Mahan

Many-Particle
Physics
THIRD EDITION
PHYSICS OF SOLIDS AND LlaUIDS
Editorial Board: Jozef T. Devreese • University of Antwerp, Belgium
Roger P. Evrary • University of Liege, Belgium
Stig Lundqvist • Chalmers University of Technology, Sweden
Gerald D. Mahan • University of Tennessee, USA
Norman H. March • University of Oxford, England
Current Volumes in the Series

CRYSTALLINE SEMICONDUCTING MATERIALS AND DEVICES
Edited by Paul N. Butcher, Norman H. March, and Mario P. Tosi
ELECTRON CORRELATION IN MOLECULES AND CONDENSED PHASES
N. H. March
ELECTRONIC EXCITATIONS AT METAL SURFACES
Ansgar Liebsch
EXCITATION ENERGY TRANSFER PROCESSES IN CONDENSED MATTER: Theory and
Applications
Jai Singh
FRACTALS
Jens Feder
INTERACTION OF ATOMS AND MOLECULES WITH SOLID SURFACES
Edited by V. Bortolani, N. H. March, and M. P. Tosi
LOCAL DENSITY THEORY OF POLARIZABILITY
Gerald D. Mahan and K. R. Subbaswamy
MANY-PARTICLE PHYSICS, Third Edition
Gerald D. Mahan
ORDER AND CHAOS IN NONLINEAR PHYSICAL SYSTEMS
Edited by Stig Lungqvist, Norman H. March, and Mario P. Tosi
PHYSICS OF LOW-DIMENSIONAL SEMICONDUCTOR STRUCTURES
Edited by Paul Butcher, Norman H. March, and Mario P. Tosi
QUANTUM TRANSPORT IN SEMICONDUCTORS
Edited by David K. Ferry and Carlo Jacoboni
A Continuation Order Plan is available for this series. A continuation order will bring delivery of each new
volume immediately upon publication. Volumes are billed only upon actual shipment. For further information
please contact the publisher.
Many-Particle
Physics
THIRD EDITION
Gerald D. Mahan
University of Tennessee, Knoxville, Tennessee
and Oak Ridge National Laboratory
Springer Science+Business Media, LLC

Library of Congress Cataloging-in-Publication Data
Mahan, Gerald D.
Many-particle physics/Gerald D. Mahan.-3rd ed.
p. cm. - (Physics of solids and liquids)
Includes bibliographical references and index.
ISBN 978-1-4419-3339-3 ISBN 978-1-4757-5714-9 (eBook)
DOI 10.1007/978-1-4757-5714-9
1. Solid state physics. 2. Many-body problem. 3. Green's functions. I. Title. II. Series.
QC176 .M24 2000

530A'I-dc21
00-039101
ISBN 978-1-4419-3339-3
©2000, 1990, 1981 Springer Science+Business Media New York

Originally published by Kluwer Academic/Plenum Publishers, New York in 2000
http://www.wkap.nl
ill 9 8 7 6 5 4 3 2 1
A C.l.P. record for this book is available from the Library of Congress
All rights reserved
No part of this book may be reproduced, stored in a retrieval system, or transmitted in any form or by any
means, electronic, mechanical, photocopying, microfilming, recording, or otherwise, without written permission
from the Publisher
Preface
The first, second, and third editions of this book seem to occur at ten year intervals. The intent
is to keep the book up-to-date. Many-body theory is a field which continually evolves in time.
Journals only publish new results, conferences only invite speakers to report new phenomena,
and agencies only fund scientists to do new physics. Today's physics is old hat by tomorrow.
Students want to learn new material, and textbooks must be modified to keep up with the
times.
The early chapters in this book teach the techniques of many-body theory. They are
largely unchanged in format. The later chapters apply the techniques to specific problems.
The third edition increases the number of applications. New sections have been added, while
old sections have been modified to include recent applications.
The previous editions were set in type using pre-computer technology. No computer file
existed of the prior editions. The publisher scanned the second edition and gave me a disk
with the contents. This scan recorded the words accurately and scrambled the equations into
unintelligible form. So I retyped the equations using LaTeX. Although tedious, it allowed me
to correct the infinite numbers of typographical errors in the previous edition. The earlier
typesetting methods did not permit such corrections. The entire book was edited sentence-by-
sentence. Most old sections of the book were shortened by editing sentences and paragraphs.
I also contemplated removing entirely some old sections. Each time I did this, and told
somebody, they always remarked that the deleted section was their favorite, and I simply
could not remove it. While it is gratifYing to have so many sections be everyone's favorite, it
does make shortening the book somewhat hard! In the end I gave up, and no sections were
removed. Many were rewritten to shorten them. Since many new sections were added, the
book gets longer with each edition. The reference list was updated.
New sections include: Bethe lattice, different mean-free-paths, Hubbard model,
Coulomb blockade, Landauer transport, and the Quantum Hall effect. The big problem is
what to say about high-temperature superconductivity. Although much experimental infor-
mation is available regarding this important topic, the theoretical picture is quite uncertain.
There is no agreed understanding of the pairing mechanism which causes the high transition
temperature. It is hard to write a text book on a topic for which there is little agreement
regarding fundamental theory. In the end, I mentioned only some important experiments and
their results, and added little new information on the theory mechanisms. The section on the
gap equation was rewritten to use the modem method of solving it in complex frequency
space, rather than the older method of real frequency space.
v
vi Preface
I thank Steve Girvin for his proofreading twice the various versions of the section on the
Quantum Hall effect, and Koung-An Chao for teaching me about quantum dots. I also very
much thank my wife Sally for letting me spend every evening and weekend for one year
preparing this new edition.
Contents
1. Introductory Material. 1
1.1. Harmonic Oscillators and Phonons . 1
1.2. Second Quantization for Particles 11
1.3. Electron-Phonon Interactions 26
1.3.1. Interaction Hamiltonian . 27
1.3.2. Localized Electron 29
1.3.3. Deformation Potential 31
1.3.4. Piezoelectric Interaction. 32
1.3.5. Polar Coupling. 34
1.4. Spin Hamiltonians 36
1.4.1. Homogeneous Spin Systems 38
1.4.2. Impurity Spin Models 43
1.5. Photons. 48
1.5.1. Gauges 49
1.5.2. Lagrangian . 53
1.5.3. Hamiltonian 55
1.6. Pair Distribution Function 58
Problems 62
2. Green's Functions at Zero Temperature. 65

2.1. Interaction Representation. 66
2.1.1. Schr6dinger . 66
2.1.2. Heisenberg . 66
2.1.3. Interaction 67
2.2. S Matrix 70
2.3. Green's Functions 71
2.4. Wick's Theorem. 76
2.5. Feynman Diagrams. 81
2.6. Vacuum Polarization Graphs . 83
vii
viii Contents
2.7. Dyson's Equation . . . . . . 86

2.8. Rules for Constructing Diagrams. 90
2.9. Time-Loop S Matrix. . . . 95
2.9.1. Six Green's Functions. 96
2.9.2. Dyson's Equation 99
2.10. Photon Green's Functions 102
Problems . . . . . . . . 106
3. Nonzero Temperatures. 109

3.1. Introduction 109
3.2. Matsubara Green's Functions 112
3.3. Retarded and Advanced Green's Functions 118
3.4. Dyson's Equation 128
3.5. Frequency Summations 136
3.6. Linked Cluster Expansions 142
3.6.1. Thermodynamic Potential 142
3.6.2. Green's Functions. 152
3.7. Real-Time Green's Functions. 154
3.7.1. Wigner Distribution Function. 157
3.8. Kubo Formula for Electrical Conductivity 160
3.8.1. Transverse Fields, Zero Temperature 163
3.8.2. ~onzero Temperatures 168
3.8.3. Zero Frequency 170
3.8.4. Photon Self-Energy 173
3.9. Other Kubo Formulas 174
3.9.1. Pauli Paramagnetic Susceptibility 174
3.9.2. Thermal Currents and Onsager Relations . 177
3.9.3. Correlation Functions 181
Problems 183
4. Exactly Solvable Models 187

4.1. Potential Scattering . 187
4.1.1. Reaction Matrix 189
4.1.2. T Matrix. 192
4.1.3. Friedel's Theorem. 195
4.1.4. Impurity Scattering 199
4.1.5. Ground State Energy. 204
4.2. Localized State in the Continuum 207
4.3. Independent Boson Models 218
4.3.1. Solution by Canonical Transformation . 218
4.3.2. Feynman Disentangling of Operators 221
4.3.3. Einstein Model. 224
4.3.4. Optical Absorption and Emission 228
4.3.5. Sudden Switching. 236
Contents ix
4.3.6. Linked Cluster Expansion . 241

4.4. Bethe Lattice . 247
4.4.1. Electron Green's Function . 247
4.4.2. Ising Model. 251
4.5. Tomonaga Model 256
4.5.1. Tomonaga Model . 257
4.5.2. Spin Waves. 262
4.5.3. Luttinger Model 264
4.5.4. Single-Particle Properties 267
4.5.5. Interacting System of Spinless Fermions 272
4.6. Polaritons 276
4.6.1. Semiclassical Discussion 276
4.6.2. Phonon-Photon Coupling 278
4.6.3. Exciton-Photon Coupling 282
Problems 291
5. Homogeneous Electron Gas 295

5.1. Exchange and Correlation. 295
5.1.1. Kinetic Energy. 297
5.1.2. Hartree 297
5.1.3. Exchange 297
5.1.4. Seitz's Theorem 301
5.1.5. 2:(2a) 303
5.1.6. 2:(2b) 304
5.1.7. 2:(2c) 305
5.1.8. High-Density Limit 306
5.1.9. Pair Distribution Function . 308
5.2. Wigner Lattice 311
5.3. Metallic Hydrogen 315
5.4. Linear Screening . 316
5.5. Model Dielectric Functions 323
5.5.1. Thomas-Fermi. 323
5.5.2. Lindhard, or RPA . 325
5.5.3. Hubbard. 336
5.5.4. Singwi-Sj6lander . 338
5.5.5. Local Field Corrections. 341
5.5.6. Vertex Corrections 343
5.6. Properties of the Electron Gas 346
5.6.1. Pair Distribution Function . 346
5.6.2. Screening Charge. 346
5.6.3. Correlation Energies . 347
5.6.4. Compressibility 352
5.6.5. Pauli Paramagnetic Susceptibility 356
5.7. Sum Rules. 358
5.8. One-Electron Properties 362
5.8.1. Renormalization Constant ZF . 365
x Contents
5.8.2. Effective Mass. 368

5.8.3. Mean-Free-Path 369
Problems . 372
6. Strong Correlations 375

6.1. Kondo Model. 375
6.1.1. High-Temperature Scattering 376
6.1.2. Low-Temperature State . 383
6.1.3. Kondo Temperature 387
6.1.4. Kondo Resonance. 387
6.2. Single-Site Anderson Model . 389
6.2.1. No Hybridization. 391
6.2.2. With Hybridization 395
6.2.3. Self-Energy of Electrons 396
6.3. Hubbard Model 403
6.3.1. Spin and Charge Separation 404
6.3.2. Exchange Graphs . 409
6.4. Hubbard Model: Magnetic Phases . 411
6.4.1. Ferromagnetism 413
6.4.2. Antiferromagnetism . 416
6.4.3. An Example 422
6.4.4. Local Field Corrections. 427
Problems 430
7. Electron-Phonon Interaction 433

7.1. Frohlich Hamiltonian 433
7.1.1. Brillouin-Wigner Perturbation Theory . 434
7.1.2. Rayleigh-Schrodinger Perturbation Theory 438
7.1.3. Strong Coupling Theory 444
7.1.4. Linked Cluster Theory 448
7.2. Small Polaron Theory . 454
7.2.1. Large Polarons . 455
7.2.2. Small Polarons . 456
7.2.3. Diagonal Transitions. 458
7.2.4. Nondiagonal Transitions 459
7.2.5. Kubo Formula . 463
7.3. Heavily Doped Semiconductors . 467
7.3.1. Screened Interaction . 468
7.3.2. Experimental Verifications . 474
7.3.3. Electron Self-Energies 475
7.4. Metals 481
7.4.1. Phonons in Metals 482
7.4.2. Electron Self-Energies 487
Problems 495
Contents xi
8. dc Conductivities 499
8.1. Electron Scattering by Impurities 499
8.1.1. Boltzmann Equation . 499
8.1.2. Kubo Formula: Approximate Solution . 505
8.1.3. Ward Identities. 514
8.2. Mobility of Frohlich Polarons 517
8.3. Electron-Phonon Relaxation Times 524
8.3.1. Metals 526
8.3.2. Semiconductors 527
8.3.3. Temperature Relaxation . 531
8.4. Electron-Phonon Interactions in Metals 534
8.4.1. Force-Force Correlation Function 534
8.4.2. Kubo Formula. 537
8.4.3. Mass Enhancement 545
8.4.4. Thermoelectric Power 545
8.5. Quantum Boltzmann Equation 549
8.5.1. Derivation of the QBE 550
8.5.2. Gradient Expansion . 554
8.5.3. Electron Scattering by Impurities. 557
8.6. Quantum Dot Tunneling 561
8.6.1. Electron Tunneling 561
8.6.2. Quantum Dots . 567
8.6.3. Rate Equations. 571
8.6.4. Quantum Conductance 575
Problems 576
9. Optical Properties of Solids . 579

9.1. Nearly Free-Electron Systems 579
9.1.1. General Properties 579
9.1.2. Force-Force Correlation Functions 581
9.1.3. Frohlich Polarons . 585
9.1.4. Interband Transitions 588
9.1.5. Phonons. 590
9.2. Wannier Excitons 592
9.2.1. The Model 592
9.2.2. Solution by Green's Functions 596
9.2.3. Core-Level Spectra 600
9.3. X-ray Spectra in Metals 603
9.3.1. Physical Model 603
9.3.2. Edge Singularities 607
9.3.3. Orthogonality Catastrophe . 612
9.3.4. MND Theory 621
9.3.5. XPS Spectra 624
Problems 626
xii Contents
10. Superconductivity 627

10.1. Cooper Instability. . 628
10.1.1. BCS Theory. 635
10.2. Superconducting Tunneling. 644
10.2.1. Normal-Superconductor . 645
10.2.2. Two Superconductors. 648
10.2.3. Josephson Tunneling. 652
10.2.4. Infrared Absorption 660
10.3. Strong Coupling Theory . 664
10.4. Transition Temperature 670
Problems . . . . . . . . 674
11. Superfluids . . . . . . 677

11.1. Liquid 4He . . . . . . . 677
11.1.1. Hartree and Exchange . 679
11.1.2. Bogoliubov Theory of 4He . 682
11.1.3. Off-Diagonal Long-Range Order. 686
11.1.4. Correlated Basis Functions. 690
11.1.5. Experiments on nk' . . . 697
11.1.6. Bijl-Feynman Theory. . . 702
11.1.7. Improved Excitation Spectra 707
11.1.8. Superfluidity. . . . 710
11.2. Liquid 3He . . . . . . . . . 713
11.2.1. Fermi Liquid Theory. . . 714
11.2.2. Experiments and Microscopic Theories. 720
11.2.3. Interaction Between Quasiparticles: Excitations 723
11.2.4. Quasiparticle Transport 729
11.2.5. Superfluid 3He . 735
11.3. Quantum Hall Effects. . . . 742
11.3.1. Landau Levels. . . 742
11.3.2. Classical Hall Effect . 745
11.3.3. Quantum Hall Effect. 747
11.3.3.1. Fixed Density 749
11.3.3.2. Fixed Chemical Potential. 749
11.3.3.3. Impurity Dominated 750
11.3.4. Laughlin Wave Function. 752
11.3.5. Collective Excitations. . 757
11.3.5.1. Magnetorotons 757
11.3.5.2. Quasiholes 760
Problems . . 761
References . 765
Author Index. 777
Subject Index. 781
Chapter 1
Introductory Material
1.1. HARMONIC OSCILLATORS AND PHONONS
First quantization in physics refers to the property of particles that certain operators do not
commute:
[x,Px] = ifz (1.1)
E -+ ili~ (1.2)
at
Later it was realized that forces between particles were caused by other particles: photons
caused electromagnetic forces, pions caused some nuclear forces, etc. These particles are also
quantized, which leads to second quantization. The basic idea is that forces are caused by the
exchange of particles, and the number of particles is quantized: one, two, three, etc. The
quantization imparts a quantum nature to the classical force fields.
In solids the vibrational modes of the atoms are quantized because of first quantization
(1.1.1). These quantized vibrational modes are called phonons. An electron can interact with a
phonon, and this phonon can travel to another electron, interact, and thereby cause an indirect
interaction between electrons. Indeed, the phonon need not move but can vibrate until the next
electron comes by. The induced interaction between electrons is an example of second
quantization. The phonons play a role in solids similar to the classical fields of particle
physics. They cause quantized interactions between electrons.
Phonons in solids can usually be described as harmonic oscillators. A fuller description
of the effects of anharmonicity is introduced later. But, for the moment, this idea should be
sufficient motivation to study the harmonic oscillator. The one-dimensional harmonic oscil-
lator has the Hamiltonian
p2 K
H=-+-~ (1.3)
2m 2
1
2 Chap. 1 • Introductory Material
To solve this Hamiltonian, introduce a dimensionless coordinate s:

0)
2
=-
K (1.4)
m
mO)) 1/2
S=x ( T (1.5)
.a
-1-=--
p
(1.6)
as ../flmO)
and
(1.7)
The harmonic oscillator Hamiltonian has a solution in terms of Hermite polynomials. The
states are quantized such that
(1.8)
where n is an integer. Use Dirac notation for the eigenstates In) = ~n' One can also learn by
direct calculation that the following matrix elements exist for the operators x and p:
(1.9)
1/2
I _ •( mflO) 1/2 1/2
-2-)
I
(n IPln) - 1 [(n) 0n'=n+1 - (n) 0n'=n-d
It is customary to define two dimensionless operators as follows:
(1.10)
They are Hermitian conjugates of each other. They are sometimes called raising and lowering
operators, but here they are called creation (at) and destruction operators (a). The Hamil-
tonian (1.7) may be written with them as
H = flO) [aa t + ata] (1.11)

2
= fI; [~ (S+ ~) ( S- a~) + ~ ( S- :s) (s+ :1;) ] (1.12)
H = fI; (-:2 + 1;2) (1.13)

Sec. 1.1 • Harmonic Oscillators and Phonons 3
A very important property of these operators is called commutation relations. These are
derived by considering how they act, sequentially, on any functionf(~). The two operations a
and at in turn give
while the reverse order gives
at af(~) = -
2
1( 8)( + -8)
~ - -
8~
~
8~
1
f(~) = - (~ 2f
2
- f - f ") (1.15)
These two results are subtracted,
[aa t - at a]f(~) = f(~) (1.16)
and yield the original function. The operator in brackets is replaced by a bracket with a
comma,
(1.17)
which means the same thing. The relationship (1.16) is usually expressed by omitting the
functionf(~):
(1.18)
In a similar way, one can prove that
[a, a] = 0 (1.19)
[at, at] = 0 (1.20)
These three commutators, plus the Hamiltonian
H = lim [aat + at a] = lim [aa t - at a + 2at a] = lim[a t a +!] (1.21)

2 2
completely specify the harmonic oscillator problem in terms of operators. With these four
relationships, one can show that the eigenvalue spectrum is indeed (1.8), where n is an integer.
The eigenstates are
In} = (atr 10) (1.22)

Jnf
where 10} is the state of no phonons which obeys
alO} = 0 (1.23)
and where the n! is for normalization. Operating on this state by a creation operator gives
atln} = (a t r+1 10) = (n + 1)1/2 (at r+110) (1.24)

Jnf J(n + I)!
= (n + 1)1/2In + I) (1.25)
the state with the next highest integer. The only matrix element between states are
(n'latln) = (n + 1) 1/2 0n'=n+1 (1.26)
(n'laln) = (n)1/2 0n'=n_1 (1.27)

The second expression is derived from the first by taking the Hermitian conjugate of the first,
and then exchanging dummy variables nand n'. Alternately,
aln) = (n)1/2In - 1) (1.28)
So the destruction operator a lowers the quantum number. Then operating by the sequence
(1.29)
gives an eigenvalue n, which verifies the eigenvalue (1.8). Furthermore, using the original
definitions (1.10) permits us to express x and p in terms of these operators,
x= - ( Ii
2mro
)1/2
(a+a t ) (1.30)
P = i (m2 ro
Ii) 1/2 (at - a) (1.31)
and the matrix elements (1.9) follow immediately:
(n'lxln) = (2mro
Ii) 1/2 [(n'laln) + (n'latln)]
(n'lPln) = i m2 ro ( Ii )1/2
[(n'latln) - (n'laln)]
1/2
_ .( mliro) 1/2 1/2
- I -2- [en + 1) On'=n+1 - (n) 0n'=n-d
The description of the harmonic oscillator in terms of operators is equivalent to the

conventional method of using wave functions \jJn(~) of position.
The time dependence of these operators is often important. In the Heisenberg repre-
sentation of quantum mechanics, the time development of operators is given by = 1) (Ii
OCt) = eiHtOe-iHt (1.32)
so that the operator obeys the equation
i a~;t) = i[H, OCt)] (1.33)
For the destruction operator, this equation becomes

a = i[H, a] = iro[a t aa -
ita aat a] = iro[a t , ala = -iroa (1.34)
which has the simple solution
(1.35)
The reference point of time may be selected arbitrarily, so that the operators have an arbitrary
phase factor associated with them. This phase is unimportant, since it cancels out of all final
results. The Hermitian conjugate of this expression is
(1.36)
The time development of the position operator can be represented as
(1.37)
This result for x(t) will be used often in discussing phonon problems.
Another familiar problem which can be solved with operators is a charged harmonic
oscillator in a constant electric field F:
(1.38)
( Ii)
'A=eF - -
2moo
1/2
(1.39)
This Hamiltonian may be solved exactly. First consider the equation of motion for the time
development of the destruction operator:
-aa. .
at = I[H, a] = -l(ooa + 'A) (1.40)
The right-hand side is no longer just proportional to a, since there is the constant term.
However, let us define a new set of operators by the relationships
'A
A=a+- (1.41)
00
(1.42)
They obey the equation
aA
- = -iooA (1.43)
at
so they have the simple time development
A(t) = e-iOltA (1.44)
At(t) = eiOltAt (1.45)

Indeed, one can show that they have the following properties:
[A,At] = [a+~,at +~J = 1 (1.46)
[A,A] =0 (1.47)
[At, At] =0 (1.48)
H=m[(At -~)(A-~) +~J +A(A+At -2~) (1.49)
= m(AtA + D-: (1.50)
It was remarked above that any set of operators with these properties had a solution in terms
of harmonic oscillator states:
(1.51 )
(1.52)
The operator for the position is
Ii )1/2( 2A)
x(t) = ( 2mm Ae- iIDt + At e imt - ffi (1.53)
The physics of the Hamiltonian (1.38) is very simple. The spring stretches to a new equili-
brium point which is displaced a distance
Xo = _ (_li_) 1/22A
2mm m
= _ eF
K (1.54)
from the original one. It oscillates about this new equilibrium with the same frequency m as
before. These oscillations are still quantized, in units of m. The energy - A2; m = _e2F2;2K
is that gained by the spring from the displacement along the electric field. One can get the
same result directly in coordinate space. The Hamiltonian is written as
H = p2 +!i (x + eF) 2
_ e2F2
2m 2 K 2K (1.55)
and a new coordinate x' = x + eF; K is defined which obeys

[x',p] = iii (1.56)
The variable x' describes the simple harmonic motion. This result completes the discussion of
the harmonic oscillator in an electric field.
In a solid there are many atoms, which mutually interact. The vibrational modes are
collective motions involving many atoms. A simple introduction to this problem is obtained
by studying the normal modes of a one-dimensional harmonic chain:
H = L 2m
pr +"2K L(Xi - Xi+l) 2
(1.57)
I I
The classical solution is obtained by solving the equation of motion:
-mxj = ma/xj = K(2xj - xJ+I - Xj-I) (1.58)
A solution is assumed of the form Xj = Xo cos(kaj) and the force term becomes
2Xj - xj+1 - xj_1 = xo[2 cos(kaj) - cos(kaj + ka) - cos(kaj - ka)]
= 2xo cos(kaj)[1 - cos(ka)]
The normal modes have the solution
2= -2K [1 -
ffik cos(ka)] = -4K .
sm 2(ka)
- (1.59)
m m 2
The quantum mechanical solution begins by defining some normal coordinates, assuming
periodic boundary conditions:
1 ikal
(1.60)
Pk = ,JNLl e PI
This choice maintains the desired commutation relations in either real space or wave vector
space:
[XI,Pm] = iDlm (1.61)
[Xk' P k' ] ~ '""' e- ikal eik'am[xI, P m ]

= N~ (1.62)
I,m
_ .i '7:
- N
'""' eiam(k' -k) -- 1'S:
Uk,k' (1.63)
From the general result
It is easy to show that the potential energy term in (1.57) is
The Hamiltonian may be written in wave vector space as
(1.64)
The Hamiltonian has the fonn of a simple hannonic oscillator for each wave vector. Define
the creation and destruction operators as
(1.65)
a t -_ (mWk)
- - 1/2 ( X k - -i- P k ) (1.66)
k 2ft mWk -
They obey the commutation relations
t i
[ab ak,] =- 2ft {[XbPk'] - [P-k' x_k']} = Ok,k' (1.67)
[ak' ak'] =0 (1.68)
[at, at,] =0 (1.69)
and the Hamiltonian may be written as
(1.70)
These collective modes of vibration are called phonons. They are the quantized version of the
classical vibrational modes in the solid. These are the same commutator relations, and
Hamiltonian, as in the simple hannonic oscillator. Each wave vector state behaves indepen-
dently, as a hannonic oscillator, with a possible set of quantum numbers nk = 0, 1, 2, ... The
state of the system at any time is
(1.71)
so that the expectation value of the Hamiltonian is
(1.72)
In thennal equilibrium the states have an average value of nk which is given in tenns of the
temperature ~ = IjkBT:
(1. 73)
The system fluctuates around this average value.

The position operator in wave vector space, and real space, is
(1. 74)
Often mN is replaced by the equivalent quantity mN = pv, where p is the mass density and v
is the volume. At some point there is a summation over the discrete set of eigenstates for the
system of finite volume v. It is convenient to change the summation to an integration:
lim -I I:f(k)
v-->oo V k
J(2n)
d k
= -3f(k)
3
(1.75)
During this change, any delta functions must change from discrete delta functions (called
Kronecker deltas) to continuous ones. Since
-I f(k ,) = -I " f(k)8 k

V v -t .k' =
J-3f(k)8
d k
(2n)
3
.
k k' (1.76)
= (2n)3
v
J(2n)3
d 3k f(k)8(k - k') (1.77)
it is concluded that
• I: (2n)3 1:( ')

11m Uk k' = --u k - k (1.78)
v-->oo· V
In general, our preference is to write wave vector summations as discrete summations until it
is time to do the integrals and only then make the changes (1.75).
The quantum mechanical solution has the same frequencies as found in the classical
solution. Quantum mechanics only enters in a quantization of the amplitude of the oscillation.
The phonons occur in discrete numbers with zero, one, two, etc., phonons in each state k.
When the average number of phonons is large, nk » I, the quantization is irrelevant, since
the system behaves classically. The quantum nature of the field is more important when the
average number of phonons in each state k is small.
In three-dimensional solids, the theory is nearly identical except there are more indices.
Suppose there is a potential function between atoms or ions of the form
(1.79)
where Ri is the position of an atom. If it is vibrating, then denote R~O) as the equilibrium
position and Qi as the displacement from equilibrium:
Ri = R~O) +Qi (1.80)

The potential function is expanded in a Taylor series about the equilibrium position:
V(Ri - Rj ) = V(R~O) - RjD)) + (Qi - Q) . VV(R~O) - RjD))
1 ;P (0) (0)
+2(Qi-Q)~(Qi-Q)vaR aR V(Ri -Rj )
~ v
(1.81)
The term linear in displacement vanishes, because one defines the equilibrium position R(O) as
the place where the sum of the forces on an ion j is zero:
Fj = I: VV(R~O) - RjD)) =0 (1.82)

i
The important term is the one which is quadratic in the displacements. It gives the potential
energy of the phonons:
x (R(O) _ R(O») (1.83)

~v I J
 (R(O) _ R(O») = _iP_ V(R(O) _ RCa») (1.84)

~v / J aR aR v /~
J
The interaction is evaluated in wave vector space by trying an expansion of the form
Jj ) 1/2 . . (0)
Q/.(t)=i" (
~ 2MN(il
t
Sk).
(a
k,A.
e-OJk)J+at
-k,A.
e/OJk)J)e,k.R, (1.85)
k,A. kA.
where M is the ion mass. The factor of i on the right-hand side of the equation is required to
make Q; = Qi since it represents a real displacement in space. Take the Hermitian conjugate
of (1.85) and change k -+ -k. Nearly the same result is obtained as found in Qi' To make it
Hermitian requires that
(1.86)
The polarization vectors I;k,A. are assumed to be real but change sign with k direction,
Ck = -I;k and the above identity is satisfied. Since the displacement is in three dimensions,
there are 3L normal modes for each value of wave vector. Here L is the number of atoms per
unit cell of the crystal. The index A runs over these 3L values of normal mode. Each mode
will have its own eigenfrequency (ilkA.' It will also have a polarization vector I;k,A. which
specifies the vibrational direction of the ion for each wave vector and mode A. If there are
more than one atom per unit cell, one should add further subscripts to M and I;k,A. to specifY
the values for each atom per unit cell.
The right-hand side of (1.81) may be written as
= "~ V ((0)
R; - Rj(0») + M"
'2 ~ Qn . Q-kA.(ilkA.
2 (1.87)
ij ~A.
The first term is a constant which will be neglected in our discussion of vibrational modes.
The eigenvalues (iln are those solved in the harmonic approximation. In this approximation,
one retains the quadratic term only in the displacements in the Hamiltonian. To be more
careful, a third-order anharmonic term V3 can be added for the case of one atom per unit cell:
H=HO+ V3 (1.88)
1 2 2
Ho = 2M L(P n . P -n +M (iln QkA. . Q-kA.) (1.89)
k,A.
(1.90)
(1.91)
Sec. 1.2 • Second Quantization for Particles 11
and one solves Ho for the modes. These modes are harmonic oscillator states for each wave
vector k and mode A. The harmonic approximation applies to any theory of phonons which
retains only the terms which are quadratic in the displacements Qj. Actual solids are described
by potential functions which are more complicated than the central force field VCR; - R)
which we have assumed. For example, in semiconductors there are usually bond bending
forces between nearest neighbors. Nevertheless, the Hamiltonian is still written as (1.90) in
the harmonic approximation. A complete description of these calculations is given by Born
and Huang (1954) or Maradudin et al. (1963).
The terms in the Taylor series higher than quadratic are treated as perturbations. These
are called anharmonic effects and are very important in solids with light atomic masses:
hydrogen, helium, lithium, etc. They are often important in some other solids. The first term is
usually cubic in the displacements and has the form shown above. In terms of wave vectors it
becomes
(1.92)
The matrix element in the cubic term is quite complicated, and it will not be written out. It is
difficult to determine from first principles anyway. If the first two displacements Q have wave
vectors k and q, the third has -k - q to ensure wave vector conservation. This interaction
may be written in terms of creation and destruction operators by using (1.85). In this
representation, it is apparent that these cubic terms permit one phonon to decay into two and
VIce versa.
For solids in which the anharmonic terms are important, one must try to include the
effects of the cubic perturbation V3 and perhaps also higher terms such as quadratic. Including
these terms is a many-body problem. The effects are quite temperature dependent, so it is
necessary to use Green's functions at nonzero temperatures.
A word about notation. We dislike subscripts and superscripts. In discussing phonons,
the subscript A will usually be omitted, although it should be carried in every expression. The
summation over phonon modes is really meant to imply summation over wave vectors and
modes A.
1.2. SECOND QUANTIZATION FOR PARTICLES
There are two ways to introduce the subject of creation and destruction operators for
particles. The first is to describe their properties and then to omit any proofs. One could just
remark that they work, which is why we use them. The second way is to go through elaborate
justification arguments. These tend to leave the reader more confused than convinced. Here an
intermediate approach is tried. A short justification will be attempted. Our discussion follows
Schiff (1968).
The first treatment is for boson particles which cannot be destroyed. It is hard to think of
a fundamental particle with this property. The method is usually applied to composite
particles such as 4He which contain even numbers of fermions, so that it has boson-like
properties. In any case, it is assumed the discussion is for a point particle. If it is in a potential
U(r), the one-particle Schr6dinger equation is
. = H\jJ(r) = [1i2V2]
ili\jJ(r) - 2m + U(r) \jJ(r) (1.93)
This equation may be derived from the Lagrangian density
L = jll\jJt* - ~~ V\jJt . V\jJ - U(r, t)\jJt\jJ (1.94)
The wave function is complex, with real and imaginary parts. These two parts can be treated
as independent variables in the Lagrangian. An alternate procedure is to treat \jJ(r) and \jJt(r)
as independent variables. Then the usual variations give
(1.95)
(1.96)
aL _
• -
'.1:.,I,t
1ft,!, (1.97)
a\jJ
When these relations are put into Lagrange's equation,
o_aL_,,~( aL )_~aL (1.98)

- a\jJ ~ ax~ a( a\jJ / ax~) at a*
0= -U\jJt + ~ V2\jJt - if! ~\jJt (1.99)

2m at
the Hermitian conjugate of Schrodinger's equation is recovered. If the same manipulations are
tried with \jJt as the variable, then Schrodinger's equation itself is derived. In the Lagrangian
formulation, the momentum which is conjugate to the variable \jJ is
(1.100)
The Hamiltonian density is given by

f!2
1i = n* - L = -V\jJt . V\jJ + U\jJt\jJ (1.101)
2m
where one integrates over all volume to obtain the Hamiltonian
(1.102)
where the kinetic energy term was integrated by parts. Since n and \jJ are conjugate variables,
they obey commutation relations of the form
[\jJ(r, t), n(r', t)] = ifzO(r - r') (1.103)
or using (Ll 00),
[\jJ(r, t), \jJt (r', t)] = oCr - r') (1.104)
A commutation relation of this type is the fundamental basis of second quantization.

Although it has been made plausible by the derivation from a Lagrangian, it really is a basic
premise. These commutation relations may be satisfied by introducing creation and

destruction operators. Let H have eigenstates and eigenvalues of the form
H<p", = e",<p", (1.105)
fz2
H = __ V2 + U(r) (1.106)
2m
The wave function \jI{r) and its conjugate \jIt (r) are expanded in terms of this basis set:
(1.107)
\jIt(r) = Lat(t)<pr(r) (1.108)

'"
The original field commutators (1.104) are satisfied if a and at operators have their own
commutation relations:
[a",(t), a~(t)] = iiA.A.'
[aA.(t), aA.,(t)] = 0
[at(t), a~(t)] = 0 (1.109)
The commutation relations for the field variables are:

[\jI(r, t), \jI(r', t)] =0 (1.110)
[\jIt (r, t), \jIt (r', t)] = 0 (1.111)
[\jI(r, t), \jIt (r', t)] = L[a",(t), a~,(t)]<pA.(r)<pr,(r') (1.112)

A.A.'
= L <p",(r)<pHr') = ii(r - r') (1.113)

A.
One might also ask about the commutation relations at different times. How does one evaluate
(1.114)
The answer is that evaluating this expression is a many-body problem. In fact, that is one of
the goals of Green's function theory. The commutator at different times is related to the
retarded Green's function, which is defined and discussed in Sec. 3.3. The commutator at
different times is one property of the time development of the many-body system. Simple
commutation relations such as (1.109) are valid only if the operators are at the same time.
The Hamiltonian is
H = jd3r\jlt(r)'H\jI(r) = LataA.' jd3r<pr(r)'H<PA.'(r') = LeA.ataA. (1.115)

A.A.' A.
In Sec. 1.1 it was noted that any system with these commutation relations, and a Hamiltonian
(l.115), behaved as harmonic oscillators for each state A.. The eigenstates for each value of A.
have a discrete set of occupation numbers nA. = 0,1,2,3, .... All bosons have harmonic
oscillator eigenstates. For phonons, the number nA. is interpreted as the number of phonons in
state A.. For particles, the interpretation is the same. The number n", tells how many particles in
the system are in the same state A. However, for particles, unlike phonons, the total number of
particles is conserved. The many-particle wave function has the form
(1.116)
so that the Hamiltonian (1.115) has the eigenvalue of
(1.117)
In thermal equilibrium, the average number of particles in a state A is given by the usual
boson occupation factor:
(1.118)
Now there is a chemical potential, ~, which can vary with temperature and concentration. It is
absent in the phonon, and photon, cases because these excitations do not conserve particle
number. One may make as many phonons or photons as one wishes. Another operator of
interest is the density operator:
per) = ",t (r)"'(r) = L at a,,'<p~(r)(h,(r) (1.119)

",,'
The integral of per) is just the number operator:
(1.120)
Its thermal average is obtained simply by taking the thermal average of n,,:
(1.121)
This equation serves as a definition of the chemical potential and determines its variations
with temperature and particle number N.
The Hamiltonian (1.115) and number operator (1.120) are bilinear in creation and
destruction operators. They contain only two operators, one of each kind. Hamiltonians of
these kinds may always be solved, at least in principle. The problem may always be reduced to
the diagonalization of a matrix. For example, consider the solution of our Hamiltonian:
1i 22
H = __ \7 + U(r) (1.122)
2m
Suppose that we were unable to solve it exactly-which is an improbable assumption, since

SchrOdinger's equation for one-particle potentials may be solved in milliseconds on the
computer. Anyway, suppose there were another complete set of states <Pn which are the
solution to some other Hamiltonian. Expand the wave function in terms of these states,
(1.123)
where the creation and destruction operators bn have the usual commutation relations:
[b n, b~l = ()n m (1.124)
[b n, bml = 0 (1.125)
[b!, b~l = 0 (1.126)
For the Hamiltonian and number operators:
H= Lb!bmHnm (1.127)
nm
(1.128)
n
Hnm = Jd3r<p:(r)H<Pm(r) (1.129)
This Hamiltonian may be solved in the following fashion. Examine the equation of motion for
the destruction operator:
(1.130)
It is assumed that an operator has the time development

bn(t) = bn(O)e-iEt (1.131)
so that the solutions are of the form
o= L(Hnm - E()nm) (1.132)
m
The eigenvalues E are the solution to

detlHnm - EO nm I = 0 (1.133)
It is only necessary to find the eigenvalues of the Hamiltonian matrix. Usually the matrix is of
infinite dimensionality, since there are an infinite number of states in the set <Pn- But one may
often diagonalize it exactly for many problems. Computers allow very accurate solutions for
any case of interest. If all Hamiltonians had only bilinear operators, then many-body theory
would only be an exercise in matrix diagonalization. Fortunately, it is more fun than that.
Many-body theory is used to study Hamiltonians which have additional terms. These
terms may be interactions with phonons, spin effects, or particle-particle interactions. The
effects of particle-particle interactions may be understood by examining a many-particle
Hamiltonian of the form
I
H= LHi+-LV(ri -r) (1.134)
i 2i#
(1.135)
The first term contains a summation of one-particle Hamiltonians Hi' This term by itself is
just as simple to solve as Hi alone. A collection of particles which do not mutually interact
makes for a trivial problem. One solves the dynamics of one particle, and the total properties
are the summation of the individual ones. The term which makes the Hamiltonian hard to
solve is the particle-particle interaction Veri - r). This term is multiplied by one-half since
the double summation over (ij) counts each pair twice. This interaction is written in terms of
creation and destruction operators:
H = I: Hnmb~bm +! I: Vklmnb!b~bnbl
nm klmn
(1.136)
The interaction term contains two creation and two destruction particles. This term is inter-
preted as describing two-particle scattering events. One particle in state 1 scatters to state k,
while another in state n scatters to state m. The process is illustrated in Fig. 1.1. Each index
(k, I, m, n) runs over all possible values. There are processes where, for example, k = I, which
describes a process where one particle scatters from n to m, while the other does not change
its state. This term was carefully written so that both destruction operators are to the right and
the creation operators to the left. The reason for this arrangement is to eliminate processes
whereby a particle interacts with itself. For example, if this expression is written as
(1.137)
then we could have m = I, and this term would describe how one particle interacts with itself.
For example, if 10) is the particle vacuum where all nm = 0, then bklO) = O. The state btlO)
contains one particle in state ll. The interaction operator
! klmn
I: Vklmnb!b~bnblbtIO) = 0 (1.138)
on this state gives zero, since two particles cannot interact if there is only one particle in the
system. However, the incorrect version of the interaction gives
(1.139)
is nonzero, and this term (ll = n, 1 = m) must just be the particle interacting with itself. These
terms are avoided by writing the pairwise interaction in the form (1.136).
For a gas or liquid of 4He particles, a common basis set is just free-particle wave
functions,
I Ok
o/(r) = - " e' ·ra (1.140)
JvT k
FIGURE 1.1
in which case the Hamiltonian has the form
Jc2
Ek=-
2m
V(q) = Jd 3 re iq ' r V(r) (1.141)
The operator Pq is the particle density operator in the plane-wave representation:
(1.142)
The simplest way to write the interaction term in (1.141) is in the form
1
2v L V(q)pqP_q (1.143)
q
which is defective because both destruction operators are not to the right of the creation
operators. There is a term where the particle interacts with itself
A possible difficulty with (1.141) is that V(q) may not exist. The potential V(r) may not
possess a Fourier transform if the particle-particle potential is too divergent at small values of
r. This divergence happens, for example, with the Lennard-Jones potential, which is often
used to represent the helium-helium potential. This difficulty may be avoided by summing
subsets of diagrams to get a T-matrix interaction, which is always well behaved.
So far the discussion has concerned boson operators and boson Hamiltonians. Now
consider fermions. These are usually electrons, although occasionally one studies holes,
positrons, or 3He particles. Fermions have the property that any state may contain only zero or
one particle, which is the famous exclusion principle. Jordan and Wigner (1928) discovered
that Fermi statistics could be accomplished by making the fields anti commute, which is
represented by curly brackets:
",(r)",t (r') + ",t (r')"'(r) == {"'(r), ",t (r')} = o(r - r') (1.144)
{"'(r), ",(r')} = 0 (1.145)
{",t (r), ",t (r')} = 0 (1.146)
If these wave functions are expanded in a basis set <pA.(r),
(1.147)
(1.148)
the coefficients ct and CA become creation and destruction operators which obey anti-
{CA ' c~,} = 8AA, (1.149)
{cA, cd =0 (1.150)
{ct, c~,} = 0 (1.151)
For example, consider {cA' cAl = 2cAcA = O. The operator CACA acting upon anything gives
zero, since CA is a destruction operator which destroys a particle from a state A. A state may
only contain zero or one particle, which are called 10h and 11)1., respectively:
cAlih = 10h ctllh = 0
cAIO)A = 0 ctlO)A = 11)1. (1.152)

So cAcA acting upon either IIh or 10h gives zero. Similarly, the combination ctct = O. It is
zero because two particles cannot be created in the same state. Another way to see this result
is to consider the number operator for a state,
(1.153)
and its square,
NA
2 t t
= cAcAcAcA (1.154)
Using the anticommutation relations cct = 1 - ct c gives
Nf = ct(1 - ctc A) CA = ctc A- ctctcAcA = NA (1.155)

The only numbers which are equal to its square are 0 and 1. The number NA may only be 0 or
1. The anti commutation relations have built into them the fermion property that no more than
one particle may be in the same state A.
Quite often the Hamiltonians for fermion calculations are of the form
H= ~[;~ + U(ri) +!~ Veri - r)] (1.156)
with the particles interacting with a potential U(r) and with each other through particle-
particle interactions Veri - r). The Hamiltonian is written in terms of creation and
destruction operators exactly as in the boson case:
H= L Hnmchcm +! L VklmnC!C~CnCI
nm klmn
Hnm J
= d 3nPn(r)* [ - 11;:2 + U(r) ] <Pm(r)
Vk1mn = Jd3ri Jd\<pZ(rJ<ptCrJV(ri - r)<p:(r)<Pn(r) (1.157)
The difference in behavior between fermions and bosons is often due to the difference in
the commutation relations of the operators. The starting Hamiltonians are of similar form,
e.g., for liquid 3 He and 4 He.
The particle-particle interaction term is still interpreted in terms of Fig. 1.1. Great care
must be used in writing this term, or else one makes a sign mistake. For example, the term
(1.158)
is wrong. The order of the two destruction operators was exchanged. But because of the
anti commutation relation
(1.159)
this interchange causes a sign change. Equation (1.158) would be correct if the entire term
were multiplied by -1.
For the study of electrons in solids, a popular basis set is plane waves. Eigenstates are
described by (p,O'), where 0' is the spin index, which is ±1 for spin up or down. The
Hamiltonian then has the form
t I t t
H =L + L U(q)Pq + 2- L VqCk+qcrCk'_qcr,ck'cr,Ckcr
tpcpcrcpcr (1.160)
pcr q V qkk' crcr'
The second term represents the interaction between the electrons and the atoms or ions of the
solid. The interaction U(q) is often represented by a pseudopotential (Harrison, 1966; Heine,
1970).
The electron density operator
(1.161)
is the same as for bosons except for the additional summation over the spin index. The last
term in (1.160) contains the electron--electron interaction, which is just a Coulomb potential.
e-
The Fourier transform of a Coulomb potential jr occurs often and has been given a special
symbol with lowercase vq . The Fourier transform is
Jr
d 3 re,q. ·r = 2ne2
Vq = e2 _ Joo rdr JI .
d(cosEl) e,qrcos6
0-1
2ne2
= - .-
Joo .
dr(eW -
. 4ne2
e- 1qr ) = - -
Joo dr sin(qr)
lq 0 q 0
2
= -4ne
2 - ( 1 - lim cos(qr) )
q r-+oo
(1.162)
The integral is not well defined, since it oscillates at infinity. It is assumed these oscillations
damp out, so that the result is
4ne2
Vq =-2- (1.163)
q
The electron-electron interactions are a significant part of many-body theory. Most calcu-
lations involve worrying about electron--electron interactions in one form or another.
The full electron gas Hamiltonian (1.160) is often too complicated to use for the more
elaborate many-body theory. Quite often it is approximated by a model Hamiltonian which
has a simpler form. Usually these model Hamiltonians look very simple but still are
impossible to solve exactly. Often they are even difficult to solve approximately! Some of
these popular models are discussed next.
The homogeneous electron gas is a model which is studied frequently to learn about
correlation effects. It has the Hamiltonian
(1.164)
The basic premise is to get rid of the atoms and to replace them with a uniform positive
background charge of density no. The homogeneous electron gas is also called the }ellium
model. One can think of taking the positive charge of the ions and spreading it uniformly
about the unit cell of the crystal. Of course, the homogeneous electron gas then has no crystal
structure. To preserve charge neutrality, the average particle density of the electron gas must
also be no. The average density of the electrons is just the q = 0 value of the density operator,
(1.165)
since the number operator may be summed to give the number of particles Ne . In writing the
Hamiltonian (1.164), the q = 0 term in the interaction term was omitted from the summation.
The omitted term has the form
(1.166)
This term was canceled by two other terms. One of these is the Coulomb interaction of the
uniform positive background with itself:
1 2 2
ze no
Jd3Ir rd_ 3r'lr' = N; r
2v q~ Vq
(1.167)
The other is the Coulomb interaction of the uniform positive background with the electrons:
J d3rd3r' N2
- e2n o p(r)--- = __e lim v
Ir - r'l v q-+O q
(1.168)
The N; term cancels when these three terms are added, and the other term may be neglected.
The Hamiltonian (1.164) describes a system which has charge neutrality. The interaction
terms with q =1= 0 describe the fluctuations which occur because of electrons interacting with
each other.
The plane-wave model is often a poor approximation of electron behavior in ionic solids.
In many solids the electrons are localized on atomic sites and only occasionally hop to
neighboring sites. This behavior is described by the tight-binding model. One simple form of
this model is bilinear in the operators:
(1.169)
The index} denotes a site at point Rj , while} + 8 represents the nearest neighbor atoms. The
number of nearest neighbors is called the coordination number and is represented by the
symbol Z. One can think of the term
WI) = Jd r<p*(r - R)[ - ~~ V2 + U(r)]<p(r - Rj+l)

3 (1.170)
as arising from the matrix elements between the orbitals <p(r) which are localized on sites Rj
and Rj+I)' The term WI) for 8 =1= 0 represents processes where the electron jumps from site} to
j + 8, while Wo is the site energy. The potential U(r) is periodic. Simple versions of the
model usually have only a single orbital state for each atomic site. More realistic versions of
the tight binding model allow for the multiple orbitals characteristic of p- or d-electrons. The
discussion will assume a single orbital state per atomic site.
The bilinear form of the Hamiltonian is trivial to solve exactly. One defines the operator
in wave vector space in the usual fashion,
1 ik.R
Cia = IV L e JCka (1.171)
yN k
1 " -ik.R (1.172)

Cka = IV L.. e JCia
yN i
and the Hamiltonian may be rewritten as
H = L W(k)ctaCka (1.173)
ka
W(k) = Leik'OWo (1.174)
o
If the Cia are fermion operators, then so are the Cka' They obey anticommutation relations
{Cka, ct'a'} = 8k=k,8 aa ,· Each mode (kcr) becomes an independent Fermi system, which may
be treated separately in thermodynamic averages. The particle energy is W(k). A very
common model is the nearest neighbor model. Here the hopping term is limited to just the
nearest neighbor ions, which are presumed to be all alike. Usually the site energy is Wo set
equal to zero, which is just an arbitrary energy renormalization, and Wo = w. Then the energy
is W(k) = WZYk, where the factor
(1.175)
is summed over the Z nearest neighbors. This solution is exact for the tight-binding model.
For example, the partition function Zp for fermions is written as
(1.176)
where Tr denotes the trace. The many-particle partition function (1.176) is obtained by
treating each state k as independent and averaging its thermodynamic properties separately. It
appears as if each electron in each state k is behaving independently. This picture is deceptive.
The electrons are not just whizzing around independently.
For example, if one were to calculate the probability that any two electrons of the same
spin are on the same atomic site, a zero answer is obtained. Two electrons of the same spin are
never on the same site in the single orbital models since they would be in the same state. Of
course, it is a basic feature of fermion many-particle wave functions that two electrons of the
same spin can never be in the same location. The motion of the electrons is correlated, which
arises from the anti symmetry of the wave function under exchange of particle position. The
motion of the electrons is not really independent.
This correlation is built into the eigenstates but does not affect the energy in the simple
bilinear model, which ignores interactions between particles. The partition function (1.176)
has the appearance of independent particle form, even with correlation in the wave function.
The anti symmetrization of the many-particle wave function does not affect the expectation
value of one-particle operators. It does affect the expectation value of two (or more )-particle
operators.
The tight-binding Hamiltonian may also contain the Coulomb interaction between
electrons. In its most general form (1.157), the interaction term is
1 t t
-2 L
klmn
Vklmnckcmcncl (1.177)
(1.178)
The four orbitals could be centered on four different sites. These are called four-center
integrals. They are usually small and nearly always neglected in many-body calculations. The
terms which are included are just the largest Coulomb terms. One possible term is the direct
interaction between two particles on different atomic sites. For example, setting k = I and
n = m for k =1= m gives
(1.179)
Vkm = Jd 3r, Jd3r2 I<p(r, - R )1 2 1r, e-2 r21 1<p*(r2 - Rm)12

k (1.180)
Note the operators can be rearranged into nknm as long as k =1= m, so that one does not have a
particle interacting with itself. At large separation the interaction terms Vkm become just a
Coulomb potential,
e2
lim Vkm --+ - - - - (1.181)
IRk-Rml---+ oo IRk - Rml
if the orbitals have s-symmetry. A Hamiltonian with these nearest neighbor Coulomb inter-
actions is similar to the lattice gas Hamiltonians introduced in Sec. 1.4.
The Hubbard model (1963) retains only the Coulomb integral which is the very largest.
All four orbitals <p(r) are centered on the same site m. This term describes the interaction
between two electrons which are on the same atom. Since two electrons cannot be in the same
state, the two on the same atom must be in different atomic states. In the simplest model,
which considers only a single orbital state on each atom, the two electrons must have different
spin configurations. One has spin up, while the other has spin down. The Hubbard model
considers the following Hamiltonian:
H = W L c]+Il,aCja + Wo L cJaCja + U L njtnj,(, (1.182)

joa ja j
U = Vkk = Jd 3r, Jd3 r21 <P(r,)1 2 ~

Ir, - r21
1<p(r2)1 2 (1.183)
The hopping term is usually limited to nearest neighbors. This Hamiltonian was also intro-
duced by Gutzwiller (1963), who studied the properties of electrons in d-bands in ferro-
magnets. It was then extensively studied by Hubbard (1963-1966). It is thought to be a good
model for electron conduction in narrow band materials, for example, in transition metal
oxides (Adler, 1967). The Hubbard model has been investigated thoroughly over the past
forty years, and its properties are starting to be understood. They are discussed in Chapter 6.
The parameter U is the Coulomb interaction between two electrons on the same atom.
Usually the model is applied to tightly bound orbitals such as d- orf-states. Then U is quite
large, perhaps 6 eY. The bandwidth Zw is sometimes taken to be smaller. However, some of
the most interesting phenomena seem to occur for U '" Z w. The Hubbard model can be
solved exactly only in one dimension, as shown by Lieb and Wu (1968). But there are two
limiting cases where exact solutions can be obtained in other dimensions. One is where
U = 0, which is the nearest neighbor tight-binding model. The other case is where the
hopping bandwidth w = O. It is the atomic limit, since here each atom is considered indi-
vidually, independently of the others. The energy in this atomic limit is
(1.184)
where Ne is the number of electrons and ii is the number of sites with two electrons.
Creation and destruction operators are used to describe other kinds of operators besides
Hamiltonians. We shall frequently use density and current operators. The density operator is
summed over the position of all particles:
per) = L o(r - r) (1.185)

j
per) may be expressed in terms of creation and destruction operators as
(1.186)
The Fourier transform of the density operator is also needed:
(1.187)
The two most popular representations are the free-particle model:
(1.188)
and the tight-binding model when omitting overlap between neighbors:
p(q) = ll(q) Lnjcreiq.Rj (1.189)

jcr
(1.190)
Another important operator is the electrical current. It is the summation over all particles and
their velocities:
j(r) = ! L ei[vio(r -
i
rJ + o(r - ri)vi] (1.191)
The summation above is over different groups of particles. Each group of particles contains
identical particles with the same charge ei' Let ji be the particle current for each kind of
particle species. Then the electrical current operator is
j(r) = L ejJr) (1.192)
t(r) = _1. {\j!t(r)V\j!(r) - \j!(r)V\j!t(r)} (1.193)

2mz
where one sums over each particle species i and the particle current ji for each species. The
second equation is the standard quantum mechanical representation for the particle current.
The Fourier transform of the current operator has the form
(1.194)
For free particles, this expression has the form
i;(q) = ~m I)k + ! q)ct+q,crCkcr (1.195)

kcr
Another case of interest is the current operator for the tight binding model. It is easier to
consider an alternate, but equivalent, formula for the current operator. The derivation starts
from the definition of the polarization operator in terms of the particle density per),
P= Jd rrp(r)
3 (1.196)
which is a summation over all the particles and their positions. One then recalls that the time
derivative of the polarization is just the particle current:
a
-P=
at
Jd rr-p(r,t)
a
3
at (1.197)
This relationship can be proved easily by using the equation of continuity, followed by an
integration by parts,
per, t) = -V . j(r, t) (1.198)
a =-
itP Jd rrV· J(r,
3 . t) = Jd rj(r) . Vr
3 (1.199)
= J
d 3 rj(r, t) (1.200)
In the tight-binding model, the polarization operator has the form
P=LRini (1.201)
where Ri is the position of a lattice site, and ni is the number of particles at that site. The time
derivative is
j=ip=i[H,P] (1.202)
at
For example, if the Hamiltonian has the form
t 1
H = W L cj+o,crCjcr + 2" ij,ss'
jocr
L nisnjs' Vij (1.203)
then only the first term contributes to the current operator. The other term contains only the
position operator n is ' which commutes with itself. Here s is the spin index. In this case the
current operator is
(1.204)
This current operator is used in calculations on localized electrons. It applies to organic solids
and narrow band ionic solids.
The energy current operator is needed to calculate energy transport in solids, which
occurs, for example, in discussions of thermal conductivity or thermoelectric effects. Energy
currents flow whenever heat is generated or dissipated nonuniformly in the solid. The energy
current jE is defined as the energy flow through a surface. It obeys an equation of energy
conservation,
(1.205)
where the energy change aHI at equals the variation in the energy flux. An equation for the
energy current may be derived by formally introducing an operator which is the integral over
the position and Hamiltonian density:
RE = ! f d r[r1i(r) + 1i(r)r]
3 (1.206)
The equation of energy continuity (1.205) may be used to show that the time derivative of this
quantity is just the energy current:
a =
atRE IJ d3[r r ata1i(r) + ata1i(r)rJ'
2" = JE (1.207)
For a free-particle system, the energy current is
(1.208)
This result is sensible. It is just the energy Ep = p2/2m of each particle multiplied by its
velocity vp = p/m In the nearest neighbor tight-binding model, one writes RE in terms of the
site Hamiltonian hi and position R i:
(1.209)
(1.210)
jE = i L RI[h m• hi] (1.211)

1m
(1.212)
For example, the current operator from just the hopping term for electrons is in one
dimension:
(1.213)
The other terms can become quite complicated, and will be introduced only as they are
needed.
There are several other comments. First, the operators P and RE are not defined in
infinite systems, since the integral over position will diverge. One can devise an alternate
definition
P = -i lim V jd3rp(r)e-iq.r (1.214)

q..... O q
(1.215)
The second comment concerns the energy current. As discussed extensively in Sec. 3.8, the
energy current is often not the current which describes thermal conductivity or thermoelectric
power. In metals, it is customary to use the heat current, which is defined as
(1.216)
is given in terms of operators jE and j, which have already been defined.

Another observation is that these operator definitions are the same for all particles,
regardless of whether they are bosons or fermions. Of course, spinless bosons do not have the
summation over spin index cr. Otherwise, everything is the same. However, calculations using
these operators depend significantly on whether the particles are bosons or fermions.
1.3. ELECTRON-PHONON INTERACTIONS
The first two sections described the Hamiltonians for phonons and electrons, respec-
tively. This section discusses their mutual interaction. This topic is important in many-body
theory. The electron-phonon interaction causes superconductivity in many metals and
influences the transport properties of every metal. In pure semiconducting and ionic solids,
the electron-phonon interaction usually dominates the transport properties. The word polaron
is used to describe a single electron which is coupled to phonons. The modern formulation of
the polaron problem was due to Frohlich et al. (1950), and its study is an important part of the
history of many-body theory.
Sec. 1.3 • Electron-Phonon Interactions 27
1.3.1. Interaction Hamiltonian

The basic Hamiltonian is assumed to have the form
H=Hp +He+Hei (1.217)
Hp = L roqA.a~A. aqA. (1.218)

qA.
(1.219)
Hei = L V(ri - R) (1.220)

ij
The atom part Hp describes the normal modes of vibration of the solid and is the phonon
Hamiltonian of Sec. 1.1. The second term is the electron part He' which was discussed in Sec.
1.2. The third part is the electron-ion interaction. It is assumed that Hei is the summation of
the interaction between the individual atoms at Rj and the electrons at r i : The word ion is not
meant to imply a particular charge state. In metals, the atoms are ions, while in covalently
bonded semiconductors they are something else. The word ion is used to encompass all these
possibilities. Each ion is at a position Rj = RjO) + Qj' which is the sum of the equilibrium
position RjO) and the displacement Qj. The displacements are usually small, so that one can
expand in powers of them:
The linear electron-phonon interaction term is obtained from the first term in Qj. The terms in
O«(f) are neglected here, although they are retained in some circumstances. The constant
term
L Vei(ri - R?» (1.221)

j
is the potential function for the electrons when the atoms are in their equilibrium positions,
which forms a periodic potential in a crystal. The solution of the Hamiltonian for electron
motion in this periodic potential gives the Bloch states of the solids. They are usually assumed
to be known. Many problems begin by writing the Hamiltonian as H = H o + V, where H o is
a Hamiltonian which can be solved and V is the perturbation. Quite often the eigenstates of
H o are just the Bloch states of the solid, calculated by assuming that the atoms are in their
equilibrium positions. The sequence of approximations we are making is called, collectively,
the Born-Oppenheimer approximation. As always, there are circumstances where these
approximations are inadequate, and other approaches are necessary, e.g., in the dynamical
Jahn-Teller effect.
The electron-phonon interaction is
Vep(r) = ~ Qj
}
• VVe;(r - R?» (1.222)
This interaction is to be written in terms of operators. It is assumed that the electron-atom

potential possesses a Fourier transform:
Vei(r) = ~L Vei(q)eiq'r (1.223)

q
i .
VVei(r) = AT L qVe;(q)e,q'r (1.224)
q
Evaluate the combination
(1.225)
The earlier definition (1.85) of Qj is used to show that
(1.226)
(1.227)
where the summation G is over the reciprocal lattice vectors of the solid. The phonon states of
q + G are defined only within the first Brillouin zone of the solid. Here the values of q may
be outside the zone, so that q + G projects them back into it. But the Fourier transform
Vei (q + G) is defined over all values of (q + G), not just the first Brillouin zone. Write the
interaction Hamiltonian in the form (MN = pv, p = density of solid in grams per cubic
centimeter)
1/2
" eir'(q+G)Ve,.(q + G)(q + G) . Sq ( -1'1-
Vep (r) = - L.. 2pVOl
)
(a q + a~ q)
~ q
The summation over q is restricted to be within the first Brillouin zone of the crystal. The
phonons are defined only in this space, so Olq' aq , Sq have only q labels. But the summation
over reciprocal lattice vectors G permits the potential Vei (q + G) to interact with higher
Fourier components. The notation also dropped the summation over the various phonon
modes A. Such a summation should be done, and we always do it, but writing A in every
subscript is a nuisance. So just do the summation over modes, and skip the writing of A in
every subscript!
The potential Veir) is defined as the unscreened electron-atom potential. Later it is
shown that electron--electron interactions in metals cause a significant reduction of this
potential, which is called screening. The potential Yep is sometimes calculated from first
principles, but more often it is obtained from a pseudopotential (Heine, 1970; Harrison,
1966). Modem calculations usually use force constant models to obtain accurate phonon
energies Olq' and polarizations Sq throughout the Brillouin zone, for each mode of polarization
and use them in calculating electron-phonon properties.
The potential Vep(r) acts upon the electrons and also upon other particles such as
positrons. The electron-phonon interaction is obtained by integrating this potential over the
charge density of the solid per):
1/2
L p(q + G)Ve;(q + G)(q + G)· ~q (a q + a~q)

( )
= - - Ii2
qG PVOlq
The particle density operator p( q) was defined in the prior section. Several examples of it
were given there. Quite often this matrix element is abbreviated by the symbol
1/2
M q+G = -Ve;(q + G)(q + G)' ~q (

- Ii2
PVOlq
)
(1.228)
(1.229)
1.3.2. Localized Electron

There is one problem which can be solved immediately: the electron-phonon Hamil-
tonian when the electrons are fixed in space at positions rio It is assumed that they cannot
recoil, which neglects the electron kinetic energy term. This model is often applied for
localized electrons in solids. A localized electron occurs in deep core states and in some
impurity levels. Then the Hamiltonian has the form
H=Hp +Hep (1.230)
= ~ [ Olq (a~aq + ~) + ~(aq + a~q)eiq.r,
x t:M(q+G)T](q + G)eiG.r,] (1.231)
The electron density operator T](q + G) is just the Fourier transform of the localized charge
density:
p(q + G) = Jd re3 ir .(q+G) ~ I<po(r - r i )1 2 (1.232)
=L eir,.(q+G)T](q + G) (1.233)
i
(1.234)
The various interaction tenns are collected into an effective matrix element:
(1.235)
(1.236)
The function Fq(r) is periodic in the lattice, since increasing it by a lattice vector a does not
change its value. It is assumed that all localized electrons in different unit cells are in the same
position within the cell. That is, Fq(r) is assumed to be the same for all localized electrons.
This problem has been solved before. It is just the hannonic oscillator in an electric field
[Eq. (1.38)]. Now each wave vector and polarization state q is a separate hannonic oscillator,
which finds its own equilibrium configuration. Follow exactly the steps used to solve (1.38),
and the creation and destruction operators are transfonned to the new set
Fq .
Aq = aq +- 2::e,q·r, (1.237)
co q i
(1.238)
The Hamiltonian with these operators is
(1.239)
Furthennore, the new operators still obey the hannonic oscillator commutation relations
[A q , A~,] = Oqq' (1.240)
[A q , Aq,] =0 (1.241)
[A~,A~,] = 0 (1.242)
It was shown in Sec. 1.1 that the eigenstates and eigenvalues for this Hamiltonian are
(At)nq
In) =-q-IO) (1.243)
q In;;T
(1.244)
The eigenstates are interpreted in the same way as used for the simple spring: Each nonnal
mode q has stretched to a new equilibrium configuration,
(1.245)
and now oscillates about this new equilibrium point. The oscillation frequencies do not
change. The last tenn in (1.244) is the relaxation energy. It is the potential energy gained by
stretching the springs-the phonon normal modes-to the new equilibrium positions. This
energy term may be expanded in terms of the electron coordinates. The relaxation energy is
(1.246)
(1.247)
VR ( r ) -- - 2'" IF q 12 ;q·r - 2 '"

L..,--e - - vL..,
J d q IF
3 12
q1c- e ;q·r
--3- (1.248)
q {Oq 1c (2n) (Oq1c
Note that the summation over phonon polarizations has been added to the last expression,
which is the integral one has to evaluate.
The relaxation energy consists of two types of terms. The first is VR(O), which is the
relaxation energy of a single particle by itself. This energy is caused by the electron inducing
a static polarization in the phonon field which acts back upon the electron. It is a self-energy
effect. It shall be called a polaron self-energy or electron self-energy. The energy is not just
with the electron. As with the stretched spring, it is in the combined particle-oscillator
system.
The other type of term is the interaction between pairs of fixed particles VR(r; - r) Here
the physical picture is that one particle polarizes the medium, and this polarization field
changes the energy of other particles which are nearby. This potential VR(r) has different r
dependence for different types of phonons in solids. In some cases, it is very short-ranged, so
that two particles interact only when they are in the same unit cell of the crystal. In other
cases, the potential falls as slowly such as r- i , as if it were a Coulomb potential. Several of
these cases will be presented later.
The same many-body problem is discussed again in Chapter 4. The harmonic oscillator
in a linear potential is an important model, if only because there are so few models which can
be solved exactly. It is also the solution to the independent boson model.
1.3.3. Deformation Potential

In semiconductors and ionic solids, the excited electron states are usually confined to a
small location in wave vector space. In thermal equilibrium, the excited states are at an energy
band minimum, which is often at the zone center or edge. Usually the polaron effects for these
electrons involve only phonons of long wavelength. In this case the tradition has been to
parameterize the interaction rather than compute it from first principles. Most electron-
phonon interactions in semiconductors use only three types of interactions: deformation
potential coupling to acoustical phonons, piezoelectric coupling to acoustical phonons, and
polar coupling to optical phonons. Another possible coupling is the deformation coupling to
optical phonons. These interactions are valid only at long wavelength. When the electron-
phonon matrix element is needed at short-wavelength phonons, the usual method is to
calculate them from pseudopotentials.
The deformation potential coupling to acoustical phonons is just the long-wavelength
limit of (1.229). Only the G = 0 is retained, since the terms G i= 0 are of wavelengths that are
too short. The electron-ion potential is taken to be a constant (Ve;(q) -+ D) at q -+ 0, where
D is the deformation constant. At long wavelength, ~ ---+ q, and only longitudinal phonons are
important if the band is nondegenerate. The interaction has the form
(1.249)
where the summation is only over longitudinal acoustic phonons. However, valence bands in
semiconductors are often degenerate at the band maximum. Then the hole excitations have a
deformation coupling to transverse phonons, which is a very large polaron correction (Mahan,
1965).
The deformation constants are obtained by measuring how energy bands shift with
increasing pressure on the solid. The value of Dn for a band n is simply the rate of change of
band energy with pressure (Thomas, 1961).
1.3.4. Piezoelectric Interaction

Many semiconductors are piezoelectric. The macroscopic effect is that an electric field is
generated when a crystal is squeezed and vice versa. Acoustical phonons, which are periodic
density modulations, make periodic electric fields. The crystal must lack an inversion center
to be piezoelectric. The group IV semiconductors Ge and Si are not piezoelectric. The III-V
semiconductors such as GaAs are very weakly piezoelectric, while the II-VI materials such as
CdS and ZnO are extremely piezoelectric. A very detailed derivation of the electron-phonon
interaction has been given elsewhere (Mahan, 1972). Here we shall provide only a quick
sketch. If Sij is the stress on the crystal, then the electric field is proportional to the stress,
(1.250)
where the matrix Mijk is a constant which gives the proportionality. The stress is defined as the
symmetric derivative of the displacement field:
(1.251)
(1.252)
The electric field may be shown to be longitudinal and to point in the direction q of the
phonon. It may be written as the gradient of a potential ¢(r)
(1.253)
The final result is the observation that the potential is proportional to the displacement:
¢(r) ex Q(r) (l.254)
(l.255)
The electron-phonon interaction for the piezoelectric interaction is
(1.256)
The matrix element M;,.(q) does not depend on the magnitude of q, but it very much depends
on its direction. It also has the property that M,,( -q) = -M,,(q) so that (1.249) is Hermitian.
In fact, the piezoelectric interaction is quite anisotropic. The matrix element is also very
dependent on the polarization A of the acoustical phonon whether it is LA (longitudinal
acoustic) or TA (transverse acoustic). Most many-body calculations have tended to take a
constant value for the matrix element, where this constant is obtained by averaging over the
various angular directions in the crystal. This approximation is adopted here.
An interesting result is obtained if (1.256) is used to calculate the effective potential
energy between two fixed electrons:
(1.257)
In a Debye model, the phonon energy is proportional to wave vector ro q" = c"q, where the
sound velocity c" depends upon the mode. Since M" and p are constants, the potential VR(r)
is the Fourier transform of q-2. This transform was worked out earlier, in (1.163), and just
gives a potential varying as r- i :
(1.258)
The answer is a form of Coulomb's law. Actually, it is a law of dielectric screening. In

dielectric materials it is known that the potential between two fixed charges is
(1.259)
where Etotai is the total, static, dielectric constant. In the piezoelectric calculation, the piezo-
electric contribution to this dielectric screening was derived. The total dielectric function is
the summation of many contributions:
(1.260)
Etotal = Eo + Epiezo (1.261)
The first term Eoo is from interband electronic transitions. The piezoelectric contribution is the
term under discussion. The polar and electron-electron contributions are explained later. All
other contributions are EO' All of these contributions are functions of q and ro. In the present
problem, with a static, fixed charge, these quantities are needed in the limit that q -+ 0 and
ro -+ O. The electron-electron term is not finite in this limit, so probably should not have been
included in the list (1.260). It may be omitted if there are no mobile charges present, which
shall be assumed.
If there were no piezoelectric contribution, presumably the interaction potential between

two fixed charges would be e2 /(sor). The piezoelectric contribution (1.258) represents the
difference between the above result and total screening (1.259):
St~lr = ~ (:0 - Y) (1.262)
so that
1
----=--Y (1.263)
So + spiezo So
These equations can be solved to obtain the piezoelectric contribution to the static dielectric
function:
s· =---
S5Y (1.264)
plezo 1- soY
The interaction tenn (1.258) is negative because the screening lowers the potential energy.
The unscreened potential is just e2 /r, and each bit of screening lowers it by an amount
proportional to e2 /r.
The same acoustical phonon may interact with an electron by both the defonnation and
piezoelectric interactions. These two interactions do not interfere, to second order, because
they are out of phase. The sum of the two interactions (1.249) and (1.256) gives
(1.265)
(1.266)
The defonnation potential is real, while the piezoelectric is imaginary. To second order, they
do not interfere:
(1.267)
Since electron-phonon effects are usually evaluated only to second order, these two inter-
actions can be treated separately.
1.3.5. Polar Coupling

The polar coupling between electrons and optical phonons can be very large in ionic
crystals. The fonn of the Hamiltonian has been derived often, e.g., Frohlich (1954). In ionic
crystals some of the atoms are positively charged, while others are negatively charged. An
optical phonon has the different ions in the crystal vibrating out of phase. When the plus ions
and minus ions oscillate in the opposite direction, they set up a dipole polarization field. The
polarization causes a long-range electric field which scatters the electrons. The electric field is
the source of the polar coupling.
The polar coupling is only to LO (longitudinal optical) phonons and not to TO (trans-
verse optical) phonons (Mahan, 1972), because only the LO phonons set up strong electric
fields when they vibrate. These electric fields are in the direction of vibration, which at long
wavelength is in the direction of the phonon wave vector q. For a system of no free charges,
v . D = 0 = :L q . (E q + 41tPq)eiq .r (1.268)
q
For an LO phonon mode of wave vector q, the electric field Eq and polarization P q are both
parallel to q. The formula for the electric field produced by the polarization is:
Eq = -41tPq (1.269)
The next assumption is that the polarization is proportional to the displacement

Pq = eUQq (1.270)
(1.271)
where the coefficient U is to be determined. The phonon energy row is assumed to be

constant. Since the electric field points in the direction of q, it may be expressed by a
potential:
E = -Vq, = -i:Leiq.rqq,q (1.272)
q
,h( ) _ " iq.r 41teU

'P r - ~
q
e
q
(_h_)
2proq v
1/2 A(
q aq + a_t q ) (1.273)
which gives the potential produced by the LO phonons. The interaction constant U still needs
to be determined. Its value is obtained by considering the potential between two fixed elec-
trons that was calculated from our relaxation energy (1.247):
VR(r) = 2
---(41teU) 2 ( -h-) Jd q e3 iq r
--3-2 . (1.274)
hrow 2proq (21t) q
The Fourier transform integral again produces a Coulomb potential:

e2
VR(r) = -r- (1.275)
r
41tU2
r=-2- (1.276)
prow
This interaction energy is again interpreted as a contribution to the dielectric screening of the
solid. This term represents the contribution from the optical phonons. It represents the
difference between screening with just the electronic interband part Eoo and the interband plus
optical phonons Eo:
(1.277)
(1.278)
The unknown factor U is
(1.279)
It is possible to write the electron-phonon Hamiltonian in the form
(1.280)
(1.281)
where the matrix element M is given in the second line as M2. This form is a very sensible
way in which to express the matrix element. The dielectric constants Eo and Eoo are both
measurable: Eo is the low-frequency dielectric function measured by putting the solid between
the parallel plates of a capacitor at low frequency, while Eoo is the square of the refractive
index.
Besides these quantities, the matrix element depends only on the charge e and the LO
phonon energy Ilrow. In spite of the elegance of this simple form, the matrix element is not
usually expressed this way. Instead, it is customary to introduce the dimensionless polaron
constant cr, defined as
(1.282)
Correspondingly, the interaction matrix element is
(1.283)
The reason for introducing cr is simple. The self-energy expressions will have a simple and
elegant form with this choice of cr.
Electrons in crystalline energy bands have their motion determined by an effective band
mass mh' The band mass enters into the definition of cr, which has the obvious disadvantage
that one does not know the value of cr until one knows the value of mh'
1.4. SPIN HAMILTONIANS
The study of spin systems forms a very large part of many-body theory. There are many
solids which display magnetic ordering among the electrons. There are many impurity
problems where spin plays an important role. In attempting to explain these phenomena,
many different types of spin models have been introduced. Some of these involve localized
spins interacting among themselves, while others have localized spins interacting with free
electrons. It is a subject in which there are few exactly solvable models which are nontrivial.
Only a few of the models have solutions which are well understood, in spite of the fact that
many of them have been intensely studied. The transference of model results to real solids
which are strongly interacting systems has still not been very successful.
Sec. 1.4 • Spin Hamiltonians 37
The commutation relations for spin one-half operators are (Ii = 1)
[sy) , S?)] = is?)()lj
[SI(y) 'S(z)]
j
- z·S(x)s=
- j ulj
(1.284)
The subscript labell or j denotes spin site. Spins on different sites, or with different electrons,
commute. The superscripts (x), (Y), (z) refer to space coordinates. These spin operators are
often represented by Pauli spin matrices,
Sex) = ~ (0
2 1 :)
s(y) = ~ (0 ~i)
2 i
~C
S(z) =
2 0 ~1) (1.285)
but these representations are unnecessary for many-body calculations-the commutation

relations are sufficient. It is customary to introduce spin raising and lowering operators:
S(+) = s~x)
J J
+ iS~)
J
(1.286)
S~-) = s~x) - iS~) (1.287)

J J J
The names raising and lowering are applied to these operators because they raise or lower the
magnetic quantum number m of the spin state. For spin one-half, there is only spin up I+} and
spin down I-}, and the operators go from one to the other:
S(+)I-} = I+}, SHI+} = I-}

S(+)I+} = 0, SHI-} = 0 (1.288)
By direct mUltiplication, for two operators on the same site:

S(+)SH = S(x)2 + S(y)2 _ i[S(x)S(Y) _ S(Y)S(x)] = S(x)2 + S(y)2 + S(z)
SHS(+) = S(x)2 + S(y)2 + i[S(x)S(Y) _ S(Y)S(x)] = S(x)2 + S(y)2 _ S(z) (1.289)
By subtracting these two results, the commutation relations are found among the raising and
lowering operators:
[S~+), st)] = 2S?)()lj
[S?), sJ+)] = sJ+) ()lj
[SI(z) 'SH]
j
-- _SH"
j ulj (1.290)
The last two commutators can be easily obtained from the direct definition of the operators;
for example,
[ !/;z) s(+)] = [S~z) !/;x)] + i[S~z) S~)] = iS~) _ i2!/;x) = S~+)

J'J J'J J'J J J J
Spins are neither bosons nor fermions. These commutation relations are unlike any which we
have encountered previously. It is precisely these commutation relations which make spin
problems so difficult.
1.4.1. Homogeneous Spin Systems

The Heisenberg Hamiltonian puts one spin on each site of a lattice and has the spins
interact with a vector interaction. If the interaction is only between nearest neighbor spins, the
Hamiltonian has the form
H - -J" S S - -J ,,[s(x) sex) + s(y) s(y) + S(z) s(z) ] (1.291)
- ~j'J+O- ~jj+o jJ+o jJ+O
jO jcr
Again the notation j + 0 means the neighbors of site j. The Heisenberg Hamiltonian is often
solved for spin greater than one-half or for coupling between spins which may be further
neighbors. But the above form is the most common. The word solve means "approximately
solve," since it cannot be solved exactly, except in one dimension. Often the coupling constant
in one direction, say z, is taken to be different from those in the other directions. This case is
the anisotropic Heisenberg model:
H - -J,
- II
"S(z) S(z) -
~ j j+O
J
1.
"[S(x) sex)
~ j j+O
+ s(y)
j
s(y) ]
j+O
jO jO
H - -J, "s(z)s(z) - J "S(+)S(-) (1.292)

- II ~ j J+O 1. ~ j j+O
jo jO
In this case the operators SJ+) and Sj~~ can be arranged in any order. They commute
since they refer to different sites. There are two limiting cases of this Hamiltonian which have
their own names. The ISing model has J1. = 0
- J , " S(z) rdZ)
H I--II~ (1.293)
j ;)/+0
jO
It may be solved exactly in one dimension, even if one adds a magnetic field to the
Hamiltonian:
(1.294)
where Ho is the magnetic field in units of joules. In two dimensions, it may be solved exactly
without the magnetic field, as shown by Onsager (1944). Very accurate three-dimensional
results have been obtained by a variety of techniques, including Green's functions (Callen,
1966) and critical point and renormalization group techniques (Domb and Green, 1972-
1977). The XY model has only h:
(1.295)
It may be solved exactly only in one dimension.

It is conventional to write the Hamiltonian in (1.291) with the negative sign in front of
the interaction term. Then for J > 0, the spins tend to line up all parallel, which is the
ferromagnetic arrangement. For J < 0, the ordering has alternate spins up and down, if the
lattice permits, which is called antiferromagnetic.
The difficulty with solving spin problems is well illustrated by defining collective
operators. The operators are transformed into wave vector space:
s(+) = _1_ '""' eik .Rj S(+) s(+) = _1_ '""' e-ik-R S(+)
j
(1.296)
k .jN7 J' } .jN~ k
S (-) - _1_,"", -ikRS(-) S(-) - _1_ '""' ikRS(-) (1.297)

k - .jN7 e 1 i ' i - .jN~e 1 k
This transformation appears to be a reasonable approach. It was used successfully in solving

fermion and boson problems. But examine the commutation relations for these operators:
[S(+) S(-)] = ~ '""' eik-R,-ik'.Rj[S(+) S(-)]

k ' k' NL.. f ' }
if
[S(+) S(+)] = ~ '""' eiRj"(k-k')S(Z) (1.298)

k ' k' NL.. }
}
The operators sj+) and SJ-)

commute except on the same site, and then their commutator in
(1.290) is 2Sjz). The right-hand side of (1.298) is not simple. It would be preferable to find
something like
(1.299)
which would indicate that S~+) and st)
were independent operators except for k = k'. They
would behave just like bosons, and boson statistics would be used. Unfortunately, (1.298)
does not have this property. One common approximation (Callen, 1966) is to replace (1.298)
by the approximate expression
[S~+), S~;-)] ~ 20 k ,k' ~Lsf) = 20k .k,(S(z») (1.300)

}
One then tries to find a self-consistent equation for the average magnetization (S(z»), but this
approach is approximate. Except in special circumstances, the operators S~+) and S~;-) do not
describe independent eigenstates of the system. This lack of collective eigenstates is the
difficulty with solving spin systems and why innocent-looking Hamiltonians like (1.295) are
difficult to solve. Hamiltonians with bilinear boson or fermion operators may be solved
exactly. That is not true for Hamiltonians with bilinear spin operators.
If the two equations (1.289) are added together, then
sj+) sj-) + sj-) sj+) = 2[Sjx)2 + Sy)2] = 2[S . S - S;Z)2] (1.301)
= 2[S(S + 1) - S;Z)2] (1.302)

For spin one-half, SjZ)2 = 1/4, which gives the result
{sj+), sj-)} = 2[!@ -~] = 1 (1.303)
The S(+) and S(-) operators, on the same site, obey an anticommutation relation. Anti-
commutation relations are associated with fermion operators. The spin one-half operators
behave as a fermion on the same site. For fermions, each site may have zero or one particle for
each orbital and spin state. These two possibilities correspond to the two possible spin states
of up and down. For example, spin down is equivalent to zero particles and spin up to one
particle on the site. Equation (1.288) is analogous to the fermion relations
Cjll} = 10}, CJIO} = II}
c!ll}
]
= 0 (1.304)
Unfortunately, one cannot carry this analogy too far. On different sites, the spin operators
commute, while the fermion operators anticommute. The spin operators are not fermions
either.
However, in one dimension, the spin one-half operators can be made into exact fermion
operators. This transformation was discovered by Jordan and Wigner (1928). In one
dimension, the spins are aligned along a chain. A new set of operators is defined: the old
raising and lowering operators are multiplied by a phase factor which is dependent on spin
site:
d!d.
JJ
= S~+)S~-)
J J
= S(S + 1) - S(z)2
J
+ sf?)
J
= + S(z)
J
1
-2 (1.305)
The phase factor <Pj is chosen to be 1t times an operator which measures the number of spin-
up operators to the right of that position:
(1.306)
The chain is numbered from one end, say the right, with site indices j = 1,2,3, .... Set
<PI = O. The d t operators are interpreted as creating fermion particles, and dJ dj = nj is the
number operator for each site. The phase <Pj is 1t times the number of such fermions to the
right of the site). The phase factor commutes with sj-),
since the operator <Pj is the number to
the right and does not involve the number operator on the same site. On the same site, these
operators anti commute,
(1.307)
since that is the property of the spin operators themselves. On different sites, they also
anticommute, with the help of this new phase factor. By taking the anti commutator,
(1.308)
Since I =f. m, assume that I> m. The phase factor <Pm then commutes with S~-), but <PI does
not commute with S~+). The anticommutator is then
(1.309)
The right-hand term contains the operator combination

e- ill (I/2+S(Z»S(+)e ill (I/2+S(Z» = e iIl S(+) = -S(+) (1.310)
The S(+) operator must always raise the magnetic quantum number m by unity, so that S(z) on
the left always measures one integer higher value than the same operator on the right of S(+).
One gets an extra phase factor of in, which changes the sign of the term. This phase factor <PI
was chosen to produce this sign change. The anticommutator of the d operators is now
(1.311)
It is equal to the commutator of the S(+), SH operators, for different sites, which is zero. The
d operators anti commute for both the same site and different sites. They are pure fermion
operators and obey fermion statistics. In one dimension the XY model (1.295) may be
transformed into a Hamiltonian in terms of the d operators,
Hxy = -J L(sj+) S}~? + sj+) ~c.=?) (1.312)

}
= -JL(ei<!>Jd]e-i<!>j+l~+1 +ei<!>Jd]e-i<!>J-1d}_I) (1.313)

}
Hxy = -JL(~te-i1tnJdJ+l +~te-i1tnJ-ld}_I) (1.314)

}
Hxy = -JL(d]~+1 +~td}_l) (1.315)

}
which is just the tight-binding model for fermions. The phase factors vanish because in the
first term n} is zero if it precedes a raising operator, and in the second term it is zero if it
follows a lowering operator. Next the Hamiltonian is changed to collective coordinates:
= ,JNye
1
d
j = ,JNTe
1 " -ikRd
k,J dk
ikR
Jd}
d t = _1_" eikRJdt 1_" -ikRdt

dkt -__wL..e
} ,JNT k' '\IN}
J.
}
H = -2JLYkdtdk, Yk = cos(ka) (1.316)

k
and the collective operators now obey anticommutation relations as well:
- ~ "ei(kRJ-k'RI){d dt } _ ~ "eiR/k-k') _
{dk, dk't} -NL.. s: (1.317)
j' I -NL.. - Ukk'
fl }
The Hamiltonian may be written as a simple fermion problem. The exact partition function is
(1.318)
Note that there is no chemical potential for spin systems. One can also work out other
properties of this model. To prove that the transformation (1.305) is valid, one should show
that all the commutation relations (1.290) are preserved. They are.
The Jordan-Wigner transformation shows that in one dimension the spin one-half
operators may be represented exactly as fermions. This result is not valid for higher
dimensions. No one has been able to find an equivalent transformation for two or three
dimensions. Indeed, most approximate analyses assume that in two or three dimensions the
spin excitations behave approximately as bosons rather than as fermions.
In two and three dimensions, one may still transform spin one-half operators into particle
operators. In this case, the particle operators have funny commutation relations. They are, like
spins, neither fermions nor bosons. This transformation associates the creation operator with
S(+) and a destruction operator with S(-):
CIt -- S(+)
I
C1-- S(-)
I (1.319)
n1-- ctc - s(+)s(-)

I 1- I I
- ! + s(z)
-2 I (1.320)
This transformation preserves all the commutation relations (1.290). For example, one has
that
{CI , Cl} = 1 (1.321)
[CI, cl] = -2(! - nl)Olm (1.322)
[CI , nIl = [S}-) , s}z)] = CI (1.323)
The particle operators CI and cl anticommute if they are on the same site and commute if
they are on different sites. They are neither fermions nor bosons. Collective operators such as
- 1 '" I'k.r·C
Ck--~e } . (1.324)
v'Nj J
have funny commutation relations, similar to the spin case in (1.298). Nevertheless, this
Hamiltonian is a popular many-body model for certain systems. These are lattice gas models
for atoms on lattices. The atoms may be considered as classical particles, which commute on
different sites. However, there may not be more than one atom on each site, since the atoms
are large, substantial objects. "No more than one atom on each site" is an exclusion principle,
which is represented by the anti commutation relations on the same site. The same physics is
contained in a model which has the particle obey purely boson statistics but with the provision
that there is a strong repulsive interaction U if two particles were on the same atomic site. The
lattice gas results would be obtained in the limit U -+ 00. The statistics of anticommutation
relations on the same site merely represent the strong repulsive interaction between atoms at
close separation.
The usual lattice gas (LG) model has pairwise interactions U between particles in
nearest neighbor positions. A chemical potential J.l is also introduced for the particles, since
they may have variable concentration:
KLG = HLG - J!N = ! U L njnj+o - J.l L nj (1.325)

jo j
The factor of one-half is due to counting each pair twice in the summation over all (jS). The
lattice gas Hamiltonian may be transformed into an equivalent magnetic problem by inverting
the transformation of (1.319):
K
LG
= ~2 U"
~
JU
(s(Z) + ~) (s(Z) + ~) _
J 2 J+& 2
I"
(S(Z) + ~)
... L.. }
J
2 (1.326)
K = ~2 U" S(z)S(z)
LG } ~
+ (UZ
2
_II)"
J+O...
S(z) _ ~N, (UZ _II)
2 4
L.....J 0
(1.327)
JU J
where Z is the coordination number, i.e., the number of nearest neighbors. The spin version
of KLG is identical to the Ising model with magnetic field. The magnetic field is
Ho = U Z /2 - Il. If exactly one-half of the sites of the lattice gas are occupied, then one has
Il = U Z /2 so that the effective magnetic field is zero (Hill, 1956). In this case, for a half-
filled band, the chemical potential is temperature independent. For concentrations other than
one-half, the chemical potential varies with temperature.
The quantum lattice gas (QLG) model adds a nearest neighbor hopping term to the
lattice gas model:
(1.328)
One can show that the equivalent magnetic problem is the anisotropic Heisenberg model with
magnetic field. The quantum lattice gas was suggested by Matsubara and Matsuda (1956) as a
model for quantum fluids such as 4 He. The superfiuid transformation for this system occurs in
the liquid state. Nevertheless, a quantum lattice gas model appears to be a good description of
its critical properties. The parameter U may be taken to be either positive or negative,
depending on whether the nearest neighbors repulse or attract each other. The equivalent
magnetic problems are then antiferromagnetic and ferromagnetic, respectively.
1.4.2. Impurity Spin Models

The models mentioned so far are for homogeneous magnetic systems. The same spin
was on each site, and we tried to deduce the magnetic properties of the entire system. Other
kinds of popular models are for impurity spin problems. Here the spin is an isolated impurity
in an otherwise homogeneous electron gas. One can study, for example, the conditions for the
formation of a local moment on the impurity or the scattering properties of the free electrons
from the localized spin.
The derivation follows Kondo (1969). An impurity atom is located at the position Rn and
has a localized electron orbital <PL(r - Rn) when the electron is on that site. Otherwise the
electron is in a continuum state k with wave function <Pk(r) and energy ck- The wave functions
may be considered as plane waves or alternately as Bloch functions of the crystal. A
generalized state function is the summation over all possible states,
\jJ(r) = L <Pk(r)XcrCkcr (1.329)

kcr
where the Xcr are the spin wave functions, which denote spin up X t or down X.j,.. The wave
functions <Pk(r) and <PL(r - Rn) are not assumed to be orthogonal. This step will be taken
later, since their orthogonality is a many-body problem of sorts. The Hamiltonian is taken to
have terms such as
(1.330)
where U(r) is the potential for each electron, which may include the impurity potential as well
as the usual potential of the host lattice. The last term is the electron--electron interactions.
The first terms of the Hamiltonian are evaluated for the state function, and one gets
Jd3r\)tt(r)[~~ + U(r;)]\)t(r) = t;EkCZ"Ckcr + ~ELClcrCLcr

+ L Ukk,C:'crCkcr
kk'cr
(1.331)
The first two terms on the right are the unperturbed parts of the Hamiltonian for the con-
tinuum states and the impurity. The impurity states are labeled (La). The third term is
(1.332)
which involves the scattering of the continuum functions from the impurity potential. This
problem is simple to solve by ordinary scattering theory. The solution will be given in Sec.
4.1. The last term is a mixing term called H M :
(1.333)
The interaction HM is sometimes called Hsd but that name is used for another contribution
which will be discussed shortly. The mixing term describes processes whereby the electron
hops off of the impurity and becomes a continuum state or vice versa. This term essentially
arises from the nonorthogonality of the continuum and local wave functions. Note that the
spin direction is maintained during this hybridization process. A Hamiltonian of the type
H = L EkCtcrCkcr +L EL clcrcLcr + L Mk(CtcrCLcr + clcrckcr) (1.334)

kcr cr kcr
will be called a Fano-Anderson model, since it was introduced simultaneously by Fano
(1961) and Anderson (1961). It should not be confused with the famous Anderson model,
which will be described below. The Fano-Anderson model may be solved exactly, and this
solution will be presented in Sec. 4.2. There is no real conceptual difference between (1.331)
and (1.334). If the wave functions \)tk(r) are chosen to be eigenstates of the Hamiltonian
which includes the impurity scattering Uk, then the two Hamiltonians become identical.
The interesting magnetic phenomenon comes from the terms involving electron--electron
interactions. Consider
(1.335)
Many types of terms are generated by this expression. Only two of them are discussed here.
The first contains operators from two continuum functions and two localized wave functions:
(1.336)
The four spin operators must occur in pairs. One possible pairing (0' = cr', a" = 0'111) gives
the interaction
L Vkk,CL,.cLcrctcr'Ck,cr' (1.337)
kk'acr'
(1.338)
This term is usually ignored, since it does not cause magnetic phenomena. It states that the
conduction electrons interact with the impurity in a different way when the localized orbital is
occupied. It is quite a reasonable term, since the electron is charged and one expects that the
presence of this charge will influence the other electrons. This term is customarily ignored.
Another possible spin arrangement is the exchange term for the above process, which has the
form
J kk , J
= eZ d3rl <!>~(rl)(h(rl - Rn) Jrld-3rzrzl <!>1(rz - Rn)<!>k,(r
I Z) (1.339)
In this term, the electron which is scattering may change its spin state during the scattering
process. This spin change is always accompanied by an opposite spin change of the impurity
spins, so that the total spin angular momentum is conserved during the process. The spin
conservation is illustrated by writing out in detail the terms which can occur:
The first and last terms are regrouped as
!(CttCk't + ct-l- Ck'-I-)(Cl t CLt + CIt CL-I-)
+!(CttCk't - ct-l- Ck'-I-)(Cl t CLt - CItCL-I-) (1.340)
The first term in (1.340) contains the factor
cltCLt + CIt CL-I- =L clcr c Lcr (1.341)

cr
which is always unity if the localized state is occupied. In fact this first term has the form
1 t t
-2 L Jkk,CLcrCLcrCkcr'Ck'cr' (1.342)
kk' aa'
which is exactly the same as (1.337). The usual convention is to combine these terms, and
then ignore both of them. The second term in (1.340) has the combination
t C
SL(z) -- 2"1 (cLt t
Lt - CL-I- CL-I- ) (1.343)
which is just the z component of the localized spin. Similarly, some of the other combinations
are identified as the raising and lowering operators for the localized spin:
SL(+) = ctLt CL,!. (1.344)
SL(-) = ctL,!. CLt (1.345)
The spin-dependent terms in (1.339) may be collected as
Hsd = - LJkk,[SY\CttCkt - ct,!.Ck,!.) +si+)cLCk't +st)cttCk,,!.]

kk'
This interaction is called Hsd ' It is a model of the localized d-electrons interacting with the s-
like continuum wave functions. This form of the interaction is valid even if the localized spin
has S > !. The last two terms flip the spin of the continuum electron while flipping the
localized spin of the impurity in the opposite direction. The first term does not flip spin, but
the interaction does depend on the z-component of the spin of both the impurity and con-
tinuum wave function. A Hamiltonian of the type
(1.346)
is called the Kondo problem. It was not formulated by Kondo (1964) but rather much earlier
by Zener (1951). Kondo's contribution was recognizing that the spin-flip scattering processes
could cause unusual low-temperature behavior in the scattering properties. These low-
temperature anomalies had long been observed in resistivities (Gerritsen and Linde, 1951,
1954). They are derived in Chapter 6. Another type of term which may arise from the
electron--electron interaction (1.335) has four local operators:
(1.347)
(1.348)
If the orbital is nondegenerate, as in an s-orbital, then two electrons can be on the same site
only in opposite spin states. The spin indices must pair up as (crill = cr, cr" = cr'). One gets a
term similar to the one in the Hubbard model,
(1.349)
although it was historically introduced first in the Anderson model. This famous model
considers the model Hamiltonian
+ LMk(CtcrCLcr + clcr c kcr) + UnLtnL,!. (1.350)

kcr
The Hamiltonian is exactly solvable without the last term. The last term is very important,
since it causes magnetic instabilities in some circumstances. The Hamiltonian describes the
interaction of the conduction electrons with a single impurity at site Rn. The index L denotes
the orbital state of the impurity. If there is only one important orbital state, as is the case for s-
states, then the indices are often dropped from the notation.
The Anderson model and the Kondo model both describe the interaction of a continuum
electron with a localized one. The two models are not totally different. There is a canonical
transformation which, when applied to the Anderson model, will transform it into a form
similar to the Kondo model. This transformation on the Anderson model produces quite a few
terms, of which the Kondo model is a subset. The transformation does not produce exactly
the Kondo model, and the two models are not identical. Write the Anderson model as
(CLs == Cs' ns == c1 Cs)
H=Ho+HM (1.351)
Ho =L CkCZcrCkcr + cL L clcrcLcr + UnLtnU, (1.352)

kcr cr
HM = LMk(CZcrCLcr + clcrckcr ) (1.353)

kcr
then a canonical transformation (Schrieffer and Wolff, 1966) is
H
A
= e-S He ~
= Ho + HM + [S, Ho] + [S, HM] + 2[S,
1
[S, Ho]]
+ HS, [S, HM ]] + ... (1.354)
The function S is chosen so that all terms are eliminated which are linear in M k . It is
accomplished by choosing S to be
0= HM + [S,Ho] (1.355)
and then the canonical transformation produces the series
(1.356)
There are an infinite number of terms in the series. Only those which are proportional to M~
will be evaluated here, which is the second term in the series (1.356). The transformation
factor S turns out to be
(1.357)
so that one gets the commutators
When these are added, one exactly satisfies (1.355). Although the form of S appears
unwieldly, it gets the job done. The next step is to take the commutator [S, H M] to generate the
terms which are proportional to M~. As a preliminary step, define the following effective
exchange constant:
In terms of this constant, one finds the following for the commutator:
[S, HM ] = L Jkk,[n-sCLCk,s + CZ'sCk-sC!sCs - 0kk,nsn_s

kk's
(1.358)
The second term flips the spin of the electron and impurity electron while scattering-it has
exactly the Kondo form. There are many other terms. The last term is potential scattering of
the continuum electron, which must be added to our earlier result (1.331). There is a term
which is the non-spin-flip interaction between a continuum electron and an impurity electron
and this must be added to (1.337). There is also a term which contains the interaction between
two impurity electrons, one with spin up and the other with spin down. This renormalizes U
in (1.349). Finally, there is a new term which has not been previously encountered. It
describes the process
(1.359)
whereby two impurity electrons hop off of the impurity site to become two continuum
electrons and vice versa. The Anderson model certainly describes a rich set of phenomena. Of
course, many more terms are generated by the additional commutators in the series (1.356).
1.5. PHOTONS
Throughout this book the following type of Hamiltonian will be used for the discussion
of the interaction of charges ej with each other and with a radiation field. The raising (ah)
and lowering (akA) operators are for the photons:
(1.360)
The radiation field is represented by the vector potential
(1.361)
(1.362)
Sec. 1.5 • Photons 49
The term L ffikat .. akA represents its unperturbed photon Hamiltonian in the absence of
charges. The unit polarization vector is ~11. One feature of this Hamiltonian is the term
~L ejej (1.363)
2 j #j Irj - r)
which is the Coulomb interaction between charges. The Coulomb interaction is instantaneous
in time, since the potential has no retardation, or speed of light, built into it. The lack of
retardation is not an approximation but is rigorously correct in the Coulomb gauge. Of course,
there is retardation in the total interaction, which arises through the vector potential fields.
1.5.1. Gauges
Although the topic of gauges is treated correctly in a number of texts, it still seems to be
poorly understood by students. It seems appropriate to start at the beginning and reproduce
some standard material. Maxwell equations are
V·B=O (1.364)
1 aB
VxE=--- (1.365)
cat
V· E = 4np (1.366)
1 aE 4n.
VxB=--+-J (1.367)
cat c
The charge density is p(r, t) and the current density is j(r, t). There is an important theorem
that any vector function of position can be written as the sum of two terms: one is the gradient
of a potential, and the other is the curl of a vector:
S(r) = Vg + V x m(r) = Sf + St (1.368)
Sf=Vg (1.369)
St=Vxm (1.370)
The term Sf is called the longitudinal part of S, and St is called the transverse part. IfB(r) is
assumed to have this form, then Eq. (1.364) becomes
V· (Vg + V x A) = V 2g = 0 (1.371)
Usually g = 0, so that the vector potential is defined as
B=VxA (1.372)
However, this expression does not uniquely define A(r). The definition B = V x A is put into
(1.365):
VX[E+~~A]=O
cat
(1.373)
Now the factor in brackets is also the sum of a longitudinal and transverse part:
V x [Vw + V x M] = 0 = V x (V x M) (1.374)
The equation is satisfied if M = 0, so that for the electric field is

laA
E = -~at - Vw (1.375)
where Wis the scalar potential. When these two forms for B(r) and E(r) are put into (1.366)
and (1.367), the equations for the scalar and vector potentials are:
2 1a
v W+ ~ at v . A = -4rrp (1.376)
1 a2 a 4rr.
V x (V x A) + c2 at2 A + V cat W= ~J (1.377)
At first it appears that there are four equations for the four unknowns (Ax, Ay , A z, w): Eq.
(1.376) and the three vector components of (1.377). If this assertion were true, the four
unknown functions would be determined uniquely in terms of the sources (j, p). However,
these four equations are not linearly independent; only three of them are independent. The
linear dependence is shown by operating on (1.376) by (ljc)(ajat) and on (1.377) by V. Then
subtract the two equations and find
V. [V x (V x A)] = _ 4rr
c
(ap+ V.
at
j) (1.378)
The left-hand side is zero because it is the gradient of a curl. The right-hand side vanishes
since it is the equation of continuity. The four equations are not independent. There are only
three equations for the four unknowns.
Therefore, the four unknown functions (Ax, Ay ' Az , W) are not uniquely determined. It is
necessary to stipulate one additional condition, or constraint, on their values. It is called the
gauge condition. The imposed condition is that the Coulomb field w(r, t) shall act instan-
taneously, which is accomplished by insisting that
(1.379)
Equation (1.379) defines the Coulomb gauge, sometimes called the transverse gauge. The
latter name arises because (1.379) implies that V . Al = 0, so that A is purely transverse. One
should realize that any arbitrary constraint may be imposed as long as one can satisfy (1.376)
and (1.377). As long as these two equations are satisfied, one always obtains the same value
for E(r) and B(r). The arbitrary choice of gauge does not alter the final value of observable
quantities.
In the Coulomb gauge Eq. (1.376) simplifies to
V2W = -4rrp (1.380)
which is easily solved to give
w(r, t) = Jd3 r' p(r', t) (1.381)

Ir-r'l
The potential w(r, t) is instantaneous and is not retarded. This result is not an approximation
but is an exact consequence for our choice of gauge. Later it is shown that a different choice
of gauge leads to a retarded scalar potential.
Next the other equation (1.377) is evaluated. The following identity is useful:
V x (V x A) = -V2A+ V(V· A) (1.382)
The second term vanishes in the Coulomb gauge, which gives
V2 A _ ~ ~ A = _ 41t j + V a\jJ (1.383)

c2 at2 c cat
It is useful to operate a bit on the second term on the right-hand side. Using (1.381), this term
is
cat c
J
V a\jJ =! V d3 r' (ajat)p(r', t)
Ir - r'1
(1.384)
V a\jJ = -!VJd3r'_I_V'. j(r', t) (1.385)

cat c Ir - r'1
where the last identity uses the equation of continuity. Integrate by parts in the last term,
V a\jJa = !VJd3r'j(r', t)· V'-I_1_1 (1.386)

ct c r-r'
and then pull the gradient out by letting it operate on r instead of r'; the latter step requires a
sign change. Also operate on the current term itself in (1.383) by using the identity
41tj(r, t) = _V2 Jd 3r' j(r', t) (1.387)

Ir-r'l
By combining these results
V2 A _ ~~A = (V2 _ VV)!Jd3r' j(r', t) (1.388)

c2 at2 c Ir - r'l
Finally, using the identity (1.382),
V2A-~~A=-!Vx [V x Jd3r' j(r',t)] (1.389)

2 c2 at c Ir - r'1
The point of this exercise is that the right-hand side of (1.389) is now a transverse vector,
since it is the curl of something. The current is written as a longitudinal plus a transverse part,
j = jl + jt (1.390)
Then the vector potential obeys the equation
2 1 az 41t .
V A - c2 at2 A = -~JtCr, t) (1.391)
where Mr, t) is defined by the right-hand side of (1.389). The final equation for A is very
reasonable. Since the vector potential A is purely transverse, it should respond only to the
transverse part of the current. If it were to respond to the longitudinal part of the current, it
would develop a longitudinal part. The longitudinal component of A does not occur in the
Coulomb gauge.
As a simple example, consider a current of the form
(1.392)
whose transverse part is
Jt. = - k12 k X
(k X
J)
0 e
ik·r (1.393)
FIGURE 1.2
and jt is the component of j which is perpendicular to k; see Fig. 1.2. In free space, the
transverse and longitudinal parts are just those components which are perpendicular and
parallel to the wave vector k of the photon. However, solids are not homogeneous but
periodic. Along major symmetry directions in the crystal, it is often true that transverse
components are perpendicular to the wave vector. However, it is generally not true for
arbitrary points in the Brillouin zone, even in cubic crystals. The words transverse and
longitudinal do not necessarily mean perpendicular and parallel to k.
Two charges interact by the sum of the two interactions: Coulomb plus photon. The net
interaction may not have a component which is instantaneous. In fact, for a frequency-
dependent charge density, at distances large compared to c/ ro, one finds that the photon part
of the interaction produces a term -e'2/r which exactly cancels the instantaneous Coulomb
interaction. The remaining parts of the photon contribution are the net retarded interaction.
Solid state physics is usually concerned with interaction over short distances. Then retardation
is unimportant for most problems. In the study of the homogeneous electron gas, for example,
the photon part is small and may be neglected. In real solids, the photon part causes some
crystal field effects, which is an unexciting many-body effect. The main effect of retardation is
the polariton effects at long wavelength (Hopfield, 1958). In general, the Coulomb gauge is
chosen because the instantaneous Coulomb interaction is usually a large term which forms a
central part of the analysis, while the photon parts are usually secondary. Like most gener-
alizations, this one has its exceptions.
The Coulomb gauge V . A = 0 is not the only gauge condition popular in physics. Many
physicists use the Lorentz gauge:
V.A+O"'=O (1.394)
cot
which causes (1.376) and (1.377) to have the forms
2 1 a2 (1.395)
V '" - c2 ot2 '" = -4np
2 1 a2 4n.
V A- c2 ot2A = -~J (1.396)
Now both the vector and scalar potentials obey the retarded wave equation. They combine to
produce a four-vector which is invariant under a Lorentz transformation. The Lorentz
invariance is very useful in many branches of physics, and the Lorentz gauge is used
frequently. The Green's functions are not given for this gauge, but they are provided in books
on field theory. Obviously both the scalar and vector parts are retarded.
Another gauge which is often used, but which does not have a formal name, is the
condition that", = O. The scalar potential is set equal to zero. In this case it is found that the
longitudinal vector potential is not zero but now plays an important role. In fact, the longi-
tudinal part of the vector potential leads to an interaction between charges which is just the
instantaneous Coulomb interaction. When 0/ = 0 the longitudinal part of the vector potential
plays a role that is identical to that of the scalar potential in the Coulomb gauge.
The Hamiltonian (1.360) is written in a gauge which has the scalar potential acting
instantaneously, so that the Coulomb interaction is unretarded. This form of the Hamiltonian
is consistent with either gauge V . A = 0 or 0/ = O. One gets a different Hamiltonian for other
choices of gauge.
1.5.2. Lagrangian
So far it has been proved that the Coulomb gauge makes the vector potential transverse
and that the scalar potential acts instantaneously. Next it is shown that the Hamiltonian has the
form indicated in Eq. (1.360), which is done by starting from the following Lagrangian
(Schiff, 1968):
L = 2"1 ~ mVi J
d 3 r [E(r) 2 -
__2 + 81t 1 . A(ri) ]
B(r)2 ] - ~ ei [ o/(rJ - ~Vi
This Lagrangian was chosen to produce Maxwell's equations as well as the classical equations
of motion for a particle at r i, with charge ei and velocity Vi' in a magnetic and electric field.
There are three variables 11 which are 0/, A, rio In terms of these variables the electric and
magnetic fields are: E = -Ale - VO/, B = V x A. Each variable 11 is used to generate an
equation from
(1.397)
where the conjugate momentum is
oL
P'l = or, (1.398)
When the scalar potential 0/ is chosen as the variable 11,

oL
00/ = -per) = - ~ eio(r - r i ) (1.399)
(1.400)
oL
P>'I=-.=O (1.401)
00/
This choice produces the Maxwell equation (1.366) from (1.397):
V· E = 4rcp (1.402)
When one component of the scalar potential such as Ax is chosen as the variable in the
Lagrangian,
(10403)
(1.404)
8L
(10405)
When this equation is used in Lagrange's equation (1.397), the x-component of the Maxwell
equation (1.367) is
1 aE 4n.
VxB=--+-j (10406)
c at c
Choosing either Ay or A z as the active variable will generate the y- and z- components of this
equation. The Lagrangian does generate the two Maxwell equations which depend on particle
properties.
Another important feature is that the momentum conjugate to the scalar potential is zero
(10407)
The momentum variable conjugate to the vector potential is

E
PA = - - (10408)
4nc
which is just proportional to the electric field. This relationship IS Important, since the
quantization of the fields will require that the vector potential no longer commute with the
electric field, since they are conjugate variables.
The other equations generated by this Lagrangian are the equations for particle motion in
electric and magnetic fields. Here the variable in the Lagrangian is the particle coordinate r rx:
(1.409)
(10410)
In the last term of (10409), repeated indices over 8 imply summation. The equation deduced
from Lagrange's equation is
The total time derivative on the position dependence of the vector potential is interpreted as a
hydrodynamic derivative:
d a
-A=-A+v··VA
dt at I
Then the equation may be rearranged into
The last term is just the Lorentz force on a particle in a magnetic field since
v x B = v x (V x A) = V(v . A) - (v . V)A (1.411)
The equation for mv is Newton's law for a spinless particle in an electric and magnetic field.
The Lagrangian is a suitable starting point for the quantization of the interacting system of
particles and fields.
1.5.3. Hamiltonian
The Hamiltonian is derived from
(1.412)
where the first summation is over all of the variables and their conjugate moments. In the
present problem this summation includes the vector potential and the particle momentum; the
scalar potential has no momentum. The Hamiltonian has the form
1
H=-'LPi
m i
[ Pi---.!.A(rJ
e·
e
] + J d3 r[1
- A. · (1.
41te
-A+V\jJ) -
e
E2 8- B2]
1t
1
+ 'Lei\jJ(rJ - -'L [ Pi - e.]2 e· [ e.]
--.!.A - 'L ---'--A· Pi ---.!.A (1.413)
i 2m i e i me e
It may be collected into the form
(1.414)
Terms which are the cross product between the scalar and vector potential parts of the electric
field always vanish after an integration by parts,
(1.415)
since the Coulomb gauge is used, wherein V . A = O. The terms involving the square of the
scalar potential may also be reduced to an instantaneous interaction between charges using
(1.381):
(1.416)
With these simplifications, the Hamiltonian is written as
(1.417)
The last terms are just the energy density of the electromagnetic fields. The first terms arise
from the charged particles. This form of H is simple and instructive. However, our interest is
in obtaining a quantized version. Write the electric field energy density as the separate parts of
vector and scalar potential,
(1.418)
and the scalar potential just gives the interaction between charges.
The electric field is the conjugate momentum density of the vector potential. These two
field variables must obey the following equal time commutation relation:
(1.419)
This commutation relation is satisfied by defining the vector potential in terms of creation and
destruction operators in the form
(1.420)
There are two transverse modes, one for each transverse direction, and A. is the summation
over these two modes. The unit polarization vector ~IX gives the direction for each mode. The
operators obey the commutation relations
[akA.. at,A.'] = Okk'OA.A.' (1.421)
[akA.' ak'A.'] = 0 (1.422)
The electric field is
(1.423)
The scalar potential is not expressed in terms of operators, since its conjugate momentum is
zero. It does not influence the commutator in (1.419). The commutation relation (1.419) is the
commutator of the vector potential and its time derivative:
(1.424)
The vector potential in (1.420) has been chosen to give just this result:
_2 1/2 _2 1/2
[Acx(r, t), A~(r', t)] = i L (21tlic-rok) (21t1i~)
kA.k'A.' v vrok
x ~cx(k, A)~~(k,' A')ei (kor+k' or')20 k ,_k'OA.A.'
= 41ticOcx~O(r - r') (1.425)
The various factors of 21t and rok which enter (1.420) are selected so that the commutation
relation (1.419) is satisfied for the usual commutation relations for the operators akA. and atA.'
The expressions for the energy density of electric and magnetic fields have the form
(1.426)
(1.427)
In the second term, the photon energy is rok = ck. This term may be added to the first, and the
aa and at at terms both cancel. Combining the results from (1.417) and (1.418),
This form of the Hamiltonian is just the result which was asserted in the beginning in (1.360).
The vector potential in (1.420) is expressed in terms of the creation and destruction operators
of the photon field. The photon states behave as bosons: as independent harmonic oscillators.
Each photon state of wave vector k and polarization A has eigenstates pf the form
(1.428)
where nkA. is an integer which is the number of photons in that state. The state 10) is the
photon vacuum. The total energy in the free photon part of the Hamiltonian is
(1.429)
The Hamiltonian (1.360) has been derived for spinless, nomelativistic particles. Certainly the
most important relativistic term is the spin orbit interaction. The effects of spin also enter
through the direct interaction of the magnetic moment with an external magnetic field.
1.6. PAIR DISTRIBUTION FUNCTION
In crystalline solids, the atoms are arranged in a regular array. If Rj denotes the position
°
of the atoms, the summation of exp(iq . R) over all the atoms yields zero unless q has
particular values. These values are q = or else one of the reciprocal lattice vectors G of the
solid. In either case, the factor exp(iG . R) = 1 so that the summation yields the number of
atoms in the solid, which is N. This result is written as
(1.430)
°
where q = is in the term with G = 0. Many materials are not crystalline, e.g., liquids, gases,
and disordered solids. In discussing these quantities similar summations over particle loca-
tions are encountered. These summations still equal N for q = 0, but they must be evaluated
for q 1= 0. This evaluation is done in terms of a function S(q), which is called the static
structure factor or the static form factor.
Define the density operator for the atomic locations:
p(q) =L exp(iq . R) (1.431)

j
p( q) is an operator because the Rj describe the instantaneous location of each particle. It is

assumed the particles are classical objects, so there is no need to deal with the quantum
statistics associated with fermions or bosons. The average of this operator is defined as
(p(q)) =( ;= exp(iq . R)) (1.432)
where the average is taken over the various configurations of the atoms. Some of these
averages are defined here, without explaining how to find them from first principles. The latter
is done in standard references (Hill, 1956; Percus, 1964).
The first average is over a single density operator:
(p(q)) = Noq=o (1.433)
(p) is zero for nonzero q because there is no restriction on the location of each atom. That is,
in the averaging, each Rj can be anywhere in the material with equal likelihood since edge
effects are ignored. The averaging is equivalent to the integral over a continuous distribution
of values:
(1.434)
The integral on the right just gives a delta function at q = 0, which is the same result as
(1.433).
The second interesting average is over a product of two density operators:
(p(q)p(q')) = L(exp(iq . R;)exp(iq' . Rj ))

ij
(1.435)
Sec. 1.6 • Pair Distribution Function 59
Of course (p(q)p(q')) = N 2 when both q and q' are zero. If both q and q' are nonzero, the
average vanishes unless q + q' = O. To prove this result, go to a center-of-mass coordinate
system:
Q=q+q', (1.436)
k=!(q-q'), (1.437)
where the average is
(1.438)
There is a strong correlation between the difference Ri - R j of two particle locations, but
there is no correlation for the center of mass Ri + Rj . The term exp[iQ· (Ri + R)12]
averages to zero unless Q = 0, which gives q = -q' = k. Whenever q i= 0 the average is
(p(q)p(q')) = Oq+q'=o 2:(exp[iq . (Ri - R)]) (1.439)
if
There is strong correlation between the relative positions of two atoms at short distances. If
one atom is at a spot, the other will not be at the same spot because atoms repulse each other
at short separations. The static structure factor S(q) is defined as
1
Noq=o + Seq) = N L.(exp[iq . (Ri - R)]) (1.440)
lj
= ~ 2:(1 + 2:exp[iq . (Ri -

N i j#i
R)]) (1.441)
One subtracts the term Noq=o. The right-hand side contains a summation over all particles at
Ri and then a summation over all its neighbors Ri - R j . The first term has i = j and is
independent of q. When this formula is averaged over all possible positions of the impurities,
it should give the same result for each particle R i. The summation over Ri just gives N:
(1.442)
The prime on the summation means that ~R = 0 is excluded from the summation. The
crucial average is the second term on the right. It averages over the relative positions ~R of
the atoms near to the one at R i . This average can also be expressed as a function of r. Define
g(r) as the pair distribution function. It is defined as the probability that another particle is at
r if there is already one at r = O. The normalization is chosen to give g(r) -+ I at large
distances. In terms of the pair distribution function, S(q) may be written as
J
Noq=o + Seq) = 1 + n d 3 rg(r)eiq ' r (1.443)
where n = N Iv is the average density of the system. Also write
J
Noq=o = n d 3 reiq • r (1.444)
and transpose this expression to the right to give the equivalent equations
J
S(q) - 1 = n d 3 reiq • r [g(r) - 1]
g(r) - 1 =!J d 3 q e-iq'r[S(q) - 1]

n (21t)3
(1.445)
The inverse Fourier transform may now be taken to give g(r) as a function of S(q). Both g(r)
and S(q) go to unity at large values of their arguments. The functions [g(r) - 1] and
[S(q) - 1] are both quantities which are short-ranged in their arguments. Their mutual Fourier
transforms are well defined.
The function S(q) is determined experimentally by scattering from targets. Afterwards,
g(r) is found by a numerical Fourier transform. In liquids they depend only upon the
magnitude of their vector arguments: g(r) and S(q). The well-known results for liquid 4He are
shown in Fig. 1.3, which is from Mozer et al. (1974). The pair distribution function g(r)
should vanish according to the dashed line at short distances because the atoms do not
penetrate each other at the low thermal energies of the liquid. The numerical transform shown
by the solid line has a spurious peak: for R < 2 A because of inaccuracies in S(k) for k -+ 00.
The higher moments such as (p(q)p(q')p(q"») would also be interesting were they
known. But they are not easily obtainable from experiments nor from theory.
In the model of the homogeneous electron gas, the electrons are not ordered. In this case
one can also describe relative behavior of pairs of particles by a pair distribution function g(r).
This quantity is always defined as the average over the motion of the particles. In a quantum
system the diagonal density matrix is defined as the square of the N -particle wave function:
PN(r\r2 ... r N) = 1'I'(r\r2 ... rN)1 2 (1.446)
It is normalized so that unity is obtained when integrating over all the coordinates:
1 = Jd3r\d3r2·.·d3rNPN(r\r2 ... rN) (1.447)
4 8 12 16 6.0
R (A)
FIGURE 1.3 g(R) (left) and S(k) (right) for liquid 4 He. S(k) is obtained by X-ray scattering measurements, and g(R)
is found by the numerical transform. Uncertainties in S(k) at large k produce the extraneous structure in g(R) at low R,
which is to be ignored. The dashed line shows the proper extrapolation. p = 0.1628 g/cm3 , T = 2.86 K. Source:
Mozer et al. (1974) (used with permission).
Sec. 1.6 • Pair Distribution Function 61
Since the many-particle wave function qt is either even or odd under the interchange of two
identical particles, the density matrix is unchanged by this operator. Any two particles can be
called rl and r2. Their relative behavior, after averaging over all the other particles, is
obtained by integrating over the coordinates of all the other particles:
P2(rl, r2) = J d3r3d3r4 ... d 3rNPN(rl r2 ... rN) (1.448)
P2(rl, r2) is the average two-particle density matrix. The "integration" over other particles
included spatial integration, spin averages, etc. Similarly, the one-particle density matrix is
that obtained by integrating over all but one of the particles:
PI (rl) = Jd3r2d3r3 ... d3rNPN(rl r2 ... rN) (1.449)
= Jd3r2P2(rl, r2) (1.450)
1= Jd3rl p(rl) (1.451)
For homogeneous systems, the two-particle density matrix P2(rl, r2) can only depend on the
relative positions rl - r2 of the two coordinates. This quantity must be related to the pair
distribution function. They are in fact proportional. The constant of proportionality is
determined from the condition that g(r) ~ 1 at large separation, so
(1.452)
As an example of particular behavior, evaluate g(r) for electrons using the Hartree-Fock
approximation. The many-particle wave function is just an N-dimensional Slater determinant,
so the N -particle density matrix is just the square of this determinant:
2
\jI",\ (rl) \jI",/rl) \jI"'N (rl)
\jI",\ (r2) \jI"'2 (r2) \jI"'N (r2)
1
PN(rlr2··· rN) = - (1.453)
N!
\jI",\ (rN) \jI"'2 (rN) \jI",)rN)
The single-particle orbitals \jI",\ (r l ) are assumed to be orthogonal for different states A. A
theorem states (see Parr, 1964) that if one integrates over all but two coordinates, the two-
particle density matrix, from this Slater determinant, is the sum of all pairs of two-particle
wave functions:
(1.454)
The summation is taken over all pairs of occupied states. The two-dimensional Slater
determinant can be expanded to give the following equivalent result:
1 2
P2(rl' r2) = 2N(N _ 1) L [\jI",\ (rlN",/r2) - \jI",\ (r2)\jI",/rl)] (1.455)
"'\#"'2
A further integration gives the one-particle density matrix as the summation over all occupied
states of the square of the wave function:
(1.456)
(1.457)
These theorems apply to any Hartree-Fock system with any orbitals. For homogeneous
systems, it is assumed the wave functions are translationally invariant. If they are plane waves
with spin up (x, the two-particle density matrix is
eik.r
W=,.;v(X (1.458)
P2
(r, r ') -_ v N(N1 _ 1) "[1
~
_ i(k-k')o(r-r')]
e (Xl(X2 (1.459)
2
Using (1.452) gives a pair distribution function of the form
g(r - r') = _1 I:[1 - ei(k-k'Hr-r')](X1 (X2 (1.460)

N2kk'
In most systems, the electrons do not all have spin up, and one also has to average over all
spin coordinates to get g(r). This procedure is discussed in Chapter 5.
Problems
1. Solve the classical vibrational modes of a one-dimensional chain of atoms of type A and B. They
alternate on the chain with masses mA and me. The harmonic spring between atoms has spring constant
K.
2. Write down the Hamiltonian of Problem 1. Solve it, and show that it may be reduced to the form
(1.461)
where the (Ok), are the classical normal modes.
3. Find the exact solution to
H=Eoata+E\(ata t +aa) (1.462)

where Eo and E\ are constants and a and at are boson operators.
4. Solve the Hamiltonian below with a canonical transformation:
(1.463)
(l.464)
Problems 63
a. Show that A. = A.* since H is Hennitian.

b. Use the expansion (1.356), and show that only a few terms in the series are finite-the
remainder vanish.
c. Find the choice of A. which reduces H to (1.50).
5. Consider a fennion system which has three energy states with eigenvalues E I , E 2 , and E 3 • There
also exist matrix elements which connect these states and pennit transitions between them: M 12 , M 23 ,
and M13
(a) Write down the Hamiltonian for this system in terms of creation and destruction operators.
(b) Determine the eigenvalue equation for this system.
6. Consider a tight-binding solid which has alternate atoms of type A and B. The electron Hamiltonian
in the nearest neighbor model has the form
(1.465)
where ai and bj are electron operators for atoms of type A and B. Find the exact eigenvalues of this
Hamiltonian.
7. Calculate the exact partition function
(1.466)
for the Hubbard model (1.182) in the atomic limit. Then give the expression for the average number of
electrons N = -(an/a~).
8. Take a three-atom chain with periodic boundary conditions. Each atom has one orbital state with
two spin configurations.
(a) Solve the nearest neighbor tight-binding model for zero, one, two, or three electrons with spin
up, and give the partition function. Show that it is the same as (1.176).
(b) Give the partition function for the XY model, and show that it is different.
9. Give the partition function for the Hubbard model for a two-atom chain with periodic boundary
conditions.
10. What is the energy current operator jE for the Hubbard model (1.182)?
11. Show that the energy current in the one-dimensional harmonic chain (1.50) is
(1.467)
12. Assume that the Hamiltonian in the tight-binding model has additional terms describing how
phonons (b i, bi) on a site i interact with electrons ni when they are on the site. How do these terms affect
the heat current?
~]roob; bi + Mn;(bi + bi)] (1.468)

i
64 Chap, 1 • Introductory Material
13, Explicitly verify the equation of continuity,
~ per, t) + V ' j(r, t) = 0 (1.469)
for a gas of free fennion particles which have
H =L tkctcrCkcr (1.470)
kcr
(1.471)
J'(r, t ) -_ - 1 '"
L.. e
-iq' r(k t
1)Ck+qcrCkcr
+:2q (1.472)
mVkqcr
14. Find the effective interaction VR(r) between two fixed electrons (1.247) using the defonnation
potential interaction (1.249) and a Debye spectrum.
15. Find the exact solution in one dimension and the partition function for the XY model with a
magnetic field in the z direction.
16. The Holstein-Primakoff transfonnation between spin one-half operators and bosons is
s(z) -1_ ata (1.473)
I - 2 I I
(1.474)
(1.475)
where the al are boson operators.

(a) Use these definitions to evaluate the following commutators:
[sj+) , S~-)l; [Si z ), S~-)l (1.476)
(b) Write out the Heisenberg Hamiltonian in tenns of the a operators.
(c) At low temperature, where the number of excitations iii :( (aJ al) is small, one can simplify the
Hamiltonian by neglecting all tenns of the fonn at at aa which describe the scattering of two
excitations. Then one can solve the Hamiltonian exactly. What are the eigenvalues?
Chapter 2
Green's Functions at Zero Temperature
Many-body calculations are often done for model systems at zero temperature. Of course, real
experimental systems are never at zero temperature, although they are often at low
temperature. Many quantities are not very sensitive to temperature, particularly at low
temperature. Zero temperature calculations are useful even for describing real systems.
Furthermore, the zero temperature property of a system is an important conceptual quantity-
the ground state of an interacting system. A system is often described as its ground state plus
its excitations, and the ground state may be deduced from a zero temperature calculation.
Many zero temperature calculations have been done to deduce, for example, the ground state
of the homogeneous electron gas or the ground state of superfluid 4 He.
Some workers believe that zero temperature calculations are easier to perform. There are
usually more terms at nonzero temperature, which is a nuisance. We do not completely share
this viewpoint. The Matsubara methods, which are described in Chapter 3 for nonzero
temperature, are very easy to use. We usually do zero temperature calculations by first finding
the nonzero temperature formulas and then taking the limit of zero temperature. However,
sometimes it is easier to do zero temperature calculations from the beginning. The zero
temperature formalism is a necessary part of one's calculational machinery.
It is presumed that one is trying to solve a Hamiltonian which cannot be solved exactly.
One does not need Green's functions if the problem may be solved exactly. Very few exact
results were obtained by Green's functions which were not first obtained by conventional
theoretical techniques. It is assumed that one is trying to deduce the properties of some
system described by a Hamiltonian H which may not be solved exactly. The usual approach is
to set
H=Ho+V (2.1)
where Ho is a Hamiltonian which may be solved exactly. The term V represents all the
remaining parts of H. One tries to choose Ho so that the effects of V are small. The basic
procedure is to start with a system completely described by Ho. The effects of V are intro-
duced, and then calculations are done to find how V changes the properties. These steps are
the basic procedure in many-body theory.
65
66 Chap. 2 • Green's Functions at Zero Temperature
2.1. INTERACTION REPRESENTATION
Most readers of this book should be familiar with the interaction representation. A short
discussion of this topic will be presented anyway, as a refresher. There are three repre-
sentations which will be discussed: Schr6dinger, Heisenberg, and interaction.
2.1.1. Schrodinger
Elementary quantum mechanics is taught in the Schrodinger representation, which is
based on the formula (Ii = 1)
i ata\jJ(t) = H\jJ(t) (2.2)
which has the formal operator solution

\jJ(t) = e-iHt\jJ(O) (2.3)
The use of this formula requires some assumptions:
1. The wave functions \jJ(t) are time dependent, even if this dependence is only a simple
factor of exp( -iEt).
2. Operators such as the Hamiltonian H are taken to be independent of time.
2.1.2. Heisenberg
It is possible to solve quantum mechanical problems another way which gives exactly the
same answers yet uses methods that look quite different. The Heisenberg representation has
the following properties:
1. The wave functions are independent of time.

2. The operators are time dependent, and this dependence is given by
OCt) = eiHtO(O)e-iHt (2.4)
or, equivalently, one is trying to solve the equation of motion which is derived from a time
derivative:
.a
l-O(t) = [OCt), H] (2.5)
at
Note that the time dependences of these two representations appear to be contrary. Yet it was
asserted they give the same answer. To prove this identity takes too much space, so the result
is just made plausible.
In physics one is usually trying to evaluate matrix elements in order to determine
transition rates. In the Schr6dinger representation, the matrix element of the operator 0(0)
between two states is
(2.6)
In the Heisenberg representation one obtains the result
(\jJr (0)0(tN2(0)) = (\jJt (0)eiHtO(0)e-iHt\jJ2(0)) (2.7)
Sec. 2.1 • Interaction Representation 67
The two representations produce the same result. In order to understand this assertion, first
recall that the Hamiltonian H will include some interaction terms. In the Heisenberg repre-
sentation, these interactions act upon the operators, thereby changing them. When taking
matrix elements, these changed operators are projected back upon the unchanged states in
order to see how much the operators are changed by the interactions. On the other hand, the
Schrodinger picture leaves the operators fixed and instead has the interactions affect the wave
functions. Then the new wave functions are used to evaluate the matrix elements of the
unchanged operators. The result is the same either way.
2.1.3. Interaction
The interaction representation is another way of doing things. Here both the wave
functions and the operators are time dependent. The Hamiltonian is separated into two parts,
H=Ho+ V (2.8)
where Ho is the unperturbed part, while the V are the interactions. At the moment the exact
form of V is left unspecified, and we merely note that it may be either, some, or all of the
interactions discussed in Chapter 1. Usually Ho is selected as a Hamiltonian which is exactly
solvable.
Operators and wave functions in the interaction representation will be denoted by a caret,
such as in V. This notation will distinguish their time dependence from that of the other
representations:
1. Operators have a time dependence
(2.9)
2. Wave functions have a time dependence
(2.10)
It is assumed that [Ho, V] =J O. If these operators do commute, the problem is usually too
trivial to require many-body theory. When these operators do not commute, the exponentials
cannot be combined, because
(2.11)
only if [A, B] = O.
Before going any further, let us check to see that this choice of time dependence does
produce the same matrix elements as before:
('"r(t)O(t)"'2 (t)) = (\jJr(O)eiHte-iHot (eiHot Oe-iHot)eiHote-iHt\jJ2 (O))

= (\jJ t (O)e iHt O(0)e-iHt\jJ2(0)} (2.12)
It does. The time dependence of the operators is governed by the unperturbed Hamiltonian.
Next show that the time dependence of the wave functions is governed by the interactions
~~(t) = ieiHot(Ho - H)e-iHlo/(O)

at
= _ieiHotVe-iHto/(O)
= _ieiHotVe-iHot[eiHote-iHto/(O)]
a
at o/(t) =
A A A
-iV(t)o/(t) (2.13)
This result proves the assertion that the time dependence of ~(t) is determined by Vet). An
operator was introduced into Eq. (2.10) which is defined as U(t):
(2.14)
This function has the value of unity at t = 0: U(O) = 1. Furthermore, it obeys a differential
equation which can be written in the interaction representation:
a
- U(t) = ie'"Rot(Ho - H)e-'"Rt
at
= -iV(t)U(t) (2.15)
In order to solve this equation, one way of proceeding is by integrating both sides of the
equation with respect to time:
U(t) - U(O) = -i I dt\ V(t\)U(t\) (2.16)
Rearranging gives (U(O) = 1)

U(t) = 1 - i J: dt\ V(t\)U(t\) (2.17)
If this equation is iterated repeatedly, then
U(t) = 1 - i I dt\ V(t\) + (_i)2 J: dt\ Ldt2 V(t\) V(t2) + ...
= f: (_i)n JI dt\ Jt dt2 ... Jt 1

n 1
- dtn V(t\) V(t2) ... V(tn) (2.18)
n=O 0 0 0
At this point it is convenient to introduce the time-ordering operator T, which should

not be confused with the temperature. The T operator acts upon a group of time-dependent
operators,
(2.19)
and is just an instruction to arrange the operators with the earliest times to the right. For
example,
(2.20)
Sec. 2.1 • Interaction Representation 69
It helps to introduce the following step function:
Sex) =1 if x > 0
=0 ifx<O
=! ifx=O (2.21)
For two operators, the explicit definition of T ordering gives
Of course, if V(t l ) and V(t2 ) commuted with each other, then the order of the operators is
unimportant. The T ordering needs to be applied only to operators which do not commute at
different times. Now rearrange the integral by using the above identity:
(2.23)
The second term on the right-hand side is equal to the first, which is easy to see by just
redefining the integration variables t I ~ t2 , t2 ~ t I. These steps give
(2.24)
Similarly, one can show that
(2.25)
Returning to the expansion of U(t) gives
(2.26)
This expansion may be abbreviated by writing it as
(2.27)
However, it should always be kept in mind that the exponential form is really just a shorthand
for the series definition (2.26). The T-ordered series definition is really equivalent to the
original expansion (2.18).
70 Chap.2 • Green's Functions at Zero Temperature
2.2. S MATRIX
It was shown in the previous section that the wave function in the interaction repre-
sentation had a time dependence given by
(2.28)
~ow define the S matrix as the operator Set, t') which changes the wave function ~(t') into
\)I(t):
(2.29)
From the original definition
(2.30)
which produces the result
Set, t') = U(t)ut(t') (2.31)
Now examine some properties of this operator. The first two of the following identities may be
proved in a trivial way:
1. Set, t) = 1 = U(t)ut(t) = eiHote-iHt(eiHte-iHot)

2. st(t, t') = U(t')ut(t) = Set', t)
3. Set, t')S(t', t") = Set, t")
The third identity can be shown by appealing to the original definition:
~(t) = Set, t')~(t') = Set, t')S(t', t")~(t") = Set, t")~(t")

Finally, it is shown that Set, t') can also be expressed as a time-ordered operator,
a a t ~
atS(t, t') = at U(t)U (t') = -iV(t)S(t, t') (2.32)
which has the solution
Set, t') = T exp [ -i f dt, V(t,) ] (2.33)
The function ~(O) == \)1(0) was introduced in the discussion of the previous section. It is a
wave function in the Heisenberg representation, so that it is independent of time. If a
Schrodinger wave function is defined as
\)lit) = e-iHt\)l(O) (2.34)
then \)1(0) is also the Schrodinger wave function at t = O. Define
~(t) = U(t)\)I(O) (2.35)

Then \)1(0) is also the t = 0 wave function in the interaction representation. At zero
temperature the only wave function of special interest is the ground state wave function. For
Green's functions it is necessary to define "'(0) as the exact ground state wave function. Since
the total Hamiltonian is H, the exact ground state must have the lowest eigenvalue of this
Hamiltonian. An immediate problem is that the eigenvalues or eigenstates of the Hamiltonian
Sec. 2.3 • Green's Functions 71
H are not known initially. In fact, that is exactly the kind of information we are trying to
obtain by using Green's functions.
There is a problem in that the formalism is based on the wave function \jt(0) which is not
yet known. Now in the interaction representation set H = Ho + V, where Ho is chosen to be
sufficiently simple that its eigenvalues and eigenstates are known. Let the lowest eigenvalue
of Ho-its ground state-be denoted <1>0' Somehow the unknown wave function \jt(0) must be
determined in terms of the known wave function <1>0'
The relationship between the two ground states \jt(0) and <1>0 at zero temperature was
established by Gell-Mann and Low (1951):
\jt(0) = S(O, -00)<1>0 (2.36)
This result will not be proven: instead, it will be made plausible. One known result is that
\jt(t) = S(t, 0)\jt(0) (2.37)

Operate by S(O, t) and get
\jt(0) = S(O, t)\jt(t) (2.38)

since S(O, t)S(t, 0) = 1 by the use of the previous theorems. Let t --+ -00, and find
\jt(0) = S(O, -00 )~( -00) (2.39)
The important assertion is that ~(-oo) is equal to <1>0' The traditional argument is that one
starts in the dim past (t --+ -00) with a wave function <1>0 which does not contain the effects
of the interaction V. The operator S(O, -00) brings this wave function adiabatically up to the
present t = O. It is a wave function which does contain the effects ofthe interaction V, so that
it is an eigenstate of H.
There is an additional property of these states which is needed for the discussion of
Green's functions. As t --+ +00, then
~(oo) = S(oo, 0)\jt(0) (2.40)

One possible assumption is that \jt( 00) must be related to <1>0' If they are equal except for a
phase factor L:
<l>oeiL = ~(oo) = S(oo, 0)\jt(0) = S(oo, -00)<1>0 (2.41)
e iL = (oIS(oo, -00)1<1>0) (2.42)
Alternatives to this assumption are discussed in Sec. 2.9.
2.3. GREEN'S FUNCTIONS
Most of this book is concerned with Green's functions for only three types of particles:
electron, phonon, and photon. The electron and phonon cases will be discussed here. The
photon case is badly muddled by the several available choices of electromagnetic gauge.
Section 2.10 is devoted to discussing these complications, and the photon Green's functions
will be treated there. But keep in mind that the basic proofs for the photon case are essentially
identical to the phonon case.
At zero temperature the electron Green's function is defined as
(2.43)
The quantum number A can be anything depending on the problem of interest. Quite often it
shall be taken to be the quantum numbers of the free-electron gas A = (p, cr), where p is the
wave vector and cr is the spin. At zero temperature the state I) must be the ground state. If the
Hamiltonian of the problem is chosen to be H, then I) is the ground state of H, and therefore
it is an eigenstate of H.
Of course, initially the ground state or any other eigenstate of H is not known since that
is what is to be determined by using Green's functions. Write H = Ho + V, where Ho is the
unperturbed part, while V is the interaction. Choose Ho so that its eigenstates are known. In
Sec. 1.3 the CI.. are defined in terms of a complete set of states 0/1... Now select this complete
set of states as the eigenstates of the unperturbed Hamiltonian Ho. In the definition of the
Green's function the CI.. represent states of H o, while the ground state I) is an eigenstate of H.
Furthermore, (2.43) is defined in the Heisenberg representation, so that I) is independent of
time, while CI..(t) is given by the usual result
(2.44)
One way of understanding the Green's function is to observe that it describes a certain
Gedanken experiment. For t > t
(2.45)
Here one takes the real ground state, and at a time t one creates an excitation A. At a later
time t one destroys the same excitation. Now if A were an eigenstate of H, with HCtl) =
el..Ctl) and HI) = eol), then this state would propagate with a simple exponential time
dependence:
G(A, t > t') = -i exp[-i(t - t')(el.. - eo)] (2.46)
Because A is not usually an eigenstate of H, the particle in the state A gets scattered, shifted in
energy, etc., during the time interval t - t. When one measures at a later time t, to see how
much amplitude is left in the state A, the measurement provides information about the system.
For the other time arrangement t > t
(2.47)
where the sign in front is changed whenever the position of two fermion operators is inter-
changed. Now an electron is destroyed from the ground state at time t and recreated at the
later time t, which is possible only if there are electrons in the ground state at zero
temperature. One case where this is true is in the Fermi sea of a metal. The initial destruction
of an electron at time t must, roughly speaking, remove an electron from the filled Fermi sea.
This destruction creates a vacancy, often called a hole, and the hole can interact and scatter in
the interval t - t. Then ct acting at t destroys the hole state A, and this measurement
provides information about the hole excitation.
The next step is to take the Green's function, defined in the Heisenberg representation,
and convert it to the interaction representation. Call 1)0 == <Po the ground state of H o, so that
I) = S(O, -00)1)0 (2.48)
Next change the operators to this representation:
C,,(t) = eiHle-iHOIC,,(t)eiHOle-iHI = ut(t)C,,(t)U(t)

= S(O, t)C,,(t)S(t, 0)
G(A, t - tl) = -i0(t - t')o(IS( -00, O)S(O, t)C,,(t)S(t, O)S(O, tl)
x ct (t1)S(tl, O)S(O, -00)1)0
+ i0(t' - t)o(IS(-oo, O)S(O, t')Ct(tI)S(tl, O)S(O, t)
x C,,(t)S(t, O)S(O, -00)1)0
This expression may be regrouped by using the properties of the S matrix developed in the
previous section. The left-hand bracket is replaced by
-iL o(IS(oo,O)
o(IS(-oo, 0) =e o(IS(oo, -oo)S(-oo, 0) = o(IS(oo, -00)1)0
so the Green's function becomes
G(A, t - tl) =- o(IS(OO,i -00)1)0 [0(t - t')o(IS(oo, t)C,,(t)
x Set, t')ct(tI)S(tl, -00)1)0
- 0(t' - t)o(IS(oo, t')ctS(tl, t)ct(t)S(t, -00)1)0]
The first term can be simplified by writing
0(t - t')o(IS(oo, t)C,,(t)S(t, t')C,,(tI)S(tl, -00)1)0
= 0(t - t')o(ITC,,(t)ct(t')S(oo, -00)1)0 (2.49)
The operator S(oo, -00) contains operators which act in the three time intervals:
(00, t), (t, t'), and (t', -00). The T operator automatically sorts these segments so they act in
their proper sequens.es, which are, respectively, to the left of C,,(t), between C,,(t) and ctct'),
ct
and to the right of (t'). The total Green's function is expressed as
A At
G(A, t _ tl) = _io(ITC,,(t)C,,(t')S(oo, -00)1)0
(2.50)
o(I TS(oo, -00)1)0
which is the desired result. It does not matter where S( 00, -00) is placed in the numerator,
since the time ordering operator puts the pieces in the right place.
A Green's function can also be defined for the special case where the interactions V = 0
and hence the S matrix is unity. This Green's function plays a special role in the formalism,
and it is designated by G(O):
(2.51)
G(O) is often called the unperturbed Green sjunction, or sometimes the name free propagator
is used.
There are two quite different types of electronic systems in which we want to employ the
Green's function analysis. These two have quite different ground states, 1)0' and also quite
different real ground states, I}. These two systems are the following.
1. An empty band. Here the properties are studied of an electron in an energy band in
which it is the only electron. An example is an electron in the conduction band of a semi-
conductor or an insulator. In this case the ground state is the particle vacuum, which is
denoted as 10). This state has the property that
CpIO) = 0 (2.52)
aqlO) = 0 (2.53)
where the two destruction operators represent electrons and phonons. Therefore both H o and
V give zero when operating upon the vacuum. It follows that the S matrix also gives unity
when operating upon the vacuum:
Set, -00)10) = 10) (2.54)
Both of the ground states, I}o and I}, are the vacuum 10). The Green's function can exist only
for the time ordering
G(A, t - t f) = -iE>(t - 1)(OIC\(t)CI(tf)IO) (2.55)
The unperturbed Green's function G(O) is particularly easy to evaluate:
G(O)(A, t - I) = -iE>(t - l)e-iE).{t-t')(OICA,CIIO) (2.56)
= -iE>(t - l)e-iE).(t-t') (2.57)
The Fourier transform of G(A, t) is defined as
G(A, E) = J~oo dteiE1 G(A, t) (2.58)
To make the integrals converge, add the infinitesimal quantity io to the exponents:
O<°)(A, E) = -i J~ dtei(E-E). +io)1
O<0)(A, E) = 1 (2.59)
E - f.A, + io
2. A degenerate electron gas. Our second case is where the electrons are in a Fermi sea at
zero temperature. The standard example is a simple metal. The system has a chemical
potential 11, and all electron states with E < 11 are occupied. If the unperturbed electrons
(eigenstates of H o) are characterized by an energy f.k' the ground state I}o has all states f.k < 11
filled and states f.k > 11 empty. It is convenient and conventional to measure the electron's
energy relative to the chemical potential, to define ~k = f.k - 11. For a spherical Fermi surface
with Fermi wave vector PF'
o(lctCkl}o = E>(PF - k) (2.60)
o(ICkCtl}o = E>(k - PF) (2.61)

A more general way to write this result is
The unperturbed Green's function is now
oC°)(A, t - t') = -iE>(t - t')o (I C),.(t)ct(t') 1)0 + iE>(t' - t)o(ICt(t')C),.(t)l)o
= -i[E>(t - t')e(~k) - E>(t' - t)E>( _~k)]e-i~k(t-t') (2.62)
The Fourier transform is
oC°)(k, E) = - { e(~k) [ dteit(E-~k+iO) - e( -~k) Loo dteit(E-~k-iO)]

e(~k) + e( -~k)
(2.63)
E - ~k + io E - ~k - io
The energy ~k = tk - I! is measured with respect to the Fermi surface ~kF = O. Another way
to write G(O) is
oC0)(k E) = 1 (2.64)
, E - ~k + iO k
Ok = 0 sgn(~k) (2.65)
where Ok is a small infinitesimal part which changes sign at the chemical potential ~k = O.
3. Phonons. The Green's function for phonons is defined as
D(q, A, t - t') = -i(ITAq),.(t)A_q),.(t')I} (2.66)
(2.67)
The subscripts A refer to the polarization of the phonons. Most applications have one kind of
phonon in Hamiltonians which do not mix polarizations, and then these subscripts are omitted
entirely. In the interaction representation
(2.68)
At zero temperature there are no phonons. The ground states I} and 1)0 are again the particle
vacuum IO}. Note that in an electron-phonon system the notation Do means the combination
of ground states for electrons, phonons, etc. Although the phonon system has the vacuum as
its ground state, either of the two-electron ground states can be used.
The unperturbed phonon Green's function is defined as
(2.69)
At zero temperature
o(laqa~l)o = 1
o(la~aql)o = 0
D(O)(q, t - t') = -i[e(t - t')e-i(fJq(t-t') + e(t' - t)ei(fJq(t-t')] (2.70)
The Fourier transform has two terms which correspond to the two time orderings:
D(O)(q, (0) = J~oo dtei(fJtD(O)(q, t) (2.71)
1 1
D(O)(q, (0) = . (2.72)
(0 - (Oq + 10 (0 + (Oq - io
2(Oq
= (2.73)
+ iO
---=----:?--~
(02 - (O~
Sometimes it is useful to have the phonon Green's function at nonzero temperature. In this
case, the thermal average is taken of the phonon occupation numbers,
o(laqa~}o = Nq + 1
1
o(la~aql)o = Nq = eP(fJq _ 1
and the Green's function of time is

(2.74)
In some many-body systems the interactions cause changes in the phonon energies. For these
cases, the thermal average at nonzero temperature should be over the frequencies which result
from the interactions. The use of the unperturbed frequencies is valid only if they are not
altered, which happens when the perturbation is either localized or confined to a small number
of particles.
2.4. WICK'S THEOREM
The Green's function is evaluated by expanding the S matrix S(oo, -00) in (2.33) in a
series such as (2.26):
(-it+ Joo 1 Joo

G(p, t - t') = fo-n-!-
00
-00 dt 1 ••• -00 dtn (2.75)
A A A A At
0(ITCp(t)V(t 1)V(t2)· •• V(tn)Cp(t')l)o

x---~-------~--- (2.76)
o(IS(oo, -00)1)0
Let us, for the moment, ignore the phase factor o(IS(oo, -00)1)0' It will be taken care of in
Sec. 2.6. The immediate aim is to learn how to evaluate time-ordered brackets such as
(2.77)
Sec. 2.4 • Wick's Theorem 77
Suppose that V(tl) is the electron-electron interaction Eq. (1.160):
V(tl) = ~2 "L.. L..

,,41t~ d ct, ,C , ,C eitl(~k+q+~k'-q-~k-~k')
q2 k+q,s k -q,s k ,s k,s
k'kq ss'
In this case the time-ordered bracket (2.77) contains seven creation operators and seven
destruction operators. It is a very arduous task to evaluate this bracket: there are many
possible time orderings and many possible pairings between creation and destruction
operators. What is meant by pairings between operators? First note that these brackets always
contain the same number of creation and destruction operators. One is always trying to
evaluate the product of n creation operators and n destruction operators between the ground
state 1)0
(2.78)
The effect of a creation operator C!,(t') is to put an electron into the state n'. The system must
be back in the ground state before the final operator of 0 (I, so that one of the destruction
operators Cm(tm) must destroy the state n' and m = n' for some m. For example,
(2.79)
equals zero unless a = ~, while
(2.80)
equals zero unless a = ~,y = 0, or unless a = 0, ~ = y. There are many possible time
orderings and pairings in a bracket like (2.77). However, only a limited number of these
combinations are physically interesting. Our aim is to sort these in a simple way to identifY
the important terms. This sorting is achieved with the help of some theorems which simplifY
the procedures. The first of these is Wicks theorem.
This theorem is really just an observation that the time ordering can be taken care of in a
simple way. Wick's theorem states that, in making all the possible pairings between creation
and destruction operators, each pairing should be time-ordered. The time ordering of each
pair gives the proper time ordering to the entire result.
For example,
A At A At
0(ITCit)C~(tl)Cr<t2)Co(t')1)0
A At A At
= 0 (ITCI),(t)C~(tl)l)oo (I TCy(t2) Co (t') I) 0
A At A At
- 0(I TC",(t)Co(t')1)00 (I TCy(t2)C~(tl)1)0
Note that there is a time-ordering operator T in each of the two pairing brackets. For n = 3
creation and destruction operators in the original bracket, there are six possible pairings; for n
operators of each kind there are n! possible pairings. Also note that within a pairing bracket,
the labels a, ~, etc., must be the same. These labels denote eigenstates, so the creation and
destruction operators must refer to the same state.
A few simple rules should be kept in mind when making these pairings. The first is that a
sign change occurs each time the positions of two neighboring Fermi operators are inter-
changed. One keeps count of the number of interchanges needed to achieve the desired
pairing. An odd number of interchanges is the origin of the minus sign in the second term of
the example above.
The second rule concerns the time ordering of combinations of operators representing
different excitations. For example, consider the following mixture of phonon and electron
operators:
(2.81)
Because electron operators commute with phonon operators, it is not important how they are
ordered with respect to each other. The bracket can be immediately factored into separate
parts for electrons and phonons:
(2.82)
This separation is always possible with different kinds of operators, i.e., whenever operators
commute. Wick's theorem also applies to brackets of phonon operators; for example,
o(ITA ql (t 1)A q2 (t2)A q3 (t3)A q4 (t4 ) 1)0

= o(ITA ql (tl)A~ (t2)1) 00 (I TAq3 (t3)Aq4 (t4 )1) 0
+ o(ITA ql (t1)Aq3 (t3) I) 00 (ITAqJt2)A q4 (t4)1)0

+ o(ITA ql (tl)Aq/t4)1}oo(ITAq2 (t2 )A q3 (t3)1)0 (2.83)
In this case, for boson operators, there is no change of sign when the order of the operators is
changed. Each of the above brackets vanishes unless the two wave vectors are equal and
opposite:
= Oql+q2 Oq3+q4 o(ITA ql (t1)A_ ql (t2) 1)00 (I TAq3 (t3)A_ q3 (t4)1)0
+ Oql+q3 Oq2+q4 o(ITA ql (t1)A_ ql (t3) 1)00 (I TAq2 (t2)A-~ (t4 ) 1)0
+ Oql+q4 Oq2+q3 o(ITA ql (t1)A_ ql (t4 ) 1)00 (I TAq2 (t2)A-q2 (t3) 1)0 (2.84)
The third rule is a method of treating the "time ordering" of two operators which occur at the
same time, such as
(2.85)
In these cases the destruction operator always goes to the right:
(2.86)
and the term is just the number operator which is independent of time. This convention is
dependent on the convention used to write down the Hamiltonian. In constructing H the
destruction operators are put to the right of the creation operators in all terms in the
Hamiltonian.
Sec. 2.4 • Wick's Theorem 79
When two electron operators have different time arguments in a pairing, it is conven-
tional to put the creation operator to the right:
(2.87)
This term can be immediately identified as the unperturbed Green's function iG(0)(k 1, t2 - t1).
All of the pairing brackets for electron operators are either Green's functions or else number
operators. The previous examples can also be written in terms of Green's functions:
. 2 ~ ~t ~ ~t
(-i) 0(ITC<x(t)C~(t1)Cr<t2)Cll(t')1}0
= O<x~OYllG(O)(IJ(, t - t 1)G(0)(y, t2 - tf )
- O<xllO~yG(O)(IJ(, t - t')O<0)(y, t2 - t 1)
3 ,...
(-i) o(ITAql (t 1)Aq2 (t2)Aq/t3)Aq4 (t4)1) 0
A A ,..
= Oql+q2 0q3+q4 D(0)(q1' t1 - t2)D(0)(q3' t3 - t4)
+ Oql+q3 0q2+qP(0)(q1' t1 - t3)D(0)(q2' t2 - t4)
+ Oql+q4 0q2+qP(0)(q1' t1 - t4)D(0)(q2' t2 - t 3)
In summary, Wick's theorem states that a time-ordered bracket may be evaluated by

expanding it into all possible pairings. Each of these pairings will be a time-ordered Green's
function or a number operator nF or nB. This expansion gets the correct time-ordering for the
entire brackets. Wick's theorem is valid only when the Hamiltonian Ho is bilinear in creation
and destruction operators. The theorem fails if Ho contains pair-wise interactions between
particles.
A comprehensive example is done for the n = 2 term in the S matrix expansion in
(2.76). The n = 0 term is always G(O):
The interaction will be taken as the electron-phonon interaction:
(2.88)
First note that the n = 1 term must vanish because it contains the factor
(2.89)
which is zero since the factors (OlaqIO) and (Ola~IO) are zero. Similarly, all the terms where~ n
is odd vanish because their time-ordered bracket for phonons contains an odd number of Aq
factors. Only the tenns even in n contribute to the S-matrix expansion for the electron-phonon
interaction:
G(p, t - t') = dO)(p, t - t') + (~~)3 J~oo dt l J~oo dt2
X L MqIMq20(ITAql (t l )Aq/t2)1}0
qlq2
The phonon bracket gives a single-phonon Green's function:
(2.90)
The electron bracket, unfortunately, has six possible combinations of pairings. The six tenns
are evaluated using the fact that q2 = -ql' Wick's theorem gives the result
A At A At A At
0(1 TCpcr(t)Ckl +ql ,itl )Ckl ,itl )Ck2 +q2's' (t2)Ck2S' (t2)Cpcr(t')1 )0
A At A At
= 0(I TCpcr(t)Ckl +ql itl) I} 00 (I TCkl ,S(tl)Ck2+q2's' (t2) I}o
A At
x 0(ITCk2s,(t2)Cpcr(t')1}0
A At A At
+ 0(I TCpcr(t)Ck2+q2,s,(t2)1)oo (I TCkl ,s(tI )Cp,cr(t')I)o
A At
X 0(ITCk2S,(t2)Ckl+qlitl)l)O
.A At A At
+ 0(I TCpcr(t)Ckl +q,itl)l)oo (I TCkl ,.(tl)Cpcr(t')l}o
A t A
X 0(ITCk2+q2,s'(t2)Ck2s,(t2)1)o
A At A At
+ 0(I TCpcr(t)Ck2+q2,s,(t2) 1)00 (I TCk2,At2)Cpcr(t') 1)0
A t A
X o(ITCkl+qlitl)Cklitl)l}o
A At At A
+ 0(I TCpcr(t)Cpcr(t') I} 00 (ICkl +ql itl)Ckl ,s(tl)l}o
A t A
X 0(ITCk2+q2,At2)Ckj(t2)1}0
A At, A At
- o(ITCpcr(t)Cpcr(t ) I} 00 (I TCkl,.(tl)Ck2+q2,S,(t2)1)o
A At
xo (ITCk2At2)Ckl+qlitl)l)o
Sec. 2.5 • Feynman Diagrams 81
Each bracket is either G(O) functions with different time arguments, or else a number operator
nF at equal times. The six terms, taken in the same order, give
i3 0p=k2=k l+ql Os='=crdO)(p, t - tl)G(O)(p - ql' tl - t2 )dO)(p, t2 - t')
+ i2 0p=k l=k2-ql 0S='=cr G(O) (p, t - t2 )G(O)(p, +ql' t2 - tl)G(O)(P, tl - t')
+ FOQl=oop=k l os=crnF(~k)G(O)(p, t - tl)G(O)(p, tl - t')
+ FOQl=Oop=k2oS'=crnF(~k)G(O)(p, t - t2 )dO)(p, t2 - t')
+ iOQl=OoQ2=OnF(~k)nF(~k)dO)(p, t - t')
- j3 Ok l=k2-Ql 0s'=s G(O) (p, t - t')G(O)(k l , tl - t2 )dO)(k l + ql' t2 - t l ) (2.91)
2.5. FEYNMAN DIAGRAMS
Feynman introduced the idea of representing the kind of terms in (2.91) by drawings.
These drawings, called diagrams, are extremely useful for providing an insight into the
physical process which these terms represent. These diagrams can be drawn both for the
Green's function depending on time G(p, t) as well as for functions which are Fourier-
transformed and depend on energy G(p, E).
The diagrams in time are drawn by representing the electron Green's function
G(O)(p, t - t') by a solid line which goes from t' to t, as shown in Fig. 2.1. An arrow is often
included to represent the direction. The arrow is mostly for convenience, and it does not imply
or require that t > t'. The phonon Green's function is represented by a dashed line. It does not
have a directional arrow because
D(O) ( q, t - t') = D(O) ( -q, t' - t) (2.92)
and the sign of q is irrelevant. Phonons can be viewed as going in either direction in time.
Next consider the diagram for the factor
(2.93)
The occupation number nF is drawn as a solid line which loops and represents an electron line
which starts and ends at the same point in time. By using these rules, the diagrams are
constructed which represent each of the six terms in the S-matrix expansion (2.91). These six
p
d01(p.t.t' ) >
t' t
dO'(1. t -t ') Ii
t'-----"t
<cpt(t)cpt,> Op
t
V1 11
FIGURE 2.1
terms are shown in Fig. 2.2, and the diagrams (a) to (t) are the six terms in (2.91) in the same
order. Each term has the phonon line connecting the times tl and t 2.
In Fig. 2.2 the terms (c), (d), and (e) are zero. They vanish because they exist only if the
phonon wave vector q is zero, but there are no phonons with q = O. A phonon with q = 0 is
either a translation of the crystal or a permanent strain, and neither of these is meant to be in
the Hamiltonian. The sum over q in (2.88) should exclude the q = 0 term.
The two terms (a) and (b) in Fig. 2.2 are not zero. They are the contributions of primary
interest. By using the results of (2.91), the contributions of the terms (a) and (b) are
~ J~oo dtl J~oo dt2 ~ IMlD(O)(q, tl - t2)
x G(O)(p, t - t 1)G(O)(P - q, tl - t 2)G(O)(P, t2 - t')
(2.94)
The two drawings (a) and (b) look alike. They differ only in the labeling of the variables
t1 , t2 , q. But these are variables of integration in (2.94) and may be relabeled which shows the
two terms are equal. Only one term need be written, and the factor of 1/2! in front is
eliminated.
The term (t) in Fig. 2.2 is
dO)(p, t - t')F1 (2.95)
Fl = -~Joo
2. -00
dtl Joo -00
dt2 L
kq
IMlD(O)(q, tl - t2)dO)(k, tl - t2)
(2.96)
(a) (b)
, ~,
, ............
I \ I \
p
(c)
Ok2
(d)
Q k1
\ Q1=O
IQ1=O
1 ~
\
~
P P
P P
k1 Ok2 k1+Q1
(e) Q-Q;D- (1) E!1~

k1
~ ~
p P
FIGURE 2.2
Sec. 2.6 • Vacuum Polarization Graphs 83
(b)
<=>.
I"
."".--- .......
.... -... ,\ Q '
I , " I
) ') / ) I)! ) , )
FIGURE 2.3
This drawing has the property that part of it is not topologically connected to the Green's
function line G(O)(p,t - t'). Diagrams in which all parts are not connected are called
disconnected diagrams. For example, in Fig. 2.3, part (a) is disconnected, while part (b) is
connected. The disconnected parts, as in (2.95), provide just a multiplicative constant such as
FI which multiplies the contribution from the connected parts.
2.6. VACUUM POLARIZATION GRAPHS
Now consider the factor which has been ignored to this point:
Again consider the electron-phonon interaction and evaluate the term n = 2. The n = 1 term
vanishes as it did for the Green's function expansion,
(2.97)
where
o(ITV(tl)V(tz)l}o = L MqlMq2 o(ITA q, (t l )A q2 (tz)1}0

qlq2
By using Wick's theorem,

A A • (0)
0(ITAq,(tl)Aq2(tZ)1}0 = lOql+qP (ql' tl - tz )
At A At A
0(1 TCk, +ql ,sCtl )CklsCtl)Ck2-ql.s,(tZ)Ck2,Atz)I)0
= 0ql nF(~k)nF(~k) + 0k l=k2-ql G(O)(k l , tl - tz)G(O)(k l + ql' tz - t l ) (2.98)
The Feynman diagrams for the two terms in Eq. (2.98) are shown in Figs. 2.4(a) and 2.4(b).
The (a) term is zero because there are no q = 0 phonons. The (b) term is nonzero and gives a
contribution
(2.99)
O '1'-0
--~-o
(0)
FIGURE 2.4
e (b)
where F J is defined in (2.96). The constant F J appears whenever the closed bubble of
Fig. 2.4(b) occurs, regardless of whether the term arises in the disconnected diagrams of
G(p, t - t') or in the expansion of o(IS(oo, -(01)0'
The terms in the series for o(IS(oo, -(01)0 are called vacuum polarization terms. Some
terms for n = 4, where there are two phonon lines, are shown in Fig. 2.5. Each of these
diagrams or terms represents a constant Fj which one can evaluate by doing the required time
and wave vector integrals. The constant o(IS(oo, -(01)0 could be evaluated by computing all
the Fj and then summing them (Fo = 1):
00
o(IS(oo, -(01)0 = "L. Fj (2.100)

j=O
This procedure is unnecessary because of a cancellation theorem.

The next theorem also simplifies the calculation of the Green's function expansion
(2.76). This theorem is that the vacuum polarization diagrams exactly cancel the disconnected
diagrams in the expansion for G(p, t - t'). The net result is that in calculating G(p, t - t') one
needs only to evaluate the connected diagrams. The other contributions, from the discon-
nected diagrams and from o(IS(oo, -(01)0, exactly cancel one another.
This theorem will not be proven, but only explained. Call Gc(p, t - t') the summation of
all connected diagrams, and the basic theorem is that
The Green's function (2.43) is just the summation of all the connected diagrams:
G(p, t - t') = Gc(p, t - t') (2.101)
The proof of this theorem is just a counting problem. One must convince oneself that each
connected diagram has, in higher-order terms in the S-matrix expansion, all disconnected
parts which exactly add up to o(IS(oo, -(01)0' For example, the self-energy diagram in Fig.
2.2(a) has in higher order the vacuum polarization terms shown in Fig. 2.6. The summation of
all these terms, to all orders, is just the factor o(IS( 00, -(01)0' The important point is that each
disconnected part is just a constant factor Fj. This theorem is very convenient, since it states
that one can just ignore the disconnected diagrams. They do not need to be calculated. It is
just as well, since when they are evaluated they often tum out to be infinity. In fact, it is easy
FIGURE 2.5 Vacuum polarization graphs

Sec. 2.6 • Vacuum Polarization Graphs 85
~
,-,
G>
,0
t:::::>'
®
,I
,,-,
' \
,;-,,
, \ ,
,,,, .....\
,'
, ,-,,
.
, "
,--
"" ..,
+ + + + (
§
,'
,,-- " ,
[ I + 0 + G + + CD
1
FIGURE 2.6
to show that they are infinity. The disconnected term FI is defined in (2.96). The integrand is
only a functionj(t l - t2 ) of tl - t2 . Change integration variables to
and Eq. (2.96) becomes
(2.102)
The important point is that there is no dependence on s, so
Joo
-00
ds = 00 (2.103)
One can show that each disconnected part has an "extra" time integral and has the same
infinity.
It has been shown that the one-particle Green's functions consist of just connected
diagrams:
(-it Joo Joo
,
G(p, t - t) = -l fo ---;:;!
. 00
-00 dtl . .. -00 dtn
x o(ITCpcr(tl)C~cr(t')V(tI)'" V(tn)l)o (connected)

The next step is to get rid of the 1/n! factor. It is eliminated because there are just n! terms
exactly alike in each bracket of the nth term in the expansion.
Considering only different terms gives the result
G(p, t - t') = -i E( -it J: dtl ··· J~oo dtno(ITCpcr(t)C~cr(t')

x i\t l )· .. V(tn)l)o (different connected) (2.104)
The obvious question is then: how can you tell when terms are different? Usually one can tell
by inspection, although sometimes terms must be examined carefully. In Fig. 2.2, terms (a)
and (b) are the same and provide the 2 = 2! necessary for this n = 2 term. Similarly, (c) and
(e) are identical. For example, (a) and (b) differ only in the variables t\> t2 and ql' But these
are dummy variables of integration and so may be relabeled, respectively, to t2 , tl and -ql'
Then the (a) term is obviously the same as (b).
The next terms in the electron-phonon expansion have n = 4, which are diagrams with
two phonons. Here each different connected diagram is found 4! = 24 times.
2.7. DYSON'S EQUATION
The Green's function of energy is defined by taking the usual Fourier transform with
respect to the time variable:
G(p, E) = J~oo dteiE(t-t')G(p, t - t') (2.105)
This time integral has already been evaluated for the unperturbed Green's function with the
following results. For a single particle in a band, the result from (2.59) is
aC0)( E) _ 1
p, - E - e +io
p
For a fermion in a degenerate electron gas, the result (2.64) is
aC0)(p, E) = 1.
E-ep + lOp
The Fourier transform in time is applied to each term in the S-matrix summation:
G(p, E) = -i n~(_i)n J~oo dteiE(t-t') J~oo dt1··· J~oo dtn
x 0(1 TCpcit)C:cr(t')V{t1) ... V(tn)l}o (different connected) (2.106)
To see what sort of terms develop, consider the example of the electron-phonon interaction.
The first two terms [Figs. 2.2(c) and 2.2(e) are zero] are G(O) plus the self-energy term in Fig.
2.2(a):
G(p, E) = aC°)(p, E) + (-ii ~ IMq 12 J: dteiE(t-t') J~oo dt1 J~oo dt2
x aC°)(p, t - t1)G(O)(p - q, t1 - t2)
x aC°)(p, t2 - t')D(O)(q, t1 - t2) (2.107)
The phonon Green's function of energy is defined the same way:
D(q, 00) = J~oo dteitffiD(q, t)
D(q, t) = Joo -2
-00
dOl
1t
.
e-1tffiD(q, oo) (2.108)
Using the unperturbed phonon Green's function in (2.107),
(2.109)
Sec.2.7 • Dyson's Equation 87
and the remaining time integrals are easy:
J~oo dtei(f-f1)EdO)(p, t - t l ) J~oo dtlei(fl-fz)(E-ro)dO)(p - q, tl - t2)
x J~oo dt2ei(f2 -t')EdO)(p, t2 - t') = dO)(p, E)2dO)(p - q, E - (0)
The first two terms are

(2.110)
where the self-energy of the electron due to one-phonon processes is
(2.111)
This self-energy will be evaluated in Chapter 7.

The electron-phonon interaction has four connected diagrams in n = 4; they are self-
energy diagrams with two phonons and are shown in Fig. 2.7. These four terms give,
respectively, the contribution to the Green's function series (2.104):
J~oo dtl J~oo dt2 J~oo dt3 J~oo dt4G(O)(p, t - tl)(~ IM M ,1 D(O)(q, tl - q q
2 t2)
x D(O)(q', t3 - t4){[dO)(p + q, tl - t2)G(°l(p, t2 - t3)dO)(p + q', t3 - t4)
+ dO)(p + q, tl - t3)dO)(p + q + q', t3 - t2)G(°l(p + q', t2 - t4)]dO)(p, t4 - t')
+ dO)(p + q, tl - t3)dO)(p + q + q', t3 - t4)dO)(p + q, t4 - t2)d°l(p, t2 - t')}
+ L IMq I4 D(O)(q, tl - t2)D(O)(q, t3 - t4)G(O)(p + q, tl - t4)G(O)(p, t4 - t')

q
x L G(O)(k, t2 - t3)G(O)(k + q, t3 - t2))

k.cr
Figure 2.7 shows the labeling of electron and phonon Green's functions and the time label of
each vertex. The Fourier transform of these terms is taken to give their contribution to the
Green's function of energy,
dO)(p, El~:{l)(p, E)2 + dO)(p, E)[L(2a)(p, E) + L(2b)(p, E) + L(2c)(p, E)]
so that combining this result with the earlier one in (2.110) gives
(2.112)
..
/--.. e. ,----, fj , t·Ot,
'.! . '
/--,"\__ --'l:
fj fj'
, '\ t i, "., I,
.' . ' ' ' . 'IP
~t, T t. t. , t. t, t. t. t.
.' J ) I J )
,. t, ,.., t. f'
fMI ~.fj'
(0) (b) (e)
FIGURE 2.7 Two-phonon Feynman diagrams in the self-energy of the electron.

The first term, which comes from Fig. 2.7(a), contains three unperturbed Green's functions
G(O) and two one-phonon self-energy terms ~(1). Each of these factors can be associated with
a piece of Fig. 2.7(a), as illustrated in Fig. 2.8. Similarly, the last three terms in Fig. 2.7
contain the three diagrams which represent the two-phonon self-energy terms:
~(2a)(p, E) + ~(2b)(p, E)
= Joo dOl Joo dro'L IMqMq' 12 D(O) ( q, 00 )D(O)(q', 00')
-00 2n -00 2n qq'
x dO)(p + q, E + ro)dO)(p + q + q', E + 00 + 00')

x [dO)(p + q', E + 00') + G(O)(p + q, E + (0)] (2.113)
x J OO dOl'
- L G(O)(k, ro')G(O)(k + q, 00 + 00')
-00 2n kG
Dyson s equation is obtained by formally summing the series in (2.112),

G(O)(p, E)
G(p, E) = 1 _ G(O)(p, E)~(p, E) (2.114)
where the total self-energy ~(p, E) is the summation of all different self-energy contributions:
~(p, E) = L ~U)(p, E) (2.115)

j
So far the example of the electron-phonon interaction has yielded the following four terms
for the self-energy:
~(p, E) = ~(l)(p, E) + ~(2a)(p, E) + ~(2b)(p, E) + ~(2c)(p, E) + ...

Contributions such as Fig. 2.7(a) which contain (~(1»2 do not mean that ~ contains (~(1)i.
The Green's function expansion contains terms in (~(1)i because the expansion for G is
G(O)
G= = dO) + dO)2~ + dO)3~2 + G(0)4~3 + ...
1- G(O)~
Successive terms contain higher powers of each ~U) plus cross terms. Dyson's equation is
really a theorem which states that one may sum the series of self-energy terms which develops
in higher order, and the form (2.114) is obtained with each distinct contribution to ~
occurring just once.
G10I l:'Q G10I l:IQ d(/j

• > ' ==>----
til ,--f.
::::::::::::~>=~
FIGURE 2.8 These four tenns give, respectively, the contribution to the Green's function series (2.104):
Sec.2.7 • Dyson's Equation 89
The derivation of Dyson's equation has been rather complicated and consisted of a large
number of steps and theorems. Each step has to be understood before the final result is
understood. Yet the final result-Dyson's equation-has achieved a great simplification. It
states that the exact Green's function is obtained from (2.114) by just calculating the self-
energy L(p, E). The self-energy is a summation of an infinite number of distinct diagrams.
This method is useful only if L(p, E) can be approximated by the lowest few terms in the
series. Alternately, sometimes one can sum subsets of diagrams in the series. However, except
in a few rare cases, it is impossible to get L(p, E) exactly, and one must be content with an
approximate result. If the approximate result is not a very good approximation, one should not
try to solve the problem in this fashion. Some of the alternate methods will be developed in
subsequent chapters. But basically one should realize that Dyson's equation is usually useful
only in weak coupling theory where the perturbation is sufficiently weak that an adequate
approximation is obtained with a few terms in L(p, E). Later chapters discuss some strong
coupling theories which use Dyson's equation.
A great simplification has been achieved. It is only necessary to evaluate a few self-
energy diagrams in L(p, E). If they are sufficient, the calculation is finished. If they are not,
then discard the results and try something else. The formidable-looking series (2.76), which
served as the starting point, has been reduced to the evaluation of a few terms.
Dyson's equation is often written in a slightly different but equivalent form. It is obtained
by using the algebraic form for G(O). For one electron in a band,
dO)(p, E) = 1.
E -!:p + 10
1
G(p, E) =E_ !:p + io - L(p, E) (2.116)
When the electron is in a Fermi sea at zero temperature,
dO)(p, E) = l. (2.117)
E -!:p + lOp
1
G(p,E) = . (2.118)
E- !:p + lOp - L(p, E)
One aspect of this result deserves special mention. The infinitesimal part op switches sign
depending on whether !:p > 11 or !:p < 11, where 11 is the chemical potential. Taking the
imaginary part of (2.117) gives E = !:p. Therefore ImG(O)(p, E) changes sign depending on
whether E > 11 or E < 11.
The self-energy has real (LR) and imaginary (Lj) parts which are written as
(2.119)
Later it is shown that L j also switches sign at E = 11 in the same manner:
Lj(p, E) < 0, E > 11
Lj(p,E) > 0, E < 11 (2.120)
The switching of signs of op was caused by the distinction between electron excitations with
~p > 0 and hole excitations with ~p < o. This distinction is maintained even when L(p, E) is
included, i.e., in the presence of interactions. The one-electron Green's function (2.116) has
the feature that :!:[ < 0, so the imaginary part of the denominator always has the same sign.
The electron self-energy is sometimes called a mass operator.
The phonon Green's function has the same type of Dyson equation:
D(O)(q,O)
D(q,O) = 1 _ D(O)(q, O)1t(q, 0) (2.121)
where the self-energy function for phonons is 1t(q, 0). This equation may also be written in an
alternate form by utilizing the following result for D(O):
20)
D(O)(q,O) = q (2.122)
0)2 - O)ij + io
20)
D(q,O) = q (2.123)
ro2 - O)ij + iO - 20)q1t(q, 0)
The phonon self-energy term 1t(q, 0) is sometimes called apo[arization operator. This name
is quite descriptive, since the self-energy effects arise from the phonons causing polarization
in the medium.
The real and imaginary parts of the self-energies :!: and 1t each have interpretations. The
imaginary part, :!:[ or 1tb is interpreted as causing the damping of the particle motion. They
are related to the mean free path of the excitation or its energy and momentum uncertainty.
The real parts are actual energy shifts of the excitation, which may also change its dynamical
motion. The excitation may alter its effective mass or group velocity because of the self-
energy contributions.
2.S. RULES FOR CONSTRUCTING DIAGRAMS
It is certainly worthwhile for the student to evaluate some self-energy diagrams by the
method outlined above. One should expand the S matrix, decide which terms are connected
and which are zero, and finally obtain the self-energies by a Fourier transform. But this
laborious procedure can be avoided because the self-energy diagrams can be written down
directly by following a few simple rules. That is, self-energy expressions such as (2.111) and
(2.113) are easily written down in the form shown in these equations. The evaluation of the
wave vector integrals in these expressions remains a formidable task. These rules are as
follows:
1. Draw the Feynman diagram for the self-energy term, with all phonon, Coulomb, and
electron lines.
2. For each electron line, introduce the following Green's function:
(0) 0<xf3
G<xp(p, E) = E -
Ep
+ 1'0P (2.124)
The o<xP conserves the spin index and indicates that the electron line must have the
same spin at both ends of the propagator line. This feature is important in spin
problems. The factor op is 0 for one electron in a band and 0 sgn (~p) for degenerate
Fermi systems, i.e., those with a Fermi sea at zero temperature.
Sec. 2.8 • Rules for Constructing Diagrams 91
3. For each phonon line, introduce the following phonon propagator:

2co
D(O)(q co) = q (2.125)
, co2 - co q2 + if>
Also add a factor of IMql2 for each phonon Green's function, where Mq is the matrix
element for the electron-phonon interaction.
4. Add a Coulomb potential Vq = 4rcrrj q 2 for each Coulomb interaction. Note that the
Coulomb line is always drawn as a wiggly vertical line. The Coulomb interaction is
regarded as happening instantaneously in time, and time flows horizontally, from left
to right, in our diagrams. One could, of course, have time flow upwards and draw the
Coulomb interactions as horizontal wiggly lines.
5. Conserve energy and momentum at each vertex. Each electron line, phonon line, and
Coulomb line have their variables labeled to conform with this rule.
6. Sum over internal degrees of freedom: momentum, energy, and spin. If one is
calculating a self-energy term 1:(p, E) of the electron, then all momentum and
energies except p and E are internal and must be summed over.
7. Finally, multiply the result by the factor
(2.126)
where F is the number of closed fermion loops. The index m is chosen as follows:
(a) For electron self-energies, m is the number of internal phonon and Coulomb
lines. For example, m = 1 for (2.111), and m = 2 for (2.113).
(b) For phonon self-energies, m is one-half the number of vertices. The spin of the
particle is S, and the factor (2S + I) is from the summation over spin quantum
number ms' Electrons have 2S + I = 2. The factor (2rc)4m assumes taking the
limit v -+ 00, so that the wave vector summations are integrals. For box
normalization in a finite volume v, the factor is
'm
IFF
(2rcv)m(-I) (2S+ 1) (2.127)
and then wave vector summations are discrete summations.
The photon Green's functions are discussed in the last section. But it seems tidy to
present the rules for constructing diagrams with photons at this point, so that all the rules are
together. One draws photon lines between lines which represent charged particles. The photon
lines are usually dotted and usually are represented just like phonon lines. Charged particles
interact with the photons through two terms in the interaction Hamiltonian. The first is the
j . A term. For free particles, this has the form
(2.128)
This interaction brings us to the next rule for constructing diagrams.

8. For each photon line which interacts with particles through the j . A interaction, insert
a factor
~ ~:Xk + !q)JlDJlvCq, oo)(k' + !q)v

m JlV
(2.129)
where DJlvCq, (0) is the photon Green's function and k and k' are the wave vectors of
particles scattered at the two vertices. The other possible interaction of a charged
particle with photons occurs through the term
e2 2 e2
-, LA(ri ) = -2 L p(q)AJl(k)Aiq - k)
2mc..- i mqkJl
In Sec. 4.5 it is shown that this interaction contributes a self-energy term of e2no/m to
the self-energy of the photon, where no is the density of charged particles.
Other texts often define the phonon Green's function differently than the form which is
selected here. However, all of these cases have the final product of the vertex and Green's
function as
2OOqlMql2
(2.130)
002 - oo~ + io
For example, one choice has a Green's function of the form
002
D(O)(q, (0) = q (2.131)
002 - 2
00q + io
Then, in the rules for constructing diagrams, one multiplies by the factor 2lMql2/ooq for each
vertex pair. Of course, these steps give the same result.
Now examine some examples. Electron-phonon examples are given previously in
(2.111) and (2.113). Some phonon self-energies are shown in Fig. 2.9(a). The first example is
from the electron-phonon interaction, and the self-energy contribution is a closed fermion
loop, so that F = 1 in rule 7. This contribution is
n(q, (0) = IMiP(I)(q, (0)
p(1)(q, (0) = (;:;4 J J

dE d3pO<0)(p, E)G(O)(p + q, E + (0) (2.132)
(bl
£ P
f
dl
f'
fJ'f
f
FIGURE 2.9
Sec. 2.8 • Rules for Constructing Diagrams 93
The other phonon self-energy contribution arises from the lattice anharmonicity, which leads
to interaction terms in the ion Hamiltonian proportional to the third power of the phonon
displacement:
v = L V;Qi = L Mq,q,AqAq,A_q_q'
i qq'
Here the self-energy term is
n(1)(q,O)) = (2~)4 JdO)' Jd q'lM

3 q ,q,1 2 D(0)(q', O)')D(O)(q + q', 0) + 0)') (2.133)
Next consider terms in the electron self-energy arising from electron-electron interactions.
The self-energy term in Fig. 2.9(b) is called the unscreened exchange energy and is a very
important contribution to the electron's energy. Its self-energy is
~xCp, E) = ~JdO) Jd 3qVq G(0)(P + q, E + 0)) (2.134)

(2n)
This result can be simplified immediately using an important identity:
I'JdO)
2n G (p + q, E
(0)
+ 0)) -_ -nF(~p+q) (2.135)
The identity is proved by replacing the Green's function by its Fourier definition,
i J -dO)JOO dteit(E+ro)dO)(p + q, t) (2.136)

2n -00
and then inverting the order of integrations. The frequency integral gives a delta function in
time,
JdO)
2n
eit(E+ro) = oCt) (2.137)
and
i J~: dO)(p + q, E + 0)) = iG(O)(p + q, t = 0)
The right-hand side is ambiguous. Since the Green's function is a time-ordered product,
different results are obtained depending upon whether t = 0 is approached from plus time or
negative time,
G(O)(p + q, t -+ 0+) = -i(Cp+qC~+q) = -i[1 - nF(~p+q)]
G(O)(p + q, t -+ 0-) = -i(C~+qCp+q) = inF(~p+q)

In the identity (2.135), the t = 0- result is chosen. It was stated in Sec. 2.5 that equal time
operators were to be taken in this order. These steps prove the identity (2.135). The self-
energy (2.134) may be written in the simple form
~xCp) =- v
1
~ VqnF(~p+q) = -
J d3q
(2n)3 VqnF(~p+q) (2.138)
where the sUbscript x denotes exchange. The self-energy is no longer a function of the energy
E of the particle, since all E dependence has vanished from the right-hand side. This exchange
energy is easily evaluated at zero temperature (see Problem 6 at the end of the chapter).
The next electron self-energy from electron-electron interactions is shown in Fig. 2.9(c):
~H(p' E) = -2i JdE'Jd 3p'V _ dO)(p' E')

(2n)4 q-O,
(2.l39)
where the SUbscript "If" denotes "Hartree". The self-energy depends on neither p nor E. The
identity (2.l35) may be used again to produce the result
~H = 2vq=o LnF(~p') = vq=oNe (2.140)
p'
where Ne is the number of electrons. Of course, when Vq = 4n~ jq2, then the limit q ~ 0
gives vq=o equal to infinity. This term is the unscreened Coulomb energy from one electron
interacting with all the other electrons in our system. This potential energy is truly a large
number, which becomes infinity in the limit of an infinite system. But there must be an equal
amount of positive charge in the system, and the electron interaction with the positive charge
yields another large number which cancels the present divergence. The Hartree energy is
defined as the net interaction energy of an electron from both of the negative and positive
charge sources. It is zero in the jellium model of a metal, but is nonzero for actual systems
composed of ions and conduction electrons.
The third electron self-energy diagram is shown in Fig. 2.9(d). Its evaluation yields
~(p, E) = J Jd3qv~P(I)(q, oo)dO)(p + q, E + (0)

(2:)4 doo (2.141)
The factor p(1)(q, (0) is given in (2.l32). It corresponds to the closed fermion loop. This
polarization diagram occurs frequently, is very important, and will be evaluated in Chapter 5.
Finally, this section is ended with a short lecture on the way to draw diagrams. The types
of Feynman diagrams are very intimately connected with the types of terms in the Hamil-
tonian. Figures 2.1O(a) and 2.l0(b) show the two types of electron-phonon vertices and the
terms they correspond with in the Hamiltonian. In each case, an electron is destroyed in state
k (incoming arrow) and created in k + q (outgoing arrow). All of the electron-phonon
diagrams involve only diagrams which can be constructed from these basic building blocks.
Diagrams are not drawn in which one phonon makes two electrons, as shown in Fig. 2.9(c),
since such terms do not occur in the Hamiltonian. Later such terms are encountered for other
problems. Also, terms are not included in which two phonons are emitted while scattering one
electron. These are anharmonic terms in the electron-phonon interaction. However, it has
recently been suggested that they are important in some materials, so that in fact people do
draw diagrams using them. The important point is that one must start with a Hamiltonian,
decide which kind of vertex processes are permitted by each interaction term, and draw
Feynman diagrams using only these basic building blocks.
2.9. TIME-LOOP 5 MATRIX
The S matrix is defined in Sec. 2.2. The time in (2.41) is taken over the interval
(-00,00). The state at t = -00 is well defined as the ground state of the noninteracting
system. The interactions are turned on slowly. At t ,..., 0 the fully interacting ground state is
Sec. 2.9 • Time-Loop S Matrix 95
_--f
~~~f'
FIGURE 2.10
"'(0) = S(O, -00 )<Po. In condensed matter physics the state at t ~ 00 must be defined
carefully. If the interactions remain on, then this state is not well described by the non-
interacting ground state <Po. Alternately, one could require that the interactions tum off at
large times, which returns the system to the ground state <Po.
Schwinger (1961) suggested another method of handling the asymptotic limit t ~ 00.
He proposed that the time integral in the S matrix has two pieces: one goes from (-00, 't)
while the second goes from ('t, -00). Eventually 't ~ 00. The integration path is a time loop,
which starts and ends at t = -00. The advantage of this method is that one starts and ends the
S-matrix expansion with a known state "'(-00) = <Po. Usually it is the only ground state one
knows exactly.
For equilibrium phenomena the time loop method of evaluating the S matrix gives
results that are identical to the other methods such as those described earlier in this chapter. A
small advantage of the time-loop is that the formalism has a sounder philosophical basis since
the state "'(00) is avoided. However, the main advantage of the time-loop method is in
describing nonequilibrium phenomena using Green's functions. Nonequilibrium theory is
entirely based upon this formalism, or equivalent methods. The equation of motion for the
Green's function can be cast into the form of a quantum Boltzmann equation for transport
theory. This application will be developed in Chapter 8.
A disadvantage of the time-loop method is that it employs six different Green's func-
tions. They are discussed next, as a preliminary to the expansion of the S matrix. Two of these
Green's functions, called retarded and advanced, are also needed for the later discussion of
equilibrium theory at nonzero temperature.
The need for six Green's functions is explained by considering the time-loop expression
for the S matrix:
S(-oo, -00) = Ts eXP[-iJ dSIV(SI)] (2.142)

loop
The integration path is the time loop shown in Fig. 2.11. The variable SI goes (-00, 't) and
then ('tl - 00). The operator Ts orders along the entire loop, with earliest values of SI
occurring first. In expanding the S matrix, Green's functions are encountered of the form
(2.143)
Ifboth SI and S2 are on the top, or outward leg, of the loop then s-ordering is identical to time-
ordering. However, if both SI and S2 are on the lower, or backward leg, of the loop then s-
ordering is the opposite of time-ordering. This case is called anti-time-ordering. Another case
is when SI and S2 are on different legs. Then they are automatically ordered, and indepen-
dently of the values of the S arguments.
Besides the time-ordered Green's function, it is convenient to also define one that is anti-
time-ordered, and others that have no time-ordering. These different cases give rise to the four
Green's functions. The other two are called retarded and advanced. They are linear combi-
nations of these four.
2.9.1. Six Green's Functions

For the time-loop expansion, it is necessary to define six different Green's functions. It is
possible to employ fewer than six since they are not independent, but using six simplifies the
notation. They are all correlation functions which relate the field operator \j!(XI) of the particle
at one point XI - (rl' t l ) in spacetime to the conjugate field operator \j!t (x2) at another point
X2 = (r2' t2). The six functions are the advanced Gadv , retarded Gret , time-ordered Gt , anti-
time-ordered G t and G<, G> which have no name:
G>(XI,X2) = -i(\j!(XI)\j!t(X2))
G«xI' x2) = i(\j!t (x2)\j!(xI))
GlXI' X2) = e(tl - t2)G> (XI ,X2) + e(t2 - tl)G«XI' X2)
Gt(XI, X2) = e(t2 - tl)G>(XI' X2) + e(tl - t2)G«XI' X2)
Gret(XI, x2) = Gt - G< = G> - Gt
Gadv(XI, X2) = Gt - G> = G< - G t (2.144)
The brackets I) and (I have the same meaning of Sec. 2.3 as the ground state of the interacting
system. The time-ordered Green's function is the same one in (2.43). Here it has been written
in the field operator representation. For homogeneous systems in equilibrium, the Green's
-_e_:__ )T"
FIGURE 2.11 The time-loop integration path in the S matrix. Eventually 't -+ 00.
functions depend only upon the difference of their arguments (XI' Xz) = (XI - xz). Then the
most simple, and useful, quantities are the Fourier transfonns of these quantities:
(2.145)
where the symbol G represents any of the six functions. Explicit expressions for these
quantities are given below.
Often the Hamiltonian H can be solved exactly in tenns of eigenfunctions (h(r l ) and
eigenvalues CA' Two examples are the electron in a magnetic field or a free particle. Then it is
useful to have the expressions for the Green's functions in tenns of these eigenfunctions. They
are derived by expanding the field operators in tenns of the eigenfunctions and creation ct
and destruction CA operators:
q,t(XI) = LCtq,~(rl)eiE,t (2.146)

A
The Green's functions in (2.144) are evaluated with the occupation factor n A= (ctcA) and
t= tl - t z. At zero temperature nA= 0( -~A) is a step function that is zero or one depending
upon whether ~A = cA- ~ is positive or negative:
(2.147)
The above fonnulas are also valid in equilibrium at nonzero temperatures if nA=
1/[exp(~~A) + 1] is the thennodynamic average m = l/kBT) of the occupation number.
The starting point for any calculation, at least conceptually, is the behavior of the Green's
functions for systems without interactions. Then the wave functions are those for plane wave
or noninteracting Bloch states, if such can be defined. The quantum number Iv becomes the
wave vector k, and a spin index cr, which is usually not written. The eigenvalue combination
is q,A(rl)q,~(rZ) = exp[ik . (rl - rz)]/v. The superscript "(0)" or the subscript "0" on the
Green's functions means those for a noninteracting system in equilibrium. Fourier-trans-
forming the r variable to k as in (2.145) gives the free-particle Green's functions G(O)(k, t).
For fermions of band energy Ek and occupation number nk = nF(Ek - J.l) they are
G}O)(k, t) = -i[E>(t) - nk]e- hV
G(O)(k, t) = -i[E>( -t) - nk]e- iBkt
Gret(O)(k, t) = _iE>(t)e- iBkt
Gadv (O)(k, t) = iE>( _t)e- iBkt (2.148)
The t variable can be Fourier transformed, which gives the noninteracting Green's function of
energy E; the quantity 8 is infinitesimal
Go(k, E) = -2n:i[1 - nk]8(E - Ek)
GO'(k, E) = 2n:in k 8(E - Ek)
(0) 1
Gret (k, E) = E _ Ek + i8
(0) 1
Gadv (k, E) =E - Ek-/
'8
=G + G< =
dO)(k E)
t' ret
(0)
° E-
1
Ek + i8k
dO)(k E)
t'
= G<°_ G
adv
(0) =E -1
.S:
(2.149)
- Ek -!Uk
The time-ordered function G}O) is exactly the same one in (2.64). Note the two kinds of
infinitesimal deltas: 8 is always positive, while 8k is positive for k > kF and negative for
k < kF as defined earlier in (2.64). The retarded functions always have a positive 8, even for
electrons in a partially filled band. The noninteracting advanced Green's function resembles
that of an empty band in (2.59). These two Green's functions could differ as soon as inter-
actions are introduced, since they have different self-energy functions in degenerate Fermi
systems. Also note that expressions such as Gret = Gt - G< are obeyed for interacting and
noninteracting functions They are obeyed for both cases of arguments (k, t) and (k, E).
The above Green's functions are suitable for particles such as electrons, or holes in
semiconductors. Another type of Green's function is needed for boson fields such as phonons
or photons. For phonons let Q(x) be the displacement from equilibrium of the ions in the solid
at position x = (r, t) in spacetime. Displacements are vectors, but treat them as scalars for the
moment. The phonon Green's functions are defined as follows:
D>(Xt, X2) = -i(Q(Xt)Q(X2))
D«xt, x2) = -i(Q(X2)Q(Xt))
+ 0(t2 - tl)D«Xt, X2)

Dt(Xt,X2) = 0(t l - t2)D>(xt , X2)
Dt(xt, X2) = 0(t2 - tt)D>(xt, X2) + 0(t1 - t2)D«xt, x2)

D ret = Dt - D< = 0(tt - t2)[D> - D<]
D adv = D t - D> = -0(t2 - tt)[D> - D<] (2.150)
These expressions are rather similar to those in (2.144) for particles. The main difference is
that D< and D> have the same sign, since no sign change is made when interchanging the
positions of boson operators. Also the displacement operator is Hermitian [Qt = Q], which
introduces some redundancy such as D«xt, X2) = D> (x2 , Xt).
The displacement operators Q are usually represented in terms of phonon raising (at)
and lowering (a) operators. The usual case is to use Aq = a~q + aq instead of Q(x) in the
definition of the phonon Green's function. In this representation, the phonon Green's func-
tions in equilibrium are expressed in terms of the phonon occupation number Nq = (a~aq),
which equals l/[exp(~ffiq) - 1] in thermal equilibrium at nonzero temperature, and equals
zero at T = O. The noninteracting results are:
D>(q, t) = -i[(Nq + l)e- iolql + NqeiCilqt]
D«q, t) = -i[(Nq + l)eiCilqt + Nqe-iCilqt]
Dret(q, t) = -20(t) sin(ffiqt)
Dadv(q, t) = -20( -t) sin(ffiqt)
Dlq, t) = -i{(Nq + 0( _t))eiCilqt + [Nq + 0(t)]e-iCilqt }

Dt(q, t) = -i{(Nq + 0(t))eiCilqt + [Nq + 0(_t)]e-iCilqt } (2.151)
These Green's functions are used in the expansion of the S matrix.
2.9.2. Dyson's Equation

Each of the six Green's functions can be evaluated for an interacting system. They can be
expressed in the interaction representation in terms of the time-loop S matrix. For example,
one of them is
(2.152)
where the S matrix is given in (2.142). This argument has no time-ordering operator since the
order of the two state operators is fixed. However, the S matrix is time-ordered. The time tt is
on the upper loop, while t2 is on the return loop. The Green's function in the above equation
has an S -matrix expansion of the form
(2.153)
where all S integrals are over the time loop. A method is needed to evaluate expressions of this
type.
The potential V is composed of electron, phonon, or photon operators. The operators are
paired using Wick's theorem. Each pair will have a time argument such as G(Si' Sj). If both Si
and Sj are in the top loop, the expression is just the time-ordered Green's function. If they are
both in the return loop, the expression is the anti-time-ordered Green's function. If one S
variable is in the top loop and the other is in the bottom loop, then the Ts operator makes this
expression be either G< or else G>. These relationships are shown in Fig. 2.12. The nth term
in the Green's function expansion is a product of (n + 1) factors, where each factor is one of
the four Green's functions in Fig. 2.12.
It helps to have a simple example. Below is given a potential term V of the type found
for electrons scattering from impurities. The first term in the S-matrix expansion for G< with
this interaction is
G«A, tl - t2) = Go(A, tl - t2)
+ LM<x~ fdSo(ITCr(t2)C~(S)I}oo(ITC!(S)CA,(tI)I}o (2.154)

<x~
The S integral runs over the time loop. For G< remember that tl is on the top loop while t2 is
on the return loop. There are two possibilities. If S is on the top loop, the first Green's function
in the S integral is G< and the second is time-ordered. If S is on the return loop, the first is
anti-time-ordered while the second is G<. The two terms are (s = 1', 't" -+ 00)
G«A, tl - t2) = Go(A, tl - t2 ) + MA,A, f~oo dt'[G~O)(A, tl - t')Go(A, t' - t2 )
- Go(A, tl - t')G-/O)(A, t' - t2 )] (2.155)

A sign change occurred in the last term when the direction of the time integration was
changed from (00, -00) to (-00, 00). The above expression contains only the first two terms
in the S-matrix expansion, which has an infinite number of terms.
In the expansion of the S matrix, each time integral produces one set of terms for the
outward S leg, and another for the return leg. The nth term in the S-matrix expansion produces
2n arrangements. All of these terms can be managed by using a matrix formulation.
FIGURE 2.12 The four Green's functions G(SI, S2) depend upon whether the time variables (SI, s2) are on the
outgoing or return parts of the time loop.
Keldysh (1965) was the first to develop the time-loop theory for solid state applications.
Here the theory is presented using a matrix notation suggested by Craig (1968). He expressed
four of these Green's functions as the elements of a 2 x 2 matrix. The self-energy terms are
also a matrix:
_ [Gt
G=
G> =:~]
-
=:~]
~=
[~t (2.156)
~>
The actual form for the self-energy functions is discussed in the following chapters.
For systems either in equilibrium or nonequilibrium, Dyson's equation is most easily
expressed by using the matrix notation:
G(XI' X2) = GO(XI - X2) + J~oo dx3 J~oo dx4 GO (Xl - X3)
(2.157)
The matrix formulation comes directly from the time loop. Each s integral in the S matrix has
an outward and return leg. Each of these legs gives a different Green's function. So each time
integral generates two Green's functions. Hence the usefulness of the 2 x 2 matrix formalism.
Some simple expressions can be obtained for the Green's functions. First write the above
equation in a notation where the product of two functions implies an integration over the four-
variable dx, which condenses the same equation to
(2.158)
Then the equations are iterated. The following exact expressions are derived for the equations
obeyed by the various Green's functions, using the same product notation:
Gret = Gret(O)[1 + ~retGret]
Gadv = Gadv (0)[1 + ~advGadv]
G> = [I + Gret~ret]Gt[1 + ~advGadv] + Gret~>Gadv
G< = [1 + Gret~ret]Go[1 + ~advGadv] + Gret~<Gadv
Gt = [I + Gret~ret]G}O)[l + ~advGadv] + Gret~tGadv

~t = [I + Gret~ret]~t(O)[1 + ~advGadv] + Gret~tGadv (2.159)
These equations represent multiple integrals in d 3 rdt.

Equations (2.159) simplify considerably for homogeneous systems in steady state where
the arguments of the Green's functions and self-energies depend only upon (Xl - X2)' If the
equations are Fourier transformed, then all quantities depend only upon (k, co). The above
equations are, after Fourier transforming, just algebraic quantities, which are easily solved.
The time-ordered Green's function has the same equation as given in (2.118). Some of the
other Green's functions are given below. The expressions are presented for nonzero
temperature. The zero-temperature cases are found by setting the fermion occupation factor
nF( 0)) = 0(-0)). Also define a quantity A(k, 0)) = - 2ImGret (k, 0)) which is called the
spectral function:
A(k, 0)) =
2r 2' r = -ImLret > 0
cr2 + r
G< = inF(O))A(k, 0))
L> = -2i(l - nF)r
G> = -i(l- nF)A (2.160)
The notation r = - ImL ret will be used throughout, and rather often. The spectral function is
an important quantity, which is discussed further in the next chapter. These formulas are
incomplete without a prescription for calculating the different self-energy functions L<, L ret ,
etc. They will be presented in the next chapter.
2.10. PHOTON GREEN'S FUNCTIONS
The interaction of charges with themselves and with the photon field was discussed in
Sec. 1.5. For spinless particles, this interaction has the Hamiltonian in the nomelativistic limit:
e 1 eiej
= 2 .21- [ Pi --A(ri) ] t
2
H +-22.-+ 2.O)kA a kA a kA (2.161)
i m c i=l} rij kA
The vector potential is given by the expansion
1 1 "ik'r ( ~ ) (2.162)
- All = '" L.. e All k, /\', t
c "Iv kA
(2.163)
The creation and destruction operators aL,

akA obey boson statistics. Each state with wave
vector k and polarization A has its own harmonic oscillator statistics. The vector potential
represents the photon field. Two charges may interact via their common photon field or more
directly through the instantaneous Coulomb interaction eiej/rij' The division of the interaction
between photons and Coulomb field is arbitrary-both interactions come from the same basic
processes. The Hamiltonian (2.l61) is written in the Coulomb gauge where V . A = O.
Another choice of gauge will result in a different division between photon and Coulomb. The
basic forces between the particles are the same regardless of how the gauge is selected.
Sec.2.10 • Photon Green's Functions 103
Now it is time to talk about Green's functions. The scalar potential \jJ(r, t) has a Green's
function. The potential from a point charge is
fI-
\jJ(r) = - (2.164)
r
The factor
(2.165)
which has already been used for the Coulomb interaction, is in fact just the Green's function
of the longitudinal potential. It has no frequency dependence because it is instantaneous.
After all, in the rules for constructing diagrams, in Sec. 2.8, it was treated as a Green's
function. If two electrons interacted by phonons, put in the phonon Green's function and
vertex,
(2.166)
while if two electrons interact by electron-electron interactions, put in the factor Vq where
41t/q2 is regarded as the Green's function and fI- as the vertex. The Coulomb interaction is
treated as a Green's function, on equal footing with the phonon Green's function.
Since Vq is a Green's function, it has a Dyson equation of the usual form:
(2.167)
The factor P(q, co) is the self-energy or polarization operator. Its properties are discussed
extensively in Chapter 5. There is one result which can be obtained with very little effort.
Consider the form of Maxwell's equations in a homogeneous material with an isotropic
dielectric constant E:
VxE=O
EV, E = 41tp
E a 41t
VxB=--E+-j (2.168)
cat c
If these equations are solved in the usual way, further equations are obtained for the scalar and
vector potentials:
\jJ(r) =~
E
JdIr-r' per')
3
r'l
v2A -~~A = _ 41t. (2.169)
c2 at2 c Jt
Here the Coulomb Green's function is
v
V =-.!i (2.170)
q E
If this result is regarded as equivalent to (2.167), it gives a formula for the dielectric function:
E(q, co) = 1 - VqP(q, co) (2.171)
On the left-hand side this result has been generalized to include the case where the dielectric
function depends on both q and co. Equation (2.171) will serve as the definition of the
longitudinal dielectric function. It arises from the self-energy parts of the Coulomb potential.
The Green's function for the vector potential is
DllV(k, t - t') = -iL(TAll(k, lv, t)Av(-k, lv, t')) (2.172)
A
where 11, v are the x, y, z components. The vector potential is defined in (2.163). The sum over
Iv is the sum over the two transverse polarizations of the light, while ~ll are the polarization
vectors for each component. The free propagator at zero temperature is evaluated with the
states 1)0 and 0(1 as the photon vacuum:
D&~(k, t - t') = 21ti L ~/k, Iv)~v( -k, 1v)0(1 T[akAe-iffikt + a~kA eiffikt ]
co k A
(2.173)
and its Fourier transform is
D&~(k, co) = J~oo dteiffiCt-t')D&~(k, t - t') (2.174)
41t
(2.175)
co2 - co~ + i8 ~ ~ll~V
Next, consider the last factor in the above equation LA ~ll~V' The unit tensor is
811V = x.x + yy + ZZ (2.176)
Or, in a coordinate system in the direction of k,
811V = L~ll~v+kk (2.177)

A
The factor LA ~ll~V is the unit dyadic for directions perpendicular to k. The Green's function
is
DCa) = 41t[811V - (kll kv/ k2 )]

(2.178)
llv co 2 _ co~ + i8
This expression is referred to as the photon Green s function. Keep in mind that the inter-
action between two charges occurs via both the scalar and vector potentials. How the inter-
action is divided between scalar and vector potentials is somewhat arbitrary and is determined
by the gauge condition. After making this choice, the word "photon" is assigned to the vector
Sec.2.10 • Photon Green's Functions 105
potential part. This division between photon and Coulomb is arbitrary, and both parts should
really be viewed as arising from photons.
The photon Green's function also obeys a Dyson equation. Since it is a matrix quantity,
one must be careful about the treatment of indices. Previous Dyson equations have only scalar
quantities, where the Green's function is defined as a series expansion. The equation for
electrons is actually of the form
G(p, E) = d°l(p,E) + dO)(p, E)~(p, E)G(p, E) (2.179)
Since all the quantities are scalar functions, this equation could be solved immediately for
G(p, E) which obtains (2.114). But for the photon Green's function, the equation is
D JlV = D (O)
JlV + "L."
1..1)
D(O) D
JlA. 1tA.I) I)v (2.180)
where 1tA.I)(k, co) is the self-energy function, which is now a 3 x 3 matrix. Each term in the
equation is a function of (k, co). However, in homogeneous materials all matrices are of the
form
(2.181)
kky
1t
JlV
= 8
JlV
1t(1) + _Jl_1t(2)
k2 (2.182)
where the factors D(O), D, 1t(I), and 1t(2) are scalars. The self-energy function 1t Jlv (k, co) has the
form shown, which is the most general dependence on k of a matrix function. Now it is
simple to do the summations over the matrix components:
and there is a scalar equation for the photon Dyson's equation
(2.183)
(2.184)
The transverse photon has no dependence in its self-energy on the longitudinal part of the
self-energy 1t(2). The equation for the photon Green's function is (2.184).
Actual solids are periodic, rather than homogeneous, and the matrix form of 1tJlV may be
more complicated than (2.181). The self-energy function 1tJlV will be transverse but not
necessarily perpendicular to k. Then one must start from (2.180) and actually solve for the
various components of DJlv.
In a homogeneous medium with dielectric constant E, (2.169) is the equation obeyed by
the vector potential. The double time derivative is multiplied by E. Since this yields the co2
term in the Green's function, the appropriate Green's function in a medium with dielectric
function c is
4n:[Ollv - kll kv/ k2 ]

D = (2.185)
cO)2 - O)~ + iO
-----=-------,i---,--
Ilv
This equation for D llv is equated with Dyson's equation (2.184), for the dielectric function,
which gives
(2.186)
This result is not very useful for crystals, since there the form cllv(k, 0)) is a tensor, and is not
the scalar function which we have given above. The scalar result is correct in crystals in the
limit of k -+ 0:
. cllv(k, 0)) -+ 0IlV [ 1 - 2n:

hm 4n: (\) (k,O)) ] (2.187)
k~O 0)
It is interesting to note that in the limit of k -+ 0, this transverse dielectric function becomes
exactly equal to the longitudinal one (2.171) at k -+ O. This identity is not obvious yet and
will be proved in Sec. 3.7.
PROBLEMS
1. Show explicitly that
(2.188)
2. For the phonon Green's function D(q, t - t'), let Vet) be the electron-phonon interaction and
evaluate all the n = 2 diagrams. Which are connected, and which are disconnected? Draw the Feynman
graphs for each term.
3. Let Vet) be the electron-phonon interaction in the expansion for the electron Green's function
G(p, t - t'). What are the contributions from the different connected diagrams for n = 4 (two phonons).
Just draw the graphs. Also draw all the graphs for the disconnected diagrams.
4. Let V(t) be the electron--electron interaction in the expansion for the electron Green's function.
Evaluate the term for n = 1, including all equations and Feynman graphs. Also draw the connected
diagrams for n = 2.
5. Prove the Feynman result e4+B = e4eBe-(l/2)[A.Bl, which is true only if [A, B] commutes with both A
and B. Hint: Recall that
(2.189)
PROBLEMS 107
Use the same method to prove that
e'(A+B) = e'A Ts exp U: dS I e- slA Be'IA ] (2.190)
and evaluate the integral for s = 1.
6. Evaluate the wave vector integrals at zero temperature for the exchange energy in (2.138).
7. Express U I (t) = eiHote-iHt and U2 (t) = eiHte-iHot as time-ordered exponential integrals for both
cases t > 0 and t < O. You will need to use the operator T- I , which arranges operators in their inverse
time ordering.
8. Evaluate the first interaction term in Dyson's equation (2.160) and show that it does give (2.155)
when L(X3, X4)~v = 84(X3 - X4)V(X3)8~v'
9. Show that Eq. (2.159) for G is identical to (2.118) using the result in (2.160). Self-energies obey the
same relations as the Green's functions: i.e., L t = L r + L<, etc.
10. Calculate the average value of the square of an ion displacement (QJ)
(0) = ~J dd q 2Nq + 1 (2.191)

J 2p (21tt ffiq
and show that it diverges in one and two dimensions due to long wavelength fluctuations.
Chapter 3
Nonzero Temperatures
3.1. INTRODUCTION
Experiments are done at nonzero temperatures. Since one goal of many-body theory is to
explain experiments (another is to predict them), the theories should be done at nonzero
temperatures too. It is often unnecessary if the temperature is small compared to other
energies in the problem. But often temperature is important, and here it will be incorporated
into Green's functions. The nonzero temperature formalism was originated by Matsubara
(1955). It will actually be easier to use than the zero-temperature theory of Chapter 2, so that
the Matsubara method will be used throughout the remainder of the book. The zero-
temperature result is always easily obtained from the nonzero-temperature result by just
setting T = O.
At nonzero temperature, it is assumed there is something with a nonzero temperature.
That is, our particle, whether electron, phonon, or spin, is interacting with a bath of other
particles which have an average energy. The exact state of all these other particles is not
known, since they are fluctuating between different configurations. All that is known is the
temperature, which is related to the mean energy.
When defining the Green's function, one must average over all possible configurations of
the system. A possible Green's function for the electron is
Tr[e-~H CpO"(t)C~O"(t')]
(3.1)
Tr(e-~H)
CpO" (t)-eitHC itH

- pO" e- (3.2)
where the symbol "Tr" denotes trace and is the summation over some complete set of states:
Tr == L(nl· . ·In) (3.3)
n
The definition (3.1) would be suitable for a Green's function and is iG>(p; t, t'). However, it
has one drawback which makes its use unwieldy. Usually the Hamiltonian is written
H=Ho+V (3.4)
as a part Ho that can be solved exactly and a part V which remains and becomes the
perturbation. However, V now appears in two different places. First it is in exp(±iHt), which
can be expanded in the usual S matrix. But it also occurs in the factor exp( - ~H). There must
109
110 Chap. 3 • Nonzero Temperatures
also be a perturbation expansion on the thermodynamic weighting factor. Of course, it is a

nuisance to be doing two different expansions at once.
The Hamiltonian enters both terms as an exponential factor. The factor of ~ = l/kBT
can be considered to be a complex time. The Matsubara (1955) method does just the
converse; it treats time as a complex temperature. The object is to treat t and ~ as the real and
imaginary parts of a complex variable, which will require only one S-matrix expansion.
Another motivation for the Matsubara method is provided by examining the thermal
occupation numbers for bosons (e~(j)q - 1)-1 and fermions (e~~p + 1)-1. Each of these can be
expanded in a series (~p = tp - 1-1):
1 1 00 1
nF(~p) =
e~~p
L (2n + l)i1t/~ -
+ 1 = -2 + -~ n=-oo ~p
(3.5)
1 1 1 00
(3.6)
nB(coq ) = e~(j)q _ 1= - 2" + ~ 2~00 2ni1t/~ - co q
These series can be derived from a theorem which states that any meromorphic function may
be expanded as a summation over its poles and residues at those poles. The boson occupation
factor (e~(j)q - 1)-1 has poles at coq = 21tin/~ and the fermion factor (e~~p + 1)-1 has poles at
~p = (2n + 1)i1t /~. It is convenient to define the frequencies at the pole
COn = (2n + 1)1t/~, fermions

bosons (3.7)
where the fermions have poles at odd multiples of 1t /~, while bosons have poles at even
multiples, including zero. Both summations above can be written as
"
L....
--1 - or ,
L...,
-.1
----:-:- (3.8)
n ICOn - CO q n ICOn - ~p
where for fermions we sum over only odd integers and for bosons over even integers. The
factor
(3.9)
has the nature of a Green's function. Indeed, it is the noninteracting Green's function in the
Matsubara method.
In the Matsubara method, time becomes a complex quantity which is usually called 't,
where 't = it. Green's functions are functions of't with domain
(3.10)
Fourier transform theory states that if a functionJ('t) is defined over the range -~ ::::: 't ::::: ~,
then its Fourier expansion is
1
J('t) = 2"ao 00 [ an
+~ T
cos (n1t't) + bn sm T ]
. (n1t't) (3.11)
Sec. 3.1 • Introduction 111
where
1JP
an = ~ _p dif(T) cos
(mCT)
If (3.12)
1 JP
bn = ~ _p dif(T) sm
. (mCT)
If (3.13)
Another way to write the Fourier transform is to define
!
J(iro n) = ~(an + ibn) (3.14)
and hence
(3.15)
(3.16)
There is still a further simplification which can be achieved. The boson Green's functions
have the additional property that
boson:J(T) =J(T +~) when - ~ < T < 0 (and 0 < T + ~ < ~) (3.17)
Divide the integral (3.16) into its negative and positive regions:
(3.18)
and change variables in the second term from T to T + ~, which gives
1 +e'mt)
J(iro n ) =2"(1 . JP dif(T)e,mt. IA...
0 (3.19)
This expression has the feature that J(iro n ) = 0 whenever n is an odd integer. For bosons,
J(iro n ) = J: dTeiron'i(T)
1. bosons (3.20)
J(T) = ~ ~ e-,ron'i(iron )
ron = 2mtkBT
This result agrees with the previous observation (3.6) that boson frequencies contain only
even integers.
Similarly, the fermion Green's functions will have the property that
fermions:J(T) = -J(T +~) when - ~ < T < 0 (3.21)
The same manipulations on the integral in (3.16) give
1
J(iron ) = 2"(1 .
- elnll)
JP dif(T)elnll. IA,..
0 (3.22)
112 Chap.3 • Nonzero Temperatures
In this case J(iron) = 0 if n is even, while for n an odd integer,
J(iron) = J: d'teiOOn'Y('t)
1. fermions (3.23)
J('t) = ~ ~ e-·oon'i(iro n)
ron = (2n + l)nkBT
These equations are identical in form to those in (3.20). The only difference is whether the
frequency ron has even or odd integers. This pair of equations will be used often to define the
Fourier transforms of Green's functions.
At this point the reader is probably overwhelmed with complex frequencies and complex
times. One might argue that this suffering is necessary when including nonzero temperatures.
Actually, the forecast is not as gloomy as that. You will find the Matsubara method is very
easy to use. It is particularly good for evaluating high-order diagrams with many internal
lines. In fact, it is a remarkably easy method to use in practice. Except for the notion of
complex times and frequencies, it is not that much different from the zero-temperature
Green's functions which we defined in Chapter 2.
A second great merit of the Matsubara method is that it leads us directly to physical
results. In Secs. 3.7 and 3.8, some Kubo formulas are derived for the exact definitions of
physical quantities such as the electrical conductivity, thermal conductivity, magnetic
susceptibility, etc. In Sec. 3.3, it will be shown that these correlation functions are just
retarded Green's functions. Finally, it will be shown that the Matsubara Green's functions lead
directly to the retarded functions. Our Matsubara functions will be functions of the complex
frequencies iron' such asJ(iron). The equivalent retarded function is obtained by replacing iron
by ro + io, where 0 is infinitesimal and i = R. This step is called an analytical con-
tinuation. In practice, one just takes the formula one has derived for J(iro n), erases iron
everywhere, and replaces it by ro + iO. This simple procedure yields the retarded function,
which is needed for the physically measurable quantities. The Matsubara technique is a direct
method of calculating the quantities which can be compared with experiment.
3.2. MATSUBARA GREEN'S FUNCTIONS
The electron Green's function is defined as
Q(P, 't - 't') = -(TTCpa('t)C~a('t'» (3.24)

Q(p, 't - 't
') -- -
Tr
[ -~(H-!1N-Q)T
e T eT(H-!1N)Cpae-(T-T')(H-IJN)
t -T'(H-IJN)]
x Cpae (3.25)
e-~ = Tr(e-~(H-IJN) (3.26)
These definitions have several features and conventions which need to be explained. First, the
bracket (... ) in (3.24) has the definition implied by the equivalent equation (3.25). The
bracket (0) on an operator 0 means to take the thermodynamic average, which is the trace
over the complete set of states. Second, the Hamiltonian is now replaced by H - !IN, where J.l
is the chemical potential and N is the particle number operator. A grand canonical ensemble is
used, where the number of particles is variable. This definition of the Green's function applies
Sec. 3.2 • Matsubara Green's Functions 113
to a many-particle system. It can also be used very successfully for one particle in an empty
band. In the latter case, the analytical continuation is taken as iron --+ E + ~ + iO, and the
chemical potential will vanish from all expressions. One is not bothered by the fact that
~~ « 0 in one-particle systems at nonzero temperatures.
In a many-electron system, the chemical potential is retained in the fonnalism. The
analytical continuation is iron --+ E + io and energy is measured from the chemical potential
(Fermi energy). The factor Tt is a 't-ordering operator, which arranges operators with earliest 't
(closest to -~) to the right. It serves the same function as the time-ordering operator in the
zero-temperature Green's functions. The subscript 't is affixed to T to distinguish this operator
from the temperature. The thennodynamic potential n in exp( -~n) is the usual nonnal-
ization factor for a thennodynamic average. The script symbol g has been used for these
Matsubara functions. This script will always be written to alert the reader that these are
Green's functions of complex time and complex frequency.
In (3.24) the Green's function on the left has been written as a function of the difference
't - 't', although the right-hand side is not obviously a function of only the difference. Now
prove this to be the case. First, write the Green's function for the separate cases for't > 't' and
't <'t':
K=.H-J.LN (3.27)
g(p, 't - 't') = -0('t - 't')Tr[e-~(K-n)etKCpO"e-(t-t')KC:O"e-t'K]
+ 0('t' - 't)Tr[e-~(K-n)et'KC:O"e(t-t')KCpO"e-tK] (3.28)
The sign change in the second tenn appears whenever two fermion operators are inter-
changed. Next, use the theorem that the trace is unchanged by a cyclic variation of the
operators
Tr(ABC· .. YZ) = Tr(BC ... XYZA) (3.29)
to move the factors of exp( -'toK) to the left. So Eq. (3.28) can be rewritten as
g(p, 't - 't') = -0('t - 't')Tr([e-t'Ke-~(K-n)etKCpO"e-(t-t')KC:O"]
+ 0('t' - 't)Tr[e-tKe-~(K-n)et'KC:O"e(t-t')KCpO"] (3.30)
Next commute the exponential operators,
e-t'Ke-~(K-n) = e-~(K-n)e-t'K
(3.31)
since they both contain the same operator K [the thennodynamic potential n is not an
operator but is a scalar function of ~ and /l. as defined in (3.26)]:
g(p, 't - 't') = -0('t - 't')Tr[e-~(K-n)e(t-t')KCpO"e-(t-t')KC:O"]
(3.32)
The right-hand side of this equation is now a function only of the combination 't - 't'. The
Green's function can always be written as a function of this difference. It enables us to drop
one of the time variables since it is unnecessary. An equivalent definition of the Green's
function is
g(p, 't) = -(TtCpO"('t)C:O"(O») (3.33)
t )]
= -Tr[e-~(K-n)Tt (etKCpO" e-tKCpO" (3.34)
Next the Green's function is examined for 't < 0 to verify that it does have the property
asserted in (3.21):
(3.35)
By using the cyclic property of the trace several times, the above equation can be rearranged
into
(3.36)
The factor of exp(~n) is not cycled since it is not an operator. These terms can be regrouped
by adding exp(±~K) to the first terms to give
't < 0 : 9(p, 't) = Tr(e-~(K-O)e(H~)KCpO"e-(~+~)KC~O") (3.37)
The term on the right is -9(p, 't + ~) when 0 < 't + ~ < ~. These steps show that
(3.38)
as was asserted earlier in (3.21). The above identity proves that the Green's function can be
expanded in a Fourier series of the type in (3.23):
9(p, iron) = J: d'teim• ~9(p, 't) (3.39)
9(p, 't) = -1 L:e-lm·~9(p,

.
iron) (3.40)
~ n
Equation (3.39) serves as the definition of 9(p, iron), where iron is always an odd multiple of
1t/~ for fermions.
The noninteracting Green's function, or free-particle Green's function, is obtained from
(3.34) by using for the Hamiltonian
H = Ho = L EpC~O"CpO" (3.41)
pO"
K = Ko = L
pO"
~pC~O"CPO" (3.42)
~p=Ep-f.l (3.43)
The 't evolution of the operators is just

CpO"('t) = e~KoCpO"e-~Ko = e-~p~CpO" (3.44)
CtpO" ('t) -- e~Koc! e-~Ko
pO" = e~p~CtpO" (3.45)
which is easily derived from the Baker-Hausdorff theorem:

~
(:-Ce
~
= C + [A, C] + 2!1 [A, [A, C]] + 3!1 [A, [A, [A, C]]] + ... (3.46)
The 't dependence of the Green's function is

g<O)(p, 't) = -e('t)e-~p~ (CpO"C~O") + e( -'t)e-~p~ (C~O"CPO") (3.47)
= -e-~p~{e('t)[l - nF(~p)] - e(-'t)nF(~p)} (3.48)
= -e-~p~[e('t) - nF(~p)] (3.49)
Sec.3.2 • Matsubara Green's Functions 115
where nF(;p) is the expectation of the number operator: nF(;p) = (C~O"CpO"), which from
elementary statistical mechanics has the form
1
nF(;p) = P~ (3.50)
e p +1
It is also easy to obtain the Green's function of frequency:
rj0)(p, ioon) = J: d-ceiOln~rjO)(p, -c) = -(1 - nF) J: d-ce~(iOln-~p)

dO). (1 - nF)(eP(iOln-~p) - 1)
(p, lOOn)
!:I' =- .
lOOn -
;p (3.51)
The second term in the numerator may be simplified by remembering that
i~oon = i(2n + 1)1t (3.52)

(3.53)
which gives
(3.54)
(3.55)
since it is easy to show from (3.50) that [1 - nF] = l/(e-P~p + 1). Equation (3.55) is the
noninteracting Green's function for electrons.
The result for rj0) does have the form suggested in Sec. 3.1. Temperature information is
still contained in this expression but now only in the frequency (2n + 1)1t/~. Later it is shown
how the occupation factors reenter the expressions when the diagrams and correlation
functions are actually evaluated.
The phonon and photon Green's functions are defined in the same fashion. They are
obviously similar to each other, so only the derivation of the phonon Green's function is
presented. The photon results will be stated at the end. For phonons, in the interval
-~ .::: -c .::: ~ the Green's function is
V(q, -c - -c') = -(T~A(q, -c)A(-q, -c')} (3.56)

A(q, -c) = e~H (a q + a~q)e-~H (3.57)
Phonons have no chemical potential, since one can make an arbitrary number of them, and the
tau dependence is governed just by the Hamiltonian. Again one can show that the right-hand
side of (3.56) is only a function of -c - -c'. It is not necessary to keep two -c variables, and one
can instead define
V(q, -c) = -(T~(q, -c)A(-q, O)} (3.58)
Next, examine the behavior for negative tau:
-c < 0: V(q, -c) = -(A(-q, O)A(q, -c)} (3.59)
= - Tr[e-P(H-Q) A( _q)e~HA(q)e-~H] (3.60)
Using cyclic pennutations of the trace turns the trace into

't < 0: D(q,'t) = -Tr[e/3ne'tHA(q)e-(~H)HA(-q)] (3.61)
't < 0: D(q, 't) = _Tr[e-/3(H-n)e(~H)HA(q)e-(H/3)HA(-q)] (3.62)
which proves
-~ < 't < 0 : D(q, 't) = D(q, 't +~) (3.63)
The right-hand side of the equation is the Green's function when 0 < 't + ~ < ~. This identity
satisfies the general conditions asserted in (3.17) for boson correlation functions. The Fourier
transfonn has the following fonn in (3.20):
V(q, iron) = J: d'teiOOn~V(q, 't) (3.64)
1 .
V(q, 't) = ~ ~ e-lOOn~V(q, iron) (3.65)
ron = 2mckBT (3.66)
Equation (3.65) provides the definition of the frequency-dependent Green's function.

The difference between (3.17) and (3.21) is just a sign change. The fennion functions
have a sign change because their operators obey anti-commutation relations, while the bosons
have no sign change because their operators obey commutation relations. Of course, this sign
change is a result of the fundamental difference between bosons and fennions. The sign
change is also responsible for the sign change between ±1 in their two fonns of thennal
distributions: (e/3~p + 1)-' vs. (e/300q - 1)-'. This sign change is very important, and one has
to keep track of it carefully in fennion problems with many operators.
For phonons, the noninteracting or free-phonon Green's function is obtained by taking
H = Ho = Lq roqa:aq, which for the 't variation of the operators yields
aq('t) = e~HOaqe-~HO = e-~OOqaq (3.67)
a:('t) = e~Hoa:e-~Ho = e~OOqa: (3.68)
Always remember that [aq('t)]t i= a:('t). The noninteracting Green's function is

V(O)(q, 't) = -0('t)(aqe-~OOq + a!qe~OOq)(a_q + a:))
- 0( -'t)(a_q + a:)(aqe-~OOq + a!qe~OOq)) (3.69)
Capital letters are used to signify the thennal expectation value of boson number operators:
I
Nq = (a:aq) = nB(roq) = ePOOq _ 1 (3.70)
Nq + 1 = (aqa:) (3.71)
Averages such as (aqa q) and (a:a:) yield zero since they vanish for each tenn in the trace.
The Green's function of't can be written as
V(O)(q, 't) = -0('t)[(Nq + l)e-~OOq +Nqe~OOq]
- 0(-'t)[Nqe-~OOq + (Nq + l)e-~OOq] (3.72)
Sec. 3.2 • Matsubara Green's Functions 117
The Green's function of frequency is
V(O)(q, iron) = J: d1:"eim•t V(O)(q, 1:")
(e~(im.-mq) - 1) (e~(im.+mq) - 1)]

= - [ (Nq + 1). + N q - - .- - - (3.73)
Iron - roq Iron + roq
The terms in the numerators may be simplified by noting that for bosons exp(iron~) = 1 so
that the Green's function is
(3.74)
Using Eq. (3.70) shows that the first numerator equals -1, and the second is +1:
(3.75)
(3.76)
This Green's function has a simple form. It is almost identical to the zero-temperature case
(2.72), and the only difference is the use of complex frequencies instead of real ones. Note
that V(O)(q, iron) is always negative.
The photon Green's function is also identical to its zero-temperature result, except for
complex frequencies. The fundamental definition is
Vj.lv(k,1:") = - L(TtAj.I(k, A, 1:")Av(-k, A, 0) (3.77)

A.
(3.78)
where the operator Aj.I is the usual vector potential operator in (2.163). The free-photon
Green's function is
41t(0j.lv - kj.llev/k2)
(3.79)
ro~ + roi
which should be compared with (2.178).
This section is ended with a comment on notation. The following three forms for the
Green's function are equivalent and will be used interchangeably:
Q(p, iPn) = Q(p, ip) = Q(P) (3.80)

V(q, iron) = V(q, iro) = V(q) (3.81)
The form on the left has been used so far. In the second form, ip has been used instead of iPn'
They mean the same thing, since the i in ip is enough information to alert the reader that
complex frequencies are being employed, which are always discrete. Hence the n subscript is
redundant. In the last form, a four-vector notation P = (p, ip) is used, and the script form of Q
is sufficient to alert the reader to the use of Matsubara Green's functions.
3.3. RETARDED AND ADVANCED GREEN'S FUNCTIONS
The retarded and advanced Green's functions were introduced in Sec. 2.9. They play an
important role in the nonzero-temperature theory. Their properties are discussed in this
section. Their importance comes from the fact that all measurable quantities, such as
conductivities or susceptibilities, are actually retarded correlation functions. The goal of many
calculations is to calculate a retarded function. There are several different ways to obtain
them. One is to use a real time theory even at nonzero temperatures. This method is obvious
to the beginner, but is the hardest way. The second method, which is used most often, is to
first calculate the equivalent Matsubara function of imaginary frequency. It is shown below
that the retarded function is obtained from the Matsubara function by simply changing iron to
io
ro + where 0 is infinitesimal. The Matsubara function is the easiest one to calculate because
its S-matrix expansion is simple. The retarded function is most easily found from the
Matsubara function.
The retarded Green's functions may be defined for both zero and nonzero temperatures.
The retarded Green's function for an electron in state P is
Gret(p, t - t') = -i®(t - t')([Cpcr(t)C~cr(t') + C~cr(t')Cpcr(t)])

= -i®(t - t')Tr{e-f3(K-O)[Cpcr(t)C~cr(t') + C~cr(t')Cpcr(t)]} (3.82)
"AT C (t)
K = H - """, pcr = eiKtCpcre-itK (3.83)
The brackets (... ) indicate thermodynamic average, as is explicitly shown on the second line.
The square brackets mean nothing in particular; they are used to group symbols together. The
retarded Green's function depends on real time, not tau. The tip-off for this is the i factor in
front which belongs with all real-time Green's functions. The Green's function operates only
for t > t', which makes it causal. One starts a signal at one time t' and measures it later at t. Of
course, actual systems are causal, which is why these Green's functions are the ones of
physical interest. The argument of the Green's function is an anticommutator at different
times. In the limit that the times become equal, the anticommutator becomes unity,
(3.84)
since it just becomes the usual fermion anticommutator. The plus sign in the middle of the
two terms is an important feature for retarded Green's functions of fermion operators. The
left-hand side of (3.82) indicates that the retarded function depends only upon the difference
(t - t'). This feature can be shown by manipulations on the trace similar to those in the last
section.
For phonons, the retarded Green's function is
Dret(q, t - t') = -i®(t - t')(A(q, t)A( -q, t') - A( -q, t')A(q, t)} (3.85)
It is very similar to (3.82) in that it is for real time, is also thermodynamically averaged, and is
defined only for t > t'. However, the sign in the middle is now minus, which corresponds to
the fact that bosons obey commutation relations. For both electron and phonon retarded
functions, the right-hand side can be shown to be a function only of t - t', as is indicated in
the argument of the Green's function on the left-hand side of the definition.
Sec. 3.3 • Retarded and Advanced Green's Functions 119
Retarded Green's functions are needed for many types of operators. These operators will
usually be products of electron or boson operators. For example, let us define the operators
U= I:MijCJCj (3.86)
ij
v = I: MijkCJ CjCk (3.87)
ijk
The operator U is bilinear in the operators Ci , where Mij is just a matrix element. The operator
U is regarded as having boson properties, regardless of whether both Cs are fermion or both
boson operators; the case is excluded where one C is boson and one is fermion. U is boson
because it acts as a composite particle. This bilinear form will be encountered quite often,
since it is characteristic of some important operators such as the current and density operators.
The retarded Green's function for the operator U is denoted U and is defined as
Uret(t - t') = -iE>(t - t')([U(t)ut (t') - ut (t')U(t)]) (3.88)
This definition is similar to (3.85), with the important feature that it has the minus sign in the
center of the bracket, which is the case for all boson operators; that is, for any operator which
is a product of bosons or an even number of fermions.
However, an operator such as V above is considered fermion if it is a product of an odd
number of fermions. Its retarded function is
(3.89)
which now has the plus sign in the bracket.

All these retarded functions have Fourier transforms defined by the usual convention:
Grelp, E) = f~oo dteiE(f-t')Gret(p, t - t' ) (3.90)
Dret(q, 00) = f~oo dteiro{f-t'lDret(q, t - t') (3.91)
Uret(oo) = f~oo dteirofUret(t) (3.92)
The advanced Green's function for each of these is defined by
Gadv(p, t - t') = iE>(t' - t)([Cp(t)C:(t') + C:(t')Cp(t»)) (3.93)

D adv ( q, t - t') = iE>(t' - t)(A(q, t)A( -q, t')
- A( -q, t')A(q, t») (3.94)
Uadv(t - t') = iE>(t' - t)([U(t)ut (t') - ut (t')U(t»)) (3.95)
The only two differences are the sign change in front and the fact that the time domain is now
t' > t, which is just opposite of that for retarded functions. Their Fourier transforms with
respect to frequency are defined in the usual way as in (3.91).
The advanced functions of energy turn out to be the complex conjugate of the corre-
sponding retarded function. To prove this fact, first start with the advanced function, then take
its Hermitian conjugate, and finally invert its time variables. The result is the retarded function
Uadv(t' - t) = i0(t - t')([U(t')Ut (t) - ut (t)U(t')]) (3.96)

(3.97)
Now take the Fourier transform of both sides,
fjret (co) -- JOO dteiro(t-t') fjadv (t' - t)t -- JOO dt I e-irot! fjadv (t)t
I
-00 -00
where the last step is obtained by changing variables tI = t' - t. These steps prove that
(3.98)
This result can be generalized to any of the retarded and advanced Green's functions. It is
sufficient to find the retarded function, since a simple complex conjugation derives the
advanced one.
A particular representation for these Green's functions is introduced. This representation
is a formal one, which is not generally useful for calculating physical quantities and deter-
mining numbers. However, it is useful for proving theorems and in particular for relating one
Green's function to another. This representation uses the complete set of states 1m) which are
the exact eigenstates of K = H - /IN. Usually the exact eigenvalues and eigenstates of K are
not known. However, in principle they exist, which is sufficient for proving theorems. The
eigenvalues of K are denoted Em
Kim) = Emlm) (3.99)
This complete set of states will be used in the thermodynamic average: the symbol Tr denotes
trace, and the set In) are used for this summation:
Uret(t - t') = -i0(t - t')e~Q L(nle-~K[U(t) 1\ ut(t')

n
- ut (t') 1\ U(t)]ln)
The above equation has two points marked by an insertion sign "1\". In both places, a
complete set of states is inserted which is unity,
1= L Im)(ml (3.100)
m
which gives
m.n
(3.101)
The matrix elements are easy to evaluate
(3.102)
which gives for the retarded function
Urelt - t') = -iE>(t - t')eliQ L e- PEn [ei(t-t')(En-Em)I(nl U(t) 1m) 12

m,n
(3.103)
In the second term exchange the dummy summation variables n and m so that the matrix
elements (nlUlm) are the same in each term:
Uret(t - t') = -iE>(t - t')e pn L I(nlUlm) 12ei(t-t')(En-Em)

m,n
This formula is the result for the retarded Green's function of time. The Fourier transform is
taken to give the frequency function,
Uret(ro) = -iJoo eit(m+io) dte pn L l(nIUlm)12eit(En-Em)[e-PEn - e- PEm ]

o m,n
e-PEn _ e-PEm
= epn L l(nlUlm)1 2 . (3.104)
m,n ro + En - Em + 10
where io is added to the frequency to ensure convergence at large times.

The equivalent Matsubara function for the operator U is defined by a script symbol:
U('r;) = -(T~U('r)Ut(O» (3.105)
U(iro n) = J: d'teimn'U(,r) (3.106)
The representation In) is also applied to this expression
(3.107)
n,m
U('t) = _e pn L I(nl Ulm) 12e-PEnet(En-Em) (3.108)
n,m
The frequency transform is
U(iro n) = _e pn L l(nIUlm)1 2e- PEn JP

dteimnte~(En-Em)1
n,m 0
e-PEn _ e-PEm
= e pn L l(nIUlm)1 2 -.- - - - (3.109)
m,n Iron + En - Em
where again exp(Biron ) = 1 for bosons. This result should be compared with the retarded
function in (3.104). They differ only in the frequencies in the energy denominator since the
Matsubara result has iron where the retarded function has ro + iO. The Matsubara function can
be changed to a retarded one with just this alteration:
change . -
. .~ U(lro n) = Uret(ro) (3.110)
Iron ---+ ro + lu
This step is called an analytic continuation. The same method can be used to show the same
identity for the other Green's functions:
change .
.
100n -+ Ol
+ Iv's: Q(p, 100n) = Gret(p, Ol) (3.111)
change .
(3.112)
.
100n -+ Ol + Iv's: D(q, IOln) = D ret (q, Ol)
This relationship to the retarded function is one of the primary reasons that the Matsubara
functions are so useful. After they are evaluated, this simple analytical continuation then
yields the retarded function, which is the function of physical interest. The advanced func-
tions are obtained by the analytic continuation iOln -+ Ol - iD, which is obviously true since
the advanced function is the complex conjugate of the retarded function.
Another quantity of great importance is the spectral junction, which is also called the
spectral density junction. It is the imaginary part of any retarded function multiplied by 2; for
example,
R(Ol) = -2Im[Uret (0l)] (3.113)

B(q, Ol) = -2Im[DretCq, Ol)] (3.114)
A(p, Ol) = -2Im[Gret (p, Ol)] (3.115)
There is not a formal symbol for this quantity, so several are used. From the representation of
the retarded function (3.1 04), the only complex part is
1 1
------. = P - i1tD( Ol + En - Em)
Ol + En - Em + ID Ol + En - Em
which gives for R( Ol)
R(Ol) = e~Q L I(nlUlm) 12{e-~En - e-~Em }21tD(0l + En - Em)
m,n
The temperature factors can be regrouped to give
e-~En (1 - e-~(Em-En)} (3.116)
so that
R(Ol) = 21t(1 - e-~(J))e~Q L I(nl Ulm) 12D(0l + En - Em) (3.117)

m,n
It is now possible to write the retarded or Matsubara functions as integrals over these
expressions:
- Joo dOl' R(Ol')
Uret(Ol) = -00 21t Ol - Ol' + iD
U(iOln) = Joo
-00
dOl' R(Ol')
.
21t 100n - Ol
, (3.118)
These identities follow directly from the prior results (3.104), (3.109), and (3.117). An
expression of this form is called a Lehmann representation (1954), and it was first used in
quantum electrodynamics.
Much of this book is devoted to studying electrons, and the spectral function for elec-
trons is calculated often in a variety of problems. Here is a good place to discuss some of its
general features. The representation for the retarded function is:
f30 2 e- f3En + e- f3Em (3,119)

Gret(p,oo)=e LI(nICpcrlm)1 +E -E +'8
n,m 00 n m 1
This equation is similar to the boson result (3.104), the only difference is the plus sign in
e- f3En + e- f3Em . The plus sign follows directly from the plus sign between the two terms in the
definition (3.82). The spectral function for the electron is then
A(p, (0) = 2n:ef3Q L I(nICpcrlm) 12 [e- f3En + e-f3Em]8(oo + En - Em) (3.120)

n,m
This quantity is absolutely positive for all values of the variables (p, (0) since the right-hand
side of (3.120) contains only positive factors: A(P, (0) > o. This positiveness is an important
feature, since A(p, (0) is interpreted as a probability function. The spectral functions for
bosons do not have this property, since they are sometimes plus and sometimes minus. One
can show, however, that they are always plus for 00 > 0 and always minus for 00 < 0, which
follows from Eq. (3.117).
Another important feature of the electron spectral function is obtained by integrating
over all frequencies:
1= J
OO
-00
doo
-2 A(P, (0)
n:
(3.121)
This important theorem, actually a sum rule, is proved by integrating the representation
(3.120):
(3.122)
This expression can be simplified by eliminating the summations over n and m. This elim-
ination is achieved by reversing the steps by which we derived (3.104). First one relabels n
and m in the second term:
n,m
(3.123)
Then one can eliminate the summation over m to give
= ef30 Le-f3En(nl[CpcrC~cr + C~crCpcr]ln) (3.124)

n
The anti commutator in brackets yields unity, which proves
(3.125)
The spectral function may be obtained for a noninteracting electron. In the definition (3.82)
set Ko = Ho - !!N and obtain
Cpcr (t) = e-it~PCpcr (3.126)
so that
dO)(p
ret ,
t
t - t' ) = -ie(t - t')e-i(t-t')~p ((Cper Cper + Cper
t Cper )} (3.127)
= -ie(t - t')e-i(t-t')~p (3.128)
The Fourier transform is
dO) ( E) _ 1
ret p, - E - ~p + io (3.129)
The factor of iO was inserted for convergence. It has one sign 0 > 0, even in many electron
systems with a Fermi surface. The retarded functions do not have op changing sign at the
Fermi surface, which makes them easier to use than the zero-temperature Green's functions
introduced in Chapter 2.
The spectral function for the noninteracting Green's function is
(3.130)
It is just a delta function. The spectral function A(p, E) is interpreted as a probability function.
It is the probability that an electron has momentum p and energy E. For a free, or non-
interacting, particle then E = ~p so the probability distribution is a delta function: Here there
is only one value E for each ~p and vice versa. Equation (3.130) is plotted graphically in Fig.
3.1. For a fixed value of p, a plot of the spectral function vs. E is a sharp delta function (here
given a small width to aid the eye) at E = ~p. When A(p, E) is computed for interacting
systems, the broad distribution is found as the one labeled A in Fig. 3.1. There is a band of E
values for each p, which is not surprising. When the electron scatters, it has a nonzero mean
free path, and there is some uncertainty in its momentum or energy or both. So p and E are
treated as separate variables and both are summed over when evaluating physical quantities.
The spectral function A(p, E) appears in these summations and gives the proper probability
weighting between these variables.
Another quantity to evaluate, for an interacting electron system, is the number of
electrons in a momentum state p, which is
(3.131)
"
.~
u
...""
~GO
Q.
...
< f ) ' - -_ _~-.:......-
FIGURE 3.1 The noninteracting spectral function A(O) is represented by a delta function, while the interacting
spectral function A often has a nonzero width.
Sec.3.3 • Retarded and Advanced Green's Functions 125
For a noninteracting electron system, the number is trivially given by (3.50). In the interacting
system, again introduce the representation
np = e~n L(mle-~KC!crln}(nICpcrlm} (3.132)

n,m
= e~ L e-~Eml(nICpcrlm}12 (3.133)
n,m
This result np should be compared with A(p, E) in (3.120). The two expressions differ in the
factors
(e-~En + e-~Em)8(E +En -Em) = e-~Em(e-~(En-Em) + 1)8(E +En -Em)

= e-~Em(e~E + 1)8(E + En - Em)
Consider the integral using (3.120)
Joo
-00
dE 1
-2 ~IA(p,E) - e
1t e" +
_ ~n
Le
n,m
-~Em 2 _
I (nICpcrlm) I - np (3.134)
np = Joo dE
-00
-2 nF(E)A(p, E)
1t
(3.135)
The factor nF(E) is the fermion occupation factor at nonzero temperatures. Remember that the
energy E is measured with respect to the chemical potential. The thermal occupation prob-
ability is determined by the energy E of the particles. The number of particles in a state p is
obtained by summing over all energies E, weighted by the spectral function which gives the
probability that a particle in state p has energy E, and also by multiplying by the thermal
occupation factor nF(E). This reasonable expression provides further examples of the use of
the spectral function. In the limit of no interactions, then A(p, E) becomes the noninteracting
spectral function (3.130), and Eq. (3.50) is recovered again.
For phonons, the average number of phonons in a state q is N q = (a~aq), and one can
similarly show that
(3.136)
The phonon spectral function B(q, 00) was defined earlier in (3.114). The factor nB(oo)B(q, 00)
is always positive for phonons and could serve as the temperature-dependent probability of
having phonons with q and 00. The noninteracting phonon spectral function is
(3.137)
The retarded Green's functions will be calculated from the Matsubara functions. In the next
section it is shown that the Matsubara functions have a Dyson equation of the form
(3.138)
(3.139)
The self-energy functions ~ and P will be calculated according to rules described later. Define
the retarded self-energies according to (3.111):
change .
. E + .~ ~(p, 1Pn) ~ ~ret(P' E) = Re~ret(p, E) + ilm~ret(p, E)
1Pn ~ 1u
change
. (3.140)
1ffin ~ ffi + 1u.~ P( q, iffin) ~ Pret (q, ffi)
The retarded Green's function will also have a Dyson equation
1
(3.141)
Gret(p, E) = E + io - ~p - ~ret(P' E)
as derived from (3.111). The spectral function for the electron is rewritten in tenns of the
retarded self-energies,
A(p, E) = -2Im~ret~' E) 2 (3.142)

[E - ~p - Re~ret(p, E)] + [Im~ret(P' E)]
where Im~ < 0 so that A > O. One way to obtain the spectral function is to evaluate the self-
energies. The method for doing this is described in the next section.
The fonnal distinctions between the Matsubara, retarded, and advanced Green's func-
tions are best understood by some simple examples. There are some simple functions which
have the correct analytical properties. For example, consider a self-energy operator which has
the following functional fonn,
~(p, Z) = C In[f(p) - Z] (3.143)
where Z is a complex variable representing the frequency. Take C as a constant and f(p) as
some function of momentum. The Matsubara self-energy is evaluated at the points iPn:
~(p, iPn) = C In[f(p) - iPn] (3.144)
The analytic continuation ipn ~ E ± io to the real axis has the following values. For the
retarded function, iPn ~ E + io
~ret(P' E) = C In 1 f(p) - EI - i1tC0[E - f(p)] (3.145)
while for the advanced function, iPn ~ E - iO,
~adv(P, E) = C In 1 f(p) - EI + i1tC0[E - f(p)] (3.146)
These two self-energies differ in the region E > f(p), because their imaginary parts have the
opposite sign. This difference agrees with our general theorem that
Gret(p, E)* = Gadv(p, E) (3.147)
which also implies that
(3.148)
This functional behavior is shown in Fig. 3.2. There is a branch cut on the real axis for
E > f(p). This branch cut just expresses the fact that In(f - Z) is not a continuous function
of Z across the real axis for E > f, since the imaginary part changes sign.
Another example which has similar analytical properties is
~(p, Z) = C[f(p) - Z]l/2 (3.149)
Sec.3.3 • Retarded and Advanced Green's Functions 127
ImI-O (tmI<O
fc,l
lXmI>O
FIGURE 3.2
This function also has a branch cut for E > f(p), with 1m ~ < 0 above the branch cut and
1m ~ > 0 below. In fact, a branch cut is a necessary feature whenever 1m ~ =f. O. Whenever
Im~ =f. 0, then
~(p, E + iO) =f. ~(p, E - iO) (3.150)

which requires a branch cut in the analytic function. When self-energy functions are eval-
uated, they are often given by logarithmic or square root functions.
When a branch cut occurs and ~ =f. 0, then the spectral function is given by (3.142). In
frequency regions where 1m ~ = 0 and there is no branch cut, then take the limit of 1m ~ -+ 0
and obtain
1m ~ = 0: A(p, E) = 21t0(E - ~p - Re ~ret(P' E» (3.151)
Here the spectral function is again a delta function, but now the real part of the self-energy
may be nonzero, and usually is, so that it affects the spectral function. Denote by E(p) the
solution to the equation
E(p) - I! = ~p + Re ~ret(P' E(p) - I!) (3.152)
Assume that there is a problem in which (3.152) is satisfied when 1m ~ = O. Recall that if
g(x) = 0 at x = xo, then delta functions have the property
O[g(x)] = o(x - xo) (3.153)

Ig'(xo) I
and the spectral function is written as
A(p, E) = 21tZ(p)0(E - E(p) + I!) (3.154)
1
Z(p) = I1 - aEa ~ret(p, E) I (3.155)
E=E(P)-11
The factor Z(P) is called a renormalization factor. If we recall the facts that (1) A(p, E) > 0
and (2) f dE/(21t)A(p, E) = 1, then it is easy to show that Z(p) < 1. The strength of the delta
function peak is always less than or equal to unity. An example is illustrated in Fig. 3.3. The
FIGURE 3.3 The spectral function A(p, 0) will have a delta function peak whenever:rnu: = 0, which is shown here
at 0) = Ep - 11. The spectral function also has values in regions where Im~ =F 0, as shown by the crosshatched
distribution.
spectral function is shown with a sharp delta function peak at 0) = Ep - Il and a continuous
spectra in the region 0)1 < 0) < 0)2' The latter region is where 1m L ::J O. Since the total
integrated area under the entire spectra is unity, the existence of any areas where 1m L ::J 0
implies that the renormalization factor Z(p) for the delta function is less than unity.
Equation (3.l52) may be used to define the effective mass. Assume that the noninter-
acting states are free particles, so that
(3.l56)
Furthermore, assume at low momentum that Ep == E(p) varies quadratically with momentum,
2
Ep = Eo + :m* + O(p4) (3.157)
The proportionality constant is the inverse effective mass m*

!!!... _ aEp (3.158)
m* acp
and the derivative of (3.l52) is
(3.l59)
The last term on the right contains the factor (3.l58), which is brought to the left to finally
obtain
(3.160)
This formula will be used frequently to obtain the effective mass from self-energy calcula-
tions.
3.4. DYSON'S EQUATION
The Matsubara Green's functions are evaluated by the same method of Feynman diagram
techniques that were introduced in Chapter 2 for the case of zero temperature. These methods
are slightly modified to account for complex times and frequencies, but otherwise the S-
matrix expansion appears to be very similar. The ideas behind expanding the S matrix are
presented, and Dyson's equation is rederived for the Matsubara Green's functions.
Consider the case of the electron Green's function:
Q(p, ,) = -e~nTr[e-~KTtCe1:KCpae-1:K)C~a] (3.161)
e-~n = Tr(e-~K) (3.162)
Again consider the general case where

K = Ko + V = Ho - j.lN + V (3.163)
H=Ho +V (3.164)
Sec. 3.4 • Dyson's Equation 129
where Ko is a problem which can be solved, so that we know its complete set of states. The
Hamiltonians under consideration usually have the property that they commute with the
number operator:
[Ho,N] =0 (3.165)
[H,N] = 0 (3.166)
In this case, simultaneous eigenstates of Ho and N can be defined and also of Hand N. Call
Ho or Ko the noninteracting problem, and V is the perturbation whose effects are being
evaluated. Consider the operators in the interaction representation
U(r) = eTKoe-TK (3.167)
U-\r) = eTKe-TKo (3.168)
A caret is put above operators to denote their 't dependence with respect to the noninteracting
operators:
(3.169)
The Green's function in (3.161) is written for 't > 0 as

9(p, 't) = _e~QTr[e-~KO(e~KOe-~K)(eTKe-TKO)(eTKoCpcre-TKO)
x (eTKOe-TK)C;cr]
Tr(e-~Ko U(P) U-\ ('t)Cpcr ('t)U( 't)C;cr]
(3.170)
Tr[e-~Ko UCP)]
The thermodynamic potential in the denominator has been replaced by the equivalent factor
e-~Q = Tr(e-~K) = Tr[e-~KO(e~KOe-~K)]
= Tr[e-~KOU(P)] (3.171)
A similar substitution has been made on the factor of exp( - PK) in the numerator.
The operator U('t) can be solved in terms of 't-ordered products. Consider the derivative
(3.172)
which can be expressed in the interaction representation as
~ U('t) = _(eTKO Ve-TKO)(eTKoe-TK) = - V('t)U('t) (3.173)

ifr
This equation for U('t) may be solved, at least formally, by repeated integrations and using
U(O) = 1:
U('t) = 1- J: d't\ V('t\)U('t\) (3.174)
= 1- 1: d't\V('t\)+(-li 1: fd't\ d't2V('t\)V('t2)U('t2)
= f(-l
n=O
r fT d't\ ... JTn-l
0 0
d'tnV('t\) ... V('t n) (3.175)
The summation can be expressed as an ordered product,
(3.176)
= T1 exp [ - f: d'tJ'('t I )] (3.177)
and finally as a 't-ordered exponential integral. The strict definition of (3.177) is just (3.176).
r
Next consider the definition
S('tI,'t2) = T1 exp [ - d'tIV('t I)] (3.178)
S('t) = U('t) (3.179)

It is easy to prove the following operator identities:
S('t2, 'tl) = U('t2)U- I('tI) (3.180)
S('t3' 't2)S('t2' 'tl) = S('t3' 'tl) (3.181)
In this notation the Green's function (3.170) may be rewritten for 't > 0 as
~K ~ ~t
Tr[e- 0 T1S(~, 't)Cpcr('t)S('t)Cpcr(O)]
9(p, 't) =- Tr[e-~KoS(~)] (3.182)
Using the above properties of S(~, 't) and taking advantage of the freedom to rearrange terms
within the 't-ordering operator, the numerator may be expressed as
(3.183)
From now on the trace over exp( - ~Ko) is written with a subscript 0,
Tr[e-~KO] = 0(9) (3.184)
where 9 is any operator. The Green's function is
(3.185)
We continue to write a caret over cJcr(O). Although it has no 't dependence when 't = 0, the
't = 0 label is important for ordering the operator. Equation (3.183) has been proved only for
the case 't > 0, but this case is the only one of interest. The Tr operator in (3.183) really
means (3.182). The S matrix is divided into parts, which are ordered with respect to the other
operators Cpcr('t) and CJcr(O). This form of the Green's function is similar to the zero-
temperature result (2.50).
The Green's function is evaluated, at least formally, by expanding the S matrix in the
numerator:
(3.186)
Each of the nth terms are evaluated by applying Wick's theorem to the brackets and thereby
expressing the brackets as combinations of the non-interacting Green's functions g<.o) and
D(O). Wick's theorem is applied as before to give, for example,
°(TTCpcr(
A At
't )C
A
ks ( 't1)Ck,
At
s<'t 1)Cpcr (0)])
= 0p=kOp=k,Os=crdO)(p, 't - 't1)dO)(p, 't1) - 0k=k,nkdO)(p, 't) (3.187)
Wick's theorem has a double meaning for Green's functions at nonzero temperatures. The first
meaning is just the pairing feature: start with a state In) (in the trace) and operate on this state
by some CI.... In order to get back to (nl at the other side of the trace, there must be another
operator of the form ct appearing in the product of operators. The pairing was used in Wick's
theorem for zero temperatures, and the same use is made here. Whenever there is a bracket
with M creation and M destruction operators, each of the creation operators must correspond
to the same state as one of the destruction operators and vice versa.
The other feature of Wick's theorem concerns the thermodynamic average. The brackets
o{· .. ) mean the average in (3.184), and usually
(3.188)
where A and B are two arbitrary operators. Usually, the product of averages is not equal to the
average of products. However, in Wick's theorem, we do just that: a product of operators is
paired up, and each pair is thermodynamically averaged separately. This procedure does give
the right answer as long as H o is only bilinear in the operators. Any errors that are made
vanish in the limit of infinite volume. One exception is for macroscopic quantum states as
occur in superfluids. As an example, consider the evaluation of
(3.189)
where the operators refer to fermions. First, do it exactly. The combination
o{C~.s<'t)Cp.s<'t)) = np (3.190)
is the number operator. It is assumed to commute with the Hamiltonian and has no 't
dependence. In the thermodynamic average, the cases of p = k, s' = s must be treated
separately from p i= k or s' i= s. The occupation numbers are assumed to be independent of
the spin index:
+ o{C~,sCp,sC~,_sCp,_s)] (3.191)
= L L npnk + L[n p +n~] (3.192)

S,s' p#k p,s
(3.193)
°
The average of (CLCp,sCLCp,s) gives only one factor of np since the occupation number
C~,sCp,s is either zero or one. Squaring it still gives zero or one. However, the average
o{CLCp,sc~,-sCp,_s) gives n~ since the two spin states are averaged independently.
The same problem can be solved by the application of Wick's theorem which gives
(3.194)
Recall that g<O)(p, 't) = -[0('t) - npJ exp( -~p 't). Then the above combination of Green's
functions gives ('t' = 0)
and this method yields the same answer (3.193) as the exact method.
Diagrams in the S-matrix expansion are classified as connected or disconnected
according to the conventions discussed in Sec. 2.6. Only connected diagrams are retained,
since the disconnected diagrams are canceled by the vacuum polarization diagrams. For
Matsubara Green's functions, the vacuum polarization terms come from the denominator:
(3.195)
The expansion of this quantity produces a series of diagrams which just cancel the discon-
nected parts of the expansion in (3.186). In the zero-temperature case of Chapter 2, it was
found that each vacuum polarization graph gave infinity, which was perhaps not a worry since
they canceled out of the answer. In the Matsubara formalism, the vacuum polarization terms
are all nonzero. The "time" integrals, which are now 't integrals, are only over the range
o < 't < ~ and the extra integration does not diverge. The Matsubara formalism eliminates
this one divergence, which is another small advantage of the method.
The summation in (3.195) actually evaluates a useful quantity: the thermodynamic
potential Q. One might actually wish to calculate it. The method of evaluation is presented in
Sec. 3.6, and is called a linked cluster expansion.
The Matsubara Green's function can be reduced to an evaluation of all connected,
different diagrams:
(3.196)
where only the different, disconnected diagrams are evaluated. Each diagram is evaluated as a
function of't, and the Green's function of energy is found from the Fourier transform:
(3.197)
and Pn = (2n + l)nkBT. The terms in the series (3.196) yield self-energy diagrams, which
may be collected into Dyson equations:
. )- 9(0)(p,lPn
. )
9(p,lPn - (0). . (3.198)
1 - 9 (p, IPn)L(P, IPn)
V(O) ( . )
'7"'0( . )_ q, Iffin
1./ q, Iffin - (0). . (3.199)
1- V (q, Iffin)II( q, Iffin)
/
./ --- \
\
,) I
> )
FIGURE 3.4
As an example, consider the basic electron self-energy contribution from the electron-phonon
interaction shown in Fig. 3.4. This self-energy was evaluated in (2.111) using zero-
temperature methods. Here it is evaluated again using Matsubara methods. The interaction is
(3.200)
The n = 2 term in the S-matrix expansion is
(3.201)
The factor of
(3.202)
gives the phonon Green's function. Applying Wick's theorem to the electron bracket yields six
terms, but there is only one different disconnected term. It gives the contribution:
(3.203)
The 't integrals may be performed easily, since the 't dependence of the non-interacting
Green's functions was derived earlier in (3.49) and (3.72). However, this step is rarely
performed. Instead, it is easier to take the Fourier transform in (3.197). The 't integrals are
done by using the expansion given in (3.39) and (3.65), such as
v(O)(q, 'tJ - 't2) = i- ~:::e-ifin(TI-T2)V(O)(q, iron)

P fin
(3.204)
g(0)(p, 't - 'tJ) = i- L iPm)

e-iPm(T-Tl)g<0)(p, (3.205)
PPm
so that
x g(0)(p - q, iPn,,)9(O)(p, iPn"') J: J: J:

d, d'i d'2
x exp[iPn' - iCOm('1 - '2) - iPn'(' - 'I)

- iPn"('1 - '2) - iPn""2] (3.206)
All three , integrals are of the form
~J~ d,e i1 (Pn-Pm) =. 1 (ei~(Pn-Pm) - 1)

~ ° 1~(Pn - Pm)
1 (e2in(n-m) _ 1) =8 (3.207)
2in(n - m) nm
which are written in the shorthand notation
(3.208)
(3.209)
(3.210)
Combining these results yields the final result

92(p, iPn) = g(0)(p, iPn)2 L (1)(p, iPn) (3.211)
where the one-phonon self-energy term is now
L(I)(p, iPn) = - A1 L M~V(O)(q, ico m )9(O)(p - q, iPn - ico m) (3.212)

pV qcom
The self-energy (3.212) in the Matsubara notation is similar to (2.111). The only difference is
the use of complex frequencies rather than real ones. Also note there is frequency conser-
vation in this expression. An electron starts with frequency iPn, and it emits or absorbs a
phonon with frequency ico m, so the electron goes to an intermediate state with frequency
iPn - icom. Again the summation is taken over internal variables, in this case the momentum q
and the frequency icom. In general, the complex frequency is conserved at each vertex in the
Feynman diagram. This conservation does maintain the oddness of fermion frequencies and
the evenness for bosons. For example, the electron in (3.212) has the frequency iPn - ico m.
Now iPn is an odd integer, since it was associated with the original electron line, while icom is
an even integer, since it was associated with the phonon. But an odd plus an even is odd,
which preserves the oddness of electron frequencies.
The exact Matsubara Green's function may be obtained by finding the exact self-
energies. The exact solution is rarely possible, and usually one is content to write down a few
diagrams and evaluate the first few terms in the expansion for L(p, iPn). To obtain these terms,
draw the Feynman diagrams, conserve momentum and frequency at each vertex, and then
integrate or sum over internal variables. All that IS needed now are some rules for
constructing diagrams:
1. With each internal electron line, associate a quantity g(O)(p, iPn).

2. With each internal phonon line, associate a quantity M~D(O)(q, irom).
3. With each internal Coulomb line, associate a quantity Vq = 41te / qZ.
4. Conserve momentum and complex frequency at each vertex. Keep in mind that
fermion frequencies are odd integers (2n + 1)1tkBT and boson frequencies are even
integers 21tnkBT. Their oddness and evenness will be maintained in the energy
conservation.
5. Sum over internal degrees of freedom: momentum and frequency. Internal variables
are all those except the (p, ipn) of the self-energy.
6. Multiply the expression by
(_I)m+F(28 + I t
(3.213)
(~v)m
where F is the number of closed Fermion loops. The (28 + 1) factor is a summation
over spin degrees of freedom, and 28 + 1 = 2 for electrons. The integer m is the
order of the diagram, as defined earlier in Sec. 2.8.
Several examples are presented. The basic fermion loop in Fig. 3.5(a) occurs in the phonon
self-energy and also the Coulomb self-energy. It is
(3.214)
In Fig. 3.5(b) a phonon line has been added between the two electron lines, which is called a
vertex correction. This diagram gives, according to the rules,
2
p(Z)(q, iron) = --Z L g(O)(p, iPm)g(°)(p + q, iPm + iron)
(~v) q,iPm
X L M~V(O)(k, ikn,)g(O)(p + k + q, ipn + irom + ikn,)

k,;kn,
(3.215)
(a) (b)
tPn +twn
,~<>,~
tpn
FIGURE 3.5
3.5. FREQUENCY SUMMATIONS
When using the Matsubara Green's functions, one must often evaluate frequency
summations over combinations of unperturbed Green's functions. The technique for evalu-
ating these summations is discussed for both cases of unperturbed functions and also Green's
functions with self-energies. First is presented a table of results for combinations which often
occur. Their derivation is later:
(3.216)
(3.217)
_ ~ ",-,(0)(
A L... '.::1'
.
p, IPn
)'-'(O)(k'
':I' ,100m
_.)
IPn
= 1 - . nF(Sp) - nF(Sk)
_ J:: _ J:: (3.218)
f-' n 100m ~p ~k
/31"
'7 9(0) (p,lPn
.) _
-
(J::)
nF ~p (3.219)
The combination (3.216) occurs in the electron self-energy (3.212) from the electron-phonon
interaction, while the combination (3.217) is found in the basic polarization diagram (3.214).
The other combinations will be encountered later.
First consider the summation over a boson series, where the summation is over even
integer combinations OOm = 2nmkBT. For example, consider (3.216):
(3.220)
First, write this equation as
(3.221)
whereJ(iOOm) is the product of Green's functions in (3.220). This summation is evaluated by a

contour integration. The integral has the form
I = lim ,( dz. J(z)nB(z) (3.222)

R-+ooJ2m
where the contour is a large circle of radius R in the limit as R -+ 00. The function nB(z) is
chosen to generate poles at the points iOOm for all even integer m. The function which does this
is
1
nB(z) = -A--1
e>'Z -
(3.223)
The poles of nB(z) were discussed earlier at the beginning of the chapter, in (3.6). There are
poles at the points i2nmkBT for all positive and negative integer m and m = O. The residue at
these poles is 1/~. In Fig. 3.6, these poles are shown as X marks which are evenly spaced on
Sec. 3.5 • Frequency Summations 137
FIGURE 3.6
the vertical axis. The large circle is the contour of integration, which is a circle of radius R.
The functionf(z) is
fez) = 2ffiq 1 (3.224)

z2 - ffi~ iPn +z - ~p
It has poles at the points ±ffiq, which originate from the phonon Green's function. Another
pole is at the point ~q - iPno and this pole comes from the electron Green's function. The
poles, and their residues, of the integral / in (3.222) are
1 .
Zm = i2nmkBT, R j = ~ f(lffim)
Nq
R 1 = -.---,--:!.....--
lPn - ~p + ffiq
Nq +1
R2 = -._ - ' ! . . - _ -
lPn - ~p - ffiq
-2ffiqnF(~p)
R3 = 2 (3.225)
(iPn - ~p) - ffi~
where our notation is that Nq == nB(ffiq). In the last residue, since exp(ipn~) = -1 then
1 1
nB(~p - iPn) = e~(~p-jPn) _ 1= eP~p +1= -nF(~p) (3.226)
The last residue may be rewritten in partial fractions
R3 =. nF(~p) nF(~p) (3.227)

lPn - ~p + ffiq ipn - ~p - ffiq
The integral is evaluated by adding all these residues:
/ = .!. L f(iffi m) + .Nq + nF(~p) + ~q + 1 - nF(~p)

~ m lPn - ~p + ffiq lPn - ~p - ffiq
The integral vanishes (/ = 0) in the limit that R ~ 0 which gives the final result,
N q + nF(~p) N q + 1 - nF(~p)
S= + (3.228)
iPn - ~p + ffiq iPn - ~p - ffiq
which is (3.216). The method of evaluating these boson series is quite simple. To evaluate a
series such as (3.221) one just finds all the simple poles ofJ(z) which are at the points Zj with
residues rj and
S= L rjnB(zi) (3.229)
j
The same procedure is used to evaluate fermion series. Then the summations are of the form
(3.230)
where P n = (2n + l)nkBT contains odd integers. One constructs the same contour integral as
in (3.222). Now the function nF(z) is used since it has poles at the points iPn with residues
-1 /~. The integral again is I = 0, so that the summation is over the residues ri at the poles Zi
ofJ(z):
s = - L rinF(zJ (3.231)
I
The minus sign in front occurs because the residue of the fermion nF(z) is (-1/~), whereas it
is (l/~) for the boson nB(z).
As an example of summing a fermion series, again consider (3.216). The summation
variable is changed to Pn' = Pn + ffim so that the summation is now
_
S- -131 ~ V (0). . />CO)
(q, lPn' - IPnhf' (p, IPn')
•
(3.232)
_ .!.. '" 2ffiq 1

(3.233)
- L.. 2
~ n' (Pn' - Pn) + ffiij iPn' - ~p
2ffi 1
J(z) = q (3.234)
(z - iPn)2 - ffiij Z - ~p
Pn' contains odd integers, since it is a fermion frequency. The functionJ(z) has the following
poles and residues:
2ffiqnF(~p)
RJ = . 2 2
(~p - IPn) - ffiq
-nF(iPn - ffiq)
R2 = -.---,----'-
IPn - ~p - ffiq
R3
+nF(iPn
= -.---,----'-
+ ffiq) (3.235)
IPn - ~p + ffiq
The last two thermal factors may be simplified:

. 1
nF (l!p n = -=----;-- (3.236)
e~(iPn-(j)q) + 1
- ffi )
q
nF(iPn + ffiq) = -Nq (3.237)
The final result is obtained, as in (3.231), by adding these three residues, which gives again
the same answer (3.228).
Three summations were listed at the beginning of the section. The first has been derived
twice. The next two, (3.217) and (3.218), are left as an exercise for the student. The fourth
result (3.219) is now explained
1 ",,(0) .
~ ~ 1:1' (p, lPn) = nF(~p)
The left-hand side of this equation is most familiar as the Fourier transform of the Matsubara
Green's function:
e(0)(p, "C) = ~ ~ e-iPn~ejO)(p, iPn) = -o(T~CpO"("C)C:O"(O)}

The result (3.219) is just the limit as "C -+ O. This limit is ambiguous, since a different result is
obtained if "C = 0 is approached from the positive or negative direction:
e(0)(p, "C = 0+) = -o(CpcrC:cr } = -[1 - nF(~p)] (3.238)
g(O)(p, "C = 0-) = °(C:crCpcr ) = nF(~p) (3.239)

The result (3.219) is merely the convention of adopting the limit"C = 0-. This choice has been
discussed several times before. The same result at zero temperature was expressed in (2.86). It
is the convention that two operators at equal time are taken in the order which gives the
number operator:
(3.240)
The Matsubara sum in the preceding example was very easy. Summations which contain only
e(0) and V(O) always cause the contour integral I to have just poles. If the summation
contained g, instead of e(0), then the I integral has branch cuts. Next is shown how to treat
this more difficult case.
Consider the example of evaluating the summation (3.216) except that the Green's
function for the electron contains some self-energy terms:
. 1
g(p, lPn) = . ~ ~( . ) (3.241)
lPn - p - p, lPn
Letting iPn = z then the above function has branch cuts along the real z axis. These branch
cuts arise because of the self-energies ~(p, z), as discussed in Sec. 3.3. The contour integrals
used to evaluate the frequency summations become more complicated because of the branch
cuts.
The most general possibility when evaluating a summation over Matsubara frequencies
is to have all the Green's functions fully dressed. Then it is easiest to proceed by first
expressing all the Green's functions in the Lehmann representation. For example, evaluate
(3.242)
Express each Green's function as a frequency integral over its respective spectral functions:
'v1"1(q, lOOm
. ) -_ Joo doo -•B(q, 00)
....:....::.'---':.... (3.243)
-00 21t lOOm - 00
.
g(P + q, lPn +.lOOm ) -_ JOO dE A(p + q, E)
--:-.--=----:-.-=---'-- (3.244)
-00 21t 'Pn lOOm - E +
and the summation S becomes
S = Joo d2ro B(q, ro) Joo 2dc A(p + q, c)So(iPn' ro, c)

-00 1t -00 1t
1 1 1 nB(ro) + nF(C)
So(iPn' ro, c) =- A I: . . .
I-' m lro m - ro lPn + lro m - c iPn + ro - c
S = Joo dro B(q, ro) Joo dc A(p + q, c) n~(ro) + nF(C)
-00 21t -00 21t lPn + ro - c
The summation So (iPn , ro, c) is now the easy type which is evaluated using noninteracting
Green's functions. The final result provides for the most general case of fully interacting
Green's functions. In general, any frequency summation can be done by expressing all the
Green's functions in their Lehmann representation, which results in a summation of the easy
type listed in (3.216) to (3.219).
The result (3.216) has now been derived in this section by several different methods.
Now we explain the physics behind this result. It is a bit subtle. The virtue of the Matsubara
method is that these subtleties are automatically handled correctly. It is a machinery which
cranks out the correct thermal occupation factors nF and Nq even in complicated cases.
The summation (3.216) arises from the electron-phonon interaction and is a phonon
contribution to the self-energy of the electron. From the rules for constructing diagrams, the
one-phonon self-energy of the electron is
(3.245)
This result is similar to what is obtained from second-order perturbation theory. The standard
quantum mechanical expression for this quantity is
(3.246)
where Ii) is the initial state of the system, and the summation II) is over possible intermediate
states. It is assumed that the initial state is an electron in state p with energy cp . The electron-
phonon interaction has two terms. One of these is MqC:+q,crCpcra~q, which describes a
process where an electron scatters from p to p + q while creating a phonon of momentum
-q. This contribution from second-order perturbation theory is
(3.247)
The factor (Nq + 1) comes from the phonon creation operator, a~qln_q) =
In_ q + Iln_q + 1). The thermal average is used for n_ q, which is N q. Similarly, the factor
[1 - nF] is the probability that the electron state p + q is empty, so that the operator C:+qcr
can create an electron in that state. There is also a factor nF(~p), which is the probability that p
is occupied with an electron. It is assumed this factor is unity if we are trying to calculate the
properties of an electron in that state, The energy denominator gives the difference between
the initial energy cp and the energy in the intermediate state cp+q + ro q which has an addi-
tional phonon.
Similarly, the interaction term which destroys a phonon is MqC~+qcrCpcraq. In second

order it contributes an energy term
(3.248)
Here the factor Nq is the matrix element of the phonon destruction operator aqlnq} =
~Inq - I}. The electron occupation factor [1 - nF] is the same as before. The phonon
energy in the denominator has changed sign, because there is one less phonon in the inter-
mediate state.
These are the only two terms in the Hamiltonian. The summation of these two contri-
butions, however, does not equal (3.245). In fact, some important processes have been
omitted. There are two more terms. They arise from the other electrons in the system with the
same spin state cr. These other electrons are also trying to emit and absorb phonons, to alter
their energy. The other electrons may not, in this process, use the state p as their intermediate
state. For example, if another electron starts in p' and scatters to p' + q, we may not have
p = p' + q. They may not use p because our electron, whose energy we are trying to
calculate, is occupying it already. The other electrons of the system have a reduced second-
order self-energy because our electron is in the state p. This energy reduction is associated
with p. It is assigned to the self-energy of the electron in p, because the system would not
have it if there were no electron in p. A calculation of the self-energy of an electron in the
state (p, cr) must calculate how much energy is deprived to the other electrons. Assume they
start in p + q initially and get scattered into p. In analogy with the two terms above, the self-
energy of a particle in the state p + q, for emitting or absorbing a phonon and going into p, is
(3.249)
The factor [1 - nF] is the probability that p is empty. This factor is dropped, since p is
occupied. The factor nF(~p+q) is the probability that there is an electron in p + q; if there is
none, then the electron in p is not depriving any electron of energy. The calculation assumes
that the energy does not depend on spin, so the occupation factors nF should not depend on
spin either. These last two terms are then subtracted from the first two to give the result, from
second-order perturbation theory,
(3.250)
The Green's function theory (3.245) changes this result only a little bit. It allows that the
original electron may have had wave vector p, but its "energy" is ipn + J.l and not cp . It keeps
the energy and momentum as separate variables, at least at this stage in the calculation. An
obvious improvement in the calculation is to make the intermediate state energy not cp+q but
some other energy variable which reflects the possible effects of interaction. The difficulty
with such a procedure is that changes in cp+q also require changes in the occupation prob-
ability nF(~p+q). Later it will be shown that one subset of diagrams, when summed to infinity,
yield exactly this approximation; the intermediate state energies are treated exactly also.
In many-particle systems, the interactions which alter energy states of particles usually
lead to subtle counting problems about where energy has been added or subtracted from the
system. The Matsubara method has the virtue that it cranks out the correct answers auto-
matically.
3.6. LINKED CLUSTER EXPANSIONS
Another method for evaluating correlation functions is called the linked cluster method
or the cumulant expansion (Abrikosov et a!., 1963; Brout and Carruthers, 1963). It is the
method used to evaluate the thermodynamic potential n. However, the method has also been
applied to evaluating Green's functions of time G(p, t) for a few problems, which represents
an alternative to the usual Dyson equation approach. Both methods-Dyson's equation and
linked cluster-would give the same result if all terms were summed exactly. However, in the
usual case, when only a few terms in a perturbation expansion are evaluated and the rest are
neglected, Dyson's equation and the linked cluster method are both approximate solutions. In
some special cases the linked cluster expansion may be the better approximation. The art of
choosing which method to apply to a given problem is a matter of judgment, which comes
from experience.
3.6.1. Thermodynamic Potential

The thermodynamic potential is found from the equation
(3.251)
This quantity is useful to evaluate by itself. After it is found, perhaps only approximately,
some thermodynamic results are determined by taking its derivatives; for example, if Ne is the
number of electrons
(3.252)
(3.253)
(3.254)
In the last identity (3.254), the function np is meant to be the number of electrons with
momentum p. This quantity was defined earlier, in (3.135), for an interacting system.
Usually there is a Hamiltonian which is written as
H=Ho+ V (3.255)
K = Ko + V = (Ho - ~e) +V (3.256)
where Ho is exactly solvable. In the interaction representation, the thermodynamic potential is

given by the right-hand side of (3.251). If there are no interactions and V = 0, then the
definition of no is
(3.257)
Sec. 3.6 • Linked Cluster Expansions 143
For a system of electrons and phonons,
Ko = L ~pCJcrCpcr + L O)q(a~aq +!) (3.258)

pcr q
the trace is simple since the noninteracting eigenstates can be used to expand the trace
eigenstates:
(3.259)
Take the logarithm of both sides of this equation,
~no = -2 L In(1 + e-~~p) + L[ln(l - e-~roq) +! ~O)ql (3.260)

p q
The factor of 2 in front of the electron term is the spin degeneracy. The last term on the right
is the zero-point energy of the phonons. Change the summations to integrals, say for the
phonons
d3
J
LF(q) = v ---;F(q)
q (21t)
(3.261)
and a factor appears of the volume v. The thermodynamic potential is proportional to the
volume of the system. This result is also apparent from Eqns. (3.252) to (3.254), since the
right-hand side in each case is proportional to the volume or to the number of particles.
The goal is to evaluate (3.251) including S(~). The interaction V will add some
correction terms to no which changes it into n. The method of evaluating this correlation
function is by the S-matrix expansion given earlier in (3.195):
In evaluating the right-hand side, all diagrams are included, whether connected or discon-
nected, since there is no other series at hand to cancel the disconnected diagrams. The
disconnected diagrams are necessary for the resummation procedure. Basically, the terms in
the S-matrix series are examined in order to find some convenient method of resumming
them. It is efficient to evaluate each different type of diagram only once. The way of doing
this expansion is the linked cluster, or cumulant, expansion.
The answer is stated first. The following discussion will try to make the answer appear
plausible. First, introduce the parameter A. It multiplies the potential V, so that everywhere
there was a factor of V, or V(r), now write AV or AV(r). The parameter A is actually just unity,
but it is used to keep track of the number of times the potential occurs in each term in the S
matrix. The S-matrix expansion in (3.195) may be written as
= e-~no L An Wn
00
e-~n (3.262)
n=O
(3.263)
The basic linked cluster theorem is that this series can be resummed into
e-~Q = exp(-~no + f A}UZ) (3.264)

z=,
( _1)z f~ J~
(3.265)
A A
Uz = -1- 0 d,,··· 0 d,z O(TT V("),,, V('z))
where Uz contains just different, connected diagrams: The thermodynamic potential is

obtained by setting A = 1 in (3.264)
1 00
n = no - - L Uz (3.266)
~z=,
The basic theorem is that the thermodynamic potential is the summation of the different,
connected diagrams. The factor 1/1 in (3.265) turns out to be quite a nuisance, as is discussed
later. It makes further summation difficult, except by a trick.
The theorem will be proved below. First it is helpful to give a simple example. Consider
the perturbation V to be the electron-phonon interaction:
Mq t
V =L r-;AqCp+q,crCpcr (3.267)
qpcr y v
For this example, only terms with even n occur in the series, since the expectation value of
(A~) is zero if n is odd. The first nonzero term in the series, after n = 0, is n = 2:
1 J~ J~0
W2 = U2 = 2' 0 d" d'20(TT V (")V('2)) (3.268)
A A
This disconnected diagram is not evaluated here. Its Feynman diagram is shown in Fig. 3.7(a).
There are two electron lines, and the two vertices are connected by a phonon line. Next
consider the term with n = 4:
W4 = ~ J: d" J: J: J:
d'2 d'3 d'40(TJT(,,)V('2)V('3)V('4))
All contributions for n = 4 consist of diagrams with four electron lines and two phonon lines.
Some are shown in Fig. 3.7(d). However, there also occurs the diagrams shown in Fig. 3.7(b)
where the single bubble occurs twice. In fact, there are three identical terms in W4 which have
a pair of disconnected bubbles. The factor of 3 occurs because, if one chooses" as one of the
vertices of a bubble, the other, variable for the same bubble is either '2' '3, or '4' and these
three choices each give the same contribution. This factor is
0(0)
e
e(b)
ee
8 (e)
0.
b 0 ' ...... '
(d)
EB
FIGURE 3.7
Similarly, it can be shown that each of the connected diagrams in Fig. 3.7(d) occurs six times,
where the factor of 6 comes from different ways to shift the dummy variables around. The
final result is
(3.269)
Next consider terms with n = 6. There will be many connected diagrams which contain six
electron lines and three phonon lines; try to draw them all. They give U6 • There are also
disconnected diagrams. There will be diagrams containing three bubbles, as shown in Fig.
3.7(c)-in fact there will be 15 of these, all equal to each other. There will also be discon-
nected diagrams which contain one four-line term from Fig. 3.7(d) and one from Fig. 3.7(a).
The total is
1
W6 = 3! U2 + U2 U4 + U6 (3.270)
The terms in U2 appear to be generating the series

1 2
1 + U2 + - U2 + -1 U23 + ... = e
U
2 (3.271)
2! 3!
There is a simple argument which shows that the nth term is really appropriate for the
exponential series. Since n is even, let n = 2m, and there are m bubbles of the type shown in
Fig. 3.7(a). How many different ways are there to get m bubbles? Take the first variable, 't\. It
may be paired in the bubble,
U2 = ~ L M~ J~ d't\ J~ d'tjV(O)(q, 't\ - 't)

qpa 0 0
x g(O)(p, 't\ - 't)g(O)(p + q, 'tj - 't\) (3.272)
with the variable 'tj which is any of the other 2m - 1 variables. Next, take any of the 2m - 2
variables left after pairing 't\ and 'tj' say 'ts' and it may be paired with any of the 2m - 3 left.
The number of different arrangements is
(2m)!
(2m - 1)(2m - 3)(2m - 5) .. ·5·3·1 = - - (3.273)
2mm!
This factor in Wn is
which is the form proposed in (3.271). The term U2 has been shown to sum correctly. It must
be shown that the other contributions also sum correctly. This proof is not too hard. It is more
subtle to show that all the cross products such as U2 U4 occur with just the right counting.
Another identity is:
U/ = ---n-
(-Ii J~
d't\···
0
J~
d't/ O(T~V('t\)· .. V('t/))connected
0
A A
(3.274)
The same U/ is in (3.265). Comparing these two expressions shows that for each term I there
are (/- I)! contributions alike, so that the prefactor in front of the number of different
connected diagrams is (/- 1)!/I! = 1/1. The proof is just an exercise in combinatorial
counting. The S matrix is expanded in terms of Green's functions which yields expressions
such as
(3.275)
The combinatorial question is: how many different combinations are really alike if the dummy
variables 1\, 1j , 12, etc., are just relabeled? The answer is (1- I)!. The reader should try to
work some examples to convince oneself that it works. The trivial case is [= 2 so that
(1- I)! = 1 was done in (3.272). A more complex example is shown in Fig. 3.8(a), where the
(4 - I)! = 6 different terms are shown for one of the different diagrams with [ = 4. The
numbers 1,2,3,4 refer to the 1 variables: 1\,12,13,14' These six are the only different
arrangements. One might ask why the figures are drawn such that 1\ is always the upper left-
hand comer. Why not the labeling shown in Fig. 3.8(b)? The answer is that this is exactly the
same term as the first one in Fig. 3.8(a). They are topologically identical but just drawn upside
down and backwards, or rotated by 1800 • However, and most importantly, both diagrams yield
identical results when written down with Green's functions, which is the ultimate test of
whether they are topologically identical.
In the prior section, the Green's function Q(p, 1) was evaluated by a similar expansion
technique. There the connected diagrams obeyed the counting rules that
1
T!
A A A At
o(T,C pcr (1)V(1\) ... V(1/)Cpcr (0)}connected

_ A A A At
- o(T,Cpcr (1)V(1\) ... V(1/)Cpcr (0)}different connected (3.276)
Here there are l! arrangements which are identical. In the linked cluster method there are only
(1- I)!. The difference is in Q(p, 1) there is a reference variable 1, so that arrangements are
considered with respect to this variable. There tum out to be more of them by a factor of [.
Next consider the evaluation of the thermodynamic potential. It is obtained by evaluating
the series of terms in (3.266). An example is to evaluate the series of diagrams shown in Fig.
3.9 for the electron-phonon interaction. They can be summed to give a simple answer. Of
course, this series is not the exact answer, since there are additional terms in which the
bubbles may also have internal phonon lines. The rules for constructing diagrams for the
terms in the thermodynamic potential are similar to those in Sec. 3.5. One follows all those
(0)
Iq2 Ip2 Iq3 Ip3 IP4 10.4
3d4 4D 402 204 3d2 20
FIGURE 3.8
~+
'CJ o tPo·po·
0,+
--;; 0d
..
FIGURE 3.9
rules and then multiplies the result by ~ / 1. The 1/1 factor is just the one which occurs in
(3.265), which has been explained already. The factor ~ comes from the extra t integration.
For example, in (3.272), which is the first bubble, the argument is only a function of tJ - t/
J: J: dtJ dtJ(tJ - t) (3.277)
Using the Fourier expansion
J(tJ -t) =i~e-iq.(~l-~)J(iqn) (3.278)
the first integration gives
(3.279)
while the second just gives ~. Each term has this extra t integration, which gives the one extra
factor of~. For the first bubble in Fig. 3.9
~A?
U2 = 2A ~ M~VO)(q, iqn)p(J)(q, iqn) (3.280)
I-' q,lq.
where the polarization diagram p(J)(q, iqn) is for the single bubble, which was already
evaluated in (3.214):
2
p(J)(q, iqn) = A I: gCO)(p, iPm)gC°)(p + q, iPm + iqn) (3.281)
I-'Vp,iPm
The second term in Fig. 3.9 has two bubbles connected by two phonon lines. Using the rules
for constructing diagrams gives
U4 = r-: ~ [M~V(°>Cq,
I-' q,lq.
iqn)p(J)(q, iqn)f (3.282)
Momentum and frequency conservation requires both phonons to have the same variables
(q, iqn)' and both polarization diagrams are also just functions of this combination. The term
with n bubbles and n phonon lines is
_
U2n - 2n A
~ ').,,2n "[ 2V(0) ( . ) (J)( . )]n
~ Mq q, zqn P q, zqn (3.283)
I-' q,lq.
It is simple to sum the series:
n - no = -i I:
I-' n=!
U2n = 21A ~
I-'q,lq.
In[l - A2M~V(0)(q, iqn)p(J)(q, iqn)] (3.284)
The right-hand side of (3.284) is a correction to the thermodynamic potential. It is not exact,
as explained above, since only some of the diagrams are included. The right-hand side is
proportional to the volume v, as is obvious when the summation over q is changed to an
integration as in (3.261). Similarly, the summation over pin p(J)(q, iqn) can be changed to an
integration:
(3.285)
The argument of the logarithm is not dependent on Y, and Y enters only by multiplying the
result.
The answer is a logarithm. The summation over iqn needs to be evaluated. This
summation is inconvenient (but still possible) when the important functions are inside the
argument of the logarithm. There is a standard trick for eliminating the logarithm, although it
does not eliminate it but just disguises it. Treat 11 = A2 as a variable of integration,
(3.286)
(3.287)
Then for A = 1
(3.288)
The logarithm has been eliminated, but another problem is introduced. The right-hand side
must be evaluated for each value of 11 and then the integral is taken over 11. Since 11 enters in
the same way as a coupling constant, this step is called a coupling constant integration. The
integration can be taken outside of the summation over iqn' which makes this summation
easier in some cases.
There are some other manipulations on this formula, which put it into a form which
appears simpler. First, define the phonon self-energy as
(3.289)
where the coupling constant 11 is included in the definition. The superscript (1) indicates that
this expression is an approximate self-energy, which only includes the single-electron bubble.
The phonon Green's function is
v(O)
V(i) (nq i ) - -------;=---
." , qn - 1 _ V(O)n(i) (3.290)
The Green's function is also a function of the coupling constant strength 11. Equation (3.288)
is rewritten as
(3.291)
This expression is an approximation to Q since it employs an approximate self-energy n(1).

There is a theorem which we are going to state but not prove. The exact correction to the
thermodynamic potential from electron-phonon interactions is
(3.292)
where nell, q, iqn) is the exact phonon self-energy and V(ll, q, iqn) is the exact phonon
Green's function evaluated with the exact self-energy. Both expressions are found as a
function of 11, and then one has to perform the integration over 11. Proofs of this theorem are
given in Abrikosov et al. (1963).
FIGURE 3.10
The exact phonon self-energy neT], q, iqn) is the sum of an infinite number of self-energy
diagrams, some of which are shown in Fig. 3.l0. There are bubbles with internal phonon lines
and also several bubbles connected by more than one phonon line.
Of course, if electron-electron interactions are included, then the phonon self-energy can
also contain internal Coulomb interactions, etc. The effect of Coulomb interaction on the
thermodynamic potential should also be included. The general theorem (3.265) and (3.266) is
true for all interactions: Coulomb, phonon, or others.
It is possible to combine the effects of the phonons and Coulomb interaction in a simple
way. Consider electrons interacting by either or both interactions, and then the single-bubble
diagram appears twice. As shown in Fig. 3.11, the two electron vertices can be connected by a
single Coulomb line or a single phonon line. The sum of these contributions is
U2 = ~ L p(1)(q, iqn)[vq + M~V(O)(q, iqn)] (3.293)

q,lqn
where Vq = 4nt? /q2. Similarly, there are four diagrams with two bubbles. The two bubbles
interact by two Coulomb lines, two phonon lines, or a mix with one of each. This contribution
is
2
U4 = ~ L {p(1)(q, iqn)[v q + M~V(O)(q, iqn)]}2 (3,294)
q,lqn
It seems desirable to introduce a combined interaction propagator, Coulomb plus phonon,

which is
(3.295)
It obeys a Dyson equation, which is
(3.296)
where W(q, iqn) is the total self-energy diagram, which is the sum of the diagrams shown in
Fig. 3.9. Of course, now the interaction lines in Fig. 3.9 refer to W(O), since they may be either
FIGURE 3.11
Coulomb or phonon. The generalization of (3.292) to include both Coulomb and phonon
effects is
(3.297)
The first term contains the Coulomb self-energy of an electron interacting with itself, and this
unwanted contribution is subtracted out in the second term. The two interactions in (3.295) do
not complete the possibilities. If the Hamiltonian has additional interactions due to impurities,
ions, etc., then one has additional contributions which may contribute to the thermodynamic
potential.
The expression (3.292) seems to favor the phonon self-energies. It is reasonable to ask
what happened to the electron self-energies. The answer is that they are in there too. The
diagrams under consideration can also be viewed as corrections to the electron self-energy.
This viewpoint does not introduce an additional contribution but is another equivalent way of
expressing the above answer. For example, the basic bubble diagram, shown in Fig. 3.7(a) and
expressed in (3.272), may also be written as
(3.298)
where l:(l)(p, ipn) is the electron self-energy from one-phonon processes, which was given
earlier in (3.212):
We repeat, for emphasis, that the above U2 is not a different result than (3.280). It is the same
result, expressed with different symbols; the same Green's functions have been grouped
differently. Similarly, the exact answer is
(3.299)
where l:(ll, p, iPn) is the exact self-energy of the electron and g(ll, p, ipn) is the exact Green's
function obtained from Dyson's equation with the exact self-energy. If the propagator W in
(3.295) is used for all intemallines so that the electrons can interact via phonons or Coulomb
interactions, then (3.299) is exactly equal to (3.297). The two results represent the same set of
corrections. They are not different contributions but are the same contribution expressed in
different notation.
These formulas are often used to calculate the ground state energy in the limit of T ~ O.
There are two limits which are taken: v ~ 00 and T ~ O. In studying the electron gas, Kohn
and Luttinger (1960) observed that the right answer was obtained if they were taken first in
the limit of v ~ 00 and then T ~ O. The reverse order omits important terms, which are of
the form
(3.300)
Ifv i= 00, then all levels are discrete in the finite volume, and as T ~ 0, these terms give zero
since either nF or [1 - nF] is zero. However, if one first takes v ~ 00, so that the levels are
continuous, then the limit T ~ 0 gives
(3.301)
There are many terms of this kind in the perturbation expansion for the ground state energy of
the electron gas. They are called dangerous diagrams. They do not threaten your health, but
they do threaten the accuracy of your answer.
It is illustrative to show a simple example of evaluating the thermodynamic potential.
Assume that the only effect of the phonon self-energy is to change the unperturbed
frequencies oo~ to a new set of renormalized frequencies O~. Since the Green's function is
200q 200q
D = (.)2
lOOn - OOq2 - 2ooq1t(p, lOOn
. ) = (.lOOn )2 - 1"\2
u q
(3.302)
This form for D can be accomplished choosing
2OOq1t(11, p, iOOn) = 11(0~ - oo~) (3.303)

The choice (3.303) is not the only possible one to renormalize the frequencies. It is the choice
one gets by assuming that the change from OOq to Oq is accomplished by the one-bubble
polarization diagram given in (3.289). Then the coupling constant 11 enters as just a multi-
plicative factor, as shown in (3.303). Iffurther self-energy diagrams are needed to get a good
phonon self-energy 1t, then 11 would enter in a more complicated fashion. It is shown in
Chapter 6 that the single-bubble approximation is often adequate, so that the present deri-
vation applies to many systems.
Consider the steps necessary to evaluate (3.292), or perhaps just (3.288). The expression
to be evaluated is
0 2- 002
o- 0 0 = - - 1 L J1 d11 q q (3.304)
2~q.iq. 0 (iqni - oo~ -11(0~ - oo~)
First, introduce a frequency which depends on a coupling constant:
O~ = oo~ + 11(0~ - oo~) (3.305)

The summation over Matsubara frequencies may be done by the techniques suggested in Sec.
3.5. Alternately, one might recognize the answer as
1 1 1 - 1
A ~. 2 2 = 20 1)T)('t = 0) = - 20 [2nB(0T)) + 1] (3.306)
I-' lq. (lqn) - OT) T) T)
where i>T)('t) is the Green's function which is the Fourier transform of

- 20T)
1)T)(iqn) = (. )2 _ 1"\2 (3.307)
lqn UT)
The evaluation of (3.304) is now just the coupling constant integral,
o- 1
0 0 = -2 L(Oq2- OOq)
q
2 J1 21"\
d11 ( 1 + -'p=n--
OUT) e~-1
2) (3.308)
which can be done exactly since (n 2 - ff(2)dll/(2nTJ) = dnTJ:
n - no = ~ L[nTJ + ~ In(1 - e_~n")]TJ=l (3.309)

2 q ~ TJ=O
= ~U(nq - ffiq) +~ InC =:=:::)] (3.310)
The right-hand side is the thermodynamic potential from the phonons at the new frequency
nq minus the contribution from the phonons at the old frequencies ffi q. When this result is
combined with (3.260) for no, the terms with ffiq all cancel. The final answer is
2VJ --In(1+e-
n=-- d 3p ~~ p)
~ (2n)3
d 3[
+ -v -
~
J(2n)3
q -1 ~n + In(1 - e-'"Rn q)]
2 q
(3.311)
The thermodynamic potential is just the summation of the unperturbed electron contribution
plus the contribution from the phonons at the new frequencies nq . Of course, this form is
expected from simple considerations. If the only effect of the electron-phonon interactions is
to change the phonon frequencies to the new values nq , then one must be able to solve the
Hamiltonian exactly and write it as
(3.312)
The thermodynamic potential for this simple Hamiltonian is just (3.311). Unfortunately, most
corrections to the thermodynamic potential are not as easy to evaluate as this simple example.
3.6.2. Green's Functions

So far the linked cluster, or cumulant, expansion has only been used to evaluate the
thermodynamic potential. One could use similar expansion techniques to evaluate other types
of correlation functions, such as Green's functions (Brout and Carruthers, 1963). This method
has been used with success on two problems. One is the polaron problem for a single particle
in a band, studied by Dunn (1975). The other has been in treating X-ray processes, where the
hole dynamics have been studied by a linked cluster method (Mahan, 1975). Here the hole
has also been treated as a single-hole problem. It has not been applied to a many-polaron, or
many-hole, problem. The procedure begins with the S-matrix expansion for a particle Green's
function, given in (3.196):
00
9(p, T) =L AnWn(p, T)
n=O
(3.313)
where the coupling parameter A has been added to the expansion. In the cumulant expansion,
it is assumed that the series can be regrouped as an exponential power series in A:
9(p, T) = g(O)(p, T)exp[,~ AnFn(p, T)] (3.314)

The functions Fn(P, t) are found by equating tenns with like powers of "A in the series
expansions. That is, write
eXPC~A!FJ) = 1 +"AFI + "A2(F2 +~F?)

(3.315)
so that, tenn by tenn, we have the equalities

WI = g(O)FI (3.316)
W2 = !:/'
f'fO) (
F2 + 2!1 FI 2) (3.317)
(3.318)
The functions Wn(p, t) are scalar functions which are calculated using Wick's theorems. The
unknown functions Fn(P, t) can be evaluated in tenns of Wn by solving the above equations
(Brout and Carruthers, 1963):
WI
FI = gO) (3.319)
W2 1 2
F2 = gO) - 2iFI (3.320)
W3 1 3
F3 = - - FIF2 --FI (3.321)
gO) 3!
The basic idea is very simple. It is just a different way of regrouping the series which results
from the S matrix. It provides an alternative to using Dyson equations. Presumably, one gets
the exact answer if all tenns are taken and evaluated. However, since one can only get the
exact answer in trivial problems, the method is really useful in problems where one can
calculate a few tenns in the perturbation expansion.
There are two possible difficulties with this procedure. The first is that the regrouping
may not make mathematical sense. Nonsensical results are found in cases where g(p, t)
describes a Green's function of a particle in a system with Ne like particles, where Ne <X v. In
that case the tenns Fn do not converge when put into the exponent. The offending tenns are
those which contribute to the total energy of the system and hence change the chemical
potential. These energy tenns are of order O(v). The chemical potential is not changed by V
in one-particle or one-hole problems, since the one particle does not contribute any energy
tenn proportional to v. The series in Fn(P, t) shows excellent convergence in these cases.
These results will be discussed in Chapters 7 and 9. The other problem with this method is
that, when it works, one has some fairly complicated functions of t in the exponent. One is
invariably forced to use a computer to perfonn the Fourier transfonns in frequency space.
This step is not really a difficulty-just some work!
The physical model implied by this regrouping will be discussed in Chapter 7. It is a
systematic development of the Tomonaga (1947) model of pion emission in particle theory.
Tomonaga assumed that all pion emissions from a particle were statistically independent-
there was no correlation between successive emissions. His model corresponds to the
approximation of just using FI alone. However, if one uses FI + F2, then F2 puts in the
correlation between pairs of emissions. Similarly, F3 puts in correlations between three
particle events, etc. The advantage of the method is that the Green's function includes many-
particle emission processes, even when one keeps only a few terms in the series for Fn. Solid
state problems do not have pions but other boson excitations such as phonons or the boson-
like excitations of the electron gas.
3.7. REAL-TIME GREEN'S FUNCTIONS
Six different Green's function of time were introduced in Chapter 2. Two of them, the
retarded and advanced functions, have been discussed at nonzero temperature. The other four
are discussed here. The usefulness of the real-time functions is in the treatment of non-
equilibrium phenomena. Nonequilibrium transport theory using real-time Green's function is
treated in Chapter 8. The present section discusses their properties at nonzero temperature.
The Matsubara method is unsuitable for nonequilibrium since there is no thermo-
dynamic basis for temperature for a system out of equilibrium. The entire Matsubara method
is based upon temperature, and no method has been found so far for extending it to non-
equilibrium processes.
The real-time functions at nonzero temperature have formal definitions very similar to
those at zero temperature. Comparing with (2.144), the Green's functions for electrons of
momentum p are
G«p, t l , t2) = i(C:cr(t2)Cpcr(tl»)

G>(p, t l , t2) = -i(Cpcr(tl)C:cr(t2»)
G,(p, t l , t2) = 8(t l - t2)G>(p, t l , t2) + 8(t2 - tl)G«p, t l , t2)
Gi(P, t l , t2) = 8(t2 - tl)G>(p, t l , t2) + 8(tl - t2)G<(P, t l , t2) (3.322)
These definitions appear to be identical to those at zero temperature. The important difference
at nonzero temperatures is that the brackets (... ) have a different meaning. They have several
different meanings, depending upon the circumstance:
Case 1: At zero temperature, and in equilibrium, the brackets (... ) denote the ground
state of the interacting system.
Case 2: At nonzero temperature, and in equilibrium, the brackets denote the thermo-
dynamic average as given in (3.25) or (3.82).
Case 3: When not in equilibrium, the brackets denote an average over the accessible
phase space. However, the available phase space depends upon the recent
history of the system, and kinetic constraints such as energy conservation. The
meaning of the brackets is poorly understood for systems out of equilibrium.
Another feature of (3.322) is that the Green's functions are not expressed on the left as
functions of (t l - t2) but of (tl' t2) separately. In equilibrium, they are functions of (t l - t2),
but that may not be true for systems out of equilibrium.
In thermal equilibrium, the real-time functions each have a simple relation to the retarded
function. Each is found easily from Gret . Since Gret is found easily from the Matsubara
Sec.3.7 • Real-Time Green's Functions 155
function, the easiest way to find the real-time functions is by first finding the Matsubara
function of complex frequency. These statements apply only in thermal equilibrium.
In thermal equilibrium the following derivation establishes the relationships among the
real-time Green's functions. Using the states In) and 1m), which are exact eigenstates of the
Hamiltonian, two of the real-time functions in (3.322) can be written as
(3.323)
nm
G>(p, tl - t2) = _i2>-~Enl(nICp(J"lm)12eit(En-Em)(tl-t2) (3.324)

nm
The time dependence is Fourier transformed to m, which produces the delta function
2no(m + En - Em) in each term. These terms have many of the same factors that occur in the
spectral function in (3.120). They differ mainly in the thermal factors exp( -~En,m)' They can
be made to be identical by utilizing some algebraic relationships such as
e-~Emo(m + En - Em) = e-~Em[l + e~ro]nF(m)o(m + En - Em)

= [e-~Em + e-~En]nF(m)o(m + En - Em)
In the first line the factor of nF(m) has been added to both numerator and denominator. In the
second line, one of these factors has been changed to the sum of two exponential factors by
using Em - m = En" The part in brackets is exactly the combination that appears in the
spectral function:
(3.325)
The other real-time functions can be derived using the same technique:
G>(p, m) = -i[1 - nF(m)]A(p, m) (3.326)

Gt(p, m) = [1 - nF(m)]Gret(p, m) + nF(m)Gadv(p, m) (3.327)
Gt(p, m) = -[1 - nF(m)]Gadv(p, m) - nF(m)Gret(p, m) (3.328)
These expressions are identical to (2.160). In equilibrium the real-time Green's functions are
found easily from the retarded and advanced functions. The latter two are found easily from
the Matsubara functions. In equilibrium there is no need to set up a separate formalism for the
real-time functions.
The primary usefulness of the real-time Green's functions is in the theory of non-
equilibrium phenomena. Some of these ideas and techniques are introduced here. The first
step is to derive Dyson's equation. In fact the result (2.157) at zero temperature is also correct
for nonzero temperature. The matrices have the definitions that are given in (2.156). There is
no reason to rederive the same equation. Instead it seems appropriate to comment on the
differences in the derivation between zero and nonzero temperatures.
Equation (2.157) is generally regarded as being the correct form for Dyson's equation,
even for systems out of equilibrium. However, we have never seen a satisfactory proof for
nonequilibrium systems. For systems in equilibrium, one can derive (2.157) at nonzero
temperature by starting from Dyson's equation for the Matsubara Green's functions. Treating
'tas a complex variable, one can deform the contour of integration, and end up with (2.157)
(Langreth, 1976).
For nonequilibrium, the Matsubara functions are obviously an invalid starting point.
Instead, one can try to expand the S matrix in real time. Then one encounters the problems
mentioned in Sec. 3.1 regarding expanding two S matrices: one for exp(-~H) and another
one for exp(±itH). All derivations ignore the former. Including only the S matrix of time, one
again finds (2.157). It appears correct to omit the S-matrix expansion of exp( - ~H). This
point is discussed in the next section. Equation (2.157) is used for Dyson's equation for
nonzero temperature, for equilibrium and nonequilibrium, since there is nothing else avail-
able.
Nonequilibrium theory usually proceeds by first deriving an equation of motion for the
Green's function, which is similar to a Boltzmann equation. This equation is usually derived
for systems in equilibrium, or slightly out of equilibrium. Then the equation is applied to
systems far from equilibrium.
The first step is to find an equation of motion for the interacting Green's function when it
is not in equilibrium. Such an equation can be found from (2.157) by operating by
(ia/at - H) on both sides of the equation. First one needs to know the time evolution of the
noninteracting Green's functions. This behavior is deduced easily from (2.144) and (2.148).
The time derivatives can be expressed compactly in the matrix notation. Remember that
d0(t)/dt = oCt).
(i ata) - t) = oCt)!-
- Go(k,
Ek (3.329)
( i ata - Ho(x) )-Go(x) = 04 -

(x)! (3.330)
On the right-hand side of (2.157), this operator acts only upon Go for which the above result
is used to find
(3.331)
This formula provides the equation of motion for the interacting Green's function. It is the
basis for the nonequilibrium theory of interacting systems.
The structure of this equation is interesting. On the left-hand side, only the noninter-
acting terms are contained in the Hamiltonian Ho. The contribution from the interactions is
provided by the self-energy functions on the right-hand side.
So far, the equation of motion has been derived for the variable Xl in G(XI' X2). It is
useful to have an equation of motion for the other variable x2. Since the definition (2.144) of
the Green's functions contains the conjugate wave function '"t (X2), the equation of motion on
this variable is the complex conjugate of Schr6dinger's equation. Furthermore, Dyson's
equation can be written in an alternate form. Instead of (2.157), an alternate form is to have
Go on the right in the interaction terms:
The equation of motion on the X2 variable, when using this form for Dyson's equation, still
acts only upon Go on the right and produces delta functions. These steps produce the alternate
equation of motion for the Green's function:
( -i ~- HO(r2' -P2)) G(Xl' x2) = O\,Xl - X2)!
+ Jdx3G(xl , x3)f(X3' x2) (3.332)
The sign change on P2 comes from the fact that the left-hand side of the above equation is the
complex conjugate of Schrodinger's equation. The operator P = -iV changes sign under
complex conju~ation. This behavior is different from the Hermitian conjugate, where Ho acts
*
to the left on (X2) and P does not change sign. These two equations of motion will be used
in Chapter 8 to develop a quantum Boltzmann equation for nonequilibrium phenomena. The
quantum Boltzmann equation is based upon a many-body distribution function first suggested
by Wigner (1932).
3.7.1. Wigner Distribution Function

The traditional Boltzmann equation is expressed in terms of the distribution function
f (r, v, t). The three variables are position, r, velocity, v, and time, t. The point of view is
semiclassical since it is assumed that the position and velocity (momentum) of the particle
can be defined simultaneously. In order to use this distribution for quantum systems, it is
necessary to perform some type of averaging in order to remove effects due to the uncertainty
principle.
If quantum effects are important, it is necessary to introduce another variable into the
distribution function. This variable could either be energy E or equivalently a frequency
00 = E. The resulting distribution function is often called a Wigner distribution function
f (k, 00; r, t). The velocity v has been changed to the wave vector k = mv. The distribution
function is derived from the Green's function G«Xl' X2) defined in Eq. (2.144). The Wigner
distribution function (WDF) is derived using the following series of steps. First go to a center-
of-mass coordinate system:
(R, T) = !<Xl +x2) (3.333)
(r,t) =Xl -X2 (3.334)
Note that T means center-of-mass time, rather than temperature or time-ordering. The symbol
~ is used for the inverse of Boltzmann's constant times temperature. The notation on the
Green's function is altered to these center-of-mass coordinates:
(3.335)
The next step is to Fourier transform the relative variables (r, t) into (k, (0):
G«k, 00; R, T) = Jd 3re-ik. JdteimtG«r, t; R, T)

r (3.336)
The relation to the WDF is quite simple:
f(k, ro; R, T) = -iG«k, ro; R, T) (3.337)
Regard this assertion as the definition of f(k, ro; R, T). This choice is made reasonable by
showing that various moments of f provide the macroscopic quantities of particle density
nCR, T), particle current jiR, T), and energy density nE(R, T):
d3k Joo dro

nCR, T) = J-(21t)
3
-00 1t
t
-2 f(k, ro; R, T) = (\jJ (R, T)\jJ(R, T»)
k Joo dro
j(R, T) = J-d3k
3- -2f(k, ro; R, T) (3.338)
(21t) m -00 1t
d3k Joo dro

nE(R, T) = J-(21t)
-3
-00
-2 rof(k, ro; R, T)
1t
= i[-aa (\jJt(R, T - !t)\jJ(R, T + !t»)]

t 1=0
= (\jJt (R, T)H\jJ(R, T») (3.339)
The technique for solving nonequilibrium problems is very simple. The equation of motion in
(3.331) or (3.332) for G«k, ro; R, T) is just the quantum Boltzmann equation (QBE). This
equation is then solved, which yields directly the Wigner distribution functionf(k, ro; R, T).
Macroscopic variables such as current and density are found by taking the above integrals
over d 3 k and dro. This technique is also useful in systems that are not homogeneous because
there is a slowly varying potential VCR).
The above expressions are useful for nonequilibrium situations. For example, the answer
is zero if the equilibrium function (3.325) for G< is used to calculate the currentjl1 in (3.338).
That is the right answer. A system carrying current can be in steady state but not in equili-
brium. In order to obtain a nonzero current it is necessary to solve for the nonequilibrium
Green's function G< when the Hamiltonian contains an electric field. This procedure is
described in Chapter 8.
The semiclassical Boltzmann distribution function f(R, v, T) is found by taking the
frequency integral of the WDF:
feR, v, T) = J OO
-00
dro
2
1t
f (mv, ro; R, T) (3.340)
This method of deriving the semiclassical Boltzmann equation is an alternative to the usual
technique of coarse grain averaging.
The Green's function G< plays a central role in the nonequilibrium theory. In order to
solve it one also needs to know the retarded function.
These two Green's functions are the most important. The matrix equation in (3.331) and
(3.332) is untangled to present the individual equation of motion for these separate Green's
functions.
[i a~t - Ho(Xt)] G«Xt, X2) = Jdx3[1:t(xt, X3)G«X3' X2)

-1:«Xt, X3)Gi(X3' X2)]
[i a~t - Ho(xt) ] Gret(Xt ,X2) = ~4 (Xt - X2)
J 31: (Xt, X3)G (X3' X2)

+ dx ret ret (3.341)
[-i a~2 -Ho(X2)*]G«Xt,X2) = Jdx3[GlXt,X3)1:«X3,X2)
- G«Xt, X3)1: i(X3' X2)]
[ -i i.
at - HO(X2)*] Gret(Xt ' X2)
2
= ~\Xt - X2)
(3.342)
The equations are the starting point for the derivation of the quantum Boltzmann equation.
The WDF is an elegant and useful formalism. It does have some liabilities. If one
constructs the WDF for any nontrivial Hamiltonian, thenf(r, 00; R, T) has regions where it is
negative. The function is not positive definite, which means that it cannot be interpreted as a
probability density. This feature can be shown by a simple example. Consider in one
dimension a particle in a box of length L. The eigenvalues and eigenfunctions are
(3.343)
(3.344)
Using the representation in (2.147), and setting t = tt - t2 , gives
G<(xt, X2, t) = ~ L nF(En) sin(knxt) sin(knX2)e-itw2 (3.345)

n
Here xl,2 are not four-vectors but one-dimensional coordinates of position. The product of
two sine functions can be written as the difference of two cosine functions. Taking the Fourier
transform of time gives the function of frequency. Since the system is in equilibrium, write
G< = inF(oo)A. In the center-of-mass notation
f(x, 00; X) = nF(oo)A(x, 00; X) (3.346)

21t
A(x, 00; X) =T ~[cos(knx) - cos(2knX)]~(oo - En2) (3.347)
where x = Xl - X2 is relative position and X = (xl + x2)/2 is center-of-mass position. The

function A(x, ro; X) is a sum of delta functions. The amplitudes of the delta functions have
variable sign, depending upon the values ofx and X. Neither A norf are positive at all points.
For many Hamiltonians of interest the function f is not positive definite. That creates
philosophical problems iff is interpreted as a probability function. Do not interpret f this
way. Instead, f is regarded as one step in the calculation. It is never the last step, since it is
used to calculate other quantities that can be measured. Physically sensible results are always
found for measurable quantities such as the particle density and current. No problems are
encountered as long as one avoids interpreting f as a probability density.
3.8. KUBO FORMULA FOR ELECTRICAL CONDUCTIVITY
Many experiments in condensed matter physics measure the linear response to an

external perturbation. The experimentalists may put the sample in a magnetic field, electric
field, optical field, temperature gradient, or pressure field and measure the magnetization,
electrical current, light absorption, or whatever. Linear response means that the signal is
directly proportional to the intensity of the external perturbation. Usually the assumption of
linear response is valid at low magnitude of perturbing field.
Kubo formulas are the name applied to the correlation function which describes the
linear response. There are many of them, since there are many possible perturbations and
many linear responses for each perturbation. Formulas of this type were first proposed by
Green (1952, 1954) for transport in liquids. Kubo (1957, 1959) first derived the equations for
electrical conductivity in solids and in the form which will be used here. His derivation and
result will be given below. Other Kubo formulas will be derived in this section. In fact, several
derivations are given of the formula for the electrical conductivity in order to familiarize the
reader with different types of derivations in the literature. One finds formulas which appear to
be quite different but are really identical. It is also demonstrated that the longitudinal and
transverse conductivities are identical at long wavelength.
In electrical conduction, a time-dependent external electric field,
(3.348)
is applied to the solid, where (J( = X, y, z are the directions in space. In linear response, the
induced current is proportional to the applied electric field
(3.349)
Although the symbol cr' has the appearance and dimensions of a conductivity, it is not the
appropriate conductivity. Instead, the conductivity is needed which is the response to the total
electric field in the solid. The applied or external field F!aext) induces currents which in tum
make other electric fields. The summation of all these fields is the total electric field, which is
called Eir, t). The conductivity cr is the one which responds to the actual electric field in the
solid:
Jir, t) = L: cr(Xp(q, ro)Ep(r, t) (3.350)

p
E(X(r, t) = S(X exp[i(q· r - rot)] (3.351)
Sec. 3.8 • Kubo Formula for Electrical Conductivity 161
Take Eq. (3.350) as the fundamental definition of the microscopic conductivity. It also
introduces the coordinated problem of determining the proportionality between the external
electric field E(ext) and the total internal electric field E,ir, t). For a static, longitudinal electric
field, the two electric fields are related by the continuity of the normal component of D at the
surface. A similar relation for transverse fields is derived below. Equation (3.350) is correct
for a homogeneous material. Actual solids are not homogeneous, although crystals are
periodic. However, (3.350) implies the spacetime response may be written as
(3.352)
where repeated indices imply summation. This equation is the Fourier transform of (3.350). It
assumes that the current response of a material at r is only a function of the separation (r - r')
from the external electric field at r'. This assumption is incorrect on an atomic scale. It is
certainly not valid in the atomic-like core states of a solid or in any case where the separation
(r - r') is a few angstroms. A rigorous formulation would have the conductivity as a function
ofr and r' separately, cr(r, r'; t - t'). In solids, it is permissible to use (3.350) only when it is
understood that the current is to be averaged over many unit cells of the solid. Usually it is
applied when q is small and long-wavelength excitations are being studied.
Quite often our interest is in the dc conductivity, which is obtained by taking the limits
q -+ 0 and (0 -+ 0 in that order. Then the conductivity is only real. The Kubo formula is
derived assuming that only a single frequency is perturbing the system and that crap(q, (0) is
the response to this single frequency. Actually, using a single frequency is not a restriction. It
is assumed that the system is linear and perturbations at different frequencies act indepen-
dently. The total current is then the summation of the responses at different frequencies.
The Hamiltonian for the system is taken to have the form
H+H' (3.353)
The term H' contains the interaction between the total electric field and the particles of the
system. Equation (2.161) is used as the basic form of the interaction between electromagnetic
fields and charges. The electric field is expressed as a vector potential, so that
H' = - ~ J
d 3rja(r)A,ir, t) (3.354)
1 -i
-Aa(r, t) = -Ea(r, t) (3.355)
C (0
Select the Coulomb gauge, where V . A = 0, as explained in Sec. 2.10. Also, the electric and
vector potentials are taken to be transverse, so the scalar potential <P is set equal to zero. The
terms in (2.161) with A2 are dropped, since their effects are nonlinear in the electric field. The
term H contains all the other terms and interactions in the solid or liquid. There are inter-
actions such as electron-electron, electron-phonon, spin-spin, with impurities, etc. The goal
is to calculate the electrical conductivity when all these other interactions are present.
The current operator in (3.354) was discussed in Chapter 1. It has the form of
(3.356)
The notation is more compact if the dr integral is evaluated in (3.354), which gives H' in
terms of the Fourier transform of the current operator:
H' = ~j,xCq)Eece-iCDt (3.357)

0)
1 ..
jec(q) = 2m ~ei(piece,q·r,
, + e,q·r'Piec] (3.358)
In terms of creation and destruction operators, the current operator is conventionally written
as
jec(q) = LP~AO)Ct CO (3.359)
AO
where A, 0 are the states associated with some unperturbed Hamiltonian Ho which is chosen
as the basis for the perturbation expansion. Several of these possibilities were discussed in
Chapter 1. A distinction is made between the current operator jec in (3.358) and the induced
current J" in (3.350). The operatorj" is used in the Hamiltonian, while J" is the actual current
measured by the experimentalist. The measured value of the current is the average value for
the velocity of the particles in the system, which is taken as the summation over all the
particle velocities divided by the volume:
Jec(r, t) =:'(L
v
vi"O(r - r i») =:. L(Viec)
i v i
(3.360)
When quantizing the particle velocity in (1.410), the velocity is the momentum minus the
vector potential:
v· =
,
!m [po' -:.c A(r.)]
'
(3.361)
e e- (3.362)
J" = mv~(Pi") - mcv~A,xCri)
The momentum operator Pi is proportional to the current operator, j = ep;/m. The last term
uses the relationship (3.354) between the vector potential and the electric field. For this term,
the summation over particles divided by volume is replaced by the density, which is appro-
priate for a long-wavelength disturbance:
. .noe-
J,,(r, t) = (j,xCr, t») + l-E,xCr, t)
mO)
(3.363)
The second term in the current is proportional to the electric field and the first term given by
the expectation value of the local current operator. The latter term is also proportional to the
electric field. Later it is shown that the constant of proportionality is given by the Kubo
formula.
These two terms are called J(l) and JC2):
J = J(l) + J(2) (3.364)
JCl) = inoe- E(r, t) (3.365)

mO)
J(2) = (j(r, t») (3.366)

The next step is to derive J(2), which is the derivation of the Kubo formula.
3.S.1. Transverse Fields, Zero Temperature

The following derivation of the Kubo formula is valid at zero temperature. The first step
is to consider the expectation value of the current operator as a function of time:
.!J.2)(r, t) = (\jIllei(H+H')~i.x(r)e-i(H+H')tl\jl') (3.367)
The Heisenberg representation is used, as explained in Sec. 2.2. Next go to the interaction
representation, where H' is treated as the perturbation.
e-i(H+H')t = e-itHU(t) (3.368)
U(t) = eitH e-i(H+H')1 (3.369)
J~2)(r, t) = (\jI'lUt (t)eitHj", (r)e- itH U(t)W) (3.370)
The operator U(t) was defined earlier in (2.14), and a formal solution was derived in (2.27):
U(t) = T exp [ -i 1: dt'H' (t')] (3.371)
where the usual definitions are

H'(t) = eitHH'e-itH (3.372)
j(t) = eitHje-itH, etc. (3.373)
The wave function 1\jI'} in (3.367) is the Schrodinger wave function at t = 0 for an interacting
system with both H + H' as the Hamiltonian. As discussed in Sec. 3.1, the wave function I\jI}
is appropriate when H' is absent from the system. The relationship between the two wave
functions is [see (2.36)]
(3.374)
The time development of the system is given by combining these results:
U(t)I\jI'} = Texp [ -i Loo dt'HI(t')}\jI} (3.375)
= S(t, -oo)I\jI} (3.376)
The expectation value of the current is now

J~2) = (\jIISt (t, -00»",(r, t)S(t, -00)I\jI) (3.377)
An exact solution of this equation is not needed for the derivation of the Kubo formula. Only
terms linear in the electric field E", are desired, which requires us to only keep terms linear in
H'. It is sufficient, for linear response, to keep the terms
S(t, -oo)I\jI) = [1 - Loo dt'HI(tl)] 1\jI} + O(H'i

i (3.378)
The Hermitian conjugate of this wave function is
(\jIISt (t, -00) = (\jI1 [ 1 + i Loo dt'H' (tl)] + O(H'i (3.379)

From (3.377), the expectation value of the current operator is
J~2)(r, t) = Loo dt'H' (t') ]jc,(r, t)[ 1 - i Loo dt' H'(t')}W)

(wi( 1 + i
= (wi( j(X(r, t) - i Loo dt'U(X(r, t)H'(t') - H' (t')jir, t)] }W)
The first term is assumed to vanish,

(wlj(X(r, t)lw) = 0 (3.380)
since there is usually no current in the solid in the absence of an electric field or something
equivalent such as a time-varying magnetic field. The first nonzero term is the one linear in
H', which is the important contribution. There are two terms, which can be expressed as a
commutator:
J~2)(r, t) = -i Loo dt' (wIUir, t), H'(t')]lw) (3.381)
The goal is to derive Eq. (3.350). A factor of E~(r', t') must be removed from the integrand.
With H' given in (3.358), the integrand has the factors
i . t'
[j(X(r, t),H'(t')] = -3~e-1ffi Uir, t),j~(q, t')]
co
= ~E~(r, t)e-iq.reiro(/-t'JU(X(r, t),h(q, t')] (3.382)
co
which are regrouped in the following way:
J~2) = ~E~(r, t)e-iq .r fl dt'eiro(/-t'J (wi [j(X(r, t),h(q, t')]lw)

co -00
A factor of exp[i( q . r - cot)] was taken outside of the integral and, to compensate, the inverse
of this factor was left inside the integral. Comparing this result with (3.350) shows that they
have the same form. The current JI2) is proportional to the electric field. The constant of
proportionality is the conductivity:
cr(X~(q, co) = -1 e-'q·r

. fl . ( ')
dt' e'ro I-I (wIUir, t),j~( q, t')]lw)
co -00
no~
+i--O(X~ (3.383)
mco
However, this result is not quite right. The final step is to average over the space variable r in
order to eliminate atomic fluctuations. Take this average by integrating over all volume d 3 r
and then divide by v. The only r dependence of the expression are the factors
fd re-iq.rj(X(r, t)
3 = j(X( -q, t) = j!( q, t) (3.384)
The final result for the conductivity is
cr(X~(q, co) = -
Ifl .
dt'e,ro(/-t'J(wl[j!(q, t),h(q, t')]lw) + i_o-o(X~
n~
(3.385)
COy -00 mco
Equation (3.385) is the Kubo formula.
The correlation function in (3.385) is only a function of the time difference t - t'. If this
difference is made the variable of time integration (- (' -+ t, the formula can be expressed as
which is the way it is usually written. The right-hand side is definitely not a function of t.
The wave function 1\)1) in (3.385) is the ground state of the many-body Hamiltonian H.
This Hamiltonian contains all the possible interactions in the solid except the interaction with
the vector potential H'. The conductivity is calculated using (3.385), which has no mention of
photon field. The conductivity is an intrinsic property of the ground state of the system.
Equation (3.350) can be viewed as a Taylor series in the applied electric field ext) 3b
Ji3 p) = J,,(O) + (~~) 3bext) + O(3bext»)2 (3.386)
3b
The conductivity cr"p = (aJj83 p) is calculated for ext) = o. It is a characteristic of all linear
response correlation functions that they are ground state properties. A major difficulty is that
the ground state \)I is not known for most many-body systems. Then one must evaluate the
correlation function using the S-matrix expansion and Green's function analysis, which has
been described in Chapter 2 and in this chapter. Equation (3.385) is often the starting point for
a many-body calculation, as shall be demonstrated in later chapters.
The Kubo formulas contain a retarded, two-particle correlation function. From the
definition of the retarded Green's function in (3.88), the retarded correlation function of the
current operator is defined as
II"p(q, t - t') = - ~0(t - t')(\)II[J!(q, t),ip(q, t')]I\)I) (3.387)

v
Its Fourier transform is
Comparing this definition with (3.338), we find that
(3.388)
The conductivity is the retarded correlation function of the current multiplied by i and divided
by ffi. The correlation function II"p( q, ffi) is usually called the current-current correlation
function.
It is usually easiest to calculate the retarded correlation function in the Matsubara
formalism. The procedure follows the prescription outlined in (3.105) and (3.110). First
define the equivalent current--current correlation function in the Matsubara formalism:
1
II"p(q,1:) = -- (T,J!(q, 1:)Jp(q, 0)) (3.389)
v
II"p(q, iffin) = J: d1:eiOl"II"p(q,1:) (3.390)

The Matsubara function is evaluated as best one can, perhaps by using the diagrammatic
techniques described in Sec. 3.4. Then the desired retarded function is obtained from
change
. + .~ ll,,~(q, iron) -+ ll,,~(q, ro)
Iron -+ ro lu
(3.391)
The conductivity is found from (3.388). These steps are always used in calculating the
conductivity.
The dc conductivity is obtained by taking the limit q -+ 0 and then the limit ro -+ O. The
wrong answer may be obtained if the order of these limits is reversed, which may be
understood on physical grounds. The limit ro = 0, q i- 0 describes a static electric field,
which is periodic in space. Here the charge will seek a new equilibrium, after which no
current will flow. It is usually important to first take the limit of q -+ 0:
cr,,~(q, ro) = cr,,~(ro)
lim { ll,,~( q, iron) = ll,,~(iron)
(3.392)
q--+O ll,,~(q, ro) = lllX~(ro)
i,,(q, t) = iit)
The limit q -+ 0 presents no problem. The current operator is well behaved in this limit, so
that the correlation functions are well behaved. In fact, when only the dc conductivity is
needed, it simplifies the derivation to set q = 0 at the beginning of the calculation.
The limit of ro -+ 0 is more delicate. Here the conductivity is real:
Recr,,~ = - lim .!..Im[ll,,~(ro)]

ro--+oro
(3.393)
The right-hand side contains the imaginary part of the retarded correlation function. The
combination -2Im[ll,,~(ro)] is just the spectral function of that operator; call it R,,~(ro). By
using the general formula (3.117) for the spectral function, a formal solution for Imll is
-Im[ll,,~(ro)] = !R,,~(ro)
= ~(1- e-~ro)e~a L>-~EN(nli!lm)(mlhln}
v nm
(3.394)
Now it is straightforward to take the limit ro -+ 0, since the prefactor of the above equation is
lim.!..(1 - e-~ro) = ~ (3.395)

ro--+oro
and
(3.396)
The conductivity is finite as ro -+ O. There is no divergence in this limit, even though there is
the ro- 1 factor in (3.393).
It is a curious feature of the Kubo formula that, in order to calculate the dc conductivity,
it is necessary to calculate the ac conductivity and then to take the limit ro -+ O. At least this is
the easiest way to do it. Later another Kubo formula will be derived strictly at ro = O.
However, the latter formula is cumbersome to evaluate, and it is faster to evaluate cr( ro = 0)
by the procedure of taking the ac conductivity and the limit that ro -+ O. It is also easier, at
zero temperature, to use the Matsubara formalism and then to take the limit of T --+ 0 at the
end. Below it is shown that (3.385) is the right Kubo formula at nonzero temperatures.
The current-current correlation function is a two-particle correlation jUnction. The
current operator contains a product of one creation and one destruction operator, so the
correlation function (3.390) contains at least four such operators:
The correlation function describes how two particles are created and destroyed. The
conductivity arises from correlations between these two events.
The conductivity is a measurable quantity. It is the same quantity as measured in a circuit
or by an optical probe. Measurable quantities always involve retarded correlation functions of
at least two particles. The one-particle Green's function can never be measured-at least not
in the rigorous sense. The one-particle Green's function describes a Gedanken experiment
which can never happen in practice. One creates a particle at time 't',
(3.397)
in state per and then destroys it at another time 'to One asks for the correlation between these
events. This experiment is impossible, since real particles cannot be created or destroyed. In a
two-particle Green's function, one describes a sequence of events which is realistic. A particle
(3.398)
is changed from one state to another, say from v to 11, and then at a later time it is changed
back. The two-particle correlation function, wherein a particle changes its state, is always
found in the correlation functions oflinear response. In elementary particle physics, a particle
can be absolutely destroyed or created. But this event always happens in conjunction with
some other event which involves other particles. For example, an electron and positron can
mutually annihilate and make several photons. Then one would have terms in the current
operator involving the creation or destruction of two particles:
(3.399)
Even in this case, the current-current correlation function involves four operators: two for one
particle and two for another. If the particles interact, i.e., by a Coulomb interaction, then the
correlation function may not be divided into two independent Green's functions:
(3.400)
We shall encounter current operators of the form (3.399). First, positron annihilation is an
important experiment in metal physics (Brandt and Dupasquier, 1981; Mills et al., 1986). The
form of (3.399) is also used in other contexts. If an electron is destroyed in a filled band, the
resulting excitation is called a hole. This terminology is used in the filled valence band of a
semiconductor and also in the core levels of all solids. There is an identity
filled ) empty )
C" electron = dt hole (3.401)
band band
where d~ is the creation operator for the hole. When taking an electron from a filled band and
moving it to an empty or partially filled band, then one has a current operator of the form
(3.399), which is used to describe these electron-hole excitation processes.
3.S.2. Nonzero Temperatures

The preceding derivation of the Kubo formula has several restrictions. It was limited to
zero temperature and transverse electric fields. Both of these restrictions are unnecessary,
since (3.385) is correct for both nonzero temperature and longitudinal electric fields. At
nonzero temperature, the bracket is interpreted as a thermodynamic average. The restriction to
transverse electric fields was actually unnecessary. One could use a gauge wherein the scalar
potential <p was zero but not the longitudinal vector potential. Then the longitudinal electric
field can be expressed in terms of the longitudinal vector potential. The derivation is the same,
step by step, with the same answer.
Another derivation is presented which is valid at nonzero temperature. A time-varying
perturbation H'(t) is put into the system at nonzero temperature. The central question
concerns the degree to which the thermodynamic averaging is influenced by the time-varying
interaction: is H' put into the thermodynamic weighting factor exp[~(Q - H - H' + vN)] as
well as into the time development of the operators? As an example, is the following equation
a correct expression for the time development of the current operator?
(3.402)
Although this expression is reasonable, it is believed to be wrong. It is incorrect to just put H'
into the thermodynamic weighting factor. If the time oscillation is fast enough, the heat bath
will not follow the oscillations. The oscillatory field causes louIe heating, but that contri-
bution is proportional to the square of the electric field, and is not part of the linear response.
The above formula may be manipulated by moving the right-hand operator in the trace to
the left, and by expressing the time operators as a time-dependent density operator:
J(2) = Tr[p(tli] (3.403)

p(t) = e-it[H+H'(tl]ef3(n-H-H'(t)+!1N)eit[H+H'(tl] (3.404)
However, since the middle exponential factor also contains the operator combination H + H',
it commutes with the time operators, so that the above predicts
p(t) l: e f3(n-H-H'(t)+!1N) (3.405)
The time dependence of the current operator plays no role if the above expression is correct. It
was derived by assuming H' goes into the thermodynamic weighting factors. Since our
intuition is that the time development of the current operator should not cancel, the above
ansatz is incorrect.
Kubo's derivation of the time-dependent density matrix proceeds in the following
fashion. The density matrix Po = exp[~(Q - H + !IN)] applies to the equilibrium system in
the absence of H'(t). It is assumed that the system is described by this density matrix at the
initial point in time, which is taken at t = -00. The perturbation H' is adiabatically switched
on as the system is brought forward in time to the present. Now the time-dependent density
matrix, for the interacting system including H' (t), is called pet). It obeys a Heisenberg
equation of motion, which when solved yields an expression for the electrical current:
d .
dt pet) = -z[H + H'(t), pet)] (3.406)
J~2) = Tr[p(t)j/X] (3.407)
Kubo's starting point is tantamount to not including H'(t) in the thermodynamic weighting
factor. The density matrix is defined as pet) = Po +I(t), where Po is the density matrix in the
absence of H'. The equilibrium density matrix Po is time independent so that the above
equations simplify to
z. dt
d 1= [H, Po] + [H,f] + [H , Po] + [H'f]
I
, (3.408)
[H, Po] = 0 (3.409)
The objective is to solve for the term in J(2), which is proportional to H', which is treated as
infinitesimal. Since I is proportional to H', it follows that I is small. Terms proportional to
O(H'i, such as [H' ,f], are neglected.
i :t l = [H,f] + [H', Po] (3.410)
This equation is solved by moving the first term on the right to the left:
i :t l - [H,f] = [H', Po] (3.411)
and then expressing the left-hand side as
(3.412)
The linear differential equation may be integrated to give
i~(eitHle-itH) = eitH[H', po]e-itH = [H'(t), Po] (3.413)

dt
I(t) = 1(-00) - ie- itH {Loo dt'[H'(t'), Po] }eitH (3.414)
The interaction is switched on slowly in time, so that at t = -00 there is no interaction, and
H' = O. Of course, then/(-oo) = 0, since this term only exists when H' exists. A solution
for I has been derived which is proportional to H', which is adequate for a description of
linear response.
The evaluation of (3.407) is
(3.415)
where the first term on the right vanishes (Tr[poja] = 0) since there is no current if there is no
external applied field. The expectation value of the current, proportional to H', is the
remaining term:
J~2)(r, t) = -iTr{ e-itH[Loo dt'[H'(t'), po]}itHjolr )} (3.416)
= -iTr{Loo dt'[H'(t'), po]eitHjolr)e-itH } (3.417)
= -lTr{Loo dt'[H'(t'), PO]ja(r, t)} (3.418)
The quantity in curly braces was rearranged by using the cyclic properties of the trace. The
factors on the right are just the time development of the current operator: by using the cyclic
properties of the trace, the three operators can be rearranged:
Tr[H'(t'), PO]ja(r, t) = Tr[H'(t')pojolr, t) - poH'(t')ja(r, t)]

= Tr{po[ja(r, t)H'(t') - H'(t')jaCr, t)]}
= Tr{po[ja(r, t), H'(t')]} (3.419)
The term in curly braces is just a commutator. The final result is derived for the expectation
value of the current, which is just the thermodynamic average of the commutator of H' (t') and
jaCr, t):
J~2)(r, t) = -i Loo dt' ([jaCr, t), H'(t')]) (3.420)
Equation (3.420) is precisely the zero-temperature equation which was derived earlier in
(3.381). It leads, in the same way, to the Kubo formula (3.385). Previously the angle brackets
meant to take the expectation value in the ground state at zero temperature, while now they
mean a trace over the thermal distribution at nonzero temperatures. Equation (3.385) is valid
at nonzero temperatures.
3.8.3. Zero Frequency

The Kubo formula (3.385) has been shown to be valid at nonzero temperatures and for
transverse electric fields. Also, it was suggested that it is valid for longitudinal electric fields.
However, the purist might still object. All the derivations introduce a frequency (i), and the dc
conductivity is obtained by letting 00 ~ 0 at the end of the derivation. Is it possible to provide
a derivation of the dc conductivity which assumes a constant electric field throughout the
derivation? Of course, the answer is affirmative, which provides the motivation for the next
derivation, using a method suggested by Luttinger (1964).
Begin by assuming an electric field Eo which is static in time and constant in space. It
causes a scalar potential
<p(r) = -Eo' r (3.421)

and the external vector potential is zero. This scalar potential introduces a perturbation term
into the Hamiltonian which is
rft F J
= rft p(r)<p(r) (3.422)
where per) is the charge density operator of the system. A factor of exp(st) was added to this
term so that it vanishes as t ---+ -00. This factor represents the switching on of the potential as
time develops from the past. It is assumed that the perturbation on the density matrix,
p = Po + jrft (3.423)
also has this term. The first term Po = exp[p(Q - H)] is the density matrix when F is absent.
The perturbationje't is due to F. The starting point is the equation of motion for the density
operator,
(3.424)
which is solved as before. The term proportional to the perturbation F is given in (3.414):
jest = -i Loo dt'e-iHtrft'[F(t'), po]eitH = -i Loo dt'rft'[F(t' - t), Po]
It is convenient to change integration variables to t' - t =} -t:
j = -i J~ dte-st[F( -t), Po] (3.425)
The crucial trick is to write the following identity for the commutator:
[F( -t), Po] = -ipo Jo~ dp' -ataF( -t - iP') (3.426)
The identity is easy to prove,
a
Jo~ dp' -F(-t - iP') = -i
a
J~ dP'-,F(-t - iP')
at 0 ap
= -i[F( -t - iP) - F( -t)]
so that
i2 Po[F( -t - iP) - F( -t)] = -po[e~H F( -t)e-~H - F( -t)]
= -[F( -t), Po] (3.427)
The expectation value of the current is the expression
J~(r) = Tr[fj~(r)] = - J~ dte- st J: dP'Tr[po aF(-~t- iP')j~(r)] (3.428)
The reason that the trick identity is introduced is that one can now do the following
manipulations to produce the electric field. First, the time development of F( -t) is given by
F( -t) = Jd rp(r, -t)<p(r)

3 (3.429)
The scalar potential is time independent. The time derivative of the density is related to the
current through the equation of continuity:
aF( -t) =
at
Jd3r<!>(r) ap(r,at -t) (3.430)
= - J d 3r<!>(r)V . j(r, -t) (3.431)
Integrate by parts on the space variable, and thereby obtain the electric field Eo:
aF(-t)
-a-t- =
Jd rj(r, V<!>(r)
3 -t) • (3.432)
= - Jd rj(r, -t) . Eo
3 (3.433)
JIJ.(r) = J~ dte-st J: dWTr[PO Jd3r'j~(r', -t - iW)Eo~jo:(r)]

Since the electric field is a constant, it can be removed from the integrals, and the conductivity
is given by
Re(crIJ.~) = J~ dte-st J: J
dWTr[po d 3r'h(r' , -t - iW)Mr)]
The right-hand side appears to depend on the special variable r. This dependence is elimi-
nated, as before, by averaging over the volume v of the sample. This manipulation just
introduces the q -+ 0 limit of the current operator j IJ. (q)
(3.434)
The Kubo formula for the dc electrical conductivity is
Re( crIJ.~) = ~ J~ dte-st J: J

dWTr[po dVh( -t - iW)jIJ.] (3.435)
The result does not contain the frequency. The elimination of (0 has been achieved by paying a
penalty: there is a second integration, this time over the temperature-like variable W. The
second integration makes the results appear different than the prior result (3.385). The two
results are identical for (0 = O. To prove this assertion, again introduce the representation
In), 1m}, which are exact eigenstates of the Hamiltonian H. The important matrix element is
(nlh( -t - i~')lm) = (nle-iH(t+i~l)heiH(t+iP')lm) (3.436)
(3.437)
so that
(3.438)
Both integrals can be done easily. For the real part of the conductivity, the only complex term
is the t integral:
Re[Joo dte-i(En-Em-isl] = Re[. 1 . ] = 1to(En - Em)

o I(En - Em - IS)
~
8(En - Em) J0 dW e~ (En-Em)
, = ~8(En - Em)
which makes the integral over Weasy. The final result is
Re(crll(~) = 1t~~::=e-~En(nli~lm)(mlill(ln)8(En -Em)

V nm
i!
This equation is identical to the earlier result (3.396), since = ill( for q = O. It has been
proved that (3.435) is identical to the ill = 0 limit of the usual Kubo formula (3.385). In the
limit of ill = 0, the longitudinal and transverse conductivities are identical.
3.8.4. Photon Self-Energy

A formula which is identical to the Kubo formula can also be derived from the photon
self-energy. Consider the photon Green's function (3.78),
V!'vCk, 't) = - (T,A!,(k, 't)AvC -k, 0)) (3.439)
and treat as the perturbation
V= - ~ Jd r[j(r) • A(r) + ~:~) A(d]

3 (3.440)
1 i2no
=- JV~j(k)' A(k) + 2m ~A(k)' A(-k) (3.441)
The A(ki term contributes a self-energy of IT(1) = noi2 1m. The other term in the self-energy
comes from the n = 2 term in the expansion for the S matrix for the j . A interaction,
(3.442)
x (T,A!,(k, 't)A,(k', 't1)Am(k", 't2)Av( -k, 0))
= L J~ d'tl J~ d't2V~~)(k, 't - 't1)IT'm(k, 'tl - 't2)V~~(k, 't2) (3.443)

k 0 0
where
1
IT'm(k,'t1 -'t2) = --(T,j,(-k,'tI)im(k,'t2)) (3.444)
v
1
il
= - - (T, (k, 'tl)im(k, 't2))
v
(3.445)
The Fourier transform of this expression gives the self-energy contribution as
(3.446)
Recall Eq. (2.186) for the dielectric function, in retarded or Matsubara notation:
.
c(k, Iron) = 1- 411:
-.--2
[ lICk, Iron)
. + -no~J (3.447)
(IroJ m
cret(k, ro) = no~J

2 [ lICk, ro) + -
1 - -411: (3.448)
(ro) m
where II is the scalar part of the self-energy 1I(X~ for homogeneous, isotopic systems:
(3.449)
The long-wavelength dielectric function is given by the conductivity
. 411:cr(ro)
c(ro) = k---+O
11m crel(k, ro) = 1 - --.-
lro
(3.450)
Comparing (3.448) with (3.450), the expression for the conductivity cr(ro) in terms of the
retarded correlation function 1I(ro) is
no~J
cr(ro) = ~i [ lI(ro) +--;;;- (3.451)
Since II is the current-current correlation function, Eq. (3.388) is rederived.

The important point is reemphasized that, in evaluating the Kubo formula for the
current-current correlation function, do not include the self-energy terms which arise from
the vector potential A(r, t). These terms have already been included in the formalism,
indirectly, by changing the external field E(ext) to the total field E. They do not reenter the
formalism again except as internal interactions within the polarization bubble.
3.9. OTHER KUBO FORMULAS
There are other measurements besides electrical conductivity. They require additional
correlation functions or Kubo formulas. Some of these will be derived here. The first is
magnetic susceptibility. Next comes the derivation of transport coefficients such as thermal
conductivity or thermoelectric power.
3.9.1. Pauli Paramagnetic Susceptibility

The magnetic susceptibility of a solid depends rather strikingly on whether the system is
already spontaneously magnetized. If it is, then an additional weak, external magnetic field
may not change the magnetization appreciably-there may be no linear response. This case is
not considered, and it is assumed the solid is not in a magnetic state; it is not ferromagnetic,
antiferromagnetic, etc. A magnetic state has long-range order, although there is usually short-
range ordering, from the interactions, even in the absence of long-range order.
Sec. 3.9 • Other Kubo Formulas 175
The perturbation term in the Hamiltonian for Pauli paramagnetism has just the inter-
action with the electron spin:
v= -m . Hoe- iro / (3.452)

H(t) = Hoe- iro / (3.453)
ma = gllo L Si,a (3.454)
where H(t) is the ac magnetic field and m is the magnetization, which is the sum over all spins
Si a at position i with vector direction a. The other factors are the Bohr magneton 110 =
le'IJij2mc and gyromagnetic ratio g ~ 2. Another contribution to the magnetization is the
orbital, or Landau, contribution, which is not discussed here. In Pauli paramagnetism the
objective is to evaluate the magnetization,
(3.455)
for terms linear in the magnetic field. The derivation proceeds exactly as in the preceding
section. In analogy with (3.381), the magnetization is a correlation function
Mit) = -i Loo dt' ([mit), V(t')]} (3.456)
and the magnetic susceptibility is
(3.457)
Equation (3.457) is just the Fourier transform of a retarded correlation function. It may be
evaluated conveniently in the Matsubara formalism. Define a correlation function
(3.458)
and (3.457) is just the retarded function obtained by the analytic continuation iron -+ ro + i8.
The spin operators are often written in terms of raising and lowering operators:
S(±) = Sx ± isy (3.459)

S(+) -- Let
k
c
kt k,(,
(3.460)
The three operators S(+), SH, Sz are used instead of Sx, Sy, Sz. Of course, susceptibility
functions can also be constructed using these operators.
The correlation function will be evaluated for a simple example. It is the Pauli
susceptibility of a free-electron gas at zero temperature. The answer is first derived using
simple considerations. In a magnetic field Ho in the z direction, the spin-up and spin-down
electrons have the following energies and occupation functions
~t(p) = ~p - 1l0Ho, nt = nF(~p - 1l0Ho) (3.461)

~,!Jp) = ~p + 1l0Ho, n,(, = nF(~p + 1l0 H o) (3.462)
The net magnetization will be the difference between the number of spin-up and spin-down
electrons. The occupation numbers are expanded in a power series in the magnetic field.
Keeping the first two terms gives
(3.463)
(3.464)
The Pauli susceptibility Xp is given by the square of the Bohr magneton 110 times the density
of states NF at the Fermi energy. This density of states is the total for both spin components.
The notation is confusing, since often the same symbol is used as the density of states for
each spin component, so that the value is only one-half as large.
The same result is obtained from the correlation function. The operator rna is given in
terms of creation and destruction operators as
rn z = 110 L aC~aCpa (3.465)

pa
where a = ± is used to denote electron spin up or down. The Kubo formula for the corre-
lation function for the static susceptibility (iro = 0) is
(3.466)
For the noninteracting system, the Hamiltonian is just the kinetic energy term, so that
cta(t) = et~kCta and Cka(t) = e-t~kCka' Then the operator cta(t)Cka(t) has no t depen-
dence, so the t integral just gives ~. The combination of operators can be averaged directly.
Some care must be taken with the terms which have a = a' and p = k. Using the fact that
(ni) = (nk) if both refer to the same spin state, the answer is
(3.467)
(3.468)
The second term on the right vanishes, since there are as many up spins as down spins in
equilibrium. The correlation function is evaluated, of course, in the absence of any external
magnetic field. There remains the combination of operators
(3.469)
This formula has exactly the same combination of factors as obtained previously, which
confirms the result of the simple derivation.
Another method of evaluating the correlation function is to define the wave-vector-

dependent susceptibility X(q, im) by
X",~(q, im) = f: dr:ei())n' (T,miq, r:)m~(-q, 0)) (3.470)
mz(q) = 110 L (J"C~+q,crCpcr (3.471)

pcr
(3.472)
(3.473)
Evaluate Xzz for nonzero values of q and im = 0 and formally take the limit q ~ 0 at the end
of the calculation:
For noninteracting electrons, this correlation function is nonzero only if s = s', and then it is
identical to the polarization operator p(l)(q, 0) which has been evaluated previously:
(3.474)
Now take the limit of q ~ 0 on the right-hand side. There results anF(~p)/a~p in the
summation over p. All methods give the same result for the noninteracting electron system.
3.9.2. Thermal Currents and Onsager Relations

There are two conceptual difficulties in deriving a formula for the thermal conductivity.
The first is that one must put a temperature gradient on the solid. However, all of our thermal
averaging assumes a constant temperature. Of course, the thermal conductivity is still defined
in the limit that the temperature difference I1.T goes to zero, so that the correlation function
can be evaluated at a single temperature. But the problem lies in formally deriving the
correlation function without assuming that I1.T exists somewhere in the formalism. Another
way to express this difficulty is to consider how one raises the temperature by heating the
sample. The heating changes the energy of the system, so that one might try to calculate the
time rate of change of the energy. But the energy operator is the Hamiltonian, and its rate of
change is if = i[H, H] = 0, which does not help much. This problem has been well studied in
nonequilibrium statistical mechanics (de Groot, 1952, 1962), and it has been found that the
important quantity is the rate of change of the entropy.
Another problem with the thermal conductivity is that there are several definitions of
heat current. We shall mention three. The same definition of experimental quantities is
obtained for the thermal conductivity, thermo-electrical power, etc., regardless of which
definition is adopted. But one must be careful to do the right calculation for each definition of
heat current.
In linear response, there are current J i which flow as a result of "forces" Xi on the
system. These forces might be temperature gradients VT, or electric fields E = - V V, or
concentration gradients which are expressed as gradients of the chemical potential VJl. Linear
response assumes these are proportional:
J.I = "Z(ij)
L...
.X
1 (3.475)
j
The coefficients Z(ij) are the measurable constants which are defined in terms of correlation
functions. There are also Onsager relationships, which specify that Z(ij) = Z!ji). A moment's
reflection will show that the Onsager relationship is not valid for any arbitrary choice of
currents and forces. For example, if they are valid for a force V(1IT) = -(VT)IT2, then
changing the force to VT, so that 11T2 is absorbed in Z, means the Onsager relation is no
longer valid. There must be a criterion for choosing the forces and currents, which has been
given in nonequilibrium statistical mechanics by de Groot (1952). In a nonequilibrium
process, there is a net generation of entropy, so that asI at > O. Here S is just the part of the
entropy which is generated irreversibly. By requiring that entropy generation be expressed as
(3.476)
then the Onsager relations are valid. There is not a unique set of currents and forces, and
many possibilities satisfy (3.476). Each choice defines a different set of coefficients Z(ij), but
each set obeys the relationship Z(ij) = Z!ji).
The symbol J will refer to particle current. There are several definitions of energy
current. The discussion will follow Barnard (1972). The first energy current which comes to
mind is the one called the energy current J E' For a free-particle system it is just the velocity of
the particles times the energy times the number:
(3.477)
(3.478)
Other definitions were given in Sec. 1.2 for particles which were not free. For a system with
only these two currents, their forces are (Luttinger, 1964)
XI =-fVV-V(~)
J2=J E, X2=V(~) (3.479)
where e is the charge of the particle. The rate of entropy production is
~~ = -J. [fVV + V(~)J +JE · v(~) (3.480)
and the linear response equations (3.475) are
(3.481)
J
E,a
= -M(21)[-=-V V + V
a~ T ~ ~
(1:)J
T + M(22) V (~)
a~ ~T (3.482)
With these definitions, the Onsager relation is M~~2) = M~~I).

The trouble with (3.477) is that the energy current is not the heat current. As discussed
by Taylor (1970), the heat current should be
(3.483)
which for free particles has the form
i Q = l::=Vpnp(Ep - /l) (3.484)

pO"
so that the reference energy is the chemical potential /l. The latter definition makes sense only
for particles with a positive chemical potential, e.g., electrons in metals. If one takes an
electron below the chemical potential and moves it down to the other end of the sample, one
has moved energy. But when the electron arrives at the new location, it finds locally a filled
Fermi distribution, so that the only states available to it are above the Fermi energy. But the
energy gain required to increase its energy, to get it above the Fermi energy, must come from
the surroundings. The energy gain from the surroundings must cool the locality, so that the
electron has brought coldness with it. Only electrons which arrive with energy above the
Fermi energy, and give energy to the surroundings, bring heat.
With this choice of thermal current, the forces are
(3.485)
where ~ = /l + eV. With this choice, the rate of entropy production is
(3.486)
and the linear response equations are
(3.487)
(3.488)
where L(12l = L(21). It is easy to show that these equations are consistent with the set (3.479)
and (3.481). In (3.479) regroup the terms,
e
-VV+V
T
(/l) e
- =-VV+-V/l+/lV
T T
-
T
1 (1) =-V/l+/lV
T
1 _ (1)-
T T
so that as/at becomes
~~ = -J . [~V~ + /lV (~) ] + J E . V (~) (3.489)
= -~J. Vii + (JE - /11), v(~) (3.490)

which is identical to (3.486). Similarly, in order to find the relationship between the coeffi-
cients M(i}), and L(i}), operate on the current equations (3.481) with the same regrouping
(3.491)
(3.492)
Next, consider the following combination of these equations:
(3.493)
This equation for J Q•rx now has the same form as the currents in (3.487) which gives the
relationships
L(1I) = M(1I) (3.494)
L(21) = L(12) = M(12) _ ~(11) (3.495)
L(22) = M(22) _ 2~(12) + I.J? M(ll) (3.496)
A third possible definition of heat or energy current is the choice

J w = JE + eVJ = JQ + ~J (3.497)
Here one includes the fact that the potential VCr) may not be constant in the sample, so that a
charged particle which moves will acquire a change in potential energy. Determining the
forces which go with this choice of energy current is a homework assignment.
The thermal conductivity is usually measured under conditions of no particle current
J = O. From (3.481), this condition leads to
(3.498)
which for the energy current gives
(3.499)
The M tensors are being treated as scalar quantities, which is permissible if all the forces are
in the same symmetry direction. Since the thermal conductivity K is usually written as
(3.500)
(3.501)
Equations (3.475) and (3.476) and the constraint as/at> 0 can be used to show that the
right-hand side is always positive (Kubo, 1959). If there is no particle flow, then J Q = JE . The
thermal conductivity can be found from (3.487) in a similar fashion. The restriction that no
current flows gives:
1
TVft =
L(12)
L(ll)V
(1)T (3.502)
The heat current and thermal conductivity are
JQ = V (T1) [22 (L(l2))2]

L()-lJIil (3.503)
_ ~[ (22) _ (L(12))2]
(3.504)
K ~ T2 L L(ll)
It is easy to use the identities (3.495) to show that this definition is consistent with (3.501).
The electrical conductivity is usually defined when there is no temperature gradient
VT = 0 and no concentration gradient Vj.l = 0:
2
(J = ~L(ll) (3.505)
T
The coefficient L(l2) may also be measured, since it is the thermoelectric coefficient. If a solid
has a temperature gradient I1T, no particle currents J = 0, and no concentration gradients
(Vj.l = 0), then a voltage difference I1V is measured which is proportional to I1T and is given
by
1 L(12)
I1V = ---(ll)I1T
eTL
(3.506)
This equation is the definition of the thermoelectric coefficient S, which is also called the
thermopower:
11 V 1 L(l2)
S=--=-- (3.507)
I1T eT L(ll)
The thermopower S may have either sign. The sign usually indicates the sign of the charge on
the conducting particle. The correlation function is also related to the Peltier coefficient
(Barnard, 1972).
3.9.3. Correlation Functions

The coefficients MCii), and L(i)), are correlation functions of current operators. They are
expressed by formulas similar to those of the preceding section. One wishes to obtain the
measurable currents J; by evaluating the expectation value of the operator j; of the same
quantity:
(3.508)
The derivation is identical to the steps used to get (3.428). One introduces a term F into the
Hamiltonian, and this term causes the change in the density matrix. In direct analogy with
(3.428), the result is
(3.509)
The only remaining step is to make the proper choice for aF / at. It is the time rate of change of
the energy of the system. At least it is the dissipative part brought on by the transport and
forces. It is the heat production or equivalently the irreversible rate of change of entropy S:
aF dQ as 1 •
- = - = T-=-LJI,XI (3.510)
at dt at ~ I
The small ji symbols are for current operators, not their averages. Using this result for aF/ at
in (3.509),
(3.511)
Since the forces XI are constants, they may be removed from the integrals. Then J i is
proportional to Xl> and the constants of proportionality are just the transport coefficient:
zg~ = - ~ J~ dte-Sf J: dWTr [Po Ai-t - iW)ji.~(r)] (3.512)
Equation (3.512) is a very simple result. The transport coefficient between force XI and
current J i is just the current-current correlation function of jl and ji' For example, the
coefficients in (3.487) are
(3.513)
(3.514)
L~2:) = - ~ J~ dte-Sf J: dWTr [PojQ.(X( -1 - i~/)jQ.~] (3.515)
The M(if) coefficients in (3.481) are similar; the only difference is that jE is used instead of jQ'
Since jQ = jE - J.lj, it is easy to show that the relations (3.495) are automatically satisfied. It is
also easy to prove the Onsager relations. First one writes the correlation function in the In)
and 1m) notation of (3.396):
z<iT) = 7t L e-~En (nl jllm) (ml jiln)8(En - Em) (3.516)
nm
z<li) = 7t L e-~En (nl jdm) (ml jlln)8(En - Em) (3.517)
nm
These two expressions can be shown to be identical by interchanging dummy summation
variables n and m and using En = Em.
These correlation functions can also be evaluated by using the Matsubara formalism.
One just evaluates the correlation function
(3.518)
by the usual S-matrix techniques. After a suitable result is obtained, one gets the retarded
function by letting iCOn -+ CO + io and then zero frequency by letting co = O. It was shown in
Sec. 3.7 that these steps are identical to the evaluation of (3.512).
Sec. Problems 183
PROBLEMS
1. Prove the following moments of the spectral functions:
foo dOJ 2
-OJ nF(OJ)A(p, OJ)
-00 21t
t
= -([H, Cp,,][H, Cp,,]) (3.519)
foo
-00
dOJ OJnB(OJ)B(q, OJ) = (A(q)[H, A( -q)])
21t
(3.520)
Evaluate these moments for noninteracting particles and phonons.
2. Take the Fourier transform of the correlation function (3.1). Define its spectral function as C(p, OJ).
Show that
A(p, OJ) = (1 + e-Pru)C(p, OJ) (3.521)
What is the result when (3.1) is time-ordered?
3. Let U(t) be the density operator p( q, t) of wave vector q. Prove that

oo
=f
dOJ
Seq) -nB(OJ)S(q, OJ) (3.522)
-ex 21t
where Seq) = (p(q)p(-q)) is the pair distribution function and Seq, OJ) is the spectral function for the
density-density correlation function.
4. At zero temperature,
(a) Show that np = f~00(dOJ/21t)A(p, OJ).

(b) Evaluate this integral for a Lorentzian: A(p, OJ) = 2r p/[(OJ - ~p)2 + r~].
(c) Plot np vs. ~p for cases (1) rp = ro = constant, and (2) rp = rol~pl/Jr6 + ~~.
5. Let CJ be a creation operator describing a spinless boson whose number is conserved; this model is
sometimes applied to 4He atoms if one can ignore the internal structure (excitations, degrees of
freedom). The particles will have a chemical potential I!. Define
(3.523)
(a) Show that gO)(p, iOJ) = l/(iOJ - ~p).

(b) What is the definition of the retarded function?
6. Prove the following general theorem for particles which are fermions or bosons:
Gret (
p,)
t . foo
= -le(t) dOJ A(p, OJ )e-irut
-2 (3.524)
-00 1t
7. Evaluate the thermodynamic average for the case where the operators represent bosons
( ~ a~Cr)akCr) ~ a~cr/)apcr/)) (3.525)
by doing it (a) directly and (b) using Wick's theorem. Both should give the same answer. Terms with
p = k have to be treated carefully.
8. Use the rules for constructing diagrams to write down the two-phonon self-energy diagrams of the
electron: ~(2)(p, ipn)'
9. The two diagrams in Fig. 3.12 contribute to the electron self-energy and arise from the electron-
electron interaction. Use the rules for constructing diagrams, and write down these contributions in the
Matsubara fonnalism. However, do not evaluate them, i.e., do neither the frequency summations nor the
wave vector integrals.
10. How is (3.215) altered if the internal line is a Coulomb interaction?
11. Derive the summation fonnulas (3.217) and (3.218). Also evaluate
1
R LV(O)(q, ioon)V(O)(k, ioon + iqm) (3.526)
J-' lOOn
12. Evaluate the Matsubara frequency summation for

1
R L gO)(p, ipn + iqm)V(q, iqm) (3.527)
P Jqm
13. Prove that
(3.528)
where gO) is for electrons and no is the free-electron part of the thennodynamic potential (Luttinger and
Ward, 1960).
14. Consider the tenns Ws for the electron-phonon interaction. How many times do each of the tenns
appear: Ul, ui,
U6 U2 ' U4 U}?
15. Evaluate W](p, t) in (3.314) using V(t]) as the electron--electron interaction. Which tenns would
not make sense if one tried to write this as exp(F])?
16. Show that at nonzero temperatures the real part of the conductivity can also be written as
Re[cro~(oo)l =
1 - e-~w
2
Joo .
dte,IWU!(q, t)j~(q, O)} (3.529)
oov -00
17. Find the forces Xi which go with the currents J, J w in (3.497). What are the Kubo fonnulas for the
transport coefficients?
18. Show that the relations (3.495) are consistent with (3.512).
FIGURE 3.12
Sec. Problems 185
19. Show that the second-order phonon self-energy, from the electron-phonon interaction, may be
written as II(q, (0) = MJP(l)(q, (0):
p(I)(q, (0) = 2 Jd 3P3 np ( I . _ I .)

(2n) (0 - Ep + Ep+q + 18 (0 + Ep - Ep +q + 18
Evaluate this expression by doing the wave vector integrals. Assume a free-electron gas of Fermi
momentum kF at zero temperature. Give both the real and imaginary parts of p(1).
20. Derive the change in the thermodynamic potential when the electron self-energy is ~(ll) = ll~o.
21. Derive the change in the thermodynamic potential when the electron self-energy is ~(ll) = 112~0.
22. The thermodynamic potential Q, internal energy U, and entropy S are related by Q =
U - TS -!!N, where
S= -(~~) ~,v (3,530)
Cv = T(:~) ~,v
(3.531)
Derive the expressions for S, U, and C v for the free-electron gas. Show that the specific heat is
proportional to T at low T, and find the constant of proportionality.
23. Derive the relationship between the total internal and external electric fields by solving Maxwell's
equations.
24. Show that Wick's theorem is invalid when Ho contains particle-particle interactions. As an
example, evaluate the correlation function W below for a system of two states. Do it exactly and also by
Wick's theorem.
2
Ho =L ~aC!Ca + Unln2 (3.532)
0=1
W = L(TTC!('I)C"('I)C~('2)C~('2)) (3.533)
"~
25. Derive the equilibrium real-time Green's functions in (3.327) starting from (3.322).
26. In one dimension find the Wigner distribution function for the simple harmonic oscillator of mass
m and frequency (00' In center-of-mass coordinates show that its spectral function is given in terms of
Laguerre polynomials
A(k, (0; X) = 4ne- ll L( _1)nLn(2i1)8[(O - (Oo(n + 1)]
n
x2 22 Z2 = _1_
i1=-Z2+ kZ ,
m(OO
27. Prove the following identity:
~ret(k, T) = - JOO d~' '8 [~>(k, T) - ~«k, T)]

+I
(0; (0'; (0';
-00 (0 - (0
This result is easy to prove in equilibrium. Can you also prove it for systems out of equilibrium?
Chapter 4
Exactly Solvable Models
Every many-body theorist should be knowledgeable about the available exactly solvable
models. First, there are not many of them. Second, they are useful for gaining insight into
many-particle systems. If the problem to be solved can be related to an exactly solvable one,
however vaguely, one can usually gain some insight.
Several models are presented here to introduce the discussion of Green's functions.
Usually, the model is first solved without Green's functions and again with them. Models
which are exactly solvable may be solved by a variety of techniques.
The models presented here are only a small sampling of the possible exactly solvable
ones. Others have been presented by Schweber (1961). Many one-dimensional models may be
solved exactly (Lieb and Mattis, 1966; Baxter, 1982; Mattis, 1993).
4.1. POTENTIAL SeATIERING
A very simple problem is an impurity potential V('t) in an otherwise free-particle system.

All other interactions are ignored except that of the free particles with the impurity, which is
assumed to be at the origin. The potential is assumed to have no internal structure: spin,
excited states, etc. It is a simple function of position r of the particle from the origin and is
spherically symmetric.
The wave functions \jf)., and energy f:)., for each particle may be obtained by solving a one-
electron Schr6dinger equation:
(4.1)
The solution may be accomplished by numerical means if the Schr6dinger equation is not
solvable analytically.
In many-body theory, the impurity problem is usually encountered as a scattering center.
The free-particle states are plane waves,
(4.2)
187
188 Chap. 4 • Exactly Solvable Models
The spin index cr is dropped, since spin effects are unimportant. The Hamiltonian is expressed
as operators
(4.3)
Vkk, = Jd rV(r)e-
3 ir '(k-k') = V(k - k') (4.4)
The last term is the potential scattering of the free particles. The object is to diagonalize the
Hamiltonian (4.3). Of course, the solutions are given by (4.1). However, the problem is not
entirely solved. Since the equation is second order, it has two solutions. These may be chosen
as the ingoing and outgoing waves. Alternately, these two may be combined to give standing
waves. These choices must be related to the scattering problem implied in (4.3).
The discussion will benefit from the knowledge that the solution is of the form (4.1).
Consider the boundary conditions on the wave function. The integral equation for the wave
function is:
(4.5)
(4.6)
(4.7)
This form of the integral equation is valid for the free-particle states with energy Ek' The
SchrOdinger equation (4.1) may also have bound states. They shall not be discussed explicitly.
They obey an equation similar to (4.5), but the energy is changed to the binding energy
Ek ---+ -EB (EB > 0), and the <l>k term on the right is absent.
The first step is to prove that (4.5) is equivalent to (4.1). Operate on both sides of the
equation by Ho - Ek'
and use the fact that
(4.8)
to get
(Ho - Ek)"'k = - L
k'
<l>k,(r) JdV <l>~,(r')V(r')"'k(r') (4.9)
The completeness relation for the summation over the set of states is
L <l>k,(r)<l>~,(r') = 8(r - r') (4.10)
k'
(4.11)
Sec. 4.1 • Potential Scattering 189
If the potential is transferred to the left, the final equation is
(4.12)
which is the desired answer. The above equation is precisely (4.1). It has been shown that
(4.5) is equivalent to the usual SchrOdinger equation. Next discuss boundary conditions. The
differential equation (4.1) is second order, so there are two independent solutions. As a
mathematical problem, these two solutions may be combined in any possible way, depending
on the choice of boundary conditions. As a physics problem the two solutions are combined
to give a desirable physical situation. One choice is to have the wave function a standing
wave. This choice leads to the reaction matrix equation. Other choices are to have the wave
function an incoming wave or an outgoing wave, which leads to T-matrix theory. From
scattering theory and causality a natural choice is the outgoing wave. It has the asymptotic
form exp(ikr)jr at large distance. This choice of phase [rather than exp( -ikr)jr] arises from
the physics convention that the time development is exp( -irot). The integral equation (4.5) is
a convenient starting point for this discussion of boundary conditions, since the various
choices of standing, outgoing, or ingoing waves are determined only by the complex part of
the energy denominator. The factor is f.k - f.k" for standing waves, so that the principal part is
chosen for the denominator. The factor is f.k - f.k' + io for outgoing waves, and f.k - f.k' - io
for ingoing waves.
4.1.1. Reaction Matrix

Here the energy denominator is chosen to be real and given by the principal part. The
free-particle Green's function is then defined as
-l. ()-l. * (r') P ik' • (r-r')

Go(k,r- r') =P:E'I'k' r'l'k' =_ :E_
e __ (4.13)
k' f.k - f.1c' V k' f.k - f.1c'
_ Jd k' e
- P
3 ik" (r-r')
--3----,- (4.14)
(2n) f.k - f.~
The last step on the right takes v ~ 00 and changes the summation into an integration over
wave vectors. The integral is standard and gives
G (k r - r') = n (k) cos[klr - r'll (4.15)

o ' P klr - r'l
A factor p(k) has been introduced which is the density of states of the particles:
(4.16)
For particles with spin, this factor is the density of states per spin configuration. In electron
systems, for example, the net density of states is twice this result when spin degeneracy is
considered. The Green's function may be expanded as a function of rand r':
Go(k, r - r') = np(k) :E(2l + 1)PI(r • r')Hkr<)l1l(kr» (4.17)

I
where the notation r < and r> means the lesser and greater of r and r', respectively. The jl(kr)
and 111(kr) are spherical Bessel functions of the first and second kind, and the PI(cos 9) are
Legendre functions. These results may be put into the integral equation (4.5):
J
Wk(r) = <her) + d 3r'Go(k, r - r')V(r')Wk(r') (4.18)
At this point it is convenient to reduce the integral equation to angular momentum compo-
nents I. A plane wave may be expanded as
eik • r = L: (21 + l)iPI(k' r)Hkr) (4.19)
I
A similar expansion is used for the actual wave function:

Wk(r) = L:(21 + l)i PI(k r)RI(kr)
0 (4.20)
I
The radial function RI(kr) is the quantity to be determined. It satisfies a radial Schrodinger
equation of the form
/1 2[10 oR 1(1 + 1) ]
- - - - ? - - - - R +[V(r)-ck]R=O (4.21)
2m r2 or or r2
which does not determine the boundary conditions. They are obtained by substituting these
forms into (4.18). Each angular momentum component is selected by multiplying the
equation by PI(k r) and then integrating over all spatial angles. One uses the fact that
0
Jdn'pl(kor)Pm(rop) = 2/4: 10ImPI(kop) (4.22)
to reduce the equation down to one which involves only the same angular momentum
component:
RI(kr) = jl(kr) + 4n2p(k) J~ r,2dr'Hkr<)l1l(kr>)V(r')RP,r')

= jl(kr) + 4n2p(k{ 11 I (kr) L r t2 dr'Hkr')V(r')R I(kr')
+ jl(kr) J~ r'2dr'111(kr')V(r')R[(kr')] (4.23)
It is important that the potential is spherically symmetric. Otherwise the scattering term would
mix angular momentum components, which usually makes the equation much harder to
solve.
The solution is examined in the limit as kr -+ 00. From the differential equation (4.21) it
can be shown that the radial wave function must asymptotically approach the value
J~oo RI(kr) -+ C~) sin [ kr + ol(k) - I;] (4.24)
The prefactor CI is determined below. The asymptotic limit of the integral equation (4.23) is
limR[(kr) -+ jljr) + D[(k)l1l(kr)
sin(kr - In/2) D cos(kr - In/2)
-+ - 1--'------'---'- (4.25)
kr kr
where
(4.26)
The potential VCr) is assumed to be of short range: It falls off faster than r- 2 at large
distances. This assumption is necessary for the integral in (4.26) to be well defined. The
scattering is then described by a phase shift D/(k) which depends on angular momentum and
wave vector. The two asymptotic expansions (4.24) and (4.25) must be identical, which is
accomplished by setting
DI = - tan(D/) = 41t2p(k) J~ ?d':h(kr)V(r)RI(kr). (4.27)
The complicated integral just defines the tangent of the phase shift. Now (4.25) becomes
lim RI(kr)
kr-+oo
~ kr cos
1 (D) [COS(D) sin(kr - 11t/2) + sin(D) cos(kr - 11t/2)]
sin(kr + DI - 11t/2)
= -.........,...-","=..,.-- (4.28)
kr COS(D/)
The normalization coefficient in (4.24) is
C - -I- (4.29)
I - COS(D/)
When solving the radial wave function (4.21), the solution is obtained which is well behaved
at the origin. This solution is followed outward in r until the region is reached where VCr) ~ 0
and the centrifugal barrier /j2/(1 + 1)/(2mr2) is small. Then the solution has the form (4.24)
with CI = 1/ COS(D/). The wave function is now properly normalized at large r, and by
following it back toward the origin it is normalized everywhere. These steps provide the
proper solution to the scattering equation (4.5) which was the starting point in the calculation.
The reaction matrix is defined as the quantity
(4.30)
It may be expanded in angular momentum states by using the expansions for the wave
function and the plane wave. The angular integrals give
Rk'k = 41t 2:(21 + I)PI(k· k)RI(k', k) (4.31)

I
where the radial integral is
RI(k', k) = J~ ?drHk'r)V(r)RI(kr) (4.32)
It is defined for the general case where k' =f:. k. Of course, if they happen to be equal, then the
answer is just (4.27):
R (k k) = _ tan(D/) = _ /j2 tan(D/) (4.33)

I , 41t2p(k) 2mk
where the density of states is p(k). The reaction matrix is only proportional to tan(D) for the
diagonal terms where k' = k. Otherwise, one must do the integral in (4.32).
The reaction matrix obeys an integral equation which is deduced by putting (4.30) into
the integral equation (4.5) for \)Jk(r)
(4.34)
This integral equation is frequently found in scattering problems. Now it is solved easily. For
each angular momentum state, the one-particle radial equation (4.21) is solved with the
boundary condition that the asymptotic limit goes to (4.24). This wave function is used in
(4.32), which determines the answer. It is a simple procedure to solve (4.34). The important
point is that the wave functions must be normalized correctly. The designation of principal
parts on the energy denominator in (4.34) specifies the unique way this normalization is to be
done.
4.1.2. T Matrix
The other common choice of boundary conditions uses outgoing waves, which is
accomplished by adding an infinitesimal complex part io to the energy denominator in (4.5).
The integral equation is
(4.35)
where the Green's function for outgoing waves is
Go(k, r - r)
, = Jd k'
3
--3
e- ik' . (r-r')
. (4.36)
(2n) Ck - Ck' + 10
This integral is evaluated to give
eikR np(k)
Go(k, R) = -np(k) kR = - ~ [cos(kR) + i sin(kR)] (4.37)
The cosine term was evaluated earlier in (4.17). The sine term is
sin(klr - r'l) = "(21 + l)P (r. r\' (kr\' (kr')

klr - r'l 7 I JlI JlI
(4.38)
These results can be put into the integral equation (4.35) to give an equation for the wave
function. Averaging over angles gives for each angular momentum state:
(4.39)
The radial part of the wave function is denoted by RI . It will be a different wave function than
found for the standing-wave boundary conditions. The difference between the two is only a
different choice of normalization coefficient C/(k) in (4.24). This different choice is what is to
be determined. The above integral equation may be rewritten as
Rikr) = j/(kr{ 1 + 47t2 p(k) J~ y'2dr'Tl/(kr')V(r')Rpcr')

-47t2 ip(k) J~ y'2dr'Nkr')V(r')R/(kr') ]
+ 47t2 p(k)Tl/(kr) 1: y'2dr'j/(kr')V(r')R/(kr') (4.40)
Then one can take the limit kr --+ 00 and obtain
lim R/(kr) --+ j/(kr)[1 - ii)/]

kr-+oo
+ D/(k)Tl/(kr) (4.41)
D/(k) = 47t2 p(k) J~ "zd':h(kr)V(r)R/(kr) (4.42)
This equation has to have the form (4.24) because the radial wave function R/(kr) also obeys
the differential equation (4.21). The result (4.41) has that form if
(4.43)
so that the factor multiplying j/(kr) is
(4.44)
For the asymptotic limit
lim R/(kr) --+ eio/[h(kr) cos(o/) - Tl/(kr) sin(o/)]

kr-+oo
eiO /
= kr [sin(kr -17t/2) cos(o/) + cos(kr - 17t/2) sin(o/)]
(4.45)
which is indeed the right form, (4.24). For outgoing wave boundary conditions, the proper
choice of normalization coefficient is
(4.46)
One should proceed by solving (4.21) and insist that it have this asymptotic limit. The easiest
way is to first solve it for a real wave function and make this real wave function go to
[sin(kr + 0 - 17t/2)]fkr [which determines o/(k)]. Then one can multiply the wave function
everywhere by the phase factor exp[io/(k)].
The T matrix is defined as
Tk'k = Jd3r<p~,(r)V(r)\iJk(r)
= 4n 'L(21 + l)P/(k' k')T/(k', k) (4.47)
I
TI(k', k) = J~ ?drHk'r)V(r)RI(kr) (4.48)
The original integral equation for the wave function (4.35) may be used to generate the T-
matrix equation
(4.49)
This equation is often encountered in scattering problems. The solution is easy. One solves
Schr6dinger's equation (4.l) for the wave function \jIk(r). Only the radial part of the wave
function is difficult, since one must solve (4.21). The solution is normalized by insisting that
the radial wave function have the form (4.24) at long distance from the potential. The
coefficient is CI = exp[i<>l(k)]. These radial solutions are used in (4.47), which then gives an
exact solution for the T-matrix equation. This result differs from the reaction matrix result
only in the choice of the coefficient CI which multiplies the radial wave function. The angular
momentum components of the two scattering functions are related by the ratio of these
normalization coefficients:
RI(kr) = eiO [ cos(<>I)Rlkr) (4.50)

TI(k', k) = eiO [ cos(<>/)Rlk', k) (4.51)
The T matrix is a complex quantity since it was defined with a complex exp[i<>/] phase factor.
The diagonal T matrix is
-1 '0 1 '0
Tlk, k) = 4n2 p(k) e' [ sin[<>/(k)] =- 2mk e' [ sin[<>/] (4.52)
We repeat again that this simple form is valid only when k = k'. The imaginary part of the
diagonal T matrix is
sin2 (<>I)
-Im[TI(k, k)] = 4n 2 p(k) (4.53)
-Im[Tkk] = np~k) y(21 + 1) sin2 [<>I(k)] (4.54)
which is related to the total scattering cross section for the potential:
(J = 4~ 'L(21 + 1) sin2 [<>/(k)] (4.55)

k I
One can also prove the optical theorem
(4.56)
It is important to realize that the reaction matrix is not identical to the real part of the T
matrix. Either of these two different quantities may be selected as the energy shift of a particle
interacting with a potential.
4.1.3. Friedel's Theorem

There are several important theorems involving phase shifts. These theorems are
conveniently proved by letting the solid have a spherical shape of radius R, with the impurity
located at the center of the sphere. The distance R is very large, and eventually the limit is
taken of R --+ 00. A finite size sample is used since we are going to count nodes in the wave
function. Start counting at the surface of the sphere and move inward. It is assumed, for
convenience, that the wave functions vanish at the surface of the sphere.
In the absence of the potential, the solutions which are regular at the origin have a radial
partj[(kr). The condition that they vanish at the surface of the sphere is
(4.57)
Since small values of I are important, say I < 5 for most applications, and since R is very
large, an adequate approximation at large argument is
jz(knR) --+ k:R sin ( knR _ I;) (4.58)
The condition for the vanishing of the wave function at the surface is
knR = (n +Dn (4.59)
For each value of I, there is a solution for each additional integer n, where each additional
integer has a solution with an extra node in the wave function. Similarly, the continuum wave
functions (i.e., for Ek > 0) in the presence of the impurity at large r is
RI(kr) --+ kr
In)
c[ SID. (kr + 01 - "2 (4.60)
The solution is obtained for each kn which satisfies the equation
knR + ol(kn) = (n +Dn (4.61)
Count the additional particle states in the presence of the impurity. The number of states dn
between k and k + dk is
dn R 1 dOl
-=-+-- (4.62)
dk n ndk
The first term Rjn is just what one would have without the impurity. The extra states On from
the impurity are given by the formula
dOn 1 dOl
= (4.63)
dk n dk
The quantity on the right (djdk)(8z(k)jn) is interpreted as the change in the number of
particle states caused by the impurity. For example, if the potential is repulsive so that
particles are pushed away from the impurity region, then dozl dk will be negative. Similarly,
an attractive potential will draw particles inward, so that dozl dk is positive. Many potentials
have d0zldk positive for some wave vectors and negative for others. The quantity dozldk is
the change in the density of states for each angular momentum state, each magnetic quantum
number ml, and each spin quantum number ms. The total change in the density of states is
obtained by summing over all these quantum numbers:
.!!... N -.!!... L (Olm,m,) (4.64)

dk - dk m ,m
,I 1t
In a metal, the electron states are occupied up to the Fermi wave vector kF . The Friedel sum
rule (Friedel, 1952, 1953) is obtained by integrating up to the Fermi wave vector,
(4.65)
where Z is the charge on the impurity. It is assumed that the phase shift is zero at k = O.
Levinson's theorem is that the phase shift is zero at k = 0 unless there are bound states, in
which case it is 1t times the number of bound states. The Friedel sum rule is a statement of
charge neutrality. In a static electron gas, there are no long-range Coulomb potentials of the
form r- i . Instead, the electron gas is displaced in the vicinity of an impurity charge. The
displaced electronic charge exactly cancels the impurity charge. For example, an impurity of
positive valence Z has electrons drawn in, until an extra Z of electrons, or eZ of charge,
surrounds the impurity. This extra charge is called a screening charge. It is shown in Fig.
4.1(a). The phase shifts are calculated for the screened potential of impurity plus screening
charge. If the potential is repulsive, electrons are depleted around the impurity, so that charge
neutrality is maintained. This depletion is shown in Fig. 4.1 (b). The Friedel sum rule is a
statement of charge neutrality: the change in electron charge around an impurity is exactly
equal in magnitude and opposite in sign to the charge of the impurity. A neutral impurity
would have a Friedel sum of zero. The sum rule does not mean that all the phase shifts vanish.
Generally, some are plus, some are minus, and all vary with the wave vector. Only the
summation at the Fermi energy is zero. Another way to express that the screening charge
equals the impurity charge is
Z = 41t J~ r2 dr[p(r) - Po] (4.66)
where Po is the equilibrium charge density in the metal.

The charge densities in Fig. 4.1 are schematic. The method of actually calculating these
curves is discussed in Chapter 5. All such curves, when calculated correctly, have the
oscillations in the charge density at large distance. These variations are called Friedel
oscillations. Their magnitude has been exaggerated in Fig. 4.1, since they are small in
amplitude. But they occur in real solids and have been observed by several techniques. The
(all >0 p(rl (bIZ<O
r r
FIGURE 4.1 Electron density per) when an impurity charge of Z is put at the origin. (a) For Z > 0, (b) for Z < o.
Friedel oscillations may be derived by taking the asymptotic limit of the change in charge
density,
(4.67)
lim [per) - Po] =

r-->oo
~ L:(21 + 1) Jk
(2Tt) r2 1 0
F
dk[sin2(kr + 01 - ITt/2) - sinZCkr -/Tt/2)] (4.68)
and
[sin2 (kr + 01 - ITt/2) - sin2 (kr - ITt/2)]
= Hcos(2kr - ITt) - cos(2kr + 201 -ITt)] (4.69)
The wave vector integral is difficult because the phase shifts ol(k) depend on k. An
approximate answer is derived by writing this dependence as 8z(k) = 01(kF) +
(k - kF)(do/dk). The k integral is then elementary and the r-dependent part is
(-Ii J:F dk{ cos(2kr) - cos[2k(r + ~!) + 201 - 2kF ~!J}

= - (~!i {sin(2kFr) - sin[2kFr + 201(kF)]} (4.70)
and the change in density is
lim [per) - Po] = 4

r-->oo
~
Tt r
3 L:(21 + 1) sin[ol(kF)] x cos[2kFr + 01(kF)] + o(_~)
1 r'
(4.71)
At large distances, the changes in charge density oscillate with a period of 2kF and decrease in
amplitude as r- 3 • This asymptotic equation is independent of the nature of the impurity. The
impurity determines only the values for the phase shift 01(kF).
The Friedel sum rule is believed to be exact in real systems. Langer and Ambegaokar
(1961) have shown it to be valid even in an interacting many-particle system. If one knows the
exact impurity potential and the exact screening charge profile, the Friedel sum rule will be
valid. In practice, a calculation of an impurity potential is done by pseudopotential or other
means, and it is screened by a good dielectric function. The phase shifts are evaluated
numerically and it is found that the Friedel sum rule errs by a small percentage. This error is
no fault of the rule but the choice of potential or screening function. In fact, the theoretical
system is not charge neutral. Usually the potential is adjusted slightly, for example, by altering
a screening length, to force the Friedel sum rule to be satisfied. Then one has a consistent
model of the neutral system, and the phase shifts are probably reasonably accurate.
Quite often the phase shifts do not depend on ml or ms , but they strongly depend on I.
Then one can represent the answer for electrons as
(4.72)
which is the way that it is usually presented. The factor of 2 is spin degeneracy and (21 + 1) is
orbital degeneracy. The quantity (2/Tt)(21 + 1)01(kF) is interpreted as the amount of screening
charge in the angular momentum channell.
In real solids, the impurity may occupy the site ordinarily occu&ied by a host atom. Then
the normal host atom would have its own set of phase shifts 01 ) which characterize the
nonnal metal. For example, if Mg2+ is a substitutional impurity in sodium metal, it could be
at a site where a Na+ is usually place.d. The Na+ ion has its own phase shifts 'O~h), and the
Mg2+ ion has its own phase shifts 'O~'). If one repeats the arguments leading to (4.72), the
Friedel sum rule is given by the difference of the two sets of phase shifts:
(4.73)
Note that the host phase shifts do not obey the Friedel sum rule.
Another exact result may be obtained in tenns of phase shifts. It is the total energy of the
impurity as caused by its interactions with the surrounding electrons in a metal. The theorem
is due to Fumi (1955) and relates the total energy Ei to an energy integral over the phase
shifts:
(4.74)
(4.75)
The energy integral starts at the bottom of the conduction band. It is assumed there are no
bound states. The presence of bound states will change the answer. One can also use the
second fonn of (4.75), which includes just the sum over wave vector states for the conduction
electrons.
Fumi's theorem is proved in the following way. The Hamiltonian is the kinetic energy
plus the potential energy. Well outside the impurity, or its screening charge, the potential is
zero, and the Hamiltonian only has the kinetic energy tenn. Here the energy must just be the
summation of the kinetic energies of all the particles, which is a discrete summation over the
states allowed in the sphere of radius R:
11212 112 2
E = L -n = -2 L[(n +!l)n - '01] (4.76)
lmn 2m 2mR Imn
The desired quantity is actually the change in kinetic energy, so the result without the
impurity is subtracted:
Ei = 2
m
"~2 Im[m,n
L {[(n +! l)n - 'Olf - [(n +! l)n]2} (4.77)
112
~ -R L kn'Ol(kn) (4.78)
m Im[nm,
Next let the radius of the sphere go to infinity, which changes the wave vector summation to a
continuous integration. This change and the result are
(4.79)
(4.80)
which proves the theorem for the conduction band states. The result can be extended to
include bound states. They rarely occur in three dimensions, but are common in lower
dimensions.
4.1.4. Impurity Scattering

How are Green's functions used to describe impurity scattering? They do not work very
well for describing the scattering of a particle from a single impurity. In Dyson's equation, the
change in energy of the particle is calculated from the interactions. One impurity in a large
system of volume v changes the energy ofa particle only by order l/v. This result assumes, of
course, that the particle does not get bound by the impurity. A free-particle state has a density
l<l>kl 2 = l/v, so that a local impurity changes the energy only by l/v. To alter the energy, one
must add Ni impurities, where NJv = ni as v ~ 00. Then the particle energy gets changed as
a function of the concentration ni of impurities. The isolated impurity case should be studied
by the following procedure: one solves the self-energy as a function of ni and then takes the
limit ni ~ O. Terms of order O(ni) are from an electron interaction with single impurities.
The potential which scatters the electron is taken as a summation of impurity potentials:
V(r) = L V(r - Rj ) (4.81)

j
The Fourier transform is V(q):
V(q) = Pi(q)V(q)P(-q) (4.82)
where p( -q) and Pi(q) are the particle densities of electrons and impurities, respectively:
p(q) = L ct+q,crCkcr (4.83)

kcr
Pi(q) = Ldq ' Rj (4.84)
j
The expansion of the S matrix encounters products of the impurity density operator:
(4.85)
These products must be averaged. This average is not taken over temperature. Instead, it is
averaged over the possible positions which the impurities may have in the solid. Usually it is
assumed that the impurities are randomly located and that there is no correlation between their
positions. This method of impurity averaging was suggested by Kohn and Luttinger (1957).
Their result will be derived by first examining the results for a small number of operators. The
first term is n = 1:
(4.86)
If the Rj are located randomly, this sum is zero unless q = 0, and then the sum gives the
number of impurities N i • The second case is n = 2:
(~ IJ
eiql • R j +iq2 • RJ) = (~ei(ql +q2)' R+ ~ eiql •R +iq2 • R
1=)
j
1#)
j j )
(4.87)
The first term is zero unless q\ + q2 = 0, since it is the same average as infi. The second
term is zero unless both q\ and q2 equal zero:
h(q\, q2) = Nio q ,+IIz=o + Nj(Ni - 1)oq,=oOq2=o (4.88)
The general result found by Kohn and Luttinger is
!,,(q\, q2' ... , qn) = Ni0r..q=o + N;(Ni - 1) L

I t m
or.J=I!1
m q.=o0r. n q=O
j=m+l !J
(4.89)
In terms with products of several delta functions or. -0' one must take all possible combi-
nations of distributing the qj among the delta functions. The final result is obtained by
generalizing the low-order examples above. The result for n = 3 is
.I3(q\, qz, q3) = Nioq,+IIz+q3 + Ni(Ni - 1)(oq,oq2+q3 + Oq2 0q,+q3 + 0IbOq,+q)

+ Ni(N; - 1)(N; - 2)oq, 0q2 0q3 (4.90)
The next step is to learn to draw Feynman diagrams with these impurity averages. Since
N; > 10 10 for real systems, this expression is approximated by
!,,(q\, ... , qn) = N;or. q +Nl0r.q or. q +N?0r.q or. q or. q +... (4.91)
There occurs the combinations of the function

No",
• uq
(4.92)
This factor is interpreted as the particle scattering from a single impurity. The factor of or. q
states that momentum is conserved for the particle while scattering from the impurity. In Fig.
4.2(a), the impurity is represented by an x, and the solid line is the particle. The dashed lines
represent impurity interactions V(qj). Momentum conservation requires the momenta from a
single impurity to sum to zero. Diagrams with two factors of (4.92) involve the scattering
from two impurities, as shown in Figs 4.2(b) and 4.2(c).
In Fig. 4.2(b), the scattering from the two impurities interfere-the momentum lines
cross each other. Such interference occurs when the impurities are nearby, and such diagrams
are important when the concentration of impurities is large. In Fig. 4.2(c), the two scattering
events are disconnected, since the interaction lines do not overlap. Each connected diagram
gives a separate self-energy contribution. Edwards (1958) has shown that one has the Dyson
equation
1
~(p, iPn) = . ~ :E( . ) (4.93)
IPn - 'P - p, IPn
where the self-energy :E(p, iPn) from impurity scattering contains all the connected contri-
, ,
butions. The diagram in Fig. 4.2(b) is connected, so it contributes a term to :E. The self-energy
~\ ~ ~
t
I "I \ \
M+CI.-O 'I'\
, I I I ., "'"\ ,' ,
~'4\ - _'OK]
,I- '"/\ ,, " "
/:lID\
)1)1)')
I
«) , ')' I) ,),
, I \\
\ )'
, ,
\)
(01 (bl (e I
FIGURE 4.2
fI
,
I ... I
I
I
~\\
\
,
+ ,
I
I~'
I
I
I
,
,
\,
..... ,
) I ) > I ) , ) I
I
~ I J:'
FIGURE 4.3
diagrams for scattering from a single impurity are shown in Fig. 4.3. They are the terms which
are important when the concentration of impurities is small. The first self-energy diagram in
Fig. 4.3 gives a contribution
(4.94)
The second diagram in the series has a particle line as an intermediate state. This self-energy
term is
(4.95)
which may also be written as (v ~ 00)
~(2)( i ) = n.
p, 'Pn
Jd p' V(p -iPnp')V(p'
'(2n)3
3
- ~P'
- p) (4.96)
There is no summation over internal frequency variables. The impurity is considered rigid and
cannot absorb or transfer energy to the particle which is scattering. Consequently, all the
internal lines have the same energy ipn as the initial particle. This result is not assumed but
instead is derived directly from the Green's function expansion. For example, the above self-
energy term arose from second-order terms and had the t integrals
J: dtejPn~ J: J: dt2~(O)(P,
dtl t - tl)~(O)(p + ql' tl - t2)~(O)(P, t2)
= ~(O)(p, iPni~(O)(P + ql' iPn) (4.97)
The fact that all electron lines have the same energy comes directly from the t integrals. It is
true for an arbitrary order of diagram-all particle Green's functions have the same energy
iPn' As stated before, there is no other excitation in the model to which energy may be
transferred.
The third term in the series in Fig. 4.3 is
~(3)(p . ) _
.&oJ ,IPn - n, 3
J
. d 3pI d 3p2 V(p - PI)V(PI - P2)V(P2 - p)
3 (~ ~
(2n) (2n) ipn - ~p)(iPn - ~p)
At this point, one may deduce the nth term by inspection. It is
~(n)(p, iPn) = nj J~~~ ... ~~; V(p - Pl)V(Pl - P2)'" V(Pn - p)
X IlJ=l ~(O)(Pj' iPn) (4.98)
Now that the nth term is evident, it is also possible to sum the series of terms and get the total
self-energy which is proportional to the concentration n j • The first step in this summation is to
write an integral equation for the self-energy. This integral equation is in terms of a vertex
function r(p', p) which has the equation
r(p', p) = V(p' - p) J
+ d3p~ V(p' ~ Pl)r(Pl' p) (4.99)
(2n:) lPn - ~pl
L(p, iPn) =f
n=l
L(n)(p, iPn) J(2n:)
= niV(O) + d3p~ V(p ~ Pl)r(Pl, p)
lPn - ~pl
= nir(p, p) (4.100)
Repeated iteration of Eq. (4.99) will generate the series of terms which give the successive
diagrams in Fig. 4.3, of which (4.98) is the nth term. The series is summed by solving the
integral equation. To this end, define the function
d 3P 1 eiP"rr(Pl'p)
n:(r, p) = J(2n:) 3 -.------:-:-=---'--
lPn - ~pl
(4.101)
This quantity and r(p', p) are both functions of ipn, but this dependence is not explicitly
added to the notation. In terms of this new function, the self-energy (4.100) is
L(p, iPn) = niV(O) + ni Jd 3re- ip ' rV(T)n:(r, p) (4.102)
To solve for the function n:(r, p), consider the effects of the differential operator acting on it:
li2 ]
[ __ V2 -ll- iPn n:(r, p) =
Jd 3p'3 S. P' - ipn e ., .rr(p', p)
'P
2m (2n:) lPn - Sp'
li2 2
[ - 2m . ]
V -ll- lPn n:(r, p) = -
J(2n:)3
d 3P'
e
i' . r
P
,
r(p, p) (4.103)
The right-hand side of this equation may be simplified. First, note that the vertex function may
be written as
r(p', p) J
= d3re-ip'.rV(r)[eip'r + n:(r, p)] (4.104)
which follows directly from its definition (4.99). From this relation, it is easy to show that
J--r(p'
d3p'
(2n:)3 '
., .
p)e ,p · r = V(r)[e ,p · r + n:(r, p)] (4.105)
It is deduced that the function obeys an inhomogeneous differential equation:
li2
[ - 2m V2 -ll- iPn + V(r) ] n:(r, p) = .
_V(r)e ,p · r (4.106)
This equation is solved in the following fashion. First find the solution to the following
Schr6dinger equation:
(4.107)
This differential equation was discussed extensively at the beginning of this section. Take the
outgoing wave boundary conditions. The reason for this choice will be clear below. In terms
of these wave functions, the special function nCr, p) has the following solution to the inho-
mogeneous equation (4.1 06):
(4.108)
where the T matrix is defined in (4.47). This result for nCr, p) can be directly verified by
operating on both sides by [H - I.l - iPn] and then using the completeness property:
(4.109)
The summation A is taken over all the eigenstates of H, which include the bound states as well
as the continuum states. When this result is used in (4.102), the final result is obtained for the
self-energy in terms of the T matrices which were defined in (4.47):
(4.110)
Equation (4.110) is the exact term in the self-energy which is proportional to the concen-
tration nj. It has been obtained by summing the set of diagrams shown in Fig. 4.3, where the
particle has multiple scattering events from the same impurity. There are other terms in the
self-energy which are proportional to higher powers of the concentration (nj)m. For example,
one term which contributes to order nr
is shown in Fig. 4.2(b). However, in the limit of dilute
impurities, as nj -+ 0, the result (4.110) is the most important.
Our interest is usually in the retarded self-energy, which is called L(p, ro), omitting the
"ret" subscript:
(4.111)
There is one particular case which is very important: when the energy ro is set equal to ~p.
Then the self-energy (4.111) is just nj times the diagonal T matrix:
2nn·
L(p, ~p) = njTpp = --12:,(21 + l)e ·0 (P) l I sin[D[(P)] (4.112)
mp [
This identification is made by noting that the T-matrix equation (4.49) is identical to the vertex
equation (4.99) when ipn -+ ~p + iD. This result is frequently used in representing the self-
energy of the particle. The imaginary part of the self-energy has a simple formula:
-2 Im[L(p, ro)] = 2nnj 2:, I TAP 12 D(ro - ~A) (4.113)

A
The right-hand side may be further reduced in the special case ro = ~p. If the two prior results
(4.55) and (4.56) are combined, the result may be shown to be proportional to the scattering
cross section:
(4.114)
The imaginary part of the self-energy for 0) = ~p is the cross section times vpnj. The factor
vpnj is just the rate at which the particle encounters the impurities. The lifetime and mean-
free-path of the particle may be identified as
(4.115)
(4.116)
These results provide an example of the earlier assertion that the imaginary part of the
retarded self-energy is related to the damping of the particle.
An interesting question is the relationship between these self-energy expressions and
Fumi's theorem (4.75). The two results are quite compatible. The self-energy expressions
describe the effect of the impurities on single-particle states. Fumi's theorem gives the energy
change of the system. Fumi's theorem can be derived by averaging the single-particle
properties. This derivation is shown in the next section.
4.1.5. Ground State Energy

The ground state energy is derived starting from the self-energy expression (4.111).
When averaged correctly, this expression gives Fumi's theorem (4.75) for the energy per
impurity. What is proved, say at zero temperature, is that the impurities cause a change in the
ground state energy of
.Ml = -Nj - 2 JkF

pdp E(21 + 1)01(P) + O(Nl) (4.117)
nm 0 1
Terms proportional to O(Nl) give energy terms arising from interactions between impurities.
The term proportional to a single power Nj is the average energy per impurity from inter-
actions with the electrons.
The result (4.117) has an interesting history. Most of our references have been to solid
state physics work such as that of Friedel and Fumi. A parallel development was occurring in
nuclear theory. Brueckner et al. (1954) proposed that the ground state energy of the system
was an average over the energy of the single-particle states. The energy per particle was taken
to be the reaction matrix, so they took the energy change to be
This equation has tan[od instead of 01 and so is incorrect. Fukuda and Newton (1956) showed
that the correct result should be 01 rather than tan[OI]. In fact, they give results identical to
those of Fumi. Fukuda and Newton, and also DeWitt (1956), proved some important theo-
rems regarding the energy (4.117). The result is not dependent on the spherical box which
was used in the derivation; the same result was demonstrated for a cube and other shapes.
Similarly, it does not depend on the assumption that the wave function vanished at the surface
of the box. The same result is obtained for other boundary conditions. Fumi's theorem appears
to be a general result.
The ground state energy is calculated using the formula (3.300) derived earlier, which
contains a coupling constant integration
vL
,Ml = A
J-' ip,a
J(21t)
dp J'd'l'].
3
-3
0
-
'I']
.
~(p, IP, 'I'])~(P, IP, '1']) (4.118)
Since only the term proportional to ni is needed, and since the self-energy ~ is already
proportional to ni' then replace ~ by ~(O). The difference is just self-energy terms which give
higher powers in (nit, which are going to be neglected. The coupling constant integration is
important for weighting each factor V of impurity potential (and not each factor of the T
matrix). From (4.98), this self-energy is
(4.119)
where the coupling constant integration has been done. There are several ways to evaluate this
expression.
The first derivation will give tan(o/), instead of 0[, so that it gives the wrong answer. This
derivation is included, not just for fun, but to show how simple it can be to get the wrong
result. While reading this derivation, try to find the incorrect step, which leads to the wrong
result. The incorrect derivation proceeds by doing the summation on Matsubara frequencies
in (4.119) term by term:
(4.120)
(4.121)
The nth term is found from the contour integral,
(4.122)
where the contour is taken at infinity so that it includes all the poles. The contour integral
leads to the identity
(4.123)
The expression for the ground state energy becomes
i\Q - 2N.{] d 3p n (~ )v.

- I (21t)3 F 'P PP +2
!] (21t)3
d3 p ] d 3p'
(21t)3
x V~p,[nF(~p)<;§(O)(p', ~p) + nF(~p')<;§(O)(p,~p')] + ... } (4.124)
There are two terms proportional to V2 . They are identical, after changing around the dummy
!
variables of integration, so this cancels the factor. Similarly, there are three terms with three
t
powers of V which are all equal, which cancels the factor. The lin factor is canceled in the
nth term:
(4.125)
The series in brackets is just the reaction matrix equation (4.34), so that
(4.126)
The derivation gives tan 01, which is the wrong answer. Did you find the wrong step? The
reaction matrix equation (4.126) contains a P for principal parts. It is an instruction to omit
the term where the energy denominator vanishes. The T-matrix equation does not have this
instruction, because one is supposed to include such terms. The derivation mishandled these
terms. The "theorem" (4.123) is correct only as long as the ~j are different; it is incorrect if
any are alike. The correct result, valid for all cases, is obtained by evaluating the contour
integral (4.122) by drawing a branch cut along the real axis. The summation over Matsubara
frequencies is the difference between the integration along the top and bottom of the branch
cut (see Sec. 3.5). When taking the imaginary part of the right-hand side, it is possible to get
products of odd numbers of delta functions. For n = 3 the result is
1 1
~ ~ (ip - ~l)(ZP - ~2)(iP - ~3)
= nF(~l) + nF(~2) + nF(~3)
(~l - ~2)(~1 - ~3) (~2 - ~1)(~2 - ~3) (~3 - ~2)(~3 - ~l)
(4.127)
The last term contains the product of delta functions. This term was omitted in the first
derivation. Similar correction terms, with odd powers of delta functions, occur in the fourth
and all higher terms in the perturbation series (4.124). An examination of these terms shows
that they are generating a series of reaction matrices:
Sec. 4.2 • Localized State in the Continuum 207
The wave vector integrals may be done term by term. The d 3Pn integral eliminates the delta
function o(~p - ~p) and makes Pn = p. The angular integrations force all angular momentum
I components to be the same. These steps derive the series
(4.129)
The factor 41t2pR/(p,p) = -tan[o/(p)]. The series is just that for the tan-lex) =
x - xl3 + xl5 ... where x = - tan(o). The ground state energy per particle as given by
Fumi's theorem is:
2NJk pdp '£(21 + l)o/(p)

.10 = _ _I
F
(4.130)
1tm 0 /
This derivation contains two messages. The first is that Fumi's theorem with o/(p) is correct
and may be obtained by using the conventional formulas for the ground state energy of the
system. The second message is that the energy of the system is not just a simple average of
the single-particle energies, such as ~pnF(~p)Rpp or ~pnF(~p)Tpp. Both of these guesses are
incorrect.
A more elegant derivation of Fumi's theorem is possible by following the techniques of
Langer and Ambegaokar (1961). They avoided evaluating the series (4.119) on a term-by-
term basis. Instead, they used operator manipulations to attain the above result.
4.2. LOCALIZED STATE IN THE CONTINUUM
In this section the following Hamiltonian is solved exactly:
H = Ecbt b + ,£[EkCtck + Ak(ctb + bt Ck)] (4.131)

k
It describes a localized state of fixed energy Ec and operators b and bt. The localized state will
be called the impurity, and it is assumed only one exists. There is a continuous set of states of
energy Ek with operators ck and ct. This set of states could have a finite bandwidth, as often
occurs in tight-binding models in solids, or else it could be a free-particle model. The last
term in the Hamiltonian includes the mixing between these two kinds of states. It contains
processes whereby the continuum particle hops onto the impurity (b t Ck) and where the
particle hops off the impurity into the continuum (ctb). If the particles have spin, it is
assumed that the hopping on and off the impurity preserves the particle spin state. The spin
never changes and is unimportant, so its dependence is suppressed in all SUbscripts and labels.
If the hopping particle could change its spin orientation, the problem would become harder.
Since the Hamiltonian is quadratic in operators, its solution is equivalent to diag-
onalizing a matrix. The solution may be obtained in this fashion, but here it is done analy-
tically. Since the Hamiltonian is quadratic, the statistics are irrelevant since the same
eigenvalues are obtained for fermions or bosons. It is also irrelevant whether there are one or
many particles in the system. For a Hamiltonian which contains only quadratic operators of
fermions or bosons, one just diagonalizes the Hamiltonian to find the eigenstates, and then all
eigenstates behave independently.
This model Hamiltonian was introduced simultaneously by Anderson (1961) and Fano
(1961). Anderson applied it to solid-state physics, while Fano used it in atomic spectra. It
tends to be called the Anderson model or the Fano model depending on whether the speaker is
a solid state or atomic physicist. Here it is called the Fano-Anderson model. As explained in
Chapter 1, it is related to the famous Anderson model, which is (4.131) plus another term.
The Hamiltonian will first be solved without using Green's functions. Afterwards, the
Green's function solution will be given. The nature of the solution depends critically on
whether the energy ec is within the band of states ek' If the continuous band of states ek is
confined to the range
(4.132)
then the solution depends on whether ec is also within this range. Actually, this statement is
incorrect. Because of the interactions with the continuous band of states, the energy of the
localized state is altered to a new value Ec. Of course, since the Hamiltonian is not yet solved,
the renormalized energy Ec is not yet known. Jumping ahead, the final result is that the new
energy is
(4.133)
If Ec is within the range
(4.134)
then the solution has an important property: there are no localized states in the system. A
continuum particle may hop onto the impurity, but after a while it may hop off again. Particles
spend only part of their time on the impurity, so it is not a well-defined eigenstate. The
impurity state has become a scattering resonance.
Of course, if the new energy Ec is outside of the band of continuum states, then a true
localized state will exist. Then the solution has a distinct form-a real bound state.
First solve for the case where Ec is within the continuous band and no bound state exists.
All states are continuum states. A new set of operators (Xk and (Xt are defined which refer to
the eigenstates of (4.131). The old operators b and Ck can be expanded in terms of the new
set:
b= LVk(Xk
k
Ck = L 'Ilk,k'(Xk' (4.135)
k'
One impurity in the presence of N ~ 1023 particles changes their energy by a negligible
amount-remember the assumption of no bound states-so that these new operators still have
the energy ek' All that is needed is the vector Vk and the matrix 'Ilk,k' for a complete solution.
The model is solved first in one dimension. The extension to two and three dimensions will be
easy to describe at the end. It is also assumed that the dispersion ek does not permit two states
k and k' to have the same energy. Of course, this constraint is never realized in practice, since
states of k and - k usually have the same energy. The problem is that the hopping on and off
the impurity preserves energy but not wave vector information. The hopping will mix the
states which exist at the same energy. It is a nuisance to describe, so it will be omitted for the
moment, and k becomes a scalar. One way around the problem in one dimension is to
describe symmetric and anti-symmetric combinations of the states with ±k:
1
cks = J2[Ck + C_k] (4.136)
1
Cka = J2 [Ck - C_k] (4.137)
These two states have the same energy Ek' Only the symmetric state is assumed to interact
with the localized state, so that the interaction has the form LkAk[bt Cks + ctb]. This
interpretation of the Hamiltonian makes physical sense. The SUbscripts s will no longer be
written.
Since the new operators rY.k and rY.! describe eigenstates, the Hamiltonian may be written
as
H =L EkrY.!rY.k (4.138)
k
The commutator [b, H] is evaluated using both the new operators and the old:
[b,H] = bEe + LAkCk = LEkVkrY.k (4.139)
k k
Then the old operators are expressed as (4.135), and this equation is
EeLVkrY.k+ LA{Ilk,k'rY.k' = LEkVkrY.k (4.140)
k k/c' k
Each coefficient of rY.k is independent; i.e., take the commutator of the above equation with rY.!
and get
(4.141)
The same procedure is used with the commutator [ck' H], which gives
[Ck, H] = Ekck + Akb = L llk,k'Ek'rY.k' (4.142)
k'
or expressing b and Ck as rY.b gives another equation:

llk,k'(Ek - Ek') = -AkVk' (4.143)
The last equation gives a result for llk,k'> when Ek =1= Ek" which means that k =1= k', because of
our assumptions. It provides no information about llk,k' for the case where k = k'. It is
necessary to introduce another unknown function Zk which is proportional to the value of
llk,k' when k = k':
(4.144)
The other factors Ak Vk are added to the last term for convenience. The energy denominator in
the first term is taken as a principal part; the term is omitted when Ek = Ek" All energy
denominators are assumed to be principal parts, and the conventional symbol P will be
omitted. If the expression for llk,k' is used in (4.141), the equation for Vk becomes
- - + Ok/c'ZkVkAk
Ak'vk )
Vk(Ek - Ee) = LAk' ( - - (4.145)
k' Ek' - Ek
The Vk dependence factors out completely. This factoring leaves an equation in which the only
unknown quantity is Zk>
(4.146)
so that Zk quantity is now determined. It also simplifies the notation to introduce the self-
energy function
(4.147)
so that Zk is written as
(4.148)
All quantities on the right-hand side of (4.148) are known, so that Zk is now known. To make
further progress toward obtaining Tlk k' and Vh more equations are needed. They are obtained
from the commutation relations for the old operators, which must still be obeyed when they
are expressed in terms of the ak operators::
[b,b t ] = 1 = Lvi (4.149)

k
[Ck' 4] = Okk' = Lk" Tlk,k',Tlk',k" (4.150)
(4.151)
The last equation is used first. If the result (4.144) for Tlk,1e' is used, then (4.151) becomes
(4.152)
or
(4.153)
Since Ak is not zero, the quantity in parentheses must vanish. This result will be used below.
Next, take Eq. (4.150), insert the result for Tlk,Ie" and find
which equals
(4.154)
The last term must be rearranged. Since the energy denominators are principal parts, use
Poincare's theorem:
P lip
E"" - E" E"" - Ek' = E" -
1 (1 E" - E"" -1)
Ek' Ek" - Ek'
+ 1t 2 0(Ek'1 - E")O(E",, - Ek') (4.155)
The delta functions of continuous energy must be changed back to Kronecker deltas, since we
are using box normalization in a box of length L. This alteration is
(4.156)
where v" is the velocity of the particle:

(4.157)
The terms can be regrouped in (4.154):
okk' = okk'viAi [Z; + (2~J 2]

(4.158)
The last two quantities in parentheses vanish, since each is identical to (4.153). All the
remaining terms are proportional to 0"Ic'> and so exist only when k = k'. These steps give the
equation for v,,:
(4.159)
Of course, once v" is known, then ll",k' is obtained easily from (4.144). The result for v" may
be reworked into a form in which the physics is more transparent. The earlier result (4.148)
for Z" is used to rewrite (4.159) as
A2
vt- " (4.160)
" - [E" - Ec - ~(E,,)f + (LAi!2v"i
Furthermore, the self energy ~(E,,) will be interpreted as the real part of the retarded self-
energy:
(4.161)
(4.162)
Similarly, the imaginary part of the retarded self-energy is
Im[~ret(E)] = -1t LA~o(E - Ek') (4.163)

k'
By using the previous identity (4.156) for the delta function, the imaginary part of the
retarded self-energy becomes
(4.164)
The expression on the right is recognized as the same factor which occurs in the denominator
of (4.160) for v~. In fact, this quantity may be rewritten as
2 (2Vk) Im[L(ek)]
(4.165)
Vk = - L [ek - ec - Re L(ek)f + [1m L(ek)f
= -C~k)Im{ ek - e ~ Lret(ek)}
c
(4.166)
so that v~ is proportional to the imaginary part of a retarded Green's function. It was shown in
Sec. 3.3 that the imaginary part of a retarded Green's function is proportional to the spectral
function:
(4.167)
(4.168)
The remaining equation (4.149) which must be satisfied may now be considered:
1 = LV~ (4.169)
k
Since v~ is now known, it is readily substituted, and the integral is:
(4.170)
In the second equality the integration variable was altered to the energy of the particle. This
integral is just the sum rule for spectral functions which was proved in (3.121). The sum rule
applies to any spectral function and must apply to the one derived above. Equation (4.149) is
satisfied. The transformation (4.135) to the IXk has been achieved. The form derived for vk
satisfies all the commutation relations (4.149)--(4.151) as well as the commutation relations
with the original Hamiltonian. The problem has been solved exactly.
Since the quantity
(4.171)
is identified as a retarded Green's function, for which particle is it the Green's function? It is
the Green's function of the localized particle, or what became of the localized particle after it
became delocalized. This assertion is somewhat evident by considering the definition of the
localized retarded Green's function for fermions:
G(t) = -i0(t)(b(t)bt (O) + bt(O)b(t))

= -i0(t) L v~(ak(t)ak(O) + ak(O)ak(t))
k
= -i0(t) L v~e-iEkl = -i 0(t) L vkA(ck)e-iEkl

k L k
= -i0(t) foo dCk A(ck)e- iEkl (4.172)

-00 2n
In the last step the summation is changed to an integral over ck' The last line is the identity for
the retarded Green's function assigned in Problem 6 of Chapter 3.
The Fano-Anderson model is just a description of a localized scattering resonance. The
continuum particles come to the impurity, spend some time in the resonant state, and then
depart in another continuum state. This model is really no different from that of the prior
section, on impurity scattering, if the impurity potential is made to have a resonance. The
phase shifts for the present model are defined as
tan [u1:(C)] = __Im[Lret(c)]

---=--~-'----C.:=---_ (4.173)
C- Cc - Re Lret(c)
The phase shift is all that is needed to use the results of the prior section to describe the Fano-
Anderson model. The resonant behavior comes from approximating -1m L ~ r as a
constant, or at least as a slowly varying function of energy. Then the spectral function A(c) is
Lorentzian. If the width r is small, this function describes a resonance.
As an example of a scattering potential, consider the potential VCr) shown in Fig. 4.4(a).
The resonance behavior is chosen by first solving for the bound states of Fig. 4.4(b). For the
latter case, bound states exist when
tan(ka) = --k (4.174)

a
a= C~~o J? y/2 _ (4.175)
A bound state at half of the well depth is made by choosing ka = aa = 3n/4. Here k 2 is the
particle energy relative to the bottom of the well. The well shape of Fig. 4.4(a) will have a
resonance at nearly the same energy. The finite extent of the repulsive part will allow the
particle to leave and impart a width to the state and to the scattering resonance. The phase
shift is shown in Fig. 4.4(c) for the potential of part (a). The steep rise in the phase shift, of
about n, occurs at the value ka ~ 3n/4.
The change in phase shift by n across the resonance may be understood from the Friedel
sum rule. As the eigenstates of the system are filled up, one particle will reside in the vicinity
of the impurity and contribute unity to the Friedel sum over () / n.
The initial Hamiltonian was written as (4.131) in terms of eigenstates which were not
orthogonal. All of our effort was merely an exercise in orthogonalization. This work could
have been avoided by writing H in an orthogonal basis. The basis would have the particles
with a scattering resonance from the impurity. The Hamiltonian describes one-particle
behavior.
VIr) (a) VIr) (b)
a 2a r- a
(ell
•G
C
:a
--.f!..
co
-4
kG 2 !1T/4 3
FIGURE 4.4 (a) A potential whose phase shifts can show resonance behavior. The resonance occurs at the bound
state of the potential shown in (b). (c) The phase shift jumps in value by almost 1t radians at the wave vector of the
resonance.
The model is very important in a number of applications. The d-states of transition

metals seem well described as scattering resonances of the sp electrons (see Anderson and
McMillan, 1967). Another example is in surface physics. When atoms come from a gas and
are absorbed on the surface of a metal, the conduction electrons of the metal may hop onto the
atomic-like states of the absorbed atom. The charging effect on the absorbed atoms seems
well described by the model (SchOnhammer and Gunnarsson, 1977). It has also found
numerous applications in atomic physics.
The model will now be solved by using Green's functions. This solution is much quicker
than the prior method, since there is only one self-energy diagram. The Hamiltonian is written
as Ho + V, where V is the last term in (4.131) and Ho is the first two terms. The Green's
function for the localized state is:
(4.176)
The first self-energy term comes from the n = 2 term in the S-matrix expansion:
-"21 J~ dTe J~ dT! J~ dT2 L

iP1 Akl Ak2 (T1 b(T)
o 0 0 klk2
X [Ck l (Tl)bt(T 1) + b(T1)Ct (Tl)][Ck2 (T2)b t (T 2) + b(T2)Ck,cT2)]bt(O») (4.177)

The correlation function is easily evaluated in terms of the unperturbed Green's functions of
the localized state <;§(O)(r) and the continuum states <;§(O)(k, 't) to give
LA~ J~ d'te J~ d't] J~ d't2<;§(O)('t -

ip1 't])<;§(O)(k, 't] - 't2)<;§(O)('t2)
k 0 0 0
= LA~<;§(O)(ip)2<;§(O)(k, ip) = <;§(O)(ip)2'L(ip) (4.178)
k
A2
'L(ip) = LA~<;§(O)(k, ip) = L-._k_ (4.179)
k k Ip - ck
'L(ip) turns out to be the only self-energy diagram. The higher terms in the S matrix only
produce higher powers of this self-energy contribution:
(4.180)
The series may be summed to give the Dyson equation for the Matsubara Green's function:
1
<;§(ip) = ----- (4.181)
ip - ce - 'L(ip)
Changing ip ----+ C - I-l + io gives the retarded Green's function, which is the same as noted
earlier in (4.171). This derivation proves that v~ is really proportional to the spectral function
ofthe G(c). The equivalence of this result to (4.172) is just an example of the general theorem
proved in Problem 6 of Chapter 3:
(4.182)
These results may be generalized to higher dimension and other energy bands. This
extension is most easily accomplished using the Green's function technique, since the steps
are the same as in one dimension. The Matsubara form of the Green's function of the b
operators (4.172) is still given by Eq. (4.181) where the self-energy operator is summed over
all states in the system:
A2
'L(ip) = L-._k_ (4.183)
k Ip - ck
The real part of the denominator of the retarded Green's function is (ip ----+ c)
c - ce - Re['Lret(c)] (4.184)
The Green's function has poles at the points where the denominator vanishes. The poles of the
Green's function corresponds to excitations of the system. The poles are at the energy Ee
which satisfies
(4.185)
The above expression for Ee is the result asserted earlier in (4.133). If the pole lies within the
continuum of states, then the resonance occurs, and there is no bound state. But if the pole
occurs outside of the band of continuum states, then the system has a real bound state. A pole
occurs when 1m 'L = 0 at the same point that the real part of the denominator vanishes.
Generally, 1m 'L is not zero throughout the continuum band, so that 1m 'L = 0 only outside of
the band.
If it is assumed that the bound state occurs outside of the band, the spectral function has
the form
-21m L(E)
A(E) = 2no(E - Ec - Re[L(E)]) + 2 2 (4.186)
[E - Ec - Re L] + [1m L]
The first term comes from the pole of the Green's function, while the second term is from
those parts where 1m L -=1= O. The first term may be rewritten as a simple delta function plus a
renormalization factor. This procedure was described earlier in (3.155):
-2lmL
A(E) = 2nZo(E - Ec) + 2 2 (4.187)
[E - Ec - Re L] + [1m L]
Z= ( 1 - - -
dRe L)-l (4.188)
dE -
E,
In other examples there may be more than one bound state, in which case the first term
becomes a series of delta functions.
These points are well illustrated by an example. Take a one-dimensional tight-binding
model on a solid of unit separation a = 1 and length L with a constant coupling constant:
Ek = -2w cos(k) (4.189)

A2 _ C
k-T, (4.190)
Also, take the initial bound state energy Ec in the middle of the band Ec = O. The band
structure is shown in Fig. 4.5. The Brillouin zone is defined by -n ::s k ::s n. The self-energy
function Lret(E) is now elementary to evaluate:
L (E) =~L 1 = .£ f1t dk

ret L k E + w cos(k) + io 2n -1t E + w cos(k) + iO
The real part is
Re[L] = .£f1t dk = C sgn(E) if E2 > ~

2n -1t E +w cos(k) JE2 - w 2 '
=0 (4.191)
FIGURE 4.5 Tight-binding energy bands in one dimension.

where sgn(e) is the sign of e. The imaginary part is
-Im[L(e)] = ~~ [11 OlE + w cos(k)] (4.192)
= __C__ _~C==7 if ~ > e2 (4.193)

wi sin(k) I ,Jw2 - e2
The spectral function of the localized state is
(4.194)
The step function Sex) is unity if the argument is positive and zero if negative. The first term
is nonzero only outside of the band of continuum states, while the last term is nonzero only
inside the band. Figure 4.6 shows a plot of the entire spectral function. There are two sharp
bound states, one below the band and one above. The continuous contribution throughout the
band is due to the last term in (4.194). According to the sum rule, the total area under all the
contributions must be 27t. The solution to the equation
(4.195)
is
E = ±_l [~+ Jw4 + 4C2]1/2 (4.196)

c v'2
and the renormalization factor is
(4.197)
and is the same at both poles. For strong coupling, where C becomes very large, Ec » w. The
!,
renormalization factors approach so that one-half the spectral weight is in each pole.
A(E)
-0.5 o 0.5 1.0

E
FIGURE 4.6 Spectral function from Eqn. (4.196). C = I, ro = 1.0.

4.3. INDEPENDENT BOSON MODELS
The independent boson model is very important in many-body physics. It is an exactly

solvable model which describes some relaxation phenomena. It has become very useful for
describing a wide variety of effects in solid state physics. This section discusses it at great
length. Two different derivations of the basic mathematical result are provided. In addition,
several variations on the model are described briefly. An exact solution may be obtained by a
variety of techniques. The usual procedure is followed: first solve it by ordinary operator
algebra, and later solve it using Green's functions.
The first Hamiltonian which will be solved is
H = CtC[EC + ~(aq + a:)] + ~COqa:aq (4.198)
The Hamiltonian describes a fixed particle of energy Ec interacting with a set ofphonons with
frequencies coq • The interaction occurs only when the state is occupied and etc = 1. The
phonons are the independent bosons. An alert reader will note that this model was solved
exactly in Chapter 1. By making a canonical transformation, the Hamiltonian may be
rewritten as [see (1.239)]
(4.199)
where the self-energy is

M2
Ll=I:---.!!. (4.200)
q co q
The solution to this problem is identical with the problem of a charge on a harmonic spring in
a uniform electric field. The electric field causes a displacement of the charge to a new
equilibrium position, about which it vibrates with the same frequency as before; see (1.38).
The present objective is to obtain a better description of the fluctuations about equili-
brium. The self-energy Ll is for the zero temperature ground state configuration of the system.
To study relaxation effects, the fluctuations need to be understood. They are obtained from
evaluating the time variation of the Green's function.
(4.201)
It is solved for the real-time Green's function, which is permissible in the present case because
the single impurity state etc will not alter the phonon energies. In the thermodynamic
averaging over the phonon states, the perturbation V has no effect, so there is no need to
worry about the perturbation expansion for the factor exp( -~H). First solve the Green's
function of time at nonzero temperature.
4.3.1. Solution by Canonical Transformation

The Hamiltonian is first solved by a canonical transformation. This solution is nearly
identical to the procedure described in Chapter 1, but here it is done with more rigor. A new
Hamiltonian is desired by a transformation of the type
if = £I He- s = c t C(Ec - Ll) + I: coqa:a q (4.202)
q
Sec. 4.3 • Independent Boson Models 219
The transformation must be done so that st = -So The transformation on any product of
operators is done by taking the product of the transformed operators. The last theorem is
shown by inserting 1 = e- s~ between each operator:
It is assumed that any function of operators may be expressed as a power series, and the
transformation on a function of operators is just the function of the transformed operators:
00 00
~1(A)e-s =~ L anAne- s =L an(At = I(A) (4.203)

n=O n=O
The transformation is considered on each operator separately, and the transformed Hamil-
tonian is the old one with the new operators. These are evaluated using
- s S 1
A = e-Ae- = A + [S,A] +,[S, [S,AJ] + ... (4.204)
2.
(4.205)
which gives
c=cX
ct = ctxt
a- q = aq - -
Mq t
cc (4.206)
co q
M
a~ = a~ - --..!!.ctc
co q
where the operator is
x = exp [- LMq
co (atq - a q )] (4.207)
q q
Since X commutes with the c operator, the number operator is the same in the new repre-
sentation,
(4.208)
since xt = X-I. The transformed Hamiltonian is
(4.209)
(4.210)
This transformed Hamiltonian is precisely the form which was the objective in (4.202). It is
also the form which was derived earlier in Chapter 1. The reason for repeating the analysis is
that the important factors X in (4.207) are needed for the discussion of fluctuations.
This transformation is now applied to the Green's function (4.201). The factor 1 = e-seS
is inserted into the trace, say for t > 0:
(4.211)
Using the cyclic properties of the trace,
(4.212)
By using the previous theorems, everything in the trace is now changed to the transformed
representation. The Green's function may be written as (t > 0)
(4.213)
It should be emphasized that this G(t) will be exactly equal to the earlier definition (4.201).
The new equation for G(t) is just another way of evaluating the same thing.
At first glance it appears that this evaluation is now trivial, since the Hamiltonian is
diagonal in the c t c-operators. But this conclusion is untrue and misleading, since c and c t do
not commute with a or at because of the X factor in (4.206). It is necessary to stick with the e
and a representation and to put in X explicitly. The Green's function becomes
(4.214)
However, it is now possible to achieve a great simplification. Since iI is diagonal in e and a, it

is easy to commute it through eX and to obtain the time development of the operators:
eiiIteXe-iiIt = e- it(E,-8)eX(t) (4.215)
X(t) = exp [- LMq (a~eiWqt _ aqe-iWqt)] (4.216)

q co q
The phonon and electron parts of the trace may now be completely separated. For (t > 0):
(4.217)
Ec = Ec - Ll (4.218)
Hp = L coqa~aq (4.219)
q
The result for G(t) has the great simplification mentioned above. The particle part is trivial
(assuming they are fermions),
(4.220)
and there remains only the problem of evaluating the phonon part of the trace. This evaluation
is nontrivial, although it may be done exactly. The operator disentangling method was
introduced by Feynman (1951).
4.3.2. Feynman Disentangling of Operators

The objective is now to evaluate the trace over the phonon distributions of the operator:
(4.221)
Each wave vector state q is averaged independently, and the final result is the product over q
states:
(4.222)
For each state q, the trace is merely a summation over all possible integer values of nq
between zero and infinity. The prefactor is the normalization:
(4.223)
The present notation is simplified by first dropping all q subscripts
!F(t) = (1 - e- Iko ) f: e-~nm(nle-A.(ate'.,t-ae-j.,t)eA.(at-a)ln} (4.224)

n=O
The state In} is the one with n excitations and is given in terms of the operators as
In} = (att IO) (4.225)

0if
The first step is to separate the operators in the exponential. Use Feynman's theorem on the
disentangling of operators, which is as follows:
Theorem: If the operators A and B have the property that their commutator C = [A, B]
commutes with both A and B, then
e4+B = e4~e-l/2[A.Bl (4.226)
This theorem was proved in Problem 5 of Chapter 2. It is used to separate the exponents in
X(t) and xt(O). To evaluate X(t), set
(4.227)
B = Aae- imt (4.228)
(4.229)
and
X(t) = e4+B = e-'A.2/2 e -Aat ,;rot ~e-;rot (4.230)

xt(O) = e-')..2/2 e Aat e- Aa (4.231)
X(t)Xt (0) = e-')..2 e- Aat ,;"" eAae-;ro, eAat e- Aa (4.232)
The result for xt(O) is just the Hennitian conjugate of X(t) evaluated at t = O. The factor of
X(t)Xt(O) is what is needed in the average in (4.224).
The next step is to get all the destruction operators on the right and the creation operators
on the left. The center two operators need to be exchanged. Since they do not commute, this
exchange will produce another complex phase factor. These two operators are written as
(4.233)
The factor in brackets has exactly the form derived earlier in (4.203) where S = -1..a t .
Consider
t 1..2 t t
- a - 1..[a ,a] + 2! [a [a ,a]] ...
_Aa t Aa t _
e ae (4.234)
=a+1.. (4.235)
then
(4.236)
so (4.224) is finally arranged into the desired form:
ff(t) = (1 _ e- PCO )e-')..2(I-e-;ro,) (4.237)
x f e-Pcon(nleAat(1-,;rot)e-Aa(I-e-;ro')ln) (4.238)
n=O
All the terms with a can be collected together in one exponential since all of these terms
commute. Likewise for all the terms with at. Next prove that
(1 - e- PCO ) f e- pnco (nleu*a t e- ua In) = e-lul2N (4.239)

n=O
1
N=-- (4.240)
e Pm - 1
where, for our case, U = 1..(1 - e- i(01 ). Equation (4.240) is proved by expanding the exponents
in a power series:
(4.241)
Now recall the properties of boson destruction operators acting on a hannonic oscillator state:
aln} == n l/2 ln -- I} (4.242)

a2ln} == [n(n -- 1)]1/2In -- 2} (4.243)
1 n., ] 1/2
a In} == [ (n __ I)! In -- I} (4.244)
The feature that a1ln} == 0 for 1 > n is very useful: It tenninates the power series after n tenns:
-ua
e In} == E-l!-)1 [
n (__ U n.,
(n --I)!
] 1/2
In -- I} (4.245)
Of course, this truncation is why the destruction operators were arranged to the right. The
other operator may be taken to operate to the left and just produces the Hennitian conjugate of
the above result:
] 1/2
(nleu*a t == Ln_u_
( *)m [
n.
,
(n -- ml (4.246)
m=O m! (n -- m)!
These two results must be multiplied together. Using the basic orthogonality of the states
(n -- min -- I} == bn-m=n-I == bm=1 (4.247)
produces the compact result
( I u*at -ual } _ ~ (--luI 2i n!

nee n - to (l!i (n -- l)!
(4.248)
This power series should be familiar to every student of physics: It is just the Laguerre
polynomial of order n (remember hydrogen wave functions):
(4.249)
The final step is to sum the series over n. This last series is just the generating function of
Laguerre polynomials:
L
00 2
(1 -- z) LnCluI2)~ == e1ul zl(z-I) (4.250)
n=O
In our case, to prove the theorem (4.240), identifY
(4.251)
z 1
- - == --N == - - - - (4.252)
z -- 1 e~ro -- 1
When these factors are collected, the result does prove (4.240). The expression for .?F(t) is
.?F(t) == e-(p(t) (4.253)
<p(t) == 1..1[(1 -- e- irot ) +Nll -- eirot l 2] (4.254)
<p(t) == A?[(N + 1)(1 -- e- irot ) + N(1 -- eirot )] (4.255)
Return to (4.222) and reintroduce the product over all q states. The function F(t) contains a
summation of the exponential factor:
F(t) = IIqff q(t) = exp [ - ~ <Pq(t)] == exp[-<l>(t)] (4.256)
<l>(t) = L (Mq) 2
m [Nq (1 - eiO)qt) + (Nq + 1)(1 _ e-iO)qt)] (4.257)
q q
1
N =-;<"""-- (4.258)
q e~O)q-l
The final result for the particle Green's function for t > 0 is
G(t) = _ie- it(oc- li )e-<lJ(t)(1 - nF) (4.259)
This equation is the exact result. Next proceed to a description of the physics.
4.3.3. Einstein Model

The physics is best understood by examining a simple application of the model. All the
phonons are taken to have the same energy mo, which is called the Einstein model. The case of
zero temperature will be discussed first. The modifications for nonzero temperature are
derived afterwards. For zero temperature, all the phonon occupation factors are zero:
(4.260)
Furthermore, the summation over wave vector just produces a coupling constant g:
(4.261)
The Green's function still has the form (4.259), but now the factors are quite simple:
Ll = gmo (4.262)
<l>(t) = g(1 - e-iO)ot) (4.263)
The particle Green's function will be evaluated for the case of a single particle, so set nF = O.
Evaluate
G(t) = -i0(t) exp[ -itEc - gel - imot - e-iO)ot)] (4.264)
The same result is obtained for the retarded function. The spectral function is then the
imaginary part of the retarded Green's function of frequency:
A(m) = -2 Im{ -i f~ dteiO)t exp[-itEc - g(1 - imot - e-iO)ot)]}
(4.265)
The time integral may be evaluated by expanding the ge-i(f)ot part of the exponent in a power
series,
(4.266)
so that the time integral contains terms such as
Jo oo exp[it(m - Ec + Ll - mol)] =
m- Ec
~
+ - mo
I
+I
'S (4.267)
The factor is is added to force the convergence of the oscillating integrand at large values of
time. Then take the limit S --+ 0 and obtain
i i
-----,-----::-__::_ = P + 1tS(m - Ec + Ll- mol)
m- Ec + Ll - mol + is m- Ec + Ll - mol
The spectral function is the real part of this time integral, which has just the delta function:
g/
= 21te-g Eo T!S(m -
00
A(m) Ec + Ll- mol) (4.268)
The spectral function is a series of delta functions, spaced exactly mo apart. The distribution
of peak heights follows a Poisson distribution.
Remember that the spectral function is the probability that the particle has frequency m.
If there were no interactions, the particle would always have energy Ec and there would be a
single delta function at m = Ec' This limit is obtained from (4.268) by setting g = O. For
g =1= 0 the particle has a nonzero probability of occupying other states which have I phonons.
These configurations are not excited states. Since the temperature has been set equal to zero,
the solution must be describing the ground state. In the ground state of the coupled system of
particle and phonon, some probability exists that the system will have the different sets of
frequencies m/ == Ec - Ll + Imo. The different values of m/ obviously correspond to the
particle being coupled to some phonons, which is certainly to be expected in this system. In
the ground state, the particle energy fluctuates among these different values m/.
The spectral function (4.268) is shown in Fig. 4.7 for two different values of the coupling
constant g = 0.5 and 5.5. These values correspond to the weak and strong coupling cases,
respectively. Weak coupling is g < 1. Here the I = 1 peak is smaller than I = 0, and higher I
peaks get smaller very rapidly. For strong coupling, the peak strength increases with I up to
values of approximately I ~ g, and then the peaks decrease again.
It is useful to test the sum rules. The first one is [see (3.121)]
oo dm g/ Joo /
1= J -2 A(m) = e-g I: -II dmS(m - m/) = e-g I: gil = 1
-00 1t / . -00 / .
The factor exp( -g) in (4.268) is now recognized as the normalization factor which maintains
the sum rule. These spectral functions also have the property that
Ec = J
oo
-00
dm
2mA(m)
1t
= (m) (4.269)
This moment is easily proved by direct evaluation,
(m) = J
oo
-00
dill
-2 mA(m)
1t / .
gl
= e-g I: -II (Ec - Ll + mol) = Ec - Ll + gmo = Ec
(a)
-Wo 0
2 3 4
(b)
• ' r ; 1, 1,1.1.1. [, I.r.y, t, o ,
1-0 1 2 3 4 5 6 7 B 9 10 11 12
FIGURE 4.7 The spectral function of the independent boson model, Eqn. (4.270), shown for an Einstein model and
two values of coupling constant. (a) For g == 0.5, (b) for g = 5.5.
which does give (4.269). The last integral is called the first moment. Higher moments (ron)
may also be evaluated, and they will depend on coupling constant. By using the relation
Gret(t) = -i8(t) J OO
-00
dro .
-2 A(ro)e- ,oot
IT
(4.270)
the moments may also be shown to equal derivatives of the retarded Green's function:
(4.271)
Since the retarded Green's function is nonzero only at t > 0, one takes the time derivatives
first and the limit of t -+ 0+ second.
An inspection of Fig. 4.7(b) verifies that the first moment is independent of coupling
strength g. The delta function peaks have an intensity envelope which does appear to have a
maximum near ro ~ Ee' As one increases the coupling constant g, the self-energy L\ = groo
becomes larger in magnitude, so that the lowest energy peak shifts downward in energy. But
its intensity lowers also because of the factor exp( -g), and the envelope of the delta function
peaks becomes a Gaussian. A Poisson distribution becomes a Gaussian for large g, which
may be shown in the following way. Large g requires large values of I, so that Stirling's
approximation may be used for the factorial:
(4.272)
In this approximation the Poisson intensity becomes
e-gg / ~_1_ exp[-g+ 1+ 11n(¥.)]

I! .j2iCl 1
= _1_ exp[-(g -I) + lln(1 + g

.j2iCl
-I)]
1
1 g-I 1 g-I
~ .j2iCl exp [ -(g -I) + 1(
-1-
) - 2 ( -1-) 2
+ ... ]
~
1 [(g
.j2iCl exp -
_1)2]
21 (4.273)
which is a Gaussian.
The zero-temperature result was described first, with an Einstein model, to emphasize
that these additional peaks in the spectral function are really ground state properties of the
system. For acoustical phonons, one gets a continuous distribution of possible phonon
energies up to the bandwidth. The spectral function A(00) becomes a continuous function of 00.
The spectral shapes are dependent on whether the coupling to the acoustical phonons is
piezoelectric or deformation potential.
Next consider the Einstein model at nonzero temperature. The phonon contribution cI>(t)
in the Green's function is
cI>(t) = g[(N + 1)(1 - e- imot ) + N(1 _ eimot )] (4.274)
N- 1 (4.275)
-e pmo - l
Because of the relationship
N+l=e pmo ,
N
( N; 1)
1/2
= e pmo / 2 (4.276)
the exponent cI>(t) may have its terms grouped in the following fashion:
cI>(t) = g{ 2N + 1 - IN(N + 1)(e- imo (t+i P/2) + eiOlo (t+i P/2)} (4.277)
The reason for this grouping will now become apparent. The objective is to expand
exp[-cI>(t)] in a power series in exp(iooot), in order to evaluate the spectral function. Recall the
series which generates the Bessel functions of complex argument, In(z):
00
e"cos(O) = L I/(z)ei18 (4.278)
/=-00
The function exp[ -cI>(t)] has this form using
Z = 2gJN(N + 1) (4.279)
e= ooo(t + i~/2) (4.280)
The retarded Green's function at nonzero temperatures may be expanded as

Gret(t) = -i8(t) exp[ -it(Ec - L\) - (t)] (4.281)
= -i8(t)exp[-it(Ec - L\) - g(2N + 1)]
00
x L Id2gJN(N + 1)}iCilo(-it+~/2) (4.282)
1=-00
and its spectral function is
A(ro) = 21t exp[-g(2N + 1)] f

1=-00
It{2gJN(N + 1)}elCilo~/2
(4.283)
The spectral function contains a summation over the frequencies rol = Ec - L\ + rool and is a
delta function at these values. In this sense it is similar to the zero-temperature result (4.268).
Now the coefficient of the delta function is far more complicated. Another important
difference is that negative values of I are permitted. Although LI = II' the factor exp[/~roo/2)]
skews the envelope of intensities to positive values of I.
4.3.4. Optical Absorption and Emission

The relaxation effects described above can, in some cases, actually be measured. Of
course, one can never measure the properties of a one-particle Green's function, as was stated
in Sec. 3.7. Linear response theory always gives a two-particle Green's function, which
describes how the system responds when an external probe causes the system to change its
state. However, in the many-boson model, there are some two-particle Green's functions
which have properties nearly identical to the one-particle properties which have just been
derived.
One important model is a localized defect with several possible localized electronic
states. Each of these states may have a different matrix element for coupling to the phonon
field:
H = Lroqa:aq + L Vi (4.284)
q i
Vi = C] Ci [ Ei + ~Mqi(aq + a:)] == hiC] C i (4.285)
This Hamiltonian may also be exactly diagonalized. A canonical transformation of the

previous form is done below:
(4.286)
(4.287)
Tenns such as ninj will be set equal to zero, since it is assumed there is only one electron on
the impurity. It may be in different states but not in two different electronic states at once. The
Hamiltonian (4.284) was written with the electronic states not interacting with each other,
except through the phonons. Any tenns which pennit a direct interaction between the states
usually render the Hamiltonian unsolvable, at least exactly. For example, tenns such as
[CJ Cj + CJ Ci] LMq,ij(aq + a~) (4.288)

q
are not included. They pennit the particle to change its state by emitting a phonon. Such tenns
probably exist in real systems, but their addition to the Hamiltonian makes the problem much
more difficult. They are customarily omitted.
In an optical absorption process, an electron may change its electronic state, say from i to
j, by the absorption of a photon of frequency co. This process is described by the Kubo
fonnula, using the current-current correlation function. As in Sec. 4.3.3, the real-time
correlation function may be employed. It is convenient to use the version of the Kubo fonnula
given in Problem 16 of Chapter 3. For optical frequencies such that ~co » 1, the fonnula is
(4.289)
where (01, ~) are (x, y, z) indices. The relation ~co» 1 is easily satisfied, since typically
flco i :::::; 2-3 eY is in the visible spectrum, while at room temperature ~ = 40 eY-'. For the
transition between two localized states, the current operator is [see (1.194)]
j" = LPij,,,CJ Cj (4.290)

ij
The matrix element Pij,,, is treated as a constant in this problem. It plays no role in the many-
body physics which follows. The correlation function is:
(j,,(t)j~(O») = LPij,rxPkZ.~ (CJ (t)C/t)C!C z)

ijkZ (4.291)
U = (CJ (t)Cit)C!Cz) = e~Q Tr(e-~H eiHtCJ Cje-iHtC!Cz)
The Hamiltonian H is that in (4.284). This correlation function may be solved exactly by
using the same steps that were used to solve the Green's function (4.201) to get (4.259). First,
the unit operator I = e- s ~ is inserted into the trace:
(4.292)
Using the cyclic property of the trace,
(4.293)
The canonical transfonnation is taken on each tenn inside of the trace, which gives
(4.294)
The factors X; result from the transformation of the particle operators:

Ci = ,fCie-S = ,f;Cie-S; = CiA! (4.295)
X; = exp [- z=Mqi (a; - a q )] (4.296)

q ffiq
The transformed Hamiltonian H in (4.286) is diagonal in the operators C and aq- The time
development of the correlation function may be found at once:
eifltxtxct Ce- iiIt
I J I ]
= Xt(t)X.(t)Ct
I J
Ceit(e;-ej-t'J.;+t'J.
I }
j}
(4.297)
U = eit(e;-ej-t'J.;+t'J.)Tr[e-~iIx;t (t)J0(t)Xl (O)Xz(O)C; CjC!C/]
The electron and phonon parts of the trace may be separated, which is permissible because the
X; operators do not depend on particle states:
U = eit(e;-ert'J.;+t'J.) Ue/ Uph (t) (4.298)
Ue/ = Tr[e-~iI C; CjC!CIl (4.299)
Uph(t) = Tr[e-~iIX/ (t)J0(t)Xl (O)X;(O)] (4.300)
The electron part is quickly evaluated. Using Wick's theorem, the subscripts referring to
particle states must be paired. In fact, one must have that j = k and i = l:
Ue/ = Tr[e-~iI CJ CkC!Cz] = n/(1 - nk) (4.301)
Uph(t) = Tr[e-~iIX/ (t)Xk(t)Xl (O)Xz(O)] (4.302)
This result is useful when evaluating the phonon part of the trace. It is similar in form to the
earlier trace for the one-particle Green's function. The only difference is that there are now
four factors instead of two. The evaluation procedure is the same, since the four operators can
be paired and combined into two. This simplification happens because operators which are at
the same time can be combined, since their exponents commute:
~t (t)Xk(t) = eiHt (I1qeAl(a~ -aq)e -Ak(a~ -aq}) e-iHt (4.303)
= eiHt (I1qeCA1-Ad(a~-aq) )e-iHt (4.304)
(4.305)
It is only possible to add two exponential operators in this fashion when they commute. They
do in this case when taken at the same time. The correlation function Uph(t) above is
immediately simplified to
Uph(t) = I1q Tr{e-~H exp[-A(a;e iOlql - aqeiOlqt)]eA{a~-aq}} (4.306)

A = _ Mq/-Mqk (4.307)
ffiq
This equation for Uph(t) is precisely the form (4.222) which was untangled previously, and
now the effective coupling constant is A = [Mqk - Mq/]/ffiq .
The previous untangling result (4.257) can be used to obtain the evaluation of Uph(t):
Uph(t) = exp{ -<l>ktCt)}
<l>k/(t) = L
q
(Mq/ :2
Mqki [(Nq + 1)(1 - e-i(J)qt)]
q
+ N q(1 - ei(J)qt)]
These various results are collected, and the result for the correlation function for the
conductivity is
Re[cr"n(co)] =_1 Ln/(1-nk)Pk/rxP%/n

f' 2co k/ • ,f'
x f~oo dt exp{it[co + c/ - ck -11/ + 11k - <l>k/(t)]} (4.308)
The function <l>k/ in the exponent is similar to the one for the particle Green's function. The
only difference is that the effective matrix element is the difference between the two single-
state matrix elements: Mk/,q = Mqk - M q/. For example, if the two matrix elements Mqk and
Mq/ happen to be equal, then the effective matrix element vanishes. In this unlikely
circumstance the spectral function is just a delta function at the frequency co = ck - c/.
Usually Mqk and Mq/ are not equal, at least not for all different wave vectors, so that phonon
effects are present in the transition.
The above model describes dynamic relaxation. In the initial state of the system, the
electron is in a state I, and the phonons are relaxed about their equilibrium configuration for
the state I. Recall that if the phonon part of the Hamiltonian is written in terms of harmonic
oscillator coordinates, the phonons relax to an equilibrium configuration given by [see
(1.245)]
Q~l) = _2(_11_)
2pcoq
1/2 Mq /
co q
(4.309)
In the optical absorption, the electron starts in state I and ends in state k. The phonons start
with an equilibrium configuration about the Roint Q~l) but end the optical transition with the
equilibrium configuration about the point Q~ . The phonon system must alter its equilibrium
configuration during the transition. This change is a relaxation process, since it must relax to
the new equilibrium configuration during the optical step. The physics problem is to deter-
mine how the phonon relaxation process affects the absorption spectra. This information is
contained in the result (4.308).
The process is indicated schematically in Fig. 4.8, which shows a potential energy
diagram for each oscillator coordinate Qq. There are two parabolic curves, with parabolicity
co~. The lower curve describes the ground state of the system. The electronic energy is c/. If
there were no coupling to the particle in state I, the phonon parabola would be a minimum at
this point Qq = O. However, because of the coupling Mq/ to this particle, the potential
minimum is at Q~). The upper curve describes the final state potential energy curves of the
phonons plus particle. The particle energy is cb and the curve crosses here because the
phonon system has this energy when Qq = O. This potential energy curve has the minimum at
Q~). This minimum has been drawn on the other side of the axis to emphasize that it is
usually a point different than the ground state minimum. Figure 4.8 is called a configurational
coordinate diagram (Williams and Hebb, 1951). Such diagrams were originally constructed
with a single-particle coordinate Q, which represented the real atom displacement of the first
o Q
FIGURE 4.8 Configurational coordinate drawing of the independent boson model. The two parabolas represent the
phonon potential energy of the initial and final electronic states in the transition. Transition path A is most likely,
while path B is less likely but gives the zero-phonon probability.
shell of atoms about the impurity. However, the use of configurational coordinate diagrams is
much more rigorous when interpreted as the potential energy of the individual phonon modes.
The physical question is whether the optical transition happens vertically on this diagram
or along some other trajectory. For example, the arrow B is from one potential minimum to
the other. This transition is called the zero-phonon line. In reality, all these transitions are
possible, and each has a probability of occurrence. These probabilities are given by (4.308).
The physics is well illustrated by again using the Einstein model for phonons. This formula is
expanded in the same manner used to derive the one-particle Green's function (4.283):
(4.310)
rom = ck - ci - Ak + Al + mroo (4.311)
Ykl = 2gkI JN(N + 1) (4.312)
This formula for Re[cr(ro)] has exactly the same form as the one-particle spectral function
(4.283), as an expansion in Bessel functions of complex argument. Now the coupling constant
to the phonons is determined by Mqk - M ql . If Mqk = Mq/o then g = 0, Y = 0, and the
conductivity is a single delta function at ro = ck - ci'
The zero-phonon line is given by the term with m = O. It has the energy
rom=o = Ek - E/ - !ik + !i/, which corresponds to the B arrow in Fig. 4.8. The probability of
this transition is proportional to
(4.313)
The envelope of delta function heights is determined by the m dependence of the factor
Im(Yk/) exp(m~roo/2). For coupling constants gk/ greater than unity, the delta function inten-
sifies with m and then falls off at higher values. The strongest delta function occurs at positive
nonzero values of m, which is illustrated in Fig. 4.9. The circles show the peak heights
calculated for gk/ = 4 and N = !. The points are the emission spectra, which will be derived
below.
The usual interpretation is that terms with m > 0 correspond to the creation of m
phonons during the optical transition. The zero-phonon line is just the transition where no
phonons are created or destroyed. Transitions in which m > 0 are phonon emission, and
transitions in which m < 0 are phonon absorption. The phonon absorption process is possible
at nonzero temperature, when phonons are thermally present in the initial system. This
interpretation ignores the fact that both the ground and excited states are an admixture of
phonon states. The spectral functions (4.283) of both initial and final states contain admix-
tures of different numbers of phonons. These differing admixtures are due to fluctuations in
the system. Sometimes the particle has one phonon around it, while other times it has three or
four, etc. The probability of different fluctuations is given by the Bessel function coefficients
in (4.283).
The other interpretation is that the absorption spectra of Fig. 4.9 correspond to transi-
tions between states of different fluctuations. The initial state may be in a state with a fluc-
tuation of plus four phonons and the final state with plus six. It takes a net of plus two phonon
energies to complete the transition with energy conservation. This model is not really
different from the above interpretation. The phonons in the fluctuation still have to dissipate
away. However, one state may have a different number of average phonons, in its fluctuating
0.15 0J5
x 00
x
x 0
x 0
EMiSSiON ..............
x O-ABSORPTION
0.10
0.10
x 0
x 0
~
0.05 0
x
x 0 x 0
x 0 x 0
x 0 x 0
-10 -8 -8 -4 -2 0 2 4 8 8 10
m-
FIGURE 4.9 The emission and absorption spectra for an Einstein model, in the independent boson approximation,
from Eq. (4.239). The o's and x's mark the peak intensities of each phonon sideband.
cloud, than the other. The number of phonons which must be created or destroyed to make up
this average is not really creating a net amount to dissipate into the system. They will, instead,
stay at the impurity site and take part in the fluctuations.
The emission spectra will now be considered. It is assumed that the particle has been
elevated to an excited state at an earlier time, perhaps by an optical absorption process.
However, experimentally it is often done by electron bombardment. The phonons in the
excited state relax around the particle and eventually attain the equilibrium configuration plus
fluctuations. This assumption is valid only for long-lived excited states. Then a photon is
emitted while the particle drops to a lower energy state. Of course, the phonon system must
adjust during the emission process just as it did during the absorption. The same relaxation
processes are encountered again.
The emission spectra is derived without doing any more work. It may be obtained from
the absorption result by a simple argument. The emission is the rate per unit time that photons
are produced in the solid. Of course, the absorption is the rate at which they disappear. Write a
simple rate equation for the average number of photons N of frequency ro,
dN
-=-wN (4.314)
dt
which has the solution
(4.315)
where w is the rate of absorption. Since the conductivity cr(oo) also has the units of S-I, one
might expect that w ()( cr. In fact, the exact relationship is
47tcr
w=-- (4.316)
1>1
where 1>1 is the square of the refractive index. The rate of photon absorption is just propor-
tional to the real part of the conductivity, as given by the Kubo formula. Another way to
understand this is to consider rewriting the Kubo formula with some additional factors:
(4.317)
The two additional factors are the creation and destruction operators for photons b and bt .
The right-hand part of the correlation factor,
jiO)b(O) = LPkl,,,CtClb (4.318)

kl
describes the process whereby a photon is destroyed (by being absorbed) and an electron has
its state changed from I to k. This term arises from the P' A interaction. The left-hand part of
the correlation function contains the factors
jf,l(t)bt (t) = LPkl,f,lCJ (t)Ck(t)b t (t) (4.319)

kl
which describe the inverse process at another time t. The correlation function between these
two events is the Kubo formula. The photon parts may be factored out of the expression using
bt(t) = bteiOO / and (btb) = N
(4.320)
This quantity is, except for the factor of 4n, just the right-hand side ofEq. (4.314). It is the
rate that photons are destroyed in the sample by optical absorption.
For emission, calculate the rate at which photons are created. The first step in the
correlation function would describe a photon being created while the electron changes its
state. This correlation function should be
(4.321)
where the prefactor (1 - e- PW ) is dropped since it is near unity. The first operator on the right
bt creates a photon as part of the emission process. Take these photon factors out of the
integrand, using bet) = be- iwt and (bb t ) = N + 1, to obtain
(4.322)
This result, multiplied by 41t, must be the rate that photons are being made in the solid:
dN
dt =J(N + 1) (4.323)
J(w) = 4n J~oo dte-iwt(jp(t)lrxC0)) (4.324)
The formula for J(w) is identical with the Kubo formula except that the frequency has
changed sign. Compare this formula with the Einstein model result (4.310), for example, and
only change the sign of w to get
L 8(w + wm)Jm(y)emPwo/2
00
J(w) ex: e-g (2N+lj (4.325)
m=-oo
The outcome is very simple. The emission spectra is just the mirror image of the absorption
spectra reflected across the zero-phonon line. The emission is illustrated by the points marked
X in Fig. 4.9. The emission spectra are mostly on the low-frequency side of the zero-phonon
line. In general, the emission spectra have a lower average frequency than the corresponding
absorption spectra for the same processes. The difference is a consequence of relaxation,
which is illustrated using the configurational coordinate diagram of Fig. 4.10. The vertical
arrow "abs" shows the most likely absorption event. Of course, other absorption events are
possible, but the peak of the envelope of delta functions is at this vertical transition.
Similarly, the most likely emission event is the downward arrow marked "emiss." It is
shorter than the absorption arrow, which indicates that the average emission takes less energy.
No matter how one draws those two parabolas with respect to each other, the downward
emission arrow is always shorter than (or equal to) the upward arrow. This fact can also be
demonstrated by using the various theoretical formulas which have been derived. However,
the simple diagram of Fig. 4.10 is probably the clearest proof that the emission has an average
lower energy than absorption.
The lower average energy in emission, compared to absorption, is understood using the
first interpretation mentioned after (4.310). During absorption, one usually makes phonons.
The photon energy is given by the change in particle energy plus the phonon energy
w= ~(particle energy) + phonons (4.326)

FIGURE 4.10 Configurational coordinate drawing shows that the average absorption frequency is higher than the
average emission frequency.
This energy is higher than the particle transition energy. In the emission event, the energy
comes from the particle transition, but here too one makes phonons. Energy conservation is
Ll(particle energy) = 0) + phonons (4.327)
and the average photon energy 0) is less than the particle transition energy.
Figure 4.11 shows an experimental result for absorption and emission at an impurity
center in ZnTe, reported by Dietz et al. (1962). The emission and absorption are mirror
images about the zero-phonon lines. The arrows at the top show the separation of the optical
phonon lines, which are very clear in the emission spectra. The two other peaks between
successive LO phonons are sidebands due to LA and TA phonons. These data were taken at
20 K. It is an unusually good example of the relation between emission and absorption.
4.3.5. Sudden Switching

There is another way to derive the result for optical emission and absorption. Of course it
obtains the same answer, since both derivations are exact. This other derivation emphasizes
the switching aspects of the problem. It is just an alternate method of obtaining the same
answer but perhaps provides a slightly different physical insight. The many-body problem is
presented as the system response when a potential is suddenly switched on. Return to the
Kubo formula and consider the time correlation function:
(4.328)
In writing this expression, the subscripts of the particle operators are paired up. Now the
Hamiltonian H occurs at three places in this expression. It has two different forms: one for
each different place it occurs in the sequence of operators. The factor on the right is
(4.329)
VJ
CD
!"
.r:o.
W
•
:;-
Q.
40. .40 CD
1---+-,,'1' 'r - +----1-- '1 oj. ·1· --+-- ...--f-- >1 '1' '1 0" .--1 'C
CD
10.026 eV ::I
Q.
1 ," CD
I' I \
1 ::I
I' -, I \ 0+
I ~ ". I \ .... - '
30 D:J
! 1.9860eV I \/ ' o
'\ 1/1
., o
A J \ ::I
j' J "'
2
I
o I ,(IJ \, s:
o
lAo • I ' Q.
"; \ 20 '§ CD
'0 IV \ ,,; iii
I "
••~ '" I "
!
.E ,I'.,"J '
10
I·' -
,II
,
Fluorescence Absorption
r ... "'~
,I I. ) '0
O· 1.82 1.86 1.90 1.94 1.98 2.02 206 2.10 2.14 2.18
eV
FIGURE 4.11 The fluorescence and absorption at 20 K from an impurity center in ZnTe. The positions of the peak are mirrored about a central no-phonon line common to both
spectra. The LO phonon energy intervals are indicated. The other two peaks in each interval are from TA and LA phonons. Source: Dietz et al. (1962) (used with permission).
N
Co)
"""
The operators ct CI act to the right. They destroy the particle in state t and create one in state
k. The operator H acts on a system with a particle in k but not in t. According to the form
(4.284) for the Hamiltonian,
(4.330)
Hp = L: coqa~aq (4.331)
q
When H operates on a state Ik) = CtIO) then Hlk) = (Hp + hk)lk). In this case, e-iHtCtcl
may be replaced in the operator sequence by
(4.332)
This result is exact, and no approximation is involved in the replacement. The other two
factors of H are at the left of the operator sequence. On their right is the inunediate sequence
of particle operators,
e-PHeiHtctc (4.333)
I k
which returns the particle back to the state t. These two Hamiltonians operate on the system
with a particle in the state t and therefore produce
e-P(Hp+h/leit(Hp+h/lCt C (4.334)
I k
The correlation function U(t) in (4.328) is exactly equal to

U(t) -- Tr(e-P(Hp+h/leit(Hp+h/lctI Ck e-it(Hp+hklctk C)
I (4.335)
The particle operators may now be removed from the trace. The time development of the
correlation function involves only the operators Hp and hj' and these contain only phonon
operators. The particle operators have done their job, in the correlation function, by deter-
mining which effective Hamiltonians operate at various points in the sequence. There remains
only to evaluate the phonon part of the trace:
U(t) = eit(E/-Ekl nl (1 - nk)Uph(t) (4.336)
Uph(t) = Tr(e-P(Hp+ii/leit(Hp+ii/le-it(Hp+iikl) (4.337)
- _ t
hj - L:Mqj(aq + aq) (4.338)
q
This correlation function could be evaluated by Feynman disentangling. However, another

method will now be developed.
When a Hamiltonian is solved in the interaction representation one writes H = Ho + V,
where Ho is a part one can solve and V is the perturbation. Here the procedure is slightly
different. The Hamiltonian Ho is for the state with a particle in t,
Ho =Hp+h l (4.339)
and Ho + V is the Hamiltonian for the particle in the state k,
Ho + V =Hp +hk (4.340)
V = hk - hi = tk - tl + L:(Mqk - Mql)(aq + a~) (4.341)

q
The perturbation V is the difference between these two Hamiltonians. The phonon part of the
correlation function may be written as
(4.342)
The next step is to make Ho a diagonal Hamiltonian, which is not difficult, since the unitary
transformation is given earlier in (4.286):
flo = ff'Hoe- s, = EI - Al + L: Olqataq

q
flo + V = ff'(Ho + V)e- si = Ek - Ak + L:Olqataq

q
(Mqk -Mqi t
+ L:q Olq
+ L:(Mqk
q
- Mq/)(aq + aq)
(4.343)
_ Mql t
SI- L:-(aq - a q)
q Olq
,,(Mqk -Mqi
Akl=~
q Olq
The transformed potential V is obtained by subtracting flo from flo + V:
V = Ek - EI - Ak + Al + Akl + 0V (4.344)
oV- = L:(Mqk - t
Mq/)(aq + aq) (4.345)
q
The unitary transformation in (4.343) is applied to all the operators in the correlation function
(4.342). It is done following the same steps used to transform (4.291), except that here only
exp(S/) is used,
(4.346)
where flo now has the diagonal form in (4.343).

This correlation function has the form of the switching phenomena, as mentioned at the
beginning of this subsection. It describes the response of the system to suddenly switching on
the potential V at time t = O. This switching on causes numerous transients in the phonon
system, and these transients are the cause of the phonon sidebands observed in absorption.
To illustrate this switching, consider a system which has a Hamiltonian given by
(4.347)
where d t and d are creation and destruction operators for some particle. The Green's function
of the d particle is given for (t > 0)
Gd(t) = _i(OleiHtde-iHtdtIO) (4.348)
The particle d is created at time t = O. Then e- iHt operates on the state with a particle in d and
Hd t 10) = (flo + V)d t 10). At a later time t the particle is destroyed by the destruction operator
d, and eiHt operates on a state with no d-particle. The Green's function for t > 0 is
(4.349)
which is exactly the fonn of the correlation function (4.346). All the d particle did for us was
to switch on the potential V at t = 0 and switch it off at time t. This switching-on causes the
phonons to respond and adjust to the new potential V. These transients are the same
relaxation processes which were discussed earlier. The correlation function (4.346) is eval-
uated by first recalling the interaction representation result (2.1. 7):
eiHote-it(Ho+vl = Tex p [ -i 1: dtIV(t l )] (4.350)
In this case, the operator Vet) is
Vet) = eiHot Ve- iHot (4.351)
= Ek - EI -11k + 111 + I1kl + OV(t) (4.352)

oV(t) = 2:)Mqk - Mql)(aqe-iOlqt + a~eiOlqt) (4.353)
q
The constant t~nns may be immediately removed from the correlation function. Only the
operator part oV(t) needs to be time-ordered:
eiHote-it(Ho+Vl = exp[-it(Ek - EI -11k + 111 + I1 kl )]T exp[-i J: dtlo-V(tl)]
Now oV(t l ) contains only two tenns, which are, respectively, proportional to at and a. Since
the commutator of these results is a constant, one immediately considers evaluating these
time-ordered exponents by using Feynman's theorem (4.226). It is not correct to use precisely
the fonn of (4.226). The two tenns in the exponent operate at different times, and this fact
must be considered. The proper procedure is done in Problem 5 of Chapter 2, which is to
separate the two tenns in the time-ordered exponent. Again it is done for a single value of
wave vector q, and this subscript is suppressed. The equations are:
Tex p [ -i J: dtIOV(t l )] = eat<l>(tlexp[ -fA 1: dtle-at<l>(tllaeat<l>(tlle-iOltl]
t . "A . (4.354)
<p(t) = -i"A J dtle'Oltl = -(1 - e'Olt )
o 0)
The time dependence of the exponent in the integrand is

e-at<l>(tllaeat<l>(tll = a + <p(tl ) (4.355)
This time integral may now be evaluated and gives
- i"A Jt dt Ie
0 -IOlt
.
co
a + (1 - e'Olt
["A .] ) (4.356)
= -a -"A ( 1 - e -iOlt).
+ It -"A2 - -"A2( 1 - e-iOlt) (4.357)
0) 0) 0)2
The phonon correlation function is now
U h(t)
P
= e-it(ek-e,-!J.k+!J.,l exp[z= "A2 (1 _
0)2
e-iOlqt)]Tr(e-~Hoeat<l>(tle-a<l>*(tl)
q q
Note that the factor !J.kZ has canceled out of this expression. The trace is evaluated using the
previous result (4.240). The normalization factor exp[~g] has not been written explicitly in
these equations but obviously should be included. The final result is
Uph(t) = exp[ -it(Ek - EZ - !J.k + !J.z) - fl>kZ(t)]
fl>k/(t) = L (Mqk
q
:2
Mqzi [(Nq + 1)(1 - e- iffiqt )
q
+ Nq{1 _ eiffiqt )]
(4.358)
which is precisely the earlier result in (4.308).

The following physical picture emerges from this derivation. The coupled system of
particle and phonon has been sitting in a state, say I, where the phonons have been fluctuating
around the equilibrium position. Suddenly the particle is moved to state k by the optical
absorption process. From the point of view of the phonon system, it appears that the potential
oV = L(Mqk - Mqz)(aq + a~) (4.359)

q
has been suddenly switched on at that time which is taken to be t = O. The transient response
in the phonon system is measured by switching off the potential at a later time and measuring
the correlations which result. The resulting function of time describes the temporal evolution
of the phonon system. Its Fourier transform gives the phonon sideband structure which is
observed in the optical spectra. This picture explains the mirror relation between emission and
absorption. In absorption the potential which is switched on is (4.359), while in emission, if
the particle transition is in the opposite direction, the potential which is suddently switched on
is just the negative of (4.359).
4.3.6. Linked Cluster Expansion

Another way to evaluate the time-ordered exponential operator
U(t) = T( exp [ -i 1: dt) V(t)) J) (4.360)
is by the linked cluster expansion, which was discussed in Sec. 3.6. It is sufficient to derive
the result for a single-phonon state of wave vector q. The summation over the q vector in the
exponent is easy to do at the end. Consider the case where Vet) is written as
Vet) = M(t) (4.361)

(4.362)
The S matrix is expanded directly, and each term is evaluated. The series is presumed to be an
exponential series. In the present example, there is only one term which remains after the
resummation. The nth-order term is
00
U(t) = L(-itUn(t) (4.363)
n=O
(4.364)
Since A(t) describes the creation or destruction of a phonon, the A operators always exist in
pairs. Only tenns with n even are nonzero; the others are zero. Since n is even, define n = 2m:
00
U(t) = L (-ii mU2m (t) (4.365)

m=O
(4.366)
According to Wick's theorem, the A operators pair up to fonn phonon Green's function
(TA(t l )A(t2)} = iD(t l - t2) = (N + 1)e-iwltl-t21 + Neiwltl-t21
1 (4.367)
N=--
e~w - 1
Each phonon Green's function depends on the difference of two time variables D(ti - 0). The
time integrals over those two variables define a function
(t) =i 1: 1:dt l dt2D(t l - t 2) (4.368)
= ~[(N + 1)(1- e- iwt ) +N(1- eiwt )] (4.369)

co
2
The mth tenn has m phonons:
(TA(t l )· .. A(t2m )} = L D(tl - 0)D(0 - t;) ... D(tu - t2m )

aU combinations
The quantity U2m (t) is just proportional to (t)m, since each combination of time-ordered
products gives this same result. The only remaining question is the combinatorial determi-
nation of the number of different arrangements. The number of such combinations is
(4.370)
which is obtained in the following way. The first variable, tl> may be paired with any of the
2m - 1 other variables. The next time variable, which is t2 if it was not paired with t l , is
paired with any of the other 2m - 3 variables. The number of combinations is
(2m)!
(2m - 1)(2m - 3) .. ·3,1 = - - (4.371)
2mm!
The result is
)..?m [(t)]m (4.372)

U2 (t)=- -
m m! 2
which is summed to obtain the correlation function:
U(t) = exp [ _ ~2 (t)] (4.373)
The function )..? /2 is now recognized as the exponential function of time which has already
been derived several different ways. It is just the double integral over the phonon Green's
function. The linked cluster expansion is obviously simple, since there is only one distinct
linked cluster, which is <J>(t).
Another way to evaluate the Green's function is to leave the particle operators in the S-
matrix expansion. There are occasions when it is not immediately apparent that they can be
factored out, as was done in (4.328), to give the starting point for (4.346). There are several
exactly solvable models which can be evaluated this way. It is useful to learn the technique.
Consider the Green's function for a single particle in a band, which is coupled to phonons.
The subscript q is suppressed again in the notation:
H = Ho + V, Ho = EoCtC + roata (4.374)

V = ACtC(a + at) == ACtCA, A = a + at (4.375)
G(t) = -i(TC(t)Ct(O)) = _i9(t)(eiHotCte-iHtc) (4.376)
G(O)(t) = _i0(t)e- iEot (4.377)
Since there is only one particle in the band, the creation operator must be to the right, so
t > O. The unperturbed Green's function G(O)(t) has a simple form. The interacting Green's
function is written, following Chapter 2, as
G(t) = -i(TC(t)U(t)ct(O)) (4.378)
The U(t) matrix, or S matrix, is now expanded in an infinite series of terms. This infinite
series will generate the linked cluster expansion. However, now the particle operators C t and
C are included, along with the phonon operators a and at. Again only terms with n even are
nonzero, and the 2mth term is
(_iA)2m
G2m (t) = -i0(t) (2m)!
JI
0 dtt . ..
JI
0
~ ~ ~
dt2m (TA(tl )A(t2) ... A(t2m ))
x (TC(t)Ct(tI)C(tI)'" Ct(t2m)C(t2m)Ct(0)) (4.379)
Now there is a time-ordered product of particle operators. At first the time-ordering appears to
complicate the evaluation procedure. After all, this time-ordered product of operators,
(4.380)
must be expanded, according to Wick's theorem, into all possible pairings. However, a careful
inspection shows that the time-ordered product has a trivial evaluation. The number operator
Ct C is time independent in the interaction representation since Ho commutes with the number
operator,
(4.381)
so that the time-ordered correlation function just contains a product of number operators:
~ 2m ~ t ~ ~ t
W(t, t l , ... , t2m ) = (TC(t)n C (0)) = (TC(t)C (0)) (4.382)
The number operator gives unity when operating upon the state with one particle. The W
function does not actually depend on any time variable except t. The term remaining, on the
right-hand side, is just proportional to G(O) (t). Since this same factor occurs in each term in
the S-matrix expansion it may be factored out of the series:
00
G(t) = L G2m (t)

m=O
(4.383)
What remains in the series is just the same linked cluster expansion which was evaluated
above. The exact Green's function is, again, our result which is now very familiar:
G(t) = G(O)(t) exp [ _ ~2 <P(t)] (4.384)
The presence of the C operators in the linked cluster expansion did not, in this case, change
the result. Usually that does not happen. For most Hamiltonians, the presence of the C
operators changes the evaluation of each term in the S-matrix expansion. These cases are
often not exactly solvable and certainly are not by the method under discussion.
The usual polaron problem, for one free particle in a band, is described by the Hamil-
tonian derived in Sec. 1.3:
H = L EkC~Ck + L[{Oqa~aq + MqAqp(q)] (4.385)

k q
Aq = aq + a~q (4.386)
p(q) = L C~+qCk (4.387)

k
The coupling between the particle and phonons depends on the particle density operator,
(4.388)
which is time dependent. If one were to attempt to solve for the particle Green's function by a
linked cluster expansion, the particle part of the correlation function becomes far more
complicated. This Green's function is
G(k, t) = -iE>(t)(eiHtCke-iHtC~) = _iE>(t)(eiHotCke-i(Ho+V)tc~)
(4.389)
V = LMqAqp(q).
q
When evaluated by expanding the S matrix, it produces the 2mth term of the same general
form as (4.379):
(4.390)
Now the particle part of the correlation function depends on products of the density operators
at different times. Usually this function does not have a simple time dependence, and the
polaron problem is not exactly solvable. One may still evaluate it by a linked cluster
expansion. But this process now becomes an approximate procedure, whereby one evaluates
only a few terms in an infinite series, and the remaining terms are omitted. This approx-
imation will be discussed in Chapter 7.
Van Haeringen (1965) proved the theorem which specifies the most general possible
conditions for which (4.389) may be solved exactly by the linked cluster method. An exact
solution is obtained whenever p(q, t) has the form
p(q, t) = eiif(q)p(q, 0) (4.391)
p(q, 0) == p(q) = L: Ct+qCk (4.392)

k
Since the general time dependence is given by the kinetic energy difference I(k, q) =
Ek+q - Ek, his condition is that this difference is independent of k:
(4.393)
If these conditions are met, the exact solution to the Green's function may still be obtained
from the linked cluster method. Since the factor exp[iif(q)] is a c number, and not an operator,
it may be removed from the time-ordered product of particle operators:
(TC(t)p(ql' t l )· .. p(q2m, t2m )ct(0») = exp[ij~ tf('t)] (TC(t)p(ql)' .. p(q2m)Ct(0»)

(4.394)
This step leaves, in the time-ordered product, just the time-independent density operators
p(q). These operators have the same effect that the number operators did previously. They do
not change the value of the correlation function. It is exactly equal to
2m tf(qj)] (TC(t)Ct(O») = iG(O)(k, t)exp [2m]

exp [ij'f A A
ij'f tf(qj) (4.395)
The validity of (4.395) may be shown in the following way. The effect of p( q) on ct 10) is just
to change the particle from the state k to k + q:
(4.396)
The effect of a product of such density operators is to change CtlO) by a summation of all
their q vectors:
P(ql)P(q2)' .. p(q2m)CtIO) = C~IO) (4.397)
Q=k+L:q.=k
. ~ (4.398)
}
The result requires that the summation over all of the phonon wave vectors is zero. This
cancellation is required by the phonon part of the correlation function. When this correlation
function is evaluated, by pairing according to Wick's theorem, the pairing forces the q values
to be equal and opposite in sign:
(TAq;(taAIJj(~») = l>q;=_1Jj (TAq;(tj)A_qJ~») = il>q;=_qP(qj' tj - tj)

The summation of all the q values is certainly zero, since they are paired in sets of qi and -qi.
The assertion (4.395) is proven.
The extra time factors exp(iif) go into the evaluation of the linked cluster expansion.
Define a new function:
(4.399)
It is evaluated for the case f( -q) = -f(q), which is the interesting physical case discussed
I
below:
<1>f(t) =
2/- itffi q
~Mq ffi~ _ j2 + (Nq + 1)
[1 - e-i/(fJ)q-fl]
(ffiq _ f)2 + Nq
[1 - e-i/(fJ)q+fl]
(ffiq + f)2 (4.400)
Terms with ffiq - f(q) are identical to those with ffiq + f(q) by the variable change: q --+ -q.
The exact result for the Green's function (4.389) is
G(k, t) = -iE>(t)e-iEk1e-/(/l (4.401)
Equation (4.401) is the most general form for G(k, t), which may be exactly solvable,
according to van Haeringen.
What kind of systems will obey his condition that tk+q - tk = f(q)? One possibility is
that the energy is a constant, tk = to so that f = O. It is this example which has been
repeatedly worked throughout this section. However, there is another case which is now
recognized as an exactly solvable model. It has the particle moving with a constant velocity v,
so that its energy is
(4.402)
f(q) = v·q = -f(-q) (4.403)
It should be remarked that the system of equations is not Galilean invariant. Even though the
particle is moving with constant velocity, one cannot transform the equations into a system
where it is standing still, in a stationary system of phonons. The phonons, particularly the
acoustical phonons, have their own velocity and are not invariant under a Galilean trans-
formation. Instead, the phonon frequencies appear Doppler shifted to the particle. The particle
Green's function could be evaluated for the case that it was going faster or slower than the
acoustical phonons system. Presumably the phonon relaxation around the particle will vary
considerably between these two cases. This model is illustrated in the assigned problems. The
one-dimensional constant velocity model for polarons was solved by Engelsberg and Varga
(1964).
In the polaron problem the particle-phonon coupling produces fluctuations in the
number of phonons surrounding the particle. This number fluctuates from time to time and
from particle to particle. In the independent boson model, each phonon, in this fluctuating
cloud around the particle, is assumed to exist with a probability which is independent of
whether other phonons are also simultaneously present. Each phonon is fluctuating inde-
pendently of the others. If a particle is fixed, after emitting a phonon it is still fixed. The
probability of emitting the second phonon is the same as the probability of emitting the first.
When the probability of each emission or absorption is the same, then one has the inde-
pendent boson model. Of course, a particle moving with a constant velocity v also has its
motion unaffected by how many phonons have been emitted. Again the probability of
Sec. 4.4 • Bethe Lattice 247
emitting each phonon is independent of others which may have been emitted, so that all
phonon emissions are independent.
A free particle with kinetic energy Ek = k2/2m is not described by the independent
boson model. After emission of one phonon with wave vector q, the particle goes to an
intermediate state with wave vector k + q and energy (k + qi
12m. The intermediate state is
different from the starting one, so that the emission of the second phonon has a probability
that is different from the emission of the first. Each emission has a different probability, so
that the probability of n phonon emissions is not just a Poisson distribution,
an
Pn=e-()( , (4.404)
n.
where a is the probability of a single emission. The Poisson distribution is the characteristic
zero-temperature distribution of phonon sidebands only when each emission has the same
probability, which is independent of the number of other phonons emitted.
The independent boson model has been used widely, with a number of variations.
Almbladh and Minnhagen (1978) solved the Fano--Anderson model with phonon coupling to
the localized level. A number of related models have been reviewed by Cini and D' Andrea
(1988).
4.4. BETHE LATTICE
The terms Bethe lattice and Cayley tree both refer to a type of lattice in which there are
no paths which are loops. Each site has z nearest neighbors, where z is an integer greater than
one. Since z = 2 is a one-dimensional chain, one usually takes z > 2. Figure 4.l2(a) shows
the Bethe lattice for z = 3, where a site is at each vertex. Starting from the central point, there
are three neighbors. Each of those neighbors also has three neighbors. Of course, each point
has the same symmetry.
4.4.1. Electron Green's Function

The Green's function for the Bethe lattice will be calculated for the Hamiltonian with the
nearest neighbor hopping term
H = -t L CJ Cj +O - ~ L CJ Cj (4.405)
J.O J
The summation over 8 runs over the z neighbors. Spin plays no role in this calculation and
will be omitted. The last term contains the chemical potential ~. So how does one calculate
the band energy of a lattice without a regular crystal structure? The solution was given by
Brinkman and Rice (1970).
As usual, the Hamiltonian is written as H = Ho + V. In this case set Ho = -~. The
hopping term is put into V. Define the Green's function '!f(l, iPn) as that needed to advance I
(a)
j=1
j=O
FIGURE 4.12 (a) Bethe lattice for z = 3 as usually drawn. (b) The same lattice with a boundary.
steps in the lattice. Since Ho = -IlN the noninteracting Green's function for I = 0 is given
below:
'§(l, iPn) =- J: d'teiPnT(TTCJ+lr)C](O)} (4.406)
'§(O)(O iP)=-JP d'teiPnT(TC.('t)d(o)} =_1_ (4.407)

'n 0 T ) J iPn + II
Every site has the same Green's function, so there is no need to put on a label such as "j".
The self-energy L(O, iPn) is found using perturbation theory. Since the interactions are
simple, it is possible to find the exact expression. The self-energy is found from processes
where an electron hops to its neighbors and then hops back. Each hop has a matrix element t.
The total number of hops in each self-energy is an even number such as 2n. For 2n hops,
which return the particle to the original site, the self-energy is denoted as L(n). The lowest-
order self-energy (n = 1) is from the process where a particle hops to the neighbors and then
hops back to the original site. There are z neighbors, and the intermediate Green's function at
the neighboring site must be ~(O)(O, ipn). The self-energy from double hops is
zt2
L(l)(O, ip ) = -.- - (4.408)
n lPn + 11
The next term in the self-energy comes from the fourth order of perturbation theory, and
includes terms where the particle hops twice to a second-nearest neighbor, and then hops
back. The first hop has z choices, but the second only has z - 1 since the hop back to the
starting point has already been included. This term is
L(2)(0 . ) _ z(z - 1)t4 (4.409)

,IPn - 3
(iPn + 11)
A better way to write the self-energy, including both terms, is
L(I+2)(0 ip ) = ----,------,.
zP
(4.410)
'n. (z _ 1)t2
lPn + 11 - iPn + 11
=~ [1 + (z - l)P + (Z - 1)P)2+ ...] (4.411)

lPn + 11 (iPn + Ilf (iPn + Ilf
The terms in the series in brackets describe processes where, after hopping to the first
neighbor, the particle hops to-and-from its neighbors mUltiple times before hopping back to
the original site. It can only hop back to the original site once. When it hops away from the
original site the second time, it counts as another self-energy function. Equation (4.411)
includes terms which are higher order in the perturbation expansion.
The factor multiplying zt2 in Eq. (4.410) has the appearance of a Green's function. It is
the Green's function for the particle as it hops around. The same Green's function should be in
its own denominator. This process leads to a continued fraction
zt2
L(O, iPn) = ----------,2:---- (4.412)
(z - l)t
iPn + 11 - ----'---''----,-,..---
(z - 1)t2
iPn + 11 - - - - - -
iPn + 11- ".
Every denominator is the same. Further terms in the continued fraction describe processes
where the particle hops further away in the lattice. The continued fraction contains all forward
hops to all sites. This expression is the exact self-energy.
The self-energy (4.410) can be solved in a simple way by defining the self-energy in the
denominator as ~D
~D = (z-l)P (4.413)
ipn +j..l- ~D
The continued fraction yields a quadratic equation which is solved easily
o= ~1- (iPn + j..l)~D + (z - l)r (4.414)
~D =! [iPn + j..l- J(iPn + j..li - 4(z - l)t2 ] (4.415)
zP
~(O, iPn) = . ~ (4.416)
lPn + j..l- D
2zP
(4.417)
iPn + j. l + J(iPn + j..li - 4(z - l)t2
1
~(O, iPn) = . . (4.418)
lPn + j. l - ~(O, lPn)
Eq. (4.417) is the exact self-energy, and (4.418) is the exact Green's function, for the Bethe
lattice in the hopping model.
The self-energy can be used to calculate the density of states. First, recall how it is
evaluated for the tight-binding model of a regular crystal. The density of states in d
dimensions for a crystal is
N(E) = J dd kd o(E - e(k»

(21t)
(4.419)
e(k) = _tL:eik./i (4.420)

ii
This expression is also just the imaginary part of the retarded real space Green's function
Gret(r, E) evaluated at r = 0:
Gret(r, E) =
J ddk e,'k·r
(21t)d E - e(k) + ill
(4.421)
-2 Im{G(r = 0, E)} = 21t J dd kd O(E - e(k»

(21t)
(4.422)
= 21tN(E) (4.423)
Sec. 4.4 • Bethe lattice 251
For the Bethe lattice, the above Green's function is in real space at lattice spacing equal to
zero. Obtaining the retarded function (iPn + Jl ~ E + ill), and taking its imaginary part gives
nN(E) = -Im[Gret(O, E)] = -Im[E + '0 _ L1 (0 E _ )] (4.424)

I ret' Jl
nN(E) = E;)E1- E2 0(E2 _ E2) (4.425)

E: - 4t2E2 t
E2s = 2zP (4.426)

(4.427)
It is useful to renormalize the energy by 0 = E/(zt) and to write the result in dimensionless
form
p(O) = 2ntN(E) (4.428)
/02 02
=V 0 0(02 _ 02) (4.429)
1-02 0
02 _ 4(z -1)
0- z2 (4.430)
One can verify that for z = 2 then 0 0 = 1 and the density of states is
p(O) = 0(l - 0 2) (4.431)
)1-02
This result is identical to that of a particle on a one-dimensional chain with nearest neighbor
hopping. Then its energy is f.k = -2t cos(k) and the density of states for this dispersion is the
inverse square root function given above. The theory gives the exact result for a chain of
atoms, when z = 2. For larger values of z > 2 then 0 0 < 1. In these cases the denominator
1 - 0 2 is never zero. The function p(O) is shown in Fig. 4.13 for the cases z = 4 and z = 6.
The density of states has a smooth shape over a continuous band of energies. The offsite
Green's function is evaluated in the homework problems.
4.4.2. Ising Model

The Ising model for a lattice of spins is derived in Chapter 1. The usual case is to have
spin one-half particles at each lattice site. Their z component of spin can point either up or
down, which is called 0' = ±1. The spins on neighboring sites (ij) interact by -JO'iO'j' The
Ising model with a magnetic field can be solved exactly, analytically, for a one-dimensional
chain. In two dimensions it can be solved exactly without a magnetic field, as first done by
Onsager. It is interesting that it can be solved exactly for the Bethe lattice. The Hamiltonian is
H = -J L O'jO'j - hL O'j (4.432)
(if) j
where h == J..LH is the Zeeman energy in a magnetic field.

The parameter z is the coordination number. It is the number of first neighbors for each
site. A related parameter is r = z - I is called the "branching ratio". When a particle comes
to a site, it has r choices of paths to move forward. Here the particles are not moving, but only
their spins interact. Figure 4. 12(a) shows a Bethe lattice with r = 2. The same lattice will be
2.0
1.8
1.6
z=3
1.4
-
1.2
g 1.0
0.
0.8
0.6
0.4
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0
FIGURE 4.13 Normalized density of states for the Bethe lattice for z = 3, z = 4, and z = 6.
drawn in a different way in Fig. 4.12(b). If the lattice is composed of sticks with hinges at the
vertices, then if it is picked up at one point, all of the sticks will fall, thereby producing the
version of the lattice in Fig. 4.12(b). This version is convenient for solving the Ising model.
The bottom of the figure is called the "boundary" of the lattice. The partition function is
obtained by averaging the spins at the boundary, and then moving inward row-by-row.
The first step is to average all of the spins in the j = 0 row. Each spin in the j = 1 row is
connected to r spins in the j = 0 row. Let cr denote a spin in the j = 1 row, and let
( cr I' cr 2' ... , crr) denote the r spins connected to it in the j = 0 row. Averaging just this small
complex of spins gives a contribution to the partition function
L exp{f3Jcr(crl + cr2 + ... + crr) + ~h(cr + crl + cr2 + ... + crr)}

(Jl"'(Jr=±1
(4.433)
Each spin has an effective interaction with its r neighbors in the lower row of
cr[h +J(crl + ... + crr)]. In the nonmagnetic state, the sum over the neighbors will average
out to a small number. In the magnetic state, the sum over the neighbors could add up to ±r. It
is useful to define the effective magnetic energy hj which acts upon the spins in row j. A
shorthand notation will be that Bj = ~hj' For the first row there are no spins below, so that
ho = h, Bo = ~h. For the next row, Eq. (4.433) is used to define BI as
(4.434)
Setting cr = ± 1 gives two equations which are solved for the two unknowns (A I' B I):
Altl1 = e~h[2cosh(Bo + I3J)Y (4.435)

A I e-B1 = e- Ph [2 cosh(Bo - I3J)Y (4.436)
B = ~h + ~ In [COSh(Bo + 13J)] (4.437)

I 2 cosh(Bo - I3J)
Al = 2r[cosh(Bo + I3J) cosh(Bo -13J)y/2 (4.438)
The factor of Bo is inserted instead of ~h in the argument of the hyperbolic cosines since the
effective field is associated with the spins on the lower row. The factor of ~h to the right of the
equal sign in (4.437) is from the energy of the spin in row j = 1. The formula (4.437) can be
manipulated by taking
cosh(Bo + 13J) cosh(Bo) cosh(l3J) + sinh(Bo) sin(l3J)

cosh(Bo - 13J) = cosh(Bo) cosh(13J) - sinh(Bo) sin(13J)
(4.439)
1 + tanh(Bo) tanh(l3J)
= (4.440)
1 - tanh(Bo) tanh(13J)
In[I+X]
I-x
=2 tanh-Ix (4.441)
These identities combine to give the result
(4.442)
The second step is to average the spins in the row j = 1 to give the effective field B2 = ~h2 in
the next row. The procedure is exactly the same. The general recursion relation, as the rows
are averaged one-by-one, is
(4.443)
As the spins are averaged, row-by-row, the effective field Bj = ~hj converges to the value in
the interior of the Bethe lattice. This bulk: value is denoted as B*. It obeys the self-consistent
nonlinear equation
B* = ~h + r tanh-l[tanh(B*) tanh(I3J)] (4.444)

tanh[(B* - ~h)/r] = tanh(B*) tanh(l3J) (4.445)
The solutions to this equation describe the collective states of the Ising model on the Bethe
lattice.
The first example is to assume: (i) zero magnetic field (h = 0), and ferromagnetic
coupling (J > 0). The transition temperature Te, (~e = IjkB Te), is where the ordering begins
as one lowers the temperature. At the transition temperature the order parameter B* is zero,
and it increases in value as the temperature is lowered. It has an infinitesimal value at a

temperature which is infinitesimally smaller than Tc. In this case the above equation is
B* = rB* tanh(~c-f) (4.446)
tanh(~c-f) = !r (4.447)
(4.448)
(4.449)
The case for r = 1 is a one-dimensional chain. In this case Tc = 0 and there is no ordered
state at nonzero temperature. For all other branching ratios r> 1 there is a well-defined
transition temperature.
It is interesting to compare these results of the Bethe lattice with the results for the Ising
model on crystalline lattices. What is actually tabulated in Table 4.1 is the quantity Kc for
various values of r. Exact results are also shown for crystalline lattices in two and three
dimensions (Mahan and Claro, 1977). The lattices are: honeycomb (hc), square (sq), plane-
triangular (pt), simple-cubic (sc), body-centered-cubic (bcc) and face-centered-cubic (fcc).
The value of Kc for the Bethe lattice is typically below the crystalline result by 10-20%.
However, the Bethe lattice has the correct trend that Kc decreases, and Tc increases, as the
branching ratio r increases. The crystalline results are analytically known in two dimensions,
!
since the Ising model can be solved exactly. For example, Kc = In(2 + .J3) for the hc lattice
and Kc = !!n(1 + ,J2) for the sq lattice.
Another solution to Eq. (4.444) is to solve for the limit that the temperature goes to zero
when the magnetic field is zero. In the limit that K = /3.l --+ 00 the equation to solve is
tanh(B* /r) = tanh(B*) tanh(K) (4.450)
In the limit that the argument of the hyperbolic tangent function becomes large, its value
approaches one. The limit is
eK -K
lim tanh(K)
K-+oo
= eK +
- e -K = 1 -
e
2e-2K. + O(e- 4K ) (4.451)
TABLE 4.1 Ising model ferromagnetic

transition termperatures (Kc = J /kBTc). The
dimension is d and r is the branching ratio.
The column Kc is the exact crystalline result,
while Kc(B) is the result from Bethe lattice
d Crystal r Kc Kc(B)
2 hc 2 0.659 0.549
2 sq 3 0.441 0.347
2 pt 5 0.274 0.203
3 sc 5 0.222 0.203
3 bcc 7 0.157 0.144
3 fcc 11 0.102 0.091
so that (4.450) becomes
(4.452)
B* ~rK (4.453)
h* =rJ (4.454)
where B* = Ph*. The solution to (4.450) is that the effective local field from the r spins in the
row below is just h* = rJ. At zero temperature, all of the spins are aligned in the ferro-
magnetic arrangement.
The derivation at the beginning of the section began by trying to evaluate the partition
function. It was used to derive a self-consistent equation for the order parameter B*. Now
return to the task of finding the partition function. The first step is to count the number of
spins in the Bethe lattice. The counting is done using Fig. 4.12(b). The top row (j = L) has
one spin. The next row down has z spins. The next row below that has zr. Successive rows
have z';. The boundary row has (j = 0) has No = zyL-l. The number in each row is
~ = zyL-l-j exceptNL = 1. The total number of spins isN = L~ = I +z(yL - l)/(r -1).
The partition function was averaged for a cluster of r spins in the boundary row. The
number of such clusters is N 1 • So the total partition function after averaging the first row is
(4.455)
= .s;1~o exp[BI ~ criJ

,=1
(4.456)
.s;10 = A:/r = 2[cosh(Bo + j3J) cosh(Bo _ j3J)]I/Z (4.457)
This process is repeated for the second row (j = 1) which gives the result, including the first
two rows, of
Zz = .s;1~o .s;1fl exp [Bz ~ criJ

1=1
(4.458)
.s;11 = 2[cosh(B 1 + j3J) cosh(B 1 _ j3J)]I/Z (4.459)
This process is repeated for the subsequent rows. The important contribution to the partition
function for each row is the factor of .s;1;. The final partition function is
Z = IT ..s;1f>!i ~ .s;1*N (4.460)

J J
.s;1* = 2[cosh(B* + j3J) cosh(B* _ j3J)]I/Z (4.461)
In the last equation, the bulk partition function is assumed to be independent of edge effects.
The value of the order parameter for the bulk (B*) is used in the definition of .s;1*. That is,
assume that Bj rises to its bulk value after a few rows, so that the bulk value is relevent for the
majority of the spin sites. The expression for d* can be shortened. The quantity in brackets is
(where K = j3J):
cosh(B* + K) cosh(B* - K) = cosh2(B*) cosh2(K) - sinh2(B*) sinh2(K)

= cosh2(B*) cosh2(K)[1 - tanh 2 (B*) tanh2(K)]
= cosh2(B*) cosh2(K)[1 - tanh2«B* - ~h)/r)]
cosh2(B*) cosh2(K)
(4.462)
- cosh 2[(B* - ~h)/r]
d* = 2 cosh(B*) cosh(K) (4.463)
cosh[(B* - ~h)/r]
where Eq. (4.444) is used to simplify the expression.

The ferromagnetic system has a sharp phase transition only at zero magnetic field
(h = 0). In this case B* = 0 above the transition temperature, and then d* = 2 cosh(K). The
expression (4.463) gives the formula in the ordered state by setting h = O. In the case of
nonzero magnetic field (h =1= 0), the ordering of the spins is gradual as the temperature is
lowered. Even at high temperatures, the parameter B* does not vanish but approaches
B* --+ ~h. In this case one should use (4.463) with all of its various factors.
4.5. TOM ONAGA MODEL
The Tomonaga model (Tomonaga, 1950) describes a one-dimensional electron gas. The
procedure is to examine the Hamiltonian of the one-dimensional electron gas and make some
approximations on it. As a consequence of these approximations, the Hamiltonian becomes
exactly solvable. The one-dimensional electron gas is not exactly solvable but only an
approximate version of it.
The important physics is the recognition that the excitations of the electron gas are
approximate bosons, although the elementary particles, electrons, are fermions. The excita-
tions involve two-particle states, for example, moving an electron from one state to another.
The wave function of the two fermion states has boson properties. The Tomonaga model
assumes that the excitations are exactly bosons, which is the important approximation.
The model has been useful in several kinds of problems. First, there are organic solids
such as TTF-TCNQ whose conductivity is thought to be largely one dimensional (see
Heeger, 1977). The Tomonaga model has played a role in the interpretation of electrical
conductivity in these materials (see Luther and Emery, 1974). Second, in impurity problems,
or X-ray absorption problems, the response of the electron gas to the central impulse can be
factored into spherical harmonics associated with different angular momentum states I. Each
angular momentum channell then becomes a one-dimensional electron gas to which one may
apply the Tomonaga model. Recently, semiconductor nanotechnology permits the construc-
tion of semiconductor channels which act as one-dimensional conductors. The Tomonaga
model is used in the theory of these systems. Single wall carbon nanotubes are another one-
dimensional conductor.
Sec.4.5 • Tomonaga Model 257
4.5.1. Tomonaga Model

The original model of Tomonaga (1950) discusses the following Hamiltonian for the
one-dimensional interacting electron gas:
I
H = VF ~ IklataA;, + 2L ~ VkP(k)p( -k) (4.464)
p(k) = Lps a;-k/2,Sap+k/2,s (4.465)
The system has length L, and VF is the Fermi velocity of the particles, which are assumed to
have a linear dispersion relation. The label s = ±l denotes spin, and p(k) is the electron
density operator. The electron-electron interaction term Vk will be specified below. It is not
41te2 / ~, which is dimensionally incorrect in one dimension, since Vk has units of Joule-
meter. Dimensional analysis suggests the form Vk <X e2(kF/k)n, where n is any exponent. The
summation over k states may be turned into integrals by the usual transformation as L -+ 00:
~f(k) = ~ Jdkf(k) (4.466)
The basic step in the Tomonaga model is to divide the density operator into two terms:
(4.467)
(4.468)
p(k) = PI (k) + P2(k) (4.469)
The density operator p(k) commutes with any other density operator p(k'). However, the two
parts PI and P2 do not commute with the same parts for other wave vectors. Examine the
= L [a;_k/2,Sap+k'+k/2,s0(p + k/2 + k' /2)

s,p>O
- a;-k'-k/2,Sap+k/2,s0(p - k/2 - k' /2)] (4.470)
An important special case is k' = - k
[PI (k), PI (-k)] = L [np-k/2,s - np+k/2,s] = L L np,s (4.471)

s,p>O s -k/2SJ~k/2
The right-hand side shows that the commutation relations depend on the operator nps over a
range of p values. The operator nps is replaced by its average in the ground state of the free-
particle system.
L L n = 2 L 0(kF - IPI) = { 2(kL/21t), k < 2kF

s -k/2SJ~k/2 ps -k/2SJ~k/2 2kFL/1t, k > 2kF
and the commutation relations (4.471) can be written for k < 2kF as
[PI (k), PI (-k)] = (~) (4.472)
[p2(k), P2( -k)] = - (~) (4.473)
[PI (k), P2( -k)] = 0 (4.474)
The analogous results are included for the other commutators, which can be derived in the
same fashion. The Tomonaga model assumes that these density operators obey the exact
commutation relations of
[PI (k), PI (-Ie') = ok,k' (~)

(4.475)
[p2(k), P2( -k')] = -ok,k' (~)
[PI (k), P2( -k')] = 0
These relations are the central approximation of the Tomonaga model. The commutation
relations are not exact, since the commutators give operators, as in (4.470). However, these
results are obtained when taking the expectation value of the exact commutation relations. For
example, in (4.470)
([PI (k). PI (k')]) = E [(a;_k/2,Sap +k'+k/2,s)E>(P + k/2 + Ie' /2)

s.p>O
(4.476)
In the right-hand side, the averages are zero unless Ie' = -k, so that
(4.477)
Although the commutation relations (4.475) are not exact, the expectation values of these
commutators are given exactly. The approximation is not a very bad one.
It is convenient to express the density operators pi±k) in terms of creation and
destruction operators. This step is done so that the creation operators are dimensionless and
the commutation relations (4.475) are obeyed. The creation and destruction operators are for
bosons. These definitions are given below, where the symbol k is always positive:
PI(k) =bk~
PI(-k)=b!~
(4.478)
P2(k) =b~k~
P2( -k) = b_ k (ki
'Ire
[b k• b!,] = Ok,k' (4.479)
When k is positive PI (k) ex: bk> and when k is negative PI (k) ex: b~k' The operators PI always
commute with P2' The choice (4.478) does satisfy the approximate commutation relations
(4.475).
The second term in the Hamiltonian (4.465) may be written in terms of these boson
operators:
(4.480)
(4.481)
The electron-electron interaction term has been recast into an interaction between the boson
excitations of the electron gas.
The first term in (4.465) is the particle kinetic energy. It requires some additional work in
order to express it in terms of boson coordinates. It is not immediately obvious how to express
a!ak in terms of the new boson operators. When faced with this predicament, it is useful to
examine the commutation relations of this operator. The objective is to find a boson repre-
sentation of the kinetic energy operator which reproduces the commutation relations. If this
cannot be done exactly, at least try to find a good approximation. The commutator algebra
completely specifies the excitation spectrum of the system, so that the excitations are
adequately described by operators with accurate commutation relations.
Call the kinetic energy term Ho. Its commutator with PI (k) is
[PI (k). Ho] = VF L L 1k'I[a;_k/2,sap +k/2,s, aL,ak's']

s,p>o s'k'
= VF L a;-k/2,Sap +k/2,i1P + kl21 - IP - kl21) (4.482)

s,p>o
k i f P > kl2
lP+k121 - IP - kl21 = { (4.483)
2p if p < kl2
For small values of k, then p > k/2 over most of the p summation. In this case the above
commutator is approximately given by
[PI (k), Hol = vFk I: a;-k/2,Sap +k/2,S = vFkPI (k) (4.484)

s,p>O
The above is a desirable form for the commutator, since the right-hand side is also propor-
tional to PI (k). With the boson representations (4.475) and (4.478) for PI (k), the approximate
commutation relation (4.484) is
(4.485)
Of course, the same result would be given by the choice of Ho = I:k (Okblbk' Next consider
the commutator of Ho with P2' The same approximation in this case leads to
(4.486)
Both of these approximate commutators are satisfied with the following choice for Ho:
(4.487)
(4.488)
The one-dimensional electron gas (4.465) has been recast into the boson Hamiltonian (4.488).
The latter is exactly solvable, as will soon be shown. The Tomonaga model (4.488) has been
derived from (4.465) with several key approximations on commutation relations. The form
(4.488) is a description of the boson excitations of the electron gas.
Equation (4.488) may be solved exactly by a variety of techniques. Probably the easiest
method is to change to a coordinate representation for the boson operators:
(4.489)
(4.490)
(4.491)
In this representation the Hamiltonian is written as
Ho = ! I:(P-kPk + (O~QkQ-k) (4.492)

k
H = ! I:(P-kPk + EfQkQ-k) (4.493)

k
Ek = (Ok + 4(Ok Vk
22-
(4.494)
The new eigenfrequencies are E k . Now change back to a new set of boson normal mode
operators, which are normalized to the new eigenfrequencies.
I t
Qk = J2Ek ((J(k + (J(_k) (4.495)
P k = il¥((J(k - (J(_k) (4.496)
[(J(k' (J(~] = iOk,/c' (4.497)
H = L:Ek((J(k(J(k +t) (4.498)

k
These series of steps may be summarized by the observation that the boson operators are
changed in the following way:
(4.499)
(4.500)
These transformations are useful for other problems.

The Hamiltonian of the one-dimensional electron gas (4.465) has been solved
approximately. Only the excitation spectrum has been obtained. Some of these excitations are
fluctuations in the density operator p(k). Very similar results to the Tomonaga model are
obtained by writing an equation of motion for the density operator and solving it approxi-
mately. This approach is used in Chapter 5.
So far the form of the interaction potential Vk has not been specified. In fact, physicists
choose a variety of forms for this interaction to suit their problem. The units of Vk are the
same as VF: Joule-meter (IiVF is J-m). One possible choice is to take Vk ()( e'2 = constant = Yo.
The energy spectrum is just altered by having the Fermi velocity increased:
(4.501)
(4.502)
The constant Vo is assumed to be positive, since it describes interactions between electrons.

The interactions increase the velocity of the acoustic plasmon.
Another possible choice is to take Vk = 2/3(e'2k'j.jk?"). This choice leads to long-
wavelength modes with a constant frequency, which is the plasma frequency:
Ek = Jk2V} + ro~ (4.503)
41te'2no
2
rop = 4rok V- k = - m
- (4.504)
where no = k],./31t2 is not the electron density in one dimension, but is a collection of
constants. In the electron gas, there are two different types of excitations. One is the plasma
modes at long wavelength, and the other is the electron-hole excitations at shorter wave-
length. The latter are probably best described by the choice Vk = Yo.
4.5.2. Spin Waves

The Hamiltonian (4.464) of the one-dimensional electron gas has other collective
excitations besides the density oscillations which were discussed above. These other exci-
tations have the character of spin waves, or magnons. Overhauser (1965) has shown that the
excitation spectrum is completely described by the sum of these two types of excitations:
density oscillations and spin waves. This feature of one dimension does not apply to three
dimensions. The density oscillations are the excitations which occur when there are external
perturbations such as electric fields. The spin waves respond to magnetic perturbations and
contribute to the spin susceptibility.
The spin waves are described by the operators
cr(k) = crl(k) + crz{k) (4.505)
cr l (k) = L sa;_k/2,Sap +k/2,s (4.506)

p>O,s
(4.507)
where the spin index is s = ± 1 for t, -t" The nature of the spin wave excitations is shown in
Fig. 4.14. The spin-up and spin-down densities have opposite variations, so there is no net
change in the particle density. There is a variation in Pt - Pt
(4.508)
P=Pt+ Pt (4.509)
cr=Pt- Pt (4.510)
The spin operators are examined in the same fashion used for the density operators. The
commutation relations are found among these operators and between them and the density
operators. Some typical results are
[crl (k), crl (-k')] = bk,k' (k:) (4.511)
[cr2(k), cr2( -k')] = -bk,k' (~) (4.512)
[crl(k), cr2(k')] = 0 (4.513)

[criCk), p/-k')] =0 (i, j = 1,2) (4.514)
The commutator [criCk), p/k)] contains one factor of s, and the term s = 1 cancels s = -1.
This cancellation occurs when the two spin states are occupied with equal probability and the
system is not magnetic. The spin operators commute with the density operators and so
P1 C"'-.'C7c::>. } Pt + p, = 0
p, ~ / P1 - p, -;. 0
C7~
FIGURE 4.14 Spin-up and spin-down charge densities.
describe an independent set of excitations. These excitations can be represented by a new set
of creation and destruction operators, which for k > 0 are
crl (k) = Ck/r;
crl(-k) = Ck/r;
(4.515)
cr2(k) = C~k/r;
cri -k) = C-k/r;
[Ck' 4] = bk,k' (4.516)
[Ck' bk'] = 0 (4.517)
The next step is to examine the commutation relation of crik) with the Hamiltonian (4.465),
which will establish the energy spectrum of these spin wave operators. They commute with
the second term in (4.465), from electron-electron interactions, since they commute with the
density operators. The commutator with the kinetic energy term Ho is:
[crl (k), Ho] = VF L sa;-k/2,Sap +k/2,A1P + kl21 - IP - k121] (4.518)

s,p>o
(4.519)
The commutator is evaluated using the same approximation to get (4.484). Exactly the same
result is obtained by representing the spin wave part of Ho by Lk ffikCkck' The spin wave part
of the Hamiltonian is
Hsw = L ffikCkck (4.520)

k
H = L{ffikbkbk + ffikckck + Vk(bk + b~k)(bk + b_k)} (4.521)

k
H = L{EkOCkock + ffikCkck} (4.522)

k
The density operator parts in (4.488), (4.522), and (4.498) are combined with the spin wave
parts to give the total Hamiltonian H for the excitation spectra of the one-dimensional
electron gas. The original model of Tomonaga actually described a spinless electron gas. For
spin one-half systems, the two possible spin orientations lead to another type of independent
excitation which are called spin waves. The total Hamiltonian (4.522) has the density and spin
wave excitations decoupled.
The original Hamiltonian (4.465) did not contain any terms which would cause inter-
actions between spin waves; there were no terms of the type cr(k)cr( -k). The spin wave
excitation spectrum is unchanged by electron-electron interactions, at least in the Tomonaga
model.
The spin wave part of the excitation spectrum can be used to derive the Pauli spin
susceptibility. The starting point for this calculation is (3.458):
X(k, iO)) = - J: d'te ioH (Ttcr(k, 't)cr( -k, O)} (4.523)
In the Tomonaga model, the correlation function may be evaluated exactly by using the
operator representation (4.515):
cr(-k, O) = crJ(-k) + cr2(-k) = (~)(c! + Ck) (4.524)
cr(k, 't) = (~)(C~ketOlk + Cke-tOlk) (4.525)
The 't dependence is determined by HsW" The further steps in the evaluation of the correlation
function are identical to the derivation of the unperturbed phonon Green's function in (3.76):
X(k, iO)) = (lkIL) 20)k (4.526)

1t (i0)i - O)~
lkIL) 20)k
Xret(k, 0)) = ( ----;- 0)2 _ O)~ + i20)o (4.527)
The retarded correlation function is found from the analytical continuation iO) -+ 0) + io.
The susceptibility is found to be proportional to the length L of the electron gas. This
dependence on L is correct, since the susceptibility is the total magnetization M divided by
the magnetic field, and the total magnetization is indeed proportional to the size of the system.
A more meaningful quantity would be the magnetization per unit volume, which is the above
result divided by L. The susceptibility demonstrates a resonance phenomenon, so that it is
singular whenever the external perturbations (k,O)) exactly match those of the excitation
spectrum 0) = O)k = kVF'
4.5.3. Luttinger Model

A model proposed by Lurtinger (1963) is a slight variation on the Tomonaga model. It
has the advantage of being exactly solvable, with fewer approximations, yet is identical to the
Tomonaga model in some of its essential properties. The basic feature of the Lurtinger model
is that the system has two types offermions. One has an energy spectrum given by tk = kVF'
while the other has an energy spectrum given by tk = -kVF' They are shown by the solid and
dashed lines in Fig. 4.l5(a). There is an infinite number of each kind of particle, since the
occupied energy states stretch to negative infinity.
In the Tomonaga model (4.465) it is assumed the energy spectrum is as shown in Fig.
4.15(b). The particles have a linear dispersion relation, but the same kind of particle is
represented throughout the band of states.
The two kinds offermions in the Lurtinger model are denoted by the operators aJ,k,s and
a2,k,s , where the subscript 1 or 2 designates the particle. The two bands are quite independent,
so the two fermion operators anticommute:
(4.528)
(a) (bl
FIGURE 4.15 (a) The Luttinger model has two distinct particles, with separate energy bands. (b) The Tomonaga
model has one particle, whose energy band is vFlkl.
The operators p;(k) and crik) are defined as in the Tomonaga model (p > 0):
p;(P) = L a;'k+p,sa;,k,S (4.529)

ks
p;( -p) = f a;'k,sa;,k+p,s = p;(P)t (4.530)
cr;(p) = Lsa;'k+p,sa;,k,s (4.531)

ks
cr;( -p) = L a;'k,Sa;,k+p,s = cr;(P)t (4.532)

ks
The advantage of the Luttinger model is that it has the same kind of commutation relations as
found for the Tomonaga model. However, they are valid for all p, whereas they were valid
only for p < 2kF in the Tomonaga model:
[PJ(-p), PJ(P')] = Op,p'r~) (4.533)
,
[P2(P), P2(-P)] n
= Op,p' rL) (4.534)
[PJ(P), P2(P')] =0 (4.535)
[crJ(-p), crJ(P')] = Op,p,r~) (4.536)
[cr2( -p), cr2(P')] = Op,p{p~) (4.537)
[crJ(P)' cr2(P')] =0 (4.538)

[cr;(P), Pj(P')] = 0 (4.539)
These commutation relations depend, in an important way, on the assumption that there is an
infinite number of negative-energy particles. For example, the first commutator is
(4.540)
The factor of 2 comes from the summation over the two spin configurations s = ± 1. For a
finite number of particles, each summation over particle number would just give the number
of I-particles N I ,
(4.541)
and the commutator would be zero. However, when there is an infinite number of particles in
negative-energy states, a nonzero result is obtained. For a finite band, the difference
(4.542)
equals pL/2n at the top end of the band, but it equals the negative of this at the bottom end of
the band, so that there is no net difference. For a semi-infinite band, there is no bottom
contribution, so only the top difference is counted.
The kinetic energy term in the Luttinger model is
(4.543)
H o has the exact commutation relations with the operators (P > 0)
[Ho, PI(P)] = VFPPI(P) [Ho, P2(P)] = -VFPP2(P) (4.544)

[Ho, O'I(P)] = VFPO'I(P) [Ho, 0'2(P)] = -VFP0'2(P) (4.545)
The kinetic energy term is exactly represented by the operator

nVF
Ho = L I]PI(P)PI(-p) + P2(-P)P2(P) + O'I(P)O'I(-P) + 0'2(-P)0'2(P)] (4.546)
p>o
In the Tomonaga model, the boson approximation applies only for excitation with small k.
This restriction is removed in the Luttinger model. The transformation to boson operators is
(4.547)
The Hamiltonian is now
(4.548)
The operator PI (P) for P > 0 takes a particle from state k and puts it into P + k. This
operation will make an electron-hole pair when k < kF and P + k > kF . The summation over
all such electron-hole pairs is represented by the boson creation operator blp. For particle 2,
the Fermi "surface" is at the negative wave vector -kF . Electron-hole pairs are made mostly
at negative wave vectors. The operator P2( -p) = Lk aika2,k+p for P > 0 creates these
bosons, since it takes an electron from the occupied state kF < k + P to the unoccupied state
k < -kF' where k is negative.
Various kinds of interaction terms may be added to the Luttinger model. Those which
arise from electron-electron interactions are expressed as the product of four fermion
operators, or two density operators. These Hamiltonians are exactly solvable, since they
describes linear coupling between two harmonic oscillator systems.
The Luttinger model has the advantage of being exactly solvable. Of course, one could
add other terms which might render it no longer exactly solvable. The disadvantage of the
model is that it is unphysical, since it contains the infinite reservoir of negative-energy
particles.
4.5.4. Single-Particle Properties

Some of the most interesting applications of the Tomonaga-Luttinger models are
concerned with single-particle properties of the electron gas. An important quantity is the
occupation number ni,k,s = (al.k,sai,k,s)' in the interacting system. A more ambitious calcu-
lation would be the one-particle Green's function
(4.549)
To obtain these quantities requires a representation of the single fermion operator ai,k,s in
terms of boson operators. The discussion follows Mattis and Lieb (1965) and Luther and
Peschel (1974).
The representation of the single-fermion operator in terms ofbosons is found, as always,
by examining the commutation relations. A representation of ai,k,s is satisfactory if it obeys
all the proper commutation relations with the other operators. The first step is to Fourier-
transform into a real-space representation:
1 ikx
'PisCX) = .JI ~ e ai,k,s (4.550)
\Tlt ( )
T is X = .JI "T e ai,k,s
1 " -ikx t (4.551)
The advantage of this representation becomes clear when considering the commutator of
'PiS(X) with the density operators. This discussion uses the Luttinger form of the Tomonaga
model. Typical commutators are
['Pi/X), pip)] = oijeipX'Pis(x) (4.552)
['Pis (x) , crip)] = oijseiPX'PisCX) (4.553)
which is derived in the following way:

_ 1 ikx t
['Pi/X), Pj(P)] - "LLe
VLkk's'
[ai,k,s' aj,k'+p,s,aj,k',s']
Oi,j ikx
= "L L e aj,k',s,Oss,Ok=k'+p
VLkk's'
(4.554)
The commutator ['I'iS<X). Pj(P)] has a simple form, since it is just proportional to 'l'is(x). The
solution would be simpler if the commutator were a constant or even proportional to a density
operator. It is not, so the solution of (4.552) is more complicated. One possible solution has
the form
'I' \sex)= FI (x) exp[JI (x)] (4.555)

1t 1 . .
J I(x) = - - L - {e-1PX[PI (P) + SO"l (P)] - eIPX[PI (-p) + SO"( -pm (4.556)
Lp>OP
The prefactor F I (x) can be a function of x but is a c number in the sense that it must commute
with both PI (P) and 0"1 (P). Next show that this choice does satisfy Eq. (4.552)
['I'ls. pd =FI(e"PI - ple'l) =FI(e"Ple-J, _1)e'1

(4.557)
The last line is valid only when the commutator [JI , pd is a c number which commutes with
the operator J I . It does for the J I (x) in (4.556):
(4.558)
The next observation is that the factor
(4.559)
since the factor (1 + ss') = 0 unless s = s', and then it is 2. The equations can be condensed
by introducing the notation of a spin-dependent density operator (P > 0):
PlS· (P) = L.,

'" afk
I, +p,s a·I, k,S
(4.560)
all k
PiS< -p) = L ai,k,sai,k+p,s (4.561)

aUk
(4.562)
O"i(P) = LSPiS(P) (4.563)

s
21t e- ipx
JIs<X) = - - L -PIs(P) (4.564)
L allp P
'l'1S<X) = FI (x) exp[JIs(x)] (4.565)
The spin-dependent density operators can be represented by boson operators similar to

(4.547) with an additional spin subscript.
The form of 'l'1S<x) in (4.556) is a solution to the commutator equation (4.552).
Unfortunately this solution has some undesirable properties which will force a modification.
The need for changes in 'l'is(X) may be understood by examining the form of the operator
('I't(x)'I' \sex») for the noninteracting electron system, which is the Luttinger model with just
the Hamiltonian Ho in (4.543). At zero temperature, the noninteracting system has the feature
that the momentum distributions for particles 1 and 2 have the form
nl,k,s = 9(kF - k) (4.566)

n2,k,s = 9(kF + k) (4.567)
This fact can be used to evaluate the correlation function ('P;,(x)'PIsCx')} by using the inverse
of the transformation (4.550):
('TIt ()\TI ('))

TIs X TIs X = L-1 '"
L.... e
-i(kx-k'x) ( t
alksalk's
) (4.568)
kk'
(4.569)
(4.570)
(4.571)
(4.572)
where the factors of ±ill are added to aid convergence at infinity.

The objective in choosing the representation (4.565) is to make the result for
('Pi,(x)'PIsCx')} be like (4.571) for the noninteracting Luttinger model. A method of doing
this was suggested by Luther and Peschel (1974). It uses a limiting process, where the wave
function 'P IsCX) contains a parameter IX, and the limit IX ~ 0 is taken at the end of the
calculation. Including the parameter IX, the position space operators are represented as
(4.573)
(4.574)
(4.575)
(4.576)
This new form for ~sCIX, x) may be expressed in terms of the boson operators:
(IX' x)
J Is '" e-rxP/2~1t(b
= L.... L Is,p eipx _ bt
Is,p
e- ipX ) (4.577)
p>o P
(IX x) = '"
J 2s' L.... e-rxP/2~1t(bt ipx
L 2s,-p e - b 2s,-p e- ipX ) (4.578)
p>o P
The exponential factor JI,(rx, x) has the same form as (4.556) in the limit where rx ~ 0, so the
commutator (4.552) is obeyed in this limit. The prefactor 1/.J2nrx is explained below. Now
consider the evaluation of the quantity
('Pt(x)'P Is(X' )) = 2!rx e-ikF(x-x) (e-JL'(rx,x) £!1,(rf.,X)) (4.579)
= _1_ e-ikF(X-X) rr eikx b

2nrx k>0 Vki
(exp [e -rxkl2 f2it(e -ikx bt _
Is,k Is,k
)]
ki ikx'b IS,k _e-ikx'b tIs,k )])

xexp[e-rxkI2 f2it(e
V (4.580)
The right-hand side of this expression is an average of exponential functions of boson

operators. These expressions are evaluated in Sec. 4.3.2. Each exponent is separated by using
the Feynman theorem exp(A + B) = exp(A) exp(B) exp( -[A, B]/2). Then the factors are
commuted until all the destruction operators are on the right:
('P t (x)'P (x')) = _l_ e- ikF (x-X)-<l>o(x-X)rr (exp [e-rxkI2 f2itbt (e- ikx _ e-ikx')]
Is Is 2nrx k>0 ki IS,k V
x exp[e-rxk/2 /rrbIS,k(e ikx' _ eikx )])
<Po(x) = 2n L e- rxk (1 _ eikx ) = Joo dk e-rxk (l _ eikx ) (4.582)

L k>0 k 0 k
At zero temperature, the quantity in the final brackets gives unity, which gives the following
prediction for the noninteracting electron gas:
('P t (x)'P (x')) = _1_ e-ikF(x-x')-<!>o(x-x') (4.583)

Is Is 2nrx
The expression for <Po(x) has the form of an infrared divergence as discussed in Sec. 9.3.
Expand the exponential exp(ikx) and integrate term by term. The factor exp( -rxk) ensures the
convergence of these integrals, which is the primary role played by rx:
<Po(x) = - L -;-
00 (. )1 Joo dkkl-Ie- rxk = - L 00 (. )1
IX I
1=0 I. 0 1=0 Irx
(4.584)
e-<!>O(X) = _ __ (4.585)
I - ix/rx
The series for <Po(x) is recognized as a logarithm, which gives the final result
('Pt (x)'P (x')) = _1_ e-ikF(x-x') 1 (4.586)

Is Is 2nrx 1 - i(x - x') / rx
= ~e-ikF(X-X) 1 (4.587)
2n rx - i(x - x')
Sec. 4.5 • Tomonaga Model 271
The limit IY. -+ 0 does indeed recover the noninteracting value (4.571). The parameter IY.
becomes the convergence factor 11.
It is easy to check that the factor ('Pis(x)'P 2,(,x')) is also given correctly. The repre-
sentation (4.575) for 'P1sCx) reproduces the commutation relation with the density operators
and also gives the correct ground state momentum distribution for the noninteracting system.
All these results, of course, apply in the limit where IY. -+ O. The commutator of 'P Is(X) with
Ho is also given correctly, since the latter is expressed in terms of the density operators, which
have the correct commutators. The representation (4.575) is suitable for the single-fermion
operators.
This representation can be used to calculate many interesting properties of the Luttinger
model. For example, the electron Green's function is
G1sCx -x', t) = -i(T'PIsCX, t)'Pis(x', 0))

= _i8(t)(eiHot 'P I sCx)e- iHo t'Pis (x'))
+ i8(t)('PIsCx')eiHot'P IsCx)e- iHot ) (4.588)
The correlation functions can be evaluated at zero temperature, using the same steps which
led to (4.581). The time dependence of 'P1sCx, t) for the noninteracting Hamiltonian is
(4.589)
(4.590)
The time dependence of 'P Is(X, t) = 'P IsCx - vFt) merely changes the factor x in J 1s ( IY., x) to
X - vFt. This rather trivial change makes it possible to use the previous result for
('Pis(x)'P IsCx')) to evaluate ('Ptcx')'P IsCX, t)):
(4.591)
2ni(x - x' - vFt - ilY.)
('P IsCX, t)'Pis(x')) 1

= -2
nlY.
exp[ikF(x - x') - <p~(x' - x + vFt)]
(4.592)
2ni(x - x' - vFt + ilY.)
The factor ('P IsCX, t)'Ptcx')) has just the Hermitian conjugate of <Po. The Green's function for
the noninteracting system is easily obtained (IY. -+ 0):
(4.593)
This equation can be Fourier-transformed to obtain the Green's function in the wave vector
J:
representation:
GIs(k, t) = dxe-ikxGIs(X, t) (4.594)
= -ie-iVpl(k-kp)[0(t)0(k - kF ) - 0(-t)0(kF - k) (4.595)
This result is the same G(k, t) which is obtain in the fermion representation:
GIs(k, t) = -i(Taj,k,sCt)ai,k"cO)) (4,596)
where the energy has been normalized to the Fermi energy: f.k = vF(k - kF)' The correct
result for GIs(k, t) again illustrates that the boson representation (4,575) for the single-particle
operators will faithfully reproduce the results obtained directly from the fermion repre-
sentation. The virtue of the Boson representation is that more difficult problems can be
solved. In particular, interaction terms can be added to the Hamiltonian. Exact expressions
can be found for Green's functions, or other correlation functions, although they are usually
difficult to evaluate analytically.
4.5.5. Interacting System of Spinless Fermions

An exact solution can be obtained for various correlation functions, even for the
interacting electron gas in one dimension. First solve for the occupation number. This solution
relies upon the representation of the single-particle operators which was developed in the
prior subsection. The Hamiltonian in this part is taken to be the Luttinger model for spinless
fermions (Mattis and Lieb, 1965):
(4.597)
The interaction term comes from particle-particle interactions between the two types of
fermions. Other interaction terms could be considered.
The first step in the solution is to learn the method of diagona1izing this Hamiltonian.
There are several ways to do this, and all give the same result. A canonical transformation is
used to obtain a new set of boson operators rxp, ~p, which are defined as
bl,p = ~p cosh(Ap) - rx~ sinh(Ap) (4.598)
bi,p = ~; cosh(Ap) - rxp sinh(Ap) (4,599)
b2,_p = rxp cosh(Ap) - ~; sinh(Ap) (4.600)
bt_p = rx~ cosh(Ap) - ~p sinh(Ap) (4.601)
[bl,p, bLp] = [~p, ~;] cosh2 (Ap) + [rx~, rxp] sinh2 (Ap) (4.602)
= cosh2 (Ap) - sinh2 (Ap) = 1 (4.603)
[rxp, ~;] = 0, [rxp, rx~] = 1 (4.604)

The various commutation relations are still obeyed in this new representation. The parameter
Ap is chosen so that the Hamiltonian (4.597) is diagonalized. It is first written out in terms of
the transformed operators:
H = L](~;~p + 1X;lXp){[cosh2(Ap) + Sinh2(Ap)]OOp - 2~ Sinh(Ap) cosh(Ap)}

p>O
Since these are boson operators, the ordering of terms such as IX~ = ~IX does not matter. The
zero-point motion terms are ignored. Two combinations of hyperbolic functions seem to
occur:
cosh2(Ap) + sinh2(Ap) = cosh(2Ap) (4.605)

2 sinh(Ap) cosh(Ap) = sinh(2f..p ) (4.606)
The Hamiltonian is diagonalized by setting to zero the coefficient of the term (~;IX; + IXp~p).
This step gives tanh(2Ap) = Vp/oop' so that the diagonalized Hamiltonian is
(4.607)
Ep = Joo~ - V; (4.608)
cosh(2Ap) = l
00
p
(4.609)
The transformation to the new operators is used to evaluate the properties of the interacting
system. The IXp and ~p operators refer to the actual boson normal modes in the interacting
system. The ground state of the system is the vacuum of ap and ~p particles; i.e.,
IXpIO) = 0, ~pIO) = O. These are the same set of normal modes in the Tomonaga model
(4.498).
Consider the evaluation of the fermion occupation number, which is given in (4.581) as
the ground state expectation value of the operator combination:
1 exp[-ikF(x - x') - <Po(x - x')]

('PI(x)'P1(x'» = -2 (4.610)
1ta
e-o(x) = (e-J1 (<x,x) e'1(<X'O) (4.611)
where J 1(a, x) is given in (4.577). The ground state of the system must be the particle vacuum
of the bosons with excitation energy Ep in (4.608), since these are the normal modes. The
J 1(a, x) operator must be expressed in the ap and ~p representation. The transformation
(4.600) produces a redefined operator form
J 1(a, x) = I: e-rxpj2 ~L1t{ejpX[~p cosh(Ap) - a; sinh(Ap)]

p>O VPL
- e-jPX[~; cosh(Ap) - ap sinh(Ap)]} (4.612)
It contains operators of both types ap and ~p. These operators are independent, since they
each describe an independent Boson system. Each of these boson systems are averaged
independently. The ground state average gives
<PO(X) = <Pa(x) + <Pb(x) (4.613)
e-a(x) = (e-Ja(x)e'a(x)) (4.614)
e-b(X) = (e-Jb(X)e'b(X)) (4.615)
=L P;p2~ sinhO"p)(e-iPXap _ eiPXa;)

e- rxp /2
Ja(x)
p>O yii (4.616)
J b(x) = " e- rxP /2!*1t

L..
ipXA
pL cosh(Ap)(e I-'p _ e-ipxAt)
I-'p (4.617)
p>O
The separate averages for <Pa(x) and <Pb(X) are similar to those found earlier in (4.582). The
average for <Pb(x) is identical to the earlier average for <PO (x), except for the extra kernel
cosh(Ap). The average for <Pa(x) also contains a unique kernel sinh(Ap) and has x -+ -x. By
analogy with (4.582), at zero temperature
21t e- rxp .
<Pb(x) = - L - cosh2 (Ap)(1 - e'PX ) (4.618)
L p>O P
<Pa(X) = 21t L e- rxp sinh2 (Ap)(1 _ e- ipX ) (4.619)

L p>O P
The result for <Pb(x) is manipulated by replacing cosh2 (Ap) by its equivalent 1 + sinh2 (Ap).
The term with" 1" is identical to <Po(x)
21t e- rxp .
<Pb(x) = <Po(x) + - L - sinh2 (Ap)(1 - e'PX ) (4.620)
L p>O P
so that
(4.621)
(4.622)
The effect of the interactions on the electron gas is contained in the exponential factor
exp( -<Ps)' The other terms in (4.622) are the same as for the noninteracting electron gas.
By using the relation
sinh2 (Ap) = 2"1 [cosh(2Ap) - 1] = 2"1 ~ [co - 1] (4.623)
<Ps(x) = Jo _e-rxp
dp
OO (coJ.. - )
1 [1 - cos(px)] (4.624)
P Ep
which uses the prior result (4.609).

Any evaluation of the factor cl>s(x) must assume some specific form of the potential Vp
between electrons. One possible model is to take ~ = p Vo. Vo = constant. This form of the
potential is obtained from a delta function interaction in real space. This model assumes that
the particles interact only when they directly collide. In this case (rop = pvF)
Ep =pJv} - VJ (4.625)
g == sinh2 (Ap) = ~ [2 VF
JVF- V6 - 1]
(4.626)
cl>ix) = 2g Jooo Pd
£e-rlJ'[1 - cos(px)] (4.627)
The factor Sinb2 (Ap) is a constant, which is called g. The integral for the exponential factor
cl>s(x) is now simple to evaluate, since it has the same form as earlier for cl>o(x) in (4.582):
cl>s(x) = g[cI>o(x) + cI>~(x)] = g In(1 + ~) (4.628)
The delta function model makes the following prediction for the momentum distribution of
the I-particles:
e-ikF(x-x') 1
('I't (x)'I' (x'») - - -::--::----:----:----.,.---
1 1 - 21ti(X - X + iC()[1 + (x - jC(2]g xi (4.629)
nl,k = J~oo dxeik(x-x') ('I't (x)'I' 1(x'»)

oo dx rJx(k-k
= -
J -00
F) C(2g
21ti x + iC( (x2 + C(2)g

(4.630)
Setting g = 0 recovers the noninteracting case nl,k = 0(kF - k), which is obtained by
closing the contour of integration in the UHP (upper half-plane) when kF < k and in the LHP
(lower half-plane) when kF > k. The pole at x = -iC( is circled only in the latter case.
Mattis and Lieb (1965) showed that a more interesting result is found for the case where
the coupling constant g is nonzero. Then nl,k = constant, independent of k, so the Fermi
distribution is totally destroyed. This happens even in the limit where g is infinitesimally
small. As g -+ 0, then nl,k = t.
This result is obtained by changing the integration variable to y = xjC(:
nl,k = -
Joo
-00
dy dyrx(k-kF ) 1
21ti Y + i (y2 + If
(4.631)
The only C( dependence is in the exponential factor. This exponential factor is needed for
g = 0, since it tells us whether to close the integration contour in the upper or lower half-
plane. However, for a nonzero value of g the integral converges even without the exponential
factor. Therefore set C( = 0 before doing the integral and consider
(4.632)
The right-hand side is no longer a function of k - kF' and is a constant. The integral for y > 0
is added to that for y < 0 by changing the variable y --+ -y in the latter to give the real
integral:
(_I I)
n1,k=- J0oodY 1
2ni(y2+1f\y+i+ -y+i
(4.633)
I r(! + g)
=
J ITdy (1 + y2)1+g
oo
0
I
= 2.JitT(1 + g)
(4.634)
The integral is in a standard form, which is given in tables [G&R, 3.194(3) after changing
I = x] in terms of gamma functions. In the limit where g --+ 0 then r(!) = ..,fit, r(1) = 1 so
that
·
Iun
g-+O
n, ,
k =-2
•
(4.635)
The distribution function is a nonanalytic function of the coupling constant g. The usual
noninteracting distribution function is found in the case of g = O. The introduction of an
arbitrarily weak delta function potential destroys the Fermi distribution, and each wave vector
state is occupied with an equal probability. For the case where g --+ 0, this probability
!.
approaches The g = 0 result is not obtained in the g --+ 0 limit.
This result would be difficult to prove by perturbation theory and shows the value of an
exact solution. These results pertain only to the one-dimensional electron gas. Behavior of
this type is called non-Fermi liquid behavior. Fermi liquid theory is discussed in Chapter 11.
Another name to describe such systems is Luttinger liquids.
4.6. POLARITONS
4.6.1. Semiclassical Discussion

The word polariton was coined by Hopfield (1958) to describe the normal modes in
solids which propagate as electromagnetic waves. The word is a combination of polarization
and photon, because these modes are combinations of free photons and the polarization
modes of the solid. A new word was needed, because a new view was then emerging about
the optical properties of solids. Hopfield popularized this new physics, although similar ideas
had been discussed earlier by Fano (1956, 1960) and by Born and Huang (1954).
In the old view of electromagnetic wave propagation in solids, the light shone upon the
surface of a sample and went into it. The polarization modes of the solid, e.g., TO phonons,
could absorb some of this light.
The new view is that the light and the polarization modes in the solid are coupled into a
new set of normal modes. These new modes are called polaritons. When light is shone upon
the surface, polaritons are created which propagate inward. The mathematics is trivial; since
both the photons and the polarization modes are usually described by harmonic oscillator
equations, the new modes are obtained by solving coupled harmonic oscillator equations. The
physical effect is semiclassical and need not involve quantum mechanics. The photon Green's
function f0~V<q, (0) in a system with dielectric function e~v(q, (0) was derived in (2.185). A
Sec. 4.6 • Polaritons 277
transverse wave will propagate with ~ 1- q, and the normal modes are the poles of the Green's
functions. Assuming they exist, these poles are at
(4.636)
where Et(q,CO) is the transverse dielectric function.
The normal modes are the values of co which satisfy this equation; call them Qq. For
example, consider the phenomenological equation from optical phonons in ionic crystals for
the transverse dielectric function, which is the same as the longitudinal dielectric function at
long wavelength:
(4.637)
When this form of the dielectric function is used in (4.636), there results a quadratic equation
for co2. It may be solved in order to obtain the following two solutions:
Q2q = 12 (iiiq + coLa
2 ) ± 1 [(iii + co2 )2 _ 4&2co2 ]1/2
2 q La q ro
. 2 (4.638)
- 2 (cq)
co = - -
q Eoo
These solutions are plotted in Fig. 4.16. The solid lines are the solutions to (4.638) and are the
actual normal modes in the solid. The dashed lines show the unperturbed photon mode &q
and phonon mode coro. It is these two which are coupled to form the new set of normal
modes.
The picture of energy propagation is illustrated in Fig. 4.17. It shows a slab of polar
material upon which light of frequency co is incident I from the left. Some of the light is
reflected R, while other parts may be transmitted through the slab and out the other side T.
However, inside of the slab, the normal modes which propagate are those given by Eq.
(4.638). The wave vectors of the polariton modes are determined by co = Q q and not by the
free-photon value q = (co/c),J£;;. This point is further illustrated by the observation that
there are no polariton modes of real frequency in the range COw > co > coro, as shown in Fig.
4.16. Since there are no polariton modes in this gap, no energy can be transmitted through
the slab (actually modes in the range COw > co > COTO have complex wave vector qR + iq[ =
(CO/C)JE(CO) so that energy can be transmitted through the slab of thickness d with the
probability exp( - 2dq[), which is neglected). All the energy is reflected from the slab, as
'2.. Reflectance
FIGURE 4.16 Polariton energy bands are shown at the left. There are no bands allowed for (f)TO < (fJ < (fJw, so that
the solid must have a reflectivity of unity in this interval.
T
R
FIGURE 4.17 When light has nonnal incidence I upon a slab, there are the usual reflected R and transmitted
components T. Inside the slab, the normal modes are polaritons.
shown in Fig. 4.16. The complete reflectivity for row > ro > roTO was known earlier for a
simple oscillator in a solid, but the physics is clarified when cast into the form of an energy
gap in the spectrum of normal modes.
The above equations have no damping in them, so that light is not absorbed in this
model. All the electromagnetic energy, incident upon the sample, is either reflected or
transmitted. The transmitted parts get carried through the slab by the polariton modes. The
absorption of electromagnetic energy can be introduced into the equations by introducing
damping for the phonons or additional scattering for the photons. However, if the TO-phonon
system has no damping, then it does not cause energy absorption. The polariton picture is a
rather different physical model than the older one in which the phonons caused the absorption
of energy in the solid, and this absorption in turn led to reflection.
The most direct experimental verification of the theory is to measure the polariton
dispersion curves, which has been done for the polaritons associated both with optical
phonons and excitons. The first measurement was by Henry and Hopfield (1965), who
measured the polaritons associated with the optical phonons in GaP. Their results are shown
in Fig. 4.18, where the solid line is the theoretical curve (4.638), which compares well with
their experimental points. They measured these modes by Raman scattering. The frequency of
the Raman shift gave the frequency of the excited mode, while the directional change during
the Raman scattering gives the information needed to deduce the wave vector of the polariton.
There is obviously excellent agreement between theory and experiment.
4.6.2. Phonon-Photon Coupling

The semiclassical theory will be calculated using Green's functions. The self-energy
operators for the photon, at least the parts due to the phonons, may be evaluated exactly,
which gives the semiclassical theory. An interesting aspect to this derivation is that there are
contributions from both the first- and second-order self-energy. Fortunately that is all, and the
exact self-energy is found after the first two orders of perturbation theory.
The Hamiltonian has the form
H=Ho+H' (4.639)
Ho = roTO L b~bq + L roqa~A. aqA. (4.640)

q qA.
,e"
H = - 11
L.... P j • A(Rj ) i l2 "
+ 211 L.... A(Rj ) • A(Rj ) (4.641)
lY~C j lY~C j
.~2r---r--'--------------------------------,
.050
,,,
.048
,
, I' I ,
I " I
.046
--7-,-...J-L--'---f------------- I
?!WTO I ' t
....> .044
I '
6.18·
.!! I :
I I
:3 .042 o .ilL IN 111 DIRECTION
.,,:
~o.66' h IN 100 DIRECTION
, ,
: I o
, I o POLYCRYSTALLINE SAMPLE
, : ,,
.040
I '
.038 I
,,, ,
, -+0·
I , , I
.036 I
I , ,
0 0.1 02 0.4 0.5 0.6 0.7 0.8

hcq (eV)
FIGURE 4.18 A plot of the observed energies and wave vectors of the polaritons and of the LO phonons in GaP. The
theoretical dispersion curves are shown by the solid lines. The dispersion curves for the uncoupled dispersion curve
are shown by the dashed lines. Source: Henry and Hopfield (1965) (used with pennission).
The operators bq , b~, are for phonons and aqA.' a~A. for photons. The interaction between the
photons and optical phonons is contained in the tenn H'. The ions in the solid are treated as
spherical balls with charge +e, equilibrium position Rj , and momentum Pj . The reduced mass
M is used for the relative motions of positive and negative ions in the optical mode of
oscillation. In tenns of these coordinates, the current Jj and vector potential A(R) are
expressed (see Secs. 1.1 and 2.10) in tenns of phonon and photon operators:
_ e _.
Jj-MPj-le~2NM
)1/2A t
~ibq-b_q)e
-iq'R _ 1 (roo
1-JN~J(q)e
-iq'R
1
(4.642)
(4.643)
The first self-energy is from the photon self-energy, which arises from the tenn in A2. The
interaction in the second tenn in (4.641) is
e2
HA2 = 2M LA~(q)A~(-q) (4.644)
Vo q,~
where Yo = yiN is the volume of the unit cell in a binary crystal of positive and negative ions;
i.e., Yo is the volume per phonon mode. The self-energy comes from the first-order term in the
r
expansion of the S matrix:
f0~y(k, ico) = - d'te i0l1 (T1A/k, 't)Ay(-k, O)}
= f0~~(k, ico) + ~ J~ d'tei0l1 J~ d't 1 L

2Myo 0 0 q,1)
(4.645)
x (T1A~(k, 't)AI)(q, 'tl)AI)( -q, 'tl)Ay( -k, O)} + ...
A (k , 't) = e1HA e-1H
~ ~
A~(k, 't) = e1HoA~e-1Ho
The four-operator sequence in the interaction term is equal to
(4.646)
= 2 LI) f0~oJ(k, ico)f0~~(k, ico)
The factor of 2 arises because of the two choices of pairing: k = q or k = -q. The series in
(4.645) is
f0~y(k, ico) = f0~oJ(k, ico) + n(1) L f0~~(k, ico)f0~~(k, ico) + ...

I)
(4.647)
n(l) =~
Myo
According to (2.184) the self-energy n(1)(k, ico) enters into the denominator of the photon
Green's function with an additional factor of 4rr:
(4.648)
4rrn(1) = 4rre2 = co2 (4.649)

My o p
The quantity 4rrn(1) = co~ is just a plasma frequency. It would be the longitudinal frequency
COL = COp of the ion vibrational oscillation if there were no other restoring force. Of course,
there is a restoring force, which leads to the transverse frequency COro == COo. The longitudinal
frequency is then determined by the combination of the short-range restoring force co}o and
the long-range Coulomb force co~, coio = co}o + co~. All this discussion is slightly irrelevant
because we are solving for transverse modes, not longitudinal modes. For the transverse case,
the modes are at the poles of the Green's function in (4.648). The poles are at the frequencies
co 2 = co~ + 4rrnretCk, co), which would be at c2k 2 + co~ if there were no other contributions to
the self-energy. However, there is another contribution to n(k, co) from the P . A term in H',
which cancels co~.
The A2 term in the interaction H' does not contribute any further terms to the self-energy
of the photon. The self-energy contribution n(l) provides an exact evaluation of this term in
the Hamiltonian. This exact result could have been deduced without the aid of Green's
functions. The Hamiltonian for free photons can be written in terms of harmonic oscillator
coordinates by identifying the displacement qik) = AJ.l(k)/ J41t. The Hamiltonian has the
form
HO,photon = 21 ~[P(k)' p( -k) + cokq(k)'

2
q( -k)] (4.650)
When the A2 term is written in the same notation, it is co;q(k) . q( -k)/2. Add the A2 term to
H o, and then the new potential energy is n~q(k)' q( -k)/2, where the new modes are
n~ = co~ + co;. The same result was obtained using Green's functions,
The next step is to evaluate the self-energy arising from the J . A term in the interaction
H' in (4.641). The symbol Jj denotes the current of the vibrating ion pair. The J. A term is
rewritten in wave vector space as
1
Hp . A = --L:J(k)'A(-k) (4.651)
Yo k
One factor in the self-energy expressions for the photon is the correlation function of the ion
momentum Jik) with itself. Define a correlation function
<l>J.lv(k, ico) =- J: d'te ion (T1[e1HoJJ.l(k)e-1HO]JvC -k))
=b ~cooJ~0 d'tei0l1[e1ffiO(btb
J.lV 2M k k
)+e-1ffiO (b bt )]
k k
= b ~ co~ (4,652)
J.lV M(·)2
ICOn -
2
COo
which is easily evaluated since the properties of the phonon operators are known.
The J. A interaction does not contribute any term to the first-order self-energy
expression n(1), because the first-order term has only a single factor of the operator JJ.l (k), and
the average of this operator is zero. The second-order term in the expansion for the S matrix is
which is evaluated by the usual rules to give
(4.653)
where <l>88,(k, ico) is given in (4.652). A factor of 2 comes from the two possible pairings
k = k j or k = k 2 , This self-energy expression is
(4.654)
This self-energy term is the only one from this interaction. The total self-energy is obtained by
adding the two contributions:
41tII(k, iro) = 41t[II(I) + II(2)] = ro 2 [1 + ro~ ] = (iro)2roi

P (iro)2 - ro~ (iro)2 - ro~
The photon self-energy for the Hamiltonian (4.641) has been found exactly. The energy
denominator in the photon Green's function (4.648) can now be written in terms of the
dielectric function:
(iro)2 - ro~ - 41tII(k, iro) = (iro)2E(iro) - ro~
ro2
E(iro) = 1 + 2 P 2
roo - (iro) (4.655)
. 41t(cS~v - k~kv/ k 2 )
(jfi~v(k, zro) = (.zro)2 E(zro
.)
-
2
rok
This dielectric function should be compared with (4.637) from optical phonons. The present
discussion assumes that Eoo = 1. Otherwise, a value of Eoo other than unity will modify the
vector potential with factors of Eoo. Similarly, identify the static dielectric constant as
EO = 1 + roi/ro~. Another feature of the dielectric function (4.655) is that the frequency of the
longitudinal mode oscillation is determined by the condition that E(row) = 0, which has the
solution roIa = ro~ + roi. The transverse frequency is roTa = roo. With these identifications,
the two expressions (4.637) and (4.655) for the dielectric functions are identical.
The Green's function solution of the Hamiltonian for the coupled phonon-photon modes
yields exactly the same equation ro 2E(ro) = ro~ as found from the semiclassical arguments in
Sec. 4.6.1. The two modes, photon and phonon, are coupled and form two new modes which
are polaritons. In fact, the original Hamiltonian (4.641) could be solved by a canonical
transformation to a new set of operators, say !Xk and ~k' which diagonalize the Hamiltonian.
These new creation and destruction operators are for the polariton modes. This exercise is
assigned as a problem.
4.6.3. Exciton-Photon Coupling

The polariton concept applies to the mixing of any polarization waves with the photons.
Although the phonons are an obvious example, the physics is the same for waves which are
associated with electronic polarization. Here we analyze a simple case of this phenomena. The
solid is treated as a collection of atoms which are only weakly interacting, say by van der
Waals forces. The electrons in each atom can occupy a set of discrete energy levels Em. This
model is applied to molecular solids, in which molecules are weakly bound together, and the
optical properties of the solid are often only weak perturbations on the optical properties of
the free molecules (see Davydov, 1971, or Mahan, 1975).
A light wave, which is traveling through the solid, can induce the electrons to change
their electronic state from n to m. This excitation process can really happen if the photon
energy liro matches the energy of electronic excitation liro = Em - En. However, for photon
frequencies ro which do not match any of the excitation frequencies ro =f. ro", = Em - En' real
excitations cannot occur. Then the light wave will induce a virtual transition to this excited
state: a virtual process is an excitation which starts to occur but violates energy conservation,
so the excitation process ceases. Nevertheless, the electronic system is polarized during this
virtual excitation process. The atom will have electric dipole moment 11 whose magnitude is
proportional to the amplitude of the electric field of the light which is acting upon the atom:
(4.656)
An expression for the atomic polarizability IXVJJro) will be derived below. The Hamil-
tonian for each atom of atomic number Z will be taken to have the nonrelativistic form
H="£--Ze
PT 2 Ii? I
"£-+-"£- (4.657)
i 2m i ri 2 (ij) rij
It is assumed to have solutions which are single-particle orbitals <Pn(r) with energy En- In the
Hartree-Fock approximation, the state of the atom is a Slater determinant of the occupied
orbital states. The ground state is the Slater determinant of the set of orbitals with the lowest
energy. An excited electronic state is obtained by raising an electron from one of the ground
state orbitals to another one of higher energy, which is unoccupied in the ground state. Denote
by Greek subscripts IX or ~ the possible excitation energies ro" = Em - En- Each excitation IX
corresponds to moving one electron from a ground state to an excited state orbital. The
operator b! is introduced to describe this excitation, which can be defined in terms of the
operators an of the electronic states:
(4.658)
The an operators are fermions. The b" operators are the product of two fermion operators and
have boson properties. However, they are not bosons since the number in each state is limited:
one has that (b!)N = 0, where the value N depends on the maximum number of electrons
which are in state n or may be in state m. However, the b" operators are approximated as pure
bosons. It is an adequate approximation as long as the density of such bosons is low, so that
the average number ofbosons on each atomic site is much less than unity. Then the restriction
on the site occupancy of the bosons is irrelevant, and the b" operators obey good boson
statistics. The Hamiltonian of a single atom is approximated as
H = Eg + "£ /iro"b!b" (4.659)

"
(4.660)
where Eg is the ground state energy and b!b" is the number of excitations.
The model of an atom is far too simple by today's theoretical standards. The Hartree-
Fock (HF) approximation is seldom adequate to describe an atom, and usually one has to take
a linear combination ofHF states (configurational interaction, or CI in the chemist's notation).
The single-particle orbitals <Pn(r) also vary with the excited state configuration, which
requires a more complicated Hamiltonian than (4.659) in order to describe excitations. These
complications are omitted, which hopefully are only a side issue for the discussion of exci-
tons.
An important quantity for our discussion is the dipole matrix element X(IX) for the states
<Pn and <Pm which comprise the transition
(4.661)
This matrix element will be zero in most cases, since it is nonzero only when the two
electronic orbitals <Pn and <Pm differ by one unit of angular momentum. This matrix element is
important for the case the excitations occur by the interaction with light. The rate of excitation
is determined by the dipole matrix element.
Now introduce an operator P 1.1 which is the polarization operator for the atom, where J.L
denotes vector direction and IX denotes excitation:
Pil = e LXIl(IX)(ba + b!) (4.662)

1.1
It should not be confused with the momentum, which has the same symbol. This operator has
the units of polarization, which is charge times length. One can think of the quantity
XIl(b a + b!) as a typical displacement operator, which in this case refers to the displacement
of electronic charge in the atom. It is quite analogous to the displacement operator for the
harmonic oscillator system, which is Xq(a q + a~q), where Xq = (fI/2Mro q )1/2 is the unit of
length.
The symbol !F Ilv(iro) is used to denote the correlation function of the polarization
operator P 1.1 with itself:
(4.663)
It is easily evaluated for the noninteracting Hamiltonian (4.659):
!F~~(iro) = - ; J13 d1:e iOOt LXilX)XvClX)(e-too. (bab!) + etoo• (b!b a)) (4.664)
o a
(4.665)
The quantity !F~~(iro) is actually the polarizability IXllvCiro) of the atom. To keep this
discussion simple, assume that this quantity is isotropic:
!F~~(iro) = IXllv(iro) = 0llvlX(ro) (4.666)

X 2 ro
lX(iro) = 2; L 1.1 a 2 (4.667)
a ro; - (iron)
So far the discussion has been confined to the properties of a single atom. Next consider the
additional properties which result when the atoms are collected together in the solid. The
discussion about photons is deferred to later, and now consider only the original Hamiltonian
(4.657) as applied to a collection of atoms. The electrons on one atom will interact with the
nucleus and electrons on adjacent atoms. Denote the position of an electron ri = Ri + Xi'
where R j is the position of the atom and Xi is the displacement of the electron from the center
of the atom. For two electrons on different atoms, the interaction is expanded assuming that
IRi - Rjl > IXi - x):
__ 1_= 1 + (xi-x)o(Ri-R)
Iri - rjl IRi - R) IRi - Ri
+ 12 "(x.
L- I
- x.),h
"} 11 't'IlV
(RI - R)(x.
"} I
- x.)
"} v
+ ... (4.668)
0IlV 3R IlRv
<l>llv(R) = R3 - ] i s
The first two terms can usually be neglected, since the first is a constant, and the second will
vanish for crystals with inversion symmetry. The first important term, resulting from the
interaction between electrons on different atoms, is the dipole-dipole interaction:
(4.669)
The standard practice is to retain only this term in the theory, which is called the dipole
approximation. Sometimes it is improved by also including the next terms, which involve
quadrupoles and octopoles. For a treatment of exciton theory without making a monopole
expansion but instead retaining all the terms in the electron-electron interaction, see Agra-
novitch (1960, 1961) for the formalism and Mahan (1975) for a sample calculation.
The notation needs to be expanded to include the site I of the atom in the solid. The
exciton operators are referred to as brxl and polarization operators as PIl(I). For one atom per
unit cell, each quantity may also be expanded in wave vector space:
P/l) = e LXIl(a)(brx ,/ + b!,/) (4.670)

rx
PIl(k) = ~"2( eik'RIPIl(1) = e ~XIl(a)(brx'k + b!,_k) (4.671)
- _1_" ik.R1b
b
rx,k - .fN7
e rx,/ (4.672)
In the dipole-dipole interaction (4.669), the dipole moment of the electron eXill is replaced by
the equivalent displacement operator PIl(i). The dipole-dipole interaction between excitations
on different atoms becomes:
21t
= - L P/k)TllvCk)PvC-k) (4.673)
Vo Ilvk
Tllv(k) = 4VO L eik'Rj<l>llvCR) (4.674)

1t#0
The interaction between excitations on the same atom is omitted, since this contribution is
already included implicitly in the free-atom Hamiltonian (4.657). The last step above gives
the interaction in wave vector notation. The excitations in wave vector space interact with the
lattice transform of the dipole-dipole interaction, which is called T~y(k). This function is only
needed at small wave vectors, which for cubic crystals is (Cohen and Keffer, 1955)
. _ k~ky 1
2 - 30~y + O(ka)
hm T~y{k) - - k (4.675)
k-*O
where a is a lattice constant. If one tries to evaluate this function exactly at the point k = 0, it
amounts to summing the dipole-dipole interaction over neighboring sites, which gives zero.
The function is conditionally convergent, since the limit as k goes to zero gives a different
result than at the point that k = O. The limiting value is correct.
The Hamiltonian under discussion is the original atomic part (4.659) plus the dipole-
dipole interaction term. The atomic part can also be expressed in wave vector notation, so
consider
(4.676)
This form of a Hamiltonian has been diagonalized before, see (4.438). However, the present
case is more complicated, because of the summation over the different excitations IX and ~.
Equation (4.676) is solved by the use of a Green's function. The correlation function (4.663)
is redefined to include wave vectors:
'?~y{k, iro) = - J: dte iID'" (T",Pi k , t)Py(-k, 0») (4.677)
This correlation function is evaluated by a diagrammatic expansion. The first term in the
Hamiltonian (4.676) is treated as Ho, and the second term is the interaction V. For this Ho the
unperturbed Green's function was evaluated earlier in (4.664) and is the negative of the atomic
polarizability:
ab(O) . _
.? ~y (k, lffi) - I:.2~X~Xyro",
2
_
2 -
.
-1X~y{lron) (4.678)
'" (lffin) - ro",
The self-energy term for the perturbation V occurs in first order. An important feature of the
evaluation of this term is the separability of the potential V",~,
(4.679)
so that the different IX and ~ summations just go into the factors Pik) and Py{-k). For
example, the first-order term in the expansion of the S matrix is
x (Ti~(k, t)P~,( -k', tl)PAk', tl)Py( -k, 0») + ...

= '?~~(k, iro) + 47t I: ,?~o~,(k, iro)T~,y,(k)'?~~~(k, iro) + ...
Vo ~'y'
The self-energy is 41tT~v/vo. It is the only self-energy term, and higher-order terms in the S-
matrix expansion just produce multiples of this term. The Dyson equation for the correlation
function is the matrix equation
ff ~v(k, iro) = ff&~(k, iro) + 41t L ff~~,(k, iro)T~'v,(k)ff v'v(k, iro) (4.680)
Vo ~'v'
This equation is not too hard to solve in the general case, since it is only a 3 x 3 matrix in the
direction variables (11, v) = (x, y, z). We shall only consider the simplest possible case, which
has an isotropic polarizability ff&~ = -()~vtX and the long-wavelength form of the interaction
T~v as given in (4.675). Then the equation to be solved simplifies to
_ 41ttX (k~kl.. ()~I..) ar

ff~v - -()~vtX - Vo ~ y -3 $'I..V (4.681)
At small values of wave vector the matrix ff ~v depends only on the scalar functions C and D,
ar _
$' ~v - C()~V + Dk~kv
y (4.682)
We can generate equations in only C and D by inserting the above results in (4.681) and then
equating coefficients of the terms ()~v and k~kv/ Jc2
41ttX
C=-tX+-C (4.683)
3vo
41ttX 2
D=--(C+:3D) (4.684)
Vo
These two equations for C and D can be solved algebraically:
C=- tX
1 - 41ttX/3vo
D = tX(41ttX/vo)
(4.685)
(1 - 41ttX/3vo)(1 + 81ttX/3vo)
ff - - tX - - - k~kvJ
[ () - - -41ttX
~v - 1 - 41ttX/3vo ~v 1 + 81ttX/3vo k2
Equation (4.685) is the long-wavelength form of ff ~v(k, iro).

The quantity -ff ~v is the polarizability of the solid, just as -ff&~ is the polarizability of
the atom. The interactions in V modify the atomic polarizability. These modifications are just
the local field corrections. In a cubic dielectric, the dielectric function at long wavelength has
the Lorenz-Lorentz form:
41ttX/vo
E(ro) = 1 + 1 - 41ttX/3 Vo (4.686)
The factor 41ttX/3vo in the denominator is a local field correction. The same factors are
derived in the first term of ff ~v in (4.685). The atomic polarizability tX has dimensions ofm3 ,
so the polarizability per unit volume tX/vo is dimensionless. Often this combination itself is
written as tx. In any case, the dielectric function (4.686) may be derived from the correlation
function ff ~v in (4.685).
Now solve the Hamiltonian for the interaction between excitons and photons:
H = L t 1 [
O)ka"ka"k + -2 L Pi -
e
- A(ri)
]2 + -2e L I
2 1
I (4.687)
"k m i C ij ri - rj
This Hamiltonian can be rewritten in terms of the operators for excitons and photons. The
electron-electron interaction is written in terms of the dipole--dipole approximation, as
discussed above
(4.688)
J
Pjl(rx) = -i d 3r<Pm (r)* ~ <Pn(r)
aXjl
where the operator Jjl (k) is the current operator for the exciton system. The two operators Jjl
and P jl are simply related. The classical expression has the current as the time derivative of
the polarization:
(4.689)
(4.690)
This expression is also correct for the operators, which can be derived from the relation
P = -im[r, H] between momentum and position.
The photon Green's function is now obtained by the same procedure as used in Sec.
4.6.2. The two self-energy terms are evaluated as before. The first-order self-energy, which
comes from the A2 term, is evaluated with the same result [see (4.647)]:
n(l) = ~Z 8 (4.691)
jlV myo jlV
The factor of Z is the number of electrons in each atom in unit volume Yo. It enters into the A2
term because the Hamiltonian in (4.687) contains the summation over i, which is the
summation over all electrons, i.e., the summation over each atom and then the summation
over each electron on each atom. The plasma frequency in this system is then
4nn(1) = 4n~Z/(mYo) == 0);.
The other self-energy contribution comes from the J . A term in the Hamiltonian. In
analogy with (4.652), this term in the self-energy arises from the correlation function:
n~J = <l>jlv(k, iO)) = ~J~ d'tei(J)!(T!Jik , 't)Jy{-k, 0)) (4.692)

Yo 0
This correlation function needs to be evaluated. It will be found by relating it to the corre-
lation function for polarizability ff jlV which is defined in (4.677) and solved in (4.685). This
solution is the appropriate one for our case, since ff jlV has been evaluated with the full
exciton Hamiltonian, which includes all terms in the exciton-photon Hamiltonian (4.688)
except the J . A and A2 tenns. The exact solution is found for the photon self-energy, for the
Hamiltonian (4.688), if we can relate WIlV to ff IlV. They are related by using the operator
identity (4.690),
Wllv(k, iro) = -~J~

Vo 0
dteiro~(T~ipll(k,
at
't)[-;Pv(-k, 'tf)]
at ~I=O
}
and then integrating by parts on the 't variable. The first integration by parts brings
The argument of the correlation function is rewritten as
(4.693)
and another integration by parts brings us to the expression
Wllv(k, iro) = -~([Pik,~) -

Vo
PIl(k, O)](a~~)
U~ ~I=O
}
iro
- - (PIl(k, O)[pv( -k, -~) - P v ( -k, 0)])
Vo
+ -(irof J~ d'teim (T~PIl(k, 't)Pv( -k, O)}

Vo 0
The last tenn has the desired fonn of (iro n )2 ff IlV. The first two tenns, which are constants of
integration, must be evaluated next. The second tenn is easy, since it is zero:
(PIl(k, O)[Pv(-k, -~) - Pv(-k, 0)]) = Tr{e-~H[Pik)e-~HPv(-k)e~H - PIl(k)Pv(-k)]}

= -Tr{e-~H[PIl(k),PV<-k)]} = 0 (4.694)
where the cyclic properties of the trace are used to rearrange the first tenn. The first tenn in
WIlV requires more care, since it is nonzero and makes an important contribution to the result:
The commutators of J and P can be evaluated from their definitions:
(4.695)
To evaluate the remaining summation over 0(, rewrite the momentum matrix element as
iPx/m = [x, H], so that this term is
1
-Pv(O() = icoIXXv(O() (4.696)
m
2e2 e2
R = --I:COIXX~(O()Xv(O() = --o~v I: fa (4.697)
Vo IX mvo IX
2coIXX~(0()m
fa = Ii (4.698)
The summation over oscillator strength fa can be trivially evaluated by using the fsum rule,
I:fa=Z (4.699)
IX
where Z is the number of electrons on each atom. These steps show that the first term in
(4.694) is just
e2 Z
R=--o
mvo ~v
(4.700)
rr(2) -
~v -
<l> -
~v - - 4n ~v
co~ 0 + (lCO
. )2 .'!/'
cZ
~v (4.701)
The two self-energy expressions may be added,

4n(rr(1) + rr(2)) = 4n(ico)2 ff ~v(k, ico) (4.702)
The plasma frequency cancels from the result. The photon self-energy function is just
proportional to the correlation function ff ~v of the polarization operator. Earlier it was called
the polarizability ofthe solid. It is given in (4.685). Next, solve the Dyson equation for the
photon Green's function:
1M _
.;zi ~v -
IM(O)
.;zi ~v + L..,.;zi
"IM(O)rr 1M
~A AIX.;zi IXV . (4.703)
AIX
By using the fact that ~~v and ~~j both have the factor (o~v - k~kv/k2), it is easy to show
that the term D(k~kv/k2) in ff ~v makes no contribution. The solution to (4.703) is
4n(o~v - k~kv/k2)
~ v = ---=-2--'------ (4.704)
~ (ico) I':(ico) - c2 k 2
where the dielectric function is given in (4.686). The photon Green's function is shown to be
governed by the transverse dielectric function I':(ico), which has the Lorenz-Lorentz local field
correction. The propagating normal modes are governed by the relationship co2 1':(co) = co~,
which gives the polariton modes. The quantity ~~v is the Green's function for the polariton,
since it has the correct dispersion relation. This interpretation is proper, since the function has
the true eigenfrequencies of the exciton-photon system.
Since both the exciton and photon system are bosons, the problem could have as easily
been solved from the exciton viewpoint. Then the photons merely give rise to self-energy
corrections for the exciton system. When calculating the self-energy of the exciton, with all
photon effects included, the same dispersion relation is derived for polaritons. Since the
photon and exciton systems have equal footing, the problem can be treated successfully from
either starting point. The solution from the exciton system is assigned as a problem.
Problems 291
The only real difference between the phonon and exciton solutions, in Sec. 4.6.2 and this
section, is the inclusion of the local field corrections for the excitons. They arise from the term
in the Hamiltonian describing the dipole-dipole interaction between polarization modes on
different atoms. Of course, the phonon system also has this term in the Hamiltonian, which
should be included in the analysis.
Problems
1. For the hard sphere potential, show that the diagonal reaction matrix is
(4.705)
Next show that the exact off-diagonal reaction matrix is
R (k', k) = .!!:jz(k'a) (4.706)

z 2mkllz(ka)
(Hint: Let the hard sphere be a finite step with height Vo, and solve for Vo -+ 00.) What is the T matrix
for the hard sphere?
2. Consider a one-dimensional harmonic chain of infinite length. Let one atom have a mass m different
from the others of mass M. Solve exactly for the vibrational normal modes, and obtain the condition for
the formation of a local mode of vibration. Does this happen when the impurity mass m is lighter or
heavier than M?
3. Consider the vibrational modes of a three-dimensional solid. Let all atoms be alike except for one.
Obtain the equation which detennines the frequency of the normal mode. (a) Isotope effect: let the one
atom have a different mass but the same force constants. (b) Impurity: let both mass and force constants
be different.
4. For the two-dimensional electron gas where the particles have kinetic energy Ilk = k2 /2m, derive:
• The formula for the cross section in terms of the phase shifts. (Question: what are the dimensions
of cross section in 2D?)
• The formula for the on-shell T -matrix in terms of phase shifts.
• The Friedel oscillations lin(r) in terms of phase shifts.
• The change in the ground state energy due to the impurity in terms of phase shifts.
5. Let two particles move with the same constant velocity v in a phonon field. The Hamiltonian is
(a) Show that the interaction energy due to phonon exchange is (r = RJ - R2 )
VCr, v) = -2 LM~ cos(q· r) (4.707)

q v ffiq - q'v
(b) For a Debye model with piezoelectric coupling, this expression may be approximated as
VCr, v) = -Mo JOO
-00
dq JdQq
-2
1t
4
dq ' r
1t Cs - q .v
A
(4.708)
Evaluate this integral as a function of rand v.
6. Let the following Hamiltonian describe the phonon interaction with a collection of constant velocity
particles: H = Ho + V, where
Ho = L Vj • Pj + L ooqa~aq (4.709)
j q
(4.710)
where rj and Pj are the position and momentum of the particles. Evaluate
U(t) = (eiHote-iHt) (4.711)
where the average is over the phonon thermodynamics.
7. The coherent neutron scattering from the atoms in a solid is described by the correlation function
(4.712)
For a crystal, write R/t) = R?) + Q/t), where Qj(t) is the displacement due to phonons. Find the exact
result for this correlation function.
8. Derive a result for the second moment
(4.7l3)
for the spectral function (4.283). See (4.269) for definitions of (00").
9. Show that the free energy of the Fano-Anderson model can be written exactly as F = Fo + 8F
where
~Fo = 2 Lln[l +e-~ekl + In[l + e-~e'l (4.714)
Ii
k
8F = die L 'L,(ip) (4.715)

o ip ip - Ee - 1e'L,(ip)
Evaluate this expression at zero temperature, and show that it agrees with Fumi's theorem.
10. Solve for the density of state of the Bethe lattice with z = 2: Show that it is exactly equal to the
results for a simple one-dimensional chain of sites.
11. Integrate the density of state p(Q) for the Bethe lattice and show that the total number of states in
the band is one per site, as expected.
Problems 293
12. Use perturbation theory to evaluate the Green's function for the Bethe lattice
~(1, iPn) = - I~ d'teip"(T,Cj+l('t)CJ(O)) (4.716)
(4.717)
where j + 1 is a near neighbor of j. Show that the Green's function has the expansion indicated in the
second line. Find the first three functions fi ,./3 ,Is, where z is the number of neighbors. Can these be
summed into a Dyson's type of equation?
13. Solve the Ising model for a one-dimensional chain. Start at one end of the chain, and start
averaging over spins, and derive a self-consistent equation for the partition function.
14. Solve the Ising model for the Bethe lattice for the case of antiferro-magnetic interactions (J < 0).
For h = 0 prove that it has the same transition temperature as the case of ferromagnetic ordering.
15. Solve the Tomonaga model when the interaction
U = -1 L Uk[p(k)p( -k) + cr(k)cr( -k)l (4.718)

2L k
is added to the original Hamiltonian (4.465). Diagonalize this new Hamiltonian, and use it to evaluate
the Pauli spin susceptibility (4.523).
16. Consider the static density-density correlation function
1 IL
c@=L-dxp(x)p(x +~) = 2" L cos(k~)(p(k)p( -k))
1
a L k
(a) Evaluate this for the noninteracting electron system at T = O.

(b) Use the Tomonaga model to evaluate this for the interacting electron system.
17. In the spinless Luttinger model, compare the commutator ['P 1(x), Hal as obtained in the fermion
representation and the boson representation.
18. Calculate the retarded density-density correlation function for the spinless Luttinger model:
X(p, ioo) = I~ d'teion (T,p(P, 't)p( -P, 0)) (4.719)
pep) = Pl(P) + P2(P) (4.720)
19. Interesting effects happen to the susceptibilities X(q, (0) when q ~ 2kF . This wave vector region
may be evaluated in the Luttinger model by taking operators which are the product of partic1e-l and-2
fermions. For example, evaluate the correlation function X(q, (0) in the noninteracting spinless Luttinger
model, where
A(q) = I
dxe-iQX'Pi,(x)'P lsCx) (4.721)
x(q, t) = (A(q, t)At(q, 0)) (4.722)

20. Find the eigenvalue equation of the boson modes of a one-dimensional electron gas described by
the spinless Luttinger model, which interacts with phonons. The ds are phonon operators:
(4.723)
21. Write the Hamiltonian (4.641) in terms of harmonic oscillator coordinates for both the photons and
the phonons. Solve it as a coupled oscillator problem.
22. Solve the polariton Hamiltonian (4.688) from the exciton viewpoint. Calculate the self-energies of
the exciton, including those arising from the interaction with photons. You may simplify the problem by
neglecting the local field term. Your answer should have the poles of the exciton Green's function at the
polariton modes.
Chapter 5
Homogeneous Electron Gas
The use of diagram techniques in many-particle physics began in the early 1950s, soon after
their introduction into field theory. Although these methods were applied to a variety of
problems, some areas of work were more successful than others. The two areas which enjoyed
early success were the homogeneous electron gas and the polaron problem. Later there were
other successes such as the theories of superconductivity and superfluidity. However, the
theory of the homogeneous electron gas, as it was initially understood, was worked out by
many contributors during the period 1957-1958. They brought a variety of theoretical
approaches to this problem, but all used diagrammatic techniques in some form. On the other
hand, during the past twenty years, there has been achieved an understanding of electron-
electron interactions in strongly correlated metals. This latter topic is covered in the next
chapter.
5.1. EXCHANGE AND CORRELATION
The homogeneous electron gas is described by the Hamiltonian
t I t
H = I>,pCpcrCpcr + -2 L L VqCk+qcrCk'_q,cr,Ck,cr,CkO"
pcr V kk' crcr' q,oO
p2
cp = 2m (5.1)
41t~
Vq = - 2 - (5.2)
q
which was derived in (1.164). The electrons are free particles, which mutually interact by
Coulomb's law e2 /r. There are Ne electrons in a large volume v, with an average density
no = Ne/v. A positive charge of density no is spread uniformly through the volume v. The
positive background maintains the overall charge neutrality of the system. The homogeneous
electron gas is also called the jellium model of a solid.
There are two electronic properties which will be evaluated in this section. First is the
self-energy of an electron of momentum p. Second, the properties of all the electrons will be
averaged to obtain the total energy of the system. This average is done at zero temperature,
which derives the ground state energy. The important quantity is the ground state energy per
295
296 Chap. 5 • Homogeneous Electron Gas
particle Eg , which can depend only on the particle density Eino). The total energy of the Ne -
particle system is just NeEg = E T , since surface effects are ignored.
The parameter rs is universally used to describe the density of an electron gas,
41tnoa~ ~ = 1 (5.3)
3 s
where ao is the Bohr radius. In an electron gas with uniform density no. rs is the radius in
atomic units of the sphere which encloses one unit of electron charge. Thus rs is small for a
high-density electron gas and it is large for a low-density gas. Other properties of an electron
gas may be expressed in terms of this parameter. The density may be related to the Fermi
wave vector,
no=2 J(21t)
d3p
- - 3 np
1
=2"
1t
Jk
0
F 2
pdp=-32
1t
kj.. (5.4)
so that the Fermi wave vector and energy are related to rs '
_ ( 91t)1/3 ..!.. _ 1.9192

kFaO -_ (31t2no )1/3 ao -_ ( 91t ) 1/3(41tnoao3)1/3- -
4 3 4 rs rs
(5.5)
E =
F
1i2k~ =
2m
(kFa
0
i (~)
2ma2
= 3.6832 E
r2 ry
o s
where Ery = 13.60 eV will be the standard unit of energy. Similarly, the plasma frequency is
(5.6)
In the homogeneous electron gas, the average kinetic energy of the electrons is going to be
proportional to EF ~ (K.E.) ~ k~, which, by dimensional analysis, is inversely proportional
to the square of the characteristic length of the system, which is rs. Therefore (K.E.) ()( 1/";.
Similarly, dimensional analysis suggests that the average Coulomb energy per particle will be
e'2 divided by the characteristic length, or (P.E.) ()( l/rs' When the electron gas has suffi-
ciently high density, which is small rs , the kinetic energy term will be larger than the potential
energy term. In this case, the electrons will behave as free particles, since the potential energy
is a perturbation on the dominant kinetic energy. In the high-density limit, the free-particle
picture is expected to be valid. This limiting case will be investigated below. The kinetic
energy and potential energy contributions will be calculated. The potential energy terms
cannot be found exactly, but the result can be expressed as a power series-with logarithm
terms-in the parameter rs' This series should be accurate at small values of rs' A term-by-
term derivation ofthis series is derived below. Later, in Sec. 5.2, the other limit oflow density
is considered where the potential energy is larger than the kinetic energy.
Sec. 5.1 • Exchange and Correlation 297
5.1.1. Kinetic Energy

The first energy term is the kinetic energy. For a single particle it is £k = fi1 12m. The
contribution to the ground state energy is obtained by summing over all the particles in the
ground state:
(5.7)
(5.8)
The average kinetic energy is ~EF' which is given in terms of rs.
5.1.2. Hartree
All the remaining terms in the energy come from the Coulomb interaction between the
particles. This contribution has not been evaluated exactly. Instead, approximate expressions
are obtained by a variety of means. Our derivation follows that of Gell-Mann and Brueckner
(1957). They start by examining the terms generated by ordinary perturbation theory. The first
term which occurs is the Coulomb interaction between the electrons and the uniform positive
background, which is called the Hartree interaction. In the model of the homogeneous
electron gas, the time-averaged electron density is uniform throughout the system, as is the
positive background. These equal and opposite charge densities exactly cancel, so that the net
system is charge neutral. The Hartree energy is zero. That is, this energy is given by the
equation
(5.9)
but the ion and electron particle densities are Pe = Pi = no, and the contribution is zero. This
fact has already been used in writing Eq. (5.1) by the omission of the q = 0 term from the
Coulomb interaction. The q = 0 term is the direct Coulomb interaction among the electrons.
This term is omitted because it is canceled by the direct interaction with the positive back-
ground.
5.1.3. Exchange
The Coulomb interactions in (5.1) provide other energy contributions in addition to the
direct term. The Hartree term corresponds to the following pairing of the operators in (5.1)
when the average value of H is evaluated:
Another way to pair the same operators is
This arrangement requires that cr' = (J and k' = k + q. The minus sign is from exchanging
the order of the fermion operators. This term is called the exchange energy, or the Fock
energy. Retaining both terms is called Hartree-Fock. The exchange term was derived
previously in (2.138). It gives a contribution to the energy of an individual electron, as well as
a contribution to the ground state energy of the collection of electrons:
(S.11 )
(S.12)
The self-energy Lx(k) depends only on the magnitude of the wave vector k of the particle and
not upon its energy variable ipn or w. The wave vector integrals are elementary (p = k + q,
v = cos 9):
L (k) - -
x -
I d 3p 4n~
(2n)3 Ip _ kl 2
n
P
(S.13)
-n~ fk p 2dp II dv
F
(S.14)
= 0 -1 k 2 + p2 - 2pkv
= _ ~ IkF PdPlnlk+pl (S.lS)

nk k-p
-llnl +YI)
0
LxCk) =- ~kF
n
(1 + 1
2y
1
1- Y
(S.16)
k
y=- (S.17)
kF
A particle at the Fermi energy k = kF has y = 1 and
L (kF) = _ e2kF (S.18)

x n
It is convenient to write
L (k) = e2 kF S(y) (S.19)

x n
1-I
S(y)=- ( 1 + 11 - -
--ln
2y l-y
+YI) (S.20)
where S(y) is a function which gives the wave vector dependence of the exchange energy.
This function is shown in Fig. S.l. Its value at y = 0 is S = -2. In the vicinity of y = I it
rises steeply and approaches zero at large values of y. At the Fermi energy the value is
FIGURE 5.1 The unscreened exchange energy of the electron gas.
S( 1) = -1. This point is interesting because the function S(y) has an infinite slope. The
derivative of S(y) is
!!...S(y)
dy
=~(1
2y y
+1 Inl 1+yl-2)
1- Y
(5.21)
The derivative has a logarithmic divergence as y ---* 1. This fact is interesting because it
predicts that the effective mass is zero. The effective mass of a particle was defined in (3.160).
Since the exchange self-energy is not frequency dependent, the effective mass is given by
(:*) = 1 + a: ~x(k) k
(5.22)
= e'2m !!...S(y) (5.23)

rck dy
= e'2m ~
2rckF jl
(1 +1 lnl~l- 2)
y 1- Y
(5.24)
which diverges at the Fermi energy y ---* 1. If the inverse effective mass really diverged at the
Fermi surface, it would have several observable consequences. The electron gas would be
unstable at low temperatures, and the specific heat would diverge. Many metals have unusual
properties at low temperatures, e.g., some become superconducting, while others become
magnetic. However, simple metals such as the alkalis are stable at low temperatures, so that
this exchange instability is regarded as being absent. In fact, an examination of further terms
in the perturbation theory produces another divergence in the effective mass which exactly
cancels the one due to exchange. The effective mass and specific heat are not divergent. This
subject is pursued further in Sec. 5.8.
The exchange energy contribution to the ground state energy is obtained from (5.12).
Summing over spins gives an expression which is easy to evaluate
(5.25)
The average exchange energy per electron is ~ of the value at the Fermi energy. The additional
factor of! enters the ground state energy to account for the fact that the exchange energy is a
pair interaction. In tenus of the parameter rs the total ground state exchange energy per
electron is
E = -~(k a )(~) = _ 0.9163 (5.26)

~ 2n F 0 2ao rs
where the earlier result for (kFaO) was used in the final expression.
So far two tenus have been found for the energy of the particle,
h 2k 2
E(k) = 2m + Lx(k) + ... (5.27)
The corresponding two tenus for the ground state energy per particle are,
2.2099 0.9163
E = - - - - - + ... (5.28)
g r; rs
The ground state energy has the appearance of a power series, in increasing powers of rs.
Although it is usually unsafe to extrapolate from just two tenus, in fact Eg is a series in rs' The
next tenu will be of order O(r~) . The zeroth power could be interpreted as either a constant or
as In(rs). In fact, both of these tenus are present. The series has the fonu
2.2099 0.9163
Eg = - - 2- - --- 0.094 + 0.0622 In(rs) + ... (5.29)
rs rs
where the last two tenus were first given by Gell-Mann and Brueckner (1957). This result has
since been obtained by a variety of perturbation techniques. The Gell-Mann and Brueckner
derivation was by Rayleigh-Schr6dinger perturbation theory. Alternate derivations have been
presented by Sawada et al. (1957), Hubbard (1957), Nozieres and Pines (1958), and Quinn
and Ferrell (1958). Our derivation follows Quinn and Ferrell.
The above energy tenus comprise the Hartree-Fock theory. It is defined to be the kinetic
energy, the Hartree energy which is zero, and the exchange energy. Wigner and Seitz (1933,
1934) suggested the name correlation energy to mean the energy tenus beyond Hartree-Fock.
The name is applied both to the additional energy tenus in the self-energy of an electron of
wave vector k, and to the ground state energy obtained by averaging over all of the electrons.
The total ground state energy per particle is written as
(5.30)
where the correlation energy Ee needs to be detenuined. Of course, some tenus are known for
its power series in rs:
Ee = -0.094 + 0.06221n(rs) + O(rs) (5.31)
This result is accurate in the limit of rs ~ O. There is some uncertainty regarding the
maximum value of rs for which these few tenus provide an accurate description. The radius of
convergence of the power series is about rs :::s 1. Actual metals have values of rs up to about 6,
and this series does not give sensible numbers at these low densities.
The tenu correlation energy is often applied to other quantities besides the total ground
state energy. For example, the correlation energy of a particle of wave vector k are those tenus
beyond Hartree-Fock:
(5.32)
The self-energy from correlation LcCk, ikn) depends upon the particle energy ikn' The energy
and wave vector can be averaged to obtain the contribution Ec to the ground state correlation
energy.
5.1.4. Seitz's Theorem

The derivation of Quinn and Ferrell (1958) does calculate the correlation energy per
particle. They then average over electron states to obtain the correlation contribution to the
ground state energy. This averaging is based on a theorem of Seitz (1940) which relates the
ground state energy to the chemical potential. The chemical potential is defined as the energy
it takes to add or remove an electron from the material. It is the energy which divides the
empty from the occupied states at zero temperature. Of course, it is just the Fermi energy of
the metal. The chemical potential is the energy of an electron of momentum kF :
(5.33)
where ikn = 0 is the chemical potential. The chemical potential j.l is only a function of the
density of the electron no. The theorem of Seitz is that
(5.34)
The proof is based on the definition of the chemical potential. It is the energy difference
between a system with Ne particles and one with Ne + 1:
(5.35)
For an Ne-particle system, the total energy is ET = NeEg. For fixed volume v, an (Ne + 1)-
particle system has a density (l/v)(Ne + 1) = no + l/v. The total energy of the (Ne + 1)-
particle system is
(5.36)
The difference between this expression and NeEg is the chemical potential and gives the
assertion in (5.34). In proving this theorem, the volume v is kept fixed, as is the amount of
positive charge. The (Ne + 1)-particle system has a slight charge imbalance, since it has one
more unit of electron charge than positive charge. However, the extra charge is a negligible
contribution to the energy. For example, if a body of average dimension L is uniformly
charged with one unit of charge, then its Coulomb energy is of order fl- / L. But L for a
macroscopic body is L ~ 1 cm, and e2 / L is 10-7 eV, which is negligible. For very small
particles, where the dimension L is nanometers, the charging energy e2 / L is important.
The chemical potential is the negative of the work function. It is the energy required to
remove an electron from the solid and take it to infinity with zero kinetic energy. If this
vacuum level is called zero, then the measured work function just gives the negative of the
value of the chemical potential. However, there is a surface correction to the work function, or
chemical potential, but not to the volume part of the ground state energy per particle. That is,
write the total energy as
(5.37)
where A is the total surface area and Es is the energy per unit surface area. For macroscopic
bodies, Eg does not depend on the surface area. However, J..L does have a term which depends
on the surface--actually on the surface dipole layer. Write it as J..L = J..LB + L\J..L, where L\J..L is the
contribution from the surface dipole layer and J..LB is from the bulk: terms. The surface
contribution L\J..L to the work function was first recognized by Bardeen (1936) and calculated
accurately by Lang and Kohn (1971). The theorem of Seitz actually just refers to J..LB'
The relationship (5.34) is valid term by term in the energy series. The Hartree-Fock
terms may be expressed in powers of no:
E - 2.2099 _ 0.9163 _ lXn2/ 3 _ 1 Yn 1/ 3

g,HF - r2 r - 5 0 4 0
s s
3Xi 2/3 _ 2.2099 _ 3 EF _ 3 (3 2 3)2/3
"5 no - - - 2- - - - -"5 1t noao (5.38)
rs 5Ery
3 Yi 1/3 _ 0.9163 _ 3 (3 2 3)1/3
'I no - - - - -2 1t noao
rs 1t
Applying the theorem for the chemical potential gives
d
J..LB,HF = dno (nOEg,HF) =Xno - Yn o = EF -
2/3 1/3
ni2 kF (5.39)
The theorem correctly gives that the average kinetic energy is ~ of the value at the Fermi
i
energy, and the average exchange energy is the value at the Fermi energy. These results
follow from their respective powers of no: ~ and l.
The correlation energy for the ground state is a series in rs or n~/3. The theorem may be
applied to this series on a term-by-term fashion:
=A--Bln - -3)
3
1
3
(4 1tnoao (3
+C - -3)1/3 + ...
41tn oa0
(5.40)
d
J..Lc = -d
no
(noEc) = A - 3B + B In(rs) + 3Crs + ...
1 2
(5.41)
Quinn and Ferrell used this theorem to make a backward deduction. They first found J..Lc
by calculating the self-energy terms of an electron at the Fermi energy. Then the above
relationships may be used to find the correlation energy for the ground state energy. The
logarithm term has the same coefficient B for the two energy terms. Once this term is
determined, the constant term for the ground state energy is obtained by adding B /3 to the
constant term found for the chemical potential.
The self-energy of an electron, from Coulomb interactions, may be calculated by the
methods described in Chapters 2 and 3. The self-energy function has an infinite number of
terms. The basic procedure is to start examining some low-order terms and to deduce which
terms contribute to the constant and In(rs) terms. The term low-order refers to the order of the
self-energy diagram, which is the number of internal Coulomb lines.
5.1.5. ~(2a)
The exchange energy graph is shown as the first diagram in Fig. 5.2. It is the only
contribution with one Coulomb line. The correlation energy is the sum of all contributions
with two or more Coulomb lines. There are three diagrams, or self-energy terms, with two
Coulomb lines. One of these is shown in Fig. 5.2(a). By using the rules for constructing
diagrams in Sec. 3.4, this self-energy contribution is
1
:E(2a)(k) = -2 L VqVq' L f§(O)(k + q)f§(O)(k + q')f§(O)(k + q + q')
V2~ qq' iq.iq.'
The first summation over frequencies is given in (3.217):
~L f§(O)(k + q)f§(O)(k + q + q') = n:(~k+q) - nF(~k+q+q') (5.42)

~ iq. zqn' + ~k+q - ~k+q+q'
The second summation requires a new derivation, since one energy denominator has odd
multiples of i7tj~, and the other has even multiples. The answer is
~ L _._f§--,(_O)..:...(k_+_q::....:')_ _ nB(~k+q+q' - ~k+q) + nF(~k+q')

~ iq.' lqn' + ~k+q - ~k+q+q' ikn + ~k+q+q' - ~k+q - ~k+q'
where one occupation function is a boson distribution and the other fermion. The boson
distribution may be eliminated by using the identity
[nF(~k+q) - nF(~k+q+q' )]nB(~k+q+q' - ~k+q) = nF(~k+q+q,)[1 - nF(~k+q)]
which gives the final self-energy term:
:E(2a)(k) = _ ~ L . VqVq'
v qq' lkn + ~k+q+q' - ~k+q' - ~k+q
X {nF(~k+q')[nF(~k+q) - nF~k+q+q')]
+ nF(~k+k+q,)[1 - nF(~k+q)]} (5.43)
The self-energy of a particle is needed on the Fermi surface. Set k = kF and ikn =
~kF = kj.j2m - Il = O. The terms in the energy denominator largely cancel,
q.q'
~k + ~k+q+q' - ~k+q - ~k+q' = --;;;- (5.44)
(bl
£l (e I
FIGURE 5.2
The self-energy is
F (21t)6
J
'L(2a)(k ,0) = - (41teim Jd3q d3q'_1_
q2 (q'i q . q'
x {nF(~k+q,)[nF(~k+q) - nF(~k+q+q'] + ... } (S.4S)
This integral is evaluated by using dimensionless units. All wave vectors are normalized to the
Fermi wave vector x = q/kF, Y= q'/kF, Z = kF/kF:
3
-:a- {nF(Z + y)[nF(Z + x) - nF(z + x + y)]
3y 1
'L(2a) = - 2E~ Jd
-2 XJd
1t X Y X'y
+ nF(Z + x + y)[l - nF(Z + x)]} (S.46)
The quantity on the right is independent of electron density. The integral is convergent and
gives a nonzero result. It contributes to the constant term A in (S.41). Since Seitz's theorem is
true for each term, this integral must equal the same constant as found for the equivalent term
for Ee' The similar contribution to Ee was evaluated by Onsager et al. (1966) and found to be
i
'L(2a) = ln(2) - (3/21t2)1;(3) = 0.0436.
5.1.6. 1:(2b)
The second self-energy term involving two Coulomb lines is shown in Fig. S.2(b). This
contribution is omitted because it is part of a sum of diagrams which gives zero. Consider the
summation shown in Fig. S.3(a). Each term has one more exchange diagram. All terms may
be summed by evaluating the exchange energy with an electron Green's function in the self-
energy which includes the exchange energy. This summation is given by the self-energy
(S.47)
. 1
~(k + q, lkn) = 1'kn _ ~ k+q _ 'Lx (k + q) (S.48)
where the Green's function has a self-energy due to exchange. Since the self-energy 'Lx(k)
does not depend on frequency, the frequency summation of the Green's function yields the
simple number operator as in (3.219),
1
~ f ~(k + q, ikn ) = nF[~k+q + 'Lx(k + q)] =
1
eP(~+l:) +1
,.) j} --1]) § + +
'''£L+ d +6+··FIGURE 5.3

and the self-energy is
(5.49)
At zero temperature the electron distribution function nF[~p + ~x(p)] is just a step function
which equals unity for electrons beneath the chemical potential and zero for those above. This
step function must also be normalized so that the electron density is still no, as in
(5.50)
The effect of the exchange energy ~x(P) in the argument of nF is to change the chemical
potential. However, the Fermi wave vector kF is the same, even after the exchange energy has
been included in the occupation. function. The Fermi wave vector determines the density no,
which is unchanged by the interactions. The addition of the exchange energy must be
canceled by an equal change in chemical potential. The result is that at zero temperature,
(5.51)
At zero temperature the Fermi distribution is still a step function at k = kF . The exchange
energy ~~(k) in (5.49) is exactly equal to the one calculated earlier. In Fig. 5.3(a), the
summation of all terms just yields the value of the first term alone. All of the subsequent
terms in that series sum to zero, because of a shift in the chemical potential.
The evaluation of these self-energy terms has become more subtle. It is not just a
straightforward evaluation of terms in a series. One must understand each term in order to
gauge its actual contribution. Further examples of this behavior are found in the other second-
order term.
The exchange self-energy has a remarkable effect upon the zero temperature electron
distribution. It leaves kF unchanged. This simple result is a consequence of the feature that
~X<k) is independent of frequency. Most self-energy terms depend on frequency and so have a
larger effect upon the wave vector distributions: recall Problem 4 in Chapter 3.
5.1.7. ~(2c)
The third self-energy with two Coulomb lines is shown in Fig. 5.2(c). This self-energy
diagram may be written as
(5.52)
The closed fermion loop gives the polarization diagram p(1)(q, iqm)' Although one could
proceed with the evaluation of this self-energy, it obviously has one drawback. The wave
vector integral appears to diverge at small values of q, when v ~ 00, because of the factor v~
Jd3q4q => J°dqq2 ~ 00 (5.53)
Indeed it does diverge, so this self-energy term is infinite. The divergence is removed by
summing a series of self-energy diagrams. This series is shown in Fig. 5.3(b). Each term has
one more polarization bubble and one more Coulomb line-hence an additional factor of
vqp(1). The summation of these tenns gives the simple series
L(3b)(k) = - ~ LV ,§(O)(k + q)
v~ q q
x {Vqp(1)(q) + [Vqp(1)(q)]2 + [v qp(1)(q)]3 + ... }

_ 1 (0) vqp(1)(q)
- - v~ ~ vq'§ (k + q) 1 _ vqp(1)(q) (5.54)
The summation oftenns shown in Fig. 5.3 is called the random phase approximation (RPA).
The "approximation" in RPA is that this series oftenns is used to represent the entire answer.
For example, the RPA result for the electron correlation energy uses Fig. 5.3(b) or (5.54). RPA
ignores other tenns, such as (5.46), which was found from the diagram in Fig. 5.2(a). RPA
ignores many more tenns in each higher order of perturbation series.
The denominator in (5.54) is a very important quantity because it is very often used in
calculations. It is the RPA approximation to the dielectric function and is defined as
(5.55)
The self-energy series may be written as
(5.56)
The "-1" tenn in the brackets is easy to evaluate. It is
(5.57)
which is just the negative of the exchange energy. Adding the exchange energy to the above
result gives a tenn containing the RPA dielectric function:
L () = L(3b)(k) + L (k) = _ ~ " ,§(O)(k + q) (5.58)

RPA q x v~ "q Vq tRPA(q)
The summation of these two tenns is defined as the RPA self-energy. It is frequently used to
evaluate the self-energy of the electron (see Lundqvist, 1969). It will be discussed in Sec. 5.8.
The exchange energy is logically included as the first tenn in the series shown in Fig. 5.3(b).
The RPA self-energy, defined above, is also called the screened exchange energy. This self-
energy has one feature which is worth mentioning now, although it will be shown later in Sec.
5.8: the inverse effective mass no longer diverges at the Fenni surface. A detailed evaluation
of this self-energy shows that the coefficient B = (2/n 2 )(1 - In 2) = 0.0622.
5.1.8. High-Density limit

The first two tenns in the series for the ground state energy in the RPA are
Ec,RPA = -0.142 + 0.06221nrs + O(rs' rs In(rs)) (5.59)

Adding the Onsager result, for the second-order diagram ~(2a) [see (5.46)], gives the result of
Gell-Mann and Brueckner:
Ec = -0.094 + 0.0622 In rs + ... + (5.60)
Another term in the series has been obtained by Carr and Maradudin (1964):
Ec = -0.094 + 0.06221nrs + 0.018rs Inrs + ars + 0(";) (5.61)
They were unable to obtain all the terms contributing to the coefficient a of the linear term in
rs. They obtained all its contributions except one, and they were able to put limits on its value.
Carr and Maradudin also discussed the convergence of this series at large values of rs.
Because of the term In rs, this series is not analytic as rs --+ 0, which is the limit of very high
densities. But the series is accurate in this limit. Metallic densities are roughly 1.8 < rs < 6. It
is an important question whether the formula (5.61) is valid for these values of rs. Figure 5.4
shows a plot of the correlation energy found by Carr and Maradudin. Only the terms shown in
(5.61) have been plotted (a = 0). It diverges at small rs, which is just the effect of the In(rs)
term. It is negative up to values about rs ~ 2.5, where it crosses the axis and becomes
positive. Now there are simple arguments which indicate that the correlation energy must be
negative. These arguments are given below in great detail. They show that the perturbation
formula for Ec is not valid in the range of metallic densities.
Large values of rs correspond to the low-density electron gas. Wigner (1934) showed
that the low-density electron gas behaved very differently from the high-density electron gas.
At low density, the electrons become localized, and the calculation of the correlation energy
becomes quite different. The low-density limit ofWigner is discussed in the next section. The
correlation energy is determined for all values of rs by extrapolation procedures between the
low- and high-density limits. These extrapolation procedures provide a formula for the
correlation energy which is valid in the region of metallic densities.
0.05
iii
C)
0
a::
w
ID
C
>-
a:: 0.05
>-
C)
a::
w
z
w -010
Z
0
fi..J
w
a:: -0.15
a::
0
(,)
-0.20
FIGURE 5.4 The correlation energy of the electron gas as given by the expansion of Carr and Maradudin.
5.1.9. Pair Distribution Function

The correlation energy is the improvement in the ground state energy beyond the
Hartree-Fock approximation. The correlation energy may be understood, at least concep-
tually, only after understanding Hartree-Fock. It is useful to examine an important property of
the Hartree-Fock theory: the pair distribution function g(r), which was introduced in Sec.
1.6. The pair distribution function g(r) is the probability per unit volume that an electron is at
r if there is already one at r = O.1t depends on spin, so there are two different pair distribution
functions in an unmagnetized electron gas: gtt(r) = gj.,j.(r) and gt,j.(r) = g,j.t(r). In the
notation gss'(r) the first spin index s is for the central electron, while the second Sf is for the
electrons at r. Thus g,j.t(r) is the probability that a spin-up electron is at r if a spin-down
electron is at r = O. Of course, the electrons are not fixed, and they are usually moving
rapidly. These pair distribution functions are averages for the moving particles. Even the
electron at r = 0 is not fixed, so that this reference point moves with the electron.
The pair distribution functions are rather easy to calculate in the Hartree-Fock
approximation. Here the N -particle wave function is a Slater determinant,
<Pl.! (rl) <Pl.! (r2) <Pl.! (rN)
1
<pA.2(rl) <PA.2 (r2) <pA.2(rN)
q\"'z,,·A.N(rl ... rN) = ,.fNf (5.62)
<pA.N(rl) <pA.N(r2) <pA.N(rN)
where the Aj are the quantum numbers which describe the states, and there is one wave
function for every occupied electron state. The square of this many-particle wave function is
the N -particle density matrix. The pair distribution function is given by the two-particle
density matrix. It is obtained by integrating the N-particle density over all but two space
coordinates. A similar expectation value is taken over the spin variables:
(5.63)
If the one-electron orbitals <PI. (r) are assumed orthogonal, then this integration just yields the
sum over all possible pair w~ve functions:
(5.64)
The sum over Ai' Aj is over all occupied states, so each pair is summed twice. For the
homogeneous electron gas, the orbitals must describe plane waves,
(5.65)
where the Xs are the spin functions. The pair distribution function just describes the orbital
part, so that one averages over spin functions. These spin averages are (Xt Xt) = 1,
(Xt X{) = 0, etc. The two pair distribution functions are
gH (rj - r2) = N(NI _ 1) L (Ieikl . fl +ik2 • f'I 2 + leikl . f,+ik 2 ' f112)
klk,
= N(N2_ 1) (~y =! (5.66)
gtt(r j - r2) = N(N1_ 1) L leikl 'fl+ik, 'f, - eikl 'f,+ik, ' f1 12

klk,
2 L [1 _ ei(kl -k')'(fl -f2 )]
N(N -lhlk,
= ! [1 - A(r j - r2)2] (5.67)
A(r) = ~ Leik ' f (5.68)

Nk
The summation over k runs only over occupied states. The antiparallel spin distribution
function gH = gH = ! in the Hartree-Fock approximation. The cross term which results
from the determinant is zero because of the orthogonality of the two spin functions. There is
no correlation in the position of electrons of opposite spin. The parallel spin gtt = gH has a
definite spatial dependence, which comes from the cross term which is retained since the
spins are parallel and the spin averages are unity. The function A(r) is calculated to be
A(r) = 2 d3kJ
- --3nke 3 JkF kdksin(kr)
ik'f =-3
no (2n:) rkF 0
3 .
= -klI(rkF ) (5.69)
r F
where j j (x) is a spherical Bessel function. The two pair distribution functions are shown as the
solid lines in Fig. 5.5. The parallel spin gtt(r) vanishes at r = 0 and approaches! at large
distances. The pair distribution function gtt(r) must vanish at r = 0 as a consequence of the
exclusion principle for fermions. This feature is built into the many-particle wave function
(5.62): if any pair of orbitals have the same spin and position, these two rows of the deter-
minant are identical, which makes the determinant vanish.
~"
1/2t----~ _ _-
FIGURE 5.5 The pair distribution functions for parallel (left) and antiparallel (right) spins. The solid liine is the
Hartree-Fock approximation, while the dashed line includes correlation.
The total pair distribution function for a particle is the combination of the result for
parallel and antiparallel spin distributions:
(5.70)
In a nonmagnetic system, the same result is obtained if the central particle has spin down. The
pair distribution function has two features which are important for our discussion. First, there
is the normalization integral:
J
no d 3 r[g(r) - 1] = -1 (5.71)
The second is a statement about the ground state energy of the electron gas. It has kinetic
energy plus Coulomb parts. The Coulomb energy around the hole in the Hartree-Fock
approximation is
E - llno Jd 3 g(r) - 1 (5.72)

coul-T r r
The integral ECoul is related to the exchange energy which is discussed in Sec. 5.6.3. The
above two relations for g(r) can be checked in the Hartree-Fock approximation. In this case
!
g(r) = + gtt(r). The normalization integral yields (x = rkF)
J
no d 3 r[gtt(r) -!] = - n; Jd 3 rA(r)2 = - ~ J~ dxil (X)2 = -1
while the Coulomb integral yields
Ec 1= llno Jd 3 r [g (r) _1] = _ llno Jd 3 r A(r)2

ou 2 rtt 2 4 r
= -~e2kFJoo dx il (x)2 = _ 311kF (5.73)

n 0 x 4n
These results are correct for Hartree-Fock. It is the same exchange energy per particle which
was derived by diagrammatic means.
The pair distribution function g(r) is the distribution of electrons, on the average, about
any electron. It goes to unity at large distances, which is a result of the uniform distribution no
of electron charge. That is, the electron charge density is -enog(r). Since g(r) is less than
unity near r --+ 0, the electron charge is depleted in the vicinity of the electron. This reduction
may be viewed as a hole in the electron density. Wigner suggested the name exchange and
correlation hole. The hole moves with the electron.
According to the normalization (5.71), the total charge missing from this hole is one
electron charge. The factor -no(g - 1) is the density of hole charge, which integrates to 1.
The sum rule is really a statement of charge neutrality. The hole has a positive charge density,
since it is the absence of electrons. The integral of this charge density must be e. Each
electron of charge -e has in its hole the amount of charge e, and the system is neutral. The
homogeneous electron gas has a uniform charge density only on the average. To a particular
electron, the system is not uniform at all, since other electrons are not as likely to venture near
to it as they are to other points.
The Coulomb integral gives the potential energy of the system. The electron interacts
with its own hole charge. It is not influenced by the electrons or uniform charge density
farther away, since they cancel on the average. Near the electron the positive background
charge is not canceled by the other electrons, since the other electrons are not as likely to be
Sec. 5.2 • Wigner Lattice 311
nearby. In the Hartree-Fock approximation, only electrons of parallel spin make the exchange
and correlation hole. Since gt,l. = !, the antiparallel spins are not affected.
It is possible to arrange g(r) so that the sum rule (5.71) is still obeyed, yet the potential
energy is lower (i.e., a larger negative value) than found in Hartree-Fock. This lowering is
done by permitting gt,l.(r) to become less than! near the point r --+ 0, as shown by the dashed
line in Fig. 5.5. There is no particular reason it must go to zero. The only fixed rule is that it
may not be negative, since it is a probability. Since the total hole charge must be unity, the
change in gt,l.(r) must be accompanied by a change in gtt(r) which causes it to rise more
steeply with r, as shown by the dashed line in Fig. 5.5. It is these changes which give rise to
the correlation energy. Changes in gt,l. and gtt will usually cost some kinetic energy, so that
one cannot just adjust gss' to maximize the potential energy alone. The system will seek the
lowest energy state at T = 0, which is the state with some "correlation" between the motion
of electrons with antiparallel spins. This "correlation" between the motion of pairs of elec-
trons is what gives rise to the "correlation energy."
5.2. WIGNER LATriCE
Actual electron systems exist over a range of densities. The range in metals is
approximately 1.8 < rs < 6. Electron densities inside atoms are also quite variable. It is
necessary to have a description of the correlation energy which is valid for a distribution of
densities. In the previous section, a formula was derived for the high-density limit of a
homogeneous electron gas. The next step is to derive the result for the low-density limit. The
final formulas are found by interpolating between these extremes. This procedure was first
carried through by Wigner (1934), and his formula!
E = 0.88 E (5.74)
C rs + 7.8 ry
is still widely used. Recent interpolation formulas are only slightly better. At low densities,
Wigner speculated that the electrons would become localized and form a regular lattice. He
used the jellium model where the positive charge is uniformly spread through the system. The
electron lattice would presumably be a close-packed structure such as bcc, fcc, or hcp, in
which electrons would vibrate around their equilibrium positions. There would be vibrational
modes of the electrons, and they would be at the plasmon frequency. At very low density the
potential energy becomes more important than the kinetic energy. The kinetic energy is in the
zero-point motion of the vibrational modes. Localization cannot occur until the zero point
energy is less than the potential energy.
Wigner calculated the potential energy in the following fashion. A Wigner-Seitz model
was taken for the unit cell of the lattice. It is a sphere of radius rsao, with the electron at the
center. Each sphere has overall neutrality, since the one-electron charge at the center is
canceled by the positive charge inside the volume of the sphere. For a sphere of radius rsflo,
there is one unit of charge, which follows from the definition of rs. Outside of each sphere, the
electric field is zero. If all unit cells are neutral spheres, then they exert no electric fields on
each other. The electric fields inside a sphere arise only from the electron and positive charge
within that sphere. Of course this model is only approximate, since the unit cells are not truly
spheres. Spheres cannot be packed together to cover all volume. The error made by the
Wigner-Seitz approximation is remarkably small.
I Wigner's original paper contained an error. The corrected result is given.

The first term is the potential energy between the electron and the uniform positive
background. It is
3c? Jr,a o
E
ep
= Jd r (- -c?)
3
r
no = - - -
r;a~ 0
rdr
= - 2~~s = -~ (:aJ (5.75)
where -c? /r is the potential energy and no = 3/(41tr;a~) is the density of positive charge at
each distance r. The result is -3/rs in atomic units, which is a large energy term.
The second term in the potential energy is the interaction of the positive charge with
itself. The potential energy VCr) from the positive charge at a distance r from the center is
obtained by solving first for its equivalent electric field,
_ dV = eE(r) = c? no (41ty3) = c?r (5.76)

dr y2 3 r;a~
which is c? /?- times the total charge within the sphere of radius r. Integrating this equation to
obtain V(r), there is a constant of integration. It is determined by the condition that the total
potential, from electron and positive charge, must vanish at the surface of the sphere. The
constant is chosen to make VCr) be c? /rsao at the surface. The result is
(5.77)
The potential energy of the positive charge interacting with itself is found using
E 3
= !Jd rV(r)n o = ~~Jr,ao ?dr[3 _ (_r )2]
pp 2 4 (rsao) 0 rsao
3 c? 6
= - - = -E (5.78)
5 rsao 5rs ry
!
The result is multiplied by because it is a self-energy. The final result is 1.2/rs in atomic
units. There are only two potential energy terms. The interaction of the electron with itself is
not included. Aside from the fact that it is infinity, it does not change in the metal and so does
not contribute to the cohesive energy of the system.
The sum of these two terms is -1.8/rs . It is the total potential energy in the Wigner
lattice in the Wigner-Seitz approximation. It is a large term, e.g., it has a larger coefficient
than the exchange and correlation energies which were found for the free-particle system. The
system apparently has gained energy by the localization of the electrons.
The actual energy of several Wigner lattices was calculated by Sholl (1967) using
Madelung summation methods. He found the energy of a lattice of point charges in a uniform
positive background. He expressed his results as -A/rs , where the parameter A was found for
several lattices:
Lattiee A
se 1.760
fee 1.79175
bee 1.79186
hep 1.79168
Sec. 5.2 • Wigner Lattice 313
The sc lattice has a unit cell which is not very spherical, and its A is different. But the
other lattices are more close-packed, and the coefficient 1.792 is remarkably close to the
Wigner-Seitz value of 1.8. The latter approximation errs by less than !%.
Is the Wigner lattice stable? To answer that question, first consider the potential energy
on an individual electron in the vicinity of its equilibrium point. In the Wigner-Seitz model, it
is just the negative of the potential energy from the positive charge:
(5.79)
At r = 0 it equals the prior result -3/rs. Away from r = 0 the potential increases quad-
ratically. If each electron moved independently of the others, then the electron would have
harmonic vibrations about the equilibrium point. The zero-point motion of these vibrations is
another term in the ground state energy, and includes the kinetic energy as well as some
potential energy. The zero point energy was evaluated by Wigner, who found that its
contribution toward the ground state energy was proportional to r;3/2. At large rs this term
becomes smaller than the potential energy term, which falls off as r; 1• He concluded the
lattice was stable at sufficiently large values of rs. Of course the particles do not move
independently. Their movements form collective vibrational modes. A better test of lattice
stability is to calculate all the phonon modes and show that their frequencies are all real so the
lattice is stable for collective motions. This calculation was done by Carr (1961). A summary
of the theory of the Wigner lattice was given by Care and March (1975).
In the limit of low density, as rs --+ 00, the potential energy of the Wigner lattice is
proportional to -1.792/rs. This term is the total Coulomb energy, which is correlation plus
exchange. If the exchange energy -0.9163/rs is subtracted from this result, there remains the
low-density limit of the correlation energy:
(5.80)
The numerator is rounded off to 0.88. Wigner also estimated that the high-density limit of the
correlation energy was
lim Ec
rs""",*O
= -0.113 (5.81)
His estimate is now known to be incorrect, because of the 1n rs term which causes a diver-
gence in Ec at small values of rs' The final formula (5.74) is a simple interpolation scheme
which satisfies these two limits. This formula is plotted in Fig. 5.6 as the line marked W The
interpolation procedure assumes that the correlation energy is a smooth function between the
limits of high and low density. The interpolation would be reasonable if the system were in
the same state in both limits. But the two limits describe systems in different phases. The
high-density system has free-electron states, and the low-density system has localized elec-
trons. The smooth interpolation through a phase boundary is not obviously correct, nor
incorrect.
Other interpolation schemes have been proposed. One by Nozieres and Pines (1958) is
Ec = -0.115 + 0.0311n(rs) (5.82)

I
i
1.0 2.0 / 3.0 4.0 5.0 6.0 7.0 8.0 r.
,i
-0.02 i
i
-004 i eM
/
u; i
(!)
a:
LLI
-006 j
III I
0
>- i
a: -008
!
>-
(!)
'/1'-/
,
a:
LLI
Z i
LLI . //
,ill//
Z -0.12 I .
0
i=
«
...J -0.14
LLI
a:
a:
..
8 -016
I I.
-0.18
II
1
FIGURE 5.6 The correlation energy of the electron gas. The line CM (Carr-Maradudin) is the same small rs
expansion given in Fig. 5.4. The other curves are interpolation schemes according to Wigner (W), Nozieres and Pines
(NP), and Lindgren and Rosen (LR).
It is shown in Fig. 5.6 as the curve marked NP. It is quite similar to Wigner's result in the
range of metallic densities. Another interpolation formula has been proposed by Lindgren and
Rosen (1970) (LR),
0.08]-1
E c =- [ rs +3+4 v's
/r-~
- (5.83)
which is also shown in Fig. 5.6. It was chosen to agree with Carr and Maradudin at low values
of rs' Although the LR formula does not have a In rs term, it does agree well with the exact
high-density expansion at low values ofrs . In the range of metallic densities, it has a lower (in
magnitude) correlation energy than the previous two formulas.
There have been other calculations of the correlation energies at metallic densities. These
are usually based on (5.56). The quantity €RPA(q, iro) is replaced by a better dielectric
function-one which is more accurate at low densities. This approximation leads to an
improved correlation energy. Even a direct evaluation of (5.56) gives a reasonable correlation
energy for metallic densities (Lundqvist, 1969). There is no need to evaluate only the low rs
expansion of this result, as we did before. One can numerically solve it for all values of rs.
Some of these improved dielectric functions are discussed in Sec. 5.5.
Sec. 5.3 • Metallic Hydrogen 315
5.3. METALLIC HYDROGEN
It has not been possible to study the three-dimensional Wigner lattice in a laboratory.
There is another system, which is superficially quite similar, for which one could consider
attempting experiments. It is metallic hydrogen. If one takes the Wigner lattice and changes
the sign of all the charges, one has the model for metallic hydrogen. The protons are well
approximated as point charges which are embedded in a free-electron gas. Since all other
monovalent atoms form metals, perhaps hydrogen may also be made metallic.
Of course, at ordinary pressures, hydrogen does not form a metal. It is the gas H2 at
room temperatures. The lowering of temperature turns it into a liquid and then a solid. In both
condensed phases the molecule H2 retains its identity, so that the solid is a molecular crystal
and an insulator. It may be possible to make hydrogen a metal by the application of pressure.
As modern techniques achieve higher laboratory pressures, new phases of solid hydrogen
continue to be discovered. However, they all seem to be molecular crystals and insulators
(Mao and Hemley, 1994).
The discussion of metallic hydrogen will adopt the approximation that the electrons are a
uniform electron gas. Better calculations show that the electron density is slightly nonuni-
form, since the electron density is higher near the proton than at the edge of the unit cell.
Nevertheless, the assumption of uniformity makes only a small error in the analysis of the
ground state energy. For a homogeneous electron gas, the electronic contribution to the
ground state energy is the familiar result
Eg = 2.2099 _ 0.9163 E () E
2
rs rs
+ c rs + H (5.84)
where EH is the Hartree energy, and Ec(rs) is one of the formulas for the correlation energy.
The first three terms for the ground state energy are for the case where the positive charge is
spread uniformly throughout the system. In metallic hydrogen, the positive charge is a regular
lattice of points. The extra Coulomb energy which results from the localization of the positive
charge was calculated above for the Wigner lattice. Changing the sign of all the charges yields
the same result: EH = -1.792/rs' The Hartree energy is the Coulomb energy calculated in
the Hartree approximation, which neglects exchange and correlation.
There are, however, some subtle aspects to this calculation worth mentioning. The
interaction energy between the protons and the uniform electron gas is calculated as before. It
includes the interactions among the protons, but omits the proton interacting with itself. The
electron--electron interactions are evaluated assuming a constant charge density no. Consider
the Wigner-Seitz model of a spherical unit cell. Although there is one unit of electron charge
in each unit cell, it is not treated as a single electron, since the electron--electron interaction
would then be an electron interacting with itself. Instead, each electron is spread uniformly
throughout the solid. In the free-electron model, the unit cell has one charge because 1023
electrons each contribute 10-23 of their charge. Then the calculation of electron--electron
interactions has a negligible contribution from one electron interacting with itself. This
picture is changed by the concept of the exchange and correlation hole. The electrons are not
charges which are uniformly spread throughout the material. They are points which, on the
average, may be found with uniform probability anywhere. But if one electron is at a point,
the others are not, on the average, within its exchange and correlation hole. The radius of the
exchange-correlation hole is similar to the radius of the Wigner-Seitz cell. The influence of
the correlation hole is included in the correlation energy. Once it is included, the lattice energy
from the protons is found by assuming the electron density is uniform.
The net result is that the energy from electron-electron interactions is small. In the
Wigner-Seitz spherical model, the total energy from electron-electron interactions is
(5.85)
Eee is the sum of exchange, correlation, and the term 1.2/rs which comes from the uniform
density of electrons interacting with themselves in the spherical unit cell. The term 0.28/rs is
small and positive, while Ec is small and negative. They largely cancel, so the net electron-
electron energy is small. The concept of the exchange-correlation hole is correct. If one
electron is put into a unit cell, then other electrons will not likely be there because of this hole.
And if other electrons are not in the cell, there are no electron-electron interactions. Thus it is
easy to understand why the net electron-electron interaction is small.
The ground state energy for metallic hydrogen is
2.2099 1
Eg = - - 2- - -(0.916 + 1.792) + Ec(rs)
rs rs
(5.86)
Eg(rs) must be minimized, with respect to rs' to obtain the predicted density of metallic
hydrogen; the predicted value of rs is defined as rso. The minimization will be done in the
Hartree-Fock approximation by ignoring the correlation energy Ec, because the variation of
Ec(rs) with respect to rs is small at metallic densities. The correlation energy will have little
influence upon the choice of rso. The simple equation to minimize is:
0= dEHF =..!!...- (2.210 _ 2.708) = _ 4.420 + 2.~08 (5.87)

drs drs r; rs r,o ~o rso
The minimum occurs at rso = 1.632, and the Hartree-Fock energy at this value is
__ 2.210 _ 2.708 _ -0 830E
EHF (rso ) 2 -. ry (5.88)
rso rso
The Wigner correlation energy at this value of rso is -0.093, so that the predicted ground
state energy is -0.923Ery. It is the binding energy per electron. It is not even as large as
atomic hydrogen, which is 1.00 Rydberg per electron. The prediction is that metallic
hydrogen is not as bound as atomic hydrogen. Of course, it is possible to do a better
calculation of Eg for metallic hydrogen, which includes the nonuniformity of the electron gas.
The best result so far is Eg = -1.048Ery per atom at rso = 1.60 by Hammerberg and Ashcroft
(1974). Although this value is more bound than atomic hydrogen, it is still not as bound as
molecular hydrogen or solid molecular hydrogen which has rs = 3.3. The conclusion is that
hydrogen would rather be a molecule than a metal, and this conclusion is in accord with
experiments.
5.4. LINEAR SCREENING
Screening is one of the most important concepts in many-body theory. Charges, which
are able, will move in response to an electric field. This charge movement will stabilize into a
new distribution of charge around the electric field. This new distribution is just the right
Sec. 5.4 • Linear Screening 317
amount of charge to cancel the electric field at large distances. The proof of this is rather
trivial. If the electric field is not canceled at large distances, more charge will still be attracted
until it is sufficient for cancellation. If the electric field is caused by an impurity charge
J
distribution pj(r), with net charge Qj = d 3 rp;(r), the amount of mobile charge attracted to
the surroundings is exactly -Qj. This fact was already used in discussing the Friedel sum rule
in Sec. 4.1. The name screening charge is applied to the mobile charge attracted by the
impurity electric field. It will also have its own distribution in space ps(r). The screened
potential from the impurity charge and the screening charge is given by
<j>(r) = Jd r' pj(r') + pir')

3
(5.89)
Ir-r'l
This formula is an exact result, as long as ps(r) is found exactly.
The screening charge is not necessarily in bound states due to the electric field from the
impurity. Of course, that could happen if the electric field from the impurity is strong enough.
But quite often the screening charge is from the unbound conduction electrons of the metal or
semiconductor. In their motion through the crystal, they spend a little more time near the
impurity potential, if it is attractive, than they do elsewhere in the solid. When these motions
are averaged, there is more electron density near the impurity than elsewhere, which is the
screening charge. If the impurity potential is repulsive for electrons, they tend to spend less
time near the impurity, so the average charge is depleted near the impurity. Here the screening
charge is positive, since it signifies a reduction in the average density of electrons, which have
negative charge.
The classical macroscopic theory is quite familiar. The electric field E and displacement
field D obey the equations
v . D(r) = 4nPi(r) (5.90)

V • E(r) = 4n[pj(r) + ps(r)] (5.91)
All equations are Fourier-transformed to give
iq • D(q) = 4npj(q) (5.92)

iq • E(q) = 4n[Pi(q) + piq)] (5.93)
The components of D(q) and E( q) along the direction q are the longitudinal fields D/( q) and
E/(q). The longitudinal electric field is related to the scalar potential E/(r) = -V<j>(r) or its
transform <j>(q) = iE/(q)/q:
4n
D/(q) = iq p;(q) (5.94)
4n
E/(q) = iq [p;(q) + piq)] (5.95)
4n
<j>(q) = q2 [pj(q) + Ps(q)] (5.96)
The dielectric response jUnction is defined as the ratio D/( q) / E/(q) in the limit where pj --+ 0:
E(q) = lim D/(q) = lim [ pj(q) ] (5.97)

p;~O E/( q) p;~O Pj(q) + piq)
In this limit e(q) becomes a property of the material and is independent of the charge
distribution. One of our goals is to calculate this dielectric function. The linear screening
model assumes this definition is true for nonzero Pi(q), which gives for the potential
'(q) = 41t p;(q) (5.98)

q2 e(q)
'(r) = J(21t)3 41t Pi(q)

d 3q
q2 e(q)
e iq .r
(5.99)
The potential <1>' is the total potential from screening charge and impurity charge. That is
obvious from its definition, since E(r) is the electric field in (5.90) from both screening and
impurity charges. The potential ' should be similar to the exact screened potential in
(5.89). They are identical in the limit where Pi is small. The linear screening approximation is
to calculate '(r) in place of (r). It is a much easier function to calculate, at least after e(q)
has been determined. Another feature of the linear screening model is that the screening
charge Ps(q) density is proportional, in q space, to the impurity charge density Pi (q). Linear
screening models assume that
e( ) - Pi(q) (5.100)
q - Pi(q) + Ps(q)
p.(q) = _1__ 1 (5.101)

Pi(q) e(q)
are valid for finite values of Pi' rather than infinitesimal ones.
A simple example will illustrate when linear screening does not apply. Consider a point
charge Z = -1 in the homogeneous electron gas with high density Po. Here the screening
charge is positive, which requires a depletion of the electron density around the impurity. This
behavior is sketched in Fig. 5.7(a). Assume that the linear screening model works well for
pI r)
Po
plr I
FIGURE 5.7 (Top) Change in electron charge density from linear screening (neglecting Friedel oscillations).
(Bottom) A nonlinear screening model must be used for a much larger impurity charge.
Z = -1, although that may not be a good assumption in metals. Now consider the response to
a large charge of Z = -10. In the linear response theory, the screening charge is multiplied
1O-fold, so that it predicts the electron density per) to have the dashed line in Fig. 5.7(b). Of
course, this prediction is nonsense, since it has the electron particle density being negative for
small values ofr. The actual electron density has the form shown by the solid line. In this case
the screening charge is not just 10 times that for Z = -1.
The macroscopic theory defined the dielectric function E( q). But it did not provide a clue
to the technique of finding it. That is the role of a microscopic theory. The immediate goal is
to derive a rigorous definition of E( q) in terms of microscopic operators. In the following
section it is shown how this exact equation is solved, approximately, to give model dielectric
functions which are used in calculations.
The derivation of the exact equation for E( q) starts by considering the interaction
between two impurity charges ZI e and Z2e in the homogeneous electron gas. The interaction
potential between these two charges, in the linear screening model, is proportional to the
product ZI Z2' The ground state energy of the system is evaluated, and all energy terms are
extracted which are proportional to ZIZ2' The summation of these terms is defined as the
interaction potential between the two charges. Of course, there will also be terms proportional
to ZI or Z2 (n = 2, 3, ... ), which are the energies needed to put each separate charge in by
itself. The terms ZIZ2: for m, n 2: 2 are contributions to the nonlinear interaction. All other
terms are ignored except ZI Z2' since the immediate interest is the derivation of linear
screening.
The Hamiltonian of the homogeneous electron gas, with two impurity charges ZI e and
Z2e at RI and Rb is written as
(5.102)
(5.103)
The first term Ho is the Hamiltonian for the homogeneous electron gas (5.1). It includes the
electron-electron interactions, as well as the kinetic energy. The term ~ZIZ2/IRI - R21 is the
direct interaction between the two charges. The last term in (5.102) is the interaction potential
between each impurity charge ~e and the electrons of the homogeneous electron gas. They
are represented by their density operator p(q).
The ground state energy is calculated from the thermodynamic potential, which is found
from the linked cluster theorems of (3.264)-(3.266):
1 00
n=no--L:uz
~Z=I
(-Ii J~ J~
Uz = - [ - 0 d'tl ... 0 d'tz(T,Y('tI)'" V('tz)) different connected
A A
where the Uz are the different connected diagrams. Only the terms in this series will be
evaluated which are proportional to ZI Z2' The noninteracting thermodynamic potential no
comes from Ho. At zero temperature, no is the ground state energy of the homogeneous
electron gas, which was evaluated in Secs. 5.1 and 5.2. The last two terms in (5.102) are the
interaction potential V which enters the perturbation expansion for the thermodynamic
potential.
The term U I in the expansion has only one power of V. In this order, the only contri-
bution proportional to ZI Zz is from the direct interaction
U = _A
I
zI ZZe2 == _AZ z
.... IR I -Rzi
Jd q ve
.... I z (2n)3
3
q
iq .(R t -R2) (5.104)
The other first-order term is zero, since the average of p( q) is zero in the electron gas unless
q = O. Of course, these averages are taken with the Hamiltonian B o, which is without the
impurities present.
The term Uz has two powers of V. Two powers of the last term in (5.102) give a
contribution
(5.105)
This term is the only one in Uz; the other perturbation e2zl zz/IR I - Rzi does not enter again
since it has only the one connected diagram. In Uz there is a term proportional to and Zr
another proportional to Zi.
They are part of the energy needed to put each charge, separately,
into the homogeneous electron gas. It is not the total energy, since there are terms in higher
orders of perturbation theory, proportional to Zj(n ::: 3), which also contribute. The term Uz
also has a cross term, which is proportional to ZI Zz
Uz = ZI~Z L eiq.Rt+iq' 'R2VqVq' J~ dT.I J~ dT.z(T~p(q, T.I)p(q', T.z))

v qq' 0 0
There are no other terms proportional to ZIZZ' The higher linked cluster terms U/ have only
higher powers of the charges. The derivation is completed. The net interaction between the
two charges is the sum of the terms from U I and Uz. The Uz term is simplified by the fact that
in a homogeneous system it is nonzero only when q + q' = 0, since the density-density
correlation function of the electron operators is nonzero only in this circumstance:
,Ml = Z
I
zz Jd q ve
3
(2n)3 q
iq '(R t -R2 ) (5.106)
x [1- J:
:~J: dT.I dT.z(T~P(q'T.I)P(-q,T.z))] (5.107)
This formula is compared with the linear screening model (5.99) for the potential from a
charge distribution. One charge, say Zl> is the impurity (Pi(q) = ZI)' The other charge Zz is
the test charge which measures the strength of the screened potential. The net interaction
between the charges can be written as a screened coulomb interaction of the form
VCR - R) = An = Z z
s I Z I
J d 3 q ~eiq'(Rt-R2)
z (2n)3 E(q)
(5.108)
which provides a rigorous definition of the dielectric function:
(5.109)
The correlation term can be further simplified. The correlation function depends only on the
difference of the two arguments 1: 1 - 1:2' This fact, along with the periodicity (3.17) of the
argument, permits one of the 1: integrals to be eliminated to give
so the inverse dielectric function is
v J~ d1:(T,p(q, 1:)p(-q, 0))

-1() = 1 -...!!.. (5.110)
sq v 0
The exact result (5.110) is very important. It relates the dielectric function to the density-
density correlation function. The time variation of the operators p(q, 1:) = e"Ho p( q)e-1Ho is
governed by Ho which is the full Hamiltonian for the homogeneous electron gas but without
the potential of the impurities.
The static density-density correlation function (p(q)p(-q)) is related to the static
structure factor Seq), which was defined in Sec. 1.6. For a system of Ne electrons
(5.111)
The pair distribution function g(r) of the electron gas can be obtained from a knowledge of
Seq). The latter quantity is important in describing correlation in the electron gas. This
observation raises the question of whether there is any relationship between Seq) and l/s(q).
In fact, yes. In order to show this relationship, generalize (5.110) to nonzero values of
frequency:
(5.112)
This result is true, although it has not been proved here. Actually, it could be regarded as the
definition of the longitudinal dielectric function s(q, iron).
Equation (5.112) is used to prove the following important theorem:
Nb +S(q) = - 1- Joo -dro 1 [ 1- -]

Im (5.113)
e q=O nOVq -00 2rc 1 - e-~ffi seq, ro)
The function seq, ro) is the retarded function obtained from seq, iron) by iron -+ ro + ib. The
subscript "ret" will be omitted, since the retarded function is always intended when writing
seq, ro). The factor no = kj..j3rc2 is the particle density, and Vq = 4rce2/q2.
To prove this theorem, again introduce the sets of states 1m) and In), which are exact
eigenstates of the Hamiltonian. These states were used in Sec. 3.3 for proving several
theorems. The Hamiltonian of the homogeneous electron gas has Holn) = Enln). These
complete sets of states are introduced into (5.112):
The t-integral can be done and the retarded function is found in a formal manner:
I v eJ3(En- Em) - I
---,-----.--:- = I _...i. L:e- J3En l(nlp(q)lm}12-.- - - -
seq, leo) v nm leo + En - Em
(5.114)
I v eJ3(En- Em) - I
- - = 1-...i.L:e- J3En l(nlp(q)lm}1 2 •
seq, eo) V nm eo + En - Em + zo
The quantity on the right is real except for the factor of iO. The imaginary part of this term just
gives a delta function:
(5.115)
1m[_I_J =
seq, eo)
-(1- e- J3m ) 1tVq L:e- J3En l(nlp(q)lm}1 2 o(eo +En - Em)
V nm
Divide by (1 - e- J3m ) and integrate over all eo:
Joo
-00
deo
1t
1_
I-e J3
mlm[-I-J
s(q, eo)
= _ VqL:e-J3Enl(nlp(q)lm}12
V nm
V
= _...i. (p(q)p(_q)} (5.116)
v
The term on the far right is the static density-density correlation function. It gives the static
structure factor Seq), as shown in (5.111). This completes the proof of (5.113). At zero
temperature, the above formula becomes (q =1= 0)
Seq) = - -I Joo -
deoI m
[ - I -] (5.117)
nOvq 0 1t seq, eo)
which is the way it is often written. The pair distribution function g(r) or the static structure
factor S(q) is obtained from a knowledge of the frequency-dependent dielectric function
s(q, eo). The latter formula is not dependent on any assumptions regarding linear screening. It
is exact. It arises because both Seq) and seq, eo) are related to the density-density correlation
function. The assumption of linear screening is merely using (5.99) to calculate the screened
potential from the impurity charge distribution Pieq).
The density-density correlation function
-J: dteimn~(T~p(q,t)p(-q,O)} (5.118)
has the appearance of a Green's function in the Matsubara representation. There is an operator
p( -q, 0) acting at t = 0 and its inverse at t. Since the operator is the density, the Green's
function provides the response of the system to a density fluctuation. To develop the analogy
further, the function
Seq, eo) = __I_Im[_I_J (5.119)

nOvq seq, eo)
is the spectral function of this operator, since it is proportional to the imaginary part of the
retarded Green's functions associated with this correlation function. This observation is
important, since the spectral functions provide direct physical information. For the electron or
the phonon, peaks in their spectral functions are interpreted as excitations of these operators.
Sec. 5.5 • Model Dielectric Functions 323
In a similar way, the peaks in S(q, ())) are interpreted as longitudinal excitations of the electron
gas. These are two-particle excitations, since the density operator itself contains two opera-
tors-one creation and one destruction. In fact, some of the excitations, such as plasmons, are
collective excitations of many particles.
The density operator has boson properties, since it is the product of two fermion
operators. So S(q, ())) is a spectral function for boson operators. Consequently, it has many
features in common with other spectral functions for bosons; i.e., compare (5.113) with the
similar phonon result (3.136):
2Nq +1= J OO
-00
d())
-2 nB«()))B(q, ()))
1t
Another feature of S(q, ())) is that it is positive for ()) > 0 and negative for ()) < 0, with
Seq, -())) = -Seq, ())). This identity may be shown directly from (5.115).
The dielectric function has been defined in terms of the interaction between two fixed
impurity charges. The assumption has been that the impurity charges are classical objects.
There remains the question of the effective interaction between two electrons, which are
surely not classical objects. The present theory suggests that there is an additional factor in the
effective interaction between two electrons, which is a vertex correction
v
W(q) = E(;) r(q) (5.120)
The main contribution to the vertex correction r(q) arises from the ladder diagrams at the end
points of the interaction. The method of evaluating this vertex correction is discussed in later
sections of this chapter.
5.5. MODEL DIELECTRIC FUNCTIONS
No one has yet derived the exact dielectric function of the homogeneous electron gas.
Instead, approximate solutions have been obtained to (5.110). Some of these have been very
successful, perhaps because they are simple or perhaps because they are accurate. They
acquire the name of their inventor(s). Four of them are described here, which is only a small
subset of the vast number which are available.
5.5.1. Thomas-Fermi
The Thomas (1927)-Fermi (1928) theory is also called the Fermi-Thomas theory about
half the time. The derivation begins with the exact equation for the screened potential energy
from an impurity charge distribution Pier),
(5.121)
where p,(r) is the screening charge. The symbol P is used for charge density, and nCr) for
particle density (p = -en), and p(q) for the density operator. This equation can be obtained
from (5.89) by the application of -eV 2 . The factor -e arises because VCr) is the potential
energy for electrons. Now start to make approximations. In Thomas-Fermi theory, the
electron density n(r) is represented locally as a free-particle system. Write the screening
charge as the difference between n(r) and the equilibrium charge density no.
ps(r) = -e[n(r) - no] (5.122)
where the electron charge is -e. For a free-particle system, the local density is n(r) =
kj..(r)/3n 2 , where the Fermi wave vector is now a local quantity. It, in tum, is determined by
the condition that the chemical potential /l is independent of position:
kJ..(r)
2m = EF(r) = /l - V(r) (5.123)
The physics is illustrated in Fig. 5.8. Assume the potential V(r) is slowly varying in space.
Draw a little box in each region of space as the shaded region and treat it as a gas of fermions.
If the absolute Fermi level is at /l, then the effective Fermi level is reduced or raised by the
value of the local potential V. If these approximations are collected, there results the equation
v2V = 4ne\ Pi(r) + eno - en o[ 1 - ~~)r/2} (5.124)
For atoms, this approximate equation is solved exactly with EF = 0 and Pi = Zi)3(r) to give a
good description of atomic potentials and charge distributions (Landau and Lifshitz, 1958).
The assumption that V(r) is slowly varying does not seem unduly restrictive. To get a linear
screening model, and hence a dielectric function, a further assumption is needed. It is
assumed that V /EF «1, so the root is expanded (1 - V / EF )3/2 ~ 1 - 3 V /2EF to obtain the
equation
(5.125)
The term on the right, proportional to V, is now moved to the left. Its coefficient is defined as
the square of the Thomas-Fermi screening wave vector qTF
(V 2 - q}F )V(r) = 4nePi(r) (5.126)

2 6ne2no
qTF=-- (5.127)
EF
This equation may be solved in Fourier transform space to give
V(r) = -4neJ d 3q Pi(q) eiq · r (5.128)

(2n)3 q2 + q}F
r-
fL-.---u:r------
FIGURE 5.8 The Thomas-Fermi model has a Fermi degeneracy energy EF which varies with position according to
the variation in the potential energy.
Compare this equation with (5.98), and conclude that the Thomas-Fermi dielectric function is
2
c(q) = 1 + qTF
q2
(5.129)
It has a simple form, which makes it easy to use in a variety of calculations. That is the chief
explanation for its popularity.
For example, an analytical result can be obtained when the impurity is a point charge
p;(q) = Q;. The integrals to evaluate are (v = cos 9)
V(r) =- eQ; Joo 2ldq2 J1 dve;qrv (5.130)

n 0 q + qTF -1
= - 2 -eQ; Joo 2
qdq . ( )
2 sm qr (5.131)
nr 0 q + qTF
= _ eQ; JOO qdqe;qr
(5.132)
inr -00 q2 + q}F
eQ· qTFr
V(r) =-- ' e- (5.133)
r
The last integral is done by closing the integration contour in the upper half plane and taking
the residue of the pole at iqTF' The screened interaction has the form of a Yukawa potential.
The interaction declines rapidly at large distances because of the exponential dependence
exp( -qTFr). In metals, the Thomas-Fermi wave vector has a typical value of 1 A-I. The
screened Coulomb potential declines rapidly on the scale of a unit cell. The screening wave
vector may be expressed in atomic units as
4 ) 1/2 1.5632
aoqTF = ( - kFao = --- (5.134)
n ,Jr;
For example, at sodium density rs = 3.96, one finds qTF = 1.48 A-I.
Thomas-Fermi theory provides only a static model for c(q). It is not usually used to
describe the dynamic response c(q,O).
5.5.2. Lindhard, or RPA

The Lindhard (1954) dielectric function is more commonly called the RPA, for random
phase approximation. It is a model for a static c(q) or dynamic c(q, 0) dielectric function. It
was already introduced in Sec. 5.1 for the discussion of correlation energies. It is rather easy
to derive and has a simple conceptual basis. It also predicts correctly a number of properties
of the electron gas such as plasmons. In the early days of electron gas theory, it was the
dielectric function. Nowadays there is a tendency to use one of the recent models, which are
better for describing the response of the electron gas. Two derivations of the RPA will be
presented: one from equations of motion and the second using Green's functions and
diagrams.
The derivation by equations of motion is also called the method of self-consistent field
(Ehrenreich and Cohen, 1959). One introduces an impurity charge density p;(r, t) or its
equivalent Fourier transform p;(q,O). The equivalent impurity potential is Vj(q,O) =
-ecMq, (0). There are three potential energies in the problem. The first is from the impurity
V;(q, (0), the second is from the screening charge Viq, (0), and the third is the total potential,
which is the sum of these two:
V(q, (0) = V;(q, (0) + V.(q, (0) (5.135)
2 4ne
V Vir, t) = 4nePs(r, t), or Viq, (0) = - -2 piq, (0) (5.136)
q
2 4ne
V V;(r, t) = 4nep;(r, t), or V;(q, (0) = - - 2 Pi(q, (0) (5.137)
q
The major assumption in the derivation is that the electrons respond to the total energy V.
When solving the equations of motion for the electron, use the potential V(r, t) or its
transform V(q, (0). This choice presents a minor problem, since initially V(q, (0) is not
known. That is the object of the calculation, and V;(q, (0) is assumed known. Once V(q, (0) is
known, the dielectric function in the linear screening model is
( ) _ V;(q, (0) (5.138)

E q, (0 - V(q, (0)
and the calculation is completed. In the method of self-consistent field, it is assumed that the
electrons respond to V; then try to determine this function self-consistently. Write the
effective Hamiltonian for the electrons as
(5.139)
The time dependence V(q, t) is put directly into the Hamiltonian (5.139). The impurity charge
is regarded as a classical system which is oscillating. The goal is to find the quantum response
of the electron gas to this classical oscillation. Furthermore, the impurity is assumed to
oscillate at a single frequency: Pi(r, t) = Pi(r)e- irot and V;(r, t) = V;(r)e- irot . The average
response of the system will depend on (0, so write the average of p(q, t) as (p(q, t)} =
p(q, (O)e- irot and the average of V(q, t) as (V(q, t)} = V(q, (O)e- irot . For the homogeneous
electron gas, the density operator p(q) has an expectation value of zero for q =f. O. When the
impurity is present, the expectation value is nonzero and is proportional to the average for the
screening charge:
(Ps(q, t)} = -e(p(q, t)} = -e L:(C:+q,erCper}

per
= -ep(q, (O)e- rot (5.140)
Since the averages for Ps and p are proportional, it simplifies the discussion to use only one
symbol, which we choose to be p. For example, in terms of the average (Ps) = -e(p), Eq.
(5.136) is
4ne2
Vs(q, (0) = - 2 p(q, (0) (5.141)
q
The dielectric response function is defined as the ratio (5.138). The first term in (5.139) is the
kinetic energy of the electrons, and the second is their interaction with the self-consistent
potential V(q, t). Note that there are no explicit electron--electron interactions. They are
included, indirectly, in the interaction term: the part of V( q, t) from the screening Vi q, t) is
caused by electron--electron interactions. It is a rather crude way to include these interactions,
since it includes screening but neglects all other effects of correlation and exchange. In later
sections, other dielectric functions are introduced which are more accurate.
To obtain the screened potential V, in (5.136), an expression is derived for the screening
particle density p,(q, t). It is obtained by writing an equation of motion for this operator and
then solving it approximately. A perturbation on the system of (q, (0) will cause polarization
of the electron system, so that the average (p(q, t)} will now have a nonzero value. In the
linear screening model, this average is assumed to be proportional to the potential causing the
perturbation, so (p(q, t)} ex (V(q, t)}. The goal is to determine the constant of proportionality.
The equation of motion of the density operator comes from
d
i dtP(q, t) = [H, p(q, t)] (5.142)
Actually, it is more convenient to evaluate the equation of motion for the operator:
.d t t
Idt Cp+q,crCpcr = [H, Cp+q,crCpcr] (5.143)
After it has been evaluated, the result is summed over (po) to obtain p(q, t). The impurity
potential V(q, t) is assumed to be oscillating at a single frequency exp( -ioot) so that the time
derivative on the left of (5.102) gives -iooC~+q,crCpcr. The commutators on the right are
evaluated for the Hamiltonian (5.139),
(5.144)
(5.145)
which gives the equation
(Ep - Eq+q + oo)C~+q,crCpcr = -vI L V(q', t)(C~+q+q"cr - C~+q,crCp-q"cr)

q'
(5.146)
The last term on the right is approximated by taking only the term in the summation which
has q' = -q. The terms with other values of q' are neglected. It is assumed they average out
to zero. They do not average to zero, but that is what is meant by the random phase
approximation. The approximate equation can now be solved:
ct C = V(q, t) (C~crCpcr - C~+q,crCp+q,cr) (5.147)

p+q,cr pcr V E - E
P q+q
+ 00
The above equation is now summed over (pO') to give
p(q, t) = L C~+q.crCpcr = V(q, t) L C~crCpcr - C~+q,crCp+q,cr

pcr V pcr Ep - Eq+ q + 00
This equation shows that the operator on the left, p(q, t), is proportional to the operator on the
right, V(q, t). The average is taken of this equation, so that (p) and (V) are replaced by
p(q, ro)e- irot and V(q, ro)e- irot . In addition, the number operators C~crCpcr and C:+q,crCp+q,cr
are replaced by their averages nF(~p) and nF(~p+q)' These steps give an equation which
relates the average of these operators:
(S,148)
which can now be used in (S.14l):
_ 4ne _ (I)
Vs(q, ro) - - 2 p(q, ro) - V(q, ro)VqP (q, ro) (S,149)
q
The result has the screening particle density p(q, ro) proportional to the self-consistent
potential V( q, ro), The constant of proportionality is p(I)( q, ro), which is evaluated below.
The equations may now be solved to obtain the dielectric function. In the equation for
V(q, ro), substitute the new result for VsCq, ro) and then solve for V(q, ro) in terms of Vi(q, ro):
V(q, ro) = Vi(q, ro) + Vs(q, ro) = Vi(q, ro) + vpp(I)V(q, ro)
Vi( q, ro)
(S.1S0)
The ratio of these two quantities is just the RPA dielectric function:
ERPA(q, ro) = 1 - vqp(l)(q, ro) (S.lSl)

p(l)(q, ro) = ! L nF(~p) - nF(~p+q! (S.lS2)
v pcr Ep - Ep+q + ro + 18
The above result completes the derivation using self-consistent fields.

The second method of deriving ERPA is a diagrammatic analysis using Green's functions.
The basic definition of l/E(q, ro) in (S.1l2) is rewritten in the interaction representation:
(S.lS3)
(S.lS4)
where Ho is the kinetic energy term in the homogeneous electron gas and V is the electron-
electron interactions. The operator p has its time dependence determined by Ho in the
interaction representation. The S matrix will be expanded term by term to see what sort of
terms develop. One particular subset of these terms will be summed and will yield the RPA.
The first term in the expansion for the S matrix is just p(l)(q, iro):
(5.155)
(5.156)
(5.157)
The Feynman diagram is shown as a single-fermion closed loop in Fig. 5.9. The wiggly lines
at each end are just added to define the two vertices of the polarization diagram. They could
indicate that the polarization term is in response to an excitation with wave vector q and
frequency iron' The calculation for cRPA is not terminated at this point, since the terms so far
are ljc = 1 + vqp(l) + ... rather than cRPA = I - vqp(l). Obviously, more terms are needed
to get RPA.
The next term in the S-matrix expansion is
(5.158)
There are four terms which result when Wick's theorem is applied to this correlation function.
All contributions have four electron Green's functions and one Coulomb interaction vq" Their
diagrams are shown in Fig. 5.10. The first one is a vertex correction to the basic bubble
diagram. The next two are exchange energy diagrams for the Green's functions in the bubble;
they contribute to the self-energy of these Green's functions. The last diagram contains two
bubbles which are connected by the Coulomb line vq .
FIGURE 5.9
FIGURE 5.10
Earlier it was remarked that the density-density correlation function had the appearance
of a Green's function. It also has a Dyson equation. The exact evaluation of this correlation
function may be written as
I J~ .
P(q, ioo)
T
- - d'te'ffin (Ttp(q, 't)p(-q, O)} =
----'-=---
V o l - VqP(q, (0)
(5.159)
Here the density operators p(q, 't) have the 't dependence governed by H = Ho + V, instead
of only Ho as in p(l). The polarization diagram P(q, ioo) is the summation of all "different"
polarization terms. Polarization diagrams are not "different" if any of their parts are linked by
a single Coulomb line vq . For example, the last diagram in Fig. 5.10 is not a different
polarization diagram. This term arises from the expansion
_p_(I_)+-----,----p-=(2:-)+_
.._. = (p(l) + ...)[1 + v (p(l) + ...) + ...]
I - vi P (l)+ ... ) q
= p(l) + Vip (I»)2 + ... (5.160)
where it is the term viP(I))2. There are terms in P(q, ioo) which have more than one bubble,
but they must be connected by more than one Coulomb line.
The random phase approximation is approximating the exact polarization diagram
P(q, ioo) by its first term, which is p(l)(q, ioo). The RPA is found from the summation of all
single bubble polarization diagrams
1 v p(l)
-- = 1+ q - ------;= (5.161)
ERPA 1 - vqp(l) 1 - vqP(I)
which does give ERPA = 1 - vqp(l)(q, ioo). This derivation makes clear the approximate
nature ofthe RPA. The total polarization operator P( q, ioo) has an infinite number of terms,
while the RPA retains one. The exact dielectric function is easily shown to be
(5.162)
An obvious way to improve the dielectric function is to include more terms in the summation
of polarization contributions (Geldart and Taylor, 1970). This step is not as simple as it
sounds. There are an infinite number of possibilities, so some physics must be used to guide
the choice. Neither does it help that the obvious possibilities, such as the first diagram in Fig.
5.10, are not simple to evaluate analytically. In fact, most progress has been made by
nondiagrarnmatic means, as will be discussed below.
The RPA dielectric function is evaluated once p(l)(q, ioo) is obtained. The retarded
function is obtained by taking the analytical continuation ioo --+ 00 + io. The retarded
dielectric function is complex, and its real and imaginary parts are called EI and E2:
(5.163)
(5.164)
• For q < 2kF
(5.165)
=0
• For q > 2kF
The real part E[ (q, 0)) may be represented by a single formula. The imaginary part E2(q, 0))
has a variety of functional relations for different values of (q, 0)).
Figure 5.11 shows graphs for E[, E2, and -Im(l/E) for rs = 3 and all relevant
frequencies. For large values of q, say q ::: 2kF , one finds that E[ ~ 1, 82 is small, so that
82 ~ -Im(1/8). There is only a single line drawn for these cases.
(0) (b)
80 10
2 2
( d)
3 (c)
Im(lI€j
.,,
./ .
~
2
, ..... , I .
'
..... I .
/~) ./'-',i E Im(1/~)
__ ...
'.-.-'
2")..~
I ",.
$0-:"-'
0 ,2 2 4 6 8
w/E F w/EF
FIGURE 5.11 The RPA predictions regarding EJ , E2, Im(l/E} as a function of q and ro. Results shown for rs = 3.0.
The real part cl always approaches 1 at large co. At co ~ 0 then c2 ~ 0, and the static
cRPA is just cl(q, 0). Using the notation (x = q/2kF ),
c(q, 0) = 1 + ~~ [1 + ~(1-~)lnl~ ~:IJ (5.166)
which is always positive. The prefactor to the bracket includes the square of the Thomas-
Fermi wave vector qTF' For values of q < kF' then cl (q, co) becomes negative for intermediate
values of co. This requires two crossings of the cl = 0 axis. The low-frequency crossing
always happens when c2 is large, so that -lm(l/c) = c2/(cI + cD is well behaved when
cl ~ O. However, the high-frequency point where cl = 0 has c2 = O. In that case
-lim(1/c) = c2/(cI + cD = 1I:8(cl), so that a delta function is obtained. This delta function
is the plasmon peak which is the sharp singularity on the right of the graph. It is given a
nonzero width in Fig. 5.11 to aid the eye. Remember that peaks in S(q, co) are interpreted as
excitations of the system
Seq, co) = --1m

1
nOVq
[lJ
- - = --2--2
c(q, co)
1 c2
Cl + c2
nOVq
(5.167)
Plasmons are excitations which exist in real metals and in any electron gas. They were
discovered in ionized gases by Langmuir. They occur at small values of q, as is evident in Fig.
5.11.
Examine the limit of (5.164) when q ~ 0 while co remains nonzero. The first step is to
expand p(1)(q, co) while assuming cq « co, qVF « co. Let A = cp +q - cp and assume that
A < co so that
(5.168)
(5.169)
This result is inserted into c = 1 - vq p(1) which derives the high frequency limit of the
dielectric function
(5.170)
(5.171)
For c2 = 0 the condition that cl = 0 predicts that the plasmon peak occurs at the frequency
3 q2 v} )
co = cop ( 1 + 10 co~ + ... (5.172)
At q = 0 the prediction is quite simple, co = cop. The quantity cop is called the plasma
frequency of the electron gas. It depends only on the electron density no and mass m, which is
actually the effective band mass, although these are close to the free-electron value for many
metals. Table 5.1 shows some actual plasma frequencies measured in metals, which are
TABLE 5.1 Plasma frequencies in eY. The values for are tiro;,
corrected for the dielectric screening of the core electrons
Metal Experimental tloop tloo~ Reference
Li 7.12 8.03 7.95 a

Na 5.71 5.90 5.77 a
5.85 b
K 3.72 4.36 4.02 a
3.87 c
Mg 10.6 10.88 10.70 d
Al 15.3 15.77 15.55 e
a C. Kunz, Phys. Lett. 15, 312 (l965).

b 1. B. Swan, Phys. Rev. 135, AI467 (l964).
c 1. L. Robins and F. E. Best, Proc. Phys. Soc. London 79, 110 (l962).
d C. 1. Powell and 1. B. Swan, Phys. Rev. 116, 81 (l959).
e C. 1. Powell and 1. B. Swan, Phys. Rev. 115, 869 (1959).
compared with this simple formula. The predictions for rop are found to be amazingly
accurate. One great reason for the popularity of the RPA dielectric function is that it obviously
describes plasmons very well. Later it is shown that it is exact in the limit where q --+ 0, so
that this success is understandable.
The formulas for E2(q, ro) in (5.165) are complicated. They may be derived in the
following fashion. First, begin with the definition of the imaginary part of the retarded
function:
E2(q, ro) = -vq Im[p~l(q, ro»)
= 2nvq I d3p
- - 3 O(Ep
(2n)
- Ep+q + ro)[nF(~p) - nF(~p+q)] (5.173)
A variable change is made in the term with nF(~p+q) by replacing p --+ -p - q, so the above
equation becomes
Vq
E2(q, ro) = (2n)2 I
d3pnF(~p)[O(Ep - Ep+q + ro) - O(Ep - Ep+q - ro»)
= ~~I:F p 2dp tl dV[O(Eq + ~V - ro) - O(Eq + ~V + ro)] (5.174)
This formula shows that E2(q, ro) is anti symmetric in frequency E2(q, ro) = -E2(Q, -ro). It
will be evaluated first for ro > O. The angular integrations (v = cos 8) are done first, and they
eliminate the delta function:
I I
-1
qpv
dvo Eq + - ± ro
(
m
) m
= -9(p
pq
-
PI,2) (5.175)
m
, =-lro±Eql
Pl2 q (5.176)
E2(q, ro) = 2me'2

-3- (IkF pdp - IkF)
pdp (5.177)
q PI P2
E2(Q, ro) =
me'2
- 3 [9(kF
2 2
- PI)(kF - PI) - 8(kF - P2)(trp - P2)]
,2 2
(5.178)
q
The complicated aspect of the integral comes from the lower limits of integration PI and P2'
These limits are imposed by the angular integral over the delta function. For example, in the
first integral the delta function forces v = cos 8 to equal
(5.179)
and v must have values between ± 1. The latter conditions restrict the value of P to
(5.180)
m
P :::: -1(0 - Eq I = PI (5.181)
q
PI.
The integral contributes only when kJ;. :::: p2 :::: In the second integral, the result is similar
and only the sign of (0 is changed. The result in (5.178) is identical with (5.165), although
there the result is simplified. For example, kF > P2 may be satisfied only when q < 2kF, so
the second term may be eliminated when q > 2kF .
At small values of q, E2 (q, (0) is proportional to (0 at small values of (0. This dependence
is shown in (5.165) and is evident from the graphs in Fig. 5.11. The proportionality to (0 is an
important feature of E2' It arises in (5.178) whenever both inequalities kF > PbP2 are
satisfied, sincePI - p~ = 2m(O. The linear dependence of E2 on (0 must also occur for the
exact dielectric function, which may be shown by a simple argument. The physical process
under consideration is the rate at which electron-hole pairs are made in the electron gas. A
hole is a state which has an electron removed from the filled Fermi sea. An initial electron of
momentum p and energy Sp is excited by a perturbation with (q, (0). The electron is excited to
a new state with momentum p + q and energy Sp+q = Sq + (0. The electron can only be
scattered into states which are previously unoccupied, so Sp+q must be above the occupied
Fermi sea. The basic process takes an electron from below to above the Fermi level. It leaves a
vacancy in the Fermi sea, which is the hole. The excitation process makes electron-hole pairs.
The net rate of making such pairs is
where nF(Sp)[1 - nF(Sp+q)] is the rate of making pairs, while nF(Sp+q)[1 - nF(Sp)] is the
return rate. A formula similar to (5.173) is derived, except that neither Sp nor nF(Sp) need to
be their free-particle values. They could be interpreted as the energy and occupation number
for the fully interacting system. Now change integration variables to pdp = mdsp == mds: and
pdv = (m/q)dS p+q = (m/q)ds', the above integral may be written as
(5.183)
(5.184)
where the limits of integration are ignored in this simple argument. The remaining integral is
over a distribution nF(~) minus the same integral over the displaced distribution nF(~ + co).
The difference between these two must be proportional to co, at small co, since
(5.185)
(5.186)
The rate of making electron-hole pairs is proportional to co. The derivation applies only to
small values of co, since at large values the limits of integration change the result as they did
in the case of the RPA. The area in (q, co) space over which I'.z(q , co) is nonzero is shown in
Fig. 5.12 by the shaded region. These are bounded by the two lines co = I'.q ± qVF. The
excitation spectrum of the electron gas is given by S(q, co) in (5.167). It is nonzero when 1'.2 is
nonzero, so that it also exists in the shaded area of Fig. 5.12. In addition, S(q, co) has the
plasmon peak which exists where 1'.2 = o. It is also shown in Fig. 5.12. This figure actually
describes the excitation spectrum of the electron gas to density fluctuations. At large values of
q/kF' S(q, co) is given accurately by the approximation S(q, co) ~ 1'.2/nOVq, so that it describes
the process whereby single electron-hole pairs are made. At intermediate values of q/kF , the
excitation spectrum has the complex shape shown in Fig. 5.12. These curves can be measured
in metals by the inelastic scattering of energetic electrons through large momentum transfers.
However,the comparison with experiments is deferred until later sections, after several other
dielectric models have been introduced.
5.0
4.0
W
...
..... 3.0
~
a
2.0
1.0
2.5 3.0
FIGURE 5.12 The excitation region of the electron gas in (q,ro) space. The plasmon line rop becomes highly
damped in the region of electron-hole pairs, which is shown hatched.
5.5.3. Hubbard
Hubbard (1957) introduced a correction factor to the RPA of the form
(5.187)
(5.188)
The factor GH(q) is a local field correction. There is an analogy with the Lorenz-Lorentz
dielectric function, which has the form in cubic insulators
4mx
10=1+ 4TCOC (5.189)
1--
3
The two formulas have the same structure if we associate -vqp(l) => 4TCOC and GH => ~. The
factor of 4TCOC/3 comes from summing the local dipoles in the lattice. It is called a "local field
correction". Similarly, GH ( q) in the Hubbard dielectric function comes from summing over
the surrounding structure of the dielectric, which in this case is the exchange-correlation hole.
This curious formula was regarded as an improvement to the RPA in many properties.
Yet its true worth was unappreciated, because it could not be compared with the better
dielectric functions which only became available later. The best dielectric functions today are
written in precisely the form (5.187), with a GH(q) which is only slightly different from the
simple form proposed by Hubbard. His result, which was somewhat of a stab in the dark, is
now regarded as being well ahead of its time.
The factor GH(q) is introduced to account for the existence of the exchange and
correlation hole around the electron. The dielectric function describes how the conduction
electrons of the metal rearrange their positions to screen the Coulomb potential. Because of
the exchange and correlation hole around each electron, when one electron is participating in
the dielectric screening, others are less likely to be found nearby. The exchange-correlation
hole should have some affect upon the nature of the dielectric screening.
The Hubbard local field comes from the vertex corrections in the polarization function
P(q, co). The most important vertex corrections are the ladder diagrams in Fig 5.13. They can
be evaluated by introducing a vertex function rck, k + q) in a four vector notation
k == (k, ikn ). This approximation to the polarization diagrams is denoted PL where "I.;'
denotes ladder
2
PL(q, co) =A L ,§(O)(k),§(O)(k + q)rck, k + q)
tJv k
(5.190)
1
rck, k + q) = 1 - A L W(k - p),§(O)(p),§(O)(p + q)r(p,p + q)
tJv p
The above expression is rather difficult to evaluate exactly. The degree of complexity depends
upon the choice for the effective interaction W(k - p) which are the ladders.
In the Hartree-Fock approximation, to the ladder diagrams, the function W(k - p) is the
unscreened Coulomb interaction W(q) = vq • Since the interaction has no dependence upon
q--~':
?--.
--(. (I + -.
q
r
p+q
I
I W
I
k+q
r p k
FIGURE 5.13 Ladder diagrams of the polarization bubble.
frequency, neither does the vertex function r(k, k + q) depend upon ikn- In this special case
the frequency summation ipn is simple and the above expression is
r(k, k
.
+ q: lqm) = 1-
Jd p Vlk-pr(p, p + q: .
--3
3 np - n p +q
lqm)-.-..!--~~
(21t) lqm + cp - cp +q
This expression has been evaluated by Sham (1973) and by Brosens et al. (1977). They also
added the exchange self-energies ~xCp) - ~x(p + q) to the energy denominator. In this case
the theory is called "conserving" (Baym and Kadanoff, 1961) in that the vertex corrections
and self-energies are both evaluated to the same level of approximation. This dielectric
function has some interesting structure.
A better approximation is to have W(q) be a screened interaction such as
W(q) = vq/c(q). In this case the dielectric function c(q) depends upon frequency (iqm) which
complicates the evaluation of the integral equation for the ladder diagrams. An approximate
result is obtained by making some drastic approximations. The first one is to assume that
r(k, k + q) ~ r(q) and the k dependence is ignored. This approximation signficantly reduces
the complexity of the analysis, since the equations now have the form
PL(q) = r(q)pCJ)(q) (5.191)
1
r(q) = 1 - r(q)A(q) = 1 + A(q) (5.192)
A(q) = ~L W(k _ p)<§CO)(p)<§CO)(p + q) (5.193)

~v p
There is a problem with the last expression. If r(q) does not depend upon (q), then neither
can A(q). Therefore, the factor of W(k - p) must have its argument replaced by another one.
Here is where Hubbard made his important approximation. He replaced W(k - p) by a
statically screened interaction
41te2
W(k - p) => 2 2 (5.194)
q +qTF
A(q) = 21te2 p(l)(q) = v G (q)pCJ)(q) (5.195)
q2+q}F q H
1 q2
GH(q) = -2
q
2 + qTF
2 (5.196)
Once the factor of W(q) is removed from the summation over p, then that summation just
gives !p(l)(q). The factor of! comes because pCJ) contains a summation over spins which
multiplies the result by two. The ladder terms do not have a summation over spin states. Each
electron line in the polarization bubble diagram has the same spin configuration. So the result
!
is multiplied by to account for the factor of 2 in pCI). Note that the derivation gave a
Hubbard factor with a denominator of (q2 + q}F) rather than the original Hubbard expression
of (q2 + kj). Actually, Hubbard waffled, and published both results in separate papers.
Another time he suggested
1 q2
GH(q) = -2 2
q
+ kF2 + qTF
2 (5.197)
This latter result is remarkably close to the best theories today.

Why do the ladder diagrams come from the exchange-correlation hole? The polarization
diagram is a bubble with two electron lines. One is going forward in time ("electron") and
one is going backward in time ("hole"). The hole is an excitation from below the chemical
potential. The ladder diagrams introduce scattering, and therefore correlations, between the
electrons and holes. The correlation hole around the electron must be created by the low
energy excitations of the electron gas, which are just the electron-hole pairs. The correlation
hole is due to correlations between electrons and holes.
5.5.4. Singwi-Sjolander
The 1967 Singwi-Sj6lander dielectric function has the same form (5.187) as Hubbard's
but with a different choice of GH(q). Singwi has collaborated with a variety of authors to
develop improvements in the method of choosing GH(q). Once could remark, in a humorous
way, that GH(q) is time dependent because of its improvements over the years. This variety
makes the discussion difficult, since there are several possible choices to describe. The
original version of Singwi-Sj6lander is derived, and the others are cited in the references (see
Vashishta and Singwi, 1972; Singwi and Tosi, 1981).
Their original derivation is very attractive, because it explicitly includes the exchange-
correlation hole. They derive an equation of motion for the screening charge and then insert
the pair distribution function g(r) into the Coulomb integral between particles. They solved
these equations and obtained
vq p(1)(q, (0)
I::(q, (0) = 1 - 1 + vq GH (q)p(1)(q, (0)
(5.198)
GH(q) 1
= -- J d3k Sk( q - k) - 1]
- - [ q'
no (21t)3 k 2
where S(q) is the static structure factor associated with g(r). Their original derivation employs
Wigner distribution functions. An alternate derivation is presented here which was given by
Singwi et al. (1968). It uses equations of motion. They find the second time derivative of the
particle density operator p(q, t) and derive an expression for the plasma frequency. The
plasma frequency is given by I::(q, (0) = o. Remember that at high frequency Vqp(l) ~ (O~/(O2,
then using the equation for the dielectric function gives
I:: 0 1- ~ [1 - GH(q) + o(q:~}) ]

= = (5.199)
Corrections to the plasma frequency can be associated with the Hubbard local field GH .
The first time derivative is a single commutator with H, so the second time derivative is a
double commutator,
jj(q, t) = -[H, [H, p(q, t)]] (5.200)
(5.201)
where H is the Hamiltonian for the homogeneous electron gas. Consider the first commutator,
(5.202)
where p(q, t) commutes with the electron-electron term in the Hamiltonian. That is not
surprising, since electron-electron interactions can be written as Lq Vqp(q)p( -q) except for
self-interaction effects. An operator commutes with itself. The electron-electron term
contains just density operators, which is why it commutes with p(q, t). The second
commutator provides the interesting result. Another commutator with the kinetic term in H
just gives a repetition of the energy difference factor:
[L
k,s
~k,sCLCk,s' Lpcr (Ep+q - Ep)C~+q,crCpcr] = L(Ep+q
pcr
- Ep)2Cp+q,cr Cpcr
The commutator with the interaction gives four terms, which can be combined to give two
different ones by rearranging the order of operators. They are
1 t t t
-2 L L (Ep+q - Ep)Vq,[CP'+q',sCk+q',s,Ck,s'Cp's, Cp+q,crCpcr]
v kpp'q' s,s',cr
The change in dummy variables p -+ p + q' in the last term makes the operator sequence the
same as the first term:
1 t t
= - L Vq,Cp+q+q'crCk_q',sCksCpcr[(Ep+q - Ep) - (Ep+q+q' - Ep+q')]
v kpq'scr
The kinetic energies largely cancel in this expression. The bracket is simply -q' q'lm. In
writing this term, the operator sequence C~+q+q'crCLq"SCksCpcr will be rearranged into
C~+q+q'crCpcrCLq"sCks' After summing over internal variables, this expression becomes
p(q + q', t)p(-q', t). The rearrangement is permissible as long as the term is omitted which
has a particle acting on itself; a self-interaction is included in pp unless care is taken to leave
it out. The various terms are collected into the equation of motion for the density operator:
-jj(q, t) = L(E p+q - Ep)2C~+q,crCpcr (5.204)
pcr
+~ L vq' (-q' q')P(q + q', t)p( -q', t) (5.205)

v q' m
An obvious way to evaluate the last term is to take q' = -q in the summation. This term
alone gives the equation [p(O) = N e ]
v q2 N
-jj(q, t) = _q_--E. p(q, t) = ro;p(q, t) (5.206)
m v
which describes the plasma oscillations in the electron gas. Since the other terms all vanish
when q ~ 0, it has been shown that the long-wavelength density oscillations are the same
plasma oscillations as predicted in the RPA. The RPA is exact in this limit.
The Singwi-Sjolander result is obtained from treating (p(q + q', t)p( -q', t)) more
carefully. They wrote it in terms of a direct summation over particle positions, as was done in
Sec. 1.6:
p(q + q')p( -q') = L er, . (q+q') L e- irj .q' (5.207)
}#i
The term i = j is excluded, since it is a particle interacting with itself. The factors in the
exponent are rearranged to have one summation over a particle rj and another summation over
the particle separations r i - r/
p(q + q')p(-q') = Leiri'q Lei(r,-rj}'(q+q'} (5.208)
j i=Jj
By following the definitions in Sec. 1.6, the summation over r i - rj can be replaced by its
average in the electron gas:
L ei(r,-rj}'(q+q'} :::} no Jd rg (r)eir .(q+q'} = Neoq+q, + S(q + q') -

3 1
i=Jj
The other summation L e iq · rj is just p( q, t), so that

(p(q + q')p( -q'))' = (p(q, t))[Neo q+q, + S(q + q') - 1] (5.209)
The prime on the bracket means, in doing the averages, to omit the interaction of a particle
with itself. This omission produces the -1 term on the right. The term Neo q+q, is the same
one included in the above derivation, which gave plasmons. The other term gives GH(q). It is
already proportional to p(q, t), so that GH(q) is obtained without difficulty. The electron-
electron interactions give a term
(p(q, t)) ~ L Vq , (-q' q')[NeO q+q, + S(q + q') - 1]

v q' m
= m~(p(q, t))[1 - GH(q)] (5.210)
GH(q)
1 Jd q' q' q' [S(q + q') -
= no (21t)3 7
3
1]
m2p(q, t) = L(E p+q - Ep)2(C~+q,crCpcr) + m~(p(q, t))[1 - GH(q)] (5.211)

pcr
The factor of GH(q) provides a modification of the plasmon dispersion at nonzero q due to
electron correlations. The formula for GH(q) is the same one which was announced in (5.198)
after changing dummy variables k = -q'.
The Singwi-Sjolander formula for GH(q) can be evaluated with only a knowledge of
S(q). It, in tum, is obtained from a knowledge of E(q, m) through the relationship (5.117)
derived earlier:
S(q) 1 Joo dmIm[_I_]

= __
nOvq 0 1t E(q, m)
But E(q, m) depends on GH(q). The three equations (5.198), (5.187), and (5.117) form a triad
which link the three functions GH(q), S(q), and E(q, m). They are solved self-consistently on
the computer, which must be done for each value of r s ' since the results depend on density. In
TABLE 5.2 Constants A and B in the Hubbard local field
Vashista and Singwi (1972) Singwi et al. (1970)

rs
A B A B
0.70853 0.36940 0.7756 0.4307

2 0.85509 0.33117 0.8994 0.3401
3 0.97805 0.30440 0.9629 0.2924
4 1.08482 0.28430 0.9959 0.2612
5 1.17987 0.26850 1.0138 0.2377
6 1.26569 0.25561 1.0216 0.2189
discussing some later versions of GH(q), they (Singwi et al., 1970; Vashishta and Singwi,
1972) remarked that their results could be adequately fitted by the simple expression
(5.212)
The constants A and B, both dimensionless, are given in Table 5.2 for different values of rs'
Since they depend smoothly on density, the parameters A and B may be obtained for other
values of rs by interpolation. This form for GH(q) fits their computed one well at small and
intermediate values of q / kF but not at larger values. However, GH is relatively uninportant at
large q, so this drawback is not serious. In the next section a number of properties of the
electron gas are calculated using all four dielectric functions. In general the accuracy
increases in the order they were present: Thomas-Fermi is least accurate, while Singwi-
Sjolander is the best.
5.5.5. Local Field Corrections

The Hubbard local field correction is actually a function of frequency GH(q, ro). The
frequency dependence is much less well known than the dependence upon wave vector.
Numerous theorists (Niklasson, 1974; Santoro and Giuliani 1988; Vignale, 1988) have
derived the exact limit of GH(q, ro) and Gs(q, ro) in the limit of large q. They are shown in
Table 5.3. The local field for spin correlations Gs(q, ro) is introduced later in the chapter. In
taking these limits, the result depends upon the behavior of ro while q -+ 00. The case for
ro = 0 and nro = cq give the same limit in two (2D) and three (3D) dimensions. The case that
ro =I 0, but not diverging, give different results in 2D and 3D. Here g(O) is the pair distribution
function at r = O. The usual method of deriving these results is to derive equations for the
TABLE 5.3 Exact asymptotic limits (q ..... 00) of the local fields
for charge GH(q, (0) and spin Gs(q, (0). The first row is for (0 = 0
the second row is for nonzero (0. The third row scales "'(0 with f. q •
From Santoro and Giuliani (1988, used with permission)
3D 2D
GH Gs GH Gs
(0=0 1 - g(O) g(O) 1 - g(O) g(O)
(0#0 ~[1 - g(O)] H4g(0) -1] 1 - g(O) g(O)
"'(0 = f. q 1 - g(O) g(O) 1 - g(O) g(O)
response functions using the equations of motion. The correction terms vanish in the limit of
large wave vector and frequency, so that it is possible to obtain exact limits. They find at large
q, and for 0) =f. 0, the asymptotic limits
_ 1 [(q. q/iv(q/)] /
GH(q,O)) - 2N L L
e q' 0-0-'
4 (q)
q V
- 1 [So-o-,(q) - 00-0-']
(5.213)
I
Gs(q,O)) = 2N L L [(q. q/iv(q/)
4 (q) -
/]
sgn(crcr) [So-o-,(q) - 00-0-']
/
e q' 0-0-' q v
So-o-,(q) = ~ (Po-( -q)po-' (q)} - ~e 0q=o (5.214)

e
Po- = Lp C:+q,o-Cp,o- (5.215)
These formulas have the flavor of the Singwi-Sj6lander theory, which is not surprising since
they were derived in a similar fashion. In practice, the local field corrections are rarely needed
in these limits of large variables. However, exact results are always useful as benchmarks for
approximate theories.
Another method to obtain information regarding the local field corrections is from
experiments. Larson et al. (1996) have determined the local field correction GH(q) in metallic
aluminum from an analysis of x-ray scattering. The scattering measures S(q, 0)) for a fixed
value of q as a function of 0). This data is compared to the theoretical result which is written
in the form
S(q,O)) ex: -Im[ 1 _ vq[l _ G~(q)]P(1)(q, 0))] (5.216)
The RPA polarization p(1)(q, 0)) was calculated by Fleszar et al. (1995) using the accurate
numerical wave functions for the energy bands of aluminum. For each value of q, the quantity
GH(q) is treated as an adjustable constant to improve the agreement between the computed
and experimental values. Repeating the experiment for many different points q gives a plot of
GH(q) vs. q, which is shown in Fig. 5.14. There is reasonable agreement between the
measured values, and the predictions for the jellium model.
1.5 • Gexp
- Gv.s
l'a - - GTDLDA
tf 1.0
C5'
0.5
Jt' )lj'i T
I i; I: I
0.01..l.......1II!!!!!I=::r.:..-L-1l-L....L... :......1...
'
..J..1...J..............:......I.....J..:;:..cJ·''Ic....JL.''''''l:~·.!~L_J_.J
0.0 0.5 1.0 1.5 2.0 2.5
q (kF)
FIGURE 5.14 Measured local field function GH(q) in metallic aluminum by Larson et al. (1996). Also shown are
several theoretical curves based upon the homogeneous electron gas (used with pennission).
5.5.6. Vertex Corrections

The Hubbard local field factor GH(q) was derived as a vertex correction to the polar-
ization diagram of the electron gas. This vertex correction is denoted as r H(q), and the
Hubbard form of the dielectric function is written as
EH(q) = 1 - Vqp(I)(q)rH(q) (5.217)
1
(5.218)
r H(q) = 1 + vqGH(q)p(l)(q)
Where else might vertex corrections enter into the calculations of the electron gas?
The first point is to distinguish between the interaction between two impurity charges
ZI 2 in the electron gas, compared to the interaction between two electrons. The impurities
w~re assumed not to be electrons. The above screening function is appropriate only for the
interaction between two protons in the electron gas, or between other charged ions.
Additional vertex corrections are required when electrons interact among themselves. In
some cases one factor is required, while at other times two are required. This confusing
situation is now discussed. The first case is the calculation of the screened exchange energy
~sx of an electron. The self-energy is
._
~sx(k, lkn) -
-~"J
A L..
I-'
d 3q
iqm
3 Vq
rH(q,iqm) (0)
(2n)
( . ) <'§ (k + q, lkn + lqm
E q, lqm
..)
(5.219)
E(q, iqm) = 1 - vqp(ll(q, iqm)rH(q, iqm) (5.220)

1
(5.221)
r H(q, iqm) = 1 + vqGH(q)p(l)(q, iqm)
r H(q, iqm) _ I
(5.222)
E(q, iqm) - 1 - vq[1 - GH(q)]p(l)(q, iqm)
The integrand contains a single vertex function r H(q, iqm)' The vertex function should
depend upon the k variable. In four-vector notation, it is a function ofr(k, k + q). Neglecting
the k dependence is an approximation. All dielectric functions of the Hubbard type, with a
GH(q) factor, make this approximation of neglecting the k dependence of the vertex function.
Figure 5.15 shows the vertex corrections of the screened exchange energy. The wiggly
line denotes the interaction W(q) = Vq/E(q), where E(q) is the Hubbard dielectric function.
The first diagram has a single wiggly line and corresponds to the interaction Vq/E(q). The
second diagram has two wiggly lines. Because of the symmetry, the diagram could be a vertex
correction to either end point. Since the diagram occurs only once, it cannot be a vertex
correction to both end points. Keeping only the ladder diagrams, then the vertex corrections to
the first end point are shown by several more terms in the series. The summation of these
terms, the vertex correction, is shown in Feynman diagrams as a shaded triangle.
The key point is that only one end of the large wiggly interaction has a vertex correction.
Do not put a vertex correction at each end, since that is overcounting. In Fig. 5.l5(b) is shown
two Feynman diagrams. The first has a vertex correction at each end of the large wiggly line.
This term seems to suggest a correction for each end. However, the second Feynman diagram
in Fig. 5.15(b) is topologically identical to the first. The only difference is that a different
wiggly line has been drawn "large". That diagram can be considered to be a vertex correction
to the first end point, where one of the ladder diagrams itself has a vertex correction. In fact,
the correct way to include all ladder diagrams is to have each ladder also have a vertex
(8)
~ + +
(b)
=
FIGURE 5.15 Vertex corrections to the screened exchange energy.
correction at each end. This set of Feynman diagrams has fractal character (Mahan, 1993).
Writing the equations for all ladder corrections makes it obvious that the vertex correction
only goes on one end in the screened exchange energy.
There is another way to show the correctness ofEq. (5.219). Ting et al. (1975) showed
the vertex function r H enters only once in effective interaction. Furthermore, they also
showed the same vertex function enters the numerator as enters the dielectric function.
Ting et al. (1975) derived (5.219) by starting from the definition of the ground state
energy of the interacting electron system
Eg = Ego - 2V~J~ d; J(~:;3 ~[E~(~' iq) -1] (5.223)
. Tlv q p(1)(q, iq)

E~(q, zq) = I - I + TlVq GH (q)p(1)(q, iq) (5.224)
The derivation of this formula starts by defining the ground state energy as the expectation
value of the Hamiltonian (H). The term EgO is from the kinetic energy. The second term is
from the expectation value of the electron-electron interaction (p(q)p(-q»). The latter is
related to the inverse dielectric function using (5.112). The coupling constant TI has been
made dimensionless by having TI = 0 at e'2 = 0 and TI = 1 at e'2. GH(q) may also be a
functional of the product TlX, where X = vqp(1).
The self-energy can be obtained as the functional derivative of the ground state energy
with respect to the occupation number nk' For example, the kinetic energy term can be written
as
(5.225)
(5.226)
(5.227)
The screened exchange energy is the derivative of the energy in (5.223). Assume the only
dependence upon npcr in the integrand is in the susceptibility X. The functional derivative has
the form forf = l/E(q) is
of of 0]><1)
---v --- (5.228)
onpcr - q ox onpcr
p(1) _ 1, , (
--L."npcr 1 . + 1 .) (5.229)
v pcr ~p - ~p+q + zqn ~p - ~p+q - zqn
Op(l) _ 1
(p + q, Ep + zqn) + ~(0) (p + q, Ep -
(0) . .
-1::- - -[~ zqn)] (5.230)
unpcr v
The two Green's function terms in op(1) /on pcr contribute equally, since E(q, iqn) is a
symmetric function of complex frequency iqno This identity removes the factor of 2 in front of
the interaction term in (5.223). The two steps of coupling constant integral and functional
derivative can be arranged to cancel. Introduce the variable y = 11X, which becomes the
variable of both integration and differentiation:
JoI 11
d11 df(l1X)
dX
= ! JX dy df(y) = ! [f(X) -
X 0 dy X
f(O)] (5.231)
When f(x) = I/E = 1/(1 - rX) then f(x) - f(O) = Xr /E. Collecting all of these results
produces (5.219). This derivation is independent of the particular form for the vertex function
r. It is also valid for any form for GH(q) as long as one assumes that GH(q) is a functional of
the susceptibility x. This derivation makes it clear that the screened exchange energy has only
a single factor of the vertex function in its effective interaction.
There are occasions where an effective Coulomb interaction between two electrons has a
double power (r1) of the vertex correction. The ladder diagrams of the polarization diagrams
could themselves have vertex corrections at each end of the wiggly line. The wiggly line is
still the interaction Vq/E(q). An equation for the vertex correction in this case is, using a four
vector notation
v,
r(p,p + q) =1 - Lq' E(q,)
q
r(p + q',p + q + q')r(P,p + q')
X r(p + q,p + q + q')~(O)(P + q')~(O)(P + q + q') (5.232)
In the approximation of taking r(p, k) ~ r(p - k) then this equation simplifies to

v r(q,)2
r(q) = 1 - r(q) L q' , ~(O)(p + q')~(O)(P + q + q') (5.233)
q' E(q)
The effective interaction for the ladder diagram has a double vertex correction. A double term
always happens when two different electrons are interacting. A single vertex occurs when one
electron interacts with itself through the average polarization of the media.
The effective interaction in the screened exchange energy is given in Eq. (5.222). The
Coulomb interaction Vq is multiplied by the factor of [I - GH(q)]. This expression is often
written down as the effective dielectric function. In writing it this way, a vertex correction has
been included. If GH(q) = 1 for some value of q then the Coulomb interaction is effectively
unscreened. In the x-ray scattering from aluminum, that occurs at q = 1.65kF (Larson et aI.,
1996).
5.6. PROPERTIES OF Tt-IE ELECTRON GAS
There are a number of properties of the electron gas which may be predicted with a
knowledge of the dielectric function c(q, 00). These properties may also be measured for
metals. However, one should be careful in applying predictions of the homogeneous electron
gas to real metals, because the ion cores often have a significant influence. A favorable
comparison to experiment may not, in his case, be the signature of the best theory. Four
properties are discussed: the pair distribution function, the screening charge, the compres-
sibility, and the correlation energy.
5.6.1. Pair Distribution Function

The pair distribution function g(r) may be obtained from a knowledge of the static
structure factor Seq) through the transform relation (1.445):
g(r) = 1 + -3k3
Joo kdk sin(kr)[S(k) - 1] (5.234)
2r F 0
The static structure factor S(k) is obtained from the dielectric function through the relation
(5.117). The Thomas-Fermi theory is a static model of dielectric screening, so that it has no
prediction regarding g(r). So g(r) is calculated using the theories of RPA, Hubbard, and
Singwi-Sjolander. All results are obtained only after a numerical computation.
The three results for g(r) are shown in Fig. 5.16 for six values of rs. The RPA and
Hubbard theories are negative at small values of r. This behavior is viewed as a great defi-
ciency, since g(r) is, by definition, strictly a positive function. The pair distribution function is
defined as the probability that another electron is a distance r away from the first; as a
probability, it must be positive. The negative values predicted by the RPA and Hubbard
theories are regarded as significant failures of these theories. The Singwi-SjOlander theory is
slightly negative at large rs but otherwise positive, so it is much better in this respect. For most
metallic densities, it predicts g(r) to be positive for all r. The Singwi-Sjolander theory is a
significant improvement over the earlier theories.
Kimball (1973) proved thatg'(O) = g(O)jao where ao is the Bohr radius andg' = dgjdr.
His proof is based upon the assumption that when two electrons become very close together,
the other electrons can be ignored and it becomes a two-body problem. The relative wave
function is similar to the hydrogen atom, but with a repulsive interaction. The same result was
obtained more rigorously (i.e., including the other electrons) by Vignale (1988).
5.6.2. Screening Charge
The screened potential VCr) about a point charge Zeiler) may be calculated and plotted. It
is slightly inconvenient, since VCr) diverges as r- i at small r, as do all Coulomb potentials.
Instead, it is customary to plot the density of screening charge ns(r) about the point impurity.
This quantity is nonzero at the origin r = 0 and also has the Friedel oscillations at large
distance. In the linear screening model, derived in Sec. 5.4, the potential energy from an
impurity charge of Ze is
VCr) = -zJ d 3q ~eiq.r

(21t)3 c(q)
(5.235)
Sec. 5.6 • Properties of the Electron Gas 347
Operate on both sides of this equation with V 2 . From Poisson's equation, V2 V is proportional
to the excess charge density in the system. There is charge from the central impurity Ze8(r)
and the screening particle density nsCr). The equation for this screening charge is:
V 2 V = 4ne2 [Z8(r) - n,(r)] = 4ne2 Z Jd3q ec q iq • r

--3 - (
(2n)
) (5.236)
nr=Z-
s
() J(2n)3
d3-qeiq . r [ 1 - - 1]
c(q)
(5.237)
The distribution is spherically symmetric in the homogeneous electron gas, so the angular
integrals may be performed at once:
nsCr) = --;- JOO qdp sin(qr)

2n r o c q
[1 - (1 )] (5.238)
In this case a knowledge of the static dielectric function c(q) is sufficient to determine the
density of screening charge. The Thomas-Fermi model may also be used to make predictions.
It was shown in Sec. 5.5.1 that it predicted the screened potential energy to be
Ze 2 exp( -qTFr)/r, and the screening particle density is predicted to be Zq}F exp( -qTFr)/r.
This result is not plotted in Fig. 5.l7. The rest, which are shown in Fig. 5.l7, have to be
obtained by numerical Fourier transform, which was first done by Langer and Vosko (1960)
for the RPA. All curves are very similar, except for Thomas-Fermi, which is the only one to
diverge at r -+ O. The RPA (1), Hubbard (2), Singwi-Sj6lander (3), and Singwi et al. (4)
results all show Friedel oscillations which are similar. They are lacking in the Thomas-Fermi
theory. The Friedel oscillations are real features of impurities in metals, so that the Thomas-
Fermi theory is deficient in several respects. However, the other theories predict remarkably
similar results, which should not be surprising. Screening is basically a one-body property,
and little correlation is evident in one-body amplitudes. The effects of GH(q) are much more
apparent in properties involving two-body correlations.
5.6.3. Correlation Energies

A knowledge of c(q, ro) is sufficient for predicting the correlation energy of the electron
gas. This fact should be expected, since the correlation energy arises from the correlation hole
around the electron. The exchange-correlation hole was put into the derivation of GH(q). Now
c(q, ro) is manipulated to obtain the correlation energy. The discussion follows Singwi et at.
(1968).
The Coulomb interaction energy of the electron gas may be written as
(5.239)
This way of writing the interaction allows an electron to interact with itself. These contri-
butions should be omitted. The average value of p(q)p(-q) in the electron gas is
(5.240)
The prime on the bracket means one is to omit the particle interacting with itself. It explains
why this differs by -1 from the similar average in (5.111). This equation may be derived by
- O . 5 3 L - - - - - - - - -.......-------------~
...
FIGURE 5.16 The pair distribution function g(r) for the electron gas at various densities for several model dielectric
functions. Source: Singwi et al. (1968) (used with permission).
following the same steps used to get (5.209). The interaction energy Eint per electron of the
electron gas is
(5.241)
1.0
Q5
.
......
CII
-0.5
-0.92
-ID
-1.5
-1.70
-2.0
FIGURE 5.16 (Continued)
nI5~------------------------------------------------------~
0.10
"'-
~
.....
Q.
flO
nos
6 8 10
o 2 3 4
FIGURE 5.17 Screening charge density near a point impurity in the electron gas with rs == 3 according to several
model dielectric functions: (I) RPA, (2) Hubbard, (3) Singwi-SjOiander, and (4) Singwi et al. Source: Singwi et al.
(1970) (used with permission).
The summation has been converted to an integration over q. In doing this step, remember that
the q = 0 term is not in the summation, so that the term Nebq=o has no effect. The quantity
Eint(e2 ) is not the Coulomb contribution to the ground state energy. To obtain the ground state
energy per particle Eg , one must do a coupling constant integration of the type discussed
earlier in Sec. 3.6:
t? dA
Eg = ~EFO + J0 TEinl A) (5.242)
This equation may be derived starting from (3.299), which is assigned as a problem. Equation
(5.242) is an exact result for the ground state energy of the homogeneous electron gas. One
must calculate the Coulomb interaction energy for each value of A up to e2 . In practice, this
calculation is the same as finding Seq) as a function of density. The first term ~EFO is the
average kinetic energy, which is calculated assuming that ~ = o.
The angular integrals in the definition (5.241) of Eint may be done, yielding
Eint(e2 ) = e2 Joo dq[S(q) - 1] (5.243)

rt 0
This expression depends on ~ through the prefactor on the right. It also has a dependence
through Seq), which is decidedly dependent on electron-electron interactions. It is conven-
tional to introduce a dimensionless function
1
Y= -- Joo dq[S(q) - 1] (5.244)
2kF 0
which depends on density. In the coupling constant integration (5.242), Y also depends on A.
The interaction energy may be written as
2 2 2
Eint(e ) = --e kFy (5.245)
rt
= _~ (9rt) '/3 yE (5.246)
rtrs 4 ry
The result has been put into Rydberg units in the second line by using kFao = (9rt/4)'/3/ rs .
Next examine the coupling constant (e 2 ) dependence of these terms. The Rydberg energy Ery
o'
is proportional to e4 or 1..2 • Similarly, rs ex A since rs ex a and the Bohr radius is inversely
proportional to e2 . Write these dependences as rs -+ Ars or e4 -+ 1..2 e4 , where 0 < A < 1. The
interaction energy in this notation is
4 (9rt) '/3
2
Eint(e A) = -1..- - yeArs) (5.247)
rtrs 4
The factor y is also dependent on density, which we write as yeArs) or as y(rs) when A = 1.
The coupling constant integration may be done, at least formally, and the ground state energy
per particle may be obtained from (5.242):
E =---- -
g
2.2099
~
4 (9rt) '/3
rt~ 4
J0' dAY(Ar)s
(5.248)
The coupling constant integration involves only the factor y. All the other A's have factored
out of the expression. Of course, y was chosen so for this reason. The correlation energy
consists of all contributions to the ground state energy except kinetic and exchange. Thus the
correlation energy is
Ec = 0.9163 - ~ ( 9n )
rs nrs 4
J
1/3 1
0
d"}.,:y("Ars) (5.249)
One can verify that this vanishes when Seq) and y(rs) are calculated in the Hartree-Fock
approximation. This expression is evaluated numerically after obtaining Seq) and y(rs) as
functions of density. Some results are given in Fig. 5.18. Four curves are shown, corre-
sponding to RPA, Hubbard, Singwi-Sjolander, and Nozieres-Pines. The last is included to
provide a comparison with the earlier plot of the correlation energy in Fig. 5.6. The RPA
curve predicts a correlation energy with a magnitude larger than the others. The other three
are similar and also similar to that of Wigner. The correlation energy may be obtained with a
knowledge of the static structure factor Seq) if it is known for all densities. The Singwi-
Sjolander correlation energy is presumed best, because of its success in predicting a positive
g(r). The rather direct relationship between g(r) and Seq) implies that the best g(r) is the best
Seq), which in turn is the best correlation energy.
The correlation energy is negative. Correlation lowers the energy of the interacting
electron gas, and increases its binding. This result is surprising from a naive viewpoint. The
main contribution of the correlation energy is to screen the electron-electron interactions.
Whereas the Hartree-Fock approximation has a bare Coulomb interaction vq , the exchange
and correlation energy has the screened interaction Vq/s(q, 0)). The naive viewpoint is that
screening weakens the interaction, which lowers the magnitude of the binding energy. That is,
if a Thomas-Fenni dielectric function is employed, then seq) > I so Vq/s(q) < Vq and
IExcl < IExl. The Thomas-Fermi model for the dielectric function actually predicts a positive
correlation energy. The magnitude of the binding energy is reduced. The missing variable is
2 3 4 6
-
( I)
(,!)
ffi -0.04
CXI
0 STLS
>-
5 -0.06
>-
~Z -008
L1J
-010
~
~
...J
L1J -OJ2
It:
15
(,)
-QJ4
FIGURE 5.18 Correlation energy of the electron gas according to different model dielectric functions: RPA,
Hubbard, and Singwi et al. Also shown is the interpolation formula of Nozieres and Pines to facilitate comparison
with Fig. 5.6.
plasmons. The proper dielectric function has plasmon poles, and these excitations strongly
couple to the electron. The interaction with the plasmons provides a significant contribution to
the binding energy, and makes the correlation energy negative.
5.6.4. Compressibility
The compressibility of the electron gas is defined as follows. Let E be the total energy of
the system, which was calculated in Sec. 5.1. It is an extrinsic quantity; i.e., it is proportional
to the number of particles Ne in the system of volume v. The pressure P is defined as the rate
of change in E with volume at constant N e • The inverse compressibility is the rate of change
of P under the same conditions:
P= _(dE) (5.250)
dv N,
~= _v(dP) (5.251)
K dv N,
The definition of K- i is multiplied by v so that K is not dependent on the size of the

system-it is not dependent on Ne . The compressibility is usually compared to that of a free-
particle Fermi gas at zero temperature. There the only energy term is the kinetic energy, which
was shown in Sec. 5.1 to be
E
3
= sEFNe = 10m Ne
31i 2 (31t
-v-
2Ne) 2/3
(5.252)
Differentiate this equation twice with respect to v, while keeping Ne fixed, and find
P=f.E n
5 P 0
=~(31t2)2/3(Ne)5/3
5m v
(5.253)
~ = f.Epno = ~ Ne ( 31t2Ne)2/3 (5.254)

Kf 3 3m v v
The free-particle compressibility is Kf = 3/(2EFno). Most of the numerical results are

presented as the ratio Kf / K, where Kf is the above result. An effective mass of unity is
assumed.
This procedure for finding the compressibility may be generalized. Ignoring surface
effects, the total energy of the system is Ne times the exact ground state energy per particle
E = NeEg. For a fixed N e, the only volume dependence is through the density dependence of
Eino), where no = Ne/v. Twice differentiating the total energy yields
2 dEg
P = -Ne -dEg
dv
= no-dno
(5.255)
~
K
= _v(dP)
~
= no dP(no)
•
= 2nijE'g + n~E"g (5.256)
Ne 0
d 2
= nij -d2 (noEg) (5.257)
no
The last line shows that the compressibility may also be represented as the second derivative
of the quantity f(no) = noEg. Since E = NeEg = vf,J(no) is just the energy per unit volume.
The compressibility is easily found as the second derivative of this quantity multiplied by n~.
The calculation of the ground state energy Eg was described in Sec. 5.6.3. Using this method,
one may findf(no) and its derivatives numerically.
Another relation for the compressibility may be obtained by using the theorem of Seitz,
which was given in (5.34):
df(no)
J.l=-- (5.258)
dno
1 2 dJ.l
-=no- (5.259)
K dno
Here J.l is the chemical potential of the electrons, evaluated ignoring surface effects. The
compressibility is related to the change in chemical potential with density. In some cases it
may be easier to calculate the properties of a particle on the Fermi surface rather than the total
energy. In particular, only a single derivative is needed to evaluate (5.259).
As an example of using this formula, the compressibility is calculated in the Hartree-
Fock approximation. The starting point is the energy of an electron at the chemical potential
1i2k~ i2kF
J.l=--- (5.260)
2m 1t
It is a quadratic equation for kF(J.l). Solve it, and then find the electron density from
=
no k],.j31t 2, where ao is the Bohr radius
(5.261)
3
1 [ 1 1
n - -2 -+
o - 31t 1tao
- - +2mJ.l
(1taol
-]
1i2
(5.262)
(5.263)
(5.264)
Kf =1 __1_=1_~ (5.265)
K 1tkFaO 6.02
The Hartree-Fock prediction is that the inverse compressibility is a linear function of rs , and
becomes negative around rs = 6.02. Other predictions are given below.
Another method of evaluating K is through the compressibility sum rule. This rule is an
exact relationship between the compressibility and the long-wavelength limit of the static
dielectric function (see Nozieres, 1964). Take the longitudinal dielectric function e(q, 00), set
00 = 0, and then take the limit q ---* O. In this limit, one obtains
lim e(q, 0) = 1 + 41t~ n~K (5.266)

q->O q
By substituting the free-particle compressibility KJ = 3/(2n oEF ), the compressibility sum

rule may also be written as
lim e(q, 0)
q->O
= 1 + q}F (K)
qZ K J
(5.267)
The RPA predicts K/KJ = 1, since eRPA ---* 1 + q}P/qz. It also provides a method of relating
K to GH(q). If one takes the limit q ---* 0 for the dielectric function (5.187) and remembers
that V/,(I) ---* -q}F/qZ, then
(5.268)
(5.269)
The function GH(q) vanishes as O(qz), so that the limit GH(q)/qZ yields a constant, which is
called a/k'J.. The constant a is dimensionless. The Hubbard theory predicts that a = so the !,
compressibility is
KJ = lim[1 _ q}P GH(q)] = 1 _ a (qTF)Z (5.270)

K q->O qZ kF
(K)
-.1.
K Hubbard -
- 1 - -1 ( qTF
2 kF
)Z -- 1- ~
3.01
(5.271)
The constant a may also be obtained from the Singwi-S6lander theory, where GH ( q) is given
by (5.198):
(5.272)
Change variables of integration to Q = q - k, and evaluate the expression in the limit where
q ---* 0:
. 31t 1 Joo
= - -k
z
~dQ[S(Q) -
JI dv Q2
q(q - vQ)
lim GH(q)
q->O
3 -Z
F (21t) 0
1]
-I + qZ - 2 vqQ
= - 4~ J~ ~dQ[S(Q) - 1] LI dV[~: (1 - 2vZ) + O(q4)]
= - -qZ3 Joo dQ[S(Q) - 1] (5.273)

2kF 0
The combination offactors on the right is the same as in the definition of y in (5.244). The
Singwi-Sjolander theory predicts that a = y or that
Kf = 1 _ y(qTP)2 (5.274)
K kp
The value of a may be deduced from Table 5.2, since the limit q ~ 0 yields a = AB, where A
and B are the constants listed. This value of a varies with density, but it is generally smaller
than the Hubbard value of! and is nearer to !.
Two ways have been given to evaluate the compressibility. The first is by finding the
ground state energy per particle and differentiating twice. The second is from the long-
wavelength limit of the dielectric function. The two methods would give the same result if all
quantities could be found exactly. But when using approximate theories for E(q, ro), the two
methods of finding the compressibility will give different results. For example, consider the
compressibility in the Hartree-Fock approximation. The ground state energy now includes the
exchange term, and a second derivative of the ground state energy gives a = as derived !
earlier. But when the exchange hole is used to get g(r), as at the end of Sec. 5.1, and then
S(q), and then GH(q), and finally a, a different result is found: a = The Hartree-Fock i-
approximation predicts different results in the two methods of calculation. The form of GH(q)
suggested by Vashishta and Singwi (1972) and Schneider et al. (1970) has the virtue that it
gives almost the same compressibility by both methods of calculation. Denoting their func-
tion as Gvs(q) while G(q) is the function in (5.198), they are related by
GvsCq) = (1 + ~no d~JG(q) (5.275)
The derivative is taken with respect to density. This form is regarded as being particularly
accurate for the compressibility.
Figure 5.19 shows some theoretical compressibilities as a function ofrs ' The solid lines
are results obtained by differentiating the ground state energy. All the theories except Hartree-
Fock predict identical compressibilities when evaluated in this fashion. And the Hartree-Fock
result is not very different. Presumably the curve labeled RPA, SS (for Singwi-Sjolander),
---------------,- RPA
o 2
FIGURE 5.19 The compressibility K of the electron gas compared to that of the noninteracting gas Kf . The five
theories shown are Hartree-Fock, RPA, Singwi-Sjolander, Hubbard, and Vashishta-Singwi. Each theory makes two
predictions: the solid line from the derivative of the ground state energy and the dashed line from the compressibility
sum rule.
Hubbard, or VS (for Vashishta-Singwi) is the proper result, since it is given by so many

theories.
The dashed lines are the results obtained from the compressibility sum rule: from the
limit of c(q) or GH(q) as q ~ O. These vary widely among the theories, except for the
coincidence of Hubbard with the original Singwi-Sj61ander result (see Singwi et al., 1968).
The Vashishta-Singwi curve is nearly identical to that calculated by the other method.
The compressibility is predicted to go negative about rs = 5. This result suggests that the
electron gas is unstable at densities lower than this critical value. Real metals exist with larger
values than rs = 5; i.e., Cs has rs = 5.63, but they are not a homogeneous electron gas.
However, metals do not exist with rs values greater than Cs, which implies that there may be a
fundamental limit beyond which the density may no longer be reduced. Unfortunately, this
point cannot be tested in three dimensions.
The two-dimensional electron gas has a negative compressibility for rs above a value of
2.2-2.3. This prediction can be tested. A quasi-two-dimensional gas can be created in a field-
effect transistor. The electron density in the surface layer can be controlled by a gate voltage.
In silicon devices with very high mobility, it has been found that there is a critical density of
nc = 0.845 x 101lcm-2 (Kravchenko et at., 1994, 1995). Above this density the layer is
metallic, while below this density the layer is insulating. The properties of the insulating state
are still under investigation.
5.6.5. Pauli Paramagnetic Susceptibility

The Pauli susceptibility is another one of the many parameters of the electron gas which
are calculated using Green's functions. The susceptibility of a free-electron metal such as
sodium or aluminum has a number of contributions. One is from the ion cores, and another is
the orbital diamagnetism, which is called Landau susceptibility. These two effects must be
subtracted from the experimental value in order to deduce the Pauli part, which is due to the
spin of the electron.
One way to calculate the Pauli susceptibility is by the use of the Kubo formula derived in
Sec. 3.9.1,
Xct~(q, iro) = J: dr:e- ron ' (T,Miq, r:)M~(-q, 0») (5.276)
Mct(q) = J..lo L cr~~,C;+q,crCpcr' (5.277)

pcrcr'
where cr(ct) are the Pauli spin matrices. Up or down spin are signified as cr = ±1 for the s = !
system. Evaluate the static susceptibility, which has iro = O. The Hamiltonian is the homo-
geneous electron gas. Consider the type of terms which occur when expanding the S matrix.
The first term, which has no interactions, was evaluated in Sec. 3.9.1. It gave the result for the
free-electron gas. which is called XF = J..l6NF' The first interaction term is of the form
lim XzzCq)
q->O
= XF - 1
-2
V pkq'
L vq' f~0 dr: f~0 dr:!
L crcr'
x (T,MzCq, r:)C;+q',cr(r:!)CLq',cr,(r:!)Ckcr,(r:!)Cpcr(r:!)MzC-q, 0») (5.278)
MzCq) = J..lo L crC;+q,crCpcr (5.279)
pcr
(a)
'0'
,o-
(d) 5
'<2=>,'
s'
( b)
+
s~s'
sCDS'
(e)
(c)
FIGURE 5.20
The several connected tenus which result from this contribution are shown in Fig. 5.20. The
free-electron tenu is given by the bubble diagram shown in Fig. 5.20(a). The vertices are
labeled s to indicate that they contain a spin operator and so are not the same vertex as found
for the density or current operators. Figure 5.20(b) contains the exchange self-energy
diagrams for the electron Green's function. These can be incorporated by using '!J instead of
'!J(O) for this propagator, where '!J contains the self-energy tenus of the electron gas. Figure
5.20(c) has a Coulomb line connecting the two propagators. It is called a vertex correction. It
is actually a type of exchange scattering, which contributes to the local field correction Gs(q)
of Hubbard and Singwi-Sjolander. Figure 5.20(d) has two closed loops, connected by a
Coulomb line. This diagram is zero. It vanishes because each bubble has the correlation
function of an M(f, operator and a density operator
(5.280)
This combination is zero because of the averaging over cr = ±1. All diagrams with more than
one bubble are zero, because the Coulomb interaction is spin independent. Therefore the
correlation function is limited to tenus with a single bubble, which has an s vertex on each
end. The sum of all such contributions is indicated schematically in Fig. 5.20(e). This bubble
diagram is called P(q,O)). It is not the same as the exact P(q, 0)) which enters into the
definition of the longitudinal dielectric function, since the latter may have contributions from
more than one bubble if connected by two or more Coulomb lines.
One possible approximation for P is to use the Hubbard or Singwi-Sjolander theories:
(5.281)
As discussed in Sec. 5.5, the local field correction for spin Gs(q) is different than the one
GH(q) for charge fluctuations. For the static susceptibility 'Xzz == X, take the limit iO) = 0 and
then the limit q -+ O. In this limit p(1) goes to minus the density of states NF =
mkF/n2, 'XF == 1l5N F:
(5.282)
In this approximation, the ratio XIxF is similar to the ratio of the compressibilities K 1Kf
which was found in Sec. 5.6.4. However, the spin parameter as is not the same as the charge
parameter.
The local field correction for spin Gs(q) has also been extensively investigated. Its exact
value is known in the limit of large (q, (0) from Santoro and Giuliani (1988), and Vignale
(1988). These results are presented in Table 5.3. They were derived in the same fashion as the
similar results for charge fluctuations.
5.7. SUM RULES
The longitudinal dielectric function E(q, (0) of the homogeneous electron gas has some
exact moments which are useful for checking results. Some of these have already been used
and mentioned. For example, the long-wavelength limit is limq~o E(q, (0) = 1 + Vqn6K,
where K is the isothermal compressibility. Similarly, the high-frequency limit is
(02
lim E(q, (0)
(O~oo
= I ---1
(0
(5.283)
where (Op is the plasma frequency. The frequency (0 is large when it is larger than other
energies in the system, such as the plasma frequency or the Fermi energy.
Another class of exact results are called sum rules. The "sums" are actually integrals
over frequency. Several of them are
J~ d(o(o E 2(q, (0) = ~(O; (5.284)
lim
q~O
J -Ez(q,
d(O
oo
0 (0 2
1t
(0) = -(vqnoK) (5.285)
J~ d(O(O Im[E(q~ (O)J = -~(O~ (5.286)
!~J~d:Im[E(q~(O)J =-~ (5.287)
These last two identities are exact results. The validity of the first two are debated. The last
two are rigorous since there is a density-density correlation function for the inverse of the
dielectric function, which means that its retarded function is absolutely causal. However, there
is no correlation function which defines E(q, (0). Its retarded function is not required to be
causal. However, it is hard to find examples where it is not.
Approximate formulas for E(q, (0) may not automatically obey these relationships.
Indeed, approximate formulas are considered virtuous according to how accurately they
satisfy these identities. The sign convention has E2(q, (0) > 0 for (0 > 0, which implies that
Im(l/E) = -E2/1E12 is a negative quantity for (0 > O. The last two sum rules may also be
written as
Jo d(O(O-2--2
OO E2
E\ + E2
= -(Op
2
1t 2
(5.288)
limfood(O ~=~ (5.289)

q~O 0 (0 E\ + E2 2
Sec.5.7 • Sum Rules 359
These exact results may be proved by a variety of methods. The third relation (S.286) is often
called the longitudinal [sum rule. It may be proved from the double commutator
C = ([[H, p(q)], p(-q)]) (S.290)
where the average is taken, at nonzero temperature, over the thermodynamic states of the
system. The Hamiltonian H is for the full homogeneous electron gas. However, the density
operator p(q) commutes with all terms in this Hamiltonian except the kinetic energy. The first
commutator was evaluated in Sec. S.S.4.
(S.291)
The term proportional to p(q) will commute with p( -q). Only the other term needs to be
further evaluated,
C = ([[H, p(q)], p(-q]) = !(Eq 'p[C:+q,crCpcr' P(-q)])

m pcr
1
= -m E q . p{C:+q,crCp+q,cr - C~crCpcr)
pcr
1 q2 q2
=-"{ct pcr Cpcr )[(p-q)·q-p·q]=--"n m L..." p =--N
m e (S.293)
m L..."
pcr pcr
and yields the simple result that this double commutator is just -Ne q2Im. Next, this double
commutator is shown to be proportional to the sum in (S.286). It is evaluated by inserting the
complete sets of states In), 1m), which are exact eigenstates of the Hamiltonian, and then
collecting terms:
C = ([[H, p(q)], p( -q)]) = E{e-~En (nl[H, p(q)]lm){mlp( -q)ln)

nm
- e-~Em (mlp( -q)ln) {I[H, p(q)]lm}}
= E l{nlp(q)lm)12(e-~En - e-~Em)(En - Em) (S.294)
nm
Earlier in (S.IIS) it was shown that
Im[-(1 )] = -n(1- e-~ro)

c q, 00
Vq
V
Ee-~Enl{nlp(q)lm)128(oo +En -
nm
Em)
so that the sum rule is
Joo
-00
doooo 1m[_I_Jc(q,
(0)
= -n Vq E I(nlp(q)lm) 12 (En -
v nm
Em)(e-~En - e-~Em)
Compare the sum rule with the double commutator, and note that they are proportional, which
provides the derivation of the sum rule (S.286):
Joo
-00
doooo 1m[1]
--(0)
c(q,
nvq C = - _
= -
v
4nen
m
_0 = -noo; (S.29S)
The integral is from -00 to 00, but the negative and positive parts contribute equally, so the
left-hand side is twice the integral from 0 to 00.
The other three sum rules may be derived by using the Kramers-Kronig relations
(Kronig, 1926; Kramers, 1927). Let B(m) be a function of complex variable m, which is
analytic in the upper half plane. Analyticity means that it has no poles or branch cuts above
the real axis. It is also assumed that B(m) -+ 0 as Iml -+ 00.
1
B(m) = -:- Joo 1
dt' B(m')P-- (5.296)
l1t -00 m' - m
The integral is along the real axis. On the real axis, the function B( m) has real and imaginary
parts, which are called B j and B z, so that B(m) = B j (m) + iB2(m). They are related by taking
the real and imaginary parts of (5.296):
1
B j (m) = -
Joo dt,B2 (m,) P 1- - (5.297)
1t -00 m' - m
B2 (m) = - -1 Ji ") 1-
nftydt Bj(m P - (5.298)
1t -00 m' - m
These two identities are the Kramers-Kronig relations. They are useful throughout all areas of
physics. The present interest is in applying them to the real and imaginary parts of the
dielectric function c(q, m) = Cj + iC2' Another application is to define the refractive index
n( m) and extinction coefficient k( m), where n = n + ik = -JE. The complex refractive index n
obeys the conditions of the theorem and hence the preceding identities. There are other
applications in optics, scattering theory, and virtually all branches of physics. All of these
applications, including the present sum rules, assume that the dielectric function is causal so
has no poles in the upper half plane of frequency. This assumption is not firmly established for
E but is for I/E.
The other sum rules are easy to prove as simple applications of the Kramers-Kronig
relations. First take the case where B(m) = I/E(q, m) - 1. The -1 term is included to make B
vanish as Iml -+ 00. Using the first Kramers-Kronig relation gives
Re [ -1 -] - 1 = -1 Joo , [
dm 1m - -1-] P - 1- (5.299)
E(q, m) 1t -00 E(q, m') m' - m
The right-hand side may be simplified by using the fact that Im[l/c(q, m')] is asymmetric in
frequency. The integrand (-00,0) is evaluated by changing m -+ -m and combining this
with the (0,00) part to give
[ 1]
E(q, m)
2JOO dm'lm - - - P---,----"
Re - - = 1 +-
1t 0 E(q, m')
m'
m'2 - m2
[1] (5.300)
This identity is also a sum rule but not one of the simple ones in (5.284)--(5.287). However,
they are special cases of this result. First take the limit where m -+ 0:
1 2 foo dm' [ 1 ] (5.301)

E(q) = 1 + 1t 0 ffiI 1m E(q, m')
Sec.5.7 • Sum Rules 361
The quantity E(q, co = 0) == E(q) is real, so (5.301) expresses the static dielectric function as
an integral over the loss function. The sum rule (5.287) is obtained by taking the limit where
q ---+ O. The quantity E(q) is evaluated from the compressibility sum rule (5.266),
lim E(q) = 1 + vqnoK ---+ 0(-;) ---+ 00 (5.302)

q~O q
lim _1_ = lim q2 =0 (5.303)

q~O E(q) q~O q2 + 41te2n oK
which shows that the left-hand side of (5.301) is zero. The terms on the right give
lim
q~O 0
Joo dco'
co'
Im[_l_] - - ~
E(q, co') - 2
(5.304)
which is (5.287).
The longitudinal/-sum rule was proved by taking commutators. The remaining two sum
rules, (5.284) and (5.285), are obtained from the Kramers-Kronig relations using E - 1 for
B(co). Again the 1 term is included so that E - 1 vanishes at large values of lcol. From (5.298)
EJ(q, co) - 1 = - IJoo '(')

dco E2 q, co P-,-- 1 (5.305)
1t -00 co - co
EJ(q, co) = 1 +-
2 Joo dCO'E2(q,co')P,2
co'
2 (5.306)
1t 0 co -co
The second equation here is derived using the fact that E2(q, co) is anti symmetric in frequency.
This equation is evaluated for co = 0:
EJ (q)
2
= 1+-
Joo -,
dco'
E2(q, co') (5.307)
1t 0 co
Equation (5.307) is an exact relation between the imaginary part of the dielectric function
Eiq, co) and the static part EJ (q). In the limit where q ---+ 0, the compressibility sum rule
(5.302) gives EJ = 1 + vqnijK, so that
. Joo -Eiq,
lIm
dco' , 1t 2
co) = -2 vqnoK (5.308)
q~O 0 co'
which establishes (5.285).
The remaining sum rule (5.284) is found from the limit co ---+ 00 in (5.306):
lim E2(q, co) = 1 - -;'Joo dCO'CO'E2(q, co') + o(~) (5.309)

ro~oo 1tCO 0 co
By using (5.283), the left-hand side becomes 1 - co;/co2 , and equating terms of order co- 2
produces the result
Jooo"co
dco f.2(q, co ) =
, 2"1t2
cop (5.310)
All four sums rules have been demonstrated. These are not the only sum rules which exist.
Others are sometimes found useful, in particular the third moment (Mihara and Puff, 1968;
Hopfield, 1970).
It is curious that the first moment sum rules of both f.2(q, co) and Im[l/E(q, co)] yield the
same result (1t/2)co;. And this result is independent of q. One should not conclude that E2 and
Im( I / E) are identical. Earlier, in Fig. 5.11, it was shown that these two functions are similar at
large q but are quite different at small values of q/kF .
Plasmons playa particularly important role in these sum rules at small values of q. In the
limit where q --+ 0, from (5.170) then
002
lim E(q, (0) = I - ~ + 0(00- 4) (5.311)
00
[I
q~O
lim - - - I =
q~O E(q, (0)
] 00
2
002 - oo~
-I =
002
p
002 - oo~
(5.312)
_OOp(_1 _ _ I ) (5.313)
- 2 00 - OOp 00 - OOp
For a retarded function, let 00 --+ 00 + iO, so that the imaginary part of this expression gives
lim
q~O
Im[_I_]
E(q, (0)
= - noop
2
[0(00 - 00 ) - 0(00 - 00 )]
p p
(5.314)
This relation is rigorously valid only in the limit where q --+ O. Nevertheless, it is curious that
it satisfies the last two sum rules exactly:
lim
q~O
Joo0 oodoo Im[-(I )] =
E q, 00
-::200; (5.315)
~~J~d:Im[E(q:oo)] =-i (5.316)
This result shows that plasmons provide all the contributions to Im[I/E(q, (0)] in the limit
where q --+ o. Since the interpretation of Im[I/E(q, (0)] is that it describes longitudinal
excitations of the electron gas, plasmons are the only excitation in the limit where q --+ o.
Any other excitation would make a contribution to the sum rule, which is impossible since
plasmons provide the entire result. Any other excitation has zero strength in the limit where
q --+ o.
The single-mode approximation is to assume the spectral function is given by a single
excitation oo/q). The approximation is only accurate in the electron gas at small values of
wave vector q, but is often used for all wave vectors. It will be used again in Chapter 11 in
describing the excitations in correlated liquids.
5.8. ONE·ELECTRON PROPERTIES
This section discusses the single-particle properties of electrons in the homogeneous

electron gas. They include the effective mass and mean free path. The present section is put at
the end of this chapter deliberately to make a point: the one-electron properties are not very
important for discussions of most features of the electron gas. The discussions of ground state
energies and pair correlations did not rely significantly on the properties of the individual
electrons. The excitation spectra of the electron gas are also not very dependent on the
properties of single electrons. In fact, the converse is true: the collective properties of the
electron gas need to be known in order to discuss the behavior of a single electron. This single
electron interacts with the other electrons, which are represented as collective excitations. The
Sec. 5.8 • One-Electron Properties 363
order of presentation has been to first describe the collective excitations. It was done assuming
that the particles were nearly free (i.e., RPA) or had some correlation (i.e., Hubbard, Singwi-
Sjolander). These calculations ignored the finite mean free path, or possible effective mass
changes, of the electron. They do not change these collective properties to a significant
degree, as long as the mass changes are small, and the mean free path is long.
The original assumption is that electron states in the electron gas are noninteracting
plane waves, and the final description is not very different. The self-energy effects do not alter
this basic picture.
If the calculations were to show that something drastic happened to the individual
electrons-that they were magnetic or superconducting or semiconducting-obviously the
collective properties would be altered in a significant fashion. It is important that the initial
eigenstates are an approximate description of the actual physical system. It is only in this
circumstance that one finds that the one-particle self-energies have little influence on
collective properties.
The one-electron properties are obtained from a calculation of the electron self-energy
due to electron--electron interactions. This self-energy was discussed earlier, in Sec. 5.1, in the
discussion of electron correlation. If we had chosen to follow the original discussion of Gell-
Mann and Brueckner (1957), the correlation energy could be derived without first discussing
the electron self-energies. The point is made again that the collective properties are not
significantly dependent on first deriving the one-electron properties. The discussion of
electron correlation followed Quinn and Ferrell (1958) and first considered the electron self-
energy. The results of that discussion can be used as the starting point for the present deri-
vations. The best calculations which have been reported of single-electron properties calculate
only the screened exchange interaction.
The electron self-energy from the screened exchange energy is
(5.317)
(5.318)
(5.319)
The integrand contains a vertex function qq. iqm)' as discussed in Sec. 5.5. Calculations
using (5.317) are shown in Fig. 5.21. Results are obtained by a numerical evaluation of the
multidimensional integral. Figure 5.21(a) shows the real part of the self-energy ~sx as a
function of k for rs values appropriate to common metals. All of the curves have a similar
shape. The strong dip in the curve is due to the coupling of the electron to the plasmon. The
imaginary self-energy Im[~sx] has a big jump in magnitude when the electron has enough
kinetic energy to emit a plasmon. The Kramers-Kronig relation between Re[~] and Im[~]
guarantees that Re[~] has a dip when Im[~] has a jump.
A similar dip in the self-energy is obtained whenever the electron couples strongly to an
oscillator. The self-energy of electrons coupled to LO phonons in insulators has the same
shape, as discussed in Problem 7.7. Of course, there the dip happens at the energy char-
acteristic of the phonon, while here it occurs at the energy of the plasmon.
a -4.0 K(4.86)
:>
Q)
-6.0
~
c: -8.0~---L
b 1.0 r---~-"--""T"--r---'----'''''
I
/--- ...... ...... ,
I ........... r= 1
I --
I
I
:> I
~
(!J I I
Z I I
~ , I
~ 0.0 , I
c: , I
<I:
Z ,I
Cl ,I
z ,I
<I:
,,
co ,I
,I
'I
,:'
-1 .0
,
I
I--_-'--_~_--L._--' _ _'___..J.......J
o 2 3 4 5
FIGURE 5.21 (a) The real part of the screened exchange energy I:",,(k, ~k) as a function of k for values of rs
appropriate for simple metals. The dip is caused by plasmon emission. Source: Shung et al. (1987). (b) The change in
energy bandwidth for electrons interacting with electron-electron interactions. The RPA result (G = 0) is compared
to the GH(q) ofVashishta and Singwi. The results are quite similar. Source: Mahan and Semelius (1989).
The electron self-energy also contributes to the energy width of the occupied band of
electrons. This change in width can be crudely estimated as the difference between the self-
energy calculated at k = 0 and at k = kF •
(5.320)
The energy ~ = 0 at the Fermi energy, and equals -EF at the bottom of the band. Figure
5.21(b) compares AW calculated using RPA and the Vashishta-Singwi (VS) form for GH(q)
given in Table 5.2. The results are quite similar. The local field factor of GH(q) has only a
small influence upon the bandwidth of the electrons. The curve labeled r = 1 uses GH(q) in
c(q, 0)) but not in r. This inconsistent theory is called the "GW" approximation.
5.S.1. Renormalization Constant ZF

The renormalization coefficient of the one-electron Green's function is defined for a
point (k, c = ~k) as [see (3.155)]
Z(k) = 1 (5.321)
[1 - aLsx(k, C)/aC]E=~k
Usually the retarded self-energy Lsx is complex, as is the renormalization coefficient. At the
particular point, k = kF and ~ = 0, the self-energy Lsx is exactly real, since Im(L) vanishes.
Electrons on the Fermi energy have an infinitely long mean free path, or lifetime. They have
no unoccupied electron states into which they might scatter. The renormalization coefficient is
real at this point. At the Fermi energy it is called ZF = Z(kF)
1
ZF = c:-c--:-=----:-O---:---c--::-- (5.322)
[1 - aLsxCkF' c)/ac]E=O
The renormalization coefficient is interpreted as the amount of single-particle behavior of the
particle-like excitation in the electron gas. Since the excitations behave like particles, they are
called quasiparticles. Generally the spectral function A(k, c) has a central peak plus some
smooth background wings. This behavior is illustrated schematically in Fig. 5.22. The
renormalization coefficient is defined as the area under the central peak of the spectral
function. If the peak is broad, then the definition is ambiguous, since the limits of the central
peak are ill-defined. Precisely at the Fermi surface the quasiparticle central peak is a sharp
delta function, because 1m L = O. Here the renormalization coefficient has a precise meaning:
the spectral weight under the delta function peak. As discussed in Sec. 3.3, the renormali-
zation coefficient ZF must be less than or equal to unity since the total area under A(k, c) is
unity.
The quantity ZF has an additional interpretation, which can be measured experimentally.
The momentum distribution of the electron gas is given by [see (3.135)]
nk = Jde
2n A (k, c)nF(c) (5.323)
Actual results for an interacting electron gas have the form illustrated by the solid line in Fig.
5.23 (Daniel and Vosko, 1960). The dashed line is for a noninteracting Fermi system, which
has all the particles below k = kF at T = O. For the interacting Fermi system, even at T = 0,
A(p,w)
FIGURE 5.22 Spectral function of A.

.2r-- _ _ _ _ _ _ _ _ _ _- - .
1.0-- - ---I
,
I
0.8-
Np 0.6
0.4
0.2 -
°0~----~1-~~---*2-~
i~
p/,tF
FIGURE 5.23 Momentum density of the homogeneous electron gas for r, = 3.97. Source: Daniel and Vosko (1960)
(used with permission).
the momentum distribution is the solid line, with some components for k > kF . The quantity
nk should be distinguished from the energy distribution nF(f.) = l/(e~€ + 1), which is a sharp
step at zero temperature. The renormalization coefficient ZF is the magnitude of the step at
k = kF in the momentum distribution nk' This assertion is easily shown using the definition of
nk and the fact that
(5.324)
wheref(f.) is a smooth function. The step in the momentum distribution in the Fermi gas may
be measured in actual metals by Compton scattering.
There have been two major types of calculations with these formulas. The first is the
RPA, where we use Hedin's (1965) results. The other is the revised Hubbard model work of
Rice (1965). The two calculations of ZF are shown in Table 5.4 for different values ofrs . Rice
only reported values of rs < 4. The two calculations are in very good agreement. The two
models of the dielectric function make the same prediction regarding this quantity.
Eisenberger et at. (1972) reported Compton scattering experiments in metallic sodium
and lithium which demonstrated the existence of the high momentum tail in nk' Compton
scattering is the inelastic scattering of photons by the electrons in the solid. The photons are
scattered by all the electrons, both in core states and the conduction band. The conduction
band profiles, which are our present interest, are obtained after subtracting the core scattering.
TABLE 5.4 ZF as calculated by (a)

Hedin (1965) and (b) Rice (1965)
rs RPAa Hubbardb
0
1 0.859 0.87
2 0.768 0.77
3 0.700 0.70
4 0.646 0.63
5 0.602
6 0.568
The best results are obtained for metals with low atomic number, since they have fewer core
levels to subtract. Eisenberger et al. analyzed their data in terms of the sudden approximation.
This theory assumes that the scattering rate of photons is proportional to
d 2 cr
d rod nu ()(
Jd p npo(ro +
3
--3 Ep - Ep +q ) (5.325)
(2n)
where ro is the energy transferred by the photon and q is the momentum transferred. The
electrons are scattered from p to p + q, and the average is taken over the initial distributions
np of the interacting electron system. When the angular integral is done to eliminate the delta
function, it is found that the experimental quantity is only a function of the combination of
variables called Q = (m/q)(Eq - ro):
J-1 dVO(ro -
1 E
q
_pqV) =
m
~8(p -IQI)
pq
d 2 cr
drodQ ()( J(Q) (5.326)
J(Q) = Joo pdpnp (5.327)

IQI
For a noninteracting electron gas, one has that np = 8(kF - p) and the scattering function
J(Q) is an inverted parabola:
J(Q) = !(~ - ~)8(kF - IQI) (5.328)
Thus J(Q) is shown as the dashed line in Fig. 5.24(a). The solid line shows the type of
behavior expected for an interacting electron gas. There is a long tail at high values of Q,
which arises from the similar high tail in np for values of p > kF .
The experimental and theoretical results of Pandey and Lam (1973) are shown in Fig.
5.24(b). The triangular points are the experiment, and they are joined by the solid line. The
1.5
(b)
1.2
&--c. No experiment
(a)
• Theory (present)
1- 0.9
:;
.!
~ 0.6
":)
0.3
INTERACTING
x
\ 1
0
"F Q
1.2
FIGURE 5.24 (a) The dashed line shows the momentum distribution in Compton scattering from a noninteracting
electron gas, and the solid line is the prediction for an interacting system. (b) The experimental results in metallic
sodium (solid points) compared with the theory (open triangles) for this metal. Source: Pandey and Lam (1973) (with
permission).
circular points are the theory. The agreement is obviously excellent. Sodium is the ideal metal
for testing the distribution nk since its Fermi surface is quite spherical. Other experimentalists
(Eisenberger et al., 1972) find that the Compton scattering depends only upon Q, which can
be tested by different choices of q and 0).
5.8.2. Effective Mass

The formula for the effective mass was given previously in (3.160). It will be evaluated
for electrons on the Fermi surface of the metal:
(.!!!....)
m*
= 11 + oI:.(k, E)/OEk I
1 - oI:.(k, E)/OE k=kF,e=O
= ZF [1 + OI:.(k,OEk E)] k=kF,e=O (5.329)
The results of Rice were obtained by a slightly different formula, which should give similar
results. His values are shown in Table 5.5. The ratio m/m* is less than unity for values of rs
less than 2, while it is slightly larger than unity for larger values of rs. The effective masses of
electrons on the Fermi surface of the metal may be measured quite accurately by many types
of cyclotron resonance experiments. The effective mass values for real metals are determined
by three major contributions. The first is the electron-electron interactions, and they make
only a small contribution to the mass. The second is due to band structure, which is quite
variable. It is small in sodium but large in lithium and many other metals. The third
contribution is due to electron-phonon interactions, which is quite sizable. A comparison
between theory and experiment is deferred until the phonon contribution is calculated in Sec.
7.4.
Another application of the effective mass theory is in the low-temperature specific heat,
or heat capacity. The free-electron theory of metals predicts that the specific heat of a metal, at
low temperature, is linear in temperature. A free-electron gas of Ne particles has the result
C - 31t2 N JdT m ( )
vO - 4 e B (31t2noi/3/i2 5.330
For an interacting system, the prediction is that the free-electron mass is replaced by the
effective mass at the Fermi energy m*. The ratio of the linear term in the specific heat, at low
temperature, is just the ratio of the effective masses:
Cv m*
(5.331)
Cvo m
The quantity Cvo/Cv is calculated in a similar manner as m/m*. This ratio is also influenced
by phonon effects.
TABLE 5.5 Results of Rice

(1965) for m/ m * using the
Hubbard dielectric function
rs Hubbard
o 1
0.96
2 0.99
3 1.02
4 1.06
5.S.3. Mean-Free-Path
The mean-free-path Ik is the average distance the electron travels between scattering
events. Similarly, the lifetime 't'k is the average time between scattering events and is related
by Ik = vk't'k' In the electron gas, most scattering processes are inelastic, and the electron gains
or loses energy each time it scatters. These quantities are defined in terms of the imaginary
part of the self-energy:
(5.332)
(5.333)
Most calculations have been done with the screened exchange energy in (5.56). The rate of
scattering is proportional to Seq, co) = Im[l/E(q, co)]. The scattering has an energy transfer
co = ~k+q - ~k' where the initial state has ~k > 0 and the final state ~k+q > 0 is empty. A
simple expression for the mean-free-path is
~=
Ik
_2..J
fzvk
d 3q3 vq[1-
(21t)
nF(~k+q)]Im[rH(q, ~k+q - ~k)]
E(q, ~k+q - ~k)
(5.334)
This formula has formed the basis for most treatments of electron scattering in metals. The
expression is expected once Im[l/E(q, co)] is identified as the rate of making excitations (q, co)
in the solid.
For small values of initial energy ~k it is necessary to numerically integrate this
expression. It is a double integral, since one must do both the dq and dv (v = cos 9) inte-
grations. This calculation has been done by Lundqvist (1969). For more energetic particles,
the excitation spectrum of Im[l/E(q, co)] is dominated by long-wavelength plasmons. One
way to evaluate this expression approximately is to assume that the plasmons totally dominate
the excitation spectrum. For small q use the limiting form given in (5.311H5.314).
An analytical result may be derived for the mean-free-path. Insert the delta function for
Im(l/E) into the integrand and change integration variables to p = k + q:
I ~co
-I =--?e(~k - cop)
Joo
p 2dp8(Ek - cop - Ep)
J1 dv k2
1
2 k (5.335)
k nVk kp -1 +p 2 - p v
The factor e(~k - cop) arises from energy conservation ~k - co = ~k+q > O. The integrals
may be evaluated:
~ = _1_ (COp)e(~k _ COp)ln[Ft + ,JEk cop] (5.336)

Ik 2ao Ek ,.ffl -JEk cop
The formula is given in terms of the kinetic energy Ek measured from the bottom of the
conduction band, although plasmons are emitted only when ~k > cop or Ek > cop + EF'
Equation (5.336) is almost identical to an old formula of Bethe (1930) for the rate of energy
loss of charged particles in solids.
Figure 5.25 shows a comparison of these theories with the experiments in aluminum.
The metal aluminum is chosen as a basis of comparison because the best experimental results
are available of any free-electron metal. The data is due to Tracy (1974). The plasmon
formula (5.3 36) only needs the aluminum plasmon energy offz cop = 15.5 eV to determine the
dashed line. The plasmon theory in Fig. 5.25 is too low by about 30-40 %. The electron
mean-free-path is longer than would be predicted by the simple plasmon formula. The
I
I
I
ALUMINUM -
o Experiment: Tracy oo I
I "
I
0<1: I
I
I
UJ 0 I
""-0\ 20 _ "
~ 00 I
~ 0 ,
"" 01
a: I
UJ
UJ I
'
~ I
u.. 0 I
I
~ 10 - I
~ x R~,/' ,,/ "PLASMON
\ \.t)',/ THEORY
, ae x -e ",/
x........... e-x....E'f<!J__ - ... /
------ ---.1 ,I
10 100 1000
E (eV)
FIGURE 5.25 The mean-free-path of electrons in aluminum as a function of energy above the Fermi surface. Circles
are data points from Tracy (1974). The dashed line with x points is the RPA calculation of Lundqvist (1969). The
other dashed line is the plasmon theory which ignores pair production, and its predictions are inaccurate.
plasmon theory is unreliable at lower energies. Experiments show that the rate of making
plasmons vanishes below 40 eV (Flodstrom et al., 1977). The other theoretical curve is the
RPA result, which used the calculations of Lundqvist (1969) for Ys = 2, which is close to the
aluminum value of Ys = 2.07. He numerically evaluated (5.334). His curve is in excellent
agreement with the experimental results. Similar measurements have been done on other
metals, although few on free-electron metals. See the reviews by Echenique et al. (2000).
Another experiment is to measure the inelastic scattering of electrons by thin metal foils.
The electrons are shot through the foil, and their cross section for inelastic scattering is
measured as a function of angle and energy. The electrons must be fairly energetic to
penetrate the foil. The experiments show that the electrons mostly lose energy by exciting
long-wavelength plasmons (Marton et al., 1962).
First calculate the angular dependence of the scattering when the electron emits one
long-wavelength plasmon. The process is that the energetic electron, usually in the kilovolt
range, loses one plasmon of energy and is scattered through an angle e. The energetic electron
scatters from k to p. The scattering rate may be obtained from (5.335) by eliminating the
angular integrals. The result is the differential scattering rate as a function of angle:
(5.337)
The delta function determines the value of p. For energetic electrons, this momentum is
approximately
(5.338)
The typical scattering angle is very small, and it is possible to use the small-angle approx-
imation for cos 9 = 1 - 92 /2. The denominator can be simplified to
J?- + p2 - 2pk cos(9) ~ (k - p)2 + kp9 2 ~ J?-(9~ + 92 ) (5.339)

92 _ wp
0- 2s k (5.340)
and the cross section has the simple form derived by Ritchie (1957):
dw i2wp
(5.341)
dO. = 21tVk 9~ + 92
(a)
o 2 3 4 !I 6 7 a
ENERGY LOSS, UNITS OF 6E
NO. OF ENERGY LOSSES N

e
FIGURE 5.26 Energy loss spectrum at = 0 for electrons going through aluminum films. The relative peak heights
are shown in (b) and are well fit by a Poisson distribution shown by the solid line. Source: Marton et al. (1962) (used
with permission).
The scattering distribution is Lorentzian. The angular width 90 is typically a fraction of a

degree when ek is a kilovolt. The angular distribution is sharply peaked in the forward
direction. This behavior is found experimentally (Sueoka, 1965).
Another aspect of electron energy loss through metal foils is shown in Fig. 5.26, which is
the data of Marton et al. (1962) for aluminum. The peak on the left of Fig. 5.26(a) is from the
electrons which go through the foil with no energy loss. The next peak is from those which
excite one plasmon, and the nth is from electrons which excite n - 1 plasmons. The angular
dependence of these peaks was measured and found to be Lorentzian. Another feature of
these peaks is shown in Fig. 5.26(b), which is the peak intensity, or the area under each curve.
The background is subtracted according to the dashed line in Fig. 5.26(a). The peak inten-
sities follow a Poisson distribution (solid line), so that the intensity of the nth peak is given by
d
0(=- (5.342)
I
The parameter 0( is the thickness of the film d divided by the mean-free-path for plasmon
emission. The Poisson distribution is a characteristic of the independent boson model of Sec.
4.3. A Poisson distribution occurs whenever the bosons, in this case plasmons, are emitted
independently of previous emissions. This model is reasonable for these fast electrons. The
electron energy was 20 keY. An electron of this energy does not lose a significant fraction of
its energy when it emits a plasmon of 15 eY. Similarly, its angle does not change appreciably,
since the scattering is mostly in the forward direction. The electron trajectory is largely
unchanged by the plasmon emissions. All emission events have the same likelihood and are
independent of previous emissions. It makes sense that the independent boson model should
be applicable here. Of course the zero-temperature version of the theory is used, since even at
room temperatures, the number of plasmons thermally present 1/ (ef3 oop - 1) is negligible.
Problems
1. Consider the ground state of the ferromagnetic electron gas. The particles are plane waves, but all
spins point in the same direction.
(a) Find the ground state energy of the system in the Hartree-Fock approximation.
(b) Compare this with the paramagnetic state (equal numbers of up and down spins) calculated in
the same approximation. Are there values of rs for which the ferromagnetic state is lower in
energy?
2. In the Wigner lattice the electron feels a harmonic potential VCr) = -3/rs +,-2 /r;. Find the zero-
point energy from oscillations in this well by assuming that each electron moves independently of the
others.
3. The correlation energy may also be evaluated by Rayleigh-Schrodinger perturbation theory (Gell-
Mann and Brueckner, 1957). The second-order ground state energy is
(5.343)
Obtain the ground state energy terms which contribute in second order, and write them down in terms of
wave vector summations.
Problems 373
4. Calculate the equilibrium value of rso for metallic hydrogen using NP correlation. Also calculate the
ground state energy per electron.
5. Calculate the cohesive energy of metallic helium assuming it is a uniform electron gas, with IX
particles spaced on a crystal lattice. Compare your result to atomic helium.
6. Consider putting a single point charge Q into a homogeneous electron gas. Show that to order Q2
the energy required to do this is
(5.344)
Estimate this result for the Thomas-Fermi model.
7. Use (5.117) to estimate S(q) at large q in the RPA. At large q, one has that Im(l/E) ~ -E2.
8. Use (5.117) to find the behavior of S(q) as q ~ O.
9. Set l-g(r) =Aexp(-rkF).

(a) Determine A by the normalization condition (5.71).
(b) Determine S(q).
(c) Determine G(q). Compare this on a piece of graph paper with Hubbard and Singwi-Stilander
for rs = 3.
10. Calculate the compressibility of the electron gas from the ground state energy:
(a) in the Hartree-Fock approximation.
(b) including both HF and Wigner correlation energy.
Plot both results on a piece of graph paper as Kf / K vs. rs for metallic densities.
11. Start from the Singwi-SjOlander expression (5.198) for GH(q). Take the limit that q ~ 00 and find
their prediction for this limit.
12. Calculate S(q) in the Hartree-Fock approximation. First show that the starting point is S(q) =
1 - (2/N) Lk nknk+q·
13. Calculate y(rs) in the Hartree-Fock approximation using the results of Problem 12. Verify that the
correlation energy (5.249) is zero with this choice of y(rs).
14. Use the form of the dielectric function (5.187) to show that the plasma dispersion relation is
miq) = miO) + IX (11!2) + o(l) (5.345)
IX = 130:~ [1-~a(~:y] (5.346)
where a is the parameter which enters into the compressibility (5.270).
15. Prove the ground state energy theorem (5.242) starting from (3.297) using just the electron-
electron interaction.
16. In the Vashishta-Singwi theory, the constant a in the compressibility is no longer equal to y as
given by (5.244). Find the new relationship to y.
17. Does the RPA satisfY the sum rule (5.307)? Show by explicitly doing the integrations.
18. Calculate ez(q, (0) in RPA for a classical gas which obeys Maxwell-Boltzmann statistics.
19. Derive the sum rule for the third moment for a Hamiltonian of particles in a potential field VCr).
Do this by evaluating the expectation of the triple commutator with H:
H = Lpu SPC:uCpu + Lq p(q)V(q) (5.347)
(003 ) = ([[H, [H, [H, p(q)]]] , p( -q)]) (5.348)
20. Prove that the step in nk at k = kF is ZF'
21. The screened exchange energy (5.317) may be calculated exactly in the Thomas-Fermi approx-
imation. Do this, and use your answer to obtain analytical expressions for
(a) the self-energy at p = O.
(b) the self-energy at p = kF •
(c) the effective mass at the Fermi energy.
Plot all three results as a function of rs on a piece of graph paper.
22. Calculate e(q, (0) in RPA for the two-dimensional electron gas at zero temperature. Show that the
plasmon dispersion can be given by an analytical expression.
23. Calculate the average kinetic and exchange energies per electron for a two-dimensional electron
gas with Coulomb interactions eZ /r. Define rs by 1tnoa6"; = 1, and your answer should be
Eg = 1/"; - 1.20/rs .••• Find the analytical form of the coefficient of the exchange energy (it is not
6/5). For a review of correlation calculations, see Jonson (1976).
24. In two dimensions show that the exchange energy at the Fermi surface is ~x = -2eZkF /1t. Use this
result to calculate the compressibility in the Hartree-Fock approximation. At what density rs does the
compressibility go negative in two dimensions?
25. Draw the connected Feynman diagrams for the electron self-energy from electron--electron
interactions that contain two or three internal Coulomb lines. Show that all terms can be written as
vqr/e in (5.317). Which diagrams contribute to r and which to e?
Chapter 6
Strong Correlations
The previous chapter discussed metals in which correlation plays a minor role in the
dynamics of the electron. The properties of the electron gas could be calculated simply and
accurately. The present chapter discusses other metallic systems in which correlation plays a
much larger role in the dynamics of the electron. These topics come under the title of strong
correlation. Strong correlation causes itinerant magnetism.
The theory of magnetism in metals continues to be one of the challenging subjects of
modem theoretical physics. Magnetic phenomena are important, and are observed in a wide
variety of materials. The equations are difficult to solve. A variety of theoretical techniques
are applied to magnetic phenomena, such as Monte Carlo, renormalization group, and Green's
functions. Only the latter technique is discussed here.
Strong correlation is usually treated by solving the Anderson model or the Hubbard
model. They are discussed here, along with the Kondo model (1964) which was the historical
precedent. Most Green's function analyses have focused on these three model Hamiltonians.
Each of these three models is not a single mathematical problem. Instead, each has a number
of variations. For example, the original Kondo model describes a system of conduction
electrons interacting with a single localized spin. The local spin can have any value of angular
momentum S, and the conduction electrons can be treated in one, two, or three dimensions.
Recent variations include the dense or periodic Kondo problem, where there are N local spins
arranged on a lattice. Another feature of all three models is that they can be solved exactly in
one dimension, using the Bethe Ansatz, but not in higher dimension. One-dimensional results
are not a useful guide to collective effects in higher dimension, since there are neither phase
transitions nor long-range order in one dimension at nonzero temperatures.
6.1. KONDO MODEL
For many years experimentalists noticed that magnetic impurities in nonmagnetic metals
caused anomalous behavior in the low-temperature resistivity p(T). Magnetic impurities are
those with a moment caused by partially filled d- or f-electron shells. An example is
manganese impurities in copper. The electron scattering from a nonmagnetic impurity makes
a contribution to the resistivity that is independent of temperature. However, a magnetic
impurity causes a resistance minimum at a nonzero temperature. Kondo explained this
behavior as due to spin-flip scattering between the conduction electrons and the localized
375
376 Chap. 6 • Strong Correlations
spin. The resistance minimum is now called the Kondo effect. The Kondo model is described
by the Hamiltonian
H -- '" Ct C J '" iR.(k-p)[(Ckt

L." tk kG" kG" - N L." e 1
t C t
pt - C k.j, C p.j, )S(z)
j
kG" jkp
t C S(-) + C t c S(+)]
+ Ckt (6.1)
Pt j k.j, pt j
The first term represents the kinetic energy of the conduction electrons. If the local spin is
from a d-shell, the conduction band is usually formed from atomic orbitals that have s- and p-
symmetry. If the local spin is from an f-orbital, then the conduction band could be from s, p,
or d electrons.
The remaining terms represent the scattering from the local spin at the site Rj . The
number of local spins is N i , and the concentration is c = NJN, where N is the number of
atoms in the solid. Usually it is assumed that c is quite small. The electron self-energy is
calculated to order O( c). This approximation is equivalent to assuming that the conduction
electron scatters from one impurity at a time, or that the impurities are widely separated.
The interaction term can be written compactly as s· S, where the small s is the spin of the
conduction electron, and the large S is for the localized spin. The first scattering term in (6.1)
is the s(z) S(z) component. For s(z) we have plus one (ct t Cpt) when the spin of the conduction
electron is up, and minus one (-CL Cp.j,) when it is down. The other two terms in the
scattering interaction represent spin-flip processes, where there is a mutual spin flip between
the conduction and local spin. The conduction electrons are assumed to be electrons with spin
one-half. The local spin can have any value permitted by quantum mechanics. Here it is taken
to have spin S with magnetic quantum number m. Both S and m are either integers or half
integers, where m has the range of values -S :s m :s s.
S(z)[m) = m[m)
S(+)[m) = [S(S + 1) - m(m + 1)]1/2[m + 1) (6.2)
S(-)[m) = [S(S + 1) - m(m - 1)]1/2[m - 1)
The operators S(±) raise or lower the magnetic quantum number m. The interaction term in
(6.1) can be written compactly as
J", . t
V".d = - - L." exp[zRj ' (k - p)](J'o:~ • SCko:Cp~ (6.3)
N kpo:~
The indices rx and ~ denote the spin of the electron after and before the scattering event. The
vector (J' is a Pauli spin matrix for different x, y, and z components. If the meaning of the
terms in (6.3) is unclear, just consult (6.1) for the actual formula.
The constant J is called the "exchange energy." Values with J > 0 are called "ferro-
magnetic," since the local spin tends to line up parallel with the conduction band spins. Values
with J < 0 are called "antiferromagnetic," since the local spin tends to line up antiparallel
with the conduction band spins. This nomenclature is dependent upon writing the Hamilto-
nian with a negative sign before the interaction term. Most magnetic impurities have J < O.
6.1.1. High-Temperature Scattering

The self-energy is calculated for an electron scattering from the impurity. The local spin
and the conduction band spins are assumed to have equal probability for any value of m.
Sec. 6.1 • Kondo Model 377
These assumptions are valid at high temperature. Whether the temperature is high or low
depends upon a reference temperature called the Kondo temperature TK , which will be
defined below.
The interaction is (6.3). The average over impurity positions is taken following the
prescription in Sec. 4.1.4. The first term is found in first-order perturbation theory. The
average of Vsd over impurity positions forces k = p. The first-order self-energy is a constant
which is independent of wave vector or energy:
~I = -cJ(CT(J.(1. • S} (6.4)
The notation CT means to take the z component. It is usually assumed that the system is
(1.(1.
nonmagnetic. Then the average value of cr(z) and S(z) are both zero, since there are equal
numbers of states occupied with spin up as with spin down. Then the above self-energy
expression is zero.
Next consider the second-order scattering. The n = 2 term in the S-matrix expansion for
the Green's function is
~ (2)
(1.~ (k, 't) =! J~0 d'tl J~0 d't2 (TCkrx('t)vsA'tI)Vsd('t2)Ck~(O)}
A t A
= -J2~J~ d'tl J~ d't2 L L CT~~CT~~(TS(!1)('tI)S(V)('t2)}

o 0 ss'uu' kl k2P1P2
X (TCkrxC't)CL('tI)CP1A'tI)ctuC't2)Cp2U'('t2)Ct~(O)}O(kl - k2 - PI + P2) (6.5)
The last bracket contains three raising and three lowering operators for conduction electrons.
One can pair them up using Wick's theorem. The correlation function of the local spin is not
treated as easily. Local spins are neither fermions nor bosons. The thermodynamic average is
obtained by averaging over the states 1m} with different magnetic quantum number. If there is
no external magnetic field, then each state of different m is equally likely, which is assumed
here.
The major problem with evaluating the correlation function of the local spin is that they
do not obey a Wick's theorem. Here it is no problem, since there are only two operators,
which must be paired. However, the lack of Wick's theorem is a major algebraic difficulty in
higher orders of perturbation theory. The S(v) operators must be paired: each S(+) must be
paired with a S(-). Each bracket can have an arbitrary number of S(z) operators, and they are
not paired. The correlation function for the local spin has three possible combinations which
are evaluated using (6.2):
(TS(+)('tI)S(-)('t2)} = S(S + 1) - m2 - m sgn('tl - 't2) (6.6)
(TS(-)('tI)S(+)('t2)} = S(S + 1) - m2 + m Sgn('tl - 't2) (6.7)
(6.8)
The only 't dependence is in the order ofthe operators. At this point it is useful to average over
the values of m. Then the term linear in m vanishes, and the first two of the above expressions
become identical.
Assume that the initial spin ex of the conduction electron is ±. If the scattering does not
flip the spin, the intermediate and final states are also ex. Then the local spin correlation
function has the form (S(z)S(z)}. The other possibility is that the spin flips so that the inter-
mediate state has the opposite spin to ex. Then one must use one of the two combinations with
S(+) and S(-). The Green's functions for the conduction electron are the same for the two
cases of spin-flip or no spin-flip. Just add these two cases, which causes the m Z terms to
cancel. The second-order self-energy for the Kondo model is
I
Lz{ikn ) = cS(S + 1).P- L-. -- (6.9)
N k l kn - Ek
This self-energy expression has the same form as ordinary impurity scattering in the second
Born approximation. The spin enters only as an effective interaction of S(S + l).1z. Early
workers in the field stopped here, and concluded that spin-flip scattering provided no inter-
esting behavior. This conclusion is wrong, since the unusual effects are found in the higher
orders of perturbation theory.
The self-energy of the conduction electron is now calculated in the third Born
approximation, which repeats Kondo's original calculation. The anomalous behavior which
gives rise to the Kondo effect becomes evident. The n = 3 term in the Green's function
expansion is
<;§~J(k, 1:) = - cJ: L J~ d1:1 J~ d1:z J~ d1:3<;§(k, 1: - 1:1)<;§(P, 1:1 - 1:z)

N pq 0 0 0
x <;§(q, 1:z - 1:3)<;§(k, 1:3)L(1:I, 1:z, 1:3) (6.10)

L(1:I' 1:z, 1:3) = L 0'~10'~~~ 0'~~2 (TS(V)(1:I)S(IJ)(1:Z)S(A)(1:3)) (6.11 )
IJVA
The self-energy must be the same for spin-up and for spin-down electrons, so set IX = + for
spin up. The (s, s') can be either of the four combinations (+, +), (+, -), (-, +), (-, -). For
IX = + then each of these four combinations of(s, s') has a unique choice for (VilA). Since the
electron Green's functions do not depend upon the spin label s, the summation over (s, s') is
included in L:
L(1:I' 1:z, 1:3) = (TS(z)S(z)S(z») + (TS(z)(1:I)S(-)(1:z)S(+)(1:3)) + (TS(-)(1:I)S(+)(1:z)S(z)(1:3))
+ (TS(-)(1:I)S(z)(1:z)S(+)(1:3)) (6.12)
No time variables are written in the first term on the right since this expression is independent
of the 1: variables. This term equals m3 for any 1: ordering. The other terms depend upon the 1:
ordering. The factor of S(+) is always to the right of S(-). This choice follows from the ansatz
that IX = +. Spin-flip scattering must lower the spin of the conduction electron, while raising
the spin of the local electron. The local raising operator is S(+) which must act at the earlier
0'
time. The factors ss' are not written for each term. They are always plus except for one case.
When the factor S(z) immediately follows in time (1:) the operator S(+) then a minus sign is
inserted. The minus sign arises because the factor S(+) raises the local spin while flipping the
conduction spin from up to down. The O'(z) operates on the down conduction spin and gives
minus one.
The above expression must be evaluated for all six arrangements of 1: ordering. The
ordering 1:1 > 1:z > 1:3 gives
L = m3 + [S(S + 1) - mZ - mUm + m - (m + 1)] (6.13)
L = m[S(S + 1) + 1] - S(S + 1) (6.14)
The terms are written in the same order they occur in (6.12). The minus sign before the factor
of (m + 1) is due to the sign reversal mentioned above. The term proportional to m vanishes
when we average over m. After averaging, the above expression equals -S(S + 1).
Next consider the 't-ordering of't3 > 't2 > 'tl, which gives
L = m3 + [S(S + 1) - m2 + mUm - m + (m + 1)] (6.15)

L = m[S(S + 1) + 1] + S(S + 1) (6.16)
After averaging over m this term gives S(S + 1). This result is similar to the one above, except
that the sign has changed. One can work out the other four cases. The three cyclic
arrangements 'tl > 't2 > 't3' 't2 > 't3 > 'tJ> and't3 > 'tl > 't2 each have L = -S(S +1). The
other three arrangements 't3 > 't2 > 'tl' 't2 > 'tl > 't3, and 'tl > 't3 > 't2 each have
L = S(S + 1). The three 't integrals in (6.10) must be broken up into the six separate regions
of different 't ordering. Three of them are multiplied by S(S + 1) while the other three are
multiplied by -S(S + 1).
The notation (123) means 'tl > 't2 > 't3' In this notation the following result is obtained
for L after averaging over values of m
L = -S(S + 1)[(123) + (231) + (312) - (321) - (213) - (132)]

1 = (123) + (231) + (312) + (321) + (213) + (132)
The second line above expresses the idea that the summation of all possible time orderings is
just the unity operator. This result can be used to rewrite L as
L = -2S(S + 1)[(123) + (231) + (312) -!] (6.17)
The amount of work has been cut in half, since now only three time orderings need to be
evaluated.
The third-order Green's function of frequency is evaluated by multiplying Eq. (6.10) by
exp(ikn't) and integrating d't between (0, ~). The integrals are most easily evaluated in reverse
order:
<;§(3)(k, ikn) = -cS(S + 1):; J: d't3e~3(ikn-Ek)<;§(k, 't3)
x J: d't2e(ikn-Ep)(~2-~3)<;§(p, J: d'tle(ikn-Eq)(~I-~2)<;§(q,
't2 - 't3) 'tl - 't2)
x J: d'teikn-Ek)(~-~I)<;§(k,'t-'tI)L('tI,'t2,'t3) (6.18)
!.
In Eq. (6.17) the last term in brackets is It has no limits on the order of the 't integrals. Then
all of the above integrals have their usual limits of (0, ~), and the above expression is
-S( S + 1) N2 <;§
cJ3 (0)( 'k)2 " (6.19)
k, I n ~ ·k=---_------:-)-:-c(.k=----------:-)
-:-c(
pq I n Ep I n Eq
This expression is a contribution to the third-order Green's function. If one eliminates the
prefactor of <;§(O)(k, ikni then the remaining expression is a contribution to the third-order
self-energy. It is just the third term in the T -matrix equation generated by iterating (4.49). This
contribution to the self-energy is rather dull.
The Kondo effect comes from the terms in L which restrict the order of the 't integrals.
The restrictions are on 't I' 't2 and 't3' There is no restriction on the d't integral. It can be done
immediately, and yields <;9'(O)(k, ikn). The first interesting integral is (123), whose limits can be
written as
(6.20)
Doing the integral in (6.18) with these limits gives the following result. Drop all of the
prefactors and summations, and list only the results of the three 't integrals:
(123) = [1 - n(p)][l - n(q)] + [1 - n(q)][n(k) - n(p)]
(ikn - Ek)(ikn - Ep)(ikn - Eq) (ikn - Ep)(Ek - Ep)(Ep - Eq)
[1 - n(p)][l - n(q)]
(6.21)
The expression has a multitude of fermion occupation factors n(k), etc. Similar expressions
are obtained for the other two 't orderings (231) and (312):
(231) :::} Jo~ d't3 J~ d't2 J'2 d't 1

'3 0
(231) = . 1 \[1 - n(q)][n(k) - n(p)] _ [1 - n(p)][n(k) - n(q)]

(zkn - Ep)(Ep - Eq) Ep - Ek Eq - Ek
n(p)[l - n(q)] n(k)n(p)[l - n(q)] - n(q)[l - n(k)][l - n(P)]}

+ (6.22)
+'k z n - Ek Z'kn - Ek - Ek - Ep + Eq
(312) :::} J~ d't3 J'3 d't2 J'3 d'tl

o 0 '2
(312) = . [1 - n(p!][n(k) - n(q)] . n(q~[l - n(p)]
(zkn - Ep)(zkn - Eq)(Ep - Ek) (zkn - Ed(zkn - Eq)(Ep - Eq)
n(k)n(p)[l - n(q)] - n(q)[l - n(k)][l - n(p)]
(6.23)
(ikn - Ep)(ikn - Ek - Ep + Eq)(Ep - Eq)
An interesting thing happens when we add these three expressions. Most of the terms cancel:
(123) + (231) + (312) =
_. 1 2 [ . 1. +. n(p) +. n(q) ] (6.24)

(zkn - Ek) (zkn - Ep)(zkn - Eq) (zkn - Eq)(Eq - Ep) (zkn - Ep)(Ep - Eq)
A further simplification occurs by noticing that the last two terms are identical after inter-
changing the variables of integration p and q. The result for the third-order self-energy is
.
L3(zkn) = -S(S+ 1)N2
cJ3
LCk -
pq zn
1 [1
Eq
) Ck _
Z n Ep
4(n(P)]
) --_-
Ep Eq
(6.25)
The final expression is rather simple. There are two terms. The first term in the brackets is the
rather dull term, which is the third-order Born scattering from a simple potential. The second
term in the brackets was first derived by Kondo. He used it to explain the temperature
dependence of the resistivity, which is now called the Kondo effect. Today it is known that
similar anomalous terms occur in all higher orders of perturbation theory. The modem theory
of the Kondo effect employs tenns from all orders of perturbation theory. Nevertheless,
Kondo's simple arguments will be presented.
The self-energy expression has a factor of the fennion occupation number n(p). These
factors do not occur in scattering from a simple potential. They occur here because of the
spin-flip scattering. Interesting physics happens whenever they occur, such as in the BCS
theory of superconductivity or in the edge singularities in X-ray absorption in metals.
The second tenn in the above brackets produces a tenn in the resistivity that goes as
In(T), where T is the temperature. A quick and sloppy derivation is presented of this
behavior. The conduction band density of states is defined as gee). It is assumed to be a
smooth function of energy E, which extends from W < E < B. Summations over wave vector
are changed to integrations over energy using the prescription
~ "Lf(Ep) =} IB dEg(E)f(E) (6.26)

N p -w
A typical alloy system that shows the Kondo effect is manganese impurities in copper. The
density of states gee) of conduction electrons in metallic copper has a large energy depen-
dence because of the occupied d-bands. However, the conduction electrons that are involved
in the Kondo effect have energies within several kBT of the Fenni energy. On that small
energy scale, the density of states of copper and most metals is smooth and featureless. The
custom is to call the density of states gee) a constant g(O), where E = 0 signifies the Fenni
energy.
For the Kondo effect, the interesting third-order self-energy is
""f. 3b (ikn ) = -4S(S + 1)cJ3 IB dEl. g(EI) IBdE2 g(E2)n(E2) (6.27)

-w zkn - EI -w EI - E2
The ultimate goal is to calculate the resistivity. Then the quantity of interest is the scattering
rate, which is related to the imaginary part of the self-energy. In this expression let
ikn --+ m + ill, where 11 is infinitesimal, and then take the imaginary part. The first energy
denominator gets replaced by -1t8(m - EI)' The first integral gives -1tg(m). In order to
evaluate the second integral, it is assumed that the temperature is low, so the fennion
occupation factor can be replaced by the step function n(E2) ~ 0( -E2)' The most important
tenn is the logarithmic singUlarity which comes from this Fenni cutoff
(6.28)
Since the density of states g( m) is a smooth function of energy m, this quantity is usually
replaced by g(O). The standard expression for the third-order self-energy in the Kondo effect
IS
(6.29)
The real part of ""f.3b has an uninteresting m dependence near m = O.

This result is combined with the larger tenn from the second order of perturbation
theory:
Im[""f.(m)] = -1tS(S+ 1)CJ2g(0)[1 +4Jg(0)lnl;I"'] (6.30)

In calculating the conductivity, the energy ro of the electron is averaged over a region near the
Fermi surface of width kBT. This replaces lnlrojWI by In(kBTjW). Kondo derived a
temperature-dependent resistivity of the form
p(T) = p(O)[1 + 4Jg(0) In(kBT jW)] + bT5 (6.31)
The first term peT = 0) is the temperature-independent part of the impurity scattering. The
second term with In(T) is the Kondo effect. The last term '" T 5 is the low-temperature
contribution from electron-phonon scattering. Usually the experiments are done in metals or
alloys where phonon drag does not eliminate the T 5 law.
The factor of In( kB T j W) is negative at low temperature. If J < 0 then the Kondo term is
positive. At very low temperatures, the resistivity increases in value, since this term becomes
large. This expression predicts a minimum in the resistivity, which is observed in Kondo
systems. Define Tmin as the temperature at the minimum resistance. By taking d p j dT = 0 one
finds that (Tmin)5 '" c, where c is the concentration of impurities (Note p(O) ()( c). This
relationship is in agreement with experiments on Kondo alloys. A typical measurement of
p(T) is shown in Fig. 6.1 for the system of La'_xCexB6' The resistance minimum is evident.
The resistance increases at low temperature, and finally saturates. The reason for the
saturation is given in the following section.
The theory presented in this section is a high-temperature theory. A key assumption in
the derivation is that the local spin has equal energy whether it is up or down. At low
temperature, this assumption ceases to be valid. Each local spin becomes locked into a
collective state with the conduction band spins. This collective state has a binding energy,
which must be overcome during a spin-flip. The spin-flips become frozen out at low
temperature, and the Kondo effect saturates. This process is described in the next section.
Kondo's formula (6.31) is only valid at temperatures high above the formation of the
collective state.
Our derivation of the Kondo resistance was sloppy for the following reasons. In eval-
uating the integral over dE2 the Fermi factor n(E2) was replaced by a step function 0( -E2)'
This approximation is only valid at zero temperature. Later in the derivation, we averaged
InlrojWI over the distribution of electrons while assuming the temperature is nonzero. A
better derivation would have evaluated the integral over dE2 at nonzero temperature, which
broadens the Fermi distribution a few degrees in energy. Of course, this realism makes the
integral harder to evaluate.
10 K 102
T-
FIGURE 6.1 Total resistivity vs T for LaB6 and Lal_xCexB6' The Kondo temperature is TK = 1.05 K. Source:
Samwer and Winzer (1976) (used with permission).
6.1.2. Low-Temperature State

At low temperature the local spin forms a collective state with the conduction band
spins. The collective state is formed from a linear combination of conduction band states, and
is a many-particle entity. Earlier chapters have often discussed the role of screening charge in
reducing the long range Coulomb interaction in conducting systems. The screening charge is
not from a single electron in the conduction band, but usually is formed by a linear combi-
nation of conduction states. The present collective state can be viewed, in a similar way, as a
screening effect. However, now it is screening the spin of the local state, rather than its charge.
If the impurity is charged there could coexist the screening clouds for spin and charge.
For J < 0 the state is a singlet. In order to keep the discussion simple, let us assume that
!.
the coupling is antiferromagnetic (J < 0) and that the local spin has S = The up and down
states ofthe local spin are denoted as IX and B. Yosida (1966) first considered the collective
state. Its wave function can be written in two possible ways:
(6.32)
(6.33)
(6.34)
The symbol IF) denotes the filled Fermi sea of electrons. The state %a forms the collective
state by adding an electron above the Fermi sea. It has the usual form for a spin singlet, as an
anti symmetric combination of up and down states of the local and conduction spins. The
coefficient ak is a parameter that needs to be determined.
The second choice of collective wave function %b needs further explanation. Here one is
forming the collective state by taking linear combinations of hole states, where the hole is a
missing electron from the filled Fermi sea. The easiest way to understand this wave function is
to examine one of its terms, which include the relevent factor from the Fermi sea:
(6.35)
The right-hand expression is obviously a spin singlet, composed of an antisymmetric

combination of down and up spin states. The relative plus sign on the left becomes a minus
sign on the right. The two states %a and %b are orthogonal. Each describes a different
collective state, and the system will choose the one with the lowest energy. One is a ground
state, and one is an excited state.
The first term in (6.1) is calledHo and the second term is Vsd' If there is no local spin, the
ground state energy for this model Hamiltonian is
HoW) = EGW) (6.36)

EG =2 L Ek (6.37)
Ikl<kF
Now add one local spin at the point Rj = O. In this calculation, one does not average over the
spatial positions of the impurity, and one does not set p = k in the interaction term. Only one
impurity site is being considered.
The goal is to calculate the additional energy 8E < 0 gained by the system because of
the formation of the collective spin state. It is analogous to the energy found for charge
screening using Fumi's theorem in Sec. 4.1. The eigenvalue equation is
(6.38)
This equation is not strictly valid since neither \jJOa nor \jJOb is an exact eigenstate of the
Hamiltonian. In order to understand this assertion, consider the effect of one term in Vsd
acting upon one term in \jJOa
(6.39)
The spin-lowering operator S(-) changes a to ~. If q = k there is a term
(6.40)
which is recognized as being similar to one of the other terms in \jJOa' However, there are also
terms in (6.39) in which k, p, and q are all different wave vectors. These terms describe a state
with one hole and two excitations above the Fermi surface. An electron-hole pair has been
excited in addition to the original excitation. This state is perfectly valid, since it is generated
by the interaction term in the Hamiltonian. Our failure to include it in the ansatz eigenstate
\jJOa is simply an approximation. A more accurate ansatz would include these additional terms.
There are an infinite number of them, which correspond to the multiple excitation of electron-
hole pairs. The more pairs that are included, the more accurate is the wave function.
The best way to understand the ansatz wave function in (6.32) is to consider it a
variational calculation on the ground state. The coefficients ak and bk are the variational
parameters. The ground state energy is calculated, using Dirac notation, with j = a or b, as
(6.41)
After evaluating the right-hand side of this expression, say for j = a, one finds the minimum
value of 8Ea by taking the functional derivative 8(8Ea)/8ak' There results an equation for ak
that can be solved. These steps produce the minimum ground state energy for the ansatz
(6.32).
The method of solution is equivalent to the variational procedure, but is algebraically
simpler. Equation (6.38) is evaluated. The multipair states are neglected, and only project out
the terms with one conduction band excitation. For example, when solving for \jJOa define
(FlaCkt as the Hermitian conjugate of aCZ t IF) and evaluate the expression
(6.42)
As an example, the easy terms are done first:
1 t t
{Ho - EG - 8Ea}\jJOa = r.r L adEG + Ek - EG - 8Ea][aCkt - ~CktllF)
vNlkl>kF
1
(FlaCkt {Ho - EG - 8Ea}\jJOa = .jNak[Ek - 8Eal (6.43)
A similar evaluation must be done for Vsd. One term was previously evaluated in (6.39).
Another term in Vsd~Oa is
(6.44)
The operator S(z) acts upon a and gives a12. The electron operators give several contributions.
For p = q the part in parentheses gives the number of electrons with spin up minus the
number with spin down. The state ctJ, IF) has one more electron with spin down, so the
summation gives - 1. There is also a term where q = k. These two contributions are
(6.45)
Other terms contain an electron-hole pair, and these contributions are being ignored. In the
above expression, the first term can be neglected since it is of order O( 1IN) compared to the
second. The second has a double summation, while the first has only a single summation. By
employing these kinds of approximations, there results
(6.46)
This result is combined with (6.43) to produce the equation for ak:
(6.47)
This expression is then summed over k. These summations are changed to integrals over the
particle energy. The eigenvalue equation is
1=_3JJBdE g(E) (6.48)

2 0 E- 8Ea
The solution to the wave function ~ab follows the same steps used to obtain ~Oa' The form of
the wave function in (6.35) has the same form as (6.32), so that all steps in the derivation are
similar. There are two minor differences. First, since Ikl < kF' the integral over energy is over
negative energy states. Second, the result of Ho ~Ob has the factor of (EG - Ek); the minus sign
preceding Ek comes from the fact that ~Ob has an electron missing from the Fermi sea. This
changes the sign of E in the integral. So the eigenvalue equation for ~Ob has the form
(6.49)
(6.50)
The two eigenvalue equations (6.48) and (6.50) are very similar. The major difference is that
one has an integral over the empty states in the conduction band, while the other has an
integral over the occupied states. Since 8Ea ,b < 0 neither energy denominator is singular.
These two expressions will be evaluated for a simple model of the conduction band. The
density of states gee) equals a constant g. Then the two integrals give
1 = (3Jgj2) lnl W ~8Ea I (6.51)
1 = (3Jgj2) lnl 8E~

B- Eb
I (6.52)
which have the solution

WA
8E = - - - (6.53)
a 1 +A
BA
8Eb = - - - (6.54)
1 +A
2
In(A) = - (6.55)
3Jg
The largest value of 8Ea ,b is determined by whether B or W is largest. If the band is less than
half full, then W < B, 8Ea > 8Eb , and \jJOb is the ground state. If the band is more than half
full, then W > Band \jJOa is the ground state. If the band is exactly half full, the states are
degenerate. These results only apply to the case of a constant density of states. Other
examples are assigned in the problems.
Remember it is assumed that J < O. In fact, there is no solution for singlet states when
J > O. The present approximations only make sense when 8E « (B, W) so that the logarithm
terms in (6.52) are negative. Then the equation has a solution only for J < O. For anti-
ferromagnetic coupling, the factor of A = exp(2j3Jg) is generally much less than unity. This
expression also has the feature that the coupling constant J appears as an inverse power in an
exponent. It would be impossible, or at least very difficult, to derive an expression for 8E
from perturbation theory.
This mathematical form, of an inverse power in an exponent, will be encountered in
Chap. lOin the BCS theory of superconductivity. The energy gap and the formula for Tc have
a similar factor of exp[-ljNFVol
The Kondo temperature is defined by the expression
kBTK = W exp { - 2g(~)IJI} (6,56)
This expression is similar to the result for the change in ground state energy. The main
difference is the factor of ~ in the exponent has become!. The reason for this change is given
below. A general definition is that the change in ground state energy defines an energy scale
that also defines a characteristic temperature. The Kondo temperature is also very close to the
temperature at which the Kondo effect becomes important in the resistivity. The self-energy
correction given by the second term in (6.30) becomes equal to the first at a temperature of
T ~ TK' It is not the same temperature as the resistance minimum, since the minimum
depends upon the concentration of impurities. The Kondo temperature defines an energy scale
for a single impurity. Values of TK for actual alloys span a large range in temperature, from
10-3 to 103 K.
So far the derivation suggests that the collective state has a similar behavior regardless of
the sign of J. However, further work has shown that the collective state is not formed for
ferromagnetic coupling (J > 0). If further terms are evaluated in the perturbation expansion,
they destabilize the collective state for J > O. For J < 0, further terms in the perturbation
theory continue to predict a collective state. The Kondo effect does depend upon the sign of J,
and exists only for J < O.
6.1.3. Kondo Temperature

The two prior sections discussed the high- and low-temperature regimes of the Kondo
effect. The most interesting behavior occurs for temperatures near the Kondo temperature TK,
which is defined in (6.56). A short and somewhat qualitative discussion of this temperature
region is presented here. This region is actually very well understood because the exact
solution to the Kondo problem is now known. See the solutions by Andrei et al. (1983) and
Wiegmann (1981).
The high-temperature scattering theory has been carried out to much higher orders of
perturbation theory. Abrikosov (1965), and Silverstein and Duke (1967) have suggested that
the perturbation expansion for the retarded Green's function can be represented by the simple
function
cJ
(6.57)
~ret(co) = 1 - 2Jg(co) lnlco/WI - ircJg(co)S(S + 1)
The function calculated in Sec. 6.1.1 was the imaginary part of the self-energy. The imaginary
part of the above expression is
~ ncJ2g(co)S(S + 1)
(6.58)
Im[ ret (co)] = [1 _ 2Jg(co) Inlco/Wlf + [rcJg(co)S(S + 1)]2
If the denominator is expanded in a power series in Jg( co), the first term is the perturbation
expression in (6.30). Another feature of this expression is that the denominator has a reso-
nance when the energy co ~ Wexp[1/2Jg(0)1 which makes sense when J < O. There is no
resonance for J > O. This observation provides further evidence that there is no collective
behavior for J > O.
A factor of2Jg(co) can be taken out of the denominator in (6.57). Using the definition of
the Kondo temperature in (6.56) allows the real part of the denominator to have a simple
form:
c I
~ret(co) = - 2g(co) Inl CO/kBTK I + inS(S + 1)/2
(6.59)
~ cnS(S + 1)/4g(co)
Im[ ret (co)] = In2 Ico/kBTK I + [nS(S + 1)/2f
The interesting feature of this expression is that the co dependence of the self-energy scales
only with the Kondo temperature. One should be able to describe all of the self-energy effects
by this single parameter. It also has a resonance behavior. The square of the logarithm has its
smallest value of zero at COK = kBTK' and the denominator increases in magnitude for values
of co different from COK'
6.1.4. Kondo Resonance

Wilkins (1982) observed that many of the properties associated with the Kondo effect
can be described by a simple expression. The system acts as if the density of states has a
resonance at the Fermi surface for each impurity with a local moment. Let the concentration
of impurities be Cj = NJ N. A simple expression which well describes the density of states is
C Y
gK(m) = gem) +....!.]tOO2 +y2 = gem) + og(m) (6.60)
where y = 1.6kB TK . The change og(m) integrates to one electron for each impurity with a
local spin.
For example, to calculate the change in the specific heat OC due to the change og((0),
first find the change OU in the internal energy
a
oc = aT oU = cJ(T ITK)
(6.61)
oU = JdmmnF(m)og(m)
Figure 6.2 shows how the function/(TITK) (dashed line) compares to the exact result
(solid line). The two results are quite similar. Note that the temperature scale spans several
decades. The specific heat per magnetic impurity is a function only of T ITK. At high
temperature the specific heat per impurity goes as/ '" TKIT. At low temperatures it goes as
/ '" T ITK, which can be large if the Kondo temperature is low. Wilkins observes that the
entire effect of the many-body interaction is to add a Kondo Resonance at the Fermi surface,
such as the term og(m). Numerical calculations of the Kondo resonance show that it is usually
asymmetric in frequency and is not centered precisely at 00 = O.
The resonance behavior also affects the resistivity. Assume that the Kondo resonance
affects the quasiparticle self-energy r = Im[~]. Instead of the behavior predicted in (6.59) use
the resonance form in (6.60). The temperature-dependent conductivity a(T) is given in terms
of the lifetime ,(T)
n e2 ,
a =_0_ (6.62)
m
,= Joo dm[- dnF(m)] _1 (6.63)
-00 dm 2r
2r(m) = (1 + 00
/y
+y 2)/'0 (6.64)
0.2
m
~
.....
U 0.1
o~--~----~--~----~--~
0.1 1.0 10 100
T/TK
FIGURE 6.2 Specific heat per impurity for the Kondo model. The solid line is the numerical solution using the
renormalization group. The dashed line is the calculation using the Lorentzian resonance with unit weight and width
equal to 1.6TK . Adapted from Wilkins (1982) (used with permission).
Sec. 6.2 • Single-Site Anderson Model 389
14~---------r----------~--------~----------~
12
10
!Q..
8
.....
;: 6
Q.
4
o~--------~----------~--------~----------~ 2
102 10·' 10° 10' 10
T/TK
FIGURE 6.3 Resistivity vs temperature in the Kondo model.
where 0'0 = noe2'tolm is the conductivity from nonresonant impurity scattering for T » TK .
The parameter r comes from the ratio of resonant to nonresonant impurity scattering. The
frequency integral in (6.63) can be done numerically. The result is given in Fig. 6.3, for
r = 10, which shows the resistivity p(T) = 1/0'(T). At high temperature (T » TK ) one finds
p = Po. At low temperature (T« TK ) one finds that p = Po(1 + r). This behavior of the
resistivity is similar to the experimental dependence shown in Fig. 6.1. At high temperature
the resonance effects from spin-flip scattering are smeared out by thermal broadening. As the
temperature is lowered to the vicinity of TK, the thermal smearing decreases, and the Kondo
resonance becomes more important in scattering the electron. The resistivity increases. At still
lower temperatures, the collective state begins to form, and the local spin becomes harder to
flip. Then the resistance saturates, and no longer increases with decreasing temperature.
6.2. SINGLE-SITE ANDERSON MODEL
The Anderson model (1961) was introduced in Sec. 1.4. It is another model for a system
of conduction electrons that interact with a local spin. Equation (1.354) gives the transfor-
mation of Schrieffer and Wolff (1966), which shows that the model has some terms that are
similar to those of the Kondo model. Early workers thought that the two models made very
similar predictions. Now it is known that the Anderson model has a greater variety of
behavior. It has the more interesting physics.
The Kondo model treats the local spin as a separate entity. The Anderson model treats
the local spin as just another electron. It can undergo exchange and other processes with the
conduction electrons. The Anderson model is more realistic.
The Anderson model is written here with a slightly altered notation. Most applications of
the model consider the localized state to have spin and orbital degeneracy. It could be a d
electron of 10 states or an f electron of 14 states. Usually the spin-orbit interaction splits these
states into 6 and 4, and into 8 and 6, respectively. The crystal field may cause further
splittings. Let Nf denote the actual degeneracy of the local level: it may be an even number
such as 2, 4, 6, 8, 10, or 14. One of the theoretical developments is a perturbation expansion,
where the expansion parameter is 1/Nf. Obviously this expansion works better for larger
values of Nf . Many recent applications of the Anderson model have been for heavy lermion
systems, where the local orbital is one or several electrons in f orbitals. That is the origin of
the subscript I on the factor of Nf . The symbol N without the subscript is the number of
atoms sites in the solid. For reviews of heavy fermions see Stewart (1984) or Lee et al. (1986).
The Hamiltonian is written as
H=Ho+V (6.65)
Ho =L EkC!vCkv + f'j L nj.l + U L nj.lnV (6.66)
kv j.l j.l>V
nj.l =IJ/j.l (6.67)

_ 1 t t
V- /iT
'\IN
Lkv Vkv[fv Ckv + Ckvfv] (6.68)
where V is the hybridization interaction between the band electrons (k) and the localized
electrons Ij.l' The onsite Coulomb repulsion is U and Ef is the eigenvalue for a single f-
electron. The summations over (Il, v) run between 1 and Nf . This index denotes a combi-
nation of orbital and spin states. Only the band states (k, v) with the same index interact with
the local orbitals. The wave vector k can be represented by the magnitude k plus various spin
and angular momentum states. The latter are represented by the index v. An important point is
that of all the many symmetries available to the conduction band states, the only ones which
couple to the local state are those values ofv between 1 and Nf . The summations over k will
eventually be changed to an integral over the energy E. The density of states is geE) and the
matrix element as a function of energy is written as Vk = VeE).
How does the local spin get flipped in the Anderson model? Actually it does not flip.
Because of the hybridization term, a local spin that is down can become a conduction state
with spin down, and then wander away. Later, a different conduction electron with spin up can
come and reside in the local orbital. This process gives the appearance ofthe local spin having
flipped from down to up, while conduction states have flipped from up to down. Now one can
see the important role of the electron-electron interaction U. The spin-flip process has two
steps: (1) the departure of the old local electron with spin down, and (2) the arrival of the new
local electron with spin up. If U = 0 these two steps are totally independent and can occcur in
any order. There is no appearance of flipping, since down and up electrons come and go
independently. However, once U» kBT, then the two steps become correlated. It is ener-
getically unfavorable to have an up and down spin electrons both on the local orbital. The two
steps tend to become sequential: one leaves before the other comes. Then the local spin
appears to flip.
The important parameters of the Anderson model are the bandwidths W, the conduction
band density of states geE), the local site energy Ef' and the on-site Coulomb interaction U.
There is also an energy parameter associated with the hybridization term which is defined as
reE):
(6.69)
(6.70)
This function is assumed to be a constant, independent of energy. Its value for electrons in
f-orbitals is about 0.1 eV
Most theories have been calculated for the case that the ground state has only one
electron in the f-orbital. This corresponds to the case of cerium triply ionized (Ce3+). Typical
values are U "-'6 eV, r ~0.1 eV, and Ef "-' -0.5 eV (Herbst and Wilkins, 1987). The zero of
energy is taken to be the chemical potential. Since the spin-orbit splitting is large, one uses
only the lower state which has Nf = 6. The energy of one local electron is Ef' while the
energy of two is 2Ef + U. In cerium one has that Ef < 0 but 2Ef + U > O. Then the cerium
ion prefers to have just one electron in the local orbital. A related case is for triply ionized
ytterbium which has all f-orbitals filled except one. The ground state has one f-hole. The
theory is almost the same, but one uses Nf = 8 for the upper multiplet of the spin-orbit split
f-state. When the local orbital is a d electron the values of r are larger.
6.2.1. No Hybridization
The term in the Hamiltonian involving U comes from the on-site Coulomb interaction
between two electrons on the same atom. Usually this term is far too large to be treated by
perturbation theory. It must be included in H o, which is the noninteracting Hamiltonian.
However, this term makes problems in doing the Green's function expansion. Including this
term invalidates Wick's theorem for local electrons. The Green's function expansion becomes
awkward. In evaluating correlation functions, each time ordering must be evaluated sepa-
rately. This increases the effort, as was evident in the prior section on the Kondo effect. Here
we use the noncrossing approximation (Bickers, 1987).
There are several other approaches to solving the Anderson model besides the non-
crossing approximation. The method of Gunnarsson and Sch6nhammer (1983, 1987) was
described in our second edition. Grewe and Keiter (1981) developed an alternate formalism
for expanding the S-matrix for the local operators. They found formal but rigorous results for
the local self-energies. Another method of finding the same results was developed by
Coleman (1984) using the slave boson idea of Barnes (1976, 1977). One introduces a ficti-
tious boson which is created every time there is a change in the occupancy of the local level.
One can control the f occupancy by controlling the chemical potential of the slave bosons. A
similar method was used earlier by Abrikosov (1965) in solving the Kondo problem. Another
method is renormalization group (Hewson, 1993).
The first step is to evaluate the Hamiltonian Ho while setting V = O. This step is easy,
even while including the Hubbard interaction U. The eigenvalues of the electrons, in the
absence of hybridization, depend upon the number n of electrons in the f-orbital. This integer
ranges in value 0 ~ n ~ Nf . The eigenvalues are
n(n - 1)
En = nEf + 2 U (6.71 )
where the number of pair interactions multiplying U is n(n - 1)/2. The eigenvalue expression
(6.71) is parabolic in the level occupancy n. An example is shown in Fig. 6.4. Since Ef is
usually negative, the minimum energy is found near to n "-'! -
Ef/U. Denote as L the value
of n which has the lowest value of En. The f-configuration with the lowest energy has an
occupancy of L electrons. Note that L is increased by changing the value of Ef towards larger
negative values, while keeping U fixed.
Mixed valence occurs when the number of f-electrons can vary between Land L ± 1. In
Fig. 6.4 it varies between the states with n = 3 and n = 4, which are at the same low energy.
The hybridization interaction causes mixing whenever IEL - EL±ll is a small energy. By
• •
-20
• •
-30 • •
• •
-40 '--'-~'---"-~~-'--~~-"- ~~-'--~----'---'
-2 o 2 n 4 6 8
FIGURE 6.4 Eigenvalues En vs n for the Anderson model with no hybridization. Values assumed in the plot were:
Ef = -18.0eV and U = 6.0eV
"small energy" is meant values in the range of 0.1-1.0 eV The standard nomenclature is to
define the two excitation energies L1± out of the ground state as
L1_ = E L _! - EL = -[Ef + (L - l)U] (6.72)

L1+ = EL+! - EL = Ef + LU (6.73)
U = L1+ + L1_ (6.74)
The definition of the Hubbard U as being equal to L1_ + L1+ follows by summing the prior
two equations. Equation (6.74) can also be used as a phenomenological definition. Using
band structure methods to calculate L1_ and L1+, one can add them to obtain a theoretical value
for U.
Values of L1± have been calculated for the rare-earth metals. The results of Herbst and
Wilkins (1987) are shown in Table 6.1. Generally they find one value which is small
(L1 ~ 1. 0 eV) while the other is larger so that U ~ 6 - 8 eV It is assumed that similar values
apply to the rare-earths ions when they are in metallic compounds. Note that samarium and
ytterbium have small values of L1+ and they have a mixed valence in many compounds.
Cerium also has a mixed valence in many compounds, although its value for L1_ does not
seem to be small. An important point is that these values do not have to be on the order of
thermal energies, but in fact are generally much larger: a sizable fraction of an electron volt.
Even with these large values, mixed valent behavior occurs often.
Next calculate the partition function Z which is related to the thermodynamic potential
Z = exp( - ~n). This quantity is needed for all of the thermodynamic averaging. Define the
degeneracy factor Zn
(6.75)
This factor is obtained by asking for the number of different ways that n identical particles can
be distributed among Nf identical states.
The ground state degeneracy of a rare-earth ion are given according to Hund's rules.
They are shown in Table 6.1. The Anderson model ignores the splittings of the f-orbital due to
spin-orbit and other factors, and gives a degeneracy which is far too large. The Anderson
Table 6.1 Ground staste degeneracies (2J + 1) for valence (3+) ofrare
earths according to Hund's rules: d± = En±1 - En in eY. The second entry
for Eu has valence 2+. From Herbst and Wilkins (1987, used with
pennission)
nf S L J 2J + 1 d d+
Ce 1 1
3 2 6 1.8 3.2
2 2
Pr 2 1 5 4 9 3.8 1.9
Nd 3 1. 6 9
10 5.2 1.5
2 2
Pm 4 2 6 4 9 5.5 1.6
5 5
8m 5 2 6 2 6 5.4 0.3
Eu 6 3 3 0 1 7.9
Eu 7 1 0 1 8 1.9 8.6
2 2
7 7
Gd 7 2 0 2 8 8.5 3.7
Th 8 3 3 6 13 3.2 2.2
Dy 9 2 5 15
16 4.9 1.5
2 "2
Ho 10 2 6 8 17 5.8 1.4
Er 11 1. 6 12 16 5.9 1.8
2 2
Tm 12 1 5 6 13 5.5 1.1
1 7
Yb 13 2 3 2 8 7.4 0.9
model is solved as given in (6.68), which has a degeneracy given by Zn rather than Hund's
rules. The partition function is
NI
Z= L Zn e- PEn (6.76)
n=O
When U = 0 this summation simplifies to Z = [1 + e-PsI(I.

The Green's function for the f-states in the absence of the hybridization term is
(6.77)
The expression for t§f can be written exactly for the noninteracting states. The trace is taken
over the states In). The notation Nr denotes the number of states for a single f-electron. The
degeneracy of having n electrons in the f-orbital is Zn as defined above. The 't integral is
(6.78)
The matrix elements LiJ (nl~ln + 1) (n + 11ft In) are easily evaluated since they are just the
number of empty states into which one can add an electron into the f-orbital. If there are
already n, then the number of empty spaces is Nf - n. In the term with exp( -~En+l) change
summation variables n + 1 --+ n. Since Zn(Nj - n) = (n + l)Zn+l these steps produce the
final expression
<§j(iron) =! t Zne-~En
Z n=O
{. Nj - n +. n
zrom + En - E n+1 zrom + E n- 1 - En
} (6.79)
Aj(ro) = 2 Im[Gj(ro)] (6.80)

2n Nf
Aj(ro) = - L Zne-~En{(Nf - n)o(ro + En - E n+1) + no(ro + E n - 1 - En)} (6.81)
Z n=O
The spectral function Aj( ro) is obtained by the analytic continuation iro m --+ ro + iT] and then
taking twice the imaginary part. The factors of Nj - n in the first term, and n in the second,
arise from the degeneracy of the state. In the state In) there are n different states that can be
destroyed by the lowering operator fJ.l when summing over all j.l. This accounts for the factor
of n. Similarly, the factor Nf - n is the number of empty states into which the raising operator
can add an f-electron.
Let EL be the lowest eigenvalue (n = L) and assume that L is neither 0 nor Nj . The most
important terms in the summation are those for n = L - 1, L, L + 1. Factor ZL exp( -~EL)
from each term. Energy differences EL - EL±1 are expressed as Ll±. These approximations
give for the partition function
Z ~ ZL-l e-~EL-l + ZLe-~EL + ZL+l e-~EL+1 (6.82)
Z ~ :?l'ZLe-~EL (6.83)
:?l' = 1 + ZL-l e-~(EL-I-EL) + ZL+l e-~(EL+1-EL) (6.84)
ZL ZL
L
:?l' = 1+ e-~LL + Nj-L
_ _ e-~Ll+ (6.85)
Nj+l-L L+l
Other terms can be ignored since they have exponential factors exp( - ~M) with very large
values of M. These exponential factors are vanishing small. The same approximation for the
spectral function gives six terms which have four different energy differences
(6.86)
The approximate spectra has four peaks. Two are below the chemical potential and two above.
We follow custom and ignore the last two terms in Ar( ro) and include only the peaks at
ro = - Ll_, Ll+. For the case that L = 1 then the third term is missing anyway since there is no
state L - 2. Usually it is a good approximation to neglect terms of O(e-~Ll±), and in this case
the approximate spectral function is
(6.87)
(6.88)
This spectral function has only two peaks: one at 00 = ~+ and the other one at 00 = -~_.
They have relative strength of Nf - Land L, respectively. There are only two peaks, since the
Green's function gives the probability for two processes: (1) adding an electron, or (2)
removing an electron. In a well-defined ground state, these two processes each have a peak,
and there are two peaks. Without the Hubbard energy U both peaks would occur at the same
energy: there would be a single peak with weight Nf . The spectral function integrates to Nf ,
rather than to one, because the present definition included a summation over all of the state v,
of which there are ~.
The solution to the Anderson model is rather easy without hybridization. The next step is
to include this important process.
6.2.2. With Hybridization

An expression for the partition function is derived including the interaction V. Define the
eigenstates of Ho by In) and those of H = Ho + V by 10().
Holn) = Enln) (6.89)
HIO() = cIXIO() (6.90)
The states In) have n electrons in the f orbital. The nature of the states 10() is to be determined.
In principle, they exist, and the derivation proceeds using that assumption. They form a
complete set, so an insertion ofLIX IO()(O(I is equivalent to the identity operator. The derivation
below proceeds by inserting this operator whenever the total Hamiltonian H appears in any
expression. Using this process to evaluate the partition function gives:
(6.91)
n
(6.92)
nIX
Z = L Zn I(nI0()12e-~e. (6.93)
nIX
Next introduce the spectral function An(OO) for the many-electron state n. In prior usage the
symbol An(OO) may have meant the spectral function of a single electron in the state n. Here
the meaning is different. The symbol In) denotes a state of n electrons in the f orbital, and
An(OO) is its spectral function. It is the spectral function for a system ofn electrons. It is not the
spectral function of a single electron, unless n = 1.
An(OO) = 21t(nI8(oo - H)ln) (6.94)

= 21t L I(nI0()1 2 8(oo - clX ) (6.95)
IX
where the complete set of states LIX IO()(O(I is inserted in order to evaluate the delta function
containing H. This formula for the spectral function can be used to evaluate our above
formula for the partition function. From now on, the expression LIX l(nI0()1 2 will be evaluated
as an integral over the spectral function
(6.96)
The partition function can be expressed as a summation over the noninteracting states n. For
each value of n, integrate over 00 with the factor in the integrand of exp( - Boo )An (00). The
method of calculating An(OO) is described below.
6.2.3. Self-Energy of Electrons

The hybridization interaction V in Eq. (6.68) describes the processes whereby a
conduction electron can hop on or off the f shell of the local atom. This hopping preserves the
spin and orbital quantum numbers of the electron. At first it appears to have nothing to do
with spin fluctuations. As explained in the introduction to this section, the fluctuations in
occupancy can change the spin of the local electron. The local spin fluctuations are due to the
hopping process.
This process leads to several important new effects. The first is a new energy parameter
defined as r in Eq. (6.69). It is taken to be a constant. The second effect is that the resonance
lines at 00 = -L\_. L\+ become broadened by an amount proportional to r. The third new
effect is that the Kondo resonance peak appears in Af(OO) near to zero energy. It is an
important feature in the Anderson model, which is due to spin fluctuations.
The prior subsection discussed the energy levels associated with having n electrons in
the f state. Introduce a Green's function for this state
(6.97)
Most of the Green's functions in this book are for a single electron. There are also two-particle
correlation functions, such as density-density or current--current correlation functions. The
present Green's function is neither of these (unless n = 1). It is a Green's function for a state
of n electrons in the f orbital, where values of n have the range 0 :::: n :::: "Nt. Similarly, assume
there is a self-energy ~n for this n-particle Green's function, which can be evaluated using
Dyson's equation.
The hybridization interaction causes the self-energy ~n(OO). This function is calculated
using the noncrossing approximation (Bickers, et al., 1987). The result will be derived by an
intuitive argument. A rigorous definition is provided below. First consider the result for the
self-energy in the second order of perturbation theory
~n(OO) = ni,.~I(OO) + (Nf - n)S~~I(OO) (6.98)
~~I(OO) = 2. L Vink (6.99)

n N k 00 - (En+l - Ek) + ill
S'(O) (00) _ 2.L VI(l - nk) (6.100)
n-l N k 00 - (En- 1 + Ek) + ill
There are two terms. The first term (ft Ck ) in V adds an electron to the f levels, which
changes n to n + 1, while destroying a band electron in the state k. This process is propor-
tional to the occupation number nk of the band electrons, and gives the term S~~ 1(00). Again
the factor of Nf - n is the number of empty states into which the electron can be inserted into
the f orbital with n electrons. The second term (Ckf) in V takes an electron away from the f
level, which changes n to n - 1, while adding an electron to the band. This process is
Pfofortional to the probability (l - nk) that the band state is unoccupied. It gives the term
Sn~I(OO). These two processes provide the two terms in the above expression for ~n.
Note that the arguments of S(O) are Green's functions Gn without the self-energies.
1
S~O)(oo) = N~ VlnkG~O)(oo + Ek) (6.101)
The "non-crossing approximation" (NCA) is to evaluate these Green's functions while

including these self-energies. Put the self-energies in the denominators
~n(OO) = nS~_l(OO) + (Nf - n)Sn+l(OO) (6.102)
1 2
Sn+l (00) = N ~ VknkGn+l (00 + Ek) (6.103)
I
Sn_l(OO) NT
= 1" Vk(1-
2
nk)Gn_1(oo - Ek) (6.104)
The above nonlinear equations comprise a self-consistent definition for the self-energies ~n'
The values of n range over a limited number of possibilities (0 :::: n :::: Nf ), so only a few
functions of energy need to be determined. They can be solved by iteration on the computer.
The temperature enters into the expressions through the Fermi-Dirac occupation numbers nk'
The Kondo resonance is quite temperature dependent, and must be calculated anew for each
temperature.
The results are illustrated by a simple example. Assume a density of states with a
Lorentzian shape.
2D
geE) = E2 +D2 (6.105)
and place the chemical potential at E = O. This choice has the advantage of not introducing
band edges. The parameter D is an effective band width and r = 1tg(0)V(0)2 = 21tV(Oi ID.
In S~(oo), change variables E --+ -E which now makes S~ = Sn' The main integral to evaluate
is
Sn(OO) =!::J dE 2 nF(E)Gn(oo + E)

1t 1 + (EID)
(6.106)
Next discuss the derivation ofEq. (6.102). Again define In} as the exact eigenstate of Ho
while la) is the exact eigenstate of H = Ho + V. The definition of the Green's function is
1
Gn(oo) = {n l - H In} (6.107)
00-
= L l(a ln}1 2 • (6.108)

ex 00 - Eex + '11
Again a summation over the states la)(al was inserted in order to evaluate the energy
denominator with the Hamiltonian H. This result could also be obtained using the Lehmann
representation.
The self-energy ~n can be obtained by a similar argument. One starts with the state In}
and evaluates the self-energy in the second order of perturbation theory. The interaction V is
regarded as small, so an expression valid to O(V2) is considered adequate. The general
expression is
(6.109)
The intennediate states II) are taken to be the set la). The perturbation V has two tenns: one
(ft Ck ) adds an electron to the f orbital while removing an electron from the conduction band.
The other tenn (Clf) removes an electron from the f orbital while adding one to the
conduction band. These two processes generate the two tenns:
(6.110)
(6.111)
The summation over intennediate states (II includes the exact states (al and the eigenstates of
the band electrons. They are summed explicitly in the second line. A factor of In ± 1) (n ± 11
was also inserted, to use the fact that (n + Ilftln) and (n - Ilfln) are well defined. This
produces in (6.111) an expression which is obviously the Green's function. The last equation,
in the above set, is the expression for the self-energy in the non-crossing approximation.
A similar technique can be used to derive the spectral function Af( ffi). Start the derivation
with the general definition of the one-electron Green's function:
(6.112)
This Green's function is the usual one, similar to those in other chapters, where the definition
entails adding or removing a single electron from the system. It is the function which is of
physical interest, since it directly relates to the density of states and other quantities. On the
other hand, the Green's function Gn{ikn) is that of the n-particle state. It is the result of adding
n electrons to the system. It generally is hard to measure. In the present case Gn is useful since
it is used to calculate rgf.
The expression in brackets can be written exactly by inserting summations over exact
states la) and IA), as well as noninteracting states In + 1) and In' + 1).
L (nla)e-(~-t)E"(aln')(n'lfln' + I)(n' + 1IA)e- tEA (A In + I)(n + Ilftln) (6.113)

CXAn'
The usual approximation is to restrict the summation to the tenns with n' = n. Gunnarsson
and SchOnhammer (1983) consider the tenns with n' = n ± 1 and find they make a small
contribution. They are a type of vertex correction. Setting n' = n simplifies the expression.
After perfonning the integral over t
The matrix elements such as l(aln)1 2 and I(Aln + 1)1 2 are expressed in tenns of the spectral
functions An and An+! which produces the result
Taking the imaginary part of the retarded function gives the delta function 0«(0 + E - E')
which can be used to eliminate one of the energy integrals. After rearranging a bit,
A/(O) = z1 Eo Zn IdE e-~ An(E){nAn_1(E - (0) + (Nf - n)An+1(E + (O))

Nf
21t
E
(6.114)
The final equation is the result (6.114) for the spectral function A/(O). Note that the integral
for A/(O) is a convolution of the spectral functions for adjacent values ofn. In practice, the
only important terms in the summation are those with n = L - 1, L, L + 1. Again ignoring
terms of O(e-~A±) then this expression simplifies to
Af«(O) ~ -;:;;- IdE

1 -2 e- ~EAL(E){LAL_1(E - (0) + (P'i - L)AL+1(E + (O)}
ZL 1t
(6.115)
A numerical example of these equations is presented. For cerium, one typically chooses
values such as Ef = -0.5 eV, U = 6.0 eV, and r = 0.1 eY. This gives Eo = 0, El = -0.5
eV, E2 = 5.0eV, and the other levels are unimportant. The width parameter in the Lorentzian
density of states is D = 1.0 eY. The energy differences are LL = 0.5 eV and.d+ = 5.5 eY. The
ground state has L = 1 and the mixed valence is between L = 1 and L = O. Figure 6.5 shows
these spectral functions evaluated at T = 300 K. The spectral function for the ground state
A 1(E) is almost a pure Lorentzian with small width. This small width is provided by the small
value ofIm ~l due to its interactions with the neigboring state with L ± 1. From Eq. (6.102)
-2 Im[~l«(O)] = ~ L Vl[(l - nk)Ao«(O - Ek) (6.116)

Nk
+ (P'i - 1)n02«(O + E)] (6.117)
= r I+
1
dE 2 nF(E)[Ao«(O + E) + (Nf - 1)A 2«(O + E)]
(E/D)
The integration variable was changed in the first term E -+ -E which is only possible when
!! = 0 and with a symmetric density of states. Both terms in the integral contribute about
equally: Although E2 is further away in energy, and less effective, the factor of P'i - 1
20 160
.-
--
(a) (b)
W 140
.-
--«0
W 15
«~
120
100
10 80
60
5 40
0
-1
~~
-0_5 0 0_5 1 1_5 2
20
0
-1
\-
-0.5 0 0.5
E(eV) E(eV)
FIGURE 6.5 Spectral functions for cerium at T = 300 K: (a) Ao, (b) AI' Energy units are eY.
multipling A2 increases the importance of this term. The 1/Nf expansion is based upon the
premise that terms of O(~) are much larger than those of 0(1) and the latter can be
neglected. Our experience is that this assumption is not accurate numerically.
The spectral function for Ao has a main peak at Eo which is nearly Lorentzian. The
broadening is caused by r. However, there is also a narrow peak at the energy EI of the
ground state. It is due to the self energy Im[~o] defined by AI: the latter function has a peak
there. The peak at the ground state energy, in A o, is evidence that this state is fluctuating with
the ground state. The spectral function of A2 does not show a peak at the energy Eb since E2
is too far away.
The spectral function A/C) is found by the integral in Eq. (6.114). It is shown in Fig.
6.6. Since both AI and Ao have sharp peaks at the same energy Eb their convolution has a
sharp peak near to zero energy. The peak near to zero energy is the Kondo resonance. The
Kondo resonance has a narrow width. This narrow width is close in value to kBTK' where TK
is the Kondo temperature. The Kondo phenomena is caused by this sharp peak, and the width
of the peak-its characteristic energy-defines the energy scale which in turn defines the
temperature scale. The Kondo temperature for this case is about TK = 300 K. The height of
the Kondo resonance decreases rapidly with increasing temperature.
The Kondo resonance was first derived in the prior section, where it arose from the spin-
flip scattering of a conduction electron from a local impurity spin. The present Kondo
resonance is the same peak, calculated from the Anderson model rather than from the Kondo
model. In both cases, it arises from the interaction between the conduction and local electrons.
In the Anderson model, the Hubbard U appears not to playa major role. It only enters into the
energy denominators. However, this term is the major reason for the Kondo resonance.
Setting U = 0 causes the energy denominators to become equal. The Fermi factors cancel in
the numerator of the self-energy expression, and one has a simple function with dull features.
For U = 0 the Kondo resonance is not present. The energy U for pair interactions is the major
cause of the Kondo resonance. The Kondo resonance is the major macroscopic manifestation
of the spin-flip scattering. This resonance explains the unusual behavior in the heat capacity
and other measurements.
120 ~-- _ _ _ _ _ _ _ _ _ _ _ _ _- ,
100
80
~ 60
40
20
·0.3 -0.2 -0.1 0.0 0.1 0.2 0.3
E
FIGURE 6.6 Kondo resonance as a function of energy for cerium. Note the expanded energy scale.
As an example, calculate the heat capacity for a system with a concentration c of such
Kondo resonances. This calculation uses the formula
C(T) = JdEAtCE)E( - dn;iE») (6.118)
The heat capacity rises linearly with temperature at small temperatures. The formula for the
free-electron gas C '" m*k~T /EF is used to estimate the effective mass m*. One finds a value
for this system of m* = xme, where me is the mass of the electron and x is very large. The
system appears to have a very large effective mass. This result is the origin of the name
"heavy-fermion". The fermions are not that heavy. There is just a peak in the density of states
due to the spin fluctuations, which explains the large value of the heat capacity at low
temperature.
The low-temperature properties were not calculated correctly. If there is a concentration c
of such cerium atoms, then at low temperature the system develops a band of f-states, as the f-
electron hops coherently from cerium to cerium. The f-electrons develop their own Fermi
surface. This subject is rather interesting, and is reviewed by Zwicknagl (1992). The band of
f-states vanishes at higher temperature, as hopping becomes less probable. This behavior is an
example of electronic polaron model of Holstein (see Liu, 1988).
These results can be used to calculate the electrical resistivity peT) and the Seebeck
coefficient SeT). The usual assumption is that the scattering rate for the conduction band
electrons-their inverse lifetime-is proportional to a constant term from impurity scattering,
a term from phonon scattering, and a term from scattering from the local f states. The first two
terms are lumped into 'to(T). The latter term is taken to be proportional to their concentration
c and the spectral function AJ along with another constant u
1 1
't(ro) = 'to(T) + cuAtCro) (6.119)
't(ro) = 'to(T) (6.120)

1 + cu'toAJ (00 )
aCT) = eal JdE't(E)g(E) ( - dn;iE») (6.121)
TaS = al JdEE't(E)g(E) ( - dn;iE») (6.122)
where a I is a constant. The factor of geE) is the density of states. The factor of the square of
the Fermi velocity is part of al' This expression was derived by Mahan (1997).
The Kondo peak has an energy scale very different to that of the scattering due to
phonons or impurities, and the latter contributions are treated as constants. Note that the effect
of the Kondo resonance is to cut a hole in the energy spectrum of 't(00) since the factor of
AJ(ro) is found in the denominator. Figure 6.7 shows the calculated results of SeT) for this
model for cerium. Calculations of this kind were first done by Bickers et al. (1987). They
show there is nearly a universal curve for the different cases when plotting the results on an
axis of the logarithm of T / TK' Different choices of the model parameters (f.J. U. r. NJ ) give
different values for the Kondo temperature TK' The curve for the Seebeck has a universal
shape and maximum value ("'100 J.1V/K.) when plotted on a log scale. Note that S is the ratio
of two integrals: If the concentration c off levels is large, then the factor of c cancels between
the two integrals. This cancellation is the reasons the curve for S is universal.
402 Chap, 6 • Strong Correlations
1 2 0 , - - -_ _ _ _ _ _ _ _ _ _ _ _ _--,
100
80
CIJ 60
40
20
O + - - - - - - - - , - -_ _ _ _ _ _ ~
10 100 1000
T(K)
FIGURE 6.7 Seebeck coefficient S(T) for a model appropriate to cerium.
Figure 6.8 shows the Seebeck coefficient for CePd3 and YhAl3 plotted as a function of
temperature using data from Jaccard and Sierro (1982). Both the peak values, and the general
shape agree with the model calculations. These Seebeck coefficients are large for metals.
Most metals have values in the range of2-5IlVIK.. The Seebeck coefficient is large whenever
g(E}r:(E) has sharp, asymmetric, structure within 2kBT of the chemical potential (Mahan and
Sofo, 1996). In this case, the sharp structure is the Kondo resonance.
The Anderson model describes spin fluctuations. There are also charge fluctuations as
the electron hops on and off of the local orbital. The present theory neglects them. They could
be described by including a term in the Anderson Hamiltonian of the form
(6.123)
There is a Coulomb interaction between the local f electrons and band electrons. The latter
scatter from k' to k. The matrix element, and the scattering, depends upon how many local
electrons are present. One can call the ion valence Z and the sum over local electrons controls
the value of Z. When an electrons hops, say, off the orbital, then this sum changes. The theory
of this charge fluctuation encounters singularities identical to the Anderson renormalization in
-
100
• •
0 0 0 0
:III:: 80 0
i
• • • •
"0 • 0
==0 •0
>
eu 60 .0
0
'e
....... 40
o·
0
i!l
~ •
20
•••
•
0
0 100 T(K) 200 300
FIGURE 6.8 Absolute Seebeck coefficient as a function of temperature for CePd3 and YbAI3 . The results for cerium
are positive and for ytterbium are negative.
Sec. 6.3 • Hubbard Model 403
the x-ray edge problem, which is discussed in Chapter 9. They occur whenever one adds or
removes a local electron which has a Coulomb interaction with the local electrons. This
theory is discussed by Liu (1988).
6.3. HUBBARD MODEL
The Hubbard model has become the standard Hamiltonian for the investigation of
electron behavior in metals where the correlation effects are quite strong. Many analytical and
numerical techniques have been applied to this Hamiltonian. It is routine for young theorists
to announce that they have "solved the Hubbard model". In spite of this huge effort,
extending over many years, there are many features of this Hamiltonian in two and three
dimensions which are not understood.
This section discusses the Hubbard Hamiltonian HH and also the extended Hubbard
Hamiltonian HxH
HH = -tI:
jocr
C]+B,crCjcr + U~njtnn,
J
(6.124)
njcr = CJcr0cr (6.125)

HxH = HH +!j=ll,crcr'
I: njcrnlcr' V(Rj - R 1) (6.126)
i2
V(R) ~ (6.127)
R
The summation j is over lattice sites. The summation 0 is over the neighbors of j. The
summations over a, a ' are over spin indices. The term t is a hopping overlap energy between
first neighbors. It is assumed to be the same for all neighbors. This choice implies s-wave
symmetry: the orbitals are the same in all directions. Each site has the same symmetry. It is
possible to take a model with more complications, but the present one is hard enough.
The extended Hubbard Hamiltonian includes the Coulomb interaction between electrons
at distance sites. This term is essential to obtain the correct charge fluctuations. The pure
Hubbard Hamiltonian HH does a good job describing spin fluctuations, but does a poor job on
charge fluctuations.
The Hubbard model has several energy scales. The first is the on-site Coulomb inter-
action U, and the second is the bandwidth W = 2zt, where z is the number of nearest
neighbors: it is the number of terms in the summation over O. The behavior is very different
depending upon the ratio W /U. If this ratio is large, then the value of U is small compared to
the bandwidth. Here the free-electron model of metals works well. The case of simple metals
such as aluminum or sodium have W» u. The other limit is when U» w. Here electron
correlations are dominent. This case is only poorly understood. It is known that if the band is
exactly half-full, with one electron per site, then the system will be antiferromagnetic when
U » w. There have been many attempts, beginning with Penn (1966), to derive a phase
diagram at zero temperature to predict all of the different order parameters at different
concentrations n and different values of U / W. Phase diagrams are discussed in the next
section.
6.3.1. Spin and Charge Separation

Both the Hubbard model, and the extended Hubbard model, can be written in a form
which demonstrates the separation of charge and spin degrees of freedom. The extended
Hubbard model is discussed here. The Hubbard model is obtain, at any point in the derivation,
by merely setting to zero the interaction V(R).
The first step is to change the Hamiltonian to wave vector space.
(6.128)
(6.129)
(6.130)
The term in c(k) is the kinetic energy. The other terms are the various Coulomb interactions.
The Fourier transform of the extended Coulomb interaction has an interesting form
v( q) = L:>iq ' RW(R) (6.131)

#0
O=LV(q) (6.132)
q
The term withj 1= 0 is omitted from the summation since it is the Hubbard interaction U. The
term U is included separately from v(q). The second formula, in the above set, shows that the
summation of v( q) over all values of q gives zero. This happens because the summation of
v(q) over all q gives V(R = 0) and that term is omitted from the definition of v(q).
The actual summation must be done with Ewald methods. At small values of q,
v(q) ~ 41te2 j(q2Vo) where Vo is the volume of the unit cell. The potential v(q) behaves as a
normal Coulomb interaction at small values of q. However, since the summation of v(q) over
the crystal Brillouin zone of q values must give zero, then every region of positive value must
be cancelled by another region where v(q) is negative. Generally the negative regions of v(q)
occur at the points in the Brillouin zone with larger values of q. Negative values occur from
the phase factor exp(iq· R) in Eq. (6.131). It can be negative when the exponent is near to 1t
radians.
The Coulomb interactions are written in terms of the density operators of the electrons. It
is convenient to introduce the electron density operator Pcr(q) of a single spin, where
Cf = t,.J,. The charge density operator Pc(q) is the sum of both spin components, while the
spin density operator piq) is their difference
Pcr(q) = Lk CZ+q,crCkcr (6.133)
Pc(q) = Pt(q) + p./.(q) (6.134)

Ps(q = Pt(q) - p./.(q) (6.135)
The terms in the Hubbard interaction and in the extended Coulomb interaction can be written
in terms of these operators.
. U
U~nltnj{, = NLPt(q)P{,(-q) (6.136)
] q
U
= 4N L[Pc(q)pc( -q) - Ps(q)pS< -q)] (6.137)
q
L njcrnlcr' V(RjI) = 2~ L v(q)pc(q)Pc(-q)

!j#l,crcr' (6.138)
q
The two interactions can be combined into a charge and a spin interaction
(6.139)
(6.140)
It is interesting that the effective interaction for charge fluctuations is vc ( q) = v( q) + U /2.

This quantity can also be negative at large values of q. There is a simple proof due to Chen
and Mattis (1991) that v(q) + U must always be positive. Assume that the charge distribution
on an atomic site is (r), then
vr(q) = U + v(q) (6.141)
vr(q) = Jd3rd3r'(r)(r') L
j
~+ Rjl eiq ' Rj
Ir - r
(6.142)
The summation over j includes Rj = 0 since that is U. The summation can be written as a
periodic potential and therefore expanded in reciprocal lattice vectors G
(6.143)
(6.144)
The right-hand side of Eq. (6.144) is a sum of positive terms, so it is strictly positive.
Although v(q) + U is strictly positive, the effective Coulomb interaction vc(q) = Vr -
U/2 = v(q) + U/2 may be negative at large q.
The next step is to evaluate the correlation function of these operators. Assume the
system is nonmagnetic, so the average spin-up and spin-down correlations are the same. First
assume Vc = v.. = 0 then
PcCq, iqn) = !J~ d'teiq·~(T~Pc(q, 't)p!(q, O)} (6.146)

v 0
= -!J~
v 0
d'teiq·~(T~[p.j.(q, 't)pl(q, 0) + Pt(q, 't)p~(q, O)]} (6.147)
=p(1)(q, iqn) (6.148)
Piq, iqn) = -!J~ d'teiq·~(T~piq, 't)p!(q, O)} (6.149)

v 0
= -!J~
v 0
d'teiq.~(T~[pt(q, 't)p~(q, 0) + P.j.(q, 't)pl(q, O)]} (6.150)
= p(1)(q, iqn) (6.151)
Pciq, iqn) = -!J~ d'teiq·~(T~pcCq, 't)p!(q, O)} (6.152)

v 0
= -!J~ d'teiq·~(TApt(q, 't)p~(q, 0) - P.j.(q, 't)pl(q, O)]} (6.153)

v 0
=0 (6.154)
where p(1)(q, iqn) is the usual polarization function for the electron gas, which was discussed
extensively in Chapter 5. Terms such as (PtP.j.) are set to zero since different spins do not
correlate if there is no interaction.
These simple results are the basis for the separation of the spin and charge degrees of
freedom. This separation is exact in the Random Phase Approximation (RPA), but is only
approximate when evaluated with additional terms such as vertex corrections. The RPA is a
series of bubble diagrams obtained by evaluating (ppt) = p(1). In a nonmagnetic system
(Pcpt) = 0 and the spin and charge interactions do not mix.
For example, now include the interactions Vc•s and evaluate the correlations in Eqns.
(6.146) and (6.149) using the RPA. In both cases the result is obtained by summing the bubble
diagrams
(6.155)
(6.156)
The denominators of these two expressions are very different. One gives the modes associated
with charge fluctuations, while the other the modes from spin fluctuations. In Chapter 5 these
types of expressions were evaluated for electrons in energy bands with parabolic dispersion.
Let iqn ~ 00 + ill where 11 is infinitesimal. The limits depended on the dimensionless ratio of
x = oo/(qVF)' For large x one had plasmon modes, while for small x one had static screening.
In the present case, evaluate these expressions for energy bands described by a tight-
binding model, such as Eq. (6.130). For tight-binding bands define an average velocity
(v) = (VkE(k»). Again define x = O)/(q(v). The polarization function is
p(1)( 0) _ ~" nk - nk+q

(6.157)
q, - NT E(k) - E(k + q) + 0) + iT]
where the factor of two is for spin summation. Note that the normalization here is slightly
different than in Chapter 5: the summation is divided by N rather than by v = VoN where Vo
is the volume of a unit cell in real space.
x = 0: First examine the case that x = 0 which is achieved by setting 0) = O. Then the
polarization function is static. Next, consider the limit that q --+ O. For tight-binding bands
take E(k + q) --+ E(k) + Vk· q. The latter expression is also used in the occupation function
nk+q = nF(E(k) + Vk· q» which is expanded for small q
lim p(1)(q, 0) = ~ L nF[E(k)] - [nF(E(k» + vk· q(dnF/dE(k»]

q..... O N k E(k) - [E(k) + Vk . q]
=-2Nj (6.158)
N =
F
~
N k
L(- dnF(E(k»)
dECk)
(6.159)
The factors of v k • q cancel in numerator and denominator, and the final result depends only
on the energy derivative of the occupation function. This derivative is a delta function at zero
temperature. The density of states for a single spin state is defined as NF . Using this result in
the two susceptibilities (6.155) and (6.156) gives as q --+ 0
-2NF -2N!
(6.160)
Pc(q, 0) = 1 + NF[2v(q) + U] = 1 + k; /q2
P _ -2NF
s - 1- UNF (6.161)
N*- NF (6.162)
F -1 + UNF
12 = 8neN! (6.163)
s Vo
The Hubbard interaction causes a renormalization of the density of states, for charge fluc-
tuations, to N!. The screening at small wave vector has the Thomas-Fermi form, with a
screening wave vector ks define in Eq. (6.163).
The spin susceptibility Ps has only constants in this limit of x --+ O. Its denominator is
interesting since it is possible that 1 - UNF < 0, which is the Stoner Criteria for the onset of
a magnetic phase. The phase could either be ferromagnetism or else antiferromagnetism.
These possibilities are discussed in the following section. The Stoner criteria assumes that the
polarization function p(1)(q, 0) has its largest negative value at the center of the Brillouin zone
q = O. However, antiferromagnetism occurs when 1 < Up(1)(q, 0) for nonzero values of wave
vector q. Other interesting choices might have q at the edge or comer of the Brillouin zone. In
this section, it is assumed that the Stoner criteria is not satisfied, and there are no collective
magnetic states. Our interest is in the spin and charge fluctuations for the paramagnetic state.
Magnetic states are considered in the next section.
Large x: Now consider the other limit where 0) »q(v). Start with Eq. (6.157) and
rewrite it by changing variables k --+ -q - k' in the term with nk+q' which gives
p{l)(q, 0)) = -
N
2~ n (€(k)) [
F
1
€(k) - €(k + q) + 0)
+
€(k) - €(k
1+ q) - ] 0)
Next expand the denominators in a series in inverse powers of the frequency. The expansion
parameter is [€(k) - €(k + q)]/O). The first nonzero term is
lim p{l)(q,O)) --+

Ol-HX)
--i- 0) N
I>F(€(k))[€(k) - €(k
k
+ q)] (6.164)
The part in square brackets is expanded in powers of q. The first nonzero term is proportional
to q2. In the first-neighbor tight-binding model it is
2 2t .. 32q2
N ~ nF(€(k))[€(k) - €(k + q)] = N ~ nF(€(k)) ~(q . 8)2e1k Ii = d (-EK) (6.165)
2
(-EK) = -Z:>F[€(k)]€(k) (6.166)
Nk
p{l)(q,O)) --+ (3q);~~EK) (6.167)
The summation over k removes the angular dependence from exp(ik' 8) and then the
summation over 8 averages (q . 8)2 to (q3i /d. The symbol d is the dimension. A similar
result is obtained for the free electron gas. Note that the average kinetic energy (EK) is a
negative number, so (-EK) is positive. This result for p{l) in three dimensions is used in
Eqns. (6.155) and (6.156) to give
1 (q3i(-EK )
Pc(q,O)) =.3 0)2 _ (0); + cijq2)
(6.168)
:/:.2 2 _ 4rre2 32 (-EK )

ft O)p - 3Vo (6.169)
2 32 U(-EK )
Cu = 611 2 (6.170)
P( )_ ~ (q3)2(-EK)
s q, 0) - 3 0)2 + cijq2 (6.171)
The correlation functions Pc,s have denominators which depend upon frequency. Zeros of the
denominators represent excitations in the system of particles. The charge fluctuations have
their normal behavior of a plasma excitation O)p at long wavelength. The Hubbard interaction
gives the plasmons a quadratic dependence upon wave vector. Note that in the pure Hubbard
model, where the long range Coulomb interactions are set to zero, then the excitations in the
charge fluctuations become acoustic 0) = cuq. The Hubbard model gets the charge fluctua-
tions wrong, in that it predicts linear waves at long wavelength. The extended Hubbard model,
which includes the long range of the Coulomb interaction, gets the correct behavior of a
plasma frequency at long wavelength.
The spin fluctuations give damped modes 0) = ±iqcu. They have very different behavior
than the charge fluctuations. These results were derived only using the RPA. However, the
general trend of the results is thought to be valid for values of U ~ w.
6.3.2. Exchange Graphs

The Hubbard Hamiltonian contains the interaction U multiplied by the combination of
njtnj-l- for every site j. The argument in favor of this form of the interaction is that the
exclusion principle forbids two electrons of the same spin from occupying the same orbital on
the same site. Only electrons of the opposite spin can occupy the same orbital, and have the
interaction U. This feature is built into the Hamiltonian, which assumes the orbital is an s-
state which has only spin degeneracy.
There is another way to treat the problem. Instead of building the symmetry into the
Hamiltonian, one builds it into the wave function. In this formulation, the Hubbard and
extended Hubbard Hamiltonians are
HH = -t L C]+O.erCjer + U L njernjer' (6.172)

joer jerer'
njer = CJerCjer (6.173)
1
HxH = HH + 2" L njernler' V(Rj - R l) (6.174)
Ill.erer'
Now the interactions are between all electrons, regardless of the spin. The feature that no two
electrons, with parallel spin, can occupy the same site must be built into the wave function. In
our case, it must result from summing the correct sets of diagrams. Which set of diagrams
ensure the correct symmetry?
First convert the Hamiltonian to wave vector representation. For the extended Hubbard
model
(6.175)
Note the Coulomb interaction contains only the charge fluctuations, with the full interaction
vr(q). Writing this interaction as pp allows a particle to interact with itself. This contribution
is excluded. Starting from this interaction, the separation of charge and spin is less obvious.
Equation (6.175) is called the "full Hubbard model".
The various terms in perturbation theory are examined to find the cancellation which
ensures that only electrons of opposite spin occupy and interact on the same lattice site.
Generally, the cancellation is done by "exchange graphs". Repeat the discussion in Sec. 5.1.
and examine a few terms in the perturbation expansion. In the first order of perturbation
theory there are two terms: Hartree and exchange. In the discussion of the jellium model the
Hartree term gave no contribution: it utilizes the q = 0 term in the interaction vr(O), which is
absent in jellium. However, here there is a q = 0 term which is U.
LHer = U(ner + ncr) (6.176)

1
Lxer(P) = -Uner - N L v(q)np+q (6.177)
q
1
LHer + Lxer = Uncr - N Lq v(q)np+q.er (6.178)
where cr is the spin of the electron, while cr signifies the opposite spin. The symbols ner
denote the average concentration. A paramagnetic system is assumed, which has ncr = ner'
The total first-order self-energy, including Hartree and exchange, is Uncr for the Hubbard
(·)s
(b) (c)
FIGURE 6.9
model, which is precisely what one gets from the usual Hubbard Hamiltonian with the
interaction U Lj njanFr The extended Hubbard model also has an exchange term of the usual
form for charge fluctuations. The same cancellation occurs in the second order of perturbation
theory. The RPA part of the self-energy is the diagram in Fig. 6.9(b), while Fig. 6.9(a) shows
Hartree and exchange. Figure 6.9(b) is a closed loop connected to the Green's function by two
interactions. Let P a denote the closed loop polarization diagram for electrons of spin cr. In the
usual Hubbard model, the closed loop must have opposite spin of the electron, so it gives a
self-energy term with an effective interaction of U 2P ij. In the usual extended Hubbard model,
the interaction v( q) can interact with a closed loop of either spin. The self-energy in second
order, from the usual extended Hubbard model, is
(6.179)
The term with 2Uv(q) comes from Fig. 6.9(b) where the two interaction lines have U on
one side and v( q) on the other: the factor of two since there are two ways of doing this. For the
full Hubbard model in (6.175), with all spin interactions, the same diagram gives
(6.180)
which is different. Including the exchange diagram in Fig. 6.9(c) makes these two self-
energies more alike.
1
L~~(P) = - N2 L vr(q)vr(q'W'(P + q)'§(P + q')'§(P + q + q')
qq'
1
= - NL '§(p + q)Pij(q)[U2 + 2Uv(q)]
q
1
- N2 L v(q)v(q')'§(P + q)'§(P + q')'§(P + q + q')
qq'
Sec. 6.4 • Hubbard Model: Magnetic Phases 411
Adding the two self-energy tenns gives
!:~)(p) = N1 L ~(p + q)Pij(q)[2v(qi + 2Uv(q) + U 2 ]

q
1
- N2 L v(q)v(q')~(p + q)~(P + q')~(P + q + q') (6.182)
qq'
The first tenn is identical to (6.179). The exchange graph Fig. 6.9(c) cancels out the second-
order parallel spin tenns in the RPA tenn (Fig. 6.9(b» for the Hubbard model. Note that the
exchange graphs also produce exchange contributions from the charge fluctuations.
The RPA expansion for the extended Hubbard model treats the interaction U differently
from v(q). The exchange graphs must be included in order to cancel out the parallel spin parts
of the U interation. When doing RPA for the charge fluctuation interaction v(q) the exchange
contributions are usually omitted. Starting from (6.175) demonstrates this difference. This
different treatment may not cause problems. The exchange graphs from charge fluctuations
give large contributions to the energy of the electrons, but contribute little to the dynamics:
collective modes, etc.
6.4. HUBBARD MODEL: MAGNETIC PHASES
The Hubbard model has various phases with magnetic ordering. The usual metallic
phase, with no magnetic ordering, is called paramagnetic and is abbreviated "P" on phase
diagrams. It has equal number of electrons with spin up and down in the various bands, and
the spins have no ordering except local correlation. They have short-range but not long-range
order among the spin alignments. Other magnetic phases which are considered here are: (i)
ferromagnetic, where there is a net magnetic moment in some direction, which is caused by
having a majority of conduction electrons with spins pointing in the same direction. It is
denoted as "F" in phase diagrams (ii) antiferromagnetic ordering where there are an equal
number of conduction electrons with spins up and with spin down. The symbol "A" is used
for this phase. The distinction between PM and AF is that in AF the spins alignments have
both long-range order and short-range order. These three phases will be discussed in great
detail below.
Other magnetic orderings are possible. A charge-density wave is denoted CDW Its
signature is that the average value of the charge density operator
(6.183)
has a value of O(N) for some nonzero value of q. CDWs usually occur when the vector q
connects two regions of the Fermi surface, which is called nesting. Similarly, one fonn of
spin-density waves (SDW) is given by
(6.184)
The existence of CDW and SDW was first suggested by Overhauser (1964, 1978), and they
have been found in many materials. They are not discussed here. Some work has suggested
that the Hubbard model has regions of its phase space where there exist CDW and SDW
The ground state energy of the Hubbard model will be solved for the various magnetic
phases. Depending upon the variables, each of the three phases may have the lowest energy,
and be the ground state. What are the important variables? There are two energy scales: the
Hubbard U and the bandwidth W = 2zt of the kinetic energy, where z is the coordination
number. Usually the results are plotted according to the ratio of these two variables, such as
the combination U /W. The inverse ratio is used sometimes. Another important variable is the
concentration n of conduction electrons on the lattice. In the usual Hubbard model, the band
is full when n = 2. This case is uninteresting since none of the electrons can move and the
system is an insulator. The lattice has an average of one electron per site when n = 1. This
case is important since the AF phase occurs when n ~ 1. The temperature is an important
variable, since the magnetic phases disappear at high temperature. In two dimensions, they
disappear at any nonzero temperature, since the Mermin-Wagner theorem (1966) states there
is never long-range order in 2D.
The Hubbard model has particle hole symmetry, which means that the phase diagram is
symmetric. It has reflection symmetry around the point n = 1. Any phase that exists for
o < n < 1 also exists for n ~ 2 - n. This simple theorem is easy to prove. The reference
state for electron calculations is the quasiparticle vacuum 10}e' We define another reference
state IFB} which is the state if the electronic band is completely full of electrons: the state with
n = 2. The symbol "FB" denotes "full band". It is easy to calculate the energy of this state in
the real space representation. Since all states are full, the kinetic energy term CJ+oCjIFB} = O.
The state on j + () is already occupied and the creation operator gives zero. In real space,
all site occupancies nja are one, so the Hubbard interaction is UN. The result is
HIFB} = NUIFB}.
A hole is defined as an electron missing from the state IFB}. Let 10h denote the "hole
vacuum" and the hole operators are d, d t . Then
CjaIFB} = dj~IO}h (6.185)
CJaI FB } = djalOh (6.186)
Destruction of an electron is equivalent to the creation of a hole. The spin changes sign
between the electron and hole operator. The destruction of an electron of spin up leaves the
band with net spin down, which is assigned to the hole. Using the commutator for Ferrnions,
the number operator for electrons is changed to a number operator for holes
CJa0aI FB } = dja~7aI0}h (6.187)
= [1 - dJadjallO}h (6.188)
He =Hh + UN(I -n) (6.189)
Hh = tI:dJ+o.adja
joa
+ UI:n)~n)~
j
(6.190)
n)~ = dJadja (6.191)
The hole Hamiltonian Hh has nearly the same form as the original electron Hamiltonian He'
The difference is that the hopping term t has changed sign. This inverts the bands, which is no
problem. In the square lattice we can select as the center of the Brillouin zone the wave vector
Q = (n, n)/a. Since cos(9x + Qx) = - cos(9x ), this choice of band center inverts the bands
back to their original form. An alternate, but equivalent procedure, is to define the hole
operators with a minus sign (CjaIFB) = -d!aIO}h) on one of the two sublattices, which then
. th }
retams e negative sign in front of the hopping term. The hole problem is identical to the
electron problem. Any electron state of n has the same hole state of 2 - n. This proves
particle-hole symmetry. The constant terms in the transformation, NU(1 - n), are phase
independent so do not influence the selection of the magnetic ordering. This discussion
completes the digression into the explanation of particle-hole symmetry.
Another important variable for magnetic ordering is to choose the lattice for the
calculation. The results depend upon the choice of the lattice, and also depend upon whether
the calculation is done in two or three dimensions. The early work on the Hubbard model
concentrated on three dimensions, and most calculations were done for the simple-cubic (sc)
or the body-centered-cubic (bcc) structures. The discovery of high temperature super-
conductivity in the copper oxides, called "cuprates", heightened interested in layered
compounds. This focussed attention on the two dimensional lattice. Most recent work has
been done on the square lattice (sq) in two dimensions. This lattice has the advantage that the
Brillouin Zone is a plane and is represented nicely on a sheet of paper. The sq lattice has
coordination number z = 4 and W = 8t. The hopping energy, from the interaction with
nearest neighbors, is
E(k) = -2t[cos(ex ) + cos(ey )] (6.192)

eJ.1 =kJ.1a (6.193)
where a is the lattice constant.
6.4.1. Ferromagnetism
In thinking about the ferromagnetic state, it is useful to keep in mind two theorems. The
earliest was by Kanamori (1963), who proved that in the limit oflarge U, a system which is
doped slightly away from half filling is always ferromagnetic. The second theorem, by Vilk et
al. (1994), is that the the Hubbard model on the sq lattice is never ferromagnetic. These two
theorems directly disagree, and show that the Hubbard model is still not completely under-
stood.
The ferromagnetic state has a majority of spins pointing along one axis. This axis can be
in any direction, since the theory is isotropic. Take it to be spin up. Denote the concentrations
of up spin as n+ and down spins as n_. Use the symbol no where 0" = ±. Also, ncr is the
density of spins opposite to 0": the overbar denotes a negative value. The net concentration is
n = n+ + n_. Ferromagnetism occurs whenever n+ i= n_.
Denote by m the fractional magnetization.
(6.194)
Values of the magnetization range -1 S m S 1. The spin densities may be expressed as
n
n+ =2"(1 +m) (6.195)
n
n_ = 2"(1 - m) (6.196)
The Hubbard model was solved earlier using RPA. The first-order self energy was from the
Hartree interaction, and gives that Ecr(k) = e(k) + Una. In this approximation the density of
up spins is
(6.197)
(6.198)
The e integrals range from -n .:s eJ.L .:s n for the sq lattice.
Let pee) be the density of states. Equation (6.194) can be used to derive a self-consistent
equation for the magnetization at zero temperature. Then the occupation number becomes a
step function nF(e - 11*) -+ 0(11* - c)
pee) = J(2n)
d e O(e - e(k))
2
2 (6.199)
m =~ Jdep(e)[0(1l* - e + mnU/2) - 0(11* - e - mnU/2)]

n
1l*=1l-2 U (6.200)
n= Jdep(e)[0(1l* - e + mnU /2) + 0(11* - e - mnU /2)] (6.201)
The renormalized chemical potential 11* is determined by the concentration n as given in Eq.
(6.201).
The phase boundary is the point where the magnetization begins: it is the line where
there is a solution for the above equation with m = 0+. Expand the right-hand side of the
equation for m in a power series in the small parameter m. This side is proportional to m in
this limit. Cancelling the factor of m from both sides of the equation gives the equation which
determines the onset of ferromagnetism at zero temperature
1 = Up(ll*) (6.202)
Equation (6.202) is the first one to predict a phase boundary. There are several caveats. The
first is that the equation is very approximate. Only Hartree theory is used in its derivation,
which is neither sophisticated nor accurate. Secondly, this same result earlier was derived
earlier: it is just the Stoner criteria found in the prior section using RPA. The Stoner criteria
was given in terms of the polarization operator p(1)(q, co) in the case that co = 0 while q -+ 0
lim {I
q~O
+ U2 p(1)(q, O)} = 0 (6.203)
At zero temperature, in the indicated limit the polarization diagram is exactly equal to
-2p(Il*). Then the Stoner criteria is identical to Eq. (6.202). It is interesting that the simple
theory is identical to RPA.
The next step is to evaluate the density of states, which is given in tenns of an elliptic
function. First, define a dimensionless energy x = E/(4t) and dimensionless state density p(x)
f1t
pee) = 41t2
1
-1t d9x f1t-1t d9/)[E + 2t(cos(9x ) + cos(9y))] (6.204)
1 _(
= 4t p x) (6.205)
p(x) = -22 f1t d9 x J1t d9yi5[2x + cos(9J + cos(9y )] (6.206)

41t -1t -1t
p(x) = 22 K(~) (6.207)
1t
The elliptic function K(z) diverges when z = 1, which in our case is when x = O. This point
turns out to be half-filling: when n = 1. The resulting density of states is shown in Fig. 6.10.
Note that the dimensionless energy spans the range of -1 :::: x :::: 1. The band is half-full at
x = 0 where the density of states diverges to infinity at the van Hove singularity. The
divergence is logarthmic and the density of states is integrable through this divergence. In
fact, the total density of states integrates to unity:
1= fl dxp(x) (6.208)
which makes sense since p( E) is the density of states for a single spin configuration. The other
energy variables are also nonnalized to 4t: u = U /4t, v = 11/4t.
It may be possible to have a ferromagnetic state with partial magnetization 0 < Iml < 1.
The onset ofthe FM state is defined as m i= O. However, all the examples we have solved have
the feature that at the phase boundary m jumps to unity. Therefore, that will be assumed here,
2.0
1.8
1.6
1.4
1.2
x 1.0
Z-
0.8
0.6
0.4
0.2
0.0
-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
x
FIGURE 6.10 Density of states for an electron on a square lattice.
in order to simplify the algebra. In the PM state m = 0 and in the FM state Iml = 1. In this
case if ncr = n and ncr = 0, the chemical potential for the majority spins is just v* = v. The
feature that ncr = 0 requires that v - un < -1, which can be shown from the definition
ncr = J1
-1
dxp(x)E>(v - un - x) (6.209)
The integral is zero if the argument of the theta function is never positive.
The Stoner criteria for the phase boundary between FM and PM is found by solving
together the two equations
(6.210)
(6.211)
The first equation (6.210) is solved to give n(v), where v is the dimensionless chemical
potential which ranges -1 .:s: v .:s: 1. The second equation (6.211) is solved to give the phase
boundary (Uc /4t) also as a function ofv. This result is combined with the first calculation to
give the phase boundary (Uc /4t) as a function of n. The predicted phase boundary reaches
Uc = 0 at n = 1, which occurs because of the divergence in the density of states for the sq
lattice at half filling. Most other lattices do not have this divergence, and at n = 1 ferro-
magnetism only occurs above a critical value of the Hubbard parameter Uc' The sq lattice is
unusual in having this phase boundary at Uc = 0 at n = 1. All of this discussion is actually
pointless, since at n = 1 the system prefers to be in the antiferromagnetic state for small
values of U, rather than in the ferromagnetic state.
6.4.2. Antiferromagnetism
Figure 6.1 1(a) shows a lattice of circles and squares. If all of the points were alike, this
structure is the square lattice. The area of the unit cell is one square. In the AF state, the
circles form one sub lattice, while the squares form another. Antiferromagnetism occurs when
the circle sublattice has all electrons with spin up, while the square sublattice has all electrons
with spin down. Or vice versa. In real space, the unit cell in the antiferromagnetic state is the
square formed by four nearest neighbor circles. This cell has exactly twice the area of the
original unit cell. The occurrence of antiferromagnetism, in this case, doubles the size of the
unit cell. If Ao is the area of the cell in real space, and ABz is the area of the Brillouin zone,
then the theorem of areas states that (21ti = AoABz, When one doubles the area in real space,
the area of the Brillouin zone reduces in half
Figure 6.1 1(b) shows the Brillouin zone for the sq lattice. The units assume a bond
length of a = 1 so the minimal reciprocal lattice vector is G = 21t(±1, 0) or 21t(0, ±l). The
Brillouin zone for the sq lattice is the large square whose edges are solid lines. In the AF state,
the Brillouin zone is half as large. Then its boundaries are the dashed lines in the shape of a
diamond. The dashed line is also the Fermi surface in the paramagnetic state for half-filling
(n = 1). This line has an energy E(k) = O. Also shown in the figure is the vector Q = (1t, 1t)
which spans the Fermi surface at half-filling. Another similar wave vector Q' = (1t, -1t) is not
shown, but it spans the Fermi surface in the direction perpendicular to Q. The vector Q ± Q'
equals a reciprocal lattice vector, so that Q and Q' are regarded as equivalent. They play a
(a)
e
(-x,x)
,
(x,x)
/
/
/
,,
/
/
,,
/
/
,,
/
/
,,
/
, (0,0)
/
,, /
/
,, /
/
,, /
/
,, /
/
(b) (It,-x)
(-11:,-11:)
FIGURE 6.11 The square lattice. (a) In AF phase, circles have electrons of spin up, squares have electrons of spin
down. (b) Brillouin zone for sq lattice.
special role in the theory of antiferromagnetism. From now on the symbol Q is used to denote
either of the four vectors (±1t, ±1t).
Let k denote any wave vector in the Brillouin zone for the AF state. These are wave
vectors in the diamond shaped area in Fig. 6. 11 (b). The combination ofk + Q takes one to a
state outside of this reduced zone, but still within the original Brillouin zone of the para-
magnetic state. In a reduced zone representation of the AF state, the wave vector points k and
k + Q are the same point. In the PM state they are different points. So the onset of anti-
ferromagnetism is governed by order parameters of the form
(6.212)
(6.213)
where N is the number of lattice sites. These order parameters are similar to those for charge
and spin density waves. The AF state is a special kind of SDW. Either of the above two order
parameters can be used to describe the correlation in the AF state. The feature that a single
wave vector Q connects large segments of the Fermi surface is an example of nesting.
The sq lattice is a typical lattice for the description of antiferromagnetism. It naturally
divides into two sublattices, which doubles the unit cell in real space and halves the Brillouin
zone. There are other lattices which have a different behavior. Figure 6.12(a) shows the
honeycomb (hc) lattice which has two different lattice sites in each unit cell. Each site has
three nearest neighbors. Antiferromagnetism occurs when one site has electrons with spin a-
while the other site has electrons with spin cr. In this case the unit cell does not change size,
and the Brillouin zone in the AF state is identical to that of the PM state. Then the correlation
(8)
(b)
FIGURE 6.12 (a) The hc lattice. (b) The pt lattice.

functions for the onset of the AF state have Q = O. The hc lattice occurs in nature as the
structure of a single plane of graphite, where the vertices are carbon atoms. In graphite the
chemical a-bonds are the lines in the plane and are fully occupied. The conduction band is
composed of Pz orbitals which are perpendicular to the plane. They bond isotropically to the
three first neighbors, and the band structure is given accurately by the first neighbor tight-
binding model with an overlap parameter t = 2.4 eV The band is half full (n = 1). There is no
experimental evidence for an antiferromagnetic state, probably because U is too small. Both
the sq and hc lattices are bipartite, which means they have an obvious way of supporting an
antiferromagnetic structure with alternate up and down spins.
Figure 6.l2(b) shows the plane triangular lattice (pt). The vertices are lattice sites, while
the lines show bond directions. Each lattice point has six first neighbors. The pt lattice is not
bipartite. The basic structure in the unit cell is the equilateral triangle. This figure cannot
support antiferromagnetism: if one vertex has spin up, and another has spin down, then the
third has an equal tendency to be up or down, and is frustrated. It is believed that the pt lattice
does not have an antiferromagnetic structure at zero magnetic field at half-filling. All three of
these lattices (sq,hc,pt) support a ferromagnetic structure.
Now consider the antiferromagnetic transition for the square lattice. Start by examining
the function
(6.214)
(6.215)
The first expression is the charge density wave correlation function expressed in wave
vector space. The second equation is the same expression expressed in real space. For the
paramagnetic state, each sitej has equal probability of having an up or down spin. In this case
(nja ) = n12. Then the correlation function equals zero unless q = O.
A different result is obtained in the antiferromagnetic state. Then one sub lattice has all
spins a while the other sublattice has all spins cr. If one sublattice has spin up, and the other
has spin down, then
(6.216)
= n L eiQ ' Rj [1 _ eiq ' T ] (6.217)
N every other Rj
T = (±1, 0), or (0, ±1) (6.218)
The largest value of the above correlation function is when exp(iq . T) = -1 which is when
q = Q. This is another argument for why the wave vector Q is special.
Bogoliubov theory is used for the calculation of the AF state. Start with the standard
Hubbard Hamiltonian
(6.219)
In the Hartree approximation for the paramagnetic state, evaluate the Hubbard energy at
q = 0 to get two energy terms
(6.220)
= U L [ntnk", + n",nkt] (6.221)

k
H = L [s(k) + Uncr]nkcr - UNncrncr (6.222)

kcr
The last term in the effective Hamiltonian is a constant -UNncrncr which is included to get the
correct total energy. The Bogoliubov approximation overcounts the interaction, so this term is
subtracted to get the correct value. For the antiferromagnetic state, include the correlations
with q = Q. In this case the pairing of the last term in (6.219) is
The brackets are AQcr . The effective Hamiltonian in the Bogoliubov approximation is
H = L [s(k) + Uncr]nkcr +L [s(k + Q) + Uncr]nk+Qcr

kcr kcr
+ ULAQcr[C~+QcrCkcr + C~crCk+Qcr] - UN(ncrncr +AQcrAQcr) (6.223)
kcr
Constant terms are again subtracted from the effective Hamiltonian to get the correct ground
state energy. The summation over k is now only for the Brillouin zone of the AF state. The
states k + Q are added in order to sum over all of the states in the PM zone. For the sq lattice
s(k + Q) = -s(k). This Hamiltonian is easily diagonalized by using the transformation
IXkcr = cos(9k)Ckcr + sin(9k)Ck+Qcr (6.224)

~kcr = - sin(9k)Ckcr + cos(9k)Ck+Qcr (6.225)
Ckcr = cos(9k)lXkcr - sin(9k)~kcr (6.226)
Ck+Qcr = sin(9 k)lXkcr + COS(9k)~kcr (6.227)
The two operators Ckcr and Ck+Qcr are replaced by IXkcr and ~kcr' The angle 9k is determined
below.
This transformation is inserted into the Hamiltonian (6.223). Some algebra gives the
expression below. We omit writing the subscripts (kcr) on each term.
H = L{[lX t IX + ~t~]Uncr + [lXt IX - ~t~][s cos(29k ) + UA Qcr sin(29k)]

kcr
+ (IXt~ + ~t 1X)[UA Qcr cos(29k ) - s sin(29 k)]} (6.228)
Since a and ~ are to be pure eigenstates, the last term in the above Hamiltonian must vanish. It
vanishes by selecting the angle Elk to be
tan(2E1k) = UA Qcr (6.229)

e
(6.230)
H = ~::::£atcrakcr[Uncr + E(k)] + ~tcr~kcr[Uncr - E(k)]}

kcr
(6.231)
The energy spectrum has developed a gap which is twice UA Qcr . The dimensionless gap is
Ll = uA Qcr ' The energy E(k) is similar to that found in the BCS theory of superconductivity in
Chapter 10. There are some important differences. The main one is that the chemical potential
J..l does not go into the expression for E(k): note it is not written as J(e -
J..li + (UAQcri.
This difference is significant since the energy gap is only important if there are regions of
occupied electron states where e(k) = O. In the sq lattice, there are such energy states along
the dashed-diamond shaped curve in Fig. 6.12(b). However, if the density of electrons n is
changed away from n = 1 then these zero-energy states may not be occupied. The fact they
could have a gap becomes less important. Then the tendency for the AF state diminishes. The
lack of the chemical potential in E(k) is significant, and makes the theory much different from
the BCS theory of superconductivity.
As in BCS theory, in Chapter 10, the final step in the calculation is to obtain a self-
consistent equation for the order parameter AQcr . Again restricting the summations over wave
vector to the Brillouin zone in the AF state, and using the transformation to the states (a, ~)
1 t t
AQcr = Nt:[(Ck+QcrCkcr} + (CkcrCk+Qcr)] (6.232)
= ~ I: sin(2E1k)[atcrakcr - ~tcr~kcr]
Nk
(6.233)
U 1
AQcr = AQcr N t: E(k) [nF(E(k) - J..l*) - nF( -E(k) - J..l*)] (6.234)
The ~ state has the negative energy with respect to the Fermi surface, and it is the state which
is most occupied at low temperatures. The expression in brackets in the last equation, which is
the difference in occupation numbers for the alpha and beta states, is negative. The solution to
this equation requires that AQcr = -A Qcr . This choice is in agreement with Eq. (6.212). The
maximum order parameter occurs when AQcr = -AQcr . The self-consistent equation for the
energy gap in the AF state is
U I
1 = - I:-[nF(-E(k) - J..l*) - nF(E(k) - J..l*)] (6.235)
N k E(k)
U I sinh(~E(k))
(6.236)
= Nt: E(k) cosh(~J..l *) + cosh(~E(k))
At zero temperature the expression is written as

U 1
1 - - I:-[0(E(k) + J..l*) - 0(J.l* - E(k))] (6.237)
N k E(k)
Again let the density of states in the sq lattice be p(E). Recast this equation as
1 = U JO dE p(E) sinh(~E) (6.238)

-41 E(E)cosh(~Il*) + cosh(~E)
E(E) = JE 2 + (UAQol (6.239)
Note that the upper limit of integration is now zero, since the summation over k only
extended over the reduced Brillouin zone. The effective chemical potential 11* = 11 - Una is
determined by requiring that the number of particles remain n
1
n = N L [nkcr + nk+Qcr] (6.240)
kcr
1 t t
= N
kcr
L
[ockcrOCkcr + ~kcr~kcr] (6.241)
= Jo
-41
d Ep() cosh(~E)
E ----,-::----,-:--"'-----,--:-=--,-
cosh(~Il*) + cosh(~E)
(6.242)
If 11* = 0 then the right-hand side integrates to one, as it should. This identity is valid
regardless of the value of the energy gap. For values of n away from n = lone has to solve
this equation in order to obtain n(Il*), which therefore gives Il*(n).
The gap equation is (6.238). It is identical to the Stoner criteria. Recall that the Stoner
criteria for the AF state is given in terms of the polarization function P cr(Q, 0) for a single
spin component
1 = -UPcr(Q,O) (6.243)
1 nk - nk+Q
P cr(Q, 0) = N ~ E(k) - E(k + Q)
(6.244)
2 nk
(6.245)
= N ~ E(k) - E(k + Q)
The formula simplifies when using the feature that E(k + Q) = -E(k).
1 nk
P cr(Q, 0) = N ~ E(k) (6.246)
= J41 dE p(E) nF(E - 11*) (6.247)

-41 E
= _ JO dE p(E) sinh(~E) (6.248)

-41 E cosh(~Il*) + cosh(~E)
Here the summations over k extend over all of the Brillouin zone of the PM state. The Stoner
criteria is valid only at the transition temperature where the gap is zero. In this case the Stoner
theory is identical to (6.238) derived using the Bogoliubov approximation.
6.4.3. An Example
It is useful to give an example of these results. A simple model is best, in which all
results are analytical. To this end assume that the density of states p(E) is a simple constant
over the range of energy - W :::: s :::: W and is zero elsewhere. The normalization which
integrates to one electron is
_I -W<s<W
p(s} = { 2W -- (6.249)
o lsi ~ W
It is convenient to use dimensionless units such as x = s/W, v = J.l/W, and u = U/W.
Assume zero temperature so that Fermi functions become step functions.
In the paramagnetic state, then ncr = ncr = n/2. Define J.l* = J.l- Un/2 and v* =
v - un/2. The density of particles is
Lw dsp(s} = 1 + v*
Il*
n= 2 (6.250)
The factor of two is for spin degeneracy. Define the reduced ground state energy per electron
as Gj = Eo/(NW} where the symbo1j denotes P,F, or A for our three possible states. For the
PM state
(6.251)
Gp = ! [Ow dS[S + U~]p(s} - ~ Gr (6.252)
n un2
= H(v*}2 - 1] + (nJ.l* + l}u2: - 4 (6.253)
un2
Gp =!n(n-2}+4 (6.254)
In Eq. (6.251) the last term is subtracted since the Hubbard term is taken twice in the kinetic
energy part, where it is overcounted. In the last expression (6.254), v* = (n - I) is used to
obtain an expression for the reduced ground state energy as a function of occupation number
n and the Hubbard parameter (u = U / W).
The assumption in the ferromagnetic state is that the up (cr) and down (6") spins have
unequal occupation. Assume that up is dominent ncr > ncr. Again define the magnetization as
m = (ncr - ncr}/n. An interesting result is obtained by separately calculating these two
occupation numbers by integrating over the density of states
J
Il-un cr
ncr = dsp(s) = ![u + 1 - uncr] (6.255)
-w
Il - un "
ncr =J dsp(s} = Hu + 1 - uncr] (6.256)
-w
Subtract these two equations, and the difference in the occupations is the magnetization,
which gives the interesting result
u
nm = nm2: (6.257)
This equation has two solutions for n i= 0: either m = 0 or else u = 2. The latter is the Stoner
criteria: 1 = Up(O} = U/(2W}. What these equations actually mean is that there is no
solution for m i= 0 until u ~ 2. At this point then m jumps to one. The onset of the FM state is
given by the inequality U > 2 W. The calculation of the ground state energy is rather easy.
Since ncr = 0 there is no Hubbard energy: all the spins are alike to avoid any Hubbard energy.
The ground state occupation and energy for the FM state are
n cr =n=!(1+v) (6.258)
EG =N rw d€€p(€) (6.259)
iffF = ~(V2 - 1) = n(n - 1) (6.260)

where the expression for yen) is used to obtain the last expression. The corrrectness of this
result can be confirmed by comparing the ground state energies for the FM and PM states.
The FM state has the lowest energy whenever iffF < iffp which gives
un 2
n(n - 1) < !n(n - 2) +4 (6.261)
which has the solution that 2 < u for all values of occupation n. The phase diagram for a
system with just two ordering possibilities, PM or FM, is quite simple. The boundary is a
horizontal line at U j W = 2. Above the line is the FM phase and below it is the PM phase.
The line is flat because a constant density of states is assumed. The curve is actually a plot of
U(n) = Ijp[j..l(n)]. If the density of states is not constant, then the phase boundary has
curvature.
The third phase is antiferromagnetic (AF). First evaluate the Stoner criteria for the phase
boundary between the AF and the PM states
d€ u
1 = -U J
~
-pet) = --lnlvl (6.262)
-w € 2
2W
Uc(n) =- lnln - 11 (6.263)
This expression has three singular points. At n = 0,2 then lnln - 11 = In(1) = 0 and the
value for Uc diverges to positive infinity. The third point is at n = 1. There one has In(O)
which diverges to negative infinity, so that U/l) = O. This behavior gives the phase boundary
shown in Fig. 6.13(a). There are listed three phases where they exist alone. In the top-middle
region there could exist either AF or PM. The ground state energy of the AF state must be
calculated in order to compare the ground state energies of these two phases. The calculation
of the properties of the AF state proceeds in several steps, since there are several quantities to
determine. The first is the relationship between the occupation number n and the reduced
chemical potential v*. The relevant expression at zero temperature is
n= Ll dx[8(v* - J x2 + ~2) + 8(v* + J x2 + ~2)l (6.264)
(6.265)
Recall that the energy is E = !€2 + (UA Qcr )2 which when divided by W becomes Jx2 + ~2.
The upper limit of integration is zero, rather than one, since the integral is over the reduced
Brillouin zone which only goes halfway in energy. The two theta functions are the Fermi-
Dirac occupation functions for the ak and ~k operators. First consider the case that n < 1 so
that v* < O. Then the first theta function is always zero, while the second gives v* +
Jx2 + ~2 > 0 which can be solved for negative x to give
_J(V*)2 - ~2 > X (6.266)

U/W A or F
F F
3
P P
'I
I
0
(a) 0 0.5 n 1.5 2
Rectangular Density of States

5
4
F F
U/W
3
2 A
p p
0
(b) 0 0.5 1.5 2
n
FIGURE 6.13 (a) Preliminary phase diagram for the Hubbard model. Region with AF and FM together is to be
determined. (b) Final phase diagram for the Hubbard model.
so that the integral for the occupation number gives
n = 1 - J(V*)2 _ /),2 (6.267)
V* = -J /),2 + (1 - n)2 (6.268)
This equation provides one relation between v*, i1 and n. A similar derivation for n > 1 gives
n = 1 + J(v*i - i12 (6.269)
v* = Ji12 + (1 - n)2 (6.270)

The second step in the derivation of the AF state is to solve the integral (6.237) for the gap
parameter A. In dimensionless units it is the equation
(6.271)
(6.272)
The upper limit of the x integral for v* < 0 is (n - 1), as shown in the derivation of the
occupation number. Solve the above equation by defining S = exp(2/u), and then expo-
nentiate the equation. Also define y = 1 + A2 J
JA2 + (n - I i = S(y - 1) + (1 - n) (6.273)
After squaring this equation, there is a quadratic equation for y, which is solved to obtain
-S(1 - n - S) + IS(1 - n) - 11
y = ------:-c;:------- (6.274)
S2 - 1
If the factor of S(l - n) - 1 is positive, then y = 1 and A = O. In this case there is no

AF state. This result is expected, since the criteria that S(1 - n) > 1 is equivalent to
u < -2/1n(1 - n). From the Stoner criteria, the system is not in the AF state for this small
value of u. For S(l - n) < 1 there is a solution with y > 1 which gives the order parameter
y = J A2 + 1 = coth(2/u) - (1 - n)csch(2/u) (6.275)
A
2= {coth(l/u) - (1 - n) }2
sinh(l/u) -1 (6.276)
Jr-A- +-(n---1-)2 = csch(2/u) -

2 (1 - n)coth(l/u) (6.277)
This nice result depends only upon nand u. It can be evaluated, and then the result used to
find v* in Eq. (6.268).
The third and final step in solving the properties of the AF state is to evaluate the integral
for the ground state energy, again at zero temperature. In dimensionless units it is the integral
in (6.231)
cffA = fl dx{[~ + Jx 2 + A2]e(v* - Jx 2 + A2)

+[u2n - Jx 2 + A2 ]e(v* + Jx2 + A2)} - u(n/2i + A2/u (6.278)
The constant term Ij,2 ju has changed sign since AQcr = -A Qcr . The integral can be evaluated
easily by changing integration variables to ~ where x = -[\sinh(~) and Jx2 + [\2 =
[\cosh(~).
I
n I
-
-I
dxJx2 + [\2 = [\2 J~l
~
d~cosh2 ~ (6.279)
= [\2
2
[~+ 2!sinh(2~)]~1~n (6.280)
= [\2 ju + HJl + [\2 - (1 - n))(1 - ni + [\2]

1 - n = [\ sinh(~n) (6.281)
1 = [\ sinh(~I) (6.282)
Collecting these results gives
SA = ~n2 +H(1- n)v11 - ni + [\2 - )(1- n)2 + [\2 - )(1- ni + [\2 - Jl - [\2]
(6.283)
SA
u
= 4n2 +! [2(I-n)
sinh(1ju) - [2 - n(2 - n)]coth(lju)
] (6.284)
The first expression is the direct result of the prior algebra. The second equation is found by
using Eq. (6.276) for [\2 and doing more algebra. It has the advantage that the ground state
energy is a direct function of the variables (n, u). This expression is the final result needed for
the calculation of the phase diagram in the Bogoliubov approximation. For each value of n
and u, one can calculate the order parameter [\ and also the ground state energy per electron
SA-For u > 2 the AF energy is compared to that of the FM state [SF = n(n - 1)] which is
independent of u. The resulting phase diagram is shown in Fig. 6.13(b). For u > 2 the FM
state occupies an increasingly large part of the phase space, for different values of n. The
region of AF state gets shrunk to an increasingly small region near to n ~ 1. The reader is
cautioned that these results are quite approximate, and are based solely on the Bogoliubov
approximation to the initial Hamiltonian. Probably the AF phase does not extend down to
zero value of U at n = 1, but instead there is a critical value of Uc for the onset of the AF
state. This feature is discussed in the following section.
Figure 6.13(b) shows why the Kanamori theorem is perhaps reasonable. As U becomes
large the FM phase approaches the line n = 1 on both sides. His theorem is that in the limit
U --+ 00, then a bipartite lattice with N sites will be ferromagnetic if the number of electrons
differs slightly from N.
6.4.4. Local Field Corrections

The homogeneous electron gas was discussed in Chapter 5. There the RPA was found to
be a poor approximation for most quantities. It was necessary to add some additional
correlation into the response functions, by postulating the existence of Hubbard local field
corrections: GH(q) for charge fluctuations, and a similar function for spin fluctuations. A
similar approach to the Hubbard model would suggest the RPA result is changed to
(6.285)
(6.286)
where p(1)(q) is the usual polarization function of the electronic system, as calculated using a
single bubble diagram without vertex or other corrections. A better evaluation of the polar-
ization diagram would include vertex corrections. The present idea is to call these vertex
corrections the constants gc,s' which just mUltiply the correlation function. These vertex
constants will be a function of U / W. This simple idea follows directly from the Hubbard
approach. The question is how to calculate these vertex constants.
A simple and clever way was suggested by Vilk and Tremblay (1997). They used sum
rules on the correlation functions to provide formulas for the vertex constants. These sum
rules are derived from the exact definitions
Xc(q, iqn) = -~ J: dte iqn ' (T,Pc(q, t)Pc(-q, O)} (6.287)
Xs(q, iqn) = - ~ J: dte iqn ' (T,Ps(q, t)p,( -q, O)} (6.288)
The sum rules are derived by taking a four-vector summation over these correlation functions:
both over wave vector q and imaginary frequency qn' First do the summation over frequency.
This step simplifies the expression, since the summation over frequency gives a delta function
oCt)
1 .
-l:>,qn' = oCt) (6.289)
~ qn
1 1
A L XcCq, iqn) = - N (pcCq, O)Pc( -q, O)} (6.290)
I-' qn
1 1
A L X,(q, iqn) =- N (p,(q, O)Ps( -q, O)} (6.291)
I-' qn
The next step is to write the density operators in real space such as
Pc(q) = L [njt +nj.j,]eiq ' Rj (6.292)

j
p,(q) = L [njt - nj.j,]eiq'Rj (6.293)

j
1 2
NL(Pc(q)Pc(-q)} =L [njt +nj.j,] (6.294)
q J
1 2
N L(P,(q)ps( -q)} =L [njt - nj.j,] (6.295)
q J
These expressions are evaluated using nJcr = njcr to finally get

I
ncr = N "f;(njcr)
}
(6.296)
I
(nt n ,/) = -N L(njtnj
j
",) (6.297)
1
rw L Xc(q, iqn) = -[n + 2(nt n ",)] (6.298)
P q,q.
(6.299)
These two results are close to the final formula for the sum rule. The one for the spin
correlation is in its final form. A change must be made in the formula for the charge
correlation. In the expression for the charge fluctuations, there is a contribution to the sum
rule which is omitted in RPA. This contribution is the term with q = O. Then the density
operator is p(q, t) = Ne , which is independent of t. This term gives a contribution which is
N;l5 q=ol5n=o' It only contributes at zero frequency and at zero wave vector. It is usually
omitted when considering fluctuations. However, this term is included in the sum rule which
sums over all q and all n. Since it is omitted from RPA, also omit it from the sum rule.
Rewrite the sum rule for the charge correlation to be
(6.300)
The two sum rules are Eqs. (6.299) and (6.300). The derivation is unfinished without an
algorithm for determining the important correlation (ntn",). Vilk et al. (1994) argued that it
was just gs = (ntn",)/«(nt)(n",)). The sum rule for the spin correlation is now just a self-
consistent nonlinear equation for (ntn",). It is determined first. Afterwards, the charge fluc-
tuation sum rule can be evaluated to give gc. In this way the two vertex constants gc,s are
found self-consistently. Results of Vilk and Tremblay for the two-dimensional sq lattice at
half-filling (n = 1) are shown in Fig. 6.14. They plot the effective Hubbard constants
Uc = gcU, Us = gsU as a function of U, where the hopping energy t = 1. The charge
interaction Uc is significantly increased by the vertex interactions, while the spin interaction
15
u.
10
5
u,
o 2 U 3 4 5
FIGURE 6.14 Uc and Us for the Hubbard model on the sq lattice as a function of U at half-filling (n = I). From Vilk
and Tremblay (1997) (used with permission).
Us is significantly decreased. Similar changes occur at other values of filling n. Obviously,

this change in the effective interaction changes the locations of the phase boundaries.
Their theorem about the lack of an FM phase is based upon these types of calculations.
As U --+ 00 the effective Us saturates at a small value, which is less than needed for the onset
of the FM phase.
They also calculate the self-energy of the electron including the scattering by charge and
spin fluctuations
(6.301)
where a four-vector notation is used, as usual. The summation is over wave vector q and
frequency qn- Note that only one vertex is dressed, so that the interaction with the charge
fluctuations is UUc' rather than ul. The latter would be overcounting the vertex corrections.
They find this simple analytic self-energy function agrees quite well with numerical results
obtained using quantum Monte Carlo simulations.
Problems
1. Consider the bound states of the Kondo Hamiltonian for a localized spin one-half particle
interacting with a single particle in a band of continuum states:
H = L Ekcre~crekcr - L [Jzs(z)(e~tek't - e~t e k,,J + J-L(e~tek' ,S(-) + e~t ekkts(+)l (6.302)

kcr kk'
where S(z), SH and S(+) refer to the localized spins. Show that the following are exact bound states of
H:
<PI = Lake~tlt) (6.303)
k
<P2 = L bke~t I ,j,.) (6.304)

k
<P3 = Lddettl t) - eLI t)l (6.305)

k
<P4 = Lede~tl t) + eLI t)l (6.306)

k
where ak, bk , d k , and ek are coefficients and I t), I t) are configurations of the localized spins. Find the
eigenvalue equations for each case, and identify the spin triplet and singlet states.
2. Evaluate the integral A(E) in the Kondo model for each geE) and show that the coefficient of the
term In(EjW) is proportional to p(O). Also verify that the integral of geE) over the entire band
(-W < E < W) is unity.
A(E) = JO dE' ~(E') (6.307)

-w e - e
(a) gee) = Ij(2W)
(b) gee) = 2v'W2 - e2j(nW2) (6.308)
(c) gee) = (W - lel)jW2
3. Construct the zero-temperature collective state for the Kondo Hamiltonian when the local spin has
S = 1. Find the states with) = ~ and} =~. Which is preferred for J < O?
Problems 431
4. Consider the collective state for the antiferromagnetic Kondo model when the bands are half-filled.
Calculate ~ = -BEaIW by solving (6.48) using the density of states in part (a) and (c) of Problem 2.
Make a graph of ~ vs A defined in (6.52). Which case predicts the lowest binding energy?
5. Assume that an f level is split by the spin-orbit interaction into states with degeneracy 8 and 6,
separated by an energy ~ = 0.3 eV. Derive an expression for 0 = eJ - oE assuming that 0 « ~.
6. Derive the high-temperature (oe '" liT) and low-temperature (oe '" T) limits of the specific heat
per impurity in the Anderson model.
7. Derive some vertex corrections to Eq. (6.113) by taking n' = n ± 1. Can you draw a Feynman
diagram for these processes?
8. Derive the energy current for the Anderson model. Then use it to derive the expression (6.122) for
the Seebeck coefficient.
9. Calculate U and W for metallic aluminum. Estimate the bandwidth W ~ EF' Calculate U using a
charge density for the 3s atomic state in aluminum of P3s(r) = Nr4 exp( -lXr), where IX = 2.8a.u. What is
the ratio of U I W?
10. How large a negative number can you make Vc = v(q) + U /2 for the square lattice in two
dimensions? Assume that the charge distribution on each atom is a Gaussian of the form
(r) = exp(-r Ib 2 )/(1tb2 )3/2 and vary b to find the largest negative value of Vc at the comer of the
Brillouin zone.
11. Evaluate the function p(l)(q, 0) for the first-neighbor tight-binding model of the two-dimensional
square lattice. Choose q to be the comer of the Brillouin zone Qa = 1t(±, ±). Show that for this case that
e(k + Q) = -e(k) for all values of wave vector. Then show that p(l)(Q, 0) is infinity at zero temperature
when the band is exactly half full. In this case the Stoner criteria is satisfied for any value of U.
12. Use RPA in the extended Hubbard model to derive expressions for the susceptibilities Xn (q).
13. Find p(l)(q, (0) for the jellium model of the free-electron gas, in the limits: (a) x = 0, and (b) x
becomes large, where x = oo/(qvF)'
14. Evaluate the charge fluctuations in two dimensions using (6.155) in the limit oflarge x. For d =2
the small wave vector limit is v(q) --+ 21ti! /(qAo) where Ao is the area of a unit cell.
15. For the pure Hubbard model (v(q) = 0):

(a) Prove there are no RPA terms in the electron self-energy with two bubbles.
(b) Start from the all-spin form in (6.175) and show that there is a term such as U 3p2.
(c) Find the exchange graphs which cancel the term in part (b), in order to agree with (a).
16. Derive the energy band dispersion e(k) for the hc lattice in the tight binding approximation.
Include only first neighbor interactions, which are assumed to be isotropic. Find the reciprocal lattice
vectors, and the shape of the Brillouin zone. What are the minimum and maximum values of e(k).
17. Derive the energy band dispersion for the pt lattice in the tight-binding approximation. Include
only first neighbor interactions, which are assumed to be isotropic. Find the reciprocal lattice vectors,
and the shape of the Brillouin zone. What are the minimum and maximum values of e(k).
18. Use the order parameter A",,(Q) to solve for the AF phase of the sq lattice. In this case, the
effective Hamiltonian is
H = L [E(k) + Un"lnk" +L E(k + Q) + Un"lnk+Q" - UL A""(Q)[C~+Q,,Ck" + C~"Ck+Q"l

~ ~ ~
(6.309)
where k is summed over the AF Brillouin zone. One should get the same expression for the gap equation
as found using the order parameter A q ".
19. Derive a theory for the AF state in the hc lattice, where there is no Q vector since the Brillouin
zone does not change size in going to the AF state. The first step is to choose a suitable order parameter.
20. Assume the density of one-particle states has the form peE) = (2/nW2)./W2 - E2. Use Bogoliu-
bov theory to find the phase boundary between the PM and FM states.
21. Use the density of states of the prior problem to find the boundary of the AF phase using Stoner
theory.
22. A single sheet of graphite has the honeycomb lattice which is bipartite. The sigma bonds are fully
occupied by electrons. The n-bonds are half full which makes it a conductor. There is one electron per
site when undoped. The density of states vanishes at half-filling. Assume it has the form:
peE) = I~I, for -W::;E::; W (6.310)
(a) Derive the relation between D and W assuming the band holds two electrons.
(b) Find a formula for the chemical potential as a function of filling J.!(n).
(c) Derive the Stoner criteria for the onset of the FM state Uc(n) in the Hubbard model.
23. Derive the Vilk-Tremblay sum rules for the extended Hubbard model.
24. Solve the Hubbard model approximately by taking
Ho = -J.! L C]"Cj" +UL njtnIl-

j" j
V = -tLCJ+o."Cj"
jO"
Calculate the Free energy to second order in V which are the terms (Q o, Qj, (2).
(a) Use perturbation theory to evaluate the free energy of the Hubbard model to O(U2 ).
(b) Next show that the double site occupancy can be found from
1 aQ
(njtnjt) = N au
(c) Use the result of (a) to show that the above average behaves as (n/2)2[1 - U /Uel and estimate
the critical value Uc .
Chapter 7
Electron-Phonon Interaction
7.1. FROHLICH HAMILTONIAN
The Frohlich Hamiltonian describes the interaction between a single electron in a solid and
LO (longitudinal optical) phonons:
t t Mo 1 t t
H =L
pcr
EpCpcrCpcr + ffiO Lq aqaq + L 1.;-11Cp+q,crCpcr(aq + a_q)
qpcr V V q
(7.1)
c? ( m )1/2( 1 1) (7.2)
IX = -,; 2liffio Eoo - Eo
This Hamiltonian was derived in Chapter 1, with the form of the interaction given in Sec.
1.3.5. The LO phonons are usually represented by an Einstein model, i.e., the phonon
frequency ffio == ffiLO is taken to be a constant. Since there is a single electron, the Hamil-
tonian may also be written as
(7.3)
where r and p are the conjugate coordinates of the electron. The unperturbed electron is taken
to have free-particle motion with an effective mass m. Since there is only one electron in the
problem, the results are independent of the statistics of the particle. The same results are
obtained for any fermion or boson in the solid, such as holes, positions, etc., as long as they
are free to move. The phonon modes are unaffected by the one electron in the solid, so the
phonon self-energy is zero [actually of order D(1/v)], and the phonon Green's function qfi is
always qfi(O). The model also assumes that the motion is isotropic in direction and that the
energy bands of the solid are nondegenerate. These rather restricted conditions describe what
is called the Frohlich polaron problem. The model actually applies in some cases to
conduction bands of semiconductors and ionic solids which have their minimum at the r
point and have an isotropic effective mass. For other symmetry points in the solid, one usually
has to improve the model by taking into account anisotropy in the effective mass or degen-
eracy of the bands. The latter is important for holes in semiconductors. The Frohlich polaron
433
434 Chap. 7 • Electron-Phonon Interaction
problem was an important problem in mathematical physics during the 1950s. Numerous
mathematical techniques were tried out on this problem. Several are described here: Bril-
louin-Wigner perturbation theory, Rayleigh-SchrMinger perturbation theory, strong coupling
theory, and linked cluster theory. The Green's function method will be shown to be equivalent
at zero temperature to Brillouin-Wigner theory. Several other methods were tried, including
that of Low et al. (1953). The problem was most accurately solved by Feynman (1955). He
introduced a variational method based on path integrals. After lengthy algebra, he obtained a
result which even today is the best available. His theory is very lengthy and is not presented
here. The Feynman results will, however, be used as the standard to which other theories are
compared. The methods not discussed here, such as that of Low et al., give poor results when
compared with Feynman's method.
The polaron Hamiltonian (7.2) describes the motion of a particle while it is linearly
coupled to a system of boson particles. In the Frohlich Hamiltonian, the bosons are optical
phonons in a polar solid. The classical picture has the particle exerting forces upon the ions,
which respond and move. The ion motion creates new forces which act back upon the
particle. The nonzero ion frequency makes the reaction forces, of the ions on the particle,
retarded in time. The quantum nature of the phonons makes these forces occur in discrete
units. In both the classical and quantum pictures, the ion motion is pictured as a polarization
of the surrounding medium by the particle. The particle must drag this polarization with it
during its motion through the solid, which affects its energy and effective mass.
7.1.1. Brillouin-Wigner Perturbation Theory

The Brillouin (1932, 1933}-Wigner (1935) perturbation theory method is a historical
predecessor of the modem Green's functions method. It is equivalent to solving the equation
for the energy spectrum Ep of a particle of momentum p,
(7.4)
where ~ret(P' E) is the retarded self-energy. The problem is isotropic, so the functions depend
only upon the magnitude of the wave vector.
The method ignores the imaginary part of the self-energy of the electron. The particle
properties are actually described by the spectral function:
A(p, E) = -2 Im[E _ _1~ (p E)]

Ep ret'
(7.5)
If the imaginary part of the retarded self-energy is zero, it is replaced by an infinitesimal value
if>, and the spectral function just becomes a delta function, whose argument is (7.4). The
Brillouin-Wigner method is exact if the imaginary part of the retarded self-energy is zero.
Then if the retarded self-energy function ~ret is found exactly, the exact result is obtained for
the particle motion.
The self-energy function cannot be obtained exactly unless the Hamiltonian can be
solved exactly. This result has not yet been achieved for (7.2). In practice, one usually
evaluates a few terms in the perturbation expansion and thereby obtains an approximate
~ret(P' E). Equation (7.4) is solved with this approximate self-energy, which gives an
approximate solution. This procedure is Brillouin-Wigner perturbation theory.
For one particle interacting with a set of optical phonons, the imaginary self-energy does
vanish at zero temperature. This statement is only true for particles whose kinetic energy is
Sec. 7.1 • Frohlich Hamiltonian 435
less than the phonon energies (00. For a particle of initial energy E j to emit a phonon, it must
go to a final state Ef = E j - (00. However, if E j is less than (00, this equation cannot be
satisfied since the energy Ef cannot be below the bottom of the band. The process is forbidden
by energy conservation. Of course, a particle can absorb a phonon and increase its energy to
Ef = E j + (00. This step is always possible as long as there are phonons in the system. But the
number of phonons is proportional to the thermal occupation factor,
I
No = -e~=-m-o---l (7.6)
which vanishes as T ---+ O. At T = 0, the low-energy particle can neither absorb nor emit
phonons. Since these are the only two loss mechanisms, the particle cannot lose or gain
energy. Its mean-free-path is infinite, and the imaginary part of the retarded self-energy is
zero. Brillouin-Wigner perturbation theory is exact at T = O. Of course, it is exact only if the
retarded self-energy is found exactly. In practice, that is impossible, and Brillouin-Wigner
perturbation theory is usually a poor approximation.
The self-consistent energy function Ep is a smooth function of momentum at small
values ofp. It can be expanded in a power series:
p2
Ep = Eo + 2m* + O(p4) (7.7)
The quantity Eo is the downward shift of the band minimum from polaron effects. Eo is
negative and gives the amount an electron with zero momentum lowers its energy by inter-
acting with phonons. In Sec. 4.3 the self-energy of a fixed particle is - Lq M~ / (Oq' but it will
be different now that the particle can move. The quantity m* comes from the coefficient of the
p2 term in the momentum expansion. It is called the effective mass. In solids there are several
effective masses. Each energy band will have its own curvature at the band minimum, which
defines the effective band mass mb. The band mass should be used for min (7.2) and (7.3).
The "b" subscript is omitted, but m is identical to mb. The effective mass m* in (7.7) is that
resulting from the band curvature and polarons. It expresses the way that m (i.e., mb) is
changed by the polaron interactions. A formula for m/m* was given previously in (3.160).
The first term in the perturbation series for the self-energy has one phonon in the self-
energy diagram. The Feynman diagram is shown in Fig. 7.I(a). From the rules for
constructing diagrams, the self-energy is
(7.8)
The Matsubara summation was given in Sec. 3.5:
~(1)(p, lp. ) -_ M~" I ( No + nF __

. N.-=.-. o_+_I_------.,,-n::....F_)
+ (00 - ~p+q + lPn
'" - L.. 2 . (7.9)
v q q lPn (00 - ~p+q
(a) (b)
,. ",,--.
.---
....,
.....
,,-/......~- .........
,. ,,,.---.............. I
1«
I
..., \
,\
,
/
'" I I ' \ \
'
I / I ,\
I "
FIGURE 7.1
At zero temperature, No = O. The electron occupation factors nF are all zero if there is only
one particle in a band. Set ip = E - I.l + ill so the real part of the retarded self-energy is
(v--+oo)
Re[I:(I)(
p,
E)] _
- (2)3
MJ
Jd3q 1 (7.10)
ret n q2 E -COo- (p+q
) 2 / 2m
This result is also easy to obtain from the zero-temperature techniques of Chapter 2. The
energy E is now measured from the bottom of the conduction band. A free particle has E = 0
when it has no kinetic energy. The wave vector integrals may be done exactly. Results are only
given for E < coo. The formulas at nonzero temperature, for all values of E, are given else-
where (Mahan, 1972). Let v = cos(9) = P. q be the angular variable. The three-dimensional
integration is d 3q = q2dqdvdcp and the integral f dcp = 2n. The two remaining integrals are:
Re[I:(I)(p E)] - - MJ Joo d JI dv 1

ret' - (2ni 0 q -I COo - E + (P2 + q2 + 2pqv)/2m
The integrand remains unchanged if q --+ -q, v --+ -v. The answer is unchanged if the q
limits are altered to (-00,00) and the result is divided by 2; the two intervals (-00,0) and
(0, 00) contribute equally. The integral can be simplified by changing the q variable to
x=q+pv (7.11)
ffm
(I) MJffmJ oo JI 1 (7.12)
Re[I:ret (p, E)] = - 8n2 -00 dx -I dv COo - E + Ep(1 - v2) + x2
The x integral can be done first. The integral is of the form
Joo
_oox2
dx _ n0(A)
+A - .JA
(7.13)
where A represents everything else in the denominator of (7.12). The integral is nonzero only
when A > 0, which is always the case when COo > E. The remaining integral is just an arcsine:
J_J[COO-E+Edv(I-V2)]1/2= ,JE;sm
I
p
2 . -I ( Ep
COO-E+Ep
)
(7.14)
Re[I:~l(p, E)] = _ rt.CO~/2 sin-I ( Ep ) (7.15)

,JE; COo - E + Ep
Equation (7.15) is the result for the one-phonon part of the retarded self-energy when E > coo.
The parameter Eo is obtained from the self-energy at p = O. By using the expansion for the
arcsine, sin-I(x) = x + x 3 /6 + O(x5 ), the p --+ 0 limit is
3/2
Re[I:~(O, E)] = - rt.coo (7.16)
.jcoo - Eo
Equation (7.16) is the result for a zero-momentum particle. The result for a fixed particle is
quite different. In Sec. 4.3 it is shown to be
AE'= - MJL~ (7.17)

v q q2
The integral does not converge unless some cutoff is used at the high values of momentum.
Of course, the cutoff could be the size of the Brillouin zone or alternately the spatial extent of
the localized particle wave function. No cutoff is needed in the free-polaron problem because
of the recoil energy of the electron. The zero-momentum result is different from the result for
a fixed particle, a fact which is easily understood from the classical picture. The ions in the
vicinity of the electron get either repulsed or attracted to the electron. Their motion changes
the potential field felt by the electron. If the electron is fixed, it does not respond to this
change. The ions seek their new equilibrium position without any change in the potential
exerted upon them by the electron. A free electron can respond to the changing potential of
the ions. When the ions move in response to the potential exerted upon them by the electron,
the electron in turn begins to move in response to the changing potential of the ions. The final
solution describes the coordinated motion of the electron and ions. For a free electron, the
recoil of the electron makes polaron theory harder. A free polaron of zero momentum does
not mean the electron is fixed. It is constantly moving, in response to the continual interplay
with the force field of the ion. For p = 0, the motion has, on the average, zero momentum. But
the electron is moving in a random, stochastic fashion. When it has nonzero momentum, it
still has this random part of its motion, even as it is drifting. The total polaron momentum
(7.18)
commutes with the Hamiltonian and therefore is a constant of motion. Momentum is an

acceptable eigenvalue to assign to the polaron.
The imaginary part of the one-phonon self-energy may also be obtained from (7.9). With
iPn -+ E - Il and nF = 0, the imaginary part is
(I)
- Im[L ret (p, E)] = -nMJ Jd 3q
- 3 - 2 {N08(E + (00 - Eq +q ) + (No + 1)8(E - (00 - Ep +q )} (7.19)
(2n) q
There are two terms in the brackets. The first term corresponds to the absorption of a phonon
by the electron. This term is proportional to the phonon number density No, which vanishes at
zero temperature. The second term comes from the emission of a phonon by the electron. It is
proportional to the factor No + 1, which is nonzero even at T = 0. However, this term is
nonzero only if E > (00' so that the particle has more energy than the phonon, and is zero
when E < (00' Therefore, Im[L~~l(p, E)] = Oat T = 0, unless E > (00' The latter is possible in
a high electric field.
The Tarnm-Dancoff (TD) approximation constitutes solving Brillouin-Wigner pertur-
bation theory with only the one-phonon self-energy:
(7.20)
For zero momentum, the TD approximation gives the particularly simple equation
rt.(03/2
Eo =- 0 (7.21)
,J(Oo - Eo
The above is a cubic equation for Eo, which is easily solved. These results are summarized in
Table 7.1 for several values of rt.. They are labeled E TD .
The other columns are the Feynman results EF and also the results of Rayleigh-
Schri:idinger perturbation theory E~~2). E~2 is also evaluated in the one-phonon approximation
Table 7.1 Ground state energy of polaron in various

theories: Feynman (EF), Tarnrn-Dancoff (ETD ), and
Rayleigh-Schriidinger (E~l) for n phonons.
IX EF/roo ETD/roo E~~/roo Ell/roo

1 -1.013 -0.76 -1.00 -1.016
2 -2.055 -1.31 -2.00 -2.064
3 -3.133 -1.80 -3.00 -3.143
to provide a fair comparison. This equation is very simple, E~ = -IXffiO, as will be shown
below. E~ also includes the two-phonon terms. The table shows that the Tamm-Dancoff
approximation is a poorer approximation than Rayleigh-Schrodinger. It provides the larger
error when compared with the Feynman result, particularly for values of IX ,...., 1, which is
called the intermediate coupling regime.
The Tamm-Dancoff gives poor results in the intermediate coupling regime. The poor
answers occur because a gap was introduced in the excitation spectrum. The energy
denominator in (7.10) has the difference between the initial energy of the particle E and the
value in the intermediate state ep +q ' However, E has the minimum value Eo < 0, while ep +q
has the minimum value of zero. In order to get to the excited state, in this approximation,
there is an excitation energy of Eo. Tamm-Dancoff is a poor approximation because there
really is no excitation energy in the spectrum. The Tamm-Dancoff approximation just
happens to insert a gap, which explains why it gives poor results for intermediate coupling.
The above results are suitable for a single particle in a band. A different approach must
be used if there is a small but nonzero density no of particles in the band. Then one must
consider the change in the chemical potential oj.!. In insulators, the chemical potential j.! has a
negative value if the band minimum Em is defined as zero energy. If no is the density of
particles, and A is the deBroglie wavelength, then for the noninteracting system with
Maxwell-Boltzmann statistics one has that j.! - Em = kBTln(noA\ j.! - Em is negative if the
argument of the logarithm is less than unity, which happens when no is small. If polaron
interactions cause a change oEm in the energy of the band minimum, and if no is unchanged,
then the chemical potential must change according to oj.! = oEm.
In the polaron problem, Em = 0 and oEm = Eo. How does oj.! enter into the calculation?
In going from (7.9) to (7.10), there was an analytic continuation ip ~ E - j.!- oj.! + io.
Effectively E is replaced by E - oj.! in all of the following equations. Since oj.! = Eo, (7.16) is
replaced by
(7.22)
The right-hand side no longer depends upon Eo. The change in band minimum is Eo = -IXffiO'
This result now agrees with the Rayleigh-Schrodinger formula, which is given below. It is in
much better agreement with the Feynman result than is the Tamm-Dancoff approximation.
Brillouin-Wigner perturbation theory does better when more terms are included in the self-
energy expression. The anticipated improvement is difficult to test since the next terms are
quite formidable.
7.1.2. Rayleigh-Schrodinger Perturbation Theory

The Rayleigh-Schrodinger form of perturbation theory is the standard kind which is
described in quantum mechanical textbooks such as Schiff (1955). It is also called on the
mass shell perturbation theory. Energy and momentum are no longer separate variables. In
evaluating I:.(p, E) the energy E is set equal to cP' so the self-energy is just a function of one
variable p, or, equivalently, cp ' Of course, if the imaginary part of the self-energy is actually
zero, then energy and momentum are uniquely related. Rayleigh-Schr6dinger is an exact
procedure under the same conditions that Brillouin-Wigner perturbation theory is valid-a
zero value for the imaginary self-energy. The discussion is confined to the case of zero
temperature and cp < 000 for which Im[I:.] = O.
The two perturbation theories have a direct link to the potential scattering theories of
Sec. 4.1. The Rayleigh-Schr6dinger perturbation theory is analogous to the reaction matrix
theory of potential scattering, while Brillouin-Wigner is analogous to T-matrix theory. In the
former, one calculates real quantities on the mass shells, i.e., with E = cp . In the latter, one
calculates a complex quantity for a general value of E. Recall that the reaction matrix result
was not equal to the real part of the T-matrix result. Similarly, the Rayleigh-Schr6dinger
electron-phonon self-energy of the electron is not found from the Brillouin-Wigner form by
just setting E = cp and taking the real part. This happens to work for the one-phonon self-
energy but not when higher-order terms are included.
Successive terms in the perturbation theory for the self-energy function are called
I:.~(p). The superscript denotes the number of phonons in the self-energy term. The self-
energy is real, and the signature "Re" before the self-energy is omitted. The one-phonon self-
energy is
I:.(I) _ Mg Jd 3q 1 (7.23)
RS(P) - (2)3
1t
q2 cP - cp+q - 000
The energy denominator contains the difference between the initial state energy cp and the
intermediate state energy cp +q + 000; the latter has one phonon excited. The summation over
wave vectors is a summation over all intermediate states subject to momentum conservation.
The self-energy is the same as the Brillouin-Wigner result (7.10) after replacing E by cp- This
identity is true only for the one-phonon self-energies. For higher orders, in order to get
Rayleigh-Schr6dinger from the Brillouin-Wigner, replace E by cp and add some additional
terms.
The wave vector integrals in (7.23) are elementary. Indeed, just take (7.15) and replace E
by cP' with the result for cp < 000
I:.~(P) = -
3/2
IXOOO
(c
sin- I ...E.
)1/2
(7.24)
..;e; 000
In the limit where cp --+ 0, the simple result is:
I:.~(O) = -IXOOO (7.25)
It is no accident that the self-energy is just -IXOOO' The definition of the constant IX was chosen
to obtain a simple formula for the self-energy.
The polaron effective mass is derived by expanding the arcsine in a power series in
Jcp/ooo:
I:.~(p) = -1X000[1 +! cp +0(cp )2] (7.26)

6000 000
Ep = cp - IXOOO (1 + ~ ~) + 0(p4) (7.27)
= -IXOOO + cp (1 -~) + 0(p4) (7.28)

The effective mass is m* = m/(l - rJ./6). The polaron effects make the particle appear to be
heavier than the band mass m. The extra mass arises because of the interaction between the
electron and the ions. The electron causes a change in the equilibrium positions of the ions in
its vicinity. The change in equilibrium position arises from the mutual interaction between the
electron and ion, which was mentioned earlier. When the electron moves with a momentum p,
it must drag this ionic deformation with it. It takes energy to move this deformation. The drag
is what causes the mass to increase as the polaron coupling constant is increased.
In the Tamm-Dancoff approximation, the formula (3.160) gives the effective mass
m* = m(l + rJ./2)/(1 + rJ./3). The details of this calculation are assigned as a problem. This
result agrees with the Rayleigh-Schr6dinger result at small values of rJ., since both formulas
are proportional to m* = m[l + rJ./6 + O(rJ.2)]. However, they behave quite differently in the
intermediate coupling regime. The Rayleigh-Schr6dinger result predicts that something
calamitous happens at rJ. ~ 6. A similar catastrophe is not implied by the Brillouin-Wigner
formula, which is well behaved for all values of rJ.. The Rayleigh-Schr6dinger result is better.
Something quite important does indeed happen at rJ. ,... 6: The particle becomes localized. This
statement will be proved in the next subsection on strong coupling theory. It has already been
remarked, in connection with Table 7.1, that the zero-momentum values are much better in the
Rayleigh-SchrOdinger picture.
At zero temperature, the two-phonon self-energy in the Rayleigh-Schr6dinger theory is
I:.RS(P) = M4 L } 2 ( 1 ) [ 1
v2 ql q2 q 1q2 Ep - Ep+ql - 000 Ep - Ep+ql +q2 - 2000
x (E p - EP: ql - 000 + Ep - EP: q2 - 000)
- (Ep - Ep+1~ - oooi]

(7.29)
The wave vector integrals have been evaluated analytically for zero momentum. The integrals
are complicated, with the result:
I:.~(O) = -rJ.2 ooO[ In( + 2~) - ~J

1 = -0.1059rJ.2wo (7.30)
This result is combined with the earlier result for the one-phonon self-energy. The three-
phonon self-energy was evaluated numerically by Smondyrev (1986), who computed the next
terms in the Rayleigh-Schr6dinger series for Eo and m*:
Eo = -ooo[rJ. + 0.0159rJ.2 + 0.000806rJ.3 + O(rJ.4 )] (7.31)
; = 1 + ~ + 0.023627rJ.2 + o(rJ.3) (7.32)
Some values are tabulated in Table 7.1 and agree well with the Feynman values. The exact
ground state energy is a power series in rJ.. Just the first three terms in this series are known. It
is remarkable that the coefficients of the rJ.2 and rJ.3 terms are so small. Even for intermediate
coupling values I < rJ. < 6, the term 0.0159rJ.2 makes less than a 10% contribution to the total
value of Eo. Another way to say this is that first-order perturbation theory is a good
approximation even for intermediate coupling strengths. It is further evidence for the
superiority of Rayleigh-Schr6dinger over Brillouin-Wigner for the zero-temperature polaron
problem. It is assumed that higher-order terms also have a small coefficient and will also not
contribute much toward Eo for intermediate values of coupling constant.
It is natural to ask why the Rayleigh-Schrodinger method is so good. This point will be
explained in detail in the next section. But it seems worthwhile to summarize some of these
findings here. The independent boson model for a model system of electrons and phonons
was solved in Chapter 4. The polaron model may not be solved exactly, but the Rayleigh-
Schrodinger perturbation theory is related to the independent boson model. The self-energy
for the interacting system of electrons and phonons arises from terms in which various
numbers of phonons are virtually emitted into intermediate states. The one-phonon self-
energy in the Rayleigh-Schrodinger picture corresponds to the assumption that all the
phonons are virtually emitted independently of the others as in the independent-boson model.
The two-phonon self-energy describes the correlations between the virtual emission of pairs
of phonons. That is, I:~(P) contains the basic contribution from emitting one phonon, two
phonons and all numbers of phonons. The next term I:~~? describes the correlations
between pairs of phonon emissions. The following term I:is(P) describes three-phonon
correlations, which are not just pairwise correlations. The term 0.01590(2 is small, apparently,
because correlations are not important at intermediate coupling. For this reason, the 0(0(3)
terms are similarly small.
At first it seems surprising that I:~(P) has the self-energy from emissions of all different
numbers of phonons but without correlations. However, it is similar to our findings for the
independent boson model. There the exact self-energy was - Lq M~ / ffi q, yet this described a
ground state which had a mixture of large numbers of phonons. If you take the Rayleigh-
Schrodinger theory and set all kinetic energy terms equal to the same constant EO, you
immediately recover the independent boson model in which the particle had constant energy.
The one-phonon self-energy becomes
(7.33)
It is the exact self-energy when the kinetic energy is constant, which means that all higher
self-energy terms are zero. An inspection of I:~(P) in (7.29) shows that it vanishes when all
kinetic energies are a constant. However, it must vanish, since the exact self-energy for the
independent boson model is just a linear function of 0(.
The independent boson model was solved in Sec. 4.3 by a variety of methods. It was not
solved using Dyson's equation with all of its self-energy diagrams. At zero temperature, this
procedure would correspond to trying to solve it by Brillouin-Wigner perturbation theory. In
fact, the simple exact self-energy cannot be obtained this way-at least no one has ever
succeeded in doing it. Apparently one has to evaluate all the terms in all orders of perturbation
theory. It is not surprising that this approach does not work well for the polaron problem
either.
A lover of antiques will enjoy that, for the polaron problem, the old-fashioned pertur-
bation theory works better than Dyson's equations. One might ask whether this is a general
feature. If so, why bother to learn about Green's functions? A general rule cannot be given. In
some Hamiltonians the Rayleigh-SchrOdinger method is best, and in others Dyson's equation
is better. Coupled mode problems, such as solved for polaritons in Sec. 4.6, need to be
described by Brillouin-Wigner perturbation theory. Another example: whenever the
imaginary part of the self-energy is significant, it seems necessary to use a Green's function
approach. The old-fashioned techniques do not allow for imaginary energies in a systematic
manner. The theory of strongly coupled superconductors, for example, could not be done
without Green's functions. The actual message is that Green's functions are not the best way
to solve all problems. Which problems are best solved by Green's functions? Unfortunately, at
the moment, the only way to tell is by trial and error. No general rules are available which
predict when one perturbation method is better than another. Of course, for the polaron
problem, Brillouin-Wigner perturbation theory works fine as long as the chemical potential is
renormalized along with changes in the band energies.
The word polaron describes the coupled system of electrons and ions. The self-energy is
calculated for the mutually interacting system of electrons and phonons. Although it is the
electron self-energy, part of the energy resides in the ions themselves. The ions are moved
from their equilibrium position by the presence of the electron. The displacement takes some
vibrational energy, which is part of the electron self-energy, because it follows the electron in
its journey through the crystal. A simple analogy is the charge on a spring, which was
discussed in Sec. 1.1. The application of the electric field causes a deformation of the spring.
The final self-energy -(eF)2 12K contains a part which is the energy needed to compress the
spring to the new equilibrium. In the polaron motion, the springs governing ion vibration
occur locally wherever the polaron is at the moment.
The classical picture is that the ions in the polar lattice deform around the electron. The
quantum picture is the same, except that the motion of the ions is quantized. The number of
phonons is discrete. The amplitude of the ion displacements may not have a continuous range
of values but only discrete amounts, which are phonons. If the average number of phonons
around the electron is large, the quantization makes little difference. A laboratory spring does
not appear to be quantized, although it surely is, but the displacements which usually are
observed are so large, with such large quantum numbers, that the quantum nature is irrelevant.
The same is true with our polaron if it has a large average number of phonons. These phonons
are called the phonon cloud.
One may try to calculate the average number of phonons in the polaron. As with all
things in polaron theory, it may only be calculated approximately. Since the Rayleigh-
Schr6dinger method is the best, it is used here. From first-order perturbation theory, the wave
function of the electron at zero temperature is
(7.34)
where IO) is the phonon vacuum, so a~IO} has one phonon with wave vector q. The total
number of phonons is found by taking the expectation of the phonon number operator
Lk atak with this state:
(7.35)
The symbol Nu is used for number, to prevent confusion with other N symbols such as the
thermal average number ofphonons. The number operator on the first term in (7.34) is zero,
and on the second term it is unity. The first-order Rayleigh-Schrodinger prediction is

A
Nu(P) = 1 + A (7.36)
A(P) = M1i L 12 2 (7.37)

v q q (Ep - Ep+q - roo)
This wave vector integral is done by making the same variable changes that were used to
obtain (7.12) [v = cos 9, x = (q + pv)I.J2m]
(7.38)
(7.39)
(7.40)
so the final result is
(7.41)
For a particle with zero momentum, the result is A = a12, Nu = al(2 + a). Nu(O) is always
less than one, since the eigenstate only allowed for the creation of one phonon. It should be
remembered that Nu(P) is an average number of phonons. The actual number must be an
integer, which fluctuates from time to time about this average number.
These results apply for Ep < roo. An attempt to calculate the formula for Ep > roo gives
the result infinity. This answer is nonsensical, but the question was also nonsensical. A
particle with kinetic energy larger than roo will eventually emit a real phonon and lower its
energy to Ep - roo. The number of phonons about the electron is not a stationary quantity
when Ep > roo. It is not reasonable to try to evaluate a stationary matrix element such as Nu.
Another quantity to calculate is the mean-free-path of the electron. At zero temperature,
it is infinity when Ep < roo. However, at nonzero temperatures, there is a nonzero probability
No that some thermally excited phonons do exist. One of them may be absorbed by the
electron, thereby changing its energy and momentum. According to the golden rule (Schiff,
1955), the transition probability is
W
21t Jd q M1i
3
= -;::- - - 3 2 Noo(Eq + roo - Ep+q) (7.42)
" (21t) q
This formula is nearly the same as twice the imaginary part of the retarded self-energy given
in (7.19). The main difference is the replacement of E by Ep. The second term of (7.19),
proportional to No + 1, has been omitted. It gives the rate of phonon emission; it is zero if
Ep < roo. The integrals in the above expression are evaluated by changing variables to dxdv to
give
(7.43)
The two integrals may be evaluated in tum,
J OO
-00
2
dx8(roo + v ep -~) =
(roo + v
1
2
Ep)
1/2 (7.44)
Ji l l [Jroo + Ep + ,JE;]
dv
-I
1/2 =-In
(roo + V 2 Ep) ,JE; JroO + Ep - ,JE; (7.45)
to give the scattering rate for phonon absorption. The scattering rate is the same as the inverse
lifetime of the particle w = lit:
~= fro[ In[Jroo + + ,JE;]

tp
aNo
vE; Jroo + -,JE;
Ep
Ep
(7.46)
The mean-free-path Ip is found from the classical expression Ip = tpVp' so that one multiplies
tp by plm At zero momentum, the lifetime is evaluated as
1
- = 2IXrooNo (7.47)
to
This lifetime is used to calculate the electron mobility J..L = etolm* as discussed in Chapter 8.
7.1.3. Strong Coupling Theory

The strong coupling theory for polarons was invented by Landau and Pekar (1946).
Their theory was the first work on polarons, which even preceded the word polaron. Their
theory, and its subsequent improvements, is now known to be valid at large values of IX: hence
the current name of strong coupling theory. The method of calculation is radically quite
different from the prior perturbation theories. It is basically a variational calculation on a
Gaussian wave function. The calculation is done first and the physics is discussed afterward.
The Hamiltonian (7.3) is rewritten so that the phonon operators appear as displacements
Qq and their conjugate momenta Pq , which are chosen so they are dimensionless:
1 t 1.
Qq = .J2(aq + a_q), aq = .J2(Qq + iPq) (7.48)
_ __'_' (a -at)
Pq- .J2 q -q' (7.49)
H = p2 + roo L(~ + {fa) + M. A2 L Qq eiq . r (7.50)

2m 2 q q q °v q q
The wave function of the many-particle system «I>(r; Qq) must contain the coordinates of the
electron r and the ion displacements Qq. We make what appears to be a drastic assumption:
the electron is localized with a Gaussian wave function. Later this interpretation will soften,
and it is shown that the particle is not quite localized. But for the moment assume that the
total wave function is a simple product of electron and phonon coordinates:
«I>(r; Qq) = q,(r)"'iQq + 8Qq) (7.51)
q,(r) = (5nY/2 exp ( _ ~2 ? ) (7.52)

where ~ is a variational parameter. The phonon wave functions ~n are the usual harmonic
oscillator wave functions, except that they are centered about an equilibrium displacement
-oQq, which also needs to be determined. The first step in the calculation is to take the
expectation of the Hamiltonian over the electron part of the coordinates. For the two r-
dependent terms the Gaussian wave functions give
.1f(Qq) = J d3r<l>(r)H(r) (7.53)
p2 = 3~2
Jd3r<l> 2m 4m
(7.54)
Jd 3r<l>(r)2eiq . r = e- lf4ri (7.55)
The second integral is most easily evaluated in (x, y, z) coordinates. The expectation value of
the Hamiltonian is
(7.56)
(7.57)
The next step is to choose the equilibrium displacement oQq so that the term linear in Qq is
eliminated:
(7.58)
(7.59)
(7.60)
The first term on the right describes the harmonic vibrations of the phonons about their new
equilibrium positions -oQq. They have the harmonic wave functions ~n in (7.51) and the
eigenvalues (n + !)O)o. The second term 3~2 14m is the kinetic energy of the electron in the
Gaussian wave function. The last term is the potential energy of interaction between the
phonons and electron. The parameter ~ is varied to give the lowest energy for these last two
terms. First the potential term must be evaluated:
_1_ L2 =M~J d 3q e-q2/2P2 =1X(~20)0)1/2 (7.61)

20)0 ~ q 0)0 (21t)3 q2 m1t
E(~) = ~ -
2 IX (2)
~ 0)0
1/2
(7.62)
4m m1t
Set B = ~2 /2m, the last two terms are
2 )1/2
E(B) = ~B - .../EIX ( :0 (7.63)
dE _ 3 IX (2<00) 1/2
(7.64)
dB - '5. - 2,JE 1t
The variation shows that the minimum value is Po = (1X/3)J2<oo/1t so that the minimum
energy is
1X <oO
2 2
E(Bo) = -3"rr = -0.1061X <00 (7.65)
As usual with bound state variational calculations, the potential energy is twice the kinetic
energy, in agreement with the virial theorem. This result was obtained by Landau and Pekar.
The energy is proportional to 1X2 rather than IX.
The physics behind the variational theory is simple. The electron mass is much smaller
than the ion mass, so the electron moves much faster. Its motion creates a charge density, and
the ions respond to this average density. The small polaron occurs because of a strong
feedback loop. The ions move to create a local potential which traps the electron in a bound
state. The extent of the bound state depends upon the average motion of the electrons. The
ions displace in response to the average motion of the electrons.
In strong coupling theory, it is assumed that IX is very large and one tries to evaluate the
energy as a power series in 0(1/1X). The philosophy is very similar to that for deducing the
correlation energy of the homogeneous electron gas in Secs. 5.1 and 5.2. One perturbation
expansion was developed for small rs and another for high rs' The same is done for the
polaron. The Rayleigh-SchrOdinger expansion is valid at small IX, and the strong coupling is
valid at large IX. The interpolation between these two limits is remarkably easy. The best
available result for the strong coupling limit (Miyake, 1976) is
lim
IX---+OO
EO(IX) = -<Oo[0.10851X2 + 2.836 + 0(1/1X2)] (7.66)
The coefficient of the 1X2 term is 0.1085, which is remarkably close to the Landau-Pekar result
of 0.1061. This agreement shows that the Gaussian assumption is very accurate.
Figure 7.2 shows a plot of the ground state energy for Rayleigh-Schrodinger pertur-
bation theory as the solid curve labeled RS; also shown by the dashed line is the strong
coupling theory using (7.66). The strong coupling theory appears to predict a lower energy
state for small values of IX. This value is deceptive, since terms of O( 1/ 1X2) have omitted in the
asymptotic series, and these obviously diverge at small IX. The strong coupling theory results
should be believed only for values of IX above, say, about 5 or 6. The two curves almost touch
for values of IX rv 5. The value IX rv 5 is believed to be the crossover region between the two
theories. The correct theory is the Rayleigh-Schrodinger result for values up to IX ~ 5 and the
strong coupling theory for IX > 5. Such a curve is precisely the behavior of the Feynman
theory.
How large in space is the wave function of the small polaron? From the minimal value of
B o, the value of ~o is
(7.67)
a:
2 3 4 5 6 7
-I
-2
- -- ---
-0
-3
-...... ....
3 sc-l........ "
OF -4
.......
0
L&J
>-
(,!) -5
II:
'"
Z
'" -6
-7
-8
FIGURE 7.2 Energy of a polaron with zero momentum as a function of coupling constant. The solid line is the
Rayleigh-Schrodinger theory, and the dashed line is the strong coupling theory.
Po has the dimensions of meters-', so its inverse is roughly the size of the localized wave
function. For flOl o ~ 0.03 eVand m an electron mass, the polaron size is PO" '" 40 A /rx. For rx
values around 5 to 6, the size is 7 A, or about the size of the atomic unit cell. The localized
electron will be influenced by the fact that the ions are atoms. The initial Hamiltonian (7.3)
assumes a continuum theory for the ions, which is probably a reasonable approximation for
small values of rx but surely fails when the polaron size is of atomic dimension. Strong
coupling theory cannot be applied to real solids without additional modifications to account
for the atomic nature of the phonons. From now on polaron theory for rx > 5 must be
considered a mathematical model with interesting properties. The useful piece of physics is
that polarons become localized on atomic sites for values of rx larger than 5 to 6. The Frohlich
Hamiltonian does not describe their behavior after they are localized.
The theory of polaron localization has a catch. One needs to know the band mass m to
calculate rx. Today one can deduce m from a good energy band calculation. In the early days
they seldom had the required accuracy. Instead, it was customary to measure the polaron
mobility and the effective mass, and the combined results will yield m, m*, and rx (see Hodby,
1972). The catch is that this procedure works only if the polaron is mobile. After it becomes
localized, its mobility drops precipitously. The effective mass of a localized polaron cannot be
measured by cyclotron resonance. It is difficult to deduce what the polaron constant rx should
have been after the particle is localized. It can only be deduced accurately if the particle is not
localized. There are no experimental values of rx above 4. Perhaps this value is the cutoff for
localization of the polaron.
There are several other properties worth mentioning for polarons in strong coupling
theory. The first is that the localization may occur anywhere. If the phonons were really a
continuum, then the Gaussian packet would drift about and have an effective mass. The
effective mass is quite large: Allcock (1956) has estimated it to be m* = O.0208m(X4. For
example, at (X = 5 it is m* /m = 13. A similar estimate, m* = O.0227m(X4, is given by Miyake
(1976). The polaron is heavy because it has to drag with it the potential well of the phonons.
The prediction from Rayleigh-Schrodinger perturbation theory is that the effective mass
m*/m was proportional to 1/(1 - (X/6) so that something important seemed to happen at
(X '" 6. What happens is localization, which makes the polaron much heavier. The heavy mass
is not measurable because, as we said earlier, strongly coupled polarons are a model which we
cannot apply to real systems without modification.
Another feature of strongly coupled polarons is that they have excited states which are
also localized. For example, try to construct a polaron state which has p-wave symmetry. A
suitable trial wave function might be
(7.68)
One could repeat the variational procedure with ~ as a variational parameter (see Problem 9 at
the end of this chapter). It will be different from the s-wave value ~o, and it will be smaller.
The wave function is more spread out, and the electron is less bound. It is called the relaxed
excited state (see Kartheuser et al., 1969). If the excited state were stable, one could observe
actual optical transitions to this state from the ground state of s symmetry. The excited state is
not stable, since it is degenerate in energy with the ground s state plus L phonons, where L is
whatever number is necessary to make up the energy difference. Nevertheless, theoretical
calculations show a sharp line in the optical spectra, which has been interpreted as this s- to p-
wave transition. The strongly coupled polaron can create its own internal structure.
The strong coupling limit is calculated in the adiabatic approximation (see Allcock,
1962). The electron has sufficient binding energy that its oscillatory motion in the potential
well is much faster than the vibrational frequency of the phonons. The phonons do not have
time to adjust to the individual oscillations of the electron. Instead they adjust to the average
motion of the electrons. The ions are treated as a rigid potential well, in which the electron
adiabatically oscillates. A quite different picture applies to the weak coupling limit. There the
phonon energy is larger than that of the electrons. The picture is that the phonons, or ion
polarization, follow the electron during its motion.
7.1.4. Linked Cluster Theory

The linked cluster theory was advocated by Brout and Carruthers (1963) as a general
method to attack many-body problems. It is surprising that the method waited so long to be
applied to the Frohlich polaron. It was not done until the work of Dunn (1975), although some
earlier work of Mahan (1966) was similar. The surprise is even greater when it is realized that
this method is an excellent way of obtaining the Green's function G(p, t) for the polaron in the
weak and intermediate coupling regimes. In the prior section, it was concluded that the
Frohlich polaron Hamiltonian is useful only for these coupling values. The linked cluster
theory is applicable for all relevant coupling strength. One disagreeable aspect of the tech-
nique is that the final numbers have to be generated on the computer, since the formulas are
too complicated to permit analytical evaluation. This work is the price one pays for good
Green's functions.
The linked cluster methods were described in Sec. 3.6. A problem with one electron may
be worked in real time. The real-time formalism is valid even at nonzero temperatures. The
Matsubara technique is not needed since the interactions do not change the thermodynamic
averages. That is, when writing the Hamiltonian as H = Ho + V, Ho can be used in
exp( -~H), rather than H, since V does not affect the thermodynamic averages. The only
averaging is done over the phonons, and one electron in a macroscopic solid does not alter the
phonon energies. This substitution, of using Ho for H in the thermodynamic averaging, is not
appropriate in a many-particle system where the particles influence the phonon modes-as in
a metal or a heavily doped semiconductor. It is acceptable in the one-electron Frohlich
Hamiltonian. For the phonon Green's function at nonzero temperature, take
D(O)(t) = -i[(No + l)e-irooltl + Noeiroolll] (7.69)
1
No = eProo _ I (7.70)
where D(O)(t) has no q dependence, since the phonons are assumed to have no dispersion. In
this case D(O)(t) = D(t) since all the phonon self-energies are zero. The only concern is to
evaluate the electron Green's function. By adapting the formulas of Sec. 3.6 to real time, the
Green's function for the electron may be written as an exponential function of momentum and
time:
G(p, t) = dO)(t) exp[F(p, t)] (7.71)
dO)(p, t) = -i0(t)e- Epl (7.72)
The function F(p, t) is generated as a series of terms, which are obtained by the following
systematic procedure. The electron Green's function
G(p, t) = -i(OITeiHICpcre-iHIC~crIO) = -i0(t)(TCpcr(t)U(t)C~cr)
is evaluated in the interaction representation:

G(p, t) = -i0(t) L Wn(p, t)
n
_(-iinJI JI A A A At
Wn(p, t) - (2n)! 0 dt l • .. 0 dt2n (TCpcr(t)V(t l )· .. V(t2n )Cpcr(0»)

(7.73)
The S-matrix expansion generates a series of scalar functions Wn(p, t). They are resummed as
an exponential series of terms Fn(p, t)
G(p, t) = -i0(t) ~ Wip, t) = -i0(t) exp [ - iEpt + ~ Fn(p, t)]

(7.74)
00
F(p, t) = L Fn(p, t)
n=l
These two series may be equated term by term by assuming that both Wn and Fn are
proportional to powers of the coupling constant ocn • That is, define Wn to be all of the terms in
the S-matrix expansion which are proportional to ocn . Similarly, when the series is resummed
as an exponential, Fn contains all the terms proportional to ocn. For the polaron problem, the
terms in the S-matrix are zero when I is odd, since they contain an odd number of phonon
creation or destruction operators. Only the terms with even 1 contribute, and n = 2/. With this
convention, the relationship between some terms in the two series are:
FI = eisptWI (7.75)
F
2 = e
is
p
tw2 - 2!1 F2I (7.76)
F3 = eisp tw3 -
FF
I 2 - 3!1 F3I (7.77)
The terms are listed only up to n = 3, since that is already beyond what has been evaluated for
the polaron problem. So far only the first two terms have been computed. However, these
appear to be adequate to describe the polaron Green's function for low and intermediate
coupling. The term WI' or equivalently F I , was first evaluated by Mahan (1966). This term
has the form
(7.78)
Write the Frohlich Hamiltonian as H = Ho + V,
Ho = Lpcr EpC~crCpcr + roo Lq a:aq (7.79)
Mo e-it(Sp-Sp+q)
L
A A
Vet) = '" Cp+q,crCpcrAq(t) (7.80)

-vV qpcr q
Aq (t) = aq e- iCOqt + at-q e- iCOqt (7.81)
When evaluating the correlation function in (7.78), the phonon operators just give the phonon
Green's function (7.69), and the electron correlation function can be written as a product of
three G(O) by using Wick's theorem. The expression to evaluate is
(7.82)
For one particle in a band, the electron Green's function is G(O)(p, t) = -i®(t) exp[ -iEpt]. The
above expression simplifies to
(7.83)
The two time integrals may be done without difficulty:

dEpf WI (p, t) = F1(p, t) (7.84)
= - J(21t)3
d3 q MJ [. ( No + 1 NO)
qz 1t O_(p, q) + O+(p, q)
1- 1_ d
+ (No + 1) °_ (P ,qi + No p, q)2]
e itfL tn+
n (
(7.85)
u+
O±(p, q) = Ep - Ep+q ± roo (7.86)
The function Fl (p, t) is determined. It has several features which are immediately interesting.
There is a linear term in t. Its coefficient is just Rayleigh-Schrodinger one-phonon self-energy
I;(l)
RS(P) -
_ J(21t)3q MJq2 ( O_(p,
d3 No + 1 No)
q) + O+(p, q)
(7.87)
It is just the real part of the self-energy, since the principal part is taken of the energy
denominator. The self-energy is appropriate for nonzero temperatures. The result (7.23), for
zero temperature, is obtained by setting No --+ O. The first term describes processes whereby
an electron of momentum p emits a phonon of wave vector q and goes to the state p + q. The
second term describes processes where an electron in state p absorbs a phonon of q and goes
to p + q. The energy denominator contains ±roo depending on whether the intermediate state
p + q has one more or one less phonon than the initial state. The evaluation of these wave
vector integrals is assigned as a problem.
The energy denominator O+(p, q) can vanish during the q integration. It enters one term
in Fl as (0+)-2, which gives a possible divergence. However, there is no real divergence.
Collect together all the terms in 0+, including those from the self-energy, and find the
combination
(7.88)
which is well behaved when 0+ --+ O. There is no actual divergence in the wave vector
integration.
Most numerical work has been done for the momentum state p = O. In this case Fl (0, t)
can be obtained analytically in terms of the Fresnel integrals (Mahan, 1966) (s = troo)
E2 (x) = J"o vdteit21tt

r;c: (7.89)
VI (0, t) = - ~ {(No + 1)[-i,JSe-is + (2s + i)~ E2 (s) *]
+ No [i,JSeiS + (2s - i)~ E2(S)J} (7.90)
The easiest way to derive this result is to follow the method of Dunn (1975). His suggestion is
to go back to (7.83) and do the wave vector integrals before the time integrals tl> and t2
(7.91)
The wave vector integral has the fonn, for (t > 0)
Jo dq
oo e-itq2j2m =
y2ii
fiiiii (7.92)
Since the integrand is only a function of tl - t2, change the integration variable to t' = tl - t2,
and then change the order of integrations to get
(7.93)
The time integrals are now recognized as just Fresnel integrals, or integrals related to them.
The result (7.90) is derived easily. An approximate Green's function is obtained when we
approximate F(O, t) by its first tenn FI (0, t). The approximate expression can be numerically
Fourier-transfonned to obtain the spectral function
A(I)(E) = -2 1m [(-i) J~ dteiEt?I(O,t)] (7.94)
In this way, an approximation is derived for the imaginary part of the Green's function. The
real part may also be obtained.
The accuracy of this approximate Green's function is not obvious. About the only way to
judge is to also evaluate the next tenn F 2 (p, t), which was done by Durm. The next tenn
W2 (p, t) produces three different tenns, which correspond to the three Feynmann diagrams in
Fig. 7.3. Usually such Feynman diagrams are associated with Dyson's equation, which is not
the case here. Such diagrams are convenient to use, but here they have a different inter-
pretation. For example, the diagram in Fig. 7.3(c) is not evaluated in Dyson's theory, since it is
just the square of the one-phonon tenn. This identity is not true in the linked cluster theory, so
it must now be evaluated. In fact the theory is not one of "linked clusters" at all, so the name
is inappropriate. All diagrams are evaluated. A more descriptive name would be something
like exponential resummation, which lacks pizzazz.
Some numerical results ofDurm (1975) are shown in Fig. 7.4. In Fig. 7.4(a) the spectral
functions for p = 0 and kBT = 0.4000 are shown for three values of 0(. Durm found that for
0( < I the two-phonon results did not change the spectral function at all. The curves calcu-
lated with F = FI + F2 were identical with Mahan's, which used F = Fl' For higher values
of 0( there were changes, but they were not dramatic. Figure 7.4(b) shows a comparison of the
one-phonon and two-phonon spectra functions for 0( = 6. The theory is not applicable at this
high a value, because one should be using strong coupling theory with its internal excited
states. However, one can conclude that the linked cluster theory converges rapidly for low and
intennediate coupling strengths.
The spectral functions shown in Fig. 7.4 describe the Green's functions for all
frequencies 00. The low-energy exponential tail is for states below the conduction band
minimum. The peaks at multiples of 000 are states with different numbers of phonons asso-
ciated with the polaron. These are similar to those found for the independent boson model,
(a)
- .... , ( b) (e)
,
".-- ,..--;,.>c'(-- . . . ,
,\
",!
/
I __ \ / /
I I ,
(/,
I,
\
( ! \ \
FIGURE 7.3
rJ)
CD
!"I
.......
•
."
...
0'
2:
(b) a=6 r;'
1.0 :r
::t
DI
1.2
3
14 o
=
=
iii·
1.6 =
3. .-;:::-0 1.8
o
3
C>
E
2C> -2
H
'--' E r;
.,.o H -3 "
.J .,.
o ,,
.J -4 /
I 151 order
/
-5 I 2nd order
/
/
-6 I
I
I
-7 I
I
I
-8 -6 -4 -2 o -10 -8 -6 -4 -2 o
W,n-- wln-
FIGURE 7.4 The spectral function of the polaron with zero momentum and ~roo = 2.5. (a) Results for three coupling strengths, calculated with one- and two-phonon clusters. (b) A
comparison of the results for one-phonon (dashed line) and one- plus two-phonon (solid line) clusters. Source: Dunn (1975) (used with permission).
8i
Co)
except that here the peaks are broadened by the recoil of the electron. For (X = 2 the lowest
two peaks are the main ones. But for (X = 4 and 6 the lowest peaks have small amplitude,
which shows the influence of the renormalization factor.
The one-phonon term F J is very similar in form to the independent boson model.
Compare (7.84) with (4.257) and (4.259). The only difference is the substitution of
n± = Eq - Ep+Q ± Olo in the independent boson model, instead of ±0l0' F J is given the same
interpretation. It describes the Green's function when successive emissions of phonons are
independent. The approximate Green's function, with only Fb does describe an electron
coupled to many different numbers of phonons-but all uncorrelated. The addition of F2
included correlations between pairs of phonon emissions. F2 has little effect on either the self-
energy or the spectral function itself at p = O.
The linked cluster theory is quite different from the Dyson's equation approach. There
the one-phonon self-energy L(1)(P, E) adds only a one-phonon peak to the self-energy. The
two-phonon peak comes from the two-phonon self-energy, etc. In order to get a full Green's
function, with many phonon peaks, one needs to evaluate many self-energy diagrams. These
higher-order diagrams are difficult to evaluate and have never been calculated. The Dyson's
equation method has not been solved to the same accuracy as the linked cluster theory. Of
course, one can also obtain G(p, t) from Feynman theory.
7.2. SMALL POLARON THEORY
Polarons become "small" when they become localized, as in the strong coupling theory.
In contrast, Frohlich polarons are sometimes called "large." Small-polaron theory assumes
that the size of the polaron corresponds with atomic dimensions. It recognizes the periodicity
of the solid and thereby assumes that the motion of the particle is no longer translationally
continuous. Instead, it assumes that the particle, usually an electron, may occupy an orbital
state <p(r - R) centered on atomic site Rj . The orbital states are identical on each site, so
there is periodicity. The particle may move from site to site, exactly as in the tight-binding
model. The motion from site to site may be caused by the overlap, or nonorthogonality, of the
orbitals on adjacent sites. The phonons are coupled to the particle at whichever site it is on.
The following Hamiltonian describes the motion of small polarons:
(7.95)
Q]. = ""
L.... X q e · RJ(a q + at-q ) ,
q
iq X-
q -
~2MNOl q
(7.96)
QJ. = "" ~ eiq · RJ (a q+-at)

L.... X q~q q (7.97)
q
Ve~) = DJ L C] CjQj (7.98)

jq
(7.99)
(7.100)
Sec. 7.2 • Small Polaron Theory 455
where Cj is the destruction operator for a particle on site Rj . The summation 8 is over the
nearest neighbors. Spin indices are not important in this problem and are omitted. If (7.95) is
compared with the Frohlich Hamiltonian (7.2), only the first term is different. The present
Hamiltonian (7.95), which includes the periodicity of the solid through the tight-binding
model, is much more realistic.
Three different forms are provided for the electron-phonon coupling Vep- The first one
Ve~) treats the ion displacement Qi as a scalar. This model is appropriate for organic solids,
where the vibrational modes are those of the molecule. As an electron hops from molecule to
molecule, it can excite vibrational modes of the molecule it is resting on. Then Qi is the
amplitude of the normal mode. The second form of the interaction vJ;) arises from the
Coulomb interaction between the electron and the neighboring ions. When an electron is on
ion Rj then there is a contraction or expansion of the separation between this ion and its
neighbors. The third form of the electron--electron interaction V£) is due to the phonon
modulation of the hopping. The first term in (7.95) has a hopping parameter J(8) which must
depends upon the distance to the neighbors. If this distance is modulated by ion vibrations,
there is a corresponding electron-phonon interaction. In this case the coupling constant
D3 = dJ /dr. Both interactions V(2) and V(3) are present in all tight-binding solids. The first is
appropriate for molecular solids.
The discussion of small polarons will follow the classic treatment of Holstein (1959),
which followed the Eioneering work of Tiablikov (1952) and Yamashita and Kurosawa
(1958). They took Ve~ as the electron-phonon interaction. The same analysis works for Ve~).
The small-polaron Hamiltonian exhibits two types of behavior which are very different.
Each type of behavior is described in detail. Afterwards, the interesting problem is discussed
of the transition region between these limiting cases.
7.2.1. Large Polarons

The first class of behavior is "large" polaron motion, of the Frohlich type. It occurs
whenever the bandwidth zJ is large, where z is the coordination number. The condition on the
bandwidth is stated later with more precision. When the bandwidth is large, the Hamiltonian
is solved in wave vector space. Transform to collective coordinates:
(7.101)
(7.102)
M(k, q) = Xq{D J - D2~q . F(q) + D3~q • [F*(k + q) - F*(k)]} (7.103)

F(k) = L 8eik • II = -iVYk (7.104)
II
where ~q is the unit vector for phonon polarization. The electron-phonon interaction has the
usual form, except that the matrix element M(k, q) depends upon k as well as on q. The
dependence upon k comes from the phonon-modulated hopping. This term will be dropped
(D3 = 0), so that the electron-phonon matrix element M(q) == Mq will depend only upon q.
The small-polaron Hamiltonian has a close resemblance to the Frohlich polaron

Hamiltonian. The only difference is the replacement of the free-particle energy Ek = k2/2m
by the tight-binding form Ek = zJYk' and Mq has a different dependence upon q.
The wave vector summation for particles extends only over the Brillouin zone. In a
realistic model of a solid, there would be many bands to be summed. For a large bandwidth,
the particle will confine its motion to states near the bottom of the band. For J negative, the
bottom of the band occurs at k = O. Expand Yk about the point k = 0, and find in cubic
crystals Ek = zJ(1 - (k'Oi /6), so that the particle has an effective mass ofm- 1 = zIJI'02/3. If
the polaron coupling strength Mq is small, this Hamiltonian can be described by weak
coupling theory. There is a slight change in the particle's energy and effective mass because of
polaron effects. These changes must correspond, in the tight-binding model, to a change in
bandwidth. The polaron self-energy, in first-order Rayleigh-Schrodinger perturbation theory,
is
(I)
~RS(k)
, , 2 {Nq
= ~Mq
+ 1- nF(Ek+q) NqnF(Ek+q)]
+ ---'----'--'-- (7.105)
q Ek - Ek+q - Wq Ek + Ek+q + Wq
The wave vector integrals are rather hard to evaluate for most forms of M~, because the
energy denominators have the inconvenient expressions Ek+q = zJYk+q' Numerical results
have been obtained in one dimension (Brown et al., 1997). Frohlich-type polaron effects are
obtained when the bandwidth is large and the polaron effects are small.
7.2.2. Small Polarons

Next consider the other limiting case of the small-polaron Hamiltonian (7.95). The
polaron effects are assumed to be dominant and the bandwidth is small. The physical picture
is that the polaron effects localize the particle on a site and that hopping occurs infrequently
from site to site. The tight-binding term is the perturbation, while the particle-phonon term is
large. The Hamiltonian is solved in position space for the particles without resorting to
collective coordinates. The first step is to apply a canonical transformation which diagonalizes
the last two terms in the Hamiltonian. These last two terms are the same as found in the
exactly solvable models of Sec. 4.3 for the many-boson model. The canonical transformation
has the same form:
(7.106)
(7.107)
(7.108)
(7.109)
(7.110)
The polaron self-energy is d. The factors ~ were encountered previously in Sec. 4.3.4. They
arise from the canonical transformation of the particle operators eSCje-S = Ch. The number
operator nj = CJ Cj commutes with S and is unaffected by the transfonnation. But the tight-
binding tenn JCJ+8 Cj produces factor AJ~8Xj. The first tenn is not solvable exactly, so the
canonical transfonnation did not diagonalize this Hamiltonian. However, so far no approx-
imations have been made. The eigenstates and energy levels of if are the same as H. It is
fEuitful to investigate the solutions of if. Use the interaction representation to set
H=Ho+V:
Ho = L (i)qa~aq - A L nj (7.111)
q j
V= J Lj8 CJ+8CjAJ~8Xj (7.112)
The exponents in the factor AJ~8Xj can be combined since they commute. Also assume that
M~q =Mq-
AJ~8Xj = exp [ ~ eiq · Ri(l - eiq · 8) :~) (aq - a~q)] (7.113)
Combining the two factors simplifies the fonn of this operator for taking the expectation of
phonon operators. The perturbation V describes the hopping of the polaron from one site Rj
to the neighboring site Rj + O. The amplitude of this process is
(7.114)
where Ii) and If) describe the phonon occupation numbers in the initial and final states of the
transition. Since the factor AJ~8Xj pennits phonons to be made or destroyed, the states Ii) and
If) may have different numbers of phonons.
Holstein (1959) made a distinction between events in which the number ofphonons were
changed during the hop and events where they did not change. If the number of phonons is
changed in the hop, it is an inelastic scattering process. The particle loses its phase coherence
by this emission or absorption of phonons. Each hop becomes a statistically independent
event. The particle motion is diffusive, since it has Brownian motion by randomly hopping
from site to site. These events are called nondiagonal transitions.
The other situation is a diagonal transition in which all the phonon occupation numbers
remain the same during the hop. The number of phonons nq in each state q remains unaltered.
(Note: nq is the integer number of phonons, while N q is the thennal average of nq-) This
condition is rather strict. For example, if the number of phonons is unchanged but one state
gains a phonon while another loses one, it is a nondiagonal transition. The phase coherence of
a particle is maintained during a diagonal transition. If it is the most likely hopping event, the
particle will hop from site to site while retaining phase coherence. Then it is a Bloch particle
and fonns energy bands. The diagonal part of the hopping probability is defined as
e -ST -- ('Ix t j I
I }+8XI') (7.115)
If the nondiagonal transition probability is small, the particle will fonn energy bands with a
band energy
(7.116)
The effect of the polarons is to reduce the bandwidth by a factor exp(-ST)' The effective
mass is increased by the same factor. In the Holstein picture, the diagonal hops contribute to
458 Chap, 7 • Electron-Phonon Interaction
the real part of the self-energy and the nondiagonal transitions contribute to the imaginary
part of the self-energy,
The large-polaron behavior, of the Frohlich type, occurs when the diagonal transitions
dominate. The small-polaron behavior, with diffusive hopping, occurs whenever the
nondiagonal transitions dominate,
7.2.3. Diagonal Transitions

The diagonal and nondiagonal transition rates are obtained by using the same type of
operator techniques introduced in Sec. 4.3. The diagonal transition rate is calculated from
(6.2.5), The factor (fIXj~8Xjli) has two operators in the exponential which are Feynman
disentangled, with the destruction operators to the right. Using the techniques introduced into
Sec. 4.3 gives
(7.117)
e-ST = IIqze
('I 'Aa-A* at I')
z (7.118)
The result is given by events with the same number of phonons in each state q for both (il and
Ii). It is assumed that the number ofphonons in each state is determined by a thermal average.
The summation is taken over each possible number nq of phonons in each state, with the
thermal probability exp( -~ffiqnq) for this occurrence [see (4.239)]:
e- ST = IIq exp (IAi) [1 - e-~roq] f e-~roqn (nleA*a e-'Aal n )

t
2 n=O
= IIq exp[-IAqI2(Nq + !)] (7.119)
So the thermal factor Sr has the form
Sr = L:(M(q))2[1 - cos(q . 0)](2Nq + 1) (7.120)

q ffiq
(7.121)
The factor Sr is temperature dependent and increases with increasing temperature. For
temperatures larger than the Debye temperature, one may often use the expansion
!
N q ~ kBTI ffiq - + O(~ffiq) to write
Sr ~ 2kBT L:q (M(q))

ffiq
2[1 - cos(q . 0)] + 0(1 IT) (7.122)
so that Sr increases linearly with temperature at high temperature. The factor exp( -Sr)
determines the rate of diagonal transitions and decreases with increasing temperature, The
Frohlich-type polaron effects, with coherent motion, are most likely to be found at lower
temperatures. As the temperature is increased, the polaron bandwidth zJ exp( -Sr) becomes
smaller and smaller. It is reasonable to expect that the type of motion will change at some
temperature. In this model, the change is to a hopping type motion.
The factor 0 is the distance the polaron hops in a single motion. It is presumably a lattice
constant, or at least a distance associated with some fundamental unit oflength on the scale of
the crystal unit cell. The factor [1 - cos(q . 8)] suggests that phonons of short wavelength are
important, since this factor vanishes for q ---+ O. The integrations in (7.120) and (7.122)
probably have significant contributions from phonons near the edge of the Brillouin zone.
These integrals must be done carefully and probably numerically. Simple models, such as the
Debye model, are probably very inaccurate. Of course, the matrix elements M( q) must also be
found accurately for zone edge phonons. The Frohlich result M(q) 11q is a valid
r...,
approximation only at long wavelength for LO phonons. It is not adequate for the q inte-
grations in (7.120) or (7.122).
7.2.4. Nondiagonal Transitions

Nondiagonal transitions are hops, from site Rj to Rj + 8, in which the number of
phonons is not conserved. It is an inelastic scattering process. The matrix element
(fIXj~i5Ajli) describes the transition from one quantum state to another. By itself, it does not
convey meaningful information. Instead, the transition rate w must be evaluated. It is defined
as the rate per unit time at which the hopping occurs. It is calculated by first finding the
correlation function Wet), obtained by summing over all final states except the state If) = Ii),
Wet) =J2 L(iIAjt(t)Aj+i5(t)lf)(fIXj~i5Ajli) (7.123)
Ni
= J2[ (ilAjt (t)Aj+o(t)Xj~i5Ajli) - (ilAjt (t)Aj+i5(t)li) (iIXj~i5AjI)] (7.124)

and then evaluating the Fourier transform:
w = /1 2
1 Joo
-00 dtW(t) (7.125)
This formula is just the Fermi golden rule (Schiff, 1955). The transition rate is directly
proportional to the dc conductivity calculated from the Kubo formula. The latter relationship
is cr = ~e2 02 no w13. The derivation of cr is provided later, after the evaluation of w. The first
term in Wet) is a time-dependent correlation function, which is identical in mathematical form
to that found previously for the optical absorption in the many-boson model [see (4.308)].
The previous result may be utilized, by direct analogy, to give the following result:
exp[-<1>(t)] = (iIAjt(t)Aj+i5(t)Xj~oAjli)
(7.126)
<1>(t) = LM(q)211_eiq'/)12[(N
ffi2 q
+ 1)(1 _e-iffiqt)+Nq (1 _eiffiqt )]
q q
The second term in Wet) arises from the diagonal transition rate. Note that (ilAjt (t)Aj+i5(t)li) is
not actually dependent upon time. Repeat the above evaluation of ST with time-dependent
factors A. Since the result depends upon IAI2, the time cancels out. So the first impulse is to set
this term equal to the s~uare of the amplitude, or J2 exp( - 2ST ). This step is too hasty. The
thermal average of I(ilAj (t)Aj+o(t)li) 12 is not usually equal to the square of the average. The
diagonal transition matrix element, for a state with n phonons, is
(nIXj~oAjln) = e-IUqI2/2Ln(luqI2) (7.127)
U = M(q) (1 _ eiq ./» (7.128)

q ffi
q
where Ln(x) is the Laguerre polynominal. The quantity exp( -ST) is obtained by averaging
this amplitude over all possible values of n. A slightly different result is obtained if we
average the square of the amplitude over all possible values of n:
t
L
2 00
1(iIAf+oX/li) 12 = IIqe-luql [1 - e-~IDq] e-~IDqnLn(luqI2)2 (7.129)
n=O
= II/0[2I uqI2 JNq(Nq + 1)J exp(-2S T) (7.130)
which should be compared to the summation in (7.120). Note that IAql = Iuql. The average
over the square contains another factor, a modified Bessel function, which is obtained for each
state q. The final diagonal hopping probability is obtained by taking the product of this result
over all q states.
Now take the limit where the volume of the solid v goes to infinity. The particle-phonon
matrix element M( q) actually has a factor of 1/ vv.
Write this matrix element as
M(q) = M(q)/ VV, where M does not have any volume dependence and is only a function of
q. In the limit v ~ 00, the summations which determine ST and <D(t) become integrals which
are well behaved. However, the Bessel function 10 in (7.130) has an argument which vanishes
in this limit. When Z ~ 0, the Bessel function is
lim Io(z) = 1 + z24 + O(z4) (7.131)

z--*o
Z2
lim In[Io(z)] = -4 + O(z4) (7.132)
z--*o
lim Lln[Io(9(q))]
v--*oo q v
=~J d 3q39 (q)2 ~ 0
4v (21t)
(7.133)
In the limit of v ~ 00, the factor of Io(z) in (7.130) goes to one. The function Wet) is then
Wet) = J 2e- 2ST [e<P(t) - 1] (7.134)
3 - 2
2S = J~ M(q) 11-eiq .oI 2 (2N + 1) (7.135)
T (21t)3 co~ q
<p(t) = 2ST - <D(t) (7.136)
= J d 3q M(qi Il_eiq . o I2 [(N

(21t)3 co~ q
+ 1)e-iIDql +N eiIDql ]
q
(7.137)
The time integral must be done next to obtain the final transition rate in (7.125).
W = J2
h 2 e- 2ST
Joo dt(e<P(I) -
-00 1) (7.138)
which is the Golden Rule of quantum mechanics.

In evaluating this expression, some assumptions have to be made regarding the nature of
the spectrum of phonons. One case is an Einstein model where all phonons have the same
energy. This case is interesting because there are many combinations of phonons which give
the net same change in energy. This case was discussed by Nagaev (1963) but is not treated
here. Instead, the phonons will be assumed to possess a realistic spectrum in that most of the
phonon states q have different energies h coq- Of course, in a solid the states in the star of the
wave vector have the same energy, so there are always several q states with the same energy.
But the special features from Einstein phonons do not occur from a few states, but from large
sections of the Brillouin zone which are degenerate. The restriction to dispersive phonons is
asssumed because it is complicated to include events in which energy can be conserved while
phonons are being destroyed in one state, say q" while being created in another, q2' Energy is
conserved easily if co ql = co q, but not otherwise.
For a general spectrum of dispersive phonons, containing both acoustical and optical
types, the exponential function <J>(t) has a complicated time dependence, and the time inte-
grals should be done numerically. An approximate result may be obtained by a saddle-point
integration, which is done below. First, it is useful to discuss the convergence and properties
of the time integral.
At very large values of t, then <J>(t) approaches zero, which is fortunate since otherwise
the integral would diverge. However, in one dimension DeWit (1968) has shown that the
approach to unity is too slow. He showed that lim(--+oo <J>(t) ---+ 0(1/ -J(), so that the integral in
(7.138) does not converge. Since there are no one-dimensional solids, perhaps this anomaly is
no disaster. However, from the pedagogical viewpoint, it is unfortunate since one-dimensional
problems are nice for homework. In three dimensions the convergence is fast enough, so that
w has a definite value.
The physics is understood by investigating the frequency spectrum of just the first factor
in Wet). Call its Fourier transform U(co):
(7.139)
A typical spectrum is shown in Fig. 7.5. There is a delta function at co = 0 plus a smooth
spectrum. The delta function is the no-phonon transition. It is the probability that the tran-
sition takes place without any phonons being emitted or absorbed. As long as this event has a
nonzero probability, one will see a delta function in the spectrum. The intensity of the delta
function is just proportional to exp( -2ST ), which is called the Debye-Waller factor. In the
hopping transition, the equivalent of the no-phonon line is hopping by a diagonal transition.
The diagonal terms should be eliminated. The subtraction of the -1 factor from the time
integrand, in the definition of w, is just what is needed to eliminate the delta function. The
value of w is just the smooth part of the curve in Fig. 7.5 evaluated at co = O.
An approximate result for w may be obtained by a saddle-point integration (Schotte,
1966). This approximate result is very accurate in the limit where the particle-phonon
coupling is large, so the polaron effects are significant.
First we rewrite <J>(t) in (7.137) as
(7.140)
FIGURE 7.5 The function U(m).

The t integration is treated as a contour integral in a complex space. The path of integration is
deformed to go over the saddle point. The saddle point ta is located at the point ta = -i~/2.
The deformed contour is shown in Fig. 7.6, with the saddle point at €B. The function
(t) = ~(z) is expanded as a power series about the saddle point, where the distance from the
saddle point is Z = t + i~/2
~(z) = 2 L IUql2 JNq(Nq + 1)cos(coqz) (7.141)

q
~ ~(O) - ,,?? + O(z4) (7.142)
~(O) = 2 L lui JNq(Nq + 1) (7.143)

q
(7.144)
An approximate result is obtained by neglecting all terms except the first two. The time
integration, in the vicinity of the saddle point, is now just a Gaussian integral:
J dzed;!(z) ~ ed;!(O) J dze -yil = ed;!(O) ~
J2 ~ _
w= fl2 Vyexp[-2ST + <1>(0)] (7.145)
The saddle-point integral does the exp[-(t)] part of the integrand and not the -1 term.
However, the latter contribution just eliminates the delta function corresponding to the no-
phonon transitions. The saddle-point integration does not have this contribution. It provides
the smooth background part of U(co) at co = 0, which is precisely the contribution which is
needed for w.
The two exponential terms in (7.145) can be combined,
2ST - ~(O) = Lq IUqI2[JNq + 1- ~r (7.146)
which is a positive definite form. The exponent in (7.145) is always negative. At high
temperatures, the phonon occupation numbers may be expanded: N q ~ (kBT /coq) -
! + O(coq/kBT). Define x = ~COq, and the factors in the exponent give
!~[ J(Nq + 1 - JNJ 2
= 2Nq + 1 - 2JNq(Nq + 1)
2 (1
= ~ - 2 x2 -
1)1/2
4" =
X 3
4" + O(x ) (7.147)
- 1
2ST - <1>(0) ~ 4k T L IUiCOq (7.148)
B q
FIGURE 7.6 Path of integration in complex t plane.

The high-temperature expansion defines an activation energy A for the hopping rate
(7.149)
- -
The same factor ~ defines y = 4~kBT at high temperature.
Small-polaron theory predicts that the hopping rate is thermally activated. The rate of
hopping increases at higher temperatures. This behavior is in direct contrast to the rate
JZ exp( - 2ST ) of diagonal transitions, which decreases with increasing temperatures. As
temperature increases, the diagonal transitions become less likely, while the nondiagonal
transitions become more likely. The band-type polaron motion, or large polarons of the
Frohlich type, will exist only at small temperatures. Holstein estimated the transition
temperature between band motion and hopping motion to occur around 40% of the Debye
temperature. The estimate is remarkably insensitive to the magnitude of J. In the Holstein
model, the low-temperature motion should be band-like, while the high-temperature motion is
hopping. There have been many experimental systems with these characteristics which have
been ascribed to small-polaron theory. One example is TiO z (Bogomolov et a!., 1968). They
observe the transition from band to hopping conductivity at about 300°C. The conduction
bands of transition metal oxides are often d bands, and conduction in them seems well
described by small-polaron theory. See the reviews by Adler (1967), Appel (1967), and
Bottger and Bryksin (1976).
There are many other systems in which the electron conductivity is thermally activated.
They may often be otherwise explained, e.g., by an activation energy for freeing bound
electrons from defects or alternately by a static barrier to the electron motion. However, it is
thought that in most systems in which the electron motion is by hopping there is some
polarization of the electron by its immediate surroundings. The hopping electron must carry
its local polarization along, which gives the activation energy. The small-polaron picture
applies when this polarization is due to phonons or atomic realignment near the electron. Of
course, there could also be polarization of the electronic states, for example, the dielectric
screening by the material of the electron charge. Electronic polarization can be described by
an electronic polaron model, which has the same mathematical form as small-polaron theory.
Here the bosons are not phonons but density operators representing electron-hole pair
excitations of the system. This concept applies even in insulators, where the pair oscillations
have an energy gap.
The important message in small-polaron theory is that the motion is thermally activated.
If the particle polarizes its surroundings, then it can hop only by moving this polarization
along. The greater the polarization, the less likely hopping occurs. Each jump may occur only
when the polarization arrangements on initial and final sites are the same. Since the system is
in a state of continual fluctuation, this coincidence sometimes does happen. It is less likely
when the polarization is most severe. That is the reason for the thermal activation.
7.2.5. Kubo Formula

The Kubo formula for small polarons was evaluated by Lang and Firsov (1963, 1964).
The conductivity 0'(0)) is evaluated from the current-current correlation function. The limit of
0) ~ 0 provides the dc conductivity, while the optical absorption is given by the result for
nonzero frequencies. Both results are interesting for small polarons. The discussion will begin
with the form of the Kubo formula given in Problem 16 of Chapter 3:
(7.150)
(7.151)
The electrical current operator is the hopping form (1.204), which has been subsequently
altered by the canonical transformation (7.108). The spatial subscripts IX are replaced by the
diagonal sum in cubic crystals:
(7.152)
In the electron correlation function, the operators have a time development governed by
eiHt Ce- iilt , where iI also contains the hopping term V in (7.112). In the interaction repre-
sentation, the expansion of the S-matrix will result in terms of higher power in J. These terms
are seldom considered, and it is customary to evaluate only the term of order J2, which is
certainly the leading term. However, the neglect of higher terms in J is mostly expediency,
since they are hard to calculate. There is generally no proof available that these higher-order
terms are smaller.
In the interaction representation, the zeroth-order term in the S-matrix expansion just
replaces the electron part of the correlation function by its value obtained using the time
development governed by Ho rather than iI. This leading term in the conductivity will be
called cr(O) ((()). In this approximation, the electron correlation function is easy to evaluate
since there is no time dependence:
(7.153)
The correlation function is just equal to e(l - e), which is the probability e that the initial site
is occupied times the probability (1 - e) that the final site is empty. It is the correct average
for this correlation function whenever there is no correlation between the site occupations of
neighboring particles. Such correlations exist, for example, whenever the forces between
particles on different sites are included. Most model calculations omit such forces.
The electron correlation function is nonzero only when j = j' + 0' and 8 = -8'. The
calculation of the real part of the conductivity cr(O)((()) is now reduced to the evaluation of the
correlation function for phonon coordinates, which is the nondiagonal transition rate for each
site, summed over sites
(0) 1 - e -~ro J2 c:_2 2

JOO irot t t
Re[cr ((())] = 6 J:.2 e(l - e) ~ 0 dte (Xj (t)Xj+o(t)Xj+oXj) (7.154)
(() n fo-oo
The solid is assumed to have sufficient symmetry that the rate for each pair of sites is the
same. The summation over sites j just yields the number of sites, which is called N. The
summation over 0 yields the coordination number z. The number of particles is Ne = Ne,
where e is the concentration. For dimensional reasons, divide by the volume of the solid v, so
that the real part of the conductivity is
Re[O'(O)(ro)] =
1 - e-~Ol
6ro
(NZ) J fl-tl 2
-; ~e(1 - e)U(ro) (7.155)
U(ro) = f~oo dteiOli(X}(t)J0+o(t)JS~oX) (7.156)
The conductivity is proportional to the function U(ro), which was mentioned above. It has a
delta function at zero frequency, which should also be eliminated from Re[ 0'(0) (ro)]. The real
part of the conductivity O'(O)(ro) is just proportional to the Fourier transform of the hopping
correlation function Wet). The dc result ro -+ 0 was evaluated earlier in (7.95) by a saddle-
point integration, which for the dc conductivity gives
0'(0)
de
= e(l _ e)(ZNfl-)~w
6v
(7.157)
J2
w = 112
(1t)
Y 1/2 [
exp - kBT
Li] (7.158)
The same saddle-point integration also provides an estimate of the frequency-depen~ent

conductivity (Reik, 1972). Again the integration variable is changed to Z = t + i~/2, and <p(z)
is expanded about the point z = 0:
U(ro) = e- 2Sr f~oo dteiOlieq,(t) (7.159)
= e-2Sr+~0l/2 f~oo dzeiOlZ+(z) (7.160)
~ e-2srHOl/2+if,(0) f~oo dzeiOlz-yz1 (7.161)
~ G) 1/2 exp [ -2S T + ~ro/2 + <1>(0) - :~J (7.162)
Re[O'(O)(ro)] = 0'(0) sinh(~ro/2) exp [- ro2J (7.163)

de (~ro/2) 4y
At nonzero frequency the integral over z is still of the Gaussian form and yields a Gaussian
function of frequency. The result for Re[O'(O)(ro)] is important for relating theory and
experiment. It provides a prediction of proportionality between the dc conductivity and the
optical absorption. For el'-ample, it pr~dicts that the optical absorption is Gaussian, with a
width given by 4y = l6AkB T, where A is the activation energy observed in the dc conduc-
tivity. This prediction may be tested experimentally. It also predicts that the magnitude of the
absorption is proportional to the magnitude of the dc conductivity at each temperature. These
predictions are confirmed in Ti02 . Figure 7.7 shows the optical data of Kudinov et al. (1970).
The absorption data is in satisfactory agreement with the theoretical curve, which is calcu-
lated using parameters which also fit the dc conductivity. The physics of the absorption
process may be understood using a configurational coordinate picture suggested by Polder
(unpublished; see Reik, 1972). The coordinates are illustrated in Fig. 7.8. A particle sitting at
site j has a parabolic potential energy curve which represents the potential energy of the
0.3
02
j"
E /
o
j"
/
c: / o
~ er(O) /
_ 0.1 /
b /
/
WLO
O~l__~__~____~__~____~__~__~~__~______
2 4 6 B 10 12 14 16
v,I0 3 cm- 1
I I I I I I I I I I
0.5 1.0 1.5 20 E,eV
FIGURE 7.7 Optical absorption by small polarons in Ti02 for Ellc at T = 600 K given by points. Dashed curve is
theory with ~coo = 800 cm- l = 0.1 eV. Source: Kudinov et al. (1970) (used with pennission).
phonons for a displacement Q about the equilibrium point. The neighboring site j + 0 has an
identical parabola, displaced to the new equilibrium point, at a distance Qo. The dc
conductivity occurs by thermal activation over the intervening potential barrier ~, which is
midway between the two parabolas. If the parabolas have a curvature given by tX, so that the
potential energy curves are V(Q) = rxQ'2, then the activation energy is ~ = V(Qo/2) =
rx{fo/4. Similarly, the most probable optical transition is the vertical arrow in Fig. 7.8. It puts
the particle on the neighboring parabola, and later the particle relaxes down to the minimum
at the neighboring point. The length of this arrow is the energy Iiro1ll. = V(Qo) = rx{fo, which
is four times the activation energy. The proportionality lirom = 4~ agrees with the result
(7.163). The configurational coordinate model may be used, in fact, to derive (7.163) directly.
FIGURE 7.8 Configurational coordinate drawing of small-polaron motion. The probable optical absorption is the
transition COm' while the hopping barrier is ,1..
Sec. 7.3 • Heavily Doped Semiconductors 467
7 .3. HEAVILY DOPED SEMICONDUCTORS
An interesting application of polaron theory is in many-particle systems. If the solid has

many electrons, one has to consider electron-electron interactions along with the electron-
phonon interactions. They are very important in metals, which are discussed in the next
section. Other systems, which mix the two theories of polarons and electron-electron inter-
actions, are semiconductors which are heavily doped with impurities. Many III-V semi-
conductors are polar and may also be doped sufficiently with impurities to form a degenerate
electron gas. Then at low temperatures the electron gas has a well-defined Fermi energy, and
the electrical resistivity is finite.
These systems are interesting from several viewpoints. As an electron gas, they may
have an effective density which is very high. The actual density is quite low, perhaps only
no '" 10 18 _10 19 cm- 3 . However, the value of rs is calculated with the effective Bohr radius,
a* _ fi2EO
o - e2 m* (7.164)
41tn a*3)-1/3
r = ( 0 0 (7.165)
s 3
which has the effective mass m* and static dielectric function Eo for the semiconductor.
Because typical values are m* 1m '" 0.1-0.2 and Eo '" 10, the effective Bohr radius may be as
large as 5 nm. Even for an electron concentration of no '" 10 18 cm-3, a Bohr radius of
a1j '" 5 nm gives rs '" 1, which is a smaller value than found in metals. Even smaller values
of rs are obtainable in many semiconductors. Semiconductors are a suitable environment in
which to study the high-density electron gas.
Many semiconductors are polar. The interaction between electrons in the conduction
band and LO phonons is well described by the Frohlich Hamiltonian. Many of these semi-
conductors may also be doped to high electron concentration, so that experimentally one may
study polaron theory in a many-particle system. However, nature does not seem to like these
two conditions-strong polaron interactions and high doping levels-to occur in the same
materials. Materials which are very polar, such as alkali halides or II-VI semiconductors, can
seldom be doped to where the electron gas is degenerate. Group IV (Si, Ge) and III-V
semiconductors may be doped readily to high levels but are either weakly polar or not polar.
The theory of polarons, in many-particle semiconductors, needs only to treat systems with
weak coupling between electrons and phonons. An advantage of these experimental systems
is that all the parameters of the Hamiltonian are known from other measurements. The only
parameters of the theory are m*, no, Eo, and Eoo ' which all can be obtained by simple
measurements. Many-body theory may be tested in situations where there are no adjustable
parameters.
The self-energy of an electron from electron-phonon interactions is dramatically altered
by the presence of other electrons. Several effects are described. First, the phonon energies
themselves are altered by the screening properties of the electron gas. Second, the electron-
phonon interaction is screened by the electron gas. The screening is a significant feature, so
that even intermediate values of IX are reduced to weak coupling. Third, the nature of the
electron's self-energy is changed dramatically by the Pauli exclusion principle. When elec-
trons interact with phonons, they can only scatter into states not occupied by other electrons.
The sum of these three effects is that polaron theory is entirely changed from the simple one-
particle theory of the previous sections.
7.3.1. Screened Interaction

A little thought shows the system is complicated. The phonons and the electron gas are
mutually interacting and affecting each other's properties. Should the first step, in the
calculation, be to calculate the effect of the phonons on the electrons, or vice versa? The
answer is neither. Instead, one should calculate the total longitudinal dielectric function of the
system, which includes parts from both electrons and phonons. This function will be suffi-
cient to describe all the relevant physics.
The effective interaction between electrons may be written as
41te2
Veff(q, (0) = 2 ( ) (7.166)
q Etotal q, 00
where Etotal(q, (0) is the total dielectric function from all sources. The total dielectric function
will be derived in three ways. The first will be a phenomenological derivation. The other two
derivations use Green's functions.
The simple derivation is given first. The dielectric function is the summation of the three
contributions: Eoo ' the electron-electron interactions, and the electron-phonon interactions.
The first two are known, so that
!:total(q, (0) = Eoo - VqP(q, (0) + phonon term (7.167)
where P(q, (0) is the polarization of the electron gas, which was discussed in Sec. 5.5. The
phonon term will be calculated assuming that there are no electrons present.
For optical phonons in polar crystals, the ions will vibrate in response to an oscillating
electric field. If X is the distance between a pair of plus and minus ions with reduced mass M,
the classical equations of motion are
MX +KX = eEoe- imt = eE(t) (7.168)
The force constant is K = MOO}o. The oscillatory solution has a periodic displacement of
X(t) = eE(t) 1 (7.169)

M OO}o - 002
The net polarization of the system is P = enol(, where e and no are the charge and density of
the ion pairs. The contribution to the dielectric function is
X(t) 41te2no 1
L\E = 4mx = 41teno -E (t) = -M- OOro
2
- 00
2 (7.170)
The collection of constants 41te2nolM is an ion plasma frequency. It may be determined

phenomenologically by realizing that in the absence of the electron-electron term, the long-
wavelength dielectric function of a polar crystal may be written as
(7.171)
(7.172)
At zero frequency E(O) = EO' while at frequencies above the reststrahl 00» OOTO then
E(OO) = Eoo' These limits define the two quantities EO and Eoo' The total dielectric function is
EO - Eoo
Etotal(q,oo) = Eoo + 1 _ ro2 lOO}o - VqP(q, 00) (7.173)
In the derivation, the three contributions to the dielectric function are treated as independent
and additive. This assumption is not quite right. The rigorous derivation shows that the
phonons affect the electron-electron contribution. When calculating the polarization diagrams
for P(q, 00), there are diagrams in which the basic electron bubble has internal phonon lines.
However, these terms are never included in actual calculations. Instead, the term vqP(q, 00) is
approximated by either the RPA or the Thomas-Fermi model. In this approximation, the three
terms are independent.
The rigorous derivation of the total screening function proceeds by summing diagrams.
The Hamiltonian has the form
H =Ho +Hee +Hep +Hpp (7.174)
Ho = L~pC~erCper
per
+ LOOqAa~AaqA
qA
(7.175)
1
Hee = 2vLV~00)P(q)p(-q) (7.176)
q
1 t
Hep = r.; LMqAP(q)AqA (7.177)
"Iv qA
Hpp = ! LqA VA(q)At(q)AA(q) (7.178)
(7.179)
The electron density operator is p(q). The electron-phonon matrix element is MqA . The
phonon-phonon matrix element is VA(q). The electron-electron interaction has been written
as the product of two density operators. This form has the liability that it permits an electron
to interact with itself. These terms are meant to be absent and are ignored. The shorthand
notation is convenient for the following discussion.
The term Ho describes the noninteracting systems of electrons and phonons. The elec-
tron's energy ~p = Ep - Il is measured from its chemical potential. The kinetic energy is
usually assumed to have the form Ep = p2 12m*. Band structure effects are included by an
effective mass m*. In semiconductors m* is usually smaller than a bare electron mass. The
electron-phonon part of the Hamiltonian has the usual form. Our applications will mostly be
to polar coupling, where MqA has the form in (7.2). The electron-electron interaction has an
. . Vq(00) = Vq I E '
mteractlOn oo
The factor Eoo is included to account for the dielectric screening of the material. There
are three sources of dielectric screening in the system. The first is the electron-electron
interactions from the mobile electrons in the conduction band; their contribution to the
screening is included explicitly by the term Hee. The second is from the optical phonons, and
they are included through the term Her The third is from the high-energy electronic exci-
tations across the band gap of the semiconductor. They give rise to Eoo' They could also be
included as an additional term in the Hamiltonian, which would be solved to give Eoo'
However, for the low-frequency excitations of interest in the present discussion, these high-
energy excitations are uninteresting because they give a constant contribution to c(q, ro). It is
simpler to include these effects through the constant coo' which is introduced in the Hamil-
tonian at the beginning.
There are several different ways to solve these equations which give the same answer.
The easiest method is done first. In order to keep the discussion simple, the treatment is
specialized to longitudinal optical (LO) phonons. The ions are treated as point masses with an
effective charge eZ. The various interaction terms are approximated as
(7.180)
X(q)2=~ (7.181)
2MroTO
The phonon frequency ro qA is chosen to be the TO frequency, rather than the LO frequency.
The difference between these frequencies arises from the long-range Coulomb interactions in
Hpp' As long as the phonon-phonon interactions are explicitly included in the interaction term
of the Hamiltonian, then the TO frequency must be used in Ho. Later it is shown how the term
Hpp changes roTO to row.
All of the interactions are due to long-range Coulomb interactions: electron-electron,
electron-phonon, and phonon-phonon. All of the interaction terms can be combined into one
term
Hint = Hee + Hep + Hpp (7.182)
=! L v~OO)P~(q)PT(q)
q
(7.183)
PT(q) = ' " + ZqX(q)A(q) (7.184)
The total density operator PTeq) is the summation of the density operators for electron and
phonons.
The dielectric function is calculated for this interaction by following the same steps used
in Sec. 5.5.2 to obtain RPA. The difference between that case and the present one is the
phonons. By following the same steps, one can show that
c(q, iro) = coo [1- v~OO)p~)(q, iro)] (7.185)
p~)(q, iro) = - J: d"CeiOln~(T~PT(q, "C)p~(q, O)} (7.186)
The electron and phonon operators act independently. The most important terms in p(1) are
the summation of the separate susceptibilities for electrons and phonons:
p~)(q, iro) = p~l)(q, iro) + P~~(q, iro) (7.187)
P~~(q, iro) = Z2 q2X(q)2f!JJ(q, iro) (7.188)
(1). ro;i 2 41te2z2no

-VqPph (q, lro) = ro2 2' ropi = (7.189)
TO - ro M
The electron polarization pill is identical to the RPA result in Sec. 5.5.2. This dielectric
function is identical to the one found in (7.173). It is the RPA result for a system of electrons
and LO phonons. This derivation is rather simple because all long-range interactions were
included in Hint.
The third derivation of this same result is the one presented in most reviews. The first
step is derive an effective interaction W(q) between electrons. It has two terms. The first is the
direct Coulomb interaction v~oo). The second term is the interaction between two electrons
which results by the exchange of a phonon.
W(q) = v~oo) + Vph(q) (7.190)
Vph(q) = MiD(q) (7.191)
The phonon Green's function f0(q) should include the phonon self-energy due to phonon-
phonon interaction Hpp- This self-energy is VA = ffi;J2ffiTQ' The phonon Green's function is
D( ) _ -2ffiTQ -2ffiTQ
(7.192)
q, ffi - 2 2 2 V - 2 2
ffi - ffiTQ - ffiTQ A ffi - ffiLO
2 2
ffiLO = ffiTQ + ffipi2 (7.193)
The RPA expansion for the dielectric function is now evaluated with the effective interaction
W(q) which gives that
W(q)
Veff(q) = 1_ W(q)p(1)(q) (7.194)
The last equal sign states that this expression is identical to the previous one in terms of the
total dielectric function. Proving this identity is assigned as a homework problem.
The total dielectric function is often well approximated by the sum of the independent
contributions. For example, the piezoelectric contribution from acoustical phonons provides
another term in the above series (Mahan, 1972) for crystals which are piezoelectric. The
starting point of a many-body calculation for any system with several contributions should be
to find the dielectric function. Often a satisfactory approximation is obtained by just adding
the separate contributions. It will be shown later that this does provide a description of the
mixing and interference between these modes. The mixing occurs because all these various
modes-LO phonons, plasmons, piezoelectric phonons-have longitudinal electric fields.
The longitudinal electric fields of the various modes cause the mutual coupling. This coupling
is well described by the dielectric function.
A curious feature of the optical phonon part of the dielectric function (£0 - £(0)/
(1 - ffi 2 /ffi}O) is that the resonance frequency is ffiTQ' The resonance is the frequency where
the energy denominator vanishes. This position of resonance is in contrast to the Green's
function for optical phonons, which has its pole or resonance frequency at the LO phonon
frequency ffiw. There is no disagreement between these two results. The longitudinal exci-
tations of the system are not at the poles of £(ffi) but are at the poles of l/£(ffi). For example, if
the polar solid had no conduction electrons and was an insulator, the total dielectric function
would just be the first two terms of (7.173) or
(7.195)
(7.196)
(7.197)
(7.198)
The poles of I/E(co) are indeed at cow. Of course, in a polar solid there are both LO and TO
phonons. However, only the LO phonons make long-range electric fields, so only they
respond to a longitudinal electric field. The last relationship between the phonon frequencies
and dielectric constants is due to Lyddane et al. (1941).
The combined system of electrons and phonons has the dielectric function Etatal(q, co)
given in (7.173). The excitations of this system must be given by the poles in I/E tatal(q, co).
Let Zi(q) denote the complex values of co at which these poles are found. According to the
theorem of residues, the complex function may be expressed exactly as the summation over
these poles and their residues R i :
---= 1+ L R;(q) 2 (7.199)

Etatal(q, Z) i Z2 - Zi(q)
The poles always occur in the pairs ±Zi since Etatal is an even function of co. Each pole is then
interpreted as an excitation of the system, with a coupling strength given by the residue. The
expansion in poles and residues would be an exact way to describe the excitations, and their
couplings, if 1/ Etatal were only a sum of poles. However, previous experience indicates that
there are branch cuts in the complex function. In the electron gas, these described the pair
excitations, and the same excitations should occur here. The exact description of the exci-
tation spectrum is all contributions where there is a nonzero contribution from
S(q, co) = _1_ 1m [----,--1------=-] (7.200)

nOvq Etatal(q, co + i8)
which would include poles as well as branch cuts.

There are two important frequencies in the coupled system: the LO phonon frequency
COw and the plasma frequency of the electron gas:
(7.201)
The coupled modes have a quite different character depending on whether COw > cop or
cop > COTO' The case cop » COw is discussed here. In semiconductors, the experimentalist may
vary no, and hence cop, so the two situations cop < COw or cop > COw may be achieved in the
same material.
The total dielectric function is denoted as Etatal' Perhaps this should just be called E(q, co).
However, it is conventional to reserve the latter name for only the electron-electron parts of
the dielectric function
E(q, co) = 1 - v~oo)P(q, co) (7.202)
In the case where cop > cow, it is customary-although perhaps not necessary-to explicitly
separate the electron-electron part of the effective interaction:
V (00)
q Vq
Veff = -Etatal = -(--)
E q, co
+ Vsc-ph(q, co) (7.203)
The second term on the right is the screened electron-phonon interaction. It is defined by this
equation, so Vse - ph is the difference between Veff and the screened Coulomb interaction. This
definition may be manipulated by combining these two terms:
(00) V
V = __v...:q-;-,+~p,,-h_ _ (7.204)
se-ph 1 _ (v~oo) + Vph)P
(7.205)
(7.206)
From the definition of Vph = MiD(O), rewrite the effective interaction as
(7.207)
(7.208)
(7.209)
The screened electron-phonon interaction Vse - ph is expressed as the product of the ren~r
malized phonon Green's function D(q, ro) times the square of the screened matrix element M q .
The phonon Green's function contains self-energy terms arising from the polarization of the
electron gas. The quantity P(q, ro) describes the polarization of the electron gas, while iIi
is
the coupling between the electrons and phonons. It is important to note that this phonon
Green's function is different from the one which would be derived from Dyson's equation by
ignoring the electron-electron interactions. The latter result would be Dyson's equation with
no electron-electron interactions:
D(O)
D' = 1 -M2D(O)P (7.210)
q
The difference is the additional factor of 1/E in the self-energy term. The derivation of
D(q, ro) includes electron-electron interactions properly and show that the factor of I/E
actually belongs. Since E is usually greater than unity, including the 1/ E factor will consid-
erably weaken the self-energy effects.
Another interesting feature of the screened electron-phonon interaction is that the matrix
element Mq is divided by the dielectric function E. Since Mq is something in the nature of the
electron-ion potential, it is reasonable to expect it to be divided by the dielectric function. The
conduction electrons screen the electron-ion interaction. The ions are heavy and move with
small frequencies compared to electronic response times. The electrons are able to follow the
ion motion, which provides the screening.
There is a good reason for separating the effective interaction into the Coulomb and
screened phonon parts. When rop » row, the dielectric function E(q, ro) which occurs in the
phonon part needs to be evaluated at ro ~ row. For rop » row, E(q, row) is well approximated
by the static limit of c(q). Then it is a good approximation to rewrite the effective interaction
as
(00)
Veff
Vq
= -(--)
c q, co
+ MqD(q, co)
- 2
(7.211)
if = Mq (7.212)
q c(q)
(7.213)
In evaluating the screened Coulomb term, the full frequency dependence must be retained in
c(q, co). But the static limits c(q) and P(q, co = 0) are satisfactory in the phonon term. Most
calculations have used either Thomas-Fermi theory or the RPA for the dielectric function.
Although it would be more accurate to use a Singwi-Sjolander dielectric function, the
additional labor may not be worth the incremental increase in accuracy.
7.3.2. Experimental Verifications

First-order Raman scattering is permitted in crystals which lack an inversion center. Two
examples are zincblende and wurtzite, which are the crystal structure of most III-V and II-VI
semiconductors. The Raman experiment measures the frequency of excitations which have
the same magnitude of wave vector as the incident light. For optical frequencies, the wave
vector is typically 10 5 cm -I, which is essentially the limit of q -+ 0 for P(q , co). In this limit,
the electron-gas term -vqP(q, co) becomes -CO~/C02, so the total dielectric function is
(7.214)
The poles of I/Ctotal(CO) occur where Ctotal(co) = O. Setting the above equation to zero gives a
quadratic equation for co2 , which is easily solved to give
(7.215)
These two solutions are plotted as the solid lines in Fig. 7.9 as co~ is changed by increasing no.
The parameters are appropriate for GaAs, which has m* = 0.072, Coo = 11.3, COTO = 268
cm-I, and COw = 291 em-I. Also shown is the experimental data of Mooradian and Wright
(1966) for GaAs; they measured the frequencies by Raman scattering. There is excellent
agreement between theory and experiment.
The LO phonons and the plasmons are two modes which are mutually coupled. Figure
7.9 represents a typical crossing phenomenon of two coupled modes. It is a particularly clear
example, since both modes are completely classical. Even if h is restored to our equations, it
would not appear in (7.215). It is also useful to understand the asymptotic limits of very high
and low density. At high values of no, the analytical solution of (7.215) gives the roots when
2 2
cop» COw
(7.216)
(7.217)
'E
I
~
r- 500
I.L.
I
450
Cfl 400
>-
u 350
z 300
w
::> 250
0
w 200
a:
I.L.
Id 7 10'8 Ide
CARRIER CONCENTRATION
(e m3 )
FIGURE 7.9 Plasma frequency in GaAs as a function of electron density no, as determined by Raman scattering.
Solid line is theory using ffiro = 268 em-I. ffiw = 291 cm- i , Coo = 11.3, and m* = 0.07. Source: Mooradian and
Wright (1966) (used with permission).
The phonon-like mode has a frequency roro, not row. The choice roro occurs because the
frequency difference between roro and row is caused by long-ranged Coulomb interactions.
The electron gas screens these long-ranged Coulomb effects whenever ro; » roEa. In this
case, the electron gas can respond with a higher frequency than the phonons and can follow
the motion of the phonons. The electron gas will screen the phonons and prevent long-range
electric fields. The frequency difference between row and roro vanishes at long wavelengths,
and the longitudinal phonon excitation has frequency roro. Similarly, the phonons cannot
follow the plasma oscillations of the electron gas, so the phonons do not contribute to the
screening of the electron-electron interactions. The plasma frequency in (7.201) has a
dielectric constant Eoo ' not Eo. The terms which contribute to Eoo are high-frequency interband
transitions, which can certainly follow these low-frequency plasma oscillations.
The other limit of density has low values of no. When ro; « roEa, the two roots of
(7.215) are
(7.218)
(7.219)
The Lyddane-Sachs-Teller relationship has been used in the last identity. Now the phonon-
like mode has a frequency row. The electron gas cannot oscillate as fast as the phonons and
does not screen the long-ranged Coulomb fields of the phonons. The phonons have a long-
itudinal frequency given by row, which must be the correct limit as no -+ O. Since the
phonons can now follow the oscillations of the electron gas, they do contribute to the
screening of the plasma oscillations. Hence the plasma frequency now contains the screening
factor EO rather than Eoo'
7.3.3. Electron Self-Energies

The motions of the electrons are affected by the other electrons and by the phonons.
These interactions determine the self-energy of the electron. The self-energy will have
contributions from electron-electron as well as electron-phonon interactions. They combine
into the screened interaction, described in Sec. 7.3.1 and summarized in (7.166) and (7.173).
It is a rather cumbersome expression in its general form. Actual evaluation, to get analytical
results, will require making some approximations at some point in the calculation. The only
calculations which have been reported, and verified experimentally, are those for the limit that
O)p » O)w. In this case it is permissible to separate the self-energy effects into an electron part
and a screened phonon part. The electron-electron part may be evaluated by the procedures
for the homogeneous electron gas given in Chapter 5. Nothing further needs to be mentioned
about this term.
The present objective is to calculate the electron self-energy from electron-phonon
interactions. The first step is to calculate the one-phonon self-energy term. For the screened
electron-phonon interaction in (7.209), it is
'f.(p, ip) = - A1 L Jd q
3
--3
M2
: 2 ~(q, iO)n)<;9'(O)(p + q, ip + iO)n) (7.220)
I-' iron (21t) r.(q,/o))
There should also be a summation over the different phonon modes A, which is omitted, since
the discussion will be confined to polar coupling. Later it will be decided whether this one-
phonon term is sufficient. Perhaps more terms will be needed, with higher numbers of
phonons. It will tum out that the one-phonon term is adequate, since the coupling is weak.
The screening weakens the effective coupling constant. Materials which have intermediate
values of rx when they are insulators become weak coupling when the electron gas screens.
The screening will be shown later in great detail. For the moment, the one-phonon self-energy
is sufficient. Also defer, for the moment, the question of whether to use the Rayleigh-
Schrodinger or Brillouin-Wigner form of the self-energy. The energy is designated as ip,
which will be ~p for the Rayleigh-SchrOdinger form of perturbation theory and 0) for the
other.
The one-phonon self-energy (7.220) will be evaluated for the degenerate semiconductor.
A degenerate semiconductor has been doped with a sufficient number of carriers that they
become an electron gas, with a well-defined Fermi surface, in the limit of zero temperature.
The degenerate limit will simplifY the calculation in a number of ways, since the homo-
geneous electron-gas results are used to describe the dielectric function r.(q, iO)).
The self-energy in (7.220) has a free-particle electron Green's function <;9'(0) but a
renormalized phonon propagator ~. The procedure is to solve the phonon system assuming
that the electrons are a free-particle gas. The self-energy terms calculated for the phonons
have already been discussed. The phonons change their frequency at long wavelength,
because of the screening of the electron gas. These newly found modes could be used in
calculating the properties of the electrons; the phonon Green's function in (7.213) could be
used in (7.220). Although this procedure is reasonable and systematic, it makes the analytical
calculation too difficult. Instead, an unperturbed phonon Green's function ~(O) is used in
(7.220). An Einstein model is assumed for the optical phonon frequencies:
~(O)(q iO) ) = ~(0)(i0) ) = -20)0 (7.221)

, n n O)~ +
0)5
The basic problem with (7.220) is that there is too much physics in it. It describes how the
electron gas screens and modifies the phonons and their interactions with electrons. It does
this sufficiently well that it is too cumbersome to evaluate without making some approx-
imations. One may as well make these approximations at the beginning of the calculation,
since doing it early saves a lot of work later on. The first approximation has already been
described, which is to replace the phonon Green's function ~ by the unperturbed propagator
~(O).
The next approximation is much more drastic. The dielectric function c(q, iro) will be
approximated by its static value c(q). The frequency dependence of the dielectric function is
ignored. This approximation is reasonable as long as the phonon frequencies are much less
than the plasma frequency rop' The approximation c(q, ro) ~ c(q) is allowable whenever
row « rop . In fact, the dielectric function may now be combined with the matrix element to
give a screened interaction V,(q) = Mi
/c(q)2. The self-energy function in (7.220) now has
I
the form
= - A1 L + q, ip + iron)
3
'2:(p, ip) -d- 3
q Vs(q)E0(O)(iron)~(O)(p (7.222)
I-' i(O, (2n)
The summation over Matsubara frequencies now has the simple form which was evaluated in
Sec. 3.5:
'2:(p, ip) = I d 3q3 V,(q)[l ~No - nF(~p+q) + . NonF(~p+q) ] (7.223)

(2n) Ip - ~p+q - roo Ip - ~p+q + roo
A much more complicated result is obtained when the more accurate forms of c(q, iro) and
E0(q, iro) are retained in doing the summations. The simple result (7.223) is the basis for our
discussion.
The previous calculations of polaron theory always occurred in one-electron systems in
which the Fermi occupation factors nF could be set equal to zero. In the present discussion,
they must be retained, and they play an important role in the nature of the final result. Only the
case for the zero temperature is discussed, where the phonon thermal occupation factors
No = l/(e~(Oo - 1) can be set equal to zero. Next consider the wave vector integrals, where the
integration variables are changed: k = P + q
'2:(p, ip) = I d3k3 Vs(p - k)[.l - nF(~k) +. nF(~k) ] (7.224)

(2n) Ip - ~k - roo Ip - ~k + roo
A model form must be chosen for the dielectric function c(q). Of course, it would be super-
accurate to use the Singwi-Sjolander form, but it is sufficiently complicated that a computer
is needed to find any results. The Thomas-Fermi form is used instead. Although less accurate,
it permits an analytical answer. Later it is argued that the answer is insensitive to the
screening, so that the Thomas-Fermi approximation is probably adequate. However, this
assertion has never been thoroughly checked by doing the self-energy calculation with the
several alternate dielectric models of Chapter 5.
The retarded self-energy has both real and imaginary parts. The imaginary part will be
done first. It can be evaluated exactly now that enough approximations have been made. It is
given by (ip -+ ro + i8)
Im['2:(P, ro)] = -~I~dk{[l - nF(~k)]8(ro - ~k - roo)

(2n)
+ nF(~k)8( ro - ~k + roo)} I dn k ~(p - k) (7.225)
The delta functions take out the dk integrals. There remains just the angular integrals. Let
v = cos e denote the angle between p and k. For polar coupling, with Thomas-Fermi
II
screening, the angular integrals are
I dnkvsCp - k) = ~
8n21Xro3/2
-I
(p - ki
dV-[q-}F-+=--(P-_--'----k---=)2---=-f (7.226)
The integrals are straightforward and give a lengthy result:
(7.227)
(7.228)
The imaginary self-energy is shown in Fig. 7.10. It is plotted as a function of the energy ffi
relative to the Fermi energy for some value ofp. The striking feature in the result is that !meL)
is zero for energies within ffiO of the Fermi surface. Those electrons outside of this range can
lose energy by emitting phonons. But those electrons within ffiO of the Fermi energy cannot
emit phonons and have a zero value for Im(L).
The gap in the allowed values of Im(L) can be understood in a simple way. The exci-
tations with energy ffi > ffiO above the Fermi sea are electrons, or electron-like quasiparticies.
They can decay by emitting an LO phonon of energy ffio. If they initially have an energy co,
then after emitting the optical phonon, they have a final energy ffi - ffio. The electron states up
to the energy EF , or (ffi = 0), are all occupied, so that the final state must have an energy larger
than this value. Otherwise the transition cannot occur since the final states are all occupied.
This reasoning explains why ffi - ffiO > 0 for electron emission of a phonon, and why
Im(L) = 0 for 0 < ffi < ffiO' Phonon emission by electrons can occur only for initial energies
at least ffio above the Fermi energy. These processes are illustrated qualitatively in Fig.
7.1 1(a). The allowed transition has enough initial energy that the final state is above the
occupied levels. The forbidden transition would go to a state already occupied.
The behavior for ffi < 0 may be understood by realizing that the excitations are not
electrons. Instead, the excitations below the Fermi energy are states where an electron is
missing. These are called Holes in a metal. Unfortunately, the word hole in a semiconductor
usually means an excitation of the valence band. Our word Hole, with a capital "H," is an
excitation of the conduction band. It has an electron missing from the otherwise filled Fermi
sea. These Holes decay by having an electron jump into this empty state, which is illustrated
in Fig. 7.11 (b). The decay of a Hole is allowed whenever the Hole can be filled by an electron
already present. If the Hole has initial energy ffi < -ffio, then an electron in the Fermi sea can
-1m!
o CAl
FIGURE 7.10 The imaginary part of the electron self-energy from scattering by optical phonons. The energy (0 is
measured from the Fermi surface.
J ALLOWEO
(0) (b)
FIGURE 7.11 The explanation of why Im(L) = 0 for excitation energies within 000 of the Fermi surface. (a)
Electrons with 0 < 00 < 00 0 cannot decay since the final state is occupied (this transition is listed as forbidden). (b) For
-000 < 00 < 0, hole states cannot decay since no electrons are available.
emit an optical phonon and jump into this Hole state. This process is shown as the "allowed"
transition in Fig. 7.11 (b). The process is equivalent to the Hole jumping to lower energy by an
amount (00 since the zero energy for a Hole is the Fermi energy. The "forbidden" process of
Hole decay in Fig. 7.11 cannot occur because it requires the initial electron to be above the
Fermi sea. It is not possible at zero temperature if the system is in equilibrium. The forbidden
hole decay explains why Im(~) is zero for -(00 < (0 < O.
These arguments explain the energy gap in Im(~) for -(00 < (0 < (00' Of course, this
gap exists only for the self-energy contribution from optical phonons. The electron will have
nonzero values of Im(~) in this energy region arising from scattering by acoustical phonons
and from electron-electron interactions.
The self-energies describe the excitation spectrum of the interacting system. They are
electron-like quasiparticles for energies above the Fermi energy and Holes below the Fermi
energy.
The function g/Z) in (7.227) is a smooth function of both p and Z. This lack of structure
is due to the screening of the electron gas, which eliminates all sharp features. Since g/Z) is
sufficiently dull, it is a good approximation to treat it as a constant. A simple place to evaluate
this constant is at Z = 0 and p = O. This limit defines a coupling constant called 10:
3/2 ( E )3/2
{' _ l'1m l'1m -
JO -
0((00
- g (Z) -_ 0( (00 F
(7.229)
p.....,.oz.....,.04,J£; P (EF + I:s )2
--'---=--'--'--=-
Im[~((O)] = -10[8((0 - (00) - 8(-(0 - (00)] (7.230)
The real part of the self-energy may now be obtained by a simple argument. The retarded self-
energy function ~(p, (0) is causal and hence obeys a Kramers-Kronig relation (see Sec. 5.7).
The real part may be derived from the imaginary part by
Re[~((O)] = J_oooo dw' Im[~(w')] (7.231 )

TC w' - W
= _fcl (J-Olo ~ + Joo ~) (7.232)

TC -00 w' - (0 Olo (0' - w
= _fcllnl ww+- wol
TC (00
(7.233)
The result (7.233) was first obtained by Englesberg and Schrieffer (1963). They were actually
discussing metals with optical phonons, but the result also applies to doped semiconductors.
The total effect of the screened polar interaction is contained in the coupling constantfo. The
same type of answer is obtained for any electron gas with an Einstein phonon, but the detailed
definition of fo is altered. A consequence of electron screening is that the details of the
coupling constant are not terribly important.
There have been several attempts to verify this theory, in degenerate semiconductors,
through electron tunneling experiments. Usually the tunneling is done through the Schottky
barrier at a metal-semiconductor interface. The metal effects are negligible, and all structure
in the electron tunneling conductance can be attributed to the semiconductor unless the metal
is superconducting, and then the Bardeen-Cooper-Schrieffer (BCS) gap is apparent. But for
normal metals, even if ferromagnetic, the metal-semiconductor junction is determined by the
properties of the semiconductor (Conley and Mahan, 1967).
The most direct confirmation of the theory was obtained by Tsui (1974), who measured
the constant fo in InAs as the magnitude of the step increase in the scattering rate of the
electrons. He was measuring metal-semiconductor junctions in a magnetic field. The
differential conductance d 2I / d 2 V showed oscillations, similar to the deHaas-van Alphen
b
N
> "g
"H
N
"g
o 20 40 60
BIAS (mV)
FIGURE 7.12 d 2I/dV2 vs V data from an InAs---<>xide-Pbjunction. The InAs sample has no = 5.5 x 10 17 cm- 3 and
T = 4.2 K. (a) Taken with H = 22 kG, (b) taken with H = 2 kG, and (c) the difference between (a) and (b). Source:
Tsui (1974) (used with permission).
Sec. 7.4 • Metals 481
(dHvA) oscillations in a metal. Tsui's data is shown in Fig. 7.12. The oscillations change their
amplitude at voltages greater than 30 meV, which is the energy of the LO phonon in InAs. The
tunneling voltage is the difference of the two chemical potentials and is the energy in the
semiconductor which the electrons have after they tunnel from the metal. Using parameters
for InAs, Tsui used (7.229) to predict a value for 'to = 11/(210) and compared it with his
experimental number:
Theory: 'to = 5.3 X 10- 13 s (7.234)
Experiment: 'to = (5.1 ± 0.3) x 10- 13 s (7.235)
There is excellent agreement between theory and experiment.

In many respects, the degenerate semiconductor is the ideal environment in which to
study the electron gas and polaron effects in an electron gas. The electron gas can be made to
have an effective high density at which the theory is supposed to be valid. All parameters of
the theory, such as rx or m*, can be determined by independent experiments in the insulating
semiconductor. Theory can be done with no adjustable parameters. The experimentalists have
systems which are relatively easy to measure and which have an electron density which may
be varied continuously.
7.4. METALS
Electron-phonon interactions in metals have been studied for several reasons. The
phonons in metals are significantly affected by the electron gas, so any study of the phonon
system requires an understanding of the electron-phonon interaction. The electrons near the
Fermi surface have their motions influenced by the phonons, so that many experiments which
measure these electrons, such as transport or cyclotron resonance, are affected by electron-
phonon interactions. The interaction also causes superconductivity in many metals. There are
many experimental methods which have been applied to this topic, and ample reasons to do
so.
There is an initial problem facing any theoretical treatment. Since the two systems of
electrons and phonons affect each other very much, it is not obvious where to begin the many-
body calculations. Does one first solve the phonons assuming the electrons have no phonon
influences, or solve for the electron properties using the free propagators for phonons? That
is, does one first solve for the phonon or the electron self-energies? Neither approach may
work if the two systems are strongly coupled. Actually, there is a right way to begin, as
discussed by Migdal (1958). The justification of the procedure requires, to some extent,
knowing the solution to the problem. One should start by first discussing the phonons using
electron states which do not contain phonon effects. Although the phonons affect the elec-
trons, they alter electron states only within a Debye energy of the Fermi surface. These states
are only a small fraction of the electrons in the system. However, in solving for the electrons'
influence upon the phonons, the average is taken over all the occupied states of the electron
gas. This average is influenced to a negligible fashion by the few electrons at the Fermi
surface, which themselves are influenced by phonons. A good answer is obtained by first
solving for the phonon states using electron states which have no phonons included. Then one
can solve for the electron states using these renormalized phonon states. The standard
procedure is to first solve for the properties of the phonons. An electron Green's function is
employed which does not have a self-energy contribution from phonons. But it does contain
the effects of electron-electron interactions, which are very important for obtaining the
correct physical description.
7.4.1. Phonons in Metals

For simplicity, the discussion is limited to metals, such as sodium or aluminum, which
have only one atom per unit cell. The many-body system is a collection of ions and electrons.
It is charge neutral, so the ions and electrons have the same average charge density
lelno = eZn i , where no and ni are the particle densities of electrons and ions. The ions of
valence Z are considered as rigid objects. The excitations of the inner core electrons can be
neglected, since these energetic events are not induced by the motion of the ions. The core
excitations will contribute a small amount of dielectric screening, which can be included as a
phenomenological high-frequency dielectric constant ci. The Hamiltonian was discussed
(7.236)
The potential Ve;(ri - R,,) is between an electron at r i and an ion at R". It was called VA in
Sec. 1.3. Similarly, the potential Vi;(R" - R~) is between two ions. Both ofthese potentials
are unscreened, so they behave at large distances as
Ze 2
lim Vei(r) = - - (7.237)
r-+oo cir
(7.238)
In a free-electron metal such as sodium or aluminum, the ions are sufficiently far apart that the
Coulomb form (7.238) of Vii is probably valid for all ion pairs. However, the electron-ion
interaction Vei must always be treated realistically at small distances. The customary proce-
dure is to use a pseudopotential (Harrison, 1966). The electron-electron interactions are also
included in the Hamiltonian. They cause the potentials to be screened.
The first discussion is at an introductory level, approximately the same as Schrieffer
(1964). In this model the ions are point charges, and the electron gas is jellium. Later real
metal effects are discussed at a more advanced level. The introductory treatment will try to
explain the physics and avoid the bewildering notation of the correct treatment.
The first step is to solve the Hamiltonian of the electrons in the absence of phonons. The
ions are fixed in their equilibrium positions R~O). Then (7.236) becomes the Hamiltonian of
electrons in a periodic, neutral system:
2 2 1
H. =".!!.i... + =--2 "~ I + "L.. V.(r - R(O)) + 12 "L.. V"
.. (R(O) _ R(O)) (7.239)
Oe ~2
i mi i#j ri - rj I i"
el I "
,,~
" ~
The solution of this problem is quite formidable. The electrons states are Bloch waves in the
periodic potential. They must be calculated with the full effects of correlation and exchange in
the metallic environment, which has only been done approximately. The problem is quite
difficult, but it is not the problem at hand. For the discussion of electron-phonon effects, it is
assumed that the electron part of the Hamiltonian has been solved. Furthermore, in this
introductory treatment, the electron states of our quasi-jellium model of a solid are
approximated as plane waves.
The Hamiltonian of the phonons is first solved without reference to the electrons. In the
harmonic approximation (see Sec. 1.1) the ions are assumed to have small displacements QIX
about their equilibrium positions R~O), so the Hamiltonian can be written as
RIX = R~O) +Q" (7.240)
H = HOe + Hop +Hep (7.241)
Hop = I::; + ~ I:(QIX - Q~VQ" - Q~)v<l>~V<R~O) - R~O) (7.242)

IX IX ,,~
Hep = I: QIX . VVe;(rj - R~O) (7.243)

j"
(7.244)
The term Hop is called the bare-phonon Hamiltonian. These phonons are calculated by
ignoring the electron-electron and electron-phonon interactions. The ion motions are
calculated using only the direct ion-ion interaction potential Vi;. The bare-phonon Hamil-
tonian, once solved, describes a set of eigenfrequencies and normal modes which are quite
unrealistic. The poor results are obtained because the phonon Hamiltonian is for a system
which is not charge neutral. When the ions vibrate, the bare-phonon Hamiltonian Hop treats
the electron gas as a rigid background, which does not follow the ion motion. Hence the ion
vibrations cause long-range Coulomb fields. These vibrations are just plasma oscillations, of
the ions, so the long-wavelength phonons will have the frequency
(7.245)
In usual solids, and metals, the long-wavelength oscillations of the atoms or ions are
described by acoustical phonons, which have the dispersion relation ro(q) = Csq proportional
to the wave vector, where Cs is the speed of sound. In solids with more than one atom per unit
cell there may also be optical phonons, which have a constant frequency at long wavelength.
However, the above plasma modes of Hop are not from optical modes of vibration, since the
present theory is for solids with only one atom per cell and no optical modes. In fact, the bare-
phonon Hamiltonian Hop is sufficiently unrealistic that it does not have acoustical mode
solutions, which are necessary for real metals. The ion plasma modes are the correct long-
wavelength normal modes of Hop but do not realistically approximate the actual modes in a
metal.
In normal metals, the electrons and ions respond differently because of the great
difference in their masses. When the ions vibrate, the electrons follow the ion motion. There
are no long-range Coulomb fields and no ion plasma waves. When the electrons oscillate at
their natural frequency rop' the ions are too heavy to follow, so the electron motion is not
screened. The electrons can freely oscillate at their plasma frequency, while the ions cannot.
Our bare-phonon Hamiltonian gives the wrong answer because it did not let the electrons
follow the ion motion. Permitting the electrons to follow the ion motion is another set of
words to describe the screening by the electrons of the motion of the ions.
The method of solving (7.242) for the bare phonons was described in Sec. 1.1. The ion
displacements QIX and conjugate momenta PIX are expanded in a set of normal modes,
Q __l_"Q ik'R.
(7.246)
IX -,IN;~ ke
P = _l_"p ik·R. (7.247)

IX ,IN;~ ke
Hop = ~[2~Pk • P -k + !QkIlQ-kV<Pllv(k)] (7.248)
<pllv(k) = <D(O) - <D(k) (7.249)
<D(k) = L eik • R~O) cI>llv(R~O) (7.250)

IX
If the ions were point charges, these terms would be
(7.251)
(7.252)
(7.253)
where G are the reciprocal lattice vectors for the crystal. The summation over G converges
slowly, which may be helped by using Ewald techniques. The normal modes of the bare-A
phonon system are found as the eigenfrequencies nqA. and eigenstates ~qA. of the equation
(7.254)
These frequencies and eigenstates are used to define a set of creation and destruction
operators through the expansion in Eq. (1.85):
Q, = ~(2P:Q,,) ~,A" 1/2 (7.255)
AqA. = aqA. + a~qA. (7.256)
Hop = LnqA.(a~A.aqA. +!) (7.257)

qA.
where p is the density. The formal solution of the bare-phonon system is complete. The
modes which are obtained are conceptually useful, but there is no need to calculate them. The
set of eigenfrequencies and eigenstates are only used as the basis for solving the electron-ion
interaction Her When solving the interacting system, a new equation is obtained for the
phonon modes. These new equations contain electron screening and provide a realistic
description of the atomic motions. These are worth solving numerically. The eigenstates of the
bare-phonon system are useful only as a concept upon which to build our solutions in tenns
of Green's functions. The interaction tenn between the electron and phonons is written as
Hep = rvL..
_1 "M (q
I..
+ G)ei(q+G)' rA ql.. (7.258)
Y Vql..G
M,(q + G) ~ (2~J '1'4 . (q + G)V,,(q + G) (7.259)

The phonon states in this basis are the bare phonons. It is an unrealistic basis but serves only
as the starting point for the Green's function calculation. The electron states are put into the
second quantized notation, so that the effective Hamiltonian has the fonn
H = Lper ~pC~erCper + Lql.. nql..a~1..aql.. + 21V pkq

L L VqcZ+q,sC;_q,s' Cps' Ck,s
ss'
1 t
+ r;; L MI..(q)Cp+q,erCperAql.. (7.260)
yV ql..per
The solution to this Hamiltonian was discussed in Sec. 7.3. The present problem corresponds
to the case where the electron plasma frequency oop is very much larger than the phonon
frequencies. In this case the effective interaction between two electrons may be written as
(7.209) with a screened Coulomb interaction and a screened phonon interaction:
Veff(q, ioo) = (V q. ) + LMf(q)E!&I..(q, ioo) (7.261)

E q, zoo I..
v
E(q, ioo) = 1 - J.P(q, ioo) (7.262)
Ei
(7.263)
where MI.. is the screened matrix element M = M IE. The quantity E!&1.. is the phonon Green's
function. Its denominator contains the factor Mf E!&~O) PIE. The tenn 1IE comes from the
screening of the electron gas. It describes how the electrons can follow the motion of the ions
and hence screen out the long-range Coulomb forces. The screening eliminates the ion-
plasma solutions and makes the modes acoustical. To show this, first rewrite the phonon
Green's function as
Dret(q, (0) = 002 _ n~1.. _ 2nql..(Mf P IE) (7.264)
At long wavelength, the factors in the denominator of this Green's function have the following
limits:
lim (7.265)
q-+O
The last term on the right shows that the phonon energy denominator becomes the simple
expression co 2 - co~/c. For the small frequencies of phonons, the static dielectric function
c(q, 0) may be used. In the limit q ~ 0 it is given by the compressibility sum rule (5.266).
The zero of the phonon energy denominator defines the new phonon modes co~(q) = co~/c(q)
lim co,..(q) = qCs (7.266)
q--->O
C = coip fKf (7.267)

s qTFVK
The long-wavelength excitation of the coupled system of ions and electrons are now
acoustical phonons. The speed of sound is given by Cs' The formula for Cs is similar to the
result first obtained by Bohm and Staver (1951), who found that Cs = COip/qTF =
vpjZm/3M. These ordinary sound waves are obtained because the electron gas follows the
vibrations of the ions. When the ion of charge Ze moves, an amount of screening charge
nearly -Ze follows it, so the vibrating entity is nearly neutral. No long-range Coulomb fields
are present, and there are no waves at the plasma frequency of the ions. The screening charge
has contributions from all electrons in the Fermi sea.
The situations in metals is quite analogous to that of semiconductors. There the electron-
electron interactions caused the LO phonon modes to be given by
co2 _ co 2
co2 = co2 + LO TO (7.268)
TO c(q)
To apply this formula to metals, set COw = co ip and COTO = O. The TO frequency is zero in
metals since there is no restoring force for transverse oscillations. These steps give the same
formula co2 = co~/c(q).
The phonon Green's function can be written approximately as
2Qq"
D,,(q, co) = 2 2() (7.269)
co - co" q
This approximation is valid only when the damping of the phonon states can be ignored. The
phonon eigenfrequencies co" (q) are the ones found with the inclusion of electron-electron
interactions and represent the actual phonon modes measured in the solid. The electron-
phonon matrix element M~" in (7.258) contains the factor Q~r In the screened phonon
interaction, the unphysical frequency Q q / completely cancels from the product M~"D,,(q, co),
and the final formula contains no reference to Qq" whatsoever.
The preceding theory was much too simple to find realistic phonon modes in actual
metals. The arguments applied only to longitudinal modes of vibration and only at long
wavelength. A better description is needed to calculate transverse modes. Dyson's equation
for phonons is really a matrix equation, so that the phonon energy denominator is not just a
scalar quantity but is derived from the determinant of a matrix.
The reader is referred to Joshi and Rajagopal (1968) for a thorough treatment of phonons
in metals. They suggest one method of using the dielectric function for the homogeneous
electron gas to calculate realistic phonon modes for metals. If the Fourier transforms of filr)
and Vei(r) are vtlq) and Ve;(q), the screened interaction Vsi(q) potential between ions is given
by the effective potential:
(7.270)
The first term is from the direct interaction between ions. The second term is from the
interaction mediated by the electrons. The latter term arises when the ions polarize the
electrons, and this polarization acts upon other ions. The motivation for choosing this
interaction comes from the form of the denominator of the phonon's Green's function, whose
poles predict that the phonon frequencies are given by
2 2 M~'AP(q)
co = Oq'A + 20q'A c(q) (7.271)
The first term on the right, O~'A' is roughly given by <PJly(q)/M '" qJlq)iM)/p. The second
term on the right is
20 M~'AP(q) = q2 V2( ) P(q) (7.272)

q'A c(P) p el q c(q)
The two terms combined have the form
2 q2 [- - 2 P(q)J
co '" p Vii + VeJq) c(q) (7.273)
The term in brackets is just the effective interaction in (7.270). The Green's function deri-
vation has indicated this form to be the appropriate one for the potential between two ions.
The effective interaction will be used to solve for the phonon modes in metals. These modes
are found by solving a determinantal equation. For real metals it is not a good approximation
to express the screened ion potential as Vii(q)/c(q).
The phonon modes may be obtained in the same fashion as described in (7.249) and
(7.254). A dynamical matrix is constructed from the screened ion potential <PJly(q) which is
used in the determinantal equation (7.254). That procedure was followed by Woll and Kohn
(1962) to calculate phonon modes in aluminum. Other calculations are described by Ziman
(1960), Sham and Ziman (1963), and Harrison (1966). Other calculations are Na and K by
Shukla and Taylor (1976); K, Sn, and Sb by Kay and Reissland (1976); In by Garrett and
Swihart (1976); and Na, K, and Rb by Srivastava and Singh (1976).
A significant feature of phonon modes in metals is the Kahn anomaly (1959). In the
RPA, the electron polarization operator for co = 0 is (x = q /2kF )
p(1)(q) = _ mkF
2~
[1 +~(1-~)lnll
b
+xlJ I-x
(7.274)
In the equations for cot(q), from p(1) there appears the factor
(q2 _ 4~)ln(q - 2kF ) (7.275)
which is zero for q = 2kF but has an infinite slope. The phonon modes have a dispersion
dco'A(q)/dq which is logarithmically divergent at q = 2kF , which is the anomaly.
7.4.2. Electron Self-Energies

The electron-phonon interactions in metals have an important influence upon electron
states near the Fermi energy. The energy scale is set by the phonon energies themselves and
so is roughly a Debye energy. It is only 20-30meV in many metals, compared to Fermi
degeneracies EF of several electron volts. The actual self-energy L(k, u) of electrons from
electron-phonon interactions is a small energy, which is negligible compared to EF . However,
its derivative dr.(k, u)/du is large, so that it makes a large contribution to the electron
effective mass. Exactly the opposite behavior was found earlier in Chapter 5 for the electron
self-energies from electron--electron interactions: There the energies are large, but the
contributions to the effective mass are small.
For the electron-phonon interaction, it is now assumed that the phonon states have been
determined satisfactorily, perhaps by the methods already discussed. Having started with a set
of bare phonons with frequency Qq", and matrix element M q", the actual phonon states in the
metal have been determined by numerical means. There is available a new set of phonon
frequencies oo",(q), eigenvectors, and matrix element Mq", for electron-phonon interaction,
where A is the mode index: TA, LA, etc. The matrix element is obtained as follows: the
screened electron-phonon interaction (7.209) is rewritten in terms of the new matrix elements
and frequencies:
v (q) = LM",(qi 2Qq", (7.276)

sp '" E(q, 0)2002 - Q~", - 2Qq",P(q)Mi!E(q)
= '" M2(q) 200",(q) = '" M2 (q)D(O) (q, 00) (7.277)

T '" 002 - oo~(q) T '"
The frequency dependence in P(q, 00) and E(q, 00) is ignored since it has no influence at small
frequencies. The preceding equation serves to define the renormalized matrix element from
(7.258),
(7.278)
which is expressed in terms of the renormalized frequencies oo",(q). This renormalized version
of the screened electron-phonon interaction is exactly what we would obtain from a
Hamiltonian of the form
(7.279)
Equation (7.279) is just a simple electron-phonon Hamiltonian using the renormalized

phonon frequencies and matrix elements. Its significant feature is that electron--electron
interactions are omitted. When they are included, they serve to screen the electron-phonon
interactions. This phenomenon was used to derive the renormalized frequencies and matrix
elements. Adding electron--electron interactions again would be double counting. It would
give a second renormalization which is incorrect. Use the effective Hamiltonian (7.279), with
renormalized frequencies, and do not include screening again.
The self-energy of the electron, from electron-phonon interactions, is calculated using
the renormalized phonons and matrix elements. The only diagram which is ever calculated is
the one-phonon bubble diagram:
r.(k, ikn) = - ~ L M",(qi~(O)(k + q, ikn + ioom)D",(q, ioom)

pV q""ioom
= LJ d3q M (i.!.L 200",(q) I (7.280)

'" (2n)3 ",q ~iOOmoo~+00",(q)2ikn+ioom-~k+q
The one-phonon self-energy makes only a small contribution to the energy of the electron.
Higher-order terms, involving two or more phonons, are neglected on the basis that their
contribution is even smaller, as first shown by Migdal (1958). The one-phonon self-energy
diagram is evaluated by making several approximations. The free-particle propagator <§(O) is
used for the electron Green's function. The electron-phonon matrix elements M,,(q) are
treated as functions only of q, not of ro. The frequency dependence is ignored in the long-
itudinal dielectric function €(q, iro). With these approximations, the Matsubara summation is
the same one which has been evaluated previously in Sec. 3.5 for unperturbed electron and
phonon Green's functions:
(7.281)
The retarded function is obtained by ikn -+ u + if>. The discussion is confined to zero-
temperature properties, so set nB = O. Consider the expression
The self-energy is customarily evaluated using the Brillouin-Wigner form of perturbation

theory. The energy u is retained as a separate parameter, and the self-energy is calculated as a
function of both u and k.
The variation of I:(k, u) with respect to u is far more important than with respect to k.
The variation with respect to u is on the scale of a phonon energy, while the variation with
respect to k is on the scale of kF . An order of magnitude estimate of these derivatives is
(7.283)
where ron is the Debye energy of the solid. The derivative ofI:(k, u) with respect to u is larger
than the derivative with respect to ;k by the factor EF/ron ~ 10 2• It is usually possible to
neglect aI: / a;k> so that the effective mass and renormalization factors are given by the same
expression:
(7.284)
The electron self-energy I:(k, u) from electron-phonon interactions has appreciable value
only for u within a Debye energy of the Fermi surface, or -ron < u < ron. In this narrow
energy range, k hardly changes from kF . For metals with a spherical Fermi surface, the k
dependence of I:(k, u) is unimportant, and the k dependence is often even suppressed in the
notation: I:(u) == I:(kF' u). When the metal has its Fermi sphere cut by Bragg planes, so that
the Fermi surface is divided into pockets, the self-energy may depend on the position on the
Fermi surface in the region near these Bragg planes. In that case one should retain the k
dependence.
The two terms in the integrand of (7.282) which contain the factor nF are far more
important than the factor which contains the 1. It is natural to separate ~ into two terms:
~ = ~(a) + ~(b) (7.285)
~(a)(k, u) = L Jd 3q3 nF(~k+q)MA.<qi

I.. (2n)
x [u _~k+q +1 ())I..(q) + io - u_~k+q _1 ())I..(q) + iO] (7.286)
~(b)(k, u) = L
I..
J :~q 2 1
(2n) MI..(q) u - ~k+q - ())I..(q) + 10
(7.287)
The term ~(b) is a rather dull function of its two arguments. Not only is this self-energy term
small in magnitude but so are its derivatives with respect to k and u. It makes a negligible
contribution to the energy, effective mass, or other properties of the electron. An estimate of
this term can be obtained in the following way. Since the chemical potential does not enter the
evaluation of this self-energy, it behaves as the self-energies calculated for one particle in an
empty band. It contributes to the self-energy for states near to the bottom of the occupied
band, but is quite small for electrons near to the Fermi energy. Only the latter electrons matter,
so ~(b) can be neglected.
The other self-energy term, ~(a) in (7.286), is also rather small in value. Its importance
comes from its contribution to the electron effective mass. The self-energy has a strong
dependence upon energy for u '" 0, which is near the Fermi energy. Consider the real part of
a~(a) /au
The integral is evaluated by integrating by parts on the angular variable v = cos e = P. q,

where the two factors are
Judv = uv - Jvdu (7.289)
m
V = - kq (~k+q
-I
± ())I..) , dv = dV(~k+q ± ())1..)-2
u = nF(~k+q), du = ~ dV[:~ nF(~k+q)J
The integration by parts gives, with d 3 q = 2nq2 dqdv
(7.290)
The first term in (7.290) is smaller than the second by a factor ffiD/EF and may be neglected.
The second term is the important one. The factor dnF('f.J/d~ = -()(~) at zero temperature.
Changing the integration variable to p = k + q gives, for this last term alone,
Re(aL(a))
au u=o
= -2l:J d 3p ()(~ )M,,(k- P)2
,,(21t)3 P ffi,,(k - p)
(7.291)
___ 2_l:J d2PM,,(k- pi (7.292)

- (21t)3" vF ffi,,(k - p)
where d 3p = 21tp 2 dpdv and Jdv8(~) = 1/(pvF), which leaves d 2p = 21tpdp. The remaining
integration is over the two-dimensional area of the Fermi surface. For spherical Fermi
surfaces, this factor is d 2p/VF = mkFdOp ' where Op is solid angle, but most Fermi surfaces
are not spherical. The term [a Re L(a) /du] is quite large because of the factor ffi;:\ since
phonon energies are small. This important expression was first derived by Nakajima and
Watabe (1963). It is called A,
(7.293)
which is somewhat unfortunate, since A is also used to designate the summation over the
polarization modes ofthe phonon system. No confusion should arise over these quite different
parameters, however.
The parameter A(.h as defined in (7.293), is a function of the position on the Fermi
surface of the metal. The point k is the reference point, and the average is taken over other
points p on the Fermi surface. Quite often the average value of A is given for a metal, which is
obtained by averaging A(k) over the Fermi surface:
A= Jd2 kA(k) (7.294)

Jd 2 k
The other contribution to the effective mass of a particle at the Fermi surface is from the
derivative ofL(k, u) with respect to ~k' However, aL/a~k is much smaller than A by a factor
ffiD/EF, so that the most significant term is A. The other terms are usually entirely neglected,
and the effective mass of electrons at the Fermi energy, from electron-phonon contributions,
is given by
(:*) = I +A (7.295)
The first calculations of this quantity were by Swihart et al. (1965), and by Ashcroft and
Wilkins (1965), and many have been done since. The quantity A can be determined from
electron tunneling experiments, so that experimental values are available. Both theory and
experiment have been reviewed by Grimvall (1981). His suggested values are shown in Table
7.2 for some nontransition metals. Some of the values are quite large-they are 1.6 and 1.5
for mercury and lead. These large values show that the electron-phonon coupling is large for
these metals. This observation is significant for the theory of superconductivity, which in
elemental metals is due to electron-phonon coupling as described in Chapter 10. Metals
which have a low value of A, such as the alkali metals, are not superconducting.
Table 7.2 Electron phonon mass enhancement A.
Li Be
0.41 ± 0.15 0.24 ± 0.05
Na Mg Al
0.16 ± 0.04 0.35 ± 0.04 0.43 ± 0.05
K Zn Ga
0.13 ±0.03 0.37 ± 0.05 0.97 ±0.05
Rb Cd In Sn
0.16 ± 0.04 0.40 ± 0.05 0.8 0.72
Cs Hg TI Pb
0.15 ± 0.04 1.6 0.8 1.5
From Grimvall (198\).
The phonon density of states is defined to be
L J~O[ffi -
d3
F(ffi) = ffi,Jq)] (7.296)
A (21t)
where the summation Ie is over different modes, and the integration is over the Brillouin zone.
McMillan (1968) has introduced another function, which is called rx 2F( ffi). It is treated as a
single function, although the notation implies incorrectly that it is the product of two func-
tions. It is defined as
(7.297)
leek) = 2
A JroD -rxiF(ffi)
dffi
(7.298)
o ffi
rxiF( ffi) is the frequency spectrum one obtains by starting at a point on the Fermi surface k
and integrating over all other points on the Fermi surface p. It will vary from point to point on
the Fermi surface of a metal. The same quantity without the subscript k is the average over all
different points on the Fermi surface:
rx 2F(ffi) = d2k /Jd2k

J-rxiF(ffi) - (7.299)
VF VF
Nowadays it is calculated with a computer, using realistic electron wavefunctions, realistic

phonon eigenstates and frequencies, and realistic matrix elements. The quantity rx 2F( ffi) can
be obtained experimentally, as will be explained in Chapter 10, from electron tunneling
experiments. As an example, rx 2F( ffi) for Pb and Sn are given in Fig. 7.13. These results are
from the experimental data of Rowell et al. (1969a, 1971). The dashed line in Fig. 7. 13 (a) is
F( ffi) for Pb, which is similar to rx 2 F( ffi) because the matrix elements vary only by about a
factor of 2 over the various phonon states. Other physical quantities may be calculated from
rx 2F( ffi). For example, the lifetime of electrons on the Fermi surface is
't(~) = 41t fD dffirxiF(ffi)(e~rol+ 1 + e~rol_l) (7.300)
Note that Ij't(k) is zero at zero temperature but is nonzero for a nonzero temperature. This
lifetime is between scatterings of the electron by phonon emission or absorption. It is not the
same lifetime which enters a calculation of the resistivity, for example. These differences are
Co)'"
! 10
,
lAo
'i
.0; 0.8
,
,," ,
0
z \
or I
i_ l
> OS
I
!
-.!
lAo 0.4
I
I
I
I
I
'/"~
0.2
0
0 4 5 6 7 8 9
- 31469 IblSn
• 92867
0.4 +
~ 0.3
1&1
~
i5
! 0.2
~
.!
lAo
.!
"a
0.1
O~~---~~~~---~~---~---L-~~
o 4 6 8 10 12 14 16 18 20
ENERGY.meV
FIGURE 7.13 The energy dependence of the electron-phonon coupling a2 F(ro) in (a) lead and (b) tin as determined
by electron tunneling in superconductors. In (a) the dashed line is the phonon density of state F(ro) alone. In (b) the
points indicate the spread of values for different tunnel junctions. Source: Rowell et al. (1969a, 1971) (used with
permission).
treated in Chapter 8. The electron lifetime in (7.300) is derived from the imaginary part of the
retarded self-energy in (7.281):
1)= -
-( 2 Im[~(k, u = 0)] = 21t L J-d 3q MA(q)
3 - 2
tk A ~~
xHI + nB«(i)A) - nF(~k+q)]O(~k+q + (i)A(q»

+ [nB«(i)A) + nF(~k+q)]O(~k+q)]O(~k+q - (i)A(q))} (7.301)
Change variables to p = k + q, and the integral is
(7.302)
The phonon energy can be neglected in the delta functions o(~p ± (i)A(k - p» since their
energy is small compared the the Fermi energy. Then the two delta functions are alike, and
each just scatter the electron to another point in the Fermi surface
(7.303)
which is the same as (7.300). Results for aluminum and lead have been reported by
Tomlinson and Carbotte (1976). The same calculation for nonzero u gives
1
-(-)
t k, U
= -2 Im[~(k, u)] (7.304)
(7.305)
Equations (7.300) and (7.305) are the formulas for calculating Im(~).
Next a formula is needed for Re[~(k, u)]. From (7.286), integrate by parts on the
v = cos e variable in the same fashion as was done in deriving (7.290):
(7.306)
Problems 495
>.,
.s
.,
In
Pb
'2'
., -Im~(u)
•
15
- -------,
c
~ ,------ ----
I
\ ~ I
\ VJ 10 I
"
I
I
I
I
I
I
L'__-L,__~I __~~~r_~~--~---L--_4_u(meV)
-20 -15 -10 -5 5 10 15 20
~(u)
-10
FIGURE 7.14 The real and imaginary self-energies at zero temperature of an electron in lead, from the electron-
phonon interaction. Source: Grimvall (1981) (used with permission).
The term dnF(~)/d~ = -3(~) is the important one, and the other may be neglected in
comparison. A careful analysis shows it to be smaller by the factor roD/EF . The result for zero
temperature is:
lu + ro,..{k - p)1
Re[~(k, u)] = - LA. J(2n)
d3p
- 3 3(~p)MA'<k
-
- p) In
2
(k)
u-roA. -p
= - fo drocxi(ro)In lro-+-ul
OOD
ro-u
(7.307)
This expression vanishes at u = 0 and is positive for u < 0 and negative for u > o. The usual
procedure is to find cx2F(ro) from superconducting tunneling experiments. It is a property of
both the normal metal and the superconductor. It is used in (7.305) and (7.307) to find the
electron self-energy from electron-phonon interactions. Corresponding experimental infor-
mation may be obtained from electron tunneling in normal metals, as shown by Rowell et al.
(1969b). Their results for Pb are shown in Fig. 7.14. The self-energy has no k dependence
since they use an averaged value of cx2F(ro). Re[~(u)] is quite small--of the order of milli-
electron volts. However, it has a very steep slope at point u = 0, which gives d~/du its large
value.
PROBLEMS
1. The polaron self-energy -0(0)0 contains some kinetic energy of the electron and some potential
energy. Use the Rayleigh-Schrodinger wave function to show these have the ratio of 1 : 3 at p = o.
2. Show that the total momentum of the polaron in (7.18) commutes with the Hamiltonian in (7.2).
3. Calculate m/m* for the Frohlich polaron in the Tamm-Dancoff approximation.

4. Use the two-phonon wave function, in Rayleigh-Schrodinger perturbation theory, to find the 1X2
term for the number of phonons in the polaron cloud at p = O. Be sure your wave function is normalized
to unity. You should get -0.10361X2, so correlations reduce the number ofphonons in the cloud.
5. Calculatel:Rs(P) at nonzero temperature for all values of Ep. Also calculate its imaginary part,
which is the imaginary part of the retarded self-energy with E = Ep ' which is also half the mean-free-
path for one-phonon emissions or absorptions.
6. Use the answer of Problem 5 to plot, on a piece of graph paper, l:RS(Ep) for IX = I and 0 < Ep < ffio.
Evaluate the self-energy in the Tamm-Dancoff approximation for the same parameters, and plot it on the
same set of graph paper.
7. Consider the one-phonon self-energy in Rayleigh-Schrodinger perturbation theory at nonzero

temperatures:
l:RS(P) =J d3q3MJ(
(2n)
Nq + 1
Ep - Ep+q - ffiq
+
Ep -
Nq
Ep+q + ffiq
) (7.308)
At high temperatures, make the approximation N q ~ kaT /ffi q , and keep only the term proportional to
kaT. Evaluate it by assuming a Debye spectrum and a piezoelectric matrix element (1.256). You should
find that the self-energy vanishes when vp is less than the speed of sound.
8. Minimize the ground state energy, in strong coupling theory, with the variational wave function
<j>(r) = A exp( -~r). Show that the energy is -0.09801X2ffio, which is not as low as found with a Gaussian
wave function.
9. In strong coupling theory, calculate the energy of the lowest p state by using the variational wave
function <j>(r) = Ar cos(9) exp( _~2y2), where ~ is a variational parameter.
10. In the limit of small wave vectors, the interaction between an electron and the homogeneous
electron gas has the form
(7.309)
The electron-plasmon interaction is identical in form to the Frohlich polaron Hamiltonian. What is the
equivalent IX for the electron gas? Find the numerical values for metallic sodium (rs = 3.96) and
aluminum (rs = 2.07).
11. Consider a small-polaron Hamiltonian which has the phonons localized on each site. This model
could apply to molecular crystals, where the molecular vibrations are highly localized:
H = ~[J ~ C~&Cj + ffioa; aj + AffioC; qa j + al)] (7.310)
Show that this may be derived from (7.95) with the choice Mq = constant. Find the canonical
transformation which diagonalizes the last two terms in the Hamiltonian. Use the transformed
Hamiltonian to calculate:
(a) The thermally averaged amplitude and probability that a hop occurs without any change in the
number of phonons at each site.
(b) The thermally averaged amplitude and probability that the hop occurs without any change in
the number of phonons but that they may be exchanged between the two sites.
Problems 497
12. Show that the effective interaction (7.173) with the dielectric function (7.174) may be written as
Veff = V{ Eoo ~ vqP - Q{~~q~ 002 (Eoo ~ VqP - ~ Vqp) ]

EO
where Q{q) is the renormalized phonon modes of (7.220). At 00 = 0, then Veff is Vq/{EO - vqP), and at
00 ~ 00 it is Vq/{B oo - vqP) (Appel, 1966).
13. Prove that the following two expressions are identical for polar scattering by optical phonons
W
Veff = -vq = --- (7.311)
Etotal 1 - WP
where Btotal is given in (7.173) and W is given in (7.194).
14. The Raman scattering of Fig. 7.11 occurs because the light couples nonlinearly to the phonons but
not to the p1asmons. Assume a term in the Hamiltonian of the form
I L [btbk,Ak'-k + he] (7.312)

kk'
which describes the scattering of light from k' ~ k with photon operators b~bk' and phonon operators
Aq = a +a_t .
q q
(a) Derive a correlation function which describes the Raman scattering to order [2.
(b) Solve it at long wavelengths to get the Raman intensity of the modes oo± in (7.2l4) as a
function of no.
15. Compare Rayleigh-SchrOdinger and Brillouin-Wigner perturbation theory by plotting both

theories using the self-energy in (7.233):
RS: E(P) = ~p + Re[l:{~p)] (7.313)
BW: E(P) = ~p + Re{l:[E(p)]} (7.314)
Use the parameters/o = 10meV and 000 = 30meY. Which theory appears more reasonable?
16. Use the RPA for E2{q, 00) to estimate the damping of phonons from electron-hole creation in a
metal.
17. Prove that the sum rule f dOOOOIX2F{ 00) is independent of phonon properties (McMillan, 1968).
Evaluate the sum rule analytically for metallic hydrogen, and estimate the formula numerically.
18. Derive Eq. (7.305) for t{k, U)-l. Show that it goes to a constant at u ~ 00.
19. Use (7.305) and (7.307) to show at zero temperature that Re[l:{u)] and Im[l:{u)] obey a Kramers-
Kronig relation. Calculate by direct integration of (5.297) and (5.298).
20. Show that in the jellium model of a metal, with the ions as point charges, the longitudinal modes at
long wavelength are described by a total dielectric function [see (7.174)]:
€total = El (1 - OO~)
002 - VqP{q, 00) (7.315)
With this choice, one can write the effective interaction in (7.173) as the correct e1ectron-electron and
electron-phonon parts.
21. If :r.(T, k, E) is the retarded self-energy at temperature T, show that the electron self-energy in a
metal from electron-phonon interactions obeys
Re[:r.(T, k, E)] = [00 dE' [ - dn;~E')JRe[:r.(O, k, E - E')] (7.316)
22. Use (7.300) to show that at high temperatures lit = 2rcAkB T.

Chapter 8
dc Conductivities
8.1. ELECTRON SCATTERING BY IMPURITIES
This chapter is mostly concerned with the methods for calculating the electrical conductivity.
Four different methods are discussed: (1) solving the Boltzmann equation, (2) evaluating the
Kubo formula for the current-current correlation function, (3) evaluating the force-force
correlation function, and (4) solving the quantum Boltzmann equation. For scattering from
fixed impurities they all give the same answer. For scattering by phonons two different
answers are obtained. One is called the Ziman (1960) formula, and the other the Holstein
(1964) formula. Two criteria are important in comparing these methods: which is the easiest
to use, and which gives the most accurate answer?
The electrical resistivity, or conductivity, from impurity scattering is an important topic.
From the experimental viewpoint, all solids have impurities which make a contribution to the
total resistivity. In many metals and semiconductors, the low-temperature resistivity is
dominated by impurities, since all other contributions are temperature dependent and vanish
at low temperature. In metals, the resistivity from impurity scattering is largely temperature
independent, except for temperature variations on the scale of the Fermi temperature
TF = EF/kB • The subject is important, on the theoretical side, because it was one of the
earliest evaluations of the Kubo formula. The importance of vertex corrections became
apparent. Indeed, the derivation showed that vertex corrections are usually very important and
should be assumed important until shown otherwise. This conclusion, and message, continues
to be relevant even for calculations of other quantities.
The present chapter is really about vertex corrections.
8.1.1. Boltzmann Equation

The electrical resistivity from impurity scattering is easily derived by using the Boltz-
mann equation. This derivation is presented for several reasons. First, it is the easiest way to
get the answer. Second, the resistivity was first found this way; the Green's function evaluation
of the Kubo formula only confirmed the result known earlier from transport theory. The
Green's function derivation is complicated and subtle, and it is useful to know and believe the
right answer in order to recognize it when it is finally derived.
499
500 Chap_ 8 • de Conductivities
Our objective is to derive a formula for the electrical resistivity with the least possible
fuss. The simplest possible model is adopted for the solid. It is a homogeneous system except
for randomly located impurities. The electron states are plane waves except for occasional
scattering from isolated impurities. The impurities are very dilute, so that interference
between successive scatterings can be neglected. In the Boltzmann theory, the electrons are
described by a classical distribution function f(r, k, t). The time rate of change of this
distribution function is governed by the Boltmann equation
dt
#
if
at
ak
O=-=-+v-Vr!+--Vkf+
at
-
dt
(#) collisions
(8.1)
The last term is the time rate of change due to collisions with the impurities. There is no r
dependence inf(r, k, t) since the material is assumed to be homogeneous. Also, for the dc
conductivity, there is no time dependence. The system has a weak external electric field and
the current flows in a steady-state fashion. The distribution function is only a function of wave
vector f(k) and obeys the equation
ak
O=-'Vkf+ (d-f ) (8.2)
at dt collisions
In a solid, the factor ak/ at is equivalent to an acceleration which is equal to the forces on the
electron (see Kittel, 1963):
ak =
-
e
-eE - -v x Ho (8.3)
at c
In the present problem, there is an electric field E and no magnetic field Ho = 0, so that
eE'Vkf(k)= ( -df ) (8.4)

dt collisions
The collision term is the most interesting. It is evaluated in the relaxation time approximation.
This approximation assumes that collisions seek to return the system to the equilibrium
configuration fo(k), which is the configuration the system would have in the absence of an
electric field. The rate of change off(k) due to collisions is assumed to be proportional to the
degree thatf(k) is different fromfo(k):
[f(k) - fo(k)]
( ~) collisions 'tt(k)
(8.5)
2
fo(k) = e~~k + 1 (8.6)
The factor of two info is due to the spin degeneracy. The above equation defines the transport
relaxation time 'tlk). A more detailed derivation can be found in Ziman (1960). Here an
equation for 'tlk) is derived and solved. Then the distribution function is
f(k) = fo(k) - e'tt(k)E . V k f(k) (8.7)
When the electric field is small, only a small amount of current flows. The system is only
slightly out of equilibrium. The distribution functionf(k) = fo(k) + fi (k), where the change
fi (k) is small. It is only necessary to retain terms of O(E). Iterate the above equation, which
Sec. 8.1 • Electron Scattering by Impurities 501
effectively replaces f by Jo on the right-hand side. Then f(k) is evaluated to O(E).

Furthennore, it is assumed that the system is isotropic andfo(k) = Jo(k)
f(k) ~ Jo(k) - etlk)E' Vkfo(k) (8.8)
~ !coCk) - et (k) E· kdfo(k) (8.9)

t m dEk
The electrical current density J is the product of the electron charge e, the electron's density
no, and the average velocity (v), which is obtained by averaging over the electron distribution:
J = eno (v) = e J--3f(k)-

d3k
(2n)
lik
m
(8.10)
no = J--3f(k)
d3k
(2n)
(8.11)
The distribution functionf(k) is nonnalized to give the electron density no. By using the result
(8.9) forf(k), the tennfo gives an average (v) of zero, since no current flows when there is no
electric field; as many electrons are going one way as another. The current is proportional to
the second tenn, and it is proportional to the electric field:
(8.12)
In a homogeneous, isotropic system, the current J flows in the direction of E. The quantity
Jo(k) is independent of k direction. The only angular factors are vk(Vk' E). The angular
integrals will average this to viE/3 in three dimensions. The conductivity cr is the ratio of J to
E:
(8.13)
It is a positive quantity since dJo/ dEk is always negative. Equation (8.13) is the basis of all the
calculations. There remains the important task of deriving a fonnula for the relaxation time
tt(k). It is not just the time between scattering events, which is derived from the imaginary
part of the retarded self-energy. This distinction is important, since it makes life difficult. The
relaxation time in the Boltzmann equation is a special quantity.
The impurities are assumed to be static, fixed, objects with a spherically symmetric
potential. They have no internal excitations, so the electron scatters from them elastically. The
impurity causes the electron in state k to scatter to k', which has the same energy, so that
Ikl = Ik'i and Ek = Ek" The net rate of scattering out of the state k is the rate of going from k
to k', which is proportional to f(k)[1 - f(k')] minus the rate from k' to k, which is
proportional to f(k')[l - f(k)]:
- (df )
dt collisions tt(k) (2n)
J
= [f(k) - Jo(k)] = 2nnj d 3k'3 O(Ek - Ek')
X {ITk1df(k)[1 - f(k')] - ITk'kI 2f(k')[1 - f(k)]} (8.14)

502 Chap. 8 • de Conductivities
where ni is the concentration of impurities. The quantity Tkk" is the T-matrix element for
scattering from k to k' which was defined in Sec. 4.1. It is symmetric in its indices
Tkk' = Tk'k, which simplifies the above equation
The integrand contains the factorf(k) - f(k'). The integral is evaluated by assuming the form
in (8.9), which is written as
f(k) =fo(k) + k 0 EC(k) (8.16)

f(k') = fo(k) + k' 0 EC(k) (8.17)
f(k) - f(k') = (k - k') EC(k)
0 (8.18)
Since Ikl = Ik'i then and fo(k') = fo(k). The quantities f(k) and f(k') differ only in the
angular part of the second term. The angular part is treated as follows: define a coordinate
z
system in which the direction is le, so that
leoE = cos 9 (8.19)

leo'" = cos 9' (8.20)
le' E = cos 9 cos 9' + sin 9 sin 9' cos <p
0 (8.21)
where the law of cosines is used in the last identity. The difference of the two distribution
functions is now
f(k) - f(k') = kEC(k)[cos 9(1 - cos 9') - sin 9 sin 9' cos <p] (8.22)
The last term on the right, which contains the factor cos <p, will vanish when doing the integral
Jd <p. There is no other <p dependence in the integrand of (8.15), and the average of cos <p is
zero. The remaining term may be written as
J J
drok' [f(k) - f(k')] = kEC(k) cos 9 dn",(1 - cos 9') (8.23)
= [f(k) - fo(k)] Jdn",(1 - cos 9') (8.24)
The termf(k) - fo(k) is factored from both sides of(8.15), which leaves the definition for the
reciprocal of the relaxation time:
(8.25)
The important factor in the integrand is (1 - cos a') = 1 - k . k' I k 2 • It makes the relaxation
time in the Boltzmann equation different from the usual relaxation time t(k), which is the
time between scattering events. The latter quantity is simply
t
1
(k) = 21tnj -
J
d-k' 3
3 O(Ek - Ek')ITkk,1 2
(21t)
(8.26)
41tnj~( . 2~
= - k L- 21 + 1) sm uI(k) = vknjcr(k) (8.27)
m I
cr(k) = 1~ ~(21 + 1) sin2 ol(k) (8.28)
The usual inverse relaxation time is just the scattering cross section cr(k) times the particle
velocity Vk times the impurity density nj. An equivalent result may be obtained for the
relaxation time t,(k) of the Boltzmann equation. The T matrix is expanded as in (4.47) and
(4.52) for the case where Ikl = Ik'l,
Tkk, =- 21tkL (21+ 1)PiCcosa)ej1W)sin(01(k)) (8.29)

m I
j
(lk ) = 41tkn L 1sin [01(k) - 01-1 (k)]
2 (8.30)
tt m 1=1
where a is the angle between k and lc. The angular integrals in (8.25) are straightforward but
cumbersome, so only the result for tt is just listed. The formula for 1/tt has a different
combination of phase shifts than the formula for the cross section.
The factor (1 - cos a') weights the amount of scattering of the electron by the impurity.
Small-angle scattering, where cos a' ~ 1, is relatively unimportant in contributing to 1/tt.
These events do little to impede the flow of electrons and so contribute little to the resistivity.
The factor (1 - cos a') obviously favors large-angle scattering events, which are more
important for the electrical resistivity. The relaxation time in the transport equation is not
identical to the average scattering rate because there is an additional factor to weight the
amount of scattering.
An example of evaluating the conductivity in (8.13) is given for a free-electron metal. At
low temperature the distribution function (8.6) becomes a delta function in energy which sets
k=kF
(8.31)
The angular integrals have already been done, so that
_ 2e2 41t 2 k (k) e2not,(kF)

cr - 3 (21t)3 vFm Ftt F = m (8.32)
where no = kJ../31t 2 . The relevant relaxation time is for electrons at the Fermi surface. Yet the
conductivity is proportional to the density no of all conduction electrons and not just to those
at the Fermi surface. This surprising result is quite reasonable once the physics is understood.
When the electric field is first imposed, the equation k = -eE shows that all electrons in the
Fermi sea start accelerating equally. The Fermi sea is translationally shifted in wave vector
space. The scattering tends to relax the Fermi distribution back to its undisturbed config-
uration. As shown in Fig. 8.1, electrons in the leading edge of the displaced Fermi distribution
FIGURE 8.1 The circle represents the Fermi sea, which begins to move in response to an applied electric field.
Steady state is maintained by relaxation back to other points on the Fermi surface.
are scattered back to the rear regions. Only those electrons at the Fermi surface can scatter.
The electrons well below the Fermi surface cannot elastically scatter, since all states with the
same energy are already occupied with other electrons. Above the Fermi surface there are no
thermally excited electrons. Only electrons at the Fermi surface are available to elastically
scatter to other points on the Fermi surface. The lifetime only involves 'tt(kF ).
The conductivity is relatively insensitive to temperature as long as the density of states of
the metal is a smooth function of energy near the Fermi surface. The resistivity p = 1/ cr is the
inverse of the conductivity, and is proportional to the concentration of impurities. This
proportionality is experimentally verified.
Since impurity scattering is elastic, it does not change the energy of the electron. As the
current of electrons moves through the solid, each electron gains energy from the electric
field. How does the electron lose this energy, if it only scatters elastically? The next few
paragraphs will answer this question.
Let v = cos e, where e is the angle between k and E. In order to keep the discussion
simple, it is assumed that the distributionJ(k) is isotropic in the absence of a field. When the
field is present, the distribution function can be expanded in a Legendre series in v. The first
few terms are
J(k) = fs + v/P(k) + P2(V) JaCk) + ... (8.33)
where fs(k) is the isotropic part of the distribution while /P(k) is the 1 = 1 part of the
distribution. Note thatfs(k) is not the equilibrium part of the distribution, which isfo(k). The
electrical current is determined by the distribution /p.
The conduction process can be viewed as having the following steps:
1. For t < 0, E = 0 and the initial distribution isfo.

2. At t = 0 the field E is switched on. It accelerates the particles and creates the
distribution /p.
3. The elastic scattering takes the particles from the p distribution/p to the s distribution
fs. This step has a time constant 'tt.
4. The isotropic distributionfs relaxes back to the equilibrium distributionfo. This step
has a different time constant.
The energy relaxation occurs in the last step where fs is brought to equilibrium. The
electrons can lose energy to their heat bath, usually phonons. This process has a very different
time constant than 'tt and is usually much slower. The relaxation time 't( from elastic scattering
determines the rate at which particles scatter out of the p distribution /p into other distributions
such asfs andJd. The current is determined by 'tt since it gives the steady state amplitude of/p.
The energy relaxation occurs elsewhere in the chain of events. The rate for energy relaxation
te is calculated below.
8.1.2. Kubo Formula: Approximate Solution

The electrical conductivity can be calculated from the Kubo formula by using the
technique described in (3.390)-(3.393). The correlation function is evaluated for nonzero
temperatures and frequencies:
1t(iron ) = -~J~
3v 0
dteiOln"(T"j(t)·j(O») (8.34)
The retarded function 1tret ( ro) is obtained by letting iron -+ ro + iD, and the dc conductivity is
given by the limit of ro -+ 0:
cr = - lim {Im[1tret (ro)]} (8.35)

Ol--->O ro
This correlation function for cr is evaluated for the same model system described in Sec.
8.1.1. There is a free-particle system with a dilute concentration of simple scattering centers.
Equation (8.34) is evaluated for the following Hamiltonian and current operator:
H = L ~pCJerCper +V (8.36)
per
1
V =-L V(q)p;(q)p(q) (8.37)
v q
Pi(q) = Leiq ' Rj (8.38)

i
(8.39)
J"_e"
- - L.. PCtper Cper (8.40)
m per
The impurities are at positions Ri , and an average will be taken over the possible distributions
of impurity positions. This averaging technique was described earlier in Sec. 4.1.4.
The theoretical calculation is divided into two parts. The first part, in this section, is
simply to reproduce the Boltzmann result which was derived above. How is a conductivity
derived which is proportional to the relaxation time tt? The derivation entails a summation
over a set of vertex diagrams. The treatment is kept as introductory as possible, since it is one
of the first summations over vertex diagrams. A formally exact solution to the correlation
function due to Langer (1960) is presented in Sec. 8.1.3.
The logical way to evaluate the correlation function (8.34) is as a power series in the
concentration of impurities. Averages over the impurity density operators in (4.91) are
expressed as a power series in the number of impurities N i :
fn(qj' ... , qn) = (Pi(qj)P;(q2)' .. Pi(qn»)
= ND"<"
I ... qj
+NI2 D"<"...qj D"<""'jqj + ... (8.41)
At first sight it appears possible to evaluate (8.34) by just expanding the S matrix and
collecting all terms proportional to N i, then Nl, etc. A simple expansion in powers of Ni does
506 Chap. 8 • dc Conductivities
not work, as is apparent from the desired answer. It has ( j <X 'tt <X lln j • The first term has the
conductivity inversely proportional to n j , which is impossible to obtain as a simple series in N j
except by summing a set of diagrams. The expected answer is only going to be obtained by
summing a series of diagrams.
The correlation function (8.34) is written in the interaction representation as
(8.42)
The logical way to evaluate this expression is to expand the S matrix and consider each term.
The first term has S = 1, and this correlation function is called 1to:
(8.43)
= 2e: J~ d'tej())n~ LP2J~ dtd())n~~(O)(p, 't)~(O)(P, -'t)
3m v 0 p 0
This expression is zero unless iron = 0, since the number operators C~crCpcr are 't-indepen-
dent. The 't-integral gives zero unless iron = 0 and then it gives ~. The term 1to gives a
conductivity of zero. The result is not surprising, since it is the correlation function of
noninteracting particles, and they have zero resistivity. Perhaps a better answer is the corre-
lation function for the conductivity is infinity. The zero is sufficient to alert us that a
nonsensical question was asked, and a nonsensical answer was obtained. A resistive system
requires putting damping into the particle motion.
The next logical step is to replace all ~(O) by ~. The self-energy of the particles, from
impurity scattering, is included in all the particle Green's functions. Of course, ~ is obtained
by summing a series of diagrams, which is Dyson's equation. The self-energy ~(p, iPn) from
impurity scattering was evaluated in Sec. 4.1 in the limit of low n j • It is a retarded function
with real and imaginary parts, where the imaginary parts are due to the damping of the
particle motion. The step of replacing ~(O) by ~ does put in damping of the particle motion.
The first correlation function which will be evaluated is shown in Fig. 8.2. It is a simple
bubble diagram, with the smooth lines denoting ~ and the two vertices having the vector
vertex p. This correlation function is called 1t(O)(iro):
(8.44)
The wiggly lines at the two ends of the bubble, which are connected to the vertices, represent
the incoming frequency iron. The first step is to evaluate the summation over Matsubara
frequencies.
FIGURE 8.2
The procedure for doing this summation was described in Sec. 3.5. The easiest way is to
represent the interacting Green's functions by the Lehmann representation
1
S(iron) = A L ~(p, ip + iro)~(p, ip)
I-' lp
(8.45)
(8.46)
The next step is to convert to a retarded function (iron --+ ro + io) and then to take the
imaginary part:
(8.47)
The next step in the derivation (8.35) is to divide by ro and then to take the limit ro --+ O. The
important frequency dependence is in the last factor:
(8.48)
and the conductivity from this contribution is called 0'(0)
(8.49)
In the last step, the limit v --+ 00 changed the summation over p into a continuous integral.
The right-hand side is positive since dnF/dE is negative. Before discussing this result, it is
useful to review the order of the steps in the derivation. They will be used in all evaluations of
the Kubo formula:
l. Do all summations over Matsubara frequencies ipn'

2. Analytically continue iro --+ ro + io to get the retarded function 1tret (ro).
3. Take the imaginary part Im[1tret (ro)].
4. Divide by co, and then take the limit ro --+ O.
These steps cannot be taken out of order. Equation (8.49) for 0'(0) has several interesting
features. The factor dnF/dE = -O(E) at zero temperature, which is rather convenient, since it
serves to eliminate the integral over dE. The sharp step in nF(E) is in contrast to the
momentum distribution (3.135),
(8.50)
which no longer has a discontinuous step at p = kF since the impurity scattering causes a
general smearing of this distribution. However, the energy distribution nF(€) is always a sharp
function of € at zero temperature regardless of the interactions.
Another crucial feature of (8.49) is that the spectral function is squared, A(p, €)2. That
this is important may be shown by examining the limit as the impurity concentration ni
vanishes. Since the self-energy is proportional to ni' it will vanish in this limit. Define
~ = - !meL), and then
lim A(p, €) = lim 2~ 2 2 = 2rc8(€ - ~p) (8.51)

nj-+O ,1-+0 [€ - ~p - Re(L)] +~
The spectral function becomes a delta function when ni ~ O. This limiting behavior is
reasonable, since in the absence of self-energy effects, the particles are all free, and the
spectral function is indeed a delta function. The question at hand is what happens to A2 as
ni ~ 0, since it appears to go as the square of a delta function. In fact, A2 does diverge as
ni ~ 0, which makes the conductivity diverge to infinity when ni ~ O. A method is needed
for handling this divergence. The answer is provided by considering the integrals
J oo
-00
dm
-
2rc m2
2~
+ ~2
-1
-
(8.52)
JOO dm ( 2~ )2 1
(8.53)
-00 2rc m2 + ~2 ~
where m = € - ~p - Re(L). The first integral has the right behavior as ~ ~ 0, since it gives
the same result as does A = 2~/(m2 + ~2) = 2rc8(m). The second integral suggests the
replacement
·
11mA 2 = 1·1m ( 2~ )2 = 1.
1m---
2rc8(m) (8.54)
,1-+0 ,1 ..... 0 m2 + ~2 ,1-+0 ~
which will give the right behavior as ~ ~ O. The replacementA 2 ~ 2rc8(m)/~ will be made
in the limit as ni ~ O. Furthermore, the quantity 2~ is recognized as the inverse scattering
time of electrons on the Fermi surface,
1
2~(kF' € = 0) = 't(kF) = -2 Im(L) (8.55)
The conductivity formula may now be written as
cy(O) = 2i2
3m 2
J(2rc)3
d p 8(~ )p 't(p)
3
'P
2 (8.56)
This equation looks like the right answer for cy since it seems to have exactly the same
combination offactors as (8.13). But there is a very important difference between (8.56) and
the Boltzmann result (8. 13)--in the relaxation time. The formula (8.56) has a relaxation time
without the (1 - cos 8') factor, since the relaxation time in (8.56) is from Im(L), which is just
the average time between scattering events. The (1 - cos 8') factor was important for
weighting the large-angle scattering processes, which were important for the resistivity. The
preliminary answer (8.56) is not the Boltzmann result and has serious deficiencies.
The above derivation has one achievement. It succeeds in deriving a term in cy which is
inversely proportional to n i . The relaxation time 't(p), although the wrong one for the
conductivity, at least has the virtue that it is inversely proportional to nj, which makes cr(O)
also inversely proportional to nj •
The calculation has not yet yielded the right result. More diagrams need to be evaluated.
Among those remaining, there must be a subset which, when evaluated, will give the right
answer. The S matrix is expanded for impurity scattering, and the remaining terms contain
higher powers in the impurity interaction V. At first it appears that higher powers in the
impurity interaction must imply that the additional terms are higher powers in the impurity
concentration nj. This conclusion is false, as is obvious from the answer. The final cr is
proportional to ljnj, while our preliminary term cr(O) is also proportional to ljnj. The
important correction terms in the S-matrix expansion must also yield terms in the conductivity
which are proportional to ljnj. The S-matrix expansion is examined to find the terms which
cause cr to diverge as ljnj when nj -+ 0, although these terms must come from higher terms
in the S-matrix expansion. Higher-order terms in S can be proportional to Ijnj if they also
contain higher powers of the spectral function An.
The correlation function cr(O) , contained Green's functions '§ which include all self-
energy effects. The remaining diagrams are called vertex corrections. They are defined as
diagrams in which the impurity scattering links the Green's functions on both sides of the
bubble. Some vertex diagrams are shown in Fig. 8.3(a). There is a single impurity with a
varying number of scattering events from the electron line on either side of the bubble. If
there were no scattering line connecting one side of the bubble, the diagram would be a self-
energy term on the other side. A diagram in which the two electron lines, on both sides of the
bubble, scatter from the same impurity cannot be a self-energy diagram of either one and so is
called a vertex diagram. Figure 8.3(a) only shows vertex diagrams with a single impurity
participating in the scattering. Vertex diagrams can occur with scattering from several
impurities. These are equally important and are considered later.
(b)
(e) -o-~<D+®-+®-+ ...

(d)~
FIGURE 8.3
The sum of all the diagrams in Fig. 8.3(a) can be evaluated in a simple way and gives the
correlation function:
W~~(ip, ip + iro) = niTpp,(ip)Tp'p(ip + iro) (8.57)

2e'2 1
n(1)(iro) = - 32 2 I>A L p' p'~(p, ip)~(p, ip + iro)~(p', ip)
mV pp'l-'l}1
x ~(p', ip + iro)~~(ip, ip + iro) (8.58)
Equation (8.58) is easy to prove once the rules are recalled for scattering from a single
impurity: (1) total momentum is conserved, and (2) energy is not changed from ip or ip + iro.
The second rule says that all electron lines in one string, say the top, are at the same energy ip,
while the ones on the bottom are at ip + iro. The momentum conservation requires that the
momentum transfer on the top electron line be p - p', which is exactly the opposite of the
other: p' - p. The evaluation is easy, since the two sides decouple, as is illustrated sche-
matically in Fig. 8.3(b). The effective interaction W(1) is represented by a diamond in the
figures, which is illustrated to the right of Fig. 8.3(a). The diamond is the total vertex scat-
tering from a single impurity.
Figure 8.3(c) shows the sum of correlation functions which have increasing numbers of
diamonds in them. These contributions are called ladder diagrams. Summing the ladder
diagrams achieves the objective of a relaxation time with the factor (1 - cos 9'). It is
important to realize that the sum of diagrams in Fig. 8.3(c) is not the only contribution with
scattering from several impurity sites. An example of a nonladder diagram is shown in Fig.
8.3(d). This contribution is not included in the series shown in Fig. 8.3(c). It is of order O(n;>
and is neglected. The sum of ladder diagrams omits many terms. However, the omitted terms
are not as important in the limit where n i ~ O.
The first two terms in the sum of ladder diagrams have already been derived; they are
n(O) , and n(1), in (8.44) and (8.58). The superscript denotes the number of ladders. The
superscript (L) denotes sum ofladder diagrams. The series of terms in the ladder sum can be
generated by representing them as a vector vertex function r(L):
2e'2 1
n(L)(iro) = - 3 L
2 LA p' r(L)(p, ip, ip + iro)~(P, ip)~(p, ip + iro) (8.59)
m v p I-' ip
r(L)(p; ip, ip + iro) = p + -1 L r(L)(p'; ip, ip + iro)W~~(ip, ip + iro)

v p'
x ~(p', ip)~(p', ip + iro) (8.60)
Repeated iteration of (8.60) will generate the series of terms in the ladder summation shown
in Fig. 8.3(c). The ladder summation will be evaluated, approximately, in order to obtain the
factor of (1 - cos 9') in the relaxation time.
The correlation function is a function of ip and iro in the combination ip and ip + iro.
Define the quantity
2e'2
P(ip, ip + iro) = -32 L p' r(L)(p; ip, ip + iro)~(p, ip)~(p, ip + iro)
m v p
(8.61)
1
n(L)(iro) = A 1;. P(ip, ip + iro)
I-' Ip
The summation over Matsubara frequencies ip is evaluated, as usual, by examining the

contour integral JdznF(z)P(z, z + iro). The integrand has the poles of nF(z), which give the
summation over ip, and also branch cuts along the two axes z = E and z = E - iro, where E is
real. These cuts are shown in Fig. 8.4. The integral equals the contribution from the two
branch cuts, where one has to subtract the parts above and below each cut:
(8.62)
The infinitesimal part ±io is unnecessary when there is a term with ±iron- Next find the
retarded function from the analytical continuation iron -+ ro + iO. A variable change
E -+ E + ro in the last two terms brings us to the point
1t~~l(ro) = Joo 2dE. {[nF(E) - nF(E + ro)]P(E - iO, E + ro + io)

-00 m
- nF(E)P(E + iO, E + ro + io) + nF(E + ro)P(E - iO, E + ro - io)} (8.63)
The next step is to take the imaginary part of this expression. Because of the factor of i in the
denominator 21ti, this step gives the real part of the integrand. The subsequent step is to take
the limit that ro -+ O. The function peE - iO, E + ro + io) becomes peE - iO, E + io) at zero
frequency, which is real. It is real because the function is symmetric in its arguments,
P(z\, zz) = P(zz' z\), so that the complex conjugate of peE - iO, E + io) equals itself. The last
two factors of P in the above equation are complex conjugates of each other. Taking the real
part of the integrand removes their imaginary parts, and the real parts are equal.
The limit ro -+ 0 again only involves [nF(E) - nF(E + ro)]/ro. These steps brings us to a
formal expression for the dc conductivity:
0' = _ lim
Ol--->O
(Im[ro1t~l])
= JOO dE. (_ dndF(E)) {P(E _ iO, E + iO) - Re[P(E + iO, E + iO)]}
-00 2m E
There are only two functions which need to be found, peE - iO, E + io) and peE + iO, E + io).
At zero temperature, where -dnF/dE = O(E), they need to be found only at E = O. These two
functions have quite different behavior and are obtained by different methods. Both are
usually important, but the most singular is peE - iO, E + iO), and this term leads to the factor
of (1 - cos 9') in the lifetime.
~
CD-
Zo.-Iw
l~' zo•
x
FIGURE 8.4 Branch cuts in the contour integral.
Consider the definition of Pin (8.61):
2il
P(e - iO, e + iO) = - 32 L p' r(L)(p, e - iO, e + iO)~(p, e - iO)~(p, e + iO)
m v p
The product of Green's functions is entirely real:
1
~(p, e- iO)~(p, e + iO) = Gady(pe)Gret(p, e) = 2
(02 +A
A(p, e)
(8.65)
2A(p, e)
(0 = e - ~p - Re[~(p, e)], A= -Im[~]
. . _ il d 3p A(p, e) (L) J
. .
P(e-lO,e+lO)--32 --3~()p.r (p,e-lO,e+lO)
(8.66)
m (2n) L1 p, e
The combination GadyGret is rigorously defined as the spectral function A(p, e) divided by
2A(p, e). There is no assumption that A2 = A/2A. The same combination is found in the
equation for the vertex function:
r(L)(p, e - iO, e + iO) = P + J(~~~ r(L)(p', e - iO, e + iO)
x W~~(e - iO, e + iO)Gady(p', e)GretCP', e) (8.67)
The vector function r must point in the direction of p since that is the only vector in its
function arguments. It is convenient to define an integral equation for the scalar function:
py(p, e) = r(L)(p, e - iO, e + iO) (8.68)
Jd3p' A(p'p ,e
yep, e) = 1 +
(2n)
3 2A(;
e) p' p'
) 2
p
W~~(e - iO, e + iO)y(p', e) (8.69)
. . il Jd p 3 A(p, e) 2
P(e - lO, e + lO) = - 32 - - 3 ~()p yep, e) (8.70)
m (2n) L1 p, e
Equation (8.69) is a one-dimensional integral equation for the scalar function yep, e), where p'
is the integration variable. The angular integrals do.p ' just average the quantity p . p' ~;! over
angles to provide the kernel for the integral equation. The integral equation should not be
difficult to solve with modem computers for realistic self-energies and T-matrices.
The Boltzmann result is obtained in the twin limits T -+ 0 (e -+ 0) and ni -+ 0, where
A(P', 0) -+ 2no(~p')' Equation (8.69) then reduces to the integral equation
yep, e) = 1 + Jd3p' 2A(p', e) p'p2p' ndTpp,

(2n)3
2no(~p') 2,
1 y(p, e) (8.71)
A-A,
y(kF) = 1 + y(kF)-A- (8.72)
(8.73)
which is easily solved to give

A
y(kF ) = Al (8.74)
The factor (1 - P' p'lp2) = (1 - cos a') since Ip'l = Ipl. The solution y = AI Al is put into
(8.65) to give
. . e'-
pee - 10, e + 10) = -32 --3P A ( )
Jd p 3 2 A(p, E)
(8.75)
m (21t) I p, e
which gives the same conductivity as (8.13) when put into (8.64). The term peE - iO, E + iO)
leads to the important contribution as T ~ 0 and nj ~ O.
The other term pee + iO, e + iO) should not be neglected. It is complex but not very
singular in the limit where nj ~ O. As is evident from the definition (8.61) the singular parts
should arise from the Green's function product:
. 2 2 1 (ji - A2 - 2i(OA
G(p, E + 10) = Gret(p, E) = . 2= (8.76)
«(0 - IA) «(02 + A2i
1 2A2 2i(OA
(8.77)
= (02 + A2 (roZ + A2)2
= ~ - ~A2 - i Re[G)A (8.78)
As nj ~ 0 then A2 ~ AI A and the real part of this expression vanishes. The imaginary part
becomes (00«(0), which is also zero. The singular parts of this expression vanish as nj ~ O.
The vertex corrections to peE + iO, e + iO) are not of order unity but of order O(nj). In this
case, the vertex corrections may actually be a series of terms which are successively smaller,
so that the vertex corrections may be obtained by just evaluating the first few. The situation is
quite different than for pee - iO, e + iO), where one has to solve the vertex equation and sum
all the ladder diagrams.
One can evaluate pee + iO, E + iO) by solving a vertex equation similar to (8.69). Define
the scalar vertex y' by r(L)(p, E + iO, e + iO) = py'(p, e), and then
Y'(P)
,e = 1 + ni J--3 P'P'r
d 3p'
p
(21t)
pp' E Y
-2- ()2 '(p' ,e)Gret (p' ' e)2 (8.79)
The vertex function y'(P, e) is complex, as is the vertex Tpp,(Ei, and the product Gret(p', ei.
There are two coupled equation for the real and imaginary parts. It may be obtained from the
Ward identity (1950),
r(p, E + iO, E + iO) = P + mVp~(p, e + iO) (8.80)
which is an exact identity between the exact vertex function rep, e + iO, e + iO) and the exact
retarded self-energy ~(p, e + iO) from impurity scattering. This self-energy is not the one
derived in Sec. 4.1, since that was only for scattering from a single impurity and only had
unperturbed Green's functions ~(O) as the internal lines in the scattering equation. The exact
self-energy is found from scattering from all numbers of impurities and using exact Green's
functions as internal lines of diagram-although this procedure must be done carefully in
order not to count the same contribution twice. The Ward identity is very convenient, since it
permits the vertex function to be obtained from the self-energy function by a simple opera-
tion. It will be proved in the next section.
The present derivation has an approximate vertex function which contains only the
ladder diagrams. In that case the Ward identity is still valid as long as the self-energy is found
using only diagrams where an electron scatters from one impurity at a time.
The Green's function derivation of this result is certainly harder than the transport
equation of Sec. 8.1.1. The one advantage of the Green's function method is that there is no
need to make the approximation that ni --+ O. The ni --+ 0 limit is made implicitly in solving
the Boltzmann equation when it is assumed the particles are plane waves except for occa-
sional scattering events from isolated impurities.
There are several lessons to be learned from this summation of ladder diagrams. The
ladder diagrams, although they appear in a higher order of perturbation theory, do not lead to
terms in the final answer which are smaller than lower-order terms. A term is not necessarily
small if it occurs in a higher order of perturbation theory. It is this feature of vertex
corrections which means they can never be dismissed without an investigation. They may be
small, but one should always check. It would be nice to have some rules of thumb which
would establish whether vertex corrections are important. The way to tell is to examine the
scattering process which causes the vertex functions. If the two particle states, on each side of
the bubble, can scatter quasielastically so that their relative energy changes little, then vertex
corrections are large. The vertex correction is basically a potential divided by an energy
denominator V / M, where M is the change in energy of the two particles. If it is small, the
small denominator will compensate for the small potential V, so that vertex corrections
become sizable. Repeated scatterings, as in a series of ladder diagrams, just cause additional
powers of the factor (V / M)n. Vertex corrections are large when the scattering by the potential
causes only a small change in the relative energy of the two particles.
8.1.3. Ward Identities

The evaluation of a two-particle correlation function, such as the Kubo formula for the
conductivity, often requires an evaluation of a vertex function. The Ward (1950) identity is an
exact relationship between the vertex and self-energy functions in the problem. As an
example, two types of Ward identities permit the evaluation of the scalar vertex function
rep, ip) or the vector vertex function rep, ip) which satisfy the equations
d3p'
rep, ip) = 1 + f - - 3 rep', ip)~(p', ipiWpp,(ip, ip) (8.81)
(2n)
rep, ip) = P+ f d3p' rep', ip)~(p',

--3
(2n)
ipiWpp,(ip, ip) (8.82)
The Ward identity states that these two functions are given by
rep, ip) = 1_ [aL~' Z)] . (8.83)

Z' Z=IP
rep, ip) = p + mVpL(p, ip) (8.84)
An evaluation of the self-energy function L(p, ip) permits an easy evaluation of these two
vertex functions. These relationships are proved below.
An important point regarding the Ward identities is that they are not useful for evaluating
all vertex functions. An example is provided in the last section, where the Ward identities
were useful for finding P(E + i8, E + i8) but not P( E - i8, E + i8). The Ward identities cannot
be applied blindly; they must be used only when appropriate. These circumstances are
delineated after the identities are proved.
The Ward identities for impurity scattering were derived by Langer (1961). The similar
theorems for the electron-phonon interaction were derived by Engelsberg and Schrieffer
(1963). We shall prove the result for the ladder diagrams obtained by scattering from a single
impurity. In this case, the self-energy diagram is that for scattering from a single impurity:
(8.85)
An important condition is that the Green's functions in this self-energy diagram are those
calculated with the self-energy <'§ = 1/[ip - ~p - L]. Equation (8.85) is a self-consistent
equation for the self-energy L, since it depends functionally on itself. Unfortunately, the Ward
identities do not let one avoid solving an integral equation. Instead, one integral equation is
exchanged for another. In this sense the Ward identities are not very useful in practice.
Rather than prove the two separate identities (8.83) and (8.84), a general theorem is
proved for which these are two limiting cases. The general theorem is obtained by subtrac-
ting the expressions (8.85) for L(p, ip) == L(P) by the same result for
L(p + q, ip + im) == L(P + q):
L(P + q) - Jd3p\
L(P) = nj (2n)3 VpPI VpIP[<'§(P\ + q) - <'§(P\)]
d3p d3p
+ nj J (2\n)6 2 VpPI VplPZ Vpzp[<'§(p\ + q)<'§(P2 + q) - <'§(P\)<'§(P2)]
n. Jd3p\d3P2d3P3 v: v: v: v:
+ I (2n)9 PPI PIPZ P2P3 P3P
(8.86)
By purely algebraic manipulations, this series can be shown to be identical to
L(P + q) - L(P) = nj J~~~ TpPI (ip)Tpl+q,p+q(ip + im)[<'§(p\ + q)<'§(P\)]

x [L(P\ + q) - L(P\) + ~PI+q - ~PI - im] (8.87)
= nj J~~~ TpPI (ip)Tpl+q,p+q(ip + im)[<'§(p\ + q) - <'§(P\)] (8.88)
This rather startling result may be demonstrated term by term. The first nonvanishing term has
the integrand
VpPI VpIP[<'§(P\ + q) - <'§(P\)] = VpPI VpIP<'§(P\ + q)<'§(P\)
X [L(P\ + q) - L(P\) + ~PI+q - ~PI - im] (8.89)
,
p" I • pi
J
I
p+q ...---1'"--<..- p'+ q
(a)
FIGURE 8.5
which is just the vertex diagram shown in Fig. 8.5(a). The next term in the series (8.86) has
the integrand
VpPI VPIP2 Vp2p[~(P1 + q)~(P2 + q) - ~(P1)~(P2)] (8.90)
The Green's function factors in brackets may be rearranged into

~(Pl + q)~(P2 + q) - ~(P1)~(P2) = ~(Pl + q)[~P2 + q) - ~(P2)]
- ~(P2)[~(P1 + q) - ~(Pl)] (8.91)
The first bracket on the right is the first diagram in Fig. 8.5(b). The factor VpPI VPIP2 ~(Pl + q)
is just the multiple scattering from the impurity by the electron on the lower line. Similarly,
the second term in (8.91) corresponds to the second diagram in Fig. 8.5(b), where the factor
VPIP2 Vp2p~(P2) is the multiple scattering of the top electron line. These terms are just the first
in the series which generates the T-matrlces. The further terms in (8.86) provide the remaining
terms. In this manner, one can establish the validity of (8.88).
Equation (8.87) shows the quantity
A(p,p + q) = ~(P + q) - ~(P) + ~p+q - ~P - ioo (8.92)
obeys the vertex equation

A(p,p + q) = ~p+q - ~P - ioo
+ ni J(2n)3
d3PI
~(PI)~(P1 + q)TpPI Tpl+q,p+qA(PI ,PI + q) (8.93)
The two equations (8.92) and (8.93) provide the most general type of Ward identity. They are
useful, since any equation which can be cast into the form of (8.93) has the solution (8.92).
Langer (1961) and Engelsberg and Schrieffer (1963) show that this equation is related to the
equation of continuity V . j + p = O.
The first Ward identity (8.83) is obtained by taking the limit q = 0 and then dividing Eq.
(8.93) by -ioo with the result
A(p, ip, ip + ioo)
---'- = 1 + ni J-(2n)
d PI . 3
..
----"'----"----"-.
-lOO
- 3 ~(PI' lp)~(PI' lp + lOO)
')T. (. +. )A(PI' ip, ip + ioo)

T.PPI ( lp (8.94)
x PIP lp lOO -ioo
The quantity rep, ip) in (8.81) obeys the same equation as A/(-ioo) in the limit ioo --+ 0 so
they are equal. From (8.92) one has the solution
rep, = .lim [A(P, ip, ~P + iOO)] = 1 _ [a~(p, Z)]

az
l'P' )
10l-+0 -lOO Z=ip (8.95)
which proves the Ward identity (8.83).

Sec. 8.2 • Mobility of Frohlich Polarons 517
The other Ward identity is found as the limit im = 0, followed by letting q -+ O. The
latter limit is taken slowly, so that one can retain terms proportional to q. In this limit Eqns.
(8.92) and (8.93) become
lim A = q' p + q' VpL(p, ip) (8.96)

q---+O m
q'P
= --;;;- + ni J(27t)3
d 3P I . .
~(PI' Ip)~(PI + q, Ip)TpPI Tp,+q,p+qA(P1 ,PI + q) (8.97)
The vertex function A is proportional to q, so define the vector vertex function by the limit
(im = 0):
lim A ="!"q. rep, ip) (8.98)

q---+O m
Then the preceding two equations can be expressed in terms of this vector vertex function:
rep, ip) = P + mV pL(p, ip) (8.99)
(8.100)
This equation is the same as (8.82), which proves the other Ward identity (8.84). Both are
now understood to be limiting cases of the general result (8.93). The Ward identities are
useful anytime one can cast the vertex equation into the form (8.93).
The factor [1 - aL/aL] is recognized as the inverse of the renormalization Z defined
earlier and discussed, for example, in Sec. 5.8.1. This quantity is sometimes called the
effective charge. Similarly, the vector vertex is one of the factors which give the effective
mass of the particle. The Ward identities relate the vertex corrections to a change in the
effective charge and mass of the particle, which is why they are related to the equation of
continuity.
S.2. MOBILITY OF FROHLICH POLARONS
The Frohlich Hamiltonian between electrons and Einstein phonons (mo = mw) is
_ t t Mo 1 t t
H - L EpCpcrCpcr + mo L aqaq + h. L - Cp+q,crCpcr(aq + a_q)
pcr q '\Iv pq Iql
(8.101)
M, 2 _ 41t1xll(llm0 )3/2
0-
,J2mB
(8.102)
It was discussed in Sec. 7.1, where several important quantities were derived, such as the
effective mass m* and the ground state energy Eo.
The effective mass m* of a particle can be measured by cyclotron resonance. Such
experiments have been done for polarons (Hodby, 1972). The effective mass m* is a function
of the band mass mB and the polar coupling constant ex. A separate measurement of the two
dielectric functions Eo and Eoo ' as well as mw, permits a determination of the band mass mB
518 Chap. 8 e de Conductivities
and IX from m*. This analysis takes a theory of the polaron mass m*(mB' IX). which was
provided in Sec. 7.1. Another way to check the theory is to measure the mobility of electrons
in insulators. The mobility also depends upon mB and IX.
A typical experimental result for the mobility is shown in Fig. 8.6. It shows the Hall
mobility of CdTe measured by Segall et al. (1963). The steep rise around 200K is due to
optical mode scattering. At lower temperatures the mobility saturates because of the scattering
from impurities in the crystals. Impurity scattering varies from sample to sample, depending
on the concentration and type of impurity.
The average value of the current operator is the particle density no times the charge e
times the average velocity (v). The average velocity (v) is proportional to the applied electric
field F, and the constant of proportionality is the mobility, where e > 0 so the electron charge
is -e:
(v) = J!F (8.103)

J = -enoJ!F (8.104)
105
8
4 +
+
U
...
II
I
2 o
15>
N
"- o
E 4 o
2 10
>-
..... 8
• ••
::i • • ••
••
iii 6
0
2
•
..J 4 •
..J
«
:I:
103
•
8
••••
•aD aD aD
6 ~..! ___ - - e - _ _ _ --_ea
4 X102
10 20 30 40 60 80 100 200 300 400 600 800
T(·K)
FIGURE 8.6 The temperature dependence of the electron mobility of several samples of n-type CdTe. Different
samples have different kinds and concentrations of defects or impurities. The solid line is the theoretical mobility
from optical mode scattering. including the temperature dependence of the static dielectric constant. The dashed line
neglects this change. Source: Segall et al. (1963) (used with permission).
The mobility /l is the average velocity of each electron per unit applied electric field. Of
course, it is strictly defined in the limit of vanishing electric field. Since the electrical
conductivity (J = noe2Ttim = -e/lno is the ratio of the current J to the field F, then
/l = -eTtim.
The theories of the electron mobility in insulating materials, such as alkali halides and
II-VI semiconductors, treat it as a property of a single electron. The electron lifetime is
calculated for the scattering from impurities and by acoustical and optical phonons. The
electron-electron interactions can be ignored in the limit where the concentration of electrons
is very low. Spin is also unimportant, and the spin index is omitted in this section.
There are just as many different ways to calculate the polaron mobility as there are to
calculate the effective mass m* or ground state energy Eo. Each theoretical technique was
applied to the mobility as well as to the other quantities. As summarized by Langreth (1967),
these various methods usually agree in the limit of weak coupling (rY.« 1) and low
temperature kBT « limo. In this case
(8.105)
1
No = -e~;C-(O-o-_-l (8.1 06)
The lifetime TO is the result obtained in (7.46) as liTo = -2 Im[L(1)(p, E)] in the limit where
p -+ 0 and E -+ 0, where L(l) is the one-phonon self-energy. This limit is appropriate, since at
very low temperatures the electrons are in states within kBT « limo of the bottom of the band.
These low-energy particles cannot emit phonons, since this event is prevented by energy
conservation. They can only absorb them, and the rate of absorption is proportional to the
thermal average density of phonons No. The factor No makes the mobility increase expo-
nentially with decreasing temperature, since the electron scattering becomes less likely as the
number density of phonons declines. The exponential increase in the mobility is evident in the
experimental data of Fig. 8.6. The behavior of large polarons is opposite to that of small
polarons, whose mobility increases with increasing temperature.
One feature of the mobility formula (8.105) is that it is proportional to the inverse ofrY..
Our starting point for the theoretical calculation is again the Kubo formula, which will be
evaluated for using electron-phonon interactions. The expansion of the S matrix for this
potential will generate a series in the parameter rY.. To obtain a leading term in the inverse
power of rY. requires the summation of a subset of diagrams. The situation is similar to the
mobility from impurity scattering, where diagrams were summed to get the conductivity
inversely proportional to the impurity concentration n i . There are important differences
between the lifetime from opical phonon scattering and that of impurity scattering. This
conclusion is evident from the result presented in (8.105). Here the relaxation time is not
calculated with the factor of (1 - cos 9') in the angular average. The polaron mobility is
calculated in a different way than the scattering from impurities. Actually it is calculated in
the same way, but a different result is obtained. This difference arises from the inelastic nature
of the polaron scattering as first shown by Howarth and Sondheimer (1953).
Langreth and Kadanoff(1964) showed that the polaron mobility /l is a power series in rY.,
with the leading term in (8.105) of order 1I rY.:
(8.107)
They tried to calculate the coefficient a_l' aoin this series in the limit of low temperature. This
objective is simple but has slippery aspects (Mahan, 1966). It is a subtle procedure to examine
each term in the S-matrix expansion and to determine its leading term in (J(. The situation is
similar to that for impurity scattering. Each term in the series (8.107) is obtained by summing
subsets of diagrams.
The first term of Langreth and Kadanoff (a-l) is quite simple, and it is probably
worthwhile to state it in advance. Define J.1o as the result (8.105) for the limits (J( ---+ 0, T ---+ O.
They found
1:. = 1 - ~ + O«(J(2) (8.108)

J.1o 6
They observed that J.1o/J.1 is precisely the expansion given by the equations
e't
J.1=-- (8.109)
m*
't
- = 1 + O«(J(2) (8.110)
'to
m* (J(
- = 1 +_+0«(J(2) (8.111)
mB 6
The ballistic formula J.1 = -e't/m* supports the quasiparticle picture that the particle acts as if
it has an effective mass m* and lifetime 't. They speculate that the quasiparticle picture would
be valid for all values of (J( and that the inclusion of all terms in (J( would just reproduce the
product series of m* and't.
Before the derivation of the electron mobility, it is necessary to derive some single-
particle properties. These will be needed in the limit of zero temperature. All terms are
dropped of order No compared to unity. Many of these single-particle properties were derived
in Sec. 7.1. The first self-energy term, proportional to (J(, is the one-phonon result in (7.15):
(8.112)
At zero temperature, this self-energy is evaluated at small Ep and small 0),
Re[~(1)(p, 0)] = -(J([O)o + to) - tEp + 0(0)2, E;, O)Ep)] (8.113)
This expansion permits a quick derivation of the effective mass m* and the renormalization
coefficient Z. The zero subscript means p = 0, 0) = 0:
(8.114)
(8.115)
Another important quantity is the lifetime 't which is defined as

1
't(p) = Z(P){ -2 Im[~(P, Ep)]} (8.116)
where Z(P) is the renonnalization coefficient evaluated at ro = Ep' which is the ground state
energy. The ground state energy is only needed to order (J(, which is the simple result
Ep = -(J(roo + p2 12m* + O(P4).
Equation (8.116) shows that the renonnalization coefficient Z(P) enters into the defini-
tion of the lifetime. The argument for this is as follows. The spectral function is defined as
A(P, ro) = -2 Im[Gret(P, ro)] (8.117)

-2 Im[I:(P, ro)]
(8.118)
- {ro - cp - + {Im[I:(p, ro)]}2
Re[I:(p, ro)]f
A suitable definition of't(p) is obtained by examining this limit more carefully when Im(I:) is
small but not infinitesimal. In the vicinity of the peak ro ~ Ep of the spectral function the tenn
in the denominator is
aRe I:
ro - cp - Re[I:(p, ro)] ~ ro - cp - Re[I:(p, Ep)] - (ro - Ep)---ac;;-
~ ( _ E )[1 _ aRe I:J = (ro - Ep) (8.119)

ro p aro Z(P)
so that the spectral function is approximately
(8.120)
(8.121)
The last step used the definition (8.116) of the relaxation time. The relaxation time 't(p) is
treated as a function of p but not ro. This fonn for the spectral function is used in the Green's
function (Problem 6 in Chapter 3),
(8.122)
= -iZ(P)0(t) exp[ -itEp - tl(2't)] (8.123)
The relaxation time is defined from the decay of the Green's function. It has the desired fonn,
with the relaxation time 't(p) detennining the decay of the excitation.
There is another way to understand the factor Z in the definition (8.116) of't(p). The
quantity -2 Im(I:) is the rate of decay of a state (p, ro). The factor Z is the fraction of the
quasiparticle strength at the value (p, ro). The rest of the quasiparticle strength is usually
dispersed throughout the spectrum.
The one-phonon self-energy is evaluated for the quasiparticle lifetime. The imaginary
self-energy is evaluated at the quasiparticle energy co = Eo = -acoo and is multiplied by the
factor of Z given in (8.115). The imaginary self-energy is calculated from the expression
-2 Im[L(p, co)] = 21t J--3--T

d3q M2
(21t) q
(8.124)
X [N0 8(co + COo - t q +p ) + (No + 1)8(co - COo - t q+p )]
= a( coo)3/2 [N0 8(CO + COo) lnl fo+COO + ~ I

~ Jco+coo - ~
+ (No + 1)8(co - coo) lnl ~ + ~

Jco-coo - ~
I] (8.125)
Expanding for small co < COo and small tp gives
2aNoco~/2 1
-2 Im[L(O, co)] = ~ ~ 2aNo[co o - 2CO + ... ] (8.126)
y COo + co
- 2 Im[L(O, co = -acoo)] = 2aNocoo [1 + ~ + O(a2 ) ] (8.127)
~, = Z[-2 Im(L)] = ~
'0
(1 -::) (1 +::)
2 2
= ~[1 + O(a 2 )] (8.128)

'0
The first correction terms in a from Z = 1 - a/2 and Im[L] = 1 + a/2 cancel to order O(a).
The one-phonon term provides no correction term to , of O(a).
The electron mobility is evaluated be the same method which was used for impurity
scattering. Starting from the Kubo formula for the electrical conductivity, the first important
diagram is the bubble with interacting Green's functions shown in Fig. 8.7(a). The solid lines
are total Green's functions '§(p, iPm) which include the self-energies
(8.129)
(8.130)
The solid lines are total Green's functions '§(p, iPm) which include the self-energies. The
evaluation of this term is identical to that used for the same bubble diagram for impurity
0- <D
( 0) (b)
~
''-.LV
(c)
FIGURE 8.7
scattering. The contribution to the Kubo formula contains the two Green's functions as in Eq.
(8.44), which are evaluated by the standard series of steps to yield Eq. (8.49):
(8.131)
In the present application, the self-energies in the Green's functions are evaluated from the
Frohlich Hamiltonian rather than impurity scattering. In a more realistic model of a solid, both
self-energy expressions should be included: from phonons and impurity scattering.
An approximate evaluation of this contribution to the conductivity is obtained in the
limit where IX « 1. The electron distribution is assumed to be Maxwell-Boltzmann:
(8.132)
where nF(s) is the energy distribution for each spin state, while no is the total concentration
for both spin states. This term in the conductivity is divided by -eno to get the corresponding
term in the mobility of each electron:
(8.133)
Using the expression (8.121) for the spectral function, the square of the spectral function is
(8.134)
The electron lifetime 't' is inversely proportional to IX, so that the mobility ~(o) is inversely
proportional to IX. The preceding integral is evaluated in the limit where the temperature
T -+ O. The integral is easy if it is assumed at low T that Z(P) and 't'(p) are evaluated at p = 0
(8.135)
where Z(O) = 1 - 1X/2 and m* /mB = 1 + 1X/6. The result (8.135) is the contribution from the
simple bubble diagram of Fig. 8.8(a). This result does not resemble (8.108). The differences
disappear when higher-order diagrams are considered, such as those in Fig. 8.7.
Other contributions to ~ can be derived from the other diagrams, which are the vertex
corrections. In the limits T -+ 0 and IX « 1 the vertex corrections do not contribute to the
mobility a term which goes as O(I/IX). The simple bubble result is the final answer at low
temperature and weak coupling. This conclusion is quite different from the situation for
impurity scattering. There a series of ladder diagrams was summed in order to derive the final
answer, and each ladder diagram gave a term which was the same inverse power of the
coupling constant as the impurity concentration ni' The vertex diagrams for optical phonon
scattering are less important because of the inelastic nature of the phonon scattering. Mahan
(1966) showed that the two-phonon ladder diagrams in Fig. 8.7(c) provide the largest vertex
corrections, which are of OCT /(00)'
8.3. ELECTRON-PHONON RELAXATION TIMES
This section is concerned with the calculation of electron relaxation times in semi-
conductors from the scattering by acoustical phonons. The prior sections have introduced two
different relaxation times for the electron in a solid. One of these is the average time between
scattering events and is denoted as t(k). The mean-free-path (mfp) for this relaxation time is
denoted as /(k) = vk t(k). The energy bands in the solid are assumed to be isotropic, so that
the relaxation time and mfp depend only upon the magnitude of the wave vector. The other
relaxation time tt enters into the electrical conductivity (J = noett/m*). The equivalent
quantity, before averaging over wave vector, is called the momentum relaxation time and is
denoted with the subscript "t" since it is used in transport of current. The equivalent mfp is
It(k) = vk tt(k).
To further confuse the topic, the transport lifetimes tt for scattering from impurities had
a factor of (1 - cos 9') in the integrand of the scattering integral. The scattering from optical
phonons did not have such a factor. The difference is that at low temperature, impurity
scattering is elastic, while optical phonon is highly inelastic. Vertex corrections are relatively
unimportant for inelastic scattering.
In writing the Boltzmann equation, the last term in the time development is the rate of
change of the distribution function from collisions. The discussion of the lifetime from
impurity scattering in Sec. 8.1.1 started with this collision term. Here the similar collision
term is presented from the scattering due to the electron-phonon interaction:
( af(p))
at ep
= 2n Jd q 1M 12{f(p)[1 - f(p + q)]
h (2n)3
3
q
x [(Nq + l)O(Ep - Ep+q - hcoq) + NqO(Ep - Ep+q + hcoq)]

- f(p + q)[l - f(p)] [(Nq + l)o(Ep - Ep+q + hco q) + NqO(Ep - Ep+q - hcoq)]}
(8.136)
The integrand has four terms. The first two correspond to an electron initially in state p
scattering to p + q by either the emission (Nq + 1) or absorption (Nq ) of a phonon. This term
is multiplied by the occupation factorsf(p)[l - f(p + q)] which ensures that the initial state p
is occupied and the final state is empty. The other two terms correspond to processes whereby
electrons initially in p + q scatter back into the state p by either phonon emission or
absorption. The back scattering has the occupation factors off(p + q)[l - f(p)]. The above
expression vanishes when the system is in thermal equilibrium and f(p) = nF(Sp),
Nq = nB(co q ).
An immediate simplification of the above formula is attained by grouping together the
terms with the same delta function for energy conservation
( af(p))
at ep
= 2n
h
Jd q 1M 12
3
(2n)3 q
x [O(Ep - Ep+q - hcoq){f(p)[Nq + 1 - f(p + q)] - Nq!(p + q)}

+ O(Ep - Ep+q + hcoq)]{f(p)[Nq + f(p + q)] - f(p + q)[Nq + I]}] (8.137)
The above expression is quite general.

Sec. 8.3 • Electron-Phonon Relaxation Times 525
(8.138)
X {B(Ek - Ek+q - Iiroq)[Nq + 1 - nF(Ek+q)]

+ B(Ek - Ek+q + Iiroq)[Nq + nF(Ek+q)]} (8.139)
1M 12 = D2 Iiq2 (8.140)
q 2proq
N q = nB(roq ) (8.141)
The subscript "p" denotes scattering by phonons. In Eq. (8.138) the relaxation time from
phonon scattering is all of the terms which multiply f(P) in (8.137). It is identical to the
expression obtained from the imaginary part of the electron self-energy, as calculated in the
one-phonon approximation. The occupation factors f(p + q), Nq are represented by their
values in thermal equilibrium. The matrix element is from deformation potential interactions.
The transport lifetime 'tpt from the electron scattering by phonons is defined in (8.140). It
has an additional factor of ( -q . kj 12) in the integrand. If the scattering were elastic, then this
factor equals (1 - cos a'). The difference between the impurity scattering times 'ti' and 'tit' is
the factor of [1 - cos(a)] in the integrand of lj'tit, where a is the scattering angle
k.p = cos(a).
1
(k)
21tni d 3p J 2
= -----;;:- --3 ITkpl B(Ek - Ep) (8.142)
'ti Tl (21t)
1 21tni
=-l!.-
J--3ITkpl
dp3 2
[1-k'p]B(Ek- Ep)
A A
(8.143)
'tit (k) TI (21t)
The identical factor is used to calculate the momentum relaxation ('tft ) from scattering by
phonons. Ifp = k + q then the angular factor [1 - cos(a)] ~ -q . kjk . The identity is exact
if p = k. Using this factor of [1 - cos a] in the impurity scattering is rigorously correct. Using
this factor in the scattering by phonons is not rigorously correct. The difference is that
the scattering by phonons is inelastic. Instead, the correct result is found by solving the
Boltzmann equation, which is done in Sec. 8.4. However, the above approximation is actually
quite good, and therefore is useful. The lifetimes for scattering by phonons are evaluated
below.
Besides these two relaxation times, there are several others which are occasionally
useful. The most important is the scattering by electron--electron interactions in a metal. The
formulas for this case are derived in Chapter 11 in the discussion of Fermi liquid theory. Two
others are the relaxation times for temperature, and the relaxation times for energy. These two
are discussed below.
The relaxation time for energy determines the rate at which the electron loses or gains
energy from the scattering. Since impurity scattering is elastic, the electron does not change
its energy. There is no contribution to energy relaxation from impurity scattering. However,
in the scattering by phonons the electron changes its energy by Ek+q - Ek = ±liooq . One
process gains energy (+liooq ) while the other loses energy (-lioo q ). There is a net gain or loss
of energy if one process dominates over the other. For example, if an energetic electron
is injected into the solid, it will gradually lose energy until it equilibrates thermally. The
process of coming to thermal equilibrium requires that it emit more phonons than it
absorbs.
The rate of energy relaxation is calculated by starting from (8.138) and inserting the
energy change Ek+q - Ek into the integrand
(~)
t pe
= - J
2lirc d 3q3 IMiIiOOq{O(Ek -
(2rc)
Ek+q -IiOOq)[Nq + 1 - nF(Ek+q)]
- O(Ek - Ek +q + IiOOq)[Nq + nF(Ek+q)]} (8.144)
Converting this expression to a relaxation time ('tpe ) and mfp (lpe) is done below.
The evaluation of the relaxation times for the electron-phonon systems is done for
metals and semiconductors. In metals, the evaluation assumes the existence of a Fermi
surface. The electron-phonon scattering affects those few electrons within about kBT of the
Fermi surface. In semiconductors, the interesting effects are for a single electron in the band,
and the reference energy is the band edge. These two cases are different and are treated
separately.
8.3.1. Metals
In metals the derivation is simplified by introducing the McMillan function denoted as
ri.2F(oo). In this function are collected all of the complicated parts of the phonon dispersion,
and the matrix element (see Sec. 7.4)
(8.145)
(8.146)
(8.147)
The dimensionless functions r:x.ft)F(E, (0) depends upon the variable E. However, this
dependence is similar to the variation in the density of states at the Fermi surface. The
variation with E is usually smooth on the energy scale of the Debye energy, which is all that is
relevant for interactions with phonons. The important variation is
(8.148)
The second term is usually smaller than the first by a factor ofliro/W, where liro is a phonon
energy while W is an electronic bandwidth. The usual approximation is to neglect the
dependence upon the factor of E The above expression is rewritten as
1 27t J(j)D
-(k) = h drolXft)F(ro)[2nB(ro) + nF(ro - Ek) + nF(ro + Ek)] (8.149)
'tp(t) 0
High temperature is defined as T greater than the Debye temperature. Most solids have
Debye temperatures less than room temperature, so that T = 300 K is a high temperature. In
this limit, the largest term in the bracket is nB(ro) ~ kBT /liro and the lifetimes have the simple
expression
(8.150)
(8.151)
(j)D dro 2
A(t) = 2J -1X(t)F(ro) (8.152)
o ro
The inverse lifetime is proportional to temperature, and the constant of proportionality is
the dimensionless electron-phonon coupling constant A(t). The resistivity of metals
[p = m/(eno'tt)] at high temperature is linear in T, and the slope is given by At. A
measurement of peT), gives At, which can then be used to predict the transition temperature
for superconductivity in that meta1. This process works quite wel1.
The equation for energy relaxation is
( dE) J(j)D
dt pe = 27t1i 0 droroIX2F(ro)[1 - nF(Ek - ro) - nF(Ek + ro)]
(8.153)
A(ro2 } = 2 J:D droroIX2F(ro) (8.154)
The frequency integrals go from zero to the Debye frequency roD = kB 0 D /Ii. The occupation
numbers were expanded nF(Ek ± ro) ~ nF(Ek) ± ron~. The first derivative terms cancel and
the second derivative terms are small and are neglected. The factor [1 - 2nF(Ek)] is one for
electrons and minus one for holes. The energy relaxation always takes the particle to the
chemical potential.
The energy relaxation (dE/dt) is governed by the quantity A(ro2 }. It is a single function,
although the notation gives the impression that it is a product of two functions. The quantity
IiA(ro2 } has the units of Watts. In this case there is no obvious lifetime 'tpe nor mfp (lpe)' The
phonons give a constant value to the energy relaxation as long as the electron's energy
~ = Ek - I! > Ii roD'
8.3.2. Semiconductors
The relaxation time for a semiconductor is calculated assuming that there is only one
electron in the band. Most semiconductors have electrons or hole in one or several band
minimum. In order to keep the discussion simple, the present calculation will be done
assuming there is a single conduction band at the center of the Brillouin zone. This situation
applies to GaAs and other III-V and II-VI semiconductors. The phonon wave vectors are
rather small. An electron with energy ck ~ kBT has a small wave vector k. If it emits or
absorbs a phonon and goes to energy ck+q ± Jjw q, then Ik + ql is a small wave vector. The
consequence is that q is also a small wave vector. At small wave vectors, it is a good
approximation to represent acoustical phonons by the Debye model (Wq = csq) and optical
phonons by an Einstein model (Wq = wo). The results for acoustical phonons are derived here.
The case of optical phonons is assigned as problems, although the derivation is similar to that
for Frohlich polarons in the last section.
The wave vector integrals can be done analytically since the Debye approximation is
accurate. Define
(8.155)
(8.156)
(8.157)
(8.158)
The lower limit WI can be set equal to zero since the factor of w2 makes this contribution
negligible. The lifetime is
'tp~k) = 2rc Jdw{r?F(-)(k, w)[nB(w) + 1] + 2F(+)(k, w)nB(w)}

rt
There are two interesting limits to this expression. The first is at zero temperature where the
phonon occupation numbers nB(w) = o. The particle is assumed to be energetic so that
k » ks . In this case the answer is
1
'tik) = Ypck
(8.159)
4 D 2m*2
(8.160)
Yp = 3rc Jj4pc s
The lifetime depends upon the value of kinetic energy. The second case is at high temperature,
where nB ~ kBT IJjw so the result is proportional to temperature. In this case the natural
quantity is the m:tp [lp(k) = vk'tp(k)]
Vk
(8.161)
'tp(k) lp
D2kBTm*2
= (8.162)
lp rcJj4pc;
Another m:tp for the electron is called let>. The symbol is the distance over which the electron travels before it breaks its phase coherence.
For electrons in a pure semiconductor, where there are no electron-electron interactions, then
let> = lp. The distance for phase coherence is given by the m:tp for scattering by phonons. The
phonon can carry away an arbitrary amount of phase, so such scattering does change the
phase of the electron in a random fashion. In contrast, the scattering by impurities does not
change the phase coherence. Impurity scattering changes the phase of the electron, but it
changes the phase of each electron by the same amount. So coherence is maintained in
scattering by impurities, while it is not maintained in the scattering by phonons. The impu-
rities are just part of the one-electron potential which guides the electron as it wanders
through the crystal. Electron-electron interactions also break phase coherence,
A calculation similar to finding ip can be done for the momentum relaxation
D2m*
16n2h2 pk 3cf ro41(1)[<(1)<(1)"
The results are at low temperature
(8.163)
(8.164)
and at high temperature
D2kBTm*2
(8.165)
ipt nh 4 pc; Ip
The results are the same at high temperatures for the transport and regular scattering mfP:
Ipt = Ip. This equality is expected for the following reason. At high temperatures, the wave
vector dependence of the interaction is effectively
IMi dq2 d
(8.166)
hro q 2pro~ 2 pc;
The interaction is independent of wave vector, so the scattering is isotropic. The angular
factor of cos(S) averages to zero for isotropic scattering, and the two mfP's are identical.
Some numerical results are presented in Table 8.1 for the high-temperature case using
T = 300 K. The factor of pc; = ClI , where ClI is the elastic constant for LA phonons. The
values for Ipt are several microns.
Also included are data for the polar scattering by optical phonons, which produces a
lifetime '0, which can be converted to a mfP using the thermal velocity VT = J2kB T /m*.
Energy relaxation in semiconductors is discussed using the formula (8.144)
At low temperature the phonon occupation number nB = 0 and the only process is phonon
emission. Assuming that the particle has enough energy to emit phonons, the above formula
gives
3
( dE) = IYm*k = _ vkEk (8.167)
dt 1tJi2p lpe
(~) = :k (~) = - ~: (8.168)
2IYm*3
= (8.169)
lpe 1tJi4p
The mfp path for energy relaxation lpe from phonons is quite long. Some values are shown in
Table 8.1. Typical distances are millimeters. Most semiconductor devices are much smaller
than these length scales, so that electrons do not achieve energy relaxation in traversing most
devices unless there are sufficient impurities to cause them to random walk this distance.
The other interesting case is for high temperature. Expand nB = kBT /ro q and find -!
( dE) = _21tJOlD droro{kBT [1X2FH _ 1X2F(+)] + 1 [1X2F(+) + 1X2F(+)]}
dt 0 Jiro 2
The last term gives the same mfp lpe as was found for zero temperature. The first term, which
is proportional to kBT, is nonzero when the limits to the frequency integrals are taken to be
2csCk ± ks), so the integral is
J
2CS(k-kS )
droro2 -
J2Cs(k+ks)
droro2 = (8.170)
-~(2cs)3[(k + ks)3 - (k - ks)3] = -16~m*c: (8.171)
( dE) = -J.-[Ek - 2kB T] (8.172)

dx lpe
The energy relaxation has the same formula lpe for the mfp at room temperature as at low
temperature. The only change at higher temperature is that the energy Ek relaxes to the value
2kB T, which was zero at zero temperature.
TABLE 8.1 Electron mfp data at T = 300 K. The top lines have material constants. The second group of lines has
estimated mpf from optical phonon scattering. The last two lines are from defonnation potential scattering by LA
phonons
Units GaAs InP InAs InSb
m* me 0.064 0.078 0.027 0.013

EO 12.8 12.5 15.15 17.7
Eoo 10.9 9.5 12.25 15.7
ell GPa 118 101 83 66.7
p g/cm3 5.32 4.81 5.67 5.78
1zrow meV 35.4 42.8 29.6 23.6
D eV 8.0 6.4 6.0 14.0
IX 0.067 0.125 0.055 0.020

'to ps 0.80 0.51 0.85 2.09
VT km/s 378 341 581 837
vT'to !lm 0.30 0.18 0.50 1.75
Ipt !lm 2.0 1.8 13.9 8.8

Ipe mm 3.2 2.5 80 134
8.3.3. Temperature Relaxation

The relaxation time for temperature is required when the electron temperature Te and the
phonon temperature Tp are different. The electron-phonon interaction will allow energy
exchange between these two systems, and gradually bring them to the same temperature. The
rate at which they equilibrate defines the temperature relaxation time. The result for metals
was introduced by Allen (1987). The same quantity is important in semiconductors. There are
several situations where an experimentalist might encounter Tp =1= Te' Since the electrons
absorb electromagnetic radiation, an intense laser pulse could raise Te above Tp. Similarly, a
strong dc electric field could accelerate electrons to have an average kinetic energy well above
the phonon temperature, and then the two systems would mutually relax to the same
temperature.
The starting expression for this calculation is the collision term (8.137) in the Boltzmann
equation. The expression is simplified with the following assumptions:
• The electrons are in thermal equilibrium among themselves with an effective

temperature Te , Pe = l/kB Te and distribution function f(p) = nF(~p). This relative
equilibrium is maintained by rapid electron-electron scattering.
• The phonons are in thermal equilibrium among themselves with an effective
temperature Tp' Pp = l/kBTp which is maintained by rapid anharmonic phonon-
phonon scattering. They are described by a distribution function nB(ro q ).
• The electrons and phonons exchange energy according to (8.137).
Always keep in mind that the boson distribution function nB(ro) is at a different temperature
than the electron distribution functions nF(E), nF(E'). The expression can be further simplified
by using the function rx2F( ro) to express the integrals over the phonon states
(a~~E») ep = 21t Jdrorx2F(ro) JdE'

x {8(E - E' - ro)[nF(E)[l - nF(E')][nB(ro) + 1] - nF(E')nB(ro)[l - nF(E)]]
+ 8(E - E' + ro)[nF(E)nB(ro)[l - nF(E')] - nF(E')(nB(ro) + 1)[1 - nF(E)]]}
(8.173)
where E = Ep' E' = Ep+q' Next, the change in the internal energy is calculated. The change in
energy is accomplished by inserting under in the integral the factor of E' - E = ±hro.
e~;E») ep = -21th JdroriF(ro)ro JdE'(8.174)

x {8(E - E' - ro)[nF(E)[l - nF(E')][nB(ro) + 1] - nF(E')nB(ro)[l - nF(E»))
- 8(E - E' + ro)[nF(E)nB(ro)[l - nF(E')] - nF(E')(nB(ro) + 1)[1 - nF(E)]]}
= -21th Jdrorx2F(ro)ro
x {nF(E)[l - nF(E - ro)][nB(ro) + 1] - nF(E - ro)nB(ro)[l - nF(E)]
(8.175)
In order to average over the entire system of electrons, it is necessary to also integrate over dE.
If N(E) is the density of states for the electron system, the energy change in the electron
system is
dU = C dTe = Jdf.N(E)(aE(E») (8.176)

dt dt at ep
where C is the heat capacity. The integrals over dE all converge within a thermal energy of the
chemical potential. On this small energy scale, the density of states N(E) can be taken to be a
constant N(O), where the zero of energy is the chemical potential. Then all of the integrals
over dE have the typical form
(8.177)
The subscript "e" is added to the boson occupation function nBiO) to emphasize that it is
evaluated at the electron temperature Te' The above result can be derived by changing vari-
ables of integration to s = exp(BeE), ds = SBedE, t = exp(BeO) which changes the integral to
1(0) = kBTe Joo ( + 1)(ds + teem

)=
kBTe [s + IJoo = O)nBe(O)
~
In -
o s s - 1 +t s 0
The other integrals over dE are evaluated in a similar fashion. The result for the energy
relaxation is
dT fOOD
C dte = -4nI;2N(0) 0 dO)0)20(2F(0)[nBiO) - nBiO)] (8.178)
The final formula is quite simple. The right-hand side of this equation obviously vanishes in
equilbrium when Te = Tp. Another simplification occurs because the heat capacity is
proportional to the density of states C = n2k~TN(0)/3 which simplifies the expression to
dT 121;2 JmD
-de = - - 2k dO)0)20(2F(0)[nBe(0) - nBiO)] (8.179)
t n BTe 0
The only factor which relates to the particular solid is 0(2F(0).
There are two interesting limits when evaluating this expression. The first is at very low
temperature, Te,p « e, where the Debye temperature is e. Then the integral over 0) has its
main contribution at small values of 0). The small values of frequency come from sound
waves, where it is suitable to use the Debye approximation. The evaluation of 0(2F( 0) in this
limit is identical to the semiconductor case, which gives that 0(2F = y0)2 where y is a
constant. The frequency limits can be extended to infinity, which gives the expression
dTe = _ r[T5 _ T 5] (8.180)

dt e p
r _ 12kly (8.181)
- nT 1;3 14
e
14 = JOO dxx41 = 4!~(5) (8.182)

o eX-
The relaxation obeys a T 5 law, which means very little heat is exchanged between electrons
and phonons at low temperatures.
The other interesting limit is at room temperature, or at least above the Debye
temperature. Then the boson occupation factors can be expanded nB ~ kBT fliro which gives
the simple result
(8.183)
(8.184)
At high temperature the temperature relaxation is determined by the temperature difference

'OT = Te - Tp ' as well as by the function A(ro2 }. The latter function also determines the
energy relaxation.
The temperature relaxation is measured by lasers using pulse-probe techniques. The first
laser pulse is absorbed by the electrons, which thermally excites them, which causes their
temperature to differ from the phonons. The time-delayed probe measurement determines
how some property, such as the refractive index, varies with the time interval after the initial
pulse. An observation of an exponential relaxation curve is a measurement of tpT'
The above result is valid for a metal. A similar derivation can be done for a semi-
conductor, which is assigned in the problems. The same calculation for scattering of electrons
in a semiconductor, by optical phonons, can be done in a interesting and rigorous way. For
optical phonon scattering in a semiconductor, all of the phonons have the same energy liroo so
there is no need for an integral over frequency. In fact, just set
2 Aroo
IX F(ro) = 2'O(ro - roo) (8.185)
where A is the dimensionless coupling constant. Starting from (8.173) gives
at
( af(E)) = -TtArooUf(E)(No + 1) - Nof(E - roO)]0(E - roo)
+f(E)No - feE + roo)(No + I)} (8.186)
The states feE + [roo) can be regarded as a ladder of energy states, where each level is
separated by roo. The above equation can be solved exactly by matrix methods. Define
fi =feE + [roo) where the energy E is now limited to the range 0 < E < roo. The vectorj has
fi as its elements. The above equation can now be cast into a matrix equation, where No is
the occupation number of the optical phonons.
d - 1 -
-f=--A·f (8.187)
dt toT
A = Co-:0 -(No + 1)
2No + 1
-No
0
-(No + 1)
2No + 1
""")
... (8.188)
1
- = TtAroo (8.189)
toT
The matrix A has tridiagonal form. Only three rows are shown, but its dimensionality is
infinite. All row except the first have elements: -No, 2No + 1, -(No + 1) to the left of the
diagonal, on the diagonal, and to the right of the diagonal. If the matrix ~ has eigenvalues Ej
and eigenfunctions \)Ij then the solution is
J(t) = I: aj\)lje-E/ltoT (8.190)

j
where the coefficients aj are determined by the initial conditions. The equilibrium distribution
is given by Eo = 0, \)10 = exp( -1~0)0), where ~ is determined by the phonon temperature. The
system relaxes towards the equilibrium distribution, which is why its eigenvalue has to be
zero. The other eigenvalues and eigenvectors are given by Mahan (1985).
8.4. ELECTRON-PHONON INTERACTIONS IN METALS
8.4.1. Force-Force Correlation Function

In pure metals the electrical resistivity has two components. There is usually a constant
resistivity from electron scattering by impurities, which is the largest part of the resistance at
small temperatures. There is also a resistivity from electron scattering by phonons, which is
temperature dependent and becomes large at high temperature. "Matthiessen's rule" (1862) is
that these two contributions to the resistance are additive. It should be regarded as a rule of
thumb, rather than an ironclad rule. There are enough "deviations from Matthiessen's rule" to
make the abbreviation DMR a familiar acronym (Bass, 1982).
Electron scattering by acoustical phonons presents a hard problem in transport theory.
The scattering is slightly inelastic. This problem is solved using neither the elastic scattering
theory of Sec. 8.1 nor the inelastic scattering theory of Sec. 8.2. Instead an integral equation
must be derived and solved for the energy dependence of the scattering process. The slightly
inelastic nature of the scattering process makes this calculation much harder than the previous
cases.
Two methods for obtaining the electrical conductivity are emphasized in this book. One
uses equilibrium methods and evaluates the Kubo formula for the current-current correlation
function. The resistivity from phonons will be found using this method, which follows the
original derivation by Holstein (1964). The second method utilizes the quantum Boltzmann
equation (QBE), which is a nonequilibrium theory. Mahan and Hansch (1983) used the QBE
to derive the Holstein formula. Both of these derivations are complicated. They end by
deriving the same integral equation for the scattering function, which must be solved by
further work. Their virtue is that they are formally exact starting points, although approx-
imations are made in obtaining the solution.
Other methods for obtaining the resistivity have been proposed, partly to avoid all of the
work associated with the exact methods. These other methods are approximate. However,
often the theories are both simple and accurate, which make them useful approximations. One
of them is the force-force correlation function. If F(t) is the fluctuating force that acts on the
electron, then define R(iO)) as the force-force correlation function:
(8.191)
(8.192)
Sec. 8.4 • Electron-Phonon Interactions in Metals 535
This formula is just the quantum analogy of the Nyqvist theorem (Kubo et al., 1985). After
calculating this correlation function, the retarded function is obtained by letting ioo --+ 00 + iO.
The resistance p is found by dividing by 00 and taking the limit of 00 --+ O. For example,
assume that the force on the electron has two terms: F; from impurities and Fph from phonons.
If they are uncorrelated, then the correlation function has no cross terms. Symbolically write
(8.193)
In this case the resistivities from impurities and phonons are additive, in agreement with
Matthiessen's rule.
As an example, the resistivity is calculated from impurity scattering. The potential
energy of the electron scattering from the impurities at R; is discussed in Sec. 4.1.5
VCr) =L Ve;(r - RJ = -1 L .
V(q) exp[zq . (r - RJJ (8.194)
; v iq
The force F is the gradient of the potential. The factor of exp(iq· r) can also be written as the
electron density operator p(q). The factor of exp( -iq' RJ can be written as the impurity
density operator p;( -q)
F(r) = - -i L qV(q) exp[iq· (r - RJJ = - -i L qV(q)p(q)p;( -q)

v~ v~
The next step is to evaluate the force-force correlation function R(ioo). In correlating F with
itself, there are two separate factors. One is (p;( -q)Pi( -q')), which equals the number of
impurities Ni if q = -q'. The other is the electron density-density correlation function, which
is given exactly in terms of the inverse dielectric function
. Ni q2
R;(zOO)=--32L-V(q)
2[ ( 1 . )-1 ] (8.195)
v q Vq E q, zoo
The next step is to take the imaginary part ofthe retarded function. The only retarded function
on the right-hand side of the above equation is the inverse dielectric function E = E\ + iE2' Its
imaginary part is -E2 /(ET + E~). At low frequency E2 = 2oo~m2/q3. Dividing by 00 eliminates
the factor of frequency. The formula for the resistivity from impurity scattering is
ni m2 J
d 3q 1 V(q) 12
P = 6nn6e2 (2n)3 q E(q)
(8.196)
This formula is the exact result for the zero-temperature resistivity from impurity scattering,
when the scattering is calculated in the second Born approximation. If V(q)/E(q) is replaced
by the Tmatrix for scattering, then it is the exact result, period. It is the formula
p = m/(noe2'tt), where 'tt is defined in (8.25). It even includes the factor of (1 - cos e'),
although this assertion is not immediately obvious. One has to perform the angular integral in
(8.25), which eliminates the delta function, in order to show its equivalence with the above
formula for the resistivity.
The force-force correlation function gives the right resistivity for impurity scattering. No
vertex correction or integral equation was needed in the derivation. The ease of derivation has
made this approach popular.
Several caveats are needed. One is that impurity scattering is the only known example
where the force-force correlation function gives the correct answer. In other cases it give an
approximate answer. The second caveat is that the right answer is obtained by a wrong
derivation. The derivation contains two important limits. One is setting the volume v -+ 00,
while the second is 0) -+ O. The above answer is obtained by taking these limits in the wrong
order. If they had been done correctly, in the right order, a different answer is obtained. Using
the right order in evaluating the force-force correlation function gives a zero result as 0) -+ O.
These points are discussed by Argyres and Sigel (1974), Huberman and Chester (1975), Kubo
et al. (1985), and Fishman (1989).
The force-force correlation function may also be evaluated for the electron scattering by
phonons. The result is
peT) f
= C' ~ qd3 ql W(q)12(~" . q)2 [ - dn:~O))J (8.197)
c' = 3livo (8.198)

16Me2 v}kj.
Here W(q) is the screened electron-ion interaction and Vo is the unit cell volume. This
formula was first derived by Ziman (1960) as a variational solution to the Boltzmann equa-
tion. It is the formula that is most often evaluated when calculating the temperature depen-
dence of the resistivity of metals. Figure 8.8 shows a theoretical calculation of Dynes and
Carbotte (1968) compared with experiments for Na and K. An important feature of these
calculations is numerically integrating over the Brillouin zone for all the phonon states, ~hile
employing accurate values for the phonon frequencies O),,(q) and polarization vectors 1;.
K 0
Eu
~
E
2
.. ~ .8
o
.6
.4
.2
40 60 80 100 120 140 160

T(OK)
FIGURE 8.8 Resistivity as a function of temperature for Na and K. The solid line is experiment, and points are
theory. Source: Dynes and Carbotte (1968) (used with permission).
8.4.2. Kubo Formula

A rigorous theory of the dc electrical conductivity is calculated for the scattering of
electrons by all phonons: acoustical and optical. The starting point is the Kubo formula, and
the derivation follows the original one by Holstein (1964). His theory sums the ladder
diagrams for phonons and reduces the vertex function to an integral equation which is solved
numerically. So far a solution is available only for a spherical Fermi surface, so that the result
is the Kubo formula analogy of (8.198). The results are expressed in terms of the McMillan
function rx2F(ro) and a similar function rx;F(ro), which is used in transport theory.
The goal is to evaluate the current-current correlation function in the presence of the
electron-phonon interaction. This correlation function can always be expressed as a product
of two Green's functions and the vertex function. The Green's function <:'9(p, ip) used here
represents fully interacting particles, with a self-energy found from the electron-phonon
interaction plus any additional interactions of interest. The Feynman diagram for the corre-
lation function is shown in Fig. 8.9 where the vertex function is put only at one end of the
bubble in order not to overcount the vertex terms:
n(iro) = -~J~ d't"eiOln~(T~j('t)·j(O»)

3v 0
= -3e22 L
m v pp' 0"0"'
p.p'J~0 d't"eiOln~(T~C~O"('t")CpO"('t")C~,O",(O)Cp'O"'(O») (8.199)
n(iro) = 32e22
m
J
(2n)
d 3P3 ~I-' Lip <:'9(p, ip)<:'9(p, ip + iro)p· r(p; ip, ip + iro) (8.200)
The dc conductivity is found by the same steps used in Secs. 8.1 and 8.2. One evaluates the
correlation function for values of iro and analytically continues iro -+ ro + io to find the
retarded function. The dc conductivity is the imaginary part of the retarded function divided
by ro, in the limit where ro -+ O.
The vertex function r(p; ip, ip + iro) is evaluated below. It depends on both the
frequency variables ip and iro, which is written in the combination (ip, ip + iro). The two
arguments ip and ip + iro come from the electron Green's functions which have the same
frequency arguments. In a homogeneous electron gas, the vector vertex function
r(p; ip, ip + iro) must point in the vector direction p, although in real metals the crystalline
potential defines other possible directions. However, the approximation of treating the Fermi
surface as strictly spherical is equivalent to neglecting crystal directions, so assume r points
ip
FIGURE 8.9
in the direction p. It is convenient to introduce the scalar function yep; ip, ip + im), which is
the amplitude of the vector vertex function:
r(p; ip, ip + im) = py(p; ip, ip + im) (8.201)
.
n(lm) 2£?-2
= -3
m
fd 3p 2 .
- - 3 P S(p, 1m)
(2n)
(8.202)
1
S(p, im) = A L 'fJ(p, ip)'fJ(p, ip + im)y(p; ip, ip + im) (8.203)
I-' lp
The scalar function y is not the same as the scalar vertex function in (8.81).
The next step is to evaluate the summation over Matsubara frequencies ip to obtain
S(p, im). To this end, construct the usual contour integral which has cuts along the axes where
ip --+ real and also ip + im --+ real. These series of algebraic operations are the same as used
to derive (8.64) for impurity scattering. That earlier result can be used here by identifYing
P=GGy
x [Gadv(p, E)Gret(p, E + m)y(p; E - i8, E + m + i8)
- Gret(p, E)Gret(p, E + m)y(p; E + i8, E + m + i8)] } (8.204)
and the conductivity is
(j = 2£?- f d 3p p2fOO dE [_ dnF(E)]

3m 2 (2n)3 -00 2n dE
x {I Gret(p, E)12y(p, E - i8, E + i8) - Re[Gret(p, E)2y(p, E + i8, E + i8)]} (8.205)
Equation (8.205) is exact for the exact vertex function. It is expressed in terms of the two
functions yep; E - i8, E + i8) and yep; E + i8, E + i8). These two functions are expected to
be quite different, as they were in Sec. 8.1 for impurity scattering. The function
yep; E + i8, E + i8) could be obtained from a Ward identity. That is true here, as first shown
for the electron-phonon system in metals by Englesberg and Schrieffer (1963). This Ward
identity can be expressed in terms of r(p, E + i8, E + i8) or the scalar yep, E + i8, E + i8):
rep; E + i8, E + m + i8) = P + mVpL(p, E) (8.206)
yep; E + i8, E + m + i8) = I +T

a L(p, E) (8.207)
a~p
Gret(p, E)2y(p; E + i8, E + m + i8) = T

a Gret(p, E) (8.208)
a~p
For metals, the electron-phonon system has the feature that the self-energy function L(p, E) is
not very p dependent, and the derivative of L(p, E) with respect to ~p is small:
dLld~p ~ LIEF « 1. A good approximation is to set yep, E + i8, E + i8) = 1.
There is another result which is even stronger, since it applies for all values of E near the
Fermi energy. Since the self-energy ~(p, E) does not have significant p dependence near
p '" kF, call it ~(kF' E) = Re[~(E)] - ireE) where the p notation is suppressed. The function
reE) = -Im[~(kF' E)]. The retarded and advanced Green's function have their important
dependence on p through the kinetic energy term, ~ == ~p. Define neE) = E - Re[~(E)]:
1 1
G t(P, E) = = (8.209)
re E - ~ - Re[~(E)] + ir(E) neE) - ~ + ireE)
1 1
Gad (p, E) = = -------,-- (8.210)
v E - ~ - Re[~(E)] - ireE) neE) - ~ - ir(E)
Similarly, the vertex function yep; E - is, E + (0 + is) is assumed to not have a significant
dependence upon p except on the order of kF . This assumption will be justified later. The
second term in brackets in (8.205) can be neglected; it vanishes by doing the kinetic energy
integration. By neglecting terms of order O(~/ EF ) then
d 3pp2 = 4rcp4 dp = 4rcmk~d~[1 + O(~/EF)] = 12rc3mnod~ (8.211)
where the electron density is no = k}/3rc2 • When the kinetic energy integral is evaluated, the
only ~ variation is in the Green's functions. Of the three combinations which occur, only one
makes a nonzero contribution:
J~oo d~Gret(P, E)Gret(P' E') = 0 (8.212)
oo
J d~Gret(P' E)Gadv(P, E') = neE) - neE') +2rcii[reE) + reE')] (8.213)
-00
J~oo d~Gadv(P' E)Gadv(P, E') = 0 (8.214)
Each integral is evaluated by closing the contour at infinity. The two integrals which vanish
have both their poles in the same half plane (upper or lower), so the integration contour can be
chosen to avoid them both, which encircles no poles and hence gives zero. The integral over
the combination GadvGret has one pole in each plane. Closing the contour always picks up one
pole whose residue produces the result. By using these integration results, the dc conductivity
(8.205) becomes
0" = ilno
2m
Joo dE A(E)
-00
[d ]
reE) - dE nF(E) (8.215)
A(E) = y(kF' E - is, E + is) (8.216)

reE) = -2 Im[~(kF' E)] (8.217)
Equation (8.215) is the final result of the formal derivation. There only remains the evaluation
of the vertex function A(E) and the imaginary self-energy reE). If the vertex function were
absent (A = 1), then the evaluation would be easy. The imaginary self-energy r(E) has been
evaluated for many metals, and a result was given in Fig. 7.14 for Pb. The quantity
I/2r = 'teE), where 'teE) is the relaxation time defined as the average time between scattering
events. The result (8.215) when A = 1, is just the average of the relaxation time over the
thermally smeared Fermi distribution. Of course, the earlier solution for impurity scattering
showed that the neglect of A( E) is a serious error. The vertex function A(E) serves the
important role of weighting the scattering events and favoring those at high momentum
transfer. For impurity scattering, A = r /r t or A/r = l/r t , where r t is the scattering rate
which contains the equivalent of (1 - cos 9').
The vertex equation for phonon scattering is now solved to derive t . Since the phonon r
scattering is inelastic, the results are not identical to those for impurity scattering. In summing
the ladder diagrams for the vertex function, however, the vertex contributions are found to be
important and significantly different from unity. Migdal's (195S) theorem, which asserts that
vertex terms are unimportant, is contradicted.
The vertex function rep, ip, ip + iro) is calculated by solving the integral equation
rep, ip, ip + iro) = p +~ L

I-' iq.A
J(21t)
d 3q3 MA(q)2~(q, iq)~(p + q, ip + iq)
X ~(p + q, ip + iq + iro)r(p + q, ip + iq, ip + iq + iro) (S.21S)
This vertex sums the ladder diagrams for phonons. It is illustrated in Fig. S.10. Iteration of
Eq. (S.21S) produces a series in which each additional term has one more ladder diagram. The
solution to the integral equation produces an expression which contains all terms with any
number of phonon ladder diagrams. This solution is not an exact evaluation of the vertex
function rep, ip, ip + iro), since other vertex contributions occur which are not ladders but
have the phonon lines crossed. One expects these terms to be smaller, but detailed calcula-
tions are lacking, so this is only a supposition.
It is unfortunate that the vertex function is a vector. The scalar vertex function obtained
by replacing p by 1 in Eq. (S.21S) is easily obtained from a Ward identity:
rep, ip, ip + iro) = 1+~ L

I-' iq.A
J(21t)
d 3q3 MA(qf~(q, iq)~(p + q, ip + iq)
X ~(p + q, ip + iq + iro)rep + q, ip + iq, ip + iq + iro)

1
= 1 - -;- [~(p, ip
lro
+ iro) - ~(p, ip)] (S.219)
The Ward identity is not helpful for our problem with a vector vertex function. The integral
equation must be attacked and solved.
Define the scalar function yep, ip, ip + iro) as the scalar amplitude of the vector vertex
function r = py. It obeys the integral equation
L J-d- 3
q p' (p + q)
... 1 3
yep, IP, Ip + /ffi) = 1 + A
2
MA(q) ~(q) 2
I-' iq.A (21t) P
X ~(p + q)~(P + q + iro)r(p + q, ip + iq, ip + iq + iro)
.£iP+iW
IW
= +
'
-;(t
ip+iq+iw
Iq
I ip+iw
r r I.
ip ip ,/ Ip
ip+iq
FIGURE 8.10
This equation is not the same one which is obeyed by the scalar vertex function in (8.219),
and these two functions are quite different.
First do the integrals over angles and wave vector. First write d 3q = 21tq2dqdv, where
e
v = cos is the angle between P and q. The angle variable is changed to PI, defined as
PI = (p +qi =p2 +q2 +2pqv (8.220)
dv =PldPI (8.211)
pq
Jd3 q = 21t Joo qdqJlP+q1p dpI = 21t Joo qdqJ~P+q d~1

P 0 IP-ql
I
Vo 0 ~p_q
(8.222)
Next uncouple the limits of integration. Only electrons at the Fermi surface contribute, so that
values of ~I = Epl - 1.1 ~ 0 and ~P ~ 0 are important in the integration process. Approximate
vp = vF as the Fermi velocity. Similarly, the integral qdq is understood to be over the spherical
Fermi surface from one point p to all other points PI' where P and PI both have magnitude kF :
(8.223)
The limits on the d~1 integral are extended between ±oo, since most of the integrand has
large q values, where the actual limits on ~I are from a very negative number to a very
positive one. Since the main contribution is in the region ~I ~ 0, this error is small. Consider
the other angular factors in the integrand:
P • (p + q) = 1 + qv = 1 + _1_ rl _p2 _ q2] (8.224)

p2 P 2p21
~ l_L+o(h ~p) (8.225)

2k~ EF' EF
The terms ~dEF' ~pIEF are neglected, since it is expected that ~I and ~p will be small--on
the order of a Debye energy. On the other hand, the factor q2/2k}. need not be small, since the
integration over phonon states has a significant contribution from high values of q near the
edge of the Brillouin zone. This term must be retained. The difference in treating the factor
(Pr- p2)1k}., which is neglected, and q2 Ik~, which is retained, is that the former enters the
average over electron states and the latter over phonon states. The integrals over the phonon
wave vector can be expressed in terms of the function rrF(oo), which was introduced in Sec.
7.4:
C(
2
F(oo) = -1
(21t)
VF
L
A.
J
d 3q 2
- 3 MA.(q) 0[00
- OOA.(q)] (8.226)
2 1 d 3q 2 q2
C(tF(OO) = - L--3 MA.(q) 2k2 0[00 - OOA.(q)] (8.227)
VF A. (21t) F
The first of these is just the McMillan function, which was defined in Sec. 7.4. Since the
Fermi surface has been assumed to be spherical, C(2F has the same value at each point on the
surface, and the P SUbscript is omitted. The other form of coupling C(TF is called the
"transport form of alpha-squared-F," which was introduced by Allen (1971). It differs from
the McMillan form by having the additional factor of q2/2k}. in the integrand, which gives
more weight to the scattering processes at large wave vector. The factor of q2/2k],. is identical
to (1 - cos 9') when the scattering is elastic.
After completing all these angular and wave vector integrations, the vertex function
yep, ip, ip + iro) is found to be not very dependent onp. The only variation is as a function of
p, which can be set equal to kF with an error of only ~pIEF. Define y(ip, ip + iro) =
y(kF' ip, ip + iro) and arrive at the equations:
y(ip, ip + iro) = 1 + [D du[0(2F(u) - iX;F(u)]

x J~oo d~18(~I' u; ip, ip + iro) (8.228)
1 2u
8(~, u; ip, ip + iro) = A L . 2 y(ip + iq, ip + iq + iro)
..... iq (lq) - u2
X ~(~, ip + iq)~(~, ip + iq + iro) (8.229)
This integral equation is not too difficult to solve. The basic approximation has been to
decouple the integrations over dq and d~I' which permits all the phonon information to be
collected into the functions iX2F, iX;F. The primary assumption in this decoupling is that the
Fermi degeneracy EF is very much larger than other energies such as kBT or hOlD. The present
integral equation is actually much easier to solve than the one for polarons in Sec. 8.2, since
there the integration variables cannot be accurately decoupled in the same way.
The preceding is the basic integral equation which needs to be solved for the vertex
function. The ~ dependence of 8 is only in the Green's functions ~(~, ip + iq) and
~(~, ip + iq + iro). This integral is done later, using the result (8.213) that only the integral
over the pair GretGadv == IGret l2 is nonzero. The factor 2ul[(iqi - u2] is the phonon Green's
function for a phonon of energy u. The next step in the derivation is to do the summation over
Matsubara frequency iq, which is done in the usual way by constructing a contour integral.
f dz 2u
-2·-2--2 nB(z)y(z + ip, z + ip + iro)G(z + ip)G(z + ip + iro)
1tlZ - U
The contour in integration is a circle at infinity. The integrand has poles from the phonon
Green's functions at z = ±u and cuts along the axes where the electron Green's functions are
real, z = -ip + E' and z = -ip - iro + E'. The contour integrals from these three contribu-
tions are
8 = 8 1 +82 +83 (8.230)

8 1 = nB(u)y(ip + u, ip + iro + u)G(ip + u)G(ip + iro + u)
+ [nB(u) + l]y(ip - u, ip + iro - u)G(ip - u)G(ip + iro - u) (8.231)
82 = J-2.
dE' 2u
. 2
(E' - lp) - u
1tl
. .
G(E' + lro)[y(E' + iO, E' + lro)Gret(E')
2
- y(E' - iO, E' + iro)Gadv(E')] (8.232)
J
dE'
82 = -2. .
2u
. 2
1tl (E' - lp - lro) - u
2 G(E' - iro)[y(E' - iro, E' + iO)Gret(E')
- y(E' - iro, E' - iO)GadvCE')] (8.233)

The next step is to perform the analytic continuations. The final result should be A(E) =
Y(E - i?i, E + i?i). First set ip -+ E - i?i. The next step is to set ip + iro -+ E + iro -+
E + ro + i?i -+ E + i?i, since ro -+ O. The Green's functions become advanced or retarded
according to the side of the cut, so this analytical continuation produces
S = nB(u)A(E + u)IGret(~' E + u)1 2 + [nB(u) + l]A(E - u)IGrei~, E - u)1 2
+ J dE'
21t·nF
I
(') [G (~ 2 ') A(E') (2U
E I ret ':>' E I E - E - .~)2
(' lu - U
2U)
.~)2
2 - (E, - E + lu
2 - U
+ O(G;"t, G;dV)] (8.234)
The next step is to do the integration over d~I' According to (8.213), this integral eliminates
all combinations of the Green's functions except GretGadv == IGret12, which removes the terms
O(G;.,t, ~dv)' Also note that the factor below is the phonon spectral function
2u 2u
------;;;2-- = 21t[?i(E' - E - u) - ?i(E' - E + u)]
(E' - E - i?ii - u2 +
(E' - E i?i) - u2
which eliminates the integral over dE'. These manipulations provide the final form of the
integral equation for A(E). It was first derived by Holstein (1964), although his result is
modified by expressing it in the rx2F formalism:
A(E) = 1 + 1t [D du[rx2F(u) - rx;F(u)]

A(E + u) A(E - U)]
x [ [nB(u) + nF(E + u)] r(E + u) + [nB(u) + 1 - nF(E - u)] r(E _ u) (8.235)
cr = not? JOO dE A(E) [_ dnF(E)]

2m -00 r(E) dE
The integral (8.215) for the dc conductivity was rewritten in order to present the two
important results together. The integral equation for the vertex function A(E) must be solved,
and the solution is used in the integral over E for the conductivity. Recall that A(E) is a real
function, so the integral equation is not complicated. The form of the equations suggests the
unknown quantity is actually i(E) = 2A/r, which might be called the effective relaxation
time for transport.
The energy uncertainty r(E) of the electron from the electron-phonon interaction was
given previously in (7.305). The equivalent result for the transport kernel is:
(8.236)
If the self-energy function r and the vertex function A(E) were both constants and inde-
pendent of E, the solution of the vertex equation (8.235) would be easy:
A
A = 1 +-[r -
r
rt ] (8.237)
r
(8.238)
This model gives Ajr = 1jrt , so that the conductivity integral has only the transport form of
the relaxation time. The earlier solution to impurity scattering also assumed that Ajr = ljr t .
For impurity scattering the functions reE) and A(E) are usually insensitive to E near E ~ 0; the
exception is when the impurity has a scattering resonance near the Fermi surface or if the
density of states is not smooth. However, for the electron-phonon interaction in metals, it is
not a good approximation to treat reE), rl(E) or A(E) as constants. The calculated results in
Sec. 7.4, shown in Fig. 7.14 for Pb, illustrate that reE) has substantial energy variations near
the Fermi energy. The vertex function does also.
Takegahara and Wang (1977) evaluated (8.235) and (8.215) for metallic rubidium and
cesium. Their results are shown in Fig. 8.11. In each case the solid line is calculated assuming
that the ratio r jr t is a constant, while the dashed line is calculated by solving (8.235) for the
E dependence of A(E). The latter curve is in very good agreement with the experiments, which
are indicated by the points. The differences between the solid and dashed curves are similar to
the differences between the Ziman formula (8.198) and the Holstein formula (8.235). Note
that there is no region with a well-defined r5 law for the resistivity in these metals.
1.0 1.0
Q.0
0.8 0.8
"
Q. 0.6 0.6
10- 7 0.4
10- 7
10- 8
-E
u
C 10- 9
Q.
10-'O~~~~~~-L~~ ____~~-LLU~__~WW~
1 50 100 5 10 50 100
r (K)
FIGURE 8.11 Calculated constant volume phonon limited electrical resistivity for (a) Rb and (b) Cs. Solid lines are
calculations that treat A.(e) as a constant. Dashed lines are full solution to integral equation for A.(e). Points are
experimental values. Source: Takegahara and Wang (1977) (used with pennission).
8.4.3. Mass Enhancement

The electron-phonon mass enhancement factor A was introduced in Sec. 7.4. It is from
the real part of the electron self-energy due to the electron-phonon interaction. Since this self-
energy was found to be energy dependent but not very wave vector dependent, the electron
effective mass m* is approximated by
a
A(O) = - aO) ,£(kF' 0) (8.239)
m* 1
-;;; = 1 + A(O) = Z(O) (8.240)
The mass enhancement factor is also related to the quasiparticle renormalization factor Z(O).
The values of these quantities at the Fermi energy 0) = 0 are A and ZF' The mass renor-
malization factor A can also be expressed as an average over the Fermi surface of a weighted
average over the phonon density of states. This definition is given in Sec. 7.4. Values of A in
real metals range from 0.1 to 3.
An important question is the role which A plays in the dc transport properties. The most
obvious approximation is to use the effective mass m* = m(1 + A) in transport formulas
whenever classical theory says to use m. This substitution would make sense, since m* is the
effective mass which governs the motion of electrons on the Fermi surface, and these are
involved in dc transport properties. However, this sensible procedure is wrong in most cases.
The important point is that the factor 1 + A can enter the final formulas in several ways. It
changes the effective mass, relaxation time, and quasiparticle renormalization factor ZF' The
formula for the transport coefficient will have a number of factors of 1 + A. Often they all
cancel, which is the case for the electrical conductivity.
Prange and Kadanoff(1964) investigated which transport measurements were influenced
by the electron-phonon mass enhancement factor 1 + A. They concluded that the enhance-
ment did affect the following measurements: specific heat, low-field cyclotron resonance, and
the amplitude of the deHaas-van Alphen effect. The following measurements are not affected:
dc electrical conductivity, thermoelectric power, thermal conductivity, the period of the
deHaas-van Alphen effect, spin susceptibility, and the electron tunneling rate. Their
conclusion on the thermoelectric power was challenged by Opsal et al. (1976), who detected
a dependence on 1 + A. The list of quantities which are affected is much shorter than the
list of quantities which are not affected. The usual case is that the transport property
is not influenced by the mass enhancement factor; see Grimvall (1981) for a further
discussion.
8.4.4. Thermoelectric Power

The thermoelectric power is a subject which is seldom discussed in most solid-state
courses, as the lecturer is busy treating subjects which are more fashionable. This tendency is
regrettable, since it is an important measurement. Experiments show that diverse behaviors
are found for simple metals, and even the sign of this quantity shows no regularity. The low-
temperature theory in metals is poorly understood, although there has been extensive work.
The thermoelectric power was introduced briefly in Sec. 3.8. It is a simple measurement,
at least conceptually. A conducting bar is insulated so electrical currents cannot exit from its
ends, and then a temperature difference I1T is maintained along the length of the bar. The two
ends of the bar are found to have a voltage difference L\ V which is proportional to L\T. The
constant of proportionality is the thermoelectric power S:
L\V 1 L(12)
S=- L\T=-YL(11) (8.241)
L(ll) = k~T J~ d-reimnt(Tj(-t)· j(O)} (8.242)

3,ooV 0 t
L(12) = k~T J~0 d-reimnt(Tj Q (-r)"j(O)}

3,oov t
(8.243)
The thermopower is defined theoretically as the ratio of two correlation functions. One is just
proportional to the dc electrical conductivity (J' = ~L(11), which has already been evaluated for
several models. The other correlation function has the argument of heat current jQ and the
electrical current j, where jQ = jE - (/l/e)j. Equation (8.241) differs by a factor of charge e
from its earlier version in Sec. 3.9, since earlier j meant the particle current, whereas now it is
the electrical current. They differ by the unit of charge, which causes the change in (8.241).
The correlation function L(12) is now evaluated for its dc value. Both correlation func-
tions are evaluated for values of ioo; then analytically continue ioo ~ 00 + iii. Finally take the
limit 00 ~ 0 of the imaginary part of the retarded function.
In the evaluation of most correlation functions, there is usually a leading term which
provides the dominant part of the answer. There are numerous small correction terms which
can usually be ignored. In calculating the correlation function L(12) for the thermoelectric
power, the dominant term vanishes, and one is left with obtaining all the numerous small
correction terms. This feature makes it difficult to obtain an accurate answer.
The heat current operator has many terms. The one which is expected to provide the
dominant term is from the kinetic energy of the electron: jQ = LVp~pCJerCper. This heat
current operator is used in most theories. Similarly, for the electrical current the operator is
j = e L VpCJerCper. The correlation function is called L(12a):
The important feature of this correlation function is the bracket containing four electron
operators. This type of operator sequence was encountered in earlier sections of this chapter.
It is evaluated as
() 1 2e
L 12a = - 2 - . -32 L ~k L k" r(k, ik, ik + ioo)r&(k)r&(k + ioo)
V~ 100 m k ik
where r(k, ik, ik + ioo) is the vector vertex function of the bubble diagram. This same vertex
function enters into the correlation function for the conductivity. Write it again as r = ky.
There is the same summation over Matsubara frequency ik which was done in the previous
section for the correlation function L(ll). The result was given in (8.204). The retarded fonn
of this correlation function is
2e J d 3 k 12 JOO dE
-2 [nF(E + (0)
(12a)
Im[Lret ] = - -
A3 2 - - 3 "~k - nF(E)]
pm (0 (2n) -00 n
x {Gadik, E)Gret(k, E + (O)y(k, E - iO, E + (0 + iO)

- Re[Gret(k, E)Gret(k, E + (O)y(k, E + iO, E + (0 + iO)]}
The next step is to take the limit (0 -+ 0, which causes the electron occupation factors to
become dnF(E)/dE. The integral over wave vector is changed to an integral over
k 2 d 3k = 12n3mnod~k:
Im[L(12a)] = eno JOO dE [_ dnF(E)]JOO d~~{lG (k, E)1 2

ret ~m -00 2n dE -00 ':> ret
X y(k, E - iO, E + iO) - Re[Grelk, Eiy(k, E + iO, E + iO)]} (8.244)
Again it is assumed that the retarded and advanced Green's functions are significantly
dependent only on ~ in their kinetic energy tenn and that the self-energy and vertex functions
have negligible dependence on ~. Then the following integrals are evaluated by a contour
integration in analogy with (8.213):
J~oo d~~Gret(~' Ei = in
J_ood~~IGret(~'
OO
E)I
2 nQ(E)
= r(E) (8.245)
J~oo d~~Gadv(~' Ei = -in

Two of the integrals equal ±in, which comes from the semicircle closing the contour at
infinity. These integrals give zero when taking the real part. The important contribution must
arise from the combination IGret(~' E)1 2 , since it has a nontrivial contribution from the integral
over d~:
Im[L(12a)] = eno fOO dE[- dnF(E)] Q(E)A(E) =0 (8.246)

ret 2~m -00 dE r(E)
This integral is zero after one evaluates the integral dE. It vanishes because the integrand is an
anti symmetric function of E. The quantities dnF(E)/dE, A(E), and r(E) are all symmetric
functions of E, while Q(E) is an anti symmetric function. The correlation function L(12a) is zero.
The integral vanishes because of the single power of ~ in the d~ integral. It makes the
single power of Q(E) in the integral over dE. The single power of ~ comes from the heat
current operator.
A nonzero result for the correlation function is obtained by repeating this derivation and
retaining all the correction tenns. One important tenn is from the argument of the wave vector
integration:
(8.247)
548 Chap, 8 • de Conductivities
The second term in parentheses makes a nonzero contribution to the thermopower. This term
is now evaluated. Kinetic energy integrals such as (8.245) must now be done, except there is a
factor of ~2 in the integrand which mUltiplies the Green's functions. The evaluation of the
integrand is tricky, since technically the integral diverges. At large values of~, then G ~ 1/~
and (~Gi ~ 1. The integrand does not fall off at large values of ~, and taking the limits to
±oo gives an infinite integral. Usually this problem is solved by ignoring it. The product
\Gretl 2 = A/2r ~ 1t8(~ - E)/r is replaced by a delta function for energy conservation. The
integral over (~Grei is set equal to zero:
f 2 ): 2 ~ 1tE2
d~~ \Gret ('" , E)\ ~ r(E) (8.248)
The term in E2 is thermally averaged according to (8.244) using G&R 3.531(3);
J dEE2 [_ dnF(E)]
oo
-00 dE
=
3
1t2 (8.249)
2
Im[L(I2a)] = 1t enO't neT) (8.250)
ret 3pmEF "
S = -(~)Im[L~;?a)] = _ ~2:;; l1(T) (8.251)
where the conductivity is cr = nO~'I/m. The dimensions of S are volts per degree. The
parameter l1(T) is dimensionless. So far 11 = ~ from the coefficient of ~/EF in (8.247). Other
contributions to l1(T) are derived below.
Taylor and MacDonald (1986) evaluated this expression for the alkali metals at high
temperature. Rather good agreement is obtained, as shown below. First it is necessary to find
more contributions to l1(T). At high temperature it is a good approximation to set = Cpt, 'I
where the transport lifetime is entirely from phonons, and is given by the transport form of
alpha-squared-J in (8.226):
1 m J2k J21t
'pl(k) = (21tik3 0 q3dq 0 d<l> ~Mi(q){nB[roA,(q)] + nF[roA(q)]}
Contributions to l1(T) are obtained by expanding k about kF and keeping the first-order terms
in (k - kF)/kF ~ ~/2EF' The prefactor of k- 3 gives another contribution of 11 = ~ to 11. The
integration limit of 2k gives a contribution to 11 of
2m J21t
2q(T) = 2"pl d<l> LMi(2kF){nB[roA(2kF)] + nF[roA(2kF)]}
1t 0 A
Another contribution to 11(T) comes from the matrix element MA(q). It is usually calculated
using a screened pseudopotential for the electron-ion interaction. The better pseudopotentials
are nonlocal, which means they depend upon MA(k, k + q) rather than just on q. This k
dependence can also be expanded around the point kF . The pseudopotential gives another
contribution to 11(T) which is called reT). The contributions to l1(T) are
l1(T) = 3 - 2q(T) -1r(T) (8.252)
Table 8.2 shows the evaluation of these terms for the alkali metals at various temperatures.
The comparison with the experimental data is good for Na, K, and Rb. The high-temperature
thermopower seems to be understood in these cases. Both q(l) and reT) are small for Na
Sec. 8.5 • Quantum Boltzmann Equation 549
TABLE 8.2 Thennoelectric parameters in the alkali metals.

11 = 3 - 2q - r /2 (Taylor and MacDonald, 1986, used with pennis-
sion).
Metal T(K) 2q(1) r(1)/2 11
Theory Experiment
Li 424 9.26 -1.43 -5.33 -6.3

Na 300 0.04 -0.09 3.05 2.9
K 200 0.83 -1.87 4.04 4.0
Rb 100 4.78 -4.49 2.71 2.8
Cs 100 9.32 -7.15 0.83 0.0
because the electron-ion pseudopotential is nearly zero at q = 2kF • The thermopower is

easier to evaluate at high rather than at low temperature. At high temperature the resistance is
dominated by the electron-phonon interaction, which is well approximated by using the
transport form of alpha-squared-J At low temperatures one also has to include the k
dependence of impurity scattering, as well as the ordinary form of alpha-squared:! while
solving (8.235). In addition, phonon drag is important at low temperature.
8.5. QUANTUM BOLTZMANN EQUATION
There are several different methods of doing transport theory. The theory used in the
preceding sections uses the Kubo relation for the conductivity and is called "linear response."
One assumes that currents are proportional to fields. The proportionality constants can be
evaluated in equilibrium. This method works because one assumes that the applied fields are
small, and the system is only infinitesimally disturbed from equilibrium.
A second method of transport theory is discussed in this section. One assumes the
existence of a distribution function f, which describes the behavior of the particles. One writes
a differential equation for the motion off through phase space. The differential equation is a
Boltzmann equation. One then tries to solve the Boltzmann equation for a system out of
equilibrium. For fields that are small, the system is only slightly out of equilibrium, and one
reproduces the linear response solutions described earlier. The advantage of the Boltzmann
equation method is that one can also try to solve the equation when the system is far from
equilibrium.
The original Boltzmann equation described the behavior of a distribution function
f(v, R, t) of three variables: velocity, position, and time. The Wigner distribution function
(WDF) was introduced in Sec. 3.7. It is equivalent to a distribution functionf(k, 00; R, t) with
four variables: wave vector k, energy ro, position R, and time t. This latter distribution
function is the one needed for many-particle systems. Since f(k, 00; R, t) is not positive
definite, calling it a "distribution function" is probably misleading. This phrase is used since
it is widespread, but the warning should be kept in mind.
The differential equation obeyed by f(k, 00; R, t) is called the quantum Boltzmann
equation, which is abreviated QBE. It is derived rigorously in the following sections. Here a
quick derivation is provided using semiclassical arguments. According to the Liouville

theorem a distribution functionf(qj, t) is stationary when it obeys the equation
of = 0 = af + L af qj + (af ) (8.253)
at j aqj at s
where the last term is from scattering. For the WDF, k =F (force), R = v (velocity), and
cO = v . F (Joule heating). The QBE is
0= (i+V'VR +F' V +V'F~)r+ (af )

k (8.254)
at aro at s
This equation has nearly the right form. It is derived rigorously in the next section, which
produces a few more terms from self-energy contributions. That derivation also provides a
prescription for obtaining the scattering term.
Equation (8.254) has one feature that is important. The additional variable ro also causes
a new driving term on the left of the form V' Fa/ aro. This term was first derived by Mahan
and Hansch (1983). The semiclassical distribution functionf(v, R, t) lacks this driving term
since it lacks the energy variable roo The QBE for the WDF is a different equation from the
traditional Boltzmann equation.
8.5.1. Derivation of the aBE

The QBE is the equation of motion for the Green's function G<. The method of deriving
transport equations was pioneered by Kadanoff and Baym (1962). Recall from Sec. 3.7 that
fer, t; R, T) = -iG«r, t; R, T), where (r, t) are the relative variables and (R, T) are the
position and time in center-of-mass. The Green's function was defined in terms of the field
operator as
G«r, t; R, T) = i(",t(R - !r, T - !t)"'(R + !r, T + !t)} (8.255)
The next step is to Fourier transform the relative variables (r, t) into (k, ro):
G«k, ro; R, T) = J J
d 3 re- jk . r dtej(J)tG«r, t; R, T) (8.256)
The QBE will be derived for a particle in a weak electric field. The intent is to describe
interacting many-particle systems that have a small current flowing in response to a small
electric field. The derivation will be sufficiently general to include any kind of particles and
nearly any kind of interactions.
The electric field can be introduced as either a scalar or a vector potential. The QBE is
independent of this choice, as required by gauge invariance. Here both are included, in order
to provide the most general derivation. There will be an electric field E v , which is from a
vector potential, and another electric field Es from a scalar potential. The final version of the
QBE will include only the total electric field E = Ev + Es. The scalar potential is introduced
through the interaction term
(8.257)
The vector potential is introduced by changing the momentum of each charged particle to
(p - eA/c), where the vector potential is A = -cEvt. A vector potential proportional to time
could occur in a wire loop with a slowly varying magnetic flux through the center. The present
theory is for a dc electric field.
Equations (3.331) and (3.332) are equations of motion for the four Green's functions in
the 2 x 2 matrix for G. The two electric field terms are added to the left-hand side of these
equations-they are included in Ho. The derivation of the QBE involves several algebraic
manipulations on the left-hand side of these equations. In order to avoid a lot of cumbersome
notation, the right-hand side of these equations are not written during these steps. On the left,
the steps are the same for all six of the real-time Green's functions. The generic symbol G
applies to anyone of them.
The two equations for G in (3.331) and (3.332) are first added and then subtracted:
[i(~-~)
at at -HI -H2 JG=
l 2
(8.258)
[i(a~1 + a~J -HI +H2JG = (8.259)
where HI = Ho(rl' PI) and H2 = Ho(r2' -P2)' The two equations contain time derivatives
that relate either to the relative or center-of-mass motion
a a a
-+-=-,
a a
---=2-
a (8.260)
atl at2 aT atl at2 at
For particles with parabolic band dispersion, the sum and difference of the two Hamiltonians
produce simple expressions in relative coordinates. As a first step, consider what happens to
the two momentum terms which are in the form of (p - eA/ c)
PI + eEA = !
(p + eEvT) + (P + eEv t) (8.261)
P2 - eEvlz = -(p + eEvT) + !(P + eEvt) (8.262)
where P and P are the relative and center-of-mass momentum. It is important to understand all
of the various plus and minus signs. Since A = -cEvf then (p - eA/c) = (p + eEvt) for PI'
However, in H2 the momentum enters as -P2 which explains the sign change on the bottom.
On the right one uses tl,2 = T ± t/2 and PI,2 = ±p + P/2. These results make it easy to see
the form of HI ± H 2 :
(8.263)
(8.264)
Divide the top equation by two, and obtain the following two equations for the Green's
function:
2 [i -a - - I (p + eE T)2- 1
- (P + eE t) 2 + eE • R ] G = (8.265)
at 2m v m v S
[i a~ - ~ (p + eEvT) • (P + eEvt) + eEs • rJ G = (8.266)
These two equations describe the relative and center-of-mass motion of the function
G(r, t; R, T). The goal is to derive the QBE for the WDF G«k, m; R, T} Fourier transform
the variables (r, t) to the set (q, Q) as in (8.256). This transform changes p to q, r to iVq , djdt
to -iQ, and t to -iajaD.. The two transformed equations are
2[Q + eEs'R - 2~ (q + eEvTi + 8~ (VR +eEv ~r]G(q, Q; R, T) =

(8.267)
{a~ +~(q + eEvT)' (VR + EEv ~) + eEs ' Vq]G(q, Q; R, T) =
The second of these equations has, on the left-hand side, exactly the same terms as are found
in the Boltzmann equation. This similarity suggests that the QBE is the same as the BE.
However, this conclusion is incorrect. The above set of equations have several things wrong
with them
1. There is the term eEs • R. This term seems to combine with Q to produce a center-of
mass energy 0) = Q + eEs • R Together they suggest that the energy of a particle
depends upon its location. The energy is different at one end of a sample than at the
other. However, this behavior is contrary to common sense. When there is a small
electric field along the sample, and a small current flowing, we expect the system to
be uniform. There is the same particle density, current density, etc. at each point in the
solid. There is no dependence upon R. This undesirable term has to be eliminated.
2. The relative momentum seems to enter in the combination of q + eEvT. It depends
upon the center-of-mass time T. This feature is also unphysical, and needs to be
eliminated.
3. The result is not gauge invariant, since the two fields Es and Ev enter differently.
All of these problems can be eliminated through a variable transformation. Of course, this
transformation also causes some derivatives to change:
Q + eEs • R => 0)
q +eEvT => k
a (8.268)
VR => VR + eEs aO)
a a
aT => aT + eEv' V k
These transformations cause the two equations in (8.267) for the Green's function to now have
the form
[ 0) - Ek + 8~ ( VR + eE a:) 2] G(k, 0); R, T) =

(8.269)
{a~ +Vk 'V R + eE· (Vk +Vk a:) ]G(k, 0); R, T) =

The notation on the Green's function has been changed again. The arguments (q, Q) have
been changed to (k, 0). The left-hand sides of these two equations are now in the form that is
useful. The lower equation has exactly the same terms as in (8.254), and is the quantum
Boltzmann equation. The electric field is E = Ev + Es. Both electric fields contribute in the
same way, and the result is now gauge invariant.
The above variable transformation makes the QBE gauge invariant. The results are valid
for dc electric fields. A more complicated variable transformation is required for ac electric
fields (Levanda and Fleurov, 1994).
Now it is time to restore the scattering terms to the right-hand side of the equal sign. The
scattering terms for G< and Gret are in (3.341). They are the most important Green's functions
in applications using real time. The self-energy terms on the right are found by following the
same steps to bring the left-hand side to (8.269). In doing these operations, keep in mind that
the scattering terms are of the form
First add and subtract the two equations for each Green's function. There follows a series of
variable transformations. They can be understood by examining just one term in the right-
hand side of Eq. (8.269). The various variable arguments yield a fairly complicated expres-
sion, which is derived by the following steps:
1. Change the two sets of variables (Xl' X3)(X3' XZ) to the center-of-mass grouping
(Xl -X3' (Xl +X3)/2)(X3 -Xz, (X3 +xz)/2).
2. Change the integration variable from x3 to Y = Xl - x3, so the variable grouping
become (y, Xl - y/2)(xl - Xz - y, (Xl + Xz - y)/2).
3. Change to center of mass variables X = (Xl - xz), X = (Xl + xz)/2, which produces
the variable grouping (y, X + (x - y)/2)(x - y, X - y/2).
4. Change the integration variable X to z = X - Y which produces the final arguments of
(y, X + z/2)(z, X - Y /2).
Below are the full scattering equations for G< and Gret . The explicit variables in the scattering
terms are only written out in the first term, but are identical for the other three. The order of
the terms in the scattering integral is important. The first one always has argument
(y,X + z/2) while the second one has argument (z,X - y/2). The four-vector in the exponent
is q = (k, 00):
[00 - Ek + 8~ (VR + eE a:Y]Gret(k, 00; R, T)

J .J .
= 1 + 2"1 dze- lqz dye-lqY[~ret(y' X + z/2)Gret (z, X - y/2)
+ Gret(y,X +z/2)~ret(z,X - y/2)] (8.270)
{a~ + Vk· VR + eE· (Vk + Vk a:) ]Gret(k, 00; R, T)

= Jdze- iqz Jdye-iqY[~retGret - Gret~ret] (8.271)
J .J .
= 2'1 dze-· qz dye-·qY[~tG< - ~<Gt + Gt~< - G<~t] (8.272)
{~+ vk· VR +eE· (Vk +Vk a:) ]G«k' 00; R, T)

= J
dze- iqz Jdye-iqY[~tG< - ~<Gt - Gt~< + G<~t] (8.273)
These four equations are the important ones for nonequilibrium calculations. Although one
primarily wants to find G<, it is always necessary to first find Gret • These equations were first
derived by Fleurov and Kozlov (1978).
So far no approximations have been made, and the equations are exact. The QBE has a
linear term in the electric field E. However, the equation is exact to all powers of E, not just to
the first power.
8.5.2. Gradient Expansion

Equations (8.271) are usually too hard to solve because the scattering terms on the right
have a complicated form. Some sort of approximation has to be introduced to simplify the
right side. The approximation described below is only valid to first order in the field.
Kadanoff and Baym (1962) introduced an approximation for evaluating these scattering
terms, which is called the "gradient expansion." They assume that the center-of-mass time T
is very large, and take the limit that T --+ 00. At large values of T, they assume that the system
is approaching its asymptotic limit, so that variations with respect to T are small. Obviously
the gradient expansion is not suitable for studying transients, since it is poor at small values of
T. Neither is it useful for steady-state ac phenomena (Mahan, 1987). Indeed, the T dependence
is so poorly described in the gradient approximation that it should not be used and the T
derivative terms should be dropped from the QBE. Nevertheless, the gradient expansion is
used since it is applicable for homogeneous (small R dependence) steady state (small T
dependence) systems.
The center-of-mass variables are all in the form (R + AR, T + AT), which are expanded
in a Taylor series about the point X == (R, T). The integrals can be done for each term in the
series. These integrals usually cause further derivatives. The gradient expansion is shown
below for a typical scattering term:
1= Jdze-iqz Jdye-iqy~(y,X + zj2)G(z, X - yj2) (8.274)
= Jdze- iqz Jdye-iqy{~(y,X)G(Z'X)
+ z a~(y, X) G(z X) _ ~(y X)~ aG(z, X) + ... } (8.275)

2aX' '2aX
(8.276)
The first term on the right are usually the largest. It is customary to retain the first derivative
terms and ignore higher derivatives. They can be expressed using a Poisson bracket notation:
[~, G] = (Vq~)VxG - (Vx~)VqG (8.277)

a~aG aGa~
= anaT- anaT-Vq~.VRG+VqG.VR~ (8.278)
These frequency derivatives are with respect to the n variables, since the variable change
(k = q + eEvT, co = n + eEs • R) has not yet been made. If this step is taken now, the
derivatives get altered according to Eq. (8.268), which changes the Poisson brackets to
(8.279)
where E is again the total electric field. This analysis finally derives from (8.271) the
following expression for the nonequilibrium retarded Green's function in an electric field,
when Gret and ~ret depend upon (k, co; R, T):
(8.280)
(8.281)
The additional terms from the Poisson bracket, which are linear in the field E, have been
transferred to the left of the equal sign.
These equations simplify for nonequilibrium systems which are both homogeneous
(VR = 0) and steady state (a/aT = 0). The Poisson brackets vanish, as do several terms on
the left. Also ignored are terms nonlinear in the electric field O(E2 ), and find for the above
two equations
(8.282)
(8.283)
The first equation is easily solved, to yield
1 2
Gret(k, co) = ~ (k ) + O(E ) (8.284)
co - ck - ret , CO
The retarded Green's function appears to have no first-order term in the electric field. It
actually does, since the self-energy ~ret has a term linear in the field due to the electron-
phonon interaction. This term is small and seems to have little effect. If it is ignored then the
retarded Green's function is unchanged from its value in equilibrium. This result considerably
simplifies the solution to the QBE. The solution (8.284) also satisfies the equation (8.283).
Related quantities such as the advanced function Gadv = Giet and the spectral function
A(k, co) are also unchanged to first order in the electric field. This completes the discussion of
the retarded Green's function in a static electric field.
Next the gradiant expansion is applied to (8.271) for the equations for G<. Again
keeping only first-order derivatives, the final results are the quantum Boltzmann equation
(QBE):
(8.285)
0
i { oT+vk'VR+eE' [ ( 1- OL 0 ]} G<
owret ) Vk+(Vk+VkLret)ow
- ieE' [OL< V Re[G ] - oGret V L<] = L>G< - L<G>

ow k ret ow k
+ i[Re[L ret ]' G<] + i[L<, Re[Gret ]] (8.286)
Standard relationships have been used in deriving this equation, such as Gt - Gt = 2 Re[Gret ]
and Gt + Gt = G< + G> .
Equation (8.286) is the quantum Boltzmann equation. It is rather formidable. It is also
difficult to solve, since it is usually an integral equation. Sometimes it is nonlinear in the
particle density because the self-energy functions L< and L> are also functions of G< and
G>.
The QBE also contains the functions G> and L>. Similar equations for these functions
can be derived by starting from the general equations (3.331) and (3.332). This derivation
shows that the equation for G> is almost identical to the one for G<. In fact, one can prove
that the following identities are valid:
G> = G< - iA (8.287)

L> = L< + 2i Im[L ret ] = L< - 2ir (8.288)
These relations are trivial to show for equilibrium, but they are also valid for nonequilibrium
situations. These identities will be used often to simplify expressions. For example, the main
scattering term in the QBE can be immediately simplified to
(8.289)
This result will be employed in the calculations. The quantities G< and L< are generally
proportional to the density of particles, while retarded functions are only indirectly dependent
upon the density of particles-only through the self-energy function Lret . In the QBE, each
term has one factor that is either G< or L <, so each term is proportional to the density of
particles. This equation does have the character of a transport equation.
The QBE simplifies for the treatment of systems that are homogeneous (VR = 0) and
steady (ajar = 0). These derivatives are dropped, as well as the Poisson brackets, since the
latter contain similar derivatives. In this case the QBE is
eE· [( 1 - aLret)
aro Vk + (Vk a] G<
+ VkLret) aro
- ieE' [aL< V Re[G ]- aGret V L<]

aro k ret aro k
= L<A - 2rG< (8.290)
This equation can be simplified. The QBE is only valid to first power in the electric field, since
terms of O(E2) have systematically been ignored. For example, in the gradient expansion the
second derivative terms would give contributions of O(E2) were they retained. So the
equation is exact to first order in the electric field, but is not exact to higher orders in the field.
This fact is utilized to simplify the left-hand side of the equation. Since the field multiplies
each term, on this side of the equal sign the Green's functions can be taken to have E = O. Of
course, these expressions are just the equilibrium quantities in Sec. 3.7.
Consider the frequency derivatives of the left-hand side of the equation. Write the
spectral function in the shorthand notation
2r
A = 0'2 + r2' (8.291)
The left-hand side will have three types of frequency derivatives: anFjaro, ar jaro, aO' jaro.
The coefficients of the latter two vanish, which leaves only terms in anFjaro. All of the terms
proportional to nF vanish, leaving only terms multiplied by anFjaro:
A(k, ro)2 anF eE' {(Vk + Vk Re[Lret]W + O'Vkr} = L>G< - L<G> (8.292)
aro
The factor of A(k, ro)2 appears in each term, and was taken outside. The left-hand side of this
equation now contains only known quantities, which can be calculated in equilibrium. The
scattering terms remain on the right-hand side. Finding them still involves work, usually in the
form of an integral equation. This final form for the QBE is exact for transport which is linear
in the field, and for steady state, homogeneous systems. It is quite analogous to the similar
expression for the classical BE, which is
aI'CO)
-eE . Vk _u_ = L> G< - L< G> (8.293)
aro
The classical equation has aICO) jaro, while (8.292) has anFjaro.
Equation (8.292) is the steady state, homogeneous form of the QBE. It should be the
starting point for many transport calculations. It is exact, and is an alternative to using the
Kubo formalism, which is also exact. The derivation of this equation has been complicated,
and has entailed some work. However, once derived, it is often the easiest starting point for
deriving the transport coefficients. Calculations using Eq. (8.292) entail less work in getting
to the answer than any other formalism.
8.5.3. Electron Scattering by Impurities

The quantum Boltzmann equation (QBE) (8.292) will be solved for the electron scat-
tering by impurities. This case was solved in Sec. 8.1 using the Kubo formula. This exercise is
useful for two reasons: (1) producing the known result demonstrates the correctness of the
QBE; and (2) this case is the easiest one to solve, and provides an introduction to the
techniques for solving the QBE. The present example makes two assumptions that are
intended to make the solution as easy as possible: (1) the impurities are dilute, so that the
simultaneous scattering from several impurities can be neglected; and (2) the impurities have
no internal degrees of freedom, such as spin or vibrations, which can be altered by the
electron scattering. The second assumption implies that impurity is a simple potential that
elastically scatters the electron. The impurities are randomly located in the solid. The method
of Sec. 4.1.5 is employed for averaging over the random distributions.
The first step in solving the QBE is to find the retarded functions. For impurity scattering
they are independent of the electric field, at least for small fields. The retarded self-energy for
scattering from impurities of density n i is given in Sec. 4.1
(8.294)
The T matrix is energy dependent, and hence off-shell.
The next step is to express the self-energies 1:< and 1:> in terms of the Green's functions
G< and G>. The retarded functions Gret and 1:ret are known, since they are unchanged by the
electric field. However, G< and G> are, at this point, unknown since they are affected by the
electric field. The self-energy functions are
1:<'>(k, ro) = ni J(~~3ITpk(ro)12G<'>(p, ro) (8.295)
This result is derived below. The off-diagonal T matrix is the one in Eq. (4.111), which also
depends upon the energy ro.
Equation (8.295) is now derived. An impurity at R = 0 is represented by an electron
potential V(r), whose Fourier transform is V(q). The self-energy 1: from a single scattering
event is
(8.296)
where I is the unit tensor. This self-energy is inserted into (2.157). That equation is iterated in
order to find the effects of repeated scattering from the same impurity. The resulting self-
energy series is rewritten in a symbolic notation, where the product of two functions implies
an integral over dx. Iteration of these equations gives the series for 1::
(8.297)
After summing this series, the matrix ± is examined for its individual components. The ones
for 1:<'> are
(8.298)
Note the analogy with the equation for G< in (2.159). The resemblance is expected, since the
series for 1: has the same mathematical structure as the one for G.
For impurity scattering, the unperturbed self-energies 1:;'> = 0 are zero. There is only
the last term in (8.298). Since the self-energies 1:ret and 1:adv are T matrices, (8.298) gives the
result in (8.295). The damping function r(k, ro) and other self-energy 1:>(k, ro) are
J d3p
2r(k, ro) = ni --3 ITpk I2A(p, ro)
(2n)
(8.299)
(8.300)
For dilute impurities, it is sufficient to retain only those terms that are first order in the
impurity concentration ni . The broadening due to the spectral function A(p, co) is from
impurity scattering. Since the self-energy is multiplied by ni , one can replace the spectral
function A(p, co) ;::; 27to(co - Ep). This expression is then equal to the imaginary part of the T
matrix: 2r = nivkcrT, where crT is the total cross section from impurity scattering in Sec. 4.1.
The starting point for solving the QBE is (8.292). On the left of the equals sign is the
factor of
(8.301)
This expression is simplified by neglecting terms of G(nt) such as r Re[:E ret ]. The term in
cr = co - Ek - Re[:E ret l is small since the factor of A(k, CO)2 tends to force cr ;::; O. The above
expressions can be approximated by VkrA(k, CO)2:
A(k, coi dn;~CO) eE' Vkr = 2irG< - i:E<A (8.302)
These steps complete the derivation of the left-hand side of the QBE.
On the right side of (8.302) the scattering terms vanish in equilibrium since :E< = 2inFr
and G< = inFA. Since current is flowing in response to the field, the system is slightly out of
equilibrium. The right-hand side is expected to have factors similar to those on the left. This
discussion suggests the following ansatz for the nonequilibrium Green's function
G< = iA(k, co{nF(co) - en;~CO))eE ·vkA(k, CO)] (8.303)
The function A(k, co) is unknown, and needs to be determined by solving the QBE. The
factors that multiply A are for later convenience. The above choice does not make any
assumptions for the value of A. Using this ansatsz in the self-energy function gives
(8.304)
Putting these two expressions into the right-hand side of (8.302), the equilibrium terms
cancel, and the remaining terms each have the common factor of (dnF/dco)A:
A2 dnF(CO) eE· Vk r = - (dn F(CO))A{2reE' Vk A

dco dco
- ni J(27t)
d3p
--3
2
ITkpl A(p, co)eE' vpA(p, co)} (8.305)
After canceling all of the common factors, there is an integral equation for the unknown
function A(k, co):
(8.306)
This integral equation for A(k, co) is nearly identical to the one found in solving the
conductivity from the Kubo relation. The quantity VkA(k, co) is similar to the factor of
r(k, ro - iO, ro + io) in Sec. 8.1.3. As the derivation proceeds, the differences between these
quantities disappears, and the two approaches give the same resistivity.
After solving this equation for A(k, ro), the Green's function G< in (8.303) is known. It
is used in (3.340) for the current. The first term in (8.303) for G< gives a zero current. The
second term gives a current proportional to the field E, and this proportionality defines the
electrical conductivity
O"I1V = e2 J --3
d 3p JOO -2
dro v vpV (dnF(ro»)
pl1 - - d - A(p, ro)A(p, ro) (8.307)
(21t) -00 1t ro
The two equations (8.306) and (8.307) provide the solution for the conductivity from the
QBE.
For electrons in metals, with a spherical Fermi surface, the conductivity from impurity
scattering is found easily from these equations. First, do the integral over dro in (8.307). At
low temperature the factor (-dnF/dro) is nearly a delta function which sets ro ~ O. At low
temperature
O"I1V =; Jd3p
--3 vpl1vpvA(p, O)A(p, 0)
(21t)
(8.308)
The equation is reduced as far as possible. Next one must solve the equation (8.306) for A.
z
Adopt a vector coordinate system where the direction is k. Then the various scalar products
of vectors can be found from the law of cosines:
k E=
0 kop = cos(9)
cos(9 0 ), (8.309)
poE = cos(90 ) cos(9) + sin(9) sin(90 ) cos(<1» (8.310)
Equation (8.306) is multiplied by the vector E. The integral over the d<l> part of d3p makes the
cos( <1» term vanish. Each of the remaining terms has a factor of cos(90 ) which can be
canceled. Then one finds the scalar equation
1
A(k, 0) = 2A(k, d 3p
ni -3lTkpl
0) + 2r 2 J
A(p, O)A(p, 0)cos(9) (8.311)
(21t)
The spectral functions A(k,O) and A(P,O) force k = p = kF . The factor of A(p, 0) under the
integral can be set equal to A(kF' 0) and taken out of the integral. Then the above equation
can be solved to find
(8.312)
where t = 1/2r is the time between scattering from impurities, while tt is the lifetime which
is important for resistivity; it has the factor of (1 - cos 9f ):
ne2tt
0"=-- (8.313)
m
tt
J
d3p ITkpl 2A(p, ro)[1 - cos(9)]
-1 = ni --3
(21t)
(8.314)
This formula for the resistivity is identical to (8.25). The QBE gives the same formula for the
resistivity as found earlier from the Kubo formula. The two methods are also in exact
agreement for much more complex cases, such as electron-phonon scattering at nonzero
temperatures and frequency. Most of the steps in the derivation were spent in getting to
Sec. 8.6 • Quantum Dot Tunneling 561
(8.302). This equation is the starting point for any calculation for homogeneous systems. The
number of steps between this equation and the final resistivity is small. The QBE is an
efficient method of finding the resistivity or other transport coefficients.
8.6. QUANTUM DOT TUNNELING
8.6.1. Electron Tunneling

Cohen et al. (1962) introduced the concept of the tunneling Hamiltonian, which has
become universally adopted for the discussion of tunneling. Their idea was to write the
Hamiltonian as three terms:
H=HR+HL+Hr (8.315)
Hr = kpcr
I:(TkpCZcrCpcr + hc) (8.316)
The first term HR is the Hamiltonian for particles on the right side of the tunneling junction. It
contains all many-body interactions. Similarly, HL has all the physics for particles on the left
side of the junction. These two are considered to be strictly independent. Not only do these
two operators commute, [Hv H R ] = 0, but they commute term by term. The Hamiltonian on
the right can be expressed in terms of one set of operators Ckcr and those on the left by another
set Cpcr' and these operators are independent {Ckcr ' C~cr} = O. This assumption is probably
reasonable. They further assumed that the tunneling is caused by the term Hr in (8.316). The
tunneling matrix element Tkp can transfer particles through an insulating junction. This
transfer rate is assumed to depend only on the wave vectors on the two sides k and p and not
on other variables, such as the energy or spin of the particles.
The theory of electron tunneling was mainly applied to superconductors. It developed
very rapidly and was entirely based on the tunneling Hamiltonian. The theory showed
excellent agreement with the many experiments. The history books were written describing
this satisfactory situation, and the scientists in this field wandered off to do something else.
About this time there began a serious investigation, starting with Zawadowski (1967), Caroli
et al. (1975), and Feuchtwang (1975), about the validity of the tunneling Hamiltonian. Of
course it was found to be a poor approximation, since the tunneling rate depends on the
energy of the particle as well as its wave vector. What does this turn of events do to the lovely
agreement between theory and experiment for tunneling in superconductors? Actually, it
probably changes none of it. The tunneling in superconductors takes place over a very narrow
span of energies in the metal, i.e., within a Debye energy of the Fermi surface. Also, all the
electrons involved have their wave vector very near the Fermi wave vectors kF and PF on the
two sides of the junction. It is an adequate approximation to treat the transfer rate Tkp as a
constant To which is evaluated at kF and PF' because the variations in Tkp with energy must be
on the scale EjEF which are negligible for E ~ Ll ~ 1 meY. Similarly, the variation of Tkp
with p or k is on the scale of the Fermi wave vectors. One can treat the transfer rate To as a
constant if the energies involved are small. The tunneling Hamiltonian is believed to be an
improper formalism only when the applied voltages are large, say 1 eY.
For quantum dots the energy scales are also rather small, and are near to the chemical
potentials of the metal electrodes. Here the tunneling Hamiltonian is also thought to be a valid
formalism.
The general model of a tunneling junction is shown in Fig. 8.12. It describes a non-
equilibrium situation, since the chemical potential on the left-hand side ~L is not the same as
~R on the right. They differ by the applied voltage eV = ~L - ~R' The potential drop of eV
occurs in the insulating region between the electrodes.
The tunneling Hamiltonian (8.316) is used to derive a correlation function for electron
tunneling currents. This correlation function has the form of a Kubo formula, except for an
important difference. The Kubo formula for the conductivity in Sec. 3.7 expresses the ratio
between the current and the voltage (actually electric field). In tunneling theory, the corre-
lation function gives the current as a function of voltage.
The tunneling current through the insulating region is expressed as the rate of change of
the number of particles on, for example, the left-hand side of the junction NL . This rate is
found from the commutator of NL = L C~a Cpa with the tunneling Hamiltonian. Only the
term H T fails to commute with Nv
NL = i[H. N L] = i[HT• N L] = i L [Tkp cta Cpa - T~pC~aCka] (8.317)

kpa
The total current I through the tunneling interface is defined as the average value of this
operator:
(8.318)
The average value of (NL(t)) is obtained by following the same steps used to derive the Kubo
formula in Sec. 3.8. The total Hamiltonian is written as H = H' + H T, where H' = HR + HL.
Go to the interaction representation, where the tunneling term H T is treated as the interaction
and everything else H' is Ho. Then the S matrix is expanded in terms of the perturbation H T.
The objective is to obtain a formula where I ex ITkp 12 , so only the first term needs to be
retained in the expansion of the S matrix. These steps bring us to the formula
I(t) = -ei Loo dt' ([NL(t). HT(t'))) (8.319)
HT(t') = eiH't' HTe-iH't' (8.320)

NL(t) = eiH'tNLe-iH'l (8.321)
where the time dependence of HT(t) and NL(t) is governed by H'.

An important step in the calculation is to insert the chemical potentials ~L and ~R for the
two sides of the junction. This insertion must be done with more care than usual, because the
chemical potential is not the same on the two sides of the system. The initial Hamiltonian
(8.316) has been written to not include the chemical potentials, so the energy is measured on
FIG URE 8.12 Tunneling between two nonnal metals. The arrow shows the electron path through the oxide interface.
an absolute scale rather than relative to the chemical potentials. However, now insert the
chemical potentials into the time developments, so that the energy can be measured with
respect to the different chemical potentials on each side of the tunnel junction. The symbols
KR and KL denote the Hamiltonian with respect to the respective chemical potentials:
KR = HR - llRNR (8.322)
KL = HL - llLNL (8.323)
K' =KR+KL (8.324)
For a free-particle system, KR = L ~kCtO'CkO" while HR = L ckCtO'CkO' since ~k = ck - Ilk

Since the number operators commute with H', it is possible to write H' = K' + llRNR + llLNL
and exp(iH't) = exp(iK't) exp[it(IlLNL + llRNR)] since the exponentials can be separated
when the operators commute. The time development of HT is
HT(t) = eiH'tHTe-iH't = eiK'I[eit(IlLNL+IlRNR)HTe-il(IlRNR+IlLNLl]e-iIK'
-- eiK'1 ,,[r,
L.... kp
t C
eil(llrIlL)ckO' pO'
+ eil(IlL-IlR)r,*kp ctpO' CkO' Je-ilK'
kpO'
The commutator of HT with the number operators produces the factor ilL - llR = eV, which
is identified as the applied voltage, as in Fig. 8.12. The applied voltage appears in the
correlation function. The correlation function will now be evaluated by assuming that both
sides of the junction are in separate thermodynamic equilibrium.
The current operator in (8.319) now becomes
I(t) = eJt dtl( L[Tkpe-ieVlctO'(t)CpO'(t) - T~peieVlC;O'(t)CkO'(t)],

-00 kpO'
"['7' t p'O" (t') + 1k'p,e

L.... 1k'p,e-ieVt'ctk'O" (')C '7'* t (')C
ieVt'cp'O" t k'O" (t I)])
k'p/cr'
From now on the time development of C kO' operators is governed by CkO'(t) = eiKRICkO'e-iKRI
and CpO' operators by CpO'(t) = eiKLtCpO'e-iKLI. Define the operator A as
A(t) = L TkpCtO'(t)CpO'(t) (8.325)

kpO'
and the current is written as the summation of two currents
I = Is +IJ (8.326)
Is(t) = e J~oo dt'8(t - t'){eieV(t'-I) ([A(t), At (t')]) - eieV(t-t') ([At (t), A(t' )])} (8.327)
I;(t) = e J~oo dt' 8(t - t'){e-ieV(I'+I) ([A(t), A(t' )]) - eieV(I+t') ([At (t), At (t')])} (8.328)
The term Is is for single-particle tunneling, which is important for quantum dots. The other
term I J describes the tunneling currents associated with the Josephson effect in super-
conductors. It is evaluated in Chapter 10.
The terms in Is have just the right combination of factors to be a retarded Green's
function. The integrand depends only on the difference of(t - t'), so set t' = 0:
Urelt) = -i8(t) ([A(t), At (0)]) (8.329)
Uret ( -eV) = J~oo dte-ieVturet(t) (8.330)
The second term in Is is the Hermitian conjugate of Uret ( -e V) except for a sign. But the
Hermitian conjugate of the retarded function is just the advanced function. The single-particle
tunneling current is written as;
Is = ie[Uret(-eV) - Uadv(-eV)] (8.331)

= -2e Im[Uret(-eV)] (8.332)
It is twice the imaginary part of a retarded correlation function, which has the form of a
spectral density function. Using the relationship between the Matsubara and the retarded
correlation functions discussed in Sec. 3.3, the way to calculate the single-particle tunneling is
to evaluate in the Matsubara formalism the correlation function
U(iro) = - J:d-ceiO)nT(T~(-c)At(O))
= - L L TkpT;,p' J~ d-ceiO)nT(TTCtcr(-C)Cpcr(-c)C;'cr,Ck'cr') (8.333)
kpcrk'p'cr' 0
The single-particle tunneling current is just the spectral function of this operator evaluated at
the real frequency -eVjfl, as shown in (8.330). The Matsubara frequency ron = 2n1tj~ is
boson, since the correlation function has pairs of fermion operators. In the tunneling
Hamiltonian, the right- and left-hand sides of the tunneling junction are independent. Then
the correlation function factors into a product of the Green's functions for the right and left
sides of the junction:
U(iro) = L ITklJ~ d-ceiO)nT(TTCkcr(O)Ctcr(-c))(TTCpcr(-C)C~cr(O))

kpcr 0
= L ITkPI2J~ d-ceiO)nTCdR(k, --c)CdL(p,-c)

kpcr 0
1
= L ITkP I2"A L CdR(k, ip - iro)CdL(p, ip) (8.334)
kpcr t-' ip
The only terms which enter are those for k = k', P = P and 0" = 0"'. A type of Feynman
diagram is shown in Fig. 8.13. The symbols T in circles are the tunneling vertices. The
vertical dashed line is meant to divide the right from the left side of the junction. The solid
lines are the interacting Green's functions CdR and Cd L.
The correlation function is the product of the two Green's functions for the two sides of
the junction. Such a simple result is obtained only when one can neglect the vertex diagrams
of the correlation function. The argument for neglecting them in tunneling is that vertex
corrections require some interaction between electrons on the two sides of the junction, which
is improbable. Actually, there is another case where vertex corrections are needed in
FIGURE 8.13 Feynman diagram for tunneling, where the Ts are the vertices which link the right (R) and left (L)
sides of the junction.
tunneling, and that is to account for processes where the tunneling electron can excite
vibrations or other excitations in the interface. That happens in quantum dot tunneling.
The first example of solving (8.334) will be for the tunneling between two normal
metals. The experiments show the tunneling is linear in voltage V for small values of V, and
becomes slowly nonlinear at larger values of V. The latter behavior is a failure of the tunneling
Hamiltonian, which predicts only linear behavior. The present discussion is limited to small
voltages, for which the tunneling Hamiltonian formalism is valid.
For a normal system, consider the electrons to be simple quasiparticles, and approximate
the Green's functions by C§kO) and c§iO). The Matsubara summation is then familiar:
(8.335)
U(im) = L ITkp l2 n~(~k) - nF(~p) (8.336)

kpcr l(f) + ~k - ~p
1= -2e Im[Uret(eV)]
= 4n:e L ITkpI2[nF(~k) - nF(~pW)(eV + ~k - ~p)
kp
The energies ~k and occupation factor nF(~k) refer to the right side of the junction, and ~p and
nF(~p) refer to the left side. An additional factor of two in front is for the summation over
spins. The spin is usually preserved in tunneling. The summations over k and p are just over
wave vectors. For small voltages, only electrons very near to the Fermi energy are involved in
tunneling. For these small energies, the density of states on both sides is assumed to be a
constant:
(8.337)
Lp -+ Jd 3P3 -+ NL
(2n:)
Jd~L (8.338)
1= 4n:eNRNLITI2 Jd~L Jd~R[nF(~R) - nF(~L)]o(eV + ~R - ~L)

J
= 4n:eNRNLITI2 d~R[nF(~R) - nF(~R + eV)] (8.339)
At zero temperature, the occupation numbers are step functions, so (8.339) becomes
1= 41teNRNLITI 2 revd~R = cro V (8.340)
(8.341)
The tunneling device behaves as a simple resistor, with a conductance (inverse resistance)
given by cro. With a little more care, one can show that the integral in (8.339) also equals eVat
nonzero temperatures, since the thermal smearing cancels between nF(~) and nF(~ + eV).
The result is temperature independent, at least for a range of low temperatures kBT « EF .
A formal expression for the tunneling current can also be derived when interacting
Green's functions '§R, '§L are retained in the Matsubara summation. This type of summation
was evaluated in Sec. 7.1 using the Lehmann representation
1
X(im) = AL: '§L(P, ip)'§R(k, ip - im)
I-' lp
= J~; AL(P, £L) J~~ Ar(k, £R)S (8.342)
S = ~ L:_l_ 1 nF(£R) - nF(£L) (8.343)

~0~-~~-~-~ ~+~-~
The retarded function is obtained by the analytical continuation im -+ e V + ic, and then take
the imaginary part.
These steps bring us to the formula for the tunneling current of Schrieffer et al. (1963):
1= 2e L: ITkpl 2 Jd£R
-2 AL(p, £R + eV)AR(k, £R)[nF(£R) - nF(£R + eV)] (8.344)
kp 1t
Again a factor of two is added for the spin summation. The tunneling current is expressed in
terms of the spectral functions on the two sides of the junction. Equation (8.344) is the exact
formula for I within the model of the tunneling Hamiltonian. The earlier expression (8.339)
is recovered with the free-quasiparticle approximation AR = 21tC,(£R - ~k) and AL =
21tC,(£L - ~p). The virtue of (8.344) is that it contains all many-body effects on the two sides
of the junction. Its drawback is that it is based on the tunneling Hamiltonian formalism.
If the tunneling matrix element Tkp can be approximated as a constant, then the
summations over wave vectors give the interacting density of states
L:AR(k, £R) = 21tNR(£R) (8.345)

k
L:AL(p, £L) = 21tNL(£L) (8.346)

p
which is a useful formula for describing many-body effects in tunneling which occur on a
small energy scale near to the Fermi surface.
(8.347)
~ 4nelTI2NRNL JdEnF(E + eV)[1 - nF(E)] = eJ}~o~ I (8.348)
IRL - hR = 0"0 V (8.349)
The previous result is obtained by subtracting the currents in the two directions.
8.6.2. Quantum Dots

Figure 8.14 shows the typical geometry with a quantum dot (QD). There are two metal
electrodes on the right and left, and a spherical metal particle between them. The sphere is
imbedded in an insulating material through which the electron can tunnel. The tunneling of
electrons from the right to the left electrode is facilitated by the QD. The tunneling becomes a
two-step process: the electron tunnels from one electrode to the QD, and then tunnels from the
QD to the other electrode. There is also coherent tunneling from one electrode to the other
through the states of the QD, which act as virtual intermediate states. The first experiments of
this type were done by Giaever and Zeller (1968), and they have become popular recently.
Now the QD and electrodes might be composed of semiconductors, where the nanostructure
is made by lithography.
First examine the properties of the QD. Assume it is a sphere of radius a in atomic units:
the dimensional radius R = aao where ao = 0.05292 nm is the Bohr radius. The volume of
the sphere is Vo = 4nR 3 /3. The number of electrons is N = no Vo, where the density
FIGURE 8.14 Geometry for quantum dot tunneling.

no = kJ../3rr. 2 is assumed to be the same as in the bulk metal. The density of states per spin is
NF = mkF/(2rr. 2 /i 2 ). The quantity NFVO has the units of inverse energy: it is the number of
electrons per unit energy. Conversely, its inverse is the average separation between energy
levels on the QD
E
IlE = -1- = 3rr. ry (8.350)
NF Vo (kFao)a 3
e'2 Ery
U=eV = - = - (8.351)
c 2R a
The quantity U is called the charging energy. It is the energy required to add a charge of e to a
neutral metal sphere. All of the charge collects on the surface, and the energy is just
U = e'2/2R. Table 8.3 compares these two energies as a function of sphere radius. The
energies are given in terms of electron volts. The two energy scales are interesting. The
charging energy is large compared to the thermal energy kBT, regardless of which temperature
is used. If ii is the number of electrons on the QD for it to be charge neutral, then the energy
for n electrons on the QD is
En = llE(n - ii) + U(n - iii (8.352)
The system behaves as an Anderson model, in that the energy of the QD depends quad-
ratically upon the total number of electrons n. The energy is for a many-electron system. The
energy U is due to electron--electron interactions. A many-electron description is required for
treating the QD.
The other energy IlE is also interesting. It is much smaller than the charging energy for
typical systems with N ,. . ., 10-1000 electrons. Here the re1event question is whether IlE is
greater or smaller than the thermal energy kBT. Of course, this comparison depends upon the
temperature of the experiment: whether it is at 300 K, 1 K, or 1 mK. If ~1lE « 1 then the
energy levels on the QD can be treated as a continuum, and no error is made by replacing the
summation over states by a continuous integral. However, if ~1lE » 1, the states are well
separated in energy compared to kBT. The summation over states must remain a discrete
summation. Either condition could apply, depending upon the size of the QD and the
temperature of the experiment. Different theories in the literature assume one case or the
other. Some simple statistical calculations are presented of the fluctuation of the number of
Table 8.3 Number of electrons N, charging

energy U, and energy level separation !!E of a
quantum dot of radius a in atomic units.
Energies in units of electron volts. A value of
rs = 3.96 (sodium) is assumed
a N U !!E
5 2 2.72 2.116
10 16 1.36 0.265
15 54 0.91 0.078
20 129 0.68 0.033
25 252 0.54 0.017
30 435 0.45 0.010
35 690 0.39 0.006
40 1030 0.34 0.004
electrons on quantum dots. These fluctuations are important for controlling, or else not
controlling, the flow of current onto the quantum dot. These fluctuations play a role in the
phenomena called "Coulomb Blockade". The two different formulas in the literature for the
tunneling rate are obtain as two limits of the fluctuations.
The statistical averaging over the states n of the quantum dot includes both the single-
electron states (n - n)M and the charging energy U(n - ni. Sometimes they can be
decoupled. There are, in fact, many configurations of one-electron energy states. Denote their
quantum numbers by a, so that the energy without the charging energy is Erx.
Calculate the partition function for the electrons on the quantum dot. Assume there are a
number of single-electron energy levels with energy Erx as well as a charging energy
U(n - ni. Let nrx be zero or one, and it denotes whether an electron is, or is not, in the state IX.
Then the total energy of the system is
n=Lnrx (8.353)
rx
E = LnrxErx + U(n - n)2 (8.354)
rx
The label IX includes the spin and other quantum numbers. Many combinations of states may
have the same value of n. In this case, the partition function can be written exactly as
Z = L e-W(n-ii)2 J21t d9 einBIIrx t e-n.(~(E.-iJ)+iB) (8.355)

n 0 2n n.=O
= L e- W (n-ii)2 J21t d9 ein9 S(9) (8.356)

n 0 2n
S(9) = IIrx[1 + e-~(E.-iJ)-i9] (8.357)
For 9 = 0 then S(O) is just the usual Fermi-Dirac distribution function. The integral over 9
selects out the terms with a certain value of n. If U = 0 then the summation over n gives a
delta function at 9 = 0 and the partition function reverts to being a usual product of Fermi-
Dirac distributions for noninteracting electrons.
The role of the fluctuations is determined by the function S(9). This function is evaluated
by breaking the summation over IX into two parts depending upon whether ~rx = Erx - Il is
positive or negative, i.e., whether exp( -~~rx) is less than or larger than one.
In(S(9)) = L In[e- i9 (e i9 + e-~~')] + L In [(1 + e-~~')(1 + e-~~.(e-~ - 1»)]

~.<O ~.>o 1 + e ~~.
= In[S(O)] - i9n + 8S(9) (8.358)
n= L 1 (8.359)
~,<O
nrx = e~~.1+ 1 = nF(~rx) (8.360)
8S(9) = L In[1 + (1 - nrx)(ei9 - 1)] + L In[1 + nie-i9 - 1)] (8.361)

~d ~~
where n is the number of electrons below the chemical potential, and nrx is the Fermi-Dirac
distribution of electrons for the state IX.
First evaluate 8S for the case that the energy separation between the states of energy Erx
is less than the thermal energy kBT = l/~. Then the summation over rx can be replaced by a
continuous integral. Let NF(~)VO = PD(~)' PD(O) = II tw be the density of energy states on
the dot. Then one gets that
(8.362)
8S = JO d~PD(~) In[1 + nF( _~)(ei8 - 1)] + JW d~PD(~) In[1 + nF(~)(e-i8 - 1)] (8.363)
-w 0
8S = - L rx,[1 - cos(l8)] (8.364)

'=1
8S ~ -2y[1 - cos(8)] (8.365)
(8.366)
The integral has been approximated by expanding the log function in a Taylor series and
keeping the first term. Higher terms (l > 1) in the series for 8S are much smaller: the series
converges rapidly. In general y > 1 means that kBT > tw.
With this approximation the partition function can be expressed in closed form
Z = ZFDZC (8.367)
ZFD = IIrx[1 + e-~~'] (8.368)
Zc = L j\(y)e-~UV2 (8.369)
v
v=n-n (8.370)
" (y) = J21t d8 eiv8-2Y(1-cos(8)) (8.371)

v 0 2rc
= e- 2Y l (2y) ~ _1_e- v2 / 4Y (8.372)

v 2"fifY
This formula is an important result. It shows that the Fermi-Dirac statistics of the states
labeled rx are decoupled from the charging of the QD, which is given by the last term Zc.
When y > 1, then for small values of v the factor of "v(y) is independent of v, and the
averaging over the value of n involves only the term exp( _~UV2). This result is assumed by
most theoretical papers. The charging is treated as a separate statistical process from the
Fermi-Dirac statistics of the electrons on the QD.
The other case is to evaluate OS(9) while assuming that the separation between energy
levels is greater than the thennal energy. In that case this function is evaluated as
oS ~ -y<(1 - ei9 ) - y>(1 - e- i9 ) (8.373)

y< = L nF( -~"') (8.374)
;0<0
(8.375)
(8.376)
Z=ZFDZC (8.377)
Zc = L J\e- WV2 (8.378)

v
Here y < 1. The fonnulas for the two cases become identical when y = y< = y> .
However, if y < 1 then f v does depend upon y and this factor should be included in the
analysis. Our derivation shall assume that y > 1, so that the energy levels on the QD can be
considered to be continuous. The thennodynamic averaging over the charging energy is
accomplished by including the factor of exp(-BUn 2 ). In this case, the system is a type of
Hubbard model.
8.6.3. Rate Equations

The Coulomb blockade is derived using a set of rate equation, which describe how an
electron hops on or off the QD. Define in as the probability of there being n electrons on the
QD. If the number is ii in equilibrium, then we expect the probability to be something like
in = _l- exp [-B(n - ii)~ - B(n - ii)2U] (8.379)

ZQD
1= Lin (8.380)
n
where ~ is the energy to put a single electron on the QD, irrespective of the charging energy.
It is convenient to change the notation and to define n' = n - ii as the number of excess
electrons (n' > 0) or holes (n' < 0) on the QD. Similarly, the label on the site probability is
changed to in,. The SUbscript denotes the deviations from neutrality.
Define RL(m, n) as the probability that the number of electrons on the QD changes from
n to m by a jump to the electrode to the left. Similarly, RR(m, n) is the probability of going
from n to m by a jump to the electrode on the right. Since only single jumps are considered,
then m = n ± 1. These quantities are given earlier as IRL,LR in Eq. (8.347) and (8.348).
~LEL(n)
RL(n + 1, n) = e ~EL (n) - 1 (8.381)
~LEL(n - 1)
RL(n - 1, n) = 1 _ e-~EL(n-I) (8.382)
~RER(n)
RR(n + 1, n) = e~ER(n) _ 1 (8.383)
~RER(n - 1)
RR(n - 1, n) = 1 - e- ~ER (n- I) (8.384)
~R,L = 4nlTL,RI2NL,RNQD (8.385)

EL(n) = eVQD + U(2n + 1) (8.386)
ER(n) = e(VQD - V) + U(2n + 1) (8.387)
The left electrode is taken to have zero voltage, while the right one has a voltage of V. The
quantum dot has a voltage of VQD . The voltage differences VQD and VQD - V are the
differences in the chemical potential between the left electrode and the QD, and between the
QD and the right electrode. The factor of(2n + 1) = (n + Ii -
n2 is the difference between
the charging energies before and after a hop that adds an electron to the QD. The factors of
Nv NR, N QD are the density of states at the chemical potential for the left and right electrodes,
and for the QD.
As an example, consider the subsystem of just the QD and the left electrode. The rate
equation for the process of having electrons hop between the QD and the electrode is
The first term on the right of the equals sign is from events where the system starts in state n
and changes by having an electron hop either on to or off of the QD. The remaining two terms
are events where the QD starts with either n ± 1 electrons and gets to n by a hop. In equi-
librium, the average rate of change is zero. Set the right-hand side of the above equation to
zero. This step can be accomplished with the identities
fnRL(n + 1, n) =/"+IRL(n, n + 1) (8.389)

/"_IRL(n, n - 1) = /,.RL(n - I.n) (8.390)
Adding these two equation gives the right-hand side of (8.388). However, the above two
equations are actually identical. The second is the same as the first by changing n ~ n - 1
everywhere. Only consider the solution to the first equation, which can be manipulated to
write it as
(8.391)
(8.392)
The second equation above is the solution deduced from the first. The result is expected for
thermal equilibrium.
The case for the QD tunneling is different. There is a voltage Vacross the device, and a
current flowing, so that the system is not in thermal equilibrium. The relationship among the
factors of fn must be derived anew. The rate equation for this case, including both electrodes,
IS
+ RR(n - 1, n)] + fn+l [RL(n, n + 1) + RR(n, n + 1)]

+ fn-l[R L(n, n - 1) + RR(n, n - 1)] (8.393)
In steady state, the average derivative is zero, and the right-hand side vanishes. As in the
above case, this constraint is accomplished by
RL(n + 1, n) +RR(n + 1, n) (8.394)

RL(n, n + l)RR(n, n + 1)
The above equation determines the occupation numbers fn for the quantum dot. Another
important equation is for the current. One can count either the electrons leaving the first
electrode, or else those arriving at the second one. These two expressions for the current are
h = eL fn[RL(n + 1, n) - RL(n - 1, n)]

n
=eL [fnRL(n + 1, n) - fn+lRL(n, n + 1)] (8.395)

JR = e L fn[RR(n - 1, n) - RR(n + 1, n)]
n
=eL [fn+lRR(n, n + 1) - fnRR(n + 1, n)] (8.396)

n
In the second equal sign we changed n -+ n + 1 in one of the terms, which is permitted since
there is a summation over n. In steady state the two currents are identical: h = JR' This
identity is automatically satisfied by Eq. (8.394).
As a simple example, consider the case that the charging energy U = O. In that case the
solution to the rate equations is that
~LEL = -~RER (8.397)

~LeVQD = -~Re(VQD - V) (8.398)
~R
VQD = V~ ~ (8.399)
SR + SL
Putting this result into Eq. (8.394) gives that the ratio fn+llfn = 1. All of the occupation
numbers fn are the same. An evaluation of the current gives that
(8.400)
This result is the same as that obtained by two resistors in series. A simple tunnel junction acts
like a resistor. A series of two junction behaves as two resistors in series. The net resistance is
the sum of the two resistances. In the present problem, the resistance for each connection are
rL ,R = 1/(e2~L ,R)' The above formula shows that the net resistance of the two-step tunneling
is the same as adding the two resistances. This result is expected once the energy levels in the
QD are treated as continuous, and the charging energy U is neglected.
The Coulomb blockade is obtained by including the charging energy. The example is
simplified by assuming that the two electrodes are identical, so SL = SR' Furthermore, it is
assumed that the QD is exactly midway between the two electrodes. Then one must choose
VQD = V /2 by symmetry. For each value of V the ratio (8.394) is solved for the relevent
number of n-values. Define this ratio as r(2n + I). By normalizing all occupation numbers to
10, we find for the partition function and the current
ZQD = 10 {I + r(l)[1 + r(3)(1 + ... )]
+_1_[1 +_1_(1 + ... )] (8.401)

r(-l) r(-3)
1= 10 {RL(I,0)-RL(-1,0)+r(1)[RL(2, l)-RL(O, 1)]

ZQD
+ r(~1)[RL(0'-I)-RL(-2,-1)]+"'} (8.402)
A typical result from this formula is shown in Fig. 8.15. The current is an asymmetric
function of the voltage. The onset of current begins at el VI = 2U since the current can only
occur if el VI/2 > U. Electrons can only hop from either electrode on to the QD if the voltage
difference supplies the charging energy U Since the QD is in the middle of the space between
the electrodes, a voltage of V gives only V/2 between the QD and either electrode. Note that
there is a second weak threshold when eWI/2 = 3U at eWI = 0.6eY. This point is the
threshold for having two extra electrons on the QD: the factor of three is 22 - 12.
Since the geometry is perfectly symmetric, the current can also proceed by the
generation of holes (n < 0). In this case, the first step is for an electron to hop from the QD to
-1 ·0.5 o 0.5
V(volts)
FIGURE 8.15 Current-voltage characteristic for a quantum dot placed symmetrically between two electrodes.
U = 0.1 eVand T = 10K.
the left electrode, making a hole on the QD. The second step is for an electron to tunnel from
the right electrode to the QD, which fills this hole state.
This example is typical for the case that the QD is large enough that its electron states
can be considered to be continuous. A much different case, for a small QD, is when the
energy levels are large compared to the thermal energy. Then the conductance is a series of
steps, where each step is another available energy state on the QD.
An interesting experiment has a magnetic impurity in the QD (Goldhaber-Gordon et al.,
1998). The tunneling becomes a type of spectroscopy which can measure the density of states
of the magnetic impurity. The experiment allows a test of solutions to the Anderson model.
8.6.4. Quantum Conductance

Ballistic transport is the case where an electron carries current without any scattering
(Landauer, 1981, 1989). There are several relevent length scales: the length L of the sample,
and the mean-free-path A of the electron. Earlier in this chapter it was shown there are
separate values of the mfP for: momentum scattering, energy scattering, electron-electron
interactions, etc. Here the most relevant mfP is that for momentum scattering (At), since it
enters into the definition of the electrical conductivity. If L < At then the current is carried
across the material without much chance of scattering. In this case the transport is called
ballistic.
A special formula is needed for the current from ballistic transport. One cannot employ
the Boltzmann equation or the Kubo formula, since they all assume a system with quasi-
equilibrium, which means that the mfP is smaller than the dimensions of the sample. Instead,
the correct picture is from simple quantum mechanics. Consider a particle moving to the right
in one dimension, with a wave vector k. The number of particles per unit distance with wave
vector k is
(8.403)
The Fermi-Dirac occupation number is for the source of electron on the left electrode.
Assume there is another electrode on the right which is at a voltage V with respect to the one
on the left. Then the number of electrons going from right to left is given by
(8.404)
The net current is obtained by subtracting these two formula, and multiplying by ev/o where Vk
is the velocity
1= ;n fdkvk[n/~k) - nF(~k + eV)] (8.405)
= Jd~dnf(~k)
2:1i - nF(~k + eV)] (8.406)
1= aoV (8.407)
r?
ao = - (8.408)
h
The quantity e2 /h is the quantized conductance. Its units are interesting. In cgs units, e2 /lic is
dimensionless, so that r? / h has the units of velocity. That is not the correct result here.
Instead, in SI units e has the units of Coulombs, and Planck's constant has the units of Joule-
second, which gives that 0'0 has the units of Siemens, which is the inverse of the Ohm. The
approximate value is 0'0 = 38.8 J..LS.
The above derivation applies to a single channel of electrons. Consider the conduction
down a quantum wire. Treat the wire as a type of wave guide, which has many conducting
modes, although each new one occurs at higher energy. The number of conducting modes as a
function of energy is defined as Nc(E). This quantity is a series of step functions as each new
mode opens up along the quantum wire. Then the current is actually given by the expression
(8.409)
The factor of two in front is for spin degeneracy. Each orbital mode usually has spin
degeneracy. This symmetry is broken in a magnetic field. Some writers prefer to define 20'0 as
the quantum of conductance.
PROBLEMS
1. Use the piezoelectric electron-phonon interaction in Sec. 1.3 to calculate the temperature
dependence of the electrical conductivity in a semiconductor.
2. Consider the self-energy of an electron from unscreened exchange,

1
L(p) = A LVq L~(P + q, ip) (8.410)
pV q ip
which was evaluated in Sec. 5.1.6. Derive the Ward identity for this self-energy-the equivalent of
(8.81). Use this result to show that the Coulomb ladder diagrams have negligible effect upon the basic
polarization p(1)(q, iro) in the limit where q ~ o.
r
3. The correlation function
x(q, iro) = - dr:e iOlt (Ttp(q, r:)p( -q, 0)) (8.411)
vanishes when q ~ O. Show that the bare bubble p(1)(q, iro) has this feature. Use the Ward identity to
show that it still vanishes when self-energy functions and vertex functions are included in the evaluation.
4. Derive the rate of temperature relaxation in a semiconductor due to the deformation potential
scattering of electrons by acoustical phonons. Give the answer for high and low temperatures. Evaluate
the numerical value of the lifetimes using the data in Table 8.1.
5. Use polar coupling to optical phonons to calculate the rates of relaxation for an electron in a
semiconductor: (a) for momentum, (b) for energy, and (c) temperature.
6. Use deformation coupling to optical phonons to calculate the rates of relaxation for an electron in a
semiconductor: (a) for momentum, (b) for energy, and (c) for temperature.
7. Use (8.228) to evaluate y(e + if>, e + if», and show that it equals unity, in agreement with the Ward
identity.
8. Derive the Ward identity (8.219) for the phonon ladder diagrams.
Problems 577
9. Solve the scalar vertex equation (8.219) using the same techniques used to solve
y(p, e + iO, e + iO). Show that r(e + iO, e + iO) = 1 - a"i./ae but that r(e + iO, e - iO) does not obey
the Ward identity. Furthermore, show for every solution R(e) = r(e + iO, e - iO) to its vertex equation
that R(e) + a Im["i.(e)] is also a solution, where a is an arbitrary constant.
10. Write down the collision rate (at/at) for electron--electron interactions. Evaluate it for a
semiconductor where the few conduction electrons obey Maxwell-Boltzmann statistics. Use Debye
screening for the dielectric function.
11. Show that the leading term in the thermopower for a free-electron gas is
1[ ldT
2
S= ___ B_
(8.412)
2eEF
12. Evaluate the correlation function L(22) for the thermal conductivity of a metal. Use the free-electron
heat current operator, and show that the integral diverges for the first term in the vertex summation.
13. Solve Eqns. (8.215) and (8.235) in the limit of small temperature and show that the resistivity
p ~ T5 from phonons. Assume that rx.;F(u) = gu2 at small u, where g is a constant.
14. Compare the relaxation time for transport deduced from the force-balance theory (8.198) with that
given by (8.236). How do they differ? Show they become identical in the limit of high T.
15. Write down the rate equations for current between two electrodes separated by two identical
quantum dots in series. What is the minimum voltage for current to flow at zero temperature?
Chapter 9
Optical Properties of Solids
9.1. NEARLY FREE-ELECTRON SYSTEMS
9.1.1. General Properties

This section is concerned with the optical properties of free-particle systems, such as elec-
trons in metals or semiconductors. A completely free-electron system does not absorb light at
all, so that its optical properties are uninteresting. The ability of the nearly free-particle system
to absorb light is due to its imperfections or deviations from homogeneity. If these effects are
small, then so is the light absorption. This situation is described here. The optical properties
are calculated for simple metals and semiconductors. The best way to do this is using the so
called force-force correlation function.
First it is useful to recall some of the fundamental formulas regarding the optical
properties of solids. As shown in Sec. 4.5.3, the transverse dielectric function of the solid in
the limit of long wavelength is
(9.1)
(9.2)
where n(O)) is the retarded form of the current-current correlation function n(iO)). The optical
wave vector k is so short that the correlation function is evaluated in the limit where k -+ O.
The results for k -+ 0 are sometimes different from those for k = 0, in which case the limit
result is appropriate. The k symbol is omitted in the notation. The system is assumed to be
isotropic. The index I.l is any direction (x, y, z), since the average value of Uxjx} is equal to
that of (j • j) /3. The conductivity 0'(0)) of the solid is given by
4niO'( 0))
c..L (0)) = 1 +------'-----'- (9.3)
0)
() ine2 in( 0))

0' 0) =mO)
- + -0)- (9.4)
579
580 Chap. 9 • Optical Properties of Solids
with the usual identity Re( cr) = - Im[n( ro / ro]. The relationships are also valid for the real part
of n(ro) which contributes to the imaginary part of the conductivity. These general formulas
will be the basis for most of our discussion.
A general theorem which can be proved is the f-sum rule,
£-1(ro) = £,(ro) + i£2(ro) (9.5)
Jooo rodro£2(ro) = nro22 P (9.6)
The similar theorem for the longitudinal dielectric function was shown in Sec. 5.7. The
theorem for the transverse dielectric function can also be written in terms of the conductivity
since £2 = 4n Re( cr fro):
Jo droRe[cr(ro)]
OO
=
nn
2~
e2
(9.7)
The real part of the conductivity is the absorbing power of the solid. This theorem states that
the absorbing power depends only on the density of particles no and not on the detailed form
of the interactions. Many-body theory can rearrange the absorbing power by moving it from
one part of the spectrum to another, but it cannot alter the net amount. The statement in the
opening paragraph, that the light absorption is small if the perturbations are small, needs
clarification. Insight is provided by examining the Drude formula.
The Drude formula is an empirical relationship which is found to be valid, over a
restricted frequency range, in most nearly free-electron systems. It states that
n(ro) = _ noe2 1. (9.8)

m 1 - 1roto
no e2t o (9.9)
Re[cr(ro)] = (1
m + ro2to2)
The quantity to is a relaxation time, which is usually taken to be a constant. Assume that the
Drude formula is valid for all frequencies and with a constant relaxation time. Then the f-sum
rule is satisfied independently of the size of the relaxation time:
no e2t o Joo _ nnoe2

JoOOdroRecrro
[()] _
-
mol
dro
+ (roto)
2-
2m
(9.10)
Re[cr(ro)] is a Lorentzian distribution, whose width is l/to and whose height is to, so that the
area under the curve is independent of to.
The discussion of dc conductivity showed that inverse relaxation times are proportional
to scattering rates. The purely free-particle system, without any inhomogeneity, corresponds
to the case where l/to = 0 or Re[cr(ro)] <X o(ro). The scattering processes contribute to to. If
the Drude expression were rigorous, one would only need to find to, which in fact is finding
l/to. The limit ro -+ 0 in (9.9) gives the dc conductivity Re(cr) = noe2to/m.
If the scattering rate l/to is small, then to is large, and the dimensionless product roto is
large except at very small frequencies. Expand (9.9) for roto » 1 and find
4nRe[cr(ro)] = ~~
ro to
[1 + 0(_1_ (roto)
2)] (9.11)
Sec. 9.1 • Nearly Free-Electron Systems 581
At higher frequencies the real part of the conductivity is proportional to the scattering rate
1/1"0' which is small by our assumption. This result is what is meant by the limit of weak
scattering. The method of force-force correlation function casts Re[ 0'(00)] in this form, as an
expansion in the scattering rate. The resulting theories are valid at high frequencies where
001"0 » 1.
9.1.2. Force-Force Correlation Functions

The first theorem to be proved has already been mentioned, namely that the homo-
geneous electron gas does not absorb any light. This assertion is froved by showing that the
Hamiltonian commutes with the momentum operator P = .L kCk,crCkcr
_ t ~ t t
Ho - .L
pcr
~pCpcrCpcr + 2 .L.L VqCk+q,cr,Cp_q.crCpcrCkcr'
Y pkq crcr'
(9.12)
d
0= d/ = i[Ho, P] (9.13)
The momentum operator has no time dependence and is a constant of motion.

The absorption oflight is described by the Kubo formula, which is the correlation of the
momentum operator with itself:
Re[O'(oo)] = 6~Y C_;-~W) J~oo dte iW( (P(t) • P(O)) (9.14)
where the current is proportional to the momentum jet) = (elm)P(t). The lack of time
variation in P means that the real part of the conductivity is identically zero for any nonzero
frequency. There is no light absorption for 00 f. O.
A simple way of deriving the force-force correlation function was given by Hopfield
(1965). The argument proceeds in two steps. The first step is to replace pet) by the identical
operator (i 100 )dP1dt, which is justified on the grounds that in an ac field of frequency 00 the
time dependence of interest is just exp( -ioot). This replacement is done for both P operators,
which gives the expression
Re[O'(oo)] = e'- (1 - e-~W)Joo dteimt(dP(t) • [dP(t')] ) (9.15)

6m 2 oo 3 y -00 dt dt' ('=0
The second step in the argument is to notice that the time derivative of the momentum is just
the force F(t) = dPldt, so the force-force correlation function is
Re[O'(oo)] = e'-
2 3 (1 - e-~W) Joo dte iw( (F(t) • F(O)) (9.16)
6m 00 y -00
This derivation is essentially correct and will be justified later with more rigor. The force-
force correlation function then serves as another possible starting point for the evaluation of
the conductivity. For free-particle systems, it usually leads to a usable answer with much less
work than the current-current correlation function, because the first term in the perturbation
expansion of the force-force correlation function is usually the answer. For example, if the
force is due to phonons, then F ex MA(q), where M,Jq) is the electron-phonon matrix
element. The leading term in the correlation function is proportional to Mf, which is often the
largest term. The remaining parts of the correlation function are evaluated assuming the
particles are free, which is quite easy. Similarly, if the absorption is due to impurities, the
leading tenn is proportional to concentration ni of impurities, which is often the important

tenn. The advantage of the force-force correlation function is that the leading tenn is often
adequate.
One calculation for which the force-force correlation function would be a poor starting
point would be the de conductivity. The correlation function has a prefactor of co- 3 , and
obviously a great deal of cancellation in the correlation function has to take place in the limit
co --+ 0 in order to get (J' = noil'to/m.
A rigorous derivation of the force-force correlation function was given independently by
Mahan (1970) and Hasegawa and Watabe (1969). The trick is to start from the Kubo fonnula
and integrate by parts on the 't variable:
(9.17)
(9.18)
(9.19)
The integration by parts gives
n(ico) = - ~
lCOY
[(ji~)J~(O) - JiO)J~(O)]} + -.lCOY
1_ f~ d'teiffint(T/Jd'i't) JiO))
0 't
(9.20)
The first tenn is the constant of integration. These constants have the general fonn of
commutators:
(A(~)B(O) - A(O)B(O)) = Tr(Ae-~HB - e-~H AB) (9.21)
= Tr[e-~H(BA -AB)] = -([A,B]) (9.22)
where the cyclic properties of the trace are used to rearrange the first tenn. The one in (9.20)
is zero since it is the commutator ofJiO) with itself.
The next step in the derivation is to take the integration by parts on the other current
operator JiO). Since the correlation function depends only on the difference of the two 't
values in its argument, this correlation function is also equal to (Tr[(dJ~('tI)/(d't')t'=oJ~( -'t)).
The integration by parts on the tenn J~ (-'t) gives
n(ico) = ~(rJ~(O), [dJd~('t)J J)

(ICO) Y l 't t=O
-~f~d'teiffint(Tt[dJ~('t)J [dJi~')J
(ICO) Y 0 d't d't t'=O
) (9.23)
The first tenn contains the commutator ofJ~ with dJ~/d't. This commutator yields a constant,
which is real and so does not contribute to the real part of the conductivity or to the
absorption of light. But Simanek (1971) has shown that this tenn does contribute to the
renonnalization of the optical mass so it is retained. The other tenn in (9.23) is the frequency-
dependent part, which is indeed the correlation function of dJ / d't with itself. The current is
just proportional to the momentum j = (e/m)P, so (9.15) is derived again.
The Hamiltonian is written as H = Ho + V, where Ho is the homogeneous electron gas
in (9.12), while V is the potential which causes other forces besides electron-electron inter-
actions. For example, the potential could be the sum of the interactions with the crystalline
potential VG where the G are reciprocal lattice vectors, with impurities of density operator
Pi(q) (see Sec. 4.1.5), or with phonons:
V= L p(q)«l>(q) (9.24)
q
(9.25)
The time derivative of the current operator is
(9.26)
(9.27)
(9.28)
(9.29)
The last commutator is just Newton's law: the time derivative of the momentum is the force,
which is the gradient of the potential
FIl(r) = -VIlV(r) = -iLV(q)qlleiq ' r (9.30)

q
The evaluation of the constant term in (9.23) requires the additional commutator
(9.31)
(9.32)
Using these results in (9.23) gives
1t(iro) = il. 2 [- Lq;(P(q)«l>(q)(q») + Lqllq~J(3 d'te iffint
vm 2 (lro) q qq' 0
x (TtP(q,'t)«l>(q, 't) p(q' , O)«l>(q/, 0))] (9.33)
The generalized potential «l>(q) is given a 't dependence, which applies only to the phonon part
of the three terms in (9.25). The second term is the force-force correlation function, which is
equivalent to (9.16). It is often sufficient to evaluate this expression only to order «l>2. The
second term is already evaluated to this order, so that the remaining correlation function is
evaluated ignoring the interaction V. In this case q' = -q. The first term in (9.33) must also be
evaluated to order (J>2 by expanding the S matrix for the interaction potential V:
(p(q)(J>(q)) = Tr[e-~HOS(~)p(q)(J>(q)]
= Tr[e-~HOp(q)(J>(q)] - J: d"C(TtV(t)p(q, O)(J>(q, O)} + O«(J>3) (9.34)
Only the term in V is retained in order to have a result proportional to (J>2. The first term on the
right o(p(q)(J>(q)}o is zero. Combining both terms in (9.33) gives
1t(ioo) = - e2.
vm 2(lOO)
2 [Lq~J~ d"C(eiOlt -
q 0
1)(Ttp(q, "C)p(-q, O»)(Tt(J>(q, "C)(J>(-q, 0)}](9.35)
The term with exp(ioo"C) comes from the force-force correlation function, while the -1 term
comes from the constant term.
This formula will be evaluated for several different types of potentials in the remaining
parts of this section. The type of formula which results can be illustrated by a simple example.
Take the case where the potential (J> is independent of "C. One example is the scattering from
impurities, where, according to Sec. 4.1.5,
(9.36)
(9.37)
where V;( q) is the unscreened potential between the electron and the impurity. The remaining
part of the "C integral in (9.35) is the definition of the longitudinal dielectric function from Sec.
5.4:
1J~
- -
v 0
d"Ce iron t (Ttp ( q, "C ) P( -q,O ) ) = 4q2_,
1te-
[I
e
( .) - 1
q, lOO
] (9.38)
c) ni
1ti lOO = 41tm2v(iooi
2 2 V( i
Lq qllq iq
[1 1]
e(q, ioo) - e(q)
(9.39)
The subscript i denotes the impurity contribution to 1t(ioo). The current-current correlation
function is expressed as the difference between l/e(q, ioo) and I/e(q, 0). Since e(q, ioo) is an
even function of ioo, in the limit where ioo --+ 0 it must behave as e(q, ioo) = e(q, 0) + O(ioo i,
which shows that 1ti does not diverge in this limit. The retarded function is found by
ioo --+ 00 + iO:
ni ,,2 2 2[ I 1 ] (9.40)
1ti(OO) = 41tm2vr02 Lqqllq V;(q) e(q, (0) - e(q)
3
Re[crloo)] = - 4 ni2 3 -d 3
1tm 00 (21t)
J
q qllq
2 2
V;(q) 2 1m [ - (1- )]
e q, 00
+ O(V;3) (9.41)
Equation (9.41) is the Hopfield (1965) formula, which shows that the real part of the
transverse conductivity is proportional to the imaginary part of the inverse longitudinal
dielectric function. The latter quantity is interpreted as the rate of making excitations in the
electron gas. The light absorption occurs because electronic excitations are created, while the
creation rate must depend on the concentration of impurities ni in the system. This inter-
pretation agrees with our original assertion that the electron gas can absorb light only when
the impurities, or other inhomogeneities, are present to dissipate the momentum. An obvious
advantage of the Hopfield formula is that the complicated aspects of electron-electron
interactions are naturally included in the dielectric function, for which one can use the model
forms described in Chapter 5.
9.1.3. Frohlich Polarons

The optical absorption of free polarons was first calculated by Gurevich et al. (1962),
who started from the usual form of the Kubo formula. In the Frohlich Hamiltonian (7.1),
(9.42)
M.2 _ 41tow)03/2
0 - -/2ni
the electron-phonon interaction V is expanded using the usual S-matrix techniques. The real
part of the conductivity is calculated to order O(a), which is the first nonvanishing term. The
three diagrams which enter proportional to a are shown in Fig. 9.1. The first two are self-
energy diagrams, while the third is a vertex contribution. The final result is obtained by
adding contributions from all three diagrams. At zero temperature it is
(9.43)
The free polaron can absorb light only when the frequency co > coo. Polarons can emit LO
phonons only when the polaron energy exceeds the phonon energy coo. The light wave excites
the polaron to this energy, and then the polaron can emit the phonon. The light absorption
happens by the emission of LO phonons, where the electron serves as the intermediary in the
process. Of course, there is also a direct coupling between the photon and the TO phonons,
which was discussed in Sec. 4.6 on polaritons.
Formula (9.43) can be derived from the correlation functions in (9.35). The part
involving the potential is now just the Green's function for LO phonons:
(9.44)
(T,<lJ(q, 't)<lJ(-q, 0)) = _ M~ .@(O)('t) (9.45)

vq
FIGURE 9.1
Denote the density-density correlation function by the symbol X(q, 't). The 't integral in
(9.35) is evaluated by changing it into a summation over Matsubara frequencies, which is
evaluated by the usual contour integral:
=
.
x(q, 't) -(T,p(q, 't)p( -q, 0)) (9.46)
n(iro) = -
M,2€?
? 2 L -q2- ':':. J~ d't(e 'ffi ' - l)X(q, 't).@(O)('t) + O(Mg)
m2(lffi) v q q2 0
M,2€? q2 1
=- ? 2 L ~ R Liq .@(O)(iq)[X(q, iro + iq) -
m2(lro) v q q I-'
X(q, iq)]
M,2€? q2 JOO dE
=- O 2 L ~ -nB(E){2Im[D(O)(E)]X(q, iro+ E)
2
m (iro) v q q -00 2n
+ 2 Im[x(q, E)].@(O)(E - iro) - 2 Im[D(O)(E)x(q, Em (9.47)
Take the retarded function iro --+ ro + ib, and the real part of the conductivity is the imaginary
part of the preceding expression, where the last term does not contribute:
2MJe
Re[cr(ro)] = - -J-
2 d 3- q~ JOO -nB(E)
q - dE (9.48)
m 2 ro 3 (2n)3 q2 -00 2n
x {Im[D(O)(E)]Im[x(q, E + ro)] - Im[x(q, E)]Im[D(O)(E - rom (9.49)
This result is general and applies to all electron-phonon systems. Sometimes it is useful to
change variables E --+ E + ro in the second term, which gives the equivalent expression
22
2M, e J d q q2 JOO 3
Re cr(ro) = -T---J- --3 ~
dE
- [nB(E) - nB(E + ro)] (9.50)
m ro (2n) q -00 2n
x Im[D(O)(E)]Im[x(q, E + ro)] (9.51)
The spectral function for the phonon Green's function is easy to evaluate:
(9.52)
The last step in the formal derivation is to determine X(q, ro). The free polarons are assumed
to have small concentration no, so that electron-electron interactions can be neglected. Then
the density-density correlation function is adequately approximated by the simple bubble
diagram:
X(q, iro) = p(1)(q, iro) = -2 L q;(O)(pY!J(O)(p + q) (9.53)

V~p,ip
(9.54)
The factor of 2 in front is for the spin degeneracy. The correction terms to this simple bubble
term, from the electron-phonon interaction, contributes to higher-order terms in lI..
The polarons are assumed to obey Maxwell-Boltzmann statistics, so the particle density
np per spin state is
no (27t~)3/2
np =2- - m exp[-~E]
p
(9.55)
The present calculation is for zero temperature. The particles all approach the state with
p ---+ 0, so from (9.54) the spectral function is
(9.56)
Now evaluate the real part of the conductivity given in (9.51). Both spectral functions are
given by delta functions, for phonons in (9.52) and for electrons in (9.56). One set of delta
functions are used to eliminate the dE integral, which gives
Re[cr(ffi)] = 7t 2M.m~ t?n

2 Jd q q2 3
3 0 --3 ~ UnB(ffiO) - nB(ffiO + ffi)]
ffi (27t) q
X [O(ffio + ffi - Eq) - O(ffio + ffi + Eq)]
+ [1 + nB(ffiO) + nB(ffi - ffio)][O(ffi - ffiO - Eq) - O(ffi - ffio + Eq)]} (9.57)
At zero temperature, the boson occupation factor is nB(ffi) = -0( -ffi). This expression
simplifies for ffi > 0, cos2 {} = q~/q2 to
Re[cr(ffi)] = 7t 2M.~ t?n

2
m ffi
3 J
3 0 -d- q3 cos2 {}O(ffi - ffiO - Eq)
(27t)
(9.58)
2 n
= 7t 2M.0 t? o3 (27t)
-3 (2m) 3/2 .Jffi - ffi o0(ffi - ffiO) (9.59)
m2ffi3(27t)
Using the form for the electron-phonon matrix element M1; gives the result (9.43) of
Gurevich et al. The derivation has been remarkably short. The general form of the answer in
(9.51), as the product of two spectral functions, is quite familiar from other calculations, and
in this case both spectral functions are delta functions.
The optical absorption of Frohlich polarons has been calculated extensively by Devreese
(1972). He calculated the terms to order 0(0(2), which are the two phonon terms. The polaron
can be excited by the light and make two phonons as well as one. Results are shown in Fig.
9.2 for the value of the coupling constant 0( = l. The solid line is the result using (9.43) which
includes just the one-phonon processes, while the dashed line is the result including both one-
and two-phonon processes. The rather surprising result emerges that there is no sudden rise in
the absorption at the two-phonon threshold ffi = 2ffio. The sharp rise at ffi = ffio is not repeated
as a sharp rise at 2ffio. Instead, there is a very gradual rise starting at 2ffio and a slight decrease
in the peak height. The two-phonon events make only a small change in the result. The
polaron would rather emit two phonons as two separate events of one-phonon emission and
not as a correlated event where two phonons are emitted. This conclusion is in accord with the
remarks in Sec. 7.1 about the correlation between two phonons being small in the polaron
cloud. Devreese (1972) has also calculated the optical absorption ofpolarons by the Feynman
method of path integrals as well as by strong coupling theory. At larger values of polaron
constant 0( he finds sharp absorption lines, which are interpreted as arising from internal states
of the localized electron in the small polaron.
,.... -Iphonon
015 I ~
I ~
I
I
~
~
,
~ 01
l-
lL
~
CD
«
0.05
°o~--------~------~------~--~
.. 1... - - - -
FIGURE 9.2 Theoretical prediction for the optical absorption of free polarons with 0( = I and T = 0 K. The solid
line is one-phonon theory, and the dashed line has two-phonon correlations. Source: Devreese (1972, p. 93).
9.1.4. Interband Transitions

Another application of the force-force correlation function is to calculate the rate of light
absorption by interband transitions in a metal. Here the potential which provides the inho-
mogeneity is the crystalline potential of the atoms. This potential is periodic in crystalline
solids, so it contributes momentum in discrete units called reciprocal lattice vectors G. This
contribution was mentioned earlier as the first term in (9.25):
(9.60)
This potential is fixed and has no correlation with itself: (cI>(q)cI>(q')} = (cI>(q)}(cI>(q')}.
However, the other correlation function in (9.35) is (p(q, "C)p(q', O)} and is nonzero only when
q=-q':
(9.61)
This potential is static, so the correlation function of the potential is a constant. This case is
similar to that of impurity scattering, which was given in (9.41). The impurity result can be
used for the crystalline potential by changing nj V;(q)2 to V~()q=G' The interband absorption is
Re[oAO))] = - 4 1L
2 3
1tm 0) G
GIl222 [1] +
G VG 1m - ( G)
E ,0)
3
O(VG) (9.62)
The subscript I stands for interband contribution. Note the following:
(9.63)
where 1/IE12 is usually grouped with V~ to have a screened electron-ion interaction

IVG/E(G, 00)12. If the RPA result in Sec. 5.5 is used for E2(G, 00), then when G > 2kF ,
E2(G,oo) = ~7 [~ - (~r (00 - EGiJ

(9.64)
This fonnula is nearly identical to a standard fonnula for describing interband transitions in
metals: the Wilson-Butcher fonnula.
The Wilson-Butcher fonnula is used to describe the interband transition in the alkali
metals; Figure 9.3(a) shows the band structure in an alkali metal in the k = (110) direction.
The Fermi level is shown as the dashed horizontal line. The interband transitions are shown,
where the transition appears vertical in wave vector space since the wave vector of light is too
small to show up as a horizontal deflection of the arrow. This transition is shown in the
reduced zone scheme. Of course, in the extended zone description of Fig. 9.3(b), the optical
transition really changes the wave vector of the electron by a reciprocal lattice vector G. This
change in the momentum of the electron, during the optical transition, is precisely what it
needs to have the transition. The optical absorption occurs because the electron gains energy
from the light wave and momentum from the rigid crystal lattice.
The optical absorption one expects from an alkali metal is shown in Fig. 9.4. At low
frequency there is a Drude contribution, which comes from phonons. This contribution is well
described by the Drude fonnula (9.9), where 'to is largely provided by the intrinsic scattering
from phonons. The interband transition of Fig. 9.4 starts about lioo = 2 eV, with a linear rise at
threshold. Since the Fermi surface does not touch the zone face, kF < G/2, and (9.64) is the
appropriate fonn to use for E2(G, 00). Figure 9.4 shows the experimental result of Smith
(1969) for sodium, which has exactly this behavior. Similar interband tenns were found by
Smith (1970) for K, Rb, and Cs and in Li by Myers and Sixtensson (1976). The agreement
between theory and experiment appears satisfactory for sodium. There is no reason to expect
the Wilson-Butcher theory to give perfect agreement, since there are additional tenns which
are higher powers in O(ViD. Similarly, most calculations of the Wilson-Butcher fonnula use
the RPA fonn for E(G, 00), whereas, the Singwi-Sjolander fonn would be an improvement.
There is also a slight uncertainly as to the precise value of the pseudopotential VI 10· With
these qualifications in mind, the agreement between theory and experiment in Fig. 9.4 is
satisfactory.
There is one difference between the original fonnula of Wilson-Butcher and the result
(9.64) derived from the force-force correlation function. The original fonnula had the
( 0) (b)
G12 " (110)
FIGURE 9.3 Interband transitions in alkali metals as shown in (a) reduced zone and (b) extended zone
representations.
3r----,------------------------------------,
cdw) SODIUM
- DRUDE
'VO EXPERIMENT
- - INTERBAND
2
.!
b I
8 0 0 iJ '00 8
o '0
4
1\", (eV)
FIGURE 9.4 The optical conductivity for metallic sodium. The points are the experiments of Smith, while the lines
are the theoretical contribution from Drude and interband transitions. Source: Smith (1969).
electron ion potential divided by the static dielectric function s(G, 0). The present derivation
by the force-force correlation function is exact to order V~, so its factor ofs(G, co) is correct,
and the old formula is not. The difference is more philosophical than numerical for sodium,
since the frequency dependence of s(G, co) is slight over the range of frequencies shown in
Fig. 9.4. Actually these frequencies, are not much lower than the plasma frequency at
Jjcop ~ 5 eV, but this fact is irrelevant, since at these large values of wave vector G the
dielectric function becomes frequency dependent only at much larger values of co.
The alkali metals are good examples of where the Wilson-Butcher formula works well.
Here the lowest interband transition is in the visible region of the spectrum and is isolated in
frequency from any other feature except the Drude contribution, which is always present.
Ashcroft and Sturm (1971) have shown that another formula is needed for polyvalent metals
such as aluminum, since there are transitions between parallel bands which cause sharp
structure at lower photon frequencies.
9.1.5. Phonons
Equation (9.25) lists three contributions to (q). The first two terms are Bragg scattering
and impurity scattering, which have been discussed. The third term is treated here. It is due to
the electron scattering by the phonons.
Equation (9.35) is evaluated, with (q) being the electron-phonon interaction. Only
terms are retained that are of O(Mi} This point is rather tricky, since a subset of higher-order
contributions are also retained. It is assumed that the phonon modes are the actual modes in
the solid. The calculation of these modes includes electron-electron and electron-phonon
interactions, which means they include M).... Any time the actual phonon modes are used, then
one is including a subset of higher-order diagrams.
In Eq. (9.35) the phonon part is

1
-(T~(q, 't)( -q, 0») = - LMt~(q, 't) (9.65)
v A,
Equation (9.35) is now evaluated. Only keep the term that depends upon frequency iro:
't(iro) = -
2
e-
L q2MA,(qi
Il
1
L~A,(q, iq)
[1
. . - 1
]
vm (iroi qA, Vq ~ iq e(q, 1ro - 1q)
The summation over Matsubara frequencies is evaluated using the techniques of Sec. 3.5. The
phonon Green's function is expressed in the Lehmann representation using its spectral
function B(q, ro). The inverse dielectric function is expressed similarly, with its spectral
function being Im(1/e)
't iro = - 122M 2 Jdro' B ,ro' Jde' Im[_I_J nB(ro') - nB(e')

() 47tvm2(iroi ~ qllq A,(q) 7t (q ) 27t e(q, e') iro + ro' - e'
(9.66)
The retarded function is obtained by taking iro --+ ro + iO. Then take the imaginary part.
These steps cause the denominator (iro + ro' - e') to be replaced by the numerator
7to(ro + ro' - e'). The delta function is used to eliminate the integral over de'.
This expression is usually evaluated at relatively low frequencies. When ro is much less
than the plasma frequency of the metal, use the low-frequency limit for e2(q, ro) =
2rom 2 e-q- 3 0(2kF - q). The various terms involving the phonon wave vector, matrix element,
and spectral function are just the definition of the transport form of rx;'F(u). The remaining
factors combine to give
Re[cr(ro)] = _ Im[7t(ro)] (9.67)

ro
27te- no JOO 2
= durxtrF(u){(ro - u)[nB(u) - nB(u - ro)]
°
--3-
mro
+ (ro + u)[nB(u) - nB(u + ro)]} (9.68)
At zero temperature, the boson functions nB arrange to limit the range of integration to
o < u < ro. This limit gives the simple formula first derived by Allen (1971):
27te- no Jeo 2
Re[cr(ro)] = - durxtrF(u)(ro - u) (9.69)
°
-3-
mro
The Allen formula is approximate, since it is based upon the force-force correlation function.
A more accurate expression would be based upon solving the Kubo relation for the
conductivity as in Sec. 8.4. This derivation gives the result
e-n_o
Re[cr(ro)] = __ Joo de[nF(e + ro) - nF(e)] (9.70)
mro -00
x Re[y(e - iO, e + ro + io)I(e, e + ro)] (9.71)
I( ee') - - , . - - -27ti
------ (9.72)
, - nee) - n(e') + i[ree) + ree')]
0.14
0.12
I
"'- ..... ..........
II " ,
0.08 I ~
I
or I
I
0.04 I
0~~0.~4~~,~.2~~~2.0
FREQUENCY (w)
FIGURE 9.5 Phonon-induced absorption in calcium as a function of frequency. The vertical runs IS
Re[cr(ro)mro/e2no]. The horizontal axis is ro/roD' where the Debye frequency is roD' The solid line is the solution
to the ac transport equation, while the dashed line is using the Allen fonnula from the force-force correlation
function. The latter method is approximate. Source: Wu and Malian (1984).
This result can be derived directly from (8.204) and (8.213). The vertex function
r(1': - if>, I': + ro + if» obeys the integral equation (8.228). A comparison of these two
expressions is given in Fig. 9.5 for metallic calcium with Ellc. What is plotted is Re(O'/O'o) vs.
ro/ron, where 0'0 = noe? /mro. The solid line is the exact result from (9.70) while the dashed
line is the Allen formula in (9.69). The Allen formula has an error of about 10-15%. This
makes it a good approximation, since (9.69) is very much easier to evaluate than (9.70). The
latter requires solving numerically an integral equation for the vertex function.
9.2. WANNIER EXCITONS
9.2.1. The Model

Exciton states play an extremely important role in the understanding of interband
transitions in semiconductors. The word exciton is used here to signify the modification of the
absorption rate of photons due to the Coulomb interaction between the electron and the
valence band hole. The easiest case to understand is when the lowest interband transition is
direct: which means that the conduction band minimum and valence band maximum are at the
same point in k space. This configuration is shown schematically in Fig. 9.6(a). The valence
band states are all filled, and the conduction band states are all empty. The vertical arrow
shows a possible interband transition which can occur when a photon is absorbed in the solid.
The arrow starts in one of the valence states, since these are occupied by electrons, and goes
to an unoccupied state in the conduction band. The valence state is shown as nondegenerate
(except for spin) at k = 0, although that is seldom the case; usually the band has an orbital
degeneracy and is anisotropic.
The point of view in Fig. 9.6(a) is a single-particle picture of the transition process.
According to this picture, the transition rate for the absorption of photons is given by the
golden rule:
A(ro) = :1t L I(c, k'ie • plv, k) 12f>[l':v(k) + flro -

TlV kk'
I':cCk')] (9.73)
k2
I': (k) = - - (9.74)
v 2mv'
Sec. 9.2 • Wannier Excitons 593
CONDUCTION
(a)
VALENCE
(b) ~:~
FIGURE 9.6 Optical transition in a semiconductor between the occupied valence band and empty conduction band
for a direct transition. (a) Conventional band picture; (b) Wannier picture where the photon makes an electron-hole.
The energy zero is chosen to be the top of the valence band, so the bottom of the conduction
band starts at the energy gap Eg • The matrix element (e, k/le • plv, k) is between the one-
electron initial and final states. The wave functions are taken to be Bloch functions; which are
the product of a cell-periodic part unk(r) and an envelope function exp(ik· r). It is a
reasonable approximation, for our simple level of discussion, to take the cell-periodic parts as
independent of wave vector uck(r) = uc(r), uvk(r) = uv(r). It is also assumed that the valence
band has p symmetry and the conduction band has s symmetry, which is true in many
semiconductors but not all. Then uv(r) is a periodic orbital with angular momentum 1= 1,
while ucCr) is a periodic orbital with 1=0. With these approximations the optical matrix
element is a constant except for wave vector conservation:
eik • r
lv, k} = uvCr) v'v (9.75)
ik'· r
Ie, k/} = uc(r) e v'v (9.76)
(e, k/le • plv, k) = (ele • plv}0kk' == e • PcvOkk' (9.77)
(ele • plv) = ~J d3ru~e· pU v (9.78)

Vo cell
where Vo is the volume of a unit cell. The matrix element (elplv) should also contain the
factor exp(iq . r), where q is the wave vector of the photon. This wave vector is small enough
to be neglected for photons in the optical frequencies. Now that the matrix element is eval-
uated, it is possible to do the integrals over the electron wave vector:
A(ro) = 21t Ie • P 12 d k
" cv (21t)3
J o("ro -
3
E -
g
~)
2~
(2 )3/2
= Ie • Pcvl 2 ~1t Jliro - Eg0("ro - Eg) (9.79)
1 1 1
-=-+- (9.80)
~ me mv
Equation (9.79) predicts that the absorption rate begins at the energy gap of the semi-
conductor and rises as the square root of the factor (liro - E)g. It is a very definite prediction,
which is not observed. In fact, the one-particle theory is totally inadequate and does not come
close to describing the absorption spectra A(ro) observed in experiments.
Wannier (1937) observed that the interband transition in semiconductors was really a
two-particle process, which is indicated in Fig. 9.6(b). Since the valence band states are all
occupied, removing an electron from this band creates an excitation called a hole. This hole,
which is the absence of an electron in an otherwise filled band, can be treated as a particle
with an effective mass mv and a positive charge. The energy of the hole is Eh(k) = k2 j2m v' In
the two-particle picture, the photon of energy creates two excitations in the semiconductor:
the electron in the conduction band of wave vector k and energy Eg + e j2me and the hole in
the valence band of wave vector -k and energy k2 j2mv' Energy conservation is
(9.81)
which is exactly the same as the argument of the delta function for energy conservation in
(9.79). So far the two-particle picture leads to the same absorption rate as the one-particle
prediction in (9.79). Both the matrix and the energy conservation are identical, so the
prediction is exactly the same.
The point made by Wannier is that there is now additional physics which can occur in the
two-particle picture. The electron and hole are particles with charges of opposite signs, so that
there is a Coulomb attraction -e'2 j(Eor) between them, where EO is the static dielectric
function. This dielectric function is assumed to be a constant which is independent of
frequency. This approximation is poor, since most semiconductors are polar, and the dielectric
function has significant dispersion at frequencies near the optical phonon frequencies; see
Sec. 6.3.1. The frequency-dependent screening is an interesting problem which is ignored by
treating Eo as a constant. The attractive Coulomb interaction between the electron and hole
can cause hydrogenic bound states between them.
The optical absorption rate for this process was calculated by Elliott (1957). The final
state of the system is described by a two-particle SchrOdinger equation:
(9.82)
(9.83)
<ll(re. rh) can be factored into relative r = re - rh and center of mass coordinates
M = me + mv in the standard fashion:
(9.84)
(9.85)
(9.86)
(9.87)
The center of mass motion is plane-wave-like, with a wave vector P which in optical
experiments is equal to the photon wave vector. This wave vector is small. It is neglected in
the center of mass motion and set P = O. The relative motion of the electron and hole is
usually more interesting. For relative energy er less than zero, the two particles form bound
hydrogenic states with energies er = en = -ERln2. For relative energy er greater than zero
they form scattering states <her). Elliott showed that the optical transition rate depends on the
relative wave function at r = 0, <h(O). Instead of (9.79), the transition rate is determined by
(9.88)
The summation j is over the bound and continuum states of the relative motion of the
electron-hole pair. The dependence on the relative wave function, evaluated at r = 0, can be
understood by a physical argument. First, consider the corresponding emission process,
whereby the electron and hole recombine to emit a photon with energy lim. The emission rate
should depend on the relative wave function at r = 0, since that is the probability that the
electron and hole find themselves at the same spot in the solid, which is necessary for the
recombination. The dependence of the emission rate on l<PiO)12 is reasonable from physical
intuition. But the matrix elements for absorption and emission are identical, so that absorption
should have the same dependence. The absorption depends on the probability of making the
electron and hole at the same point in the solid.
The former result, for the noninteracting solid, is recovered from (9.88) by setting
l<PiO)12 = l/v and ej = k 2 12~. The results, for the interacting case are quite different. The
relative motions of the electron and hole are in s-wave hydrogenic states, either bound or
unbound, because of the angular momentum selection rule. The one-unit change in I, in the
photon absorption, is taken by the change of band symmetry, and the relative motion is not
permitted any additional angular momentum. For s states, the bound states have an amplitude
given by the principal quantum number n and the Bohr radius ao:
<PnCO) = ~
1ta on
(9.89)
For continuum states, with energy ek = ~ 12l!, the relative wave function at the origin is
l<Pk,l=O(O) = 21tTJ/v[1 - exp( -21tTJ)], where l/TJ = ka o. Then (9.88) predicts that the
absorption is a constant in frequency at the energy gap E g , and does not rise with a square
root dependence on (lim - Eg). The actual shape is shown in Fig. 9.7. A few sharp, distinct
exciton lines are observed at low frequency which correspond to 1s, 2s, etc. In actual
experiments, these absorption bands are hard to see because they are too strong. Only in the
thinnest crystals can enough light be transmitted through the experimental sample to measure
the absorption rate at the bound states of the excitons. Usually the absorption is so strong that
all the light is attenuated before transversing the sample. At higher frequencies, the ns states
"W'\ ~
EG-E.
rz:-
EG w
FIGURE 9.7 Typical absorption spectra ofa direct gap semiconductor. Distinct exciton lines are labeled Is, 2s, etc.
The solid line is the theory with final state interactions, and the dashed line is the theory without them.
are closer in frequency and are broadened, so that their envelope becomes a continuous
distribution which merges with the continuum absorption which starts at lim = Eg . This
experiment shows no anomaly at lim = Eg, and the experimental value of Eg is obtained only
by extrapolating the Rydberg series for the positions of the ns exciton lines:
ER
mn = Eg + En = Eg - 2
n
(9.90)
In most semiconductors, the Rydberg unit ER is of the order of several millivolts because of
the high values of the dielectric constant Eo ~ 10 and the small values of the effective mass
j.l ~ 0.1.
This discussion is only a brief review of exciton theory in semiconductors. The interested
reader is encouraged to seek additional information in the books by Knox (1963) and
Reynolds and Collins (1981). The theory has been verified by many detailed experiments,
such as exciton properties in static electric or magnetic fields, under stress, etc. Exciton effects
are important for determining the absorption rate of photons within a frequency interval of ER
near the energy gap Eg •
9.2.2. Solution by Green's Functions

The Elliott formula (9.88) is derived using Green's functions. The starting point is the
Hamiltonian (9.86) which contains the properties of electrons, holes, and their mutual
interaction. The Kubo formula is evaluated for the optical conductivity, which is proportional
to A(m). The Elliott forumula is obtained by a summation of all the vertex diagrams of the
current--current correlation function. The Green's function analysis is straightforward and can
be solved exactly, because it is still only a two-body problem. Indeed, this section could have
been included in Chapter 4 on exactly solvable models. The motivation for this analysis is to
stress the importance of final state interactions. The latter is a fancy name for the interactions
between the particles in the final state of the optical transition. In this case they are the
Coulomb interaction between the electron and hole.
The starting point for the Green's function calculation is the Kubo formula for the
current--current correlation function:
(9.91)
(9.92)
(9.93)
The 't development of the operators is governed by a Hamiltonian which contains the
Coulomb interaction Vq = -4ne2/Eoq2 between the electron (Cka , cta) and the hole
(dka , dta)
H=Ho +V (9.94)
Ho = L ~V<k)dtdk + L ~c(k)CtCk (9.95)

k k
(9.96)
Spin does not playa role in this analysis, so the spin index is no longer written. The procedure
is to solve the Kubo formula (9.92) with this Hamiltonian. The optical matrix element Wk is
taken here to be a function of k, although in the theory for Wannier excitons it is a constant.
An important assumption, when evaluating this correlation function, is that the densities of
electrons ne and holes nh are both negligibly small. An equivalent assumption is that the
chemical potential ~ in the semiconductor is in the forbidden energy gap, somewhere between
energy zero and Eg . The chemical potential ~h for the holes is the negative of that for the
electrons ~h = -~e' and the energies in the preceding Hamiltonian are
12
~ (k) = - + E -~ (9.97)
e 2mc g
k2 k2
~h(k) = -
2mv
- ~h = -
2mv
+~ (9.98)
The definitions are finished, and it is time to evaluate the correlation function.
The electron-hole interaction V is treated as the perturbation for the Hamiltonian. The
first term is that for no interaction, which is called 1t(O)(iro):
1t(O)(iro) = -~ L~J~ d-reiffint{TtCk(-r)d_k(-r)d!k(O)Ct(O») (9.99)

v k 0
= - ~ L ~J~ d-reiffint~cCk, -r)~h(-k,-r) (9.100)

v k 0
1 1
= - - L~- L~c(k, ik)~i-k, iro - ik) (9.101)
v k ~ ik
= ~ L w~ 1 - nF(~e) - nF(~h) (9.102)
V k iro - ~e(k) - ~h(k)
The last step involves a standard frequency summation, as given in Sec. 3.5. Both Green's
functions ~c = l/(ik - ~e) and ~h = I/(ik - ~h) are for noninteracting particles, but the
superscripts (0) are omitted since all the Green's functions in this section are noninteracting.
The occupation factors nF are zero. The absorption A(ro) is the spectral function of this
correlation function. Take the analytical continuation iro -+ ro + iO and define
A(ro) = -2 Im[1t(ro)] (9.103)
A(O)(ro) = -2 Im[1t(O)(ro)] = 21t J(21t)

d3k3 ~o(ro - ~vCk) - ~eCk»
= 21t J(21t)3
d k w~o(ro _ E
3
g
_ 12)
2~
(9.104)
The result for A(O)(ro) is recognized as the one-particle theory of (9.79). The remaining terms
in the expansion of the S matrix are now examined. Fortunately, most of them are zero. All
self-energy terms, for either the electron or the hole, are zero. All self-energy terms for the
electron or hole contain closed loops and vanish since they are proportional to nF' The only
nonzero contributions are vertex diagrams in the form of ladder diagrams.
The method of summing the ladder diagrams follows Mahan (1967) and was used for
impurity scattering in Sec. 4.1.5. They are shown as the diagrams in Fig. 9.8. The first vertex
correction A(l) contains a single ladder diagram, where the dashed interaction represents the
FIGURE 9.8
Coulomb interaction between the electron and the hole. The two particles are represented by
the solid line segments. The wiggly line at the vertex represents the photon. The lines for both
the electron and hole are shown coming from the photon vertex and are traveling parallel in
time. The diagram with n ladders is called A(n). They can be constructed according to the rules
for diagram construction. The first ladder diagrams gives
1 1
rc(l)(iOl) = '2 L WkVk-k,Wk' A L ~e(k, ik)~h( -k, iOl - i~)
v kk' I-' ik
1
x A L ~eCk ', ik')~h( -k', iOl - ik') (9.105)
)J ik'
The two summations over Matsubara frequency are identical in form:

1 . .. 1 - nF(~e) - nF(~h)
~ ~ ~eCk, lk)~h( -k, 10l - lk) = iOl _ ~e(k) - ~h(k)
iOl - Eg - k 2 /2/1
(I). _ 1 1
rc (10l) - v2 ~ wkVk_k'Wk' iOl _ Eg _ k 2/2/1 iOl - Eg - kt2 /2/1
(9.106)
Additional ladder diagrams in A(n) just produce additional factors of the Green's function pair
~c~h' which are summed over Matsubara frequency to produce another energy denominator
(iOl - Eg - kJ /2/1). Each ladder also has another factor of the Coulomb potential Vk,_k/V. For
this problem it is easy to write the term with n ladder diagrams:
rc(n) = _1_ L wkVk-k,'" Vkn_,-knwkn II. 1
v n+! kk, ... k n iOl - Eg - k2 /2/1 J iOl - Eg - kJ /2/1
The simple form of these terms permits them to be summed exactly to obtain the expression
for the correlation function. The other vertex diagrams, which are not ladder diagrams but
have a crossing of the interaction lines, are all proportional to the numbers of electrons or
holes in the semiconductor. They can be neglected, and the only nonzero vertex corrections
are the ladder diagrams.
The summation of the ladder diagrams can be expressed as a vertex correction to the
correlation function, which is written as
(9.107)
(9.108)
An exact summation of the ladder diagrams is obtained by solving for this vertex function.
The integral equation (9.108) for the vertex function is an equation of the relative motion of
the two particles.
The first step in the solution of the integral equation (9.108) is to define the auxiliary
function
(9.109)
For the Wannier exciton system, the matrix element Wk is independent of k and is a constant
Woo In this case, the correlation function in (9.107) is the value of this auxiliary function at the
origin:
1tL(ioo) = woP(r = 0, ioo) (9.110)
The integral equation (9.108) for the vertex function becomes
rek, ioo) = Wo - Jd3r~e-ik' rp(r, ioo) (9.111)

Eor
The auxiliary function per, ioo) is introduced so that one can write a differential equation for
it, which is then solved to obtain the exact solution to the function. This differential equation
is obtained by operating on both sides of (9.109) with the operator [ioo - Eg + V2 /21!], which
is chosen to eliminate the energy denominator on the right-hand side:
(zoo. - Eg + -21!V2). 1
per, zoo) = - I: e
v k
ik. r . ioo - Eg - Jc2 /21!
rek, zoo) .
zoo - Eg -
k 2 /2
I!
(9.112)
= ~ I: e,J( . rrek, ioo) (9.113)

v k
= ~v I:k eik ' r [W0 - Jd3r' --==--e-,J(·

EOr'
r' P(r', iOO)]
= woo3(r) - ~p(r, ioo) (9.114)

EOr
The integral equation (9.111) for the vertex function was used to obtain a differential equation
for the function per, ioo). The interaction term on the right is moved to the left-hand side of
the equation, which derives the inhomogeneous differential equation:
(ioo - Eg - Hex)P(r, ioo) = woo3(r) (9.115)
H
,,2 v2 __
= __ e2
(9.116)
ex 21! Eor
This equation is solved in the following fashion. The factors ioo - Eg on the left are constants,
and the important operator parts are the same as found in the exciton Hamiltonian (9.86) for
their relative motion:
(9.117)
The Hamiltonian Hex may be solved exactly in terms of hydrogenic eigenfunctions <p/r) and
eigenvalues Ej . In terms of these exact solutions, the solution to the differential equation
(9.115) is written as
.)
P( r,/CO " <p/r)<pj(O)
= Wo L... . (9.118)
j lCO - Eg - Ej
which is verified by examining the original equation:

. . * ico - Eg - E
(/CO - Eg - Hex)P(r, lCO) = Wo I: <p/r)<pj (0) -.---"-~j
j lco-Eg-Ej
= Wo I: <p/r)<pj(O) = wo83(r) (9.119)
j
The last step is a summation over the complete set of hydrogenic states, which produces a
delta function. This result agrees with (9.115), so that an exact solution is found for per, ico).
The current-current correlated function nL(ico) is obtained from (9.110) as the value of
woP(r = 0, ico):
(9.120)
The optical absorption function A(co) is the spectral function of this operator:
A(co) = 2nw6 I: I<p/O)1 28(co - Eg - E)

j
This result is exactly the Elliott form in (9.88), with the transition rate depending on the
square of the relative wave function evaluated at r = O. As discussed in Sec. 8.2.1, this
dependence causes a dramatic and important change in the absorption spectra. The final state
interactions are very important. They are obtained by summing the ladder diagrams of the
correlation function. It is an exact solution to the current-current correlation function (9.92)
for the model Hamiltonian in (9.96) and with the restriction that there is a negligible density
of electrons and holes in the semiconductor.
9.2.3. Core-Level Spectra

Core-level spectra in insulators is another problem which is similar, both physically and
mathematically, to the Wannier exciton. The initial electronic state in the optical transition is a
core level of the atom, such as the 1s or 2s shell. The optical transition in the insulator takes
the core electron to an unoccupied state in the conduction band. The empty core level can still
be described as a hole, but now it is very localized in space and so has an infinite effective
mass. The hole state has a constant energy ~h = Ec + /1, where Ec is the core-level binding
energy, as measured from the top of the valence band. The threshold for the optical transition,
in the absence of any final state interactions, is the frequency COT = Eg + Ec.
This problem can be solved exactly by the same techniques which were used for Wannier
excitons. The advantage of the present model is that it is possible to explain the physics with
greater clarity. From the point of view of the conduction electron, it has two possible potential
functions. There is a potential for the initial state of the system, which is called Vi(r). It is the
potential acting on a conduction electron in the initial state, when there is no core hole. The
effective Hamiltonian for the conduction electron is called Hi> and its eigenfunctions are
'Plk, r) with energy Elk):
(9.121)
After the optical transition has occurred, the hole appears in the core state, which alters the
potential of the conduction electron to 1'j(r). This final state potential is set equal to the initial
state potential plus a term due to the core hole:
1'j=V;+Vh (9.122)
Hf'Pf = [ - ~~ V2 + Vf(r) ]'Pf(k, r) = Ef(k)'Pf(k, r) (9.123)
The core hole potential Vh(r) has the form of a Coulomb potential -e
j(Eor) at long range,
but there are atomic effects at short range. The final state Hamiltonian Hf has its own set of
eigenfunctions 'Pf(k, r) and eigenstates Ef(k) which are different from those of the initial
state. These states may be bound or unbound, to the core hole, and k is a general index which
can represent either of these possibilities. The conduction electron has two possible sets of
eigenstates which could enter into the calculation for the absorption. Which one should be
used?
The conventional approach to this problem is to expand the Hamiltonian and the current
operator in the basis states of the initial Hamiltonian Hi' A set of creation and destruction
operators (Cks ' ct) are associated with the basis set 'Pi' The effective Hamiltonian is
H = ~h L d] ds + L ~i(k)CtCks + L Vh(k, k')CtCk'sd;,ds'

s ks kk'ss'
(9.124)
(9.125)
This Hamiltonian has the same general form as (9.96). The last term in the Hamiltonian is the
interaction between the conduction electron and the hole; the latter is represented by the
operators (ds ' d]). The current operator, for the Kubo formula (9.92), is also written in terms
of these functions:
(9.126)
(9.127)
where <pc(r) is the core wave function of the hole.

The problem has the same form as the Wannier exciton which was solved previously and
is treated similarly. The correlation function, from the Kubo formula, is evaluated as a set of
ladder diagrams. Their summation is achieved by a set of equations which are nearly identical
to (9.107) and (9.108):
(9.128)
(9.129)
The vertex function riCk, iro) is solved by introducing the auxiliary function
.) " 'Pi(k, r}ri(k, iro)
P (r, zro = ~ (9.130)
k iro - roT - Ei(k)
n(iro) = ~ Jd 3 nPc(r)E • pP(r, iro) (9.131)
The equation for the vertex function is
(9.132)
The correlation function per, iro) is solved by writing it as a differential equation, which is
found by first operating by
(iro - roT - Hi)P(r, iro) =L 'Pi(k, r)ri(k, iro) (9.133)
k
= ~ 'Pi(k, r) Jd r''P7(k, r')[E . p<pc(r') + Vh(r')P(r', iro)]

3
= Jd r8(r - r')[E . p<pc(r') + Vh(r)P(r', iro)]

3
= E . p<pJr) + Vh(r')P(r, iro) (9.134)
where the completeness relation is

L 'PJk, r)'P;(k, r') = 8(r - r'). (9.135)
k
This relation produces the delta function which eliminates the integration over r'. The
potential term is moved to the left of the equals sign, which makes the combination
Hi + Vh = Hf . The auxiliary function per, iro) obeys the equation
(iro - roT - Hf )P(r, iro) = E • p<pc(r) (9.136)
This equation is solved in terms of the basis functions 'Pf of the final state Hamiltonian Hf :
. J
" 'Pf(k, r) d 3r''Pj(k, r')E . p<pcCr')
per, zro) = ~ (9.137)
k iro - roT - Ef(k)
An obvious notation is to define the optical matrix element in terms of the final state basis
functions:
(9.138)
Sec. 9.3 • X-ray Spectra in Metals 603
The correlation function in (9.131) is
. 1" IwtCk )12

1t(lCO) = - L- (9.139)
V ks iro - roT - EtCk)
21t
A(ro) = -v Lks IwtCk )1 2o[ro- roT - Ef(k)] (9.140)
The absorption spectrum is the spectral function of this correlation function. Equation (9.140)
is actually a simple physical result. It shows that the absorption spectrum A(ro) can be
obtained from a one-particle spectrum by using the golden rule. However, one has to use the
wave functions 'l'tCk, r) and eigenvalues Ef of the electron in the final state of the transition.
During the optical transition, the effective potential of the system changes from Vi to Jif.
Formula (9.140) shows that the optical spectrum is calculated using Jif. This conclusion is
important, not only in solid-state spectra but also in atomic physics.
In a single-particle picture, this conclusion is obvious. The core hole wave function <pe(r)
is also calculated for the potential Jif, since in calculating such states one does not include the
interaction of a particle with itself. That is, presumedly <P e is found from a SchrOdinger
equation with a potential which also does not have the core level occupied. In a single-particle
picture, this potential is Jif. The final formula (9.140) merely describes a one-particle tran-
sition between two different energy levels of the same Hamiltonian Hf . It is an unremarkable
application of the golden rule for transition rates. It is possible to solve this many-body
problem because it is a one-body problem.
Unfortunately the present model is too simple to apply accurately to real spectra. One of
the assumptions is that the changeover from Vi to Jif occurs instantly with the creation of the
hole. In actual systems, the other electrons and ions take time to adjust to the core hole
potential. This time can be viewed as a frequency-dependent screening process. A realistic
calculation should confront the dynamic screening, which was carefully neglected at the
beginning of the calculation.
9.3. X-RAY SPECTRA IN METALS
9.3.1. Physical Model

The previous section discussed exciton effects in the core-level spectra of semi-
conductors or insulators. Now consider the same optical transitions in metals. The photons
usually have higher energy, so the spectroscopy is labeled X-ray or soft X-ray. The many-
body theory of this spectroscopy is considerably more complicated, since the conduction
electrons in the metal respond dramatically to the X-ray transition. These processes cause
several new effects which were predicted theoretically and verified experimentally. Among
them are (1) the prediction of Mahan (1967) that the absorption edges had a power law
divergence near threshold, which has been found for the p-shell spectra of simple metals such
as sodium, magnesium, and potassium, and (2) the prediction of Doniach and Sunjic (1970)
of asymmetries in the XPS (X-ray photoelectron spectroscopy) line shapes from core levels of
these metals, which is also now well documented experimentally. The latter phenomenon is a
consequence of (3) the orthogonality catastrophe, first explained by Anderson (1967). The
general problem is often called "MNO" after the work of Mahan (1967) and Nozieres and
deDominicis (1969).
The model for free-electron metals is shown in Fig. 9.9. It is an energy-level diagram and
shows the parabolic states in the conduction band, which is filled with electrons to the Fermi
energy J.l. The horizontal lines on the bottom represent core levels in the ions in the metal;
they are localized, have an infinite effective mass, and are drawn flat on an energy-level
diagram. In the X-ray absorption process, the photon energy is used to lift an electron from a
core level to an unoccupied state in the conduction band. At zero temperature, the only empty
states are above the chemical potential 11. In a one-election description of this process, the
absorption must start at the threshold frequency floor = EF + Ec where Ec is the core-level
binding energy, as measured from the bottom of the conduction band, and EF is the width of
the occupied conduction band:
(9.141)
The factor [1 - nF] is included to limit the final states to above the Fermi level. As indicated
schematically in Fig. 9.9, the golden rule result (9.141) predicts that the absorption starts with
a nonzero value at threshold, so there is a step in the spectrum. The step occurs because the
electron gas has a nonzero density of states at the threshold frequency, since the electrons are
going to states right above the Fermi energy.
The core levels of ions have angular momentum as a good quantum number. The angular
momentum of the electron state must change by one unit during the photon transition. If the
core level has an initial angular momentum I, the final state in the conduction band is either
1+ 1 or 1 - 1. An atomic description of this absorption process would write the one-particle
theory in terms of these two final state channels:
(9.142)
where the constants AI±J (k) are found from Clebsch-Gordan coefficients and radial integrals.
In a metal, the final electron state is a Bloch function, which is not an eigenfunction of the
angular momentum operator. The use of Bloch functions leads to a more complicated
A(w) ABSORPTION
ABSORPTION EMISSION I(w)
I w
FIGURE 9.9 X-ray transition between core levels and parabolic conduction band of a metal. Electron sates are
occupied below Fermi energy which is EF above the conduction band minimum. Absorption must remove a core
electron and out it into an unoccupied band state, while emission must remove an electron from an occupied band
state. Also shown are the predicted absorption and emission spectra in the one-body picture, neglecting energy band
variations.
description, since the final electron states in (9.141) must be summed over the full Brillouin
zone and its various energy bands.
The word hole has two different meanings in the present discussion. The first is a core
hole, which is an electron missing from a core state. The other is a conduction band hole,
where the electron is absent from an occupied state in the Fermi sea. Core hole is used for the
first case and hole for the second.
After the core hole has been created, it has several possible decay channels. The most
likely is an Auger process, whereby an electron from a higher energy state falls into this core
hole while its energy is transferred to another electron whose energy is correspondingly
increased. The electron which gains energy can also be measured, and this Auger spectro-
scopy is another important experiment (Gallon, 1978).
The second decay channel is the emission of a photon by an electron which falls into the
core hole. When the electron starts in the occupied states in the conduction band, this X-ray
emission spectroscopy is an experiment which is complementary to the absorption. The
emission provides a measurement over the states, below the Fermi level, which are occupied
in the ground state of the Fermi sea. The absorption measures the unoccupied states above the
Fermi energy. The emission spectrum also has a sharp step at the threshold frequency
IiroT = EF + Ee , as shown schematically in Fig. 9.9. These steps, in emission and absorption,
are called Fermi edges.
There are several physical mechanisms which can change the shape of the spectrum, in
emission or absorption, from the simple one-particle theory in (9.141). The first of these is
due to the lifetime of the core hole, which is usually dominated by the Auger decay process as
just described. This core hole lifetime is quite variable among the various core levels of an ion
or atom. The core hole lifetime is longer for the outer valence shells than for the inner states
closer to the nucleus. Some experimental values for core hole lifetimes in simple metals are
shown in Table 9.1, from data of Citrin et al. (1977). The same group recently remeasured
these values, Wertheim et al. (1992). The Auger lifetime is expressed as an energy uncertainty
r = lilT. (where r is the full width at half maximum). Values as high as several electron volts
can be found for inner shells of atoms with high Atomic number. This Auger decay generally
imparts a Lorentzian broadening to the X-ray spectra. It can be included, semiempirically, in
the Kubo formula as a factor exp(-rjtI/2). If the X-ray measurement is to determine other
phenomena, it is desirable to keep r, as small as possible, which is done by measuring the
spectra of outer atomic shells.
TABLE 9.1 Core level data at T = 300 K. r is

Auger width while y is phonon width. From
Citrin et al. (1977)
Metal State r(eV) y(eV) IX
Li Is 0.03 0.37 0.23

Na 2p 0.02 0.18 0.20
2s 0.28 0.20 0.20
Is 0.28 0.18 0.21
Mg 2p 0.03 0.18 0.13
2s 0.46 0.19 0.13
Is 0.35 0.20 0.15
AI 2p 0.04 0.11 0.12
2s 0.78 0.12
Another many-body effect is the broadening due to phonons. The only term which is
usually kept in the Hamiltonian is the phonon coupling to the core hole, which is a term such
as
(9.143)
It was shown in Sec. 4.3 that this type of phonon coupling to a localized level is a type of
Hamiltonian which is exactly solvable. The optical absorption from such a coupled level was
also solved, and the Kubo formula has a factor of exp[ - (t)], where
(t) = ~ I~~i~ 12 {[I - itro,Jq) - e-itCO)Jq)][nB(ro,Jq)) + 1]
+ [1 + itro,Jq) - eitco,(q)]nB(ro",(q))} (9.144)
~ "/r, y2 = ! LqA IM,Jq)1 2 [1 + 2nB(ro",(q)] (9.145)
The term linear in t is the self-energy term. The factor exp( -<I» is added to the argument of
the time integral in (8.3.3). Usually the Auger width r is large enough that only the short time
response is needed of the correlation function. Anything that is going to happen must do so
quickly, before the core hole decays by the Auger process. The short time limit of the phonon
contribution is ~ y2 t2 as shown above. The spectral function is broadened by the Auger and
phonon contributions
(9.146)
The phonons add a Gaussian broadening to the spectral function for the X-ray process, where
the Gaussian component y is temperature dependent. It is relatively easy to experimentally
separate the phonons from the Auger width, which is a temperature-independent Lorentzian.
The only remaining problem for the phonon contribution is to calculate the matrix elements
MA(q) for the coupling to the core hole. However, this calculation is quite delicate (Hedin and
Rosengren, 1977).
The phonon broadening for the emission spectra is a much more subtle calculation. Here
the problem is that the phonon system does not equilibrate to the presence of the core hole
before the emission occurs. One expects that the minimum time for the phonons to come to
equilibrium around the new core hole, and its screening charge, is the inverse Debye
frequency tph ~ l/roD' This time is generally much longer than the lifetime for the Auger
effect, except for the outer shell of electrons, as shown in Table 9.1. The theory of phonon
broadening of emission must account for the nonequilibrium states of the phonons. This
theory was developed simultaneously by Mahan (1977) and Almbladh (1977). The most
interesting case yet known is lithium, for which phonons are the most important of all the
mechanisms which broaden the edge. The theory predicts a double shoulder in the emission
spectra because of incomplete relaxation of the phonons. The prediction was verified by the
experiments of Callcott and Arakawa (1977).
These many-body processes, Auger and phonon, are reasonably straightforward. The
more interesting aspects are the response of the many-electron system to the appearance ofthe
core hole, which are discussed in the remaining parts of this section.
9.3.2. Edge Singularities

Mahan (1967) predicted that the absorption edges for X-ray transitions in metals would
have a power law divergence of the form
(9.147)
where ~o "-' EF is a bandwidth, and the exponent 0( is discussed below. His theory resulted
from an investigation into the properties of excitons in metals. The Wannier picture applies
equally well to metals, so that the optical transition should be viewed as the simultaneous
creation of an electron and a core hole in the metal. These two particles will interact in the
final state. Presumably this interaction is screened, although a detailed theory of the screening
process is still an active subject of research (Canright, 1988). Mahan assumed that the
electron core hole Coulomb potential is screened instantly at the time the core hole is created.
This approximation may be valid near the absorption edges but surely is unreliable farther
from threshold.
The exciton theory predicts that the exponent 0( in (9.147) is positive, so the absorption
diverges as a power law singularity at threshold. Later work has shown that the renormali-
zation catastrophe of Anderson tends to make 0( negative. The final values of 0( may be either
positive or negative, depending on which of these two factors is most important. Not all edges
are singular.
The calculation assumes there is instantaneous screening of the electron-hole interaction
in the metal. The model Hamiltonian for the X-ray absorption process is:
(9.148)
The spin indices have been dropped from the notation. The last term is the screened Coulomb
interaction between the electron and the core hole, while the first terms are Ho for the
electrons and the core holes. The Hamiltonian is written in the initial state basis, so the
electron operators (Ck , ct) refer to the state which does not have a core hole. For the Wannier
exciton theory it was shown that there is a great advantage to working in the final state basis
set, but that is not the case here. The calculation is about equally difficult in either basis set,
and the same answer is found from either starting point. The difference between the present
calculation for the metal, and the earlier calculation for the insulator, is that there are now
N "-' 1023 conduction electrons which also interact with the core hole, and they are initially in
states described by the initial state basis. In the calculation, it is now necessary to keep the
terms nF(~c) for the electron occupation number. They are the cause of the dramatic change in
the theory, and the prediction (9.147) of the edge singularity.
In the exciton calculation, the vertex diagrams to the correlation function are included by
summing the set of diagrams in Fig. 9.10. To keep the physics simple, the calculation is first
done by assuming some approximations which tend to reduce the number of superscripts and
subscripts: (1) the interband matrix element (fIE' pic) is taken to be a constant wo; (2) the
screened Coulomb interaction is taken to be a constant V(k, k') = Vo up to a cutoff energy
~o ~ EF which is a typical bandwidth. With these approximations, the ladder diagrams in the
(0) (b) (e) (d)
0+0+<1>+<2)
FIGURE 9.10
vertex summation in Fig. 9.10 can be written as in the previous section:
1tL(iro) = L 1t(n)(iro) (9.149)

n=O
(9.150)
Removing the wave vector dependence of the scattering potential uncouples the wave vector
integrals. One is left with a product of (n + I) integrals which are identical. The integrand of
the wave vector integral contains the factor I - nF(~k) - nF(~h)' where the core hole term
nF(~h) is set equal to zero. This step leaves I - nF(~k)' which ensures that the electron
scatters into unoccupied states. The summation over wave vectors can be changed approxi-
mately to d 3k --+ (27t)3 NFd~k" where NF is the density of states. The d~k' integral is just a
logarithm, and the ladder diagrams give the result at zero temperature
NFJ~O d~k'. I - nF(~k) = NF In (roT - iro) (9.151)

-~O lro - roT - ~k ~o
1t(n)(iro) = -%NF ( -NF vot [In ( roT ~~ iro) r+! (9.152)
These approximations are fairly crude. They are meant only to apply to the threshold region,
where ro ~ roT, so that the cutoff in the wave vector integration by the Fermi occupation
factors nF is most important. The vertex summation leads to a summation over logarithms of
increasing powers.
This series should not yet be summed, since other vertex diagrams contribute terms
which also contribute to the series. The first diagram in the series, which is not a ladder, is that
shown in Fig. 9.10(d). It contains six Green's functions as internal lines. The following
summations over Matsubara frequencies must be done:
I
3" L <;§eC k , ik)<;§eCk + q, ik + iq)<;§eCk + q + q', ik + iq + iq')
~ ik,iq,iq'
x <;§h(-k, iro - ik)<;§h(-k - q', iro - ik - iq')

x <;§h( -k - q - q', iro - ik - iq - iq') (9.153)
These summations are done in sequence. The core hole Green's functions all have the same
energy ~h which is independent of wave vector. The conduction electron energies will be
denoted by the symbols ~ = ~e(k),~' = ~e(k + q), ~" = ~e(k + q + q'). When doing these
summations, the core hole occupation factor nF(~h) is set to zero wherever it occurs. With
these conventions, the three summations are, in sequence,
1
A L ~e(k + q + q')~h(iro - ik - iq')~h(iro - ik - iq - iq')
t-' iq'
1 - nF(~I/)
= (iro - ~I/ - ~h)(iro + iq - ~I/ - ~h)
!L ~e(k + q) = nF(~')
~~~+~-r-~ ~-~+~-r-~
.!. L ~eCk)~h(iro - ik) = 1 - nF(~)
~ ik iro - ik + ~' - ~I/ - ~h (iro - ~ - ~h)(iro - ~ + ~' - ~I/ - ~h)
The vertex correction is obtained from the integral (~h = roT)
(9.154)
This integral is more difficult to evaluate than those in the previous cases, since the middle
energy denominator contains all three conduction energies, -~ +~' - ~I/. The integral over
d~' is over the occupied states nF(~'). These terms could be ignored in the insulator where no
is zero. They must be included in the metal where no =j:. O.
The triple integral in (9.154) is obviously complicated. The most divergent term it yields
r
is
n(2b)(iro) = l%NF(NFVoi[ In(ro T ~~ iro) (9.155)
This term is one-third the contribution of the double ladder diagram in Fig. 9.10(c) and with
the opposite sign.
A similar calculation can be done for the third order vertex corrections. Each of the six
terms with three vertex lines in Fig. 9.10(e) produces terms which have {In[(roT - iro)/~o]}3.
The summation of all the terms shown in Fig. 9.10 gives the series
(9.156)
(9.157)
The series is now summed to produce the final result:
(9.158)
(9.159)
The optical absorption is the spectral function of this operator. Note that the right-hand side is
real for ro < roT, but complex parts are generated when ro > roT
A(ro) = -2 Im[n(ro)]
(9.160)
Equation (9.160) is Mahan's (1967) result that the absorption diverges as a power law, where
the exponent is the dimensionless quantity oc = 2NF Yo. The exponent is obtained by summing
the most divergent diagrams in each order of perturbation theory. These terms are divergent,
because the retarded function In[(roT - ro)/~o] can become quite large near threshold ro ~ roT'
In the limit where Vo = 0, the above formula goes to the noninteracting result that
A(O) = 2nNF%0(ro - roT), and the absorption edge is a simple step.
The correlation function (9.77) for the model Hamiltonian (9.148) can be evaluated
accurately using simple analytical formulas. Pardee and Mahan (1973) showed that the edge
singularities could be derived in a simple way using dispersion theory. Penn et al. (1981,
called PGM) obtained an analytical solution to the multiple scattering problem, including all
vertex corrections. They showed that the leading term in the series was the result of Pardee
and Mahan. Ohtaka and Tanabe (1983, 1986) obtained a more elegant analytical solution,
which they illustrated with numerical examples. This work has provided a complete solution
to the MND problem, and in an analytical form that is useful for computation.
The result of Pardee and Mahan is obtained by a simple argument. Let T(ro) be the
complex scattering amplitude for an ingoing wave of an electron scattering from a central
potential, which in this case is the screened core hole. If T'(ro) is the amplitude of the
outgoing scattered wave, they differ only by a phase factor which is twice the phase shift:
T' (ro) = T( ro) exp[2io/(ro )0(ro - roT)] (9.161)
where I is the angular momentum of the outgoing electron. The step function is a reminder
that the phase factor exists only in the region of absorption, since T and T' must both be real
in regions where there is no absorption. The next assumption is that T and T' both originate
from the same analytical function of frequency:
t(ro + iO) = T'(ro) (9.162)

t(ro - iO) = T(ro) (9.163)
Combining these results shows that
t(ro ± iO) = exp[<J>(ro ± iO)] = IT(ro) I exp[±io/(ro)0(ro - roT)] (9.164)

Im[<J>(ro ± iO)] = ±0/(ro)0(ro - roT) (9.165)
Dispersion theory assumes that <J>(z) in (9.164) is an analytical function of complex frequency
z. The only function that has the property (9.164) is
1 J~O+roT o/(e)
<J>(z) =- de- (9.166)
n roT e- z
As usual ~o is the bandwidth. The function q,(z) is a dispersion integral. The integral can be
evaluated by writing O/(E) = Otero) + [O/(E) - Otero)] and the first tenn is an easy integral:
q,(ro + iO) = °l(ro)

n
[In I~o roT+roT- roro 1+ in0(ro - roT)] +K(ro) (9.167)
K(ro) = .!.f~O dE 0l(E) - °l(ro) (9.168)

n roT E-ro
The function K(ro) is a smooth function of ro since the integrand is real and not singular. The
amplitude of the X-ray absorption is proportional to T'(ro) and the intensity is proportional to
IT'(ro)1 2. This factor can be evaluated using the dispersion integrals
12'Mro)/1t
IT'(ro)12 = I
!O
,",0 + roT - ro e'lK{ro) (9.169)
roT - ro
A(ro) ~ ~
...Q sin(20)
(~)2'O/1t
__0 - 0(ro - roT) (9.170)
Vo ro - roT
This equation has the same edge singularity as (9.160). Here the exponent of the power law is
IX = 20/(ro)/n rather than 2NF VO' Note that if one is calculating the phase shift in the Born
approximation, then one gets that sin(o) ~ 0 ~ nNFVO' Nozieres and deDorninicis (1969)
were the first to realize that the exponents of the edge singularities were functions of the phase
shifts. At the threshold frequency ro ~ roT then the exponent depends upon o/(roT ), which are
the phase shifts of conduction electrons at the Fermi surface for scattering from the core hole.
The factor of K(ro) is a smooth function of frequency and does not contribute to the singular
behavior. It has an influence upon the calculation of the X-ray absorption for other values of
frequency.
In X-ray absorption an electron absorbs an X-ray, departs the atomic core, and becomes
a conduction electron. The core potential changes from having a charge of Z to having a
charge of Z + 1. All of the other conduction electrons have their central potential suddenly
changed by the appearance of this new core hole. The switching-on of the potential may cause
electron-hole pairs to be created. The pairs act as bosons. The theory for this process can be
treated by the independent Boson model of Sec. 4.3.
The electron that departs the core state and becomes a conduction electron may also
create excitations of electron-hole pairs. The departing electron has only a small probability
of creating pairs. This process may be accurately calculated by perturbation theory.
X-ray absorption has a number offactors that contribute to the shape of the spectra. The
above discussion derived the contributions of phonons, Auger decay, and the final state
Coulomb interactions. Another important contribution is the shakeup mentioned above. The
shakeup contributes a time evolution pet) which is calculated in the next section. All of these
various factors contribute a time dependence to the evolution of the absorption. The
experimental spectra are found by Fourier transforming the product of all of these factors
A(ro) = J~oo dtp(t)A(t)exp[it(ro - roT) - ntl - y2?] (9.171)
The factor A(t) is the Fourier transfonn of A(ro) in (9.170). Instead of transforming A(ro), an
alternative is to represent (9.171) as a convolution integral. The evaluation of pet) is presented
in the next section.
9.3.3. Orthogonality Catastrophe

The X-ray absorption creates a core hole in the midst of the electron gas. It also creates
an additional conduction electron, but the latter process has been discussed in the previous
section. Now consider the impact on the N ~ lO23 other conduction electrons of this sudden
appearance of a core hole. The core hole is again represented as a static potential V. The
physics is to investigate the response of the free-electron gas to the sudden appearance of a
new potential V. The result has been labeled a catastrophe because the transition is forbidden.
The word forbidden is used in an unusual way, which is now explained.
The important physics is that the ground state wave functions of the conduction electron
system, with and without the core hole potential V, are orthogonal. The transition is forbidden
because it is the transition between two orthogonal states. The X-ray absorption is observed in
actual metals, because the conduction electron system is excited by the creation of electron-
hole pairs. The pair creation is a symmetry-breaking process, which allows the forbidden
transition to be observed. The orthogonality was first suggested by Hopfield (1967, 1969) and
proved by Anderson (1967).
It is important to realize that the orthogonality between the two wave functions, with and
without the core hole potential, is between the N-particle wave functions in the limit where
N ~ 00. The single-particle wave functions are not orthogonal. As a simple example, assume
that in the ground state of the electron gas, before the appearance of the core hole, the
conduction electrons can be described by single-particle wave functions with no potential,
such as plane waves. The derivation follows Anderson who considered only s waves, so that
the wave functions have the form (kr) = sin(kr)jkr. In the Hartree-Fock approximation, the
N-particle wave function can be described by an N-dimensional Slater determinant of these
orbitals. The spin indices are omitted, so write the initial state of the system as
(k1rl) (k1r2) (k1rN)

1 (k2r l) (k2r2) (k2rN)
<l>;(rl' r2 ... rN) = .jN! (9.172)
(kNr1) (kNr2) (kNrN)
After the appearance of the core hole potential, the conduction electrons adjust their indi-
vidual wave functions to the presence of this potential. Far from the potential region, the wave
functions can be described by '(kr) = sin(kr + b)jkr, where b is the s-wave phase shift for
one-electron scattering from the core hole potential. The new N-particle wave function is just
a Slater determinant with these new orbitals <1>':
'(k1r1) <1>' (k 1r2) '(k1rN)

I <1>' (k2r l) <1>' (k2r2) <1>' (k2 r N)
<l>j(rl, r2 ... rN) = .jN! (9.173)
'(kNrl) <1>' (kN r 2) <j>'(kNrN)
During the X-ray transition, one takes the matrix element between the initial and final states of
J
the system. In a one-particle picture, the matrix element d3r<l>j(k1 r)E • pcCr) is between the
core wave function <l>e(r) and a conduction electron <l>j(kr). The many-particle calculation
computes this matrix element between the N-particle states. It should include the overlap
between these two Slater determinants,
S= J d3rl ... d 3rN(J)j(rl ... rN)(J)i(rl ... rN) (9.174)
which has been shown to be

S =N- rx / 2 (9.175)
02
1X=2- (9.176)
1t2
The many-particle overlap S turns out to be a negative power of N, where the exponent is a
function of the phase shifts. This overlap vanishes in the limit where N --+ 00 and is small for
real systems which have N '" 1023 and IX is typically 0.1-0.2.
There is no orthogonality between the individual matrix elements of single-particle
orbitals <p(kr) and <P' (kr). This single integral is nonzero. When these overlaps are evaluated
in the determinant for all the possible combinations, the result is asymptotically zero as
N --+ 00. The analogous situation occurs in atoms but on a less drastic scale. For an atom
with N electrons, the optical absorption can cause one electron to change from one energy
state to another. The matrix element for this process is calculated for the full N-particle wave
function of the atom. The N-particle matrix element can be reduced to a one-particle matrix
element between the two primary states which are the initial and final levels of the electron,
times a factor S which gives the overlap of the other N - 1 electrons. They each change their
orbitals a small amount during this change of state by one electron, and the N - 1 overlap
function has a typical value of 0.95. Then S is near unity, and the N-particle overlap is a small
effect in atoms with a few electrons. The metal case is the extrapolation to N '" 1023 elec-
trons, where the product of one-particle overlaps, each slightly less than unity, eventually
produces a vanishing matrix element.
It is necessary to turn these ideas into a dynamical theory of the absorption process.
Nozieres and deDominicis (1969) started by examining the Green's function for the core hole:
(9.177)
They manipulate this correlation function to show that it can be treated as the many-electron
response to the sudden switching on of the core hole, in the same way that the phonon
response to sudden switching was studied in Sec. 4.3.5. The core hole creation operator d t I}
acts on the ground state I} to create a single core hole. The operator exp( -itH) on this state is
represented by the final state potential with the core hole:
(9.178)
The core hole destruction operator d destroys the core hole state. The last time operator
exp(itH) operates on a configuration with no core holes, so that it is Hg for the ground state.
The hole core Green's function can be exactly given by
Gh(t) = _ie-itmrE>(t)(leitHge-itHfl} (9.179)
The core hole operators (d, d t ) are now omitted because they have done their work of
determining the proper order of the Hamiltonian operators. There are several different ways to
evaluate the time dependence of this function. One way is to recognize that the Hamiltonian
Hf = L j hj is a summation of the single-particle Hamiltonians for the individual conduction
electrons. The hole Green's function can be expressed as a Slater detenninant analogous to
(9.166) for the absorption rate. In fact, it is just the function pet) in (9.169) which is the matrix
element over the ground state orbitals of exp( -ithj )
pet) = (leitHge-itHfl) = eitEg det[<ppp,(t)]P.P'<k F (9.180)

Gh(t) = _i0(t)e- itwT pet) (9.181)
<ppp,(t) = (ple- ithj Ip') (9.182)
The evaluation of this detenninant is one possible way to find the core hole Green's function.
Methods of numerical evaluation have been discussed by Schi:inhammer and Gunnarsson
(1978).
A second way to find the core hole Green's function is by a linked cluster expansion. The
correlation function in (9.181) may be written as a time-ordered operator, which is evaluated
by the techniques described in Sec. 3.6:
(9.183)
(9.184)
The first tenn in the exponential resummation F J(t) is linear in t, and is a self-energy tenn:
(9.185)
The exact self-energy of the core hole is known from Fumi's theorem in Sec. 4.1.3 to be
Ei
2h2
= --2:: (21 + 1)
Jk F
kdko[(k) (9.186)
m [ 0
The factor V(k, k) is the first tenn in an expansion which should give Fumi's result Ei when
all tenns are summed. It seems reasonable to use Fumi's result (9.186) for the self-energy,
which includes all the tenns linear in t in higher order. The threshold energy for the X-ray
transition was previously given as ffiT, which is the value in the one-electron approximation,
which neglects the final state interactions. They provide a self-energy Ei which further lowers
the threshold by this amount. The new threshold energy in the interacting system is
roT = ffiT + E i , which accounts for the increased binding of the core hole due to the polar-
ization of the electron gas around it.
The second tenn in the exponential series F 2 (t) is the first tenn which is interesting,
since it predicts an orthogonality catastrophe:
(9.187)
(9.188)
The operators in the brackets are paired, according to Wick's theorem, into Green's functions
of time. The time integrals are then evaluated directly:
(9.189)
x [9(t l - t2) - nd[9(t2 - t l ) - n2] (9.190)
F 2(t) = ~L
2
IV(kl' k 2)1 2 [~ - nl (1 - n2; (1 _ eit(~d2))]
V k 1 k 2 0" SI - S2 (SI - S2)
!
The factor of in front vanished, because each term in the final result appears twice. The
summation over spin index (J can be taken to produce another factor of 2 in front. The first
term, which is linear in time, is dropped because it contributes only to the self-energy, which
is already known from Fumi's theorem. Consider the expression
This term does predict an orthogonality catastrophe and has been investigated extensively; see
Minnhagen (1977) for an evaluation. Here the topic is discussed briefly. Rewrite this
expression by assuming that the hole potential V(k, k') depends only on the difference of its
arguments V(k - k') and change variables to k2 = kl + q:
(9.191)
1
Re(u) =- L V(q)2 A(q, u) (9.192)
nv q
(9.193)
The factor A(q, u) is recognized as the imaginary part of the polarization diagram p(l)(q, u) of
a single-electron bubble. It was evaluated in Sec. 5.5:
A(q, u) = -Im[p(I)(q, u)] = u( m2 ) o (2kF - q) (9.194)

2nq
The important feature of this result is its linear dependence on the frequency u, which is valid
at small u. The wave vector integration over q can be done for realistic potentials V(q). A
crude approximation to the result is 2(NFVO)2 where NF = mkF/2rc2 is the density of single
spin states at the Fermi energy, while Vo is the magnitude of the electron-hole potential.
(9.195)
(9.196)
The upper limit of u is ~o, which is determined by the range over which ReCu) is linear in u.
This range is approximately ~o rv EF , as shown in Sec. 5.5:
(9.197)
The remaining du integral is interesting. The integrand has the factor Io du/u which could
diverge as In(O). In this case it does not diverge, since the other factor (l - e- iut ) rv iut in the
limit where u --+ O. The integral is finite. However, for large times (~ot » 1) this cutoff of the
logarithmic divergence happens at lower and lower frequencies. This dependence can be
approximately expressed as
du
J
~o
F2 (t) ~ -g - ~ -g In(l + it~o) (9.198)
I(i! U
The logarithm has the correct two limits of i~ot at small time and as In(it~o) at large times.
This approximation completes the evaluation of the renormalization factor:
pet) = exp[-itEi - g In(l + it~o)] (9.199)
Its Fourier transform determines the spectral function of the core hole:
(9.200)
The real part of the integral is the same as the integral plus its complex conjugate:
(9.201)
The variable change t --+ -t in the second term, along with p*( -t) = pet), brings us to the
equivalent expression
(9.202)
(9.203)
(9.204)
FIGURE 9.11 Path of integration in evaluating (9.205).
The integral is evaluated by a contour integration. Change the integration variable to z = ~ot.
There is a branch point at z = i. The branch cut is drawn vertically from this point, up the
imaginary axis, as shown in Fig. 9.11:
Ah = -1 Joo eizO
dt-:-:------,:-;; (9.205)
~o -00 (1 + iz'f
For Q < 0, the integral (-00, 00) is closed by a contour in the lower half plane (LHP). This
closed contour encloses no poles or branch cuts and so gives zero and shows that Airo) = 0
for Q < O. The other situation is Q > 0, and here the contour of integration is closed in
the UHP, as shown in Fig. 9.11. The contour encloses no poles, so the contribution on the
real axis (-00, 00) is equal to that along the cut. For the integration along the branch cut,
change variables to z = i(1 + y). The denominator in (9.205) becomes (1 + iz'f =
[1 - (1 + y)]g = [-y]g = ~ exp(±ing), where the choice of (±) in the exponent depends on
the side of the branch cut.
The integrals along the branch cut give
Ah(ro) = 0(Q) iJoo dYe-n(1+Y)(e-iltg _ eiltg ) (9.206)

~o 0 ~
= 0(Q)r(1 _ ) 2 sin(ng) e- n (9.207)
g ~o QI-g
The shape of this spectral function is shown in Fig. 9.12. It is zero for Q < 0 and nonzero for
Q > O. It diverges as a power law for Q ...., 0+, with the exponent (1 - g). It is easy to cheek
that the area under the spectral function is 2n:
I = Joo dro A (ro) = sin(ng) r(1 _ )Joo dQ e-n (9.208)

-00 2n h n g 0 QI-g
1 = sin(ng) r(1 - g)r(g) (9.209)

n
The last equation is an identity among gamma functions.
The response of the electron gas has a dramatic effect upon the spectral function of the
hole. The spectral function has a power law divergence QI-g near the threshold energy. In the
absence of interactions (V = 0), the spectral function is a ddta function Ah(ro) =
2n<>(ro - roT)' As illustrated schematically in Fig. 9.l3(a), the influence of interactions is to
o 1.0
FIGURE 9.12 Spectral function of the hole Green's function. The orthogonality catastrophe causes the power law
distribution and eliminates any delta function character.
change the delta function into a power law divergence. All quasiparticle behavior, in the form
of a remnant delta function, has vanished.
In X-ray absorption, the unperturbed spectrum for V = 0 is a step 8(ro - roT) rather than
a delta function, which is illustrated in Fig. 9.13(b). How does this step become modified by
the potential V in the orthogonality catastrophe? The answer is that the threshold becomes a
power law ~ which goes to zero as Q ~ O. The threshold behavior may be derived in the
following way. The Fourier transform of a step function 9(ro - roT) is r 1 :
=J
OO dt e it( ro-roT )
8(ro - roT)
-00 2m
. t - zb
. (9.210)
The orthogonality catastrophe modified this spectrum by the inclusion of the term pet) in the
integrand. Call the resulting X-ray spectra X(ro):
=J
oo dt 1 eitQ~o
(9.211)
X(ro) -00 27ti t - ib (1 + itSo)g
The easiest way to evaluate this integral is to note that its derivative is just the hole spectral
function
dX(ro) = ~A (ro) (9.212)

dro 27t h
i
WO
V=O
(0 )
J n n
v=o wo
( b)
L n
L n
FIGURE 9.13 The effects of the orthogonality catastrophe upon spectral distributions. (a) The delta function of a
spectral function is changed to a power law divergence. (b) The step function in absorption is changed to a converging
threshold.
This equation has the solution
X(ro) = J
ro dro' ,
-2 Ah(ro) = r()
0(Q) fO dQ,e-O'
~ (9.213)
-00 11: g 0 (Q) g
The behavior near threshold may be deduced by examining the integral for very small values
of Q. Then the slowly varying term exp( -Q') can be ignored, and the integral gives
0(Q) JO dQ' gg0(Q)

(9.214)
X(ro) = r(g) 0 (Q,)l-g = r(1 - g)
Equation (9.214) shows that the absorption behaves as Qg near threshold. The threshold has a
converging behavior, with the absorption starting at zero when Q = O. It is sketched in Fig.
9.l3(b).
The absorption is zero at the threshold frequency Q = 0+ because of the orthogonality
catastrophe. The transition is not allowed because of orthogonality and has a zero probability.
The absorption is nonzero for Q > 0 because of the creation of a number of electron-hole
pairs in the system. This serves as a symmetry-breaking process, which makes the transition
allowed.
The role of electron-hole pairs in the X-ray transition was elucidated by Schotte and
Schotte (1969). The interaction term between the core hole and the electron gas can be written
in terms of the electron density operator
(9.215)
(9.216)
The density operator can be represented by the boson operators for the electron-hole exci-
tations of the electron gas p(q) 4- (b q + b~q). The Hamiltonian between the core hole and
the electron gas is transformed into a problem of a localized level and a boson system. This
model Hamiltonian can be solved exactly as the independent boson model of Sec. 4.3. The
solution is entirely analogous to the phonon core hole system. The response of the boson
system to the sudden appearance of the core hole is described by including the factor of
exp[-<p(t)] in the time integrals The factor <p(t) for phonons is presented in (9.144) and is
rewritten as
(9.217)
The self-energy term iLt is put first, while the other term is the transient behavior. The
approximate response of the electron system to the core hole was given earlier as Fit) in
(9.192). It has exactly the same form as the preceding, except the phonon system is at nonzero
temperature. The electron-electron response at nonzero temperature is
(9.218)
The quantity Rph(u) for the phonon system is the phonon density of states, which is weighted
by the matrix element for the process. Similarly, the factor ReCu) serves the same role for
electron-electron interactions. It is interpreted as the effective coupling to the electron hole
pairs. The equations are exactly alike for the response of the core hole phonon and core hole
pairs, and enter into the spectral function for absorption in the same fashion as factors
exp( -<p(t)). This similarity is reasonable, since both the phonons and electron-hole pairs are
boson systems which respond in the same way to the appearance of the core hole.
The only difference between these two boson systems is their respective coupling
constants Rph(u) and Re(u). This difference is big, since they behave quite differently at small
values of u. The electron-electron interactions were shown to have a linear relation Re ex: u
which leads to power law behavior in A(ro). The phonon system is governed by the acoustic
phonons at small values of u and usually behaves as Rph ex: u2 in metals. Consequently, the
exponential factor for the phonon system does not diverge as "-' In(it~o) at large ~ot but goes to
a constant except for the self-energy term. The spectral function A(ro) from the phonon part
alone is a dull function, which was represented as a Gaussian. The core hole phonon and core
pair couplings cause quite different behaviors in the time response of the system and affect the
spectral shape differently. The difference arises from the dissimilar nature of their coupling
function Rph(u) and ReCu).
The electron-electron response is an example of a type of phenomenon known as an
infrared divergence. The divergence occurs in the limit where u -+ 0, which is the infrared
part of the spectrum. The quantity R/u)/u2 is the probability (at T = 0) of emitting an
excitation of energy u during the sudden switching on of the core bole potential. For example,
the number Nu of excitations created is
Nu = Joo zReCu)
du
"-' g
J~o du
- = -gln(O) (9.219)
o u 0 u
The average number Nu of bosons around the particle was discussed earlier for polarons in
Sec. 4.1. It is infinity for the electron-hole excitations, which shows that an infinite number
are created. This infinity does not imply a physical disaster, since all measurable quantities are
nonzero. For example, the energy which is released by the creation of these electron-hole
pairs IS
(9.220)
which is nonzero. However, a divergence in the number Nu of excitations is usually an

indication that the spectral function will have an unusual shape, since quasiparticle behavior is
eliminated. A further discussion is given by Hopfield (1969) and Mahan (1974). The phonon
system does not have an infrared divergence in the X-ray problem in metals, although it can in
semiconductors where the piezoelectric interaction is not screened.
The exponential series for F(t) = L; F;(t) in (9.183) has been approximated by the term
F 2 (t) plus iE;t. Further terms in the series also contribute to the orthogonality catastrophe and
should be summed in order to obtain an accurate description of the phenomenon. The exact
method of summing all of these terms has been given by Mahan (1982), and more elegantly
by Ohtaka and Tanabe (1983). They provide an exact expression for ReCu).
Several other types of results have been obtained. The most important is by Nozieres and
deDominicis (1969), who show that the exact coefficient of the term which diverges like
In(itso) is
limF(t) = -itE;
1-+0
+ a In(itso) (9.221)
a= L [bl~F)J2 (9.222)
ms,m,,l
The same phase shifts bl(kF) were mentioned earlier and arise from the scattering of the
conduction electron. with wave vector kF from the core hole. The core hole is in the center of
a spherical coordinate system, and the conduction electron states are described by the
quantum numbers of ms, I, ml and k. The phase shifts are usually taken to be independent of
ms and ml, in which case the coefficient becomes
a = 2 "f(21 + l)[bl~F)r (9.223)
The coefficient a should be used in describing the asymptotic behavior of the orthogonality
catastrophe very near the threshold.
The orthogonality catastrophe is an important feature of many-electron physics. The
process of switching on a potential occurs in many circumstances, and the physics can be
applied to a variety of phenomena. For example, the Kondo effect has this feature, which
plays an important role in its theory. Whenever the potential is switched on, the system
responds by making electron-hole pairs, which affects the spectral response of all correlation
functions which are being evaluated.
9.3.4. MND Theory

The absorption of X-rays in a metal causes several different types of excitations and
phenomena. Two have been discussed which cause power law behavior at the absorption
edges: excitons and the orthogonality catastrophe. An important contribution was made by
Nozieres and deDominicis (1969), who solved the two processes together. Their solution is
asymptotically exact near threshold roT = roT + E;. It expresses the power law exponents as
functions of the angular momentum I of the conduction electron in its final state. The many-
body correction to formula (9.142) with two angular momentum channels A/±! is
A(ro) = e(ro - roT) [AI+! ( _ _S0___)""+1 + AI _ 1

ro - roT
(s )"'I-IJ
_ _0___
ro - roT
(9.224)
2b l (kF )
al =----(X
1t
(X = 2 "f(21 + 1)[bl~F)r (9.225)

The power law exponents IY.I are the sum of two terms. The first is 28z1rc, which is from the
exciton phenomenon; this term is usually positive in free-electron metals and tends to make
the edges diverge and become singular. The other factor IY. comes from the orthogonality
catastrophe and tends to make the edges converging at threshold. The final value of IY.I for an
absorption edge depends on the difference of these two quantities and may be either positive
or negative.
There have been many calculations of these phase shifts and exponents for comparison
with those deduced from experiments. Typical results for sodium are shown in Table 9.2, as
calculated by Minnhagen (1977). A recent calculation is by Ohmura and Ogiwara (1994).
Minnhagen used a pseudopotential, which was screened by a Singwi-Sji:ilander dielectric
function. The phase shifts should always obey the Friedel sum rule in Sec. 4.1,
2
1= - 'L(21 + 1)81(kF ) (9.226)
rc 1
which is just a statement that the screening around the core hole must be one unit of charge.
The exponents are typical for simple metals. The X-ray absorption from a p shell, where the
initial angular momentum is I = 1, has the final values for the conduction electron of I = 0
and I = 2. The asymptotic limit of the Mahan-Nozieres-deDominicis (MND) theory for
0) > roT is
The first term is the s-wave channel. The exponent 1Y.0 is positive, so this term diverges at
threshold. The second is the d-wave channel where IY.z < 0 and the term goes to zero at
threshold. The summation of these two terms diverges at threshold. The theory predicts that
absorption edges are divergent for electrons from p shells, which is generally observed in
free-electron metals such as Na, Mg, AI, and K. On the other hand, if the initial core state is
an s shell with 1= 0, the final state of the conduction electron has 1= 1, IY.I ~ -0.05 for
sodium. This value is so small that it predicts there is no measurable effect from the many-
electron response. This prediction is also in accord with experimental observations. The value
ofa l is small because of the partial cancellation of28 1/rc = 0.14 and IY. = 0.19. It seems to be
a general feature of free-electron metals that the transitions from core hole s states shows no
additional peaking or rounding of the threshold, in agreement with the predictions that IY.I is
very small. MND theory seems to qualitatively explain the behavior of the edges of simple
metals.
Figure 9.14 shows the experimental results of Callcott et al. (1978) for the absorption
edge of the L Z ,3 shell of metallic sodium. The p-shell is split by the spin-orbit interaction into
the P3/Z and Pl/z components, which comprise the L3 and L z shells. In sodium, this splitting is
TABLE 9.2 Phase shifts and exponents for

sodium. IX = 0.19. From Minnhagen (1977)
(used with permission)
IXI
o 0.90 0.38
1 0.22 -0.05
2 0,01 -0.19
0.17 eV and accounts for the double edge shown in absorption. Each of these two edges
should have a power law singularity. Figure 9.14 also shows the deconvolution of the sodium
spectrum into the part due to the one-particle spectrum and broadening processes such as the
Lorentzian width from Auger and the Gaussian width from phonons. The absorption edges
show definite spikes which are from the edge singularities. The spikes are much smaller in
emission, although that is partly from the increased phonon width in emission due to the
incomplete relaxation phenomenon, as mentioned earlier. The value of the edge singularity
exponent deduced by Callcott et al. to fit their absorption data is 0(0 = 0.25, which is smaller
than the 0(0 = 0.38 predicted from the phase-shift calculations. The theoretical values of 0(0
have been computed by many different workers, and values similar to those shown in Table
9.2 are obtained by many different groups, so the theoretical values seem reliable.
There have been several explanations for this lack of detailed agreement between the
measured and theoretical exponents: (1) omission of frequency dependent screening in the
theory; (2) exchange type of effects between the core hole spin and electron spin, as discussed
by Girvin and Hopfield (1976); (3) scattering of the core hole among its different orbital states
in the L shell; and (4) the asymptotic solution (9.217) is not valid very far from threshold.
This subject is still an active area of research. A summary of the present status of this field
seems to be that the edge singularities are observed in some thresholds of simple metals, and
the refinements of the theory are still under investigation.
No. Theory (a)

r ,fl =·09.a=.25
".,..- ..... ,
,,"' .... ' .............
.- .-
/ """Gupta - Freeman
g ... ' ..... "." '
H
~ (b)
'c
~
J:l
S
::t
u:
UJ
8
z
Q
t-
o.
cr
0
CIl
CD
«
30
(eV)
FIGURE 9.14 The X-ray absorption edge ofthe L2. 3 shell of metallic sodium. The data is shown in part (b). Part (a)
shows how the band structure prediction of Gupta and Freeman (1976) is convoluted with the edge singularity
(IX = 0.25) and broadening functions to fit the spectra. Source: Callcott et al. (1978) (used with permission).
9.3.5. XPS Spectra

X-ray photoelectron spectroscopy (XPS) measures the line shapes of photo-electrons
excited from the core levels of atoms. A photon, which is usually in the kilovolt energy range,
excites a core electron to very high kinetic energy. Since the binding energies EB of the L-
shell electrons in N a, Mg, and Al are only 50-100 eV, the final kinetic energy of the election is
Ef = lim - EB . In a one-electron model, all the photoelectrons excited by the same frequency
m would have exactly the energy Ef , and a measurement of the distribution of kinetic energies
of the photoelectrons would show a delta function at this value. The actual kinetic energies of
photoelectrons, measured from simple metals, show the main peak at energy E "-' Ef plus
satellite peaks which correspond to the emission of bulk and surface plasmons (Pardee et aI.,
1975). The main peak at E "-' Ef has a nonzero width, which must be partly caused by the
usual mechanisms of (1) Auger lifetime of the core hole and (2) phonon broadening due to the
coupling between the core hole and the phonons.
Doniach and Sunjic (1970) predicted that the orthogonality catastrophe should make the
shape of the main line, at E "-' Ef , asymmetric with an asymmetry parameter which is a direct
measurement of the index IX. Their argument is that this experiment directly measures the
spectral function of the core hole, since the outgoing electron has too much energy to be
affected by the interaction with phonons or electron-hole pairs. In the calculation of the Kubo
formula for the absorption rate, the conduction electron operators [C~a, Cpa] can be removed
from the correlation function on the grounds that the electrons leave the core hole too rapidly
to be influenced by exciton or other final state processes. The correlation function is factored:
1
A(m) = ffi Joo
-00 dteiIDtCl(t),/(O)) (9.227)
= -1 Joo dte iIDt L w(k)*w(k/)(Ck,S<t)C~,s,(O)) (ds(t)d;,(O))

co -00 kk'ss'
(9.228)
The Kubo formula for the absorption becomes proportional to the spectral function
Ah(m - ~k) for the core hole. In this experiment, the photon energy m is usually fixed, and
measurement is made of the distribution peE) of final kinetic energies E == ~k - Ew of the
conduction electron, where Ew is the work function of the metal. XPS spectra can now be
measured with millivolt energy resolution, although the electron energy E is often over a
kilovolt. The matrix elements w(k) are a constant over the narrow energy range of ~k which
are measured for each XPS line. The probability peE) is just the spectral function of the core
hole Ah(m - E - Ew)
pee) = w 2f(E - Ef ) (9.229)
fee) = J~oo dtexp[-itE - !rjtl - y2 t2 - 1X1n(it~o)] (9.230)
where Ef = lim - Ew. It has the power law singularity in (9.207) and is illustrated in Fig.
9.l2, which is then broadened by the phonons and the Auger decay. The line shape is
predicted to be asymmetric because of the orthogonality catastrophe, and the degree of
asymmetry is just given by the index IX. The formula used for interpreting the experimental
line-shape function feE) contains the Lorentzian width from Auger processes, the Gaussian
width from phonons, and the asymmetry from the electron-hole pair creation.
An early experimental result is shown in Fig. 9.15 from data of Citrin (1973) for metallic
sodium. The dashed line indicates the excess line shape on the low-energy side. Many more
spectra have been published by Citrin et al. (1977). One feature of the experimental results is
that the same index rt fits the asymmetry for different core holes in the same atom: the same rt
is found for the different shells K, L2 , L3 , etc. The ion with a core hole looks the same, to
electron-hole pairs which are outside the ion, regardless of the particular state of the core
hole.
The XPS line shape is asymmetric, because of the emission of electron-hole pairs during
the creation of the core hole. The photon energy is fixed at /'too, and any energy used to make
the pairs must be subtracted from that carried away by. the conduction electron. The
conduction electron has a high probability of having energy less than Ef . This explanation
accounts for the sign of the asymmetry and explains why the line tails to the left in Fig. 9.15.
These asymmetric line shapes had been observed for many years, as reviewed by Parratt
(1959). He speculated then that the "excess width on the low energy side is essentially
8 No 2s in No
(a)
en
o No 2s in NoOH
z
ct
en (b)
:::::>
o
i: 2
75 70 65
BINDING ENERGY leV)
FIGURE 9.15 X-ray photoelectron spectra from sodium 2s electrons in (a) metal and (b) hydroxide. The metal
spectra is asymmetric, with additional contributions on the low-energy side due to pair production. These are lacking
in the insulator. Peaks to the left are plasmons in the metal. Source: Citrin (1973) (used with permission).
attributable to transitions between excitation states of the valence-electron-configuration

type." The quantitative theory of Doniach and Sunjic has verified this hypothesis in great
detail and provides the most direct method of measuring the orthogonality catastrophe.
PROBLEMS
1. Derive the nonzero temperature fonn of the free-polaron absorption (9.51) assuming Maxwell-
Boltzmann statistics. To order rx, an exact result can be obtained in tenns of Bessel functions
Kl(!~lro±rool)·
2. Consider the force-force correlation function for scattering from an impurity. Discuss whether the
multiple. scattering from an impurity can be represented by a T matrix or similar series.
3. Derive the Wilson-Butcher fonnula (9.64) from the golden rule. Find the matrix element (fIE' pli)
by writing both initial Ii) and final If) Bloch states to first order in the potential VG.
4. Consider a Hamiltonian which is the homogeneous electron gas plus the crystal potential
L VGP(G). Discuss the summation of tenns which occurs in higher order when evaluating the
force-force correlation function.
5. Where do the final state interactions for the Coulomb scattering of the electron and hole appear in
fonnula (9.64) for interband transitions?
6. Show that the rate of photon emission in an insulator with a nonequilibrium distribution of electrons
and holes is governed by the matrix element Iw/k)1 2 as in (9.132). Assume the initial state ofthe system
at zero temperature is Ik) = d t ariD), and use arguments analogous to those following (9.139).
7. Write out the correlation function n(I) in (9.105), with one vertex diagram, for the frequency-
dependent screening function (7.171). Do the Matsubara summations.
8. Show there is an infrared divergence in the X-ray response resulting from the piezoelectric electron-
phonon interaction in insulators (Sec. 1.3) when used in the response function (9.217).
9. Consider the X-ray edge problem with an interaction tenn Hsd in Sec. 1.4.2 between the conduction
electrons and the core hole. What is the contribution of this tenn to the orthogonality index rx from the
cumulant F 2 (t) (Girvin and Hopfield, 1976)?
10. Show that the imaginary part of the retarded correlation function (8.34) has the following sum rule
relating to the average kinetic energy:
J oo
-00
dro e'2 no
-2 ns(-ro)Jm[n(ro)] = -EKE
n m
(9.231)
Chapter 10
Superconductivity
The theory of superconductivity was formulated by Bardeen, Cooper, and Schrieffer (1957)
and is called the BCS theory. It very successfully describes the superconducting properties of
weak superconductors, such as aluminum, which are weak because of the small strength of
the electron-phonon interaction. Further refinements of the theory have led to the strong
coupling theory of Eliashberg (1960) which describes well the properties of superconductors
such as lead. The distinction between aluminum and lead is roughly determined by the value
of the electron-phonon mass enhancement factor A, as shown by McMillan (1968). The BCS
theory is discussed first. It must rank as one of the great successes of many-body formalism,
since the theory provides detailed agreement with experiments. This agreement is a refreshing
change from most comparisons between many-body theory and experiment, where the results
often depend upon vertex corrections, correlations, and computer simulations. The beauty of
BCS is that it is, mathematically, a simple theory which is exceedingly accurate. The reason it
works is that the basic coupling forces are weak, and mean field works well.
The basic idea of BCS theory is that the electrons in the metal form bound pairs. Not all
electrons do this, but only those within a Debye energy of the Fermi surface. The bound states
of the electron pairs are not described by simple orbitals such as used for the hydrogen atom
or positronium. The pair state, and the entire ground state of the superconductor, requires a
many-body description.
The Debye energy enters into the ground state description because the attractive forces
between electrons, which are responsible for the pair binding, are due to the electron-phonon
interaction. Frohlich (1950) was the first to realize that electrons could interact by exchanging
phonons and that this interaction could be attractive. He was the first to suggest that super-
conductivity was caused by the electron-phonon interaction. The phonon dependence would
explain the experimental observation that the transition temperature Tc is a function of the ion
mass for different isotopes of the same metal. This isotope effect was discovered for the metal
Hg by Maxwell (1950) and Reynolds et ai. (1950), where the dependence was proportional to
I1Tc/Tc ~ -oM /(2M). The BCS theory explains this in detail, since it shows that the tran-
sition temperature is proportional to the Debye frequency kETc ~ roD ex M- 1/ 2. The isotope
effect verified the Frohlich hypothesis that the electron-phonon interaction caused super-
conductivity.
Another piece in the theoretical puzzle was supplied by Schafroth (1955), who showed
that a charged boson gas, when undergoing a Bose-Einstein condensation, would exhibit
many of the superconducting properties known at that time such as the Meissner effect. His
627
628 Chap. 10 • Superconductivity
theory does not explain other features, such as the energy gap in the excitation spectrum.
Schafroth speculated that in the superconductor the "bosons are resonant two electron states."
The BCS theory uses a similar mechanism, since the paired electrons behave, in some
respects, as bosons. The BCS theory was not conceived in a vacuum but among many related
ideas, each of which contained some element of truth.
BCS was the first theory to explain superconductivity in metals and also made a number
of remarkable predictions. The foremost was that an energy gap existed in the excitation
spectrum of the superconductor. The actual observation of this energy gap by electron
tunneling (Giaever, 1960) provided a dramatic verification of the theory, although BCS earlier
argued that the thermodynamic data supported the existence of a gap. Many different
experiments in weak superconductors have shown that the original version of the theory is
correct in its many details. An extensive comparison between theory and experiment is
provided in the two volumes of Superconductivity edited by R.D. Parks (1969) and in the
books by Rickayzen (1965) and Schrieffer (1964).
In 1987 superconductivity was discovered in materials with planes of copper oxide,
which is now the field of high-temperature superconductivity. The highest value of the
transition temperature at ambient pressure is Tc = 133 K in the Hg-Ba-Ca-Cu-O system
(Schilling et al., 1993). Although much is known about these materials, the basic mechanism
causing superconductivity at high temperatures is still unexplained. Even the symmetry of the
energy gap is uncertain. Most experiments support the d-wave picture, but some still support
s-wave symmetry. The electron seemed to be paired in spin singlet states, with a very short
coherence length. It is also interesting that experiments which were important in under-
standing elemental superconductors, such as electron tunneling or infrared absorption, proved
to be difficult to interpret for the cuprates.
10.1. COOPER INSTABILITY
The first inkling of the BCS theory was a letter by Cooper (1956), who pointed out that
the ground state of a normal metal was unstable at zero temperature. A normal metal is
defined as one which is neither superconducting nor magnetic. The instability is an indication
that the metal prefers to be in another state, in this case the superconducting one. The
demonstration of an instability does not provide a description of the superconducting state,
but it did suggest that the instability was caused by the scattering between pairs of electrons,
where the scattering potential is the exchange of phonons.
Two electrons can scatter as shown in Fig. 10.1. The screened interaction between two
electrons was derived in Sec. 6.4:
2COAM~
V (q co) - - - + --;:-------'--;:_
Vq
(10.1)
S' - e(q, co) e(qf[co2 - co,.,(qi]
It has two terms. The first is the screened Coulomb interaction. The theory of super-
conductivity is applied at low temperatures, where the energy exchanged between particles,
while scattering, is also low. Requirements of crystal stability require that this interaction be
repulsive at zero frequency (Allen et al., 1988). There have been numerous attempts to bind
two electrons using just the first term which has only an electron-electron interaction. Then
the plasmon is the boson which is exchanged between electrons. Careful calculations,
Sec. 10.1 • Cooper Instability 629
( 0) (b)
FIGURE 10.1 The pairing force in a superconductor is due to the exchange of excitations, such as phonons, between
pairs of electrons. This interaction is frequency dependent and has attractive regions for nonzero frequencies.
including local field corrections, show that the binding force from plasmon exchange is
negligibly small.
The second term in (10.1) is the screened electron-phonon interaction. It is, on the
average, weaker than the repulsive Coulomb interaction. However, for frequencies near to the
Debye (ro < roD) the energy denominator becomes small and negative, which causes a
relatively large interaction over this narrow range of frequency. The frequency dependence of
V.( q, ro) is illustrated schematically in Fig. 10.1 (b). It may be possible for two electrons to
bind if they can construct a relative wave function which selectively uses the frequency region
which is attractive. Not all metals are superconductivity; the alkali metals are not. They have a
weak electron-phonon interaction. So the attractive interaction needs some minimum strength
in order to cause superconductivity.
There are two ways to proceed with the theory. One is to use the full frequency-
dependent interaction, which sums over all phonon wave vectors and modes. This realistic
calculation is done in a later section. The other possibility is to replace (10.1) with a model
interaction of the form
-Vo for I~ql < roD

VsCq, ro) = { 0 (10.2)
for I~ql > roD
This potential is constant and attractive (VO > 0) up to a cutoff energy which is of the order of
the Debye energy roD of the solid. This second form of the interaction permits a much simpler
discussion of the theory and allows the physics to be introduced more easily. The initial
discussion of superconductivity uses the model potential in (10.2), which follows the
historical pathway of Cooper and BCS. Later, when the physics is better understood by the
reader, the theory is redone from the beginning using the realistic potential (10.1). The latter
is called strong-coupling theory.
Cooper's model of a normal metal at low temperature was a free-electron system. In the
limit of zero temperature, the Fermi surface has a sharp step in energy. The electrons are
allowed to have a weak attractive interaction as given in (10.2). Consider the mutual scattering
of two electrons. Assume they initially have states of equal and opposite momentum k and
-k, or zero center of mass. Why this is necessary for the Cooper instability is shown later. It
is also assumed the particles have opposite spin states t and . )" so that exchange scattering
does not occur. The interaction potential does not flip the electron spin and the spin states are
preserved in the scattering process.
Figure 10.2 shows a double scattering event between two electron lines which are
moving in the same direction in time. This process is the scattering in the second Born
approximation, where the first Born approximation is shown in Fig. lO.1(a). Each dashed line
represents an interaction of the type shown in (10.2). If two electrons initially start in opposite
momentum states k and -k, then a momentum transfer q leaves them still in opposite
FIGURE 10.2
momentum states k + q and -(k + q). This process is shown in the figure as the pairs
(k 1, - k 1) in the intermediate state and (k', - k') in the final state of the double scattering. The
effective scattering in the first and second Born approximation is
v:
eff
(k - k') = V(k - k') + J d 3k1 V(k - k1)V(k 1 - k')
(2n)3 2~k - 2~kl
x ([I - nF(~k)][1 - nF(~k)] - nF(~kY} (10.3)
The second term on the right is the contribution of Fig. 10.2. The energy denominator
contains the initial state energy ~k + ~-k = 2~k minus the intermediate state energy 2~kl' The
numerator contains the important factors [1 - nF(~kl )]2 - nF(~kY = 1 - 2nF(~k) which
occur because the two particles, which are being scattered, can only go into the states (k 1 , t)
and (-k l' -!-) if they are not already occupied by an electron. This explains the factors
[1 - nF(~k 1)f. The other term nF(~k)2 1
represents the scattering back into this state,
(k 1 , t) ----+ (k, t), etc., since the result depends upon the net scattering. What is left are the
remaining factors of 1 - 2nF(~kl)' These occupation factors of nF(~kl) playa crucial role in
the theory and are the cause of the instability.
The integral in (10.3) may be evaluated. The key is that the interaction acts only over a
small energy interval near the Fermi energy. Over this interval, of a Debye energy, the
electron density of states in most metals is nearly constant. Change integration variables to
f f
d 3kd(2n)3 = d~IN(~I) and treat N(~ ~ 0) == NF as a constant. At zero temperature the
result is (~kl = ~l' ~k = ~)
Veff(k - k') = V(k - k') + NF V6JroD d~l! ~ ~~~l) (lOA)

-roD 1
The integrand contains the factor! - nF(~I)' and the! term may be ignored. It does not cause
the singularity, and its inclusion leads to little effect. The term in nF(~I) leads to a logarithm
which is singular at the Fermi energy:
roD d~l nF(~I) = JO ~ = -In(~) (10.5)

J
-roD ~ - ~l -roD ~ - ~l WD
The results of the first and second Born approximation may be summarized as
Veff = -VO[1 -NFVoln(w:)] (10.6)
The term -NFVO In(~/wD) is regarded as the vertex correction which results from the
additional scattering between electrons. The scattering becomes very large for electrons near
the Fermi energy I~/WDI « 1. Further insight is gained by considering the sum of diagrams in
~ ~ "-
I
?'
I
"-':--~-.--(?'
I
I
I
I +
I I
+
I I
+ I I + I
I
I
I
I
I
~
I
~ ,/ \, I
/ )--.J.....----'--4.~
FIGURE 10.3
Fig. 10.3. Each additional interaction (dashed line) causes two more Green's functions which
are going parallel in the intermediate state. Each new set of intermediate states has the same
type of integrand, so that a term with (n + 1) ladder diagrams gives a net contribution of
(10.7)
Each term with n ladder diagrams has a factor of [-NF Vo In(~/ ())D)r, and the summation of
these terms produces the series
(10.8)
(10.9)
This series can be summed to produce a net potential with an energy denominator (10.9). This
energy denominator equals zero at
(10.10)
The energy denominator has a pole at this energy, since the denominator can be written as
NF Vo In(~/~o)· In the vicinity of ~o the denominator can be approximated using
~ = ~o + (~ - ~o) and the scattering potential has a pole:
1 1 1 1
Veff = - - = - - :-::-:----::c---::--:-:-:-::
NF In(~/~o) NF In[l + (~ - ~o)/~o]
~ _...".~-,-,o'---c--:- (10.11)
NF(~ - ~o)
This pole is sufficient to cause the instability. The electrons near the Fermi energy will interact
with their pair on the opposite side of the Fermi sea. The mutual scattering produces a pole in
the scattering amplitude, which will make the pair of electrons try to bind together. Of course,
all electron pairs are doing this simultaneously, so that the entire metal undergoes a phase
transition. The existence of this pole depended on the sharpness of the electron distribution
nF(~l) = e(-~l)' If all the electrons near the Fermi energy become paired, one must
reconsider whether this sharp distribution still exists. A theory of superconductivity must self-
consistently determine the properties of bound electron pairs. This derivation is the basis of
the BCS theory.
Another way to describe the instability is as a function of temperature. The calculation is
repeated as a function of temperature. It enters into the electron distribution np(~l) by
changing the step 8( -~l) into a smooth function with an energy width of several kBT. The
influence of nF(~l) can be approximated by expressing the integral (10.5) as
]
ro
D
d~l nF~~l) ::::; _~ln[~2 + (~BTi] (10.12)
-roD ~ ~l 2 OlD
Follow the same steps through the summation of all the ladder diagrams, and conclude that
the effective potential now has an energy denominator of the form
Vo
V ff - - ------;::=====:--- (10.13)
e - 1 + NFVO In[J~2 + (kBT)2/0lD]
At zero energy (~ = 0), Veff becomes singular when the temperature is lowered to the critical
temperature Tc:
kBTc = OlD exp [ - N:VJ (10.14)
This argument correctly predicts the right form for the transition temperature of the BCS
theory, which is
kBTc = 1.140lD exp [-_I_J (10.15)

NFVO
The result is indeed proportional to the Debye energy, in agreement with the isotope effect. Of
course, to explain the isotope effect, it must be shown that the exponent I/(NF Vo) does not
change with the ion mass. It actually does depend on the ion mass M, which is contained in
the electron-phonon matrix element, in the interaction potential Vo. Some experimental
values of Tc for elemental superconductors are given in Table 10.1.
The theory of the Cooper instability should be compared, for example, with the ordinary
binding of two isolated particles. If the two particles are isolated, they do not have to obey the
statistics of a collection of identical particles. Then the scattering theory does not contain any
of the occupation factors; all states may be used as intermediate states since there are no other
particles. In this case the multiple scattering theory was described earlier, in Sec. 9.2, in the
theory of Wannier excitons. The multiple scattering theory may be described by a vertex
function:
(10.16)
TABLE 10.1 Critical temperatures

Tc of elemental superconductors (K).
Mg AI
1.2
Zn Ga
0.91 1.1
Cd In Sn
0.56 3.37 3.73
Hg TI Pb
4.16 2.38 7.22
The solution to this vertex function is equivalent to solving the two-particle Schr6dinger
equation in relative coordinates:
(10.17)
The problem is factored into the relative and center of mass motions:
r = rl - r2' \jJ(rl' r2) = eip.R<!>(r) (10.18)
2
R=-21(rl +r2)' E=L+
2m
E (10.19)
[ - : + V(r)-E]<!>(r) =0 (10.20)
The center of mass motion is plane wave, and the relative motion becomes a one-body
problem. Without the occupation factors, the relative scattering of two particles by an
instantaneous potential is a trivial problem. When bound states occur, they are at negative
binding energy in the relative coordinates, i.e., at E < O. This behavior is in great contrast to
the Cooper instability, where the pole occurs at a small negative energy relative to the Fermi
energy E F, so the pole is at a positive energy E F - ~o. Two electrons cannot really bind at that
energy, since their net energy is positive. The instability occurs because it appears to them as
if they should bind, although if they tried, they would find they could not. The role of the
occupation factors [1 - 2nF(~I)] in the argument of the scattering integral is what moved the
apparent pole out to the Fermi energy.
One can now see the reason the electrons must be paired with opposite momentum. The
instability is caused by the sharpness of the Fermi surface, which is fixed in momentum space.
Two electrons with arbitrary wave vectors kl and k2 interact by exchanging phonons and this
interaction can be attractive. Nevertheless, this does not lead to an instability when the center
of mass motion is nonzero, since the zero energy of relative motion does not coincide with the
location of the discontinuity in the momentum distribution. The center of mass transformation
P=k,+k2, k,=!P+k (10.21)
k = !(k, - k 2), k2 =!p - k (10.22)
k2 p2
+
~k1 2
~k m= - + -4m - 2EF (10.23)
changes the relative scattering integral of two particles into
J(21t)3
d3q V( ) 1 - nF(~p-k-q) - nF(~p+k+q)
q k21m - (k + q)21m
(10.24)
The energy denominator factors into just the difference of the relative energies, as it should,
but the Fermi surface discontinuities are not at (k + q)2 Im . Instead, they are at
(±! P + k + q)2/2m, which gets smeared with the averaging over angles. This smearing
makes the pole disappear rapidly for nonzero center of mass momentum P.
The same results can also be derived from a Green's function analysis. If two particles
interact with an instantaneous interaction, as in Fig. 10.2, and propagate parallel in time or't
space, then the vertex has the combination of Green's functions
(10.25)
X(iro) = r
The frequency spectrum is obtained by integrating this pair over (, - ,I)
d,eiOl(T-T')~(O)(kl" - ,1)~(O)(k2" - ,I) (lO.26)
This integral can be evaluated directly in , space by using the definitions of ~(O) in Sec. 3.2.
Alternately, it can be converted to a frequency summation, which has the form
(lO.27)
1 - nF(~k) - nF(~k)
(lO.28)
iro - ~kl - ~k2
The expression contains one of the standard Matsubara summations which was given in Sec.
3.5. It has exactly the combination of occupation number factors stated in (lO.24). It also has
the same energy denominators. The binding of particles is described by parallel motion in
time.
It is interesting to generalize the Cooper instability to other circumstances in which
similar effects could occur. For example, consider the interaction between two different
fermions, e.g., electrons and holes in a semimetal or semiconductor. They have the Fermi
distributions neC~k) and nh(~k)' respectively. The scattering theory contains the vertex term for
the scattering (ke' k h ) -+ (k e + q, kh - q).
(lO.29)
This vertex would also cause Veff to be resonant if one had kh = - ke' but this cannot happen
in a solid in equilibrium, since otherwise the electrons and holes would recombine. However,
one can still get instabilities out of this process, as was originally discussed by Keldysh and
Kopaev (1965). The instability leads to a state called the excitonic insulator, which has been
reviewed by Halperin and Rice (1968).
A logarithmic singularity is obtained in (10.29) even if one of the occupation factors is
zero, say nh = 0, and only one particle obeys many-particle statistics. One example is the edge
singularities in X-ray spectra first predicted by Mahan (1967) and discussed in Sec. 9.3.2. In
electron-hole scattering, the electron occupation factors remain in the scattering function
even if nh = 0
(10.30)
This behavior is in sharp contrast to the multiple scattering theory for an electron scattering
from a fixed potential such as impurity scattering. Then there is only one fermion, sayan
electron, which has a factor (1 - ne) for scattering k -+ k' and the factor (-ne) for the rate of
back scattering k' -+ k. The back scattering enters with a sign change because of the anti-
symmetry of the single fermion state. The two rates are added, 1 - ne + ne = 1, and the
occupation number terms ne cancel out. The electron scattering from an impurity is treated as
a one-body problems (except for self-consistent screening) even in a many-particle system.
This cancellation explains why such factors as nF(~) do not appear in the vertex equations for
impurity scattering, which were solved in Sec. 8.1.
Another possibility is to have one or both of the particles be bosons. Then the scattering
probability is multiplied by [1 + nB(~I)l for scattering into state kl and by nB(~I) for scat-
tering out of it. For a single boson scattering from a fixed potential, the scattering integral has
the rate (1 + nB) for k ---+ k J and the rate nB back, so the net is 1 + nB - nB = 1, which does
not contain occupation factors. The potential scattering is again a one-body problem even in a
many-particle system. However, two bosons which are mutually scattering have the factors
(1 + nB)(1 + nB) - nBnB= 1 + nB + nB, so that the scattering is enhanced because of the
occupation of other particles.
10.1.1. BeS Theory

The basic feature of the BCS theory is that pairing occurs between electrons in states
with opposite momentum and opposite spins, e.g., between states (k, t) and (-k, ~). The two
spins are combined into a spin singlet, with 8 = o. The singlet was chosen in BCS theory on
the basis that the other choices of spin combination would lead to a triplet state with 8 = 1.
The latter choice (8 = 1) implies the superconducting state has magnetic properties, which in
fact are absent for simple metals. The choice 8 = 0 seems most reasonable. Later work by
Balian and Werthamer (1963), who solved the BCS equations for 8 = 1, showed that the
triplet state had smaller binding energy and was therefore less favored. However, the recent
theories of superfluidity in 3He are based on the premise that the pairing occurs in the triplet
state (see Chapter 11). Triplet pairing is possible and may exist in heavy fermion solids such
as UPt3 and UBe13 (Han et al., 1986). The discussion will assume that the spin arrangement is
singlet.
The pairing of electrons in the BCS theory must cause correlations in their relative
motions. The pairing is described by introducing a new correlation function, similar to a
Green's function, for particles of opposite spin. Following Abrikosov et al. (1963), these are
~(p, 1:' - 1:") = -(Tt Cpcr (1:')C:cr (1:"))

$'(p,1:' - 1:") = (Tt C_ p,!,(1:')Cpt (1:")) (10.31)
$'t(p, 1:' - 1:") = (TtC:t(1:')C~p,!,(1:"))
The Green's function ~ has the same definition as usual, although it has a different algebraic
form in the superconducting state. The $I' and $'t functions are identically zero in the normal
state. Even in the superconducting state, the bra (I and ket I) notation must have special
meaning. For instance, in the definition of the $I' function, the ket I) is operated on by two
destruction operators and then closed by (I. The state I) must have two more electrons than (I.
A basic feature of the BCS ground state wave function is that it is composed of a super-
position of electronic states containing a different number of electrons. Such a formulation is
possible in a grand canonical ensemble. The procedure will be to find a self-consistent
equation for the correlation function $I' or its Hermitian conjugate $'t. At high temperatures
the only solution is $I' = $'t = 0, but a nonzero solution becomes possible at low
temperature.
The order of the spin indices (t, ~) is important in specifying the Green's functions.
Suppose by mistake for $I' one wrote
(10.32)
Since $I' does not depend on the sign of p, this correlation function is the same as
(10.33)
The Cpcr operators anti commute, and doing this gives
(10.34)
-&i""(p, "C' - "C). These kinds of sign errors can be avoided by always following the definition
closely.
The objective is to provide the simplest possible derivation of the BCS theory. First
assume a model Hamiltonian of the form
(10.35)
The interaction potential V(q) between electrons is taken to have the form in (10.2), which is
an attractive constant V(q) = - Vo over a range of energies within a Debye energy of the
Fermi surface. With this Hamiltonian, a set of self-consistent equations will be derived for the
Green's functions '§, &i"", and &i""t. The derivation is done using the equation of motion.
Consider
(10.36)
From the definition of the "C-ordered product, the first derivative of the equation for the
Green's function is:
a '§(p, "C -
m "C') =
a
- m[8("C - "C')(Cpcr("C)cJcr("C')) - 8("C' - "C)(CJcr("C')Cpcr("C))] (10.37)
= -0("C - "C') - (T, [! Cpcr("C)]CJcr("C 1

)) (10.38)
Using the result (10.36) for dCpcr/d"C gives
( -~
m - ~p)'§(P' "C - "C' ) + ~ L V(q)
v qp's
(10.39)
(10.40)
The bracket of four operators in the interaction term must be reduced to products of pair
operators. There are many ways of doing the pairing, since in addition to the normal
combinations such as (Cct)( cct), there are also arrangements such as (CC) (ct ct). One
simplification is to assume that long-wavelength phonons give a zero potential, so that
V(q = 0) = O. Neglect the pairing which occurs when q = O. For a normal metal, there
would only remain the pairing 0p=p,os'=crnp_q'§(p, "C - "C' ) which gives the exchange energy.
This pairing occurs in the superconductor as well but is not the only term. The pairings which
include the &i"" functions must pay attention to the spin variables. The combination
cr = _S' = t gives
(T, C!, -q.s' ("C )Cp's'("C )Cp_q,cr("C ) CJcr ("C')) (10.41)
= -ocr=_s,oP'=_p+q&i""(p - q, O)&i""t (p, "C' - "C) (10.42)

where the sign change resulted from an odd number of operator rearrangements. Similarly,
the choice cr = -s' = -!- gives
(T~C!, -q,s' ('t )Cp's'('t)Cp_q,cr('t)C~cr('t')} (10.43)
= -ocr=_s,oP'=_p+q~(-p + q, O)~t(-P. 't - 't') (10.44)
These two results are identical, since later it is shown that the ~ and ~t functions do not
depend on the sign of their arguments either momentum or 'to The last term in (10.40) gives
the expression
1
= - L V(q)[~(p. 't - 't')n p_q - ~(p - q. O)~t (P. 't - 't')] (10.45)
Vq
The first term is the exchange self-energy of the electron due to the phonon induced inter-
action between electrons. It provides the self-energy term I:x = -I:V(q)np _ q ' A careful
investigation shows that this self-energy does not change much between the normal and
superconducting states. The self-energy of the electrons, from phonons, causes a change in
the electron effective mass given by the parameter A.. This effect is not large in weak
superconductors, so it may be ignored. In metals where A. is large, such as lead, the super-
conducting state can be expected to significantly alter the properties of electrons near the
Fermi surface. Hence the need, in these cases, for strong coupling theory. This self-energy
term is neglected in our weak coupling theory.
In the second term of(10.45) there arises the combination offactors which are defined as
i\(p) = --V1 Lq V(q)~(p - q. 't = 0) (10.46)
The quantity i\(p) is the gap function in the BCS theory and plays a central role in the
properties of the superconductor state. The quantity i\(p) is defined to be positive, since the
right-hand side of the definition is positive, with an attractive potential V(q) < O. Collecting
these results in (10.40) gives
(-! -~p)~(P. 't - 't') + i\(p)~t(P. 't - 't') = o('t - 't') (10.47)
The equation has two unknowns in ~ and ~t, so another equation is needed to link these two
quantities. It comes from the equation of motion for the ~t function:
! ~t(P. 't - 't') = - ! [S('t - 't')(C!t('t)C~p./. ('t')} (10.48)
- S('t' - 't)(C~p./.('t')C!t('t)}]
= -o('t - 't')({C!t. C~p./.}} + (T~[! C!t('t)]C~p./.('t'))

= (T~[! C!t('t)]C~p/'t')) (10.49)
It lacks a term o( t - t') because the ct operators anti commute. The time development of the
fft operator is determined by
This equation is not the Hermitian conjugate of dCpcrC'r;)ldt in (10.36), because C~cr(t) is not
the Hermitian conjugate of Cpcr(t). The result for dCt Idt brings us to the equation for
aff t 1m:
( -~+
m
~p)fft(P' t - t') +~v qp's
L V(q) (10.50)
X (T,C;_q,t(t)C;'+q,s,(t)Cp's,(t)C!p,j,(t')) = 0 (10.51)
Again there is an expression for four operators. The operator C!p,j, (t') is unique, since it is the
only one which is not operating at time t. Pairing it with each of the other three operators
gives three terms:
Cp's,(t): oS'=,j,dp'=_pAt(p)<;9'(-p, t - t')
C;_qt(t): oq=onp,fft(p, t - t') (10.52)
C;'+q,s,(t): - os'=tOp'=p_qnp_qfft(p, t - t')
The first term has At (p), which is the important term; the second requires q = 0, which was
assumed is zero; and the last gives the exchange potential Lx = - L V(q)n p+q ' which is
ignored again. The exchange potential just renormalizes the chemical potential. This term is
omitted. These steps brings us to the final equation:
( - :t + ~p )fft(P' t - t') + At(p)<;9'(p, t - t') = 0 (10.53)
The gap function is assumed real, At (p) = A(p), which is verified later.
The Fourier transforms of the correlation functions are defined in the usual way:
(10.54)
fft (p, t) = ~ L e-iPn' fft(p, ipn)

I-' Pn
After transforming, the two equations (10.47) and (10.53) are
+ A(p)fft(p, iPn) = 1
(iPn - ~p)<;9'(p, iPn) (10.55)
(ipn + ~p)fft(p, iPn) + A(p)<;9'(p, iPn) = 0 (10.56)
Sec. , O. , • Cooper Instability 639
These algebraic equations are easily solved:
'§(p i ) _ _ ipn + ~p (10.57)

,'Pn - p~+~;+A2(p)
az:(p, lPn
.:,- =t(p, lPn
. ) =.T . )= A(p)
2 2 (10.58)
p~+~p+A (p)
The equivalence of :y; and :y;t can be shown by deriving similar equations for :y;.
The results (10.58) can be used to test some of the assumptions in the derivation. For
example, consider the Fourier transform (10.45) of :Y;(p, -t). Changing the sign of the
dummy variable of summation gives
(10.59)
Similarly,
1 1
:Y;(p, t = 0) = ~ ~ :Y;(p, iPn) = ~ ~:y;t (p, ipn) = :y;t (p, t = 0)
This relation causes the energy gap equation to be real, and for At (p) = A(p)
A(p) = -~ EV(q):Y;(p - q, 0) = -~ E V(q):y;t(p - q, 0)

v q v q
= At(p) (10.60)
Note that if the energy gap A(p) is set equal to zero, there is found the usual form for the
normal state Green's function and zero for the others:
A = 0: {'§(P' ip) = ip ~ ~p (10.61)

:y; =:y;t = 0
It must still be shown that there is a self-consistent solution to the equations for A i= O.
The Green's functions '§,:Y; and:y;t have poles at the points ±Ep = ±i~; + A(pi. Ep
is called the excitation energy of the superconductor. It will occur repeated y m the various
formulas. In BCS theory the gap function A(p) for s-wave superconductors is not dependent
on momentum p and is treated as a constant for each temperature. Of course, it very much
depends on temperature and vanishes at the transition temperature of the superconductor.
However, A is treated as a constant at a fixed temperature. [Later, in strong coupling theory,
A(iron ) is found to be a function of energy.] The excitation energy Ep depends on~; > O. The
minimum excitation energy is A, and Ep ::: A. An important feature of the BCS theory is that
the particles are paired, and it is not possible to excite just one quasiparticle with excitation
energy Ep. Instead, one must break a pair of particles and excite them both to the band of
excitations. This pair breaking is shown in Fig. 10.4, where the horizontal line represents the
bound state pairs at the chemical potential of the superconductor. To break a pair, one must
excite both particles to the excitation line, so that it takes energy Ep + Ep' ::: 2A. The mini-
mum excitation energy of the superconductor is the energy to break a pair, which is Eg == 2A,
where Eg is the energy gap of the superconductor. This fact must be kept in mind when
comparing thermodynamic data with the BCS theory.
640 Chap.10 • Superconductivity
FIGURE 10.4 BeS model of superconductor. (a) The ground state has electrons paired at the chemical potential. (b)
Excited states occur by breaking a pair and forming two quasipartic1es, each with excitation energy Ep.
The existence of condensed pairs also clarifies the meaning of the bra (I and ket I)
symbols in the definition of
(10.62)
At zero temperature, the ket I) is the ground state of the superconductor, which has all bound
pairs at the chemical potential Il. The operators C!t and C~p,l. create two excitations with
energy Ep + E_p = 2Ep. The bra (I state has the same number of particles but two less in the
ground state.
The Feynman diagrams for the three Green's functions are shown in Fig. 10.5. A
diagrammatic representation is useful when evaluating correlation functions. The way of
doing the diagrams was suggested by Abrikosov et al. (1963). The Green's function f§
represents creating a particle at one point in time and destroying it at a later point in time. It is
drawn as an arrow with points at both ends and both points are in the same direction. The
inward point at one end symbolizes particle destruction, and the outward point on the other is
particle creation. $'t is represented by an arrow with both points outward, since a particle is
created at both times. Similarly, the Green's function $' is an arrow with both points inward,
which represents particle destruction at both points in time.
It is conventional to introduce the following coherence factors:
(10.63)
q ,.~--->~
<::---___
qt .... >~
q .,>~--~<
FIGURE 10.5
which are misnamed since they have little to do with coherence. Their usefulness stems from
the fact that they are the residues of the poles of the Green's functions. Write the Green's
functions (10.58) as
u2 V2
~(p, ip) = ~E +. +P E (10.64)
Ip - P Ip p
G;(P,iP)=g;t(P,iP)=-UpVp(.
Ip
~E
p
-. +IE)
Ip P
(10.65)
This form is useful for doing the contour integrals associated with the summations over
Matsubara frequencies. The excitations of the superconductor are fermions, and the
frequencies ipn in ~ and G; are for fermions ipn = (2n + l)ni/~. The spectral function for the
Green's function is
(10.66)
These equations are not really solved until.1 is evaluated. It is found from the definition of the
gap function in (10.46), where
(10.67)
The summation over frequencies is evaluated in the usual fashion, by the contour integral
f dz .1
-nF(z)--=O
2ni z2 - E2P
(10.68)
The integral is zero since the contour is taken to infinity. The poles of nF(z) give the
summation over iPn while the poles at z = ±Ep give the hyperbolic tangent result. The
equation for the gap function is
.1(p) =- J(2n)
d 3q3 V(q) .1(p - q) tanh(~Ep_q)
2E p_ q 2
(10.69)
Since .1 is constant, it can be factored from both sides, which leaves the integral equation
for .1
1 = NFVO JroD d~tanh(~E) (10.70)

2 -roD E 2
E = J~2 +.12 (10.71)
The solution is found in two limiting cases. The first is at zero temperature, where the
hyperbolic function tanhWE /2) = 1
1 _ NFVO JroD d~
(10.72)
- 2 -roD J~2 +.12
(10.73)
which may be solved to produce the equation for the energy gap
Eg = 2A = 4ffiD exp ( _ _1_) (10.74)

NFVO
In BCS theory the gap equation (10.70) was solved as a function of temperature. The energy
gap gets smaller as the temperature is increased, as shown in Fig. 10.6. In BCS theory the
critical temperature was the result stated earlier in (10.15):
kBTc = 1.14ffiD exp ( - N:VJ (10.75)
The ratio of these two results predicts

Eg 4.0
kBTc = 1.14 = 3.52 (10.76)
Both the energy gap Eg and the transition temperature Tc can be measured, so this can be
tested. It is found to work well in weak superconductors such as aluminum, which is generally
well described by the BCS theory. For strong coupling superconductors, the ratio increases in
value.
The BCS theory is often called a mean field theory, because the gap equation (10.69) has
that form. It has a similar form to the mean field theory for magnetism in metals (see Kittel,
1966; he uses the term molecular field approximation). The order parameter, in this case the
gap A, is set equal to a thermodynamic average over the excitations, which are also functions
of A. The self-consistent solution determines A. There is a critical temperature Tc above
which no solution is possible. It is seldom that the mean field theory is an accurate description
0.8
••
o
~
..... o Tin 33.5 Mc
<i 04 • Tin 54 Mc
- BCS theory
o
T/~
FIGURE 10.6 The ratio t1(T)/t1(O) vs. T/Tc in tin. The solid curve is the BCS theory. The points are the ultrasonic
attenuation data of Morse and Bohm (1957) (used with permission).
of physical reality, but it certainly works well for BCS. In statistical mechanics, mean field
theory works well when the forces between particles are long range. In the case of super-
conductivity, the forces are short range, but the bound state orbits of the particles extend over
a long distance.
Besides the existence of an energy gap, another dramatic prediction of the BCS theory
was that the density of states peE) for excitations has a square root singularity. It is defined as
dE)-1 E
pee) = ( d~ = Jw _A2 (10.77)
This spectral shape is shown in Fig. 10.7, with a gap until A and then a square root singularity.
The density of states arises from the variable change
J
J~ d~ = dE(~!) = J~ dEp(E) (10.78)
This density of states should not be regarded as just a function which results from a change of
variables but as a real property of superconductors which can be measured.
The above theory is valid for a superconductor with an isotropic energy gap. Electron
states in crystals belong to representations of the group which leaves the crystal invariant.
Angular momentum eigenstates, such as S-, p-, etc., do not belong to representations of
crystal groups. Instead, the crystal representations are mixtures of angular momentum states.
The energy gap in crystals is found to vary somewhat with direction of the wave vector,
because of this mixture of angular momentum states. However, the phrase "s-wave super-
conductivity" is used to describe the state which belongs to the most isotropic representation.
In a cubic crystal, the gap would have the same value along all ±X, ±y, ±z axes. Similarly, the
phrase "p-wave superconductivity" applies to crystals in which order parameters might have
different signs in the ±z directions.
The cuprate superconductors may have "d-wave symmetry". The conduction electrons
are confined to move in the (x, y) planes composed of copper and oxygen atoms. In some
cases the x and y directions have the same lattice constant, and have identical symmetry. Then
the planes have the same crystal symmetry as a square. In polar coordinates the angle givese
ptE)
FIGURE 10.7 BeS density of states of a superconductor.

the wave vector direction in the plane (tan 9 = ky/kJ. In this case the superconducting order
parameter for d-wave symmetry is either of the two functions
L1d1 (9) = L11 cos(29) (10.79)

L1d2 (9) = L12 sin(29) (10.80)
Similarly, another possible order parameter called "anistropic s-wave" is
L1,(9) = L10 + L1a cos(49) (10.81)
Note that the term cos(49) preserves the symmetry of the crystal: at 9 = TC/2 the function has
the same value as at 9 = O. The x- and y-axes are identical for anisotropic s-wave super-
conductivity. In contrast, the d-wave superconductivity breaks crystal symmetry.
An important feature of d-wave superconductivity is that the order parameter changes
sign as the angle 9 is varied. There are nodes, where the gap function vanishes. The aniso-
tropic s-wave can also have nodes if L1a > L1 o, but that is unlikely.
10.2. SUPERCONDUCTING TUNNELING
The most important verifications of the BCS theory came from electron tunneling
experiments. They measured the energy gap, as a function of temperature, in perfect agree-
ment with the BCS theory. They also measured the density of states function peE). Later,
Josephson (1962) predicted the coherent tunneling of pairs, which was also quickly observed.
These experiments provided a detailed verification of the BCS theory.
The theory of electron tunneling between two metal electrodes, through a barrier
consisting of a metal oxide, was derived in Sec. 8.6. The tunneling Hamiltonian was used to
obtain the following expression for the tunneling current
I = Is +IJ (10.82)
Is = ie[Uret( -e V) - Uadv ( -e V)] (10.83)
= -2e Im[Uret(-eV)] (10.84)
U(im) = - 1: d'teiffinT(TTA(r)At(O)} (10.85)
A =L TkpCt,Cpcr (10.86)
kpcr
The term Is is the single-particle tunneling. The term IJ is for the Josephson effect, and is
evaluated below. In Sec. 8.6 this expression was evaluated, and reduced to an expression in
terms of spectral functions and occupation numbers
(10.87)
The subscripts L, R refer to which electrode. The tunneling matrix element Tkp can be
approximated as a constant for the small voltages used in superconductive tunneling. Then
Sec.10.2 • Superconducting Tunneling 645
the summations over wave vectors give the interacting density of states NL,R(E) on the two
sides of the junction.
This expression is also valid in a superconductor with an appropriate interpretation of the

density of states.
10.2.1. Normal Superconductor

Electron tunneling experiments are often done with different metals on the two sides of
the junction. Since different metals have different transition temperatures (and some are not
superconducting), it is quite easy to arrange that one metal be normal and the other be a
superconductor. This experiment is interesting, to the theorist, because it provides a direct
measurement of the BCS density of states p(E). Take the right side to be superconducting and
the left side to be normal. The derivation is simple, since the many-body expression (10.87) is
valid in this case. Just use the respective spectral functions for the normal and super-
conducting sides of the tunneling junction:
AR(k, E) = 2n[U~O(E - Ek ) + V~O(E + Ek )] (10.88)

AL(p, E + eV) = 2nO(E + eV - ~p) (10.89)
Is = 4nelTI2 ~)u~o(eV + Ek - ~p)[nF(Ek) - nF(~p)]
kp
+ v~o(eV - Ek - ~p)[nF( -Ek) - nF(~p)]} (10.90)
Next use the fact that nF( -E) = 1 - nF(E). Consider the situation at zero temperature, where
the density of excitations in the superconductor is negligible [nF(E) ~ (ef3E + 1)-1 ~ e- f3E
~ 0]. The variable change (8.337) brings us to
Is = 4nelTI 2NRNL J~oo d~p J~oo d~k{v~o(eV - Ek - ~p)[l - nF(~p)]

- u~o(eV + Ek - ~p)nF(~p)} (10.91)
The limits of integration have been extended to infinity, which has no effect on the result since
the limits are narrowly confined. There are two terms in the curly braces: the first is positive,
and the second is negative. The first will contribute only when eV > 0 (actually eV > L\1 and
Is is positive in this case because this direction of electron current is defined to be positive.
The second term contributes only when eV < -L\, and Is is negative in this case, in agree-
ment with the definition, since Is has the same sign as V.
At zero temperature, the first term has a factor 1 - nF(~p) which limits ~p > O. The
integral d~k extends over positive and negative values of ~k' Factors which depend on
Ek = lE,,~ + 112 are the same for both signs of Sk' The tenn linear in Sk in the coherence
factor vi in (10.63) averages to zero because of the cancellation of the +Sk and -Sk parts:
J OO
-00
dskvi I(Ek) = -21 Joo dS k ( 1 - ~
-00 Ek
)r
S (Ek) = -1 Joo dS k I(Ek)
2 -00
= J~ dskl(Ek) = J~ dEkP(Ek)f(Ek) (10.92)
where I is any function of Ek . The integration is changed to dEk> which produces a factor of
the density of states p(E) = dS/ dE in the superconductor. The integral f dSp can be used to
eliminate the delta function for energy conservation Sp = eV - Ek > O. Since Sp > 0 then
eV > Ek ~ 11, which gives for eV > 0
(10.93)
The current is a simple square root function of the voltage. This type of dependence is
observed experimentally.
Another quantity which can be measured (by lock-in amplifier techniques) is the
dynamical conductance dI/ dV as a function of voltage:
( -dIs) =eV(JOr=====
0(eV -11)
(10.94)
dV SN /(evi
_112
G~)
---7-::c::-:--"S.:..:...N = 0(eV -11)
eV
= p(eV) (10.95)
( dIS) /(eVi _112
dV NN
The experimental technique is to compare dI/ dV for the tunneling between the nonnal metal
and the superconductor (dI/dV)SN' with that for the tunneling between both metals when they
are nonnal (dI/dV)NN = (Jo. Low-temperature superconductors can be made nonnal by the
application of a small magnetic field, so that the two measurements are done at the same low
temperature. The ratio of these two experimental quantities provides a direct measurement of
the BCS density of states p(eV). The current is antisymmetric in voltage, which can be shown
from the second tenn in (10.91).
Figure 10.8 shows the experimental results of Giaever et al. (1962) for dI/ dVof a tunnel
junction between Pb and Mg. The experiments were done at a temperature of 0.3 K with a 3He
refrigerator. At this low temperature, the Mg is nonnal, and the Pb is well below its transition
temperature of Tc = 7.2 K. The factor nF(E) '"" exp( -~I1) is indeed small, and there should
be few thennal excitations in the superconducting Pb. The figure shows the derivative
(dI/dV), which well illustrates the BCS density of states. Another feature of their data are
bumps in dI/ dV at higher voltages. These are due to phonons and are not predicted by the
BCS theory. They are explained by the strong coupling theory. These experimental results by
Giaever et al. (1962) were one of the first experimental indications of the need for strong
coupling theory.
Sec. 10.2 • Superconducting Tunneling 647
Pb/MgO/Mg
E=1.34xIO· 3 eV
3 T=.33·K
o 4E 8E 12E 16E
ENERGY UN UNITS OF E)
FIGURE 10.8 Electron tunneling between a nonnal metal (Mg) and a superconductor (Pb). The relative conductance
of a Ph-MgO--Mg sandwich plotted against energy. At higher energies there are definite divergences from the BCS
density of states, as can be seen from the bumps in the experimental curve. Source: Giaever et al. (1962) (used with
permission).
The measured current is an antisymmetric function of the voltage J( - V) = -J(V) for

the small voltages of interest in this experiment. An energy gap of A = 1.34 meV is observed
in the tunneling current at zero temperature (which can only be approached asymptotically),
and no current flows unless leVI> A. An interpretation of these results is provided in Fig.
10.9. Figure 10.9(a) shows the two sides of the tunnel junction with no applied voltage
V = O. The chemical potentials are the same on the two sides of the junction. On the normal
side (left) the chemical potential divides the occupied states from the empty states in the
Fermi sea of electrons. On the superconductor side (right), the chemical potential is the
energy of the paired electrons. The excitation spectrum is shown as the hyperbolic-shaped line
above the chemical potential. No net current flows when V = O.
The situation for a forward voltage eV > A is shown in Fig. 10.9(b). The electrons from
the normal side of the junction can tunnel through and become a quasiparticle excitation of
energy Ek on the superconducting side. The arrow, which shows the tunneling path, is drawn
horizontal to indicate the energy-conserving aspect of tunneling. The electron cannot go into
the pair state on the right as a single step in the tunneling, since it needs two electrons for this.
It must become an excited quasiparticle during its tunneling step. Later it will find some other
excited electron and join with it to become a pair state. This pairing is expected to happen
well after the tunneling process and so will not affect the tunneling rate.
The situation for reverse bias is shown in Fig. 10.9(c). At zero temperature, all the
electrons in the superconductor are in the pair states at the chemical potential, and none are
thermally excited to the excitation state. The single-particle tunneling process must break up
the pair state. For each electron which tunnels, its partner in the initial pair state must become
(b)
------fLR
FIGURE 10.9 Electron tunneling between a nonnal metal and a superconductor at T = O. (a) No applied voltage. (b)
For eV > A, electrons can tunnel from the nonnal metal and becomes a quasiparticle in the superconductor. (c) For
eV < -A, a pair is broken in the superconductor: one electron tunnels, while the other becomes a quasiparticle in the
superconductor.
an excited quasiparticle. The electron which tunnels does not do this at constant energy, since
it must give up some of its energy to its partner to permit it to overcome the excitation gap A.
Energy conservation can be understood as follows. The initial energy of the paired electrons
is Einitial = 21lR' The final energy is IlR + Ek for the one in the superconductor and ilL + ~p for
the one in the normal metal. The total energy of the original pair in the final state is
(10.96)
The entire process conserves energy, so this is set equal to the initial energy. The quasiparticle
energy in the normal metal ~p must be positive since the electron must go into an unoccupied
state. The condition for the tunneling current to flow is:
IlR -ilL = eV = ~p +Ek > A (10.97)
This condition is that the magnitude of the voltage must be larger than A, which is the same
condition for the other direction of bias voltage. The actual physical processes for the two
directions of voltage are dissimilar, since in one direction there is direct pair breaking.
10.2.2. Two Superconductors

The electron tunneling between two superconductors has two kinds of currents. One is
the tunneling of a single electron, which is similar to that between normal metals or between
normal metals and superconductors. The other is the tunneling of pairs, which is the
Josephson effect. This interesting subject will be described in the next section. Now consider
only the single-particle tunneling, which can happen between two similar or two dissimilar
superconductors. Take the case where they are different, so there are gap functions AR and AL
for the two sides of the junction. At any point in the calculation the two superconductors can
be made identical, theoretically, by just setting AR = AL .
The many-body fonnula (8.344) is still valid when the two sides are superconductors
(10.98)
The spectral function for the superconducting BCS Green's function is used for both sides of
the tunneling junction:
A(k, E) = 21t[U~O(E - E k) + viO(E + E k)] (10.99)

A(p, E+ eV) = 21t[U~O(E + eV - Ep) + V;0(E + eV + Ep)] (10.100)
Their product gives four tenns, which are regrouped using the identity that nF( -E) = 1
-nF(E):
1= 41te L ITkl{u~u~o(eV + Ek - Ep)[nF(Ek) - nF(Ep)]

kP
+ u~v~o(eV + Ek + Ep)[nF(Ek) - nF( -Ep)]
+ v~z4o(eV - Ek - Ep)[nF( -Ek) - nF(Ep )]
+ v~v~o(eV - Ek + Ep)[nF( -Ek) - nF( -Ep)]} (10.101)

1= 41te L ITkp12{[1 - nF(Ep) - nF(Ek)][v~u~o(eV - Ep - E k)
kp
- uiv~o(eV + Ek + Ep)] + [nF(Ek) - nF(Ep )]
x [u~u~o(eV + Ek - Ep) - viv~o(eV - Ek + Ep)]} (10.102)
This fonnula is the most general result, which is valid at nonzero temperatures. The variable
change to d~kd~p is done next. These integrals over the coherence factors u2 and v2 will
eliminate their linear tenn in~, as described in (10.92). Then the variables can be changed to
E = Ek = J~~ + L\~ and E' = Ep = J~; + .11, which brings us to the expression
oo
1= 41teNLN RITI 2Joo dEPR(E)J dE'PL(E')
AR AL
x {[1- nF(E) - nF(E')][o(eV -E -E') - o(eV +E +E')]

+ [nF(E) - nF(E')][o(eV +E -E') - o(eV - E +E')]} (10.103)
Some of the tenns are positive, while others are negative. They will ensure that I is positive
for V > 0 and negative for V < O. These integrals must be evaluated numerically at nonzero
temperatures. The physics is illustrated by taking the simplest case, which is zero temperature
and identical superconductors L\L = L\R' At zero temperature, set to zero all the thennal
factors nF(E) and nF(E'):
1= 0"0
e
JooA dEP(E)JAoo dE' p(E')[o(eV - E - E') - o(eV + E + E')]
(10.104)
0"0 = 41te2NRNLITI2
The two terms in square brackets give identical results, where the first is for e V > 0 and the
second for e V < 0, in such a way that! ( - V) = - / (V). Only the first term needs to be done,
where E' = eV - E and eV > 2Ll:
ev-t.
/ = (:O)e(eV - 2Ll) ft. dEPR(E)p(eV - E)
= (CJO)e(eV _ 2Ll) fev-t. dE E(eV - E) (10.105)

e t. JE2 _ Ll 2 J(eV _ E)2 _ Ll 2
This integral is done in the following way. Change variables of integration to x, where
2E = eV + x(eV - 2Ll). After some algebra, the integral is in the form of complete elliptic
integrals:
eV - 2Ll
rx=--- (10.106)
eV +2Ll
/ = (CJO) e(eV - 2Ll) JI d)eVi - x2(eV - 2Lli
e eV + 2Ll -I [(1 - x2)(1 - rx 2x2)]1/2
= (:O)e(eV - 2Ll)1 e~:)~Ll K(rx) - (eV + 2Ll)[K(rx) - E(rx)] I (10.107)
Equation (10.107) is the final result, which is compared with the experiments. One check on
the correctness of the answer is to set the energy gap Ll = O. The coefficient of K(rx) vanishes,
so the part in curly braces becomes eVE(1) = eV. The current is / = CJoV, which is the
correct result, since in the absence of superconductivity the tunneling device is a resistor at
low voltage with the conductance CJo.
The tunneling current is zero for voltages less than twice the gap function 2Ll. It is
interesting to examine the threshold phenomenon, for voltages slightly larger than 2Ll. Since
K(O) = E(O) = rc/2, at eV = 2Ll + the current is
/(2Ll +) = (CJO) rcLl = CJo V ~ (10.108)

e 2 4
It has a nonzero value, so there is a discontinuous jump in the tunneling current at e V = 2Ll.
The current above threshold is rc/4 of the value of a normal tunneling current at the
same voltage. Exactly this behavior was found by Giaever et al. (1962). Their data for a
Sn-SnOx-Sn junction is shown in Fig. 10.10. The entire experimental curve fits the theo-
retical curve from (10.107), except for the finite slope on the "discontinuity." They also point
out an interesting feature of this discontinuous jump in current: the discontinuity does not get
thermally broadened with increasing temperature. As the temperature is increased, the energy
gaps get smaller, so the value 2Ll of the jump moves to small voltages. However, the BCS
energy gap does not get smeared at higher temperature; it only gets smaller. At higher
temperatures, the nonzero value of nF(E) will lead to some tunneling current for eV < 2Ll,
but there is still a discontinuous jump at this critical voltage.
The physical process of the single-particle tunneling is shown in Fig. 10.11. For electron
flow to the right, a pair on the left is broken up, and one electron tunnels, while the other
becomes a quasiparticle on the left. The initial energy of the pair is E; = 2/-lv and the final
energy of the pair is Ef = (/-lL + EL) + (/-lR + ER). By equating these two energies, we derive
Ul
t- 10
Z
:;) Sn/Sn0 2/S n
o 8 T=0.300K
I.IJ
N E=0.60'10-~v
:::i 6
<t
:z
a:
o 4
z
3
t-
Z
I.IJ
a:
a:
:;)
u
2
THEORETICAL
/ ' CONDUCTANCE
I T=O
: ... EXPERIMENTAl
CONDUCTANCE
IE 2E 3E
VOLTAGE(IN UNITS OF E)
FIGURE 10.10 Electron tunneling between two identical superconductors. Current voltage characteristic for a
sandwich compared with the BCS theory. The experimental curve has a finite slope at 2A because of the anisotropic
nature of the energy gap. Source: Giaever et al. (1962) (used with permission).
~L - ~R = eV = EL + E R • The summation of the two excitation energies must be equal to the

applied voltage, which is in agreement with the theoretical formulas.
The reader should now be convinced that the BCS theory rather well describes the theory
of electron tunneling in low temperature superconductors. Indeed, the BCS theory describes
rather well all the properties of weakly coupled superconductors, and other good agreements
between theory and experiment will be shown later. Several severe approximations were made
in the derivation of the BCS theory. Apparently none of these really matter. Two of the major
approximations were (1) calling the gap function fl.(p) a constant fl. and (2) replacing the
retarded (Le., frequency-dependent) interaction between electrons by a static potential, which
is a constant. A constant potential in Fourier space is a delta function in real space, so it was
assumed that the interaction between electrons is localized in space. The electrons themselves
- - - - - - fLR
FIGURE 10.11 Physical process of electron tunneling between two superconductors. A pair breaks on the left, one
electron tunnels, and both electrons become excited quasiparticies.
are not localized. Indeed, the pairing orbits are extended in space, but the particles interact
only when they are in contact. In any case, the essential feature ofBCS theory is the existence
of the additional correlation functions g; and g;t. This feature must be right, and the
particular method of getting to these correlation functions is not very crucial.
Electron tunneling is a less successful experimental tool for the investigation of high
temperature superconductivity. Although many experiments have been done, the results are
inconsistent and difficult to interpret. There are several problems. The cuprates are layered
materials. The electron conduction is within a layer, and the layers are weakly coupled. That
means the top few layers of a material may not even be superconducting. Tunneling down to
the top layer may be going into a normal layer, even if the bulk is superconducting. The other
geometry is to tunnel into the layers from the edge. In this case, the electron is injected into a
single layer, which is a two-dimensional electron gas. The last chapter discussed MND
effects, which is the renormalization of a transition intensity due to the process of injecting an
electron into an electron gas. Such effects are accentuated in two dimensions. So the spectral
function measured in tunneling may be severely distorted due to electron-electron interac-
tions (Shekhtrnan and Glazman, 1995). Finally, if the gap has d-wave symmetry there are
nodes where the gap vanishes. Depending upon the directional dependence of the tunneling,
there may be no energy gap to observe.
10.2.3. Josephson Tunneling

The derivation of the tunneling current led to two terms, which are Is and lJ. They are
the single-particle term and the Josephson term. The single-particle tunneling has been
discussed in great detail, and now examine the Josephson effect. The discussion of this
current contribution closely follows Ambegaokar and Baratoff (1962, 1963). The Josephson
current was derived in Sec. 8.6
(10.109)
The single most interesting feature of this equation is that it contains, in the two exponents,
the sum of the two times t + t'. This dependence has not been previously encountered in any
of the correlation functions, since they usually come out containing t - t'. The factor of t + t'
is the first indication, mathematically, that the Josephson effect is rather special. The problem
of t + t' is solved by writing it as 2t + t' - t and pulling the factor 2t outside of the correlation
function. The current is written as
lAt) = e{ e-2iteV/h f~oo dt'E>(t - t')eieV(t-t') ([A(t), A(t')]) + h.c.}
Define the retarded correlation function
f:
<Pret(t - t') = -iE>(t - t')([A(t), A(t')]) (10.110)
<Pret(eV) = dteietV(t-t')<Pret(t - t') (10.111)
lAt) = -ie[e-2ietv/h<PretCeV) - e2ietV/h<Padv(eV)] (10.112)

The part in brackets is purely imaginary since it is a quantity minus its complex conjugate.
The Josephson current is
(10.113)
It is important to include the factor exp( - 2ite V I Ii) in the expression whose imaginary part is
evaluated. The Josephson tunneling is not just the imaginary part of a retarded function-it is
not just a spectral function. Instead, one must multiply by the phase factor exp( -2iteVIii)
before taking the imaginary part. The phase factor contains most of the interesting physics.
For a nonzero voltage, the current is actually oscillatory in time, with a frequency ffi = 2eVIii.
This prediction of the theory is so absolute, and simple, that the frequency can be used as a
method of measuring the ratio of fundamental constants elh (see Parker et al., 1969).
The first task is to develop a method of calculating the retarded correlation function
<l>ret(eV). As usual, start from a Matsubara function,
<l>(iffi) = - J:d'teiOln~(T~A('t)A(O»
= - L.: L.: TkP TJ{p,jf3 d'teiOln~(T~CtJ('t)Cpcr('t)CLJ'Cp'cr') (10.114)
kpcr k' p' cr' 0
and later make the analytical continuation iffi -+ eV + iO. The spin index has been explicitly
added to this expression. The correlation function can be factored into a term for the left and
right sides of the tunnel junction. In this case the factoring produces the Green's function !F
and !Ft:
<l>(iffi) = 2 L.: TkpL k._Pjf3d'te iOOn ~ (T~C~t('t)C!k-l- (0» (T~C_p-l-(O)Cpt('t»

kp 0
= 2 L.:TkPLk._pJf3 d'teiOln~!Ft(k, 't)!F(p, -'t) (10.115)

kp 0
1
= 2 L.:TkpLk.-PA L.:!Ft(k, ip)!F(p, ip - iffi) (10.116)
kp .... ip
The factor of 2 is for spin arrangement. One spin configuration is shown explicitly, and
the other one gives the same result. The Feynman diagram for this tunneling is shown in Fig.
10.12. There is an!F t on the right side of the junction and!F on the left, where the arrow
conventions agree with those set forth in Fig. 10.5. The direction of the arrows is continuous
through the symbol T in a circle, which must always happen with any diagram where particle
number is conserved.
FIGURE 10.12 Feynman diagram for Josephson tunneling.
The summation over Matsubara frequencies in (10.116) is done by the usual method of
contour integral:
1
A L ~t(k, ip)~(p, ip - 0)) (10.117)
I-' lp
= ::'2 {[I - nF(EJ ) - nF(Ep)] (iO) +;p + Ek - iO) - L- Ep)
+ [nF(Ep) - nF(Ek)] (iO) +;p - Ek - iO) +;k - Ep) I

<l>(i0)) = I1LI1R L TkpLk,_p (10.118)
kp 2EkEp
x {[I - nF(EJ ) - nF(Ep)] (iO) + E~ + Ek) - iO) -;k - EJ

+ [nF(Ep) - nF(Ek)] (iO) +;p - Ek - iO) + L- Ep) I
This expression is rather complicated, particularly for nonzero temperatures and different
superconductors. The physics may be best understood by examining the simplest situation,
which is zero temperature [nF(E) = 0] and identical superconductors (I1L = I1R)' With the
analytical continuation iO) -+ e V + if>, consider the quantity
(10.119)
An important feature of these equations is that the tunneling matrix elements Tkp are not
necessarily real, so the combination TkpT_k,_p is not real. Write this factor as an amplitude
ITI2 and a phase factor exp(i<l» and consider both of these to be independent ofk and p. The
summations over k and p can be changed to integrations over dSkdSp and then to dEkdEp :
1 '<!J
<l>ret(eV) = -Js(eV)e' (10.120)
2e
J (eV) = 4eL\21T1 2N N foo dE peE) fOO dE' peE') ( 1 _ 1 ) (10.121)
s L R Ll E Ll E' e V + E + E' e V - E - E'
(10.122)
The integral is an elliptic integral. The retarded correlation function is written as a phase
factor times a function of voltage Js(eV), where Js is real for leVI < 2L\. The Josephson
current can be written as the imaginary part of (10.113), or
Is(t) = Js(eV) sin(rot + <1» (10.123)

2eV
ro=T (10.124)
The Josephson current oscillates with time, with a frequency ro = 2eV /Ii. The amplitude of
the oscillation is determined by the function of voltage Js(eV), where it is assumed
leVI < 2L\. The easiest way to evaluate Js(eV) at nonzero temperatures is to use a different
sequence of steps. Ambegaokar and Baratoff showed that it is best to return to (10.116) and to
do the two integrals dSkdSp before the summation over Matsubara frequency iPn- This step
uses the result
f OO dS k:#'
oz-t· foo dS k 1tL\
-00
(k, lp) = L\
-00
,2
Sk + L\
2
+ p~
= --;===
J
L\2 + p~
(10.125)
The function Js (e V) is the retarded part (iro --+ e V) of the function Js(iro):
1
Js(iro) = 4e2:=iTkp12 A L Ji't(k, ip)Ji'(p, ip - iro) (10.126)
kp t-' ip
(10.127)
The summation over Matsubara frequencies iPn should be done before the step of analytical
continuation iro --+ eV. An error is made if these two steps are reversed, except at the point
eV = O. At zero voltage, the result for identical superconductors is [see (10.67)]
222 21 1
Js(eV = 0) = 4e1t N L\ ITI AL 2 (10.128)
t-' iPn L\ + p~
(1.129)
At zero temperature there is the particularly simple result that
(10.130)
The coefficient of the Josephson current is the same current found in a nonnal junction at a
voltage of e V = rtA./2.
Josephson tunneling has a wide variety of behavior in junctions in static electric and
magnetic fields. A very simple characteristic current is obtained in the case where there is
neither a voltage V = 0 nor a magnetic field. Then one gets that the Josephson current is
(10.131)
where is the phase difference between the two superconductors on the two sides of the
tunnel junction.
The experimental arrangement is to put a current I through the tunnel junction, so that
current flows through the interface oxide. The voltage change V is measured as a function of
the applied current 1. It is found that a dc current can flow without any voltage V. This
behavior is sketched in Fig. to.13. The dc current is the coherent tunneling of electron pairs
from one superconductor to another. The pair tunneling costs no energy when the two
superconductors have the same chemical potential. When the experimentalist increases the
current, the phase difference between the two superconductors adjusts to make the
Josephson current Js(O) sin just the experimental value I of current. The phase increase
continues for increasing I until the point is reached where I = Js(O), or = rt/2. Then the
current jumps from the zero-voltage line to the right and on to the experimental curve for
single-particle tunneling, which is shown as the dashed line. In fact, in most tunnel junctions,
the maximum phase = rt/2 cannot be reached, and the jump to the single-particle tunneling
curve occurs at applied currents smaller than the maximum Js(O).
Another experiment is to take a tunnel junction between two superconductors and to
apply a static voltage e V < 2A.. Then one observes an oscillatory current according to
(10.124). This oscillatory current is the tunneling of pairs from the chemical potential of one
superconductor to the chemical potential of the other. This tunneling of pairs does not
conserve energy but takes an energy 2e V for the pair of charge 2e to overcome the voltage V.
The pair must tunnel back, since it cannot complete a transition where energy is not
conserved. This ac current can be observed by the external radiation it produces (Langenburg
et aI., 1965).
The preceding two experiments must be perfonned in zero static magnetic field. The
Josephson effect is so sensitive to magnetic field that care must be used to cancel out the
effects of the earth's magnetic field. This sensitivity to magnetic field, which was pointed out
by Josephson (1962), is useful for verifying that the tunneling is due to the Josephson effect
/
/
/
/
/
I /
/
/
/
----.,)1
I
I
I
2~ eV
FIG URE 10.13 Current-voltage characteristics of a Josephson junction. The dashed line rising abruptly at e V = 2A
is the single-particle tunneling curve in Fig. 10.10 at T = O. The arrows on the vertical axis show that a current can
flow with no voltage. But an imposition of too much current causes the device to jump to the single-particle curve.
FIGURE 10.14 Coordinate system used in discussing Josephson tunneling.
and not some other process such as a superconducting short in the oxide-the latter would be
expected to behave like the current characteristic in Fig. 10.13. The method of including
magnetic field is as follows. The total phase difference between the superconductors is written
as 11>. The phase is now meant to include the applied voltage V, so that we write it as
(10.132)
11> = 11>0 +h
2eJt
0dt'
Jba dx dV
dx (10.133)
The spatial coordinates are shown in Fig. 10.14. The integral f dx in (10.133) is taken across
the junction from one side to the other. The dx integral is actually V(b) - V(a) == V, since a
and b are on two different sides of the oxide and each is deep within the bulk of the
superconductor.
The phase 11> needs to be expressed in a manner which is gauge invariant. Write the
phase as an electric field E(r, t) which is always gauge invariant (see Sec. 1.5):
dV x 1 aA aq,
-~-E
dx
=--+-
x cat ax
(10.134)
11> = 11>0 + 2e
lie
Jdx Jdt' aAat'x + 2eIi Jdx'dt' aq,ax' (10.135)
The vector potential Ax term is manipulated in the following fashion. It is split into time-
dependent and time-independent terms. The time-independent term is obtained by doing the
time integral
(10.136)
Second, the space integral f dx' Ax is recognized as a line integral f dl· A along a particular
path, and any path will give the same result. Then the use of Stokes' theorem brings us to
Jdl· A = Jds . V x A = Jds . Bo (10.137)
11> = 11>0 + ~: J ds' Bo - ~e f J

dt' dl'· E(r', t') (10.138)
The identity V x A = Bo gives the term f ds . Bo, which is the magnetic flux under the area
of the integral. The external magnetic field Bo = BoY is constant when A is constant. The area
ds is the distance z from some reference point, which can be the end of the junction, times the
effective distance in the x direction. The latter is the thickness of the oxide d plus the distance
the magnetic field penetrates into the superconductor on each side of the junction. The latter
distances are the two penetration depths AR + AL. In a fixed magnetic field the phase is
 = <1>0 + kz + rot (10.139)

2e
k = lie Bo(d + AR + AL) (10.140)
The magnetic field generates an effective wave vector k. The Josephson current IJ now
depends on the distance z from the reference point. The total current in the sample of length L
in the z direction is found by integrating in this direction (eV = 0):
j(z) = Js(O) sin(o + kz) (10.141)
(10.142)
The theoretical prediction is that the maximum current at zero voltage has a Fraunhofer
pattern sin(8)/8, where 8 <X Bo. The Fraunhofer pattern was indeed observed by Fiske, as
shown in Fig. 10.15 for Sn-oxide-Sn junctions. This experiment is another spectacular
verification of Josephson's predictions. Even more interesting results are obtained when there
is both a static dc magnetic field Bo and dc voltage V. Then the Josephson current has the form
j(z) = Js(eV) sin(o + kz + rot) (10.143)
This form of current is also observed in a variety of experiments which are described in the
review volume edited by Parks (1969).
10
:~ii,.'·~eNI 10m-a
OXIDE
sn/sn
2.70·
1
_~ 5
H
2.70 3 I 0 3
H" (GAUSS)
FIGURE 10.15 The Fraunhofer pattern of a Josephson junction. The experimental curve is the maximum dc current
which can flow with no voltage as a function of applied magnetic field. Source: Fiske (1978, unpublished) (used with
permission).
Another way to evaluate the integral in (10.142) is to separate the negative values of z,
and then to change z' = -z in this integrand
IJ T(0) [Jo dzsin(o + kz) + JL°2dzsin(o + kz) ]

=J -L12
I
(10.144)
=T
J (0) [JL 2
° dzsin(o - kz) + JL°2dzsin(o + kz) ]
I I
(10.145)
where the first integral is for the left side of the junction, and the second integral is for the
right side.
A dc Josephson current at zero magnetic field is observed between lead and the high
temperature cuprates. This observation is evidence that the cuprates are also spin singlet
states. A Josephson current is not allowed between two different superconductors if one has
the spins in a singlet and the other has the spins in a triplet. This experiment proves that the
cuprates are not p-wave superconductors, since symmetry demands that the p-state must have
a spin triplet. See Chapter 11 for a discussion of p-wave superconductivity.
The same Fraunhofer pattern is observed for flat junctions, if one side has lead and the
other side has a cuprate superconductor. Wollman et al. (1995) made a tunnel junction that
wrapped around a ninety degree comer. The junction now has the geometry shown in Fig.
1O.16(a). The inside comer is a cuprate superconductor, while the outside is lead. It is known
that lead is an s-wave superconductor. If the cuprates are also s-wave, one should get the same
Fraunhofer pattern as for a flat junction. However, if the cuprates have d-waves symmetry, the
tunneling currents in the two arms of the comer should be out of phase. This case can be
0.8
i(<I»
0.6
0.4
0.2
0
-10 -5 o 5 10
FIGURE 10.16 (a) Geometry of a comer junction between s and d-wave superconductors. (b) Critical current vs.
applied magnetic field for a comer junction. The predicted current is zero at zero magnetic field since the currents are
in opposite directions in the two arms.
evaluated using (10.145), where the two integrals represent the two anns of the comer.
Changing the relative sign between the two integrals gives
(10.146)
(10.147)
The pattern of maximum current is no longer Fraunhofer. The predicted behavior of a comer
junction is shown in Fig. 1O.16(b). Precisely this behavior was found for comer junctions by
Wollman et al. (1995). The Josephson current vanishes at zero magnetic field because the
current is in a different direction in each ann. The two anns have to have the same length for
perfect cancellation. Their experimental result is strong evidence for d-wave super-
conductivity in the cuprates.
10.2.4. Infrared Absorption

A unique feature of the BCS theory was the prediction of an energy gap in the excitation
spectrum of an s-wave superconductor at zero temperature. Electron tunneling (Giaever,
1960) was the first measurement to convincingly demonstrate the existence of the energy gap.
However, this experiment was not commonplace at the time of the BCS theory (1957), and
the conventional method of seeking an energy gap would be to measure the optical absorption
of the metal in the far infrared. For example, Pb has Tc = 7.2K so kBTc = 0.62meY. The
BCS prediction is that the energy gap is Eg = 2.1\ = 3.52(kB TJ = 2.2 meV = 17.7 cm- I . The
value found experimentally is Eg = 22.5 cm- I . The difference from the BCS prediction is due
to the fact that Pb is a strongly coupled superconductor. The frequency range is very difficult
experimentally so that good experimental results for Pb were not reported until 1968 (Palmer
and Tinkham, 1968).
An s-wave superconductor at zero temperature should not absorb radiation for co < 2.1\.
The absorption breaks up an electron pair, which creates two quasiparticles. Each quasi-
particle has an energy of the form E = J~2 + .1\2, so absorption starts at lico = 2.1\. The
theoretical discussion follows the original theory by Mattis and Bardeen (1958).
The rate of absorption in the infrared is given by the real part of the conductivity, which
is evaluated from the Kubo formula:
1 J130 dte,ron~(T~j{q,
1t(q, ico) = - 3v . t)· j(-q, O)} (10.148)
. 1
cr(q, lCO) = -Im[1tret (q, co)] (10.149)
co
j(q) = :: L(P + iq)C:+q,O'CpO' (10.150)
m pO'
The expression for the current is appropriate for noninteracting particles. There results a
correlation function of four operators
1t(q,iO))=-4 I: (p+!q)'(p'-!q)
3m Ypp'crcr'
(10.151)
In prior discussions of optical absorption processes, it was convenient to set to zero the
photon wave vectors q. The step q = 0 is not done here because later it would be found that
the conductivity is zero. It is formally necessary to retain a nonzero value for q, although the
magnitude of q ~ 102 cm -1 can certainly be neglected compared to the wave vectors of the
electrons kF ~ 108 cm- I .
To evaluate the bracket of operators in (10.151), fix one spin, say cr = t. The answer is
twice this result, since the same contribution is found for cr = l The operators in the angle
brackets can be paired two ways (the pairings with q = 0 are not allowed since q is nonzero):
(T'rC~+q,tCt)CptCt)C~'_q,cr'(O)Cp'cr'(O)) = -8cr,=t8p'=p+q'§(p, T)'§(p + q - T) (10.152)

+ 8cr,=t8p'=_pff(p, T)fft(p + q, T) (10.153)
These pairings evaluate the correlation function, which is written in both T-space and Fourier
space:
2e2
1t(q, iO)) = --I:(p + lq)2 J~'dTelffin'r
3m2 y p 2 0
X ['§(p, T)'§(p + q, -T) + ff(p, -T)fft(p + q, T)] (10.154)

. 2e2 1 2 1
1t(q, 10)) = - 32 I:(p + zq) A I:
mY p fJip
x ['§(p, ip)'§(p + q, ip + 0)) + ff(p, ip)fft(p + q, ip + iO))] (10.155)
Note that the vector vertex of the current operator changes the relative minus sign between the
two terms in (10.152) to a relative plus in (10.155). The two terms in (10.155) can be
represented by the two diagrams shown in Fig. 10.17. The photon of (q, 0)) enters and leaves,
and couples to the two solid lines of the polarization diagram. The solid lines can be either the
pair '§'§ or fffft as these are the only two possibilities which conserve particle number at the
two vertices. One could not, for example, have the combination '§ff since that does not
conserve particle number at one of the vertices. The crucial aspect of the above expression is
the relative sign between the two terms '§'§ and fffft. There seems to be no easy way to
deduce this sign, except by being careful.
FIGURE 10.17
The next step in the derivation is the summation over Matsubara frequencies. The
summation for $' $'t was given previously in (10.117), while the other one is
1
A I: ~(p, ip)~(k, ip + iw)
I-' lp
(10.156)
At nonzero temperatures, there are a number nF(E) of quasiparticle excitations with energy E
which can absorb the radiation even for hw < 2L1. The interesting case is the limit of zero
temperature, where there is no absorption until hw > 2L1. At zero temperature the nonzero
terms are
2r?
n(q, iw) = - 3
2 I:(p +1q)2
m v p
(10.157)
The retarded function is obtained from iw -+ w + if>, and taking the imaginary part of the
retarded function brings us to the real part of the conductivity as v -+ 00 for w > 0:
(10.158)
This expression has a reasonable form. The photon is absorbed by having its energy hw
converted into two quasiparticles with separate energies Ep and Ep+q.
The combination of coherence factors u~+qV~ - upvpup+qvp+qdetermines the rate of
absorption. This quantity vanishes as q -+ O. It was remarked that q «p, so that one should
indeed take this limit. Then one would conclude that the absorbing power of the super-
conductor was very weak, at least for the process of creating two quasiparticles. This
conclusion is valid in the case where momentum is a good quantum number, so the two
quasiparticles made by the photon have momenta which are related, i.e., p and p + q.
Mattis and Bardeen assumed that momentum was not conserved. Instead, the two excited
quasiparticles have momentum p and pi which are unrelated. The lack of momentum
conservation is characteristic of the so-called dirty superconductor. The momentum breaking
might occur, for example, because the experiments are done on evaporated metal films which
are not crystalline. Alternately, there might be phonons or impurities which scatter the
particles. In any case, it is assumed that energy is conserved but not momentum. Separately
sum over p and pi which leads to the expression
where Co is a constant which will be determined later. The terms in the coherence factors v~
and u~, which are linear in ;p and ;p" vanish because they are odd functions of ;p and ;p"
Then change variables to E = Ep and E' = Epl, and derive the expression
cr(ro) = Co
ro
JooA dEP(E)JAoo dE'p(E')i3(ro - E - E')(l -~)
EE'
(10.159)
The E' integral is done by energy conservation:
C
cr(ro) = ~0(ro
ro
- 2A) Jro A
A
- dEp(E)p(ro (~?
- E) 1 -
(
Ero-E
) )
This integral has the same form as (10.105) and is evaluated in the same manner: change
variables of integration to x, where 2E = ro + x(ro - 2Ll), and the integral has the form for
complete elliptic integrals:
ro - 2Ll
oc=--- (10.160)
ro + 2Ll
CO
cr(ro) = 2ro 0(ro - 2Ll)(ro - 2Ll)
JI dx [(1 _ x2)(1
-I
1 - rxx2
_ oc x
2 2)]1/2
(10.161)
cr(ro) = Co 0(ro - 2Ll)[ro + 2Ll)E(oc) - 4M(oc)]
ro
The absorption for a normal metal has Ll = 0, which in the same units is Co. The ratio of the
conductivity in the superconductor to that of the normal metals is
cr (ro) 1
~() = -0(ro - 2Ll)[(ro + 2Ll)E(oc) - 4M(oc)] (10.162)
crn ro ro
which is the result of Mattis and Bardeen.
The experiment measures the conductivity in the normal metal and the superconductor at
the same temperature. The metal is made normal, at the low temperature, by the application of
a small magnetic field. The experimental results of Palmer and Tinkham (1968) for Pb are
shown as the points in Fig. 10.18. They compare well with the theory of Mattis and Bardeen,
Eq. (10.162), which is shown as the solid line. There is no absorption until the energy gap of
22.5 cm- I . Then the absorption rises gradually with increasing roo It does not, as in the case of
tunneling between two superconductors, rise discontinuously at the threshold frequency.
The agreement between theory and experiment is obviously excellent. In fact it is
initially surprising that the agreement is that good. After all, Pb is a strongly coupled
superconductor, and there is no particular reason to expect the BCS theory to work well for
this metal. This question was investigated by Shaw and Swihart (1968), who solved the theory
of infrared absorption using the strong coupling theory. They found that the results were
nearly identical to the predictions of the BCS theory. This experiment is not sensitive to the
strength of the coupling. There are, however, numerous other experiments which can
distinguish between these theories.
Many far-infrared experiments have been performed on high-temperature super-
conductors. The results have generally been inconclusive. There is no peak which can be
identified as due to the energy gap of the superconductors. The cuprates are oxides, and have
many optical phonons in the same range of frequency, which complicates the interpretation.
Of course, if the order parameter has d-wave character, then there is no energy gap in the
infrared. The node in the gap function means that there is a distribution of energy gaps, which
extends down to zero energy.
1.0
+ SAMPLE A
o SAMPLE B
o SAMPLE C
- THEORY
0
o +
bZ + +
0.5 ~+
"b- 'I-
O.Ot-----h:-~~!i-"I!o~~tt----+_---_+_----+_---___l
o 10 20 30 50
e
40 60
FREQUENCY (em-I)
FIGURE 10.18 Far-infrared absorption in superconducting lead at 2K. The solid line is BCS theory, and points are
experimental. Source: Palmer and Tinkham (1968) (used with permission).
10.3. STRONG COUPLING THEORY
The BCS theory is based on a simple approximation (10.2) to the attractive interaction
between two electrons. This theory successfully describes the properties of many elemental
superconductors. Nevertheless, there are some superconductors for which the e1ectron-
phonon coupling is quite strong, and here the BCS theory is not an accurate description of the
superconducting state. Instead, one must solve the gap equation for a realistic interaction
between two electrons, which includes both the attraction due to phonons and the repulsion
due to the screened Coulomb interaction. It is important that the phonon term be fully
retarded, i.e., that the interaction be frequency dependent. This strong coupling theory was
described by E1iashberg (1960). One of its important features is that the gap function A(ro)
becomes a complex function of the real frequency ro. Alternately, it is a real function A(iPn) of
the imaginary frequency Pw In addition, the self-energy function of the electrons, due to
electron-phonon interaction, must be retained and plays an important role in the analysis. The
discussion of strong coupling theory follows Sca1apino et al. (1966).
The starting point in the theory is the screened interaction between two electrons, which
is written in the conventional way as the sum of a screened Coulomb interaction and a
screened phonon interaction (see Sees. 7.3 and 7.4):
Veff(q, iro) = Ve(q) + I: Mi(q)9h(q, iro) (10.163)
41te2
"
V/q) = q2E(q) (10.164)
if (q) = M,,(q) (10.165)

" E(q)
Sec. 10.3 • Strong Coupling Theory 665
The Coulomb term Ve is taken to be instantaneous, since the frequency dependence of the
dielectric function seq, co) is not important for the low frequencies of interest, co:::: COD.
Similarly, the screened electron-phonon coupling constant M.,(q) is also taken to be a
function of wave vector but not frequency. These were the same approximations used in Sec.
7.4 to discuss the electron-phonon effects in normal metals.
The objective is to calculate, self-consistently, the properties of the correlation functions
t'§, lJ', and lJ't. The last two are again equal, so it is sufficient to find one of them. They have
the same definition (10.31) as before. Now they will be evaluated by directly expanding the S
matrix and summing the terms which are found in higher order:
t'§(p, , - ,') = -(T~Cpcr(')C:cr("») = -(Ti:pcr(')SC: cr (,'»)
lJ'(p, , - ,') = (T~C_p.J-(')Cpt("») = (T~C_p",(,)SCpt("»)
lJ't(p, , - ,') = (T~C~t(')C!p/"») = (T~C~t(')SC!p",("»)
Two different self-energy functions are retained in this expansion. Both are one-phonon self-
energies, where the electron line in the self-energy can be either t'§ or lJ':
S(p, ip) = - J 3q3i L Veff(q, iq)t'§(p + q, ip + iq)

d
(2n) I-' iq
(10.166)
W(p, ip) = - Jd 3q3i L Veff(q, iq)lJ'(p + q, ip + iq)

(2n) I-' iq
(10.167)
These self-energies are shown as Feynman diagrams in Fig. 10.19. The dashed line is Veff in
(10.163). The electron line, in the self-energy expression, contains the Green's functions t'§ or
lJ' which are to be determined self-consistently. This theory neglects the vertex corrections,
which are the self-energy diagrams of the type shown in Fig. 10.19(c). This neglect is
customarily justified on the basis of a theorem due to Migdal (1958), who argued that such
terms were smaller by a factor of Jmj M, where m is the electron mass and M is the ion mass.
The difficulty with his argument is that superconductivity itself (i.e., the Cooper instability) is
caused by a vertex diagram. A theorem which asserts that all vertex diagrams may be
neglected is obviously unreliable. The strong coupling theory is described in its customary
form, which does neglect these vertex corrections.
First discuss the wave vector dependence of W(p, iPn). The wave vectors of interest are
on the Fermi surface. Since the Fermi surface is not a sphere, the magnitude of the Fermi
wave vector will depend upon the direction in the crystal. For an isotropic s-wave super-
conductor, then just ignore the variation in the Fermi wave vector. This approximation is
tantamount to neglecting the wave vector dependence, and writing the gap as W(ipn)' There
are many cases where this approximation is poor, and one should include the dependence of
the gap upon direction in the crystal. Because of crystal symmetry, this dependence can
usually be accomplished with a few functions, which are the basis functions of the group
---
(el
""""-- .......
/ '" .... ... /
/ ....
,
/
I(
'~ / , \ }
FIGURE 10.19
which describes the crystal symmetry. For cubic crystals these are called Kubic Harmonics
(van der Lage and Bethe, 1947). For other crystals, similar functions can be constructed. For
example, in high-temperature superconductors, recent experiments have indicated that the gap
may have d-wave symmetry. The symmetry is xl - y. Define the angle cI> for the two-
dimensional square lattice by tan( cI» = Icy/ kx • Then a good approximation to the gap function
is W(p, ipn) = Wz(ipn)cos(2c1». The angular function cos(2c1»gives the proper symmetry of
being plus one along the ±X axes and minus one along the ±y axes. This ansatz form could be
used in the gap equation. In this case it is relatively easy to integrate out the angle cI> so that
one is left again with solving a matrix equation where the only important variable is the
imaginary frequency.
Normally the expansion of the S matrix leads to a Dyson equation of the typical form:
t;§(P) = t;§(0)(p)[1 + ~(p)t;§(P)] (10.168)
The equivalent expansion is derived for the superconductor. The situation is obviously more
complicated, since there are two self-energy functions S(P) and W(P), and two Green's
functions t;§ and ff. The noninteracting function t;§(0)(p) = 1/(ip - ~p) as usual, but obviously
there is not an equivalent form for ,9'(0), so ,9'(0) = O. The relevant Dyson's equations are, in a
four-vector notation p == (p, ip)
t;§(P) = t;§(0)(p)[1 +S(p)t;§(P) - W(P),9't(P)]
(10.169)
,9'(P) = t;§(o(-p)[W(p)t;§(p) + S(-p),9't(P)]
These equations are shown diagrammatically in Fig. 10.20. The key to constructing the
equations is to remember that the vertex conserves particle number, so that the two electron
arrows must point in the same direction; i.e., one can only have vertices of the form below.
These equations are simplified because W(-p) = W(P). Dyson's equations are quite similar
, I
I I
I I
to I ... ill ' ..
to the BCS equations (10.56). This resemblance is made more apparent by rewriting (10.169)
using t;§(o(±p) = 1/(±ip - ~p) to obtain
lip - ~p - S(P)]t;§ + W(P),9't = 1 (10.170)
lip + ~p + S( -p)],9't + W(P)t;§ = 0 (10.171)
The BCS equations (10.56) are obtained in the approximation where W(p, ip) =
Ll = constant, and the other self-energy S(p,ip) is ignored. The more general equations
(10.171) are now solved. The gap function W(P) is an even function of ip, as already stated
-.
--q
11ft .
FIGURE 10.20
Sec. 10.3 • Strong Coupling Theory 667
above. The other self-energy S(P) is not. It is convenient to break: it up into its symmetric and
anti symmetric parts:
(10.172)
where Se and So are both even functions of frequency ip. Furthermore, the function Z(p, ip) is
defined as
Z(p, ip) = 1 - So(P, ip) (10.173)
The notation Z usually is applied to a renormalization coefficient for the single-particle

Green's function. Here the definition is different, but the physics is similar, since Z again turns
out to be another type of renormalization function. Now it is easy to solve the coupled
equations (10.171) by writing
ip - ~p - S(P) = ipZ(P) - ~P' (10.174)
(10.175)
t;§(P) = [ipz(p)f - [~; + w(Pi]
/F(p) _ _ _-=--_W--=(p~)_ _-=- (10.176)
- [ipz(p)f - [~; + w(Pi]
These equations are similar to those for the BCS theory (10.58). The present theory requires
that the functions W(P), Z(P), and Se(P) are found self-consistently.
Eliashberg examined the retarded function Re[Sip, (0)] in great detail. He showed that it
was a constant, independent of (p, (0), which therefore just renormalizes the chemical
potential. This conclusion should not be surprising. The earlier investigation in Sec. 7.4 of the
electron-phonon properties of normal metals showed that Re[~(p, (0)] was an asymmetric
function of frequency, so that the symmetric part of the self-energy is small. This conclusion
does not change in the superconductor. Neglect Se.
The other functions of interest Z(P) and W(P) are both functions of (p, ip). The wave
vector dependence, for s-wave superconductors, is unimportant, since everything happens at
the Fermi surface p ~ kF . It is the energy dependence of these functions which provides the
interesting phenomena. Actual crystalline superconductors have anisotropic Fermi surfaces,
which cause the functions to vary with angle around the Fermi surface. This dependence is
neglected, and isotropy is assumed. In d-wave or p-wave superconductors the angular
dependence of the order parameter is also important around the Fermi surface.
For s-wave superconductors the notation is altered to suppress the dependence on wave
vector. These functions are written as_ So(ip), W(ip), Z(ip), etc. The main dependence of t;§(P)
and /F (P) on p is through the factor ~p ~ ~p. All the wave vector dependence enters through
the integrals of the Green's functions:
J d~ t;§( i ) - -Joo d~ iPnZ(P) + ~p _ _ rcipn Z

'P
OO
-00 p, 'Pn - -00 'P~; + W2 + (PnZ)2 - Jp~Z2 + W2
Joo d~'P /F(p ,'Pn

-00
i ) - Joo
- -00
d~ W(P) _
'P~; + W2 + (Pnzi - Jp~z2 + W2
rcW(P)
These integrals are evaluated by contour integration, say by closing the contour in the upper
half plane.
The earliest work on superconductivity converted these gap equations to real frequency
and solved them as integral equations over the energy and frequency. That process is
cumbersome, since each retarded function (w, Z) has a real and imaginary part. An alternate
procedure was done by Vidberg and Serene (1977). They solved the gap equations as a
function of complex frequency. The advantage is that all functions are real when expressed as
a function of complex frequency. The solution to the self-consistent gap equation is much
easier using only real functions. The hard part of the numerical work is then the analytical
continuation of the numerical results to the real frequency axis. Since Pn = (2n + l)nkBT, the
relevent information is (2n + 1). Label these functions as W(iPn) == W(2n + 1), S(iPn) ==
S(2n + 1), etc. The end result of the computer solution is a pair of vectors representing these
two functions. The best way to convert them to the axis of real frequency is to use Pade
approximants (Baker, 1975). Vidberg and Serene showed this method gives excellent results.
The goal is to evaluate the gap Eqns. (10.166) using the Green's functions in (10.176).
These integral equations will be reduced to a single independent variable, which is the
imaginary frequency. The first step is to do all the wave vector integrals, whose variables are
(10.178)
where v = cos(e) is the angle between p and q. Change the integration variable to
P' = Jp 2 + q2+ 2pqv, or p'dp' = pqdv, and as a second step change p'dp' = md~':
J d 3q = ~ J~ qdq Jd~' J: rr
d<p = :p J Jd~'
d 2q (10.179)
The q integral is over the Brillouin zone; the inclusion of Umklapp processes extends the
integration to infinity. The above change of variables restricts IP - p'l ::::; q ::::; IP + p'l. Since
both (P,p') are near to kp , the wave vector integral is restricted to 0::::; q ::::; 2kp . The d~'
integral will be limited to values of I~'I < roD at the Fermi surface, since the integral
converges rapidly outside of this range. With this change of variables, the wave vector
integrals are quite simple to do, at least in a formal sense. The integral over d~' was already
performed in (10.177). The remaining integration over the coordinates d 2 q provides a
constant Ilc for the electron-electron interaction, and the exact definition of the function
(X2 F( ro) for the phonon term. The latter was defined in Sec. 6.4
1J d 2q Vq i2 J2kF dq
Ilc = hvp (2n)3 c(q) = nhvp 0 qc(q)
(10.180)
(X
2
F(ro) 1 J d q I:M,,(q)
=~
2
-3
- 2
o[ro - ro,,(q)] (10.181)
flVp (2n) "
After performing these wave vector integrations, the self-consistent equations (10.166) for the
two self-energy functions have been reduced to manageable proportions:
(10.182)
The Coulomb term Ilc is a dimensionless constant, which acts to reduce the gap function.
Sec.10.3 • Strong Coupling Theory 669
The formulas (10.182) for S(iPn) and W(ipn) can be simplified. First change the notation
to S(2n + I), W(2n + I). Secondly, introduce the gap function A(2n + I) and phonon
interaction V(m)
.1(2 + I) = W(2n + I) (10.183)

n Z(2n + I)
V(m) =2 Jrodro 0(: F(ro~

ro + qm
(10.184)
Note that V(O) = J.., where here J.. is the dimensionless strength of the electron-phonon
interaction. Also note that V(m) < J.. for m =f:. O. The coupled gap equations are, after chan-
ging summation variable to km = Pn + qm
S(2n + 1) = -13ire ~A(2m + 1)[V(n - m) -Ilc] (10.185)
k
A(2m + 1) = m (10.186)
J(kmi + A(2m + Ii
re
Z(2n + I)A(2n + 1) = ~ ~ cI>(2m + 1)[V(n - m) -Ilcl (10.187)
cI>(2m + 1) = A(2m + 1) (10.188)

J(kmi + A(2m + Ii
Examine the parity of these functions. The phonon interaction V( -m) = V(m) is an even
function of its argument. The discrete frequency is odd: kn = (2n + l)rekBT changes sign if
(2n + 1) -+ -(2n + 1). This feature makes A(2n + 1) an odd function
(A(-2n - 1) = -A(2n + 1», which also makes S(2n + 1) an odd function. This term is
identified as S(2n + 1) = iPnSo(ipn), so the factor of Z is
1
Z(2n + 1) = 1 + - 21 LA(2m + 1)[V(n - m) -Ilcl (10.189)
n+ m
1
= 1 + --LA(2m + 1)V(n - m) (10.190)
2n + 1 m
where the term in Ilc vanishes when one sums over positive and negative values of m due to
the oddness of A. The above formula (10.190) can be used to show that Z(2n + 1) is an even
function of its argument (i.e., change m = -m' - 1, n = -n' - 1).
The remaining issue is the symmetry of the gap function A(2n + 1) as defined by Eq.
(10.187). Since Z and Yare both even functions, then the gap function can be either pure even
or pure odd. A gap function which is odd in frequency has the feature that it must vanish at
zero frequency A(iPn = 0) = o. In this case it is not superconducting. The physical solution
has the gap an even function of frequency. The two equations to be solved are recollected
here:
1
Z(2n+l)=I+-
2n+ l LA(2m+l)V(n-m)
m
(10.191)
Z(2n + I)A(2n + 1) = rekBTLcI>(2m + 1)[V(n - m) -Ilc]
m
They are called "the gap equations".

r,
I '" " ,\
.. 3 1
> \
\
,
\
oS N ",
<I 2 \
\
+
.
<I
"3
1
I
" t::..
\
\
\
,
I
<l I
Z 0
Q
t-
V
Z
;:) ·1
LI..
II.
<l
(!)
-2
FIGURE 10.21 The real and imaginary parts 8 1 and 8 2 of the complex gap function as a function of energy co for
lead at T = O. Source: Shaw and Swihart (1968) (used with permission).
These two equations must be solved self-consistently for the two real functions
~(2n + 1) and Z(2n + 1). The inputs for the calculation are the kernels Vern), which depends
on the phonon properties through a2 F, and the Coulomb factor j..le' Values of the Coulomb
constant j..le for actual superconductors are about 0.1-0.2. The temperature Tenters explicitly
in the prefactor T in the equation for ~. It also enters into the function kn which is found in Ie,
<1>, and V
The first solution to the gap function, using real frequency, was by Shaw and Swihart
(1968) for lead at T = 0 as shown in Fig. lO.21. They used the a2F of McMillan and Rowell
(1965), which is shown in Fig. 7. 13 (a). The real part of the gap function Re[~(ro = 0)] == ~1
equals the value 1.34 me V found from electron tunneling. The imaginary part of the gap
function Im[~(ro)] == ~2 is zero for ro < ~o. Both real and imaginary parts of ~(ro) are
obviously strong functions of roo Shaw and Swihart also calculated the infrared absorption and
showed excellent agreement with the results of Palmer and Tinkham. The electron tunneling
data for a2 F, along with the strong coupling theory, provides an excellent description of other
experiments such as infrared absorption. The later calculations of Vidberg and Serene (1977)
using the method of imaginary frequency and Pade approximants, obtained the same result
for Pb.
10.4. TRANSITION TEMPERATURE
The transition temperature of a superconductor is called Te. It is one of the easier

quantities to calculate. A simple method was suggested by Owen and Scalapino (1971). It has
been adopted widely, and is described here. The trick is to retain the imaginary frequency
throughout the calculation. One advantage of using imaginary frequency is that all of the
quantities are real functions, which simplifies the computation. A second advantage of the
method is that it is relatively easy to include embellishments such as the influence on Te of
impurity scattering, electron-electron interactions, etc. Here a calculation is presented for the
Sec.10.4 • Transition Temperature 671
simple case of an isotropic s-wave superconductor. The extension to more complex systems is
described by Allen and Mitrovic (1982).
For temperatures larger than Te the superconducting energy gap is zero. At temperatures
equal to Te the gap starts to have nonzero value. The transition temperature is the place where
the energy gap is infinitesimally small. The value of Te is found by setting Ll = 0 in all of the
denominators of (10.191)
1
Z(2n + 1) = 1 + - 21 Lsgn(2m + 1)V(n - m) (10.192)
n+ m
Ll(2m + 1)
Z(2n + I)Ll(2n + 1) = ~ 12m + 11 [V(n - m) - /-lcl (10.193)
where sgn(x) has the sign of x. The equation for Z(2n + 1) no longer depends upon the gap
Ll(2n + 1). It can be evaluated at the start of the calculation. Recall that Z comes from the self-
energy of the electron. At Te the gap is zero so one can use the self-energy of the normal state.
The function Z(2n + 1) now has a simple form. This fact is demonstrated by evaluating
it for small values of (2n + 1). In the summation over m on the right, the terms are presented
in two brackets: the first is when (2m + 1) has positive values as m = 0, 1, 2, etc. The second
is when 2m + 1 has negative values as m = -1, -2, -3 etc. Note that V(-m) = V(m) is
symmetric. Examine the four cases that n = 0, -1, 1, -2
Z(I) = 1 + {[V(O) + V(I) + V(2) + ... ] - [V(1) + V(2) + ... ]}
= 1 + V(O) = 1 + /..
Z( -1) = 1 + (-I){[V(I) + V(2) + V(3) + ... ] - [V(O) + V(1) + V(2) + ... ]}
= 1 + V(O) = Z(1)
Z(3) = 1 + l {[V(1) + V(O) + V(I) + V(2) + ... ] - [V(2) + V(3) + ... ]}
Z(3) = 1 + l[V(O) + 2V(I)]
Z(-3) = 1 + l[V(O) + 2V(1)] = Z(3)
Z(2n+l)=1+2111[V(0)+2tV(l)]=Z(-2n-l)
I n+ 1=1
The renormalization factor Z(2n + 1) is an series with 2n + 1 terms. Note that V(O) = /..,
where /.. is the dimensionless electron-phonon coupling constant. The renormalization
function is an even function of imaginary frequency.
Now examine the gap equation (10.193). The gap Ll(2n + 1) is infinitesimally small. The
gap equation is a linear matrix equation. The unknown quantity Ll(2n + 1) is a vector. The gap
equation is actually an eigenvalue equation. The solution to this equation is obtained by
setting the determinant of the matrix equal to zero
detlMnml =0 (10.194)
M = 0 Z(2n 1) _ V(n - m) - /-le (10.195)
nm nm + 12m+ll
This "eigenvalue" equation is unusual, since there is no "eigenvalue". The temperature Tis
contained in the factor V(m) which enters into Z and Mnm. This temperature is varied until one
finds the highest value at which the determinant vanishes. This temperature is then Te' In
practice the determinant does not have to have a large dimension. For /.. > 1 then its
dimension need only be about 10(-9 S 2n + 1 S +9) for accurate answers. This dimension
of determinent is easy for present computers. The solution is a symmetric function of
frequency, in that A(-(2n + 1)) = A(2n + 1). Knowing this fact, the dimension of the
matrices can be reduced to half.
The Einstein model provides a simple example. Then all of the phonons have the same
frequency COo and ('J.2p(co) = cooAo(co - co o)/2. In this case it is useful to introduce the
dimensionless constant b = 2nkBT /(licoo). The gap equation becomes a search for the largest
value of b which satisfies the equation
A
V(m) = 1 + b2m2 (10.196)
A(2n + 1)Z(2n + 1) = A L A(2n + 2m + 1) (10.197)

m 12n + 2m + 11[1 + b2m2]
A [ n 1 ] (10.198)
Z(2n + 1) = 1 + 12n + 11 1 + 2t; 1 + b2[2
The gap equation (10.197) for the Einstein model ofphonons is easily solved to give a graph
of b vs. A, which is shown in Fig. 10.22. Above A > 1 the curve rises slowly with increasing
coupling strength A. In fact the dependence goes as b ~ -JA. There is no saturation of
b = 2nkBTc/licoo, which means no saturation of Teo The higher is A then the higher is Tc.
The asymtotic limit of large A can be found in a simple way. Assuming that b becomes
very large, then V(m) becomes increasingly small for modest values of m. In the limit oflarge
b retain only the values V(O) = A and V(l) = 1.,/(1 + b2 ). The gap equation can be solved by
using only a matrix of dimension two. It is necessary to retain only the gap components A(1)
and A( -1), and the renormalization parameter Z(1) = Z( -1) = 1 + A. When /-lc = 0 the
equations are
A(1)(1 + A) = A{ A(1) + ~~;~} (10.199)
,1(-1)(1 + A) = A{A(-l) + 1A~1~2} (10.200)
The matrix equation is 2 x 2. Setting the determinent equal to zero gives two solutions for
b(A). The only one for real values of b is
b2 =A-1 (10.201)
kBT =licoo~ (10.202)

c 2n
Keep in mind that this solution is approximate and only valid in the limit that A ~ 00.
However, this solution clearly shows that Tc increase as the square root of the electron-
phonon coupling constant at large values of the coupling constant. The exact asymptotic limit
as A ~ 00 is kBTc ~ 0.15Iicoo-JA (Allen and Mitrovic, 1982).
The electron-electron parameter /-lc plays the same role as V(m) for the electron-phonon
interaction. Combining these two interactions gives a gap equation with an effective inter-
action of (V - /-lc). For A > /-lc the interaction is attractive for small values of m but becomes
negative for larger values of m. The electron-phonon interaction is attractive only over a
limited range of imaginary frequency.
Figure 1O.22(a) shows the solution of the gap equation for the Einstein model assuming
that /-lc = 0.1. The solid line has /-lc = 0 compared to the dashed line with /-lc = 0.1. Electron-
Sec. 10.4 • Transition Temperature 673
1.4
a
1.2
b 1
/'
/'
0.8
/
0.6 /
0.4 /
0.2 /
0
/
0 0.5 1 A. 1.5 2
0.4
b
-+
.,....
-c::
(\J
<l
0.3
0.2
0.1
"\ 11*=0
0
"" '-....
11*=O.i'- - _
-0.1
0 5 10 15 20 25 30
2n+1
FIGURE 10.22 (a) The solution to the gap equations is given as b('A.). The solid line has Ilc = 0, while the dashed
line has Ilc = 0.1. (b) The eigenvectors ~(2n + I) for the two cases.
electron interactions tend to reduce the value of b = 21tkBTc/l'zroo. Another interesting

function are the eigenvectors A(2n + 1). Figure 10.22(b) shows the relative values of this
function for the case that A = 1. The two curves are the cases that /lc = 0.1 and o. Since the
gap function is symmetric in (2n + 1) only positive values are shown. Of course, the gap
function vanishes at Tc. Here we assume the temperature is approaching Tc and the gap
functions are very small. In this case only their relative size matters. Note that the two curves
are very similar near to small values of frequency (2n + 1). However, they differ markedly at
large frequency. For the case of no electron-electron interaction (/lc = 0), the relative values
of A(2n + 1) get very small and vanish at large (2n + 1) which justifies the use of matrices of
relatively small dimension in solving the gap equation.
The opposite behavior is found when /lc =1= O. At large values of (2n + 1), the gap
function A(2n + 1) does not go to zero but goes to a constant. Then there is no argument for
using a determinent of small dimension. The problem is that the effective interaction in the
gap equation V(m) - /lc does not vanish at large m but goes to a constant -/lc. Therefore one
should use large determinents. However, there is a method of using determinents of smaller
dimension, which entails rescaling the Coulomb coefficient. This procedure is described in
the review by Allen and Mitrovic (1982).
Problems
1. Consider the mutual scattering of a fermion and a boson, i.e., a 3He and a 4He particle moving
parallel in time. Use Green's functions to derive the occupation number factors which enter the
scattering integral.
2. Calculate how the pole in the Cooper instability varies with the center of mass momentum P of the
pair.
3. What is the momentum distribution nk of electrons in the superconductor at T = O?

4. The entropy per unit volume of a superconductor in the BCS model is
S = - -2kB
v k
L {nF(Ek) In[nF(Ek)] + [1 - nF(Ek)]ln[1 - nF(Ek)]}
Use this expression to derive the specific heat of the superconductor. Then use L\( T) ~ ~ to show
that the specific heat is discontinuous at the critical temperature.
5. Write the Hamiltonian (10.35) in the following form:
Heff = L ~kC~aCka - LL$k)(C~tC~k~ + Ck~Ckt)

ka k
L\(k) = - L V(q)(Ck-q~Ck+qt)
q
Solve the effective Hamiltonian Heff by a canonical transformation, and reduce it to the form
Heff = Eo + LEk(octOCk + ~t~k) (10.203)

k
Ckt = cos(8k)ock + Sin(8k)~t (10.204)
C_k,j. = cOS(8)~k - sin(8k)oct (10.205)
6. Calculate the Pauli spin susceptibility of the BCS superconducting state. Show that the formula at
nonzero temperatures is
7. Derive the formula for the coefficient in the Josephson current Js(O) = (rr.L\/2)(cro/e) by doing the
integral in (10.121) for eV = O.
8. Evaluate Js(eV) in (10.127) at zero temperature by making the change kBTLiP ~ dp/(2rr.) and J
then the analytical continuation jro ~ eV. Define A(eV) at T = 0 by using Js(eV) = (cro/e)A(eV).
Show that:
(a) A(eV = 0) = L\sK(q): q= (1 - ~i2) 1/2

where L\s and L\L are the smaller and larger of the two energy gaps.
(b) When L\L = L\s = L\, then A(eV) = L\K(eV/2L$.
Sec. Problems 675
9. Use strong coupling theory to derive a fonnula for the coefficient Js(eV) for the Josephson
tunneling.
10. Derive the strong coupling theory fonnula for the infrared absorption of a dirty superconductor.
II. At large values of", the gap equation for Jle = 0 goes as b --+ rxo~ where the best numerical
solution gives rxo = 0.1481 (Mahan, 1997). Our first estimate solved the gap equation retaining only
Ll(±l), and gave rxo = 1. Make a better estimate of this constant by solving the gap equation while
retaining Ll(±I), Ll(±3). (Hint: one cannot solve for b analytically. But letting ~ = "'lb 2 so that
V(m) = ~/m2 gives a simple equation for ~.)
12. How does the electron--electron parameter Jle change the result in (10.202)?
13. For impurity scattering, take the self-energy of an electron to be given in tenns of a T matrix
. J d 3q
L(p, IPn) = n i --3ITp,p+ql 2
. 1 (10,206)
(2n) IPn - Ep +q
This interaction does not change the energy of the electron, but does change its wave vector, Use this
function to calculate a parameter
= ~ J(2n)3ITp,p+ql
d3q 2
Yi 8(E - ~p+q) (10,207)
which enters into the gap equation and also into Z. Show that these two contributions cancel in the gap
equation for an isotropic superconductor. In this case, impurity scattering does not change the value of
Te' This was first shown by P.w. Anderson,
14. Use the Thomas-Fermi model for the dielectric function to evaluate Jlc for the homogeneous
electron gas. Give numerical estimates for rs = 2, 3, 4.
15. Repeat the above calculation for the two dimensional electron gas, where the electron dispersion is
parabolic and the Fenni surface is a circle. The major change is in the dielectric screening function.
Chapter 11
Superfluids
Helium has two common isotopes, 3He and 4He. Each isotope can be separated and a liquid
formed at low temperatures which is nearly pure 3He or pure 4He. Each has unusual prop-
erties and displays collective behavior of a unique character. The boson liquid 4He shows a
phase transition at T" = 2.172 K to a superfluid state which is similar to Bose-Einstein
condensation, although vastly modified by the strong interparticle interactions. Similarly, the
fermion liquid 3He develops a Fermi distribution at low temperature, and the particles have a
superfluid transition which is similar to the superconducting transition in a metal. Of course,
now it is occurring in a liquid, of electrically neutral atoms, so there is no Meissner effect, but
there is pairing. However, it also has a unique character, since the atoms avoid the usual
singlet pairing common to metals and instead pair with the spins aligned parallel. The triplet
pairing, in turn, leads to many new phenomena and a richer phase diagram, which was
discovered by Osheroff et al. (1972). The third superfluid in this chapter is the two-dimen-
sional electron gas under strong magnetic field, which shows the quantum Hall effect and the
fractional quantum Hall effect. It is also a highly correlated fluid.
Quantum fluids is a subject where Green's functions are often abandoned in favor of
other techniques. There are some calculations for which other techniques are better and for
which the Green's function method gives awful results. One example is the discussion of the
ground state properties of quantum liquids. Green's function techniques have been unsuc-
cessful in the study of the ground state properties of classical fluids, and other techniques are
now used for these high-density, strongly interacting, and highly correlated systems (see
Percus, 1964). It is understandable that the ground state of quantum liquids is similarly
difficult to treat with Green's functions. The description of the ground state properties of the
system will require techniques beyond those introduced here, and often these results only
summarize. However, when discussing the excitation spectra of the liquid, which are crucial
for their superfluid properties, familiar techniques are used, such as Green's functions or
perturbation theory.
11.1. LIQUID 4He
The first discussion of 4He uses pairing theory. The important approximation in pairing
theory is that the particles' interactions can be approximated as a summation of pairwise
events. Of course, the potentials between particles are a summation of pair interactions:
677
678 Chap. 11 • Superfluids
L Veri - r). But pairing theory makes a much stronger assumption: when one particle
interacts with another, it is not simultaneously interacting with other particles. Pairing theory
would be a good approximation in a gas of low density, where the particles collide occa-
sionally, and most collision events are binary. But this assumption is not obviously valid in a
liquid. In fact, it is now known that the assumption is terrible and that pairing theory is a very
bad approximation when applied to liquid helium. It makes numerical predictions of
measurable quantities, such as the speed of sound in 4He, which are a factor of 10 higher than
experiment. One should not regard the pairing theory as a serious theory of liquid helium.
There are still several reasons for introducing pairing theory. Historically it played an
important role in the history of many-particle physics. The major ideas were introduced by
Bogoliubov (1947), who proposed the idea that the particles interacted in pairs. Although this
theory was unsuccessful when applied to helium, virtually the same theory works for
superconductivity. The BCS theory is just the Bogoliubov theory, changing the operators from
bosons to fermions. Of course, there it works since the particles do have pairwise interactions
and the interactions are weak. In superfluid helium it does not work, although much effort was
expended over a period of years by many investigators before this conclusion became
obvious.
Another reason for introducing pairing theory is that it contains many ideas which are
qualitatively correct such as the condensate. It provides a good introduction to the subject.
Later it is shown how the quantum liquid properties are actually calculated using correlated
basis functions.
The starting point for this discussion is a Hamiltonian which contains the kinetic energy
of the particles and also the pairwise potential VCr)
2
H = L Pi +~LV(ri - r) (11.1)
i 2m i#i
The helium particles are treated as spherically symmetric objects, and the electronic excita-
tions are ignored. This approximation is accurate, since the atoms have a kinetic energy of
roughly 15 K, using temperature units, while the electronic excitations have a minimum of
20 eV The 4He atom is considered to be entirely structureless and is represented as a single-
boson particle with a potential VCr) when interacting with other similar bosons. The 3He atom
has additional degrees of freedom associated with its nuclear spin. The spin is an important
property, since it is what makes 3He have its fermion character. The helium-helium potential
which was used most frequently in the early days of physics was a Lennard-Jones potential,
VCr) = 4£[ (~)'2 _(~)6J (11.2)
£ = 1.484 X 10- 22 J (11.3)

cr = 2.648 A (11.4)
which is often called a 6-12 potential. The parameter £ ~ 10K is the maximum well depth,
and cr is the hard core radius, where the potential rises steeply. These parameters were
obtained by fitting various experimental results to this potential function (see Hirschfelder et
al., 1954). It should be appreciated that this potential form was chosen for its mathematical
simplicity and not because it was a good approximation to the actual potential shape.
Certainly the long-range attractive potential behaves as r- 6 from van der Waals forces, but the
repulsive part is probably not r-12. Recent efforts to deduce this potential function have
produced better versions.
Sec. 11.1 • Liquid 4He 679
In the pairing theory the Hamiltonian is written H = Ho + V, where Ho is the kinetic

energy term. All operators are expanded in the basis set of H o, which are plane-wave states.
Rewrite the Hamiltonian (11.1) as
(11.5)
(11.6)
An additional spin index should be added when describing 3He. The effective interaction
V(q) is the Fourier transform of the interparticle potential VCr). One immediate problem is
that the Fourier transform may not exist. Certainly there is no Fourier transform for a
Lennard-Jones potential. It is shown later that the potential can be replaced by a T matrix,
which always exists regardless of the potential strength or shape. This divergence is formally
eliminated until the T matrix is introduced by assuming the divergence is cut off by a
parameter g and by later letting g -+ 00. The nonexistence of V(q) is not really a serious
problem, since the replacement of scattering properties by a T matrix is formally exact for
binary collisions. This theory would give a good description of a dilute gas.
11.1.1. Hartree and Exchange

Liquid helium is not a weakly interacting system, but it is treated as such in pairing
theory. It is assumed, for the present discussion, that the potential terms are weak. The self-
energy is found from the first terms which occur in the S-matrix expansion. The first two
contributions are identical to the terms found for the homogeneous electron gas, which has
the same formal kind of Hamiltonian as (11.5). They are the Hartree and exchange energies:
1
~(k) = noV(q = 0) ±- Ln(~p+q)V(q) (11. 7)
v q
The ± signs refer to 4He( +) and 3He( -), respectively. The 3He case is identical to that of the
electron gas, where the fermion nature makes the exchange energy have the opposite sign as
the Hartree energy. The fermions of like spin wish to avoid each other. For 4He, its boson
character makes the opposite occur: the quantum nature of the particles tends to make them
prefer to collect at the same spot, so that the exchange energy is positive. For a system with
only two spinless bosons, the two-particle wave function must be symmetric under an
interchange of coordinates, which can be achieved with the choice
1
'Pk1k/rl, r2) = .J2 [<Pk1(r,)<pk (r2) + <PkJr 2)<Pk (rl)]
2 2 (11.8)
<Pk(r) = _1 dk ' r (11.9)

.;v
The expectation value of the potential energy for this wave function is
(V) = klLk2 n(Sk )n(Sk2) f d 3r\ d3r21'1\ k2(r\, r2)1 2V(r\

I I
- r2) (11.1 0)
= V12 Ln(Skl)n(Sk2)fd3r\d3r2[1 + ei(kl-k2)·(rl-r2)]V(r\ - r2)

klk2
1
=- L n(Sk)n(Sk )[V(q = 0) + V(k\ - k 2)] (11.11)
V klk2 2
In the integral in the middle equation, the integration variables were changed to center of mass
f
coordinates R = (r\ + r2)/2 and r = (r\ - r2), and d 3R = v. The remaining integral over
d 3 r produces V(q = 0) in the first tenn and V(k\ - k 2) in the second. The expression for (V)
has two tenns, which are the Rartree and exchange energies. They have a relative plus sign,
whose origin is the relative plus sign between the two tenns in the two-particle wave function
(11.8). The two-particle fennion wave function, when the spins are parallel, has a relative
minus sign, which causes the negative sign in front of the exchange energy. The occupation
factors n(Sk) are either nB for 4Re or nF for 3Re particles. The subscript is omitted in the
fonnulas, since they can apply to either case equally. The diagrams for Rartree and exchange
are shown in Figs. 11.1 (a) and 11.1 (b).
The next step is to improve the Rartree and exchange energies by replacing the potential
V( q) by the equivalent T -matrix result. The Rartree energy is the potential energy of a particle
in k\ from the interaction with the other particles in the state k2 by the q = 0 interaction
V(q = 0). This interaction can be viewed as a process whereby the two particles interact by
mutually scattering. Go to center of mass coordinates: k = (k\ - k 2)/2, K = (k\ + k2). The
mutual scattering of two particles, which interact by central forces, does not alter the center of
mass wave vector K. It is unaffected by the scattering process, so that the scattering can be
viewed in a rest frame where K = O. This scattering is shown in Fig. l1.l(c). The two
particles approach with k and -k, scatter with a q = 0 interaction, and depart with their same
wave vectors. In the center of mass transfonnation, the labeling convention is that the particle
Ok2
I
,,11=0 ~l
) )
kl k, k,
(a)
o
( b)
: 1=0
~
(e) (d)
FIGURE 11.1
k is the one whose self-energy is being calculated, while - k is the other particle from which
the first is scattering. The particle line of -k is part of a closed loop.
The scattering of two particles is described by a T matrix Tkk" which was introduced in
Sec. 4.1. The choice of T matrix, rather than reaction matrix, is based on wanting retarded
self-energies, since the T matrix has the proper dependence on ib in the energy denominators.
Recall that the on-shell T matrix Tkk' applies only for elastic scattering, when the particle
energy was E = ~k = ~k" The kinetic energy ~k is conserved in relative coordinates, so this
requirement is satisfied. The retarded self-energy for particles of energy ~k in the Hartree
approximation should be
(11.12)
In the scattering from other particles, whose wave vector is k2' the scattering rate depends on
the difference k = (k 1 - k 2)/2 rather than on q = O. The forward scattering with k' = k is
easy to obtain from Sec. 4.1,
Tkk = - 41tkL (21 + l)eii>t(k) sin[bl(k)] (11.13)

2J..l I
where J..l is the reduced mass. By changing integration to k = (k 1 - k 2)/2 and d 3 k2 = 8d3 k,
the Hartree energy is
LH(k 1, ~k ) = - -161t J--3

d k 3
n(~k -2k)-k1 L (21 + l)e'Mk)
'.
sm[bl(k)]
1 J..l (21t) 1 I
The same considerations should now be applied to the scattering term for the exchange
energy. Examine the scattering in the center of mass coordinates, where the incoming
particles have wave vectors k and -k. This diagram is shown in Fig. 11.1(d). Here the two
particles exchange momentum 2k, so the final states are still k and -k. The solid line
indicates the particle lines which are connected to make the self-energy diagram for
exchange. This term has aT-matrix expression:
1
Lx(k 1, ~kl) = ~ Ln(~k)T(kl-k2)/2,(k2-kl)/2
k2
=- J
161t d3k3
J..l (21t)
n(~k _2k)-k1 L(21 + 1)( -lieil>t(k) sin[bl(k)]
1 I
(11.14)
When making the expansion in angular momentum states I, then PI ( -1) = (-1 ( The two
self-energy terms, Hartree and exchange, are either added or subtracted, respectively, for
bosons and fermions. These two self-energy expressions are added for 4He:
L(k, ~k) = - ~ foo k'dk' L

J..l1t 0 even I
(21 + 1)eiO,(k') sin[bl(k')] Jdnk,n(~k_2k')
For 4He, the two terms combine to eliminate angular momentum components I which are odd.
Only the phase shifts bl(k) for I even need to be computed, and the others are irrelevant.
For 3He, there is exchange scattering between two particles only when the spins are
parallel, and then the scattering rate has no dependence upon the angular momentum terms
with I even but contains only terms with I being odd. Since there is no exchange scattering
between particles with antiparallel spin, the energy of a particle contains an average over
events with spin parallel and antiparallel. For the self-energy for a particle in 3He then
L(k) = 2LH - Lx,
11.1.2. Bogoliubov Theory of 4He

The Bogoliubov (1947) theory describes Bose-Einstein condensation in a weakly
interacting system. It is basically a form of pairing theory for the interacting bosons. The
Hamiltonian (11.5) for a system of interacting bosons is expressed in the plane-wave
representation. The main feature of the theory is the assumption that a Bose-Einstein
condensation occurs at a nonzero temperature TA• The condensation occurs because the
chemical potential j.l goes to zero, and there is a macroscopic occupation of the zero-
momentum state. The collection of particles in the zero-momentum state is called the
condensate.
The 4He particles are treated as bosons. Because the number of particles is fixed, the
number operator
(11.15)
(11.16)
contains a chemical potential j.l. It is usually temperature dependent and varies to satisfY the
implied definition in (11.16) that the integral over the occupation number gives the total
number of particles N. The particle energy (Ok is yet to be determined, self-consistently, but it
is positive and the zero of energy is set at (Ok=O = O. With these conventions, the chemical
potential for most boson systems is negative, so that the energy denominator in nB«(Ok) is
never zero because (Ok > j.l. However, in Bose-Einstein condensation, the chemical potential
j.l vanishes, as illustrated in Fig. 11.2. The occupation number nB«(Ok) diverges as k --+ 0, and
the number of particles in the k = 0 states is not well defined by this limit. Instead, assume
that the number of particles in the state k = 0 is some number No which is a finite fraction of
the particle number N. Indeed, for a weakly interacting system, the condensate fraction
fo = No/N approachs a number near unity as the temperature is lowered below TA• For
T < TA, the particle number operator is expressed as
(11.17)
The delta function at k = 0 gives the number of particles in the condensate, and the second
term applies only for k =j:. O. One of the goals of the theory is to calculate this fractionfo self-
consistently.
The creation and destruction operators ct
and Ck must have special properties for
k = O. Normally, for bosons, when there are nk particles in the state k, the creation operator
T~ T
FIGURE 11.2 Chemical potential as a function of temperature. Note that J.l becomes zero at T). in Bose-Einstein
condensation.
gives Ctlnk) = Jnk + link + 1). In normal systems, the occupation numbers nk are integers
on the order of unity. For the state with k = 0, the average value of (no) = No ~ 1023. Then
the operations give
CJINo) = JNo + 11No + 1) ~ PaiNo) (11.18)

CoINo) = PaiNo - 1) ~ PaiNo) (11.19)
where the numbers No ± 1 ~ No when No "-' 1023 . Both the creation and destruction opera-
tors have a similar effect upon the system and give the result $0 as an effective eigenvalue.
The number No is sufficiently large that the operators CJ and Co are treated as scalars whose
value is $0.
(11.20)
The next step is to examine the Hamiltonian and to isolate all terms in which any of the C
operators has k = O. These terms are replaced by $0 for both CJ and Co. Since Il = 0, write
the kinetic energy term as
, ,
~);kCtCk = coNo + I)kCtCk = LCkCtCk (11.21)
k k k
The prime on the summation means to omit the term with k = O. The kinetic energy has no
dependence on the condensate, since the particles have a zero kinetic energy in this state. The
interesting terms arise from the potential energy:
(11.22)
When the condensate fraction/o is near unity, the largest term is when k = k' = q = 0, which
gives an energy contribution in terms of the condensate density Po = Nolv = fop:
(11.23)
The next largest terms are those in which only three of the C operators has zero wave vector,
but none exist. If any three particles have zero wave vectors, then so does the fourth, and that
term has been counted already. The next largest terms are those in which two of the wave
vectors are zero and two are not. There are six ways of pairing the four C operators in the
potential energy term:
k=k'=O 1POLq V(q)C~C!q

k = q = -k' 1POLq V(q)CqC_q
q=O=k 1Po V(O)Lk'CZ, Ck'
(11.24)
q=O=k' 1Po V(O)LkCtCk
k=O,k' =-q 1POLq V(q)C~Cq
k' = O,k=q 1POLq V(q)C~Cq
It is customary to stop at this point and to keep only these tenus in an effective Hamiltonian:
(11.25)
The tenu Po V(O) is dropped, since this constant tenu just causes a shift in the chemical
potential, which is still zero. The effective Hamiltonian given in (11.25) can be diagonalized
exactly, which is the strongest argument for using it as the effective Hamiltonian. A better
theory would retain additional tenus. There are tenus with three C operators with nonzero
wave vectors, and of course tenus with four. The Hamiltonian is written as Ho + V, where Ho
is given in (11.25). It contains the tenus quadratic in the operators, while V is the other tenus
which contain three or four creation-destruction operators. The tenus are separated in this
manner because Ho is a Hamiltonian which can be solved exactly. The other tenu V must be
treated as the interaction, which is evaluated by the usual Green's function method to give
additional self-energy corrections.
The particles in the k = 0 state are the condensate, which is the ground state of the
system. Quasiparticles with k I- 0 are excitations, and Ho is the effective Hamiltonian of
these excitations. It contains combinations of operators such as CtC~k which correspond to
the excitation of two particles from the condensate. Since momentum is conserved, they must
have k and -k. Similarly, the tenu with CkC_ k is the destruction of two quasiparticles with k
and -k when both are returned to the condensate. The tenus with ctck correspond to the
scattering of quasiparticles by the condensate, and there are both direct and exchange
processes which lead to the tenus with V(O) and V(k), respectively.
The Hamiltonian (11.25) for Ho is solved to obtain the eigenvalues and eigenstates of the
system. This solution is approximate, since it omits the effects of the potential V. The
approximation of neglecting V and using only the eigenstates of Ho is the pairing theory of
Bogoliubov. The Hamiltonian is rather easy to diagonalize. Each wave vector state is treated
independently, and each of these is just the problem encountered in Problem 3 in Chapter 1.
Introduce a set of harmonic oscillator coordinates Qk and P k , and the Hamiltonian is an
example where the interactions merely shift the eigenfrequency of the oscillator:
(11.26)
(11.27)
Ho = 12:k {PkPk + QkQdE~ + 2PoEk V(k)]}

(11.28)
I
Ho = 12:k {PkPk + n~QkQd

(11.29)
The first two tenus in Ho come from ctck , while the last tenu comes from the second tenu in
the Hamiltonian Ho. The tenus multiplying QkQ-k are combined into the square of an
effective frequency nk- The bosons behave as a hanuonic oscillator system, for each wave
vector k, at the new frequency Ok' The Hamiltonian is diagonalized by defining a new set of
creation and destruction operators
(11.30)
(11.31)
(11.32)
In the pairing theory, the energy Ok is that of the quasiparticles, and (J(k are the operators
which describe the destruction of these quasiparticles.
The equation for the quasiparticle energy Ok = {Ek[Ek + 2PoV(k)]}I/2 makes no sense
whatever, since the Fourier transform V(k) does not exist. It should be replaced by a T matrix
or reaction matrix, and the latter is correct. According to Eq. (4.31), for each angular
momentum state lone can replace the potential by the on-shell reaction matrix
(11.33)
where J.1 = ml2 is the reduced mass for the scattering of two identical particles of mass m.
The reaction matrix has the important feature that it goes to a constant in the limit where
k ---+ O. The phase shifts behave in this limit as
lim o/(k) = m/n - a~/+I (11.34)

k-+O
lim tan(o/) = -a~/ (11.35)

k-+O k
where m/ is an integer which denotes the number of bound states in that scattering channel.
The quantity a is called a scattering length.
The largest term is from the s-wave channel, in which Ro(O, 0) = 2nalJ.1, while
R/(O, 0) = 0 for I > O. Only the s-wave needs to be included in the dispersion relation of the
excitations as k ---+ 0
V(k) ---+ 2lta = 4na (11.36)

J.1 m
(11.37)
(11.38)
The scattering length a is positive for repulsive potentials. For a hard sphere potential of
radius 0', then 0' = a.
In the earlier days of helium theory, the helium-helium interaction was approximated as a
hard core at the radius a = 0' = 0.265 nm of the Lennard-Jones potential. All the other
parameters are known since Po is the 4He density. The velocity of the excitation may be
calculated:
a = 0.265nm (11.39)
m = 6.648 x 10-27 kg (11.40)
Po = 2.2 x 1022 atoms/cm3 (11.41)
C = 130mls (11.42)
The predicted velocity is close to the speed of ordinary sound in helium, which is
Cs = 220m/s. This agreement was viewed as a great success ofthe pairing theory, since the
collective excitations of the liquid at long wavelength must certainly be the longitudinal sound
waves.
Unfortunately, this agreement is entirely superfluous. The interaction potential between
two 4He particles has attractive regions, and the phase shifts for this potential show there is a
low-energy scattering resonance. The exact value of the scattering length a is uncertain
because it is very sensitive to the potential. The expected values are more in the range of 500"
rather than 0". A value of a '" 500" will make the calculated speed of sound be 4 times larger
than the experiment. Of course, the sign of a could be negative, in which case the predicted
frequencies are complex, and the vibrations are unstable. A possible way out of this difficulty
is to do the theory more carefully. This was done by Brown and Coopersmith (1969), who
found even worse agreement with experiment. Their calculation represented all direct and
exchange scattering by reaction matrices, and they self-consistently found the condensate
fraction fo and the chemical potential. They obtained a speed of sound about 10 times that of
the experiment and also predicted a roton energy about 10 times too big. Indeed, their
theoretical excitation spectrum of 4He is similar to the actual one but scaled up by a factor of
10. This calculation was done carefully and correctly. Its lack of success is caused by the
failure of the pairing approximation. Indeed, the history of the pairing theory is that each time
the theoretical calculations were done better, the answers got worse. The conclusion is that the
pairing theory is bad and should be discarded as a way of calculating the excitation spectrum
of 4He. Some of the concepts are retained, in particular that of the condensate.
11.1.3. Off-Diagonal Long-Range Order

The modem theory of superfluidity in 4He treats the liquid as a highly correlated and
strongly interacting system. The pairing theory is completely abandoned, as is the repre-
sentation of operators in plane-wave basis states. The one concept carried over is the
condensate, or zero-momentum state, in which there is a nonzero fraction of the particles in
the superfluid. There is an obvious problem of trying to decide the definition of "zero-
momentum state" in a basis set which is not plane waves. It is really part of a larger problem
of trying to decide what is meant by superfluidity and Bose-Einstein condensation in a
system which is strongly interacting and highly correlated. The whole concept of Bose-
Einstein condensation is based on particles occupying zero-momentum states, or not occu-
pying them, and another basis set is needed for a more general definition. The method of
doing this was introduced by Penrose (1951), and his ideas are described. Yang (1962)
suggested the name of off-diagonal long-range order (ODLRO) for the type of ordering
introduced by Penrose and also discussed by Penrose and Onsager (1956).
ODLRO was introduced as the type of ordering for superfluids, such as Bose-Einstein
condensation in 4He, and electron pairs in superconductors. It is distinguished from diagonal
long-range order (DLRO), which is the usual ordering one finds, for example, in crystalline
solids. When 4He atoms arrange themselves into a solid, as happens under pressure, they are
exhibiting DLRo. But when they go superfluid, they exhibit ODLRO. The distinction
between these two processes stems from the different behavior in the density matrix for each
type of ordering. These differences will be explained. Further discussion is given by Kohn
and Sherrington (1970).
The system has N '" 1023 identical bosons, which are spinless. Their ground state is
described by a many-particle wave function 'I'o(r,. r2."" rN)' The actual method of
calculating this wave function, or at least some of its properties, is given in the following
sections. The present discussion only needs to assume that the wave function exists. The
subscript zero on '1'0 indicates it is for the ground state. Note that there is no assumption that
each atom is in the ground state (i.e., k = 0) but that the system is in the ground state. When
the particles are strongly interacting and highly correlated, they will spend little time with
k = 0 and will fluctuate between many momentum states. The ground state is the lowest
possible energy state for the whole liquid. It is not a static or rigid structure, since each atom
fluctuates with zero-point energy. The present estimates for 4He at T = 1 K give the average
kinetic energy per particle as 15 K and the average potential energy as -22 K, so the average
binding energy is 7 K. The large value of average kinetic energy shows the large amount of
zero-point motion in the fluid, even at low temperature, which comes from the quantum nature
of the fluid. The classical estimate (K.E.) = 3kB T /2 is obviously inaccurate. The ground state
wave function 'I'o(r,. r2."" rN) for 4He describes a system with a large amount of zero-
point motion.
The square of the wave function gives the probability density for finding particles at
positions rj in the system and is called the diagonal density matrix:
(11.43)
The subscript N indicates that it applies to N particles. PN is normalized so that the integral
over all coordinates gives unity:
(11.44)
The one-particle density matrix is obtained from PN by integrating over all but one coordi-
nate:
p, (r,) = Jd3r2d3r3 ... d3rNPN(r,. r2 ....• rN) (11.45)
1= J r, p, (r,)
d3 (11.46)
The notation is a bit confusing, since p, (r,) is not the probability in a one-particle system but
rather the probability of one particle being at r, in the N-particle system. Since all points are
equivalent, Pieri) is really independent of position-neglecting edges-so p, = l/v, where v
is the volume of the system.
Another useful quantity is the two-particle density matrix P2(rl. r2) which is obtained
from PN by integrating over all but two coordinates:
P2(rl. r2) = Jd3r3d3r4 ... d3rNPN(rl' r2 •...• rN) (11.47)
-;1 = J r2P2(rl. r2)

d3 (11.48)
Now P2(rl' r2) is the probability that any boson is at r2 if there is one at rl' In a homogeneous
system it must depend only on the difference rl - r 2. In fact, it must be proportional to the
pair distribution function g(r), and this relationship is
1
P2(rl. r2) = 2"
v
g(rl - r2) (11.49)
which follows from the fact that g(r) is normalized to be unity at large r.
Crystalline solids have DLRO, which is indicated by structure in g(r). Since the atoms
are regularly spaced, g(r) has large values where the atoms are located and is zero otherwise.
Thermal vibrations smear out the answer.
Penrose (1951) used the idea of a general density matrix which is defined as the product
of two wave functions with different coordinates:
PN(rl. r2.···, r N; r'1' r~ •...• r'N) (11.50)
= 'P(j(rl' r2.···' rN)'PO(r;.~ ....• r'N) (11.51)
This quantity is denoted by P. where the tilde is to distinguish it from the diagonal density
matrix introduced earlier. Of course, they are identical if the two sets of coordinates in Pare
set equal: PN(rl' r2."" rN; rl' r2"'" rN) = PN(rl. r2."·' rN)' The concept ofODLRO is
contained in the function p(rl' r;) obtained from PN when all but one set of coordinates are
set equal and averaged over
PI(rl. r'1) = Jd3r2d3r3d3r4" ·d3rNPN(rl. r2.···, rN; r'1. r2.···, rN)

1
PI (rl. rl) = PI (rl) = - (11.52)
v
where PI(rl. r'1) becomes the diagonal density matrix PI when rl = r2' In a liquid, the
dependence on rl and r2 can only be their difference, since there is no absolute frame of
reference. It is convenient to define the quantity
R(rl - r;) = VPI (rl. r;) (11.53)
R(O) = 1 (11.54)
which is normalized to unity at r = O. R(r) is the function which is important in under-
standing ODLRO.
Some insight into p(rl' r;) is obtained by reconsidering the techniques used in the
weakly interacting systems. There a one-particle state function is defined in the plane-wave
representation as
(11.55)
(11.56)
The number of particles in state k is
nk = (ctck ) = ~ Jd3rd3r'e- ik .(r-r')(Cl>t(r)Cl>(r'») (11.57)
nk is found to be the Fourier transfonn of the quantity (Cl> t (r)Cl>(r'»), which must be a function
of r - r'. Recall what this average means. Take a particle and find its wave function at two
different points r and r'. This product is averaged, which must be taken over the other
particles and their positions. This procedure is exactly the one which was used to obtain
PI (rl' rD, except now it is recognized from the outset that the particle is part of a many-
particle system. The procedure is to take a product wave function with one particle at two
points rl and r'1 and average it over the other particles and their positions. The quantity
PI (rl' ri) or its equivalent R(rl - r2), is the many-body definition of the quantity
(Cl>t (r)Cl>(r'»). It is also important to appreciate that R(r) is quite a different function from the
pair distribution function g(r) and the two are not related. The quantity nk is also related to the
Wigner distribution function where nk(R, T) = f dooi(k, 00; R, T).
The quantity nk is the number of particles, on the average, in the momentum state k. It is
an important quantity in Bose-Einstein condensation, since one expects that one momentum
state ko will have a macroscopic occupation such that nko =!oN, where 10 is still the fraction
of particles in the condensate. This fraction is of order unity, rather than O(l/N). Nonnally
ko = 0, which applies when the fluid is at rest. There are circumstances when another state
has the macroscopic occupation, e.g., when the fluid is flowing at a unifonn rate or when it is
rotating. For k =1= ko,
nk = J~: nB(E)A(k, E) (11.58)
where A(k, E) is the spectral function for the boson particles.

The procedure for finding nk is the same as in the free-particle case (11.57). There one
took the Fourier transfonn of (Cl>t(r)Cl>(r'»), while now one takes the Fourier transfonn of the
equivalent quantity R(r - r'). One difficulty with this procedure is that R(r) may not possess a
Fourier transfonn. For example, how does it behave as r ~ oo? For fluids at rest, so that the
macroscopic occupation is in the zero-momentum state, R(r) goes to a constant as r ~ 00.
Call this constant R( 00). The Fourier transfonn is
J
nk = Po d 3 rR(r)eik • r (11.59)
J
= poR(00)(2n)3()3(r) + Po d 3r[R(r) - R(oo)]eik • r (11.60)
= NO()k=o + Po Jd r[R(r) - R(oo)]e

3 ik • r (11.61)
The density is Po = N lv, No = ioN. There is a delta function tenn at k = 0 whose amplitude
is poRe 00). The quantity R( 00) = 10 is the fraction of particles in the zero-momentum state.
This fraction is the "order" which exists in the off-diagonal density matrix R(r). For k =1= 0,
the momentum distribution of particles nk is found from the Fourier transfonn of [R(r) -10],
Figure 11.3 shows the function R(r) calculated for liquid 4He by McMillan (1965) using
Monte Carlo techniques. He used the Lennard-Jones potential between helium atoms and
found 10 = 0.11. His fraction is similar to that obtained earlier by Penrose and Onsager, who
gotfo = 0.08 for a gas of hard spheres. The latter estimate is also based on a Monte Carlo
calculation. It appears, unfortunately, that reliable results can be obtained in this field only by
-
a:
FIGURE 11.3 Single-particle density marix as a function of separation. The solid line is the Monte Carlo calculation
of McMillan (1965), which as asymptotically approaches a density fraction offo = 0.11. The dashed curve is the
Gaussian approximation.
extensive computer calculations. There have been many calculations of these quantities; e.g.,
see Francis et al. (1970). Figure 11.3 shows that R(r) starts at unity and falls smoothly to its
asymptotic value, which it reaches at about r = 0.4 nm. This distance is rather short, since it
is only 1.5 atomic diameters.
The quantity fo = R( 00) is the fraction of time a particle spends in the condensate.
Alternately, it is also the fraction of particles in this state at anyone time. However, in
calculating this quantity, it has not been assumed that the system is superftuid. The value of
R( 00) is found by solving the dynamical properties of normal fluids, which may seem
paradoxical and is probably best understood in the next section which gives more details of
the method. Normal fluids do not have a condensate and havefo = R(oo) = O. However, the
point is that most methods of calculatingfo are based on the properties of normal fluids. For
example, the estimate of Penrose and Onsager came from the equation of state of a classical
hard sphere gas, which is certainly not a superftuid system.
11.1.4. Correlated Basis Functions

Correlated basis functions (CBFs) are the type of wave function most often employed in
the study of the ground state properties of 4He. They have the form
(11.62)
where L~ is a normalization constant which is specified by satisfying the integral (11.44).

They were used by Bijl (1940), Dingle (1949), and Jastrow (1955) and are sometimes named
after these authors; different writers preferring various combinations of their names. The
CBFs have several advantages which make them desirable for a description of 4He. The most
important is that the wave function possesses the necessary symmetry of being symmetric
under the exchange of any two coordinates. This symmetry is automatically satisfied by the
summation of pairwise correlations. Another advantage is that 'Po has a simple form for
which one needs to determine only one function u(r), which is usually found variationally by
minimizing the ground state energy.
A third advantage of CBFs is that the diagonal density matrix has a mathematical form
which is identical to another problem which has been studied extensively-the classical
fluid. For a classical fluid, the kinetic energy terms are irrelevant for determining particle
correlations, and only the potential energy veri - r) between particles i andj is important.
The density matrices for CBFs and the classical fluid are
CBF: pN = 1'1'012 = L~ exp[-2~.U(ri

'>j
- r)] (11.63)
Classical: PN(r" r2,"" rN) = Qn exp[-~L.V(ri

'>j
- r)] (11.64)
These have the same mathematical form when setting 2u(r) = ~V(r). The classical problem
has been extensively studied, and accurate calculational techniques have been developed
which work well for liquids of neutral atoms such as the rare gases. The most successful
methods are based on the Percus-Yevick equation (see Percus, 1964). The availability of this
successful computational technology is one reason for the popularity and success of CBFs.
One important quantity is the off-diagonal density matrix R(r), which is defined in terms
ofCBFs as
R(r, - r;) = vL~ Jd3r2d3r3 ... d 3rN
x exp [ - j t [u(r, - rj ) + u(r; - r)] - 2i~2u(ri - r)] (11.65)
Let us determine the classical analog of this expression. That is, regard u(ri - r) as the
effective "potential" ~V(ri - rj)/2 between particles and see where that leads us. Obviously
there is a system of N - 1 particles which mutually interact with each other with the
"potential" 2u(r). There is a particle at r, which interacts with these N - 1 particles with the
different potential u(r). Another particle at r; also interacts with the N - 1 particles with the
same relative potential u(r), so that it is identical to the particle at r,. Finally, the particles at
r, and r', have no mutual interaction. This description is of a system of two impurity particles,
at r, and r'" in a system of N - 1 other particles. The impurities are different, since they have
a different interaction with the N - 1 particles and no interaction between themselves. This
classical system can be solved by using the Percus-Yevick equations for binary mixtures,
where one constituent is very dilute. This technique was used by Francis et al., (1970) for
their investigation of superfiuid 4He. It is another example of where classical equations are
solved to determine the properties of the superfiuid.
The ground state properties which are calculated with CBFs include the condensate
fraction 10, the momentum distribution nk' R(r), g(r), and S(k). Another property is the
ground state energy per atom, which is usually presented as the separate contributions from
potential energy (P.E.) and kinetic energy (K.E.). The potential energy per atom is given by
(11.66)
The density matrix PN(r, ... rN) is unchanged when any pair of coordinates r i and rj are
interchanged. With this fact, one can show that the average value of each potential energy
term Veri - r) is the same as any other term VCr; - rj). The average potential of any particle
is the same as any other. The summation over i andj gives N(N - 1)/2, which is the number
of pairs, and all pairs contribute equally:
The integration over all variables except rl and r2 produces the reduced density matrix
P2(rl, r2) = g(rl - r2)/v2. Then changing to relative coordinates r = (rl - r2), R =
(rl + r2)/2, there is no dependence on R, so its integral yields v. The final answer is
(P.E.) = Po
N 2
Jd rV(r)g(r)
3 (11.67)
The average potential energy is obtained from the pair distribution function. This result is
obvious for a system with pairwise interaction between particles. It does not depend on CBFs
but applies to any liquid and is an exact identity. The CBFs are used to calculate g(r), which is
t
then used in the evaluation of the average potential energy per particle. The factor of occurs
because each pair interaction is shared between two atoms.
The average kinetic energy per particle is obtained from the fundamental definition
(11.68)
The result for each term j in the summation is identical, so the sum gives N times the result
for one term:
(11.69)
The evaluation of this quantity depends sensitively on the use of CBFs. The gradient of the
wave function produces the gradient acting upon u(r):
This expression is squared in the evaluation of the average kinetic energy. At first the
evaluation appears to involve three-particle correlations. One can integrate all but the three
coordinates rl' rj. rm , which produces the reduced density matrix P3(rl' rj • rm). It would be
difficult to evaluate P3 accurately, since the knowledge of three-particle correlations is
imperfect. However, McMillan (1965) suggested a method of avoiding this problem and
expressing the result in terms of only two-particle correlations. His method is to write one
derivative as acting upon the density matrix itself:
N
PN LV1u(rl - r) = -tVIPN(r1• r2.···. rN)
j=2
(11. 70)
All coordinates were integrated over except rl and rj . Eliminate the summation overj since
each term contributes identically:
(11.71)
Again changing integration variables to relative coordinates, the final form of the result is
(11.72)
The average kinetic energy is expressed as an integral over the derivative of u(r) and g(r).
Neither quantity has an angular dependence, so the gradient acts only upon the radial variable.
This form is used for the evaluation of the average kinetic energy per particle. The numerical
results obtained by different investigators vary somewhat, depending on the details. Generally
the results show that the average potential energy per particle has the magnitude of 20-22 K,
while the average kinetic energy per particle is 14-15 K, and the net binding energy per
particle is 6-7 K. These values depend on the particle density.
So far the form of u(r) in CBFs has not been specified. That will be done now. Usually
the contribution of u(r) is divided into a short-range component and a long-range component.
The short-range component is the most straightforward and is discussed first. One expects that
the atoms do not interpenetrate. There must be a term in u(r) which keeps them apart, which
is done by having a term in u(r) which becomes very large as r ~ 0, which makes
exp( -u) ~ O. The form of this term can be deduced from Schrodinger's equation and the
repulsive part of the interparticle potential. Take a two-particle system, so that u(r) describes
only the effects of correlation between the motion of two particles. In center-of-mass coor-
dinates, the relative motion of the two helium atoms is described by the following CBF wave
functions:
\
_ ",2 [~+ ~~
2J.l dr2 rdr
_1(1 +I)J +
r2
V(r)}e-u(r) = Ee-u(r) (11.73)
e-u(7)\_Ii} [_ d 2u + (dU)
2
_~du _1(1+ 1)] + VCr) -E} = 0 (11.74)
2J.l dr2 dr rdr r2
Ifu(r) is diverging sharply as r ~ 0, the dominant term among all the derivatives is (du/dri.
For example, if u <X r- n, then (du/ dri <X r- 2n - 2, while the second derivative diverges only as
r- n - 2 • The two most divergent terms must cancel, where one is from the kinetic energy and
the other from the potential energy:
= _ [2J.l VCr) J
I/2
du (11.75)
dr h2
Students of physics should recognize this result and derivation. It is just the Wentzel-
Kramers-Brillouin-Jeffreys (WKBJ) wave function for a particle penetrating into a repulsive
potential: as a specific example, in the Lennard-Jones potential the repulsive part is 4E( cr jr)12.
In this case the short-range part of the CBF must be
du = _(8~E)1/2(~)6 (11.76)
dr li2 r
u(r)= - ( 8~E)1/2
li2
cr6
-
5r5
= (a)5
-
r
(11.77)
a = cr (8~Ecr2) 1/10
(11.78)
25li2
For the potential between two 4He atoms, E = 1.484 X 10-22 J, cr = 0.22648 nm, and the
numerical result is a = 1.01cr. The constant a is nearly identical to the hard sphere diameter
cr. McMillan (1965) used the Lennard-Jones potential for his variational calculation and
included no long-range potential. He tried correlations of the form u(r) = (ajrl, where a and
n were both variational parameters. He used cr = 0.2556 nm so a = 1.01cr = 0.259nm is our
prediction. His minimum energy was found with n = 5 and a = 0.261 nm. These values are
nearly identical to the results using the WKBJ between two particles. The short-range part of
the CBF can be obtained from the wave function for a particle penetrating into a repulsive
potential. This part of u(r) is determined once the potential is specified.
The long-range part of u(r) is more subtle. There is a term which falls off as u(r) ()( r- 2
as r ---+ 00. This contribution arises from the zero-point motion of the phonons. Bogoliubov
and Zubarev (1955) showed, in the pairing theory, that the phonons lead to long-range
correlations between the motion of particles in the liquid. Reatto and Chester (1967) showed
that these fluctuations lead to long-range contributions to u(r). They recommended
u(r) = uir) + u/(r), where the short-range result uir) = (ajr)5, and the long-range part is
me 1
(11.79)
u/(r) = 21t2pO r2 + r~
where the cut-off is rc ~ 0.20 nm. The parameter e is the speed of sound. The function u/(r)
goes to a constant as r ---+ 0 and only becomes proportional to r- 2 at large distances. An
inverse square law is a rather slow fall off with distance and is the long-range behavior.
The same u(r) was used by Francis et al. (1970) in the numerical study referred to
previously. The inclusion of the long-range correlations will change R(r) at large r and nk at
small k. In particular, the distribution nk diverges as k ---+ 0 as a result of the long-range
correlations. The divergence is shown in Fig. 11.4 where the results for nk of Francis et al. are
shown for both the short-range correlations only (SR) and the short plus long-range corre-
lations (LR). Their short-range result is nearly identical to that of McMillan. This agreement
is expected, since they used the same interparticle potential but a different method of
obtaining results, i.e., the Percus-Yevick equations for mixtures. The long-range correlations
have a dramatic effect upon the theoretical values of nk' This low k divergence is found in the
experiments, as discussed by Griffin (1985), and by Svensson and Sears (1986).
Amid all these equations, it is worthwhile to gain a physical picture of the particle
motions on the microscopic level. The average potential energy of 20 to 22 K per atom can be
achieved by having four neighbors at the minimum well depth of 10K, so the average
potential energy per particle is 20 K. However, an inspection of g(r) shows that it peaks at
larger values of r, and the neighboring shell of atoms is farther away than the minimum well
depth at 0.3 nm. The potential energy comes from having about eight neighbors, each with an
0.6
~ 0.4
c
0.2
o •
o
kcr
FIGURE 11.4 Theoretical calculations of the momentum distribution of atoms in liquid 4He. The points are due to
McMillan (1965), and the solid line is from Francis et al., who used the Percus-Yevick equation both with and without
long-range order. Source: Francis et aI., (1970) (used with permission).
average interaction energy of 5 K, which gives the per particle average of 20 K. In a classical
fluid such as argon, g(r) peaks at the maximum well depth, so the first neighbor atoms are
sitting at the distance of the maximum potential energy. Helium does not behave this way,
because the quantum nature of the motion prevents that much localization of the particles, so
the first neighbors are farther away on the average.
The average kinetic energy is the more interesting number. The estimates of this quantity
range from 10 to 15 K. What kinds of particle motion give this large number? There are two
microscopic pictures which come to mind. The first has the particle motion as part of the zero-
point motion of the long-wavelength phonons. Here the particles keep their nearest neighbor
positions rather fixed, but the system executes long-range fluctuations. This Jell-O model has
the system wiggling like an elastic medium. The amount of energy in this zero-point motion
can be estimated as
(Z.P.E.) = -2 d3 q J
h -3(cq)0(qc - q) (11.80)
Po (2n)
,. . ., hcq~
(11.81)
,...., l6n2 po
Here c is the speed of sound and Po is the particle density. The cutoff wave vector qc is the
maximum value of the excitation spectrum, which is considered to be a sound wave, i.e., over
which ffiq = cq is an accurate approximation. Francis et al., estimate it to be qc = 0.5
A-I. With this value and the other known parameters for 4He, one can estimate that (Z.P.E.) =
0.33 K. The result 0.3 K is a negligible part of the total kinetic energy per particle. This
low estimate is confirmed by the calculations of Francis et al., who found that the long-range
correlations (and fluctuations) changed the average energy per particle only by 0.5 K.
The average kinetic energy per particle does not come from the long-range fluctuations.
Instead, it seems to come from the short-range motion. The first peak in g(r) is at 4 A, so the
atom sits in its own space in the liquid, which is roughly a sphere with a radius of about 2 A.
A particle in a spherical box of radius a has a kinetic energy of fhc 2 /2ma 2 • Using a = 2 A
and the mass of helium yields a kinetic energy of 15 K, which is just the right magnitude. The
kinetic energy comes from the short-range fluctuations of the particle bouncing around inside
its own small space in the liquid. In the ground state of the liquid, at very low temperatures,
the individual particles are moving rapidly with this motion.
It is also useful to have a simple approximation for nk and its Fourier transform R(r).
Puff and Tenn (1970) observed that neutron scattering experiments, which will be described
later, showed that the particle distribution was Gaussian. They suggested the form
(11.82)
One advantage of the Gaussian distribution is that one can calculate quantities easily. For
example, the average kinetic energy per particle is simply
From their fits to the neutron scattering data, they deduced that this average kinetic energy is
about 15 K. It is their result we have mentioned several times previously in this chapter. This
form has a certain appeal. The classical distribution is also Gaussian, except then w = kBT,
and the average kinetic energy is (3/2)kB T. In 4He there is still a Gaussian distribution, but
the width w is much larger than predicted by the temperature; one estimates that w = 11 K in
the limit where T -+ o.
The off-diagonal density matrix R(r) is the Fourier transform of nk. When nk is Gaus-
sian, so is R(r):
(11.83)
(11.84)
This Gaussian is the dashed curve in Fig. 11.3, compared with the calculations of McMillan.
The fit is reasonable, considering that the parameters were taken from the estimate of Puff and
Tenn that w = 11 K, which gives b = 1.48 A.
It is likely that neither nk (for k # 0) nor R(r) are exact Gaussians in the superfluid state
and that the fit to this functional form is only an approximation. Nevertheless, it is useful to
have approximate analytical forms, and the Gaussian fits best. Puff and Tenn, in fitting
to
the neutron data to a Gaussian plus a delta function at k = 0, estimated that = 0.06 at
T = 1.27 K. Their estimate is in agreement with later measurements and extrapolates to a
value of10 = 10.8% in the limit where T -+ 0 if one uses the Bose-Einstein extrapolation
10(0)[1 - (T /Tc)3/2].
11.1.5. Experiments on ~
Feynman first suggested that neutron scattering would be an excellent means of inves-
tigating the structure of liquid 4He. His remarks were aimed at a measurement of the exci-
tation spectrum. These measurements by neutron scattering have been very successful and are
described in the next section. However, in the past three decades another series of neutron
scattering experiments have been performed on liquid 4He in order to measure the momentum
distribution of particles nk' This latter experiment involves using higher-energy neutrons and
analyzing the data in a fashion different from that used to find the excitation energy. The
primary objective of this effort is a direct measurement of the condensate fractionfo.
Hohenberg and Platzman (1966) made the suggestion that inelastic scattering by very
energetic neutrons (say 0.1 eV) would provide a measurement ofnk and hence the condensate
fraction 10, which would show up as a delta function on the nk distribution. Of course, the
delta function would be broadened by experimental resolution, but should still be apparent if
the estimates offo = 0.1 were correct. The reason for using very energetic neutrons is that the
sudden approximation becomes valid when the neutron energy is much higher than the
kinetic or potential energy of the 4He atom. In the sudden approximation, one has a direct
measurement of nk'
In the neutron scattering experiment, the neutrons with initial wave vector k i and initial
energy Ei = Ii?!if 12m are directed toward the scattering chamber. Some neutrons are scattered
and leave the sample with a final wave vector kf = k i + Q, where Q is transferred to the
liquid. Similarly, the final energy is Ef = (k i - Q)2/2m = Ei - 0), where 0) is the energy
transferred to the liquid. The scattering cross section for 4He can be expressed in terms of the
dynamic liquid structure factor SeQ, 0)) as
d2 (j kf
dOdO) = k- (joS(Q, 0))
I
(11.85)
where (jo is a cross section for neutron scattering from a single alpha particle. Equation
(11.85) is a general result which is always valid for neutron scattering. Generally a calculation
of SeQ, 0)) is quite complicated. However, in the sudden approximation, it can be approxi-
mated by the expression
S(Q,O)) = 2n I~nk[O(O) - Q'-

Po (2n)3 2m
- Q.
m
k) - 0(0) + Q'- + Q'
2m m
k)] (11.86)
where nk is the momentum distribution of particles in the ground state of the liquid. The
outline is to first discuss the implications of this result; the derivation is provided later. The
momentum distribution nk is written as the condensate term NOok=o plus the other term which
is called nk' These two are used in (11.86), and the wave vector integrals are done as far as
possible (EQ = Q'- 12m)
nk = NOok=o + nk (11.87)
S(Q,O)) = SeQ, 0)) - seQ, -0)) (11.88)
S(Q, 0)) = 2nfoo(0) - EQ) + 2n I

Po (2n)
d3k3 nko(O) _ Q'- _ Q.
2m m
k)
-
S(Q,O)) = 2nfoo(0) - EQ) + -2-
1
npo
Joo ~dknk
0
- II (
-1
dvo 0) - -Q'- -QkV)
2m
-
m
(11.89)
The angular integral dv = d( cos 9) over the delta function equals m I kQ if k>
(mIQ)lco - f:QI == kQ(co), so the final result is (co> 0)
(11.90)
Equation (11.90) predicts that S(Q,co) for co>O has a delta function contribution at the
energy loss co = f:Q = (j2 12m. The delta function corresponds to the neutron knocking
particles out of the condensate, and the fraction of the spectral strength in this process isfo. Of
course, this delta function is broadened by the resolution of the measuring apparatus but
should still be observable. The second term in (11.90) gives the scattering from the particles
not in the condensate. This term is left as an integral. However, nk may be obtained from data
with small numerical scatter by taking the following derivative (co =1= f:Q):
This equation provides a direct method of measuring nk' As pointed out by Puff and Tenn, the
Gaussian distribution in (11.82) fits the scattering data well. For this distribution, the integral
can be evaluated analytically
(11.91)
(11.92)
Equation (11.91) predicts that the spectral function S(Q, co) is the difference of two Gaus-
sians, peaked at co = ±f:Q and with a width given by 2,Jf:Qw, There is also the delta function
at co = ±f:Q from the scattering from the condensate. The results appear as a single Gaussian
whenever f:Q > co, which is the usual experimental case.
There have been numerous measurements of nk, including those of Harling (1971),
Rodriquez et ai., (1974), Svensson et ai., (1981), and Svensson and Sears (1986). The data of
Woods and Sears (1977) is shown in Fig. 10.5. The solid line in Fig. 1 1.5 (a) is a fit to their
points taken at T = 4.2 K when the liquid 4He is not superfluid. This line is a remarkably
good fit to a Gaussian. The other set of points are taken at T = 1.1 K in the superfluid state.
There is an obvious increase in the distribution nk at small values of k, as shown by the
difference spectra in Fig. 11.5(b). The increase at k ~ 0 has the width of their resolution
function and is interpreted as being caused by the condensate fraction 10. They estimate
fo = 6.9% ± 0.8% from this data. If this value offo is extrapolated to zero temperature by
using the Bose-Einstein formula, one finds that fo(T = 0) = 10.8% ± 1.3%. The value
10 = 0.11 is in good agreement with the theoretical estimates, with the earlier results of
Harling, and with the analysis of Puff and Tenn. However, there is little justification for using
the Bose-Einstein formula for extrapolating fo(T) to zero temperature except expediency,
since no other formula is available. It also appears from the data of Woods and Sears that the
values of nk in the superfluid state are not as good a fit to a Gaussian as they were in the
normal fluid. This feature is not understood.
Equation (11.86) for the sudden approximation is now derived. The first step begins
with the definition of the spectral function S(Q, co). As in the case of the electron gas in
Chapter 5, the spectral function is derived from the density-density correlation function. If
(0 )
500
( b)
. E
c:
•'Q 0.4
-CONDENSATE PEAK
.."
N
~
c:
0.2
~
~
c
0.0 ~-- 00000000000000000000

o 000
0 40
FIGURE 11.5 (a) Neutron scattering results for the momentum distribution in liquid 4He at T = 1.1 and 4.2 K as a
function of p2 (b) The difference between the two distributions in part (a) shows a peak at small wave vector whose
width equals the experimental resolution. The peak is interpreted as the condensate fraction. Source: Woods and Sears
(1977) (used with permission).
this correlation function is called X(Q, iwn ) in the Matsubara representation, then S(Q, w) is
-2 times the imaginary part of its retarded function:
X(Q, iwn ) .
= - N1 f~0 d1:e,ro,t(Tt P(Q, 1:)p(-Q, O)} (11.93)
S(Q, w) = -2 Im[Xret(Q, w)] (11.94)

According to the result of Problem 16 in Chapter 3, S(Q, w) can written as
S(Q, w) I
= -(1- e-~ro) Joo dteiro((p(Q, t)p(-Q, O)} (11.95)
N -00
The experiments are done with conditions of temperature and energy transfer Iiw such that
~w » 1, and the second term in the parentheses exp( - ~w) can be neglected. The space
representation is used for the density operator p(Q), which brings us to the expression
p(Q) = L:>iQ'Rj (11.96)
j
S(Q, w) = ~ I:Joo dteiro((eiQ'R/t)e-iQ'R;(O)} (11.97)

N ij -00
The correlation function is evaluated by determining how the particle motion Rit) proceeds
in time. The two exponents Q' Ri and Q' Rit) cannot be combined, since they do not
commute at different times. This point will be shown later.
The sudden approximation is obtained by solving the correlation function for small
values of time t by expanding Rit) about point t = O. The physical argument is that the
scattering by energetic neutrons happens rapidly, so that only the short time response of the
systems is applicable. Near t = 0, the coefficients in the Taylor series are found from the
(11.98)
(11.99)
(11.100)
The next step is to separate the several terms in the exponent into the product of several
exponents. The Feynman theorem is used: exp(A + B) = exp(A) exp(B) exp( C) with
B = iQ' Rj , A = i(Q/m)' (tPj + t 2 F./2), and C = -[A, B]/2 = -itEQ. The next step is to
separate the factor itQ· P)m from it~Q' Fj /(2m). They do not commute, but such corrections
terms contribute on the order 0(t3 ), which are being neglected. The short time response is
written as
The sudden approximation is achieved by making two important approximations to this

expression. The first is to take only the term in the series with i = j. This choice is a valid
approximation when Q is large, since then the terms with i =1= j average to a small contri-
bution. The second approximation is to neglect the term iPQ . F)2m. The quantity Fj is the
force on the particle at R), which must average to zero. If this force is evaluated by a cumulant
expansion, for example, the average (F)' Q) is zero, although the average (F) . Q)2) is not.
This force contributes a term in the time expansion on the order of 0(t4 ) rather than 0(t2 ). It
is neglected.
In the sudden approximation of high Q and high m, the remaining expression to evaluate
is:
SeQ, m) = ~ L JOO dteit(m-EQ) (eitQ'Pj/m)

N) -00
(11.101)
(eitQ'Pj/m) = Jd3R 1 •• .d3RN'P~(Rl·· .RN)eitQ'Pj/m'Po(Rl·· ·RN)
The quantity exp(itQ' P}im) operates on the wave function to the right and displaces the
position variable R) by the increment tQ/m:
eitQ'Pj/m'Po(Rl, R 2 ,···, R N) = 'PO(Rl' R 2 ,.··, R) + tQ/m, ... ,~)
The next step is to do all the position integrals except d 3R). The multiple integrals just
produce the off-diagonal density matrix peRi' R) + tQ/m). Since the off-diagonal density
matrix is just a function of the difference of its arguments, the expression gives
peRi' R) + tQ/m) = R(tQ/m)/v:
SeQ, m) = -1 L Joo dteit(m-EQ)R(tQ/m) (11.102)

N) -00
seQ, m) = J~oo dteit(m-EQ)R(tQ/m) (11.103)
The summation over j gives N, since all terms in the summation are alike. Earlier it was
observed that R(r) was approximately a Gaussian function of r. In this case, the integrand of
the time integral is also a Gaussian. A Gaussian is found since the exponent is evaluated to
order O(?).
To derive to the standard formula for the sudden approximation, remember that nk is the
Fourier transform of R(r):
- -
R( r ) - d3 k
1 --nke J
-ik' r
(11.104)
Po (2ni
S(Q,m)=~J d3k3 nkJoo dtexp[it(m-cQ-
Po (21t) m
Q'k)] (11.105)
k)
-00
= 21t Jd 3 k nko(m _ cQ _ Q' (11.106)

Po (21t)3 m
which is the first term in (11.86). The second term in (11.86) comes from realizing that
SeQ, m) is antisymmetric in m.
Two important assumptions are made in the derivation of the sudden approximation. The
first is that Q is large, and the second is that m is large so that the Fourier transform involves
only small values of t. The experiments are done such that both conditions are well met. Then
the approximation is accurate and the experiments determine nk'
The sudden approximation has the appearance of treating the particles in the liquid as
being free. The same result could be derived by assuming there was a gas of free particles
with a distribution nk' When the neutron strikes a particle of momentum k and kinetic energy
Ek = Jc2 12m, the particle is changed to wave vector k + Q and kinetic energy Ek+q =
(k + Qi 12m. Energy conservation demands that the energy transfer be (0 + Ek = Ek+Q' The
probability of having a momentum-energy transfer of (Q, (0) is proportional to
(11.107)
This equation is exactly the form of the sudden approximation. What happened to the
potential energy? It is not really ignored. In the sudden approximation the essential physics is
that when the neutron strikes the 4He particle, its potential energy does not change imme-
diately. The impulse will alter the particle velocity, and after a time duration the subsequent
motion will alter its potential energy. However, by that time the neutron is gone and does not
record this change in potential energy. The potential energy depends only on the position of
the particles, and when the neutron strikes one particle and redirects its motion, the potential
energy only changes later. The fast-moving neutron does not remain around to record this
alteration in potential energy, so that the change in energy appears to involve only the kinetic
energy component.
11.1.6. Bijl-Feynman theory

The excitation spectrum of 4He is shown in Fig. 11.6 as determined by neutron scat-
tering by Cowley and Woods (1971). These points are found as maxima in S(Q, (0). For some
100
__ "tat
2M
80 + Res
+ TACS
60
.
:.:
>-
C)
II::
UJ 40
Z
UJ
20
'ONE PHONON
FIGURE 11.6 Neutron scattering results for the excitation spectrum of liquid 4He at T = 1.1 K. The lower solid line
is the one-phonon curve, while the dashed line is for the noninteracting particle. A broad second peak becomes the
single-particle curve at large Q and the two-roton bound state at small Q. Source: Cowley and Woods (1971) (used
with permission).
values of Q there are two values of co, which indicates that S(Q, co) has two peaks. The dashed
line shows the free-particle spectrum Ek = Ji2 12m. The excitation spectrum approaches this at
large values of k, as shown by Harling (1971). At small values of k, the lower branch of the
dispersion curve becomes linear in k. Small wave vector is the phonon region of sound waves,
and the slope is the sound velocity, 240 mls. The lower dispersion curve has a second
minimum which is called the roton. This point has a high density of states and so has a large
weight when averaging over the density of excitation states. Indeed, the necessity of this
minimum was deduced by Landau (1941, 1947) on the basis of thermodynamic data. The
roton parameters are ko = 1.91 A and .1 = 8.68 K. There is a two-roton bound state at 2.1
-I
which forms the lower limit of the upper dispersion curve (Ruvalds and Zawadowski, 1970)
Zawadowski et al. (1972). The temperature dependence of the excitations was measured by
Stirling and Glyde (1990).
This experimental curve forms the basis for the low-temperature and superfiuid prop-
erties of 4He. It explains all macroscopic behavior, based on the Landau two-fluid model.
Once the excitation spectrum is derived, superfiuidity can be explained. Deriving the exci-
tation spectrum is the objective of this section.
The successful method of finding the excitation spectrum of liquid 4He was popularized
by Feynman (1954). He suggested that the excitation spectrum is given by the formula
EkIS(k), where S(k) is the liquid structure factor. This formula is qualitatively right but not
quantitatively accurate. However, its importance is that it led to the formalism which does
produce the right result. The same formula EdS(k) was obtained much earlier by Bijl (1940),
whose suggestion attracted little notice since neither S(k) nor COk were known in those early
days. The current usage is to call this expression the Bijl-Feynman formula:
(11.108)
This result is called coo(k) and is derived next.

The goal is to construct a wave function 'Pk(rl, r2"'" rN) which describes an exci-
tation in the system with a wave vector k For the boson system of liquid 4He, this excitation
wave function must still have the property of being symmetric under the interchange of any
two-particle coordinates. For long-wavelength excitations, the wavefunction will still have the
same kind of short-range correlations which were found for the ground state wave function. It
seems reasonable to assume that the excited state wave function is only a slight perturbation
upon the ground state wave function. A suitable wave function is obtained by writing the
excited state as the product of the ground state wave function and another symmetric function
Ak(rl, ... ,rN) of all the positions:
(11.109)
The idea is simple enough. The stipulation on Ak (r I' ... , r N) is that it is symmetric in the
coordinates and contains the wave vector k Two possible choices are
A~)(rl' ... , r N) = Lk ~>ik'rl = LkP(k) (11.110)

I
A~\rb"" rN) = eXP(ik' ~rl) (11.111)

Perhaps the reader can think of others. Each choice of Ak corresponds to some kind of excited
state of the system. What is the physics behind each one? It is helpful to use operators such as
the total momentum of the system:
Ii
P=-2:V. (11.112)
i} }
For example, the total momentum on 'I'(2) = Af)'I' 0 gives two terms, since P can act upon
either Af ' or 'I' o. However, the average momentum in the ground state must be zero, so this
can be evaluated to give
('I'f)IPI'I'f)) J
= d3rl ... d3rN'I'~Af)*['I'oPAf) + Af)p'I'ol
= ;= lik Jd3rl ... d3rN'I'~Af)* Af)'I'o = Nlik
since A(2)* A(2) = 1. The total momentum of a system of N particles is just Nk, which
corresponds to the uniform flow of the entire fluid with velocity k/m. That is the correct
interpretation of the excitation spectrum represented by the choice 'I'(2) = Af)'I' o. Another
check on this interpretation is to find the energy by evaluating the Hamiltonian:
(11.113)
li
= -2mNT
2 "J d 3 rl'" d 3 rNTO
H/*A(2)*
k [HI n 2A(2)
TOV} k +2 V}A(2)
k . V}TO+
HI A(2)n2\T1
k ViTO ]
+N
1 Jd rl ... d rN l'I'o I
3 3 2~ Very) (11.114)
(11.115)
The average energy per particle is Ek = k2/2m plus the ground state value Eo/N. The cross
term 2V}Af) • V} 'I'0 in the preceding expression averages to zero, since it contains the average
momentum f 'I'~V'I' 0 in the ground state of the stationary liquid.
The other excited state wave function 'I'~I) = A~I)'I' 0 is the important one for the
excitation spectrum of the stationary liquid. The superscript (1) will be dropped, and it will be
called just 'I' k = Ak 'I'0 = LkP(k)'I'o. This term will be investigated systematically. The first
step is to find the normalization constant Lk for the wave function, which is done by setting
the normalization integral to unity:
(11.116)
The terms in the double summation with I = j just give unity since the wave vector depen-
dence drops out. Special care must be taken whenever I =1= j. Then the space integrals can each
be evaluated, at least formally, except those with r} and rl' These integrals over d3rl ... d3rN
just produce the pair distribution function P2(r}, rl) = v-2g(r} - rl):
1= L~ [2: 1 + r#2: :2 Jd\d3rlg(r} -

;=1
rt)eik.(rrr l ) ] (1LlI7)
The summation for each particle I is the same, since all particles are identical. This
summation is eliminated and instead multiply by N:
1 = NLi[l L ~Jd3r.d3rlg(r. -
+#1 r1)eik.(rrr 1) ] (11.118)
v2 } }
Each term in the summation over j gives the same result, so just do one term and multiply by
N - 1 ~ N. Also, change integration variables to relative coordinates d3rjd3rl = vd3r, so the
integral is over g(r). In fact, this expression is precisely the definition of S(k) which is given
1 = NLi(l +P Jd3rg(r)eikor) = NS(k)Li (11.119)
L _ 1
(11.120)
k - JNS(k)
Equation (11.120) provides the rigorous evaluation of the normalization constant L k . It is

useful to call the structure factor S(k) although it is only a function of Ikl = k in the liquid.
Now that the wave function is normalized properly, it is possible to investigate its
properties by taking the expectation of different operators. The first is the momentum operator
P given in (11.112):
(11.121)
First examine a single gradiant of position Vj acting upon the wave function '¥k' Since the
function Ak is a summation over different coordinates, the one term with I = j will be
different from the others:
Vj '¥k(rl."·' rj •...• rN) = Lk LVieik'rl'¥o) (11.122)

1
= Lk[Vieik.rj'¥o) + L el1<.r/Vj '¥o]

1#
= I'kLke11<'r'\II
] TO + A k Vj \IIT 0 (11.123)
The momentum has a simple result
P'¥k = lik'¥k + AkP'¥o (11.124)

(P) = lik + ('¥kIAkP'¥O) (11.125)
= lik + ('¥oIPI'¥o) = lik (11.126)
The first term gives lik. The second term averages to zero because the operators AtAk
average to unity, and then ('1'0 IPI'¥o) is the momentum in the ground state, which is zero. The
momentum in state '¥k is lik. The momentum operator commutes with the original Hamil-
tonian, so that states can be described by the quantum number of momentum lik.
The next step is to calculate the energy of this excitation whose wave vector is k. The
energy is found by taking the expectation value of the Hamiltonian operator between these
states:
(11.127)
It is desirable to commute the Hamiltonian operator to the right, so that it operates on the
ground state '1'0' The ground state is an exact eigenstate of the Hamiltonian H 0/0 = Eo '1'0 and
is the one with the lowest eigenvalue Eo.
Hp(k) = Hp(k) - p(k)H + p(k)H = [H, p(k)] + p(k)H (11.128)
Ek = L~('I'op(-k)[H, p(k)]'I'o) +L~('I'op(-k)p(k)H'I'o) (11.129)
The second term on the right gives EoL~('I'op( -k)p(k)'I'o) = Eo. The first term on the right
contains the commutator of H with the density operator p(k). Only the kinetic energy term
fails to commute with p(k) so
[H, p(k)] = -~ 2]V;, eik . rj ] (11.130)

2m if
li2
= - 2::>lk. rj[k2
.
- 2ik· V] (11.131)
2m j }
('I' ope -k)[H, p(k)]'I' 0) = Ek('I' ope -k)p(k)'I'0) (11.132)

li2
- i- L ('I' ope _k)e ik . rjk· Vj'l' 0) (11.133)
m j
The second term on the right must be treated carefully. Since the ground state wave function is
real, we can replace '1'0Vj 'I'0 by Vi'l' 0)2/2 and then integrate by parts, so that this term is
transformed to
('I' Ope -k)[H, p(k)]'I' 0)
(11.134)
(11.135)
The gradient operator in the last term will produce a factor ik in each term except the one
where I = j. The two terms will cancel term by term except the one where I = j, which exists
only in the first term. The result is
('I'op( -k)[H, p(k)]'I'o) = Ek L 1 = NEk (11.136)
j
Since L~ = l/NS(k), the first term in (11.129) for Ek is Ek/S(k), and the second is Eo. The
energy in the system for the excitation k is
Ek = ('I'kIHI'I'k) = Eo + moCk) (11.137)
Ek
moCk) = S(k) (11.138)
The excitation energy is defined as moCk), since it is the energy above the ground state Eo. The
result is the Bijl-Feynman excitation energy.
The solid line in Fig. 11.7 shows moCk) and is compared with the actual excitation
spectrum of liquid 4He. The two curves have exactly the same shape, but the Bijl-Feynman
spectrum is about a factor of two larger. The roton minimum in the Bijl-Feynman dispersion
curve is caused by the peak in S(k) at k ~ 2 A. At small values of k, moCk) goes to zero
70
60
50
20
1.0 2.0 3.0

Wave number K ('&-')
FIGURE 11.7 The excitation spectrum of liquid 4He. The solid line is the Bijl-Feynman theory, compared with the
experimental data as the points.
linearly with k, and the slope is the speed of sound. In the Bijl-Feynman theory, this happens
because ck vanishes as k 2 while S(k) vanishes as k, so their ratio does vanish as k. One cannot
use experimental values of S(k) in this region, since the experimental values of S(k) have an
additional contribution from thermal fluctuations which provides a small constant to S(k) in
the limit where k -+ O. This constant, which is proportional to kBT times the compressibility,
is easily subtracted from experimental S(k) to obtain the S(k) one desires to use in the Bijl-
Feynman formula.
The agreement between theory coo(k) and experiment coCk) is not satisfactory. The theory
must press onward, and one must look for additional interactions or mechanisms which will
improve the comparison. The interactions between these excitations must be included. Next
consider a system with multiple excitations, where each is obtained by multiplying the ground
state by the factor Ak = LkP(k):
\fIk = Ak \flo = Ik)

\fIk, ,k2 = Ak, Ak2 \flo = Ik J , k 2) (11.139)
\fIk" ... ,kn = A k, ... Ak n \flo =Ik J , ••• , kn)
The Ak are scalar functions, so it does not matter in which order they are arranged. These
states are examined carefully in the next section to see whether the different excitations are
independent or whether they interact.
11.1.7. Improved Excitation Spectra

The method of Feenberg (1969) will be described, which is based upon the states in
(11.139) which contain multiple excitations. They are examined to test their orthogonality and
their interaction. The state of one excitation Ik) has an interaction with the state of two
excitations Ik', k - k'). They contribute self-energy effects for the state of one interaction.
These self-energy effects are just what is necessary to give a good spectrum. The final result
will be w(k) = wo(k) + :E(k), where :E(k) is the self-energy of the excitation k. The method
of calculating this self-energy is only outlined; a full description is in Feenberg's book.
The reader should be aware that the present procedure is dramatically different from any
used elsewhere in this book. All the other problems have been based on the interaction
representation, where the Hamiltonian is written as H = Ho + V, where V is a well defined
perturbation. The present approach does not do this at all. Instead, a set of states such as
(11.139) is written down as an ansatz. These states cannot be readily identified with any part
of the Hamiltonian and are not orthogonal eigenstates of any operator which is obvious. Nor
is it straightforward to do perturbation theory, since there is no V in which to expand. These
problems can all be overcome, but the reader should be alerted that the present derivation is
quite different from previous ones.
Feenberg and his associates established the following properties for these excitation
states:
(kplkp) = 1, (11.140)
(kklkk) = 2 (11.141)
(kpIHlkp) = Eo + wo(k) + wo(p) (11.142)
These properties are exactly those for operators Ak which are were treated as boson creation
operators at. For example, the state with two excitations in k is
Ik) =atlO} = 11 k } (11.143)
Ikk) = atatlO} = atll k } = .J212k } (11.144)
where the .J2 comes from the usual raising operators for bosons. This factor explains the
additional factor of 2 in (11.141). The state Ik 1k2) has an excitation energy given by
wo(k 1) + wo(k2) and seems to be composed of states with independent boson excitations.
However, this free-boson analogy should not be pushed too literally. For example, states
with different numbers of bosons but with the same total momentum are orthogonal only to
order O(lj..;N). For example,
IP
(k 1 ,k2Ik 1 +k2) = [S(kl)S(k2~(kl +k2)J (11.145)
(k 1, k21Hlki + k2) = o(lv) (11.146)
The last two matrix elements are important for calculating the self-energy of the excitations.
There exists a matrix element for one excitation of wave vector k going to two states with k'
and k - k'. The self-energy diagram is shown in Fig. 11.8. The excitations are represented by
wavy lines, and the one excitation has a self-energy from making two excitations. The same
Feynman diagram is calculated for the self-energy of a phonon from anharmonic interactions.
The matrix elements in (11.146) show that the excitations are not truly independent.
Anharmonic processes contribute to the self-energy. It is derived in the following fashion.
Since there is no obvious interaction V in the present problem, instead use H - E. The basic
matric element is (p, k - pl(H - E)lk) which uses both results in (11.146). From second
order Brillouin-Wigner perturbation theory, the self-energy is
:E(k,E)=!L: l(p,k-pl(H-E)lk)1 2 (11.147)

p E - Eo - wo(p) - wo(k - p)
k-k'
FIGURE 11.8
A self-consistent equation is derived for roCk) = roo(k) + I:(k). Note that E appears in
the matrix element and the energy denominator. Feenberg and his collaborators took the
quantity (H - E) ~ (H - Eo - roo(k)) in the matrix element, and found only an approximate
eigenvalue from their results. The details of calculating the matrix element of H between the
excitation states is tedious, since it involves an evaluation of three-particle correlation
functions for the liquid. Their numerical results are shown in Fig. 11.9. as the dashed line.
They compare quite favorably with the experimental spectrum as found by neutron scattering.
The result is a dramatic improvement over the Bijl-Feynman curve, which is marked as B-F.
Probably the first calculation of this kind was done by Feynman and Cohen (1956, 1957), who
did a variational calculation of the backfiow around the excitation. That is exactly the same
physics as the admixing of the double excitation states, so the two methods of calculation are
really the same but dressed in a slightly different language.
The theory of liquid 4He is remarkably successful. Starting from the potential function
VCr) between atoms and other basic numbers such as the particle mass and density, the theory
has calculated most of the features of the liquid with great success. By using correlated basis
functions, the ground state properties such as g(r), S(k), nk' condensate fraction/o, and off-
70
60
50
20
10
W 20 3D
Wave number K (A -I)
FIGURE 11.9 The excitation spectrum of liquid 4He. The points are the experimental result for the superfluid. The
other lines are various theories, including Bijl-Feynman (B-F) and backflow theories which use the Kirkwood
superposition approximation (KSA) or convolution approximation (CA). Solid points are the Feynman-Cohen theory.
Source: Feenberg (1969) (used with permission).
diagonal density matrix R(r) are obtained. Most of these quantities, or their Fourier trans-
forms, can be compared favorably with experimental results. Similarly, the excitation spec-
trum can also be calculated. After some work, a decent curve for coCk) is now found which
compares well with the experimental results. All of the features of liquid 4He have been
derived starting from a microscopic theory. The excitation spectrum coCk) is all that is required
to find the macroscopic properties, such as the superfluid density in the two-fluid model, and
the superfluid flow properties. The macroscopic theory of superfluidity in 4He can be derived,
step by step, from a knowledge of only VCr) plus some fundamental constants and masses.
This achievement is remarkable for such a strongly interacting and highly correlated system.
11.1.S. Superfluidity
This section explains why the liquid is superfluid, and shows how to compute its
superfluid properties. The explanation was first provided by Landau (1941, 1947), whose
derivation is given below.
For temperatures below T)." in the superfluid state, the fluid density can be viewed as
consisting of a normal PN and superfluid Ps component. The sum of these two is the normal
density P = PN + Ps· The two components PN and Ps vary considerably with temperature
where Ps(T).,,) = 0 and PN(T = 0) = O. The actual variation with temperature of PN/P is
shown in Fig. 11.10. In this two-fluid model, each component of the fluid, normal and
superfluid, carries its own momentum, energy, etc. The normal component is given by the
following expression for T < T)."
(11.148)
IDDr-------------------------------~r_~
8.0
C.,
0 6.0
~
~
...z 4.0
2.0
DL-____- L____ ~ ______ ~ ____~ ____~~
1.2 1.6 2.0 Xpl

Temperature (OK)
FIGURE 11.10 The density of the nonnal fluid PN as a function of temperature: 0, Derived from oscillating disk
experiments; ., from the velocity of second sound. Source: Lifshits and Andronikashivile (1959).
which is explained later. The superfluid component is just Ps = P - PN and has nothing at all
to do with the condensate fractionfo. If one takes the measured excitation spectrum m(q) and
uses it to theoretically calculate PN(T), one obtains the solid line in Fig. 11.10. This calcu-
lation is in remarkably good agreement with a number of ways to measure PN(T) such as the
viscous drag on rotating plates, etc. Indeed, Landau was able to deduce a number of properties
of m(q) before it was measured directly from neutron scattering by fitting Eq. (11.148) to the
measured component of normal density. In particular, the curves for PN(T) show an activation
energy, which is now associated with the roton minimum Ll. Landau was able to deduce the
necessity for this roton minimum, as well as determine the speed of sound, and predicted
quite well the entire shape of the excitation spectrum, including phonon and roton parts. His
was a remarkable theoretical achievement.
The Landau argument may be understood by considering a Gedanken experiment, which
has the liquid helium flowing down a pipe with a uniform velocity v. This velocity is small,
say 1 cm/s, which is much smaller than the velocities in the zero-point motion of the fluid.
The total momentum of the particle flow is calculated, and it is not just Nmv, where N is the
number of particles.
First construct a wave function which describes the uniformly moving liquid, which uses
the operator A~) for uniform flow in (11.110). Label this wave function as 'I'v(rl, r 2 , •.. , r N )
to denote the state of uniform flow. The energy and momentum of this wave function were
calculated earlier:
'I'v(rl, rz, ... , rN) = exp (imv. ;= rj )'I'o(r l ... r N) (11.149)
('I'vIPI'I'v) = mNv (11.150)
('I'vI H I'I'v) = Eo
mv
+N 2 (11.151)
The next step is to consider the excitation spectrum of this moving fluid. The multiplication of
'I'v by Ak = LkP(k) produces an excitation with momentum hk and excitation energy
moCk) + k· v, as shown by
<Pv(k) = LkP(k)'I'v (11.152)

(<Pv(k)lPl<Pv(k)} = mNv + hk (11.153)
(<Pv(k)IHI<Pv(k)} = Eo
mv
+ N 2 + moCk) + k· v (11.154)
The result moCk) + k . v is quite reasonable: If moCk) is the excitation spectrum with respect to
the frame moving with the fluid, then the Doppler-shifted energy, as viewed by an observer at
rest in the laboratory, is moCk) + k· v.
Of course, the excitation spectrum should be m(k) + k· v rather than moCk) + k· v. This
improvement can be achieved by including the self-energy in (11.147) which produces the
improved excitation spectra. This step is done and the excitations are indeed the ones which
are physically realistic.
The momentum operator gives mNv + hk for a single excitation of energy m(k) + k . v.
In the actual system, there will be a number of excitations whose distribution is given by the
thermal occupation factor nB for boson excitations. The average value of the momentum
operator is
+ NJ
- -d k 3
('¥yP'¥y}av = Nmv - 3 hknB[co(k) + k· v] (11.155)
P (21t)
The argument of the boson distribution factor nB(Z) = 1/[expWZ) - 1] is the excitation
energy co(k) + k· v in the laboratory frame. The chemical potential is zero in the superfluid
state, since it is a form of Bose-Einstein condensation. Note that this expression is not the
average momentum in the ground state of flowing fluid, since that would be found from the
average of m f d 3kl(21t)3hkn k _ mv = Nmv, where nk-my is the momentum distribution of the
particles in the ground state of the superfluid. The total momentum in the ground state of the
flowing fluid is just Nmv. Instead, (11.155) is the total momentum in the flowing fluid as
contributed by the ground state plus the excitations. The excitations are produced by the
thermal fluctuations, so that the total momentum will be temperature dependent.
For small velocities the argument of the distribution function can be expanded in powers
of k· v: nB(co(k) + k· v) ~ nB(co) + ~k· vno(co) + O(v2). The first term with just nB[co(k)]
gives a zero average of hk, and the important result is from the second term in the series
expansion:
+ -~NJ -d k3
(P}av = Nmv - 3 hk(k· v)no[co(k)] (11.156)
P (21t)
Since the spectrum no is isotropic in k, the angular integrals are done easily and give a vector
in the direction v in f dQkk(k· v) = 41tvk2 /3. Collect together the remaining factors shows
that the total momentum is expressed in terms of the normal fraction of superfluid which was
defined in (11.148):
(P}av = Nmv( 1 - P;) = ~ Nmv (11.157)
The final result is that the total momentum in the flowing fluid is given only by the fraction
Ps/p which is superfluid. The normal fraction PNlp gives no contribution to the momentum.
In a normal liquid, uniform flow is not an eigenstate because viscosity damps the flow.
The superfluid has no viscosity, and uniform flow is an eigenstate. At nonzero temperatures
0< T < T"J..' only the superfluid component Psi P can flow without damping and is an
eigenstate with a well defined momentum.
The physical picture of this result is very subtle. The flow of liquid is down a pipe, which
remains in the laboratory frame. The normal component of the fluid is pinned by the pipe and
does not participate in the flow. There are two important points to this derivation. The
reference point for the excitation spectra is the rest frame of the superfluid. The excitations are
made from this flowing fluid and have an energy co(k) with respect to the rest frame of the
superfluid. However, the influence of the excitations was calculated in the rest frame of the
walls of the container, in this case the pipe. The coordinates were shifted to the laboratory
frame and which gave the Doppler-shifted energy co(k) + k • v before doing the thermal
averaging. The rest frame of the container and the superfluid both enter the calculation. The
result is that only the superfluid density carries momentum when the walls are stationary.
The flowing superfluid is in an eigenstate of the system, which can persist without
damping. The state '¥y has an eigenvalues (ps/p)Nmv of the momentum operator. In a
completely normal fluid, such as neon or 4He for T> T"J..' uniform flow down a fixed pipe is
Sec. 11.2 • LIQUID 3He 713
not an eigenstate, and the flow is dissipated by viscosity. In the superfluid, the component Ps
is not dissipated.
This calculation should be compared with other similar systems. What if the entire
container of superfluid is just picked up and moved at a constant velocity; i.e., what if a
Dewar is driven on a truck? Then the walls and the superfluid are moving at the same velocity,
and the thermal averaging should be done in their mutual rest frame, which is moving with the
truck. Then the momentum average gives
= Nmv + NJ
- -d- k
3
(P}av 3 hknB[ro(k)] = Nmv (11.158)
P (21t)
Now the result is Nmv since both normal and superfluid components are moving with the
same velocity.
11.2. LIQUID 3He
It is certainly a misnomer to call liquid 3He a "normal liquid." The title is used here to
distinguish it from the superfluid phase of the liquid. This system is interesting because its
properties are so exceptional. It is really the only ideal Fermi liquid in earthbound nature,
since electrons in metals are not ideal because their properties are perturbed by the crystal
lattice of ions. The 3He atoms are charge neutral, so that the properties of the Fermi system
are much different from electrons in metals. They are also strongly interacting and highly
correlated, so the noninteracting Fermi gas is not a good starting point for a description of its
ground state. The discovery that 3He has a superfluid phase, which is akin to the BCS
superconductor, has heightened interest in this liquid.
The theory of these systems operates at several levels. One example is the two-fluid
model of hydrodynamic flow, and another example is the microscopic derivation of S(k) and
g(r). The theoretical description can be viewed as a sequence of plateaux, where each plateau
is a self-consistent description of the system. For example, the excitation spectrum of liquid
4He is such a plateau. One feature of these plateaux is that they can be compared directly with
experiment. The theoretical effort generally is devoted to deducing the theory of one plateau
by using the model of the lower one. For example, in liquid 4He, one does not try to derive in
one step the hydrodynamic flow properties from the microscopic theory. Instead, one uses
microscopic theory to find the excitation spectrum roCk), which is one plateau. Then one uses
Landau theory to derive the two-fluid model in terms of Ps(T) and PN(T), which are described
in terms of roCk). The theory is a sequence of plateaux, each of which can be compared with
experiment, and one uses the theory to try to relate one step to the next one higher in the
sequence.
The same sequence applies to the properties ofliquid 3He. The first, and most important
plateau, is the Fermi liquid theory which was developed by Landau (1946, 1956, 1957), who
has provided the main insight into both 4He and 3He. This theory was developed as a
phenomenological model for the behavior of low-lying excitations of the strongly interacting
Fermi system. These excitations can be described by a few parameters, which can be deduced
from experiments. A major goal of microscopic theory is to derive these parameters from first
principles. The major effort along this line, as was also the case for 4He, is by Feenberg and
his associates. So far they have achieved a good level of agreement, which is on the order of
30% at the worst and is better for some quantities. Since the Landau theory of the Fermi liquid
is the major plateau of the theory, this section is devoted to describing it as well as
summarizing the microscopic theory.
The next plateau is a description of the superfluid state, which is usually done using the
parameters of Fermi liquid theory. These parameters play a central role in the theoretical
development. The superfluid properties are well described in this fashion using the experi-
mentally measured parameters. In this sense a quantitative theory is available for the super-
fluid state.
11.2.1. Fermi Liquid Theory

There are excellent reviews of Fermi liquid theory as applied to liquid 3He in the normal
phase. Several of these are Pines and Nozieres (1966), Baym and Pethick (1978), Ron (1975),
and Leggett (1975). These treatments are usually more detailed than the one here, and the
reader is encouraged to read some of these others.
The symbol np is defined in (3.135) as the average number of atoms in the Fermi liquid
with momentum p. In a noninteracting Fermi system, at zero temperature, this distribution is
np = 8(kF - Ipl). Of course, in a strongly interacting Fermi system, this distribution will be
strongly smeared. Figure 11.11 shows the momentum distribution for 3He measured by
neutron scattering at T = O.37K by Mook (1985). The points are experimental. The solid line
is a fit to the Fermi distribution n(p) = l/[exp(~p/kBTK) + 1] with TK = 1.8 K. The smearing
caused by particle--particle interactions can be represented by an effective temperature in the
Fermi distribution.
Throughout the remaining part of this chapter, the symbol np is used to mean something
entirely different from its usage elsewhere in this book. Whereas before np always is the
distribution of states with free-particle momentum p, now it has a different meaning. Fermi
liquid theory has a well-entrenched set of notation, with a universal set of symbols, whose
meaning is well agreed upon. These strong conventions are followed here, and now np is the
distribution of excitations of momentum p.
The objective of Fermi liquid theory is to describe the low-lying excited states of the
fermion system. There is a ground state which the system would be in at zero temperature
except for superfluid phases and this ground state will be discussed in the next section. In
analogy with the electron gas, two kinds of low-lying excitations are expected. The first are
density oscillations, which are plasmons in the charged electron gas. In the neutral Fermi
liquid they are sound waves. The second type of excitation is particle--hole pairs. There is also
a third kind of excitation, which is a damped spin wave, called a paramagnon. Fermi liquid
1.2
0.8
0.4
OL-~0~.2~~0~.4~~0~.6~~0~.8~~~~1.~2~~1.4
p(I-')
FIGURE 11.11 Momentum distribution np of particles in 3He at T = 0.37 K. The solid line is a least-squares fit to
the Fenni distribution with an effective temperature of Tp = 1.8 K. Source: Mook (1985) (used with pennission).
Sec. 11.2 • LlaUID 3He 715
theory is primarily aimed at a description of the properties of low-lying particles and holes.
However, the other excitations-density oscillations and paramagnons-can be viewed as
collective resonances of these particle and hole excitations. Fermi liquid theory also describes
them. Indeed, many of the so-called Fermi liquid effects actually come from the collective
properties. The discussion will follow the historical pathway and first treat the classical Fermi
liquid theory due to Landau. Then the modem interpretation in terms of collective properties
will be discussed.
By using the parameters in Table 11.1, one would calculate a Fermi energy of EFO =
4.97 K for a noninteracting gas of the same density and mass of 3He. The actual degeneracy
energy in liquid 3He is about two-thirds of this value. Fermi liquid theory is valid only for the
particle-hole excitations whose energy is a small fraction of the degeneracy energy-say
5-10%. The excitations in this theory have low energy, say less than 0.1-0.2 K. If these
excitations are used to describe the thermodynamics of the excited states of the system, it can
only apply for temperatures below 0.1-0.2 K. Otherwise, the thermodynamic properties will
involve states beyond the range of the theory. Fermi liquid theory is concerned with very low-
energy excitations, which are important only in the liquid at very low temperatures. Since the
superfluid properties begin at a temperature two decades lower than this, there is an appre-
ciable temperature range over which Fermi liquid theory is applicable.
In the homogeneous electron gas, the imaginary part of the electron self-energy, from
electron-electron interactions, vanishes for electrons whose energy is right at the chemical
potential j..l. Furthermore, 1m L is also small for states near the chemical potential, since it
vanishes as (cp - j..ll The low-lying excited states of the Fermi liquid have a long lifetime.
The same phenomenon is assumed to happen in 3He, and general theorems can be proved
which demonstrate this with rigor (Luttinger and Nozieres, 1962). In cases where 1m L ~ 0,
the spectral function A(P , co) is sharply peaked and has the character of a delta function. There
is a unique relationship between energy and momentum in these cases, and the excitation is
satisfactorily described by only one of these two variables. In Fermi liquid theory this variable
is conventionally taken to be the momentum p. An excitation of momentum p is assigned an
energy cp .
The energy zero is defined where the excitation energy cp is zero at p = O. Furthermore,
if the energy is expanded in a Taylor series about point p = 0, the odd terms in this series
p2m+l must vanish because of the homogeneity of the liquid and the isotropy between p and
-po The first term in the expansion is p2, and other terms are usually ignored. Write cp ex.. p2,
and the constant of proportionality defines the effective mass m*, cp = p2 /2m*. In liquid 3He,
the effective mass has a value of m* = 2.76m, where m is the bare mass of 3He. These and
other experimental properties are reviewed by Glyde (1994) and by Vollhardt and Wolfe
(1990).
TABLE 11.1 Properties ofliquid helium (pressure = I bar). The parameters

cr, E are for the Lennard-Jones potential
Parameter 4He 3He
m 6.65 x 10-27 kg 5.01 X 10-27 kg

P 2.18 X 1028 atoms/m] 1.64 x 1028 atoms/m]
Cj(T = 1 K) 238 mls 182 mls
(J 2.648 A 2.648 A
E 1.484 X 10-22 J = 10.7 K 1.484 x 10-22 J = 10.7 K
EIN -7.20 Klatom -2.52 Klatom
The ground state of the fennion system does not allow all the excitations to be in the
state with p = O. The exclusion principle forces many excitations into states of nonzero p, and
all energy states are occupied at zero temperature for Ep < j.l. Define n~ as the number of
excitations of momentum p in the ground state, in which case the ground state energy is
(11.159)
where Eo is the energy when all the quasiparticles are in p = 0, which would be the case if
they were bosons. Of course, the result (11.159) is actually nonsense, since the approximation
of treating the excitations as free particles with an effective mass m* and no damping is valid
only when IEp - j.ll :s kBT and does not apply throughout the degenerate Fermi sea. The
quantity n~ is a totally fictitious concept. However, it is never used for any property of n~
except at the chemical potential, where the following fonn is assumed at nonzero temperature:
o 1
np = nF(E p - j.l) = e~(Ep-l1) + 1 (11.160)
This approximation is surely valid only very near the chemical potential ~(Ep - j.l) :s 1, which
is the only place it is used.
The important feature is the excitation of the system above the ground state. When
excitations are present, the distribution np of these excitations will be different from the
ground state distribution n~. The difference between these two distributions is defined as
onp = np - n~. The important physical idea is that the crucial quantity is neither np nor n~ but
rather on p , since it gives the number of excitations in the excited state which is important for
the thennodynamics of the system at low temperature. Neither np nor n~ can be detennined
with great accuracy, but onp can be detennined, which is all that matters. For example, the
total energy of the system, including the excited states, is
E =Eo + LEpnp (11.161)
per
= Eo + L Epn~ + L Ep[np - n~] (11.162)

per per
(11.163)
The excited state energy is L Epon p .

The important quantity is not the energy but the free energy F = E - j.lN, where N is the
total number of particles. The particle number is defined in tenns of the number of particles in
the ground state No
No0- L.., nO
- " p (11.164)
per
- "n p-L..,
N -L.., - " nOp + "on
L.., p-- No0 + oN (11.165)
per per per
F =E - j.lN = Eg - flNo + L(Ep - j.l)onp (11.166)

per
There are two key assumptions contained in these equations. The first is that the chemical
potential j.l does not change because of the excited states, although it does change with
temperature. The second assumption is that there is a one-to-one correspondence between
quasiparticles and the excited states of the noninteracting Fenni gas. Adding a quasiparticle
also adds a particle and changes N by unity. Of course, usually there are about equal numbers
Sec.11.2 • LIQUID 3He 717
of quasiparticles above sp > j..l, which have 8np > 0, as there are a depletion of them for sp < j..l,
which have 8np < O. The latter excitations are called holes.
So far the Fermi liquid theory has been equivalent to calling the excitations free particles
with an effective mass m*. If that were all of it, the theory would be unexceptional and not
quantitatively accurate. The genius of Landau was in realizing that the expansion for the free
energy must have another term, which is due to the interactions. This contribution is required
for a self-consistent theory. The formulas derived so far are the terms in an expansion of the
free energy in the excited distribution 8np. That is, F(8np) is a functional of 8n p and can be
expanded formally in a Taylor series in this parameter. The Landau theory assumes that 8np is
small and that an accurate quantitative theory is achieved after a few terms in this expansion.
The terms derived so far are given in (11.165). The first term is Fo = Eg - J.1N. The next
term appears to be first order in 8np. A closer inspection shows that this term is actually of
order (8npf The summation over p will run over all values for which 8n p is nonzero.
However, the integrand contains (sp - j..l), which vanishes at sp = j..l. If 8np has value up to ~
away from the chemical potential, the kinetic energy integral Lp(Sp - j..l)8np <X ~2. This term
is second order in the quantity 8np. Landau saw the need to include all terms of order (8np)2.
There is another term which describes the interactions between the excited quasiparticles and
so contains the dependence 8n p8n p" This interaction is very dependent on the spin of the
quasiparticles. The free-energy expansion has the form
1 3
F = Fo + Z)sp - j..l)8npcr + -2 L f pcr.p'cr,8npcr8np'cr' + O(8n)
pcr v pp', crcr'
where spin has been added to various quantities. This dependence on the particle spin is not
due to the spin-dependent interactions. Indeed, the dipole-dipole forces between the nuclear
moments are extremely small and may be neglected when discussing most phenomena. The
spin dependence of the interaction is merely due to particle statistics. The anti symmetrization
of the many-particle wave function for fermions causes an exchange hole around each
particle. The exchange hole causes an effective exchange energy, just as it did for the electron
gas. The exchange energy may be qualitatively understood by a consideration of the mutual
scattering of two 3He atoms. Since the two-particle wave function must be antisymmetric, the
scattering properties depend on the total spin state of the two spin one-half particles as
described in Sec. 11.1.1. If they are in an S = 0 state, the orbital cross section contains only
even angular momentum components; if they have a total S = 1 state, only odd angular
momentum components enter the cross section. The scattering and interaction depend on the
relative spins of the two particles. It is a very important effect in liquid 3He, since the effective
interaction fpcr,p'cr' is very spin dependent.
The spin dependence of the effective interaction is written as
(11.167)
where 0' = (O'x, O'y, O'z) refer to Pauli spin matrices. Equation (11.167) is a very particular
choice for the form of the interaction. Other possible spin dependencies are ignored, such as
spin-orbit effects (0" p)(O'" p') or spin-other-orbit effects (0" p')(O'" p). The choice of a
simple 0" 0" term derives from our understanding that the basic spin mechanism is due to
exchange forces, which can be written this way.
Fermi liquid theory is really only accurate for describing the interactions of quasi-
particles on the Fermi surface. In the interaction termsf~p' andf;p" the momentum variables
have the magnitude kF' and the only important variable is the angle 8 between p and p'. The
interaction terms are represented by an expansion in Legendre polynomials. It is also

conventional to normalize the coefficients in this expansion by removing a factor of the
density of states at the Fermi energy NF = m*kF /n 2 fi3:
1 00
I:p~ = -N L Ff'a PI ( cos 9) (11.168)
FI=O
The coefficients Fi and Fi are sometimes called FI and ZI' They are the fundamental
parameters in the theory and are dimensionless. They can be deduced from experiments, as
shown later. Reliable experimental numbers are now available for Fo' Fj, Fo and Fl' A major
goal of the microscopic theory is to derive them.
The term Ipcr,p'cr' describes the interaction between excited quasiparticles. The interac-
tions cause the energy of an excited quasiparticle to depend on the number i5npcr of other
excited ones. The quasiparticle energy can be formally derived as a functional derivative of
the total energy E:
(11.169)
(11.170)
The energy of the excited quasiparticle is called Ep. It depends on the number of other excited
quasiparticles through the interaction term. One result of the interactions is that the total
excitation energy is no longer equal to L Epi5npcr' since this expression overcounts the
pairwise interaction term. The chemical potential Il is not altered by the density of excited
quasi particles as long as this density is very small compared to the density of atoms.
It is time to reconsider the distribution of excited quasiparticles. The same notation is
used: i.e., that np is the total number of excited quasiparticles and that ng is the number of
excited quasiparticles in the ground state with energy cp . Another symbol is needed to denote
the quasiparticle density in terms of the actual energy variable Ep. Call iig the equilibrium
density of excited quasiparticles with energy Ep' which is defined in the usual way in terms of
the Fermi distribution function:
(11.171)
(11.172)
Similarly, the quantity i5iipcr is the departure of the total number npcr from this equilibrium
concentration. The two quantities i5npcr and i5iipcr are not independent and are related through
the Fermi liquid parameters. This dependence can be derived by assuming that i5npcr is a small
quantity and expanding i5iipcr in this parameter. The key step is the recognition that the
difference (Ep - cp) is proportional to i5npcr
i5iipcr = npcr - nF(Ep - Il) (11.173)

= npcr - nF(cp - Il + Ep - cp) (11.174)
_ dnF(c p -Il)
= npcr - nF(C p - Il) - (c p - cp) d (11.175)
cp
(11.176)
The first two terms in (11.175) are just 8npcr' The third term contains the factor (Ep - Ep)
which is also proportional to 8iipcr' It has the factor -dnF(E p - 11)/dEp, which becomes the
delta function 8(Ep - 11) at zero temperature. One can show that every term in Eq. (11.176)
has this factor. Divide it out of the expression. Furthermore, the spin dependence of 8iipcr is
written as a symmetric or antisymmetric combination oiip± = 8n~ ± 8n~. All these results are
collected, and the distribution functions are expanded in spherical harmonic functions Yim.
The first two equations below define 8n:::, etc., while the third equation is the integral
equation (11.176):
_ dnF(E p - 11) a
L Ylm(9, )(8nlm ± 8n lm )
S
onp± - - d (11.177)
Ep 1m
dn (E -11)
oiip± = - F/ L Ylm (9, )(8iiim ± 8iiZm ) (11.178)
Ep 1m
+ 2 J--3~-4
3
s:-s,a _ s: s,a
unlm -unlm dn F) .lIm
d p' "dnp ( --d v* (9 "")f.s,a v
,'+' p'p.ll'm'
(9' ''+'
""')S:unl'm'
s,a (11.179)
(2n) I'm' n Ep'
Integrate over angle and collect all terms with the same angular dependence, which provides
the relationship between the two distributions of excited quasiparticles:
8iis ,a
Im
= 8n s1m,a ( 1 + _l_
Fs,a )
21 + 1
(11.180)
These results are useful later.

The distribution 8iipcr seems more important than 8npcr' For example, Pines and Nozieres
show that the particle current operator is given correctly by both of the following expressions:
(11.181)
(11.182)
The two current operators (11.181) and (11.182) can be used to derive a relationship between
m* and m. The current operators are nonzero if the distribution functions oiipcr and 8npcr have
a nonzero component of angular momentum 1 = 1, say the value 1 = 1, m = O. Then the two
current expressions can be evaluated by using the definitions of 8iipcr and 8npcr in (11.180):
-m* _ 8iiio _ 1 iFs

m - -8ns - +-3 i (11.183)
10
A measurement of m* gives the value of Ff. It was previously remarked that m* ~ 2.76m, so
that Ff ~ 5.3. The Fermi liquid parameters can be large relative to unity.
Table 11.2 shows some predictions of Fermi liquid theory regarding the effect of the
quasiparticle interactions upon measurable quantities. The first is the specific heat, which is
enhanced by the effective mass m*, as it is for the electron gas. The quantity Fj is obtained by
measuring this quantity. The isothermal compressibility and also the velocity of ordinary
sound are changed by the ratio of (1 + Fij)/(1 + FU3). For a noninteracting gas of 3He
particles, one would estimate the sound velocity from the formula 20 = V}/3 to be 95 mls.
The experimental value is 182 mls. The ratio is c/co = 1.91 gives the value of Fo
= 9.15
since Ff is known. A recent compilation of these parameters is found in Table 11.3, as given
TABLE 11.2 Fenni liquid theory:NFo = mkF7rH~J
Quantity Free particles Fenni liquid theory
rc 2 ~_ m* - I +lp
Specific heat e vo = 3k~TNFO e vo - m - J 1
n2
_0 K 1 +Fil
Compressibility
Kf N FO K( 1 +Ffl3
Sound velocity
2 no v}
co=--=-
mNFO 3
(CY
;:;; = K
Kf
X 1 + Ff /3
Spin susceptibililty Xo = J.l6 N FO Xo I +FIl
TABLE 11.3 Fenni liquid parameters: Ae = FeI[1 + Fe(U + 1)]

(from Greywall 1983)
Parameter Experiment Parameter Experiment
F'0 9.15 AS0 0.90

F S1 5.27 AS1 1.91
Fa0 -0.70 A0a -2.33
Fa1 -0.55 Aa1 -0.67
by Greywall (1983) who found that m* = 2.76m. The other parameters Af·a shown in Table
1l.3 are explained later.
o'
The other interesting number is F which is the isotropic term in the spin-dependent
part of the interaction. It is deduced from the spin susceptibility, which is found to be much
larger than the free-particle value Xo. The enhancement of X over Xo suggests that is Fo
negative and near unity, so that the denominator is small in the theoretical expression for the
susceptibility. Later it is shown that this has an important consequence for the theory of
Fo
superfluidity. The large positive value of and the negative value of combine to suppress Fo
the quasiparticle interaction in the s-wave channel and enhance it in the p-wave channel. It is
this reason that the superfluid phases in 3He pair up in the spin triplet state, which has a p-
wave orbital part.
11.2.2. Experiments and Microscopic Theories

The microscopic theories of liquid 3He are even more complicated than the theories of
4He, because the ground state is more complicated. The complexity is due to the anti-
symmetrization of the many-particle wave function. If this ground state wave function is
defined as 'P g (X1, ... ,XN), where Xj = (rj' cr) are general coordinates for both position and
spin, then the wave function changes sign under the exchange of any pair of coordinates Xi
and x/
For a noninteracting system of fermions, the particle states are described by single-particle
orbitals <pp(x), and the many-particle wave function in the Hartree-Fock approximation is a
Slater determinant of these orbitals. The determinant has the property that no two particles
can occupy the same single-particle orbital state, or else the wave function is trivially zero.
The nature of the ground state wave function is less clear in this strongly interacting and
highly correlated liquid. The microscopic picture of the liquid has the atoms bouncing around
inside of a small cage defined by the average position of neighboring atoms. This picture is
correct for liquid 4He and leads to a correlated wave function of the CBF form, which is
highly successful for describing bosons. It is not nearly so obvious how to construct a ground
state wave function for highly correlated fermions. There probably should be something
equivalent to the CBFs but for fermions. The question is how the fermion nature of the
particles is expressed in real space. When visualizing the atoms bouncing around, how is this
motion different when they are fermions rather than bosons? How does one construct a CBF
which has the fundamental antisymmetrization property (11.184)?
Feenberg and his associates have provided the most successful microscopic theory to
date, which compares favorably to the available experiments. Their ground state wave
function for liquid 3He has the form of an algebraic product of two familiar wave functions:
the CBF with the form used for ground state of 4He and a Slater determinant of plane-wave
states with spin:
q'iX ),X2"" ,XN) = 'I'o(r) , r2"'" rN)<l>PI""PN(X) , ... ,XN)

(11.185)
'P ,(r,. r, ..... rNJ ~ LN exp [- t; (Ir, ~ rJ]

<l>PI'''PN (Xl' ... , XN) = det I<PPI (X))<PP2 (X2) ••• I (11.186)
The CBF was taken to be the form (11.62) of a product of pairwise correlations, with a single
short-range repulsive term as in (11.77). The value of b changes slightly from the value used
in 4He, because of the mass difference. The Slater determinant is the same one which would
be used for a system of noninteracting particles and contains single-particle wave functions
<p/x) which are plane-wave states and spin functions for spin up (IX) or down (~). The Slater
determinant provides the necessary anti symmetrization, while the CBF provides the necessary
short-range correlation in the atomic motion.
This wave function is used to calculate ground state properties such as the energy
Eg = (gIHlg) / {gig}, S(k), etc. These calculations are extremely complicated and will not be
described here. Interested readers are referred to Feenberg's book (1969). All many-particle
matrix elements are evaluated by a cumulant expansion which must be carried to high order.
They involve the evaluation ofthree- and four-particle correlations in the liquid state. Some of
their results will now be summarized.
The ground state energy is discussed first. Feenberg and his associates first calculated the
ground state energy assuming 3He was a boson. That is, they evaluated Eo = {'I'oIHI'I'o}
where 'I'0 is the CBF for liquid 3He. Since this wave function is symmetric under coordinate
exchange, Eo is the energy of the equivalent boson liquid. The value of Eo differs somewhat
for the different ways of doing the cumulant expansion and estimating three- and four-particle
correlations but has the typical value of Eo = -2.9 Klatom. A boson liquid with the mass of
3He is less bound than liquid 4He, which reflects the increased zero-point motion of the
lighter mass. Although the density ofliquid 3He is lower than liquid 4He, the average value of
kinetic energy per particle is nearly the same value of 13.4 Klatom. The average value of
potential energy is -16.3 Klatom, giving the difference of -2.9 Klatom. Thus the average
kinetic energy per atom is still high, and the particles still have short-range fluctuations in
their position within the cage defined 'by the neighboring atoms.
The next calculation is to add the Slater determinant of the single-particle orbitals to the
ground state wave function and to compute the new ground state energy. It is defined as
Eg = Eo + 8E, where 8E is the change in energy due to the motion of the particles in the
Fermi sea. For a gas of noninteracting particles, this change would be 8E =
3EFO/ 5 = 0.6(4.97) = 3.0 K. Feenberg and his associates find the smaller value of 8E = 1.7
Klatom, which is a small number compared to the average kinetic energy in the boson ground
state, ~14 Klatom. The short-range motions of the particles are more important than the
kinetic energy of motion in the Fermi sea. The final prediction of the ground state energy per
atom Eg = -1.2 Klatom is only in fair agreement with the experimental value of -2.52 KI
atom. However, the least accurate theoretical number is probably Eo. There could easily be a 1
K error in the calculation of this quantity, as there was in the similar calculation for 4He,
where it was less noticeable as the difference between 6 and 7 Klatom. This number is
difficult to find accurately because of the large cancellation between the kinetic and potential
energies and their sensitivity on the trial function for the pairwise part of the CBF. The
agreement between theory and experiment, in liquid 3He, must be regarded as satisfactory.
The next ground state property is the liquid structure factor S(k). It can be measured by
both X-ray and neutron scattering. Figure 1l.12 summarizes theory and X-ray data. The
points are experimental, and the lines are the theories of Massey (1966) and Massey and Woo
(1967). The agreement is obviously excellent. Since the theories were published as predic-
tions several years before the experiments, the theoretical success is even greater. There is
great similarity between S(k) for liquid 3He and 4He. Both have peaks near k ~ 2.0 A-\ and
then rapidly approach unity with small oscillations at large values of k. The S(k) values for
3He are about 40% higher than for 4He in the low k region. The good agreement between
theory and experiment for S(k) means that similar good agreement exists between their
Fourier transforms, which are the pair distribution functions g(r).
1.5
",,~
,
'7 0. 0. 00
,.0 0000
00000.0.0 0
...
III
o ACHTER 8 MEYER
Qs.o o HALLOCK
0
O~ol -MASSEY
0,0
" 0»
~/
--- MASSEY 8 WOO
ee/
0.5 LO 1.5 2.0 2.5 3.0 3.5 4.0
.6(A-')
FIGURE 11.12 Plot of S(k) for liquid 3He. The points are the X-ray scattering data of Achter and Meyer (1969) and
Hallock (1972). The solid curves are the theories of Massey (1966) and Massey and Woo (1967). The agreement is
obviously excellent.
-
1.4
>at 1.2 • §•
E • 8 ~ ~ o 0
.......
W 1.0 •
0
8 ~
0
0.80 •0
0
0.60 ~
0.40
0.0 0.20 0.40 0.60 O.BO 1.0 1.2 1.4 1.6
FIGURE 11.13 Zero sound mode energy in liquid 3He. The points are from Skold and Pelizzari (1980). From Glyde
(1994, used with permission).
The liquid must have a long-wavelength collective oscillation. In ordinary liquids it is

just the longitudinal sound wave. In the electron gas it is at the plasma frequency ffip ' where
the charge of the electron gas makes the mode have a nonzero frequency even in the limit
where q -+ O. Since liquid 3He is electrically neutral, the long-wavelength density oscillation
will just be a sound wave, which is called the zero sound mode. Figure 11.13 shows the zero
sound mode in liquid 3He as measured by neutron scattering. The cross-hatched region in the
lower comer of the figure are the particle-hole pair excitations out of the Fermi sea, as is
found in metals. They are calculated using m* = 3m.
There are several different kinds of sound waves which can propagate in these systems.
The ordinary sound waves at very long wavelengths are called first sound. Their speed is
calculated by including the particle collisions, since such collisions are very frequent
compared to the sound frequency. This results in a hydrodynamic version of the sound
propagation. At much higher sound frequencies, the collision rate between particles becomes
small compared to the sound frequency, and the speed must be calculated in the collisionless
approximation. This regime is called zero sound and has a higher velocity.
The third excitation of liquid 3He are the spin fluctuations at low energy. These will be
treated fully in the following section.
11.2.3. Interaction Between Quasiparticles: Excitations

Fermi liquid theory is not a complete description of all the dynamical motion of
quasiparticles in liquid 3He. The theory was constructed by Landau to explain thermo-
dynamic and static measurements such as specific heat and spin susceptibility, and it does that
very well. In order to explain other measurements, it must be decided whether enough
information is available in the Fermi liquid parameters or whether additional facts are needed
about the quasiparticles. The additional measurements are transport experiments such as
thermal conductivity and viscosity. A theory is also needed for low-energy spin fluctuations
and also superfluidity. So far all that is known about the quasiparticles are their effective mass
and the four interaction parameters Fg'~. It must be decided what these parameters mean in
terms of the microscopic scattering theory, and what additional information is needed to fully
describe the interaction between quasiparticles.
It is useful to remember the main features of electrons in metals, which is the most
understood Fermi liquid. The electrons at the Fermi surface have an effective mass m* which
is comprised of several contributions. Those of the electron-electron interactions, and also the
band structure, gave a contribution to m* 1m which was not very energy dependent. However,
the contribution from electron-phonon interactions was often very large and varied strongly
with energy, so that it was negligible a Debye energy away from the Fermi surface. In liquid
3He, one should ask the question whether the ratio m* 1m ~ 3 changes rapidly or slowly as the
energy of the quasiparticles is varied. Or, to rephrase the question, what is the appropriate
energy scale which determines the variation of these parameters? Another motivation for this
question is provided by recalling the BCS equation for the transition temperature of a
superconductor:
(11.187)
The exponent has the factors N F Vo which are the density of states and the effective interaction
between quasiparticles at the Fermi surface. These two parameters can be estimated from the
parameters of Fermi liquid theory. However, the prefactor in this equation contains the Debye
energy hOOf> which is the effective energy range over which the phonon attraction between
quasiparticles is operative. This effective energy range of the interaction cannot be obtained
from the parameters of Fermi liquid theory and must be deduced from other considerations.
In elemental metals, superconductivity is usually caused by the electron-phonon inter-
action. This theory has been made very quantitative. The accuracy is helped by a number of
smallness parameters, such as cjvF' OODIEF, and mlM, in which the theoretical expressions
can be expanded. None of these apply to liquid 3He. It has Cs > VF, and the energy of high
frequency sound exceeds the Fermi energy of EF = kETF ~ 3 K = 0.26 meV In fact, no
obvious smallness parameter comes to mind. Nevertheless, there probably is one since the
ratio of TclTF ~ 10- 3 is similar between metals and liquid 3He.
The scattering of two quasiparticles can be described in the following manner. Two
quasiparticles initially have momenta PI and P2. They interact and exchange momentum q
and end up in the states P3 = PI + q and P4 = P2 - q. The scattering of the quasiparticles is
described by a phenomenological matrix element Mcrcr,(PI' P2; P3' P4). The spin indices are
included, since the effective scattering is spin dependent. The goal is to obtain a quantitative
description of this matrix element. It is defined to include exchange events, so it describes the
direct scattering plus the exchange scattering.
The first step is to find the relationship between this matrix element and the Fermi liquid
parameters. This identification is found by an intuitive argument. In a weakly interacting
Fermi gas, the Hartree energy is given by
(11.188)
The self-energy diagram for LH has a particle of momentum P scatter from the other particles
with momentum pi and density nF(;p'). No momentum is exchanged, so the self-energy is
expressed in terms of the potential V(O) at zero-momentum transfer (q = 0), which is
equivalent to the forward scattering amplitude for the two particles P and p'. The exchange
self-energy 'Lx(P) just provides the exchange corrections to this forward scattering. Applying
this theory to calculating the self-energies of quasiparticles in liquid 3He gives
1
'LH + 'Lx = - L Mcrcr,(p, pi; p, p')np'cr' (11.189)
v p'c;'
The potential V(O) has been replaced by the forward scattering matrix element
Mc;cr'(p, pi; p, pi), and the distribution of quasiparticles is written as np'cr" When an excitation
is created, the change in this self-energy function due to the excitation is
! L Mcrcr,(p, pi; p, p')onp'cr' (11.190)

v p'cr'
This expression is very familiar, since it is exactly the form of the interaction energy in Fermi
liquid theory, where /pcr,P'cr' = Mcrcr,(p, pi; p, pi). The Fermi liquid parameters represent the
forward scattering amplitude between bare quasiparticles. It will be necessary to obtain the
scattering amplitude for other directions besides forward and also to have it for dressed
quasiparticles of energy Ep rather than bare ones of energy cp .
The objective is to have a model for the scattering of two quasiparticles. The parameters
of the model are chosen to reproduce the Fermi liquid theory for scattering in the forward
direction. One method is to write an effective quasiparticle Hamiltonian with creation and
destruction operators:
(11.191)
The matrix element M is different from Mcrcr" because M(q) does not include exchange
events. In the theory of the homogeneous electron gas, which has a similar Hamiltonian, the
interactions between particles led to a screening of the interactions. The word screening
sometimes implies a charge redistribution, so the alternate word dressing is used in liquid
3He. But the physics of the two concepts is the same: the quasiparticles rearrange all the other
quasiparticles in their vicinity to alter the effective interaction between any pair of them. In
the random phase approximation, this effective interaction is
M () M(q) (11.192)
eff q, (0 = 1 _ M(q)P(1)(q, (0)
where p(1) is the simple bubble contribution for the polarization diagram. The denominator
has the dressing function 1 - MP(l), which is equivalent to the dielectric screening function in
the electron gas. the forward direction is obtained by first taking (0 --+ 0, and afterwards
q --+ 0:
r [~~q~=-;[;N F
(11.193)
O),~~o N M = NFM«}) = --.!L = AS
F eff 1 +NFM(O) 1 +Fa - 0
The effective interaction Fa is replaced by Aij. The quantity Fa is the s-wave part of the
interaction between bare quasiparticles, while Ao
is the same interaction between dressed
quasiparticles. This difference is the same as between the bare Coulomb interaction e2 /r and
the screened interaction. The dressed interaction between quasiparticles is changed, from the
bare interaction, by the other quasiparticles which have also been influenced to alter their
equilibrium configuration by the effective interaction between quasiparticles.
Since the static limit of (11.193) has been obtained, the next step is to examine the
dynamical predictions of this model. Consider the case whereby 00 » qVF' For the electron
gas in this same limit recall that
nq2
p(1)(q,oo) = - -
m*oo2
F.s k 2q2 dq2
M{q)P(1)(q,oo) = ~_F_ == _1- (11.194)
3 m*2oo2 002
2 FOkj. vJ..o Fo
3 1 + Fi 13
C ----
1 - 3m*2 -
This result may be used to obtain the dressed susceptibility, which is the correlation function
for density-density operators:
(11.195)
(11.196)
There is a resonance at 00 ~ cl q. The large value of Fo ensures that cl > vF so 00 > vFq is valid
when 00 ~ clq.
Table 11.2 shows that the exact sound velocity in the long-wavelength limit is
2 = (vJ..o/3)(1 + Fo)/(1 + FI/3). The approximate result in (11.196) is close to the exact c,
since Fo is large and there is not much difference between Fo and 1 + Fo. The factor of
(1 + F1/3) comes from the effective mass. The exact density-density correlation function has
the exact same form as (11.196) but with the correct speed of sound c rather than the
approximate result CI' The difference between these two calculations is that the long-wave-
length sound velocity C is found in the hydrodynamic regime and is the first sound. The above
calculation is for the collisionless regime, and CI is the velocity of zero sound. They approach
the same value for large Fo. The density-density correlation function has the long-wavelength
limit of sound waves, which dominates the excitation spectrum.
Define Apcr,p'cr' as the Fermi liquid theory function which gives the interaction between
dressed quasiparticles. It is also the forward scattering amplitude of the screened matrix
element for the interaction between two quasiparticles. It may be derived from the bare
interaction between quasiparticles/pcr,p'cr' by the following simple argument. The self-energy
of a quasiparticle may be expressed as either the product of the bare interaction/pcr,p'cr" and
the bare additional density on pcr' or else the product of the dressed interaction Apcr,p'cr' and the
dressed additional density oiipcr:
(11.197)
The interaction between dressed quasiparticles Apcr,p'cr' is also defined only right on the Fermi
surface, where both p and p' are equal to kF . All quantities are expanded in spherical
harmonics, and doing the angular integrals gives the following obvious formula relating these
coefficients:
1
Apcr,p'cr' = N- ~]Al + 0'. O"Af]PI(cos 9) (11.198)
F I
A:,aoii:: = Fi,aon:: (11.199)
Using the previous relationship (11.180) between oii:: and on::, the final formula is deduced
between Al and FI
s Fi (11.200)
Al = 1 + Ff/(21 + 1)
a Fi (11.201)
Al = 1 + Fi /(21 + 1)
The case 1 = 0 agrees with the RPA result obtained in (11.193). These numbers are tabulated
in Table 11.3. The effective interaction between dressed quasiparticles is qualitatively much
Fo
different than the interaction between bare quasiparticles. The very large value of makes Ao
less than unity, so the dressing effects have a drastic effect on reducing the effective inter-
action. Of course, the same thing happens for electron-electron interactions in metals, where
screening makes a similar reduction. But the most interesting effect is the increase in the spin-
dependent parts Af. These become larger in magnitude than Fi
because the latter are negative.
This enhancement of the spin-dependent interaction is one step in the argument, which is
given later, toward the conclusion that superfluidity in liquid 3He is caused by triplet pairing.
The next question is the method of calculating the spin susceptibility. In (11.194) and
(11.196), RPA and Fermi liquid parameters were used to derive an approximate formula for
the density-density correlation function. Is it possible to do the same for the spin suscept-
ibility? By reasoning deductively, Leggett (1965) suggested that a possible form for the spin
susceptibility in the RPA is
p(1)(q, (0)
(11.202)
o
X.(q, (0) = 1 - (F /NF )P(1)(q, (0)
This equation is deduced, in analogy with (11.192) and (11.193), by just replacing Fo Fo.
by
A better treatment would start from a model Hamiltonian which would give this result for the
correlation function. For the moment, accept (11.202) as a reasonable hypothesis and examine
its dynamical predictions.
The spectral function for this retarded correlation function is called Siq, (0). In analogy
with the electron gas, two types of dynamical modes are expected from this correlation
function. The first is a collective mode equivalent to the plasmon in the electron gas and the
sound waves in the density-density correlation function. Such modes can exist only when
00 » qVF' The theory has already been done for the sound case, in (11.194) and (11.196), so
change Fo Fo
to in these equations. The prediction is that the collective spin sound has the
eigenfrequency
oo=qVF~ (11.203)
This frequency is imaginary, since Fo is negative. These modes are totally damped and do not
exist as excitations.
The other excitation of the fermion gas is the particle-hole pairs. The theory of spin
pairs can be developed in an analogous fashion. Consider the case where 00 <qvF and q<kF •
Then the polarization function p(I) can be approximated for its real and imaginary parts by
p(l)(q, (0) = -NF [ 1 - i; C:J0(qV F - (0)]

(11.204)
Ss(q, (0) = -2 Im[xsCq, (0)] = As
00
2
00
+ ooa
0(qVF - (0)
2
(11.205)
(11.206)
There are no collective modes since the pole in the denominator is found at imaginary
frequency. The spectral function is linear in energy for small values of E = noo. The same
linear behavior was found for the electron gas when the pair contribution to Im[ljs(q, (0)] was
evaluated, as in Fig. 5.12. The same theory can be applied to the pair spectrum from the
density-density correlation function, and the only difference is in replacing by and AD Fo Fo
by Ao. It is instructive to compare these two cases when pairs are made by either the spin or
density response functions. The energy widths ooa and oos are similar for the two cases, since
AD and Ao differ only by a factor of 2. However, the significant difference is in the coupling
strength As,a' The factor 1F~,a(1 + F~,a)rl is quite different for the two cases, since it is 4.5
for spins and 0.009 for density, which differ by a factor of 500. The RPA predicts that the
spectral function for making pairs from the spin correlation function is 103 larger than for the
density correlation function.
The correlation function (11.202) must be regarded as ad hoc without a derivation based
on a realistic model Hamiltonian for the interaction between quasiparticles. The only attempt
in this direction has been using paramagnon theory, which is briefly described. This theory
starts with a model Hamiltonian for quasiparticles of the form (11.191), with the matrix
element if only scattering particles of opposite spin. This matrix element is taken to be
independent of momentum and is denoted by the symbol I:
V= I Jnt (r)nj, (r)d 3 r (11.207)
-_ I L Ct
-Vkk'q t
k +qt C kt Cp_q~ Cp~ (11.208)
This interaction term was first suggested by Berk and Schrleffer (1966) in their discussion of
the superconductivity effects in transition metals, such as Pd, which are nearly ferromagnetic.
They are actually paramagnetic, with a large spin susceptibility and a low transition
temperature for superconductivity. They reasoned that the spin arrangements in Pd have short-
range correlation, whereby if any spin was pointing in a direction, then nearby spins were
highly correlated and were likely pointing in the same direction. The spin arrangements were
ordered locally, although for short durations, and these spin fluctuations are called para-
magnons. They proposed that these spin fluctuations suppressed the onset of super-
conductivity, since an electron with the spin pointing in the opposite direction would find it
hard to be in the same vicinity because of Coulomb repulsion. They constructed this ad hoc
Hamiltonian, which had a point repulsion (I> 0) between particles of opposite spin.
Sec. 11.2 • LIQUID 3 He 729
This idea was adopted immediately for liquid 3He by Doniach and Engelsberg (1966)
and Rice (1967). They used it to calculate a number of properties ofliquid 3He. Levin and
Valls (1978) show that paramagnon theory gives an accurate quantitative theory of dressed
quasiparticle interaction for both superfluidity and quasiparticle transport.
An alert reader will notice that the above Hamiltonian is identical to the Hubbard model
which is discussed in Chapter 6. Of course, the Hubbard model was for a lattice, while here it
is for a liquid. In the wave vector representation, the two theories are identical. The reader can
refer back to Chapter 6 for the predictions of this model Hamiltonian.
11.2.4. Quasiparticle Transport

Fermi liquid theory can also be used to describe the transport of quasiparticles at very
low temperatures. The agreement between theory and experiment provides a further verifi-
cation of Fermi liquid theory and the derived parameters. The theoretical framework was
provided by Abrikosov and Khalatnikov (1959), who presented an approximate theory of
viscosity 11 and thermal conductivity K. The third transport coefficient is spin diffusion D, and
an approximate theory was first given by Hone (1961). All these theories involve the solution
of the Boltzmann equation for transport in Fermi systems, where the scattering mechanism is
particle-particle interactions. In this case the particles are quasiparticles. Later there was an
exact solution of the Boltzmann equation in the limit where T ---+ 0 by Jensen et at. (1968)
and Brooker and Sykes (1968). It would take us too far afield to solve the Boltzmann equation
for these three transport coefficients, so the results are just quoted. The following formulas are
exact in the limit of zero temperature:
K = 8n 2 kj. [(W(8, <1»(1 - COS8»)]-IH(1e )

3T(m*)4 cos(8/2) K
= 64kfr [(W(8,<I»(1-COS8)2sin2<1»]-IC(Ie)
11 45(m*)\kB Ti cos(8/2) 11
D = 32n2k~(1 + Fa) [(WH (8, <1»(1 - cos 8)(1 - cos <1»)]-1 C Ie

3(m*)\kBT)2 cos(8/2) ( D)
3 - Ie 00 4n + 5
H(Ie) = -4-fo(n + 1)(2n + 3)[(n + 1)(2n + 3) - Ie]
(11.209)
1 - Ie 4n + 3
00
C(1e) = -4-Io(n + 1)(2n + 1)[(n + 1)(2n + 1) - Ie]
Ie = 3 _ 2 (W(8, <1»(1 - cos 8)/ cos(8/2)}

K (W(8, <1»/ cos(8j2)}
Ie = 1 _ 3 (W(8, <1»(1 - cos 8)2 sin2 <1>/ cos(8/2)}

11 4 (W(8,<I»/cos(8/2)}
1 (Wn J8, <1»(1- cos 8)2 sin2 <1>/ cos(8/2)}

Ie = 1-
D 2 (W(8,<I»/cos(8/2)}
The functions H(A) and C(A) are correction factors. The original solution of Abrikosov,
Khalatnikov, and Hone had an approximate result with these quantities set equal to unity. In
the exact solution, one also has to evaluate these correction terms. The first step is to find the
appropriate A and then to use it in the series for H(A) or C(A). Both series converge rapidly, so
that the answer is obtained easily. The factor C turns out to be about 0.8, so that the older
theories are reduced by about 20%. However, the correction for the thermal conductivity is
much larger, since H(AK) is almost exactly 0.5, so the older expression erred by a factor of2.
These correction factors, resulting from the exact solution, play an important role in
improving the agreement between theory and experiment.
These formulas contain one or several brackets which have a factor wee, <j» plus some
angular terms. This average is taken over the 41t solid angle:
( W 1cos(9/2)
- cos 9) 1 J21t J1t 1 - cos 9 (11.210)
= 41t 0 d<j> 0 d9sin(9)W(9, <j» cos(e/2)
The angles 9 and <j> are explained later. The symbol W(9, <j» is the matrix element for the
scattering of quasiparticles. It is conventionally defined as
(11.211)
where Att and An- are the matrix elements for the scattering of two spin-up particles, and a
spin-up with a spin-down particle. These matrix elements are given by the Fermi liquid
parameters. The factor (!) in front of IAtt 12 occurs because the scattering of identical
quasiparticles into (k, -k) is indistinguishable from (-k, k), and the angular average counts
both of these and so overcounts the scattering by a factor of 2.
The factors in the numerators of (11.209) come from the effective lifetime appropriate
to the transport coefficient. For the thermal conductivity, the effective lifetime contains the
factor (1 - cos 9), just as it does for the electrical conductivity in metals. The other factor in
this angular average is 1/ cos(9/2), whose origin is now described. First examine the
expression for the lifetime of a quasiparticle as it scatters from the other quasiparticles.
Assume that the initial quasiparticle has momentum PI' that it scatters from a quasiparticle P2'
and that they go to P3 and P4' The lifetime for this scattering rate is written in a symmetical
form
1 21t Jd3P2d3P3d3P4 2 3
'(PI) =h (21t)9 ~ lAta' I (21t) O(PI + P2 - P3 - P4)
X O(EI + E2 - E3 - E4)n2(1 - n3)(1 - n4) (11.212)
The occupation factors nj = nF(E) express the feature that P2 is in an occupied state, while P3
and P4 must go to empty ones. The integrand contains the delta functions for conservation of
energy and momentum, and the matrix element lAta' 12 for scattering (but IAtt 12 is divided by
twO). The evaluation of this quantity follows Pines and Nozieres (1966).
At very low temperature, all the quasiparticles will be quite close to the Fermi surface.
The magnitude of all the momentums p/i = 1,2,3,4) are very near to kF and therefore very
nearly the same length. This feature simplifies the calculation. The delta function for
momentum conservation is used to eliminate the d 3p4 integral and leaves the combination
d3p2d3p3' Let P = PI + P2 be the center of mass momentum of the two quasiparticles, which
Sec.11.2 • LlaUID 3He 731
is unchanged by the collision. The d 3p3 integral is nested inside d 3pz so that the vector Pis
used as the basis of choosing the angles in doing the d 3p3 integral. In particular, let
P . P3 = PP3 cos(8 3), cos(83) == v3' Then the vector P4 has the magnitude
P4 = IP - P31 = jPz + p~ - 2PP3 v3 (11.213)
This identity is used to change the variables of integration from d8 3 sin(83) = dV3 to dP4'
wherep3dv3 = -P4dp4/P, The next step is to define P in terms of the variables used in the
d 3pz integral. Use 8 to denote the angle between PI and pz and use the fact that both are near
kF to find P = IpI + Pzl = Jpf + p~ + 2PIPz cos(8) ~ kFJ2(1 + cos 8) = 2kF cos(8/2).
Finally, we set p~dpz ~ kFd(PD/2 which brings us to the identity
3 3 * 3 sin(8)d8
d pzd P3 = (m ) 21 cos(8/2)1 d(pzd(P3dEZdE3dE4 (11.214)
The six integration variables in d 3p zd 3p3 have been reexpressed in terms of coordinates which
are more useful. It is usually assumed that the matrix element has no significant energy
variation. Then the energy integrals can be done exactly. The first one is trivial, since it just
uses the delta function for energy conservation. The remaining two integrals can also be done
exactly, which is assisted by the indicated variable changes:
1= JdEZdE3dE48(EI + E2 - E3 - E4)nF(Ez)nF( -E3)nF( -E4) (11.215)
1= JdEzdE3nF(Ez)nF( -E3)nF(E3 - EI - EZ) (11.216)
x = e~(e2-~), y = e~(e3-~), z = e-~(el-~) (11.217)
1= (kBTi Jooo dx foo dy.,.------,-,--,--------.,.-:-:----:-

0
1
(x + 1)(y + )(x + yz)1
(11.218)
= (kBT)z Joo ~ In(yz) (11.219)

o y + lyz - 1
1
= -z (rcz + lnz z) =!(1 - nl)[rcZ(kBT)z + (EI - z
Jl) ] (11.220)
2~ z+ 1
The last integral is from G&R, 4.232(3). The form of the answer is very interesting. The result
is proportional to two terms: one is (rckBT)Z and the other is (EI - Jlf
The latter vanishes at
the chemical potential EI = Jl and the former vanishes at zero temperature. This form of the
answer is found in particle-particle scattering of any like fermions. The same factor enters
into the expression for the inverse lifetime of an electron in a metal, from electron-electron
scattering.
The last step is to do the angular integrals. First it is helpful to visualize the coordinate
system. Take the case where all four vectors Pj have the same length kF . Define 8 as the angle
between PI and Pz' It is also the angle between P3 and P4' which is true only when all four
vectors are the same length. Only one more angle is needed to define the relative orientation
of all four vectors. This angle could be the relative orientation of PI and P3' but this choice is
not conventional. Instead, note that the two initial vectors PI and pz form a plane whose
orientation is represented by a unit vector ni in the direction PI x Pz. Similarly, the scattering
plane in the final state has an orientation nf in the direction P3 X P4' Define (p as the angle
between these unit vectors, cos (p = ni' nf. The angle (p is shown in the vector diagram in Fig.
FIGURE 11.14 Angular variables in quasi particle scattering in liquid 3He.
11.14. All four vectors Pj make an angle 9/2 with respect to the center of mass momenta P,
and is the angle between the two scattering planes.
In the variable list in (11.214), the azimuthal angle <1>3 is actually <1>, while <1>2 is
unnecessary and may be integrated to give 2n. The description of the angles in the scattering
process is now complete . The final lifetime is the following expression from Pines and
Nozieres (1966):
1 (m*)3 2 2 2( W )
-r(p\) = 16n4h6 (1 - n\)[n (kBT) + (E\ - ~) ] cos(9/2) (11.221)
The symbol W(9, <1» has been given its definition in (11.211), where all the angular variables
have been defined explicitly. The result for l/-r(p\) has two terms which show that at the
chemical potential E\ = ~ the lifetime is determined by the temperature, but at high energies
~(E\ - ~) » 1 it is determined by phase space availability. The equivalent quantity using
Green's functions is -2 Im[~(p\, ~\)] for the appropriate diagram. The imaginary part of the
self-energy produces the same lifetime if the diagrams are selected correctly, except the
(1 - n\) factor is absent (see Problem 7 at the end of this chapter). The factor (1 - n\) should
not be included in the quasiparticle lifetime.
The evaluation of the angular integrals for the quasiparticle lifetime require an expres-
sion for W(9, <1». The Landau-Fermi liquid theory gives only the forward part of the scat-
tering amplitude, where P3 = p\ ' P4 = P2' or the opposite. Forward scattering corresponds to
the case where = 0 or n. A method is needed of extending this scattering theory to other
values of angle besides the forward direction. The extension to other angles is useful not only
for this calculation of the quasiparticle lifetime but also for the similar calculations of the
transport coefficients in (11.209) which have other angular averages of W(9, <1». In transport
theory, the quasiparticles not only scatter in the forward direction but to any other point on the
Fermi surface. How can this be described by using only known parameters?
An ingenious and very successful method for this was proposed by Dy and Pethick
(1969), which they called the s-p approximation. The goal is to obtain the scattering
amplitudes Att(e, <1» and At{, (e, <1» for the scattering of two spin-up particles and a spin-up
with a spin-down particle. If two particles both have spin up, they must be in a relative triplet
spin state, and this scattering amplitude is labeled At(e, <1». The singlet scattering amplitude is
called As(e, <1». Two particles in opposite spin states have equal likelihood of being in a
singlet or triplet state:
Att(e, <1» = AH(e, <1» = At(e, <1» (11.222)

At{,(e, <1» = A H (9, <1» = HAt(e, <1» +A.(e, <1»] (11.223)
The next step is to use the symmetry of the two-particle wave function to deduce the possible
angular variations. In center of mass coordinates, the only relevant angular variable in the
Sec. 11.2 • LIQUID 3 He 733
scattering process is that between the initial and final relative momenta, which is <1>. Singlet
states must be symmetric under the exchange of the orbital part of the particle coordinates, so
only even values of relative angular momentum are allowed, as explained in Sec. 11.1.
Similarly, triplet spin states permit only odd values of relative angular momentum, so that the
symmetry of the two-particle wave functions dictates the following possible choices for these
scattering amplitudes:
AI(8, <1» = L C/(8)P/(cos<l» (11.224)

/ odd
AsC8, <1» = L C[(8)PJCcos<l» (11.225)

/ even
The next step is the s-p approximation. Dy and Pethick argue that since only the I = 0 and 1
values are known in the spherical harmonic expansion for 8, consistency dictates that only
similar terms are retained in the expansion for Legendre functions on the variable. In this
case there is only one term allowed in each expansion, which is I = 0 for singlet states and
I = 1 for triplet states:
Al8, <1» = CI(8)cos(<I» (11.226)

AsC8, <1» = Cb(8) (11.227)
The next step in the derivation is to find the amplitudes C I (8) and q(8) by evaluating them in
the forward direction, where the answer is known. Set = 0 and equate these coefficients to
the interaction term between dressed quasiparticles:
1 I
C I (8) = AI(8, 0) = - L(A, + A't)P/(cos 8) (11.228)
NF/=o
1 I
Cb(8) = AsC8, 0) = - L(AI - 3A't)P/(cos 8) (11.229)
NF/=o
These various steps are collected together to finally provide the angular variation for the
scattering amplitude between dressed quasiparticles:
Att(8, <1» = C~F [(A(j + Ag) + (Al + Ai) cos 8] (11.230)
AH(8, <1» = 2~F {cos [(A(j +Ag) + (Al + Ai) cos 8]
+ [(A(j - 3Ag) + (Al - 3A~) cos 8]} (11.231)
In older calculations of the transport coefficients, the angular functions Acrcr , (8, <1» were
approximated by the results in the forward direction = 0, since they are given by the
Landau Fermi liquid parameters. The improvement of Dy and Pethick is to add the factor
cos to the amplitude for the triplet scattering. The factor of cos is a necessary addition
from the point of symmetry, since the triplet amplitude must change sign when = 1t,
because of the wave function anti symmetry. Going from = 0 to = 1t effectively changes
the scattering (PI ---+ P3' P2 ---+ P4) to the exchange event (PI ---+ P4, P2 ---+ P3). This small
change of adding cos is a big one insofar as improving the agreement between theory and
experiment for the transport coefficients. Table 11.4 shows their calculation of the quantities
KT, DT2, and 11 T2, as compared with the experimental values, using the theoretical formulas
in (11.209). The agreement is obviously excellent. Previous calculations without the cos 
TABLE 11.4 3He transport results as T ~ 0
KT (erg/cm-s)
Experiment" 35 l.4x 10-6 1.8xlO-6

Theoryh 33 1.6x 10-6 1.6x 10-6
a. Wheatley (1975).
b. Dy and Pethick (1969).
factor in At disagreed with experiment by a factor of two. Dy and Pethick also predicted the
transport results for higher temperatures.
Another important success of the s-p approximation has been to explain the superfluid
properties of liquid 3He. In particular, it explains why the BCS type of pairing is not in the
singlet state but rather is in the spin triplet state. Here the discussion follows Patton and
Zaringhalam (1975). They calculated the temperature at which the Cooper instability occurs
in liquid 3He, which is a property of the normal state of the liquid. The construction of a wave
function for the superfluid state is a different step, which is done in the next section. But the
Cooper instability is a property of the normal fluid and is discussed here.
In a BCS of pairing state, two dressed quasiparticles of momentum p and -p are
coupled into a collective bound state. One important question is whether this pairing occurs
with the two spins in a relative singlet or triplet state. To answer this question, examine the
effective interaction for each configuration. Consider the scattering of a bound pair with
e
(p, -p) into another pair state (pi, _pi). The two scattering particles have = n, so that the
relevant amplitudes are
AtCn, <1» = C1(n)p· ji (11.232)

AsCn, <1» = C~(n) (11.233)
The Fermi liquid parameters are used to derive C 1(n) and C~(n) from (11.229):
NFC1(n) = -2.57 (11.234)

NFC~(n) = 3.35 (11.235)
The singlet state amplitude CO(n) is positive. Two dressed quasiparticles have a repulsive
interaction when paired in the spin singlet state. They do not attract each other and do not
form bound states. The superfluid state is not a spin singlet in liquid 3He.
For the spin triplet state, C 1(n) is negative, which shows that the two quasiparticles have
an attractive interaction. Pairing is possible in this state, and the normal fluid will show a
Cooper instability at a nonzero temperature because of this tendency toward triplet pairing.
The prediction of Fermi liquid theory is that liquid 3He has to form a BCS pairing state
in the spin triplet arrangement. This prediction seems in good accord with the experiments on
the superfluid state, which are described in the next section. The triplet pairing is a conse-
quence of the negative value of C 1(n) and a positive value of CO (n). The physics question is:
Why does this happen? The answer seems poorly understood. The parameters C 1 and CO are
derived from the interaction between dressed quasiparticles, which in turn are derived from
the interaction between bare quasiparticles. The latter numbers are taken from experiment,
although the microscopic theory for them gives good results. However, the chain of argument
starting at C 1 and C~ runs back to other parameters (Ff'a ) which are hard to calculate. It is
difficult to have any intuitive insight into any of these parameters.
11.2.5. Superfluid 3He

Superfluidity in liquid 3He was discovered by Osheroff et al. (1972), who discussed it in
a series of papers. Their Nobel Lectures are a nice review: Lee et al. (1997). They identified
two phases in the liquid, which are called the A and B phases. The phase boundary on a
pressure vs. temperature curve is shown in Fig. 11.15 as measured by Greywall (1986). The
line marked Tc separates normal liquid 3He from the superfluid phases. It is a second-order
phase boundary with a jump in the specific heat, which are both characteristics of a BCS state.
Both A and B phases are superfluid and are experimentally different. The line marked TAB is a
first-order phase boundary. The location of this phase boundary is strongly affected by small
magnetic fields. Indeed, both the A and B phases have interesting magnetic properties which
reinforce the notion that pairing is in a spin triplet state and that the pairs have a net magnetic
moment. The theory of triplet pairing is given below.
The theory of superconductivity for electron spins in a triplet state was derived by Balian
and Werthamer (1963). They developed this theory in the expectation that some metals might
be superconducting with this spin configuration. The first important application of their result
is in explaining superfluidity in liquid 3He. Their theory is similar to the BCS theory, with
weak coupling between the pairs of fermions, except that the spins are assumed to be in a
relative triplet state. However, triplet pairing makes a nontrivial change in the nature of the
answer because ofthe additional spin degrees of freedom. The spin state (S = 1, ms = 1, 0, -1)
can project in different ways on the relative orbital motion (L = 1, m[ = 1, 0, -1). The paired
states have a number of possible values of total angular momentum J = L + S. The gap
equation is a matrix equation. It has a number of possible solutions, and the order parameter is
actually a spinor of dimensionality 7. This feature leads to a rich description of the types of
motion and ordering of the superfluid. All this complexity is just to describe one of the two
phases. The theory of Balian and Werthamer is believed to describe the B phase of the
superfluid. The A phase is another type of triplet pairing, which is treated later.
40
......
solid
35
-
as
.a 30
Do
25
superflljld
20
a·phase
15
10
5 normal Fermi liquid
0
0 0.5 1.5 2 2.5 3
T(mk)
FIGURE 11.15 Phase diagram of 3 He as a function of pressure and temperature. The line Tc divides normal from
superfluid phases. The line TAB divides the two superfluid phases, and the horizontal line at 34 bars divides solid from
liquid. Source: Greywall (1986) (used with permission)
The BW (Balian and Werthamer) theory is solved by following the same steps used to
derive the BCS theory in Chapter 10. The first step is to write an effective Hamiltonian
between dressed quasiparticles:
t I t t
H =L Lq crcr'
+ -2v pp'
~pCpcrCpcr L V(q)Cp+q,crCP'_q,cr'Cp'cr'Cpcr (11.236)
pcr
The interaction potential V(q) is actually a function of the momentum change between
relative wave vectors Pi and Pj in the scattering process, where Pj = Pi + q. The interaction
potential is expanded in spherical harmonics, and the first two terms are
(11.237)
The first term Vo acts when the orbital motion of the two particles is in a relative s state, which
happens for a spin singlet. The discussion of the previous section showed the s-wave inter-
action is repulsive (Vo > 0) and cannot cause bound states. The second term Vi)i" h
is for
relative p states and applies for spin triplets. This coefficient is negative and can cause pairing,
where VI ~ C I (n) in the notation of the previous section. This term is the only one that is
kept in the interaction potential. It has the feature that it is anti symmetric in either of the
momentum variables: V(Pi' Pj) = Vi)i"h = -V(-Pi' Pj) = -V(Pi' -Pj)' In the case of
weak coupling theory, such as BCS, this potential is assumed to be a constant VI in energy up
to a cutoff roc from the chemical potential. The physical nature of this cutoff is not clear.
Obviously, a better theory is impossible without a better understanding of the retarded nature
of the potentials between dressed or bare quasiparticles.
The next step is to define the correlation functions appropriate to the superfluid. They
will be of the same type we used in the BCS theory of superconductivity, One expects
superfluidity in liquid 3He to be similar to a BCS state because of the obvious experimental
similarities. The theory requires the introduction of seven correlation functions:
=
~(p, , - ,I) -(TTCpcr(,)C~cr(,I))
ffl(p" - ,I) = (TTC-Pt(')Cpt("))
ffo(p" - ,I) = (TTCp,J-(')Cpt("))
= (TTC-P,J-(')Cp,J-("))
ff -I(P,' - ,I) (11.238)
ff!(p" - ,I) = (TTC!t(,)C~Pt(,I))
ff6(P,' - ,I) = (TTC!t(,)C~p,J-(,I))
ff~l(p" - ,I) = (TTC!,J-(,)C~p,J-(,I))
The quasiparticle Green's function ~ has the usual definition, although it will have a different
form in the superfluid state. The others are pairing functions for different spin arrangements.
The first one has both spins up and belongs to the triplet state. The next has one spin up and
another down, and appears identical to the correlation function used for the BCS singlet state.
However, now it is for the spin triplet state and represents the configuration where
S = 1, ms = 0, whereas in Chapter 10 it was the correlation function for S = 0, ms = O. The
difference between these two correlation functions becomes clearer when they are written as
ffS=I,m=O(P,' - ,I) = -![(TTCpt(,)Cp,J-(,I)) + (TTC-p,J-(')Cpt("))]

(11.239)
ffS=O,m=O(P" - ,I) = -![(TTCPt(,)Cp,J-(,I)) - (TTC-P,J-(')Cpt("))]
The first one is for the spin triplet state and the second one for the spin singlet. The triplet
state is anti symmetric in momentum, while the singlet state is symmetric:
~ S=l,m=O( -p, , - ,') = -~ S=l,m=O(P,' - ,')

(11.240)
~S=O,m=O( -P, , - ,') = ~S=O,m=O(P, , - ,')
These parity relations are easily proved from the definitions (11.239), by changing the sign on
P and commuting the operators. The simpler definition in (11.238) can be used for either
triplet or singlet with the appropriate choice of symmetry (11.240). The simple definition is
used, and only the anti symmetric version is used when solving the equations self-consistently
for the correlation functions of the triplet state.
There are four possible spin configurations for the two spin one-half quasiparticles: three
triplets and one singlet. The most general description would have four correlation functions
~s m (one for each spin arrangement) plus t§, and the gap equation becomes a 5 x 5 matrix
equ~tion. The problem can be simplified, at the outset, to a matrix equation which is only
3 x 3. The simplification is achieved by using the relationships in the previous paragraph,
where ms = -1 is the Hermitian conjugate of ms = 1 and only one ms = 0 function is used
with momentum anti symmetry. The simplification to a matrix of dimension 3 is obviously
desirable and is conventional.
The equations of motion method is used to obtain the gap equation in the weak coupling
theory. The first step is to recall (10.36) for acpcr/at, which is then put into the, derivative of
the Green's functions:
(11.241)
(11.242)
The interaction term has an expectation of four operators. These are paired in different
arrangements, which produces a large number of terms, since pairings are allowed between
like and unlike operators (i.e., (CC), (ctct), (CCt )) with parallel and antiparallel spin
arrangements. In weak coupling theory all terms are ignored except those which contain a gap
function ~ m' In the preceding equation, the terms which are retained are those with the
pairing (C;'_q,cr,C~cr}(Cp',cr,Cp_qcr). There are two spin arrangements, cr' = cr and cr' = -cr,
which gives the first equation of motion
-(:, + ~p )t§(P,' - ,') - ~~r(p" - ,') ~ V(q)~l(P - q, 0)
1
- -~6(P' t - ,') L V(q)~o(p - q, 0) = oCt - ,') (11.243)
v q
Equations of motion are also needed for ffJ(p, ,) and ffo(p, c). They can both be done at the
same time by considering
ff a(P,' - ,') = (T,C_pa(')Cpt (")) (11.244)
( u,~, + ~p)ff a(P,' - ,') + ~V p'qa'

L V(q)
(11.245)
where cr = ± 1 gives the two gap functions. In the bracket of four operators, pair the first two
operators to get a Green's function '§( -p, , - ,') = '§(p, , - ,I), which sets cr = crt. The
other two operators are paired into a gap function. The pairings which lead to Hartree and
exchange energies such as (C-pa Cp'a' ) (C~, -qa' Cp- qt ) are omitted in weak coupling theory.
The pairing terms which are kept give the following two equations of motion:
(~
a~
+ ~p)ffJ(P" - ,') - ~'§(p"
v
- ,') L V(q)ffJ(p - q, 0)
q
=0
(~
a~
+ ~p)ffo(P" - ,') - ~'§(p"
v
- ,') L V(q)ffo(p - q, 0)
q
=0
These are now Fourier-transformed in the usual way to obtain functions of Matsubara
frequency. Then they become algebraic equations which can be solved. Important terms in
these equations are the gap functions, which are defined as
(11.246)
The equations of motion are
(ip - ~p)'§(p) + ff1 (P)Li J + ff6(P)Lio = 1 (11.247)

(ip + ~p)ff m(P) + '§(P)Li m = 0 (11.248)
(ip + ~p)ff~(P) + ,§(P)Li~ = 0 (11.249)
The equation is also given for the Hermitian conjugate ff~. Then the solution is straight-
forward:
E(p) =~; + ILi J(p)1 2 + ILio(p)12 (11.250)

'§(P) = _ ip + ~p (11.251)
p~ +E(pi
fft (P) = ff (P) = Lim (11.252)
m m p~+E(p)2
The effective gap function is Li(p) = JlLiJI2 + ILiol2 and the energy gap Eg = 2Li. In some
respects, the BW state appears to be similar to a BCS singlet state but with an effective energy
gap of Li(p) and excitation energy E(p). This similarity does not extend to magnetic
phenomena, such as spin susceptibility, which are much more interesting in triplet pairing.
The last step in deriving the self-consistent theory is to obtain the equation for the gap
function. The correlation functions JT~(p, 't) are found for 't = 0, with the familiar result
=t (
:#' m p, 't -
_ 0) _!~ '"
-
=t ( .) _
~ ~ m p,lp -
Am(p)tanh[~E(p)/2]
2E(p) (11.253)
A ( )= -
m p
fd 3p ' V(
(21t)3 p, P
') Am(P') tanh[~E(p')/2]
2E(p')
(11.254)
In the second equation, the definition (11.246) is used to derive the self-consistent equation
for the gap function. This gap function must be solved for the energy gaps Am(P)
The gap equation is obviously a nonlinear matrix equation. Each term in the matrix
depends on the excitation energy E(p), which depends on the two energy gaps Ao and AI' The
above is a set of coupled equations for Ao and AI' There is also the restriction that the gap
function is an odd function of momentum, Am(-P) = -Am(P), which follows from the
condition that JT m(P) have this property. This condition is automatically satisfied when the
interaction potential is taken to be the p-wave type V = VIP' p'. For example, by changing
dummy variables of integration in the gap equation from pi to _pi, the equation is unchanged
when both V(p, pi) and Am(P') are odd functions of p'.
The interesting aspect of the gap equation in triplet pairing is that the solutions are not
unique but have orbital degeneracy. There are two sources of angular momentum from each
pair of particles: the spin angular momentum and the orbital angular momentum in the p state.
The first question is the relative orientation of these angular momenta. If two particles were
binding in a gas to form an He2 molecule, then one could just couple these momenta with
Clebsch-Gordon coefficients into all possible values of J = 0, 1, and 2. In the Fermi liquid,
the superfluid state is a collective property, and each pair cannot be treated as individuals.
Nevertheless, there is still some freedom in selecting the basis for the spin and the orbital
angular momentum. This idea is best illustrated by an example.
The gap function Am(P) must have p-wave symmetry. The obvious configuration has the
spin and orbital states aligned along the same axis. The first solution described below was
found by BW. For the gap function Am(P) they chose the orbital function according to
Ylm(9p, <pp), where (9p' <pp) are the angular orientation of the p vectors. Specifically,
A±I(p) = (ftx ± ipy)A(P) = sin(9p)e±iq,pA(p) (11.255)

Ao(p) = ftzA(P) = cos(ep)A(p) (11.256)
IA(p)12 = IAol2 + IAI12 = A(Pi (11.257)
The functions E(p) and A(p) are even functions of p and are scalar.
First verify that this choice satisfies the gap equation. Use the angles (e, <p) for the
direction of p and the variables (9' , <1>') for p'. From the law of cosines, the angle between p
and pi is
P.pi = cos 9 cos e' + sin 9 sin 9' cos( <p - <P')
A = - NF VI dO.' A •
I 161t
J AI
p P s e
J
in(9 ' ) W d';.l A(P') tanh[~E(PI)J
~ E(P') 2 (11.258)
A = - NFVI JdOIA • AI cos(e' ) Jd';.l A(P') t nh[~E(PI)J

o 161t p P ~ E(P') a 2
The energy gap E(P') does not depend on the angles. The angular integrals in the gap equation
are easy to do and give (gt = -NP VI /12 = -Cd24>0)
Ll±l = A sin(9)e±i<\> (11.259)

Llo = A cos(9) (11.260)
_ Jro c
, Ll(P') [~E(P')J (11.261)
A - gt -roc d~ E(P') tanh 2
For each gap function Llm(P), the angular functions are the same on both sides ofthe equation.
These can be canceled from both sides, and each equation produces the same formula for the
effective gap function:
Ll(P) =A= Jro c

. dr-' Ll(P') tanh[~E(P')J (11.262)
gt -roc '" E(P') 2
The equation is identical to the BCS equation for the singlet gap. The coupling constant is
effectively gt. This equation is easiest to solve at zero temperature where the hyperbolic
tangent is unity, Ll = constant:
1 = gt Jro c ,d~' ~ 2gt In (21:cl ) (11.263)

-roc J~ 2 + Ll 2
ILlI Bw = 2(Oc exp ( __1_) (11.264)
2gt
The solution is an equation of the BCS type for the gap function ILlI at zero temperature. The
variation with temperature below Tc is also similar to the BCS prediction.
This equation describes the Balian-Werthamer state. Their theory is thought to apply to
the B state of superfluid liquid 3He, although the theory must be modified to account for the
strong coupling between quasiparticles. However, even the weak coupling theory, as just
derived, makes a number of predictions which agree with the experimental findings in the
superfluid B state. Some of these will now be described.
One important property is that the B state is isotropic, which is defined as having the
energy gap Ll(P) not depend on angle in the superfluid. The best way to understand this
definition is to give an example of a state which is nonisotropic. As a random choice, consider
the possibility
Lll= Ll_l = 0 (11.265)

Llo(p) = .J3 cos(9)Ll(P) (11.266)
E(p) = J~; + 3 cos2 9Ll(Pi (11.267)
The p-wave pairing requires that Llm(p) always depend on a p-symmetry angular function.
The choice in (11.266) provides an entirely self-consistent solution to the gap equation
(11.253). The normalization factor .J3 is chosen so that Ll(pi averages to Ll(P)2 around the
Fermi surface. The energy gap is no longer isotropic but now depends on the angle 9 between
the spin and orbital motion. The energy gap and excitation energy will vary around the Fermi
surface. A more detailed discussion of nonisotropic superfluid states is presented below. The
motivation is provided by the superfluid A phase, which is thought to be a nonisotropic state.
The BW state (11.256) chose axes where the coordinate bases for the spin and orbital
motions were aligned. There is no need to make this choice, and the same gap equation for
A(p) (and the same solution) are obtained if these axes are rotated with respect to each other.
For example, give the orbital coordinates a 90° rotation, and the gap functions are
A±I = (jJy± iftz)A(P) (11.268)

Ao =PxMp) (11.269)
IA(p)12 = A(Pi (11.270)
The gap function is isotropic and given by the same value (11.264) as before. In weak
coupling the choice of orbital coordinates does not influence the energy gap. The BW
superfluid state has orbital degeneracy in that the orbital motion is not coupled to the spin
motion and can proceed independently. This feature depends on the isotropic nature of the
superfluid phase.
Many of the thermodynamic properties of the superfluid phase depend only on the
energy gap. One example is the specific heat. These properties are exactly the same as
predicted by the BCS theory for singlet pairing, which also has an isotropic energy gap. (In
real metals, with anisotropic Fermi surfaces, the BCS gap can also be anisotropic, which is a
complication we shall not discuss.) The BW theory predicts that the transition from the
normal to the superfluid phase is a second-order phase transition and that the specific heat is
discontinuous at this phase transition. These properties are observed in superfluid 3He, as
mentioned earlier. The interesting difference between singlet and triplet pairing is in magnetic
phenomena. There are interesting nuclear magnetic resonance phenomena in the triplet states
which were very important for identification of the superfluid phases and which verify the
theory. These are described by Leggett (1975).
The A phase has a different type of triplet pairing states than the B phase. The latter is
well described by the theory of Balian and Werthamer (BW). The states of the A phase are
also solutions to the gap equation (11.253) for triplet pairing. They are called equal spin
pairing (ESP). They are composed of only the ms = ±1 components of the spin angular
momentum and not with the ms = 0 components. Now consider the theory when Ao = 0
while retaining the two gaps A±I.
The first realization is that there is no relationship between the orbital momentum of the
two gap functions Al and A_I. The equation which is solved for each A±I involves the gap
function Ao. If the latter is zero, there is no connection between m = ±1. The excitation
energy Em(P) = J~; + IAml 2 involves only one gap function. Another way to understand the
decoupling of Al and A_I is to start from the beginning and write the equations of motion. In
describing the response of a spin-up particle, the only nozero terms are with the correlation of
the motion of other spin-up particles. There is no correlation with spin-down particles once
Ao = o. The spin-up particles live in one world, and the spin-down particles live in another.
The orbital motion of spin-up and spin-down particles is arbitrary. There are two independent
superfluids, one of spin-up and one of spin-down, which coexist in the liquid.
The orbital motions of each superfluid need not be correlated. A consequence of the
orbital independence is that the excitation energies Em(P) and gap functions IAm(P)12 no
longer need to have the same values for m = 1 and m = -1. For example, a possible set of
solutions are
AI(p) = -A<ftx + ipy)A(P) = -A sin(9)ei <PA(P)

(11.271)
A_I(p) = -A(jJy + pz)A(P) = -A[sin(9)cos(<l» + icos(9)]A(P)
It is obvious that E](8, q,) =f. E_](8, q,) except at special angular points. The excitation and
energy gaps for spin-up and spin-down particles would be quite different. Each energy gap
function has the same shape relative to its axis (north and south poles), but these poles are
rotated relative to each other for spin-up and spin-down particles.
The superfluid is called unitary if the spin-up and spin-down particles have the identical
angular dependence of their energy gaps E](8, q,) = E_](8, q,). ESP states are unitary when
the north-south poles are aligned for the two systems. The superfluid is called nonunitary
when the excitation energies are not identical as a function of angle. The BW state is unitary.
The A phase appears to be a unitary ESP state. The north and south poles are aligned for
the two gap functions. This is an experimental conclusion, and there is no particular reason
the state has to be unitary. The important question is the nature of the forces which tend to
align the poles. There are no forces in the Hamiltonian (11.236) which cause alignment. The
Hamiltonian neglects the very, very weak forces due to the dipole--dipole interactions between
the nuclear moments. Leggett has shown that these dipolar forces will tend to align the poles.
Although it is a weak effect, there is nothing to oppose it. Of course, any external perturbation
such as a magnetic field will also tend to align the poles, since alignment lowers the magnetic
energy. Nonunitary states are discussed in reviews by Leggett (1975) and Anderson and
Brinkman (1975).
11.3. QUANTUM HALL EFFECTS
The quantum Hall effect (QHE), and the fractional quantum Hall effect (FQHE), are
phenomena which occur in highly correlated electron fluids. The QHE was discovered by von
Klitzing et al. (1980), and the FQHE was discovered by Tsui et al. (1982).
Electrons are allowed to move in only two dimensions using semiconductor technology
to confine the motion in the third direction (z). The application of a strong magnetic field B
along the z direction causes the electrons to move in periodic orbits in the xy plane, which are
quantized. This quantization leads to energy levels En = hroc(n + D, where the cyclotron
frequency is roc = eBjm*c, and m* is the effective mass of the electron in the semiconductor.
The center of the cyclotron orbit can be anywhere in the plane. The electrons select the
centers of their orbits to be arrangements which are highly correlated. At certain densities the
electron gas becomes quite rigid, or incompressible, which is the simple explanation of the
QHE and FQHE.
The topic shares many attributes with superfluid helium. Perturbation theory does not
work. The best wave functions are guessed, and are written down by intuition. The excitation
spectrum is calculated using Bijl-Feynman theory.
11.3.1. Landau Levels

In a large magnetic field, electrons tend to have orbits which oscillate in time with an
angular frequency given by the cyclotron frequency roc = eBjmc where B is the magnetic
field. If the relaxation time for scattering is 't, then the behavior of electrons depends upon
whether roc't is greater or smaller than one. If roc't » 1 then the electrons form very periodic
orbits. Their energy levels are quantized in units of hroc' which are called Landau levels.
These orbits are simple to describe in two dimensions.
Sec. 11.3 • Quantum Hall Effects 743
In the theory of electricity and magnetism, the potentials may be either scalar potentials
<p(r) or vector potentials A(r). The magnetic field is introduced through a vector potential.
The basic equation is B = V X A. When the magnetic field is a constant there are many
different ways to choose a vector potential to satisfy this relation. Ifthe magnetic field is in the
z direction, then the following choices all give the correct magnetic field
A = B(-y, 0, 0) (11.272)
A = B(O,x, 0) (11.273)
B
A= "2 (-y,x, 0) (11.274)
Each different vector potential is a choice of gauge condition. Since the eigenvalues do not
depend upon this choice of A, they are called gauge invariant. The first gauge is chosen, for
no particular reason, so that the Hamiltonian is
H = L. -2m.
1 [P -
]
-e A(r)
c ]
J2 - L. fl.J • B (11.275)
} ] }
1
= 2m .y [( Pjx + -eYjB)
c-
2
+ Pjy2] -.y fl.j • B (11.276)
The electron has a spin, and a magnetic moment fl.j. The last term in the interaction is the
Pauli interaction between the electron moment and the magnetic field. For electrons in free
space, the Pauli term and the orbital eigenvalues have similar energies. The QHE and FQHE
measurements are usually done on semiconductors with very small effective mass m*, such as
GaAs with m* 1m = 0.07. The effective mass should be used in calculating the cyclotron
frequency COc = eBlm*c. The symbol m in our equations is actually m*. The level splitting
fico c is greatly enhanced in the semiconductor because of the small effective mass. The Pauli
energy is altered much less by the band structure, so that it is much smaller than fico c .
Although small, it is important for discussing the experiments. It is ignored for now, in order
to concentrate on the Landau term.
The Hamiltonian does not yet contain interactions between particles. In this case, the
solution is found for one electron, and the many-particle ground state is composed of elec-
trons which occupy these one-electron states. Consider the Hamiltonian for one electron,
H
1
= 2m [(Px
eyB
)2]
+ ---;;- + P y
2
(11.277)
The potential energy term in (11.277) contains y as the only space variable. The Hamiltonian
commutes withpx so the eigenstates can be chosen to be plane waves in this coordinate. Write
the solution as
eik,x
\jJ(x,y) = .jr;f(Y) (11.278)
e' xX 1
'k
H\jJ(x,y) = .jr;2m [ ( fikx
eBy
)2]
+ ---;;- + Py
2
fey) (11.279)
The effective Hamiltonian operating on the y variable is a harmonic oscillator. This obser-
vation is made clearer by defining some constants, where the electron charge is (e<O)
lelB
00 =- (11.280)
c me
z2 =~= he (11.281)
mooc lelB
_ hekx _ k 12 (11.282)
Yo - - eB - x
eikxx [p2
T
moo 2 ]
H",(x, y) = ..;r; 2~ + (y - Yoi f(y) (11.283)
The last equation provides the effective Hamiltonian in the y direction in the explicit form of
the harmonic oscillator equation. Its eigenstates are harmonic oscillator functions <Pn(y - Yo)
with eigenvalue En = hoo/n + !). The solution to the Hamiltonian is
eikxx
",(x, y) = ..;r; <Pn(y - Yo) (11.284)
En,m, = hOOc[n +!] + ms~ (11.285)
The magnetic length scale is I. The last term is for the Pauli interaction, where ms = The ±!.
Landau levels are states with different values of n. The lowest Landau level has n = O. It is
interesting that the eigenvalue does not depend upon the quantum number kx . The state has a
high degeneracy, in that many different values of kx have the same eigenvalue. The classical
picture has the electron going in a circular orbit. The center of the orbit can be located any
where in the system, without altering the energy. For the quantum system, the value of kx
relates to the location of the center of the harmonic oscillator.
A system of one electron will have a ground state with the electron in the lowest Landau
level (n = 0). The interesting question is how these states are occupied for a set of Ne
electrons. The question can be rephrased: In a system of area A = LxLy, how many electrons
can be put into the lowest Landau level? The constant Yo is the center of the cyclotron orbit.
The orbit center must be in the area, which provides the constraint
o ::: Yo = ki2 ::: Ly (11.286)

k = 21tnx < Ly (11.287)
x Lx -12
LxLy mooc
nx ::: 21t[2 = 21th A == nmax (11.288)
Although the energy does not depend upon the value of kx' this quantum number has a
restricted range of value. The requirement of keeping the orbit center in the area constrains
nx ::: nmax'
Define p(E) as the density of states per unit area. This state density is
(11.289)
Sec. 11.3 • Ouantum Hall Effects 745
where En,m, is the eigenvalue in (11,285). The summation over nx just gives nmax ' Then the
factor of area cancels
p(E) = m:e L 8(E - En,m)

n,ms
(11,290)
no = JdEp(E)nF(E) = m:e n~, nF(En,m) (11,291)
mme = _1_ = ~B = ~ (11,292)

h 2nz2 he <Po
v
no = 2nz2 (11,293)
v= L nF(En,m) (11.294)
n,ms
The density of electrons no (#/meterZ) can be expressed as a prefactor times the number of
occupied Landau levels, where the Fermi-Dirac occupation number is nF(E). Note that the
prefactor can be written as mme/h = B/<po where the quantum of flux is <Po = he/e. The
factor of v is called the filling factor and determines the number of occupied Landau levels.
There are two length scales in the problem. One of them is the magnetic length I. It
determines the radius of the harmonic oscillator orbits. The other is the average separation
between electrons, which is called a. Highly correlated liquids usually have short range order
in the form of a triangular lattice with each electron having six neighbors at a distance a. As
an example, examine a woodpile stacked with equally sized logs. The area per electron in the
plane triangular lattice is a2 .J3/2. Setting this formula equal to the inverse of the electron
density gives that a2 = 2/n o.J3. At what value of the magnetic field does the separation
between electrons a become larger than the orbital diameter? The harmonic oscillator wave
function for the lowest Landau level is exp[ -(y - YO)2 /2z2]. The exponent is -1 at ily = .J21.
Taking this to be the radius of the orbit, the separation between electrons is greater than the
orbit diameter when
(11.295)
(11.296)
(11,297)
The number on the left is about 0.907. The summation over the occupied levels, on the right
in the above equation, must be a less than 0.9 before the Landau levels are truly separated in
1
space. These values are reached in the FQHE for fractions such as v = or ~.
11.3.2. Classical Hall Effect

The classical Hall effect must be understood before its quantum version can be appre-
ciated. The measurement geometry is shown in Fig. 11,16. The shaded area is the experi-
mental sample. A current I is sent through the material, along the direction labeled "x". The
voltage change Vx is measured between the two leads on top. This configuration is a standard
four-probe measurement. The voltage contacts are not put at the end of the sample, in order to
avoid the influence of nonohmic contacts where the current enters and leaves. Another pair of
FIGURE 11.16 Experimental geometry for the measurement of the Hall effect. The magnetic field is perpendicular
to the plane.
voltage contacts measure the voltage Vy across the sample. This voltage is zero in the absence
of a magnetic field.
The Hall effect occurs when there is a magnetic field B perpendicular to the plane of the
sample. Then a nonzero value is found for the voltage v;" which is proportional to both Band
I. Since the current I is along the x direction, define two components of resistance
(11.298)
(11.299)
For most materials, the magnetic field hardly changes Rxx' The main experimental observation
is the nonzero value of RyX'
The physics is simple when considering the motion of the electrons as classical. Assume
the particles obey Newtonian mechanics. In two dimensions, the magnetic field creates a
Lorentz force
. [1
mv=e E+-;;vXB ]mv --:r (11.300)
mv. x =e [ Ex +-
VyB]
C
-mvx
-
~
(11.301)
(11.302)
where E is the electric field and ~ is the relaxation time for scattering. Consider that the
experiment is steady state, so that the time derivatives are zero. The currents densities are given
by Jx = enovx, Jy = enOvy There is no current in the y direction since the devices to measure
voltage have very high impedance. Setting Vy = 0 in the last equation gives the identity
(11.303)
(11.304)
(11.305)
The Lorentz force tends to drive the electrons in the perpendicular direction. The flow of
electrons along the x direction is diverted to the y direction. Since no current can flow in the y
direction, then a voltage Vy must be be built up to force the electrons back along the x
direction. The voltage is due to an accumulation of excess charge along the top of the device.
The quantity Jx is the current density. In a three-dimenensional system it has the units of
amperes per area, so that Ix = JxAo where Ao is the cross sectional area of the sample. In the
present case Ao = Lyt, where t is the thickness of the sample. One then gets in 3D that
V = Ix B (11.306)
y noect
The Hall effect is very useful in the laboratory for measuring the density no of current
carrying particles in the system. It is a relatively easy measurement if a sample of material is
available which is several millimeters on each side. The Hall measurement gives the density
of carriers in semiconductors and metals. The result is only simple to interpret if all carriers
are in the same energy band of the crystal.
In two dimensions the Hall effect is even easier to derive. The particle density no has the
units of number of particles per area. The current density Jx has the units of amperes per unit
width. Therefore Ix = JxLy and the expression for the Hall voltage in 2D is
V =~B
- ( 11.307)
y noec
R
yx
= ~
noec
(11.308)
This equation does not depend upon any dimension of the sample. Samples of differing
shapes and size give the same measurement.
The classical Hall effect has two characteristics:
• The longitudinal resistance Rxx is independent of magnetic field.
• The transverse resistance Ryx is proportional to magnetic field.
These characteristics change in the Quantum Hall Effect.
11.3.3. Quantum Hall Effect

A quasi-two-dimensional electron gas can be created in semiconductor microstructures.
There are several ways to accomplish this feat, and most of them have shown the QHE. One
system is the channel of a field-effect transistor. The gate voltage is used to confine the
electrons in the direction perpendicular to the plane, where the barrier oxide traps them along
the interface. The electrons are free to move in the plane, and their motion is largely two
dimensional. In order to observe the QHE (quantum Hall effect), the semiconductor sample
must have exceptionally high purity and a large value for the mobility. Experiments originally
showed the QHE in silicon MOSFETs. Another kind of semiconductor microstructure is a
J 2 112
2/5 1/3
3n
10 20 30
Magnetic Field (T)
FIGURE 11.17 Quantum Hall effect and fractional quantum Hall Effect in GaAs. (Stiinner, 1999 used with
permission).
quantum well. It is formed by growing n-layers of one semiconductor, such as GaAs, between
another semiconductor such as AlAs. Since AlAs has a larger energy gap, electrons in the
conduction band of GaAs are confined in the direction perpendicular to the layers. They move
readily along the layer, so the motion is quasi-two-dimensional. Since GaAs and AlAs have
the same lattice constant, they can be alloyed continuously. The barrier layers are often
All_xGa"As. Most measurements have been done in single quantum wells ofGaAs, although
the phenomena has been observed in a few other semiconductors.
A typical result for the QHE and FQHE is shown in Fig. 11.17, which shows the
resistivity as a function of the magnetic field B. The two resistances have the following
features:
• The transvere resistance Ryx == RH develops plateaux at values of RH = Ro/i, where
Ro = h/e2 ~ 25812.8n, and i is an integer or simple fraction. The QHE is when i is
integer, the FQHE is when i is a fraction.
• The longitudinal resistance Rxx == R -+ 0 during these plateaux.
The quantity Ro is a combination of fundamental constants. It has become the world standard
for maintaining electrical resistance. Since the resistance steps appear at the same values,
regardless of the shape of the sample, any laboratory in the world will find the steps at the
same numerical values.
In the earlier discussion of Landau levels, an expression was derived for the density of
electrons in the plane. Using that expression (11.291) gives
1 noec e2
-=--=-v (11.309)
Ryx B h
v = L nF[~(En,m, - Il)] (11.310)
n,ms
R =Ro (11.311)
yx v
The QHE has a simple explanation if the function v has integer values. The FQHE requires
that the filling factor v be a simple fraction.
The key to understanding the QHE is to understand the behavior of the function v as the
magnetic field in increased. The important parameter is the chemical potential Il. How does it
change with magnetic field? There are three different models for the behavior of the chemical
potential in these systems.
11.3.3.1. Fixed Density

The first model assumes that the density of electrons n is fixed in the sample. When the
magnetic field B is varied, the chemical potential must also be varied so that the density no is
unchanged. When considering a function such as (11.291), in order to keep the right-hand
side of this equation a constant as B is varied, the chemical potential must go through some
gyrations. The product no<po = Bo has the units of magnetic field. For small values of
B, (B« Bo), the summation over n needs many terms. Then the chemical potential varies
with magnetic field in a manner which is periodic in liB. This behavior is well known, and
shows up experimentally in de Haas-van Alphen, Shubnikov-deHaas, and other measure-
ments of these oscillations. However, as B increases and becomes B 2: Bo then all electrons
are in the lowest Landau level. The requirement of constant density gives the expression for
the chemical potential (neglecting the Zeeman term)
(11.312)
(11.313)
The variation of the chemical potential with magnetic field is an important feature of electrons
in large magnetic fields.
This model is applied to most three-dimensional solids. There the number of electrons is
fixed by considerations of charge neutrality.
11.3.3.2. Fixed Chemical Potential

The second model for the chemical potential assumes that it does not change with
magnetic field. Note that in the experimental measurements, there are at least five metallic
wire leads soldered to the sample. These metal contacts form a reservoir of electrons which
could fix the chemical potential in the device. The two-dimensional device has only about
1011 conduction electrons. That is a very small number compared to the number of electrons
in the metallic reservoirs. The change in energy of these small number of electrons has a
negligible contribution to the energy of the system. The chemical potential is thereby fixed.
The assumption of a fixed chemical potential certainly applies to MOSFET ("Metal-Oxide-
Semiconductor-Field Effect Transistor") devices. The usefulness of the gate voltage is that
one can use it to modify the density of electrons in the conducting channel. The electrons in
the channel have a negligible effect on the chemical potential, which is determined by the
source and drain.
In the case that Il is a constant, then the function! will have plateaux at low temperature
when plotted as a function of magnetic field.
Figure 11.18 shows a plot of the series
(11.314)
n
which is the number of electrons in Landau levels, of a single spin component, as a function
of chemical potential. When hroc » kBT there are plateaux in the occupation number f.
Figure 11.18(b) shows the value of no as the magnetic field is varied. Note that no ~ 0 as the
lowest Landau level exceeds the chemical potential (hroc > /l). This prediction is in direct
contradiction to the experimental data, which shows that no is relatively constant at high
magnetic field. Note that the resistance Ryx tends linearly upward in Fig. 11.17, except for the
plateaux. The linear behavior is only possible when no is relatively constant.
The model with constant chemical potential explains well the QHE (yennie, 1987). The
experimental plateaux are very well fit by this model. The plateaux appear exactly as in Fig.
11.18, which plots f, while the experiments measure 1If. However, this model fails to explain
the FQHE, and fails to explain the data at large magnetic field which shows that no is rela-
tively constant.
11.3.3.3. Impurity Dominated

The third model for the chemical potential invokes localization caused by static impu-
rities (prange, 1987). All semiconductors have impurities. They tend to bind the Landau
levels near potential mimima in the plane. These minima are due to impurities in the
underlying substrate. In one dimension, all states are localized by any impurity scattering. In
the QHE, the system is two dimension. However, one quantum number becomes localized
into a harmonic oscillator state. The second quantum number (kx ) defines a type of one-
dimensional motion. The particles behave, in some respects, as if they were in one dimension.
However, their bound states are not given by a Hamiltonian in one dimension, since there is
no kinetic energy term associated with the motion in the x direction.
The argument in favor of localization caused by impurities, as controlling the chemical
potential, is that as the experimental samples are made with higher purity, the plateaux
become narrower. Inpurities are essential for pinning the value of chemical potential, and
thereby observing the FQHE.
The fractional quantum Hall effect (FQHE) is the appearance of additional plateaux at
noninteger value of the function v. Tsui et al. (1982) discovered that increasing the magnetic
field, beyond the plateaux for v = 1, produced additional plateaux which appear at fraction
t
occupation such as or~. The experimental technique continues to improve, with samples of
higher purity, at larger magnetic fields, and at lower temperature. Figure 11.17 shows one of
their recent experimental result on a sample of very high purity. Numerous fractional plateaux
are apparent.
The plateaux at fractional filling are due to electron correlation. The theory of this
phenomena shares some features with liquid helium. It is impossible to describe the correlated
states with any kind of perturbation theory. The theory has no small parameter on which to
base a perturbation expansion. Instead, progress has occurred by writing down ad hoc wave
functions, and seeing what they give. The best wave function is described in the next section.
6.--.--------------------------------------____--.
a
5
f(B)
o +--------.--------~------_,._------_.~------r_------~
o 5 10 15 20 25 30
B(T)
1.0.-----------------------------------------------.
b
0.8
0.6
-iil
c:
0.4
0.2
0.0 +-----.-------.------.----,....-"'----..------i
o 5 10 15 20 25 30
FIGURE 11.18 (a) Number of electrons in Landau levels as a function of magnetic field for a fixed chemical
potential. (b) The density of electrons vs. magnetic field, in units of 10 16 m- 2 . Parameters are T = 0.1 K, m* = 0.07,
no = 0.47 x 10 12 m- 2 , and 11 = 30 J.tERy"
11.3.4. Laughlin Wave Function

Laughlin proposed a wave function for the FQHE which has become the standard basis
for all theories. It is discussed in this section. As with all wave functions in this chapter, it is
written down by intuition, and contains correlation. The correlated wave function is not
derived by any form of perturbation theory. It is just written down, and its properties are
studied.
The first step in the derivation is to solve the Hamiltonian for the symmetric gauge
(11.274). For a single electron, the Hamiltonian in this gauge is
(11.315)
x
The space variables are put in dimensionless form using = x/l,y = y/l where the magnetic
length is P. = lie/ eB. After some algebra, the Hamiltonian is rewritten as (dropping the
primes)
(11.316)
(11.317)
(11.318)
where tan( <1» = y /x and p2 = xl +T in polar coordinates. Consider the eigenfunction

\jIm(P, <1»
(11.319)
(11.320)
(11.321)
(11.322)
The eigenfunctions \jim have an eigenvalue ofliIDe/2 which indicates that the electron is in the
lowest Landau level (n = 0). The quantum number m indicates angular momentum, since
Lz\jlm = m\jl. The value of angular momentum m does not alter the eigenvalue. In the prior
section the silent quantum number was kx' which also did not affect the eigenvalue. In the
symmetric gauge the silent quantum number is m. The last expression shows that these
functions are mutually orthogonal, because the integral over d<l> vanishes unless m = m'.
In the FQHE, the high value of magnetic field at low temperature ensures that
IiIDe » kBT. All electrons are in the lowest Landau level for the case that the fractional filling
v < 1. Then the eigenfunction \jim is the basis for all electrons in the ground state. An
interesting question is to ask where are the electrons? Use Stirling's approximation to the
factorial to write the electron density for large m as
p (p) = I\j! 12 = I p2me-p2/2 (11.323)

m m 21tm!2 m
~ I
...fIi[ii1. - -2 + m In(p2) + m - m In(2m) ]
exp [p2 (11.324)
21t 21tm
~ 21t ~exp[-!(p2
21tm
-2m)+mln(~)]
2m
(11.325)
1
~ 21t...fIi[ii1. exp [ -! (p2 - 2m) + mInI
( - 2m)]
+ p2 2m
~ 1 [18m
21tJ21tm
exp --(p - 2m) 2 2] (11.326)
For large value of angular momentum m, the peak in the electron density occurs at p '" ffm.
The electrons are in a circle, where m = 0 is the orbit at the center of the circle, and higher
values of m are further from the center.
The above discussion is for a single electron. The next step is to construct a many-
electron wave function. In this case it is useful to describe the position vector (x, y) by the
complex number Z = x + iy = pei<l>. The basis set is now written as
(11.327)
In the Rartree-Fock approximation, the wave function for a many-electron system is a Slater
determinant. All electrons are assumed to have the same spin direction since they are aligned
by the magnetic field. If they have spin up, the eigenfunction is ('f.j for the j-th electron. Each
electron must be in a different state, so that they occupy the angular momentum states
m = 0, 1, 2, ... ,N - 1. Then the Rartree-Fock wave function for N electrons is
\j!O(ZI) \j!1(zl) \j!2(ZI) \j!N-I (ZI)
\j!O(Z2) \j! I (Z2) \j!2(Z2) \j!N-I (z2)
1 \j!O(z3) \j! I (z3) \j!2(z3)
'¥N(zl,z2,'" ,zN) = &N! \j!N-I (z3)
\j!O(zN) \j!1 (zN) \j!2(zN)

x ('f.1 ···('f.N (11.328)
= ~ [IIkNdDN('f.1 ... ('f.N exp

vN!
[-! t 4 j =1
Izi] (11.329)
ZI zT zf- '
1 Z2 z22 z!f-l
2
if-I
DN= Z3 ~ 3 (11.330)
ZN 4 ~-I
N
The determinant is due to Vandermonde, and is exactly equal to the product of the difference
for all pairs of coordinates
'I'N(ZI,Z2,'" ,zN) = 1
t.TIIIkNkIIj>i(zj -z;)al" ·aNexp [ - -1 L N Izi ] (11.331)
vN! 4 j =1
The first product is the normalization coefficients. The second product contains (Zj - zi) for
all pairs of electrons. This factor ensures the many-electron wave function is anti-symmetric
under exchange of any two electrons, as is required for fermions.
The wave function 'I'N can be used to calculate some properties of the many-electron
system. Squaring it, and then integrating over all but one coordinate, gives the average density
of one electron. Using the theorems in Sec. 5.1.9, and the fact that the eigenfunctions are
orthogonal, gives the result
PI (p) = N fo I\jim I =
1 N-I 2 e- p2 / 2
21tN ~ 2mm!
p2m
(11.332)
Again the notation PI means the density of one particle in the many-particle system. For small
values of p, the summation gives exp(p2/2) which cancels the other exponent, and
PI ~ 1/(21tN). The density of N such electrons, after restoring the factors of length I, are
(11.333)
Equation (11.293) has almost the same expression except the right-hand side is multiplied by
v. Equating these two expressions gives that v = 1. The many-particle wave function was
chosen to fill up the lowest states of angular momentum. This case corresponds to the
maximum packing of electrons in the lowest Landau level, which is a filling factor v = 1.
The pair distribution function g(r) is found from P2(rl, r2), which is obtained by inte-
grating over all but two coordinates. Since the functions are orthogonal, P2 is found from all
pairs of two-particle wave functions
(11.334)
e-(p~+p~)/2 N-I 1 (2' 2' 2' 2' +" )

= 21tN(N _ 1) ~ 2i+1i[jl PI'pi + P{P2' - 2(PIP2)' J cos[(i - j)(<pI - <P2)]
(11.335)
For small values of PI.2 the series converge to an exponential. In this case the pair distribution
function is
pirl' r 2) = 1tN(~ _ 1) [1 - exp(-![pI + p~ - 2PIP2 COS(<PI - <P2)])]

(11.336)
g(rl - r2) = 1 - exp [ - 2~2 (rl - r2)2]
S(k) = 1 + no Jd r[g(r) -
2 l]eik ' r = 1 - exp[-k2p /2] (11.337)
Note this pair distribution function is for two particles in a system of N » 1 particles. A
different result is obtained for only a few particles. One can also derive the three- and four-
particle distribution functions (Jancovici, 1981). The static structure factor S(k) is also given.
Note that it goes to zero as O(k2) rather than as O(k) as in a fluid of neutral particles. These
fonnulas apply only for the case of v = 1.
In Chap. 5 the exchange energy per electron of the electron gas is provided by the
fonnula
E = -noil
x
- -[1
2 r
Jd
r - g(r)]
2
(11.338)
where it has been adopted to two dimensions by changing d3 r -+ d2 r. Using the above
expression for g(r), and no = 1/(2nz2) for v = 1, the exchange energy is (r = pII)
E = - il Joo dre- r /2 = _ il ~ (11.339)

x 21 0 21'/"2
This fonnula is valid for filling factor v = 1. For fractional filling, the exchange energy in the
Hartre-Fock approximation is written as Ex = -evil II. Using the Laughlin wave function,
described below, gives the following values for the dimensionless constant:
I/v c.
III 0.627
1/3 0.416
1/5 0.334
The Laughlin wave function for the FQHE was written down by intuition starting from
Eq. (11.331). It is
'¥(zl' Z2' ... ,ZN) = NpNITj>i(Zj - Z;'jJ IX1 .. ·IXN exp [- f.IZ/] i J=l (11.340)
where the nonnalization constant is NpN . The factor of (Zj - z;'! is the difference between the
coordinates raised to a power of p. The fennion wave function has to change sign under the
exchange of any two coordinates. This fundamental anti symmetrization is achieved as long as
p is any odd integer. The case with p = 1 is the integer QHE with v = 1 in (11.331). Other
values such as p = 3, 5, 7 ... provide an accurate wave function for the FQHE. The Laughlin
wave function cannot be derived from a single Slater determinent and is not a Hartree-Fock
wave function. It contains correlation and is a highly correlated wave function.
The exponent p is related to the filling factor p = 1/v. The choice p = 3 gives a wave
function for the fractional state v = t.
The proof of this assertion starts by examining the
density matrix
PN(r1, r2' ... ,rN) = 1'¥(zl' Z2' •.• ,zN)1 2 (11.341)
= N;NITi»ri - rip exp [ -~;=1] (11.342)
= N;Nexp[p ~ 1n Iri - r/ - ~;=1 ] (11.343)
The spin eigenstates have been eliminated. Also note that

IZ1 - z21 2 = (Xl - x2i + (Y1 - Y2i = (r1 - r2)2. (11.344)
The density matrix can be cast into the form for a classical fluid PN ""' exp[-~V] where the
potential energy ~ V == U has the form
U(rl"'" r N ) = ~ '2;1 - p L.lnlri -

j '>j
ri (11.345)
The potential has two types of terms. Each electron has a potential Yf /2 which tends to
confine it to the origin. There is also a particle-particle interaction of the form of a logarithm.
It is characteristic of the two-dimensional Coulomb gas, where the particles are not points
with V = c? /rij, but instead are charged rods perpendicular to the plane which have
V = c? In(rij)' The same potential is found for interacting vortices.
For a system of N electrons, the two-dimensional fluid forms a circle of radius
R ""' ...tiN. The electrons near to the center of the circle are in equilibrium, which means that
the force on them should be zero. For a uniform circle, the force can only be radial. The radial
force on one of them, say r I is
au N 2rl - 2rj cos( <p)
F (rl) = - - = -rl + P L -;;-----;,---"""'-----'---:- (11.346)
r arl j=2rT + 1 - 2rlrj cos <Pj
where <Pj is the angle between rl and rj. The last term is evaluated by changing the
summation to an integral, assuming the particles have a uniform density of no
(11.347)
(11.348)
The angular integral is done using
Jo A +Bcos r I and is two if rj < r I' In fact, the above expression
is Gauss's law for the two-dimensional Coulomb gas. The force comes only from the charge
inside of the circle of radius rl' The remaining integral is elementary,
4npno Jrl
Fr(rl) = -rl + - - rjdrj = -rl[l - 2nnoP] (11.351)
rl 0
The density no = v/2n in units where all dimensions are scaled by the magnetic length I.
Then the factor in brackets is [1 - pv]. The requirement that the force on particle 1 be zero
sets p = 1/ v. Since p is an odd integer, then the Laughlin wave function describes the
fractional fillings ofv = lip such as t,!, t,
etc. Numerous numerical tests have shown that the
wave function describes these states extremely well.
The wave function for p = 1 is the Slater determinant which gives (11.331). By filling up
all states m = 0, 1, 2, ... , N - 1, the system has a uniform density given by no =
1/(2n[2). The Laughlin wave function for v = lip also has a uniform density, on the average.
It is interesting to calculate the current density associated with the Laughlin ground state.
First calculate the current from a typical electron, such as ZI, using the standard quantum
mechanical expression
en . {'I'..,.
J I = -2 * *
l' 1'1' - 'I' VI'I'} (11.352)
ml
VI'I' = 'I'{-~+p f= i+ iY } (11.353)
2 i=2ZI - zi
J I = _pen 1'1'12 f=i(yl - y;) - Y(~I - Xi) (11.354)
m i=2 (rl - r;)
J
The summation over i can be changed to a continuous integral Li -+ no d 2r. The factor of
1'1'12 supplies an additional factor of I/A, where A is the area. The current from particle 1 is
J1 ---
f
- epnon d 2 i(Y1 - y;) - .Y(x I - Xi)
ri 2 (11.355)
mA (rl - r i )
The integral is zero for an infinite system. The result is nonzero for a finite system consisting
of a circle of radius R and area A = TtR2. The angular integral have the same form as
discussed in Eq. (11.347), so that the integral over dri is constrained to values ri < rl' The
result is
(11.356)
The current is rotational. This result should not be surprising, since the lowest Landau level is
composed of eigenstates with many values of angular momentum, which are all rotating in the
same direction. Also note that the current obeys the equation of continuity V . J = O.
11.3.5. Collective Excitations

Several collective excitations have been identified for the QHE and the FQHE. Two are
discussed here.
11.3.5.1. Magnetorotons
Girvin et al. (1986) used the Bijl-Feyman theory to calculate the frequency dependence
of the density fluctuations in the QHE and the FHQE. The method is identical to that used
earlier in this chapter for excitations in 4He. However, the derivation for the FQHE has some
new wrinkles. These excitations are called magnetorotons.
The excitation of wave vector k is obtained by multiplying the ground state wave
function 'I'g by the density fluctuation p(k) which gives an excited state 'I'k
p(k) = Leik ' rj (11.357)

j
(11.358)
As in the derivation for 4He, the only tenn which fails to commute with the density operator is
the kinetic energy. Repeating that earlier derivation gives the same result
h2 k 2
m(k) = 2mS(k) (11.359)
This result is quite universal. It applies to any system, and to any fractional filling.
First examine the case v = 1 where the lowest Landau level is completely filled. The
structure factor for that case is given in (11.337).
h2 k2
m(k) = 2m 1 _ e-k't' /2 (11.360)
At large values of k the excitations go to the free-particle kinetic energy m(k) = 10k' At small
values of k the result is
h2
lim m(k)
k-+O
= -/
m2
= hme (11.361)
The excitation energy is just the cyclotron energy hme. Of course, this makes physical sense:
if the lowest Landau level is filled, the excitation is to the next Landau level which is (n = 1).
This result is also in agreement with a theorem ofKohn (1961) which states that the magneto-
phonon excitation energy is always hme' regardless of the filling factor, and regardless of the
interactions.
Girvin et al. (1986) remark on the accuracy of the Bijl-Feynman fonnula. Here it gives
the exact excitation energy, while for 4He it gave a very approximate excitation energy. They
remark that the current from this excitation satisfies the equation of continuity (V . J = 0),
while it does not when using the correlated basis function for the ground state of 4He. They
also remark that the theory assumes a Single Mode Approximation. All of the excitation
spectra is in the one state. Most fennion systems have a collective mode (e.g., plasmons) plus
electron-hole excitations. At long wavelength the collective mode dominates. Then the single
mode approximation is an accurate description of the excitation.
Girvin et al. predicted the existence of a new low energy collective excitation for the
FQHE. The first guess is to use the Laughlin wave function for the ground state 'Pg for
p = 3, 5, etc. According to Kohn's theorem, the calculated excitation energy will still be hme.
This large excitation energy is obtained because the density operator p(k) has contributions
from many Landau levels. However, the low lying excitations from the Laughlin state are
probably to states which are different configuration of the lowest Landau level. So they used a
projection operator P n which projects out the states for the Landau level n. In their case, they
only wanted Po. Both numerator and denominator are evaluated in this reduced Hilbert space
m(k) =~(k) (11.362)

S(k)
p(k) = Pop(k)Po (11.363)
- 1 t
S(k) = N ('Pglp (k)p(k)I'Pg) (11.364)
- 1 t
f(k) = N ('Pglp (k)[H, p(k)]I'Pg) (11.365)
The evaluation of this expression required numerical work such as Monte Carlo. They were
able to prove some important results at long wavelength. They showed that for both the v = ~
!
and v = states that co(O) =f. 0 and the excitation spectrum has a gap. The constant result for
k -+ 0 is not due to plasmon effects. In two dimensions, the plasmon without a magnetic field
goes as co2 (q) = noe'2qjm and vanishes at long wavelength. The gap at long wavelength is due
1
to the rigidity of the Laughlin state. For v = bothj(k) and S(k) are proportional to /(4 at long
wavelength, so that their ratio goes as a constant.
Figure 11.19 shows the numerical predictions for the low energy excitation. In Fig.
11 19(a) are S(q), S(q) for the state v = 1.
In (b) is shown the excitation energy co(k) for
v= 1,!' There is a minimum at nonzero wave vector, which is similar to the roton in
superfluid 4He. The name magnetoroton has been applied to this low energy excitation at non-
zero wave vector. At large wave vectors the neutral excitation is a quasi exciton consisting of a
bound state of a quasielectron and quasihole.
0.10
0.15
0.08
<J 0.10 0.08
0.04
0.05
0.02
o. 00 L....II.-.I......I.~-'-~......&.....IL.....&.....I.~-'-..L....L.....&.....IL-I.--L-I 0.00
0.0 O.S 1.0 1.5 2.0
qt
FIGURE 11.19 (a) S(q) for states v =l
The top line is S(q) and the lower line is S(q). (b) ro(k) for the states
v= t. t·
from Girvin et al. (1986) (used with permission).
11.3.5.2. Quasiholes
Laughlin (1983) considered the excitatiOn created by adding a "quasihole" or
"quasielectron". The easiest to understand is the quasihole. It is a place in the fluid where
other electrons cannot be located. Denote the Laughlin wave function (11.340) as 'I' == IP, N)
for p = 1/ v and N electrons. The wave function for the fluid with a quasihole at position Zo is
'I'N({ZJ; zo) = W(zo)lP, N) (11.366)

W(zo) = rrf (z; - zo) (11.367)
The factor of W(zo) ensures that no electron of coordinates Z; will be found at the position zoo
For the case that p = 1, v = 1 then this orbital factor is equivalent to adding another electron
to the system at position zoo The ground state energy is that of a system with N + 1 electrons
in the lowest Landau level. The cost in energy from adding the quasihole, in Hartree-Fock, is
AEh = /iroj2 - Jrr/8 e'2/I.
For fractional filling the properties of this excitation must be determined by numerical
methods. The probability is found by taking the absolute square of the wave function. The
electron density PI (rl' ro) is found by integrating this probability over all but one space
variable. This density must be uniform except near to the point r o, where it has the form of a
pair distribution g(r - ro)
perl '" rN ; ro) = I'I'N({z;}; zO)12 (11.368)
g(r - r 0)
f Pdr2" ·drN
= ---;;-"---''-------'------- (11.369)
N f Pdrdr2dr3 ... drN
q =e Jd r[1 - g(r - ro)]

2 (11.370)
The quantity q measures the amount of electron charge which is excluded from the vicinity of
the quasihole. It is found that q = ev. The quasihole has fractional charge.
The evaluation of the above integrals again uses the analogy with interacting fluids.
Write P as a Jastrow wave function:
P = N(zo)exp(-U) (11.371)
(11.372)
where N(zo) is the normalization constant. The fluid analogy has N particles of "charge" -JP
interacting among themselves with logarthmic forces. An impurity at ro of charge 1/-JP
interacts with the other electrons. The resulting pair correlation function is solved using the
theory of interacting fluid mixtures, where one component is very dilute. The energy needed
to insert the quasihole into the fractional ground state is Eq = -Cqe'2/I. Numerical work
shows that Cq = 0.026 for v = ~. For a typical magnetic field of B = 5 T, this excitation energy
in temperature units is about 6 K. Experiments must be performed at a much lower
temperature than this value, in order that quasihole excitations be thermally quenched. A
similar energy scale is found for the magnetorotons.
Problems 761
Consider the additional current due to multiplying the Laughlin wave function by a sum
of phase factors
CI> = exp[i~q,(ri)J'¥ (11.373)
(OJ) = nohVq,(r) (11.374)

m
The additional current is given by the gradient of the phase factor. If the quasihole is located
at Zo = 0 then its factor has the above form
W = TIiz = exp [ ~In(riei6i)J (11.375)
iq,(r) = Inri + i9 (11.376)
(OJ) = noh~ (11.377)

m r
The additional current from the quasihole is also rotational. Girvin (1987) points out that the
quasihole has the form of a vortex flow.
In many two-dimensional systems the low energy excitations are vortices. The statistical
mechanics of the disordering process are dominated by the binding together of vortices of
opposite rotational motion. This phenomena was first described by Kosterlitz and Thouless
(1973). However, the vortices in the QHE and FQHE are different than those of Kosterlitz and
Thouless, and their theory does not actually apply to the present system. These differences are
elucidated by Girvin (1987).
The Laughlin wave function explains quite well the states with filling v = lip where pis
an odd integer. The experimentals show many other fractions, and these are not explained by
the Laughlin wave function. There are various schemes to explain these states, such as one
proposed by Jain (1990). Another interesting feature is that the higher Landau levels show
striped phases at very high magnetic field (Lilly et al., 1999).
PROBLEMS
1. Derive the fonnula for the Bose-Einstein transition temperature T).. of a gas of bosons in two
dimensions.
2. In the Bogoliubov theory of superfluid 4He, derive the tenns in the interaction potential which
contain three C operators whose wave vectors are not zero. Identify the physics behind each different
kind of tenn.
3. Use CBFs to derive the McMillan result that the average energy per particle is
(11.378)
4. Use the result (11.91) for S(Q, co) to evaluate the moments (T = 0)
f
oo
-00
dco
4n sgn(co)S(Q, co) = 2S(Q) (11.379)
foo
-00
dco
-coS(Q, co) =?
4n
(11.380)
foo dco
-co3 S(Q, co) =?
-00 4n
(11.381)
5. Consider the case where the liquid helium is flowing down the pipe with velocity vs , and the pipe
walls are moving with velocity VN' Derive an expression for the total momentum as a function of
Ps' PN' vs ' VN'
6. For the superfiuid 4He flowing through a pipe, derive an expression for the total energy to order
O(v). Show it is not exactly given by the superfiuid density Ps'
7. Use the paramagnon Hamiltonian (11.208) to evaluate the self-energy of quasiparticles by summing
the set of diagrams in (a) Fig. 11.20(a) and (b) Fig. l1.20(b).
(a)
Q+R + +
( b)
+ +
FIGURE 11.20
8. Use the paramagnon theory to study the effective interaction between quasiparticles. By summing
the diagrams in Fig. 11.21, show that the effective interaction can be written as
(11.382)
FIGURE 11.21
9. Calculate the quasiparticle self-energy in normal 3He arising from the diagram in Fig. 11.22, and
use the result to find the quasiparticle lifetime. Show that the result is (11.221) but without the factor
1 - nj.
Problems 763
FIGURE 11.22
10. Use the result of Problem 7 [or Eq. (11.221)] to find the numerical value of the mean-free-path ofa
dressed quasiparticle in 3He at ~J = 0 at T = 1 mK.
11. Use Eqs. (11.209) to calculate the thermal conductivity and verify the result kBT = 33 erglcm-s in
Table 1104.
12. Find the zero-temperature gap function for the solution (11.266) to the gap equation for triplet
pairing.
13. Show by explicit calculation that the choice (11.269) for the BWorbital alignment has the same
gap function 1l(P) as the choice (11.256).
14. Calculate the pair distribution function g(fJ - (2) in a system of only three particles using Eq.
(11.331) for the case v = 1.
15. Use Eq. (11.331) for N » 1 to calculate the three-body distribution function g(fJ, f2' (3) for the
=
case that v 1.
16. Use the Hartree-Fock wave function for the QHE with v = 1 to calculate the ODLRO R(p, p').
17. In the QHE for v = 1, evaluate the partition function to derive the thermodynamic potential Qo.
18. Show that the Laughlin wavefunction can be written exactly as
qt = N[ni>je-U(Z,-Zj)]CM( ~Zl) (11.383)
<l>CM(Z) = exp [ - 4~ IZI2] (11.384)
1
u(z) = -p 1n(z) + 4N IzI2 (11.385)
If there is no center of mass motion (LZi = 0) then the wave function has exactly the Jastrow
form. Use this form to show the den~ity is uniform.
References
Chapter 1
ADLER, D., in Solid State Physics, Vol. 21, eds. H. Ehrenreich, F. Seitz, and D. Turnbull (Academic Press, New York,
1967), pp. 1-113.
ANDERSON, P. w., Phys. Rev. 124, 41 (1961).
BORN, M., and K. HUANG, Dynamical theory of Crystal Lattices (Oxford University Press, London, 1954).
CALLEN, H., in Physics of Many-Particle Systems, Vol. 1, ed. E. Meeron (Gordon & Breach, New York, 1966), pp.
183-230.
DOMB, C., and M. S. GREEN, Phase Transitions and Critical Phenomena, Vo1s. 1--6 (Academic Press, New York,
1972-1977).
FANO, U, Phys. Rev. 124, 1866 (1961).
FRoHLICH, H., H. PELZER, and S. ZIENAU, Phi/os. Mag. 41,221 (1950).
FRoHLICH, H., Adv. Phys. 3, 325 (1954).
GERRITSEN, A. N., and 1. O. LINDE, Physica 17, 573, 584 (1951); 18, 872 (1954).
GUTZWILLER, M. C., Phys. Rev. Lett. 10, 159 (1963).
HARRISON, W. A., Pseudopotentials in the Theory of Metals (Benjamin, Reading, Mass., 1966).
HEINE, v., in Solid State Physics, Vol. 24, eds. H. Ehrenreich, F. Seitz, and D. Turnbull (Academic Press, New York,
1970).
HILL, T. L., Statistical Mechanics (McGraw-Hill, New York, 1956), Chap. 7.
HOPFIELD, 1. 1., Phys. Rev. 112, 1555 (1958).
HUBBARD, J., Proc. R. Soc. London Ser. A 276, 238 (1963); 277, 237 (1964); 281, 401 (1964); 285,542 (1965); 296,
82, 100 (1966).
JORDAN, P., and E. WIGNER, Z. Phys. 47, 631 (1928).
KONDO, 1., Prog. Theor. Phys. (Kyoto) 32, 37 (1964).
KONDO, 1., in Solid State Physics, Vol. 23, eds. H. Ehrenreich, F. Seitz, and D. Turnbull (Academic Press, New York,
1969), pp. 183-281.
LIEB, E. H., and F. Y. Wu, Phys. Rev. Lett. 20, 1445 (1968).
MAHAN, G. D., J. Phys. Chem. Solids 26, 751 (1965).
MAHAN, G. D., in Polarons in Ionic Crystals and Polar Semiconductors, ed. 1. T. Devreese (North-Holland,
Amsterdam, 1972), pp. 553--657.
MARADUDIN, A. A., E. W MONTROLL, and G. H. WEISS, Theory of Lattice Dynamics in the Harmonic
Approximation (Academic Press, New York, 1963).
MATSUBARA, T., and H. MATSUDA, Prog. Theor. Phys. (Kyoto) 16,416 (1956).
MOZER, B., L. A. DEGRAAF, and B. LENEINDRE, Phys. Rev. A 9, 448 (1974).
ONSAGER, L., Phys. Rev. 65, 117 (1944).
PARR, R. G., Quantum Theory of Molecular Electronic Structure (Benjamin, Reading, Mass., 1964).
PERCUS, 1. K., in The Equilibrium Theory of Classical Fluids, eds. H. L. Frisch and 1. L. Lebowitz (Benjamin,
Reading, Mass., 1964), pp. II-33-II-170.
SCHIFF, L. I., Quantum Mechanics (McGraw-Hill, New York, 1968).
765
766 References
SCHRIEFFER,1. R., and P. A. WOLFF, Phys. Rev. 149,491 (1966).

THOMAS, D. G., J. Appl. Phys. Suppl. 32,2298 (1961).
ZENER, C., Phys. Rev. 81,440 (1951).
Chapter 2
This reference list contains several books which have excellent descriptions of the material in this chapter.
ABRIKOSOY, A. A., L. P. GORKoY, and 1. E. DZYALOSHINSKl, Methods of Quantum Field Theory in Statistical Physics
(Prentice-Hall, Englewood Cliffs, N.J., 1963; Pergamon, Elmsford, N.Y., 1965).
AUERBACH, A., Interacting Electronics and Quantum Magnetism (Springer-Verlag, New York, 1994).
CRAIG, R., J. Math. Phys 9, 605 (1968).
DONIACH, S., and E. H. SONDHEIMER, Green s Functions for Solid State Physicists (Benjamin, Reading, Mass.,
1974).
FETTER, A. L., and 1. D. WALECKA, Quantum Theory of Many Particle Systems (McGraw-Hill, New York, 1971).
GROSS, E. K. u., E. RUNGE, and O. HEINONEN, Many-Particle Theory (Hilger, New York, 1991).
GELL-MANN, M., and F. Low, Phys. Rev. 84, 350 (1951).
INKSON,1. c., Many-Body Theory of Solids (Plenum, New York, 1984).
KELDYSH, L., Sov. Phys. JETP 20, 1018 (1965).
MATTOCK, R. D., A Guide to Feynman Diagrams in the Many Body Problem (McGraw-Hill, New York, 1967).
SCHWINGER, 1., J. Math. Phys. 2,407 (1961).
Chapter 3
ABRIKOSOY, A. A., L. P. GORKOY, and 1. E. DZYALOSHINSKl, Methods of Quantum Field Theory in Statistical Physics
(Prentice-Hall, Englewood Cliffs, N.J., 1963).
BARNARD, R. D., Thermoelectricity in Metals and Alloys (Taylor & Francis, London, 1972).
BRANDT, w., and A. DUPASQUIER, editors of Positron Solid State Physics, Int. School "Enrico Fermi" (Varenna, Italy,
1981).
BROUT, R., and P. CARRUTHERS, Lectures on the many-Electron Problem (Wiley-Interscience, New York, 1963).
DE GROOT, S. R., Thermodynamics of Irreversible Processes (North-Holland, Amsterdam, 1952).
DE GROOT, S. R., and P. MAzUR, Non-Equilibrium Thermodynamics (North-Holland, Amsterdam, 1962).
DUNN, D., Can. J. Phys. 53, 321 (1975).
GREEN, M. S., J. Chem. Phys. 20, 1281 (1952); 22, 398 (1954).
KOHN, w., and 1. M. LUTTINGER, Phys. Rev. 118,41 (1960).
KUBo, R., Lectures in Theoretical Physics, Vol. I (Boulder) (Wiley-Interscience, New York, 1959), pp. 120--203;
J. Phys. Soc. Japan 12, 570 (1957).
LANGRETH, D., in NATO Advanced Study Institute on Linear and Nonlinear Transport, ed. 1. T. Devreese, V. E. van
Doren (Plenum, New York, 1976), p. 3.
LEHMANN, H., Nuovo Omento 11, 342 (1954).
LUTTINGER,1. M., Phys. Rev. 135, AI505 (1964).
LUTTINGER, 1. M., and 1. C. WARD, Phys. Rev. 118, 1417 (1960).
MAHAN, G. D., Phys. Rev. B 11, 4814 (1975).
MATSUBARA, T., Prog. Theor. Phys. (Kyoto) 14, 351 (1955).
MILLS, A. P., W. S. CRANE, and K. F. CANTER, editors of Positron Studies of Solids, Surfaces, Atoms (World
Scientific, Singapore, 1986).
TAYLOR, P. L., A Quantum Approach to the Solid State (Prentice-Hall, Englewood Cliffs, N.1., 1970).
TOMONAGA, S., Prog. Theor. Phys. (Kyoto) 2, 6 (1947).
WIGNER, E., Phys. Rev. 40, 749 (1932).
Chapter 4
AGRANOVICH, V. M., Sov. Phys. JETP 10, 307 (1960); Sov. Phys. Solid State 3, 592 (1961).
ALMBLADH, C. 0., and P. MINNHAGEN, Phys. Rev. B 17, 929 (1978).
References 767
ANDERSON, P. W, Phys. Rev. 124,41 (1961).

ANDERSON, P. W, and W L. McMILLAN, in Theory of Magnetism in Transition Metals, Course XXXVII, Enrico
Fermi International School of Physics, Varenna, ed. W Marshall (Academic Press, New York, 1967), pp. 50-86.
BAXTER, R. J., Exactly Solved Models in Statistical Mechanics (Academic, New York, 1982).
BORN, M., and K. HUANG, Dynamical Theory of Crystal Lattices (Oxford University Press, New York, 1954); see Fig.
18a on p. 91.
BRINKMAN, W E, and T. M. RICE, Phys. Rev. B 2, 1324 (1970).
BRUECKNER, K. A., C. A. LEVINSON, and H. M. MAHMOUD, Phys. Rev. 95,217 (1954).
CINI, M., and A. D'ANDREA, J. Phys. C 21, 193 (1988).
COHEN, M. H., and E KEFFER, Phys. Rev. 99, 1128 (1955).
DAVYDOV; A. S., Theory of Molecular Excitons (Plenum, New York, 1971).
DEWITT, B. S., Phys. Rev. 103, 1565 (1956).
DIETZ, R. E., D. G. THOMAS, and J. J. HOPFIELD, Phys. Rev. Lett. 8, 391 (1962).
EDWARDS, S. E, Phi/os. Mag. 3, 33, 1020 (1958).
ENGELSBERG, S., and B. B. VARGA, Phys. Rev. 136, A1582 (1964).
FANO, u., Phys. Rev. 103, 1202 (1956); 118,451 (1960).
FANO, u., Phys. Rev. 124, 1866 (1961).
FEYNMAN, R. P., Phys. Rev. 84, 108 (1951).
FRlEDEL, J., Phi/os. Mag. 43, 153 (1952); Adv. Phys. 3, 446 (1953).
FUKUDA, N., and R. G. NEWTON, Phys. Rev. 103, 1558 (1956).
FUMI, E G., Phi/os. Mag. 46, 1007 (1955).
HEEGER, A., in Chemistry and Physics of One-Dimensional Metals, ed. H. J. Keller (Plenum, New York, 1977),
pp.87-135.
HENRY, C. H., and J. J. HOPFIELD, Phys. Rev. Lett. 15,964 (1965).
HOPFIELD, J. J., Phys. Rev. 112, 1555 (1958).
KOHN, W, and J. M. LUTTINGER, Phys. Rev. 108,590 (1957).
LANGER, J. S., and V AMBEGAOKAR, Phys. Rev. 121, 1090 (1961).
LIEB, E. H., and D. C. MATTIS, Mathematical Physics in One Dimension (Academic Press, New York, 1966) Chap. 4.
LUTHER, A., and V J. EMERY, Phys. Rev. Lett. 33, 589 (1974).
LUTHER, A., and I. PESCHEL, Phys. Rev. B 9, 2911 (1974).
LUTTINGER, J. M., J. Math. Phys. N Y. 4, 1154 (1963).
MAHAN, G. D., in Electronics Structure of Polymers and Molecular Crystals, eds. J. M. Andre and J. Ladik (Plenum
Press, New York, 1975), pp. 79-158.
MAHAN, G. D., and E H. CLARO, Phys. Rev. B 16, 1168 (1977).
MATTIS, D. C., The Many-Body Problem (World Scientific, Singapore, 1993).
MATTIS, D. c., and E. H. LIEB, J. Math. Phys. 6,304 (1965).
OVERHAUSER, A. W, Physics 1, 307 (1965).
SCH6NHAMMER, K., and O. GUNNARSSON, Solid State Commun. 23,691 (1977).
SCHWEBER, S. S., Relativistic Quantum Field Theory (Harper & Row, New York, 1961).
TOMONAGA, S., Prog. Theor. Phys. (Kyoto) 5, 544 (1950).
VAN HAERlNGEN, W, Phys. Rev. 137, A1902 (1965).
WILLIAMS, E E., and M. H. HEBB, Phys. Rev. 84, 1181 (1951).
Chapter 5
BARDEEN, J., Phys. Rev. 49, 653 (1936).

BAYM, G., and L. P. KADANOFF, Phys. Rev. 124,287 (1961).
BETHE, H., Ann. Phys. 5, 325 (1930).
BROSENS, E, L. E LEMMENS, and J. T. DEVREESE, Phys. Status Solidi (b) 81, 551 (1977).
CARE, C. M., and N. H. MARCH, Adv. Phys. 24, 101 (1975).
CARR, W J., Phys. Rev. 122, 1437 (1961).
CARR, W J., and A. A. MARADUDIN, Phys. Rev. 133, A371 (1964).
DANIEL, E., and S. H. VOSKO, Phys. Rev. 120,2041 (1960).
EHRENREICH, H., and M. H. COHEN, Phys. Rev. 115, 786 (1959).
EISENBERGER, P., L. LAM, P. M. PLATZMAN, and P. SCHMIDT, Phys. Rev. B 6, 3671 (1972).
768 References
ECHENIQUE, P. M., 1. M. PITARKE, E. V. CHULKOY, and A. RUBIO, Chemical Physics 251, 1 (2000).
FERMI, E., Z. Phys. 48, 73 (1928).
FLESZAR, A., A. A. QUONG, and A. G. EGUILUZ, Phys. Rev. Lett. 74, 590 (1995).
FLODSTROM, S. A., R. Z. BACHRACH, R. S. BAUER, 1. C. MCMENAMIN, and S. B. M. HAGSTROM, J. Tilc. Sci. Technol.
14, 303 (1977).
GELDART, D. 1. W, and R. TAYLOR, Can. J. Phys. 48, 155, 167 (1970).
GELL-MANN, M., and K. BRUECKNER, Phys. Rev. 106, 364 (1957).
HAMMERBERG, 1., and N. W ASHCROFT, Phys. Rev. B 9, 409 (1974).
HEDIN, L., Phys. Rev. 139, A 796 (1965).
HOPFIELD, 1. 1., Phys. Rev. B 2, 973 (1970).
HUBBARD, 1., Proc. R. Soc. London Ser. A 243, 336 (1957).
JONSON, M., J. Phys. C 9, 3055 (1976).
KIMBALL, 1. C., Phys. Rev. A 7, 1648 (1973).
KRAMERS, H. A., reprinted in Collected Scientific Papers (North-Holland, Amsterdam, 1956).
KRAVCHENKO, S. v., G. V. KRAVCHENKO, 1. E. FURNEAUX, V. M. PUDALOY, and M. D'IORIO, Phys. Rev. B 50, 8039
(1994).
KRAVCHENKO, S. v., W E. MASON, G. E. BOWKER, 1. E. FURNEAUX, V. M. PuDALOY, and D'IoRIo, Phys. Rev. B 51,
7038 (1995).
KRONIG, R., J. Opt. Soc. Am. 12, 547 (1926).
LANDAU, L. D., and E. LIFSHITZ, Quantum Mechanics: Non Relativistic Theory (Addison Wesley, Reading, Mass.,
1958).
LANG, N. D., and W KOHN, Phys. Rev. B 3, 1215 (1971).
LANGER, 1. S., and S. H. VOSKO, J. Phys. Chem. Solids 12, 196 (1960).
LARSON, B. C., 1. Z. TISCHLER, E. D. ISAACS, P. ZSCHACK, A. FLESZAR, and A. G. EQUILUZ, Phys. Rev. Lett. 77, 1346
(1996).
LINDGREN, L., and A. ROSEN, unpublished (1970).
LINDHARD, 1., K. Dan. Vidensk. Selsk. Mat. Fys. Medd. 28, (8) (1954).
LUNDQVIST, B. L., Phys. Status Solidi 32, 273 (1969).
MAHAN, G. D., and B. E. SERNELIUS, Phys. Rev. Lett. 62,2718 (1989).
MAHAN, G. D., Mod. Phys. Lett. 7, 13 (1993).
MAo, H. K., and R. 1. HEMLEY, Rev. Mod. Phys. 66,671 (1994).
MARTON, L., 1. A. SIMPSON, H. A. FOWLER, and N. SWANSON, Phys. Rev. 126, 182 (1962).
MIHARA, N., and R. D. PUFF, Phys. Rev. 174,221 (1968).
NIKLASSON, G., Phys. Rev. B 10, 3052 (1974).
NOZIERES, P., Interacting Fermi Systems (Benjamin, Reading, Mass., 1964), p. 287.
NOZIERES, P., and D. PINES, Phys. Rev. 111, 442 (1958).
ONSAGER, L., L. MITTAG, and M. 1. STEPHEN, Ann. Phys. 18,71 (1966).
PANDEY, K. C., and L. LAM, Phys. Lett. 43A, 319 (1973).
QUINN, 1. 1., and R. A. FERRELL, Phys. Rev. 112, 812 (1958).
RiCE, T. M., Ann. Phys. 31, 100 (1965).
RiTCHIE, R. H., Phys. Rev. 106, 874 (1957).
SANTORO, G. E., and G. F. GIULIANI, Phys. Rev. B 37, 4813 (1988).
SAWADA, K., K. A. BRUECKNER, N. FUKADA, and R. BROUT, Phys. Rev. 108,507 (1957).
SCHNEIDER, T., R. BROUT, H. THOMAS, and 1. FEDER, Phys. Rev. Lett. 25, 1423 (1970).
SEITZ, F., Modern Theory of Solids (McGraw-Hill, New York, 1940), Sec. 76.
SHAM, L. 1., Phys. Rev. B 7, 4357 (1973).
SHOLL, C. A., Proc. Phys. Soc. London 92, 434 (1967).
SHUNG, K. W K., B. E. SERNELIUS, and G. D. MAHAN, Phys. Rev. B 36,4499 (1987).
SINGWI, K. S., and M. P. TOSI, Solid State Phys. 36, 177 (1981).
SINGWI, K. S., M. P. TOSI, R. H. LAND, and A. SJOLANDER, Phys. Rev. 176, 589 (1968).
SINGWI, K. S., A. SJOLANDER, M. P. TOSI, and R. H. LAND, Phys. Rev. B 1, 1044 (1970).
SUEOKA, 0., J. Phys. Soc. Jpn. 20,2203 (1965).
THOMAS, L. H., Proc. Cambridge Phi/os. Soc. 23, 542 (1927).
TING, C. S., T. K. LEE, and 1. 1. QUINN, Phys. Rev. Lett. 34, 870 (1975).
TRACY, 1. C., J. Tilc. Sci. Technol. 11,280 (1974).
VASHISHTA, P., and K. S. SINGWI, Phys. Rev. B 6, 875 (1972).
VIGNALE, G., Phys. Rev. B 38, 6445 (1988).
References 769
WIGNER, E., Phys. Rev. 46, 1002 (1934).

WIGNER, E., Trans. Faraday Soc. 34, 678 (1938).
WIGNER, E., and E. SEITZ, Phys. Rev. 43, 804 (1933); 46, 509 (1934).
Chapter 6
ABRIKOSOY, A. A., Physics 2, 61 (1965).

ANDERSON, P. W, Phys. Rev. 124,41 (1961).
ANDREI, N., K. FURUYA, and 1. H. LOWENSTEIN, Rev. Mod. Phys. 55, 331 (1983).
BARNES, S. E., J Phys. F 6, 1375 (1976); 7, 2637 (1977).
BICKERS, N. E., Rev. Mod. Phys. 59, 845 (1987).
BICKERS, N. E., D. L. Cox, and 1. W WILKINS, Phys. Rev. B 36, 2036 (1987).
CHEN, H., and D. C. MATTIS, Int. J Mod. Phys. B 5, 2951 (1991).
COLEMAN, P., Phys. Rev. B 29, 3035 (1984).
GREWE, N., and H. KEITER, Phys. Rev. B 24, 4420 (1981).
GUNNARSSON, 0., and K. SCHONHAMMER, Phys. Rev. B 28,4315 (1983).
GUNNARSSON, 0., and K. SCHONHAMMER, in Handbook on the Physics and Chemistry of Rare Earths ed. K. A.
Gschneider, L. Eyring, and S. Hafner (North-Holland, Amsterdam, 1987), Vol. 10, p. 103.
HERBST, J. E, and J. W WILKINS, Handbook of the Physics and Chemistry of Rare Earths, Vol. 10, ed. K. A.
Gschneider, L. Eyring, and S. Hafner (Elsevier, New York, 1987), Chap. 68.
HEWSON, A. C., The Kondo Problem to Heavy Fermions (Cambridge University Press, Cambridge, 1993).
JACCARD, D., and J. SIERRO, in Valence Instabilities, ed. P. Wachter, and H. Boppart (North-Holland, Amsterdam,
1982), p. 409.
KANAMORI, 1., Prog. Theor. Phys. 30,275 (1963).
KONDO, J., Prog. Theor. Phys. 32, 37 (1964).
LEE, P. A., T. M. RICE, J. W SERENE, L. J. SHAM, and J. W WILKINS, Comments on Condensed Matter Phys. XII,
99-161 (1986).
Lru, S. H., Phys. Rev. B 37, 3542 (1988).
MAHAN, G. D., in Solid State Physics, ed. H. Ehrenreich, and E Spaen, Vol. 51 (Academic Press, New York, 1997),
pp. 81-157.
MAHAN, G. D., and J. O. SOFO, Proc. Nat!. Acad. Sci. USA 93, 7436 (1996).
MERMIN. N. D., and H. WAGNER, Phys. Rev. Lett. 17, 1133 (1966).
OVERHAUSER, A. W, Phys. Rev. Lett. 13, 190 (1964).
OVERHAUSER, A. W, Adv. Phys. 27,343 (1978).
PENN, D. R., Phys. Rev. 142, 350-365 (1966).
SAMWER, K., and K. WINZER, Z. Phys. B 25, 269-274 (1976).
SCHRIEFFER, 1. R., and P. A. WOLFF, Phys. Rev. 149,491 (1966).
SILVERSTEIN, S. D., and C. B. DUKE, Phys. Rev. 61,456,470 (1967).
STEWART, G. R., Rev. Mod. Phys. 56,755 (1984).
VILK, Y. M., L. CHEN, and A. M. S. TREMBLAY, Phys. Rev. B 49, 13267-13270 (1994).
VILK, Y. M., and A. M. S. TREMBLAY, J Phys. I (France) 7, 1309 (1997).
WIEGMANN, P., J Phys. C 14, 1463 (1981).
WILKINS, J. W, in Valence Instabilities, ed. P. Wachter, and H. Boppart (North-Holland, New York, 1982).
YOSIDA, K., Phys. Rev. 147, 223 (1966).
ZWICKNAGL, G., Adv. Phys. 41, 203 (1992).
Chapter 7
ADLER, D., in Solid State Physics. Vol. 21, eds. H. Ehrenreich, E Seitz, and D. Turnbull (Academic Press, New York,
1967), p. 1.
ALLCOCK, G. R., Adv. Phys. 5,412 (1956).
ALLCOCK, G. R., in Polarons and Excitons, eds. C. G. Kuper, and G. D. Whitfield (Oliver & Boyd, Edinburgh, 1962),
pp.45-70.
ApPEL, 1., Phys. Rev. Lett. 17, 1045 (1966).
ApPEL. J., in Solid State Physics, Vol. 21, eds. H. Ehrenreich, E Seitz, and D. Turnbull (Academic Press, New York,
1967), p. 193.
770 References
ASHCROFT, N. w., and J. W. WILKINS, Phys. Lett. 14,285 (1965).

BOGOMOLOV, V N., E. K. KUDINOV, and Yu. A. FIRSOV, Sov. Phys. Solid State 9,2502 (1968).
BOHM, D., and 1. STAVER, Phys. Rev. 84, 836 (1951).
BOTTGER, H., and V V BRYKSIN, Phys. Status Solidi B 78, 415 (1976).
BRILLOUIN, L., J. Phys. Paris 3, 379 (1932); 4, 1 (1933).
BROUT, R., and P. CARRUTHERS, Lectures on the Many-Electron Problem (Wiley-Interscience, New York, 1963).
BROWN, D. w., K. LINDENBERG, and Y. ZHAO, J. Chem. Phys. 107,3159,3179 (1997).
CONLEY, J. w., and G. D. MAHAN, Phys. Rev. 161,681 (1967).
DEWIT, H. J., Philips Res. Rep. 23,449 (1968).
DUNN, D., Can. J. Phys. 53, 321 (1975).
ENGELSBERG, S., and J. R. SCHRIEFFER, Phys. Rev. 131,993 (1963).
FEYNMAN, R. P., Phys. Rev. 97,660 (1955).
GARRETT, D. G., and J. C. SWIHART, J. Phys. F 6, 1781 (1976).
GRIMVALL, G., The Electron Phonon Interaction in Metals (North-Holland, New York, 1981).
HARRISON, W. A., Pseudopotentials in the Theory Of Metals (Benjamin, Reading, Mass., 1966).
HODBY, J. w., in Polarons in Ionic Crystals and Polar Semiconductors, ed. J. Devreese (North-Holland, Amsterdam,
1972), pp. 389 -459.
HOLSTEIN, T., Annals of Phys. 8,343 (1959).
JOSHI, S. K., and A. K. RAJAGOPAL, in Solid State Physics, Vol. 22, eds. H. Ehrenreich, E Seitz, and D. Turnbull
(Academic Press, New York, 1968), pp. 160-313.
KARTHEUSER, E., R. EVRARD, and J. DEVREESE, Phys. Rev. Lett. 22,94 (1969).
KAY, P., and J. A. REISSLAND, J. Phys. F 6, 1503 (1976).
KOHN, w., Phys. Rev. Lett. 2,395 (1959).
KUDINOV, R. K., D. N. MIRLIN, and Yu. A. FIRsov, Sov. Phys. Solid State 11, 2257 (1970).
LANDAU, L. D., and S. L. PEKAR, Zh. Eksp. Teor. Fiz. 16, 341 (1946).
LANG, 1. G., and Yu. A. FIRSOV, Sov. Phys. JETP 16, 1301 (1963); Sov. Phys. Solid State 5, 2049 (1964).
LOw, E, T. D. LEE, and D. PINES, Phys. Rev. 90,297 (1953).
LYDDANE, R. H., R. G. SACHS, and E. TELLER, Phys. Rev. 59,673 (1941).
MAHAN, G. D., Phys. Rev. 145, 602 (1966).
MAHAN, G. D., in Polarons in Ionic Crystals and Polar Semiconductors, ed. J. Devreese (North-Holland, Amsterdam,
1972), pp. 553-{j57.
McMILLAN, W. L., Phys. Rev. 167,331 (1968).
MIGDAL, A. B., Sov. Phys. JETP 7, 996 (1958).
MIYAKE, S. J., J. Phys. Soc. Jpn. 41, 745 (1976).
MOORADIAN, A., and G. B. WRIGHT, Phys. Rev. Lett. 16,999 (1966).
NAGAEV, E. L., Sov. Phys. Solid State 4, 1611 (1963).
NAKAJIMA, S., and M. WATABE, Prog. Theor. Phys. 29,341 (1963).
REIK, H. G., in Polarons in Ionic Crystals and Polar Semiconductors, ed. J. Devreese (North-Holland, Amsterdam,
1972), pp. 679-714.
ROWELL, J. M., W. L. McMILLAN, and W. L. FELDMANN, Phys. Rev. 178,897 (l969a); Phys. Rev. 180,658 (l969b);
Phys. Rev. B 3, 4065 (1971).
SCHIFF, L. 1., Quantum Mechanics, 2nd ed. (McGraw-Hill, New York, 1955), p. 199.
SCHOTTE, K. D., Z. Phys. 196, 393 (1966).
SCHRIEFFER, J. R., Theory of Superconductivity (Benjamin, Reading, Mass., 1964).
SHAM, L. J., and J. M. ZIMAN, Solid State Phys., Vol. 15, eds. H. Ehrenreich, E Seitz, and D. Turnbull (Academic
Press, New York, 1963), p. 221.
SHUKLA, R. C., and R. TAYLOR, J. Phys. F 6,531 (1976).
SMONDYREV, M. A., Theor. Math. Phys. (Russian) 68, 653 (1986).
SRIVASTAVA, P. L., and R. N. SINGH, J. Phys. F 6, 1819 (1976).
SWIHART, J. C., J. SCALAPINO, and Y. WADA, Phys. Rev. Lett. 14, 106 (1965).
TiABLIKOV, S. V, Zh. Eksp. Teor. Fiz. 23, 381 (1952).
TOMLINSON, P. G., and J. P. CARBOTTE, Solid State Commun. 18, 119 (1976).
TSUI, D. C., Phys. Rev. B 10, 5088 (1974).
WIGNER, E. P., Math. U. Naturw. Anz. Ungar. Akad. Wiss. 53,477 (1935).
WOLL, E. J., and W. KOHN, Phys. Rev. 126,.1693 (1962).
YAMASHITA, J., and T. KUROSAWA, J. Phys. ChBI'Q. Solids 5, 34 (1958).
ZIMAN, J. M., Electrons and Phonons (Oxford University Press, Inc., New York, 1960).
References 771
Chapter 8
ALLEN, P. B., Phys. Rev. B 3, 305 (1971).

ALLEN, P. B., Phys. Rev. Lett. 59, 1460 (1987).
ARGYRES, P. N., and 1. L. SIGEL, Phys. Rev. B 9, 3197 (1974).
BASS, 1., Landolt-Bornstein, Neue Serie, Band 15a (Springer-Verlag, New York, 1982).
CAROLl, C., R. COMBESCOT, D., LEDER-RoZENBLATT, P. NOZIERES, and D. SAINT JAMES, Phys. Rev. B 12, 3977
(1975).
COHEN, M. H., L. M. FALICOV, and 1. C. PHILLIPS, Phys. Rev. Lett. 8, 316 (1962).
DYNES, R. C., and 1. P. CARBOTTE, Phys. Rev. 175,913 (1968).
ENGELSBERG, S., and 1. R. SCHRIEFFER, Phys. Rev. 131,993 (1963).
FEUCHTWANG, T. E., Phys. Rev. B 12, 3979 (1975).
FISHMAN, R., Phys. Rev. B 39, 2994 (1989).
FLEUROV, V. N., and A. N. KOZLOV, J. Phys. F 8, 1899 (1978).
GIAEVER, I., and H. R. ZELLER, Phys. Rev. Lett. 20, 1504 (1968).
GoLDHABER-GoRDON, D., 1. GORES, M. A. KASTNER, H. SHTRlKMAN, D. MAHALU, and U. MEIRAV, Phys. Rev. Lett.
1, 5225 (1998).
GRIMVALL, G., The Electron Phonon Interaction in Metals (North-Holland, New York, 1981).
HODBY, 1. w., in Polarons in Ionic Crystals and Polar Semiconductors, ed. J. Devreese (North-Holland, Amsterdam,
1972), pp. 389-459.
HOLSTEIN, T., Ann. Phys. (N.Y.) 29, 410 (1964).
HOWARTH, D. 1., and E. H. SONDHEIMER, Proc. R. Soc. London Ser. A 219, 53 (1953).
HUBERMAN, M., and G. V. CHESTER, Adv. Phys. 24, 489 (1975).
KADANOFF, L. P., and G. BAYM, Quantum Statistical Mechanics (Benjamin, New York, 1962).
KITTEL, C., Quantum Theory of Solids (Wiley, New York, 1963) Chap. 9.
KUBO, R., M. TODA, and N. HASHITSUME, Statistical Physics II (Springer-Verlag, New York, 1985).
LANDAUER, R., Phys. Lett. 85A, 91 (1981).
LANDAUER, R., J. Phys.: Condo Matter 1, 8099 (1989).
LANGER, 1. S., Phys. Rev. 120,714 (1960).
LANGER, 1. S., Phys. Rev. 124,997, 1003 (1961).
LANGRETH, D. C., Phys. Rev. 159,717 (1967).
LANGRETH, D. C., and L. P. KADANOFF, Phys. Rev. 133, AI070 (1964).
LEVANDA, M., and V. N. FLEUROV, J. Phys.: Condo Matter 6, 789 (1994).
MAHAN, G. D., Phys. Rev. 142,366 (1966).
MAHAN, G. D., J. Appl. Phys. 58, 2242 (1985).
MAHAN, G. D., Phys. Rep. 145, 251 (1987).
MAHAN, G. D., and W. HANSCH, J. Phys. F 13, L47 (1983); Phys. Rev. B 28, 1902 (1983).
MATTHIESSEN, A., Rep. Br. Assoc. 32, 144 (1862).
MIGDAL, A. B., Sov. Phys. JETP 7, 996 (1958).
OPSAL, 1. L., B. J. THALER, and 1. BASS, Phys. Rev. Lett. 36, 1211 (1976).
PRANGE, R. E., and L. P. KADANOFF, Phys. Rev. 134, A566 (1964).
SCHRIEFFER, 1. R., D. 1. SCALAPINO, and 1. W. WILKINS, Phys. Rev. Lett. 10,336 (1963).
SEGALL, B., M. R. LORENZ, and R. E. HALSTAD, Phys. Rev. 129,2471 (1963).
TAKEGAHARA, K., and S. WANG, J. Phys. F 7, L293 (1977).
TAYLOR, R., and A. H. MAcDONALD, Phys. Rev. Lett. 57, 1639 (1986).
WARD, 1. C., Phys. Rev. 78, 182 (1950).
ZAWADOWSKI, A., Phys. Rev. 163, 341 (1967).
ZIMAN,1. M., Electrons and Phonons (Oxford University Press, me., New York, 1960).
Chapter 9
ALLEN, P. B., Phys. Rev. B 3,305 (1971).

ALMBLADH, C. 0., Phys. Rev. B 16, 4343 (1977).
ANDERSON, P. w., Phys. Rev. Lett. 18, 1049 (1967).
ASHCROFT, N. w., and K. STURM, Phys. Rev. B 3, 1898 (1971).
772 References
CALLCOTT, T. A., and E. T. ARAKAWA, Phys. Rev. Lett. 38, 442 (1977).
CALLCOTT, T. A., E. T. ARAKAWA, and D. L. EDERER, Phys. Rev. B 18, 6622 (1978).
CANRIGHT, G. S., Phys. Rev. B 38, 1647 (1988).
CiTRIN, P. H., Phys. Rev. B 8, 5545 (1973).
CITRIN, P. H., G. K. WERTHEIM, and Y. BAER, Phys. Rev. B 16, 4268 (1977).
DEVREESE, J. T., ed., in Polarons in Ionic Crystals and Polar Semiconductors (North-Holland, Amsterdam, 1972), pp.
92-93.
DONIACH, S., and M. SUNJIC, J. Phys. C 3, 285 (1970).
ELLIOTT, R. J., Phys. Rev. 108, 1384 (1957).
GALLON, T. E., in Electron and Ion Spectroscopy of Solids, eds. L. Fiermans, J. Vennik, and W Dekeyser (Plenum
Press, New York, 1978), pp. 230-272.
GIRVIN, S. M., and J. J. HOPFIELD, Phys. Rev. Lett. 37, 1091 (1976).
GUPTA, R. P., and A. J. FREEMAN, Phys. Rev. Lett. 36, 1194 (1976).
GUREVICH, V, I. LANG, and Yu. FIRSOY, Sov. Phys. Solid State 4, 918 (1962).
HASEGAWA, M., and M. WATABE, J. Phys. Soc. Jpn. 27, 1393 (1969).
HEDIN, L., and A. ROSENGREN, J. Phys. F 7, 1339 (1977).
HOPFIELD, J. J., Phys. Rev. A 139, 419 (1965).
HOPFIELD. J. J., unpublished (1967).
HOPFIELD, J. J., Comments Solid State Phys. 11,40 (1969).
KNox, R. S., Theory of Excitons (Academic Press, New York, 1963), Supplement 5 to Solid State Physics, eds.
H. Ehrenreich, F. Seitz, and D. Turnbull.
MAHAN, G. D., Phys. Rev. 153,882 (1967); 163,612 (1967).
MAHAN, G. D., J. Phys. Chem. Solids 31, 1477 (1970).
MAHAN, G. 0. Solid State Physics, Vol. 29, eds. H. Ehrenreich, F. Seitz, and D. Turnbull (Academic Press, New York,
1974), p. 75.
MAHAN, G. D., Phys. Rev. B 15,4587 (1977).
MAHAN, G. D., Phys. Rev. B 25, 5021 (1982).
MINNHAGEN, P., J. Phys. F 7,2441 (1977).
MYERS, H. P., and P. SIXTENSSON, J. Phys. F 6,2023 (1976).
NOZIERES, P., and C. T. DEDOMINICIS, Phys. Rev. 178, 1097 (1969).
OHMURA, Y., and A. OGIWARA, Phil. Mag. B 69, 993-999 (1994).
OHTAKA, K., and Y. TANABE, Phys. Rev. B 28,6833 (1983); 34, 3717 (1986).
PARDEE, W J., and G. D. MAHAN, Phys. Lett. 45A, 117 (1973).
PARDEE, W J., G. 0. MAHAN, 0. E. EASTMAN, R. A. POLLAK, L. LEY, F. R. McFEELY, S. P. KOWALCZYK, and D. A.
SHIRLEY, Phys. Rev. B 11, 3614 (1975).
PARRATT, L. G., Rev. Mod. Phys. 31, 616, (1959).
PENN, D. R., S. M. GIRVIN, and G. D. MAHAN, Phys. Rev. B 24, 6971 (1981).
REYNOLDS, D. c., and T. C. COLLINS, Excitons: Their Properties and Uses (Academic Press, New York, 1981).
SCHoNHAMMER, K., and O. GUNNARSSON, Phys. Rev. B 18, 6606 (1978).
SCHOTTE, K., and U. SCHOTTE, Phys. Rev. 182, 479 (1969).
SIMANMEK, E., Phys. Lett. A 37, 175 (1971).
SMITH, N. V, Phys. Rev. 183, 634 (1969); Phys. Rev. B 2, 2840 (1970).
WANNIER, G. H., Phys. Rev. 52, 191 (1937).
WERTHEIM, G. K., D. M. RIFFE, and P. H. CITRIN, Phys. Rev. B 45, 8703 (1992).
Wu, J. W, and G. 0. MAHAN, Phys. Rev. B 29, 1769 (1984).
Chapter 10
ABRlKOSOY, A. A., L. P. GORKoY, and I. E. DZALOSHINSKl, Methods of Quantum Field Theory in Statistical Physics
(Prentice-Hall, Englewood Cliffs, N.J., 1963).
ALLEN, P. B., M. L. COHEN, and D. R. PENN, Phys. Rev. B 38, 2513 (1988).
ALLEN, P. B., and B. MITROVIC, in Solid State Physics, Vol. 37, eds. H. Ehrenreich, F. Seitz, and D. Turnbull
(Academic Press, New York, 1982), pp. 1-82.
AMBEGAOKAR, V, and A. BARATOFF, Phys. Rev. Lett. 10,486 (1962); (E) 11, 104 (1963).
BAKER, G. A., Jr., Essentials of Pade Approximants (Academic Press, New York, 1975).
BALIAN, R., and N. R. WERTHAMER, Phys. Rev. 131, 1553 (1963).
References 773
BARDEEN, J., L. N. COOPER, and J. R. SCHRIEFFER, Phys. Rev. 108, 1175 (1957).
COOPER, L. N., Phys. Rev. 104, 1189 (1956).
ELIASHBERG, G. M., Sov. Phys. JETP 11,696 (1960).
FISKE, M. D., unpublished (1978).
FRoHLICH, H., Phys. Rev. 79, 845 (1950).
GIAEVER, r., Phys. Rev. Lett. 5, 147,464 (1960).
GIAEVER, I., H. R. HART, and K. MEGERLE, Phys. Rev. 126, 941 (1962).
HALPERIN, B. I., and T. M. RICE, in Solid State Physics, Vol. 21, eds. H. Ehrenreich, E Seitz, and D. Turnbull
(Academic Press, New York, 1968), p. 115.
Han, S., K. W. NG, E. L. WOLF, A MILLIS, J. L. SMITH, and Z. FISK, Phys. Rev. Lett. 57,238 (1986).
JOSEPHSON, B. D., Phys. Lett. 1,251 (1962).
KELDYSH, L. v., and Y. V. KoPAEV, Sov. Phys. Solid State 6,2219 (1965).
KITTEL, C., Introduction to Solid State Physics, 3rd ed. (Wiley, New York, 1966), p. 458.
LANGENBURG, D. N., D. J. SCALAPINO, B. N. TAYLOR, and R. E. EcK, Phys. Rev. Lett. 15, 294 (1965).
MAHAN, G. D., Phys. Rev. 153, 882 (1967); 163, 612 (1967).
MAHAN, G. D., Phys. Rev. B 56, 8322-8329 (1997).
MAms, D. C., and J. BARDEEN, Phys. Rev. 111, 412 (1958).
MAxWELL, E., Phys. Rev. 78,477 (1950).
McMILLAN, W. L., Phys. Rev. 167,331 (1968).
McMILLAN, W. L., and J. M. ROWELL, Phys. Rev. Lett. 14, 108 (1965).
MIGDAL, A. B. Sov. Phys. JETP 7,996 (1958).
MORSE, R. w., and H. V. BOHM, Phys. Rev. 108, 1094 (1957).
OWEN, C. S., and D. J. SCALAPINO, Physica 55, 691 (1971).
PALMER, L. H., and M. TINKHAM, Phys. Rev. 165, 588 (1968).
PARKER, W. H., D. N. LANGENBERG, A DENENSTEIN, and B. N. TAYLOR, Phys. Rev. 177, 639 (1969).
PARKS, R. D., ed., Superconductivity, Vo1s. I & II (Dekker, New York, 1969).
REYNOLDS, C. A., B. SERIN, W. H. WRIGHT, and L. B. NESBITT, Phys. Rev. 78, 487 (1950).
RICKAYZEN, G., Theory of Superconductivity (Wi1ey-lnterscience, New York, 1965).
SCALAPINO, D. J., J. R. SCHRIEFFER, and J. W. WILKINS, Phys. Rev. 148,263 (1966).
SCHAFROTH, M. R., Phys. Rev. 100,463 (1955).
SCHILLING, A, M. CANTONI, J. D. Guo, and H. R. Orr, Nature, 363, 56 (1993).
SCHRIEFFER, J. R., Theory of Superconductivity (Benjamin, Reading, Mass., 1964).
SHAW, w., and J. C. SWIHART, Phys. Rev. Lett. 20, 1000 (1968).
SHEKHTMAN, L., and L. r. GLAZMAN, Phys. Rev. B 52, R2297 (1995).
VAN DER LAGE, E C., and H. A. BETRE, Phys. Rev. 71, 612 (1947).
VIDBERG, H. J., and J. W. SERENE, J. Low Temp. Phys. 29, 179 (1977).
WOLLMAN, D. A, D. J. VAN HARLINGEN, J. GIAPINTZAKIS, and D. M. GINSBERG, Phys. Rev. Let. 74, 797 (1995).
Chapter 11
ABRIKOSOV, A A., and I. M. KHALATNIKOV, Rep. Prog. Phys. 22, 329 (1959).
ACHTER, E. K., and L. MEYER, Phys. Rev. 188,291 (1969).
ANDERSON, P. w., and W. E BRINKMAN, in The Helium Liquids, eds. J. G. M. Armitage, and I. E. Farquhar (Academic
Press, New York, 1975), pp. 315-416.
BALIAN, R., and N. R. WERTHAMER, Phys. Rev. 131, 1553 (1963).
BAYM, G., and C. PETRICK, in The Physics of Liquid and Solid Helium, Part 2, eds. J. B. Ketterson, and K.
Bennemann (Wiley, New York, 1978), Chap. 1.
BERK, N. E, and J. R. SCHRIEFFER, Phys. Rev. Lett. 17,433 (1966).
BIJL, A, Physica 7, 860 (1940).
BOGOLIUBOV, N. N., J. Phys. Moscow 11, 23 (1947).
BOGOLIUBOV, N. N., and D. N. ZUBAREV, Sov. Phys. JETP 1, 83 (1955).
BROOKER, G. A., and J. SYKES, Phys. Rev. Lett. 21,279 (1968).
BROWN, G. v., and M. H. COOPERSMITH, Phys. Rev. 178,327 (1969).
COWLEY, R. A, and AD. B. WOODS, Can. J. Phys. 49, 177 (1971).
DINGLE, R. M., Phi/os. Mag. 40, 573 (1949).
774 References
DONIACH, S., and S. ENGELSBERG, Phys. Rev. Lett. 17, 750 (1966).
DY, K. S., and C. J. PETHICK, Phys. Rev. 185, 373 (1969).
FEENBERG, E., Theory of Quantum Fluids (Academic Press, New York, 1969).
FEYNMAN, R. P., Phys. Rev. 94,262 (1954).
FEYNMAN, R. P., and M. COHEN, Phys. Rev. 102, 1189 (1956); 107, 13 (1957).
FRANCIS, W P., G. V CHESTER, and L. REATTO, Phys. Rev. A 1, 86 (1970).
GIRVIN, S. M., A. H. MACDoNALD, and P. M. PLATZMAN, Phys. Rev. B 33, 2481 (1986).
GIRVIN, S. M., in The Quantum Hall Effect, eds. R. E. Prange, and S. M. Girvin (Springer-Verlag, New York, 1987).
GLYDE, H. R., Excitations in Liquid and Solid Helium (Oxford University Press, Oxford, 1994).
GREYWALL, D. S., Phys. Rev. B 27, 2747 (1983).
GREYWALL, D. S., Phys. Rev. B 33, 7520 (1986).
GRIFFIN, A., Phys. Rev. B 32, 3289 (1985).
HALLOCK, R. B., J Low Temp. Phys. 9, 109 (1972).
HARLING, 0., Phys. Rev. A 3, 1073 (1971).
HIRSCHFELDER, J. 0., C. F. CURTISS, and R. B. BIRD, Molecular Theory of Gasses and Liquids (Wiley, New York,
1954), p. 11 O.
HOHENBERG, P. c., and P. M. PLATZMAN, Phys. Rev. 152, 198 (1966).
HONE, D., Phys. Rev. 121, 669 (1961).
JAIN, J. K., J Phys. Chem. Solids 51, 889 (1990).
JANCOVICI, B., Phys. Rev. Lett. 46, 386 (1981).
JASTROW, R., Phys. Rev. 98, 1479 (1955).
JENSEN, H. H., H. SMITH, and J. W WILKINS, Phys. Lett. A 27, 532 (1968).
KLITZING, Von K., G. DORDA, and M. PEPPER, Phys. Rev. Lett. 45,494 (1980).
KOHN, W, Phys. Rev. 123, 1242 (1961).
KOHN, W, and D. SHERRINGTON, Rev. Mod. Phys. 42, I (1970).
KOSTERLITZ, J. M., and D. J. THOULESS, J Phys. C 6, 1181 (1973).
LAUGHLIN, R. B., Phys. Rev. Lett. 50, 1395 (1983).
LANDAU, L. D., J Phys. Moscow 5,71 (1941); 11,91 (1947); translations in I. M. Khalatnikov, Introduction to the
Theory of Supeifluidity (Benjamin, Reading, Mass., 1965).
LANDAU, L. D., J Phys. Moscow 10, 25 (1946); Sov. Phys. JETP 3, 920 (1956); 5, 101 (1957).
LEE, D. M., D. D. OSHEROFF, and R. C. RICHARDSON, Rev. Mod. Phys. 69,645--690 (1997).
LEGGETT, A. J., Phys. Rev. 140, AI869 (1965).
LEGGETT, A. J., Rev. Mod. Phys. 47,331 (1975).
LEVIN, K., and O. T. VALLS, Phys. Rev. B 17, 191 (1978).
LIFSHITS, E. M., and E. L. ANDRONIKASHIVILE, A Supplement to Helium (Consultants Bureau, New York, 1959),
p.75.
LILLY, M. P., K. B. COOPER, J. P. EISENSTEIN, L. N. PFEIFFER, and K. W WEST, Phys. Rev. Lett. 82,394 (1999).
LUTTINGER, J. M., and P. NOZIERES, Phys. Rev. 127, 1423,4131 (1962).
MASSEY, WE., Phys. Rev. 151, 153 (1966).
MASSEY, W E., and C. W WOO, Phys. Rev. 164,256 (1967).
McMILLAN, W L., Phys. Rev. 138, A442 (1965).
MOOK, H. A., Phys. Rev. Lett. 55, 2452 (1985).
OSHEROFF, D. D., R. C. RICHARDSON, and D. M. LEE, Phys. Rev. Lett. 28, 885 (1972).
OSHEROFF, D. D., W J. GULLY, R. C. RICHARDSON, and D. M. LEE, Phys. Rev. Lett. 29, 920 (1972).
PATTON, B. R., and A. ZARINGHALAM, Phys. Lett. ASS, 95 (1975).
PENROSE, 0., Phi/os. Mag. 42, 1373 (1951).
PENROSE, 0., and L. ONSAGER, Phys. Rev. 104, 576 (1956).
PERCUS, J. K., in The Equilibrium Theory of Classical Fluids, eds. H. L. Frisch and J. L. Lebowitz (Benjamin,
Reading, Mass., 1964), pp. 11-33-11-170.
PINES, D., and P. NOZIERES, The Theory of Quantum Liquids (Benjamin, Reading, Mass., 1966).
PRANGE, R. E., in The Quantum Hall Effect, eds. R. E. Prange and S. M. Girvin (Springer-Verlag, New York, 1987).
PUFF, R. D., and J. S. TENN, Phys. Rev. A 1, 125 (1970).
REATTO, L., and G. V CHESTER, Phys. Rev. 155, 88 (1967).
RICE, M. J., Phys. Rev. 159, 153 (1967); 162, 189 (1967); 163, 206 (1967).
RODRIQUEZ, L. J., H. A. GERSCH, and H. A. MOOK, Phys. Rev. A 9, 2085 (1974).
RON, A., in The Helium Liquids, eds. J. G. M. Armitage and L. E. Farquhar (Academic Press, New York, 1975),
pp. 211-240.
References 775
RUVALDS, 1., and A. ZAWADOWSKl, Phys. Rev. Lett. 25, 333, 632 (1970).
SKoLD, K., and C. A. PELIZZARI, Phi/os. Trans. Roy. Soc. (London) B 290, 605 (1980).
STIRLING, W G., and H. R. GLYDE, Phys. Rev. B 41, 4224 (1990).
STORMER, H. L., (unpublished) (1999).
SVENSSON, E. C., V. F. SEARS, and A. GRIFFIN, Phys. Rev. B 23, 4493 (1981).
SVENSSON, E. C., and V. F. SEARS, Physica 137B, 126 (1986).
TSUI, D. C., H. L. SToRMER, and A. C. GOSSARD, Phys. Rev. Lett. 48, 1559 (1982).
VOLLHARDT, D., and P. WoLFE, The Superfluid Phases of Helium 3 (Taylor & Francis, London, 1990).
WHEATLEY, 1. c., Rev. Mod. Phys. 47,415 (1975).
WOODS, A. D. B., and V. F. SEARS, Phys. Rev. Lett. 39,415 (1977).
YANG, C. N., Rev. Mod. Phys. 34,694 (1962).
YENNIE, D. R., Rev. Mod. Phys. 59, 781 (1987).
ZAWADOWSKl, A., 1. RUVALDS, and 1. SOLNA, Phys. Rev. A 5, 399 (1972).
Author Index
Abrikosov, A. A., 142, 148,391, Bethe, H., 369, 666, 767, 773 Citrin, P. H., 605, 625, 769, 772
635,640,729,730,766,769, Bickers, N. E., 391, 396, 401, 769 Claro, F. H., 254, 767
772, 773 Bijl, A., 690, 702, 703, 742, 773 Cohen, M., 709, 774
Achter, E. K., 722, 773 Bird, R. B., 774 Cohen, M. H., 286, 325, 561, 767,
Adler, D., 23, 463, 765, 769 Bogoliubov, N. N., 678, 682, 694, 771
Agranovich, V. M., 285, 766 773 Cohen, M. L., 772
Allcock, G. R., 448, 769 Bogomolov, V. N., 463, 770 Coleman, P., 391, 769
Allen, P. 8., 531, 542, 591, 592, Bohm, D., 486, 642, 770, 773 Collins, T. C., 596. 772
628,671,672,674,771,772 Bottger, H., 463, 770 Combescot, R., 771
Almbladh, C. 0., 247, 606, 766, Born, M., II, 276, 765, 767 Conley, J. w., 480, 770
771 Bowker, G. E., 768 Cooper, K. B., 774
Ambegaokar, v., 197,207,652, Billouin, L., 434, 770 Cooper, L. N., 627, 628, 629, 773
655, 767, 772 Brandt, w., 168,766 Coppersmith. M. H., 686, 773
Anderson, P. w., 44, 208, 214, 389, Brinkman, W. F., 247, 742, 767, Cowley, R. A., 702, 773
603,612,675,742,765,767, 773 Cox, D. L., 769
769, 771, 773 Brooker, G. A., 729, 773 Craig, R., 107,766
Andrei, N., 387, 769 Brosens, F., 337, 767 Crane, W. S., 766
Andronikashivile, E. L., 710, 774, Brout, R., 142, 152,448,766,768, Curtiss, C. F., 774
Appel, J., 463, 497, 769, 770 770
Arakawa, E. T., 606, 772 Brown, D. w., 456, 770 D'Andrea, A., 247, 767
Argyres, P. N., 536, 771 Brown, G. v., 686, 773 Daniel, E., 365, 366, 767
Ashcroft, N. w., 316, 491, 590, Brueckner, K. A., 204, 297, 300, Davydov, A. S., 282, 767
768, 770, 772 307,363,767,768 Dedominicis, C. T., 603, 611, 613,
Auerbach, A., 766 Bryksin, V. v., 463, 770 621,772
De Groot, S. R., 177, 178,766
Bachrach, R. Z., 768 Callcott, T. A., 606, 622, 623, 772 Degraaf, L. A., 765
Baer, Y., 772 Callen, H., 38, 39, 765 Denenstein, A., 773
Baker, G. A., 668, 772 Canter, K. F., 766 Devreese, J. T., 584, 585, 767, 770,
Balian, R., 635, 735, 736, 740, 772, Canright, G. S., 607, 772 772
773 Cantoni, M., 773 Dewit, H. J., 461. 770
Baratoff, A., 652, 655, 772 Carbotte,1. P., 494, 524, 537, 770, Dewitt, B. S .. 204, 767
Bardeen, 1., 302, 627, 660, 662, 771 Dietz, R. E., 236, 237, 767
767, 772, 773 Care, C. M., 313,767 Dingle, R. M., 690, 774
Barnard, R. D., 178, 181,766 Caroli, c., 561, 771 D'lorio, M., 768
Barnes, S. E., 391, 769 Carr, W. J., 307, 313, 767 Domb, c., 38, 765
Bass, 1.,535,771 Carruthers, P., 142, 152,448,766, Doniach, S., 603, 624, 626, 729,
Bauer, R. S., 768 770 766, 772, 774
Baxter, R., 187,767 Chen, H., 405, 769 Dorda, G., 774
Baym, G., 337, 550, 554, 714, 767, Chen, L., 769 Duke, C. B., 387, 769
771,773 Chester, G. V., 536, 694, 771, 774 Dunn, D., 152,44,451,452,453,
Berk, N. F., 728, 773 Chulkov, E. v., 768 766,770
Best, F. E., 333 Cini, M., 247, 767 Dupasquier, A .. 168,766
777
778 Author Index
Dy, K. S., 732, 733, 734, 774 Gersch, H. A., 775 Hone, D., 729, 730, 774
Dynes, R. c., 524, 537, 771 Giaever, I., 567, 628, 646, 647, Hopfield, J. J., 52, 276, 278, 279,
Dzyaloshinski, I. E., 766, 772 650,651,660,771,773 361,584,612,620.623,626,
Giapintzakis, J., 773 765,767,768,772
Eastman, D. E., 772 Ginsberg, D. M., 773 Howarth, D. J., 519, 771
Echenique, P. M., 370, 768 Girvin, S. M., 623, 626, 757, 758, Huang, K., II, 276, 765, 767
Eck, R. E., 773 759,761,772,774 Huberman, M., 536, 771
Ederer, D. L., 772 Giuliani, G. F., 341, 358, 768 Hubbard, J., 22, 300, 336, 338,
Edwards, S. F., 200,767 Glazman, L. I., 652, 773 765, 768
Eguliuz, A. G., 768 Glyde, H. R., 703, 715, 723, 774,
Ehrenreich, H., 325,767 775 Inkson, J. C., 766
Eisenberger, P., 366, 367, 368, 767 Goldhaber-Gordon, D., 575, 771 Isaacs, E. D., 768
Eisenstein, J. P., 774 Gores, J., 771
Eliashberg, G. M., 627, 664, 667, 773 Gorkov, L. P., 766, 772 Jaccard, D., 402, 769
Elliott, R. J., 594, 595, 600, 772 Gossard, A. c., 775 Jain, J. K., 761, 774
Emery, V. J., 256, 767 Green, M. S., 38, 160,765,766 Jancovici, B., 754, 774
Engelsberg, S., 246,480,515,516, Grewe, N., 391, 769 Jastrow, R., 690, 774
539,729,767,770,771,774 Greywall, D. S., 720, 735, 774 Jensen, H. H., 729, 774
Evrard, R., 770 Griffin, A., 694, 774, 775 Jonson, M., 374, 768
Grimvall, G., 491,492,545,770, Jordan, P., 17,40,765
Falicov, L. M., 771 771 Josephson, B. D., 644, 656, 773
Fano, U., 44, 208, 276, 765, 767 Gross, E. K. U., 766 Joshi, S. K., 486, 770
Feder, J., 768 Gully, W. J., 774
Feenberg, E., 707, 708, 709, 713, Gunnarsson, 0., 214, 391, 398, Kadanoff, L. P., 337, 519, 545, 550,
721, 774 614,767,769,772 554,767,771
Feldmann, W. L., 770 Guo, J. D., 773 Kanamori, J., 413, 427, 769
Fermi, E., 323, 768 Gupta, R. P., 623, 772 Kartheuser, E., 448, 770
Fetter, A. L., 766 Gurevich, V., 585, 587, 772 Kastner, M. A., 771
Ferrell, R. A., 300, 301, 302, 363, Gutzwiller M. c., 22, 765 Kay, P., 487, 770
768 Keffer, F., 286, 767
Feynman, R. P., 81, 220, 434, 438, Hagstrom, S. B. M., 768 Keiter, H., 391,769
446,697,702,703,709,742, Hallock, R. B., 722, 774 Keldysh, L., 107,634,766,773
767,770,774 Halperin, B. 1.,634,773 Khalatnikov, I. M., 729, 730, 773
Feuchtwang, T E., 561, 771 Halstad, R. E., 771 Kimball, J. c., 346,768
Firsov, Yu., A., 463, 770, 772 Hammerberg, J., 316, 768 Kittel, c., 500, 642, 771,773
Fishman, R., 536, 771 Han, S., 635, 773 Klitzing, Von K., 742, 774
Fisk, Z., 773 Hansch, w., 535, 550, 771 Knox, R. S., 596, 772
Fiske, M. D., 658, 773 Harling, 0., 698, 703, 774 Kohn, W., 150, 199,302,487,687,
Fleurov, V. N., 553, 554, 771 Harrison, W. A., 19,482,487,765, 758, 766, 767, 768, 770, 774
Fleszar, A., 342, 768 770 Kondo, J., 43, 46, 375, 765, 769
Flodstrom, S. A., 370, 768 Hart, H. R., 773 Kopaev, Y. v., 634, 773
Fowler, H. A., 768 Hasegawa, M., 582, 772 Kosterlitz, J.M., 761, 774
Francis, W. P., 690, 691, 694, 695, Hashitsume, N., 771 Kowalczyk, S. P., 772
696, 774 Hebb, M. H., 232, 767 Kozlov, A. N., 554, 771
Freeman, A. J., 623, 772 Hedin, L., 366, 606, 768, 772 Kramers, H. A., 360, 768
Friedel, J., 196,204,767 Heeger, A., 256, 767 Kravchenko, G. v., 768
Frohlich, H., 26, 34, 627, 765, 773 Heine, V., 19,765 Kravchenko, S. v., 356, 768
Fukada,N., 204, 767, 768 Heinonen, 0., 766 Kronig, R., 360, 768
Fumi, F. G., 198,204,767 Hemley, R. J., 315, 768 Kubo, R., 160, 179,535,536,766,
Furneaux, J. E., 768 Henry, C. H., 278, 279, 767 771
Furuya, K., 769 Herbst, J. F., 391, 392, 393, 769 Kudinov, E. K., 465, 770
Hewson, A. c., 391,769 Kunz, c., 333
Gallon, T E., 605, 772 Hill, T L., 43, 58, 765 Kurosawa, T, 455, 770
Garrett, D. G., 487, 770 Hirschfelder, J. 0., 678, 774
Geldart, D. J. w., 330, 768 Hodby, J. w., 447, 517, 770, 771 Lam, L., 367, 767, 768
Gell-Mann, M., 71, 297, 300, 307, Hohenberg, P. C., 697, 774 Land, R. H., 768
363, 766, 768 Holstein, T, 62, 401,455,457,463, Landau, L. D., 324,444,703,710,
Gerritsen, A. N., 46, 765 499,535,537,543,770,771 711,713,768,770,774
Author Index 779
Landauer, R., 575, 771 Massey, W. E., 722, 774 Palmer, L. H., 660, 663, 664, 670,
Lang, I. G., 463, 770, 772 Matsubara, T, 43, 109,765,766 773
Lang, N. D., 302, 768 Matsuda, H., 765 Pandey, K. c., 367, 768
Langenburg, D. N., 656, 773 Matthiessen, A., 535, 771 Pardee, W. J., 610, 624, 772
Langer, J. S., 197,207,347,505, Mattis, D. c., 187,267,272,405, Parker, W. H., 653, 773
515,516,767,768 660,662,767,769 Parks, R. D., 628, 658, 773
Langmuir, I., 332 Mattock, R. D., 766 Parr, R. G., 61, 765
Langreth, D., 156,519,766,771 Maxwell, E., 627, 773 Parratt, L. G., 625, 772
Larson, B. c., 342, 345, 768 Mazur, P., 766 Patton, B. R., 734, 774
Laughlin, R. B., 752, 760, 774 Mao, H. K., 315, 768 Pekar, S. L., 444, 770
Leder-Rozenblatt, D., 771 McFeely, F. R., 772 Pelizzari, C. A., 723, 775
Lee, D. M., 735, 774 McMenamin, J. c., 768 Pelzer, H., 765
Lee, P. A., 390, 769 McMillan, W. L., 214, 492, 497, Penrose, 0., 686, 688, 689, 690,774
Lee, T D., 770 670,627,689,690,694,696, Penn, D. R., 403, 610, 769, 772
Lee, T K., 768 767,770,772,774 Pepper, M., 774
Leggett, A. J., 714, 727, 741, 742, Megerle, K., 773 Percus, J. K., 58, 677, 691, 765,
774 Meirav, U., 771 744
Lehmann, H., 122,766 Mermin, N. D., 412, 769 Peschel, I., 267, 269, 767
Lemmens, L. E, 767 Meyer, L., 722, 773 Pethick, c., 714, 732, 733, 734,
LeNeindre, B., 765 Migdal, A. B., 481,489, 540, 665, 773,774
Levanda, M., 553, 771 770, 773 Pfeiffer, L. N., 774
Levin, K., 729, 774 Mihara, N., 361, 768 Phillips, J. c., 771
Levinson, C. A., 767 Millis, A., 773 Pines, D., 300, 313, 714, 719, 731,
Ley, L., 772 Mills, A. P., 168,766 732, 768, 770, 774
Lieb, E. H., 23, 187,267,282,765, Minnhagen, P., 247, 615, 622, 766, Pitarke, J. M., 768
767 772 Platzman, P. M., 697, 767, 774
Lifshitz, E., 324, 710, 768, 774 Mirlin, D. N., 770 Polder, D., 465
Lilly, M. P, 761, 774 Mittag, L., 768 Pollak, R. A., 772
Linde, J. 0., 765 Mitrovic, B., 671, 672, 674, 773 Powell, C. J., 333
Lindenberg, K., 770 Miyake, S. J., 446, 448, 770 Prange, R. E., 545, 750, 771, 774
Lindgren, L., 313, 768 Montroll, E. w., 765 Primakoff, H., 62
Lindhard, J., 325, 768 Mook, H. A., 714, 774, 775 Pudalov, Y. M., 768
Liu, S. H., 401,403,769 Mooradian, A., 474, 475, 770 Puff, R. D., 361, 696, 698, 768, 774
Lorenz, M. R., 771 Morse, R. w., 642, 773
Low, E, 71,434,766,770 Mozer, B. L., 60, 765 Quinn, J. J., 300, 301, 302, 363,
Lowenstein, J. H., 769 Myers, H. P., 589, 772 768
Lundqvist, B. L., 306, 313, 369, Quong, A. A., 768
370, 768 Nagaev, E. L., 460, 770
Luther, A., 256, 267, 269, 767 Nakajima, S., 491,770 Rajagopal, A. K., 486, 770
Luttinger,1. M., 150, 170, 178, Nesbitt, L. B., 773 Reatto, L., 694, 774
199,715,766,767,774 Newton, R. G., 204, 767 Reik, H. G., 465, 770
Lyddane, R. H., 472, 475, 770 Ng, K. w., 773 Reissland, J. A., 487, 770
Niklasson, G., 341, 768 Reynolds, C. A., 627, 773
MacDonald, A. H., 548, 549, 771 Nozieres, P., 300, 313, 354, 603, Reynolds, D. c., 596, 772
Mahalu, D., 771 611,613,621,714,715,719, Rice, M. J., 729, 774
Mahan, G. D., 32, 34, 152,254, 731,732,768,771,772,774 Rice, T M., 247, 366, 634, 767,
282,285,344,364,401,402, 768. 769, 773
436, 448, 450, 451,471,480, Ogiwara, A., 622, 772 Richardson, R. C., 774
520,523,534,535,550,554, Ohmura, Y., 622, 772 Rickayzen, G., 628, 773
582,592,603,606,607,610, Ohtaka, K., 610, 621, 772 Riffe, D. M., 769, 772
620,621,634,675,765,766, Onsager, L., 38, 251, 302, 686, Ritchie, R. H., 371, 768
768, 769, 770, 773 689,690.765,768,774 Robins, J. L., 333
Mahmoud, H. M., 767 Opsal, J. L., 545, 771 Rodriquez, L. J., 698, 775
Maradudin, A. A., II, 307, 313, Osheroff, D. D., 677, 735, 774 Ron, A., 714, 775
765, 767 Ott, H. R .. 773 Rosen, A., 313, 768
March, N. H., 313, 767 Overhauser, A. w., 262, 411, 767, Rosengren, A., 606, 772
Marton, L., 370, 371, 372, 768 769 Rowell, J. M., 492, 493, 670, 770,
Mason, W. E., 768 Owen, C. S., 670, 773 773
780 Author Index
Rubio, A., 768 Smodyrev, M. A., 440, 770 Vollhardt, D., 715, 775
Runge, E., 766 Sofo, J. 0., 402, 769 Vosko, S. H., 347, 365, 366, 767
Ruvalds, J., 703, 775 Solna, J., 775
Sondheimer, E. H., 519, 766, 771 Wada, Y., 770
Sachs, R. G., 475, 770 Srivastava, P. L., 487, 779 Wagner, H., 412, 769
Saint James, D., 771 Staver, 1.,486,770 Walecka, J. D., 766
Samwer, K., 382, 769 Stephen, M. J., 768 Wang, S., 544, 545, 771
Santoro, G. E., 341, 358, 768 Stewart, G. R., 390, 769 Wannier, G. H., 594, 772
Sawada, K., 300, 768 Stirling, W. G., 703, 775 Ward. J. C., 513, 771
Scalapino, D. J., 664, 670, 770, Stormer, H. L., 748, 775 Watabe, M., 491,582,770,772
771,773 Sturm, K., 590, 771 Weiss, G. H., 765
Schafroth, M. R., 627, 628, 773 Sueoka, 0., 372, 768 Werthamer, N. R., 635, 735, 736,
Schiff, L. I., 11,53,438,443,459, Sunjic, M., 603, 624, 626, 772 740, 772, 773
765 Svensson, E. c., 694, 698, 775 Wertheim, G. w., 370, 605, 769,
Schilling, A., 628, 773 Swan, J. B., 333 772
Schmidt, P., 767 Swanson, N., 768 West, K. w., 774
Schneider, T., 355,768 Swihart, J. c., 487, 491, 663, 670, Wheatley, J. c., 734, 775
SchOnhammer, K., 214, 391, 398, 770, 773 Wiegmann, P., 387,769
614,767,769,772 Sykes, J., 729, 773 Wigner, E. P., 17,40, 157,300,
Scholle, K. D., 461, 619, 770, 772 307,311,313,313,434,765,
Schotle, U., 619, 772 Takegahara, K., 544, 545, 771 766, 768, 770
Schrieffer, J. R., 47, 389, 480, 482, Tanabe, Y., 610, 621, 772 Wilkins, J. w., 37, 38, 391, 392,
515,516,539,566,627,628, Taylor, B. N., 773 393, 49C 769, 770, 771, 773,
728,766,769,770,771,773 Taylor, P. L., 179,766 774
Schweber, S. S., 187,767 Taylor, R., 330,487,548,549,768, Williams, F. E., 232, 767
Schwinger, J., 95, 766 770,771 Winzer, K., 382, 769
Sears, V. F., 694, 698, 699, 775 Teller, E., 475, 770 Wolf, E. L., 773
Segall, B., 518, 771 Tenn,J.S.,696,698,774 Wolfe, P., 715, 775
Seitz, F., 300, 301,768 Thaler, B. J., 771 Wolff, P. A., 47,389,766,769
Serene, J. w., 668, 670, 769, 773 Thomas, D. G., 32, 766, 767 Woll, E. J., 487, 770
Sernelius, B. E., 364, 768 Thomas, H., 768 Wollman, D. A., 659, 660, 773
Serin, B., 773 Thomas, L. H., 323, 768 Woo, C. w., 722, 774
Sham, L. J., 337, 487, 768, 769, Thouless, D. J., 761, 774 Woods, A. D. B., 698, 699, 702,
770 Tiablikov, S., 455, 770 773, 775
Shaw, w., 663, 670, 773 Ting, C. S., 344,768 Wright, G. B., 474, 475, 770
Shekhtman, L., 652, 773 Tinkham, M., 660, 663, 664, 670, Wright, W. H., 773
Sherrington, D., 687, 774 773 Wu, F. Y., 23, 765
Shirley, D. A., 772 Tischler, J. Z., 768 Wu, J. w., 592, 772
Sholl, C. A., 312, 768 Toda, M., 771
Shtrikman, H., 771 Tomlinson, P. G., 494, 770 Yamashita, J., 455, 770
Shukla, R. C., 487, 770 Tomonaga, S., 153,256,257,766, Yang, C. N., 686, 775
Shung, K. w., 364, 768 767 Yennie, D. R., 750, 775
Sierro, J., 402, 769 Tosi, M. P., 338, 768 Yosida, K., 383, 769
Sigel, J. L., 536, 771 Tracy, J. C., 369, 768
Silverstein, S. D., 387, 769 Tremblay, A. M. S., 428, 429, 769 Zaringhalam, A., 734, 774
Simanek, E., 582, 772 Tsui, D. c., 480, 742, 750, 770, 775 Zawadowski, A., 561, 703, 771,
Simpson, J. A., 768 775
Singh, R. N., 47, 770 Valls, O. T., 729, 774 Zeller, H. R., 567, 771
Singwi, K. S., 338, 347, 348, 349, Van Der Lage, F. C., 666, 773 Zener, C., 46, 766
355,356, Van Haeringen. w., 245, 246, 767 Zhao, Y., 770
Sixtensson, P., 589, 772 Van Harlingen, D. J., 773 Zienau, S., 765
Sjolander, A., 338, 768 Varga, B. B., 276, 767 Ziman,1. M., 487, 499, 500, 537,
Skold, K., 723, 775 Vashishta, P., 338, 355, 341, 768 770
Smith, H., 774 Vidberg, H. 1., 668, 670, 773 Zschack, P., 768
Smith, J. L., 773 Vignale, G., 341, 346, 358, 768 Zubarev, D. N., 694, 773
Smith, N. v., 589, 590, 769, 772 Vilk, Y. M., 413, 428, 429, 769 Zwicknagl, G., 401, 769
Subject Index
Advanced Green's function: see Green's functions Compton scattering, 366

Allen formula, 591-592 Condensate, 682, 697
Alpha-squared-F Conductivity
regular, 492, 527-533, 542, 668 impurity scattering, 499-514
transport, 527-533, 542, 548 polarons, 517-523
Anderson model, 44, 46-48, 389-403, 569 Configurational coordinates, 232, 235-236, 465-466
Anharmonic interactions, 10-11,708 Conserving approximation, 337
Antiferromagnetism, 411, 416-422 Cooper instability, 628-632
Anti-time ordering, 96, 154 Coordination number, 20, 43, 251, 403, 412, 455
Auger decay, 605, 624 Correlated basis functions, 690--696, 721
Correlation energy, 295, 300
Ballistic transport, 575 Correlation function
Bardeen-Cooper-Schrieffer: see BCS currenH:urrent, 165, 182,229,463,582,596
BCS theory, 386,420,480,627-644,678,713 density-density, 321, 536, 726
Bethe ansatz, 375 force-force, 535-537, 579-585
Bethe lattice, 247-256, 292-293 magnetic susceptibility, 175,341-342,356-358,727
Bijl-Feynman formula, 703-707 two-particle, 167
Bipartite, 418 Coulomb blockade, 569-575
Bogoliubov theory, 419, 427 Coupling constant integration, 148,350
Boltzmann equation, 499-505 Creation operator
Born-Oppenheimer approximation, 27 condensate, 682
Bose-Einstein condensation, 627, 677 electrons, 97
Brillouin-Wigner perturbation theory, 434-437, 489 particles, 13
phonons, 2,99,222
Cayley tree: see Bethe lattice photons, 57
CBF: see Correlated basis functions spins, 42
CDW: see Charge-density wave Tomonaga model, 258
Charge-density waves, 411 Cumulant expansion, 142-154,241-247,319,448-
Charge-spin separation, 406 454,614-619
Charging energy, 30 I, 568 Current
Collision term electrical, 23, 161
impurity scattering, 50 I energy, 25,178
electron-phonon scattering, 525 heat, 26, 179, 546
Commutation relations particle, 24, 178
particles, 13, 118 polarization, 23
phonons,3 Current-current correlation function: see Correlation
photons, 56 function
spins, 37-39
Tomonaga model, 258 Dangerous diagrams, 151
Complex time, 110-112 Debye-Waller factor, 461
Compressibility, 352 Deformation constant, 32
Compressibility sum rule, 353 Deformation potential, 31
781
782 Subject Index
Density matrix, 60, 687 f-sum rule, 290

Density of states longitudinal, 359
Bethe lattice, 251 transverse, 580
free electrons, 189 Fano--Anderson model, 44, 208-217, 247, 292
Densi ty operator, 17, 23 Fermi liquid theory, 713-720
Destruction operator: see Creation operator Fermi wave vector, 74,196,257,296,305
Diagonal transitions, 457 Ferromagnetism, 411, 413-416
Dielectric constant, 33 Feynman diagrams, 81, 94, 565
Dielectric function, 106, 174,277,287,317 Feynman theorem, 221
definition, 321 Filling factor, 745
electrons and phonons, 468-474 Forward scattering amplitude, 725
Hubbard, 336 FQHE: see Fractional quantum Hall effect
impurity scattering, 536 Fractional quantum Hall effect, 750-761
Lindhard: see Random phase approximation Frequency summations, 136-139
Lorenz-Lorentz, 287, 336 Friedel oscillations, 196-197, 346
random phase approximation, 306 Friedel sum rule, 196, 213, 317, 622
Singwi-Sjolander, 338 Frustration, 418
Thomas-Fermi, 323 Fumi's theorem, 198,204,292,384,614
Dipole approximation, 285
Disconnected diagrams, 83 Gauges, 49
Disentangling of operators, 221-224 Coulomb, 50
Distribution function Lorentz, 52
classical, 500, 49 magnetic fields, 743
Wigner, 157-160,549 transverse, 50
Dressed interaction, 725 Grand canonical ensemble, 112
Drude formula, 580, 589 Gaussian distribution, 227
Dyson's equation, 86-90, 99-102, 105, 128-135, 155- Gradient expansion, 554--556
157,200,215 Green's functions
advanced, 96, 118-128
Edge singularities: see X-ray edge singularity anti-time-ordered, 96
Effective mass, 128,368,435,491 cumulant expansion, 152-154, 241-247, 448-454,
Einstein model for phonons, 224 619-621
Electrical current operator, 23 electron, 71,12-115
Electron-electron interactions, 19 Matsubara, 112
Electron-phonon interactions, 26-36 phonons, 75, 115-117
deformation potential, 31, 227 photons, 102-106, 117, 276
mass enhancement, 491-495, 544--545, 627 retarded, 96, 118-128, 159
piezoelectric, 32, 227 time-ordered, 71-76, 96
polar, 34 unperturbed, 73
Electron self-energy Ground state energy
correlation, 300 homogeneous electron gas, 295-311,319-320,344,350
exchange, 92, 298 impurities, 204--207
from phonons, 87, 134 thermodynamic potential, 148-150
Electronic polaron, 401 GW approximation, 365
Elliott formula, 595-600
Emission spectra, 234 H'd' 46, 376, 383
Energy current, tight-binding, 24 Hall effect, 745-747
Entropy, 142, 177-179, 185 Hamiltonian
Equal spin pairing, 741 Anderson model, 46
Equations of motion, 327, 339, 636 bare phonon, 483
Exchange, 295-300 electron-phonon, 27-36
helium, 679-682 Fano--Anderson model, 44, 207
Hubbard model, 409-411 Frohlich,34-36,433-444,517-523
screened, 306, 363 harmonic chain, 6
Excitonic insulator, 634 harmonic oscillator, I
Excitons Heisenberg model, 38
metals, 607-61 I helium, 678
Wannier, 592-600, 632 homogeneous electron gas, 22, 295
Exponential resummation, 452 Hubbard model, 22
Subject Index 783
Hamiltonian (cont.) Ladder diagrams, 510, 540, 598

Ising model, 38 Lagrangian
Kondo model, 46, 375 particles, 12
lattice gas model, 42 photons, 53
phonons,1O Landau levels, 742
particles, 12-19 Laughlin wave function, 755
photons, 55 Lennard-lones potential, 678, 694
self-consistent field, 326 Lattice gas model, 42
spin, 36-48 Lehmann representation, 122, 139,397,507
tight-binding model, 20 Levinson's theorem, 196
tunneling, 561 Linear response, 165
XY model, 3 Linked cluster expansion: see Cumulant expansion
Harmonic oscillator, I-II Local field corrections, 336, 427
in electric field, 5-6, 30-31 Lowering operator
Hartree interaction, 297 phonons,2
Hartree-Fock approximation, 283, 298, 679-{)82 particle, 14-15
Heat current, 26 spins, 37
Heavy fermions, 390, 40 I, 635 Luttinger liquid, 276
Heisenberg Hamiltonian, 38 Luttinger model, 264-276
Hole
exchange and correlation, 310 Magnetic impurities, 375
core hole, 600, 605 Magnetic susceptibility, 174-177,264
in Fermi sea, 72, 334, 605 Magneto rotons, 757-759
in filled band, 168,412 Mahan singularity: see X-ray edge divergence
valence band, 594 Mass operator, 90
Holstein formula, 537, 543-544 Matthiessen's rule, 535
Homogeneous electron gas, 19 Mean field theory, 642
Hubbard model Mean free path, 369
definition, 22, 571 Meissner effect, 627
magnetic phases, 41 1-429 Metallic hydrogen, 315-316
paramagnons, 729 Mixed valence, 391
solution, 403-429 MND, for Mahan, Nozieres and deDominicis, 603,
621-{)23
Impurity scattering, 199-207,499-514,558-561,584 Mobility
Incomplete phonon relaxation, 606, 623 acoustical phonons, 524-534
Independent boson model, 31 , 218-247, 372,441,452, 619 CdTe,518
Infrared divergence, 270, 620 polarons,517-523
Interactions Moments
anharmonic, 10-11 first, 226
electron-electron, 19 second, 292
electron-phonon, 26-36 third, 361
particle-particle, 15-19
Interband absorption, 588-590 Nesting, 411, 418
Intermediate coupling, 438 Noncrossing approximation, 391, 397
Itinerant magnetism, 375 Nondiagonal transitions, 457
Ising model, 38, 251-256 Non-Fermi liquid behavior, 276
Isotope effect, 627, 632 No-phonon transition, 232, 461
Number operator, 442, 620
lahn-Teller effect, 26 Nyqvist theorem, 535
lellium: see Homogeneous electron gas
Occupation number
Kohn anomaly, 487 bosons, 14
Kondo electrons, 75
effect, 376, 621 phonons,8
model, 46-48, 375-389 Off-diagonal long range order, 686-690
resonance, 388, 397, 400 Onsager relations, 177-182
temperature, 377, 386-389 Optical properties, 579-580
Kramers-Kronig relation, 360, 363, 479, 497 Optical theorem, 194
Kubo formula, 163-165,234,505,537,562 Orthogonality catastrophe, 612-621
784 Subject Index
Pair distribution function, 59-62, 308-311, 341, 346, Random phase approximation, 306, 325, 406
688, 754 Rayleigh-Schrodinger perturbation theory, 300,434,
Paramagnetic susceptibility, 174, 264, 356 438
Paramagnon theory, 714, 728 Reaction matrix, 189-192
Particle current, 24 Relaxation energy, 30
Particle-hole symmetry, 412 Relaxation time
Partition function, 21 acoustical phonons, 524-534
Perturbation theory approximation, 500
Brillouin-Wigner, 434 degenerate semiconductors, 481
on the mass shell, 438 energy, 526-531
Rayleigh-Schrodinger, 438 impurity scattering, 502
Phase shift, 191-199 metals, 484
Phonons, I-II polarons, 444
anharmonic interactions, 10-11 temperature, 531-534
c1oud,442 transport, 500
drag, 382 Renormalization castrophe, 607
first quantization, I Renormalization factor, 127, 216, 365
force between electrons, I, 31 Representations
Kohn anomaly, 47 Heisenberg, 66
metals, 482--487 interaction, 67
number, 442 Schrddinger, 66
Photons,48-57,276-291 Resistivity, 505, 536-537, 558-561
Piezoelectric interaction, 32, 471 Retarded Green's function: see Green's functions
Plasma frequency Rotons,703
ions,280,468.483 bound state, 703
metals, 332-333 "He, 702-703
semiconductors, 472--475 RPA: see Random phase approximation
tight-binding, 406--408 r" 296, 467
Poincare's theorem, 211 Rules for constructing diagrams, 90-92, 135
Polar interaction, 34
Polariton, 276-291 Scattering resonance, 208
Polarizability, 284 Schrieffer-Wolff transformation, 47
Polarization operator, 90, 284 Screening charge density, 317, 346
Polaron Seebeck coefficient: see Thermopower
electronic, 40 I, 463 Seitz's theorem, 30 I, 353
Frohlich, 433--444 Self-consistent field, 325
linked-cluster, 448--454 Self-energy
optical absorption, 585-588 Dyson's equation, 86-90,132
p-wave states, 44 electron
small, 454--463 electron--electron interaction, 93-94, 295-311
strong-coupling, 444--448 electron-phonon interaction, 134,475--481,487-
Pseudopotential, 19, 31, 197 495
phonon,90,92-3,482--487
QBE: see Quantum Boltzmann equation s-d Interaction: see H,"
QD: see Quantum dot SDW: see Spin-density waves
QHE: see Quantum Hall effect Single mode approximation, 362, 758
Quantum Boltzmann equation, 95, 158,549-561 Slater determinant, 61, 612
Quantum conductance, 575 Slave bosons, 391
Quantum dots, 561-571 S matrix
Quantum Hall effect, 742, 747-750 impurity averaging, 199-200
Quantum lattice gas, 43 time-loop, 95-102
Quasiparticles, 365 time-ordered, 70-76
tau-ordered, 130
Raising operator s-p approximation, 732
phonons,2 Spectral function
particles, 14-15 dielectric function, 322
spin, 37 electron, 102, 122-128.212
Raman scattering, 278-279,474--475,495 phonon, 122, 125
Subject Index 785
Spin sound, 727 Tomonaga model, 256-276

Spin waves, 262 Trace, 109, 116, 120, 130, 170,220
Spin-density waves, 411, 418 Transport
Static form factor, 57 3He, 729, 734
Static structure factor, 57, 321, 338, 703, 722, 754 impurity scattering, 499, 514
Stoner criteria, 407, 414, 422 insulators, 517-523
Strong-coupling theory metals, 527-528
polarons, 4~8 semiconductors, 528-531
superconductors, 639, 663--670 Triplet pairing, 734
Structure factor: see Static structure factor Tunneling
Sudden approximation, 367, 700--702 conductance, 646
Sudden switching, 236-241 Josephson, 652--660
Sum rules, 123,212,225,358,428-429 normal metals, 561-567
Superconductors quantum dots, 571-575
anisotropic s-wave, 644 superconductors, 644-660
coherence factors, 639 Two fluid model, 703, 710-713
density of states, 643
dirty, 662 Ultrasonic attenuation, 642
d-wave pairing, 643--644 Unitary superfluid, 742
energy gap, 639, 642
excitation energy, 639 Vacuum polarization, 83, 132
high-temperature, 628 Van der Monde determinant, 754
infrared absorption, 660--664 Van Hove singularity, 415
p-wave, 635, 643,659 Vertex corrections, 343, 509, 538-540, 607--609
renormalization coefficient, 667, 671 V,d' 46, 376, 383
strong coupling, 639, 646, 663--670
transition temperature, 627, 670--674 Ward identity, 513-517, 538-539
triplet pairing, 635, 736-740 WDF: see Wigner distribution function
weak,627 Wick's theorem, 76,100,131,391
Wigner lattice, 311-316
T-matrix, 192,503 Wigner distribution function, 157-160, 338, 549-561,
Tamm-Dancoff approximation, 437 689
Tau ordering, 112 Wigner-Seitz model, 311-312
Thermal conductivity, 180
Thermopower, 181,401-402,546-549 XPS, for X-ray photoelectron spectroscopy, 603, 624-
Tight-binding model 626
definition, 20 X-ray edge divergence, 403, 603, 607
electrical current, 23 XY model, 38
energy current, 25
interactions, 22 Yukawa potential, 325
particle current, 23
small polarons, 454 Zero phonon line, 232, 461
Time ordering operator, 68, 96, 154 Zero sound, 723
Ti02, 463, 465, 466 Ziman formula, 537

Mahan

Uploaded by

Copyright:

Available Formats

Mahan

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mahan

Uploaded by

Copyright:

Available Formats

Many-Particle

Current Volumes in the Series

Springer Science+Business Media, LLC

QC176 .M24 2000

©2000, 1990, 1981 Springer Science+Business Media New York

All rights reserved

2. Green's Functions at Zero Temperature. 65

2.7. Dyson's Equation . . . . . . 86

3. Nonzero Temperatures. 109

4. Exactly Solvable Models 187

4.3.6. Linked Cluster Expansion . 241

5. Homogeneous Electron Gas 295

5.8.2. Effective Mass. 368

6. Strong Correlations 375

7. Electron-Phonon Interaction 433

9. Optical Properties of Solids . 579

10. Superconductivity 627

11. Superfluids . . . . . . 677

1.1. HARMONIC OSCILLATORS AND PHONONS

[x,Px] = ifz (1.1)

To solve this Hamiltonian, introduce a dimensionless coordinate s:

It is customary to define two dimensionless operators as follows:

H = flO) [aa t + ata] (1.11)

= fI; [~ (S+ ~) ( S- a~) + ~ ( S- :s) (s+ :1;) ] (1.12)

H = fI; (-:2 + 1;2) (1.13)

while the reverse order gives

These two results are subtracted,

[aa t - at a]f(~) = f(~) (1.16)

In a similar way, one can prove that

[at, at] = 0 (1.20)

These three commutators, plus the Hamiltonian

H = lim [aat + at a] = lim [aa t - at a + 2at a] = lim[a t a +!] (1.21)

In} = (atr 10) (1.22)

atln} = (a t r+1 10) = (n + 1)1/2 (at r+110) (1.24)

(n'latln) = (n + 1) 1/2 0n'=n+1 (1.26)

(n'laln) = (n)1/2 0n'=n_1 (1.27)

and the matrix elements (1.9) follow immediately:

The description of the harmonic oscillator in terms of operators is equivalent to the

so that the operator obeys the equation

i a~;t) = i[H, OCt)] (1.33)

For the destruction operator, this equation becomes

which has the simple solution

The time development of the position operator can be represented as

They obey the equation

A(t) = e-iOltA (1.44)

At(t) = eiOltAt (1.45)

[A,At] = [a+~,at +~J = 1 (1.46)

[At, At] =0 (1.48)

H=m[(At -~)(A-~) +~J +A(A+At -2~) (1.49)

= m(AtA + D-: (1.50)

The operator for the position is

and a new coordinate x' = x + eF; K is defined which obeys

The classical solution is obtained by solving the equation of motion:

-mxj = ma/xj = K(2xj - xJ+I - Xj-I) (1.58)

2Xj - xj+1 - xj_1 = xo[2 cos(kaj) - cos(kaj + ka) - cos(kaj - ka)]

= 2xo cos(kaj)[1 - cos(ka)]

The normal modes have the solution

[XI,Pm] = iDlm (1.61)

[Xk' P k' ] ~ '""' e- ikal eik'am[xI, P m ]

From the general result