5 The formalism of second quantization

Systems consisting of many identical particles form the prevailing part of

the physical world. A typical example are electrons in a lattice of ions in
a solid or nucleons in a heavy nucleus. While the classical treatement of
identical particles does not differ from the treatement of nonidentical ones -
classical theory relies on the assumption that the motion of each individual
particle can always be followed1 - indistinguishability imposes additional re-
quirements on the quantum theory of identical particles: vectors representing
states must have definite symmetry properties with respect to interchanging
labels of identical particles. Imposing this requirement in the ordinary ap-
proach based on the multiparticle-wave function is rather cumbersome. In
this section a convenient formulation - called “second quantization” - of quan-
tum mechanics of systems composed of many identical particles, allowing to
automatically take into account these symmetry requirements, is presented.
Its most characteristic feature is the use of the creation and annihilation op-
erators in terms of which any operator can be expressed and whose action on
the system’s states is particularly simple. The true essence of this formula-
tion is however the introduction of the “big” Hilbert space H whose vectors
can represent states of an arbitrary, also infinite, and even indefinite numbers
of particles. (It is this Hilbert space in which the action of the creation and
annihilation operators is naturally defined). Quantum mechanics formulated
using this formalism, when restricted to a subspace of H corresponding to
a fixed number N of particles (this is possible if the system’s Hamiltonian
commutes with the particle number operator), is fully equivalent to the quan-
tum mechanics based on the N-particle Schrödinger equation supplemented
with the appropriate symmetry requirements. However, the second quanti-
zation also opens up essentially new possibilities.2 First of all, by allowing
to form in the “big” Hilbert space H superpositions of vectors representing
different (also infinite) numbers of particles, second quantization enables one
to consider systems of interacting relativistic particles because, as will be
demonstrated in section 7, Poincaré covariance of transition amplitudes (of
the S-matrix) necessarily enforces nonconservation of the number of parti-
cles by the time evolution (relativistic Hamiltonians cannot commute with
particle number operators). Second quantization constitutes therefore a link
between the ordinary quantum mechanics of many-particle systems and the
1
Indistinguishability of identical particles must, nevertheless, be taken into account
in classical statistical mechanics to avoid the Gibbs paradox, that is nonextensiveness of
entropy.
2
In view of this the sometimes encountered statement that “second quantization” is
simply a misnomer, because it is just another formulation of the ordinary quantum me-
chanics based on the multi-particle Schrödinger equation (and not a new conceptual step,
similar to the transition from classical to quantum mechanics) is not entirely true.

164
relativistic quantum field theory. Moreover, nonseparability of the “big”
Hilbert space H (the lack of a countable basis) leads to the existence of in-
finitely many unitarily inequivalent representations by operators acting in it
of the basic commutation rules (of the abstract operator algebra) defining
the quantum theory. In other words, H furnishes a reducible representation
of the abstract operator algebra. Any such representation selects in H a
separable subspace called Fock space and it is the dynamics of the consid-
ered physical system which selects the Fock space in which the physically
accessible states of the system are represented. This profound property of
the big Hilbert space H is at the heart of the possibility of describing in this
formulation such phenomena as the Bose-Einstein condensation in systems
of bosons or spontaneous (“parametrical” or dynamical) breaking of various
symmetries in nonrelativistic as well as in relativistic systems.

5.1 Many-particle Hilbert spaces

Consider first a system of N distinguishable particles (for example all having

different masses). The Hilbert space of a system of N mutually interacting
(1) (1)
such particles is the N-fold tensor product H(N ) = H1 ⊗ . . . ⊗ HN of one-
particle Hilbert spaces of individual particles;3 it is spanned by the state-
vectors having the form of the tensor product

|Ψi = |ψ1 i ⊗ |ψ2 i ⊗ . . . ⊗ |ψN i , (5.1)

in which each of the individual state-vectors |ψk i, k = 1, . . . , N, belongs to

(1)
the separate one-particle Hilbert space Hk of the k-th particle. The scalar
product of two such states of N particles is simply given by the formula

hΦ|Ψi = hφ1 |ψ1 i · . . . · hφN |ψN i , (5.2)

in which each each factor hφk |ψk i is the scalar product in the respective
(1)
space Hk . If the normalized states |lk i, lk = 1, 2, . . . , ∞ form a countable
orthonormal (in the sense of the respective scalar products) basis of the k-th
particle Hilbert space, the product state-vectors

|l1 , l2 , . . . , lN i ≡ |l1 i ⊗ |l2 i ⊗ . . . ⊗ |lN i , (5.3)

form the associated countable basis of H(N ) . For example, if all N consid-
ered particles are spinless (or their spin degrees of freedom are neglected
3
All these one-particle Hilbert spaces fall, in the nonrelativistic case, into classes of
identical (isomorphic) spaces classified by the spin of the particle they correspond to.
One-particle Hilbert spaces of particles of spin s are all isomorphic to the (2s+1)-fold
Cartesian product of L2 (R3 ).

165
altogether) and can move in the infinite three-dimensional space, the three-
dimensional isotropic harmonic oscillator normalized to unity state-vectors
|ψ(lx ly lz )k (ωk )i or |ψ(lr lθ lϕ )k (ωk )i (lθ stands here, somewhat unconventionally,
for the orbital momentum quantum number) can be taken for |lk i (lk is
then a three-index lx ly lz or lr lθ lϕ ), because in the position (or the momen-
tum) representation any normalizable wave function ψ(x) (or ψ(p)) can be
written as a superposition of the functions ψ(lx ly lz ) (x) = ψlx (x)ψly (y)ψlz (z)
or |ψ(lr lθ lϕ )k (x) = ψlr (r)Ylθ lϕ (θ, ϕ) (or, in the momentum representation, of
ψ(lx ly lz ) (p) = ψlx (px )ψly (py )ψlz (pz )) with some arbitrary ωk2 > 0 (the fre-
quencies ω can be different for diffent particles and in ψ(lx ly lz ) could even
be different for diffent directions). Similarly, if the system of N particles is
enclosed in a box of volume V = L3 (and the periodic boundary conditions
are imposed on the wave functions of individual particles), the vectors

|p1 , p2 , . . . , pN i = |p1 i ⊗ |p2 i ⊗ . . . ⊗ |pN i , (5.4)

with pk = (2π/L)nk , can be taken for the basis of H(N ) . That the vectors
(5.3) or (5.4) form a (countable) basis of H(N ) follows from the simple ob-
servation that any normalizable wave function Ψ(x1 , . . . , xN ) of N particles
can be written as a superposition
X X
Ψ(x1 , . . . , xN ) = . . . cl1 ...lN ψl1 (x1 ) · . . . · ψlN (xN ) ,
l1 lN

of products ψl1 (x1 ) · . . . · ψlN (xN ) with

X X
. . . |cl1 ...lN |2 < ∞ .
l1 lN

The completeness relation then reads (1̂(N ) is the unit operator in H(N ) )
X X
1̂(N ) = ... |lN , . . . , l1 ihl1 , . . . , lN | . (5.5)
l1 lN

In the infinite space it is also possible, as it is customary in one-particle

quantum mechanics, to take for the basis of H(N ) the generalized (non-
normalizable) vectors4

|x1 , x2 , . . . , xN i ≡ |x1 i ⊗ |x2 i ⊗ . . . ⊗ |xN i , (5.6)

whose scalar products is

hyN , . . . , y1 |x1 , . . . , xN i = δ (3) (y1 − x1 ) · . . . · δ (3) (yN − xN ) , (5.7)

4
More precisely covectors, that is elements of the dual space.

166
or the generalized vectors

|p1 , p2 , . . . , pN i ≡ |p1 i ⊗ |p2 i ⊗ . . . ⊗ |pN i , (5.8)

with the scalar product

hqN , . . . , q1 |p1 , . . . , pN i = (2π)3 δ (3) (q1 − p1 ) · . . . · (2π)3 δ (3) (qN − pN ) . (5.9)

The completeness relation in these cases read

Z Z
1̂ = d x1 . . . d3 xN |x1 , . . . , xN ihxN , . . . , x1 |
(N ) 3

Z 3 Z 3
d p1 d pN
= 3
... |p1 , . . . , pN ihpN , . . . , p1 | . (5.10)
(2π) (2π)3

Internal degrees of freedom (spin) can be easily incorporated into this for-
malism by including the spin labels σk = −sk , . . . , +sk into lk ’s or using the
variables (xk , σk ) or (pk , σk ) instead of xk or pk .
The scalar product5 of a state of the form (5.1) with the basis vectors
(5.6) gives then the N-particle wave function

Ψ(x1 , . . . , xN ) = hxN , . . . , x1 |Ψi = ψ (1) (x1 ) · . . . · ψ (N ) (xN ) . (5.11)

The wave-function (more generally, the state-vector) of a linear superposition

of the state-vectors of the form (5.1) or (5.3) cannot in general be written
as a product of one-particle wave functions (one-particle states); states |Ψi
which do not factorize as (5.1) are called entangled states and play crucial
role in modern quantum optics and quantum information theory.

Consider now a system consisting of N indistinguishable particles of one

type,6 bosons or fermions (continuing to omit in the notation their spin
labels). In this case the requirement of the Bose-Einstein or Fermi-Dirac
statistics has to be implemented: state-vectors should be symmetric with
respect to interchanges of variables of any two identical bosons and antisym-
metric with respect to interchanges of variables of any two fermions.7 This
requirement has to be added as an extra rule selecting possible states. In
the framework of the nonrelativistic quantum mechanics it cannot be given
any sound foundation; the Bose-Einstein (Fermi-Dirac) statistics obeyed by
systems of many identical particles having integer (half-integer) spin must
5
Or, more precisely, the value of the covector (5.6) on the vector (5.1).
6
Generalization of the formalism to several types of identical particles is straightfor-
ward.
7
In two spatial dimensions there are more possibilities.

167
simply be regarded as a phenomenological input.8 As will be argued in Sec-
tion 8, justification of this celebrated spin-statistics connection comes only
from the relativistic quantum field theory. It is implemented by defining
the state-vectors spanning the Hilbert space H(N ) of a system of N identical
particles in terms of the one-particle state-vectors by the formula
1 X P
|ψ1 , ψ2 , . . . ψN i = √ ζ |ψP (1) i ⊗ |ψP (2) i ⊗ . . . ⊗ |ψP (N ) i , (5.12)
N! P

in which |ψi i are N arbitrary one-particle states and P denotes their permu-
tations. We have introduced here the symbol ζ

+ 1 for bosons
ζ= ,
−1 for fermions

and by ζ P understand for fermions the sign of the permutation P . For bosons
of course ζ P = 1. For example, in the case of N = 2 particles and two one-
particle states |ai or |bi one can construct the following three states
1
|a, bi = √ (|ai ⊗ |bi + |bi ⊗ |ai) ,
2!
√
|a, ai = 2 |ai ⊗ |ai , (5.13)
√
|b, bi = 2 |bi ⊗ |bi ,

if the particles are identical bosons and only one state

1
|a, bi = √ (|ai ⊗ |bi − |bi ⊗ |ai) ,
2!
if they are identical fermions. The scalar product of such states is

hϕN , . . . , ϕ1 |ψ1 , . . . , ψN i
1 XX P Q
= ζ ζ hϕQ(1) |ψP (1) i · . . . · hϕQ(N ) |ψP (N ) i
N! P Q
1 XX R
= ζ hϕ1 |ψR(1) i · . . . · hϕN |ψR(N ) i
N! R Q
X
= ζ R hϕ1 |ψR(1) i · . . . · hϕN |ψR(N ) i , (5.14)
R

8
The Planck law of the black body radiation can be taken for the empirical proof that
photons obey the Bose-Einstein statistics (are bosons); similarly, stability of matter is the
best physical indication that electrons (and nucleons) are subject to the Pauli exclusion
principle (are fermions).

168
(in the second step a new permutation R = P Q−1 has been defined, whose
sign is ζ R = ζ P ζ Q ). If the particles are fermions, (5.14) is just the determi-
nant
 
 hϕ1 |ψ1 i . . . hϕ 1 |ψ N i 
 

 · · 

hϕN , . . . , ϕ1 |ψ1 , . . . , ψN i = 
 · · ,


 · · 

 hϕN |ψ1 i ... hϕN |ψN i 

while if they are bosons, the scalar product differs from the determinant by
having all signs in the Laplace expansion positive. Notice that in the example
(5.13) of N = 2 particles the states |a, ai and |b, bi of bosons are not prop-
erly normalized even if the one-particle states |ai and |bi are orthonormal:
ha, a|a, ai = hb, b|b, bi = 2.

Let us now construct the basis of the (anti)symmetrized N-particle Hilbert

space H(N ) . Let |li = |1i, |2i, . . . be a countable complete set of normalizable
and orthonormal one-particle state-vectors (forming a basis of the Hilbert
space H(1) of a single particle) labeled by l = 1, 2, . . . , ∞, that is such that
X
hl′ |li = δl′ l , |lihl| = 1̂(1) .
l

Again, if the system of spinless particles is enclosed in a box of finite volume

L3 , the momentum operator eigenvectors |ki with k = (2π/L)n can be taken
for |li’s; in the infinite volume one can take for |li the three-dimensional har-
monic oscillator state-vectors |lx ly lz i (or the vectors |lr lθ lϕ i in the angular
momentum representation). If particles have a nonzero spin s, the appro-
priate spin label σ = −s, . . . , +s must be included in the label l. As an
orthonormal a basis of the Hilbert space H(N ) of N identical bosons one can
then take the vectors9
1
√ |l1 , l1 , . . . , lN i with l1 ≤ l2 ≤ . . . ≤ lN , (5.15)
n1 ! n2 ! . . .

where |l1 , l2 , . . . , lN i are the states of the form (5.12) and n1 is the number of
l1 occurrences in the sequence l1 , l2 , . . . , lN , n2 is the number of l2 occurrences,
etc. Of course, n1 + n2 + . . . = N. As an ortonormal basis of the Hilbert
space H(N ) of N identical fermions one takes instead the vectors (the label l
in this case must necessarily include also the spin variable)

|l1 , l2 , . . . , lN i with l1 < l2 < . . . < l N . (5.16)

9
It is assumed that the basis state-vectors |li forming a countable set, can be ordered
in some natural way.

169
 If in the example (5.13) of N = 2 bosons the whole H(1) is spanned by
only two vectors |ai and |bi, the basis of H(2) can be formed by the vectors:
1 1
√ |a, ai , √ |b, bi , |a, bi .
2! 2!
√
The factors of 1/ 2! included in the first two basis vectors ensure their proper
normalization. More generally, recalling that
X
′
hlN , . . . l1′ |l1 , . . . , lN i = ζ P hl1′ |lP (1) i · . . . · hlN
′
|lP (N ) i , (5.17)
P

we see that the scalar product is nonzero only if each li′ finds its counterpart
among the li ’s. If there are ni occurrences of a particular li′ in the sequences
l1′ , . . . lN
′
and l1 , . . . , lN (which is possible only if the considered particles are
bosons), then there are ni !√equal and nonzero terms contributing to the sum
in (5.17). The factors 1/ ni ! in the definition of the basis state-vectorss
(5.15) of N identical bosons cancel then the factors ni ! arising in the scalar
product. The completeness relation in the space of N identical bosons or
fermions can be therefore conveniently written in the form
∞ ∞
1 X X
... |l1 , . . . , lN ihlN , . . . l1 | = 1̂(N ) . (5.18)
N! l l
1 N

Note that here the orderings of li ’s appearing in the definitions of the basis
vectors (5.15) and (5.16) are not respected. Instead, the factor 1/N! ensures
the cancellation of the multiple counting of the same states. To understand
better its working, let us consider three bosons, each of which can be in one
of the
√ two states |ai or |bi and consider the contribution of the basis vector
(1/ 2)|a, a, bi to the completeness relation. In the decomposition of the unit
operator only a single term of the form
1
1̂(N ) = . . . + |a, a, bihb, a, a| + . . .
2!
should be present. (5.18) gives
 
(N ) 1
1̂ = . . . + |a, a, bihb, a, a| + |a, b, aiha, b, a| + |b, a, aiha, a, b| + . . .
3!
which is the same taking into account the symmetry of the states. With this
convention it is possible to work (in the infinite volume) also with the bases
formed out of generalized (i.e. non-normalizable) symmetrized or antisym-
metrized state-vectors like e.g. the (we make now the spin labels explicit)
|p1 σ1 , . . . , pN σN i ones; the unit operator 1̂(N ) is then decomposed as
Z 3
(N ) 1 d p1 d3 p1 X X
1̂ = . . . ... |p1 , . . . , pN ihpN , . . . , p1 | . (5.19)
N! (2π)3 (2π)3 σ σ
1 N

170
 It is important to stress that any two bases like (5.15) (like (5.16)) of
the Hilbert space H(N ) of N bosons (N fermions) formed out of two dif-
ferent bases |li and |˜li of H(1) are unitarily equivalent. This means that
any vector of the basis (5.15) (of the basis (5.16)) √ can be written as a
linear combination of the vectors |˜l1 , ˜l2 , . . . , ˜lN i/ ñ1 !ñ2 ! . . . (of the vectors
|˜l1 , ˜l2 , . . . , ˜lN i). In other words, any vector of the basis (5.15) (of the basis
(N )
(5.16)) has nonzero scalar √ products (in H ) with at least a finite number
of vectors |˜l1 , ˜l2 , . . . , ˜lN i/ ñ1 !ñ2 ! . . . (of vectors |˜l1 , ˜l2 , . . . , ˜lN i).

All this works in the same way for arbitrary N ≥ 1. It proves convenient
to formally include the N = 0 case, by adopting the convention that the H(0)
Hilbert space is spanned by a single vector |voidi in most texts misleadingly
called the “vacuum”, or, less misleadinly, the Fock vacuum vector.10 That
is, the zero-particles Hilbert space H(0) is one-dimensional (all other Hilbert
spaces for N ≥ 1 are all countably infinite dimensional). One then formally
introduces the “big” Hilbert space H
(N )
H = ⊕∞
N =0 H , (5.20)

whose vectors are of the form

H ∈ |Ψi = |Ψ(0) i + |Ψ(1) i + |Ψ(2) i + . . . (5.21)

with |Ψ(N ) i belonging to H(N ) and |Ψ(0) i = a|voidi. This construction is

not so unnatural as it might seem at first sight. In a typical accelerator
experiment a well defined two-particle state is prepared,11 say |e+ e− i in an
e+ e− collision, and, after the interaction (collision), the state of the system is
represented by a vector which in fact is a superposition of vectors belonging
to different multi-particle spaces H(N ) , corresponding to all possible many-
particle states that can be created in this collision (of course selection rules
following from e.g the electric charge conservation, impose some constraints).
Therefore the final state is a (in general infinite) superposition of vectors
corresponding to different numbers of particles

|Ψi = a |e+ e− i + b |e+ e− γi + c |q q̄i + d |q q̄gi + e |q q̄q q̄i + . . .

where a, b, c, d, e are some complex number. Measurement of a concrete final

state by the detector has the effect of reducing this state-vector of the system.
10
We denote this vector |voidi in order to distinguish this “technical” “no particle”
vector from the state-vectors |Ω0 i and |Ωi which will be the true ground state-vectors,
that is the lowest energy eigenvectors, of the free H0 and interacting H = H0 + Vint
Hamiltonians of systems of free or interacting particles (or fields), respectively.
11
The precise way the states like |e+ e− i, representing a well defined number of physically
detectable particles, are related to the basis states formed as tensor products of one-particle
states will be elucidated in Section 7.2.

171
Thus, at least in physics of relativistic particles in which (as will be seen)
particle number conservation is impossible, the big Hilbert space (5.20) is the
right arena in which to describe physical processes. The possibility of treating
systems with variable number of particles is useful also in nonrelativistic
physics. Investigating statistical properties of physical systems one usually
prefers to work in the framework of the Grand Canonical Ensemble, in which
the number of paricles is allowed to fluctuate. Also one is usually interested
in systems having fixed density which requires considering the system of N
particles in a finite box of volume V and taking the limits N → ∞, V → ∞
with ρ = N/V kept fixed. Finally, the possibility of forming superpositions
of state-vectors corresponding to different number of particles is crucial for
description of phenomena as superconductivity in which some symmetries
are spontaneously broken.
The seemingly innocuous formal step of forming the Hilbert space H
as the direct sum of infinitely many Hilbert spaces H(N ) has a profound
mathematical consequence: the constructed space is not separable, that is
it has no countable basis - as will be shown in the next section, the power
of the set of basis vectors (5.15) or (5.16) with arbitrarily large N is equal
to the power of the continuum. Infinitely many different separable Hilbert
subspaces (called in this context Fock spaces), with countable bases can be
chosen in H (5.20), all of which are, if the volume of the space is taken to be
infnite and/or no UV cutoff (effectively discretizing the space) is imposed,
are ortghonal to one another.
In the “big” Hilbert space (5.20) the scalar product is defined as
∞
X
hΦ|Ψi = hΦ(N ) |Ψ(N ) i , (5.22)
N =0

that is, vectors belonging entirely to H(N ) and H(M ) are declared to be or-
thogonal if N 6= M. The completeness relation in H reads
∞
X 1 X
1̂ = |voidihvoid| + |l1 , . . . , lN ihlN , . . . , l1 | . (5.23)
N =1
N!
l1 ,...,lN

For example, in the position basis it takes the form (spin labels can be in-
cluded here as in (5.19))
∞
1
X Z Z
1̂ = |voidihvoid| + dx1 . . . dxN |x1 , . . . , xN ihxN , . . . , x1 | , (5.24)
N =1
N!

and the most general form of a state belonging to H is

∞
1
X Z Z
|Ψi = a|voidi + dx1 . . . dxN |x1 , . . . , xN ihxN , . . . , x1 |Ψi
N =1
N!

172
 ∞
1
X Z Z
= a|voidi + dx1 . . . dxN |x1 , . . . , xN iΨ(N ) (x1 , . . . , xN ) ,
N =1
N!

where Ψ(N ) (x1 , . . . , xN ) ≡ hx1 , . . . , xN |Ψ(N ) i is the wave function of the N-

particle component of the state |Ψi. The function Ψ(N ) (x1 , . . . , xN ) of a
system of bosons is totally symmetric while a similar function of a system of
fermions is totally antisymmetric in all groups of its arguments corresponding
to individual particles. If |Ψ(N ) i is of the form (5.12) then (we again make
the spin labels α - which must, and in nonrelativistic mechanics can, be given
a meaning in terms of some one-particle spin operator - explicit here)
 
 hx1 α1 |ψ1 i ... hx1 α1 |ψN i 

 · ... · 
(N )
 
Ψα1 ,...,αN (x1 , . . . , xN ) =  · ... ·  , (5.25)


 · ... · 

 hxN αN |ψ1 i ... hxN αN |ψN i ζ

where the symbol ζ means the determinant in the case of fermions (called
the Slater determinant) and for bosons the so-called permanent of the matrix
which is calculated as the determinant except for taking everywhere positive
signs (the precise definition is given by (5.14) with ϕi replaced by xi ).

5.2 Creation and annihilation operators

Construction of the big Hilbert space H enables the introduction of the cre-
ation and annihilation operators. Let |ϕi be a one-particle state. With any
such state it is possible to associate the corresponding creation operator a† (ϕ)
acting in H and mapping H(N ) into H(N +1) . On vectors of the form (5.12) it
is defined by the formula

a† (ϕ)|ψ1 , . . . , ψN i = |ϕ, ψ1 , . . . , ψN i , (5.26)

and extended to all H by linearity. The annihilation operator a(ϕ) is defined

as the Hermitian conjugate of a† (ϕ) through the equality (cf. (4.1))

∗
hχN −1 , . . . , χ1 |a(ϕ)|ψ1 , . . . , ψN i = hψN , . . . , ψ1 |a† (ϕ)|χ1 , . . . , χN −1 i
 ∗
 hψ1 |ϕi hψ1 |χ1 i . . . hψ 1 |χ N −1 i 
 

 · · ... · 

= · · ... ·  .


 · · ... · 

 hψN |ϕi hψN |χ1 i ... hψN |χN −1 i ζ

173
The Laplace expansion (valid for determinants and permanents alike) gives
N  ∗
∗  withoutthe k−th row 
X
k−1

hχN −1 , . . . , χ1 |a(ϕ)|ψ1 , . . . , ψN i = ζ hψk |ϕi  .
and the first column ζ
k=1

From this formula it is easy to read-off the following rule:

N
X
a(ϕ)|ψ1 , . . . , ψN i = ζ k−1hϕ|ψk i|ψ1 , . . . (no ψk ) . . . , ψN i , (5.27)
k=1

which shows that a(ϕ) acting on a state |Ψ(N ) i of the form (5.12) gives zero
if |ϕi is orthogonal to all one-particle states |ψi i building |Ψ(N ) i. The factor
ζ k−1 appearing in (5.27) can be understood as a sign factor arising when ψk
is moved from its k-th position in the ket to the first position, where a(ϕ)
annihilates it.
From the definitions (5.26) and (5.27) it is easy to show that

a† (ϕ1 ) a† (ϕ2 ) = ζ a† (ϕ2 ) a† (ϕ1 ) ,

a(ϕ1 ) a(ϕ2 ) = ζ a(ϕ2 ) a(ϕ1 ) , (5.28)

that is, the creation operators associated with one-particle states of bosons
commute (ζ = +1) and those associated with fermion states anticommute
(ζ = −1) and the same holds true for the annihilation operators. It follows,
that the fermionic operators are nilpotent:

a2 (ϕ) = (a† (ϕ))2 = 0 . (5.29)

It is also easy to prove that

a(ϕ1 ), a† (ϕ2 ) −ζ = hϕ1 |ϕ2 i ,

 
(5.30)

where the subscript −ζ denotes the commutator for bosons and the anticom-
mutator for fermions.
The creation and annihilation operators can be associated with any com-
plete set of one-particle states |li. If the vectors |li form a countable basis of
H(1) , then

[al′ , a†l ]−ζ = δl′ l . (5.31)

(If, instead, the |li’s are generalized vectors, then [al′ , a†l ]−ζ = δ(l′ − l)).
This means that the annihilation and creation operators of bosons (fermions)
commute (anticommute) to 1 (or, in the case of a continuous normalization
of the one-particle states, to the appropriate delta function). For future

174
use note also, that for fermions the rule (5.31) is symmetric with respect
to the interchange al ↔ a†l , i.e. the algebraic properties of the fermionic
annihilation and creation operators are the same. This property will be
exploited in section 5.4. The action of the fermionic operators al′ anda†l on
the basis vectors (5.16) is
N
X
al |l1 , l2 , . . . , lN i = (−1)k−1 δllk |l1 , . . . (no lk ) . . . , lN i ,
k=1

a†l |l1 , l2 , . . . , lN i = |l, l1 , l2 , . . . , lN i = ±|l1 , . . . , l, . . . , lN i , (5.32)

where the sign depends on the number of interchanges needed to put l on

the appropriate position (in agreement with the ordering specified in (5.16)).
To represent the action of the bosonic operators al′ anda†l on the basis vec-
tors (5.15) it is convenient to introduce first the so-called occupation number
representation by changing the notation used for the vectors (5.15) or (5.16)
forming countable bases of the Hilbert spaces H(N ) of N bosons or fermions,
respectively, by defining
1
|n1 , n2 , . . .i ≡ √ |1, . . . , 1, 2, . . . , 2, . . .i , (5.33)
n1 ! n2 ! . . .
In this notation the numbers n1 , n2 etc. simply indicate how many particles
occupies a given (discrete) one-particle state |li (if the particles are fermions,
only nl = 0 or 1 are possible). Of course, vectors belonging to H(N ) are
restricted by the condition n1 + n2 + . . . = N. Removing this restriction
one obtains a set of vectors belonging to the “big” Hilbert space H defined
by (5.20). The set of such vectors is uncountable - its power is equal to the
power of the continuum.12 As a result, the Hilbert space H constructed as
in (5.20) is not separable and the state-vectors |n1 , n2 , . . .i span in fact only
a subspace of the big Hilbert space: there are vectors in H which have zero
scalar products with all vectors the |n1 , n2 , . . .i.
In the occupation number representation the action of bosonic creation
and annihilation operators looks more familiar:
√
al |n1 , n2 , . . . , nl , . . .i = nl |n1 , n2 , . . . , nl − 1, . . .i ,
√
a†l |n1 , n2 , . . . , nl , . . .i = nl + 1 |n1 , n2 , . . . , nl + 1, . . .i . (5.34)
12
This is particularly easy to demonstrate in the case of fermions: it suffices to notice
that the infinite sequence of numbers n1 n2 n3 . . ., in which each nl = 0, or 1, treated as the
binary coding of an integer, can be uniquely mapped onto the infinite sequence of integer
numbers p1 p2 p3 . . ., 0 ≤ pl < 9 and the fractions 0.p1 p2 p3 . . . fill the entire segment [0, 1)
whose power is equal to the power of the continuum.

175
For completeness we give also the action of the fermionic operators on the
basis vectors (5.16) written in this notation:

0 if nl = 0
al |n1 , n2 , . . . , nl , . . .i = ,
η|n1 , n2 , . . . , nl − 1, . . .i if nl = 1

† 0 if nl = 1
al |n1 , n2 , . . . , nl , . . .i = ,
η|n1 , n2 , . . . , nl + 1, . . .i if nl = 0
†
where η = (−1)p , p =
P
l′ <l nl′ . In both cases, the operator al al counts
therefore the number of particles occupying the (one-particle) state |li.
The creation and annihiolation operators can be also associated with
non-normalizable bases of H(1) , for example with the basis of plane waves
|pi (in the case of spin zero bosons) or |pσi in general normalized to the
delta function in the infinite space13

hp′ , σ ′ |p, σi = (2π)3 δ (3) (p′ − p)δσ′ σ ,

for which hx, α|p, σi = (ψpσ )α (x) = δασ eip·x , or normalized in a box of
volume V = L3 with periodic boundary conditions:

hp′ σ ′ |pσi = δp′ ,p δσ′ σ , (5.35)

and (ψpσ )α (x) = L−3/2 δασ eip·x . The rule (which is frequently used) for tran-
sition from the discrete normalization to the continuous one is
V
X X Z
≡ ↔ 3
d3 p . (5.36)
p n ,n ,n
(2π)
x y z

In this basis

[aσ (p), a†σ′ (p′ )]−ζ = (2π)3 δ (3) (p′ − p)δσ′ σ or δp′ ,p δσ′ σ , (5.37)

and the Hilbert space H(N ) is spanned by the vectors

|p1 σ1 , p2 σ2 , . . . , pN σN i = a†σ1 (p1 ) a†σ2 (p2 ) . . . a†σN (pN )|voidi . (5.38)

One can also use the position basis in which

[aα (x), a†α′ (x′ )]−ζ = δ(x′ − x)δα′ α , (5.39)

and the Hilbert space H(N ) is spanned by the vectors

|x1 α1 , x2 α2 , . . . , xN αN i = a†α1 (x1 ) a†α2 (x2 ) . . . a†αN (xN )|voidi . (5.40)

13
In the relativistic case it will be convenient to slightly modify this normalization.

176
The operators a(x) and a† (x) are sometimes denoted by ψ̂(x) and ψ̂ † (x) and
called the field operators.
In general, if a one-particle state-vector |χi is a linear superposition of
some other vectors, e.g. if

|χi = A |ψi + B |φi , (5.41)

then from the definition (5.26) it is clear that

a† (χ) = A a† (ψ) + B a† (φ) ,

a(χ) = A∗ a(ψ) + B ∗ a(φ) . (5.42)

Since
d3 p X
Z 3 X
dp
Z
|x, αi = |p, σihp, σ|x, αi = |p, σi δσα e−ip·x , (5.43)
(2π)3 σ (2π)3 σ

we have
d3 p † d3 p
Z Z
a†α (x) = a (p) e−ip·x , aα (x) = aα (p) e+ip·x . (5.44)
(2π)3 α (2π)3

More generally, if the functions (ulσ )α (x) ≡ hx, α|l, σi form a complete or-
thonormal set (we single out the spin label σ from the general one-particle
state label l), such that
X X
|x, αi = |l, σihl, σ|x, αi = |l, σi (ul,σ )∗α (x) , (5.45)
l,σ l,σ

then
X X
a†α (x) = a†l,σ (ulσ )∗α (x) , aα (x) = al,σ (ulσ )α (x) , (5.46)
l,σ l,σ
Z X Z X
a†l,σ = dx 3
(ulσ )α (x) a†α (x) , al,σ = d3 x (u∗lσ )α (x) aα (x) . (5.47)
α α

5.3 Hamiltonian and other operators

The annihilation and creation operators allow to represent various operators

in a simple form. Let us begin with one-particle operators of the general
form
N
(1)
X
O= Oi , (5.48)
i=1

177
 (1)
where Oi denotes a one-particle operator O (1) acting only on the variables
of the i-th particle:
O|ψ1 , ψ2 , . . . , ψN i = |ψ1′ , ψ2 , . . . , ψN i + . . . + |ψ1 , ψ2 , . . . , ψN
′
i. (5.49)
The state-vectors |ψ1 , ψ2 , . . . , ψN i are defined in (5.12) and |ψi′ i = O (1) |ψi i.
(1)
Obviously, if |ψi i are all eigenstates of the respective Oi ’s with the eigen-
values oψi , one gets
O|ψ1, ψ2 , . . . , ψN i = (oψ1 + oψ2 + . . . + oψN )|ψ1 , ψ2 , . . . , ψN i . (5.50)
An example of the operator of this type is e.g. the kinetic energy operator
PN (1)2
T = i=1 P̂i /2m which in many cases plays the role of the free Hamilto-
nian H0 of the system of particles. Acting on the (generalized) state-vectors
(5.38) gives the eigenvalue (p21 + . . . + p2N )/2m.
To find the representation of a general O of the form (5.48) in terms
of the annihilation and creation operators let us consider first a particular
operator O (1) = |lihl′ | (the spin label is now included in the general label
l). The action of the corresponding O of the form (5.48) on a state-vector
|ψ1 , ψ2 , . . . , ψN i is
O|ψ1 , ψ2 , . . . , ψN i = hl′ |ψ1 i |l, ψ2 , . . . , ψN i + hl′ |ψ2 i |ψ1 , l, . . . , ψN i + . . .
It is easy to see that this action of O is identical with the action of a†l al′ .
Since any one-particle operator can be written in the form
XX X X (1)
O (1) = |lihl|O (1) |l′ ihl′ | ≡ Oll′ |lihl′ | , (5.51)
l l′ l l′

one concludes that

(1)
XX
O= Oll′ a†l al′ . (5.52)
l l′

Taking for example O (1) = 1̂(1) (the one-particle unit operator), whose
(1)
matrix elements are Oll′ = hl|1̂(1) |l′ i = δll′ , one gets the operator N̂ counting
the number of particles of the system:
Z 3 X
dp
X † Z X
3 †
N̂ = al al = d x aα (x) aα (x) = 3
a†σ (p) aσ (p) . (5.53)
l α
(2π) σ

Similarly, taking for O (1) the one-particle momentum operator P̂(1) we obtain
the operator P̂ of the total momentum of the system:14
Z 3 X
dp
Z X
† 3
P̂ = ~p a σ (p)aσ (p) = d x a†α (x)(−i~∇x )aα (x) , (5.54)
(2π)3 σ α
14
The second form of this operator follows from the matrix element
∂ (3)
hx|P̂(1) |yi = i~ δ (y − x) ,
∂y

178
(recall that p is the wave vector, not the momentum). In the similar way,
out of the corresponding one-particle operators, it is possible to construct
also the total angular momentum J and boost K operators of the system of
free particles.
If (identical) particles are not interacting with each other but all move in
some external potential V (x), their nonrelativistic Hamiltonian is the sum
(over particles) of one-particle operators H (1) = T (1) + V (1) = P̂(1)2 /2m +
V (1) (x̂) whose matrix elements are

(1) ~2 2 (3)
hx|H |yi = − ∇y δ (y − x) + V (x) δ (3) (y − x) , (5.55)
2m
and takes the form
~2 2
Z  
3 †
H= d x a (x) − ∇ + V (x) a(x) . (5.56)
2m x

To rewrite it in the momentum representation we use the formulae

~2 p2
hp|T (1) |p′ i = (2π)3 δ (3) (p − p′ ) ,
2m
and
Z Z
(1) 3
hp|V ′
|p i = dx d3 y hp|xihx|V (1) |yihy|p′i = Ṽ (p − p′ ) ,

where Ṽ (q) is the Fourier transform of the potential V (x):

Z 3
dq
Z
3
Ṽ (q) = d x V (x) e −iq·x
, V (x) = Ṽ (q) eiq·x .
(2π)3

Thus, in the momentum representation we get

Z 3
d p ~2 p2 †
Z 3 Z 3 ′
dp dp † ′
H= 3
a (p) a(p) + 3
a (p ) Ṽ (p′ − p) a(p)
(2π) 2m (2π) (2π)3
Z 3
d p ~2 p2 †
Z 3 Z 3
dp dq †
= 3
a (p) a(p) + 3
a (p + q) Ṽ (q) a(p) . (5.57)
(2π) 2m (2π) (2π)3

Of course, if instead of the one-particle bases formed by the generalized

vectors |xi or |pi one uses as the complete set the eigenvectors |li of the full
of the one-particle momentum operator P̂(1) which correctly gives
∂
Z
hx|P(1) |ψi = d3 y hx|P̂(1) |yi hy|ψi = −i~ ψ(x) .
∂x

179
one-particle Hamiltonian H (1) = P̂2 /2m + V (1) (x̂), the “second-quantized”
Hamiltonian H will take the simple form
X
H= El a†l al , (5.58)
l

(the range of the label l may consist of a discrete and a continuous parts).
Note also that the operator

ρ(x) = a† (x)a(x) , (5.59)

has the natural interpretation of the operator of the particle number density
at the point x (the number of particles per unit volume).
R 3 The operator (5.53)
counting the number of particles is given by N̂ = d x ρ(x).
Even more useful is the formalism of the creation and annihilation op-
erators for the 2-particle operators which act on variables of two particles.
The most frequently encountered operator of this kind is the potential energy
operator of the binary interactions
X 1 X (2)
Vint = V (2) (xi , xj ) = V (xi , xj ) , (5.60)
i<j
2
i6=j

where if the particles are identical the function V (2) must be symmetric:
V (2) (xi , xj ) = V (2) (xj , xi ). Furthermore, if the system is translationally in-
variant the function V (2) can depend only on the difference of the positions:
V (2) (xi , xj ) = V (2) (xi − xj ). It is easy to see that in the formalism of second
quantization on the generalized vectors |x1 , . . . , xN i of the position basis the
operator Vint must act in the following way
X
Vint |x1 , . . . , xN i = V (2) (xi , xj )|x1 , . . . , xN i . (5.61)
i<j

Indeed, if Vint acts as above, using the orthogonality relation

X
hxN , . . . , x1 |y1 , . . . , yN i = ζ P δ (3) (y1 − xP (1) ) . . . δ (3) (yN − xP (N ) ) ,
P

one can write

1
Z Z
d y1 . . . d3 yN hxN , . . . , x1 |Vint |y1 , . . . , yN ihyN , . . . , y1 |Ψi
3
N!
1
Z Z X
= d y1 . . . d3 yN
3
V (2) (yi , yj )hxN , . . . , x1 |y1 , . . . , yN ihyN , . . . , y1 |Ψi
N! i<j
X X
= V (2) (xi , xj )hxN , . . . , x1 |Ψi = V (2) (xi , xj )Ψ(x1 , . . . , xN ) .
i<j i<j

180
This shows that the action of Vint on the N-particle wave function in the
position representation is correctly reproduced. In H(2) the rule (5.61) is
satisfied by
1
Z Z
(2)
Vint = d x d3 y |x, yi V (x, y) hy, x| ,
3
2
and it is not too difficult to guess that in H the operator Vint is given by
1
Z Z
Vint = d x d3 y a† (x) a† (y) V (x, y) a(y) a(x) ,
3
(5.62)
2

(note the reversed order - important if the particles are fermions - of a’s
compared to a† ’s!). Notice also that the operator

1
Z Z
′
Vint = d x d3 y ρ(x)ρ(y)V (x, y) ,
3
2

with ρ(x) given by (5.59) differs from (5.62) by a local term (1/2) d3 x V (x, x)
R

which usually is infinite. We have seen that in the quantum theory of ra-
diation (outlined in section 3.8) such infinite terms had to be - due to the
improper formulation of that theory - subtracted by hand (using the so-called
normal ordering of operators).
If the interaction Vint is translationally invariant, using the rules (5.44),
in the momentum space representation we get
Z 3 Z 3 ′Z 3
1 dp dp dq †
Vint = 3 3 3
a (p + q) a† (p′ − q) Ṽ (q) a(p′ ) a(p) , (5.63)
2 (2π) (2π) (2π)

where Ṽ (q) is the Fourier transform of V (x − y):

Z 3
dq
V (x − y) = 3
Ṽ (q) eiq·(x−y) . (5.64)
(2π)

Thus, the action of Vint can be graphically represented as in figure 5.1. It is

also easy to check that

[Vint , N̂ ] = 0 , (5.65)

where N̂ is the particle number operator (5.53) (the same is true also for the
interaction term in (5.57)). If in addition the interaction depends only on
differences of coordinates, i.e. if V (2) (x, y) = V (2) (x − y), then also

[Vint , P̂] = 0 , (5.66)

181
 p+q p′ − q

Ṽ (q)
p p′

Figure 5.1: Graphical representation of the two-particle interaction (5.67)

which conserves the total momentum in the momentum representation.

where P̂ is the total momentum operator (5.54). Thus, the time independent
Hamiltonian of the form
Z 3
d p ~2 p2 †
H= a (p) a(p)
(2π)3 2m
Z 3 Z 3 ′Z 3
1 d p dp dq †
+ 3 3 3
a (p + q) a† (p′ − q) Ṽ (q) a(p′ ) a(p) , (5.67)
2 (2π) (2π) (2π)

commutes with the total momentum operator (5.54). Translations in space

and in time are then symmetries of the system. In this case one can also
construct the boost operator K satisfying the necessary commutation rules
(4.46). Finally, if V (2) (x − y) depends only on |x − y|, the generators J of the
rotations can also be constructed and commute with H. The whole Galileo
group discussed in section 4.3 is then the symmetry group of the system
and its generators satisfying the commutation rules (4.46) can be explicitly
constructed out of the available creation and annihilation operators.
The creation and annihilation operators allow to construct various forms
of interactions, also of the nonlocal type, like
Z 3 Z 3 ′ Z 3 Z 3 ′
dp d p dq dq
3 3 3 3
Ṽ (p′ , p, q′ , q) a† (p′ ) a† (p)a(q′ ) a(q) , (5.68)
(2π) (2π) (2π) (2π)

and similar ones having more integrals over independent wave vectors and
more creation and annihilation operators. Such interactions of real systems
of N nonrelativistic particles are constrained only by the requirement that
they commute with the operator N̂ (5.53), which means that they must
be built of equal numbers of the creation and annihilation operators. We
will see that the requirement that the amplitudes satisfy the cluster decom-
position principle (section 7.6) enforces the presence of one delta function
(which effectively reduces the number of independent wave vectors) in the
kernel Ṽ (p1 , . . . , p2M ) (in fact the main reason for which the formalism of
the creation and annihilation operators is so useful in nonrelativistic quan-
tum mechanics of many particle systems is that is allows to easily satisfy

182
the cluster decomposition principle). We will also see, that the requirement
of relativistic covariance of transition amplitudes imposes much more strin-
gent constraints on possible forms of interactions and makes it impossible to
preserve the particle number conservation.
The second quantized version of quantum theory of many-particle sys-
tems is obviously the usual quantum mechanics in which the central role is
played by the Hamiltonian operator and its spectrum. All standard rules
of quantum mechanics which are formulated in terms of state-vectors and
matrix elements of operators, such as e.g. perturbative expansions (station-
ary and time-dependent - see section 2.1), for computing energy spectra and
transition probabilities remain valid. New is only the method of calculating
the requisite matrix elements. In particular, it should be clear that as long
as the Hamiltonian operator H commutes with the particle number operator
N̂ (5.53), the formalism developed here is equivalent with the one based of
the multi-particle Schrödinger equation
∂
i~ Ψ(x1 , . . . , xN , t) = H(x1 , . . . , xN )Ψ(x1 , . . . , xN , t) . (5.69)
∂t
More precisely, the equation which determines the time evolution of state-
(N )
vectors in the big Hilbert space H = ⊕∞
N =0 H ,

∂
i~ |Ψ(t)i = H|Ψ(t)i , (5.70)
∂t
where |Ψ(t)i has components in each H(N )

a|voidi, |Ψ(1) i, |Ψ(2) i , ...

is equivalent to an infinite set of independent Schrödinger equations (5.69)

for N = 1, 2, . . . (for N = 0 the equation is trivial because for consistency
H should always be constructed so that H|voidi = 0). This is easily seen
by closing (5.70) with hxN , . . . , x1 | from the left and using the orthogonality
of vectors belonging to different H(N ) ’s, and the fact that H commutes with
N̂ (it has vanishing matrix elements between vectors belonging to different
H(N ) ’s). Relativistic Hamiltonians do not possess this property and, as a
result, in the relativistic case the equation (5.70) leads to infinitely many
coupled equations.

5.4 Ground state of a system of fermions. Holes

Consider a system of identical fermions whose unperturbed Hamiltonian H0

obtained by neglecting their mutual interactions is the sum of one-particle

183
 (1)
operators H0 = N
P
i=1 Hi . Suppose |li are eigenvectors of the one-particle
Hamiltonian H (1)

H (1) |li = El |li , (5.71)

where l = 1, 2, . . . (we assume that the spectrum of H (1) is discretized e.g. by

enclosing the system in a box of volume V = L3 ; alternatively the considered
particles can be confined by an external harmonic oscillator-like potential).
As a basis of the N-particle Hilbert space H(N ) one takes then the state-
vectors

|l1 , l2 , . . . , lN i = a†l1 a†l2 . . . a†lN |voidi , (5.72)

with l1 < l2 < . . . < lN . In this basis the free Hamiltonian H0 of the system
takes the simple form
X
H0 = El a†l al . (5.73)
l

If the complete Hamiltonian of the system (i.e. H0 plus the interaction

term Vint ) commutes with the particle number operator N̂ (5.53), one is not
interested in the state |voidi but rather in the state |Ω0 i which is the lowest
energy eigenvector of H0 in the subspace H(N ) of the big Hilbert space H. It
is clear that for N > 2 the energy of this state cannot be N × E1 because
of the Pauli exclusion principle reflected in the antisymmetry of N-particle
states with respect to interchanges of particles: each fermion has to occupy
a separate one-particle state. If the energy eigenvalues of the one-particle
(1)
Hamiltonian H0 are ordered so that15 E1 ≤ E2 ≤ E3 ≤ . . ., the state of
lowest possible energy - called the (unperturbed) ground state - is

|Ω0 i = |1, 2, 3, . . . , Ni = a†1 a†2 . . . a†N |voidi , (5.74)

and

EΩ0 = E1 + E2 + E3 + . . . + EN > N × E1 . (5.75)

One says that all one-particle states which are occupied in the ground state
are below the Fermi level, whereas those which are unoccupied are above the
Fermi level.
When mutual interactions Vint of the considered fermions are taken into
account the basis vectors (5.72) will not in general be eigenvectors of the
15
If the particle energy does not depend on the direction of its spin, there are always
2s + 1 (s is the particle’s spin) distinct one-particle states with the same energy; the
degeneracy of excited energy levels can be even higher if e.g. energy does not depend on
the direction of the particle’s momentum.

184
total Hamiltonian H = H0 + Vint : the perturbation Vint will cause transitions
between these states. Even if the system is initially in the ground state
(5.74) of H0 , the probability of finding it later in state of a higher H0 energy
is nonzero. This can be interpreted as the action of Vint which removes a
particle from a one-particle state which is filled in the ground state (that
is from a state below the Fermi level) and puts it in a higher energy one-
particle state above the Fermi level. Such an action, which reduces to moving
a particle from a one-particle state |˜li below the Fermi level to a state |li
above the Fermi level, can be viewed as creation of a hole of energy −El̃ and
of a particle of energy El > El̃ .
This intuitive picture can be formalized as follows. Let us redefine the
creation and annihilation operators corresponding to the (H0 one-particle
eigen)states |˜li for 1 ≤ ≤N
˜ and |li for N < l:

dl̃ ≡ a†l̃
 
b ≡ a
if 1 ≤ ˜l ≤ N , and l l
if N < l . (5.76)
d†l̃ ≡ al̃ b†l ≡ a†l

It is obvious that (l > N, ˜l ≤ N)

bl |Ω0 i = 0 , dl̃ |Ω0 i = 0 . (5.77)

Because the fermionic creation and annihilation operators satisfy anticommu-

tation rules which are symmetric (with respect to interchanging al and a†l ),
the redefinition (5.76) is perfectly possible: d†l̃ and dl̃ have all the necessary
properties to be interpreted as fermionic creation and annihilation operators,
respectively. Therefore one can consider states (created by the actions of Vint
on the ground state) of the form

|l1 , . . . , lm , ˜l1 , . . . , ˜ln i ≡ b†l1 . . . b†lm d†l̃ . . . d†l̃ |Ω0 i , (5.78)

1 n

where all li correspond to one-particle states above the Fermi level (unoccu-
pied in the ground state) and all ˜lj correspond to one-particle states below
the Fermi level (filled in the ground state). One interprets such states as
consisting of m “particles” and n “holes” (of course n = m, if H conserves
the number of particles).
The particle number operator N̂ (5.53) rewritten accordingly (its last
term is a c-number),
X X †
N̂ = b†α bα − dl̃ dl̃ + N , (5.79)
l>N ˜
l≤N

should be then interpreted as the fermion number operator and its form (5.79)
suggests that a hole should be ascribed the fermionic number −1 (that is, it

185
can be viewed as −1 particle, or as an “antiparticle”). The free Hamiltonian
rewritten in terms of the new operators takes the form
∞
X X X
H0 = El a†l al = El b†l bl + El̃ dl̃ d†l̃
l=1 l>N l̃≤N
X X X
= El b†l bl + (−El̃ ) d†l̃ dl̃ + El̃ . (5.80)
l>N l̃≤N l̃≤N

The last c-number term is simply the energy EΩ0 of the unperturbed ground
state |Ω0 i of H0 . This means that we consider “particles” and “holes” as posi-
tive and negative energy excitations, respectively over the ground state of en-
ergy EΩ0 . The interpretation of holes as particles carrying negative fermionic
number can be made even more suggestive it the one-particle Hamiltonian
H (1) is shifted by an appropriate negative constant, so that in (5.71) all one-
particle states |˜li with ˜l = 1, 2, . . . , N have negative energies. The last sum
in (5.80), which can simply be discarded as physically uninteresting (only dif-
ferences of energy levels are important), is then negative while each created
hole (the penultimate term in (5.80)) increases energy of the system (because
its creation corresponds to removing from the system of a particle in the neg-
ative energy state). Thus both types of system’s excitations (particles and
holes-antiparticles) carry now positive energies. Finally, the formulae (5.46)
rewritten in terms of the new operators take the forms
X X †
ψ(x) = bl ul (x) + dl̃ vl̃ (x) ,
l>N ˜
l≤N
X X
ψ † (x) = b†l u∗l (x) + dl̃ vl̃∗ (x) , (5.81)
l>N l̃≤N

where we have defined vl̃ (x) ≡ ul̃ for l˜ ≤ N. (The operators ψ(x) and ψ † (x)
must have as many components as do have the functions ul (x) and v˜l). In
this form the operators ψ(x) and ψ † (x) bear close resemblance to relativistic
field operators.
Any interaction Vint of fermions with an external potential can be decom-
posed similarly
X X X
Vint = Vll′ a†l al′ = Vll′ b†l bb′ + Vll̃ b†l d†l̃
l,l′ ll′ >N l>N,l̃≤N
X X X
+ Vl̃l dl̃ bl − Vl̃,l̃′ d†l̃ dl̃′ + Vl̃l̃ , (5.82)
l̃≤N,l>N l̃,l̃′ ≤N ˜
lN

into terms having different effects: the first term modifies energy of a “parti-
cle” (moving it from one one-particle state above the Fermi level to another

186
such state), the second term creates a “particle-hole” pair, the third term
annihilates such a pair; finally, the last term modifies energy of a “hole”
(moving it from one one-particle state below the Fermi level to another such
state). Similar decomposition can be applied to other multi-particle inter-
actions describing binary and higher interactions between fermions forming
the system.
The interpretation of excitations of a system of fermions in terms of
”holes” and particles proves extremely useful in applications to solid state
physics and condensed matter physics. Furthermore, the reasoning (whose
essence is the transition from the quantum mechanics of a single particle
to the many-particle theory) put forward by Dirac to make sense out of
the negative energy solutions of his relativistic wave equation for charged
spin 21 particles is in fact application of the same idea. The only difference
with the many-body non-relativistic theory developed in this Section is that
the energy spectrum of the one-particle Hamiltonian corresponding to the
Dirac equation is unbounded from below and therefore the c-number con-
stants (treated as unphysical) in (5.79) and in (5.80) are infinite. While
successful in solving the problem of negative energy states associated with
relativistic wave equations for fermions, this reasoning, for obvious reason,
cannot solve the analogous problem of relativistic wave equations supposed
to apply to bosons, like the Klein-Gordon equation. As will be shown, rela-
tivistic quantum mechanics of many particles of arbitrary spins can however
be consistently formulated without any reference to relativistic wave equa-
tions. Wave equations are not the basis of the relativistic quantum field
theory (although such an impression can be drawn from older textbooks). It
might therefore seem that in the modern approach to quantum field theory
the interpretation of negative energy states as holes is obsolete and should
be regarded purely as a historical curiosity were it not for the fact that the
picture of a filled sea of negative energy states comes back (in somewhat
different disguise) in the path integral approach to quantum field theories
involving fermions and seems indispensable to understand highly nontrivial
nonperturbative phenomena like nonconservation of the fermion number or
baryon number.

5.5 Analogy with coupled harmonic oscillators

In order to provide a link between quantum mechanics of many particle sys-

tems and quantization of classical fields we now consider quantization of a
system of N coupled harmonic oscillators to which, in the simple approxima-
tion of small departures from the (classical) equilibrium solution any system
of e.g. point masses connected by springs can be reduced. The classical

187
Lagrangian of such a system has in general the form
1X 1X
L0 = Tij q̇ i q̇ j − Vij q i q j , (5.83)
2 i,j 2 i,j

in which Vij and Tij are symmetric positive definite (constant) N × N matri-
ces.16 The canonical momenta conjugated to the variables q i are pj = Tji q̇ i
and the corresponding classical Hamiltonian reads
1 X −1 ij 1X
H0 = (T ) pi pj + Vij q i q j . (5.84)
2 i,j 2 i,j

Upon quantization q i (t) and pi (t) become Schrödinger picture operators sat-
isfying the standard relation [q̂ i , p̂j ] = i~δji but the Hamiltonian (5.84) ex-
pressed through the operators ai and a†i related in the standard way to q̂ i
and p̂j would not have the form of the sum of the a†i ai terms. To find the
spectrum and the Hamiltonian eigenstates one can solve first the classical
problem by introducing normal the mode coordinates Qα (t) by the formula
q i (t) = Ai(α) Qα (t) , (5.85)

where the vectors Ai(α) α = 1, . . . , N are solutions of the eigenproblem

−ωα2 Tij + Vij Aj(α) = 0 .

(5.86)

The vectors Ai(α) should be chosen orthonormal in the scalar product set by
the matrix Tij : Ai(α) Tij Aj(β) = δαβ . The Lagrangian expressed through the
normal variables is (working out the potential energy terms one uses (5.86))
1 1
L0 = Tij Ai(α) Q̇α Aj(β) Q̇β − Vij Ai(α) Qα Aj(β) Qβ
2 2
1 1 1 1
= δαβ Q̇α Q̇β − Ai(α) Qα ωβ2 Tij Aj(β) Qβ = δαβ Q̇α Q̇β − ωβ2 δαβ Qα Qβ ,
2 2 2 2
To the momenta pi , the momenta Pα conjugated to the new variables Qα are
related by17 Pα = pi Ai(α) and, since (T −1 )ij = Ai(a) Aj(a) , the corresponding
Hamiltonian is
1 1
H0 = Pα Pα + ωα2 Qα Qα . (5.87)
2 2
16
Here we are more restrictive than in classical mechanics and assume that both T and
V matrices have strictly positive eigenvalues.
17
As usually,

∂ L̃(Q, Q̇) ∂L(q(Q), q̇(Q, Q̇)) ∂ q̇ i

Pa = = ,
∂ Q̇a ∂ q̇ i ∂ Q̇a

and ∂ q̇ i /∂ Q̇a = ∂q i /∂Qa .

188
The system can be now quantized by promoting the variables Qα (t) and Pα (t)
to Schrödinger picture operators. It can be checked that the commutation
relations of the canonical variables remain unchanged: [Q̂α , P̂β ] = i~δβα .
Next, we define the operators
r   r  
ωα α i † ωα α i
Aα = Q̂ + P̂α , Aα = Q̂ − P̂α , (5.88)
2~ ωα 2~ ωα

which satisfy the relation [Aα , A†β ] = δαβ . Expressed in terms of these
operators the Hamiltonian is diagonal:
N
 
X
† 1
H0 = ~ωα Aα Aα + . (5.89)
α=1
2

The eigenvectors of H0 are therefore of the form

N
nα !
Y A†α
|n1 , n2 , . . . , nN i = √ |0, 0, . . . , 0i , (5.90)
α=1
nα !

where |0, 0, . . . , 0i (with N zeroes) is the ground state annihilated by all aα ’s.
The energy of the state (5.90) is
N N
X 1X
En1 ,n2 ,...,nN = nα ~ωα + ~ωα . (5.91)
α=1
2 α=1

The second term - the energy of the ground state can be subtracted if we
declare that we are interested only in the differences of energies of the states.
The Hamiltonian of this type can describe e.g. quantized vibrations of
a crystal lattice.18 The same structure, with N = ∞ and α replaced by
(k, λ) was also obtained in section 3.8 as a result of quantizing the free
electromagnetic radiation field enclosed in a box. By analogy with the form
of the Hamiltonian
X
H= E(p) a† (p)a(p) ,
p

of a system of N noninteracting spinless particles (with periodic boundary

conditions in a box of volume V = L3 ) in the second quantization formalism
discussed in the preceding subsections, the crystal lattice quantum states
18
In the case of a crystal lattice one usually neglects boundary effects and considers
N = ∞ coupled oscillators; after the change of variables, the normal modes are then
labeled by a continuous parameter K (called quasi-momentum) rather than by a discrete
index α. The occupation number representation (5.90) has then to be replaced by the
representation (5.94).

189
|n1 , n2 , . . . , nN i (the states |nk1 λ1 , nk2 λ2 , . . .i of the radiation field) are inter-
preted as the states consisting of n1 phonons of type 1 (nk1 λ1 photons with the
wave vector k1 and the polarization λ1 ), n2 phonons of type 2 (nk1 λ1 photons
of type two), etc. (there and N possible types of phonons corresponding to
N different lattice vibration modes of frequencies ωα ; the number of types of
photons is not limited). Since the energy of such a state has the form (5.91),
that is, it is additive (the energy contributed by each phonon or photon is
independent of the presence of other phonons or photons), the phonons and
photon described by the Hamiltonian (5.89) are noninteracting.
The action of the operators Aα and A†α on the states (5.90) is standard:
√
A†α |n1 , . . . , nα , . . . , nN i = nα + 1 |n1 , . . . , nα + 1, . . . , nN i
√
Aα |n1 , . . . , nα , . . . , nN i = nα |n1 , . . . , nα − 1, . . . , nN i . (5.92)

It corresponds to the action of the bosonic particles creation and annihilation

operators in the occupation number representation (5.33). There is however
one difference between the second-quantized version of nonrelativistic quan-
tum mechanics of particles and phonons: in the former case the number N of
particles is fixed (in nonrelativistic systems it does not change in time, even
if the interactions are taken into account), so that in the occupation number
representation always n1 + n2 + . . . = N. In the latter, the system can be
excited to an arbitrary state |n1 , . . . , nα , . . . , nN i for which the eigenvalues
of the operator A†α Aα (giving the number of phonons of type α), as well as
of the operator of total number of phonons
N
X
N̂ = A†α Aα , (5.93)
α=1

can be arbitrary (in contrast, it is the number of the phonon types that is
finite). The difference disappears however if one allows for arbitrary numbers
of particles (e.g. by working in the Grand Canonical Ensemble) in the first
case and for an infinite number of vibration modes in the second case.
The basis of states of phonons can be also labeled differently:

|α1 , α2 , . . . , αn , . . .i = A†α1 A†α2 . . . A†αm . . . |Ω0 i , (5.94)

where α1 ≤ α2 ≤ . . . and the state |Ω0 i ≡ |0, 0, . . . , 0i (with N zeroes), the

ground state of (5.90), is annihilated by all Aα ’s. The states |α1 , α2 , . . . , αn , . . .i
are related to the basis by
  n1   n2  nN
† †
|1, . . . , 1, 2, . . . , 2, . . . , N, . . . , Ni = A1 A2 . . . A†N |Ω0 i
p
= n1 !n2 ! . . . nN !|n1 , n2 . . . , nN i . (5.95)

190
For small numbers of phonons this new representation is more convenient.
The action of the creation and annihilation operators on the states (5.94) is
given by

A†α |α1 , . . . , αn , . . .i = |α, α1, . . . , αn , . . .i ,

XN
Aα |α1 , . . . , αn , . . .i = δαk ,α |α1 , . . . , (no αk ), . . .i , (5.96)
k=1

and looks the same as the action (5.26) and (5.27) of the creation and an-
nihilation operators corresponding to one-particle states. Thus, barring the
difference just explained, the description of quantized crystal lattice vibra-
tions in terms of phonons is formally the same as of many particle systems.
Consider finally perturbations of the initial Lagrangian (5.83) by polyno-
mial terms of order higher than second in the variables qi
N
X N
X
i j k
Vint = Vijk q q q + Vijkl q i q j q k q l + . . . (5.97)
i,j,k=1 i,j,k,l=1

When rewritten in terms of the operators aα , a†α they give rise (among others)
to terms of the form

Vint ∋ B A†α A†β A†γ + C A†α A†β Aγ + D A†α A†β A†γ A†κ + F Aα Aβ Aγ Aκ , (5.98)

which contain nonequal numbers of the creation and annihilation operators

and, hence, do not commute with the total number of phonons operator
(5.93). The time evolution will therefore not preserve the number of phonons.
In contrast, Hamiltonians of nonrelativistic particle systems always preserve
the number of particles. We will see however, that the requirement of rel-
ativistic covariance of transition amplitudes enforces nonconservation of the
number of particles. Hence, relativistic quantum mechanics of particles nat-
urally acquires features of phonon systems.

5.6 Summary

We have constructed the second quantized version of quantum mechanics of

nonrelativistic many-particle systems starting from its standard version based
on the multiparticle Schrödinger equation and multiparticle wave functions
Ψ(x1 , . . . , xN ).
An alternative approach would be to start from symmetry principles and
to assume that the Galileo group is realized (projectively) in some Hilbert
space (playing the role of the “big” Hilbert space H) by unitary operators.

191
One could then classify possible eigenstates |pi of commuting Galileo group
generators H, P and identify them with particles. Operators a(p) and a† (p)
could be then associated with these states. Finally interactions Vint could be
constructed from the operators a(p) and a† (p) respecting assumed symmetry
principles.
We will follow such an approach in the next chapter to construct the
Hilbert space of relativistic quantum theory of particles.

192