Classical Electrodynamics

This content has been downloaded from IOPscience. Please scroll down to see the full text.
Download details:
IP Address: 168.176.5.118
This content was downloaded on 22/02/2023 at 13:18
Please note that terms and conditions apply.
You may also like:
Diffraction in high numerical aperture systems: polarization effects

L Ciocci, R M Echarri and J M Simon
Understanding zero-point energy in the context of classical electromagnetism

Timothy H Boyer
Eigenmode generation by a given current in anisotropic and gyrotropic media

I N Toptygin and G D Fleishman
Angular momentum in classical electrodynamics

A L Barabanov
Contrasting interactions between dipole oscillators in classical and quantum theories:

illustrations of unretarded van der Waals forces
Timothy H Boyer
Classical Electrodynamics
Lecture notes
Lecture notes
Konstantin K Likharev
IOP Publishing, Bristol, UK

ª Copyright Konstantin K Likharev 2018
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system
or transmitted in any form or by any means, electronic, mechanical, photocopying, recording
or otherwise, without the prior permission of the publisher, or as expressly permitted by law or
under terms agreed with the appropriate rights organization. Multiple copying is permitted in
accordance with the terms of licences issued by the Copyright Licensing Agency, the Copyright
Clearance Centre and other reproduction rights organisations.
Permission to make use of IOP Publishing content other than as set out above may be sought
at permissions@iop.org.
Konstantin K Likharev has asserted his right to be identified as the author of this work in
accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988.
ISBN 978-0-7503-1404-6 (ebook)

ISBN 978-0-7503-1405-3 (print)
ISBN 978-0-7503-1406-0 (mobi)
DOI 10.1088/978-0-7503-1404-6
Version: 20180501
IOP Expanding Physics

ISSN 2053-2563 (online)
ISSN 2054-7315 (print)
British Library Cataloguing-in-Publication Data: A catalogue record for this book is available
from the British Library.
Published by IOP Publishing, wholly owned by The Institute of Physics, London
IOP Publishing, Temple Circus, Temple Way, Bristol, BS1 6HG, UK
US Office: IOP Publishing, Inc., 190 North Independence Mall West, Suite 601, Philadelphia,
PA 19106, USA
Contents
Preface to the EAP Series ix

Preface to Classical Electrodynamics: Lecture Notes xii
Acknowledgments xiii
Notation xiv
1 Electric charge interaction 1-1

1.1 The Coulomb law 1-1
1.2 The Gauss law 1-5
1.3 Scalar potential and electric field energy 1-11
1.4 Problems 1-21
Reference 1-24
2 Charges and conductors 2-1

2.1 Polarization and screening 2-1
2.2 Capacitance 2-6
2.3 The simplest boundary problems 2-13
2.4 Using other orthogonal coordinates 2-18
2.5 Variable separation—Cartesian coordinates 2-26
2.6 Variable separation—polar coordinates 2-32
2.7 Variable separation—cylindrical coordinates 2-37
2.8 Variable separation—spherical coordinates 2-47
2.9 Charge images 2-53
2.10 Green’s functions 2-61
2.11 Numerical methods 2-65
2.12 Problems 2-68
References 2-77
3 Dipoles and dielectrics 3-1

3.1 Electric dipole 3-1
3.2 Dipole media 3-7
3.3 Polarization of dielectrics 3-12
3.4 Electrostatics of linear dielectrics 3-18
3.5 Energy of electric field in a dielectric 3-23
3.6 Problems 3-27
References 3-32
v
4 DC currents 4-1
4.1 Continuity equation and the Kirchhoff laws 4-1
4.2 The Ohm law 4-4
4.3 Boundary problems 4-8
4.4 Energy dissipation 4-14
4.5 Problems 4-15
References 4-17
5 Magnetism 5-1
5.1 Magnetic interaction of currents 5-1
5.2 Vector-potential and the Ampère law 5-9
5.3 Magnetic energy, flux, and inductance 5-16
5.4 Magnetic dipole moment, and magnetic dipole media 5-24
5.5 Magnetic materials 5-30
5.6 Systems with magnetics 5-35
5.7 Problems 5-44
References 5-49
6 Electromagnetism 6-1
6.1 Electromagnetic induction 6-1
6.2 Magnetic energy revisited 6-4
6.3 Quasi-static approximation and skin effect 6-7
6.4 Electrodynamics of superconductivity and gauge invariance 6-12
6.5 Electrodynamics of macroscopic quantum phenomena 6-18
6.6 Inductors, transformers, and ac Kirchhoff laws 6-27
6.7 Displacement currents 6-30
6.8 Finally, the full Maxwell equation system 6-32
6.9 Problems 6-38
References 6-44
7 Electromagnetic wave propagation 7-1

7.1 Plane waves 7-1
7.2 Attenuation and dispersion 7-7
7.3 Reflection 7-18
7.4 Refraction 7-25
7.5 Transmission lines: TEM waves 7-34
vi
7.6 Waveguides: H and E waves 7-42

7.7 Dielectric waveguides, optical fibers, and paraxial beams 7-49
7.8 Resonators 7-63
7.9 Energy loss effects 7-68
7.10 Problems 7-73
References 7-79
8 Radiation, scattering, interference, and diffraction 8-1

8.1 Retarded potentials 8-1
8.2 Electric dipole radiation 8-5
8.3 Wave scattering 8-10
8.4 Interference and diffraction 8-13
8.5 The Huygens principle 8-23
8.6 Fresnel and Fraunhofer diffraction patterns 8-28
8.7 Geometrical optics placeholder 8-32
8.8 Fraunhofer diffraction from more complex scatterers 8-33
8.9 Magnetic dipole and electric quadrupole radiation 8-37
8.10 Problems 8-40
References 8-44
9 Special relativity 9-1

9.1 Einstein postulates and the Lorentz transform 9-1
9.2 Relativistic kinematic effects 9-7
9.3 Four-vectors, momentum, mass, and energy 9-17
9.4 More on four-vectors and four-tensors 9-24
9.5 Maxwell equations in the four-form 9-29
9.6 Relativistic particles in electric and magnetic fields 9-35
9.7 Analytical mechanics of charged particles 9-43
9.8 Analytical mechanics of the electromagnetic field 9-50
9.9 Problems 9-58
References 9-63
10 Radiation by relativistic charges 10-1

10.1 Liénard–Wiechert potentials 10-1
10.2 Radiation power 10-8
10.3 Synchrotron radiation 10-12
vii
10.4 Bremsstrahlung and Coulomb losses 10-23

10.5 Density effects and Cherenkov radiation 10-30
10.6 Radiation’s back-action 10-38
10.7 Problems 10-42
References 10-44
Appendices
A Selected mathematical formulas A-1
B Selected physical constants B-1
Bibliography 13-1
viii
Preface to the EAP Series
Essential Advanced Physics

Essential Advanced Physics (EAP) is a series of lecture notes and problems with
solutions, consisting of the following four parts1:
• Part CM: Classical Mechanics (a one-semester course),
• Part EM: Classical Electrodynamics (two semesters),
• Part QM: Quantum Mechanics (two semesters), and
• Part SM: Statistical Mechanics (one semester).
Each part includes two volumes: Lecture Notes and Problems with Solutions, and
an additional file Test Problems with Solutions.
Distinguishing features of this series—in brief

• condensed lecture notes (∼250 pp per semester)—much shorter than most
textbooks
• emphasis on simple explanations of the main notions and phenomena of
physics
• a focus on problem solution; extensive sets of problems with detailed model
solutions
• additional files with test problems, freely available to qualified university
instructors
• extensive cross-referencing between all parts of the series, which share style
and notation
Level and precursors

The goal of this series is to bring the reader to a general physics knowledge level
necessary for professional work in the field, regardless on whether the work is
theoretical or experimental, fundamental or applied. From the formal point of view,
this level (augmented by a few special topic courses in a particular field of
concentration, and of course by an extensive thesis research experience) satisfies
the typical PhD degree requirements. Selected parts of the series may be also
valuable for graduate students and researchers of other disciplines, including
astronomy, chemistry, mechanical engineering, electrical, computer and electronic
engineering, and material science.
The entry level is a notch lower than that expected from a physics graduate from
an average US college. In addition to physics, the series assumes the reader’s
familiarity with basic calculus and vector algebra, to such an extent that the meaning
of the formulas listed in appendix A, ‘Selected mathematical formulas’ (reproduced
at the end of each volume), is absolutely clear.
1
Note that the (very ambiguous) term mechanics is used in these titles in its broadest sense. The acronym EM
stems from another popular name for classical electrodynamics courses: Electricity and Magnetism.
ix
Origins and motivation

The series is a by-product of the so-called ‘core physics courses’ I taught at Stony
Brook University from 1991 to 2013. My main effort was to assist the development
of students’ problem-solving skills, rather than their idle memorization of formulas.
(With a certain exaggeration, my lectures were not much more than introductions to
problem solution.) The focus on this main objective, under the rigid time restrictions
imposed by the SBU curriculum, had some negatives. First, the list of covered
theoretical methods had to be limited to those necessary for the solution of the
problems I had time to discuss. Second, I had no time to cover some core fields of
physics—most painfully general relativity2 and quantum field theory, beyond a few
quantum electrodynamics elements at the end of Part QM.
The main motivation for putting my lecture notes and problems on paper, and
their distribution to students, was my desperation to find textbooks and problem
collections I could use, with a clear conscience, for my purposes. The available
graduate textbooks, including the famous Theoretical Physics series by Landau and
Lifshitz, did not match the minimalistic goal of my courses, mostly because they are
far too long, and using them would mean hopping from one topic to another,
picking up a chapter here and a section there, at a high risk of losing the necessary
background material and logical connections between the course components—and
the students’ interest with them. In addition, many textbooks lack even brief
discussions of several traditional and modern topics that I believe are necessary
parts of every professional physicist’s education3.
On the problem side, most available collections are not based on particular
textbooks, and the problem solutions in them either do not refer to any background
material at all, or refer to the included short sets of formulas, which can hardly be
used for systematic learning. Also, the solutions are frequently too short to be useful,
and lack discussions of the results’ physics.
Style
In an effort to comply with the Occam’s Razor principle4, and beat Malek’s law5, I
have made every effort to make the discussion of each topic as clear as the time/
space (and my ability :-) permitted, and as simple as the subject allowed. This effort
has resulted in rather succinct lecture notes, which may be thoroughly read by a
student during the semester. Despite this briefness, the introduction of every new
2
For an introduction to this subject, I can recommend either a brief review by S Carroll, Spacetime and
Geometry (2003, New York: Addison-Wesley) or a longer text by A Zee, Einstein Gravity in a Nutshell (2013,
Princeton University Press).
3
To list just a few: the statics and dynamics of elastic and fluid continua, the basics of physical kinetics,
turbulence and deterministic chaos, the physics of computation, the energy relaxation and dephasing in open
quantum systems, the reduced/RWA equations in classical and quantum mechanics, the physics of electrons
and holes in semiconductors, optical fiber electrodynamics, macroscopic quantum effects in Bose–Einstein
condensates, Bloch oscillations and Landau–Zener tunneling, cavity quantum electrodynamics, and density
functional theory (DFT). All these topics are discussed, if only briefly, in my lecture notes.
4
Entia non sunt multiplicanda praeter necessitate—Latin for ‘Do not use more entities than necessary’.
5
‘Any simple idea will be worded in the most complicated way’.
x
physical notion/effect and of every novel theoretical approach is always accom-

panied by an application example or two.
The additional exercises/problems listed at the end of each chapter were carefully
selected6, so that their solutions could better illustrate and enhance the lecture
material. In formal classes, these problems may be used for homework, while
individual learners are strongly encouraged to solve as many of them as practically
possible. The few problems that require either longer calculations, or more creative
approaches (or both), are marked by asterisks.
In contrast with the lecture notes, the model solutions of the problems (published
in a separate volume for each part of the series) are more detailed than in most
collections. In some instances they describe several alternative approaches to the
problem, and frequently include discussions of the results’ physics, thus augmenting
the lecture notes. Additional files with sets of shorter problems (also with model
solutions) more suitable for tests/exams, are available for qualified university
instructors from the publisher, free of charge.
Disclaimer and encouragement

The prospective reader/instructor has to recognize the limited scope of this series
(hence the qualifier Essential in its title), and in particular the lack of discussion of
several techniques used in current theoretical physics research. On the other hand, I
believe that the series gives a reasonable introduction to the hard core of physics—
which many other sciences lack. With this hard core knowledge, today’s student will
always feel at home in physics, even in the often-unavoidable situations when
research topics have to be changed at a career midpoint (when learning from
scratch is terribly difficult—believe me :-). In addition, I have made every attempt to
reveal the remarkable logic with which the basic notions and ideas of physics
subfields merge into a wonderful single construct.
Most students I taught liked using my materials, so I fancy they may be useful to
others as well—hence this publication, for which all texts have been carefully
reviewed.
6
Many of the problems are original, but it would be silly to avoid some old good problem ideas, with long-lost
authorship, which wander from one textbook/collection to another one without references. The assignments
and model solutions of all such problems have been re-worked carefully to fit my lecture material and style.
xi
Preface to Classical
Electrodynamics: Lecture Notes
The structure of this classical electrodynamics course is quite traditional7; in order to

address the most important problems of the field—which involve not only charged
point particles, but also conducting, dielectric, and magnetic media—the electro-
magnetic interactions of charges are discussed in parallel with the simplest models of
the electric and magnetic properties of materials.
Also following tradition, I use this volume (mostly chapter 2) as a convenient
platform for the discussion of various methods of the solution of partial differential
equations, using:
• the polar, cylindrical, spherical, and other curvilinear orthogonal coordinates
(sections 2.3 and 2.4), including conformal mapping (section 2.4);
• the variable separation method8, also in various coordinates (sections 2.5–
2.8), involving the Bessel functions (section 2.7), and the Legendre poly-
nomials and the associated Legendre functions (section 2.8)9;
• the spatial (section 2.10), temporal (section 7.2)10, and spatial-temporal
(chapter 6) Green’s functions; and
• the finite-difference variety of numerical methods, with a very brief discussion
of the finite-element approach (section 2.11).
I hope the discussion of these methods of mathematical physics, using the

platform of specific problems of electrodynamics, makes them more clear and vivid.
One more traditional part of classical electrodynamics is an introduction to
special relativity, because—although this field includes a substantial classical
mechanics component—it is the electrodynamics which makes the relativistic
approach obligatory. The narrative logic accepted in this course dictates the
discussion of the special relativity effects in the very end of the course—chapters 9
and 10.
7
A notable exception from tradition is the famous Theoretical Physics series by L Landau and E Lifshitz, with
two different volumes (#2 and #8) devoted to, respectively, electromagnetic interactions in free space and the
electrodynamics of continua. I believe that such approach may only be justified if the first topic is discussed (as
it is in the Landau and Lifshitz series) together with general relativity, for which I unfortunately could not find
time in this series.
8
This method was already briefly discussed in Part CM section 6.5, and then used in Part CM sections 6.6 and
8.4. One more family of special functions, the Fresnel integrals, are discussed in section 8.6.
9
Due to their key importance for quantum mechanics, these functions will be also discussed in Part QM
section 3.5.
10
Such functions were also discussed in Part CM section 5.1.
xii
Acknowledgments
I am extremely grateful to my faculty colleagues and other readers of the preliminary

(circa 2013) version of this series, who provided feedback on certain sections; here
they are listed in alphabetical order11: A Abanov, P Allen, D Averin, S Berkovich,
P-T de Boer, M Fernandez-Serra, R F Hernandez, A Korotkov, V Semenov, F
Sheldon, and X Wang. (Obviously, these kind people are not responsible for any
remaining deficiencies.)
A large part of my scientific background and experience, as reflected in these
materials, came from my education, and then research, in the Department of Physics
of Moscow State University from 1960 to 1990. The Department of Physics and
Astronomy of Stony Brook University provided a comfortable and friendly
environment for my work during the following 25+ years.
Last but not least, I would like to thank my wife Lioudmila for all her love, care,
and patience—without these, this writing project would have been impossible.
I know very well that my materials are still far from perfection. In particular, my
choice of covered topics (always very subjective) may certainly be questioned. Also,
it is almost certain that despite all my efforts, not all typos have been weeded out.
This is why all remarks (however candid) and suggestions from readers will be
greatly appreciated. All significant contributions will be gratefully acknowledged in
future editions.
Stony Brook, NY
11
I am very sorry for not keeping proper records from the beginning of my lectures at Stony Brook, so I cannot
list all the numerous students and TAs who have kindly attracted my attention to typos in earlier versions of
these notes. Needless to say, I am very grateful to all of them as well.
xiii
Notation
Abbreviations Fonts Symbols

12 .
c.c. complex conjugate F, F scalar variables time differentiation operator (d/dt)
h.c. Hermitian conjugate F, F vector variables ∇ spatial differentiation vector (del)
Fˆ , F ˆ scalar operators ≈ approximately equal to
ˆ Fˆ vector operators
F, B of the same order as
F matrix ∝ proportional to
Fjj′ matrix element ≡ equal to by definition (or evidently)
⋅ scalar (‘dot-’) product
× vector (‘cross-’) product
__
time averaging
〈 〉 statistical averaging
[ , ] commutator
{ , } anticommutator
Prime signs
The prime signs (′, ″, etc) are used to distinguish similar variables or indices (such as j
and j′ in the matrix element above), rather than to denote derivatives.
Parts of the series

Part CM: Classical Mechanics Part EM: Classical Electrodynamics
Part QM: Quantum Mechanics Part SM: Statistical Mechanics
Appendices
Appendix A: Selected mathematical formulas
Appendix B: Selected physical constants
Formulas
The abbreviation Eq. may mean any displayed formula: either the equality, or
inequality, or equation, etc.
12
The same letter, typeset in different fonts, typically denotes different variables.
xiv
IOP Publishing
Lecture notes
Chapter 1
Electric charge interaction
This brief chapter describes the basics of electrostatics, the study of interactions
between static (or slowly moving) electric charges. Much of this material should be
known to the reader from his or her undergraduate studies; because of that, the
explanations will be very brief 1.
1.1 The Coulomb law

A serious discussion of any part of classical electrodynamics, starting from the
electrostatics, requires a common agreement on the meaning of the following notions2:
• electric charges qk, as revealed, most explicitly, by observation of the

electrostatic interaction between the charged particles;
• the electric charge conservation, meaning that the algebraic sum of qk of all
particles inside any closed volume is conserved, unless the charged particles
cross the volume’s border; and
• a point charge, meaning an ultimately small (‘point’) charged particle, whose
position in space is completely described (in a given reference frame) by its
radius-vector r.
I will assume that these notions are well known to the reader—although my
strong advice is to give some thought to their vital importance. Using these notions,
the Coulomb law3 for the interaction of two point, stationary charges may be
formulated as follows:
1
For remedial reading, virtually any undergraduate text on electricity and magnetism may be used; I can
recommend, for example, [1].
2
On top of the more general notions of the classical Cartesian space, point particles, and forces, which are used
in classical mechanics—see, e.g. Part CM section 1.1.
3
Discovered experimentally in the early 1780s, and formulated in 1785 by C-A de Coulomb.
doi:10.1088/978-0-7503-1404-6ch1 1-1 ª Konstantin K Likharev 2018

Classical Electrodynamics: Lecture notes
Figure 1.1. Coulomb force directions (for the case qkqk′ > 0).
rk − rk ′ qq
Fkk ′ = κqkqk ′ 3
≡ κ k 2k ′ n kk ′, with R kk ′ ≡ rk − rk ′, (1.1)
rk − rk ′ R kk ′
where Fkk′ denotes the electrostatic (‘Coulomb’) force exerted on charge number k
by charge number k′, and nkk′ ≡ Rkk′/Rkk′ is a unit vector directed along the vector
Rkk′ of the relative position of the charges—see figure 1.1.
I am confident that this law is very familiar to the reader, but a few comments
may still be due:
(i) Flipping indices k and k′, we see that Eq. (1.1) complies with Newton’s third law:
the reciprocal force is equal in magnitude but opposite in direction: Fk′k = −Fkk′.
(ii) Since the vector Rkk′ ≡ rk − rk′ is directed from point rk′ toward point rk
(figure 1.1), Eq. (1.1) implies that charges of the same sign (i.e. with qkqk′ > 0) repulse,
while those with opposite signs (qkqk′ < 0) attract each other.
(iii) In some textbooks, the Coulomb law (1.1) is given with the qualifier ‘in free
space’ or ‘in a vacuum’. However, Eq. (1.1) actually remains valid even in the
presence of other charges, for example the charges of quasi-continuous media that
may surround the two charges (number k and k′) under consideration.
The confusion stems from the fact (to be discussed in detail in chapter 3) that in
some cases it is convenient to formally represent the effect of other charges as an
effective modification of the Coulomb law.
(iv) The constant κ in Eq. (1.1) depends on the system of units we use. In the
Gaussian units, κ is set to 1, for the price of introducing a special unit of charge
(the statcoulomb) that would fit the experimental data for (1.1), for forces Fkk′
measured in the Gaussian units (dynes). On the other hand, in the International
System (SI) of units, the charge unit is one coulomb (abbreviated C)4, close to 3 × 109
statcoulombs, and κ is different from unity5:
4
In the formal metrology, one coulomb is defined as the charge carried over by a constant current of one
ampere (see chapter 5 for its definition) during 1 s.
5
ε0 is called either the electric constant or the free-space permittivity; from Eq. (1.2) with the free-space speed of
light c ≈ 3 × 108 m c−1, ε0 ≈ 8.85 × 10−12 SI units. For more accurate values of the constants, and their brief
discussion, see appendix B.
1-2
1
κ SI = ≡ 10−7c 2 , (1.2)
4πε0
where c is the speed of light. Unfortunately, the continuing struggle between zealot
proponents of these two systems resembles the not-so-nice features of a religious
war, with a similarly slim chance of any side winning in the foreseeable future. In my
humble view, each of these systems has its advantages and handicaps (to be noted
below on several occasions), and every educated physicist should have no problem
with using either of them. Following the insistent recommendations of international
scientific unions, I am using SI units throughout this series. However, for the readers’
convenience, in this course (where the difference between the Gaussian and SI
systems is particularly significant) I will write the most important formulas with the
constant (1.2) clearly displayed—for example, Eq. (1.1) as
q q r − rk ′
Fkk ′ = k k ′ k , (1.3)
4πε0 rk − rk ′ 3
so that the transfer to the Gaussian units may be performed just by the formal
replacement 4πε0 → 1. Moreover, in the cases when the transfer is not obvious, I will
duplicate such formulas in the Gaussian units.
In addition to Eq. (1.3), another key experimental law of electrostatics is the

linear superposition principle: the electrostatic forces exerted on some point charge
(say, qk) by other charges do not affect each other and add up as vectors to form the
net force:
Fk = ∑ Fkk ′, (1.4)
k ′≠ k
where the summation is extended over all charges but qk, and the partial force Fkk′ is
described by Eq. (1.3). The fact that the sum is restricted to k′ ≠ k means that a point
charge, in statics, does not interact with itself. This fact may look trivial from
Eq. (1.3), whose right-hand side diverges at rk → rk′, but becomes less evident (though
still true) in quantum mechanics where the charge of even an elementary particle is
effectively spread around some volume, together with the particle’s wavefunction6.
Now we may combine Eqs. (1.3) and (1.4) to obtain the following expression for
the net force F acting on some charge q located at point r:
1 r − rk ′
F( r ) = q ∑ qk ′
4πε0 r ≠ r r − rk ′ 3
. (1.5)
k′
This equality implies that it makes sense to introduce the notion of the electric field
at point r (as an entity independent of the probe charge q), characterized by the
following vector:
6
Moreover, there are some widely used approximations, e.g. the Kohn–Sham equations in the density
functional theory (DFT) of multi-particle systems, which essentially violate this law, thus limiting the accuracy
and applicability of these approximations—see, e.g. Part QM section 8.4.
1-3
F( r )
E(r) ≡ , (1.6)
q
formally called the electric field strength—but much more frequently, just the
‘electric field’. In these terms, Eq. (1.5) becomes
1 r − rk ′
E(r) = ∑ qk ′
4πε0 r ′≠ r r − rk ′ 3
. (1.7)
k
The notion of field is just convenient is electrostatics, but becomes virtually

unavoidable for description of time-dependent phenomena (such as electromagnetic
waves), where the electromagnetic field shows up as a specific form of matter, with
zero rest mass, and hence different from the usual ‘material’ particles.
Many real-world problems involve multiple point charges located so closely that it
is possible to approximate them with a continuous charge distribution. Indeed, let us
consider a group of many (dN ≫ 1) close charges, located at points rk′, all within an
elementary volume d3r′. For relatively distant field observation points, with ∣r − rk′∣ ≫
dr′, the geometrical factor in the corresponding terms of Eq. (1.7) is essentially the
same. As a result, these charges may be treated as a single elementary charge dQ(r′).
Since at dN ≫ 1 this elementary charge is proportional to the elementary volume d3r′,
we can define the local 3D charge density ρ(r′) by the relation7
ρ(r′)d 3r′ ≡ dQ(r′) ≡ ∑ qk ′, (1.8)
rk ′ ∈ d 3r ′
and rewrite Eq. (1.7) as an integral (over the volume containing all essential
charges):
1
E(r) =
4πε0
∫ ρ(r′) rr −− rr′′3 d 3r′. (1.9)
Note that for a continuous, smooth charge density ρ(r′), the integral in Eq. (1.9) does
not diverge at R ≡ r − r′ → 0, because in this limit the fraction under the integral
increases as R−2, i.e. slower than the decrease of the elementary volume d3r′,
proportional to R3.
Let me emphasize the dual role of Eq. (1.9). In the case if ρ(r) is a continuous
function representing the average charge, defined by Eq. (1.8), Eq. (1.9) is not valid
at distances ∣r − rk′∣ of the order of the distance between the adjacent point charges,
i.e. does not describe rapid variations of the electric field at these distances. Such an
approximate, smoothly changing field E(r) is called macroscopic; we will repeatedly
return to this notion8 in the following chapters. On the other hand, Eq. (1.9) may be
7
The 2D (areal) charge density σ and the 1D (linear) density λ may be defined absolutely similarly: dQ = σd2r,
dQ = λdr. Note that a finite value of σ and λ means that the volumic density ρ is formally infinite at the charge
location; for example for a plane z = 0, charged with a constant areal density σ, ρ = σδ(z), where δ(z) is the
Dirac delta-function.
8
It was formally introduced by H Lorentz in the early 1900 s for the description of dielectrics—see section 3.3.
1-4
also used for the description of the exact (‘microscopic’) field of discrete point charges,
employing the notion of Dirac’s δ-function9, which is the mathematical approxima-
tion for a very sharp function equal to zero everywhere but one point, and still having
a finite integral (equal to 1). Indeed, in this formalism, a set of point charges qk′
located in points rk′ may be represented by the pseudo-continuous density
ρ(r′) = ∑qk ′δ(r′ − rk ′). (1.10)
k′
Plugging this expression into Eq. (1.9), we return to its exact, discrete version
(1.7). In this sense, Eq. (1.9) is exact, and we may use it as the general expression for
the electric field.
1.2 The Gauss law

Due to this extension of Eq. (1.9) to point (‘discrete’) charges, it may seem that we do
not need anything else for solving any problem of electrostatics. In practice,
however, this is not quite true, first of all because the direct use of Eq. (1.9)
frequently leads to complex calculations. Indeed, let us try to solve a very simple
problem: finding the electric field produced by a spherically symmetric charge
distribution with density ρ(r′). We may immediately use the problem symmetry to
argue that the electric field should also be spherically symmetric, with only one
component in the spherical coordinates: E(r) = E(r)nr, where nr ≡ r/r is the unit
vector in the direction of the field observation point r (figure 1.2).
Figure 1.2. One of the simplest problems of electrostatics: an electric field produced by a spherically symmetric
charge distribution.
Taking this direction as the polar axis of a spherical coordinate system, we can
use the evident independence of the elementary radial field dE, created by the
elementary charge ρ(r′)d3r′ = ρ(r′)r′2sinθ dr′dθ′dϕ′, of the azimuthal angle ϕ′, and
reduce the integral (1.9) to
π ∞
1 ρ(r′)
E=
4πε0
2π ∫0 ∫0
sin θ ′dθ ′ r′2 dr′
R2
cos θ , (1.11)
where θ and R are the geometrical parameters marked in figure 1.2. Since they may
all be readily expressed via r′ and θ′, using the auxiliary parameters a and h,
9
See, e.g. appendix A, section A.14.
1-5
r−a
cos θ = , R2 = h 2 + (r − r′ cos θ )2 ,
R (1.12)
where a ≡ r′ cos θ′ , h ≡ r′ sin θ′ ,
Eq. (1.11) may be eventually reduced to an explicit integral over r′ and θ′, and
worked out analytically, but that would require some effort.
For other problems, the integral (1.9) may be much more complicated, defying
analytical solution. One could argue that with the present-day abundance of
computers and numerical algorithm libraries, one can always resort to numerical
integration. This argument may be enhanced by the fact that numerical integration is
based on the replacement of the integral by a discrete sum, and summation is much
more robust to the (unavoidable) discretization and rounding errors than the finite-
difference schemes typical in the numerical solution of differential equations. These
arguments, however, are only partly justified, since in many cases the numerical
approach runs into a problem sometimes called the curse of dimensionality—the
exponential dependence of the number of needed calculations on the number of
independent parameters of the problem10. Thus, despite the proliferation of
numerical methods in physics, analytical results have an ever-lasting value, and
we should try to get them whenever we can. For our current problem of finding the
electric field generated by a fixed set of electric charges, large help may come from
the so-called Gauss law.
Let us consider a single point charge q inside a smooth, closed surface S (figure 1.3),
and calculate product End2r, where d2r is an elementary area of the surface (which may
be well approximated with a plane fragment of that area), and En is the component of
the electric field in that point, normal to that plane.
Figure 1.3. Deriving the Gauss law: a point charge q is (a) inside volume V and (b) outside of that volume.
This component may be calculated as E cos θ, where θ is the angle between the
vector E and the unit vector n normal to the surface. (Equivalently, En may be
10
For a more detailed discussion of this problem, see, e.g. Part CM section 5.8.
1-6
represented as the scalar product E·n.) Now let us notice that the product cosθ d2r is
nothing more than the area d2r′ of the projection of d2r onto the plane perpendicular
to vector r connecting charge q with this point of the surface (figure 1.3), because the
angle between the planes d2r′ and d2r is also equal to θ. Using the Coulomb law for
E, we obtain
1 q 2
End 2r = E cos θ d 2r = d r′ . (1.13)
4πε0 r 2
But the ratio d2r′/r2 is nothing more than the elementary solid angle dΩ under
which the areas d2r′ and d2r are seen from the charge point, so that End2r may be
represented as just a product of dΩ by a constant (q/4πε0). Summing these products
over the whole surface, we obtain
q q
∮S End 2r = 4πε ∮
0 S
dΩ ≡ ,
ε0
(1.14)
since the full solid angle equals 4π. (The integral in the left-hand side of this relation
is called the flux of electric field through surface S.)
The relation (1.14) expresses the Gauss law for one point charge. However, it is
only valid if the charge is located inside the volume limited by the surface. In order to
find the flux created by a charge outside the volume, we still can use Eq. (1.13), but
have to be careful with the signs of the elementary contributions EndA. Let us use the
common convention to direct the unit vector n out of the closed volume we are
considering (the so-called outer normal), so that the elementary product End2r =
(E · n)d2r and hence dΩ = End2r′/r2 is positive if the vector E is pointing out of the
volume (as in the example shown in figure 1.3a and the upper-right area in figure
1.3b), and negative in the opposite case (for example, in the lower-left area in
figure 1.3b). As the latter figure shows, if the charge is located outside the
volume, for each positive contribution dΩ there is always an equal and opposite
contribution to the integral. As a result, at the integration over the solid angle the
positive and negative contributions cancel exactly, so that
∮S End 2r = 0. (1.15)
The real power of the Gauss law is revealed by its generalization to the case of
many charges within volume V. Since the calculation of flux is a linear operation, the
linear superposition principle (1.4) means that the flux created by several charges is
equal to the (algebraic) sum of individual fluxes from each charge, for which either
Eq. (1.14) or Eq. (1.15) is valid, depending on the charge position (in or out of the
volume). As the result, for the total flux we obtain:
QV 1 1
∮S End 2r = ε0
≡ ∑ qj ≡
ε0 r ∈ V ε0
∫V ρ(r′)d 3r′, (1.16)
j
where QV is the net charge inside volume V. This is the full version of the Gauss law.
1-7
In order to appreciate the problem-solving power of the law, let us return to the
problem illustrated by figure 1.2, i.e. a spherical charge distribution. Due to its
symmetry, which has already been discussed above, if we apply Eq. (1.16) to a
sphere of radius r, the electric field should be perpendicular to the sphere at each of
its points (i.e. En = E), and its magnitude the same at all points: En = E = E(r). As a
result, the flux calculation is elementary:
∮ End 2r = 4πr 2E (r). (1.17)
Now, applying the Gauss law (1.16), we obtain

r
1
4πr 2E (r ) =
ε0
∫r′<r ρ(r′)d 3r′ = 4επ0 ∫0 r′2 ρ(r′)dr′ , (1.18)
so that, finally,
r
1 1 Q (r )
E (r ) = 2
r ε0
∫0 r′2 ρ(r′)dr′ =
4πε0 r 2
, (1.19)
where Q(r) is the full charge inside the sphere of radius r:

r
Q ( r ) ≡ 4π ∫0 ρ(r′)r′2 dr′ . (1.20)
In particular, this formula shows that the field outside a sphere of a finite radius R
is exactly the same as if all its charge Q = Q(R) was concentrated in the sphere’s
center. (Note that this important result is only valid for a spherically symmetric
charge distribution.) For the field inside the sphere, finding the electric field still
requires an explicit integration (1.20), but this 1D integral is much simpler than the
2D integral (1.11), and in some important cases may be readily worked out
analytically. For example, if the charge Q is uniformly distributed inside a sphere
of radius R,
Q Q
ρ(r′) = ρ = = , (1.21)
V (4π /3)R3
the integration is elementary:
r
ρ ρr 1 Qr
E (r ) = 2
r ε0
∫0 r′2 dr′ =
3ε0
=
4πε0 R3
. (1.22)
We see that in this case the field is growing linearly from the center to the sphere’s
surface, and only at r > R starts to decrease in agreement with Eq. (1.19) with
constant Q(r) = Q. Another important observation is that the results for r ⩽ R and
r ⩾ R give the same value (Q/4πε0R2) at the charged sphere’s surface, r = R, so that
the electric field is continuous.
In order to underline the importance of the last fact, let us consider one more
elementary but very important example of the application of the Gauss law. Let a
thin plane sheet (figure 1.4) be charged uniformly, with an areal density σ = const. In
1-8
Figure 1.4. The electric field of a charged plane.
this case, it is fruitful to use the Gauss volume in the form of a planar ‘pillbox’ of
thickness 2z (where z is the Cartesian coordinate perpendicular to the charged plane)
and certain area A—see the dashed lines in figure 1.4. Due to the symmetry of the
problem, it is evident that the electric field should be: (i) directed along axis z,
(ii) constant on each of the upper and bottom sides of the pillbox, (iii) equal and
opposite on these sides, and (iv) parallel to the side surfaces of the box. As a result,
the full electric field flux through the pillbox’s surface is just 2AE(z), so that the
Gauss law (1.16) yields 2AE(z) = QA/ε0 ≡ σA/ε0, and we get a very simple but
important formula
σ
E (z ) = = const. (1.23)
2ε 0
Notice that, somewhat counter-intuitively, the field magnitude does not depend
on the distance from the charged plane. (From the point of view of the Coulomb
law (1.5), this result may be explained as follows: the farther the observation point
from the plane, the weaker the effect of each elementary charge, dQ = σd2r, but the
more such elementary charges give contributions to the vertical component of
vector E.)
Note also that though the magnitude E ≡ ∣E∣ of the electric field is constant, its
component En normal to the plane (for our coordinate choice, Ez) changes its sign at
the plane, experiencing a discontinuity (jump) equal to
σ
ΔEn = . (1.24)
ε0
This jump disappears if the surface is not charged (σ = 0). Returning for a minute to
our charged sphere problem, very close to its surface it may be considered planar, so
that the electric field should indeed be continuous, as it is.
Admittedly, the integral form (1.16) of the Gauss law is immediately useful
only for highly symmetrical geometries, such as in the two problems discussed
above. However, it may be recast into an alternative, differential form whose field
of useful applications is much wider. This form may be obtained from Eq. (1.16)
using the divergence theorem that, according to the vector algebra, is valid for any
1-9
space-differentiable vector, in particular E, and for volume V limited by any

closed surface S11:
∮S End 2r = ∫V (∇ ⋅ E)d 3r, (1.25)
where ∇ is the del (or ‘nabla’) operator of spatial differentiation12. Combining

Eq. (1.25) with the Gauss law (1.16), we obtain
⎛ ⎞
∫V ⎜⎝∇ ⋅ E − ερ0 ⎟d
⎠
3
r = 0. (1.26)
For a given distribution of electric charge (and hence of the electric field), this
equation should be valid for any choice of volume V. This can hold only if the
function under the integral vanishes at each point, i.e. if13
ρ
∇⋅E= . (1.27)
ε0
Note that in a sharp contrast to the integral form (1.16), Eq. (1.27) is local: it relates
the electric field divergence to the charge density at the same point. This equation,
being the differential form of the Gauss law, is frequently called one of the Maxwell
equations.
Another, homogeneous Maxwell equation’s ‘embryo’ (valid for the stationary
case only!) may be obtained by noticing that the curl of the point charge’s field, and
hence that of any system of charges, equals zero14:
∇ × E = 0. (1.28)
(We will arrive at two other Maxwell equations, for the magnetic field, in chapter 5,
and then generalize all the equations to their full, time-dependent form by the end of
chapter 6. However, Eq. (1.27) would stay the same.)
Just to obtain a better gut feeling of Eq. (1.27), let us apply it to the same example
of a uniformly charged sphere (figure 1.2). The vector algebra tells us that the
divergence of a spherically symmetric vector function E(r) = E(r)nr may be simply
expressed in spherical coordinates15: ∇ · E = [d(r2E)/dr]/r2. As a result, Eq. (1.27)
yields a linear, ordinary differential equation for the scalar function E(r):
11
See, e.g. Eq. (A.79). Note that the scalar product under the integral in Eq. (1.25) is nothing more that the
divergence of vector E—see, e.g. Eq. (A.53).
12
See, e.g. appendix A, sections A.8–A.10.
13
In Gaussian units, just as in the initial Eq. (1.6), ε0 has to be replaced with 1/4π, so that the Maxwell
equation (1.27) looks like ∇ · E = 4πρ, while Eq. (1.28) stays the same.
14
This follows, for example, from the direct application of Eq. (A.69) to any spherically symmetric vector
function of the type f(r) = f(r)nr (in particular, to the electric field of a point charge placed at the origin), giving
fθ = fϕ = 0 and ∂fr/∂θ = ∂fr/∂ϕ = 0, so that all components of the vector ∇ × f vanish. Since nothing prevents us
from placing the reference frame’s origin at the point charge’s location, this result remains valid for any
position of the charge.
15
See, e.g. Eq. (A.68) for this particular case (when ∂/∂θ = ∂/∂ϕ = 0).
1-10
1 d 2 ⎧ ρ / ε0 , for r ⩽ R ,
( r E ) = ⎨ (1.29)
r 2 dr ⎩ 0, for r ⩾ R ,
which may be readily integrated on each of the segments:

⎧
1 1 ⎪ ρ r 2dr = ρr 3/3 + c1,
∫ for r ⩽ R ,
E (r ) = × ⎨ (1.30)
ε0 r 2 ⎪
⎩c2 , for r ⩾ R .
In order to determine the integration constant c1, we can use the following boundary
condition: E(0) = 0. (It follows from the problem’s spherical symmetry: in the center
of the sphere, the electric field has to vanish, because otherwise, where would it be
directed?) This requirement gives c1 = 0. The second constant, c2, may be found from
the continuity condition E(R − 0) = E(R + 0), which has already been discussed
above, giving c2 = ρR3/3 ≡ Q/4π. As a result, we arrive at our previous results (1.19)
and (1.22).
We can see that in this particular, highly symmetric case, using the differential
form of the Gauss law is more complex than its integral form. (For our second
example, shown in figure 1.4, it would be even less natural.) However, Eq. (1.27) and
its generalizations are more convenient for asymmetric charge distributions,
and invaluable in the cases where the charge distribution ρ(r) is not known a priori
and has to be found in a self-consistent way. (We will start discussing such cases in
the next chapter.)
1.3 Scalar potential and electric field energy

One more help for solving electrostatics (and more complex) problems may be
obtained from the notion of the electrostatic potential, which is just the electrostatic
potential energy U of a probe point charge placed into the field in question,
normalized by its charge:
U
ϕ≡ . (1.31)
q
As we know from classical mechanics16, the notion of U (and hence ϕ) makes most
sense for the case of potential forces, for example those depending just on the
particle’s position. Eqs. (1.6) and (1.9) show that, in static situations, the electric field
clearly falls into this category. For such a field, the potential energy may be defined
as a scalar function U(r) that allows the force to be calculated as its gradient (with
the opposite sign):
F = −∇U . (1.32)
16
See, e.g. Part CM section 1.4.
1-11
Dividing both sides of this equation by the probe charge, and using Eqs. (1.6) and
(1.31), we obtain17
E = −∇ϕ . (1.33)
In order to calculate the scalar potential, let us start from the simplest case of a
single point charge q placed at the origin. For it, Eq. (1.7) takes a simple form
1 r 1 nr
E= q = q . (1.34)
4πε0 r 3 4πε0 r 2
It is straightforward to verify that the last fraction in the right-hand side of
Eq. (1.34) is equal to −∇(1/r).18 Hence, according to the definition (1.33), for this
particular case
1 q
ϕ= . (1.35)
4πε0 r
(In the Gaussian units, this result is spectacularly simple: ϕ = q/r.) Note that we
could add an arbitrary constant to this potential (and indeed to any other
distribution of ϕ discussed below) without changing the force, but it is convenient
to define the potential energy to approach zero at infinity.
In order to justify the further exploration of U and ϕ, let me demonstrate (perhaps,
unnecessarily) how useful the notions are, on a very simple example. Let two similar
charges q be launched from afar, with the same initial speed v0 ≪ c each, straight toward
each other (i.e. with the zero impact parameter)—see figure 1.5. Since, according to the
Coulomb law, the charges repel each other with increasing force, they will stop at some
minimum distance rmin from each other, and then fly back. We could of course find rmin
directly from the Coulomb law. However, for that we would need to write Newton’s
second law for each particle (actually, due to the problem symmetry, they would be
similar), then integrate them over time to find the particle velocity v as a function of
distance, and then recover rmin from the requirement v = 0.
The notion of potential allows this problem to be solved in one line. Indeed, in the
field of potential forces the system’s total energy E = T + U ≡ T + qϕ is conserved.
In our non-relativistic case v0 ≪ c, the kinetic energy T is just mv2/2. Hence, equating
the total energy of two particles in the points r = ∞ and r = rmin, and using Eq. (1.35)
for ϕ, we obtain
Figure 1.5. A simple problem of charged particle motion.
17
Eq. (1.28) could also be derived from this relation, because according to vector algebra, any gradient field
has vanishing curl—see, e.g. Eq. (A.71).
18
This may be done either by Cartesian components or using the well-known expression ∇f = (df/dr)nr valid
for any spherically symmetric scalar function f(r)—see, e.g. Eq. (A.66) for the particular case ∂/∂θ = ∂/∂ϕ = 0.
1-12
mv02 1 q2
2 +0=0+ , (1.36)
2 4πε0 rmin
immediately giving us the final answer: rmin = q2/4πε0mv02. So, the notion of the
scalar potential is indeed very useful.
With this motivation, let us calculate ϕ for an arbitrary configuration of charges.
For a single charge in an arbitrary position (say, rk′), r in Eq. (1.35) should be
evidently replaced for ∣r − rk′∣. Now, the linear superposition principle (1.3) allows
for an easy generalization of this formula to the case of an arbitrary set of discrete
charges,
1 q
ϕ(r) = ∑ k′ .
4πε0 r ≠ r r − rk ′
(1.37)
k′
Finally, using the same arguments as in section 1.1, we can use this result to argue
that in the case of an arbitrary continuous charge distribution
1 ρ(r′) 3
ϕ(r) =
4πε0
∫ r − r′
d r′ . (1.38)
Again, the Dirac’s delta-function allows the use of the last equation to recover
Eq. (1.37) for discrete charges as well, so that Eq. (1.38) may be considered the
general expression for the electrostatic potential.
For most practical calculations, using this expression and then applying
Eq. (1.33) to the result, is preferable to using Eq. (1.9), because ϕ is a scalar, while
E is a 3D vector, mathematically equivalent to 3 scalars. Still, this approach still may
lead to technical problems similar to those discussed in section 1.2. For example,
applying it to the spherically symmetric distribution of charge (figure 1.2), we obtain
the integral
π ∞
1 ρ(r′)
ϕ=
4πε0
2π ∫0 sin θ′dθ′ ∫0 r′2 dr′
R
cos θ , (1.39)
which is not much simpler than Eq. (1.11).

The situation may be much improved by re-casting Eq. (1.38) into a differential
form. For that, it is sufficient to plug the definition of ϕ, Eq. (1.33), into
Eq. (1.27):
ρ
∇ ⋅ ( −∇ϕ) = . (1.40)
ε0
1-13
The left-hand side of this equation is nothing more than the Laplace operator of ϕ
(with the minus sign), so that we obtain the famous Poisson equation19 for the
electrostatic potential:
ρ
∇2 ϕ = − . (1.41)
ε0
(In Gaussian units, the Poisson equation is ∇2ϕ = −4πρ.)20 This differential equation
is so convenient for applications that even its particular case for ρ = 0,
∇2 ϕ = 0, (1.42)
has earned a special name—the Laplace equation21.

In order to obtain a feeling of the Poisson equation’s value as a problem-solving
tool, let us return to the spherically symmetric charge distribution (figure 1.2) with a
constant charge density ρ. Using the symmetry, we can represent the potential as
ϕ(r) = ϕ(r), and hence use the following simple expression for its Laplace operator22:
1 d ⎛ 2 dϕ ⎞
∇2 ϕ = ⎜r ⎟, (1.43)
r 2 dr ⎝ dr ⎠
so that for the points inside the charged sphere (r ⩽ R) the Poisson equation yields
1 d ⎛ 2 dϕ ⎞ ρ d ⎛ 2 dϕ ⎞ ρ
⎜r ⎟=− , i.e. ⎜r ⎟ = − r 2. (1.44)
r dr ⎝ dr ⎠
2
ε0 dr ⎝ dr ⎠ ε0
Integrating the last form of the equation over r once, with the natural boundary
condition dϕ/dr∣r=0 = 0 (because of the condition E(0) = 0, which has been discussed
above), we obtain
r
dϕ ρ ρr 1 Qr
dr
(r ) = − 2
r ε0
∫0 r′2 dr′ = −
3ε 0
=−
4πε0 R3
. (1.45)
Since this derivative is nothing more than −E(r), in this formula we can readily
recognize our previous result (1.22). Now we may like to carry out the second
integration to calculate the potential itself:
r
Q Qr 2
ϕ(r ) = −
4πε0R3
∫0 r′dr′ + c1 = −
8πε0R3
+ c1. (1.46)
19
Named after S D Poisson (1781–1840), also famous for the Poisson distribution—one of the central results of
probability theory—see, e.g. Part SM section 5.2.
20
As a sanity check, a point charge q at the origin, according to Eq. (1.10), ρ(r) = qδ(r), and with the well-
known (or readily verifiable) identity ∇2(1/r) = −4πδ(r), Eq. (1.41) immediately yields Eq. (1.35).
21
After mathematician (and astronomer) P S de Laplace (1749–1827) who, together with A Clairault, is
credited for the development of the very concept of potential.
22
See, e.g. Eq. (A.67) for ∂/∂θ = ∂/∂ϕ = 0.
1-14
Before making any judgment on the integration constant c1, let us solve the
Poisson equation (in this case, just the Laplace equation) for the range outside the
sphere (r > R):
1 d ⎛ 2 dϕ ⎞
⎜r ⎟ = 0. (1.47)
r 2 dr ⎝ dr ⎠
Its first integral,
dϕ c
(r ) = 22 , (1.48)
dr r
also gives the electric field (with the minus sign). Now using Eq. (1.45) and requiring
the field to be continuous at r = R, we obtain
c2 Q dϕ Q
2
=− , i.e. (r ) = − , (1.49)
R 4πε0R2 dr 4πε0r 2
in an evident agreement with Eq. (1.19). Integrating this result again,

Q
ϕ(r ) = −
4πε0
∫ dr
r 2
=
Q
4πε0r
+ c3, for r > R , (1.50)
we can select c3 = 0, so that ϕ(∞) = 0, in accordance with the usual (though not
compulsory) convention. Now we can finally determine the constant c1 in Eq. (1.46)
by requiring that this equation and Eq. (1.50) give the same value of ϕ at the
boundary r = R. (According to Eq. (1.33), if the potential had a jump, the electric
field at that point would be infinite.) The final answer may be represented as
Q ⎛ R2 − r 2 ⎞
ϕ(r ) = ⎜ + 1⎟, for r ⩽ R . (1.51)
4πε0R ⎝ 2R2 ⎠
We see that using the Poisson equation to find the electrostatic potential distribution
for highly symmetric problems may be more cumbersome than directly finding the
electric field—say, from the Gauss law. However, we will repeatedly see below that if
the electric charge distribution is not fixed in advance, using Eq. (1.41) may be the
only practicable way to proceed.
Returning now to the general theory of electrostatic phenomena, let us calculate the
potential energy U of an arbitrary system of point electric charges qk. Despite the
apparently straightforward relation (1.31) between U and ϕ, the result is not that
straightforward. Indeed, let us assume that the charge distribution has a finite spatial
extent, so that at large distances from it (formally, at r = ∞) the electric field tends to
zero, so that the electrostatic potential tends to a constant. Selecting this constant, for
convenience, to equal zero, we may calculate U as a sum of the energy increments ΔUk
1-15
Figure 1.6. Deriving Eqs. (1.55) and (1.60) for the potential energies of a system of electric charges.
created by bringing the charges, one by one, from infinity to their final positions rk—
see figure 1.623. According to the integral form of Eq. (1.32), such a contribution is
rk rk
ΔUk = − ∫∞ F(r) ⋅ d r = −qk ∫∞ E(r) ⋅ d r ≡ qkϕ(rk ), (1.52)
where E(r) is the total electric field, and ϕ(r) is the total electrostatic potential during
this process, besides the field created by the very charge qk that is being moved.
This expression shows that the increment ΔUk, and hence the total potential
energy U, depend on the source of the electric field E. If the field is dominated by an
external field Eext, induced by some external charges, not being a part of the charge
configuration under our analysis (whose energy we are calculating, see figure 1.6),
the spatial distribution ϕ(r) is determined by this field, i.e. does not depend on how
many charges we have already brought in, so that Eq. (1.52) is reduced to
r
ΔUk = qkϕext(rk ), where ϕext(r) ≡ − ∫∞ Eext(r′) ⋅ d r′. (1.53)
Summing up these contributions, we get what is called the charge system’s energy in
the external field24:
Uext ≡ ∑ΔUk = ∑qkϕext(rk ). (1.54)
k k
Now repeating the argumentation that has led us to Eq. (1.9), we see that for a
continuously distributed charge, this sum turns into an integral:
Uext = ∫ ρ(r)ϕext(r)d 3r. (1.55)
(As was discussed above, using the delta-functional representation of point charges,
we may always return from here to Eq. (1.54), so that Eq. (1.55) may be considered
as a final, universal result.)
23
Indeed, by the very definition of the potential energy of a system, it should not depend on the way we are
arriving at the its final configuration.
24
An alternative, perhaps more accurate term for Uext is the energy of the system’s interaction with the external
field.
1-16
The result is different in the opposite limit, when the electric field E(r) is created only
by the very charges whose energy we are calculating. In this case, ϕk(rk) in Eq. (1.52) is
the potential created only by the charges with numbers k′ = 1, 2, …, (k − 1) already in
place when the kth charge is moved in (in figure 1.6, the charges inside the dashed
boundary), and we may use the linear superposition principle to write
ΔUk = qk ∑ ϕk ′(rk ), so that U = ∑Uk = ∑ qkϕk ′(rk ).
k ′< k k k, k ′ (1.56)
(k ′< k )
This result is so important that it is worth rewriting it in several other forms. First,
we may use Eq. (1.35) to represent Eq. (1.56) in a more symmetric form:
1 qkqk ′
U=
4πε0
∑ rk − rk ′
.
(1.57)
k, k ′
(k ′< k )
The expression under this sum is evidently symmetric with respect to the index swap,
so that it may be extended into a different form,
1 1 qkqk ′
U=
4πε0 2
∑ rk − rk ′
,
(1.58)
k ′, k
(k ′≠ k )
where the interaction between each couple of charges is described by two, equal
terms under the sum, and the front coefficient ½ is used to compensate this double-
counting. The convenience of the last form is that it may be readily generalized to the
continuous case:
1 1
U=
4πε0 2
∫ d 3r∫ d 3r′ ρr(r−)ρ(rr′′) . (1.59)
(As before, in this case the restriction expressed in the discrete charge case as k ≠ k′ is
not important, because if the charge density is a continuous function, the integral
(1.59) does not diverge at point r = r′.)
To represent this result in one more form, let us notice that according to Eq.
(1.38), the inner integral over r′ in (1.59), divided by 4πε0, is just the full electrostatic
potential at point r, and hence
1
U=
2
∫ ρ(r)ϕ(r)d 3r. (1.60)
For the discrete charge case, this result becomes

1
U= ∑q ϕ(rk ),
2 k k
(1.61)
but now it is important to remember that here the ‘full’ potential’s value ϕ(rk) should
exclude the (infinite) contribution of the point charge k itself. Comparing the last
two formulas with Eqs. (1.54) and (1.55), we see that the electrostatic energy of
1-17
charge interaction within the system, as expressed via the charge-by-potential

product, is twice less than that of the energy of charge interaction with a fixed
(‘external’) field. This is evidently the result of the fact that in the case of mutual
interaction of the charges, the electric field E in the basic Eq. (1.52) is proportional to
the charge magnitude, rather than constant25.
Now we are ready to address an important conceptual question: can we locate
this interaction energy in space? Expressions (1.58)–(1.61) seem to imply that
contributions to U come only from the regions where the electric charges are
located. However, one of the most beautiful features of physics is that sometimes
completely different interpretations of the same mathematical result are possible. In
order to obtain an alternative view of our current result, let us write Eq. (1.60) for a
volume V so large that the electric field on the limiting surface S is negligible, and
plug into it the charge density expressed from the Poisson equation (1.41):
ε0
U=−
2
∫V ϕ ∇2 ϕd 3r. (1.62)
This expression may be integrated by parts as26

ε0 ⎡ ⎤
U=−
2⎣
⎢∮S ϕ (∇ϕ)nd 2r − ∫V (∇ϕ)2d 3r⎥⎦. (1.63)
According to our condition of negligible field E = −∇ϕ on the surface, the first
integral vanishes, and we obtain a very important formula
ε0
U=
2
∫ (∇ϕ)2d 3r = ε20 ∫ E 2d 3r. (1.64)
This result, represented in the following equivalent form27:

ε0 2
U= ∫ u(r)d 3r, with u(r) ≡
2
E (r), (1.65)
certainly invites an interpretation very much different than Eq. (1.60): it is natural to
interpret u(r) as the spatial density of the electric field energy, which is continuously
distributed over all the space where the field exists—rather than just its part where
the charges are located.
Let us have a look how these two alternative pictures work for our testbed
problem, a uniformly charged sphere. If we start from Eq. (1.60), we may limit the
integration by the sphere volume (0 ⩽ r ⩽ R) where ρ ≠ 0. Using Eq. (1.51), and the
spherical symmetry of the problem (d3r = 4πr2dr), we get
25
The nature of this additional factor ½ is absolutely the same as in the well-known formula U = (½)κx2 for the
potential energy of an elastic spring, providing the returning force F = −κx proportional to the deviation x
from equilibrium.
26
This transformation follows from the divergence theorem (A.79) applied to the vector function f = ϕ∇ϕ,
taking into account the differentiation rule (A.74a): ∇ · (ϕ ∇ϕ) = (∇ϕ) · (∇ϕ) + ϕ ∇ · (∇ϕ) = (∇ϕ)2 + ϕ∇2ϕ.
27
In Gaussian units, the standard replacement ε0 → 1/4π turns the last of Eqs. (1.65) into u(r) = E2/8π.
1-18
1 R
1 Q R ⎛ R2 − r 2 ⎞
U=
2
4π ∫0 ρϕ r 2dr =
2
4πρ
4πε0R
∫0 ⎜
⎝ 2R2
+ 1⎟ r 2dr
⎠
(1.66)
6 1 Q2
= .
5 4πε0R 2
On the other hand, if we use Eq. (1.65), we need to integrate the energy density
everywhere, i.e. both inside and outside of the sphere:
ε0 ⎛ R ∞ ⎞
U= 4π ⎜ ∫0 E 2r 2dr + ∫R E 2r 2dr⎟ . (1.67)
2 ⎝ ⎠
Using Eqs. (1.19) and (1.22) for, respectively, the external and internal regions, we
obtain
⎡R
⎛ Q ⎞2 2 ⎤⎥ ⎛ 1
∞
ε0 ⎢ ⎛ Qr ⎞ 2 ⎞ 1 Q2
2
U = 4π ⎜ ∫⎟ r dr + ∫ ⎜ ⎟ r dr = ⎜ + 1⎟ . (1.68)
2 ⎢⎣ ⎝ 4πε0 ⎠ ⎝ 4πε0r 2 ⎠ ⎥ ⎝5
⎦
⎠ 4πε0R 2
0 R
This is (fortunately:-) the same answer as given by Eq. (1.66), but to some extent it is
more informative, because it shows how exactly the electric field energy is distributed
between the interior and exterior of the charged sphere28.
We see that, as we could expect, within the realm of electrostatics, Eqs. (1.60) and
(1.65) are equivalent. However, when we examine electrodynamics (in chapter 6 and
beyond), we will see that the latter equation is more general, and that it is more
adequate to associate the electric energy with the field itself rather than its sources—
in our current case, the electric charges.
Finally, let us calculate the potential energy of a system of charges in the general
case when both the internal interaction of the charges, and their interaction with an
external field are important. One might fancy that such a calculation should be very
difficult, since in both ultimate limits, when one of these interactions dominates, we
have obtained different results. However, once again we get help from the almighty
linear superposition principle: in the general case, for the total electric field we may
write
E(r) = E int(r) + Eext(r), ϕ(r) = ϕint(r) + ϕext(r), (1.69)
where the index ‘int’ now marks the field induced by the charge system under
analysis, i.e. the variables participating (without indices) in Eqs. (1.56)–(1.68). Now
let us imagine that our system is being built up in the following way: first, the charges
are brought together at Eext = 0, giving the potential energy Uint expressed by
Eq. (1.60), and then Eext is slowly increased. Evidently, the energy contribution from
28
Note that U → ∞ at R → 0. Such divergence appears at application of Eq. (1.65) to any point charge. Since it
does not affect the force acting on the charge, the divergence does not create any technical difficulty for
analysis of charge statics or non-relativistic dynamics, but it points to a conceptual problem of classical
electrodynamics as the whole. This issue will be discussed in the very end of the course (section 10.6).
1-19
the latter process cannot depend on the internal interaction of the charges, and hence
may be expressed in the form (1.55). As the result, the total potential energy29 is the
sum of these two components:
1
U = Uint + Uext =
2
∫ ρ(r)ϕint(r)d 3r + ∫ ρ(r)ϕext(r)d 3r. (1.70)
Now making, in the first integral, the transition from the potentials to the fields,
absolutely similar to that performed in Eqs. (1.62)–(1.65), we may rewrite this
expression as
ε0 2
2
[E int(r) + 2E int(r) ⋅ Eext(r)]. (1.71)
One might think that this result, more general than Eq. (1.65) and perhaps less
familiar to the reader, is something entirely new; however, it is not. Indeed, let us
add and subtract Eext2(r) from the sum in the brackets, and use Eq. (1.69) for the
total electric field E(r); then Eq. (1.71) takes the form
ε0
U=
2
∫ E 2(r)d 3r − ε20 ∫ Eext2(r)d 3r. (1.72)
Hence, in the most important case when we are using the potential energy to analyze
the statics and dynamics of a system of charges in a fixed external field, i.e. when the
second term in Eq. (1.72) may be considered as a constant, we may still use for U an
expression similar to the familiar Eq. (1.65), but with the field E(r) being the sum
(1.69) of the internal and external fields.
Let us see how this works in a very simple problem. A uniform external electric
field Eext is applied normally to a very broad, planar layer that contains a very large
and equal number of free electric charges of both signs—see figure 1.7. What is the
equilibrium distribution of the charges over the layer?
Figure 1.7. A simple model of the electric field screening in a conductor.
Since the equilibrium distribution should minimize the total potential energy of
the system, Eq. (1.72) immediately gives the answer: the distribution should provide
E ≡ Eint + Eext = 0 inside the whole layer—the effect called the electric field
screening. The only way to ensure this equality is to have enough free charges of
opposite signs residing on the layer’s surfaces to induce a uniform field Eint = −Eext,
exactly compensating the external field at each point inside the layer—see figure 1.7.
29
This total U (or rather its part dependent on our system of charges) is sometimes called the Gibbs potential
energy of the system. (I will discuss this notion in detail in section 3.5.)
1-20
According to Eq. (1.24), the areal density of these surface charges should equal ±σ,
with σ = Eext/ε0. This is a rudimentary but reasonable model of the conductor’s
polarization—to be discussed in detail in the next chapter.
1.4 Problems
Problem 1.1. Calculate the electric field created by a thin, long, straight filament,
electrically charged with a constant linear density λ, using two approaches:
(i) directly from the Coulomb law, and
(ii) using the Gauss law.
Problem 1.2. Two thin, straight parallel filaments, separated by distance ρ, carry
equal and opposite uniformly distributed charges with linear density λ—see the
figure below. Calculate the electrostatic force (per unit length) of the Coulomb
interaction between the wires. Compare the result with the Coulomb law for the
force between the point charges.
Problem 1.3. A sphere of radius R, whose volume has been charged with a constant
density ρ, is split with a very narrow, planar gap passing through its center.
Calculate the force of mutual repulsion of the resulting two hemispheres.
Problem 1.4. A thin spherical shell of radius R, which has been charged with a
constant areal density σ, is split into two equal halves by a very narrow, planar cut
passing through the sphere’s center. Calculate the force of electrostatic repulsion
between the resulting hemispheric shells, and compare the result with that of the
previous problem.
Problem 1.5. Calculate the distribution of the electrostatic potential created by a

straight, thin filament of finite length 2l, charged with a constant linear density λ,
and explore the result in the limits of very small and very large distances from the
filament.
Problem 1.6. A thin plane sheet, perhaps of an irregular shape, carries electric charge
with a constant areal density σ.
(i) Express the electric field’s component normal to the plane, at a certain distance from
it, via the solid angle Ω at which the sheet is visible from the observation point.
(ii) Use the result to calculate the field in the center of a cube, with one face charged
with constant density σ.
1-21
Problem 1.7. Can one create, in a non-vanishing region of space, electrostatic fields
with the Cartesian components proportional to the following products of Cartesian
coordinates {x, y, z},
(i) {yz, xz, xy},
(ii) {xy, xy, yz}?
Problem 1.8. Distant sources have been used to create different electric fields on two
sides of a wide and thin metallic membrane, with a round hole of radius R in it—see the
figure below. Besides the local perturbation created by the hole, the fields are uniform:
⎧E , at z < 0,
E(r) r ≫R = nz × ⎨ 1
⎩ E2, at z > 0.
Prove that the system may serve as an electrostatic lens for charged particles
flying along axis z, at distances ρ ≪ R from it, and calculate the focal distance f of
the lens. Spell out the conditions of validity of your result.
Problem 1.9. Eight equal point charges q are located at the corners of a cube of side
a. Calculate all Cartesian components Ej of the electric field, and their spatial
derivatives ∂Ej/∂rj′, at the cube’s center, where rj are the Cartesian coordinates
oriented along the cube’s sides—see the figure below. Are all your results valid for
the center of a planar square, with four equal charges in its corners?
Problem 1.10. By a direct calculation, find the average electric potential of the spherical
surface of radius R, created by a point charge q located at a distance r > R from the
sphere’s center. Use the result to prove the following general mean value theorem: the
electric potential at any point is always equal to its average value on any spherical
surface with the center at that point, and containing no electric charges inside it.
Problem 1.11. Two similar thin, circular, coaxial disks of radius R, separated by
distance 2d, are uniformly charged with equal and opposite areal densities ±σ—see
1-22
the figure below. Calculate and sketch the distribution of the electrostatic potential
and the electric field of the disks along their common axis.
Problem 1.12. In a certain reference frame, the electrostatic potential created by

some electric charge distribution, is
⎛1 1 ⎞ ⎧ r⎫
ϕ(r) = C ⎜ + ⎟ exp ⎨ − ⎬ ,
⎝r 2r0 ⎠ ⎩ r0 ⎭
where C and r0 are constants, and r ≡ ∣r∣ is the distance from the origin. Calculate the
charge distribution in space.
Problem 1.13. A thin flat sheet, cut in a form of a rectangle of size a × b, is

electrically charged with a constant areal density σ. Without an explicit calculation
of the spatial distribution ϕ(r) of the electrostatic potential induced by this charge,
find the ratio of its values at the center and at the corners of the rectangle.
Hint: Consider partitioning the rectangle into several similar parts and using the
linear superposition principle.
Problem 1.14. Calculate the electrostatic energy per unit area of the system of two
thin, parallel planes with equal and opposite charges of a constant areal density σ,
separated by distance d.
Problem 1.15. The system analyzed in the previous problem (two thin, parallel,
oppositely charged planes) is now placed into an external, uniform, normal electric
field Eext = σ/ε0—see the figure below. Find the force (per unit area) acting on each
plane, by two methods:
(i) directly from the electric field distribution, and
(ii) from the potential energy of the system.
Problem 1.16. Explore the relation between the Laplace equation (1.42) and the
condition of minimum of the electrostatic field energy (1.65).
1-23
Problem 1.17. Prove the following reciprocity theorem of electrostatics30: if two

spatially confined charge distributions ρ1(r) and ρ2(r) create respective distributions
ϕ1(r) and ϕ2(r) of the electrostatic potential, then
∫ ρ1(r) ϕ2(r)d 3r = ∫ ρ2 (r) ϕ1(r)d 3r.

Hint: Consider integral ∫ E1 ⋅ E2d 3r.
Problem 1.18. Calculate the energy of electrostatic interaction of two spheres, of
radii R1 and R2, each with a spherically symmetric charge distribution, separated by
distance d > R1 + R2.
Problem 1.19. Calculate the electrostatic energy U of a (generally, thick) spherical

shell, with a charge Q uniformly distributed through its volume—see the figure
below. Analyze and interpret the dependence of U on the inner cavity’s radius R1, at
fixed Q and R2.
Reference
[1] Griffiths D 2015 Introduction to Electrodynamics 4th edn (Pearson)
30
This is only the simplest of several reciprocity theorems in electromagnetism—see, e.g. section 6.8.
1-24
IOP Publishing
Lecture notes
Chapter 2
Charges and conductors
This chapter will start a discussion of the very common situations where the electric
charge distribution in space is not known a priori, but rather should be calculated in a
self-consistent way together with the electric field it creates. The simplest situations of
this kind involve conductors, and lead to the so-called boundary problems, in which
partial differential equations describing the field distribution have to be solved with
appropriate boundary conditions. Such problems are also broadly used in other areas
of electrodynamics (and indeed in other fields of physics as well), so that following
tradition, I will use this chapter’s material as a playground for the discussion of various
methods of boundary problem solution, and the special functions most frequently
encountered in the process.
2.1 Polarization and screening

The basic principles of electrostatics outlined in chapter 1 present the conceptually
full solution for the problem of finding the electrostatic field (and hence Coulomb
forces) induced by electric charges distributed over space with density ρ(r). However,
in most practical situations this function is not known but should be found self-
consistently with the field. For example, if a volume of relatively dense material is
placed into an external electric field, it is typically polarized, i.e. acquires some local
charges of their own, which contribute to the total electric field E(r) inside, and even
outside it—see figure 2.1a.
The full solution of such problems should satisfy not only the fundamental
Eq. (1.7), but also the so-called constitutive relations between various macroscopic
variables describing the body’s material. Due to the atomic character of real
materials, such relations may be very involved. In this part of my series, I will
have time to address these relations, for various materials, only very superficially1,
1
More detailed discussions may be found, e.g. in section 13.5 of [1], or the section on electric field screening in
chapter 17 of [2].

Figure 2.1. Schematic representations of two typical electrostatic situations involving conductors: (a) the
polarization by an external field and (b) the conductor’s own charge redistribution over its surface. Here and
below, the red and blue points are used to show charges of opposite signs.
focusing on their simple approximations. Fortunately, in most practical cases such

approximations work very well.
In particular, for the polarization of good conductors, a very reasonable
approximation is given by the so-called macroscopic model, in which the free
charges in the conductor are treated as a charged continuum, which is free to
move under the effect of the average force F = qE exerted by the averaged,
macroscopic electric field E—see the discussion of this notion at the end of section
1.1. In electrostatics (which excludes the case of dc currents, to be discussed in
chapter 4), there should be no such motion, so that everywhere inside the conductor
the macroscopic electric field should vanish:
E = 0. (2.1a )
This is the electric field screening2 effect, meaning, in particular, that the conductor’s
polarization in an external electric field has the extreme form shown (schematically)
in figure 2.1a, with the field of the induced surface charges completely compensating
the external field in the conductor’s bulk. Eq. (2.1a) may be rewritten in another,
frequently more convenient form:
ϕ = const, (2.1b)
where ϕ is the macroscopic electrostatic potential, related to the macroscopic electric
field by Eq. (1.33). Note, however, that if a problem includes several unconnected
conductors, the constant in Eq. (2.1b) may be different for each of them.
Now let us examine what we can say about the electric field just outside a
conductor, within the same macroscopic model. At a close proximity, any smooth
surface (in our current case, that of a conductor) looks planar. Let us integrate
Eq. (1.28) over a narrow (d ≪ l ) rectangular loop C encircling a part of such plane
conductor’s surface (see the dashed line in figure 2.2a), and apply to the electric field
vector E the well-known vector algebra equality, the Stokes theorem3
2
This term, used for the electric field, should not be confused with shielding—the word used for the description
of the magnetic field reduction by magnetic materials—see chapter 5.
3
See, e.g. Eq. (A.78).
2-2
Figure 2.2. (a) The surface charge layer at a conductor’s surface, and (b) the electric field lines and
equipotential surfaces near it.
∮S (∇ × E)nd 2r = ∮C E ⋅ d r, (2.2)
where S is any surface limited by the contour C. In our case the contour is
dominated by two straight lines of length l, so that if l is much smaller that the
characteristic scale of field change, but much larger that the inter-atomic distances,
the right-hand side of Eq. (2.2) may be approximated as [(Eτ)in − (Eτ)out]l, where
(Eτ) is the component of the corresponding macroscopic field, parallel to the surface.
On the other hand, according to Eq. (1.28), the left-hand side of Eq. (2.2) equals zero.
Hence, Eτ should be continuous at the surface, and in order to satisfy Eq. (2.1a) inside
the conductor, the component has to vanish immediately outside it: (Eτ)out = 0.
This means that the electrostatic potential immediately outside a conducting
surface does not change along it. In other words, the equipotential surfaces outside a
conductor should ‘lean’ to the conductor’s surface, with their potential values
approaching the constant potential—see figure 2.2b.
So, the electrostatic field just outside the conductor has be normal to its surface.
In order to find this normal field, we may use the universal relation (1.24). Since in
our current case En = 0 inside the conductor, we obtain
∂ϕ
σ = ε0(En)out = −ε0(∇ϕ)n ≡ −ε0 , (2.3)
∂n
where σ is the areal density of conductor’s surface charge. Note that in deriving this
universal relation between the normal component of the field and the surface charge
density, we have not used any cause-versus-effect arguments, so that Eq. (2.3) is
valid regardless of whether the surface charge is induced by an externally applied
field (the case of polarization of the conductor, shown in figure 2.1a), or the electric
field is induced by the electric charge placed on the conductor and then self-
redistributed over its surface (figure 2.1b), or it is some mixture of both effects.
Before starting to use the macroscopic model for the solution of particular
problems of electrostatics, let me use the balance of this section to briefly discuss its
limitations. (The reader in a rush may skip this discussion and proceed to section 2.2;
however, I believe that every educated physicist has to understand when the model
works, and when it does not.)
Since the argumentation which has led to Eq. (1.24) and hence Eq. (2.3) is valid
for any thickness d of the Gauss pillbox, within the macroscopic model, the whole
surface charge is located within an infinitely thin surface layer. This is of course
2-3
impossible physically: for one, this would require an infinite volume density ρ of the
charge. In reality the charged layer (and hence the region of electric field’s crossover
from the finite value (2.3) to zero) has a non-vanishing thickness λ. At least three
effects contribute to λ.
(i) Atomic structure of matter. Within each atom, and frequently between the
adjacent atoms as well, the genuine (‘microscopic’) electric field is highly non-
uniform. Thus Eq. (2.1) is valid only for the macroscopic field in a conductor (see the
discussion following Eq. (1.9) in section 1.1), averaged over distances much larger
than the atomic size scale a0 ∼ 10−10 m,4 and cannot be applied to the field changes
on that scale. As a result, the surface layer of charges cannot be much thinner
than a0.
(ii) Thermal excitation. In the conductor’s bulk, the number of protons of atomic
nuclei (n) and electrons (ne) and per unit volume are balanced, so that the
net charge density, ρ = e(n − ne), vanishes5. However, if an external electric
field penetrates a conductor, free electrons can shift in or out of its affected
part, depending on the field addition to their potential energy, ΔU = qeϕ = −eϕ.
(Here the arbitrary constant in ϕ is chosen to give ϕ = 0 well inside the
conductor.) In the classical statistics, this change is described by the Boltzmann
distribution6:
⎧ U (r) ⎫
n e(r) = n exp ⎨ − ⎬, (2.4)
⎩ kBT ⎭
where kB ≈ 1.38 × 10−23 J K−1 is the Boltzmann constant, and T is the absolute
temperature in SI units (kelvins). As a result, the net charge density is
⎛ ⎧ eϕ(r) ⎫⎞
ρ(r) = en⎜1 − exp ⎨ ⎬⎟ (2.5)
⎝ ⎩ kBT ⎭⎠
The penetrating electric field polarizes the atoms as well. As will be discussed in the
next chapter, such polarization results in the reduction of the electric field by a
material-specific dimensionless factor κ (larger, but typically not too much larger
than 1), called the dielectric constant. As a result, the Poisson equation (1.41) takes
the form7,
4
This scale originates from the quantum-mechanical effects of electron motion, characterized by the Bohr
radius rB ≡ ℏ2/me(e2/4πε0) ≈ 0.53 × 10−10 m—see, e.g. Part QM Eq. (1.13). It also defines the scale EB = e/
4πε0rB2 ∼ 1012 SI units (V m−1) of the microscopic electric fields inside the atoms. (Please note how large these
fields are.)
5
In this series, e denotes the fundamental charge, e ≈ 1.6 × 10−19 C > 0, so that the electron’s charge equals
(−e).
6
See, e.g. Part SM section 3.1.
7
This equation and/or its straightforward generalization to the case of charged particles (ions) of several kinds
is frequently (especially in the theories of electrolytes and plasmas) called the Debye–Hückel equation.
2-4
d 2ϕ ρ en ⎛ ⎧ eϕ ⎫ ⎞
= − = ⎜exp ⎨ ⎬ − 1 ⎟, (2.6)
dz 2 κε0 κε0 ⎝ ⎩ kBT ⎭ ⎠
where we have taken advantage of the 1D geometry of the system to simplify the
Laplace operator, with the axis z normal to the surface.
Even with this simplification, Eq. (2.6) is a nonlinear differential equation
allowing an analytical but rather bulky solution. Since our current goal is just to
estimate of the field penetration depth λ, let us simplify the equation further by
considering the low-field limit: e∣ϕ∣ ∼ e∣E∣λ ≪ kBT. In this limit we can extend the
exponent into the Taylor series, and keep only two leading terms (of which the first
one cancels with the unity). As a result, Eq. (2.6) becomes linear,
d 2ϕ en eϕ d 2ϕ 1
2
= , i.e. 2
= 2 ϕ, (2.7)
dz εε0 kBT dz λ
where the constant λ in this case is equal to the so-called Debye screening length λD,
defined by the relation
κε0kBT
λ D2 ≡ (2.8)
e 2n
As the reader certainly knows, Eq. (2.7) describes an exponential decrease of the
electric potential, with the characteristic length λD: ϕ ∝ exp{−z/λD}, where axis z is
directed inside the conductor. Plugging in the fundamental constants, we obtain the
following estimate: λD[m] ≈ 70 × (κ × T[K]/n[m−3])1/2. According to this formula, in
semiconductors at room temperature, the Debye length may be rather substantial.
For example, in silicon (κ ≈ 12) doped to the free charge carrier concentration n =
3 × 1024 m−3 (the value typical for modern integrated circuits)8, λD ≈ 2 nm, still well
above the atomic size scale a0. However, for typical good metals (n ∼ 1029 m−3, κ ∼
10) the same formula gives an estimate λD ∼ 10−11 m, less than a0. In this case
Eq. (2.8) should not be taken literally, because it is based on the assumption of a
continuous charge distribution.
(iii) Quantum statistics. Actually, the last estimate is not valid for good metals (and
highly doped semiconductors) for one more reason: their free electrons obey the
quantum (Fermi–Dirac) statistics rather that the Boltzmann distribution (2.4)9. As a
result, at all realistic temperatures the electrons form a degenerate quantum gas,
8
There is a good reason for making an estimate of λD for this case: the electric field created by the gate
electrode of a field-effect transistor, penetrating into doped silicon by a depth ∼λD, controls the electric current
in this most important electronic device—on whose back all the current information technology rides. Because
of that, λD establishes the possible scale of semiconductor circuit shrinking, which is the basis of the well-
known Moore’s law. (Practically, the scale is determined by integrated circuit patterning techniques, and
Eq. (2.8) may be used to find the proper charge carrier density n and hence the necessary level of silicon
doping.)
9
See, e.g. Part SM section 2.8. For a more detailed derivation of Eq. (2.10), see Part SM chapter 3.
2-5
occupying all available energy states below certain energy level EF ≫ kBT, called the
Fermi energy. In these conditions, the screening of relatively low electric field may be
described by replacing Eq. (2.5) with
ρ = e(n − n e ) = −eg(EF )( −U ) = −e 2g(EF )ϕ , (2.9)
where g(E ) is the density of quantum states (per unit volume) at electron’s energy E .
At the Fermi surface, the density is of the order of n/EF .10 As a result, we again
obtain the second of Eqs. (2.7), but with a different characteristic scale λ, defined by
the following relation:
κε0 κε E
2
λ TF ≡ 2
∼ 02 F , (2.10)
e g(EF ) e n
and called the Thomas–Fermi screening length. Since for most good metals, n is of
the order of 1029 m−3, and EF is of the order of 10 eV, Eq. (2.10) typically gives λTF
close to a few a0, and makes the Thomas–Fermi screening theory valid at least semi-
quantitatively.
To summarize, the electric field penetration into good conductors is limited to a

depth λ ranging from fractions of a nanometer to a few nanometers, so that for
problems with the characteristic size much larger than that scale, the macroscopic
model (2.1) gives a very good accuracy, and we will use them in the rest of this
chapter. However, the reader should remember that in many situations involving
semiconductors, as well as at some nanoscale experiments with metals, the electric
field penetration effect should be taken into account.
Another important condition of the macroscopic model’s validity is imposed on the
electric field’s magnitude, which is especially significant for semiconductors. Indeed, as
Eq. (2.6) shows, Eq. (2.7) is only valid if eϕ ≪ kBT, so that E ∼ ϕ/λD should be much
lower than kBT/eλD. In the example given above (λD ≈ 2 nm, T = 300 K), this means
E ≪ Et ∼107 V m−1 ≡ 105 V cm−1—a value readily reachable in the lab. At larger
fields, the field penetration becomes nonlinear, leading to the important effect of
carrier depletion; it will be discussed in Part SM chapter 6. For typical metals, such
linearity limit, Et ∼ EF/eλTF is much higher, ∼1011 V m−1, but the model may be
violated at lower fields (also ∼107 V m−1) by other effects, such as the impact-
ionization, leading to electric breakdown.
2.2 Capacitance
Let us start with the systems consisting of charged conductors only, with no stand-
alone charges in the space outside them. Our goal here is calculating the distribu-
tions of electric field E and potential ϕ in space, and the distribution of the surface
10
See, e.g. Part SM section 3.3.
2-6
charge density σ over the conductor surfaces. However, before doing that for
particular situations, let us see if there are any integral measures of these
distributions, that should be our primary focus.
The simplest case is of course a single conductor in the otherwise free space.
According to Eq. (2.1b), all its volume should have a constant electrostatic potential
ϕ, evidently providing one convenient global measure of the situation. Another
integral measure is provided by the total charge
Q≡ ∫V ρ d 3r ≡ ∮S σ d 2r, (2.11)
where the latter integral is extended over the whole surface S of the conductor. In the
general case, what can we tell about the relation between Q and ϕ? At Q = 0, there is
no electric field in the system, and it is natural (although not necessary) to select the
arbitrary constant in the electrostatic potential to have ϕ = 0. Then, if the conductor
is charged with a finite Q, according to the Coulomb law, the electric field in any
point of space is proportional to Q. Hence the electrostatic potential everywhere,
including its value ϕ on the conductor, is also proportional to Q:
ϕ = pQ. (2.12)
The proportionality coefficient p , which depends on the conductor’s size and shape,
but not on ϕ or Q, is called the reciprocal capacitance (or, not too often, ‘electric
elastance’). Usually, Eq. (2.12) is rewritten in a different form,
1
Q = Cϕ , with C ≡ , (2.13)
p
where C is called self-capacitance. (Frequently, C is called just capacitance, but as we
will see very soon, for more complex situations the latter term may be too
ambiguous.)
Before calculating C for particular geometries, let us have a look at the electro-
static energy U of a single conductor. In order to calculate it, of the several relations
discussed in chapter 1, Eq. (1.61) is the most convenient, because all elementary
charges qk are now parts of the conductor’s surface charge, and hence reside at the
same potential ϕ—see Eq. (2.1b) again. As a result, the equation becomes very
simple:
1 1
U= ϕ∑q = ϕ Q . (2.14)
2 k k 2
Moreover, using the linear relation (2.13), the same result may be re-written in two
more forms:
Q2 C
U= = ϕ2 . (2.15)
2C 2
2-7
We will discuss several ways to calculate C in the next sections, and right now will
have a quick look at just the simplest example, for which we have calculated
everything necessary in the previous chapter: a conducting sphere of radius R.
Indeed, we already know the electric field distribution: according to Eq. (2.1), E = 0
inside the sphere, while Eq. (1.19), with Q(r) = Q, describes the field distribution
outside it, because of the evident spherical symmetry of the surface charge
distribution. Moreover, since the latter formula is exactly the same as for the point
charge placed in the sphere’s center, the potential distribution in space can be
obtained from Eq. (1.35) by replacing q with the sphere’s full charge Q. Hence, on
the surface of the sphere (and, according to Eq. (2.1b), through its interior),
1 Q
ϕ= . (2.16)
4πε0 R
Comparing this result with the definition (2.13), for the self-capacitance we obtain a
very simple formula
C = 4πε0R . (2.17)
This formula, which should be very familiar to the reader11, is convenient to
obtain some feeling of how large the SI unit of capacitance (1 farad, abbreviated as
F) is: the self-capacitance of Earth (RE ≈ 6.34 × 106 m) is below 1 mF! Another
important note is that while Eq. (2.17) is not exactly valid for a conductor of
arbitrary shape, it implies an important estimate
C ∼ 2πε0a (2.18)
where a is the scale of the linear size of any conductor12.
Now proceeding to a system of two arbitrary conductors, we immediately see why
we should be careful with the capacitance definition: one constant C is insufficient to
describe such system. Indeed, here we have two, generally different, conductor
potentials, ϕ1 and ϕ2, that may depend on both conductor charges, Q1 and Q2.
Using the same arguments as for the single-conductor case, we may conclude that
the dependence is always linear:
ϕ1 = p 11 Q1 + p 12 Q2,
(2.19)
ϕ2 = p 21 Q1 + p 22 Q2,
but now has to be described by more than one coefficient. Actually, it turns out that
there are three rather than four different coefficients in these relation, because
11
In Gaussian units, using the standard replacement 4πε0 → 1, this relation takes an even simpler form: C = R,
easy to remember. Generally, in Gaussian units (but not in the SI system!) the capacitance has the
dimensionality of length, i.e. is measured in centimeters. Note also that a fractional SI unit, 1 picofarad
(10−12 F), is very close to the Gaussian unit: 1 pF = (1 × 10−12)/(4πε0 × 10−2) cm ≈ 0.8998 cm. So, this unit is
rather close to the capacitance of a metallic ball with a 1 cm radius, making it very convenient for human-scale
systems.
12
These arguments are somewhat insufficient to say which size should be used for a in the case of narrow,
extended conductors, e.g. a thin, long wire. Very soon we will see that in such cases the electrostatic energy,
and hence C, depends mostly on the larger size of the conductor.
2-8
p 12 = p 21 . (2.20)
This relation may be proved in several ways, for example, using the general
reciprocity theorem of electrostatics (whose proof was the subject of one of the
problems of chapter 1):
∫ ρ1(r) ϕ(2)(r)d 3r = ∫ ρ2 (r) ϕ(1)(r)d 3r, (2.21)
where ϕ(1)(r) and ϕ(2)(r) are the potential distributions induced, respectively, by two
electric charge distributions, ρ1(r) and ρ2(r). In our current case, each of these
integrals is limited to the volume (or, more exactly, the surface) of the corresponding
conductor, where each potential is constant, and may be taken out of the integral. As
a result, Eq. (2.21) is reduced to
Q1ϕ(2)(r1) = Q1ϕ(1)(r2). (2.22)
In terms of Eq. (2.19), ϕ(2)(r1) is just p 12Q2, while ϕ(1)(r2) equals p 21Q1. Plugging
these expressions into Eq. (2.22), and cancelling the products Q1Q2, we arrive at
Eq. (2.20).
Hence the 2 × 2 matrix of coefficients p jj′ (called the reciprocal capacitance
matrix) is always symmetric, and using the natural notation p 11 ≡ p 1, p 22 ≡ p 2,
p 12 = p 21 ≡ p , we may write it in a simple form
p1 p
( p p ). 2
(2.23)
Plugging the relation (2.19), in this new notation, into Eq. (1.61), we see that the full
electrostatic energy of the system may be expressed by a quadratic form:
p p
U = 1 Q12 + pQ1Q2 + 2 Q 22. (2.24)
2 2
It is evident that the middle term on the right-hand side of this equation describes
the electrostatic coupling of the conductors. (Without it, the energy would be just a
sum of two independent electrostatic energies of conductors 1 and 2.13) Still, even
with this simplification, Eqs. (2.19) and (2.20) show that in the general case of
arbitrary charges Q1 and Q2, the system of two conductors should be characterized
by three coefficients, rather than just one coefficient (‘the capacitance’). This is why
we may attribute a certain single capacitance to the system only in some particular
cases.
13
This is why systems with p ≪ p 1 , p 2 are called weakly coupled, and may be analyzed using approximate
methods—see, e.g. figure 2.4 and its discussion below.
2-9
For practice, the most important of them is when the system as the whole is
electrically neutral: Q1 = −Q2 ≡ Q. In this case the most important function of Q is
the difference of the conductors’ potentials, called the voltage14:
V ≡ ϕ1 − ϕ2 , (2.25)
For that function, the subtraction of the two Eqs. (2.19) gives
Q 1
V= , with C ≡ , (2.26)
C p 1 + p 2 − 2p
where the coefficient C is called the mutual capacitance between the conductors—or,
again, just ‘capacitance’, if the term’s meaning is absolutely clear from the context.
The same coefficient describes the electrostatic energy of the system. Indeed,
plugging Eqs. (2.19) and (2.20) into Eq. (2.24), we see that both forms of
Eq. (2.15) are reproduced if ϕ is replaced with V, Q1 with Q, and with C meaning
the mutual capacitance:
Q2 C
U= = V 2. (2.27)
2C 2
The best known system for which the mutual capacitance C may be readily
calculated is the plane (or ‘parallel-plate’) capacitor, a system of two conductors
separated with a narrow plane gap of thickness d and area A ∼ a2 ≫ d2—see
figure 2.3.
Figure 2.3. A schematic representation of the plane capacitor.
Since the surface charges, that contribute to the opposite charges ±Q of the
conductors of this system, attract each other, in the limit d ≪ a they sit entirely on
the opposite surfaces limiting the gap, so there is virtually no electric field outside of
the gap, while (according to the discussion in section 2.1) inside the gap it is normal
to the surfaces. According to Eq. (2.3), the magnitude of this field is E = σ/ε0.
Integrating this field across thickness d of the narrow gap, we obtain V ≡ ϕ1 − ϕ2 =
Ed = σd/ε0, so that σ = ε0V/d. But due the constancy of the potential of each
14
A word of caution: in condensed matter physics and electrical engineering, voltage is frequently defined as
the difference of electrochemical rather than electrostatic potentials. These two notions coincide if the
conductors have equal workfunctions—for example, if they are made of the same material. In this course
this condition will be implied, and the difference between the two voltages ignored—to be discussed in detail in
Part SM section 6.4.
2-10
electrode, V should not depend on the position in the gap area. As a result, σ should
be also constant over all the gap area A, and hence Q = σA = ε0V/d. Thus, we may
write V = Q/C, with
ε
C = 0 A. (2.28)
d
Let me offer a few comments on this well-known formula. First, it is valid even if
the gap is not quite planar, for example if it gently curves on a scale much larger than
d. Second, Eq. (2.28), valid if A ∼ a2 is much larger than d2, ignores the electric field
deviations from uniformity15 at distances ∼d near the gap edges. Finally, the same
condition (A ≫ d2) assures that C is much larger than the self-capacitance Cj ∼ ε0a of
each conductor—see Eq. (2.18). The opportunities open by this fact for electronic
engineering and experimental physics practice are rather astonishing. For example, a
very realistic 3 nm layer of high-quality aluminum oxide (which may provide a
nearly perfect electric insulation between two thin conducting films) with an area of
0.1 m2 (which is a typical area of silicon wafers used in semiconductor industry)
provides C ∼ 1 mF,16 larger than the self-capacitance of the whole planet Earth!
In the case shown in figure 2.3, the electrostatic coupling of the two conductors is
evidently strong. As an opposite example of a weakly coupled system, let us consider
two conducting spheres of the same radius R, separated by a much larger distance d
(figure 2.4).
Figure 2.4. A system of two well separated, similar conducting spheres.
In this case the diagonal components of the matrix (2.23) may be approximately
found from Eq. (2.16), i.e. by neglecting the coupling altogether:
1
p1 = p 2 ≈ . (2.29)
4πε0R
Now, if we had just one sphere (say, number 1), the electric potential at distance d
from its center would be given by Eq. (2.16): ϕ = Q1/4πε0d. If we move into this point
a small (R ≪ d) sphere without its own charge, we may expect that its potential
should not be too far from this result, so that ϕ2 ≈ Q1/4πε0d. Comparing this
expression with the second of Eq. (2.19) (taken for Q2 = 0), we obtain
1
p≈ ≪ p 1,2 . (2.30)
4πε0d
15
Frequently referred to as ‘fringe’ fields resulting in an additional ‘stray’ capacitance C′ ∼ ε0a ≪ C ∼ ε0a × (a/d).
16
Just as in section 2.1, in order for the estimate to be realistic, I took into account the additional factor κ (for
aluminum oxide, close to 10) which should be included into the nominator of Eq. (2.28) to make it applicable
to dielectrics—see chapter 3 below.
2-11
From here and Eq. (2.26) the mutual capacitance

1
C≈ ≈ 2πε0R . (2.31)
p1 + p 2
We see that (somewhat counter-intuitively), in this case C does not depend

substantially on the distance between the spheres, i.e. does not describe their
electrostatic coupling. The off-diagonal coefficients of the reciprocal capacitance
matrix (2.20) play this role much better—see Eq. (2.30).
Now let us consider the case when only one conductor of the two is charged, for
example Q1 ≡ Q, while Q2 = 0. Then Eqs. (2.19) and (2.20) yield
ϕ1 = p 1 Q1. (2.32)
Now, we may follow Eq. (2.13) and define C1 ≡ 1/ p 1 (and C2 ≡ 1/ p 2), just to see that
such partial capacitances of the conductors of the system differ from its mutual
capacitance C—cf Eq. (2.26). For example, in the case shown in figure 2.4, C1 = C2
≈ 4πε0R ≈ 2C.
Finally, let us consider one more frequent case when one of the conductors carries
a certain charge (say, Q1 = Q) but the potential of its counterpart is a sustained
constant, say ϕ2 = 0.17 (This condition is particularly easy to implement if the second
conductor is much larger that the first one. Indeed, as the estimate (2.18) shows, in
this case it would take a much larger charge Q2 to make the potential ϕ2 comparable
with ϕ1.) In this case the second of Eq. (2.19), with an account of Eq. (2.20), yields
Q2 = −( p / p 2)Q1. Plugging this relation into the first of those equations, we obtain
⎛ p 2⎞
−1
p2
Q1 = C1ef ϕ1, with C1ef ≡ ⎜p 1 − ⎟ ≡ . (2.33)
⎝ p2⎠ p1 p 2 − p 2
Thus, this effective capacitance of the first conductor, is generally different both from
both its partial capacitance C1 and the mutual capacitance C of the system,
emphasizing again how accurate one should be using this term.
Note also that none of these capacitances is equal to any element of the matrix
reciprocal to the matrix (2.23):
p1 p −1 1 − p2 p
( p p2 ) = (
p − p1 p 2 p − p1
2
. ) (2.34)
For this reason, the last matrix (sometimes the called the physical capacitance
matrix), which expresses the vector of conductor charges via the vector of their
potentials, is less useful for applications than the reciprocal capacitance matrix
(2.23). The same conclusion is valid for multi-conductor systems, which are most
17
In electrical engineering, such a constant-potential conductor is called the ground. This term stems from the
fact that in many cases the Earth surface may be considered a good electric ground, because its potential is
unaffected by laboratory-scale electric charges.
2-12
conveniently characterized by an evident generalization of Eq. (2.19). Indeed, in this

case even the mutual capacitance between two selected conductors may depend on
the electrostatics conditions of other components of the system.
Logically, at this point I would need to discuss the particular, but practically very
important, case when the regions where the electric field between each pair of
conductors is most significant, do not overlap—such as in the example shown in
figure 2.5a. In this case the system’s properties may be discussed using the equivalent
circuit language, representing each such region as a lumped (localized) capacitor,
with a certain mutual capacitance C, and the whole system as some connection
of these capacitors by conducting ‘wires’, whose length and geometry are not
important—see figure 2.5b.
Figure 2.5. (a) A simple system of conductors, with three well-localized regions of the electric field and surface
charge concentration, and (b) its representation with an equivalent circuit of three lumped capacitors.
Since the analysis such equivalent circuits is covered in typical introductory

physics courses, I will save time by skipping their discussion. However, since such
circuits are very frequently encountered in physical experiments and electrical
engineering practice, I would urge the reader to self-test his or her understanding
of this topic by solving a couple of problems given at the end of this chapter18, and if
their solution present any difficulty, review the corresponding section in an under-
graduate textbook.
2.3 The simplest boundary problems

In the general case when the electric field distribution in the free space between the
conductors cannot be easily found from the Gauss law or by any other special
methods, the best approach is to try to solve the differential Laplace equation (1.42),
with boundary conditions (2.1b):
∇2 ϕ = 0, ϕ Sk = ϕk , (2.35)
where Sk is the surface of the kth conductor of the system. After such a boundary
problem has been solved, i.e. the spatial distribution ϕ(r) has been found at all points
18
These problems have been selected to emphasize the fact that not every circuit may be reduced to the
simplest connections of the capacitors in parallel and/or in series.
2-13
outside the conductors, it is straightforward to use Eq. (2.3) to find the surface
charge density, and finally the total charge
Qk = ∮S k
σ d 2r (2.36)
of each conductor, and hence any component of the reciprocal capacitance matrix.
As an illustration, let us implement this program for three very simple problems.
(i) Plane capacitor (figure 2.3). In this case, the easiest way to solve the Laplace
equation is to use the linear (Cartesian) coordinates with one coordinate axis
(say, z), normal to the conductor surfaces—see figure 2.6. In these coordinates, the
Laplace operator is just the sum of three second derivatives19. It is evident that due
to problem’s translational symmetry in the {x, y} plane, deep inside the gap (i.e. at
the lateral distance from the edges much larger than d) the electrostatic potential
may only depend on the coordinate perpendicular to the gap surfaces: ϕ(r) = ϕ(z).
For such a function, the derivatives over x and y vanish, and the boundary problem
(2.35) is reduced to a very simple ordinary differential equation
d 2ϕ
(z ) = 0, (2.37)
dz 2
with boundary conditions
ϕ(0) = 0, ϕ(d ) = V . (2.38)
Figure 2.6. The plane capacitor as the system for the simplest illustration of the boundary problem (2.35) and
its solution.
(For the sake of notation simplicity, I have used the discretion of adding a constant
to the potential to make one of the potentials vanish, and also the definition (2.25) of
the voltage V.) The general solution of Eq. (2.37) is a linear function: ϕ (z) = c1z + c2,
whose constant coefficients c1,2 may be readily found from the boundary conditions
(2.38). The final solution is
z
ϕ=V . (2.39)
d
19
See, e.g. Eq. (A.56).
2-14
From here the only non-vanishing component of the electric field is

dϕ V
Ez = − =− , (2.40)
dz d
and the surface charge of the capacitor plates
V
σ = ε0En = ∓ε0Ez = ±ε0 , (2.41)
d
where the upper and lower sign correspond to the upper and lower plate,
respectively. Since σ does not depend on coordinates x and y, we can obtain the
full charges Q1 = −Q2 ≡ Q of the surfaces by its multiplication by the gap area A,
giving us again the already known result (2.28) for the mutual capacitance C ≡ Q/V.
I believe that this calculation, though very easy, may serve as a good introduction to
the boundary problem solution philosophy.
(ii) Coaxial-cable capacitor. The coaxial cable is a system of two round cylindrical,
coaxial conductors, with the cross-section as shown in figure 2.7.
Figure 2.7. The cross-section of a coaxial capacitor.
Evidently, in this case the cylindrical coordinates {ρ, φ, z}, with the axis z
coinciding with the common axis of the cylinders, are most convenient. Due to the
axial symmetry of the problem, in these coordinates E(r) = nρE(ρ), ϕ(r) = ϕ(ρ), so
that in the general expression for the Laplace operator20 we can take ∂/∂φ = ∂/∂z = 0.
As a result, only the first (radial) term of the operator survives, and the boundary
problem (2.35) takes the form
1 d ⎛ dϕ ⎞
⎜ρ ⎟ = 0, ϕ(a ) = V , ϕ(b) = 0. (2.42)
ρ dρ ⎝ dρ ⎠
The sequential integration of this ordinary differential equation is elementary (and

similar to that of the Poisson equation in spherical coordinates, carried out in
section 1.3), giving
ρ
dϕ dρρ ρ
ρ
dρ
= c1, ϕ(ρ) = c1 ∫a ρ′
+ c2 = c1 ln + c2 .
a
(2.43)
20
See, e.g. Eq. (A.61).
2-15
The constants c1,2 may be found using boundary conditions (2.42):

b
V = c2 , 0 = c1 ln + c2 , (2.44)
a
giving c1 = −V/ln(b/a), so that solution (2.43) takes the following form:
⎛ ln(ρ / a ) ⎞
ϕ(ρ) = V ⎜1 − ⎟. (2.45)
⎝ ln(b / a ) ⎠
Next, for our axial symmetry the general expression for the gradient of a function
is reduced to its radial derivative, so that
dϕ(ρ) V
E (ρ ) ≡ − = . (2.46)
dρ ρ ln (b / a )
This expression, plugged into Eq. (2.2), allows us to find the density of the
conductors’ surface charges. For example, for the inner electrode
ε0V
σa = ε0Ea = , (2.47)
a ln(b / a )
so that its full charge (per unit length of the system) is
Q 2πε0V
= 2πaσa = . (2.48)
l ln(b / a )
(It is straightforward to check that the charge of the outer electrode is equal and
opposite.) Hence, by the definition of the mutual capacitance, its value per unit
length is
C Q 2πε0
≡ = . (2.49)
l lV ln(b / a )
This expression shows that the total capacitance C is proportional to the systems
length l (if l ≫ a,b), while being only logarithmically dependent on the dimensions
of its cross-section. Since the logarithm of a very large argument is an extremely
slow function (sometimes called a quasi-constant), if the external conductor is made
large (b ≫ a) the capacitance diverges, but very weakly. Such a logarithmic
divergence may be cut by any miniscule additional effect, for example by the finite
length l of the system. This allows one to obtain a crude but very useful estimate of
the self-capacitance of a single round wire of radius a:
2πε0l
C≈ , for l ≫ a . (2.50)
ln(l / a )
On the other hand, if the gap between the conductors is very narrow, b = a + d with
d ≪ a, then ln(b/a) = ln(1 + d/a) may be approximated as d/a, and Eq. (2.49) is
2-16
reduced to C ≈ 2πε0aL/d, i.e. to Eq. (2.28) for the plane capacitor, with the
appropriate area A = 2πal.
(iii) Spherical capacitor. This is a system of two conductors, with the central cross-
section similar to that of the coaxial cable (figure 2.7), but now with the spherical
rather than axial symmetry. This symmetry implies that we would be better off using
spherical coordinates, so that potential ϕ depends only on one of them, the distance
r from the common center of the conductors: ϕ(r) = ϕ(r). As we already know from
section 1.3, in this case the general expression for the Laplace operator is reduced to
its first (radial) term, so that the Laplace equation takes the simple form (1.47).
Moreover, we have already found the general solution to this equation—see
Eq. (1.50):
c
φ(r ) = 1 + c2 . (2.51)
r
Now acting exactly as above, i.e. determining the constant c1 from the boundary
conditions ϕ(a) = V, ϕ(b) = 0, we obtain
⎛1 1⎞ V ⎛1 1 ⎞−1
V = c1⎜ − ⎟ , so that ϕ(r ) = ⎜ − ⎟ + c2 . (2.52)
⎝a b⎠ r ⎝a b⎠
Next, we can use the spherical symmetry to find electric field, E(r) = nrE(r), with
dϕ V ⎛ 1 1 ⎞−1
E (r ) = − = 2⎜ − ⎟ , (2.53)
dr r ⎝a b⎠
and hence its values on conductors’ surfaces, and then the surface charge density σ
from Eq. (2.3). For example, for the inner conductor’s surface,
V ⎛1 1 ⎞−1
σa = ε0E (a ) = ε0 ⎜ − ⎟ , (2.54)
a2 ⎝ a b⎠
so that, finally, for the full charge of that conductor we obtain the following
result:
⎛1 1 ⎞−1
Q = 4πa 2σ = 4πε0⎜ − ⎟ V . (2.55)
⎝a b⎠
(Again, the charge of the outer conductor is equal and opposite.) Now we can use
the definition (2.26) of the mutual capacitance to obtain the final result
Q ⎛1 1 ⎞−1 ab
C≡ = 4πε0⎜ − ⎟ = 4πε0 . (2.56)
V ⎝a b⎠ b−a
2-17
For b ≫a, this result coincides with Eq. (2.17) for the self-capacitance of the inner
conductor. On the other hand, if the gap between two conductors is narrow, d ≡
b − a ≪ a,
a(a + d ) a2
C = 4πε0 ≈ 4πε0 , (2.57)
d d
i.e. the capacitance approaches that of the planar capacitor of area A = 4πa2—as it
should.
All this seems (and is) very straightforward, but let us contemplate what the
reason was for such easy successes. In each of the cases (i)–(iii) we have managed to
find such coordinates that both the Laplace equation and the boundary conditions
involve only one of them. The necessary condition for the former fact is that the
coordinates are orthogonal. This means that three vector components of the local
differential dr, due to small variations of the new coordinates (say, dr, dθ, and dφ for
the spherical coordinates), are mutually perpendicular.
2.4 Using other orthogonal coordinates

Since the cylindrical and spherical coordinates used above are only the simplest
examples of the curvilinear orthogonal (or just ‘orthogonal’) coordinates, this
methodology may be extended to other coordinate systems of this type. As an
example, let us calculate the self-capacitance of a thin, round conducting disk. The
cylindrical or spherical coordinates would not give too much help here because,
although they have the appropriate axial symmetry about axis z, they would make
the boundary condition on the disk too complex—involving two coordinates, either
ρ and z, or r and θ. The help comes from noting that the flat disk, i.e. the area z = 0,
r < R, may be thought of as the limiting case of an axially symmetric ellipsoid—the
result of rotation of the usual ellipse about one of its axes—in our case, the
symmetry axis of the disk (in figure 2.8, axis z)21.
Figure 2.8. Solving the disk’s capacitance problem. (The cross-section of the system by the vertical plane
y = 0.)
Analytically, such an ellipsoid may be described by the following equation:
21
Alternative names for this surface are the ‘degenerate ellipsoid’, ‘ellipsoid of rotation’, and ‘spheroid’.
2-18
x2 + y2 z2
+ = 1, (2.58)
a2 b2
where a and b are the so-called major semi-axes whose ratio determines the ellipse
eccentricity (the degree of squeezing). For our problem, we will only need oblate
ellipsoids with a ⩾ b; according to Eq. (2.58), they may be represented as surfaces of
constant α in the degenerate ellipsoidal (or ‘spheroidal’) coordinates {α, β, φ}, which
are related to the Cartesian coordinates as follows:
x = R cosh α sin β cos φ ,
y = R cosh α sin β sin φ , (2.59)
z = R sinh α cos β.
Such ellipsoidal coordinates are the evident generalization of the spherical coor-
dinates, which correspond to the limit α ≫ 1 (i.e. r ≫ R). In the opposite limit, the
surface of constant α = 0 describes our thin disk of radius R, with the coordinate β
describing the distance ρ ≡ (x2 + y2)1/2 = R sin β of its point from the axis z. It is
almost evident (and easy to prove) that the curvilinear coordinates (2.59) are also
orthogonal, so that the Laplace operator may be expressed as a sum of three
independent terms:
1
∇2 =
R (cosh α − sin2 β )
2 2
⎡ 1 ∂ ⎛ ∂ ⎞ ⎤
⎢ ⎜cosh α ⎟ ⎥ (2.60)
⎢ cosh α ∂α ⎝ ∂α ⎠ ⎥
×⎢ 2 ⎥
.
1 ∂ ⎛ ∂ ⎞ ⎛ 1 1 ⎞ ∂
⎢+ ⎜sin β ⎟ + ⎜ 2 − ⎟ ⎥
⎣ sin β ∂β ⎝ ∂β ⎠ ⎝ sin β cosh2 α ⎠ ∂φ 2 ⎦
Although this expression may look a bit intimidating, let us notice that in our
current problem, the boundary conditions depend only on coordinate α:22
ϕ α=0 = V , ϕ α=∞ = 0. (2.61)
Hence there is every reason to believe that the electrostatic potential in all space is a
function of α alone; in other words, that all ellipsoids α = const are equipotential
surfaces. Indeed, acting on such a function ϕ(α) by the Laplace operator (2.60), we
see that the two last terms in the square brackets vanish, and the Laplace Eq. (2.35)
is reduced to a simple ordinary differential equation
d ⎡ dϕ ⎤
⎢⎣cosh α ⎥⎦ = 0. (2.62)
dα dα
Integrating it twice, just as we did in the previous problems, we obtain
22
I have called disk’s potential V, to distinguish it from the potential ϕ at an arbitrary point of space.
2-19
dα
ϕ(α ) = c1 ∫ cosh α
. (2.63)
This integral may be readily worked out, using the substitution ξ ≡ sinh α (with dξ ≡
cosh α dα, cosh2 α = 1 + sinh2 α = 1 + ξ2):
sinh α
dξ
ϕ(α ) = c1 ∫0 1 + ξ2
+ c2 = c1 tan−1(sinh α ) + c2 . (2.64)
The integration constants c1,2 are again simply found from boundary conditions, in
this case Eqs. (2.61), and we arrive at the following final expression for the
electrostatic potential:
⎡ 2 ⎤
ϕ(α ) = V ⎢1 − tan−1(sinh α )⎥ . (2.65)
⎣ π ⎦
This solution satisfies both the Laplace equation and the boundary conditions.
Mathematicians tell us that the solution of any boundary problem of the type (2.35)
is unique, so we do not need to look any further.
Now we may use Eq. (2.3) to find the surface density of electric charge, but in the
case of thin disk, it is more natural to add up such densities on its top and bottom
surfaces at the same distance ρ = (x2 + y2)1/2 from the disk’s center (which are
evidently equal, due to the problem symmetry about the plane z = 0): σ = 2ε0En∣z=+0.
According to Eq. (2.65), the electric field on the upper surface is
∂ϕ ∂ϕ(α )
En α=+0 = − =−
∂z z=+0
∂(R sinh α cos β ) α=+0 (2.66)
2 1 2 1
= V = V 2 ,
π R cos β π (R − ρ 2 )1/2
and we see that the charge is distributed along the disk very non-uniformly:
4 1
σ= ε0V 2 , (2.67)
π (R − ρ 2 )1/2
with a singularity at the disk edge. Below we will see that such singularities are very
typical for sharp edges of conductors23. Fortunately, in our current case the
divergence is integrable, giving a finite disk charge:
R R
4 ρ dρ
Q= ∫disk σ d 2ρ = ∫0 σ (ρ)2πρ dρ =
π
∫0
ε0V 2π
(R − ρ 2 )1/2
2
surface
(2.68)
1
dξ
= 4ε0VR ∫0 (1 − ξ )1/2
= 8ε0RV .
23
If you seriously worry about the formal infinity of the charge density at ρ → R, please remember that this
mathematical artifact disappears for any non-vanishing disk thickness.
2-20
Thus, for disk’s self-capacitance we get a very simple result,

2
C = 8ε0R ≡ 4πε0R , (2.69)
π
a factor of 2/π ≈ 0.64 lower than that for the conducting sphere of the same equal
radius, but still complying with the general estimate (2.18).
Can we always find such a ‘good’ system of orthogonal coordinates?
Unfortunately, the answer is no, even for highly symmetric geometries. This is
why the practical value of this approach is limited, and other, more general methods
of boundary problem solution are clearly needed. Before proceeding to their
discussion, however, let us note that in the case of 2D problems (i.e. cylindrical
geometries24), the orthogonal coordinate method gets much help from the following
conformal mapping approach.
Let us consider a pair of Cartesian coordinates {x, y} in the cross-section plane as
a complex variable z = x + iy,25 where i is the imaginary unity (i2 = −1), and let
w (z ) = u + iv be an analytic complex function of z .26 For our current purposes, the
most important property of an analytic function is that its real and imaginary parts
obey the following Cauchy–Riemann relations27:
∂u ∂v ∂v ∂u
= , =− . (2.70)
∂x ∂y ∂x ∂y
For example, for the function
w = z 2 ≡ (x + iy )2 ≡ (x 2 − y 2 ) + 2ixy , (2.71)
whose real and imaginary parts are
u ≡ Re w = x 2 − y 2 , v ≡ Im w = 2xy , (2.72)
we immediately see that ∂u/∂x = 2x = ∂v/∂y, and ∂v/∂x = 2y = −∂u/∂y, in accordance
with (2.70).
Let us differentiate the first of Eqs. (2.70) over x again, then change the order of
differentiation, and after that use the latter of those equations:
24
Let me remind the reader that the term cylindrical describes any surface formed by a translation, along a
straight line, of an arbitrary curve, and hence more general than the usual circular cylinder. (In this
terminology, for example, a prism is also a particular form of cylinder, formed by a translation of a polygon.)
25
The complex variable z should not be confused with the (real) third spatial coordinate z! We are considering
2D problems now, with the potential independent of z.
26
The analytic (or ‘holomorphic’) function may be defined as the one that may be expanded into the complex
Taylor series, i.e. is infinitely differentiable in the given point. (Almost all ‘regular’ functions, such as
z n, z1/n, exp z , ln z , etc, and their combinations are analytic at all z , maybe apart from certain special points.)
If the reader needs to brush up his or her background on this subject, I can recommend a popular (and very
inexpensive:-) textbook [3]
27
These relations may be used, in particular, to prove the Cauchy integral theorem—see, e.g. Eq. (A.91).
2-21
∂ 2u ∂ ∂u ∂ ∂v ∂ ∂v ∂ ∂u ∂ 2u
= = = = − = − , (2.73)
∂x 2 ∂x ∂x ∂x ∂y ∂y ∂x ∂y ∂y ∂y 2
and similarly for v. This means that the sum of the second-order partial derivatives
of each of the real functions u(x,y) and v(x,y) is zero, i.e. that both functions obey the
2D Laplace equation. This mathematical fact opens a nice way of solving problems
of electrostatics for (relatively simple) 2D geometries. Imagine that for a particular
boundary problem we have found a function w (z ) for which either u(x, y) or v(x, y)
is constant on all electrode surfaces. Then all lines of constant u (or v) represent
equipotential surfaces, i.e. the problem of the potential distribution has been
essentially solved.
As a simple example, consider a practically important problem: the quadrupole
electrostatic lens—a system of four cylindrical electrodes with hyperbolic cross-
sections, whose boundaries obey the following relations:
⎧+ a 2 , for the left and right electrodes,
x2 − y2 = ⎨ 2 (2.74)
⎩− a , for the top and bottom electrodes,
voltage-biased as shown in figure 2.9a. Comparing these relations with Eq. (2.72),
we see that each electrode surface corresponds to a constant value of the function u:
u = ±a2. Moreover, the potentials of both surfaces with u = +a2 are equal to +V/2,
while those with u = −a2 are equal to −V/2. Hence, we may conjecture that the
electrostatic potential at each point is a function of u alone; moreover, a simple
linear function,
ϕ = c1u + c2 = c1(x 2 − y 2 ) + c2 , (2.75)
is a valid (and hence the unique) solution of our boundary problem. Indeed, it does
satisfy the Laplace equation, while its constants c1,2 may be selected in a way to
satisfy all the boundary conditions shown in figure 2.9a:
V x2 − y2
ϕ= . (2.76)
2 a2
so that the boundary problem has been solved.
Figure 2.9. (a) The quadrupole electrostatic lens’s cross-section and (b) its conformal mapping.
2-22
According to Eq. (2.76), all equipotential surfaces are hyperbolic cylinders,

similar to those of the electrode surfaces. What remains is to find the electric field
at an arbitrary point inside the system:
∂ϕ x ∂ϕ y
Ex = − = −V 2 , Ey = − = V 2. (2.77)
∂x a ∂y a
These formulas show that if charged particles (e.g. electrons in an electron optics
system) are launched to fly ballistically through the lens, along axis z, they
experience a force pushing them toward the symmetry axis and proportional to
particle’s deviation from the axis (and thus equivalent in action to an optical lens
with a positive refraction power) in one direction, and a force pushing them out
(negative refractive power) in the perpendicular direction. One can show that letting
charged particles fly through several such lenses, with alternating voltage polarities,
in series enables beam focusing28.
Hence, we have reduced the 2D Laplace boundary problem to that of finding the
proper analytic function w (z ). This task may be also understood as that of finding a
conformal map, i.e. a correspondence between components of any point pair, {x, y}
and {u, v}, residing, respectively, on the initial Cartesian plane z and the plane w of
the new variables. For example, Eq. (2.71) maps the real electrode configuration
onto a plane capacitor of an infinite area (figure 2.9b), and the simplicity of
Eq. (2.75) is due to the fact that for the latter system the equipotential surfaces are
just parallel planes.
For more complex geometries, the suitable analytic function w (z ) may be hard to
find. However, for conductors with piece-linear cross-section boundaries, substantial
help may be obtained from the following Schwarz–Christoffel integral
w (z ) = const × ∫ (z − x1)k (z − x2d)kz …(z − xN −1)k

1 2 N −1
, (2.78)
that provides the conformal mapping of the interior of an arbitrary N-sided polygon
on the plane w = u + iv, onto the upper-half (y > 0) of the plane z = x + iy. In
Eq. (2.78), xj ( j = 1, 2, N − 1) are the points of axis y = 0 (i.e. of the boundary of the
mapped region on plane z ) to which the corresponding polygon vertices are mapped,
while kj are the exterior angles at the polygon vertices, measured in the units of π,
with −1 ⩽ kj ⩽ +1—see figure 2.1029. Of the points xj, two may be selected arbitrarily
(because their effects may be compensated by the multiplicative constant in
Eq. (2.78), and the additive constant of integration), while all the others have to
be adjusted to provide the correct mapping.
28
See, e.g. the textbook [4] or the review collection [5], in particular the review by K-J Hanszen and R Lauer,
pp 251–307.
29
The integral (2.78) includes only (N − 1) rather than N poles, because a polygon’s shape is completely
determined by (N − 1) positions w j of its vertices and (N − 1) angles πkj. In particular, since the algebraic sum
of all external angles of a polygon equals π, the last angle parameter kj = kN is uniquely determined by the set
of the previous ones.
2-23
Figure 2.10. The Schwartz–Christoffel mapping of a polygon’s interior onto the upper half-plane.
In the general case, the complex integral (2.78) may be difficult to tackle.
However, in some important cases, in particular those with right angles (kj = ±½)
and/or with some points w j at infinity, the integrals may be readily worked out,
giving explicit analytical expressions for the mapping functions w (z ). For example,
let us consider a semi-infinite strip, defined by restrictions −1 ⩽ u ⩽ +1 and 0 ⩽ v, on
the plane w —see the left-hand panel of figure 2.11.
Figure 2.11. A semi-infinite strip mapped onto the upper half-plane.
The strip may be considered as a polygon, with one vertex at the infinitely distant
vertical point w 3 = 0 + i∞. Let us map the polygon on the upper half of plane z ,
shown in the right-hand panel of figure 2.11, with the vertex w 1 = −1 + i0 mapped
onto the point x1 = −1, y1 = 0, and the vertex w 2 = +1 + i0 mapped onto the point
x2 = +1, y2 = 0. Since both external angles are equal to +π/2, and hence k1 = k2 = +½,
Eq. (2.78) is reduced to
dz dz
w (z ) = const × ∫
(z + 1)1/2 (z − 1)1/2
≡ const ×
(z 2 − 1)1/2
∫
(2.79)
dz
≡ const × i ∫
(1 − z 2 )1/2
.
2-24
This complex integral may be worked out, just as for real z , by the substitution
z = sinξ, giving
sin−1 z
w(z ) = const′ × ∫ dξ = c1 sin−1 z + c2 . (2.80)
Determining the constants c1,2 from the required mapping, i.e. from the equations
w (−1 + i0) = −1 + i0 and w (+1 + i0) = +1 + i0 (see the arrows in figure 2.11), we
finally obtain
2 −1 πw
w (z ) = sin z , i.e. z = sin . (2.81a )
π 2
Using the well-known expression for the sine of a complex argument30, we may
rewrite this elegant result in either of the following two forms for the real and
imaginary components of z and w :
2 −1 2x
u= sin ,
π [(x + 1) + y ] + [(x − 1)2 + y 2 ]1/2
2 2 1/2
2 [(x + 1)2 + y 2 ]1/2 + [(x − 1)2 + y 2 ]1/2 (2.81b)

v= cosh−1 ,
π 2
πu πv πu πv
x = sin cosh , y = cos sinh .
2 2 2 2
It is amazing how perfectly the last formula manages to keep y ≡ 0 at the different
borders of our w -region (figure 2.11): at its side borders (u = ±1, 0 ⩽ v < ∞), this is
performed by the first multiplier, while at the bottom border (−1 ⩽ u ⩽ +1, v = 0),
the equality is insured by the second multiplier.
This mapping may be used to solve several electrostatics problems with the
geometry shown in figure 2.11; probably the most surprising of them is the following
one. A straight gap of width 2t is cut in a very thin conducting plane, and voltage V
is applied between the resulting half-planes—see the bold straight lines in figure 2.12.
Selecting a Cartesian coordinate system with axis z directed along the cut, axis y
perpendicular to the plane, and the origin in the middle of the cut (figure 2.12), we
can write the boundary conditions of this Laplace problem as
⎧+ V /2, at x > t , y = 0,
ϕ=⎨ (2.82)
⎩− V /2, at x < −t , y = 0.
(Due to the problem’s symmetry, we may expect that in the middle of the gap, i.e. at
−t < x < +t and y = 0, the electric field is parallel to the plane and hence ∂ϕ/∂y = 0.)
The comparison of figures 2.11 and 2.12 shows that if we normalize our coordinates
{x, y} to t, Eqs. (2.81) provide the conformal mapping of our system on the plane z
to the plane capacitor on plane w , with voltage V between two planes u = ±1. Since
30
See, e.g. Eq. (A.20).
2-25
Figure 2.12. The equipotential surfaces of the electric field between two thin conducting semi-planes (or rather
their cross-sections by the perpendicular plane z = const).
we already know that in that case ϕ = (V/2)u, we may immediately use the first of
Eq. (2.81b) to write the final solution of the problem31:
V V 2x
ϕ= u= sin−1 . (2.83)
2 π [(x + t )2 + y 2 ]1/2 + [(x − t )2 + y 2 ]1/2
The thin lines in figure 2.12 show the corresponding equipotential surfaces32; it is
evident that the electric field concentrates at the gap edges, just as it did at the edge
of the thin disk (figure 2.8). Let me leave the remaining calculation of the surface
charge distribution and the mutual capacitance between the half-planes (per unit
length) for the reader’s exercise.
2.5 Variable separation—Cartesian coordinates

The general approach of the methods discussed in the last two sections was to satisfy
the Laplace equation by a function of a single variable that also satisfies the
boundary conditions. Unfortunately, in many cases this cannot be done (at least,
using practicably simple functions). In this case, a very powerful method, called
variable separation33, may work, frequently producing ‘semi-analytical’ results in the
form of series (infinite sums) of either elementary or well-studied special functions.
Its main idea is to express the solution of the general boundary problem (2.35) as the
sum of partial solutions,
31
This result may be also obtained using the so-called elliptical (not ellipsoidal!) coordinates, and by the
Green’s function method, to be discussed in section 2.10 below.
32
Another graphical representation of the electric field distribution, by field lines, is much less convenient. As
a reminder, the field lines are defined as lines to whom the (in our current case, electrostatic) field vectors are
tangential at each point. By this definition, the field lines are always normal to the equipotential surfaces, so
that it is always straightforward to sketch them from the equipotential surface pattern—such as shown in
figure 2.12.
33
Again, this method was already discussed in Part CM section 6.5, and then used in Part CM sections 6.6 and
8.4. However, the method is so important that I need to repeat its discussion in this part of my series, for the
benefit of the readers who have skipped my Classical Mechanics course for any reason.
2-26
ϕ= ∑ckϕk , (2.84)
k
where each function ϕk satisfies the Laplace equation, and then select the set of
coefficients ck to satisfy the boundary conditions. More specifically, in the variable
separation method the partial solutions ϕk are looked for in the form of a product of
functions, each depending of just one spatial coordinate.
Let us discuss this approach on the classical example of a rectangular box with
conducting walls (figure 2.13), with the same potential (that I will take for zero) at all
the walls, but a different potential V fixed at the top lid. Moreover, in order to
demonstrate the power of the variable separation method, let us carry out all the
calculations for a more general case when the top lead potential is an arbitrary 2D
function V(x, y)34.
Figure 2.13. The standard playground for the variable separation discussion: a rectangular box with five
conducting, grounded walls and a fixed potential distribution V(x,y) on the top lid.
For this geometry, it is natural to use the Cartesian coordinates {x, y, z} and
hence represent each of the partial solutions in Eq. (2.84) as a product
ϕk = X (x )Y (y )Z (z ). (2.85)
Plugging it into the Laplace equation expressed in the Cartesian coordinates,

∂ 2ϕk ∂ 2ϕk ∂ 2ϕk
+ + = 0, (2.86)
∂x 2 ∂y 2 ∂z 2
and dividing the result by the product XYZ, we obtain
1 d 2X 1 d 2Y 1 d 2Z
+ + = 0. (2.87)
X dx 2 Y dy 2 Z dz 2
Here comes the punchline of the variable separation method: since the first term
of this sum may depend only on x, the second one only on y, etc, Eq. (2.87) may be
satisfied everywhere in the volume only if each of these terms equals a constant.
Shortly we will see that for our current problem (figure 2.13), these constant x- and
34
Such distributions may be implemented in practice using the so-called mosaic electrodes consisting of many
electrically insulated and individually biased panels.
2-27
y-terms have to be negative; hence let us denote these variable separation constants as
(−α2) and (−β2), respectively. Now Eq. (2.87) shows that the constant z-term has to
be positive; if we denote it as γ2, we obtain the following relation:
α 2 + β 2 = γ 2. (2.88)
Now the variables are separated in the sense that for the functions X(x), Y(y), and
Z(z) we have obtained separate ordinary differential equations,
d 2X d 2Y d 2Z
+ α 2X = 0, + β 2Y = 0, − γ 2Z = 0, (2.89)
dx 2 dy 2 dz 2
which are related only by Eq. (2.88) for their parameters. Let us start from the
equation for function X(x). Its general solution is the sum of functions sinαx and
cosαx, multiplied by arbitrary coefficients. Let us select these coefficients to satisfy
our boundary conditions. First, since ϕ ∝ X should vanish at the back vertical wall
of the box (i.e. with the choice of coordinate origin shown in figure 2.13, at x = 0 for
any y and z), the coefficient at cosαx should be zero. The remaining coefficient (at
sinαx) may be included in the general factor ck in Eq. (2.84), so that we may take X
in the form
X = sin αx . (2.90)
This solution satisfies the boundary condition at the opposite wall (x = a) only if its
argument αa is a multiple of π, i.e. if α is equal to any of the following numbers
(commonly called eigenvalues)35:
π
αn = n , n = 1, 2, … (2.91)
a
(Terms with negative values of n would not be linearly independent from those with
positive n, and may be dropped from the sum (2.84). The value n = 0 is formally
possible, but would give X = 0, i.e. ϕk = 0, at any x, i.e. no contribution to sum
(2.84), so it may be dropped as well.) Now we see that we indeed had to take α real,
(i.e. α2 positive); otherwise, instead of the oscillating function (2.90) we would have a
sum of two exponential functions, which cannot equal zero in two independent
points of axis x.
Since Eq. (2.89) for function Y(y) is similar to that for X(x), and the boundary
conditions on the walls perpendicular to axis y (y = 0 and y = b) are similar to those
for x-walls, the absolutely similar reasoning gives
π
Y = sin βy , βm = m , m = 1, 2, …, (2.92)
b
35
Note that according to Eqs. (2.91) and (2.92), as the spatial dimensions a and b of the system are increased,
the distances between the adjacent eigenvalues tend to zero. This fact implies that for spatially infinite, non-
periodic systems, the eigenvalue spectra are continuous, so that the sums of the type (2.84) become integrals. A
few problems of this type are provided in section 2.9 for the reader’s exercise.
2-28
where the choice of the integer m is independent of that of n. Now we see that
according to Eq. (2.88), the separation constant γ depends on two indices, n and m,
so that the relation may be rewritten as
⎡⎛ n ⎞2 ⎛ m ⎞2 ⎤1/2
γnm ⎡ 2 2 ⎤1/2
= ⎣αn + βm⎦ = π ⎢⎜ ⎟ + ⎜ ⎟ ⎥ . (2.93)
⎣⎝ a ⎠ ⎝b⎠ ⎦
The corresponding solution of the differential equation for Z may be represented as

a sum of two exponents exp{±γnmz}, or alternatively as a linear combination of two
hyperbolic functions, sinhγnmz and cosh γnmz, with arbitrary coefficients. At our
choice of coordinate origin, the latter option is preferable, because cosh γnmz cannot
satisfy the zero boundary condition at the bottom lid of the box (z = 0). Hence, we
may take Z in the form
Z = sinh γnmz , (2.94)
which automatically satisfies that condition.
Now it is the right time to combine Eqs. (2.84) and (2.85) in a more explicit form,
replacing the temporary index k for full set of possible eigenvalues, in our current
case of two integer indices n and m:
∞
πnx πmy
ϕ(x , y , z ) = ∑ cnm sin a
sin
b
sinh γnmz , (2.95)
n, m = 1
where γnm is given by Eq. (2.93). This solution satisfies our boundary conditions on
all walls of the box, apart from the top, for arbitrary coefficients cnm. The only job
left is to choose these coefficients from the top-lid requirement:
∞
πnx πmy
ϕ(x , y , c ) ≡ ∑ cnm sin a
sin
b
sinh γnmc = V (x , y ). (2.96)
n, m = 1
It looks bad to have just one equation for the infinite set of coefficients cnm.
However, decisive help comes from the fact that the functions of x and y that
participate in Eq. (2.96) form full, orthogonal sets of 1D functions. The last term
means that the integrals of the products of the functions with different integer indices
over the region of interest equal zero. Indeed, a direct integration gives
a
πnx πn′x a
∫0 sin
a
sin
a
dx = δnn ′,
2
(2.97)
where δnn′ is the Kronecker symbol36, and similarly for y (with the evident
replacements a → b, n → m). Hence, a fruitful way to proceed is to multiply both
sides of Eq. (2.96) by the product of the basis functions, with arbitrary indices n′ and
m′, and integrate the result over x and y:
36
Let me hope that the reader knows what it is; if not, see Eq. (A.82).
2-29
∞ a b
πnx πn′x πmy πm′y
∑ cnm sinh γnmc ∫0 sin
a
sin
a
dx ∫0 sin
b
sin
b
dy
n, m = 1 (2.98)
a b
πn′x πm′y
= ∫0 dx ∫0 dy V (x , y )sin
a
sin
b
.
Due to Eq. (2.97), all terms on the left-hand side of the last equation, apart from
those with n = n′ and m = m′, vanish, and (replacing n′ with n, and m′ with m) we
finally obtain
a b
4 πnx πmy
cnm =
ab sinh γnmc
∫0 dx ∫0 dyV (x , y )sin
a
sin
b
. (2.99)
The relations (2.93), (2.95), and (2.99) give the complete solution of the posed
boundary problem; we can see both good and bad news here. The first bit of bad
news is that in the general case we still need to work out the integrals (2.99)—
formally, an infinite number of them. In some cases, it is possible to achieve this
analytically. For example, if the top lid in our problem is a single conductor, i.e. has
a constant potential, we may take V(x,y) = const ≡ V0, and both 1D integrations are
elementary; for example
a
πnx 2a ⎧1, for n odd,
∫0 sin dx = ×⎨ (2.100)
a πn ⎩ 0, for n even,
and similarly for the integral over y, so that
16V0 ⎧1, if both n and m are odd,

cnm = ×⎨ (2.101)
π nm sinh γnmc ⎩ 0,
2 otherwise.
The second item of bad news is that even at such a happy occasion, we still have to
sum up the series (2.95), so that our result may only be called analytical with some
reservations, because in most cases we need a computer to obtain the final numbers
or plots.
Now the first good news. Computers are very efficient for both operations (2.95)
and (2.99), i.e. for the summation and integration. (As was discussed in section 1.2,
random errors are averaged out at these operations.) As an example, figure 2.14
shows the plots of the electrostatic potential in a cubic box (a = b = c), with an
equipotential top lid (V = V0 = const), obtained by a numerical summation of the
series (2.95), using the analytical expression (2.101). The remarkable feature of this
calculation is a very fast convergence of the series; for the middle cross-section of the
cubic box (z/c = 0.5), already the first term (with n = m = 1) gives an accuracy about
6%, while the sum of four leading terms (with n, m = 1, 3) reduces the error to just
0.2%. (For a longer box, c > a, b, the convergence is even faster—see the discussion
below.) Only close to the corners between the top and the side walls, where the
2-30
Figure 2.14. The electrostatic potential’s distribution inside a cubic box (a = b = c) with a constant voltage V0
on the top lid (figure 2.13), calculated numerically from Eqs. (2.93), (2.95), and (2.101). The dashed line on the
left panel shows the contribution of the main term of the series (with n = m = 1) to the full result, for z/c = 0.5.
potential changes very rapidly, are several more terms necessary to obtain a
reasonable accuracy.
The second bit of good news is that our ‘semi-analytical’ result allows its ultimate
limits to be explored analytically. For example, Eq. (2.93) shows that for a very flat
box (c ≪ a, b), γn,mz ⩽ γn,mc ≪ 1 at least for the lowest terms of series (2.95), with n,
m ≪ c/a, c/b. In these terms, sinh functions in Eqs. (2.96) and (2.99) may be well
approximated with their arguments, and their ratio by z/c. So, if we limit the
summation to these terms, Eq. (2.95) gives a very simple result
z
ϕ(x , y ) ≈ V (x , y ), (2.102)
c
which means that each segment of the flat box behaves just as a plane capacitor.
Only near the side walls, the higher terms in the series (2.95) are important,
producing some deviations from Eq. (2.102). (For the general problem with an
arbitrary function V(x,y), this is also true at all regions where this function changes
sharply.)
In the opposite limit (a, b ≪ c), Eq. (2.93) shows that, in contrast, γn,mc ≫ 1 for all
n and m. Moreover, the ratio sinh γn,mz/sinh γn,mc drops sharply if either n or m is
increased, if z is not too close to c. Hence in this case a very good approximation
may be obtained by keeping just the leading term, with n = m = 1, in Eq. (2.95), so
that the issue of summation disappears. (As was discussed above, this approxima-
tion works reasonably well even for a cubic box.) In particular, for the constant
potential of the top lid, we can use Eq. (2.101) and the exponential asymptotic for
both sinh functions, to obtain a very simple formula:
2-31
16 πx πy ⎧ (a 2 + b 2 )1/2 ⎫
ϕ= sin sin exp ⎨ − π ( c − z ) ⎬. (2.103)
π2 a b ⎩ ab ⎭
These results may be readily generalized to some other problems. For example, if
all walls of the box shown in figure 2.13 have an arbitrary potential distribution,
one can use the linear superposition principle to argue that the electrostatic
potential distribution inside the box is the sum of six partial solutions of the type
of Eq. (2.95), each with one wall biased by the corresponding voltage and all others
grounded (ϕ = 0).
To summarize, the results given by the variable separation method in Cartesian
coordinates are closer to what we could call a genuinely analytical solution than to
purely numerical solutions. Now, let us explore the issues that arise when this
method is applied in other orthogonal coordinate systems.
2.6 Variable separation—polar coordinates

If a system of conductors is cylindrical, the potential distribution is independent of
the coordinate z along the cylinder axis: ∂ϕ/∂z =0, and the Laplace equation
becomes two-dimensional. If conductor’s cross-section is rectangular, the variable
separation method works best in Cartesian coordinates {x, y}, and is just a
particular case of the 3D solution discussed above. However, if the cross-section
is circular, much more compact results may be obtained by using the polar
coordinates {ρ, φ}. As we already know from the last section, these 2D coordinates
are orthogonal, so that the 2D Laplace operator is a sum of two separable terms37.
Requiring, just as we have done above, each component of the sum (2.84) to satisfy
the Laplace equation, we obtain
1 ∂ ⎛ ∂ϕk ⎞ 1 ∂ 2ϕk
⎜ρ ⎟+ 2 = 0. (2.104)
ρ ∂ρ ⎝ ∂ρ ⎠ ρ ∂φ 2
In a full analogy with Eq. (2.75), let us represent each particular solution as a
product: ϕk = R (ρ)F (φ). Plugging this expression into Eq. (2.104) and then dividing
all its parts by RF /ρ2, we obtain
ρ d ⎛ dR ⎞ 1 d 2F
⎜ρ ⎟ + = 0. (2.105)
R dρ ⎝ dρ ⎠ F dφ 2
Following the same reasoning as for the Cartesian coordinates, we obtain two
separated ordinary differential equations
d ⎛ dR ⎞
ρ ⎜ρ ⎟ = ν 2R , (2.106)
dρ ⎝ dρ ⎠
37
See, e.g. Eq. (A.61) with ∂/∂z = 0.
2-32
d 2F
+ ν 2F = 0, (2.107)
dφ 2
where ν2 is the variable separation constant.
Let us start their analysis from Eq. (2.106), plugging into it a probe solution R =
cρα, where c and α are some constants. The elementary differentiation shows that if
α ≠ 0, the equation is indeed satisfied for any c, with just one requirement on
constant α, namely α2 = ν2. This means that the following linear superposition
R = aνρ+ν + bνρ−ν , for ν ≠ 0, (2.108)
with any constant coefficients aν and bν, is also a solution to Eq. (2.106). Moreover,
the general theory of linear ordinary differential equations tells us that the solution
of a second-order equation such as Eq. (2.106) may only depend on just two constant
factors that scale two linearly independent functions. Hence, for all values ν2 ≠ 0,
Eq. (2.108) presents the general solution of that equation. The case when ν = 0, in
which functions ρ+ν and ρ−ν are just constants and hence are not linearly
independent, is special, but in this case the integration of Eq. (2.106) is straightfor-
ward38, giving
R = a 0 + b0 ln ρ , for ν = 0. (2.109)
In order to specify the separation constant, let us explore Eq. (2.107), whose
general solution is
⎧cν cos νφ + sν sin νφ , for ν ≠ 0,
F=⎨ (2.110)
⎩c0 + s0φ , for ν = 0.
There are two possible cases here. In many boundary problems solvable in
cylindrical coordinates, the free space region, in which the Laplace equation is
valid, extends continuously around the origin point ρ = 0. In this region, the
potential has to be continuous and uniquely defined, so that F has to be a 2π-
periodic function of angle φ. For that, one needs the product ν(φ + 2π) to be equal to
νφ + 2πn, with n an integer, immediately giving us a discrete spectrum of possible
values of the variable separation constant:
ν = n = 0, ±1, ±2, … (2.111)
In this case both functions R and F may be labeled with the integer index n. Taking
into account that the terms with negative values of n may be summed up with those
with positive n, and that s0 should equal zero (otherwise the 2π-periodicity of
function F would be violated), we see that the general solution to the 2D Laplace
equation may be represented as
38
Actually, we have already done this in section 2.3—see Eq. (2.43).
2-33
∞ ⎛ bn ⎞
ϕ(ρ , φ) = a 0 + b0 ln ρ + ∑⎜anρn + ⎟(cn cos nφ + sn sin nφ). (2.112)
n=1
⎝ ρn ⎠
Let us see how all this machinery works on the famous problem of a round
cylindrical conductor placed into an electric field that is uniform and perpendicular to
the cylinder’s axis at large distances (say, created by a large plane capacitor)—see figure
2.15a. First of all, let us explore the effect of system’s symmetries on the coefficients in
Eq. (2.112). Selecting the coordinate system as shown in figure 2.15a, and taking the
cylinder’s potential for zero, we immediately obtain a0 = 0. Moreover, due to the mirror
symmetry about the plane [x, z], the solution has to be an even function of the angle φ,
and hence all coefficients sn should also equal zero. Also, at large distances (ρ ≫ R) from
the cylinder axis its effect on the electric field should vanish, and the potential should
approach that of the uniform field E = E0nx:
ϕ → −E 0x = −E 0ρ cos φ , for ρ → ∞ . (2.113)
This is only possible if in Eq. (2.112), b0 = 0, and also all coefficients an with n ≠ 1
vanish, while the product a1c1 should be equal to (−E0). Thus the solution is reduced
to the following form
∞
Bn
ϕ(ρ , φ) = −E 0ρ cos φ + ∑ cos nφ , (2.114)
n=1
ρn
in which the coefficients Bn ≡ bncn should be found from the boundary condition on
the cylinder’s surface, i.e. at ρ = R:
ϕ(R , φ) = 0. (2.115)
This requirement yields the following equation,
∞
⎛ B1 ⎞ B
⎜ − E 0R +
⎝ R⎠
⎟ cos φ + ∑ Rnn cos nφ = 0, (2.116)
n=2
Figure 2.15. A conducting cylinder inserted into an initially uniform electric field perpendicular to is axis: (a)
the problem’s geometry and (b) the equipotential surfaces given by Eq. (2.117).
2-34
which should be satisfied for all φ. This equality, read backwards, may be considered
as an expansion of a function identically equal to zero into a series over mutually
orthogonal functions cosnφ.39 It is of course valid if all coefficients of the expansion,
including (−E0R + B1/R) and all Bn for n ⩾ 2, are equal to zero. As a result, our final
answer (valid only outside of the cylinder, i.e. for ρ ⩾ R), is
⎛ R2 ⎞ ⎛ R2 ⎞
ϕ(ρ , φ) = −E 0⎜ρ − ⎟ cos φ ≡ −E 0⎜1 − 2 ⎟x . (2.117)
⎝ ρ⎠ ⎝ x + y2 ⎠
This result (figure 2.15b) shows a smooth transition between the uniform field
Eq. (2.113) far from the cylinder, to the equipotential surface of the cylinder (with
ϕ = 0). Such smoothening is very typical for Laplace equation solutions. Indeed, as
we know from chapter 1, these solutions correspond to the lowest potential energy
(1.65), i.e. the lowest integral of the potential gradient’s square, possible at the given
boundary conditions.
To complete the problem, let us calculate the distribution of the surface charge
density over the cylinder’s cross-section, using Eq. (2.3):
∂ϕ ∂ ⎛ R2 ⎞
σ = ε0En surface ≡ −ε0 = ε0E 0 cos φ ⎜ρ − ⎟ = 2ε0E 0 cos φ . (2.118)
∂ρ ∂ρ ⎝ ρ⎠
ρ=R ρ=R
This very simple formula shows that with the field direction shown in figure 2.15a
(E0 > 0), the surface charge is positive on the right side of the cylinder and negative
on its left side, thus creating a field directed from the right to the left, which
compensates the external field inside the conductor, where the net field is zero.
(Please take one more look at the schematic figure 2.1a.) Note also that the net
electric charge of the cylinder is zero, in the correspondence with the problem
symmetry. Another useful by-product of the calculation (2.118) is that the surface
electric field equals 2E0cos φ, and hence its largest magnitude is twice the field far
from the cylinder. Such electric field concentration is very typical for all convex
conducting surfaces.
The last observation receives additional confirmation for the second possible
topology, when Eq. (2.110) is used to describe problems with no angular periodicity.
A typical example of this situation is a cylindrical conductor with a cross-section
that features a corner limited by straight lines (figure 2.16). Indeed, at we may argue
that at ρ < R (where R is the scale of radial extension of the straight sides of the
corner, see figure 2.16), the Laplace equation may be satisfied by a sum of partial
solutions R (ρ)F (φ), if the angular components of the products satisfy the boundary
conditions on the corner sides. Taking (just for the simplicity of notation) the
conductor’s potential to be zero, and one of the corner’s sides as the axis x (φ = 0),
these boundary conditions are
39
Mathematics tells us that such expansions are unique, so this is the only possible solution of Eq. (2.116).
2-35
Figure 2.16. The cross-sections of cylindrical conductors with (a) a corner and (b) a wedge.
F (0) = F (β ) = 0, (2.119)
where the angle β may be anywhere between 0 and 2π—see figure 2.16.
Comparing this condition with Eq. (2.110), we see that it requires all cν to vanish,
and ν to take one of the values of the following discrete spectrum:
νmβ = πm , with m = 1, 2, …. (2.120)
Hence the full solution of the Laplace equation takes the form
∞
πmφ
ϕ= ∑ amρπm/β sin β
, for ρ < R , 0 ⩽ φ ⩽ β, (2.121)
m=1
where constants sν have been incorporated into am. The set of constants am cannot be
universally determined, because it depends on the exact shape of the conductor
outside the corner and the externally applied electric field. However, whatever the set
is, in the limit ρ → 0, the solution (2.121) is almost40 always dominated by the term
with lowest ν (corresponding to m = 1),
π
ϕ → a1ρ π /β sin φ , (2.122)
β
because the higher terms go to zero faster. This potential distribution corresponds to
the surface charge density
∂ϕ πa
σ = ε0En∣ surface = −ε0 ∣ ρ=const, φ→+0 = −ε0 1 ρ(π /β−1) . (2.123)
∂(ρφ) β
(It is similar on the opposite face of the angle.)
The result (2.123) shows that if we are dealing with a usual, concave corner (β < π,
see figure 2.16a), the charge density (and the surface electric field) tends to zero. On
the other case, at a ‘convex corner’ with β > π (actually, a wedge—see figure 2.16b),
both the charge and the field’s strength concentrate, formally diverging at ρ → 0. (So,
40
Exceptions are possible only for highly symmetric configurations when the external field is specially crafted
to make a1 = 0. In this case the solution at ρ → 0 is dominated by the first non-vanishing term of the series
(2.121).
2-36
do not sit on a roof’s ridge during a thunderstorm; rather hide in a ditch!) We have
already seen qualitatively similar effects for the thin round disk and the split plane.
2.7 Variable separation—cylindrical coordinates

Now, let us discuss whether it is possible to generalize our approach to problems
whose geometry is still axially symmetric, but with a substantial dependence of the
potential on the axial coordinate (∂ϕ/∂z ≠ 0). The classical example of such a
problem is shown in figure 2.17. Here the side wall and the bottom lid of a hollow
round cylinder are kept at a fixed potential (say, ϕ = 0), but the potential V fixed at
the top lid is different. Evidently, this problem is qualitatively similar to the
rectangular box problem solved above (figure 2.13), and we will also try to solve
it first for the case of arbitrary voltage distribution over the top lid: V = V(ρ, φ).
Figure 2.17. A cylindrical volume with conducting walls.
Following the main idea of the variable separation method, let us require that
each partial function ϕk in Eq. (2.84) satisfies the Laplace equation, now in the full
cylindrical coordinates {ρ, φ, z}:41
1 ∂ ⎛ ∂ϕk ⎞ 1 ∂ 2ϕk ∂ 2ϕk
⎜ρ ⎟+ 2 + = 0. (2.124)
ρ ∂ρ ⎝ ∂ρ ⎠ ρ ∂φ 2 ∂z 2
Plugging in ϕk in the formR (ρ)F (φ)Z (z) into Eq. (2.124) and dividing all terms by
the product RF Z , we obtain
1 d ⎛ dR ⎞ 1 d 2F 1 d 2Z
⎜ρ ⎟ + 2 + = 0. (2.125)
ρR dρ ⎝ dρ ⎠ ρ F dϕ 2
Z dz 2
Since the first two terms of Eq. (2.125) can only depend on the polar variables ρ
and φ, while the third term only on z, at least that term should be a constant.
Denoting it (just like in the rectangular box problem) by γ2, we obtain, instead of
Eq. (2.125), a set of two equations:
d 2Z
= γ 2Z , (2.126)
dz 2
41
See, e.g. Eq. (A.61).
2-37
1 d ⎛ dR ⎞ 1 d 2F
⎜ρ ⎟ + γ 2 + 2 = 0. (2.127)
ρR dρ ⎝ dρ ⎠ ρ F dφ 2
Now, multiplying all terms of Eq. (2.127) by ρ2, we see that the last term, (d2F /
dφ )/F , may depend only on φ, and thus should be constant. Calling that constant ν2
2
(just as in section 2.6 above), we separate Eq. (2.127) into an angular equation,
d 2F
+ ν 2F = 0, (2.128)
dφ 2
and a radial equation:
d 2R 1 dR ⎛ ν2 ⎞
+ + ⎜γ 2
− ⎟R = 0. (2.129)
dρ 2 ρ dρ ⎝ ρ2 ⎠
We see that the ordinary differential equations for functions Z (z) and F (φ) (and
hence their solutions) are identical to those discussed earlier. However, Eq. (2.129)
for the radial function R (ρ) (called the Bessel equation) is more complex than in the
2D case, and depends on two independent constant parameters, γ and ν. The latter
challenge may be readily overcome if we notice that any change of γ may be reduced
to a re-scaling of the radial coordinate ρ. Indeed, introducing a dimensionless
variable ξ ≡ γρ,42 Eq. (2.129) may be reduced to an equation with just one
parameter, ν:
d 2R 1 dR ⎛ ν2 ⎞
+ + ⎜1 − ⎟R = 0. (2.130)
dξ 2 ξ dξ ⎝ ξ2 ⎠
Moreover, we already know that for angle-periodic problems, the spectrum of

eigenvalues of Eq. (2.128) is discrete: ν = n, with integer n.
Unfortunately, even in this case, Eq. (2.130) cannot be satisfied by a single
‘elementary’ function, and is the canonical form of the Bessel equation. Its solutions
that we need for our current problem are called the Bessel function of the first kind, of
order ν, commonly denoted as Jν(ξ). Let me review in brief the properties of these
functions that are most relevant for our problem—and many other problems
discussed in this series43.
First of all, the Bessel function of a negative integer order is very simply related to
that with the positive order:
42
Note that this normalization is specific for each value of the variable separation parameter γ. Also, please
notice that the normalization is meaningless for γ = 0, i.e. for the case Z(z) = const. However, if we need the
partial solutions for this particular value of γ, we can always use Eqs. (2.108) and (2.109).
43
For a more complete discussion of these functions, see the literature listed in appendix A, section 2.16, for
example, chapter 9 (written by F Olver) in [6].
2-38
Figure 2.18. Several Bessel functions Jn(ξ) of integer order. The dashed lines show the envelope of the
asymptotes (2.135).
J−n(ξ ) = ( −1)nJn(ξ ), (2.131)

enabling us to limit our discussion to the functions with n ⩾ 0. Figure 2.18 shows
four of these functions with the lowest positive n.
As its argument is increased, each function is initially close to a power law: J0(ξ) ≈ 1,
J1(ξ) ≈ ξ/2 = ξ/2, J2(ξ) ≈ ξ2/8, etc. This behavior follows from the Taylor series
∞
⎛ ξ ⎞n ( −1)k ⎛⎜ ξ ⎟⎞2k
Jn(ξ ) = ⎜ ⎟ ∑ , (2.132)
⎝2⎠ k ! (n + k )! ⎝ 2 ⎠
k=0
which is formally valid for any ξ, and may even serve as an alternative definition of
the functions Jn(ξ). However, this series is converging fast only at relatively small
arguments, ξ < n, where its leading term is
1 ⎛ ξ ⎞n
Jn(ξ ) → ⎜ ⎟ . (2.133)
ξ→0 n! ⎝ 2 ⎠
At ξ ≈ n + 1.86n1/3, the Bessel function reaches its maximum44

0.675
max ξ[Jn(ξ )] ≈ , (2.134)
n1/3
44
These two formulas for the Bessel function peak are strictly valid for n ≫ 1, but may be used for reasonable
estimates starting already from n = 1; for example, maxξ [J1(ξ)] is close to 0.58 and is reached at ξ ≈ 2.4, just
about 30% away from the values given by the asymptotic formulas.
2-39
and then starts to oscillate with a period gradually approaching 2π, a phase shift that
increases by π/2 with each unit increment of n, and an amplitude that decreases as
ξ−1/2. These features are described by the following asymptotic formula:
⎛ 2 ⎞1/2 ⎛ π nπ ⎞⎟
Jn(ξ ) → ⎜ ⎟ cos⎜ξ − − , for ξ / n → ∞ , (2.135)
⎝ πξ ⎠ ⎝ 4 2⎠
which starts to give reasonable results very soon above the function peaks—see
figure 2.1845.
Now we are ready to return to our case study (figure 2.17). Let us select functions
Z (z) to satisfy the Eq. (2.126) and the bottom-lid boundary condition Z (0) = 0, i.e.
proportional to sinhγz—cf Eq. (2.94). Then
∞
ϕ= ∑∑Jn(γρ)(cnγ cos nφ + snγ sin nφ) sinh γ z . (2.136)
n=0 γ
Next, we need to satisfy the zero boundary condition at the cylinder’s side wall
(ρ = R). This may be ensured by taking
Jn(γR ) = 0. (2.137)
Since each function Jn(x) has an infinite number of positive zeros (see figure 2.18),
which may be numbered by an integer index m = 1, 2, …, Eq. (2.137) may be
satisfied with an infinite number of discrete values of the separation parameter γ:
ξnm
γnm = , (2.138)
R
where ξnm is the mth zero of the function Jn(x)—see the top numbers in the cells of
table 2.1. (Very soon we will see what we need the bottom numbers for.)
Hence, Eq. (2.136) may be represented in a more explicit form:
∞ ∞
⎛ ρ ⎞⎟ ⎛ z⎞
ϕ(ρ , φ , z ) = ∑ ∑ Jn⎜⎝ξnm (cnm cos nφ + snm sin nφ) sinh ⎜ξnm ⎟ . (2.139)
n=0 m=1
R⎠ ⎝ R⎠
Here the coefficients cnm and snm have to be selected to satisfy the only remaining
boundary condition—that on the top lid:
∞ ∞
⎛ ⎞ ⎛ ⎞
ϕ(ρ , φ , l ) ≡ ∑ ∑ Jn⎜⎝ξnm ρ ⎟⎠(cnm cos nφ + snm sin nφ) sinh ⎝⎜ξnm L ⎠⎟
n=0 m=1 R R (2.140)
= V (ρ , φ).
45
Eq. (2.135) and figure 2.18 clearly show the close analogy between the Bessel functions and the usual
trigonometric functions, sine and cosine. In order to emphasize this similarity, and help the reader to develop a
better gut feeling of the Bessel functions, let me mention one fact of elasticity theory: while sine functions
describe, in particular, possible modes of standing waves on a guitar string, functions Jn(ξ) describe, in
particular, possible standing waves on an elastic round membrane, with J0(ξ) describing their lowest
(fundamental) mode.
2-40
Table 2.1. Approximate values of a few first zeros, ξnm, of a few lowest-order Bessel functions Jn(ξ) (the top
number in each cell), and the values of dJn(ξ)/dξ at these points (the bottom number).
m=1 2 3 4 5 6
n=0 2.40482 5.52008 8.65372 11.79215 14.93091 18.07106

−0.51914 +0.34026 −0.27145 +0.23245 −0.20654 +0.18773
1 3.83171 7.01559 10.17347 13.32369 16.47063 19.61586
−0.40276 +0.30012 −0.24970 +0.21836 −0.19647 +0.18006
2 5.13562 8.41724 11.61984 14.79595 17.95982 21.11700
−0.33967 +0.27138 −0.23244 +0.20654 −0.18773 +0.17326
3 6.38016 9.76102 13.01520 16.22347 19.40942 22.58273
−0.29827 +0.24942 −0.21828 +0.19644 −0.18005 +0.16718
4 7.58834 11.06471 14.37254 17.61597 20.82693 24.01902
−0.26836 +0.23188 −0.20636 +0.18766 −0.17323 +0.16168
5 8.77148 12.33860 15.70017 18.98013 22.21780 25.43034
−0.24543 +0.21743 −0.19615 +0.17993 −0.16712 +0.15669
To use it, let us multiply both parts of Eq. (2.140) by Jn(ξnm′ρ/R) cos n′φ, integrate
the result over the lid area, and use the following property of the Bessel functions:
1
1
∫0 Jn(ξnms )Jn(ξnm ′s ) s ds =
2
[Jn+1(ξnm )]2 δmm ′. (2.141)
The relation (2.141) expresses a very specific (‘2D’) orthogonality of the Bessel
functions with different indices m—do not confuse them with the function’s order n,
please46! Since it relates two Bessel functions with the same index n, it is natural to
ask why its right-hand side contains the function with a different index (n + 1). Some
clue may come from one more very important property of the Bessel functions, the
so-called recurrence relations47:
2nJn(ξ )
Jn−1(ξ ) + Jn+1(ξ ) = , (2.142a )
ξ
46
The Bessel functions of the same argument but of different orders are also orthogonal, but in a different way:
∞ dξ 1
∫0 Jn(ξ )Jn ′(ξ )
ξ
=
n + n′
δnn ′.
47
These relations provide, in particular, a convenient way for fast numerical computation of all Jn(ξ) after J0(ξ)
has been computed. (The latter is usually done with an algorithm using Eq. (2.132) for smaller ξ and an
extension of Eq. (2.135) for larger ξ.) Note that most mathematical software packages, including all those
listed in appendix A, section A.16(iv), include ready subroutines for calculation of the functions Jn(ξ) and other
special functions used in this lecture series. In this sense, the line separating these ‘special functions’ from
‘elementary functions’ is rather fine.
2-41
dJn(ξ )
Jn−1(ξ ) − Jn+1(ξ ) = 2 , (2.142b)
dξ
which in particular yield the following relation (convenient for working out some
Bessel function integrals):
d n
[ξ Jn(ξ )] = ξ nJn−1(ξ ). (2.143)
dξ
For our current purposes, let us apply the recurrence relations at the special
points ξnm. At these points, Jn vanishes, and the system of two equations (2.142) may
be readily solved to obtain, in particular,
dJn
Jn+1(ξnm ) = − (ξnm ), (2.144)
dξ
so that the square bracket on the right-hand side of Eq. (2.141) is just (dJn/dξ)2 at ξ =
ξnm. Thus the values of the Bessel function derivatives at the zero points, given by the
lower numbers in the cells of table 2.1, are as important for boundary problem
solutions as the zeros themselves.
Since the angular functions cos nφ are also orthogonal—both to each other,
2π
∫0 cos(nφ) cos(n′φ) dφ = πδnn ′, (2.145)
and to all functions sin nφ, the integration over the lid area kills all terms of both
series in Eq. (2.140), apart from just one term proportional to cn′m′, and hence gives
an explicit expression for that coefficient. The counterpart coefficients sn′m′ may be
found by repeating the same procedure with the replacement of cos n′φ by sin n′φ.
This evaluation (left for the reader’s exercise) completes the solution of our problem
for an arbitrary lid potential V(ρ,φ).
Still, before leaving the Bessel functions (for a while only:-), we need to address
two important issues. First, we have seen that in our cylinder problem (figure 2.17),
the set of functions Jn(ξnmρ/R) with different indices m (which characterize the degree
of the Bessel function’s stretch along axis ρ) play a role similar to that of functions
sin(πnx/a) in the rectangular box problem shown in figure 2.13. In this context, what
is the analog of functions cos(πnx/a) which may be important for some boundary
problems? In a more formal language, are there any functions of the same argument
ξ ≡ ξnmρ/R that would be linearly independent of the Bessel functions of the first
kind, while satisfying the same Bessel Eq. (2.130)?
The answer is yes. For the definition of such functions, we first need to generalize
our prior formulas for Jn(ξ), and in particular Eq. (2.132), to the case of arbitrary,
not necessarily real order ν. Mathematics says that the generalization may be
performed in the following way:
∞
⎛ ξ ⎞ν ( −1)k ⎛ ξ ⎞2k
Jν(ξ ) = ⎜ ⎟ ∑ ⎜ ⎟ , (2.146)
⎝2⎠ k !Γ ( ν + k + 1) ⎝2⎠
k=0
2-42
where Γ(s) is the so-called gamma function that may be defined as48
∞
Γ( s ) ≡ ∫0 ξ s−1e−ξ dξ. (2.147)
The simplest, and the most important property of the gamma function is that for
integer values of argument it gives the factorial of a number smaller by one:
Γ(n + 1) = n ! ≡ 1 ⋅ 2 ⋅ … ⋅n , (2.148)
so it is essentially a generalization of the notion of the factorial to all real numbers.

The Bessel functions defined by Eq. (2.146) satisfy, after the replacements n → ν
and n! → Γ(n + 1), virtually all the relations discussed above, including the Bessel
Eq. (2.130), the asymptotic formula (2.135), the orthogonality condition (2.141), and
the recurrence relations (2.142). Moreover, it may be shown that ν ≠ n, functions
Jν(ξ) and J-ν(ξ) are linearly independent of each other, and hence their linear
combination may be used to represent a general solution of the Bessel equation.
Unfortunately, as (2.131) shows, for ν = n this is not true, and a solution
independent of Jn(ξ) has to be formed in a different way.
The most common way of overcoming this difficulty is first to define, for all ν ≠ n,
functions
Jν(ξ )cos νπ − J−ν(ξ )
Yν(ξ ) ≡ , (2.149)
sin νπ
called the Bessel functions of the second kind, or more often the Weber functions49,
and then to follow the limit ν → n. At this, both the nominator and denominator of
the right-hand side of Eq. (2.149) tend to zero, but their ratio tends to a finite value
called Yn(x). It may be shown that the resulting functions are still the solutions of the
Bessel equation and are linearly independent of Jn(x), though are related just as
those functions if the sign of n changes:
Y−n(ξ ) = ( −1)nYn(ξ ). (2.150)
Figure 2.19 shows a few Weber functions of the lowest integer orders. The plots
show that the asymptotic behavior is very much similar to that of Jn(ξ ),
⎛ 2 ⎞1/2 ⎛ π nπ ⎟⎞
Yn(ξ ) → ⎜ ⎟ sin⎜ξ − − , for ξ → ∞ , (2.151)
⎝ πξ ⎠ ⎝ 4 2⎠
but with the phase shift necessary to make these Bessel functions orthogonal to those
of the first order—cf Eq. (2.135). However, for small values of argument ξ, the Bessel
48
See, e.g. Eq. (A.34). Note that Γ(s) → ∞ at s → 0, −1, −2,….
49
They are also sometimes called the Neumann functions, and denoted as Nν(ξ).
2-43
Figure 2.19. A few Bessel functions of the second kind (a.k.a. the Weber functions, a.k.a. the Neumann
functions).
functions of the second kind behave completely differently from those of the first
kind:
⎧2⎛ ξ ⎞
⎪ ⎜ln
⎪π ⎝ 2
+ γ⎟, for n = 0,
⎠
Yn(ξ ) → ⎨ (2.152)
⎪− (n − 1)! ⎜⎛ ξ ⎟⎞ ,
−n
⎪ for n ≠ 0,
⎩ π ⎝2⎠
where γ is the so-called Euler constant, defined as follows:

⎛ 1 1 1 ⎞
γ ≡ lim n→∞⎜1 + + + … + − ln n⎟ ≈ 0.577157 … (2.153)
⎝ 2 3 n ⎠
As Eq. (2.152) and figure 2.19 show, the functions Yn(ξ ) diverge at ξ → 0 and hence
cannot describe the behavior of any physical variable, in particular the electrostatic
potential. One may wonder: if this is true, when do we need these functions in
physics? This does not happen too often, but still does.
Figure 2.20 shows an example of a simple boundary problem of electrostatics,
whose solution by the variable separation method involves both functions Jn(ξ ) and
Yn(ξ ). Here two round, conducting coaxial cylindrical tubes are kept at the same
(say, zero) potential, but at least one of two lids has a different potential. The
problem is almost completely similar to that discussed above (figure 2.17), but now
we need to find the potential distribution in the free space between the tubes, i.e. for
R1 < ρ < R2. If we use the same variable separation as in the simpler counterpart
problem, we need the radial functions R (ρ) to satisfy two zero boundary conditions:
at ρ = R1 and ρ = R2. With the Bessel functions of just the first kind, Jn(γρ), it is
2-44
Figure 2.20. A simple boundary problem that cannot be solved using just one kind of Bessel functions.
impossible to do, because the two boundaries would impose two independent (and
generally incompatible) conditions, Jn(γR1) = 0, and Jn(γR2) = 0, for one ‘stetching
parameter’ γ. The existence of the Bessel functions of the second kind immediately
saves the day, because if the radial function solution is represented as a linear
combination,
R = cJ Jn(γρ) + cY Yn(γρ), (2.154)
two zero boundary conditions give two equations for γ and ratio c ≡ cY/cJ.50 (Due to
the oscillating character of both Bessel functions, these conditions would be typically
satisfied by an infinite set of discrete pairs {γ, c}.) Note, however, that generally none
of these pairs would correspond to zeros of either Jn nor Yn, so that having an analog
of table 2.1 for the latter function would not help much. Hence, even simple
problems of this kind (such as the one shown in figure 2.20) typically require
numerical solutions of algebraic (transcendental) equations.
In order to complete the discussion of variable separation in the cylindrical
coordinates, one more issue to address is the so-called modified Bessel functions: of
the first kind, Iν(ξ), and of the second kind, Kν(ξ). They are two linearly independent
solutions of the modified Bessel equation,
d 2R 1 dR ⎛ ν2 ⎞
+ − ⎜1 + ⎟R = 0, (2.155)
dξ 2 ξ dξ ⎝ ξ2 ⎠
which differs from Eq. (2.130) ‘only’ by the sign of one of its terms. Figure 2.21
shows a simple problem that leads to this equation: a round conducting cylinder is
sliced, perpendicular to its axis, to rings of equal height h, which are kept at equal
but sign-alternating potentials.
50
A pair of independent linear functions, used for representation of the general solution of the Bessel equation,
may be also chosen in a different way, using the so-called Hankel functions
(1,2)
Hn (ξ ) ≡ Jn(ξ ) ± iYn(ξ ).
For representing the general solution of Eq. (2.130), this alternative is completely similar, for example, to using
the pair of complex functions exp{±iαx} = cos αx ± isin αx instead of the pair of real functions {cos αx, sin αx}
for representing the general solution of Eq. (2.89) for X(x).
2-45
Figure 2.21. A typical boundary problem whose solution may be conveniently described in terms of the
modified Bessel functions.
If the gaps between the sections are narrow, t ≪ h, we may use the variable
separation method for the solution to this problem, but now we evidently need
periodic (rather than exponential) solutions along axis z, i.e. linear combinations of
sin kz and cos kz with various real values of the constant k. Separating the variables,
we arrive at a differential equation similar to Eq. (2.129), but with the negative sign
before the separation constant:
d 2R 1 dR ⎛ ν2 ⎞
+ − ⎜k 2 + 2 ⎟R = 0. (2.156)
dρ 2
ρ dρ ⎝ ρ ⎠
The radial coordinate’s normalization, ξ ≡ kρ, immediately leads us to Eq. (2.155),

and hence (for ν = n) to the modified Bessel functions In(ξ) and Kn(ξ).
Figure 2.22 shows the behavior of such functions, of a few lowest orders. One can
see that at ξ → 0 the behavior is virtually similar to that of the ‘usual’ Bessel
functions—cf Eqs. (2.132) and (2.152), with Kn(ξ) multiplied (due to purely historical
reasons) by an additional coefficient, π/2:
⎧ ⎡ ⎛ξ ⎞ ⎤
⎪− ⎢ln ⎜ ⎟ + γ ⎥ , for n = 0,
1 ⎛ξ ⎞n ⎪ ⎣ ⎝2⎠ ⎦
In(ξ ) → ⎜ ⎟ , K n(ξ ) → ⎨ (2.157)
⎝
n! 2 ⎠ ⎪ (n − 1)! ⎛⎜ ξ ⎞⎟− n
⎪ , for n ≠ 0,
⎩ 2 ⎝2⎠
However, the asymptotic behavior of the modified functions is very different, with
In(x) exponentially growing, and Kn(ξ) exponentially dropping at ξ → ∞:
⎛ 1 ⎞1/2 ⎛ π ⎞1/2
In(ξ ) → ⎜ ⎟ eξ, K n(ξ ) → ⎜ ⎟ e−ξ . (2.158)
⎝ 2πξ ⎠ ⎝ 2ξ ⎠
To complete our brief survey of the Bessel functions, let me note that all them
discussed so far may be considered as particular cases of Bessel functions of the
complex argument, say Jn(z ) and Yn(z ), or, alternatively, Hn(1,2)(z ) = Jn(z ) ±
iYn(z ).51 The ‘usual’ Bessel functions Jn(ξ) and Yn(ξ) may be considered as the
51
These complex functions still obey the general relations (2.143) and (2.146), with ξ replaced with z .
2-46
Figure 2.22. The modified Bessel functions of the first kind (left panel) and the second kind (right panel).
sets of values of these generalized functions on the real axis (z = ξ), while the
modified functions as their particular case at z = iξ, also with real ξ:
π
Iν(ξ ) = i −νJν(iξ ), K ν(ξ ) = i ν+1Hν(1)(iξ ). (2.159)
2
Moreover, this generalization of the Bessel functions to the whole complex plane z
enables the use of their values along other directions on that plane, for example
under angles π/4 ± π/2. As a result, one arrives at the so-called Kelvin functions:
berνξ + i bei νξ ≡ Jν(ξe−iπ /4) ,
π (2.160)
kerνξ + i kei νξ ≡ i Hν(1)(ξe−i 3π /4) ,
2
which are also useful for some important problems of physics and engineering.
Unfortunately, I do not have time to discuss these problems in this course52.
2.8 Variable separation—spherical coordinates

The spherical coordinates are very important in physics, because of the (approx-
imate) spherical symmetry of many physical objects—from nuclei and atoms, to
water drops in clouds, to planets and stars. Let us again require each component ϕk
of Eq. (2.84) to satisfy the Laplace equation. Using the full expression for the
Laplace operator in spherical coordinates53, we obtain
52
Later in the course we will also run into the so-called spherical Bessel functions jn(ξ) and yn(ξ), which may be
expressed via the Bessel functions of a semi-integer order. Surprisingly enough, the spherical Bessel functions
turn out to be much simpler than Jn(ξ) and Yn(ξ).
53
See, e.g. Eq. (A.67).
2-47
1 ∂ ⎛ 2 ∂ϕk ⎞ 1 ∂ ⎛ ∂ϕk ⎞ 1 ∂ 2ϕk

⎜r ⎟ + ⎜sin θ ⎟ + = 0. (2.161)
r 2 ∂r ⎝ ∂r ⎠ r 2 sin θ ∂θ ⎝ ∂θ ⎠ r 2 sin2 θ ∂φ 2
Let us look for a solution of this equation in the following variable-separated form:
R (r )
ϕk = P (cos θ )F (φ), (2.162)
r
Separating the variables one by one, just as has been done in cylindrical coordinates,
we obtain the following equations for the functions participating in this solution:
d 2R l (l + 1)
− R = 0, (2.163)
dr 2 r2
d ⎡ dP ⎤ ⎡ ν2 ⎤
⎢(1 − ξ 2 ) ⎥ + ⎢l (l + 1) − ⎥P = 0, (2.164)
dξ ⎣ dξ ⎦ ⎣ 1 − ξ2 ⎦
d 2F
+ ν 2F = 0, (2.165)
dφ 2
where ξ ≡ cos θ is a new variable in lieu of θ (so that −1 ⩽ ξ ⩽ +1), while ν2 and l(l +
1) are the separation constants. (The reason for selection of the latter one in this
form will be clear in a minute.)
One can see that Eq. (2.165) is very simple, and is absolutely similar to the
Eq. (2.107) we obtained for the cylindrical coordinates. Moreover, the equation for
the radial functions is simpler than in the cylindrical coordinates. Indeed, let us look
for its solution in the form crα—just as we have done with Eq. (2.106). Plugging this
solution into Eq. (2.163), we immediately obtain the following condition on the
parameter α:
α(α − 1) = l (l + 1). (2.166)
This quadratic equation has two roots, α = l + 1 and α = −l, so that the general
solution to Eq. (2.163) is
bl
R = al r l +1 + . (2.167)
rl
However, the general solution of Eq. (2.164) (called either the general or
associated Legendre equation) cannot be expressed via what is usually called
elementary functions—although there is no generally accepted line between ‘ele-
mentary’ and ‘special’ functions.
Let us start its discussion from the axially symmetric case, when ∂ϕ/∂φ = 0. This
means F (φ) = const, and thus ν = 0, so that Eq. (2.164) is reduced to the so-called
Legendre differential equation:
2-48
d ⎡ dP ⎤
⎢(1 − ξ 2 ) ⎥ + l (l + 1)P = 0. (2.168)
dξ ⎣ dξ ⎦
One can readily check that the solutions of this equation for integer values of l are
specific (Legendre) polynomials54 that may be described by the following Rodrigues
formula:
1 dl 2 l
P l (ξ ) = (ξ − 1) , l = 0, 1, 2, …. (2.169)
2l l! dξl
As this formula shows, the first few Legendre polynomials are pretty simple:
P 0(ξ ) = 1,
P 1(ξ ) = ξ,
1
P 2(ξ ) = (3ξ 2 − 1) ,
2 (2.170)
1 3
P 3(ξ ) = (5ξ − 3ξ ) ,
2
1
P 4(ξ ) = (35ξ 4 − 30ξ 2 + 3) , …,
8
though such explicit expressions become more and more bulky as l is increased. As
figure 2.23 shows, all these polynomials, which are defined on the [−1, +1] segment,
start at the same point, P l(+1) = +1, and end up either at the same point or at the
Figure 2.23. A few lowest Legendre polynomials P l (ξ).
54
For the reader’s reference: if l is not integer, the general solution of Eq. (2.168) may be represented as a linear
combination of the so-called Legendre functions (not polynomials!) of the first and second kind, P l (ξ ) and Q l (ξ ).
2-49
opposite point: P l(−1) = (−1)l. Between these two end points, the lth polynomial has
l zeros. It is straightforward to use Eq. (2.169) to prove that these polynomials form
a full, orthogonal set of functions, with the following normalization rule:
+1
2
∫−1 P l (ξ )P l ′(ξ )dξ=
2l + 1
δll ′, (2.171)
so that any function f(ξ), defined on the segment [−1, +1], may be represented as a
unique series over the polynomials55.
Thus, taking into account the additional division by r in Eq. (2.162), the general
solution of any axially symmetric Laplace problem may be represented as
∞
⎛ bl ⎞
ϕ(r , θ ) = ∑⎜⎝al r l + ⎟P l (cos θ ). (2.172)
l=0
r l +1 ⎠
Please note a strong similarity between this solution and Eq. (2.112) for the 2D
Laplace problem in the polar coordinates. However, apart from the difference in
angular functions, there is also a difference (by one) in the power of the second radial
function, and this difference immediately shows up in problem solutions.
Indeed, let us solve a problem similar to that shown in figure 2.15: find the electric
field around a conducting sphere of radius R, placed into an initially uniform
external field E0 (whose direction I will take for axis z)—see figure 2.24a. If we select
the arbitrary constant in the electrostatic potential so that ϕ∣z=0 = 0, then in
Eq. (2.172) we should take a0 = b0 = 0. Now, just as has been argued for the
cylindrical case, at r ≫ R the potential should approach that of the uniform field:
ϕ → −E 0z = −E 0r cos θ , (2.173)
so that in Eq. (2.172), only one of the coefficients al survives: al = −E0δl1. As a result,
from the boundary condition on the surface, ϕ(R, θ) = 0, we obtain:
⎛ b ⎞ bl
0 = ⎜ −E 0R + 12 ⎟ cos θ + ∑ P (cos θ ).
l +1 l (2.174)
⎝ R ⎠ l⩾2
R
Now repeating the argumentation that led to Eq. (2.117), we may conclude that
Eq. (2.174) is satisfied if
bl = E 0R3δl ,1, (2.175)
so that, finally, Eq. (2.172) is reduced to

⎛ R3 ⎞
ϕ = −E 0⎜r − 2 ⎟ cos θ . (2.176)
⎝ r ⎠
55
This is the reason why, at least for the purposes of this course, there is no reason in pursuing (more complex)
solutions to Eq. (2.168) for non-integer values of l, which were mentioned in the previous footnote.
2-50
Figure 2.24. Conducting sphere in a uniform electric field: (a) the problem’s geometry, and (b) the
equipotential surface pattern given by Eq. (2.176). The pattern is qualitatively similar but quantitatively
different from that for the conducting cylinder in a perpendicular field—cf figure 2.15.
This distribution, shown in figure 2.24b, is very much similar to Eq. (2.117) for
the cylindrical case (cf figure 2.15b, with the account for a different plot orientation),
but with a different power of the radius in the second term. This difference leads to a
quantitatively different distribution of the surface electric field:
∂ϕ
En = − = 3E 0 cos θ , (2.177)
∂r r=R
so that its maximal value is a factor of 3 (rather than 2) larger than the external field.
Now let me briefly (mostly just for the reader’s reference) mention the Laplace
equation solutions in the general case (with no axial symmetry). If the free space
surrounds the origin from all sides, the solutions to Eq. (2.165) have to be 2π-
periodic, and hence ν = n = 0, ±1, ±2, … Mathematics says that Eq. (2.164) with
integer ν = n and a fixed integer l has a solution only for a limited range of n:56
−l ⩽ n ⩽ +l. (2.178)
These solutions are called the associated Legendre functions (generally, they are not
polynomials!) For n ⩾ 0, these functions may be defined via the Legendre
polynomials, using the following formula57:
dn
P ln(ξ ) = ( −1)n(1 − ξ 2 )n/2 P l (ξ ). (2.179)
dξ n
On the segment ξ ∈ [−1, +1], each set of the associated Legendre functions with a
fixed index n and non-negative l form a full, orthogonal set, with the normalization
relation,
56
In quantum mechanics, the letter n is typically reserved to be used for the ‘principal quantum number’, while the
azimuthal functions are numbered by index m. However, here I will keep using n as their index, because for this
course’s purposes, this seems more logical in view of the similarity of the spherical and cylindrical functions.
57
Note that some texts use different choices for the front factor (called the Condon–Shortley phase) in the
functions P lm , which do not affect the final results for the spherical harmonics Ylm—see below.
2-51
+1
2 (l + n )!
∫−1 P ln(ξ )P ln′ (ξ )dξ = δll ′, (2.180)
2l + 1 (l − n )!
that is evidently a generalization of Eq. (2.171).
Since these relations may seem a bit intimidating, let me write down explicit
expressions for a few P ln (cosθ) with the lowest values of l and n ⩾ 0, which are most
important for applications.
l = 0: P 00(cos θ ) = 1; (2.181)
⎧P 0(cos θ ) = cos θ ,
⎪
l = 1: ⎨ 11 (2.182)
⎩P 1 (cos θ ) = −sin θ ;
⎪
⎧P 0(cos θ ) = (3 cos2 θ − 1)/2,

⎪ 2
l = 2: ⎨P 21(cos θ ) = −2 sin θ cos θ , (2.183)
⎪ 2
⎩P 2 (cos θ ) = −3 cos2 θ .
The reader should agree there is not much to fear is these functions—they are
just the sums of products of functions cos θ ≡ ξ and sinθ ≡ (1 − ξ2)1/2. Figure 2.25
shows the plots of a few lowest functions P ln (ξ).
Figure 2.25. A few lowest associated Legendre functions. (Adapted from an original by Geek3, available at
https://en.wikipedia.org/wiki/Associated_Legendre_polynomials, under the GNU Free Documentation
License.)
2-52
Using the associated Legendre functions, the general solution (2.162) to the
Laplace equation in the spherical coordinates may be expressed as
∞
⎛ b ⎞
l
ϕ(r , θ , φ) = ∑⎜al r l + l +l 1 ⎟ ∑ P ln(cos θ )F n(φ),
l=0
⎝ r ⎠n=0 (2.184)
F n(φ) = cn cos nφ + sn sin nφ .
Since the difference between angles θ and φ is somewhat artificial, physicists prefer
to think not about functions P and F in separation, but directly about their products
that participate in this solution58.
As a rare exception for my courses, in order to save time, I will skip giving an
example of using the associated Legendre functions here, because quite a few such
examples of that will be given in the quantum mechanics part of these series.
2.9 Charge images

So far, we have discussed various methods of solution of the Laplace boundary
problem (2.35). Let us now move on to the discussion of its generalization, the
Poisson equation (1.41), that we need when besides the conductors, we also have
‘stand-alone’ charges with a known spatial distribution ρ(r). (This will also allow us,
better equipped, to revisit the Laplace problem again in the next section.)
Let us start with a somewhat limited, but very useful charge image (or ‘image
charge’) method. Consider a very simple problem: a single point charge near a
conducting half-space—see figure 2.26.
Figure 2.26. The simplest problem readily solvable by the charge image method. Point colors are used, as
before, to denote the charges of the original (red) and opposite (blue) sign.
58
In quantum mechanics, it is more convenient to use a slightly different, alternative set of basic functions,
namely the following complex functions, called the spherical harmonics,
⎡ 2l + 1 (l − n)! ⎤1/2
Yln(θ , φ) ≡ ⎢ ⎥ P ln(cos θ )einφ,
⎣ 4π (l + n)! ⎦
which are defined for both positive and negative n (within the limits −l ⩽ n ⩽ +l )—see, e.g. Part QM sections
3.6 and 5.6. Note again that in that field, the index n is traditionally denoted as m, and called the magnetic
quantum number.
2-53
Let us prove that its solution, above the conductor’s surface (z ⩾ 0), may be
represented as,
1 ⎛q q⎞ q ⎛ 1 1 ⎞
ϕ(r) = ⎜ − ⎟= ⎜ − ⎟, (2.185)
4πε0 ⎝ r1 r2 ⎠ 4πε0 ⎝ r − r′ r − r″ ⎠
or in a more explicit form, using the cylindrical coordinates shown in figure 2.26:
q ⎛ 1 1 ⎞
ϕ(r) = ⎜ − ⎟, (2.186)
4πε0 ⎝ [ρ 2 + (z − d )2 ]1/2
[ρ 2 + (z + d )2 ] ⎠
1/2
where ρ is the distance of the observation point from the ‘vertical’ line on which the
charge is located. Indeed, this solution evidently satisfies both the boundary
condition ϕ = 0 at the surface of the conductor (z = 0), and the Poisson equation
(1.41), with the single δ-functional source at point r′ = {0, 0, d} on its right-hand
side, because the second singularity of the solution, at point rʺ = {0, 0, −d}, is outside
the region of the solution’s validity (z ⩾ 0).
Physically, this solution may be interpreted as the sum of the fields of the actual
charge (+q) at point r′, and an equal but opposite charge (−q) at the ‘mirror image’
point rʺ (figure 2.26). This is the basic idea of the charge image method. Before
moving on to more complex problems, let us discuss the situation shown in figure
2.26 in a little bit more detail, due to its fundamental importance. First, we can use
Eqs. (2.3) and (2.186) to calculate the surface charge density:
∂ϕ q ∂⎛ 1 1 ⎞
σ = − ε0 =− ⎜ 2 − ⎟
∂z 4π ∂z ⎝ [ρ + (z − d )2 ]1/2 [ρ 2 + (z + d )2 ]1/2 ⎠z=0
z=0 (2.187)
q 2d
=− .
4π (ρ + d 2 )3/2
2
The total surface charge is

∞ ∞
q d
Q= ∫A σ d 2r = 2π∫0 σ (ρ)ρ dρ = −
2
∫0 ( ρ 2
+ d 2 )3/2
2 ρ dρ . (2.188)
This integral may be easily worked out using the substitution ξ ≡ ρ2/d2 (giving dξ =
2ρ dρ/d2):
∞
q dξ
Q=−
2
∫0 (ξ + 1)3/2
= −q. (2.189)
This result is very natural, because the conductor ‘wants’ to bring as much surface
charge from its interior to the surface as necessary to fully compensate the initial
charge (+q) and hence to kill the electric field at large distances as efficiently as
possible, hence reducing the total electrostatic energy (1.65) to the lowest possible
value.
2-54
For a better feeling of this polarization charge of the surface, let us take our
calculations to the extreme, to q equal to one elementary change e, and place a
particle with this charge (for example, a proton) at a macroscopic distance—say
1 m—from the conductor’s surface. Then, according to Eq. (2.189), the total polar-
ization charge of the surface equals that of an electron, and according to Eq. (2.187), its
spatial extent is of the order of d2 = 1 m2. This means that if we consider a much
smaller part of the surface, ΔA ≪ d2, its polarization charge magnitude ΔQ = σΔA is
much less than one electron! For example, Eq. (2.187) shows that the polarization
charge of quite a macroscopic area ΔA = 1 cm2 right under the initial charge (ρ = 0) is
eΔA/2πd2 ≈ 1.6 × 10−5 e. Can this be true, or our theory is somehow limited to the
charges q much larger than e? (After all, the theory is substantially based on the
approximate macroscopic model (2.1); maybe this is the culprit?)
Surprisingly enough, the answer to this question has become clear (at least to
some physicists:-) only as late as in the mid-1980s when several experiments
demonstrated, and theorists accepted, some rather grudgingly, that the usual
polarization charge formulas are valid for elementary charges as well, i.e. the
polarization charge ΔQ of a macroscopic surface area can indeed be less than e. The
underlying reason for this paradox is the nature of the polarization charge of a
conductor’s surface: as was discussed in section 2.1, it is due not to new charged
particles brought into the conductor (such charge would be in fact quantized in the
units of e), but to a small shift of the free charges of a conductor by a very small
distance from their equilibrium positions that they had in the absence of the external
field induced by charge q. This shift is not quantized, at least on the scale relevant for
our problem, and neither is ΔQ.
This understanding has opened a way toward the invention and experimental
demonstration of several new devices including so-called single-electron transistors59,
which are used, in particular, for ultrasensitive measurement of polarization charges
as small as ∼10−6 e. Another important class of single-electron devices is the current
standards based on the fundamental relation I = −ef, where I is the dc current
carried by electrons transferred with the frequency f. The experimentally achieved60
relative accuracy of such standards is of the order of 10−7, and is not too far from
that provided by the competing approach based on the combination of the
Josephson effect and the quantum Hall effect61.
Second, let us find the potential energy U of the charge-to-surface interaction. For
that we may use the value of the electrostatic potential (2.185) at the point of the real
charge (r = r′), of course ignoring the infinite potential created by the charge itself, so
that the remaining potential is that of the image charge
59
Actually, this term (for which the author of these notes should be blamed:-) is misleading: the operation of
the ‘single-electron transistor’ is based on the interplay of discrete charges (multiples of e) transferred between
conductors, and sub-single-electron polarization charges—see, e.g. [7].
60
See, e.g. [8, 9].
61
See [10].
2-55
1 q
ϕimage(r′) = − . (2.190)
4πε0 2d
Looking at the definition of the electrostatic potential, given by Eq. (1.31), it may be
tempting to immediately write U = qϕimage = −(1/4πε0)(q2/2d) (WRONG!), but this
would be incorrect. The reason is that potential ϕimage is not independent of q, but is
actually induced by this charge. This is why the correct approach is to calculate U
from Eq. (1.61), with just one term:
1 1 q2
U= qϕimage = − , (2.191)
2 4πε0 4d
giving a twice lower energy than in the wrong result cited above. In order to double-
check Eq. (2.191), and also obtain a better feeling of the factor ½ that distinguishes it
from the wrong guess, we can recalculate U as the integral of the force exerted on the
charge by the conductor (i.e. in our formalism, by the image charge):
d d
1 q2 1 q2
U=− ∫∞ F (z )dz =
4πε0
∫∞ (2z )2
dz = −
4πε0 4d
. (2.192)
This calculation clearly accounts for the gradual build-up of the force F, as the real
charge is brought from afar (where we have opted for U =0) toward the surface.
This result has several important applications. For example, let us plot the
electrostatic energy U for an electron, i.e. particle with charge q = −e, near a metallic
surface, as a function of d. For that, we may use Eq. (2.191) until our macroscopic
model (2.1) becomes invalid, and U transitions to some negative constant value (−ψ)
inside the conductor—see figure 2.27a. Since our calculation was for an electron
with zero potential energy at infinity, at relatively low temperatures, kBT ≪ ψ,
electrons in a metal may occupy only the states with energies below −ψ (the so-called
Fermi level62). The positive constant ψ is called the workfunction, because it
describes the smallest work necessary to remove the electron from a metal. As
was discussed in section 2.1, in good metals the electric field screening takes place at
interatomic distances a0 ≈ 10−10 m. Plugging d = a0 and q = −e into Eq. (2.191), we
Figure 2.27. (a) Schematic representations of the origin of the workfunction and (b) the field emission of
electrons.
62
More discussion of these states may be found in Part SM sections 3.3 and 6.4.
2-56
obtain ψ ≈ 6 × 10−19 J ≈ 3.5 eV. This crude estimate is in a surprisingly good

agreement with the experimental values of the workfunction, ranging between 4 and
5 eV for most metals63.
Next, let us consider the effect of an additional uniform external electric field E0
applied normally to a metallic surface, and toward it, on this potential profile. We
can add the potential energy the field gives to the electron at distance d from the
surface, Uext = −eE0d, to that created by the image charge. (As we know from
Eq. (1.53), since field E0 is independent of the electron’s position, its recalculation
into the potential energy does not require the coefficient ½.) As the result, the
potential energy of an electron near the surface becomes
1 e2
U (d ) = −eE 0d − , for d ≳ a 0, (2.193)
4πε0 4d
with a similar crossover to U = −ψ inside the conductor—see figure 2.27b. One can
see that at the appropriate sign, and a sufficient magnitude of the applied field, it
lowers the potential barrier that prevented the electron from leaving the conductor.
At E0 ∼ ψ/a0 (for metals, ∼1010 V m−1), this suppression becomes so strong that
electrons with energies at, and just below the Fermi level start quantum-mechanical
tunneling through the remaining thin barrier. This is the field emission effect, which
is used in vacuum electronics to provide efficient cathodes that do not require
heating to high temperatures64.
Returning to the basic electrostatics, let us consider some other geometries where
the method of charge images may be effectively applied. First, let us consider a right
corner (figure 2.28a). Reflecting the initial charge in the vertical plane we obtain the
image shown in the top left corner of the panel. This image makes the boundary
condition ϕ = const satisfied on the vertical surface of the corner. However, in order
for the same to be true on the horizontal surface, we have to reflect both the initial
charge and the image charge in the horizontal plane, flipping their signs. The final
configuration of four charges, shown in figure 2.28a, evidently satisfies all the
boundary conditions. The resulting potential distribution may be readily written as
the generalization of Eq. (2.185). From it, the electric field and electric charge
distributions, and the potential energy and forces acting on the charge may be
calculated exactly as above—an easy exercise, left for the reader.
Next, consider a corner with angle π/4 (figure 2.28b). Here we need to repeat the
reflection operation not two but four times before we arrive at the final pattern of
eight positive and negative charges. (Any attempt to continue this process would
lead to an overlap with the already existing charges.) This reasoning may be readily
63
More discussion of the workfunction, and its effect on the electrons’ kinetics, is given in Part SM section 6.4.
64
The practical use of such ‘cold’ cathodes is affected by the fact that, as follows from our discussion in section
2.4, any nanoscale surface irregularity (a protrusion, an atomic cluster, or even a single ‘adatom’ stuck to the
surface) may cause a strong increase of the local field well above the applied field E0, making the emission
reproducibility and stability in time significant challenges. In addition, the impact-ionization effects may lead
to avalanche-type electric breakdown already at fields as low as ∼3 × 106 V m−1.
2-57
Figure 2.28. Charge images for (a, b) the corners with angles π and π/2, (c) a plane capacitor, and (d) a
rectangular box. (e) The typical equipotential surfaces for the last system.
extended to a corner of angle β = π/n, with any integer n, which requires 2n charges
(including the initial one) to satisfy all the boundary conditions.
Some configurations require an infinite number of images, but are still tractable.
The most important of them is a system of two parallel conducting surfaces, i.e. a
plane capacitor of an infinite area (figure 2.28c). Here the repeated reflection leads to
an infinite system of charges ±q at points
x j± = 2aj ± d , (2.194)
where d (with 0 < d < a) is the position of the initial charge, and j an arbitrary
integer. The resulting infinite sum for the potential of the real charge q, created by
the field of its images,
⎡ ⎤
1 ⎢ q ±q ⎥
ϕ(d ) = − + ∑∑
4πε0 ⎢⎣ 2d j≠0 ± d − x ± ⎥
j ⎦
(2.195)
⎧ ∞ ⎫
q ⎪1 d2 1 ⎪
≡− ⎨ + 3∑ 2 ⎬,
4πε0 ⎩⎪ 2d a j = 1 j [j − (d / a )2 ] ⎭
⎪
is converging (in its last form) very fast. For example, the exact value, ϕ(D/2) =
−2ln 2(q/4πε0D), differs by less than 5% from the approximation using just the first
term of the sum.
2-58
The same method may be applied to 2D (cylindrical) and 3D rectangular

conducting boxes that require, respectively, a 2D or 3D infinite rectangular lattices
of images; for example in a 3D box with sides a, b, and c, charges ±q are located at
points (figure 2.28d)
r±jkl = 2ja + 2kb + 2lc ± r′ , (2.196)
where r′ is the location of the initial (real) charge, and j, k, and l are arbitrary
integers. Figure 2.28e shows the results of summation of the potentials of such
charge set, including the real one, in a 2D box (within the plane of the real charge).
One can see that the equipotential surfaces, concentric near the charge, are naturally
leaning along the conducting walls of the box, which should be equipotential.
Even more surprisingly, the image charge method works very efficiently not only
for rectilinear geometries, but also for spherical ones. Indeed, let us consider a point
charge q at distance d from the center of a conducting, grounded sphere of radius R
(figure 2.29a), and try to satisfy the boundary condition ϕ = 0 for the electrostatic
potential on sphere’s surface using an imaginary charge q′ located at some point
located beyond the surface, i.e. inside the sphere.
Figure 2.29. Method of charge images for a conducting sphere: (a) the concept and (b) the resulting potential
distribution in the central plane containing the charge, for the particular case d = 2R.
From the problem’s symmetry, it is clear that the point should be at the line
passing through the real charge and the sphere’s center, at some distance d′ from the
center. Then the total potential created by the two charges at an arbitrary point with
r ⩾ R (figure 2.29a) is
1 ⎡ q q′ ⎤
ϕ(r , θ ) = ⎢ 2 + 2 ⎥. (2.197)
4πε0 ⎣ (r + d − 2rd cos θ )
2 1/2
(r + d ′ − 2rd ′ cos θ ) ⎦
2 1/2
This expression shows that we can make the two fractions to be equal and opposite
at all points on the sphere’s surface (i.e. for any θ at r = R), if we take65
65
In geometry, such points, with dd′ = R2, are referred to as the result of mutual inversion in a sphere of radius R.
2-59
R2 R
d′ = , q′ = − q. (2.198)
d d
Since the solution to any Poisson boundary problem is unique, Eqs. (2.197) and (2.198)
give us such a solution for this problem. Figure 2.29b shows a typical equipotential
pattern calculated using Eqs. (2.197) and (2.198). It is surprising how formulas that
simple may describe such a non-trivial field distribution.
Now let us calculate the total charge Q induced by charge q on the grounded
sphere’s surface. We could do this, as we have done for the conducting plane, by the
brute-force integration of the surface charge density σ = −ε0∂ϕ/∂r∣r = R. It is more
elegant, however, to use the following Gauss law argument. The expression (2.197)
is valid (at r ⩾ R) regardless of whether we are dealing with our real problem (charge
q and the conducting sphere) or with the equivalent charge configuration (with the
point charges q and q′, but no sphere at all). Hence, according to Eq. (1.16), the
Gaussian integral over a surface with radius r = R + 0, and the total charge inside
the sphere should be also the same. Hence we immediately obtain
R
Q = q′ = − q. (2.199)
d
The similar argumentation may be used to find the charge-to-sphere interaction

force:
q′ q2 R 1
F = qE image(d ) = q = −
4πε0(d − d ′) 2
4πε0 d (d − R2 / d )2
(2.200)
q2 Rd
=− .
4πε0 (d − R2 )2
2
(Note that this expression is legitimate only at d > R.) At large distances, d ≫ R, this
attractive force decreases as 1/d3. This unusual dependence arises because, as
Eq. (2.198) specifies, the induced charge of the sphere, responsible for the force, is
not constant but decreases as 1/d. In the next chapter we will see that such force is
also typical for the interaction between a point charge and a point dipole.
All the previous formulas were for a sphere that is grounded to keep its potential
equal to zero. But what if we keep the sphere galvanically insulated, so that its net
charge is fixed, for example, equals zero? Instead of solving this problem from
scratch, let us use (again!) the linear superposition principle. For that, we may add to
the previous problem an additional charge, equal to −Q = −q′, to the sphere, and
argue that this addition gives, at all points, an additional, spherically symmetric
potential that does not depend of the potential induced by charge q, and was
calculated in section 1.2—see Eq. (1.19). For the interaction force, such addition
yields
2-60
qq′ qq′ q2 ⎡ Rd R⎤
F= + = − ⎢ − ⎥. (2.201)
4πε0(d − d ′)2 4πε0d 2 4πε0 ⎣ (d 2 − R2 )2 d3 ⎦
At large distances, the two terms proportional to 1/d3 cancel each other, giving F ∝
1/d5. Such a rapid force decay is due to the fact that the field of the uncharged sphere
is equivalent to that of two (equal and opposite) induced charges +q′ and −q′, and
the distance between them (d′ = R2/d) tends to zero at d → ∞. The potential energy
of such an interaction behaves as U ∝ 1/d6 at d → ∞; in the next chapter we will see
that this is the general law of the induced dipole interaction.
2.10 Green’s functions

I have spent so much time/space discussing the potential distributions created by a
single point charge in various conductor geometries, because for any of the
geometries, the generalization of these results to the arbitrary distribution ρ(r) of
free charges is straightforward. Namely, if a single charge q, located at point r′,
creates the electrostatic potential
1
ϕ(r) = qG (r , r′), (2.202)
4πε0
then, due to the linear superposition principle, an arbitrary charge distribution
(either discrete or continuous) creates the potential
1 1
ϕ(r) = ∑
4πε0 j
qj G (r , rj ) =
4πε0
∫ ρ(r′)G(r, r′)d 3r′. (2.203)
The function G(r, r′) is called the (spatial) Green’s function—the notion is very
fruitful and hence popular in all fields of physics66. Evidently, as Eq. (1.35) shows, in
the unlimited free space
1
G (r , r′) = , (2.204)
r − r′
i.e. the Green’s function depends only on one scalar argument—the distance
between the field observation point r and the field-source (charge) point r′.
However, as soon as there are conductors around, the situation changes. In this
course I will only discuss the Green’s functions that vanish as soon as the radius-
vector r points to the surface of any conductor67:
G (r , r′) r∈A = 0. (2.205)
With this definition, it is straightforward to deduce the Green’s functions for the
solutions of the last section’s problems in which conductors were grounded (ϕ = 0).
66
See, e.g. Part CM section 5.1, Part QM sections 2.2 and 7.4, and Part SM section 5.5.
67
G thus defined is sometimes called the Dirichlet function.
2-61
For example, for a semi-space z ⩾ 0 limited by a conducting plane (figure 2.26),

Eq. (2.185) yields
1 1
G= − , with ρ″ = ρ′ and z″ = −z′ . (2.206)
r − r′ r − r″
We see that in the presence of conductors (and, as we will see later, any other
polarizable media), the Green’s function may depend not only on the difference r − r′,
but on each of these two arguments in a specific way.
So far, this looks just like re-naming our old results. The really non-trivial result
of this formalism for electrostatics is that, somewhat counter-intuitively, the
knowledge of the Green’s function for a system with grounded conductors (figure
2.30a) enables the calculation of the field created by voltage-biased conductors
(figure 2.30b), with the same geometry.
Figure 2.30. The Green’s function method allows the solution of a simpler boundary problem (a) to be used to
find the solution of a more complex problem (b), for the same conductor geometry.
In order to show this, let us use the so-called Green’s theorem of the vector
calculus68. The theorem states that for any two scalar, differentiable functions f(r)
and g(r), and any volume V,
∫V (f ∇2 g − g∇2 f )d 3r = ∮S (f ∇g − g ∇f )nd 2r, (2.207)
where S is the surface limiting the volume. Applying the theorem to the electrostatic
potential ϕ(r) and the Green’s function G (also considered as a function of r), let us
use the Poisson equation (1.41) to replace ∇2ϕ with (−ρ/ε0), and notice that G,
considered as a function of r, obeys the Poisson equation with the δ-functional
source:
∇2 G (r , r′) = −4πδ(r − r′). (2.208)
(Indeed, according to its definition (2.202), this function may be formally

considered as the field of a point charge q = 4πε0.) Now swapping the notation
68
See, e.g. Eq. (A.80). Actually, this theorem is a ready corollary of the better-known divergence (‘Gauss’)
theorem, Eq. (A.79).
2-62
of the radius-vectors, r ↔ r′, and using the Green’s function symmetry, G(r, r′) =
G(r′, r)69, we obtain
⎛ ⎞
−4πϕ(r) − ∫V ⎜⎝− ρε(r0′) ⎟⎠G(r, r′)d 3r′
(2.209)
⎡ ∂G (r , r′) ∂ϕ(r′) ⎤ 2
∮
= ⎢ϕ(r′)
S⎣ ∂n′
− G (r , r′)
∂n′ ⎦
⎥d r′ .
Let us apply this relation to the volume V of free space between the conductors,
and the boundary S drawn immediately outside of their surfaces. In this case, by its
definition, the Green’s function G(r, r′) vanishes at the conductor surface (r ∈ S)—
see Eq. (2.205). Now changing the sign of ∂n′ (so that it would be the outer normal
for conductors, rather than free space volume V), dividing all terms by 4π, and
partitioning the total surface S into the parts (numbered by index j) corresponding to
different conductors (possibly, kept at different potentials ϕk), we finally arrive at the
famous result70:
1
ϕ(r) =
4πε0
∫V ρ(r′)G(r, r′)d 3r′ + 41π ∑ϕk∮S k
∂G (r , r′) 2
∂n′
d r′ . (2.210)
k
While the first term on the right-hand side of this relation is a direct and evident
expression of the superposition principle, given by Eq. (2.203), the second term is
highly non-trivial: it describes the effect of conductors with non-zero potentials ϕk
(figure 2.30b), using the Green’s function calculated for a similar system with
grounded conductors, i.e. with all ϕk = 0 (figure 2.29a). Let me emphasize that since
our volume V excludes conductors, the first term on the right-hand side of
Eq. (2.210) includes only the stand-alone charges in the system (in figure 2.30,
marked q1, q2, etc), but not the surface charges of the conductors—which are taken
into account, implicitly, by the second term.
In order to illustrate what a powerful tool Eq. (2.210) is, let us use to calculate the
electrostatic field in two systems. In the first of them, a planar, circular, conducting
disk, separated with a very thin cut from the remaining conducting plane, is biased
with potential ϕ = V, while the rest of the plane is grounded—see figure 2.31. If the
width of the gap between the circle and rest of the plane is negligible, we may apply
Eq. (2.210) without stand-alone charges, ρ(r′) = 0, and the Green’s function for the
uncut plane—see Eq. (2.206)71. In the cylindrical coordinates, with the origin at the
disk’s center (figure 2.31), the function is
69
This symmetry, evident for the particular cases (2.204) and (2.206), may be readily proved for the general
case, by applying Eq. (2.207) to functions f (r) ≡ G(r, r′) and g(r) ≡ G(r, rʺ). With this substitution, the left-hand
side becomes equal to −4π [G(rʺ, r′) − G(r′, rʺ)], while the right-hand side is zero, due to Eq. (2.205).
70
In some textbooks, the sign before the surface integral is negative, because their authors use the outer normal
to the free-space region V rather than that occupied by conductors—as I do.
71
Indeed, if all parts of the cut plane are grounded, a narrow cut does not change the field distribution, and
hence the Green’s function, significantly.
2-63
Figure 2.31. A voltage-biased conducting circle separated from the rest of a conducting plane.
1
G (r , r′) =
[ρ + ρ′ − 2ρρ′ cos(φ − φ′) + (z − z′)2 ]1/2
2 2
(2.211)
1
− 2 .
[ρ + ρ′ − 2ρρ′ cos(φ − φ′) + (z + z′)2 ]1/2
2
(The sum of the first three terms under the square roots in Eq. (2.211) is just the
squared distance between the horizontal projections ρ and ρ′ of vectors r and r′ (or
rʺ), correspondingly, while the last terms are the squares of their vertical spacings.)
Now we can readily calculate the derivative participating in Eq. (2.210):
∂G ∂G 2z
= = . (2.212)
∂n′ S
∂z′ z ′=+0
[ρ + ρ′ − 2ρρ′ cos(φ − φ′) + z 2 ]3/2
2 2
Due to the axial symmetry of the system, we can take φ for zero. With this, Eqs.
(2.210) and (2.212) yield
V
ϕ=
4π
∮S ∂G∂(rn,′ r′) d 2r′
2π R (2.213)
Vz ρ′dρ′
=
2π
∫0 dφ ′∫0 ( ρ 2
+ ρ′2
− 2 ρρ′ cos φ′ + z 2 ) 3/2
.
This integral is not too pleasing, but may be readily worked out for points on the
symmetry axis (ρ = 0):
R
ρ′dρ′ V R 2 /z 2
dξ ⎡ z ⎤
ϕ = Vz ∫0 2
(ρ′ + z )2 3/2
=
2
∫0 (ξ + 1)3/2
= V ⎢
⎣
1 − ⎥ . (2.214)
(R2 + z 2 )1/2 ⎦
This expression shows that if z → 0, the potential tends to V (as it should), while at z
≫ R,
R2
ϕ→V . (2.215)
2z 2
Now, let us use the same Eq. (2.210) to solve the (in:-)famous problem of the cut
sphere (figure 2.32). Again, if the gap between the two conducting semi-spheres is
2-64
Figure 2.32. A system of two separated, oppositely biased semi-spheres.
very thin (t ≪ R), we may use the Green’s function for the grounded (and uncut)
sphere. For a particular case r′ = dnz, this function is given by Eqs. (2.197) and
(2.198); generalizing the former relation for an arbitrary direction of the vector r′, we
obtain
1
G=
[r + r′ − 2rr′ cos γ ]1/2
2 2
(2.216)
R / r′
− , for r , r′ ⩾ R ,
[r + (R / r′) − 2r(R2 / r′)cos γ ]1/2
2 2 2
where γ is the angle between the vectors r and r′, and hence rʺ—see figure 2.32.
Now, calculating the Green’s function’s derivative,
∂G (r 2 − R 2 )
=− , (2.217)
∂r′ r ′=R +0
R[r 2 + R2 − 2Rr cos γ ]3/2
and plugging it into Eq. (2.210), we see that the integration is again easy only for the
field on the symmetry axis (r = rnz, γ = θ), giving:
V⎡ z 2 − R2 ⎤
ϕ= ⎢1 − ⎥. (2.218)
2⎣ z(z 2 + R2 )1/2 ⎦
For z → R, ϕ → V/2 (just checking), while for z ≫ R,

3R2
ϕ→V . (2.219)
4z 2
As will be discussed in the next chapter, such a field is typical for an electric dipole.
2.11 Numerical methods

Despite the richness of analytical methods, for many boundary problems (in
particular in geometries without a high degree of symmetry), numerical methods
are the only way to the solution. Despite the current abundance of software codes
2-65
and packages offering their automatic numerical solution72, it is important for every
educated physicist to understand ‘what is under their hood’, at least because most
universal programs exhibit mediocre performance in comparison with custom codes
written for particular problems, and sometimes do not converge at all, in particular
for fast-changing (say, exponential) functions. The very brief discussion presented
here73 is a (hopefully, useful) fast glance under the hood, though it is certainly
insufficient for professional numerical research work.
The simplest of the numerical approaches to the solution of partial differential
equations, such as the Poisson or the Laplace equations (1.41) and (1.42), is the
finite-difference method74, in which the sought function of N scalar arguments f(r1,
r2,…rN) is represented by its values in discrete points of a rectangular grid
(frequently called the mesh) of the corresponding dimensionality—see figure 2.33.
Each partial second derivative of the function is approximated by the formula that
readily follows from the linear approximations of the function f and then its partial
derivatives—see figure 2.33a:
∂ 2f ∂ ⎛ ∂f ⎞ 1 ⎛ ∂f ∂f ⎞ 1⎡f − f f − f← ⎤
= ⎜ ⎟≈ ⎜ − ⎟≈ ⎢→ − ⎥
∂rj2 ∂rj ⎝ ∂rj ⎠ h ⎝ ∂rj ∂rj ⎠ h⎣ h h ⎦
rj +h /2 rj −h /2 (2.220)
f + f← − 2f
= → ,
h2
where f→ ≡ f(rj + h) and where f← ≡ f(rj − h). (The relative error of this
approximation is of the order of h4∂4f/∂rj4.) As a result, the action of a 2D
Laplace operator on the function f may be approximated as
∂ 2f ∂ 2f f→ + f← − 2f f↑ + f↓ − 2f f→ + f← + f↑ + f↓ − 4f
+ = + = , (2.221)
∂x 2 ∂y 2 h2 h2 h2
and of the 3D operator, as
Figure 2.33. The general idea of the finite-difference method in (a) one, (b) two, and (c) three dimensions.
72
See, for example, appendix A, sections A.16(iii) and (iv).
73
It is almost similar to that given in Part CM section 8.5, and is reproduced here for the reader’s convenience,
being illustrated with examples from this (Part EM) course.
74
For more details see, e.g. [11].
2-66
∂ 2f ∂ 2f ∂ 2f f→ + f← + f↑ + f↓ + f⊗ + f× − 6f
+ + = . (2.222)
∂x 2 ∂y 2 ∂z 2 h2
(The notation used in Eqs. (2.221) and (2.222) should be clear from figures 2.33b
and c, respectively.)
Let us apply this scheme to find the electrostatic potential distribution inside a
cylindrical box with conducting walls and square cross-section a × a, using an
extremely coarse mesh with step h = a/2 (figure 2.34). In this case our function, the
electrostatic potential ϕ(x, y), equals zero at the side and bottom walls, and V0 at the
top lid, so that, according to Eq. (2.221), the 2D Laplace equation may be
approximated as
0 + 0 + V0 + 0 − 4ϕ
= 0. (2.223)
(a /2)2
The resulting value for the potential in the center of the box is ϕ = V0/4.
Figure 2.34. Numerically solving an internal 2D boundary problem for a conducting, cylindrical box with a
square cross-section, using a very coarse mesh (with h = a/2).
Surprisingly, this is the exact value! This may be proved by solving this problem
by the variable separation method, just as was done for the similar 3D problem in
section 2.5. The result is
∞
πnx πny 4V0 ⎧1, if n is odd,
ϕ(x , y ) = ∑cn sin sinh , cn = ×⎨ (2.224)
n=1
a a πn sinh(πn ) ⎩ 0, otherwise.
so that at the central point (x = y = a/2),

∞
4V0 sin [π (2j + 1)/2] sinh [π (2j + 1)/2]
ϕ= ∑ (2j + 1)sinh [π (2j + 1)]
π j=0
∞ (2.225)
2V ( −1) j
≡ 0∑ .
π j = 0 (2j + 1)cosh [π (2j + 1)/2]
The last series equals exactly to π/8, so that ϕ = V0/4.

For a similar 3D problem (a cubic box), with a similar 3D mesh, Eq. (2.222)
yields
2-67
0 + 0 + V0 + 0 + 0 + 0 − 6ϕ
= 0, (2.226)
(a /2)2
so that ϕ = V0/6. Unbelievably enough, this result is also exact! (This follows from
our variable separation result expressed by Eqs. (2.95) and (2.99).)
Though such exact results should be considered as a happy coincidence rather
than the general law, they still show that numerical methods, even with relatively
crude meshes, may be more computationally efficient than the ‘analytical’
approaches, such as the variable separation method with its infinite-sum results
that, in most cases, require computers anyway—at least for the result’s compre-
hension and analysis.
A more powerful (but also much more complex) approach is the finite-element
method in which the discrete point mesh, typically with triangular cells, is
(automatically) generated in accordance with the system geometry75. Such mesh
generators provide higher point concentration near sharp convex parts of conductor
surfaces, where the field concentrates and hence the potential changes faster, and
thus ensure a better accuracy-to-speed trade off than the finite-difference methods on
a uniform grid. The price to pay for this improvement is the algorithm’s complexity,
which makes manual adjustments much harder. Unfortunately, I do not have time
to go into the details of that method, and have to refer the reader to special literature
on this subject76.
2.12 Problems
Problem 2.1. Calculate the force (per unit area) exerted on a conducting surface by
an external electric field, normal to it. Compare the result with the definition of the
electric field, given by Eq. (1.6), and comment.
Problem 2.2. Certain electric charges QA and QB have been placed on two metallic,
concentric spherical shells—see the figure below. What is the full charge of each of
the surfaces S1–S4?
Problem 2.3. Calculate the mutual capacitance between the terminals of the lumped
capacitor circuit shown in the figure below. Analyze and interpret the result for
major particular cases.
75
See, e.g. Part CM figure 8.14.
76
See, e.g. [12] or [13].
2-68
Problem 2.4. Calculate the mutual capacitance between the terminals of the semi-
infinite lumped-capacitor circuit shown in the figure below, and the law of decay of
the applied voltage along the system. Analyze and interpret the result.
Problem 2.5. A system of two thin conducting plates is located over a ground plane as
shown in the figure below, where A1 and A2 are the areas of the indicated plane parts,
while d′ and dʺ are the distances between them. Neglecting the fringe effects, calculate:
(i) the effective capacitance of each plate, and
(ii) their mutual capacitance.
Problem 2.6. A wide, thin plane film, carrying a uniform electric charge density σ, is
placed inside a plane capacitor whose plates are connected with a wire (see the figure
below), and were initially electroneutral. Neglecting the edge effects, calculate the
surface charges of the plates, and the net force acting on the film (per unit area).
Problem 2.7. Following up the discussion of two weakly coupled spheres in section
2.2, find an approximate expression for the mutual capacitance (per unit length)
between two very thin, parallel wires, both with round cross-sections, but each with
2-69
its own radius. Compare the result with that for two small spheres, and interpret the
difference.
Problem 2.8. Use the Gauss law to calculate the mutual capacitance of the following
two-electrode systems, with the cross-section shown in figure 2.7 (reproduced
below):
(i) a conducting sphere inside a concentric spherical cavity in another conductor, and
(ii) a conducting cylinder inside a coaxial cavity in another conductor. (In this case,
we speak about the capacitance per unit length).
Compare the results with those obtained in section 2.2 using the Laplace equation
solution.
Problem 2.9. Calculate the electrostatic potential distribution around two barely
separated conductors in the form of coaxial, round cones (see the figure below), with
voltage V between them. Compare the result with that of a similar 2D problem, with
the cones replaced by plane-face wedges. Can you calculate the mutual capacitances
between the conductors in these systems? If not, can you estimate them?
Problem 2.10. Calculate the mutual capacitance between two rectangular, planar
electrodes of area A = a × l, with a small angle φ0 ≪ a/ρ0 between them—see the
figure below.
2-70
Problem 2.11. Using the results for a single thin round disk, obtained in section 2.4,
consider a system of two such disks at a small distance d ≪ R from each other—see
the figure below. In particular, calculate:
(i) the reciprocal capacitance matrix of the system,
(ii) the mutual capacitance between the disks,
(iii) the partial capacitance, and
(iv) the effective capacitance of one disk,
all in the first non-vanishing approximation in d/R ≪ 1. Compare the results (ii)–(iv)
and interpret their similarities and differences.
Problem 2.12.* Calculate the mutual capacitance (per unit length) between two
cylindrical conductors forming a system with the cross-section shown in the figure
below, in the limit t ≪ w ≪ R.
Hint: You may like to use elliptical (not ‘ellipsoidal’!) coordinates {α, β} defined by
the following equation:
x + iy = c cosh(α + iβ ), (2.227)
with the appropriate choice of the constant c. In these orthogonal 2D coordinates,
the Laplace operator is very simple77:
1 ⎛ ∂2 ∂2 ⎞
∇2 = ⎜ + ⎟.
c 2(cosh2 α − cos2 β ) ⎝ ∂α 2 ∂β 2 ⎠
77
This fact should not be surprising, because Eq. (2.227) is essentially the conformal map z = ccosh w , where
z = x + iy , and w = α + iβ —see the discussion in section 2.4.
2-71
Problem 2.13. Formulate 2D electrostatic problems that can be solved using each of
the following analytic functions of the complex variable z ≡ x + iy:
(i) w = ln z ,
(ii) w = z 1/2,
and solve these problems.
Problem 2.14. On each side of a cylindrical volume with a rectangular cross-section

a × b, with no electric charges inside it, the electric field is uniform, normal to the
side’s plane, and opposite to that on the opposite side—see the figure below.
Calculate the distribution of the electric potential inside the volume, provided that
the field magnitude on the vertical sides equals E. Suggest a practicable method to
implement such a potential distribution.
Problem 2.15. Complete the solution of the problem shown in figure 2.12, by
calculating the distribution of the surface charge of the semi-planes. Can you
calculate the mutual capacitance between the plates (per unit length)? If not, can you
estimate it?
Problem 2.16.* A straight, long, thin, round-cylindrical metallic pipe has been cut,
along its axis, into two equal parts—see the figure below.
(i) Use the conformal mapping method to calculate the distribution of the electro-
static potential, created by voltage V applied between the two parts, both outside
and inside the pipe, and of the surface charge.
(ii) Calculate the mutual capacitance between pipe’s halves (per unit length), taking
into account a small width 2t ≪ R of the cut.
Hints: In task (i) you may like to use the following complex function:
⎛R + z ⎞
w = ln ⎜ ⎟,
⎝R − z ⎠
while in task (ii) it is advisable to use the solution of the previous problem.
2-72
Problem 2.17. Solve task (i) of the previous problem using the variable separation
method, and compare the results.
Problem 2.18. Use the variable separation method to calculate the potential
distribution above the plane surface of a conductor, with a strip of width w
separated by very thin cuts, and biased with voltage V—see the figure below.
Problem 2.19. The previous problem is now slightly modified: the cut-out and
voltage-biased part of the conducting plane is now not a strip, but a square with side
w. Calculate the potential distribution above the conductor’s surface.
Problem 2.20. Each electrode of a large plane capacitor is cut into long strips of
equal width w, with very narrow gaps between them. These strips are kept at the
alternating potentials, as shown in the figure below. Use the variable separation
method to calculate the electrostatic potential distribution, and explore the limit
w ≪ d.
Problem 2.21. Complete the cylinder problem started in section 2.7 (see figure 2.17),
for the cases when the top lid’s voltage is fixed as follows:
(i) V = V0J1(ξ11ρ/R)sin φ, where ξ11 ≈ 3.832 is the first root of the Bessel function
J1(ξ);
(ii) V = V0 = const.
2-73
For both cases, calculate the electric field at the centers of the lower and upper lids.
(For task (ii), an answer including series and/or integrals is satisfactory.)
Problem 2.22. Solve the problem shown in figure 2.21. In particular:

(i) calculate and sketch the distribution of the electrostatic potential inside the
system for various values of the ratio R/h, and
(ii) simplify the results for the limit R/h → 0.
Problem 2.23. Use the variable separation method to find the potential distribution
inside and outside a thin spherical shell of radius R, with a fixed potential
distribution: ϕ(R, θ, φ) = V0 sin θ cos φ.
Problem 2.24. A thin spherical shell carries charge with areal density σ = σ0cos θ.
Calculate the spatial distribution of the electrostatic potential and the electric field,
both inside and outside the shell.
Problem 2.25. Use the variable separation method to calculate the potential
distribution both inside and outside a thin spherical shell of radius R, separated
with a very thin cut, along plane z = 0, into two halves, with voltage V applied
between them—see the figure below. Analyze the solution; in particular, compare
the field at the axis z, for z > R, with Eq. (2.218).
Hint: You may like to use the following integral of a Legendre polynomial with odd
index l = 1, 3, 5…= 2n − 1:78
1
1 ⎛1 ⎞ ⎛ 3⎞ ⎛ 5⎞ ⎛3 ⎞ (2n − 3) !!
In ≡ ∫0 P2n−1(ξ )dξ = ⋅⎜ ⎟ ⋅ ⎜ − ⎟ ⋅ ⎜ − ⎟…⎜ − n⎟ ≡ ( −1)n−1 .
n! ⎝ 2 ⎠ ⎝ 2⎠ ⎝ 2⎠ ⎝2 ⎠ 2n(2n − 2) !!
Problem 2.26. Calculate, up to the terms O(1/r2), the long-range electric field induced
by a cut and voltage-biased conducting sphere—similar to that discussed in section
2.7 (see figure 2.32) and in the previous problem, but with the cut’s plane at an
arbitrary distance d < R from the center—see the figure below.
78
As a reminder, the double factorial (also called ‘semi-factorial’) operator (!!) is similar to the usual factorial
operator (!), but with the product limited to numbers of the same parity as its argument (in our particular case,
of the odd numbers in the nominator, and even numbers in the denominator).
2-74
Problem 2.27. A small conductor (in this context, called the single-electron island) is
placed between two conducting electrodes, with voltage V applied between them. The
gap between the island and one of the electrodes is so narrow that electrons may tunnel
quantum-mechanically through this ‘junction’—see the figure below. Neglecting
thermal excitations, calculate the equilibrium charge of the island as a function of V.
Hint: To solve this problem, you do not need to know much about the quantum-
mechanical tunneling between conductors, except that such tunneling of an electron,
followed by energy relaxation of the resulting excitations, may be considered as a
single inelastic (energy-dissipating) event. At negligible thermal excitations, such an
event takes place only when it decreases the total potential energy of the system79.
Problem 2.28. The system discussed in the previous problem is now generalized as
shown in the figure to right. If the voltage V′ applied between two bottom electrodes
is sufficiently large, electrons can successively tunnel through two junctions of this
system (called the single-electron transistor), carrying dc current between these
electrodes. Neglecting thermal excitations, calculate the region of voltages V and V′
where such a current is fully suppressed (Coulomb-blocked).
79
Strictly speaking, this model, implying negligible quantum-mechanical coherence of the tunneling events, is
correct only if the junction transparency is sufficiently low, so that its effective electric resistance is much higher
than the quantum unit of resistance (see, e.g. Part QM section 3.2), RQ ≡ πℏ/2e2 ≈ 6.5 kΩ—which is usually
the case.
2-75
Problem 2.29. Use the image charge method to calculate the surface charges induced
in the plates of a very broad plane capacitor of thickness D by a point charge q
separated from one of the electrodes by distance d.
Problem 2.30. Prove the statement, made in section 2.9, that the 2D boundary
problem, shown in the figure below, can be solved using a finite number of image
charges if the angle β equals π/n, where n = 1, 2,…
Problem 2.31. Use the image charge method to calculate the potential energy of the
electrostatic interaction between a point charge placed in the center of a spherical
cavity that was carved inside a grounded conductor, and the cavity’s walls. Looking
at the result, could it be obtained in a simpler way (or ways)?
Problem 2.32. Use the method of images to find the Green’s function of the system
shown in the figure below, where the bulge on the conducting plane has the shape of
a semi-sphere of radius R.
Problem 2.33.* Use the spherical inversion, expressed by Eq. (2.198), to develop an
iterative method for more and more precise calculation of the mutual capacitance
between two similar metallic spheres of radius R, with centers separated by distance
d > 2R.
Problem 2.34.* A metallic sphere of radius R1, carrying electric charge Q, is placed
inside a spherical cavity of radius R2 > R1, cut inside another metal. Calculate the
2-76
electric force exerted on the sphere if its center is displaced by a small distance δ ≪
R1, R2 −R1 from that of the cavity—see the figure below.
Problem 2.35. Within the simple models of the electric field screening in conductors,
discussed in section 2.1, analyze the partial screening of the electric field of a point
charge q by a plane, uniform conducting film of thickness t ≪ λ, where λ is
(depending on the charge carrier statistics) either the Debye or the Thomas–Fermi
screening length—see, respectively, Eq. (2.8) or Eq. (2.10). Assume that the distance
d between the charge and the film is much larger than t.
Problem 2.36. Suggest a convenient definition of the Green’s function for 2D

electrostatic problems and calculate it for:
(i) the unlimited free space, and
(ii) the free space above a conducting plane.
Use the latter result to re-solve problem 2.18.
Problem 2.37. Calculate the 2D Green’s function for the free spaces:
(i) outside a round conducting cylinder, and
(ii) inside a round cylindrical hole in a conductor.
Problem 2.38. Solve task (i) of problem 2.16 (see also problem 2.17), using the
Green’s function method.
Problem 2.39. Solve the 2D boundary problem that was discussed in section 2.11
(figure 2.34) using:
(i) the finite difference method, with a finer square mesh, h = a/3, and
(ii) the variable separation method.
Compare the results at the mesh points, and comment.
References
[1] Hook J and Hall H 1991 Solid State Physics 2nd edn (Wiley)
[2] Ashcroft N and Mermin N 1976 Solid State Physics (Brooks Cole)
2-77
[3] Spiegel M et al 2009 Complex Variables 2nd edn (New York: McGraw-Hill)
[4] Grivet P 1972 Electron Optics 2nd edn (Pergamon)
[5] Septier A (ed) 1967 Focusing Charged Particles vol 1 (New York: Academic)
[6] Abramowitz M and Stegun I (eds) 1965 Handbook of Mathematical Formulas (New York:
Dover)
[7] Likharev K 1999 Proc. IEEE 87 606
[8] Keller M et al 1996 Appl. Phys. Lett. 69 1804
[9] Stein F et al 2017 Metrologia 54 S1
[10] Brun-Pickard J et al 2016 Phys. Rev. X 6 041051
[11] Leveque R 2007 Finite Difference Methods for Ordinary and Partial Differential Equations
(SIAM)
[12] Johnson C 2009 Numerical Solution of Partial Differential Equations by the Finite Element
Method (New York: Dover)
[13] Hughes T J R 2000 The Finite Element Method (New York: Dover)
2-78
IOP Publishing
Lecture notes
Chapter 3
Dipoles and dielectrics
In contrast to conductors, in dielectrics the charge motion is limited to the interior of

an atom or a molecule, so that the electric polarization of these materials by an
external field takes a different form. This issue is the main subject of this chapter, but
in preparation for its analysis, we have to start with a general discussion of the electric
field induced by a spatially restricted system of charges.
3.1 Electric dipole

Let us consider a localized system of charges, of a linear size scale a, and derive a
simple but approximate expression for the electrostatic field induced by the system at
a distant point r. For that, let us select a reference frame with the origin either
somewhere inside the system, or at a distance of the order of a from it (figure 3.1).
Figure 3.1. Deriving the approximate expression for the electrostatic field of a localized system of charges at a
distant point (r ≫ r′ ∼ a).
Then positions of all charges of the system satisfy the following condition:
r′ ≪ r. (3.1)
Using this condition, we can expand the general expression (1.38) for the
electrostatic potential ϕ(r) of the system into the Taylor series in small

parameter r′. For any spatial function of the type f(r − r′), the expansion may be
represented as1
3 3
∂f 1 ∂ 2f
f (r − r′) = f (r) − ∑rj′ ∂rj
(r) + ∑
2! j, j ′= 1
rj ′rj ′′
∂rj ∂rj ′
( r ) − …. (3.2)
j=1
Applying this formula to the free-space Green’s function 1/∣r − r′∣ in Eq. (1.38), we
obtain the so-called multipole expansion of the electrostatic potential:
⎛ 3 3 ⎞
1 ⎜1 1 1 ⎟,
ϕ(r) =
4πε0 ⎜⎝ r
Q + ∑rjp + 2r 5
r3 j=1 j
∑ j j′ jj′
r r Q + …⎟ (3.3)
j , j ′= 1 ⎠
whose r-independent parameters are defined as follows:
Q≡ ∫ ρ(r′)d 3r′, pj ≡ ∫ ρ(r′)rj′d 3r′, Q jj ′ ≡ ∫ ρ(r′)(3rj′rj′′ − r′2 δjj′) d 3r′. (3.4)

(Indeed, the two leading terms of the expansion (3.2) may be rewritten in the vector
form f(r) − r′·∇f(r), and the gradient of such a spherically symmetric function f(r) =
1/r is just nr df/dr, so that
1 1 d ⎛1 ⎞ 1 r
≈ − r′ ⋅ n r ⎜ ⎟ = + r′ ⋅ 3 , (3.5)
r − r′ r dr ⎝ r ⎠ r r
immediately giving the two first terms of Eq. (3.3). The proof of the third, quadrupole
term in Eq. (3.3) is similar but a bit longer, and is left for the reader’s exercise.)
Evidently, the scalar parameter Q in Eqs. (3.3) and (3.4) is just the total charge of
the system. The constants pj are the Cartesian components of a vector
p≡ ∫ ρ(r′)r′d 3r′, (3.6)
called the system’s electric dipole moment, and Q jj′ are the Cartesian components of a
tensor—the system’s electric quadrupole moment. If Q ≠ 0, all higher terms on the
right-hand side of Eq. (3.3), at large distances (3.1), are just small corrections to the
first one, and in many cases may be ignored. However, the net charge of many
systems is exactly zero, the most important examples being neutral atoms and
molecules. For such neural systems, the second (dipole-moment) term, ϕd, in
Eq. (3.3) is, most frequently, the leading one. (Such systems are called electric
dipoles.) Due to its importance, let us rewrite the expression for this term in three
other, equivalent forms:
1 r⋅p 1 p cos θ 1 pz
ϕd ≡ ≡ ≡ , (3.7)
4πε0 r 3
4πε0 r 2
4πε0 [x + y 2 + z 2 ]3/2
2
1
See, e.g. Eq. (A.14).
3-2
that are more convenient for some applications. Here θ is the angle between the
vectors p and r, and in the last (Cartesian) representation, the axis z is directed along
the vector p. Figure 3.2a shows equipotential surfaces of the dipole field (or rather
their cross-sections by any plane in which the vector p resides).
Figure 3.2. (a) The equipotential surfaces and (b) the electric field lines of a dipole. (Panel (b) adapted from
http://en.wikipedia.org/wiki/Dipole under the GNU Free Documentation License.)
The simplest example of a system whose field, at large distances, approaches the
dipole field (3.7) (and which gave the dipole the name), is a system of two equal but
opposite point charges (‘poles’), +q and −q, with the radius-vectors, respectively, r+
and r−:
ρ(r) = ( +q )δ(r − r+) + ( −q )δ(r − r−). (3.8)
For this system, Eq. (3.5) yields
p = ( +q )r+ + ( −q )r− = q(r+ − r−) = qa , (3.9)
where a is the vector connecting points r− and r+. Note that in this case (and indeed
for all systems with Q = 0), the dipole moment does not depend on the choice of the
reference frame’s origin.
A less trivial example of a dipole is a conducting sphere of radius R in a uniform
external electric field E0. As a reminder, this problem was solved in section 2.8, and
its result is expressed by Eq. (2.176). The first term in the parentheses of that relation
describes just the external field (2.173), so that the field of the sphere itself (i.e. that of
the surface charge induced by E0) is given by the second term:
E 0R3
ϕs = cos θ . (3.10)
r2
Comparing this expression with the second form of Eq. (3.7), we see that the sphere
has an induced dipole moment
p = 4πε0E 0R3. (3.11)
3-3
This is an interesting example of an almost purely dipole field: at all points outside
the sphere (r > R), the field has neither a quadrupole moment, nor any higher
moments.
Other examples of dipole fields are given by two more systems discussed in
chapter 2—see Eqs. (2.215) and (2.219). Those systems, however, do have higher-
order multipole moments, so that for them Eq. (3.7) gives only the long-distance
approximation.
Returning to the general properties of the dipole field (3.7), let us calculate its
characteristics. First of all, we may use Eq. (3.7) to calculate the electric field of a
dipole:
1 ⎛⎜ r ⋅ p ⎞⎟ 1 ⎛ p cos θ ⎞
Ed = −∇ϕd = − ∇ =− ∇⎜ ⎟. (3.12)
4πε0 ⎝ r 3 ⎠
4πε0 ⎝ r 2 ⎠
The differentiation is easiest in the spherical coordinates, using the well-known
expression for the gradient of a scalar function in these coordinates2 and taking axis
z parallel to the dipole moment p. From the last form of Eq. (3.12) we immediately
obtain
p 1 3r(r ⋅ p) − pr 2
Ed = 3
(2nr cos θ + n θ sin θ ) ≡ . (3.13)
4πε0r 4πε0 r5
Figure 3.2b shows the electric field lines given by Eq. (3.13). The most important
features of this result are a faster drop of the field’s magnitude (Ed ∝ 1/r3, rather than
E ∝ 1/r2 for a point charge), and the change of the signs of all field components as
functions of the polar angle θ.
Next, let us use Eq. (1.55) to calculate the potential energy of interaction between
a dipole and an external electric field. Assuming that the external field does not
change much at distances of the order of a (figure 3.1), we may expand the external
potential ϕext(r) into the Taylor series, and keep only its two leading terms:
U= ∫ ρ(r)ϕext(r)d 3r ≈ ∫ ρ(r)[ϕext(0) + r ⋅ ∇ϕext(0)] d 3r (3.14)

≡ Qϕext(0) − p ⋅ Eext .
The first term is the potential energy the system would have if it were a point charge.
If the net charge Q is zero that term disappears and the leading contribution is due to
the dipole moment:
U = −p ⋅ Eext , for p = const. (3.15a )
Note that this result is only valid for a fixed dipole, with p independent of Eext. In the
opposite limit, when the dipole is induced by the field, i.e. p ∝ Eext (see again
2
See, e.g. Eq. (A.66) with ∂/∂φ = 0.
3-4
Eq. (3.11) as an example of such a proportionality), we need to start with Eq. (1.60)
rather than Eq. (1.55), and obtain
1
U = − p ⋅ Eext , for p ∝ Eext . (3.15b)
2
In particular, combining Eqs. (3.13) and (3.15a), we may obtain the following
important formula for the interaction of two independent dipoles
1 p1 ⋅ p2r 2 − 3(r ⋅ p1)(r ⋅ p2) 1 p1x p2x + p1y p2y − 2p1z p2z
Uint = 5
= , (3.16)
4πε0 r 4πε0 r3
where r is the vector connecting the dipoles, and axis z is directed along this vector. It
is easy to prove (this exercise is left for the reader) that if the magnitude of each
dipole moment is fixed (an approximation valid, in particular, for weak interaction
of so-called polar molecules), this potential energy reaches its minimum at, and hence
favors the parallel orientation of the dipoles along, the line connecting them. Note
also that in this case Uint is proportional to 1/r3. On the other hand, if each moment p
has a random value plus a component due to its polarization by the electric field of
its counterpart, Δp1,2 ∝ E2,1 ∝ 1/r3, their average interaction energy (which may be
calculated from Eq. (3.16) with the additional factor ½) is always negative and
proportional to 1/r6. Such negative potential describes, in particular, the long-range,
attractive part (the so-called London dispersion force) of the interaction between
electrically neutral atoms and molecules3.
According to Eqs. (3.15), in order to reach the minimum of U, the electric field
‘tries’ to align the dipole direction along its own. The quantitative expression of this
effect is the torque τ exerted by the field. The simplest way to calculate it is to sum up
all the elementary torques dτ = r × dFext = r × Eext(r)ρ(r)d3r exerted on all elementary
charges of the system:
τ= ∫ r × Eext(r)ρ(r)d 3r ≈ p × Eext(0), (3.17)
where to make the last step, the spatial dependence of the external field was again
neglected. The spatial dependence of Eext cannot, however, be ignored at the
calculation of the total force exerted by the field on the dipole (with Q = 0).
Indeed, Eqs. (3.15) shows that if the field is constant, the dipole energy is the same at
all spatial points, and hence the net force is zero. However, if the field has a non-zero
gradient, a total force does appear; for a field-independent dipole,
F = −∇U = ∇(p ⋅ Eext), (3.18)
3
This force is calculated, using several models, in Part QM of this series.
3-5
where the derivative has to be taken at the dipole’s position (in our notation, at r = 0).
If the dipole that is being moved in a field retains its magnitude and orientation, then
the last formula is equivalent to4
F = (p ⋅ ∇)Eext . (3.19)
Alternatively, the last expression may be obtained similarly to Eq. (3.14):
F= ∫ ρ(r)Eext(r)d 3r ≈ ∫ ρ(r)[Eext(0) + (r ⋅ ∇)Eext]d 3r (3.20)

= Q Eext(0) + (p ⋅ ∇)Eext .
Finally, let me add a note on the so-called coarse-grain model of the dipole. The
dipole approximation explored above is asymptotically correct only at large
distances, r ≫ a. However, for some applications (including the forthcoming
discussion of the molecular field effects in section 3.3) it is important to have an
expression that would be approximately valid everywhere in space, though maybe
without exact details at r ∼ a, and also give the correct result for the space average of
the electric field,
1
E≡
V
∫V Ed 3r, (3.21)
where V is a regularly shaped volume much larger than a3, for example a sphere of a
radius R ≫ a, with the dipole at its center. For the field Ed given by Eq. (3.13), such
an average is zero. Indeed, let us consider the Cartesian components of that vector in
a reference frame with the axis z directed along vector p. Due to the axial symmetry
of the field, the averages of the components Ex and Ey evidently vanish. Let us use
Eq. (3.13) to spell out the ‘vertical’ component of the field (parallel to the dipole
moment vector):
p 1
Ez ≡ Ed ⋅ = (2nr ⋅ p cos θ − n θ ⋅ p sin θ )
p 4πε0r 3
(3.22)
p
= (2cos2 θ − sin2 θ ).
4πε0r 3
Integrating this expression over the whole solid angle Ω = 4π, at fixed r, using a
convenient variable substitution cos θ ≡ ξ, we obtain
π π
p
∮4π Ez d Ω = 2π∫0 Ez sin θ dθ =
2ε0r 3
∫0 (2cos2 θ − sin2 θ ) sin θ dθ
+1 (3.23)
p
=
2ε0r 3
∫−1 (3ξ3 − ξ ) dξ = 0.
4
The equivalence may be proved, for example, by using Eq. (A.76) with f = p = const and g = Eext, taking into
account that according to the general Eq. (1.28), ∇ × Eext = 0.
3-6
On the other hand, the exact electric field of an arbitrary charge distribution, with
the total dipole moment p, obeys the following equality:
1 4π
∫V E(r)d 3r = − 3pε0 ≡−
4πε0 3
p, (3.24)
where the integration is over any sphere containing all the charges. A proof of this
formula for the general case requires a straightforward, but somewhat tedious
integration5. The origin of Eq. (3.24) is illustrated in figure 3.3 on the example of the
dipole created by two equal but opposite charges—see Eqs. (3.8) and (3.9). The zero
average (3.23) of the dipole field (3.13) does not take into account the contribution of
the region between the charges (where Eq. (3.13) is not valid), which is directed
mostly against the dipole vector (3.9).
Figure 3.3. A sketch illustrating the origin of Eq. (3.24).
In order to be used as a reasonable coarse-grain model, Eq. (3.13) may be

modified as follows:
1 ⎡ 3r(r ⋅ p) − pr 2 4π ⎤
Ecg = ⎢ − pδ ( r ) ⎥, (3.25)
4πε0 ⎣ r5 3 ⎦
so that its integral satisfies Eq. (3.24). Evidently, such a modification does not
change the field at large distances r ≫ a, i.e. in the region where the expansion (3.3),
and hence Eq. (3.13), are valid.
3.2 Dipole media

Let us generalize Eq. (3.7) to the case of several (possibly, many) dipoles pj located at
arbitrary points rj. Using the linear superposition principle, we obtain
1 r − rj
ϕd(r) = ∑
4πε0 j
pj ⋅ 3
. (3.26)
r − rj
If our system (medium) contains many similar dipoles, distributed in space with
density n(r), we may approximate the last sum with a macroscopic potential, which is
the average of the ‘microscopic’ potential (3.26) over a local volume much larger
than the distance between the dipoles, and as a result is given by the integral
5
See, e.g. the end of section 4.1 in the textbook [1].
3-7
1 r − r′ 3
ϕd(r) =
4πε0
∫ P(r′) ⋅ r − r′ 3
d r′ , with P(r) ≡ n(r)p , (3.27)
where vector P(r), called the electric polarization, has the physical meaning of the net
dipole moment per unit volume. (Note that by its definition, P(r) is also a
‘macroscopic’ field.)
Now comes a very impressive trick, which is the basis of all the theory of the
‘macroscopic’ electrostatics (and eventually, electrodynamics). Just as was done at
the derivation of Eq. (3.5), Eq. (3.27) may be rewritten in the equivalent form
1
ϕd(r) =
4πε0
∫ P(r′) ⋅ ∇′ r −1 r′ d 3r′, (3.28)
where ∇′ means the del operator (in this particular case, the gradient) acting in the
‘source space’ of vectors r′. The right-hand side of Eq. (3.28), applied to any volume
V limited by surface S, may be integrated by parts to give6
1 Pn(r′) 2 1
ϕd(r) =
4πε0
∮S r − r′
d r′ −
4πε0
∫V ∇′r ⋅−Pr(′r′) d 3r′. (3.29)
If the surface does not carry an infinitely dense (δ-functional) sheet of additional
dipoles, or it is just very distant, the first term on the right-hand side is negligible7.
Now comparing the second term with the basic equation (1.38) for the electric
potential, we see that this term may be interpreted as the field of certain effective
electric charges with density
ρef = −∇ ⋅ P. (3.30)
Figure 3.4 illustrates the physics of this key relation for a cartoon model of a
simple multi-dipole system: a layer of uniformly distributed two-point-charge units
oriented perpendicular to the layer surface. (In this case ∇ · P = dP/dx.) One can see
that the ρef defined by Eq. (3.30) may be interpreted as the density of the
uncompensated surface charges of polarized elementary dipoles.
Next, from section 1.2, we already know that Eq. (1.38) is equivalent to the
inhomogeneous Maxwell equation (1.27) for the electric field, so that the macro-
scopic electric field of the dipoles (defined as Ed = −∇ϕd, where ϕd is given by
Eq. (3.27)) obeys a similar equation, with the effective charge density (3.30).
Now let us consider a more general case when a system, in addition to the
compensated charges of the dipoles, also has certain stand-alone charges (not parts
of the dipoles already taken into account in the polarization P)8. As was discussed
in section 1.1, if we average this charge over the inter-charge distances,
6
To prove this (almost evident) formula strictly, it is sufficient to apply the divergence theorem (A.79) to the
vector function f = P(r′)/∣r − r′∣, in the ‘source space’ of radius-vectors r′.
7
Just as in the case of Eq. (1.9), if we want to describe such a dipole sheet, we may always do that using the
second term in Eq. (3.29), by including a delta-functional part into the polarization distribution P(r′).
8
In some texts, these charges are called ‘free’. This term is somewhat misleading, because the stand-alone may
well be bound, i.e. unable to move freely.
3-8
Figure 3.4. The spatial distributions of the polarization and effective charges in a layer of similar elementary
dipoles (schematically).
i.e. approximate it with a continuous ‘macroscopic’ density ρ(r), then its macro-
scopic electric field also obeys Eq. (1.27), but with the stand-alone charge density.
Due to the linear superposition principle, for the total macroscopic field E of the
charges and dipoles we may write
1 1
∇⋅E= (ρ + ρef ) = (ρ − ∇ ⋅ P). (3.31)
ε0 ε0
This is already the main result of the ‘macroscopic’ electrostatics. However, it is
evidently tempting (and very convenient for applications) to rewrite Eq. (3.31) in a
different form by carrying the dipole-related term of this equation over to the left-
hand side. The resulting equality is called the macroscopic Maxwell equation for D:
∇ ⋅ D = ρ, (3.32)
where D(r) is a new ‘macroscopic’ field, called the electric displacement, defined as9
D ≡ ε0E + P. (3.33)
The comparison of Eqs. (3.32) and (1.27) shows that D (or more strictly, the
fraction D/ε0) may be interpreted as the ‘would-be’ macroscopic electric field that
would be created by stand-alone charges in the absence of the dipole medium
polarization. In contrast, the E participating in Eqs. (3.31) and (3.33) is the genuine
macroscopic electric field, exact at distances much larger than that between the
adjacent elementary point stand-alone charges and dipoles.
9
Note that according to its definition (3.33), the dimensionality of D in the SI units is different from that of E.
In contrast, in the Gaussian units the electric displacement is defined as D = E + 4πP, so that ∇ · D = 4πρ (the
relation ρef = −∇ · P remains the same as in SI units), and the dimensionalities of D and E coincide.
Philosophically, this coincidence is a certain perceptional handicap, because it is frequently convenient to
consider the scalar components of E as generalized forces, and those of D as generalized coordinates (see
section 3.5 below), and it is somewhat comforting to have their dimensionalities different.
3-9
In order to have a better look into the physics of the fields E and D, let us first
rewrite the macroscopic Maxwell equation (3.32) in the integral form. Applying the
divergence theorem to an arbitrary volume V limited by surface S, we obtain the
following macroscopic Gauss law:
∮S Dnd 2r = ∫V ρ d 3r ≡ Q, (3.34)
where Q is the total stand-alone charge inside volume V.

This general result may be used to find the boundary conditions for D at a sharp
interface between two different dielectrics. (The analysis is clearly applicable to a
dielectric/free-space boundary as well.) For that, let us apply Eq. (3.34) to a pillbox
formed at the interface (see the solid rectangle in figure 3.5), which is sufficiently
small on the spatial scales of the dielectrics’ non-uniformity and the interface’s
curvature, but still containing many elementary dipoles. Assuming that the interface
does not have stand-alone surface charges, we immediately obtain
(Dn)1 = (Dn)2 , (3.35)
i.e. the normal component of the electric displacement has to be continuous. Note
that a similar statement for the macroscopic electric field E is generally not valid,
because the polarization vector P may have, and typically does have a leap at a
sharp interface (say, due to the different polarizability of the two different
dielectrics), providing a surface layer of the effective charges (3.30)—see again the
example in figure 3.4.
However, we still can make an important statement about the behavior of E at the
interface. Indeed, the macroscopic field electric fields, defined by Eqs. (3.29) and
(3.31), are evidently still potential ones, and hence obey another macroscopic
Maxwell equation, similar to Eq. (1.28):
∇ × E = 0. (3.36)
Integrating this equation along to a narrow contour stretched along the interface
(see the dashed rectangle in figure 3.5), we obtain
Figure 3.5. Deriving the boundary conditions at an interface between two dielectrics, using a Gauss pillbox
(shown as a solid rectangle) and a circulation contour (dashed-line rectangle). n and τ are the unit vectors
which are, respectively, normal and tangential to the interface.
3-10
(E τ )1 = (E τ )2 . (3.37)
Note that this condition is compatible with (and may be derived from) the continuity
of the macroscopic electrostatic potential ϕ (related to the macroscopic field E by the
relation similar to Eq. (1.33), E = −∇ϕ), at each point of the interface: ϕ1 = ϕ2.
In order to see how do these boundary conditions work, let us consider the simple
problem shown in figure 3.6. A very broad plane capacitor, with zero voltage
between its conducting plates (as may be fixed, e.g. by their connection with an
external wire), is partly filled with a material with a constant polarization P0,10
oriented normal to the plates. Let us calculate the spatial distribution of the fields E
and D, and also the surface charge density of each conducting plate.
Figure 3.6. A simple system whose analysis requires Eq. (3.35).
Due to the symmetry of the system, vectors E and D are all evidently normal to
the plates, and do not depend on the position in the capacitor’s plane, so that we can
limit the field analysis to the calculation of their z-components E(z) and D(z). In this
case, Eq. (3.32) is reduced to dD/dz = 0 inside each layer (but not at their border!), so
that within each of them D is constant—say, D1 in the layer with P = P0, and D2 in
the free-space layer (where P = 0). As a result, according to Eq. (3.33), the
(macroscopic) electric field inside each layer is also constant:
D1 = ε0E1 + P0, D2 = ε0E2. (3.38)
Since the voltage between the plates is zero, we may also require the integral of E,
taken along a path connecting the plates, to vanish. This gives us one more relation:
E1d1 + E2d2 = 0. (3.39)
Still, the three Eqs. (3.38) and (3.39) are insufficient to calculate the four fields in the
system (E1,2 and D1,2). The decisive help comes from the boundary condition (3.35):
D1 = D2 . (3.40)
(Note that Eq. (3.35) is valid because the layer interface does not carry stand-alone
electric charges, even though it has a polarization surface charge, whose areal density
may be calculated by integrating Eq. (3.30) across the interface: σef = P0. Note also
that in our simple system, Eq. (3.37) is identically satisfied due to the system’s
symmetry, and does not give any additional information.)
Now solving the resulting system of four equations, we readily obtain
10
As will be discussed in the next section, this is a good approximation for the so-called electrets, and also hard
ferroelectrics, in not very high electric fields.
3-11
P0 d2 P0 d1 d1
E1 = − , E2 = , D1 = D2 = D = P0 . (3.41)
ε0 d1 + d2 ε0 d1 + d2 d1 + d2
The areal densities of the electrode surface charges may be readily calculated now by
the integration of Eq. (3.32) across each surface:
d1
σ1 = −σ2 = D = P0 . (3.42)
d1 + d2
Note that due to the spontaneous polarization of the lower layer’s material, the
capacitor plates are charged even in the absence of voltage between them, and that
this charge is a function of the second electrode position (d2)11. Also notice a
substantial similarity between this system (figure 3.6) and the system whose analysis
was the subject of problem 2.6.
3.3 Polarization of dielectrics

The general relations derived in the previous section may be used to describe the
electrostatics of dielectrics—materials with bound electric charges (and hence with
negligible dc electric conduction). However, in order to form the full system of
equations necessary to solve electrostatics problems, they have to be complemented
by certain relations between the vectors P and E.12
In most materials, in the absence of external electric field, the elementary dipoles
p either equal zero or have a random orientation in space, so that the net dipole
moment of each macroscopic volume (still containing many such dipoles) equals
zero: P = 0 at E = 0. Moreover, if the field changes are sufficiently slow, most
materials may be characterized by a unique dependence of P on E. Then using the
Taylor expansion of function P(E), we may argue that in relatively low electric fields
the function should be well approximated by a linear dependence between these two
vectors. Such dielectrics are called linear. In an isotropic media, the coefficient of
proportionality should be just a scalar. In the SI units, this scalar is defined by the
following relation:
P = χe ε0E, (3.43)
with the dimensionless constant χe called the electric susceptibility. However, it is
much more common to use, instead of χe, another dimensionless parameter13,
11
This effect is used in most modern microphones. In such a device, the sensed sound wave’s pressure bends a
thin conducting membrane playing the role of one of capacitor’s plates, and thus modulates the thickness (in
figure 3.6, d2) of the air gap adjacent to an electret layer. This modulation produces a proportional electric
charge variation that is picked up by device’s electronics.
12
This is one more example of constitutive relations (already mentioned in section 2.1). In the problem solved
in the end of the previous section, the role of this relation was played by the equality P0 = const.
13
In older physics literature, the dielectric constant is often denoted by letter εr, while in the electrical
engineering literature, its notation is frequently K.
3-12
κ ≡ 1 + χe , (3.44)
which is sometimes called the ‘relative electric permittivity’, but much more often,
the dielectric constant. This parameter is very convenient, because combining Eqs.
(3.43) and (3.44),
P = (κ − 1)ε0E, (3.45)
and then plugging the resulting relation into the general Eq. (3.33), we obtain simply
D = κε0 E, or D = εE, (3.46)
where another popular parameter14,
ε ≡ κε0 ≡ (1 + χe )ε0. (3.47)
ε is called the electric permittivity of the material15. Table 3.1 gives approximate
values of the dielectric constant for several representative materials.
In order to understand the range of these values, let me discuss (rather super-
ficially) the two simplest mechanisms of electric polarization. The first of them is
typical for liquids and gases of polar atoms/molecules, which have their own,
spontaneous dipole moments p.16 In the absence of the external electric field, the
Table 3.1. Dielectric constants of a few representative (and/or practically important) dielectrics.
Material κ
Air (at ambient conditions) 1.00054

Teflon (polytetrafluoroethylene, CnF2n) 2.1
Silicon dioxide (amorphous) 3.9
Glasses (of various compositions) 3.7–10
Castor oil 4.5
Silicona 11.7
Water (at 100 °C) 55.3
Water (at 20 °C) 80.1
Barium titanate (BaTiO3, at 20 °C) ∼1,600
a
Anisotropic materials, such as silicon crystals, require a susceptibility tensor to give an exact description of the
linear relation of vectors P and E. However, most important crystals (including Si) are only weakly
anisotropic, so that they may be reasonably well characterized with a scalar (angle-average) susceptibility.
14
The reader may be perplexed by the use of three different but uniquely related parameters (χe, κ ≡ 1 + χe, and
ε ≡ κε0) for the description of just one scalar property. Unfortunately, such redundancy is typical for physics,
whose different sub-field communities have different, well-entrenched traditions.
15
In the Gaussian units, χe is defined by the following relation: P = χeE, while ε is defined just as in the SI units,
D = εE. Because of that, in the Gaussian units ε is dimensionless and equals (1 + 4πχe). As a result, εGaussian =
(ε/ε0)SI ≡ κ, so that (χe)Gaussian = (χe)SI/4π, sometimes creating a confusion with the numerical values of the
latter parameter (dimensionless in both systems).
16
A typical example is the water molecule H2O, with the negative oxygen ion offset from the line connecting
two positive hydrogen ions, thus producing a spontaneous dipole moment p = ea, with a ≈ 0.38 × 10−10 m ∼ rB.
3-13
Figure 3.7. Crude cartoons of two mechanisms of the induced electrical polarization: (a) a partial ordering of
spontaneous elementary dipoles and (b) an elementary dipole induction. The upper two panels correspond to
E = 0, and the lower two panels, to E ≠ 0.
orientation of such dipoles may be random, with the average polarization P = n〈p〉
equal to zero—see the top panel of figure 3.7a.
A relatively weak external field does not change the magnitude of the dipole
moments significantly but, according to Eqs. (3.15a) and (3.17), tries to orient them
along the field and thus creates a non-zero vector average 〈p〉 directed along the
vector 〈Em〉, where Em is the microscopic field at the point of the dipole’s location—
cf the two panels of figure 3.7a. If the field is not two high (p〈Em〉 ≪ kBT), the
induced average polarization 〈p〉 is proportional to 〈Em〉. If we write this propor-
tionality relation in the following traditional form,
p = α E m, (3.48)
where α is called the atomic (or, sometimes, ‘molecular’) polarizability, this means
that α is positive. If the concentration n of such elementary dipoles is low, the
contribution of their own fields into the microscopic field acting on each dipole is
negligible, and we may identify 〈Em〉 with the macroscopic field E. As a result, the
second of Eq. (3.27) yields
P ≡ n p = αn E. (3.49)
Comparing this relation with Eq. (3.45), we obtain
α
κ = 1 + n, (3.50)
ε0
so that κ > 1 (i.e. χe > 0). Note that at this particular polarization mechanism
(illustrated on the lower panel of figure 3.7a), the thermal motion ‘tries’ to
randomize the dipole orientation, i.e. reduce its ordering by the field, so that we
may expect α, and hence χe ≡ κ − 1 to increase as temperature T is decreased—the
3-14
so-called paraelectricity. Indeed, the elementary statistical mechanics17 shows that in

this case the electric susceptibility follows the law χe ∝ 1/T.
The materials of the second, much more common class consist of non-polar atoms
without intrinsic spontaneous polarization. A crude classical image of such an atom
is an isotropic cloud of negatively charged electrons surrounding a positively
charged nucleus—see the top panel of figure 3.7b. The external electric field shifts
the positive charge in the direction of the vector E, and the negative charges in the
opposite direction, thus creating a similarly directed average dipole moment 〈p〉.18
At relatively low fields, this average moment is proportional to E, so that we again
arrive at Eq. (3.48), with α > 0, and if the dipole concentration n is sufficiently low,
also at Eq. (3.50), with κ − 1 > 0. So, the dielectric constant is larger than 1 for both
polarization mechanisms—please have one more look at table 2.1.
In order to make a crude but physically transparent estimate of the magnitude of
the difference κ − 1, let us consider the following toy model of a non-polar dielectric: a
set of similar conducting spheres of radius R, distributed in space with a low density n
≪ 1/R3. At such density, the electrostatic interaction of the spheres is negligible, and
we can use Eq. (3.11) for the induced dipole moment of a single sphere. Then the
polarizability definition (3.48) yields α = 4πε0R3, so that Eq. (3.50) gives
κ = 1 + 4πR3n . (3.51)
Let us use this result for a crude estimate of the dielectric constant of air at the so-
called ambient conditions, meaning the normal atmospheric pressure P = 1.013 × 105
Pa and temperature T = 300 K. At these conditions the molecular density n may be, with
a few-percent accuracy, found from the equation of state of an ideal gas19:
n ≈ P /kBT ≈ (1.013 × 105)/(1.38 × 10−23 × 300) ≈ 2.45 × 1025m−3. The molecule
of the air’s main component, N2, has a van der Waals radius20 of 1.55 × 10−10 m.
Taking this radius for the R of our crude model, we obtain χe ≡ κ − 1 ≈ 1.15 × 10−3.
Comparing this number with the first line of table 3.1, we see that the model gives a
surprisingly reasonable result: in order to obtain the experimental value, it is
sufficient to decrease the effective R of the sphere by just ∼30%, to ∼1.2 × 10−10 m.21
This result may encourage us to try using Eq. (3.51) for a larger density n. For
example, as a crude model for a non-polar crystal, let us assume that the conducting
spheres form a simple cubic lattice with the period a = 2R (i.e. the neighboring
spheres virtually touch). With this n = 1/a3 = 1/8R3, and Eq. (3.44) yields κ = 1 +
17
See, e.g. Part SM chapter 2.
18
Realistically, these effects are governed by quantum mechanics, so that the average here should be
understood not only in the statistical-mechanical, but also (and mostly) in the quantum-mechanical sense.
Because of that, for non-polar atoms, α is typically a very weak function of temperature, at least on the usual
scale T ∼ 300 K.
19
See, e.g. Part SM sections 1.4 and 3.1.
20
Such radius is defined by the requirement that the volume of the corresponding sphere, if used in the van der
Waals equation (see, e.g. Part SM section 4.1), gives the best fit to the experimental equation of state n = n
(P, T).
21
As discussed in Part QM chapter 6, for a hydrogen atom in its ground state, the low-field polarizability may
be calculated analytically, α = (9/2) × 4πε0rB3, corresponding to our metallic-ball model with a close value of
the effective radius: R = (9/2)1/3rB ≈ 1.65 rB ≈ 0.87 × 10−10 m.
3-15
4π/8 ≈ 2.5. This estimate provides a reasonable semi-qualitative explanation for the
values of κ listed in first few lines of table 3.1. However, at such small distances the
electrostatic dipole–dipole interaction should already be essential, so that such a
simple model cannot even approximately describe values of κ much larger than 1,
listed in the last rows of the table.
Such high values may be explained by the molecular field effect: each elementary
dipole is polarized not only by the external field (as Eq. (3.49) assumes), but by the
field of neighboring dipoles as well. In 1850, O-F Mossotti and (perhaps, independ-
ently, but almost 30 years later) R Clausius suggested what is now known, rather
unfairly, as the Clausius–Mossotti formula, which works remarkably well for a broad
class of non-polar materials22. In our notation it reads23
κ−1 αn 1 + 2α n /3ε0
= , so that κ = . (3.52)
κ+2 3ε 0 1 − αn /3ε0
If the dipole density is low in the sense n ≪ 3ε0/α, this relation is reduced to
Eq. (3.50) corresponding to independent dipoles. However, at higher dipole density,
both κ and χe increase much faster, and tend to infinity as the density-polarizability
product approaches a critical value, in the simple Clausius–Mossotti model equal to
3ε0.24 This means that the zero-polarization state becomes unstable even in the
absence of an external electric field.
This instability is a linear-theory (i.e. low-field) manifestation of the substantially
nonlinear effect—the formation of a spontaneous polarization even in the absence of
external electric field. Such materials are called ferroelectrics and may be exper-
imentally recognized by the hysteretic behavior of their polarization as a function of
the applied (external) electric field—see figure 3.8. As the plots show, the polar-
ization of a ferroelectric depends on the applied field’s history. For example, the
direction of its spontaneous remnant polarization PR may be switched by applying
and then removing a sufficiently high field (larger than the so-called coercive field
EC—see figure 3.8). The physics of this switching is rather involved; the polarization
vector P of a ferroelectric material is typically constant only within each of the
spontaneously formed spatial regions (called domains), with a typical size of a few
tenths of a micron, and different (frequently, opposite) directions of the vector P in
adjacent domains. The change of the applied electric field results not in the switching
of the direction of P inside each domain, but rather in a shift of the domain walls,
resulting in the change of the average polarization of the sample.
22
Applied to the high-frequency electric field, with κ replaced by the square of the refraction coefficient n at the
field’s frequency (see chapter 7), this formula is known as the Lorenz–Lorentz relation.
23
I am leaving the proof of Eq. (3.52), with help from a formula that will be derived in the next section, for the
reader’s exercise.
24
The Clausius–Mossotti model does not give qualitatively correct results for most ferroelectric materials. For
a review of modern approaches to the theory of their polarization, see, e.g. the paper by Resta and Vanderbilt
in the recent review collection [2].
3-16
Figure 3.8. A schematic representation of the average polarization of soft and hard ferroelectrics as functions
of the applied electric field.
Depending on the ferroelectric’s material, temperature, and the sample’s geom-

etry (a solid crystal, a ceramic material, or a thin film), the hysteretic loops may be
rather different, ranging from a rather smooth form in the so-called soft ferro-
electrics (with include most ferroelectric thin films) to an almost rectangular form in
hard ferroelectrics—see figure 3.8. In low fields, soft ferroelectrics behave essentially
as linear paraelectrics, but with a very high average dielectric constant—see the
bottom line of table 3.1 for such a classical material as BaTiO3 (which is a soft
ferroelectric at temperatures below Tc ≈ 120 °C, and a paraelectric above this
critical temperature). On the other hand, the polarization of a hard ferroelectric in
the fields below its coercive field remains virtually constant, and the analysis of their
electrostatics may be based on the condition P = PR = const—already used in the
problem discussed at the end of the previous section25. This condition is even more
applicable to the so-called electrets—synthetic polymers with a spontaneous polar-
ization that remains constant even in very high electric fields.
Some materials exhibit even more complex polarization effects, for example
antiferroelectricity, helielectricity, and (practically very valuable) piezoelectricity.
Unfortunately, we do not have time for a discussion of these exotic phenomena in
this course26; the main reason I am mentioning them is to emphasize again that the
constitutive relation P = P(E) is material-specific rather than fundamental. However,
most insulators, in practicable fields, behave as linear dielectrics, so that the next
section will be committed to the discussion of their electrostatics.
25
Due to this property, hard ferroelectrics, such as the lead zirconate titanate (PZT) and strontium bismuth
tantalite (SBT), with high remnant polarization PR (up to ∼1 C m−2), may be used in non-volatile random-
access memories (dubbed either FRAM or FeRAM)—see, e.g. [3]. In a cell of such a memory, binary
information is stored in the form of one of two possible directions of spontaneous polarization at E = 0 (see
figure 3.8). Unfortunately, the time of spontaneous depolarization of ferroelectric thin films is typically well
below 10 years—the industrial standard for data retention in non-volatile memories, and this time may be
decreased even more by ‘fatigue’ from the repeated polarization recycling at information recording. Due to
these reasons, the industrial production of FRAM is currently just a tiny fraction of the non-volatile memory
market, which is dominated by floating-gate memories—see, e.g. section 4.2.
26
For a detailed coverage of ferroelectrics, I can recommend the encyclopedic monograph [4] and the recent
review collection [2].
3-17
3.4 Electrostatics of linear dielectrics

First, let us discuss the simplest problem: how is the electrostatic field of a set of
stand-alone charges of density ρ(r) modified by a uniform linear dielectric medium,
which obeys Eq. (3.46) with a space-independent dielectric constant κ. For this case,
we may combine Eqs. (3.32) and (3.46) to write
ρ
∇⋅E= . (3.53)
ε
As a reminder, in the free space we had the similar equation (1.27), but with a
different constant, ε0 = ε/κ. Hence all the results discussed in chapter 1 are valid
inside a uniform linear dielectric, for the macroscopic field E (and the corresponding
macroscopic electrostatic potential ϕ), if they are reduced by the factor of κ > 1.
Thus, the most straightforward result of the induced polarization of a dielectric
medium is the electric field reduction. This is a very important effect, in particular
taking into account the very high values of κ in such dielectrics as water—see table 3.1.
Indeed, this is the reduction of the attraction between positive and negative ions (called,
respectively, cations and anions) in water that enables their substantial dissociation and
hence almost all biochemical reactions, which are the basis of the biological cell
functions—and hence of the life itself.
Let us apply this general result to the important particular case of the plane
capacitor (figure 2.3) filled with a linear, uniform dielectric. Applying the macro-
scopic Gauss law (3.34) to a pillbox-shaped volume on the conductor surface, we
obtain the following relation,
∂ϕ
σ = Dn = εEn = −ε , (3.54)
∂n
which differs from Eq. (2.3) only by the replacement ε0 → ε ≡ κε0. Hence the charge
density, calculated for the free-space case, should be increased by the factor of κ—
that is it. In particular, this means that all the mutual capacitance (2.28) has to be
increased by this factor:
κε0A εA
C= ≡ . (3.55)
d d
(As a reminder, this increase of C by κ has already been incorporated, without
derivation, into some estimates made in sections 2.1 and 2.2.)
If a linear dielectric is non-uniform, the situation is more complex. For example,
let us consider the case of a sharp interface between two otherwise uniform
dielectrics, free of stand-alone charges. In this case, we still may use (3.37) for the
tangential component of the macroscopic electric field, and also Eq. (3.36), with
Dn = εEn, which yields
∂ϕ1 ∂ϕ
(εEn)1 = (εEn)2 , i.e. ε1 = ε2 2 . (3.56)
∂n ∂n
3-18
Let us apply these boundary conditions, first of all, to the very illuminating case
of two very thin (t ≪ d) slits cut in a uniform dielectric with an initially uniform27
electric field E0 (figure 3.9). In both cases, a slit with t → 0 cannot modify the field
distribution outside it substantially.
Figure 3.9. Fields inside two narrow slits cut in a linear dielectric with a uniform field E0.
For the slit A, with the plane normal to the applied field, we may apply Eq. (3.56)
to the ‘major’ (broad) interfaces, shown horizontal in figure 3.9, to see that the
vector D should be continuous. But according to Eq. (3.46), this means that in the
free space inside the gap the electric field equals D/ε0, and hence is κ times higher
than the applied field E0 = D/κε0. This field, and hence D, may be measured by a
sensor placed inside the gap, showing that the electric displacement is by no means a
purely mathematical construct28. In contrast, for the slit B parallel to the applied
field, we may apply Eq. (3.37) to the major (now, vertical) interfaces of the slit, to see
that now the electric field E is continuous, while the electric displacement D = ε0E
inside the gap is a factor of κ lower than its value in the dielectric. (Similarly to the
case A, any perturbations of the field uniformity, caused by the compliance with
Eq. (3.56) at the minor interfaces, are settled at distances ∼t from them.)
For other problems with piecewise-constant ε and with more complex geometries
we may need to apply the methods studied in chapter 2. As in the free space, in the
simplest cases we can select such a set of orthogonal coordinates that the electro-
static potential depends on just one of them. Consider, for example, two types of
plane capacitor filling with two different dielectrics—see figure 3.10.
In case (a), the voltage V between the electrodes is the same for each part of the
capacitor and, at least far from the dielectric interface, the electric field is vertical,
uniform, and similar (E = V/d). Hence the boundary condition (3.37) is satisfied even
if such a distribution is valid near the surface as well, i.e. at any point of the system.
The only effect of different values of ε in the two parts is that the electric
displacement D = εE and hence the electrodes’ surface charge densities σ = D are
different in the two parts. Thus we can calculate the electrode charges Q1,2 of the two
27
Actually, the following arguments and results are valid for any external field distribution, provided that the
slits are much smaller than the characteristic scale of the field’s change in space.
28
Superficially, this result violates the boundary condition (3.37) at the vertical (‘minor’) interfaces. This
apparent contradiction may be resolved, taking into account that the slit deforms the field outside it, near its
edges. These fringe effects extend from the edges only by the horizontal distances ∼t, so that the above
relations for E and D are valid at most of the slit area.
3-19
Figure 3.10. Plane capacitors filled with two different dielectrics.
parts independently, and then add up the results to obtain the total mutual
capacitance
Q1 + Q2 1
C= = (ε1A1 + ε2A2 ). (3.57)
V d
Note that this formula may be interpreted as the total capacitance of two separate
lumped capacitors connected (by wires) in parallel. This is natural, because we may
cut the system along the dielectric interface, without any effect on the fields in either
part, and then connect the corresponding electrodes by external wires, again without
any effect on the system—besides very close vicinities of capacitor’s edges.
Case (b) may be analyzed just as in the problem shown in figure 3.6 by applying
Eq. (3.34) to a Gaussian pillbox with one lid inside (for example) the bottom
electrode, and the other lid in any of the layers. From this we see that D anywhere
inside the system should be equal to the surface charge density σ of the electrode, i.e.
constant. Hence, according to Eq. (3.46), the electric field inside each dielectric layer
is also constant: in the top layer E1 = D1/ε1 = σ/ε1, while in the bottom layer, E2 =
D2/ε2 = σ/ε2. Integrating the field E across the whole capacitor, we obtain
d1+d 2 ⎛d d ⎞
V= ∫0 E (z )dz = E1d1 + E2d2 = ⎜ 1 + 2 ⎟σ , (3.58)
⎝ ε1 ε2 ⎠
so that the mutual capacitance per unit area
C σ ⎡d d ⎤
−1
≡ = ⎢ 1 + 2⎥ . (3.59)
A V ⎣ ε1 ε2 ⎦
Note that this result is similar to the total capacitance of an in-series connection of
two plane capacitors based on each of the layers. This is also natural, because we
could insert an uncharged, thin conducting sheet (rather than a cut as in the previous
case) at the layer interface, which is an equipotential surface, without changing the
field distribution in any part of the system. Then we could thicken the conducting
sheet as much as we like (turning it into a wire), also without changing the fields and
hence the capacitance.
Proceeding to problems with more complex geometry, let us consider the system
shown in figure 3.11a: a dielectric sphere placed into an initially uniform external
electric field E0. According to Eq. (3.53) for the macroscopic electric field, and the
definition of the macroscopic electrostatic potential, E = −∇ϕ, the potential satisfies
3-20
Figure 3.11. Dielectric sphere in an initially uniform electric field: (a) the problem and (b) the equipotential
surfaces, as given by Eq. (3.62), for κ = 3.
the Laplace equation both inside and outside the sphere. Due to the spherical
symmetry of the dielectric sample, this problem invites the variable separation
method in spherical coordinates, which was discussed in section 2.8. From that
discussion, we already know, in particular, the general solution (2.172) of the
Laplace equation outside of the sphere. In order to avoid the divergence, and satisfy
the uniform-field condition at r → ∞, we have to reduce this solution to
∞
b
ϕr⩾R = −E 0r cos θ + ∑ r l +l 1 P l (cos θ ). (3.60)
l=1
Inside the sphere we can also use Eq. (2.172), but keeping only the radial functions
finite at r → 0:
∞
ϕr⩽R = ∑al r l P l (cos θ ). (3.61)
l=1
Now, spelling out the boundary conditions (3.37) and (3.56) at r = R, we see that for
all coefficients al and bl with l ⩾ 2 we obtain (just as for the conducting sphere,
discussed in section 2.8) homogeneous equations that have only trivial solutions.
Hence, all these terms may be dropped, while for the only surviving angular
harmonic, proportional to P 1(cos θ ) ≡ cos θ , we obtain two equations:
2b1 b1
−E 0 − = κa1, −E 0R + = a1R . (3.62)
R3 R2
Solving this simple system of linear equations for a1 and b1, and plugging the result
into Eqs. (3.60) and (3.61), we obtain the final solution of the problem:
⎛ κ − 1 R3 ⎞ 3
ϕr⩾R = E 0⎜ −r + ⎟ cos θ , ϕr⩽R = −E 0 r cos θ . (3.63)
⎝ κ + 2 r2 ⎠ κ+2
3-21
Figure 3.11b shows the equipotential surfaces given by this solution, for a
particular value of the dielectric constant κ. Note that according to Eq. (3.62), at
r ⩾ R the dielectric sphere, just as the conducting sphere in a similar problem,
produces (on the top of the uniform external field) a purely dipole field with the
dipole moment
κ−1 κ−1 4π 3
p = 4πR3 ε0E 0 ≡ 3V ε0 E 0 , where V = R, (3.64)
κ+2 κ+2 3
an evident generalization of Eq. (3.11), to which Eq. (3.64) tends at κ → ∞. By the
way, this property is common: from the point of view of their electrostatic (but not
transport!) properties, conductors may be adequately described as dielectrics with
κ → ∞.
Another remarkable feature of Eq. (3.63) is that the electric field and polarization
inside the sphere are uniform, with R-independent values29
3 3κ κ−1
E= E 0, D ≡ κε0E = ε0 E 0, P ≡ D − ε 0 E = 3ε 0 E 0. (3.65)
κ+2 κ+2 κ+2
In the limit κ → 1 (the ‘sphere made of free space’, i.e. no sphere at all), the electric
field inside the sphere naturally tends to the external one, and its polarization
disappears. In the opposite limit κ → ∞, the electric field inside the sphere vanishes.
Curiously enough, in this limit the electric displacement inside the sphere remains
finite: D → 3ε0E0.
More complex problems with piecewise-uniform dielectrics also may be addressed
by the methods discussed in chapter 2, and I leave a few of them for the reader’s
exercise. Let me discuss just one of such problems, because it exhibits the new
features of the charge image method which was discussed in section 2.9 (and is the
basis of the Green’s function approach—see section 2.10). Consider the system
shown in figure 3.12, a point charge near a dielectric half-space, which evidently
parallels the simplest problem discussed in section 2.9—see figure 2.26.
Figure 3.12. Charge images for a dielectric half-space.
29
The first of these relations is used in the standard derivation of the Clausius–Mossotti formula (3.52).
3-22
As for the case of a conducting half-space, the Laplace equation for the
electrostatic potential in the upper half-space z > 0 (besides the charge point
ρ = 0, z = d) may be satisfied using a single image charge q′ at point ρ = 0, z = −d,
but now q′ may differ from (−q). In addition, in contrast to the case analyzed in
section 2.9, we should also calculate the field inside the dielectric (at z ⩽ 0). This field
cannot be contributed by the image charge q′, because it would provide a potential
divergence at its location. Thus, in that half-space we should try to use the real point
source only, but maybe with a re-normalized charge qʺ rather than the genuine
charge q—see figure 3.12. As a result, we may look for the potential distribution in
the form
1
ϕ(ρ , z ) =
4πε0
⎧⎡ q q′ ⎤
⎪⎢ + ⎥, for z ⩾ 0, (3.66)
⎪ ⎣ [ρ 2 + (z − d )2 ]1/2 [ρ 2 + (z + d )2 ]1/2 ⎦
×⎨
⎪ q″
⎪ , for z ⩽ 0,
⎩ [ρ 2
+ ( z − d )2 ]1/2
at this stage with unknown q′ and qʺ. Plugging this solution into the boundary
conditions (3.37) and (3.56) at z = 0 (with ∂/∂n = ∂/∂z), we see that they are indeed
satisfied (so that Eq. (3.66) does express the unique solution of the boundary
problem), provided that the effective charges q′ and qʺ obey the following relations:
q − q′ = κq″ , q + q′ = q″ . (3.67)
Solving this simple system of linear equations, we obtain
κ−1 2
q′ = − q, q″ = q. (3.68)
κ+1 κ+1
If κ → 1, then q′ → 0, and qʺ → q—both facts are very natural, because in this limit
(no polarization at all!) we have to recover the unperturbed field of the initial point
charge in both semi-spaces. In the opposite limit κ → ∞ (which, according to our
discussion of the last problem, should correspond to a conducting half-space), q′→
−q (repeating the result we have discussed in much detail in section 2.9) and qʺ → 0.
The last result means that in this limit, the electric field E in the dielectric tends to
zero—as it should.
3.5 Energy of electric field in a dielectric

In chapter 1, we have obtained two key results for the electrostatic energy: Eq. (1.55)
for a charge interaction with an independent (‘external’) field, and a similarly
structured formula (1.60), but with an additional factor ½, for the field produced by
the charges under consideration. These relations are of course always valid for
dielectrics as well, provided that the charge density includes all charges (including
3-23
those bound into the elementary dipoles), but it is convenient to recast them into a
form depending on the density ρ(r) of only stand-alone charges.
If a field is created only by stand-alone charges under consideration, and is
proportional to ρ(r) (requiring that we deal with linear dielectrics only), we can
repeat all the argumentation of the beginning of section 1.3, and again arrive at
Eq. (1.60), provided that ϕ is now the macroscopic field’s potential. Now we can
recast this result in the terms of fields—essentially as was done in Eqs. (1.62)–(1.64),
but now making a clear difference between the macroscopic electric field E = −∇ϕ
and the electric displacement field D that obeys the macroscopic Maxwell equation
(3.32). Plugging ρ(r), expressed from that equation, into Eq. (1.60), we obtain
1
U=
2
∫ (∇ ⋅ D) ϕ d 3r . (3.69)
Using the fact30 that for any differentiable functions ϕ and D,

(∇ ⋅ D) ϕ = ∇ ⋅ (ϕD) − (∇ϕ) ⋅ D , (3.70)
we may rewrite Eq. (3.69) as
1
U=
2
∫ ∇ ⋅ (ϕ D) d 3r − 12 ∫ (∇ϕ) ⋅ D d 3r. (3.71)
The divergence theorem, applied to the first term on the right-hand side, reduces it to
a surface integral of ϕDn. (As a reminder, in Eq. (1.63) the integral was of ϕ(∇ϕ)n ∝
ϕEn.) If the surface of the volume we are considering is sufficiently far, this surface
integral vanishes. On the other hand, the gradient in the second term of Eq. (3.71) is
just (minus) field E, so that it gives
1
U=
2
∫ E ⋅ D d 3 r = 12 ∫ E (r) ⋅ ε(r)E (r) d 3 r = ε20 ∫ κ(r)E 2(r) d 3 r. (3.72)
This expression is a natural generalization of Eq. (1.65), and shows that we can, as
we did in the free space, represent the electrostatic energy in a local form31
1 ε D2
U= ∫ u(r)d 3r, with u =
2
E ⋅ D = E2 =
2 2ε
. (3.73)
As a sanity check, in the trivial case ε = ε0 (i.e. κ = 1), this result is reduced to
Eq. (1.65).
Again, Eq. (3.73) is not valid for nonlinear dielectrics, because our starting point,
Eq. (1.62), is only valid if ϕ is proportional to ρ. In order to make our calculation
more general, we should intercept the calculations of section 1.3 at an earlier stage at
which we have not yet used this proportionality. For example, the first of Eqs. (1.56)
may be rewritten, in the continuous limit, as
30
See, e.g. Eq. (A.74a).
31
In the Gaussian units the last expression should be divided by 4π.
3-24
δU = ∫ ϕ(r)δρ(r)d 3r, (3.74)
where symbol δ means a small variation of the function, e.g. its change in time,
sufficiently slow to ignore the relativistic and magnetic-field effects. Applying such
variation to Eq. (3.32) and plugging the resulting relation δρ = ∇· δD into Eq. (3.74),
we obtain
δU = ∫ (∇ ⋅ δ D)ϕd 3r. (3.75)
(Note that in contrast to Eq. (3.69), this expression does not have the front factor ½.)
Now repeating the same calculations as in the linear case, for the energy density
variation we get a remarkably simple (and general!) expression,
3
δu = E ⋅ δ D ≡ ∑EjδDj , (3.76)
j=1
where the last expression uses the Cartesian components of the vectors E and D. This
is as far as we can go for the general dependence D(E). If the dependence is linear
and isotropic, as in Eq. (3.46), then δD = εδE and
⎛E2 ⎞
δu = ε E ⋅ δ E ≡ εδ⎜ ⎟ . (3.77)
⎝ 2 ⎠
Integration of this expression over the variation, from the field equal to zero to a
certain final distribution E(r), brings us back to Eq. (3.73).
An important role of Eq. (3.76), in its last form, is to indicate that the Cartesian
coordinates of E may be interpreted as generalized forces, and those of D as
generalized coordinates of the field effect on a unit dielectric’s volume32. This allows
one, in particular, to form the proper Gibbs potential energy33 of a system with an
electric field E(r) fixed, at every point, by some external source:
UG = ∫V u G(r)d 3r, u G(r) = u(r) − E(r) ⋅ D(r). (3.78)
As an analytical mechanics reminder, if a generalized external force (in our case,

E) is fixed, the stable equilibrium of the system corresponds to the minimum of UG,
rather than of the potential energy U as such—in our case, that of the field in our
system. As the simplest illustration, let us consider a very long cylinder (with an
32
This is one of the cases where the SI units, prescribing different dimensionalities to the fields E and D, are
more revealing than the Gaussian units.
33
See, e.g. Part CM section 1.4, in particular Eq. (1.41), and Part CM section 2.1. Note that as Eq. (3.78)
clearly illustrates, once again, that the difference between the potential energies UG and U, usually discussed in
courses of statistical physics and/or thermodynamics as the difference between the Gibbs and Helmholtz free
energies (see, e.g. Part SM section 1.6), is more general than the effects of random thermal motion, addressed
by these disciplines.
3-25
Figuree 3.13. A cylindrical dielectric sample in a longitudinal external electric field.
arbitrary cross-section’s shape), made of a uniform linear dielectric, placed into a

uniform external electric field, parallel to the cylinder’s axis—see figure 3.13.
The equilibrium value of D inside the cylinder may, of course, be readily found
without any appeal to energies. Indeed, the solution of the Laplace equation inside
the cylinder, with the boundary condition (3.37), is evident: E(r) = Eext, and so that
Eq. (3.46) immediately yields D(r) = εEext. However, one may wonder why the
minimum of the potential energy U, given by Eq. (3.73) in its last form,
U D2
= , (3.79)
V 2ε
corresponds to a different (zero) value of D. The Gibbs potential energy (3.78)
immediately removes the contradiction. For our uniform case, this energy per unit
volume of the cylinder is
UG U D2
3 ⎛ D j2 ⎞
= −E⋅D= −E⋅D≡ ∑⎜⎜ − EjDj ⎟⎟ , (3.80)
j = 1⎝ ⎠
V V 2ε 2ε
and its minimum as a function of every Cartesian component of D corresponds to

the correct value of the displacement: Dj = εEj, i.e. D = εE = εEext. So, the minimum
of the Gibbs potential energy indeed corresponds to the systems’ equilibrium, and it
may be very useful for analyses of the polarization dynamics. Note also that
Eq. (3.80) at this equilibrium point may be rewritten as
UG U D2 D D2
= −E⋅D= − ⋅D≡− , (3.81)
V V 2ε ε 2ε
i.e. formally coincides with Eq. (3.79), besides the opposite sign. Another useful
general relation (not limited to linear dielectrics) may be obtained by taking the
variation of the uG expressed by Eq. (3.78), and then using Eq. (3.76):
δu G = δu − δ(E ⋅ D) = E ⋅ δ D − (δ E ⋅ D + E ⋅ δ D) ≡ −D ⋅ δ E. (3.82)
3-26
In order to see how do these expressions (with their perhaps counter-intuitive

negative signs34) work, let us plug D from Eq. (3.33):
⎛ε E2 ⎞
δu G = −(ε0E + P) ⋅ δ E ≡ −δ⎜ 0 ⎟ − P ⋅ δ E. (3.83)
⎝ 2 ⎠
So far, this relation is general. In the particular case when the polarization P is field-
independent, we may integrate Eq. (3.83) over the electric field, from 0 to some finite
value E, obtaining
ε0E 2
uG = − − P ⋅ E. (3.84)
2
Again, the Gibbs energy is relevant only if E is dominated by an external field Eext,
independent of the orientation of the polarization P. If, in addition, P(r) ≠ 0 only in
some finite volume V, we may integrate Eq. (3.84) over the volume, obtaining
UG = −p ⋅ Eext + const, with p ≡ ∫V P(r)d 3r, (3.85)
where ‘const’ means the terms independent of p. In this expression, we may readily
recognize Eq. (3.15a) for an electric dipole p, which was obtained in section 3.1 in a
different way.
This comparison shows again that UG is nothing extraordinary; it is just the
relevant part of the potential energy of the system in a fixed external field, including
the energy of its interaction with the field. Still, I would strongly recommend the
reader obtain a better gut feeling of the relation between the two potential energies,
U and UG—for example, by using them to solve a very simple problem: calculate the
force of attraction between the plates of a plane capacitor.
3.6 Problems
Problem 3.1. Prove Eqs. (3.3) and (3.4), starting from Eqs. (1.38) and (3.2).
Problem 3.2. A planar thin ring of radius R is charged with a constant linear density
λ. Calculate the exact electrostatic potential distribution along the symmetry axis of
the ring, and prove that at large distances, r ≫ R, the three leading terms of its
multipole expansion are indeed correctly described by Eqs. (3.3)–(3.4).
Problem 3.3 In suitable reference frames, calculate the dipole and quadrupole
moments of the following systems (see the figures below):
(i) four point charges of the same magnitude, but alternating signs, placed in the
corners of a square;
(ii) a similar system, but with a pair charge sign alternation; and
34
Some psychological relief may be provided by the fact that you may add to UG (and U) any constant—
positive if you like.
3-27
(iii) a point charge in the center of a thin ring carrying a similar but opposite charge,
uniformly distributed along its circumference.
Problem 3.4. Without carrying out an exact calculation, can you predict the spatial
dependence of the interaction between various electric multipoles, including point
charges (in this context, frequently called monopoles), dipoles, and quadrupoles?
Based on these predictions, what is the functional dependence of the interaction
between dumbbell-shaped diatomic molecules such as H2, N2, O2, etc, on the
distance between them, if the distance is much larger than the molecular size?
Problem 3.5. Two similar electric dipoles of fixed magnitude p, located at a fixed
distance r from each other, are free to rotate, changing their directions. What stable
equilibrium position(s) may they take as a result of their electrostatic interaction?
Problem 3.6. An electric dipole is located above an infinite, grounded conducting

plane (see the figure below). Calculate:
(i) the distribution of the induced charge in the conductor,
(ii) the dipole-to-plane interaction energy, and
(iii) the force and the torque acting on the dipole.
Problem 3.7. Calculate the net charge Q induced in a grounded conducting sphere of
radius R by a dipole p located at point r outside the sphere—see the figure below.
3-28
Problem 3.8. Use two different approaches to calculate the energy of interaction
between a grounded conductor and an electric dipole p, placed in the center of a
spherical cavity of radius R, carved in the conductor.
Problem 3.9. A plane separating two parts of otherwise free space is densely and
uniformly (with a constant areal density n) filled with dipoles, with similar dipole
moments p oriented in a direction normal to the plane.
(i) Calculate the boundary conditions for the electrostatic potential on both sides of
the plane.
(ii) Use the result of task (i) to calculate the potential distribution created in space
by a spherical surface, with radius R, densely and uniformly filled with radially
oriented dipoles.
(iii) What condition should be imposed on the dipole density n for your results to be
qualitatively valid?
Problem 3.10. Prove the Clausius–Mossotti relation (3.52) for the case of a cubic
lattice of similar dipoles obeying (3.48): p = αEm, where Em is the microscopic
electric field at the dipole’s location point.
Hint: Use Eq. (3.65) to account for the difference between the external field and the
macroscopic field.
Problem 3.11. A sphere of radius R is made of a material with a uniform, fixed

polarization P0.
(i) Calculate the electric field everywhere in space—both inside and outside the sphere.
(ii) Compare the result for the internal field with Eq. (3.24).
Problem 3.12. Calculate the electric field at the center of a cube with side a, made of
material with the uniform spontaneous polarization vector P0 parallel to one of
cube’s sides.
Problem 3.13. A stand-alone charge Q is distributed, in some way, in the volume of a

body made of a uniform linear dielectric with a dielectric constant κ. Calculate the
polarization charge Qef residing on the surface of the body, provided that it is
surrounded by free space.
Problem 3.14. In two separate experiments, a thin, planar sheet of a linear dielectric
with κ = const is placed into a uniform external electric field E0:
(i) with sheet’s surface parallel to the electric field, and
(ii) the surface normal to the field.
3-29
For each case, find the electric field E, the electric displacement D, and the
polarization P inside the dielectric (far from sheet’s edges).
Problem 3.15. A point charge q is located at a distance r ≫ R from the center of a

uniform sphere of radius R, made of a uniform linear dielectric. In the first non-
vanishing approximation in small parameter R/r, calculate the interaction force, and
the energy of interaction between the sphere and the charge.
Problem 3.16. A fixed dipole p is placed in the center of a spherical cavity of radius
R, cut inside a uniform, linear dielectric. Calculate the electric field distribution
everywhere in the system (both at r < R and at r > R).
Hint: You may start with the assumption that the field at r > R has a distribution
typical for a dipole (but be ready for surprises :-).
Problem 3.17. A spherical capacitor (see the figure below) is filled with a linear
dielectric whose permittivity ε depends on spherical angles θ and φ, but not on the
distance r from the system’s center. Derive an explicit expression for its capacitance C.
Problem 3.18. For each of the two capacitors shown in figure 3.10, calculate the
electric forces (per unit area) exerted on the interface between the dielectrics, in terms
of fields in the system.
Problem 3.19. A uniform electric field E0 has been created (by external sources)
inside a uniform linear dielectric. Find the change of the electric field, created by
cutting out a cavity in the shape of a round cylinder of radius R, with the axis
perpendicular to the external field—see the figure below.
3-30
Problem 3.20. Similar small spherical particles, made of a linear dielectric, are
dispersed in free space with a low concentration n ≪ 1/R3, where R is the particle’s
radius. Calculate the average dielectric constant of such a medium. Compare the
result with the apparent, but wrong, answer
κ − 1 = (κ − 1) nV (WRONG ! )
(where κ is the dielectric constant of the particle’s material and V = (4π/3)R3 is its
volume), and explain the origin of the difference.
Problem 3.21.* Calculate the spatial distribution of the electrostatic potential

induced by a point charge q placed at distance d from a very wide parallel plate,
of thickness D, made of a uniform linear dielectric—see the figure below.
Problem 3.22. Discuss the physical nature of Eq. (3.76). Apply your conclusions to a
material with a fixed (field-independent) polarization P0, and calculate the electric
field’s energy of the uniformly polarized sphere (see problem 3.11).
Problem 3.23. Use Eqs. (3.73) and (3.81) to calculate the force of attraction of the
plane capacitor’s plates (per unit area), for two cases:
(i) the capacitor is charged to a voltage V, and then disconnected from the battery,
and
(ii) the capacitor remains connected to the battery.
References
[1] Jackson J 1999 Classical Electrodynamics 3rd edn (Wiley)
[2] Rabe K, Ahn C and Triscone J-M (eds) 2010 Physics of Ferroelectrics: A Modern Perspective
(Berlin: Springer)
[3] Scott J 2000 Ferroelectric Memories (Berlin: Springer)
[4] Lines M and Glass A 2001 Principles and Applications of Ferroelectrics and Related Materials
(Oxford: Oxford University Press)
3-31
IOP Publishing
Lecture notes
Chapter 4
DC currents
The goal of this chapter is to discuss the laws governing the distribution of stationary
(‘dc’) currents inside conducing media. In the most important case of linear (‘Ohmic’)
conductivity, the partial differential equation governing the distribution is reduced to
the same Laplace and Poisson equations whose solution methods were discussed in
detail in chapter 2—although sometimes with different boundary conditions. Because
of this, the chapter is rather brief.
4.1 Continuity equation and the Kirchhoff laws

Until this point, our discussion of conductors has been limited to the cases when they
are separated by insulators (meaning either the free space or some dielectric media),
preventing any continuous motion of charges from one conductor to another, even if
there is a non-zero voltage (and hence electric field) between them—see figure 4.1a.
Figure 4.1. Two oppositely charged conductors: (a) in the electrostatic situation, (b) at the charge relaxation
through an additional narrow conductor (‘wire’), and (c) in a system sustaining a dc current I.
Now let us connect the two conductors with a wire—a thin, elongated conductor
(figure 4.1b). Then the electric field causes the motion of charges in the wire—from
the conductor with a higher electrostatic potential toward that with a lower

potential, until the potentials equilibrate. Such process is called charge relaxation.
The main equation governing this process may be obtained from the fundamental
experimental fact (already mentioned in section 1.1) that electric charges cannot
appear or disappear—though opposite charges may recombine with the conserva-
tion of the net charge. As a result, the charge Q in the top conductor may change
only due to the electric current I through the wire:
dQ
= −I (t ), (4.1)
dt
the relation that may be understood as the definition of the current1.
Let us express Eq. (4.1) in a differential form, introducing the notion of the
current density vector j(r). This vector may be defined via the following relation for
the elementary current dI crossing an elementary area dA (figure 4.2):
dI = jdA cos θ = (j cos θ )dA = jn dA, (4.2)
where θ is the angle between the direction normal to the surface and the carrier
motion direction (which is taken for the direction of vector j).
Figure 4.2. The current density vector j.
With that definition, Eq. (4.1) may be re-written as

d
dt V
∫
ρ d 3r = − jn d 2r,
S
∮ (4.3)
where V is an arbitrary stationary volume limited by the closed surface S. Applying

to this volume the same divergence theorem as was repeatedly used in previous
chapters, we obtain
⎡ ⎤
∫V ⎢⎣ ∂∂ρt + ∇ ⋅ j⎥⎦d 3r = 0. (4.4)
Since volume V if arbitrary, this equation may be true only if

∂ρ
+ ∇ ⋅ j = 0. (4.5)
∂t
1
Just as a (hopefully, unnecessary) reminder, in the SI units the current is measured in amperes (A). In the legal
metrology, the ampere (rather than the coulomb, which is defined as 1 C = 1 A × 1 s) is a primary unit. (Its
formal definition will be discussed in the next chapter.) In the Gaussian units, Eq. (4.1) remains the same, so
that the current’s unit is the statcoulomb per second—the so-called statampere.
4-2
This is the fundamental continuity equation—which is true even for the time-
dependent phenomena2.
The charge relaxation, such as illustrated by figure 4.1b, is of course a dynamic,
time-dependent process. However, electric currents may also exist in stationary
situations, when a certain current source, for example a battery, drives the current
against the electric field, and thus replenishes the conductor charges and sustains
currents at a certain time-independent level—see figure 4.1c. (This process requires a
persistent replenishment of the electrostatic energy of the system from either a source
or a large storage of energy of a different kind—say, the chemical energy of the
battery.) Let us discuss the laws governing the distribution of such dc currents. In this
case (∂/∂t = 0), Eq. (4.5) reduces to a very simple equation
∇ ⋅ j = 0. (4.6)
This relation acquires an even simpler form in the particular but important case of
dc electric circuits (figure 4.3), the systems that may be represented as connections of
components of two types:
(i) small-size (lumped) circuit elements (also called ‘two-terminal devices’), mean-
ing a passive resistor, a current source, etc—generally, any ‘black box’ with two or
more terminals, and
(ii) perfectly conducting wires, with negligible drop of the electrostatic potential along
them, that are galvanically connected at certain points, called nodes (or ‘junctions’).
Figure 4.3. A typical system obeying the Kirchhoff laws.
In the standard circuit theory, the electric charges of the nodes are considered
negligible3, and we may integrate Eq. (4.6) over the closed surface drawn around any
node to get a simple equality
2
Similar differential relations are valid for the density of any conserved quantity, for example for mass in the
classical fluid dynamics (see, e.g. Part CM section 8.3), and for the probability in the statistical physics (Part
SM section 5.6) and quantum mechanics (Part QM section 1.4).
3
In many cases, the charge accumulation may be described without an explicit violation of Eq. (4.7a), but just
by adding other circuit elements, lumped capacitors (see figure 2.5 and its discussion), to the circuit under
analysis. The resulting circuit may be used to describe not only the transient processes of charge accumulation/
relaxation, but also ac currents. However, it will be more convenient for me to postpone the discussion of such
ac circuits until chapter 6, where one more circuit element type, lumped inductances, will be introduced.
4-3
∑I j = 0, (4.7a )
j
where the summation is over all the wires (numbered with index j) connected in the
node. On the other hand, according to its definition (2.25), the voltage Vk across
each circuit element may be represented as the difference of the electrostatic
potentials of the adjacent nodes, Vk = ϕk − ϕk−1. Summing such differences around
any closed loop of the circuit (figure 4.3), we obtain all terms cancelled, so that
∑Vk = 0. (4.7b)
k
These relations are called, respectively, the first and second Kirchhoff laws—or
sometimes the node rule (4.7a) and the loop rule (4.7b). They may seem elementary,
and their genuine power is in the mathematical fact that a set of Eqs. (4.7), covering
every node and every circuit element of the system, gives a system of equations
sufficient for the calculation of all currents and voltages in it—provided that the
relation between current and voltage is known for each circuit element.
It is almost evident that in the absence of current sources, the system of equations
(4.7) has only a trivial solution: Ij = 0, Vk = 0—with the exotic exception of
superconductivity, to be discussed in section 6.3. The current sources, that allow
non-vanishing current flows, may be described by their electromotive forces (e.m.f.)
V k , having the dimensionality of voltage, which have to be taken into account in the
corresponding terms Vk of the sum (4.7b). Let me hope that the reader has some
experience of using Eqs. (4.7) for analyses of simple circuits—say consisting of
several resistors and batteries, so that I can save time by skipping their discussion.
Still, due to their practical importance, I would recommend the reader to carry out a
self-test by solving the two problems given at the beginning of section 4.6.
4.2 The Ohm law

As was mentioned above, the relations spelled out in section 4.1 are sufficient for
forming a closed system of equations for finding currents and an electric field in a
system only if they are complemented with some constitutive relations between the
scalars I and V in each lumped circuit element, i.e. between the vectors j and E in
each point of the material of such an element. The simplest, and most frequently met
relation of this kind is the famous Ohm law whose differential (or ‘local’) form is
j = σ E, (4.8)
where σ is a constant called the Ohmic conductivity (or just the ‘conductivity’ for
short)4. Although this is not a fundamental relation, and is approximate for any
conducting medium, we can argue that if:
4
In SI units, the conductivity is measured in S m−1, where one siemens (S) is the reciprocal of one ohm: 1 S ≡
(1 Ω)−1 ≡ 1 A/1 V. The constant reciprocal to conductivity, 1/σ, is called resistivity, and is commonly denoted by
letter ρ. I will, however, try to avoid using this notion, because I am already overusing this letter in these notes.
4-4
Table 4.1. Ohmic conductivities for some representative (or practically important) materials at 20 °C.
Material σ (S/m)
Teflon ([C2F4]n) 10−22–10−24

Silicon dioxide 10−16–10−19
Various glasses 10−10–10−14
Deionized water ∼10−6
Sea water 5
Silicon n-doped to 1016 cm−3 2.5 × 102
Silicon n-doped to 1019 cm−3 1.6 × 104
Silicon p-doped to 1019 cm−3 1.1 × 104
Nichrome (alloy 80% Ni + 20% Cr) 0.9 × 106
Aluminum 3.8 × 107
Copper 6.0 × 107
Zinc crystal along a-axis 1.65 × 107
Zinc crystal along c-axis 1.72 × 107
(i) there is no current at E = 0 (mind superconductors!),

(ii) the medium is isotropic or almost isotropic (a notable exception: some
organic conductors),
(iii) the mean free path l of current carriers is much smaller than the character-
istic scale a of the spatial variations of j and E,
then the Ohm law may be viewed as a result of the Taylor expansion of the local
relation j(E) for relatively small fields, and thus is very common.
Table 4.1 gives the experimental values of the dc conductivity for some practically
important (or just representative) materials. The reader can see that the range of its
values is very broad, covering more than 30 orders of magnitude, even without going
to such extremes as very pure metallic crystals at very low temperatures, where σ
may reach ∼1012 s m−1.
In order to obtain some feeling as to what these values mean, let us consider a
very simple system (figure 4.4): a plane capacitor of area A ≫ d2, filled with a
material that has not only a dielectric constant κ, but also some Ohmic conductivity
σ, with much more conductive electrodes.
Figure 4.4. A ‘leaky’ plane capacitor.
4-5
Assuming that these properties are compatible with each other5, we may assume
that the distribution of the electric potential (not too close to the capacitor’s edges)
still obeys Eq. (2.39), so that the electric field is normal to the plates and uniform,
with E = V/d. Then, according to Eq. (4.6), the current density is also uniform, j =
σE = σV/d. From here, the total current between the plates is
V
I = jA = σEA = σ A. (4.9)
d
On the other hand, from Eqs. (2.26) and (3.45), the instant value of the plate charge
is Q = CV = (κε0A/d)V. Plugging these relations into Eq. (4.1), we see that the speed
of charge (and voltage) relaxation does not depend on the geometric parameters A
and d of the capacitor:
dV V ε0κ ε
=− , with τr ≡ ≡ , (4.10)
dt τr σ σ
so that where the relaxation time constant τr may be used to characterize the gap
filling material as such.
As we already know (see table 3.1), for most practical materials the dielectric
constant κ is within one order of magnitude from 10, so that the nominator of
Eq. (4.10) is of the order of 10−10 (SI units). As a result, according to table 4.1, the
charge relaxation time ranges from ∼1014 s (more than a million years!) for the best
insulators such as teflon, to ∼10−18 s for the least resistive metals. What is the physics
behind such a huge range of σ, and why, for some materials, does table 4.1 give them
such a large uncertainty? As in chapters 2 and 3, I have time only for a brief, and
admittedly superficial, discussion of these issues6.
If the charge carriers move as classical particles (e.g. in plasmas or non-
degenerate semiconductors), a very reasonable description of the conductivity is
given by the famous Drude formula7. In his picture, due to a weak electric field, the
charge carriers are accelerated in its direction (possibly on top of their random
motion in all directions, i.e. with a vanishing average velocity vector),
dv q
= E, (4.11)
dt m
and as a result their velocity acquires the average value
dv q
v = τ = Eτ , (4.12)
dt m
where the phenomenological parameter τ = l/v (not to be confused with τr!) may be
understood as the effective average time between carrier scattering events. From
here, the current density8:
5
As will be discussed in chapter 6, such simple analysis is only valid if σ is not too high.
6
A more detailed discussion may be found in Part SM chapter 6.
7
It was suggested by P Drude in 1900.
8
Note that j is usually defined as a macroscopic variable, by taking the area dA in Eq. (4.2) much larger than
the square of inter-particle distances, so that no additional average sign is necessary in Eq. (4.13a).
4-6
q 2nτ q 2nτ
j = qn v = E, i.e. σ = . (4.13a )
m m
(Notice the independence of σ of the carrier charge sign.) Another form of the same
result, more popular in the physics of semiconductors, is
τ
σ = q 2nμ, with μ = , (4.13b)
m
where the parameter μ, defined by relation 〈v〉 ≡ μE, is called the charge carrier
mobility.
Most good conductors (e.g. metals) are essentially degenerate Fermi gases (or
liquids), in which the average thermal energy of a particle, kBT is much lower that
the Fermi energy εF. In this case, a quantum theory is needed for the calculation of σ.
Such a theory was developed by the godfather of quantum physics A Sommerfeld in
1927 (and is sometimes called the Drude–Sommerfeld model). I have no time to
discuss it in this course9, and here will only note that for an ideal, isotropic Fermi gas
the result is reduced to Eq. (4.13), with a certain effective value of τ, so it may be
used for estimates of σ, with due respect to the quantum theory of scattering. In a
typical metal, n is very high (∼1023 cm−3) and is fixed by the atomic structure, so that
the sample quality may only affect σ via the scattering time τ.
At room temperature, the scattering of electrons by thermally excited lattice
vibrations (phonons) dominates, so that τ and σ are high but finite, and do not change
much from one sample to another. (Hence, the more accurate values given for
metals in table 4.1.) On the other hand, at T → 0, a perfect crystal should not exhibit
scattering at all, and conductivity should be infinite. In practice, this is never true
(for example, due to electron scattering from imperfect boundaries of finite-size
samples), and the effective conductivity σ is infinite (or practically infinite, at least
above the measurable value ∼1020 s m−1) only in superconductors10.
On the other hand, the conductivity of quasi-insulators (including the deionized
water) and semiconductors depends mostly on the carrier density n, which is much
lower than in metals. From the point of view of quantum mechanics, this happens
because the ground-state eigenenergies of charge carriers are localized within an
atom (or molecule) and separated from excited states, with space-extended wave-
functions, by a large energy gap (called the bandgap). For example, in SiO2 the
bandgap approaches 9 eV, equivalent to ∼4000 K. This is why, even at room
temperature, the density of thermally excited free charge carriers in good insulators
is negligible. In these materials, n is determined by impurities and vacancies, and
may depend on a particular chemical synthesis or other fabrication technology,
rather than on fundamental properties of the material. (In contrast, the carrier
mobility μ in these materials is almost technology-independent.)
9
For such a discussion see, e.g. Part SM section 6.3.
10
The electrodynamic properties of superconductors are so interesting (and fundamentally important) that I
will discuss them in more detail in chapter 6.
4-7
The practical importance of the technology may be illustrated by the following

example. In the cells of the so-called floating-gate memories, in particular the flash
memories which currently dominate non-volatile digital memory technology, data
bits are stored as small electric charges (Q ∼ 10−16 C) of highly doped silicon islands
(so-called floating gates) separated from the rest of the integrated circuit with a ∼10
nm thick layer of the silicon dioxide, SiO2. Such layers are fabricated by high-
temperature oxidation of virtually perfect silicon crystals. The conductivity of the
resulting high-quality (though amorphous) material is so low, σ ∼ 10−19 s m−1, that
the relaxation time τr, defined by Eq. (4.10), is well above 10 years—the industrial
standard for data retention in non-volatile memories. In order to appreciate how
good this technology is, the cited value should be compared with the typical
conductivity σ ∼ 10−16 s m−1 of the usual, bulk SiO2 ceramics11.
4.3 Boundary problems

For an Ohmic conducting medium, we may combine equations (4.6) and (4.8) to
obtain the following differential equation
∇ ⋅ (σ ∇ϕ) = 0. (4.14)
For a uniform conductor (σ = const), Eq. (4.14) is reduced to the Laplace equation
for the electrostatic potential ϕ. As we already know from chapters 2 and 3, its
solution depends on the boundary conditions. These conditions depend on the
interface type.
(i) Conductor–conductor interface. Applying the continuity equation (4.6) to a

Gauss-type pillbox at the interface of two different conductors (figure 4.5), we obtain
(jn )1 = (jn )2 , (4.15)
so that if the Ohm law (4.8) is valid inside each medium, then
Figure 4.5. DC current’s ‘refraction’ at the interface between two different conductors.
11
This course is not an appropriate platform to discuss the details of floating-gate memory technology.
However, I think that every educated physicist should know its basics, because such memories are currently the
drivers of all semiconductor integrated circuit technology development, and hence of the entire progress of
information technology. Perhaps the best available book is [1].
4-8
∂ϕ1 ∂ϕ
σ1 = σ2 2 . (4.16)
∂n ∂n
Also, since the electric field should be finite, its potential ϕ has to be continuous
across the interface—the condition that may also be written as
∂ϕ1 ∂ϕ2
= . (4.17)
∂τ ∂τ
Both these conditions (and hence the solutions of the boundary problems using them)
are similar to those for the interface between two dielectrics—cf Eqs. (3.46) and (3.47).
Note that using the Ohm law, Eq. (4.17) may be rewritten as
1 1
(j )1 = (jτ )2 . (4.18)
σ1 τ σ2
Comparing it with Eq. (4.15) we see that, generally, the current density’s magnitude
changes at the interface: j1 ≠ j2. It is also curious that if σ1 ≠ σ2, the current line slope
changes at the interface (figure 4.5), qualitatively to the refraction of light rays in
optics—see chapter 7.
(ii) Conductor–electrode interface. An electrode is defined as a body made of a

‘perfect conductor’, i.e. of a medium with σ → ∞. Then, at a fixed current density at
the interface, the electric field in the electrode tends to zero, and hence it may be
described by equation
ϕ = ϕj = const, (4.19)
where constants ϕj may be different for different electrodes (numbered with index j).
Note that with such boundary conditions, the Laplace boundary problem becomes
exactly the same as in electrostatics—see Eq. (2.35)—and hence we can use all the
methods (and some solutions) of chapter 2 for finding the dc current distribution.
(iii) Conductor–insulator interface. For the description of an insulator, we can use

σ = 0, so that Eq. (4.16) yields the following boundary condition,
∂ϕ
= 0, (4.20)
∂n
for the potential derivative inside the conductor. From the Ohm law (4.8) in the form
j = −σ∇ϕ, we see that this is just the very natural requirement for the dc current not
to flow into an insulator.
Now, note that this condition makes the Laplace problem inside the conductor
completely well-defined, and independent of the potential distribution in the
adjacent insulator. In contrast, due to the continuity of the electrostatic potential
at the border, its distribution in the insulator has to follow that inside the conductor.
Let us discuss this conceptual issue on the following (apparently, trivial) example: dc
4-9
current in a uniform wire of length l and a cross-section of area A. The reader

certainly knows the answer:
V V l
I= , where R ≡ = , (4.21)
R I σA
where the constant R is called the resistance12. However, let us derive this result
formally from our theoretical framework. For the simple geometry shown in figure
4.6a, this is easy to do. Here the potential evidently has a linear 1D distribution
x
ϕ = const − V , (4.22)
l
both in the conductor and the surrounding free space, with both boundary conditions
(4.16) and (4.17) satisfied at the conductor–insulator interfaces, and the condition
(4.20) satisfied at the conductor–electrode interface. As a result, the electric field is
constant and has only one component Ex = V/l, so that inside the conductor
jx = σEx, I = jx A , (4.23)
giving us the well-known Eq. (4.21).
Figure 4.6. (a) An elementary problem and (b) a (slightly) less obvious problem of the field distribution at dc
current flow (schematically).
However, what about the geometry shown in figure 4.6b? In this case the field
distribution in the free space around the conductor is dramatically different, but
according to boundary problem defined by equations (4.14) and (4.20), inside the
conductor the solution is exactly the same as it was in the former case. Now, the
Laplace equation in the surrounding insulator has to be solved with the boundary
values of the electrostatic potential, ‘dictated’ by the distribution of the current (and
hence potential) in the conductor. Note that as the result, the electric field lines are
12
The first of Eqs. (4.21) is essentially the integral form of the Ohm law (4.8), and is valid not only for a
uniform wire, but also for any Ohmic conductor with a geometry in which I and V may be clearly defined.
4-10
generally not normal to the conductor’ surface, because the surface is not
equipotential.
Let us solve a problem in which this conduction hierarchy may be followed
analytically to the very end. Consider an empty spherical cavity cut in a conductor
with an initially uniform current flow with a constant density j0 = nzj0 (figure 4.7a).
Following the conduction hierarchy, we have to solve the boundary problem in the
conducting part of the system, i.e. outside the sphere (r ⩾ R), first. Since the problem
is evidently axially symmetric, we already know the general solution of the Laplace
equation—see Eq. (2.172). Moreover, we know that in order to match the uniform
field distribution at r → ∞, all coefficients al but one (a1 = −E0 = −j0/σ) have to be
zero, and that the boundary conditions at r = R will give zero solutions for all
coefficients bl but one (b1), so that
j b
ϕ = − 0 r cos θ + 21 cos θ , for r ⩾ R . (4.24)
σ r
In order to find the coefficient b1, we have to use the boundary condition (4.20) at r = R:
∂ϕ ⎛ j 2b ⎞
= ⎜ − 0 − 31 ⎟ cos θ = 0. (4.25)
∂r r=R ⎝ σ R ⎠
This gives b1 = −j0R3/2σ, so that, finally,
j⎛ R3 ⎞
ϕ(r , θ ) = − 0 ⎜r + 2 ⎟ cos θ , for r ⩾ R . (4.26)
σ⎝ 2r ⎠
(Note that this potential distribution corresponds to the dipole moment p = −E0R3/2.
It is straightforward to check that if the spherical cavity was cut in a dielectric, the
potential distribution outside it would be similar, with p = −E0R3(κ − 1)/(κ + 2). In the
limit κ → ∞, these two results coincide, despite the rather different type of problem: in
the dielectric case, there is no current at all.)
Now, as the second step in the conductivity hierarchy, we may find the electro-
static potential distribution ϕ(r,θ) in the insulator, in this particular case inside the
Figure 4.7. A spherical cavity cut in a uniform conductor: (a) the problem’s geometry and (b) the equipotential
surfaces as given by equations (4.26) and (4.28).
4-11
empty cavity (r ⩽ R). It should also satisfy the Laplace equation with the boundary
conditions at r = R, ‘dictated’ by the distribution (4.26):
3 j0
ϕ(R , θ ) = − R cos θ . (4.27)
2σ
We could again solve this problem by the formal variable separation (keeping in the
general solution (2.172) only the term proportional to al, which does not diverge at r
→ 0), but if we notice that the boundary condition (4.27) depends on just one
Cartesian coordinate, z = R cos θ, the solution may be just guessed:
3 j0 3j
ϕ(r , θ ) = − z = − 0 r cos θ , at r ⩽ R . (4.28)
2σ 2σ
It evidently satisfies the Laplace equation and the boundary condition (4.27), and
corresponds to a constant electric field parallel to the vector j0, and equal to 3j0/2σ—
see figure 4.7b. Again, the cavity surface is not equipotential, and the electric field
lines at r ⩽ R are not normal to it at almost all points.
The conductivity hierarchy says that static electrical fields and charges outside
conductors (e.g. electric wires) do not affect currents flowing in the wires, and it is
physically very clear why. For example, if a charge in the free space is slowly moved
close to a wire, it (in accordance with the linear superposition principle) will only
induce an additional surface charge (see chapter 2) that screens the external charge’s
field, without participating in the current flow inside the conductor.
Apart from this conceptual issue, the two examples given above may be considered
as the further demonstrations of the first two methods discussed in chapter 2 (the
orthogonal coordinates (figure 4.6) and the variable separation (figure 4.7)) to dc
current distribution problems. If we have a glance at other methods discussed in that
chapter, we may notice that there is also an analog of the method of charge images.
Indeed, let us consider the spherically symmetric potential distribution of the electro-
static potential, similar to that given by the basic Eq. (1.35):
c
ϕ= . (4.29)
r
As we know from chapter 1, this is a particular solution of the 3D Laplace equation
at all points but r = 0. In the free space, this distribution would correspond to a point
charge q = 4πε0c; but what about the conductor? Calculating the corresponding
electric field and current density,
c c
E = −∇ϕ = 3 r , j = σ E = σ 3 r, (4.30)
r r
we see that the total current flowing from the origin through a sphere of an arbitrary
radius r does not depend on the radius:
I = Aj = 4πr 2j = 4πσ c . (4.31)
4-12
Figure 4.8. Applying the method of images for the current injection analysis.
Plugging the resulting coefficient c into Eq. (4.29), we obtain

I
ϕ= . (4.32)
4πσr
Hence the Coulomb-type distribution of the electric potential in a conductor is
possible (at least at some distance from the singular point r = 0), and describes the dc
current I flowing out of a small-size electrode—or into such a point, if the coefficient
c is negative. Such current injection may be readily implemented experimentally;
think for example about an insulated wire with a small bare end, inserted into a
poorly conducting soil—an important method in geophysical research13.
Now let the current injection point r′ be close to a plane interface between the
conductor and an insulator (figure 4.8). In this case, besides the Laplace equation,
we should satisfy the boundary condition,
∂ϕ
jn = σEn = −σ = 0, (4.33)
∂n
at the interface. It is clear that this can be done by replacing the insulator with an
imaginary similar conductor with an additional current injection point, at the mirror
image point r″. Note, however, that in contrast to the charge images, the sign of the
imaginary current has to be similar, not opposite, to the initial one, so that the total
electrostatic potential inside the conducting semi-space is
I ⎛ 1 1 ⎞
ϕ(r) = ⎜ + ⎟. (4.34)
4πσ ⎝ r − r′ r − r″ ⎠
(The image current’s sign would be opposite at the interface between a conductor
with a moderate conductivity and a perfect conductor (‘electrode’), whose potential
should be virtually constant.)
This result may be readily used, for example, to calculate the current density at a
plane surface of a uniform conductor, as a function of distance ρ from point 0 (the
surface’s point closest to the current injection site)—see figure 4.8. At such surface,
Eq. (4.34) yields
13
Such injection is even simpler in 2D situations—think about a wire soldered, in a small spot, to a thin
conducting foil. (Note that here the current density distribution law is different, j ∝ 1/r rather than 1/r2.)
4-13
I 1
ϕ= , (4.35)
2πσ (ρ + d 2 )1/2
2
so that the current density is:

∂ϕ I ρ
jρ = σEρ = −σ = . (4.36)
∂ρ 2π (ρ 2 + d 2 )3/2
Deviations from equations (4.35) and (4.36) may be used to find and characterize
conductance inhomogeneities, say those due to mineral deposits in the Earth’s crust14.
4.4 Energy dissipation

Let me conclude this brief chapter with an ultra-short discussion of energy
dissipation in conductors. In contrast to the electrostatics situations in insulators
(vacuum or dielectrics), at dc conduction the electrostatic energy U is ‘dissipated’
(i.e. transferred to heat) at a certain rate P ≡ −dU /dt , with the dimensionality of
power15. This so-called dissipation power may be evaluated by calculating the power
of the electric field’s work on a single moving charge:
P 1 = F ⋅ v = q E ⋅ v. (4.37)
After the summation over all charges, Eq. (4.37) gives us the dissipation power. If
the charge density n is uniform, multiplying by it the both parts of this relation, and
taking into account that qnv = j, for the energy dissipation in a unit volume we
obtain the Joule law
P PN
p≡ = 1 = P 1n = q E ⋅ vn = E ⋅ j. (4.38)
V V
In the case of the Ohmic conductivity, this expression may also be rewritten in two
other forms:
j2
p = σE 2 = . (4.39)
σ
At the dc conduction, the electrostatic energy has to be permanently replenished by
an equal flow of power from the current source(s). With our electrostatics back-
ground, it is also straightforward (and hence left for the reader’s exercise) to prove that
the dc current distribution in a uniform Ohmic conductor, at a fixed voltage applied at
its borders, corresponds to the minimum of the total dissipation power
P = σ∫ E 2d 3r. (4.40)
V
14
The current injection may be produced, due to electrochemical reactions, by an ore mass itself, so that one
need only measure (and correctly interpret) the resulting potential distribution—the so-called self-potential
method—see, e.g. section 6.1 in [2].
15
Since the electric field and hence the electrostatic energy are time-independent, this means that the energy is
replenished at the same rate from the current source(s).
4-14
4.5 Problems
Problem 4.1. A dc voltage V0 is applied to the end of a semi-infinite chain of lumped
Ohmic resistors, shown in the figure below. Calculate the voltage across the jth link
of the chain.
Problem 4.2. It is well known that properties of many dc current sources (e.g.
batteries) may be reasonably well represented as a connection in series of a perfect
voltage source and an Ohmic internal resistance. Discuss the option, and possible
advantages, of using a different equivalent circuit that would include a perfect
current source.
Problem 4.3. Calculate the resistance between two large, uniform Ohmic conductors
separated with a very thin, planar, insulating partition, with a circular hole of radius
R in it—see the figure below.
Hint: You may like to use the degenerate ellipsoidal coordinates, which had been
used in section 2.4.
Problem 4.4. Calculate the effective (average) conductivity σef of a medium with many
empty spherical cavities of radius R, carved at random positions in a uniform Ohmic
conductor (see the figure below), in the limit of a low density n ≪ R−3 of the spheres.
Hint: Try to use the analogy with a dipole medium—see, e.g. section 3.2.
4-15
Problem 4.5. In two separate experiments, a narrow gap, possibly of an irregular

width, between two close metallic electrodes is filled with some material—in the first
case, with a uniform linear insulator with an electric permittivity ε, and in the second
case, with a uniform conducting material with an Ohmic conductivity σ. Neglecting
the fringe effects, calculate the relation between the mutual capacitance C between
the electrodes (in the first case) and the dc resistance R between them (in the second
case).
Problem 4.6. Calculate the voltage V across a uniform, wide resistive slab of
thickness t, at distance l from the points of injection/pickup of the dc current I passed
across the slab—see the figure below.
Problem 4.7. Calculate the voltage V between two corners of a square cut from a
uniform, resistive sheet of a very small thickness t, induced by dc current I that is
passed between its two other corners—see the figure below.
Problem 4.8. Calculate the distribution of the dc current’s density in a thin, round,
uniform resistive disk, if the current is inserted into a point at its rim, and picked up
at the center.
Problem 4.9.* The simplest model of a vacuum diode consists of two planar, parallel
metallic electrodes of area A, separated by a gap of thickness d ≪ A1/2: a ‘cathode’
which emits electrons to vacuum, and an ‘anode’ which absorbs the electrons
arriving at its surface. Calculate the dc I–V curve of the diode, i.e. the stationary
relation between the current I flowing between the electrodes and the voltage V
applied between them, using the following simplifying assumptions:
(i) due to the effect of the negative space charge of the emitted electrons, the current
I is much smaller than the emission ability of the cathode,
(ii) the initial velocity of the emitted electrons is negligible, and
4-16
(iii) the direct Coulomb interaction of electrons (besides the space charge effect) is
negligible.
Problem 4.10.* Calculate the space-charge-limited current in a system with the same
geometry, and using the same assumptions as in the previous problem, apart from
assuming that now the emitted charge carriers move not ballistically, but drift in
accordance with the Ohm law, with the conductivity given by (4.13): σ = q2μn, with a
constant mobility μ.
Hint: In order to obtain a realistic result, assume that the medium in which the
charge carriers move16 has a certain dielectric constant κ, unrelated to the carriers.
Problem 4.11. Prove that the distribution of dc currents in a uniform Ohmic

conductor, at a fixed voltage applied at its boundaries, corresponds to the minimum
of the total power dissipation (‘Joule heat’).
References
[1] Brewer J and Gill M (eds) 2008 Nonvolatile Memory Technologies with Emphasis on Flash
(Piscataway, NJ: IEEE)
[2] Telford W et al 1990 Applied Geophysics 2nd edn (Cambridge University Press)
16
As was mentioned in section 4.2 of the lecture notes, the assumption of a constant (charge-density-
independent) mobility is most suitable for semiconductors.
4-17
IOP Publishing
Lecture notes
Chapter 5
Magnetism
Despite the fact that this chapter addresses a completely new type of electric charge
interaction, its discussion (for the stationary case) will take not too much time/space,
because it recycles many ideas and methods of electrostatics, although with a twist or two.
5.1 Magnetic interaction of currents

DC currents in conductors usually leave them electroneutral, ρ(r) = 0, with very
good precision, because even a minor disbalance of positive and negative charge
density results in extremely strong Coulomb forces that restore their balance by an
additional shift of free charge carriers1. This is why we start the discussion of
magnetic interactions from the simplest case of two spatially separated, current-
carrying, electroneutral conductors (figure 5.1).
Figure 5.1. The magnetic interaction of two currents.
According to the Coulomb law, there should be no force between them. However,
several experiments carried out in the early 1820s2 proved that such non-Coulomb
1
The most important case when the electroneutrality does not hold is the motion of electrons in vacuum. In
this case, the magnetic forces we are going to discuss coexist with (typically, stronger) electrostatic forces—see
Eq. (5.3) below and its discussion. In some semiconductor devices, local violations of electroneutrality also
play an important role—see, e.g. Part SM chapter 6.
2
Most notably, by H C Ørsted, J-B Biot and F Savart, and A-M Ampère.

forces do exist, and are the manifestation of a different, magnetic interaction

between the currents. In the contemporary notation, their results may be summar-
ized with just one formula, in SI units expressed as3:
μ0
F=−
4π
∫V d 3r∫V ′ d 3r′[j(r) ⋅ j′(r′)] rr −− rr′′3 . (5.1)
Here the coefficient μ0/4π (where μ0 is called either the magnetic constant or the free
space permeability), by definition, equals exactly 10−7 SI units, thus relating the
electric current unit definition to that of force—see below.
Note an almost complete similarity of this expression to the Coulomb law (1.1),
written for a continuous charge distribution, with the account of the linear super-
position principle:
1
F=
4πε0
∫V d 3r∫V ′ d 3r′ρ(r)ρ′(r′) rr −− rr′′3 . (5.2)
Apart from the different coefficient and a different sign, the ‘only’ difference between
Eq. (5.1) and Eq. (5.2) is the scalar product of current densities, evidently necessary
because of the vector character of the current density. We will see that this difference
brings certain complications in applying the approaches discussed in the previous
chapters to magnetostatics.
Before moving on to their discussion, let us have one more glance at the
coefficients in Eqs. (5.1) and (5.2). To compare them, let us consider two objects
with uncompensated charge distributions ρ(r) and ρʹ(r), each moving parallel to each
other as a whole, with certain velocities v and vʹ, as measured in an inertial
(‘laboratory’) frame. In this case, j(r) = ρ(r)v, so that j(r) · jʹ(r) = ρ(r)ρ ʹ(r)vvʹ, and
the integrals in Eqs. (5.1) and (5.2) become functionally similar, and differ only by
the factor
Fmagnetic μ vv′ 1 vv′
=− 0 =− 2. (5.3)
Felectric 4π 4πε0 c
(The last expression holds in any consistent system of units.) We immediately see
that the magnetism is an essentially relativistic phenomenon, very weak in
comparison with the electrostatic interaction at the human scale velocities, v ≪ c,
and may dominate only if the latter interaction vanishes—as it does in electroneutral
systems4.
3
In the Gaussian units, the coefficient μ0/4π is replaced with 1/c2 (i.e. implicitly with μ0ε0) where c is the speed
of light, in modern metrology considered exactly known—see, e.g. appendix B.
4
The discovery and initial studies of such a subtle, relativistic phenomenon as magnetism in the early
nineteenth century was much facilitated by the relative abundance of natural ferromagnets, materials with a
spontaneous magnetic polarization, whose strong magnetic field may be traced back to relativistic effects (such
as spin) in the constitute atoms. Ferromagnetism will be (briefly) discussed in section 5.5 below, and then in
Part SM chapter 4.
5-2
Also, Eq. (5.3) points at an interesting paradox. Consider two electron beams
moving parallel to each other, with the same velocity v with respect to a lab reference
frame. Then, according to Eq. (5.3), the net force of their total (electric plus
magnetic) interaction is proportional to (1 − v2/c2), and tends to zero in the limit
v → c. However, in the reference frame moving together with the electrons, they are
not moving at all, i.e. v = 0. Hence, from the point of view of such a moving
observer, the electron beams should interact only electrostatically, with a repulsive
force independent of the velocity v. Historically, this had been one of several
paradoxes that led to the development of special relativity; its resolution will be
discussed in chapter 9, which is devoted to this theory.
Returning to Eq. (5.1), in some simple cases the double integration in it may be
carried out analytically. First of all, let us simplify this expression for the case of two
thin, long conductors (wires) separated by a distance much larger than their
thickness. In this case we may integrate the products jd3r and jʹd3rʹ over the wires’
cross-sections first, neglecting the corresponding change of (r − rʹ). Since the
integrals of the current density over the cross-sections of the wires are just the
currents I and Iʹ flowing in the wires, and cannot change along their lengths (say,
l and lʹ, respectively), they may be taken out of the remaining integrals, reducing
Eq. (5.1) to
μ0II ′
F=−
4π
∮l ∮l′ (d r ⋅ d r′) rr −− rr′′3 . (5.4)
As the simplest example, consider two straight, parallel wires (figure 5.2),
separated by distance d, with length l ≫ d. In this case, due to symmetry, the vector
of the magnetic interaction force has to:
(i) lie in the same plane as the currents, and
(ii) be perpendicular to the wires—see figure 5.2.
Figure 5.2. Magnetic force between two straight parallel currents.
Hence we can limit our calculations to just one component of the force, normal to
the wires. Using the fact that with the coordinate choice shown in figure 5.2, the
scalar product dr · drʹ is just dxdxʹ, we obtain
5-3
μ0II ′ +∞ +∞
sin θ
F=−
4π −∞
dx ∫−∞
dx′ 2 ∫
d + (x − x′)2
(5.5)
μ II ′ +∞ +∞
d
=− 0
4π −∞
∫
dx
−∞
dx′ ∫
[d + (x − x′)2 ]3/2
2
.
Introducing, instead of xʹ, a new, dimensionless variable ξ ≡ (x − xʹ)/d, we may

reduce the internal integral to a table integral which we have already met in this
course:
μ0II ′ +∞ +∞ μ0II ′ +∞
dξ
F=− ∫−∞ dx ∫−∞ =− ∫−∞ dx . (5.6)
4πd ( 1 + ξ 2 )3/2 2πd
The integral over x formally diverges, but this means merely that the interaction
force per unit length of the wires is constant:
F μ II ′
=− 0 . (5.7)
l 2πd
Note that the force drops rather slowly (only as 1/d) as the distance d between the
wires is increased, and is attractive (rather than repulsive as in the Coulomb law) if
the currents are of the same sign.
This is an important result5, but again, the problems so simply solvable are few
and far between, and it is intuitively clear that we would strongly benefit from the
same approach as in electrostatics, i.e. from breaking Eq. (5.1) into a product of two
factors via the introduction of a suitable field. Such decomposition may be done as
follows:
F= ∫V j(r) × B(r)d 3r, (5.8)
where the vector B is called the magnetic field (in our particular case, induced by
current jʹ)6:
μ0 r − r′ 3
B(r) ≡
4π
∫V ′ j′(r′) × r − r′ 3
d r′ . (5.9)
5
In particular, Eq. (5.7) is used for the legal definition of the SI unit of current, one ampere (A), via the SI unit
of force (the newton, N), with the coefficient μ0 considered exactly fixed as listed above.
6
The SI unit of the magnetic field is called tesla (T) after N Tesla, an electrical engineering pioneer. In
Gaussian units, the already discussed constant 1/c2 in Eq. (5.1) is equally divided between Eqs. (5.8) and (5.9),
so that in them both, the constant before the integral is 1/c. The resulting Gaussian unit of field B is called the
gauss (G); taking into account the difference of units of electric charge and length, and hence the current
density, 1 G equals exactly 10−4 T. Note also that in some textbooks, in particular old ones, B is called either
the magnetic induction, or the magnetic flux density, while the term ‘magnetic field’ is reserved for the
macroscopic vector H, which will be introduced in section 5.5 below.
5-4
The last relation is called the Biot–Savart law, while the force F expressed by Eq.
(5.8) is sometimes called the Lorentz force7. However, more frequently the later term
is reserved for the full force,
F = q(E + v × B), (5.10)
exerted by electric and magnetic fields field on a point charge q, moving with
velocity v.8
Now we have to prove that the new formulation, given by Eqs. (5.8) and (5.9), is
equivalent to Eq. (5.1). At first glance this seems unlikely. Indeed, first of all, Eqs.
(5.8) and (5.9) involve vector products, while (5.1) is based on a scalar product.
More profoundly, in contrast to Eq. (5.1), Eqs. (5.8) and (5.9) do not satisfy
Newton’s third law, applied to elementary current components jd3r and jʹd3rʹ, if
these vectors are not parallel to each other. Indeed, consider the situation shown in
figure 5.3.
Figure 5.3. The apparent violation of Newton’s third law in magnetism.
Here the vector jʹ is perpendicular to the vector (r − rʹ), and hence, according to
Eq. (5.9), produces a non-zero contribution dBʹ to the magnetic field, directed (in
figure 5.3) normal to the plane of drawing, i.e. is perpendicular to the vector j.
Hence, according to Eq. (5.8), this field provides a non-zero contribution to F. On
the other hand, if we calculate the reciprocal force Fʹ by swapping the indices in Eqs.
(5.8) and (5.9), the latter equation immediately shows that dB(rʹ) ∝ j × (r − rʹ) = 0,
because the two operand vectors are parallel (figure 5.3). Hence, the current
component jʹd3rʹ does exert a force on its counterpart, while jd3r does not.
Despite this apparent problem, let us still go ahead and plug Eq. (5.9) into
Eq. (5.8):
μ0 ⎛ r − r′ ⎞
F=
4π
∫V d 3r∫V ′ d 3r′j(r) × ⎜⎝j′(r′) × ⎟.
r − r′ 3 ⎠
(5.11)
7
Named after H Lorentz, already mentioned in section 3.3, but famous mostly for his numerous contributions
to the development of special relativity—see chapter 9. To be fair, the magnetic part of the Lorentz force was
first correctly calculated by O Heaviside.
8
From the magnetic part of Eq. (5.10), Eq. (5.8) may be derived by the elementary summation of all forces
acting on n ≫ 1 particles in a unit volume, with j = qn〈v〉—see the footnote for Eq. (4.13a). On the other hand,
the reciprocal derivation of Eq. (5.10) from Eq. (5.8) with j = qvδ(r − r0), where r0 is the current particle’s
position (so that dr0/dt = v), requires care, and will be performed in chapter 9.
5-5
This double vector product may be transformed into two scalar products,
using the vector algebraic identity called the bac minus cab rule, a × (b × c) =
b(a · c) − c(a · b).9 Applying this relation, with a = j, b = jʹ, and c = R ≡ r − rʹ, to
Eq. (5.11), we obtain
⎛ ⎞
μ0 ⎜ 3 j(r) ⋅ R ⎟
μ0 R
F=
4π
∫ 3
d r′j′(r′)∫⎜
dr
R3 ⎟⎠ 4π
− ∫ d 3r ∫ d 3r′j(r) ⋅ j′(r′)
R3
. (5.12)
V′ ⎝V V V′
The second term on the right-hand side of this relation coincides with the right-hand
side of Eq. (5.1), while the first term equals zero because its internal integral vanishes.
Indeed, we may break the volumes V and Vʹ into narrow current tubes—the stretched
elementary volumes whose walls are not crossed by current lines (so that on their
walls, jn = 0). As a result, the elementary current in each tube, dI = jdA = jd2r, is the
same along its length and, just as in a thin wire, jd2r may be replaced with dIdr, with
the vector dr directed along j. Because of this, each tube’s contribution into the
internal integral in the first term of Eq. (5.12) may be represented as
dI ∮l d r ⋅ RR3 = −dI∮l d r ⋅ ∇ R1 = −dI∮l dr ∂∂r R1 , (5.13)
where the operator ∇ acts in the r space, and the integral is taken along the tube’s
length l. Due to the current continuity, each loop should follow a closed contour,
and an integral of a full differential of some scalar function (in our case, 1/r12) along
such a contour equals zero.
So we have recovered Eq. (5.1). Returning for a minute to the paradox illustrated
with figure 5.3, we may conclude that the apparent violation of Newton’s third law
was the artifact of our interpretation of Eqs. (5.8) and (5.9) as the sums of
independent elementary components. In reality, due to the dc current continuity
expressed by Eq. (4.6), these components are not independent. For the whole
currents, Eqs. (5.8) and (5.9) do obey the third law—as follows from their already
proved equivalence to Eq. (5.1).
Thus we have been able to break the magnetic interaction into two effects: the
induction of the magnetic field B by one current (in our notation, jʹ), and the effect of
this field on the other current (j). Now comes an additional experimental fact: other
elementary components jd3rʹ of the current j(r) also contribute to the magnetic field
(5.9) acting on the component jd3r.10 This fact allows us to drop the prime sign after j
in Eq. (5.9), and rewrite Eqs. (5.8) and (5.9) as
μ0 r − r′ 3
B(r) =
4π
∫V ′ j(r′) × r − r′ 3
d r′ , (5.14)
9
See, e.g. Eq. (A.47).
10
Just in electrostatics, one needs to exercise due caution at a transfer from these expressions to the limit of
discrete classical particles, and extended wavefunctions in quantum mechanics, in order to avoid the (non-
existent) magnetic interaction of a charged particle upon itself.
5-6
F= ∫V j(r) × B(r)d 3r, (5.15)
Again, the field observation point r and the field source point rʹ have to be clearly
distinguished. We immediately see that these expressions are similar to, but still
different from the corresponding relations of the electrostatics, namely Eq. (1.9),
1
E(r) =
4πε0
∮V ′ ρ(r′) rr −− rr′′3 d 3r′, (5.16)
and the distributed-charge version of Eq. (1.5):
F= ∮V ρ(r) E(r)d 3r. (5.17)
(Note that the sign difference has disappeared, at the cost of the replacement of
scalar-by-vector multiplications in electrostatics with cross-products of vectors in
magnetostatics.)
For the frequent case of a field of a thin wire of length lʹ, Eq. (5.14) may be re-
written as
μ0I r − r′
B(r) =
4π
∮ l ′ d r′ × r − r′ 3
. (5.18)
Let us see how this formula works for the simplest case of a straight wire (figure 5.4a).
The magnetic field contribution dB due to any small fragment drʹ of the wire’s length is
directed along the same line (perpendicular to both the wire and the perpendicular d
dropped from the observation point to the wire line), and its magnitude is
μ0I dx′ μI dx′ d
dB = 2
sin θ = 0 . (5.19)
4π r − r ′ 4π (d + x ) (d + x 2 )1/2
2 2 2
Summing up all such elementary contributions, we obtain

μ 0 Iρ ∞ μ0I
B=
4π
∫−∞ (x 2 +dxd 2)3/2 =
2πd
. (5.20)
This is a simple but very important result. (Note that it is only valid for very long
(l ≫ d), straight wires.) It is particularly crucial to note the ‘vortex’ character of the
Figure 5.4. Magnetic fields of (a) a straight current and (b) a current loop.
5-7
field: its lines go around the wire, forming rings with their centers on the current line.
This is in the sharp contrast to the electrostatic field lines that can only begin and end
on electric charges and never form closed loops (otherwise the Coulomb force qE
would not be conservative). In the magnetic case, the vortex field may be reconciled
with the potential character of the magnetic forces, which is evident from Eq. (5.1),
due to the vector products in Eqs. (5.14) and (5.15).
Now we may use Eq. (5.15), or rather its thin-wire version
F=I ∮l d r × B(r), (5.21)
to apply Eq. (5.20) to the two-wire problem (figure 5.2). Since for the second wire
vectors dr and B are perpendicular to each other, we immediately arrive at our
previous result (5.7), which was obtained directly from Eq. (5.1).
The next important example of application of the Biot–Savart law (5.14) is the
magnetic field at the axis of a circular current loop (figure 5.4b). Due to the
problem’s symmetry, the net field B has to be directed along the axis, but each of its
components dB is tilted by angle θ = tan−1(z/R) to this axis, so that its axial
component
μ0I dr′ R
dBz = dB cos θ = . (5.22)
4π R2 + z 2 (R2 + z 2 )1/2
Since the denominator of this expression remains the same for all wire components
drʹ, the integration over rʹ is trivial (∫drʹ = 2πR), giving finally
μ0I R2
B= . (5.23)
2 (R2 + z 2 )3/2
Note that the magnetic field in the loop’s center (i.e. for z = 0),
μ0I
B= , (5.24)
2R
is π times higher than that due to a similar current in a straight wire, at distance d = R
from it. This difference is readily understandable, since all elementary components of
the loop are at the same distance R from the observation point, while in the case of a
straight wire, all its points but one are separated from the observation point by a
distance larger than d.
Another notable fact is that at large distances (z2 ≫ R2), the field (5.23) is
proportional to z−3:
μ 0 I R2 μ 2m
B≈ 3
= 0 3, with m ≡ IA , (5.25)
2 z 4π z
where A = πR2 is the loop area. Comparing this expression with Eq. (3.13), for the
particular case θ = 0, we see that such field is similar to that of an electric dipole (at
least along its direction), with the replacement of the electric dipole moment
magnitude p with m (besides the front coefficient). Indeed, such a plane current
5-8
loop is the simplest example of a system whose field, at a large distance, is that of a
magnetic dipole, with the dipole moment m—the notions that will be discussed in
much more detail in section 5.4.
5.2 Vector-potential and the Ampère law

The reader can see that the calculations of the magnetic field using Eqs. (5.14) or
(5.18) are still cumbersome even for the very simple systems we have examined. As
we saw in chapter 1, similar calculations in electrostatics, at least for several
important systems of high symmetry, could be substantially simplified using the
Gauss law (1.16). A similar relation exists in magnetostatics as well, but has a
different form, due to the vortex character of the magnetic field.
To derive it, let us notice that in an analogy with the scalar case, the vector
product under integral (5.14) may be transformed as
j (r′) × (r − r′) j(r′)
3
=∇× , (5.26)
r − r′ r − r′
where the operator ∇ acts in the r space. (This equality may be really verified by its
Cartesian components, noticing that the current density is a function of rʹ and hence
its components are independent of r.) Plugging Eq. (5.26) into Eq. (5.14), and
moving the operator ∇ out of the integral over rʹ, we see that the magnetic field may
be represented as the curl of another vector field11:
B(r) = ∇ × A(r), (5.27)
namely of the so-called vector-potential, defined as
μ0 j(r′) 3
A(r) ≡
4π
∫V ′ r − r′
d r′ . (5.28)
Please note a beautiful analogy between Eqs. (5.27) and (5.28) and, respectively,
Eqs. (1.33) and (1.38). This analogy implies that the vector-potential A plays, for the
magnetic field, essentially the same role as the scalar potential ϕ plays for the electric
field (hence the name ‘potential’), with a due respect to the vortex character of B.
This notion will be discussed in more detail below.
Now let us see what equations we may get for the spatial derivatives of the
magnetic field. First, vector algebra says that the divergence of any curl is zero12. In
application to Eq. (5.27), this means that
∇ ⋅ B = 0. (5.29)
Comparing this equation with Eq. (1.27), we see that Eq. (5.29) may be interpreted
as the absence of a magnetic analog of an electric charge on which magnetic field
lines could originate or end. Numerous searches for such hypothetical magnetic
11
In the Gaussian units, Eq. (5.27) remains the same, and hence in Eq. (5.28), μ0/4π is replaced with 1/c.
12
See, e.g. Eq. (A.72).
5-9
charges, called magnetic monopoles, using very sensitive and sophisticated exper-
imental set-ups, have not given convincing evidence of their existence in Nature.
Proceeding to the alternative, vector derivative of the magnetic field (i.e. its curl)
and using Eq. (5.28), we obtain
μ0 ⎛ j(r′) 3 ⎞
∇ × B(r) =
4π
∇ × ⎜∇ ×
⎝
∫V ′ r − r′
d r′⎟ .
⎠
(5.30)
This expression may be simplified by using the following general vector identity13:
∇ × (∇ × c) = ∇(∇ ⋅ c) − ∇2 c , (5.31)
applied to vector c(r) = j(rʹ)/∣r − rʹ ∣:
μ0 μ
∇×B=
4π
∇ ∫V ′ j(r′) ⋅ ∇ r −1 r′ d 3r′ − 4π0 ∫V ′ j(r′)∇2 1
r − r′
d 3r′ . (5.32)
As was already discussed during our study of electrostatics in section 3.1,

1
∇2 = −4πδ(r − r′), (5.33)
r − r′
so that the last term of Eq. (5.32) is just μ0j(r). On the other hand, inside the first
integral we can replace ∇ with (−∇ʹ), where prime means differentiation in the space
of radius-vector rʹ. Integrating that term by parts, we get
μ0
∇×B=−
4π
∇ ∮S′ jn (r′) r −1 r′ d 2r′ + ∇∫V ′ ∇r′ ⋅−j(rr′′) d 3r′ + μ0 j(r). (5.34)
Applying this equation to the volume Vʹ limited by a surface Sʹ either sufficiently

distant from the field concentration, or with no current crossing it, we may neglect
the first term on the right-hand side of Eq. (5.34), while the second term always
equals zero in statics, due to the dc charge continuity—see Eq. (4.6). As a result, we
arrive at a very simple differential equation14
∇ × B = μ0 j. (5.35)
This is (the dc form of) the inhomogeneous Maxwell equation, which in

magnetostatics plays a role similar to Eq. (1.27) in electrostatics. Let me display,
for the first time in this course, this fundamental system of equations (at this stage,
for statics only), and give the reader a minute to stare at their beautiful symmetry—
which inspired so much of later physics development:
13
See, e.g. Eq. (A.73).
14
As in all earlier formulas for the magnetic field, in the Gaussian units the coefficient μ0 in this relation is
replaced with 4π/c.
5-10
∇ × E = 0, ∇ × B = μ0 j,
ρ (5.36)
∇⋅E= , ∇ ⋅ B = 0.
ε0
Their only asymmetry, the two zeros on the right-hand sides (for the magnetic field’s
divergence and electric field’s curl), is due to the absence in Nature of magnetic
monopoles and their currents. I will discuss these equations in more detail in section
6.7, after the equations for the field curls have been generalized to their full (time-
dependent) versions.
Returning now to our current, more mundane but important task of calculating
the magnetic field induced by simple current configurations, we can benefit from an
integral form of Eq. (5.35). For that, let us integrate this equation over an arbitrary
surface S limited by a closed contour C, and apply to the result the Stokes theorem15.
The resulting expression,
∮C B ⋅ d r = μ0∮S jn d 2r ≡ μ0I , (5.37)
where I is the net electric current crossing surface S, is called the Ampère law.
As the first example of its application, let us return to the current in a straight wire
(figure 5.4a). With the Ampère law in our arsenal, we can readily pursue an even
more ambitious goal then was achieved in the previous section—calculate the
magnetic field both outside and inside a wire of an arbitrary radius R, with an
arbitrary (albeit axially symmetric) current distribution j(ρ)—see figure 5.5.
Figure 5.5. The simplest application of the Ampère law: the magnetic field of a straight current.
Selecting two contours C in the form of rings of some radius ρ in the plane
perpendicular to the wire axis z, we have B · dr = Bρdφ, where φ is the azimuthal
angle, so that the Ampère law (5.37) yields:
⎧ ρ
⎪
⎪
2π ∫0 j (ρ′ )ρ′dρ′ , for ρ ⩽ R ,
2πρB(ρ) = μ0 × ⎨ R (5.38)
⎪ 2π
⎪
⎩ ∫0 j (ρ′)ρ′ dρ′ ≡ I , for ρ ⩾ R.
15
See, e.g. Eq. (A.78) with f = B.
5-11
Thus we have not only recovered our previous result (5.20), with the notation
replacement d → ρ, in a much simpler way, but could also find the magnetic field
distribution inside the wire. (In the most common case when the wire’s conductivity
σ is constant, and hence the current is uniformly distributed along its cross-section,
j(ρ) = const, the first of Eq. (5.38) immediately yields B ∝ ρ for ρ ⩽ R).
Another important system is a straight, long solenoid (figure 5.6a), with dense
winding: n2A ≫ 1, where n is the number of wire turns per unit length, and A is the
area of the solenoid’s cross-section.
Figure 5.6. Magnetic fields of the (a) straight and (b) toroidal solenoids.
From the symmetry of this problem, the longitudinal (in figure 5.6a, vertical)
component Bz of the magnetic field may only depend on the horizontal position ρ of
the observation point. First taking a plane Ampère contour C1, with both long sides
outside the solenoid, we obtain Bz(ρ2) − Bz(ρ1) = 0, because the total current piercing
the contour equals zero. This is only possible if Bz = 0 at any ρ outside of the
(infinitely long!) solenoid16. With this result on hand, from the Ampère law applied
to the contour C2 we obtain the following relation for the only (z-) component of the
internal field:
Bl = μ0NI, (5.39)
where N is the number of wire turns passing through the contour of length l. This
means that regardless of the exact position on the internal side of the contour, the
result is the same:
N
B = μ0 I = μ0nI. (5.40)
l
Thus, the field inside an infinitely long solenoid (with an arbitrary shape of its cross-
section) is uniform; in this sense, a long solenoid is a magnetic analog of a wide plane
capacitor.
As should be clear from its derivation, the obtained results, in particular that the
field outside the solenoid equals zero, are conditional on the solenoid length being
16
Applying the Ampère law to a circular contour of radius ρ, coaxial with the solenoid, we see that the field
outside (but not inside!) it has an azimuthal component Bφ, similar to that of the straight wire (see Eq. (5.38)
above) and hence (at N ≫ 1) much weaker than the longitudinal field inside the solenoid—see Eq. (5.40).
5-12
very large in comparison to its lateral size. (From Eq. (5.25), we may predict that for
a solenoid of a finite length l, the close-range external field is only a factor of ∼A/l2
lower than the internal one.) A much better suppression of this external (‘fringe’)
field may be obtained using the toroidal solenoid (figure 5.6b). The application of
Ampère law to this geometry shows that, in the limit of dense winding (N ≫ 1), there
is no fringe field at all (for any relation between two radii of the torus), while inside
the solenoid, and distance ρ from the center,
μ0NI
B= . (5.41)
2πρ
We see that a possible drawback of this system for practical applications is that the
internal field does depend on ρ, i.e. is not quite uniform; however, if the torus is thin,
this problem is minor.
How should we solve the problems of magnetostatics for systems whose low
symmetry does not allow obtaining easy results from the Ampère law? (The
examples are of course too numerous to list; for example, we cannot use this
approach even to reproduce Eq. (5.23) for a round current loop.) From the deep
analogy with electrostatics, we may expect that in this case we could recover the field
from the solution of a certain boundary problem for the field’s potential, in this case
the vector-potential A defined by Eq. (5.28). However, despite the similarity of this
formula and Eq. (1.38) for ϕ, which was emphasized above, there are two additional
issues we should tackle in the magnetic case.
First, calculating the vector-potential distribution means determining three scalar
functions (say, Ax, Ay, and Az), rather than one (ϕ). To reveal the second, deeper
issue, let us plug Eq. (5.27) into Eq. (5.35),
∇ × (∇ × A) = μ0 j , (5.42)
and then apply to the left-hand side of this equation the now-familiar identity (5.31).
The result is
∇(∇ ⋅ A) − ∇2 A = μ0 j . (5.43)
On the other hand, as we know from electrostatics (please compare Eqs. (1.38) and
(1.41)), the vector-potential A(r) defined by Eq. (5.28) has to satisfy a simpler
(‘vector-Poisson’) equation
∇2 A = −μ0 j, (5.44)
which is just a set of three usual Poisson equations for each Cartesian component
of A.
In order to resolve this paradox, let us note that Eq. (5.43) is reduced to Eq. (5.44)
if ∇ · A = 0. In this context, let us discuss what discretion we have in the choice of a
potential. In electrostatics, we might add to the scalar function ϕʹ that satisfied
5-13
Eq. (1.33) for the given field E, not only an arbitrary constant, but also an arbitrary
function of time:
−∇[φ′ + f (t )] = −∇φ′ = E. (5.45)
Similarly, using the fact that the curl of the gradient of any scalar function equals
zero17, we may add to any vector function Aʹ that satisfies Eq. (5.27) for the given
field B, not only any constant, but even a gradient of an arbitrary scalar function χ
(r, t), because
∇ × (A′ + ∇χ ) = ∇ × A′ + ∇ × (∇χ ) = ∇ × A′ = B. (5.46)
Such additions, which keep the fields intact, are called gauge transformations18. Let
us see what such a transformation does to ∇ · Aʹ:
∇ ⋅ (A′ + ∇χ ) = ∇ ⋅ A′ + ∇2 χ . (5.47)
For any choice of such a function Aʹ, we can always choose the function χ in such a
way that it satisfies the Poisson equation ∇2χ = −∇ · Aʹ, and hence makes the
divergence of the transformed vector-potential, A = Aʹ + ∇χ, to be equal to zero
everywhere,
∇ ⋅ A = 0, (5.48)
thus reducing Eq. (5.43) to Eq. (5.44).
To summarize, the set of distributions Aʹ(r) that satisfy Eq. (5.27) for a given field
B(r), is not limited to the vector-potential A(r) given by Eq. (5.44), but is reduced to it
upon the additional Coulomb gauge condition (5.48). However, as we will see in a
minute, even this condition still leaves some degrees of freedom in the choice of the
vector-potential. In order to illustrate this fact, and also to obtain a better feeling of the
vector-potential’s distribution in space, let us calculate A(r) for two very basic cases.
First, let us revisit the straight wire problem shown in figure 5.5. As Eq. (5.28)
shows, in this case vector A has just one component (along the axis z). Moreover,
due to the problem’s axial symmetry, its magnitude may only depend on the distance
from the axis: A = nzA(ρ). Hence, the gradient of A is directed across the axis z, so
that Eq. (5.48) is satisfied. For our symmetry (∂/∂φ = ∂/∂z = 0), the Laplace operator,
written in cylindrical coordinates, has just one term19, reducing (5.44) to
1 d ⎛ dA ⎞
⎜ρ ⎟ = −μ0j (ρ). (5.49)
ρ dρ ⎝ dρ ⎠
Multiplying both parts of this equation by ρ and integrating them over the
coordinate once, we obtain
17
See, e.g. Eq. (A.71).
18
The use of the term ‘gauge’ (originally meaning ‘a measure’ or ‘a scale’) in this context is purely historic,
so the reader should not try to find too much hidden sense in it.
19
See, e.g. Eq. (A.61).
5-14
ρ
dA
ρ
dρ
= −μ0 ∫0 j (ρ′)ρ′dρ′ + const. (5.50)
Since in the cylindrical coordinates, for our symmetry20, B = −dA/dρ, Eq. (5.50) is
nothing other than our old result (5.38) for the magnetic field21. However, let us
continue the integration, at least for the region outside the wire, where the function
A(ρ) depends only on the full current I rather than on the current distribution.
Dividing both parts of Eq. (5.50) by ρ, and integrating them over it again, we obtain
μ0I R
A(ρ) = −
2π
ln ρ + const, where I = 2π ∫0 j (ρ)ρdρ , for ρ ⩾ R . (5.51)
As a reminder, we had similar logarithmic behavior for the electrostatic potential

outside a uniformly charged straight line. This is natural, because the Poisson
equations for both cases are similar.
Now let us find the vector-potential for the long solenoid (figure 5.6a), with its
uniform magnetic field. Since Eq. (5.28) prescribes the vector A to follow the
direction of the inducing current, we may start with looking for it in the form
A = nφ A(ρ). (This is particularly natural if the solenoid’s cross-section is circular.)
With this orientation of A, the same general expression for the curl operator in
cylindrical coordinates yields ∇ × A = nz(1/ρ)d(ρA)/dρ. According to Eq. (5.27), this
expression should be equal to B, in our current case equal to nzB, with a constant
B—see Eq. (5.40). Integrating this equality, and selecting such integration constant
so that A(0) is finite, we obtain
Bρ Bρ
A(ρ) = , i.e. A = nφ. (5.52)
2 2
Plugging this result into the general expression for the Laplace operator in the
cylindrical coordinates22, we see that the Poisson equation (5.44) with j = 0 (i.e. the
Laplace equation), is satisfied again—which is natural since for this distribution,
the Coulomb gauge condition (5.48) is satisfied: ∇ · A = 0.
However, Eq. (5.52) is not the unique (or even the simplest) vector-potential that
gives the same uniform field B = nzB. Indeed, using the well-known expression for
the curl operator in Cartesian coordinates23, it is straightforward to check that each
of the vector functions Aʹ = nyBx and Aʺ = −nxBy also has the same curl, and also
satisfies the Coulomb gauge condition (5.48)24. If such solutions do not look very
natural because of their anisotropy in the [x,y] plane, please consider the fact that
they represent the uniform magnetic field regardless of its source—for example,
regardless of the shape of the long solenoid’s cross-section. Such choices of the
20
See, e.g. Eq. (A.63) with ∂/∂φ = ∂/∂z = 0.
21
Since the magnetic field at the wire axis has to be zero (otherwise, being normal to the axis, where would it be
directed?), the integration constant in Eq. (5.50) has to equal zero.
22
See, e.g. Eq. (A.64).
23
See, e.g. Eq. (A.54).
24
The axially symmetric vector-potential (5.52) is just a weighed sum of these two functions: A = (Aʹ + Aʺ)/2.
5-15
vector-potential may be very convenient for some problems, for example for the
quantum-mechanical analysis of the 2D motion of a charged particle in the
perpendicular magnetic field, giving the famous Landau energy levels25.
5.3 Magnetic energy, flux, and inductance

Considering the currents flowing in a system as generalized coordinates, the
magnetic forces (5.1) between them are their unique functions, and in this sense
the energy U of their magnetic interaction may be considered a potential energy of
the system. The apparent (but somewhat deceptive) way to derive an expression
for this energy is to use the analogy between Eq. (5.1) and its electrostatic analog,
Eq. (5.2). Indeed, Eq. (5.2) may be transformed into Eq. (5.1) with just three
replacements:
(i) ρ(r)ρʹ(rʹ) should be replaced with [j(r) · jʹ(rʹ)],
(ii) ε0 should be replaced with 1/μ0, and
(iii) the sign before the double integral has to be replaced with the opposite one.
Hence we may avoid repeating the calculation made in chapter 1, by making

these replacements in Eq. (1.59), which gives the electrostatic potential energy of
the system with ρ(r) and ρʹ(rʹ) describing the same charge distribution, i.e. with ρʹ
(r) = ρ(r), to obtain the following expression for the magnetic potential energy in the
system with, similarly, jʹ (r) = j(r)26:
μ0 1
Uj = −
4π 2
∫ d 3r∫ d 3r′ j(rr) −⋅ jr(′r′) . (5.53)
However, this is not the unique answer, and not even the most convenient answer.
Actually, Eq. (5.53) describes the energy that is adequate (i.e. whose minimum
corresponds to the stable equilibrium of the system) only in the case when the
interacting currents are fixed—just as Eq. (1.59) is adequate when the interacting
charges are fixed. Here comes a substantial difference between electrostatics and
magnetostatics: due to the fundamental fact of charge conservation (already
discussed in sections 1.1 and 4.1), keeping electric charges fixed does not require
external work, while the maintenance of currents generally does. As a result,
Eq. (5.53) describes the energy of the magnetic interaction plus of the system
keeping the currents constant—or rather of its part depending on the system under
our consideration27.
Now in order to exclude from Uj the contribution due to the interaction with the
current-supporting system(s), i.e. calculate the potential energy U of our system as
25
See, e.g. Part QM section 3.2.
26
As was repeatedly discussed above, for the interaction of two independent charge distributions ρ(r) and ρʹ(rʹ),
the front factor ½ has to be dropped; the same is true for the interaction of two independent current
distributions j(r) and jʹ(rʹ).
27
In the terminology already used in section 3.5 (see also a general discussion in Part CM section 1.4.), Uj is
essentially the Gibbs potential energy of our magnetic system.
5-16
such, we need to know this contribution. The simplest way to do this is to use the
Faraday induction law, which describes this interaction. This is why let me postpone
the derivation until the beginning of the next chapter, and for now ask the reader to
believe me that its account leads to an addition to Uj of a term of twice larger
magnitude, so that the result is given by an expression similar to equation (5.53), but
with the opposite sign:
μ0 1
U=
4π 2
∫ d 3r∫ d 3r′ j(rr) −⋅ jr(′r′) , (5.54)
I promise to prove this fact in section 6.2 below.

Due to the importance of Eq. (5.54), let us rewrite it in several other forms,
convenient for different applications. First of all, just as in electrostatics, it may be
recast into a potential-based form. Indeed, using the definition (5.28) of the vector-
potential A(r), Eq. (5.54) becomes28
1
U=
2
∫ j(r) ⋅ A(r)d 3r. (5.55)
This formula, which is a clear magnetic analog of Eq. (1.60) of electrostatics, is very
popular among field theorists, because it is very handy for their manipulations.
However, for many calculations it is more convenient to have a direct expression of
energy via the magnetic field. Again, this may be done very similarly to what was
done for electrostatics in section 1.3, i.e. by plugging, into Eq. (5.55), the current
density expressed from Eq. (5.35), and then transforming it as29
1 1
U=
2
∫j ⋅ Ad 3r =
2μ 0
∫
A ⋅ (∇ × Β) d 3r
. (5.56)
1 1
=
2μ 0
∫B ⋅ (∇ × Α)d 3r −
2μ 0
∫
∇ ⋅ (A × B)d 3r
Now using the divergence theorem, the second integral may be transformed into a
surface integral of (A × B)n. According to Eqs. (5.27) and (5.28) if the current
distribution j(r) is localized, this vector product drops, at large distances, faster than
1/r2, so that if the integration volume is large enough, the surface integral is
negligible. In the remaining first integral in Eq. (5.56) we may use Eq. (5.27) to recast
∇ × A into the magnetic field. As a result, we obtain a very simple and fundamental
formula.
1
U=
2μ 0
∫ B 2d 3r. (5.57a )
28
This relation remains the same in the Gaussian units, because in those units, both Eqs. (5.28) and (5.54)
should be stripped of their μ0/4π coefficients.
29
For that, we may use Eq. (A.77) with f = A and g = B, giving A · (∇ × B) = B · (∇ × A) − ∇ · (A × B).
5-17
Just as with the electric field, this expression may be interpreted as a volume integral
of the magnetic energy density u:
1 2
2μ 0
B (r), (5.57b)
clearly similar to Eq. (1.65)30. Again, the conceptual choice between the spatial
localization of magnetic energy—either at the location of electric currents only, as
implied by Eqs. (5.54) and (5.55), or in all regions where the magnetic field exists, as
apparent from Eq. (5.57b)—cannot be done within the framework of magneto-
statics, and only the electrodynamics gives the decisive preference for the latter
choice.
For the practically important case of currents flowing in several thin wires, Eq.
(5.54) may be first integrated over the cross-section of each wire, just as was done at
the derivation of Eq. (5.4). Again, since the integral of the current density over the
kth wire’s cross-section is just the current Ik in the wire, and cannot change along its
length, it may be taken from the remaining integrals, giving
μ0 1 d rk ⋅ d rk ′
U= ∑IkIk
4π 2 k , k ′ ′
∮l ∮l
k k′ rk − rk ′
, (5.58)
where l is the full length of the wire loop. Note that Eq. (5.58) is valid if the currents
Ik are independent of each other, because the double sum counts each current pair
twice, compensating the coefficient ½ in front of the sum. It is useful to decompose
this relation as
1
U= ∑IkIk′Lkk′,
2 k, k ′
(5.59)
where the coefficients Lkkʹ are independent of the currents:

μ0 d rk ⋅ d r k′
Lkk′ ≡
4π
∮l ∮l
k k′ rk − r k′
, (5.60)
The coefficient Lkkʹ with k ≠ kʹ, is called the mutual inductance between current the
loops with numbers k and kʹ, while the diagonal coefficient Lk ≡ Lkk is called the self-
inductance (or just inductance) of the kth loop31. From the symmetry of Eq. (5.60)
30
The transfer to the Gaussian units in Eqs. (5.77) and (5.78) may be accomplished by the usual replacement
μ0 → 4π, thus giving, in particular, u = B2/8π.
31
As is evident from Eq. (5.60), these coefficients depend only on the geometry of the system. Moreover, in the
Gaussian units, in which Eq. (5.60) is valid without the factor μ0/4π, the inductance coefficients have the
dimension of length (centimeters). The SI unit of inductance is called the henry, abbreviated as H, after
J Henry, 1797–1878, who, in particular, discovered the effect of electromagnetic induction (see section 6.1)
independently of M Faraday.
5-18
with respect to the index swap, k ↔ kʹ, it is evident that the matrix of coefficients Lkkʹ
is symmetric32:
L kk′ = L k′k , (5.61)
so that for the practically important case of two interacting currents I1 and I2, Eq.
(5.59) reads
1 1
U= L1I12 + MI1I2 + L 2I22, (5.62)
2 2
where M ≡ L12 = L21 is the mutual inductance coefficient.
These formulas clearly show the importance of the self- and mutual inductances,
so I will demonstrate their calculation for at least a few basic geometries. Before
doing that, however, let me recast Eq. (5.58) into one more form that may facilitate
such calculations. Namely, let us notice that for the magnetic field induced by
current Ik in a thin wire, Eq. (5.28) is reduced to
μ0 d rk
A k(r) =
4π
Ik ∫l′ r − rk
, (5.63)
so that Eq. (5.58) may be rewritten as

1
U= ∑Ik
2 k, k ′
∮l
k
A k′(rk ) ⋅ d r k′. (5.64)
But according to the same Stokes theorem that was used earlier in this chapter to
derive the Ampère law and Eq. (5.27), the integral in this expression is nothing more
than the magnetic field’s flux (more frequently called just the magnetic flux) through
a surface S limited by the contour l33:
∮l A(r) ⋅ d r = ∫S (∇ × A)nd 2r = ∫S Bnd 2r ≡ Φ. (5.65)
As a result, Eq. (5.64) may be rewritten as

1
U= ∑Ik Φ kk′,
2 k, k ′
(5.66)
where Φkkʹ is the flux of the field induced by the kʹ th current through the loop of the
kth current. Comparing this expression with Eq. (5.59), we see that
32
Note that the matrix of the mutual inductances Ljjʹ is very similar to the matrix of reciprocal capacitance
coefficients pkkʹ —for example, compare Eq. (5.62) with Eq. (2.21).
33
The SI unit of magnetic flux is called the weber, abbreviated as Wb, after W Weber, who, in particular,
co-invented (with C Gauss) the electromagnetic telegraph, and in 1856 was the first, together with R
Kohlrausch, to notice that the value of (in modern terms) 1/(ε0μ0)1/2, derived from electrostatic and
magnetostatic measurements, coincides with the independently measured speed of light c, giving an important
motivation for Maxwell’s theory.
5-19
Φ kk′ ≡ ∫S k
(B k′)nd 2r = L kk′Ik′, (5.67)
This expression not only gives us one more means for calculating coefficients Lkkʹ,
but also shows their physical sense: the mutual inductance characterizes what part of
the magnetic field (colloquially, ‘how many field lines’), induced by the current Ikʹ,
pierces the kth loop—see figure 5.7.
Figure 5.7. A schematic representation of the physical sense of the mutual inductance coefficient Lkkʹ ≡ Φkkʹ/Ikʹ.
Due to the linear superposition principle, the total flux piercing kth loop may be
represented as
Φk ≡ ∑Φ kk′ = ∑L kk′Ik′ (5.68)

k′ k′
For example, for the system of two currents this expression is reduced to a clear
analog of equations Eqs. (2.19):
Φ1 = L1I1 + MI2,
(5.69)
Φ2 = MI1 + L 2I2.
For the even simpler case of a single current,
Φ = LI, (5.70)
so that the magnetic energy of the current may be represented in several equivalent
forms:
L 2 1 1 2
U= I = IΦ = Φ . (5.71)
2 2 2L
These relations, similar to Eqs. (2.14) and (2.15) of electrostatics, show that the self-
inductance L of a current loop may be considered as a measure of the system’s
magnetic energy, but, as we will see in section 6.1, this measure is adequate only if
the flux Φ, rather than the current I, is fixed.
5-20
Now we are well equipped for the calculation of inductance coefficients, having
three options. The first is to use Eq. (5.60) directly34. The second is to calculate the
magnetic field energy from Eq. (5.57) as the function of all currents Ik in the system,
and then use Eq. (5.59) to find all coefficients Lkkʹ. For example, for a system with
just one current, Eq. (5.71) yields
U
L= . (5.72)
I 2 /2
Finally, if the system consists of thin wires, so that the loop areas Sk and hence fluxes
Φkkʹ are well defined, we may calculate them from Eq. (5.65) and then use Eq. (5.67)
to find the inductances.
Actually, the first two options may have technical advantages over the third one
even for some system of thin wires, in which the notion of magnetic flux is not quite
apparent. As an important example, let us find the self-inductance of a long solenoid—
see figure 5.6a again. We have already calculated the magnetic field inside it—see
Eq. (5.40)—so that, due to the field uniformity, the magnetic flux piercing each wire
turn is just
Φ1 = BA = μ0nIA, (5.73)
where A is the area of solenoid’s cross-section—for example πR2 for a round
solenoid, although Eq. (5.40), and hence Eq. (5.73), are valid for any cross-section.
Comparing Eq. (5.73) with Eq. (5.70), one might wrongly conclude that L = Φ1/I =
μ0nA (WRONG!), i.e. that the solenoid’s inductance is independent of its length.
Actually, the magnetic flux Φ1 pierces each wire turn, so that the total flux through
the whole current loop, consisting of N turns, is
Φ = N Φ1 = μ0n 2lAI, (5.74)
and the correct expression for the long solenoid’s self-inductance is

Φ
L= = μ0n 2lA, (5.75)
I
i.e. the inductance per unit length is constant: L/l = μ0n2A.
Since this reasoning may seem not quite apparent, it is prudent to verify it by
using Eq. (5.72), with the full magnetic energy inside the solenoid (neglecting minor
fringe and external field contributions) given by Eq. (5.57) with B = const within the
internal volume V = lA, and zero outside it:
1 2 1 2 I2
U=
2μ 0
B Al =
2μ 0
( μ0nI ) Al = μ0n 2lA .
2
(5.76)
Plugging this relation into Eq. (5.72) immediately confirms the earlier result (5.75).
34
Numerous applications of this Neumann formula to electrical engineering problems may be found, for
example, in the classical text [1].
5-21
This approach becomes virtually inevitable for continuously distributed currents.

As an example, let us calculate the self-inductance L of a long coaxial cable with the
cross-section shown in figure 5.835, with the full current in the outer conductor equal
and opposite to that (I) in the inner conductor.
Figure 5.8. The cross-section of a coaxial cable.
Let us assume that the current is uniformly distributed over the cross-sections of
both conductors. (As we know from the previous chapter, this is indeed the case if
both the internal and external conductors are made of a uniform resistive material.)
First, we should calculate the radial distribution of the magnetic field (that of course
has only one, azimuthal component, because of the axial symmetry of the problem).
This distribution may be immediately found from the application of the Ampère law
to the circular contours of radii ρ within four different ranges:
⎧ ρ2 a 2 , for ρ < a ,
⎪
⎪1, for a < ρ < b ,
2πρB = μ0I ∣piercing the contour = μ0I × ⎨ 2 (5.77)
⎪(c − ρ ) (c − b ), for b < ρ < c ,
2 2 2
⎪
⎩ 0, for c < ρ .
Now, an elementary integration yields the magnetic energy per unit length of the
cable:
∞
U 1
=
l 2μ 0
∫ B 2d 2r = μπ ∫0 B 2ρdρ
0
2⎡ b ⎛ ⎞2 c⎛ ⎤
c 2 − ρ2 ⎞
2
μ0I ⎛ρ⎞ a 2
1
= ⎢ ⎜
∫ ⎟ ρdρ + ⎜ ⎟ ρ∫
d ρ + ⎜ ∫ ⎟ ρd ρ⎥ (5.78)
4π ⎢⎣ 0 ⎝ a 2 ⎠ a ⎝ ρ⎠ b ⎝ ρ(c 2 − b 2 ) ⎠ ⎥⎦
μ ⎡ b c2 ⎛ c2 c 1 ⎞⎤ I 2
= 0 ⎢ln + 2 ⎜ ln − ⎟⎥ .
2π ⎣ a c − b2 ⎝ c 2 − b2 b 2 ⎠⎦ 2
From here, and Eq. (5.72), we obtain the final answer:
L μ ⎡ b c2 ⎛ c2 c 1 ⎞⎤
= 0 ⎢ln + 2 ⎜ ln − ⎟⎥ . (5.79)
l 2π ⎣ a c − b2 ⎝ c 2 − b2 b 2 ⎠⎦
35
As a reminder, the mutual capacitance C between the conductors of such a system was calculated in section 2.3.
5-22
Note that for the particular case of a thin outer conductor, c − b ≪ b, this
expression reduces to
L μ ⎛ b 1⎞
≈ 0 ⎜ln + ⎟ , (5.80)
l 2π ⎝ a 4⎠
where the first term in the parentheses may be traced back to the contribution of the
magnetic field energy in the free space between the conductors. This distinction is
important for some applications, because in superconductor cables, as well as the
normal-metal cables at high frequencies (to be discussed in the next chapter), the field
does not penetrate the conductor’s bulk, so that Eq. (5.80) is valid without the last
term, 1/4, in the parentheses, which is due to the magnetic field energy inside the wire.
As the last example, let us calculate the mutual inductance between a long straight
wire and a round wire loop adjacent to it (figure 5.9), neglecting the thickness of both
wires. Here there is no problem with using the last of the approaches discussed
above, based on the direct magnetic flux calculation. Indeed, as was discussed in
section 5.1, the field B1 induced by the current I1 at any point of the round loop is
normal to its plane—e.g. to the plane of drawing of figure 5.9. In the Cartesian
coordinates shown in that figure, Eq. (5.20) reads B1 = μ0I1/2πy, giving the following
magnetic flux through the loop:
μ0I1 +R R +(R 2−x 2 )1/2
1
Φ21 =
2π
∫−R dx ∫R−(R −x )2 2 1/2
dy
y
R
μ I1 R + (R2 − x 2 )1/2
= 0
π
∫0 ln
R − (R2 − x 2 )1/2
dx (5.81)
μ0I1R 1
1 + (1 − ξ 2 )1/2
=
π
∫0 ln
1 + (1 − ξ 2 )1/2
dξ.
This is a table integral equal to π,36 so that Φ21 = μ0I1R, and the final answer for the
mutual inductance M ≡ L12 = L21 = Φ21/I1 is finite (and very simple):
M = μ0R, (5.82)
despite the magnetic field’s divergence at the lowest point of the loop (y = 0).
Figure 5.9. An example of the mutual inductance calculation.
36
See, e.g. Eq. (A.40), with a = 1.
5-23
Note that in contrast with the finite mutual inductance of this system, the self-
inductances of both wires are formally infinite in the thin-wire limit—see, e.g.
Eq. (5.80), which in the limit b/a ≫ 1 describes a thin straight wire. However, since
this divergence is very weak (logarithmic), it is quenched by any deviation from this
perfectly axial geometry. For example, a good estimate of the inductance of a wire of a
large but finite length l may be obtained from Eq. (5.80) via the replacement of b with l:
μ0 l
L∼ l ln . (5.83)
2π a
(Note, however, that the exact result depends on where from/to the current flows
beyond that segment.) A close estimate, with l replaced with 2πR, and b replaced
with R, is valid for the self-inductance of the round loop. A more exact calculation of
this inductance, which would be asymptotically correct in the limit a ≪ R, is a very
useful exercise, highly recommended to the reader37.
5.4 Magnetic dipole moment, and magnetic dipole media

The most natural way in which the magnetic media description parallels that described in
chapter 3 for dielectric media, is based on properties of magnetic dipoles—the notion
close (but not identical!) to that of the electric dipoles discussed in section 3.1. To
introduce this notion quantitatively, let us consider, just as in section 3.1, a spatially
localized system with a current distribution j(rʹ), whose magnetic field is measured at
relatively large distances r ≫ rʹ (figure 5.10).
Figure 5.10. Calculating the magnetic field of localized currents, as observed from a distant point (r ≫ a).
Applying the truncated Taylor expansion (3.5) of the fraction 1/∣r − rʹ∣ to the
vector potential given by Eq. (5.28), we obtain
μ0 ⎡ 1 ⎤
A(r) ≈ ⎢
4π ⎣ r
∫V j(r′)d 3r′ + r13 ∫V (r ⋅ r′)j(r′)d 3r′⎥⎦. (5.84)
Now, due to the vector character of this potential, we have to depart slightly from
the approach of section 3.1 and use the following vector algebra identity38:
∫V [f (j ⋅ ∇g ) + g(j ⋅ ∇f )] d 3r = 0, (5.85)
37
Its solution may be found, for example, in section 34 of [2].
38
See, e.g. Eq. (A.81) with the additional condition jn∣S = 0, pertinent for space-restricted currents.
5-24
which is valid for any pair of smooth (differentiable) scalar functions f(r) and g(r),
and any vector function j(r) that, as the dc current density, satisfies the continuity
condition ∇ · j = 0 and whose normal component vanishes on the surface of the
volume V. First, let us use Eq. (5.85) with f = 1 and g equal to any Cartesian
component of the radius-vector r: g = rl (l = 1, 2, 3). Then it yields
∫V (j ⋅ nl )d 3r = ∫V jl d 3r=0, (5.86)
so that for the vector as a whole

∫V j(r)d 3r = 0, (5.87)
showing that the first term on the right-hand side of Eq. (5.84) equals zero. Next, let
us use Eq. (5.85) again, now with f = rl, g = rlʹ (l, lʹ = 1, 2, 3); then it yields
∫V (rl jl′ + rl′jl ) d 3r = 0, (5.88)
so that the lth Cartesian component of the second integral in Eq. (5.84) may be
transformed as
3 3
∫V (r ⋅ r′)jl d 3r′ = ∫V ∑rl′r′l′jl d 3r′ = 12 ∑rl′∫V (r′l′jl + r′l ′jl )d 3r′
l ′= 1 l ′= 1
3
1
= ∑ rl ′
2 l ′= 1
∫V (r′l′jl − r′l ′jl ′ )d 3r′ (5.89)
1⎡ ⎤
= − ⎢r ×
2 ⎣ V
∫
(r′ × j)d 3r′⎥ .
⎦l
As a result, Eq. (5.84) may be rewritten as
μ0 m × r
A(r) = , (5.90)
4π r 3
where the vector m, defined as39
1
m≡
2
∫V r × j(r) d 3r, (5.91)
is called the magnetic dipole moment of our system—that itself, within the long-range
approximation (5.90), is called the magnetic dipole.
Note a close analogy between the m defined by Eq. (5.91) and the orbital40
angular momentum of a non-relativistic particle with mass mk:
39
In the Gaussian units, the definition (5.91) is kept valid, so that Eq. (5.90) is stripped of the factor μ0/4π.
40
This adjective is used, particularly in quantum mechanics, to distinguish the motion of a particle as a whole
(not necessarily along a closed orbit) from its intrinsic angular momentum, the spin—see, e.g. Part QM
chapter 4.
5-25
Lk ≡ rk × pk = rk × mk vk , (5.92)
where pk = mkvk is its mechanical momentum. Indeed, for a continuum of such
particles with the same electric charge q, with the spatial density n, j = qnv, and
Eq. (5.91) yields
m= ∫V 12 r × j d 3r = ∫V nq2 r × v d 3r, (5.93)
while the total angular momentum of such a system of particles of the same mass
(mk = m0) is
L= ∫V nm0r × vd 3r, (5.94)
so that we obtain a very straightforward relation

q
m= L. (5.95)
2m 0
For the orbital motion, this classical relation survives in quantum mechanics for
linear operators, and hence for the eigenvalues of the observables. Since the orbital
angular momentum is quantized in units of the Plank’s constant ℏ, for an electron
the orbital magnetic moment is always a multiple of the so-called Bohr magneton
eℏ
μB ≡ , (5.96)
2m e
where me is the free electron mass41. However, for particles with spin, such a
universal relation between the vectors m and L is no longer valid. For example, the
electron’s spin s = ½ gives a contribution of ℏ/2 to the mechanical angular
momentum, but its contribution to the magnetic moment it still very close to μB.
The next important example of a magnetic dipole is a planar thin-wire loop,
limiting area A (of an arbitrary shape), and carrying current I, for which m has a
surprisingly simple form,
m = I A, (5.97)
where the modulus of vector A equals loop’s area A, and its direction is normal to
the loop’s plane. This formula may be readily proved by noticing that if we select the
coordinate origin on the plane of the loop (figure 5.11), then the elementary
component of the magnitude of the integral (5.91),
1 1
m=
2
∮C r × Id r =I ∮C 2
r × dr = I ∮C 12 r 2dφ, (5.98)
is just the elementary area dA = (1/2)rdh = (1/2)rd(rsinφ) = r2dφ/2.
41
In SI units, me ≈ 0.91 × 10−30 kg, so that μB ≈ 0.93 × 10−23 J/T.
5-26
Figure 5.11. Calculating the magnetic dipole moment of a planar current loop.
The combination of Eqs. (5.96) and (5.97) allows a useful estimate of the scale of
atomic currents, by finding what current I should flow in a circular loop of atomic
size scale (the Bohr radius) rB ≈ 0.5 × 10−10 m, i.e. of area A ≈ 10−20 m2, to produce a
magnetic moment equal to μB.42 The result is surprisingly macroscopic: I ∼ 1 mA
(quite comparable to the currents driving your earbuds). Although this estimate
should not be taken too literally, due to the quantum-mechanical spread of the
electron’s wavefunctions, it is very useful for obtaining a feeling of how significant
atomic magnetism is and hence why ferromagnets may provide such a strong field.
After these illustrations, let us return to equation (5.90). Plugging it into the
general formula (5.27), we may calculate the magnetic field of a magnetic dipole43:
μ0 3r(r ⋅ m) − mr 2
B(r) = . (5.99)
4π r5
The structure of this formula exactly replicates that of Eq. (3.13) for the electric
dipole field (including the sign). Because of this similarity, the energy of a dipole of a
fixed magnitude m in an external field, and hence the torque and force exerted on it
42
Another way to arrive at the same estimate is to take I ∼ ef = eω/2π with ω ∼ 1016 s−1 being the typical
frequency of radiation due to atomic interlevel quantum transitions.
43
Similarly to the situation with the electric dipoles (see Eq. (3.24) and its discussion), it may be shown that the
magnetic field of any closed current loop (or any system of such loops) satisfies the following equality:
∫v B(r)d 3r = (2/3)μ0m,
where the integral is over any sphere confining all the currents. On the other hand, as we know from section
3.1, for a field with the structure (5.99), derived from the long-range approximation (5.90), such integral
vanishes. As a result, in order to obtain a course-grain description of the magnetic field of a small system,
located at r = 0, which would give the correct average value of the magnetic field, Eq. (5.99) should be modified
as follows:
μ 0 ⎛ 3r(r ⋅ m) − mr 2 8π ⎞
B cg(r) = ⎜ 5
+ mδ(r)⎟,
4π ⎝ r 3 ⎠
in a conceptual (but not quantitative) similarity to Eq. (3.25).
5-27
by a fixed external field, are also absolutely similar to the expressions for an electric
dipole—see Eqs. (3.15)–(3.19)44:
U = −m ⋅ Bext , (5.100)
and, as a result,
τ = m × Bext , (5.101)
F = ∇(m ⋅ Bext). (5.102)

Now let us consider a system of many magnetic dipoles (e.g. atoms or molecules),
distributed in space with a macroscopic (i.e. average) density n. Then we can use
Eq. (5.90) (generalized in the evident way for an arbitrary position, rʹ, of a dipole), and
the linear superposition principle, to calculate the macroscopic vector-potential A:
μ0
A(r) =
4π
∫ M(r′r) ×− (rr′ 3− r′) d 3r′, (5.103)
where M ≡ nm is the magnetization, i.e. the average magnetic moment per unit
volume. Transforming this integral absolutely similarly to how Eq. (3.27) had been
transformed into Eq. (3.29), we obtain:
μ0
A(r) =
4π
∫ ∇′ r×−Mr′(r′) d 3r′. (5.104)
Comparing this result with Eq. (5.28), we see that ∇ × M is equivalent, in its
effect, to the density jef of a certain effective ‘magnetization current’. Just as the
electric-polarization charge ρef discussed in section 3.2 (see figure 3.4), the vector
jef = ∇ × M may be interpreted as the uncompensated part of the loop currents
representing single magnetic dipoles m (figure 5.12). Note, however, that since the
atomic dipoles may be due to the particles’ spins, rather than the actual electric
currents due to the orbital motion, the magnetization current’s nature is not as direct
than that of the polarization charge.
Figure 5.12. A cartoon illustrating the physical nature of the effective magnetization current jef = ∇ × M.
44
Note that the fixation of m and Bext effectively means that the currents producing them are fixed—please
have one more look at Eq. (5.35) and Eq. (5.97). As a result, Eq. (5.100) is a particular case of Eq. (5.53) rather
than Eq. (5.54)—hence the minus sign.
5-28
Now, using Eq. (5.28) to add the possible contribution from ‘stand-alone’
currents j, not included into the currents of microscopic magnetic dipoles, we obtain
the general equation for the vector-potential of the macroscopic field:
μ0
A(r) =
4π
∫ [j(r′) +r∇−′ ×r′ M(r′)] d 3r′. (5.105)
Repeating the calculations that have led us from Eq. (5.28) to the Maxwell equation
(5.35), with the account of the magnetization current term, for the macroscopic
magnetic field B we obtain
∇ × B = μ0(j + ∇ × M). (5.106)
Following the same philosophy as in section 3.2, we may recast this equation as
∇ × H = j, (5.107)
where the field H, defined as
B
H≡ − M, (5.108)
μ0
for historic reasons (and very unfortunately) is also called the magnetic field45. It is
crucial to remember that the physical sense of field H is very much different from
field B.
In order to understand the difference better, let us use Eq. (5.107) to bring
Eqs. (3.32), (3.36), (5.29), and (5.107) together, writing them as the system of the
macroscopic Maxwell equations (again, so far for the stationary case ∂/∂t = 0)46:
∇ × E = 0, ∇ × H = j,
(5.109)
∇ ⋅ D = ρ, ∇ ⋅ B = 0.
They clearly show that the roles of vector fields D and H are very similar: they both may
be called the ‘would-be fields’—which would be induced by the charges ρ and currents j, if
the medium had not modified them by its dielectric and/or magnetic polarization.
Despite this similarity, let me note an important difference of signs in the relation
(3.33) between E, D, and P, on one hand, and the relation (5.108) between B, H, and
M, on the other hand. This is not just a matter of definition. Indeed, due to the
similarity of equations (3.15) and (5.100), including similar signs, the electric and
magnetic fields both try to orient the corresponding dipole moments along the field.
45
This confusion is exacerbated by the fact that in Gaussian units, Eq. (5.108) has the form H = B − 4πM, and
hence the fields B and H have one dimensionality (and are equal in free space!)—although the unit of H has a
different name (oersted, abbreviated as Oe). Mercifully, in the SI units, the dimensionality of B and H is
different, with the unit of H being called the ampere per meter.
46
Let me remind the reader once again that in contrast with the system (5.36) of the Maxwell equations for the
genuine (microscopic) fields, the right-hand sides of Eq. (5.109) represent only the stand-alone charges and
currents, not included in the microscopic electric and magnetic dipoles.
5-29
Hence, in the media that allow such an orientation (and as we will see momentarily,
for magnetic media this is not always the case), the induced polarizations P and M
are directed along, respectively, the vectors E and B of the genuine (although
macroscopic) fields. According to Eq. (3.33), if the would-be field D is fixed—say, by
a fixed stand-alone charge distribution ρ(r)—such a polarization reduces the electric
field E = (D − P)/ε0. On the other hand, Eq. (5.108) shows that in a magnetic
medium with a fixed would-be field H, the magnetic polarization with M parallel to
B enhances the magnetic field B = (H + M)/μ0. This difference may be traced back to
the sign difference in the basic relations (5.1) and (5.2), i.e. to the fundamental fact
that the electric charges of the same sign repulse, while the currents of the same
direction attract each other.
5.5 Magnetic materials

In order to form a complete system of differential equations, the macroscopic
Maxwell equations (5.109) have to be complemented with the constitutive relations
describing the medium: D ↔ E, j ↔ E, and B ↔ H. In the previous two chapters we
already discussed, in brief, the first two of them; let us proceed to the last one.
A major difference between the dielectric and magnetic constitutive relations
D(E) and B(H) is that while a dielectric medium always reduces the external field,
magnetic media may either reduce or enhance it. In order to quantify this fact, let us
consider the most widespread materials—linear magnetics in which M (and hence H)
are proportional to B. For isotropic materials, this proportionality is characterized
by a scalar—either the magnetic permeability μ, defined by the following relation:
B ≡ μH, (5.110)
47
or the magnetic susceptibility defined as
M = χm H. (5.111)
Plugging these relations into Eq. (5.108), we see that these two parameters are not
independent, but are related as
μ = (1 + χm )μ0 . (5.112)
Note that despite the superficial similarity between Eqs. (5.110)–(5.112) and
relations (3.43)–(3.47) for linear dielectrics,
47
According to Eqs. (5.110) and (5.112) (i.e. in SI units), χm is dimensionless, while μ has the same the same
dimensionality as μ0. In the Gaussian units, μ is dimensionless: (μ)Gaussian = (μ)SI/μ0, and χm is also introduced
differently, as μ = 1 + 4πχm, Hence, just as for the electric susceptibilities, these dimensionless coefficients are
different in the two systems: (χm)SI = 4π(χm)Gaussian. Note also that χm is formally called the volumic magnetic
susceptibility, in order to distinguish it from the atomic (or ‘molecular’) susceptibility χ defined by a similar
relation, 〈m〉 ≡ χH, where m is the induced magnetic moment of a single dipole—e.g. an atom. Evidently, in a
dilute medium, i.e. in the absence of a substantial dipole–dipole interaction, χm = nχ, where n is the dipole
density. (Note that χ is an analog of the electric atomic polarizability α—see Eq. (3.48) and its discussion.)
5-30
D = εE, P = χe ε0E, ε = (1 + χe )ε0, (5.113)

there is an important conceptual difference between them. Namely, while the vector
E on the right-hand sides of Eqs. (5.113) is the actual (although macroscopic) electric
field, the vector H on the right-hand side of Eqs. (5.110) and (5.111) represents a
‘would-be’ magnetic field, in all aspects similar to D rather than E—see, for example,
Eq. (5.109).
This historic difference in the traditional way to write the constitutive relations for
the electric and magnetic fields is not without its physical reasons. Most key
experiments with electric and magnetic materials have been performed by placing
their samples into nearly uniform electric and magnetic fields, and the simplest
systems for their implementation are, respectively, the plane capacitor (figure 2.3)
and the solenoid (figure 5.6). The field in the former system may be most
conveniently controlled by measuring the voltage V between its plates, which is
proportional to the electric field E. On the other hand, the field provided by the
solenoid may be controlled by the current I in it. According to Eq. (5.107), the field
proportional to this stand-alone current is H, rather than B.48
Table 5.1 lists the values of magnetic susceptibility for several materials. It shows
that in contrast to linear dielectrics whose susceptibility χe is always positive, i.e. the
dielectric constant κ = χe + 1 is always larger than 1 (see table 3.1), linear magnetics
may be either paramagnets (with χm > 0, i.e. μ > μ0) or diamagnets (with χm < 0,
μ < μ0).
Table 5.1. Magnetic susceptibility (χm)SI of a few representative (and/or important) materials.a
‘Mu-metal’ (75% Ni + 15% Fe + a few %% of Cu and Mo) ∼20 000b

Permalloy (80% Ni + 20% Fe) ∼8000b
‘Electrical’ (or ‘transformer’) steel (Fe + a few %% of Si) ∼4000b
Nickel ∼100
Aluminum +2 × 10−5
Oxygen (at ambient conditions) +0.2 × 10−5
Water −9 × 10−6
Diamond −2 × 10−5
Copper −7 × 10−5
Bismuth (the strongest non-superconducting diamagnet) −1.7 × 10−4
a
The table does not include bulk superconductors, which may be described, in a crude (‘coarse-grain’)
approximation, as perfect diamagnets (with B = 0, i.e. χm = −1 and μ = 0), although the actual physics of this
phenomenon is more intricate—see section 6.3.
b
The exact values of χm ≫ 1 for soft ferromagnetic materials (see, e.g., the upper three rows of the table)
depend not only on their exact composition, but also on their thermal processing (“annealing”). Moreover, due
to unintentional vibrations, the extremely high χm of such materials may somewhat decay with time, though
may be restored to the original value by new annealing. The reason for that behavior is discussed below.
48
This fact also explains the misleading term ‘magnetic field’ for H.
5-31
The reason for this difference is that in dielectrics, two different polarization
mechanisms (schematically illustrated by figure 3.7) lead to the same sign of the
average polarization—see the discussion in section 3.3. One of these mechanisms,
illustrated by figure 3.7b, i.e. the ordering of spontaneous dipoles by the applied field,
is also possible for magnetization—for the atoms and molecules with spontaneous
internal magnetic dipoles of magnitude m0 ∼ μB, due to their net spins. Again, in the
absence of an external magnetic field the spins, and hence the dipole moments m0 may
be disordered, but according to Eq. (5.100), the external magnetic field tends to align
the dipoles along its direction. As a result, the average direction of the spontaneous
elementary moments m0, and hence the direction of the arising magnetization M, is the
same as that of the microscopic field B at the points of the dipole location (i.e. for a
diluted media, of H ≈ B/μ0), resulting in a positive susceptibility χm, i.e. in the
paramagnetism—such as that of oxygen and aluminum (see table 5.1).
However, in contrast to the electric polarization of atoms/molecules with no
spontaneous electric dipoles, which gives the same sign of χe ≡ κ − 1 (see figure 3.7a
and its discussion), the magnetic materials with no spontaneous atomic magnetic
dipole moments have χm < 0—the effect called orbital (or ‘Larmor’49) diamagnetism.
As the simplest model of this effect, let us consider the orbital motion of an atomic
electron about an atomic nucleus as that of a classical particle of mass m0, with an
electric charge q about an immobile attracting center. As classical mechanics tells
us, the central attractive force does not change the particle’s angular momentum
L ≡ m0r × v, but the applied magnetic field B (that may be taken as uniform on the
atomic scale) does, due to the torque (5.101) it exerts on the magnetic moment
(5.95):
dL q
=τ=m×B= L × B. (5.114)
dt 2m 0
The vector diagram in figure 5.13 shows that in the limit of a relatively weak field,
when the magnitude of the angular momentum L may be considered constant, this
equation describes rotation (called the torque-induced precession50) of the vector L
about the direction of the vector B, with the angular frequency Ω = −qB/2m0,
independent of the angle θ. According to Eqs. (5.91) and (5.114), the resulting
additional (field-induced) magnetic moment Δm ∝ qΩ ∝ −q2B/m0 has, irrespectively
of the sign of q, a direction opposite to the field. Hence, according to Eq. (5.111) with
H ≈ B/μ0, χm ∝ χ ≡ Δm/H is indeed negative. (Let me leave its quantitative estimate
within this model for the reader’s exercise.) The quantum-mechanical treatment
confirms this qualitative picture of Larmor diamagnetism, while giving quantitative
corrections to the classical result for χm.51
49
After J Larmor (1857–1947) who was the first to describe the underlying mechanical effect, the torque-
induced precession, mathematically.
50
For a detailed discussion of the effect see, e.g. Part CM section 4.5.
51
See, e.g. Part QM section 6.4. The quantum mechanics also shows why in the most important s-states,
the contribution (5.95) of the basic angular momentum L to the average m vanishes—see, e.g. Part QM
section 3.6.
5-32
Figure 5.13. The torque-induced precession of a classical charged particle in a magnetic field.
A simple estimate (also left for the reader’s exercise) shows that in atoms with
uncompensated spins, the magnetic dipole orientation mechanism prevails over the
orbital diamagnetism, so that the materials incorporating such atoms usually exhibit
net paramagnetism—see table 5.1. Due to possible strong quantum interaction
between the spin dipole moments, the magnetism of such materials is rather complex,
with numerous interesting phenomena and elaborate theories. Unfortunately, all this
physics is well outside the framework of this course, and I have to refer the interested
reader to special literature52, but still need to mention some key notions we will need.
Most importantly, a sufficiently strong dipole–dipole interaction may lead to their
spontaneous ordering, even in the absence of an applied field. This ordering may
correspond to either parallel alignment of the magnetic dipoles (ferromagnetism) or
anti-parallel alignment of the adjacent dipoles (antiferromagnetism). Evidently, the
external effects of ferromagnetism are stronger, because such a phase corresponds to a
substantial spontaneous magnetization M even in the absence of an external magnetic
field. (The corresponding magnitude of B = μ0M is called the remanence field, BR). The
direction of the vector BR may be switched by the application of an external magnetic
field, with a magnitude above a certain value HC called coercivity, leading to the well-
known hysteretic loops on the [B, H] plane (see figure 5.14 for a typical example),
similar to those in ferroelectrics, already discussed in section 3.3.
Similarly to ferroelectrics, ferromagnets may also be hard or soft—in the magnetic
sense. In hard ferromagnets (also called permanent magnets), the dipole interaction is
so strong that B stays close to BR in all applied fields below HC, so that the hysteretic
loops are virtually rectangular. Correspondingly, the magnetization M of a permanent
magnet may be considered constant, with the magnitude BR/μ0. Such hard ferromag-
netic materials, with high remanence fields (typically, from 1 to 2 T) have numerous
practical applications. Let me give just two, perhaps the most important, examples.
First, permanent magnets are core components of most electric motors. By the
way, this venerable (∼150-year-old) technology is currently experiencing a quiet
revolution, driven mostly by the development of electric cars. In the most advanced
type of motors, called permanent-magnet synchronous machines (PMSM), the
remnant magnetic field BR of a permanent-magnet central part (called the rotor)
52
See, e.g. [3] or [4].
5-33
Figure 5.14. Experimental magnetization curves of specially processed (cold-rolled) electrical steel—a solid
solution of a few Si in Fe. (Reproduced from www.thefullwiki.org/Hysteresis under the Creative Commons
BY-SA 3.0 license.)
interacts with ac currents passed through wire windings in the external, static part of
the motor (called the stator). The resulting torque drives the rotor to extremely high
speeds, exceeding 10 000 rotations per minute, enabling the motor to deliver several
kilowatts of mechanical power from each kilogram of its mass.
As the second important example, despite the decades of the exponential (Moore’s
law) progress of semiconductor electronics, most computer data storage systems are
still based on hard disk drives whose active medium is a submicron-thin layer of a
hard ferromagnet, with the data bits stored in the form of the direction of the
permanent magnetization of small film spots. This technology has reached a
fantastic sophistication54, with a recorded data density of the order of 1012 bits
per square inch. Only recently has it started to be seriously challenged by the so-
called solid state drives based on the floating-gate semiconductor memories already
mentioned in chapter 3.55
In contrast, in soft ferromagnets, with their lower magnetic dipole interactions,
the magnetization is constant only within spontaneously formed magnetic domains,
while the volume and shape of the domains are affected by the applied magnetic
field. As a result, the hysteresis loop’s shape of soft ferromagnets is dependent on the
54
‘A magnetic head slider [the read/write head—KKL] flying over a disk surface with a flying height of 25 nm
with a relative speed of 20 m s−1 [all realistic parameters—KKL] is equivalent to an aircraft flying at a physical
spacing of 0.2 μm at 900 km h−1.’ B Bhushan, as quoted in the (generally good) book by G Hadjipanayis [5].
55
High-frequency properties of hard ferromagnets are also very non-trivial. For example, according to
Eq. (5.101), an external magnetic field Bext exerts torque τ = M × Bext on the spontaneous magnetic moment M
of a ferromagnetic sample. In some nearly-isotropic, mechanically fixed ferromagnetic samples, this torque
causes the precession around the direction of Bext (very similar to that illustrated in figure 5.13) of not the
sample as such, but of the magnetization M inside it, with a certain frequency ωr. If the frequency ω of an
additional RF field becomes very close to ωr, its absorption sharply increases—the so-called ferromagnetic
resonance. Moreover, if ω is somewhat higher than ωr, the effective RF magnetic permeability μ(ω) of the
material for the RF field may become negative, enabling a series of interesting effects and practical
applications. Most unfortunately, I do not have time for their discussion, and have to refer the interested
reader to special literature, for example to the monograph [6].
5-34
cycled field’s amplitude and cycling history—see figure 5.14. At high fields, their B
(and hence M) are driven into saturation, with B ≈ BR, but at low cycled fields they
behave essentially as linear magnetics with very high values of χm and hence μ—see
the top rows of table 5.1. (The magnetic domain interaction and hence the low-field
susceptibility of such soft ferromagnets are highly dependent on the material’s
fabrication technology and its post-fabrication thermal and mechanical treatment.)
Due to these high values of μ, the soft ferromagnets, in particular iron and its alloys
(e.g. various special steels), are extensively used in electrical engineering—for
example in the cores of transformers—see the next section.
Due to the relative weakness of the magnetic dipole interaction in some materials,
their ferromagnetic ordering may be destroyed by thermal fluctuations, if the
temperature is increased above some value called the Curie temperature TC.
The transition between the ferromagnetic and paramagnetic phase at T = TC is
the classical example of a continuous phase transition, with the average polarization
M playing the role of the so-called order parameter that (in the absence of external
fields) becomes different from zero only at T < TC, increasing gradually with the
further temperature reduction.56
5.6 Systems with magnetics

Similarly to the electrostatics of linear dielectrics, the magnetostatics of linear
magnetics is very simple in the particular case when the stand-alone currents are
embedded into a medium with a constant permeability μ. Indeed, let us assume that
we know the solution B0(r) of the magnetic pair of the genuine (‘microscopic’)
Maxwell equations (5.36) in free space, i.e. when the genuine current density j
coincides with that of stand-alone currents. Then the macroscopic Maxwell
equations (5.109) and the linear constitutive equation (5.110) are satisfied with the
pair of functions
B 0(r) μ
H(r) = , B(r) = μH(r) = B 0(r). (5.115)
μ0 μ0
Hence the only effect of the complete filling of a system of fixed currents with a
uniform, linear magnetic is the change of the magnetic field B at all points by the
same constant factor μ/μ0 ≡ 1 + χm, which may be either larger or smaller than 1.
(As a reminder, a similar filling of a system of fixed stand-alone charges with a
uniform, linear dielectric always leads to a reduction of the electric field E by a
factor of ε/ε0 ≡ 1 + χe—the difference whose physics was already discussed at the
end of section 5.4.)
However, this simple result is generally invalid in the case of non-uniform (or
piece-wise uniform) magnetic samples. To analyze them, let us first integrate the
macroscopic Maxwell equation (5.107) along a closed contour C limiting a smooth
56
A quantitative discussion of such transitions may be found, in particular, in Part SM chapter 4.
5-35
surface S. Now using the Stokes theorem, we obtain the macroscopic version of the
Ampère law (5.37):
∮C H ⋅ d r = I . (5.116)
Let us apply this relation to a sharp boundary between two regions with different
magnetics, with no stand-alone currents on the interface, absolutely similarly to how
this was done for field E in section 3.4—see figure 3.5. The result is similar as well:
Hτ = const. (5.117)
On the other hand, the integration of the Maxwell equation (5.29) over a Gaussian
pillbox enclosing a border fragment (again just as shown in figure 3.5 for the field D)
yields the result similar to Eq. (3.35):
Bn = const. (5.118)
For linear magnetics, with B = μH, the latter boundary condition is reduced to
μHn = const. (5.119)
Let us use these boundary conditions, first of all, to see what happens with a long
cylindrical sample of a uniform magnetic material, placed parallel to a uniform
external magnetic field B0—see figure 5.15. Such a sample cannot noticeably disturb
the field in the free space outside it, at most of its length: Bext = B0, Hext = μ0Bext =
μ0B0. Now applying Eq. (5.117) to the dominating side surfaces of the sample, we
obtain Hint = H0.57 For a linear magnetic, these relations yield Bint = μHint = (μ/μ0)
B0.58 For the high-μ, soft ferromagnetic materials, this means that Bint ≫ B0. This
effect may be vividly represented as the concentration of the magnetic field lines in
high-μ samples—see figure 5.15. (The concentration affects the external field
distribution only at distances of the order of (μ/μ0)t ≪ l near the sample’s ends.)
Such concentration is widely used in such practically important devices as
transformers, in which two multi-turn coils are wound on a ring-shaped (e.g.
toroidal, see figure 5.6b) core made of a soft ferromagnetic material (such as the
Figure 5.15. Schematic representation of the magnetic field concentration in long, high-μ magnetic samples.
57
The fact of constancy of H in this geometry explains why this field’s magnitude is used as the argument in
the plots such as figure 5.14: such measurements are typically carried out by placing an elongated sample of the
material under study into a long solenoid with a controllable current I, so that according to Eq. (5.116),
H0 = nI, regardless of the properties of the material under study.
58
The reader is highly encouraged to carry out a similar analysis of the fields inside narrow gaps cut in a linear
magnetic, similar to that carried in section 3.3 out for linear dielectrics—see figure 3.6 and its discussion.
5-36
transformer steel, see table 5.1) with μ ≫ μ0. This minimizes the number of ‘stray’
field lines, and makes the magnetic flux Φ piercing each wire turn (of either coil)
virtually the same—the equality important for the secondary voltage induction—see
the next chapter.
Samples of other geometries may create strong perturbations of the external field,
extended to distances of the order of the sample’s dimensions. In order to analyze
such problems, we may benefit from a simple partial differential equation for a scalar
function, e.g. the Laplace equation, because in chapter 2 we have learned how to
solve it for many simple geometries. In magnetostatics, the introduction of a scalar
potential is generally impossible due to the vortex-like magnetic field lines. However,
if there are no stand-alone currents within the region we are interested in, then the
macroscopic Maxwell equation (5.107) for the field H is reduced to ∇ × H = 0, similar
to Eq. (1.28) for the electric field, showing that we may introduce the scalar potential
of the magnetic field, ϕm, using the relation similar to Eq. (1.33):
H = −∇ϕm . (5.120)
Combining it with the homogenous Maxwell equation (5.29) for the magnetic field,
∇ · B = 0, and Eq. (5.110) for a linear magnetic, we arrive at a single differential
equation, ∇ · (μ∇ϕm) =0. For a uniform medium (μ(r) = const), it is reduced to our
beloved Laplace equation:
∇2 ϕm = 0. (5.121)
Moreover, Eqs. (5.117) and (5.119) give us very familiar boundary conditions: the
first of them
∂ϕm
= const, (5.122a )
∂τ
being equivalent to
ϕm = const, (5.122b)
with the second one giving
∂ϕm
μ = const. (5.123)
∂n
Indeed, these boundary conditions are absolutely similar for (3.37) and (3.56) of
electrostatics, with the replacement ε → μ.59
Let us analyze the geometric effects on magnetization, first using the (too?) familiar
structure: a sphere, made of a linear magnetic material, placed into a uniform external
field H0 ≡ B0/μ0. Since the differential equation and the boundary conditions are
59
This similarity may seem strange, because earlier we have seen that the parameter μ is physically more
similar to 1/ε. The reason for this paradox is that in magnetostatics, the introduced potential ϕm is traditionally
used to describe the ‘would-be field’ H, while in electrostatics, the potential ϕ describes the actual electric
field E. (This tradition persists from the days when H was perceived as a genuine magnetic field.)
5-37
similar to those of the corresponding electrostatics problem (see figure 3.11 and its
discussion), we can use the above analogy to reuse the solution we already have—see
Eqs. (3.63). Just as in the electric case, the field outside the sphere, with
⎛ μ − μ0 R3 ⎞
(ϕm)r>R = H0⎜−r + ⎟ cos θ , (5.124)
⎝ μ + 2μ 0 r 2 ⎠
is a sum of the uniform external field H0, with the potential −H0r cos θ ≡ −H0z, and the
dipole field (5.99) with the following induced magnetic dipole moment of the sphere60:
μ − μ0 3
m = 4π R H 0. (5.125)
μ + 2μ0
In contrast, the internal field is perfectly uniform, and directed along the external one:
3μ
(ϕm)r<R = −H0 μ + 20 μ r cos θ ,
0
(5.126)
H 3μ 0 Bint μHint 3μ
so that int = , = = .
H0 μ + 2μ 0 B0 μ0H0 μ + 2μ 0
Note that the field Hint inside the sphere is not equal to the applied external field H0.
This example shows that the interpretation of H as the ‘would-be’ magnetic field
generated by external currents j should not be exaggerated into saying that its
distribution is independent of the magnetic bodies in the system. In the limit μ ≫ μ0,
Eq. (5.126) yield Hint/H0 ≪ 1, Bint/H0 = 3μ0, the factor 3 being specific for the particular
geometry of the sphere. If a sample is strongly stretched along the applied field, with its
length l much larger than the thickness scale t, this geometric effect is gradually
decreased and Bint tends to its value μH0 ≫ B0, as was discussed above—see figure 5.15.
Now let us calculate the field distribution in a similar, but slightly more complex
(and practically important) system: a round cylindrical shell, made of a linear
magnet, placed in a uniform external field H0 normal to its axis—see figure 5.16.
Since there are no stand-alone currents in the region of our interest, we can again
represent the field H(r) by the gradient of the magnetic potential ϕm—see
Eq. (5.120). Inside each of three constant-μ regions, i.e. at ρ < b, a < ρ < b, and
b < ρ (where ρ is the distance from the cylinder’s axis), the potential obeys the
Laplace equation (5.121). In the convenient, polar coordinates (see figure 5.16), we
may, guided by the general solution (2.112) of the Laplace equation and our
experience in its application to axially symmetric geometries, look for ϕm in the
following form:
60
To derive Eq. (5.125), we may either calculate the gradient of the ϕm given by Eq. (5.124), or use the similarity of
Eqs. (3.13) and (5.99), to derive from Eq. (3.7) a similar expression for the magnetic dipole’s potential:
1 m cos θ
ϕm = .
4π r 2
Now comparing this formula with the second term of Eq. (5.124), we immediately obtain Eq. (5.125).
5-38
Figure 5.16. A cylindrical magnetic shield.
⎧( −H0ρ + b1′/ ρ) cos φ , for b ⩽ ρ ,

⎪
ϕm = ⎨ (a1ρ + b1/ ρ) cos φ , for a ⩽ ρ ⩽ b , (5.127)
⎪
⎩ −Hintρ cos φ , for ρ ⩽ a .
Plugging this solution into the boundary conditions (5.122) and (5.123) at both
interfaces (ρ = b and ρ = a), we obtain the following system of four equations:
− H0b + b1′ / b = a1b + b1/ b , (a1a + b1/ a ) = −Hinta ,
(5.128)
( 2
)
μ0 −H0 − b1′/ b H0 = μ(a1 − b1/ b ) , 2
μ(a1 − b1/ a 2 ) = −μ0Hint,
for four unknown coefficients a1, b1, b1ʹ, and Hint. Solving the system, we obtain, in
particular:
Hint αc − 1 ⎛ μ + μ ⎞2
= , with αc ≡ ⎜ 0
⎟ . (5.129)
H0 αc − (a / b)2 ⎝ μ − μ 0⎠
According to these formulas, at μ > μ0, the field in the free space inside the
cylinder is lower that the external field. This fact allows using such structures, made
of high-μ materials such as permalloy (see table 5.1), for the passive shielding61 from
unintentional magnetic fields (e.g. the Earth’s field)—a task very important for the
design of many physical experiments. As Eq. (5.129) shows, the larger μ is, the closer
αc is to 1, and the smaller is the ratio Hint/H0, i.e. the better is the shielding (for the
same a/b ratio). On the other hand, for a given magnetic material, i.e. for a fixed
parameter αc, the shielding is improved by making the ratio a/b < 1 smaller, i.e.
making the shield thicker. On the other hand, a smaller a leaves less space for the
shielded equipment/samples, calling for a compromise.
Now let us discuss a curious (and practically important) approach to systems with
relatively thin, closed magnetic cores made of sections of (possibly, different) high-μ
magnetics, with the cross-section areas Ak much smaller than the squared lengths lk
of the sections—see figure 5.17. If all μk ≫ μ0, virtually all field lines are confined to
the interior of the core. Then, applying the macroscopic Ampère law (5.116) to a
contour C that follows a magnetic field line inside the core (see, for example, the
61
A complementary approach to the undesirable magnetic fields’ reduction is the ‘active shielding’—the field
compensation with the counter-field induced by controlled currents in specially designed wire coils.
5-39
Figure 5.17. Deriving the ‘magnetic Ohm law’ (5.131).
dashed line in figure 5.17), we obtain the following approximate expression (exactly
valid only in the limit μk/μ0, lk2/Ak → ∞):
B
∮ Hl dl ≈∑lkHk ≡ ∑lk k = NI. (5.130)
C
k k
μk
However, since the magnetic field lines stay in the core, the magnetic flux Φk ≈ BkAk
should be the same (≡ Φ) for each section, so that Bk = Φ/Ak. Plugging this condition
into Eq. (5.130), we obtain
NI lk
Φ= , where R k ≡ .
∑R k μk Ak (5.131)
k
Note a close analogy of the first of these equations with the usual Ohm law for several
resistors connected in series, with the magnetic flux playing the role of electric current,
with the product NI, of the voltage applied to the resistor chain. This analogy is fortified
by the fact that the second of Eqs. (5.131) is similar to the expression for resistance R =
l/σA of a long, uniform conductor, with the magnetic permeability μ playing the role of
the electric conductivity σ. (In order to sound similar, but still different from the
resistance R , the parameter R is called reluctance.) This is why Eq. (5.131) is called the
magnetic Ohm law; it is very useful for approximate analyses of systems such as ac
transformers, magnetic energy storage systems, etc.
Now let me proceed to a brief discussion of systems with permanent magnets.
First of all, using the definition (5.108) of the field H, we may rewrite the Maxwell
equation (5.29) for the field B as
∇ ⋅ B ≡ μ0∇ ⋅ (H + M) = 0, i.e. as ∇ ⋅ H = −∇ ⋅ M, (5.132)
While this relation is general, it is particularly convenient in permanent magnets,

where M may be considered field-independent. In this case, Eq. (5.132) for H is an
exact analog of Eq. (1.27) for E, with the fixed term −∇ · M playing the role of the
fixed charge density (more exactly, of ρ/ε0). For the scalar potential ϕm, defined by
Eq. (5.120), this gives the Poisson equation
∇2 ϕm = ∇ ⋅ M, (5.133)
similar to those solved, for many electrostatic situations, in previous chapters.
5-40
In the particular case when the magnetization vector M is not only field-
independent, but also constant inside a permanent magnet’s volume, then the
right-hand sides of Eqs. (5.132) and (5.133) vanish both inside the volume and in the
surrounding free space, and give a non-zero effective charge only on the magnet’s
surface. Integrating Eq. (5.132) along a short path normal to the surface and
crossing it, we obtain the following boundary conditions:
ΔHn ≡ (Hn)in free space − (Hn)in magnet = Mn ≡ M cos θ , (5.134)
where θ is the angle between the magnetization vector and the outer normal to the
magnet’s surface. This relation is an exact analog of Eq. (1.24) for the normal component
of the field E, with the effective surface charge density (or rather σ/ε0) equal to Mcosθ.
This analogy between the magnetic field induced by a fixed, constant magnet-
ization and the electric field induced by surface electric charges enables one to reuse
quite a few problems considered in chapters 1–3. Leaving a few such problems for
the reader’s exercise (see section 5.7), let me demonstrate the power of this analogy
with two examples specific to magnetic systems. First, let us calculate the force
necessary to detach the flat ends of two long, uniform rod magnets, of length l and
cross-section area A ≪ l2, with the saturated magnetization M0 directed along their
length—see figure 5.18.
Figure 5.18. Detaching two magnets.
Let us assume we have succeeded in detaching the magnets by an infinitesimal

distance t ≪ A1/2, l. Then, according to Eqs. (5.133) and (5.134), the distribution of
the magnetic field near this small gap should be similar to that of the electric field in
a system of two equal but opposite surface charges with the surface density σ
proportional to M0. From chapters 1 and 2, we know the properties of such a system
very well: within the gap, the electric field is virtually constant, uniform, propor-
tional to σ, and independent of t, while outside the gap it is negligible. (Due to the
condition A ≪ l2, the effect of the similar effective charges at the ‘outer’ ends of
the rods on the field near the gap t is negligible.) Hence the magnetic field H inside
the gap is proportional to M0, and independent of A and t. Specifically, for its
magnitude, Eq. (5.134) gives simply H = M0, and hence B = μ0M0.
Now we could calculate Fmin as the force exerted by this field on the effective
surface ‘charges’. However, it is easier to find it from the following energy argument.
Since the magnetic field energy localized inside the magnets and near their outer ends
cannot depend on t, this small detachment may only alter the energy inside the gap.
For this part of energy, Eq. (5.57) yields:
2
B2 (μ0M0) μ M02A
ΔU = V= At ≡ 0 t. (5.135)
2μ0 2μ0 2
5-41
The gradient of this potential energy is equal to the attraction force F = −∇(ΔU),
trying to reduce ΔU by decreasing the gap, with the magnitude
∂(ΔU ) μ M02A
F = = 0 . (5.136)
∂t 2
The magnet detachment requires an equal and opposite force.
Now let us consider the situation where similar long permanent magnets (such as
the magnetic needles used in magnetic compasses) are separated, in otherwise free
space, by a larger distance d ≫ A1/2—see figure 5.19.
Figure 5.19. Schematic representations of (a) ‘magnetic charges’ at the ends of a thin permanent-magnet
needle and (b) the result of its breaking into two parts.
For each needle (figure 5.19a) of length l ≫ A1/2, the right-hand side of Eq. (5.133)
is substantially different from zero only in two relatively small areas at the needle’s
ends. Integrating the equation over each end, we see that at distances r ≫ A1/2 from
each end, we may reduce Eq. (5.132) to
∇ ⋅ H = qmδ(r − r+) − qmδ(r − r−), (5.137)
where r± are the ends’ positions, and qm ≡ M0A, with A being the needle’s
cross-section area. This equation is completely similar to Eq. (3.32) for the displace-
ment D, for the particular case of two equal and opposite point charges, i.e. with
ρ = qδ(r − r+) − qδ(r − r+), with the only replacement q → qm. Since we know the
resulting electric field all too well (see, e.g. equation (1.7) for E ≡ D/ε0), we may
immediately write a similar expression for the field H:
1 ⎛ r − r+ r − r− ⎞
H(r) = qm⎜ − ⎟. (5.138)
4π ⎝ r − r+ 3 r − r− 3⎠
The resulting magnetic field B(r) = μ0H exerts on another ‘magnetic charge’ qʹm,
located at point rʹ, force F = qʹmB(rʹ).62 Hence if two ends of different needles are
separated by an intermediate distance R (A1/2 ≪ R ≪ l, see figure 5.19b), we may
neglect one term in Eq. (5.138), and obtain the following ‘magnetic Coulomb law’
for the interaction of the nearest ends:
62
The simplest way to verify this (perhaps, obvious) expression is to check that for a system of two ‘charges’
±qʹm, separated by vector a, placed into a uniform external magnetic field Bext, it yields the potential energy
(5.100) with the correct magnetic moment m = qma—cf Eq. (3.9) for an electric dipole.
5-42
μ0 R
F=± qmq′m 3 . (5.139)
4π R
The ‘only’ (but conceptually, crucial!) difference of this interaction from that of the
electric point charges is that the ‘magnetic charges’ (quasi-monopoles) of one needle
cannot be fully separated. For example, if we break a magnetic needle in the middle in an
attempt to bring its two ends further apart, two new ‘charges’ appear—see figure 5.19b.
There are several solid state systems where more flexible structures, similar in their
magnetostatics to the needles, may be implemented. First of all, certain (‘type-II’)
superconductors may sustain so-called Abrikosov vortices—colloquially, flexible tubes
with field-suppressed superconductivity inside, each carrying one magnetic flux
quantum Φ0 = ℏ/πe ≈ 2 × 10−15 Wb. Ending on the superconductor’s surfaces, these
tubes let their magnetic field lines spread into the surrounding free space, essentially
forming magnetic monopole analogs—of course, with equal and opposite ‘magnetic
charges’ qm on each end of the tube. Such flux tubes are not only flexible but readily
stretchable, resulting in several peculiar effects—see section 6.4 for more detail.
Another, recently found, example of such paired quasi-monopoles includes spin chains
in the so-called spin ices—crystals with paramagnetic ions arranged into a specific
(pyrochlore) lattice—such as dysprosium titanate Dy2Ti2O7.63 Let me emphasize again
that any reference to magnetic monopoles in such systems should not be taken literally.
In order to complete this section (and this chapter), let me briefly discuss the
magnetic field energy U, for the simplest case of systems with linear magnetics. In
this case we still may use Eq. (5.55), but if we want to operate with the macroscopic
fields, and hence the stand-alone currents, we should repeat the manipulations that
have led us to Eq. (5.57), replacing j not from Eq. (5.35), but from Eq. (5.107). As a
result, instead of Eq. (5.57) we obtain
B⋅H B2 μH 2
U= ∫V u(r)d 3r, with u =
2
=
2μ
=
2
. (5.140)
This result is evidently similar to Eq. (3.73) of electrostatics.

As a simple but important example of its application, let us again consider a long
solenoid (figure 5.6a), but now filled with a linear magnetic material with perme-
ability μ. Using the macroscopic Ampère law (5.116), just as we used Eq. (5.37) for
the derivation of Eq. (5.40), we obtain
H = In , and hence B = μIn , (5.141)
where n ≡ N/l, just as in Eq. (5.40), is the winding density, i.e. the number of wire turns
per unit length. (At μ = μ0, we immediately return to that old result.) Now we may plug
Eq. (5.141) into Eq. (5.140) to calculate the magnetic energy stored in the solenoid,
μH 2 μ(nI )2 lA
U = uV = lA = , (5.142)
2 2
63
See, e.g. [7] and references therein.
5-43
and then use Eq. (5.72) to calculate its self-inductance64:
U
L= = μn 2lA . (5.143)
I 2 /2
We see that L ∝ μV, so that filling a solenoid with a high-μ material may allow
one to make it more compact by preserving the same value of inductance. In
addition, as the discussion of figure 5.15 has shown, such filling reduces the fringe
fields near the solenoid’s ends, which may be detrimental for some applications,
particularly in precise physical experiments.
Still, we need to explore the issue of the energy in magnetics beyond Eq. (5.140),
not only to obtain a general expression for it in materials with an arbitrary
dependence B(H), but also to finally prove Eq. (5.54) and explore its relation with
Eq. (5.53). I will do this at the beginning of the next chapter.
5.7 Problems
Problem 5.1. A circular wire loop, carrying a fixed dc current, has been placed inside
a similar but larger loop, carrying a fixed current in the same direction—see the
figure below. Use semi-quantitative arguments to analyze the mechanical stability of
the coaxial, coplanar position of the inner loop with respect to its possible angular,
axial, and lateral displacements relative to the outer loop.
Problem 5.2. Two straight, planar, parallel, long, thin conducting strips of width w,
separated by distance d, carry equal but oppositely directed currents I—see the figure
below. Calculate the magnetic field in the plane located at the middle between the
strips, assuming that the flowing currents are uniformly distributed across the strip
widths.
64
Admittedly, we could obtain the same result just by arguing that since the magnetic material fills the whole
volume of a substantial magnetic field in this system, the filling simply increases the vector B at all points and
hence its flux Φ, and hence L ≡ Φ/I by a factor of μ/μ0 in comparison with the free-space value (5.75).
5-44
Problem 5.3. For the system studied in the previous problem, but now only in the
limit d ≪ w, calculate:
(i) the distribution of the magnetic field (in the simplest possible way),
(ii) the vector-potential of the field,
(iii) the magnetic force (per unit length) exerted on each strip, and
(iv) the magnetic energy and self-inductance of the loop formed by the strips (per
unit length).
Problem 5.4. Calculate the magnetic field distribution near the center of the system
of two similar, planar, round, coaxial wire coils, fed by equal but oppositely directed
currents—see the figure below.
Problem 5.5. The two-coil-system, similar to that considered in the previous

problem, now carries equal and similarly directed currents—see the figure below.
Calculate what should be the ratio d/R for the second derivative ∂2Bz/∂z2 at z = 0 to
vanish65.
Problem 5.6. Calculate the magnetic field’s distribution along the axis of a straight
solenoid (see figure 5.6a, partly reproduced below) with a finite length l, and a round
cross-section of radius R. Assume that the solenoid has many (N ≫ 1, l/R) wire
turns, uniformly distributed along its length.
65
Such a system, producing a highly uniform field near its center, is called the Helmholtz coils, and is broadly
used in physics experiment.
5-45
Problem 5.7. A thin round disk of radius R, carrying electric charge of a constant
areal density σ, is being rotated around its axis with a constant angular velocity ω.
Calculate:
(i) the induced magnetic field on the disk’s axis,
(ii) the magnetic moment of the disk,
and relate these results.
Problem 5.8. A thin spherical shell of radius R, with charge Q uniformly distributed
over its surface, rotates about its axis with angular velocity ω. Calculate the
distribution of the magnetic field everywhere in space.
Problem 5.9. A sphere of radius R, made of an insulating material with a uniform

electric charge density ρ, rotates about its diameter with angular velocity ω.
Calculate the magnetic field distribution inside the sphere and outside it.
Problem 5.10. The reader is (hopefully:-) familiar with the classical Hall effect when it
takes place in the usual rectangular Hall bar geometry—see the left panel of the figure
below. However, the effect takes a different form in the so-called Corbino disk—see
the right panel below. (Dark shading shows electrodes, with no appreciable resist-
ance.) Analyze the effect in both geometries, assuming that in both cases the
conductors are thin, planar, have a constant Ohmic conductivity σ and charge carrier
density n, and that the applied magnetic field B is uniform and normal to the
conductors’ planes.
5-46
Problem 5.11.* The simplest model of the famous homopolar motor66 is a thin, round
conducting disk, placed in a uniform magnetic field normal to its plane, and fed by a
dc current flowing from the disk’s center to a sliding electrode (‘brush’) on its rim—
see the figure below.
(i) Express the torque, rotating the disk, via its radius R, the magnetic field B, and
the current I.
(ii) If the disk is allowed to rotate about its axis, and the motor is driven by a battery
with e.m.f. V , calculate its stationary angular velocity ω, neglecting friction and the
electric circuit’s resistance.
(iii) Now assuming that the current circuit (battery + wires + contacts + disk itself)
has a non-zero resistance R , derive and solve the equation for the time evolution of
ω, and analyze the solution.
Problem 5.12. Current I flows in a thin wire bent into a planar, round loop of radius
R. Calculate the net magnetic flux through the plane in which the loop is located.
Problem 5.13. Prove that:

(i) the self-inductance L of a current loop cannot be negative, and
(ii) any inductance coefficient Lkkʹ, defined by Eq. (5.60), cannot be larger than
(LkkLkʹkʹ)1/2.
Problem 5.14.* Estimate the values of magnetic susceptibility due to

(i) orbital diamagnetism, and
(ii) spin paramagnetism
for a medium with negligible interaction between the induced molecular dipoles.
Compare the results.
Hints: For task (i), you may use the classical model described by Eq. (5.114)—see
figure 5.13. For task (ii), assume the mechanism of ordering of spontaneous
magnetic dipoles m0, with a magnitude m0 of the order of the Bohr magneton μB,
similar to the one sketched for electric dipoles in figure 3.7a.
66
It was invented by M Faraday in 1821, i.e. well before his celebrated work on electromagnetic induction. The
adjective ‘homopolar’ refers to the constant ‘polarity’ (sign) of the current; the alternative term is ‘unipolar’.
5-47
Problem 5.15.* Use the classical picture of the orbital (‘Larmor’) diamagnetism,
discussed in section 5.5, to calculate its (small) correction ΔB(0) to the magnetic field
B felt by an atomic nucleus, modeling the electrons of the atom by a spherically
symmetric cloud with an electric charge density ρ(r). Express the result via the value
ϕ(0) of the electrostatic potential of the electron cloud, and use this expression for a
crude numerical estimate of the relative correction, ΔB(0)/B, for the hydrogen atom.
Problem 5.16. Calculate the (self-)inductance of a toroidal solenoid (see figure 5.6b)
with the round cross-section of radius r ∼ R (see the figure below), with many
(N ≫ 1, R/r) wire turns uniformly distributed along the perimeter, and filled with a
linear magnetic material of a magnetic permeability μ. Check your results by
analyzing the limit r ≪ R.
Problem 5.17. A long straight, thin wire, carrying current I, passes parallel to the
plane boundary between two uniform, linear magnetics—see the figure below.
Calculate the magnetic field everywhere in the system, and the force (per unit length)
exerted on the wire.
Problem 5.18. Solve the magnetic shielding problem similar to that discussed in
section 5.6, but for a spherical, rather than cylindrical shell, with the central cross-
section shown in figure 5.16. Compare the efficiency of these two shields, for the
same permeability μ of the shell, and the same b/a ratio.
Problem 5.19. Calculate the magnetic field distribution around a spherical perma-
nent magnet with a uniform magnetization M0 = const.
Problem 5.20. A limited volume V is filled with a magnetic material with a fixed
(field-independent) magnetization M(r). Write explicit expressions for the magnetic
field induced by the magnetization, and its potential, and recast these expressions
into forms more convenient when M(r) = M0 = const inside the volume V.
Problem 5.21. Use the results of the previous problem to calculate the distribution
of the magnetic field H along the axis of a straight permanent magnet of length 2l,
5-48
with a round cross-section of radius R, and a uniform magnetization M0 parallel to

the axis—see the figure below.
Problem 5.22. A very broad film of thickness 2t is permanently magnetized

normally to its plane, with a periodic checkerboard pattern, with the square of
area a × a:
⎛ πx πy ⎞⎟
M∣ z <t = nzM (x , y ), with M (x , y ) = M0 × sgn ⎜cos cos .
⎝ a a⎠
Calculate the magnetic field’s distribution in space.67
Problem 5.23.* Based on the discussion of the quadrupole electrostatic lens in

section 2.4, suggest permanent-magnet systems that may similarly focus particles
moving close to the system’s axis, and carrying:
(i) an electric charge,
(ii) no net electric charge, but a spontaneous magnetic dipole moment m.
References
[1] Grover F 1946 Inductance Calculations (New York: Dover)
[2] Landau L et al 1984 Electrodynamics of Continuous Media 2nd edn (Butterworth: Heinemann)
[3] Jiles D J 1998 Introduction to Magnetism and Magnetic Materials 2nd edn (Boca Raton FL:
CRC Press)
[4] O’Handley R 1999 Modern Magnetic Materials (Wiley)
[5] Hadjipanayis G 2001 Magnetic Storage Systems Beyond 2000 (Springer)
[6] Gurevich A and Melkov G 1996 Magnetization Oscillations and Waves (Boca Raton FL:
CRC Press)
[7] Jaubert L and Holdworth P 2011 J. Phys.: Condens. Matter 23 164222
67
This problem is of an evident relevance for the perpendicular magnetic recording (PMR) technology, which
currently dominates high-density digital magnetic recording.
5-49
IOP Publishing
Lecture notes
Chapter 6
Electromagnetism
This chapter discusses two major new effects that arise when the electric and magnetic
fields are changing in time: the ‘electromagnetic induction’ of an additional electric
field by the changing magnetic field, and the reciprocal effect of ‘displacement
currents’—actually, the induction of an additional magnetic field by the changing
electric field. These two phenomena, which make the time-dependent electric and
magnetic fields inseparable (hence the term ‘electromagnetism’), are reflected in the
full system of Maxwell equations, valid for an arbitrary electromagnetic process. On
the way toward this system, I will make a pause for a brief review of the electro-
dynamics of superconductivity, which (in addition to its own significance), provides a
perfect platform for the discussion of the general issue of gauge invariance.
6.1 Electromagnetic induction

As Eqs. (5.36) show, in static situations (∂/∂t = 0) the Maxwell equations describing
the electric and magnetic fields are independent, and are coupled only implicitly, via
the continuity equation (4.5) relating their right-hand sides ρ and j. (In statics, this
relation imposes a restriction only on the vector j.) In dynamics, when the fields
change in time, the situation is different.
Historically, the first discovered explicit coupling between the electric and
magnetic fields was the effect of electromagnetic induction1. The summary of
Faraday’s numerous experiments has turned out to be very simple: if the magnetic
flux, defined by Eq. (5.65),
Φ≡ ∫S Bn d 2r, (6.1)
1
The induction e.m.f. was discovered independently by J Henry, but is was the brilliant series of experiments
by M Faraday, carried out mostly in 1831, which resulted in the general formulation of the induction law.

through a surface S limited by contour C, changes in time for whatever reason (e.g.
either due to a change of the magnetic field B (as in figure 6.1), or the contour’s
motion, or its deformation, or any combination of the above), it induces an
additional, vortex-like electric field Eind—see figure 6.1.
Figure 6.1. The two simplest ways to observe the Faraday electromagnetic induction.
The exact distribution of Eind in space depends on system geometry details, but its
integral along the contour C, called the inductive electromotive force (e.m.f.), obeys a
very simple Faraday induction law2:
V ind ≡ ∮C Eind ⋅ d r = − ddtΦ . (6.2)
It is straightforward (and hence left for the reader’s exercise) to show that this e.m.f.
may be measured, for example, either by inserting a voltmeter into a conducting loop
following contour C, or by measuring the small current I = V ind /R it induces in a thin
wire with a sufficiently large Ohmic resistance R,3 whose shape follows the contour—
see figure 6.1. (Actually, these methods are not entirely different, because some
practical voltmeters measure voltage by the small Ohmic current it drives through a
known high resistance.) In the context of the latter approach, it is easy to formulate the
so-called Lenz rule used for the description of the minus sign in Eq. (6.2): the additional
magnetic field of the induced current I provides a partial compensation of the change of
the original flux Φ(t) with time4.
In order to recast Eq. (6.2) in a differential form, let us apply to the contour
integral in it, the same Stokes theorem that was repeatedly used in chapter 5. The
result is
V ind = ∫S (∇ × Eind )n d 2r. (6.3)
Now combining Eqs. (6.1)–(6.3), for a contour C whose shape does not change in time
(so that the integration along it is interchangeable with the time derivative), we obtain
⎛ ⎞
∫S ⎜⎝∇ × Eind + ∂∂Bt ⎟⎠ d 2r = 0. (6.4)
n
2
In Gaussian units, the right-hand side of this formula has the additional coefficient 1/c.
3
Such induced current is sometimes called the eddy current, although most often this term is saved for the
distributed currents induced by a changing magnetic field in a bulk conductor—see section 6.3.
4
Let me hope that the reader is also familiar with the paradox arising in attempts to measure V ind with a
voltmeter, without opening the wire loop; if not, I would highly recommend solving problem 6.2.
6-2
Since the induced electric field is additional to the gradient field (1.33) created by
electric charges, for the net field we may write E = Eind − ∇ϕ. However, since the
curl of any gradient field is zero5, ∇ × (∇ϕ) = 0, Eq. (6.4) remains valid for the net
field E. Since this equation should be correct for any closed area S, we may conclude
that
∂B
∇×E+ =0 (6.5)
∂t
at any point. This is the final (time-dependent) form of this Maxwell equation.
Superficially, it may look as if Eq. (6.5) is less general than Eq. (6.2); for example it
does not describe any electric field—and hence any e.m.f. in a moving loop—if the
field B is constant in time, even if the magnetic flux (6.1) through the loop does
change in time. However, this is not true; in chapter 9 we will see that in the reference
frame moving with the loop such an e.m.f. does appear6.
Now let us re-formulate Eq. (6.5) in terms of the vector-potential A. Since the
induction effect does not alter the fundamental relation ∇ · B = 0, we still may
represent the magnetic field as prescribed by Eq. (5.27), i.e. as B = ∇ × A. Plugging
this expression into Eq. (6.5), and changing the order of the temporal and spatial
differentiation, we obtain
⎛ ∂A ⎞
∇ × ⎜E + ⎟ = 0. (6.6)
⎝ ∂t ⎠
Hence we can use the same argumentation as in section 1.3 (there applied to the
vector E alone) to represent the expression in the parentheses as −∇ϕ, so that we
obtain
∂A
E=− − ∇ϕ , B = ∇ × A. (6.7)
∂t
It is very tempting to interpret the first term on the right-hand side of the
expression for E as describing the electromagnetic induction alone, and the second
term representing a purely electrostatic field induced by electric charges. However,
the separation of these two terms is, to a certain extent, conditional. Indeed, let us
consider the gauge transformation already mentioned in section 5.2,
A → A + ∇χ , (6.8)
5
See, e.g. Eq. (A.71).
6
I have to admit that from the beginning of the course, I was carefully sweeping under the rug a very important
question: in exactly which reference frame(s) are all the equations of electrodynamics valid? I promise to
discuss this issue in detail later in the course (in chapter 9), and for now would like to get away with a very
short answer: all the formulas discussed so far are valid in any inertial reference frame, as defined in classical
mechanics—see, e.g. Part CM section 1.3. It is crucial, however, to have fields E and B measured in the same
reference frame.
6-3
which, as we already know, does not change the magnetic field. According to
Eq. (6.8), in order to keep the full electric field intact (gauge-invariant) as well, the
scalar electric potential has to be transformed simultaneously as
∂χ
ϕ→ϕ− , (6.9)
∂t
leaving the choice of a time-independent addition to ϕ restricted only by the Laplace
equation—since the full ϕ should satisfy the Poisson equation (1.41) with a gauge-
invariant right-hand side. We will return to the discussion of gauge invariance in
section 6.4.
6.2 Magnetic energy revisited

Now we are sufficiently equipped to return to the issue of magnetic energy, in particular
to finally prove Eqs. (5.57) and (5.140) and discuss the dichotomy of the signs in Eqs.
(5.53) and (5.54)7. For that, let us consider a sufficiently slow, small magnetic field
variation δB. If we want to neglect the kinetic energy of the system of electric currents
under consideration, as well as the wave radiation effects, we need to prevent its
acceleration by the arising induction field Eind. Let us suppose that we do this by the
virtual balancing this field by an external electric field Eext = −Eind. According to
Eq. (4.38), the work of that field8 on the stand-alone currents of the system during time
interval δt, and hence the change of the potential energy of the system, is
δU = δt ∫V j ⋅ Eext d 3r=−δt∫V j ⋅ Eind d 3r, (6.10)
where the integral is over the volume of the system. Now expressing the current
density j from the macroscopic Maxwell equation (5.107), and then applying the
vector algebra identity9
(∇ × H) ⋅ E ind ≡ H ⋅ (∇ × E ind) − ∇ ⋅ (E ind × H), (6.11)
we obtain
δU = − δt ∫V H ⋅ (∇ × E)d 3r + δt∫V ∇ ⋅ (E × H)d 3r. (6.12)
According to the divergence theorem, the second integral on the right-hand side is
equal to the flux of the vector S ≡ E × H through the surface limiting the considered
volume V. Later in the course we will see that this flux represents, in particular, the
power of electromagnetic radiation through the surface. If such radiation is
negligible (as it always is if the field variation is sufficiently slow) the surface may
7
Actually, this dichotomy should not be too puzzling to the reader who has understood a similar sign duality
in electrostatics—cf, e.g. Eqs. (3.73) and (3.81).
8
As a reminder, the magnetic component of the Lorentz force (5.10), v × B, is always perpendicular to the
particle’s velocity, so that the magnetic field B itself cannot perform any work on moving charges, i.e. on
currents.
9
See, e.g. Eq. (A.77) with f = Eind and g = H.
6-4
be selected sufficiently far such as that the flux of S is negligible. In this case, we may
express ∇ × E from the Faraday induction law (6.5) to obtain
⎛ ⎞
δU = − δt ∫V ⎜⎝− ∂∂Bt ⎟⎠ ⋅ H d 3r = ∫V H ⋅ δ B d 3r. (6.13)
Just as in electrostatics (see Eqs. (1.65) and (3.73), and their discussion), this relation
may be interpreted as the variation of the magnetic field energy U of the system, and
represented in the form
δU = ∫V δu(r)d 3r, with δu ≡ H ⋅ δ B . (6.14)
This is a keystone result; let us discuss it in some detail.

First of all, for a system filled with a linear magnetic material, we may use
Eq. (6.14) together with Eq. (5.110): B = μH. Integrating the result over the variation
of B from 0 to a certain final value, we obtain Eq. (5.140)—so important that it is
worth rewriting again:
B2
U= ∫V u(r)d 3r, with u=
2μ
. (6.15)
In the simplest case of free space (no magnetics at all, so that j above is the complete
current density), we may take μ = μ0, and reduce Eq. (6.15) to Eq. (5.57). Now
performing backwards the transformations that took us, in section 5.3, to derive that
relation from Eq. (5.54), we finally have the latter formula proved—as was
promised.
It is very important, however, to understand the limitations of Eq. (6.15). For
example, let us try to apply it to a very simple problem, which was already analyzed
in section 5.6 (see figure 5.15): a very long cylindrical sample of a linear magnetic
material placed into a fixed external field Hext, parallel to the sample’s axis. It is
evident that in this simple geometry, the field H and hence the field B = μH have to
be uniform inside the sample, apart from negligible regions near its ends, so that
Eq. (6.15) is reduced to
B2
U= V, (6.16)
2μ
where V = Al is the cylinder’s volume. Now if we try to calculate the static
(equilibrium) value of the field from the minimum of this potential energy, we obtain
evident nonsense: B = 0 (WRONG!)10.
The situation may be readily rectified by using the notion of the Gibbs potential
energy, just as was done for the electric field in section 3.5 (and implicitly in the end
of section 1.3). According to Eq. (6.14), in magnetostatics the Cartesian components
of the field H(r) play the role of generalized forces, while those of the field B(r) play
10
Note that this erroneous result cannot be corrected by just adding the energy (6.15) of the field outside the
cylinder, because in the limit A → 0, this field is not affected by the internal field B.
6-5
the role of generalized coordinates (per unit volume)11. As a result, the Gibbs
potential energy, whose minimum corresponds to the stable equilibrium of the
system under the effect of a fixed generalized force (in our current case, in a fixed
external field Hext), is
UG = ∫V u G(r)d 3r, with u G(r) ≡ u(r) − H ext(r) ⋅ B(r), (6.17)
an expression to be compared with Eq. (3.78). For a system with linear magnetics,
we may use Eq. (6.15) for u, obtaining the following Gibbs energy’s density:
1 1
u G (r) = B ⋅ B − H ext ⋅ B ≡ (B − μH ext)2 + const, (6.18)
2μ 2μ
where ‘const’ means a term independent of the field B. For our simple cylindrical
system, with its uniform fields, Eq. (6.18) gives the following full Gibbs energy of the
sample:
(B int − μH ext)2
UG = V + const, (6.19)
2μ
whose minimum immediately gives the correct stationary value Bint = μHext, i.e.
Hint ≡ Bint/μ = Hext—which was already obtained in section 5.6 in a different way,
from the boundary condition (5.117).
Now notice that with this result on hand, Eq. (6.18) may be rewritten in a
different form,
1 B B2
u G (r) = B⋅B− ⋅B≡− , (6.20)
2μ μ 2μ
similar to Eq. (6.15) for u(r), but with an opposite sign. This sign dichotomy explains
that in Eqs. (5.53) and Eq. (5.54); indeed, as was already noted in section 5.3, the
former of these expressions gives the potential energy whose minimum corresponds to
the equilibrium of a system with fixed currents. According to Eq. (5.107), ∇ × H = j,
this condition is fulfilled if the field H is fixed12. So, the energy Uj given by Eq. (5.53)
is essentially the Gibbs energy UG defined by Eqs. (6.17) and (for the case of linear
magnetics, or no magnetic media at all) by Eq. (6.20), while Eq. (5.54) is just another
form of Eq. (6.15)—as was explicitly shown in section 5.313.
Let me complete this section by stating that the difference between the energies U
and UG is not properly emphasized (or is even left obscure) in some textbooks, so
that the reader is advised to seek additional clarity—for example, by spelling them
11
Note that in this respect, the analogy with electrostatics is not quite complete. Indeed, according to
Eq. (3.76), in electrostatics the role of a generalized coordinate is played by ‘would-be’ field D, and that of the
generalized force, by the actual electric field E. This difference may be traced back to the fact that electric field
E may perform work on a moving charged particle, while the magnetic field cannot. However, this difference
does not affect the full analogy of expressions (3.73) and (6.15) for the field energy density in linear media.
12
The opposite is not always true—see, e.g. the example discussed in section 5.6.
13
As was already noted in section 5.4, one more example of the energy Uj is given by Eq. (5.100).
6-6
out for a simple case of a long straight solenoid (figure 5.6a), and then using these
formulas to calculate the pressure exerted by the magnetic field on the solenoid’s
walls (windings) and the longitudinal forces exerted on its ends.
6.3 Quasi-static approximation and skin effect

Perhaps the most surprising experimental fact concerning time-dependent electro-
magnetic phenomena is that unless they are so fast that one more new effect,
displacement currents (to be discussed in section 6.7 below), becomes noticeable, all
formulas of electrostatics and magnetostatics remain valid, with only the exception:
the generalization of Eq. (3.36) to Eq. (6.5), describing the Faraday induction. As a
result, the system of macroscopic Maxwell equations (5.109) is generalized to
∂B
∇×E+ =0, ∇×H=j,
∂t (6.21)
∇⋅D=ρ, ∇⋅B=0.
(As follows from the discussions in chapters 3 and 5, the corresponding system of
microscopic Maxwell equations for genuine fields E and B may be obtained from
Eq. (6.16) by the formal substitutions D = ε0E and H = B/μ0, and the replacement of
the stand-alone charge and current densities ρ and j with their full densities14.) These
equations, whose range of validity will be quantified in section 6.7, define the so-
called quasi-static approximation of electromagnetism, and are sufficient for
adequate description of a broad range of physical effects.
Let us use them first of all for an analysis of the so-called skin effect, the
phenomenon of self-shielding of the alternating (ac) magnetic field by the eddy
currents induced by the field in an Ohmic conductor. In order to form a complete
system of equations, Eq. (6.21) should be augmented by constituent equations
describing the medium. Let us take them, for a conductor, in the simplest (and
simultaneously, most common) linear and isotropic forms already discussed in
chapters 4 and 5:
j = σ E, B = μH . (6.22)
If the conductor is uniform, i.e. the coefficients σ and μ are constant inside it, the
whole system of Eqs. (6.21) and (6.22) may be reduced to just one equation. Indeed,
a sequential substitution of these equations into each other, using a well-known
vector-algebra identity15 in the middle, yields:
14
Obviously, in free space the last replacement is unnecessary, because all charges and currents may be treated
as ‘stand-alone’ ones.
15
See, e.g. Eq. (A.73).
6-7
∂B 1 1
= − ∇ × E = − ∇ × j = − ∇ × (∇ × H)
∂t σ σ
1 1
=− ∇ × (∇ × B) ≡ − [∇(∇ ⋅ B) − ∇2 B] (6.23)
σμ σμ
1 2
= ∇ B.
σμ
Thus we have arrived, without any further assumptions, at a rather simple partial
differential equation. Let us use it to analyze the skin effect in the simplest geometry
(figure 6.2a) when an external source (which, at this point, does not need to be
specified) has produced, near a plane surface of a bulk conductor, a spatially
uniform ac magnetic field H(0)(t) parallel to the surface16.
Figure 6.2. (a) The skin effect in the simplest, planar geometry and (b) two Ampère contours, C1 and C2, for
deriving the ‘microscopic’ (C1) and the ‘macroscopic’ (C2) boundary conditions for H.
Selecting the coordinate system as shown in figure 6.2a, we may express this
condition as
H x=−0 = H (0)(t )ny . (6.24)
The translational symmetry of our simple problem within the surface plane [y, z]
implies that inside the conductor ∂/∂y = ∂/∂z = 0 as well, and H = H(x, t)ny even at
x ⩾ 0, so that Eq. (6.23) for the conductor’s interior is reduced to a differential
equation for just one scalar function H(x, t) = B(x, t)/μ:
∂H 1 ∂ 2H
= , for x ⩾ 0. (6.25)
∂t σμ ∂x 2
16
Due to the simple linear relation B = μH between the fields B and H, it does not matter too much which of
them is used for the solution of this problem, with a slight preference for H, due to the simplicity of Eq. (5.117)—
the only boundary condition relevant for this simple geometry.
6-8
This equation may be further simplified by noticing that due to its linearity, we may
use the linear superposition principle for the time dependence of the field17, via
expanding it, as well as the external field (6.24), into the Fourier series:
H (x , t ) = ∑Hω(x)e−iωt , for x ⩾ 0,
ω
(6.26)
H (0)(t ) = ∑Hω(0)e−iωt , for x = 0,
ω
and arguing that if we know the solution for each frequency component of the series,
the whole field may be found through the elementary summation (6.26) of these
solutions. For each a single-frequency component, Eq. (6.25) is immediately reduced
to an ordinary differential equation for the complex amplitude Hω(x):18
1 d2
−iωHω = Hω. (6.27)
σμ dx 2
From the theory of linear ordinary differential equations we know that Eq. (6.27)
has the following general solution,
Hω(x ) = H+e κ+x + H−e κ−x , (6.28)
where constants κ± are roots of the characteristic equation that may be obtained by
the substitution of any of these two exponents into the initial differential equation.
For our particular case, the characteristic equation, following from Eq. (6.27), is
κ2
−iω = (6.29)
σμ
and its roots are complex constants
1−i
κ± = ( −iμωσ )1/2 ≡ ± (μωσ )1/2 . (6.30)
2
For our problem, the field cannot grow exponentially at x → +∞, so that only one
of the coefficients, namely H− corresponding to the decaying exponent with Re κ < 0
(i.e. κ = κ−), may be non-vanishing, so that Hω(x) = Hω(0)exp{κ−x}. In order to find
the constant factor Hω(0), we can integrate the macroscopic Maxwell equation ∇ ×
H = j along a pre-surface contour—say, the contour C1 shown in figure 6.2b. The
right-hand side’s integral is negligible, because the stand-alone current density j does
17
Another important way to exploit the linearity of Eq. (6.25) is to use the spatial-temporal Green’s function
approach to explore the dependence of its solutions on various initial conditions. Unfortunately, due to a lack
of time, I have to leave an analysis of this opportunity to a reader’s exercise.
18
Let me hope that the reader is not intimidated by the (very convenient) use of such complex variables; their
imaginary parts always disappear at the final summation (6.26). For example, if the external field is purely
sinusoidal, with the actual (positive) frequency ω, each sum in Eq. (6.26) has just two terms, with complex
amplitudes Hω and H−ω = Hω*, so that their sum is always real. (For a more detailed discussion of this issue,
see, e.g. Part CM section 5.1.)
6-9
not include the ‘genuinely surface’ currents responsible for the magnetic properties
of the material (see figure 5.12). As a result, we obtain the ‘microscopic’19 boundary
condition similar to Eq. (5.117) for the stationary magnetic field, Hτ = const at x = 0,
i.e.
H (0, t ) = H (0)(t ), i.e. Hω(0) = Hω(0), (6.31)
so that the final solution of our problem may be represented as

⎧ x⎫ ⎧ ⎛ x ⎞⎫
Hω(x ) = Hω(0) exp ⎨ − ⎬ exp ⎨ −i ⎜ωt − ⎟⎬ , (6.32)
⎩ δs ⎭ ⎩ ⎝ δs ⎠⎭
where the constant δs, with the dimension of length, is called the skin depth:
1 ⎛ 2 ⎞1/2
δs ≡ − =⎜ ⎟ . (6.33)
Re κ− ⎝ μσω ⎠
This solution describes the skin effect: the penetration of the ac magnetic field of
frequency ω into a conductor only to a finite depth of the order of δs.20 Let me give a few
numerical examples of the skin depth. For copper at room temperature, δs ≈ 1 cm at
the standard ac power distribution frequency of 60 Hz, and is of the order of just 1 μm
at a few GHz, i.e. at typical frequencies of cell phone signals and kitchen microwave
magnetrons. For sea water, δs is close to 250 m at just 1 Hz (with big implications for
radio communications with submarines), and is of the order of 1 cm at a few GHz
(explaining the non-uniform heating of a soup bowl in a microwave oven).
In order to complete the skin effect discussion, let us consider what happens with
the induced eddy currents21 and the electric field at this effect. When deriving our
basic equation (6.23), we have used, in particular, relations j = ∇ × H = μ−1∇ × B,
and E = j/σ. Since a spatial differentiation of an exponent yields a similar exponent,
the electric field and current density have the same spatial dependence as the
magnetic field, i.e. penetrate inside the conductor only by distances of the order of
δs(ω), but their vectors are directed perpendicularly to B, while still being parallel to
the conductor surface:
κ
jω (x ) = κ−Hω(x ) nz , Eω(x ) = − Hω(x ) nz . (6.34)
σ
We may use these expressions to calculate the time-averaged power density (4.39)
of the energy dissipation, for the important case of a sinusoidal (‘monochromatic’)
19
This common name is awkward, because Eq. (6.31) results from the macroscopic Maxwell equations (6.21),
but is partly justified as the counterpart to the ‘macroscopic’ (better called ‘coarse-grain’) boundary conditions
(6.38) and (6.59)—to be discussed in a minute.
20
Let me hope that the physical intuition of the reader makes it evident that the ac field penetrates into a
sample of any shape only by a distance of the order of δs.
21
The loop (vortex) character of the induced current lines, responsible for the term ‘eddy’, is not very
apparent in the 1D geometry explored above, with the near-surface currents (figure 6.2b) looping only
implicitly, at z → ±∞.
6-10
field H(x, t) = ∣Hω(x)∣ cos(ωt + φ), and hence sinusoidal eddy currents: j(x, t) =
∣jω(x)∣ cos(ωt + φ′):
2 2
j 2 (x , t ) jω (x ) cos2 (ωt + φ′) jω (x )
p(x ) = = =
σ σ 2σ (6.35)
2 2 2 *
H (x )H (x )
κ Hω(x ) H (x )
= − = ω2 ≡ ω 2 ω .
2σ δs σ δs σ
Now the (elementary) integration of this expression along the normal to the surface,
i.e. along axis x (through all the skin depth), using the exponential law (6.32), gives
us the following average power of energy loss per unit area:
∞
dP 1 2 μωδs 2
dA
≡ ∫0 p(x )dx =
2δsσ
Hω(0) ≡
4
Hω(0) . (6.36)
We will extensively use this expression in the next chapter to calculate the energy
losses in waveguides and resonators with conducting (practically, metallic) walls, but
for now just note that according to Eqs. (6.33) and (6.36) at a fixed applied field the
losses grow with frequency as ω1/2.
One more important remark concerning Eq. (6.34): integrating the first of them
over x, with the help of Eq. (6.32), we may readily prove that the linear density J of
the surface currents (measured in A m−1) is simply and fundamentally related to the
applied magnetic field:
∞
Jω ≡ ∫0 jω (x )dx = Hω(0)nz . (6.37)
Since this relation does not have any frequency-dependent factors, we may sum it up
for all frequency components, and obtain a universal relation
J(t ) = H (0)(t )nz ≡ H (0)(t )( −ny × nx) = H (0)(t ) × ( −nx) = H (0)(t ) × n , (6.38)
where n = −nx is the outer normal to the surface—see figure 6.2b. This simple
relation (whose last form is independent of the choice of coordinate axes) is
independent of the used constituent relations (6.22) and is by no means occasional.
Indeed, it may be readily obtained from the macroscopic Ampère law Eq. (5.116)
applied to a contour drawn around a fragment of the surface, extending under it
much deeper than the skin depth—see the contour C2 in figure 6.2b—regardless of
the exact law of the field penetration. The ‘coarse-grain’ relation (6.38) is sometimes
called the ‘macroscopic’ boundary condition for the magnetic field near a con-
ductor’s surface, to distinguish it from the ‘microscopic’ boundary condition (6.31).
For the skin effect, this fundamental relation between the linear current density and
the external magnetic field implies that the skin effect’s implementation does not
necessarily require a dedicated ac magnetic field source. For example, the effect takes
place in any wire that carries an ac current, leading to a current concentration in a
surface sheet of thickness ∼δs. (Of course, the quantitative analysis of this problem in a
wire with an arbitrary cross-section may be technically complicated, because it requires
solving Eq. (6.23) for a 2D geometry; even for a round cross-section, the solution
6-11
involves the Bessel functions.) In this case, the ac magnetic field outside the conductor,
which still obeys Eq. (6.38), is better interpreted as an effect of, rather than the reason
for, the ac current flow.
Finally, the reader should mind the validity limits of all these results—apart from
the universal Eq. (6.38). First, in order for the quasi-static approximation to be
valid, the field frequency ω should not be too high, so that the displacement current
effects are negligible. (Again, this condition will be quantified in section 6.7; it will
show that for usual metals, the condition is violated only at extremely high
frequencies above ∼1018 s−1.) A more practical upper limit on ω is that the skin
depth δs should stay much larger than the mean free path l of charge carriers22.
Beyond this point, the relation between the vectors j(r) and E(r) becomes essentially
non-local. Both theory and experiment show that at δs < l the skin effect still persists,
but acquires a frequency dependence slightly different from Eq. (6.33), δs ∝ ω−1/3
rather than ω−1/2. This so-called anomalous skin effect has useful applications, for
example, for experimental measurements of the Fermi surface of metals23.
6.4 Electrodynamics of superconductivity and gauge invariance

The effect of superconductivity24 takes place (in certain materials only, mostly
metals) when temperature T is reduced below a certain critical temperature Tc,
specific for each material. For most metallic superconductors, Tc is typically of the
order of a few kelvins, although several compounds (the so-called high-temperature
superconductors) with Tc above 100 K have been found since 1987. The most notable
property of superconductors is the absence at T < Tc of measurable resistance to (not
very high) dc currents. However, the electromagnetic properties of superconductors
cannot be described by just taking σ = ∞ in our previous results. Indeed, for this
case, Eq. (6.33) would give δs = 0, i.e. no ac magnetic field penetration at all, while
for the dc field we would have the uncertainty σω → ? Experiments show something
substantially different: weak magnetic fields do penetrate into superconductors by a
material-specific distance δL ∼ 10−7–10−6 m, the so-called London penetration
depth25, which is virtually frequency-independent until the skin depth δs—measured
in the same material in its ‘normal’ state, i.e. the absence of superconductivity—
becomes less than δL. (This crossover happens typically at frequencies ∼ 1013–1014 s−1.)
The smallness of δL on the human scale means that the magnetic field is pushed out of
macroscopic samples at their transition into the superconducting state.
This Meissner–Ochsenfeld effect, discovered experimentally in 193326, may be
partly understood using the following classical reasoning. The discussion of the Ohm
22
A brief discussion of the mean free path may be found, for example, in Part SM chapter 6. In very clean
metals at very low temperatures, δs may approach l at frequencies as low as ∼1 GHz, but at room temperature
the crossover from the normal to the anomalous skin affect takes place only at ∼100 GHz.
23
See, e.g. [1].
24
Discovered experimentally in 1911 by H Kamerlingh Onnes.
25
Named to acknowledge the pioneering theoretical work of brothers F and H London—see below.
26
It is hardly fair to shorten the name to just the ‘Meissner effect’, as is frequently done, because of the
reportedly crucial contribution by R Ochsenfeld, then W Meissner’s student, to the discovery.
6-12
law in section 4.2 implied that the current’s (and hence the electric field’s) frequency
ω is either zero or sufficiently low. In the classical Drude reasoning, this is acceptable
while ωτ ≪ 1, where τ is the effective carrier scattering time participating in
Eqs. (4.12) and (4.13). If this condition is not satisfied, we should take into account
the charge carrier inertia; moreover, in the opposite limit ωτ ≫ 1 we may neglect the
scattering at all. Classically, we can describe the charge carriers in such a ‘perfect
conductor’ as particles with a non-zero mass, which are accelerated by the electric
field in accordance with Newton’s second law (4.11),
mv ̇ = F = q E, (6.39)
so that the current density j = qnv they create changes in time as
q 2n
j̇ = E. (6.40)
m
In terms of the Fourier amplitudes of the functions j(t) and E(t), this means
q 2n
−iω jω = Eω . (6.41)
m
Comparing this formula with the relation jω = σEω implied in the last section, we see
that we can use all its results with the following replacement:
q 2n
σ→i . (6.42)
mω
This change replaces the characteristic equation (6.29) with
κ 2mω μq 2n
−iω = , i.e. κ 2 = , (6.43)
iq 2nμ m
i.e. replaces the skin effect with the field penetration by the following frequency-
independent depth:
1 ⎛ m ⎞
1/2
δ≡ =⎜ 2 ⎟ . (6.44)
κ ⎝ μq n ⎠
Superficially, this means that the field decay into the superconductor does not
depend on frequency:
H (x , t ) = H (0, t )e−x/δ , (6.45)
thus explaining the Meissner–Ochsenfeld effect.
However, there are two problems with this result. First, for parameters typical for
good metals (q = −e, n ∼ 1029 m−3, m ∼ me, μ ≈ μ0), Eq. (6.44) gives δ ∼ 10−8 m, a
factor of ∼10 to ∼102 lower than the typical experimental values of δL. Experiments
also show that the penetration depth diverges at T → Tc, which is not predicted by
Eq. (6.44). Another, much more fundamental problem with Eq. (6.44) is that it has
been derived for ωτ ≫ 1. Even if we assume that somehow there are no collisions at
6-13
all, i.e. τ = ∞, at ω → 0 both parts of the characteristic equation (6.43) vanish, and
we cannot make any conclusion about κ. This is not just a mathematical artifact we
can ignore. For example, let us place a non-magnetic metal into a static external
magnetic field at T > Tc. The field would completely penetrate into the sample. Now
let us cool it. As soon as the temperature is decreased below Tc, the above
calculations would become valid, forbidding penetration into the superconductor
of any change of the field, so that the initial field would be ‘frozen’ inside the sample.
Experiments shows something completely different: as T is lowered below Tc, the
initial field is pushed out of the sample.
The resolution of these contradictions has been provided by quantum mechanics.
As was explained in 1957 in a seminal work by J Bardeen, L Cooper, and J Schrieffer
(commonly referred to as the BSC theory), superconductivity is due to the correlated
motion of electron pairs, with opposite spins and nearly opposite momenta. Such
Cooper pairs, each with the electric charge q = −2e and zero spin, may form only in a
narrow energy layer near the Fermi surface, of certain thickness Δ(T ). Parameter
Δ(T ), which may also be considered as the binding energy of the pair, tends to zero
at T → Tc, while at T ≪ Tc it has a virtually constant value Δ(0) ≈ 3.5 kBTc, of the
order of a few meV for most superconductors. This fact readily explains the
relatively low spatial density of the Cooper pairs: np ∼ nΔ(T )/εF ∼ 1026 m−3.
With the correction n → np, our Eq. (6.38) for the penetration depth becomes
⎛ m ⎞1/2
δ → δL = ⎜ 2 ⎟ . (6.46)
⎝ μq n p ⎠
This expression diverges at T → Tc and generally fits the experimental data

reasonably well, at least for so-called ‘clean’ superconductors (with the mean free
path l = vFτ, where vF ∼ (2mεF)1/2 is the r.m.s. velocity of electrons on the Fermi
surface, much longer that the Cooper pair size ξ—see below).
The smallness of the coupling energy Δ(T ) is also a key factor in the explanation of
the Meissner–Ochsenfeld effect, as well as several macroscopic quantum phenomena in
superconductors. Because of Heisenberg’s quantum uncertainty relation δrδp ∼ ℏ, the
spatial extension of the Cooper pair’s wavefunction (the so-called coherence
length of the superconductor) is relatively large: ξ ∼ δr ∼ ℏ/δp ∼ ℏvF/Δ(T ) ∼ 10−6 m.
As a result, npξ3 ≫ 1, meaning that the wavefunctions of the pairs are strongly
overlapped in space. Now, due to their integer spin, Cooper pairs behave like
bosons, which means in particular that at low temperature they exhibit the so-
called Bose–Einstein condensation onto the same ground energy level εg.27 This
means that the frequency ω = εg/ℏ of the time evolution of each pair’s wavefunction
27
A qualitative discussion of the Bose–Einstein condensation of bosons may be found in Part SM section 3.4,
although the full theory of superconductivity is more complex, because it describes the condensation taking
place simultaneously with the formation of effective bosons (Cooper pairs) from fermions (single electrons).
For a more detailed, but very readable coverage of physics of superconductors, I would refer the reader to the
monograph [2].
6-14
Ψ = ψ exp{−iωt} is the same, i.e. that the phases φ of the wavefunctions, defined by
the relation
ψ= ψ e iφ , (6.47)
become equal, so that the electric current is carried not by individual Cooper pairs
but rather their Bose–Einstein condensate described by a single wavefunction. Due to
this coherence, the quantum effects (which are, in the usual Fermi-liquids of single
electrons, masked by the statistical spread of their energies and hence of their phases)
become very explicit—‘macroscopic’.
To illustrate this, let us write the well-known quantum-mechanical formula for
the probability current density of a free, non-relativistic particle28,
iℏ 1
jw = (ψ ∇ψ ⁎ − c.c.) ≡ [ψ ⁎( −i ℏ∇)ψ − c.c.], (6.48)
2m 2m
where c.c. means the complex conjugate of the previous expression. Now let me
borrow one result that will be proved later in this course (in section 9.7) when we
discuss the analytical mechanics of a charged particle moving in an electromagnetic
field. Namely, in order to account for the magnetic field effects, the particle’s kinetic
momentum p ≡ mv (where v ≡ dr/dt is particle’s velocity) has to be distinguished from
its canonical momentum29,
P ≡ p + q A. (6.49)
where A is the vector-potential of the field, defined by Eq. (5.27). In contrast with the
Cartesian components pj = mvj of the momentum p, the canonical momentum’s
components are the generalized momenta corresponding to the Cartesian compo-
nents rj of the radius-vector r, considered as generalized coordinates of the particle:
Pj = ∂L /∂vj , where L is the particle’s Lagrangian function. According to the general
rules of transfer from classical to quantum mechanics30, it is the vector P whose
operator (in the Schrödinger picture) equals −iℏ∇, so that the operator of the kinetic
momentum p = P − qA is equal to −iℏ∇ − qA. Hence, the in order to account for the
magnetic field31 effects, we should make the following replacement,
−i ℏ∇ → −i ℏ∇ − qA, (6.50)
in all field-free quantum-mechanical relations. In particular, Eq. (6.48) has to be
generalized as
28
See, e.g. Part QM section 1.4, in particular Eq. (1.47).
29
I am sorry to use traditional notations p and P for the momenta—the same symbols which were used for the
electric dipole moment and polarization in chapter 3. I hope there will be no confusion, because the latter
notions are not used in this section.
30
See, e.g. Part CM section 10.1, in particular Eq. (10.26).
31
The account of the electric field is easier, because the related energy qϕ of the particle may be directly
included into the potential energy operator—not participating in our current discussion.
6-15
1
jw = [ψ ⁎( −i ℏ∇ − q A)ψ − c.c.]. (6.51)
2m
This expression becomes more transparent if we take the wavefunction in the form
(6.47):
ℏ ⎛
2 ⎜ q ⎞⎟
jw = ψ ∇φ − A . (6.52)
m ⎝ ℏ ⎠
This relation means, in particular, that in order to keep jw gauge-invariant, the
transformation (6.8)–(6.9) has to be accompanied by a simultaneous transformation
of the wavefunction’s phase:
q
φ → φ + χ. (6.53)
ℏ
It is fascinating that the quantum-mechanical wavefunction (or more exactly, its
phase) is not gauge-invariant, meaning that you may change it in your mind—at
will! Again, this does not change any observable (such as jw or the probability
density ψψ*), i.e. any experimental results.
Now for the electric current density of the whole superconducting condensate,
Eq. (6.52) yields the following constitutive relation:
ℏqn p 2 ⎜
⎛ q ⎞⎟
j ≡ j w qn p = ψ ∇φ − A , (6.54)
m ⎝ ℏ ⎠
This equation shows that this supercurrent may be induced by the dc magnetic field
alone and does not require any electric field. Indeed, for the simplest, 1D geometry
shown in figure 6.2a, j(r) = j(x)nz, A(r) = A(x) nz, and ∂/∂z = 0, so that the Coulomb
gauge condition Eq. (5.48) is satisfied for any choice of the gauge function χ(x) and,
for the sake of simplicity, we can choose it to provide φ(r) ≡ const32, so that
q 2n p 1
j=− A≡− A. (6.55)
m μδ L2
where δL is given by Eq. (6.46), and the field is assumed to be small and hence not
affecting the probability ∣ψ∣2 (normalized to 1 in the absence of the field). This is the
so-called London equation, proposed (in a different form) by brothers F and H
London in 1935 to explain the Meissner–Ochsenfeld effect. Combining it with
Eq. (5.44), generalized for a linear magnetic medium by the replacement μ0 → μ, we
obtain
1
∇2 A = A, (6.56)
δ L2
32
This is the so-called London gauge; for our simple geometry it is also the Coulomb gauge (5.48).
6-16
This simple differential equation, similar to Eq. (6.23), for our simple geometry has
an exponential solution similar to Eq. (6.32):
⎧ x⎫ ⎧ x⎫
A(x ) = A(0)exp ⎨ − ⎬ , B (x ) = B (0)exp ⎨ − ⎬ ,
⎩ δL ⎭ ⎩ δL ⎭
(6.57)
⎧ x⎫
j (x ) = j (0)exp ⎨ − ⎬ ,
⎩ δL ⎭
which shows that the magnetic field and supercurrent penetrate into a super-
conductor only by the London penetration depth δL, regardless of frequency33. By
the way, integrating the last result through the penetration layer and using the
vector-potential’s definition B = ∇ × A (for our geometry, giving B(x) = dA(x)/dx =
−δLA(x)) we may readily check that the linear density J of the surface supercurrent
still satisfies the universal coarse-grain relation (6.38).
This universality should bring to our attention the following common feature of
the skin effect (in ‘normal’ conductors) and the Meissner–Ochsenfeld effect (in
superconductors): if the linear size of a bulk sample is much larger than, respectively,
δs or δL, then B = 0 in the dominating part of its interior. According to Eq. (5.110), a
formal description of such conductors (valid only on a coarse scale much larger than
either δs or δL) may be achieved by formally treating the sample a as an ideal
diamagnet, with μ = 0. In particular, we can use this description and Eq. (5.124) to
immediately obtain the magnetic field’s distribution outside a bulk sphere:
⎛ R3 ⎞
B = μ0H = −μ0∇ϕm , with ϕm = H0⎜ −r − 2 ⎟ cos θ , for r ⩾ R . (6.58)
⎝ 2r ⎠
Figure 6.3 shows the corresponding surfaces of equal potential ϕm. It is evident
that the magnetic field lines (which are normal to the equipotential surfaces) bend to
become parallel to the surface near it. This pattern also helps to answer the question
that might arise when making the assumption (6.24): what happens to bulk conductors
placed into in a normal ac magnetic field (and to superconductors in a dc magnetic
field)? The answer is: the field is deformed outside the conductor to sustain the ‘coarse-
grain’ boundary condition34
Bn surface = 0, (6.59)
which follows from Eq. (5.118) and the requirement B∣inside = 0.
33
Since not all electrons in a superconductor form Cooper pairs, at any frequency ω ≠ 0 the unpaired electrons
provide energy-dissipating Ohmic currents, which are not described by Eq. (6.54). These losses become very
substantial when frequency ω becomes so high that the skin-effect length δs of the material (as measured with
superconductivity suppressed, say by a high magnetic field) becomes less than δL. For typical metallic
superconductors this crossover takes place at frequencies of a few hundred GHz, so that even for microwaves
Eq. (6.57) still gives a fairly good description of the field penetration.
34
Sometimes this boundary condition, as well as the (compatible) Eq. (6.38), is called ‘macroscopic’. However,
this term may lead to confusion with the boundary conditions (5.117) and (5.118), which also ignore the
atomic-scale microstructure of the ‘effective currents’ jef = ∇ × M, but (as was shown in this section) still allow
an explicit, detailed account of the skin currents (6.34) or supercurrents (6.55).
6-17
Figure 6.3. Equipotential surfaces ϕm = const around a (super)conducting sphere of radius R ≫ δs (or δL),
placed into a uniform magnetic field, within the coarse-grain model μ = 0.
This answer should be taken with a grain of salt. First, for normal conductors it is
only valid at sufficiently high frequencies, so that the skin depth (6.33) is sufficiently
small: δs ≪ a, where a is the scale of the conductor’s linear size—for a sphere, a ∼ R.
In superconductors, this simple picture is valid not only if δs ≪ a, but also only in
sufficiently low magnetic fields, because strong fields do penetrate into super-
conductors, destroying superconductivity (completely or partly) and thus disrupting
the Meissner–Ochsenfeld effect—see the next section.
6.5 Electrodynamics of macroscopic quantum phenomena35

Despite this superficial similarity of the skin effect and the Meissner–Ochsenfeld
effect, the electrodynamics of superconductors is much richer. For example, let us
use Eq. (6.54) to describe the fascinating effect of magnetic flux quantization.
Consider a closed ring/loop (of any form) made of a superconducting ‘wire’ with
a cross-section much larger than δL2 (figure 6.4a).
Figure 6.4. (a) A closed, flux-quantizing superconducting ring, (b) a ring with a narrow slit, and (c) a
superconducting quantum interference device (SQUID).
35
The material of this section is not covered in most E&M textbooks, and will not be used in later sections of
this course. Thus the ‘only’ loss to the reader from skipping this section would be a lack of familiarity with one
of the most fascinating fields of physics. Note also that we already have virtually all the tools necessary for its
discussion, so that reading this section should not require much effort.
6-18
From the last section’s discussion, we know that deep inside the wire the
supercurrent is exponentially small. Integrating Eq. (6.54) along any closed contour
C that does not approach the surface closer than a few δL at any point (see the
dashed line in figure 6.4), so that j = 0 at all its points, we obtain
∮C ∇φ ⋅ dr − ℏq ∮C A ⋅ d r = 0. (6.60)
The first integral, i.e. the difference of φ in the initial and final points, has to be equal
to either zero or an integer number of 2π, because the change φ → φ + 2πn does not
change the Cooper pair condensate’s wavefunction:
ψ ′ = ψ e i (φ+2πn) = ψ e iφ = ψ . (6.61)
On the other hand, according to Eq. (5.65), the second integral in Eq. (6.60) is just
the magnetic flux Φ through the contour36. As a result, we obtain a wonderful result:
2π ℏ
Φ = n Φ0 , where Φ0 ≡ , with n = 0, ±1, ±2, …, (6.62)
q
saying that the magnetic flux inside any superconducting loop can only take values
multiple of the flux quantum Φ0. This effect, predicted in 1950 by the same Fritz
London (who expected q to be equal to the electron charge −e), was confirmed
experimentally in 196137, but with ∣q∣ = 2e (so that Φ0 ≈ 2.07 × 10−15 Wb).
Historically, this observation gave decisive support to the BSC theory of super-
conductivity based on Cooper pairs, with charge q = −2e, which had been put
forward just 4 years earlier.
Note the truly macroscopic character of this quantum effect: it has been
repeatedly observed in human-scale superconducting loops and, from what is
known about superconductors, there is no doubt that if we made a giant super-
conducting wire loop extending, say, over the Earth’s diameter, the magnetic flux
through it would still be quantized—although with a very large flux quanta number
n. This means that the coherence of the Bose–Einstein condensates may extend over
(using H Casimir’s famous expression) ‘miles of dirty lead wire’. (Lead is a typical
superconductor, with Tc ≈ 7.2 K, and indeed retains its superconductivity even when
highly contaminated by impurities.)
Moreover, hollow rings are not entirely necessary for flux quantization. In 1957,
A Abrikosov explained the counter-intuitive high-field behavior of superconduc-
tors with δL > ξ√2, known experimentally as their mixed (or ‘Shubnikov’) phase
since the 1930s. He showed that a sufficiently high magnetic field may penetrate
into such superconductors in the form of self-formed magnetic field ‘threads’ (or
‘tubes’) surrounded by vortex-shaped supercurrents—the so-called Abrikosov
vortices. In the simplest case, the core of such a vortex is a straight line, on which
36
Due to the Meissner–Ochsenfeld effect, the exact path of the contour is not important, and we may discuss Φ
just as the magnetic flux through the ring.
37
Independently and virtually simultaneously by two groups: B Deaver and W Fairbank, and R Doll and M
Näbauer; their reports were published back-to-back in Physical Review Letters.
6-19
Figure 6.5. The Abrikosov vortex: (a) a 3D structure sketch and (b) the main variables as functions of the
distance ρ from the axis (schematically).
the superconductivity is completely suppressed (∣ψ∣ = 0), surrounded by circular,

axially symmetric, persistent supercurrents j(ρ), where ρ is the distance from the
vortex axis—see figure 6.5a. At the axis the current vanishes, and with distance it
first rises and then falls, so that j(∞) = 0, reaching its maximum at ρ ∼ ξ, while the
magnetic field B(ρ), directed along the vortex axis, drops monotonically at
distances ∼δL (figure 6.5b).
The total flux of the field equals exactly one flux quantum Φ0, given by Eq. (6.62).
Correspondingly, the wavefunction’s phase φ performs just one ±2π revolution
along any contour drawn around the vortex’s axis, so that ∇φ = ±nφ/ρ, where nφ is
the azimuthal unit vector38. This topological feature of the wavefunction’s phase is
sometimes called the fluxoid quantization—in order to distinguish it from the flux
quantization, which is valid only for relatively large contours not approaching the
axis by distances ∼δL.
A quantitative analysis of the Abrikosov vortex requires, in addition to the equations
we have discussed, one more constituent equation that would describe the changes of
the number of Cooper pairs (quantified by ∣ψ∣2) by the magnetic field—or rather by the
field-induced supercurrent. In his original work, Abrikosov used for this purpose the
famous Ginzburg–Landau equation39, which is qualitatively valid only at T ≈ Tc.
The equation may be conveniently represented using either of the following forms,
1
( −i ℏ∇ − q A)2 ψ = aψ − bψ ψ 2 ,
2m
(6.63)
⎛ q ⎞2
ξ 2ψ *⎜∇ − i A⎟ ψ = (1 − ψ 2 ) ψ 2 ,
⎝ ℏ ⎠
where a and b are certain temperature-dependent coefficients, with a → 0 at T → Tc.
The first of these forms clearly shows that the Ginzburg–Landau equation (together
with the similar Gross–Pitaevskii equation describing uncharged Bose–Einstein
condensates) belongs to a broader class of nonlinear Schrödinger equations, differing
38
The last (perhaps, evident) expression formally follows from Eq. (A.60) with f = ±φ + const.
39
This equation was derived by V Ginzburg and L Landau from phenomenological arguments in 1950, i.e.
before the advent of the ‘microscopic’ BSC theory, and may be used for simple analyses of a broad range of
nonlinear effects in superconductors. The Ginzburg–Landau and Gross–Pitaevskii equations will be further
discussed in Part SM section 4.3.
6-20
only by the additional nonlinear term from the usual Schrödinger equation, which is
linear in ψ. The equivalent, second form of Eq. (6.63) is more convenient for
applications, and shows more clearly that if the superconductor’s condensate
density, proportional to ∣ψ∣2, is suppressed only locally, it restores to its unperturbed
value (with ∣ψ∣2 = 1) at distances of the order of the coherence length ξ ≡ ℏ/(2ma)1/2.
This fact enables a simple quantitative analysis of the Abrikosov vortex in the
limit ξ ≪ δL. Indeed, in this case (see figure 6.5) ∣ψ∣ 2 = 1 at most distances (ρ ∼ δL)
where the field and current are distributed, so that these distributions may be readily
calculated without any further involvement of Eq. (6.63), just from Eq. (6.54) with
∇φ = ±nφ/ρ, and the Maxwell equations (6.21) for the magnetic field, giving ∇ × B =
μj, and ∇ · B = 0. Indeed, combining these equations just as this was done at the
derivation of Eq. (6.23), for the only Cartesian component of the vector B(r) = B(ρ)nz
(where the axis z is directed along the vortex’ axis), we obtain a simple equation
ℏ
δ L2∇2 B − B = − ∇ × (∇ × φ) ≡ ∓Φ0δ 2(ρ), at ρ ≫ ξ, (6.64)
q
which coincides with Eq. (6.56) at all regular points ρ ≠ 0. Spelling out the Laplace
operator for our current case of axial symmetry40, we obtain an ordinary differential
equation,
1 d ⎛ dB ⎞
δ L2 ⎜ρ ⎟ − B = 0, for ρ ≠ 0. (6.65)
ρ dρ ⎝ dρ ⎠
Comparing this equation with Eq. (2.155) with ν = 0 and taking into account that we
need the solution decreasing at ρ → ∞, making any contribution from the function I0
unacceptable, we obtain
⎛ ρ⎞
B = CK 0⎜ ⎟ (6.66)
⎝ δL ⎠
—see the plot of this function (black line) in the right-hand panel of figure 2.22. The
constant C should be calculated from fitting the 2D delta-function on the right-hand
side of Eq. (6.64), i.e. by requiring
∞ ∞
∫vortex B(ρ)d 2ρ ≡ 2π∫0 B(ρ)ρ dρ ≡ 2πδ L2C ∫0 K 0(ζ )ζ dζ = ∓Φ0 . (6.67)
The last, dimensionless integral equals 1,41 so that finally

Φ0 ⎛ ρ⎞
2 0⎜ ⎟,
B (ρ ) = K at ρ ≫ ξ. (6.68)
2πδ L ⎝ δ L ⎠
40
See, e.g. Eq. (A.61) with ∂/∂φ = ∂/∂z = 0.
41
This equality follows, for example, from the integration of both sides of Eq. (2.143) (which is valid for any
Bessel functions, including Kn) with n = 1, from 0 to ∞, and then using the asymptotic values given by Eqs.
(2.157) and (2.158): K1(∞) = 0, and K1(ζ) → 1/ζ at ζ→ 0.
6-21
The function K0 (the modified Bessel function of the second kind), drops
exponentially as its argument becomes larger than 1 (i.e. in our problem, at
distances ρ much larger than δL), and diverges as its argument tends to zero—see,
e.g. the second of Eqs. (2.157). However, this divergence is very slow (logarithmic)
and, as was repeatedly discussed in this series, is avoided by the account of virtually
any other factor. In our current case, this factor is the decrease of ∣ψ∣2 to zero at ρ ∼ ξ
(see figure 6.5), not taken into account in Eq. (6.68). As a result, we may estimate the
field on the axis of the vortex as
Φ0 δ
B(0) ≈ ln L ; (6.69)
2πδ L2 ξ
the exact (much more involved) solution of the problem confirms this estimate with a
minor correction: ln(δL/ξ) → [ln(δL/ξ) − 0.28].
The current density distribution may be now calculated using the Maxwell
equation ∇ × B = μj, giving j = j(ρ)nφ, with42
1 ∂B Φ0 ∂ ⎛ ρ⎞ Φ0 ⎛ ρ⎞
j (ρ ) = − =− K 0⎜ ⎟ ≡ 3 1⎜
K ⎟, at ρ ≫ ξ, (6.70)
μ ∂ρ 2πμδ L ∂ρ ⎝ δ L ⎠ 2πμδ L ⎝ δ L ⎠
2
where the same identity Eq. (2.158), with Jn → Kn and n = 1, was used. Now looking
at Eqs. (2.157) and (2.158) with n = 1, we see that the supercurrent is exponentially
small at ρ ≫ δL (thus outlining the vortex’s periphery), and is proportional to 1/ρ
within the broad range ξ ≪ ρ ≪ δL. This rise of the current at ρ → 0 (which could be
readily predicted directly from Eq. (6.54) with ∇φ = ±nφ/ρ, and the A-term negligible
at ρ ≪ δL) is quenched at ρ ∼ ξ by a rapid drop of the factor ∣ψ∣2 in the same
Eq. (6.54), i.e. by the suppression of the superconductivity near the axis (by the same
supercurrent!)—see figure 6.5 again.
This vortex structure may be used to calculate, in a straightforward way, its
energy per unit length (i.e. its linear tension)
U Φ 20 δ
T≡ ≈ ln L , (6.71)
l 4πμδ L2 ξ
and hence the ‘first critical’ value Hc1 of the external magnetic field43, at which the
vortex formation becomes possible (in a long cylindrical sample parallel to the field):
T Φ0 δ
Hc1 = ≈ 2
ln L . (6.72)
Φ0 4πμδ L ξ
Let me leave the proof of these two formulas for the reader’s exercise.
42
See, e.g. Eq. (A.63), with fρ = fφ = 0, and fz = B(ρ).
43
This term is used to distinguish Hc1 from the higher ‘second critical field’ Hc2, at which the Abrikosov
vortices are pressed to each other so tightly (to distances d ∼ ξ) that they merge, and the remains of
superconductivity vanish: ψ → 0. Unfortunately, I do not have time/space to discuss these effects; the interested
reader is referred to chapter 5 of [2].
6-22
The flux quantization and the Abrikosov vortices discussed above are just two of
several macroscopic quantum effects in superconductivity. Let me discuss just one more,
but perhaps the most interesting of such effects. Let us consider a superconducting ring/
loop interrupted with a very narrow slit (figure 6.4b). Integrating Eq. (6.54) along any
current-free path from point 1 to point 2 (see, e.g. dashed line in figure 6.4b), we obtain
2 ⎛ q ⎞⎟ q
0= ∫1 ⎜∇φ −
⎝ ℏ
A ⋅ d r = φ2 − φ1 − Φ .
⎠ ℏ
(6.73)
Using the flux quantum definition (6.62), this result may be rewritten as
2π
φ ≡ φ1 − φ2 = Φ, (6.74)
Φ0
where φ is called the Josephson phase difference. Note that in contrast to each of the
phases φ1,2, their difference φ is gauge-invariant, because it is directly related to the
gauge-invariant magnetic flux Φ.
Can this φ be measured? Yes, using the Josephson effect44. Let us consider two (for
the argument simplicity, similar) superconductors, connected with some sort of weak
link, for example a tunnel barrier, or a point contact, or a narrow thin-film bridge,
through that a weak Cooper pair supercurrent can flow. (Such system of two weakly
coupled superconductors is called a Josephson junction.) Let us think what this
supercurrent I may be a function of. For that, reverse thinking is helpful: let us
imagine we change the current; what parameter of the superconducting condensate
can it affect? If the current is weak, it cannot perturb the superconducting condensate’s
density, proportional to ∣ψ∣2; hence it may only change the Cooper condensate phases
φ1,2. However, according to Eq. (6.53), the phases are not gauge-invariant, while the
current should be. Hence the current may affect (or, if you like, may be a function of)
only the phase difference φ defined by Eq. (6.74). Moreover, just has already been
argued during the flux quantization discussion, a change of any of φ1,2 (and hence of
φ) by 2π or any of its multiples should not change the current. In addition, if the
wavefunction is the same in both superconductors (φ = 0), the supercurrent should
vanish due to the system’s symmetry. Hence the function I(φ) should satisfy conditions
I (0) = 0, I (φ + 2π ) = I (π ). (6.75)
With these conditions on hand, we should not be terribly surprised by the following
Josephson’s result that for the weak link provided by tunneling45,
I (φ) = Ic sin φ , (6.76)
where constant Ic, which depends on the weak link’s strength and temperature, is
called the critical current. Actually, Eqs. (6.54) and (6.63) enable not only a
straightforward calculation of this relation, but even obtaining a simple expression
44
It was predicted in 1961 by B Josephson (then a PhD student!), and observed experimentally by several
groups soon after that.
45
For some other types of weak links, the function I(φ) may deviate from the sine form Eq. (6.76) rather
considerably, still satisfying the general requirements (6.75).
6-23
of the critical current Ic via the link’s normal-sate resistance—a task left for the
(creative:-) reader’s exercise.
Now let us see what happens if a Josephson junction is placed into the gap in a
superconductor loop—see figure 6.4c. In this case, we may combine Eqs. (6.74) and
(6.76), obtaining
Φ
I = Ic sin 2π . (6.77)
Φ0
This effect of a periodic dependence of the current on the magnetic flux is called the
macroscopic quantum interference46, while the system shown in figure 6.4c is a
superconducting quantum interference device, abbreviated as SQUID (with all letters
capital, please). The low value of the magnetic flux quantum Φ0, and hence the high
sensitivity of φ to external magnetic fields, allows the use of SQUIDs as ultra-
sensitive magnetometers. Indeed, for a superconducting ring of area ∼1 cm2, one
period of the change of the supercurrent (6.77) is produced by magnetic field change
of the order of 10−11 T (10−7 Gs), while sensitive electronics allow us to measure a
tiny fraction of this period—limited by thermal noise at a level of the order of a few
fT. Such sensitivity allows measurements, for example, of the magnetic fields
induced by the beating human heart, and even by brain activity, outside the body47.
An important aspect of quantum interference is the so-called Aharonov–Bohm (AB)
effect (which actually takes place for single quantum particles as well)48. Let the
magnetic field lines be limited to the central part of the SQUID ring, so that no
appreciable magnetic field ever touches the superconducting ring material. (This may
be done experimentally with very good accuracy, for example using high-μ magnetic
cores—see their discussion in section 5.6.) As predicted by Eq. (6.77), and confirmed
by several careful experiments carried out in the mid-1960s49, this restriction does not
matter—the interference is observed anyway. This means that not only the magnetic
field B, but also the vector-potential A represents physical reality, albeit in a quite
peculiar way—remember the gauge transformation (5.46), which you may carry out
off the top of your head, without changing any physical reality? (Fortunately, this
transformation does not change the contour integral participating in Eq. (5.65), and
hence the magnetic flux Φ, and hence the interference pattern.)
Actually, the magnetic flux quantization (6.62) and the macroscopic quantum
interference (6.77) are not completely different effects, but just two manifestations of
the inter-related macroscopic quantum phenomena. In order to show that, one
46
The name is due to the deep analogy between this phenomenon and the interference between two coherent
waves, to be discussed in detail in section 8.4.
47
Other practical uses of SQUIDs include MRI signal detectors, high-sensitive measurements of magnetic
properties of materials, and weak field detection in a broad variety of physical experiments—see, e.g. [3]. For a
comparison of these devices with other sensitive magnetometers see, e.g. the review collection [4].
48
For a more detailed discussion of the AB effect see, e.g. Part QM section 3.2.
49
Similar experiments have been carried out with single (unpaired) electrons—moving either ballistically, in a
vacuum, or in ‘normal’ (non-superconducting) conducting rings. In the last case, the effect is much harder to
observe that in SQUIDs, because the ring size has to be very small and temperature very low to avoid the so-
called dephasing effects due to unavoidable interactions of the electrons with their environment—see, e.g. Part
QM chapter 7.
6-24
should note that if the critical current Ic (or rather its product by the loop’s self-
inductance L) is high enough, the flux Φ in the SQUID loop is due not only to the
external magnetic field flux Φext, but also has a self-field component—cf Eq. (5.68)50:
Φ = Φext − LI , where Φext ≡ ∫S (Bext)nd 2r. (6.78)
Now the relation between Φ and Φext may be readily found by solving this equation
together with Eq. (6.77). Figure 6.6 shows this relation for several values of the
dimensionless parameter λ ≡ 2πLIc/Φ0.
Figure 6.6. The function Φ(Φext) for SQUIDs with various values of the normalized LIc product. Dashed
arrows show the flux leaps as the external field is changed. (The branches with dΦ/dΦext < 0 are unstable.)
These plots show that if the critical current (or the inductance) is low, λ ≪ 1, the
self-field effects are negligible, and the total flux follows the external field (i.e. Φext)
faithfully. However, at λ > 1, the function Φ(Φext) becomes hysteretic and at λ ≫ 1
the stable (positive-slope) branches of this function are nearly flat, with the total flux
values corresponding to Eq. (6.62). Thus, a superconducting ring closed by a high-Ic
Josephson junction exhibits nearly perfect flux quantization.
The self-field effects described by Eq. (6.78) create certain technical problems for
SQUID magnetometry, but they are the basis for one more application of these
devices: ultrafast computing. Indeed, figure 6.6 shows that at the values of λ
modestly above 1 (e.g. λ ≈ 3), and within a certain range of applied field, the
SQUID has two stable flux states that differ by ΔΦ ≈ Φ0 and may be used for coding
binary 0 and 1. For practical superconductors (such as Nb), the time of switching
50
The sign before LI would be positive, as in Eq. (5.70), if I was the current flowing into the inductance.
However, in order to keep the sign in Eq. (6.76) intact, I should mean the current flowing into the Josephson
junction, i.e. from the inductance, thus changing the sign of the LI term in Eq. (6.78).
6-25
between these states (see the dashed arrows in figure 6.4) are of the order of a
picosecond, while the energy dissipated at such an event may be as low as ∼10−19 J.
(This bound is determined not by the device’s physics, by the fundamental require-
ment for the energy barrier between the two states to be much higher than the
thermal fluctuation energy scale kBT, ensuring a sufficiently long information
retention time.) While the picosecond switching speed may also be achieved with
some semiconductor devices, the power consumption of the SQUID-based digital
devices may be 5–6 orders of magnitude lower, enabling VLSI integrated circuits
with 100 GHz scale clock frequencies. Unfortunately, the range of practical
application of these rapid single-flux-quantum (RSFQ) digital circuits is still very
narrow, due to the inconvenience of their deep refrigeration to temperatures below
Tc.51
Since we have already obtained the basic relations (6.74) and (6.76) describing
macroscopic quantum phenomena in superconductivity, let me mention in brief two
other members of this group, called the dc and ac Josephson effects. Differentiating
Eq. (6.74) over time, and using the Faraday induction law (6.2), we obtain52
dφ 2e
= V. (6.79)
dt ℏ
This famous Josephson phase-to-voltage relation should be valid regardless of the
way how the voltage V has been created53, so let us apply Eqs. (6.76) and (6.79) to
the simplest circuit with a non-superconducting source of dc voltage—see figure 6.7.
If the current’s magnitude is below the critical value, Eq. (6.76) allows the phase φ
to have a time-independent value
I
φ = sin−1 , if − Ic < I < +Ic, (6.80)
Ic
and hence, according to Eq. (6.79), a vanishing voltage drop across the junction:
V = 0. This dc Josephson effect is not quite surprising—indeed, we have postulated
from the very beginning that the Josephson junction may pass a certain
Figure 6.7. A dc-voltage-biased Josephson junction.
51
For more on that technology see, e.g. the review paper [5] and references therein.
52
Since the induced e.m.f. V ind cannot drop on the superconducting path between the Josephson junction
electrodes 1 and 2 (see figure 6.4c), it should be equal to (−V), where V is the voltage across the junction.
53
Indeed, it may also be obtained from simple Schrödinger equation arguments—see, e.g. Part QM
section 1.6.
6-26
supercurrent. Much more fascinating is the so-called ac Josephson effect, which

occurs if the voltage across the junction has a non-vanishing average (dc) component
V0. For simplicity, let us assume that this is the only voltage component: V(t) = V0 =
const54, then Eq. (6.79) may be easily integrated to give φ = ωJt + φ0, where
2e
ωJ ≡ V0. (6.81)
ℏ
This result, plugged into Eq. (6.76), shows that the supercurrent oscillates,
I = Ic sin (ωJt + φ0), (6.82)
with the so-called Josephson frequency ωJ (6.81), proportional to the applied dc
voltage. For practicable voltages (above the typical noise level), the cyclic frequency
fJ = ωJ/2π corresponds to the GHz or even THz ranges, because the proportionality
coefficient in Eq. (6.81) is very high: fJ/V0 = e/πℏ ≈ 483 MHz μV−1.55
An important experimental fact is the universality of this coefficient. For
example, in the mid-1980s, a Stony Brook group led by J Lukens proved that this
factor is material-independent with a relative accuracy of at least 10−15. Very few
experiments, particularly in solid state physics, have ever reached such precision.
This fundamental nature of the Josephson voltage-to-frequency relation (6.81)
allows an important application of the ac Josephson effect in metrology. Namely,
phase locking56 the Josephson oscillations with an external microwave signal from
atomic frequency standards, one can obtain a more precise dc voltage than from any
other source. In NIST and other metrological institutions around the globe, this
effect is used for the calibration of simpler ‘secondary’ voltage standards that can
operate at room temperature.
6.6 Inductors, transformers, and ac Kirchhoff laws

Let a wire coil (meaning either a single loop as illustrated in figure 5.4b, or a series of
such loops, such as one of the solenoids shown in figure 5.6) have a self-inductance L
much larger than that of the wires connecting it to other components of our system:
ac voltage sources, voltmeters, etc. (Since, according to Eq. (5.75), L scales as the
number N of wire turns squared, this is condition is easier to satisfy at N ≫ 1.) Then
in a quasi-static system consisting of such lumped induction coils and external wires
(and other circuit elements such as resistors, capacitances, etc), we may neglect the
electromagnetic induction effects everywhere outside the coil, so that the electric
field in those external regions is potential. Then the voltage V between coil’s
54
In experiments, this condition is hard to implement, due to relatively high inductance of the current leads
providing the dc voltage supply. However, these complications do not change the main conclusion of the
analysis.
55
This 1962 prediction by B Josephson was confirmed experimentally—first implicitly (by phase locking of the
oscillations with an external oscillator) in 1963, and then explicitly (by the detection of microwave radiation) in
1967.
56
For a discussion of this very important (and general) effect, see, e.g. Part CM section 5.4.
6-27
terminals may be defined (as in electrostatics) as the difference of values of scalar

potential ϕ between the terminals, i.e. as the integral
V= ∫ E ⋅ dr (6.83)
between the coil terminals along any path outside the coil. This voltage has to be
balanced by the induction e.m.f. (6.2) in the coil, so that if the Ohmic resistance of
the coil is negligible57, we may write
dΦ
V= , (6.84)
dt
where Φ is the magnetic flux in the coil. If the flux is due to the current I in the same
coil only (i.e. if it is magnetically uncoupled from other coils), we may use Eq. (5.70)
to obtain the well-known relation
dI
V=L , (6.85)
dt
where the compliance with the Lenz sign rule is achieved by selecting the relations
between the assumed voltage polarity and the current direction as shown in figure
6.8a.
If similar conditions are satisfied for two magnetically coupled coils (figure 6.8b),
then, in Eq. (6.84), we need to use Eqs. (5.69) instead, obtaining
dI1 dI dI2 dI
V1 = L1 + M 2, V2 = L 2 + M 1. (6.86)
dt dt dt dt
Such systems of inductively coupled coils have numerous applications in electrical
engineering and physical experiments. Perhaps the most important of them is the ac
transformer, in which the coils share a common soft-ferromagnetic core with the
toroidal (‘doughnut’) topology—see figure 6.8c58. As we already know from the
discussion in section 5.6, such cores with μ ≫ μ0 ‘try’ to suck in all magnetic field
Figure 6.8. Some lumped ac circuit elements: (a) an induction coil, (b) two inductively coupled coils, and (c) an
ac transformer.
57
If the resistance is substantial, it may be represented by a separate lumped circuit element (resistor)
connected in series with the coil.
58
The first practically acceptable form of this device, called the Stanley transformer, in which multi-turn
windings could be easily mounted onto a toroidal ferromagnetic (then a silicon–steel plate) core, was invented
in 1886.
6-28
lines, so that the magnetic flux Φ(t) in the core is nearly the same in each of its cross-
sections. With this, Eq. (6.84) yields
dΦ dΦ
V1 ≈ N1 , V2 ≈ N2 , (6.87)
dt dt
where the voltage ratio is completely determined by the ratio N1/N2 of the number of
wire turns.
Now we may generalize to the ac current case the Kirchhoff laws already
discussed in chapter 4—see figure 4.3, reproduced in figure 6.9a. Let not only wire
inductances but also the wire capacitances and resistances be negligible in compar-
ison to those of the lumped (compact) circuit elements, whose list now would include
not only resistors and current sources (as in the dc case), but also the induction coils
(including magnetically coupled ones) and capacitors—see figure 6.9b. In the quasi-
static limit, the current flowing in each wire is conserved, so that the ‘node rule’, i.e.
the first Kirchhoff law (4.7a),
∑I j = 0, (6.88a )
j
remains valid. Also, if the electromagnetic induction effect is restricted to the interior
of lumped induction coils as discussed above, the voltage drops Vk across each circuit
element may be still represented, just as in dc circuits, with differences of potentials of
the adjacent nodes. As a result, the ‘loop rule’, i.e. the second Kirchhoff law (4.7b),
∑Vk = 0, (6.88b)
k
is also valid. In contrast to the dc case, Eqs. (6.88) are now the (ordinary) differential
equations. However, if all circuit elements are linear (as in the examples presented in
figure 6.9b), these equations may be readily reduced to linear algebraic equations
using the Fourier expansion. (In the common case of sinusoidal ac sources, the final
stage of the Fourier series summation is unnecessary.)
My experience suggests that the potential readers of this text are very familiar with
the application of Eqs. (6.88) to such problems from their undergraduate studies, so I
Figure 6.9. (a) A typical quasi-static ac circuit obeying the Kirchhoff laws and (b) the simplest lumped circuit
elements.
6-29
would like to save time/space, skipping discussions of even the simplest examples of
such circuits, such as LC, LR, RC, and LRC loops and periodic structures59.
However, since these problems are very important in practice, my sincere advice to
the reader is to carry out a self-test by solving a few problems of this type, provided at
the end of this chapter, and if they cause difficulty, pursue remedial reading.
6.7 Displacement currents

Electromagnetic induction is not the only new effect arising in non-stationary
electrodynamics. Indeed, although Eqs. (6.21) are adequate for the description of
quasi-static phenomena, a deeper analysis shows that one of these equations, namely
∇ × H = j, cannot be exact. To see that, let us take the divergence of both its sides:
∇ ⋅ (∇ × H) = ∇ ⋅ j . (6.89)
But, as for the divergence of any curl60, the left-hand side should equal zero. Hence
we obtain
∇ ⋅ j = 0. (6.90)
This is fine in statics, but in dynamics this equation forbids any charge accumu-
lation, because according to the continuity relation (4.5)
∂ρ
∇⋅j=− . (6.91)
∂t
This discrepancy was recognized by James Clerk Maxwell who suggested, in the
1860s, a way out of this contradiction. If we generalize the equation for ∇ × H by
adding to j (that describes real currents) the so-called displacement current term,
∂D
jd ≡ , (6.92)
∂t
(which of course vanishes in statics), then the equation takes the form
∂D
∇ × H = j + jd = j + . (6.93)
∂t
In this case, due to the equation (3.32), ∇ · D = ρ, the divergence of the right-hand
side equals zero due to the continuity equation (6.92), and the discrepancy is removed.
This incredible theoretical feat61, confirmed by the 1886 experiments by Heinrich
Hertz (see below) was perhaps the main triumph of the theoretical physics of the
nineteenth century.
59
Curiously enough, these effects include wave propagation in periodic LC circuits, even within the quasi-
static approximation! However, the speed 1/(LC)1/2 of these waves in lumped circuits is much lower than the
speed 1/(εμ)1/2 of electromagnetic waves in the surrounding medium—see section 6.8 below.
60
Again, see Eq. (A.72)—if you need.
61
It looks deceptively simple with current mathematical tools (much superior to those available to Maxwell),
and after the fact.
6-30
Figure 6.10. The Ampère law applied to a recharged capacitor.
The Maxwell’s displacement current concept, expressed by Eq. (6.93), is so

important that it is worth having one more look at its derivation using the particular
model shown in figure 6.1062. Neglecting the fringe field effects, we may use Eq. (4.1)
to describe the relation between the current I flowing through the wires and the
electric charge Q of the capacitor63:
dQ
= I. (6.94)
dt
Now let us consider a closed contour C drawn around the wire. (The solid points in
figure 6.10 show the places where the contour pierces the plane of drawing.) This
contour may be seen as the line limiting either the surface S1 (crossed by the wire) or
the surface S2 (avoiding such crossing by passing through capacitor’s gap). Applying
the macroscopic Ampère law (5.116) to the former surface we obtain
∮C H ⋅ d r = ∫S jn d 2r = I ,
1
(6.95)
while for the latter surface the same law gives a different result,
∮C H ⋅ d r = ∫S jn d 2r =
2
0, (WRONG), (6.96)
for the same integral. This is just an integral-form manifestation of the discrepancy
outlined above, but it shows clearly how serious the problem is (or rather was—
before Maxwell).
Now let us see how the introduction of the displacement currents saves the day,
considering for the sake of simplicity a plane capacitor of area A with a constant
electrode spacing. In this case, as we already know, the field inside it is uniform with
62
No physicist should be ashamed of doing this. For example, J Maxwell’s main book, A Treatise of Electricity
and Magnetism, is full of drawings of plane capacitors, inductance coils, and voltmeters. More generally, the
whole history of science teaches us that snobbishness toward particular examples and practical systems is a
sure way toward producing nothing of either practical value or fundamental importance. In any productive
science, all ways leading to novel, correct results should be welcome.
63
This is of course just the integral form of the continuity equation (6.91).
6-31
D = σ, so that the total capacitor’s charge Q = Aσ = AD and the current (6.94) may
be represented as
dQ dD
I= =A . (6.97)
dt dt
So, instead of Eq. (6.96), the modified Ampère law gives
∂Dn 2 dD
∮C H ⋅ d r=∫S (jd )nd 2r = ∫S
2 2 ∂t
d r=
dt
A = I, (6.98)
i.e. the Ampère integral becomes independent of the choice of the surface limited by
the contour C—as it has to, because the surface exists only in our imagination.
6.8 Finally, the full Maxwell equation system

This is a very special moment in this course: with the displacement current inclusion,
i.e. with the replacement of Eq. (5.107) with Eq. (6.93), we have finally arrived at the
full set of macroscopic Maxwell equations for time-dependent fields64,
∂B ∂D
∇×E+ = 0, ∇×H− = j, (6.99a )
∂t ∂t
∇ ⋅ D = ρ, ∇ ⋅ B = 0, (6.99b)
whose validity has been confirmed in by an enormous body of experimental data.
Indeed, despite numerous efforts, no other corrections (e.g. additional terms) to the
Maxwell equations have ever been found, and these equations are still considered
exact within the range of their validity, i.e. while the electric and magnetic fields may
be considered classically. Moreover, even in the quantum case, these equations are
believed to be strictly valid as relations between the Heisenberg operators of the
electric and magnetic field65.
The most striking feature of these equations is that, even in the absence of stand-
alone charges and currents, when all the equations become homogeneous,
∂B ∂D
∇×E=− , ∇×H= , (6.100a )
∂t ∂t
∇ ⋅ D = 0, ∇ ⋅ B = 0, (6.100b)
they still describe something very non-trivial: electromagnetic waves, including light.
The physics of the waves may be clearly seen from Eq. (6.100a): according to the first
64
This vector form of the equations, magnificent it its symmetry and simplicity, was developed in 1884–85 by
O Heaviside, with substantial contributions by H Lorentz. (The original Maxwell’s result looked like a system
of 20 equations for Cartesian components of the vector and scalar potentials.) Note that the microscopic
Maxwell equations for genuine (‘microscopic’) fields E and B may be formally obtained from Eqs. (6.99) by the
substitutions D = ε0E and H = B/μ0, and the simultaneous replacement of the stand-alone charge and current
densities in their right-hand sides with the full ones.
65
See, e.g. Part QM chapter 9.
6-32
of them, the change of magnetic field creates a vortex-like (divergence-free) electric

field. On the other hand, the second of Eqs. (6.100a) describes how the changing
electric field, in turn, creates a vortex-like magnetic field. Thus exchanging energy,
the electomagnetic field may propagate as waves.
We will carry out a detailed quantitative analysis of the waves in the next chapter;
here I will only use this notion to fulfil the promise given in section 6.3, namely to
establish the condition of validity of the quasi-static approximation (6.21). For
simplicity, let us consider an electromagnetic wave with a time period T , velocity v,
and hence the wavelength λ = vT in a linear medium with D = εE, B = μH, and j = 0.
Then the magnitude of the left-hand side of the first of Eqs. (6.100a) is of the order of
E /λ = E /vT , while that of its right-hand side may be estimated as B /T = μH /T .
Using similar estimates for the second of Eqs. (6.100a), we arrive at the following two
(approximate) requirements66:
E 1
∼ μv ∼ . (6.101)
H εv
In order to insure the compatibility of these two relations, the waves’ speed should
satisfy the estimate
1
v∼ , (6.102)
(εμ)1/2
reduced to v ∼ 1/(ε0μ0)1/2 ≡ c in free space, while the ratio of the electric and
magnetic field amplitudes should be of the following order:
E 1 ⎛ μ ⎞1/2
∼ μv ∼ μ = ⎜ ⎟ . (6.103)
H (εμ)1/2 ⎝ε⎠
(In the next chapter we will see that these are indeed the exact results for a planar
electromagnetic wave.)
Now, let a system of size ∼a carry currents producing a certain magnetic field H.
Then, according to Eq. (6.100a), their magnetic field Faraday-induces the electric
field of magnitude E ∼ μHa/T , whose displacement currents in turn produce an
additional magnetic field with magnitude
aε aε μa ⎛ aλ ⎞2 ⎛ a ⎞2
H′ ∼ E∼ H∼⎜ ⎟ H = ⎜ ⎟ H. (6.104)
T T T ⎝ vTλ ⎠ ⎝λ⎠
Hence, the displacement current effects are negligible for a system of size a ≪ λ.67
In particular, the quasi-static picture of the skin effect, which was discussed in
section 6.3, is valid while the skin depth (6.33) remains much smaller than the
corresponding wavelength,
66
The fact that T cancels, shows that these estimates are valid for waves of arbitrary frequency.
67
Let me emphasize that if this condition is not fulfilled, the lumped-circuit representation of the system (see
figure 6.9 and its discussion) is typically inadequate—apart from some special cases, to be discussed in chapter 7.
6-33
2πv ⎛ 4π 2 ⎞
1/2
λ = vT = =⎜ ⎟ . (6.105)
ω ⎝ εμω 2 ⎠
The wavelength decreases with the frequency as 1/ω, i.e. faster than δs ∝ 1/ω1/2, so
that they become comparable at the crossover frequency
σ σ
ωr = ≡ , (6.106)
ε κε0
which is nothing else than the reciprocal charge relaxation time (4.10). As was discussed
in section 4.2, for good metals this frequency is extremely high (about 1018 s−1), so the
validity of Eq. (6.33) is typically limited by the anomalous skin effect (which was briefly
discussed in section 6.3), rather than the wave effects.
Before going after the analysis of the full Maxwell equations in particular
situations (that will be the main goal of all the next chapters of this course), let us
have a look at the energy balance they yield for a certain volume V, which may
include both charged particles and the electromagnetic field. Since according to
Eq. (5.10) the magnetic field does no work on charged particles even if they move,
the total power P being transferred from the field to the particles inside the volume is
due to the electric field alone—see Eq. (4.38):
P= ∫V p d 3r , p = j ⋅ E, (6.107)
Expressing j from the corresponding Maxwell equation of the system (6.99), we

obtain
⎡ ⎤
P= ∫V ⎢⎣E ⋅ (∇ × H) − E ⋅ ∂∂Dt ⎥⎦ d 3r . (6.108)
Let us pause here for a second and transform the divergence of vector E × H,
using the well-known vector algebra identity68:
∇ ⋅ (E × H) = H ⋅ (∇ × E) − E ⋅ (∇ × H). (6.109)
The last term on the right-hand side of this equation is exactly the first term in the
square brackets of Eq. (6.108), so that we can rewrite that formula as
⎡ ⎤
P= ∫V ⎢⎣−∇ ⋅ (E × H) + H ⋅ (∇ × E) − E ⋅ ∂∂Dt ⎥⎦ d 3r. (6.110)
However, according to the Maxwell equation for ∇ × E, it is equal to −∂B/∂t, so

that the second term in the square brackets of Eq. (6.110) equals −H · ∂B/∂t and,
according to Eq. (6.14), is just the (minus) time derivative of the magnetic energy per
unit volume. Similarly, according to Eq. (3.76), the third term under the integral is
the minus time derivative of the electric energy per unit volume. Finally, we can use
68
See, e.g. Eq. (A.77) with f = E and g = H.
6-34
the divergence theorem to transform the integral of the first term in the square
brackets to a 2D integral over the surface S limiting the volume V. As a result, we
obtain the so-called Poynting theorem69 for the power balance in the system:
⎛ ⎞
∫V ⎝p + ∂∂ut ⎠d 3r + ∮S Sn d 2r = 0.
⎜ ⎟ (6.111)
Here u is the density of the total (electric plus magnetic) energy of the electro-
magnetic field, with
δu ≡ E ⋅ δ D + H ⋅ δ B, (6.112)
so that for an isotropic, linear, and dispersion-free medium, with D(t) = εE(t), B(t) =
μH(t),
E⋅D H⋅B εE 2 B2
u= + ≡ + , (6.113)
2 2 2 2μ
and S is the Poynting vector defined as70
S ≡ E × H. (6.114)
The first integral in Eq. (6.111) is evidently the net change of the energy of the
system (particles + field) per unit time, so that the second (surface) integral has to be
the power flowing out from the system through the surface. As a result, it is tempting
to interpret the Poynting vector S locally, as the power flow density at the given
point. In many cases such a local interpretation of the vector S is legitimate;
however, in some cases it may lead to incorrect conclusions. Indeed, let us consider
the simple system shown in figure 6.11: a charged plane capacitor placed into a static
and uniform external magnetic field, so that the electric and magnetic fields are
mutually perpendicular.
Figure 6.11. The Poynting vector paradox.
In this static situation, with no charges moving, both p and ∂/∂t are equal to zero,
and there should be no power flow in the system. However, Eq. (6.114) shows that
the Poynting vector is not equal to zero inside the capacitor, being directed as the red
arrows in figure 6.11 show. From the point of view of our only unambiguous
corollary of the Maxwell equations, Eq. (6.111), there is no contradiction here,
69
Named after J Poynting, although this fact was independently discovered by O Heaviside, while a similar
expression for the intensity of mechanical elastic waves had been derived earlier by N Umov.
70
Actually, an addition to S of the curl of an arbitrary vector function f(r, t) does not change Eq. (6.111). Indeed,
we may use the divergence theorem to transform the corresponding change of the surface integral in Eq. (6.111)
to a volume integral of scalar function ∇ · (∇ × f) that equals zero at any point—see, e.g. Eq. (A.72).
6-35
because the fluxes of vector S through the side boundaries of the volume shaded in
figure 6.11, are equal and opposite (and they are zero for other faces of this
rectilinear volume), so that the total flux of the Poynting vector through the volume
boundary equals zero, as it should. It is, however, useful to recall this example each
time before giving a local interpretation to the vector S.
The paradox illustrated in figure 6.11 is closely related to the radiation recoil
effects, due to the electromagnetic field’s momentum—more exactly, it linear
momentum. Indeed, acting as at the Poynting theorem derivation, it is straight-
forward to use the microscopic Maxwell equations71 to prove that, neglecting the
boundary effects, the vector sum of the mechanical linear momentum of the particles
in an arbitrary volume and the integral of the following vector,
S
g≡ , (6.115)
c2
over the same volume is conserved, allowing us to interpret g as the density of the
linear momentum of the electromagnetic field. (It will be more convenient for me to
prove this relation and discuss the related issues in section 9.8, using the four-vector
formalism of the special relativity.) Due to this conservation, if some static fields
coupled to mechanical bodies are suddenly decoupled from them and are allowed to
propagate in space, i.e. change their local integral of g, they give the bodies an
opposite and equal impulse of force.
Finally, to complete our initial discussion of the Maxwell equations72, let us
rewrite them in terms of potentials A and ϕ, because this is more convenient for the
solution of some (though not all!) problems. Even when dealing with the system
(6.99) of the more general Maxwell equations than discussed before, Eqs. (6.7) are
still used for the potential definitions. It is straightforward to verify that with these
definitions, the two homogeneous Maxwell equations (6.99b) are satisfied automati-
cally. Plugging Eq. (6.7) into the inhomogeneous equations (6.99a) and considering,
for simplicity, a linear, uniform medium with frequency-independent ε and μ, we
obtain
∂ ρ ∂ 2A ⎛ ∂ϕ ⎞
∇2 ϕ + (∇ ⋅ A) = − , ∇2 A − εμ − ∇⎜∇ ⋅ A + εμ ⎟ = −μ j . (6.116)
∂t ε ∂t 2 ⎝ ∂t ⎠
This is a more complex result than we would like to obtain. However, let us select
a special gauge, which is frequently called (particularly for the free space case, when
v = c) the Lorenz gauge condition73
71
The situation with the macroscopic Maxwell equations is more complex, and is still the subject of some
lingering discussions (usually called the Abraham–Minkowski controversy, despite contributions by many other
scientists including A Einstein), because of the ambiguity of the momentum’s division between the field and
particle parts—see, e.g. the recent review paper [6].
72
We will return to their general discussion (in particular, to the analytical mechanics of the electromagnetic
field and its stress tensor) in section 9.8, after we have equipped ourselves with the theory of special relativity.
73
This condition, named after L Lorenz, should not be confused with the so-called Lorentz invariance condition
of the relativity theory, due to H Lorentz (note the last names’ spellings), to be discussed in section 9.4.
6-36
∂ϕ
∇ ⋅ A + εμ = 0, (6.117)
∂t
which is a natural generalization of the Coulomb gauge (5.48) to time-dependent
phenomena. With this condition, Eq. (6.107) are reduced to a simpler, beautifully
symmetric form,
1 ∂ 2ϕ ρ 1 ∂ 2A
∇2 ϕ − 2 2
=− , ∇2 A − = −μ j , (6.118)
v ∂t ε v 2 ∂t 2
where v2 ≡ 1/εμ. Note that these equations are essentially a set of four similar
equations for four scalar functions (namely, ϕ and three Cartesian components of
the vector A) and thus clearly invite the four-component vector formalism of the
theory of relativity; it will be discussed in chapter 9.74
If ϕ and A depend on just one spatial coordinate, say z, in a region without field
sources ρ = 0, j = 0, Eqs. (6.118) are reduced to the well-known 1D wave equations
∂ 2ϕ 1 ∂ 2ϕ ∂ 2A 1 ∂ 2A
− 2 2 = 0, − 2 2 =0. (6.119)
2 2
∂z v ∂t ∂z v ∂t
These equations describe waves with arbitrary waveforms, propagating with the
same speed v in either of the directions of axis z. Due to the definitions of the
constants ε0 and μ0, in the free space v is just the speed of light:
1
v= ≡ c. (6.120)
(ε0μ0)1/2
Historically, the experimental observation of relatively low-frequency (GHz-scale)

electromagnetic waves and the proof that their speed in free space is equal to that of
light was the decisive proof of Maxwell’s theory. This was first accomplished in 1886
by H Hertz, using electronic circuits and antennas he had invented for this purpose.
Before proceeding to the detailed analysis of these waves in the following
chapters, let me mention that the invariance of Eqs. (6.119) with respect to the
wave propagation direction is not occasional; it is just a manifestation of one more
general property of the Maxwell equations (6.99), called the Lorentz reciprocity. We
have already met its simplest example, for time-independent electrostatic fields, in
74
Here I have to mention in passing the so-called Hertz vector potentials Πe and Πm (whose introduction
may be traced back to at least the 1904 work by E Whittaker). They may be defined by the following
relations:
∂Π e 1
A=μ + μ∇ ⋅ Πm, φ = − ∇ ⋅ Π e,
∂t ε
which make the Lorentz gauge condition (6.117) automatically satisfied. These potentials are particularly
convenient for the solution of problems in which the electromagnetic field is excited by external sources
characterized by externally fixed electric and magnetic polarizations P and M—rather than fixed charge and
current densities ρ and j. Indeed, it is straightforward to check that both Πe and Πm satisfy the equations
similar to Eq. (6.118), but with their right-hand sides equal to, respectively, −P and −M. Unfortunately, I
would not have time for a discussion of such problems.
6-37
one of the problems of chapter 1. Let us now consider a much more general case
when two time-dependent electromagnetic fields, say {E1(r, t), H1(r, t)} and {E2(r, t),
H2(r, t)}, are induced, respectively, by spatially localized stand-alone currents j1(r, t)
and j2(r, t). Then it may be proved75 that if the medium is linear, and either isotropic
or even anisotropic, but with symmetric tensors εjj′ and μjj′, then for any volume V,
limited by a closed surface S,
∫V (j1 ⋅ E2 − j2 ⋅ E1) d 3r = ∮S (E1 × H2 − E2 × H1) n d 2r. (6.121)
This property implies, in particular, that the waves propagate similarly in two
reciprocal directions even in situations much more general than the 1D case
described by Eqs. (6.119). For some important practical applications (e.g. for low-
noise amplifiers and detectors) such reciprocity is rather inconvenient. Fortunately,
Eq. (6.121) may be violated in anisotropic media with asymmetric tensors εjj′ and/or
μjj′. The simplest, and most important case of such an anisotropy, the Faraday
rotation of the wave polarization in plasma, will be discussed in the next chapter.
6.9 Problems
Problem 6.1. Prove that the electromagnetic induction e.m.f. V ind in a conducting
loop may be measured as shown on two panels of figure 6.1:
(i) by measuring the current I = V ind /R induced in the closed loop with Ohmic
resistance R, or
(ii) using a voltmeter inserted into the loop.
Problem 6.2. The flux Φ of the magnetic field that pierces a resistive ring is being
changed in time, while the magnetic field outside the ring is negligibly low. A
voltmeter is connected to a part of the ring, as shown in the figure below. What
would the voltmeter show?
Problem 6.3. A weak, uniform magnetic field B is applied to an axially symmetric

permanent magnet, with the dipole magnetic moment m directed along the
symmetry axis, rapidly rotating about the same axis, with an angular momentum
L. Calculate the electric field resulting from the magnetic field’s application and
formulate the conditions of your result’s validity.
75
A warning: the proof of Eq. (6.121), given in many textbooks and online sites, is deficient.
6-38
Problem 6.4. The similarity of Eq. (5.53), obtained in section 5.3 without any use of
the Faraday induction law, and Eq. (5.54), proved in section 6.2 using the law,
implies that the law may be derived from magnetostatics. Prove that this is indeed
true for a particular case of a current loop, being slowly deformed in a fixed
magnetic field B.
Problem 6.5. Could problem 5.1 (i.e. the analysis of the mechanical stability of the
system shown in the figure below) be solved using potential energy arguments?
Problem 6.6. Use energy arguments to calculate the pressure exerted by the
magnetic field B inside a long uniform solenoid of length l, and a cross-section of
area A ≪ l2, with N ≫ l/A1/2 ≫ 1 turns, on its ‘walls’ (windings), and the force
exerted by the field on the solenoid’s ends, for two cases:
(i) the current through the solenoid is fixed by an external source, and
(ii) after the initial current setting, the ends of the solenoid’s wire, with a negligible
resistance, are connected so that it continues to carry a non-zero current.
Compare the results and give a physical interpretation of the direction of these forces.
Problem 6.7. The electromagnetic railgun is a projectile launch system consisting of

two long, parallel conducting rails and a sliding conducting projectile, shorting the
current I fed into the system by a powerful source—see panel (a) in the figure below.
Calculate the force exerted on the projectile using two approaches:
(i) by a direct calculation, assuming that the cross-section of the system has the
simple shape shown on panel (b) of the figure below, with t ≪ w, l, and
(ii) using the energy balance (for simplicity, neglecting the Ohmic resistances in the
system),
and compare the results.
6-39
Problem 6.8. A uniform, static magnetic field B is applied along the axis of a long
round pipe of a radius R, and a very small thickness τ, made of a material with
Ohmic conductivity σ. A sphere of mass M and radius R′ < R, made of a linear
magnetic with permeability μ ≫ μ0, is launched with an initial velocity v0 to fly
ballistically along the pipe’s axis—see the figure below. Use the quasi-static
approximation to calculate the distance the sphere would pass before it stops.
Formulate the conditions of validity of your result.
Problem 6.9. AC current of frequency ω is being passed through a long uniform

wire with a round cross-section of radius R comparable to the skin depth δs. In the
quasi-static approximation, find the current’s distribution across the cross-section
and analyze it in the limits R ≪ δs and δs ≪ R. Calculate the effective ac resistance of
the wire (per unit length) in these two limits.
Problem 6.10. A very long, round cylinder of radius R, made of a uniform

conductor with Ohmic conductivity σ and magnetic permittivity μ, has been placed
into a uniform ac magnetic field Hext(t) = H0cos ωt directed along its symmetry axis.
Calculate the spatial distribution of the magnetic field’s amplitude, and in particular
its value on the cylinder’s axis. Spell out the last result in the limits of relatively small
and large R.
Problem 6.11.* Define and calculate an appropriate spatial-temporal Green’s

function for Eq. (6.25), and then use this function to analyze the dynamics of
propagation of the external magnetic field, suddenly turned on at t = 0 and then left
constant:
⎧ 0, at t < 0,
H (x < 0, t ) = ⎨
⎩ H0, at t > 0,
into an Ohmic conductor occupying the half-space x > 0—see figure 6.2.
Hint: Try to use a function proportional to exp{−(x − x′)2/2(δx)2}, with a suitable

time dependence of the parameter δx, and a properly selected pre-exponential factor.
Problem 6.12. Solve the previous problem using the variable separation method,
and compare the results.
Problem 6.13. A small, planar wire loop, carrying current I, is located far from a
plane surface of a superconductor. Within the ‘coarse-grain’ (ideal-diamagnetic)
description of the Meissner–Ochsenfeld effect, calculate:
6-40
(i) the energy of the loop-superconductor interaction,

(ii) the force and torque acting on the loop, and
(iii) the distribution of supercurrents on the superconductor surface.
Problem 6.14. A straight, uniform magnet of length 2 l, cross-section area A ≪ l2,

and mass m, with a permanent longitudinal magnetization M0, is placed over a
horizontal surface of a superconductor—see the figure below. Within the ideal-
diamagnet description of the Meissner–Ochsenfeld effect, find the stable equilibrium
position of the magnet.
Problem 6.15. A planar superconducting wire loop of area A and inductance L may
rotate, without friction, about a horizontal axis 0 (in the figure below, perpendicular
to the plane of drawing) passing through its center of mass. Initially the loop was
horizontal (with θ = 0), and carried supercurrent I0 in such direction that its
magnetic dipole vector was directed down. Then a uniform magnetic field B,
directed vertically up, was applied. Using the ideal-diamagnet description of the
Meissner–Ochsenfeld effect, find all possible equilibrium positions of the loop,
analyze their stability, and give a physical interpretation of the results.
Problem 6.16. Use the London equation to analyze the penetration of a uniform
external magnetic field into a thin (t ∼ δL), planar superconducting film, whose plane
is parallel to the field.
Problem 6.17. Use the London equation to calculate the distribution of supercurrent
density j inside a long, straight superconducting wire, with a circular cross-section of
radius R ∼ δL, carrying dc current I.
Problem 6.18. Use the London equation to calculate the inductance (per unit length)
of a long, uniform superconducting strip placed close to the surface of a similar
superconductor—see the figure below, which shows the structure’s cross-section.
6-41
Problem 6.19. Calculate the inductance (per unit length) of a superconducting cable
with a round cross-section, shown in the figure below, in the following limits:
(i) δL ≪ a, b, c − b, and
(ii) a ≪ δL ≪ b, c − b.
Problem 6.20. Use the London equation to analyze the magnetic field shielding by a
superconducting thin film of thickness t ≪ δL, by calculating the penetration of the
field induced by current I flowing in a thin wire which runs parallel to a wide, planar
thin film, at distance d ≫ t from it, into the half-space behind the film.
Problem 6.21. Use the Ginzburg–Landau equations (6.54) and (6.63) to calculate
the largest (‘critical’) value of the supercurrent in a uniform, long superconducting
wire of a small cross-section Aw ≪ δL2.
Problem 6.22. Use the discussion of a long, straight Abricosov vortex in the
approximation ξ ≪ δL in section 6.5 to prove Eqs. (6.71) and (6.72) for its energy
per unit length and the first critical field.
Problem 6.23.* Use the Ginzburg–Landau equations (6.54) and (6.63) to prove the
Josephson’s relation (6.76) for a small superconducting weak link, and express its
critical current Ic via the Ohmic resistance Rn of the same weak link in its normal
state.
Problem 6.24. Use Eqs. (6.76) and (6.79) to calculate the coupling energy of a
Josephson junction, and the full potential energy of the SQUID shown in figure 6.4c.
Problem 6.25. Analyze the possibility of wave propagation in a long, uniform chain
of lumped inductances and capacitances—see the figure below.
6-42
Hint: Readers without prior experience with electromagnetic wave analysis may like
to use a substantial analogy between this effect and mechanical waves in a 1D chain
of elastically coupled particles76.
Problem 6.26. A sinusoidal e.m.f. of amplitude V0 and frequency ω is applied to an

end of a long chain of similar, lumped resistors and capacitors, shown in the figure
below. Calculate the law of decay of the ac oscillation amplitude in the chain.
Problem 6.27. As was discussed in section 6.7, the displacement current concept
allows one to generalize the Ampère law to time-dependent processes as
∮C H ⋅ d r = IS + ∂∂t ∫S Dn d 2r.
We also have seen that such generalization makes ∫H·dr over an external contour,
such as the one shown in figure 6.10, independent of the choice of the surface S
limited by the contour. However, it may look like the situation is different for a
contour drawn inside the capacitor—see the figure below. Indeed, if the contour’s
size is much larger than the capacitor’s thickness, the magnetic field H created by the
linear current I on the contour’s line is virtually the same as that of a continuous
wire, and hence the integral ∫H · dr along the contour apparently does not depend on
its area, while the magnetic flux ∫Dn d2r does, so that the equation displayed above
seems invalid. (The current IS piercing this contour evidently equals zero.) Resolve
the paradox, for simplicity considering an axially symmetric system.
76
6-43
Problem 6.28. A straight, uniform, long wire with circular cross-section of radius R,
is made of an Ohmic conductor with conductivity σ, and carries dc current I.
Calculate the flux of the Poynting vector through its surface, and compare it with the
Joule energy losses.
References
[1] Abrikosov A 1972 Introduction to the Theory of Normal Metals (New York: Academic)
[2] Tinkham M 1996 Introduction to Superconductivity 2nd edn (New York: McGraw-Hill)
[3] Clarke J and Braginski A (ed) The SQUID Handbook vol 2 (Wiley)
[4] Grosz A et al (ed) 2017 High Sensitivity Magnetometers (Springer)
[5] Bunyk P et al 2001 Int. J. High Speed Electron. Syst. 11 257
[6] Pfeiffer R et al 2007 Rev. Mod. Phys. 79 1197
6-44
IOP Publishing
Lecture notes
Chapter 7
Electromagnetic wave propagation
This (rather long) chapter focuses on the most important effect that follows from the
time-dependent Maxwell equations, namely electromagnetic waves, at this stage
avoiding discussion of their origin—radiation—which will the subject of chapters 8
and 10. The discussion starts from the simplest plane waves in a uniform and isotropic
medium, and then proceeds to non-uniform systems, in particular those with sharp
boundaries between different materials, bringing up such effects as reflection and
refraction. Then we will discuss the so-called guided waves, propagating along various
long transmission lines—such as coaxial cables, waveguides, and optical fibers.
Finally, the end of the chapter is devoted to final-length fragments of such lines
serving as resonators and to the effects of energy dissipation in transmission lines and
resonators.
7.1 Plane waves

Let us start from considering a spatial region that does not contain field sources
(ρ = 0, j = 0), and is filled with a linear, uniform, isotropic medium, which obeys
Eqs. (3.46) and (5.110):
D = εE, B = μH . (7.1)
Moreover, let us assume, for a while, that these constitutive equations hold for all
frequencies of interest. (Of course, these relations are exactly valid for the very
important particular case of free space, where we may formally use the macroscopic
Maxwell equations (6.100), but with ε = ε0 and μ = μ0.)
As was already shown in section 6.8, in this case the Lorenz gauge condition
(6.117) allows the Maxwell equations to be recast into the wave equations (6.118) for
the vector and scalar potentials. However, for most purposes it is more convenient to
use directly the homogeneous Maxwell equations (6.100) for the electric and magnetic

fields (which are independent of the gauge choice). After the elementary elimination of
D and B using Eqs. (7.1)1, these equations take a simple, symmetric form
∂H ∂E
∇×E+μ = 0, ∇×H−ε = 0, (7.2a )
∂t ∂t
∇ ⋅ E = 0, ∇ ⋅ H = 0. (7.2b)
Now, acting by operator ∇ × on each of Eqs. (7.2a), i.e. taking their curl, and then
using the vector algebra identity (5.31), whose first term for both E and H vanishes due
to Eqs. (7.2b), we obtain similar wave equations for the electric and magnetic fields:
⎛ 1 ∂2 ⎞ ⎛ 1 ∂2 ⎞
⎜∇2 − 2 2 ⎟E = 0, ⎜∇2 − 2 2 ⎟H = 0, (7.3)
⎝ v ∂t ⎠ ⎝ v ∂t ⎠
where the parameter v is defined by relation

1
v2 ≡ , (7.4)
εμ
with v2 = 1/ε0μ0 ≡ c2 in free space—see Eq. (6.120) again.

The two vector equation (7.3) are of course just a shorthand for six similar
equations for the three Cartesian components of E and H. These equations allow, in
particular, the following solution,
f = f (z − vt ), (7.5)
where z is the Cartesian coordinate along a certain (arbitrary) direction n. This
solution describes a specific type of wave, i.e. a certain field pattern moving without
deformation along axis z with velocity v. According to Eq. (7.5), each variable f has
the same value in each plane perpendicular to the direction n of wave propagation,
hence the name—plane wave.
According to Eqs. (7.2), the independence of the wave equations (7.3) for vectors
E and H does not mean that their plane-wave solutions are independent. Indeed,
plugging the solutions of the type (7.5) into Eqs. (7.2a), we obtain
n×E
H= , i.e. E = Z H × n , (7.6)
Z
where the constant Z is defined as
E ⎛ μ ⎞1/2
Z≡ =⎜ ⎟ . (7.7)
H ⎝ε⎠
1
Although B rather than H is the actual magnetic field, mathematically it is a bit more convenient (just as it
was in section 6.2) to use the latter vector in the following discussion, because at sharp media boundaries, H
obeys the boundary condition (5.117) similar to that for E—cf Eq. (3.37).
7-2
Figure 7.1. Field vectors in a plane electromagnetic wave propagating along direction n.
The vector relation (7.6) means, first of all, that the vectors E and H are
perpendicular not only to vector n (such waves are called transverse), but also to
each other (figure 7.1)—at any point of space and at any time instant. Second, this
relation does not depend on the function f, meaning that the electric and magnetic
fields increase and decrease simultaneously.
Finally, the field magnitudes are related by the constant Z, called the wave
impedance of the medium. Very soon we will see that the impedance plays a pivotal
role in many problems, in particular at the wave reflection from the interface
between the two media. Since the dimensionality of E, in SI units, is V m−1, and that
of H is A m−1, Eq. (7.7) shows that Z has the dimensionality of V/A, i.e. ohms (Ω).2
In particular, in free space,
⎛ μ ⎞1/2
Z = Z0 ≡ ⎜ 0 ⎟ = 4π × 10−7 c ≈ 377 Ω . (7.8)
⎝ ε0 ⎠
Now plugging Eq. (7.6) into Eqs. (6.113) and (6.114), we obtain
u = εE 2 = μH 2, (7.9a )
E2
S≡E×H=n = nZH 2, (7.9b)
Z
so that, according to Eqs. (7.4) and (7.7), the wave’s energy and power densities are
universally related as
S = nuv . (7.9c )
In the view of the Poynting vector paradox discussed in section 6.8 (see figure
6.11), one may wonder whether this expression may be interpreted as the actual
density of power flow. In contrast to the static situation shown in figure 6.11 that
limits the electric and magnetic fields to a vicinity of their sources, waves may travel
far from them. As a result, they can form wave packets of a finite length in free
space—see figure 7.2. Let us apply the Poynting theorem (6.111) to the cylinder
shown with dashed lines in figure 7.2, with one lid inside the wave packet and
another lid in the region already passed by the wave. Then, according to Eq. (6.111),
the rate of change of the full field energy E inside the volume is dE /dt = −SA
2
In the Gaussian units, E and H have a similar dimensionality (in particular, in a free-space wave, E = H),
making the (very useful) notion of the wave impedance less manifestly exposed—so that in some older physics
textbooks it is not mentioned at all!
7-3
Figure 7.2. Interpreting the Poynting vector in an electromagnetic wave.
(where A is the lid area), so that S may be indeed interpreted as the power flow (per
unit area) from the volume. Making a reasonable assumption that the finite length of
a sufficiently long wave packet does not affect the physics inside it, we may indeed
interpret the S given by Eqs. (7.9b) and (7.9c) as the power flow density inside a
plane electromagnetic wave.
As we will see later in this chapter, the free-space value Z0 of the wave impedance,
given by Eq. (7.8), establishes the scale of Z of virtually all wave transmission lines,
so we may use it, together with Eq. (7.9), to obtain a better feeling of how different
the electric and magnetic field amplitudes in the waves are, on the scale of typical
electrostatics and magnetostatics experiments. For example, according to Eqs. (7.9),
a wave of a modest intensity S = 1 W m−2 (this is what we get from a usual electric
bulb a few meters away from it) has E ∼ (SZ0)1/2 ∼ 20 V m−1, quite comparable with
the dc field created by a standard AA battery right outside it. On the other hand, the
wave’s magnetic field H = (S/Z0)1/2 ≈ 0.05 A m−1. For this particular case, the
relation following from Eqs. (7.1), (7.4), and (7.7),
E E E
B = μH = μ =μ 1/2
= (εμ)1/2 E = , (7.10)
Z (μ / ε ) v
gives B = μ0H = E/c ∼ 7 × 10−8 T, i.e. a magnetic field thousand times lower than the
Earth’s field, and about eight orders of magnitude lower than the field of a typical
permanent magnet. This huge difference may be interpreted as follows: the scale of
magnetic fields B ∼ E/c in the waves is ‘normal’ for electromagnetism, while the
permanent magnet fields are abnormally high, because they are due to the ferromag-
netic alignment of electron spins, essentially relativistic objects—see the discussion in
section 5.5.
As soon as ε and μ are simple constants, the wave speed v is also constant, and Eq.
(7.5) is valid for an arbitrary function f—defined by the initial conditions. In plain
English, a medium with frequency-independent ε and μ supports the propagation of
plane waves with an arbitrary waveform—without either decay (attenuation) or
deformation (dispersion). However, for any real medium but pure vacuum, this
approximation is valid only within limited frequency intervals. We will discuss the
effects of attenuation and dispersion in the next section and see that all our prior
formulas remain valid even for an arbitrary linear medium, provided that we limit
7-4
them to single-frequency (i.e. sinusoidal, frequently called monochromatic) waves.

Such waves may be most conveniently represented as3
f = Re [fω e i (kz−ωt ) ], (7.11)
where fω is the complex amplitude of the wave, and k is its wave number (the
magnitude of the wave vector k ≡ nk), sometimes also called the spatial frequency.
The last term is justified by the fact, evident from Eq. (7.11), that k is related to the
wavelength λ exactly as the usual (‘temporal’) frequency ω is related to the time
period T :
2π 2π
k= , ω= . (7.12)
λ T
Requiring Eq. (7.11) to be a particular form of Eq. (7.5), i.e. the argument (kz − ωt)
≡ k[z − (ω/k)t] to be proportional to (z − vt), so that ω/k = v, we see that the wave
number should equal
ω
k= = (εμ)1/2 ω, (7.13)
v
showing that in this ‘dispersion-free’ case the dispersion relation ω(k) is linear.
Now note that Eq. (7.6) does not mean that vectors E and H retain their direction
in space. (The simple case when they do is called the linear polarization of the wave.)
Indeed, nothing in the Maxwell equations prevents, for example, a joint rotation of
this pair of vectors around the fixed vector n, while still keeping all these three
vectors perpendicular to each other at any instant—see figure 7.1. An arbitrary
rotation law, or even an arbitrary constant frequency of such rotation, however,
would violate the single-frequency (monochromatic) character of the elementary
sinusoidal wave (7.11). In order to understand what is the most general type of
polarization the wave may have without violating that condition, let us represent
two Cartesian components of one of these vectors (say, E) along any two fixed axes x
and y, perpendicular to each other and axis z (i.e. to vector n), in the same form as
used in Eq. (7.11):
Ex = Re [Eωxe i (kz−ωt ) ], Ey = Re [Eωye i (kz−ωt ) ]. (7.14)
In order to keep the wave monochromatic, the complex amplitudes Eωx and Eωy
must be constant; however, they may have different magnitudes and an arbitrary
phase shift between them.
In the simplest case when the arguments of the complex amplitudes are equal,
Eωx,y = Eωx,y e iφ , (7.15)
3
As we have already seen in the previous chapter (see also Part CM section 5.1), such a complex-exponential
representation of sinusoidally changing variables is more convenient for mathematical manipulation than
using sine and cosine functions, particularly because in all linear relations, the operator Re may be omitted
(implied) until the very end of the calculation. Note, however, that this is not valid for the quadratic forms such
as Eqs. (7.9a) and (7.9b).
7-5
Figure 7.3. Time evolution of the electric field vector in monochromatic waves with: (a) a linear polarization,
(b) the circular polarization, and (c) an elliptical polarization.
the real field components have the same phase,

Ex = Eωx cos(kz − ωt + φ), Ey = ∣Eωy∣ cos(kz − ωt + φ), (7.16)
so that their ratio is constant in time—see figure 7.3a. This means that the wave is
linearly polarized, within the polarization plane defined by the relation
tan θ = ∣Eωy∣ / Eωx . (7.17)
Another simple case is when the moduli of the complex amplitudes Eωx and Eωy
are equal, but their phases are shifted by +π/2 or −π/2:
Eωx = Eω e iφ , Eωy = Eω e i(φ±π /2). (7.18)
In this case
Ex = Eω cos (kz − ωt + φ),
⎛ π⎞ (7.19)
Ey = Eω cos ⎜kz − ωt + φ ± ⎟ = ∓ Eω sin (kz − ωt + φ).
⎝ 2⎠
This means that on the [x, y] plane, the end of the vector E measured at a fixed
coordinate z moves with the wave’s frequency ω either clockwise or counter-
clockwise around a circle—see figure 7.3b:
θ (t ) = ∓(ωt − φ). (7.20)
Such waves are called circularly polarized4. These particular solutions of the
Maxwell equations are very convenient for quantum electrodynamics, because
single electromagnetic field quanta with a certain (positive or negative) spin direction
may be considered as elementary excitations of the corresponding circularly
polarized wave5. (This fact does not exclude, from the quantization scheme, waves
4
In the dominant convention, the wave is called right-polarized (RP) if it is described by the lower sign in
Eqs. (7.18)–(7.20) (i.e. if the vector ω of the angular frequency of the field vector’s rotation coincides with the
wave propagation’s direction n), and left-polarized (LP) in the opposite case.
5
This issue is closely related to that of the radiation’s angular momentum; it will be more convenient for me to
discuss it later in this chapter (in section 7.7).
7-6
of other polarizations, because any monochromatic wave may be represented as a

linear combination of two opposite circularly polarized waves—just as Eqs. (7.14)
represent it as a linear combination of two linearly polarized waves.)
Finally, in the general case of arbitrary complex amplitudes Eωx and Eωy, the
electric field vector end moves along an ellipse on the [x, y] plane (figure 7.3c), such
wave is called elliptically polarized. The eccentricity and orientation of the ellipse are
completely described by one complex number, the ratio Eωx/Eωy, i.e. by two real
numbers: ∣Eωx/Eωy∣ and φ = arg(Eωx/Eωy).6
7.2 Attenuation and dispersion

Let me start the discussion of the dispersion and attenuation effects by considering a
particular case of the time evolution of the electric polarization of a dilute, non-polar
medium, with negligible interaction between its elementary dipoles p(t). As was
discussed in section 3.3, in this case the local electric field acting on each elementary
dipole equals the macroscopic field E(t). Then, the polarization p(t) ≠ 0 may be
caused only by the values of the field E at the same moment of time (t), or at the
earlier moments of time, t < t′. Due to the linear superposition principle, the
macroscopic polarization P(t) ≡ np(t) should be a linear sum (integral) of the values
of E(t′) at all previous moments of time, t′ < t, weighed by some function of t and t′:7
t
P (t ) = ∫−∞ E (t′)G(t, t′)dt′. (7.21)
The condition t′ < t, which is implied by this relation, expresses a key principle of
physics, the causal relation between a cause (in our case, the electric field applied to
each dipole) and its effect (the polarization it creates). The function G(t, t′) is called
the temporal Green’s function for the electric polarization8. In order to understand its
physical sense, let us consider the case when the applied field E(t) is a very short
pulse at the moment t0 < t, that may be approximated with Dirac’s delta-function:
E (t ) = δ(t − t0). (7.22)
6
Note that the same information may be expressed via four so-called Stokes parameters s0, s1, s2, s3, which are
popular in practical optics, because they may be used for the description of not only completely coherent waves
that are discussed here, but also of party coherent or even fully incoherent waves—including the natural light
emitted by thermal sources such as our Sun. (In contrast to the coherent waves (7.14), whose complex
amplitudes are deterministic numbers, the amplitudes of incoherent waves should be treated as stochastic
variables.) For more on the Stokes parameters, as well as about many optics topics I will not have time to
cover (in particular geometrical optics), I can recommend the classical text [1].
7
In an isotropic medium, vectors E, P, and hence D = ε0E + P, are all parallel, and for simplicity of notation I
will drop the vector sign in the following formulas. I am also assuming that P at any point r is only dependent
on the electric field at the same point, and hence drop the term ikz from the exponent’s argument. This
assumption is valid if the wavelength λ is much larger than the elementary media dipole’s size a. In most
systems of interest, the scale of a is atomic (∼10−10 m), so that the last approximation is valid up to very high
frequencies, ω ∼ c/a ∼ 1018 s−1, corresponding to hard x-rays.
8
The idea of these functions is very similar to that of the spatial Green’s functions (see section 2.10), but with a
new twist, due to the causality principle. A discussion of the temporal Green’s functions in application to
classical mechanics may be found in Part CM section 5.1.
7-7
Figure 7.4. An example of the temporal Green’s function for electric polarization.
Then Eq. (7.21) yields just P(t) = G(t, t0), showing that the Green’s function is just
the polarization at moment t, created by a unit δ-functional pulse of the applied field
at moment t′ (figure 7.4).
What are the general properties of the temporal Green’s function? First, the
function is evidently real, since the dipole moment p and hence polarization P = np
are real by definition—see Eq. (3.6). Next, for systems without infinite internal
‘memory’, G should tend to zero at t − t′ → ∞, although the type of this approach
(e.g. whether the function G oscillates approaching zero—see figure 7.4) depends on
the elementary dipole’s properties. Finally, if parameters of the medium do not
change in time, the polarization response to an electric field pulse should be
dependent not on its absolute timing, but only on the time difference θ ≡ t − t′
between the pulse and observation instants, i.e. Eq. (7.21) is reduced to
t ∞
P (t ) = ∫−∞ E (t′)G(t − t′)dt′ ≡ ∫0 E ( t − θ ) G ( θ ) dθ . (7.23)
For a sinusoidal waveform, E(t) = Re [Eωe−iωt], this equation yields

∞ ⎧⎡ ∞ ⎤ ⎫
P(t ) = Re ∫0 Eωe−iω(t−θ )G (θ )dθ = Re ⎨⎢Eω ∫0 G (θ )e iωθ dθ ⎥ e−iωt⎬ . (7.24)
⎩⎣ ⎦ ⎭
The expression in square brackets is of course nothing more that the complex
amplitude Pω of the polarization. This means that even if the static linear relation
(3.43), P = χeε0E, is invalid for an arbitrary time-dependent process, we may still
keep its Fourier analog,
∞
1
Pω = χe (ω)ε0Eω, with χe (ω) ≡
ε0
∫0 G (θ )e iωθ dθ , (7.25)
for each sinusoidal component of the process, using it as the definition of the
frequency-dependent electric susceptibility χe(ω). Similarly, the frequency-dependent
electric permittivity may be defined using the Fourier analog of Eq. (3.46):
Dω ≡ ε(ω)Eω. (7.26a )
Then, according to the definition (3.33), the permittivity is related to the temporal
Green’s function by the usual Fourier transform:
∞
Pω
ε( ω ) ≡ ε0 +
Eω
= ε0 + ∫0 G (θ )e iωθ dθ . (7.26b)
7-8
This relation shows that ε(ω) may be complex,

∞
ε(ω) = ε′(ω) + iε″(ω), with ε′(ω) = ε0 + ∫0 G (θ )cos ωθ dθ ,
∞ (7.27)
ε″( ω ) = ∫0 G (θ )sin ωθ dθ ,
and that its real part ε′(ω) is always an even function of frequency, while the
imaginary part εʺ(ω) is an odd function of ω.
Although the particular causal relationship (7.21) between P(t) and E(t) is
conditioned by the elementary dipole independence, the frequency-dependent
complex electric permittivity ε(ω) may be introduced in a similar way if any two
linear combinations of these variables are related by a similar formula. Absolutely
similar arguments show that magnetic properties of a linear, isotropic medium may
be characterized with frequency-dependent, complex permeability μ(ω).
Now rewriting Eq. (7.1) for the complex amplitudes of the fields at a particular
frequency, we may repeat all calculations of section 7.1, and verify that all its results
are valid for monochromatic waves even for a dispersive (but necessarily linear!)
medium. In particular, Eqs. (7.7) and (7.13) now become
⎛ μ(ω) ⎞1/2
Z (ω ) = ⎜ ⎟ , k (ω) = ω[ε(ω)μ(ω)]1/2 , (7.28)
⎝ ε( ω ) ⎠
so that the wave impedance and the wave number may be both complex functions of
frequency.
This fact has important consequences for the electromagnetic wave propagation.
First, plugging the representation of the complex wave number as the sum of its real
and imaginary parts, k(ω) ≡ k′(ω) + ikʺ(ω), into Eq. (7.11),
f = Re {fω e i [k(ω)z−ωt ]} = e−k ″(ω)z Re {fω e i [k ′(ω)z−ωt ]}, (7.29)
we see that kʺ(ω) describes the rate of wave attenuation in the medium at frequency
ω.9 Second, if the waveform is not sinusoidal (and hence should be represented as a
sum of several/many sinusoidal components), the frequency dependence of k′(ω)
provides for wave dispersion, i.e. the waveform deformation at the propagation,
because the propagation velocity (7.4) of component waves is now different10.
9
It may be tempting to attribute this effect to wave absorption, i.e. the dissipation of the wave’s energy, but we
will see very soon that wave attenuation may also be due to different effects.
10
The reader is probably familiar with the most noticeable effect of the dispersion, namely the difference
between that group velocity vgr ≡ dω/dk′, giving the speed of the envelope of a wave packet with a narrow
frequency spectrum, and the phase velocity vph ≡ ω/k′ of the component waves. The second-order dispersion
effect, proportional to d2ω/d2k′, leads to the deformation (gradual broadening) of the envelope itself.
Following tradition, these effects are discussed in more detail in the quantum-mechanics part of this series
(Part QM section 2.1), because they are the crucial factor of Schrödinger’s wave mechanics. (See also their
brief discussion in Part CM section 6.3.)
7-9
As an example of such a dispersive medium, let us consider a simple but very

representative Lorentz oscillator model11. In dilute atomic or molecular systems (e.g.
gases), electrons respond to the external electric field particularly strongly when
frequency ω is close to certain eigenfrequencies ωj corresponding to the spectrum of
quantum transitions of a single atom/molecule. An approximate, phenomenological
description of this behavior may be obtained from a classical model of several
externally driven harmonic oscillators, generally with non-zero damping. For a
single oscillator, driven by electric field’s force F(t) = qE(t), we can write Newton’s
second law as
m( x ̈ + 2δ0x ̇ + ω 02x ) = qE (t ), (7.30)
where ω0 is the own frequency of the oscillator, and δ0 its damping coefficient. For
the electric field of a monochromatic wave12, E(t) = Re[Eωexp{−iωt}], we may look
for a particular, forced-oscillation solution of this equation in a similar form x(t) =
Re[xωexp{−iωt}].13 Plugging this solution into Eq. (7.30), we can readily find the
complex amplitude of these oscillations:
q Eω
xω = 2
. (7.31)
m (ω 0 − ω 2 ) − 2iωδ0
Using this result to calculate the complex amplitude of the dipole moment as pω =
qxω, and then the electric polarization Pω = npω of a dilute medium with n
independent oscillators for unit volume, for its frequency-dependent permittivity
(7.26) we obtain
nq 2 1
ε( ω ) = ε0 + . (7.32)
(
2
m ω 0 − ω − 2iωδ0
2
)
This result may be readily (and obviously) generalized to the case when the system
has several types of oscillators with different eigenfrequencies:
q2 fj
ε( ω ) = ε0 + n ∑
m j (ω j − ω 2 ) − 2iωδj
2
, (7.33)
where fj ≡ nj/n is the fraction of oscillators with eigenfrequency ωj, so that the sum of
all fj equals 1. Figure 7.5 shows a typical behavior of the real and imaginary parts of
the complex dielectric constant, described by Eq. (7.33), as functions of frequency.
11
This example is focused on the frequency dependence of ε, because electromagnetic waves interact with
‘usual’ media via their electric field much more than via the magnetic field. However, as will be discussed in
section 7.6, forgetting about the possible dispersion of μ(ω) might result in missing some remarkable
opportunities for manipulating the waves.
12
According to Eq. (7.7), the magnetic field of the wave B = μH =E/v ∼ E/c, so that the magnetic component of
the Lorentz force (5.10), acting on a non-relativistic partice (Fm ∼ quB ∼ (u/c)qE), is much smaller than that of
its electric component (Fe = qE), and may be neglected.
13
If this point is not absolutely clear, please see Part CM section 5.1 for a more detailed discussion.
7-10
Figure 7.5. A typical frequency dependence of the real and imaginary parts of the electric permittivity of a
complex dielectric constant according to the generalized Lorentz oscillator model.
The oscillator resonances’ effect is clearly visible and dominates the medium’s
response at ω ≈ ωj, in particular in the case of low damping, δj ≪ ωj. Note that in the
low-damping limit, the imaginary part of the dielectric constant εʺ, and hence the
wave attenuation kʺ, are negligibly small at all frequencies apart from small vicinities
of frequencies ωj, where derivative dε′(ω)/dω is negative14. Thus, for a system of
weakly damped oscillators, Eq. (7.33) may be approximated at most frequencies as a
sum of odd singularities (‘poles’):
q2 fj
ε( ω ) ≈ ε0 + n ∑
2m j ωj − ω
, for δj ≪ ∣ω − ωj ∣ ≪ ∣ωj − ωj ′∣ . (7.34)
This result is particularly important because, according to quantum mechanics15,

Eq. (7.34) is also valid for a set of non-interacting, similar quantum systems (whose
dynamics may be completely different from that of a harmonic oscillator!), provided
that ωj are replaced with frequencies of possible quantum interstate transitions,
and coefficients fj are replaced with the so-called oscillator strengths of the
transitions—which obey the same sum rule, Σj fj = 1. At ω → 0, the imaginary
part of the permittivity also vanishes (for any δj), while its real part approaches its
electrostatic (‘dc’) value
nj
ε(0) = ε0 + q 2∑ . (7.35)
mω2 j j j
Note that according to Eq. (7.30), the denominator in Eq. (7.35) is just the effective
spring constant κj = mjωj2 of the jth oscillator, so that the oscillator masses mj as such
are actually (and quite naturally) not involved in the static dielectric response.
14
In optics, such behavior is called the anomalous dispersion.
15
See, e.g. Part QM chapters 5–7.
7-11
In the opposite limit of very high frequencies, ω ≫ ωj, δj, the permittivity (7.33)
also becomes real and may be represented as
⎛ ωp2 ⎞ q2 nj
ε(ω) = ε0⎜1 − 2 ⎟⎟ ,
⎜ where ωp2 ≡ ∑ . (7.36)
⎝ ω ⎠ ε0 j mj
The last result is very important because it is also valid at all frequencies if all ωj and
δj vanish, i.e. for a gas of free charged particles, in particular for plasmas—ionized
atomic gases, provided that the ion collision effects are negligible. (This is why the
parameter ωp defined by Eq. (7.36) is called the plasma frequency.) Typically, the
plasma as a whole is neutral, i.e. the density n of positive atomic ions is equal to that
of the free electrons. Since the ratio nj/mj for electrons is much higher than that for
ions, the general formula (7.36) for the plasma frequency is usually well approxi-
mated by the following simple expression:
ne 2
ωp2 ≡ . (7.37)
ε0m e
This expression has a simple physical sense: the effective spring constant κef =
meωp2 = ne2/ε0 describes the Coulomb force that appears when the electron
subsystem of a plasma is shifted, as a whole, from its positive-ion subsystem, thus
violating electroneutrality. (Indeed, let us consider such a small shift, Δx, perpen-
dicular to the plane surface of a broad, plane slab filled with a plasma. The
uncompensated ion charges, with equal and opposite surface densities σ = ±enΔx,
that appear at the slab surfaces, create inside it, according to Eq. (2.3), a uniform
electric field with Ex = enΔx/ε0. This field exerts force −eE = −(ne2/ε0)Δx = −κefΔx
on each electron, pulling it back to its equilibrium position.) Hence, there is no
surprise that the function ε(ω) vanishes at ω = ωp: at this resonance frequency, the
electric field E may oscillate, i.e. have a non-zero amplitude Eω = Dω/ε(ω), even in
the absence of external forces induced by external (stand-alone) charges, i.e. in the
absence of the field D these charges induce—see Eq. (3.32).
The behavior of electromagnetic waves in a medium that obeys Eq. (7.36), is very
remarkable. If the wave frequency ω is above ωp, the dielectric constant ε(ω), and
hence the wave number (7.28) are positive and real, and waves propagate without
attenuation, following the dispersion relation,
1 2 1/2
k (ω) = ω[ε(ω)μ0 ]1/2 =
c
(ω − ωp2 ) , (7.38)
which is shown in figure 7.6. At ω → ωp the wave number k tends to zero. Beyond
that point (at ω < ωp), we still can use Eq. (7.38), but it is instrumental to rewrite it in
the mathematically equivalent form
i 2 1/2 i c
k (ω ) = (ωp − ω 2 ) = , where δ ≡ 1/2
. (7.39)
c δ (ωp2 − ω2 )
7-12
Figure 7.6. The plasma dispersion law (solid line) in comparison with the linear dispersion of the free space
(dashed line).
According to Eq. (7.29), this means that the electromagnetic field exponentially
decreases with distance:
f = Re fω e i (kz−ωt ) ≡ exp { }−
z
δ
Re fω e−iωt . (7.40)
Does this mean that the wave is being absorbed in the plasma? Answering this
question is a good pretext to calculate the time average of the Poynting vector
S = E × H of a monochromatic electromagnetic wave in an arbitrary dispersive (but
still linear!) medium. First, let us spell out the fields’ time dependences:
1
E (t ) = Re [Eωe−iωt ] ≡ [Eωe−iωt + c.c.] ,
2
(7.41)
1⎡ E ⎤
H (t ) = Re [Hωe−iωt ] ≡ ⎢ ω e−iωt + c.c.⎥.
2 ⎣ Z (ω ) ⎦
Now, a straightforward calculation yields16
EωE ω* ⎡ 1 1 ⎤ EωE ω* 1 E 2 ⎡ ε(ω ) ⎤1/2

S = E (t )H (t ) = ⎢ + ⎥≡ Re ≡ ω Re ⎢ ⎥ . (7.42)
4 ⎣ Z (ω ) Z*(ω ) ⎦ 2 Z (ω ) 2 ⎣ μ( ω ) ⎦
Let us apply this important general formula to our simple model of plasma at ω <
ωp. In this case the magnetic permeability equals μ0, i.e. μ(ω) = μ0 is positive and
16
For an arbitrary plane wave the total average power flow may be calculated as an integral of Eq. (7.42) over
all frequencies. By the way, combining this integral and the Poynting theorem (6.111), is it straightforward to
prove the following interesting expression for the average electromagnetic energy density of a narrow (Δω ≪
ω) wave packet propagating in an arbitrary dispersive (but linear and isotropic) medium:
u =
1
∫
2 packet { d [ωε′(ω)]
dω
EωE ω* +
d [ωμ′(ω)]
dω }
HωHω* dω
7-13
real, while ε(ω) is real and negative, so that 1/Z(ω) = [ε(ω)/μ(ω)]1/2 is purely
imaginary, and the average Poynting vector (7.42) vanishes. This means that energy,
on average, does not flow along axis z. However, this does not mean that the waves
with ω < ωp are absorbed in the plasma. (Indeed, the Lorentz model with δj = 0 does
not describe any energy dissipation mechanism.) Instead, as we will see in the next
section, the waves are instead reflected from the plasma’s boundary.
Note also that in the limit ω ≪ ωp, Eq. (7.39) yields
c ⎛ c 2ε m ⎞1/2 ⎛ m e ⎞1/2
δ→ = ⎜ 02 e ⎟ = ⎜ ⎟ . (7.43)
ωp ⎝ ne ⎠ ⎝ μ0ne 2 ⎠
But this is just a particular case (for q = e, m = me, and μ = μ0) of the expression
(6.44), which was derived in section 6.4 for the depth of the magnetic field’s
penetration into a lossless (collision-free) conductor in the quasi-static approxima-
tion. This fact shows again that, as was already discussed in section 6.7, that this
approximation (in which the displacement currents are neglected) gives an adequate
description of the time-dependent phenomena at ω ≪ ωp, i.e. at δ ≪ c/ω = 1/k =
λ/2π.17
There are two very important examples of plasmas. For the Earth’s ionosphere,
i.e. the upper part of its atmosphere, which is almost completely ionized by the UV
and x-ray components of the Sun’s radiation, the maximum value of n, reached at
about 300 km over the Earth’s surface, is between 1010 and 1012 m−3 (depending on
the time of the day and the Sun’s activity phase), so that that the maximum plasma
frequency (7.37) is between 1 and 10 MHz. This is much higher than the particles’
typical reciprocal collision time τ−1, so that the first of Eq. (7.36) gives a good
description of the plasma’s electric polarization. The effect of the reflection of waves
with ω < ωp from the ionosphere enables long-range (over-the-globe) radio
communications and broadcasting at the so-called short waves, with frequencies of
the order of 10 MHz: they may propagate in the flat channel formed by the Earth’s
surface and the ionosphere, reflected repeatedly by these ‘walls’. Unfortunately, due
to the random variations of the Sun’s activity, and hence ωp, such a natural
communication channel is not very reliable and, in our age of trans-world fiber-optic
cables, its practical importance has diminished.
Another important example of plasmas is free electrons in metals and other
conductors. For a typical metal, n is of the order of 1023 cm−3 ≡ 1029 m−3, so that
(7.37) yields ωp ∼ 1016 s−1. Note that this value of ωp is somewhat higher than the
mid-optical frequencies (ω ∼ 3 × 1015 s−1). This explains why planar, clean metallic
surfaces, such as the aluminum and silver films used in mirrors, are so shiny: at these
frequencies their complex permittivity ε(ω) is almost exactly real and negative,
17
One more convenience of the simple model of a collision-free plasma, which has led us to Eq. (7.36), is that it
may be readily generalized to the case of an additional strong dc magnetic field B0 (much higher that that of the
wave) applied in the direction n of wave propagation. It is straightforward (and hence left for the reader) to
show that such plasma exhibits the Faraday effect of the polarization plane’s rotation, and hence gives an
example of an anisotropic media that violates the Lorentz reciprocity relation (6.121).
7-14
leading to light reflection with very little absorption. However, the simple model
(7.36), which neglects electron scattering, becomes inadequate at lower frequencies,
ωτ ∼ 1.
A phenomenological way of extending the model to account for scattering is to
take, in Eq. (7.33), the lowest frequency ωj to be equal zero (to describe the free
electrons), while keeping the damping coefficient δ0 of this mode finite to account for
their energy loss due to scattering. Then Eq. (7.33) is reduced to
n 0q 2 1 i n 0q 2 1
εef (ω) = εopt(ω) + 2
= εopt(ω) + , (7.44)
m −ω − 2iωδ 0 ω 2δ 0m 1 − iω /2δ 0
where the response εopt(ω) at high (in practice, optical) frequencies is still given by
Eq. (7.33), but now with j ≠ 0. The result (7.44) allows for a simple interpretation. To
show that, let us incorporate into our calculations the Ohmic conduction of the
medium, generalizing Eq. (4.7) as jω = σ(ω)Eω to account for the possible frequency
dependence of the Ohmic conductivity. Plugging this relation into the Fourier image
of the relevant macroscopic Maxwell equation, ∇ × Hω = jω − iωDω ≡ jω − iωε(ω)Eω,
we obtain
∇ × H ω = [σ (ω) − iωε(ω)]Eω. (7.45)
This relation shows that for a sinusoidal process, the addition of the Ohmic current
density jω to the displacement current density is equivalent to the addition of σ(ω) to
−iωε(ω), i.e. to the following change of the ac electric permittivity18:
σ (ω )
ε(ω) → εef (ω) ≡ εopt(ω) + i . (7.46)
ω
Now the comparison of Eqs. (7.44) and (7.46) shows that they coincide if we take
n 0q 2τ 1 1
σ (ω ) = = σ (0) , (7.47)
m 0 1 − iωτ 1 − iωτ
where the dc conductivity σ(0) is described by the Drude formula (4.13) and the
phenomenologically introduced coefficient δ0 is associated with 1/2τ. Relation (7.47),
which is frequently called the generalized (or ‘ac’, or ‘rf’) Drude formula19, gives a
very reasonable (semi-quantitative) description of the ac conductivity of many
metals almost all the way up to optical frequencies.
Now returning to the discussion of the generalized Lorentz model (7.33), we see
that the frequency dependences of the real (ε′) and imaginary (εʺ) parts of the
complex permittivity it yields are not quite independent. For example, let us have
one more look at the resonance peaks in figure 7.5. Each time the real part drops
18
Alternatively, according to Eq. (7.45), it is possible (and in the field of infrared spectroscopy, conventional)
to attribute the ac response of a medium at all frequencies to an effective complex conductivity: σef (ω) ≡ σ(ω) −
iωε(ω) ≡ −iωεef(ω).
19
It may also be derived from the Boltzmann kinetic equation in the so-called relaxation-time approximation
(RTA)—see, e.g. Part SM section 6.2.
7-15
with frequency, dε′/dω < 0, its imaginary part εʺ has a positive peak. R de L Kronig
(in 1926) and H A Kramers (in 1927) independently showed that this is not an
occasional coincidence pertinent only to this particular model. Moreover, full
knowledge of the function ε′(ω) allows one to calculate the function εʺ(ω) and
vice versa. The reason is that both these functions are always related to a single real
function G(θ) by Eq. (7.28).
To derive the Kramers–Kronig relations, let us consider Eq. (7.26b) on the
complex frequency plane, ω → ω ≡ ω′ + iωʺ:
∞ ∞
f (ω) ≡ ε(ω) − ε0 = ∫0 G (θ )e i ωθ dθ ≡ ∫0 G (θ )e iω′ θe−ω″ θ dθ . (7.48)
For all stable physical systems, G(θ) has to be finite for all important values of the
real integration variable (θ > 0), and tend to zero at θ → 0 and θ → ∞. (Indeed,
according to Eq. (7.23), a non-vanishing G(0) would mean an instantaneous
response of the medium to the external force, while G(∞) ≠ 0 would mean that is
has an infinitely long memory.) Because of that, and thanks to factor e−ωʺθ, the
expression under the integral tends to zero at ∣ω∣ → ∞ in all the upper half-plane
(ωʺ ⩾ 0). As a result, we may claim that the complex function f(ω), defined by
Eq. (7.48), is analytical in that half-plane. This fact allows us to use for it the general
Cauchy integral formula20
1 dΩ
f (ω) =
2πi C
f (Ω)
Ω−ω
∮ , (7.49)
where Ω ≡ Ω′ + iΩʺ, with Ω′ ≡ Ω, is a complex variable, for the case when ω = ω + i0

≡ ω is real. Let us take the integration contour shown in figure 7.7, with the radius R
of the larger semicircle tending to infinity, and the radius r of the smaller semicircle
(around the singular point Ω = ω) tending to zero.
Figure 7.7. Deriving the Kramers–Kronig dispersion relations.
Due to the exponential decay of ‫׀‬f(Ω)‫ ׀‬at ‫׀‬Ω‫∞ → ׀‬, the contribution to the right-
hand side of Eq. (7.49) from the larger semicircle vanishes21, while the contribution
from the small semicircle, where Ω = ω + rexp{iφ} with −π ⩽ φ ⩽ 0, is
20
See, e.g. Eq. (A.92).
21
Strictly speaking, this also requires ∣f(Ω)∣ to decrease faster than Ω−1 at the real axis (at Ω″ = 0), but due to
the inertia of charged particles, this requirement is fulfilled for all realistic models of dispersion—see, e.g.
Eq. (7.36).
7-16
0
1
lim
r → 0 2πi
∫Ω=ω+r exp{iφ) f (Ω) Ωd−Ωω = f2(πωi ) ∫−π ir rexp{ iφ}dφ
exp{iφ}
(7.50)
0
f (ω ) 1
≡
2π
∫
−π
dφ ≡ f (ω).
2
As a result, for our contour C, Eq. (7.49) yields
1 ⎛ ω−r +∞ ⎞ dΩ 1
f (ω) = lim ∫−∞
⎜
r → 0 2πi ⎝
+ ∫ω+r ⎟f (Ω)
⎠ Ω−ω
+ f (ω).
2
(7.51)
Such an integral, excluding a symmetric infinitesimal vicinity of a pole singularity, is

called the principal value of the (formally, diverging) integral from −∞ to +∞, and is
denoted by letter P before it22. Using this notation, subtracting f(ω)/2 from both
parts of Eq. (7.51), and multiplying them by 2, we obtain
+∞
1 dΩ
f (ω ) =
πi
P ∫−∞ f (Ω)
Ω−ω
. (7.52)
Now plugging into this complex equality the polarization-related difference f(ω)
≡ ε(ω) − ε0 in the form [ε′(ω) − ε0] + i[εʺ(ω)], and requiring both real and imaginary
components of two sides of Eq. (7.52) to be equal separately, we obtain the famous
Kramers–Kronig dispersion relations
+∞
1 dΩ
ε′(ω) = ε0 +
π
P ∫−∞Ω−ω
,ε″( Ω )
+∞
(7.53)
1 dΩ
ε″( ω ) = − P
π −∞
∫
[ε′(Ω) − ε0 ]
Ω−ω
.
We may use the already mentioned fact that ε′(ω) is always an even function, while
εʺ(ω) an odd function of frequency, to rewrite these relations in the following
equivalent form,
+∞
2 Ω dΩ
ε′(ω) = ε0 +
π
P ∫0Ω2 − ω 2
ε″( Ω )
,
+∞
(7.54)
2ω dΩ
ε″( ω ) = − P
π 0
∫
[ε′(Ω) − ε0 ] 2
Ω − ω2
,
which is more convenient for most applications, because it involves only physical
(positive) frequencies.
Although the Kramers–Kronig relations are ‘global’ in frequency, in certain cases
they allow an approximate calculation of dispersion from experimental data for
absorption, collected even in a limited (‘local’) frequency range. Most importantly, if
a medium has a sharp absorption peak at some frequency ωj, we may describe it as
22
I am typesetting this symbol in a Roman (upright) font, to avoid its confusion with the medium’s
polarization.
7-17
ε″(ω) ≈ cδ(ω − ωj ) + a more smooth function of ω, (7.55)
and the first of Eq. (7.54) immediately gives

2c ωj
ε′(ω) ≈ ε0 + 2
+ another smooth function of ω, (7.56)
π ω j − ω2
thus predicting the anomalous dispersion near such a point. This calculation shows
that such behavior observed in the Lorentz oscillator model (see figure 7.5) is by no
means occasional or model-specific.
Let me emphasize again that the Kramers–Kronig relations (7.53) and (7.54) are
much more general than the Lorentz model (7.33), hinging only on the causal, linear
relation (7.21) between the polarization P(t) with the electric field E(t′). Hence, these
relations are also valid for the complex functions relating Fourier images of any
cause/effect-related pair of variables. In particular, at a measurement of any linear
response r(t) of any experimental sample to any external applied field f(t′), whatever
the nature of this response and the physics behind it, we may be confident that there
is a causal relation between the variables r and f, so that the complex function χ(ω) ≡
rω/fω does obey the Kramers–Kronig relations. However, it is still important to
remember that a linear relation between the Fourier amplitudes of two variables
does not necessarily imply the causal relationship between them23.
7.3 Reflection
The most important new effect arising in non-uniform media is the wave reflection. Let
us start its discussion from the simplest case of a plane electromagnetic wave that is
normally incident on a sharp interface between two uniform, linear, isotropic media.
As the simplest example, let us assume that one of the two media (say, located at
z > 0, see figure 7.8) cannot sustain any electric field at all24:
E z⩾0 = 0. (7.57)
This condition is evidently incompatible with the single traveling wave (7.5).
However, this solution may be readily corrected using the fact that the dispersion-
free 1D wave equation,
⎛ ∂2 1 ∂2 ⎞
⎜ 2 − 2 2 ⎟E = 0, (7.58)
⎝ ∂z v ∂t ⎠
23
For example, the function φ(ω) ≡ Eω/Pω in the Lorentz oscillator model does not obey the Kramers–Kronig
relations. This is evident not only from the fact that E(t) is not a causal function of P(t), but even
mathematically. Indeed, the Green’s function describing a causal relationship has to tend to zero at small
time delays θ ≡ t − t′, so that its Fourier image has to tend to zero at ω → ± ∞. This is certainly true for the
function f(ω) given by Eq. (7.32), but not for the reciprocal function φ(ω) ≡ 1/f(ω) ∝ (ω2 − ω02) − 2iδω, which
diverges at large frequencies.
24
Such equality is given, in particular, by the macroscopic model of a good conductor—see Eq. (2.1).
7-18
Figure 7.8. A snapshot of the electric field at the reflection of a sinusoidal wave from a perfect conductor: a
realistic pattern (red lines) and its macroscopic, ideal-mirror approximation (blue lines). Dashed lines show the
snapshots after a half-period time delay (ωΔt = π).
supports waves propagating, with the same speed, in opposite directions. As a result,
the following linear superposition of two such waves,
E z⩽0 = f (z − vt ) − f ( −z − vt ), (7.59)
satisfies both the equation and the boundary condition (7.57) for an arbitrary function
f. The second term in Eq. (7.59) may be interpreted as the total reflection of the incident
wave (described by its first term)—in this particular case, with the change of the electric
field’s sign. This means, in particular, that within the macroscopic model a conductor
acts as a perfect mirror. By the way, since the vector n of the reflected wave is opposite
to that incident one (see the arrows in figure 7.8), Eq. (7.6) shows that the magnetic field
of the wave does not change its sign at the reflection:
1
H z⩽0 = [f (z − vt ) + f ( −z − vt )]. (7.60)
Z
The blue lines in figure 7.8 show the resulting pattern (7.59) for the simplest,
monochromatic wave:
E z⩽0 = Re [Eωe i (kz−ωt ) − Eωe i (−kz−ωt )]. (7.61a )
Depending on convenience in a particular context, this pattern may be legitimately

represented and interpreted either as the linear superposition (7.61a) of two traveling
waves, or as a single standing wave:
E z⩽0 = − 2Im(Eωe−iωt ) sin kz
(7.61b)
≡ 2 Re (iEωe−iωt ) sin kz ≡ 2 Re [Eωe−i (ωt−π /2)] sin kz ,
in which the electric and magnetic field oscillate with phase shifts of π/2 both in time
and space:
7-19
⎡E E ⎤ ⎛E ⎞
H z⩽0 = Re ⎢ ω e i (kz−ωt ) + ω e i (−kz−ωt )⎥ ≡ 2 Re ⎜ ω e−iωt ⎟ cos kz . (7.62)
⎣Z Z ⎦ ⎝Z ⎠
As the result of this shift, the time average of the Poynting vector’s magnitude,
1
S (z , t ) = EH = Re [Eω2e−2iωt ] sin 2kz , (7.63)
Z
equals zero, showing that at the total reflection there is no average power flow. (This
is natural, because a perfect mirror can neither transmit a wave nor absorb it.)
However, Eq. (7.63) shows that the standing wave provides local oscillations of
energy, transferring it periodically between the concentrations of the electric and
magnetic fields, separated by the distance Δz = π/2k = λ/4.
In the case of sinusoidal waves, the reflection effects may be readily explored even
for the more general case of dispersive and/or lossy (but still linear) media in which
ε(ω) and μ(ω), and hence the wave vector k(ω) and the wave impedance Z(ω),
defined by Eq. (7.28), are certain complex functions of frequency. The ‘only’ new
factors we have to account for is that in this case the reflection may not be full, and
that inside the second medium we have to use the traveling-wave solution as well.
Both these factors may be taken care of by looking for the solution of our boundary
problem in the form
E z⩽0 = Re [Eω(e ik−z + R e−ik−z ) e−iωt ] , E z⩾0 = Re [EωTe ik+ze−iωt ], (7.64)
and hence, according to Eq. (7.6),
⎡ E ⎤
H z⩽0 = Re ⎢ ω (e ik−z − R e−ik−z ) e−iωt ⎥ ,
⎣ Z−(ω) ⎦
(7.65)
⎡ Eω ⎤
H z⩾0 = Re ⎢ Te ik+ze−iωt ⎥ .
⎣ Z+(ω) ⎦
(Indices + and − correspond to, respectively, the media at z > 0 and z < 0.) Please
note the following important features of these relations:
(i) Due to the problem’s linearity, we could (and did) take the complex amplitudes
of the reflected and transmitted wave proportional to (Eω) of the incident wave,
scaling them with dimensionless, generally complex coefficients R and T. As the
comparison of Eqs. (7.64) and (7.65) with Eqs. (7.61) and (7.62) shows, the total
reflection from an ideal mirror, that was discussed above, corresponds to the
particular case R = −1 and T = 0.
(ii) Since the incident wave we are considering arrives from one side only (from
z = −∞), there is no need to include a term proportional to exp{−ik+z} into Eqs.
(7.64) and (7.65), in our current problem. However, we would need such a term if
the medium at z > 0 had been non-uniform (e.g. had at least one more interface or
7-20
any other inhomogeneity), because the wave reflected from that additional inho-
mogeneity would be incident on our interface (located at z = 0) from the right.
(iii) The solution (7.64) and (7.65) is sufficient even for the description of the cases when
waves cannot propagate at z ⩾ 0, for example a conductor or a plasma with ωp > ω.
Indeed, the exponential drop of the field amplitude at z > 0 in such cases is automatically
described by the imaginary part of the wave number k+—see Eq. (7.29).
In order to calculate the coefficients R and T, we need to use boundary conditions

at z = 0. Since the reflection does not change the transverse character of the partial
waves, at the normal incidence both vectors E and H remain tangential to the
interface plane (in our notation, z = 0). Reviewing the arguments that have led us, in
statics, to the boundary conditions (3.37) and (5.117) for these components, we see
that they remain valid for the time-dependent situation as well25, so that for our
current case of normal incidence we may write:
E z=−0 = E z=+0 , H z=−0 = H z=+0 . (7.66)
Plugging equations (7.64) and (7.65) into these conditions, we obtain two equations
for the coefficients R and T:
1 1
1 + R = T, (1 − R ) = T. (7.67)
Z− Z+
Solving this simple system of linear equations, we obtain26

Z+ − Z− 2Z+
R= , T= . (7.68)
Z+ + Z− Z+ + Z−
These formulas are very important, and much more general than one might think,
because they are applicable for virtually any 1D waves—electromagnetic or not—if
only the impedance Z is defined in a proper way27. Since in the general case the wave
impedances Z±, defined by Eq. (7.28) with the corresponding indices, are complex
functions of frequency, Eq. (7.68) show that coefficients R and T may have
imaginary parts as well. This fact has most important consequences at z < 0, where
the reflected wave, proportional to R, mixes (‘interferes’) with the incident wave.
25
For example, the first of conditions (7.66) may be obtained by integrating the full (time-dependent) Maxwell
equation ∇ × E + ∂B/∂t = 0 over a narrow and long rectangular contour with dimensions l and d (d ≪ l )
stretched along the interface. In the Stokes theorem, the first term gives ΔEτl, while the contribution of the
second term is proportional to product dl and vanishes as d/l → 0. The proof of the second boundary condition
is similar—as was already discussed in section 6.2.
26
Please note that only the medium impedances (rather than wave velocities) are important for the reflection in
this case! Unfortunately, this fact is not clearly emphasized in some textbooks that discuss only the case μ± =
μ0, when Z = (μ0/ε)1/2 and v = 1/(μ0ε)1/2 are proportional to each other.
27
See, e.g. the discussion of elastic waves of mechanical deformations in Part CM sections 6.3, 6.4, 7.7, and
7.8.
7-21
Indeed, with R = ∣R∣eiφ (where φ ≡ arg R is a real phase shift), the expression in
parentheses in the first of Eqs. (7.64) becomes
e ik−z + R e−ik−z = (1 − R + R )e ik−z + R e iφ e−ik−z
= (1 − R )e ik−z + 2 R e iφ /2 sin [k−(z − δ−)] , (7.69)
φ−π
where δ− ≡ .
2k−
This means that the field may be represented as a sum of a traveling wave and a
standing wave, with amplitude proportional to ∣R∣, shifted by distance δ− toward the
interface, relative to the ideal-mirror pattern (7.61b). This effect is frequently used
for the experimental measurement of an unknown impedance Z+ of some medium,
provided than Z− is known (e.g. for the free space, Z− = Z0). For that, a small
antenna (the probe), not disturbing the fields’ distribution too much, is placed into
the wave field, and the amplitude of the ac voltage induced in it by the wave in the
probe is measured by a detector (e.g. a semiconductor diode with a quadratic I–V
curve) as a function of z (figure 7.9). From the results of such a measurement, it is
straightforward to find both ∣R∣ and δ−, and hence restore the complex R, and then
use Eq. (7.68) to calculate both the modulus and the argument of Z+. (Before the
advent of computers, a specially lined paper, called the Smith chart, was commer-
cially available for performing this recalculation graphically; it is occasionally used
even now for the presentation of results.)
Figure 7.9. Measurement of the complex impedance of a medium (schematically).
Now let us discuss what these results give for waves incident from the free space
(Z−(ω) = Z0 = const, k− = k0 = ω/c) onto the surface of two particular, important
media.
(i) For a collision-free plasma (with negligible magnetization) we may use
Eq. (7.36) with μ(ω) = μ0, to represent the impedance in either of two equivalent forms:
ω ω
Z+ = Z0 1/2
≡ −iZ0 . (7.70)
(ω2 − ωp )
2
(ωp − ω2 )1/2
2
The former of these forms is more convenient in the case ω > ωp, when the wave
vector k+ and the wave impedance Z+ of the plasma are real, so that part of the
incident wave does propagate into the plasma. Plugging this expression into the
latter of Eqs. (7.68), we see that T is real:
7-22
2ω
T= 1/2
. (7.71)
ω + (ω 2 − ωp2 )
Note that according to this formula, and somewhat counter-intuitively, T > 1 for
any frequency (above ωp). How can the transmitted wave be more intense than the
incident one that has induced it? For a better understanding of this result, let us
compare the powers (rather than the electric field amplitudes) of these two waves,
i.e. their average Poynting vectors (7.42):
1/2
E 2 4ω(ω − ωp )
2 2
E 2 TEω 2
S¯incident = ω , S+ = = ω . (7.72)
2Z 0 2Z+ 2Z0 ⎡ω + ω 2 − ω 2 1/2 ⎤2
⎣ ( p) ⎦
It is easy to check that the ratio of these two values28 is always below 1 (and tends to
zero at ω → ωp), so that only a fraction of the incident wave power may be
transmitted. Hence the result T > 1 may be interpreted as follows: the interface
between two media also works as an impedance transformer. Although it can never
transmit more power than the incident wave provides (i.e. can only decrease the
product S = EH), since the ratio Z = E/H changes at the interface, the amplitude of
one of the fields may increase at the transmission.
Now let us proceed to case ω < ωp, when the waves cannot propagate in the
plasma. In this case, the latter of the expressions (7.70) is more convenient, because
it immediately shows that Z+ is purely imaginary, while Z− = Z0 is purely real. This
means that (Z+ − Z−) = (Z+ + Z−)*, i.e. according to the first equation of Eqs.
(7.68), ∣R∣ = 1, so that the reflection is total, i.e. no incident power (on average) is
transferred into the plasma—as was already discussed in section 7.2. However, the
complex R has a finite argument,
ω
φ ≡ arg R = 2 arg(Z+ − Z0) = −2 tan−1 , (7.73)
(ωp2 − ω2 )1/2
and hence provides a finite spatial shift (7.69) of the standing wave toward the
plasma surface:
φ−π c ω
δ− = = tan−1 . (7.74)
2k 0 ω (ωp2 − ω2 )1/2
On the other hand, we already know from Eq. (7.40) that the solution at z > 0 is
exponential, with the decay length δ that is described by Eq. (7.39). Calculating from
28
This ratio is sometimes also called the transmission coefficient, but in order to avoid its confusion with the T
defined by Eq. (7.64), it is better to call it the power transmission coefficient.
7-23
the coefficient T the exact coefficient before this exponent, it is straightforward to

verify that the electric and magnetic fields are indeed continuous at the interface,
forming the pattern shown with red lines in figure 7.8. This penetration may be
experimentally observed, for example, by bringing close to the interface another
material transparent as frequency ω. Even without solving this problem exactly, it is
evident that if the distance between these two interfaces becomes comparable to δ, a
part of the exponential ‘tail’ of the field is picked up by the second material, and
induces a propagating wave. This is an electromagnetic analog of the quantum-
mechanical tunneling through a potential barrier29.
Note that at low frequencies, both δ− and δ tend to the same frequency-
independent value,
c ⎛ c 2ε0m e ⎞1/2 ⎛ m e ⎞1/2 ω

δ , δ− → =⎜ ⎟ =⎜ ⎟ , at → 0, (7.75)
ωp ⎝ ne 2 ⎠ ⎝ μ0ne 2 ⎠ ωp
which is just the field penetration depth δ (6.44) calculated for a perfect conductor
model (assuming m = me and μ = μ0) in the quasi-static limit. This is natural, because
the condition ω ≪ ωp may be recast as λ0 ≡ 2πc/ω ≫ 2πc/ωp ≡ 2πδ, justifying the
quasi-static approximation.
(ii) Now let us consider the electromagnetic wave reflection from an Ohmic, non-
magnetic conductor. In the simplest low-frequency limit, when ωτ is much less
than 1, the conductor may be described by a frequency-independent conductivity
σ.30 According to Eq. (7.46), in this case we can take
⎛ μ0 ⎞1/2
Z+ = ⎜ ⎟ . (7.76)
⎝ εopt(ω) + iσ / ω ⎠
With this substitution, Eqs. (7.68) immediately give us all the results of interest. In
particular, in the most important quasi-static limit (when δs ≡ (2/μ0σω)1/2 ≪ λ0 ≡ 2πc/
ω, i.e. σ/ω ≫ ε0 ∼ εopt), the conductor’ impedance is low:
⎛ μ0ω ⎞1/2 ⎛ 2 ⎞1/2 δs Z+
Z+ ≈ ⎜ ⎟ ≡ π⎜ ⎟ Z, i.e. ≪ 1. (7.77)
⎝ iσ ⎠ ⎝ i ⎠ λ0 0 Z0
The impedance is complex, and hence some fraction f of the incident wave is
absorbed by the conductor. This fraction may be found as the ratio of the dissipated
power (either calculated, as was done above, from Eq. (7.68), or just taken from Eq.
(6.36) with the magnetic field amplitude ∣Hω∣ = 2∣Eω∣/Z0—see Eq. (7.62)) to the
incident wave’s power given by the first of Eq. (7.72). The result is
29
30
In a typical metal, τ ∼ 10−13 s, so that this approximation works well all the way up to ω ∼ 1013 s−1, i.e. up to
far-infrared frequencies.
7-24
2ωδs δ
f= ≡ 4π s ≪ 1. (7.78)
c λ0
This important result is widely used for crude estimates of the energy dissipation
in metallic-wall waveguides and resonators, and immediately shows that in order to
keep the energy losses low, the characteristic size of such systems (which gives a
scale of the free-space wavelengths λ0 at which they are used) should be much larger
than δs. A more detailed theory of these structures, and the effects of energy loss in
them, will be discussed later in this chapter.
7.4 Refraction
Now let us consider the effects arising at a plane interface between two uniform
media if the wave’s incidence angle θ (figure 7.10) is arbitrary (rather than equal to
zero as in our previous analysis), for the simplest case of fully transparent media
with real ε±(ω) and μ±(ω). (For the sake of notation simplicity, the argument of these
functions will be dropped, i.e. just implied in most formulas below.)
Figure 7.10. Plane wave reflection, transmission, and refraction at a plane interface. The plane of the drawing
is selected to contain all three wave vectors: k+, k−, and k′−.
In contrast with the case of normal incidence, here the wave vectors k-, k′−, and
k+ of the three component (incident, reflected, and transmitted) waves may have
different directions. (This change of the transmitted wave’s direction is called
refraction.) Hence now we have to start our analysis by writing a general expression
for a single-plane, monochromatic wave for the case when its wave vector k has
all three Cartesian components, rather than one. An evident generalization of
Eq. (7.11) for this case is
f (r , t ) = Re⎡⎣fω e i (kxx+kyy+kzz )−ωt )⎤⎦ = Re⎡⎣fω e i (k⋅r−ωt )⎤⎦ . (7.79)
This relation enables a ready analysis of ‘kinematic’ relations that are independ-
ent of the media impedances. Indeed, it is sufficient to notice that in order to satisfy
any linear, homogeneous boundary conditions at the interface (z = 0), all waves must
have the same temporal and spatial dependence on this plane. Hence if we select the
plane xz so that the vector k− lies in it, then (k−)y = 0, and k+ and k′− cannot have
7-25
any y-component either, i.e. all three vectors lie in the same plane—which is selected
as the plane of drawing in figure 7.10. Moreover, due to the same reason their
x-components should be equal:
k− sin θ = k −′ sin θ′ = k+ sin r . (7.80)
From here we immediately obtain the well-known laws of reflection

θ′ = θ , (7.81)
and refraction31
sin r k
= −. (7.82)
sin θ k+
In this form, the laws are valid for plane waves of any nature. In optics, the Snell law
(7.82) is frequently represented in the form
sin r n
= −, (7.83)
sin θ n+
where n± is the index of refraction, also called the ‘refractive index’ of the
corresponding medium, defined as its wave number normalized to that of the free
space (at the particular wave’s frequency):
k± ⎛ ε±μ± ⎞
1/2
n± ≡ =⎜ ⎟ . (7.84)
k0 ⎝ ε0μ0 ⎠
Perhaps the most famous corollary of the Snell law is that if a wave propagates
from a medium with a higher index of refraction to that with a lower one (i.e. if n− >
n+ in figure 7.10), for example from water into air, there is always a certain critical
value θc of the incidence angle,
n+ ⎛ ε+μ+ ⎞1/2
θc = sin ≡ sin ⎜
−1 −1
⎟ , (7.85)
n− ⎝ ε−μ− ⎠
at which the refraction angle r (see figure 7.10 again) reaches π/2. At a larger θ, i.e.
within the range θc < θ < π/2, the boundary conditions (7.80) cannot be satisfied by a
refracted wave with a real wave vector, so that the wave experiences so-called total
internal reflection. This effect is very important in practice, because it means that
dielectric surfaces may be used as optical mirrors, in particular in optical fibers—to
be discussed in more detail in section 7.7. This is very fortunate for telecommuni-
cation technology, because light’s reflection from metals is rather imperfect. Indeed,
31
This relation is traditionally called the Snell law, after seventeenth century author W Snellius, though it has
been traced all the way back to a circa 984 manuscript by Abu Saad al-Ala ibn Sahl.
7-26
according to Eq. (7.78), in the optical range (λ0 ∼ 0.5 μm, i.e. ω ∼ 1015 s−1), even the
best conductors (with σ ∼ 6 × 108 s m−1 and hence the normal skin depth δs ∼ 1.5
nm) suffer losses of at least a few percent at each reflection.
Note, however, that even within the range θc < θ < π/2, the field at z > 0 is not
identically equal to zero: it penetrates into the less dense medium by a distance of the
order of λ0, exponentially decaying inside it, just as it does at normal incidence—see
figure 7.8. However, at θ ≠ 0 the penetrating field still propagates, with the wave
number (7.80), along the interface. Such a field, exponentially dropping in one
direction but still propagating as a wave in another direction, is commonly called the
evanescent wave.
One more remark: just as at the normal incidence, the field’s penetration into
another medium causes a phase shift of the reflected wave—see, e.g. Eq. (7.69) and
its discussion. A new feature of this phase shift, arising at θ ≠ 0, is that it also has a
component parallel to the interface—the so-called called the Goos–Hänchen effect.
In the geometric optics, this effect leads to an image shift (relative to its position in a
perfect mirror) with components both normal and parallel to the interface.
Now let us carry out an analysis of the ‘dynamic’ relations that determine the
amplitudes of the refracted and reflected waves. For this we need to write explicitly
the boundary conditions at the interface (i.e. the plane z = 0). Since now the electric
and/or magnetic fields may have components normal to the plane, in addition to the
continuity of their tangential components, which were repeatedly discussed above,
Ex,y z=−0 = Ex,y z=+0 , Hx,y z=−0 = Hx,y z=+0 , (7.86)
we also need relations for the normal components. As follows from the homogeneous
macroscopic Maxwell equations (6.99b), these conditions are also the same as in statics,
i.e. Dn = const and Bn = const, for our reference frame choice (figure 7.10) giving
ε−Ez z=−0 = ε+Ez z=+0 , μ−Hz z=−0 = μ+Hz z=+0 . (7.87)
The expressions of these components via the amplitudes Eω, REω, and TEω of the
incident, reflected, and transmitted waves depend on the incident wave’s polar-
ization. For example, for a linearly polarized wave with the electric field vector
perpendicular to the plane of incidence, i.e. parallel to the interface plane, the
reflected and refracted waves are similarly polarized—see figure 7.11a. As a result,
all Ez are equal to zero (so that the first of Eq. (7.87) is inconsequential), while the
tangential components of the electric field are just equal to their full amplitudes, just
like at the normal incidence, so we still can use Eq. (7.64), expressing these
components via the coefficients R and T. However, at θ ≠ 0 the magnetic fields
have not only tangential components
⎡E ⎤ ⎡E ⎤
Hx = Re ⎢ ω (1 − R)cos θ e−iωt ⎥, Hx = Re ⎢ ω T cos r e−iωt ⎥, (7.88)
⎣ Z− ⎦ ⎣ Z+ ⎦
z=−0 z=+0
but also normal components (see figure 7.11a):
7-27
Figure 7.11. Reflection and refraction at two different linear polarizations of the incident wave.
⎡E ⎤ ⎡E ⎤
Hz = Re ⎢ ω (1 + R )sin θ e−iωt ⎥ , Hz = Re ⎢ ω T sin r e−iωt ⎥ . (7.89)
⎣ Z− ⎦ ⎣ Z+ ⎦
z=−0 z=+0
Plugging these expressions into the boundary conditions expressed by Eq. (7.86)
(in this case, for y components only) and the second of Eqs. (7.87), we obtain three
equations for two unknown coefficients R and T. However, two of these equations
duplicate each other because of the Snell law, and we obtain just two independent
equations,
1 1
1 + R = T, (1 − R) cos θ = T cos r , (7.90)
Z− Z+
which are a very natural generalization of Eq. (7.67), with replacements Z− →

Z−cos r, Z+ → Z+cos θ. As a result, we can immediately use Eqs. (7.68) to write the
solution of the system (7.90)32:
Z+ cos θ − Z− cos r 2Z+ cos θ
R= , T= . (7.91a )
Z+ cos θ + Z− cos r Z+ cos θ + Z− cos r
If we want to express these coefficients via the angle of incidence alone, we should
use the Snell law (7.82) to eliminate the angle r, obtaining the commonly used, more
bulky expressions:
Z+ cos θ − Z−[1 − (k−/ k+)2 sin2 θ ]1/2

R= ,
Z+ cos θ + Z−[1 − (k−/ k+)2 sin2 θ ]1/2
(7.91b)
2Z+ cos θ
T= .
Z+ cos θ + Z−[1 − (k−/ k+)2 sin2 θ ]1/2
32
Note that we may calculate the reflection and transmission coefficients R′ and T′ for the wave traveling in
the opposite direction just by making the following parameter swaps: Z+ ↔ Z− and θ ↔ r, and that the
resulting coefficients satisfy the following Stokes relations: R′ = −R, and R2 + TT′ = 1, for any Z±.
7-28
However, my strong preference is to use the kinematic relation (7.82) and the
dynamic relations (7.91a) separately, because Eq. (7.91b) obscures the very
important physical fact that and the ratio of k±, i.e. of the wave velocities of the
two media, is only involved in the Snell law, while the dynamic relations essentially
include only the ratio of wave impedances—just as in the case of normal incidence.
In the opposite case of linear polarization of the electric field within the plane of
incidence (figure 7.11b), it is the magnetic field which does not have a normal
component, so it is now the second of Eqs. (7.87) that does not participate in the
solution. However, now the electric fields in two media have not only tangential
components,
Ex z=−0 = Re [Eω(1 + R)cos θ e−iωt ] , Ex z=+0 = Re [EωT cos r e−iωt ], (7.92)
but also normal components (figure 7.11b):

Ez z=−0 = Eω( −1 + R )sin θ , Ez z=+0 = −EωT sin r . (7.93)
As a result, instead of Eqs. (7.90), the reflection and transmission coefficients are
related as
1 1
(1 + R)cos θ = T cos r , (1 − R ) = T. (7.94)
Z− Z+
Again, the solution of this system may be immediately written using the analogy
with Eqs. (7.67):
Z+ cos r − Z− cos θ 2Z+ cos θ

R= , T= , (7.95a )
Z+ cos r + Z− cos θ Z+ cos r + Z− cos θ
or, alternatively, using the Snell law, in a bulkier form:
Z+[1 − (k− / k+ )2 sin2 θ ]1/2 − Z− cos θ

R= ,
Z+[1 − (k− / k+ )2 sin2 θ ]1/2 + Z− cos θ
(7.95b)
2Z+ cos θ
T= .
Z+[1 − (k− / k+ )2 sin2 θ ]1/2 + Z− cos θ
For the particular case μ+ = μ− = μ0, when Z+/Z− = (ε−/ε+)1/2 = k−/k+ = n−/n+
(which is approximately correct for traditional optical media), Eqs. (7.91b) and
(7.95b) are called the Fresnel formulas33. Most textbooks are quick to point out that
there is a major difference between them: for the electric field polarization within the
33
After A-J Fresnel (1788–1827), one of the pioneers of wave optics, who is credited, among many other
contributions (see, in particular, the discussions in chapter 8), for the concept of light as a purely transverse
wave.
7-29
plane of incidence (figure 7.11b), the reflected wave amplitude (proportional to the
coefficient R) turns to zero34 at a special value of θ (called the Brewster angle)35
n
θ B = tan−1 + , (7.96)
n−
while there is no such angle in the opposite case (figure 7.11a). However, note that
this statement, as well as Eq. (7.96), is true only for the case μ+ = μ−. In the general
case of different ε and μ, Eqs. (7.91) and (7.95) show that the reflected wave vanishes
at θ = θB with
⎧
tan2 θ B =
ε−μ+ − ε+μ− ⎪
×⎨
(μ+ /μ−), for E⊥nz (figure 11a), (7.97)
⎪
ε+μ+ − ε−μ− ⎩( − ε+ / ε−), for H ⊥nz (figure 11b).
Note the natural ε ↔ μ symmetry of these relations, resulting from the E ↔ H

symmetry for these two polarization cases (figure 7.11). These formulas also show
that for any set of parameters of the two media (with ε±, μ± > 0), tan2θB is positive
(and hence a real Brewster angle θB exists) only for one of these two polarizations. In
particular, if the interface is due to the change of μ alone (i.e. if ε+ = ε−), the first of
Eq. (7.97) is reduced to the simple form (7.96) again, while for the polarization
shown in figure 7.11b there is no Brewster angle, i.e. the reflected wave has a non-
vanishing amplitude for any θ.
Such an account of both media parameters, ε and μ, on an equal footing is in
especially necessary to describe the so-called negative refraction effects36. As was
shown in section 7.2, in a medium with electric-field-driven resonances, the function
ε(ω) may be almost real and negative, at least within limited frequency intervals—
see, in particular, Eq. (7.34) and figure 7.5. As has already been discussed, if the
function μ(ω) is real and positive at these frequencies, then k2(ω) = ω2ε(ω)μ(ω) < 0
and k may be represented as i/δ with real δ, meaning that the exponential field decays
into the medium. However, let us consider the case when both ε(ω) < 0 and μ(ω) < 0
at a certain frequency. (This is evidently possible in a medium with both E-driven
and H-driven resonances, at a proper choice of their resonant frequencies.) Since in
this case k2(ω) = ω2ε(ω)μ(ω) > 0, the wave vector is real, so that Eq. (7.79) describes
a traveling wave, and one could think that there is nothing new in this case. Not
quite so!
34
This effect is used in practice to obtain linearly polarized light, with the electric field vector perpendicular to
the plane of incidence, from the natural light with its random polarization. An even more practical application
of the effect is a partial reduction of undesirable glare from wet surfaces (for the water/air interface, n+/n− ≈
1.33, giving θB ≈ 50°) by making car-light covers and sunglasses of vertically polarizing materials.
35
A very simple interpretation of Eq. (7.96) is based on the fact that, together with the Snell law (7.82), it gives
r + θ = π/2. As a result, the vector E+ is parallel to the vector k′−, and hence oscillating electric dipoles of the
medium at z > 0 do not have the component which could induce the transverse electric field E′− of the potential
reflected wave.
36
Despite some important background theoretical work by A Schuster (1904), L Mandelstam (1945),
D Sivikhin (1957), and in particular V Veselago (1966–67), the negative refractivity effects have only recently
become a subject of intensive scientific research and engineering development.
7-30
Figure 7.12. Directions of the main vectors of a plane wave inside a medium with (a) positive and (b) negative
values of ε and μ.
First of all, for a sinusoidal, plane wave (7.79), the operator ∇ is equivalent to the
multiplication by ik. As the Maxwell equations (7.2a) show, this means that at a
fixed direction of vectors E and k, the simultaneous reversal of the signs of ε and μ
means the reversal of the direction of the vector H. Namely, if both ε and μ are
positive, these equations are satisfied with mutually orthogonal vectors E, H, and k
forming the usual, right-hand system (see figure 7.1 and figure 7.12a), the name
stemming from the popular ‘right-hand rule’ used to determine the vector product’s
direction. However, if both ε and μ are negative, the vectors form a left-hand system
—see figure 7.12b. (Due to this fact, the media with ε < 0 and μ < 0 are frequently
called left-handed materials, LHM for short.) According to the fundamental relation
(6.114), which does not involve media parameters, this means that for a plane wave
in a left-hand material, the Poynting vector S = E × H, i.e. the energy flow, is
directed opposite to the wave vector k.
This fact may look strange, but is in no contradiction with any fundamental
principle. Let me remind you that, according to the definition of vector k, its
direction shows the direction of the phase velocity vph = ω/k of a sinusoidal (and
hence infinitely long) wave that cannot be used, for example, for signaling. Such
signaling (by sending wave packets—see figure 7.13) is possible with the group
velocity vgr = dω/dk. This velocity in left-hand materials is always positive (directed
along the vector S).
Figure 7.13. An example of a wave packet moving along axis z with a negative phase velocity, but positive
group velocity. Blue lines show a packet snapshot a short time interval after the first snapshot (red lines).
Maybe the most fascinating effect possible with left-hand materials is the wave
refraction at their interfaces with the usual, right-handed materials—first predicted
by V Veselago in 1960. Consider the example shown in figure 7.14a. In the incident
7-31
Figure 7.14. Negative refraction: (a) waves at the interface between media with positive and negative values of
εμ, and (b) the hypothetical perfect lens: a parallel plate made of a material with ε = −ε0 and μ = −μ0.
wave, coming from the usual material, the directions of vectors k− and S− coincide,
and so they are in the reflected wave with vectors k′− and S′−. This means that the
electric and magnetic fields in the interface plane (z = 0) are, at our choice of
coordinates, proportional to exp{ikxx} with a positive component kx = k−cos θ. In
order to satisfy any linear boundary conditions, the refracted wave, going into
the left-handed material, should match that dependence, i.e. have a positive
x-component of its wave vector k+. But in this medium, this vector has to be
antiparallel to the vector S that, in turn, should be directed out of the interface,
because it represents the power flow from the interface into the material bulk. These
conditions cannot be reconciled by the refracted wave propagating along the usual
Snell-law direction (shown with the dashed line in figure 7.14a), but are all satisfied
at refraction in the direction given by Snell’s angle with the opposite sign. (Hence the
term ‘negative refraction’)37.
In order to understand how unusual the results of the negative refraction may be,
let us consider a parallel slab of thickness d, made of a hypothetical left-handed
material with exactly selected values ε = −ε0 and μ = −μ0—see figure 7.14b. For such
a material, placed in free space, the refraction angle r = −θ, so that the rays from a
point source located in free space, at a distance a < d from the slab, propagate as
shown in the figure, i.e. all meet again at the distance a inside the plate and then
continue to propagate to the second surface of the slab. Repeating our discussion for
this surface, we see that a point’s image is also formed beyond the plate at distance
2a + 2b = 2a + 2(d − a) = 2d from the object.
Superficially, this system looks like the usual lens, but the well-known lens
formula, which relates a and b with the focal length f, is not satisfied. (In particular, a
parallel beam is not focused into a point at any finite distance.) As an additional
difference to the usual lens, the system shown in figure 7.14b does not reflect any part
of the incident light. Indeed, it is straightforward to check that in order for all above
formulas for R and T to be valid, the sign of the wave impedance Z in left-handed
37
Inspired by this fact, in some publications the left-handed materials are prescribed a negative index of
refraction n. However, this prescription should be treated with care (for example, it complies with the first form
of Eq. (7.84), but not its second form), and the sign of n, in contrast to that of the wave vector k, is the matter
of convention.
7-32
materials has to be kept positive. Thus, for our particular choice of parameters (ε =
−ε0, μ = −μ0), Eqs. (7.91a) and (7.95a) are valid with Z+ = Z− = Z0 and cos r = cos θ
= 1, giving R = 0 for any linear polarization, and hence for any other wave
polarization—circular, elliptic, natural, etc.
The perfect lens suggestion has triggered a wave of efforts to implement left-
handed materials experimentally. (Attempts to find such materials in nature have
failed so far.) Most progress in this direction has been achieved using the so-called
metamaterials, which are essentially quasi-periodic arrays of specially designed
electromagnetic resonators, ideally with high density n ≫ λ−3. For example, figure
7.15 shows the metamaterial that was used for the first demonstration of negative
refractivity in the microwave region, i.e. a few GHz frequencies38. It combines
straight strips of a metallic film, working as lumped resonators with a large electric
dipole moment (hence strongly coupled to the wave’s electric field E), and several
almost-closed film loops (so-called split rings), working as lumped resonators with
large magnetic dipole moments, coupled to the field H. The negative refractivity is
achieved by designing the resonance frequencies close to each other. More recently,
metamaterials with negative refractivity were demonstrated in the optical range39,
although to the best of my knowledge, their relatively large absorption still prevents
practical applications.
Figure 7.15. An artificial left-handed material providing negative refraction at microwave frequencies (∼10
GHz). The original image by Jeffrey D Wilson (in the public domain) is available at https://en.wikipedia.org/
wiki/Metamaterial.
This progress has stimulated the development of other potential uses of

metamaterials (not necessarily the left-handed ones), in particular designs of non-
uniform systems with engineered distributions ε(r, ω) and μ(r, ω), which may provide
electromagnetic wave propagation along desired paths, e.g. around a certain region
38
See [2, 3].
39
See, e.g. [4].
7-33
of space, making it virtually invisible for an external observer—so far, within a

limited frequency range40.
As was mentioned in section 5.5, another way to reach negative values of μ(ω) is
to place a ferromagnetic material into such an external dc magnetic field that the
frequency ωr of the ferromagnetic resonance is somewhat lower than ω. If thin layers
of such a material (e.g. nickel) are interleaved with layers of a non-magnetic, very
good conductor (such as copper), the resulting metamaterial has an average value of
μ(ω)—say, positive, but substantially below μ0. According to Eq. (6.33), the skin-
depth δs of such a material may be larger than that of the good conductor alone,
enforcing a more uniform distribution of the ac current flowing along the layers, and
hence making the energy losses lower than in the good conductor alone. This effect
may be useful, in particular, for electronic circuit interconnects41.
7.5 Transmission lines: TEM waves

So far, we have analyzed plane electromagnetic waves, implying that their cross-
section is infinite—evidently an unrealistic assumption. The cross-section may be
limited, still sustaining wave propagation, using wave transmission lines42: cylindri-
cally shaped structures made of either good conductors or dielectrics. Let us first
discuss the first option, using the following simplifying assumptions:
(i) the structure is a cylinder (not necessarily with a round cross-section, see figure
7.16) filled with a usual (right-handed), uniform dielectric material with negligible
energy losses (εʺ = μʺ = 0), and
(ii) the wave attenuation due to the skin effect is also negligibly low. (As Eq. (7.78)
indicates, for this to be the case the characteristic size a of the line’s cross-section has to
be much larger than the skin-depth δs of its wall material. The energy dissipation
effects will be analyzed in section 7.9 below.)
Figure 7.16. Electric field’s decomposition in a transmission line.
40
For a review of such ‘invisibility cloaks’, see, e.g. [5].
41
See, for example, [6] and references therein.
42
Another popular term is the waveguide, but it is typically reserved for transmission lines with a singly
connected cross-section, to be analyzed in the next section.
7-34
With such exclusion of energy losses, we may look for a particular solution of the
macroscopic Maxwell equations in the form of a monochromatic wave traveling
along the line:
E(r , t ) = Re [Eω(x , y )e i (kzz−ωt ) ], H(r , t ) = Re [H ω(x , y )e i (kzz−ωt )], (7.98)
with real kz, where the axis z is directed along the transmission line—see figure 7.16.
Note that this form allows an account for a substantial coordinate dependence of the
electric and magnetic field in the plane {x, y} of the transmission line’s cross-section,
as well as for longitudinal components of the fields, so that the solution (7.98) is
substantially more complex than the plane waves we have discussed above. We will
see in a minute that as a result of this dependence, the parameter kz may be very
different from its plane-wave value (7.13), k ≡ ω(εμ)1/2, in the same material, at the
same frequency.
In order to describe these effects qualitatively, let us decompose the complex
amplitudes of the wave’s fields into their longitudinal and transverse components
(figure 7.16)43
Eω = Ez nz + Et , H ω = Hz nz + Ht . (7.99)
Plugging Eqs. (7.98) and (7.99) into the source-free Maxwell equations (7.2), and
requiring the longitudinal and transverse components to be balanced separately, we
obtain
ikz nz × Et − iωμHt = −∇t × (Ez nz), ikz nz × Ht + iωε Et = −∇t × (Hz nz),
∇t × Et = iωμHz nz , ∇t × Ht = −iεωEz nz , (7.100)
∇t ⋅ Et = −ikzEz , ∇t ⋅ Ht = −ikzHz .
where ∇t is the 2D del operator acting in the transverse plane [x, y] only (i.e. the usual
∇, but with ∂/∂z = 0). These equations may look even more bulky than the original
equations (7.2), but actually are much simpler for analysis. Indeed, eliminating the
transverse components from these equations (or, even simpler, just plugging Eq. (7.99)
into Eq. (7.3) and keeping only their z-components), we obtain a pair of self-consistent
equations for the longitudinal components of the fields44,
(∇ 2
t )
+ kt2 Ez = 0, (∇ 2
t )
+ kt2 Hz = 0, (7.101)
where k is still defined by Eq. (7.13), k = (εμ)1/2ω, while

kt2 ≡ k 2 − k z2 = ω 2εμ − k z2. (7.102)
After the distributions Ez(x,y) and Hz(x,y) have been found from these equations,
they provide the right-hand sides for rather simple, closed system of Eq. (7.100) for
43
Note that for notation simplicity, I am dropping index ω in the complex amplitudes of the field components,
and also have dropped the argument ω in kz and Z, although these parameters of the wave may depend on its
frequency rather substantially—see below.
44
The wave equation represented in the form (7.101), even with the 3D Laplace operator, is called the Helmholtz
equation, after H von Helmholtz (1821–94)—the mentor of H Hertz and M Planck, among many others.
7-35
the transverse components of field vectors. Moreover, as we will see below, each of
the following three types of solutions,
(i) with Ez = 0 and Hz = 0 (called transverse, or TEM waves),
(ii) with Ez = 0, but Hz ≠ 0 (called either TE waves or, more frequently, H-modes),
and
(iii) with Ez ≠ 0, but Hz = 0 (so-called TM waves or E-modes)
has its own dispersion law and hence its own wave propagation velocity; as a result,
these modes (i.e. the field distribution patterns) may be considered separately.
In this section we will focus on the simplest, TEM waves, with no longitudinal
components of either field. For them, the top two equations of the system (7.100)
immediately give Eqs. (7.6) and (7.13), and kz = k. In plain English, this means that
E = Et and H = Ht are proportional to each other and mutually perpendicular (just
as in the plane wave) at each point of the cross-section, and that the TEM wave’s
impedance Z ≡ E/H and dispersion law ω(k), and hence the propagation speed, are
the same as in a plane wave in the same material. In particular, if ε and μ are
frequency-independent within a certain frequency range, the dispersion law is linear,
ω = k/(εμ)1/2, and the wave’s speed does not depend on its frequency. For practical
applications to telecommunications, this is a very important advantage of TEM
waves over their TM and TE counterparts—to be discussed below.
Unfortunately, such waves cannot propagate in every transmission line. In order
to show this, let us have a look at the two last lines of Eq. (7.100). For the TEM
waves (Ez = 0, Hz = 0, kz = k), they yield
∇t × Et = 0, ∇t × Ht = 0,
(7.103)
∇t ⋅ Et = 0, ∇t ⋅ Ht = 0.
In the macroscopic approximation for the conducting walls of the line (i.e.
completely neglecting the skin effect), we have to require that inside them, E =
H = 0. Close to a wall but outside it, the normal component En of the electric field
may be different from zero, because surface charges may sustain its jump—see
section 2.1, in particular Eq. (2.3). Similarly, the tangential component Hτ of the
magnetic field may have a finite jump at the surface due to skin currents—see section
6.3, in particular Eq. (6.38). However, the tangential component of the electric field
and the normal component of magnetic field cannot experience such a jump, and in
order to have them equal to zero inside the walls they have to equal zero outside the
inside the walls as well:
E τ = 0, Hn = 0. (7.104)
But the left columns of Eqs. (7.103) and (7.104) coincide with the formulation of
the 2D boundary problem of electrostatics for the electric field induced by electric
charges of the conducting walls, with the only difference that in our current case the
value of ε actually means ε(ω). Similarly, the right columns of those relations
coincide with the formulation of the 2D boundary problem of magnetostatics for the
7-36
magnetic field induced by currents in the walls, with μ → μ(ω). The only difference is
that in our current ac case the magnetic fields cannot penetrate inside the
conductors.
Now we immediately see that in waveguides with a singly connected wall, for
example a hollow conducting tube (see, e.g. the example shown in figure 7.16), TEM
waves are impossible because there is no way to create a finite electrostatic field
inside a conductor with such cross-section. However, such fields (and hence TEM
waves) are possible in structures with cross-sections consisting of two or more
disconnected (galvanically insulated) parts—see, e.g. Figure 7.17.
Figure 7.17. An example of the cross-section of a transmission line that may support TEM wave propagation.
In order to derive ‘global’ relations for such a line, let us consider the contour C
drawn very close to the surface of one of its conductors—see, e.g. the red dashed line
in figure 7.17. We can consider it, on one hand, as the cross-section of a cylindrical
Gaussian volume of a certain elementary length dz ≪ λ ≡2π/k. Using the generalized
Gauss law (3.34), we obtain
∮C (Et)n dr = λεω , (7.105)
where λω (not to be confused with the wavelength λ!) is the complex amplitude of the
linear density of electric charge of the conductor. On the other hand, the same
contour C may be used in the generalized Ampère law (5.116) to write
∮C (Ht)τ dr = Iω, (7.106)
where Iω is the total current flowing along the conductor (or rather its complex
amplitude). But, as was mentioned above, in the TEM wave the ratio Et/Ht of the
field components participating in these two integrals is constant and equal to Z =
(μ/ε)1/2, so that Eqs. (7.105) and (7.106) give the following simple relation between
the ‘global’ variables of the conductor:
λω / ε λω ω
Iω = ≡ ≡ λω . (7.107)
Z (εμ)1/2 k
This important relation may be also obtained by a different means; let me
describe it as well, because it has an independent value. Let us consider a small
segment dz ≪ λ = 2π/k of the line’s conductor, and apply the electric charge
7-37
conservation law (4.1) to the instant values of the linear charge density and current.
The cancellation of dz in both parts yields
∂λ(z , t ) ∂I (z , t )
=− . (7.108)
∂t ∂z
If we accept the sinusoidal waveform, exp{i(kz − ωt)}, for both these variables, we
immediately recover Eq. (7.107) for their complex amplitudes, showing that this
relation expresses just the charge continuity law.
The global Eq. (7.108) may be made more specific in the case when the frequency
dependence of ε and μ is negligible, and the transmission line consists of just two
isolated conductors (see, e.g. Figure 7.17). In this case, in order to have the wave
localized in the space near the two conductors, we need a sufficiently fast decrease of
its electric field at large distances. For that, their linear charge densities for each
value of z should be equal and opposite, and we can simply relate them to the
potential difference V between the conductors:
λ(z , t )
= C 0, (7.109)
V (z , t )
where C0 is the mutual capacitance of the conductors (per unit length)—which was
repeatedly discussed in chapter 2. Then Eq. (7.108) takes the form
∂V (z , t ) ∂I (z , t )
C0 =− . (7.110)
∂t ∂z
Next, let us consider the contour shown with the red dashed line in figure 7.18,
which shows a cross-section of the transmission line by a plane containing the wave
propagation axis z, and apply to it the Faraday induction law (6.3).
Figure 7.18. Electric current, magnetic flux, and voltage in a two-conductor transmission line.
Since the electric field is zero inside the conductors (in figure 7.18, on the
horizontal segments of the contour), the total e.m.f. equals the difference of voltages
V at the end of the segment dz, while the only sources of magnetic flux through the
area limited by the contour are the (equal and opposite) currents ±I in the
conductors, we can use Eq. (5.70) to express it. As a result, canceling dz in both
parts of the equation, we obtain
7-38
∂I (z , t ) ∂V (z , t )
L0 =− , (7.111)
∂t ∂z
where L0 is the mutual inductance of the conductors per unit length. The only
difference between this L0 and the dc mutual inductances discussed in chapter 5 is
that at the high frequencies we are analyzing now, L0 should be calculated neglecting
its penetration into the conductors. (In the dc case, we had the same situation for
superconductor electrodes, within their coarse-grain, ideal-diamagnetic description.)
The system of Eqs. (7.110) and (7.111) is frequently called the telegrapher’s
equations. Combined, they give for any ‘global’ variable f (either V, or I, or λ) the
usual 1D wave equation,
∂ 2f ∂ 2f
− L 0C 0 = 0, (7.112)
∂z 2 ∂t 2
which describes the dispersion-free TEM wave’s propagation.
Again, this equation is only valid within the frequency range where the frequency
dependence of both ε and μ is negligible. If it is not so, the global approach may still
be used for sinusoidal waves f = Re[fωexp{i(kz − ωt)}]. Repeating the above
arguments, instead of Eqs. (7.110) and (7.111) we obtain a system of two algebraic
equations
ωC 0Vω = kIω, ωL 0Iω = kVω, (7.113)
in which L0 ∝ μ and C0 ∝ ε may now depend on frequency. These equations are
consistent only if
k2 1
L 0C 0 = ≡ 2 ≡ εμ. (7.114)
ω2 v
In addition to the fact we have already learned (that the TEM wave’s speed is the
same as that of the plane wave), Eq. (7.114) gives us a result that I confess was not
emphasized enough in chapter 5: the product L0C0 does not depend on the shape or
size of the line’s cross-section (provided that the magnetic field penetration into the
conductors is negligible). Hence, if we have calculated the mutual capacitance C0 of
a system of two cylindrical conductors, the result immediately gives us their mutual
inductance: L0 = εμ/C0. This relation stems from the fact that both the electric and
magnetic fields may be expressed via the solution of a 2D Laplace equation for
system’s cross-section.
With Eq. (7.114) satisfied, any of Eqs. (7.113) gives the same result for the
following ratio:
Vω ⎛ L ⎞1/2
ZW ≡ = ⎜ 0⎟ , (7.115)
Iω ⎝ C0 ⎠
which is called the transmission line’s impedance. This parameter has the same
dimensionality (in SI units, ohms) as the wave impedance (7.7),
7-39
Eω ⎛ μ ⎞1/2
Z≡ =⎜ ⎟ , (7.116)
Hω ⎝ ε ⎠
but these parameters should not be confused, because ZW depends on the cross-
section’s geometry, while Z does not. In particular, ZW is the only relevant
parameter of a transmission line for matching with a lumped load circuit
(figure 7.19), in the important case when both the cable cross-section’s size and
the load’s linear dimensions are much smaller than the wavelength45.
Figure 7.19. Passive, lumped termination of a TEM transmission line.
Indeed, in this case we may consider the load in the quasi-static limit and write
Vω(z0) = ZL(ω)Iω(z0), (7.117)
where ZL(ω) is the (generally complex) impedance of the load. Taking V(z, t) and
I(z, t) in the form similar to Eqs. (7.61) and (7.62), and writing the two Kirchhoff’s
laws for the point z = z0, we obtain for the reflection coefficient a result similar to
Eq. (7.68):
ZL(ω) − ZW
R= . (7.118)
ZL(ω) + ZW
This formula shows that for perfect matching (i.e. the total wave absorption in the
load), the load’s impedance ZL(ω) should be real and equal to ZW—but not
necessarily to Z.
As an example, let us consider one of the simplest (and the most important)
transmission lines: the coaxial cable (figure 7.20)46. For this geometry, we already
know the expressions for both C0 and L047, although they have to be modified for
the dielectric and magnetic constants, and the magnetic field’s non-penetration into
the conductors. As a result of such modification, we obtain the formulas,
2πε μ
C0 = , L0 = ln(b / a ), (7.119)
ln(b / a ) 2π
illustrating that the universal relation (7.114) is indeed valid. In contrast, for the
cable’s impedance (7.115), Eqs. (7.119) yield a geometry-dependent value
45
The ability of TEM lines to have such a small cross-section is their other important practical advantage.
46
It was invented by O Heaviside in 1880.
47
See, respectively, Eqs. (2.49) and (5.79).
7-40
Figure 7.20. The cross-section of a coaxial cable with a (possibly, dispersive) dielectric filling.
⎛ μ ⎞1/2 ln(b / a ) ln(b / a )

ZW = ⎜ ⎟ ≡Z ≠ Z. (7.120)
⎝ε⎠ 2π 2π
For the standard TV antenna cables (such as RG-6/U, with b/a ∼ 3, ε/ε0 ≈
2.2), ZW = 75 ohms, while for most computer component connections, cables
with ZW = 50 ohms (such as RG-58/U) are prescribed by electronic engineering
standards. Such cables are broadly used for the transmission of electromagnetic
waves with frequencies up to 1 GHz over distances of a few km, and up to ∼20
GHz on the tabletop scale (a few meters), limited by wave attenuation—see
section 7.9.
Moreover, the following two facts enable a broad application in electrical
engineering and physical experiments of the coaxial-cable-like systems. First, as
Eq. (5.78) shows, in a cable with a ≪ b, most wave energy is localized near the
internal conductor. Second, the theory to be discussed in the next section shows that
waves of other (H- and E-) modes in the cable are impossible until the wavelength λ
becomes smaller than ∼π(a + b). As a result, the TEM mode propagation in a cable
with a ≪ b < λ/π is not much affected even if the internal conductor is not straight,
but is bent—for example, into a helix—see, e.g. Figure 7.21.
Figure 7.21. A typical traveling-wave tube: (1) electron gun, (2) RF input, (3) beam-focusing magnets, (4)
wave attenuator, (5) helix coil, (6) RF output, (7) vacuum tube, (8) electron collector. Adapted from https://en.
wikipedia.org/wiki/Traveling-wave_tube under the Creative Commons BY-SA 3.0 license.
In such a system, called the traveling-wave tube (TWT), the quasi-TEM wave
propagates with velocity v = 1/(εμ)1/2 ∼ c along the helix’s length, so that the
velocity’s component along the cable’s axis may be made close to the velocity u ≪ c
of the electron beam moving ballistically along the tube’s axis, enabling their
7-41
effective (length-accumulating) interaction and, as a result, a broadband amplifica-

tion of the wave48.
Another important example of a TEM transmission line is a set of two parallel
wires. In the form of twisted pairs49, they allow communications, in particular long-
range telephone and DSL Internet connections, at frequencies up to a few hundred
kHz, as well as relatively short, multi-line Ethernet and TV cables at frequencies up
to ∼1 GHz, limited mostly by the mutual interference (‘crosstalk’) between the
individual lines of the same cable and the unintended radiation of the wave into the
environment.
7.6 Waveguides: H and E waves

Let us now return to Eqs. (7.100) and explore the H- and E-waves—with, respectively,
either Hz or Ez different from zero. At first sight they may seem more complex.
However, Eqs. (7.101), which determine the distribution of these longitudinal
components over the cross-section, are just 2D Helmholtz equations for scalar
functions. For simple cross-section geometries, they may be solved using the methods
discussed for the Laplace equation in chapter 2, in particular variable separation.
After the solution of such an equation has been found, the transverse components of
the fields may be calculated by differentiation, using the simple formulas
i i ⎡ k ⎤
2⎢
Et = [kz ∇t Ez − kZ ( nz × ∇tHz )], Ht = kz ∇tHz + (nz × ∇tEz )⎥ , (7.121)
kt2 kt ⎣ Z ⎦
which follow from the two equations in the first line of Eqs. (7.100)50.
In comparison with the boundary problems of electro- and magnetostatics, the
only conceptually new feature of Eqs. (7.101) is that they form the so-called
eigenproblems, with typically many solutions (eigenfunctions), each describing a
specific wave mode and corresponding to a specific eigenvalue of the parameter kt.
The good news here is that these values of kt are determined by this 2D boundary
problem and hence do not depend on kz. As a result, the dispersion law ω(kz) of any
mode, which follows from the last form of Eq. (7.102),
⎛ k 2 + kt2 ⎞1/2 1/2

ω=⎜ z ⎟ = (v 2k z2 + ωc2 ) , (7.122)
⎝ εμ ⎠
is functionally the same and is also absolutely similar to that of plane waves in a
plasma—see Eq. (7.38), figure 7.6, and their discussion in section 7.2. The only
48
Very unfortunately, in this course I will not have time/space to discuss the (rather elegant) theory of such
devices. The reader seriously interested in this field may be referred, for example, to the detailed monograph [7].
49
Such twisting, around the line’s axis, reduces mutual induction (‘crosstalk’) between adjacent lines, and the
parasitic radiation at their bends.
50
For that, one of these two linear equations should first be vector-multiplied by nz. Note that this approach
could not be used to analyze the TEM waves, because for them kt = 0, Ez = 0, Hz = 0, and Eqs. (7.121) yield
uncertainty.
7-42
differences are that the speed in light c is now replaced with v = 1/(εμ)1/2, the speed of
plane (or any TEM) waves in the medium filling the waveguide, and ωp is replaced
with the so-called cutoff frequency
ωc ≡ vkt , (7.123)
specific for each mode. (As Eq. (7.101) implies, and as we will see from several
examples below, kt has the order of 1/a, where a is the characteristic dimension of
waveguide’s cross-section, so that the critical value of the free-space wavelength λ ≡
2πc/ω is of the order of a.) Below the cutoff frequency of each particular mode, such
wave cannot propagate in the waveguide51. As a result, the modes with the lowest
values of ωc present special practical interest, because the choice of the signal
frequency ω between the two lowest values of the cutoff frequency (7.123)
guarantees that the waves propagate in the form of only one mode, with the lowest
kt. Such a choice allows one to simplify the excitation of the desired mode by wave
generators, and to avoid the parasitic transfer of electromagnetic wave energy to
undesirable modes by (virtually unavoidable) small inhomogeneities of the system.
The boundary conditions for the Helmholtz equation (7.101) depend on the
propagating wave type. For the E-modes, with Hz = 0 but Ez ≠ 0, the condition
Eτ = 0 immediately gives
Ez C = 0, (7.124)
where C is the contour limiting the conducting wall’s cross-section. For the
H-modes, with Ez = 0 but Hz ≠ 0, the boundary condition is slightly less obvious
and may be obtained using, for example, the second equation of the system (7.100),
vector-multiplied by nz. Indeed, for the component perpendicular to the conductor
surface the result of such multiplication is
k ∂Hz
ikz(Ht )n − i (nz × Et)n = . (7.125)
Z ∂n
But the first term on the left-hand side of this relation must be zero on the wall
surface, because of the second of Eqs. (7.104), while according to the first of Eqs.
(7.104), the vector Et in the second term cannot have a component tangential to the
wall. As a result, the vector product in that term cannot have a normal component,
so that the term should equal zero as well, and Eq. (7.125) is reduced to
∂Hz
= 0. (7.126)
∂n C
Let us see how all this machinery works for a simple but practically important
case of a metallic-wall waveguide with a rectangular cross-section—see figure 7.22.
51
An interesting recent twist in the ideas of electromagnetic metamaterials (mentioned in section 7.5) is the so-
called ε-near-zero materials, designed to have an effective product εμ much lower than ε0μ0 within certain
frequency ranges. Since at these frequencies the speed v (7.4) becomes much lower than c, the cutoff frequency
(7.123) virtually vanishes. As a result, the waves may ‘tunnel’ through very narrow sections of metallic
waveguides filled with such materials—see, e.g. [8].
7-43
Figure 7.22. A schematic representation of a rectangular waveguide and the transverse field distribution in its
fundamental mode H10.
In the natural Cartesian coordinates shown in this figure, both equations (7.101)
take the simple form
⎛ ∂2 ∂2 ⎞ ⎧E , for E -modes,
2
⎜ 2 + 2 + kt ⎟ f = 0, where f = ⎨ z (7.127)
⎝ ∂x ∂y ⎠ ⎩ Hz , for H -modes.
From chapter 2 we know that the most effective way of solving of such equations in
a rectangular region is variable separation, in which the general solution is
represented as a sum of partial solutions of the type
f = X (x )Y (y ). (7.128)
Plugging this expression into Eq. (7.127) and dividing each term by XY, we obtain
the equation
1 d 2X 1 d 2Y
+ + kt2 = 0, (7.129)
X dx 2 Y dy 2
which should be satisfied for all values of x and y within the waveguide’s interior.
This is only possible if each term of the sum equals a constant. Taking the X-term and
Y-term constants in the form (−kx2) and (−ky2), respectfully, and solving the
corresponding ordinary differential equations52, for the eigenfunction (7.128) we obtain
f = (cx cos kxx + sx sin kxx )(cy cos kyy + sy sin kyy ) , with k x2 + k y2 = kt2, (7.130)
where the constants c and s should be found from the boundary conditions. Here the
difference between the H-modes and E-modes pitches in.
For the H-modes, Eq. (7.130) is valid for Hz, and we should use the boundary
condition (7.126) on all metallic walls of the waveguide (x = 0 and a; y = 0 and b—
see figure 7.22). As a result, we obtain very simple expressions for eigenfunctions and
eigenvalues:
πnx πmy
(Hz )nm = Hl cos cos , (7.131)
a b
52
Let me hope that the solution of equations of the type d 2X /dx 2 + k x2X = 0 does not present a problem for
the reader, due to his or her prior experience with problems such as standing waves on a guitar string,
wavefunctions in a flat 1D quantum well, or (with the replacement x → t) a classical harmonic oscillator.
7-44
πn πm 1/2 ⎡⎛ n ⎞2 ⎛ m ⎞2 ⎤
1/2
kx = , ky = , (kt )nm = (k x2 + k y2 ) = π ⎢⎜ ⎟ + ⎜ ⎟ ⎥ , (7.132)
a b ⎣⎝ a ⎠ ⎝b⎠ ⎦
where Hl is the longitudinal field’s amplitude, and n and m are two arbitrary integer
numbers, except that they cannot be equal to zero simultaneously. (Otherwise, the
function Hz(x,y) would be constant, so that according to Eq. (7.121) the transverse
components of the electric and magnetic field would equal zero. As a result, as the
last two lines of Eq. (7.100) show, the whole field would be zero for any kz ≠ 0.)
Assuming for certainty that a ⩾ b (as shown in figure 7.22), we see that the lowest
eigenvalue of kt, and hence the lowest cutoff frequency (7.123), is achieved for the
so-called H10 mode with n = 1 and m = 0, and hence with
π
(kt )10 = (7.133)
a
(thus confirming our prior estimate of kt).
Depending on the a/b ratio, the second lowest kt (and hence ωc) belongs to either
the H11 mode with n = 1 and m = 1,
1 ⎞1/2 ⎡ ⎛ a ⎞2 ⎤
1/2
⎛1
(kt )11 = π ⎜ 2 + 2 ⎟ = ⎢1 + ⎜ ⎟ ⎥ (kt )10 , (7.134)
⎝a b ⎠ ⎣ ⎝b⎠ ⎦
or to the H20 mode with n = 2 and m = 0,

2π
(kt )20 = = 2(kt )10 . (7.135)
a
These values become equal at a/b = √3 ≈ 1.7; in practical waveguides, the a/b ratio
is made not too far from this value. For example, in the standard X-band (∼10 GHz)
waveguide WR90, a ≈ 2.3 cm (fc ≡ ωc/2π ≈ 6.5 GHz), and b ≈ 1.0 cm.
Now let us have a fast look at the alternative E-modes. For them, we still should
use the general solution (7.130) with f = Ez, but now with the boundary condition
(7.124). This gives us the eigenfunctions
πnx πmy
(Ez )nm = El sin sin , (7.136)
a b
and the same eigenvalue spectrum (7.132) as for the H modes. However, now neither
n nor m can be equal to zero; otherwise Eq. (7.136) would give the trivial solution
Ez(x,y) = 0. Hence the lowest cutoff frequency of TM waves is achieved in the so-
called E11 mode with n =1, m = 1, and with the eigenvalue given by Eq. (7.134),
always higher than (kt)10.
Thus the fundamental H10 mode is certainly the most important wave in
rectangular waveguides; let us have a better look at its field distribution. Plugging
7-45
the corresponding solution (7.131) with n = 1 and m = 0 into the general relation
(7.121), we easily obtain
kza πx
(Hx )10 = −i Hl sin , (Hy )10 = 0, (7.137)
π a
ka πx
(Ex )10 = 0, (Ey )10 = i ZHl sin . (7.138)
π a
This field distribution is (schematically) shown in figure 7.22. Neither of the fields
depends on the coordinate y—the feature very convenient, in particular, for
microwave experiments with small samples. The electric field has only one (in figure
7.22, vertical) component that vanishes at the side walls and reaches its maximum at
the waveguide’s center; its field lines are straight, starting and ending on wall surface
charges (whose distribution propagates along the waveguide together with the
wave). In contrast, the magnetic field has two non-vanishing components (Hx and
Hz) and its field lines are shaped as horizontal loops wrapped around the electric
field maxima.
An important question is whether the H10 wave may be usefully characterized by
a unique impedance introduced similarly to ZW of the TEM modes—see Eq. (7.115).
The answer is ‘no’, because the main value of ZW is a convenient description of the
impedance matching of a transmission line with a lumped load—see figure 7.19 and
Eq. (7.118). As was discussed above, such simple description is possible (i.e. does not
depend on the exact geometry of the connection) only if both dimensions of the line’s
cross-section are much less than λ. But for the H10 wave (and more generally, any
non-TEM mode) this is impossible—see, e.g. Eq. (7.129): its lowest frequency
corresponds to the TEM wavelength λmax = 2π/(kt)min = 2π/(kt)10 = 2a.53
Now let us consider metallic-wall waveguides with a round cross-section (figure
7.23a). In this single-connected geometry, the TEM waves are impossible again,
while for the analysis of H-modes and E-modes, the polar coordinates {ρ, φ} are
most natural. In these coordinates, the 2D Helmholtz equation (7.101) takes the
following form:
⎡1 ∂ ⎛ ∂ ⎞ 1 ∂2 ⎤ ⎧ E , for E -modes,
⎢ ⎜ρ ⎟ + 2 2 + kt ⎥ f = 0,
2
f=⎨ z (7.139)
⎣ ρ ∂ρ ⎝ ∂ρ ⎠ ρ ∂φ ⎦ ⎩ Hz , for H -modes.
Separating the variables as f = R (ρ )F (φ ), we obtain

1 d ⎛ dR ⎞ 1 d 2F
⎜ρ ⎟ + 2 + kt2 = 0. (7.140)
ρR dρ ⎝ dρ ⎠ ρ F dφ 2
53
The reader is encouraged to find a simple interpretation of this equality.
7-46
Figure 7.23. (a) Metallic and (b) dielectric waveguides with circular cross-sections.
But this is exactly the Eq. (2.127) that was studied in section 2.7 in the context of
electrostatics, just with a replacement of notation: γ → kt. So we already know that
in order to have 2π-periodic functions F (φ ), and finite values R (0) (which are
evidently necessary for our current case—see figure 7.23a), the general solution must
have the form given by Eq. (2.136), i.e. the eigenfunctions are expressed via integer-
order Bessel functions of the first kind:
fnm = Jn(knmρ)(cn cos nφ + sn sin nφ) ≡ const × Jn(knmρ) cos n(φ − φ0), (7.141)
with the eigenvalues knm of the transverse wave number kt to be determined from
appropriate boundary conditions, and an arbitrary constant φ0.
As for the rectangular waveguide, let us start from the H-modes (f = Hz). Then
the boundary condition on the wall surface (ρ = R) is given by Eq. (7.126), which for
the solution (7.141) takes the form
d
Jn(ξ ) = 0, where ξ ≡ kR . (7.142)
dξ
This means that eigenvalues of Eq. (7.139) are
ξ′nm
kt = knm = , (7.143)
R
where ξ′nm is the mth root of the function dJn(ξ)/dξ. The approximate values of these
roots for several lowest n and m may be read out from the plots in figure 2.18; their
more accurate values are given in table 7.1. The table shows, in particular, that the
lowest of the roots is ξ′11 ≈ 1.84.54 Thus, a bit counter-intuitively, the fundamental
mode, providing the lowest cutoff frequency ωc = vknm, is H11, corresponding to n =
1 rather than n = 0:
⎛ ρ⎞
Hz = Hl J1⎜ξ11′ ⎟ cos (φ − φ0). (7.144)
⎝ R⎠
54
Mathematically, the lowest root of Eq. (7.142) with n = 0 equals 0. However, it would yield k = 0 and hence
a constant field Hz, which, according to the first of Eqs. (7.121), would give zero electric field.
7-47
Table 7.1. Roots ξ′nm of the function dJn(ξ)/dξ for a few lowest values of the Bessel function’s index n and the
root’s number m.
m=1 2 3
n=0 3.83171 7.015587 10.1735

1 1.84118 5.33144 8.53632
2 3.05424 6.70613 9.96947
3 4.20119 8.01524 11.34592
It has the transverse wave number kt = k11 = ξ′11/R ≈ 1.84/R, and hence the cutoff
frequency corresponding to the TEM wavelength λmax = 2π/k11 ≈ 3.41 R. Thus the
ratio of λmax to the waveguide diameter 2R is about 1.7, i.e. is close to the ratio λmax/
a = 2 for the rectangular waveguide. The origin of this proximity is clear from figure
7.24, which shows the transverse field distribution in the H11 mode. (It may be
readily calculated from Eq. (7.121) with Ez = 0, and Hz given by Eq. (7.144).)
Figure 7.24. Schematic representation of transverse field components in the fundamental H11 mode of a
metallic, circular waveguide.
One can see that the field structure is actually very similar to that of the
fundamental mode in the rectangular waveguide, shown in figure 7.22, despite
the different nomenclature (which is due to the different coordinates used for the
solution). However, note the arbitrary constant angle φ0, indicating that in circular
waveguides the transverse field’s polarization is arbitrary. For some practical
applications, such degeneracy of these ‘quasi-linearly polarized’ waves creates
problems; they may be avoided by using waves with circular polarization.
As table 7.1 shows, the next lowest H-mode is H21, for which kt = k21 = ξ′21/R ≈
3.05/R, almost twice larger than that of the fundamental mode, and only then comes
the first mode with no angular dependence of any field, H01, with kt = k01 = ξ′01/R ≈
3.83/R.55
55
The electric field lines in the H01 mode (as well as all higher H0m modes) are directed straight from the
symmetry axis to the walls, recalling those of the TEM waves in a coaxial cable. Due to this property, these
modes provide, at ω ≫ ωc, much lower energy losses (see section 7.9 below) than the fundamental H11 mode,
and are sometimes used in practice, despite the inconvenience of working in the multimode frequency range.
7-48
For the E modes, we may still use Eq. (7.141) (with f = Ez), but with the boundary
condition (7.124) at ρ = R. This gives the following equation for the problem
eigenvalues:
ξnm
Jn(knmR ) = 0, i. e. knm = , (7.145)
R
where ξnm is the mth root of function Jn(ξ)—see table 2.1. The table shows that the
lowest kt equals to ξ01/R ≈ 2.405/R. Hence the corresponding mode (E01), with
⎛ ρ⎞
Ez = El J0⎜ξ01 ⎟ , (7.146)
⎝ R⎠
has the second lowest cutoff frequency, approximately 30% higher than that of the
fundamental mode H11.
Finally, let us discuss one more topic of general importance—the number N of
electromagnetic modes that may propagate in a waveguide within a certain range of
relatively large frequencies ω ≫ ωc. It is easy to calculate for a rectangular
waveguide, with its simple expressions (7.132) for the eigenvalues of {kx, ky}.
Indeed, these expressions describe a rectangular mesh on the [kx, ky] plane, so that
each point corresponds to the plane area ΔAk = (π/a)(π/b), and the number of modes
in a large k-plane area Ak ≫ ΔAk is N = Ak/ΔAk = abAk/π2 = AAk/π2, where A is the
waveguide’s cross-section area56. However, it is frequently more convenient to
discuss transverse wave vectors kt of arbitrary direction, i.e. with an arbitrary sign of
their components kx and ky. Taking into account that the opposite values of each
component actually give the same wave, the actual number of different modes of
each type (E- or H-) is a factor of 4 lower than was calculated above. This means
that the number of modes of both types is
Ak A
N=2 . (7.147)
(2π )2
Let me leave it for the reader to give hand-waving (but convincing) arguments
that this mode counting rule is valid for waveguides with a cross-section of any shape,
and any boundary conditions on the walls, provided that N ≫ 1.
7.7 Dielectric waveguides, optical fibers, and paraxial beams

Now let us discuss electromagnetic wave propagation in dielectric waveguides. The
conceptually simplest, step-index waveguide (see figures 7.23b and 7.25) consists of
an inner core and an outer shell (in the optical fiber technology lingo, called the
cladding) with a higher wave propagation speed, i.e. lower index of refraction:
This formula ignores the fact that, according to our analysis, some modes (with n = 0 and m = 0 for
56
H modes, and n = 0 or m = 0 for E modes), are forbidden. However, for N ≫ 1, the associated corrections of
Eq. (7.147) are negligible.
7-49
Figure 7.25. Wave propagation in a thick optical fiber.
v+ > v−, i.e. n+ < n−, k+ < k−, ε+μ+ < ε−μ− . (7.148)
at the same frequency. (In most cases the difference is achieved due to that in the
electric permittivity, ε+ < ε−, while magnetically both materials are virtually passive:
μ− ≈ μ+ ≈ μ0, so that their refraction indices n±, defined by Eq. (7.84), are very close
to (ε±/ε0)1/2; I will limit my discussion to this approximation.) The idea of the
waveguide operation may be readily understood in the limit when the wavelength λ
is much smaller than the characteristic size R of the core’s cross-section. If this
‘geometric optics’ limit, at the distances of the order of λ from the core-to-cladding
interface, which determines the wave reflection, we can approximate the interface with a
plane. As we know from section 7.4, if the angle θ of the wave’s incidence on such an
interface is larger than the critical value θc specified by Eq. (7.85), the wave is totally
reflected. As a result, the waves launched into the fiber core at such ‘grazing’ angles
propagate inside the core, repeatedly reflected from the cladding—see figure 7.25.
The most important type of dielectric waveguides are optical fibers57. Due to a heroic
technological effort, in about three decades starting from the mid-1960s the attenuation
of glass fibers has been decreased from the values of the order of 20 db km−1 (typical for
window glass) to the fantastically low values of about 0.2 db km−1 (meaning virtually
perfect transparency of 10 km long fiber segments!), combined with extremely low
plane-wave (so-called chromatic) dispersion below 10 ps km−1 · nm−1.58 In conjunction
with the development of inexpensive erbium-based quantum amplifiers, this break-
through has enabled inter-continental (undersea), broadband59 optical cables, which are
the backbone of all the modern telecommunication infrastructure.
The only bad news is that these breakthroughs were achieved for just one kind of
material (silica-based glasses)60 within a very narrow range of chemical composition.
As a result, the dielectric constants κ± ≡ ε±/ε0 of the cladding and core of practical
57
For a comprehensive discussion of this vital technology see, e.g. [9].
58
Both these parameters are best not in the visible light range (from 380 to 740 nm), but in the near-infrared, with
the attenuation lowest between approximately 1500 and 1650 nm, so that two windows—the so-called C-band
(1530–1565 nm) and L-band (1570–1610 nm) are used in modern optical communication systems.
59
Each of the frequency bands mentioned above, at a typical signal-to-noise ratio S/N > 105 (50 db),
corresponds to the Shannon bandwidth Δf log2(S/N) exceeding 1014 bits per second, five orders of magnitude
(!) higher than that of a modern Ethernet cable. The practical bandwidth of a fiber is somewhat lower, but an
optical cable, with many fibers in parallel, has a proportionately higher aggregate bandwidth. A recent (circa
2017) example is the transatlantic (6600 km long) cable Marea, with eight fiber pairs and an aggregate useable
bandwidth of 160 Tbits per second.
60
The silica-based fibers were suggested in 1966 by C Kao (awarded with the 2009 Nobel Prize in physics), but
the idea of using optical fibers for communications may be traced back to at least the 1963 work by J
Nishizawa.
7-50
optical fibers are both close to 2.2 (n± ≈ 1.5) and hence very close to each other, so
that the relative difference of the refraction indices,
n− − n+ ε−1/2 − ε+1/2 ε − ε+
Δ≡ = 1/2
≈ − , (7.149)
n− ε− 2ε±
is typically below 0.5%. This factor limits the fiber bandwidth. Indeed, let us use the
geometric-optics picture to calculate the number of quasi-plane-wave modes that
may propagate in the fiber. For the complementary angle (figure 7.25)
π
ϑ ≡ − θ, so that sin θ = cos ϑ , (7.150)
2
Eq. (7.85) gives the following propagation condition:
n
cos ϑ > + = 1 − Δ. (7.151)
n−
In the limit Δ ≪ 1, when the incidence angles θ > θc of all propagating waves are
very close to π/2 and hence the complimentary angles are small, we can keep only the
two first terms in the Taylor expansion of the left-hand side of Eq. (7.151) and obtain
ϑ2max ≈ 2Δ. (7.152)
(Even for the higher-end value Δ = 0.005, this critical angle is only ∼0.1 radian, i.e.
close to 5°.) Due to this smallness, we can approximate the maximum transverse
component of the wave vector as
(kt )max = k (sin ϑ)max ≈ k ϑmax ≈ 2 k Δ, (7.153)
and use Eq. (7.147) to calculate the number N of propagating modes:
N≈2
(
(πR2 ) πk 2 ϑ2max ) = (kR)2Δ. (7.154)
(2π )2
For typical values k = 0.73 × 107 m−1 (corresponding to the free-space wavelength
λ0 = nλ = 2πn/k ≈ 1.3 μm), R = 25 μm, and Δ = 0.005, this formula gives N ≈ 150.
Now we can calculate the geometric dispersion of such a fiber, i.e. the difference of
the mode propagation speed, which is commonly characterized in terms of the
difference between the wave delay times (traditionally measured in picoseconds per
kilometer) of the fastest and the slowest mode. Within the geometric optics
approximation, the difference of time delays of the fastest mode (with kz = k) and
the slowest mode (with kz = k sin θc) at distance l is
⎛l ⎞ ⎛k l ⎞ l l l⎛ n ⎞ l
Δt = Δ⎜ ⎟ = Δ⎜ z ⎟ = Δkz = (1 − sin θc ) = ⎜1 − + ⎟ = Δ. (7.155)
⎝ vz ⎠ ⎝ω⎠ ω v v⎝ n− ⎠ v
For the example considered above, the TEM wave speed in the glass, v = c/n ≈ 2 ×
108 m s−1, and the geometric dispersion Δt/l is close to 25 ps m−1, i.e. 25 000 ps km−1.
7-51
(This means, for example, that a 1 ns pulse, being distributed between the modes,
would spread to a ∼25 ns pulse after passing through a just 1 km fiber segment.) This
result should be compared with the chromatic dispersion mentioned above, below
10 ps km−1 · nm−1, which gives dt/l of the order of only 1000 ps km−1 in the whole
communication band dλ ∼ 100 nm. For this reason, such relatively thick (2R ∼
50 nm) multi-mode fibers with high geometric dispersion are used for the transfer of
signals/power over only short distances below ∼100 m. (In return, they may carry
relatively large power, beyond 10 mW.)
Long-range telecommunications are based on single-mode fibers, with thin cores
(typically with diameters 2R ∼ 5 μm, i.e. of the order of λ/Δ1/2). For such structures,
Eq. (7.154) yields N ∼ 1, but in this case the geometric optics approximation is not
quantitatively valid, and for the fiber analysis, we should get back to the Maxwell
equations. In particular, this analysis should take into an explicit account the
evanescent wave in the cladding, because its penetration depth may be comparable
to R.61
Since the cross-section of an optical fiber lacks metallic walls, the Maxwell
equations describing them cannot be exactly satisfied with TEM-wave, H-mode, or
E-mode solutions. Instead, the fibers can carry the so-called HE and EH modes, with
both vectors H and E having longitudinal components simultaneously. In such
modes, both Ez and Hz inside the core (ρ ⩽ R) have a form similar to Eq. (7.141):
f− = fl Jn(ktρ)cos n(φ − φ0), where kt2 = k −2 − k z2 > 0, and k −2 ≡ ω 2ε−μ− , (7.156)
where the constant angles φ0 may be different for each field. On the other hand, for
the evanescent wave in the cladding, we may rewrite Eqs. (7.101)–(7.102) as
( ∇2 − κt2 )f+ = 0, where κt2 ≡ k z2 − k+2 > 0, and k+2 ≡ ω 2ε+μ+ . (7.157)
Figure 7.26. Relation between the transverse exponents kt and κt for waves in optical fibers.
Figure 7.26 illustrates the relation between kt, κt, kz, and k±; note that the
following sum,
kt2 + κt2 = ω 2(ε− − ε+)μ0 ∼ 2k 2Δ, (7.158)
61
The following quantitative analysis of the single-mode fibers is very valuable—both in practice and as a very
good example of the solution of the Maxwell equations. However, its results will not be used in the following
parts of the course, so that if the reader is not interested in this topic, he or she may safely jump to the text
following Eq. (7.181). (I believe that the discussion of the angular momentum of electromagnetic radiation,
starting at that point, is a compulsory reading for every professional physicist.)
7-52
is fixed (at fixed frequency) and, for typical fibers, is very small (≪ k2). In particular,
figure 7.26 shows that neither kt nor κt can be larger than ω[(ε− − ε+)μ0]1/2 =
(2Δ)1/2 k. This means that the depth δ = 1/κt of the wave penetration into the
cladding is at least 1/k(2Δ)1/2 = λ/2π(2Δ)1/2 ≫ λ/2π. This is why the cladding layers in
practical optical fibers are made as thick as ∼50 μm, so that only a negligibly small
tail of this evanescent wave field reaches their outer surfaces.
In the polar coordinates, Eq. (7.157) becomes
⎡1 ∂ ⎛ ∂ ⎞ 1 ∂2 ⎤
⎢ ⎜ρ ⎟ + 2 2 − κt ⎥ f+ = 0,
2
(7.159)
⎣ ρ ∂ρ ⎝ ∂ρ ⎠ ρ ∂φ ⎦
the equation to be compared with Eq. (7.139) for the circular metallic-wall wave-
guide. From section 2.7 we know that the eigenfunctions of Eq. (7.159) are the
products of the sine and cosine functions of nφ by a linear combination of the
modified Bessel functions In and Kn, shown in figure 2.22, now of argument κtρ. The
fields have to vanish at ρ → ∞, so that only the latter functions (of the second kind)
can participate:
f+ ∝ K n(κtρ)cos n(φ − φ0). (7.160)
Now we have to reconcile Eqs. (7.156) and (7.160) using the boundary conditions
at ρ = R for both longitudinal and transverse components of both fields, with the
latter components first calculated using Eq. (7.121). Such a conceptually simple, but
somewhat bulky calculation (which I am leaving for the reader’s exercise), yields a
system of two linear, homogeneous equations for complex amplitudes El and Hl,
which are compatible if
⎛k 2 J ′ k 2 K ′ ⎞⎛ 1 Jn′ 1 K n′ ⎞ n2 ⎛ k 2 k+2 ⎞⎛ 1 1 ⎞
⎜ − n + + n ⎟⎜ + ⎟ = 2 ⎜ −2 + 2 ⎟⎜ 2 + 2 ⎟ , (7.161)
⎝ kt Jn κt K n ⎠⎝ kt Jn κt K n ⎠ R ⎝ kt κt ⎠⎝ kt κt ⎠
where the prime sign denotes the derivative of each function over its full argument:
ktρ for Jn, and κtρ for Kn.
For any given frequency ω, the system of equations (7.158) and (7.161)
determines the values of kt and κt, and hence kz. Actually, for any n > 0, this
system provides two different solutions: one corresponding to the so-called HE
wave, with a larger ratio Ez/Hz, and the EH wave, with a smaller value of that ratio.
For angular-symmetric modes with n = 0 (for which we might naively expect the
lowest cutoff frequency), the equations may be satisfied by the fields having just one
finite longitudinal component (either Ez or Hz), so that the HE modes are the usual E
waves, while the EH modes are the H waves. For the H modes, the characteristic
equation is reduced to the requirement that the second parentheses on the left-hand
side of Eq. (7.161) equal zero. Using the Bessel function identities J′0 = −J1 and K′0
= −K1, this equation may be rewritten in a simpler form:
1 J1(ktR ) 1 K1(κtR )
=− . (7.162)
kt J0(ktR ) κt K 0(κtR )
7-53
Using the simple relation between kt and κt, given by Eq. (7.158), we may plot
both parts of Eq. (7.162) as a function of the same argument, say, ξ ≡ ktR—see
figure 7.27.
Figure 7.27. Two sides of the characteristic equation (7.162), plotted as functions of ktR, for two values of its
dimensionless parameter: V = 8 (blue line) and V = 3 (red line). Note that according to Eq. (7.158), the
argument of the functions K0 and K1 is κtR = [V2 − (ktR )2]1/2 ≡ (V2 − ξ 2 )1/2 .
The right-hand side of Eq. (7.162) depends not only on ξ but also on the
dimensionless parameter V defined as the normalized right-hand side of Eq. (7.158):
V 2 ≡ ω 2(ε− − ε+)μ0R2 ≈ 2Δk±2R2 . (7.163)
(According to Eq. (7.154), if V ≫ 1, it gives twice the number N of the fiber modes
—the conclusion confirmed by figure 7.27, taking into account that it describes only
the H modes.) Since the ratio K1/K0 is positive for all values of the functions’
argument (see, e.g. the right-hand panel of figure 2.22), the right-hand side of Eq.
(7.162) is always negative, so that the equation may have solutions only in the
intervals where the ratio J1/J0 is negative, i.e. at
ξ01 < ktR < ξ11, ξ02 < ktR < ξ12, …, (7.164)
where ξnm is the mth zero of function Jn(ξ)—see table 2.1. The right-hand side of the
characteristic equation (7.162) diverges at κtR → 0, i.e. at ktR → V , so that no
solutions are possible if V is below the critical value Vc = ξ01 ≈ 2.405. At this cutoff
point, Eq. (7.163) yields k±≈ ξ01/R(2Δ)1/2. Hence, the cutoff frequency of the lowest
H mode corresponds to the TEM wavelength
2πR
λ max = (2Δ)1/2 ≈ 3.7RΔ1/2 . (7.165)
ξ01
For typical parameters Δ = 0.005 and R = 2.5 μm, this result yields λmax ∼ 0.65 μm,
corresponding to the free-space wavelength λ0 ∼ 1 μm. A similar analysis of the first
parentheses on the left-hand side of Eq. (7.161) shows that at Δ → 0, the cutoff
frequency for the E modes is similar.
7-54
This situation may look exactly like that in metallic-wall waveguides, with no
waves possible at frequencies below ωc, but this is not so. The basic reason for the
difference is that in the metallic waveguides, the approach to ωc results in the
divergence of the longitudinal wavelength λz ≡ 2π/kz. On the other hand, in dielectric
waveguides this approach leaves λz finite (kz → k+). Due to this difference, a certain
linear superposition of HE and EH modes with n = 1 can propagate at frequencies
well below the cutoff frequency for n = 0, which we have just calculated62. This
mode, in the limit ε+ ≈ ε− (i.e. Δ ≪ 1) allows a very interesting and simple
description using the Cartesian (rather than polar) components of the fields, but still
expressed as functions of the polar coordinates ρ and φ. The reason is that this mode
is very close to a linearly polarized TEM wave. (For this reason, this mode is
referred to as LP01.)
Let us select axis x parallel to the transverse component of the magnetic field
vector, so that Ex∣ρ=0 = 0, but Ey∣ρ=0 ≠ 0, and Hx∣ρ=0 ≠ 0, but Hy∣ρ=0 = 0. The only
suitable solutions of the 2D Helmholtz equation (which should be obeyed not only
by z-components of the field, but also the x- and y-components) are proportional to
J0(ktρ) with zero coefficients for Ex and Hy:
Ex = 0, Ey = E 0J0(ktρ), Hx = H0J0(ktρ), Hy = 0, for ρ ⩽ R . (7.166)
Now we can use the last two of Eqs. (7.100) to calculate the longitudinal components
of the fields:
1 ∂Ey k
Ez = = −i t E 0J1(ktρ)sin φ ,
−ikz ∂y kz
(7.167)
1 ∂Hx k
Hz = = −i t H0J1(ktρ)cos φ ,
−ikz ∂x kz
where I have used the mathematical identities J′0 = −J1, ∂ρ/∂x = x/ρ = cos φ, and
∂ρ/∂y = y/ρ = sinφ. As a sanity check, we see that the longitudinal component or
each field is a (legitimate!) eigenfunction of the type (7.141), with n = 1. Note also
that if kt ≪ kz (this relation is always true if Δ ≪ 1—see either Eq. (7.158) or figure
7.26), the longitudinal components of the fields are much smaller than their
transverse counterparts, so that the wave is indeed very close to the TEM one.
Because of that, the ratio of the electric and magnetic field amplitudes is also close to
that in the TEM wave: E0/H0 ≈ Z− ≈ Z+.
Now in order to satisfy the boundary conditions at the core-to-cladding interface
(ρ = R), we need to have a similar angular dependence of these components at ρ ⩾ R.
The longitudinal components of the fields are tangential to the interface and thus
should be continuous. Using solutions similar to Eq. (7.160) with n = 1, we obtain
62
This fact becomes less surprising if we recall that in the circular metallic waveguide, discussed in section 7.6,
the fundamental mode (H11, see figure 7.23) also corresponded to n = 1 rather than n = 0.
7-55
kt J1(ktR )
Ez = −i E 0K1(κtρ)sin φ ,
kz K1(κtR )
(7.168)
k J (k R )
Hz = −i t 1 t H0K1(κtρ)cos φ , for ρ ⩾ R .
kz K1(κtR )
For the transverse components, we should require the continuity of the normal
magnetic field μHn, for our simple field structure equal to just μHxcos φ, of the
tangential electric field Eτ = Eysin φ, and of the normal component of Dn = εEn =
εEycos φ. Assuming that μ− = μ+ = μ0, and ε+ ≈ ε−,63 we can satisfy these conditions
with the following solutions:
J0(ktR )
Ex = 0, Ey = E 0K 0(κtρ),
K 0(κtR )
(7.169)
J0(ktR )
Hx = H0K 0(κtρ), Hy = 0, for ρ ⩾ R .
K 0(ktR )
From here, we can calculate components from Ez and Hz, using the same approach
as for ρ ⩽ R:
1 ∂Ey κ J (k R )
Ez = = −i t 0 t E 0K1(κtρ)sin φ ,
−ikz ∂y kz K 0(κtR )
(7.170)
1 ∂Hx κ J (k R )
Hz = = −i t 0 t H0K1(κtρ)cos φ , for ρ ⩾ R .
−ikz ∂x kz K 0(κtR )
We see that this relation provides the same functional dependence of the fields as Eq.
(7.167), i.e. the internal and external fields are compatible, but their amplitudes at
the interface coincide only if
J1(ktR ) K (κ R )
kt = κt 1 t . (7.171)
J0(ktR ) K 0(κtR )
This characteristic equation (which may be also derived from Eq. (7.161) with n =
1 in the limit Δ→ 0) looks similar to Eq. (7.162), but functionally is very different
from it—see figure 7.28. Indeed, its right-hand side is always positive, and the left-
hand side tends to zero at ktR → 0. As a result, Eq. (7.171) may have a solution for
arbitrary small values of the parameter V , defined by Eq. (7.163), i.e. for arbitrary
low frequencies. This is why this mode is used in practical single-mode fibers: there
are no other modes that can propagate at ω < ωc, so that they cannot be
unintentionally excited on small inhomogeneities of the fiber.
It is easy to use the Bessel function approximations by the first terms of the Taylor
expansions (2.132) and (2.157) to show that in the limit V → 0, κtR tends to zero
63
This is the core assumption of this approximate theory, which accounts only for the most important effect of
the difference of dielectric constants ε+ and ε−: the opposite signs of the differences (k+2 − kz2) = kt2 and
(k−2 − kz2) = −κt2. For more discussion of the accuracy of this approximation and some exact results, let me
refer the interested reader either to the monograph [10], or to chapter 3 and appendix B in [9].
7-56
Figure 7.28. Two sides of the characteristic equation (7.171) for the LP01 mode, plotted as a function of ktR,
for two values of the dimensionless parameter: V = 8 (blue line) and V = 1 (red line).
much faster than ktR ≈ V : κtR → 2exp{−1/V } ≪ V . This means that the scale ρc ≡
1/κt of the radial distribution of the LP01 wave’s fields in the cladding becomes very
large. In this limit, this mode may be interpreted as a virtually TEM wave
propagating in the cladding, just slightly deformed (and guided) by the fiber’s
core. The drawback of this feature is that it requires a very thick cladding, in order to
avoid energy losses in outer (‘buffer’ and ‘jacket’) layers that defend the silica
components from the elements, but lack their low optical absorption. Due to this
reason, the core radius is usually selected so that the parameter V is just slightly less
than the critical value V c = ξ01 ≈ 2.4 for higher modes, thus ensuring the single-
mode operation and eliminating the geometric dispersion problem.
In order to reduce the field spread into the cladding, the step-index fibers
discussed above may be replaced with graded-index fibers whose dielectric constant
ε is gradually and slowly decreased from the center to the periphery64. Keeping only
the main two terms in the Taylor expansion of the function ε(ρ) at ρ = 0, we may
approximate such reduction as
⎛ ζ ⎞
ε(ρ) ≈ ε(0)⎜1 − ρ 2 ⎟ , (7.172)
⎝ 2 ⎠
where ζ ≡ −[(d2ε/dρ2)/ε]ρ=0 is a positive constant characterizing the fiber composition
gradient65. Moreover, if this constant is sufficiently small (ζ ≪ k2), the field
distribution across the fiber’s cross-section may be described by the same 2D
Helmholtz equation (7.101), but with the space-dependent transverse wave vector66:
64
Due to the technological difficulties of achieving wave attenuation below a few dm km−1, the graded-index
fibers are still not used as broadly as the step-index ones.
65
For an axially symmetric smooth function ε(ρ), the first derivative dε/dρ always vanishes at ρ = 0.
66
Such approach is invalid at arbitrary (large) ζ. Indeed, in the macroscopic Maxwell equations, ε(r) is under
the differentiation sign, and the exact Helmholtz-type equations for fields have additional terms containing ∇ε.
7-57
⎡⎣ ∇2 + k 2(ρ)⎤⎦ f = 0, where
t t
⎛ ζ ⎞ (7.173)
kt2( ρ) ≡ k 2( ρ) − k z2 = ω 2ε(ρ)μ0 − k z2 = kt2(0)⎜1 − ρ 2 ⎟ .
⎝ 2 ⎠
Surprisingly for such an axially symmetric problem, because of its special depend-
ence on the radius, this equation may be most readily solved in Cartesian
coordinates. Indeed, rewriting it as
⎡ ∂2 ∂2 2 ⎛ ζ ζ ⎞⎤
⎢ 2 + 2 + kt (0)⎜1 − x 2 − y 2 ⎟⎥ f = 0, (7.174)
⎣ ∂x ∂y ⎝ 2 2 ⎠⎦
and separating variables as f = X(x)Y(y), we obtain

d 2X d 2Y ⎛ ζ ζ ⎞
+ + kt2(0)⎜1 − x 2 − y 2 ⎟ = 0, (7.175)
Xdx 2
Ydy 2 ⎝ 2 2 ⎠
so that the functions X and Y obey similar differential equations, for example
d 2X ⎡ ζ ⎤
+ ⎢kx2 − kt2(0) x 2⎥X = 0, (7.176)
dx 2 ⎣ 2 ⎦
with the separation constants satisfying the following relation:
k x2 + k y2 = kt2(0) = ω 2ε(0)μ0 − k z2. (7.177)
The ordinary differential equation (7.176) is well known from the elementary
quantum mechanics, because the Schrödinger equation for the perhaps most
important quantum system, a 1D harmonic oscillator, may be rewritten in this
form. Its eigenvalues are described by a simple formula
2 ⎛ ζ ⎞1/2
(kx )n = kt(0)⎜⎝ ⎟⎠ (2n + 1),
2
(7.178)
⎛ ζ ⎞1/2
(ky2)m = kt(0)⎜ ⎟ (2m + 1),
⎝2⎠
n , m = 0, 1, 2, ...
but eigenfunctions Xn(x) and Ym(y) have to be expressed via not quite elementary
functions—the Hermite polynomials67. For most practical purposes, however, the
lowest eigenfunctions X0(x) and Y0(y) are sufficient, because they correspond to the
lowest kx,y and hence the lowest
⎛ ⎞1/2
[kt2(0)]min = (kx2 )0 + ky2( )0 = 2[kt(0)]min⎝ 2ζ ⎠ ⎜ ⎟ , i.e. [kt2(0)]min = 2ζ, (7.179)
and hence the highest propagation speed. The eigenfunctions corresponding to these
lowest eigenvalues are simple:
67
7-58
⎧ ζx 2 ⎫
X0(x ) = const × exp ⎨ − ⎬, (7.180)
⎩ 2 ⎭
and similarly for Y0(y), so that the field distribution follows the Gaussian (‘bell
curve’) function
⎧ ζ (x 2 + y 2 ) ⎫
f0 (ρ) = f0 (0)exp ⎨ − ⎬
⎩ 2 ⎭
(7.181)
⎧ ζρ 2 ⎫ ⎧ ρ2 ⎫
≡ f0 (0)exp ⎨ − ⎬ ≡ f0 (0)exp ⎨ − 2 ⎬ ,
⎩ 2 ⎭ ⎩ 2a ⎭
where a ≡ 1/ζ1/2 ≫ 1/k is the effective width of the field’s extension in the radial
direction, normal to the wave propagation axis z. This is the so-called Gaussian
beam, very convenient for some applications.
The Gaussian beam (7.181) is just one example of the so-called paraxial beams,
which may be represented as a result of modulation of a plane wave with a wave
number k with an axially symmetric envelope function f(ρ), where ρ ≡ {x, y}, with a
relatively large effective radius a ≫ 1/k.68 Such beams give me a convenient
opportunity to deliver on the promise made in section 7.1: calculate the angular
momentum L of a circularly polarized wave, propagating in free space, and prove its
fundamental relation to the wave’s energy U. Let us start from the calculation of U
for a paraxial beam (with an arbitrary, but spatially limited envelope f) of
the circularly polarized waves, with the transverse electric field components given
by Eq. (7.19):
Ex = E 0 f (ρ) cos ψ , Ey = ∓E 0 f (ρ) sin ψ , (7.182a )
where E0 is the real amplitude of the wave’s electric field at the propagation axis, ψ ≡
kz − ωt + φ is its total phase, and the two signs correspond to two possible directions
of the circular polarization69. According to Eq. (7.6), the corresponding transverse
components of the magnetic field are
E0 E0
Hx = ± f (ρ) sin ψ , Hy = f (ρ) cos ψ . (7.182b)
Z0 Z0
These expressions are sufficient to calculate the energy density (6.113) of the wave70,
68
Note that propagating in a uniform medium, i.e. outside grade-index fibers or other focusing systems, such
beams gradually increase their width a due to diffraction—to be analyzed in the next chapter.
69
For our task of calculation of two quadratic forms of the fields (L and U), their real representation (7.182) is
more convenient then the complex-exponent one. However, for linear manipulations, the latter representation
of the circularly polarized waves, Et = E0f(ρ)Re[(nx ± iny)exp{iψ}], Ht = (E0/Z0)f(ρ)Re[(∓inx + ny)exp{iψ}], is
usually more convenient, and is broadly used.
70
Note that, in contrast to a linearly polarized wave (7.16), the energy density of a circularly polarized wave
does not depend on the full phase ψ—in particular, on t at fixed z, or vice versa. This is natural, because its field
vectors rotate (keeping their magnitude) rather than oscillate—see figure 7.3b.
7-59
ε0(Ex2 + E y2 ) μ0(Hx2 + H y2 ) ε0E 02f 2 μ0E 02f 2

u= + = + 2
≡ ε0E 02f 2 , (7.183)
2 2 2 2Z0
and hence the full energy (per unit length in the direction z of the wave’s
propagation) of the beam:
∞ ∞
U= ∫ u d 2r ≡ 2π∫0 uρ dρ = 2πε0E 02 ∫0 f 2 ρ dρ . (7.184)
However, the transverse fields (7.182) are insufficient to calculate a non-vanishing

average of L. Indeed, following the angular moment’s definition in mechanics71,
L ≡ r × p, where p is a particle’s (linear) momentum, we may use Eq. (6.115) for the
electromagnetic field momentum’s density g in free space, to define the field’s
angular momentum’s density as
1 1
l≡r×g≡ 2
r × S ≡ 2 r × (E × H). (7.185)
c c
Let us use the familiar bac minus cab rule of the vector algebra72 to transform this
expression to
1
l= [E(r ⋅ H) − H(r ⋅ E)]
c2
(7.186)
1
≡ 2 {nz[Ez(r ⋅ H) − Hz(r ⋅ E)] + [Et(r ⋅ H) − Ht(r ⋅ E)]}.
c
If the field is purely transverse (Ez = Hz = 0), as it is in a strictly plane wave, the first
square brackets in the last expression vanish, while the second bracket gives an
azimuthal component of l, which oscillates in time, and vanishes at its time
averaging. (This is exactly the reason why I have not tried to calculate L at our
first discussion of the circularly polarized waves in section 7.1.)
Fortunately, our discussion of optical fibers, in particular, the derivation of Eqs.
(7.167), (7.168), and (7.170), gives us a very clear clue how to solve this paradox. If
the envelope function f(ρ) differs from a constant, the transverse wave components
(7.182) alone do not satisfy the Maxwell equations (7.2b), which necessitate
longitudinal components Ez and Hz of the fields, with73
∂Ez ∂E ∂Ey ∂Hz ∂H ∂Hy

=− x − , =− x − . (7.187)
∂z ∂x ∂y ∂z ∂x ∂y
However, as these expressions show, if the envelope function f changes very slowly in
the sense df/dρ ∼ f/a ≪ kf, the longitudinal components are very small and do not
71
See, e.g. Part CM Eq. (1.31).
72
See, e.g. Eq. (A.47).
73
The complex-exponential versions of these equalities are given by the bottom line of Eq. (7.100).
7-60
have a back effect on the transverse components, so that the above calculation of U
is still valid (asymptotically, at ka → ∞). Hence, we may still use Eqs. (7.182) on the
right-hand side of Eq. (7.187),
∂Ez ⎛ ∂f ∂f ⎞
= E 0⎜ − cos ψ ± sin ψ ⎟ ,
∂z ⎝ ∂x ∂x ⎠
(7.188)
∂Hz E 0 ⎛ ∂f ∂f ⎞
= ⎜∓ sin ψ − cos ψ ⎟ ,
∂z Z0 ⎝ ∂x ∂x ⎠
and integrate them over z as

⎛ ⎞
Ez = E 0 ∫ ⎜⎝− ∂∂xf cos ψ ± ∂∂xf sin ψ ⎟⎠dz
E 0 ⎛ ∂f ∂f ⎞
= ⎜− ∫ cos ψ dψ ± ∫ sin ψ dψ ⎟ (7.189a )
k ⎝ ∂x ∂x ⎠
E 0 ⎛ ∂f ∂f ⎞
≡ ⎜− sin ψ ∓ cos ψ ⎟ .
k ⎝ ∂x ∂x ⎠
Here the integration constant is taken for zero, because evidently no wave field
component may have a time-independent part. Integrating, absolutely similarly, the
second of Eq. (7.188), we obtain
E 0 ⎛ ∂f ∂f ⎞
Hz = ⎜± cos ψ − sin ψ ⎟ . (7.189b)
kZ0 ⎝ ∂x ∂y ⎠
With the same approximation we may calculate the longitudinal (z-)component

of l, given by the first term of Eq. (7.186), keeping only the dominant, transverse
fields (7.182) in the scalar products:
lz = Ez(r ⋅ Ht) − Hz(r ⋅ Et) ≡ Ez(xHx + yHy ) − Hz(xEx + yEy ). (7.190)
Plugging in Eqs. (7.182) and (7.189), and taking into account that in free space,
k = ω/c, and hence 1/Z0c2k = ε0/ω, we obtain:
ε0E 02 ⎛ ∂f ∂f ⎞
lz = ∓ ⎜xf +y ⎟
ω ⎝ ∂x ∂y ⎠
(7.191)
ε E 2 ⎡ ∂( f 2 ) ∂( f 2 ) ⎤ ε0E 02 ε E 2 d( f 2 )
≡ ∓ 0 0 ⎢x +y ⎥≡∓ ρ ⋅ ∇( f 2 ) ≡ ∓ 0 0 ρ .
2ω ⎣ ∂x ∂y ⎦ 2ω 2ω dρ
Hence the total angular momentum of the beam (per unit length), is
7-61
∞
Lz = ∫ lz d 2r ≡ 2π∫0 lzρ dρ
(7.192)
ε0E 02 ∞
d (f 2 ) ε0E 02 ρ=∞
=∓π
ω
∫0 ρ2
dρ
dρ ≡ ∓ π
ω
∫ρ=0 ρ 2 d (f 2 ) .
Taking this integral by parts, with the assumption that ρf → 0 at ρ → 0 and ρ → ∞

(at is true for the Gaussian beam (7.181) and all realistic paraxial beams), we finally
obtain
ε0E 02 ∞ ε0E 02 ∞
Lz = ± π
ω
∫0 f 2 d ( ρ 2 ) ≡ ± 2π
ω
∫0 f 2 ρ dρ . (7.193)
Now comparing this expression with Eq. (7.184), we see that remarkably, the ratio
Lz/U does not depend on the shape and the width of the beam (and of course on the
wave’s amplitude E0), so these parameters are very simply and universally related:
U
Lz = ± . (7.194)
ω
Since this relation is valid in the plane-wave limit a → ∞, it may be attributed to
plane waves as well, with the understanding that in real life they always have some
kind of the wave width (‘aperture’) restriction.
As the reader certainly knows, in quantum mechanics the energy excitations of
any harmonic oscillator of frequency ω are quantized in units of ℏω, while the
components of the internal angular momentum of a particle are quantized in units of
sℏ, where s is its spin. In this context, the classical relation (7.194) is used in quantum
electrodynamics as the basis for treating the electromagnetic field excitation quanta
(photons) as a sort of quantum particles with spin s = 1. (Such integer spin also fits
the Bose–Einstein statistics of the electromagnetic radiation.)
Unfortunately, I do not have time for a further discussion of the (very interesting)
physics of paraxial beams, but cannot help noticing, at least in passing, the very curious
effect of helical waves—the beams carrying not only the ‘spin’ momentum (7.194), but
also an additional ‘orbital’ angular momentum. The distribution of their energy in space
is not monotonic, and it is in the Gaussian beam (7.181), but resembles several threads
twisted around the propagation axis—hence the term ‘helical’74. Mathematically, this
structure is described by the associate Laguerre polynomials—the same special functions
that are used for the quantum-mechanical description of hydrogen-like atoms75.
Presently there are efforts to use such beams for the so-called orbital angular momentum
(OAM) multiplexing for high-rate information transmission76.
74
Note that such solutions of the Maxwell equations may be traced back to at least the 1943 theoretical work
by J Humblet; however, this issue had not been much discussed in the literature until the spectacular 1992
experiments by L Allen et al who demonstrated a simple way of generating helical optical beams—see, e.g.
[11]. For a review of later work see, e.g. [12], and references therein.
75
76
See, e.g. [13].
7-62
7.8 Resonators
Resonators are distributed oscillators, i.e. structures that may sustain standing waves
(in electrodynamics, oscillations of the electromagnetic field) even without a source,
until the oscillation amplitude slowly decreases in time due to unavoidable energy
losses. If the resonator quality (described by the so-called Q-factor, which will be
defined and discussed in the next section) is high, Q ≫ 1, this decay takes many
oscillation periods. Alternatively, high-Q resonators may sustain oscillating fields
permanently, if fed with a relatively weak incident wave.
Conceptually the simplest resonator is the Fabry–Pérot interferometer77 that may
be obtained by placing two well-conducting planes parallel to each other78. Indeed,
in section 7.3 we have seen that if a plane wave is normally incident on such a
‘perfect mirror’, located at z = 0, its reflection, at negligible skin depth, results in a
standing wave described by Eq. (7.61b):
E (z , t ) = Re (2Eωe−iωt+iπ /2 ) sin kz . (7.195)
Hence the wave would not change if we had suddenly put the second mirror
(isolating the segment of length l from the external wave source) at any position z = l
with sin kl = 0, i.e.
kl = pπ , where p = 1, 2, …. (7.196)
This condition, which determines the eigen- (or resonance-) frequency spectrum of the
resonator of fixed length l,
πv 1
ωp = vkp = p, with v= , (7.197)
a (εμ)1/2
has a simple physical sense: the resonator’s length l equals exactly p half-waves of
frequency ωp. Though this is all very simple, please note a considerable change
of philosophy from what we have been doing in the previous sections: the main task
of resonator’s analysis is finding its eigenfrequencies ωp that are now determined by
the system geometry rather than by an external wave source.
Before we move to more complex resonators, let us use Eq. (7.62) to represent the
magnetic field in the Fabry–Pérot interferometer:
⎛ E ⎞
H (z , t ) = Re ⎜2 ω e−iωt ⎟ cos kz . (7.198)
⎝ Z ⎠
Expressions (7.195) and (7.198) show that in contrast to traveling waves, each field
of the standing wave changes simultaneously (proportionately) at all points of the
Fabry–Pérot resonator, turning to zero everywhere twice a period. At the instants
when the energy of the corresponding field vanishes, the total energy of oscillations
77
The device is named after its inventors, M Fabry and A Pérot; it is also called the Fabry–Pérot etalon
(meaning ‘gauge’), because of its initial use for light wavelength measurements.
78
The resonators formed by well conducting (usually, metallic) walls are frequently called resonant cavities.
7-63
stays constant, because the counterpart field oscillates with the phase shift π/2. Such
behavior is typical for all electromagnetic resonators.
Another, more technical remark is that we can readily get the same results
(7.195)–(7.198) by solving the Maxwell equations from scratch. For example, we
already know that in the absence of dispersion, losses, and sources, they are reduced
to the wave equations (7.3) for any field components. For the Fabry–Pérot
resonator’s analysis, we can use the 1D form of these equations, say, for the
transverse component of the electric field:
⎛ ∂2 1 ∂2 ⎞
⎜ 2 − 2 2 ⎟E = 0, (7.199)
⎝ ∂z v ∂t ⎠
and solve it as a part of an eigenvalue problem with the corresponding boundary

conditions. Indeed, separating time and space variables as E (z, t ) = Z (z )T (t ), we
obtain
1 d 2Z 1 1d 2T
− 2 = 0. (7.200)
Z dz 2
v T dt 2
Calling the separation constant k2, we obtain two similar ordinary differential
equations,
d 2Z d 2T
+ k 2Z = 0, + k 2v 2T = 0, (7.201)
dz 2 dt 2
both with sinusoidal solutions, so that the product Z (z )T (t ) is a standing wave
with the wave vector k and frequency ω = kv.79 Now using the boundary conditions
E(0, t) = E(l, t) = 0,80 we obtain the eigenvalue spectrum for kp and hence for ωp =
vkp, given by Eqs. (7.196) and (7.197).
Lessons from this simple case study may be readily generalized for an arbitrary
resonator: there are (at least) two approaches to finding the eigenfrequency
spectrum:
(i) We may look at a traveling wave solution and find where reflecting mirrors may
be inserted without affecting the wave’s structure. Unfortunately, this method is
limited to simple geometries.
(ii) We may solve the general 3D wave equations,
⎛ 1 ∂2 ⎞
⎜∇2 − 2 2 ⎟ f (r , t ) = 0, (7.202)
⎝ v ∂t ⎠
79
In this form, the equations are valid even in the presence of dispersion, but with a frequency-dependent wave
speed: v2 = 1/ε(ω)μ(ω).
80
This is of course the expression of the first of the general boundary conditions (7.104). The second of these
conditions (for the magnetic field) is satisfied automatically for the transverse waves we are considering.
7-64
for field components as an eigenvalue problem with appropriate boundary con-

ditions. If the system parameters (and hence the coefficient v) do not change in time,
the spatial and temporal variables of Eq. (7.202) may be always separated by taking
f (r , t ) = R (r)T (t ), (7.203)
where the functionT (t ) always obeys the same equation as in Eq. (7.201), having the
sinusoidal solution of frequency ω = vk. Plugging this solution back into Eq. (7.201),
for the spatial distribution of the field we obtain the 3D Helmholtz equation,
(∇2 + k 2 )R (r) = 0, (7.204)
whose solution (for non-symmetric geometries) may be much more complex.

Let us use these approaches to find the eigenfrequency spectrum of a few simple,
but practically important resonators. First of all, the first method is completely
sufficient for the analysis of any resonator formed as a fragment of a uniform TEM
transmission line (e.g. a coaxial cable), confined with two conducting lids perpen-
dicular to the line direction. Indeed, since in such lines kz = k = ω/v, and the electric
field is perpendicular to the propagation axis, e.g. parallel to the lid surface, the
boundary conditions are exactly the same as in the Fabry–Pérot resonator, and we
again arrive at the eigenfrequency spectrum (7.197).
Now let us analyze a slightly more complex system: a rectangular metallic-wall
cavity of volume a × b × l—see figure 7.29. In order to use the first approach outlined
above, let us consider the resonator as a finite-length (Δz = l ) section of the
rectangular waveguide stretched along axis z, which was analyzed in detail in
section 7.6. As a reminder, at a < b, in the fundamental H10 traveling wave mode,
both vectors E and H do not depend on y, with E having only an y-component. In
contrast, H has two components, Hx and Hz, with the phase shift π/2 between them,
with Hx having the same phase as Ey—see Eqs. (7.131), (7.137), and (7.138). Hence,
if a plane, perpendicular to axis z, is placed so that the electric field vanishes on it, Hx
also vanishes, so that both boundary conditions (7.104) pertinent to a perfect
metallic wall are fulfilled simultaneously.
As a result, the H10 wave would not be perturbed by two metallic walls separated
by an integer number of half-wavelength λz/2 corresponding to the wave number
given by the combination of Eqs. (7.102) and (7.133):
Figure 7.29. A rectangular metallic-wall resonator as a finite section of a waveguide with the cross-section
shown in figure 7.22.
7-65
1/2 ⎛ ω2 π2 ⎞
kz = ( k 2 − kt2 ) = ⎜ 2 − 2 ⎟. (7.205)
⎝v a ⎠
Using this expression, we see that the smallest of these distances, l = λz/2 = π/kz, gives
the resonance frequency81
⎡⎛ π ⎞2 ⎛ π ⎞2 ⎤1/2
ω101 = v⎢⎜ ⎟ + ⎜ ⎟ ⎥ , (7.206)
⎣⎝ a ⎠ ⎝l ⎠ ⎦
where the indices of ω show the numbers of half-waves along each dimension of the
system. This is the lowest (fundamental) eigenfrequency of the resonator (if b < a, l ).
The field distribution in this mode is close to that in the corresponding waveguide
mode H10 (figure 7.22), with the important difference that the magnetic and electric
fields are shifted by phase π/2 both in space and time, just as in the Fabry–Pérot
resonator—see Eqs. (7.195) and (7.198). Such a time shift allows for a very simple
interpretation of the H101 mode that is particularly adequate for very flat resonators,
with b ≪ a, l. At the instant when the electric field reaches its maximum (figure
7.30a), i.e. when the magnetic field vanishes in the whole volume, the surface electric
charge of the walls (with the areal density σ = En/ε) is largest, being localized mostly
in the middle of the broadest (in figure 7.30, horizontal) faces of the resonator. At
immediate later times, the walls start to recharge via surface currents whose density
J is largest in the side walls, and reaches its maximal value in a quarter period of
the oscillation period of frequency ω101—see figure 7.30b. The currents generate the
vortex magnetic field, with looped field lines in the plane of the broadest face of the
resonator. The surface currents continue to flow in this direction until (in one more
quarter period) the broader walls of the resonator are fully recharged in the polarity
opposite to that shown in figure 7.30a. After that, the surface currents start to flow in
the direction opposite to that shown in figure 7.30b. This process, which repeats
again and again, is conceptually similar to the well-known oscillations in a lumped
LC circuit, with the role of (now, distributed) capacitance played mostly by the
broadest faces of the resonator, and that of (now, distributed) inductance, mostly by
its narrower walls.
Figure 7.30. A schematic representation of the fields, charges, and currents in the fundamental (H101) mode of
a rectangular metallic resonator, at two instants separated by Δt = π/2ω101.
81
In most electrical engineering handbooks, the index corresponding to the shortest side of the resonator is
listed last, so that the fundamental mode is nominated as H110 and its eigenfrequency as ω110.
7-66
In order to generalize the result (7.206) to higher oscillation modes, the second of
the approaches discussed above is more prudent. Separating variables as
R (r) = X (x )Y (y )Z (z ) in the Helmholtz equation (7.204), we see that X, Y, and Z
have to be either sinusoidal or cosinusoidal functions of their arguments, with wave
vector components satisfying the characteristic equation
ω2
k x2 + k y2 + k z2 = k 2 ≡ . (7.207)
v2
In contrast to the wave propagation problem, now we are dealing with standing
waves along all three dimensions and have to satisfy the macroscopic boundary
conditions (7.104) on all sets of parallel walls. It is straightforward to check that
these conditions (Eτ = 0, Hn = 0) are fulfilled at the following field component
distribution,
Ex = E1 cos kxx sin kyy sin kzz , Hx = H1 sin kxx cos kyy cos kzz ,
Ey = E2 sin kxx cos kyy sin kzz , Hy = H2 cos kxx sin kyy cos kzz , (7.208)
Ez = E3 sin kxx sin kyy cos kzz , Hz = H3 cos kxx cos kyy sin kzz ,
with each of the wave vector components having an equidistant spectrum, similar to
Eq. (7.196),
πn πm πp
kx = , ky = , kz = , (7.209)
a b l
so that the full spectrum of eigenfrequencies is given by the following formula,
⎡⎛ π n ⎞2 ⎛ π m ⎞2 ⎛ π p ⎞2 ⎤1/2
ωnmp = vk = v⎢⎜ ⎟ + ⎜ ⎟ + ⎜ ⎟ ⎥ , (7.210)
⎣⎝ a ⎠ ⎝ b ⎠ ⎝ l ⎠⎦
which is a natural generalization of Eq. (7.206). Note, however, that of the three
integers m, n, and p, at least two have to be different from zero in order to keep the
fields (7.206) from vanishing at all points.
We may use Eq. (7.210), in particular, to evaluate the number of different modes
in a relatively small range d3k ≪ k3 of the wave vector space, which is still much
larger than the reciprocal volume, 1/V = 1/abl, of the resonator. Taking into account
that each eigenfrequency (7.210), with nml ≠ 0, corresponds to two field modes with
different polarizations82, argumentation absolutely similar to that used at the end of
section 7.7 for the 2D case yields
d 3k
dN = 2V . (7.211)
(2π )3
82
This fact becomes evident from plugging Eqs. (7.208) into the Maxwell equation ∇ · E = 0. The resulting
equation, kxE1 + kyE2 + kzE3 =0, with the discrete, equidistant spectrum (7.209) for each wave vector
component, may be satisfied by two linearly independent sets of the constants E1,2,3.
7-67
This property, valid for resonators of arbitrary shape, is broadly used in classical
and quantum statistical physics83 in the following form. If some electromagnetic
mode functional f(k) is a smooth function of the wave vector k and the volume V is
large enough, then Eq. (7.211) may be used to approximate the sum of the
functional’s values over the modes by an integral:
∑f (k) ≈∫
N
f (k)dN ≡ ∫k f (k) ddN3k d 3k = 2 (2Vπ )3 ∫k f (k)d 3k. (7.212)
k
Leaving the similar analyses of resonant cavities of other shapes for reader’s
exercises, let me finish this section by noting that low-loss resonators may be also
formed by finite-length sections of not only metallic-wall waveguides with different
cross-sections, but also of the dielectric waveguides. Moreover, even a simple slab
of a dielectric material with a μ/ε ratio substantially different from that of its
environment (say, the free space) may be used as a high-Q Fabry–Pérot interfer-
ometer (figure 7.31), due to an effective wave reflection from its surfaces at normal
Figure 7.31. A dielectric Fabry–Pérot interferometer.
and in particular inclined incidence—see, respectively, Eq. (7.68), and Eqs. (7.91)
and (7.95).
Actually, such dielectric Fabry–Pérot interferometers are frequently more con-
venient for practical purposes than metallic-wall resonators, not only due to possibly
lower losses (particularly in the optical range), but also due to a natural coupling to
the environment that enables a ready method of wave insertion and extraction—see
figure 7.31 again. However, this coupling to environment provides an additional
mechanism of power losses, limiting the resonance quality—see the next section.
7.9 Energy loss effects

The inevitable energy losses (‘dissipation’) in passive media lead, in two different
situations, to two different effects. In a long transmission line fed by a constant wave
source at one end, the losses lead to a gradual attenuation of the wave, i.e. to a
decrease of its amplitude, and hence its power P , with the distance z along the line.
In linear materials, the time-averaged losses are proportional to the wave amplitude
squared, i.e. to the time-averaged power P of the wave itself, so that the energy
balance on a small segment dz takes the form
83
See, e.g. Part QM section 1.1 and Part SM section 2.6.
7-68
dP loss
dP = − dz = −αP dz. (7.213)
dz
The coefficient α, participating in the last form of Eq. (7.213) and defined by that
relation,
dP loss / dz
α≡ , (7.214)
P
is called the attenuation constant84. Comparing the solution of Eq. (7.213),
P (z ) = P (0)e−αz , (7.215)
with Eq. (7.29), where k is replaced with kz, we see that α may expressed as
α = 2Im kz , (7.216)
where kz is the component of the wave vector along the transmission line. In the
most important limit when the losses are low in the sense α ≪ ∣kz∣ ≈ Re kz, its effects
on the field distribution along the line’s cross-section are negligible, making the
calculation of α rather straightforward. In particular, in this limit the contributions
to attenuation from two major sources, energy losses in the filling dielectric and the
skin effect-losses in conducting walls, are independent and additive.
The dielectric losses are especially simple to describe. Indeed, a review of our
calculations in sections 7.5–7.7 shows that all of them remain valid if either ε(ω) or
μ(ω), or both, and hence k(ω), have small imaginary parts:
k″ = ωIm[ε1/2(ω)μ1/2 (ω)] ≪ k′ . (7.217)
In TEM transmission lines kz = k and hence Eq. (7.216) yields

αfilling = 2k″ = 2ωIm[ε1/2(ω)μ1/2 (ω)]. (7.218)
For dielectric waveguides, in particular optical fibers, these losses are the main
attenuation mechanism. As we already know from section 7.7, in practical optical
fibers κtR ≫ 1, i.e. most of the field propagates (as an evanescent wave) in the
cladding, with a field distribution very close to the TEM wave. This is why
Eq. (7.218) is approximately valid if it is applied to the cladding material alone.
In waveguides with non-TEM waves, we can readily use the relations between kz and
k, derived in the previous sections, to re-calculate kʺ into Im kz. (Note that at such a
re-calculation, the values of kt have to be kept real, because they are just the
eigenvalues of the Helmholtz equation (7.101), which does not include the filling
media parameters.).
84
In engineering, attenuation is frequently measured in decibels per meter (acronymed as db m−1 or just dbm):
P (z = 0) 10
α ≡ 10log10 = 10log10e α[1/m] = α[m−1] ≈ 4.34α[m−1].
db/m P (z = 1m) ln 10
7-69
In transmission lines and waveguides and with metallic walls, much higher energy
losses may come from the skin effect. If the wavelength λ is much larger than δs, as it
usually is85, we may use Eq. (6.36)86:
dP loss 2 μωδ s
= Hwall , (7.219)
dA 4
where Hwall is the real amplitude of the tangential component of the magnetic field at
the wall’s surface. The total power loss dPloss /dz per unit length of a waveguide, i.e.
the right-hand side of Eq. (7.213), now may be calculated by the integration of the
ratio dPloss /dA along the contour(s) limiting the cross-section of all conducting walls.
Since our calculation is only valid for low losses, we may ignore their effect on
the field distribution, so that the unperturbed distributions may be used both in
Eq. (7.219), i.e. in the nominator of Eq. (7.214), and also for the calculation of the
average propagating power, i.e. the denominator of Eq. (7.214)—as the integral of
the Poynting vector over the cross-section of the waveguide.
Let us see how this approach works for the TEM mode in one of the simplest
transmission lines, the coaxial cable (figure 7.20). As we already know from section
7.5, in the absence of losses, the distribution of TEM mode fields is the same as in
statics, namely:
a
Hz = 0, Hρ = 0, Hφ(ρ) = H0 , (7.220)
ρ
where H0 is the field’s amplitude on the surface of the inner conductor, and
a ⎛ μ ⎞1/2
Ez = 0, Eρ(ρ) = ZHφ(ρ) = ZH0 , Eφ = 0, Z≡ ⎜ ⎟ . (7.221)
ρ ⎝ε⎠
Now we can, neglecting losses for now, use Eq. (7.42) to calculate the time-averaged
Poynting vector
2
Z H0 2 ⎛ a ⎞
2
Z Hφ(ρ)
S = = ⎜ ⎟ , (7.222)
2 2 ⎝ ρ⎠
and from it, the total power propagating through the cross-section:
H0 2 a 2 b
P= ∫A S d 2r = Z 2
2π ∫a ρdρ
ρ 2
b
= πZ H0 2 a 2 ln .
a
(7.223)
For the particular case of the coaxial cable (figure 7.20), the contours limiting the
wall cross-sections are circles of radii ρ = a (where the surface field amplitude Hwalls
equals, in our notation, H0), and ρ = b (where, according to Eq. (7.214), the field is a
factor of b/a lower). As a result, for the power loss per unit length, Eq. (7.219) yields
85
As follows from Eq. (7.78), which may be used for crude estimates even in cases of arbitrary incidence, this
condition is necessary for low attenuation: α ≪ k only if f ≪ 1.
86
For a normally incident plane wave, this formula would bring us back to Eq. (7.78).
7-70
dP loss ⎛ a 2⎞μ
0 ωδ s π ⎛⎜ a⎞
= ⎜2πa H0 2 + 2πb H0 ⎟ = a 1 + ⎟μωδs H0 2 . (7.224)
dz ⎝ b ⎠ 4 2 ⎝ b⎠
Note that at a ≪ b the losses in the inner conductor dominate, despite its smaller
surface, because of the higher surface field. Now we may plug Eqs. (7.223) and
(7.224) into the definition (7.214) of α to calculate the skin-effect contribution to the
attenuation constant:
dP loss / dz 1 ⎛1 1 ⎞ μωδs kδ s ⎛ 1 1⎞
αskin ≡ = ⎜ + ⎟ = ⎜ + ⎟. (7.225)
P 2 ln(b / a ) ⎝ a b⎠ Z 2 ln(b / a ) ⎝ a b⎠
We see that the relative (dimensionless) attenuation, α/k, scales approximately as the
ratio δs/min[a, b]. This result has to be compared with Eq. (7.78) for the normal
incidence of plane waves on a conducting surface.
Let us use this result to evaluate α for the standard TV cable RG-6/U (with
copper conductors of diameters 2a = 1 mm, 2b = 4.7 mm, and ε ≈ 2.2ε0, μ ≈ μ0).
According to Eq. (6.33), for frequency f = 100 MHz (ω ≈ 6.3 × 108 s−1) the skin
depth of pure copper at room temperature (with σ ≈ 6.0 × 107 s m−1) is close to 6.5 ×
10−6 m, while k = ω(εμ)1/2 = (ε/ε0)1/2(ω/c) ≈ 3.1 m−1. As a result, the attenuation is
rather low: αskin ≈ 0.016 m−1, so that the attenuation length scale ld ≡ 1/α is about
60 m. Hence the attenuation in a cable connecting a roof TV antenna to a TV set in
the same house is not a big problem, although using a worse conductor, e.g. steel,
would make the losses rather noticeable. (Hence the current worldwide shortage of
copper.) However, an attempt to use the same cable in the X-band (f ∼ 10 GHz) is
more problematic. Indeed, though the skin depth δs ∝ ω−1/2 decreases with
frequency, the wavelength drops, i.e. k increases even faster (k ∝ ω), so that the
attenuation αskin ∝ ω1/2 becomes close to 0.16 m−1, i.e. ld to ∼6 m. This is why at such
frequencies, it may be necessary to use rectangular waveguides, with their larger
internal dimensions a, b ∼ 1/k, and hence lower attenuation. Let me leave the
calculation of this attenuation, using Eq. (7.219) and the results derived in section
7.7, for the reader’s exercise.
The power loss effect on free oscillations in resonators is different: here it leads to
a gradual decay of the oscillating fields’ energy U in time. A useful dimensionless
measure of this decay, called the Q-factor, may be introduced by writing the
temporal analog of Eq. (7.213)87:
ω
dU = −P loss dt = − U dt, (7.226)
Q
where ω is the eigenfrequency in the loss-free limit and
87
As losses grow, the oscillation waveform deviates from sinusoidal, and the very notion of ‘oscillation
frequency’ becomes vague. As a result, the parameter Q is well defined only if it is much higher than 1.
7-71
ω P
≡ loss (7.227)
Q U
is the temporal analog of Eq. (7.214). The solution of Eq. (7.226),
Q Q /2π QT
U (t ) = U (0)e−t/τ , with τ ≡ = = , (7.228)
ω ω /2π 2π
which is the temporal analog of Eq. (7.215), shows the physical meaning of the Q-
factor: the characteristic time τ of the oscillation energy’s decay is (Q/2π) times
longer than the oscillation period T = 2π/ω. (Another useful interpretation of Q
comes from the relation88
ω
Q= , (7.229)
Δω
where Δω is the so-called FWHM89 bandwidth of the resonance, namely the
difference between the two values of the external signal frequency, one above and
one below ω, at which the energy of the forced oscillations induced in the resonator
by an input signal is twice lower than its resonance value.)
In the important particular case of resonators formed by insertion of metallic
walls into a TEM transmission line of small cross-section (with the linear size scale a
much less than the wavelength λ), there is no need to calculate the Q-factor directly,
provided that the line attenuation coefficient α is already known. Indeed, as was
discussed in section 7.8 above, the standing waves in such a resonator, of the length
given by Eq. (7.196): l = p(λ/2) with p = 1, 2,…, may be understood as an overlap of
two TEM waves running in opposite directions or, in other words, a traveling wave
plus its reflection from one of the ends, the whole roundtrip taking time Δt = 2 l/v =
pλ/v = 2πp/ω = p T . According to Eq. (7.215), at this distance the wave’s power
drops by the factor of exp{−2αl} = exp{−pαλ}. On the other hand, the same decay
may be viewed as happening in time and, according to Eq. (7.228), results in the
drop by exp{−Δt/τ} = exp{−(pT )/(Q/ω)} = exp{−2πp/Q}. Comparing these two
exponents, we obtain
2π k
Q= = . (7.230)
αλ α
This simple relation neglects the losses at the wave reflection from the walls
limiting the resonator length. Such approximation is indeed legitimate at a ≪ λ; if
this relation is violated, or if we are dealing with more complex resonator modes
(such as those based on the reflection of E or H waves), the Q-factor may be smaller
than that given by Eq. (7.230) and needs to be calculated directly. A substantial relief
for such a direct calculation is that, just at the calculation of small attenuation in
waveguides, in the low-loss limit (Q ≫ 1) both the nominator and denominator of
the right-hand side of Eq. (7.227) may be calculated neglecting the effects of the
88
89
This is the acronym for ‘full width at half-maximum’.
7-72
power loss on the field distribution in the resonator. I am leaving such a calculation,
for the simplest (rectangular and circular) resonators, for the reader’s exercise.
To conclude this chapter, let me make a final remark: in some resonators
(including certain dielectric resonators and metallic resonators with holes in their
walls), additional losses due to the wave radiation into the environment are also
possible. In some simple cases (say, the Fabry–Pérot interferometer shown in figure
7.31) the calculation of these radiative losses is straightforward, but sometimes it
requires more elaborated approaches, which will be discussed in the next chapter.
7.10 Problems
Problem 7.1.* Find the temporal Green’s function of a medium whose complex
dielectric constant obeys the Lorentz oscillator model, given by Eq. (7.32), using:
(i) the Fourier transform, and
(ii) the direct solution of Eq. (7.30).
Hint: For the Fourier-transform approach, you may like to use the Cauchy
integral90.
Problem 7.2. The electric polarization of a material responds in the following way to
an electric field step91:
⎧ 0, for t < 0,
P(t ) = ε1E 0(1 − e−t /τ ) , if E (t ) = E 0 × ⎨
⎩1, for 0 < t ,
where τ is a positive constant. Calculate the complex permittivity ε(ω) of this
material and discuss a possible simple physical model giving such dielectric response.
Problem 7.3. Calculate the complex dielectric constant ε(ω) for a material whose
dielectric-response Green’s function, defined by Eq. (7.23), is
G (θ ) = G 0(1 − e−θ /τ ),
with some positive constants G0 and τ. What is the difference between this dielectric
response and the apparently similar one considered in the previous problem?
Problem 7.4. Use the Lorentz oscillator model of an atom, given by Eq. (7.30), to
calculate the average potential energy of the atom in a uniform, sinusoidal ac electric
field, and use the result to calculate the potential profile created for the atom by a
standing electromagnetic wave with the electric field amplitude Eω(r).
Problem 7.5. The solution of the previous problem shows that a standing plane wave
exerts a time-averaged force on a non-relativistic charged particle. Reveal the
90
See, e.g. Eq. (A.92).
91
This function E(t) is of course proportional to the well-known step function θ(t)—see, e.g. Eq. (A.87).
I am not using this notation just to avoid possible confusion between two different uses of the
Greek letter θ.
7-73
physics of this force by writing and solving the equations of motion of a free,
charged particle in:
(i) a linearly polarized, monochromatic, plane traveling wave, and
(ii) a similar but standing wave.
Problem 7.6. Calculate, sketch, and discuss the dispersion relation for electro-
magnetic waves propagating in a medium described by the Lorentz oscillator model
(7.32), for the case of negligible damping.
Problem 7.7. As was briefly discussed in section 7.292, a wave pulse of a finite but
relatively large spatial extension Δz ≫ λ ≡ 2π/k may be represented with a wave
packet—a sum of sinusoidal waves with wave vectors k within a relatively narrow
interval. Consider an electromagnetic plane wave packet of this type, with the
electric field distribution
+∞
E(r , t ) = Re ∫−∞ Eke i (kz−ωkt )dk , with ωk[ε(ωk )μ(ωk )]1/2 ≡ k ,
propagating along axis z in an isotropic, linear, and loss-free (but not necessarily
dispersion-free) medium. Express the full energy of the packet (per unit area of
wave’s front) via the complex amplitudes Ek, and discuss its dependence of time.
Problem 7.8.* Analyze the effect of a constant, uniform magnetic field B0, parallel to
the direction n of electromagnetic wave propagation, on the wave’s dispersion in
plasma, within the same simple model that was used in section 7.2 for the derivation
of Eq. (7.38). (Limit your analysis to relatively weak waves, whose magnetic field is
negligible in comparison with B0.)
Hint: You may like to represent the incident wave as a linear superposition of two
circularly polarized waves, with opposite polarization directions.
Problem 7.9. A monochromatic, plane electromagnetic wave is normally incident,

from the free space, on a uniform slab of a material with electric permittivity ε and
magnetic permeability μ, with the slab thickness d comparable with the wavelength.
(i) Calculate the power transmission coefficient T , i.e. the fraction of the incident
power that is transmitted through the slab.
(ii) Assuming that ε and μ are frequency-independent and positive, analyze in detail
the frequency dependence of T . In particular, how does the function T (ω) depend
on the slab’s thickness d and the wave impedance Z = (μ/ε)1/2 of its material?
Problem 7.10. A monochromatic, plane electromagnetic wave, with free-space wave

number k0, is normally incident on a planar, conducting film of thickness d ∼ δs ≪
1/k0. Calculate the power transmission coefficient of the system, i.e. the fraction of
92
And in more detail in Part CM section 5.3, and particularly in Part QM section 2.2.
7-74
incident wave’s power propagating beyond the film. Analyze the result in the limits
of small and large ratio d/δs.
Problem 7.11. A plane wave of frequency ω is normally incident, from free space, on
a planar surface of a material with real values of electric permittivity ε′ and magnetic
permeability μ′. To minimize the wave reflection from the surface, you may cover it
with a layer of thickness d of another transparent material—see the figure below.
Calculate the optimal values for ε, μ, and d.
Problem 7.12. A monochromatic, plane wave is incident from inside a medium with
εμ > ε0μ0 onto its plane surface, at an angle of incidence θ larger than the critical
angle θc = sin−1(ε0μ0/εμ)1/2. Calculate the depth δ of the evanescent wave penetration
into the free space, and analyze its dependence on θ. Does the result depend on the
wave’s polarization?
Problem 7.13. Analyze the possibility of propagation of surface electromagnetic

waves along a plane boundary between a plasma and the free space. In particular,
calculate and analyze the dispersion relation of the waves.
Hint: Assume that the magnetic field of the wave is parallel to the boundary and
perpendicular to the wave’s propagation direction. (After solving the problem,
justify this mode choice.)
Problem 7.14. Light from a very distant source arrives to an observer through a
planar layer of a non-uniform medium with a certain refraction index distribution,
n(z), at angle θ0—see the figure below. What is the actual direction θi to the source, if
n(z) → 1 at z → ∞? (This problem is obviously important for high-precision
astronomical measurements from the Earth’s surface.)
7-75
Problem 7.15. Calculate the impedance ZW of the long, straight TEM transmission
lines formed by metallic electrodes with the cross-sections shown in the figure below:
(i) two round, parallel wires, separated by distance d ≫ R,
(ii) a microstrip line of width w ≫ d,
(iii) a stripline with w ≫ d1 ∼ d2,
in all cases using the macroscopic boundary conditions on metallic surfaces. Assume
that the conductors are embedded into a linear dielectric with constant ε and μ.
Problem 7.16. Modify the solution of task (ii) of the previous problem for a
superconductor microstrip line, taking into account the magnetic field penetration
into both the strip and the ground plane.
Problem 7.17.* What lumped ac circuit would be equivalent to the TEM-line system
shown in figure 7.19, with an incident wave’s power Pi ? Assume that the wave
reflected from the lumped load circuit does not return to it.
Problem 7.18. Find the lumped ac circuit equivalent to a loss-free TEM transmission
line of length l ∼ λ, with a small cross-section area A ≪ λ2, as ‘seen’ (measured) from
one end, if the line’s conductors are galvanically connected (‘shortened’) at the other
end—see the figure below. Discuss the result’s dependence on the signal frequency.
Problem 7.19. Represent the fundamental H10 wave in a rectangular waveguide

(figure 7.22) with a sum of two plane waves, and discuss the physics behind such a
representation.
Problem 7.20.* For a metallic coaxial cable with the circular cross-section (figure
7.20), find the lowest non-TEM mode and calculate its cutoff frequency.
Problem 7.21. Two coaxial cable sections are connected coaxially—see the figure
below, which shows the cut along the system’s symmetry axis. Relations (7.118) and
(7.120) seem to imply that if the ratios b/a of these sections are equal, their
impedance matching is perfect, i.e. a TEM wave incident from one side on the
7-76
connection would pass it without any reflection at all: R = 0. Is this statement

correct?
Problem 7.22. Prove that TEM-like waves may propagate, in the radial direction, in
the free space between two coaxial, round, metallic cones—see the figure below. Can
this system be characterized by a certain transmission line impedance ZW, as defined
by Eq. (7.115)?
Problem 7.23.* Use the recipe outlined in section 7.7 to prove the characteristic
equation (7.161) for the HE and EH modes in a round, step-index optical fiber.
Problem 7.24. Neglecting the skin-effect depth δs, find the lowest eigenfrequencies,
and the corresponding field distributions, of the standing electromagnetic waves
inside a round cylindrical resonant cavity—see the figure below.
Problem 7.25. A plane, monochromatic wave propagates through a medium with an

Ohmic conductivity σ, and negligible electric and magnetic polarization effects.
Calculate the wave’s attenuation, and relate the result to a certain calculation carried
out in chapter 6.
Problem 7.26. Generalize the telegrapher’s equations (7.110) and (7.111) by

accounting for small energy losses:
(i) in the transmission line’s conductors, and
(ii) in the medium separating the conductors,
using their simplest (Ohmic) models. Formulate the conditions of validity of the
resulting equations.
7-77
Problem 7.27. Calculate the skin-effect contribution to the attenuation coefficient α,

defined by Eq. (7.214), for the fundamental (H10) mode propagating in a metallic-
wall waveguide with a rectangular cross-section—see figure 7.22. Use the results to
evaluate the wave decay length ld ≡ 1/α for a 10 GHz wave in the standard X-band
waveguide WR-90 (with copper walls, a = 23 mm, b = 10 mm, and no dielectric
filling), at room temperature. Compare the result with that, made in section 7.9, for
the standard TV coaxial cable, at same frequency.
Problem 7.28.* Calculate the skin-effect contribution to the attenuation coefficient

α of
(i) the fundamental (H11) wave, and
(ii) the H01 wave,
in a metallic-wall waveguide with the circular cross-section (see figure 7.23a), and
analyze the low-frequency (ω → ωc) and high-frequency (ω ≫ ωc) behaviors of α for
each of these modes.
Problem 7.29. For a rectangular metallic-wall resonator with dimensions a × b × l

(b ⩽ a, l ), calculate the Q-factor in the fundamental oscillation mode, due to the
skin-effect losses in the walls. Evaluate the factor for a 23 × 23 × 10 mm3 resonator
with copper walls, at room temperature.
Problem 7.30.* Calculate the lowest eigenfrequency and the Q-factor (due to the
skin-effect losses) of the toroidal (axially-symmetric) resonator with metallic walls,
and interior’s cross-section shown in the figure below, in the case when d ≪ r, R.
Problem 7.31. Express the contribution to the damping coefficient (the reciprocal
Q-factor) of a resonator, from small energy losses in the dielectric that fills it, via the
complex functions ε(ω) and μ(ω) of the material.
Problem 7.32. For the dielectric Fabry–Pérot resonator (figure 7.31) with the normal
wave incidence, calculate the Q-factor due to radiation losses, in the limit of a strong
impedance mismatch (Z ≫ Z0), using two approaches:
(i) from the energy balance, using Eq. (7.227), and
(ii) from the frequency dependence of the power transmission coefficient, using Eq.
(7.229).
Compare the results.
7-78
References
[1] Born M et al 1999 Principles of Optics 7th edn (Cambridge University Press)
[2] Shelby R et al 2001 Science 292 77
[3] Wilson J and Schwartz Z 2005 Appl. Phys. Lett. 86 021113
[4] Valentine J et al 2008 Nature 455 376
[5] Wood B 2009 C. R. Phys. 10 379
[6] Sato N et al 2012 J. Appl. Phys. 111 07A501
[7] Whitaker J 2012 Power Vacuum Tubes Handbook 3rd edn (Boca Raton, FL: CRC Press)
[8] Silveirinha M and Engheta N 2006 Phys. Rev. Lett. 97 157403
[9] Yariv A and Yeh P 2007 Photonics 6th edn (Oxford University Press)
[10] Snyder A and Love D 1983 Optical Waveguide Theory (Chapman and Hall)
[11] Allen L et al 2003 Optical Angular Momentum (Bristol: IOP Publishing)
[12] Marrucchi L et al 2011 J. Opt. 13 064001
[13] Wang J et al 2012 Nat. Photon. 6 488
7-79
IOP Publishing
Lecture notes
Chapter 8
Radiation, scattering, interference, and
diffraction
This chapter continues the discussion of electromagnetic wave propagation, now

focusing on the results of wave incidence on various objects with more complex
shapes. Depending on the shape, the result of this interaction is called either scattering,
or diffraction, or interference. However, as the reader will see, the boundaries between
these effects are blurry, and their mathematical description may be conveniently based
on a single key calculation—the electric dipole radiation of a spherical wave by a
localized source. Naturally, I will start the chapter from this calculation, deriving it
from an even more general result—the ‘retarded potentials’ solution of the Maxwell
equations.
8.1 Retarded potentials

Let us start from finding the general solution of the macroscopic Maxwell equations
(6.99) in a dispersion-free, linear, uniform, isotropic medium, characterized by
frequency-independent, real ε and μ.1 The easiest way to perform this calculation is
to use the scalar (ϕ) and vector (A) potentials of electromagnetic field, defined via the
electric and magnetic fields by Eqs (6.7):
∂A
E = −∇ϕ − , B=∇×A. (8.1)
∂t
As was discussed in section 6.8, by imposing on the potentials the Lorenz gauge
condition (6.117),
1
When necessary (e.g. in the discussion of Cherenkov radiation in section 10.4), it will not be too difficult to
generalize the results to a dispersive medium.

1 ∂ϕ 1
∇⋅A+ = 0, with v 2 ≡ , (8.2)
v 2 ∂t εμ
which does not affect the fields E and B, the Maxwell equations may be used to
obtain a pair of very similar, simple equations (6.118) for the potentials:
1 ∂ 2ϕ ρ
∇2 ϕ − 2 2
=− , (8.3a )
v ∂t ε
1 ∂ 2A
∇2 A − = −μj. (8.3b)
v 2 ∂t 2
Let us find the general solution of these equations, assuming that the densities
ρ(r, t) and j(r, t) of the stand-alone charges and currents are known2. The idea of
such a solution is borrowed from electro- and magnetostatics. Indeed, for the
stationary case (∂/∂t = 0), the solutions of Eqs. (8.3) are given by the ready
generalization of, respectively, Eqs. (1.38) and (5.28) to a uniform, linear medium:
3
1
ϕ(r) =
4πε
∫ ρ(r′) rd−r′r′ , (8.4a )
3
A(r) ≡
μ
4π
∫ j(r′) rd−r′r′ . (8.4b)
As we know, these expressions may be derived by first calculating the potential of a

point source and then using the linear superposition principle for a system of such
sources.
Let us do the same for the time-dependent case, starting from the field induced by
a time-dependent point charge at the origin3:
ρ(r , t ) = q(t )δ(r). (8.5)
In this case Eq. (8.3a) is homogeneous everywhere but the origin:

1 ∂ 2ϕ
∇2 ϕ − = 0, at r ≠ 0. (8.6)
v 2 ∂t 2
2
This assumption will not prevent the results from being valid for the case when ρ(r, t) and j(r, t) should be
calculated self-consistently.
3
Admittedly, this expression does not satisfy the continuity equation (4.5), but this deficiency will be corrected
imminently, at the linear superposition stage—see Eqs. (8.17) below.
8-2
Due to the spherical symmetry of the problem, it is natural to look for a spherically
symmetric solution to this equation4. Thus, we may simplify the Laplace operator5
correspondingly and reduce Eq. (8.6) to
⎡1 ∂ ⎛ ∂ ⎞ 1 ∂2 ⎤
⎢ 2 ⎜r 2 ⎟ − 2 2 ⎥ ϕ = 0, at r ≠ 0. (8.7)
⎣ r ∂r ⎝ ∂r ⎠ v ∂t ⎦
If we now introduce a new variable χ ≡ rϕ, Eq. (8.7) is reduced to a 1D wave

equation
⎛ ∂2 1 ∂2 ⎞
⎜ 2 − 2 2 ⎟ χ = 0, at r ≠ 0. (8.8)
⎝ ∂r v ∂t ⎠
From discussions in chapter 7,6 we know that its general solution may be
represented as
⎛ r⎞ ⎛ r⎞
χ (r , t ) = χout ⎜t − ⎟ + χin ⎜t + ⎟ , (8.9)
⎝ v⎠ ⎝ v⎠
where χin and χout are (so far) arbitrary functions of one variable. The physical sense
of ϕout = χout/r is a spherical wave propagating from our source (at r = 0) to outer
space, i.e. exactly the solution we are looking for. On the other hand, ϕin = χin/r
describes a spherical wave that could be created by some distant spherically
symmetric source that converges exactly on our charge located at the origin—
evidently not the effect we want to consider here. Discarding this term, and returning
to ϕ = χ/r, we can write the solution (8.9) as
1 ⎛ r⎞
ϕ(r , t ) = χout ⎜t − ⎟ . (8.10)
r ⎝ v⎠
In order to calculate the function χout, let us consider the solution (8.10) at distances
r so small that the time derivative in Eq. (8.3a), with the right-hand side (8.5),
1 ∂ 2ϕ q(t )
∇2 ϕ − 2 2
=− δ(r), (8.11)
v ∂t ε
is much smaller that the spatial derivative (which diverges at r → 0). Then Eq. (8.11)
is reduced to the electrostatic equation, whose solution (8.4a) for the source (8.5) is
q(t )
ϕ(r → 0, t ) = . (8.12)
4πε r
4
Let me confess that this is not the general solution to Eq. (8.6). For example, it does nor describe the possible
waves created by other sources that pass by the considered charge q(t). However, such fields are irrelevant for
our current task: to calculate the field created by the charge q(t). The solution becomes general when it is
integrated (as it will be) over all charges of interest.
5
See, e.g. Eq. (A.67).
6
See also Part CM section 6.3.
8-3
Now requiring the two solutions, (8.10) and (8.12), to coincide at r ≪ vt, we obtain
χout(t) = q(t)/4πεr, so that Eq. (8.10) becomes
1 ⎛⎜ r⎞
ϕ(r , t ) = q t − ⎟. (8.13)
4πε r ⎝ v⎠
Just as was done in statics, this result may be readily generalized for the arbitrary
position rʹ of the point charge,
ρ(r , t ) = q(t )δ(r−r′) ≡ q(t )δ(R), (8.14)
where R is the distance between the field observation point r and the source position
point r′, i.e. the length of the vector,
R ≡ r − r ′, (8.15)
connecting these points—see figure 8.1.
Figure 8.1. Calculating the retarded potentials of a localized source.
Obviously, now Eq. (8.13) becomes

1 ⎛ R⎞
ϕ(r , t ) = q⎜t − ⎟ . (8.16)
4πε R ⎝ v⎠
Finally, we may use the linear superposition principle to write, for the arbitrary
charge distribution,
1 ⎛ ⎞ 3
ϕ(r , t ) =
4πε
∫ ρ⎜⎝r′, t − Rv ⎟⎠ dRr′ , (8.17a )
where integration is extended over all charges of the system under analysis. Acting
absolutely similarly, for the vector-potential we obtain7
⎛ ⎞ 3
A(r , t ) =
μ
4π
∫ j⎜⎝r′, t − Rv ⎟⎠ dRr′ . (8.17b)
The solutions (8.17) are called the retarded potentials8, the name signifying that
the observed fields are ‘retarded’ (delayed) in time by Δt = R/v relative to the source
7
Now nothing prevents functions ρ(r, t) and j(r, t) from satisfying the continuity relation.
8
As should be clear from the analogy of Eqs. (8.17) with their stationary forms (8.4), which were discussed,
respectively, in chapters 1 and 5, in the Gaussian units the retarded potential formulas are valid with the
coefficient 1/4π dropped in Eq. (8.17a), and replaced with the coefficient 1/c in Eq. (8.17b).
8-4
variations, due to the finite speed v of the electromagnetic wave propagation. These
solutions are so important that they deserve at least a couple of general remarks.
First, very remarkably, these simple expressions are exact solutions of the
macroscopic Maxwell equations (in a uniform, linear, dispersion-free) medium for
an arbitrary distribution of stand-alone charges and currents. They also may be
considered as the general solutions of these equations, provided that the integration
is extended over all field sources in the Universe—or at least in its part that affects
our observations.
Second, due to the mathematical similarity of the microscopic and macroscopic
Maxwell equations, Eqs. (8.17) are valid, with the coefficient replacement ε → ε0 and
μ → μ0, for the exact, rather than the macroscopic fields, provided that the functions
ρ(r, t) and j (r, t) describe not only stand-alone but all charges and currents in the
system. (Alternatively, this statement may be formulated as the validity of Eqs.
(8.17), with the same coefficient replacement, in the free space.)
Finally, Eqs. (8.17) may be plugged into Eq. (8.1), giving (after an explicit
differentiation) the so-called Jefimenko equations for fields E and B—similar in
structure to Eqs. (8.17), but more cumbersome. Conceptually, the existence of such
equations is good news, because they are free from the gauge ambiguity pertinent to
the potentials ϕ and A. However, the practical value of these explicit expressions
for the fields is not too high: for all applications I am aware of, it is easier to use
Eqs. (8.17) to calculate the particular expressions for the potentials first, and only
then calculate the fields from Eq. (8.1). Let me now present an (apparently, the most
important) example of this approach.
8.2 Electric dipole radiation

Consider again the problem that was discussed in electrostatics (section 3.1), namely
the field of a localized source with linear dimensions a ≪ r (see figure 8.1 again), but
now with time-dependent charge and/or current distributions. Using the arguments
of that discussion, in particular the condition expressed by Eq. (3.1), rʹ ≪ r, we may
apply the Taylor expansion (3.3), truncated to two leading terms,
f (R) = f (r) − r′ ⋅ ∇f (r) + ⋯, (8.18)
to the function f(R) ≡ R (for which ∇f(r) = ∇R = n, where n ≡ r/r is the unit vector
directed toward the observation point—see figure 8.1) to approximate the distance
R as
R ≈ r − r′ ⋅ n. (8.19)
In each of the retarded potential formulas (8.17), R participates in two places: in
the denominator and in the source’s time argument. If ρ and j change in time on
scale ∼1/ω, where ω is some characteristic frequency, then any change of the argument
(t − R/v) on that time scale, for example due to a change of R on the spatial scale
∼v/ω = 1/k, may substantially change these functions. Thus, the expansion (8.19)
may be applied to R in the argument (t − R/v) only if ka ≪ 1, i.e. if the system’s size a
is much smaller than the radiation wavelength λ = 2π/k. On the other hand, the
8-5
function 1/R changes relatively slowly, and for it even the first term of the expansion
(8.19) gives a good approximation as soon as a ≪ r, R. In this approximation,
Eq. (8.17a) yields
1 ⎛ ⎞ ⎛ ⎞
ϕ(r , t ) ≈
4πε r
∫ ρ⎜⎝r′, t − Rv ⎟⎠d 3r′ ≡ 4πε1 r Q⎜⎝t − Rv ⎟⎠, (8.20)
where Q(t) is the net electric charge of the localized system. Due to charge
conservation, this charge cannot change with time, so that the approximation
(8.20) describes just a static Coulomb field of our localized source, rather than a
radiated wave.
Let us, however, apply the similar approximation to the vector potential (8.17b):
⎛ ⎞
A(r , t ) ≈
μ
4π r
∫ j⎜⎝r′, t − Rv ⎟⎠d 3r′. (8.21)
According to Eq. (5.87), in statics the right-hand side of this expression would
vanish, but in dynamics this is no longer true. For example, if the current is due to a
non-relativistic motion9 of a system of point charges qk, we can write
∫ j(r′, t )d 3r′ = ∑qk rk̇ (t ) = dtd ∑qk rk(t ) ≡ p(̇ t ), (8.22)

k k
where p(t) is the dipole moment of the localized system, defined by Eq. (3.6). Now,
after the integration, we may keep only the first term of the approximation (8.19) in
the argument (t − R/v) as well, obtaining
μ ⎛⎜ r⎞
A(r , t ) ≈ ṗ t − ⎟ . (8.23)
4π r ⎝ v⎠
Let us analyze exactly what this result, valid in the limit ka ≪1, describes. The
second of Eqs. (8.1) allows us to calculate the magnetic field by the spatial
differentiation of A. At large distances r ≫ λ (i.e. in the so-called far field zone),
where Eq. (8.23) describes a virtually plane wave, the main contribution into this
derivative is given by the dipole moment factor:
μ ⎛ r⎞ μ ⎛ r⎞
B(r , t ) = ∇ × ṗ⎜t − ⎟ = − n × p̈⎜t − ⎟ . (8.24)
4π r ⎝ v⎠ 4π rv ⎝ v⎠
This expression means that the magnetic field, at the observation point, is
perpendicular to the vectors n and (the retarded value of) p̈, and its magnitude is
9
For relativistic particles, moving with velocities of the order of speed of light, one has to be more careful. As a
result, I will postpone the discussion of their radiation until chapter 10, i.e. until after the detailed discussion of
special relativity in chapter 9.
8-6
Figure 8.2. Far-zone fields of a localized source, contributing into its electric dipole radiation.
μ ⎛⎜ r⎞ 1 ⎛⎜ r⎞
B= p ̈ t − ⎟ sin Θ, i.e. H = p ̈ t − ⎟ sin Θ, (8.25)
4πrv ⎝ v⎠ 4πrv ⎝ v⎠
where Θ is the angle between those two vectors—see figure 8.210.
The most important feature of this result is that the time-dependent field decreases
very slowly (only as 1/r) with the distance from the source, so that the radial
component of the corresponding Poynting vector (7.9b)11,
Z ⎡ ⎛⎜ r ⎞⎟⎤2 2
Sr = ZH 2 = ⎢p ̈ t − ⎥ sin Θ (8.26)
(4π vr )2 ⎣ ⎝ v ⎠⎦
drops as 1/r2, i.e. the full instant power P of the emitted wave,
π
P≡ ∮r=const Sr d 2r = (4πZv)2 p2̈ 2π∫0 sin3 Θ d Θ =
Z 2
6π v 2
p̈ , (8.27)
does not depend on the distance from the source—as it should for radiation12.
This is the famous Larmor formula13 for the electric dipole radiation; it is the
dominant component of radiation by a localized system of charges—unless p̈ = 0.
Please notice its angular dependence: the radiation vanishes at the axis of the
retarded vector p̈ (where Θ = 0), and reaches its maximum in the plane perpendicular
to that axis.
In order to find the average power, Eq. (8.27) has to be averaged over a
sufficiently long time. In particular, if the source is monochromatic, p(t) = Re
[pωexp{−iωt}], with a time-independent vector pω, such averaging may be carried
out just over one period, giving an extra factor 2 in the denominator:
Zω4
P = ∣p ∣2 . (8.28)
12π v 2 ω
10
From the first of Eqs. (8.1), for the electric field in the first approximation (8.23) we would obtain −∂A/∂t =
−(1/4πεvr) p̈ (t − r/v) = −(Z/4πr)p̈ (t − r/v). The transverse component of this vector (see figure 8.2) is the proper
electric field E = ZH × n of the radiated wave, while its longitudinal component is exactly compensated by
(−∇ϕ) in the next term of the expansion of Eq. (8.17a) with respect to the small parameter r/λ ≪ 1.
11
Note the ‘doughnut’ dependence of Sr on the direction n, frequently used to visualize the dipole radiation.
12
In the Gaussian units, for free space (v = c), Eq. (8.27) reads P = (2/3c3)p 2̈ .
13
After J Larmor, who was first to derive it (in 1897) for the particular case of a single point charge q moving
with acceleration r ,̈ when p̈ = q r .̈
8-7
The easiest example of application of the formula is to a point charge oscillating,

with frequency ω, along a straight line (which we may take for axis z), with amplitude
a. In this case, p = qnzz(t) = nzqa Re [exp{−iωt}], and if the charge velocity amplitude,
aω, is much less than the wave speed v, we may use Eq. (8.28) with pω = qa, giving
Zq 2a 2ω4
P = . (8.29)
12π v 2
Applied to an electron (q = −e ≈ −1.6 × 10−19 C), initially rotating about a nucleus
at an atomic distance a ∼ 10−10 m, the Larmor formula shows14 that the energy loss
due to the dipole radiation is so large that it would cause the electron to collapse on
the atom’s nucleus in just ∼10−10 s. In the beginning of the 1900 s, this classical result
was one of the main arguments for the development of quantum mechanics, which
prevents such collapse of electrons in their lowest-energy (ground) quantum state.
Another example of a very useful application of Eq. (8.28) is the radio wave
radiation by a short, straight, symmetric antenna which is fed, for example, by a
TEM transmission line such as a coaxial cable—see figure 8.3.
Figure 8.3. The dipole antenna.
The exact solution of this problem is rather complex, because the law Iω(z) of the
current variation along antenna’s length should be calculated self-consistently with
the distribution of the electromagnetic field induced by the current in the surround-
ing space. (This fact is unfortunately ignored in some textbooks.) However, one
may argue that at l ≪ λ, the current should be largest in the feeding point (in
figure 8.3, taken for z = 0), vanish at antenna’s ends (z = ±l/2), and that the only
possible scale of the current variation in the antenna is l itself, so that the linear
function,
⎛ 2 ⎞
Iω(z ) = Iω(0)⎜1 − z ⎟ , (8.30)
⎝ l ⎠
gives a good approximation to the actual distribution—as it indeed does. Now we
can use the continuity equation ∂Q/∂t = I, i.e. −iωQω = Iω, to calculate the complex
14
Actually, the formula needs a numerical coefficient adjustment to account for the electron’s orbital (rather
than linear) motion—the task left for reader’s exercise. However, this adjustment does not affect the order-of-
magnitude estimate given above.
8-8
amplitude Qω(z) = iIω(z)sgn(z)/ω of the electric charge Q(z, t) = Re[Qωexp{ iωt}] of

the wire beyond point z, and from it, the amplitude of the linear density of charge
dQ ω(z ) 2I (0)
λ ω (z ) ≡ = −i ω sgn z . (8.31)
d z ωl
From here, the dipole moment’s amplitude is
l /2
Iω(0)
pω = 2 ∫0 λω(z )z dz = −i
2ω
l, (8.32)
so that Eq. (8.28) yields

ω4 Iω(0) 2 2 Z (kl )2 Iω(0) 2
P =Z l = , (8.33)
12πv 2 4ω 2 24π 2
where k = ω/v. The analogy between this result and the dissipation power,
P = Re Z Iω 2 /2 , in a lumped linear circuit element, allows the interpretation of
the first fraction in the last form of Eq. (8.33) as the real part of the antenna’s
impedance:
(kl )2
Re ZA = Z , (8.34)
24π
as felt by the transmission line.
According to Eq. (7.118), the wave traveling along the line toward the antenna is
fully radiated, i.e. not reflected back, only if ZA equals ZW of the line. As we know
from section 7.5 (and the solution of related problems), for typical TEM lines ZW ∼
Z0, while Eq. (8.34), which is only valid in the limit kl ≪ 1, shows that for radiation
into the free space (Z = Z0), ReZA is much less than Z0. Hence in order to reach the
impedance matching condition ZW = ZA, the antenna’s length should be increased—
as a more involved theory shows, to l ∼ λ/2. However, in many cases, practical
considerations make short antennas necessary. The example most frequently
encountered nowadays is the cell phone antennas, which use frequencies close to 1
or 2 GHz, with free-space wavelengths λ between 15 and 30 cm, i.e. much larger than
the phone size15. The quadratic dependence of the antenna’s efficiency on l,
following from Eq. (8.34), explains why every millimeter counts in the design of
such antennas, and why the designs are carefully optimized using software packages
for (virtually exact) numerical solution of time-dependent Maxwell equations for the
specific shape of the antenna and other phone parts16.
To conclude this section, let me note that if the wave source is not monochro-
matic, so that p(t) should be represented as a Fourier series,
15
The situation will be partly remedied by the planned transfer of the wireless mobile technology to its next (5
G) generation, with the frequencies moving to the 28 GHz, 37–39 GHz, and possibly even the 64–71 GHz
bands.
16
A partial list of popular software packages of this kind includes both publicly available codes such as NEC-2
(whose various versions are available online, e.g. at http://alioth.debian.org/projects/necpp/ and http://www.
qsl.net/4nec2/), and proprietary packages—such as Momentum from Aglient Technologies (now owned by
Hewlett-Packard), FEKO from EM Software & Systems, and XFdtd from Remcom.
8-9
p(t ) = Re ∑pωe−iωt (8.35)

ω
the terms corresponding to interference of spectral components with different

frequencies ω are averaged out at the time averaging of the Poynting vector, so
that the average radiated power is just a sum of contributions (8.28) from all
substantial frequency components.
8.3 Wave scattering

The formalism described above may be immediately used in the theory of
scattering—the phenomenon illustrated by figure 8.4. Generally, scattering is a
complex problem. However, in many cases it allows the so-called Born approx-
imation17, in which the scattered wave field’s effect on the scattering object is
assumed to be much weaker than that of the incident wave, and is neglected.
Figure 8.4. Schematic representation of scattering.
As the first example of this approach, let us consider the scattering of a plane
wave, propagating in free space (Z = Z0, v = c), by a free18 charged particle whose
motion may be described by non-relativistic classical mechanics. (This requires, in
particular, the incident wave not to be too powerful, so that the speed of the induced
charge motion remains much lower than the speed of light.) As was already
discussed at the derivation of Eq. (7.32), in this case the magnetic component of
the Lorentz force (5.10) is negligible in comparison with the force Fe = qE exerted by
its electric field. Thus, assuming that the incident wave is linearly polarized along
some axis x, the equation of the particle’s motion in the Born approximation is just
mx ̈ = qE(t), so that for the x-component px = qx of its dipole moment we can write
q2
p ̈ = qx ̈ = E (t ). (8.36)
m
17
Named after M Born, one of the founding fathers of quantum mechanics. Note, however, the basic idea of
this approach was developed in electromagnetic theory much earlier (1881) by Lord Rayleigh (born J Stuff,
1842–1919), whose numerous contributions to science include the discovery of argon.
18
As Eq. (7.30) shows, this calculation is also valid for an oscillator with its own frequency ω0 ≪ ω.
8-10
As we already know from section 8.2, oscillations of the dipole moment lead to
radiation of a wave with a wide angular distribution of intensity; in our case this is
the scattered wave—see figure 8.4. Its full power may be found by plugging
Eq. (8.36) into Eq. (8.27),
Z0 2 Z0q 4
P= 2
p̈ = E 2(t ), (8.37)
6π c 6π c 2m 2
so that for the average power we obtain
Z0q 4
P = Eω 2 . (8.38)
12π c 2m 2
Since the power is proportional to incident wave’s intensity S, it is customary to
characterize the scattering ability of the object by the ratio,
P P
σ≡ ≡ , (8.39)
Sincident Eω 2 /2Z0
which has the dimension of area and is called the full cross-section of scattering19.
For this measure, Eq. (8.38) yields the famous result
Z02q 4 μ02 q 4
σ= = , (8.40)
6π c 2m 2 6π m 2
which is called the Thomson scattering formula20, in particular when applied to an
electron. This relation is most frequently represented in the form21
8π 2 q2 1 −7 q
2
σ= rc , with rc ≡ ⋅ = 10 . (8.41)
3 4πε0 mc 2 m
This constant rc is called the classical radius of the particle (or sometimes the
‘Thomson scattering length’); for an electron (q = −e, m = me) it is close to 2.82 ×
10−15 m. Its possible interpretation is evident from the first form of Eq. (8.41) for rc:
at that distance between two similar particles, the potential energy q2/4πε0r of their
electrostatic interaction is equal to the particle’s rest-mass energy mc2.22
19
This definition parallels those accepted in the classical and quantum theories of particle scattering—see, e.g.
respectively, Part CM section 3.5 and Part QM section 3.3.
20
Named after Sir J J Thomson (1856–1940), the discoverer of the electron—and of isotopes as well! He should
not be confused with his son, G P Thomson, who discovered (simultaneously with C Davisson and L Germer)
the quantum-mechanical wave properties of the same electron.
21
In Gaussian units, this formula looks like rc = q2/mc2 (giving, of course, the same numerical value: for the
electron, rc ≈ 2.82 ×10−13 cm). This classical quantity should not be confused with the particle’s Compton
wavelength λc ≡ 2πℏ/mc (for the electron, close to 2.24 × 10−12 cm), which naturally arises in quantum
electrodynamics—see a brief discussion in the next chapter, and also Part QM section 1.1.
22
It is fascinating how smartly has the relativistic expression mc2 sneaked into the result (8.40)–(8.41), which
was obtained using a non-relativistic equation of particle motion. This was possible because the calculation
engaged electromagnetic waves, which propagate with the speed of light, and whose quanta (photons), as a
result, may be frequently treated as relativistic (moreover, ultra-relativistic) particles—see the next chapter.
8-11
Now we have to go back and establish the conditions under which the Born
approximation, when the field of the scattered wave is negligible, is indeed valid for a
point-object scattering. Since the scattered wave’s intensity, described by Eq. (8.26),
diverges as 1/r2, according to the definition (8.39) of the cross-section, it may become
comparable to Sincident at r2 ∼ σ. However, Eq. (8.38) itself is only valid if r ≫ λ, so
the Born approximation does not lead to a contradiction only if
σ ≪ λ2 . (8.42)
For the Thompson scattering by an electron, this condition means λ ≫ rc ∼ 3 × 10−15
m and is fulfilled for all frequencies up to very hard γ -rays with energies ∼100 MeV.
Possibly the most notable feature of result (8.40) is its independence of the
wave frequency. As follows from its derivation, particularly from Eq. (8.37), this
independence is intimately related to the unbound character of charge motion.
For bound charges, say for electrons in gas molecules, this result is only valid if
the wave frequency ω is much higher than all eigenfrequencies ωj of molecular
resonances. In the opposite limit, ω ≪ ωj, the result is dramatically different. Indeed,
in this limit we can approximate the molecule’s dipole moment by its static value (3.48)
p = α E. (8.43)
In the Born approximation, and in the absence of the molecular field effects
discussed in section 3.3, E in this expression is just the incident wave’s field, and
we can use Eq. (8.28) to calculate the power of the wave scattered by a single
molecule:
Z0ω4 2
P = α Eω 2 . (8.44)
4π c 2
Now, using the last form of the definition (8.39) of the cross-section, we obtain a
very simple result,
Z02ω4 2
σ= α , (8.45)
6πc 2
showing that in contrast to Eq. (8.40), at low frequencies σ grows as fast as ω4.
Now let us explore the effect of such Rayleigh scattering on wave propagation in a
gas, with a relatively low volumic density n. We may expect (and will prove in the
next section) that due to the randomness of molecule positions, the waves scattered
by individual molecules may be treated as incoherent ones, so that the total
scattering power may be calculated just as the sum of those scattered by each
molecule. We can use this fact to write the balance of the incident’s wave intensity in
a small volume dV of length (along the incident wave direction) dz, and area A
across it. Since such a segment includes ndV = nA dz molecules and, according to
definition (8.39), each of them scatters power Sσ = Pσ /A, the total scattered power
is nPσ dz ; hence the incident power’s change is
dP ≡ −nσP dz. (8.46)
8-12
Comparing this equation with the definition (7.213) of the wave attenuation
constant, applied to scattering23,
dP ≡ −αscatP dz. (8.47)
we see that the scattering gives the following contribution to attenuation: αscat = nσ.
From here, using Eq. (3.50) to write α = ε0(κ − 1)/n, where κ is the dielectric constant,
and Eq. (8.45) for σ, we obtain
k2 2π ω
αscat = (κ − 1)2 , where k ≡ = . (8.48)
6π n λ0 c
This is the famous Rayleigh scattering formula, which in particular explains the
colors of blue skies and red sunsets. Indeed, through the visible light spectrum, ω
changes almost two-fold; as a result, the scattering of blue components of sunlight is
an order of magnitude higher than that of its red components. More qualitatively,
for the air near the Earth’s surface, κ − 1 ≈ 6 × 10−4, and n ∼ 2.5 × 1025 m−3—see
section 3.3. Plugging these numbers into (8.47), we see that the characteristic length
lscat ≡ 1/αscat of scattering is ∼30 km for blue light and ∼200 km for red light24. The
effective thickness h of the Earth’s atmosphere is ∼10 km, so that the Sun looks just a
bit yellowish during most of the day. However, elementary geometry shows that at
the sunset, the light should travel the length l ∼ (REh)1/2 ≈ 300 km to reach an Earth-
surface observer; as a result, the blue components of Sun’s light spectrum are almost
completely scattered out, and even the red components are weakened substantially.
Now let us discuss scattering by objects with size of the order of, or even larger
than λ. For such extended objects, the phase difference factors (neglected above) step
in, leading in particular to the important effects of interference and diffraction, to
whose discussion we now proceed.
8.4 Interference and diffraction

These effects show up not as much in the total power of the scattered radiation, as in
its angular distribution. It is traditional to characterize this distribution by the
differential cross-section defined as
dσ Srr 2
≡ , (8.49)
dΩ Sincident
where r is the distance from the scatterer, at which the scattered wave is observed25.
Both the definition and notation may become more clear if we notice that according
to Eq. (8.26), at large distances (r ≫ a), the nominator on the right-hand side of
23
Sorry for using the same letter (α) for both the molecular polarizability and the wave attenuation, but both
notations are traditional. Hopefully, the subscript ‘scat’, marking α in the latter meaning, excludes any
possibility of confusion.
24
These values are approximate, because both n and (κ − 1) vary through the atmosphere’s thickness.
25
Just as in the case of the full cross-section, this definition is also similar to that accepted at the particle
scattering—see, e.g. Part CM section 3.5 and Part QM section 3.3.
8-13
Eq. (8.49), and hence the differential cross-section as the whole, does not depend on
r, and that its integral over the total solid angle Ω = 4π coincides with the full cross-
section defined by Eq. (8.39):
∮4π ddΩσ d Ω = 1
Sincident
∮
r2
4π
Srd Ω =
1
Sincident
∮r=const Srd 2r
(8.50)
P
= ≡ σ.
Sincident
For example, according to Eq. (8.26), the angular distribution of the radiation
scattered by a single dipole is rather broad; in particular, in the quasi-static case
(8.43) and in the Born approximation,
dσ ⎛ αk 2 ⎞2
=⎜ ⎟ sin2 Θ. (8.51)
d Ω ⎝ 4πε0 ⎠
If the wave is scattered by a small dielectric body, with a characteristic size a ≪ λ (i.e.
ka ≪ 1), then all its parts re-radiate the incident wave coherently. Hence, we can
calculate it in the similar way, just replacing the molecular dipole moment (8.43)
with the total dipole moment of the object—see Eq. (3.45):
p = PV = (κ − 1) ε0EV , (8.52)
where V ∼ a is the body’s volume. As a result, the differential cross-section may be
3
obtained from Eq. (8.51) with the replacement αmol → (κ − 1)ε0V:

dσ ⎛ k 2V ⎞2
=⎜ ⎟ (κ − 1)2 sin2 Θ, (8.53)
d Ω ⎝ 4π ⎠
i.e. follows the same sin2Θ law. The situation for extended objects, with at least one
dimension of the order of or larger than the wavelength, is different: here we have to
take into account the phase shifts introduced by various parts of the body. Let us
analyze this issue first for an arbitrary collection of similar point scatterers located at
points rj.
If the wave vector of the incident plane wave is k0, the wave’s field has the phase
factor exp{ik0 · r}—see (7.79). At the location of the jth scattering center, the factor
equals exp{ik0 · rj}, so that the local dipole vectors p and the scattered wave they
create are proportional to this factor. On its way to the observation point r, the
scattered wave, with the wave vector k (with k = k0), acquires an additional phase
factor exp{ik · (r − rj)}, so that the scattered wave field is proportional to
exp{i k 0 ⋅ rj + i k(r − rj )} = exp{i (k 0 − k) ⋅ rj + i k ⋅ r}
(8.54)
= e i k⋅r exp{ −i (k − k 0) ⋅ rj }.
Since the first factor in the last expression does not depend on rj, in order to calculate
the total scattering wave it is sufficient to sum up the last phase factors, exp{−iq · rj},
where the vector
q ≡ k − k0 (8.55)
8-14
has the physical sense of the wave vector change at scattering26. It may look like the
phase factor depends on the choice of the reference frame. However, according to
Eq. (7.42), the average intensity of the scattered wave is proportional to EωEω*, i.e.
to the following real scalar function of vector q:
⎛ ⎞⎛ ⎞*
F (q) = ⎜∑ exp{ −i q ⋅ rj }⎟⎜∑ exp{ −i q ⋅ rj ′}⎟⎟
⎜ ⎟⎜
⎝ j ⎠⎝ j ′ ⎠ (8.56)
= ∑ exp{i q ⋅ (rj − rj ′)} ≡ I (q) , 2
j, j ′
where the complex function

I (q ) ≡ ∑ exp {−iq ⋅ rj } (8.57)
j
is called the phase sum, and may be calculated in any reference frame, without
affecting the final result (8.56).
So, in addition to the sin2Θ factor, the differential cross-section (8.49) of
scattering by an extended object is also proportional to the scattering function
(8.56). Its double-sum form is convenient to notice that for a system of many (N ≫ 1)
of similar but randomly located scatterers, only the terms with j = jʹ accumulate at
summation, so that F(q) and hence dσ/dΩ scale as N rather than N2—thus justifying
again our treatment of the Rayleigh scattering problem in the previous section.
Let us start using Eq. (8.56) by applying it to a simple problem of just two similar
small scatterers, separated by a fixed distance a:
2
F (q) = ∑ exp{i q ⋅ (rj − rj ′)} = 2 + exp{ −iqaa} + exp{iqaa}
j , j ′= 1 (8.58)
qa
= 2(1 + cos qaa ) = 4 cos a , 2
2
where qa ≡ q · a/a is the component of the vector q along the vector a connecting the
scatterers. The apparent simplicity of this result may be a bit misleading, because the
mutual plane of the vectors k and k0 (and hence of the vector q) does not necessarily
coincide with the mutual plane of vectors k0 and Eω, so that the scattering angle θ
between k and k0 is generally different from (π/2 − Θ)—see figure 8.5.
Moreover, the angle between the vectors q and a (within their common plane) is
one more parameter independent of both θ and Θ. As a result, the angular
dependence27 of the scattered wave’s intensity (and hence dσ/dΩ) that depends on
all three angles may be rather involved, but some of its details do not affect the basic
physics of the interference/diffraction.
26
In quantum mechanics, ℏq has a very clear sense of the momentum transferred from the scattering object to
the scattered particle (for example, a photon) and this terminology is sometimes smuggled even into classical
electrodynamics texts.
27
In optics, such patterns, observed as dark and bright spots on a screen, are called interference fringes.
8-15
Figure 8.5. The angles important for the general scattering problem.
Thus let me consider only the simple case when the vectors k, k0, and a are all in
the same plane (figure 8.6a), with k0 perpendicular to a. Then, with our choice of
coordinates, qa = qx = ksin θ, and Eq. (8.58) is reduced to
ka sin θ
F (q) = 4 cos2 . (8.59)
2
This function always has two maxima, at θ = 0 and θ = π, and, if the product ka is large
enough, also other maxima at special angles θn that satisfy the simple condition
ka sin θn = 2πn , i.e. a sin θn = nλ . (8.60)
Figure 8.6. The simplest cases of (a) interference and (b) diffraction.
As figure 8.6a shows, this condition may be readily understood as that of the in-
phase addition (the constructive interference) of two coherent waves scattered from
the two points, when the difference between their paths toward the observer, asin θ,
equals an integer number of wavelengths. At each such maximum, F = 4, due to the
doubling of the wave amplitude and hence quadrupling its power.
If the distance between the point scatterers is large (ka ≫ 1), the first maxima
(8.60) correspond to small scattering angles, θ ≪ 1. For this region, Eq. (8.59) is
reduced to a simple sinusoidal dependence of function F on the angle θ. Moreover,
within the range of small θ, the wave polarization factor sin2 Θ is virtually constant,
so that the scattered wave intensity and hence the differential cross-section
dσ kaθ
∝ F (q) = 4 cos2 . (8.61)
dΩ 2
8-16
This is of course the simple interference pattern, well known from Young’s two-slit
experiment28. (As will be discussed in the next section, the theoretical description of
the two-slit experiment is more complex than that of the Born scattering, but is
preferable experimentally, because at such scattering, the wave of intensity (8.61) has
to be observed on the backdrop of a stronger incident wave that propagates in
almost the same direction, θ = 0.)
A very similar analysis of scattering from N > 2 similar, equidistant scatterers,
located along the same straight line (left for the reader’s exercise), shows that the
positions (8.60) of the constructive interference maxima do not change (because the
derivation of this condition is still applicable to each pair of adjacent scatterers), but
the increase of N makes these peaks sharper and sharper.
Now let me jump to the limit N → 0, in which we may ignore the scatterers’
discreteness. The resulting pattern is similar to that at scattering by a continuous thin
rod (see figure 8.6b), so let us first spell out the Born scattering formula by an
arbitrary an extended, continuous, uniform dielectric body. Transferring Eq. (8.56)
from the sum to an integral, for the differential cross-section we obtain
dσ ⎛ k 2 ⎞2 ⎛ k 2 ⎞2
= ⎜ ⎟ (κ − 1)2 F (q)sin2 Θ ≡ ⎜ ⎟ (κ − 1)2 I (q) 2 sin2 Θ, (8.62)
d Ω ⎝ 4π ⎠ ⎝ 4π ⎠
where I(q) now becomes the phase integral29,
I (q) = ∫V exp{ −i q ⋅ r′}d 3r′ , (8.63)
with the dimensionality of volume.

Now we may return to the particular case of a thin rod (with both dimensions of
the cross-section’s area much smaller than λ, but an arbitrary length a), otherwise
keeping the same simple geometry as for two point scatterers—see figure 8.6b. In this
case the phase integral is just
+a /2 exp{ − iqxa /2} − exp{ − iqxa /2} sin ξ
∫−a /2
I (q ) = A exp{ − iqxx′}dx′ = A
− iq
=V
ξ
, (8.64)
where V = Aa is the volume of the rod, and ξ is the dimensionless parameter defined
as
qxa ka sin θ
ξ≡ ≡ . (8.65)
2 2
28
This experiment was described as early as in 1803 by T Young—one more universal genius of science, who
has also introduced the Young modulus in the elasticity theory (see, e.g. Part CM chapter 7), in addition to
numerous other achievements—including deciphering Egyptian hieroglyphs! The two-slit experiment firmly
established the wave picture of light, only to be replaced by the dualistic photon-versus-wave picture,
formalized by quantum electrodynamics 100+ years later.
29
Since the observation point’s position r does not participate in this formula explicitly, the prime sign in rʹ
could be dropped, but I keep it as a reminder that the integral is taken over points rʹ of the scattering object.
8-17
The fraction participating in Eq. (8.64) is met in physics so frequently that is has
deserved the special name sinc (not ‘sync’, please!) function:
sin ξ
sinc ξ ≡ . (8.66)
ξ
Obviously, this function, plotted in figure 8.7, vanishes at all points ξn = πn, with
integer n, apart from the point with n = 0: sinc ξ0 = sinc 0 = 1.
Figure 8.7. The sinc function.
The function F(q) = V2sinc2 ξ, resulting from Eq. (8.64), is plotted with the red
line in figure 8.8 and is called the Fraunhofer diffraction pattern.
Figure 8.8. The Fraunhofer diffraction pattern (solid red line) and its envelope 1/ξ2 (dashed red line). For
comparison, the blue line shows the standard interference pattern cos2 ξ —cf equation (8.59).
Note that it oscillates with the same argument period Δ(kasin θ) = 2π/ka as the
interference pattern (8.59) from two point scatterers (shown with the blue line in
figure 8.8). However, at the interference, the scattered wave intensity vanishes at
angles θnʹ that satisfy the condition
ka sin θn′ 1
=n+ , (8.67)
2π 2
8-18
i.e. when the optical paths difference asin α equals to a semi-integer number of
wavelengths λ/2 = π/k, and hence the two waves from the scatterers arrive to the
observer in anti-phase—the so-called destructive interference. On the other hand, for
the diffraction from a continuous rod the minima occur at a different set of
scattering angles:
ka sin θn
= n, (8.68)
2π
i.e. exactly where the two-point interference pattern has its maxima—see figure 8.8
again. The reason for this relation is that the wave diffraction on the rod may be
considered as a simultaneous interference of waves from all its elementary frag-
ments, and exactly at the observation angles when the rod edges give waves with
phases shifted by 2πn, the interior points of the rod give waves with all possible
phases, with their algebraic sum equal to zero. As is even more visible in figure 8.8,
at the diffraction, the intensity oscillations are limited by a rapidly decreasing
envelope function 1/ξ2 (while at the two-point interference, the oscillations retain the
same amplitude). The reason for this fast decrease is that with each Fraunhofer
diffraction period, a smaller and smaller fraction of the road gives an unbalanced
contribution to the scattered wave.
If the rod’s length is small (ka ≪ 1, i.e. a ≪ λ), then the sinc function’s argument ξ
is small at all scattering angles θ, so I(q) ≈ V, and Eq. (8.62) is reduced to Eq. (8.53).
In the opposite limit, a ≫ λ, the first zeros of the function I(q) correspond to very
small angles θ, for which sin Θ ≈ 1, so that the differential cross-section is just
dσ ⎛ k 2 ⎞2 kaθ
= ⎜ ⎟ (κ − 1)2 sinc 2 , (8.69)
d Ω ⎝ 4π ⎠ 2
i.e. figure 8.8 shows the scattering intensity as a function of the direction toward the
observation point—if this point is within the plane containing the rod.
Finally, let us discuss a problem of a large importance for applications: calculate
the positions of the maxima of the interference pattern arising at the incidence of a
plane wave on a very large 3D periodic system of point scatterers. For that, first of
all, let us quantify the notion of 3D periodicity. The periodicity in one dimension is
simple: the system we are considering (say, the positions of point scatterers) should
be invariant with respect to the linear translation by some period a, and hence by
any multiple sa of this period. (Here s is any integer.) Anticipating the 3D
generalization, we may require any of the possible translation vectors R to be equal
sa, where the primitive vector a is directed along the (only) axis of the 1D system.
Now we are ready for the common definition of the 3D periodicity—as the
invariance of the system with respect to the translation by any vector of the
following set:
3
R= ∑sl al , (8.70)
l=1
8-19
where sl are three independent integers and {al} is a set of three linearly independent
primitive vectors. The set of geometric points described by Eq. (8.70) is called the
Bravais lattice; perhaps the most non-trivial feature of this relation is that the vectors
al should not necessarily be orthogonal to each other. (That requirement would
severely restrict the set of possible lattices, and make it unsuitable for the
description, for example, of many solid-state crystals.) For the scattering problem
we are considering, let us assume that the position rj of each scatterer coincides with
one of the points R of some Bravais lattice, with a given set of primitive vectors al, so
that the index j is coding the set of integers {s1, s2, s3}.
Now let us consider a similar Bravais lattice, but in the reciprocal (wave-number)
space, numbered by independent integers {t1, t2, t3}:
3
am ″ × am ′
Q= ∑ tm b m , with bm = 2π
am ⋅ (am ″ × am ′)
, (8.71)
m=1
where the indices m, mʹ, and mʺ are all different. This is the so-called reciprocal
lattice, which plays an important role in all physics of periodic structures, in
particular in the quantum energy-band theory30. To reveal its most important
property, and thus justify the above definition of the primitive vectors bm, let us
calculate the scalar product
3 3
am ″ × am ′
R ⋅ Q≡ ∑ sl tmal ⋅ bm ≡ 2π ∑ sl tmal ⋅ am ⋅ (am ″ × am ′)
l, m = 1 l, m = 1
3
(8.72)
a ⋅ (am ″ × am ′)
≡ 2π ∑ sl tk l .
l, m = 1
am ⋅ (am ″ × am ′)
Applying to the nominator of the last fraction the operand rotation rule of vector
algebra31, we see that it is equal to zero if l ≠ m, while for l = m the whole fraction is
evidently equal to 1. Thus the double sum (8.72) is reduced to a single sum,
3 3
R ⋅ Q = 2π ∑sl tl = 2π ∑nl , (8.73)
l=1 l=1
where each of the products nl ≡ sltl is an integer and hence their sum,
3
n≡ ∑nl ≡ s1t1 + s2t2 + s3t3, (8.74)
l=1
is an integer as well, so that the main property of the direct/reciprocal lattice couple
is very simple:
30
See, e.g. Part QM section 3.4, where several particular Bravais lattices R, and their reciprocals Q, are
considered.
31
See, e.g. Eq. (A.48).
8-20
R ⋅ Q = 2πn , and exp { −i R ⋅ Q} = 1. (8.75)

Now returning to the scattering function (8.56), we see that if the vector q ≡ k − k0
coincides with any vector Q of the reciprocal lattice, then all terms of the phase sum
(8.57) take their largest possible values (equal to 1), and hence the sum as the whole
is largest as well, giving a constructive interference maximum. This equality, q = Q,
where Q is given by Eq. (8.71), is called the von Laue condition of the constructive
interference; it is, in particular, the basis of all field of the x-ray crystallography of
solids and polymers—the main tool for revealing their atomic/molecular structure32.
In order to recast the von Laue condition in a more vivid, geometric form, let us
consider one of the vectors Q of the reciprocal lattice, corresponding to a certain
integer n in Eq. (8.75), and notice that if that relation is satisfied for one point R of
the direct Bravais lattice (8.70), i.e. for one set of integers {s1, s2, s3}, it is also
satisfied for a 2D system of other integer sets, which may be parameterized, for
example, by two integers S1 and S2:
s1′ = s1 + S1t3, s2′ = s2 + S2t3, s3′ = s3 − S1t1 − S2t2. (8.76)
Indeed, each of these sets has the same value of the integer n, defined by Eq. (8.74),
as the original one:
n′ ≡ s1′t1 + s2′t2 + s3′t3 ≡ (s1 + S1t3) t1 + (s2 + S2t3) t2 + (s3 − S1t1 − S2t2 ) t3 = n . (8.77)
Since according to Eq. (8.75), the vector of distance between any pair of the
corresponding points of the direct Bravais lattice (8.70),
ΔR = ΔS1t3a1 + ΔS2t3a2 − (ΔS1t1 + ΔS2t2 )a3, (8.78)
satisfies the condition ΔR · Q = 2πΔn = 0, this vector is normal to the (fixed) vector
Q. Hence, all the points corresponding to the 2D set (8.76), with arbitrary integers S1
and S2, are located on one geometric plane, called the crystal (or ‘lattice’) plane. In a
3D system of N ≫ 1 scatterers (such as N ∼1023 atoms in a ∼1-cm3 solid crystal),
with all linear dimensions comparable, such a plane contains ∼N2/3 ≫ 1 points. As a
result, the constructive interference peaks are very sharp.
Now rewriting (8.75) as a relation for the vector R’s component along the
vector Q,
2π Q
RQ = n, where RQ ≡ R ⋅ nQ ≡ R ⋅ , and Q ≡ Q , (8.79)
Q Q
we see that the parallel crystal planes, corresponding to different numbers n (but the
same Q) are located in space periodically, with the smallest distance
32
For more reading on this important topic I can recommend, for example, the classical monograph [1]. Note
that it uses the alternative popular name of the field, once again illustrating how blurry the boundary between
interference and diffraction is.
8-21
2π
d= , (8.80)
Q
so that the von Laue condition q = Q may be rewritten as the following rule for the
possible magnitudes of the scattering vector q ≡ k − k0:
2πn
q= . (8.81)
d
Figure 8.9a shows the diagram of the three wave vectors k, k0, and q, taking into
account the elastic scattering condition ∣k∣ = ∣k0∣ = k ≡ 2π/λ:
Figure 8.9. Deriving the Bragg rule: (a) from the von Laue condition (in the reciprocal space) and (b) from a
direct-space diagram. Note that the scattering angle θ equals 2α.
From the diagram, we immediately obtain the famous Bragg rule33 for the (equal)
angles α ≡ θ/2 between the crystal plane and each of the vectors k and k0:
q πn
k sin α = = , i.e. 2d sin α = nλ . (8.82)
2 d
The physical sense of this relation is very simple—see figure 8.9b (drawn in the
‘direct’ space of the radius-vectors r, rather than in the reciprocal space of the wave
vectors as figure 8.9a). It shows that if the Bragg condition (8.82) is satisfied, the
total difference 2dsin α of the optical paths of two waves, partly reflected from the
adjacent crystal planes, is equal to an integer number of wavelengths, so these waves
interfere constructively.
Finally, note that the von Laue and Bragg rules, as well as the similar condition
(8.60) for the 1D system of scatterers, are valid not only in the Born approximation,
but also follow from any adequate theory of scattering. This is because the phase
sum (8.57) does not depend on the magnitude of the wave propagating from an
elementary scatterer, provided that they are all similar.
33
Named after Sir William Bragg and his son Sir William Lawrence Bragg who were the first to demonstrate
(in 1912) x-ray diffraction by atoms in crystals. The Braggs’ experiments made the existence of atoms (before
that, a hypothetical notion, which had been ignored by many physicists) indisputable.
8-22
8.5 The Huygens principle

The Born approximation is very convenient for tracing the basic features of (and the
difference between) the phenomena of interference and diffraction. Unfortunately,
this approximation, based on the relative weakness of the scattered wave, cannot be
used for more typical experimental implementations of these phenomena, for
example, the Young’s two-slit experiment or the diffraction on a single slit or
orifice—see, e.g. Figure 8.10. Indeed, in such experiments, the orifice size a is
typically much larger than the light’s wavelength and, as a result, no clear
decomposition of the fields to the ‘incident’ and ‘scattered’ waves is possible.
Figure 8.10. Deriving the Huygens principle.
However, for such experiments, another approximation called the Huygens (or
‘Huygens–Fresnel’) principle34 is very instrumental. In this approach, the wave
beyond the screen is represented as a linear superposition of spherical waves of the
type (8.17), as if they were emitted by every point of the incident wave’s front that
has arrived at the orifice. This approximation is valid if the following strong
conditions are satisfied:
λ ≪ a ≪ r, (8.83)
where r is the distance of the observation point from the orifice. In addition, as we
have seen in the last section, at small λ/a the diffraction phenomena are confined
to angles θ ∼ 1/ka ∼ λ/a ≪ 1. For observation at such small angles, the
mathematical expression of the Huygens principle, for the complex amplitude
fω(r) of a monochromatic wave f(r, t) = Re[fωe−iωt], is given by the following
simple formula
ikR
fω (r) = C ∫orifice fω (r′) eR d 2r′ . (8.84)
34
Named after C Huygens (1629–95) who conjectured the wave nature of light (which remained controversial
for more than a century, until T Young’s experiments) and A-J Fresnel (1788–1827) who developed a
quantitative theory of diffraction, and in particular gave a mathematical formulation of the Huygens principle.
(Note that Eq. (8.91), sufficient for the purposes of this course, is not its most general form.)
8-23
Here f is any transverse component of any of wave’s fields (either E or H)35, R is the
distance between point rʹ at the orifice and the observation point r (i.e. the magnitude
of vector R ≡ r − rʹ), and C is a complex constant.
Before describing the proof of Eq. (8.84), let me carry out its sanity check—which
also will give us the constant C. Let us see what the Huygens principle gives for the
case when the field under the integral is a plane wave with the complex amplitude
fω(z), propagating along axis z, with an unlimited x–y front (i.e. when there is no
opaque screen at all), so we should take the whole [x, y] plane, say with zʹ = 0, as the
integration area in Eq. (8.84)—see figure 8.11.
Figure 8.11. Applying the Huygens principle to a plane incident wave.
Then, for the observation point with coordinates x = 0, y = 0, and z ≫ λ,

Eq. (8.84) yields
exp{ ik (x′2 + y′2 + z 2 )1/2 }
∫ ∫
fω (z ) = Cfω (0) dx′ dy′
(x′2 + y′2 + z 2 )
1/2
. (8.85)
Before specifying the integration limits, let us consider the range ∣xʹ∣, ∣yʹ∣ ≪ z. In this
range the square root, participating in Eq. (8.85) twice, may be approximated as
1/2
⎛ x′2 + y′2 ⎞
1/2
⎛ x′2 + y′2 ⎞
(x′2 + y′2 + z 2 ) ≡ z⎜1 + ⎟ ≈ z ⎜1 + ⎟
⎝ z2 ⎠ ⎝ 2z 2 ⎠ (8.86)
x′2 + y′2
≡z + .
2z
The denominator of Eq. (8.85) is a much slower function of xʹ and yʹ than the
exponent, and in it (as we will check a posteriori), it is sufficient to keep just the main,
first term of the expansion (8.86). With that, Eq. (8.85) becomes
e ikz 2 2 ikz
fω (z ) = Cfω (0)
z
∫ dx′∫ dy′ exp ik(x′ 2z+ y′ ) = Cfω (0) ez IxIy, (8.87)
where Ix and Iy are two similar integrals; for example,
35
The fact that the Huygens principle is valid for any field component should not be too surprising. Due to
condition a ≫ λ, the real boundary conditions at the orifice edges are not important; what is important is only
that the screen that limits the orifice is opaque. Because of this, the Huygens principle’s expression (8.84) is part
of the so-called scalar theory of diffraction. (In this course, I will not have time to pursue the more accurate,
vector theory of these effects - see, e.g. chapter 11 of the classical monograph [2].)
8-24
ikx′2 ⎛ 2z ⎞1/2
Ix = ∫ exp dx′ = ⎜ ⎟ ∫ exp {iξ 2}dξ
2z ⎝k⎠
(8.88)
⎛ 2z ⎞1/2 ⎡ ⎤
= ⎜ ⎟ ⎢⎣ ∫ 2
cos (ξ )dξ + i ∫ 2
sin (ξ )dξ⎥⎦ ,
⎝k⎠
where ξ ≡ (k/2z)1/2x′. These are the so-called Fresnel integrals. I will discuss them in
more detail in the next section (see, in particular, figure 8.13), and for now, only one
property36 of these integrals is important for us: if taken in symmetric limits [−ξ0, +ξ0],
both of them rapidly converge to the same value, (π/2)1/2, as soon as ξ0 becomes much
larger than 1. This means that even if we do not impose any exact limits on the
integration area in Eq. (8.85) this integral converges to the value
e ikz ⎧⎛ 2z ⎞1/2 ⎡⎜⎛ π ⎟⎞1/2 ⎛ π ⎞1/2 ⎤⎫

2
⎛ 2π i ⎞
fω (z ) = Cfω (0) ⎨⎜ ⎟ ⎢ + i ⎜ ⎟ ⎥⎬ = ⎜C ⎟f (0)e ikz , (8.89)
z ⎩⎝ k ⎠ ⎣ 2 ⎝ ⎠ ⎝ ⎠
2 ⎦⎭ ⎝ k ⎠ω
due to contributions from the central area with linear size of the order of Δξ ∼ 1, i.e.
⎛ z ⎞1/2
Δx ∼ Δy ∼ ⎜ ⎟ ∼ (λz )1/2 , (8.90)
⎝k ⎠
so that the contributions from the front points rʹ well beyond the range (8.90) are
negligible37. (Within our assumptions (8.83), which in particular require λ to be
much less than z, the diffraction angle Δx/z ∼ Δy/z ∼ (λ/z)1/2, corresponding to the
important area of the front, is small.) According to Eq. (8.89), in order to sustain the
unperturbed plane wave propagation, fω(z) = fω(0)eikz, the constant C has to be
taken equal to k/2πi. Thus, the Huygens principle’s prediction (8.84), in its final
form, reads
ikR
k
fω (r) =
2πi
∫orifice fω (r′) eR d 2r′ , (8.91)
and describes, in particular, the straight propagation of the plane wave (in a uniform
media).
Let me pause to emphasize how non-trivial this result is. It would be a natural
corollary of Eqs. (8.25) (and the linear superposition principle) if all points of the
orifice were filled with point scatterers that re-emit all the incident waves into
spherical waves. However, as it follows from the above proof, the Huygens principle
is also valid if there is nothing in the orifice but the free space!
36
See, e.g. (A.37).
37
This result very is natural, because the function exp{ikR} oscillates fast with the change of rʹ, so that the
contributions from various front points are averaged out. Indeed, the only reason why the central part of plane
[xʹ, yʹ] gives a non-zero contribution (8.89) to fω(z) is that the phase exponents stops oscillating as (xʹ2 + yʹ2) is
reduced below ∼z/k—see Eq. (8.86).
8-25
This is why let us discuss a proof of the principle,38 based on Green’s theorem
(2.207). Let us apply this theorem to the function f = fω, where fω is the complex
amplitude of a scalar component of one of the wave’s fields, which satisfies the
Helmholtz equation (7.204),
(∇2 + k 2 )fω (r) = 0, (8.92)
and the function g = Gω, which is the time Fourier image of the corresponding
Green’s function. The latter function may be defined, as usual, as the solution to the
same equation, but with the added delta-functional right-hand side with an arbitrary
coefficient, for example,
(∇2 + k 2 ) Gω(r , r′) = −4πδ(r − r′). (8.93)
Using Eqs. (8.92) and (8.93) to express the Laplace operators of the functions fω and
Gω, we may rewrite Eq. (2.207) as
∫V {fω [−k 2Gω(r, r′) − 4πδ(r − r′)] − Gω(r, r′)⎡⎣−k 2fω ⎤⎦} d 3r
⎡ ∂Gω(r , r′) ∂f ⎤ (8.94)
= ∮ ⎢fω
S⎣ ∂n
− Gω(r , r′) ω ⎥d 2r ,
∂n ⎦
where n is the outward normal to the surface S limiting the integration volume V.
Two terms on the left-hand side of this relation cancel, so that after swapping the
arguments r and rʹ we obtain
⎡ ⎤
−4π fω (r) = ∮S ⎢⎣fω (r′) ∂Gω∂(nr′′, r) − gω(r′, r) ∂G∂ωn(′r′) ⎥⎦d 2r′. (8.95)
This relation is only correct if the selected volume V includes the point r
(otherwise we would not obtain its left-hand side from the integration of the
delta-function), but does not include the genuine source of the wave—otherwise Eq.
(8.92) would have a non-zero right-hand side. Let r be the field observation point, V
be all the source-free half-space (for example, the half-space right of the screen in
figure 8.10), so that S is the surface of the screen, including the orifice. Then the
right-hand side of Eq. (8.95) describes the field at the observation point r, induced by
the wave passing through the orifice points rʹ. Since no waves are emitted by the
opaque parts of the screen, we can limit the integration by the orifice area39.
Assuming also that the opaque parts of the screen do not re-emit the waves
38
This proof was given in 1882 by G Kirchhoff.
39
Actually, this is a non-trivial point of the proof. Indeed, it may be shown that the solution of Eq. (8.94)
identically equals zero if f(rʹ) and ∂f(rʹ)/∂nʹ vanish together at any part of the boundary. A more careful analysis
of this issue (which is the task of the formal vector theory of diffraction, which I will not have time to pursue)
confirms our intuition-based conclusion.
8-26
‘radiated’ by the orifice, we can take the solution of Eq. (8.93) to be the retarded
potential for the free space40:
e ikR
Gω(r , r′) = . (8.96)
R
Plugging this expression into Eq. (8.82) we obtain
⎡ ⎛ ikR ⎞ ⎛ e ikR ⎞ ∂fω (r′) ⎤
−4π fω (r) = ∮orifice ⎢⎣fω (r′) ∂∂n′ ⎜⎝ eR ⎟−⎜ ⎟
⎠ ⎝ R ⎠ ∂n′ ⎦
⎥d 2r′ . (8.97)
This is the so-called Kirchhoff (or ‘Fresnel–Kirchhoff’) integral. (Again, with the
integration extended over all boundaries of the volume V, this would be an exact
mathematical result.) Now, let us make two additional approximations. The first of
them stems from Eq. (8.83): at ka ≫ 1, the wave’s spatial dependence in the orifice
area may be represented as
fω (r′) = (a slow function of r′ ) × exp {i k 0 ⋅ r′}, (8.98)
where ‘slow’ means a function that changes on the scale of a rather than λ. If, also,
kR ≫ 1, then the differentiation in Eq. (8.97) may be, in both instances, limited to
the rapidly changing exponents, giving
ikR
−4π fω (r) = ∮orifice i(k + k 0) ⋅ n′ eR f (r′)d 2r′ , (8.99)
Second, if all observation angles are small, we can take k · nʹ ≈ k0 · nʹ ≈ −k. With
that, Eq. (8.99) is reduced to Eq. (8.91) expressing the Huygens principle.
It is clear that the principle immediately gives a very simple description of the
interference of waves passing through two small holes in the screen. Indeed, if the
holes’ sizes are negligible in comparison with the distance a between them (though
still much larger than the wavelength!), Eq. (8.91) yields
kf1,2 A1,2
fω (r) = c1e ikR1 + c2e ikR2, with c1,2 ≡ , (8.100)
2πiR1,2
where R1,2 are the distances between the holes and the observation point, and A1,2
are the hole areas. For the wave intensity, Eq. (8.100) yields
S ∝ fω f ω* = c1 2 + c2 2
+ 2 c1 c2 cos [k (R1 − R2 ) + ϕ ] ,
(8.101)
where ϕ ≡ arg c1 − arg c2 .
The first two terms in this relation clearly represent the intensities of the partial
waves passed through each hole, while the last one is the result of their interference.
The interference pattern’s contrast ratio
40
It follows, e.g. from Eq. (8.16) with a monochromatic source q(t) = qωexp{−iωt}, with the amplitude qω =
4πε that fits the right-hand side of Eq. (8.93).
8-27
Smax ⎛ c1 + c2 ⎞
2
R ≡ =⎜ ⎟ , (8.102)
Smin ⎝ c1 − c2 ⎠
is largest (infinite) when both waves have equal amplitudes.
The analysis of the interference pattern is simple if the line connecting the holes is
perpendicular to wave vector k ≈ k0—see figure 8.6a. Selecting the coordinate axes
as shown in that figure, and using for distances R1,2 the same expansion as in
Eq. (8.86), for the interference term in Eq. (8.101) we obtain
⎛ kxa ⎞
cos [k (R1 − R2 ) + ϕ ] ≈ cos ⎜ + ϕ⎟ . (8.103)
⎝ z ⎠
This means that the intensity does not depend on y, i.e. the interference pattern in the
plane of constant z is a set of straight, parallel strips, perpendicular to the vector a,
with the period given by Eq. (8.60), i.e. by the Bragg law41. Note that this result is
strictly valid only at (x2 + y2) ≪ z2; it is straightforward to use the next term in the
Taylor expansion (8.73) to show that farther from the interference pattern center the
strips start to diverge.
8.6 Fresnel and Fraunhofer diffraction patterns

Now let us use the Huygens principle to analyze a more complex problem: a plane
wave’s diffraction on a long, straight slit of a constant width a (figure 8.12).
Figure 8.12. Diffraction on a slit.
According to Eq. (8.83), in order to use the Huygens principle for the problem
analysis we need to have λ ≪ a ≪ z. Moreover, the simple version (8.91) of the
principle is only valid for small observation angles, ∣x∣ ≪ z. Note, however, that the
relation between two dimensionless numbers, z/a and a/λ, both much less than 1, is
41
The phase shift ϕ vanishes at the normal incidence of a plane wave on the holes. Note, however, that the
spatial shift of the interference pattern following from Eq. (8.103), Δx = −(z/ka)ϕ, is extremely convenient for
the experimental measurement of the phase shift between two waves, in particular if it is induced by some
factor (such as insertion of a transparent object into one of the interferometer’s arms, etc) that may be turned
on/off at will.
8-28
so far arbitrary. As we will see in a minute, this relation determines the type of
observed diffraction pattern.
Let us apply Eq. (8.91) to our current problem (figure 8.12), for the sake of
simplicity assuming normal wave incidence, and taking z = 0 at the screen plane:
k +a +∞
exp {ik [(x − x′)2 + y′2 + z 2 ]1/2 }
fω (x , z ) = f0
2π i
∫−a dx′ ∫−∞ dy′
[(x − x′)2 + y′2 + z 2 ]1/2
, (8.104)
where f0 ≡ fω(xʹ, 0) = const is the incident wave’s amplitude. This is the same integral
as in Eq. (8.85), except for the finite limits for the integration variable xʹ, and may be
simplified similarly, using the small-angle condition (x − xʹ)2 + yʹ2 ≪ z2:
k e ikz +a /2 +∞
ik [(x − x′)2 + y′2 ]
fω (x , z ) ≈ f0
2πi z −a /2
∫ dx′ ∫−∞ dy′ exp
2z
(8.105)
k e ikz
≡ f0 IxIy.
2π i z
The integral over yʹ is the same as in the last section,
+∞
iky′2 ⎛ 2πiz ⎞1/2
Iy ≡ ∫−∞ exp dy′=⎜ ⎟ , (8.106)
2z ⎝ k ⎠
but the integral over xʹ is more complicated, because of its finite limits:
+a /2
ik (x − x′)2
Ix ≡ ∫−a /2 exp
2z
dx′ . (8.107)
It may be simplified in the following two (opposite) limits.

(i) Fraunhofer diffraction takes place when z/a ≫ a/λ—the relation which may be
rewritten either as a ≪ (zλ)1/2, or as ka2 ≪ z. In this limit, the ratio kxʹ2/z is negligibly
small for all values of xʹ under the integral, and we can approximate it as
+a /2
ik (x 2 − 2xx′ + x′2 ) +a /2
ik (x 2 − 2xx′)
Ix = ∫−a /2 exp
2z
dx′ ≈
−a /2
exp ∫ 2z
dx′
(8.108)
ikx 2 +a /2 ⎧ ikxx′ ⎫ 2z ⎧ ikx 2 ⎫ kxa
≡ exp ∫ exp⎨ − ⎬dx′ = exp ⎨ ⎬ sin ,
2z −a /2 ⎩ z ⎭ kx ⎩ 2z ⎭ 2z
so that Eq. (8.105) yields
k e ikz 2z ⎛ 2π i z ⎞1/2 ⎧ ikx 2 ⎫ kxa

fω (x , z ) ≈ f0 ⎜ ⎟ exp ⎨ ⎬ sin , (8.109)
2π i z kx ⎝ k ⎠ ⎩ 2z ⎭ 2z
and hence the relative wave intensity is
8-29
2
S (x , z ) f (x , z ) 8z kxa 2 ka 2 ⎛ kaθ ⎞
= ω = sin2 ≡ sinc 2⎜ ⎟, (8.110)
S0 f0 πkx 2
2z π z ⎝ 2 ⎠
where S0 is the (average) intensity of the incident wave and θ ≡ x/z ≪ 1 is

the observation angle. Comparing this expression with Eq. (8.69), we see that
this diffraction pattern is exactly the same as that of a similar (uniform, 1D) object in
the Born approximation—see the red line in figure 8.8. Note again that the angular
width δθ of the Fraunhofer pattern is of the order of 1/ka, so that its linear width
δx = z δθ is of the order of z/ka ∼ zλ/a.42 Hence the condition of the Fraunhofer
approximation’s validity may be also represented as a ≪ δx.
(ii) Fresnel diffraction. In the opposite limit of a relatively wide slit, with a ≫ δx =
z δθ ∼ z/ka ∼ zλ/a, i.e. ka2 ≫ z, the diffraction patterns at two slit edges are well
separated. Hence, near each edge (for example, near xʹ = −a/2) we may simplify
Eq. (8.107) as
+∞
ik (x − x′)2 ⎛ 2z ⎞1/2 +∞
Ix(x ) ≈ ∫−a /2 exp dx′ ≡ ⎜ ⎟ ∫(k /2z ) exp {iζ 2}dζ, (8.111)
2z ⎝k⎠ 1/2(x+a /2)
and express it via the special functions called the Fresnel integrals43,
⎛ 2 ⎞1/2 ξ ⎛ 2 ⎞1/2 ξ
C (ξ ) ≡ ⎜ ⎟ ∫ cos(ζ 2 )dζ, S (ξ ) ≡ ⎜ ⎟ ∫ sin(ζ 2 )dζ, (8.112)
⎝π ⎠ 0 ⎝π ⎠ 0
whose plots are shown in figure 8.13a. As was mentioned above, at large values of
their argument (ξ), both functions tend to ½.
Figure 8.13. (a) The Fresnel integrals and (b) their parametric representation.
42
Note also that since in this limit ka2 ≪ z, Eq. (8.97) shows that even the maximum value S(0, z) of the
diffracted wave intensity is much less than the intensity S0 of the incident wave. This is natural, because the
incident power S0a per unit length of the slit is now distributed over a much larger width δx ≫ a, so that S(0, z)
∼ S0 (a/δx) ≪ S0.
43
Slightly different definitions of these functions, mostly affecting the constant factors, may also be met in
literature.
8-30
Plugging this expression into Eqs. (8.105) and (8.111), for the diffracted wave
intensity in the Fresnel limit (i.e. at ∣ x + a/2 ∣ ≪ a) we obtain
⎧ 2⎫
1 ⎪⎡⎢ ⎛⎛ k ⎞1/2 ⎛⎜ a ⎞⎟⎞ 1 ⎤⎥ ⎡ ⎛⎛ k ⎞1/2 ⎛ ⎞⎞ 1 ⎤ ⎪
2
S (x , z ) a
= ⎨ C ⎜⎜ ⎟ x + ⎟+ + ⎢S ⎜⎜ ⎟ ⎜x + ⎟⎟ + ⎥ ⎬ . (8.113)
2⎪ ⎝ ⎠ ⎝ 2 ⎠⎠ 2 ⎥⎦ ⎣⎢ ⎝⎝ 2z ⎠ ⎝ 2 ⎠⎠ 2 ⎥⎦ ⎪
S0 ⎩⎢⎣ ⎝ 2z ⎭
A plot of this function (figure 8.14) shows that the diffraction pattern is very
peculiar: while in the ‘shade’ region x < −a/2 the wave intensity fades monotonically,
the transition to the ‘light’ region within the gap (x > −a/2) is accompanied by
intensity oscillations, just as at the Fraunhofer diffraction—cf figure 8.8.
Figure 8.14. The Fresnel diffraction pattern.
This behavior, which is described by the following asymptotes,
⎧ 1 sin(ξ 2 − π /4) ⎛ k ⎞1/2

⎪ 1+ , for ξ ≡ ⎜ ⎟ (x + a /2) → +∞ ,
S ⎪ π ξ ⎝ 2z ⎠
→⎨ (8.114)
S0 ⎪ 1
⎪ , for ξ → −∞ ,
⎩ 4πξ 2
is essentially an artifact of observing just the wave intensity (i.e. its real amplitude)
rather than its phase as well. Indeed, as may be seen even more clearly from the
parametric representation of the Fresnel integrals, shown in figure 8.13b, these
functions oscillate similarly at large positive and negative values of their argument.
(This famous pattern is called either the Euler spiral or the Cornu spiral.) Physically,
this means that the wave diffraction at the slit edge leads to similar oscillations of its
phase at x < −a/2 and x > −a/2; however, in the latter region (i.e. inside the slit) the
diffracted wave overlaps the incident wave passing through the slit directly and their
interference reveals the phase oscillations, making them visible in the measured
intensity as well.
Note that according to Eq. (8.113) the linear scale δx of the Fresnel diffraction
pattern is of the order of (2z/k)1/2, i.e. is complies with the estimate (8.90). If the slit is
8-31
gradually narrowed so that its width a becomes comparable to δx,44 the Fresnel
diffraction patterns from both edges start to ‘collide’ (interfere). The resulting wave,
fully described by Eq. (8.107), is just a sum of two contributions of the type (8.111)
from both edges of the slit. The resulting interference pattern is somewhat
complicated, and only when a becomes substantially less than δx, it is reduced to
the simple Fraunhofer pattern (8.110). Of course, this crossover from the Fresnel to
Fraunhofer diffraction may be also observed at fixed wavelength λ and slit width a
by increasing z, i.e. by measuring the diffraction pattern farther and farther away
from the slit.
Note also that the Fraunhofer limit is always valid if the diffraction is measured
as a function of the diffraction angle θ alone, i.e. effectively at infinity, z → ∞. This
may be done, for example, by collecting the diffracted wave with a ‘positive’
(converging) lens, and observing the diffraction pattern in its focal plane.
8.7 Geometrical optics placeholder

Behind all these details, I would not like the reader to miss the main feature of the
Fresnel diffraction, which has an overwhelming practical significance. Namely, apart
from narrow diffraction ‘cones’ (actually, parabolic-shaped regions) with the lateral
scale δx ∼ (λz)1/2, the wave far behind a slit of width a ≫ λ, δx, repeats the field just
behind the slit, i.e. reproduces the unperturbed incident wave inside the slit, and has
negligible intensity in the shade regions outside it. An evident generalization of this fact
is that when a plane wave (in particular an electromagnetic wave) passes any opaque
object of a large size a ≫ λ, it propagates around it by distances z up to
∼a2/λ along straight lines with virtually negligible diffraction effects. This fact gives
the strict foundation for the very notion of the wave ray (or beam), as the line
perpendicular to the local front of a quasi-plane wave. In a uniform medium such a ray
follows a straight line, but it refracts in accordance with the Snell law at the interface of
two media with different wave speeds v, i.e. different values of the refraction index. The
notion of rays enables the whole field of geometric (or ‘geometrical’) optics, devoted
mostly to ray tracing in various (sometimes very complex) systems.
This is why, at this point, an E&M course that followed scientific logic more
faithfully than this one would give an extended discussion of the geometric and
quasi-geometric optics, including (as a minimum45) such vital topics as
• the so-called lensmaker’s equation expressing the focus length f of a lens via
the curvature radii of its spherical surfaces and the refraction index of the lens
material,
• the thin lens formula relating the image distance from the lens via f and the
source distance,
• the concepts of basic optical instruments such as telescopes and microscopes, and
44
Note that this condition may be also rewritten as a ∼ δx, i.e. z/a ∼ a/λ.
45
Admittedly, even this list leaves aside several spectacular effects, including such a beauty as conical refraction
in biaxial crystals—see, e.g. chapter 15 of the textbook [2].
8-32
• the concepts of the spherical, angular, and chromatic aberrations (image

distortions).
However, since I have made a (possibly, incorrect) decision to follow the common
tradition in selecting the main topics for this course, I do not have time left for such
discussion. Still, I am placing this ‘placeholder’ pseudo-section to relay my deep
conviction that any educated physicist has to know the basics of geometric optics. If
the reader has not had exposure to this subject during his or her undergraduate
studies, I highly recommend at least browsing one of the available textbooks46.
8.8 Fraunhofer diffraction from more complex scatterers

So far, our discussion of diffraction has been limited to a very simple geometry—a
single slit in an otherwise opaque screen (figure 8.12). However, in the most
important Fraunhofer limit, z ≫ ka2, it is easy to obtain a very simple expression
for the plane wave diffraction/interference by a plane orifice (with a linear size ∼a) of
an arbitrary shape. Indeed, the evident 2D generalization of the approximation
(8.106) and (8.107) is
2 2
IxIy = ∫orifice exp ik[(x − x′) 2z+ (y − y′) ] dx′dy′
(8.115)
⎧ ik (x 2 + y 2 ) ⎫ ⎧ kxx′ kyy′ ⎫
≈ exp ⎨ ⎬ ∫orifice exp⎨ −i −i ⎬dx′dy′ ,
⎩ 2z ⎭ ⎩ z z ⎭
so that apart from the inconsequential total phase factor, Eq. (8.105) is reduced to
f (ρ) ∝ f0 ∫orifice exp{−i κ ⋅ ρ′}d 2ρ′ ≡ f0 ∫all screen T (ρ′)exp{−i κ ⋅ ρ′}d 2ρ′, (8.116)
where the 2D vector κ (not to be confused with wave vector k that is virtually
perpendicular to κ!) is defined as
ρ
κ ≡ k ≈ q ≡ k − k 0, (8.117)
z
ρ = {x, y} and ρʹ = {xʹ, yʹ} are 2D radius-vectors in the, respectively, observation
and screen planes (both nearly normal to vectors k and k0). In the last form of
Eq. (8.116), the function T(ρʹ) describes the screen’s transparency at point ρʹ, and the
integral is over the whole screen plane zʹ = 0. (Although the strict equivalence of the
two forms of Eq. (8.116) is only valid if T(ρʹ) equals either 1 or 0, its last form may be
readily obtained from Eq. (8.91) with f(rʹ) = T(ρʹ)f0 for any transparency profile,
provided that T(ρʹ) is an arbitrary function, but changes only at distances much
larger than λ ≡ 2π/k.)
46
My top recommendation for that purpose would be chapters 3–6 and section 8.6 in [2]. A simpler alternative
is chapter 10 in [3]. Note also that the venerable field of optical microscopy is currently revitalized by
holographic/tomographic methods, using the scattered wave’s phase information. These methods are
particularly productive in biology and medicine—see, e.g. [4, 5].
8-33
From the mathematical point of view, the last form of Eq. (8.116) is just the 2D
spatial Fourier transform of the function T(ρʹ), with the variable κ defined by the
observation point’s position: ρ = (z/k)κ = (zλ/2π)κ. This interpretation is useful because
of the experience we all have with the Fourier transform, mostly in the context of its
time/frequency applications. For example, if the orifice is a single small hole, T(ρʹ) may
be approximated by a delta-function, so that Eq. (8.116) yields |f(ρ)| ≈ const. This
corresponds (at least for the small diffraction angles θ ≡ ρ/z for which the Huygens
approximation is valid) to a spherical wave spreading from the point-like orifice. Next,
for two small holes, Eq. (8.116) immediately gives the Young interference pattern
(8.103). Let me now use Eq. (8.116) to analyze the simplest (and most important) 1D
transparency profiles, leaving 2D cases for the reader’s exercise.
(i) A single slit of width a (figure 8.12) may be described by transparency
⎧1, for x′ < a /2,
T (ρ′) = ⎨ (8.118)
⎩ 0, otherwise.
Its substitution into Eq. (8.116) yields

+a /2
exp{ −iκxa /2} − exp {iκxa /2}
f (ρ) ∝ f0 ∫−a /2 exp{ −iκxx′}dx′ = f0
−iκx
(8.119)
⎛ κxa ⎞ ⎛ kxa ⎞
∝ sinc⎜ ⎟ = sinc⎜ ⎟,
⎝ 2 ⎠ ⎝ 2z ⎠
naturally returning us to Eqs (8.64) and (8.110), and hence to the red lines in figure
8.8 for the wave intensity. (Please note again that Eq. (8.116) describes only the
Fraunhofer, but not the Fresnel diffraction!)
(ii) Two narrow, similar, parallel slits with a much larger distance a between
them, may be described by taking
T (ρ′) ∝ δ(x′ − a /2) + δ(x′ + a /2), (8.120)
so that Eq. (8.116) yields the generic 1D interference pattern,
⎡ ⎤
⎣
iκ a
f (ρ) ∝ f0 ⎢exp − x
2 { } + exp { }iκxa
2
⎥ ∝ cos
⎦
κxa
2
= cos
kxa
2z
, (8.121)
whose intensity is shown with the blue line in figure 8.8.

(iii) In a more realistic version of the Young-type two-slit experiment, each slit has
a width (say, w) which is much larger than light wavelength λ, but still much smaller
than the slit spacing a. This situation may be described by the following trans-
parency function
⎧1, for x′ ± a /2 < w /2,
T (ρ′) = ∑⎨⎩0, otherwise,
(8.122)
±
8-34
Figure 8.15. Young’s double-slit interference pattern for a finite-width slit.
for which Eq. (8.116) yields a natural combination of results (8.119) (with a replaced
with w) and (8.121):
⎛ kxw ⎞ ⎛ kxa ⎞
f (r) ∝ sinc⎜ ⎟ cos ⎜ ⎟. (8.123)
⎝ 2z ⎠ ⎝ 2z ⎠
This is the usual interference pattern, modulated with a Fraunhofer-diffraction
envelope (shown with the dashed blue line in figure 8.15). Since the function sinc2 ξ
decreases very fast beyond its first zeros at ξ = ±π, the practical number of
observable interference fringes is close to 2a/w.
(iv) A structure very useful for experimental and engineering practice is a set of
many parallel, similar slits, called a diffraction grating47. Indeed, if the slit width is
much smaller than the grating period d, then the transparency function may be
approximated as
+∞
T (ρ′) ∝ ∑ δ(x′ − nd ) (8.124)
n =−∞
and Eq. (8.116) yields

n =+∞ n =+∞
⎧ nkxd ⎫
f (ρ) ∝ ∑ exp{ −inκxd } = ∑ exp⎨ −i ⎬. (8.125)
⎩ z ⎭
n =−∞ n =−∞
This sum vanishes for all values of κxd that are not multiples of 2π, so that the
result describes sharp intensity peaks at the following diffraction angles:
⎛x ⎞ ⎛κ ⎞ 2π λ
θm ≡ ⎜ ⎟ = ⎜ x⎟ = m = m. (8.126)
⎝ z ⎠m ⎝ k ⎠m kd d
Taking into account that this result is only valid for small angles ∣θm ∣ ≪ 1, it may be
interpreted exactly as Eq. (8.59)—see figure 8.6a. However, in contrast with the
interference (8.121) from two slits, the destructive interference from many slits kills
the net wave as soon as the angle is even slightly different from each value (8.60).
This is very convenient for spectroscopic purposes, because the diffraction lines
47
The rudimentary diffraction grating effect, produced by the parallel fibers of a bird’s feather, was discovered
as early as in 1673 by J Gregory—who also invented the reflecting (‘Gregorian’) telescope.
8-35
produced by multi-frequency waves do not overlap even if the frequencies of their

adjacent components are very close.
Two unavoidable features of practical diffraction gratings make their properties
different from this simple, ideal picture. First, the finite number N of slits, which may
be described by limiting the sum (8.125) to the interval n = [−N/2, +N/2], results in a
non-zero spread, δθ/θ ∼ 1/N, of each diffraction peak, and hence in the reduction of
the grating’s spectral resolution. (Unintentional variations of the inter-slit distance d
have a similar effect, so that before the advent of high-resolution photolithography,
special high-precision mechanical tools have been used for grating fabrication.)
Second, a finite slit width w leads to the diffraction peak pattern modulation by a
sinc2(kwθ/2) envelope, similarly to the pattern shown in figure 8.15. Actually, for
spectroscopic purposes such modulation is sometimes a plus, because only one
diffraction peak (say, with m = ±1) is practically used, and if the frequency spectrum
of the analyzed wave is very broad (covers more than one octave), the higher peaks
produce undesirable hindrance. Because of this reason, w is frequently selected to be
equal exactly to d/2, thus suppressing each other diffraction maximum. Moreover,
sometimes semi-transparent films are used to make the transparency function T(rʹ)
continuous and close to the sinusoidal one:
T⎛ 2πx′ ⎞
T (ρ′) ≈ T0 + T1 cos
2πx′
d
≡ T0 + 1 ⎜exp
2⎝ { }
i
2πx′
d {
+ exp −i
d }⎟ . (8.127)
⎠
Plugging the last expression into Eq. (8.116) and integrating, we see that the output
wave consists of just three components: the direct-passing wave (proportional to T0)
and two diffracted waves (proportional to T1) propagating in the directions of the
two lowest Bragg angles, θ±1 = ±λ/d.
The same relation (8.116) may be also used to obtain one more general (and
rather curious) result, called the Babinet principle. Consider two experiments with
diffraction of similar plane waves on two ‘complementary’ screens who together
would cover the whole plane, without a hole or an overlap. (Think, for example,
about an opaque disk of radius R and a large opaque screen with a round orifice of
the same radius.) Then, according to the Babinet principle, the diffracted wave
patterns produced by these two screens in all directions with θ ≠ 0 are identical. The
proof of this principle is straightforward: since the transparency functions produced
by the screens are complementary in the following sense:
T (ρ′) ≡ T1(ρ′) + T2(ρ′) = 1, (8.128)
and (in the Fraunhofer approximation only!) the diffracted wave is a Fourier
transform of T(ρʹ), which is a linear operation, we obtain
f1 (ρ) + f2 (ρ) = f0 (ρ), (8.129)
where f0 is the wave ‘scattered’ by the composite screen with T0(ρʹ) ≡ 1, i.e. the
unperturbed initial wave propagating in the initial direction (θ = 0). In all other
directions, f1 = −f2, i.e. the diffracted waves are indeed similar apart from the
8-36
difference in sign—which is equivalent to a phase shift by ±π. However, it is

important to remember that the Babinet principle notwithstanding, in real experi-
ments, with screens at finite distances, the diffracted waves may interfere with the
unperturbed plane wave f0(ρ), leading to different diffraction patterns in the cases 1
and 2—see, e.g. Figure 8.14 and its discussion.
8.9 Magnetic dipole and electric quadrupole radiation

Throughout this chapter, we have seen how many important results may be obtained
from Eq. (8.26) for the electric dipole radiation by a small-size source (figure 8.1).
Only in rare cases when this radiation is absent, for example if the dipole moment p
of the source equals zero (or does not change at time—either at all, or at the
frequency of our interest), higher-order effects may be important. I will discuss the
main two of them, the quadrupole electric radiation and the dipole magnetic
radiation.
In section 8.2 above, the electric dipole radiation was calculated by plugging the
expansion (8.19) into the exact formula (8.17b) for the retarded vector-potential
A(r, t). Let us make a more exact calculation, by keeping the second term of that
expansion as well:
⎛ R⎞ ⎛ r r′ ⋅ n ⎞⎟ ⎛ r′ ⋅ n ⎞⎟ r
j⎜r′ , t − ⎟ ≈ j⎜r′ , t − + ≡ j⎜r′ , t′ + , where t′ ≡ t − . (8.130)
⎝ v⎠ ⎝ v v ⎠ ⎝ v ⎠ v
Since the expansion is only valid if the last term in the time argument of j is relatively
small, in the Taylor expansion of j with respect to that argument we may keep just
two leading terms:
⎛ r′ ⋅ n ⎞ 1 ∂
j⎜r′ , t′ + ⎟ ≈ j(r′ , t′) + j(r′ , t′)(r′ ⋅ n) , (8.131)
⎝ v ⎠ v ∂t′
so that Eq. (8.17b) yields A = Ae + Aʹ, where Ae is the electric dipole contribution as
given by Eq. (8.23), and Aʹ is the new term of the next order in the small parameter
rʹ ≪ r:
μ ∂
A′(r , t ) =
4π rv ∂t′
∫ j(r′, t′)(r′ ⋅ n)d 3r′. (8.132)
Just as it was done in section 8.2, let us evaluate this term for a system of non-
relativistic particles with electric charges qk and radius-vectors rk(t):
μ ⎡d ⎤
A′(r , t ) = ⎢ ∑qk rk̇ (rk ⋅ n)⎥ . (8.133)
4π rv ⎢⎣ dt k ⎥⎦
t =t ′
Using the ‘bac minus cab’ identity of the vector algebra again48, the vector operand
of Eq. (8.133) may be rewritten as
48
If you need, see, e.g. Eq. (A.47).
8-37
1 1
rk̇ (rk ⋅ n) ≡ rk̇ (rk ⋅ n) + rk̇ (n ⋅ rk )
2 2
1 1 1
= (rk × rk̇ ) × n + rk (n ⋅ rk̇ ) + rk̇ (n ⋅ rk ) (8.134)
2 2 2
1 1 d
≡ (rk × rk̇ ) × n + [rk (n ⋅ rk )],
2 2 dt
so that the right-hand side of Eq. (8.133) may be represented as a sum of two terms,
Aʹ = Am + Aq, where
μ μ ⎛ r⎞
A m(r , t ) = ṁ (t′) × n = ṁ ⎜t − ⎟ × n ,
4π rv 4π rv ⎝ v⎠
(8.135)
1
with m(t ) ≡ ∑rk (t ) × qk rk̇ (t ) ,
2 k
μ ⎡ d2 ⎤
A q(r , t ) = ⎢ ∑ q rk ( n ⋅ rk ) ⎥ . (8.136)
8π rv ⎢⎣ dt 2 k k ⎦⎥t=t′
Comparing the second of Eqs. (8.135) with Eq. (5.91), we see that m is just the
total magnetic moment of the source. On the other hand, the first of Eqs. (8.135) is
absolutely similar in structure to Eq. (8.23), with p replaced with (m × n)/v, so that
for the corresponding component of the magnetic field it gives (in the same
approximation r ≫ λ) a result similar to Eq. (8.24):
μ ⎡ ⎛ r⎞ ⎤ μ ⎡ ⎛ r⎞ ⎤
B m(r , t ) = ∇ × ⎢ṁ ⎜t − ⎟ × n⎥ = − n × ⎢m̈ ⎜t − ⎟ × n⎥ . (8.137)
4π rv ⎣ ⎝ v⎠ ⎦ 4π rv 2 ⎣ ⎝ v⎠ ⎦
According to this expression, just as at the electric dipole radiation, the vector B is
perpendicular to the vector n, and its magnitude is also proportional to sinΘ, where
Θ is now the angle between the direction toward the observation point and the
second time derivative of the vector m—rather than p:
μ ⎛ r⎞
Bm = m̈ ⎜t − ⎟ sin Θ.
2 ⎝
(8.138)
4π rv v⎠
As the result, the intensity of this magnetic dipole radiation has a similar angular
distribution:
Z ⎡ ⎛ r ⎟⎞⎤2 2
Sr = ZH 2 = ⎢m̈ ⎜t −
⎥ sin Θ, (8.139)
(4π v 2r )2 ⎣ ⎝ v ⎠⎦
—cf Eq. (8.26), except for the (generally) different meaning of the angle Θ.
Note, however, that this radiation is usually much weaker than its electric-dipole
counterpart. For example, for a non-relativistic particle with electric charge q,
moving on a trajectory of size ∼a, the electric dipole moment is of the order of qa,
8-38
while its magnetic moment scales as qa2ω, where ω is the motion frequency. As a
result, the ratio of the magnetic and electric dipole radiation intensities is of the
order of (aω/v)2, i.e. the squared ratio of the particle’s speed to the speed of emitted
waves—that has to be much smaller than 1 for our non-relativistic calculation to be
valid.
The angular distribution of the electric quadrupole radiation, described by
Eq. (8.136), is more complicated. In order to show this, we may add to Aq a vector
parallel to n (i.e. along the wave’s propagation), obtaining
μ ⎛ r⎞
A q(r , t ) → Q ⎜̈ t − ⎟ , where Q ≡
24π rv ⎝ v⎠
∑qk{ 3rk (n ⋅ rk ) − nrk2}, (8.140)
k
because this addition does not give any contribution to the transverse component of
the electric and magnetic fields, i.e. to the radiated wave. According to the above
definition of the vector Q , its Cartesian components may be represented as
3
Qj = ∑ Q jj′nj′, (8.141)
j ′= 1
where Q jj ʹ are the elements of the electric quadrupole tensor of the system—see the
last of Eqs. (3.4):
Q jj ′ = ∑qk(3rjrj′ − r 2δjj′)k . (8.142)
k
Let me hope that the reader has already obtained some expertise in the calculation of
this tensor’s components for the simple systems listed in problems 3.2 and 3.3.
Now taking the curl of the first of Eq. (8.140) at r ≫ λ, we obtain
μ ⎛ r⎞
B q (r , t ) = − n × Q ⃛ ⎜t − ⎟ . (8.143)
24π rv 2 ⎝ v⎠
This expression is similar to Eqs (8.24) or (8.137), but according to Eqs (8.140) and
(8.142) the components of the vector Q do depend on the direction of the vector n,
leading to a different angular dependence of Sr.
As the simplest example, let us consider a system of two equal point electric
charges moving symmetrically, at equal distances d(t) ≪ λ from a stationary
center—see figure 8.16.
Figure 8.16. The simplest system emitting electric quadrupole radiation.
8-39
Due to the symmetry of the system, its dipole moments p and m (and hence its
electric and magnetic dipole radiation) vanish, but the quadrupole tensor (8.142) still
has non-zero components. With the coordinate choice shown in figure 8.16, these
components are diagonal:
Q xx = Q yy = −2qd 2, Q zz = 4qd 2. (8.144)
With the axis x selected within the common plane of the axis z and the direction n
toward the source (figure 8.16), so that nx = sin Θ, ny = 0, and nz = cos Θ, Eq. (8.141)
yields
Q x = −2qd 2 sin Θ, Q y = 0, Q z = 4qd 2 cos Θ, (8.145)
and the vector product in Eq. (8.143) has only one non-vanishing Cartesian
component:
3
d
(n × Q ⃛ ) y = nzQ x⃛ − nxQ z ⃛ = −6q sin Θ cos Θ 3 [d 2(t )]. (8.146)
dt
As a result, the quadrupole radiation intensity, S ∝ Bq2, is proportional to sin2 Θ
cos2 Θ, i.e. vanishes not only along the symmetry axis of the system (as the electric-
dipole and the magnetic-dipole radiations would do), but also in all directions
perpendicular to this axis, reaching its maxima at Θ = ±π/4.
For more complex systems, the angular distribution of the electric quadrupole
radiation may be different, but it may be proved that its total (instant) power always
obeys the simple formula
3
Z
Pq =
1440πv 4
∑ (Q jj⃛ ′)2 . (8.147)
j , j ′= 1
Let me finish this section by giving, also without proof, one more fact important
for some applications: due to their different spatial structures, the magnetic-dipole
and electric-quadrupole radiation fields do not interfere, i.e. the total power of
radiation (neglecting the electric-dipole and higher multipole terms) may be found as
the sum of these components, calculated independently. In contrast, if the electric-
dipole and magnetic-dipole radiations of the same system are comparable, they
typically interfere coherently, so that their radiation fields (rather than powers)
should be added.
8.10 Problems
Problem 8.1. In the electric-dipole approximation, calculate the angular distribution
and the total power of electromagnetic radiation using the following classical
model of the hydrogen atom: an electron rotating, at a constant distance R,
about a much heavier proton. Use the latter result to evaluate the classical
lifetime of the atom, borrowing the initial value of R from quantum mechanics:
R(0) = rB ≈ 0.53 × 10−10 m.
8-40
Problem 8.2. A non-relativistic particle of mass m, with electric charge q, is placed into
a uniform magnetic field B. Derive the law of decrease of the particle’s kinetic energy
due to its electromagnetic radiation at the cyclotron frequency ωc = qB/m. Evaluate the
rate of such radiation cooling for electrons in a magnetic field of 1 T, and estimate the
electron energy interval in which this result is qualitatively correct.
Hint: The cyclotron motion will be discussed in detail (for arbitrary particle
velocities v ∼ c) in section 9.6 below, but I hope that the reader knows that in the
non-relativistic case (v ≪ c) the above formula for ωc may be readily obtained by
combining the Newton’s second law mv⊥2/R = qv⊥B for the circular motion of the
particle under the effect of the magnetic component of the Lorentz force (5.10), and
the geometric relation v⊥ = Rωc. (Here v⊥ is particle’s velocity within the plane
normal to the vector B.)
Problem 8.3. Solve the dipole antenna radiation problem discussed in section 8.2 (see
figure 8.3) for the optimal length l = λ/2, assuming49 that the current distribution in
each of its arms is sinusoidal:
I (z , t ) = I0 cos (πz / l ) cos ωt .
Problem 8.4. Use the Lorentz oscillator model of a bound charge, given by
Eq. (7.30), to explore the transition between the two scattering limits discussed in
section 8.3, in particular the resonant scattering taking place at ω ≈ ω0. In this
context, discuss the contribution of scattering into the oscillator’s damping.
Problem 8.5.* A sphere of radius R, made of a material with a uniform permanent

electric polarization P0 and a constant mass density ρ, is free to rotate about its
center. Calculate the average total cross-section of scattering, by the sphere, of a
linearly polarized electromagnetic wave of frequency ω ≪ R/c, propagating in free
space, in the limit of a small wave amplitude, assuming that the initial orientation of
the polarization vector P0 is random.
Problem 8.6. Use Eq. (8.56) to analyze the interference/diffraction pattern produced
by a plane wave’s scattering on a set of N similar, equidistant points on a straight
line normal to the direction of the incident wave’s propagation—see the figure
below. Discuss the trend(s) of the pattern in the limit N → ∞.
Problem 8.7. Use the Born approximation to calculate the differential cross-section
of the plane wave scattering by a non-magnetic, uniform dielectric sphere of an
49
As was emphasized in section 8.2, this is a reasonable guess rather than a controllable approximation. The
exact (rather involved!) theory shows that this assumption gives errors ∼5%.
8-41
arbitrary radius R. In the limits kR ≪ 1 and 1 ≪ kR (where k is the wave number),

analyze the angular dependence of the differential cross-section, and calculate the
full cross-section of scattering.
Problem 8.8. A sphere of radius R is made of a uniform, non-magnetic, linear

dielectric material, with an arbitrary dielectric constant. Derive an exact expression
for its full cross-section of scattering of a low-frequency monochromatic wave, with
k ≪ 1/R, and compare the result with the solution of the previous problem.
Problem 8.9. Use the Born approximation to calculate the differential cross-section
of the plane wave scattering on a right, circular cylinder of length l and radius R, for
an arbitrary angle of incidence.
Problem 8.10. Formulate the quantitative condition of the Born approximation’s

validity for a uniform linear-dielectric scatterer, with all linear dimensions of the
order of the same scale a.
Problem 8.11. If a scatterer absorbs some part of the incident wave’s power, it may
be characterized by an absorption cross-section σa defined similarly to Eq. (8.39) for
the scattering cross-section:
Pa
σa ≡ ,
Eω 2 /2Z0
where the nominator is the time-averaged power absorbed in the scatterer. Calculate
σa for a very small sphere of radius R ≪ k−1, δs, made of a non-magnetic material
with Ohmic conductivity σ, and with high-frequency permittivity εopt = ε0. Can σa of
such a sphere be larger than its geometric cross-section πR2?
Problem 8.12. Use the Huygens principle to calculate the wave’s intensity on the
symmetry plane of the slit diffraction experiment (i.e. at x = 0 in figure 8.12), for an
arbitrary ratio z/ka2.
Problem 8.13. A plane wave with wavelength λ is normally incident on an opaque,

plane screen, with a round orifice of radius R ≫ λ. Use the Huygens principle to
calculate the passing wave’s intensity distribution along the system’s symmetry axis,
at distances z ≫ R from the screen (see the figure below) and analyze the result.
8-42
Problem 8.14. A planar monochromatic wave is now normally incident on an

opaque circular disk of radius R ≫ λ. Use the Huygens principle to calculate the
wave’s intensity at distance z ≫ R behind the disk’s center (see the figure below).
Discuss the result.
Problem 8.15. Use the Huygens principle to analyze the Fraunhofer diffraction of a
plane wave normally incident on a square-shaped hole, of size a × a, in an opaque
screen. Sketch the diffraction pattern you would observe at a sufficiently large
distance, and quantify the expression ‘sufficiently large’ for this case.
Problem 8.16. Use the Huygens principle to analyze the propagation of a mono-
chromatic Gaussian beam described by Eq. (7.181), with the initial characteristic
width a0 ≫ λ, in a uniform, isotropic medium. Use the result for a semi-quantitative
derivation of the so-called Abbe limit50 for the spatial resolution of an optical
system,
wmin = λ /2 sin θ ,
where θ is the half-angle of the wave cone propagating from the object, and captured
by the system.
Problem 8.17. Within the Fraunhofer approximation, analyze the pattern produced
by a 1D diffraction grating with the periodic transparency profile shown in the figure
below, for the normal incidence of a plane, monochromatic wave.
Problem 8.18. N equal point charges are attached, at equal intervals, to a circle
rotating with a constant angular velocity about its center—see the figure below. For
what values of N does the system emit:
(i) the electric dipole radiation?
(ii) the magnetic dipole radiation?
(iii) the electric quadrupole radiation?
50
Reportedly, due to not only E Abbe (1873), but also to H von Helmholtz (1874).
8-43
Problem 8.19. What general statements can you made about:

(i) the electric dipole radiation, and
(ii) the magnetic dipole radiation
due to a collision of an arbitrary number of similar classical, non-relativistic

particles?
Problem 8.20. Calculate the angular distribution and the total power radiated by a
small round loop antenna with radius R, fed by ac current I(t) with frequency ω and
amplitude I0, into the free space.
Problem 8.21. The orientation of a magnetic dipole m, of a fixed magnitude, is

rotating about a certain axis with angular velocity ω, with angle α between them
staying constant. Calculate the angular distribution and the average power of its
radiation (into the free space).
Problem 8.22. Solve problem 8.8 (also in the low-frequency limit kR ≪ 1), for the
case when the sphere’s material has a frequency-independent Ohmic conductivity σ,
and εopt = ε0, in two limits:
(i) of a very large skin depth (δs ≫ R), and
(ii) of a very small skin depth (δs ≪ R).
Problem 8.23. Complete the solution of the problem started in section 8.9, by
calculating the full power of radiation of the system of two charges oscillating in
antiphase along the same straight line—see figure 8.16. Also, calculate the average
radiation power for the case of harmonic oscillations, d(t) = acosωt, compare it with
the case of a single charge performing similar oscillations, and interpret the difference.
References
[1] Cullity B 1978 Elements of x-ray Diffraction 2nd edn (New York: Addison-Wesley)
[2] Born M et al 1999 Principles of Optics 7th edn (Cambridge University Press)
[3] Fowles G 1989 Introduction to Modern Optics 2nd edn (New York: Dover)
[4] Brezinski M 2006 Optical Coherence Tomography (New York: Academic)
[5] Popescu G 2011 Quantitative Phase Imaging of Cells and Tissues (New York: McGraw-Hill)
8-44
IOP Publishing
Lecture notes
Chapter 9
Special relativity
This chapter starts with a review of the basics of special relativity, including the very
convenient four-vector formalism. This background is then used for the analysis of the
relation between the electromagnetic field’s values measured in different reference
frames moving relative to each other. The results of this discussion enable the analysis
of the relativistic particle dynamics in the electric and magnetic fields, the analytical
mechanics of the particles, and of the electromagnetic field as such.
9.1 Einstein postulates and the Lorentz transform

As was emphasized at the derivation of expressions for the dipole and quadrupole
radiations in the previous chapter, they are only valid for systems of non-relativistic
particles. Thus, these results cannot be used for the description of such important
phenomena as the Cherenkov radiation or synchrotron radiation, in which rela-
tivistic effects are essential. Moreover, an analysis of the motion of charged
relativistic particles in electric and magnetic fields is also a natural part of electro-
dynamics. This is why I will follow the tradition of using this course for a (by
necessity, brief) introduction to the theory of special relativity. This theory is based
on the fundamental idea that measurements of physical variables (including the
spatial and even temporal intervals between two events) may give different results in
different reference frames, in particular in two inertial frames moving relative to
each other translationally (i.e. without rotation), with a certain constant velocity v
(figure 9.1).
In non-relativistic (Newtonian) mechanics the problem of transfer between such
reference frames has a simple solution, at least in the limit v ≪ c, because the basic
equation of particle dynamics (Newton’s second law)1
1
Let me hope that the reader does not need a reminder that in order for Eq. (9.1) to be valid, the reference
frames 0 and 0′ have to be inertial—see, e.g. Part CM section 1.2.

Figure 9.1. The translational, uniform mutual motion of two reference frames.
mk rk̈ = −∇k∑U (rk − rk ′), (9.1)

k′
where U, the potential energy of inter-particle interactions, is invariant with respect

to the so-called Galilean transformation (or just ‘transform’ for short)2. Choosing the
coordinate axes of both frames so that the axes x and x′ are parallel to the vector v
(figure 9.1), the transform3 may be represented as
x = x′ + vt′ , y = y′ , z = z′ , t = t′ , (9.2a )
and plugging Eq. (9.2a) into Eq. (9.1) we obtain an absolutely similar looking
equation of motion in the ‘moving’ reference frame 0′. Since the reciprocal
transform,
x′ = x − vt , y = y′ , z′ = z , t′ = t , (9.2b)
is similar to the direct one, with the replacement of (+v) by (−v), we may say that the
Galilean invariance means that there is no ‘master’ (absolute) spatial reference frame
in classical mechanics, although the spatial and temporal intervals between different
instant events are absolute, i.e. reference-frame invariant: Δx = Δx′,…, Δt = Δt′.
However, it is straightforward to use Eq. (9.2) to check that the form of the wave
equation
⎛ ∂2 ∂2 ∂2 1 ∂2 ⎞
⎜ 2 + 2 + 2 − 2 2 ⎟f = 0, (9.3)
⎝ ∂x ∂y ∂z c ∂t ⎠
describing in particular the electromagnetic wave propagation in free space4, is not

Galilean-invariant5. For the ‘usual’ (say, elastic) waves, which obey a similar
2
It was first formulated by G Galilei as early as in 1638—4 years before Newton was born!
3
Note the very unfortunate term ‘boost’ sometimes used for the description of the transfer between reference
frames. (It is particularly unnatural in special relativity, which does not describe any accelerations.) In these
notes, this term is avoided.
4
The discussions in this chapter and most of the next chapter will be restricted to the free space (and hence
dispersion-free) case; some media effects on the radiation by relativistic particles will be discussed in section
10.4.
5
It is interesting that the usual Schrödinger equation, whose fundamental solution for a free particle is a similar
monochromatic wave (albeit with a different dispersion law), is Galilean-invariant, with a certain addition to
the wavefunction’s phase—see, e.g. Part QM chapter 1. This is natural, because this equation is non-
relativistic.
9-2
equation albeit with a different speed6, this lack of Galilean invariance is natural and
is compatible with the invariance of Eq. (9.1), from which the wave equation
originates. This is because elastic waves are essentially the oscillations of interacting
particles of a certain medium (e.g. an elastic solid), which makes the reference frame,
connected to this medium, special. So, if the electromagnetic waves were oscillations
of a certain special medium (which was first called the ‘luminiferous aether’7 and
later the aether, or just ‘ether’), similar arguments might be applicable to reconcile
Eqs. (9.2) and (9.3).
The detection of such a medium was the goal of the Michelson–Morley measure-
ments (carried out between 1881 and 1887 with better and better precision), that are
sometimes called ‘the most famous failed experiment in physics’. Figure 9.2 shows a
crude scheme of their experiments.
Figure 9.2. The Michelson–Morley experiment.
A nearly monochromatic wave from a light source is split in two parts (nominally,
of equal intensity), using a semi-transparent mirror tilted by 45° to the incident wave
direction. These two partial waves are reflected back by two full-reflection mirrors
and arrive at the same semi-transparent mirror again. Here a half of each wave is
returned to the light source area (where they vanish without affecting the source),
but the other half travels toward the detector, forming, with its counterpart, an
interference pattern similar to that in the Young experiment. Thus each of the
interfering waves has traveled twice (back and forth) along each of the two mutually
perpendicular ‘arms’ of the interferometer. Assuming that the aether, in which light
propagates with speed c, moves with speed v < c along one of the arms, of length ll, it
is straightforward (and hence left for the reader’s exercise) to obtain the following
expression for the difference between the light roundtrip times:
2⎡ lt ll ⎤ l ⎛ v ⎞2
Δt = ⎢ − ⎥≈ ⎜ ⎟ , (9.4)
c ⎣ (1 − v 2 / c 2 )1/2 1 − v 2 /c 2 ⎦ c ⎝ c ⎠
where lt is the length of the second, ‘transverse’ arm of the interferometer

(perpendicular to v), and the last, approximate expression is valid at lt ≈ ll and
v ≪ c.
6
See, e.g. Part CM sections 6.5 and 7.7.
7
In the ancient Greek mythology, aether is the clean air breathed by the gods residing on Mount Olympus.
9-3
Since the Earth moves around the Sun with a speed vE ≈ 30 km s−1 ≈ 10−4 c, the
arm positions relative to this motion alternate, due to the Earth’s rotation about its
axis, every 6 h—see the right-hand panel of figure 9.2. Hence if we assume that the
aether rests in the Sun’s reference frame, Δt (and the corresponding shift of the
interference fringes) has to change its sign with this half-period as well. The same
alternation may be achieved, at a smaller time scale, by a deliberate rotation of the
instrument by π/2. In the most precise version of the Michelson–Morley experiment
(1887), this shift was expected to be close to 0.4 of the fringe pattern period. The
results of the search for such a shift were negative, with the error bar about 0.01 of
the fringe period8.
The most prominent immediate explanation of this zero result9 was suggested in
1889 by G FitzGerald and (independently and more qualitatively) by H Lorentz in
1892: as is evident from Eq. (9.4), if the longitudinal arm of the interferometer itself
experiences so-called length contraction,
⎛ v2 ⎞
1/2
ll (v ) = ll (0)⎜1 − 2 ⎟ , (9.5)
⎝ c ⎠
while the transverse arm’s length is not affected by its motion through the aether,
this kills the shift Δt. This extremely radical idea received strong support from the
proof, in 1887–1905, that the Maxwell equations, and hence the wave Eq. (9.3), are
form-invariant under the so-called Lorentz transform10. For the choice of coor-
dinates shown in figure 9.1, the transform reads
x′ + vt′ t ′ + ( v / c 2 ) x′
x= , y = y′ , z = z′ , t= . (9.6a )
(1 − v 2 / c 2 )1/2 (1 − v 2 / c 2 )1/2
It is elementary to solve these equations for the primed coordinates to obtain the
reciprocal transform
x − vt t − (v / c 2 )x
x′ = , y′ = y , z′ = z , t′ = . (9.6b)
(1 − v 2 / c 2 )1/2 (1 − v 2 / c 2 )1/2
(I will soon represent Eqs. (9.6) in a more elegant form.)
The Lorentz transform relations (9.6) are evidently reduced to the Galilean
transform formulas (9.2) at v2 ≪ c2. As will be proved in the next section, Eqs. (9.6)
also yield the Lorentz length contraction (9.5). However, all attempts to give a
8
Through the twentieth century, Michelson–Morley-type experiments were repeated using more and more
refined experimental techniques, always with the zero result for the apparent aether motion speed. For
example, recent experiments, using cryogenically cooled optical resonators, have reduced the upper limit for
such a speed to just 3 × 10−15 c—see [1].
9
The zero result of a slightly later experiment, namely precise measurements of the torque which should be
exerted by the moving aether on a charged capacitor, carried out in 1903 by F Trouton and H Noble
(following G FitzGerald’s suggestion) seconded Michelson and Morley’s conclusions.
10
The theoretical work toward this goal (which I do not have time to review in detail) included important
contributions by W Voigt (in 1887), H Lorentz (1892–1904), J Larmor (1897 and 1900), and H Poincaré (1900
and 1905).
9-4
reasonable interpretation of these equations while retaining the notion of the aether
have failed, in particular because of the restrictions imposed by the results of earlier
experiments carried out in 1851 and 1853 by H Fizeau—which were repeated with
higher accuracy by the same Michelson and Morley in 1886. These experiments have
shown that if one sticks to the aether concept, this hypothetical medium should be
partially ‘dragged’ by any moving dielectric material with a speed proportional to
(κ − 1). Such a local drag is irreconcilable with the assumed continuity of the aether.
In his famous 1905 paper Albert Einstein suggested a bold resolution of this
contradiction, essentially removing the concept of the aether altogether. Moreover,
he argued that the Lorentz transform is a general property of time and space, rather
than of the electromagnetic field alone. He has started with two postulates, the first
one essentially repeating the principle of relativity, formulated earlier (1904) by H
Poincaré in the following form:
…the laws of physical phenomena should be the same, whether for an observer
fixed, or for an observer carried along in a uniform movement of translation;
so that we have not and could not have any means of discerning whether or not
we are carried along in such a motion.11
The second Einstein’s postulate was that the speed of light c, in free space, should
be constant in all reference frames. (This is essentially a denial of the aether’s
existence.)
Then, Einstein showed how naturally the Lorenz transform relations (9.6) follow
from his postulates, with a few (very natural) additional assumptions. Let a point
source emit a short flash of light, at the moment t = t′ = 0 when the origins of the
reference frames shown in figure 9.1 coincide. Then, according to the second of
Einstein’s postulates, in each of the frames the spherical wave propagates with the
same speed c, i.e. the coordinates of points of its front, measured in the two frames,
have to obey equations
(ct )2 − (x 2 + y 2 + z 2 ) = 0,
(9.7)
(ct′)2 − (x′2 + y′2 + z′2 ) = 0.
What might be the general relation between the combinations on the left-hand side
of these equations—not for this particular wave’s front, but in general? A very
natural (essentially, the only justifiable) choice is
[(ct )2 − (x 2 + y 2 + z 2 )] = f (v 2 )[(ct′)2 − (x′2 + y′2 + z′2 )]. (9.8)
Now, according to the first postulate, the same relation should be valid if we
swap the reference frames (x ↔ x′, etc) and replace v with (−v). This is only possible
11
Note that although the relativity principle excludes the notion of the special (‘absolute’) spatial reference
frame, its verbal formulation still leaves the possibility of the Galilean ‘absolute time’ open. The quantitative
relativity theory kills this option—see Eq. (9.6) and their discussion below.
9-5
if f 2 = 1, so that excluding the option f = −1 (which is incompatible with the

Galilean transform in the limit v/c → 0), we obtain
(ct )2 − (x 2 + y 2 + z 2 ) = (ct′)2 − (x′2 + y′2 + z′2 ) . (9.9)
For the line with y = y′ = 0, z = z′ = 0, Eq. (9.9) is reduced to

(ct )2 − x 2 = (ct′)2 − x′2 . (9.10)
It is very illuminating to interpret this relation as the one resulting from a mutual
rotation of the reference frames (that now have to include clocks to measure time) on
the plane of the coordinate x and the so-called Euclidian time τ ≡ ict—see figure 9.3.
Figure 9.3. The Lorentz transform as a mutual rotation of two reference frames on the [x, τ] plane.
Indeed, rewriting Eq. (9.10) as

τ 2 + x 2 = τ′2 + x′2 , (9.11)
we may consider it as the invariance of the squared radius at the rotation that is
shown in figure 9.3 and described by the evident geometric relations
x = x′ cos ψ − τ′ sin ψ ,
(9.12a )
τ = x′ sin ψ + τ′ cos ψ ,
with the reciprocal relations
x′ = x cos ψ + τ sin ψ ,
(9.12b)
τ′ = −x sin ψ + τ cos ψ .
So far, the angle ψ has been arbitrary. In the spirit of Eq. (9.8), a natural choice is
ψ = ψ(v), with the requirement ψ(0) = 0. In order to find this function, let us write the
definition of the velocity v of the reference frame 0′, as measured in the frame 0
(which was implied above): for x′ = 0, x = vt. In the variables x and τ, this means
x x v
≡ = . (9.13)
τ x ′=0 ict x ′=0 ic
On the other hand, for the same point x′ = 0, Eqs. (9.12a) yield
x
= −tan ψ . (9.14)
τ x ′=0
These two expressions are compatible only if
9-6
iv
tan ψ = , (9.15)
c
so that
tan ψ iv / c
sin ψ ≡ 1/2
= ≡ iβγ ,
2
(1 + tan ψ ) (1 − v 2 / c 2 )1/2
(9.16)
1 1
cosψ ≡ 1/2
= ≡ γ,
2
(1 + tan ψ ) (1 − v 2 / c 2 )1/2
where β and γ are two very convenient and commonly used dimensionless
parameters defined as
v 1 1
β≡ , γ≡ 2 2 1/2
= 1/2
. (9.17)
c (1 − v / c ) (1 − β 2 )
(The vector β is called the normalized velocity, while the scalar γ is the Lorentz
factor12.)
Using the relations for ψ, Eqs. (9.12) become
x = γ (x′ − iβτ′), τ = γ (iβx′ + τ′), (9.18a )
x′ = γ (x + iβτ ), τ′ = γ ( −iβx + τ ). (9.18b)

Now returning to the real variables [x, ct], we obtain the Lorentz transform relations
(9.6), in a more compact form:
x = γ (x′ + β ct′), y = y′ , z = z′ , ct = γ (ct′ + β x′), (9.19a )
x′ = γ (x − β ct ), y′ = y , z′ = z , ct′ = γ (ct − β x ). (9.19b)

An immediate corollary of Eqs. (9.19) is that for γ to stay real, we need v2 ⩽ c2, i.e.
that the speed of any physical body (to which we could connect a meaningful
reference frame) cannot exceed the speed of light, as measured in any other
meaningful reference frame13.
9.2 Relativistic kinematic effects

In order to discuss other corollaries of Eqs. (9.19), we need to spend a few minutes
discussing what these relations actually mean. Evidently, they are trying to tell us
that the spatial and temporal intervals are not absolute (as they are in Newtonian
space), but do depend on the reference frame they are measured in. So, we have to
12
Note the following identities: γ2 ≡ 1/(1 − β2) and (γ2 − 1) ≡ β2/(1 − β2) ≡ γ2β2, which are frequently handy for
relativity-related algebra. One more function of β, the rapidity φ ≡ tanh−1β (so that ψ = iφ), is also useful for
some calculations.
13
All attempts to rationally conjecture particles moving with v > c, called tachyons, have failed—so far, at
least. Possibly the strongest objection against their existence is the fact that tachyons could be used to
communicate backwards in time, thus violating the causality principle—see, e.g. [2].
9-7
understand very clearly what exactly may be measured—and thus may be discussed
in a physics theory. Recognizing this necessity, Einstein has introduced the notion of
numerous imaginary observers that may be distributed all over each reference frame.
Each observer has a clock and may use it to measure the instants of local events. He
also conjectured, very reasonably, that:
(i) all observers within the same reference frame may agree on a common length
measure (‘a scale’), i.e. on their relative positions in that frame, and synchronize
their clocks14, and
(ii) the observers belonging to different reference frames may agree on the
nomenclature of world events (e.g. short flashes of light) to which their respective
measurements refer.
Actually, these additional postulates have been already implied in our ‘derivation’
of the Lorentz transform in section 9.1. For example, by {x, y, z, and t} we mean the
results of space and time measurements of a certain world event, about which all
observers belonging to the frame 0 agree. Similarly, all observers of the frame 0′
have to agree about the results {x′, y′, z′, t′}. Finally, when the origin of frame 0′
passes some sequential points xk of frame 0, observers in that frame may measure its
passage times tk without fundamental error, and know that all these times belong to
x′ = 0.
Now we can analyze the major corollaries of the Lorentz transform, which are
rather striking from the point of view of our everyday (rather non-relativistic)
experience.
(i) Length contraction. Let us consider a rigid rod, stretched along axis x, with its
length l ≡ x2 − x1, where x1,2 are the coordinates of the rod’s ends as measured in its
rest frame 0, at any instant t (figure 9.4). What would be the rod’s length l′ measured
by the Einstein observers in the moving frame 0′?
Figure 9.4. The relativistic length contraction.
At a time instant t′ agreed upon in advance, the observers who find themselves
exactly at the rod’s ends may register that fact and then subtract their coordinates
14
A posteriori, the Lorenz transform may be used to show that consensus-creating procedures (such as clock
synchronization) are indeed possible. The basic idea of the proof is that at v ≪ c, the relativistic corrections to
space and time intervals are of the order of (v/c)2, they have negligible effects on clocks being brought together
into the same point for synchronization very slowly. The reader interested in detailed discussion of this and
other fine points of special relativity is referred to, e.g. either [3] or [4].
9-8
x′1,2 to calculate the apparent rod length l′ ≡ x′2 − x′1 in the moving frame.
According to Eq. (9.19a), l may be expressed via l′ as
l ≡ x2 − x1 = γ (x2′ + βct′) − γ (x1′ + βct′) = γ (x2′ − x1′) ≡ γ l ′ . (9.20a )
Hence, the rod’s length, as measured in the moving reference frame is
l ⎛ v2 ⎞
1/2
l′ = = l ⎜1 − 2 ⎟ ⩽ l , (9.20b)
γ ⎝ c ⎠
in accordance with the FitzGerald–Lorentz hypothesis (9.5). This is the relativistic

length contraction effect: an object is always the longest (has the so-called proper
length l) if measured in its rest frame. Note that according to Eq. (9.19), the length
contraction takes place only in the direction of the relative motion of two reference
frames. As has been noted in section 9.1, this result immediately explains the zero
result of the Michelson–Morley-type experiments, so that they provide very
convincing evidence (if not irrefutable proof) for Eq. (9.20).
(ii) Time dilation. Now let us use Eq. (9.19a) to find the time interval Δt as measured in
the frame 0 between two world events—say, two ticks of a clock moving with the frame
0′ (figure 9.5), i.e. having constant values of x′, y′, and z′.
Figure 9.5. The relativistic time dilation.
Let the time interval between these two events, measured in the clock’s rest frame
0′, be Δt′ ≡ t′2 − t′1. At these two moments, the clock would fly by two certain
Einstein observers at rest in the frame 0, so that they can record the corresponding
moments t1,2 shown by their clocks, and then calculate Δt as their difference.
According to the second of Eqs. (9.19a),
γ
Δt ≡ t2 − t1 = [(ct2′ + βx′) − (ct1′ + βx′)] ≡ γ Δt′ , (9.21a )
c
so that, finally,
Δt′
Δt = γ Δt′ ≡ ⩾ Δt′ . (9.21b)
(1 − v 2 / c 2 )1/2
This is the famous relativistic time dilation (or ‘dilatation’) effect: a time interval is
longer if measured in a frame (in our case, frame 0) moving relatively to the clock,
while that in the rest frame is the shortest—the so-called proper time interval.
9-9
This counter-intuitive effect is the everyday reality in experiments with high-

energy elementary particles. For example, in a typical (and by no means record-
breaking) experiment carried out in Fermilab, a beam of charged 200 GeV pions
with γ ≈ 1400 travelled a distance l = 300 m with a measured loss of only 3% of the
initial beam intensity due to pion decay (mostly, into muon–neutrino pairs) with the
proper lifetime t0 ≈ 2.56 × 10−8 s. Without time dilation, only an exp{−l/ct0} ~ 10−17
fraction of the initial pions would survive, while the relativity-corrected number
exp{−l/ct} = exp{−l/cγt0} ≈ 0.97 was in a full accordance with experimental
measurements.
As another example, global positioning systems (say, GPS) are designed to take
into account the time dilation due to the velocity of their satellites (and also some
gravity-induced, i.e. general-relativity corrections, which I do not have time to
discuss) and would give large errors without such corrections. So, there is no doubt
that time dilation (9.21) is a reality, although the precision of all its experimental
tests I am aware of has been limited to a few percent, because of the almost
unavoidable involvement of less controllable gravity effects15.
Before the first reliable observation of time dilation (by B Rossi and D Hall in
1940), there had been serious doubts on the reality of this effect, the most famous
being the twin paradox first posed (together with an immediate suggestion of its
resolution) by P Langevin in 1911. Let us send one of two twins on a long star
journey with a speed v approaching c. Upon his return to Earth, which of the twins
would be older? The naïve approach is to say that due to the relativity principle,
neither one would be older (and hence there is no time dilation), because each twin
could claim that his counterpart, rather than himself, was moving, with the same
speed, just in the opposite direction. The resolution of the paradox in the general
theory of relativity (which can handle gravity and acceleration effects) is that one of
the twins had to be accelerated to be brought back, and hence the reference frames
have to be dissimilar: only one of them may stay inertial all the time. Because of
that, the twin who had been accelerated (‘actually traveling’) would be younger than
his or her sibling, when they finally come together.
(iii) Velocity transformation. Now let us calculate velocity u of a particle, as
observed in the reference frame 0, provided that its velocity, as measured in the
frame 0′, is u′ (figure 9.6).
Figure 9.6. The relativistic velocity addition.
15
See, e.g. [5].
9-10
Keeping the usual definition of velocity, but with due attention to the relativity of
not only spatial but also temporal intervals, we may write
dr d r′
u≡ , u′ ≡ . (9.22)
dt dt′
Plugging in the differentials of the Lorentz transform relations (9.6a) into these
definitions, we obtain
dx dx′ + vdt′ ux′ + v

ux ≡ = 2
= ,
dt dt′ + vdx′ / c 1 + ux′v / c 2
(9.23)
dy 1 dy′ 1 u y′
uy ≡ = = ,
dt γ dt′ + vdx′ / c 2 γ 1 + ux′v / c 2
and a similar formula for uz. In the classical limit v/c → 0, these relations are reduced
to
ux = ux′ + v , u y = u y′ , uz = uz′, (9.24a )
and may be merged into the familiar Galilean vector form
u = u′ + v, for v ≪ c. (9.24b)
In order to see how unusual the full relativistic rules (9.23) are, let us first consider
a purely longitudinal motion, uy = uz = 0; then16
u′ + v
u= , (9.25)
1 + u′v / c 2
where u ≡ ux and u′ ≡ u′x. Figure 9.7 shows u as the function of u′, given by this
formula, for several values of the reference frames’ relative velocity v. The first
sanity check is that if v = 0, i.e. the reference frames are at rest relative to each
other, then u = u′, as it should be—see the diagonal straight line. Next, if the
magnitudes of u′ and v are both below c, so is the magnitude of u. (Also good,
because otherwise ordinary particles in one frame would be tachyons in the other
one and the theory would be in big trouble.) Now strange things begin: even as u′
and v are both approaching c, then u is also close to c, but does not exceed it. As an
example, if we fired ahead a bullet with the relative speed 0.9c from a spaceship
moving from the Earth also at 0.9c, Eq. (9.25) predicts the speed of the bullet
relative to Earth to be just [(0.9 + 0.9)/(1 + 0.9 × 0.9)]c ≈ 0.994c < c, rather than
16
With an account of the well-known trigonometric identity tan(a + b) = (tan a + tan b)/(1 − tan a tan b) and
Eq. (9.15), Eq. (9.25) shows that that the rapidities ψ add up exactly as the longitudinal velocities in non-
relativistic motion, making that notion very convenient for the analysis of transfer between several frames.
9-11
Figure 9.7. The addition of longitudinal velocities.
(0.9 + 0.9)c = 1.8c > c as in the Galilean kinematics. Actually, we could expect this
strangeness, because it is necessary to fulfil Einstein’s second postulate: the
independence of the speed of light in any reference frame. Indeed, for u′ = ±c,
Eq. (9.25) yields u = ±c, regardless of v.
In the opposite case of purely transverse motion, when a particle moves across the
relative motion of the frames (for example, at our choice of coordinates, u′x = u′z =
0), Eqs. (9.23) yield a much less spectacular result
u y′
uy = ⩽ u y′ . (9.26)
γ
This effect comes purely from the time dilation, because the transverse coordinates
are Lorentz-invariant.
In the case when both ux′ and uy′ are substantial (but uz′ is still zero), we may
divide Eqs. (9.23) by each other to relate the angles θ of particle propagation, as
observed in the two reference frames:
uy u y′ sin θ′
tan θ ≡ = = . (9.27)
ux γ (ux′ + v ) γ (cos θ′ + v / u′)
This expression describes, in particular, the so-called stellar aberration effect, the
dependence of the observed direction θ toward a star on the speed v of the telescope’s
motion relative to the star—see figure 9.8. (The effect is readily observable
experimentally as the annual aberration due to the periodic change of speed v by
2vE ≈ 60 km s−1 because of the Earth’s rotation about the Sun. Since the aberration’s
main part is of the first order in vE/c ∼ 10−4, the effect is very significant and has been
known since the early 1700s.)
9-12
Figure 9.8. The stellar aberration.
For the analysis of this effect, it is sufficient to take, in Eq. (9.27), u′ = c, i.e. v/u′ =
β, and interpret θ′ as the ‘proper’ direction to the star that would be measured at v =
0.17 At β ≪ 1, both Eq. (9.27) and the Galilean result (which the reader is invited to
derive directly from figure 9.8),
sin θ′
tan θ = , (9.28)
cos θ′ + β
may be well approximated by the first-order term
Δθ ≡ θ′ − θ ≈ β sin θ . (9.29)
Unfortunately, it is not easy to use the difference between Eqs. (9.27) and (9.28), of
second order in β, for the special relativity confirmation, because other components
of the Earth’s motion, such as its rotation, nutation, and torque-induced preces-
sion18, provide masking first-order contributions to the aberration.
Finally, at a completely arbitrary direction of vector u′, Eqs. (9.23) may be
readily used to calculate the velocity magnitude. The most popular form of the
resulting expression is for the square of the relative velocity (or rather the relative
reduced velocity β) of two particles,
(β1 − β2)2 − ∣β1 ⋅ β2∣
β2 = ⩽ 1. (9.30)
(1 − β1 ⋅ β2)2
where β1,2 ≡ v1,2/c are their normalized velocities as measured in the same reference
frame.
17
Strictly speaking, in order to reconcile the geometries shown in figure 9.1 (for which all our formulas,
including Eq. (9.27), are valid) and figure 9.8 (giving the traditional scheme of the stellar aberration), it is
necessary to invert the signs of u (and hence sinθ′ and cosθ′) and v, but as is evident from Eq. (9.27), all the
minus signs cancel and the formula is valid as it is.
18
See, e.g. Part CM sections 4.4 and 4.5.
9-13
(iv) The Doppler effect. Now let us consider a plane, monochromatic wave moving
along axis x:
f = Re [fω exp {i (kx − ωt )}] ≡ ∣fω ∣ cos (kx − ωt + arg fω ). (9.31)
Its total phase Ψ ≡ kx − ωt + arg fω (in contrast to the real amplitude ∣fω∣!) cannot
depend on the observer’s reference frame, because all fields of a traveling wave
vanish simultaneously at Ψ = 2πn (for all integer n), and such ‘world events’ should
take place in all reference frames. The only way to keep Ψ = Ψ′ at all times is to
have19
kx − ωt = k′x′ − ω′t′ . (9.32)
First, let us use this general relation to consider the Doppler effect in the usual
non-relativistic waves, e.g. oscillations of particles of a certain medium. Using the
Galilean transform (9.2), we may rewrite Eq. (9.32) as
k (x′ + vt ) − ωt = k′x′ − ω′t . (9.33)
Since this transform leaves all space intervals (including wavelength λ = 2π/k) intact,
we can take k = k′ so that Eq. (9.33) yields
ω′ = ω − kv. (9.34)
For a dispersion-free medium, the wave number k is the ratio of its frequency ω,
as measured in the reference frame bound to the medium, and the wave velocity vw.
In particular, if the wave source rests in the medium, we may bind the reference
frame 0 to the medium as well, and the frame 0′ to wave’s receiver (so that v = vr), so
that
ω
k= , (9.35)
vw
and for the frequency perceived by the receiver, Eq. (9.34) yields
v − vr
ω′ = ω w . (9.36)
vw
On the other hand, if the receiver and the medium are at rest in the reference frame
0′, while the wave source is bound to the frame 0 (so that v = −vs), Eq. (9.35) should
be replaced with
ω′
k = k′ = , (9.37)
vw
and Eq. (9.34) yields a different result:
19
Strictly speaking, Eq. (9.32) is valid to an additive constant, but for simplicity of notation it may always be
made equal to zero by selecting (as has already been done in all relations of section 9.1) the reference frame
origins and/or clock turn-on times so that at t = 0 and x = 0, t′ = 0, and x′ = 0 as well.
9-14
vw
ω′ = ω . (9.38)
vw − vs
Finally, if both the source and detector are moving, it is straightforward to combine
these two results to obtain the general relation
v − vr
ω′ = ω w . (9.39)
vw − vs
At low speeds of both the source and the receiver this result simplifies,
v − vs
ω′ ≈ ω(1 − β ), β ≡ r , (9.40)
vw
but at speeds comparable to vw we have to use the more general Eq. (9.39). Thus, the
usual Doppler effect is generally affected not only by the relative speed (vr − vs) of
the wave’s source and detector, but also of their speeds relative to the medium in
which the waves propagate.
Somewhat counter-intuitively, for electromagnetic waves the calculations are
simpler, because for them the propagation medium (aether) does not exist, the wave
velocity equals ±c in any reference frame, and there are not two separate cases: we
can always take k = ±ω/c and k′ = ±ω′/c. Plugging these relations, together with the
Lorentz transform (9.19a), into the phase-invariance condition (9.32), we obtain
ω ct′ + β x′ ω′
± γ (x′ + β ct′) − ωγ = ± x′ − ω′t′ . (9.41)
c c c
This relation has to hold for any x′ and t′, so we may require that the net coefficients
before these variables vanish. These two requirements yield the same equality:
ω′ = ωγ (1 ∓ β ). (9.42)
This result is already quite simple, but may be transformed further to be even more
illuminating:
1∓β ⎡ (1 ∓ β )(1 ∓ β ) ⎤1/2

ω′ = ω ≡ ω ⎢ ⎥ . (9.43)
(1 − β 2 )
1/2
⎣ (1 + β )(1 − β ) ⎦
At any sign before β, one pair of parentheses cancel, so that

⎛ 1 ∓ β ⎞1/2
ω′ = ω ⎜ ⎟ . (9.44)
⎝1 ± β ⎠
(It may look like the reciprocal expression of ω via ω′ is different, violating
the relativity principle. However, in this case we have to change the sign of β,
because the relative velocity of the system is opposite, so we return to Eq. (9.44)
again.)
9-15
Thus the Doppler effect for electromagnetic waves depends only on the
relative velocity v = βc between the wave source and detector—as it should be,
given the aether’s absence. At velocities much below c, Eq. (9.43) may be
approximated as
1 ∓ β /2
ω′ ≈ ω ≈ ω (1 ∓ β ), (9.45)
1 ± β /2
i.e. in the first approximation in β ≡ v/c it coincides with the corresponding limit
(9.40) of the usual Doppler effect. However, even at v ≪ c there is still a difference,
of the order of (v/c)2, between the Galilean and Lorentzian relations.
If the wave vector k is tilted by angle θ to the vector v (as measured in frame 0),
then we have to repeat the calculations, with k replaced by kx, and components ky
and kz left intact at the Lorentz transform. As a result, Eq. (9.42) is generalized as
ω′ = ωγ (1 − β cos θ ). (9.46)
For the cases cos θ = ±1, Eq. (9.46) reduces to our previous result (9.42). However,
at θ = π/2 (i.e. cos θ = 0), the relation is rather different:
ω
ω′ = γω = 1/2
. (9.47)
(1 − β 2 )
This is the transverse Doppler effect—which is completely absent in non-

relativistic physics. Its first experimental evidence was obtained using electron
beams (as suggested in 1906 by J Stark) by H Ives and G Stilwell in 1938 and
1941. Later, similar experiments were repeated several times, but the first
unambiguous measurements were only performed in 1979 by D Hasselkamp
et al who confirmed Eq. (9.47) with a relative accuracy of about 10%. This
precision may not look too spectacular, but in addition to the special tests
discussed above, the Lorentz transform formulas have been also confirmed, less
directly, by a huge body of other experimental data, especially in high energy
physics, in agreement with calculations incorporating the transform as a part.
This is why, with all respect to the spirit of challenging authority, I should warn
the reader: if you decide to challenge the theory of relativity (which is called
‘theory’ by tradition only), you would also need to explain all of these data20.
Best of luck with that!
20
The same fact, ignored by crackpots, is also valid for other favorite points of their attacks, including the
expansion of the Universe and quantum mechanics in physics, and the theory of evolution in biology.
9-16
9.3 Four-vectors, momentum, mass, and energy

Before proceeding to relativistic dynamics, let us discuss a mathematical formalism
which makes all the calculations more compact—and more beautiful. We have
already seen that the three spatial coordinates {x, y, z} and the product ct are
Lorentz-transformed similarly—see Eqs. (9.19). So it is natural to consider them as
components of a single four-component vector (or, for short, four-vector),
{x0, x1, x2 , x3} ≡ {ct , r}, (9.48)
with components
x0 ≡ ct , x1 ≡ x , x2 ≡ y , x3 ≡ z . (9.49)
According to Eqs. (9.19), its components are Lorentz-transformed as
3
xj = ∑ Ljj′x′ j′ , (9.50)
j ′= 0
where Ljj′ are the elements of the following 4 × 4 Lorentz transform matrix
⎛ γ βγ 0 0 ⎞
⎜ ⎟
⎜ βγ γ 0 0 ⎟ . (9.51)
⎜⎜ 0 0 1 0 ⎟⎟
⎝ 0 0 0 1⎠
Since four-vectors are a new notion for our course, and are used for many more
aims than the just the space–time transform, we need to discuss the mathematical
rules they obey. Indeed, as was mentioned in section 8.9, the usual (three-
component) vector is not just any ordered set (string) of three scalars {Ax, Ay,
Az}; if we want it to represent a reference-frame-independent physical reality, the
vector’s components have to obey certain rules at transfer from one reference frame
to another. In particular, the vector’s norm (its magnitude squared),
A2 = Ax2 + A y2 + Az2 , (9.52)
should be an invariant at the Galilean transform (9.2). However, a naïve extension

of this formula to four-vectors would not work because, according to the calcu-
lations of section 9.1, the Lorentz transform keeps intact combinations of the type
(9.7), with one sign negative, rather than the sum of all components squared. Hence
for the four-vector all the rules of the game have to be reviewed and adjusted—or
rather redefined from the very beginning.
An arbitrary four-vector is a string of four scalars,
{A0 , A1, A2 , A3}, (9.53)
9-17
defined in 4D Minkowski space21, whose components Aj, as measured in systems 0

and 0′, shown in figure 9.1, obey a Lorentz transform similar to Eq. (9.50):
3
Aj = ∑ Ljj′A′ j′ . (9.54)
j ′= 0
As we have already seen on the example of the space–time four-vector (9.48), this
means in particular that
3 3
A02 − ∑A j2 = (A′0 )2 − ∑(A′ j )2 . (9.55)
j=1 j=1
This is the so-called Lorentz invariance condition of the norm of the four-vector.
(The difference between this relation and Eq. (9.52), pertaining to the Euclidian
geometry, is the reason why the Minkowski space is called pseudo-Euclidian.) It is
also straightforward to use Eqs. (9.51) and (9.54) to check that an evident general-
ization of the norm, the scalar product of two arbitrary four-vectors,
3
A0 B0 − ∑Aj Bj , (9.56)
j=1
is also Lorentz-invariant.
Now consider the four-vector corresponding to a small interval between two close
world events:
{dx0, dx1, dx2 , dx3} = {c dt , d r}; (9.57)
its norm,
3
(ds )2 ≡ dx02 − ∑dx j2 = c 2(dt )2 − (dr )2 , (9.58)
j=1
is of course also Lorentz-invariant. Since the speed of any particle (or signal) cannot
be larger than c, for any pair of world events that are in a causal relation with each
other, dr cannot be larger than cdt, i.e. such time-like interval (ds)2 cannot be
negative. The 4D surface separating such intervals from space-like intervals (ds)2 < 0
is called the light cone (figure 9.9).
Now let us assume that these two close world events happen with the same
particle that moves with velocity u. Then in the frame moving with a particle (v = u),
the last term on the right-hand side of Eq. (9.58) equals zero, so that
21
After H Minkowski, who was the first to recast (in 1907) the relations of special relativity in a form in which
the spatial coordinates and time (or rather ct) are treated on an equal footing.
9-18
Figure 9.9. A 2+1-dimensional image of the light cone (which is actually 3+1-dimensional).
ds = c dτ , (9.59)
where dτ is the proper time interval. But according to Eq. (9.21) this means that we
can write
dt
dτ = , (9.60)
γ
where dt is the time interval in an arbitrary (besides being inertial) reference frame,
while
u 1 1
β≡ and γ ≡ 1/2
= (9.61)
c 2
(1 − β ) (1 − u 2 / c 2 )1/2
are the parameters (9.17) corresponding to the particle’s velocity (u) in that frame, so
that ds = c dt/γ.22
Let us use Eq. (9.60) to explore whether a four-vector may be formed using the
spatial components of particle’s velocity
⎧ dx dy dz ⎫
u=⎨ , , ⎬. (9.62)
⎩ dt dt dt ⎭
Here we have a slight problem: as Eqs. (9.23) show, these components do not obey
the Lorentz transform (9.54). However, let us use dτ ≡ dt/γ, the proper time interval
of the particle, to form the following string:
⎧ dx0 dx1 dx2 dx3 ⎫ ⎧ dx dy dz ⎫
⎨ , , , ⎬ ≡ γ ⎨c , , , ⎬ ≡ γ{c , u}. (9.63)
⎩ dτ dτ dτ dτ ⎭ ⎩ dt dt dt ⎭
22
I have opted against using special indices (e.g. βu, γu) to distinguish Eqs. (9.17) and (9.61) here and below, in
hope that the suitable velocity (of either a reference frame or a particle) will be always clear from the context.
9-19
As follows from comparison of the first form of this expression with Eq. (9.48), since the
time–space vector obeys the Lorentz transform, and τ is Lorentz-invariant, the string
(9.63) is a legitimate four-vector. It is called the four-velocity of the particle.
Now we are well equipped to proceed to relativistic dynamics. Let us start with
such basic notions as the momentum p and the energy E —so far, for a free
particle23. Perhaps the most elegant way to ‘derive’ (or rather guess24) the
expressions for p and E as functions of the particle’s velocity u is based on analytical
mechanics. Due to the conservation of v, the trajectory of a free particle in the 4D
Minkowski space is always a straight line. Hence, from the Hamilton principle25, we
may expect its action S , between points 1 and 2, to be a linear function of the space–
time interval (9.59):
2 2 t2
dt
S = α∫ ds ≡ α c∫ dτ ≡ αc∫ , (9.64)
1 1 t1 γ
where α is some constant. On the other hand, in analytical mechanics the action is
defined as
t2
S= ∫t
1
L dt, (9.65)
where L is the particle’s Lagrangian function26. Comparing these two expressions,

we obtain
αc ⎛ u2 ⎞
1/2
L= ≡ αc⎜1 − 2 ⎟ . (9.66)
γ ⎝ c ⎠
In the non-relativistic limit (u ≪ c), this function tends to

⎛ u2 ⎞ αu 2
L ≈ αc⎜1 − 2 ⎟ = αc − . (9.67)
⎝ 2c ⎠ 2c
In order to correspond to Newtonian mechanics27, the last (velocity-dependent) term

should equal mu2/2. From here we find α = −mc, so that, finally,
⎛ u2 ⎞
1/2
L = −mc ⎜1 − 2 ⎟ .
2 (9.68)
⎝ c ⎠
23
I am sorry for using, as in section 6.3, for the particle’s momentum the same traditional notation (p) as had
been used for the electric dipole moment. However, since the latter notion will be virtually unused in the
balance of the notes, this is unlikely to lead to confusion.
24
Indeed, such a derivation uses additional assumptions, however natural (such as the Lorentz-invariance of
S ), so it can hardly be considered as a real proof of the final results, so that they require experimental
confirmation. Fortunately, such confirmations have been numerous—see below.
25
26
27
See, e.g. Part CM Eq. (2.19b).
9-20
Now we can find the Cartesian components pj of the particle’s momentum as the
generalized momenta corresponding to the components rj ( j = 1, 2, 3) of the 3D
radius-vector r:28
∂L ∂L ⎛ u12 + u 22 + u 32 ⎞
1/2
muj
2 ∂
pj = = = −mc ⎜1 − ⎟ =
∂rj̇ ∂uj ∂uj ⎝ c 2
⎠ (1 − u 2 / c 2 )1/2 (9.69)
= mγ uj .
Thus for the 3D vector of momentum, we can write the result in the same form as in
non-relativistic mechanics,
p = mγ u ≡ M u, (9.70)
where the reference-frame-dependent scalar M (called the relativistic mass) is defined
as
m
M ≡ mγ = ⩾ m, (9.71)
(1 − u 2 / c 2 )1/2
m being the non-relativistic mass of the particle. (It is also called the rest mass,
because in the reference frame in which the particle rests, Eq. (9.71) yields M = m.)
Next, let us return to the analytical mechanics to calculate the particle’s energy E
(which for a free particle coincides with the Hamiltonian function H )29:
3
E =H = ∑pj uj − L = p ⋅ u − L
j=1
(9.72)
mu 2 ⎛ u2 ⎞ mc 2
= + mc 2
⎜1 − ⎟ ≡ .
(1 − u 2 / c 2 )1/2 ⎝ c 2 ⎠ (1 − u 2 / c 2 )1/2
Thus, we have arrived at the most famous of Einstein’s formulas (and probably the
most famous formula of physics as a whole),
E = mγ c 2 = Mc 2 , (9.73)
which expresses the relation between the free particle’s mass and its energy30. In the
non-relativistic limit, it reduces to
mc 2 ⎛ u2 ⎞ mu 2
E= ≈ mc 2⎜1 + 2 ⎟ = mc 2 + , (9.74)
2 2 1/2
(1 − u / c ) ⎝ 2c ⎠ 2
the first term mc2 being called the rest energy of a particle.
28
See, e.g. Part CM section 2.3, in particular Eq. (2.31).
29
See, e.g. Part CM Eq. (2.32).
30
Let me hope that the reader understands that all the layman talk about the ‘mass to energy conversion’ is
only valid in a very limited sense of the word. While the Einstein relation (9.73) does allow the conversion of
‘massive’ particles (with m ≠ 0) into particles with m = 0, such as photons, each of the latter particles also has a
non-zero relativistic mass M, and simultaneously the energy E related to M by Eq. (9.73).
9-21
Now let us consider the following string of four scalars:
{ E
,p,p ,p
c 1 2 3 } { }
=
E
c
,p . (9.75)
Using Eqs. (9.70) and (9.73) to represent this expression as
{ }E
c
,p = mγ{c , u}, (9.76)
and comparing the result with Eq. (9.63), we immediately see that, since m is a
Lorentz-invariant constant, this string is a legitimate four-vector of energy–momen-
tum. As a result, its norm,
⎛ E ⎞2
⎜ ⎟ − p2 , (9.77a )
⎝c ⎠
is Lorentz-invariant, and in particular has to be equal to the norm in the particle-

bound frame. But in that frame p = 0 and, according to Eq. (9.73), E = mc 2 , and so
the norm is just
⎛ E ⎞2 ⎛ mc 2 ⎞2
⎜ ⎟ = ⎜ ⎟ ≡ (mc )2 , (9.77b)
⎝c ⎠ ⎝ c ⎠
so that in an arbitrary frame

⎛ E ⎞2
⎜ ⎟ − p 2 = (mc )2 . (9.78a )
⎝c ⎠
This very important relation31 between the relativistic energy and momentum (valid
for free particles only!) is usually represented in the form32
E 2 = (mc 2 )2 + (pc )2 . (9.78b)
According to Eq. (9.70), in the ultra-relativistic limit u → c, p tends to infinity,

while mc2 stays constant, so that pc ≫ mc2. As follows from Eq. (9.78), in this limit
E ≈ pc . Although the above discussion was for particles with finite m, the four-
vector formalism allows us to consider particles with zero rest mass as ultra-
relativistic particles for which the above energy-to-moment relation,
E = pc , for m = 0, (9.79)
is exact. Quantum electrodynamics33 tells us that under certain conditions, the
electromagnetic field quanta (photons) may also be considered as such massless
31
Please note one more simple and useful relation following from Eqs. (9.70) and (9.73): p = (E /c 2 )u.
32
It may be tempting to interpret this relation as the perpendicular-vector-like addition of the rest energy mc2
and the ‘kinetic energy’ pc, but from the point of view of the total energy conservation (see below), a better
definition of the kinetic energy is T (u ) ≡ E (u ) − E (0).
33
Briefly reviewed in Part QM chapter 9.
9-22
particles, with momentum p = ℏk. Plugging (the modulus of) the last relation into
Eq. (9.78), for the photon’s energy we obtain E = pc = ℏkc = ℏω. Please note again
that according to Eq. (9.73), the relativistic mass of a photon is not equal to zero:
M = E /c2 = ℏω/c2, so that the term ‘massless particle’ has a limited meaning: m = 0.
For example, the relativistic mass of an optical phonon is of the order of 10−36 kg.
This is not too much, but still a noticeable (approximately one-millionth) part of the
rest mass me of an electron.
The fundamental relations (9.70) and (9.73) have been repeatedly verified in
numerous particle collision experiments in which the total energy and momentum of
a system of particles are conserved—at the same conditions as in non-relativistic
dynamics. (For the momentum this is the absence of external forces, and for the
energy this is the elasticity of particle interactions—in other words, the absence of
alternative channels of energy escape.) Of course, generally, the total energy of the
system is conserved, including the potential energy of particle interactions. However,
at typical high-energy particle collisions, the potential energy vanishes so rapidly
with the distance between them that we can use the momentum and energy
conservation laws using Eq. (9.73).
As an example, let us calculate the minimum energy Emin of a proton (pa),
necessary for the well-known high-energy reaction that generates a new proton–
antiproton pair, pa + pb → p + p + p + p, provided that before the collision the
proton pb has been at rest in the lab frame. This minimum evidently corresponds to
the vanishing relative velocity of the reaction products, i.e. their motion with
virtually the same velocity (ufin), as seen from the lab frame—see figure 9.10.
Figure 9.10. Schematic representation of a high-energy proton reaction at E ≈ Emin .
Due to the momentum conservation, this velocity should have the same direction
as the initial velocity (umin) of proton pa. This is why two scalar equations, for the
energy conservation
mc 2 4mc 2
2 2 1/2
+ mc 2 = 2 (9.80a )
(1 − u min / c ) (1 − u fin / c 2 )1/2
and for the momentum conservation
mu 4mu fin
2 2 1/2
+0= 2
, (9.80b)
(1 − u min / c ) (1 − u fin / c 2 )1/2
are sufficient to find both umin and ufin. After a conceptually simple but rather
tedious solution of this system of two nonlinear equations, we obtain
4 3 3 (9.81)
u min = c, u fin = c.
7 2
9-23
Finally, we can use Eq. (9.73) to calculate the required energy; the result is
Emin = 7mc 2 . (Note that of the kinetic energy of the initial moving particle, 6mc2,
only 2mc2 goes into the ‘useful’ proton–antiproton pair production.) The proton’s
rest mass, mp ≈ 1.67 × 10−27 kg, corresponds to mpc2 ≈ 1.502 × 10−10 J ≈ 0.938 GeV,
so that Emin ≈ 6.57 GeV .
The second, more intelligent way to solve the same problem is to use the center-
of-mass (c.o.m.) reference frame that, in relativity, is defined as the frame in which
the total momentum of the system vanishes34. In this frame, at E = Emin , the velocity
and momenta of all reaction products are equal to zero, while the velocities of
protons pa and pb before the collision are equal and opposite, with some magnitude
u′. Hence the energy conservation law becomes
2mc 2
= 4mc 2 , (9.82)
(1 − u′2 / c 2 )1/2
readily giving u′ = (√3/2) c. (This is of course the same result as Eq. (9.81) gives for
ufin.) Now we can use the fact that the velocity of the proton pb in the c.o.m. frame is
(−u′), and hence the velocity of the proton pa is (+u′). Hence we may find the lab-
frame speed of the proton pa using the velocity transform formula (9.25):
2u′
u min = . (9.83)
1 + u′2 / c 2
With the above result for u′, this relation gives the same result as the first method,
umin = (4√3/7)c, but in a much simpler way.
9.4 More on four-vectors and four-tensors

This is a good moment to describe a formalism that will allow us, in particular, to
solve the same proton collision problem in one more (and arguably, the most
elegant) way. Much more importantly, this formalism will be virtually necessary for
the description of the Lorentz transform of the electromagnetic field, and its
interaction with relativistic particles—otherwise the formulas would be too cum-
bersome. Let us call the four-vectors we have used before,
Aα ≡ {A0 , A}, (9.84)
contravariant, and denote them with the top index, and introduce also covariant
vectors,
Aα ≡ {A0 , −A}, (9.85)
marked by the lower index. Now if we form a scalar product of these vectors using
the standard (3D-like) rule, just as a sum of the products of the corresponding
components, we immediately obtain
34
Note that according to this definition, the c.o.m.’s radius-vector is R = ΣkMkrk/ΣkMk ≡ Σkγkmkrk/Σkγkmk, i.e.
is generally different from the well-known expression R = Σkmkrk/Σkmk of the non-relativistic mechanics.
9-24
Aα Aα ≡ Aα Aα ≡ A02 − A2 . (9.86)
Here and below the sign of the sum of four components of the product has been
dropped35.
The scalar product (9.86) is just the norm of the four-vector in our former
definition, and as we already know, is Lorentz-invariant. Moreover, the scalar
product of two different vectors (also a Lorentz invariant), may be written in any of
two similar forms36:
A0 B0 − A ⋅ B ≡ Aα B α = Aα Bα; (9.87)
again, the only caveat is to take one vector in the covariant, and another in the
contravariant form.
Now let us return to our sample problem (figure 9.10). Since all the components
(E /c and p) of the total four-momentum of our system are conserved at the collision,
its norm is conserved as well:
(pa + pb )α (pa + pb )α = (4p )α (4p )α . (9.88)
Since now the vector product is the usual math construct, we know that the
parentheses on the left-hand side of this equation may be multiplied as usual. We
may also swap the operands and move constant factors around as convenient. As a
result, we obtain
(pa )α (pa )α + (pb )α (pb )α + 2(pa )α (pb )α = 16pα p α . (9.89)
Thanks to the Lorentz-invariance of each of the terms, we may calculate it in the

reference frame we like. For the first two terms on the left-hand side, as well as for
the right-hand side term, it is beneficial to use the frames in which that particular
proton is at rest. As a result, according to Eq. (9.77b), each of the left-hand side
terms equals (mc)2, while the right-hand side equals 16(mc)2. In contrast, the last
term of the left-hand side is more easily evaluated in the lab frame, because in that
frame the three spatial components of the four-momentum pb vanish, and the scalar
product is the just the product of scalars E /c for protons a and b. For the latter
proton this ratio is just mc, so that we obtain a simple equation,
Emin
(mc )2 + (mc )2 + 2 mc = 16(mc )2 , (9.90)
c
immediately giving the final result Emin = 7mc 2 we had already obtained in two more
complex ways.
35
This compact notation may take some time to become accustomed to, but is very convenient (compact) and
can hardly lead to any confusion, due to the following rule: the summation is implied when, and only when an
index is repeated twice, one on the top and another at the bottom. In these notes, this shorthand notation will
be used only for four-vectors, but not for the usual (spatial) vectors.
36
Note also that, by definition, for any two four-vectors, AαBα = BαAα.
9-25
Let me hope that this example was a convincing demonstration of the conven-
ience of representing four-vectors in the contravariant (9.84) and covariant (9.85)
forms37 with Lorentz-invariant norms (9.86). To be useful for more complex tasks,
the formalism should be developed a little bit further. In particular, it is crucial to
know how the four-vectors change under the Lorentz transform. For contravariant
vectors, we already know the answer (9.54); let us rewrite it in the new notation:
Aα = L βαA′β . (9.91)
where L βα is the matrix (9.51), generally called the mixed Lorentz tensor38:
⎛ γ βγ 0 0 ⎞
⎜ ⎟
βγ γ 0 0 ⎟
L βα = ⎜ , (9.92)
⎜⎜ 0 0 1 0 ⎟⎟
⎝ 0 0 0 1⎠
Note that although the position of indices α and β in the Lorentz tensor notation is
not crucial, because it is symmetric, it is convenient to place them using the general
index balance rule: the difference of the numbers of the upper and lower indices
should be the same in both parts of any four-vector/tensor equality. (Check yourself
that all the formulas above do satisfy this rule.)
In order to rewrite Eq. (9.91) in a more general form that would not depend on
the particular orientation of the coordinate axes (figure 9.1), let us use the contra-
variant and covariant forms of the four-vector of the time-space interval (9.57),
dx α = {c dt , d r}, dxα = {c dt , −d r}; (9.93)
then its norm (9.58) may be represented as39
(ds )2 ≡ (c dt )2 − (dr )2 = dx α dxα = dxα dx α . (9.94)
37
These forms are four-vector extensions of the notions of contravariance and covariance (introduced in the
1850s by J Sylvester) for the description of the change of the usual (three-component) vectors at the transfer
between different reference frames—e.g. resulting from the frame rotation. In this case, the contravariance or
covariance of a vector is uniquely determined by its nature: if the Cartesian coordinates of a vector (such as the
non-relativistic velocity v = dr/dt) are transformed similarly to the radius-vector r, it is called contravariant,
while the vectors (such as ∇f) that require a reciprocal transform are called covariant. In the Minkowski space,
both forms may be used for any four-vector.
38
Just as the four-vectors, the four-tensors with two top indices are called contravariant, and those with two
bottom indices, covariant. The tensors with one top and one bottom index are called mixed.
39
Another way to write this relation is (ds)2 = gαβ dxαdxβ = gαβdxα dxβ, where double summation over indices α
and β is implied, and g is the so-called metric tensor,
⎛1 0 0 0 ⎞
⎜0 − 1 0 0 ⎟
g αβ ≡ gαβ ≡ ⎜ ⎟,
⎜0 0 − 1 0 ⎟
⎝0 0 0 − 1⎠
which may be used, in particular, to a transfer a covariant vector into the corresponding contravariant one and
back: Aα = gαβAβ, Aα = gαβ Aβ. The metric tensor plays a key role in general relativity, in which it is affected by
gravity—‘curved’ by particles’ masses.
9-26
Applying Eq. (9.91) to the contravariant form of vector (9.93), we obtain

dx α = L βαdx′β . (9.95)
But, with our new shorthand notation, we can also write the usual rule of
differentiation of each component xα, considering it as a (in our case, linear)
function of four arguments x′β, as follows40:
∂x α
dx α = dx′β . (9.96)
∂x′β
Comparing Eqs. (9.95) and (9.96), we can rewrite the general Lorentz transform rule
(9.92) in the new form,
∂x α β
Aα = A′ . (9.97a )
∂x′β
which evidently does not depend on the orientation of the coordinate axes.
It is straightforward to verify that the reciprocal transform may be represented as
∂x′α β
A′α = A . (9.97b)
∂x β
However, the reciprocal transform has to differ from the direct one only by the sign
of the relative velocity of the frames, so that the transform is given by the inverse
matrix ∂x′α/∂xβ; for the coordinate choice shown in figure 9.1 the matrix is
⎛ γ −βγ 0 0⎞
∂x′ α ⎜ ⎟
= ⎜−βγ γ 0 0⎟. (9.98)
∂x β ⎜⎜ 0 0 1 0 ⎟⎟
⎝ 0 0 0 1⎠
Since, according to Eqs. (9.84) and (9.85), covariant four-vectors differ from the
contravariant ones by the sign of the spatial components, their direct transform is
given by the matrix (9.98). Hence their direct and reciprocal transforms may be
represented, respectively, as
∂x′β ′ ∂x β
Aα = Aβ , Aα′ = Aβ , (9.99)
∂x α ∂x′α
evidently satisfying the index balance rule. (Note that primed quantities are now
multiplied, rather than divided as in the contravariant case.) As a sanity check, let us
apply this formalism to the scalar product AαAα. As Eq. (9.96) shows, the implicit
summation notation allows us to multiply and divide any equality by the same
partial differential of a coordinate, so that we can write:
∂x′β ∂x α ∂x′β
Aα Aα = A′ β A′γ
= A′ β A′γ = δβγA′ β A′γ = A′γA′γ , (9.100)
∂x α ∂x′γ ∂x′γ
i.e. the scalar product AαAα (as well as AαAα) is Lorentz-invariant, as it should be.
40
Note that in the index balance rule, the top index in the denominator of a fraction is counted as a bottom
index in the nominator, and vice versa.
9-27
Now, let us consider the four-vectors of derivatives. Here we should be very

careful. Consider, for example, the following vector operator
∂ ⎧ ∂ ⎫
≡⎨ , ∇⎬ . (9.101)
∂x α
⎩ ∂(ct ) ⎭
As was discussed above, the operator is not changed by its multiplication and
division by another differential, e.g. ∂x′β (with the corresponding implied summation
over β), so that
∂ ∂x′β ∂
= . (9.102)
α
∂x ∂x α ∂x′β
But, according to the first of Eqs. (9.99), this is exactly how the covariant vectors are
Lorentz-transformed! Hence, we have to consider the derivative over a contravariant
space–time interval as a covariant four-vector, and vice versa41. (This result might be
also expected from the index balance rule.) In particular, this means that the scalar
product
∂ α ∂A0
α
A ≡ +∇⋅A (9.103)
∂x ∂(ct )
should be Lorentz-invariant for any legitimate four-vector. A convenient shorthand
for the covariant derivative, which complies with the index balance rule, is
∂
≡ ∂α, (9.104)
∂x α
so that the invariant scalar product may be written just as ∂αAα. A similar definition
of the contravariant derivative,
∂ ⎧ ∂ ⎫
∂α ≡ =⎨ , −∇⎬ , (9.105)
∂xα ⎩ ∂(ct ) ⎭
allows us to write the Lorentz-invariant scalar product (9.103) in any of two forms:
∂A0
+ ∇ ⋅ A = ∂ αAα = ∂αAα . (9.106)
∂(ct )
Finally, let us see how does the general Lorentz transform change four-tensors. A
second-rank 4 × 4 matrix is a legitimate four-tensor if the both four-vectors it relates
obey the Lorentz transform. For example, if two legitimate four-vectors are related
as
Aα = T αβBβ , (9.107)
we should require that
41
As was mentioned above, this is also a property of the ‘usual’ reference-frame transform of 3D vectors.
9-28
A′α = T ′αβ B′ β , (9.108)

where Aα and A′α are related by Eqs. (9.97), while Bβ and B′β are related by
Eqs. (9.99). This requirement immediately yields
∂x α ∂x β γδ ∂x′α ∂x′β γδ
T αβ = T′ , T ′αβ = T , (9.109)
∂x′γ ∂x′δ ∂x γ ∂x δ
with the implied summation over two indices, γ and δ. The rules for the covariant
and mixed tensors are similar42.
9.5 Maxwell equations in the four-form

This four-vector formalism background is already sufficient to analyze the Lorentz
transform of the electromagnetic field. Just to warm up, let us consider the
continuity equation (4.5),
∂ρ
+ ∇ ⋅ j = 0, (9.110)
∂t
which expresses the electric charge conservation, and, as we already know, is
compatible with the Maxwell equations. If we now define the contravariant and
covariant four-vectors of electric current as
j α ≡ {ρc , j}, jα ≡ {ρc , −j}, (9.111)
then Eq. (9.110) may be represented in the form
∂ αjα = ∂αj α = 0, (9.112)
showing that the continuity equation is form-invariant43 with respect to the Lorentz
transform.
Of course, such a form’s invariance of a relation does not mean that all
component values of the four-vectors participating in it are the same in both frames.
For example, let us have some static charge density ρ in the frame 0; then
Eq. (9.97b), applied to the contravariant form of the four-vector (9.111), reads
∂x′α β
j ′α = j , with j β = {ρc , 0, 0, 0}. (9.113)
∂x β
Using the particular form (9.98) of the reciprocal Lorentz matrix for the coordinate
choice shown in figure 9.1, we see that this relation yields
ρ′ = γρ , j ′x = −γβρc = −γvρ , j ′ y = j ′z = 0. (9.114)
42
It is straightforward to check that transfer between the contravariant and covariant forms of the same tensor
may be readily achieved using the metric tensor g: Tαβ = gαγT γδgδβ, T αβ = gαγTγδ gδβ.
43
In some texts, the equations preserving their form at a transform are called ‘covariant’, creating a possibility
for confusion with the covariant vectors and tensors. On the other hand, calling such equations ‘invariant’
would not distinguish them properly from invariant quantities, such as the scalar products of four-vectors.
9-29
Since the charge velocity, as observed from frame 0′, is (−v), the non-relativistic
result would be j = −vρ. The additional γ factor in the relativistic results (for both the
charge density and the current) is caused by the length contraction dx′ = dx/γ, so that
in order to keep the total charge dQ = ρ d3r = ρ dx dy dz inside the elementary
volume d3r = dx dy dz intact, ρ (and hence jx) should increase proportionally.
Next, at the end of chapter 6 we have seen that Maxwell equations for potentials
ϕ and A may be represented in similar forms (6.118), under the Lorenz (again, not
‘Lorentz’ please!) gauge condition (6.117). For the free space, this condition takes
the form
1 ∂ϕ
∇⋅A+ = 0. (9.115)
c 2 ∂t
This expression gives us a hint as to how to form the four-vector of potentials44:
Aα ≡ { }ϕ
c
,A , Aα ≡ { ϕ
c
, −A ;} (9.116)
indeed, such vector satisfies Eq. (9.115) in its four-vector form:

∂ αAα = ∂αAα = 0. (9.117)
Since this scalar product is Lorentz-invariant, and the derivatives (9.104) and
(9.105) are legitimate four-vectors, this implies that the four-vector (9.116) is also
legitimate, i.e. obeys the Lorentz transform formulas (9.97) and (9.99). More
convincing evidence of this fact may be obtained from the Maxwell equation
(6.118) for the potentials. In the free space they may be rewritten as
⎡ ∂2 ⎤ (ρc ) ⎡ ∂2 ⎤
2 ϕ
⎢ − ∇ ⎥ = ≡ μ0(ρc ), ⎢ − ∇2
⎥A = μ0 j. (9.118)
⎣ ∂(ct )2 ⎦c ε0c 2 ⎣ ∂(ct )2 ⎦
Using the definition (9.116), these equations may be merged into one45:
□Aα = μ0j α , (9.119)
where □ is the d’Alembert operator,46 which may be represented as either of two
scalar products,
∂2
□ ≡ − ∇2 = ∂ β∂β = ∂β∂ β , (9.120)
∂(ct )2
and hence is Lorentz-invariant. Because of that, and the fact that the Lorentz
transform changes both four-vectors Aα and j α in a similar way, Eq. (9.119) does not
depend on the reference frame choice. Thus we have arrived at a key point of this
chapter: we see that the Maxwell equations are indeed form-invariant with respect to
44
In the Gaussian units, the scalar potential should not be divided by c in this relation.
45
In the Gaussian units, the coefficient μ0 in Eq. (9.119) should be replaced, as usual, with 4π/c.
46
Named after J-B d’Alembert (1717–83). (Some older textbooks use notation □2 for this operator.)
9-30
the Lorentz transform. As a by-product, the four-vector form (9.119) of these

equations (for potentials) is extremely simple—and beautiful.
However, as we have seen in chapter 7, for many applications the Maxwell
equations for the field vectors are more convenient, so let us represent them in the
four-form as well. For that, we may express all Cartesian components of the usual
(3D) field vectors
∂A
E = −∇ϕ − , B = ∇ × A, (9.121)
∂t
via those of the potential four-vector Aα. For example,
∂ϕ ∂Ax ⎛∂ ϕ ∂Ax ⎞
Ex = − − = − c⎜ + ⎟ ≡ −c(∂ 0A1 − ∂1A0 ), (9.122)
∂x ∂t ⎝ ∂x c ∂(ct ) ⎠
∂Az ∂Ay
Bx = − ≡ −(∂ 2A3 − ∂ 3A2 ). (9.123)
∂y ∂z
Completing similar calculations for other field components (or just generating them
by the appropriate index shifts), we find that the following asymmetric, contra-
variant field-strength tensor,
F αβ ≡ ∂ αAβ − ∂ βAα , (9.124)
may be expressed via the field components as follows47:
⎛ 0 −Ex / c −Ey / c −Ez / c ⎞
⎜ ⎟
⎜ Ex / c 0 − Bz By ⎟
F αβ =⎜ , (9.125a )
Ey / c Bz 0 −Bx ⎟
⎜ ⎟
⎝ Ez / c −By Bx 0 ⎠
so that the covariant form of the tensor is

⎛ 0 Ex / c Ey / c Ez / c ⎞
⎜ ⎟
γδ ⎜−Ex / c 0 −Bz By ⎟
Fαβ ≡ gαγF gδβ = ⎜ . (9.125b)
−Ey / c Bz 0 −Bx ⎟
⎜ ⎟
⎝ −Ez / c −By Bx 0 ⎠
If this expression looks a bit too bulky, please note that as a reward, the pair of
inhomogeneous Maxwell equations, i.e. the two first equations of the system (6.99),
which in free space (D = ε0E, B = μ0H) may be rewritten as
47
In Gaussian units, this formula, as well as Eq. (9.131) for Gαβ, does not have the factors c in all the
denominators.
9-31
E ∂ E
∇⋅ = μ 0 cρ , ∇×B− = μ0j , (9.126)
c ∂(ct ) c
may now be represented in a very simple (and manifestly form-invariant) way,
∂αF αβ = μ0j β , (9.127)
which is comparable with Eq. (9.119) in its beauty and simplicity. Somewhat
counter-intuitively, the pair of homogeneous Maxwell equations (6.99),
∂B
∇×E+ = 0, ∇ ⋅ B = 0, (9.128)
∂t
look, in the four-vector notation, a bit more complicated48:
∂αFβγ + ∂βFγα + ∂γFαβ = 0. (9.129)
Note, however, that Eqs. (9.128) may be also represented in a much simpler form,
∂αG αβ = 0, (9.130)
using the so-called dual tensor
⎛ 0 Bx By Bz ⎞
⎜ ⎟
⎜−Bx 0 −Ez / c Ey / c ⎟
G αβ =⎜ , (9.131)
−By Ez / c 0 −Ex / c ⎟
⎜ ⎟
⎝ −Bz −Ey / c Ex / c 0 ⎠
which may be obtained from F αβ, given by Eq. (9.125), with the following
replacements:
E E
→ −B, B→ . (9.132)
c c
In addition to the proof of the form-invariance of the Maxwell equations with
respect to the Lorentz transform, the four-vector formalism allows us to achieve our
initial goal: find out how do the electric and magnetic field components change at the
transfer between (inertial) reference frames. Let us apply to the tensor F αβ the
reciprocal Lorentz transform described by the second of Eqs. (9.109). Generally, it
gives, for each field component, a sum of 16 terms, but since (for our choice of
coordinates, shown in figure 9.1) there are many zeros in the Lorentz transform
matrix, and the diagonal components of F γδ equal zero as well, the calculations are
rather doable. Let us calculate, for example, E′x ≡ −cF′01. The only non-vanishing
terms on the right-hand side are
48
To be fair, note that just as Eq. (9.127), Eq. (9.129) is also a set of four scalar equations—in the latter case
with indices α, β, and γ taking any three different values of the set {0, 1, 2, 3}.
9-32
⎛ ∂x′0 ∂x′1 ∂x′0 ∂x′1 01⎞ E

Ex′ = −cF 01 = −c⎜ 1 F 10
+ F ⎟ ≡ −cγ 2(β 2 − 1) x ≡ Ex. (9.133)
⎝ ∂x ∂x 0 0
∂x ∂x 1 ⎠ c
Repeating the calculation for the other five components of the fields, we obtain the
very important relations
Ex′ = Ex, Bx′ = Bx,
Ey′ = γ (Ey − vBz ), By′ = γ (By + vEz / c 2 ), (9.134)
2
Ez′ = γ (Ez + vBy ), Bz′ = γ (Bz − vEy / c ),
whose more compact ‘semi-vector’ form is

E′ = E , B′ = B ,
(9.135)
E′⊥ = γ (E + v × B)⊥ , B′⊥ = γ (B − v × E/ c 2 )⊥ ,
where the indices ∣∣ and ⊥ stand, respectively, for the field components parallel and
perpendicular to the relative velocity v of the two reference frames. In the non-
relativistic limit, the Lorentz factor γ tends to 1, and Eqs. (9.135) acquire an even
simpler form
1
E′ → E + v × B, B′ → B − v × E. (9.136)
c2
Thus we see that the electric and magnetic fields actually transform to each other
even in the first order of the v/c ratio. For example, if we fly across the field lines of a
uniform, static, purely electric field E (e.g. the one in a plane capacitor) we will see
not only the electric field’s re-normalization (in the second order of the v/c ratio), but
also a non-vanishing dc magnetic field B′ perpendicular to both the vector E and the
vector v, i.e. to the direction of our motion. This is of course what might be expected
from the relativity principle: from the point of view of the moving observer (which is
as legitimate as that of a stationary observer), the surface charges of the capacitor’s
plates, which create the field E, move back creating the dc currents (9.114), which
induce the magnetic field B′. Similarly, motion across a magnetic field creates, from
the point of view of the moving observer, an electric field.
This fact is very important philosophically. One can say there is no such thing in
Mother Nature as an electric field (or a magnetic field) all by itself. Not only can the
electric field induce the magnetic field (and vice versa) in dynamics, but even in an
apparently static configuration, what exactly we measure depends on our speed
relative to the field sources—hence the very appropriate term for the whole field we
are studying: electromagnetism.
Another simple but very important application of Eqs. (9.134) and (9.135) is the
calculation of the fields created by a charged particle moving in free space by inertia,
i.e. along a straight line with constant velocity u, at the impact parameter49 (the
closest distance) b from the observer. Selecting the frame 0′ to move with the particle
49
This term is very popular in the theory of particle scattering—see, e.g. Part CM section 3.7.
9-33
Figure 9.11. The field pulses induced by a uniformly moving charge.
in its origin, and the frame 0 to reside in the ‘lab’ (in that the fields E and B are
measured), we can use the above formulas with v = u. In this case the fields E′ and B′
may be calculated from, respectively, electro- and magnetostatics,
q r′
E′ = , B′ = 0, (9.137)
4πε0 r′3
because in the frame 0′ the particle does not move. Selecting the coordinate axes so
that at the measurement point x = 0, y = b, z = 0 (figure 9.11a), for this point we may
write x′ = −ut′, y′ = b, z′ = 0, so that r′ = (u2t′2 + b2)1/2, and the Cartesian
components of the fields (9.137) are:
q ut′ q b
Ex′ = − , Ey′ = ,
4πε0 (u t′ + b 2 )3/2
2 2
4πε0 (u t′ + b 2 )3/2
2 2
(9.138)
Ez′ = 0, Bx′ = By′ = Bz′ = 0.
Now using the last of Eqs. (9.19b), with x = 0, for the time transform, and the
equations reciprocal to Eq. (9.134) for the field transform (it is evident that they are
similar to the direct transform, with v replaced with −v = −u), in the lab frame we
obtain
q uγ t
Ex = Ex′ = − ,
4πε0 (u γ t 2 + b 2 )3/2
2 2
(9.139)
q γb
Ey = γEy′ = , Ez = 0,
4πε0 (u γ t + b 2 )3/2
2 2 2
γu ′ u q γb u
Bx = 0, By = 0, Bz = E = 2
2 y 3/2
≡ 2 Ey. (9.140)
c 2 2 2 2
c 4πε0 (u γ t + b ) c
9-34
These results50, plotted in figure 9.11b, reveal two major effects. First, the charge
passage by the observer generates not only an electric field pulse, but also a magnetic
field pulse. This is natural, because, as was repeatedly discussed in chapter 5, any
charge motion is essentially an electric current51. Second, Eqs. (9.139) and (9.140)
show that the pulse duration scale is
b⎛ u2 ⎞
1/2
b
Δt = = ⎜1 − 2 ⎟ , (9.141)
γu u⎝ c ⎠
i.e. shrinks to zero as the charge’s velocity u approaches the speed of light. This is of
course a direct corollary of the relativistic length contraction: in the frame 0′ moving
with the charge, the longitudinal spread of its electric field at distance b from the
motion line is of the order of Δx′ = b. When observed from the lab frame 0, this
interval, in accordance with (9.20), shrinks to Δx = Δx′/γ = b/γ, and so does the pulse
duration scale Δt = Δx/u = b/γu.
9.6 Relativistic particles in electric and magnetic fields

Now let us analyze the dynamics of charged particles in electric and magnetic fields.
Inspired by ‘our’ success in forming the four-vector (9.75) of the energy–momentum,
dx α
pα = { }E
c
,p = γ{mc , p} = m
dτ
≡ mu α, (9.142)
where uα is the contravariant form of the four-velocity (9.63) of the particle,

dx α dxα
uα ≡ , uα ≡ , (9.143)
dτ dτ
we may notice that the non-relativistic equation of motion, resulting from the
Lorentz-force formula (5.10) for the three spatial components of pα, at a charged
particle’s motion in an electromagnetic field,
dp
= q(E + u × B), (9.144)
dt
is fully consistent with the following four-vector equality (which is evidently form-
invariant with respect to the Lorentz transform):
dp α
= qF αβuβ . (9.145)
dτ
For example, according to Eq. (9.125), the α = 1 component of this equation reads
50
In the next chapter, we will re-derive them in a different way.
51
It is straightforward to use Eq. (9.140) and the linear superposition principle to calculate, for example, the
magnetic field of a string of charges moving along the same line, and separated by equal distances Δx = a (so
that the average current, as measured in frame 0, is qu/a), and to show that the time-average of the magnetic
field is given by the familiar Eq. (5.20) of magnetostatics, with b instead of ρ.
9-35
dp1 ⎡E ⎤
= qF 1βuβ = q⎢ x γc + 0 ⋅ ( −γux ) + ( −Bz )( −γu y ) + By( −γuz )⎥
dτ ⎣c ⎦ (9.146)
= qγ [E + u × B]x ,
and similarly for two other spatial components (α = 2 and α = 3). It may look like
these expressions differ from the Newton’s second law (9.144) by the extra factor γ.
However, plugging into Eq. (9.146) the definition of the proper time interval, dτ =
dt/γ, and canceling γ in both parts, we recover Eq. (9.144) exactly—for any velocity
of the particle! The only caveat is that if u is comparable with c, the vector p in
Eq. (9.144) has to be understood as the relativistic momentum (9.70) proportional to
the velocity-dependent mass M = γm ⩾ m rather than to the rest mass m.
The only remaining task is to examine the meaning of the zeroth component of
Eq. (9.145). Let us spell it out:
dp0 ⎡ ⎛ E ⎞ ⎛ Ey ⎞ ⎛ E ⎞ ⎤
= qF 0βuβ = q⎢0 ⋅ γc + ⎜ − x ⎟( − γux ) + ⎜ − ⎟( − γuy ) + ⎜ − z ⎟( − γuz )⎥
dτ ⎣ ⎝ c ⎠ ⎝ c ⎠ ⎝ c ⎠ ⎦ (9.147)
E
= qγ ⋅ u.
c
Recalling that p0 = E /c , and using the basic relation dτ = dt/γ again, we see that
Eq. (9.147) looks exactly like the non-relativistic relation for the kinetic energy
change52,
dE
= qE ⋅ u, (9.148)
dt
except that in the relativistic case the energy has to be taken in the general form
(9.73).
There is no question that the four-component equation (9.145) of the relativistic
dynamics is beautiful in its simplicity. However, for the solution of particular
problems, Eqs. (9.144) and (9.148) are frequently more convenient. As an illus-
tration of this point, let us now use these equations to explore the relativistic effects
at charged particle motion in uniform, time-independent electric and magnetic fields.
In doing that, we will, for the time being, neglect the contributions into the field by
the particle itself 53.
(i) Uniform magnetic field. Let the magnetic field be constant and uniform in the
‘lab’ reference frame 0. Then in this frame, Eqs. (9.144) and (9.148) yield
dp dE
= q u × B, = 0. (9.149)
dt dt
52
See, e.g. Part CM Eq. (1.20) with dp/dt = F = qE. (As a reminder, the magnetic field cannot affect the
particle’s energy, because the magnetic component of the Lorentz force is perpendicular to its velocity.)
53
As was emphasized earlier in this course, in statics this contribution has to be ignored. In dynamics, this is
generally not true; these self-action effects will be discussed in section 10.6.
9-36
From the second equation, E = const , we obtain u = const, β ≡ u/c = const,

γ ≡ (1 − β2)−1/2 = const, and M ≡ γm = const, so that the first of Eqs. (9.149) may
be rewritten as
du
= u × ωc, (9.150)
dt
where ωc is the vector directed along the magnetic field B, with a magnitude equal to
the cyclotron frequency (sometimes called ‘gyrofrequency’)
qB qB qc 2B
ωc ≡ = = . (9.151)
M γm E
If the particle’s initial velocity u0 is perpendicular to the magnetic field, Eq. (9.150)
describes its circular motion with a constant speed u = u0 in a plane perpendicular to
B, with the angular velocity (9.151). In the non-relativistic limit u ≪ c, when γ → 1, i.e.
M → m, the cyclotron frequency ωc = qB/m, i.e. is independent of the speed. However,
as the kinetic energy is increased to become comparable with the rest energy mc2 of the
particle, the frequency decreases, and in the ultra-relativistic limit
B qB
ωc ≈ qc ≪ , at u ≈ c. (9.152)
p m
The cyclotron motion’s radius may be calculated as R = u/ωc; in the non-
relativistic limit it is proportional to the particle’s speed, i.e. to the square root of its
kinetic energy. However, as Eq. (9.151) shows, in the general case the radius is
proportional to the particle’s relativistic momentum rather than its speed:
u Mu mγu 1 p
R= = = = , (9.153)
ωc qB qB qB
so that in the ultra-relativistic limit, when p ≈ E /c , R is proportional to the kinetic
energy.
This dependence of ωc and R on energy are the major factors in the design of
circular accelerators of charged particles. In the simplest of these machines (the
cyclotron, invented in 1929 by E Lawrence), the frequency ω of the accelerating ac
electric field is constant, so that even if it is tuned to the ωc of the initially injected
particles, the drop of the cyclotron frequency with energy eventually violates this
tuning. For to this reason, the largest achievable particle speed is limited to just ~0.1c
(for protons, corresponding to the kinetic energy of just ~15 MeV). This problem may
be addressed in several ways. In particular, in synchrotrons (such as Fermilab’s
Tevatron and the CERN’s Large Hadron Collider, LHC54) the magnetic field is
gradually increased in time to compensate the momentum increase (B ∝ p), so that
both R (9.148) and ωc (9.147) stay constant, enabling the proton acceleration to
energies as high as ∼7 TeV, i.e. ∼2000 mc2.55
54
https://home.cern/topics/large-hadron-collider
55
For more on this topic, I have to refer the interested reader to special literature, for example either [6] or [7].
9-37
Returning to our initial problem, if the particle’s initial velocity has a component
u∣∣ along the magnetic field, then it is conserved in time, so that the trajectory is a
spiral around the magnetic field lines. As Eqs. (9.149) show, in this case Eq. (9.150)
remains valid, but in Eqs. (9.151) and (9.153) the full speed and momentum have to
be replaced with magnitudes of their (also time-conserved) components, u⊥ and p⊥,
normal to B, while the Lorentz factor γ in those formulas still includes the full speed
of the particle.
Finally, in the special case when particle’s initial velocity is directed exactly along
the magnetic field’s direction, it continues to move in a straight line along vector B.
In this case, the cyclotron frequency still has its non-zero value (9.151), but does not
correspond to any real motion, because R = 0.
(ii) Uniform electric field. This problem is (technically) more complex than the
previous one, because in the electric field, the particle’s energy may change.
Directing the axis z along the field E, from Eq. (9.144) we obtain
dpz d p⊥
= qE , = 0. (9.154)
dt dt
If the field does not change in time, the first integration of these equations is
elementary,
pz (t ) = pz (0) + qEt , p⊥(t ) = const = p⊥(0), (9.155)
but the further integration requires care, because the effective mass M = γm of the
particle depends on its full speed u, with
u 2 = u z2 + u ⊥2, (9.156)
making the two motions, along and across the field, mutually dependent.
If the initial velocity is perpendicular to the field E, i.e. if pz(0) = 0, p⊥(0) = p(0) ≡
p0, the easiest way to proceed is to calculate the kinetic energy first:
E 2 = (mc 2 )2 + c 2p 2 (t ) = E02 + c 2(qEt )2 , where E0 ≡ [(mc 2 )2 + c 2p02 ]1/2 . (9.157)
On the other hand, we can calculate the same energy by integrating Eq. (9.148),
dE dz
= q E ⋅ u ≡ qE , (9.158)
dt dt
over time, with a simple result:
E = E0 + qEz(t ), (9.159)
where (for simplicity of notation) I took z(0) = 0. Requiring Eq. (9.159) to give the
same E 2 as Eq. (9.157), we obtain a quadratic equation for z(t),
E02 + c 2(qEt )2 = [E0 + qEz(t )]2 , (9.160)
whose solution (with the sign before the square root corresponding to E > 0, i.e.
z ⩾ 0) is
9-38
⎧ ⎫
E0 ⎪⎡ ⎛ cqEt ⎞2 ⎤
1/2
⎪
z (t ) = ⎨⎢1 + ⎜ ⎟ ⎥ − 1⎬ . (9.161)
⎩⎢⎣
qE ⎪ ⎝ E0 ⎠ ⎥⎦ ⎪
⎭
Now let us find the particle’s trajectory. Selecting the axis x so that the initial
velocity vector (and hence the velocity vector at any further instant) is within the
[x, z] plane, i.e. that y(t) ≡ 0, we may use Eqs. (9.155) to calculate the trajectory’s
slope, at its arbitrary point, as
dz dz / dt Muz p qEt
≡ ≡ ≡ z = . (9.162)
dx dx / dt Mux px p0
Now let us use Eq. (9.160) to express the nominator of this fraction, qEt, as a
function of z:
1
qEt = [(E0 + qEz )2 − E02 ]1/2 . (9.163)
c
Plugging this expression into Eq. (9.161) we obtain
dz 1
= [(E0 + qEz )2 − E02 ]1/2 . (9.164)
dx cp0
This differential equation may be readily integrated, separating the variables z and
x, and using the following substitution: ξ ≡ cosh−1(qEz/E 0 + 1). Selecting the origin
of axis x at the initial point, so that x(0) = 0, we finally obtain the trajectory
E0 ⎛ qEx ⎞
z= ⎜cosh − 1⎟ . (9.165)
qE ⎝ cp0 ⎠
This curve is usually called catenary, but sometimes the ‘chainette’, because it
(with the proper constant replacement) describes the stationary shape of a heavy,
uniform chain in a uniform gravity field, directed along axis z. At the initial part of
the trajectory, where qEx ≪ cp0(0), this expression may be approximated by the first
non-zero term of its Taylor expansion in small x, giving a parabola,
E0qE ⎛ x ⎞
2
z= ⎜ ⎟ , (9.166)
2 ⎝ cp0 ⎠
so that if the initial velocity of the particle is much less than c (i.e. p0 ≈ mu0,
E0 ≈ mc 2 ), we obtain the familiar non-relativistic formula:
qE 2 a 2 F qE
z= x ≡ t , where a = = . (9.167)
2mu 02 2 m m
The straightforward generalization of this solution to the case of an arbitrary

direction of the particle’s initial velocity is left for the reader’s exercise.
9-39
(iii) Crossed uniform magnetic and electric fields (E ⊥ B). In the view of the
somewhat bulky solution of the previous problem (i.e. the particular case of the
current problem for B = 0), one might think that the new problem should be
forbiddingly complex for an analytical solution. Counter-intuitively, this is not the
case, due to the help from the field transform relations (9.135). Let us consider two
possible cases.
Case 1: E/c < B. Let us consider an inertial reference frame moving (relatively the
‘lab’ reference frame 0 in which fields E and B are defined) with the following
velocity:
E×B
v= , (9.168)
B2
whose magnitude v = c × (E/c)/B < c. Selecting the coordinate axes as shown in
figure 9.12, so that
Ex = 0, Ey = E , Ez = 0; Bx = 0, By = 0, Bz = 0, (9.169)
we see that the Cartesian components of this velocity are vx = v, vy = vz = 0.
Figure 9.12. The particle’s trajectory in crossed electric and magnetic fields (at E/c < B).
Since this choice of the coordinates complies with the one used to derive
Eqs. (9.134), we can readily use that simple form of the Lorentz transform to
calculate the field components in the moving reference frame:
⎛ E ⎞
Ex′ = 0, Ey′ = γ (E − vB ) = γ ⎜E − B⎟ = 0, Ez′ = 0, (9.170)
⎝ B ⎠
Bx′ = 0, By′ = 0,
⎛ vE ⎞ ⎛ vE ⎞ ⎛ v2 ⎞ B (9.171)
Bz′ = γ ⎜B − 2 ⎟ = γB⎜1 − ⎟ = γ B ⎜1 − ⎟= ⩽ B,
⎝ c ⎠ ⎝ Bc 2 ⎠ ⎝ c2 ⎠ γ
where the Lorentz parameter γ ≡ (1 − v2/c2)−1/2 corresponds to the velocity (9.168)

rather than that of the particle. These relations show that in this special reference
frame the particle only sees a (re-normalized) uniform magnetic field B′ ⩽ B, parallel
to the initial field, i.e. perpendicular to the velocity (9.168). Using the result of the
above example (i), we see that in this frame the particle moves along either a circle or
9-40
a spiral winding about the direction of the magnetic field, with the angular speed
(9.151),
qB′
ωc′ = , (9.172)
E ′/c 2
and radius (9.153):
p′
R′ = ⊥ . (9.173)
qB′
Hence in the lab frame, the particle performs this orbital motion plus a ‘drift’
with the constant velocity v (figure 9.12). As a result, the lab-frame trajectory of the
particle (or rater its projection onto the plane perpendicular to the magnetic field) is
a trochoid-like curve56 that, depending on the initial velocity, may be either prolate
(self-crossing), as in figure 9.12, or curtate (stretched so much that it is not self-
crossing).
Such looped motion of electrons (in practice, with v ≪ c) is used, in particular, in
magnetrons—very popular generators of microwave radiation. In such device (figure
9.13), the magnetic field, usually created by specially shaped permanent magnets, is
nearly uniform (in the region of electron motion) and directed along the magnet-
ron’s axis (in figure 9.13, normal to the plane of drawing), while the electric field of
magnitude E ≪ cB, created by the dc voltage applied between the anode and
cathode, is virtually radial.
Figure 9.13. The cross-section of a typical magnetron. Figure adapted from https://en.wikipedia.org/wiki/
Cavity_magnetron under the Free GNU Documentation License.
As a result, the above simple theory is only approximately valid, and the electron
trajectories are close to epicycloids rather than trochoids. The applied electric field is
adjusted so that these looped trajectories pass close to the anode’s surface, and hence
to the gap openings of the cylindrical microwave cavities drilled in the anode’s bulk.
56
As a reminder, a trochoid may be described as the trajectory of a point on a rigid disk rolled along a straight
line. Its canonical parametric representation is x = Θ + acos Θ, y = asin Θ. (For a > 1, the trochoid is prolate, if
a < 1, it is curtate, and if a = 1, it is called the cycloid.) Note, however, that for our problem, the trajectory in
the lab frame is exactly trochoidal only in the non-relativistic limit v ≪ c (i.e. E/c ≪ B), because otherwise the
Lorentz contraction in the drift direction squeezes the cyclotron orbit from a circle into an ellipse.
9-41
The fundamental mode of such a cavity is quasi-lumped, with the cylindrical walls
working mostly as lumped inductances, and the gap openings as lumped capaci-
tances, with the microwave electric field concentrated in these openings. This is why
the mode is strongly coupled to the electrons passing nearby, and their interaction
creates a large positive feedback (equivalent to a negative damping), which results in
intensive microwave self-oscillations at the cavities’ own frequency57. The oscillation
energy, of course, is taken from the dc-field-accelerated electrons; due to the energy
loss, the looped trajectory of each electron gradually moves closer to the anode and
finally lands on its surface. The widespread use of such generators (in particular, in
microwave ovens, which operate in a narrow frequency band around 2.45 GHz,
allocated for these devices to avoid their interference with wireless communication
systems) is due to their simplicity and high (up to 65%) efficiency.
Case 2: E/c > B. In this case, the speed given by Eq. (9.168) would be above the speed
of light, so let us introduce a reference frame moving with a different velocity,
E×B
v= , (9.174)
(E / c )2
whose direction is the same as before (figure 9.12), and magnitude v = c × B/(E/c)
is again below c. A calculation absolutely similar to the one performed above for
case 1, yields
Ex′ = 0, Ey′ = γ (E − vB )
⎛ vB ⎞ ⎛ v2 ⎞ E (9.175)
= γE ⎜1 − ⎟ = γE ⎜1 − ⎟= ⩽ E, Ez′ = 0,
⎝ E ⎠ ⎝ c2 ⎠ γ
⎛ vE ⎞ ⎛ EB ⎞
Bx′ = 0, By′ = 0, Bz′ = γ ⎜B − 2 ⎟ = γ ⎜B − ⎟ = 0. (9.176)
⎝ c ⎠ ⎝ E ⎠
so that in the moving frame the particle sees only an electric field E′ ⩽ E. According
to the solution of our previous problem (ii), the trajectory of the particle in the
moving frame is the catenary (9.165), so that in the lab frame it has an ‘open’,
hyperbolic character as well.
To conclude this section, let me note that if the electric and magnetic fields are
non-uniform, the particle motion may be much more complex, and in most cases the
integration of Eqs. (9.144) and (9.148) may be carried out only numerically.
However, if the field’s non-uniformity is small, approximate analytical methods
may be very effective. For example, if the magnetic field has a small transverse
gradient ∇B in a direction perpendicular to the vector B itself, such that
∇B 1
η≡ ≪ , (9.177)
B R
57
9-42
where R is the cyclotron radius (9.153), then it is straightforward to use Eq. (9.150)
to show58 that the cyclotron orbit drifts perpendicular to both B and ∇B, with the
drift speed
η ⎛1 2 ⎞
vd ≈ ⎜ u ⊥ + u 2⎟ ≪ u . (9.178)
ωc ⎝ 2 ⎠
The physics of this drift is rather simple: according to Eq. (9.153), the instant
curvature of the cyclotron orbit is proportional to the local value of the field. Hence
if the field is non-uniform, the trajectory bends slightly more on its parts passing
through stronger field, thus acquiring a shape close to a curate trochoid.
For experimental physics and engineering practice, the effects of longitudinal
gradients of magnetic field on the charged particle motion are much more
important, but let me postpone their discussion until we have developed a few
more analytical tools in the next section.
9.7 Analytical mechanics of charged particles

Eq. (9.145) gives a full description of relativistic particle dynamics in electric and
magnetic fields, just as Newton’s second law (9.1) does in the non-relativistic limit.
However, we know that in the latter case the Lagrange formalism of analytical
mechanics allows an easier solution of many problems59. We can fully expect that to
be true in relativistic mechanics as well, so let us expand the analysis of section 9.3 to
particles in the field.
Let recall that for a free particle, our main result was Eq. (9.68), which may be
rewritten as
γL = −mc 2 , (9.179)
with γ ≡ (1 − u2/c2)−1/2, showing that the product on the left-hand side is Lorentz-
invariant. How can the electromagnetic field affect this relation? In non-relativistic
electrostatics, we could write
L = T − U = T − qϕ . (9.180)
However, in relativity the scalar potential ϕ is just one component of the potential
four-vector (9.116). The only way to obtain a Lorentz-invariant contribution to γL
from the full four-vector (9.116), that would also be proportional to the Lorentz
force, i.e. to the first power of the particle’s velocity (to account for the magnetic
component of the Lorentz force), is evidently
γL = −mc 2 + const × u αAα , (9.181)
58
See, e.g. section 12.4 in [8].
59
See, e.g. Part CM section 2.2 and beyond.
9-43
where uα is the four-velocity (9.63). In order to comply with Eq. (9.180) in

electrostatics, the constant factor should be equal to (−qc), so that Eq. (9.181)
becomes
γL = −mc 2 − qu αAα (9.182)
and, finally, we obtain an extremely important equality,
mc 2
L=− − qϕ + q u ⋅ A , (9.183)
γ
whose Cartesian form is
⎛ u x2 + u y2 + u z2 ⎞
1/2
2⎜ ⎟ (9.184)
L = −mc ⎜1 − ⎟ − qϕ + q(uxAx + u yAy + uzAz ).
⎝ c2 ⎠
Let us see whether this relation (which admittedly was obtained by an educated
guess rather than by a strict derivation) passes a natural sanity check. For the case
of unconstrained motion of a particle, we can select its three Cartesian coordinates
rj ( j = 1, 2, 3) as the generalized coordinates, and its linear velocity components uj as
the corresponding generalized velocities. In this case, the Lagrange equations of
motion are60
d ∂L ∂L
− = 0. (9.185)
dt ∂uj ∂rj
For example, for r1 = x, Eq. (9.184) yields

∂L mux ∂L ∂ϕ ∂A
= + qAx ≡ px + qAx , = −q + qu ⋅ , (9.186)
∂ux (1 − u 2 / c 2 )1/2 ∂x ∂x ∂x
so that Eq. (9.185) takes the form
dpx ∂ϕ ∂A dA
= −q + qu ⋅ − q x. (9.187)
dt ∂x ∂x dt
In the equations of motion, the field values have to be taken at the instant position of
the particle, so that the last (full) derivative has components due to both the actual
field’s change (at a fixed point of space) and the particle’s motion. Such addition is
described by the so-called convective derivative61
d ∂
= + u ⋅ ∇. (9.188)
dt ∂t
60
61
Alternatively called the ‘Lagrangian derivative’; for its (rather simple) derivation see, e.g. Part CM
section 8.3.
9-44
Spelling out both scalar products, we may group the terms remaining after
cancellations as follows:
dpx ⎡⎛ ∂ϕ ∂Ax ⎞ ⎛ ∂Ay ∂Ax ⎞ ⎛ ∂Ax ∂Az ⎞⎤
= q⎢⎜ − − ⎟ + u y⎜ − ⎟ − uz⎜ − ⎟⎥ . (9.189)
dt ⎣⎝ ∂x ∂t ⎠ ⎝ ∂x ∂y ⎠ ⎝ ∂z ∂x ⎠⎦
But taking into account the relations (9.121) between the electric and magnetic fields
and potentials, this expression is nothing more than
dpx
= q(Ex + u yBz − uzBy ) = q(E + u × B)x , (9.190)
dt
i.e. the x-component of Eq. (9.144). Since other Cartesian coordinates participate in
(9.184) in a similar way, it is evident that the Lagrangian equations of motion along
other coordinates yield other components of the same vector equation of motion.
So, Eq. (9.183) does indeed give the correct Lagrangian function, and we can use
it for further analysis, in particular to discuss the first of Eqs. (9.186). This relation
shows that in the electromagnetic field, the generalized momentum corresponding to
particle’s coordinate x is not px = mγux, but62
∂L
Px ≡ = px + qAx . (9.191)
∂ux
Thus, as was already discussed (at that point, without a proof) in section 6.4, the
particle’s motion in a magnetic field may be described by two different momentum
vectors: the kinetic momentum p, defined by Eq. (9.70), and the canonical (or
‘conjugate’) momentum63
P = p + q A. (9.192)
In order to facilitate the discussion of this notion, let us generalize expression (9.72)
for the Hamiltonian function H of a free particle to the case of a particle in the field:
⎛ mc 2 ⎞
H ≡ P ⋅ u − L = (p + q A) ⋅ u − ⎜ − + q u ⋅ A − qϕ⎟
⎝ γ ⎠
(9.193)
2
mc
=p⋅u+ + qϕ .
γ
Merging the first two terms exactly as it was done in Eq. (9.72), we obtain an
extremely simple result,
H = γmc 2 + qϕ , (9.194)
which may leave us wondering: where is the vector-potential A here—and the
magnetic field effects it has to describe? The resolution of this puzzle is easy: for a
practical use (e.g. for the alternative derivation of the equations of motion), H has
to be represented as a function of the particle’s generalized coordinates (in the case
of unconstrained motion, these may be the Cartesian components of vector r that
62
With regret, I have to use the same (common) notation as was used earlier for the electric polarization—
which is not discussed below.
63
In Gaussian units, Eq. (9.192) has the form P = p + qA/c.
9-45
serves as an argument for potentials A and ϕ), and the generalized momenta, i.e. the
Cartesian components of the vector P (plus, generally, time). Hence, the velocity u
and the factor γ should be eliminated from (9.194). This may be done using the
relation (9.192), γmu = P − qA. For such an elimination, it is sufficient to notice that
according to Eq. (9.194), the difference (H − qϕ) is equal to the right-hand side of
Eq. (9.72), so that the generalization of Eq. (9.78) is64
(H − qϕ)2 = (mc 2 )2 + c 2(P − q A)2 , giving
⎡ ⎛ p ⎞2 ⎤
1/2
(9.195)
H = mc ⎢1 +
2 ⎜ ⎟ ⎥ + qϕ .
⎣ ⎝ mc ⎠ ⎦
It is straightforward to verify that the Hamilton equations of motion for three

Cartesian coordinates of the particle, obtained in the regular way65 from this H ,
may be merged into the same vector equation (9.144). In the non-relativistic limit,
performing the Taylor expansion of the latter of Eqs. (9.195) in p2, and limiting it to
two leading terms, we obtain the following generalization of Eq. (9.74):
p2 1
H ≈ mc 2 + + qϕ , i.e. H − mc 2 ≈ (P − q A)2 + U ,
2m 2m (9.196)
with U = qϕ .
These expressions for H , and Eq. (9.183) for L , give a clear view of the
electromagnetic field effect account in analytical mechanics. The electric part of the
total Lorentz force q(E + u × B) can perform work on the particle, i.e. change its
kinetic energy—see Eq. (9.148) and its discussion. As a result, the scalar potential ϕ,
whose gradient gives a contribution into E, may be directly associated with the
potential energy U = qϕ of the particle. In contrast, the magnetic component qu × B
of the Lorentz force is always perpendicular to the particle’s velocity u, and cannot
perform non-zero work on it, and as a result cannot be described by a contribution
to U. However, if A did not participate in the functions L and/or H at all,
the analytical mechanics would be unable to describe the effects of magnetic field
B = ∇ × A on the particle’s motion. The relations (9.183), (9.195) and (9.196) show
the wonderful way in which physics (or Mother Nature herself?) solves this problem:
the vector-potential gives such contributions to the functions L and H (if the latter
is considered, as it should be, a function of P rather than p) that cannot be uniquely
attributed to either kinetic or potential energy, but ensure the correct equation of
motion (9.144) in both the Lagrange and Hamilton formalisms.
I believe I still owe the reader a discussion of the physical sense of the canonical
momentum P. For that, let us consider a particle moving near a region of localized
magnetic field B(r, t), but not entering this region (see figure 9.14), so that on its
trajectory B ≡ ∇ × A = 0. If there is no electrostatic field (no other electric charges
64
Alternatively, this relation may be obtained from the expression for the Lorentz-invariant norm, pαpα =
(mc)2, of the four-momentum (9.75), pα = {E /c , p} = {(H − qϕ)/c, P − qA}.
65
9-46
Figure 9.14. A particle’s motion around a localized magnetic field with a time-dependent flux.
nearby), we may select such a local gauge that ϕ(r, t) = 0 and A = A(t), so that Eq.
(9.144) is reduced to
dp dA
= qE = −q , (9.197)
dt dt
so that Eq. (9.192) immediately gives
dP dp dA
≡ +q = 0. (9.198)
dt dt dt
Hence, even if the magnetic field is changed in time, so that the induced electric field
does accelerate the particle, its conjugate momentum does not change. Hence P is a
variable more stable to magnetic field changes than its kinetic counterpart p. This
conclusion may be criticized because it relies on a specific gauge, and generally P ≡
p + qA is not gauge-invariant, because the vector-potential A is not66. However, as
was already discussed in section 5.3, the integral ∫A · dr over a closed contour does
not depend on the chosen gauge and equals the magnetic flux Φ through the area
limited by the contour—see Eq. (5.65). So, integrating Eq. (9.197) over a closed
trajectory of a particle (figure 9.14), and over the time of one orbit, we obtain
Δ ∮C p ⋅ d r = −qΔΦ, so that Δ ∮C P ⋅ d r = 0, (9.199)
where ΔΦ is the change of flux during that time. This gauge-invariant result confirms
the above conclusion about the stability of the canonical momentum to magnetic
field variations.
Generally, Eq. (9.199) is invalid if a particle moves inside a magnetic field and/or
changes its trajectory at the field variation. However, if the field is almost uniform,
i.e. its gradient small in the sense of Eq. (9.177), this result is (approximately)
applicable. Indeed, analytical mechanics67 tells us that for any canonical coordinate-
momentum pair {qj, pj}, the corresponding action variable,
1
Jj ≡
2π
∮ pj dqj , (9.200)
remains virtually constant at slow variations of motion conditions. According to

Eq. (9.191), for a particle in a magnetic field, the generalized momentum,
66
The kinetic momentum p = Mu is just the usual mu product modified for relativistic effects, so that this
variable is evidently gauge- (although not Lorentz-)invariant.
67
9-47
corresponding to the Cartesian coordinate rj, is Pj rather than pj. Thus forming the
net action variable J ≡ Jx + Jy + Jz, we may write
2πJ = ∮ P ⋅ d r = ∮ p ⋅ d r + qΦ = const. (9.201)
Let us apply this relation to the motion of a non-relativistic particle in an almost

uniform magnetic field, with a small longitudinal velocity, ·u∣∣/u⊥ → 0 (figure 9.15).
Figure 9.15. A particle in a magnetic field with a small longitudinal gradient ∇B∣∣B.
In this case, Φ in Eq. (9.201) is the flux encircled by a cyclotron orbit, equal to
(−πR2B), where R is its radius given by Eq. (9.153) and the negative sign accounts
for the fact that the ‘correct’ direction of the normal vector n in the definition of flux,
Φ = ∫B · n d2r, is antiparallel to the vector B. At u ≪ c, the kinetic momentum is just
p⊥ = mu⊥, while Eq. (9.153) yields
mu⊥ = qBR. (9.202)
Plugging these relations into Eq. (9.201), we obtain
qRB
2πJ = mu⊥2πR − qπR2B = m 2πR − qπR2B ≡ (2 − 1)qπR2B ≡ −qΦ . (9.203)
m
This means that even if the circular orbit slowly moves in the magnetic field, the
flux encircled by the cyclotron orbit should remain virtually constant. One manifes-
tation of this effect is the result already mentioned at the end of section 9.6: if a small
gradient of the magnetic field is perpendicular to the field itself, then the particle
orbit’s drift is perpendicular to ∇B, so that Φ stays constant. Now let us analyze the
case of a small longitudinal gradient, ∇B∣∣B (figure 9.15). If the small initial
longitudinal velocity u∣∣ is directed toward the higher field region, in order to keep
Φ constant, the cyclotron orbit has to gradually shrink. Rewriting Eq. (9.202) as
πR2B Φ
mu⊥ = q =q , (9.204)
πR πR
we see that this reduction of R (at constant Φ) should increase the orbiting speed u⊥.
But since the magnetic field cannot perform any work on the particle, its kinetic energy,
m
E = (u 2 + u ⊥2 ), (9.205)
2
should stay constant, so that the longitudinal velocity u∣∣ has to decrease. Hence
eventually the orbit’s drift has to stop, and then the orbit has to start moving back
9-48
toward the region of lower fields, being essentially repulsed from the high-field
region. This effect is very important, in particular, for plasma confinement systems.
In the simplest of such systems, two coaxial magnetic coils inducing magnetic fields
of the same direction (figure 9.16) naturally form a ‘magnetic bottle’ which traps
charged particles injected, with sufficiently low longitudinal velocities, into the
region between the coils. More complex systems of this type, but working on the
same basic principle, are the most essential components of the persisting large-scale
efforts to achieve controllable nuclear fusion68.
Figure 9.16. A magnetic bottle (schematically).
Returning to the constancy of the magnetic flux encircled by free particles, it

reminds us of the Meissner–Ochsenfeld effect, which was discussed in section 6.4,
and gives a motivation for a brief revisit of the electrodynamics of superconductiv-
ity. As was emphasized in that section, superconductivity is a substantially quantum
phenomenon; nevertheless the classical notion of the conjugate momentum P helps
us understand its theoretical description. Indeed, the general rule of quantization of
physical systems69 is that each canonical pair {qj, pj} of a generalized coordinate and
the corresponding momentum are described by quantum-mechanical operators that
obey the following commutation relation
[qˆj , pˆj ′ ] = i ℏδjj ′. (9.206)
According to Eq. (9.191), for the Cartesian coordinates rj of a particle in the

magnetic field the corresponding generalized momenta are Pj, so that their operators
should obey the following commutation relations:
[rˆj , Pˆj ′] = i ℏδjj ′. (9.207)
In the coordinate representation of quantum mechanics, the canonical operators

of the linear momentum are described by Cartesian components of the vector
operator −iℏ∇. As a result, ignoring the rest energy mc2 (which gives an incon-
sequential phase factor exp{−imc2t/ℏ} in the wave function), we can use Eq. (9.196)
to rewrite the usual non-relativistic Schrödinger equation,
∂ψ
iℏ = Hˆψ , (9.208)
∂t
as follows:
68
For further reading on this technology the reader is referred, for example, to a simple monograph [9], and/or
a graduate-level theoretical treatment in [10].
69
9-49
∂ψ ⎛ pˆ 2 ⎞ ⎡ 1 ⎤
iℏ =⎜ + U ⎟ψ = ⎢ ( −i ℏ∇ − q A)2 + qφ⎥ψ . (9.209)
∂t ⎝ 2m ⎠ ⎣ 2m ⎦
Thus, I believe I have finally delivered on my promise to justify the replacement

(6.50), which was used in sections 6.4 and 6.5 to discuss the electrodynamics of
superconductors, including the Meissner–Ochsenfeld effect70.
9.8 Analytical mechanics of the electromagnetic field

We have just seen that the analytical mechanics of a particle in an electromagnetic
field may be used to obtain some important results. The same is true for the
analytical mechanics of the field as such, and the field–particle system as a whole.
For such a space-distributed system as the field, governed by local dynamics laws
(Maxwell equations), we need to apply analytical mechanics to the local densities l
and h of the Lagrangian and Hamiltonian functions, defined by relations
L= ∫ l d 3r, H= ∫ h d 3r. (9.210)
Let us start, as usual, from the Lagrange formalism. Some clue toward the
possible structure of the Lagrangian density l may be obtained from that of
the description of the particle–field interaction in this formalism, discussed in the
previous section. As we have seen, for the case of a single particle, the interaction is
described by the last two terms of Eq. (9.183):
Lint = −qϕ − qu ⋅ A. (9.211)
It is virtually obvious that if charge q is continuously distributed over some volume, we
may represent L as a volume integral of the following Lagrangian density:
lint = −ρϕ + j ⋅ A ≡ −jα Aα . (9.212)
Notice that the density (in contrast to Lint itself) is Lorentz-invariant. (This is due
to the contraction of the longitudinal coordinate, and hence volume, at the
Lorentz transform.) Hence we may expect the density of the field’s Lagrangian to
be Lorentz-invariant as well. Moreover, in the view of the simple, local structure
of the Maxwell equations (containing only the first spatial and temporal derivatives
of the fields), lfield should be a simple function of the potential’s four-vector and its
four-derivative:
lfield = lfield(Aα , ∂αAβ ) . (9.213)
Also, the density should be selected in such a way that the four-vector analog of the
Lagrangian equations of motion,
70
Eq. (9.209) is also the basis for discussion of numerous other magnetic field phenomena, including the AB
and quantum Hall effects—see, e.g. Part QM sections 3.1 and 3.2.
9-50
∂lfield ∂l
∂α β
− field = 0, (9.214)
∂(∂αA ) ∂Aβ
gave us the correct inhomogeneous Maxwell equation (9.127)71. It is clear that the
field part lfield of the total Lagrangian density l should be a scalar, and a quadratic
form of the field strength, i.e. of F αβ, so that the natural choice is
lfield = const × FαβF αβ , (9.215)
with implied summation over both indices. Indeed, adding to this expression the
interaction Lagrangian (9.212),
l = lfield + lint = const × FαβF αβ − jα Aα , (9.216)
and performing the differentiation (9.214), we see that the relations (9.214) and
(9.215) indeed yield Eqs. (9.127), provided that the constant factor equals (−1/4μ0).72
So, the field’s Lagrangian density is
1 1 ⎛E2 ⎞ ε0 B2
lfield = − FαβF αβ = ⎜ 2 − B 2⎟ ≡ E 2 − ≡ u e − u m, (9.217)
4μ 0 2μ 0 ⎝ c ⎠ 2 2μ 0
where ue is the electric field energy density (1.65), and um is the magnetic field energy
density (5.57). Let me hope the reader agrees that Eq. (9.217) is a wonderful result,
because the Lagrangian function has a structure absolutely similar to the well-
known expression L = T − U of the classical mechanics. So, for the field alone, the
‘potential’ and ‘kinetic’ energies are separable again73.
Now let us explore whether we can calculate a four-form of the field’s
Hamiltonian function H . In the generic analytical mechanics,
∂L
H= ∑ ∂q ̇ qj̇ − L . (9.218)
j j
However, just as for the Lagrangian function, for a field we should find the spatial
density h of the Hamiltonian, defined by the second of Eqs. (9.210), for which a
natural four-form of Eq. (9.218) is
∂l
h αβ = ∂ βAγ − g αβl . (9.219)
∂(∂αAγ )
Calculated for the field alone, i.e. using Eq. (9.217) for lfield , this definition yields
71
Here the implicit summation over index α plays a role similar to the convective derivative (9.188) in
replacing the full derivative over time, in a way that reflects the symmetry of time and space in special
relativity. I do not want to spend more time justifying Eq. (9.214) for reasons that will be clear imminently.
72
In the Gaussian units, the coefficient is (−1/16π).
73
Since the Lagrange equations of motion are homogeneous, the simultaneous change of the signs of T and U
does not change them. Thus, it is not important which of the two energy densities, ue or um, we count as the
potential, and which as the kinetic energy. (Actually, such duality of the two field energy components is typical
for all analytical mechanics, and was discussed already in Part CM section 2.2.)
9-51
αβ
h field = θ αβ − τDαβ , (9.220)
where the tensor

1 ⎛ αγ 1 ⎞
θ αβ ≡ ⎜g FγδF δβ + g αβFγδF γδ⎟ (9.221)
μ0 ⎝ 4 ⎠
is gauge-invariant, while the remaining term,

1 αγ
τDαβ ≡ g Fγδ ∂ δAβ , (9.222)
μ0
is not, so that it cannot correspond to any measurable variables. Fortunately, it is
straightforward to verify that the last tensor may be represented in the form
1
τDαβ = ∂γ(F γαAβ ) (9.223)
μ0
and as a result obeys the following relations,
∂ατDαβ = 0, ∫ τD0β d 3r = 0, (9.224)
so it does not interfere with the conservation properties of the gauge-invariant,

symmetric energy–momentum tensor (also called the symmetric stress tensor) θαβ, to
be discussed below.
Let us use Eq. (9.125) to express the components of the latter tensor for the
electric and magnetic fields. For α = β = 0, we obtain
ε B2
θ 00 = 0 E 2 + = u e + u m ≡ u, (9.225)
2 2μ 0
i.e. the expression for the total energy density u—see Eq. (6.113). The other three
components of the same row/column turn out to be just the Cartesian components of
the Poynting vector (6.114), divided by c:
1 ⎛E ⎞ ⎛E ⎞ Sj
θ j0 = ⎜ × B⎟ = ⎜ × H⎟ ≡ , for j = 1, 2, 3. (9.226)
⎝
μ0 c ⎠ j
⎝ c ⎠ j c
The remaining nine components θjj′ of the tensor, with j′ = 1, 2, 3, are usually
represented as
θ jj ′ = −τ jj(M )
′ , (9.227)
where τ(M) is the so-called Maxwell stress tensor:
9-52
⎛ δjj ′ 2⎞ 1⎛ δjj ′ 2⎞
τ jj(M )
′ = ε0⎜EjEj ′ − E ⎟ + ⎜BjBj ′ − B ⎟, (9.228)
⎝ 2 ⎠ μ0 ⎝ 2 ⎠
so that the whole symmetric energy–momentum tensor may be conveniently
represented in the following symbolic way:
⎛ u ← S / c →⎞
⎜↑ ⎟
⎜ ⎟
θ αβ = ⎜S (M ) ⎟. (9.229)
−τ jj ′
⎜c ⎟
⎜ ⎟
⎝↓ ⎠
The physical meaning of this tensor may be revealed in the following way.
Considering Eq. (9.221) just as the definition of tensor θαβ,74 and using the four-
vector form of the Maxwell equations given by Eqs. (9.127) and (9.129), it is
straightforward to verify an extremely simple result for the four-derivative of the
symmetric tensor:
∂αθ αβ = −F βγjγ . (9.230)
This expression is valid in the presence of electromagnetic field sources, e.g. for any
system of charged particles and the fields they have created. Of these four equations
(for four values of the index β), the temporal one (with β = 0) may be simply
expressed via the energy density (9.225) and the Poynting vector (9.226),
∂u
+ ∇ ⋅ S = −j ⋅ E, (9.231)
∂t
while three spatial equations (with β = j = 1, 2, 3) may be represented in the form
3
∂ Sj ∂
− ∑ ∂r τ jj(M )
′ = − (ρE + j × B) j . (9.232)
∂t c 2 j ′= 1 j′
Integrated over a volume V limited by surface S, with the account of the

divergence theorem, Eq. (9.231) returns us to the Poynting theorem (6.111),
⎛ ⎞
∫V ⎝ ∂∂ut + j ⋅ E⎠d 3r + ∮S Sn d 2r = 0,
⎜ ⎟ (9.233)
while Eq. (9.232) yields75
74
In this way, we are using Eq. (9.219) just as a useful guess, which has led us to the definition of θαβ, and may
leave its strict justification for more in-depth field theory courses.
75
Just like the Poynting theorem Eq. (9.233), Eq. (9.234) may be obtained directly from the Maxwell
equations, without resorting to the four-vector formalism — see, e.g. section 8.2.2 in [11]. However, the
derivation discussed above is preferable, because it shows the wonderful unity between the laws of
conservation of energy and momentum.
9-53
⎡ ⎤ 3
∫V ⎢⎣ ∂∂t cS2 + f⎥ d 3r =
⎦j
∑∮
S
τ jj(M )
′ dAj ′ , with f ≡ ρE + j × B , (9.234)
j ′= 1
where dAj = nj dA = nj d2r is the jth component of the elementary area vector dA =
n dA = nd2r that is normal to volume’s surface, and directed out of the volume—see
figure 9.1776.
Figure 9.17. The force dF exerted on a boundary element dA of the volume V occupied by the field.
Since, according to Eq. (5.10), the vector f is nothing else than the density of
volume-distributed Lorentz forces exerted by the field on the charged particles, we
can use Newton’s second law, in its relativistic form (9.144), to rewrite Eq. (9.234)
for a stationary volume V as
d⎡ ⎤
⎢
dt ⎣
∫V cS2 d 3r + ppart⎥⎦ = F, (9.235)
where ppart is the total mechanical (relativistic) momentum of all particles in the
volume V, and the vector F is defined by its Cartesian components:
3
Fj = ∑∮ τ jj(M )
′ dAj ′ . (9.236)
A
j ′= 1
Relations (9.235) and (9.236) are our main new results. The first of them shows
that the vector
S
g≡ , (9.237)
c2
already discussed in section 6.8 without derivation, may be indeed interpreted as the
density of momentum of the electromagnetic field (per unit volume). This classical
relation is consistent with the quantum-mechanical picture of photons of ultra-
relativistic particles, with the momentum’s magnitude E /c , because then the total
flux of the momentum carried by photons through a unit normal area per unit time
may be represented either as Sn/c or as gnc. It also allows us to revisit the Poynting
vector paradox that was discussed in section 6.8—see figure 6.11 and its discussion.
As has been emphasized at this discussion, the vector S = E × H in this case does not
correspond to any measurable energy flow. However, the corresponding momentum
76
The same notions are used in the mechanical stress theory—see, e.g. Part CM section 7.2.
9-54
of the field, equal to the integral of the density (9.237) over a volume of interest, is
not only real, but may be measured by the recoil impulse77 it gives to the field
sources—say, to a magnetic coil inducing the field H, or to the capacitor plates
creating the field E.
Now let us turn to our second result, Eq. (9.236). It tells us that the 3 × 3-element
Maxwell stress tensor complies with the general definition of the stress tensor78
characterizing the force F exerted by external forces on the boundary of a volume, in
this case occupied by the electromagnetic field (figure 9.17). Let us use this important
result to analyze two simple examples of static fields.
(i) Electrostatic field’s effect on a perfect conductor. Since Eq. (9.235) has been
derived for a free space region, we have to select volume V outside the conductor,
but we may align one of its faces with the conductor’s surface (figure 9.18).
Figure 9.18. The electrostatic field near a conductor’s surface.
From chapter 2, we know that the electrostatic field has to be perpendicular to the
conductor’s surface. Selecting axis z in this direction, we have Ex = Ey = 0, Ez = ±E,
so that only diagonal components of the tensor (9.228) are not equal to zero:
ε ε
(M )
τxx (M )
= τ yy = − 0 E 2, τzz(M ) = 0 E 2, (9.238)
2 2
Since the elementary surface area vector has just one non-vanishing component,
dAz, according to Eq. (9.236), only the last component (which is positive regardless
of the sign of E) gives a contribution to the surface force F. We see that the force
exerted by the conductor (and eventually by external forces that hold the conductor
in its equilibrium position) on the field is normal to the conductor and directed out of
the field volume: dFz ⩾ 0. Hence, by Newton’s third law, the force exerted by the field
on conductor’s surface is directed toward the field-filled space:
ε
dFsurface = −dFz = − 0 E 2 dA . (9.239)
2
This important result could be obtained by simpler means as well. (Actually, this
was the task of one of the problems given in chapter 2.) For example, one could
argue quite convincingly that the local relation between the force and the field
77
This impulse is sometimes called the hidden momentum; this term makes sense if the field sources have finite
masses, so that their velocity change at the field variation is measurable.
78
9-55
should not depend on the global configuration creating the field, and thus consider the
simplest configuration, a planar capacitor (see e.g. figure 2.3) with the surfaces of both
plates charged by equal and opposite charges of density σ = ±ε0E. According to the
Coulomb law, the charges should attract each other, pulling each plate toward the
field region, so that the Maxwell-tensor result gives the correct direction of the force.
The force’s magnitude (9.239) can be verified either by the direct integration of the
Coulomb law, or by the following simple reasoning. In the plane capacitor, the field
Ez = σ/ε0 is equally contributed by two surface charges; hence the field created by the
negative charge of the counterpart plate (not shown in figure 9.18) is E− = σ/2ε0, and
the force it exerts on the elementary surface charge dQ = σdA of the positively charged
plate is dF = E dQ = σ2 dA/2ε0 = ε0E2 dA/2, in accordance with Eq. (9.239)79.
Quantitatively, even for such a high electric field as E = 3 MV m−1 (close to
electric breakdown in air), the ‘negative pressure’ (dF/dA) given by Eq. (9.239) is of
the order of 500 Pa (N m−2), i.e. below one thousandth of the ambient atmospheric
pressure (1 bar ≈ 105 Pa). Still, this negative pressure may be substantial (above 1
bar) in some cases, for example in good dielectrics (such as high-quality SiO2, grown
at high temperature, which is broadly used in integrated circuits) that can withstand
electric fields up to ~109 V m−1.
(ii) Static magnetic field’s effect on its source80—say, a solenoid’s wall or a
superconductor’s surface (figure 9.19).
Figure 9.19. The static magnetic field near a current-carrying surface.
With the choice of coordinates shown in figure 9.19, we have Bx = ±B, By = Bz =

0, so that the Maxwell stress tensor (9.228) is diagonal again:
(M ) 1 2 (M ) 1 2
τxx = B , τ yy = τzz(M ) = − B . (9.240)
2μ 0 2μ 0
However, since for this geometry only dAz differs from 0 in Eq. (9.236), the sign of
the resulting force is opposite to that in electrostatics, dFz ⩽ 0, and the force exerted
by the magnetic field upon the conductor’s surface,
79
By the way, repeating these arguments for a plane capacitor filled with a linear dielectric, we may readily see
that Eq. (9.239) may be generalized for this case by replacing ε0 for ε. The similar replacement (μ0 → μ) is valid
for Eq. (9.241) in a linear magnetic medium.
80
The causal relation is not important here. Especially in the case of a superconductor, the magnetic field may
be induced by another source, with the surface supercurrent j just shielding the superconductor’s bulk from its
penetration—see section 9.6.
9-56
1 2
dFsurface = −dFz = B dA , (9.241)
2μ0
corresponds to a positive pressure. For good laboratory magnets (B ~ 10 T), this
pressure is of the order of 4 × 107 Pa ≈ 400 bars, i.e. is very substantial, so the
magnets require a solid mechanical design.
The direction of the force (9.241) could also be readily predicted by elementary
magnetostatics arguments. Indeed, we can imagine the magnetic field volume limited
by another, parallel wall with the opposite direction of surface current. According to
the starting point of magnetostatics, Eq. (5.1), such surface currents of opposite
directions have to repulse each other—doing that via the magnetic field.
Another explanation of the fundamental sign difference between the electric and
magnetic field pressures may be provided on the electric circuit language. As we
know from chapter 2, the potential energy of the electric field stored in a capacitor
may be represented in two equivalent forms,
CV 2 Q2
Ue = = . (9.242)
2 2C
Similarly, the magnetic field energy in an inductive coil is
LI 2 Φ2
Um = = . (9.243)
2 2L
If we do not want to consider the work of external sources at a virtual change of the
system dimensions, we should use the latter forms of these relations, i.e. consider a
galvanically detached capacitor (Q = const) and an externally shorted inductance
(Φ = const)81. Now if we let the electric field forces (9.239) drag the capacitor’s plates
in the direction they ‘want’, i.e. toward each other, this would lead to a reduction of
the capacitor thickness, and hence to an increase of its capacitance C, and to a
decrease of Ue. Similarly, for a solenoid, allowing the positive pressure (9.241) to
move its walls away from each other would lead to an increase of the solenoid
volume, and hence of its inductance L, so that the potential energy Um would be also
reduced—as it should be. It is remarkable (actually, beautiful) how the local field
formulas (9.239) and (9.241) ‘know’ about these global circumstances.
Finally, let us see whether the major results (9.237) and (9.241), obtained in this
section, match each other. For that, let us return to the normal incidence of a plane,
monochromatic wave from the free space upon the plane surface of a perfect
conductor (see, e.g. figure 7.8 and its discussion), and use those results to calculate
the time average of the pressure dFsurface/dA imposed by the wave on the surface. At
elastic reflection from the conductor’s surface, the electromagnetic field’s momen-
tum retains its amplitude but changes its sign, so that the momentum transferred to a
unit area of the surface (i.e. average pressure) is
81
Of course, this condition may hold ‘forever’ only for solenoids with superconducting wiring, but even in
normal-metal solenoids with practicable inductances, the flux relaxation constants L/R may be rather large
(practically, up to a few minutes), quite sufficient to carry out the force measurement.
9-57
dFsurface Sincident EH EωHω*

= 2cgincident = 2c = 2 c = , (9.244)
dA c2 c2 c
where Eω and Hω are complex amplitudes of the incident wave. Using the relation
(7.7) between these amplitudes (for ε = ε0 and μ = μ0 giving Eω = cBω) we obtain
dFsurface 1 B* Bω 2
= cBω ω = . (9.245)
dA c μ0 μ0
On the other hand, as was discussed in section 7.3, at the surface of a perfect
mirror the electric field vanishes while the magnetic field doubles, so that we can use
Eq. (9.241) with B → B(t) = 2Re[Bωexp{−iωt}]. Averaging the pressure given by
Eq. (9.241) over time, we obtain
dFsurface 1 Bω 2
= (2 Re [Bωe−iωt ])2 = , (9.246)
dA 2μ 0 μ0
i.e. the same result as Eq. (9.245).

For the development of physics intuition, it is useful to estimate the electro-
magnetic radiation pressure’s magnitude. Even for a relatively high wave intensity
Sn of 1 kW m−2 (close to that of the direct sunlight at the Earth’s surface), the
pressure 2cgn = 2Sn/c is somewhat below 10−5 Pa ~ 10−10 bar. Still, this extremely
small effect was experimentally observed (by P Lebedev) as early as in 1899, giving
one more confirmation of the Maxwell’s theory.
9.9 Problems
Problem 9.1. Use the non-relativistic picture of the Doppler effect, in which light
propagates with velocity c in a Sun-bound aether, to derive Eq. (9.4).
Problem 9.2. Show that two successive Lorentz space/time transforms in the same
direction, with velocities u′ and v, are equivalent to a single transform, with the
velocity u given by Eq. (9.25).
Problem 9.3. N + 1 reference frames, numbered by index n (taking values 0, 1, …,

N), move in the same direction as a particle. Express the particle’s velocity in the
frame number 0 via its velocity uN in the frame number N, and the set of velocities vn
of the frame number n relative to the frame number (n − 1).
Problem 9.4. A spaceship, moving with constant velocity v directly from the Earth,
sends back brief flashes of light with a period ΔtS—as measured by the spaceship’s
clock. Calculate the period with which Earth’s observers receive the signals—as
measured by the Earth’s clock.
Problem 9.5. From the point of view of observers in a ‘moving’ reference frame 0′, a
straight thin rod, parallel to axis x′, is moving, without rotation, with a constant
velocity u′ directed along axis y′. The reference frame 0′ is itself moving relative to
9-58
another (‘lab’) reference frame 0 with a constant velocity v along axis x, also without
rotation—see the figure below. Calculate:
(i) the direction of the rod’s velocity, and
(ii) the orientation of the rod on the [x, y] plane,
both as observed from the lab reference frame. Is the velocity, in this frame,
perpendicular to the rod?
Problem 9.6. Starting from the rest at t = 0, a spaceship moves directly from the
Earth with a constant acceleration as measured in its instantaneous rest frame. Find
its displacement x(t) from the Earth as measured from the Earth’s reference frame
and interpret the result.
Hint: The instantaneous rest frame of a moving particle is the inertial reference
frame that, at the considered moment of time, has the same velocity as the particle.
Problem 9.7. Calculate the first relativistic correction to the frequency of a harmonic
oscillator as a function of its amplitude.
Problem 9.8. An atom, with an initial rest mass m, has been excited to an internal
state with an additional energy ΔE , still being at rest. Next, it returns into its initial
state, emitting a photon. Calculate the photon’s frequency, taking into account the
relativistic recoil of the atom.
Hint: In this problem, and in problems 9.11–9.13, treat photons as ultra-
relativistic point particles with zero rest mass, energy E = ℏω, and momentum
p = ℏk.
Problem 9.9. A particle of mass m, initially at rest, decays into two particles, with
rest masses m1 and m2. Calculate the total energy of the first decay product, in the
reference frame moving with that particle.
Problem 9.10. A relativistic particle with a rest mass m, moving with velocity u,
decays into two particles with zero rest mass.
(i) Calculate the smallest possible angle between the decay product velocities (in the
lab frame, in which the velocity u is measured).
(ii) What is the largest possible energy of one product particle?
9-59
Problem 9.11. A relativistic particle, propagating with velocity u, in the absence of

external fields, decays into two photons82. Calculate the angular dependence of the
probability of photon detection, as measured in the lab frame.
Problem 9.12. A photon with wavelength λ is scattered by an electron, initially at

rest. Calculate the wavelength λ′ of the scattered photon as a function of the
scattering angle α—see the figure below83.
Problem 9.13. Calculate the threshold energy of a γ-photon for the reaction
γ + p → p + π0,
if the proton was initially at rest.
Hint: For protons mpc2 ≈ 938 MeV, while for neutral pions mπc2 ≈ 135 MeV.
Problem 9.14. A relativistic particle with energy E and rest mass m collides with a
similar particle, initially at rest in the laboratory reference frame. Calculate:
(i) the final velocity of the center of mass of the system, in the lab frame,
(ii) the total energy of the system, in the center-of-mass frame, and
(iii) the final velocities of both particles (in the lab frame), if they move along the
same direction.
Problem 9.15. A ‘primed’ reference frame moves, with the reduced velocity β ≡ v/c =
nxβ, relative to the ‘lab’ frame. Use Eq. (9.109) to express the components T′00 and
T′0j (with j = 1, 2, 3) of an arbitrary contravariant four-tensor T γδ via its components
in the lab frame.
Problem 9.16. Static fields E and B are uniform but arbitrary (both in magnitude
and in direction). What should be the velocity of an inertial reference frame to have
the vectors E′ and B′, observed from that frame, parallel? Is this solution unique?
Problem 9.17. Two charged particles, moving with the same constant velocity u, are
offset by distance R = {a, b} (see the figure below), as measured in the lab frame.
Calculate the forces between the particles—also in the lab frame.
82
Such a decay may happen, for example, with a neutral pion.
83
This the famous Compton scattering problem.
9-60
Problem 9.18. Each of two thin, long, parallel particle beams of the same velocity u,
separated by distance d, carries electric charge with a constant density λ per unit
length, as measured in the coordinate frame moving with the particles.
(i) Calculate the distribution of the electric and magnetic fields in the system (outside
the beams), as measured in the lab frame.
(ii) Calculate the interaction force between the beams (per particle) and the resulting
acceleration, both in the lab frame and in the system moving with the electrons.
Compare the results and give a brief discussion of the comparison.
Problem 9.19. Spell out the Lorentz transform of the scalar potential and the vector
potential components, and use the result to calculate the potentials of a point charge
q moving with a constant velocity u as measured in the lab reference frame.
Problem 9.20. Calculate the scalar and vector potentials created by a time-
independent electric dipole p, as measured in a reference frame that moves relatively
to the dipole with a constant velocity v, with the shortest distance (‘impact
parameter’) equal to b.
Problem 9.21. Solve the previous problem in the limit v ≪ c for a time-independent
magnetic dipole m.
Problem 9.22. Assuming that the magnetic monopole does exist and has magnetic
charge g, calculate the change ΔΦ of magnetic flux in a superconductor ring due to
the passage of a single monopole through it. Evaluate ΔΦ for the monopole charge
conjectured by P Dirac, g = ng0 ≡ n(2πℏ/e), where n is an integer; compare the result
with the magnetic flux quantum Φ0 (6.62) with ∣q∣ = e, and discuss their relation.
Hint: For simplicity, you may consider the monopole’s passage along the symmetry
axis of a thin, round superconducting ring, in the otherwise free space.
Problem 9.23.* Calculate the trajectory of a relativistic particle in a uniform

electrostatic field E, for arbitrary direction of its initial velocity u(0), using two
different approaches—at least one of them different from the approach used in
section 9.6 for the case u(0) ⊥ E.
Problem 9.24. A charged relativistic particle with velocity u performs

planar cyclotron rotation in a uniform external magnetic field B. How much
9-61
would the velocity and orbit radius change at a slow change of the field to a new
magnitude B′?
Problem 9.25.* Analyze the motion of a relativistic particle in uniform, mutually

perpendicular fields E and B for the particular case when E is exactly equal to cB.
Problem 9.26.* Find the law of motion of a relativistic particle in uniform, parallel,
static fields E and B.
Problem 9.27. Neglecting relativistic effects, calculate the smallest voltage V that
has to be applied between the anode and cathode of a magnetron (see figure 9.13 and
its discussion) to enable electrons to reach the anode, at negligible electron–electron
interactions (including the space-charge effects), and collisions with the residual gas
molecules. You may:
(i) model the cathode and anode as two coaxial round cylinders, of radii R1 and R2,
respectively;
(ii) assume that the magnetic field B is uniform and directed along their common
axis; and
(iii) neglect the initial velocity of the electrons emitted by the cathode.
(After the solution, estimate the validity of the last assumption and of the non-
relativistic approximation for reasonable values of parameters.)
Problem 9.28. A charged, relativistic particle has been injected into a uniform
electric field whose magnitude oscillates in time with frequency ω. Calculate the time
dependence of the particle’s velocity, as observed from the lab reference frame.
Problem 9.29. Analyze the motion of a non-relativistic particle in a region where the
electric and magnetic fields are both uniform and constant in time, but not
necessarily parallel or perpendicular to each other.
Problem 9.30. A static distribution of electric charge in otherwise free space has
created a time-independent distribution E(r) of the electric field. Use two different
approaches to express the energy density u′ and the Poynting vector S′ as observed
from a reference frame moving with constant velocity v, via components of the
vector E. In particular, is S′ equal to (−vu′)?
Problem 9.31. A plane wave, of frequency ω and intensity S, is normally incident on

a perfect mirror, moving with velocity v in the same direction as the wave.
(i) Calculate the reflected wave’s frequency, and
(ii) use the Lorentz transform of the fields to calculate the reflected wave’s intensity,
both as observed from the lab reference frame.
Problem 9.32. Carry out the second task of the previous problem by using the
relations between the wave’s energy, power, and momentum.
9-62
Hint: As a by-product, this approach should also give you the pressure exerted by
the wave on the moving mirror.
Problem 9.33. Consider the simple model of capacitor charging by a lumped current
source, shown in the figure below, and prove that the momentum given by the
constant, uniform external magnetic field B to the current-carrying conductor is
equal and opposite to the momentum of the electromagnetic field that current I(t)
builds up in the capacitor. (You may assume that the capacitor is planar and very
broad, and neglect the fringe field effects.)
Problem 9.34. Consider an electromagnetic plane wave packet propagating in free

space, with the electric field represented as the Fourier integral
+∞
E(r , t ) = Re ∫−∞ Eke iψk dk , with ψk ≡ kz − ωkt , and ωk ≡ c k .
Express the full linear momentum (per unit area of wave’s front) of the packet via
the complex amplitudes Ek of its Fourier components. Does the momentum depend
on time? (In contrast with problem 7.7, in this case the wave packet is not necessarily
narrow.)
Problem 9.35. Calculate the pressure exerted on well-conducting walls of a wave-

guide with a rectangular (a × b) cross-section, by a wave propagating along it in the
fundamental (H10) mode. Give an interpretation of the result.
References
[1] Müller H et al 2003 Phys. Rev. Lett. 91 020401
[2] Benford G et al 1970 Phys. Rev. D 2 263
[3] Arzeliès H 1966 Relativistic Kinematics (Pergamon)
[4] Rindler W 1991 Introduction to Special Relativity 2nd edn (Oxford University Press)
[5] Hafele J and Keating R 1972 Science 177 166
[6] Lee S 2004 Accelerator Physics 2nd edn (Singapore: World Scientific)
[7] Wilson E 2001 An Introduction to Particle Accelerators (Oxford University Press)
[8] Jackson J 1999 Classical Electrodynamics 3rd edn (Wiley)
[9] Chen F 1984 Introduction to Plasma Physics and Controllable Fusion vol 1 2nd edn
(Springer)
[10] Hazeltine R and Meiss J 2003 Plasma Confinement (New York: Dover)
[11] Griffiths D 1999 Introduction to Electrodynamics 3rd edn (Prentice-Hall)
9-63
IOP Publishing
Lecture notes
Chapter 10
Radiation by relativistic charges
The discussion of special relativity in the previous chapter enables us to revisit the
analysis of electromagnetic radiation by charged particles, now for arbitrary speed.
For a single particle, we will be able to calculate the radiated wave fields in an explicit
form, and analyze them for such important specific cases as synchrotron radiation and
‘Bremsstrahlung’ (brake radiation). After that we will discuss the apparently
unrelated effect of the so-called Coulomb losses of energy by a particle moving in
condensed matter, because this discussion will naturally lead us to such important
phenomena as the Cherenkov radiation and transitional radiation. At the end of the
chapter, I will briefly review the effects of back action of the emitted radiation on the
emitting particle, whose analysis reveals some limitations of classical electrodynamics.
10.1 Liénard–Wiechert potentials

A convenient starting point for the discussion of radiation by relativistic charges is
provided by Eqs. (8.17) for the retarded potentials. In the free space, these formulas
(with the integration variable changed from rʹ to rʺ for clarity in what follows) are
reduced to
1 ρ(r″ , t − R / c ) 3
ϕ(r , t ) =
4πε0
∫ R
d r″ ,
(10.1a )
μ j(r″ , t − R / c ) 3
A(r , t ) = 0
4π
∫ R
d r″ , with R ≡ r − r″ .
As a reminder, Eqs. (10.1a) were derived from the Maxwell equations without any
restrictions, and are very natural for situations with continuous distributions of the
electric charge and/or current. However, for a single point charge, with
ρ(r , t ) = qδ(r − r′), j(r , t ) = q uδ(r − r′), with u = r′̇ , (10.1b)

where rʹ is the instantaneous position of the charge, it is more convenient to recast

Eqs. (10.1a) into an explicit form that would not require the integration in each
particular case. Indeed, as Eqs. (10.1) show, the potentials at the observation point
{r, t} are contributed by only one specific point {rʹ(tret), tret} of the particle’s 4D
trajectory (called its world line), which satisfies the following condition:
R ret
tret ≡ t − , (10.2)
c
where tret is called the retarded time, and Rret is the length of the following distance
vector
R ret ≡ r(t ) − r′(tret ) (10.3)
—physically, the distance covered by the electromagnetic wave from its emission to
observation.
The reduction of Eqs. (10.1) to such a simpler form, however, requires care.
Indeed, their naïve integration would yield the following apparent, but incorrect
results:
1 q ϕ(r , t ) μ qc
ϕ(r , t ) = , i.e. = 0 ;
4πε0 R ret c 4π R ret
(10.4)
μ qu
A(r , t ) = 0 ret , (WRONG ! )
4π R ret
where uret is the particle’s velocity at the retarded point rʹ(tret). Eq. (10.4) is a good
example how the theory of relativity (even the special theory) cannot be taken too
lightly. Indeed, the strings (9.84) and (9.85), formed from the apparent potentials
(10.4), would not obey the Lorentz transform rule (9.91), because according to Eqs.
(10.2) and (10.3) the distance Rret also depends on the reference frame it is measured
in.
In order to correct the error, we need, first of all, to discuss the conditions (10.2)
and (10.3). Combining them, we obtain the following equation for tret:
c(t − tret ) = r(t ) − r′(tret ) . (10.5)
Figure 10.1 depicts the graphical solution of this self-consistency equation as the
only1 point of intersection of the light cone of the observation point (see figure 9.9
and its discussion) and the particle’s word line. In Eq. (10.5), just as in Eqs. (10.1)–
(10.3), all variables have to be measured in the same ‘lab’ reference frame, in which
the observation point r rests. Now let us write Eqs. (10.1) for a point charge in
another inertial reference frame 0ʹ, whose velocity (as measured in the lab frame)
1
As figure 10.1 shows, there is always another, ‘advanced’ point {rʹ(tadv), tadv} of the particle’s world line, with
ta > t, which is also a solution of Eq. (10.5), but it does not fit Eq. (10.1), because the observation of the field,
induced at the advanced point, at the point {r, t < tadv} would violate the causality principle.
10-2
Figure 10.1. Graphical solution of Eq. (10.5).
coincides, at the moment tʹ = tret, with the velocity uret of the charge. In that frame
the charge rests, so that, as we know from the electro- and magnetostatics,
q
ϕ′ = , A′ = 0. (10.6a )
4πε0R′
(Remember that this Rʹ may not be equal to R, because the latter distance is
measured in the ‘lab’ reference frame.) Let us use the identity 1/ε0 ≡ μ0c2 to rewrite
Eq. (10.6a) in the form of components of a four-vector similar in structure to
Eq. (10.4):
ϕ′ μ c
= 0q , A′ = 0. (10.6b)
c 4π R′
Now it is easy to guess the correct answer for the whole four-potential:
μ cu α
Aα = 0 q , (10.7)
4π u β R β
where (just as a reminder), Aα ≡ {ϕ/c, A}, uα ≡ γ{c, u},and Rα is a four-vector of the
inter-event distance, formed similarly to that of a single event—cf Eq. (9.48):
R α ≡ {c(t − t′), R′} ≡ {c(t − t′), r − r′}. (10.8)
Indeed, we needed the four-vector Aα that would:
(i) obey the Lorentz transform,

(ii) have its spatial components Aj scaling at low velocity as uj, and
(iii) be reduced to the correct result (10.6) in the reference frame moving with the
charge.
Formula (10.7) evidently satisfies all these requirements, because the scalar product
in its denominator is just
uβRβ = γ{c , −u} ⋅ {c(t − t′), R} ≡ γ [c 2(t − t′) − u ⋅ R]
(10.9)
≡ γc(R − β ⋅ R) ≡ γcR(1 − β ⋅ n),
10-3
where n ≡ R/R is a unit vector in the observer’s direction, β ≡ u/c is the normalized
velocity of the particle, and γ ≡ 1/(1 − u2/c2)1/2. In the reference frame of the charge
(in which β = 0 and γ = 1), the expression (10.9) is reduced to cR, so that Eq. (10.7) is
correctly reduced to Eq. (10.6b). Now let us spell out components of Eq. (10.7) in the
lab frame (in which tʹ = tret and R = Rret):
1 q 1 ⎡ 1 ⎤
ϕ(r , t ) = = q⎢ ⎥ , (10.10a )
4πε0 (R − β ⋅ R)ret 4πε0 ⎣ R(1 − β ⋅ n) ⎦ret
μ0 ⎛ u ⎞ μ ⎡ β ⎤ uret
A(r , t ) = q⎜ ⎟ = 0 qc⎢ ⎥ ≡ ϕ(r , t ) 2 . (10.10b)
4π ⎝ R − β ⋅ R ⎠ret 4π ⎣ R(1 − β ⋅ n) ⎦ret c
These formulas are called the Liénard–Wiechert potentials2. In the non-relativistic

limit, they coincide with the naïve guess (10.4), but in the general case include an
additional factor 1/(1 − β · n). The explanation of its physical origin may be
facilitated by one more formal calculation—which we will need anyway. Let us
differentiate the geometric relation (10.5), rewritten as
R ret = c(t − tret ), (10.11)
over tret and then, independently, over t, assuming that r is fixed. For that let us first
differentiate, over tret, both sides of the identity Rret2 = Rret · Rret:
∂R ret ∂R ret
2R ret = 2R ret ⋅ . (10.12)
∂tret ∂tret
If r is fixed, ∂Rret/∂tret ≡ ∂(r − rʹ)/∂tret = −∂rʹ/∂tret ≡ −u, and Eq. (10.12) yields
∂R ret R ∂R ret
= ret ⋅ = −(n ⋅ u)ret . (10.13)
∂tret R ret ∂tret
Now let us differentiate the same Rret over t. On one hand, Eq. (10.11) yields
∂R ret ∂t
= c − c ret . (10.14)
∂t ∂t
On the other hand, according to Eq. (10.5), at the partial differentiation over time,
i.e. if r is fixed, tret is a function of t alone so that (using Eq. (10.13) at the second
step) we may write
∂R ret ∂R ret ∂tret ∂t
= = −(n ⋅ u)ret ret . (10.15)
∂tret ∂tret ∂t ∂t
Now requiring Eqs. (10.14) and (10.15) to give the same result, we obtain3:
2
They were derived in 1898 by A-M Liénard and (apparently, independently) in 1900 by E Wiechert.
3
This relation may be used for an alternative derivation of Eq. (10.10) directly from Eqs. (10.1)—an exercise
highly recommended to the reader.
10-4
∂tret c ⎛ 1 ⎞
= =⎜ ⎟ . (10.16)
∂t c − (n ⋅ u)ret ⎝ 1 − β ⋅ n ⎠ret
This relation may be readily re-derived (and more clearly understood) for the simple
particular case when the charge’s velocity is directed straight toward the observation
point. In this case its vector u resides in the same space–time plane as the observation
point’s world line r = const—say, plane [x, t], shown in figure 10.2.
Figure 10.2. Deriving Eq. (10.16) for the case β · n = β.
Let us consider an elementary time interval dtret ≡ dtʹ, during which the particle
would travel the space interval dxret = uretdtret. In figure 10.2, the corresponding
segment of its world line is shown with a solid vector. The dotted vectors in this
figure show the world lines of the radiation emitted by the particle at the beginning
and end of this interval, and propagating with the speed of light c. As follows from
the drawing, the time interval dt between the instants of arrival of the radiation from
these two points to any time-independent spatial point of observation is
dxret u
dt = dtret − = dtret − ret dtret,
c c
(10.17)
dtret 1 1
so that = ≡ .
dt 1 − u ret / c 1 − βret
This expression coincides with (10.16), because in our particular case the directions
of the vectors β ≡ u/c and n ≡ R/R (both taken at tret) coincide, and hence (β · n)ret =
βret. Now the general Eq. (10.16) may be interpreted by saying that the particle’s
velocity in the transverse directions (normal to the vector n) is not important for this
kinematic effect4—a fact almost evident from figure 10.1.
So, the additional factor in the Liénard–Wiechert potentials is just the derivative
∂tret/∂t. The reason for its appearance in Eq. (10.10) is usually interpreted along the
following lines. Let the charge q be spread along the direction of vector Rret (in figure
10.2, along axis x) by an infinitesimal speed-independent interval δxret, so that the
linear density λ of its charge is proportional to 1/δxret. Then the time rate of the
4
Note that this effect (linear in β) has nothing to do with the Lorentz time dilation (9.21), which is quadratic in
β. (Indeed, all our arguments above referred to the same, lab frame.) Rather, it is close in nature to the Doppler
effect.
10-5
charge’s arrival at some spatial point is λuret = λdxret/dtret, i.e. scales as 1/dtret.
However, the rate of the radiation’s arrival at the observation point scales as 1/dt, so
that due to the non-vanishing velocity uret of the particle, this rate differs from the
charge arrival rate by the factor of dtret/dt, given by Eq. (10.16). (If the particle
moves toward the observation point, (β · n)ret > 0, as shown in figure 10.2, this factor
is larger than 1.) This radiation compression effect leads to the field change (at
(β · n)ret > 0, its enhancement) by the same factor (10.16)—as described by Eqs. (10.10).
So, the four-vector formalism was instrumental in the calculation of field
potentials. It may be also used to calculate the fields E and B—by plugging
Eq. (10.7) into Eq. (9.124) to calculate the field strength tensor. This calculation
yields
μ 0 q 1 d ⎡ R αu β − Rβ u α ⎤
F αβ = ⎢ ⎥. (10.18)
4π u γR γ dτ ⎣ u δR δ ⎦
Now using Eq. (9.125) to identify the elements of this tensor with the field
components, we may bring the result to the following vector form5:
q ⎡ n−β n × { (n − β) × β̇} ⎤
E= ⎢ 2 + ⎥ . (10.19)
4πε0 ⎣ γ (1 − β ⋅ n)3R2 (1 − β ⋅ n)3cR ⎦ret
nret × E nret × E
B= , i.e. H = . (10.20)
c Z0
Thus the magnetic and electric fields of a relativistic particle are always
proportional and perpendicular to each other, and related just as in a plane
wave—cf Eq. (7.6), with the only difference that now the vector nret may be a
function of time. Superficially, this result contradicts the electro- and magneto-
statics, because for a particle at rest B should vanish while E stays finite. However,
note that according to the Coulomb law for a point charge, in this case E = Enret,
so that B ∝ nret × E ∝ nret × nret = 0. (Actually, in these relations, the index ‘ret’ is
unnecessary.)
As a sanity check, let us use Eq. (10.19) as an alternative way to find the electric
field of a charge moving without acceleration, i.e. uniformly, along a straight line—
see figure 9.11a reproduced, with minor changes, in figure 10.3. (This calculation will
also illustrate very well the technical challenges of practical applications of the
Liénard–Wiechert formulas for even simple cases.) In this case, the vector β does not
change in time, so that the second term in Eq. (10.19) vanishes and all we need to do
is to spell out the Cartesian components of the first term.
5
An alternative way of deriving these formulas (highly recommended to the reader as an exercise) is to plug
Eqs. (10.10) into the general relations (9.121), and carry out the required temporal and spatial differentiations
directly, using Eq. (10.16) and its spatial counterpart (which may be derived absolutely similarly):
⎡ n ⎤
∇tret = −⎢ ⎥ .
⎣ c(1 − β ⋅ n) ⎦ret
10-6
Figure 10.3. The linearly moving charge problem.
Let us select the coordinate axes and time origin in the way shown in figure 10.3, and
make a clear distinction between the actual position, rʹ(t) = {ut, 0, 0} of the charged
particle at the instant t we are considering, and its position rʹ(tret) at the retarded instant
defined by Eq. (10.5), i.e. the moment when the particle’s field, propagating with the
speed of light, had to be radiated to reach the observation point r at the time t. In these
coordinates
β = {β, 0, 0}, r = {0, b , 0}, r′(tret ) = {utret, 0, 0},
(10.21)
nret = {cos θ , sin θ , 0},
with cos θ = −utret/Rret, so that [(n − β)x]ret = −utret/Rret − β, and Eq. (10.19) yields,
in particular,
q −utret / R ret − β q −utret − βR ret
Ex = 2 3 2
≡ . (10.22)
4πε0 γ [(1 − β ⋅ n) R ]ret 4πε0 γ [(1 − β ⋅ n)3R3]ret
2
But according to Eq. (10.5), the product βRret may be represented as βc(t − tret) ≡
u(t − tret). Plugging this expression into Eq. (10.22) we may eliminate the explicit
dependence of Ex on time tret:
q −ut
Ex = . (10.23)
4πε0 γ [(1 − β ⋅ n)R ]3ret
2
The non-vanishing transverse component of the field also has a similar form:
q ⎡ sin θ ⎤ q b
Ey = ⎢ 2 ⎥ = , (10.24)
4πε0 ⎣ γ (1 − β ⋅ n) R ⎦ret 4πε0 γ 2[(1 − β ⋅ n)R ]3ret
3 2
while Ez = 0. From figure 10.3, β · nret = βcos θ = −βutret/Rret, so that (1 − β · n)Rret ≡

Rret + βutret, and we may again use Eq. (10.5) to obtain (1 − β · n)Rret = c(t − tret) +
βutret ≡ ct − ctret/γ2. What remains is to calculate tret from the self-consistency
equation (10.5), whose square in our current case (figure 10.3) takes the form
2
R ret ≡ b 2 + (utret )2 = c 2(t − tret )2 . (10.25)
This is a simple quadratic equation for tret, which (with the appropriate negative sign
before the square root, in order to obtain tret < t) yields
1/2 γ
tret = γ 2t − [(γ 2t )2 − γ 2(t 2 − b 2 / c 2 )] = γ 2t − (u 2γ 2t 2 + b 2 )1/2 , (10.26)
c
so that the only retarded-function combination that participates in Eqs. (10.23) and
(10.24) is
10-7
c 2 2 2 1/2
[(1 − β ⋅ n)R ]ret = 2
(u γ t + b 2 ) (10.27)
γ
and, finally, the electric field components are
q γut q γb
Ex = − , Ey = , Ez = 0. (10.28)
4πε0 (b + γ 2u 2t 2 )3/2
2 4πε0 (b + γ 2u 2t 2 )3/2
2
These are exactly Eqs. (9.139)6, which had been obtained in section 9.5 by much
simpler means, without the necessity to solve the self-consistency equation (10.5).
However, that alternative approach was essentially based on the inertial motion of
the particle, and cannot be used in problems in which the particle moves with
acceleration. In those problems, the second term in Eq. (10.19), dropping with
distance as 1/Rret and hence describing the wave radiation, is essential and most
important.
10.2 Radiation power

Let us calculate the angular distribution of a particle’s radiation. For that, we need
to return to Eqs. (10.19) and (10.20) to find the Poynting vector S = E × H, and in
particular its component Sn = S · nret, at large distances R from the particle.
Following tradition7, let us express the result as the energy radiated into unit solid
angle per unit time interval dtrad of the radiation, rather than that (dt) of its
measurement. (We will need to return to the measurement time t in the next section,
in order to calculate the observed radiation spectrum.) Using Eq. (10.16), we obtain
dP dE ∂t
≡− = (R2Sn)ret = (E × H) ⋅ [R2 n (1 − β ⋅ n)]ret . (10.29)
dΩ d Ω dtret ∂tret
At sufficiently large distances from the particle, i.e. in the limit Rret → ∞ (in the
radiation zone), the contribution of the first (essentially, the Coulomb-field) term in
the square brackets of Eq. (10.19) vanishes as 1/R2, and the substitution of the
remaining term into Eqs. (10.20) and then (10.29) yields the following formula, valid
for an arbitrary law of particle motion8:
2
dP Z q 2 n × [(n − β) × β̇]
= 0 2 . (10.30)
dΩ (4π ) (1 − n ⋅ β)5
Now, let us apply this important result to some simple cases. First of all,
Eq. (10.30) says that a charge moving with a constant velocity β does not radiate
6
A similar calculation of magnetic field components from Eq. (10.20) gives results identical to Eq. (9.140).
7
This tradition may be reasonably justified. Indeed, we may say that the radiation field ‘detaches’ from the
particle at times close to tret, while the observation time t depends on the detector’s position, and hence is less
relevant for the radiation process as such.
8
If the direction of radiation, n, does not change in time, this formula does not depend on the observer’s
position R. Hence, from this point on, the index ‘ret’ may be safely dropped for brevity, though we should
always remember that β in Eq. (10.30) is the reduced velocity of the particle at the instant of the radiation’s
emission, not its observation.
10-8
at all. This might be expected from our analysis of this case in section 9.5, because in
the reference frame moving with the charge it produces only the Coulomb electro-
static field, i.e. no radiation.
Next, let us consider a linear motion of a point charge with a non-vanishing
acceleration — evidently directed along the straight line of motion of the coordinate
axes directed as shown in figure 10.4a. Each of the vectors involved in Eq. (10.30)
has at most two non-vanishing Cartesian components,
n = {sin θ , 0, cos θ}, β = {0, 0, β}, β̇ = {0, 0, β }, ̇ (10.31)
where θ is the angle between the directions of the particle’s motion and radiation
propagation. Plugging these expressions into Eq. (10.30) and performing the vector
multiplications, we obtain
dP Z q2 sin2 θ
= 0 2 β 2̇ . (10.32)
dΩ (4π ) (1 − β cos θ )5
Figure 10.4b shows the angular distribution of this radiation, for three values of the
particle’s speed u.
Figure 10.4. Particle’s radiation at linear acceleration: (a) the problem’s geometry, and (b) the last fraction of
Eq. (10.32) as a function of the angle θ.
If the speed is relatively low (u ≪ c, i.e. β ≪ 1), the denominator in Eq. (10.32) is
very close to 1 for all observation angles θ, so that the angular distribution of the
radiation power is close to sin2 θ—just as follows from the general non-relativistic
Larmor formula (8.27). However, as the velocity is increased, the denominator
becomes less than 1 for θ < π/2, i.e. for the forward-looking directions, and larger
than 1 for backward directions. As a result, the radiation toward the particle’s
velocity is increased (somewhat counter-intuitively, regardless of the acceleration’s
sign!), while that in the back direction is suppressed. For ultra-relativistic particles (β
→ 1) this trend is enormously exacerbated, and radiation to very small forward
angles dominates. To describe this main part of the angular distribution we may
expand the trigonometric functions of θ participating in Eq. (10.32) into the Taylor
series in small θ, and keep only their leading terms: sin θ ≈ θ, cos θ ≈ 1 − θ 2/2, so that
(1 − βcos θ) ≈ (1 + γ2θ 2)/2γ2. The resulting expression,
dP 2Z0q 2 2̇ 8 (γθ )2
≈ β γ , for γ ≫ 1, (10.33)
dΩ π2 (1 + γ 2θ 2 )5
10-9
describes a narrow ‘hollow cone’ distribution of radiation, with a maximum at the

angle
1
θ0 = ≪ 1. (10.34)
2γ
Another important aspect of Eq. (10.33) is how fast (as γ8) the radiation density
grows with the Lorentz factor γ, i.e. with the particle’s energy E = γmc 2.
Still, the total radiated power P (into all observation angles) at linear acceleration
is not too high for any practicable values of parameters. In order to show this, let us
calculate P for an arbitrary motion of the particle. First, let me demonstrate how P
may be found (or rather guessed) from the general relativistic arguments. In section
8.2, we have derived Eq. (8.27) for the power of the electric dipole radiation for non-
relativistic particle motion. That result is valid, in particular, for one charged
particle whose electric dipole moment’s derivative over time may be expressed as
d(qr)/dt = (q/m)p, where p is the particle’s mechanical momentum (not its electric
dipole moment). As the result, the Larmor formula (8.27) in free space, i.e. with v = c,
reduces to
Z0 ⎛ q dp ⎞2 Z0q 2 ⎛ d p d p ⎞
P= ⎜ ⎟ ≡ ⎜ ⋅ ⎟. (10.35)
6πc 2 ⎝ m dt ⎠ 6πm 2 c 2 ⎝ dt dt ⎠
This is evidently not a Lorentz-invariant result, but it gives a clear hint as to how
such an invariant, that would be reduced to Eq. (10.35) in the non-relativistic limit,
may be formed:
Z0q 2 ⎛ dpα dp α ⎞ Z0q 2 ⎡⎛ d p ⎞2 1 ⎛ dE ⎞2 ⎤

P=− ⎜ ⋅ ⎟ ≡ ⎢⎜ ⎟ − ⎜ ⎟ ⎥. (10.36)
6πm 2 c 2 ⎝ dτ dτ ⎠ 6πm 2 c 2 ⎣⎝ dτ ⎠ c 2 ⎝ dτ ⎠ ⎦
Plugging in the relativistic expressions, p = γmcβ, E = γmc2, and dτ = dt/γ, the last
formula may be recast into the so-called Liénard extension of the Larmor formula9:
Z0q 2 6 ̇ 2 Z q2
P= γ [(β) − (β × β̇)2 ] ≡ 0 γ 4[(β̇)2 + γ 2(β ⋅ β̇)2 ], (10.37)
6π 6π
which may be also obtained by a direct integration of Eq. (10.30) over the full solid
angle, thus confirming our guess.
However, for some applications, it is beneficial to express P the via the time
evolution of particle’s momentum alone. For that, we may differentiate the
fundamental relativistic relation (9.78), E 2 = (mc 2 )2 + (pc )2 , over the proper time
τ to obtain
dE dp dE c 2p dp dp
2E = 2c 2p , i.e. = =u , (10.38)
dτ dτ dτ E dτ dτ
9
The second form of Eq. (10.37), which is frequently more convenient for applications, may be readily
obtained from the first one by applying Eq. (A.49a) to the vector product.
10-10
where, at the last step, the magnitude of the relativistic vector relation c2p/E = u,
mentioned in section 9.3, has been used. Plugging Eq. (10.38) into Eq. (10.36), we
may rewrite it as
Z0q 2 ⎡⎛ d p ⎞2 ⎛ ⎞2 ⎤
2 ⎜ dp ⎟ ⎥
P= ⎢⎜ ⎟ − β . (10.39)
6πm 2 c 2 ⎣⎝ dτ ⎠ ⎝ dτ ⎠ ⎦
Note the difference between the squared derivatives in this expression: in the first of
them we have to differentiate the momentum vector p, and only then form a scalar
by squaring the resulting vector derivative, while in the second case, only the
magnitude of the vector has to be differentiated. For example, for a circular motion
with a constant speed (to be analyzed in detail in the next section), the second term is
zero, while the first one is not.
However, if we return to the simplest case of linear acceleration (figure 10.4), then
(dp/dτ)2 = (dp/dτ)2, and Eq. (10.39) is reduced to
Z0q 2 ⎛ dp ⎞2 Z0q 2 ⎛ dp ⎞2 1 Z0q 2 ⎛ dp ⎞

2
P= ⎜ ⎟ ( 1 − β 2
) ≡ ⎜ ⎟ ≡ ⎜ ⎟ , (10.40)
6πm 2 c 2 ⎝ dτ ⎠ 6πm 2 c 2 ⎝ dτ ⎠ γ 2 6πm 2 c 2 ⎝ dtret ⎠
i.e. formally coincides with the non-relativistic relation (10.35). In order to obtain a
better feeling of the magnitude of this radiation, we may use the fact that dp/dtret =
dE /dzʹ, where zʹ is the particle’s coordinate at moment tret.10 It allows us to rewrite
Eq. (10.40) in the following form:
Z0q 2 ⎛ dE ⎞2 Z0q 2 dE dE dtret Z0q 2 dE dE

P= ⎜ ⎟ ≡ ≡ . (10.41)
6πm 2 c 2 ⎝ dz ⎠ 6πm 2 c 2 dz′ dtret dz′ 6πm 2 c 2u dz′ dtret
For the most important case of ultra-relativistic motion (u → c), this result reduces to
P 2 d (E / mc 2 )
≈ , (10.42)
dE / dtret 3 d (z′ / rc )
where rc is the classical radius of the particle, defined by Eq. (8.41). This formula
shows that the radiated power, i.e. the change of particle’s energy due to radiation, is
much smaller than that due to the accelerating field, unless an energy as large as
∼mc2 is gained on the classical radius of the particle. For example, for an electron,
with rc ≈ 3 × 10−15 m and mc2 ≈ 0.5 MeV, such an acceleration would require an
accelerating electric field of the order of (0.5 MV)/(3 × 10−15 m) ∼ 1014 MV m−1,
while practicable accelerating fields are below 103 MV m−1—limited by the electric
breakdown effects. (As described by the factor m2 in the denominator of Eq. (10.41),
for heavier particles such as protons, the relative losses are even lower.) Such
negligible radiative losses of energy are actually a large advantage of linear
accelerators—such as the famous 2 mile long SLAC11, which can accelerate
10
This relation follows, for example, from comparison of Eq. (9.144) with B = 0, and Eq. (9.148) with E ∣∣ u.
11
See, e.g. https://www6.slac.stanford.edu/.
10-11
electrons or positrons to energies up to 50 GeV, i.e. to γ ≈ 105. If obtaining radiation

from the accelerated particles is the goal, it may be readily achieved by bending their
trajectories using additional magnetic fields—see the next section.
10.3 Synchrotron radiation

Consider a charged particle being accelerated in the direction perpendicular to its
velocity u (for example by the magnetic component of the Lorentz force), so that its
speed u, and hence the magnitude p of its momentum, do not change. In this case,
the second term in the square brackets of Eq. (10.39) vanishes, and it yields
Z0q 2 ⎛ d p ⎞2 Z0q 2 ⎛ d p ⎞ 2
2
P= ⎜ ⎟ = ⎜ ⎟γ . (10.43)
6πm 2 c 2 ⎝ dτ ⎠ 6πm 2 c 2 ⎝ dtret ⎠
Comparing this expression with Eq. (10.40), we see that for the same acceleration
magnitude, the electromagnetic radiation is a factor of γ2 larger. For modern
accelerators, with γ ∼ 104–105, such a factor creates an enormous difference. For
example, if a particle is on a cyclotron orbit in a constant magnetic field (as was
analyzed in section 9.6), both u and p = γmu obey Eq. (9.150), so that
dp u mc 2
= ωcp = p = β 2γ , (10.44)
dtret R R
(where R is the orbit’s radius), so that for the power of this synchrotron radiation,
Eq. (10.43) yields
Z0q 2 4 4 c 2
P= β γ 2. (10.45)
6π R
According to Eq. (9.153), at a fixed magnetic field (in particle accelerators,

limited to a few tesla produced by the beam-bending magnets), the synchrotron
orbit radius R scales as γ, so that according to Eq. (10.45), P scales as γ2, i.e.
grows as the square of the particle’s energy E ∝ γ. For example, for typical
parameters of the first electron cyclotrons (such as the General Electric’s machine
in which the synchrotron radiation was first noticed in 1947), R ∼ 1 m, E ∼ 0.3
GeV (γ ∼ 600), Eq. (10.45) gives a very modest electron energy loss per one
revolution: PT ≡ P (2πP R/u) ≈ 2πR/c ∼ 1 keV. However, already by the mid-
1970 s, electron accelerators, with R ∼ 100 m, could give each particle an energy
E ∼10 GeV, and the energy loss per revolution grew to ∼10 MeV, becoming the
major energy loss mechanism. For proton accelerators, such energy loss is much
less of a problem, because γ of an ultra-relativistic particle (at fixed E ) is
proportional to 1/m, so that the estimates at the same R should be scaled back
by (mp/me)4 ∼ 1013. Nevertheless, in the giant modern accelerators such as the already
mentioned LHC (with R ≈ 4.3 km and E up to 7 TeV), the synchrotron radiation loss
per revolution is rather noticeable (P Δtret ∼ 6 keV), leading not as much to particle
10-12
deceleration as to a substantial photoelectron emission from the beam tube’s walls,

creating harmful defocusing effects.
However, what is bad for particle accelerators and storage rings is good for the
so-called synchrotron light sources—the electron accelerators designed specifically
for the generation of intensive synchrotron radiation—with a spectrum extending
well beyond the visible light range. Let us now analyze the angular and spectral
distributions of such radiation. To calculate the angular distribution, let us select the
coordinate axes as shown in figure 10.5, with the origin at the current location of the
orbiting particle, axis z directed along its instant velocity (i.e. the vector β), and axis
x toward the orbit center.
Figure 10.5. The synchrotron radiation problem’s geometry.
In the general case, the unit vector n toward the radiation’s observer is not within
any of the coordinate planes, and hence should be described by two angles—the
polar angle θ, and the azimuthal angle φ between the x axis and the projection OP of
the vector n to the plane [x, y]. Since the length of the segment OP is sin θ, the
Cartesian components of the relevant vectors are as follows:
n = {sin θ cos ϕ , sin θ sin ϕ , cos θ},
(10.46)
β = {0, 0, β}, and β̇ = {β ,̇ 0, 0} .
Plugging these expressions into the general Eq. (10.30), we obtain
dP 2Z0q 2 ̇ 2 6
= β γ f (θ , ϕ),
dΩ π2
⎡ (10.47)
1 sin2 θ cos2 ϕ ⎤
with f (θ , ϕ ) ≡ 6 ⎢1 − 2 ⎥.
8γ (1 − β cos θ ) ⎣
3
γ (1 − β cos θ )2 ⎦
According to this result, just as at the linear acceleration, in the ultra-relativistic
limit most radiation goes into a narrow cone (of a width Δθ ∼ γ−1 ≪ 1) around the
vector β, i.e. around the instant direction of particle’s propagation. For such small
angles and γ ≫ 1:
1 ⎡ 4γ 2θ 2 cos2 ϕ ⎤
f (θ , ϕ ) ≈ ⎢1 − ⎥. (10.48)
(1 + γ 2θ 2 )3 ⎣ (1 + γ 2θ 2 )2 ⎦
The left-hand panel of figure 10.6 shows a color-coded contour map of the angular
distribution f(θ, φ) of the radiation, as observed on a distant plane normal to the
particle’s instant velocity (in figure 10.5, parallel to the plane [x, y]), while the right-
10-13
Figure 10.6. The angular distribution of the synchrotron radiation at γ ≫ 1.
hand panel shows it as a function of θ in two perpendicular directions: within the

particle rotation plane (in the direction parallel along axis x, i.e. at φ = 0) and
perpendicular to this plane (along axis y, i.e. at φ = ±π/2). The result shows, first of all,
that, in contrast to the case of linear acceleration, the narrow radiation cone is now
not hollow: the intensity maximum is reached at θ = 0, i.e. exactly in the direction of
the particle’s motion. Second, the radiation cone is not axially symmetric: within the
particle rotation plane, the intensity drops faster (and even has nodes at θ = ±1/γ).
As figure 10.5 shows, the calculated angular distribution (10.47) of the synchrotron
radiation is in the (inertial) reference frame whose origin coincides with the particle’s
position at this a particular instant, i.e. its radiation pattern is time-independent only
in a frame moving with the particle. This pattern enables a semi-qualitative
description of the radiation by an ultra-relativistic particle from the point of view
of a stationary observer: if the observation point is on (or close to) the rotation
plane12, it is being ‘struck’ by the narrow radiation cone once each rotation period T
≈ 2πR/c, each ‘strike’ giving a pulse of a short duration Δtret ≪ 1/ωc—see figure 10.713.
The evaluation of the time duration Δt of each pulse requires some care: its
estimate Δtret ∼ 1/γωc is correct for the duration of the time during which its cone is
aimed at the observer. However, due to the time compression effect, discussed in
detail in section 10.1 and described by Eq. (10.12), the pulse duration as seen by the
observer is a factor of 1/(1 − β) shorter, so that
1−β 1
Δt = (1 − β )Δtret ∼ ∼ 3 ∼ γ −3T , for γ ≫ 1. (10.49)
γωc γ ωc
From the Fourier theorem, we can expect the frequency spectrum of such radiation
to consist of numerous (N ∼ γ3 ≫ 1) harmonics of the particle rotation frequency ωc,
12
It is easy (and hence is left as a reader’s exercise) to show that if the observation point is greatly off-plane
(say, is located on the particle orbit’s axis), the radiation is virtually monochromatic, with frequency ωc. (As we
know from section 8.2, in the non-relativistic limit u ≪ c this is true for any observation point.)
13
The fact that the in-plane component of each electric field’s pulse E(t) is asymmetric with respect to its
central point, and hence vanishes at that point (as figure 10.7b shows), readily follows from Eq. (10.19).
10-14
Figure 10.7. Schematic representations of (a) the synchrotron radiation cones (at γ ≫ 1) for two close values of
tret , and (b) the in-plane component of the electric field, observed in the rotation plane, as a function of time t.
with comparable amplitudes. However, if the orbital frequency fluctuates even

slightly (δωc/ωc > 1/N ∼ 1/γ3), as happens in most practical systems, the radiation
pulses are not coherent, so that the average radiation power spectrum may be
calculated as that of one pulse multiplied by the number of pulses per second. In this
case, the spectrum is continuous, extending from low frequencies all the way to
approximately
ωmax ∼ 1/ Δt ∼ γ 3ωc . (10.50)
In order to verify and quantify this result, let us calculate the spectrum of
radiation due to a single pulse. For that, we should first make the general notion of
the radiation spectrum quantitative. Let us represent an arbitrary electric field (say
that of the synchrotron radiation we are studying now), observed at a fixed point r as
a function of the observation time t, as a Fourier integral14:
+∞
E(t ) = ∫−∞ Eωe−iωt dt . (10.51)
This expression may be plugged into the formula for the total energy of the radiation
pulse (i.e. of the loss of particle’s energy E ) per unit solid angle15:
dE +∞
R2 +∞
−
dΩ
≡ ∫−∞ Sn(t )R2 dt =
Z0
∫−∞ E(t ) 2 dt . (10.52)
This substitution, plus a natural change of the integration order, yield

+∞ +∞ +∞
dE R2
−
dΩ
≡
Z0
∫ dω ∫ dω′Eω ⋅ Eω′ ∫ dte−i (ω+ω′)t . (10.53)
−∞ −∞ −∞
14
In contrast to the single-frequency case (i.e. a monochromatic wave), we may avoid taking the real part of
the complex function (Eωe−iωt) by requiring that in Eq. (10.51), E−ω = Eω*. However, it is important to
remember the factor ½ required for the transition to a monochromatic wave of frequency ω0 and real
amplitude E0: Eω = E0 [δ(ω − ω0) + δ(ω + ω0)]/2.
15
Note that the expression under the integral differs from dP /dΩ defined by Eq. (10.29) by the absence of term
(1 − β · n) = ∂tret/∂t—see Eq. (10.16). This is natural, because this is the wave energy arriving at the observation
point r during time interval dt rather than dtret.
10-15
But the inner integral (over t) is just 2πδ(ω + ωʹ ).16 This delta-function kills one of
the frequency integrals (say, one over ωʹ) and Eq. (10.53) gives us a result that may
be recast as
dE +∞
4πR2 4πR2
−
dΩ
= ∫0 I ( ω ) dω , with I (ω) ≡
Z0
Eω ⋅ E−ω ≡
Z0
EωE*ω, (10.54)
where the evident frequency symmetry of the scalar product Eω · E−ω has been
utilized to fold the integral of I(ω) to positive frequencies only. The first of
Eqs. (10.54) makes the physical sense of the function I(ω) clear: this is the so-called
spectral density of the electromagnetic radiation (per unit solid angle)17.
In order to calculate the spectral density, we need to express the function Eω via
E(t) using the Fourier transform reciprocal to Eq. (10.51):
+∞
1
Eω =
2π
∫−∞ E(t )e iωt dt . (10.55)
In the particular case of radiation by a single point charge, we can use the second
(radiative) term of Eq. (10.19):
+∞
1 q 1 ⎡ n × { (n − β ) × β̇} ⎤
Eω = ∫ ⎢ ⎥ e iωt dt . (10.56)
2π 4πε0 cR ⎣ (1 − β ⋅ n)3 ⎦ret
−∞
Since the vectors n and β are more natural functions of the radiation (retarded) time
tret, let us use Eqs. (10.5) and (10.16) to exclude the observation time t from this
integral:
+∞
q 1 1 ⎡ n × {(n − β) × β̇} ⎤ ⎧ ⎛ R ⎞⎫
Eω = ∫ ⎢ ⎥ exp⎨iω⎜tret + ret ⎟⎬dtret. (10.57)
4πε0 2π cR ⎣ (1 − β ⋅ n) 2
⎦ret ⎩ ⎝ c ⎠⎭
−∞
Assuming that the observer is sufficiently far from the particle18, we may treat the
unit vector n as a constant and also use the approximation (8.19) to reduce
Eq. (10.57) to
Eω =
1 q 1
2π 4πε0 cR
exp
iωr
c { }
+∞
⎡ n × {(n − β) × β̇} (10.58)
⎧ ⎛ n ⋅ r′ ⎞⎫⎤
× ⎢
∫ exp ⎨ iω⎜t − ⎟⎬⎥ dt ret .
⎣ (1 − β ⋅ n)2 ⎩ ⎝ c ⎠⎭⎦ret
−∞
16
See, e.g., Eq. (A.88).
17
The notion of spectral density may be readily generalized to random processes—see, e.g. Part QM
section 7.4 and Part SM section 5.4.
18
According to the estimate Eq. (10.49), for a synchrotron radiation’s pulse this restriction requires the
observer to be much farther than Δrʹ ∼ cΔt ∼ R/γ3 from the particle. With the values R ∼ 104 m and γ ∼ 105
mentioned above, Δrʹ ∼ 10−11 m, so this requirement is satisfied for any realistic radiation detector.
10-16
Plugging this expression into Eq. (10.54), and using the definitions c ≡ 1/(ε0μ0)1/2 and
Z0 ≡ (μ0/ε0)1/2, we obtain19
+∞ 2
Z q2 ⎡ n × {(n − β) × β̇} ⎧ ⎛ n ⋅ r′ ⎞⎫⎤
I (ω ) = 0 3 ∫ ⎢ exp⎨iω⎜t − ⎟⎬⎥ dt ret . (10.59)
16π ⎣ (1 − β ⋅ n) 2 ⎩ ⎝ c ⎠⎭⎦ret
−∞
This result may be further simplified by noticing that the fraction before the
exponent may be represented as a full derivative over tret,
⎡ n × n − β) × β̇ ⎤
⎢ {( } ⎥ ≡ ⎡ n × {(n − β) × d β / dt} ⎤
⎢ ⎥
⎣ (1 − β ⋅ n)2 ⎦ret ⎣ (1 − β ⋅ n)2 ⎦ret
(10.60)
d ⎡ n × (n × β) ⎤
≡ ⎢ ⎥ ,
dt ⎣ 1 − β ⋅ n ⎦ret
and working out the resulting integral by parts. At this operation, the time
differentiation of the parentheses in the exponent gives d[tret − n · rʹ(tret)/c]/dtret =
(1 − n · u/c)ret ≡ (1 − β · n)ret, leading to the cancellation of the remaining factor in
the denominator and hence to a very simple general result20:
2
+∞ ⎡
Z q 2ω 2 ⎧ ⎛ n ⋅ r′ ⎞⎫⎤
I (ω ) = 0 3 ∫−∞ ⎢n × (n × β)exp ⎨iω⎜t − ⎟⎬⎥ dt ret . (10.61)
16π ⎣ ⎩ ⎝ c ⎠⎭⎦ret
Now returning to the particular case of the synchrotron radiation, it is beneficial

to choose the origin of time tret so that at tret = 0 the angle θ between the vectors n
and β takes its smallest value θ0, i.e. in terms of figure 10.5 the vector n is within the
plane [y, z]. Fixing this direction of axes so that they do not move in further times,
we can redraw that figure as shown in figure 10.8.
Figure 10.8. Deriving the spectral density of synchrotron radiation. The vector n is static within the plane
[y, z], while the vectors rʹ(tret) and βret rotate within the plane [x, y] with the angular velocity ωc of the particle.
19
Note that for our current purposes of calculation of the spectral density of radiation by a single particle, the
factor exp{iωr/c} has been cancelled. However, as we have seen in chapter 8, this factor plays the central role at
interference of radiation from several (many) sources. Such interference is important, in particular, in
undulators and free-electron lasers—the devices to be (qualitatively) discussed below.
20
Actually, this simplification is not occasional. According to Eq. (10.10b), the expression under the derivative
in the last form of Eq. (10.60) is just the transverse component of the vector-potential A (give or take a constant
factor), and from the discussion in section 8.2 we know that this component determines the electric dipole
radiation of the particle, which dominates the radiation field in our current case of a particle with
uncompensated electric charge.
10-17
In this ‘lab’ reference frame, the vector n does not depend on time, while the
vectors rʹ(tret) and βret do depend on it via angle α ≡ ωctret:
n = {0, sin θ 0, cos θ 0}, r′(tret ) = {R(1 − cos α ), 0, R sin α},
(10.62)
βret ≡ {β sin α , 0, β cos α}.
Now an easy multiplication yields

[n × (n × β)]ret = β{sin α , sin θ 0 cos θ 0 cos α , −sin2 θ 0 sin α} , (10.63)
⎡ ⎧ ⎛ n ⋅ r′ ⎞⎫⎤ ⎧ ⎛ R ⎞⎫
⎢exp ⎨iω⎜t − ⎟⎬⎥ = exp ⎨iω⎜tret − cos θ 0 sin α⎟⎬ . (10.64)
⎣ ⎩ ⎝ c ⎠⎭⎦ret ⎩ ⎝ c ⎠⎭
As we already know, in the (most interesting) ultra-relativistic limit γ ≫ 1, most

radiation is confined to short pulses, so that only small angles α ∼ ωcΔtret ∼ γ−1 may
contribute to the integral in Eq. (10.61). Moreover, since most radiation goes to
small angles θ ∼ θ0 ∼ γ−1, it makes sense to consider only such small angles.
Expanding both trigonometric functions of these small angles participating in
parentheses of Eq. (10.64) into the Taylor series, and keeping only the leading
terms, we obtain
R R R θ 02 R ωc3 3
tret − cos θ 0 sin α ≈ tret − ωctret + ωctret + tret. (10.65)
c c c 2 c 6
Since (R/c)ωc = u/c = β ≈ 1, in the two last terms we may approximate this parameter
by 1. However, it is crucial to distinguish the difference of the two first terms,
proportional to (1 − β)tret, from zero; as we have done before, we may approximate
it with tret/2γ2. On the right-hand side of Eq. (10.63), which does not have such a
critical difference, we may be bolder, taking21
β{sin α , sin θ 0 cos θ 0 cos α , −sin2 θ 0 sin α} ≈ {α , θ 0, 0}

(10.66)
≡ {ωctret, θ 0, 0}.
As a result, Eq. (10.61) is reduced to
Z0q 2 Z q2
I (ω ) = 3
∣ax nx + ay ny∣2 ≡ 0 3 (∣ ax∣2 + ∣ay∣2 ), (10.67)
16π 16π
where ax and ay are dimensionless factors,
21
This expression confirms that the in-plane (x) component of the electric field is an odd function of tret and
hence of t − t0 (see its sketch in figure 10.7b), while the normal (y) component is an even function of this
difference. Also note that for an observer exactly in the rotation plane (θ0 = 0) the latter component equals zero
for all times—the fact which could be predicted from the very beginning because of the evident mirror
symmetry of the problem with respect to the particle rotation plane.
10-18
+∞ ⎧ iω ⎛ ⎪ ω 2 3 ⎞⎫ ⎪
ax ≡ ω ∫ ωctret exp ⎨ ⎜(θ 02 + γ −2 )tret + c tret ⎟⎬dtret,

⎩2⎝ ⎠⎭
⎪ ⎪
−∞ 3
(10.68)
+∞ ⎧ iω ⎛ ⎪ ω 2 3 ⎞⎫ ⎪
ay ≡ ω ∫ θ 0 exp ⎨ ⎜(θ 02 + γ −2 )tret + c tret ⎟⎬dtret,

⎩2⎝ ⎠⎭
⎪ ⎪
−∞ 3
which describe the frequency spectra of two components of the synchrotron

radiation, with mutually perpendicular directions of polarization. Defining the
following dimensionless parameter
ω
ν≡ (θ 02 + γ −2 )3/2 , (10.69)
3ωc
which is proportional to the observation frequency, and changing the integration
variable to ξ ≡ ωctret/(θ02 + γ−2)1/2, the integrals (10.68) may be reduced to the
modified Bessel functions of the second kind, but with fractional indices:
ω 2 +∞ ⎧3 ⎛ ξ3 ⎞⎫
ax = (θ 0 + γ − 2 ) ∫−∞ ξ exp ⎨ iν⎜ξ + ⎟⎬dξ
ωc ⎩2 ⎝ 3 ⎠⎭
2 3 i
= νK2/3(ν ),
(θ 02 + γ −2 )1/2
(10.70)
ω +∞ ⎧3 ⎛ ξ3 ⎞⎫
ay = θ 0(θ 02 + γ −2 )1/2 ∫−∞ exp ⎨ iν⎜ξ + ⎟⎬dξ
ωc ⎩2 ⎝ 3 ⎠⎭
2 3 θ0
= νK1/3(ν ).
θ 02 + γ −2
Figure 10.9a shows the dependence of the Bessel factors, defining the amplitudes
ax and ay, on the normalized observation frequency ν. It shows that the radiation
intensity changes with frequency rather slowly (note the log–log scale of the plot)
until the normalized frequency, defined by Eq. (10.69), is increased beyond ∼1. For
most important observation angles θ0 ∼ γ this means that our estimate (10.50) is
indeed correct, although formally the frequency spectrum extends to infinity22.
Naturally, the spectral density integrated over the full solid angle exhibits a
similar frequency behavior. Without performing the integration23, let me just give
the result (also valid for γ ≫ 1 only) for the reader’s reference:
∞
3 2 2 ω
∮4π I (ω)d Ω = 4π
q γζ ∫ζ K5/3(ξ )dξ, where ζ ≡
3 ωcγ 3
. (10.71)
22
The law of the spectral density decrease at large ν may be readily obtained from the second of Eqs. (2.158),
which is valid even for any (even non-integer) Bessel function index n: ax ∝ ay ∝ ν−1/2exp{−ν}. Here the
exponential factor is certainly most important.
23
For that, and many other details, the interested reader is referred, for example, to the fundamental review
collection [1] or the more concise monograph [2].
10-19
Figure 10.9. The frequency spectra of (a) two components of the synchrotron radiation, at a fixed angle θ0, and
(b) its total (polarization- and angle-averaged) intensity.
Figure 10.9b shows the dependence of this integral on the normalized frequency ζ.
(This plot is sometimes called the ‘universal flux curve’.) In accordance with estimate
(10.50), it reaches maximum at
ω
ζmax ≈ 0.3, i.e. ωmax ≈ c γ 3. (10.72)
2
For example, in the new National Synchrotron Light Source (NSLS-II) in the
Brookhaven National Laboratory near the SBU campus, with a ring circumference
of 792 m, the electron revolution periodT is 2.64 μs. Calculating ωc as 2π/T ≈ 2.4 ×
106 s−1, for the achieved γ ≈ 6 × 103 (E ≈ 3 GeV)24 we obtain ωmax ∼ 3 × 1017 s−1 (the
photon energy ℏωmax ∼ 200 eV), corresponding to soft x-rays. In the light of this
estimate, the reader may be surprised by figure 10.10, which shows the calculated
spectra of radiation that this facility was designed to produce, with the intensity
maxima at photon energies up to a few keV.
The reason for this discrepancy is that in NLLS-II, and in all modern synchrotron
light sources, most radiation is produced not by the circular orbit itself (which is, by
the way, not exactly circular, but consists of a series of straight and bend-magnet
sections), but by such bend sections, and also devices called wigglers and undulators:
strings of several strong magnets with alternating field direction (figure 10.11), that
induce periodic bending (‘wiggling’) of the electron’s trajectory, with the synchro-
tron radiation emitted at each bend.
The difference between the wigglers and the undulators is more quantitative than
qualitative: the former devices have a larger spatial period λu (the distance between
the adjacent magnets of the same polarity, see figure 10.11), giving enough space for
the electron beam to bend by an angle larger than γ−1, i.e. larger than the radiation
cone’s width. As a result, the radiation arrives to an in-plane observer as a periodic
sequence of individual pulses (figure 10.12a).
24
By modern standards, this energy is not too high. The distinguished feature of NSLS-II is its unprecedented
electron beam intensity (planned average beam current up to 500 mA) which should allow an extremely high
synchrotron ‘brightness’—see figure 10.10.
10-20
Figure 10.10. Design brightness of various synchrotron radiation sources of the NSLS-II facility. For bend
magnets and wigglers, the ‘brightness’ may be obtained by multiplication of the spectral density I(ω) from one
electron pulse, calculated above, by the number of electrons passing the source per second. (Note the non-SI
units, commonly used by the synchrotron radiation community.) However, for undulators, there is an
additional factor due to the partial coherence of radiation—see below. (Adapted from NSLS-II Source
Properties and Floor Layout, available online at https://www.bnl.gov/ps/docs/pdf/SourceProperties.pdf.)
The shape of each pulse, and hence its frequency spectrum, is essentially similar to
those discussed above25, but with much higher local values of ωc and hence ωmax—
see figure 10.10. Another difference is a much higher frequency of the pulses. Indeed,
the fundamental Eq. (10.16) allows us to calculate the time distance between them,
for the observer, as
∂t λ 1 λ λ
Δt ≈ Δtret ≈ (1 − β ) u ≈ 2 u ≪ u , (10.73)
∂tret u 2γ c c
where the first two relations are valid at λu ≪ R (a relation that is typically satisfied
very well, see the numbers in figure 10.10), and the last two relations assume the
ultra-relativistic limit. As a result, the radiation intensity, which is proportional to
25
Indeed, the period λu is typically a few centimeters (see the numbers in figure 10.10), i.e. is much larger than
the interval Δrʹ ∼ R/γ3 estimated above. Hence the synchrotron radiation results may be applied locally, to
each electron beam’s bend. (In this context, a simple problem for the reader: use Eqs. (10.19) and (10.63) to
explain the difference between the shapes of the in-plane electric field pulses emitted at opposite magnetic poles
of the wiggler, which is schematically shown in figure 10.11a.)
10-21
Figure 10.11. The generic structure of wigglers, undulators, and free-electron lasers. (Adapted from http://
www.xfel.eu/overview/how_does_it_work/.)
Figure 10.12. Schematic representation of waveforms of the radiation emitted by (a) a wiggler and (b) an
undulator.
the number of the magnetic poles, is much higher than that from the bend magnets—
see figure 10.10 again.
The situation is different in undulators—similar structures with a smaller spatial
period λu, in which the electron’s velocity vector oscillates with an angular amplitude
smaller than γ−1. As a result, the radiation pulses overlap (figure 10.12b) and the
radiation waveform is closer to a sinusoidal one. As a result, the radiation spectrum
narrows to the central frequency26
2π 2πc
ω0 = ≈ 2γ 2 . (10.74)
Δt λu
For example, for the LSNL-II undulators with λu = 2 cm, this formula predicts the
radiation peak at phonon energy ℏω0 ≈ 4 keV, in reasonable agreement with the
results of quantitative calculations, shown in figure 10.10.27 Due to the spectrum
narrowing, the intensity of undulators radiation is higher than that of wigglers using
the same electron beam.
26
This important formula may be also derived in the following way. Due to the relativistic length contraction
(9.20), the undulator structure period as perceived by beam electrons is λʹ = λu/γ, so that the central frequency
of the radiation in the reference frame moving with the electrons is ω0ʹ = 2πc/λʹ = 2πcγ/λu. For the lab-frame
observer, this frequency is Doppler-upshifted in accordance with Eq. (9.44): ω0 = ω0ʹ[(1 + β)/(1 − β)]1/2 ≈ 2γω0ʹ,
giving the same result as Eq. (10.74).
27
Some of the difference is due to the fact that those plots show the spectral density of the number of photons
n = E /ℏω per second, which peaks at a frequency below that of the density of power, i.e. of the energy E per
second.
10-22
This spectrum-narrowing trend is brought to its logical conclusion in the so-called

free-electron lasers,28 whose basic structure is the same as that of wigglers and
undulators (figure 10.11), but the radiation at each beam bend is so intense and
narrow-focused that it affects the electron motion downstream the radiation cone.
As a result, the radiation spectrum narrows around the central frequency (10.74),
and its power grows as a square of the number N of electrons in the structure (rather
than proportionately to N in wigglers and undulators).
Finally, note that wigglers, undulators, and free-electron lasers may also be used
at the end of linear electron accelerators (such as SLAC) which, as was noted above,
may provide extremely high values of γ, and hence radiation frequencies, due to the
absence of the radiation energy losses at the electron acceleration stage. Very
unfortunately, I do not have time/space to discuss (the very interesting) physics of
these devices in more detail29.
10.4 Bremsstrahlung and Coulomb losses

Surprisingly, a very similar mechanism of radiation by charged particles works at
much lower spatial scale, namely at their scattering by charged particles of the
propagation medium. This effect, traditionally called by its German name brems-
strahlung (‘brake radiation’), is responsible, in particular, for the continuous part of
the frequency spectrum of the radiation produced in standard vacuum x-ray tubes,
at the electrons’ collisions with a metallic ‘anti-cathode’30.
The bremsstrahlung in condensed matter is generally a rather complicated
phenomenon, because of the simultaneous involvement of many particles and
(frequently) some quantum electrodynamic effects. This is why I will provide only
a very brief glimpse of the theoretical description of this effect, for the simplest case
when the scattering of incoming, relatively light charged particles (such as electrons,
protons, α-particles, etc) is produced by atomic nuclei that remain virtually
immobile during the scattering event (figure 10.13). This is a reasonable approx-
imation if the energy of incoming particles is not too low; otherwise most scattering
Figure 10.13. The basic geometry of the bremsstrahlung and the Coulomb loss problems in (a) direct and (b)
reciprocal spaces.
28
This name is somewhat misleading, because in contrast to the usual (‘quantum’) lasers, the free-electron laser
is essentially a classical device, and the dynamics of electrons in it is very similar to that in other vacuum
microwave generators, such as the magnetrons briefly discussed in section 9.6 and especially klystrons (see
chapter 7).
29
The interested reader may be referred, for example, to either [3] or [4].
30
Such x-ray radiation was first observed experimentally, although not correctly interpreted, by N Tesla in
1887, i.e. before the radiation was re-discovered (in 1895) and studied in detail by W Röntgen.
10-23
is produced by atomic electrons whose dynamics is substantially quantum—see

below.
To calculate the frequency spectrum of radiation emitted during a single
scattering event, it is convenient to use a by-product of the last section’s analysis,
namely Eq. (10.59) with the replacement (10.60)31:
2
+∞ ⎡
1 q2 d n × (n × β) ⎧ ⎛ n ⋅ r′ ⎞⎫⎤
I (ω ) = ∫−∞ ⎢ exp ⎨iω⎜t − ⎟⎬⎥ dt ret . (10.75)
4π 2c 4πε0 ⎣ dt 1 − β ⋅ n ⎩ ⎝ c ⎠⎭⎦ret
A typical duration τ of a single scattering event we are discussing is of the order of

a0/c ∼ (10−10 m)/(3 × 108 m s−1) ∼ 10−18 s in solids, and only an order of magnitude
longer in gases at ambient conditions. This is why for most frequencies of interest,
from zero all the way up to at least soft x-rays32, we can use the so-called low-
frequency approximation, taking the exponent in Eq. (10.75) for 1 through the whole
collision event, i.e. the integration interval. This approximation immediately yields
2
1 q2 n × (n × βfin) n × (n × βini)
I (ω ) = − . (10.76)
4π 2c 4πε0 1 − βfin ⋅ n 1 − βini ⋅ n
In the non-relativistic limit (βini, βfin ≪ 1), this formula is reduced to the result
1 q2 q 2
I (ω ) = sin2 θ (10.77)
4π 2c 4πε0 m 2c−2
(which may be derived from Eq. (8.27) as well), where q is the momentum
transferred from the scattering center to the scattered charge (figure 10.13b)33:
q ≡ pfin − pini = mΔu = mcΔβ = mc(βfin − βini), (10.78)
and θ is the angle between the vector q and the direction n toward the observer (at
the collision moment).
31
In publications on this topic (whose development peak was in the 1920–30 s) Gaussian units are more
common, and the letter Z is usually reserved for expressing charges as multiples of the fundamental charge e,
rather than for the wave impedance. This is why, in order to avoid confusion and facilitate the comparison
with other texts, in this section I (still staying with the SI units used through my series) will use the fraction
1/ε0c, instead of its equivalent Z0, for the free-space wave impedance and write the coefficients in a form that
makes the transfer to the Gaussian units trivial: it is sufficient to replace all (qqʹ/4πε0)SI with (qqʹ)Gaussian. In the
(rare) cases when I spell out the charge values, I will use a different font: q ≡ Z e, qʹ ≡ Z ʹe.
32
A more careful analysis shows that this approximation is actually quite reasonable up to much higher
frequencies of the order of γ2/τ.
33
Please note the font-marked difference between this variable (q ) and the particle’s electric charge (q).
10-24
The most important feature of the result (10.77)–(10.78) is the frequency-

independent (‘white’) spectrum of the radiation, very typical for any rapid leaps,
which may be approximated as delta-functions of time34. (Note, however, that
Eq. (10.77) implies a fixed value of q , so that the statistics of this parameter, to be
discussed in a minute, may ‘color’ the radiation.) Note also the ‘doughnut-shaped’
angular distribution of the radiation, typical for non-relativistic systems, with the
symmetry axis directed along the momentum transfer vector q . In particular, this
means that in typical cases when q ≪ p, i.e. the vector q is nearly normal to the
vector pini (see, e.g. the example shown in figure 10.13), the bremsstrahlung produces
a significant radiation flow in the direction back to the particle source—the fact
significant for the operation of x-ray tubes.
Now integrating the result (10.77) over all wave propagation angles, just as we did
for the instant radiation power in section 8.2, we get the following spectral density of
the particle energy loss,
dE 2 q2 q 2
−
dω
= ∮4π I (ω ) d Ω =
3πc 4π ε0 m 2c 2
. (10.79)
The main new feature of bremsstrahlung (as of most scattering problems35), is the
necessity to take into account the randomness of the impact parameter b (figure
10.13). For elastic (βini = βfin ≡ β) Coulomb collisions we can use the so-called
Rutherford formula for the differential cross-section of scattering36
dσ ⎛ qq′ ⎞2 ⎛ 1 ⎞2 1
=⎜ ⎟⎜ ⎟ . (10.80)
d Ω′ ⎝ 4πε0 ⎠ ⎝ 2pcβ ⎠ sin (θ ′ /2)
4
Here dσ = 2πbdb is the elementary area of the sample cross-section (as visible from
the direction of incident particles) corresponding to particle scattering into an
elementary body angle37
34
This is the foundation, in particular, of the so-called high-harmonic generation (HHG), discovered in 1977,
which takes place at the irradiation of materials by intensive laser beams. The high electric field of the beam
strips valent electrons from initially neutral atoms and accelerates them, just to slam them back into the
remaining ions as the field’s polarity changes in time. The electrons change their momentum sharply during
their recombination with the ions, resulting in a bremsstrahlung-like radiation. The spectrum of radiation from
each such event obeys Eq. (10.77), but since the ionization/acceleration/recombination cycles repeat periodi-
cally with the frequency ω0 of the laser field, the final spectrum consists of many equidistant lines, with
frequencies nω0. The bremsstrahlung mechanism does not provide for a classical cutoff ωmax = nmaxω0 of the
spectrum, but such a limitation is imposed by quantum mechanics: ℏωmax ∼ E p, where the so-called
ponderomotive energy E p = (eE0/ω0)2/4me is the average kinetic energy given to a free electron by the periodic
electric field of the laser, with amplitude E0. In practice, nmax may be as high as ∼100, enabling alternative
compact sources of x-ray radiation. For a detailed quantitative theory of this effect, see, for example, [5].
35
See, e.g. Part CM section 3.5 and Part QM section 3.3.
36
See, e.g. Part CM Eq. (3.73) with α = qqʹ/4πε0. In the form used in Eq. (10.80), the Rutherford formula is
also valid for small-angle scattering of relativistic particles, the criterion being ∣Δβ∣ ≪ 2/γ.
37
The angle θʹ and differential dΩʹ, describing the direction of scattered particles (see figure 10.13) should not
be confused with the parameters θ and dΩ describing the direction of the radiation emitted at the scattering
event.
10-25
d Ω′ = 2π sin θ ′ dθ′ . (10.81)

Differentiating the geometric relation, which is evident from figure 10.13b,
θ′
q = 2p sin , (10.82)
2
we may represent Eq. (10.80) in a more convenient form
dσ ⎛ qq′ ⎞2 1
= 8π ⎜ ⎟ . (10.83)
dq ⎝ 4πε0 ⎠ u 2q 3
Now combining Eqs. (10.79) and (10.83) we obtain
16 q 2 ⎛ qq′ ⎞ 1 1
2
dE dσ
− = ⎜ ⎟ . (10.84)
dω d q 3 4πε0 ⎝ 4πε0mc 2 ⎠ cβ 2 q
This product is called the differential radiation cross-section. When integrated

over all values of q (which is equivalent to averaging over all values of the impact
parameter), it gives a convenient measure of the radiation intensity. Indeed, after the
multiplication by the volume density n of independent scattering centers, the integral
yields the particle’s energy loss per unit bandwidth of radiation per unit path length,
−d2E /dωdx. A minor problem here is that the integral of 1/q formally diverges at
both infinite and zero values of q . However, these divergences are very weak
(logarithmic), and the integral converges for virtually any reason unaccounted for by
our simple analysis. The standard, although slightly approximate, way to account
for these effects is to write
16 q 2 ⎛ qq′ ⎞ 1
2
d 2E q
− ≈ n ⎜ ⎟ ln max , (10.85)
dωdx 3 4πε0 ⎝ 4πε0mc ⎠ cβ
2 2
qmin
and then plug, instead of q max and q min, the scales of the most important effects
limiting the range of the momentum transfer magnitude. In classical-mechanics
analysis, according to Eq. (10.82), q max = 2p ≡ 2mu. To estimate q min, let us note
that the very small momentum transfer takes place when the impact parameter b is
very large and hence the effective scattering time τ ∼ b/v is very long. Recalling the
condition of the low-frequency approximation, we may associate q min with τ ∼ 1/ω
and hence with b ∼ uτ ∼ v/ω. Since for the small scattering angles q may be estimated
as the impulse Fτ ∼ (qqʹ/4πε0b2)τ of the Coulomb force, we obtain the estimate q min
∼ (qqʹ/4πε0)ω/u2, and Eq. (10.85) should be used with
qmax ⎛ 2mu 3 qq′ ⎞
ln = ln ⎜ ⎟. (10.86)
qmin ⎝ ω 4πε0 ⎠
This is the Bohr’s formula for what is called the classical bremsstrahlung. We see
that the low momentum cutoff indeed makes the spectrum slightly colored, with
more energy going to lower frequencies. There is even a formal divergence at ω → 0;
10-26
however, this divergence is integrable, so it does not present a problem for finding
the total energy radiative losses (−dE /dx) as an integral of Eq. (10.86) over all
radiated frequencies ω. A larger problem for this procedure is the upper integration
limit, ω → ∞, at which the integral diverges. This means that our approximate
description, which considers the collision as an elastic process, becomes invalid and
needs to be amended by taking into account the difference between the initial and
final kinetic energies of the particle due to radiation of the energy quantum ℏω of the
emitted photon, so that
2 2 2 2
pini pfin pini pfin
− = ℏω, i.e. = E, = E − ℏω. (10.87)
2m 2m 2m 2m
As a result, taking into account that the minimum and maximum values of q
correspond to, respectively, the parallel and antiparallel alignments of the vectors
pini and pfin, we obtain
qmax pini + pfin (p + pfin )2 /2m
ln = ln ≡ ln ini2
qmin pini − pfin (pini − pfin2 )/2m (10.88)
[E1/2 + (E − ℏω)1/2 ]2
= ln ,
ℏω
Plugged into Eq. (10.85), this expression yields the so-called Bethe–Heitler formula
for quantum bremsstrahlung38. Note that at this approach, q max is close to that of the
classical approximation, but q min ∼ ℏω/u, so that
qmin αZZ ′
classical
∼ , (10.89)
qmin β
classical
where Z and Z ʹ are the particles’ charges in units of e, and α is the fine structure
(‘Sommerfeld’) constant,
e2 e2 1
α≡ = ≈ ≪ 1, (10.90)
4πε0ℏc SI
ℏc Gaussian
137
which is one of the basic notions of quantum mechanics39. Due to the smallness of
the constant, the ratio (10.89) is smaller than 1 for most cases of practical interest,
and since the integral of (10.84) over q is limited by the largest of all possible cutoffs
q min, it is the Bethe–Heitler formula which should be used.
Now nothing prevents us from calculating the total radiative losses of energy per
unit length:
38
The modifications of this formula necessary for the relativistic case description are surprisingly minor—see,
e.g. chapter 15 of [6]. For even more detail, the standard reference monograph on bremsstrahlung is [7].
39
See, e.g. Part QM sections 6.3, 9.3, 9.5, and 9.7.
10-27
∞⎛
dE d 2E ⎞
−
dx
= ∫0 ⎜−
⎝ dωdz ⎠
⎟dω
(10.91)
16 q 2 ⎛ qq′ ⎞ 1
2 ω max
E1/2 − (E − ℏω)1/2
= n ⎜ ⎟
3 4πε0c ⎝ 4πε0mc 2 ⎠ β 2
2 ∫0 ln
(ℏω)1/2
dω ,
where ℏωmax = E is the maximum energy of the radiation quantum. By introducing

the dimensionless integration variable ξ ≡ ℏω/E = 2ℏω/(mu2/2), this integral is
reduced to a table one40, and we obtain
16 q 2 ⎛ qq′ ⎞ 1 u 2 16 ⎛ q′2 ⎞⎛ q 2 ⎞ 1
2 2
dE
− = n ⎜ ⎟ = n ⎜ ⎟⎜ ⎟ . (10.92)
dx 3 4πε0c ⎝ 4πε0mc 2 ⎠ β 2 ℏ 3 ⎝ 4πε0ℏc ⎠⎝ 4πε0 ⎠ mc 2
Following my usual style, at this point I would give you an estimate of the losses
for a typical case; however, let me first discuss a parallel energy loss mechanism, the
so-called Coulomb losses, due to the transfer of mechanical impulse from the
scattered particle to the scattering centers. (This energy eventually goes into an
increase of the thermal energy of the scattering medium, rather than to the
electromagnetic radiation.)
Using Eqs. (9.139) for the electric field of a linearly moving charge q, we can
readily find the momentum it transfers to the counterpart charge qʹ:41
+∞ +∞
Δp′ = (Δp′) y = ∫−∞ (p ̇ ′) y dt = ∫−∞ q′Ey dt
+∞
(10.93)
qq′ γb qq′ 2
=
4πε0
∫−∞ 2 2 2 2 3/2
(b + γ u t )
dt =
4πε0 bu
.
Hence, the kinetic energy acquired by the scattering particle (and hence to the loss of
the energy E of the incident particle) is
(Δp′)2 ⎛ qq′ ⎞2 2
−ΔE = =⎜ ⎟ . (10.94)
2m′ ⎝ 4πε0 ⎠ m′u 2b 2
Such elementary energy losses have to be summed up over all collisions, with
random values of the impact parameter b. At the scattering center density n, the
number of collisions per small path length dz per small range db is dN = n 2πb db dx,
so that
dE ⎛ qq′ ⎞2 2 bmax
db
−
dx
= ∫ ( −ΔE )dN = n ⎜ ⎟
⎝ 4πε0 ⎠ m′u 2
2π ∫b min b
. (10.95)
⎛ qq′ ⎞ ln B
2
b
= 4 π n⎜ ⎟ , where B ≡ max
⎝ 4πε0 ⎠ m′u 2 bmin
40
See, e.g., Eq. (A.41).
41
According to Eq. (9.139), Ez =0, and the net impulse of the longitudinal force qʹEx is also zero.
10-28
Here the logarithmic integral over b was treated similarly to that over q in the
bremsstrahlung theory. This approach is adequate, because the ratio bmax/bmin is
much larger than 1. Indeed, bmin may be estimated from (Δpʹ)max ∼ p = γmu. For this
value, Eq. (10.93) with qʹ ∼ q gives bmin ∼ rc (see Eq. (8.41) and its discussion), which
is, for elementary particles, of the order of 10−15 m. On the other hand, for the most
important case when the charges qʹ belong to electrons (which, according to
Eq. (10.94), are the most efficient Coulomb energy absorbers due to their extremely
low mass mʹ), bmax may be estimated from condition τ = b/γu ∼ 1/ωmax, where ωmax ∼
1016 s−1 is the characteristic frequency of electron transitions in atoms. (Below this
frequency, our classical analysis of the scatterer’s motion is invalid.) From here, we
have the estimate bmax ∼ γu/ωmax, so that
bmax γu
B≡ ∼ , (10.96)
bmin rcω0
for γ ∼ 1 and u ∼ c ≈ 3 × 108 m s−1 giving bmax ∼ 3 × 10−8 m, so that B ∼ 109 (give or
take a couple orders of magnitude—this does not change the estimate ln B ≈ 20 too
much)42.
Now we can compare the Coulomb losses (10.95) with those due to the
bremsstrahlung, given by Eq. (10.92):
−dE radiation m′ 1
∼ α ZZ ′ β 2 . (10.97)
−dE Coulomb m ln B
Since α ∼ 10−2 ≪ 1, for non-relativistic particles (β ≪ 1) the bremsstrahlung losses of

energy are much lower (this is why we did not need to rush with their estimate), and
only for ultra-relativistic particles may the relation be opposite.
According to Eq. (10.95), for electron–electron scattering (q = qʹ = −e, mʹ = me),43
at the value n = 6 × 1026 m−3 typical for air at ambient conditions, the characteristic
length of energy loss,
E
lc ≡ , (10.98)
( −dE / dx )
for electrons with kinetic energy E = 6 keV is close to 2 × 10−4 m = 0.2 mm. (This is
why we need a high vacuum in particle accelerators and electron microscope
columns!) Since lc ∝ E 2, more energetic particles penetrate deeper into matter until
the bremsstrahlung steps in and limits this trend at very high energies.
42
A quantum analysis (carried out by H Bethe in 1940) replaces, in Eq. (10.95), ln B with ln(2γ2mu2/ℏ〈ω〉) − β 2,
where 〈ω〉 is the average frequency of the atomic quantum transitions’ weight by their oscillator strength. This
refinement does not change the estimate given below. Note that both the classical and quantum formulas
describe a fast increase (as 1/β) of the energy loss rate (−dE /dx) at γ → 1, and its slow increase (as lnγ) at γ → ∞,
so that the losses have a minimum at (γ − 1) ∼ 1.
43
Actually, the above analysis has neglected the change of momentum of the incident particle. This is
legitimate at mʹ ≪ m, but for m = mʹ the change approximately doubles the energy losses. Still, this does not
change the order of magnitude of the estimate.
10-29
10.5 Density effects and Cherenkov radiation

For condensed matter, the Coulomb loss estimate made in the last section is not
quite suitable, because it is based on the upper cutoff bmax ∼ γu/ωmax. For the
example given above, the incoming electron velocity u is close to 5 × 107 m s−1, and
for the typical value ωmax ∼ 1016 s−1 (ℏωmax ∼ 10 eV), this cutoff bmax ∼ 5 × 10−9 m =
5 nm. Even for air at ambient conditions, this cutoff is larger than the average
distance (∼2 nm) between the molecules, so that at the high end of the impact
parameter range, at b ∼ bmax, the Coulomb loss events in adjacent molecules are not
quite independent, and the theory needs corrections. For condensed matter, with
much higher particle density n, most collisions satisfy the condition
nb3 ≫ 1, (10.99)
and the treatment of Coulomb collisions as independent events is inadequate.

However, the condition (10.99) enables the opposite approach: treating the medium
as a continuum. In the time domain formulation, used in the previous sections of this
chapter, this would be a very complex problem, because it would require an explicit
description of the medium dynamics. Here the frequency-domain approach, based
on the Fourier transform in both time and space, helps a lot, provided that functions
ε(ω) and μ(ω) are considered known—either calculated or taken from experiment.
Let us have a good look at this approach, because it gives some interesting (and
practically important) results.
In chapter 6, we have used the macroscopic Maxwell equations to derive Eqs.
(6.118), which describe the time evolution of potentials in a linear medium with
frequency-independent ε and μ. Looking for all functions participating in Eqs.
(6.118) in the form of plane-wave expansion44
f (r , t ) = ∫ d 3k∫ dωfk,ω ei(k⋅r−ωt ), (10.100)
and requiring all coefficients at similar exponents to be balanced, we obtain their

Fourier image45:
ρk,ω
(k 2 − ϕ 2εμ) φ k,ω = , (k 2 − ω 2εμ) A k,ω = μ jk,ω . (10.101)
ε
As was discussed in chapter 7, in such a Fourier form, the macroscopic Maxwell
theory remains valid even for the dispersive (but linear!) media, so that Eq. (10.101)
may be generalized as
44
All integrals here and below are in infinite limits, unless specified otherwise.
45
As was discussed in section 7.2, the Ohmic conductivity of the medium (generally, also a function of
frequency) may be readily incorporated into the dielectric permittivity: ε(ω) → εef(ω) + iσ(ω)/ω. In this section,
I will assume that such incorporation, which is especially natural for high frequencies, has been performed, so
that the current density j(r, t) describes only stand-alone currents—for example, the current (10.105) of the
incident particle.
10-30
ρk,ω
[k 2 − ω 2ε(ω)μ(ω)] φ k,ω = , [k 2 − ω 2ε(ω)μ(ω)] A k,ω = μ(ω) jk,ω , (10.102)
ε( ω )
The evident advantage of these equations is that their formal solution is trivial:
ρk,ω μ(ω) jk,ω
φ k,ω = 2 2
, A k,ω = , (10.103)
ε(ω)[k − ω ε(ω)μ(ω)] [k − ω 2ε(ω)μ(ω)]
2
so that the ‘only’ remaining things to do are to, first, calculate the Fourier transforms
of the functions ρ(r, t) and j(r, t), describing stand-alone charges and currents, using the
transform reciprocal to Eq. (10.100) with one factor 1/2π per each scalar dimension,
1
fk,ω =
(2π )4
∫ d 3r∫ dtf (r, t )e−i(k⋅r−ωt ), (10.104)
and then carry out the integration (10.100) of Eqs. (10.103).

For our current problem of a single charge q, uniformly moving in the medium
with velocity u,
ρ(r , t ) = qδ(r − ut ), j(r , t ) = q uδ(r − ut ), (10.105)
the first task is easy:
q
ρk,ω =
(2π )4
∫ ∫
d 3r dt qδ(r − ut )e−i (k⋅r−ωt )
(10.106)
q q
=
(2π )4
∫
e i (ωt −k⋅ut )dt =
(2π )3
δ(ω − k ⋅ u).
Since the expressions (10.105) for ρ(r, t) and j(r, t) differ only by a constant factor u,
it is clear that the absolutely similar calculation for the current gives
qu
jk,ω = δ(ω − k ⋅ u). (10.107)
(2π )3
Let us summarize what we have got by now, plugging Eqs. (10.106) and (10.107)
into Eqs. (10.103):
1 qδ(ω − k ⋅ u)
φ k,ω = ,
(2π ) ε(ω)[k 2 − ω 2ε(ω)μ(ω)]
3
(10.108)
1 μ(ω)q uδ(ω − k ⋅ u)
A k,ω = = ε(ω)μ(ω)uφ k,ω .
(2π )3 [k 2 − ω 2ε(ω)μ(ω)]
Now, at the last calculation step, namely the integration (10.100), we are starting
to pay a heavy price for the easiness of the first steps. This is why we need to consider
well what exactly we need from it. First of all, for the calculation of power losses, the
electric field is more convenient to use than the potentials, so let us calculate the
Fourier images of E and B. Plugging the expansion (10.100) into the basic relations
(6.7), and again requiring the balance of exponent’s coefficients, we obtain
10-31
Ek,ω = − i kφk,ω + iω A k,ω = i [ωε(ω)μ(ω)u − k] φ k,ω ,

(10.109)
B k,ω = i k × A k,ω = iε(ω)μ(ω)k × uφ k,ω ,
so that Eqs. (10.100) and (10.108) yield
E(r , t ) = ∫ d 3k∫ dω Ek,ωe i (k⋅r−ωt )

iq [ωε(ω)μ(ω)u − k] δ(ω − k ⋅ u) i (k⋅r−ωt ) (10.110)
=
(2π )3
∫ d 3k∫ dω ε(ω)[k 2 − ω 2ε(ω)μ(ω)]
e .
This formula may be rewritten as the temporal Fourier integral (10.51), with a
space-dependent amplitude
Eω(r) = ∫ Ek,ωei k⋅rd 3k

iq (10.111)
=
(2π )3
∫ [ωε(εω(ω)μ)[(ωk 2)u−−ωk2ε] (δω(ω)μ−(ωk)] ⋅ u) ei k⋅rd 3k .
Let us calculate the Cartesian components of this partial Fourier image Eω, at a
point separated by distance b from the particle’s trajectory. Selecting the coordinates
and time origin as shown in figure 9.11a, we have r = {0, b, 0}, so that only Ex and
Ey are different from zero. In particular, according to Eq. (10.111),
iq (ω)μ(ω)u − kx
(Ex )ω =
(2π )3ε(ω)
∫ dkx∫ dky∫ dkz ωε
k − ω 2ε(ω)μ(ω)
2
δ(ω − kxu )
(10.112)
× exp {ikyb} .
The delta-function kills one integral (over kx) of the three, and we obtain
iq ⎡ ω⎤
(Ex )ω = 3 ⎢⎣ωε(ω)μ(ω)u − ⎥⎦
(2π ) ε(ω)u u
(10.113)
× ∫ exp {ikyb}dky ∫ ω2 /u 2 + k 2 + kdk2z− ω2ε(ω)μ(ω) .
y z
The internal integral (over kz) may be readily reduced to the table integral ∫dξ/(1 + ξ2)
in infinite limits, equal to π.46 The result may be represented as
iπ qκ 2 exp {ikyb}
(Ex )ω = − ∫ dky, (10.114)
(2π )3ωε(ω) (k y2 + κ 2 )1/2
where the parameter κ (generally, a complex function of frequency) is defined as47
46
See, e.g. Eq. (A.32a).
47
Of course, the frequency-dependent parameter κ(ω) should not be confused with the dc low-frequency
dielectric constant κ ≡ ε(0)/ε0, which was discussed in chapter 3.
10-32
⎛1 ⎞
κ 2(ω) ≡ ω 2⎜ 2 − ε(ω)μ(ω)⎟ . (10.115)
⎝u ⎠
The last integral may be expressed via the modified Bessel function of the second
kind48:
iquκ 2
(Ex )ω = − K 0(κb). (10.116)
(2π )2 ωε(ω)
A very similar calculation yields
qκ
(Ey )ω = (2π )2 ε(ω)
K1(κb). (10.117)
Now, instead of rushing to make the final integration (10.51) over ω to calculate
E(t), let us realize that what we need is actually only the total energy loss through the
whole time of particle’s passage over an elementary distance dx. According to
Eq. (4.38), the energy loss per unit volume is
dE
−
dV
= ∫ j ⋅ E dt, (10.118)
where j is the current of the bound charges in the medium, and should not be
confused with the stand-alone particle’s current (10.105). This integral may be
readily expressed via the partial Fourier image Eω and the similarly defined image jω,
just as was done at the derivation of Eq. (10.54):
dE
−
dV
= ∫ dt∫ dωe−iωt∫ dω′e−iω′t jω ⋅ Eω′
(10.119)
= 2π∫ dω∫ dω′ jω ⋅ Eω′δ(ω + ω′) = 2π∫ jω ⋅ E−ω dω.
In our approach, the effective Ohmic conductivity σef(ω) is incorporated into the
complex permittivity ε(ω) just as was discussed in section 7.2, so that we may use
Eq. (7.46) to write
jω = σef (ω)Eω = −iωε(ω)Eω. (10.120)
As a result, Eq. (10.119) yields

∞
dE
−
dV
∫
= −2πi ε(ω)Eω ⋅ E−ωω dω = 4π Im ∫0 ε(ω) Eω 2 ω dω. (10.121)
(The last step was possible due to the property ε(−ω) = ε*(ω), which was discussed in
section 7.2.)
48
As a reminder, the main properties of these functions are listed in section 2.7—see, in particular, figure 2.22,
and Eqs. (2.157) and (2.158).
10-33
Finally, just as in the last section, we have to average the energy loss rate over
random values of the impact parameter b:
dE ⎛ dE 2 ⎞ ∞ ⎛ dE ⎞
−
dx
= ∫ ⎜⎝− dV ⎟d b ≈ 2π∫
⎠ b min
⎜− ⎟ b db
⎝ dV ⎠
∞ ∞
(10.122)
= 8π ∫b
2
min
b db ∫0 ( Ex 2ω + Ey
2
ω )Imε(ω)ω dω.
Note that due to the divergence of the functions K0(ξ) and K1(ξ) at ξ → 0 we have to
cut the resulting integral over b at some bmin where our theory loses legitimacy. (On
that limit, we are not doing much better than in the past section). Plugging in the
calculated expressions (10.116) and (10.117) for the field components, swapping the
integrals over ω and b, and using the recurrence relations (2.142), which are valid for
any Bessel functions, we finally obtain:
∞
dE 2 dω
−
dx
= q 2Im
π
∫0 (κ *bmin )K1(κ *bmin )K 0(κ *bmin )
ωε(ω)
. (10.123)
This general result is valid for an arbitrary linear medium, with arbitrary
dispersion relations ε(ω) and μ(ω). (The last function participates in Eq. (10.123)
only via Eq. (10.115), which defines the parameter κ.) To obtain more concrete
results, some particular model of the medium should be used. Let us explore the
Lorentz oscillator model, which was discussed in section 7.2, in its form (7.33)
suitable for the transition to a quantum-mechanical description of atoms:
nq′2 fj
ε( ω ) = ε0 + ∑
m j (ω j2 − ω 2 ) − 2iωδj
, with ∑ f j = 1; μ(ω) = μ0 . (10.124)
j
If the damping of the effective atomic oscillators is low, δj ≪ ωj, as it typically is, and
the particle’s speed u is much lower than the typical wave’s phase velocity v (and
hence c!), then for most frequencies Eq. (10.115) gives
⎛1 1 ⎞ ω2
κ 2 (ω ) ≡ ω 2 ⎜ 2 − 2 ⎟≈ 2, (10.125)
⎝u v (ω ) ⎠ u
i.e. κ ≈ κ* ≈ ω/u is real. In this case, Eq. (10.123) may be shown to reduce to
Eq. (10.95) with
1.123u
bmax = . (10.126)
ω
The good news here is that both approaches (the microscopic analysis of section
10.4 and the macroscopic analysis of this section) give essentially the same result.
The same fact may be also perceived as bad news: the treatment of the medium as a
continuum does not give any new results here. The situation somewhat changes at
relativistic velocities, at which such treatment provides noticeable corrections (called
density effects), in particular reducing the energy loss estimates.
10-34
Let me, however, skip these details and focus on a much more important effect
described by our formulas. Consider the dependence of the electric field components
on the impact parameter b, i.e. on the closest distance between the particle’s
trajectory and the field observation point. If κ2 > 0, then κ is real, and we can use,
in Eqs. (10.116) and (10.117), the asymptotic formula (2.158),
⎛ π ⎞1/2
K n(ξ ) → ⎜ ⎟ e−ξ , at ξ → ∞ , (10.127)
⎝ 2ξ ⎠
to conclude that the complex amplitudes Eω of both components Ex and Ey of the

electric field decrease with b exponentially. However, let us consider what happens at
frequencies where κ2(ω) < 0, i.e.
1 1 1
ε(ω)μ(ω) ≡ 2
< 2 < 2 ≡ ε0μ0 . (10.128)
v (ω ) u c
(This condition means that the particle’s velocity is larger than the phase velocity of
waves, at this particular frequency.) In these intervals the parameter κ(ω) is purely
imaginary49, so that the functions exp{κb} in the asymptotics (10.127) of Eqs.
(10.116) and (10.117) become just phase factors and the field components fall very
slowly:
1
Ex(ω) ∝ ∣Ey(ω)∣ ∝ . (10.129)
b1/2
This means that the Poynting vector drops as 1/b, so that its flux through a surface of
a round cylinder of radius b, with the axis on the particle trajectory (i.e. power flow
from the trajectory), does not depend on b. This is wave emission—the famous
Cherenkov radiation50.
The direction n of its propagation may be readily found by taking into account
that at large distances from the particle’s trajectory the emitted wave has to be
locally planar and transverse (n⊥E), so that the Cherenkov angle θ between the
vector n and the particle’s velocity u may be simply found from the ratio of the
electric field components—see figure 10.14a:
Ex
tan θ = − . (10.130)
Ey
49
Strictly speaking, the inequality κ2(ω) < 0 does not make sense for a medium with a complex product ε(ω)μ
(ω), and hence complex κ2(ω). However, in a typical medium where particles can propagate over substantial
distances, the imaginary part of the product ε(ω)μ(ω) does not vanish only in very limited frequency intervals,
much narrower than the intervals which we are now discussing—please have one more look at figure 7.5.
50
This radiation was observed experimentally by P Cherenkov (in older Western texts, ‘Čerenkov’) in 1934,
with the observations explained by I Frank and E Tamm in 1937. Note, however, that the effect was predicted
theoretically as early as in 1889 by the same O Heaviside, whose name was mentioned in this course so many
times—and whose genius I believe is still underappreciated.
10-35
Figure 10.14. The Cherenkov radiation’s propagation angle θ and (b) its interpretation.
This ratio may be calculated by plugging the asymptotic formula (10.127) into
Eqs. (10.116) and (10.117) and calculating their ratio:
Ex iκu 1/2
⎛ u2 ⎞1/2
tan θ = − = = [ε(ω)μ(ω)u − 1] = ⎜ 2
2
− 1⎟ , (10.131a )
Ey ω ⎝ v (ω ) ⎠
so that
v (ω )
cos θ = < 1. (10.131b)
u
Remarkably, this direction does not depend on the emission time tret, so that the
radiation of frequency ω, at each instant, forms a hollow cone led by the particle.
This simple result allows an evident interpretation (figure 10.14b): the cone is just the
set of all observation points that may be reached by ‘signals’ propagating with the
speed v(ω) < u from all previous points of the particle’s trajectory.
This phenomenon is closely related to the so-called Mach cone in fluid dynamics,
except that in the Cherenkov radiation there is a separate cone for each frequency (of
the range in which v(ω) < u): the smaller the ε(ω)μ(ω) product (i.e. the larger is wave
velocity v(ω) = 1/[ε(ω)μ(ω)]1/2) and the broader the cone, the earlier the correspond-
ing ‘shock wave’ arrives at an observer. Please note that the Cherenkov radiation is a
unique radiative phenomenon: it takes place even if a particle moves without
acceleration, and (in agreement with our analysis in section 10.2), is impossible in the
free space, where v = c is always larger than u.
The intensity of the Cherenkov radiation intensity may be also readily found by
plugging the asymptotic expression (10.127), with imaginary κ, into Eq. (10.123).
The result is
dE ⎛ Ze ⎞2 ⎛ 2 ⎞
−
dx
≈⎜ ⎟
⎝ 4π ⎠ ∫v(ω)<u ω ⎜⎝1 − v u(ω2 ) ⎟⎠ dω. (10.132)
For non-relativistic particles (u ≪ c), the Cherenkov radiation condition u > v(ω)
may be fulfilled only in relatively narrow frequency intervals where the product ε(ω)
μ(ω) is very large (usually, due to optical resonance peaks of the electric permittivity
—see figure 7.5 and its discussion). In this case the emitted light consists of a few
nearly monochromatic components. In contrast, if the condition u > v(ω), i.e. u2/ε(ω)
μ(ω) > 1 is fulfilled in a broad frequency range (as it is for ultra-relativistic particles
in condensed media), the radiated power is clearly dominated by higher frequencies
10-36
Figure 10.15. The Cherenkov radiation glow from the Advanced Test Reactor of the Idaho National
Laboratory in Arco, ID. (Adapted from http://en.wikipedia.org/wiki/Cherenkov_radiation under the
Creative Commons’ CC-BY-SA-2.0 license.)
of the range—hence the famous bluish color of the Cherenkov radiation glow from
water nuclear reactors—see figure 10.15.
The Cherenkov radiation is broadly used for the detection of radiation in high
energy experiments for particle identification and speed measurement (since it is easy
to pass the particles through layers of various density and hence of various dielectric
constant values)—for example, in the so-called Ring Imaging Cherenkov (RICH)
detectors that have been designed for the DELPHI experiment51 at the Large
Electron–Positron Collider (LEP) in CERN.
A little bit counter-intuitively, the formalism described in this section is also very
useful for the description of an apparently rather different effect—the so-called
transition radiation that takes place when a charged particle crosses a border
between two media52. The effect may be understood as the result of the time
dependence of the electric dipole formed by the moving charge q and its mirror
image qʹ in the counterpart medium—see figure 10.16. In the non-relativistic limit,
this effect allows a straightforward description combining the electrostatics picture
of section 3.4 (see figure 3.9 and its discussion) and Eq. (8.27), corrected for the
media polarization effects. However, if the particle’s velocity u is comparable with
51
See, e.g. http://delphiwww.cern.ch/offline/physics/delphi-detector.html. For a broader view at radiation
detectors (including the Cherenkov ones), the reader may be referred, for example, to the classical text [8] and a
newer treatment [9].
52
The effect was predicted theoretically in 1946 by V Ginzburg and I Frank, and only later observed
experimentally.
10-37
Figure 10.16. The physics of transition radiation.
the phase velocity of waves in either medium, the adequate theory of the transition
radiation becomes very close to that of the Cherenkov radiation.
In comparison with the Cherenkov radiation, the transition radiation is rather
weak and its practical use (mostly for the measurement of the relativistic factor γ, to
which the radiation intensity is proportional) requires multi-layered stacks53. In
these systems, the radiation emitted at sequential borders may be coherent, and the
system’s physics becomes close to that of the undulators discussed in section 10.4.
10.6 Radiation’s back-action

An attentive reader could notice that so far our treatment of charged particle
dynamics has never been fully self-consistent. Indeed, in section 9.6 we have
analyzed the particle’s motion in various external fields, ignoring the fields radiated
by the particle itself, while in section 8.2 and earlier in this chapter these fields have
been calculated (admittedly, just for a few simple cases) but, again, their back-action
on the emitting particle has been ignored. Only in few cases have we taken the back
effects of the radiation implicitly, via energy conservation. However, even in these
cases, the near-field components of the fields, such as the first term in Eq. (10.19),
which affect the moving particle most, have been ignored.
At the same time, it is clear that in sharp contrast to electrostatics, the interaction
of a moving point charge with its own field cannot be always ignored. As the
simplest example, if an electron is made to fly through a resonant cavity, thus
inducing oscillations in it, and then is forced (say, by an appropriate static field) to
return to it before the oscillations have decayed, its motion will certainly be affected
by the oscillating fields, just as if they had been induced by another source. There is
no conceptual problem with applying the Maxwell theory to such ‘field-particle
rendezvous’ effects; moreover, it is the basis of the engineering design of such
vacuum electron devices as klystrons, magnetrons, and undulators.
A problem arises only when no clear ‘rendezvous’ point is enforced by boundary
conditions, so that the most important self-field effects are at R ≡ ∣r − rʹ∣→ 0, the most
evident example being the radiation of a particle in free space, described earlier in this
chapter. We already know that such radiation takes away a part of the charge’s kinetic
53
See, e.g. section 5.3 in [9].
10-38
energy, i.e. has to cause its deceleration. One should wonder, however, whether such
self-action effects might be described in a more direct, non-perturbative way.
As a first attempt, let us try a phenomenological approach based on the already
derived formulas for radiation power P . For the sake of simplicity, let us consider a
non-relativistic point charge q in free space, so that P is described by Eq. (8.27), with
the electric dipole moment’s derivative over time equal to qu:
Z0q 2 2 2 q2 2
P= 2
u̇ ≡ 3 u̇ . (10.133)
6πc 3c 4πε0
The most naïve approach would be to write the equation of the particle’s motion in
the form
mu̇ = Fext + Fself , (10.134)
and try to calculate the radiation back-action force Fself by requiring its instant
power, −Fself · u, to be equal to P . However, this approach (say, for a 1D motion)
would give a very unnatural result,
u 2̇
Fself ∝ , (10.135)
u
that might diverge at some points of the particle’s trajectory. This failure is clearly
due to the retardation effect: as the reader may recall, Eq. (10.133) results from the
analysis of radiation fields in the far zone, i.e. at large distances R from the particle
(e.g. from the second term in Eq. (10.19)) when the non-radiative first term (which is
much larger at small distances, R → 0) is ignored.
Before exploring the effects of this term, let us, however, make one more try at
Eq. (10.133), considering its average effect on some periodic motion of the particle.
(A possible argument for this step is that at the periodic motion the retardation
effects may perhaps be averaged out—just at the transfer from Eq. (8.27) to
Eq. (8.28).) To calculate the average, let us write
T
1
u 2̇ ≡
T
∫0 u̇ ⋅ u̇ dt , (10.136)
and integrate this identity over the motion period T by parts:
2 q2 2 q2 1 ⎛ T − T ⎞
P =
3c 3 4πε0
( u )
̇ 2
= ⎜u̇ ⋅ u
3c 3 4πε0T ⎝ 0
∫0 ü ⋅ u dt⎟
⎠
. (10.137)
T
1 2 q2
=−
T
∫0 3c 3 4πε0
ü ⋅ u dt
One the other hand, the back-action force would give
10-39
1 T
P = − ∫ Fself ⋅ u dt . (10.138)
T 0
These two averages coincide if54
2 q2
Fself = ü . (10.139)
3c 3 4πε0
This is the so-called Abraham–Lorentz force of self-action. Before going after a more
serious derivation of this formula, let us estimate its scale, representing Eq. (10.139) as
2 q2
Fself = mτ u,̈ with τ ≡ , (10.140)
3mc 3 4πε0
where the constant τ evidently has the dimension of time. Recalling the definition
(8.41) of the classical radius rc of the particle, Eq. (10.140) for τ may be rewritten as
2 rc
τ= . (10.141)
3c
For the electron, τ is of the order of 10−23 s, so that the right-hand side of
Eq. (10.140) is very small. This means that in most cases the Abrahams–Lorentz
force is either negligible or leads to the same results as the perturbative treatments of
energy loss we have used earlier in this chapter.
However, Eq. (10.140) brings some unpleasant surprises. For example, let us
consider a 1D oscillator with the own frequency ω0. For it, Eq. (10.134), with the
back-action force given by Eq. (10.140), takes the form
mx ̈ + mω 02x = mτx ⃛ . (10.142)
Looking for the solution to this linear differential equation in the usual exponential
form, x(t) ∝ exp{λt}, we obtain the following characteristic equation,
λ2 + ω 02 = τλ3. (10.143)
It may look like that for any ‘reasonable’ value of ω0 ≪ 1/τ ∼ 1023 s−1, the right-hand
side of this nonlinear algebraic equation may be treated as a perturbation. Indeed,
looking for its solutions in the natural form λ± = ±iω0 + λʹ, with ∣λʹ∣≪ ω0, expanding
both parts of Eq. (10.143) in the Taylor series in the small parameter λʹ and keeping
only the terms linear in λʹ, we obtain
54
Just for the reader’s reference, this formula may be readily generalized to the relativistic case:
2 q 2 ⎡ d 2p α p α ⎛ dpβ dp β ⎞⎤
α =
Fself ⎢ + ⎜ ⎟⎥,
3mc3 4πε 0 ⎢⎣ dτ 2 (mc)2 ⎝ dτ dτ ⎠⎥⎦
the so-called Abraham–Lorentz–Dirac force.
10-40
ω 02τ
λ′ ≈ − . (10.144)
2
This means that the energy of free oscillations decreases in time as exp{2λʹt} = exp
{−ω02τ t}; this is exactly the radiative damping analyzed earlier. However,
Eq. (10.143) is deceptive; it has the third root corresponding to unphysical,
exponentially growing (so-called run-away) solutions. It is easiest to see this for a
free particle, with ω0 = 0. Then Eq. (10.143) becomes the very simple
λ2 = τλ3, (10.145)
and it is easy to find all its three roots explicitly: λ1 = λ2 = 0 and λ3 = 1/τ. While the
first two roots correspond the values λ± found earlier, the last one describes an
exponential (and extremely rapid!) acceleration.
In order to remove this artifact, let us try to develop a self-consistent approach to
the back-action effects, taking into account the near-field terms of particle fields. For
that, we need to somehow overcome the divergence of Eqs (10.10) and (10.19) at
R → 0. The most reasonable way to do this is to spread the particle’s charge over a
ball of radius a, with a spherically symmetric (but not necessarily constant) density
ρ(r), and in the end of the calculations trace the limit a → 0.55 Again sticking to the
non-relativistic case (so that the magnetic component of the Lorentz force is not
important), we should calculate
Fself = ∫V ρ(r)E(r, t )d 3r, (10.146)
where the electric field is that of the charge itself, with the field of any elementary
charge dq = ρ(r)d3r described by Eq. (10.19).
In order to make analytical calculations doable, we need to make assumption
a ≪ rc, treat the ratio R/rc ∼ a/rc as a small parameter, and expand the resulting the
right-hand side of Eq. (10.146) into the Taylor series in small R. This procedure yields
∞
2 1 ( −1)n d n+1u
Fself =− ∑
3 4πε0 n = 0 c n+2n! dt n+1
∫V d 3r∫V d 3r′ρ(r)Rn−1ρ(r′). (10.147)
The distance R cancels only in the term with n = 1,

2
2 ü
F1 =
3c 3 4πε0
∫V d 3r∫V d 3r′ρ(r)ρ(r′) ≡ 32c3 4qπε0 ü, (10.148)
showing that we have recovered (now in an apparently legitimate fashion)

Eq. (10.139) for the Abrahams–Lorentz force. One could argue that in the limit
a → 0 the terms higher in R ∼ a (with n > 1) could be ignored. However, we have to
55
Note: this operation cannot be interpreted as describing a quantum spread due to the finite extent of a point
particle’s wavefunction. In quantum mechanics, parts of wavefunction of the same charged particle do not
interact with each other!
10-41
notice that the main contribution into series (10.147) is not described by Eq. (10.148)
for n = 1, but is given by the larger term with n = 0:
2 1 u̇ ρ(r )ρ(r′)
F0 = − 2
3 4πε0 c V
∫
d 3r d 3r′
V
∫R
(10.149)
4 u̇ 1 1 ρ( r )ρ(r′) 4
≡− 2
3 c 4πε0 2 V V
∫
d 3r d 3r′ ∫
R 3c
̇ ,
≡ − 2 uU
where U is the electrostatic energy (1.59) of the static charge self-interaction. This
term may be interpreted as the inertial ‘force’56 (−mefa) with the following effective
electromagnetic mass:
4U
m ef = . (10.150)
3 c2
This is the famous (or rather infamous) 4/3 problem that does not allow one to
interpret the electron’s mass as that of its electric field. An (admittedly, rather
formal) resolution of this paradox is possible only in quantum electrodynamics with
its renormalization techniques—beyond the framework of this course. Note,
however, that these issues are only important for motions with frequencies of the
order of 1/τ ∼ 1023 s−1, i.e. at energies E ∼ ℏ/τ ∼ 108 eV, while other quantum
electrodynamics effects may be observed at much lower frequencies, starting from
∼1010 s−1. Hence the 4/3 problem is by no means the only motivation for the transfer
from classical to quantum electrodynamics.
However, the reader should not think that his or her time spent on this course has
been lost: quantum electrodynamics incorporates virtually all classical electro-
dynamics results and the basic transition to it is surprisingly straightforward57.
So, I look forward to welcoming the readers to the next, quantum-mechanics part of
this series.
10.7 Problems
Problem 10.1. Derive Eqs. (10.10) from Eqs. (10.1) by a direct (but careful!)
integration.
Problem 10.2. Derive the radiation-related parts of Eqs. (10.19) and (10.20) from the
Liénard–Wiechert potentials (10.10) by direct differentiation.
Problem 10.3. A point charge q that was in a stationary position on a circle of radius
R is carried over, along the circle, to the opposite position on the same diameter (see
the figure below) as fast as only physically possible, and then is kept steady at this
new position. Calculate and sketch the time dependence of its electric field E at the
center of the circle.
56
57
See, e.g. Part QM chapter 9.
10-42
Problem 10.4. Express the total power of electromagnetic radiation by a relativistic

particle with electric charge q and rest mass m, moving with velocity u, via the
external Lorentz force F exerted on the particle.
Problem 10.5. A relativistic particle with the rest mass m and electric charge q, initially
at rest, is accelerated by a constant force F until it reaches a certain velocity u, and then
moves by inertia. Calculate the total energy radiated during the acceleration.
Problem 10.6. Calculate

(i) the instantaneous power, and
(ii) the power spectrum
of the radiation emitted into a unit solid angle by a relativistic particle with charge
q, performing 1D harmonic oscillations with frequency ω0 and displacement
amplitude a.
Problem 10.7. Analyze the polarization and the spectral contents of the synchrotron
radiation propagating in the direction perpendicular to the particle’s rotation plane.
How do the results change if not one, but N > 1 similar particles move around the
circle, at equal angular distances?
Problem 10.8. Calculate and analyze the time dependence of the energy of a charged
relativistic particle performing synchrotron motion in a constant and uniform
magnetic field B, and hence emitting the synchrotron radiation. Qualitatively,
what is the particle’s trajectory?
Hint: You may assume that the energy loss is relatively slow (−dE /dt ≪ ωcE ), but
should spell out the condition of validity of this assumption.
Problem 10.9. Analyze the polarization of the synchrotron radiation propagating

within the particle’s rotation plane.
Problem 10.10.* The basic quantum theory of radiation shows that the electric dipole
radiation by a particle is allowed only if the change of its angular momentum’s
magnitude L at the transition is of the order of the Planck’s constant ℏ.
(i) Estimate the change of L of an ultra-relativistic particle due to its emission of a

typical single photon of the synchrotron radiation.
(ii) Does the quantum mechanics forbid such a radiation? If not, why?
10-43
Problem 10.11. A relativistic particle moves along axis z, with velocity uz,
through an undulator—a system of permanent magnets providing (in the
simplest model) a perpendicular magnetic field, whose distribution near the
axis z is sinusoidal58:
B = nyB0 cos k 0z .
Assuming that the field is so weak that it causes only very small deviations of the
particle’s trajectory from the straight line, calculate the angular distribution of the
resulting radiation. What condition does this assumption impose on the system’s
parameters?
Problem 10.12. Discuss possible effects of the interference of the undulator radiation
from different periods of its static field distribution. In particular, calculate the
angular positions of the power density maxima.
Problem 10.13. An electron, launched directly toward a plane surface of a perfect

conductor, is instantly absorbed by it at the collision. Calculate the angular
distribution and the frequency spectrum of the electromagnetic waves radiated at
this collision, if the initial kinetic energy T of the particle is much larger than the
conductor’s workfunction ψ.59 Is your result valid near the conductor’s surface?
Problem 10.14. A relativistic particle, with the rest mass m and electric charge q, flies
ballistically with velocity u, by an immobile point charge qʹ, with an impact
parameter b so large that the deviations of its trajectory from the straight line are
negligible. Calculate the total energy loss due to the electromagnetic radiation
during the passage. Formulate the conditions of validity of your result.
References
[1] Koch E et al (eds) 1983–1991 Handbook on Synchrotron Radiation vols 1–5 (Amsterdam:
North-Holland)
[2] Hofmann A 2007 The Physics of Synchrotron Radiation (Cambridge: Cambridge University
Press)
[3] Luchini P and Motz H 1990 Undulators and Free-electron Lasers (Oxford: Oxford University
Press)
[4] Salin E et al 2000 The Physics of Free Electron Lasers (Berlin: Springer)
[5] Lewenstein M et al 1994 Phys. Rev. A 49 2117
[6] Jackson J 1999 Classical Electrodynamics 3rd edn (New York: Wiley)
[7] Heitler W 1954 The Quantum Theory of Radiation 3rd edn (Oxford: Oxford University Press)
[8] Knoll G 2010 Radiation Detection and Measurement 4th edn (New York: Wiley)
[9] Kleinknecht K 1999 Detectors for Particle Radiation (Cambridge: Cambridge University Press)
58
As the Maxwell equation for ∇ × H shows, such a field distribution cannot be created in any non-vanishing
volume of free space. However, it may be created on a line—e.g. on a particle’s straight trajectory.
59
See section 2.9, in particular figure 2.27a.
10-44
IOP Publishing
Lecture notes
Appendix A
Selected mathematical formulas
This appendix lists selected mathematical formulas that are used in this lecture
course series, but not always remembered by students (and some instructors:-).
A.1 Constants
• Euclidean circle’s length-to-diameter ratio:
π = 3.141 592 653 …; π 1/2 ≈ 1.77. (A.1)
• Natural logarithm base:
⎛ 1 ⎞n
e ≡ lim n→∞ ⎜1 + ⎟ = 2.718 281 828 …; (A.2a )
⎝ n⎠
from that value, the logarithm base conversion factors are as follows (ξ > 0):
ln ξ log10ξ 1
= ln 10 ≈ 2.303, = ≈ 0.434. (A.2b)
log10ξ ln ξ ln 10
• The Euler (or ‘Euler–Mascheroni’) constant:
⎛ 1 1 1 ⎞
γ ≡ lim n→∞ ⎜1 + + + … − ln n⎟ = 0.577 156 649 0 …;
⎝ 2 3 n ⎠ (A.3)
γ
e ≈ 1.781.
A.2 Combinatorics, sums, and series

(i) Combinatorics
• The number of different permutations, i.e. ordered sequences of k elements
selected from a set of n distinct elements (n ⩾ k), is
n n!
Pk ≡ n ⋅ (n − 1) ⋯ (n − k + 1) = ; (A.4a )
(n − k )!
doi:10.1088/978-0-7503-1404-6ch11 A-1 ª Konstantin K Likharev 2018

in particular, the number of different permutations of all elements of the set

(n = k) is
k
Pk = k ⋅ (k − 1)⋯2 ⋅ 1 = k ! . (A.4b)
• The number of different combinations, i.e. unordered sequences of k

elements from a set of n ⩾ k distinct elements, is equal to the binomial
coefficient
n
Pk n!
n
Ck ≡ (kn ) ≡ k
Pk
=
k ! (n − k )!
. (A.5)
In an alternative, very popular ‘ball/box language’, nCk is the number of

different ways to put in a box, in an arbitrary order, k balls selected from n
distinct balls.
• A generalization of the binomial coefficient notion is the multinomial
coefficient,
l
n n!
C k1,k2,…kl ≡
k1! k2!…kl !
, with n = ∑ kj , (A.6)
j=1
which, in the standard mathematical language, is a number of different

permutations in a multiset of l distinct element types from an n-element set
which contains kj ( j = 1, 2,…l ) elements of each type. In the ‘ball/box
language’, the coefficient (A.6) is the number of different ways to distribute
n distinct balls between l distinct boxes, each time keeping the number (kj) of
balls in the jth box fixed, but ignoring their order inside the box. The
binomial coefficient nCk (A.5), is a particular case of the multinomial
coefficient (A.6) for l = 2 - counting the explicit box for the first one, and
the remaining space for the second box, so that if k1 ≡ k, then k2 = n − k.
• One more important combinatorial quantity is the number Mn(k) of ways to
place n indistinguishable balls into k distinct boxes. It may be readily
calculated from Eq. (A.5) as the number of different ways to select (k − 1)
partitions between the boxes in an imagined linear row of (k − 1 + n)
‘objects’ (balls in the boxes and partitions between them):
n−1+k (k − 1 + n )!
Mn(k ) = Ck − 1 ≡ . (A.7)
(k − 1) ! n!
(ii) Sums and series

• Arithmetic progression:
n
n(r + nr )
r + 2r + ⋯ + nr ≡ ∑ kr = 2
; (A.8a )
k=1
A-2
in particular, at r = 1 it is reduced to the sum of n first natural numbers:

n
n(n + 1)
1+2+⋯+n≡ ∑k = 2
. (A.8b)
k=1
• Sums of squares and cubes of n first natural numbers:

n
n(n + 1)(2n + 1)
12 + 22 + ⋯ + n 2 ≡ ∑k2 = 6
; (A.9a )
k=1
n
n 2(n + 1)2
13 + 23 + ⋯ + n3 ≡ ∑k3 = 4
. (A.9b)
k=1
• The Riemann zeta function:

∞
1 1 1
ζ (s ) ≡ 1 + s
+ s +⋯≡ ∑ ; (A.10a )
2 3 k=1
ks
the particular values frequently met in applications are

⎛3⎞ π2 ⎛5 ⎞
ζ⎜ ⎟ ≈ 2.612, ζ(2) = , ζ⎜ ⎟ ≈ 1.341,
⎝2⎠ 6 ⎝2⎠
(A.10b)
π4
ζ(3) ≈ 1.202, ζ(4) = , ζ(5) ≈ 1.037.
90
• Finite geometric progression (for real λ ≠ 1):

n−1
1 − λn
1 + λ + λ 2 + ⋯ + λ n −1 ≡ ∑ λk = 1−λ
; (A.11a )
k=0
in particular, if λ < 1, the progression has a finite limit at n → ∞ (called the

2
geometric series):
n−1 ∞
1
lim n→∞∑ λ k = ∑ λk = 1 . (A.11b)
k=0 k=0
−λ
• Binomial sum (or the ‘binomial theorem’):

n
(1 + a )n = ∑ nCka k , (A.12)
k=0
where nCk are the binomial coefficients defined by Eq. (A.5).
A-3
• The Stirling formula:

1 1 1
lim n→∞ ln (n!) = n(ln n − 1) + ln(2πn ) + − + …; (A.13)
2 12n 360n3
for most applications in physics, the first term1 is sufficient.
• The Taylor (or ‘Taylor–Maclaurin’) series: for any infinitely differentiable
function f (ξ):
df 1 d 2f 2
lim ξ̃ → 0f (ξ + ξ˜ ) = f (ξ ) + (ξ ) ξ˜ + 2
(ξ ) ξ˜ + ⋯
dξ 2! dξ
∞ (A.14a )
1 d kf k
=∑ k
(ξ ) ξ˜ ;
k=0
k! dξ
note that for many functions this series converges only within a limited,
sometimes small range of deviations ξ̃ . For a function of several arguments,
f(ξ1,ξ2,…,ξN), the first terms of the Taylor series are
lim ξ̃ → 0f (ξ1 + ξ˜1, ξ2 + ξ˜2, ⋯) = f (ξ1, ξ2, ⋯)

k
N
∂f
+∑ (ξ1, ξ2, ⋯) ξ˜k
∂ξ
k=1 k
(A.14b)
N
1 ∂ 2f ˜ ˜
+ ∑
2! k, k ′= 1 ∂kξ ∂ξk ′
ξkξk ′ + ⋯
• The Euler–Maclaurin formula, valid for any infinitely differentiable function

f(ξ):
1 1 ⎡ df df ⎤
n−1 n
1
∑ f (k ) = ∫0 f (ξ )dx −
2
[f (n ) − f (0)] + ⋅ ⎢ (n ) −
6 2! ⎣ dξ
(0)⎥
dξ ⎦
n=0
1 1 ⎡ d 3f d 3f ⎤
+ ⋅ ⎢ 3 (n ) − 3 (0)⎥ (A.15a )
30 4! ⎣ dξ dξ ⎦
1 1 ⎡ d 5f d 5f ⎤
− ⋅ ⎢ 5 (n ) − 5 (0)⎥ + ⋯;
42 6! ⎣ dξ dξ ⎦
the coefficients participating in this formula are the so-called Bernoulli
numbers2:
1 1 1
B1 = , B2 = , B3 = 0, B4 = , B5 = 0,
2 6 30 (A.15b)
1 1
B6 = , B7 = 0, B8 = , ⋯
42 30
1
Actually, this leading term was derived by A de Moivre in 1733, before J Stirling’s work.
2
Note that definitions of Bk (or rather their signs and indices) vary even in the most popular handbooks.
A-4
A.3 Basic trigonometric functions

• Trigonometric functions of the sum and the difference of two arguments3:
cos (a ± b) = cos a cos b ∓ sin a sin b , (A.16a )
sin (a ± b) = sin a cos b ± cos a sin b . (A.16b)

• Sums of two functions of arbitrary arguments:
a+b b−a
cos a + cos b = 2 cos cos , (A.17a )
2 2
a+b b−a
cos a − cos b = 2 sin sin , (A.17b)
2 2
a±b ±b − a
sin a ± sin b = 2 sin cos . (A.17c )
2 2
• Trigonometric function products:

2 cos a cos b = cos(a + b) + cos(a − b), (A.18a )
2 sin a cos b = sin(a + b) + sin(a − b), (A.18b)
2 sin a sin b = cos(a − b) − cos(a + b); (A.18c )

For the particular case of equal arguments, b = a, these three formulas yield
the following expressions for the squares of trigonometric functions, and their
product:
1 1
cos2 a = (1 + cos 2a ), sin a cos a = sin 2a ,
2 2 (A.18d )
1
sin2 a = (1 − cos 2a ).
2
• Cubes of trigonometric functions:

3 1 3 1
cos3 a = cos a + cos 3a , sin3 a = sin a − sin 3a . (A.19)
4 4 4 4
3
I am confident that the reader is quite capable of deriving the relations (A.16) by representing the exponent in
the elementary relation ei(a ± b) = eiae±ib as a sum of its real and imaginary parts, Eqs. (A.18) directly from
Eqs. (A.16), and Eqs. (A.17) from Eqs. (A.18) by variable replacement; however, I am still providing these
formulas to save his or her time. (Quite a few formulas below are included because of the same reason.)
A-5
• Trigonometric functions of a complex argument:

sin (a + ib) = sin a cosh b + i cos a sinh b ,
(A.20)
cos (a + ib) = cos a cosh b − i sin a sinh b .
• Sums of trigonometric functions of n equidistant arguments:

n
sin kξ = sin ⎛⎜ n + 1 ξ⎞⎟ sin ⎛ n ξ⎞ ⎛ξ ⎞
∑
k=1
{cos } {cos}⎝ 2 ⎠ ⎝ 2 ⎠ ⎜ ⎟ sin ⎜ ⎟ .
⎝2⎠
(A.21)
A.4 General differentiation

• Full differential of a product of two functions:
d (fg ) = (df )g + f (dg ). (A.22)
• Full differential of a function of several independent arguments, f(ξ1, ξ2,…, ξn):

n
∂f
df = ∑∂ dξk . (A.23)
ξ
k=1 k
• Curvature of the Cartesian plot of a 1D function f(ξ):

1 d 2f / dξ 2
κ≡ = . (A.24)
R [1 + (df / dξ )2 ]3/2
A.5 General integration

• Integration by parts - immediately follows from Eq. (A.22):
g (B ) B f (B )
∫g(A) f dg = fg
A
− ∫f (A) g df . (A.25)
• Numerical (approximate) integration of 1D functions: the simplest trapezoi-

dal rule,
b ⎡ ⎛ h⎞ ⎛ 3h ⎞ ⎛ h ⎞⎤
∫a f (ξ )dξ ≈ h⎢f ⎜a + ⎟ + f ⎜a + ⎟ + ⋯ + f ⎜b − ⎟⎥
⎣ ⎝ 2⎠ ⎝ 2⎠ ⎝ 2 ⎠⎦
(A.26)
⎛N
h ⎞ b−a
= h∑f ⎜a − + nh⎟ , h≡ .
n=1
⎝ 2 ⎠ N
A-6
has relatively low accuracy (error of the order of (h3/12)d2f/dξ2 per step), so
that the following Simpson formula,
b
h
∫a f (ξ )dξ ≈
3
[f (a) + 4f (a + h) + 2f (a + 2h) + ⋯ + 4f (b − h) + f (b)] ,
(A.27)
b−a
h≡ ,
2N
whose error per step scales as (h5/180)d4f/dξ4, is used much more frequently4.
A.6 A few 1D integrals5

(i) Indefinite integrals:
• Integrals with (1 + ξ2)1/2:
∫ (1 + ξ2)1/2dξ = ξ2 (1 + ξ2)1/2 + 12 ln∣ξ + (1 + ξ2)1/2∣, (A.28)
∫ (1 +dξξ2)1/2 = ln∣ξ + (1 + ξ 2 )1/2 ∣ , (A.29a )
∫ (1 +dξξ2)3/2 =
ξ
(1 + ξ 2 )
1/2
. (A.29b)
• Miscellaneous indefinite integrals:
∫ ξ(ξ2 + 2daξξ − 1)1/2 = cos−1

aξ − 1
ξ (a 2 + 1)1/2
, (A.30a )
2
2ξ sin 2ξ + cos 2ξ − 2ξ 2 − 1
∫ (sin ξ −ξξ5 cos ξ) dξ =
8ξ 4
, (A.30b)
⎡ ⎤
∫ a + bdξcos ξ = (a 2 −2b2)1/2 tan−1 ⎢⎣ (a(2a−−bb2))1/2 tan ξ2 ⎥⎦, (A.30c )
for a 2 > b 2 .
∫ 1 +dξξ2 = tan−1 ξ. (A.30d )
4
Higher-order formulas (e.g. the Bode rule), and other guidance including ready-for-use codes for computer
calculations may be found, for example, in the popular reference texts by W H Press et al [1]. In addition, some
advanced codes are used as subroutines in the software packages listed in the same section. In some cases, the
Euler–Maclaurin formula (A.15) may also be useful for numerical integration.
5
A powerful (and free) interactive online tool for working out indefinite 1D integrals is available at http://
integrals.wolfram.com/index.jsp.
A-7
(ii) Semi-definite integrals:

• Integrals with 1/(eξ ±1):
∞
dξ
∫a ξ
e +1
= ln (1 + e−a ), (A.31a )
∞
∫a>0 e ξd−ξ 1 = ln
1
1 − e −a
. (A.31b)
(iii) Definite integrals:

• Integrals with 1/(1 + ξ2):6
∞
dξ π
∫0 1+ξ 2
= ,
2
(A.32a )
∞
dξ
∫0 ( 1 + ξ 2 )3/2
= 1; (A.32b)
more generally,
∞
dξ π (2n − 3) !! π 1 ⋅ 3 ⋅ 5 … (2n − 3)
∫0 ( 1 + ξ 2 n
)
=
2 (2n − 2) !!
≡
2 2 ⋅ 4 ⋅ 6 … (2n − 2)
,
(A.32c )
for n = 2, 3, …
• Integrals with (1 − ξ2n)1/2:

1
dξ π 1/2 ⎛ 1 ⎞ ⎛ n + 1⎞
∫0 = Γ⎜ ⎟ Γ⎜ ⎟, (A.33a )
2n 1/2
(1 − ξ ) 2n ⎝ 2n ⎠ ⎝ 2n ⎠
1
π 1/2 ⎛ 1 ⎞ ⎛ 3n + 1 ⎞
∫0 1/2
( 1 − ξ 2 n ) dξ = Γ⎜ ⎟ Γ⎜ ⎟, (A.33b)
4n ⎝ 2n ⎠ ⎝ 2n ⎠
where Γ(s) is the gamma-function, which is most often defined (for Re s > 0)
by the following integral:
∞
∫0 ξ s−1e−ξ dξ = Γ(s ). (A.34a )
6
Eq. (A.32a) follows immediately from Eq. (A.30d), and Eq. (A.32b) from Eq. (A.29b)—a couple more
examples of the (intentional) redundancy in this list.
A-8
The key property of this function is the recurrence relation, valid for any
s ≠ 0, −1, −2,…:
Γ(s + 1) = s Γ(s ). (A.34b)
Since, according to Eq. (A.34a), Γ(1) = 1, Eq. (A.34b) for non-negative
integers takes the form
Γ(n + 1) = n ! , for n = 0, 1, 2, ⋯ (A.34c )
(where 0! ≡ 1). Because of this, for integer s = n + 1 ⩾ 1, Eq. (A.34a) is
reduced to
∞
∫0 ξ ne−ξdξ = n! . (A.34d )
Other frequently met values of the gamma-function are those for positive
semi-integer arguments:
⎛1 ⎞ ⎛3⎞ 1 ⎛5 ⎞ 1 3
Γ⎜ ⎟ = π 1/2, Γ⎜ ⎟ = π 1/2, Γ⎜ ⎟ = ⋅ π 1/2,
⎝2⎠ ⎝2⎠ 2 ⎝2⎠ 2 2
(A.34e )
⎛7⎞ 1 3 5
Γ⎜ ⎟ = ⋅ ⋅ π 1/2, ….
⎝2⎠ 2 2 2
• Integrals with 1/(eξ ±1):

∞
ξ s−1dξ
∫0 eξ + 1
= (1 − 21−s ) Γ(s )ζ(s ), for s > 0, (A.35a )
∞
ξ s−1dξ
∫0 eξ − 1
= Γ(s )ζ(s ), for s > 1, (A.35b)
where ζ(s) is the Riemann zeta-function—see Eq. (A.10). Particular cases:

for s = 2n,
∞
ξ 2n−1dξ 22n−1 − 1 2n
∫0 ξ
e +1
=
2n
π B2n, (A.35c )
∞
ξ 2n−1dξ (2π )2n
∫0 ξ
e −1
=
4n
B2n. (A.35d )
where Bn are the Bernoulli numbers—see Eq. (A.15). For the particular case
s = 1 (when Eq. (A.35a) yields uncertainty),
∞
dξ
∫0 eξ + 1
= ln 2. (A.35e )
A-9
• Integrals with exp{−ξ 2}:

∞
1 ⎛ s + 1⎞
∫0 Γ⎜ ⎟,
2
ξ se−ξ dξ = for s > −1; (A.36a )
2 ⎝ 2 ⎠
for applications the most important particular values of s are 0 and 2:
∞
1 ⎛ 1 ⎞ π 1/2
∫0 Γ⎜ ⎟ =
2
e − ξ dξ = , (A.36b)
2 ⎝2⎠ 2
∞
1 ⎛ 3 ⎞ π 1/2
∫0 Γ⎜ ⎟ =
2
ξ 2e−ξ dξ = , (A.36c )
2 ⎝2⎠ 4
though we will also run into the cases s = 4 and s = 6:
∞
1 ⎛ 5 ⎞ 3π 1/2
∫0 ξ 4e−ξ dξ =
2
Γ⎜ ⎟ =
2 ⎝2⎠ 8
,
(A.36d )
∞
1 ⎛ 7 ⎞ 15π 1/2
∫ 0
2
ξ 6e−ξ dξ = Γ⎜ ⎟ =
2 ⎝2⎠ 16
;
for odd integer values s = 2n + 1 (with n = 0, 1, 2,…), Eq. (A.36a) takes a

simpler form:
∞
1 n!
∫0 ξ 2n+1e−ξ dξ =
2
2
Γ(n + 1) = .
2
(A.36e )
• Integrals with cosine and sine functions:

∞ ∞ ⎛ π ⎞1/2
∫0 cos (ξ 2 ) dξ = ∫0 sin (ξ 2 ) dξ = ⎜ ⎟
⎝8⎠
. (A.37)
∞
cos ξ π −a
∫0 a2 + ξ2
dξ =
2a
e . (A.38)
∞⎛
sin ξ ⎞
2
π
∫0 ⎜
⎝ ξ ⎠
⎟ dξ = .
2
(A.39)
• Integrals with logarithms:

1
a + (1 − ξ 2 )1/2
∫0 ln
a − (1 − ξ 2 )1/2
dξ = π [a − (a 2 − 1)1/2 ] , for a ⩾ 1. (A.40)
1
1 + (1 − ξ )1/2
∫0 ln
ξ1/2
dξ = 1. (A.41)
A-10
• Integral representations of the Bessel functions of integer order:

1 +π
Jn(α ) =
2π
∫−π e i (α sin ξ−nξ )dξ,
∞ (A.42a )
so that e iα sin ξ = ∑ Jk(α )e ikξ ;
k =−∞
π
1
In(α ) =
π
∫0 e α cos ξ cos nξ dξ. (A.42b)
A.7 3D vector products

(i) Definitions:
• Scalar (‘dot-’) product:
3
a⋅b= ∑ajbj , (A.43)
j=1
where aj and bj are vector components in any orthogonal coordinate

system. In particular, the vector squared (the same as the norm squared):
3
a2 ≡ a ⋅ a = ∑a j2 ≡ a 2. (A.44)
j=1
• Vector (‘cross-’) product:
a × b ≡ n1(a2b3 − a3b2 ) + n2(a3b1 − a1b3) + n3(a1b2 − a2b1)

n1 n2 n3 (A.45)
= a1 a2 a3 ,
b1 b2 b3
where {nj} is the set of mutually perpendicular unit vectors7 along the
corresponding coordinate system axes8. In particular, Eq. (A.45) yields
a × a = 0. (A.46)
(ii) Corollaries (readily verified by Cartesian components):
• Double vector product (the so-called bac minus cab rule):
a × (b × c) = b(a ⋅ c) − c(a ⋅ b). (A.47)
7
Other popular notations for this vector set are {ej} and {r̂ j }.
8
It is easy to use Eq. (A.45) to check that the direction of the product vector corresponds to the well-known
‘right-hand rule’ and to the even more convenient corkscrew rule: if we rotate a corkscrew’s handle from the
first operand toward the second one, its axis moves in the direction of the product.
A-11
• Mixed scalar–vector product (the operand rotation rule):
a ⋅ (b × c) = b ⋅ (c × a) = c ⋅ (a × b). (A.48)
• Scalar product of vector products:

(a × b) ⋅ (c × d) = (a ⋅ c)(b ⋅ d) − (a ⋅ d)(b ⋅ c); (A.49a )
in the particular case of two similar operands (say, a = c and b = d), the last
formula is reduced to
(a × b)2 = (ab)2 − (a ⋅ b)2 . (A.49b)
A.8 Differentiation in 3D Cartesian coordinates

• Definition of the del (or ‘nabla’) vector-operator ∇:9
3
∂
∇≡ ∑ nj ∂ , (A.50)
j=1
rj
where rj is a set of linear and orthogonal (Cartesian) coordinates along

directions nj. In accordance with this definition, the operator ∇ acting on a
scalar function of coordinates, f(r),10 gives its gradient, i.e. a new vector:
3
∂f
∇f ≡ ∑nj ∂r ≡ grad f . (A.51)
j=1 j
• The scalar product of del by a vector function of coordinates (a vector field),

3
f(r) ≡ ∑nj f j (r), (A.52)
j=1
compiled formally following Eq. (A.43), is a scalar function—the divergence

of the initial function:
3
∂f j
∇⋅f≡ ∑ ≡ div f, (A.53)
j=1
∂rj
9
One can run into the following notation: ∇ ≡ ∂/∂r, which is convenient is some cases, but may be misleading in
quite a few others, so it will be not used in these notes.
10
In this, and four next sections, all scalar and vector functions are assumed to be differentiable.
A-12
while the vector product of ∇ and f, formed in a formal accordance with

Eq. (A.45), is a new vector - the curl (in European tradition, called rotor and
denoted rot) of f:
n1 n2 n3
∂ ∂ ∂ ⎛ ∂f ∂f ⎞ ⎛ ∂f ∂f ⎞
∇×f≡ = n1⎜ 3 − 2 ⎟ + n2⎜ 1 − 3 ⎟
∂r1 ∂r2 ∂r3 ⎝ ∂r2 ∂r3 ⎠ ⎝ ∂r3 ∂r1 ⎠
f1 f2 f3 (A.54)
⎛ ∂f ∂f ⎞
+ n3⎜ 2 − 1 ⎟ ≡ curl f.
⎝ ∂r1 ∂r2 ⎠
• One more frequently met ‘product’ is (f·∇)g, where f and g are two arbitrary
vector functions of r. This product should be also understood in the sense
implied by Eq. (A.43), i.e. as a vector whose jth Cartesian component is
3
∂g
[(f ⋅ ∇) g] j = ∑ f j′ ∂r j . (A.55)
j ′= 1 j′
A.9 The Laplace operator ∇2 ≡ ∇ · ∇

• Expression in Cartesian coordinates—in the formal accordance with
Eq. (A.44):
3
∂2
∇2 = ∑ . (A.56)
∂r 2
j=1 j
• According to its definition, the Laplace operator acting on a scalar function

of coordinates gives a new scalar function:
3
∂ 2f
∇2 f ≡ ∇ ⋅ (∇f ) = div(grad f ) = ∑ . (A.57)
∂r 2
j=1 j
• On the other hand, acting on a vector function (A.52), the operator ∇2 returns
another vector:
3
∇2 f = ∑nj∇2 f j . (A.58)
j=1
Note that Eqs. (A.56)–(A.58) are only valid in Cartesian (i.e. orthogonal and
linear) coordinates, but generally not in other (even orthogonal) coordinates—
see, e.g. Eqs. (A.61), (A.64), (A.67) and (A.70) below.
A-13
A.10 Operators ∇ and ∇2 in the most important systems of

orthogonal coordinates11
(i) Cylindrical12 coordinates {ρ, φ, z} (see figure below) may be defined by their
relations with the Cartesian coordinates:
ðA:59Þ
• Gradient of a scalar function:

∂f 1 ∂f ∂f
∇f = nρ + nφ + nz . (A.60)
∂ρ ρ ∂φ ∂z
• The Laplace operator of a scalar function:
1 ∂ ⎛ ∂f ⎞ 1 ∂ 2f ∂ 2f
∇2 f = ⎜ρ ⎟ + 2 2 + 2 , (A.61)
ρ ∂ρ ⎝ ∂ρ ⎠ ρ ∂φ ∂z
• Divergence of a vector function of coordinates (f = nρ fρ + nφ fφ + nz fz):
∇⋅f=
( )
1 ∂ ρfρ
+
1 ∂fφ ∂f
+ z. (A.62)
ρ ∂ρ ρ ∂φ ∂z
• Curl of a vector function:
⎛ ⎞
⎛ 1 ∂f
∇ × f = nρ⎜ z
−
∂fφ ⎞ ⎛ ∂fρ
⎟ + n φ⎜ −
∂fz ⎞
⎟ + nz
1 ⎜ ∂ ρfφ
−
∂fρ ⎟
.
( ) (A.63)
⎝ ρ ∂φ ∂z ⎠ ⎝ ∂z ∂ρ ⎠ ρ ⎜ ∂ρ ∂φ ⎟
⎝ ⎠
• The Laplace operator of a vector function:

⎛ 1 2 ∂fφ ⎞ ⎛ 1 2 ∂fρ ⎞
∇2 f = nρ⎜∇2 fρ − 2 fρ − 2 ⎟ + n φ⎜∇2 fφ − 2 fφ + 2 ⎟ + nz ∇2 fz . (A.64)
⎝ ρ ρ ∂φ ⎠ ⎝ ρ ρ ∂φ ⎠
11
Some other orthogonal curvilinear coordinate systems are discussed in Part EM, section 2.3.
12
In the 2D geometry with fixed coordinate z, these coordinates are called polar.
A-14
(ii) Spherical coordinates {r, θ, φ} (see figure below) may be defined as:
ðA:65Þ
• Gradient of a scalar function:

∂f 1 ∂f 1 ∂f
∇f = nr + nθ + nφ . (A.66)
∂r r ∂θ r sin θ ∂φ
• The Laplace operator of a scalar function:
1 ∂ ⎛ 2 ∂f ⎞ 1 ∂ ⎛ ∂f ⎞ 1 ∂ 2f
∇2 f = ⎜r ⎟ + ⎜sin θ ⎟ + . (A.67)
r 2 ∂r ⎝ ∂r ⎠ r 2 sin θ ∂θ ⎝ ∂θ ⎠ (r sin θ )2 ∂φ 2
• Divergence of a vector function f = nrfr + nθ fθ + nφ fφ :
1 ∂(r fr ) 1 ∂(fθ sin θ )

2
1 ∂fφ (A.68)
∇⋅f= + + .
r 2 ∂r r sin θ ∂θ r sin θ ∂φ
• Curl of a similar vector function:
⎛ ⎞ ⎛
∇ × f = nr
(
1 ⎜ ∂ fφ sin θ )
−
∂fθ ⎟
+ nθ
1 ⎜ 1 ∂fr
−
∂ rfφ ( ) ⎞⎟
r sin θ ⎜ ∂θ ∂φ ⎟ r ⎜ sin θ ∂φ ∂r ⎟
⎝ ⎠ ⎝ ⎠
(A.69)
1 ⎛ ∂(rfθ ) ∂f ⎞
+ n φ ⎜⎜ − r ⎟⎟ .
r ⎝ ∂r ∂θ ⎠
• The Laplace operator of a vector function:

⎛ 2 2 ∂ 2 ∂fφ ⎞
∇2 f = nr⎜∇2 fr − fr − ( fθ sin θ ) − ⎟
⎝ r2 r 2 sin θ ∂θ r 2 sin θ ∂φ ⎠
⎛ 1 2 ∂f 2 cos θ ∂fφ ⎞
+ n θ⎜∇2 fθ − 2 2 fθ + 2 r − 2 2 ⎟ (A.70)
⎝ r sin θ r ∂θ r sin θ ∂φ ⎠
⎛ 1 2 ∂fr 2 cos θ ∂fθ ⎞
+ n φ⎜∇2 fφ − fφ + 2 + 2 2 ⎟.
⎝ 2 2
r sin θ r sin θ ∂φ r sin θ ∂φ ⎠
A-15
A.11 Products involving ∇

(i) Useful zeros:
• For any scalar function f (r),
∇ × (∇f ) ≡ curl(grad f ) = 0. (A.71)
• For any vector function f(r),

∇ ⋅ (∇ × f) ≡ div(curl f ) = 0. (A.72)
(ii) The Laplace operator expressed via the curl of a curl:

∇2 f = ∇(∇ ⋅ f) − ∇ × (∇ × f). (A.73)
(iii) Spatial differentiation of a product of a scalar function by a vector

function:
• The scalar 3D generalization of Eq. (A.22) is
∇ ⋅ (f g) = (∇f ) ⋅ g + f (∇ ⋅ g). (A.74a )
• Its vector generalization is similar:
∇ × (f g) = (∇f ) × g + f (∇ × g). (A.74b)
(iv) Spatial differentiation of products of two vector functions:

∇ × (f × g) = f(∇ ⋅ g) − (f ⋅ ∇)g − (∇ ⋅ f)g + (g ⋅ ∇) f , (A.75)
∇(f ⋅ g) = (f ⋅ ∇)g + (g ⋅ ∇) f + f × (∇ × g) + g × (∇ × f), (A.76)
∇ ⋅ (f × g) = g ⋅ (∇ × f) − f ⋅ (∇ × g). (A.77)
A.12 Integro-differential relations

(i) For an arbitrary surface S limited by closed contour C:
• The Stokes theorem, valid for any differentiable vector field f(r):
∫S (∇ × f) ⋅ d 2 r ≡ ∫S (∇ × f)nd 2r = ∮C f ⋅ d r ≡ ∮C fτ dr, (A.78)
where d2r ≡ nd2r is the elementary area vector (normal to the surface), and
dr is the elementary contour length vector (tangential to the contour line).
(ii) For an arbitrary volume V limited by closed surface S:
A-16
• Divergence (or ‘Gauss’) theorem, valid for any differentiable vector field f(r):
∫V (∇ ⋅ f ) d 3r = ∮S f ⋅ d 2 r ≡ ∮S fn d 2r. (A.79)
• Green’s theorem, valid for two differentiable scalar functions f(r) and g(r):
∫V (f ∇2 g − g∇2 f ) d 3r = ∮S (f ∇g − g ∇f )nd 2r. (A.80)
• An identity valid for any two scalar functions f and g, and a vector field j
with ∇·j = 0 (all differentiable):
∫V [f (j ⋅ ∇g ) + g(j ⋅ ∇f )] d 3r = ∮S fgjn d 2r. (A.81)
A.13 The Kronecker delta and Levi-Civita permutation symbols

• The Kronecker delta symbol (defined for integer indices):
⎧ 1, if j ′ = j ,
δjj ′ ≡ ⎨ (A.82)
⎩ 0, otherwise.
• The Levi-Civita permutation symbol (most frequently used for 3 integer

indices, each taking one of values 1, 2, or 3):
⎧ +1, if the indices follow in the ‘correct’ (‘even’)
⎪
⎪ order: 1 → 2 → 3 → 1 → 2 …,
⎪
εjj ′ j ″ ≡ ⎨ −1, if the indices follow in the ‘incorrect’ (‘odd’) (A.83)
⎪ order: 1 → 3 → 2 → 1 → 3 …,
⎪
⎪
⎩ 0, if any two indices coincide.
• Relation between the Levi-Civita and the Kronecker delta products:

3 δjl δjl ′ δjl ″
εjj ′ j ″εkk ′ k ″ = ∑ δj ′ l δj ′ l ′ δj ′ l ″ ; (A.84a )
l , l ′ , l ″= 1 δj ″ l δj ″ l ′ δj ″ l ″
summation of this relation, written for 3 different values of j = k, over these

values yields the so-called contracted epsilon identity:
3
∑εjj′ j″εjk ′ k ″ = δj′ k ′δj″ k ″ − δj′ k ″δj″ k ′. (A.84b)
j=1
A-17
A.14 Dirac’s delta-function, sign function, and theta-function

• Definition of 1D delta-function (for real a < b):
b ⎧ f (0), if a < 0 < b ,
∫a f (ξ )δ(ξ )dξ = ⎨ (A.85)
⎩ 0, otherwise,
where f(ξ) is any function continuous near ξ = 0. In particular (if f(ξ) = 1 near
ξ = 0), the definition yields
b ⎧1, if a < 0 < b ,
∫a δ(ξ )dξ = ⎨ (A.86)
⎩ 0, otherwise.
• Relation to the theta-function θ(ξ) and sign function sgn(ξ)

d 1 d
δ (ξ ) = θ (ζ ) = sgn(ξ ), (A.87a )
dξ 2 dξ
where
sgn(ξ ) + 1 ⎧ 0, if ξ < 0,
θ (ξ ) ≡ = ⎨
2 ⎩1, if ξ > 1,
(A.87b)
ξ ⎧−1, if ξ < 0,
sgn(ξ ) ≡ =⎨
ξ ⎩+1, if ξ > 1.
• An important integral13:
+∞
∫−∞ e is ξds = 2πδ(ξ ). (A.88)
• 3D generalization of the delta-function of the radius-vector (the 2D general-

ization is similar):
⎧ f (0), if 0 ∈ V ,
∫V f (r)δ(r)d 3r = ⎨⎩0, otherwise;
(A.89)
it may be represented as a product of 1D delta-functions of Cartesian

coordinates:
δ(r) = δ(r1)δ(r2 )δ(r3). (A.90)
13
The coefficient in this equation may be readily recalled by considering its left-hand part as the Fourier-
integral presentation of function f(s) ≡ 1, and applying Eq. (A.85) to the reciprocal Fourier transform
1 +∞
f (s ) ≡ 1 = ∫
2π −∞
e−isξ[2πδ(ξ )]dξ .
A-18
A.15 The Cauchy theorem and integral

Let a complex function f (z ) be analytic within a part of the complex plane z , that is
limited by a closed contour C and includes point z ′. Then
∮C f (z )dz = 0, (A.91)
∮C f (z ) z d−z z′ = 2πif (z′) (A.92)
The first of these relations is usually called the Cauchy integral theorem (or the
‘Cauchy–Goursat theorem’), and the second one—the Cauchy integral (or the
‘Cauchy integral formula’).
A.16 Literature
(i) Properties of some special functions are briefly discussed at the relevant
points of the lecture notes; in the alphabetical order:
• Airy functions: Part QM section 2.4;
• Bessel functions: Part EM section 2.7;
• Fresnel integrals: Part EM section 8.6;
• Hermite polynomials: Part QM section 2.9;
• Laguerre polynomials (both simple and associated): Part QM section 3.7;
• Legendre polynomials, associated Legendre functions: Part EM section 2.8,
and Part QM section 3.6;
• Spherical harmonics: Part QM section 3.6;
• Spherical Bessel functions: Part QM sections 3.6 and 3.8.
(ii) For more formulas, and their discussions, I can recommend the following
handbooks (in the alphabetical order of the authors)14:
• Handbook of Mathematical Formulas [2]15;
• Tables of Integrals, Series, and Products [3];
• Mathematical Handbook for Scientists and Engineers [4];
• Integrals and Series volumes 1 and 2 [5].
A popular textbook Mathematical Methods for Physicists [6] may be also used as
a formula manual. Many formulas are also available from the symbolic calculation
parts of commercially available software packages listed in section (iv) below.
(iii) Probably the most popular collection of numerical calculation codes are the
twin manuals by W Press et al [1]:
14
On a personal note, perhaps 90% of all formula needs throughout my research career were satisfied by a tiny,
wonderfully compiled old book: H. Dwight, Tables of Integrals and Other Mathematical Data, 4th ed.,
Macmillan, 1961, whose copies, rather amazingly, are still available on the Web.
15
An updated version of this collection is now available online at http://dlmf.nist.gov/.
A-19
• Numerical Recipes in Fortran 77;

• Numerical Recipes [in C++ - KKL].
My lecture notes include very brief introductions into numerical methods of

differential equation solution:
• ordinary differential equations: Part CM section 3.9, and
• partial differential equations: Part CM section 8.5, and Part EM
section 2.8, which include references to literature for further reading.
(iv) The following are the most popular software packages for numerical and
symbolic calculations, all with plotting capabilities (in alphabetical order):
• Maple (http://www.maplesoft.com/);
• MathCAD (http://www.ptc.com/products/mathcad/);
• Mathematica (http://www.wolfram.com/products/mathematica/index.html);
• MATLAB (http://www.mathworks.com/products/matlab/).
References
[1] Press W et al 1992 Numerical Recipes in Fortran 77 2nd edn (Cambridge: Cambridge
University Press); Press W et al 2007 Numerical Recipes 3rd edn (Cambridge: Cambridge
University Press)
[2] Abramowitz M and Stegun I (eds) 1965 Handbook of Mathematical Formulas (New York:
Dover)
[3] Gradshteyn I and Ryzhik I 1980 Tables of Integrals, Series, and Products 5th edn (New York:
Academic)
[4] Korn G and Korn T 2000 Mathematical Handbook for Scientists and Engineers 2nd edn
(New York: Academic)
[5] Prudnikov A et al 1986 Integrals and Series vol 1 (Boca Raton, FL: CRC Press); Prudnikov A
et al 1986 Integrals and Series vol 2 (Boca Raton, FL: CRC Press)
[6] Arfken G et al 2012 Mathematical Methods for Physicists 7th edn (New York: Academic)
A-20
IOP Publishing
Lecture notes
Appendix B
Selected physical constants1
Table B.1.
Relative rms
Symbol Quantity SI value and unit Gaussian value and unit uncertainty
c speed of light 2.997 924 58 × 108 m s−1 2.997 924 58 × 1010 cm s−1 0 (defined value)
in free space
G gravitation 6.674 1 × 10−11 m3 kg−1 s−2 6.674 1 × 10−8 cm 3 g−1 s−2 ∼5 × 10−5
constant
ℏ Planck 1.054 5718 0 × 10–34 J s 1.054 571 80 × 10−27 erg s ∼1 × 10−8
constant
e elementary 1.602 176 2 × 10−19 C 4.803 203 × 10−10 statcoulomb ∼6 × 10−9
electric charge
me electron’s rest 0.910 938 35 × 10−30 kg 0.910 938 35 × 10−27 G ∼1 × 10−8
mass
mp proton’s rest 1.672 621 90 × 10−27 kg 1.672 621 90 × 10−24 G ∼1 × 10−8
mass
μ0 magnetic 4π × 10−7 N A−2 – 0 (defined value)
constant
ε0 electric 8.854 187 817 × 10−12 F m−1 – 0 (defined value)
constant
kB Boltzmann 1.380 648 × 10−23 J K−1 1.380 650 9 × 10−16 erg K−1 ∼2 × 10−6
constant
Comments:
1. The fixed value of c was defined by an international convention in 1983,
in order to extend the official definition of a second (as ‘the duration of
9 192 631 770 periods of the radiation corresponding to the transition
1
The listed numerical values of the constants are from the most recent (2014) International CODATA
recommendation (see, e.g. http://physics.nist.gov/cuu/Constants/index.html), besides a newer result for kB—
see [1].
doi:10.1088/978-0-7503-1404-6ch12 B-1 ª Konstantin K Likharev 2018

between the two hyperfine levels of the ground state of the cesium-133 atom’)
to that of a meter. The values are back-compatible with the legacy definitions
of the meter (initially, as 1/40 000 000th of the Earth’s meridian length) and
the second (for a long time, as 1/(24 × 60 × 60) = 1/86 400th of the Earth’s
rotation period), within the experimental errors of those measures.
2. ε0 and μ0 are not really the fundamental constants; in the SI system of units
one of them (say, μ0) is selected arbitrarily2, while the other one is defined via
the relation ε0μ0 = 1/c2.
3. The Boltzmann constant kB is also not quite fundamental, because its only
role is to comply with the independent definition of the kelvin (K), as the
temperature unit in which the triple point of water is exactly 273.16 K. If
temperature is expressed in energy units kBT (as is done, for example, in
Part SM of this lecture note series), this constant disappears altogether.
4. The dimensionless fine structure (‘Sommerfeld’s) constant α is numerically the
same in any system of units:
⎧ e 2 /4πε0ℏc in SI units ⎫
α≡⎨ ⎬ ≈ 7.297 352 566 × 10−3
⎩ e 2 / ℏc in Gaussian units ⎭
1
≈ ,
137.035 999 14
and is known with a much smaller rms uncertainty (currently, ∼3 × 10−10)
than those of the component constants.
References
[1] Gaiser C et al 2017 Metrologia 54 280
[2] Newell D 2014 Phys. Today 67 35–41
2
Note that the selected value of μ0 may be changed (a bit) in a few years—see, e.g. [2].
B-2
IOP Publishing
Lecture notes
Bibliography
This section presents a partial list of the textbooks and monographs used in the work
on the EAP series1,2.
Part CM: Classical Mechanics

Fetter A L and Walecka J D 2003 Theoretical Mechanics of Particles and Continua (New York:
Dover)
Goldstein H, Poole C and Safko J 2002 Classical Mechanics 3rd edn (Reading, MA: Addison
Wesley)
Granger R A 1995 Fluid Mechanics (New York: Dover)
José J V and Saletan E J 1998 Classical Dynamics (Cambridge: Cambridge University Press)
Landau L D and Lifshitz E M 1976 Mechanics 3rd edn (Oxford: Butterworth-Heinemann)
Landau L D and Lifshitz E M 1986 Theory of Elasticity (Oxford: Butterworth-Heinemann)
Landau L D and Lifshitz E M 1987 Fluid Mechanics 2nd edn (Oxford: Butterworth-Heinemann)
Schuster H G 1995 Deterministic Chaos 3rd edn (New York: Wiley)
Sommerfeld A 1964 Mechanics (New York: Academic)
Sommerfeld A 1964 Mechanics of Deformable Bodies (New York: Academic)
Part EM: Classical Electrodynamics

Batygin V V and Toptygin I N 1978 Problems in Electrodynamics 2nd edn (New York: Academic)
Griffiths D J 2007 Introduction to Electrodynamics 3rd edn (Englewood Cliffs, NJ: Prentice-Hall)
Jackson J D 1999 Classical Electrodynamics 3rd edn (New York: Wiley)
Landau L D and Lifshitz E M 1984 Electrodynamics of Continuous Media 2nd edn (Auckland:
Reed)
Landau L D and Lifshitz E M 1975 The Classical Theory of Fields 4th edn (Oxford: Pergamon)
Panofsky W K H and Phillips M 1990 Classical Electricity and Magnetism 2nd edn (New York:
Dover)
1
The list does not include the sources (mostly, recent original publications) cited in the ends of the chapters,
and the mathematics textbooks and handbooks listed in appendix A, section A.16.
2
Recently some high-quality teaching materials on advanced physics have become available online, including:
R Fitzpatrick’s text on classical electromagnetism (farside.ph.utexas.edu/teaching/jk1/Electromagnetism.pdf);
B Simons’ ‘lecture shrunks’ on advanced quantum mechanics (www.tcm.phy.cam.ac.uk/∼bds10/aqp.html);
and D Tong’s lecture notes on several advanced topics (www.damtp.cam.ac.uk/user/tong/teaching.html).

Stratton J A 2007 Electromagnetic Theory (New York: Wiley)

Tamm I E 1979 Fundamentals of the Theory of Electricity (Paris: Mir)
Zangwill A 2013 Modern Electrodynamics (Cambridge: Cambridge University Press)
Part QM: Quantum Mechanics

Abers E S 2004 Quantum Mechanics (London: Pearson)
Auletta G, Fortunato M and Parisi G 2009 Quantum Mechanics (Cambridge: Cambridge
University Press)
Capri A Z 2002 Nonrelativistic Quantum Mechanics 3rd edn (Singapore: World Scientific)
Cohen-Tannoudji C, Diu B and Laloë F 2005 Quantum Mechanics (New York: Wiley)
Constantinescu F, Magyari E and Spiers J A 1971 Problems in Quantum Mechanics (Amsterdam:
Elsevier)
Galitski V et al 2013 Exploring Quantum Mechanics (Oxford: Oxford University Press)
Gottfried K and Yan T-M 2004 Quantum Mechanics: Fundamentals 2nd edn (Berlin: Springer)
Griffith D 2005 Quantum Mechanics 2nd edn (Englewood Cliffs, NJ: Prentice Hall)
Landau L D and Lifshitz E M 1977 Quantum Mechanics (Nonrelativistic Theory) 3rd edn
(Oxford: Pergamon)
Messiah A 1999 Quantum Mechanics (New York: Dover)
Merzbacher E 1998 Quantum Mechanics 3rd edn (New York: Wiley)
Miller D A B 2008 Quantum Mechanics for Scientists and Engineers (Cambridge: Cambridge
University Press)
Sakurai J J 1994 Modern Quantum Mechanics (Reading, MA: Addison-Wesley)
Schiff L I 1968 Quantum Mechanics 3rd edn (New York: McGraw-Hill)
Shankar R 1980 Principles of Quantum Mechanics 2nd edn (Berlin: Springer)
Schwabl F 2002 Quantum Mechanics 3rd edn (Berlin: Springer)
Part SM: Statistical Mechanics

Feynman R P 1998 Statistical Mechanics 2nd edn (Boulder, CO: Westview)
Huang K 1987 Statistical Mechanics 2nd edn (New York: Wiley)
Kubo R 1965 Statistical Mechanics (Amsterdam: Elsevier)
Landau L D and Lifshitz E M 1980 Statistical Physics, Part 1 3rd edn (Oxford: Pergamon)
Lifshitz E M and Pitaevskii L P 1981 Physical Kinetics (Oxford: Pergamon)
Pathria R K and Beale P D 2011 Statistical Mechanics 3rd edn (Amsterdam: Elsevier)
Pierce J R 1980 An Introduction to Information Theory 2nd edn (New York: Dover)
Plishke M and Bergersen B 2006 Equilibrium Statistical Physics 3rd edn (Singapore: World
Scientific)
Schwabl F 2000 Statistical Mechanics (Berlin: Springer)
Yeomans J M 1992 Statistical Mechanics of Phase Transitions (Oxford: Oxford University Press)
Multidisciplinary/specialty
Ashcroft W N and Mermin N D 1976 Solid State Physics (Philadelphia, PA: Saunders)
Blum K 1981 Density Matrix and Applications (New York: Plenum)
Breuer H-P and Petruccione E 2002 The Theory of Open Quantum Systems (Oxford: Oxford
University Press)
Cahn S B and Nadgorny B E 1994 A Guide to Physics Problems, Part 1 (New York: Plenum)
13-2
Cahn S B, Mahan G D and Nadgorny B E 1997 A Guide to Physics Problems, Part 2 (New York:
Plenum)
Cronin J A, Greenberg D F and Telegdi V L 1967 University of Chicago Graduate Problems in
Physics (Reading, MA: Addison Wesley)
Hook J R and Hall H E 1991 Solid State Physics 2nd edn (New York: Wiley)
Joos G 1986 Theoretical Physics (New York: Dover)
Kompaneyets A S 2012 Theoretical Physics 2nd edn (New York: Dover)
Lax M 1968 Fluctuations and Coherent Phenomena (London: Gordon and Breach)
Lifshitz E M and Pitaevskii L P 1980 Statistical Physics, Part 2 (Oxford: Pergamon)
Newbury N et al 1991 Princeton Problems in Physics with Solutions (Princeton, NJ: Princeton
University Press)
Pauling L 1988 General Chemistry 3rd edn (New York: Dover)
Tinkham M 1996 Introduction to Superconductivity 2nd edn (New York: McGraw-Hill)
Walecka J D 2008 Introduction to Modern Physics (Singapore: World Scientific)
Ziman J M 1979 Principles of the Theory of Solids 2nd edn (Cambridge: Cambridge University
Press)
13-3

Classical Electrodynamics

Uploaded by

Copyright:

Available Formats

Classical Electrodynamics

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

Classical Electrodynamics

Uploaded by

Copyright:

Available Formats

This content has been downloaded from IOPscience. Please scroll down to see the full text.

Please note that terms and conditions apply.

You may also like:

Diffraction in high numerical aperture systems: polarization effects

Understanding zero-point energy in the context of classical electromagnetism

Eigenmode generation by a given current in anisotropic and gyrotropic media

Angular momentum in classical electrodynamics

Contrasting interactions between dipole oscillators in classical and quantum theories:

IOP Publishing, Bristol, UK

ISBN 978-0-7503-1404-6 (ebook)

IOP Expanding Physics

Published by IOP Publishing, wholly owned by The Institute of Physics, London

IOP Publishing, Temple Circus, Temple Way, Bristol, BS1 6HG, UK

Preface to the EAP Series ix

1 Electric charge interaction 1-1

2 Charges and conductors 2-1

3 Dipoles and dielectrics 3-1

7 Electromagnetic wave propagation 7-1

7.6 Waveguides: H and E waves 7-42

8 Radiation, scattering, interference, and diffraction 8-1

9 Special relativity 9-1

10 Radiation by relativistic charges 10-1

10.4 Bremsstrahlung and Coulomb losses 10-23

A Selected mathematical formulas A-1

B Selected physical constants B-1

Essential Advanced Physics

Distinguishing features of this series—in brief

Level and precursors

Origins and motivation

physical notion/effect and of every novel theoretical approach is always accom-

Disclaimer and encouragement

The structure of this classical electrodynamics course is quite traditional7; in order to

I hope the discussion of these methods of mathematical physics, using the

I am extremely grateful to my faculty colleagues and other readers of the preliminary

Abbreviations Fonts Symbols

Parts of the series

1.1 The Coulomb law

• electric charges qk, as revealed, most explicitly, by observation of the

doi:10.1088/978-0-7503-1404-6ch1 1-1 ª Konstantin K Likharev 2018

In addition to Eq. (1.3), another key experimental law of electrostatics is the

The notion of ﬁeld is just convenient is electrostatics, but becomes virtually

1.2 The Gauss law

∮ End 2r = 4πr 2E (r). (1.17)

Now, applying the Gauss law (1.16), we obtain

where Q(r) is the full charge inside the sphere of radius r:

Figure 1.4. The electric ﬁeld of a charged plane.

space-differentiable vector, in particular E, and for volume V limited by any

∮S End 2r = ∫V (∇ ⋅ E)d 3r, (1.25)

where ∇ is the del (or ‘nabla’) operator of spatial differentiation12. Combining

which may be readily integrated on each of the segments:

1.3 Scalar potential and electric field energy

Figure 1.5. A simple problem of charged particle motion.

which is not much simpler than Eq. (1.11).

has earned a special name—the Laplace equation21.

Its ﬁrst integral,

in an evident agreement with Eq. (1.19). Integrating this result again,

Uext = ∫ ρ(r)ϕext(r)d 3r. (1.55)

For the discrete charge case, this result becomes

charge interaction within the system, as expressed via the charge-by-potential

This expression may be integrated by parts as26