Lectures On Electromagnetism PDF

Lectures on
Electromagnetism
Second Edition
This page intentionally left blank
Lectures on
Electromagnetism
Second Edition
Ashok Das
University of Rochester, USA
World Scientific
NEW JERSEY • LONDON • SINGAPORE • BEIJING • SHANGHAI • HONG KONG • TA I P E I • CHENNAI
Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.
LECTURES ON ELECTROMAGNETISM
Second Edition
Copyright © 2013 Hindustan Book Agency (HBA)
Authorized edition by World Scientific Publishing Co. Pte. Ltd. for exclusive distribution worldwide except
India.
The distribution rights for print copies of the book for India remain with Hindustan Book Agency (HBA).
All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or
mechanical, including photocopying, recording or any information storage and retrieval system now known or to
be invented, without written permission from the Publisher.
ISBN 978-981-4508-26-1
Printed in India, bookbinding made in Singapore.
Lakshmi - Lecs on Electromagnetism.pmd 1 2/22/2013, 5:00 PM

To
Jayanthi, Judy
and
Silvana
Contents
1 Electrostatics . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Coulomb’s law . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Electric field . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Gauss’ law . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Potential . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.5 Electrostatic energy . . . . . . . . . . . . . . . . . . . 24
1.6 Selected problems . . . . . . . . . . . . . . . . . . . . 27
2 Potential for simple systems. . . . . . . . . . . . . . . . . . 31
2.1 Potential for a thin spherical shell . . . . . . . . . . . 33
2.2 Potential for an infinitely long wire . . . . . . . . . . 36
2.3 Potential for a circular charged disc . . . . . . . . . . 39
2.4 Potential for a charge displaced along the z-axis. . . 42
2.5 Dipole . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.6 Continuous distribution of dipoles . . . . . . . . . . . 50
2.7 Quadrupole . . . . . . . . . . . . . . . . . . . . . . . 53
2.8 Potential due to a double layer of charges . . . . . . 55
2.9 Conductors and insulators . . . . . . . . . . . . . . . 58
2.10 Capacitor . . . . . . . . . . . . . . . . . . . . . . . . 63
3 Boundary value problems . . . . . . . . . . . . . . . . . . . 69
3.1 Method of images . . . . . . . . . . . . . . . . . . . . 69
3.2 Boundary conditions for differential equations . . . . 80
3.2.1 Partial differential equations . . . . . . . . . . 82
3.2.2 Uniqueness theorem . . . . . . . . . . . . . . . 88
3.3 Solutions of the Laplace equation . . . . . . . . . . . 90
3.3.1 General properties of harmonic functions . . . 90
3.3.2 Solution in Cartesian coordinates . . . . . . . . 93
3.3.3 Solution in spherical coordinates . . . . . . . . 96
3.3.4 Circular harmonics . . . . . . . . . . . . . . . . 103
3.4 Solution of the Poisson equation. . . . . . . . . . . . 106
3.4.1 Green’s function . . . . . . . . . . . . . . . . . 107
vii
viii
3.4.2 Dirichlet boundary condition . . . . . . . . . . 109

3.4.3 Neumann boundary condition . . . . . . . . . . 114
4 Dielectrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.1 Electric displacement field . . . . . . . . . . . . . . . 121
4.2 Boundary conditions in dielectric . . . . . . . . . . . 128
5 Magnetostatics . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.1 Lorentz force. . . . . . . . . . . . . . . . . . . . . . . 141
5.2 Current. . . . . . . . . . . . . . . . . . . . . . . . . . 142
5.3 Force on a current due to a magnetic field . . . . . . 144
5.4 Nature of the magnetic field . . . . . . . . . . . . . . 150
5.5 Vector potential . . . . . . . . . . . . . . . . . . . . . 152
5.6 Multipole expansion. . . . . . . . . . . . . . . . . . . 155
5.7 Magnetization . . . . . . . . . . . . . . . . . . . . . . 157
5.8 Magnetic field intensity . . . . . . . . . . . . . . . . . 161
5.9 Boundary condition . . . . . . . . . . . . . . . . . . . 163
5.10 Faraday’s Law . . . . . . . . . . . . . . . . . . . . . . 166
5.11 Inductance . . . . . . . . . . . . . . . . . . . . . . . . 167
6 Maxwell’s equations . . . . . . . . . . . . . . . . . . . . . . 171
6.1 Generalization of Ampere’s law . . . . . . . . . . . . 171
6.2 Plane wave solution . . . . . . . . . . . . . . . . . . . 175
6.2.1 Polarization . . . . . . . . . . . . . . . . . . . . 179
6.3 Boundary conditions . . . . . . . . . . . . . . . . . . 181
6.4 Energy and the Poynting vector . . . . . . . . . . . . 190
6.5 Gauge invariance of Maxwell’s equations . . . . . . . 195
6.6 Lorentz transformation . . . . . . . . . . . . . . . . . 200
6.7 Covariance of Maxwell’s equations . . . . . . . . . . 208
6.8 Retarded Green’s function . . . . . . . . . . . . . . . 212
6.9 Kirchhoff’s representation . . . . . . . . . . . . . . . 218
7 Wave guides. . . . . . . . . . . . . . . . . . . . . . . . . . . 225
7.2 Rectangular wave guide. . . . . . . . . . . . . . . . . 231
7.2.1 TM waves . . . . . . . . . . . . . . . . . . . . . 233
7.2.2 TE waves . . . . . . . . . . . . . . . . . . . . . 235
7.3 Cylindrical wave guide . . . . . . . . . . . . . . . . . 237
7.3.1 TM waves . . . . . . . . . . . . . . . . . . . . . 240
7.3.2 TE waves . . . . . . . . . . . . . . . . . . . . . 242
7.4 TEM waves . . . . . . . . . . . . . . . . . . . . . . . 243
ix
7.5 Wave impedance. . . . . . . . . . . . . . . . . . . . . 245

7.6 Attenuation factor in wave guides . . . . . . . . . . . 246
7.7 Cavity resonators . . . . . . . . . . . . . . . . . . . . 249
7.8 Q factor of a cavity . . . . . . . . . . . . . . . . . . . 254
7.9 Dielectric wave guides (optical fibers) . . . . . . . . . 256
8 Propagation through a conducting medium . . . . . . . . . 263
8.2 Reflection at normal incidence . . . . . . . . . . . . . 266
8.3 Reflection at oblique incidence . . . . . . . . . . . . . 269
8.4 Reflection from a good conducting surface . . . . . . 275
8.5 Radiation pressure . . . . . . . . . . . . . . . . . . . 282
9 Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
9.1 Electric dipole radiation . . . . . . . . . . . . . . . . 292
9.1.1 Power radiated by an electric dipole . . . . . . 296
9.2 Magnetic dipole radiation . . . . . . . . . . . . . . . 297
9.3 Center-fed antennas . . . . . . . . . . . . . . . . . . . 301
9.3.1 Properties of antennas . . . . . . . . . . . . . . 307
9.4 Multipole expansion. . . . . . . . . . . . . . . . . . . 308
9.5 Behavior of multipole fields . . . . . . . . . . . . . . 314
10 Electromagnetic fields of currents. . . . . . . . . . . . . . . 323
10.1 Lienard-Wiechert potential. . . . . . . . . . . . . . . 323
10.2 Uniform linear motion . . . . . . . . . . . . . . . . . 329
10.3 Method of virtual photons . . . . . . . . . . . . . . . 337
10.4 Asymptotic values of the fields. . . . . . . . . . . . . 341
10.4.1 Dipole approximation . . . . . . . . . . . . . . 343
10.4.2 Linear acceleration . . . . . . . . . . . . . . . . 346
10.4.3 Uniform circular motion . . . . . . . . . . . . . 348
10.5 Čerenkov effect . . . . . . . . . . . . . . . . . . . . . 352
10.6 Self-force . . . . . . . . . . . . . . . . . . . . . . . . . 362
11 Plasma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
11.1 General features of a plasma . . . . . . . . . . . . . . 373
11.2 Plasma oscillation . . . . . . . . . . . . . . . . . . . . 377
11.3 Motion of the positive ions . . . . . . . . . . . . . . . 382
11.4 Effect of a background magnetic field . . . . . . . . . 383
11.5 Faraday rotation . . . . . . . . . . . . . . . . . . . . 389
11.6 Alfvén waves . . . . . . . . . . . . . . . . . . . . . . . 391
11.7 Collisions . . . . . . . . . . . . . . . . . . . . . . . . . 394
x

12 Electromagnetic interactions . . . . . . . . . . . . . . . . . 397
12.1 Relativistic Lagrangian description . . . . . . . . . . 397
12.2 Motion in a uniform electric field . . . . . . . . . . . 408
12.3 Motion in a uniform magnetic field . . . . . . . . . . 412
12.4 Motion in crossed fields. . . . . . . . . . . . . . . . . 414
12.5 Motion in a slowly varying magnetic field . . . . . . 419
12.6 Anomalous magnetic moment . . . . . . . . . . . . . 424
13 Scattering and diffraction . . . . . . . . . . . . . . . . . . . 431
13.1 Scattering from a perfectly conducting sphere . . . . 432
13.2 Kirchhoff’s approximation . . . . . . . . . . . . . . . 442
Preface
Over the years, I have taught a two semester graduate course on

electromagnetism at the University of Rochester. The present book
follows that material almost word-for-word. I have not attempted
to polish the writing, and these lecture notes, therefore, reflect the
informality of the class room. In fact, I even considered presenting the
material in the original format, but lectures have a way of ending and
starting in the middle of a topic, which is neither very appropriate
nor expected of a book. Nonetheless, the subject is presented exactly
in the order it was taught in class.
Some of the material is repeated in places, but this was deemed
important for clarifying the lectures. The book is self-contained, in
the sense that most of the steps in the development of the subject
are derived in detail, and integrals are either evaluated or listed when
needed. A motivated student should be able to work through the
notes independently and without difficulty. Throughout the book, I
followed the convention of representing three dimensional vectors by
bold-faced symbols, and I use CGS units because of their relevance
in special relativity.
In preparing lectures for the course, I relied, at least partially,
on the material contained in the following texts:
1. P. C. Clemmow, “Introduction to Electromagnetic Theory”,

Cambridge University Press (1973).
2. L. Eyges, “The Classical Electromagnetic Field”, Dover Publi-

cation (1972).
3. J. Frenkel, “Princı́pios de Eletrodinâmica Clássica”, Editora da

Universidade de São Paulo (1996).
4. W. Hauser, “Introduction to the Principles of Electromagnetism”,

Addison-Wesley Publishing (1971).
xi
xii
Several of my colleagues at Rochester and at other universities,

as well as many of my students, have influenced the development of
these lectures. Most important were, of course, the excellent ques-
tions raised by students in class and during private discussions. I
sincerely appreciate everyone’s input.
Part of the lecture notes were converted to LATEX by Diane Pick-
ersgill. The present format of the book in LATEX is largely due to the
meticulous work of Dr. Alex Constandache, who succeeded in giving
it a more “user friendly” appearance. Most of the figures were drawn
using PSTricks, while a few were done using Gnuplot and Xfig. I
would like to thank Drs A. Constandache and F. T. Brandt for help
with several figures.
It is also a pleasure to thank the editors of the TRiPS series, as
well as the publisher, for being so accommodating to all my requests
in connection with the book.
Finally, I thank the members of my family, in particular for
their patient support and understanding during the completion of
this book.
Ashok Das
Rochester
Preface to the second edition
The modifications in this second edition of the book arose mainly

from the requests made by various readers. Several typos in the
earlier version have been fixed and the presentation made clearer at
many places, sometimes with longer derivations of the results. The
figures now carry captions with references to them in the text. In
addition to the numerous examples and exercises that were already
present in the text, I have now included a few selected problems at
the end of (almost) every chapter in the present edition.
Ashok Das
Rochester
Chapter 1
Electrostatics
Electrostatics is the study of the properties of electric charge distri-

butions at rest. This is the first step in the understanding of elec-
tromagnetic phenomena. In the next few lectures, we will discuss
various properties of time independent charge distributions.
1.1 Coulomb’s law
It was observed quite early that when particles carrying electric charge
are brought closer, they experience a force, and this force was called
the electric force. The main question that one studies in electrostat-
ics is the analysis of the electric force experienced by a given charge
due to a complicated distribution of static electric charges in space.
(As a side remark, let us simply note here that although at the mo-
ment it may seem like the course is only concerned with developing
techniques for solving problems in electromagnetism, the techniques
are quite general and are so powerful that they may be used in any
other field of research as well. In that sense, we can think of the
material of the course as developing powerful techniques for solving
theoretical problems through examples of electromagnetism.) Fun-
damental to this study, therefore, is the understanding of the force
between a pair of static charges separated by a given distance. This
question was, in fact, studied by Coulomb in a series of impressive
experiments and he found that the electric force between a pair of
static particles carrying electric charge
1. is linearly proportional to the individual charges on each of the

particles,
2. varies inversely as the square of the distance between the two

particles,
1
2 1 Electrostatics
3. and is a vector along the line joining the positions of the two
particles carrying charge. It is attractive if the charges of the
two particles have opposite sign and is repulsive otherwise.
Quantitatively, we can, therefore, write that the force experi-

enced by a particle with charge q at the coordinate r because of the
presence of a static particle carrying charge q1 at the coordinate r1 is
given by
qq1 qq1 qq1
F(r − r1 ) = k (r − r1 ) = k R=k R̂. (1.1)
|r − r1 |3 |R|3 |R|2
Here, we have defined R = r − r1 (see Fig. 1.1) and have used the
notation (which we will use throughout the lectures) that a bold-
face quantity represents a vector while a boldface quantity with a
“hat” simply stands for a unit vector and, in the above equation, k
represents the constant of proportionality.
z q1
R
r1 q
r
y
x
Figure 1.1: Direction of the electrostatic force experienced by a charge

q at r due to a charge q1 of the same sign at coordinate r1 .
There are several things to note about the force from the ex-
pression in (1.1). First, it is translationally invariant, namely, under
a translation of the coordinate system, the expression is unchanged.
Sometimes, one takes advantage of this to translate the coordinate
system such that q1 is at the origin, in which case, the force has the
simple form
qq1 qq1
F(r) = k 3
r = k 2 r̂, (1.2)
|r| |r|
where r denotes the coordinate of the charge q in the new frame

(with respect to the new origin). Second, the value of the constant
1.1 Coulomb’s law 3
k depends on the system of units used. In the CGS (Gaussian) units

(which we will work with throughout these lectures), where distances
are measured in centimeters and the force in dynes, k = 1 while in the
MKS (SI) units (which is used mostly in engineering applications),
where distances are measured in meters and the force in Newtons,
2
k = 10−7 c2 N−m C2
. Here “N” stands for Newton, “C” for Coulomb
and “m” for meter and c is the speed of light (just the value ≈ 3× 108
in the MKS system without the dimensions). The rationale behind
the choice of different units is as follows. In applications (SI units),
currents are measured in units of Ampere which is defined to be
Coulomb/sec. Therefore, one can define the unit of electric charge,
Coulomb, from Ampere’s law which involves the magnetic force be-
tween two currents. Once the unit of charge, Coulomb, is defined
in this way, consistency determines the constant k in the Coulomb’s
law. (Basically, once the unit of charge has been defined from the
magnetic force, it has to be consistent with the definition of force in
Coulomb’s law. The peculiar value of the constant arises because it
involves the vacuum magnetic permeability since it originates from
the definition of the magnetic force and the constant can be identified
1
with k = 4πǫ 0
where ǫ0 is the dielectric permittivity of the vacuum
through usual relations.) On the other hand, in the CGS (Gaussian)
units, the desire is more to be consistent with the requirements of
relativistic invariance. We know from the studies in relativity that
electric and magnetic fields can be mapped to each other under a
Lorentz transformation. The simplest way to see this is to note that
if we have a static charge, it only produces an electric field. However,
in a different Lorentz frame, the charge would be moving giving rise to
a current and, therefore, to a magnetic field as well. Thus, relativis-
tic invariance treats electric and magnetic fields on the same footing.
From the form of Lorentz transformations, we can also see that they
have to have the same dimension. The choice k = 1 enforces this and
we see that if we choose k = 1, then, Coulomb’s law can be thought
of as defining the unit of electric charge in the CGS units, which is
an “esu”. (The electric force between two charges of magnitude 1 esu
each, at a separation of 1 cm, is defined to be 1 dyne.) It is quite
easy to see from this that (remember that 1N = 105 dynes)
1 esu = 1 statCoulomb = (10 c)−1 C = (3 × 109 )−1 C. (1.3)
In fact, let us suppose, 1 esu = α C. Then,

4 1 Electrostatics
N − m2
k = 10−7 c2
C2
105 × 104 dyne − cm2
= 10−7 c2 ×
α−2 esu2
= 102 c2 α2 = 1, (1.4)
which determines α = (10 c)−1 and, therefore, 1 esu = (10c)−1 C. (We

note here that the magnitude of electric charge on an electron or on
a proton is known to be 1.6 × 10−19 C = 4.8 × 10−10 esu.)
Experimentally, it is also observed that when more charges are
present, the force on a given charge adds vectorially as if each of the
source charges interacts (pairwise) with the given charge independent
of the presence of the others. Namely, if we have a distribution of
static source charges qi , i = 1, 2, . . . , n, at the coordinates ri , then,
the force experienced by a test charge q at the coordinate r is given
by (in the CGS system that we have chosen)
n
X qqi
F(r) = (r − ri ). (1.5)
|r − ri |3
i=1
This is known as the law of superposition of electric forces.

As a digression, let us talk briefly about the experimental va-
lidity of Coulomb’s law. Of course, there are various uncertainties
associated with any experiment and it is impossible to say experi-
mentally that the electric force is given exactly by Coulomb’s law.
One can only put limits on its validity. There are two meaningful
ways one puts limits on the experimental validity of Coulomb’s law.
The first is to assume that the force, instead of varying exactly as the
inverse square of the distance, varies as |r − r1 |−2−δ and find an ex-
perimental bound for δ. Experiments, of course, have become much
more sensitive since the time of Coulomb (although the philosophy
of the experiments remains essentially the same) and the present day
experiments give us a bound of δ ∼ 10−16 and, therefore, we can
think of the electric force as varying inversely as the square of the
distance for all practical purposes.
A second way of putting an experimental limit on the form of
Coulomb’s law is to parameterize the potential (from which the force
−µr
is obtained) as e r where r represents the magnitude of the sepa-
ration between the two charges (in the earlier notation |r − r1 |) and
µ a mass parameter so that the exponential will be dimensionless
1.2 Electric field 5
(in some units). If Coulomb’s law is exact, we should have µ = 0

giving the Coulomb potential. However, if µ 6= 0, the potential is
called the Yukawa potential and the present experimental bound on
the mass is given by µ < 10−20 me , where me represents the mass
of the electron. The reason for such a parameterization of the po-
tential is of a quantum nature. Quantum mechanically, if we think
of a force as arising due to the exchange of a particle (quantum),
then µ would represent the mass of the exchanged particle which, in
the case of electromagnetic forces, is called the photon. The exper-
imental bound suggests that the photon is massless which is what
we would expect from gauge invariance of the Maxwell’s equations
which we will discuss later. Furthermore, all these results are ob-
tained from laboratory as well as satellite experiments. Thus, to
summarize, Coulomb’s law holds quite well over a wide range of dis-
tances – from very small to very large – and experimentally, we find
that the mass of the exchanged particle (photon) associated with the
Coulomb force is consistent with zero.
1.2 Electric field
From the form of the force in (1.5) it is clear that, even in the presence
of a distribution of static charges, the force is linearly proportional to
the charge of the test particle. Therefore, by dividing out the charge
of the test particle, we can define an auxiliary quantity which we call
the electric field. For a single source charge q1 at r1 , the electric field
at the coordinate r is given by (k = 1 in our units)
q1
E(r) = (r − r1 ), (1.6)
|r − r1 |3
while, for a distribution of static charges, the electric field at r is

given by (the electric field also depends on the locations of the source
charges which we are suppressing for simplicity)
n
X qi
E(r) = (r − ri ). (1.7)
|r − ri |3
i=1
This quantity is inherently a characteristic of a given distribution of

static charges and has no reference to the test charge. The electric
field is not directly measurable, rather the electric force on a particle
is and, from the definition of the electric field, we can think of the
electric field as the force per unit charge. Namely, when an electrically
charged particle with charge q is brought to the coordinate r in the
6 1 Electrostatics
presence of a distribution of charges, it would experience an electric

force
F(r) = qE(r). (1.8)
However, the assumption here is that in bringing the test charge to
the coordinate r, we should not alter the electric field of the charge
distribution appreciably (namely, the magnitude of the test charge
should not be large). The concept of an electric field is not as crucial
for the study of problems in electrostatics as it is for the study of
time-dependent phenomena which we will deal with later. Finally,
let us simply note here that the standard unit of the electric field in
MKS (SI) system is Volt
m , while it is
statVolt
cm in the CGS system and
they are related as
Volt statVolt 1 statVolt
1 = 104 c−1 = × 10−4 . (1.9)
m cm 3 cm
So far, we have talked about a distribution of discrete charges.
However, in many physical examples, we may have a continuous dis-
tribution of charges. Of course, as we know, electric charge is quan-
tized in units of the charge of the electron. However, the magnitude
of the charge carried by an electron is quite small,
e = 4.8 × 10−10 esu = 1.6 × 10−19 C. (1.10)
Consequently, it is quite meaningful to talk about a continuous distri-
bution of charges. Furthermore, as we know, in quantum mechanics
a particle has a probabilistic description and that the probability
density associated with a particle is a continuous function (of the co-
ordinates) leading, in the case of a charged particle, to a continuous
distribution of charge. For all of these reasons, it is meaningful for us
to develop various concepts for a continuous distribution of charges
which is really quite simple. For example, if ρ(r) represents the vol-
ume density of a continuous charge distribution, then, the electric
field produced by this charge distribution at r has the form
Z
ρ(r′ )
E(r) = d3 r ′ ′ |3
(r − r′ ), (1.11)
V |r − r
where V denotes the volume containing the charge. This can be seen
simply as follows. The total charge contained in a small volume d3 r ′
around r′ is given by ∆q = d3 r ′ ρ(r′ ) which will produce an electric
field at the coordinate r,
∆q ′
′ 3 ′ ρ(r )
∆E(r) = (r − r ) = d r (r − r′ ). (1.12)
|r − r′ |3 |r − r′ |3
1.2 Electric field 7
Integrating this over the volume containing the charge, we obtain

the electric field due to a continuous distribution of charges as given
in (1.11). Similarly, if we have a surface density σ(r) or a linear
density λ(r) of charges, the corresponding expressions for the electric
field would involve a surface integral or a line integral respectively.
However, to make a connection between a continuous and a discrete
charge distribution, we need the concept of the Dirac delta function
which we will discuss next.
The Dirac delta function is one of the most fundamental con-
cepts in the study of microscopic systems and will probably be cov-
ered in greater detail in your study of quantum mechanics. However,
let us discuss here only what we need from the point of view of the
present discussions. The Dirac delta function is, in some sense, a gen-
eralization of the Kronecker delta to the case of continuous indices.
It is denoted by δ3 (r − r′ ) (this is true in three dimensions and, in
general, in n dimensions it is denoted by δn (r − r′ )) and is defined
such that
Z
d3 r ′ f (r′ ) δ3 (r − r′ ) = f (r),
Z
d3 r ′ δ3 (r − r′ ) = 1, (1.13)
for any well behaved function f (r). The second relation in (1.13), in
fact, follows from the first if we choose f (r) = 1. The two relations
imply that the delta function must vanish at points where its argu-
ment does not vanish and that at points where its argument vanishes,
it must diverge (see Fig. 1.2) in such a way that its integral is unity
(namely, the area under the curve is normalized).
This does not correspond to the behavior of any simple function
that we know of. In fact, the Dirac delta function is truly not a func-
tion, rather it can be thought of as a limit of a sequence of functions.
Without going into detail, let us note some explicit representations
for the delta function. In one dimension, for example, we can write
1 sin g(x − x′ )
δ(x − x′ ) = lim
g→∞ π (x − x′ )
Zg Z∞
1 ik(x−x′ ) dk ik(x−x′ )
= lim dk e = e ,
g→∞ 2π 2π
−g −∞
r
α −α(x−x′ )2
δ(x − x′ ) = lim e ,
α→∞ π
8 1 Electrostatics
r′ r
Figure 1.2: A graphical representation of the Dirac delta function

δ3 (r − r′ ).
dθ(x − x′ )
δ(x − x′ ) = , (1.14)
dx
where θ(x − x′ ) is the step function defined to be
(
′
1, for x − x′ > 0,
θ(x − x ) = (1.15)
0, for x − x′ < 0.
The one dimensional representations in (1.14) can be easily gener-

alized to higher dimensions. For example, in Cartesian coordinates,
δ3 (r − r′ ) = δ(x − x′ )δ(y − y ′ )δ(z − z ′ ).
The delta function has some important and useful properties
which can be easily derived from its definition in (1.13). Let us simply
note some of them here (in one dimension)
(δ(x − x′ ))∗ = δ(x − x′ ),

δ(x′ − x) = δ(x − x′ ),
d d
δ(x − x′ ) = − ′ δ(x − x′ ),
dx dx
(x − x′ )δ(x − x′ ) = 0,
d
(x − x′ ) δ(x − x′ ) = −δ(x − x′ ),
dx
1
δ(a(x − x′ )) = δ(x − x′ ), (1.16)
|a|
and so on. It is important to note from the defining relation for the
delta function in (1.13) that it has the dimension of a density (inverse
1.3 Gauss’ law 9
volume of the space on which it is defined). Consequently, it is clear

that we can define the charge density for a discrete charge q1 at the
point r1 as
ρ(r) = q1 δ3 (r − r1 ). (1.17)
It follows then, from the defining relation in (1.11), that the electric
field produced at the coordinate r by a point charge q1 located at the
point r1 is given by
Z
ρ(r′ )
E(r) = d3 r ′ (r − r′ )
|r − r′ |3
Z
q1
= d3 r ′ δ3 (r′ − r1 ) (r − r′ )
|r − r′ |3
q1
= (r − r1 ), (1.18)
|r − r1 |3
as it should be (see (1.6)). A similar derivation follows as well for the

electric field produced by a distribution of point charges.
1.3 Gauss’ law
Let us consider a continuous charge distribution given by the volume

charge density ρ(r) contained in a finite region bounded by a surface
as shown in Fig. 1.3. It is a simple matter to calculate the flux of
electric field out of this surface in the following manner.
z ds
r
r′
y
x
Figure 1.3: A finite volume containing a continuous distribution of

charge.
From the defining relation in (1.11), we obtain the flux of the

10 1 Electrostatics
electric field to be
Z Z Z
ρ(r′ )
ds · E(r) = d3 r ′ ds · (r − r′ ) , (1.19)
|r − r′ |3
S S
where we have interchanged the orders of integration. The surface

integral, on the right hand side, can be simplified by calculating it
around r′ . The infinitesimal surface (area) element at r would, then,
be a vector pointing radially outwards along (r − r′ ) and would have
the form
ds = dΩ |r − r′ |(r − r′ ), (1.20)
where dΩ represents the solid angle subtended at r′ by the infinites-

imal surface area so that we have
ds · (r − r′ ) = dΩ |r − r′ |3 . (1.21)
Let us add here some clarification on the expression for the sur-
face element in (1.20). First, consider an infinitesimal line element
vector in polar coordinates (see Fig. 1.4) which has the form
dℓ
r
θ dθ
x
Figure 1.4: An infinitesimal line element vector in polar coordinates.
dℓ = r dθ θ̂. (1.22)
Now, let us consider a surface (area) element in spherical coordinates

shown in Fig. 1.5. For an infinitesimal change dθ and dφ of the
angular coordinates (and no change in r), we can write the surface
(area) element vector to be (note that θ̂ × φ̂ = r̂)
ds = r dθ θ̂ × r sin θ dφ φ̂ = r 2 sin θ dθ dφ r̂ = r 2 dΩ r̂. (1.23)

1.3 Gauss’ law 11
θ
rd
r r sin θdφ
y
φ
x
Figure 1.5: An infinitesimal surface (area) element vector in spherical

coordinates.
Here dΩ is the solid angle subtended by the surface element at the

origin and we have
Z Zπ Z2π
dΩ = dθ sin θ dφ = 4π, (1.24)
0 0
showing that the total solid angle around a point is 4π. An alternative
way to understand the surface element is to note that the line element
vector in spherical coordinates has the form
dℓ = dr r̂ + r dθ θ̂ + r sin θ dφ φ̂, (1.25)
so that the area along the direction of r̂ is simply given by (here r is

fixed and θ̂ × φ̂ = r̂)
ds = dℓθ × dℓφ = r dθ θ̂ × r sin θ dφ φ̂

= r 2 sin θ dθ dφ r̂ = r 2 dΩ r̂, (1.26)
which is the result obtained earlier in (1.23). Using (1.21), the surface
integral on the right hand side of (1.19) simplifies and we have
Z Z Z
ds · E(r) = d3 r ′ ρ(r′ ) dΩ
S
Z
= 4π d3 r ′ ρ(r′ ) = 4π Qenclosed , (1.27)
12 1 Electrostatics
where we have used (1.24) (namely, the fact that the total solid angle
around r′ is 4π) and Qenclosed represents the total charge enclosed
inside the volume bounded by the surface. Note that if there are
electric charges outside the enclosing surface, then the flux of the
electric field, due to such charges, enters as well as exits the surface
the same number of times (can be more than once depending on the
topology of the volume), thereby canceling (since the directions of
the surfaces at the point of entry and exit are opposite) any further
contribution to the total flux due to charges external to the given
volume. Thus, independent of the shape of the enclosing surface, we
have the general result
Z
ds · E(r) = 4π Qenclosed . (1.28)
S
This is known as Gauss’ law which says that the total electric
flux out of a closed surface equals 4π times the total electric charge
enclosed in the volume bounded by the surface. In fact, (1.28) repre-
sents the integral form of Gauss’ law. We can also write a differential
form for Gauss’ law by appealing to Gauss’ theorem which says that,
for any vector function A(r),
Z Z
d3 r ∇ · A(r) = ds · A(r), (1.29)
V S
where S represents the surface enclosing the volume V . Using this in

(1.28), we obtain
Z
ds · E(r) = 4π Qenclosed ,
S
Z Z
3
or, d r ∇ · E(r) = 4π d3 r ρ(r),
V V
or, ∇ · E(r) = 4π ρ(r). (1.30)
This is the differential form of Gauss’ law. It is worth commenting

here that Gauss’ law is a consequence of the r12 form of the Coulomb
force (1.1). (Otherwise the coordinate dependent factors from the
surface element and the force law would not cancel.) Such laws also
arise for other forces which vary as inverse square of the distance
and, in particular, one can write down a Gauss’ law for gravitation
as well. Furthermore, an experimental test of r12 behavior of the
Coulomb force corresponds to testing Gauss’ law as well.
1.3 Gauss’ law 13
◮ Example (Modified Coulomb’s law). Let us consider the hypothetical example

of a point charge q at the origin producing an electric field at a point r of the form
q
E(r) = r̂. (1.31)
r 2+δ
For δ = 0, we have the Coulomb behavior, but if δ 6= 0 (δ can be positive or
negative), the charge will carry a nontrivial dimensionality.
In this case, the electric flux out of a large sphere of radius R enclosing the
point charge (origin) will be given by
Z Z
q 4πq
ds · E(R) = R2 dΩ 2+δ = δ . (1.32)
R R
Namely, we observe that a modification of the Coulomb’s law will lead to a mod-
ified Gauss’ law where the electric flux out of a sphere will not be 4πQenclosed ,
rather it will depend on the radius of the spherical surface. Consequently, the
electric flux out of a spherical surface will change as the radius of the sphere
changes even though the charge contained inside is the same.
Similarly, recalling that the divergence of a vector A in spherical coordinates
has the form
1 ∂(r 2 Ar ) 1 ∂(sin θAθ ) 1 ∂Aφ
∇·A = + + , (1.33)
r2 ∂r r sin θ ∂θ r sin θ ∂φ
we obtain the modified differential form of Gauss’ law on the surface of a large
sphere of radius R to be

1 ∂ 2 q qδ
∇ · E|R = 2 r 2+δ = − 3+δ , (1.34)
r ∂r r R R
which also differs from the usual differential form of Gauss’ law. ◭
Gauss’ law is quite useful in determining the electric field when

we know the charge distribution. However, the integral form is mean-
ingful (easier to solve) only when there is a symmetry in the problem.
As examples, let us solve some problems with symmetry.
◮ Example (Infinitely long charged wire). Let us consider a thin wire which is
infinitely long along the z-axis and which carries a constant linear charge density
of λ. We would like to determine the electric field produced by such a charge
distribution.
This problem has a cylindrical symmetry and hence we expect the electric
field to point radially perpendicular to the wire and that its magnitude will be the
same at any point on the surface of a cylinder whose axis lies along the wire. (Any
non-radial component of the electric field would vanish by the up-down symmetry
of the system. In other words, any non-radial component of the electric field
due to charges in the upper half of the wire would be canceled exactly by that
due to the charges in the lower half of the wire. The magnitude of the electric
field will be a constant on a cylindrical surface because of the radial symmetry.)
Thus, drawing a cylindrical Gaussian surface of radius r (normally the radius in
cylindrical coordinates is denoted by ρ, but we use r instead to avoid any confusion
with the volume charge density) and height h around the wire as shown in Fig.
1.6, we obtain from Gauss’ law (1.28) (applied to this surface and note that there
14 1 Electrostatics
r
h
Figure 1.6: The dashed surface represents a cylindrical Gaussian sur-

face enclosing a part of the infinitely long thin wire along the z-axis
carrying a constant linear charge density λ.
is no flux through the top or the bottom of the cylinder because of the radial
nature of the electric field),
Z
ds · E(r) = 4π Q,
S
or, |E(r)| 2πrh = 4π hλ,

2λ
or, |E(r)| = , (1.35)
r
and the field points radially perpendicular to the wire. (Normally, one writes this
as E(r) = 2λρ
ρ̂, where ρ represents the radial vector on the plane in cylindrical
coordinates. The cylindrical coordinates are conventionally denoted by (ρ, φ, z).)
We see that the strength of the electric field in this case decreases inversely with
the perpendicular distance from the wire. We also note here that |E(r)| denotes
the magnitude of the electric field, namely, E(r) = |E(r)|r̂. ◭
Exercise. Compare this with the behavior of the electric field due to a point charge.
◮ Example (Charged spherical shell). Let us next consider a spherical shell of

radius R, carrying a uniform distribution of charge characterized by the constant
surface charge density σ. We would like to calculate the electric field due to such
a charge distribution.
Choosing the center of the shell to be the origin, let us note that the problem
has spherical symmetry and, consequently, the only direction that is physically
meaningful is the radial direction from the origin (center of the shell). The electric
field, being a vector, can only point along this direction. Furthermore, because
of the spherical symmetry, the magnitude of the electric field at any point on
the surface of a spherical shell of radius r around the origin (concentric spherical
shell) must be the same. Therefore, let us imagine a spherical Gaussian surface of
radius r > R as shown in Fig. 1.7 and apply Gauss’ law to determine the electric
flux through this surface. This gives,
1.3 Gauss’ law 15
Figure 1.7: The dashed surface represents a spherical Gaussian sur-

face enclosing the spherical shell of radius R carrying a constant sur-
face charge density σ.
Z
ds · E(r) = 4π Q,
S
Z
or, r 2 dΩ |E(r)| = 4π Q,
Q
or, |E(r)| = , (1.36)
r2
where, Q = 4πR2 σ is the total charge carried by the spherical shell. Furthermore,
recalling that the electric field can only point radially, for r > R, we have
Q 4πR2 σ
E(r) = 2
r̂ = r̂. (1.37)
r r2
Namely, outside the spherical shell, the electric field behaves as if the entire charge
on the surface of the shell were located at the center of the shell.
On the other hand, for points inside the shell, if we apply Gauss’ law and
calculate the electric flux through a spherical shell of radius r < R (see Fig. 1.8),
we obtain,
Z
ds · E(r) = 0, (1.38)
S
since there is no charge inside the Gaussian surface and this leads to the fact that
inside the shell the electric field vanishes. Therefore, we determine the electric
field to have the general form
Q 4πR2 σ
E(r) = θ(r − R) r̂ = θ(r − R) r̂, (1.39)
r2 r2
which shows explicitly that the electric field is discontinuous across the charged
surface. This is, of course, reminiscent of the behavior of the gravitational field
which satisfies the inverse square law as well. Let us note from (1.39) that the
16 1 Electrostatics
Figure 1.8: The dashed surface represents a spherical Gaussian sur-

face inside the spherical shell of radius R carrying a constant surface
charge density σ.
discontinuity in the normal component of the electric field across the surface car-
rying charge is given by (R and L refer respectively to right and left of the shell,
or even more appropriate outside and inside the shell)
Q

r̂ · (ER − EL ) = En,R − En,L = 2 = 4πσ. (1.40)
R R R
◭
◮ Example (Charged infinite plane). As another example of a system with sym-

metry, let us consider a thin rectangular plane of infinite extent carrying a constant
surface charge density of σ and we would like to calculate the electric field that it
produces.
From the symmetry of the problem, we realize that the direction of the
electric field at any point would be perpendicular to the plane and its magnitude
will be the same at points whose perpendicular distance from the plane is the
same. Thus, drawing a rectangular Gaussian surface (see Fig. 1.9) whose end
surfaces are equidistant from the plane and have area A, we obtain from Gauss’
law that
Z
ds · E = 4πQ,
S
or, 2|E|A = 4πσA,
or, |E| = 2πσ. (1.41)
Thus, we can write
E(r) = 2πσ n̂, (1.42)
where n̂ represents the outward unit vector normal to the plane. We see that, in
this case, the magnitude of the electric field is constant at every point in space
(although the direction changes on either side of the plane). We note again that
1.4 Potential 17
Figure 1.9: The dashed surface represents a rectangular Gaussian

surface enclosing a part of the infinite plane with a constant surface
charge density σ.
the normal component of the electric field is discontinuous across the surface
carrying charge with the discontinuity given by

n̂ · (ER − EL ) = En,R − En,L = 4πσ, (1.43)
where the restriction refers to the location of the plane and (1.43) can be compared
with (1.40). ◭
This demonstrates how the integral form of Gauss’ law (1.28) is

useful in determining the electric field when there is enough symmetry
in the problem. However, we note that if there is no symmetry in
the problem, the electric field on the Gaussian surface need not be
constant and, consequently, the surface integration becomes more
involved and the method is less useful.
1.4 Potential
We have seen that the electric field produced by a static charge dis-
tribution is a vector given by (1.11). Let us discuss briefly the nature
of this vector for a static distribution of charges. The analysis of
this section obviously may not hold when the charge distribution is
non-static and we will come back to this question later. For the
time being, however, we are interested in solving problems only in
electrostatics and, therefore, this is quite meaningful as we will see
shortly.
18 1 Electrostatics
We have already seen that the electric field for an arbitrary

distribution of charges is given by
Z
ρ(r′ )
E(r) = d3 r ′ (r − r′ ), (1.44)
|r − r′ |3
where ρ(r) represents the charge density of the distribution of charges

in a given volume. Let us recall the identities

1 1
∇ ′
=− (r − r′ ),
|r − r | |r − r′ |3

1 2
∇ ′ 2
=− (r − r′ ), (1.45)
|r − r | |r − r′ |4

1 3
∇ =− (r − r′ ),
|r − r′ |3 |r − r′ |5
and so on, which are readily verified in the Cartesian coordinates.
For example, we note that

1
∇
|r − r′ |

∂ ∂ ∂ 1
= x̂ + ŷ + ẑ p
∂x ∂y ∂z (x − x ) + (y − y ′ )2 + (z − z ′ )2
′ 2
x̂(x − x′ ) + ŷ(y − y ′ ) + ẑ(z − z ′ )

=− 3
((x − x′ )2 + ((y − y ′ )2 + (z − z ′ )2 ) 2
1
=− (r − r′ ), (1.46)
|r − r′ |3
and so on.
Using the last relation in (1.45), it is now clear from the defini-
tion of the electric field in (1.44) that
Z
ρ(r′ )
∇ × E(r) = −3 d3 r ′ (r − r′ ) × (r − r′ )
|r − r′ |5
= 0, (1.47)
where we have also used the familiar vector identity
∇ × r = 0. (1.48)
We could also have seen the identity (1.47) alternatively as follows.

Note that upon using the first relation in (1.45), we can also write
1.4 Potential 19
the electric field (1.44) as

Z
ρ(r′ )
E(r) = −∇ d3 r ′ , (1.49)
|r − r′ |
from which it follows that
Z
ρ(r′ )
∇ × E(r) = −∇ × ∇ d3 r ′ = 0. (1.50)
|r − r′ |
Namely, we see from the definition of a static electric field that its
curl vanishes and, consequently, the electric field, in this case, is a
conservative field. (This is not necessarily true if the source charges
are not static as we will see later.) We know from Stokes’ theorem
that, for any vector function A(r),
Z I
ds · (∇ × A) = dℓ · A, (1.51)
C
S
where C represents a closed contour which bounds the surface S.

Since the curl of the electric field vanishes, it follows from Stokes’
theorem that along any closed curve
I
dℓ · E = 0, (1.52)
C
so that the value of the line integral of the electric field between any
two points depends only on the end points independent of the path
of integration.
It is well known that any vector, which falls off rapidly at infinite,
can be decomposed uniquely into the sum of two vectors, one of which
is divergence free while the other has vanishing curl. This is com-
monly known as the Helmholtz theorem. Explicitly, the Helmholtz
theorem says that, if A is a vector function, we can write it as
A = B + C, (1.53)
where
∇ · B = 0 = ∇ × C. (1.54)
Furthermore, we can easily express B and C in terms of A to write

Z
1 ∇′ × A(r′ )
A(r) = ∇ × d3 r ′
4π |r − r′ |
Z
1 ∇′ · A(r′ )
− ∇ d3 r ′ , (1.55)
4π |r − r′ |
20 1 Electrostatics
which can be readily verified by using the identity

2 1
∇ = −4π δ3 (r − r′ ). (1.56)
|r − r′ |
A more physical way to understand the Helmholtz theorem is
to recognize that any vector can always be uniquely decomposed into
a longitudinal and a transverse component with respect to a given
direction. The gradient operation (∇) indeed provides a direction
and any vector can be decomposed into a parallel and a perpendicular
component with respect to this direction. In modern parlance, one
says that a vector can be projected parallel to the direction of ∇ and
perpendicular to it and writes formally (repeated indices are summed)

∇i ∇j ∇i ∇j
Ai = + δij − Aj , (1.57)
∇2 ∇2
where it is understood that ∇2 = ∇i ∇i with all the components

(indices) summed. Furthermore, written this way, it is quite clear
that
(L) ∇i ∇j
Pij = ,
∇2
(T) ∇i ∇j
Pij = δij − , (1.58)
∇2
denote respectively the longitudinal (curl free) and the transverse
(divergence free) projection operators with respect to the gradient.
The Helmholtz theorem is particularly interesting in the case
of a static electric field because, in this case, the electric field has
vanishing curl. Consequently, we see that for a static distribution
of charges we can write the electric field as the gradient of a scalar
function (the negative sign is a convention whose rationale would
become clear when we deal with relativistic systems)
E(r) = −∇Φ(r), (1.59)
and the form of the scalar function follows from the general form of
an arbitrary vector in (1.55) (from the Helmholtz theorem) to be
Z ′ ′ Z ′
1 3 ′ ∇ · E(r ) 3 ′ ρ(r )
Φ(r) = d r = d r . (1.60)
4π |r − r′ | |r − r′ |
Here, we have used the differential form of Gauss’ law (1.30) in the
last step. Let us note that we could have derived the result in (1.60)
1.4 Potential 21
also directly from the definitions of the electric field in (1.49) and
(1.59), namely,
Z
ρ(r′ )
E(r) = −∇ d3 r ′ = −∇Φ(r),
|r − r′ |
Z
ρ(r′ )
or, Φ(r) = d3 r ′ , (1.61)
|r − r′ |
where we have ignored a constant of integration (which is related to
the choice of a reference point, as we will see). Alternatively, from
(1.59) we can also write
Zr
Φ(r) = − dℓ · E. (1.62)
∞
The scalar function Φ(r) is called the potential or the scalar

potential or the potential function (not to be confused with the po-
tential energy) and it is clear that if we know Φ(r), we can determine
the electric field simply by the gradient operation. The surfaces on
which Φ(r) is a constant are known as equipotential surfaces. Such
surfaces are important in the study of electrostatics because the elec-
tric field lines are normal to them (because of the gradient nature of
the electric field). The importance of Φ(r) lies in the fact that it is a
scalar function and, consequently, is much easier to handle than the
vector field E(r). Going back to the differential form of Gauss’ law in
(1.30), we see that with the identification in (1.59) the potential for a
given distribution of charges satisfies the partial differential equation
∇ · E(r) = 4πρ(r),
or, ∇2 Φ(r) = −4πρ(r), (1.63)
which is known as the Poisson equation. In regions where there are
no source charges, the Poisson equation reduces to the form
∇2 Φ(r) = 0, (1.64)
which is known as the Laplace equation. (∇2 is conventionally called
the Laplacian.) Solving problems in electrostatics, therefore, cor-
responds to solving the Poisson equation or the Laplace equation
subject to appropriate boundary conditions. (Once the potential is
determined, the electric field can be determined by taking the gra-
dient.) We will develop general methods for solving these equations
later when the meaning of the particular integral representation of
Φ(r) obtained in (1.60) will also become clear.
22 1 Electrostatics
◮ Example. Let us consider a charge distribution that produces an electric field

of the form
qr
E(r) = a , (1.65)
r
where a is a real parameter and a 6= 3. (For a = 3, we have the electric field of
a point charge that we have already studied in (1.6).) We would like to calculate
the charge density that produces such an electric field as well as the potential
associated with it.
Using (1.45) and (1.48) in (1.65) we note that
∇ × E(r) = 0, (1.66)
namely, the electric field (1.65) is curl free. Second, we obtain the divergence of
the electric field to be
r
∇ · E(r) = q∇ · a
r

1 1
=q ∇ · r + ∇ · r
ra ra

a 3 (3 − a)q
= q − a+2 r · r + a = . (1.67)
r r ra
Comparing this with the differential form of Gauss’ law (1.30), we identify the
charge density that produces such an electric field to be
(3 − a)q
ρ(r) = . (1.68)
4πr a
We can use (1.62) to calculate the potential as
Zr
Φ(r) = − dℓ · E
∞
Zr
dr ′ qr 2−a
= −q =− , (1.69)
(r ′ )a−1 2−a
∞
where we have used the fact that the electric field is conservative (curl free) and
correspondingly chosen a radial path to do the line integral. (Recall that the
electric field is along the radial direction. We have also thrown away a divergent
constant for the case a < 2.) Note that when a = 2, the above expression needs
to be calculated in a limiting manner and gives
Φa=2 (r) = −q ln r, (1.70)
where we have thrown away an infinite constant.

This problem can also be studied in an alternative manner as follows. Note
from (1.65) that the electric field can be rewritten as

qr q 1 q
E(r) = a = − ∇ = ∇ r 2−a . (1.71)
r a−2 r a−2 2−a
It follows from this that the electric field is curl free
∇ × E(r) = 0, (1.72)
1.4 Potential 23
and the potential can be identified with
qr 2−a
Φ(r) = − , (1.73)
2−a
as derived earlier. Furthermore, the divergence of the electric field is given by

2 q 2 2−a q 1 ∂ 2 ∂
∇ · E(r) = −∇ Φ(r) = ∇ (r )= r r 2−a
2−a 2 − a r 2 ∂r ∂r
q 1 ∂ (3 − a)q
= (2 − a)r 3−a = , (1.74)
2 − a r 2 ∂r ra
which coincides with the earlier result and leads to the charge density through
Gauss’ law. ◭
◮ Example. As another example, let us consider the Yukawa potential that we

have discussed earlier. Namely, let us assume that the potential due to a charge
distribution has the form
q e−µr
Φ(r) = , (1.75)
r
where µ represents a mass scale.
In this case, the electric field can be easily calculated to have the form
∂ e−µr
E(r) = −∇Φ(r) = −qr̂
∂r r
qr̂
= (1 + µr) e−µr . (1.76)
r2
We note that when µ = 0, this reduces to the Coulomb field (1.6) for a point
charge at the origin.
We can also calculate the charge density that produces this potential by
using the differential form of Gauss’ law. Namely, we note that
r
∇ · E(r) = q∇ · (1 + µr) e−µr
r3
h r r i
= q ∇ · 3 (1 + µr) e−µr + 3 · ∇ (1 + µr) e−µr
r r

1 r
= q − ∇2 (1 + µr) e−µr + 3 · −µ2 r e−µr
r r
2

µ −µr
= q 4πδ 3 (r) (1 + µr) e−µr − e
r

µ2 −µr
= q 4πδ 3 (r) − e = 4πρ(r). (1.77)
r
Here we have used (1.56) in the intermediate step. This determines the charge
density associated with the Yukawa potential to be
qµ2 −µr
ρ(r) = q δ 3 (r) − e . (1.78)
4πr
We note that for µ = 0, this reduces to (1.17). ◭
24 1 Electrostatics
Meaning of the potential. To get a feeling for the meaning of the

potential function, let us consider a test charge q being brought in
from a reference point rA to the point r in the presence of an electric
field E. Since the electric field exerts a force on the charged particle,
work has to be done to move the electric charge and the amount of
work needed to bring it to r is given by
Zr Zr Zr
W =− dℓ · F = −q dℓ · E = q dℓ · ∇Φ
rA rA rA
= q(Φ(r) − Φ(rA )). (1.79)

Namely, the work done is given by the difference of the potential at
the two coordinates up to the multiplicative factor of the charge of the
test particle. This is indeed a consequence of the conservative nature
of the electric field and the result is independent of the path along
which the test charge is brought to the final point. Furthermore, if
we choose the reference point to be at infinity, where the potential for
most physical systems vanishes (namely, if we assume that the electric
field vanishes at infinity as is done in the Helmholtz theorem), then,
we can write
W = qΦ(r). (1.80)
Namely, we can think of the potential as the work done in bringing
a unit charge from spatial infinity to a given coordinate point in the
presence of an electric field. The work done, of course, can be thought
of as the potential energy of the charge. The potential is measured in
Volts in the MKS (SI) system whereas the unit in the CGS system is
statVolt. From the definition of the potential above, it is quite clear
that
Joule 107 erg
1 Volt = 1 =
C 10 c esu
= 106 c−1 statVolt
1
= × 10−2 statVolt, (1.81)
3
where we have used (1.3). Eq. (1.81) is consistent with (1.9).
1.5 Electrostatic energy
Let us next calculate the electrostatic potential energy for a given dis-
tribution of charges. As we will see, the form of the result is different
1.5 Electrostatic energy 25
depending on whether we are calculating the energy for a discrete

distribution of charges or a continuous distribution of charges. So,
let us analyze this feature in some detail. First, let us assume that we
have a discrete distribution of point charges. Let us not worry about
how the charges were produced and assume that the potential energy
for such a system of charges, when they are infinitely separated from
one another, is zero. We want to calculate the electrostatic energy
associated with such a distribution of charges. To this end, we note
that, given a charge q1 at r1 , if we bring in a second charge q2 to the
coordinate r2 , the work done (and hence the potential energy) would
be given by
Zr2
W12 = − dℓ · F12
∞
Zr2
(r − r1 )
= −q1 q2 dr ·
|r − r1 |3
∞
2 −r1
rZ
r̂
= −q1 q2 dr ·
|r|2
∞
|rZ
1 −r2 |
dr
= −q1 q2
r2
∞
q1 q2
= , (1.82)
|r1 − r2 |
where |r| = r and, since the integral is independent of the path, owing
to the conservative nature of the electric field, we have chosen a radial
path in evaluating the integral. We can now keep adding more and
more charges and since the electric force is additive, the calculation
simplifies. For example, to bring in a third charge q3 to the point r3 ,
the total work done is given by
Zr3
W123 = W12 − dℓ · (F13 + F23 )
∞
q1 q2 q1 q3 q2 q3
= + + . (1.83)
|r1 − r2 | |r1 − r3 | |r2 − r3 |
It is obvious that this expression is completely symmetric in the per-
mutation of any pair of charged particles (charges and coordinates)
26 1 Electrostatics
and hence, the order in which the charges are brought in does not
matter. Carrying this out for n charged particles, it is easily obtained
that the total work required is
1 X qi qj
W = , (1.84)
2 |ri − rj |
i,j,i6=j
where the factor of 12 is there to avoid double counting. This is,

therefore, the electrostatic energy for a distribution of point charges.
Let us next calculate the electrostatic energy for a continuous
distribution of charges. Once again, generalizing (1.84) to a continu-
ous distribution of charges, we can write
Z
1 ρ(r)ρ(r′ )
W = d3 r d3 r ′
2 |r − r′ |
Z Z ′ Z
1 3 3 ′ ρ(r ) 1
= d r ρ(r) d r = d3 r ρ(r)Φ(r)
2 |r − r′ | 2
Z
1
=− d3 r ∇2 Φ(r) Φ(r)
8π
Z
1
= d3 r (∇Φ(r)) · (∇Φ(r))
8π
Z
1
= d3 r E2 (r), (1.85)
8π
where we have used the differential form of Gauss’ law in (1.63) (Pois-
son equation) as well as the relation between the electric field and the
potential. We have also neglected surface terms in the integration by
parts with the assumption that the electric field falls off rapidly at
infinity.
The difference between the two cases needs to be discussed.
First, we note that the energy for a continuous distribution of charges
is completely given in terms of the electric field and is non-negative.
On the other hand, for a distribution of point charges the sign of the
energy in (1.84) clearly depends on the signs of various charges and
can, in fact, be negative. Let us recall that the electrostatic energy
for a pair of similarly charged particles is positive while it is neg-
ative if the charges of the two particles are opposite in sign. This
difference in the behavior of the electrostatic energy between a dis-
crete and a continuous distribution of charges arises mainly because
of our choice of the reference system in the case of point charges.
We assumed the zero of the energy to correspond to the system of
1.6 Selected problems 27
point charges separated by an infinite distance thereby discarding the

self-energy associated with the point charges. (Experimentally, one
measures only the differences in energy and, therefore, this is quite
acceptable.) The reason for such a choice is that the self-energy for a
point charge is known to diverge and hence, it is meaningful to throw
away such a divergent constant in the calculation. In contrast, the
expression for the electrostatic energy of a continuous distribution of
charges is more complete in that it contains the self-energy.
To summarize what we have learnt so far, all of electrostatics
can be described by the two equations
∇ × E = 0 =⇒ E(r) = −∇ Φ(r), (1.86)
and
∇ · E(r) = 4π ρ(r) =⇒ ∇2 Φ(r) = −4π ρ(r). (1.87)
These equations can also be written in the respective integral forms

as
I
dℓ · E = 0,
C
Z
ds · E = 4π Q. (1.88)
S
Furthermore, the solution of the Poisson equation has the general

form
Z
ρ(r′ )
Φ(r) = d3 r ′ . (1.89)
|r − r′ |
It is worth remarking here that the true dynamical equations of
electromagnetism (which we will study later) are coupled equations,
involving both electric and magnetic fields, as one would expect from
relativistic invariance. However, in the static limit these equations
decouple, making it possible and meaningful to study electrostatics
and magnetostatics separately.
1.6 Selected problems
1. Prove the following relations from vector analysis (A and B are

three dimensional vectors)
∇ × (∇ × A) = ∇(∇ · A) − ∇2 A,
28 1 Electrostatics
∇ · (A × B) = B · (∇ × A) − A · (∇ × B),
∇ · r = 3, ∇ × r = 0,

1
∇2 = −4πδ3 (r).
|r|
2. Show that if Tij , i, j = 1, 2, 3, denotes a second rank three

dimensional tensor, then a generalization of Gauss’ theorem
would lead to Z Z
d3 r ∂j Tij = dsj Tij ,
V S
where S denotes the surface bounding the volume V and re-

peated indices are assumed to be summed.
3. Using the result from the last problem show that for an arbi-
trary vector A we have
Z Z
3
d r ∇ × A = ds × A,
V S
where S denotes the surface bounding the volume V .
4. Given the following two vectors, E, which do you think would

describe a true static electric field?
(i) E = k[xyx̂ + 2yzŷ + 3xzẑ],

(ii) E = k[y 2 x̂ + (2xy + z 2 )ŷ + 2yzẑ].
Here, k is a constant. For the true electric field, determine the

potential with the origin as the reference point.
5. Consider the hypothetical case that the Yukawa potential given

by
e−µr
Φ(r) = q ,
r
where r = |r| and µ is a mass parameter (in units of c, ~) is
produced by a point charge q at the origin. Would Gauss’ law
be valid for such a case? Show that, for r 6= 0, this potential
satisfies the equation
∇2 Φ(r) = µ2 Φ(r).
6. Calculate the electric field inside and outside of a solid sphere

of radius R carrying a constant volume charge density ρ.
7. Consider a spherical region V of radius R without any charge

and a point charge q outside the spherical region at a distance
r from the center. Evaluate explicitly the net electric flux out
of the surface S which bounds the spherical region V .
8. Consider a spherical shell of radius R with a uniform surface

charge density σ. What is the electrostatic energy stored in such
a system. What happens as the radius of the sphere decreases?
What is the electrostatic energy for a solid sphere of radius R
carrying a uniform volume charge density ρ?
9. Consider a spherical distribution of charge for which the volume

charge density is nonzero only for 0 ≤ r ≤ R and has the form
ρ(r) = k r −n ,
where both k and n are positive constants. Calculate the elec-

trostatic energy associated with such a distribution of charges.
For what values of n is the energy finite?
Chapter 2
Potential for simple systems
Let us next calculate the potential function for some simple systems
with known charge distributions, to get a feeling for its properties.
The simplest example, of course, is the potential for a point charge.
Let us assume that a charge q1 is located at the coordinate r1 . In
such a case, we can write the charge density as (see (1.17))
ρ(r) = q1 δ3 (r − r1 ), (2.1)
so that the potential at the coordinate r is easily obtained to be

Z
ρ(r′ )
Φ(r) = d3 r ′
|r − r′ |
Z
q1
= d3 r ′ δ3 (r′ − r1 )
|r − r′ |
q1
= , (2.2)
|r − r1 |
as we would expect. Similarly, we can also calculate the potential for
a system of point charges. One thing to note from this calculation is
the fact that the potential for a single charge is a continuous function
of r. Furthermore, it vanishes as |r| → ∞ for a fixed r1 . This is
consistent with our choice of the reference point for the potential,
namely, that the potential vanishes for infinite separation.
◮ Example (Hydrogen atom). The electric charge distribution (due to the elec-
tron) in the ground state of the Hydrogen atom is given by
q − 2r
ρ(r) = e a, (2.3)
πa3
where q represents the charge of the electron and a, the Bohr radius. We would
like to calculate the potential as well as the electric field due to such a charge
distribution. We note that the total charge of the distribution is given by
Z Z∞ Z∞
3 q 2 − 2r q q
d r ρ(r) = (4π) dr r e a = dr r 2 e−r = Γ(3) = q, (2.4)
πa3 2 2
0 0
31
32 2 Potential for simple systems
which is to be expected since the charge density in (2.3) is simply q times the
(quantum mechanical) probability density in the ground state.
The calculation of the potential and the electric field can be carried out
in one of two ways. First, let us note that the charge distribution is spherically
symmetric. Consequently, we expect the electric field as well as the potential
to reflect this. Using Gauss’ law to integrate the electric field over a spherical
(Gaussian) surface of radius r we obtain
Z Z
ds · E = 4π d3 r ′ ρ(r′ )
Z Zr
2 q − 2ra′
or, 4πr |E(r)| = 4π dΩ r ′2 dr ′ e
πa3
0
Zr
q 2r′
or, |E(r)| = (4π) dr ′ r ′2 e− a
πa3 r 2
0
2r
Z
a
q ′
= dr ′ r ′2 e−r
2r 2
0
q h ′
i 2r
a
= 2 (−r ′2 − 2r ′ − 2)e−r
2r 0

q 2r 2r 2 2r
= 2 1− 1+ + 2 e− a . (2.5)
r a a
Therefore, we conclude that the radial electric field is given by

qr̂ 2r 2r 2 2r
E(r) = 2 1 − 1 + + 2 e− a . (2.6)
r a a
The potential is then easily determined from

Zr
Φ(r) = − dℓ · E(r′ )
∞
Zr
q 2r ′ 2r ′2 − 2r
′
=− dr ′ 1 − 1 + + e a
r ′2 a a2
∞
Zr
d 1 r′ 2r′
=q dr ′ 1− 1+ e− a
dr ′ r′ a
∞
qh r − 2r i
= 1− 1+ e a . (2.7)
r a
The second way of solving the problem is to note that given the charge
density, we can obtain the potential simply as
Z Z 2r′
ρ(r′ ) q e− a
Φ(r) = d3 r ′ = d3 r ′
|r − r′ | πa3 |r − r′ |
2.1 Potential for a thin spherical shell 33
Z∞ Z1
q ′ ′2 − 2r
′
d cos θ
= (2π) dr r e a
1
πa3 (r 2 + r ′2 − 2rr ′ cos θ) 2
0 −1
Z∞
q (−2π) 2r′
= dr ′ r ′ e− a |r − r ′ | − (r + r ′ )
πa3 r
0
 
Zr Z∞
4q  − 2r
′
− 2r
′
= dr ′ r ′2 e a +r dr ′ r ′ e a 
ra3
0 r
qh r − 2r i
= 1− 1+ e a . (2.8)
r a
This is the same result as obtained in the earlier method in (2.7). The electric
field now follows from the definition
∂ qh r − 2r i
E(r) = −∇Φ(r) = −r̂ 1− 1+ e a
∂r r a
2

qr̂ 2r 2r 2r
= 2 1− 1+ + 2 e− a , (2.9)
r a a
which is what we had obtained earlier in (2.6). ◭
Before we proceed further with calculations, let us recall some

of the results from our earlier analyses in the last chapter. We have
noted from the calculations of the electric fields in the presence of
surface charges that the electric field is discontinuous across a surface
carrying charge. In fact, the general result is (see (1.40))
n̂ · (ER − EL ) = 4πσ, (2.10)
where ER and EL represent the electric fields infinitesimally close to

the surface to the right and to the left respectively and n̂ denotes
the unit vector normal to the surface on the right. In contrast, the
tangential components of the electric fields are continuous across the
surface. Let us next examine the behavior of the electric potential
across a surface carrying a charge density.
2.1 Potential for a thin spherical shell
Let us consider a thin spherical shell of radius R with a constant

surface charge density σ as shown in Fig. 2.1. When we have a
surface charge density, as opposed to a volume charge density, the
expression for the potential can be rewritten as a surface integral
Z
σ(r′ )
Φ(r) = ds′ . (2.11)
|r − r′ |
R σ
Figure 2.1: A thin spherical shell of radius R with a constant surface

charge density σ.
Alternatively, we note that we can express the surface charge density

as a volume charge density through the use of the Dirac delta func-
tion, which in the present case takes the form (remember that the
delta function has the inverse dimension of its argument)
ρ(r) = σδ(r − R). (2.12)
Using this, then, we obtain from (2.1),

Z
ρ(r′ )
Φ(r) = d3 r ′
|r − r′ |
Z
σδ(r ′ − R)
= r ′2 dr ′ sin θ ′ dθ ′ dφ′ 1
(r 2 + r ′2 − 2rr ′ cos θ ′ ) 2
Z1
dα
= 2πR2 σ 1
(r 2 + R2 − 2rRα) 2
−1

2 1 1 1
= 2πR σ − (r 2 + R2 − 2rRα) 2
rR −1

1 p p
= 2πR2 σ − (r − R)2 − (r + R)2 . (2.13)
rR
Here, we have simplified our calculation by choosing the z-axis to be

parallel to r and have defined α = cos θ ′ in the intermediate step.
In evaluating the final form of the potential, we have to be careful
in choosing the positive square root, particularly in the first factor.
2.1 Potential for a thin spherical shell 35
Recognizing that Q = 4πR2 σ, we obtain from (2.13)
2πRσ
Φ(r) = (r + R − |r − R|)
r

 4πR2 σ = Q for r > R,
r r
= (2.14)
 4πR2 σ = Q
for r < R.
R R
We can also rewrite this as

Q Q
Φ(r) = θ(r − R) + θ(R − r) . (2.15)
r R
There are several things to observe from the result in (2.15) (or
(2.14)). First, the potential is spherically symmetric, as it should be
because of the symmetry in the problem. Second, it is a continuous
function across r = R, namely, across the surface carrying the charge.
This has to be contrasted with the behavior of the electric field. Out-
side the shell, the potential behaves as if all the charge were located
at the origin. We also note that the potential is a constant inside
the shell. In fact, the value of the potential at the origin is readily
seen to be the average of the potential over any closed surface within
the shell. This, as we will see, is a general feature of the solutions
of Laplace equation. Finally, given the potential, we can determine
the electric field by taking the gradient (see (1.59)). Recalling from
(2.15) that the potential only depends on the radial coordinate, we
obtain
E(r) = −∇Φ(r)

∂ Q Q
= −r̂ θ(r − R) + θ(R − r)
∂r r R
Q
= θ(r − R) r̂. (2.16)
r2
The important thing to note in this derivation is that the derivatives
of the two theta functions give delta functions of opposite sign which
cancel each other. Eq. (2.16) is, of course, our previous result ob-
tained in (1.37) and (1.38), namely, the electric field is non-vanishing
only outside the shell and, at such points, it behaves as if the to-
tal charge were concentrated at the origin. Furthermore, expressed
as in (2.16), the discontinuity in the electric field across the surface
carrying charge is manifest.
2.2 Potential for an infinitely long wire
Although this problem is quite simple, it is worth going through the

derivation which brings out some particular property of the choice
of the reference point. Let us consider an infinitely long and thin
wire carrying a constant linear charge density λ. For simplicity, we
assume the wire to lie along the z-axis as shown in Fig. 2.2.
Figure 2.2: An infinitely long wire along the z-axis carrying a constant
linear charge density λ.
In this case the problem has cylindrical symmetry and, conse-

quently, it is meaningful to use cylindrical coordinates (see Fig. 2.3)
given by
z
r
φ y
x ρ̃
Figure 2.3: Cylindrical coordinates (ρ̃, φ, z).
x = ρ̃ cos φ,
y = ρ̃ sin φ,
z = z, (2.17)
2.2 Potential for an infinitely long wire 37
where ρ̃ represents the radial coordinate on the plane z = 0 (z =

constant). (Normally, it is denoted as ρ. Here, we have added a tilde
to distinguish it from the volume charge density.)
Once again, since we have a problem with a linear charge density
along the z-axis, we can write the potential as a line integral
Z
λ(r′ )
Φ(r) = dz ′ . (2.18)
|r − r′ |
Alternatively, we can express the linear density as a volume density,
which, for the present problem, can be done in the following man-
ner. We note that the wire carrying charge lies along the z-axis and,
therefore, the charge density is nonzero only for x = 0 = y. Thus, we
can write a volume charge density for the system as
λ
ρ(r) = λ δ(x)δ(y) = δ(ρ̃), (2.19)
2π ρ̃
where the normalization factor of 2π1 ρ̃ arises from the observation
that if, δ(x)δ(y) = cδ(ρ̃) (c has to be independent of the angle φ by
rotational symmetry), then,
Z
dx dy δ(x)δ(y) = 1,
Z
or, ρ̃ dρ̃ dφ cδ(ρ̃) = 1,
1
or, c = . (2.20)
2π ρ̃
With (2.19), we can now calculate the potential due to a thin
wire from the definition in (2.1).
Z
ρ(r′ )
Φ(r) = d3 r ′
|r − r′ |
Z
λ δ(ρ̃′ )
= ρ̃′ dρ̃′ dφ′ dz ′
2π ρ̃′ ((ρ̃ − ρ̃′ )2 + (z − z ′ )2 ) 12
Z∞
dz ′
=λ 1
(ρ̃2 + (z − z ′ )2 ) 2
−∞
Z∞
dz̄
=λ 1
(ρ̃2 + z̄ 2 ) 2
−∞
p ∞

= λ ln z̄ + ρ̃2 + z̄ 2 . (2.21)
−∞
It is clear that the right hand side of (2.21) diverges and the reason
for this is not hard to see. In writing an expression for the poten-
tial, we had chosen the potential to vanish at infinity which we had
taken as a reference point. However, in the present problem, such
a choice is not consistent simply because the charge density extends
to spatial infinity. (We have an infinitely long wire as is clear from
the integration limits.) The proper way to analyze this problem is
to recognize that we must choose a different reference point for this
problem (or equivalently allow for a constant potential at infinity).
In particular, let us note that a different choice of the reference point
simply corresponds to adding a constant term to the potential, be it
infinite. Thus, we can extract the finite meaningful potential from
(2.21) by writing
p Λ

Φ(r) = lim λ ln z̄ + ρ̃2 + z̄ 2
Λ→∞ −Λ
p
ρ̃2 + Λ2 )
(Λ +
= lim λ ln p
Λ→∞ (−Λ + ρ̃2 + Λ2 )
p
(Λ + ρ̃2 + Λ2 )2
= lim λ ln
Λ→∞ ρ̃2
4Λ2 + 2ρ̃2 + O( Λ12 )
≈ lim λ ln
Λ→∞ ρ̃2
= −2λ ln ρ̃ + constant, (2.22)
where the constant, on the right hand side, is a divergent constant.
(The important thing to remember is that the potential is not observ-
able, but the electric field is through the electric force. The electric
field is obtained from the potential by the gradient operation so that
a constant term in the potential does not contribute to the electric
field.) Thus, discarding the constant we determine the potential for
the infinitely long charged wire to be
Φ(r) = −2λ ln ρ̃. (2.23)
We see that the potential has cylindrical symmetry and it is con-
tinuous. The electric field can be obtained from (2.23) (or (2.22))
by taking the gradient and since the potential only depends on the
coordinate ρ̃, we obtain
ˆ ∂ 2λ ˆ
E(r) = −∇Φ(r) = −ρ̃ (−2λ ln ρ̃) = ρ̃, (2.24)
∂ ρ̃ ρ̃
which is the same result that we had obtained earlier in (1.35).
2.3 Potential for a circular charged disc 39
2.3 Potential for a circular charged disc
Let us next consider a thin circular disc of radius R which carries a

uniform surface charge density σ. For simplicity, we choose the disc
to lie in the x − y plane (at z = 0) with the center at the origin of
the coordinate system as shown in Fig. 2.4.
R σ
Figure 2.4: A thin circular disc of radius R carrying a uniform surface

charge density σ.
Let us calculate the potential due to the disc at points along

the z-axis, namely, at (x = 0, y = 0, z). Clearly, this is a problem
with cylindrical symmetry and it is meaningful to use the cylindrical
coordinates r = (r, φ, z) (namely, we are simply going to write r for
what we called ρ̃ before). Then, as before, we can write the surface
density of charge as a volume density of the form
ρ(r) = σ δ(z) θ(R − r). (2.25)
Note that the theta function implements the finite extension of the
disc (and, therefore, the charge distribution).
We can now calculate the potential at a point on the z-axis
(x = 0 = y) simply as
Z
ρ(r′ )
Φ(x = 0, y = 0, z) = d3 r ′
|r − r′ |
Z
σδ(z ′ )θ(R − r ′ )
= r ′ dr ′ dφ′ dz ′ 1
(r ′2 + (z − z ′ )2 ) 2
ZR
r ′ dr ′
= 2πσ 1
(r ′2 + z 2 ) 2
0
1 R
= 2πσ r ′2 + z 2 2
0
p
= 2πσ R2 + z 2 − |z| . (2.26)
This shows that the potential is a continuous function. Note that,

at very far off distances, namely, when |z| >> R, we can expand the
potential in (2.26) in a power series to obtain
1 !
R2 2
Φ(x = 0, y = 0, |z| >> R) = 2πσ |z| 1 + 2 − |z|
z
πR2 σ Q
≈ = , (2.27)
|z| |z|
where Q = πR2 σ is the total charge contained on the disc. Thus,

we see that very far away from the disc, the potential along the
z-axis behaves as if all the charge were concentrated at the origin
(like a point charge). We also note that, at the center of the disc
(x = 0, y = 0, z = 0), the potential has the value
Φ(x = 0, y = 0, z = 0) = 2πσR. (2.28)
Furthermore, the electric field along the z-axis can be obtained from
the potential in (2.26) by taking the gradient which gives
E(x = 0, y = 0, z) = −∇Φ(x = 0, y = 0, z)
∂ p
= −ẑ 2πσ R2 + z 2 − |z|
∂z

z
= −2πσ √ − sgn(z) ẑ. (2.29)
R2 + z 2
Here, sgn (z) stands for the sign of z which can also be represented
as an alternating step function. It is obvious from (2.29) that very
close to the disc, namely, for z ≈ 0, the electric field has the leading
behavior
E ≈ sgn(z) 2πσ ẑ. (2.30)
First of all, this explicitly shows that the electric field is discontinuous
across the surface. But, more interestingly, the form of the electric
field, at such points, is like the electric field for the infinite plane
which we have calculated earlier (see (1.42)). This can be understood
qualitatively as a consequence of the fact that very close to the disc,
2.3 Potential for a circular charged disc 41
σ y
x
Figure 2.5: A thin circular disc of radius R carrying a uniform surface

charge density σ.
the disc appears to be of infinite extent even though it has a finite

size.
For this system (see Fig. 2.5), the value of the potential (or
the electric field) at a general point is not easy to obtain in closed
form, nor does it provide any meaningful insight. However, we can
calculate the value of the potential at the rim of the disc quite easily.
By rotational invariance on the plane, the value of the potential will
be the same at any point on the rim and, for simplicity, we choose
the point (x = R, y = 0, z = 0) to calculate the potential. Then, the
potential has the form
Φ(x = R, y = 0, z = 0)
Z
ρ(r′ )
= d3 r ′
|r − r′ |
Z
σδ(z ′ )θ(R − r ′ )
= r ′ dr ′ dφ′ dz ′ 1
((R − r ′ cos φ′ )2 + r ′2 sin2 φ′ + z ′2 ) 2
ZR Z2π
r ′ dr ′ dφ′
=σ 1
(r ′2 − 2Rr ′ cos φ′ + R2 ) 2
0 0
Z2π p
′
=σ dφ r ′2 − 2Rr ′ cos φ′ + R2
0
p R
+ R cos φ′ ln 2 r ′2 − 2Rr ′ cos φ′ + R2 + 2r ′ − 2R cos φ′
0
Z2π
′
p
= σR dφ 2(1 − cos φ′ ) − 1
0
p
′ 2(1 − cos φ′ ) + (1 − cos φ′ )
+ cos φ ln
(1 − cos φ′ )
Z2π !
φ′ 1
= σR dφ′ 2 sin − 1 + cos φ′ ln 1 + ′ . (2.31)
2 sin φ2
0
Here, in the intermediate steps, we have used some standard integrals
from the tables (for example, see Gradshteyn and Ryzhik, 2.261 and
2.264). Finally, integrating by parts the last term inside the bracket
in (2.31), we have (the first term in the integration by parts vanishes
at the limits)
! Z2π 2
Z2π φ′
1 cos 2
dφ′ cos φ′ ln 1 + φ′
= dφ′
sin 2 φ′
0 0
1 + sin 2
Z2π
′ φ′
= dφ 1 − sin . (2.32)
2
0
Using this in (2.31), we obtain
Z2π
′ φ′ φ′
Φ(x = R, y = 0, z = 0) = σR dφ 2 sin − 1 + 1 − sin
2 2
0
Z2π
φ′
= σR dφ′ sin
2
0
= 4σR. (2.33)
It is interesting to compare this with the value of the potential at
the center of the disc (2.28), which makes it clear that the potential
decreases as we move out of the center of the disc. Consequently,
there must exist a radial component of the electric field on the disc
itself.
2.4 Potential for a charge displaced along the z-axis
The next example that we consider is really not that different from
what we have studied earlier and yet has many features which will be
2.4 Potential for a charge displaced along the z-axis 43
useful later. Let us consider a point charge q located at (x = 0, y =

0, z = R) as shown in Fig. 2.6.
ẑ
r−R
q, R r b
y
x
Figure 2.6: A point charge q displaced from the origin along the z-axis
by a distance R.
In this case, we can write the charge density as

q
ρ(r) = qδ(x)δ(y)δ(z − R) = δ(r − R)δ(θ), (2.34)
2πr 2 sin θ
where, in the second step, we have rewritten the charge density in

spherical coordinates. Note that the multiplicative factor arises from
the normalization of the delta function and that, even though the
factor appears to be singular (for example, sin θ in the denominator),
it is, in fact, well behaved inside an integral.
With this, we are now ready to calculate the potential at r due
to this charge located on the z-axis,
Z
ρ(r′ )
Φ(r) = d3 r ′
|r − r′ |
Z
1
=q r ′2 dr ′ sin θ ′ dθ ′ dφ′
2πr ′2 sin θ ′
δ(r ′ − R)δ(θ ′ )
× 1
(r 2 + r ′2 − 2rr ′ cos(θ − θ ′ )) 2
q
= 1 . (2.35)
(r 2 + R2 − 2Rr cos θ) 2
At this point, let us assume that we are interested in the poten-

tial very far away from the point charge. Namely, let us assume that
r >> R. Then, we can expand the denominator in (2.35) in a power
series as
1
1
(r 2 + R2 − 2Rr cos θ) 2
2 !− 21
1 2R R
= 1− cos θ +
r r r
" 2 !
1 1 2R R
= 1− − cos θ +
r 2 r r

2 !2
3 2R R
+ − cos θ + + ···
8 r r

1 R cos θ R2 (3 cos2 θ − 1)
= 1+ + + ···
r r 2r 2

1 R cos θ R2 (3 cos2 θ − 1)
= + + + ··· . (2.36)
r r2 2r 3
Substituting this back into the potential in (2.35), we obtain (for
r >> R),
q qR cos θ qR2 (3 cos2 θ − 1)
Φ(r) = + + + ··· . (2.37)
r r2 2r 3
As we will see later, the angular coefficients of the expansion
of the denominator in (2.36) can be identified with the Legendre
polynomials, namely,
P0 (cos θ) = 1,
P1 (cos θ) = cos θ,
(3 cos2 θ − 1)
P2 (cos θ) = , (2.38)
2
and so on. Thus, very far away from the point charge, we can write
the potential due to a charge displaced along the z-axis, as
X∞
qRn
Φ(r) = Pn (cos θ). (2.39)
n=0
r n+1
(Incidentally, by a similar expansion we can also show that when

R >> r the potential has the form
X qr n
Φ(r) = Pn (cos θ). (2.40)
n
Rn+1
2.4 Potential for a charge displaced along the z-axis 45
This also follows from the fact that the expression for the potential
in (2.35) is symmetric under r ↔ R.)
Such an expansion of the potential is known as the multipole
expansion. We note that, very far away from the charge, the domi-
nant term is the first term which we recognize to be the potential due
to a point charge at the origin (also called a monopole term). How-
ever, if for some reason, the first term is absent (namely, if we have
a charge neutral system), then the dominant term will be the second
term which is the potential due to a dipole. Furthermore, if we have
a system for which the first two terms vanish, then the leading term
would be the third term which is the potential due to a quadrupole
and so on.
As a parenthetical discussion, let us analyze the expansion of
the denominator in (2.35) a bit more in detail. Let us recall that a
d
translation by an amount a in one dimension is implemented by ea dx
so that we can write
d
f (x + a) = ea dx f (x) . (2.41)
The exponential operator simply generates the Taylor series for the
expansion of the function. In higher dimensions, this generalizes so
that we can write

f (r + a) = ea·∇ f (r) . (2.42)
If we apply this to the denominator in (2.35) (say, for r ≫ R), we

obtain (in spherical coordinates ẑ = r̂ cos θ − θ̂ sin θ)
1 1
1 =
(r 2 + R2 − 2rR cos θ) |r − R|
2

−R·∇ 1 −R·∇ 1
= e = e
|r| r

∂ sin θ ∂ 1
= e−R(cos θ ∂r − r ∂θ ) , (2.43)
r
which, in fact, generates the series in (2.36). The other thing to
observe is that, for r ≫ R, we can write the denominator in (2.35) as
1 1 1
1 = , (2.44)
(r 2 + R2 − 2Rr cos θ) 2 r (1 − 2z cos θ + z 2 ) 21
where we have defined

R
z= . (2.45)
r
Furthermore, it is well known that the second fraction on the right

hand side of (2.44) is the generator of Legendre polynomials
∞
X
1
1 = z n Pn (cos θ), (2.46)
(1 − 2z cos θ + z2) 2 n=0
which explains the structure of the series in (2.39).
2.5 Dipole
Let us consider next a charge configuration as shown in Fig. 2.7

where we are assuming that a charge q is located at z = R while a
second charge (−q) is located at z = −R.
q, R b
r
y
b
−q, −R
x
Figure 2.7: A dipole system with two point charges where charge q
is at z = R while charge (−q) is at z = −R.
The potential for this system can be simply obtained from what we
have already calculated (see (2.39) or (2.40)), namely, for r >> R,
we have (the potential due to the second charge, (−q) at z = −R, is
obtained by letting q → −q and θ → π − θ)

Φ(r) = + + + ···
r r2 2r 3

+ − + − + ···
r r2 2r 3
2qR cos θ
= + ··· . (2.47)
r2
Here, we note that the potential at very large distances does not
behave like that of a point charge. In fact, the total charge of the
system is zero as seen from large distances. Therefore, it is the second
2.5 Dipole 47
term in the expansion that gives the leading contribution. Such a

configuration of (equal and opposite) charges, when the separation
between them is small (or, when we are interested in the large distance
behavior), is called a dipole (centered at the origin) and the dipole
moment associated with the system is defined to be
p = qd = 2qR ẑ, (2.48)
where d represents the vector from the negative charge to the positive
charge. Incidentally, a more complete definition of the electric dipole
moment for a continuous distribution of charges is given by
Z
p = d3 r r ρ(r), (2.49)
which can be seen to reduce to the earlier definition in the case of

point charges. In terms of the dipole moment in (2.48), we can rewrite
the potential for the dipole (in (2.47)) to be
p · r̂ p·r
Φdipole (r) = 2
= 3
r r

1 p
= −p · ∇ = −∇ · , (2.50)
|r| |r|
which shows that the potential for the dipole can, in fact, be written
as a divergence (p is constant and hence can be taken inside the
gradient operation). Thus, comparing with the potential for a point
charge, we realize that the potential for the dipole behaves more
like the electric field of a charge (just the gradient nature or the
dependence on the distance and not the vector aspect). We can
also calculate the electric field associated with the dipole system by
recalling that in spherical coordinates,
∂ θ̂ ∂ φ̂ ∂
∇ = r̂ + + , (2.51)
∂r r ∂θ r sin θ ∂φ
so that the electric field for a dipole has the form
Edipole (r) = −∇Φdipole (r)

∂ 2qR cos θ θ̂ ∂ 2qR cos θ
= −r̂ −
∂r r2 r ∂θ r2
4qR cos θ 2qR sin θ
= 3
r̂ + θ̂
r r3
6qR cos θ 2qR

= 3
r̂ − 3 (cos θ r̂ − sin θ θ̂)
r r
3(p · r̂) r̂ − p
= , (2.52)
r3
where we have used the relations between the unit vectors in the
Cartesian and the spherical coordinates in the last step, namely, ẑ =
r̂ cos θ − θ̂ sin θ (see Fig. 2.8), to rewrite the expression in terms of
the dipole moment and the direction of observation.
r
θ θ̂
y
x
Figure 2.8: Unit vectors r̂, θ̂, ẑ.
Alternatively, the electric field in (2.52) can also be derived in a

simpler manner in the following way.
Edipole (r) = −∇Φdipole (r)

p·r
= −∇
|r|3

1 ∇(p · r)
= −(p · r)∇ −
|r|3 |r|3
3(p · r) r p
= 5
− 3
|r| |r|
3(p · r̂) r̂ − p
= , (2.53)
|r|3
where |r| = r. We see that both the potential and the electric field
for a dipole decrease faster than the corresponding quantities for a
point charge (monopole). Note also that the electric field for the
dipole continues to be curl free (∇ × p = 0 since p is constant and
∇ × r̂ = 0). The curl free nature also follows from the fact that
E = −∇Φdipole .
2.5 Dipole 49
◮ Example (Force on a dipole). Let us consider a dipole of length ℓ where the

vector from the origin of the coordinate system to the negative charge (−q) is r
and ℓ ≪ r (see Fig. 2.9). Let us further assume that the dipole is placed in an
electric field E(r) which is not necessarily uniform.
q
ℓ
(−q)
r
Figure 2.9: Dipole in an electric field.
In this case, the negative and the positive charges of the dipole will experi-
ence an electric force given respectively by
F(−q) = −qE(r), F(q) = qE(r + ℓ), (2.54)
where ℓ denotes the vector from the negative charge to the positive charge. As a
result, the total force acting on the dipole can be written as
F = F(−q) + F(q) = −qE(r) + qE(r + ℓ)

≈ −qE(r) + q (E(r) + (ℓ · ∇)E(r))
= q (ℓ · ∇) E(r) = (p · ∇) E(r). (2.55)
Here we have used the definition of the dipole moment in (2.48) to identify p =
qℓ. We note that the dipole experiences an electrostatic force in the presence of
a nonuniform electric field. On the other hand, if the electric field is uniform
(constant), then it is clear that the net force on the dipole vanishes, namely, the
positive and the negative charges experience equal and opposite forces.
We can also calculate the torque exerted on the dipole due to such a force
given in (2.55). The torque around the origin, for example, will have the form
τ = r × F(−q) + (r + ℓ) × F(q)
≈ −qr × E(r) + q(r + ℓ) × (E(r) + (ℓ · ∇)E(r))

≈ qℓ × E(r) + qr × (ℓ · ∇) E(r)
= p × E(r) + r × (p · ∇) E(r). (2.56)
We note that when the electric field is uniform, the second term on the right hand
side vanishes, but the first term is nonzero. Therefore, the dipole experiences a
torque even though the net force acting on the dipole is zero in this case. ◭
2.6 Continuous distribution of dipoles
It is worth pointing out here that there are many physical systems in
nature that behave like a dipole. In many molecules, for example, the
centers of positive and negative charges may not coincide giving rise
to an associated dipole moment even though the molecule as a whole
is charge neutral. A prime example of this is the water molecule
which behaves like a strong dipolar molecule and for these reasons,
the study of dipoles is quite significant. In fact, just as we can define
a continuous distribution of charges, for such dipolar material we can
also define a continuous distribution of dipole moments. Let P(r)
represent the dipole moment density centered at r so that we can
write the total dipole moment associated with such a physical system
as
Z
P = d3 r P(r), (2.57)
V
where V denotes the volume containing the distribution. From the

form of the potential for a single dipole in (2.50),
p · (r − r′ )
Φdipole (r) = , (2.58)
|r − r′ |3
where we are assuming that the dipole is centered at r′ , we see that
for a continuous distribution of dipole moments we can write the
potential as
Z
P(r′ ) · (r − r′ )
Φdipole (r) = d3 r ′
|r − r′ |3
V
Z
1
=− d3 r ′ P(r′ ) · ∇
|r − r′ |
V
Z
P(r′ )
= −∇ · d3 r ′ . (2.59)
|r − r′ |
V
Here, in the last step we have used the fact that P(r′ ) does not
depend on the coordinate r and hence the gradient operator does not
act on it and can be taken outside the integral. Thus, we see that
the potential for a continuous distribution of dipole moments can be
written as a divergence as well.
Given a potential we can always relate it to a charge distribu-
tion through the Poisson equation. Therefore, it is meaningful to
2.6 Continuous distribution of dipoles 51
ask what kind of a continuous charge distribution would give rise to

the potential (2.59) for a given continuous dipole distribution. The
answer is surprisingly not very difficult. Let us note from (2.59) that
we can write
Z
1
Φdipole (r) = − d3 r ′ P(r′ ) · ∇
|r − r′ |
V
Z
3 ′ ′ ′ 1
= d r P(r ) · ∇
|r − r′ |
V
Z Z
3 ′ ′ P(r′ ) ∇′ · P(r′ )
= d r ∇ · − d3 r ′
|r − r′ | |r − r′ |
V V
Z Z
ds′ · P(r′ ) ∇′ · P(r′ )
= − d3 r ′ . (2.60)
|r − r′ | |r − r′ |
S V
This shows that a dipole potential can equivalently be thought of as

being produced by a combination of a surface charge density as well
as a volume charge density given respectively by
σ(r) = n̂ · P(r),
ρ(r) = −∇ · P(r), (2.61)
where n̂ represents a unit vector normal to the given surface. (Here

the volume and the surface integrals refer respectively to the volume
where the dipole moments are localized and the surface bounding
such a volume.)
◮ Example. As an example of such a system, let us consider a sphere of radius

R with a uniform distribution of dipole moments given by the density P along
the z-axis as shown in Fig. 2.10. (A vector field is said to be uniform when its
magnitude as well as its direction are the same at every point.)
To calculate the potential due to such a distribution of dipoles we note that
the total charge of the system is zero. Therefore, there is no monopole contribution
and the leading term in the expansion of the potential is the potential due to the
dipoles. We choose the origin of the coordinate system to coincide with the center
of the sphere. Then, the calculation of the potential is straightforward,
Z
P θ(R − r ′ )
Φdipole (r) = −∇ · d3 r ′
|r − r′ |
Z
θ(R − r ′ )
= −P · ∇ d3 r ′ . (2.62)
|r − r′ |
Figure 2.10: A sphere of radius R with a uniform distribution of

dipole moments along the z-axis.
The integral can be evaluated in the following way. (We use the standard trick
to simplify the evaluation of this integral, namely, let us assume that r lies along
the z-axis, or alternatively that the angle θ′ is measured from r.)
Z
θ(R − r ′ )
d3 r ′
|r − r′ |
Z
θ(R − r ′ )
= r ′2 dr ′ sin θ′ dθ′ dφ′ 1
(r 2 + r ′2 − 2rr ′ cos θ′ ) 2
ZR
1 1 π
= 2π dr ′ r ′2 (r 2 + r ′2 − 2rr ′ cos θ′ ) 2
rr ′ 0
0
ZR
2π
= dr ′ r ′ (r + r ′ ) − |r − r ′ | . (2.63)
r
0
The value of this integral clearly depends on whether r > R or r < R

because of the second term. For r > R, we have
Z ZR
θ(R − r ′ ) 2π
d3 r ′ = dr ′ r ′ (r + r ′ − r + r ′ )
|r − r′ | r
0
ZR
4π 4πR3
= dr ′ r ′2 = , (2.64)
r 3r
0
2.7 Quadrupole 53
while, for r < R, we have

Z
θ(R − r ′ )
d3 r ′
|r − r′ |
 r 
Z ZR
2π  ′ ′ ′ ′ ′ ′ ′ ′
= dr r (r + r − r + r ) + dr r (r + r − r + r)
r
0 r
 r 
Z ZR
4π  4π r 3 r(R2 − r 2 )
= dr ′ r ′2 + dr ′ rr ′  = +
r r 3 2
0 r
r 2
4π(3R2 − r 2 ) 4πR3
= = 3− . (2.65)
6 6R R
Therefore, we can write the potential (2.62) for the dipole distribution to
be
r 2
1 1
Φdipole (r) = −P · ∇ θ(r − R) + θ(R − r) 3− , (2.66)
r 2R R
where we have defined the total dipole moment of the sphere (V is the volume of
the sphere) as
4π R3 P
P = PV = . (2.67)
3
Furthermore, since the polarization is along the z-axis, we obtain
Φdipole (r)
r 2
∂ 1 1
= −|P| θ(r − R) + θ(R − r) 3−
∂z r 2R R
hz z z z i
= |P| 3
θ(r − R) − 2 δ(r − R) + 2 δ(r − R) + 3 θ(R − r)
r R R R
P ·r P ·r
= θ(r − R) + θ(R − r). (2.68)
r3 R3
It is clear from the above calculation that the potential is continuous. Outside the
sphere the potential behaves as if we have a single dipole with moment P centered
at the origin while the potential inside the sphere is that of a uniform electric field
(− RP ) (the negative sign is from the definition of the potential (−E · r) for a
3
uniform field, or alternatively, from the definition E = −∇Φ). ◭
2.7 Quadrupole
Let us consider a configuration of four charges as shown in Fig. 2.11.

Once again, we can obtain the potential for this system from what we
have already calculated in (2.39). The potential for the charge at the
origin is obtained by setting R = 0 and q → −2q, while the potential
q, R b
r
−2q b
y
b
q, −R
x
Figure 2.11: A quadrupole configuration with two point charges q at

z = ±R and a point charge (−2q) at the origin.
for the charge on the lower z-axis is obtained by letting θ → π − θ.

The complete potential is given by

q qR cos θ qR2 (3 cos2 θ − 1) 2q
Φ(r) = + + + ··· −
r r2 2r 3 r

+ − + + ···
r r2 2r 3
qR2 (3 cos2 θ − 1)
= + ··· . (2.69)
r3
This shows that for this system of charges both the monopole and
the dipole terms in the potential vanish. The total charge of the
system is zero which is why the monopole term in (2.69) vanishes,
but we can also think of the system of four charges in Fig. 2.11 as
two dipoles with opposite dipole moments which makes the dipole
term vanish as well (total dipole moment is zero). As a result it is
the third term in the multipole expansion which gives the leading
behavior of the potential for large distances. Such a configuration
of charges is known as a quadrupole. We note that unlike the case
of dipoles, other configurations of quadrupoles are possible and this
is not the unique quadrupole configuration. Furthermore, while the
dipole moment is a vector, the quadrupole moment, in general, is a
second rank symmetric traceless tensor. In fact, the n-th term in the
multipole expansion (see, for example, (2.39)) leads to the 2n-th pole
moment which isR an n-th rank tensor. The monopole gives a 0-th
rank tensor q = d3 r ρ(r) which is a scalar, the dipole gives a rank
one tensor which is a vector and so on. In Cartesian coordinates the
2.8 Potential due to a double layer of charges 55
quadrupole moment tensor is defined to be

Z

Qij = d3 x 3xi xj − δij x2 ρ(x), (2.70)
which is manifestly symmetric and traceless. For the charge con-

figuration in Fig. 2.11, which can be thought of as two dipoles
aligned back to back along the z-axis, we have (note that ρ(x) =
q ((δ(x3 − R) + δ(x3 + R)) − 2δ(x3 )) δ(x1 )δ(x2 ))
Q11 = −2qR2 = Q22 ,

Q33 = 4qR2 = −(Q11 + Q22 ) = −2Q11 , (2.71)
with all other components vanishing. (Note that knowing Q11 and
Q22 , we could have predicted the value for Q33 from the tracelessness
condition of the quadrupole moment tensor.) In general, the potential
for the quadrupole can be written in terms of the quadrupole moment
as (for large values of r = |x|)

1X 3xi xj − δij x2
Φquadrupole (r) = Qij , (2.72)
6 r5
i,j
which can be readily checked to give the leading term (the quadrupole
potential) in (2.69) for the system under consideration. We can again
obtain the electric fields associated with the quadrupole system by
taking the gradient. But, we will not get into the details of this
except for noting that the potential as well as the electric field for
the quadrupole decrease even faster than those for the dipole.
2.8 Potential due to a double layer of charges
Let us consider a single plane of infinite extent carrying a constant

surface charge density σ. We have already obtained the electric field
for such a configuration in the last chapter using Gauss’ law (see
(1.42)). Let us now calculate the potential associated with such a
configuration as shown in Fig. 2.12. Let us assume that the plane is
at z = 0 and because of the symmetry in the problem, we can always
choose the z-axis to lie along the perpendicular to the plane from the
coordinate where we are interested in evaluating the potential.
The calculation of the potential is straightforward. We can write
the volume charge density for the system to be
ρ(r) = σδ(z), (2.73)

z=0
Figure 2.12: An infinite plane located at z = 0 and carrying a con-

stant surface charge density σ.
where σ denotes the constant surface charge density. Consequently,

using cylindrical coordinates, we obtain,
Z
ρ(r′ )
Φ(r) = d3 r ′
|r − r′ |
Z Z
σδ(z ′ ) r ′ dr ′ dφ′
= r ′ dr ′ dφ′ dz ′ 1 = σ 1
(r ′2 + (z − z ′ )2 ) 2 (r ′2 + z 2 ) 2
ZΛ
r ′ dr ′ 1 Λ
= lim 2πσ 1 = lim 2πσ (r ′2 + z 2 ) 2
Λ→∞ (r ′2 + z 2 ) 2 Λ→∞ 0
0
h 1
i
= lim 2πσ (Λ2 + z 2 ) 2 − |z|
Λ→∞
= −2πσ |z| + constant, (2.74)
where we recognize that the constant on the right hand side of (2.74)
is a divergent constant. We also understand the origin of this diver-
gent constant, namely, we have a charge distribution which extends
to infinity. Consequently, the reference point for the potential has to
be chosen differently or one has to allow for a (possibly) divergent
constant in the potential. However, an additive constant is not mean-
ingful if we are interested in physical quantities such as the electric
field and, therefore, ignoring the constant we can write the potential
for such a system of charge distribution to be
Φ(r) = −2πσ|z|. (2.75)

2.8 Potential due to a double layer of charges 57
Taking the gradient of (2.75) we obtain the electric field to be
∂
E(r) = −∇Φ(r) = −ẑ (−2πσ|z|) = 2πσ sgn(z) ẑ, (2.76)
∂z
which is the result we had obtained earlier in (1.42).

Let us next consider two thin planes of infinite extent separated
by a small distance d. Namely, let us assume that the two planes
are located at z = d2 and z = − d2 and carry a uniform surface charge
density of σ and −σ respectively as shown in Fig. 2.13. Thus, in some
sense we have an infinite distribution of dipoles. Since the electric
potential is additive, the potential for this distribution of charges can
be easily obtained from the calculation for a single layer of infinite
extent in (2.75) which leads to (remember that the locations of the
planes are displaced from z = 0)
σ
−σ
z = − d2 z= d
2
Figure 2.13: Two infinite planes carrying constant surface charge

densities σ and (−σ).

d d
Φ(r) = Φ1 (r) + Φ2 (r) = −2πσ z − − z+ . (2.77)
2 2
This can be simplified and written as



 2πσd for z > d2 ,

d
Φ(r) = 4πσz for − 2 < z < d2 , (2.78)



−2πσd for z < − d2 .
Taking the gradient of the potential it is now easy to obtain that



 0 for z > d2 ,

E(r) = −∇Φ(r) = −4πσ ẑ for − d2 < z < d2 , (2.79)



0 for z < − d2 .
Let us note that for this system the electric field is nontrivial
only between the two planes. Outside the two planes the electric
field vanishes because the contributions from the two planes (which
are oppositely charged) cancel exactly. Furthermore, there is one as-
pect of the results in (2.78) and (2.79) which is worth emphasizing.
Namely, when d is very small, the potential is discontinuous across
the double layer, while the normal component of the electric field is
continuous. The continuity of the normal component of the electric
can be easily understood from the fact that across the double layer,
the net surface charge density is zero. On the other hand, the discon-
tinuity in the dipole potential is understood from the fact that it has
the behavior similar to that of an electric field and, consequently, is
discontinuous across the double layer simply because there is surface
density of dipole moments in such a case.
2.9 Conductors and insulators
In studying problems of electromagnetism we are quite often in-

terested in the behavior of various materials (not simply of point
charges) in the presence of electromagnetic fields. All materials found
in nature can be broadly classified into two groups depending on their
response to an external electric field. One class of materials is known
as conductors and metals are prime examples of such systems which
contain a large number of free electrons. These electrons are free
to move around inside the material (they cannot, of course, leave
the material under normal circumstances) and they respond almost
instantaneously (in an idealized situation) to any external applied
field. When an electric field is set up within the material due to some
external source the free electrons conduct electricity.
In contrast, the second class of materials, known as insulators,
are non-conductors. In such materials, the electrons are tightly bound
to the atomic nuclei and are not free to move. When an external
electric field is applied, the electrons may be displaced slightly from
their normal position (the atomic nuclei are heavier, so they do not
move appreciably), but they are still bound to the nuclei. As a re-
sult, insulators do not conduct electricity. Dielectrics are insulators
2.9 Conductors and insulators 59
which can be polarized in the presence of an external electric field

(namely, the negative and the positive charge centers can be sep-
arated). In some dielectrics permanent atomic (molecular) dipole
moments may be present independent of any applied external field.
The water molecule that we discussed earlier is an example of such
a substance. In such materials, even though polarization (dipole mo-
ment) is non-vanishing at smaller scales (for individual molecules),
macroscopically the dielectric may be unpolarized. However, a small
external electric field would be able to orient the dipoles giving rise
to a net polarization for the dielectric. (Let us note here that, for all
practical purposes, “dielectric” and “insulator” mean the same thing.
Vacuum is the only insulator which cannot be polarized and hence is
not a dielectric.)
Although an ideal dielectric has zero conductivity (it does not
conduct electricity at all), in reality, dielectrics may have a small
conductivity. However, a typical conductor has a conductivity which
is about 1020 times larger than that of a typical dielectric and so, in
our discussions, we can safely assume the conductivity of dielectrics to
be zero. Similarly, even though conductors have a finite conductivity
(collision of the electrons during motion gives rise to a resistivity
to the flow of currents), for the purposes of our discussions we will
assume that the conductors have an infinite conductivity.
Besides conductors and insulators (dielectrics), there are also
semi-conductors and electrolytes with intermediate properties as far
as conductivity is concerned, but they have many other properties
which make them interesting independently. However, we would limit
ourselves only to conductors and dielectrics for the purposes of our
discussions.
Let us summarize here the properties of conductors. First, it is
clear that if a conductor is in static equilibrium in the presence of
an external electric field, then, the electric field inside the conductor
must be zero. This must be so, because if the electric field is nonzero
inside the conductor, the free electrons will experience an electric
force and would move, violating the assumption of static equilibrium.
The way conductors achieve static equilibrium is really quite simple.
In response to an external electric field, the electrons move to one edge
of the surface of the conductor leaving the opposite edge positively
charged (see Fig. 2.14) and set up an internal electric field which
exactly cancels the external field within the conductor.
Since E = 0 inside a conductor, it follows from Gauss’ law that
the net charge density must vanish inside a conductor. This simply
means that there is an equal number of positive and negative charges
−− +
+
−
− ++
− +
Figure 2.14: A conductor in an external electric field.
in any small volume inside a conductor. The external electric field

simply leads to a redistribution of the free electrons on the surface
of the conductor giving rise to a nonzero surface density of charges.
(It also follows from this that any charge that one puts inside a con-
ductor must necessarily move to the surface for static equilibrium.)
Furthermore, the potential must be a constant inside the conductor
(E = 0) all the way up to the surface of the conductor. Thus, the sur-
face of a conductor (in fact, any surface inside the conductor) defines
an equipotential surface. There cannot be any tangential component
of the electric field on the surface of the conductor (otherwise, static
equilibrium will not hold). Outside the conductor, of course, the elec-
tric field will not be zero and, in fact, from our earlier discussions,
we can conclude that immediately outside the conductor the electric
field would be normal to the surface with the value
En,R = 4π σ, (2.80)
where σ denotes the surface density of the redistributed charges.

The conductors need not always have a simple configuration.
Sometimes, a conductor may contain a cavity inside. If such a con-
ductor is placed in an external electric field with no charge in the
cavity (see Fig. 2.15), once again there will be a surface distribution
of charges (now there are two surfaces, one exterior and one interior)
such that there would be no electric field inside the conductor as well
as inside the cavity in the static equilibrium. The vanishing of the
electric field inside the conductor can, of course, be understood along
the lines of reasoning given above. The vanishing of the electric field
inside the cavity follows simply from the fact that along any closed
2.9 Conductors and insulators 61
E
B
Figure 2.15: A conductor with an internal cavity in an external elec-

tric field.
curve cutting the inner surface (see Fig. 2.15), we must have
I ZB ZB
dℓ · E = dℓ · E = dℓ |E| = 0, (2.81)
C
A A
implying that E = 0 inside the cavity. In deriving this result, we

have chosen the path of integration inside the cavity to lie along the
electric field (recall that E = 0 inside the conductor and because
of the conservative nature of the electric field, the choice of path is
irrelevant). This result is quite interesting because this shows that
the electric field within the conductor as well as inside the cavity
vanishes. Consequently, there is no discontinuity of the electric field
across the inner surface. This is possible only if the inner surface does
not carry any surface charge density. Therefore, we conclude that in
such a case, the free electrons redistribute themselves in a way so
that only the outer surface carries a surface charge density. This also
shows that any external field cannot penetrate inside a cavity within
a conductor. This is the principle behind electrical shielding which
is used to shield electrical equipment by putting them inside a metal
container (commonly known as the Faraday’s cage) and this is also
the reason we do not get electrocuted inside a car during lightning.
(One may think that this result can also be derived using a Gaussian
surface inside the cavity. It is true that there is no charge inside and
hence there will be no net flux. However, that does not mean that
there is no electric field in the cavity.)
Thus, the mechanism by which a conductor maintains a vanish-
−
++ −
−
−
− q
b
−
+ −
++
Figure 2.16: A cavity carrying a charge q inside a conductor.
ing E field within itself is that in the presence of external fields or

charges, induced charges appear on the surface of the conductor to
precisely cancel the external field. The actual distribution of the sur-
face charges is, of course, a hard problem to calculate, but the charges
rearrange themselves precisely in a way so as to give E = 0 within
the conductor. If there is a cavity inside a conductor and the cavity
carries a charge q (see Fig. 2.16), it would induce a charge density on
the inner as well as the outer surfaces of the conductor so as to have
a vanishing electric field within the conductor. On the other hand,
the electric field would be non-vanishing within the cavity as well as
outside the conductor.
r
b
q
Figure 2.17: A spherical Gaussian surface of radius r surrounding the

conductor.
2.10 Capacitor 63
The electric field outside the conductor can be calculated simply

by using Gauss’ law. Drawing a spherical Gaussian surface (centered
at the charge q as in Fig. 2.17) of radius r which is much bigger than
any dimension of the conductor, we see that the field at such points
must be radial with the value
q
E(r) = r̂, (2.82)
|r|2
independent of the shape of the cavity or the conductor.

This discussion shows that in the presence of external electric
fields the behavior of conductors is quite complex and interesting and
needs to be analyzed carefully which we will do.
2.10 Capacitor
When we have two conductors carrying equal and opposite charge,

they are said to form a capacitor system. The surfaces of the con-
ductors are equipotential surfaces and let us denote the potentials on
the surfaces of the two conductors as Φ1 and Φ2 respectively with
Φ1 > Φ2 . Consequently, the potential difference between the two
surfaces can be written as
V = Φ1 − Φ2 , (2.83)
and is conventionally known as the voltage between the two surfaces.

From our discussion so far we know that the potential and, there-
fore, the voltage depend linearly on the charge Q of the conductors.
Consequently, we can write
Q = C V, (2.84)
where the constant of proportionality C is known as the capacitance

of the system which can be thought of as the charge necessary to
maintain a unit voltage across the two surfaces. In the CGS system,
it is clear that the unit of capacitance is given by
esu (esu)2 (esu)2 /cm

= = cm = cm, (2.85)
statVolt esu−statVolt esu−statVolt
2
where we have used the fact that both “ (esu)cm ” and “esu−statVolt”
correspond to units of work, namely, “erg”. Since the capacitance
has the dimension of a length, it is intuitively clear that it must be a
geometrical property of the system. We also note here that the MKS
unit of capacitance is a farad (F) which is defined as
Coulomb 3 × 109 esu

1 farad = 1 = 1 = 9 × 1011 cm. (2.86)
Volt 3 × 10−2 statVolt
We see that a farad is quite large and, in fact, the capacitance of a

typical capacitor is of the order of a microfarad (µF) or even smaller,
a picofarad (pF or sometimes also written as µµF). Let us also note
here that sometimes one even talks about the capacitance of a single
conductor. It is understood, in such a case, that the second conduct-
ing surface lies at infinity.
◮ Example (Parallel plate capacitor). Let us calculate the capacitance of some
typical capacitors. The simplest is to consider a pair of parallel, rectangular
conducting plates with area A separated by a distance d and carrying charges Q
and (−Q) respectively, distributed uniformly over the two surfaces as in Fig. 2.18.
If the separation d is small compared to the area A of the plates, we can think of
them as two infinite plates carrying an equal and opposite surface charge density.
This is a problem we have worked out earlier in detail and we conclude that the
electric field will be nonzero only between the two plates and would be uniform
given by (see (2.79))
|E| = 4πσ, (2.87)

Q
pointing from the upper to the lower plate. Here σ = A
represents the magnitude
of the surface charge density.
(−Q)
Figure 2.18: Two parallel plates separated by a distance d and car-

rying charges Q and (−Q).
It follows, therefore, that the voltage across the plates is given by

4πd
V = |E|d = 4πσd = Q. (2.88)
A
As a result, we immediately identify the capacitance for the parallel plate system
to be
A
C= , (2.89)
4πd
2.10 Capacitor 65
and we see explicitly that it is determined by the geometry of the system.

We note here that in deriving this result, we have pretended as if the two
plates are of infinite extension. This is, of course, not true and consequently the
electric fields are not uniform all over. In particular, at the edges of the plates,
the electric fields are not uniform. Thus, our determination of the capacitance
is not quite correct. If one determines the correction due to the edge effect, it
turns out that the capacitance increases slightly, but to a first approximation, the
idealized answer obtained above is quite good when d is very small.
Let us now calculate the energy stored in this capacitor system. First, let
us note that if the capacitance is C and the magnitude of the charges on the two
plates is Q e with a voltage Ve , then, to increase the charge on the upper plate
infinitesimally by an amount dQ e (and, therefore, to decrease the charge on the
lower plate by dQ),e that is to take an amount of charge dQ e from the lower plate
and move it to the upper plate, we must do work against the electric force and
the amount of work is given by
e dQ
Q e
e (Φ
dW = dQ e1 − Φ
e 2 ) = Ve dQ
e= . (2.90)
C
Integrating this we can obtain the total work necessary to charge the capacitor
plates starting from the uncharged state and the result is
ZQ e e
Q dQ Q2 (CV )2 1
W = = = = CV 2 . (2.91)
C 2C 2C 2
0
This work must, of course, be stored in the capacitor as electrostatic energy. Now,
let us recall that for the parallel plate system
V = |E|d,
A
C= . (2.92)
4πd
Using these we can also write
1 A 1 2 1 2
W = × (|E|d)2 = E × Ad = E × volume, (2.93)
2 4πd 8π 8π
which is exactly the result we had obtained earlier for the electrostatic energy for
a continuous distribution of charges in the last chapter (see (1.85)). ◭
◮ Example (Spherical capacitor). As a second example, let us consider two spher-

ical conducting shells of radii R1 and R2 carrying charges Q1 and Q2 respectively.
We assume that R1 > R2 (see Fig. 2.19).
The spherical symmetry of the problem determines that the charges would
distribute uniformly over the two surfaces giving rise to the surface densities σ1
and σ2 respectively. Thus, we can write
ρ(r) = σ1 δ(r − R1 ) + σ2 δ(r − R2 ), (2.94)
where
Q1 Q2
σ1 = , σ2 = . (2.95)
4πR12 4πR22
R2
Q2
R1 Q1
Figure 2.19: Two spherical shells of radius R1 and R2 carrying charges

Q1 and Q2 respectively.
The potential for this system can now be easily calculated.

Z
ρ(r′ )
Φ(r) = d3 r ′
|r − r′ |
Z
(σ1 δ(r ′ − R1 ) + σ2 δ(r ′ − R2 ))
= r ′2 dr ′ sin θ′ dθ′ dφ′ 1
(r 2 + r ′2 − 2rr ′ cos θ′ ) 2

1
= 2π R12 σ1 − (|r − R1 | − (r + R1 ))
rR1

2 1
+R2 σ2 − (|r − R2 | − (r + R2 ))
rR2
Q1 Q1
= θ(r − R1 ) + θ(R1 − r)
r R1
Q2 Q2
+ θ(r − R2 ) + θ(R2 − r) . (2.96)
r R2
This shows that the surface at r = R1 is an equipotential surface with the
potential given by
(Q1 + Q2 )
Φ(R1 ) = , (2.97)
R1
while the surface r = R2 has a constant potential given by
Q1 Q2
Φ(R2 ) = + . (2.98)
R1 R2
This can, of course, be qualitatively understood in terms of the properties of
electric systems as well as the calculations we have already done. Namely, at
r = R1 , the system behaves as if the entire charge (Q1 + Q2 ) were concentrated
at the origin. On the other hand, at r = R2 , the charge Q2 behaves like it is
at the origin while the charge Q1 gives a constant potential characteristic of the
potential inside a shell of radius R1 .
This is quite general so far. Let us next assume that we have a spherical
capacitor system in which case
Q2 = −Q1 = Q, Q > 0. (2.99)
(The other possibility is Q2 = −Q1 = −Q which can also be worked out in a

parallel manner and leads to the same result for the capacitance.) In this case,
the voltage across the two surfaces of the spherical capacitor system is given by

1 1
V = Φ(R2 ) − Φ(R1 ) = Q −
R2 R1
(R1 − R2 )
= Q. (2.100)
R1 R2
Therefore, it follows that the capacitance of this capacitor is again a geometrical
quantity given by
R1 R2
C= . (2.101)
(R1 − R2 )
Furthermore, if we let R1 → ∞, then, we have a single spherical shell of radius
R2 with a capacitance given by
C = R2 . (2.102)
Let us note here that capacitors are widely used to store charge. ◭
1. Consider a particle as a sphere of radius R carrying a volume

charge density
3Q
ρ(r) = (R − r), r ≤ R,
πR4
where the coordinate origin is assumed to coincide with the
center of the sphere. What is the total charge carried by the
sphere? Calculate the electric fields both inside and outside the
sphere.
2. Consider a dipole centered at the origin in a uniform electric

field. Calculate the torque on the dipole due to the electric
field. If the dipole is initially parallel to the electric field, how
much work would be needed to rotate it by an angle θ?
3. a) Calculate the dipole moment for the following configuration

of charges: three, each of value q, located at (x = 0, y = 0, z =
d), (x = R, y = 0, z = 0), (x = −R, y = 0, z = 0), as well as a
charge of −3q located at (x = 0, y = 0, z = −d).
b) Calculate the quadrupole moment for the charge configura-

tion where two charges of value q are at (x = 0, y = 0, z = R)
and (x = 0, y = 0, z = −R), while two other charges of value
−q are at (x = R, y = 0, z = 0) and (x = −R, y = 0, z = 0).
4. Consider a localized distribution of charge given by a density
ρ(r) = r 2 e−r sin2 θ.
a) What is the potential due to this charge distribution at very

far away distances?
b) What are the nontrivial multipole moments present in this
potential?
5. Two long cylindrical conductors (wires) of radii a1 and a2 re-

spectively are separated by a distance d >> a1 , a2 . Find the
capacitance per unit length for the system. If d = 0.5cm and
a1 = a2 = a, what must be the diameter of the wires to give a
capacitance per unit length of 9 × 10−3 ?
6. Consider a pair of coaxial, conducting, hollow cylinders of in-

finite length and with radii R1 and R2 , where R2 > R1 . The
outer and the inner cylinders carry charges Q and −Q respec-
tively. What is the capacitance per unit length for such a sys-
tem?
Chapter 3
Boundary value problems
3.1 Method of images
When we have a system of conductors, the physical problem of in-

terest may be, in general, much more complex than calculating the
potential for a given distribution of point charges. For example, we
may have a system of conductors in an electric field with the sur-
faces of the conductors held fixed at some given value of the poten-
tial. (A conductor may be grounded meaning that the surface is
maintained at zero potential.) Such problems where boundary values
of the potential may be specified are commonly known as boundary
value problems. The proper way to solve such a problem is, of course,
by determining the solution of the Laplace or the Poisson equation
subject to the appropriate boundary conditions, which we will discuss
later in this chapter. However, sometimes the physical system may
be simple enough that one can determine the potential in a simple
manner without having to derive the explicit solution of the Laplace
or the Poisson equation. One such method is known as the method
of images. Here, the idea is very simple. If we are given a physical
system with a set of boundary conditions, we try to reproduce these
conditions by a simple choice of fictitious “image” charge distribu-
tions. If we can do this, then, solving the original problem subject
to the boundary conditions is equivalent to solving the problem with
these additional fictitious charges without any boundary condition.
It is clear that if the boundary conditions are extremely complicated,
finding the “image” charge distribution may prove difficult and, con-
sequently, the method will not be very useful. However, for relatively
simple boundary conditions this method works quite well as we will
see from the following examples.
◮ Example (Point charge above a conducting plane). Let us consider a system
consisting of a conducting plane of infinite extent which is maintained at zero
potential (grounded), and a point charge q which is above the plane at a height
69
70 3 Boundary value problems
d as shown in Fig. 3.1. Let us assume for concreteness that the charge is positive
although this is not essential for our discussion.
b q
Figure 3.1: A point charge q above an infinite conducting plane

(grounded) at a height d.
It is clear that the electric field is meaningful and nonzero only above the
plane. This is because, for an insulated conductor, the point charge would induce
charges of opposite sign on the two sides of the conductor in such a way as to
cancel the field within the conductor. The charges on the lower surface of the
conductor, positive charges in this case, would then give rise to an electric field on
the lower half of the plane. Here, however, we have a conductor that is grounded
and the ground has an infinite supply of negative charges which would move on to
the conductor to annihilate all the positive charges. As a result of this, the surface
would have a net negative charge and all the field lines originating from the point
charge q would terminate on the plane and there will be no field lines below the
plane. Namely, the electric field cannot penetrate a grounded conducting plane
of infinite extent. Another way to see this is to note that, by grounding, the
surface of the conductor is maintained at zero potential as is the surface of the
plane infinitely below the conducting plane. Consequently, the potential difference
(voltage) across these two planes is zero and there cannot be an electric field in
this region. It is only at points above the plane that the electric field will be
nonzero.
Therefore, to determine the potential and the electric field above the plane
for this physical system, we need to find a fictitious charge distribution which can
reproduce the boundary condition. Without loss of generality, we can assume that
the conducting plane lies in the x − y plane and that the charge q is located at
a height d on the positive z-axis. Clearly the problem has cylindrical symmetry
and, consequently, it is meaningful to use cylindrical coordinates to analyze this
problem. Let us next assume that a second point charge q ′ is located at a height
d′ below the plane on the negative z-axis (see Fig. 3.2). In the presence of these
two charges, the potential at any point can be easily calculated and in particular
on the plane at z = 0 it is given by
q q′
Φ(r, φ, z = 0) = √ +√ , (3.1)
r2 +d 2 r + d′2
2
3.1 Method of images 71
b q
d′
q′ b
Figure 3.2: The system in Fig. 3.1 with an image charge q ′ at a

distance d′ on the opposite side of the plane directly below the charge
q.
where r represents the radial coordinate on the plane (in cylindrical coordinates).
Requiring the potential to vanish on the plane (which is our boundary condition),
we obtain

q 2 r 2 + d′2 − q ′2 r 2 + d2 = 0,

or, q 2 − q ′2 r 2 + q 2 d′2 − q ′2 d2 = 0. (3.2)
Requiring this to be true for any value of r (namely, at any point on the plane),
we determine
q ′ = ±q, d′ = ±d. (3.3)
Putting this back into the expression for Φ in (3.1) and noting that both d and d′
are positive, we determine that the potential vanishes on the plane only for
q ′ = −q, d′ = d. (3.4)
We see that if we were to introduce a fictitious charge (−q) located at z = −d into

the problem, then we can reproduce in a natural manner the boundary condition
of the problem (namely, the plane is grounded).
Thus, as far as the calculations of the potential and the electric field for pos-
itive z are concerned, we can forget about the plane and the boundary condition
and work with only these two charges. The charge above the plane is, of course,
the physical charge. The second charge which is not real but which we can use
in lieu of the boundary condition is known as the “image” charge borrowing the
terminology from optics. At any point r above the plane (z > 0), the potential is
given by (see Fig. 3.3)
q q
Φ(r) = p −p . (3.5)
r 2 + (z − d)2 r 2 + (z + d)2
Consequently, the electric field at any point above the plane (z > 0) is obtained
to be
E(r) = −∇Φ(r)
!
∂ ∂ q q
= − r̂ + ẑ p −p
∂r ∂z r 2 + (z − d)2 r 2 + (z + d)2
!
r r
= qr̂ 3 − 3
(r 2 + (z − d)2 ) 2 (r 2 + (z + d)2 ) 2
!
(z − d) (z + d)
+qẑ 3 − 3 , (3.6)
(r 2 + (z − d)2 ) 2 (r 2 + (z + d)2 ) 2
where we have used the fact that the potential is independent of the azimuthal
angle φ and correspondingly have dropped the angular derivative in the gradient.
r
q b
−q b
Figure 3.3: The potential at a point above the plane due to the charge
q and the “image” charge (−q) at a distance d above and below the
plane respectively.
Let us emphasize that the expressions for both the potential and the electric
field are meaningful only for z > 0. As is obvious, they give the wrong result for
z < 0 as they should because we do not have a real charge for z < 0. It is worth
noting here that for z → 0+ , the electric field (3.6) takes the form
2qd
E(r)|z→0+ = − 3 ẑ. (3.7)
(r 2 + d2 ) 2
That is, there is only a normal component of the electric field on the surface along
the z-axis as we would expect. It is not a uniform electric field on the plane, but
has cylindrical symmetry and, in fact, the field becomes weaker as we move away
from the origin. (By the way, remember that r is the radial coordinate on the
plane.) On the lower surface (z < 0), of course, there is no electric field and,
consequently, the normal component of the electric field is discontinuous across
the surface. And as we have seen (see, for example, (2.10)), the discontinuity is
proportional to the surface charge density

1 qd

σ(r) = ẑ · (ER − EL ) =− 3 . (3.8)
4π z=0 2π (r + d2 ) 2
2
Namely, the presence of the point charge q induces a charge density of opposite
sign on the surface of the grounded conductor. (For our case, with the choice
q > 0, there would be a negative induced surface charge.) The surface charge
density is invariant under rotations on the plane, but is not uniform. In fact, like
the electric field, it falls off rapidly as we move away from the center. We can, of
course, obtain the total induced surface charge by integrating over the entire area
and we have
Z
Qinduced = ds σ(r)
Z
qd r dr dφ
=− 3
2π (r 2 + d2 ) 2
Z∞
r dr
= −qd 3
(r 2 + d2 ) 2
0
∞
1
= qd √ = −q. (3.9)
r 2 + d2 0
Namely, the total induced charge on the plane is equal to the “image” charge
(which, in this case, is equal in magnitude to the physical charge, but opposite in
sign).
The induced charge, of course, would lead to an attractive force between the
point charge and the plane and this can be calculated in the following way. Let
us note that at any point on the z-axis, the electric field produced by the induced
surface charge density on the plane would be along the z-axis by symmetry. We
can calculate it in a simple manner. The potential at any point along the z-axis
due to the surface charges is given by (z > 0)
Z
σ(r ′ )
Φ(x = 0, y = 0, z) = r ′ dr ′ dφ′ 1
(r ′2 + z 2 ) 2
Z
qd r ′ dr ′
=− (2π) 3 1
2π (r + d ) 2 (r ′2 + z 2 ) 2
′2 2
Z∞
q ∂ dx
= 1 1
2 ∂d (x + d2 ) 2 (x + z 2 ) 2
0
Z∞
q ∂ dx
= p
2 ∂d x2 + (d2 + z 2 )x + d2 z 2
0
q ∂ h p i∞
= ln 2 x2 + (d2 + z 2 ) x + d2 z 2 + 2x + d2 + z 2
2 ∂d 0
q ∂ q
= (−) ln(z + d)2 = − . (3.10)
2 ∂d z+d
(It can be seen in two different, but equivalent, ways that the upper limit does
∂
not contribute. If we take the derivative ∂d first and then the limit, it is obvious.
Alternatively, we note that the upper limit gives rise to an infinite constant which
∂
vanishes upon taking the derivative ∂d .) We recognize this to be the potential
at a point along the z-axis in the upper half plane due to the fictitious “image”
charge (see (3.5)). The electric field due to the induced surface charge can now
be calculated and gives the value for z = d to be (actually, this can be obtained
from the general expression for the E field derived in (3.6) as well)

∂ q q
E(x = 0, y = 0, z = d) = −ẑ − = − 2 ẑ. (3.11)
∂z z + d z=d 4d
Therefore, the force of attraction experienced by the point charge has the value
(this is the force of attraction between the conducting plane and the point charge)
q2
F = qE(x = 0, y = 0, z = d) = − ẑ. (3.12)
4d2
Once again, we see that this is exactly the attractive force between the point charge
and the “image” charge and, consequently, on the positive z-axis the “image”
charge truly reproduces the effect of the induced surface charge on the plane.
Finally, let us note that we can calculate the electrostatic energy of the
system in a simple way. Let us recall that if the point charge is at a distance
z from the plane on the positive z-axis, then, the force between the conducting
plate and the charge is given by (see (3.12))
q2
F(z) = − ẑ. (3.13)
4z 2
Using this, we can calculate the work that must be done to bring the point charge
from infinity to a distance d above the plane on the z-axis. From the defining
relation, we have
Zdẑ Zd d
q2 dz q 2 q2
W =− dℓ · F = 2
=− =− . (3.14)
4 z 4z ∞ 4d
∞ ∞
In deriving this result, we have used the fact that the work is independent of the
choice of the path and, consequently, we have chosen a path along the z-axis for
simplicity. It is worth noting here that this energy is half the energy between the
point charge and its “image”. This can be understood in the following way. The
“image” charge is not real. But, if we had calculated the work done to bring the
point charge and its “image” from infinity to the final position, we would have
done twice the work because we have to move the “image” charge as well. This
is, of course, wrong because the “image” charge is not real. So, had we calculated
the electrostatic energy using the “image” charge, we would have obtained an
erroneous result. (An alternative way to see why one would get twice the result
using the “image” charge is to note that the electrostatic energy is related to
the square of the electric field integrated over the entire volume, as discussed in
chapter 1. With the “image” charge, of course, there is an associated electric
field even in the lower half of the plane and by symmetry, it contributes an exact
amount as the physical electric field above the plane. That is why the electrostatic
energy calculated with the “image” charge would give twice the actual value. In
reality, of course, there is no electric field in the lower half plane and that is how
the error creeps in.)
Although we have found a solution to the problem of a point charge above
a grounded conducting plane of infinite extent by the method of images, it is not
clear whether this solution is unique. The uniqueness of the solution can be seen
only from an analysis of the Laplace or the Poisson equation which we will discuss
later in this chapter. ◭
◮ Example (Point charge between two grounded intersecting planes). As a sec-

ond example we study a physical system consisting of a point charge q in front of
two intersecting orthogonal conducting planes which are of infinite extent and are
grounded as shown in Fig. 3.4. We will use the method of images to solve this
problem (without going into too much detail). We will see that the solution for
this system can be determined provided we have three “image” charges.
y
d1
−q b b q
d2
2d2 x
2d1
q b b
−q
Figure 3.4: The point charge q at (x = d1 , y = d2 , z = 0) and the

three image charges.
Let the two infinite, orthogonal and intersecting conducting planes be de-
scribed by x = 0, y ≥ 0 and y = 0, x ≥ 0 respectively. If we assume that the point
charge is on the plane z = 0, then, it is easy to conclude that all the “image”
charges would also lie on the same plane. In fact, it is easy to check that with the
choice of the “image” charges shown in the figure, the potential at any point on
the plane at x = 0 is
q q
Φ(x = 0, y, z) = 1 − 1
(d21 + (d2 − y)2 + z 2 ) 2 (d21 + (d2 − y)2 + z 2 ) 2
q q
+ 1 − 1
(d21 + (d2 + y)2 + z 2 ) 2 (d21 + (d2 + y)2 + z 2 ) 2
= 0. (3.15)
Similarly, the potential at any point on the plane at y = 0 also vanishes.

q q
Φ(x, y = 0, z) = 1 − 1
((d1 − x)2 + d22 + z 2 ) 2 ((d1 + x)2 + d22 + z 2 ) 2
q q
+ 1 − 1
((d1 + x)2 + d22 + z 2 ) 2 ((d1 − x)2 + d22 + z 2 ) 2
= 0. (3.16)
Thus, these image charges indeed reproduce the boundary condition of vanishing
potential on the two infinite planes.
Once we have determined the “image” charges, we can forget about the
conducting planes and determine the potential in the region x > 0 and y > 0
simply to be
q
Φ(x, y, z) = 1
((x − d1 )2 + (y − d2 )2 + z 2 ) 2
q q
− 1 − 1
((x + d1 )2 + (y − d2 )2 + z2) 2 ((x − d1 )2 + (y + d2 )2 + z 2 ) 2
q
+ 1 . (3.17)
((x + d1 )2 + (y + d2 )2 + z 2 ) 2
The electric field can be obtained from (3.17) by taking the gradient. It has a
general structure (namely, all the x̂, ŷ, ẑ components are nonzero) at an arbitrary
point. However, close to the planes, the electric field takes a simpler form. For
example, when x = 0, the electric field has the form
E(x = 0, y > 0, z)
" #
1 1
= 2qd1 x̂ 3 − 3 , (3.18)
(d21 + (y + d2 )2 + z 2 ) 2 (d21 + (y − d2 )2 + z 2 ) 2
while, for y = 0, it has the form
E(x > 0, y = 0, z)
" #
1 1
= 2qd2 ŷ 3 − 3 . (3.19)
((x + d1 )2 + d22 + z 2 ) 2 ((x − d1 )2 + d22 + z 2 ) 2
The surface charge densities now follow from the discontinuities of the elec-
tric field across the two planes. Namely,
1
σ(x = 0, y > 0, z) = x̂ · E(x = 0, y > 0, z)
4π
" #
qd1 1 1
= 3 − 3 ,
2π (d21 + (y + d2 )2 + z 2 ) 2 (d21 + (y − d2 )2 + z 2 ) 2
1
σ(x > 0, y = 0, z) = ŷ · E(x > 0, y = 0, z)
4π
" #
qd2 1 1
= 3 − 3 . (3.20)
2π ((x + d1 )2 + d22 + z 2 ) 2 ((x − d1 )2 + d22 + z 2 ) 2
Using (3.20), the total induced charge on the conducting planes can now be de-
termined as follows.
Z∞ Z∞
Qinduced = dy dz σ(x = 0, y > 0, z)
0 −∞
Z∞ Z∞
+ dx dz σ(x > 0, y = 0, z)
0 −∞
ZZ " #
∞
qd1 1 1
= dydz 3 − 3
π 0 (d21 + (y + d2 )2 + z 2 ) 2 (d21 + (y − d2 )2 + z 2 ) 2
ZZ " #
∞
qd2 1 1
+ dxdz 3 − 3
π 0 ((x + d1 )2 + d22 + z 2 ) 2 ((x − d1 )2 + d22 + z 2 ) 2
Z∞
2qd1 1 1
=− dy − 2
π y 2 + 2d2 y + d21 + d22 y − 2d2 y + d21 + d22
0
Z∞
2qd2 1 1
− dx − 2
π x2 + 2d1 x + d21 + d22 x − 2d1 x + d21 + d22
0

2qd1 1 −1 d2 2qd2 1 −1 d1
=− tan − tan
π d1 d1 π d2 d2

2q d2 d1
=− tan−1 + tan−1
π d1 d2
2q π
=− × = −q. (3.21)
π 2
In deriving this result, we have used some standard integrals from the tables
(Gradshteyn and Ryzhik, 2.172 and 2.271) as well as the trigonometric relation
that, for x > 0,

1 π
tan−1 x + tan−1 = . (3.22)
x 2
Once again, we see that the total induced charge (on the two plates) is equal to
the sum of all the “image” charges (which is equal in magnitude to the physical
point charge, but opposite in sign). However, the amount of charges on the two
plates depends on the ratio of the perpendicular distances d1 and d2 of the point
charge from the two planes. ◭
◮ Example (Point charge outside a grounded conducting sphere). Another exa-

mple of the method of images is the system of a point charge q at a distance d
from the center of a conducting sphere of radius R where we will assume that
d > R. The sphere is grounded, namely, the surface of the sphere is maintained
at zero potential. Clearly, the potential inside the conducting sphere will be zero
(since every surface inside the conducting sphere defines an equipotential surface
with the same potential as the surface of the sphere) and the electric field within
the sphere will also be zero. Thus, the region where the potential and the electric
field would be nontrivial lies outside the sphere.
Let us assume the origin of our coordinate system to coincide with the
center of the sphere. Without any loss of generality, we can choose the charge to
lie along the z-axis. Then, let us consider the following system of charges, charge
q at z = d and another charge q ′ at z = d′ . Namely, we are considering the effect
of an additional charge located on the line connecting the point charge and the
center of the sphere as shown in Fig. 3.5. It is clear now that the potential for
this combined system of charges at any point r with r ≥ R is given by
(q ′ , d′ ) (q, d)
b b
r−
dẑ
R r
Figure 3.5: A point charge q outside a grounded conducting sphere

and its image charge.
q q′
Φ(r) = +
|r − dẑ| |r − d′ ẑ|
q q′
= 1 + 1 . (3.23)
(r 2 + d2 − 2rd cos θ) 2 (r 2 + d′2 − 2rd′ cos θ) 2
If we require that this potential vanishes when r = R for any angle θ and φ
(namely, on the surface of the sphere), we obtain,
q 2 (R2 + d′2 − 2Rd′ cos θ) − q ′2 (R2 + d2 − 2Rd cos θ) = 0,

′ 2! ′ 2! ′ 2 !
q qd q d
or, R2 1 − + d′2 1 − − 2Rd ′
cos θ 1 − = 0.
q qd′ q d′
(3.24)
Since this must hold for any θ, the coefficient of the last term in (3.24) must vanish
leading to
d′
q ′2 = q 2 . (3.25)
d
Substituting this back into (3.24), the other two terms lead to

R2 R
d′ = , q ′ = −q . (3.26)
d d
Actually, there are two solutions for the charge (as is clear from (3.25)), but this is
the one which gives a vanishing potential (see (3.23)) on the surface of the sphere
r = R.
Thus, we see that, studying the problem of a point charge outside a con-
ducting sphere which is grounded, is equivalent to studying the point charge in
the presence of an “image” charge inside the sphere. The potential at an arbitrary
point outside the sphere is now easily determined to be
q q′
Φ(r) = 1 + 1
(r 2 + d2 − 2rd cos θ) 2 (r 2 + d′2 − 2rd′ cos θ) 2
!
1 R
=q 1 − 1 . (3.27)
(r 2 + d2 − 2rd cos θ) 2 (r 2 d2 + R4 − 2rdR2 cos θ) 2
The electric field can, of course, be calculated from this by taking the gradient
and it is clear that it would, in general, have both a radial and an angular compo-
nent. However, near the surface of the sphere, namely, when r = R, the angular
components cancel out (The simplest way to see this is to note that when taking
derivative with respect to θ, we can set r = R, but then the potential vanishes and
so does the θ component of the gradient at such points. Physically, of course, this
means that there is no tangential component of E on the surface of the conductor
as we would expect.) and we have only a radial component

∂
E(r)|r=R = − ∇Φ(r)|r=R = − r̂ Φ(r)
∂r r=R
!
(R − d cos θ) R(Rd2 − R2 d cos θ)
= qr̂ 3 − 3
(d2 + R2 − 2dR cos θ) 2 R3 (d2 + R2 − 2dR cos θ) 2
q(d2 − R2 ) r̂
=− 3 . (3.28)
R (d2 + R2 − 2dR cos θ) 2
Consequently, we can determine the surface distribution of the charges from the
discontinuity of the electric field, namely,
1
σ(r = R, θ, φ) = r̂ · E(r)|r=R
4π
q(d2 − R2 ) 1
=− 3 . (3.29)
4πR 2 2
(d + R − 2dR cos θ) 2
The total induced charge on the surface of the sphere can now be obtained
from (3.29) through a simple integration and we obtain,
Z
Qinduced = R2 sin θ dθ dφ σ(r = R, θ, φ)
Z1
q(d2 − R2 ) dx
=− × 2πR2 3
4πR (d2 + R2 − 2dRx) 2
−1
1

q(d2 − R2 )R 1 1
=− 1
2 (d + R − 2dRx) −1
dR
2 2 2

q(d2 − R2 ) 1 1
=− −
2d d−R d+R

R
= −q = q′ . (3.30)
d
This again shows that the total charge induced on the sphere is identical to the
“image” charge (which is not equal in magnitude to the point charge in this case).
The force of attraction between the point charge and the sphere can again
be calculated directly or from the “image” charge.

qq ′ q Rd
F(x = 0, y = 0, z = d) = ẑ = −q × 2 ẑ
(d − d′ )2 2
d − Rd
q 2 dR
=− ẑ. (3.31)
(d2 − R2 )2
It is clear from this that, at any point on the z-axis, the force experienced by the
point charge is
q 2 zR
F(x = 0, y = 0, z) = − ẑ. (3.32)
(z 2− R 2 )2
Therefore, the work done in bringing the charge from infinity is easily obtained
to be
Zdẑ Zd
q 2 zR
W = − dℓ · F = − dz − 2
(z − R2 )2
∞ ∞
d
q2 R 1 q2 R
= × − 2 =− . (3.33)
2 z − R2 ∞ 2(d2 − R2 )
Once again, this is half of the energy that we would have found from a calculation
using the “image” charge for reasons which we have discussed earlier. The method
of images works well for all quantities except for the energy of the system. ◭
Although we have discussed the method of images only within

the context of point charges, it works well for other systems such as
line charges etc. But, it is clear from our discussions that only when
there is some symmetry in the problem, it may be easier to determine
the “image” charges, otherwise, the method may not be very useful.
3.2 Boundary conditions for differential equations
In solving dynamical equations of second order such as Newton’s

equation, we normally require two initial conditions (namely, the ini-
tial position and the initial velocity) to solve the equation uniquely.
The Laplace equation as well as the Poisson equation are also second
order equations and yet, we saw that given just one condition such
as the potential on the surface of a conductor, we could solve the
problem completely. Therefore, this raises the question of how one
determines what kind of boundary conditions (or initial conditions)
are necessary for solving a differential equation uniquely.
To examine this, let us start with an ordinary differential equa-
tion of order m of the form
dm f (x) dm−1 f (x)
a0 (x) + a 1 (x) + · · · + am (x)f (x) = F, (3.34)
dxm dxm−1
3.2 Boundary conditions for differential equations 81
where f (x) represents the unknown variable to be determined. In

general, the function F may depend on x as well as f (x). When it
does depend on f (x) in a nonlinear manner, the equation is said to
be nonlinear. However, in most of our discussions we will be inter-
ested in linear equations where F can at the most depend on x (if it
depends linearly on f (x), the linear terms in F can always be com-
bined with the terms on the left). Of course, an equation can have
an infinity of solutions in general and the one appropriate for a par-
ticular physical situation is uniquely selected by the given boundary
conditions. Thus, for example, we know from the study of the har-
monic oscillator equation (which is second order in the time variable)
that, in general, there is an infinity of solutions given by an arbitrary
linear superposition of e±iωt where ω is the natural frequency of the
oscillator. However, if we further specify the initial position as well
as the initial velocity of the oscillator, then the solution is uniquely
determined. Thus, the boundary conditions as well as the surface on
which they are prescribed (in the case of Newton’s equation, they are
known as initial conditions because they are prescribed on the initial
surface t = 0) are quite crucial in determining the unique solution of
a given physical problem.
For the case of the mth order equation in (3.34), it is clear from
our experience with Newton’s equation that a unique solution may
be possible if we know the function f (x) as well as its first (m − 1)
derivatives at some point x0 . (It is assumed that F (x) is a known
function whose value at x0 is known.) This is because at any point
in space, the function f (x) (which we assume to be continuous as is
the case for the potential) has a Taylor expansion of the form

df
f (x) = f (x0 ) + (x − x0 ) + ···
dx x0

(x − x0 )n dn f
+ + ··· . (3.35)
n! dxn x0
We see from this that determining f (x) involves a knowledge of the

derivatives of the function at x0 to all orders. However, we note that
if we know the function and its first (m − 1) derivatives at x0 , we
can determine all the higher order derivatives from the differential
equation itself. For example, from (3.34) we have
"
dm f 1 df
=− −F (x0 ) + am (x0 )f (x0 ) + am−1 (x0 )
dxm x0 a0 (x0 ) dx x0
#
dm−1 f
+ · · · + a1 (x0 ) , (3.36)
dxm−1 x0
and so on for the higher order derivatives. Thus, we see that a

unique solution of an mth order ordinary differential equation needs
m boundary conditions, namely, the values of the function as well as
its first (m−1) derivatives at a coordinate, say x0 . Note that the mth
derivative (as well as the higher order ones) can no longer be specified
independently, but is (are) determined from the differential equation
itself. Specifying the mth derivative as well will only over-specify the
solution. In contrast, specifying a fewer number of derivatives will
not determine the solution uniquely.
Let us also note that if we specify the boundary conditions at a
m
coordinate where a0 (x0 ) = 0, then it is clear from (3.36) that ddxmf |x0
as well as the higher derivatives cannot be determined and hence a
solution cannot be uniquely determined. In fact, in such a case, ei-
ther the boundary conditions are consistent with the equation itself
in which case infinitely many solutions are possible, or the boundary
conditions are inconsistent with the equation implying that no solu-
tion satisfying the given boundary conditions is possible. The point
x0 where the coefficient of the highest derivative term of the differ-
ential equation vanishes is known as the characteristic point of the
equation and it is clear that to have a unique solution, we need to
specify the right number of boundary conditions at points which are
not characteristic points of the equation. Boundary conditions where
one specifies the function as well as its first derivative on a bound-
ary (for a second order equation) are known as Cauchy conditions
and finding a solution subject to such boundary conditions is known
as the Cauchy problem (Cauchy initial value problem if time is in-
volved). However, as we will see for some equations other boundary
conditions are more appropriate to obtain a unique solution.
3.2.1 Partial differential equations. Let us next analyze the boundary

conditions for partial differential equations. For simplicity of discus-
sion, let us start with a general two dimensional second order partial
differential equation of the form

∂2f ∂2f ∂2f ∂f ∂f
A 2 + 2B +C 2 =F x, y, , . (3.37)
∂x ∂x∂y ∂y ∂x ∂y
Here, we are going to assume that the equation is linear and that the
coefficients A, B and C are, in general, functions of x, y. Furthermore,
the two independent variables x, y can both be space coordinates or
one space and one time coordinate.
Unlike the one dimensional ordinary differential equation that
we discussed earlier, here specifying the function and its (two) first
derivatives at a point will not be enough to determine the solution
uniquely. Rather, we need to specify appropriate boundary condi-
tions on a curve. (In general, the solution of a partial differential
equation in n variables needs boundary conditions specified on a
(n−1) dimensional hypersurface.) Furthermore, as we will see Cauchy
boundary conditions may not always work in these cases because they
may over-specify the solution.
To understand the nature of boundary conditions and the curve
on which they must be specified, let us represent the boundary curve
parametrically by ξ = x(s) and η = y(s) where s is the distance of a
point on the curve from some reference point. At any point on the
curve, there are two orthogonal directions – one along the tangent to
the curve and the other normal to it (see Fig. 3.6). The unit vectors
along these directions are easily determined to be
y
êt
ên
Figure 3.6: Unit vectors êt and ên which are respectively tangential
and normal to the curve at a given point.
dξ dη
êt = x̂ + ŷ,
ds ds
dη dξ
ên = ẑ × êt = − x̂ + ŷ. (3.38)
ds ds
These are orthogonal by construction and are easily checked to be

unit vectors from the relation that
2 2
dξ dη
+ = 1, (3.39)
ds ds
which follows from the fact that the infinitesimal distance between
two points on the curve can be written as
ds2 = dξ 2 + dη 2 . (3.40)
The value of the function on the boundary curve can be represented

as f (ξ, η) = f (s). It is clear then that we can define the derivative of
the function along the curve as well as along the normal to it as

dξ ∂f dη ∂f
ft = êt · ∇f |x(s),y(s) = +
ds ∂x x(s),y(s) ds ∂y x(s),y(s)
df
≡ , (3.41)
ds

dη ∂f dξ ∂f
fn = ên · ∇f |x(s),y(s) =− + .
ds ∂x x(s),y(s) ds ∂y x(s),y(s)
This analysis makes it clear that once we know f (s) along the
curve, we also know its derivative along the curve, since ft = df ds .
Therefore, the first order derivative which needs to be specified as
an independent boundary condition (for the Cauchy problem) is the
derivative normal to the curve or fn . A solution at any point will, of
course, have a Taylor expansion of the form
f (x, y) = f (x(s), y(s))

∞ X n
X (x − x(s))n−m (y − y(s))m ∂nf
+ ,
n! ∂x n−m ∂y m
n=1 m=0 x(s),y(s)
(3.42)
and if we know all the partial derivatives in the expansion, then

the solution can be uniquely determined. The question, therefore,
is whether from a knowledge of the values of f (s) and fn (s), we
can determine all the partial derivatives and, therefore, the solution
uniquely.
To start with, let us note that using ( dξ 2 dη 2
ds ) + ( ds ) = 1 in (3.41),
we obtain (this basically corresponds to inverting (3.41))

∂f dξ dη
= ft − fn = a(s),
∂x x(s),y(s) ds ds

∂f dη dξ
= ft + fn = b(s). (3.43)
∂y x(s),y(s) ds ds
Thus, we see that given f (s) and fn , we can determine the (two) first
derivatives directly from the data (remember ft = dfds ). Furthermore,
since a(s) and b(s) are known functions, by taking their derivative as
well as using the differential equation, we have

dξ ∂ 2 f dη ∂ 2 f da(s)
2
+ = ,
ds ∂x x(s),y(s) ds ∂x∂y x(s),y(s) ds

dξ ∂2f dη ∂2f db(s)
+ = ,
ds ∂x∂y x(s),y(s) ds ∂y 2 x(s),y(s) ds

∂2f ∂2f ∂2f
A(s) 2 + 2B(s) + C(s) 2 = F (s). (3.44)
∂x ∂x∂y ∂y x(s),y(s)
This is a set of three coupled inhomogeneous equations in the three

unknown second order derivatives and has a unique solution only if
the determinant of the coefficient matrix does not vanish. Thus, as
long as
dξ dη

ds ds 0

∆ = 0 dξ
ds
dη 6= 0,
ds
(3.45)

A 2B C
we can determine the second derivatives uniquely. Furthermore, once

these are known, by successive differentiation, the higher derivatives
can also be determined and, consequently, it would appear that the
Cauchy problem can be uniquely solved, but as we will see shortly it
is not that simple.
On the other hand, if the characteristic determinant vanishes,
we have (upon expanding the determinant)
2 2
dη dξ dη dξ
A − 2B +C = 0, (3.46)
ds ds ds ds
which is a quadratic equation with solutions

√
dη B± B 2 − AC dξ
= ,
ds A ds
√
B± B 2 − AC
or, dη = dξ. (3.47)
A
If this holds, then the Cauchy problem cannot be uniquely solved.
These two equations are equations for two curves which in the present
case are known as the characteristic curves and we see that the
Cauchy problem cannot be solved if the Cauchy data are specified
on any of the characteristics. This is reminiscent of the behavior in
the case of ordinary differential equations.
For a given second order equation, A, B, and C are known func-
tions and depending on the behavior of the radical in (3.47), partial
differential equations can be classified into three different groups. If
B 2 > AC, then we see that there are two real characteristic curves
of the equation. Such equations are known as hyperbolic equations.
The most familiar of the hyperbolic equations is the wave equation
of the form (in 1 + 1 dimensions)
∂2f 1 ∂2f
− = 0, (3.48)
∂x2 v 2 ∂t2
where v represents the speed of propagation of the wave.
y y
x x
Figure 3.7: A curve intersects the characteristic curves (on the left)
while it is tangential to a characteristic (on the right).
For a hyperbolic equation, there are two families of characteristic

curves. Of course, if the Cauchy data are specified along any one of
the two characteristics, the solution cannot be uniquely obtained.
On the other hand, if the Cauchy data are given on a curve which
intersects all the characteristics exactly once, then, it is easy to see
that the Cauchy problem can be solved uniquely. But, if the boundary
is a curve which is tangent to any of the characteristic curves at any

point (this, therefore, includes closed curves, see, for example, the
second plot in Fig. 3.7), specifying the Cauchy data along the entire
curve will over-specify the solution. In such a case, one needs to
specify the Cauchy data only on part of the curve and either the
function or the normal derivative alone on the other parts of the
curve. Specifying only the value of the function on the boundary
is known as the Dirichlet boundary condition while specifying the
normal derivative on a boundary is known as the Neumann boundary
condition.
If the differential equation is such that B 2 < AC, then, it is
known as an elliptic equation and clearly, in this case, the charac-
teristic curves are complex, namely, we do not have any real charac-
teristics. A familiar example of such an equation which shows up in
many physical problems is the Laplace equation (in two dimensions),
∂2f ∂2f
∇2 f = + = 0, (3.49)
∂x2 ∂y 2
or its higher dimensional generalizations. (In electrostatics, for ex-

ample, we can identify f = Φ in a region free of charge.) In such
a case, since no real characteristic curves are present, it would seem
that the Cauchy boundary conditions would be sufficient to solve the
problem. However, further analysis shows that, in this case, Cauchy
boundary conditions over-specify the solution and the problem can
be solved uniquely only if the function or the normal derivative or a
linear combination of the two is specified on a closed curve. Thus,
either Dirichlet or Neumann boundary condition or a linear combina-
tion of the two, specified along a closed boundary, solves an elliptic
equation such as the Laplace equation uniquely. This is familiar from
the examples we have already worked out involving the conductors.
We have seen that we can solve the Laplace equation (in three di-
mensions) when the potential or the charge density is specified on a
given surface. (Remember that the surface charge density is related
to the normal component of the electric field which is the normal
derivative of the potential.) When only the Dirichlet or Neumann
boundary condition is specified, the corresponding problem is known
as a Dirichlet/Neumann problem. On the other hand, if a linear com-
bination of the two is specified, then, the problem is referred to as a
mixed boundary value problem. Mixed boundary value problems are
harder to solve analytically and are, in general, solved numerically.
Finally, if the partial differential equation is such that B 2 = AC,
then, it is known as a parabolic equation. The diffusion equation,
∂2f ∂f
=α , (3.50)
∂x2 ∂t
is an example of a parabolic equation. In the case of a parabolic
equation, there is only one characteristic curve and a unique solution
can be obtained only for either Dirichlet or Neumann (or a mixed)
boundary condition on an open curve.
Although our discussion so far has been within the context of
two dimensions, it can be generalized to higher dimensions as well.
For a general second order partial differential equation in n dimen-
sions, we can always find a suitable coordinate transformation to
diagonalize the equation. When, only one of the coefficients of the
diagonalized equation is negative, the equation is known as a hyper-
bolic equation. If none of the coefficients of the diagonalized equation
is negative, then, it is known as an elliptic equation and if any one
of the coefficients vanishes, then, the equation is called a parabolic
equation. In general, hyperbolic equations are solved uniquely by
specifying Cauchy boundary conditions on an open hypersurface, el-
liptic equations by specifying Dirichlet/Neumann (or mixed) bound-
ary conditions on a closed hypersurface and parabolic equations by
specifying Dirichlet/Neumann (or mixed) boundary conditions on an
open hypersurface.
3.2.2 Uniqueness theorem. Let us now go back to the Laplace equa-

tion in three dimensions involving the electrostatic potential. Let us
assume that
∇2 Φ = 0, (3.51)
holds in a region of space denoted by the volume V whose bounding

surface is S. This is an elliptic equation and, as we have seen, for a
unique solution, we must specify either the value of the potential Φ
or ∂Φ
∂n (which is the negative of the normal component of the electric
field) on S. Suppose Φ1 and Φ2 represent two solutions of the Laplace
equation satisfying the same boundary condition on S, then we will
show that they can at the most differ by a constant.
To prove this, let us first note that the Laplace equation is a lin-
ear partial differential equation and hence a superposition of distinct
solutions also defines a solution. Therefore, let us define
Φ = Φ1 − Φ2 , (3.52)
which also satisfies the Laplace equation, namely,
∇2 Φ = ∇2 Φ1 − ∇2 Φ2 = 0. (3.53)
Using Gauss’ theorem, let us next note that we can write

Z Z
d3 r ∇ · (Φ(∇Φ)) = ds · (∇Φ)Φ,
V S
Z Z
3 2
2 ∂Φ
or, d r (∇Φ) + Φ(∇ Φ) = ds Φ,
∂n
V S
Z
or, d3 r (∇Φ)2 = 0. (3.54)
V
Here, we have used the fact that Φ satisfies the Laplace equation
(3.53) so that the second term on the left hand side does not con-
tribute. Furthermore, for either Dirichlet or Neumann boundary con-
ditions on S, either Φ or ∂Φ
∂n vanishes on S so that the right hand side
identically vanishes. The final result, as it stands, shows that the in-
tegral of a positive quantity vanishes and, consequently, the integrand
on the left in (3.54) must vanish, namely,
∇Φ = 0. (3.55)
This would seem to say that Φ = Φ1 − Φ2 is at most a constant. For

Dirichlet boundary condition this constant must vanish since it van-
ishes on the boundary S. On the other hand, for Neumann boundary
condition for which
n̂ · ∇Φ = 0, (3.56)
on S, we can only say that Φ1 and Φ2 can at most differ by a constant.

However, since a constant is not relevant for calculations of physical
quantities like the electric field, we can say that for a given set of
boundary conditions, either Dirichlet or Neumann, the solution of
the Laplace equation is unique. Namely, if we find a solution to a
given problem, it must be the unique solution. This shows that the
solutions we found earlier using the method of images must be the
unique solutions for those particular physical problems.
Incidentally, the uniqueness theorem can also be extended to the
solutions of the Poisson equation quite easily. For, suppose Φ1 and
Φ2 satisfy the Poisson equation
∇2 Φ1 = −4πρ(r), ∇2 Φ2 = −4πρ(r), (3.57)

with the same Dirichlet/Neumann boundary condition on S, then,

Φ = Φ1 − Φ2 would satisfy the Laplace equation. This would, there-
fore, imply as before in the derivation of (3.55) that
∇Φ = 0. (3.58)
For solutions satisfying the Dirichlet boundary condition on S, we

have
Φ = 0, (3.59)
and, consequently, the two solutions are the same. However, for
Neumann boundary conditions, Φ = constant which for the purposes
of calculating electric field etc. still implies that the solution is unique.
3.3 Solutions of the Laplace equation
Let us now analyze the solutions of the Laplace equation in three

dimensions, subject to specific boundary conditions. In Cartesian
coordinates the Laplace equation has the form
2
2 ∂ ∂2 ∂2
∇ Φ(r) = + + Φ(x, y, z) = 0, (3.60)
∂x2 ∂y 2 ∂z 2
which, in spherical coordinates, takes the form

1 ∂ 2∂ 1 ∂ ∂ 1 ∂2
r + sin θ + Φ(r, θ, φ) = 0,
r 2 ∂r ∂r r 2 sin θ ∂θ ∂θ r 2 sin2 θ ∂φ2
(3.61)
while, in cylindrical coordinates, it has the form

1 ∂ ∂ 1 ∂2 ∂2
r + + Φ(r, φ, z) = 0. (3.62)
r ∂r ∂r r 2 ∂φ2 ∂z 2
Depending on the symmetry in the problem, it is meaningful to solve

the equation in the appropriate coordinates. The Laplace equation
appears in many branches of physics and its solutions are known as
harmonic functions. To begin with, let us summarize some of the
general properties of harmonic functions.
3.3.1 General properties of harmonic functions. The harmonic func-

tions possess several interesting properties which can be derived with-
out explicitly solving the Laplace equation.
3.3 Solutions of the Laplace equation 91
1. Let us assume that Φ represents a solution of the Laplace equa-

tion in a given region V , namely,
∇2 Φ = 0, (3.63)
in V . Then, it follows from Gauss’ theorem that

Z Z
ds · (∇Φ) = d3 r ∇2 Φ = 0. (3.64)
S V
Here S is assumed to be the surface which bounds the volume

V . (In electrostatics this would correspond to the fact that the
integral of the electric field over a closed surface or the electric
flux out of the region must vanish.) Furthermore, using Stokes’
theorem, it follows that for any function Φ (this is independent
of whether Φ satisfies the Laplace equation or not and in elec-
trostatics corresponds to the electric field being conservative)
I Z
dℓ · (∇Φ) = ds · (∇ × ∇Φ) = 0. (3.65)
C S
2. Let us note that, given any two functions A and B, we have the
identity
∇ · (A∇B − B∇A) = (A∇2 B − B∇2 A), (3.66)
so that using Gauss’ theorem, we can write

Z Z
3 2 2
d r (A∇ B − B∇ A) = ds · (A∇B − B∇A). (3.67)
V S
1
This is known as Green’s identity. If we now choose A = |r|
and B = Φ, namely, if B represents a solution of the Laplace
equation (3.63) and A satisfies (this is known as the Green’s
function for the Laplacian)
1
∇2 A = ∇2 = −4π δ3 (r), (3.68)
|r|
then, from (3.67) (we are going to assume that S is the surface
of a sphere of radius R), we obtain
Z Z
3 3 1 r̂
4π d r Φ(r) δ (r) = ds · (∇Φ) + 2 Φ ,
V S R R
Z
1
or, Φ(r = 0) = ds Φ(R, θ, φ). (3.69)
4πR2 S
Here we have used Eq. (3.64). Relation (3.69) leads to the

result that if Φ satisfies the Laplace equation (3.63) inside a
sphere of radius R, then the value of Φ at the origin of the
sphere is the average of the value of Φ over the surface of the
sphere (recall that the surface area of a sphere of radius R is
4πR2 ). This is something which we have seen earlier in specific
calculations, but this is, in fact, a general property of harmonic
functions. This is a very important result, for it has many
interesting consequences.
3. One of the most interesting consequences of (3.69) is the fact

that a harmonic function without any singularities in a given
region (namely, satisfying the Laplace equation in the entire
region) cannot have a maximum or a minimum value in that
region. This can be proved easily in the following manner. Let
us suppose that Φ (satisfying the Laplace equation in a given
region) has a maximum at a point r0 in that region. If we draw
a small sphere around r0 , then, it is clear that ds · (∇Φ) which
is proportional to the normal derivative of Φ must be negative
at every point on the surface of the sphere since r0 represents
a maximum of Φ. It follows, then, that
Z
ds · (∇Φ) < 0. (3.70)
S
This is, however, inconsistent with the general property (3.64)

of the solution of the Laplace equation, namely,
Z
ds · (∇Φ) = 0. (3.71)
S
Consequently, there cannot be a maximum of Φ at any point

in the region. By a similar argument, it is straightforward to
show that Φ cannot have a minimum at any point in the region
either. (Incidentally, this is the reason why there can be no
stable equilibrium in pure electrostatics. To have electrostatic
equilibrium, one must apply some other form of force as well.)
4. It follows from the above property that if Φ satisfies the Laplace

equation in a given region and if its value is a constant on the
bounding surface of the region, then, it must have the same
constant value in the entire region. The proof of this is really
quite simple. As we have seen, Φ cannot have any maximum or
minimum inside the region. Therefore, the maximum and the
minimum can occur at best on the bounding surface. However,

Φ is a constant on the surface and, consequently, its maximum
and minimum are the same and equal to its value on the surface
(also equal to its value in the interior). This implies that Φ is
a constant inside the region and all the way up to the surface.
This also has the implication that if the solution of the Laplace
equation is valid in the entire space, it must vanish if it vanishes
asymptotically.
3.3.2 Solution in Cartesian coordinates. The solutions of the Laplace

equation are not hard to work out because of the very special struc-
ture of the Laplacian operator which leads to separable solutions in a
number of coordinate systems. In this section, we will work out the
solutions in the Cartesian coordinates subject to appropriate bound-
ary conditions.
L1
L3 y
L2
x
Figure 3.8: A conducting rectangular box with the faces maintained

at fixed potentials.
Let us consider the physical problem of a conducting rectangular

box of sides L1 , L2 and L3 respectively as shown in Fig. 3.8. All
the faces of the box are grounded, except for the top face which is
maintained at the given potential f (x, y). We would like to determine
the potential everywhere in the interior of the box. It is clear that
the potential can be easily obtained by solving the Laplace equation
in Cartesian coordinates, namely,
2
∂ ∂2 ∂2
+ + Φ(x, y, z) = 0, (3.72)
∂x2 ∂y 2 ∂z 2
in the region 0 ≤ x ≤ L1 , 0 ≤ y ≤ L2 and 0 ≤ z ≤ L3 , subject to the
boundary conditions
Φ(0, y, z) = Φ(x, 0, z) = Φ(x, y, 0) = 0,
Φ(L1 , y, z) = Φ(x, L2 , z) = 0,
Φ(x, y, L3 ) = f (x, y), (3.73)
in this region.
In trying to solve the Laplace equation for this system, let us
note that the Laplacian in Cartesian coordinates (see (3.72)) is really
a sum of three terms which commute with one another. In such a case,
it is a general result that the solution can be written as a product of
three terms, each depending on only one coordinate, namely, in such
a case, we expect the solution to have the factorized form
Φ(x, y, z) = X(x)Y (y)Z(z). (3.74)
Substituting this form of the solution into the Laplace equation (3.72)
and dividing by Φ throughout, we obtain
1 d2 X(x) 1 d2 Y (y) 1 d2 Z(z)
+ + = 0. (3.75)
X(x) dx2 Y (y) dy 2 Z(z) dz 2
Since each of the three terms in the above expression depends on only
one coordinate x, y or z, their sum can vanish for arbitrary values of
the coordinates only if each of the terms equals a constant such that
1 d2 X(x)
= α1 ,
X(x) dx2
1 d2 Y (y)
= α2 ,
Y (y) dy 2
1 d2 Z(z)
= α3 , α1 + α2 + α3 = 0. (3.76)
Z(z) dz 2
The boundary conditions (3.73) for Φ can now be translated to
conditions on the individual component functions as
X(0) = X(L1 ) = Y (0) = Y (L2 ) = Z(0) = 0,
Z(L3 ) = constant. (3.77)

The three ordinary differential equations in (3.76) can now be solved
subject to the boundary conditions (3.77) and they lead to

mπx (m) mπ 2
Xm (x) = sin , α1 = − ,
L1 L1
2
nπy (n) nπ
Yn (y) = sin , α2 =− , (3.78)
L2 L2
2 2
sinh(αmn z) (m,n) mπ nπ
Zmn (z) = , α3 = α2mn = + ,
sinh(αmn L3 ) L1 L2
with m, n = 1, 2, . . . and we have normalized the z solution for later
convenience. The determination of these solutions uses only the five
homogeneous boundary conditions on Φ(x, y, z) with the condition
for the surface z = L3 in (3.73) yet to be implemented.
A general solution of the problem can now be written as a linear
superposition of the form (it is not unique yet)
∞
X
Φ(x, y, z) = Amn Xm (x)Yn (y)Zmn (z)
m,n=1
∞
X

mπx nπy sinh(αmn z)
= Amn sin sin ,(3.79)
L1 L2 sinh(αmn L3 )
m,n=1
where Amn are constants which can be determined by imposing the

last of the boundary conditions in (3.73), namely,
∞
X
mπx nπy
Φ(x, y, L3 ) = Amn sin sin = f (x, y).
L1 L2
m,n=1
(3.80)
Since f (x, y) is a given function, this relation can be inverted to

determine
ZL1 ZL2
4 mπx nπy
Amn = dx dy f (x, y) sin sin , (3.81)
L1 L2 L1 L2
0 0
where we have used the standard orthonormality relation for trigono-

metric functions, namely,
ZL nπx
n′ πx L
dx sin sin = δnn′ . (3.82)
L L 2
0
With this, the determination of the (unique) solution of the

Laplace equation (namely, the potential) in the interior of the box
is complete. Physical quantities such as the electric field can now
be obtained from the potential. This method of solving the Laplace
equation is known as the method of separation of variables.
3.3.3 Solution in spherical coordinates. Let us next solve the Laplace

equation in spherical coordinates given in (3.61)

1 ∂ 2∂ 1 ∂ ∂ 1 ∂2
r + sin θ + Φ(r, θ, φ) = 0.
r 2 ∂r ∂r r 2 sin θ ∂θ ∂θ r 2 sin2 θ ∂φ2
(3.83)
Once again, the Laplace equation is separable in spherical coordinates

and, therefore, let us try a solution in the product form
Φ(r, θ, φ) = R(r)Θ(θ)Q(φ). (3.84)
Substituting this back into the Laplace equation and dividing through-
out by rΦ2 , we obtain

1 d 2 dR 1 d dΘ 1 1 d2 Q
r + sin θ + = 0.
R dr dr Θ sin θ dθ dθ sin2 θ Q dφ2
(3.85)
Since the expression within the parenthesis in the last term of (3.85)
is the only term which depends on φ, this equation cannot be satisfied
(for arbitrary values of φ) unless this term equals a constant, namely,
we must have
1 d2 Q
= −m2 , (3.86)
Q dφ2
where the choice of the sign of the constant is for convenience. Equa-
tion (3.86) can be readily integrated to give
Qm (φ) = e±imφ . (3.87)
Since we expect the solutions to be single valued the constant m is

required to be an integer (so that φ and φ + 2π lead to the same
solution). Allowing for both positive as well as negative integers (in-
cluding zero) for m, we note that we can write the solution depending
on the azimuthal angle as
Qm (φ) = eimφ , m = 0, ±1, ±2, . . . . (3.88)
Substituting the form of the solution (3.88) into (3.85), we ob-

tain

1 d 2 dR 1 d dΘ m2
r =− sin θ + = k. (3.89)
R dr dr Θ sin θ dθ dθ sin2 θ
Here we have used the fact that since the left hand side and the right
hand side of (3.89) depend on independent variables, this relation
will hold only if each side equals a constant which we denote by k.
We see from (3.89) that the θ equation takes the simple form

1 d dΘ m2
sin θ + k− Θ = 0. (3.90)
sin θ dθ dθ sin2 θ
Introducing the variable x = cos θ, this equation can be written in

the simple form

d dΘ m2
1 − x2 + k− Θ = 0. (3.91)
dx dx (1 − x2 )
A systematic analysis of this equation reveals that its solutions be-

come unphysical at x = ±1 unless k = ℓ(ℓ + 1) where ℓ = 0, 1, 2, . . .
and m = −ℓ, −ℓ+1, . . . , ℓ−1, ℓ for any given value of ℓ. (Incidentally,
if the physical problem excludes the regions x = ±1 or θ = 0, π, then
these restrictions would not apply. However, in most applications,
we will have 0 ≤ θ ≤ π, so that we will consider only these integer
values for the separation constants k, m.)
For these integer values of the separation constants, the θ equa-
tion takes the form

d
2 dPℓ,m (x) m2
1−x + ℓ(ℓ + 1) − Pℓ,m (x) = 0.
dx dx (1 − x2 )
(3.92)
This is known as the associated Legendre equation and the solutions,

Pℓ,m (x), are known as the associated Legendre polynomials. In par-
ticular, when m = 0 the equation (Pℓ,m=0 (x) = Pℓ (x))

d dPℓ (x)
1 − x2 + ℓ(ℓ + 1)Pℓ (x) = 0, (3.93)
dx dx
is known as the Legendre equation and the solutions, Pℓ (x), are poly-
nomials of order ℓ known as the Legendre polynomials. It can be eas-
ily checked that the associated Legendre polynomials are related to
the Legendre polynomials through the relation (associated Legendre
polynomials are not really polynomials for odd values of m as is clear
from (3.94))
|m| d|m| Pℓ (x)

Pℓ,m (x) = (1 − x2 ) 2 , ℓ ≥ |m|. (3.94)
dx|m|
The Legendre polynomials have a closed form expression given

by the Rodrigues’ formula
1 dℓ 2
Pℓ (x) = (x − 1)ℓ , (3.95)
2ℓ ℓ! dxℓ
from which the explicit forms of the first few Legendre polynomials
can be easily determined to be
P0 (x) = P0 (cos θ) = 1,
P1 (x) = P1 (cos θ) = x = cos θ,
1 1
P2 (x) = P2 (cos θ) = (3x2 − 1) = (3 cos2 θ − 1), (3.96)
2 2
and so on. These are precisely the functions that we encountered in
the last chapter in connection with the expansion of the potential
for a point charge displaced along the z-axis (see (2.38)). It is also
clear that we can write a closed form expression for the associated
Legendre polynomials from the Rodrigues’ formula as well.
It is also possible to write a generating function for the Legendre
polynomials in a simple manner. Consider a function of two variables
1
T (x, s) = 1 . (3.97)
(1 − 2sx + s2 ) 2
The Taylor expansion of this function around s = 0 gives the Legen-

dre polynomials as the coefficients of the expansion, namely,
∞
X
1
T (x, s) = 1 = sℓ Pℓ (x). (3.98)
(1 − 2sx + s2 ) 2 ℓ=0
In fact, this generating function leads to recursion relations satisfied

by the Legendre polynomials and using these it is straightforward to
show that the Pℓ (x)’s in the Taylor expansion in (3.98) do satisfy the
Legendre equation. This relation is particularly interesting because
it immediately gives
1 1
′
= 1
|r − r | (r 2 − 2rr ′ cos θ + r ′2 ) 2
1 1
=
r 2r ′

r′ 2
1
2
1− r cos θ + r
∞
1 X r′ ℓ
= Pℓ (cos θ), (3.99)
r r
ℓ=0
where we have assumed that r > r ′ . This is, of course, the expansion
of the potential that we have discussed earlier in (2.39) and (2.46).
A similar expansion is trivially obtained when r < r ′ , simply by
interchanging r and r ′ in the previous expression (which is symmetric
in r, r ′ ).
The Legendre polynomials can be shown to satisfy the orthonor-
mality relation
Z1
2
dx Pℓ (x) Pℓ′ (x) = δℓℓ′ . (3.100)
2ℓ + 1
−1
In general, when m 6= 0, the total angular part of the solution of the

Laplace equation is written (in the normalized form) as
s
2ℓ + 1 (ℓ − |m|)!
Yℓ,m(θ, φ) = ǫm Pℓ,m (cos θ) eimφ , (3.101)
4π (ℓ + |m|)!
m+|m|
where the phase is conventionally chosen to be ǫm = (−1) 2 .
The Yℓ,m ’s are known as the spherical harmonics and are the eigen-
functions of the angular momentum operator in quantum mechanics.
They satisfy the orthonormality relation
Z
∗
sin θ dθ dφ Yℓ,m (θ, φ)Yℓ′ ,m′ (θ, φ) = δℓℓ′ δmm′ . (3.102)
Note that when m = 0, the spherical harmonics reduce to the Legen-

dre polynomials up to the normalization constant.
Once the angular part of the solution of the Laplace equation is
determined, the radial equation (3.89) becomes

1 d 2 dRℓ (r)
r = ℓ(ℓ + 1),
Rℓ (r) dr dr
d2 Rℓ (r) 2 dRℓ (r) ℓ(ℓ + 1)
or, + − Rℓ (r) = 0. (3.103)
dr 2 r dr r2
The two independent solutions of this equation are clearly
Rℓ (r) = r ℓ , or, Rℓ (r) = r −(ℓ+1) , (3.104)
and the particular choice of the solution depends on the region in

which the problem is being investigated. Thus, if we are looking for
the solution outside an enclosed region (which includes the origin)

such that the potential vanishes asymptotically, it is the second so-
lution that is relevant. On the other hand, if we are looking at the
solution in a region which includes the origin and where we assume
the potential to be nonsingular, then the first form of the radial so-
lution is the natural choice. In general, of course, the radial solution
can be written as a linear superposition of the two independent solu-
tions. In any case, independent of which form of the radial solution
we choose, a general solution of the Laplace equation in spherical
coordinates takes the form
X
Φ(r, θ, φ) = Aℓ,m Rℓ (r) Yℓ,m (θ, φ), (3.105)
ℓ,m
where Aℓ,m ’s are constants, which can be determined from the given
boundary conditions of a physical system. We note here that if we
have a physical problem where the potential does not depend on the
azimuthal angle φ (namely, when m = 0), the corresponding solutions
are known as zonal harmonics.
◮ Example (Sphere in uniform electric field). As an example of a physical sys-
tem where spherical solutions of the Laplace equation may be used, let us con-
sider space without any free charge consisting of a uniform background electric
field along the z-axis. Thus, the field lines can be drawn as parallel lines of the
same magnitude and we can write
E(r) = E ẑ, (3.106)
which suggests an associated electrostatic potential of the form
Φ(r) = −Ez + C = −Er cos θ + C = −ErP1 (cos θ) + CP0 (cos θ), (3.107)
where C is a constant. Clearly, this has azimuthal symmetry and as a result

there is no dependence on the azimuthal angle φ. (Uniform electric fields can be
produced by large capacitors with a small separation. Uniform electric fields over
the entire space is, therefore, not a physical concept. However, we can think of all
space to mean only a relatively large region.) Since the electric field is uniform, it
follows that ∇ · E = −∇2 Φ = 0, which can also be seen explicittly from (3.106)
and (3.107).
If we now introduce a conducting sphere of radius R, then, the field lines
will be distorted around the surface of the conducting sphere (see Fig. 3.9). Far
away from the surface of the sphere, the field lines will continue to be uniform with
the potential as given in (3.107). However, there will be surface charges induced
on the sphere so that the field lines will end on the surface of the sphere and
there will be no electric field inside the sphere. The surface of the sphere will, of
course, have a constant potential which we will take to be Φ0 . We are interested
in determining the potential as well as the electric field outside the sphere.
Because of the azimuthal symmetry present in the problem, it is clear that if
we use the spherical solutions of the Laplace equation, we can write the potential
Figure 3.9: A conducting sphere of radius R in a uniform electric

field.
outside the sphere as (see (3.105))

∞
X
Φ(r) = Aℓ r ℓ + Bℓ r −(ℓ+1) Pℓ (cos θ). (3.108)
ℓ=0
Here even though we are interested in the solution outside the sphere (the origin
of the coordinate system is chosen to be at the center of the sphere), we have
allowed for both the independent forms of the radial solution since the potential
for a constant electric field is linear in r in this region. Furthermore, since there
is no free charge anywhere (so that Φ satisfies the Laplace equation) and

1
∇2 = −4πδ 3 (r), (3.109)
r
we conclude that B0 = 0. The constants Aℓ can be determined from the boundary

conditions for the problem. First, let us note that since
lim Φ(r) → −ErP1 (cos θ) + CP0 (cos θ)

r→∞
∞
X
= lim Aℓ r ℓ + Bℓ r −(ℓ+1) Pℓ (cos θ), (3.110)
r→∞
ℓ=0
using the orthonormality of the Legendre polynomials in (3.100) we determine

Aℓ = 0 for ℓ ≥ 2. Furthermore, from the asymptotic structure of the potential
(3.107), we conclude that
A0 = C, A1 = −E. (3.111)
All the terms with coefficients Bℓ vanish asymptotically and hence there is no
constraint on these coefficients from the asymptotic condition. Thus, we can
write for r ≥ R,
∞
X
Φ(r) = C − Er cos θ + Bℓ r −(ℓ+1) Pℓ (cos θ). (3.112)
ℓ=1
We still have to satisfy the boundary condition that the potential on the surface
of the sphere is a constant Φ0 . Requiring this, we obtain
∞
X
Φ0 = C − ER cos θ + Bℓ R−(ℓ+1) Pℓ (cos θ), (3.113)
ℓ=1
which determines
C = Φ0 , B1 = ER3 , Bℓ = 0 for l > 1, (3.114)

so that we can write the potential outside the sphere, satisfying the boundary
conditions, to correspond to

R3
Φ(r) = Φ0 − E 1 − 3 r cos θ. (3.115)
r
The form of the electric field outside the sphere can now be determined to be
E(r) = −∇Φ(r)
!
∂ θ̂ ∂ R3
→ − r̂ + Φ0 − E 1 − 3 r cos θ
∂r r ∂θ r

2R3 R3
= r̂ E 1 + 3 cos θ − θ̂ E 1 − 3 sin θ. (3.116)
r r
Thus, we see that, in general, the electric field has a radial as well as a θ
component. However, on the surface of the sphere (r = R), the theta component
vanishes, so that the electric field is normal to the surface. Furthermore, it is not
a uniform electric field on the surface, rather its value depends on the angle θ.
From this, we can determine the surface charge density induced on the sphere to
be (we are using the fact that E = 0 inside the condocting sphere)
1 3E
σ(R, θ) = r̂ · E(R) = cos θ. (3.117)
4π 4π
Integrating this over the surface of the sphere, we obtain
Z
Qinduced = R2 dΩ σ(R, θ)
Zπ Z2π
3ER2
= dθ sin θ cos θ dφ = 0. (3.118)
4π
0 0
This is consistent with our earlier discussion in the last chapter. Namely, the
conducting sphere remains neutral, the positive and the negative charges simply
rearrange themselves so as to cancel the electric field inside the sphere. Note from
(3.116) that asymptotically for large r, the electric field becomes
lim E → E(r̂ cos θ − θ̂ sin θ) = Eẑ, (3.119)

r→∞
consistent with (3.106). In deriving the surface charge density we have assumed
that there is no electric field inside the conducting sphere. However, this analysis
can also be carried out in the interior of the sphere as well to show that the
potential is a constant Φ0 and there is no electric field inside which also follow
from the general properties of harmonic functions that we discussed earlier. ◭
3.3.4 Circular harmonics. The Laplace equation can also be solved

by the method of separation of variables in cylindrical coordinates.
In this case, the solutions involve Bessel functions. Instead of go-
ing through this complete solution, let us consider the simpler case
where the potential is independent of the z coordinate which shows
up in many physical problems and, therefore, is more useful. When
there is no dependence on the z coordinate, the Laplace equation in
cylindrical coordinates (3.62) takes the form (in fact, this is just the
Laplace equation in the polar coordinates)

1 ∂ ∂ 1 ∂2
r + Φ(r, φ) = 0. (3.120)
r ∂r ∂r r 2 ∂φ2
Writing a separable solution of the form
Φ(r, φ) = R(r)Q(φ), (3.121)
and substituting it into the differential equation (3.120), we obtain

r d dR(r) 1 d2 Q(φ)
r =− = k. (3.122)
R(r) dr dr Q(φ) dφ2
Here, we have used the fact that both sides of (3.122) are functions
of independent variables and, therefore, the relation can be satisfied
for arbitrary values of r, φ only if both sides equal a constant which
we have identified with k. The solution for the φ equation is straight-
forward. We note that we will have a single valued function only if
k = n2 , n = 0, 1, . . ., with the φ solution written as
Qn (φ) = An cos nφ + Bn sin nφ, (3.123)
where An , Bn are constants. Furthermore, for this value of the sepa-

ration constant the radial equation in (3.122) becomes
d2 Rn (r) dRn (r)

r2 2
+r − n2 Rn (r) = 0. (3.124)
dr dr
There are two independent solutions of this equation of the forms
Rn (r) = r n , Rn (r) = r −n , for n 6= 0, (3.125)
so that a general solution for the radial equation (3.124) can be writ-
ten (for n 6= 0) as
Rn (r) = Cn r n + Dn r −n , (3.126)
with Cn , Dn constants. Let us note here that when n = 0, the solution

of the radial equation in (3.124) has the general form
R0 (r) = C0 + D0 ln r, (3.127)
where C0 , D0 are constants. Thus, for a given n we can write the

solution of the Laplace equation (3.120) as
Φn (r, φ) = Rn (r)Qn (φ), (3.128)
which are known as circular harmonics with n denoting the degree of

the harmonics. A general solution of (3.120), of course, will have the
form
X
Φ(r, φ) = Rn (r)Qn (φ). (3.129)
n
◮ Example (Cylinder in uniform electric field). As an example of the use of circu-

lar harmonics, let us consider the problem of an infinitely long conducting cylinder
of radius R in a uniform electric field perpendicular to the axis of the cylinder.
Let us assume that the axis of the cylinder is along the z-axis and that the electric
field is along the x-axis (see Fig. 3.10). There are no free charges anywhere in
space.
Figure 3.10: A conducting cylinder in a uniform electric field.
Clearly, the potential will be independent of the z coordinate by symmetry

and, therefore, we can use circular harmonics. Let us note that since the electric
field is uniform along the x-axis, we can write (at least when the cylinder is not
present or very far away from the cylinder)
E(r) = E x̂, (3.130)
so that the potential will have the form (in polar coordinates)
Φ(r) = −Ex + C = −Er cos φ + C, (3.131)

where C is a constant and r is the radial coordinate on the plane. In the presence
of the conducting cylinder, the field lines will be distorted near the surface of the
cylinder, but asymptotically they will have the form (3.131). The surface of the
cylinder, of course, would be at a constant potential which we take to be Φ0 .
The general solution for the potential in the presence of the cylinder (out-
side) would have the form (since there are no free charges, the potential will satisfy
Laplace equation and in this case will have a solution of the form (3.129))
Φ(r) = C0 + D0 ln r
∞
X
+ (Cn r n + Dn r −n )(An cos nφ + Bn sin nφ). (3.132)
n=1
However, comparing with the asymptotic form of the potential (3.131) and using
the orthonormality relations for the trigonometric functions, we determine that
the only coefficients that are nontrivial are A1 , C0 , C1 , D1 and satisfy
C0 = C, A1 C1 = −E, (3.133)
so that we can write the form of the potential outside the cylinder to be (r ≥ R)
Φ(r) = C − Er cos φ + A1 D1 r −1 cos φ. (3.134)
We see that the term with the coefficient D1 vanishes asymptotically and, as a
consequence, there is no constraint on this coefficient from the asymptotic condi-
tion. (Remember that r is the radial coordinate in cylindrical coordinates and,
therefore, there is no inconsistency with the absence of free charges. Furthermore,
this term also involves an angular function.)
Let us next impose the boundary condition on the surface of the cylinder.
Namely, on the surface of the cylinder (r = R) we have,
Φ0 = C − ER cos φ + A1 D1 R−1 cos φ, (3.135)
which determines
C = Φ0 , A1 D1 = ER2 , (3.136)
so that we can write the potential outside the cylinder (3.134) to be

R2
Φ(r, φ) = Φ0 − E r − cos φ. (3.137)
r
We can now determine the electric field which has the form
E(r, φ) = −∇Φ(r, φ)
!
∂ φ̂ ∂ R2
= − r̂ + Φ0 − E r − cos φ
∂r r ∂φ r

R2 R2
= r̂E 1 + 2 cos φ − φ̂E 1 − 2 sin φ. (3.138)
r r
Once again we see that the electric field outside the cylinder has both a radial
and an angular component. However, on the surface of the cylinder (r = R), only
the normal (radial) component is nontrivial. This allows us to determine the
induced surface charge density on the cylinder to be
1 E
σ(R, φ) = r̂ · E(R) = cos φ, (3.139)
4π 2π
where we are assuming that there is no electric field inside the cylinder since the
potential is a constant. Consequently, the total induced charge on the surface of
the cylinder is obtained to be (L is the length of the cylinder which is to be taken
to infinity at the end)
Z Z2π
ELR
Qinduced = LR dφ σ(R, φ) = dφ cos φ = 0. (3.140)
2π
0
We see that the cylinder remains neutral. The charges simply rearrange them-
selves on the surface to yield zero electric field inside the cylinder, which can be
explicitly checked by carrying out a similar analysis inside the cylinder. Let us
also note that asymptotically,
lim E → E (r̂ cos φ − φ̂ sin φ) = E x̂, (3.141)
r→∞
as we would expect. ◭
3.4 Solution of the Poisson equation
In the absence of free charges, solutions of the (homogeneous) Laplace

equation determine the electrostatic potential and, therefore, the elec-
tric field. However, when free electric charges are present we have to
solve the (inhomogeneous) Poisson equation
∇2 Φ(r) = −4πρ(r), (3.142)
with ρ(r) representing the charge density in order to determine the

potential and the electric field. Let us note that the Laplace equa-
tion represents the homogeneous part of the Poisson equation and it
is important to know the solutions of the Laplace equation for the
determination of the solution of the Poisson equation since we know
that a general solution of any inhomogeneous differential equation
consists of a sum of the solution of the homogeneous equation as well
as the particular solution of the inhomogeneous equation. The free-
dom of adding a homogeneous part to the solution allows us to satisfy
boundary conditions in a simple manner.
Since the solutions of the Laplace equation define a complete
basis, one method of solving the Poisson equation is to expand the
solution in one such (appropriate) complete basis (namely, Carte-
sian, spherical, cylindrical etc.) and impose the relevant boundary
conditions. There is, however, an alternative and powerful method
for solving inhomogeneous differential equations which goes under
the name of the method of Green’s functions. This method is quite
useful in many branches of physics and, therefore, is an important
concept which we discuss next.
3.4 Solution of the Poisson equation 107
3.4.1 Green’s function. To see how the method of Green’s function

works, let us replace the inhomogeneous source term on the right
hand side of the Poisson equation (3.142) by a delta function (source)
and consider the equation
∇2 G(r, r′ ) = −4π δ3 (r − r′ ). (3.143)
Namely, G(r, r′ ) represents the potential (solution of the Poisson

equation) at r due to a unit point source charge at r′ . This is known
as the Green’s function for the Poisson equation (or the Laplacian
operator). It is clear that if we know the Green’s function for a
given equation, then the particular solution for the inhomogeneous
equation can be trivially determined. For example, for the Poisson
equation (3.142), we can write the particular solution to be
Z
Φparticular (r) = d3 r ′ G(r, r′ )ρ(r′ ), (3.144)
which can be easily checked to satisfy the Poisson equation, namely,

Z

∇2 Φparticular (r) = d3 r ′ ∇2 G(r, r′ ) ρ(r′ )
Z
= −4π d3 r ′ δ3 (r − r′ )ρ(r′ )
= −4πρ(r). (3.145)
Thus, it is clear that the knowledge of the Green’s function for a

given equation is quite essential in solving an inhomogeneous differen-
tial equation. The determination of the Green’s function requires the
knowledge of boundary conditions. For example, we already know of
a Green’s function for the Poisson equation, namely, in three dimen-
sions we have seen that

2 1
∇ = −4π δ3 (r − r′ ). (3.146)
|r − r′ |
Therefore, let us identify
1
G(0) (r, r′ ) = . (3.147)
|r − r′ |
The important question to ask is what is the boundary condition for
which this is the appropriate Green’s function. In this case, it is
obvious from (3.147) that for a fixed r′ ,
lim G(0) (r, r′ ) → 0. (3.148)

|r|→∞
Therefore, this is the Green’s function corresponding to the boundary

condition that the solution vanishes asymptotically for large distances
away from the source charge (remember that the Green’s function
really describes a solution of the Poisson equation for a unit source
charge). In this case, we see that the general solution of the Poisson
equation, subject to the boundary condition that it vanishes asymp-
totically, has the form
Z
Φ(r) = Φhomo (r) + d3 r ′ G(0) (r, r′ )ρ(r′ ), (3.149)
V
where Φhomo denotes a solution of the homogeneous equation (van-

shing asymptotically). On the other hand, we have already seen
earlier that the solution of the homogeneous equation (the Laplace
equation), subject to the boundary condition that it vanishes asymp-
totically, is identically zero. Therefore, the unique solution of the
Poisson equation subject to this boundary condition is
Z Z
ρ(r′ )
Φ(r) = d r G (r, r )ρ(r ) = d3 r ′
3 ′ (0) ′ ′
, (3.150)
|r − r′ |
which is the familiar relation for the potential due to a charge distri-
bution that we have discussed earlier (see (1.60)).
The relation between the solution of an inhomogeneous problem
and the Green’s function can be easily seen from the Green’s identity
which we have derived earlier (see (3.67)), namely,
Z Z
′ ′
d3 r ′ A∇ 2 B − B∇ 2 A = ds′ · A∇′ B − B∇′ A . (3.151)
V S
In (3.151), if we choose A = G(r, r′ ), the generic Green’s function for

the Laplacian and B = Φ(r′ ), the solution of the Poisson equation,
we obtain
Z Z
′2 ′2
d r (G∇ Φ − Φ∇ G) = ds′ ·(G∇′ Φ − Φ∇′ G), (3.152)
3 ′
V S
which can be simplified to give

 
Z Z
4π − d3 r ′ G(r, r′ )ρ(r′ ) + Φ(r) = ds′ · (G∇′ Φ − Φ∇′ G).
V S
(3.153)
Here, we have used the Poisson equation (3.142) as well as the equa-
tion satisfied by the Green’s function (3.143). Furthermore, if we use
∂
the notation ês · ∇ = ∂n to represent the normal derivative, we can
rewrite (3.153) also as
Z ′ ′

1 ′ ′ ∂Φ(r ) ′ ∂G(r, r )
Φ(r) = ds G(r, r ) − Φ(r )
4π ∂n′ ∂n′
S
Z
+ d3 r ′ G(r, r′ )ρ(r′ ). (3.154)
V
This can be compared with the form of the result obtained earlier in
(3.149). Incidentally, the simplest way to check that the first term
in (3.154) must be a solution of the homogeneous equation is to note
that
 
Z
∇2 Φ(r) − d3 r ′ G(r, r′ )ρ(r′ ) = 0. (3.155)
V
A direct verification of this relation for the surface integral is, how-
ever, tricky.
3.4.2 Dirichlet boundary condition. It appears from the general result

(3.154) that to determine the unique solution of the Poisson equation
in the presence of some conducting surfaces we need the value of both
Φ and ∂Φ∂n on the surfaces. But, as we have seen, a unique solution of
an elliptic equation requires either Φ (Dirichlet boundary condition)
or ∂Φ
∂n (Neumann boundary condition) to be specified on the boundary
and not both. The way out of this puzzle is that we have not yet
specified a boundary condition for the Green’s function. For example,
suppose we are investigating a Dirichlet boundary value problem,
then, the value of Φ(r) on the surface is known. Consequently, if
we require the Green’s function to satisfy the homogeneous Dirichlet
boundary condition, namely,
GD (r, r′ ) = 0, for r or r′ ∈ S, (3.156)
we see that the solution (3.154) can be written as
Z
1 ∂GD (r, r′ )
Φ(r) = − ds′ Φ(r′ )
4π ∂n′
S
Z
+ d3 r ′ GD (r, r′ )ρ(r′ ), (3.157)
V
which is indeed a well defined solution for the given Dirichlet bound-
ary value problem. It is worth noting that this solution is valid even
in the region which does not contain any charges (compare this with
the method of images where one does not calculate the potential in
the region containing the image charge). In such a case, the second
term vanishes and the solution is given completely by the surface
integral. (Let us also emphasize here that even though the bound-
ary condition for the solution may be inhomogeneous, the Green’s
function is required to satisfy only the homogeneous condition.)
The main question to answer now is what is the form of the
Green’s function satisfying the homogeneous Dirichlet boundary con-
dition (3.156). The exact structure of the Green’s function would, of
course, depend on the particular problem under consideration. How-
ever, let us note that we can write a general Green’s function to be
of the form
G(r, r′ ) = G(0) (r, r′ ) + H(r, r′ ), (3.158)
where
∇2 H(r, r′ ) = 0. (3.159)
Namely, the Green’s function is unique only up to addition of a term

which satisfies the homogeneous equation. We can, therefore, take
advantage of this arbitrariness to impose a given boundary condition
on a particular Green’s function.
◮ Example (Point charge above a conducting plane). As an application of this
method, let us study the problem of a point charge above a conducting plane of
infinite extent which is maintained at some constant potential Φ0 . We can take
the plane to be at z = 0, assume that the point charge q is at z = d on the z-axis,
and we are interested in determining the potential for z > 0 (see Fig. 3.1).
This is a Dirichlet boundary value problem and, as we have seen, to deter-
mine the potential we need the Dirichlet Green’s function in the region z, z ′ ≥ 0,
satisfying the homogeneous boundary condition GD (r, r′ ) = 0 whenever z = 0 or
z ′ = 0. To determine this Green’s function, let us note from our general discussion
that
GD (r, r′ ) = G(0) (r, r′ ) + H(r, r′ )

1
= p + H(r, r′ ). (3.160)
(x − x′ )2 + (y − y ′ )2 + (z − z ′ )2
We require this Green’s function to vanish whenever z = 0 or z ′ = 0. It is clear

from this requirement that
1
H(r, r′ ) = − p . (3.161)
(x − x′ )2 + (y − y ′ )2 + (z ± z ′ )2
Although there are two possible choices for H(r, r′ ) we note that for the second
choice of the sign in the denominator in (3.161) we will have GD (r, r′ ) = 0 which
is a trivial solution. Therefore, the first choice of the sign is the natural one and
we obtain
1
GD (r, r′ ) = p
(x − x ′ )2 + (y − y ′ )2 + (z − z ′ )2
1
−p . (3.162)
(x − x′ )2 + (y − y ′ )2 + (z + z ′ )2
There are two things to note from (3.162). First, with our choice of H(r, r′ ),
we note that
!
2 ′ 2 1
∇ H(r, r ) = −∇ p
(x − x′ )2 + (y − y ′ )2 + (z + z ′ )2
= 4πδ(x − x′ )δ(y − y ′ )δ(z + z ′ ). (3.163)
Therefore, it appears that H(r, r′ ) does not satisfy the homogeneous equation.
However, note that in the region that we are interested in, namely, z, z ′ > 0, the
right hand side of (3.163) indeed vanishes and H(r, r′ ) satisfies the homogeneous
equation in this region. The second thing to note is that the Green’s function
in (3.162) is reminiscent of the structure of the potential for a grounded plane
obtained by the method of images in (3.5). This should not be surprising since
the Green’s function is the potential for a unit source charge with a homogeneous
Dirichlet boundary condition.
Substituting (3.162) into the right hand side of the solution (3.157), we
obtain (S in this case is the plane z ′ = 0. Actually, S is the closed surface
bounding the upper half plane. However, as is clear from the form of the Green’s
function, the Green’s function as well as its derivative vanish when any of the
coordinates is at infinity. Consequently, the surface S is effectively the plane at
z ′ = 0.)
Z Z
1 ∂GD (r, r′ )
Φ(r) = − ds′ Φ(r′ ) + d3 r ′ GD (r, r′ )ρ(r′ )
4π ∂n′
S
Z
Φ0 2zdx′ dy ′
= 3
4π ((x − x′ )2 + (y − y ′ )2 + z 2 ) 2
Z
+ d3 r ′ GD (r, r′ )qδ(x′ )δ(y ′ )δ(z ′ − d)
Z
Φ0 z 2dx′ dy ′
= 3
4π (x′2 + y ′2 + z 2 ) 2
!
1 1
+q 1 − 1 . (3.164)
(x2 + y 2 + (z − d)2 ) 2 (x2 + y 2 + (z + d)2 ) 2
(Note that there is an extra negative sign in the surface term because the sur-
∂ ∂
face area points along the negative z-axis so that ∂n ′ = − ∂z ′ .) Here, we have
translated the coordinates of integration in the first integral for simplicity and
recognize that it is best evaluated in polar coordinates

Z∞ Z
2dx′ dy ′ 2r ′ dr ′ dφ′
3 = 3
(x′2 + y ′2 + z 2 ) 2 (r ′2 + z 2 ) 2
−∞
Z∞
dr ′2
= 2π 3
(r ′2 + z 2 ) 2
0
∞
1
= 2π (−2) √ = 4π . (3.165)
r + z 0
′2 2 z
Substituting this back into the solution (3.164), we have
!
1 1
Φ(r) = Φ0 + q 1 − 1 . (3.166)
(x2 + y 2 + (z − d)2 ) 2 (x2 + y 2 + (z + d)2 ) 2
For Φ0 = 0, this reduces to the solution (3.5). However, we now have the solution
for the case when the conducting plane is not grounded, but is held at a nonzero
constant potential. Incidentally, from (3.166) (or (3.164)), we see that the poten-
tial in the lower half plane is given by (the second term vanishes in the lower half
plane because there is no free charge in that region)
Φ(r) = Φ0 , for z ≤ 0, (3.167)
which is consistent with what we expect, namely, that there is no electric field in
the lower half plane. All the field lines from the point charge terminate on the
conducting plane. (An alternative way to see that the surface term in (3.164) for
a constant potential Φ0 on the boundary is simply equal to Φ0 is as follows. By
definition
∇′2 GD (r, r′ ) = −4πδ 3 (r − r′ ). (3.168)
Consequently, using Gauss’ theorem we can write
Z Z
ds′ · ∇′ GD (r, r′ ) = d3 r ′ ∇′2 GD (r, r′ )
S
Z
= d3 r ′ (−4π)δ 3 (r − r′ ) = −4π, (3.169)
which gives the desired result.) ◭
◮ Example (Point charge inside a conducting sphere). Let us next analyze the
problem of a conducting sphere of radius R which contains a point charge q in-
side, at a distance d from the center of the sphere. The surface of the sphere is
maintained at a constant potential Φ0 and we are interested in determining the
potential within the sphere.
For simplicity, let us choose the origin to coincide with the center of the
sphere and the point charge q to lie on the z-axis so that we are interested in the
solution of the equation
∇2 Φ = −4π qδ(x)δ(y)δ(z − d), 0 ≤ x, y, z ≤ R, d < R, (3.170)
subject to the boundary condition that
Φ(R, θ, φ) = Φ0 . (3.171)
Therefore, this defines a Dirichlet problem. Once again, the Green’s function for
the problem, satisfying the homogeneous Dirichlet boundary condition, can be
written as (see (3.158))
1
GD (r, r′ ) = + H(r, r′ ). (3.172)
|r − r′ |
Since H(r, r′ ) has to satisfy the homogeneous equation (for r, r ′ < R), we note
that we can write it as
α
H(r, r′ ) = − , (3.173)
|r − r′′ |
where we assume that α is a constant (namely, independent of r) and r′′ = r′′ (r′ )
and lies outside the sphere (namely, r ′′ > R). Now, requiring that GD (r, r′ )|r=R =
0, we obtain
1 α
1 = 1 ,
(R2 + r ′2 − 2Rr ′ cos γ ′ ) 2 (R2 + r ′′2 − 2Rr ′′ cos γ ′′ ) 2
or, α2 (R2 + r ′2 − 2Rr ′ cos γ ′ ) = (R2 + r ′′2 − 2Rr ′′ cos γ ′′ ). (3.174)

Here, we are using the generic notation that
cos γ ′ = cos θ cos θ′ + sin θ sin θ′ cos(φ − φ′ ). (3.175)

′ ′′
It is clear that (3.174) can be satisfied if r and r lie along the same axis (this
also follows from the fact that since r′′ is a vector function of r′ , it must have the
form r′ multiplied by a scalar function of r ′ ) and a nontrivial solution is obtained
if
R R2
α= , r ′′ = α2 r ′ = , γ ′′ = γ ′ . (3.176)
r′ r′
Using (3.176), then, we can write
1 R
GD (r, r′ ) = 1 − 1 .
(r 2 + r ′2 − 2rr ′ cos γ ′ ) 2 (r 2 r ′2 + R4 − 2R2 rr ′ cos γ ′ ) 2
(3.177)
It is manifestly symmetric in r ↔ r′ and it is straightforward to verify explicitly
that it satisfies the homogeneous Dirichlet boundary condition whenever r or r′
lies on the surface of the sphere. Furthermore, its structure can be compared with
the potential obtained from the method of images.
The solution of the Poisson equation inside the sphere is now straightforward
∂ ∂ ′
(r < R, ∂n ′ = ∂r ′ ). (For simplicity of evaluation, we will measure the angle θ
′ ′
with respect to the vector r in the first integral so that γ = θ .)
Z Z
1 ∂GD (r, r′ )
Φ(r) = − ds′ Φ(r′ ) + d3 r ′ GD (r, r′ )ρ(r′ )
4π ∂n′
S V
Zπ Z2π
Φ0 (R2 − r 2 ) R2 sin θ′ dθ′ dφ′
= 3
4πR (r 2 + R2 − 2rR cos θ′ ) 2
0 0
Z
+ d3 r ′ GD (r, r′ )qδ(x′ )δ(y ′ )δ(z ′ − d)
Z1
Φ0 (R2 − r 2 )R dx
= 3
2 (r 2 + R2 − 2rRx) 2
−1
!
1 R
+q 1 − 1
(r 2 + d2 − 2rd cos θ) 2(d2 r 2 + R4 − 2R2 dr cos θ) 2
1

Φ0 (R2 − r 2 )R 1 1
= 1
2 Rr (r 2 + R2 − 2rRx) 2
−1
!
1 R
+q 1 − 1
(r 2 + d2 − 2rd cos θ) 2 (d2 r 2 + R4 − 2R2 dr cos θ) 2
!
1 R
= Φ0 + q 1 − 1 .
(r 2 + d2 − 2rd cos θ) 2 (d2 r 2 + R4 − 2R2 dr cos θ) 2
(3.178)
This can be compared with the solution obtained from the method of images. ◭
3.4.3 Neumann boundary condition. The solution of the Poisson equa-

tion subject to Neumann boundary conditions is slightly tricky for a
variety of reasons. First, let us note that if
∇2 Φ(r) = −4πρ(r), (3.179)
then, using Gauss’ theorem we can write
Z Z
ds · ∇Φ = d3 r ∇2 Φ(r),
S V
Z Z
∂Φ
or, ds = −4π d3 r ρ(r) = −4πQ. (3.180)
∂n
S V
This shows that the normal derivative of Φ cannot be specified ar-

bitrarily on the boundary surface. Rather, it should satisfy the
constraint condition (3.180) which does not make this a well posed
boundary value problem. Furthermore, as in the earlier case of Dirich-
let boundary condition, for a solution of the boundary value problem
to exist we must define the Neumann Green’s function with the ho-
mogeneous condition
∂GN (r, r′ )
= 0, (3.181)
∂n′
on the boundary surface. However, let us note from the definition of
the Green’s function that
∇′2 GN (r, r′ ) = −4πδ3 (r − r′ ). (3.182)
Consequently, using Gauss’ theorem we obtain

Z Z
′ ′ ′
ds · ∇ GN (r, r ) = d3 r ′ ∇′2 GN (r, r′ ),
S V
Z Z
∂GN
′ (r, r′ )
or, ds = −4π d3 r ′ δ3 (r − r′ ) = −4π. (3.183)
∂n′
S V
This is not compatible with the homogeneous Neumann condition

(3.181).
This second difficulty can be circumvented by defining the Neu-
mann Green’s function as satisfying the differential equation

′2 ′ 1 3 ′
∇ GN (r, r ) = −4π δ (r − r ) − , (3.184)
V
where V denotes the volume of the region under consideration. In

such a case, use of Gauss’ theorem leads to
Z Z
′ ′ ′ ∂GN (r, r′ )
ds · ∇ GN (r, r ) = ds′ = 0, (3.185)
∂n′
S S
so that a Green’s function satisfying the homogeneous Neumann

boundary condition can be defined consistently.
With such a choice of the Green’s function we note that Gauss’
theorem (Green’s identity (3.67)) leads to
Z Z
3 ′ ′2 ′2
d r (GN ∇ Φ − Φ∇ GN ) = ds′ · (GN ∇′ Φ − Φ∇′ GN ),
V S
 
Z Z
Φ ∂Φ
or, − 4π  d3 r ′ GN ρ + − Φ(r) = ds′ GN .
V ∂n′
V S
(3.186)
Here, we have used the fact that Φ satisfies the Poisson equation while
GN satisfies (3.184). Noting that for any function A(r) the average
over a given volume is defined to be
Z
1
Aavg =A= d3 r A(r), (3.187)
V
V
we now obtain the solution from (3.186) to be

Z
1 ∂Φ(r′ )
Φ(r) = Φ + ds′ GN (r, r′ )
4π ∂n′
S
Z
+ d3 r ′ GN (r, r′ )ρ(r′ ). (3.188)
V
However, the Neumann boundary condition on ∂Φ ∂n still has to be

specified consistent with the constraint (3.180) which makes such
problems extremely difficult and impractical. If we do not constrain
the boundary conditions (data), the Neumann problem may not have
a solution. (Incidentally, if V is infinite, the average term will not be
present in the solution if the potential falls off asymptotically.)
◮ Example (Point charge above a conducting plane). Let us analyze the follow-
ing example to show how this method works. We have already seen in (3.8) that a
point charge q above a (grounded) conducting plane, say, on the z-axis at a height
d, induces a surface charge density on the plane given by (here r represents the
radial coordinate on the x − y plane)
1 1 ∂Φ qd

σ= ẑ · E = − =− 3 ,
4π 4π ∂z 2π(r + d2 ) 2
2

∂Φ 2qd
or, = 3 . (3.189)
∂z (r 2 + d2 ) 2
Here the restriction stands for z = 0, namely, the location of the plane. Conse-
quently, we can use this as a Neumann boundary condition which automatically
satisfies the appropriate constraint and try to solve for the potential of the prob-
lem. We are interested in the solution in the upper half plane so that the volume
is infinite implying that the average term in (3.188) can be ignored.
The Green’s function satisfying the homogeneous Neumann boundary con-
dition can be determined as before and has the form
1
GN (r, r′ ) = p
(x − x′ )2 + (y − y ′ )2 + (z − z ′ )2
1
+p . (3.190)
(x − x ′ )2 + (y − y ′ )2 + (z + z ′ )2
The additional term as we have seen earlier in (3.163), is a solution of the homo-
geneous equation and the sign is chosen such that the z (or z ′ ) derivative vanishes
if either z or z ′ vanishes (on the surface of the plane). The solution can now
be obtained from (3.188). However, unlike the Dirichlet boundary condition it is
clear that the surface term now gives a coordinate dependent term and to compare
with what we have already done by the method of images, let us calculate the
potential along the z-axis (x = 0, y = 0, z ≥ 0). For such points we obtain (the
negative sign in the surface term arises from the direction of the outward normal
∂ ∂
to the surface as discussed earlier, namely, ∂n ′ = − ∂z ′ )
Z Z
1 ∂Φ
Φ(x = 0 = y, z) = ds′ GN (r, r′ ) ′ + d3 r ′ GN (r, r′ )ρ(r′ )
4π ∂n
S
Z
2qd 2r ′ dr ′ dφ′
=− 3 1
4π (r ′2 + d2 ) 2 (r ′2 + z 2 ) 2
Z
+ d3 r ′ GN (r, r′ )qδ(x′ )δ(y ′ )δ(z ′ − d)
Z∞
q ∂ dr ′2
= (2π) 1 1
2π ∂d (r ′2 + d2 ) 2 (r ′2 + z 2 ) 2
0

1 1
+q +
|z − d| z+d
2q q q
=− + +
z+d |z − d| z+d
q q
= − , (3.191)
|z − d| z+d
which is what we had calculated earlier in (3.5) using the method of images. (See,
for example, Gradshteyn and Ryzhik 2.261 for the value of the first integral.) ◭
1. Consider a point charge q at a distance d from the center of a

conducting sphere of radius R, where d > R. The surface of
the sphere is maintained at a constant potential Φ0 .
a) Determine the “image” charges needed to study this problem.
b) Determine the potential at any point outside the sphere.
c) Determine the induced surface charge density as well as the
total induced charge.
2. Consider a hollow metallic sphere of finite thickness, with the

inner radius a and the outer radius b. (The coordinate origin
is chosen to be at the center of the sphere.) A point charge q
is placed inside the sphere at a distance a2 from the center of
the sphere (the sphere is insulated so that the charge cannot
move).
a) What is the potential at a point r outside the sphere (r ≫ b)?
b) What are the potentials at the inner (r = a) as well as the
outer (r = b) surfaces of the sphere?
c) What is the potential at the center of the sphere? (Use the
method of images to calculate this result.)
3. Two infinite grounded parallel conducting planes are separated

by a distance d. A point charge q is placed between the planes.
Determine the induced surface charge densities as well as the

total charges on the two planes.
4. Consider a conducting sphere of radius R whose surface is main-

tained at a potential Φ(R) = Φ0 cos θ. Assuming that there are
no free charges present (inside or outside), what is the potential
inside and outside the sphere?
5. Consider a cylindrical conducting can of radius R and height h.

The side and the bottom walls of the can are grounded while the
top face is maintained at Φ0 (r, φ). Find the electrostatic poten-
tial inside the can. (You will have to solve the Laplace equation
in cylindrical coordinates, which lead to Bessel functions.)
6. Let us assume that the potentials Φ1 and Φ2 are produced

by the charge distributions (ρ1 , σ1 ) and (ρ2 , σ2 ) respectively.
Namely, both volume and surface charge distributions are re-
sponsible for the potentials that correspond to solutions of the
Poisson equations
∇2 Φ1 = −4πρ1 ,
∇2 Φ2 = −4πρ2 .
Using Green’s identity discussed in this chapter, prove the reci-

procity theorem
Z Z Z Z
3 3
d r ρ1 Φ2 + ds σ1 Φ2 = d r ρ2 Φ1 + ds σ2 Φ1 .
V S V S
Using this reciprocity theorem, determine the total induced

charges on each of two infinite grounded conducting plates sepa-
rated by a distance d, with a point charge q in the space between
them.
7. Consider the one dimensional wave equation
∂2Φ 1 ∂2Φ
2
− 2 = 0,
∂x v ∂t2
where v represents the speed of propagation of the wave. This
is a hyperbolic equation, and at every point in space-time, there
are two characteristics ξ(x, t) and η(x, t).
a) Determine the characteristics as functions of (x, t).
b) Invert the relations for the characteristics to express x and

t in terms of the characteristics ξ and η. What is the form of
the wave equation written in terms of the characteristics as the
independent coordinates.

c) With the Cauchy conditions Φ(x, t = 0) = f (x) and ∂Φ
∂t t=0 =
g(x), determine the solution to the wave equation.
8. Using Green’s identity, as well as other identities discussed in

this chapter, show that the Green’s function satisfying Dirichlet
boundary conditions is symmetric in the interchange of its ar-
guments (namely, GD (r, r′ ) = GD (r′ , r)). (In general, Neumann
Green’s functions are not symmetric.)
Chapter 4
Dielectrics
4.1 Electric displacement field
As we have discussed earlier dielectrics are materials that can be

polarized. Namely, in the presence of an external electric field the
bound electrons and the positively charged nuclei of the atoms of
such materials get displaced slightly so that each individual atom
behaves like a point dipole and all these dipoles inside the material
become aligned to give the material a macroscopic dipole moment.
We say that the material becomes polarized in the presence of an
electric field.
Experimentally it is observed that the net polarization of a ma-
terial depends on the applied electric field and the general relation
between the two is of the form
Pi = χij Ej . (4.1)
The χij are known as the components of the electric susceptibility

tensor of the material (for individual atoms the constant of propor-
tionality is known as the polarizability of the atom) and it can, in
principle, depend on the electric field itself. However, for electric
fields which are not too strong the components of the tensor are in-
dependent of the electric field and can be thought of as constants
for a given medium. Furthermore, the electric susceptibility is a sec-
ond rank tensor and, in fact, it is a symmetric tensor. In general,
therefore, it can have at the most six independent components. How-
ever, by a suitable coordinate transformation (namely, with a suitable
choice of the coordinate axes), it can be brought to a diagonal form
with the three diagonal elements, in general, distinct. On the other
hand, there is a wide class of dielectric materials which are isotropic
(or sometimes also called linear) for which the three diagonal elements
are the same. In this case, the relation between the polarization and
121
122 4 Dielectrics
the electric field is a linear one of the form
P = χE, (4.2)
with χ, which is a scalar, representing the electric susceptibility of

the linear dielectric material. We will restrict ourselves to such mate-
rials for simplicity. (The electric susceptibility is a positive quantity,
namely, χ ≥ 0 which simply represents the fact that the polarization
is along the direction of the applied electric field.)
We have already seen (see (2.60)) that when a dielectric material
is polarized it produces a potential of the form
Z Z ′
ds′ · P(r′ ) 3 ′ ∇ · P(r )
′
Φdipole (r) = − d r , (4.3)
|r − r′ | |r − r′ |
S V
which allowed us to conclude that a polarized dielectric develops a

volume charge density as well as a surface charge density given re-
spectively by (n̂ is the unit vector normal to the surface)
ρb = −(∇ · P), σb = n̂ · P. (4.4)
Here, the subscript “b” simply stands for the fact that these charges
are bound and are not free to move around unlike the charges in a
conductor. It follows from the identification in (4.4) that the total
charge in the dielectric is
Z Z
Qb = d3 r ρb (r) + ds σb
V S
Z Z
3
=− d r∇·P+ ds n̂ · P(r)
V S
Z Z
3
=− d r ∇ · P(r) + ds · P(r) = 0, (4.5)
V S
where the last identity follows from Gauss’ theorem. This is, of
course, what we would expect. Namely, the dielectric is charge neu-
tral, all that happens in the presence of an external electric field is
that the charges are displaced slightly to give it a macroscopic dipole
moment.
It is intuitively clear that because a polarized dielectric develops
a volume as well as a surface density of bound charges, the differential
form of Gauss’ law satisfied by the electric field in a dielectric would
4.1 Electric displacement field 123
modify. To determine this systematically, let us consider a dielectric

of infinite extent with some point charges embedded inside. In this
case, it is clear that if we consider a Gaussian surface as shown in
Fig. 4.1, Gauss’ law (1.28) would lead to
si b
qi b
S
b
s s2
q11 q2
Figure 4.1: A dielectric with a number of embedded free charges.

The dashed curve represents the Gaussian surface enclosing the free
charges.
Z
ds · E = 4π(Q + Qb ), (4.6)
S
where Q represents the sum of free charges embedded inside the di-
electric and Qb is the total bound charge within the Gaussian vol-
ume (which need not be zero unlike the case of the whole dielectric).
By definition (ρb is contained only in the volume excluding the free
charges)
Z Z
Qb = d3 r ρb (r) + ds σb (r)
P
V Si
i
Z Z
3
=− d r (∇ · P) + ds · P
P
V Si
i
Z Z
=− ds · P + ds · P
P P
S+ Si Si
i i
Z
=− ds · P. (4.7)
S
124 4 Dielectrics
Here Si represents the surface area of the interface between the charge
qi and the dielectric. Putting this back into the integral form of
Gauss’ law (4.6), we obtain
Z
ds · (E + 4πP) = 4πQ,
S
Z
or, ds · D = 4πQ,
S
or, ∇ · D(r) = 4πρ(r). (4.8)
Here, we have used Gauss’ theorem in deriving the last line and have
defined a new vector field
D(r) ≡ E(r) + 4πP(r), (4.9)
which is known as the electric displacement vector (electric displace-

ment field). Therefore, in a dielectric it is the flux of the electric
displacement field out of a Gaussian surface which equals (4π) times
the free charge contained in the Gaussian volume. Correspondingly,
it is the divergence of this field that is proportional to the density
of free charges which is quite useful because, in reality, we can have
information only on the free charges in a system. (The number of
bound charges is large and their distribution is clearly an impossible
thing to determine.)
Therefore, in the presence of a dielectric the differential form of
Gauss’ law takes the form
∇ · D = 4πρ. (4.10)
In spite of its similarity with the differential form of Gauss’ law (1.30)
satisfied by the electric field in the absence of a dielectric, the two
vector fields may have quite different characters in general. For ex-
ample, as we have seen in chapter 1 the electric field is conservative,
but D may not be, namely,
∇ × E = 0, ∇ × D 6= 0. (4.11)
This follows because the polarization vector may not have vanishing
curl in general. Furthermore, note that since the polarization vector
is parallel to the electric field, (for isotropic dielectrics) we can write
D = E + 4πP = (1 + 4πχ)E = ǫ E, ǫ = 1 + 4πχ. (4.12)

Thus, we see that in an isotropic (linear) dielectric the electric dis-

placement vector is parallel to the electric field and the constant of
proportionality ǫ is known as the permittivity or the dielectric con-
stant of the material. (Even when P is parallel to E, it may still have
non-vanishing curl if χ is space dependent.) From the definition of χ
it follows that the dielectric constant for a material is greater than
unity, ǫ ≥ 1. The dielectric constant is unity only for vacuum as we
have noted earlier. Note also that even though D and E are linearly
related, in general
4π
∇ · E 6= ρ, (4.13)
ǫ
which follows because the dielectric constant may be different for
different regions of space (namely, it may be space dependent). How-
ever, in a homogeneous region of space (namely, where there is no
change in the dielectric constant), we can write the differential form
of Gauss’ law also in terms of the electric field as
4π
∇·E= ρ. (4.14)
ǫ
◮ Example (Point charge in an isotropic dielectric). As an example of problems
involving dielectric materials, let us consider an isotropic dielectric material of
infinite extent and permittivity ǫ with a point charge q embedded inside at a
point which we choose to be the origin of the coordinate system (see Fig. 4.2) and
we want to calculate the electric as well as the displacement fields. This problem is
very simple and yet clarifies some of the essential properties of dielectric materials.
b q
Figure 4.2: A point charge embedded in an isotropic dielectric. The

dashed sphere represents the Gaussian surface.
From the (spherical) symmetry of the problem we see that the electric field
will be radial (with the charge q at the center) at every point. Furthermore,
drawing a Gaussian sphere of radius r we note that the magnitude of the electric
field will be the same at every point on the surface of this sphere. Therefore, we
126 4 Dielectrics
determine trivially from (4.8) that (remember that D is parallel to the electric
field and, consequently, is radial as well)
4πr 2 |D| = 4πq,

q
or, |D| = . (4.15)
r2
It follows from this that
q
D = 2 r̂,
r
1 1 q 1 q
E= D= r̂, Φ= . (4.16)
ǫ ǫ r2 ǫ r
Namely, the electric field and the potential have the same structures as in vacuum,
but are reduced in magnitude (remember that ǫ ≥ 1). Furthermore, we can also
determine the polarization of the medium from the fact that (see (4.12))
(ǫ − 1) (ǫ − 1) q
P = χE = E= r̂. (4.17)
4π 4πǫ r 2
The reason why the strength of the electric field (as well as the potential) is
reduced in the presence of the dielectric is easily understood from the fact that the
point charge polarizes the medium. In fact, from the structure of the polarization
vector in (4.17) as well as using (4.4), we see that (remember ∇ · r = 3 in three
dimensions and we are interested in points r 6= 0)

(ǫ − 1) qr
ρb = −∇ · P = −∇ · = 0, (4.18)
4πǫ r 3
so that there is no volume density of bound charges in this case and only a surface
charge density is present.
The total bound charge in the Gaussian volume (see Fig. 4.2) can, therefore,
be calculated easily to give (the negative sign is because n̂ is inward at the interface
opposite to the direction of P)
Z Z
(ǫ − 1)q (ǫ − 1)q (ǫ − 1)
Qb = ds · P = − dΩ = − × (4π) = − q. (4.19)
4πǫ 4πǫ ǫ
S
b q
Figure 4.3: The induced negative charges in the interface of the free
charge in the dielectric medium.
All of this charge lies on the inner surface as shown in Fig. 4.3 and, as a
result, at a distance far away from the origin the total effective (free) charge seen
is

(ǫ − 1) q
qeff = q + Qb = q 1 − = . (4.20)
ǫ ǫ
In other words, induced bound charges have a tendency to reduce the magni-
tude of the (free) point charge and, consequently, lead to a weaker electric field.
Conventionally, this is known as the screening of a point charge by a dielectric
medium. ◭
◮ Example (Capacitor filled with dielectric). Once we have Gauss’ law (4.10) in
the presence of a dielectric (or the integral form of it in (4.8)) solving electrostatic
problems involving dielectrics is no more difficult than what we have already done
in chapters 1 and 2. Let us recall that Gauss’ law involves the field D and the free
charge distribution. Consequently, just as we determined E earlier (in chapter 1)
from Gauss’ law we can now determine D. Furthermore, from the relation between
D and E in (4.12) we can then obtain the electric field as well.
−Q Q
D
n̂
Figure 4.4: A capacitor filled with a dielectric of permittivity ǫ. The

dashed rectangle represents the Gaussian surface.
As an example of how this is done, let us consider a large parallel plate
capacitor of area A separated by a small distance d. Let us assume that the
plates carry charges Q and −Q respectively and that the space between the two
capacitor plates is filled with a dielectric material of permittivity ǫ (see Fig. 4.4).
Clearly from the symmetry of the problem we conclude, as before, that the charge
would distribute itself uniformly over the two plates so that we have two plates
with a uniform surface distribution of charge densities σ and −σ respectively. We
expect the electric field (and, consequently, the electric displacement field where
applicable) to be perpendicular to the plates and that it has a nonzero constant
value only between the two plates in a direction from the positively charged plate
to the negatively charged plate. Calculating the flux out of a rectangular Gaussian
surface we conclude that
Z
ds · D = 4πQ,
or, |D|A = 4πQ,

4πQ
or, |D| = = 4πσ. (4.21)
A
128 4 Dielectrics
Thus, we can write the displacement field between the two plates to be
D(r) = −4πσ n̂, (4.22)
where we have defined n̂ to be the unit vector normal to the plates as in Fig. 4.4
(say, along the z-axis). Here the negative sign arises because the direction of the
D field is opposite to the vector n̂.
The electric field between the plates can now be determined from relation
(4.12)
1 4πσ
E(r) = D(r) = − n̂. (4.23)
ǫ ǫ
Thus, as before, we see that the electric field continues to be a constant between
the plates. However, its strength is reduced from the case where there was no
dielectric between the plates. From (4.23) we determine the potential difference
(voltage) between the two plates to be
4πσ 4πd −1
V = |E|d = d= σA = Cdiel. Q. (4.24)
ǫ ǫA
Namely, we determine the capacitance in the presence of the dielectric to be
ǫA
Cdiel. = . (4.25)
4πd
In other words, the capacitance of the system increases in the presence of a di-
electric (compare with (2.89)). ◭
4.2 Boundary conditions in dielectric
ǫ2 (ǫL ) ǫ1 (ǫR )
n̂
Figure 4.5: The interface of two distinct dielectric media with the
dashed curve representing the surface of a rectangular Gaussian vol-
ume.
Suppose, we have two distinct dielectric media, characterized by

the respective dielectric constants ǫ1 (or ǫR ) and ǫ2 (ǫL ), separated
by a surface (perpendicular to the z-axis). Let us draw a rectangular
4.2 Boundary conditions in dielectric 129
Gaussian volume of infinitesimal thickness projecting into both the

media as shown in Fig. 4.5.
It is clear now that in the limit of vanishing thickness of the
rectangular volume, Gauss’ law gives (namely, taking the integral
form of Gauss’ law in (4.8))
n̂ · (DR − DL ) = 4πσ, (4.26)
where σ denotes the density of free charges on the surface separating

the two media. Thus, the normal component of the D field is discon-
tinuous across a surface carrying free charges. On the other hand,
from the conservative nature of the electric field, it follows (taking
the line integral of the electric field or the surface integral of the curl
of the electric field) that the tangential component of the electric field
is continuous across the boundary, namely,
ER,t − EL,t = 0. (4.27)
Equations (4.26) and (4.27) represent the two boundary conditions

that we have to satisfy at an interface of two distinct dielectric ma-
terials.
From the fact that the electric field is defined in terms of the
potential as
E(r) = −∇Φ(r), (4.28)
it follows that in the presence of an isotropic dielectric, we have to

solve the equation
∇ · D(r) = 4πρ(r),
or, ∇ · (ǫ ∇Φ(r)) = −4πρ(r), (4.29)
where ρ(r) represents the density of free charges. Furthermore, equa-

tion (4.29) needs to be solved subject to the boundary conditions
that across a surface separating two dielectric media

∂ΦR ∂ΦL ∂
ǫR = ǫL − 4πσ, = n̂ · ∇,
∂n S ∂n S ∂n
ΦR (r)|S = ΦL (r)|S , (4.30)
which correspond to the boundary conditions (4.26) and (4.27) re-

spectively in terms of the potential (see the definition in (4.28)).
The boundary conditions in dielectrics lead to some very inter-
esting consequences. For example, let us consider two dielectric media
130 4 Dielectrics
with permittivities ǫ1 and ǫ2 separated by a surface which does not

contain any free surface charge. Let us further assume that an electric
field vector makes an angle θ1 with the surface in the first dielectric
while it makes an angle θ2 with the surface in the second medium as
shown in Fig. 4.6. Denoting the electric fields in the two media by
E1 and E2 respectively, the boundary conditions (4.26) and (4.27)
tell us that
E2
ǫ2
θ2
θ1
ǫ1
E1
Figure 4.6: An electric field in two distinct dielectric media with

permittivities ǫ1 and ǫ2 .
ǫ1 |E1 | sin θ1 = ǫ2 |E2 | sin θ2 , |E1 | cos θ1 = |E2 | cos θ2 . (4.31)
We can take the ratio of the two relations in (4.31) to write
ǫ1 tan θ1 = ǫ2 tan θ2 , (4.32)
which is known as the Snell’s law in optics. We will study this in

more detail in a later chapter.
◮ Example (Point charge in semi-infinite dielectric medium). As an example of

applications of the boundary conditions in dielectric media, let us consider a semi-
infinite medium of a dielectric material of permittivity ǫ1 . Outside of the dielectric
lies vacuum and a point charge of magnitude q is immersed on the boundary
between the two media shown in Fig. 4.7. We wish to determine the fields E and
D everywhere except at the location of the point charge.
Without loss of generality, we can take the origin of the coordinate system
to be at the location of the point charge. Furthermore, from the symmetry of
the problem we expect both the electric and the displacement fields to be radial
and have spherical symmetry in the two regions. Let us denote the fields in the
r
b
Figure 4.7: A point charge on the interface of a dielectric medium

and vacuum with the surface of the dashed sphere representing the
Gaussian surface.
dielectric as E1 , D1 , while we denote them in vacuum as E0 , D0 . The boundary

condition (4.27) for the electric field implies that
E0,t | = E1,t | . (4.33)
On the other hand, since the electric fields are radial, on the boundary surface
they are in fact tangential which determines that
|E0 | = |E1 |. (4.34)
Let us next draw a Gaussian sphere (indicated by the dashed sphere in Fig.
4.7) of radius r around the point charge. From Gauss’ law we obtain
Z
ds · D = 4πq,
or, 2πr 2 (|D1 | + |D0 |) = 4πq,

2q
or, (1 + ǫ1 )|E0 | = ,
r2
2qr̂
or, E0 (r) = , (4.35)
(1 + ǫ1 )r 2
in the upper half plane. Here we have used the fact that the surface area of the
hemisphere in both the regions has the value 2πr 2 as well as the relation between
the displacement field and the electric field (4.12) in the two regions. It follows
now that ((E0 , D0 ) and (E1 , D1 ) are defined respectively in the upper and lower
half planes)
2qr̂
E1 (r) = E0 (r) = ,
(1 + ǫ1 )r 2
2qr̂
D0 (r) = E0 (r) = ,
(1 + ǫ1 )r 2
2ǫ1 qr̂
D1 (r) = ǫ1 E1 (r) = . (4.36)
(1 + ǫ1 )r 2
We note that when ǫ1 = 1, these fields reduce to the electric field for a point
charge in vacuum that we are familiar with. ◭
132 4 Dielectrics
◮ Example (Method of images). As a second application, let us consider two

semi-infinite dielectric media of permittivities ǫ1 and ǫ2 respectively separated
by a plane at z = 0. Let us assume that a point charge q is immersed in the
second dielectric medium (with permittivity ǫ2 ) at a height d from the boundary.
We would like to calculate the electrostatic potential for this system in both the
regions using the method of images.
r
q b
ǫ2
z=0
r′
ǫ1
q′ b
Figure 4.8: A point charge q in the region z > 0 with the image
charge q ′ in the region z < 0 for calculating the electric field in the
upper half plane.
Without loss of generality, we can assume that the point charge lies on
the z-axis at a height z = d from the interface. Unlike the case of a grounded
conducting plane, here we will have nonvanishing electric fields present in both the
dielectric media and, consequently, we need to calculate these in both the regions
z ≥ 0 and z ≤ 0. Let us recall from our earlier study involving the method of
images that we need an image charge in a region where we are not calculating
the electric field. As a result, since we have to calculate the potential and the
fields in both the regions z ≥ 0 and z ≤ 0, we need two sets of image charges -
one for each calculation. When we are calculating the field in the region z ≥ 0,
we need an image charge q ′ located at z = −d as shown in Fig. 4.8. On the
other hand, when we calculate the field in the region z ≤ 0, we need an image
charge in the region z ≥ 0. In fact, if we think for a moment, we realize that the
dielectric will be polarized because of the presence of the charge q. As a result,
as we have discussed in an earlier example, the effective charge seen in the region
z ≤ 0 will be modified. Consequently, in calculating the field in region z ≤ 0, we
can imagine an image charge present at z = d so as to give rise to an effective
charge q ′′ at that point (namely, the image charge has the value q ′′ − q at that
point), as shown in Fig. 4.9.
With these introductory remarks, the calculation is now straightforward.
Let us use cylindrical coordinates for our calculations. From Fig. 4.8, we see that
we can write the potential in the region z ≥ 0 as (see, for example, the discussion
q ′′ b
ǫ2
z=0
r
ǫ1
Figure 4.9: For calculating the electric field in the lower half plane the
image charge can be chosen to lie on top of q leading to an effective
charge q ′′ .
for the electric field in (4.16))

1 q q′
Φ(r; z ≥ 0) = + ′
ǫ2 |r| |r |
!
1 q q′
= p +p . (4.37)
ǫ2 2
ρ + (z − d) 2 ρ + (z + d)2
2
Here ρ represents the radial coordinate on the plane. Similarly, the potential in
the region z ≤ 0 can be determined from Fig. 4.9 to be
1 q ′′ 1 q ′′
Φ(r; z ≤ 0) = = p . (4.38)
ǫ1 |r| ǫ1 ρ2 + (d − z)2
The components of the electric fields can now be calculated easily and have
the forms
" #
∂ 1 q(z − d) q ′ (z + d)
Ez (z ≥ 0) = − Φ(z ≥ 0) = + ,
∂z ǫ2 (ρ2 + (z − d)2 ) 23 3
(ρ2 + (z + d)2 ) 2
∂ 1 q ′′ (z − d)
Ez (z ≤ 0) = − Φ(z ≤ 0) = ,
∂z ǫ1 (ρ2 + (d − z)2 ) 32
" #
∂ 1 qρ q′ ρ
Eρ (z ≥ 0) = − Φ(z ≥ 0) = + ,
∂ρ ǫ2 (ρ2 + (z − d)2 ) 32 3
(ρ2 + (z + d)2 ) 2
∂ 1 q ′′ ρ
Eρ (z ≤ 0) = − Φ(z ≤ 0) = . (4.39)
∂ρ ǫ1 (ρ2 + (d − z)2 ) 32
Since there are no free surfaces charges, the boundary conditions (see (4.26)
134 4 Dielectrics
and (4.27)) would require
ǫ2 Ez (ρ, z = 0+ ) = ǫ1 Ez (ρ, z = 0− ),
d d
or, − 3 (q − q ′ ) = − 3 q ′′ ,
(ρ2 + d2 ) 2 (ρ2 + d2 ) 2
′ ′′
or, (q − q ) = q ,
Eρ (ρ, z = 0+ ) = Eρ (ρ, z = 0− ),
ρ (q + q ′ ) ρ q ′′
or, 3 = 3 ,
(ρ2 + d2 ) 2 ǫ2 (ρ2 + d2 ) 2 ǫ1
or, ǫ1 (q + q ′ ) = ǫ2 q ′′ . (4.40)
The image charges are determined from these relations to be

ǫ1 − ǫ2 2ǫ1
q′ = − q, q ′′ = q. (4.41)
ǫ1 + ǫ2 ǫ1 + ǫ2
We note that only when ǫ1 = ǫ2 , we have q ′′ = q. Otherwise, the effective charge
seen in the region z ≤ 0 will be different due to polarization of the two media.
Once the image charges are determined, the potential as well as the electric fields
can be determined in both the regions. Furthermore, the polarizations in the two
media can also be calculated using (4.17). In fact, since the permittivity of the
two media are different there will be a net polarization charge at the interface,
which can be calculated as follows.
We note from our results (as well as (4.17)) that the polarizations on both
sides of the interface will have the forms
(ǫ2 − 1)
Pz (ρ, z = 0+ ) = Ez (ρ, z = 0+ )
4π
(ǫ2 − 1) 2ǫ1 qd
=− ,
4πǫ2 ǫ1 + ǫ2 (ρ2 + d2 ) 23
(ǫ1 − 1)
Pz (ρ, z = 0− ) = Ez (ρ, z = 0− )
4π
(ǫ1 − 1) 2ǫ1 qd
=− . (4.42)
4πǫ1 ǫ1 + ǫ2 (ρ2 + d2 ) 23
It follows from this that the net density of surface polarization charge is given by
(see (4.4))

σb = n̂ · P = Pz (ρ, z = 0+ ) − Pz (ρ, z = 0− )
(ǫ1 − ǫ2 ) qd
= . (4.43)
ǫ2 (ǫ1 + ǫ2 ) 2π(ρ2 + d2 ) 23
This shows, in particular, that as long as ǫ1 6= ǫ2 , there will be a net polarization

charge at the interface. ◭
◮ Example (Dielectric sphere in uniform electric field). Let us consider a dielec-

tric sphere of radius R and dielectric constant ǫ placed in vacuum in a uniform
electric field along the z-axis. Thus, in the absence of the dielectric sphere we
have
E(r) = E ẑ, (4.44)
leading to the potential
Φ(r) = −Ez + C = −Er cos θ + C, (4.45)
where C is a constant and we are assuming the coordinate origin to coincide with
the center of the dielectric sphere. When we introduce the dielectric sphere, the
sphere will be polarized and would modify the electric field around and within
the sphere. However, asymptotically, the form of the potential in (4.45) would
continue to hold.
To determine the potential in the presence of the dielectric sphere let us note
that outside the dielectric sphere we simply have to solve the Laplace equation in
vacuum which has the form
∇2 Φ> (r) = 0, r ≥ R, (4.46)
and whose solutions can be written, in general, as (see (3.105) and note that the
present problem has azimuthal symmetry leading to m = 0)
∞
X
Φ> (r) = C − Er P1 (cos θ) + Aℓ r −(ℓ+1) Pℓ (cos θ), (4.47)
ℓ=1
where we have kept a linear term in r with the asymptotic condition (4.45) in
mind. (Here we have also used the fact that for Φ> (r) to satisfy the Laplace
equation, we must have A0 = 0.) Inside the dielectric sphere we also have to solve
the Laplace equation (there are no free charges inside the sphere and note also
that within the dielectric sphere ǫ is a constant so that it can be taken out of the
gradient operation)
∇2 Φ< (r) = 0, r ≤ R, (4.48)
whose general solution has the form

∞
X
Φ< (r) = Bℓ r ℓ Pℓ (cos θ). (4.49)
ℓ=0
We note that in writing these solutions, we have used the fact that the region inside
the sphere contains the origin and, consequently, should have regular solutions,
while outside the sphere the potential should fall off except for the asymptotic
behavior required by a constant electric field.
Now matching the solutions in (4.47) and (4.49) across the surface of the
sphere we obtain (see (4.30))
Φ< (r = R) = Φ> (r = R),

∞
X
or, Bℓ Rℓ Pℓ (cos θ) = C − ERP1 (cos θ)
ℓ=0
∞
X
+ Aℓ R−(ℓ+1) Pℓ (cos θ), (4.50)
ℓ=1
136 4 Dielectrics
which determines
B0 = C,
B1 R = −ER + A1 R−2 ,
Bℓ Rℓ = Aℓ R−(ℓ+1) , for ℓ ≥ 2. (4.51)
Similarly, the condition on the normal derivatives in (4.30) at the boundary surface
gives (remember, there are no free charges on the surface and the space outside
the sphere is vacuum)

∂Φ< (r) ∂Φ> (r)
ǫ = ,
∂r r=R ∂r r=R
∞
X
or, ǫℓBℓ R(ℓ−1) Pℓ (cos θ) = − EP1 (cos θ)
ℓ=1
∞
X
− (ℓ + 1)Aℓ R−(ℓ+2) Pℓ (cos θ), (4.52)
ℓ=1
which determines
ǫB1 = −E − 2A1 R−3 ,
ǫℓBℓ R(ℓ−1) = −(ℓ + 1)Aℓ R−(ℓ+2) , for ℓ ≥ 2. (4.53)
From the relations (4.51) and (4.53) it is straightforward to determine that
Aℓ = 0 = Bℓ , for ℓ ≥ 2,
B0 = C,
ǫ−1
A1 = ER3 ,
ǫ+2
3
B1 = − E, (4.54)
ǫ+2
3 !
ǫ−1 R
Φ> (r) = C − E 1− r cos θ,
ǫ+2 r
3
Φ< (r) = C − Er cos θ. (4.55)
ǫ+2
The electric fields can be determined from these and we note, in particular, that
inside the dielectric sphere the electric field is given by (z = r cos θ)
3
E< (r) = −∇Φ< (r) = Eẑ. (4.56)
ǫ+2
Namely, as a result of the external electric field the electric field present inside the
dielectric sphere is uniform. It has a reduced strength, but is along the z-axis like
the asymptotic field. This can be contrasted with the case of a conducting sphere
where there is no field inside the sphere. (Namely, even though the dielectric
is polarized, the polarization is not large enough to completely cancel the field
inside.) This shows that the boundary value problems involving dielectrics are
solved much the same way by imposing appropriate boundary conditions at the
surface separating two dielectric media.
We note here that the problem of a spherical cavity inside an infinite
isotropic dielectric medium in the presence of a uniform electric field can also
be solved exactly in the same manner. In fact, let us note that the solution is
similar except that since the electric field inside a dielectric is reduced compared
to that in vacuum, we can obtain the solution simply by letting ǫ → 1ǫ . (Namely,
the boundary conditions, in this case, lead to an inverted field.)
◭
◮ Example (Cylindrical electret). As a final example, let us analyze the electric

field associated with a cylindrical electret. We note that, in nature, there are
dielectric materials which have a permanent constant polarization such that in
the presence of an external field the polarizations orient themselves and retain
this polarization even when the external field is switched off. Such materials
are known as “electrets” and the associated “retained” polarization is normally
referred to as “ferroelectricity”.
2ℓ
Figure 4.10: A cylindrical electret of radius R and length 2ℓ.
If we have a cylindrical electret of radius R and length 2ℓ along the z-axis

with a uniform polarization P = P ẑ as shown in Fig. 4.10, then by definition (see
(4.4)), we have
ρb = −∇ · P = 0, (4.57)
so that there is no volume charge density (of bound charges) in the cylinder. At
the ends of the cylinder, however, we have a surface density of bound charges
given by
σb = n̂ · P = ±P. (4.58)
Namely, the two ends of the cylinder will have equal, but opposite surface charge
densities (because the normal vectors point in opposite directions at the two end
138 4 Dielectrics
surfaces). Thus, we can think of such an electret as the electrostatic equivalent

of a bar magnet. For purposes of calculating the electric field, therefore, we can
forget about the cylinder and consider the equivalent case of two charged disks
separated by a distance of 2ℓ along the z-axis, with surface charge densities given
by σb = ±P .
We have already calculated the electric field for a single charged disk along
the z-axis (see (2.29)) and it has the form

z
E(z) = −2πσ √ − sgn(z) ẑ, (4.59)
R2 + z 2
where σ denotes the surface charge density. Choosing z = 0 to lie at the mid-
point between the two disks, we see that when two disks carrying opposite charge
densities are present, for |z| ≥ ℓ, we have
" #
z−ℓ z+ℓ
E(z) = −2πP ẑ p −p
R2 + (z − ℓ)2 R2 + (z + ℓ)2
" #
z+ℓ z−ℓ
= 2πP p −p . (4.60)
R2 + (z + ℓ)2 R2 + (z − ℓ)2
Similarly, the electric field along the z-axis in the region between the two
disks is obtained to be (|z| ≤ ℓ)
" ! !#
ℓ−z ℓ+z
E(z) = −2πP ẑ −p +1 − p −1
R2 + (ℓ − z)2 R2 + (ℓ + z)2
" #
z+ℓ z−ℓ
= 2πP −2 + p −p . (4.61)
R2 + (z + ℓ)2 R2 + (z − ℓ)2
We note that as z → ℓ, the two expressions give respectively

2ℓ 2ℓ
E(ℓ) = 2πP √ , E(ℓ) = 2πP −2 + √ . (4.62)
R2 + 4ℓ2 R2 + 4ℓ2
This shows that the electric field is discontinuous across the surface, which is a
consequence of the surface polarization charge density on the disk.
◭
1. Consider a dielectric sphere of radius R and permittivity ǫ

placed in vacuum. A point charge q is located outside the sphere
at a distance r = d > R. Determine the electric field due to
this charge both inside and outside the sphere. (A solution us-
ing the method of “images” can be found in Am. J. Phys. 61
(1993) 39. However, you can solve the Laplace equation with
the appropriate matching condition to determine the solution.)
2. Given a dielectric material and the relation between the po-

larization vector P and the bound volume and surface charge
densities, namely,
ρb (r) = −∇ · P(r), σb (r) = n̂ · P(r),
where n̂ denotes the unit vector normal to the surface at r,

prove the following relationship
Z Z Z
d3 r P(r) = d3 r r ρb (r) + ds r σb (r).
V V S
(Physically, this corresponds to the fact that the total polariza-

tion is equal to the total dipole moment of the system.)
3. Consider a dielectric sphere of radius R, which carries a bound

charge density (in spherical coordinates)
σb = α cos θ, ρb = 0,
where α is a constant. There are no free charges either inside

or outside the sphere. Determine the potential both inside and
outside the sphere.
Chapter 5
Magnetostatics
5.1 Lorentz force
We have so far discussed only the nature of forces experienced by

stationary charges in the presence of electric fields. In addition to
such electrostatic forces a moving charged particle also experiences a
force in the presence of a magnetic field of the form
v q
Fm = q × B = v × B, (5.1)
c c
where v represents the velocity of the moving charge and B is known
as the magnetic induction vector or simply the magnetic field. The
experimentally observed relation (5.1) is known as the Lorentz force
law. Note that the magnetic force experienced by a charged particle
is perpendicular to its velocity (as well as to the magnetic field) so
that
v · Fm = 0, (5.2)
which shows that the magnetic force leads to no work.
There are a couple of things to note from the definition in (5.1).
First, it is clear from the definition (5.1) that the unit of the magnetic
field in the CGS system, which is Gauss, is the same as the unit of
the electric field (compare with (1.8)), namely, stat−Volt
cm . This is, of
course, the rationale behind the CGS system of units which is to
recognize that the electric and the magnetic fields should be treated
on an equal footing. (The unit of B in the MKS system is Weber m2
Weber 4
and 1 m2 = 10 Gauss.) The second thing to note is that even
though the Lorentz force was originally observed as an empirical law
it can also be derived from relativistic invariance which we will discuss
later. The total force on a charged particle in the presence of both
an electric and a magnetic field is, therefore, given by

1
F = Fe + Fm = q E + v × B . (5.3)
c
141
142 5 Magnetostatics
This can be taken as the defining relation for the magnetic field in
the sense that the magnetic field can be determined from the velocity
dependent part of the force experienced by a charged particle.
We have seen that the sources of electric fields are point charges
(or monopole charges or charge distributions). Correspondingly, we
can ask what are the sources of magnetic fields in nature. Exper-
imentally we know that there are no magnetic monopoles (yet) in
nature. The simplest sources of magnetic fields are known as mag-
netic dipoles. In fact, every magnet in nature including our own earth
has two magnetic poles (conventionally called north and south poles)
which cannot be separated into magnetic monopoles no matter how
hard we try. Thus, the center piece (the basic element) in the study
of magnetic phenomena is the magnet or the magnetic dipole. There
are some substances in nature which have permanent magnetic mo-
ments (or are permanent magnets). One can study the properties of
such materials and the effect of the magnetic fields, they produce,
on charged particles. However, there is an alternate mechanism for
producing magnetic fields which is what we will concern ourselves
with. It was observed in a series of experiments by Oersted, Ampere
etc. that a current produces a magnetic field. In fact, one can think
of closed current loops as magnetic dipoles which is the approach we
will take in studying magnetic phenomena.
5.2 Current
Figure 5.1: Charges with a net drift velocity leading to a current.
Charges in motion produce currents. In general, of course,

charges inside a conductor may have random motion with a zero net
drift velocity. However, if due to some external force charges move
with a net drift velocity, then the flow leads to a current as shown in
5.2 Current 143
Fig. 5.1. The current density in a conductor at a given point is sim-

ply defined to be the amount of charge crossing a unit cross-sectional
area per unit time at that coordinate and is expressed as
J = ρ v, (5.4)
where ρ is the volume charge density and v represents the net velocity
of the charge flow. The current density may or may not be uniform.
In either case, the current carried by a conductor (wire) is simply the
total charge flowing across a given cross-section of the conductor per
unit time
Z
I = ds · J. (5.5)
S
For a thin wire, the variation of J over the cross-sectional area may
not be appreciable in which case we can write
I = |J|A, (5.6)
where A represents the area parallel to the direction of the current.

Most of our discussion would involve thin wires.
When there is a current flow, charge is moving from one section
of the conductor to another. Therefore, any cross-sectional area sep-
arates a conductor into two parts and the charge has to pass through
the cross-sectional area to go from one side to the other. If there is
no sudden creation or destruction of charges we expect that charge in
one region must decrease as a result of the flow of current out of the
region through the cross-sectional area. Thus, drawing an arbitrary
volume with the cross-sectional area as the boundary we see that
Z Z
d 3
d r ρ = − ds · J,
dt
V S
Z Z
∂ρ
or, d3 r =− d3 r (∇ · J),
∂t
V V
Z
3 ∂ρ
or, d r +∇·J = 0,
∂t
V
∂ρ
or, + (∇ · J) = 0, (5.7)
∂t
where we have used Gauss’ theorem as well as the fact that the in-
tegral identity must hold for any volume V and, consequently, the
integrand itself must vanish. Equation (5.7) is known as the conti-

nuity equation and is one of the fundamental equations in the study
of electrodynamics. It expresses the fact that electric charge is con-
served which follows from gauge invariance of the theory as we will
discuss later. (Incidentally, as we will see later the continuity equa-
tion is a relativistic equation, namely, it is invariant under Lorentz
transformations.) In the study of magnetostatics we are interested in
steady state currents for which
∂ρ
= 0, or, ∇ · J = 0. (5.8)
∂t
Namely, in such a case the system has reached equilibrium and as
much charge enters a given volume as leaves through a cross-sectional
area. In a later chapter, when we study time dependent phenomena
we will also analyze currents that are not steady state.
5.3 Force on a current due to a magnetic field
Let us next try to understand some of the properties of current car-

rying conductors. Once we have understood the magnetic force expe-
rienced by a moving point charge (see (5.1)) it is not hard to derive
the force experienced by a current in the presence of a magnetic field
since a current represents a collection of charges in motion. Let us
consider an element dℓ of a current carrying conductor (see Fig. 5.2).
We can choose dℓ to lie along the direction of flow of the current so
that dℓ k v.
dℓ
| |
Figure 5.2: An infinitesimal element of a current carrying conductor

in a magnetic field.
5.3 Force on a current due to a magnetic field 145
In this case, the Lorentz force experienced by the charges in the

element dℓ is easily seen to be
ρA|dℓ| 1 I
dF = v × B = dV J × B = dℓ × B, (5.9)
c c c
where the last relation is obtained with the assumption that if the
cross-sectional area of the conductor (wire) is small, then the current
density does not vary appreciably over this area. Integrating (5.9)
we obtain the total force on a current carrying conductor due to a
magnetic field to be of the form
Z I
1 I
F= d3 r J × B = dℓ × B. (5.10)
c c
Note that for a constant B field (uniform) this integral vanishes since
the line integral around a closed loop is zero so that a closed loop of
wire carrying a current does not feel any force in a uniform magnetic
field.
z
r1
r2
y
x
Figure 5.3: Magnetic force between two current carrying loops.
In separate experiments Ampere had discovered that two current

carrying loops exert a magnetic force on each other (see Fig. 5.3). If
the two loops carry steady currents I1 and I2 respectively, then the
force on the first loop due to the second was known to be of the form
I I
I1 I2 dℓ1 × (dℓ2 × (r1 − r2 ))
F1 = 2 . (5.11)
c |r1 − r2 |3
It is clear that we can write (5.11) in the form of a magnetic force
I
I1
F1 = dℓ1 × B(r1 ), (5.12)
c
if we assume that the current in the second loop produces a magnetic

field of the form
I
I2 dℓ2 × (r − r2 )
B(r) = . (5.13)
c |r − r2 |3
Infinitesimally, we can then write the magnetic field produced at the

coordinate r by an element carrying a current I at a point r′ as given
by
I dℓ′ × (r − r′ ) 1 ′ ′
′ J(r ) × (r − r )
dB(r) = = dV . (5.14)
c |r − r′ |3 c |r − r′ |3
The infinitesimal as well as the integrated forms of the magnetic field

(see (5.13)),
I Z
I dℓ′ × (r − r′ ) 1 J(r′ ) × (r − r′ )
B(r) = ′ 3
= d3 r ′ , (5.15)
c |r − r | c |r − r′ |3
are known as the Biot-Savart law and are observed to hold experi-
mentally. Comparing (5.15) with (1.11) we conclude that an electric
current is a source for the magnetic field just as an electric charge is a
source for the electric field. (One should be careful with the infinites-
imal form of the law. Namely, the current element here is assumed to
be a part of a current in a conductor. An independent element which
is not part of a current loop would violate the continuity equation.)
◮ Example (Magnetic field due to a long straight wire). As an application of the
Biot-Savart law let us determine the magnetic field, produced by an infinitely long
straight thin wire carrying a current I, at a perpendicular distance r from the wire.
For simplicity, let us assume that the wire lies along the z-axis as shown in Fig.
5.4 and that the base of the perpendicular from the point of observation on the
z-axis defines the origin.
We know from (5.15) that the magnetic field produced by the current can
be written as
I
I dℓ′ × (r − r′ )
B(r) = . (5.16)
c |r − r′ |3
Furthermore, it is obvious from the geometry that since dℓ′ is along the z-axis, the
magnetic field lines would be along the polar angle φ̂, field lines forming circles
surrounding the wire. This can be seen explicitly from the fact that, if we define
R = r − r′ = r − z ′ ẑ, (5.17)
then (remember that dℓ′ is along the z-axis),
ẑ × R = ẑ × (r − z ′ ẑ) = ẑ × r = r φ̂. (5.18)
The magnitude of the magnetic field can be obtained trivially as

I z
r
r′ θ ′
r
r−
Figure 5.4: The magnetic field due to a current carrying long straight
wire carrying current.
Z∞ Z∞
I dz ′ r 2I dz ′ r
|B| = 3 = 3
c (r 2 + z ′2 ) 2 c (r 2 + z ′2 ) 2
−∞ 0
∞
2Ir 1 z′
= 2I .
= × 2√ (5.19)
c r r 2 + z ′2 0 cr
Thus, using (5.18) and (5.19), we can write the magnetic field produced by the
current as
2I
B(r) = φ̂. (5.20)
cr
Incidentally, it is quite easy to see now that if there are two infinite parallel
wires separated by a distance r and carrying currents I1 and I2 respectively along
the same direction (say the z-axis), then there will be a force acting between the
two. For, we can think of the current I2 as producing a magnetic field which gives
rise to a magnetic force on the wire with current I1 and we can write (see (5.12))
Z
I1
F= dℓ × B. (5.21)
c
The magnetic field is along the polar direction φ̂ and the current (or dℓ) is along
the z-axis. Consequently, the force would be along the radial direction connecting
the two wires and would be attractive. Namely, in cylindrical coordinates
r̂ × φ̂ = ẑ, φ̂ × ẑ = r̂, ẑ × r̂ = φ̂, (5.22)
so that we can write the force (5.21) as
ZL/2
I1 2I2 2I1 I2 L
F = −r̂ dz = − r̂. (5.23)
c cr c2 r
−L/2
Here we have assumed the two wires to be of length L (each) which is to be taken
to L → ∞ at the end. Therefore, we obtain the force per unit length between the
two currents to be
F 2I1 I2
= − 2 r̂. (5.24)
L c r
The force is clearly attractive. However, if we reverse the direction of one of the
currents, then, the direction of the force would reverse as well and leads to the
familiar fact that two parallel currents attract while two anti-parallel currents
repel each other. ◭
◮ Example (Magnetic field due to a circular current loop). The important result
to note from the previous example is that a straight wire carrying a steady current
produces concentric circular magnetic fields around the axis of the wire, whose
strength falls off inversely as the radius of the circle. Let us next analyze the
magnetic field produced by a circular current loop of radius R as shown in Fig.
5.5. We assume that the the current loop is in a plane perpendicular to the z-axis
and that the current moves in a clockwise direction when seen from below. The
magnetic field due to this current at an arbitrary point is difficult to calculate.
Therefore, we will calculate the magnetic field at any point on the z-axis which
has a simpler form.
r
y
φ R
x
Figure 5.5: The magnetic field due to a circular current loop.
Once again, we use Biot-Savart law (5.15) and choosing the center of the
current loop to be the origin of our coordinate system, we have
I
I dℓ × (r − R)
B(r) = . (5.25)
c |r − R|3
The simplest way to evaluate this is to note that (dℓ is orthogonal to R)
dℓ = Rdφ φ̂, R̂ = cos φ x̂ + sin φ ŷ,
φ̂ × (r − R) = φ̂ × (zẑ − RR̂) = z R̂ + Rẑ = z(cos φ x̂ + sin φ ŷ) + Rẑ,

(5.26)
where we have used the usual rules for cross products, namely, φ̂×ẑ = R̂, R̂× φ̂ =
ẑ. It follows now that
Z2π
I R dφ [z(x̂ cos φ + ŷ sin φ) + Rẑ]
B(z) = 3
c (R2 + z 2 ) 2
0
2πIR2
= 3 ẑ. (5.27)
c(R2 + z 2 ) 2
Thus, along the axis of the loop, the magnetic field is completely parallel to
the axis. At the center of the loop (z = 0), we note from (5.27) that the magnetic
field has the value
2πI
B(z = 0) = ẑ, (5.28)
cR
while very far away, namely, when z >> R, we have
2πIR2
B(z >> R) ≈ ẑ, (5.29)
cz 3
which is reminiscent of the electric field due to a dipole. Thus, we suspect that
a circular current somehow produces a magnetic field which has dipole charac-
teristics. We will see this shortly, but let us note that this suggests that the
magnitude of the magnetic dipole moment is proportional to the current times
the area enclosed by the current loop.
Figure 5.6: A long solenoid of radius R with n turns per unit length.
From (5.27), we can also calculate the magnetic field due to a long solenoid
along its axis. Let us consider a solenoid of radius R with n turns per unit length
with the z-axis representing the axis of the solenoid (see Fig. 5.6). In a length
interval dz there will be n dz loops of wire each producing a magnetic field as
derived in (5.27). Consequently, the total magnetic field produced by the solenoid
along the axis is given by
Z∞
2πnIR2 dz
B= ẑ 3
c (R2 + z 2 ) 2
−∞
∞
2πnIR2 1 z

= ẑ × 2 √
c R R2 + z 2 −∞
4πnI
= ẑ. (5.30)
c
Namely, the magnetic field is a constant along the axis of an infinitely long solenoid
determined completely by the current and the number of turns per unit length.
In fact, even though we have not shown this, the magnetic field is really constant
at any point inside the solenoid. Long solenoids are often used to produce and
maintain constant magnetic fields over short distances. ◭
5.4 Nature of the magnetic field
We have already seen that the Biot-Savart law (5.15) expresses the
magnetic field in terms of the current density as
Z
1 J(r′ ) × (r − r′ )
B(r) = d3 r ′ , (5.31)
c |r − r′ |3
which has a form similar to that of the electric field in electrostatics.

In particular, this allows us to think of the current density (current) as
the source of the magnetic field. Let us note that using the identities
we have derived earlier (see (1.45)), we can write
Z
1 3 ′ ′ 1
B(r) = − d r J(r ) × ∇
c |r − r′ |
Z
1 J(r′ )
= ∇× d3 r ′ . (5.32)
c |r − r′ |
This relation is, in fact, quite interesting. In particular, it shows that

since the magnetic field can be written as the curl of a vector its
divergence must vanish. Namely, we have
∇ · B(r) = 0. (5.33)
Comparing this with Gauss’ law in electrostatics, we see that this

is equivalent to saying that there are no magnetic monopoles (or
magnetic monopole charges).
Let us also recall the identity from vector calculus that
∇ × (∇ × A(r)) = ∇(∇ · A) − ∇2 A. (5.34)
Using this as well as (5.32), we now obtain,

Z
1 J(r′ )
∇ × B(r) = ∇ × ∇ × d3 r ′
c |r − r′ |
Z
1 3 ′ ′ 1
= ∇ d r J(r ) · ∇
c |r − r′ |
Z
1 3 ′ ′ 2 1
− d r J(r )∇
c |r − r′ |
5.4 Nature of the magnetic field 151
Z
1 3 ′ ′ ′ 1
= − ∇ d r J(r ) · ∇
c |r − r′ |
Z
4π
+ d3 r ′ J(r′ )δ3 (r − r′ )
c
Z
1 1 4π
= ∇ d3 r ′ (∇′ · J(r′ )) ′
+ J(r)
c |r − r | c
4π
= J(r), (5.35)
c
where we have used the fact that ∇ · J = 0 for steady currents (see
(5.8)). (The surface terms are neglected with the assumption that
we only have localized currents.)
Thus, we see that the two fundamental relations of magneto-
statics are
∇ · B(r) = 0,
4π
∇ × B(r) = J(r), (5.36)
c
which should be compared with the laws of electrostatics in vacuum,
∇ · E(r) = 4πρ(r),
∇ × E(r) = 0. (5.37)
Equations (5.36) describe the differential forms of the laws of mag-

netostatics and with the use of Gauss’ and Stokes’ theorems, we can
obtain the integral representations for them as well. For example, we
can write
Z Z
ds · B(r) = d3 r (∇ · B(r)) = 0,
S V
I Z Z
4π
dℓ · B(r) = ds · (∇ × B(r)) = ds · J(r)
C c
S S
4πI
= . (5.38)
c
The second of these relations (in either the differential or the integral
form) is known as Ampere’s law and says that the line integral of the
magnetic field around any closed loop is proportional to the current
flowing through the cross-sectional area of the loop. It is useful in
calculating magnetic fields for problems with symmetry, much like
the Gauss’ law in calculating electric fields.
◮ Example (Magnetic field of a long straight wire). As an application of Am-

pere’s law, let us calculate the magnetic field due to an infinitely long straight
wire carrying a current I along the z-axis as shown in Fig. 5.7. This problem has
I
z
Figure 5.7: The magnetic field due to a long straight wire carrying
current.
enough symmetry so that to begin with, we know that the magnetic field at any
point would be in the tangential direction to the circle drawn around the axis of
the wire. Furthermore, the magnitude of the magnetic field would be the same at
points (perpendicularly) equidistant from the wire. With this information, let us
draw an “Amperian” loop of radius r around the axis of the wire (clockwise as
seen from below). Then, according to Ampere’s law (5.38), we have
I
4πI
dℓ · B = ,
c
4πI
or, |B(r)| 2πr = ,
c
2I
or, |B(r)| = , (5.39)
cr
and the magnetic field is along the direction of the polar angle. This is the result
we had obtained earlier in (5.20) by explicitly evaluating the integral in the Biot-
Savart law. ◭
5.5 Vector potential
In electrostatics we have noted that the electric field is conservative

(∇ × E = 0) so that we can write it as the (negative) gradient of a
scalar potential which also follows from the Helmholtz theorem. In
magnetostatics, on the other hand, we find that the magnetic field is
not conservative in general. (If there is no current then there is no
magnetic field.) Therefore, we do not expect to be able to write the
magnetic field in terms of a scalar potential in general. On the other
5.5 Vector potential 153
hand, we know that the magnetic field has vanishing divergence (see
(5.36)). Consequently, in this case, we expect that we can write it as
the curl of a vector and, in fact, we have already seen in (5.32) that
this is true, namely,
Z
1 J(r′ ) × (r − r′ )
B(r) = d3 r ′
c |r − r′ |3
Z
1 3 ′ ′ 1
=− d r J(r ) × ∇
c |r − r′ |
Z
1 J(r′ )
=∇× d3 r ′ . (5.40)
c |r − r′ |
Consequently, let us define
B(r) ≡ ∇ × A(r), (5.41)
where A is known as the vector potential (it is a vector) and in the

particular case of (5.40), we readily identify that
Z
1 J(r′ )
A(r) = d3 r ′ . (5.42)
c |r − r′ |
We note that writing B as the curl of a vector automatically

satisfies the vanishing divergence equation in (5.36). It is also clear
that the vector potential that gives rise to a given magnetic field
cannot be unique unless further conditions are specified. Namely,
both
A(r) and A(r) + ∇α(r), (5.43)
where α(r) is an arbitrary scalar function would give rise to the same
magnetic field since the curl of a gradient vanishes. This is the first
manifestation of what we would see later as the gauge invariance
of Maxwell’s equations. For the present, let us simply note that
the vector potential A obtained in (5.42) appears to be unique only
because it also satisfies the condition
Z
1 3 ′ ′ 1
∇ · A(r) = d r J(r ) · ∇
c |r − r′ |
Z
1 1
= d3 r ′ ∇′ · J(r′ ) = 0, (5.44)
c |r − r′ |
which follows from the fact that the currents are steady state cur-
rents in magnetostatics (the surface terms arising from integration
by parts are assumed to vanish for localized currents as in (5.35)).
Such conditions, as we will see later, are called gauge conditions and
are necessary when dealing with a system of equations which has
gauge invariance.
◮ Example (Vector potential of a long straight wire). As an example of calcula-
tions of the vector potential, let us consider again the example of an infinitely
long straight wire carrying a current I along the z-axis as shown in Fig. 5.8. The
vector potential is defined in (5.42) to be
I z
r
r′ θ ′
r
r−
Figure 5.8: Vector potential for a long straight wire carrying current.
Z Z
1 J(r′ ) I dr′
A(r) = d3 r ′ = . (5.45)
c |r − r′ | c |r − r′ |
In the present case, the current is along the z-axis. Consequently, only the z-
component of the vector potential will be nonzero. If we define the perpendicular
distance of a point from the wire as r, then the vector potential takes the form
(here we assume that the point at which the field is being calculated lies in the
z = 0 plane with the origin at the foot of the perpendicular to the wire)
Z∞
I dz ′
Az (r) = √
c r2+ z ′2
−∞
Z∞
2I dz ′
= √
c r2+ z ′2
0
2I p ∞

= log r 2 + z ′2 + z ′
c 0
2I
=− log r + constant, (5.46)
c
where we have used the standard integration formula (see Gradshteyn and Ryzhik
2.261). Note that the constant of integration is a divergent constant, much like the
5.6 Multipole expansion 155
case of the scalar potential for an infinitely long wire carrying charge. As in the
example in electrostatics the constant in (5.46) does not matter in the calculation
of physical fields.
From the form of p the vector potential (5.46), we obtain the magnetic field
to be (in this case, r = x2 + y 2 )
B(r) = ∇ × A(r)
∂Az ∂Az
= x̂ − ŷ
∂y ∂x
2I (−y x̂ + x ŷ) 2I r(− sin φ x̂ + cos φ ŷ)
= =
c r2 c r2
2I
= φ̂. (5.47)
cr
This is, of course, the result we had obtained earlier for the magnetic field by
directly evaluating the integral in the Biot-Savart law (see (5.20)). ◭
5.6 Multipole expansion
Let us consider a small circular current loop with the center at the
origin of the coordinate system (see Fig. 5.9). We note from (5.42)
that we can write the vector potential as
r
r′
r−
r′
Figure 5.9: The vector potential for a small circular current loop.
Z I
1 3 ′ J(r′ ) I dr′
A(r) = d r ′
= , (5.48)
c |r − r | c |r − r′ |
where we are assuming that the wire is thin so that we can assume
the current density to be uniform. We note that, as in the case of
electrostatics (see, for example, (2.40) and discussions there), we can
expand the denominator in (5.48) for r ≫ r ′ , so that the vector
potential can be written in the form (θ denotes the angle between r

and r′ )
∞ I
IX 1
A(r) = Pℓ (cos θ) dr′ r ′ℓ
c r ℓ+1
ℓ=0
I I
I I
= ′
dr + 2 P1 (cos θ) dr′ r ′ + · · ·
cr cr
I
I
= 3 dr′ (r′ · r) + · · · , (5.49)
cr
where the first term vanishes because the line integral around a closed
loop vanishes. Thus, there is no monopole term in the expansion. The
dominant term at large distances, therefore, is the dipole term which
can be simplified by using some of the vector identities. Note that (r
is a fixed vector)
(dr′ × r′ ) × r = −dr′ (r′ · r) + r′ (dr′ · r)
= −2dr′ (r′ · r) + d(r′ (r′ · r)),
1 1
or, dr′ (r′ · r) = d(r′ (r′ · r)) − (dr′ × r′ ) × r, (5.50)
2 2
so that keeping only the dipole term in (5.49) we have (r = |r|)
I I
I ′ ′ I
A(r) ≈ d(r (r · r)) − (dr′ × r′ ) × r
2cr 3 2cr 3
I
I
=− (dr′ × r′ ) × r
2cr 3
m×r
= . (5.51)
r3
Here, the first term in (5.51) vanishes because the integral of a total
derivative around a closed loop is trivial and we have defined
I
I I
m=− dr′ × r′ = S, (5.52)
2c c
where S represents the area enclosed by the current loop (we do not
use the symbol A since it is used for the vector potential). From the
form of A in (5.51) it is suggestive that we can think of m as the
magnetic dipole moment of the current loop which we will show next.
Once we know the vector potential, we can determine the mag-
netic field (for r large) to be
B(r) = ∇ × A(r)
5.7 Magnetization 157

m×r 1
≈∇× 3
= −∇ × m × ∇
r |r|

1 1
= −m ∇2 +∇ m·∇
|r| |r|

m·r m 3(m · r)r
= −∇ 3
=− 3 +
|r| |r| |r|5
3(m · r̂)r̂ − m
= , (5.53)
|r|3
which can be compared with the electric field obtained

earlier in
(2.52) for an electric dipole. (Note that the term ∇2 |r|
1
= −4πδ3 (r)
vanishes because we are considering points far away from the origin.)
It is clear now that m can indeed be thought of as the magnetic
dipole moment of the current loop. This also demonstrates that at
large distances the current loop behaves like a magnetic dipole. Con-
sidering that there are no magnetic monopoles we see that the basic
elements in the study of magnetic phenomena are, therefore, current
loops.
5.7 Magnetization
As in the case of electric fields, we note that different materials in

nature respond differently to an applied magnetic field and based on
their response all materials can be classified broadly into three groups
- diamagnetic, paramagnetic and ferromagnetic materials. The re-
sponse of each of these three kinds of materials can be properly un-
derstood within the context of a quantum theory. However, without
getting into technical details, let us simply note the broad features of
such materials. As we have seen current loops generate a magnetic
dipole moment. Every material, as we know, consists of atoms where
electrons are moving in orbits and every such electron can be thought
of as describing a current loop and, therefore, as producing a mag-
netic dipole moment. In fact, every atom may have several electrons
moving in orbits in a random fashion so that the magnetic moments
within an atom may cancel each other leading to a vanishing net
magnetic dipole moment for the atom. In the presence of an exter-
nal magnetic field, however, these individual magnetic moments can
align themselves and the material can get magnetized. In fact, they
align themselves in such a way as to oppose the external field. This
is the basic behavior of a diamagnetic material. We can, of course,
also have materials where the magnetic moments due to each of the
electron currents in an atom do not quite cancel. In such a case, every
atom in the material may have a net magnetic dipole moment. How-
ever, the magnetic moments of different atoms in the material may be
randomly distributed, leading to a zero net magnetic moment for the
material. In the presence of an external magnetic field, once again,
the magnetic moments would orient themselves. However, in such a
case, the magnetic moments align parallel to the external field and
this is the behavior of paramagnetic materials. There are, of course,
also ferromagnetic materials. Here, the atoms do have a net magnetic
moment like paramagnetic materials. Furthermore, in such materials
the atoms are quite close together and the magnetic moments are
aligned so that inside the material there are domains with large mag-
netic moments which, however, are randomly distributed and can give
rise to a net zero magnetic moment. In the presence of a magnetic
field, however, they all align to give rise to a large magnetic moment
which does not vanish even when the magnetic field is switched off.
(This phenomenon is known as hysteresis.)
Just as in the case of dielectric materials, for a magnetic ma-
terial we can define a magnetic dipole moment M per unit volume
(analogous to polarization, see for example, (2.57) and discussions
there), which is also known as the magnetization of the material.
Experimentally, it is observed, for both diamagnetic and paramag-
netic materials, that in the presence of an external magnetic field the
magnetization is linearly related to the applied magnetic field so that
we can write
M ∝ B. (5.54)
For ferromagnetic materials, on the other hand, there is no such sim-

ple linear relation. In our discussions, we will not be concerned with
ferromagnetic materials at all.
Let us next calculate the magnetic field produced by a magne-
tized material. If M(r) represents the magnetic dipole moment per
unit volume, then, we see from (5.51) that we can write the vector
potential that would be associated with this as
Z Z
M(r′ ) × (r − r′ )
3 ′ 3 ′ ′ 1
A(r) = d r = − d r M(r ) ×∇
|r − r′ |3 |r − r′ |
Z
3 ′ ′ ′ 1
= d r M(r ) × ∇
|r − r′ |
5.7 Magnetization 159
Z Z ′
3 ′ ′ M(r′ ) 3 ′ ∇ × M(r )
′
=− d r ∇ × + d r
|r − r′ | |r − r′ |
Z ′ ′ Z ′
3 ′ ∇ × M(r ) ′ n̂ × M(r )
= d r − ds , (5.55)
|r − r′ | |r − r′ |
V S
where n̂ represents the unit vector normal to the surface and we have
used the identity from vector calculus that for any arbitrary vector
A,
Z Z
3
d r ∇ × A = ds × A. (5.56)
V S
The relation (5.55) is quite interesting, for it says that the vector
potential produced by a magnetized material can be thought of as
due to both a volume current density and a surface current density
of the forms
J(r) = c ∇ × M(r),
Jσ (r) = −c n̂ × M(r), (5.57)
and the surface current density is tangential to the surface.
Figure 5.10: A sketch of the infinitesimal current loops inside a ma-

terial.
The current densities in (5.57) can be compared with the charge

densities (2.61) or (4.4) which describe a polarized dielectric material.
Furthermore, the existence of the volume and surface currents can be
understood in the following way. There are many small current loops
inside a material. If the current loops are not of the same strength,
then inside the material, they would not cancel each other giving rise
to a volume current density, while on the surface, of course, there
will be a current density since there is nothing to cancel this (see Fig.
5.10 which shows a sketch of current loops in a magnetic material).
Let us note that if the small current loops are all uniform, then the
current density would cancel in the interior of the material, giving
rise to a net vanishing volume current density. We also note from
the structure of the surface current density in (5.57) that it vanishes

when integrated over any surface, namely,
Z Z Z
ds·Jσ = −c ds·(n̂×M) = −c ds n̂·(n̂×M) = 0, (5.58)
S S S
which follows from the fact that n̂ is normal to n̂ × M.

The magnetic field produced by the magnetized material can
now be easily obtained from (5.55) to be
B(r) = ∇ × A(r)
Z
3 ′ ′ 1
= −∇ × d r M(r ) × ∇
|r − r′ |
Z
3 ′ ′ 2 1
=− d r M(r ) ∇
|r − r′ |

′ 1
−(M(r ) · ∇)∇
|r − r′ |
Z Z
M(r′ ) · (r − r′ )
= d r M(r ) 4πδ (r − r ) − ∇ d3 r ′
3 ′ ′ 3 ′
|r − r′ |3
Z
M(r′ ) · (r − r′ )
= 4π M(r) − ∇ d3 r ′ . (5.59)
|r − r′ |3
Here, we have used various identities from vector calculus such as
∇ × (A × C) = A(∇ · C) + (C · ∇)A − (A · ∇)C − C(∇ · A),

∇ (A · C) = (A · ∇)C + A × (∇ × C) + (C · ∇)A
+ C × (∇ × A). (5.60)
Thus, we see that, in general, a magnetized material gives rise to a

magnetic field which is a sum of two parts – the first is simply the
magnetization of the material up to a multiplicative factor and the
second is the gradient of a scalar. Let us also note from (5.57) and
(5.59) that
4π
∇ × B = 4π ∇ × M = J, (5.61)
c
where J represents the volume current density due to the current
loops in the magnetized material.
5.8 Magnetic field intensity 161
5.8 Magnetic field intensity
Let us now ask what will be the modifications in the laws of magneto-
statics (5.36) in the presence of a magnetic material. The discussion
is completely parallel to the case of the laws of electrostatics in the
presence of a dielectric material. First, we note that when we have
a magnetic material present we will have two kinds of currents, one
that is maintained by batteries etc. and the other due to the inter-
nal motion of bound electrons inside the magnetic material. Thus,
analogous to the case of the dielectrics, let us refer to them as the
free and the bound currents respectively. In this case, Ampere’s law
would say that the magnetic field integrated around any closed curve
will be related to the total current,
I
4π
dℓ · B = (I + Ib ), (5.62)
C c
where the right hand side represents the total current through the
surface bounded by the curve, with I denoting the free current and
Ib representing the current due to the bound electrons of the magnetic
material completely in analogy with the case of electrostatics in the
presence of a dielectric. We can, of course, write
Z
Ib = ds · Jb , (5.63)
S
where S is the surface bounded by the closed contour C and Jb

represents the current density due to the bound electrons. As we
have seen, this consists of two parts, a volume density and a surface
density. However, we also noted in (5.58) that the surface integral of
the surface current density actually vanishes so that only the volume
current density contributes to the current in (5.63). (The surface
current density lies on the surface and is normal to the direction of
the surface area.) Furthermore, we had identified the volume current
density with c∇ × M (see (5.57)). Consequently, we can write
Z Z I
Ib = ds · Jb = c ds · (∇ × M) = c dℓ · M. (5.64)
C
S S
Using this in (5.62), we see that we can write

I
4π
dℓ · (B − 4π M) = I,
C c
I
4π
or, dℓ · H = I, (5.65)
C c
where we have defined (analogous to the case of electrostatics) a new
field
H = B − 4π M. (5.66)
The new field, H, is known as the magnetic field intensity (also
known simply as the magnetic field) and is the analog of the electric
displacement field. Ampere’s law (5.65), in the presence of a magnetic
material, is written in terms of this field and the right hand side, in
this case, involves only the free currents in a magnetic material. The
differential form of Ampere’s law can now be obtained using Stokes’
theorem and takes the form
4π
∇×H= J, (5.67)
c
where J represents the free current density in the system. Thus, the
laws of magnetostatics, in the presence of a magnetic material, take
the forms
∇ · B = 0,
4π
∇×H= J, (5.68)
c
where the fields are related as
H = B − 4π M, or, B = H + 4π M. (5.69)
Since the magnetization is parallel to B all the vectors B, M,
and H are parallel. Let us define
M = χm H. (5.70)
The constant of proportionality χm is known as the magnetic sus-
ceptibility of the material. For diamagnetic materials it is negative,
while it is positive for paramagnetic materials. Furthermore, for both
diamagnetic and paramagnetic materials, the magnitude of the mag-
netic susceptibility is quite small (of the order of 10−5 − 10−4 ) which
should be compared with the electric susceptibility which is positive
and is much larger in magnitude. Let us also note that with this
definition, we can now write
B = (1 + 4πχm ) H = µ H, (5.71)
where we have identified
µ = 1 + 4πχm . (5.72)
This is known as the permeability of a magnetic material.
5.9 Boundary condition 163
5.9 Boundary condition
As we have seen, the laws of magnetostatics are given by

∇ · B = 0,
4π
∇×H= J, (5.73)
c
where J represents the free current density in the system. All the
reference to any bound currents is now completely contained in the
definition of the new field H. If we have two different magnetic
materials separated by a boundary surface, the magnetic fields in the
two media have to satisfy some boundary conditions on the surface.
These can be easily determined using the equations (5.73). (They
can be derived much the same way as we did for the dielectrics. We
do not repeat the discussion here and simply give the results.)
L R
Figure 5.11: The boundary surface of two magnetic materials with

distinct permeabilities with the dashed curve representing an Ampe-
rian loop.
It is clear from the vanishing of the divergence of B that the nor-

mal component of the B field must be continuous across the bound-
ary. On the other hand, from Ampere’s law in (5.73), we see that
if there are free surface currents across a boundary, the tangential
component of H must be discontinuous. We can write the boundary
conditions explicitly as
BR,n = BL,n ,
4π
HR,t = HL,t + Jσ , (5.74)
c
where Jσ represents the magnitude of the free surface current density.

◮ Example (Uniformly magnetized sphere). With the boundary conditions for
the magnetic fields across a surface determined in (5.74), the boundary value
problems in magnetostatics can be solved much the same way as in electrostatics.
For example, let us consider a sphere of magnetic material of radius R with a
uniform magnetization along the z-axis (see Fig. 5.12) and we are interested in
calculating the magnetic field produced by the sphere both inside and outside the
sphere.
R z
Figure 5.12: A uniformly magnetized sphere of radius R.
Since we have a magnetized sphere with a uniform magnetization along the

z-axis, we can write (we assume the center of the sphere to coincide with the
origin of the coordinate system)
M = M ẑ = M (r̂ cos θ − θ̂ sin θ). (5.75)
Furthermore, since there are no free currents present in either of the regions
(namely, inside and outside the sphere), the equations (5.73) take the forms
∇ · B = 0 = ∇ × H. (5.76)
Since the curl of H vanishes in each of the regions, we can write it as the gradient
of a scalar (magnetic) potential (analogous to the case of electrostatics) of the
form
H< = −∇Φm,< , H> = −∇Φm,> . (5.77)
In each of the regions, we see from the divergence equation that we have to solve
the Laplace equation in terms of the scalar potential (remember that B is parallel
to H).
We have already discussed the solutions of the Laplace equation in detail.
Here we have a system with spherical symmetry. Thus, using spherical coordinates
we note that, since there is no dependence on the azimuthal angle, we can write
the well behaved solutions in the two regions to be (see (3.105))
∞
X
Φm,< (r) = Aℓ r ℓ Pℓ (cos θ),
ℓ=0
5.9 Boundary condition 165
∞
X
Φm,> (r) = Cℓ r −(ℓ+1) Pℓ (cos θ). (5.78)
ℓ=0
We can now match the boundary condition for the tangential components
of the H fields. The tangential component is along the θ direction giving (we can
also equivalently write this as Φm,> |r=R − Φm,< |r=R = 0 as in (4.50))
H>,θ |r=R − H<,θ |r=R = 0,

∞
X dP (cos θ)
ℓ
or, Cℓ R−(ℓ+2) − Aℓ R(ℓ−1) = 0. (5.79)
ℓ=1
dθ
This determines
Cℓ = Aℓ R2l+1 , ℓ ≥ 1. (5.80)
The matching of the normal component of the B field can be done as follows.
We recall the relation (5.69) between the B and the H fields. Since there is no
magnetization outside the sphere, we can write (remember that B = H + 4πM
and that H = −∇Φm )
B>,n |r=R − B<,n |r=R = 0,
or, H>,n |r=R − (H<,n + 4π Mn )|r=R = 0,

∞
X
or, (ℓ + 1)Cℓ R−(ℓ+2) + ℓAℓ R(ℓ−1) Pℓ (cos θ) − 4πM cos θ = 0, (5.81)
ℓ=0
which (note that P0 (cos θ) = 1, P1 (cos θ) = cos θ) determines
(ℓ + 1)Cℓ = −ℓAℓ R2ℓ+1 , ℓ 6= 1,
2C1 = −A1 R3 + 4πM R3 . (5.82)
We can now compare the conditions in (5.80) and (5.82) to determine
Cℓ = 0 = A ℓ , for ℓ ≥ 2,
C0 = 0,
4πM
A1 = C1 R−3 = , (5.83)
3
so that, we can write
3
4πM R
Φm,> (r) = r cos θ,
3 r
4πM
Φm,< (r) = r cos θ. (5.84)
3
Here we have used the fact that even though A0 is an undetermined constant, a
constant term in the potential does not influence the fields and accordingly we
have chosen to set it to zero. It follows from this that
3
4πM R
H> (r) = −∇Φm,> (r) = (2r̂ cos θ + θ̂ sin θ),
3 r
4πM
H< (r) = −∇Φ< (r) = − ẑ. (5.85)
3
It is clear from this that outside the sphere, the magnetic field behaves like that
of a magnetic dipole of moment (recall that 3(ẑ · r̂)r̂ − ẑ = (2r̂ cos θ + θ̂ sin θ))
4πR3 4πR3 3(m · r̂)r̂ − m

m= M= M ẑ, H> (r) = , (5.86)
3 3 r3
as we would expect (see (5.53)). Inside the magnetic material, however, the mag-
netic field is anti-parallel to the magnetization M. ◭
5.10 Faraday’s Law
So far, we have discussed problems in electrostatics and magneto-

statics. As we have seen, these phenomena are decoupled from each
other. Namely, electrostatics is completely described by electric fields
while magnetostatics involves only magnetic fields. As a result, one
can study such phenomena independently. We have also seen that
the work of Ampere, Biot and Savart as well as Oersted showed how
a current can produce a magnetic field. Faraday, on the other hand,
reasoned that the phenomenon may be reversible in the sense that a
magnetic field can possibly produce an electric current as well. Thus,
for example, it may be that a magnet placed inside a circular conduct-
ing loop would cause an electric current to flow in the loop. Thus,
he tried to set up an experiment to study this phenomenon. The
experiment was not successful. Namely, he did not find any steady
current in the loop when a magnet is placed within the loop. On the
other hand, he did observe that as the magnet is brought near the
loop, there is a transient current set up in the loop. This led him to
believe that it is not the magnetic field which is likely to produce a
current, rather it is the change in the magnetic flux through the loop
which may be responsible for setting up a current. Several careful ex-
periments, primarily due to Faraday, led to the conclusion that this
expectation is indeed true and quantitatively the relation describing
this (which is actually due to Maxwell) has the form
I Z
1 ∂
dℓ · E = − ds · B, (5.87)
c ∂t
S
which is known as the Faraday’s law of induction. The presence of c,

the speed of light, in the formula is simply understood on dimensional
grounds as well as from the structure of the Lorentz force that we
have seen earlier.
5.11 Inductance 167
There are several things to note from this relation. First of all,
we have seen in electrostatics that
I
dℓ · E = 0. (5.88)
Namely, the electric field in electrostatics is conservative. However,

if we are dealing with time dependent phenomena, this is no longer
true. In fact, in the presence of other forces (mechanical, chemical
etc.), the electric field need not be conservative and the integral over
a closed contour does not have to vanish. When the contour integral
(in (5.88)) does not vanish, its value is known as the electromotive
force which is responsible for driving currents in the circuit. Thus,
Faraday’s law says that when there is a change in the magnetic flux
through a circuit the effect is to set up an electromotive force (emf)
in the circuit in a direction which will oppose the change in the mag-
netic flux. Of course, the change in the magnetic flux can happen in
several ways. First, the circuit itself may be moving (changing), or
the current (in a different circuit) producing the magnetic field may
be changing with time or even the magnetic field due to a magnet
itself may be changing. If the circuit is fixed and is not changing
with time, we can take the time derivative in Faraday’s law inside
the integral which allows us to write
I Z
1 ∂B
dℓ · E = − ds · ,
c ∂t
S
Z Z
1 ∂B
or, ds · (∇ × E) = − ds · ,
c ∂t
S S
1 ∂B
or, ∇ × E = − . (5.89)
c ∂t
This is the differential form of Faraday’s law and shows that when
we have time dependent phenomena, electric and magnetic fields es-
sentially become coupled.
5.11 Inductance
Let us consider a single circuit carrying a current I as shown in Fig.

5.13. As we have seen earlier, this will set up a magnetic field and,
therefore, a magnetic flux through the area enclosed by the circuit
itself. The flux would, of course, depend on the geometry of the
circuit as well as on the current in the circuit. In fact, the magnetic
field and, therefore, the magnetic flux would be linearly proportional

to the current in the circuit (as we have seen from the Biot-Savart
law (5.15)). If the circuit is fixed and is not changing with time, the
magnetic field and, therefore, the magnetic flux, in such a case, can
change with time only if the current changes with time.
Figure 5.13: A circuit carrying a current I.
If the current changes with time leading to a time dependence

in the magnetic flux, according to Faraday’s law (5.87), this would
produce an emf in the circuit itself leading to a modification of the
current in the circuit. Namely, the circuit would act on itself in the
following way
I Z
1 ∂
emf = dℓ · E = − ds · B
c ∂t
S
 
Z
1 ∂ dI
=−  ds · B
c ∂I dt
S
dI
= −L , (5.90)
dt
Z
1 ∂
L= ds · B. (5.91)
c ∂I
S
Here, we see that the parameter L is a property of the circuit (in-

dependent of the current if the relation between the magnetic flux
and the current is linear) and is known as the self-inductance of the
circuit, which determines how a changing current in a circuit acts on
itself. Similarly, if we had a number of nearby circuits, each produc-
ing a time dependent magnetic flux in one another, that would define
a different parameter M , known as the mutual inductance of the cir-

cuits, which describes the way different circuits act on one another.
Namely, let us assume that there are n current carrying circuits and
let us denote generically the magnetic flux as (Φm denotes the mag-
netic flux and not the scalar potential for H as we had done earlier
in the example)
Z
Φm = ds · B. (5.92)
S
Then, denoting by Φm,i the total magnetic flux through the ith circuit
due to the currents in all the circuits, we have
n
X
Φm,i = Φm,ij . (5.93)
j=1
Here Φm,ij represents the magnetic flux through the ith circuit due
to the current in the jth circuit,
Z
Φm,ij = dsi · Bj . (5.94)
Si
It is clear now from Faraday’s law that the emf in the ith circuit due
to time dependent currents in the circuits takes the form (recall that
Φm,ij ∝ Ij )
1 dΦm,i
(emf)i = −
c dt
n
1 X ∂Φm,ij dIj
=−
c ∂Ij dt
j=1
n
X dIj
=− Mij , (5.95)
dt
j=1
where the components Mij represent parameters characteristic of the

circuits. In particular, Mij with i 6= j, represents the mutual induc-
tance between the circuits i and j. On the other hand, Mii = Li (no
summation), represents the self-inductance of the ith circuit.
1. A proton of velocity 109 cm/sec is projected at right angles to

a uniform magnetic induction field of 103 Gauss.
a) What is the deflection in the path of the particle from a

straight line after it has traversed a distance of 1cm?
b) How long would it take the proton to traverse a 90 degree
arc?
2. We have seen that, for two wires carrying currents I1 and I2

respectively, the force on the first wire due to the current in the
second wire is given by
I I
I1 I2 dℓ1 × (dℓ2 × (r1 − r2 ))
F1 = 2 .
c |r1 − r2 |3
What is the force F2 on the second wire due to the current

in the first wire? Show that these forces satisfy Newton’s law,
namely,
F1 + F2 = 0.
3. Consider two magnetic media with permeabilities µ1 and µ2 ,

separated by a boundary surface without any free current. From
the boundary conditions satisfied by the magnetic fields derive
the “Snell’s law” for the present case.
4. Consider a sphere of radius R and permeability µ, placed in a

magnetic field in vacuum, which is initially uniform along the
z-axis, namely, initially, B = Bẑ. Determine the magnetic field,
in the presence of the sphere, both inside and outside, namely,
for r < R as well as for r > R.
Chapter 6
Maxwell’s equations
6.1 Generalization of Ampere’s law
As we have seen in the last chapter, in the presence of time dependent

currents one of the laws of electrostatics changes and the electric field
is no longer curl free (conservative). Similarly, Maxwell realized that
Ampere’s law must also modify in the presence of time dependent
sources. In fact, there was no experimental basis for proposing a
generalization at the time and yet Maxwell proposed his now famous
modification completely from the theoretical consistency of the set of
equations.
Let us consider the differential form of Ampere’s law (5.65)
which says that
4π
∇×H= J. (6.1)
c
From the fact that the divergence of a curl vanishes, (6.1) leads to
4π
∇ · (∇ × H) = ∇ · J,
c
4π
or, 0 = ∇ · J. (6.2)
c
Of course, this holds true in magnetostatics because we are interested
in steady currents (see (5.8)). However, it is clear immediately from
(6.2) that if we are considering a general, time dependent current
(whose divergence need not vanish), then, Ampere’s law (6.1) cannot
hold in the present form since the left hand side is divergence free by
definition while the right hand side is not in general.
Maxwell proposed to remove this inconsistency as follows. Let
us recall that the continuity equation which describes conservation of
electric charge has the form
∂ρ
+ ∇ · J = 0, (6.3)
∂t
171
172 6 Maxwell’s equations
and must hold for any time dependent sources. Furthermore, Gauss’
law has the general form
∇ · D = 4πρ. (6.4)
Consequently, it is clear that if we generalize Ampere’s law as

4π 1 ∂D 4π 1 ∂D
∇×H= J+ = J+ , (6.5)
c c ∂t c 4π ∂t
then, using (6.3) and (6.4), we obtain

4π 1 ∂(∇ · D)
∇ · (∇ × H) = ∇·J+
c 4π ∂t

4π ∂ρ
= ∇·J+ = 0, (6.6)
c ∂t
for general time dependent sources. In particular, when the charge

density has no time dependence so that the current is steady state,
then (6.5) reduces to Ampere’s law (6.1) which we know to be valid
for magnetostatics.
S1
S2
Figure 6.1: Capacitor plates filled with a dielectric and connected to

an alternating power source.
While the modification in (6.5) is quite clear from the theoretical

consistency of the equation, Maxwell also tried to envision an exper-
imental set up where the necessity for such a term would naturally
arise. Consider a time dependent current charging the plates of a ca-
pacitor (see Fig. 6.1). (For example, we can think of an alternating
current source connected to the capacitor as a different set up.) Fur-
thermore, let us suppose that the space between the capacitor plates
is filled with a dielectric material. Then, from Ampere’s law (6.1) we
6.1 Generalization of Ampere’s law 173
know that
I Z Z
4π
dℓ · H = ds · (∇ × H) = ds · J. (6.7)
c
S S
For the same closed contour, if we choose two different surfaces S1

and S2 (see Fig. 6.1) – one enclosing one of the capacitor plates and
the other without – then, it is clear that we obtain respectively
I Z Z
4π
dℓ · H = ds · (∇ × H) = ds · J = 0,
c
S1 S1
I Z Z
4π 4πI
dℓ · H = ds · (∇ × H) = ds · J = . (6.8)
c c
S2 S2
The first relation follows from the fact that there is no conduction
current within the dielectric. The two relations in (6.8) are, there-
fore, inconsistent. On the other hand, the additional term in (6.5)
will remove this inconsistency since inside the dielectric there is a dis-
placement field (even though there is no conduction current) whose
time rate of change provides the effect of a current when integrated
over a surface, namely, (here we will assume the surface to be closed)
 
Z Z
1 ∂D ∂  1
ds · = ds · D
4π ∂t ∂t 4π
S S
 
Z Z
d  1 3 d
= d x ∇ · D = d3 x ρ
dt 4π dt
V V
dQ
= = ID . (6.9)
dt
We know that a current produces a magnetic field and so, in
keeping with this concept Maxwell identified the new term added to
the right hand side of Ampere’s law with a current (since it con-
tributes to the magnetic field) known as the displacement current,
1 ∂D
JD = . (6.10)
4π ∂t
The simplest way to think of this current is to recall that in the
presence of an applied field (including an alternating one) the charge
centers in a dielectric are displaced leading to the effect of a current.
Such currents are, however, different from the usual conduction cur-
rents that we are used to, since the charges never really leave the
nucleus (or the atom). They are known as displacement currents
since they arise from a displacement of charge centers. (Sometimes,
they are also known as polarization currents. Actually, during the
time of Maxwell, it was believed that space is filled with ether which
acts like a dielectric and Maxwell himself believed that this effect
should arise in ether as well.) Maxwell’s proposal (6.5), of course,
had a purely theoretical origin and the experimental verification of
this would not come until two decades later in the experiments of
Hertz. The main difficulty in the experimental verification lies in the
fact that in a conductor, where we know how to measure a current,
the conduction current is overwhelmingly dominant over the displace-
ment current unless the frequency of the time dependent current is
extremely high. However, Hertz’s experiments clearly demonstrated
the existence of a displacement current in dielectrics and the validity
of Maxwell’s modification of Ampere’s law in (6.5).
Together with this modification, we can write all the laws of elec-
tricity and magnetism for general time dependent fields (and sources)
as
∇ · D = 4πρ,
∇ · B = 0,
1 ∂B
∇×E=− ,
c ∂t
4π 1 ∂D 4π
∇×H= J+ = (J + JD ) . (6.11)
c c ∂t c
These are the fundamental laws of electrodynamics and they are
known as Maxwell’s equations. They hold for both time dependent
as well as time independent fields and sources. As we see, these
are coupled differential equations (incidentally, one can also write
the integral forms for these equations using Gauss’ and Stokes’ the-
orems, but we would not go into this), which can be checked to be
self-consistent and which become decoupled in the static limit. The
equations (6.11), of course, have to be supplemented further by the
continuity equation (6.3) as well as various other equations describing
the effects of the medium, namely,
∂ρ
+ ∇ · J = 0,
∂t
J = σE,
6.2 Plane wave solution 175
(
q E + 1c v × B for point charges,
F= (6.12)
ρE + 1c J × B for continuous distributions.
Here, we note that the second of the supplementary equations in
(6.12) is simply Ohm’s law with σ representing the conductivity of
the conductor, while the third describes the Lorentz force law (for
continuous distributions, the third condition corresponds to the force
density).
6.2 Plane wave solution
Let us next consider Maxwell’s equations in an isotropic and homo-

geneous dielectric medium of infinite extent. In such a case, we can
write
D = ǫE, B = µH, (6.13)
in the entire space. Let us further assume that there are no free
charges or currents present in the medium in which case the set of
four Maxwell’s equations (6.11) takes the form,
∇ · E = 0,
∇ · B = 0,
1 ∂B
∇×E=− ,
c ∂t
ǫµ ∂E
∇×B= . (6.14)
c ∂t
Let us note the vector identity
∇ × (∇ × A) = ∇(∇ · A) − ∇2 A, (6.15)
which holds for any arbitrary vector A. Using this, we obtain
1 ∂(∇ × B)
∇ × (∇ × E) = − ,
c ∂t
ǫµ ∂ 2 E
or, ∇(∇ · E) − ∇2 E = − ,
c2 ∂t2
ǫµ ∂ 2 E
or, ∇2 E − = 0. (6.16)
c2 ∂t2
Similarly, taking the curl of the equation for the magnetic field (the
last equation in (6.14)), we obtain
ǫµ ∂ 2 B
∇2 B − = 0. (6.17)
c2 ∂t2
Equations (6.16) and (6.17) show that both the electric field as
well as the magnetic field satisfy the three dimensional wave equation,
with the velocity (speed) of propagation given by (see (3.48))
c
v=√ . (6.18)
ǫµ
In particular, since ǫ = 1 = µ in vacuum we see that the speed of
propagation of these waves in vacuum coincides with the speed of
light. This was the first evidence that light waves arise because of
time dependent electric and magnetic fields or that light waves are
simply electromagnetic waves.
To further understand the behavior of these waves, let us con-
sider for simplicity a plane wave solution for the electric field. A
plane wave by definition is a wave where the wave variable has the
same phase at any point on the wavefront which is an infinite plane.
Thus, we note that a plane wave solution of the equation involving
the electric field (6.16) will have the form
E(x, t) = E(0) e∓iωt+ik·x , (6.19)
where E(0) is a constant vector, provided the parameters ω and k
satisfy
ω c
= √ = v, (6.20)
|k| ǫµ
which is precisely the relation satisfied by traveling waves. Conven-
tionally, we say that ω = 2πν represents the angular frequency of
the wave, while |k| = 2π λ with λ the wavelength denotes the wave
number. This is seen by noting that the direction of propagation of
the wave is along k and along that direction, points separated by a
distance of λ (the wavelength) are in phase. With this identification,
(6.20) leads to the familiar relation of wave phenomena, namely,
c
νλ = v = √ . (6.21)
ǫµ
As we have noted, we are analyzing plane wave solutions. The
wavefronts (points where the phases are the same), in this case, are
given by
∓ωt + k · x = constant, (6.22)
which, for a fixed time, correspond to planes of infinite extent satis-
fying
k · x = constant. (6.23)
The two independent solutions we have found are characteristic of

the wave equation which is a second order differential equation. The
solution, with the negative sign in the exponent for the first term
in (6.19), represents a forward moving wave (namely, a wave moving
along k) while the one with the positive sign is known as a backward
moving wave. This can be seen as follows. The velocity of propaga-
tion of a wave is precisely the velocity with which wavefronts (namely,
planes with constant phase) move. Thus, defining ξ to represent the
component of x along k (namely, ξ = x · k̂), we have
∓ ωt + |k|ξ = constant,
dξ ω
or, =± = ±v, (6.24)
dt |k|
showing that the velocity of propagation in one case is along k while
it is in the opposite direction in the other case.
Let us emphasize here that the wave solutions that we have con-
structed in (6.19) are known as monochromatic plane waves since they
involve only a single frequency ω (also known as harmonic waves).
A monochromatic plane wave can consist of a linear superposition
of both forward and backward moving waves of the same frequency.
A general solution, on the other hand, would involve a sum (or an
integral) over distinct frequencies as well, in which case it is not a
monochromatic wave. The velocity of propagation for a monochro-
matic wave is known as the phase velocity, while for a wave packet
consisting of distinct frequencies the velocity of propagation is known
as the group velocity. The two velocities can, in general, be different.
Furthermore, let us note that while the four Maxwell’s equations
(6.14) lead to the wave equations (6.16) and (6.17) (for E and B
fields), the two wave equations are not equivalent to the set of four
Maxwell’s equations. (Namely, the solutions of (6.16) and (6.17)
would not automatically satisfy all the equations in (6.14).) There-
fore, a plane wave solution of Maxwell’s equations has to satisfy fur-
ther conditions. Thus, for example, we see from the first equation of
(6.14)
∇ · E = 0, (6.25)
that the plane wave solution for the electric field (6.19) must satisfy
k · E = 0. (6.26)
Namely, the electric field must be orthogonal to the direction of prop-

agation of the wave.
A similar analysis holds for the magnetic fields as well, leading

to the fact that Maxwell’s equations have plane wave solutions for
both E and B fields of the forms
E(x, t) = E(0) e∓iωt+ik·x , B(x, t) = B(0) e∓iωt+ik·x , (6.27)
subject to (6.20) and satisfying
k · E = 0 = k · B. (6.28)
Namely, both the electric and the magnetic fields are perpendicular
to the direction of propagation of the wave which shows that electro-
magnetic waves are transverse waves (unlike sound waves which are
longitudinal). It is also worth emphasizing here that both the electric
and the magnetic fields are real quantities. In writing a solution in
the form (6.27), the assumption is that the electric and the magnetic
fields correspond to either the real or the imaginary parts of the com-
plex solutions which would respectively give cosine or sine solutions.
Furthermore, from the third equation in (6.14)
1 ∂B
∇×E=− , (6.29)
c ∂t
we obtain for the forward moving wave,
ω
k×E= B. (6.30)
c
k
B
Figure 6.2: The electric and the magnetic fields as well as the direc-
tion of propagation as defining an orthogonal system.
In other words, the electric and the magnetic fields are not only
perpendicular to the direction of propagation, but they are orthogonal
to each other as well (see, for example, Fig. 6.2). It is also clear from
(6.20) and (6.30) that if the wave is propagating along the z-axis,
then (B = µH),
r
|Ex | |Ey | ωµ µ µ
= = =√ = , (6.31)
|Hy | |Hx | c|k| ǫµ ǫ
which is a property of the material under consideration. Such rela-

tions are quite important from the point of machine design.
Electromagnetic waves were originally identified with visible light
simply because of the fact that the speed of propagation in vacuum
coincides with the speed of light. However, experiments by Hertz
showed that the electromagnetic waves are much more than just the
visible light. In fact, the visible light forms only a small part of the
spectrum of electromagnetic waves. Radio waves, microwaves etc.
covering a wider range of frequencies are also governed by Maxwell’s
equations. The ratio of the speed of propagation of electromagnetic
waves in vacuum to that in a dielectric medium is known as the index
of refraction of the medium (a term borrowed from optics)
c √
n= = ǫµ, (6.32)
v
which follows from (6.20). As is clear, the index of refraction is com-
pletely determined by the dielectric constant and the permeability of
the medium. This relation is very well tested in radio waves. How-
ever, in the range of optical frequencies, one observes a variation in
the value of the refractive index which is understood as follows. The
dielectric constant of a medium is really not a constant. Rather, it de-
pends on the frequency of the applied field which leads to significant
changes in its value in the range of optical frequencies. Consequently,
the index of refraction of a medium also becomes dependent on fre-
quency, a phenomenon known as dispersion.
6.2.1 Polarization. We note that the plane wave solutions of Maxwell’s

equations (6.27) have two distinct aspects. The exponential charac-
terizes the wave nature of the solution. However, the amplitude is a
vector where information about other aspects of Maxwell’s equations
(such as transversality etc.) are contained. The vector amplitudes
E(0) and B(0) can, of course, contain constant phases. The direc-
tional (vectorial) properties of electromagnetic waves are known as
polarization and are completely determined from the structure of the
electric field alone. (This is because the directional property of the
magnetic field can be determined from a knowledge of the electric
field and the direction of propagation, as is clear from (6.30).) For

example, we know that the electric field is orthogonal to the direction
of propagation of the wave. If we assume that the wave propagates
along the z-axis, then it is clear that the electric field must lie in the
x − y plane, known as the plane of polarization for the wave. Thus,
for a wave propagating along the z-axis, we can write (taking the real
part to denote the electric field)

(0) (0)
E(z, t) = Re E1 x̂ + E2 ŷ e−iωt+ikz . (6.33)
x x x
z z z
E E E
y y y
Figure 6.3: Linear, elliptic and circular polarizations of electromag-

netic waves.
(0)
If we assume that the relative (constant) phase between E1
(0)
and E2 is zero, namely, that
(0) (0) (0) (0)
E1 = |E1 |eiφ , E2 = |E2 |eiφ , (6.34)
where φ is the constant phase of the two amplitudes, then we can

write

(0) (0)
E(z, t) = Re |E1 |x̂ + |E2 |ŷ e−iωt+ikz+iφ

(0) (0)
= |E1 |x̂ + |E2 |ŷ cos(ωt − kz − φ). (6.35)
It is clear from (6.35)

q that the magnitude of the electric field varies
(0) (0)
between zero and |E1 |2 + |E2 |2 with time, but the direction of
(0) (0)
the vector always lies along (|E1 |x̂ + |E2 |ŷ) which is constant.
In this case, we say that the wave is linearly polarized along this
6.3 Boundary conditions 181
direction. On the other hand, if the (constant) relative phase between

the two components of the electric field is arbitrary, then choosing
the phase of the first component to be zero we have

(0) (0)
E(z, t) = Re |E1 |x̂ e−iωt+ikz + |E2 |ŷ e−iωt+ikz+iφ

(0) (0)
= |E1 |x̂ cos(ωt − kz) + |E2 |ŷ cos(ωt − kz − φ) . (6.36)
In this case, it is easy to see that not only the magnitude of the electric
field, but also its direction changes with time tracing out an ellipse
in the x − y plane and we say that the wave is elliptically polarized.
(0) (0)
Finally, if |E1 | = |E2 | and the magnitude of the relative phase
between the two components is π2 , then we can write from (6.33)

(0)
E(z, t) = Re |E1 |(x̂ ± iŷ)e−iωt+ikz
(0)
= |E1 | (x̂ cos(ωt − kz) ± ŷ sin(ωt − kz)) . (6.37)
(0)
In this case, the magnitude of the electric field is a constant (|E1 |),
but the direction changes with time tracing out a circle in the x − y
plane and we say that the wave is circularly polarized. If the rotation
of the electric field is clockwise to an observer facing the incoming
wave, the wave is said to be right circularly polarized. For an op-
posite rotation, the wave is correspondingly known as left circularly
polarized. (Namely, the two terms x̂±iŷ in (6.37) denote respectively
left and right circular polarizations, see Fig. 6.3.) The different po-
larizations are sketched in Fig. 6.3.
6.3 Boundary conditions
One of the great triumphs of Maxwell’s equations is the prediction

of electromagnetic waves which can be identified with light waves in
the appropriate frequency range. It is, therefore, important to check
whether the solutions of Maxwell’s equations lead to familiar phe-
nomena observed in optics such as reflection, refraction etc. To be
able to study such phenomena, we have to first derive the bound-
ary conditions which electric and magnetic fields have to satisfy in
the time dependent case. When we have two distinct dielectric me-
dia separated by a boundary surface, then the boundary conditions
which time dependent electric and magnetic fields would satisfy can
be derived much the same way as we did for the static case. Let us
assume that there are no free charges and currents present on the
boundary surface separating the two media. In such a case, the first
two of Maxwell’s equations in (6.11)
∇ · D = ∇ · (ǫE) = 0 = ∇ · B, (6.38)
hold in the two dielectric media and tell us that the normal compo-
nents of the electric displacement field and the magnetic field must
be continuous across the boundary. Specifically, we have
ǫR En,R = ǫL En,L , Bn,R = Bn,L . (6.39)
The other two equations in (6.11), namely,

1 ∂B
∇×E=− ,
c ∂t

1 1 ∂D
∇×H=∇× B = , (6.40)
µ c ∂t
which are supposed to tell us about the continuity of the tangential

components of the electric and the magnetic fields appear slightly
tricky. However, the boundary conditions for these components can
also be derived in a simple manner.
L R
Figure 6.4: A boundary surface separating two dielectric media.
Let us consider a rectangular closed loop of infinitesimal width

as shown in Fig. 6.4. Then, integrating the first equation in (6.40)
over the area enclosed by this closed loop, we have
Z Z
1 ∂B
ds · (∇ × E) = − ds · ,
c ∂t
S S
I Z
1 ∂
or, dℓ · E = − ds · B. (6.41)
c ∂t
S
In the limit of vanishing width of the rectangular loop, the right hand
side of the equation vanishes because the area enclosed by the loop
does. In the same limit, the left hand side simply gives the difference
in the tangential components of the electric field on the two sides
multiplied by the horizontal length of the curve. Therefore, in this
limit, (6.41) leads to
Et,R = Et,L . (6.42)
In a similar manner, it is straightforward to show that the tan-

gential components of the magnetic fields multiplied by the appro-
priate inverses of the permeabilities are also continuous across the
boundary (namely, it is the tangential components of H which are
continuous across the boundary, Ht,R = Ht,L ),
1 1
Bt,R = Bt,L . (6.43)
µR µL
With these boundary conditions, we are now ready to discuss the
problem of reflection and refraction in a dielectric media.
◮ Example (Normal incidence). As a simple example, let us consider the question
of reflection and refraction for an electromagnetic plane wave incident perpendic-
ularly on the interface of two dielectric media. Thus, let us consider an incident
wave moving along the z-axis, incident on the boundary surface between two
(homogeneous and isotropic) dielectric media of infinite extent. We assume the
boundary surface to be the plane z = 0 as in Fig. 6.5. In the region to the left
(namely, for z ≤ 0), we have a forward moving incident wave as well as a backward
moving reflected wave, while in the region to the right (namely, for z ≥ 0), we
have only a forward moving transmitted wave. Thus, we can write (kL , kR > 0)
EL = Einc + ERefl ,

(0) (0)
Einc = Re E1 e−iωt+ikL z = E1 cos(ωt − kL z),

(0) (0)
ERefl = Re E2 eiωt+ikL z = E2 cos(ωt + kL z),

(0) (0)
ER = Etrans = Re E3 e−iωt+ikR z = E3 cos(ωt − kR z). (6.44)
Here, we are assuming that the vector amplitudes are real for simplicity. Further-
more, for these to represent solutions of Maxwell’s equations in the two media
with dielectric constants and permeabilities (ǫL , µL ) and (ǫR , µR ) respectively, we
must have (see (6.20))
√ √
ω ǫL µL ω ǫR µR
kL = = nL k, kR = = nR k, (6.45)
c c
z
kL kR
z=0
Figure 6.5: An electromagnetic wave at normal incidence on a surface

separating two dielectric media.
where k represents the wave number in vacuum. (Incidentally, the fact that the
waves are all moving along the z-axis follows from the fact that the tangential
components of the fields have to be continuous across the boundary. The copla-
narity of the waves is a general property that we will see in more detail in the
next example where we consider reflection and refraction for oblique incidence.)
Furthermore, from (6.30)
ω
k×E= B, (6.46)
c
as well as (6.45), we obtain
BL = Binc + BRefl ,
√ √ (0)
Binc = ǫL µL ẑ × Einc = ǫL µL ẑ × E1 cos(ωt − kL z),
√ √ (0)
BRefl = − ǫL µL ẑ × Erefl = − ǫL µL ẑ × E2 cos(ωt + kL z),
√ √ (0)
BR = Btrans = ǫR µR ẑ × Etrans = ǫR µR ẑ × E3 cos(ωt − kR z). (6.47)
We note here that the electric and the magnetic fields are in the plane
orthogonal to the direction of propagation which is along the z-axis. Consequently,
there are no components of these fields normal to the boundary plane. Matching
the tangential components of the electric and the magnetic fields at z = 0 (see
(6.42) and (6.43)), we obtain from (6.44) and (6.47)
(0) (0) (0)

E1 + E2 = E3 ,
r rǫ
ǫL (0) (0) R (0)
E1 − E2 = E . (6.48)
µL µR 3
(0) (0) (0)

We can solve for E2 and E3 from the two relations in (6.48) in terms of E1
and they have the forms

q q
ǫL ǫR
µ
− µR
(0) L (0)
E2 = q q E1 ,
ǫL ǫR
µL
+ µR
q
2 µǫLL
(0) (0)
E3 = q q E1 . (6.49)
ǫL ǫR
µL
+ µR
For optically transparent materials, µ ≈ 1, in which case, we can write the

refractive index of the material as (see (6.32))
√ √
n = ǫµ ≈ ǫ. (6.50)
Thus, for two optically transparent materials separated by an interface, (6.49)
leads to
(0) nL − nR (0)
E2 = E ,
nL + nR 1
(0) 2nL (0)
E3 = E . (6.51)
nL + nR 1
◭
◮ Example (Oblique incidence). Let us next consider the case where a plane elec-
tromagnetic wave is incident on a boundary surface separating two dielectric me-
dia at an oblique angle (see Fig. 6.6). We choose the boundary surface to be the
plane z = 0 and assume that the wave is incident at an angle θi (with the z-axis).
Without loss of generality, we can assume the plane of incidence to be the x − z
plane (plane of incidence is defined to be the plane containing the direction of
propagation of the incident wave k̂i and the normal to the boundary surface n̂
which corresponds to ẑ in the present example). Thus, as before, we can write
the incident wave to have the form (see (6.44))
θi
θr z
θt
x
z=0
Figure 6.6: An electromagnetic wave incident on the interface of two

dielectric media at an oblique angle.
(0)
Einc = E1 cos(ωt − ki · x), (6.52)
where, as we have noted, k̂i denotes the direction of propagation of the incident
wave which we have chosen to be in the x − z plane so that it has the form (see
Fig. 6.6)
k̂i = (sin θi x̂ + cos θi ẑ). (6.53)
As we have discussed earlier, there will also be a reflected wave in the first
medium and a transmitted wave in the second, with the forms (see (6.44))
(0)
Erefl = E2 cos(ωt − kr · x),
(0)
Etrans = E3 cos(ωt − kt · x), (6.54)
where k̂r and k̂t denote respectively the directions of propagation for the reflected
as well as the transmitted waves. (The sign of the direction of propagation for
the reflected wave has been absorbed into the definition of k̂r .) The waves need
not a priori be coplanar. However, we know that for them to satisfy Maxwell’s
equations, the magnitudes of the wave vectors must satisfy (see (6.20) and (6.32))
nL ω nR ω
|ki | = |kr | = = nL |k|, |kt | = = nR |k|, (6.55)
c c
where we have defined |k| to represent the wave number in free space.
Given the electric fields, we can also determine the magnetic fields from
(6.30) (or (6.46)) and we have
ki
Binc = × Einc = nL k̂i × Einc ,
|k|
kr
Brefl = × Erefl = nL k̂r × Erefl ,
|k|
kt
Btrans = × Etrans = nR k̂t × Etrans . (6.56)
|k|
With these fields, we can now match the boundary conditions (6.39), (6.42)
and (6.43). Let us note, for example, that the tangential components of the electric
fields have to be continuous across the boundary z = 0, namely,
tang.
(0) (0)
E1 cos(ωt − ki · x) + E2 cos(ωt − kr · x)
z=0
tang.
(0)
= E3 cos(ωt − kt · x) . (6.57)
z=0
Such a condition has two aspects. First, of course, the vector amplitudes have
to satisfy some conditions, but more important is the fact that the phases should
match as well. For example, matching the phases in (6.57) leads to
ki · x|z=0 = kr · x|z=0 = kt · x|z=0 . (6.58)
From the fact that ki lies in the x − z plane, it now follows from (6.58) that all
the three wave vectors must also lie in the x − z plane. This can be seen simply
as follows. The matching condition (6.58) explicitly gives
kix x = krx x + kry y = ktx x + kty y, (6.59)
which leads to the fact that kry = 0 = kty . In other words, all the three plane
waves have to be coplanar (have to lie on the same plane). Therefore, from the
geometry of the problem under study (see Fig. 6.6) we note that the unit vectors
have the forms
k̂i = (sin θi x̂ + cos θi ẑ),
k̂r = (sin θr x̂ − cos θr ẑ),
k̂t = (sin θt x̂ + cos θt ẑ), (6.60)
where θi , θr and θt denote respectively the angles of incidence, reflection and trans-
mission.
In addition to determining that the waves have to be coplanar, the matching
of the phases at the boundary (6.59) also requires that
kix = krx = ktx . (6.61)
Upon using this in (6.55), it now follows that
kix = krx ,
or, nL |k| sin θi = nL |k| sin θr ,

or, θi = θr , (6.62)
and similarly,
kix = ktx ,
or, nL |k| sin θi = nR |k| sin θt ,
sin θi nR
or, = . (6.63)
sin θt nL
Equation (6.62) describes the familiar law (from optics) that the angle of incidence
is equal to the angle of reflection, while (6.63) is the Snell’s law for refraction.
Let us next come to the vector amplitudes. There are two independent cases
to analyze and let us start with the simple case where the electric field for the
incident wave is normal to the plane of incidence, namely, let us assume that it
lies along the y-axis (so that the electric field has no normal component). Then,
it follows from the boundary condition (6.57) that the electric fields of all the
three waves will lie along the y-axis and the boundary condition for the normal
components of the displacement field will hold automatically. In this case, we
have chosen an incident wave polarized along the y-axis. The vector amplitudes
for the magnetic fields can now be easily calculated from (6.56) and (6.60)
(0) (0) (0)
B1 = nL E1 k̂i × ŷ = nL E1 (− cos θi x̂ + sin θi ẑ),
(0) (0) (0)
B2 = nL E2 k̂r × ŷ = nL E2 (cos θi x̂ + sin θi ẑ),
(0) (0) (0)
B3 = nR E3 k̂t × ŷ = nR E3 (− cos θt x̂ + sin θt ẑ). (6.64)
Furthermore, since for most optically transparent material µ ≈ 1, we will use such
a value. In such a case, matching the tangential components of the electric and
the magnetic fields at z = 0, we obtain (the normal components of the magnetic
fields are automatically continuous which can be seen using Snell’s law)
(0) (0) (0)

E1 + E2 = E3 ,

(0) (0) (0)
nL E1 − E2 cos θi = nR E3 cos θt . (6.65)
(0) (0) (0)
Using Snell’s law (6.63), we can now solve for E2 and E3 in terms of E1 and
we obtain
(0)
E2 nL cos θi − nR cos θt cos θi − sin θi cot θt
(0)
= =
E1 nL cos θi + nR cos θt cos θi + sin θi cot θt
tan θt − tan θi
= ,
tan θt + tan θi
(0)
E3 2nL cos θi 2 cos θi
(0)
= =
E1 nL cos θi + nR cos θt cos θi + sin θi cot θt
2 tan θt
= . (6.66)
tan θt + tan θi
The other case to analyze is when the electric field is polarized parallel to
the plane of incidence. Since the fields have to be perpendicular to the direction of
propagation, we can choose (this choice also makes the normal components of the
(0) (0)
displacement field continuous across the boundary, namely, n2L ẑ · (E1 + E2 ) =
(0)
n2R ẑ · E3 )
(0) (0)
E1 = E1 (cos θi x̂ − sin θi ẑ),
(0) (0)
E2 = −E2 (cos θi x̂ + sin θi ẑ),
(0) (0)
E3 = E3 (cos θt x̂ − sin θt ẑ), (6.67)
which will give rise to the magnetic fields (see (6.56))
(0) (0)
B1 = nL E1 ŷ,
(0) (0)
B2 = nL E2 ŷ,
(0) (0)
B3 = nR E3 ŷ. (6.68)
In this case, matching the tangential components of the electric and the magnetic
fields across the boundary we obtain (we are assuming µ ≈ 1)

(0) (0) (0)
E1 − E2 cos θi = E3 cos θt ,

(0) (0) (0)
nL E1 + E2 = nR E3 . (6.69)
(0) (0)
Once again, we can solve for E2 and E3 using Snell’s law and we obtain
(0) sin θi cos θi
E2 nR cos θi − nL cos θt sin θt
− cos θt
(0)
= = sin θi cos θi
E1 nR cos θi + nL cos θt sin θt
+ cos θt
tan(θi − θt )
= ,
tan(θi + θt )
(0)
E3 2nL cos θi 2 cos θi
(0)
= = sin θi cos θi
E1 nR cos θi + nL cos θt sin θt
+ cos θt
2 cos θi sin θt
= . (6.70)
sin(θi + θt ) cos(θi − θt )
It is worth noting from (6.70) that when

π
θi + θt = , (6.71)
2
there is no reflected wave. In such a case, Snell’s law gives
nR sin θi
= = tan θi . (6.72)
nL sin θt
The incident angle for which this holds is known as the Brewster’s angle. In
general, of course, a wave can be decomposed into a sum of waves polarized
parallel and perpendicular to the plane of incidence. What this analysis shows is
that at the Brewster’s angle of incidence, the component (of the field) polarized
parallel to the plane will not be reflected and, consequently, the reflected wave
will be polarized perpendicular to the plane of incidence.
The other important observation from all of the above analysis is that if
nL > nR , then, from Snell’s law (6.63) we obtain
sin θi nR
= ,
sin θt nL
or, θt > θi . (6.73)
π
It follows, therefore, that there is some angle of incidence for which θt = 2
. Let
us call this θint so that we have from Snell’s law
nR
sin θint = . (6.74)
nL
π
Let us note that for θi = θint , we have θt = 2
so that at this angle of incidence
(0)
Etrans = E3 cos(ωt − kt · x)
(0) (0)
= E3 cos(ωt − |kt |(sin θt x + cos θt z)) = E3 cos(ωt − |kt |x).
(6.75)
Namely, there is no z dependence in the transmitted wave. In other words, there

is no transmitted wave in the z ≥ 0 region independent of the polarization of the
wave. Furthermore, if θi > θint , then
s 2 s
p nL sin2 θi
cos θt = 1 − sin2 θt = 1− sin2 θi = 1− , (6.76)
nR sin2 θint
which becomes imaginary. As a result, the transmitted wave becomes exponen-

tially damped with z and the wave propagates only along the x-axis. Such a wave
is conventionally known as a surface wave. ◭
6.4 Energy and the Poynting vector
We have already seen in the case of electrostatics how we can calcu-

late the energy stored in an electric field configuration (see (1.85)).
Let us next ask how we can determine the energy stored in a time
dependent electromagnetic field. In order to do that let us start with
the discussion of the static case. We have seen in electrostatics that
the energy is given by (we are considering a dielectric medium of
permittivity ǫ)
Z Z
3 1
Welec = d x welec = d3 x ρ(x)Φ(x)
2
Z Z
1 3 1
= d x (∇ · D(x))Φ(x) = − d3 x D(x) · ∇Φ(x)
8π 8π
Z Z
1 3 1
= d xD · E = d 3 x ǫ E2 . (6.77)
8π 8π
Here, we have used Gauss’ law as well as integration by parts in the
intermediate steps. Therefore, we can talk of an energy density stored
in the static electric fields as given by
1 ǫ 2
welec = D·E= E . (6.78)
8π 8π
The derivation of the energy density stored in a magnetic field
is a bit more involved. However, we can note the following analogy
between the electrostatic and the magnetostatic cases to determine
the energy density stored in the magnetic fields in a simple manner,
namely,
J(x)
ρ(x) ↔ , Φ(x) ↔ A(x). (6.79)
c
With this analogy, we can intuitively determine the energy stored in
static magnetic fields as
Z Z
1 J(x)
Wmag = d3 x wmag = d3 x · A(x)
2 c
Z
1
= d3 x (∇ × H(x)) · A(x)
8π
Z
1
= d3 x H(x) · (∇ × A(x))
8π
Z Z
1 1
= d3 x H · B = d 3 x µ H2 , (6.80)
8π 8π
6.4 Energy and the Poynting vector 191
so that, we can identify the energy density stored in the static mag-
netic field as
1 1 µ 2
wmag = H·B= B·H= H . (6.81)
8π 8π 8π
Thus, we can write the total energy density stored in the electromag-
netic field, in the static case, as given by
D·E+B·H
w = welec + wmag = . (6.82)
8π
This also continues to be the energy density stored in the electromag-
netic field in the time dependent case which can be seen as follows
and which also brings out the concept of energy conservation within
the context of electromagnetic phenomena.
Let us consider a volume V containing electromagnetic fields as
well as sources (charges and currents). With time the energy stored
in the electromagnetic fields inside the volume would decrease in two
possible ways. First, there may be some dissipation of the energy
density due to conversion into heat or other forms of mechanical en-
ergy. For example, a wire carrying current may heat up due to the
resistance and in the process lose energy. Second, electromagnetic
waves may leave the volume V carrying with them energy. Let us
call the two kinds of energy losses as mechanical and radiation re-
spectively. The mechanical energy loss is easy to calculate directly
from the Lorentz force law. The rate at which the electromagnetic
field does work on a charged particle moving with velocity v is given
by
dWmech
Pmech =
dt
Z Z
3 3 1
= d xf · v = d xρ E + v × B · v
c
V V
Z Z
3
= d x E · (ρv) = d3 x E · J. (6.83)
V
Therefore, this denotes the magnitude of the rate at which energy

stored in the electromagnetic fields is lost to other forms of energy.
This can also be obtained from Ohm’s law where we know that a
wire carrying a current I and maintained at a potential difference E
(the conventional terminology for this is, of course, V which we are
avoiding in order not to have any confusion with the volume V ) loses
energy at the rate
I Z
Pmech = IE = I dℓ · E = d3 x E · J, (6.84)
V
which is the result obtained in (6.83).

The second form of energy loss is determined as follows. Let us
note the vector identity
∇ · (E × H) = −E · (∇ × H) + H · (∇ × E)

4π 1 ∂D 1 ∂B
= −E · J+ +H· −
c c ∂t c ∂t

4π ∂ D·E+B·H
=− E·J+
c ∂t 8π

4π ∂w
=− E·J+ . (6.85)
c ∂t
Thus, defining a vector
c
S= E × H, (6.86)
4π
we see that we can write (6.85) as

∂w
∇·S = − E·J+ ,
∂t
∂w
or, + ∇ · S = −E · J. (6.87)
∂t
This indeed has the structure of a continuity equation describing
conservation of energy (see, for example, (6.3)) with the term on the
right hand side representing the dissipation of energy calculated in
(6.83). This shows that w calculated in (6.82) indeed corresponds to
the energy density for time dependent electromagnetic fields. In fact,
integrating this over the volume V , we obtain
dW
= −(Pmech + Prad ), (6.88)
dt
which shows that the rate at which energy is lost has two parts with
Prad denoting the power carried out by the radiation fields and is
given by
Z Z
3
Prad = d x ∇ · S = ds · S. (6.89)
V S
6.4 Energy and the Poynting vector 193
The vector S is known as the Poynting vector and represents the

density of power carried away by the radiation fields across the surface
bounding the volume V . The “continuity” equation (6.87) indeed
describes the conservation of energy when electromagnetic fields are
involved.
In a similar manner, we can also determine the momentum as-
sociated with the electromagnetic fields. Let us consider a system
of sources (particles, currents etc) as well as electromagnetic (radia-
tion) fields in a given volume V . Let pmech represent the momentum
associated with the charge distribution of the system and prad the mo-
mentum associated with the radiation fields (electromagnetic fields).
Then, from Newton’s law as well as the Lorentz force law, we identify
Z
dpmech 3 1
=F= d x ρE + J × B
dt c
V
Z
1 3 1 ∂D
= d x (∇ · D)E + ∇ × H − ×B
4π c ∂t
V
Z
1 3 1 ∂(D × B)
= d x E(∇ · D) −
4π c ∂t
V

1 ∂B
+D× − B × (∇ × H)
c ∂t
Z
1
= d3 x E(∇ · D) + H(∇ · B) − D × (∇ × E)
4π
V
1 ∂(D × B)
− B × (∇ × H) − , (6.90)
c ∂t
where we have added a term which gives a vanishing contribution

since ∇ · B = 0.
Thus, we can write
dpmech dprad
+
dt dt
Z
1
= d3 x E(∇ · D) + H(∇ · B) − D × (∇ × E)
4π
V

− B × (∇ × H) , (6.91)

Z Z
1 3 ǫµ
prad = d x (D × B) = d3 x (E × H)
4πc 4πc
V V
Z
ǫµ
= d3 x S. (6.92)
c2
V
Furthermore, we can simplify the integrand on the right hand side of

(6.91) as follows. Let us denote the ith component of the integrand
by Ii . The integrand is symmetric in the electric and the magnetic
fields (namely, under E ↔ H, D ↔ B or under E ↔ H, ǫ ↔ µ).
Consequently, let us look at only the terms involving the electric
fields and we have (recall the identity ǫijk ǫkℓm = δiℓ δjm − δim δjℓ ),
1
IiE = (Ei ∂j Dj − ǫijk ǫklm Dj ∂l Em )
4π
1
= (Ei ∂j Dj − Dj ∂i Ej + Dj ∂j Ei )
4π
1
= (∂j (Ei Dj ) − ǫEj ∂i Ej )
4π

ǫ 1 2

= ∂j (Ei Ej ) − ∂i Ej
4π 2

ǫ 1 2
= ∂j Ei Ej − δij Ek . (6.93)
4π 2
Adding the magnetic part as well, we see that the integrand (6.91)
can be written as a total divergence

1 1 2 2
Ii = ∂j ǫEi Ej + µHi Hj − δij (ǫE + µH )
4π 2
= ∂j Tij , (6.94)
where Tij = Tji is known as the stress tensor (spatial components)

for the radiation field (Maxwell field).
Using this result, we see that we can write the conservation
equation for momentum (6.91) in components as
Z Z
dpmech dprad 3
+ = d x ∂j Tij = dsj Tij . (6.95)
dt dt i
V S
This is like a continuity equation and shows that the components

of Tij represent the density of momentum flux through the surface
6.5 Gauge invariance of Maxwell’s equations 195
bounding the volume. In an analogous manner, we can derive the

angular momentum associated with the radiation field as well.
To close this section, let us simply note that we are using com-
plex notation to represent electric and magnetic fields with the un-
derstanding that the fields are real. Similarly, the Poynting vector
c
S= E × H,
4π
is understood to be defined in terms of the real electric and magnetic
fields. However, if we are dealing with harmonic fields, then it is easy
to see that the time averaged value of the Poynting vector can be
represented in terms of complex fields as
c
S= Re (E × H∗ ) , (6.96)
8π
which is quite useful in practical calculations as we will see later.
6.5 Gauge invariance of Maxwell’s equations
The set of four Maxwell’s equations are given by
∇ · D = 4πρ,
∇ · B = 0,
1 ∂B
∇×E=− ,
c ∂t
4π 1 ∂D
∇×H= J+ . (6.97)
c c ∂t
In the static case, we saw that we can write the electric and the mag-
netic fields in terms of scalar and vector potentials and the question
is whether we can continue to do so for time dependent fields.
To analyze this, let us note that the second of Maxwell’s equa-
tions implies that the magnetic field is divergence free. This can, of
course, be solved as in the static case to give
B(x, t) = ∇ × A(x, t). (6.98)
The only difference is that the vector potential, in the present case,
would be a function of time as well. Putting this back into the third
equation of Maxwell, we obtain
1 ∂B 1 ∂(∇ × A)
∇×E=− =− ,
c ∂t c ∂t

1 ∂A
or, ∇× E+ = 0,
c ∂t
1 ∂A
or, E+ = −∇Φ,
c ∂t
1 ∂A(x, t)
or, E(x, t) = −∇Φ(x, t) − . (6.99)
c ∂t
Once again, here the scalar potential Φ depends on space and time.
We note that when Φ, A are independent of time (6.98) and (6.99)
reduce to our earlier discussion of potentials in the static case. As
in the static case, it is clear that the scalar and the vector potentials
have an arbitrariness, namely,
1 ∂Λ(x, t)
Φ′ = Φ + , A′ = A − ∇Λ(x, t), (6.100)
c ∂t
where Λ(x, t) is an arbitrary function give the same electric and mag-
netic fields. Consequently, Maxwell’s equations, written in terms of
the scalar and the vector potentials, will reflect this arbitrariness.
This is known as the gauge invariance of Maxwell’s equations which
we will study in some detail next.
We see that two of Maxwell’s equations (second and third equa-
tions in (6.97)) can be solved to express the electric and the magnetic
fields in terms of scalar and vector potentials. Let us next substitute
the solutions (6.98) and (6.99) into the other two equations. In any
medium, Gauss’ law takes the form
∇ · D = ǫ∇ · E = 4πρ,

1 ∂A
or, ǫ∇ · −∇Φ − = 4πρ,
c ∂t
1 ∂(∇ · A) 4π
or, ∇2 Φ + = − ρ,
c ∂t ǫ

2 ǫµ ∂ 2 Φ 4π 1 ∂ ǫµ ∂Φ
or, ∇ Φ − 2 2
=− ρ− ∇·A+ . (6.101)
c ∂t ǫ c ∂t c ∂t
Similarly, from the last of Maxwell’s equations in (6.97), we obtain
4π 1 ∂D
∇×H= J+ ,
c c ∂t

1 4π ǫ ∂ 1 ∂A
or, ∇ × (∇ × A) = J+ −∇Φ − ,
µ c c ∂t c ∂t
4πµ ǫµ ∂(∇Φ) ǫµ ∂ 2 A
or, ∇(∇ · A) − ∇2 A = J− − ,
c c ∂t c2 ∂t2

2 ǫµ ∂ 2 A 4πµ ǫµ ∂Φ
or, ∇ A − 2 2
=− J+∇ ∇·A+ . (6.102)
c ∂t c c ∂t
The two equations, (6.101) and (6.102), appear to be coupled
second order equations. However, let us recall that the scalar and the
vector potentials are arbitrary up to gauge transformations (6.100).
This is also reflected in the fact that the two coupled equations are
invariant under a gauge transformation. In such a case, the Cauchy
initial value problem cannot be solved uniquely unless we specify
some further conditions on the potentials. Let us, therefore, choose
the scalar and the vector potentials such that
ǫµ ∂Φ
+ ∇ · A = 0. (6.103)
c ∂t
With such a choice of the potentials, the two equations (6.101) and
(6.102) become decoupled and take the forms
ǫµ ∂ 2 Φ 4π
∇2 Φ − = − ρ,
c2 ∂t2 ǫ
ǫµ ∂ 2 A 4πµ
∇2 A − =− J. (6.104)
c2 ∂t2 c
Namely, with such a choice of the potentials, both the scalar and the
vector potentials satisfy the wave equation with sources.
The choice of a form of the potentials is known as a choice of
gauge. And the particular gauge we have chosen in (6.103) is known
as the Lorenz gauge (named after Ludvig Lorenz) which is manifestly
relativistic invariant as we will see. The choice of a gauge is subject
to the condition that it should be implementable, namely, that we
can always find potentials which would satisfy the gauge condition.
For the case at hand, for example, suppose our potentials did not
satisfy the Lorenz gauge condition, namely, if
ǫµ ∂Φ
∇·A+ 6= 0, (6.105)
c ∂t
then, we can make a gauge transformation
1 ∂Λ
Φ′ = Φ + , A′ = A − ∇Λ, (6.106)
c ∂t
requiring that the new potentials would satisfy the gauge condition
ǫµ ∂Φ′
∇ · A′ + = 0. (6.107)
c ∂t
This leads to

ǫµ ∂ 1 ∂Λ
∇ · (A − ∇Λ) + Φ+ = 0,
c ∂t c ∂t

2 ǫµ ∂ 2 ǫµ ∂Φ
or, ∇ − 2 2 Λ=∇·A+ . (6.108)
c ∂t c ∂t
This is an inhomogeneous equation and since the inverse (Green’s

function) of the operator on the left exists, this equation is soluble.
Namely, even if our potentials did not satisfy the Lorenz gauge con-
dition, we can always find a Λ and, therefore, a gauge transformation
such that the new potentials will satisfy the Lorenz gauge condition.
Let us note here that the second order operator on the left is the gen-
eralization of the Laplacian to four dimensions including both space
and time and is known as the D’Alembertian.
The physical results are, of course, independent of the choice
of gauge and one is not forced to choose the Lorenz gauge to study
(solve) Maxwell’s equations. Other gauge choices may be more suit-
able to study specific phenomena. One such gauge choice is known
as the Coulomb gauge or the transverse gauge where the potentials
are required to satisfy
∇ · A = 0. (6.109)
One can see in a straightforward manner as before that this is an

implementable gauge. Furthermore, with such a choice of gauge,
Gauss’ law (6.101) takes the form
4π
∇2 Φ = − ρ, (6.110)
ǫ
which is the Poisson equation. The solution of this, as we have seen
before, can be written in the form
Z
1 ρ(x′ , t)
Φ(x, t) = d 3 x′ . (6.111)
ǫ |x − x′ |
If we now look at the (modified) Ampere’s law (6.102) in this
gauge, we have

2 ǫµ ∂ 2 A 4πµ ǫµ ∂Φ
∇ A− 2 =− J+ ∇
c ∂t2 c c ∂t
Z ∂ρ
4πµ µ
=− J + ∇ d 3 x′ ∂t
c c |x − x′ |
Z
4πµ µ ∇′ · J
=− J− ∇ d 3 x′
c c |x − x′ |
Z
4πµ µ ∇′ · Jl
=− J− ∇ d 3 x′ . (6.112)
c c |x − x′ |
Here, we have used the continuity equation (6.3) as well as the famil-
iar decomposition (also discussed in connection with the Helmholtz
theorem in chapter 1, see (1.53), (1.54) and the discussion there) that
a given vector can be written as a sum of two terms, one longitudinal
and the other transverse with respect to the operation of ∇, namely,
J = Jl + Jt , ∇ · Jt = 0, ∇ × Jl = 0. (6.113)
Furthermore, taking ∇ inside the integral in (6.112) and integrating

by parts, we obtain
Z ′ ′
ǫµ ∂ 2 A 4πµ µ 3 ′ ∇ (∇ · Jl )
∇2 A − 2 = − J − d x
c ∂t2 c c |x − x′ |
Z
4πµ µ ∇′ × (∇′ × Jl ) + ∇′2 Jl
=− J− d 3 x′
c c |x − x′ |
Z
4πµ µ ∇′2 Jl 4πµ 4πµ
=− J− d 3 x′ ′
=− J+ Jl
c c |x − x | c c
4πµ 4πµ
=− (J − Jl ) = − Jt . (6.114)
c c
Here, in the intermediate steps, we have used integration by parts as
well as the fact that

2 1
∇ = −4πδ3 (x − x′ ). (6.115)
|x − x′ |
We have also used the fact that the term ∇ × (∇ × Jl ) vanishes by

definition (6.113).
Thus, with the choice of the Coulomb gauge (6.109), we see that
the scalar potential satisfies the Poisson equation while the vector po-
tential satisfies the wave equation with a transverse current (source),
namely,
4π
∇2 Φ = − ρ,
ǫ
ǫµ ∂ 2 A 4πµ
∇2 A − 2 2
=− Jt . (6.116)
c ∂t c
This is, of course, consistent with the choice of the gauge condition
(6.109). Since the vector potential is transverse in this case, one also
says that this is the transverse gauge (sometimes it is also called the
physical gauge). The Coulomb gauge is quite useful when there are
no sources present, namely, when ρ = 0 = J. In such a case, the
solution of the Poisson equation is trivial and the vector potential
satisfies the free wave equation
ǫµ ∂ 2 A
∇2 A − = 0. (6.117)
c2 ∂t2
It is also worth noting here that even after choosing a gauge,
there is some residual gauge invariance. For example, in the case of
the Lorentz gauge (6.103), even when
ǫµ ∂Φ
+ ∇ · A = 0, (6.118)
c ∂t
holds we can still make a gauge transformation preserving this gauge.
In other words, we can define a new set of potentials
1 ∂ Λ̃
Φ′ = Φ + , A′ = A − ∇Λ̃, (6.119)
c ∂t
which would also satisfy the Lorentz condition provided

2 ǫµ ∂ 2
∇ − 2 2 Λ̃ = 0. (6.120)
c ∂t
Unlike the case of the Laplacian, the D’Alembertian operator allows
for oscillatory solutions to this homogeneous equation.
6.6 Lorentz transformation
To appreciate Lorentz transformations and Lorentz invariance prop-

erly, it is important to look at the chronological development of
events in physics. Newton’s equation was known to be invariant un-
der Galilean transformations or Galilean boosts (in addition to being
covariant under spatial rotations)
x′ = x − vt,
t′ = t, (6.121)
where v represents the constant velocity with which an observer in

a reference frame is moving. The basic idea following from this in-
variance led to the understanding that there are an infinite number
6.6 Lorentz transformation 201
of inertial frames which move with constant velocities with respect

to one another and physical laws are independent of the choice of the
frame. Consequently, while time was assumed to be absolute, veloci-
ties were thought of as relative depending on what is the (Galilean)
inertial frame being used. Most of the laws of physics known at the
time were compatible with invariance under Galilean relativity except
for a couple of phenomena.
One such puzzle came from the plane wave solutions of Maxwell’s
equations which, as we have seen, travel in vacuum with the speed of
light which is a constant. An absolute velocity, however, was against
the spirit of Galilean relativity. Consequently, there was a major con-
flict and to avoid this conflict, Maxwell even assumed that the wave
solutions of his equations travel in a medium called ether, namely,
he tried to promote that the speed with which the waves propagate
is really the speed of light in a specific inertial frame represented by
ether. On the other hand, Michelson and Morley conclusively showed
through their experiments that there is no ether and, therefore, the
speed of light is a constant independent of the inertial frame of ref-
erence. This was indeed the turning point in thinking, for Maxwell’s
equations were not invariant under Galilean transformations. (In
retrospect even a simple experimental measurement such as the life
time of muon, performed decades later, cannot conform to the ideas
of Galilean invariance. Experimentally it is measured that the life
time of the muon decaying at rest in the laboratory is shorter by an
order of magnitude than the life time measured for the muons de-
caying in the cosmic ray showers (τlab ≈ 10−6 sec). This cannot be
explained with Galilean relativity where time is absolute and does
not depend on the choice of the reference frame.)
Until the time of Einstein, it was believed that time is a coordi-

nate very different from space where events take place. Einstein was
the first to propose that space and time should really be considered
on an equal footing and not distinct from each other. Based on the
earlier work of Lorentz, he also proposed that physical laws should
be invariant under Lorentz transformations and not under Galilean
transformations as was thought to be the case until that time. The
Lorentz transformations are very different from the Galilean trans-
formations in that they mix up space and time coordinates. For
example, under a Lorentz transformation the space and time coor-
dinates for an observer in an inertial frame, moving with velocity v
along the x-axis (with respect to another observer), would be given
by

′ β
t =γ t− x ,
c
x′ = γ(x − βct),
y ′ = y,
z ′ = z, (6.122)

v − 1
β = , γ = 1 − β2 2 , γ 2 1 − β 2 = 1. (6.123)
c
It is interesting to note from (6.122) that in the limit c → ∞, Lorentz
transformations reduce to Galilean transformations (6.121). Lorentz
transformations, of course, have far reaching consequences. But, one
of the interesting consequences that can be immediately derived from
the structure of the transformations in (6.122) is that

c2 t′2 − x′2 − y ′2 − z ′2

β 2
= γ c t − x − γ 2 (x − βct)2 − y 2 − z 2
2 2
c

= c2 t2 − x2 − y 2 − z 2 . (6.124)
All of this led Einstein to propose that space and time together
should be thought of as defining a four dimensional manifold where
events take place and that Lorentz transformations are symmetry
transformations which transform the coordinates of this four dimen-
sional manifold, much like rotations transform the coordinates of the
three dimensional space. A vector in such a manifold would consist
of four components – one time and three space – and is called a four
vector, as opposed to the vectors in three dimensional space that we
are all familiar with. However, space and time components can be
embedded into this four vector in two distinct ways. For example,
let us consider the space-time coordinates themselves which define a
four vector. We note that we can define a four component vector as
xµ = (x0 , x) = (ct, x) = (ct, x, y, z), µ = 0, 1, 2, 3. (6.125)
Alternatively, we can define a four vector with a relative negative sign

between the time and the space components, namely,
xµ = (x0 , −x) = (ct, −x) = (ct, −x, −y, −z). (6.126)

It is worth pointing out that on dimensional grounds the time com-

ponent has to be multiplied by a velocity (to be on the same footing
as the space coordinates) and the only velocity that is a constant is
the speed of light. Thus, even though in our discussions, we would
restrict to free space (ǫ = 1 = µ), if one is dealing with a dielectric
medium, the appropriate velocity should be used.
Under a Lorentz transformation, say, along the x-axis it is clear
that the two distinct four vectors would transform as (repeated in-
dices are summed)
x′µ = Λµ ν xν ,
e µ ν xν ,
x′µ = Λ (6.127)
where from the explicit transformations in (6.122), we note that

 
γ −γβ 0 0
 
 −γβ γ 0 0 
µ
Λ ν =  ,
 0 0 1 0 
0 0 0 1
 
γ γβ 0 0
 
 γβ γ 0 0 
e ν
Λµ =   . (6.128)
 0 0 1 0  
0 0 0 1
It is clear from the structure of the matrices in (6.128) that they are
inversely related,
e λ ν )T = δµ .
Λµ ν (Λ (6.129)
λ
The two distinct four vectors, therefore, behave differently under

Lorentz transformations. In fact, they transform in an inverse man-
ner under a Lorentz transformation. The four vector xµ is known as
a contravariant vector while xµ is known as a covariant vector.
From the structure of these transformation matrices, it is also
clear that
eµ ν .
det Λµ ν = 1 = det Λ (6.130)
This is very much like the relation for the rotation matrices in three
dimensional space and suggests that Lorentz transformations can be
thought of as rotations in the four dimensional space-time manifold.
Let us recall that in three dimensional space, a rotation around the

z-axis by an angle θ is defined by
 ′    
x cos θ − sin θ 0 x
 y ′  =  sin θ cos θ 0   y  . (6.131)
z ′ 0 0 1 z
Keeping this in mind, let us define (β, γ are defined in (6.123))
cosh θ = γ, sinh θ = γβ, (6.132)
so that cosh2 θ − sinh2 θ = 1 (alternatively, tanh θ = β). With this,

the Lorentz transformations (namely, Λµ ν and Λ e µ ν ) do actually cor-
respond to rotations, although the angle of rotation appears to be
imaginary. This is a consequence of the structure of the four dimen-
sional manifold that we will discuss. Indeed, just as a rotation around
the z-axis can be thought of as a rotation in the 1 − 2 (x − y) plane,
similarly a Lorentz transformation (boost) along the x-axis can be
thought of as a rotation in the 0 − 1 (t − x) plane. (In fact, the
totality of space rotations and boosts is known as Lorentz transfor-
mations.)
Since rotations leave the length of a vector invariant, here, too,
we can ask what is the invariant length under a Lorentz transforma-
tion. We have already seen in (6.124) that
c2 t2 − x2 = xµ xµ = xµ xµ = x2 , (6.133)
is invariant under Lorentz transformations and defines the invariant

length. However, it is clear now that the reason for this invariance is
that a covariant and a contravariant vector transform inversely under
a Lorentz transformation. From the structure of the invariant length
in (6.133), we can define a metric tensor for the four dimensional
manifold, namely,
x2 = ηµν xµ xν = η µν xµ xν , (6.134)
where the covariant and the contravariant metric tensors are deter-
mined from (6.133) and (6.134) to be
 
1 0 0 0
 
 0 −1 0 0 
ηµν =  ,

 0 0 −1 0 
0 0 0 −1
 
1 0 0 0
 
 0 −1 0 0 
η µν
=
 0 0 −1 0
.
 (6.135)
 
0 0 0 −1
It follows that these matrices are inverses of each other, namely,
η µν ηνλ = δλµ . (6.136)
The metric tensor also allows us to raise and lower the Lorentz
indices,
xµ = η µν xν , xµ = ηµν xν , (6.137)
so that the time component of a vector does not change sign under
raising or lowering while the space components do. Let us note here
that the three dimensional space that we are used to is known as
a Euclidean space where the metric tensor is the trivial Kronecker
delta function δij . Consequently, there is no difference between the
covariant and the contravariant vectors there. The four dimensional
space-time manifold, on the other hand, has a nontrivial metric tensor
giving rise to distinct covariant and contravariant vectors. A manifold
with such a metric (as in (6.135)) is known as a Minkowski space.
Furthermore, from the definition of the invariant length in (6.134),
we see that unlike the three dimensional case, here the length of a
nontrivial (four) vector is not necessarily positive. In fact, it can be
positive, negative, or zero. If
2
x2 = x0 − x2 > 0, (6.138)
we say that the four vector is time-like, while for

2
x2 = x0 − x2 < 0, (6.139)
the vector is called space-like. On the other hand, when

2
x2 = x0 − x2 = 0, (6.140)
the vector is called light-like. It is along such light-like directions

that a light ray travels. (The three types of vectors do not mix
under a Lorentz transformation.) Correspondingly, the structure of
the Minkowski space (see Fig. 6.7) is quite different from the three
dimensional Euclidean space that we are familiar with.
lig
ht ke
-li t-li
ke h
ct lig
time-like
space-like space-like
time-like
Figure 6.7: Four dimensional Minkowski space-time projected onto
two dimensions.
It is now clear that in a Minkowski space we would have two

kinds of vectors, the covariant and the contravariant ones. Thus, if
Aµ and B µ denote two arbitrary (contravariant) vectors in this space,
we can define an inner product (scalar product) between them as
A · B = ηµν Aµ B ν = A0 B 0 − A · B = Aµ B µ = Aµ Bµ , (6.141)
as well as the length of any vector as

2
A2 = ηµν Aµ Aν = A0 − A2 = Aµ Aµ , (6.142)
both of which will be invariant under Lorentz transformations. Any

quantity without a free Lorentz index is a Lorentz scalar and is invari-
ant under Lorentz transformations. We have already seen how vectors
transform under a Lorentz transformation. Any quantity with more
than one free Lorentz index is known as a tensor and its transfor-
mation properties follow from the transformation properties of the
vectors in a straightforward manner. Thus, for example, a third rank
tensor of the form T µν λ (the number of free indices defines the rank
of a tensor) would transform under a Lorentz transformation as
e λ λ′ T µ′ ν ′ λ′ .
T µν λ → T ′ µν λ = Λµ µ′ Λν ν ′ Λ (6.143)
Let us next note that we can define the derivatives (gradients)

on this manifold in a standard manner. There will be two kinds of
derivatives, contragradient and cogradient, defined respectively as

∂ 1∂
∂µ = = , −∇ ,
∂xµ c ∂t

∂ 1∂
∂µ = = , ∇ . (6.144)
∂xµ c ∂t
Note the change in the relative signs between the time and the space
components of the gradient vectors compared with the components
of the coordinate vectors in (6.125) and (6.126). Since gradients
define vectors we can define a scalar (which would be invariant under
Lorentz transformations) from these as
1 ∂2
= ∂ 2 = ηµν ∂ µ ∂ ν = − ∇2 . (6.145)
c2 ∂t2
As we have seen, this is the wave operator and is known as the
D’Alembertian operator (remember that we have restricted ourselves
to free space for which ǫ = 1 = µ). It is invariant under Lorentz trans-
formations just as the Laplacian is invariant under three dimensional
rotations.
There are other familiar quantities from the study of three di-
mensions which also combine into four vectors. (To be able to com-
bine distinct quantities into a four vector, they must have the right
transformation properties under a Lorentz transformation.) Of course,
the most familiar is the fact that energy and momentum combine into
a four vector such that

E E
pµ = ,p , pµ = , −p . (6.146)
c c
The length of this four vector
E2
p2 = ηµν pµ pν = − p2 , (6.147)
c2
is Lorentz invariant and we see from this that we can write the Ein-
stein relation as an invariant relation
E2
p2 = − p2 = m2 c2 ,
c2
or, E 2 = c2 p2 + m2 c4 , (6.148)
where m is known as the rest mass of the particle.

6.7 Covariance of Maxwell’s equations
Just as space-time coordinates as well as energy-momentum combine

into four vectors, similarly, let us note that the charge density and
the current density also combine into a four vector of the form
J µ = (cρ, J), Jµ = (cρ, −J), (6.149)
from which it follows that we can write the continuity equation as

(see the definition of the gradients in (6.144))
∂ρ
∂µ J µ = ∂0 J 0 + ∂i J i = + ∇ · J = 0. (6.150)
∂t
Namely, the continuity equation is nothing other than the vanishing

of the four divergence of the four vector current density. Since the
four divergence is a scalar (it has no free Lorentz index), this equation
is Lorentz invariant. In other words, the form of the continuity equa-
tion is the same in all Lorentz frames. (Incidentally, integrating the
continuity equation over a large volume, we obtain the conservation
of total charge.)
The scalar and the vector potentials also combine into a four
vector known as the (four) vector potential of the form
Aµ = (Φ, A), Aµ = (Φ, −A). (6.151)
Given the vector potential, we can construct a second rank anti-

symmetric tensor by taking its four dimensional curl in the following
way
Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ , µ, ν = 0, 1, 2, 3. (6.152)
Being anti-symmetric, this tensor has only six independent compo-

nents and from the definition of the magnetic and the electric fields
in (6.98) and (6.99) in terms of the scalar and the vector potentials,
we see that they can be expressed as components of this second rank
anti-symmetric tensor in the following way (note that µ = 0, 1, 2, 3
while i, j = 1, 2, 3)
F0i = ∂0 Ai − ∂i A0
1 ∂(A)i
=− − ∇i Φ = (E)i ,
c ∂t
6.7 Covariance of Maxwell’s equations 209
Fij = ∂i Aj − ∂j Ai
= − (∇i (A)j − ∇j (A)i ) = −ǫijk (∇ × A)k
= −ǫijk (B)k . (6.153)
The anti-symmetric tensor Fµν is known as the field strength ten-

sor since its components are none other than the electric and the
magnetic fields.
From the definition of the electric and the magnetic fields in
(6.153), it is now straightforward to determine their transformation
properties under a Lorentz transformation. Thus, with the explicit
form of the Lorentz transformations along the x-axis in (6.128) (or
(6.122)), we obtain
′
Fµν → Fµν e µ µ′ Λ
=Λ e ν ν ′ Fµ′ ν ′ . (6.154)
This gives explicitly

′
F0i e 0 0Λ
=Λ e i j F0j + Λ
e0 j Λ
e i 0 Fj0 + Λ
e0 jΛ
e i k Fjk
e 0 0Λ
= (Λ ei j − Λ
e0 j Λ
e i 0 )F0j + Λ
e0 j Λ
e i k Fjk ,
e i 0Λ
Fij′ = Λ e j k F0k + Λ
ei kΛ
e j 0 Fk0 + Λ
ei kΛ
e j l Fkl
e i 0Λ
= (Λ ej k − Λ
ei kΛ
e j 0 )F0k + Λ
ei kΛ
e j l Fkl . (6.155)
Using the identifications in (6.153), the transformations of the electric

and the magnetic fields then follows
Ex′ = Ex , Bx′ = Bx ,
Ey′ = γ(Ey − βBz ), By′ = γ(By + βEz ), (6.156)
Ez′ = γ(Ez + βBy ), Bz′ = γ(Bz − βEy ).
These are indeed the correct transformations for the electric and the
magnetic fields under a Lorentz transformation and this shows that
the electric and the magnetic fields are really not independent of each
other. (That they should have the same dimension also follows.) A
more physical way to see this is to note that if a charge is at rest, it
produces only an electric field. However, in a different inertial frame
the charge would be moving giving rise to a current, which, as we
have seen, produces a magnetic field.
In terms of the field strength tensors, two of Maxwell’s equations
(the first and the last in (6.97) in vacuum) can be written as
4π ν
∂µ F µν = J . (6.157)
c
This can be checked as follows. Let ν = 0. In this case, the equation

becomes (remember that we are restricting to free space)
4π 0
∂i F i0 = J ,
c
or, ∂i F0i = ∇i (E)i = ∇ · E = 4πρ, (6.158)
which is, of course, the Gauss’ law. On the other hand, if we choose
ν = j, then (6.157) leads to
4π j
∂0 F 0j + ∂i F ij = −∂0 F0j + ∂i Fij = J ,
c
1 ∂(E)j 4π
or, − + ∇i (−ǫijk (B)k ) = (J)j ,
c ∂t c
4π 1 ∂(E)j
or, (∇ × B)j = (J)j + ,
c c ∂t
4π 1 ∂E
or, ∇×B= J+ . (6.159)
c c ∂t
Clearly, these equations are manifestly Lorentz covariant, since they
are expressed in terms of Lorentz covariant quantities. In fact, we
note that the left hand side of (6.157) behaves like a vector under
a Lorentz transformation (namely, has only one free index) and the
same is true for the right hand side as well. Defining the dual of the
field strength tensor as
1
Feµν = ǫµνλρ Fλρ = −Fe νµ , (6.160)
2
where ǫµνλρ is the completely anti-symmetric (Levi-Civita) tensor in
four dimensions with ǫ0123 = 1 (ǫ0ijk = ǫijk ), we recognize that
1 1
Fe0i = ǫ0ijk Fjk = ǫijk (−ǫjkl (B)l ) = −(B)i ,
2 2
Feij = ǫij0k F0k = ǫ0ijk F0k = ǫijk (E)k , (6.161)
namely, the dual interchanges E and B fields (up to sign). In terms

of the dual, we can write the other two Maxwell’s equations as
∂µ Feµν = 0. (6.162)
6.7 Covariance of Maxwell’s equations 211
Note that for ν = 0, (6.162) gives

∂i Fei0 = −∂i Fe0i = 0,
or, ∇i (B)i = ∇ · B = 0. (6.163)
Similarly, choosing ν = j in (6.162), we obtain
∂0 Fe 0j + ∂i Feij = 0,
1 ∂(B)j
or, − + ∇i (ǫijk (E)k ) = 0,
c ∂t
1 ∂(B)j
or, (∇ × E)j = − ,
c ∂t
1 ∂B
or, ∇ × E = − . (6.164)
c ∂t
Of course, solving these conditions leads us to the definitions of the
electric and the magnetic fields in terms of the scalar and the vector
potentials as we have seen before. This demonstrates that the set of
four Maxwell’s equations is manifestly covariant.
Finally, let us note that the gauge invariance of the system is
completely built into the definition of the field strength tensor. In
fact, from the definition
Fµν = ∂µ Aν − ∂ν Aµ , (6.165)
we note that the field strength is invariant under (these are precisely
the same gauge transformations which we have discussed earlier in
(6.100))
Aµ → A′µ = Aµ + ∂µ Λ. (6.166)
Namely, under this redefinition,
Fµν = ∂µ Aν − ∂ν Aµ
→ ∂µ (Aν + ∂ν Λ) − ∂ν (Aµ + ∂µ Λ)
= ∂µ Aν − ∂ν Aµ = Fµν . (6.167)
Furthermore, the Lorenz gauge choice (6.103) can now be seen to be
Lorentz invariant from the fact that (remember that we are consid-
ering free space)
1 ∂Φ
∂µ Aµ = + ∇ · A = 0, (6.168)
c ∂t
is a scalar and, therefore, does not change under a Lorentz transfor-
mation.
6.8 Retarded Green’s function
As we have seen in (6.104), in the Lorenz gauge both the scalar and
the vector potentials satisfy the wave equation with sources, namely,
(for simplicity, we are choosing free space)

1 ∂2 2
Φ = − ∇ Φ = 4πρ,
c2 ∂t2

1 ∂2 2 4π
A = 2 2
−∇ A= J. (6.169)
c ∂t c
Of course, we can combine the two equations in (6.169) into a single
covariant equation of the form (which is consistent with the Lorenz
gauge condition)

1 ∂2 2 4π
Aµ = − ∇ Aµ = Jµ . (6.170)
c2 ∂t2 c
It is clear, therefore, that we can solve for the potentials if we know
the solutions to the equation of the form

1 ∂2 2
Ψ = − ∇ Ψ(x, t) = 4πf (x, t). (6.171)
c2 ∂t2
As we have discussed earlier, a simple way to solve such inho-
mogeneous equations is through the method of Green’s functions.
Namely, let us define the Green’s function G(x, t; x′ , t′ ) for the wave
operator to satisfy the equation,

1 ∂2 2
G = − ∇ G(x, t; x′ , t′ ) = −4π δ4 (x − x′ )
c2 ∂t2
= −4πδ3 (x − x′ )δ(c(t − t′ )). (6.172)
Then, the particular solution of the inhomogeneous equation (6.171)

is easily seen to be (as in (3.145))
Z
Ψ(x, t) = − d4 x′ G(x, x′ )f (x′ )
Z
=− d3 x′ cdt′ G(x, t; x′ , t′ )f (x′ , t′ ). (6.173)
This follows because

Z
Ψ(x) = − d4 x′ G(x, x′ )f (x′ )
6.8 Retarded Green’s function 213
Z
=− d4 x′ (G(x, x′ ))f (x′ )
Z
= 4π d4 x′ δ4 (x − x′ ) f (x′ ) = 4πf (x). (6.174)
Physically, it is clear that the Green’s function represents the solution

of a given equation for a delta function source.
The Green’s function is easily determined by transforming the
equation (6.172) into Fourier space. Let us define the Fourier trans-
forms (we are assuming that the Green’s function is a function of the
difference in the coordinates because of translation invariance)
Z
′
G(x, x ) = d4 k e−ik·(x−x ) G(k),
′
Z
4 ′ 1 ′
δ (x − x ) = d4 k e−ik·(x−x ) , (6.175)
(2π)4
where we have defined a wave number (four) vector kµ = ( ωc , k).

Substituting this into the equation satisfied by the Green’s function
(6.172), we obtain
4π 1
k 2 G(k) = = 3,
(2π)4 4π
1 1 1 1
or, G(k) = = 3 ω2 . (6.176)
4π 3 k 2 4π ( 2 − k2 )
c
This shows the usefulness of the method of Fourier transforms. Basi-

cally, the Fourier transform converts a (partial) differential equation
into an algebraic equation which is much easier to solve.
We can now determine the Green’s function in the coordinate
space by substituting (6.176) into the definition of the Fourier trans-
formation in (6.175), namely,
Z
′
G(x, x′ ) = d4 k e−ik·(x−x ) G(k)
Z ′ ′
1 e−iω(t−t )+ik·(x−x )
= 3 d3 kdω ω2
. (6.177)
4π c − k2
c2
The integrand in (6.177) has poles on the real axis at ω = ±c|k|

and in order to evaluate the integral, we have to specify the contour
of integration in the complex energy plane. Specifying the contour
is equivalent to specifying the boundary condition for the Green’s
function (solution). There are various possible choices of the contour,

just as there are several possible boundary conditions that one can
impose on the solutions. However, in classical electrodynamics (or
for that matter in classical physics), the boundary condition that is
most commonly used is known as the retarded boundary condition.
Since the Green’s function represents the solution at a point (x, t)
due to a delta function source (disturbance) at (x′ , t′ ), on grounds of
causality we require that
G(x, t; x′ , t′ ) = 0, for t < t′ . (6.178)
In other words, the cause and the effect are related in a retarded
manner, namely, the effect cannot precede the cause.
Im ω
b b
Re ω
−ck − iǫ ck − iǫ
Figure 6.8: Shifted poles in the complex energy plane (with k = |k|)
for the retarded Green’s function.
The choice of the contour which incorporates the retarded bound-

ary condition is the one which pushes both the poles infinitesimally
below the real axis as shown in Fig. 6.8. In other words, let us define
Z ′ ′
(R) ′ ′ 1 e−iω(t−t )+ik·(x−x )
G (x, t; x , t ) = lim 3 d3 kdω (ω+iǫ)2
. (6.179)
ǫ→0 4π c − k 2
c 2
In such a case, the poles of the integrand lie at
ω = ±c|k| − iǫ = ±ck − iǫ, (6.180)
both of which are in the lower half of the complex ω plane as in Fig.
6.8. Here we have defined k = |k| for simplicity.
To see that this indeed satisfies the retarded boundary condition,
let us evaluate the ω integral in (6.179) for t < t′ . In such a case, the
exponential in the integrand will be damped on the semi-circle only if

the contour is closed in the upper half plane (see Fig. 6.9). However,
in such a case, there is no pole inside the contour and, consequently,
the integral vanishes by the residue theorem giving
Im ω
b b
Re ω
Figure 6.9: Choice of contour in the complex ω-plane for t − t′ < 0.
G(R) (x, t; x′ , t′ ) = 0, for t < t′ , (6.181)
which indeed defines the retarded Green’s function.

Im ω
b b
Re ω
Figure 6.10: Choice of contour in the complex ω-plane for t − t′ > 0.
On the other hand, when t > t′ , the exponential will be damped

if we close the contour in the lower half plane as in Fig. 6.10. In
this case, the contour will enclose both the poles and using Cauchy’s
method of residues, we obtain (for t > t′ )
G(R) (x, t; x′ , t′ )
Z " ′ ′ ′ ′
#
i 3 e−ikc(t−t )+ik·(x−x ) eikc(t−t )+ik·(x−x )
=− d k − .
(2π)2 k k
(6.182)
The remaining k integrals are easily done in spherical coordinates. If

we define
R = x − x′ , R = |R|, (6.183)
we can write
Z h
i ′
G(R) (x, t; x′ , t′ ) = − kdkd(cos θ)dφ e−ikc(t−t )+ikR cos θ
(2π)2
′
i
−eikc(t−t )+ikR cos θ
Z∞ h
1 ′ ′
=− dk e−ik(c(t−t )−R) − e−ik(c(t−t )+R)
2πR
0
′ ′
i
−eik(c(t−t )+R) + eik(c(t−t )−R)
Z∞ h i
1 ′ ′
=− dk e−ik(c(t−t )−R) − e−ik(c(t−t )+R)
2πR
−∞
1
=− δ(c(t − t′ ) − R) − δ(c(t − t′ ) + R) . (6.184)
R
It is clear that since t − t′ > 0 (and note that R > 0), the second
delta function does not contribute. Thus, we determine the retarded
Green’s function of the wave equation to be (for t > t′ , a condition
which can be implemented through a step function)
δ(c(t − t′ ) − |x − x′ |)
G(R) (x, t; x′ , t′ ) = − . (6.185)
|x − x′ |
With this, we can now obtain a particular solution of the wave
equation (6.171) satisfying the retarded boundary condition as (see
(6.173))
Z
Ψ(x, t) = − d4 x′ G(R) (x, x′ )f (x′ )
Z
δ(c(t − t′ ) − |x − x′ |)f (x′ , t′ )
= d3 x′ cdt′
|x − x′ |
Z
f (x′ , t′ )
= d 3 x′ . (6.186)
|x − x′ | t′ =t− |x−x′ |
c
It is clear that this gives a nontrivial solution only for later times,
′|
t = t′ + |x−x c (namely, a retarded solution). Furthermore, from
this we can now write down the retarded solutions for the Maxwell’s
equations (in the Lorenz gauge) (6.170) to be
Z ′ ′

1 3 ′ Jµ (x , t )
Aµ (x, t) = d x , (6.187)
c |x − x′ | t′ =t− |x−x′ |
c
which reduces, in the static limit, to the solutions we have determined

earlier.
◮ Example (Lienard-Wiechert potential). As a simple example of the application
of the retarded Green’s function, let us calculate the vector potential associated
with the fields produced by a slowly moving charged particle. (We will study this
problem in more detail in a later chapter.) We have seen in (6.187) that we can
write the particular solution for the vector potential, in the Lorenz gauge, as
Z
1 Jµ (x′ , t′ )
Aµ (x, t) = d3 x′
c |x − x′ | t′ =t− |x−x′ |
c

Z ′ ′ ′ |x−x′ |
1 Jµ (x , t )δ t − t + c
= dt′ d3 x′ . (6.188)
c |x − x′ |
Let us now consider a point particle with charge q moving along a trajectory
ξ(t) so that we can write
Jµ (x, t) = jµ (t)δ 3 (x − ξ(t)), (6.189)
where,

dξ(t)
j µ (t) = (cq, qv) = cq, q . (6.190)
dt
Given this, we can compute the potential that such a moving charge would produce
using (6.188). Namely,
′

Z jµ (t′ )δ 3 (x′ − ξ(t′ ))δ t′ − t + |x−x |
1 ′ 3 ′ c
Aµ (x, t) = dt d x
c |x − x′ |

jµ (t′ )δ t′ − t + |x−ξc(t )|
′
Z
1 ′
= dt . (6.191)
c |x − ξ(t′ )|
Let us next define
|x − ξ(t′ )|
τ = t′ − t + . (6.192)
c
Then, the integration over t′ in (6.191) would determine the time coordinate t′ to
be the one for which the argument of the delta function vanishes, namely,
|x − ξ(t′ )|
τ = t′ − t + = 0. (6.193)
c
This can be solved for t′ once we know the trajectory of the particle. Furthermore,
dτ 1 d|x − ξ(t′ )|
′
=1+ , (6.194)
dt c dt′
and using the standard formula for integration with a delta function, namely,
Z
1
dx δ(f (x)) g(x) = df (x) g(x0 ), (6.195)
| dx |x0
where x0 represents the solution of f (x) = 0 (we assume here that there is only
one root x0 of the equation), we obtain

Z ′ ′ |x−ξ (t′ )|
1 jµ (t )δ t − t + c
Aµ (x, t) = dt′
c |x − ξ(t′ )|

1 jµ (t )′

=
c |x − ξ(t′ )| 1 + 1 d|x−ξ(t )| ′
c dt′
τ =0

1 jµ (t′ )

= . (6.196)
c |x − ξ(t′ )| + 1 d|x−ξ(t′ )|2
2c dt′ τ =0
These are known as the Lienard-Wiechert potentials. (We note here that for slow
moving particles it is not necessary to have the magnitude in the Jacobian coming
from the delta function.) From these, we can easily determine the electric and the
magnetic fields that a moving charged particle produces which, in turn, are used
in the study of radiation due to a moving charged particle. We will discuss this
in more detail in a later chapter. ◭
6.9 Kirchhoff’s representation
The Kirchhoff representation is a mathematical statement of Huy-

gen’s principle and is the starting point for the discussion of the the-
ory of diffraction. It really follows from a generalization of Green’s
identity (3.67) to the case where both time and space are involved.
Let us note that we have so far determined the particular solution of
the wave equation subject to the retarded boundary condition. How-
ever, the solution of any differential equation, as we know, consists
of a sum of a homogeneous part and the particular integral. In the
static case, we saw that the homogeneous solution was the one which
allowed us to impose the appropriate Dirichlet boundary condition
on the solution. The wave equation, on the other hand, is a hyper-
bolic equation. In such a case, as we have seen, a unique solution
6.9 Kirchhoff’s representation 219
is obtained by imposing Cauchy’s initial values, namely, by provid-

ing the solution as well as its first derivative with respect to time at
t = 0. As in the static case, the homogeneous solution allows us to
find the solution subject to the given initial value conditions and the
generalization of Green’s identity to four dimensions is quite crucial
in showing this.
Let us consider two arbitrary functions of space and time which
we denote by Ψ and Φ. (Here Φ is not necessarily the scalar poten-
tial.) It follows now that
Z Ztf Z
4 ′ ′ ′
′

d x Φ Ψ − Ψ Φ = cdt d3 x′ Φ∂ ′2 Ψ − Ψ∂ ′2 Φ
ti
Ztf Z
′

= cdt d3 x′ ∂µ′ Φ∂ ′µ Ψ − Ψ∂ ′µ Φ
ti
Z

= ds′µ Φ∂ ′µ Ψ − Ψ∂ ′µ Φ
Z t′ =tf
1 ∂Ψ
3 ′ ∂Φ
= d x Φ ′ −Ψ ′
c ∂t ∂t t′ =ti
Ztf Z

−c dt′ ds′ · Φ∇′ Ψ − Ψ∇′ Φ . (6.197)
ti
Here ds′ denotes integration over the two dimensional surface bound-
ing the three dimensional volume. This is the generalization of Green’s
identity to the four dimensional case for any two arbitrary functions
Ψ and Φ.
Let us now specialize to the case where
Φ = G(x, t; x′ , t′ ), Ψ = Ψ(x′ , t′ ), (6.198)
such that
Ψ(x) = 4πf (x),

G(x, x′ ) = −4πδ4 (x − x′ ). (6.199)
Substituting this into the identity (6.197) and assuming that ti ≤ t ≤

tf , we obtain,
Z
Ψ(x, t) = − d4 x′ G(x, x′ )f (x′ )
Z t′ =tf
1 ∂Ψ
3 ′ ∂G
+ d x G ′ −Ψ ′
4πc ∂t ∂t t′ =ti
Ztf Z
c ′

− dt ds′ · G∇′ Ψ(x′ , t′ ) − Ψ(x′ , t′ )∇′ G . (6.200)
4π
ti
If we now restrict to retarded solutions, we note that we can identify

G = G(R) and that the contribution from the upper limit in the
second term on the right in (6.200) vanishes because t − t′ < 0 at
that point. Thus, in such a case, using (6.185) we can write
Z
Ψ(x, t) = − d4 x′ G(R) (x, x′ )f (x′ )
Z " #
1 ∂Ψ ∂G(R)
− d 3 x′ G(R) ′ − Ψ
4πc ∂t ∂t′
t′ =ti
Ztf Z
c ′
− dt ds′ · G(R) ∇′ Ψ(x′ , t′ ) − Ψ(x′ , t′ )∇′ G(R)
4π
ti
Z
f (x′ , t′ )
3 ′
= d x
|x − x′ | t′ =t− |x−x′ |
c
Z " #
1 ∂Ψ ∂G(R)
− d3 x′ G(R) ′ − Ψ
4πc ∂t ∂t′ ′ t =ti
Ztf Z
c ′
− dt ds′ · G(R) ∇′ Ψ(x′ , t′ ) − Ψ(x′ , t′ )∇′ G(R) .
4π
ti
(6.201)
There are two special cases that we will consider now. First,
let us consider the case where the volume of space is infinite. In
such a case, with the assumptions of asymptotic fall off for the fields
(variables), the surface integral in (6.201) vanishes. The remaining
terms are determined completely in terms of the initial values of Ψ
and ∂Ψ∂t . Namely, we have a solution of the Cauchy initial value
6.9 Kirchhoff’s representation 221
problem subject to the retarded boundary condition. Let us choose

ti = 0 and denote

∂Ψ
Ψ(x, t = 0) = F (x), = H(x). (6.202)
∂t t=0
Furthermore, to simplify the evaluation of the integrals, let us choose

the point of observation to be x = 0. In such a case, the solution
(6.201) becomes

Z |x′ |
h f x′ , t′ = t − c
Ψ(0, t) = d 3 x′
|x′ |
′
i
1 ′ δ(ct − |x |) ′ ∂ δ(c(t′ − t) + |x′ |)
+ H(x ) − F (x ) ′ ′ ,
4πc |x′ | ∂t |x′ | t =0
(6.203)
where we have used the form of the retarded Green’s function in

(6.185). Going over to the spherical coordinates, this gives
Z
′ ′ ′ ′ ′ r′
Ψ(0, t) = dΩ r dr f r , Ω, t = t −
c
Z ′

1 ′ ′ ′ ′ ′ ′ ′ ′ ∂δ(ct − r )
+ dΩ r dr H(r , Ω )δ(ct − r ) + F (r , Ω )
4πc ∂t
Z
r′
= dΩ′ r ′ dr ′ f r ′ , Ω′ , t′ = t −
c
Z
1 ′ ′ ∂(tF (ct, Ω′ ))
+ dΩ tH(ct, Ω ) + . (6.204)
4π ∂t
This gives the solution of the initial value problem once we know the
explicit forms of F (x) and H(x).
The second case that we are interested in is when the volume is
finite. Furthermore, let us assume that there are no sources present
in this volume, namely, f (x) = 0 and that the initial values are also
trivial. In such a case, (6.201) leads to
Ztf Z
c
Ψ(x, t) = − dt′ ds′ · G(R) ∇′ Ψ(x′ , t′ ) − Ψ(x′ , t′ )∇′ G(R) .
4π
ti
(6.205)
Once again, we can use the form of the retarded Green’s function in
(6.185) and defining, for simplicity R = x − x′ , we have
′

′ (R) ′ δ(c(t − t ) − R)
∇G = −∇
R

′ ∂ δ(c(t − t′ ) − R)
= −(∇ R)
∂R R

δ(c(t − t′ ) − R) 1 ∂δ(c(t − t′ ) − R)
= −R̂ − , (6.206)
R2 cR ∂t′
where we have used (∇′ R) = −R̂. Using this, we can do the time
integral in (6.205) to obtain
Ztf Z
c ′ ′ δ(c(t − t′ ) − R) ′
Ψ(x, t) = dt ds · ∇Ψ
4π R
ti

δ(c(t − t′ ) − R) 1 ∂δ(c(t − t′ ) − R)
− R̂ − Ψ
R2 cR ∂t′
Z " #
1 ′ 1 ′ R̂ R̂ ∂Ψ
= ds · ∇ Ψ − 2Ψ − . (6.207)
4π R R cR ∂t′ ′ R
t =t− c
This is known as Kirchhoff’s representation for the solution of

the wave equation. It is important to recognize that this does not
provide an explicit solution for the equation, since the unknown Ψ
appears on the right hand side. Rather, it gives an integral repre-
sentation for the wave disturbance and this provides a mathematical
description of Huygen’s principle in the sense that the value of the
solution at any point is given completely by its value as well as the
derivatives of the function on the surface of a closed volume.
1. (a) Show that the four Maxwell’s equations (in any medium)
imply the continuity equation.
(b) For a system with well localized charges and currents (namely,
charges and currents which do not extend to infinity), show that
the total charge
Z
Q= d3 x ρ(x, t),
all space
of the system does not change with time as a consequence of

the continuity equation.
2. Consider the following one dimensional problem. A wire of

cross-sectional area A and length L carries a current I and is
maintained at a potential difference (voltage) V . Defining the
specific resistivity of the material to be ρ which is the inverse
of conductivity σ, show that Ohm’s law gives
J = σE.
L
(Just to remind you, the resistance R is defined as R = A ρ.)
3. Work out the plane wave solution for a charge neutral (without
any free charge) conducting medium, for which you may assume
J = σE. Compare the present solution with the one for the
dielectrics obtained in this chapter.
4. Show explicitly that Maxwell’s equations in terms of the scalar

and the vector potential, namely,
1 ∂(∇ · A) 4π
∇2 Φ + = − ρ,
c ∂t ǫ
2

2 ǫµ ∂ 4πµ ǫµ ∂Φ
∇ − 2 2 A=− J+∇ ∇·A+ ,
c ∂t c c ∂t
are invariant under the gauge transformations
1 ∂Λ
Φ→Φ+ , A → A − ∇Λ.
c ∂t
5. For a plane wave solution of the Maxwell’s equations, calculate

the time averaged power radiated through a surface of unit
area (namely, calculate the time averaged value of k̂ · S over
one period where k̂ is the direction of propagation).
6. It is possible to define the Poynting vector, S, even in the static

case. However, show that, in the static case in the absence of
currents, there is no power loss due to radiation through any
closed surface.
7. If we write the electric and the magnetic fields in the complex

notation, then the time-averaged Poynting vector (in vacuum)
can be written in the form
c c
S= Re (E × H∗ ) = Re (E × B∗ ) .
8π 8π
Consider the following three plane wave solutions of Maxwell’s

equations
E1 = x̂E1 e−i(ωt−kz) , B1 = ŷE1 e−i(ωt−kz) ,
E2 = ŷE2 e−i(ωt−kz+α) , B2 = −x̂E2 e−i(ωt−kz+α) ,
E3 = x̂E3 e−i(ωt−kz+α) , B3 = ŷE3 e−i(ωt−kz+α) ,
where E1 , E2 , E3 are real constants.
(a) For a solution which is a superposition of the first two (of

equal weight), show that S1+2 = S1 + S2 . Explain this result.
(b) For a solution which is a superposition of the first and the
third (of equal weight), calculate S1+3 and compare with the
earlier case.
Chapter 7
Wave guides
We have already studied simple solutions of Maxwell’s equations and

have seen that electromagnetic waves carry power. It is, therefore, an
interesting question to ask how this power can be transmitted from
one place to another. There are two kinds of transmission problems
that may be of interest. First, we may want electric power to be
transmitted from a specific point to another without an intermediate
party having access to it. The significance of this is clear, for example,
from the usual domestic power connection. It is also crucial in the
case of telephone conversations, as we may not want a third party
to overhear a private conversation. The second kind of transmission
corresponds to radio or television transmissions which may not raise
such privacy concerns. Here, anyone who can tune to a particular
frequency is capable of getting the signal. In this lecture, we will
study transmissions of the first kind relegating the second topic to
later lectures.
Transmission of electric power from one specific point to an-
other is achieved much the same way as the water supply or the
gas supply. The usual transmission of power is through parallel wire
lines or through coaxial cables. Such a mode is suitable for low fre-
quency transmissions of less than 200 MHz. For higher frequencies,
the physical dimensions of such a carrier system become unrealis-
tic. For transmissions of power of higher frequency one uses wave
guides which are basically hollow metal tubes where the metal walls
of the tube “guide” the wave along the tube (much like the role of the
water pipes). The wave guides can be of any cross-sectional shape.
However, the two most commonly used are the rectangular and the
cylindrical wave guides. We note that, in discussing wave guides,
it is commonly assumed that the metal walls of the wave guide are
perfect conductors with σ → ∞, where σ represents the conductivity
of the metal. In reality, of course, metals have a finite conductivity
and electric fields can penetrate inside a metal. However, the skin
225
226 7 Wave guides
depth (penetration depth) at high frequencies is quite small. Conse-

quently, the assumption of the walls being perfect conductors works
well and the small deviations from this assumption can be calculated
systematically if necessary.
In discussing wave guides, it is very important to understand the

boundary conditions for the system. Let us recall that Maxwell’s
equations in an arbitrary region have the form
∇ · D = 4πρ,
∇ · B = 0,
1 ∂B
∇×E=− ,
c ∂t
4π 1 ∂D
∇×H= J+ . (7.1)
c c ∂t
Various fields satisfy the relations noted earlier in (6.13)
D = ǫ E,
1
H= B, (7.2)
µ
where ǫ, µ represent respectively the permittivity and the permeabil-
ity of the medium. In addition, in a metal, we can relate the con-
duction current to the electric field through Ohm’s law as (see the
second relation in (6.12))
J = σ E, (7.3)
with σ representing the conductivity of the metal. (Incidentally, re-

lation (7.3) is true both in the CGS as well as the MKS system of
units.)
The boundary conditions in the interface of two different media
can be derived from (7.1) in the standard manner as we have done
earlier in the case of static problems. Let us note here only the
geometrical behavior of the electric and the magnetic fields when time
variations are allowed. We note that the first equation of Maxwell (in
(7.1)) implies that the electric fields must begin and end on charges
while the second implies that magnetic field lines must form closed
loops since there is no magnetic charge (magnetic monopole). From
the third equation of Maxwell, we see that electric fields must form
closed loops around time varying magnetic fields. Similarly, the last
equation of Maxwell leads to the fact that magnetic fields must form
closed loops around a conduction current or a “displacement” current
produced by a time varying electric field. This is the general behavior
of electric and magnetic fields when time dependence is present. If we
have a perfect conductor as would be the case for the walls of a wave
guide, then, we realize that the electric and the magnetic fields cannot
penetrate inside the metal. Any change in the external fields would
lead to an instantaneous response whereby charges will move to the
surface of the conductor to prevent any field within. In particular,
we note from (7.3) that, for a perfect conductor with σ → ∞, there
cannot be any tangential component of the electric field present on the
surface of the conductor (for any t), which would otherwise imply an
infinite conduction current that is physically untenable. Similarly, the
normal component of the magnetic field must vanish on the surface of
the conductor simply because there are no magnetic monopoles (on
the surface of the conductor). Mathematically, we can write these
boundary conditions as
n̂ × E| = 0,
n̂ · B| = 0, (7.4)
where n̂ represents a unit vector normal (outward) to the conduct-

ing surface and the restriction implies the validity of these at the
boundary corresponding to the locations of the conducting surfaces.
A wave guide, as we have noted earlier, is simply a hollow metal
tube without any free charge or current inside. Let us assume that
the length of the tube is along the z-axis, which is the direction in
which we would like the electromagnetic wave to be transmitted. The
shape of the transverse cross-section of the wave guide will depend on
whether we have a rectangular or a cylindrical wave guide. Therefore,
to keep things as general as is possible, let us decompose the fields
into components along the z-axis and normal to it as
E = ẑ × (E × ẑ) + ẑ (ẑ · E) = E⊥ + Ek = E⊥ + ẑEz ,

B = ẑ × (B × ẑ) + ẑ (ẑ · B) = B⊥ + Bk = B⊥ + ẑBz . (7.5)
This follows from the familiar vector identity (if A, B, C involve op-
erators, the order should be maintained)
A × (B × C) = (A · C) B − (A · B) C. (7.6)
228 7 Wave guides
Similarly, we can write
∂
∇ = ∇⊥ + ∇k = ∇⊥ + ẑ . (7.7)
∂z
Exercise. Although ẑ × E gives a component of the E field normal to the z-axis,
show that (7.5) gives the correct decomposition of the fields. Namely, show that
any arbitrary vector V can be written uniquely as
V = ẑ × (V × ẑ) + ẑ (ẑ · V). (7.8)
With the decomposition in (7.5) and (7.7), in a charge free re-

gion, the first two equations of (7.1) take the forms
∂Ez
∇⊥ · E⊥ = − ,
∂z
∂Bz
∇⊥ · B⊥ = − . (7.9)
∂z
The last two equations in (7.1), being vector equations, decompose

into two equations each. For example, the third equation in (7.1) can
be written as

∂ 1 ∂ (B⊥ + ẑBz )
∇⊥ + ẑ × (E⊥ + ẑEz ) = − , (7.10)
∂z c ∂t

∂E⊥ 1 ∂ (B⊥ + ẑBz )
or, ∇⊥ × E⊥ + ẑ × − ∇⊥ Ez = − .
∂z c ∂t
This, in turn, leads to (taking the dot as well as cross product with
ẑ)
1 ∂Bz
ẑ · (∇⊥ × E⊥ ) = − ,
c ∂t
∂E⊥ 1 ∂B⊥
− ẑ × = ∇⊥ Ez , (7.11)
∂z c ∂t
where we have used the fact that ∇⊥ × E⊥ points along the z-axis as
well as (7.6). Similarly, the last equation in (7.1) can be decomposed
into
ǫµ ∂Ez
ẑ · (∇⊥ × B⊥ ) = ,
c ∂t
∂B⊥ ǫµ ∂E⊥
+ ẑ × = ∇⊥ Bz . (7.12)
∂z c ∂t
Let us next assume that the fields are harmonically varying with
time. Furthermore, since we want the wave to be propagating along
the z-axis, we can also extract the z dependence of the fields to write
E(x⊥ , z, t) = E(x⊥ )e−i(ωt−kz) ,
B(x⊥ , z, t) = B(x⊥ )e−i(ωt−kz) , (7.13)
where ω represents the frequency of the wave and k is a constant

parameter, complex, in general. (Recall that the fields are real. When
we write it in the form as in (7.13), we are tacitly assuming that we
are looking at the real or the imaginary part of it.) Factoring out the
(t, z) dependence, the Maxwell’s equations, (7.9), (7.11) and (7.12),
take the forms
∇⊥ · E⊥ = −ikEz ,
∇⊥ · B⊥ = −ikBz ,
iω
ẑ · (∇⊥ × E⊥ ) = Bz ,
c
iω
ikE⊥ + ẑ × B⊥ = ∇⊥ Ez ,
c
iǫµω
ẑ · (∇⊥ × B⊥ ) = − Ez ,
c
iǫµω
ikB⊥ − ẑ × E⊥ = ∇⊥ Bz . (7.14)
c
There are several things to note from Eq. (7.14). First, we

note that the transverse components of the fields are completely de-
termined from a knowledge of Ez , Bz . For example, from the fourth
and the last equations in (7.14), we obtain trivially that (for example,
multiply the fourth equation with (ik) and then use the last equation)
i ω
E⊥ = ǫµω 2
k∇⊥ Ez − ẑ × ∇⊥ Bz ,
− k2 c
c2
i ǫµω
B⊥ = ǫµω 2
k∇⊥ Bz + ẑ × ∇⊥ Ez . (7.15)
− k2 c
c2
(Here, we are considering a wave traveling along the positive z-axis.

For a wave traveling in the reverse direction, we simply let k → −k.)
Furthermore, from the fourth equation in (7.14) we obtain (using the
230 7 Wave guides
first as well as the fifth equations)

iω
∇⊥ · ikE⊥ + ẑ × B⊥ = ∇2⊥ Ez ,
c

2 ǫµω 2 2
or, ∇⊥ Ez + − k Ez = 0. (7.16)
c2
Similarly, it follows from Eq. (7.14) that

2 ǫµω 2 2
∇⊥ Bz + − k Bz = 0. (7.17)
c2
The solutions of the wave equation are correspondingly classified

into three categories depending on the values of Ez , Bz . When Ez =
0 = Bz , namely, if both the electric and the magnetic fields have
only transverse components, then the solution is known as the TEM
(transverse electromagnetic) solution. This is the case already seen
for plane wave traveling solutions following from Maxwell’s equations.
From Eqs. (7.14) and (7.15), it is clear that such solutions exist only
if
√
ǫµ ω
k = k0 = , (7.18)
c
and that in such a case it follows, for example, from the last relation
in (7.14) that (remember Bz = 0)
ǫµω √
B⊥ = ± ẑ × E⊥ = ± ǫµ ẑ × E⊥ , (7.19)
kc
as we have discussed earlier (see, for example, (6.30)). (The two signs
correspond to the waves traveling along ±z directions respectively.)
If Bz = 0 but Ez 6= 0, then the corresponding solution is known
as the TM (transverse magnetic) solution since the magnetic field is
transverse to the direction of propagation in this case. In this case,
from (7.15) (or alternatively from the last relation in (7.14)) we have
iǫµω
c
B⊥ = ǫµω 2
ẑ × ∇⊥ Ez
c2
− k2
2 !
iǫµω
c
−i( ǫµω
c2
− k2 )
= ǫµω 2
ẑ × E⊥ ,
− k2 k
c2
ǫµω
or, B⊥ = ẑ × E⊥ . (7.20)
kc
7.2 Rectangular wave guide 231
Similarly, if Ez = 0 but Bz 6= 0, then the corresponding solution

is known as the TE (transverse electric) solution because, in this
case, the electric field is transverse to the direction of propagation.
It follows from (7.15) (or alternatively from the fourth relation in
(7.14)) that in this case, we can write
kc
B⊥ = ẑ × E⊥ . (7.21)
ω
Furthermore, we note that since the direction of propagation is chosen
to be along the z-axis, the boundary conditions (7.4) for a wave guide
take the simpler form

∂Bz
Etan | = 0 = , (7.22)
∂n
where Etan represents the component of the electric field tangential

∂
to the boundary surface and ∂n denotes derivative along the normal
direction and the second identity in (7.22) follows from using the
second and the last equation in (7.14). Namely, we note that
n̂ · (ẑ × E⊥ ) = −ẑ · (n̂ × E⊥ ) = −ẑ · (n̂ × (E − ẑEz ))

= −ẑ · Etan , (7.23)
which vanishes at the boundary. Furthermore, we recognize that the

second boundary condition in (7.4) implies that (recall that n̂ · ẑ = 0)
n̂ · (B⊥ + ẑBz ) = n̂ · B⊥ , (7.24)
which vanishes at the boundary. Using (7.23) and (7.24) in the last
equation of (7.14) leads to

∂
n̂ · ∇⊥ Bz | = n̂ · ∇ − ẑ Bz = n̂ · ∇Bz |
∂z

∂Bz
= = 0, (7.25)
∂n
where we have identified n = n̂ · x. With these basics, we are now

ready to analyze the solutions in the case of a rectangular wave guide.
7.2 Rectangular wave guide
Let us consider a rectangular hollow tube along the z-axis and with
transverse dimensions satisfying 0 ≤ x ≤ a, 0 ≤ y ≤ b as shown in
232 7 Wave guides
y=b
y
z
x=a x
Figure 7.1: A rectangular wave guide along the z-axis.
Fig. 7.1. As we have noted, the walls of the tube are assumed to be
perfect conductors. In this case, the solutions of Maxwell’s equations
have to satisfy the boundary conditions (7.22). It is clear that, in
this case, for the wall in the x-z plane, n̂ is parallel to ŷ and this
leads to the boundary conditions (see (7.22))

∂Bz
Ex |y=0,b = 0 = Ez |y=0,b , = 0. (7.26)
∂y y=0,b
On the other hand, for the wall in the y-z plane, n̂ is parallel to x̂
leading to the boundary conditions

∂Bz
Ey |x=0,a = 0 = Ez |x=0,a , = 0. (7.27)
∂x x=0,a
Equations (7.26) and (7.27) define all the boundary conditions in this
case. We note that Eq. (7.15) explicitly takes the forms

i ∂Ez ω ∂Bz
Ex = ǫµω2 k + ,
2 ∂x c ∂y
c2 − k

i ∂Ez ω ∂Bz
Ey = ǫµω2 k − ,
2 − k 2 ∂y c ∂x
c

i ∂Bz ǫµω ∂Ez
Bx = ǫµω2 k − ,
− k2 ∂x c ∂y
c2

By = ǫµω2 k + , (7.28)
2 − k 2 ∂y c ∂x
c
while the z-components of the fields satisfy

∂ 2 Ez ∂ 2 Ez ǫµω 2 2
+ + − k Ez = 0,
∂x2 ∂y 2 c2

∂ 2 Bz ∂ 2 Bz ǫµω 2 2
+ + − k Bz = 0. (7.29)
∂x2 ∂y 2 c2
Equations (7.28) and (7.29) have to be solved subject to the boundary

conditions (7.26) and (7.27).
7.2.1 TM waves. In this case, we assume that Bz = 0. The solution

for the z-component of the electric field in Eq. (7.29), subject to the
boundary conditions in (7.26) and (7.27), yields
πmx πny
Ez (x⊥ ) = Ez (x, y) = A sin sin , (7.30)
a b
where A is an arbitrary constant and
2 2
ǫµω 2 π m π 2 n2
k2 = 2 − + , m, n = 1, 2, . . . . (7.31)
c a2 b2
The transverse components of the fields are then determined from
(7.28) to be
i kAπm πmx πny
Ex (x⊥ ) = ǫµω 2
cos sin ,
− k2 a a b
c2
i kAπn πmx πny
Ey (x⊥ ) = ǫµω 2
sin cos ,
− k2 b a b
c2
i ǫµωAπn πmx πny
Bx (x⊥ ) = − ǫµω2 sin cos ,
− k2 cb a b
c2
i ǫµωAπm πmx πny
By (x⊥ ) = ǫµω 2
cos sin . (7.32)
− k2 ca a b
c2
It is clear from (7.32) that we can write, in this case,

ǫµω
B⊥ = ẑ × E⊥ , (7.33)
kc
consistent with (7.20). Note that all the transverse components of
the fields satisfy the boundary conditions in (7.26) and (7.27).
The solutions in (7.30)–(7.32), corresponding to fixed integers,
m, n, are known as TMmn modes. From (7.31), it is clear that the
234 7 Wave guides
constant k depends on ω as well as the values of the integers m, n.

For small values of ω, it follows from
s 2 2
ǫµω 2 π m π 2 n2
k= − + 2 , (7.34)
c2 a2 b
that the quantity inside the square root can be negative. In this
case, k will become purely imaginary. As a result, there will be no
propagation inside the wave guide. Rather, the electromagnetic signal
will be attenuated along the tube. Defining
r √ p
πc m2 n 2 ǫµ
ωmn = √ 2
+ 2, k = ω 2 − ωmn
2 , (7.35)
ǫµ a b c
we see that propagation of the TMmn wave can take place inside a
rectangular wave guide only if ω > ωmn which is known as the cut-
off frequency below which propagation of the TMmn wave will not
occur. Above this frequency, the TMmn wave will propagate without
any attenuation for a wave guide with perfectly conducting walls.
The wavelength of propagation inside the wave guide for the
TMmn wave is easily obtained to be
2π 2πc 1
λ= =√ p
k ǫµ ω − ωmn
2 2
2πc 1
=√ r . (7.36)
ǫµ π 2 c2 m2 n2
ω2 − ǫµ a2
+ b2
Correspondingly, the velocity of propagation inside the wave guide is

given by
ω c ω
v= =√ p
k ǫµ ω 2 − ωmn 2
c ω
=√ r . (7.37)
ǫµ 2 2 2 n2
ω 2 − πǫµc m a2
+ b2
Note that, in deriving all these results, we have assumed the inside of
the wave guide to be filled with an arbitrary dielectric. If we assume
that it is empty space inside, then in this case we can identify ǫµ = 1
and the formulae in this section simplify. With this, we note from
Eq. (7.37) that the velocity of propagation of the wave, inside the
wave guide, is larger than its value in free space. We note that it
is infinite at the cut-off frequency. As the frequency is increased, it

decreases and for arbitrarily large frequencies, it approaches c, the
speed of light in free space (for ǫµ = 1). The fact that the velocity of
propagation is larger than the speed of light is not disturbing since
this corresponds to the phase velocity of the wave. We note, from
(7.31) (or (7.34)), that the group velocity, in this case, is given by
∂ω c2 k c2 1
vg = = = ,
∂k ǫµ ω ǫµ v
c2
or, vg v = , (7.38)
ǫµ
so that, when the phase velocity is larger than the speed of light, the
group velocity is smaller, as it should be.
As is clear, the wave guide can support an infinite number of
TM modes. It is for the mode corresponding to m = 1 = n that
the cut-off frequency is the smallest and would correspond to the
dominant TM mode. In this case, with ǫµ = 1, we have
√
πc a2 + b2
ω11 = ,
ab
c
λ = 2π q ,
2 2 2 2
ω 2 − π c a(a2 b2+b )
ω
v=c q . (7.39)
π 2 c2 (a2 +b2 )
ω2 − a2 b2
7.2.2 TE waves. In the case of TE modes, we have Ez = 0. It follows,

then, that the solution of (7.29) subject to the boundary conditions
in (7.26) and (7.27) is given by
πmx πny
Bz (x⊥ ) = Bz (x, y) = C cos cos , (7.40)
a b
where C is an arbitrary constant and we have, as in Eq. (7.31),

2 ǫµω 2 π 2 m2 π 2 n 2
k = 2 − + 2 , m, n = 0, 1, 2, . . . . (7.41)
c a2 b
236 7 Wave guides
The transverse components of the electric and the magnetic fields, in

this case, become
i ωCπn πmx πny
Ex (x⊥ ) = − ǫµω2 cos sin ,
c2
− k 2 cb a b
i ωCπm πmx πny

Ey (x⊥ ) = ǫµω 2
sin cos ,
− k2 ca a b
c2
i kCπm πmx πny
Bx (x⊥ ) = − ǫµω2 sin cos ,
− k2 a a b
c2
i kCπn πmx πny
By (x⊥ ) = − ǫµω2 cos sin . (7.42)
− k2 b a b
c2
It follows from this that, for TE waves, we can write

kc
B⊥ (x⊥ ) = ẑ × E⊥ (x⊥ ), (7.43)
ω
consistent with (7.21). We see that the transverse components of the
fields satisfy the boundary conditions in (7.26).
Like the TM waves, we see that a rectangular wave guide can
also support an infinite number of TE modes known as TEmn modes,
but unlike the TM modes, here it is possible to have m = 0 or n = 0.
However, from the form of the transverse fields in (7.42), we see that
we cannot have m = 0 = n because, in that case, all the transverse
fields would vanish and we will have Bz = constant which would
correspond to a trivial solution. As in the case of TM modes (see Eq.
(7.35)), here, too, there is an analogous cut-off frequency for TEmn
waves given by
r
πc m2 n 2
ωmn = √ + 2, (7.44)
ǫµ a2 b
and the discussion of the wave length of propagation as well as the

velocity of propagation goes through in a completely parallel manner.
However, in this case since we can have m = 0 or n = 0, the lowest
cut-off frequency will be for the TE10 mode (where we are assuming
that a > b) given by
πc
ω10 = √ . (7.45)
ǫµa
Correspondingly, the TE10 mode is called the dominant mode in a

rectangular wave guide. It is also clear from (7.42) that in this case
7.3 Cylindrical wave guide 237
Ez = 0 = Ex . As a result, for this mode the electric field is parallel

to the y-axis everywhere. Furthermore, for the dominant TE10 mode,
we have
2π 2πc 1
λ= =√ q ,
k ǫµ ω 2 − π 2 c2
ǫµa2
ω c ω
v= =√ q . (7.46)
k ǫµ ω 2 − π 2 c2
ǫµa2
7.3 Cylindrical wave guide
The other kind of wave guide that is commonly used is the cylindrical
wave guide. Here, we have a hollow cylindrical tube of radius a with
a perfectly conducting wall and length of the tube along the z-axis
as shown in Fig. 7.2. Because of the symmetry in the problem, it is
simpler to study this problem in cylindrical coordinates defined by
r=a
Figure 7.2: A cylindrical wave guide of radius a along the z-axis.
x = r cos φ, y = r sin φ, z = z. (7.47)
Here, r, φ denote the radial and the angular coordinates on a trans-

verse plane. The unit vectors satisfy
r̂ × φ̂ = ẑ, φ̂ × ẑ = r̂, ẑ × r̂ = φ̂.

238 7 Wave guides
The unit vectors, in non-Cartesian coordinate systems, are not fixed

and their variations can be determined from the transformation rules
x̂ = r̂ cos φ − φ̂ sin φ,
ŷ = r̂ sin φ + φ̂ cos φ,
ẑ = ẑ.
These can be inverted to give
r̂ = x̂ cos φ + ŷ sin φ,
φ̂ = −x̂ sin φ + ŷ cos φ,

ẑ = ẑ.
From these, it follows that the only non-trivial variation of the unit
vectors are given by
∂r̂ ∂ φ̂
= φ̂, = −r̂. (7.48)
∂φ ∂φ
We also note here that in cylindrical coordinates,
∂ φ̂ ∂
∇⊥ = r̂ + ,
∂r r ∂φ

2 1 ∂ ∂ 1 ∂2
∇⊥ = r + 2 .
r ∂r ∂r r ∂φ2
With these, we can go back and recast all the equations (7.13)–
(7.17) and (7.22) in cylindrical coordinates. In particular, the bound-
ary conditions (7.22) in the case of cylindrical wave guides take the
form (here n̂ is parallel to r̂)

∂Bz
Ez |r=a = 0 = Eφ |r=a , = 0. (7.49)
∂r r=a
The relation for the transverse components in (7.15) take the forms

i ∂Ez ω ∂Bz
Er (x⊥ ) = Er (r, φ) = ǫµω2 k + ,
2 − k 2 ∂r cr ∂φ
c

i k ∂Ez ω ∂Bz
Eφ (x⊥ ) = Eφ (r, φ) = ǫµω2 − ,
2 − k 2 r ∂φ
c
c ∂r

Br (x⊥ ) = Br (r, φ) = ǫµω2 k − ,
− k2 ∂r cr ∂φ
c2

i k ∂Bz ǫµω ∂Ez
Bφ (x⊥ ) = Bφ (r, φ) = ǫµω2 + , (7.50)
2 − k 2 r ∂φ c ∂r
c
while the equations for the Ez , Bz (see Eqs. (7.16)–(7.17)) take the
forms

∂ 2 Ez 1 ∂Ez 1 ∂ 2 Ez ǫµω 2 2
+ + 2 + − k Ez = 0,
∂r 2 r ∂r r ∂φ2 c2

∂ 2 Bz 1 ∂Bz 1 ∂ 2 Bz ǫµω 2 2
+ + 2 + − k Bz = 0. (7.51)
∂r 2 r ∂r r ∂φ2 c2
Physical solutions have to be periodic in the angular variable.
Consequently, we can extract the φ dependence of the z-components
of the fields as

Ez (r, φ) = Ez(0) (r) An ein(φ+φ0 ) + Bn e−in(φ+φ0 ) ,

Bz (r, φ) = Bz(0) (r) Cn ein(φ+φ0 ) + Dn e−in(φ+φ0 ) . (7.52)
Furthermore, by choosing the arbitrary constants An , Bn , Cn , Dn , φ0

appropriately (which is equivalent to choosing an axis of orientation
of the wave), we can write
Ez (r, φ) = Ez(0) (r) cos nφ,
Bz (r, φ) = Bz(0) (r) cos nφ, (7.53)
where n = 0, 1, 2, . . . and the overall constants have been absorbed

(0) (0)
into the definitions of Ez , Bz . Thus, Eq. (7.51) takes the form
(0) (0)
d2 Ez 1 dEz ǫµω 2 n2
+ + − k − 2 Ez(0) = 0,
2
dr 2 r dr c2 r
(0) (0)
d 2 Bz 1 dBz ǫµω 2 2 n2
+ + − k − Bz(0) = 0. (7.54)
dr 2 r dr c2 r2
Each of the equations in (7.54) is a Bessel equation of order n.
(Fig. 7.3 shows the behavior of J0 (x) as a function of x.) Conse-
quently, we can write the solutions of (7.51) as
Ez (r, φ) = an Jn (hr) cos nφ,

Bz (r, φ) = bn Jn (hr) cos nφ, (7.55)
240 7 Wave guides
0.8
0.6
J0 HxL
0.4
0.2
-0.2
-0.4
0 2 4 6 8 10 12
x
Figure 7.3: The behavior of the Bessel function J0 (x) as a function

of x.

r
ǫµω 2
h= − k2 . (7.56)
c2
7.3.1 TM waves. In this case, we assume that Bz = 0 (namely, bn =

0), whereas Ez has the form given in (7.55). The boundary condition
on Ez , (7.49), implies that
Jn (ha) = 0. (7.57)
There are an infinite number of zeros of the Bessel function for any
n and denoting all such roots as (ha)mn (implying the mth zero of
Jn ), we note that the first few take the values
(ha)10 = 2.405, (ha)11 = 3.85, (ha)12 = 7.02,

(ha)20 = 5.52, · · · . (7.58)
From (7.56), this determines the constant k to be

r
ǫµω 2
k= − h2mn . (7.59)
c2
The transverse components of the fields can now be determined

from (7.50) to be
ian
Er (r, φ) = ǫµω 2
k Jn′ (hr) cos nφ,
c2
− k2
ian kn
Eφ (r, φ) = − ǫµω2 Jn (hr) sin nφ,
2 − k 2 r
c
ian ǫµωn
Br (r, φ) = ǫµω 2
Jn (hr) sin nφ,
− k2 cr
c2
ian ǫµω ′
Bφ (r, φ) = ǫµω 2
Jn (hr) cos nφ. (7.60)
− k2 c
c2
Here, a prime denotes a derivative with respect to the radial coordi-

nate. We note from Eq. (7.60) that the transverse field Eφ satisfies
the boundary conditions in (7.49). It is clear from Eq. (7.60) that
we can write, in this case,
ǫµω
B⊥ = ẑ × E⊥ , (7.61)
kc
consistent with (7.20).
The TM modes with different values for (ha) (see (7.58)) are
known as TMmn modes of the cylindrical wave guide. From Eq.
(7.59) we note that, as in the case of the rectangular wave guide,
there is also a cut-off frequency here given by
c
ωmn = √ hmn . (7.62)
ǫµ
For ω < ωmn , there is no transmission of the TMmn wave in a cylin-
drical wave guide. With ω > ωmn , the wave is propagated without
any attenuation for perfectly conducting outer boundaries. The wave
length of propagation, in this case, is given by
2π 2πc 1
λ= =√ p , (7.63)
k ǫµ ω 2 − ωmn
2
while the velocity of propagation takes the form

ω c ω
v= =√ p . (7.64)
k ǫµ ω 2 − ωmn2
The velocity, as in the case of the rectangular wave guide, is infinitely

large near the cut-off frequency and goes to the speed of light in a
dielectric for asymptotically large frequencies (in free space, this is
just c). As is clear from Eqs. (7.58) and (7.62), the dominant TM
mode in a cylindrical wave guide is the TM10 mode.
242 7 Wave guides
7.3.2 TE waves. In the case of TE waves, we assume that Ez = 0

(namely, an = 0) while Bz has the form given in Eq. (7.55). However,
in this case, the boundary condition on the magnetic field, (7.49),
requires that
Jn′ (ha) = 0. (7.65)
In contrast to the TM case, here we are not interested in the zeros
of the Bessel function, rather the locations where the Bessel function
has vanishing slope are important. Once again, for every value of n,
there is an infinite number of such points given by
(ha)10 = 3.83, (ha)11 = 1.84, (ha)12 = 3.05,
(ha)20 = 7.02, (ha)21 = 5.33, · · · , (7.66)
and as in (7.59), the value of k is determined from (7.56) to be
r
ǫµω 2
k= − h2mn . (7.67)
c2
The transverse components of the fields are now determined
from (7.50) and (7.55) to be
ibn ωn
Er (r, φ) = − ǫµω2 Jn (hr) sin nφ,
− k 2 cr
c 2
ibn ω ′
Eφ (r, φ) = − ǫµω2 J (hr) cos nφ,
− k 2 c n
c 2
ibn
Br (r, φ) = ǫµω 2
k Jn′ (hr) cos nφ,
c2
− k2
ibn kn
Bφ (r, φ) = − ǫµω2 Jn (hr) sin nφ. (7.68)
c2
−k r
2
It is clear from Eq. (7.68) that Eφ satisfies the boundary condition

in (7.49) by virtue of (7.65) and that we can write
kc
B⊥ = ẑ × E⊥ , (7.69)
ω
consistent with (7.21).
The TE modes with different roots in (7.66) are known as TEmn
modes and it is clear that, as in the case of the TM modes, the cut-off
frequency is given by
c
ωmn = √ hmn ,
ǫµ
7.4 TEM waves 243
and correspondingly, for the propagating modes, ω > ωmn , the wave
length and the velocity of propagation are given as in Eqs. (7.63) and
(7.64). We note from (7.66) that the dominant mode, in a cylindrical
wave guide, is the TE11 mode. Thus, we see that the rectangular and
the cylindrical wave guides have similar qualitative features.
7.4 Impossibility of having TEM waves in a wave guide
Although we have classified electromagnetic waves into three cat-

egories, in discussing propagation in a rectangular or a cylindrical
wave guide, we have only discussed the TM and TE modes of propa-
gation. The reason for this is that in a hollow wave guide, rectangular
or cylindrical, TEM waves cannot be present. This can be easily seen
in the following manner. Let us note that by definition, both the elec-
tric and the magnetic fields are transverse in a TEM wave. Namely,
for such a solution, we must have
Ez = 0 = Bz .
We know from Maxwell’s equation, ∇ ·B = 0, that the magnetic field

lines must form closed loops. In particular, when Bz = 0, they must
form closed loops in the transverse plane to the z-axis. Furthermore,
from the last equation of Maxwell in (7.1), we see that in the absence
of a conduction current
ǫµ ∂E
∇×B= ,
c ∂t
so that these closed magnetic loops must enclose the “displacement”
current. However, since the closed magnetic loops are in the trans-
verse plane, this is possible only if the “displacement” current has a
component along the z-axis, which would imply that the electric field
itself has a time varying component along the z-axis. This is, how-
ever, in contradiction to the requirement of a TEM wave for which
Ez = 0.
Thus, we see that in a hollow wave guide, there cannot be any
TEM wave present. The TM and the TE waves represent all the
modes that can be present in such a system. On the other hand,
waves guided by two infinite parallel conducting plates, for example,
can support TEM modes. This can be seen from our previous analysis
of the rectangular wave guides by taking the limit a → ∞. In this
limit, since there is no boundary in the x-direction, we can assume
that the fields are uniform along this direction (namely, there is no x
244 7 Wave guides
dependence because of translation invariance). The TM solution, in

this case, is easily seen to lead from (7.29) and (7.14) to
Bz = 0,
πmC πmy
Ez (y) = sin ,
b b
Ex (y) = 0,
πmy
Ey (y) = ikC cos ,
b
iǫµωC πmy ǫµω
Bx (y) = − cos =− Ey ,
c b kc
By = 0. (7.70)
Here, we have
r
ǫµω 2 π 2 m2
k= − , (7.71)
c2 b2
and m = 0, 1, 2, . . .. In fact, when m = 0, we see that the field
configuration has the form
Bz (y) = 0 = Ez (y),
Ex (y) = 0 = By (y),
Ey (y) = ikC,
iǫµωC
Bx = − , (7.72)
c
with
√
ǫµ ω
k= . (7.73)
c
This field configuration is clearly that of a TEM wave and we see that
waves guided by two infinite parallel conducting planes can support
TEM modes unlike a hollow rectangular wave guide. Furthermore,
in this case, we see from (7.72) and (7.73) that we can write
√
B⊥ (y) = ǫµ ẑ × E⊥ , (7.74)
as we would expect for a TEM wave (see Eq. (7.19)).

Let us note here that TEM waves can also be present if the
wave guide is not completely hollow. For example, in the case of a
7.5 Wave impedance 245
coaxial cable, we have an inner conducting medium surrounded by

an external conducting surface. In such a case, the inner conductor
can carry a conduction current along the z-axis. The last equation
of Maxwell in (7.1), in this case, has the form
4πµ ǫµ ∂E
∇×B= J+ ,
c c ∂t
and the closed magnetic loops in the transverse plane can enclose the
conduction current and we do not need a z-component of the electric
field to be present. In fact, from the radial symmetry of the coaxial
cable, we see that the electric field must be radial everywhere. As
a result the “displacement” current must also be along the radial
direction. Since the conduction current is along the z-axis and the
“displacement” current along the radial direction, it follows from the
above equation that the B field cannot have a component along the
z-axis since there is no current along the φ (angular) direction. This
shows that TEM wave is the only wave that can exist in a coaxial
cable. A similar conclusion also follows for two wire transmission
lines. Furthermore, since TEM waves do not have a cut-off frequency,
coaxial cables or two wire transmission lines are used to transmit low
frequency electromagnetic waves.
7.5 Wave impedance
Let us recall that the impedance of an electromagnetic wave is related

to the proportionality constant between the transverse magnetic and
electric fields, namely,
1
B⊥ = ± ẑ × E⊥ , (7.75)
Z
where Z is known as the impedance (and the two signs correspond
respectively to forward and backward traveling waves). For example,
in free space (ǫµ = 1), we see from Eq. (7.19) that the impedance
for a TEM wave is unity. In general, in a dielectric medium the
impedance for a TEM wave is
1
ZTEM = √ . (7.76)
ǫµ
We can, similarly, determine the impedance for the TM and the TE

waves in a rectangular as well as a cylindrical wave guide. From Eqs.
(7.33), (7.43), (7.61) and (7.69), we see that the impedance for the
246 7 Wave guides
rectangular as well as the cylindrical wave guides have similar forms,

namely,
kc c k 1
ZTM = =√ √
ǫµω ǫµ ω ǫµ
r ω 2
mn
= 1− ZTEM ,
ω
√
ω ǫµ ω 1
ZTE = = √
kc c k ǫµ
1
=q ZTEM . (7.77)
ωmn 2
1− ω
Of course, the values of ωmn are different for the two wave guides.
However, for a given wave guide (for example, rectangular), we see
from the forms of the impedances in (7.77) that if the space inside is
empty (ǫµ = 1, ZTEM = 1), then we can formally write
1
ZTM = . (7.78)
ZTE
In general, though, we have

1
ZTM = Z2 . (7.79)
ZTE TEM
7.6 Attenuation factor in wave guides
Thus far, we have discussed wave guides where the external wall is
assumed to be a perfect conductor with σ → ∞. In reality, however,
the metallic conductor has a finite conductivity, be it very large.
ǫω
Normally, a metal is considered a good conductor if σ ≫ 4π (see
(8.57)). When the conductivity is finite, as in a realistic conductor,
the electromagnetic fields within the wave guide can penetrate inside
the conducting walls. The penetration depth, also known as the skin
depth, is normally very small and this phenomenon induces a surface
current in the metal that plays a very important role. Namely, it
leads to heating and Ohmic losses in the conductor. Even though
this effect is small, it leads to an attenuation of the electromagnetic
fields that are propagated inside the wave guide. This is reflected in
the fact that in the regime of propagation (namely, for ω > ωmn ), the
7.6 Attenuation factor in wave guides 247
wave vector k becomes complex of the form (see discussion in section

8.4)
k → k + iα, α > 0, (7.80)
so that all the non-vanishing components of the fields have the z

dependence of the form
E ∼ eikz−αz , B ∼ eikz−αz . (7.81)
Here, α is known as the attenuation factor. In this case, the time

averaged power transmitted along the z-axis can be obtained from
the Poynting vector as
Z Z
c
P (z) = da ẑ · S = da Re ẑ · (E × H∗ )
A 8π A
Z
c
= da Re ẑ · (E⊥ × H∗⊥ ) ∼ e−2αz , (7.82)
8π A
where A represents the cross sectional area of the wave guide per-
pendicular to the z-axis (we do not use the conventional symbol S
to avoid confusion with the Poynting vector) and we have used the
definition of the time averaged Poynting vector in (6.96)
c
S= Re (E × H∗ ) . (7.83)
8π
It now follows that
dP (z)
= −2α P (z)
dz

1 dP (z)
or, α =
2P (z) dz
Power lost per unit length
= . (7.84)
2 × Power transmitted
Thus, we see that the attenuation factor for the wave guide can be
calculated once we know the power lost as well as the power trans-
mitted during the process of propagation.
A rigorous calculation of the attenuation factor can be carried
out systematically, starting from Maxwell’s equations with modified
boundary conditions to take care of the finite conductivity of the
metal. However, an approximate calculation that gives very good
agreement with the actual results can be described as follows. Once
we know the solutions for the case of the perfectly conducting wall,
248 7 Wave guides
it is reasonable to assume that the solutions inside the wave guide

are unaffected significantly if the conducting surface has a finite but
large conductivity. This allows us to calculate the power flow along
the wave guide as (see (7.82))
Z
cZµ
P (z) = − da Re ẑ · ((ẑ × H⊥ ) × H∗⊥ )
8π A
Z
cZµ
= da |H⊥ |2 , (7.85)
8π A
where Z represents the impedance and we have used Eqs. (7.75) as
well as (7.2). As we have seen earlier in (7.77),
q 2
1 − ωmn
ω 1
ZTM = √ , ZTE = r . (7.86)
ǫµ
ωmn 2
ǫµ 1 − ω
The power loss can be calculated in the following manner. By

assumption, the fields inside the wave guide are affected only slightly
when the conductivity is finite. This is particularly true for the tan-
gential component of the magnetic field (tangential to the the wall,
Htan = n̂ × H) when the conductivity is large. Through the bound-
ary conditions, we expect the tangential component of the magnetic
field to be continuous which gives us the tangential component of the
magnetic field on the surface of the conductor. This, in turn, leads
to the induced surface current given by (See discussion in section 8.1,
in particular, Eq. (8.9) as well as section 8.4.)
4π
Js = n̂ × H = Htan . (7.87)
c
The surface current leads to heating and, consequently, to power loss.
The power loss per unit length (along the z-axis) can be calculated
in the standard manner as (see, for example, (6.83) or (6.84))
Z
Power loss per unit length = Rs ds |Js |2
surf
Z
c2 Rs
= 2
ds |Htan |2 , (7.88)
16π surf
where we have used (6.12) as well as (7.87) and (see (8.73))

r
2πµω
Rs = , (7.89)
σc2
7.7 Cavity resonators 249
is the resistive component (real part) of the surface impedance of the

conductor which has the form (see (8.72))
r
2πµω
Zs = (1 − i) . (7.90)
σc2
In (7.88), “surf” represents a surface of the conductor of unit length
along the z-axis. Note that, for a perfect conductor, σ → ∞ and
the power loss vanishes so that waves travel unattenuated. However,
when σ is finite, there is a power loss and Eqs. (7.85) and (7.88)
determine the attenuation constant through (7.84).
A concept related to the attenuation factor is known as the
quality factor or simply the Q factor of the wave guide, defined as
Energy stored per unit length

Q=ω , (7.91)
Energy lost per unit length per second
where ω is the angular frequency of the wave. We note that, by

definition,
Power transmitted = vg ×Energy stored per unit length, (7.92)
where vg denotes the group velocity of propagation. Using this, we

note that the Q factor can be written as
ω Power transmitted
Q=
vg Power lost per unit length
ω 1 1 ǫµω
= = v
vg 2α 2α c2
√
ǫµ ω
= q 2 , (7.93)
2αc
1 − ωmn
ω
where we have used Eqs. (7.37) and (7.38). Since, in wave guides,
the attenuation factor α can be very low, it is possible to construct
wave guides with large Q factors. This becomes quite important in
the construction of resonating cavities, which we will study next.
7.7 Cavity resonators
Let us next consider a rectangular wave guide of length d along the

z-axis. Furthermore, let us close the two ends of the wave guide with
perfectly conducting metal walls. Then, it is clear that, in this case,
250 7 Wave guides
the fields have to satisfy additional boundary conditions on the two

new surfaces. From (7.4), we see, for example that the electric field
components, Ex , Ey must vanish at z = 0, d. As a result, the z de-
pendence of the fields cannot have the simple exponential form, as
has been assumed in (7.13). Rather, the fields must be expanded in
terms of sin kz and cos kz, as is suitable for the appropriate bound-
ary conditions. Let us note that, for a TM wave, Bz = 0 and the
additional boundary condition required is obtained to be

∂Ez
= 0, (7.94)
∂z z=0,d
which follows from the first equation of (7.9), since (7.1) requires that
E⊥ = 0 at z = 0, d. This, then, determines the form of Ez from Eq.
(7.16) to be (compare with Eq. (7.30))
πmx πny πℓz
Ez (x, y, z) = A sin sin cos , (7.95)
a b d
with ℓ = 0, 1, 2, . . . as well as the usual restrictions on m, n and
r
πc m2 n 2 ℓ2
ω=√ + + = ωmnℓ , (7.96)
ǫµ a2 b2 d2
which follows from (7.34) with the identification k = πℓ
d . The trans-
verse field components can now be determined from the Maxwell’s
equations, (7.9), (7.10), (7.11) and (7.12), to be
1 ∂Ez
E⊥ (x) = E⊥ (x, y, z) = ǫµω 2
∇⊥ ,
π 2 ℓ2 ∂z
c2 − d2
i ǫµω
B⊥ (x) = B⊥ (x, y, z) = ǫµω 2
ẑ × ∇⊥ Ez , (7.97)
π 2 ℓ2 c
c2
− d2
which lead explicitly to
π 2 mℓ
ad πmx πny πℓz
Ex (x, y, z) = −A ǫµω 2 2 2
cos sin sin ,
− πd2ℓ a b d
c2
π 2 nℓ
bd πmx πny πℓz
Ey (x, y, z) = −A ǫµω 2 2 2
sin cos sin ,
− πd2ℓ a b d
c2
ǫµωπn
cb πmx πny πℓz
Bx (x, y, z) = −iA ǫµω 2
sin cos cos ,
π 2 ℓ2 a b d
c2 − d2
ǫµωπm
ca πmx πny πℓz
By (x, y, z) = iA ǫµω 2 2 2
cos sin cos . (7.98)
− πd2ℓ a b d
c2
It is easy to see from the explicit forms of E⊥ in (7.98) that they

satisfy the required boundary condition (7.4) at z = 0, d (namely,
they vanish at these boundaries). Parenthetically, let us sketch here
how, for example, the first relation in (7.97) will be obtained. From
(7.9) we obtain
∂Ez
∇⊥ = −∇⊥ (∇⊥ · E⊥ )
∂z
= −∇2⊥ E⊥ − ∇⊥ × (∇⊥ × E⊥ )
iω
= −∇2⊥ E⊥ − ∇⊥ × ẑBz
c

2 ǫµω 2 π 2 ℓ2
= −∇⊥ E⊥ = − 2 E⊥ , (7.99)
c2 d
which leads to the first relation in (7.97). Here we have used the
harmonic time dependence of the fields as well as Bz = 0 for TM
waves in the intermediate step. We have also used the wave equation
for E⊥ in the last step.
It is clear from Eq. (7.96) that, in such a set up, electromagnetic
fields exist only for a single frequency depending on the given values
of m, n, ℓ. This is the behavior of an undamped resonant system such
as an oscillator. As a result, such a set up is called a cavity resonator
(or a resonant cavity). The cavity can also have resonant TE modes.
For the TE modes, we note that Ez = 0 and the additional boundary
conditions have the form
Bz |z=0,d = 0, (7.100)
which follows directly from Eq. (7.4). It is now easy to determine Bz

from (7.17) (compare with (7.40))
πmx πny πℓz

Bz (x, y, z) = C cos cos sin , (7.101)
a b d
with ℓ = 1, 2, . . . as well as the usual restrictions on m, n and (see

(7.34))
r
πc m2 n 2 ℓ2
ω=√ 2
+ 2 + 2 = ωmnℓ , (7.102)
ǫµ a b d
which has the same form as (7.96).

252 7 Wave guides
The transverse field components can now be determined from

(7.11) and (7.12) to be
ω
c
E⊥ (x) = E⊥ (x, y, z) = −i ǫµω2 ẑ × ∇⊥ Bz ,
π 2 ℓ2
c2
− d2
1 ∂Bz
B⊥ (x) = B⊥ (x, y, z) = ǫµω 2
∇⊥ . (7.103)
π 2 ℓ2 ∂z
c2 − d2
Explicitly, these give

πωn
cb πmx πny πℓz
Ex (x, y, z) = −iC ǫµω 2
cos sin sin ,
π 2 ℓ2 a b d
c2 − d2
πωm
ca πmx πny πℓz
Ey (x, y, z) = iC ǫµω 2 2 2
sin cos sin ,
− πd2ℓ a b d
c2
π 2 mℓ
ad πmx πny πℓz
Bx (x, y, z) = −C ǫµω 2
sin cos cos ,
π 2 ℓ2 a b d
c2 − d2
π 2 nℓ
bd πmx πny πℓz
By (x, y, z) = −C ǫµω 2
cos sin cos .
π 2 ℓ2 a b d
c2 − d2
(7.104)
In a similar manner, we can also find the appropriate solutions

for a cylindrical cavity resonator of radius a and length d along the z
direction. From the earlier analysis of cylindrical wave guides as well
as the discussion of the rectangular cavity resonator, it is easy to see
that for the TM waves in such a resonator, we will have
Bz (r, φ, z) = 0,
πℓz
Ez (r, φ, z) = an Jn (hr) cos nφ cos , (7.105)
d
where
r
ǫµω 2 π 2 ℓ2
h= − 2 . (7.106)
c2 d
Furthermore, for the electric fields to vanish at the cylindrical walls,
we must have (see (7.57))
Jn (ha) = 0, (7.107)
which determines that (ha)mn must correspond to the m-th zero of

the n-th Bessel function. Some of these are already listed in (7.58).
In terms of these, we see from Eq. (7.106) that
r
c π 2 ℓ2
ω=√ h2mn + 2 = ωmnℓ , (7.108)
ǫµ d
with ℓ = 0, 1, 2, . . . and m, n restricted as in the case of TM waves in

the cylindrical wave guide.
Similarly, for the TE waves, we have
Ez (r, φ, z) = 0,
πℓz
Bz (r, φ, z) = bn Jn (hr) cos nφ sin , (7.109)
d
with h still defined as in (7.106). However, for TE waves, as we have
already seen in (7.65), the radial derivative of the magnetic field at
the cylindrical walls must vanish leading to
Jn′ (ha) = 0. (7.110)
This determines (ha)mn to correspond to the m-th zero of the first

derivative of the n-th Bessel function, some of which are listed in Eq.
(7.66). In terms of these, we obtain, from Eq. (7.106),
r
c π 2 ℓ2
ω=√ h2mn + 2 = ωmnℓ , (7.111)
ǫµ d
with ℓ = 1, 2, . . . and m, n restricted as in the case of TE waves in the

cylindrical wave guide. Once we have the longitudinal components
of the fields, the transverse components can be obtained using the
Maxwell’s equations as we have done earlier (or see (7.97), (7.103)).
Let us note that a cavity resonator, with a variable length (along
the z-axis) can be used as a frequency meter. Namely, by varying the
length of the cavity, one can make a signal resonate in the cavity and
thereby determine its frequency. While cavity resonators can be of
any shape, for such a purpose, the TE101 wave in a cylindrical cavity
resonator is quite useful. This is because, in this case, the radial
component of the electric field vanishes (see, for example, (7.68) with
n = 0) and the electric fields define concentric circles in a plane with
z = constant. As a result, there is no radial current and the movable
piston (or the “plunger”) does not have to make a rubbing contact
with the walls of the cavity.
254 7 Wave guides
7.8 Q factor of a cavity
As we have seen, electromagnetic waves can exist inside a cavity only

in resonant modes of discrete frequencies. (Any other field can, of
course, be expanded in terms of these resonant modes.) However,
this result is obtained by assuming that the walls of the cavity are
perfectly conducting, which is not the case in realistic cavities. As
a result, as in the case of wave guides, there are Ohmic losses and
power is lost. A consequence of this is that the resonant frequencies
are no longer sharp, rather they spread out a little. The Q factor of
the cavity gives a measure of this spread and is defined as
Energy stored in a cavity
Q = ωr , (7.112)
Power lost
where ωr represents a resonant frequency (of the original lossless sys-
tem). Let us note that if U denotes the energy stored in a cavity,
then the power lost is the energy lost per unit time. Thus, from the
definition (7.112), it follows that
dU ωr
=− U
dt Q
− ωQr t
or, U (t) = U (0) e . (7.113)
This is clearly defined only for positive times and the energy inside
the cavity decreases as time evolves due to the losses at the walls of
the cavity (as well as possible losses in the dielectric inside).
It is clear from this that, since there is energy loss in a cavity,
if we want to excite a particular resonant mode in the cavity by, say,
introducing an external electromagnetic wave into the cavity through
a small aperture, the system would behave like a damped oscillator
with a time dependent driving force. From the form of the energy U
in (7.113), we note that we can write the form of the electric fields
inside the cavity to have the form
ωr
−i(ω̄r −i 2Q )t
E(t) = E(0) e , (7.114)
where we have suppressed the spatial dependence of the fields for

simplicity and have defined
ω̄r = ωr + δω.
The quantity, δω, has been introduced to account for the possible
smearing of the resonant frequency due to other effects. Taking the
7.8 Q factor of a cavity 255
Fourier transform of (7.114), we obtain

Z ∞ Z
1 E(0) ∞ ωr
E(ω) = dt eiωt E(t) = dt ei(ω−ω̄r +i 2Q )t
2π 0 2π 0
iE(0) 1
= ωr . (7.115)
2π (ω − ω̄r ) + i 2Q
It follows now that
|E(0)|2 1
|E(ω)|2 = 2 ωr2
. (7.116)
4π (ω − ω̄r )2 + 4Q2
This has the characteristics of a resonant behavior (Breit-Wigner

shape) and shows that the electric field no longer has a sharp, discrete
frequency. Rather, it is smeared out around the resonant value ω =
ω̄r as shown in Fig. 7.4. From (7.116), we see that the intensity has
the maximum value at ω = ω̄r . It decreases to half of its maximum
value at
|E(0)|2
4π 2
| |
ω̄r ωr ω
2Q
Figure 7.4: The behavior of the absolute square of the electric field
as a function of the frequency ω.
ωr
ω = ω̄r ± . (7.117)
2Q
Consequently, the width of the curve at half the peak value is obtained
to be
ωr
Γ = ∆ω = ,
Q
ωr
or, Q = . (7.118)
Γ
256 7 Wave guides
This shows that the Q factor, indeed, measures the spread out in the
frequency due to Ohmic losses.
As in the case of the wave guide, the power loss at the walls can
be calculated in a simple manner and, for microwave cavities, one can
obtain a Q factor as large as ten thousand. This implies that one can
construct cavities that can show very sharp resonant behavior.
7.9 Dielectric wave guides (optical fibers)
The perfectly conducting walls of a wave guide basically do not al-

low the electromagnetic waves to escape from the wave guide. As
a result, the waves bounce back and forth at the walls and prop-
agate along the axis of the wave guide. We have also seen earlier
that, while electromagnetic waves can be transmitted through a di-
electric boundary surface, under appropriate conditions, there can
be total internal reflection (without any transmission). Thus, under
appropriate conditions, a dielectric slab can also be used to guide
waves through successive total internal reflections, much like wave
guides with perfectly conducting walls. The first analysis of such a
phenomenon was carried out by Debye in 1910.
When the frequency of the wave lies in the optical or in the in-
frared range, such a transmission line is known as an optical fiber and
is extremely important in reliable, high speed telecommunications. A
common optical fiber consists of a silicon fiber of small dimensions
(of the order ∼ 50 − 100µm) and an optical fiber cable, typically of
the order of a few centimeters, contains many such fibers allowing
for multi-mode transmissions. Technology in optical fibers has ad-
vanced tremendously over the last couple of decades which signifies
the importance of this. Of course, the conditions in optical fibers
cannot always be met perfectly so as to have complete total internal
reflection. As a result, a small amount of the signal leaks through,
leading to losses much like in the metallic wave guides. Nonetheless,
this form of transmission is quite important considering that the fre-
quencies involved are large thereby allowing for large bandwidths for
transmission.
Although, in practice, one uses cylindrical dielectric wave guides
(optical fibers), the mathematical analysis of such systems is rather
involved. Therefore, just to get a feeling for the qualitative properties
of such a system, let us analyze a rectangular dielectric slab of width
−a ≤ x ≤ a which we choose to be of infinite extension in the y direc-
tion (see Fig. 7.5), for simplicity. As before, we will assume the wave
to propagate along the z-axis. Because of the infinite extension along
7.9 Dielectric wave guides (optical fibers) 257
the y direction, translations along this direction define a symmetry

and, consequently, it follows that the fields cannot depend on the y
coordinate. Therefore, we can choose the electric and the magnetic
fields to have the forms
x=a
x = −a
Figure 7.5: An infinite rectangular dielectric slab with a finite width

along the x-axis and wave propagation along the z-axis.
E(x, t) = E(x) e−i(ωt−kz) ,
B(x, t) = B(x) e−i(ωt−kz) . (7.119)
Substituting these into (7.16)–(7.17), we obtain

d2 Ez (x) ǫµω 2 2
+ − k Ez (x) = 0,
dx2 c2

d2 Bz (x) ǫµω 2 2
+ − k Bz (x) = 0, (7.120)
dx2 c2
while Eq. (7.15) leads to
ik dEz (x)
Ex (x) = ǫµω 2
,
− k2 dx
c2
iω dBz (x)
Ey (x) = − ǫµω2 c ,
−k 2 dx
c2
ik dBz (x)
Bx (x) = ǫµω 2
,
− k2 dx
c2
258 7 Wave guides
i ǫµω
c dEz (x)
By (x) = ǫµω 2
. (7.121)
− k2 dx
c2
These equations must hold in both the regions – inside and out-
side the dielectric slab. Let us assume that ǫ1 , µ1 represent the per-
mittivity and the permeability inside the dielectric (−a < x < a)
while ǫ2 , µ2 represent the same quantities outside (|x| > a). Because
of the symmetry in the problem under x ↔ −x, the solutions can
be classified into even and odd ones. Furthermore, we would like the
solutions to be exponentially damped outside the slab while oscilla-
tory inside. With this, we see that we can have four different kinds of
solutions – TM even, TM odd, TE even and TE odd. Let us simply
work out one of the solutions in detail.
Let us consider the TM even solutions. In this case, we have
Bz = 0. Furthermore defining, in the two regions,
r r
ǫ1 µ 1 ω 2 2
ǫ2 µ 2 ω 2
α= 2
−k , γ = k2 − , (7.122)
c c2
where we assume that both α, γ are real and positive, we see that
ω2 ω2
α2 + γ 2 = (ǫ1 µ1 − ǫ2 µ2 ) = ∆ . (7.123)
c2 c2
Furthermore, we note that because of the symmetry of the solutions,
we can restrict ourselves to the positive x-axis only (x ≥ 0). Using
(7.122) and (7.123), we obtain from Eqs. (7.120) and (7.121) (for
x ≥ 0)
(1) (2)
Ez (x) = A cos αx, Ez (x) = C e−γx ,
(1) (2)
Ex (x) = −iA αk sin αx, Ex (x) = iC k
γ e−γx ,
(1) (2)
Ey (x) = 0, Ey (x) = 0,
(1) (2)
Bx (x) = 0, Bx (x) = 0,
(1) (2)
By (x) = −iA ǫ1cα
µ1 ω
sin αx, By (x) = iC ǫ2 µ2 ω
cγ e−γx ,
(7.124)
where the superscripts (1), (2) denote the two regions 0 ≤ x ≤ a and
x ≥ a respectively (because of our choice x ≥ 0).
Since these are even solutions, we can apply the boundary con-
ditions only at x = a (the conditions will then automatically hold at
the boundary x = −a by symmetry). We have already worked out
7.9 Dielectric wave guides (optical fibers) 259
the boundary conditions at the interface of two dielectric surfaces

(see (6.39), (6.42) and (6.43)) and, in the present case, they take the
forms
ǫ1 Ex(1) (a) = ǫ2 Ex(2) (a),
Bx(1) (a) = Bx(2) (a),

(1) (2)
Etan (a) = Etan (a),
1 (1) 1 (2)
Btan (a) = B (a). (7.125)
µ1 µ2 tan
In the present case, since Ey , Bx (and Bz ) are zero in the two regions,
we obtain from the third relation in (7.125) that
Ez(1) (a) = Ez(2) (a)

or, A cos αa = C e−γa . (7.126)
Similarly, the last relation in (7.125) leads to
1 (1) 1 (2)
B (a) = B (a)
µ1 y µ2 y
ǫ1 ǫ2
or, A sin αa = −C e−γa . (7.127)
α γ
It is easy to see that the first relation in (7.125) leads to the same
relation as (7.127) and, therefore, (7.126) and (7.127) represent essen-
tially the boundary conditions that need to be satisfied. (The second
relation in (7.125) is trivially satisfied.)
Taking the ratio of Eqs. (7.126) and (7.127), we obtain
α γ
cot αa = −
ǫ1 ǫ2
ǫ1
or, α cot αa = − γ. (7.128)
ǫ2
Using (7.123) and defining dimensionless variables ξ = αa, η = γa,

we can write (7.128) also as
r
ǫ1 ǫ1 ∆ω 2 a2
ξ cot ξ = − η = − − ξ2. (7.129)
ǫ2 ǫ2 c2
This is a transcendental equation for ξ (or α) as a function of

ω and the solutions can be obtained graphically (see Fig. 7.6), much
260 7 Wave guides
η
3
0
0 1 2 3 4 5 6
ξ
Figure 7.6: The TM even solutions are obtained from the intersections
of η = − ǫǫ21 ξ cot ξ and η 2 + ξ 2 = constant.
like in the case of the square well potential in quantum mechanics. In

fact, let us note that we can analyze the other modes as well exactly
in this manner and the boundary conditions, in each case, would yield
r
ǫ1 ∆ω 2 a2
TM (odd) : ξ tan ξ = − ξ2,
ǫ2 c2
r
µ1 ∆ω 2 a2
TE (even) : ξ cot ξ = − − ξ2,
µ2 c2
r
µ1 ∆ω 2 a2
TE (odd) : ξ tan ξ = − ξ2, (7.130)
µ2 c2
with α, γ satisfying (7.122) and (7.123). Thus, in each case, we have a

transcendental equation to solve. Much like in quantum mechanical
potential problems, here, too, one finds that there is a cut off fre-
quency below which there does not exist any solution. Beyond this,
there is a range of values of the frequency for which there exists only
one solution and so on. In this respect, it is quite reminiscent of the
square well potential in quantum mechanics and this behavior is also
qualitatively similar to that of metallic wave guides.
1. Consider a pair of perfectly conducting parallel plates of infinite

dimensions along the y and the z axes, separated by a distance
“a” along the x-axis. Determine all the possible solutions for
an electromagnetic wave to propagate along the z-axis.
2. Calculate the attenuation factor for the TEM waves in the set
up described in the previous question.
3. A rectangular wave guide (with perfectly conducting walls) has

cross-sectional dimensions
a = 7cm, b = 4cm
Determine all the modes which will propagate at a frequency

of (a) 3000MHz, (b) 5000MHz.
4. If λ and λ̄ represent the wavelengths of the same electromag-

netic wave in vacuum and inside a wave guide respectively, then,
show that
λ̄λc
λ= p ,
λ̄2 + λ2c
where λc denotes the wavelength corresponding to the cut-off

frequency ωc of the wave guide.
5. Derive the transcendental equations (discussed in this chapter)

determining the TM (odd), TE (even) and the TE (odd) solu-
tions for a wave traveling between two parallel dielectric slabs.
6. A wave guide of right triangular cross section is bounded by

perfectly conducting walls at x = a, y = 0, x = y. Find the
cut-off frequencies and the field modes of such a wave guide.
Chapter 8
Propagation through a conducting

medium
We have so far discussed the propagation of electromagnetic waves in

vacuum or in dielectric media. Let us next analyze the propagation
of such waves through a conducting medium. Of course, the essential
change, in this case, lies in the boundary conditions. If we look
at the Maxwell’s equations in an arbitrary medium in the Fourier
transformed space (in the time variable), they take the forms
∇ · (ǫE) = 4πρ,
∇ · B = 0,
∇ × E = ik B,

1 4π 4πσ
∇× B = J − ikǫ E = E − ikǫ E, (8.1)
µ c c
ω
where we have defined, as before, k = c and have identified
J = σ E, (8.2)
with σ representing the conductivity of the medium. We note that

σ, ǫ and µ are, in general, dependent on the frequency and although
for most non-magnetic materials we can set µ ≈ 1, we will not do
so for completeness. Here all the field variables have the coordinate
dependence φ(x, ω), since time has been Fourier transformed. Alter-
natively one can also think of these equations as those for fields with
a harmonic time dependence (E(t), B(t)) = (E(0), B(0))e−iωt with
the exponential factored out.
The first two equations of Maxwell lead to the fact that, across
a boundary with no free charges, the normal components of both ǫE
263
264 8 Propagation through a conducting medium
n̂
∆ℓ
∆h
Figure 8.1: Boundary condition for the tangential component of the

electric field.
and B should be continuous. In the presence of free surface charges,

the normal component of the D field will be discontinuous across the
surface, the discontinuity being proportional to the surface charge
density. This is what we have already seen in the static case. Taking
an infinitesimal surface element as shown in Fig. 8.1, we can deduce
from the third equation that the tangential component of the electric
field must be continuous across the boundary.
Z Z
ds · (∇ × E) = ik ds · B
I Z
or, dℓ · E = ik ds · B. (8.3)
In the limit of an infinitesimal surface with area going to zero, this

leads to
n̂ × E1 | = n̂ × E2 | , (8.4)
where n̂ represents a unit vector normal to the boundary and the re-
striction represents the boundary. If the second medium corresponds
to a perfect conductor, then there is no electric field in the second
medium. Consequently, in such a case, we obtain the boundary con-
dition to be
n̂ × E| = 0. (8.5)
Similarly, taking an infinitesimal surface element across the bound-

ary, we obtain from the last equation in (8.1),

Z Z
1 4π
ds · ∇ × B = ds · J − ikǫ E
µ c
I Z
1 4π
or, dℓ · B = ds · J − ikǫ E . (8.6)
µ c
For non-singular field configurations, the right hand side of (8.6) van-
ishes in the limit of vanishing surface area and the boundary condition
takes the form

1 1
n̂ ×
B1 = n̂ × B2 . (8.7)
µ1 µ2
When the second medium is a perfect conductor, the second term on

the right hand side of (8.6) still vanishes in the limit of a vanishing
surface area. However, in the presence of time dependent incident
electromagnetic fields, a perfect conductor develops singular surface
currents so that the first term on the right hand side of (8.6) does not
vanish. (What this means is that the surface currents exist only on
the surface and, therefore, are described with delta functions which
are singular and may not yield a vanishing contribution even in the
limit that the surface area vanishes.)
n̂
t̂
Js
Figure 8.2: The induced surface current on a conductor with n̂ and

t̂ denote respectively the normal to the surface and the direction of
∆ℓ.
In fact, let us define the surface current Js as the amount of

charge crossing the Gaussian surface per second per unit length along
the long arm of the surface (see Fig. 8.2). In this case, in the limit
of vanishing surface area, (8.6) leads to

1 1 4π
n̂ × B1 − B2 ∆ℓ = Js ∆ℓ
µ1 µ2 c

1 1 4π
or, n̂ ×
B1 = n̂ × B2 + Js . (8.8)
µ1 µ2 c
Since there is no magnetic field inside a conductor, this condition can

also be written as

1 4π
n̂ × B = Js
µ c
4π
or, n̂ × H| = Js . (8.9)
c
The important thing to note from this relation is that it relates the
induced surface current in the conductor to the magnetic field outside.
It is, therefore, not a boundary condition. Rather, once all the fields
are determined, this relation can be used to calculate the induced
surface current.
8.2 Reflection from a perfect conductor at normal incidence
Let us consider a plane wave of frequency ω traveling along the z-

axis and let it be incident on a perfectly conducting surface located
at z = 0 as shown in Fig. 8.3.
z=0
Figure 8.3: Reflection of a wave from a perfectly conducting surface

at z = 0 at normal incidence.
8.2 Reflection at normal incidence 267
We assume the first medium to be vacuum for simplicity. Thus,

in the region z < 0, we expect an incident wave as well as a reflected
wave whereas, since the second medium is a perfect conductor, we do
not expect any transmitted wave in this case. (This is different from
normal incidence in the case of a dielectric medium.) Thus, in region
z < 0 we have
E(x, ω) = Einc (x, ω) + Erefl (x, ω), (8.10)
where the incident and the reflected waves have the coordinate de-
pendence given by
(0)
Einc (x, ω) = Einc eikz ,
(0)
Erefl (x, ω) = Erefl e−ikz , (8.11)
with k = ωc . (We have factored out the time dependence for simplic-
ity.)
The electric and the magnetic fields, in a plane wave, are or-
thogonal to the direction of propagation (as well as to each other)
and, therefore, are along the surface of the boundary (tangential to
the surface). As we see from the boundary condition for a perfect
conductor in (8.5),
E(x, y, z = 0, ω) = 0, (8.12)
which determines
(0) (0)
Erefl = −Einc = −E(0) , ẑ · E(0) = 0. (8.13)
Namely, the incident and the reflected electric fields have the same
amplitudes, but are out of phase. Thus, we can write
E(x, ω) = Einc (x, ω) + Erefl (x, ω)
= E(0) (eikz − e−ikz ) = 2iE(0) sin kz. (8.14)
Putting in the harmonic time dependence (and recalling that the

fields are really defined as the real part of the complex quantities),
we obtain the form of the electric field for z < 0 to be

E(x, t, ω) = Re 2iE(0) sin kz e−iωt
= 2E(0) sin kz sin ωt. (8.15)

Once we have the electric fields, the magnetic fields can be ob-
tained through the use of the relation (see (7.76) in vacuum)
B = ±ẑ × E, (8.16)
where the ± signs are related to the direction of propagation of the

wave. Thus, using (8.16), we obtain

Binc (x, ω) = ẑ × Einc (x, ω) = ẑ × E(0) eikz ,

Brefl (x, ω) = −ẑ × Erefl (x, ω) = ẑ × E(0) e−ikz . (8.17)
As a result, we obtain the total magnetic field in the region z < 0 to

be
B(x, ω) = Binc (x, ω) + Brefl (x, ω)

= ẑ × E(0) eikz + e−ikz

= 2 ẑ × E(0) cos kz. (8.18)
Putting in the time dependence, we obtain

B(x, t, ω) = Re 2(ẑ × E(0) ) cos kz e−iωt

= 2 ẑ × E(0) cos kz cos ωt. (8.19)
There are several things to note from Eqs. (8.14) and (8.18) (or
(8.15) and (8.19)). The unique solution to the problem is obtained
using only the boundary condition (8.5) (namely, we do not need (8.9)
to obtain the solution). Furthermore, the electric and the magnetic
fields define standing waves with the electric field vanishing at (z < 0)
kz = −nπ, n = 0, 1, 2, . . . , (8.20)
while the magnetic field vanishes at
(2n + 1)π
kz = − , n = 0, 1, 2, . . . . (8.21)
2
The electric field vanishes at the boundary z = 0 simply because the
incident and the reflected electric fields are out of phase. However, we
note that the magnetic field does not vanish at the boundary. This
8.3 Reflection at oblique incidence 269
can, in fact, be physically understood as follows. The (time averaged)

radiated power per unit area, in the region z < 0, is obtained to be
dP c
= Re ẑ · (E × B∗ )
da 8π
c h
= Re ẑ · E(0) × ẑ × (E(0) )∗
8π
i
× eikz − e−ikz e−ikz + eikz
= 0. (8.22)
The radiated power vanishes simply because the incident and the
reflected components of the magnetic field are in phase while those
for the electric field are out of phase. Physically, this is clear since
there cannot be any transmitted wave inside a perfect conductor and,
therefore, there cannot be any power loss.
Let us note that the non-vanishing of the magnetic field, at the
boundary, immediately shows that there must be a surface current
in the conductor. In fact, from (8.9), we conclude that (µ = 1 and
ẑ · E(0) = 0)
4π
Js = n̂ × B| = − ẑ × B|
c

= −2ẑ × ẑ × E(0) = 2E(0) ,
c (0)
or, Js = E . (8.23)
2π
This can, therefore, be thought of as the reason for the change in the
phase of the reflected component of the electric field.
8.3 Reflection from a perfect conductor at oblique incidence
Let us next consider the reflection of a plane wave of frequency ω from

a perfectly conducting surface at oblique incidence as shown in Fig.
8.4. Let us assume that the boundary surface is at z = 0 and that the
region z < 0 corresponds to vacuum. Without loss of generality, we
can assume the plane of incidence (the normal to the surface and the
direction of propagation define the plane of incidence) to be the x − z
plane. Let θi be the angle of incidence (namely, the angle between
the incident ray and the z-axis).
Therefore, we can write the incident wave in region z < 0 to be
of the form
(0) (0)
Einc (x, ω) = Einc eiki ·x = Einc eiki (sin θi x+cos θi z) , (8.24)
x
θi θr
Figure 8.4: Reflection of a wave from a perfectly conducting surface

at z = 0 at oblique incidence.
where
ki = ki (sin θi x̂ + cos θi ẑ), k̂i = x̂ sin θi + ẑ cos θi . (8.25)
In the region z < 0, we also expect a reflected wave. However, since

the region z > 0 is perfectly conducting, we do not expect a trans-
mitted wave in this region. We can write the reflected wave to be of
the form
(0) (0)
Erefl (x, ω) = Erefl eikr ·x = Erefl eikr (sin θr x−cos θr z) , (8.26)
where
kr = kr (x̂ sin θr − ẑ cos θr ), k̂r = (x̂ sin θr − ẑ cos θr ). (8.27)
Therefore, the total electric field in this region has the form

(0) (0)
= Einc eiki (sin θi x+cos θi z) + Erefl eikr (sin θr x−cos θr z) . (8.28)
The boundary condition (8.5) now leads to
n̂ × E(x, ω)|z=0 = 0,

(0) (0)
or, ẑ × Einc eiki sin θi x + ẑ × Erefl eikr sin θr x = 0. (8.29)
Clearly, this is satisfied if
ki = kr = k,
θi = θr = θ,

(0) (0)
ẑ × Einc + Erefl = 0. (8.30)
This shows that the wave numbers for the incident as well as the
reflected waves are the same (they are in the same region) and that
the angle of incidence is equal to the angle of reflection, as is also
true in the case of reflection from a dielectric surface. However, here
we have no transmission.
The last relation in (8.30) allows for two possibilities. First, we
can have the electric fields along the y-axis - perpendicular to the
plane of incidence. In this case, for the last relation in (8.30) to hold,
we must have
(0) (0)
Erefl = −Einc = −E (0) ŷ. (8.31)
It follows now that
E(x, ω) = Einc + Erefl
= 2iE (0) ŷ eikx sin θ sin(kz cos θ). (8.32)
Furthermore, using (8.25) and (8.27), we obtain
Binc (x, ω) = k̂i × Einc (x, ω)
= (x̂ sin θ + ẑ cos θ) × ŷ E (0) eik(x sin θ+z cos θ)
= (−x̂ cos θ + ẑ sin θ) E (0) eik(x sin θ+z cos θ) ,
Brefl (x, ω) = k̂r × Erefl (x, ω)
= (x̂ sin θ − ẑ cos θ) × (−ŷ) E (0) eik(x sin θ−z cos θ)
= −(x̂ cos θ + ẑ sin θ) E (0) eik(x sin θ−z cos θ) , (8.33)
= 2E (0) eikx sin θ [−x̂ cos θ cos(kz cos θ)

+iẑ sin θ sin(kz cos θ)] . (8.34)
Thus, in this case, we see that while the electric field is perpendicular
to the plane of incidence, the magnetic field is parallel to it (lies in
the plane of incidence). Furthermore, we have
B(x, y, z = 0, ω) = −2E (0) x̂ cos θ eikx sin θ 6= 0, (8.35)
so that the tangential component of the magnetic field does not vanish
at the boundary. This, therefore, leads to a surface current (see (8.9))
of the form
c c
Js = n̂ × B| = ( −ẑ × B|)
4π 4π
cE (0)
= ŷ cos θ eikx sin θ . (8.36)
2π
Furthermore, the time averaged power per unit area is obtained to

be
dP c
= Re ẑ · (E × B∗ )
da 8π
c
= Re 2i|E (0) |2 cos θ sin(2kz cos θ)
8π
= 0. (8.37)
Namely, since there is no transmission, all the energy that is radiated

in the forward direction is reflected back leading to a net zero energy
loss.
The other possibility that one can have is that the electric field
lies in the plane of incidence. Of course, it has to be orthogonal to
the direction of propagation. Therefore, we can choose
(0)
Einc = E (0) (x̂ cos θ − ẑ sin θ),
(0)
Erefl = −E (0) (x̂ cos θ + ẑ sin θ). (8.38)
This satisfies the last relation in (8.30) and leads to
= 2E (0) eikx sin θ [ix̂ cos θ sin(kz cos θ)

−ẑ sin θ cos(kz cos θ)] . (8.39)
In turn, using (8.25) and (8.27), this leads to
Binc (x, ω) = k̂i × Einc (x, ω)
= ŷ E (0) eik(x sin θ+z cos θ) ,
Brefl (x, ω) = k̂r × Erefl (x, ω)
= ŷ E (0) eik(x sin θ−z cos θ) . (8.40)
As a result, we obtain
= 2E (0) ŷ eikx sin θ cos(kz cos θ). (8.41)
In this case, we note that the magnetic field is orthogonal to the plane
of incidence while the electric field is parallel to it.
Let us note that
B(x, y, z = 0, ω) = 2E (0) ŷ eikx sin θ 6= 0, (8.42)
so that the tangential component of the magnetic field does not vanish
on the boundary. Consequently, it leads to a surface current of the
form
c c
Js = n̂ × B| = ( −ẑ × B|)
4π 4π
cE (0)
= x̂ eikx sin θ . (8.43)
2π
We can also calculate the power radiated per unit area in region
z < 0, which takes the form
dP c
= Re ẑ · (E × B∗ )
da 8π
c
= Re 2i|E (0) |2 cos θ sin(2kz cos θ)
8π
= 0. (8.44)
This shows that all the energy that is incident is reflected back and
there is no net loss of energy.
Finally, let us consider, for simplicity, the first solution (where
the electric field is orthogonal to the plane of incidence) and note
some qualitative features which hold for both the solutions. We see,
from Eq. (8.32) that the electric field forms standing waves along the
z-axis with nodes at (z < 0)
kz cos θ = −nπ, n = 0, 1, 2, . . . . (8.45)
The locations of the nodes depend on the angle of incidence and, in

particular, for normal incidence (θ = 0), we recover the conditions
in (8.20). Furthermore, from (8.32), we also note that this entire
standing wave travels along the x-axis with a wave number
k̄ = k sin θ. (8.46)
Namely, the wave length of propagation along the x-axis is given by
λ
λ̄ = , (8.47)
sin θ
corresponding to a propagation velocity
ω c
v̄ = = . (8.48)
k̄ sin θ
Thus, we see that the wave length as well as the velocity of propa-
gation are larger than their corresponding values in vacuum. This is
very similar to the behavior we have seen in wave guides. In fact,
suppose we add a parallel, perfectly conducting surface at z = −b,
then, from the solution for the electric field in (8.32), we see that
this has to satisfy the boundary condition (8.5) at the new boundary
leading to
kb cos θ = nπ, (8.49)
which gives
p
k̄ = k sin θ = k 2 − k 2 cos2 θ
r
ω 2 π 2 n2
= − 2 . (8.50)
c2 b
This shows that for ω < πcnb , there will be no propagation, while for
ω > πcn
b , there will be propagation of waves. This is, in fact, exactly
what we have seen for a wave guide. However, here we only have a
pair of parallel conducting surfaces. As we have discussed earlier, a
pair of parallel conducting plates can also guide waves and, among
other things, has a TEM mode of propagation.
8.4 Reflection from a good conducting surface 275
8.4 Reflection from a good conducting surface
We have, thus far, talked about perfect conductors for which σ → ∞.

In reality, the conductors have a large but finite conductivity. The
reflection of an electromagnetic wave from a real conducting surface
will, therefore, be different from the earlier results. In this section,
we will consider the reflection of a harmonic plane wave from a good
(but not perfect) conductor. A conductor is commonly known as a
good conductor if 4πσǫω ≫ 1, as we will see shortly.
Before analyzing reflection from a good conductor, let us discuss
some of the properties of harmonic fields. For harmonic fields (that
is when all the field variables have a harmonic time dependence of
the form e−iωt of a given frequency), the Maxwell’s equations take
the forms
∇ · D = 4πρ,
∇ · B = 0,
iω
∇×E= B,
c

1 4π iω 4πσ iωǫ
∇× B = J− D= − E. (8.51)
µ c c c c
We have used here the relation
J = σ E, (8.52)
as well as
D = ǫE, (8.53)
where ǫ represents the permittivity of the medium. Although, for

most non-magnetic material, the permeability µ ≈ 1, we will keep it
in our calculations for completeness.
The important thing to observe from (8.51) is that the first two
equations follow as a consequence of the last two and the continuity
equation. Furthermore, it follows from the continuity equation that
∂ρ
+∇·J=0
∂t
∂ρ
or, + σ∇ · E = 0, (8.54)
∂t
where we have assumed that the conductivity does not vary with
space. Furthermore, using the first of the Maxwell’s equations and
assuming that the permittivity also is independent of space (this is

true when we are considering an isotropic medium), we obtain
∂ρ 4πσ
= −σ∇ · E = − ρ,
∂t ǫ
4πσt
or, ρ(t) ∼ e− ǫ . (8.55)
This shows that the charge density dissipates with a time scale
ǫ
τ= . (8.56)
4πσ
On the other hand, the only meaningful time scale in a harmonic
problem is ω1 and so, we conclude that if
1
τ≪
ω
4πσ
or, ≫ 1, (8.57)
ǫω
the charge will dissipate quite rapidly and this is what defines a good
conductor. In such a medium, therefore, we can set the charge density
to zero. Note, from (8.55), that for a perfect conductor, σ → ∞ so
that there is no charge density produced.
Returning to the Maxwell’s equations, (8.51), and taking the
curl of the last two equations, we note that they lead to

2 ǫµω 2 4πiσ
∇ E+ 2 1+ E = 0,
c ǫω

ǫµω 2 4πiσ
∇2 B + 1+ B = 0. (8.58)
c2 ǫω
These are complex equations which we can think of as the wave equa-
tions that we have studied earlier, if we allow for a complex permit-
tivity of the form

4πiσ 4πiσ
ǫeff = ǫ 1 + =ǫ+ . (8.59)
ǫω ω
Since σ = 0 in vacuum, such a definition is compatible with the

vacuum solutions. The equations (8.58) have solutions of the forms
E ∼ eiK·x , B ∼ eiK·x , (8.60)

provided

2 ǫµω 2
2 4πiσ
K =K = 2 1+ . (8.61)
c ǫω
This shows that in a conducting medium, the wave number, in gen-
eral, becomes complex with a real and an imaginary part. In partic-
ular, if we have a good conductor, 4πσ
ǫω ≫ 1, we obtain
√ r √
ǫµω 4πiσ 2πσµω
K≈ = (1 + i) = kr + iki , (8.62)
c ǫω c
where we have used
√ (1 + i)
i= √ ,
2
and have identified
√
2πσµω
kr = ki = . (8.63)
c
Thus, in a good conductor, the solutions of the Maxwell’s equa-
tions traveling along the z-axis, for example, lead to electric and
magnetic fields of the forms
E(x, ω) = E(0) eikr z e−ki z ,
B(x, ω) = B(0) eikr z e−ki z . (8.64)
This shows that the wave attenuates as it travels because of dissipa-

tion in the medium and the length
1 c
δ= =√ , (8.65)
ki 2πσµω
is known as the penetration depth or the skin depth of the medium
beyond which the amplitude of the wave becomes negligible. As a
result, it follows that the fields and the currents cling close to the
surface of a good conductor. For such a wave, it is easy to see from
the first two of the Maxwell’s equations that
Ez = 0 = Bz , (8.66)
and the third of Maxwell’s equations gives

c cK
B= (kr + iki ) ẑ × E = ẑ × E, (8.67)
ω ω
which allows us to identify the impedance for such a TEM wave to

be
r
ω ω ω
Z= = = (1 − i) . (8.68)
c(kr + iki ) cK 8πσµ
This can be identified with
r
(1 − i) ǫω 1
Z= √ =√ ≪ 1, (8.69)
2ǫµ 4πσ ǫeff µ
for a good conductor. We see that the impedance, in this case, is

complex and, as a result, the refractive index for such a medium also
becomes complex, namely,
r
1 Kc 2πσµ
n= = = (1 + i) , (8.70)
Z ω ω
which signals absorption by the medium.
Since the electric fields fall off with large values of z, it follows,
from (8.52), that the current must also have a similar behavior. This,
therefore, allows us to define a surface current density as
Z ∞
Js = dz J(x, ω)
0
Z ∞
(0)
= σE dz eiKz
0
iσ E(0) iσ
= = E| , (8.71)
K K
where the restriction refers to the surface assumed to be at z = 0. In
analogy with electric circuits, the ratio of the tangential component
of the electric field at the surface to the surface current density is
defined as the surface impedance of the medium, namely,
E| = Zs Js ,
r
iK 2πµω 1
or, Zs = − = (1 − i) = (1 − i) , (8.72)
σ σc2 σδ
where δ represents the skin depth defined in (8.65). The real part of
the surface impedance can be thought of as the surface resistance of
the medium,
r
1 2πµω
Rs = = . (8.73)
σδ σc2
Note that when σ → ∞, the surface resistance vanishes (as does the
surface impedance). It is now straightforward to see from (8.68) and
(8.72) that
c c
Z= Zs = (1 − i) . (8.74)
4πµ 4πµσδ
With these, we are now ready to analyze reflection from a good
conducting surface. Let us assume that the conducting surface is at
z = 0 and that an electromagnetic wave is incident from vacuum in
z < 0 at an angle of incidence θi in the x − z plane as shown in Fig.
8.5. In this case, we do expect a reflected wave in the region z < 0
with angle of reflection θr as well as a transmitted wave in the region
z > 0 with the angle of transmission θt . Of course, the transmitted
wave will be highly attenuated and we assume that the thickness of
the conducting medium is much larger than the skin depth so that
the transmitted wave will practically be a surface wave. We can
parameterize the electric fields associated with the three components
as
θt
x
θi θr
Figure 8.5: Reflection from a good conducting surface at z = 0 at

oblique incidence.
Einc (x, ω) = Ei eiki (x sin θi +z cos θi ) ,
Erefl (x, ω) = Er eikr (x sin θr −z cos θr ) ,
Etrans (x, ω) = Et eikt (x sin θt +z cos θt ) , (8.75)
where we have used the results from our earlier analysis that the
propagation has to be in the same plane and have defined here
ki = ki (x̂ sin θi + ẑ cos θi ) = ki k̂i ,
kr = kr (x̂ sin θr − ẑ cos θr ) = kr k̂r ,
kt = kt (x̂ sin θt + ẑ cos θt ) = kt k̂t . (8.76)
We recall that the wave number kt inside the conducting medium is

complex.
In this case, the boundary condition (8.4) takes the form
(ẑ × Ei )eiki x sin θi + (ẑ × Er )eikr x sin θr = (ẑ × Et )eikt x sin θt . (8.77)
For this to be true, we must have
θi = θr = θ,
ki = kr = k,
k
sin θt = sin θ
kt
ẑ × Et = ẑ × (Ei + Er ). (8.78)
We note that the first of these relations tells us the familiar result
that the angle of reflection is the same as the angle of incidence while
the third gives Snell’s law, namely
sin θt k 1
= = = Zt , (8.79)
sin θ kt nt
where nt represents the index of refraction of the conducting medium
and we have used k = ωc as well as (8.70). The important thing to note
here is that, in this case, the index of refraction is complex signifying
dissipation (absorption) of energy in the medium. Furthermore, for
a good conductor (see (8.69)),
Zt ≪ 1, (8.80)
so that we expect
θt ≈ 0. (8.81)
To continue with the analysis, let us assume that the incident

electric field is orthogonal to the plane of incidence, namely, it has a
component only along the y-axis. It follows, then from the boundary
conditions, that all the electric fields will have the same polarization,
namely,
Einc (x, ω) = Ei ŷ eik(x sin θ+z cos θ) ,
Erefl (x, ω) = Er ŷ eik(x sin θ−z cos θ) ,
Etrans (x, ω) = Et ŷ eikt (x sin θt +z cos θt ) . (8.82)
Furthermore, we can obtain the components of the magnetic fields as

1
Binc (x, ω) = k̂i × Einc
Z
= (−x̂ cos θ + ẑ sin θ) Ei eik(x sin θ+z cos θ) ,
1
Brefl (x, ω) = k̂r × Erefl
Z
= (x̂ cos θ + ẑ sin θ) Er eik(x sin θ−z cos θ) ,
1
Btrans (x, ω) = k̂t × Etrans
Zt
1
= (−x̂ cos θt + ẑ sin θt ) Et eikt (x sin θ+z cos θ) . (8.83)
Zt
Here we have used Z = 1 in vacuum. Matching the tangential com-
ponents of the electric and the magnetic fields across the boundary
(see (8.7) with µ ≈ 1 for non-magnetic materials), we obtain (the ẑ
component for the magnetic fields does not lead to any new condition
because of (8.79))
Ei + Er = Et ,
1
(Ei − Er ) cos θ = Et cos θt . (8.84)
Zt
These can be solved to obtain the coefficients of reflection and refrac-
tion in a straightforward manner from
Er Zt cos θ − cos θt Zt cos θ − 1

= ≈ ,
Ei Zt cos θ + cos θt Zt cos θ + 1
Et 2Zt cos θ 2Zt cos θ
= ≈ . (8.85)
Ei Zt cos θ + cos θt Zt cos θ + 1
The other case, where the incident electric field is parallel to the
plane of incidence can also be done in a straightforward manner and
we only quote the results here. In this case, we have
Einc (x, ω) = (x̂ cos θ − ẑ sin θ) Ei eik(x sin θ+z cos θ) ,
Erefl (x, ω) = −(x̂ cos θ + ẑ sin θ) Er eik(x sin θ−z cos θ) ,
Etrans (x, ω) = (x̂ cos θt − ẑ sin θt ) Et eikt (x sin θt +z cos θt ) . (8.86)
The magnetic fields are obtained from this to have the forms
Binc (x, ω) = ŷ Ei eik(x sin θ+z cos θ) ,
Brefl (x, ω) = ŷ Er eik(x sin θ−z cos θ) ,

Et ikt (x sin θt +z cos θt )
Btrans (x, ω) = ŷ e . (8.87)
Zt
Furthermore, the boundary conditions on the tangential components
of the electric and the magnetic fields give (µ ≈ 1 for non-magnetic
materials)
(Ei − Er ) cos θ = Et cos θt ,

1
Ei + Er = Et , (8.88)
Zt
and the coefficients of reflection and transmission can be obtained
easily from
Er cos θ − Zt cos θt cos θ − Zt
= ≈ ,
Ei cos θ + Zt cos θt cos θ + Zt
Et 2Zt cos θ 2Zt cos θ
= ≈ . (8.89)
Ei cos θ + Zt cos θt cos θ + Zt
We note that, in a good conductor, θt ≈ 0 independent of the
value of the incident angle. Furthermore, the coefficients of reflection
reduce to those for reflection from a perfect conductor with Zt → 0.
8.5 Radiation pressure
As we have studied earlier, electromagnetic fields carry energy and

momentum. As a result, when electromagnetic waves are incident
on a surface, they can exert a force or pressure on the surface. This
is known as radiation pressure (in analogy with statistical mechan-
ics of gases). The examples we have studied in this chapter clearly
demonstrate this and bring out the origin behind this (which was
8.5 Radiation pressure 283
also discussed earlier along general lines when we studied the Poynt-
ing vector). Let us recall that the momentum density associated with
electromagnetic fields is related to the Poynting vector as (see (6.92))
ǫµ
p= Re S. (8.90)
c2
For harmonic fields, therefore, we can define a time averaged momen-
tum density as
ǫµ c ǫ
p= 2 Re (E × H∗ ) = Re (E × B∗ ). (8.91)
c 8π 8πc
The total momentum exerted by the EM waves on a surface of area
“a” in a time interval ∆t, therefore, follows to be
ǫa(c∆t)
∆p = p a(c∆t) = Re (E × B∗ ). (8.92)
8πc
This leads to a pressure exerted by an electromagnetic wave on a
surface of the form
F 1 ∆p ǫ
P= = = pc = Re (E × B∗ ). (8.93)
a a ∆t 8π
Let us now analyze all of this in the case of an electromagnetic
wave (in vacuum) at normal incidence on a perfectly conducting sur-
face at z = 0 that we have already studied. We have seen that, in
this case, the incident electric and the magnetic fields have the forms
(see (8.11) and (8.17))
Einc (x, ω) = E(0) eikz ,

Binc (x, ω) = ẑ × E(0) eikz . (8.94)
It follows, therefore, from (8.93) that such a wave will exert a pressure
on the surface at z = 0 of the form (in vacuum ǫ = 1)
1
P= Re E(0) × (ẑ × (E(0) )∗ )
8π
1
= ẑ |E(0) |2 . (8.95)
8π
This shows that the incident wave exerts a pressure on the surface
along the z-axis. Furthermore, if we have a perfectly conducting
surface, the incident wave is totally reflected doubling the pressure
so that, for a perfectly conducting surface, we have
1
P total = 2P = ẑ |E(0) |2 . (8.96)
4π
On the other hand, we note that, for a perfectly absorbing surface

such as a blackbody, the pressure will be given by (8.95).
The mechanism leading to the radiation pressure can also be
easily understood from this simple example. We have already seen
that, in this case, there will be induced surface currents. However, the
surface current is uniform so that there is no induced charge density.
As a result, when an electromagnetic wave is incident on the surface,
the surface current will experience a force (Lorentz force) leading to
a time averaged force per unit area as
F 1
= Re (Js × B∗ ). (8.97)
a 4c
Several comments are in order here. First, the factor of 14 arises as
follows. The time averaging leads to 12 whereas the other 12 comes
from averaging the current both above and below the surface. (An-
other way to look at it is to note that the surface current is obtained
from the jump in the magnetic field which, in the present case, is
twice that associated with the incident magnetic field (see (8.18)).)
Second, normally the Lorentz force, given in terms of a current den-
sity, gives rise to a force density. However, since we have a surface
current (that is already integrated over a unit line), the Lorentz force
leads to a force per unit area or pressure. Recalling the definitions of
B and Js in Eqs. (8.17) and (8.23) respectively, we obtain
1 c ∗
P= Re E(0) × ẑ × E(0)
4c 2π
1
= ẑ |E(0) |2 . (8.98)
8π
It follows now that, for a perfect conductor, the total pressure on the
surface due to the incident and the reflected waves is
1
P total = 2P = ẑ |E(0) |2 , (8.99)
4π
which coincides identically with (8.96).
Thus, we see explicitly the origin of the radiation pressure,
namely, the electromagnetic waves exert a force on the charged par-
ticles on the surface which generates a pressure.
1. Determine the reflection coefficient at normal incidence for sea

water, fresh water and “good” earth at frequencies 60 Hz, 1
MHz and 1 GHz. Use ǫ = 80, σ = 3.6 × 1010 /sec for sea water,
ǫ = 80, σ = 4.5 × 107 /sec for fresh water and ǫ = 15, σ =
9 × 107 /sec for “good” earth.
2. Show that two dimensional fields that are independent of z can

be written as superpositions of the following fields (in vacuum)
i) E = (0, 0, Ez ), H = (Hx , Hy , 0),
with
i ∂Ez i ∂Ez
Hx = , Hy = − ,
k ∂y k ∂x
∂ 2 Ez ∂ 2 Ez
+ + k 2 Ez = 0,
∂x2 ∂y 2
and
ii) E = (Ex , Ey , 0), H = (0, 0, Hz ),
with
i ∂Hz i ∂Hz
Ex = − , Ey = ,
k ∂y k ∂x
∂ 2 Hz ∂ 2 Hz
+ + k 2 Hz = 0.
∂x2 ∂y 2
3. Show that, when a uniform plane wave is incident normally on

a good conductor, the linear current density, Js , is essentially
independent of the conductivity σ.
Chapter 9
Radiation
So far, we have studied some of the properties of the electromagnetic

waves in various media without worrying about how they are pro-
duced. The sources of electromagnetic waves are, of course, charges
and currents. Let us now study, in some detail, the properties of
electromagnetic fields produced by a localized distribution of (time
dependent) charges and currents. Let us assume that we have a given
distribution of charges and currents in a localized volume V . We have
already seen (see (6.187) and (6.188)) that the retarded solutions of
Maxwell’s equations in the Lorentz gauge are given by (in vacuum)
Z ′ ′

1 3 ′ Jµ (x , t )
Aµ (x, t) = d x
c |x − x′ | t′ =t− |x−x′ |
c

Z |x−x′ |
1 Jµ (x′ , t′ )δ t′ − t + c
= d3 x′ dt′ . (9.1)
c |x − x′ |
In fact, let us note that since in the Lorentz gauge
1 ∂Φ
∂µ Aµ = + ∇ · A = 0, (9.2)
c ∂t
the scalar potential is related to the vector potential and it is sufficient
for us to study only the (three dimensional) vector potential. From
(9.1) we have
′

Z J(x′ , t′ )δ t′ − t + |x−x |
1 c
A(x, t) = d3 x′ dt′ ′
. (9.3)
c |x − x |
The space integral here is over the volume V which contains (charges)
currents and if we are interested in the fields at points which are far
away from the volume containing the (charges) currents, we can make
an expansion much like in the static case. This would give rise to the
287
288 9 Radiation
multipole expansion of the vector potential. In general, in the time

dependent case this expansion has to be carried out more carefully
than in the static case and would contain both electric as well as
magnetic multipoles which we will study later. In what follows in
this section, we will describe such an expansion for systems with a
simple time dependence.
Let us consider the case where the current has a simple harmonic
time dependence of the form (remember that the current is real and
corresponds to either the real or the imaginary part of this expression)
J(x, t) = J(x) e−iωt . (9.4)
It follows from the continuity equation

∂ρ
+ ∇ · J = 0,
∂t
that the charge density must also have exactly the same harmonic
time dependence, namely,
ρ(x, t) = ρ(x) e−iωt . (9.5)
In such a case, the solution for the vector potential becomes

′

Z J(x′ )e−iωt′ δ t′ − t + |x−x |
1 c
A(x, t) = d3 x′ dt′
c |x − x′ |
Z iω|x−x′ |
e−iωt 3 ′ e c
′
= d x J(x ) . (9.6)
c |x − x′ |
Therefore, the (three) vector potential also has a harmonic time de-
pendence and separating out the time dependence of the vector po-
tential,
A(x, t) = A(x) e−iωt , (9.7)
we have
Z iω|x−x′ |
1 e c
A(x) = d3 x′ J(x′ ) . (9.8)
c |x − x′ |
It is clear that all the variables (Φ, A) and (E, B) will have the same
harmonic time dependence which can be factored out. The space
dependent magnetic field can, of course, be determined from (9.8) as
B(x) = ∇ × A(x), (9.9)

289
while the electric field in the region outside the volume containing
charges and currents (in vacuum) can be obtained from the last equa-
tion of Maxwell in (7.1) as
ic i
E(x) = ∇ × B(x) = ∇ × B(x). (9.10)
ω k
So far, we have made no approximations. But, we recognize that

there are now two length scales in the problem (as opposed to the
static case, where there was only one length scale), namely, the size
d associated with the volume V which contains charges and currents
2π
and the wavelength of oscillations λ = |k| = 2πc
ω . (We are assuming
that we are in vacuum.) As a result, there are more possibilities in
the expansions that we can make. First, let us assume that we are
very far away from the sources, namely, |x| >> d (as also in the static
case). In such a case, we can expand |x − x′ | as
1
|x − x′ | = (r 2 + r ′2 − 2rr ′ cos θ ′ ) 2
12
2r ′ cos θ ′ r ′2
=r 1− + 2
r r
1
= r − r ′ cos θ ′ + O( ). (9.11)
r
Substituting this into the expression for the vector potential in (9.8),
we have
Z ′ ′
1 ei|k|(r−r cos θ )
A(x) ≈ d3 x′ J(x′ )
c (r − r ′ cos θ ′ )
Z ′ ′
!
eikr 3 ′ ′ e−ikr cos θ
= d x J(x ) ′ θ′
. (9.12)
cr 1 − r cos
r
Here, we have identified |k| = k.

Let us next assume that d << λ. This is, in fact, an excellent
ω
approximation. For, suppose ν = 2π = 60 cycles/sec, then, λ = νc =
5 × 108 cm which is large compared to laboratory sizes. In this case,
it follows that for most localized sources we have
kd << 1, (9.13)
which allows us to make a Taylor expansion of the quantity in the

290 9 Radiation
parenthesis in (9.12) as
′ ′
!
e−ikr cos θ i
r ′ cos θ ′
= 1 + (−ik)(1 + )r ′ cos θ ′
1− r kr
(−ik)2 2i 2
+ (1 + − )(r ′ cos θ ′ )2 + · · ·
2! kr (kr)2
∞
X
= Xn (r, k)(r ′ cos θ ′ )n , (9.14)
n=0
where the general form of Xn is easily seen to be

!
(1) (2) (n)
(−ik)n an an an
Xn (r, k) = 1+ + + ··· + , (9.15)
n! kr (kr)2 (kr)n
(m)
and an ’s (m ≤ n) are numerical constants. Since Xn (r, k) is inde-
pendent of the variables of integration, it can be taken outside the
integral in (9.12) leading to an expansion of the vector potential of
the form
∞ Z
eikr X 3 ′ ′ ′ ′ n
A(x) = Xn (r, k) d x J(x )(r cos θ ) . (9.16)
cr n=0
There are now two possibilities. Namely, we can either have

r << λ (kr << 1), or r >> λ (kr >> 1). The two regions are known
respectively as the near (static) zone and the far (radiation) zone. In
the two cases, the quantities Xn and, therefore, A behave differently.
(In addition, there is also the intermediate zone where kr ∼ 1 which
we will not consider.) Thus, for example, in the near zone, kr ≪ 1
and we have from (9.15)
(n) (n)
(−ik)n an ãn
Xn (r, k) ≈ n
= n , (9.17)
n! (kr) r
which is independent of k. Substituting this into (9.16), we see that,

in such a case, the vector potential takes the form
∞
X (n) Z
ãn
A(x) = d3 x′ J(x′ ) (r ′ cos θ ′ )n . (9.18)
cr n+1
n=0
In other words, in the near zone, the vector potential oscillates har-
monically with time. Otherwise, it has a purely static character (no
291
propagation). (Here we have used the fact that in the near zone,
eikr ≈ 1.) We note that, since r ′ ≪ r, the dominant term in (9.18) is
the zeroth order term leading to
Z
1
A(x) = d3 x′ J(x′ ). (9.19)
cr
(0)
Here, we have used the fact that ã0 = 1 (see (9.15)).
In contrast, in the far zone (kr >> 1), we see from Eq. (9.15)
that we have
(−ik)n
Xn (r, k) ≈ , (9.20)
n!
so that the vector potential in (9.16) takes the form
∞ Z
eikr X (−ik)n
A(x) = d3 x′ J(x′ )(r ′ cos θ ′ )n
cr n=0 n!
Z ∞
X
eikr 3 ′ ′ (−ikr ′ cos θ ′ )n
= d x J(x )
cr n=0
n!
Z
eikr ′ cos θ ′
= d3 x′ J(x′ ) e−ikr . (9.21)
cr
We note that, since kd ≪ 1 (see Eq. (9.19)), the dominant term in
(9.21) leads to
Z
eikr
A(x) = d3 x′ J(x′ ). (9.22)
cr
We see that in the far zone, the vector potential is represented
by a spherically outgoing wave. This is because the phase of the
complete vector potential is given by (ωt − kr). As we have learnt
earlier, wave fronts are described by surfaces of constant phase. Thus,
at any given time, the surfaces of constant phase are given by
ωt − kr = constant
or, kr = constant, (9.23)
which are spherical surfaces of radius r. This is like the plane wave
solutions of radiation that we studied earlier, but the waves, in the
present case, are spherically outgoing. The other thing to note is
that, in the far zone, the higher order terms in the expansion (of the
exponential) fall off rapidly, simply because kd << 1. Consequently,
292 9 Radiation
in the study of radiation, only the first term in the series contributes
significantly. We can calculate the electric and the magnetic fields
from the potentials (see (9.9), (9.10)). However, we will not go into
it now except for the observation that, asymptotically, the electric
and the magnetic fields fall off as 1r which is precisely the behavior of
radiation fields.
9.1 Electric dipole radiation
In this and in the following sections, we will study the properties of

the electric and the magnetic fields produced by some simple charge
and current distributions. Let us consider a system consisting of two
equal, but opposite charges whose magnitudes oscillate with time.
The charges are separated by an infinitesimal distance along the z-
axis as shown in Fig. 9.1. This can be thought of as an element of
an alternating current circuit (as in the second diagram in Fig. 9.1)
and is often called a Hertzian dipole.
q b
d
−q b
Figure 9.1: A Hertzian dipole element.
Let us further assume that the size of the dipole, d, is very small
and that the time dependence of the charge is harmonic as before.
Namely,
q(t) = q0 e−iωt . (9.24)
This shows that

dq(t)
I(t) = = −iωq0 e−iωt = I0 e−iωt . (9.25)
dt
Therefore, such a system can be thought of as an alternating
current element. Furthermore, both the charge and the current have
a simple harmonic dependence on time so that the earlier analysis
9.1 Electric dipole radiation 293
can be directly applied. In such a case, the expansion of the vector

potential, with the assumption that r >> d can be approximated by
the leading term, which is the first term in the expansion. As we
have seen in Eqs. (9.19) and (9.22), irrespective of whether we are in
the near zone or in the far zone, the vector potential has the leading
form (in the near zone eikr ≈ 1)
Z
eikr
A(x) = d3 x′ J(x′ ). (9.26)
cr
Next, let us use the vector identities (Ji = −xi (∂j Jj ) + ∂j (xi Jj ))
Z Z Z
d x J(x ) = − d x x (∇ · J(x )) + (ds′ · J(x′ ))x′
3 ′ ′ 3 ′ ′ ′ ′
Z
= −iω d3 x′ x′ ρ(x′ )
= −iωp, (9.27)
where we have used the fact that the currents are contained within
the volume and, consequently, n̂ · J = 0 on the surface. Furthermore,
we have used the continuity equation
∂ρ
+ ∇ · J = 0, (9.28)
∂t
as well as the definition of the electric dipole moment p for an arbi-
trary charge distribution (see (2.49))
Z
p = d3 x x ρ(x), (9.29)
in the above derivation. We note that by construction, the electric

dipole moment of our system is along the z-axis.
Substituting (9.27) into (9.26), we see that in such a case, we
can write (k = ωc )
eikr eikr
A(x) = (−iωp) = −ikp . (9.30)
cr r
This allows us to calculate the magnetic field directly as

∂ eikr
B(x) = ∇ × A(x) = −ik (r̂ × p)
∂r r
ikr
i e
= k 2 (r̂ × p) 1 + . (9.31)
kr r
294 9 Radiation
On the other hand, the electric field can be calculated from the Am-
pere’s law, namely, (since we are interested in points far away from
the sources, J = 0, see also (9.10))
1 ∂E
∇×B= , (9.32)
c ∂t
which gives
i
E(x) = (∇ × B(x))
k

2 2i 2
= −k (r̂ × (r̂ × p)) 1 + − (9.33)
kr (kr)2
ikr
i 1 e
−(θ̂ × (θ̂ × p) + φ̂ × (φ̂ × p)) − ,
kr (kr)2 r
where we have used the fact that the unit vectors in spherical coor-
∂r̂ ∂r̂ ∂r̂
dinates are not fixed. In fact, while ∂r = 0, ∂θ = θ̂ and ∂φ = φ̂ sin θ.
Furthermore, recalling that p is along the z-axis, we can simplify this
and write (recall that, in spherical coordinates, ẑ = r̂ cos θ − θ̂ sin θ)

2
E(x) = −k (r̂ × (r̂ × p))

i 1 eikr
+ (3r̂(r̂ · p) − p) − . (9.34)
kr (kr)2 r
It is clear from (9.31) that the magnetic field is always transverse

to the radial vector, namely,
r̂ · B = 0. (9.35)
However, it follows from (9.34) that the electric field is not transverse
to the radial vector in general. We note that the magnetic field
in (9.31) has two terms – one behaving as 1r which dominates for
large r, while the second depends on the radial coordinate as r12 and,
therefore, contributes significantly for small r. The second is known
as the static (or induction) term while the first is called the radiation
term for reasons that will become clear shortly. Similarly, we see from
(9.34) that the electric field has three terms out of which the r13 term
gives the most contribution for small r (and is called the static field)
while the 1r term dominates at large distances and is known as the
radiation field.
9.1 Electric dipole radiation 295
We note that in the near zone, kr << 1 and we can write the
electric and the magnetic fields as (eikr ≈ 1)
ik(r̂ × p)
B(x) = ,
r2
3r̂(r̂ · p) − p
E(x) = . (9.36)
r3
This shows that in the near zone, the magnetic field is what would be
obtained from the Biot-Savart law for a current element (except for
the trivial time dependence that has been factored out). Similarly,
in this region, the electric field is that of a static dipole (except for
the trivial harmonic time dependence which we have factored out).
Furthermore, since in this region kr << 1, the electric field dominates
over the magnetic field. In the far (radiation) zone, on the other hand,
we note that kr >> 1 and we can approximate the electric and the
magnetic fields in (9.31) and (9.34) as
eikr
B(x) = k 2 (r̂ × p) ,
r
eikr
E(x) = −k 2 (r̂ × (r̂ × p)) = −r̂ × B(x). (9.37)
r
The last relation in (9.37) can also be written as
B(x) = r̂ × E(x), (9.38)
which is the relation for traveling EM waves that we have seen earlier,
for example, in (6.30). From (9.37) we see that both the E and the B
fields fall off as 1r and are transverse to the direction of propagation
(as well as to each other) as is expected of radiation fields. We see
that the radiation terms are new compared to the behavior of static
distributions. We note that, in the static limit (ω = 0 or k = 0),
the magnetic field identically vanishes everywhere. Furthermore, the
radiation component of the electric field also vanishes in this limit.
Thus, we see that radiation is an essential feature associated with
time varying charges and currents.
Incidentally, although we have discussed a very simple system, it
is behind many physical systems such as antennas. In this discussion,
we have only retained the lowest order term which, as we see, leads to
an electric dipole description (in the static zone). The higher order
terms in the expansion, similarly, can be shown to give rise to the
description of a magnetic dipole, electric quadrupole etc.
296 9 Radiation
9.1.1 Power radiated by an electric dipole. As we have seen earlier, the

energy flow per unit time per unit area or the power flow through a
unit area is related to the real part of the Poynting vector, which in
the vacuum takes the form (this is the time averaged value defined
in (6.96)),
c
S= Re (E × B∗ ). (9.39)
8π
If we consider a large sphere of radius R, then, it is clear that the
total time averaged power radiated through the surface of this sphere
is given by
Z
Ptotal = da r̂ · S, (9.40)
where da represents a surface element on the sphere of radius R.
da
R
dΩ
Figure 9.2: The solid angle subtended by an infinitesimal surface area

on a sphere at the center.
Since the solid angle subtended by this infinitesimal area element

is given by (see Fig. 9.2)
da
dΩ = , (9.41)
R2
it follows that the average power radiated through unit solid angle
has the form
dP
= R2 r̂ · S. (9.42)
dΩ
If we take the radius of the sphere to infinity, then it is clear from
(9.42) that only the field components that decrease as 1r can con-
tribute to the power radiated out to infinity. It is for this reason that
9.2 Magnetic dipole radiation 297
these components of the fields are known as the radiation fields. They
lead to a radiation of power to infinity (which cannot come back to
the system).
If we look at the forms of the radiation components of the elec-
tric and the magnetic fields in Eq. (9.37) and recall that the dipole
moment is along the z-axis (by assumption) with ẑ = r̂ cos θ − θ̂ sin θ,
it follows that in the limit R → ∞,
dP c 4 2 2
= k |p| sin θ. (9.43)
dΩ 8π
This shows that the radiated power is highly directional (not uniform,
but depends on θ), peaking at θ = π2 . This also leads to a total
integrated average radiated power of the form
Z Z
dP c 4 2
Ptotal = dΩ = k |p| dΩ sin2 θ
dΩ 8π
c 4 2 8π ck 4 |p|2
= k |p| = . (9.44)
8π 3 3
Using k = ωc and the fact that for this simple system we can think of
(see also (9.24), (9.25))
|I0 |2 d2
|p|2 = |q0 |2 d2 = , (9.45)
ω2
we can also write the (time averaged) total power radiated to infinity,
(9.44), as
ω 2 d2 2 2ω 2 d2 2
Ptotal = |I 0 | = I . (9.46)
3c3 3c3 rms
Here, I0 represents the peak current while Irms denotes the effective
current (root mean square). This suggests that we can associate a
radiation resistance with the dipole from the standard definitions as
2ω 2 d2
Rrad = . (9.47)
3c3
9.2 Magnetic dipole radiation
Just as we studied the electric and the magnetic fields produced by a

short alternating current element (or an electric dipole), we can also
do the same for a small alternating current loop. In this case, the
298 9 Radiation
a
Figure 9.3: An alternating current loop of radius a in the x − y plane.
system would behave like a magnetic dipole. Let us consider a small

alternating current loop of radius “a” in the horizontal (x − y) plane
as shown in Fig. 9.3.
We will assume, as before, that the current has a harmonic time
dependence,
I(t) = I0 e−iωt ,
J(x, t) = J(x)e−iωt = φ̂ I0 δ(ρ − a)δ(z)e−iωt , (9.48)
where ρ denotes the radial coordinate in the x − y plane. Just as

in the case of steady currents a closed current loop gives rise to a
magnetic dipole moment, here, too, we can show that such a current
loop would have a magnetic dipole moment with a harmonic time
dependence associated with it,
m(t) = m0 e−iωt = ẑ I0 πa2 e−iωt . (9.49)
Since the current has a harmonic time dependence, we can apply

our general analysis and obtain from (9.6) that
Z ′
1 3 ′ ′ eik|x−x |
A(x) = d x J(x )
c |x − x′ |
Z ′
I0 eik|x−x |
= d x δ(ρ − a)δ(z ) φˆ′
3 ′ ′ ′
. (9.50)
c |x − x′ |
Let us note that if we write
x = r (x̂ sin θ cos φ + ŷ sin θ sin φ + ẑ cos θ) ,

x′ = ρ′ x̂ cos φ′ + ŷ sin φ′ , (9.51)
9.2 Magnetic dipole radiation 299
π
then, it follows that (note that z ′ = 0 or equivalently θ ′ = 2 because
of the delta function constraint)
1
|x − x′ | = (r 2 + ρ′2 − 2rρ′ sin θ cos(φ − φ′ )) 2
≈ r − ρ′ sin θ cos(φ − φ′ ). (9.52)
Here, we have assumed that the size of the current loop is extremely
small compared with the distance of the point of observation (r ≫ ρ′ )
which allows us to keep only the lowest order terms in the expansion.
Furthermore, let us assume that we are interested only in the
far field approximation (namely, in the radiation zone). In that case,
r is very large (kr ≫ 1) and we can approximate the denominator
by the lowest order term. However, we have to be careful with the
exponent. Let us also note that we can write φ̂′ = −x̂ sin φ′ + ŷ cos φ′ .
With these, the expression (9.50) for large values of r takes the form
(and we assume that a is very small, ka ≪ 1)
Z2π
I0 aeikr ′
A(x) ≈ dφ′ (−x̂ sin φ′ + ŷ cos φ′ ) e−ika sin θ cos(φ−φ )
cr
0
Z2π
I0 aeikr
≈ dφ′ (−x̂ sin φ′ + ŷ cos φ′ )
cr
0

× 1 − ika sin θ cos(φ − φ′ )
Z2π
ikI0 a2 sin θeikr
=− dφ′ −x̂ sin φ + sin(2φ′ − φ)
2cr
0

+ŷ cos φ + cos(2φ′ − φ)
ikI0 πa2 sin θeikr

=− (−x̂ sin φ + ŷ cos φ)
cr
ikI0 πa2 sin θeikr ik|m0 | sin θeikr
= −φ̂ = −φ̂ . (9.53)
cr cr
Here we have used standard trigonometric identities
1
sin AB = (sin(A + B) + sin(A − B)) ,
2
1
cos AB = (cos(A + B) + cos(A − B)) .
2
300 9 Radiation
We note that the form of (9.53) is, in fact, quite similar to the case
of the electric dipole radiator studied earlier, with the electric dipole
moment replaced by the magnetic dipole moment (along with various
constants needed for dimensional reasons).
We note that, unlike the electric dipole case, here the vector
potential in the radiation zone is transverse to the direction of prop-
agation, namely,
r̂ · A(x) = 0. (9.54)
The magnetic field can now be calculated easily
B(x) = ∇ × A

k 2 I0 πa2 sin θeikr 1
= −θ̂ +O . (9.55)
cr r2
Similarly, the electric field takes the form
i
E(x) = ∇×B
k

k 2 I0 πa2 sin θeikr 1
= φ̂ +O . (9.56)
cr r2
This allows us to calculate the (time averaged) power radiated per
unit solid angle in a given direction through the surface of a large
sphere to be (see (9.42))
dP
= R2 r̂ · S
dΩ
(k 2 I0 πa2 )2 k 4 |m0 |2
= sin2 θ = sin2 θ. (9.57)
8πc 8πc
We see that the angular dependence of the radiated power, in this
case, is the same as that in the case of the electric dipole radiation
in (9.43). The total power radiated through the surface of a large
sphere, averaged over a cycle, is then obtained to be
π 2 k 4 a4 2 2π 2 k 4 a4 2
Ptotal = I0 = Irms , (9.58)
3c 3c
so that we can identify the radiation resistance associated with this
system to be
2π 2 k 4 a4
Rrad = . (9.59)
3c
9.3 Center-fed antennas 301
9.3 Center-fed antennas
The previous analysis has direct application in short center-fed an-

tennas. Antennas are open wire systems connected to an alternating
current source – a transmitter. The center-fed antenna, for example,
can be a parallel transmission line with open wires or wires bent at
right angles. Such an antenna of length d (measured from one end to
the other even though it is center-fed), is known as a dipole antenna.
On the other hand, if one of the transmission lines is grounded (or
we have only one wire above a perfectly conducting plane), it also
acts like a dipole antenna together with the image inside the earth
and such a set up is known as a monopole antenna (see Fig. 9.4).
d
2
d
Figure 9.4: A dipole antenna (on the left) and a monopole antenna
(on the right).
In the earlier analysis of the alternating current element, we

assumed that the current is constant in the entire length of the wire,
which is valid only because we assumed the dipole to have a very
small length. In a realistic antenna, however, the current will be
different at different points along the length of the wire, as is known
from the study of transmission lines. Let us first consider a very
short dipole antenna of length d. We assume a harmonic variation
of the current with time as before. Since it is center-fed, it is quite
reasonable to assume that the current has a maximum at the center
and decreases linearly to zero at both ends. (From symmetry, we see
that the current pattern will be the same in both the halves of the
antenna.) Namely, let us assume that

2|z| d d
I(z, t) = I0 1− e−iωt , − ≤z≤ . (9.60)
d 2 2
302 9 Radiation
With this, we can repeat the calculation of the earlier sections. There
is a simpler method for obtaining the relevant results by noting that
the average current in the antenna, in this case, follows to be
I0
Iavg = , (9.61)
2
where we are assuming that I0 represents the peak current at the
center of the antenna.
With this, we can now extend our previous analysis simply by
letting I0 → I20 or Irms → Irms
2 . Thus, for example, for the center-fed
short dipole antenna we obtain (see (9.46)),
ω 2 d2 2
Ptotal = I . (9.62)
6c3 rms
Correspondingly, the radiation resistance of the center-fed dipole an-
tenna is smaller by a factor of 4,
ω 2 d2
Rrad = . (9.63)
6c3
On the other hand, for a short center-fed monopole antenna,
since only one half of the antenna really radiates, we have
1 ω 2 d2 2 ω 2 d2 2
Ptotal = × I = I , (9.64)
2 6c3 rms 12c3 rms
and correspondingly, the radiation resistance
ω 2 d2
Rrad = , (9.65)
12c3
is even smaller.
These theoretical predictions work quite well and can be checked
experimentally to hold for short antennas satisfying d ≪ λ where
λ denotes the wavelength of the signal. In fact, they hold up to
d ≤ λ4 . However, in transmitting radio waves (for radio waves ν =
300Hz −3000GHz, λ = 100Km −1mm), where antennas are primarily
used, it is found that the transmission is better if the dimensions of
the antenna were of the order of the wave length, d ∼ λ (see Fig. 9.5).
Of course, these are no longer short antennas and the analysis has
to be carried out more carefully. The difficulty really lies in knowing
the distribution of the current along the length of the antenna. If
this distribution is known, the calculation of the fields can be carried
out in principle. Following studies of the transmission lines where
d z
z= 2
z = − d2
Figure 9.5: A center-fed antenna of length d ∼ λ.
it is known that the current varies sinusoidally with distance, let us

assume that the current has the form

d d d
I(z, t) = I0 sin k − |z| e−iωt , − ≤z≤ . (9.66)
2 2 2
Namely, we are assuming the antenna to lie along the z-axis and the
current to be sinusoidally varying with z such that the maximum is
at the center and the ends have vanishing current. This translates to
a current density of the form
J(x, t) = J(x) e−iωt = ẑ δ(x)δ(y)I(z, t), (9.67)
where we identify

d d
J(x) = ẑ δ(x) δ(y) I0 sin k − |z| , |z| ≤ . (9.68)
2 2
Since this involves a current distribution that is harmonic in

time, we can apply our previous analysis and obtain from (9.8) and
(9.68)
Z ′
1 3 ′ eik|x−x |
′
A(x) = d x J(x )
c |x − x′ |
d
Z2 ik|x−z ′ ẑ|
I0 ′ d ′ e
= ẑ dz sin k − |z | . (9.69)
c 2 |x − z ′ ẑ|
− d2
304 9 Radiation
We note that we are interested in the radiation fields for which r ≫ d.

In such a case, we can approximate the denominator in (9.69) by
1 1
′
≈ . (9.70)
|x − z ẑ| r
We have to be more careful with the exponent where we write

1
|x − z ′ ẑ| = (r 2 + z ′2 − 2rz ′ cos θ) 2 ≈ r − z ′ cos θ. (9.71)
Furthermore, let us simplify our calculation by assuming that

λ
d = (kd = π), in which case, we have
2
π
d ′
sin k − |z | = sin − k|z ′ | = cos k|z ′ | = cos kz ′ . (9.72)
2 2
Using these in (9.69), we obtain for large r (the other term sin(kz ′ cos θ)
coming from the exponential vanishes by anti-symmetry in z ′ )
d
Z2
2I0 eikr
A(x) ≈ ẑ dz ′ cos kz ′ cos(kz ′ cos θ)
cr
0
d
Z2
I0eikr
= ẑ dz ′ cos kz ′ (1 − cos θ) + cos kz ′ (1 + cos θ)
cr
0
d
I0 eikr sin kz ′ (1 − cos θ) sin kz ′ (1 + cos θ) 2
= ẑ +
cr k(1 − cos θ) k(1 + cos θ) 0
π
I0 eikr 1 1
= ẑ + cos cos θ
ckr 1 − cos θ 1 + cos θ 2

2I0 eikr cos π2 cos θ
= ẑ . (9.73)
ckr sin2 θ
Exercise. Show that, for an antenna of arbitrary length d, the vector potential has
the large distance behavior given by
kd

2I0 eikr cos 2
− cos kd2
cos θ
A(x) = −ẑ .
ckr sin2 θ
Recalling that ẑ = r̂ cos θ−θ̂ sin θ, we can now calculate the magnetic
field for large r from (9.73) to be
B(x) = ∇ × A(x)
!
∂ θ̂ ∂ φ̂ ∂
= r̂ + +
∂r r ∂θ r sin θ ∂φ

2I0 eikr cos π2 cos θ
× (r̂ cos θ − θ̂ sin θ)
ckr sin2 θ

2iI0 eikr cos π2 cos θ 1
= −φ̂ +O . (9.74)
cr sin θ r2
This shows that, at large distances, the dominant term in the mag-
netic field is the radiation term and is along the φ̂ direction. The
electric field can also be calculated similarly and for large distances
has the form
i
E(x) = ∇×B
k

i ∂ 2iI0 eikr cos π2 cos θ 1
= − r̂ × φ̂ +O
k ∂r cr sin θ r2

2iI0 eikr cos π2 cos θ 1
= −θ̂ +O . (9.75)
cr sin θ r2
Thus, we see that at large distances, the dominant terms in the

electric and the magnetic fields are the radiation terms. They are
transverse to the direction of propagation r̂ (since as we recall, at
large distances the wave is a spherical wave) as well as to each other
which is characteristic of radiation fields. We can now calculate the
average power radiated by such a dipole per unit solid angle through
the surface of a large sphere using (9.39) and (9.42), which leads to
dP
= R2 r̂ · S
dΩ

I02 cos2 π2 cos θ
= . (9.76)
2πc sin2 θ
Similarly, we can evaluate the total power radiated, averaged over a

306 9 Radiation
cycle, to be
π
Z2 π

2I02 cos2 2 cos θ
Ptotal = dθ
c sin θ
0
π
Z 2
I2 1 + cos(π cos θ)
= 0 dθ . (9.77)
c sin θ
0
Defining x = cos θ, this leads to

Z1
I2 1 + cos πx
Ptotal = 0 dx
c 1 − x2
0
Z1
I2 1 1
= 0 dx (1 + cos πx) +
2c 1+x 1−x
0
Z1
I2 1 + cos πx
= 0 dx . (9.78)
2c 1+x
−1
Making a further change of variables, t = π(1 + x), we obtain
Z2π
I2 1 − cos t
Ptotal = 0 dt
2c t
0
I02
= (C + ln 2π − Ci(2π)) . (9.79)
2c
Here, C ≈ 0.577 is the Euler constant and Ci(x) is known as the
cosine integral defined as (see, for example, Gradshteyn and Ryzhik)
Z∞ Zx
cos t 1 − cos t
Ci(x) = − dt = C + ln x − dt . (9.80)
t t
x 0
The values of Ci(x) are tabulated in various mathematical handbooks

and using these one can show that the quantity in the parenthesis in
(9.79) has the value 0.609. Therefore, we obtain
0.609 2 0.609 2
Ptotal = I0 = Irms , (9.81)
2c c
so that we can obtain the radiation resistance, associated with such

an antenna, to be
0.609
Rrad = . (9.82)
c
For the monopole antenna, the average power radiated would be half
of this value and correspondingly the radiation resistance will also
decrease by a factor of 2 (since power is radiated only through half
the antenna).
9.3.1 Properties of antennas. Radio antennas are primarily used to

transmit signal from a given transmitter along some directions sup-
pressing transmission along others. From the results in (9.43) as well
as (9.76), we see that, for a short dipole antenna
dP
∼ sin2 θ,
dΩ
whereas for a half wavelength antenna,

dP cos2 π2 cos θ
∼ .
dΩ sin2 θ
Thus, we see that, in both the cases, the transmitted signal is the
strongest along θ = π2 or along the axis perpendicular to the antenna.
The detailed patterns of the two antennas are, of course, different.
Therefore, unlike a point sound source, which leads to uniform radia-
tion along all directions, antennas have a strong directional property
which can be used to focus the transmission along certain directions.
This enhancement of the transmitted signal along certain directions
is characterized by the antenna gain parameter which is defined as
4π dP
gd = . (9.83)
Ptotal dΩ
Namely, it measures the ratio of the power radiated per unit solid
angle in a given direction to the average power radiated per unit
solid angle. In units of decibels, this can be expressed as
Gd = 10 log10 gd . (9.84)
There are also other measures for the antenna gain, but we will not
get into these details. Let us simply note here that, by a clever choice
of an array of antennas, the antenna gain can be enhanced quite a
bit.
308 9 Radiation
9.4 Multipole expansion for electric and magnetic fields
In electrostatics, we obtained a multipole expansion for the electric

field by expressing the scalar potential in terms of spherical harmon-
ics. However, in dealing with time dependent Maxwell’s equations,
we are dealing with vector quantities and correspondingly, the mul-
tipole expansion has to be carried out in terms of vector spherical
harmonics. Such an expansion is quite useful in solving boundary
value problems (including in the study of scattering and diffraction)
as is also true in the static case.
To begin with, let us consider a scalar wave equation of the form
1 ∂2ψ
∇2 ψ − = 0, (9.85)
c2 ∂t2
where, for simplicity, we are assuming wave propagation in vacuum.
The conventional way one solves this equation is by Fourier trans-
forming the solution in the time variable
Z
ψ(x, t) = dω e−iωt ψ(x, ω). (9.86)
Substituting this into Eq. (9.85), we obtain

∇2 + k 2 ψ(x, ω) = 0. (9.87)
Here, we have defined, as before, k = ωc and this equation is known

as the Helmholtz equation. For k = 0 (ω = 0) or the static case, this
reduces to the Laplace equation which we are quite familiar with. Let
us emphasize that we are trying to solve here a general problem and
are not assuming a harmonic dependence on time as we had done
earlier for simple systems.
The solution of the Helmholtz equation is obvious in Cartesian
coordinates. However, our interest is to find solutions in spherical
coordinates where the multipole expansion becomes manifest. To
this end, drawing from our experience with spherically symmetric
equations (say, from quantum mechanics), we write a factorized form
for the solution ψ as
ψ(x, ω) = Rℓ (r, ω) Yℓ,m (θ, φ), (9.88)
where Yℓ,m represent the spherical harmonics depending only on the

angular coordinates while Rℓ is the radial function which depends
on the radial coordinate as well as the energy. Substituting this
into (9.87), we find that the Helmholtz equation separates into two
equations of the forms
2
d 2 d 2 ℓ(ℓ + 1)
+ +k − Rℓ (r, ω) = 0,
dr 2 r dr r2
(9.89)

1 ∂ ∂1 ∂2
− sin θ 2 +Yℓ,m = ℓ(ℓ + 1)Yℓ,m .
sin θ ∂θ sin θ ∂φ2
∂θ
Here, ℓ takes positive integer values including zero while, for a given
ℓ, we have m = −ℓ, −ℓ + 1, . . . , ℓ (as we know from the study of
angular momentum in quantum mechanics). The angular functions
Yℓ,m , the spherical harmonics, are eigenstates of L2 and Lz .
The radial equation in (9.89) is the equation for the spherical
Bessel functions and the two independent solutions can be written as
the spherical Bessel functions and the spherical Neumann functions
defined as
π 1
2
jℓ (x) = Jℓ+ 1 (x),
2x 2
π 1
2
ηℓ (x) = Nℓ+ 1 (x). (9.90)
2x 2
The spherical Bessel functions jℓ (x) are regular at the origin while
the spherical Neumann functions ηℓ (x) diverge. An alternative way to
write the solutions is in terms of the spherical Hankel functions, which
are defined as linear combinations of the spherical Bessel functions
and the spherical Neumann functions, namely,
(1)
hℓ (x) = jℓ (x) + iηℓ (x),
(2)
hℓ (x) = jℓ (x) − iηℓ (x). (9.91)
From the fact that the spherical Bessel functions (as well as the spher-
ical Neumann functions) are real, it follows that the two spherical
Hankel functions are complex conjugates of each other. We note that
either of the sets in (9.90) or (9.91) can be thought of as an inde-
pendent set of solutions for the spherical Bessel equation in (9.89).
Thus, we can write the most general radial solution of the form
(1) (1) (2) (2)
Rℓ (r) = aℓ hℓ (kr) + aℓ hℓ (kr), (9.92)
(1,2)
where aℓ are coordinate independent constants. The full solution
for the Helmholtz equation can now be written in the form
X (1) (1) (2) (2)

ψ(x, ω) = aℓ hℓ (kr) + aℓ hℓ (kr) Yℓ,m (θ, φ). (9.93)
ℓ,m
310 9 Radiation
There are several things to note here. First, the spherical har-
monics are normalized so that
Z
∗
sin θdθ dφ Yℓ,m (θ, φ)Yℓ′ ,m′ (θ, φ) = δℓℓ′ δmm′ . (9.94)
The spherical harmonics reduce to the Legendre polynomials when

m = 0 and that they are eigenfunctions of the angular momentum
operator. Namely, let us define (recall the definition from quantum
mechanics)
!
∂ θ̂ ∂
L = −i(r × ∇) = −i φ̂ − . (9.95)
∂θ sin θ ∂φ
As is clear, this is the rotation operator (changes only the angles)

and it is straightforward to check that

2 1 ∂ ∂ 1 ∂2
L =− sin θ + , (9.96)
sin θ ∂θ ∂θ sin2 θ ∂φ2
from which it follows that (see Eq. (9.89))
L2 Yℓ,m(θ, φ) = ℓ(ℓ + 1)Yℓ,m (θ, φ). (9.97)
We note that by construction the angular momentum operator is

transverse to the radial direction, namely, (as is obvious from (9.95))
r̂ · L = 0. (9.98)
Similarly, it satisfies commutation relations analogous to the angular

momentum operators in quantum mechanics so that we have
L × L = iL. (9.99)
As a result, we also have
r̂ · (L × L) = 0. (9.100)
With these basics, we are now ready to discuss the multipole

expansion for the electric and the magnetic fields. Let us consider
an arbitrary distribution of localized charges and currents. Outside
the region containing the sources, the Maxwell’s equations take the
forms (in vacuum, for simplicity)
∇ · E = 0, ∇ · B = 0,
(9.101)
∇ × E = − 1c ∂B
∂t , ∇×B= 1 ∂E
c ∂t .
Fourier transforming the time variable as before, we obtain

∇ · E = 0, ∇ · B = 0,
(9.102)
∇ × E = ik B, ∇ × B = −ik E,
where k = ωc . Here, both the electric and the magnetic fields have
a nontrivial dependence on frequency (because we are allowing for
an arbitrary time dependence other than the simple harmonic depen-
dence) that is understood.
The Maxwell’s equations (9.102) can be written in a simpler
form eliminating either the electric or the magnetic field. For exam-
ple, if we eliminate the electric field from (9.102), we can write
∇ · B = 0,

∇2 + k 2 B = 0,
i
E= ∇ × B. (9.103)
k
Here, we treat B as the independent field. Alternatively, if we elimi-
nate the magnetic field from (9.102), then we obtain
∇ · E = 0,

∇2 + k 2 E = 0,
i
B = − ∇ × E, (9.104)
k
where E represents the independent field.
Both the sets of equations in (9.103) or (9.104) are equivalent
and also equivalent to the Maxwell’s equations in (9.102) and we
note that the independent field in (9.103) or (9.104) can be solved
by solving a Helmholtz equation. However, in the present case, the
dynamical variable (the electric or the magnetic field) is a vector,
as opposed to the earlier case where ψ was a scalar function. Cor-
respondingly, the solutions can be expressed as before in terms of
spherical Hankel functions and spherical harmonics, but with vector
coefficients. Thus, for example, the solutions for B in (9.103) can be
written as
X (1) (1) (2) (2)

B(x) = aℓ hℓ (kr) + aℓ hℓ (kr) Yℓ,m (θ, φ), (9.105)
ℓ,m
(1,2)
where aℓ now represent arbitrary vector coefficients. These coeffi-
cients are arbitrary except that the magnetic field has to be transverse
312 9 Radiation
(see (9.103)) so that we must have

X (1) (1) (2) (2)

∇·B=∇· aℓ hℓ (kr) + aℓ hℓ (kr) Yℓ,m (θ, φ)
ℓ,m
= 0. (9.106)
(1,2)
Since hℓ represent independent solutions of the spherical Bessel
equation, it follows that, for (9.106) to hold, we must have indepen-
dently,
X (i) (i)
∇· aℓ hℓ (kr)Yℓ,m (θ, φ) = 0, i = 1, 2. (9.107)
ℓ,m
Decomposing the gradient into its radial and angular parts (see (9.95)),
∂ i
∇ = r̂ − r̂ × L, (9.108)
∂r r
we obtain from (9.107)
(i)
!
X (i) dhℓ (kr) i (i) (i)
r̂ · aℓ Yℓ,m − hℓ (kr) r̂ · (L × aℓ )Yℓ,m = 0,
dr r
ℓ,m
(9.109)
where we have used some familiar properties of products of vectors

(namely, (A × B) · C = A · (B × C)). From (9.98) and (9.100), we
note that a particular solution of (9.109) is given by
(i) (i)
aℓ ∼ aℓ L. (9.110)
The vanishing of the first term in (9.109) is, of course, obvious from
(9.98). The vanishing of the second term follows from (9.100), namely,
r̂ · (L × L) = 0. (9.111)
Thus, we can write a particular solution of the Helmholtz equation

in (9.103) satisfying the transversality of the magnetic field as
X (1) (1) (2) (2)

B(x) = aℓ hℓ (kr) + aℓ hℓ (kr) (LYℓ,m (θ, φ)),
ℓ,m
i
E(x) = ∇ × B. (9.112)
k
(The fact that this is a solution of the Helmholtz equation follows

from the observation that if ψ is a solution of the Helmholtz equation,
then so is (Lψ) since L commutes with ∇2 as well as with a scalar.)
Such an expansion of the fields leads to the electric multipole fields.
Alternatively, if we had started with (9.104), the solution would turn
out to be
X (1) (1) (2) (2)

E(x) = bℓ hℓ (kr) + bℓ hℓ (kr) (LYℓ,m (θ, φ)),
ℓ,m
i
B(x) = − ∇ × E, (9.113)
k
which leads to the magnetic multipole fields. The reason for this
nomenclature will become clear shortly.
Since the combination (LYℓ,m ) arises frequently in the study of
electrodynamics, it is given a special name, vector spherical harmon-
ics, and is defined such that it is normalized, namely,
(
√ 1 (LYℓ,m (θ, φ)), ℓ 6= 0,
ℓ(ℓ+1)
Yℓ,m (θ, φ) = (9.114)
0, ℓ = 0.
The fact that it is normalized follows from
Z
∗
sin θdθ dφ Yℓ,m (θ, φ) · Yℓ′ ,m′ (θ, φ)
Z
1
=p sin θdθ dφ (L† Yℓ,m
∗
) · (LYℓ′ ,m′ )
ℓ(ℓ + 1)ℓ′ (ℓ′ + 1)
Z
1 ∗
=p sin θdθ dφ Yℓ,m (θ, φ)(L2 Yℓ′ ,m′ (θ, φ))
ℓ(ℓ + 1)ℓ′ (ℓ′ + 1)
s Z
ℓ′ (ℓ′ + 1) ∗
= sin θdθ dφ Yℓ,m (θ, φ)Yℓ′ ,m′ (θ, φ)
ℓ(ℓ + 1)
= δℓℓ′ δmm′ , ℓ, ℓ′ 6= 0. (9.115)

Here, we have used the fact that L is a Hermitian operator as well as
Eqs. (9.94) and (9.97). With this, we can write the electric multipole
fields as
X (1) (1) (2) (2)

B(x) = aℓ hℓ (kr) + aℓ hℓ (kr) Yℓ,m (θ, φ),
ℓ,m
i
E(x) = ∇ × B, (9.116)
k
314 9 Radiation
while the magnetic multipole fields take the forms

X (1) (1) (2) (2)

E(x) = bℓ hℓ (kr) + bℓ hℓ (kr) Yℓ,m (θ, φ),
ℓ,m
i
B(x) = − ∇ × E. (9.117)
k
An arbitrary electric field can, of course, be written as a linear su-
perposition of the electric and the magnetic multipole fields.
9.5 Behavior of multipole fields
To understand the nomenclature of electric and magnetic multipole

fields, let us analyze the behavior of these fields in the near zone,
kr ≪ 1. For example, for the electric multipole fields in (9.116),
we note that the spherical Hankel functions are linear combinations
of spherical Bessel functions and spherical Neumann functions. For
small values of the arguments, these functions behave as
1
jℓ (x) → xℓ , ηℓ (x) → . (9.118)
xℓ+1
The spherical Neumann functions diverge at the origin. However,
since we are excluding the region containing charges and currents
(assumed to be in a finite region near the origin), they are allowed
in the solutions. Correspondingly, we see that at small distances,
it is the spherical Neumann functions that dominate. Choosing a
(1,2)
particular normalization for aℓ , we see that, in this case, we can
write the magnetic field in the near zone as
Xk 1
B(x) → − LYℓ,m (θ, φ)
ℓ r ℓ+1
ℓ,m
Xk
Yℓ,m
=− L . (9.119)
ℓ r ℓ+1
ℓ,m
Here, we have used the fact that L is an angular operator which does
not act on the radial coordinate as is clear from (9.95). The behavior
of the electric field in the near zone now follows from (9.116) to be
Xi
i Yℓ,m
E(x) = ∇ × B → − ∇×L . (9.120)
k ℓ r ℓ+1
ℓ,m
The expression (9.120) can be simplified in the following manner.

From the definition of L in (9.95) and the definition of the cross
9.5 Behavior of multipole fields 315
product (one should be careful about the positions of the operators),

we obtain
(∇ × L)i = −i (∇ × (r × ∇))i
= −iǫijk ǫkst ∇j rs ∇t
= −i (δis δjt − δit δjs ) ∇j rs ∇t
= −i (∇j ri ∇j − ∇j rj ∇i )

= −i δij ∇j + ri ∇2 − 3∇i − rj ∇j ∇i

= −i ri ∇2 − 2∇i − ∇i rj ∇j + δij ∇j

= −i ri ∇2 − ∇i (1 + r · ∇) . (9.121)
Thus, we can write

2 ∂
∇ × L = −i r∇ − ∇ 1 + r , (9.122)
∂r
∂
where we have used r · ∇ = r ∂r .
With (9.122), the expression for the electric field in (9.120) for
small distances takes the form
X1
∂

Yℓ,m

2
E(x) → − r∇ − ∇ 1 + r . (9.123)
ℓ ∂r r ℓ+1
ℓ,m
Let us recall here (see (9.89) with k = 0) that

2 Yℓ,m 1 ∂ ∂ 1
∇ ℓ+1
= 0, ∇2 = 2 r 2 − 2 L2 , (9.124)
r r ∂r ∂r r
so that the first term on the right hand side of (9.123) vanishes.
d
Furthermore, we observe that (r dr simply counts the power of r)

d 1 ℓ
1+r ℓ+1
= − ℓ+1 , (9.125)
dr r r
so that the expression for the electric field in (9.123) in the near zone
becomes (has the static form of a gradient)
X Yℓ,m
E(x) → − ∇
r ℓ+1
ℓ,m
X X
=− ∇Φℓ,m = Eℓ,m (x), (9.126)
ℓ,m ℓ,m
316 9 Radiation
ℓ,m Y
where we have identified Φℓ,m = rℓ+1 , the factor arising in a multipole
expansion. We recognize each term in the sum, namely Eℓ,m , to co-
incide precisely with the static electric multipole moments. For small
1
distances, these behave as ∼ rℓ+2 as opposed to the corresponding
1
terms in the magnetic fields (Bℓ,m ∼ rℓ+1 ). Therefore, the electric
fields dominate over the corresponding magnetic fields. In fact, in
the static limit, k = 0, the magnetic field vanishes and we are simply
left with the electric field which has the correct multipole expansion.
It is for these reasons that these solutions are known as the electric
multipole fields. (We have already seen this behavior in the case of
electric dipole radiation.)
Let us next analyze the behavior of the electric multipole fields in
the far away region (radiation zone) where kr ≫ 1. Asymptotically,
for large values of the argument, we know that the spherical Hankel
functions behave as
(1) eix
hℓ (x) → (−i)ℓ+1 ,
x
(2) (1) e−ix
hℓ (x) = (hℓ (x))∗ → (i)ℓ+1 . (9.127)
x
On the other hand, from our earlier discussion on fields produced by
an arbitrary distribution of charges, we know that in the radiation
zone the fields have the forms of outgoing spherical waves (see (9.22)
(2)
and (9.23)). Correspondingly, we conclude that the coefficient aℓ =
0 which leads to (for large values of kr)
X (−i)ℓ+1 ikr
e Yℓ,m
B(x) → aℓ L ,
k r
ℓ,m
X (−i)ℓ
eikr Yℓ,m
E(x) → aℓ ∇ × L (9.128)
k2 r
ℓ,m
X (−i)ℓ+1 ikr
2 ∂ e Yℓ,m
= aℓ r∇ − ∇ 1 + r ,
k2 ∂r r
ℓ,m
where we have defined the coefficient aℓ by absorbing the normaliza-

tion factor √ 1 coming from the definition of the vector spherical
ℓ(ℓ+1)
harmonics. The expression for the electric field can be further sim-
plified by noting that

2eikr Yℓ,m
r∇
r

1 ∂ ikr
eikr 2
=r 2 (ikr − 1)e Yℓ,m − 3 L Yℓ,m
r ∂r r

ℓ(ℓ + 1)
= −r̂ k 2 + eikr Yℓ,m . (9.129)
r2
On the other hand, using (9.108) we have
ikr
∂ e Yℓ,m
∇ 1+r
∂r r

= ik∇ eikr Yℓ,m
keikr
= −r̂k 2 eikr Yℓ,m + r̂ × LYℓ,m . (9.130)
r
Substituting (9.129) and (9.130) into the expression for the electric
field in (9.128), we obtain, for large distances, (namely, we are ne-
glecting terms of order r12 compared to those of order 1r )
X (−i)ℓ+1 ikr
e Yℓ,m
E(x) → − aℓ r̂×L = −r̂×B(x). (9.131)
k r
ℓ,m
There are several things to note from the structure of the radia-
tion fields in Eqs. (9.128) and (9.131). We note that the fields in the
far off zone do fall off as 1r . Furthermore, using (9.98) it is clear that
the magnetic field is perpendicular to the direction of propagation r̂.
Similarly, the electric field is also perpendicular to the direction of
propagation and the electric and the magnetic fields are orthogonal
to each other. This is the general characteristic of radiation fields.
We can also obtain the small distance as well as the large distance
behaviors for the magnetic multipole fields in a similar manner. In
fact, looking at Eqs. (9.103) and (9.104), we see that the electric
multipole fields go over to the magnetic multipole fields under the
transformations (known as duality transformations)
E → B, B → −E. (9.132)
As a result, we do not have to carry out a separate analysis for the

magnetic multipole fields. Rather, the same analysis can be taken
over with the appropriate replacements.
318 9 Radiation
Given the electric and the magnetic fields for either of the mul-
tipole expansion, we note that we can write
X
B(x) = Bℓ,m (x),
ℓ,m
X
E(x) = Eℓ,m (x). (9.133)
ℓ,m
Here, Bℓ,m (x) and Eℓ,m (x) are the multipole fields of order (ℓ, m).
For the electric multipole fields, in the radiation zone, for example,
as we have seen
ikr
(−i)ℓ+1 aℓ e Yℓ,m
Bℓ,m (x) = L ,
k r
Eℓ,m (x) = −r̂ × Bℓ,m (x)

(−i)ℓ+1 aℓ eikr Yℓ,m
=− r̂ × L . (9.134)
k r
Using (9.134) we can now calculate the time averaged radiated power
for the multipole component fields through the surface of a large
sphere per unit solid angle along a given direction as
dPℓ,m cR2
= R2 r̂ · Sℓ,m = Re r̂ · (Eℓ,m × B∗ℓ,m )
dΩ 8π
cR2
= (B∗ℓ,m · Bℓ,m )
8π
c|aℓ |2
= (LYℓ,m )∗ · (LYℓ,m ). (9.135)
8πk 2
The right hand side can be simplified by noting that
(LYℓ,m )∗ · (LYℓ,m )
1
= |L+ Yℓ,m |2 + |L− Yℓ,m|2 + |Lz Yℓ,m|2
2
h1
= (ℓ − m)(ℓ + m + 1)|Yℓ,m+1 |2
2
i
+(ℓ + m)(ℓ − m + 1)|Yℓ,m−1 |2 + m2 |Yℓ,m |2 , (9.136)
where we have defined, as in quantum mechanics, L± = Lx ± iLy

and used the properties of the action of the angular momentum op-
erators on the eigenstates of angular momenta (namely, L± Yℓ,m =
p
(ℓ ∓ m)(ℓ ± m + 1)Yℓ,m±1 ). This, therefore, gives
dPℓ,m c|aℓ |2 h 1
= (ℓ − m)(ℓ + m + 1)|Yℓ,m+1 |2
dΩ 8πk 2 2
i
+(ℓ + m)(ℓ − m + 1)|Yℓ,m−1 |2 + m2 |Yℓ,m |2 . (9.137)
We see, therefore, that by measuring the angular distribution of the

radiated power, it is possible to determine the order of the multipole
component of the wave. However, without going into details let us
note that the magnetic multipole fields would also lead to the same
exact angular pattern for radiated power, since one is obtained from
the other by replacing the electric field by the magnetic field (up
to a sign). Therefore, it is not possible to determine the nature of
the multipole fields (electric or magnetic) from a measurement of the
angular distribution of the radiated power alone. However, the two
fields have very different behavior under parity (something that we do
not get into). Therefore, one needs to measure the polarization of the
waves to determine the character of the multipole field components.
Let us note from the definition of the spherical harmonics
1
m+|m| 2ℓ + 1 (ℓ − |m|)! 2
Yℓ,m(θ, φ) = (−1) 2 Pℓ,m eimφ , (9.138)
4π (ℓ + |m|)!
and the form of (9.137) that the radiated power spectrum is indepen-
dent of the azimuthal angle φ. Therefore, the radiated power per solid
angle along a given direction depends only on the angle θ through the
associated Legendre polynomials. We note that we have denoted the
associated Legendre polynomials by Pℓ,m to avoid confusion with the
radiated power for a given multipole field component. The first few
associated Legendre polynomials have the forms
P0,0 (cos θ) = P0 (cos θ) = 1,
P1,0 (cos θ) = P1 (cos θ) = cos θ,
1
P1,±1 (cos θ) = (1 − cos2 θ) 2 = sin θ,
1
P2,0 (cos θ) = P2 (cos θ) = (3 cos2 θ − 1),
2
1
P2,±1 (cos θ) = 3 cos θ(1 − cos2 θ) 2 = 3 sin θ cos θ,
P2,±2 (cos θ) = 3(1 − cos2 θ) = 3 sin2 θ,
..
.. (9.139)
320 9 Radiation
Furthermore, by integrating (9.137) over all directions, we obtain the

time averaged total power radiated of the form
Z
total dPℓ,m
Pℓ,m = dΩ
dΩ

c|aℓ |2 1 2
= ((ℓ − m)(ℓ + m + 1) + (ℓ + m)(ℓ − m + 1)) + m
8πk 2 2
c|aℓ |2 ℓ(ℓ + 1)
= . (9.140)
8πk 2
Let us next work out explicitly the radiation pattern for a few
low order multipoles. For dipoles, ℓ = 1 and we obtain, from Eqs.
(9.137)–(9.139), that
dP1,0 c|a1 |2 3
= sin2 θ,
dΩ 8πk 2 4π
dP1,±1 c|a1 |2 3
= (1 + cos2 θ). (9.141)
dΩ 8πk 2 8π
We note that for the dipole case, the two distinct possible radiation
patterns can be represented as shown in Figs. 9.6 and 9.7. The first
describes the polar plot for the case m = 0 while the second denotes
the case for m = ±1. For m = 0, we see that the maximum power is
radiated along θ = π2 while for m = ±1, it is along θ = 0, as is also
clear from (9.141).
θ
z
Figure 9.6: Dipole radiation pattern for m = 0.
For the quadrupole radiation, ℓ = 2 and the possible forms for

θ
z
Figure 9.7: Dipole radiation pattern for m = ±1.
the differential radiated power are given by
dP2,0 c|a2 |2 45
= sin2 θ cos2 θ,
dΩ 8πk 2 4π
dP2,±1 c|a2 |2 15
= 2
1 − 3 cos2 θ + 4 cos4 θ ,
dΩ 8πk 8π
dP2,±2 c|a2 |2 15
= 2
1 − cos4 θ . (9.142)
dΩ 8πk 8π
Similarly, the angular distributions of radiated power for higher mul-
tipole fields can also be calculated.
1. A spherical balloon carries a charge Q uniformly distributed

on its surface. The balloon pulsates with a frequency “ν” and
amplitude “a” so that its radius is given by
r(t) = r0 + a sin 2πνt.
Calculate the rate of radiation of electromagnetic energy.
2. Determine the angular distribution of radiated power from a

dipole antenna of arbitrary length “d”.
3. Show that if ψ is a solution of the Helmholtz equation

∇2 + k 2 ψ = 0,
then, so is (r × ∇)ψ.
322 9 Radiation
(a) (a)
4. A time harmonic source “a” produces the fields Eω , Bω and
(b) (b)
another independent source “b” produces the fields Eω , Bω .
Show that, at any point outside the sources,
∇ · (Eω(a) × B(b) (b) (a)

ω ) = ∇ · (Eω × Bω ).
This relation is sometimes known as the Lorentz lemma.
5. a) Find the current required to radiate a power of 100W at

100MHz from a 1cm Hertzian dipole.
b) Find the magnitudes of E, B at a distance 104 cm at θ =
0◦ , 90◦ .
6. For a harmonically oscillating dipole of moment p, the Hertz

vector is defined to be
p −iω(t− r )
H= e c ,
r
at a distance r from the dipole.
a) Show that, in terms of the Hertz vector, the potentials can
be written as (up to possible multiplicative constants)
1 ∂H
A= , Φ = −∇ · H.
c ∂t
b) Show from this that, at large distances from the dipole, the
fields can be written as (up to possible multiplicative constants)

1 d2 −iω(t− r )
E= r × r × 2 pe c ,
4πcr 3 dt

1 d2 −iω(t− r )
B=− r × 2 pe c .
4πcr 2 dt
Chapter 10
Electromagnetic fields of currents
10.1 Lienard-Wiechert potential
Let us consider in this section the electromagnetic fields associated

with the simplest of physical systems. Namely, let us consider a point
particle carrying charge q and moving along a trajectory ξ(t) under
the action of some forces that we do not specify. A moving charge,
of course, produces a current with the current density of the form
Jµ (x, t) = jµ (t)δ(x − ξ(t)), (10.1)
where

µ dξ(t)
j (t) = (cq, qv) = cq, q . (10.2)
dt
We have already seen that the retarded solution for the vector
potential for an arbitrary distribution of charges and currents in the
Lorentz gauge has the form (in vacuum)
Z ′ ′

1 3 ′ Jµ (x , t )
Aµ (x, t) = d x
c |x − x′ | t′ =t− |x−x′ |
c
′

Z J (x′ , t′ ) δ t′ − t + |x−x |
1 µ c
= d3 x′ dt′ ′
, (10.3)
c |x − x |
where the space integral is over the volume containing the charges and
the currents. We can now apply this to the case of a point charge
moving along a trajectory, in which case using the form of the current
density in (10.1), we obtain
Z ′ ′ ′ ′ |x−x′ |
1 3 ′ ′ jµ (t ) δ (x − ξ(t )) δ(t − t + c )
Aµ (x, t) = d x dt ′
c |x − x |

′ ) δ t′ − t + |x−ξ(t )|
′
Z j (t
1 µ c
= dt′ ′
. (10.4)
c |x − ξ(t )|
323
324 10 Electromagnetic fields of currents
To simplify this a little further, let us define
R(t′ ) = x − ξ(t′ ),
|x − ξ(t′ )| R(t′ )
τ = t′ − t + = t′ − t + , (10.5)
c c
where we have identified R(t′ ) = |R(t′ )|. Then, the integration over
t′ in (10.4) would have contribution only from that value of the time
coordinate for which the argument of the delta function vanishes,
namely,
R(t′ )
τ = t′ − t + = 0. (10.6)
c
This can, of course, be solved for t′ in principle, once we know the
trajectory of the particle. Furthermore, we note that
dτ 1 dR(t′ )
s(t′ ) = = 1 +
dt′ c dt′
1 ′ dR(t′ ) ′
b ′ ) · v(t ) ,
=1+ R(t ) · = 1 − R(t (10.7)
cR(t′ ) dt′ c
b = R . We note
where we have used the definition of R in (10.5) and R R
′
that s(t ) is a positive quantity whenever the speed of the particle
is less than the speed of light. (This does not hold for Čerenkov
radiation which we will discuss later.) We note the standard formula
for the delta function integral,

Z
g(x)
dx δ(f (x)) g(x) = , (10.8)
df
dx
x=x0
where x0 represents the solution of f (x) = 0. If there are more than

one solution to this equation, of course, one has to sum over all the
solutions. Using (10.7) and (10.8), the vector potential in (10.4) takes
the form
′

Z j (t ′ ) δ t′ − t + R(t )
1 µ c
Aµ (x, t) = dt′ ′
c R(t )
′

1 jµ (t )
= . (10.9)
c s(t′ )R(t′ ) τ =0
These are known as the Lienard-Wiechert potentials which we had
studied briefly earlier. In deriving (10.9) we are assuming that τ =
10.1 Lienard-Wiechert potential 325
0 has only one solution. (We are also assuming that the speed of
the particle is less than the speed of light which is always true in
vacuum.) If there are more than one solution, we must sum over all
the contributions. Furthermore, the electric and the magnetic fields
associated with a moving charge can be calculated from the vector
potentials in the following manner.
We note that we can rewrite the scalar and the vector potentials
explicitly in a form convenient for our purpose as

Z R(t′ )
δ t′ − t + c
Φ(x, t) = q dt′ ,
R(t′ )
′

Z ′ δ t ′ − t + R(t )
v(t ) c
A(x, t) = q dt′ ′
. (10.10)
c R(t )
The electric field is defined in terms of these potentials to be
1 ∂A
E(x, t) = −∇Φ − . (10.11)
c ∂t
There are several things to note here. First of all, since Φ and A
depend on x only through their dependence on R, it can be easily
checked that acting on these functions, the effect of the gradient inside
the integral can be represented as ((∇R) = R) b
∂f (R) b ∂f (R) .
∇f (R) = (∇R) =R (10.12)
∂R ∂R
Using this, we obtain,
Z R
!
′b ∂ δ t′ − t + c
E(x, t) = −q dt R
∂R R
Z ′ R

q ∂ ′ v δ t −t+ c
− dt
c ∂t c R
Z " #
′ b δ t′ − t + Rc 1 ∂ ′ R
= −q dt R − − δ t −t+
R2 cR ∂t c
Z R

q ∂ v δ t′ − t +
′ c
− dt
c ∂t c R
Z ′−t+ R

b δ
′ t c
= q dt R
R2

Z Rb − v δ t′ − t + R
q ∂ c c
+ dt′
c ∂t R
 
b b v
qR q ∂  R− c 
= + . (10.13)
sR2 c ∂t sR
τ =0 τ =0
Here, we have used the relation (10.8).

The time derivative of a quantity with a restriction such as in
(10.13) is tricky and can be carried out in the following manner. Let
us note that the restriction τ = 0 can be formally written as
R(t′ )
t = t′ + = A(t′ )
c
dt dA(t′ ) 1 dR(t′ )
or, = = 1 + = s(t′ ), (10.14)
dt′ dt′ c dt′
where we have used (10.7). If the relation between t′ and t is locally

invertible (which we assume for a solution to exist), we can write
t′ = B(t),

dt′ dB(t) 1
or, = = . (10.15)
dt dt s(t′ ) t′ =B(t)
With this, we can now show that
∂ ′ ∂
f (t ) t′ =B(t) = f (B(t))
∂t ∂t
dB(t) ′
= f (B(t))
dt

df (t′ )
′
= dt ′ , (10.16)
s(t ) ′
t =B(t)
where we have used (10.15).

With the help of these, the expression in (10.13) can be simpli-
10.1 Lienard-Wiechert potential 327
fied as follows.
 
b Rb−v
qR q ∂  c

E(x, t) = +
sR2 ′ c ∂t sR
t =B(t) t′ =B(t)
  
b Rḃ − v̇ Rb − v ṡ Rb − v Ṙ
 R 1 c c c 
= q 2 +  2
− 3
− 2 2  .
sR c s R s R s R
t′ =B(t)
(10.17)
From the definitions of R in (10.5) and s in (10.7), we note that
Ṙ = −v,
Ṙ = −Rb · v,

ḃ d R(t′ ) 1 b b
R= ′ = R (R · v) − v ,
dt R(t′ ) R
ḃ · v − R
ṡ = −R b · v̇
c c
b ′) · v v
1 dR(t 1 b b b · v̇ .
=− = − R(R · v) − v · + R
c dt′ R c c
(10.18)
Using the definition in (10.7) and the relations in (10.18), as well as

some vector identities, the expression for the electric field, (10.17),
becomes

b − v 1 − ( v )2
R c c
E(x, t) = q 
s3 R 2

b×
R Rb − v × v̇
c c
+  . (10.19)
3
cs R
t′ =B(t)
Similarly, we can also calculate the magnetic field as
B(x, t) = ∇ × A(x, t)
Z
q ′ v ′ R
= dt ∇ × δ t − t +
c R c
Z ′−t+ R

q b × v)
′ ∂ δ t c
= dt (R
c ∂R R
"Z
b ′ R
q ′ (R × v) δ t − t + c
=− dt
c R2

Z Rb × v δ t′ − t + R
∂ c
+ dt′ 
∂t cR
   
Rb × v b ×v
R
q ∂   
=−  +  . (10.20)
c sR2 ∂t csR
′ t =B(t) t′ =B(t)
This can be further simplified by using (10.16) and (10.18). Further-

more, using some vector identities as well as the form for the electric
field in (10.19), it is straightforward to show that

B(x, t) = Rb × E(x, t). (10.21)
′ t =B(t)
There are several things to note from the forms of the electric
and the magnetic fields in Eqs. (10.19) and (10.21) respectively. We
see that we need to solve for t′ = B(t) to determine the electromag-
netic fields. (t′ is known as the retarded time.) This can, in principle,
be done once we know the trajectory of the particle. We note from
(10.19) that, for a charged particle at rest, v = 0 = v̇, in which case
(s = 1 in such a case)
b
qR
E(rest) (x, t) = ,
R2
b × Erest (x, t) = 0,
B(rest) (x, t) = R (10.22)
which is what we will expect from our studies in electrostatics. In

this case,
R = x − ξ, (10.23)
where ξ represents the fixed location (independent of time) of the

charged particle. In general, though, both the electric and the mag-
netic fields depend on a term of the form R12 as well as R1 . For
small values of R it is the term corresponding to R12 that dominates
while, for large R it is the one with R1 that dominates. The fields
in the far zone have the right characteristics of radiation fields and
10.2 Uniform linear motion 329
will lead to power loss through radiation. However, we note that the
term corresponding to the radiation field ( R1 ) is proportional to v̇.
Correspondingly, we see that a particle moving in vacuum will radi-
ate power only if it is being accelerated. Conversely, an accelerated
charged particle will lose energy through radiation. On the other
hand, there is no radiation of power if the particle is not being accel-
erated (in vacuum). Let us note, from (10.19), that the amplitude of
the radiation term dominates if

R|v̇|
≫ 1. (10.24)
c2 ′
t =B(t)
Thus, as long as v̇ is nonzero, we can always find a distance large

enough for (10.24) to be true. This is the analogue of the far zone
condition kr ≫ 1 that we have studied earlier for systems with a
harmonic time dependence.
10.2 Uniform linear motion
As we have seen in Eqs. (10.19) and (10.20) (or (10.21)), in order to

derive the fields, we have to determine the retarded time t′ = B(t).
This is not always easy and, therefore, the fields can be determined
in closed form only for a few special classes of motions. The simplest
motion that we can think of is, of course, a charged particle moving
in vacuum along a trajectory with a uniform velocity. In this case,
therefore, the acceleration vanishes and the radiation term is not
present in the expression for the fields. For example, the expression
for the electric field in (10.19), in this case, becomes

q Rb − v 1 − ( v )2
c c
E(uniform) (x, t) = . (10.25)
s3 R 2
′
t =B(t)
The restriction implies that the retarded time should be expressed

in terms of variables at the present time. The velocity is uniform so
that
v|t′ =B(t) = v = constant.
We can solve for the trajectory of the particle easily in this case,
ξ(t) = vt, (10.26)
where, for simplicity, we have chosen the particle to be at the coor-

dinate origin at time t = 0.
x
R
vt v
Figure 10.1: A charged particle in uniform linear motion.
From the definition in (10.5), we note that we can define (see

Fig. 10.1)
R = R(t) = x − vt. (10.27)
It follows, therefore, that we can write
x = R + vt, R(t′ ) = x − vt′ = R − v(t′ − t). (10.28)
Using this, we can solve for the retarded time
R(t′ )
t′ = t − ,
c
or, c(t′ − t) = −R(t′ ) = −|R − v(t′ − t)|,
2
or, (c2 − v 2 )(t′ − t)2 + 2Rv cos θ(t′ − t) − R = 0. (10.29)
Here, v and R stand respectively for the magnitudes of v and R and

θ denotes the angle between R and v (see also Fig. 10.2). This
is a quadratic equation which can be easily solved by the standard
method yielding
 s 
2
R v cos θ v sin θ
(t′ − t) = −  ∓ 1−  . (10.30)
c 1− v
2 c c
c2
We note that, since t′ represents a retarded time and we assume the

particle to be moving with a speed less than the speed of light, it is
the second root that is physical, namely,

 s 
2
R v cos θ v sin θ
(t′ − t) = −  + 1−  . (10.31)
c 1− v
2 c c
c2
Since R and θ are actually functions of time t, Eq. (10.31) gives the
necessary relation t′ = B(t).
Next, let us note from (10.28) and (10.31) that
R(t′ ) = R − v (t′ − t)
 s 
v 2
Rc
=R+  v cos θ + 1 − v sin θ  , (10.32)
v2 c c
1− 2 c
where we have used the physical solution in (10.31). We can calculate

and show from (10.32) that
 s 
2
R  v cos θ v sin θ 
R(t′ ) = v2
+ 1− . (10.33)
1− 2 c c
c
It follows from (10.32) and (10.33) that

2
1 − vc2 R b
b )=
′ v
R(t q +
v cos θ 2 c
c + 1 − v sin c
θ
b
b ′) − v = R R .
or, R(t (10.34)
c R(t′ )
Similarly, using (10.34), it is easily calculated that
 
2
v cos θ
b ′ ) · v = 1 − v 1 −
s(t′ ) = 1 − R(t q c 
c c2 v cos θ

v sin θ 2
c + 1− c
q
v2 θ 2
1 − c2 1 − v sinc
= q . (10.35)
v cos θ v sin θ 2
c + 1 − c
It follows, therefore, from (10.33) and (10.35) that

2 ! 12
v sin θ
s(t′ )R(t′ ) = R 1 − . (10.36)
c
Using (10.34) and (10.36) in the expression for the electric field
(10.25), we obtain

q Rb − v 1 − v22
c c
E(uniform) (x, t) =
3
s R 2

t′ =B(t)

b− v v2
q R c R 1− c 2
=
(sR)3
′
t =B(t)
b R (1 − v )
qR
2
c2
=
3 3
θ 2 2
R 1 − v sinc

qRb 1 − v2
c 2
= . (10.37)
2 3
θ 2 2
R 1 − v sinc
The magnetic field can also be calculated for this case using (10.21)
and (10.34) and it turns out to be

b ′ )
B(uniform) (x, t) = R(t × E(uniform) (x, t)
′ t =B(t)

v b 1 − v2
× qR
c c2
=
2 3
θ 2 2
R 1 − v sin
c
v
= × E(uniform) (x, t). (10.38)
c
This is, in fact, what we would expect from our earlier studies
in electrostatics. To see that, let us specialize to the case of particle
motion along the x-axis and the observation point x lying in the x−y
plane for simplicity, as shown in Fig. 10.2.
Introducing the conventional notations,
v
β= ,
c
1 1
γ=p =q , (10.39)
1− β2 1− v2
c2
we note that we can write the components of the electric and the
y
x y
R
θ
vt x
Figure 10.2: A charged particle moving along the x-axis with the
observation point in the x − y plane.
magnetic fields in (10.37) and (10.38) to be

q(x−vt)
Ex = 3 3 , Bx = 0,
γ 2 R (1−β 2 sin2 θ) 2
qy
Ey = 3 3 , By = 0,
γ 2 R (1−β 2 sin2 θ) 2 (10.40)
Ez = 0, Bz = βEy
qβy
= 3 3 .
γ 2 R (1−β 2 sin2 θ) 2
In this case, we note that

1
R = (x − vt)2 + y 2 2 ,
y y
sin θ = = 1 ,
R ((x − vt)2 + y 2 ) 2
2 2 y2
R (1 − β 2 sin2 θ) = R − β 2 y 2 = (x − vt)2 + . (10.41)
γ2
If we now go to a Lorentz frame where the particle is at rest,
the electric and the magnetic fields would transform as
q(x − vt)
Ex′ = Ex = 3 3 ,
γ 2 R (1 − β 2 sin2 θ) 2
1
Ey′ = γ(Ey − βBz ) = γ(1 − β 2 )Ey = Ey
γ
qy
= 3 3 ,
γ 3 R (1 − β 2 sin2 θ) 2
Ez′ = γ(Ez + βBy ) = 0,
Bx′ = Bx = 0,
By′ = γ(By + βEz ) = 0,
Bz′ = γ(Bz − βEy ) = γ(βEy − βEy ) = 0, (10.42)
where we have used the last relation in (10.40).

These will be the electric and the magnetic fields in the rest
frame of the particle, but still expressed in the old coordinates. We
can transform them to the coordinates of the rest frame using
x′ = γ(x − βct),
y ′ = y,
z ′ = z,

′ βx
t =γ t− . (10.43)
c
In particular, this leads to
x′
(x − vt) = (x − βct) = ,
γ
1 r′
R(1 − β 2 sin2 θ) 2 = , (10.44)
γ
where we have defined r ′ to be the distance of the observation point

in the rest frame of the particle, namely,
1
r ′ = (x′2 + y ′2 ) 2 .
Using (10.41), (10.43) and (10.44), it follows that
qx′
Ex′ = ,
r ′3
qy ′
Ey′ = ′3 ,
r
Ez′ = 0 = Bx′ = By′ = Bz′ . (10.45)
This shows that, in the rest frame of the charged particle, the mag-
netic field is zero and the electric field has the form that we will
expect for a point charge at rest. (There is no z-component of the
electric field at the observation point since it lies in the x-y plane.)
For a charged particle moving with a uniform velocity in vacuum, we
do not expect any power loss due to radiation, since the radiation
components vanish in this case.
Let us now analyze Eqs. (10.37) and (10.38) in some detail.
First, we see that, for very small velocities, the electric and the mag-
netic fields have the forms following from Coulomb’s law and Biot-
Savart’s law respectively. However, as the velocity increases and ap-
proaches the speed of light, we see, from Eq. (10.37), that the electric
field becomes negligible along the direction of motion (along θ = 0).
Its magnitude is larger as θ increases from zero and peaks at θ = π2 ,
namely, at a direction perpendicular to the motion as shown in Fig.
10.3. (Namely, the electric field is dominantly transverse to the di-
rection of motion.)
E
v= 0
v^
Figure 10.3: The electric field as a function of the velocity of the

particle. The dashed circle represents the case when the particle is
at rest.
The magnetic field is, of course, always orthogonal to the direc-

tion of motion as well as to the electric field as is clear from (10.38).
Thus, we see that the fields of an extremely relativistic charged par-
ticle in uniform motion along a straight line behave like an electro-
magnetic plane wave (where both E as well as B are perpendicular
to the direction of motion as well as to themselves). This can, in fact,
be seen in a more quantitative manner as follows.
Let us assume, as we have done earlier, that the particle is mov-
ing along the x-axis. Then, we see from (10.40), that only the z com-
ponent of the magnetic field is non-zero. Furthermore, using (10.41)
and taking the Fourier transform in the t variable (so that we go to
the frequency space), we obtain

Z∞
1
Ex (x, ω) = dt Ex (x, t) eiωt
2π
−∞
Z∞
qγ (x − vt) eiωt
= dt 3
2π (γ 2 (x − vt)2 + y 2 ) 2
−∞
iωx Z∞ − iωyξ
qe v ξe γv
= dξ 3
2πγvy (1 + ξ 2 ) 2
−∞
qe v iγv d e γv
= dξ 3
2πγvy ω dy (1 + ξ 2 ) 2
−∞
iωx
q e v πωy (1) iωy
= H
2πγvy γv 0 γv
iωx
qω e v (1) iωy
= H 0 , (10.46)
2γ 2 v 2 γv
γ(x−vt)
where we have used (10.41) and defined ξ = y in the interme-
(1)
diate steps. We note that Hn
represents the nth Hankel function
of the first kind and we have used some standard relations from the
table of integrals (see, for example, Gradshteyn and Ryzhik 8.407.1,
8.432.5, 8.472.1). Similarly, the Fourier transform of the y component
of the electric field leads to
Z∞
1
Ey (x, ω) = dt Ey (x, t) eiωt
2π
−∞
qe v e γv
= dξ 3
2πvy (1 + ξ 2 ) 2
−∞
iωx
qe v πωy (1) iωy
= − H1
2πvy γv γv
iωx
qω e v (1) iωy
=− H1 . (10.47)
2γv 2 γv
10.3 Method of virtual photons 337
It also follows from (10.40) that

iωx
qβω e v (1) iωy
Bz (x, ω) = βEy (x, ω) = − H1 . (10.48)
2γv 2 γv
Let us recall that asymptotically for large values of the argument

(i.e. when ωyγv large), the Hankel function behaves as
r
iωy 2γv − ωy π
Hn(1) → e γv −i(2n+1) 4 , (10.49)
γv iπωy
while near the origin, the behavior is of the form

(1) iωy 2i 2γv
H0 →− ln − ln 1.781 ,
γv π iωy

(1) iωy 2γv
H1 →− . (10.50)
γv πωy
Let us note that for fixed ω, y, as v → c, γ → ∞. We see from (10.46)

and (10.47) that since Ex is suppressed by a factor of γ (compared
to Ey ), it will be negligible. For large values of the frequency, all
the field components will be exponentially damped. Furthermore, as
v → c, we see from the asymptotic forms in (10.50) that (namely, we
are looking at small ω)
Ex (x, ω) → 0,
q iωx q iωx
Ey (x, ω) → e v → e c . (10.51)
πvy πcy
The non-vanishing magnetic field along the z-axis also has the same
form and magnitude as Ey . Thus, we see that every frequency com-
ponent of the fields behaves like a plane wave when v → c.
10.3 Method of virtual photons
The analysis of the fields in the Fourier transformed space, as dis-

cussed in the previous section is very important and leads to a very
useful technique called the “method of virtual photons”, originally
due to Fermi. Let us note from (10.49) that for fixed y, v, both
Ex , Ey are exponentially damped for ω > γv y . For small values of ω,
it follows from (10.46) and (10.50) that Ex is negligible while Ey dom-
inates. Thus, we can assume that Ex is negligible for any frequency
and write
γv
Ey (x, ω) = 0, ω> y ,
q iωx γv
(10.52)
Ey (x, ω) = πvy e v , ω< y .
This allows us to reconstruct the original fields through an inverse

Fourier transformation as
Z
E(x, t) = dω E(x, ω) e−iωt . (10.53)
This shows that the original field associated with the particle can be
thought of as an electromagnetic pulse.
It is, of course, much easier to solve for the fields for a given fre-
quency and once this is known, the original field can be reconstructed
by taking a linear superposition of the form in (10.53). An electro-
magnetic field configuration corresponding to a given frequency can,
of course, be associated with a photon of energy ~ω. The electromag-
netic fields associated with a moving charged particle can, therefore,
be thought of as resulting from emission and reabsorption of “vir-
tual” photons by the charged particle. The word “virtual” (also in
the method of virtual photons) comes from the following quantum
mechanical correspondence. Quantum mechanics allows a charged
particle to emit and reabsorb photons while it travels as shown in
Fig. 10.4.
k, ω
p, E p′ , E ′
Figure 10.4: An electron emitting and absorbing a virtual photon of

wave number k and frequency ω.
Such a process is, of course, not allowed classically. For example,

if k represents the wave number of the emitted photon and ω its
frequency (see Fig. 10.4), then conservation of energy and momentum
would require
p − p′ = ~k,
E − E ′ = ~ω. (10.54)
10.3 Method of virtual photons 339
It now follows from (10.54) as well as from Einstein’s relation that

µ2 c4 = (~ω)2 − (~k)2 c2 = (E − E ′ )2 − (p − p′ )2 c2
= (E 2 − p2 c2 ) + (E ′2 − p′2 c2 ) − 2EE ′ + 2p · p′ c2

= 2m2 c4 1 − γγ ′ (1 − ββ ′ cos θ)
6= 0, (10.55)
where θ represents the angle between p and p′ and we have used the
fact that for a real particle, we can write
βE
E = mc2 γ, |p| = = mcγβ. (10.56)
c
It is clear, therefore, that energy and momentum conservation will
be violated in a process where a classical charged particle emits a
photon. Quantum mechanically, however, energy can be uncertain
due to quantum mechanical fluctuations and satisfies a relation of
the form
~
∆E∆t ≥ . (10.57)
2
Therefore, during a time interval of
~ ~
∆t ≥ ∼ ,
2∆E 2µc2
such a process can take place and would involve a “virtual” photon
(it cannot be a real photon since the rest mass cannot be zero by
energy-momentum conservation). This is the reason that this method
is known as the “method of virtual photons” and the quantity µ is
known as the invariant mass of the photon.
Let us note that the electric and the magnetic fields carry en-
ergy. In vacuum, the contribution of the electric and the magnetic
energies are equal and, therefore, the total energy associated with the
transverse components of the fields (volume is considered in cylindri-
cal coordinates with the axis of the cylinder along the x-axis) can be
calculated easily to be
Z
1
U= d3 x |Ey (x, t)|2
4π
Z
v
= (2πy dy) dt |Ey (x, t)|2
4π
Z Z∞
v
= dy y dt |Ey (x, t)|2 . (10.58)
2
−∞
Here, we have used the fact that the particle is moving along the
x-axis with a uniform velocity v, to convert the x integral to a time
integral. Furthermore, from the definitions of the Fourier transform
as well as the delta function,
Z∞
E(x, t) = dω E(x, ω) e−iωt ,
−∞
Z∞ Z∞
1 1
δ(ω) = dt e iωt
= dt e−iωt , (10.59)
2π 2π
−∞ −∞
we obtain
Z Z
v ′
U= dy y dt dω dω ′ Ey (x, ω)Ey∗ (x, ω ′ ) e−i(ω−ω )t
2
Z Z
= πv dy y dω dω ′ Ey (x, ω)Ey∗ (x, ω ′ ) δ(ω − ω ′ )
Z Z∞
= πv dy y dω |Ey (x, ω)|2
−∞
Z Z∞
= 2πv dy y dω |Ey (x, ω)|2
0
γv
Z∞ Zω
q2
= 2πv dω dy
π2v2 y
0 b
γv
Z∞ Zω
2q 2 dy
= dω . (10.60)
πv y
0 b
Here, we have used (10.52) and have introduced a lower cut-off on

the y integral since it diverges. The value of this cut-off, which can
be thought of as a minimum impact parameter, can be fixed later on
physical grounds.
We recognize that we can write
Z∞
U= dω U (ω), (10.61)
0
10.4 Asymptotic values of the fields 341
where U (ω) is the energy carried by a particular frequency mode of

the field components. From (10.60), we see that we can write
γv
Zω
2q 2 dy 2q 2 γv
U (ω) = = ln . (10.62)
πv y πv ωb
b
Assuming that there are N (ω) photons with energy ~ω (associated

with frequency ω), we can now obtain
2q 2 γv
~ω N (ω) = U (ω) = ln ,
πv ωb
2q 2 γv
or, N (ω) = ln . (10.63)
πv~ω ωb
Let us assume that the charged particle is an electron with
charge q = −e. Then, we note that the fine structure constant has
the form
e2
α= . (10.64)
~c
Furthermore, identifying the minimum impact parameter from the
uncertainty relation as
~ ~
b∼ = , (10.65)
|∆p| |p − p′ |
as well as using (10.54), we can determine
2α γβc|p − p′ |
N (ω) ∼ ln
πβω (E − E ′ )

2α~ E βc|p − p′ |
= ln , (10.66)
πβ(E − E ′ ) mc2 (E − E ′ )
where we have used (10.56). This gives the flux of transverse “vir-
tual” photons. Note that, since the fine structure constant is a small
number, the number of photons associated with an electron is also
small.
10.4 Asymptotic values of the fields
As is clear from the previous example, even in the simple case of a

charged particle moving with a uniform velocity, the calculation of
the electric and the magnetic fields is nontrivial. This is primarily

because it is not easy, in general, to find a relation between the re-
tarded time and the instantaneous (or observation) time in a form
that is convenient for manipulations. It is for this reason that one
often uses approximate methods to determine the forms of the fields
(when nonuniform motion is involved). The approximations are quite
analogous to the ones we have made earlier in connection with sys-
tems with a harmonic time dependence and let us discuss these in
some detail.
As before, let us assume that the system is characterized by a
size d. For a particle in bound motion, the meaning of this size is, of
course, quite clear, namely, it corresponds to the size of the bound
system. However, for unbounded motion extending to infinity, the
meaning of a size is not quite clear. In this case, of course, we are
not looking at the motion of the particle along the entire trajectory,
rather, we are interested in a finite segment of the trajectory of the
particle (that the particle traverses during the time that we need to
make the observations). The size d can then be associated with such
a segment of the trajectory of the particle. In either case, let us as-
sume that we are interested in observations far away from the source,
namely, in the radiation zone. Correspondingly, we can assume that
R(t′ ) ≈ r ≫ d. (10.67)
Under this approximation, we see that only the second term in (10.19)
would dominate and the electric field would have the form (we are
now preparing to consider nonuniform motion)

b × ((R
qR b − β) × β̇)
E(x, t) = . (10.68)
cs3 R ′
t =B(t)
To evaluate the quantities at the retarded time, let us note that

we do not expect the retarded time to be very different from the
present time. Therefore, we can make an approximate expansion
which leads to (since by assumption r ≫ d)
|x − ξ(t′ )| r
t′ = t − ≈t− ,
c c
r
or, (t′ − t) ≈ − , (10.69)
c
where we have identified r = |x|.

With this approximation, we can now calculate various quanti-

ties of interest.
R(t′ ) = x − ξ(t′ ) ≈ x,
R(t′ ) ≈ r = |x|,
′
b ′ ) = R(t ) ≈ x̂,
R(t ′
R(t )
r¯
β(t′ ) ≈ β(t) + (t′ − t)β̇(t) ≈ β̄ − β̇ ≈ β̄,
c
b ′ ) · β(t′ )
s(t′ ) = 1 − R(t
r ¯
≈ 1 − x̂ · β̄ + x̂ · β̇
c
≈ 1 − x̂ · β̄, (10.70)
where we are using our earlier convention of defining quantities with

a bar to denote quantities at the instantaneous time. We also note
that we are assuming rc to be small (namely, the retarded time is
assumed not to be very different) and, as a result, we have neglected
some of the terms. Using (10.70) the electric field in the radiation
zone, (10.68), can now be calculated. However, since the form is
complicated, we do not give the explicit results. Rather, we would
like to discuss several special cases of this approximation.
10.4.1 Dipole approximation. As a first application of this approxi-

mation method, let us consider the case of a non-relativistic particle,
for which we can assume |β| ≪ 1. The acceleration, even if small, is
assumed to be nonzero. In this case, we see from (10.70) that
R(t′ ) ≈ R ≈ x,
R(t′ ) ≈ R ≈ r,
b ≈ x̂,
b ′) ≈ R
R(t
|β(t′ )| ≈ |β̄| ≪ 1,
s(t′ ) ≈ 1. (10.71)
In such a case, therefore, the dominant term in the electric field in

(10.68) has the form

q b × ((R
R b − β) × β̇)
E(dipole) (x, t) =
cs3 R ′
t =B(t)
¯
q x̂ × (x̂ × β̇)
≈
cr
¯ ¯
q (x̂(x̂ · β̇) − β̇)
=
cr
qa sin θ
= θ̂, (10.72)
c2 r
¯
where we have defined a = c|β̇| to be the magnitude of the acceler-
ation. Here, θ is the angle between x̂ and v̇ (see Fig. 10.5) and θ̂
represents the unit vector along this direction (which is not fixed) so
¯ ¯ ¯
that β̇ = x̂(x̂ · β̇) − θ̂|β̇| sin θ. It follows now that

b ′ )
B(dipole) (x, t) = R(t × E(dipole) (x, t)
′
t =B(t)
qa sin θ
≈ (x̂ × θ̂). (10.73)
c2 r
θ̂
x
θ
v̇
Figure 10.5: The angle θ between x and v̇.
A particle which is being accelerated radiates. Thus, we can

calculate the power loss through radiation as follows. First, we note
that the Poynting vector (we are not time averaging in this case and,
therefore, the factor differs from the earlier formula by two), in this
case, has the form
c
S= Re E(dipole) × H(dipole) ∗
4π
q 2 a2 sin2 θ
= θ̂ × (x̂ × θ̂)
4πc3 r 2
q 2 a2 sin2 θ
= x̂. (10.74)
4πc3 r 2
Here, we have used the identities involving vector products as well as
the fact that x̂ and θ̂ are orthogonal. We have also assumed that we
are in vacuum where H = B. Thus, we see that the radiated power
along x̂, through a sphere of large radius r, is given by
dP q 2 a2
= r 2 x̂ · S = sin2 θ. (10.75)
dΩ 4πc3
This is exactly the angular distribution of power radiated from a
dipole (see (9.43) as well as Fig. 10.6) and for this reason, this ap-
proximation is conventionally called the dipole approximation.
θ
z
Figure 10.6: Radiation pattern for a dipole with m = 0 as in Fig.

9.6.
Let us note that, if we write

p̈
p = qξ, a = ξ̈ = , (10.76)
q
then, the power radiated can also be written as
dP |p̈|2
= sin2 θ. (10.77)
dΩ 4πc3
We see that this is exactly the same as in (9.43) if we remember that,

for a harmonic time dependence,
ω
c= , p̈ = −ω 2 p.
k
The extra factor of two in the denominator in (9.43) comes from the
time averaging.
We can integrate (10.75) over the entire solid angle of a large
sphere to obtain the total power radiated, which has the form
Z
q 2 a2 3 2q 2 a2
Ptotal = 2π dθ sin θ = , (10.78)
4πc3 3c3
which is also known as the Larmour formula.
10.4.2 Linear acceleration. Let us next consider the case where the
particle is extremely relativistic and is subjected to an acceleration
that is small during the time scale that observations are made. We
also assume that the particle is moving in a linear trajectory so that
the acceleration is parallel to the velocity. Therefore, we have
β̇ k β, |β̇| ≪ 1. (10.79)
In this case, therefore, from (10.70) we have
R(t′ ) ≈ r,
b ′ ) ≈ x̂,
R(t
β(t′ ) ≈ β̄,
s(t′ ) ≈ 1 − x̂ · β̄. (10.80)
Furthermore, since the acceleration is parallel to the velocity, we have
β × β̇ = 0. (10.81)
With these, we see that at large distances, the electric field takes
the form (see (10.68))

q R b − β) × β̇)
b × ((R
E(linear) (x, t) =
cs3 R ′
t =B(t)
¯
q x̂ × (x̂ × β̇)
≈
cr(1 − x̂ · β̄)3
qa sin θ
= θ̂, (10.82)
c2 r(1 − β̄ cos θ)3
where, as in the case of the dipole approximation, we have defined

¯
θ to be the angle between x̂ and β̄ (or β̇ since they are parallel).
¯
We have also defined a = c|β̇| as before. We see that the form of
the electric field is very similar to the one in the case of the dipole
approximation except for the factors in the denominator. From this,
we can determine the magnetic field to be

b ′ )
B(linear) (x, t) = R(t × E(linear) (x, t)
′
t =B(t)
qa sin θ
≈ (x̂ × θ̂). (10.83)
c2 r(1 − β̄ cos θ)3
The power radiated along x̂ through the surface of a sphere of

large radius r (in vacuum) can now be calculated easily (not time
averaged)
dP cr 2
= r 2 x̂ · S = Re x̂ · E(linear) × H(linear) ∗
dΩ 4π
q 2 a2
≈ sin2 θ. (10.84)
4πc3 (1 − β̄ cos θ)6
Thus, we see that the radiated power has an angular distribution very
much like the dipole approximation except that it is modulated by
1
the relativistic correction 6 as shown in Fig. 10.7.
(1−β̄ cos θ)
Figure 10.7: Modulated pattern for a dipole radiation.
The total power radiated through the entire surface of a large

sphere, then, is obtained by integrating (10.84) over all solid angles,
which gives
Zπ
q 2 a2 sin3 θ
Ptotal = 2π dθ
4πc3 (1 − β̄ cos θ)6
0
Z1
q 2 a2 1 − x2
= 2π dx
4πc3 (1 − β̄x)6
−1
q 2 a2 8π(5 + β̄ 2 ) 2q 2 a2 (5 + β̄ 2 )
= = . (10.85)
4πc3 15(1 − β̄ 2 )4 3c3 5(1 − β̄ 2 )4
Note that this reduces to (10.78) when β̄ ≪ 1.
10.4.3 Uniform circular motion. As another application, let us analyze

the motion of a charged particle moving in a circle. This is a case
of harmonic motion and we assume that the particle motion is in
the x − y plane with the radius of the circle a small compared to
the distance where we are observing the fields. Furthermore, without
loss of generality, we can assume that the observation point lies in
the x − z plane. Then, we have
x̂ = êx sin θ + êz cos θ, (10.86)
where θ is the angle x makes with the z-axis (see Fig. 10.8, it is
different from the θ in the earlier examples) and ê{x,y,z} represent
the Cartesian unit vectors along the three axes.
xθ
ξ y
x
Figure 10.8: Circular motion of a particle in the x − y plane with the

observation point lying in the x − z plane.
If the angular frequency associated with the particle motion is

ω, then we can write
ξ(t) = a (êx cos ωt + êy sin ωt) ,
ξ̇ ωa
β= = (−êx sin ωt + êy cos ωt)
c c
= β (−êx sin ωt + êy cos ωt) ,
ω2a
β̇ = − (êx cos ωt + êy sin ωt)
c
cβ 2
=− (êx cos ωt + êy sin ωt)
a
= −β̇ (êx cos ωt + êy sin ωt) . (10.87)
Here we have defined
ωa cβ 2
β = |β| = , β̇ = |β̇| = . (10.88)
c a
Using Eqs. (10.86) and (10.87), it is easy to obtain
x̂ · β = −β sin θ sin ωt,
x̂ · β̇ = −β̇ sin θ cos ωt,
β · β̇ = 0. (10.89)
Here, the magnitude of the velocity is constant and the acceleration

is orthogonal to the direction of the velocity, a case complementary
to the earlier example where the acceleration was along the direction
of the velocity.
With the approximations in (10.70), in this case, we have
R(t′ ) ≈ r,
b ′ ) ≈ x̂,
R(t
β(t′ ) = β,
s(t′ ) ≈ 1 − x̂ · β = 1 + β sin θ sin ωt. (10.90)
It is now easy to calculate the electric field in (10.19) for large dis-
tances, which takes the form

q
E(circular) (x, t) ≈ x̂ × ((x̂ − β) × β̇)
cs3 r
q
= 3 (x̂ · β̇) (x̂ − β) − (x̂ · (x̂ − β))β̇
cs r
q
= 3 (x̂ · β̇) (x̂ − β) − sβ̇ . (10.91)
cs r
The magnetic field can also be calculated using (10.21), which

for large distances takes the form
B(circular) (x, t) ≈ x̂ × E(circular) (x, t)

q
=− 3 (x̂ · β̇) (x̂ × β) + s (x̂ × β̇) . (10.92)
cs r
It follows now that (in vacuum)

c
S= Re E(circular) × B∗ (circular)
4π
q2
=− (x̂ · β̇)(x̂ − β) − s β̇
4πcs6 r 2

× (x̂ · β̇)(x̂ × β) + s(x̂ × β̇)
!
q2 β̇ 2 (1 − β 2 )(x̂ · β̇)2
= x̂ − , (10.93)
4πcr 2 s4 s6
where we have used (10.88) and (10.89).

We can now calculate the power radiated along x̂ through the
surface of a large sphere of radius r as (not time averaged)
dP
= r 2 x̂ · S
dΩ
cq 2 β 4 (1 − β 2 ) cos2 θ + (β + sin θ sin ωt)2
= . (10.94)
4πa2 (1 + β sin θ sin ωt)6
This is, of course, time dependent and since the motion is harmonic,
we can average over one cycle of the motion. In doing so, however,
we have to be careful and note from (10.14) that
dt = s dt′ ≈ sdt, (10.95)

so that the time averaged radiated power along x̂ is given by

2π
Zω
dP cq 2 β 4ω (1 − β 2 ) cos2 θ + (β + sin θ sin ωt)2
= dt
dΩ 4πa2 2π (1 + β sin θ sin ωt)5
0
Z2π
cq 2 β 4 (1 − β 2 ) cos2 θ + (β + sin θ sin φ)2
= 2 2 dφ
8π a (1 + β sin θ sin φ)5
0
Z2π
cq 2 β 2 (1 − β 2 )(1 − β 2 sin2 θ)
= 2 2 dφ
8π a (1 + β sin θ sin φ)5
0

2(1 − β 2 ) 1
− + .
(1 + β sin θ sin φ)4 (1 + β sin θ sin φ)3
(10.96)
Thus, we see that evaluating the integral basically reduces to

evaluating an integral of the form
Z2π
1
In = dφ , (10.97)
(1 + β sin θ sin φ)n
0
which can be done in a standard manner. Let us define
z = eiφ . (10.98)
Then, we can write

n I
2i z n−1
In = (−i) dz n , (10.99)
β sin θ z2 + 2i
β sin θ z −1
where the integration is over a unit circle. The integrand has n-th
order poles at
q
i 2 2
z= −1 ± 1 − β sin θ . (10.100)
β sin θ
We note that only the first root lies within the unit circle and, con-
sequently, the integral can be evaluated in a straightforward (but
tedious manner since it is a higher order pole) using the method of
residues. Let us simply note the results that

" #
1 3
I3 = π − 3 + 5 ,
(1 − β 2 sin2 θ) 2 (1 − β 2 sin2 θ) 2
" #
3 5
I4 = π − 5 + 7 ,
(1 − β 2 sin2 θ) 2 (1 − β 2 sin2 θ) 2
"
π 3 30
I5 = −
4 (1 − β 2 sin2 θ) 52 (1 − β sin2 θ) 2
2
7
#
35
+ 9 . (10.101)
(1 − β 2 sin2 θ) 2
Using (10.101) in (10.96), we obtain

dP cq 2 β 4
= 7
dΩ 8πa2 (1 − β 2 sin2 θ) 2

2 β2 2 4
× (1 + cos θ) − (1 + 3β ) sin θ . (10.102)
4
We see that, for non-relativistic motion,

dP cq 2 β 4 2 q 2 a2 ω 4
≈ (1 + cos θ) = (1 + cos2 θ), (10.103)
dΩ 8πa2 8πc3
where we have used (10.88). This shows that the power radiated
peaks along the z-axis (θ = 0). In contrast, in the relativistic case we
see from (10.102) that the radiated power peaks at θ = π2 . The study
of this system is particularly useful in the analysis of synchrotron
radiation in accelerators.
10.5 Čerenkov effect
Earlier, we saw that particles moving with a uniform velocity in vac-

uum do not give rise to radiation. For radiation in vacuum, the
particles have to be accelerated. Let us note that in vacuum, a parti-
cle cannot travel faster than the speed of light c for causality to hold.
However, in a material medium, the speed of light changes by the in-
dex of refraction as c′ = nc which can be smaller than c depending on
√
the index of refraction of the medium. Here, n = n(ω) = ǫµ is the
index of refraction of the medium and is a function of the frequency
10.5 Čerenkov effect 353
of the traveling wave (this is also true of the speed of light in the
medium). A charged particle traveling in a medium can, therefore,
travel faster than the speed of light in the medium without violating
causality provided c′ < v < c. In such a case, radiation is produced
even when the particle is not being accelerated and this effect, known
as the Čerenkov effect, provides an important tool in detecting high
energy particles.
The Čerenkov effect is an interesting phenomenon where the
charged particle does not lead to radiation directly. Rather, it is a col-
lective phenomenon where the radiation is produced by the medium
through which the charged particle moves. Quantum mechanically,
the phenomenon can be understood as follows. The charged particle
moving through the medium excites the electrons in the atoms which,
upon return to their original state, emit a coherent radiation. Macro-
scopically, the phenomenon is analogous to the production of sound
waves (shock waves) in the case of supersonic motion in a medium,
where the fluctuations in the density of the medium produce a sound
wave. In fact, geometrically, it is easy to see that if the particle trav-
els with v < c′ , then the information spheres (spherical wave fronts
traveling with the speed of light in the medium), originating at later
times, are contained inside the earlier ones. In this case, it is easy to
see geometrically that there is only one unique retarded time for every
point as shown in Fig. 10.9. (Basically, what this means is that any
observation time can lie only on the surface of a single information
sphere and, therefore, would correspond to a unique retarded time.)
123
b b b
Figure 10.9: Information spheres for a particle travelling with v < c′ .
On the other hand, if v > c′ the particle is moving faster than

the speed with which the information spheres travel. Consequently,
the information spheres originating at later times overtake the earlier

ones (see Fig. 10.10). (In this case, the information spheres intersect
leading to the fact that any observation time may lie on the surface
of two or more information spheres and, therefore, may correspond
to two or more retarded times.) The surfaces of the information
spheres define a cone with the charged particle at the vertex and as
these waves become more and more dense, a shock wave is produced
traveling perpendicular to the conical surface. It is easy to see geo-
metrically, that if θc denotes the angle between the axis of the cone
and the perpendicular to the surface of the cone, then,
c′
cos θc = . (10.104)
v
c ′t θc
b b b
1 2 3 vt
Figure 10.10: Information spheres for a particle travelling with v > c′ .
Defining β ′ = cv′ = βn(ω), we see that for such a phenomenon

to take place, we must have
1 1
′
= = cos θc < 1
β βn(ω)
1
or, β> , (10.105)
n(ω)
which also implies β ′ > 1. This is the condition for the emission of
Čerenkov radiation. Geometrically, one can see that, in this case,
there can be more than one retarded times associated with any given
point.
To understand the Čerenkov radiation more quantitatively, let
us note that, in an arbitrary medium, the vector potential, in the
Lorentz gauge, is given by (This follows from the Maxwell’s equations

and so far we had been considering the medium to be vacuum, for
which ǫ = µ = 1.)
Z ′ ′

µ 3 ′ Jµ (x , t )
Aµ (x, t) = ′ d x
c |x − x′ | t′ =t− |x−x ′|
c′
Z ′ ′

µ 3 ′ Jµ (x , t )
= ′ d x , (10.106)
c |x − x′ | t′ =t− n(ω)|x−x′ |
c
where the refractive index of the medium is defined to be

√ c
n(ω) = ǫµ = ′ . (10.107)
c
Let us assume that we have an isotropic, non-magnetic medium (which
is fairly general) for which µ = 1 and, consequently,
√ c
n(ω) = ǫ = ′ . (10.108)
c
As we have discussed earlier, in the Lorentz gauge, the scalar and
the vector potentials are related to each other so that it is sufficient
to study only the vector potential. If we have a charged particle in
uniform motion, we know that we can write the current as
J(x, t) = qv δ3 (x − ξ(t)), (10.109)
where ξ(t) represents the trajectory of the charged particle. Using
this as well as our earlier observations, we can write
′

Z qv δ 3 (x′ − ξ(t′ ))δ t′ − t + |x−x |
1 c ′
A(x, t) = ′ dt′ d3 x′ ′
c |x − x |
Z |x−ξ(t′ )|)
qv δ(t′ − t +
′ c′
= ′ dt ′
c |x − ξ(t )|

q v

= ′ , (10.110)
b
c R|1 − R·v
′ | ′ R
c t =t− c′
where we have defined as before, R = x − ξ. Equation (10.110)

is exactly like the earlier cases (see (10.9)) except that we have the
modulus of s in the denominator, which is not necessary when v < c′
(for which s is positive definite).
For uniform linear motion, the retarded time can be determined
as before. In fact, following the discussion up to (10.30) we see that,
in this case, we can write
R h p i
c′ (t′ − t) = − β ′
cos θ ∓ (β ′ cos θ)2 + (1 − β ′2 ) .
(1 − β ′2 )
(10.111)
Here θ is the angle between R = R(t) and v (see Fig. 10.2). For t′ to
represent a retarded time, as we have argued before, only the second
root is allowed when β ′ < 1. As we will see now, for β ′ > 1, both the
roots are allowed leading to two retarded times for a given t.
For β ′ > 1, we note from (10.111) that real roots will exist only
if
(β ′ cos θ)2 + (1 − β ′2 ) > 0

2 1
or, cos θ > 1 − ′2 , (10.112)
β
which leads to the two solutions
1 1
1 2 1 2
cos θ > 1 − ′2 , or, cos θ < − 1 − ′2 . (10.113)
β β
On the other hand, for t′ to represent a retarded time, we see from
(10.111) that cos θ must be negative (because of the factor in the
denominator). Therefore, we choose θ to be an obtuse angle and
note from (10.113) that the allowed root corresponds to
1 !
1 2
arccos − 1 − ′2 < θ < π. (10.114)
β
In this case, it is easy to see from (10.111) that both the solutions
lead to retarded times and provide the two retarded times for this
problem.
Even though the retarded times are determined (and now the
contributions from both these solutions must be added in (10.110)),
evaluating the fields is extremely complicated. In what follows, we
will present an alternate derivation of the fields as well as the energy
radiated, using the method of Fourier transforms, which also brings
out some other interesting features. To keep the discussion parallel
to the earlier discussion of uniform linear motion in vacuum, let us
choose the charged particle to be traveling along the x-axis. In this
case, therefore, the current has a non-vanishing component only along
the x-axis given by
Jx (x, t) = qv δ(x − vt)δ(y)δ(z). (10.115)
Taking the Fourier transform of this (into the energy-momentum

space), we have
Z
1
Jx (k, ω) = dt d3 x Jx (x, t) ei(ωt−k·x)
(2π)4
Z
qv
= dt dx δ(x − vt) ei(ωt−kx x)
(2π)4
Z
q ω
= 4
dx ei( v −kx )x
(2π)
q ω
= δ kx − . (10.116)
(2π)3 v
In the Lorentz gauge, Maxwell’s equations (in the presence of

currents) in an arbitrary medium (with µ = 1) lead to the inhomo-
geneous wave equation (recall that in a medium ∂µ = ( c1′ ∂t
∂
, ∇))

1 ∂2 2 4π
− ∇ A(x, t) = ′ J(x, t), (10.117)
c′2 ∂t2 c
which in the Fourier transformed space becomes

ω2 2 4π
− ′2 + k A(k, ω) = ′ J(k, ω),
c c
4π J(k, ω)
or, A(k, ω) = . (10.118)
c′ k2 − ω′22
c
It is now straightforward to obtain

Z
4π J(k, ω) ik·x
A(x, ω) = ′ d3 k 2 e . (10.119)
c k2 − cω′2
The magnetic field, at any point, can now be obtained as
B(x, ω) = ∇ × A(x, ω)
Z
4πi k × J(k, ω) ik·x
= ′ d3 k 2 e . (10.120)
c k2 − cω′2
Putting in the form of the current in (10.116), we see that the mag-
netic field is orthogonal to the direction of motion of the charged
particle (namely, Bx = 0) and has the form
iωx Z
iq e v (kz êy − ky êz )
B(x, ω) = dky dkz ei(ky y+kz z) .
2π 2 c′ ky2 2 ω2
+ kz − c′2 1 − β ′2 1
(10.121)
The integral over kz in (10.121) can be carried out using the

method of residues. We note that the integrand has poles at
s s
ω2 1 1 1
k∗ = kz = ± 1 − ′2 − ky2 =± ω2 − − ky2 .
c′2 β c′2 v 2
(10.122)
If β ′ < 1 (v < c′ ), we note that the quantity inside the square root is
negative and, therefore, the poles are on the imaginary axis. Enclos-
ing the contour in the upper half plane would pick up the contribution
of the pole on the positive imaginary axis and it is easy to check that
the integrals, in this case, lead to fields of the forms discussed in
(10.47) and (10.48).
On the other hand, if β ′ > 1 (v > c′ ), then, the quantity inside
the square root in (10.122) can be positive for some values of ky ,
while it will be negative for other values. When ky is such that the
quantity is negative, the poles will lie on the imaginary axis and
the integration can be carried out, much like in the earlier case, by
enclosing the contour in the upper half plane. For values of ky for
which the quantity inside the square root is positive, the poles will
lie on the real axis. However, since we are interested in retarded
solutions, as discussed in (6.179), the proper prescription is obtained
by letting (this gives the pole prescription for the retarded Green’s
function or solution)
ω → ω + iǫ,
whose effect, in the present case, is to push the pole at the positive
value of kz to the upper half plane while pushing the other pole at the
negative value to the lower half plane. Thus, enclosing the contour
in the upper half plane would pick up two contributions – one from
the pole on the positive imaginary axis and the other from the pole
slightly above the positive real axis. This is the basic difference from
the case β ′ < 1 and is a reflection of the existence of two retarded
times in this language. Denoting by k∗ the generic location of the
poles, we obtain from (10.121)

iωx Z
iq e v k∗ i(ky y+k∗ z)
By (x, ω) = 2 ′
(2πi) dky e
2π c 2k∗
iωx Z
qe v
=− dky ei(ky y+k∗ z) ,
2πc′
iωx Z
iq e v ky i(ky y+k∗ z)
Bz (x, ω) = − 2 ′
(2πi) dky e
2π c 2k∗
iωx Z
qe v ky i(ky y+k∗ z)
= dky e . (10.123)
2πc′ k∗
The electric fields can now be calculated from Maxwell’s equa-
tions, which implies that, away from the trajectory of the charged
particle, we have (recall that we are assuming µ = 1)
ǫ ∂E(x, t)
∇ × B(x, t) = , (10.124)
c′ ∂t
leading to
ic′
E(x, ω) = ∇ × B(x, ω). (10.125)
ωǫ
It follows now in a straightforward manner from (10.123) that

ic′ ∂Bz ∂By
Ex (x, ω) = −
ωǫ ∂y ∂z
iωx Z
qω e v c′2 ei(ky y+k∗ z)
=− 1 − dky ,
2πc′2 v2 k∗

ic′ ∂Bx ∂Bz
Ey (x, ω) = −
ωǫ ∂z ∂x
c′
= Bz ,
vǫ

ic′ ∂By ∂Bx
Ez (x, ω) = −
ωǫ ∂x ∂y
c′
=− By . (10.126)
vǫ
Equations (10.123) and (10.126) give the electric and the mag-
netic fields produced by an unaccelerated charged particle moving in
a medium with a speed larger than the speed of light in the medium.
To see that there is radiation, in such a case, let us calculate the
energy radiated through the conical surface along the ẑ direction as
the particle travels a distance d along the trajectory. Let us consider
a surface parallel to the x − y plane defined by z = z0 . Further-
more, let us consider a strip of the surface of width d along the x-axis
(x0 ≤ x ≤ x0 + d). Then, the total energy radiated through this
strip (considering contributions from both the surfaces at ±z0 , as we
will see shortly, the contribution is independent of the value of z0 ) is
given by
Z∞ Z∞ xZ0 +d
c′
E = 2 Re dt dy dx ẑ · (E(x, t) × H∗ (x, t))
4π
−∞ −∞ x0
Z∞ Z∞ xZ0 +d
c′
= Re dt dy dx Ex (x, t)By∗ (x, t)
2π
−∞ −∞ x0
Z∞ Z∞ xZ0 +d
c′
= 4π Re dω dy dx Ex (x, ω)By∗ (x, ω)
2π
0 −∞ x0
Z∞ Z∞ xZ0 +d
= 2 Re dω dy dx c′ Ex (x, ω)By∗ (x, ω), (10.127)

0 −∞ x0
where we have used the fact that B = H when µ = 1 as well as the

fact that Bx = 0.
From the forms of the fields in (10.123) and (10.126), we see
that since both Ex , By have the same x dependence in the phase, the
integrand in (10.127) is independent of x and, consequently, the x
integration is trivial leading to a factor of d. Furthermore, defining
s
ω2 1
k∗ = 1 − ′2 − ky2 ,
c′2 β
s
′ ω2 1
k∗ = 1 − ′2 − ky′2 , (10.128)
c′2 β
we note from (10.127) that we can write
Z∞ Z∞
E = 2d Re dω dy c′ Ex (x, ω)By∗ (x, ω)
0 −∞
Z∞ Z∞ Z∞ Z∞
q 2
= 2d − Re dω dy dky dky′
2π
0 −∞ −∞ −∞
′ ′
ω c′2 ei(ky −ky )y+i(k∗ −k∗ )z0
× ′2 1− 2
c v k∗
Z∞ Z∞ Z∞
q2d ω c′2
= Re dω dky dky′ 1− 2
π c′2 v
0 −∞ −∞
′
δ ky − ky ei(k∗ −k∗ )z0
′
×
k∗
Z∞ Z∞
q2d ω c′2 1
= Re dω ′2 1− 2 dky . (10.129)
π c v k∗
0 −∞
As noted earlier, this expression is independent of z0 .

The ky integration can be done in a trivial manner. We note
from the definition in (10.128) that k∗ is imaginary for
1
ω 1 2
|ky | > k0 = ′ 1 − ′2 . (10.130)
c β
Therefore, we can cutoff the integral over ky at these limiting values
and obtain
Zk0
dk
q y = π, (10.131)
k 2 − k2
−k0 0 y
where we have used standard results from the table of integrals (see,
for example, Gradshteyn and Ryzhik, 2.274 or this can also be done
by elementary methods). With this, we obtain
Z∞
2 ω c′2
E = q d dω ′2 1 − 2
c v
0
Z∞
2 1 1
=q d dω ω ′2
− 2 . (10.132)
c v
0
We can also define the energy radiated per unit length as
Z∞
E 2 1 1
W = =q dω ω ′2 − 2 . (10.133)
d c v
0
This can be thought of as the energy radiated per unit length of the
trajectory of the charged particle. Thus, we see that, even though the
charged particle is not being accelerated, by virtue of the fact that
it travels faster than the speed of light in the medium, it leads to
emission of radiation. This effect is known as the Čerenkov radiation
(effect). (Equation (10.133) shows that if v > c′ , energy is radiated
since E > 0. On the other hand, if v < c′ , no energy is radiated since
E < 0.)
10.6 Self-force
So far, we have talked about a point charge which is an idealization.

In fact, classically, we know that a point charge leads to singular
field configurations (and energy) and, therefore, a better way is to
think of a classical charged particle as an extended object of small
dimensions with a charge distribution. In electrostatics, for example,
one calculates the fields and energy of such a system by assuming a
spherical charge distribution of a small radius. Such a description of a
classical charged particle leads to interesting effects when the particle
is not at rest. As we have seen, for a moving charged particle, the
fields at any point are determined by the retarded time (and not by
the instantaneous time). If the particle has an extension, then, of
course, each element of the object will exert a force on every other
element. If the particle is not moving with a uniform velocity, the
retarded times associated with the action (force) will be different from
that of the reaction (force). As a result, there will be a net force acting
on the charged particle, commonly known as the “self-force” which
will be proportional to the acceleration of the particle. Intuitively,
we can see that something like this should happen from the following
simple argument. As we have seen, an accelerated charged particle
radiates energy and thereby loses kinetic energy. Therefore, radiation
of energy leads to a deceleration implying that there must be a force
acting on the particle due to the fields it produces. This is the “self-
force” or sometimes also known as the radiation reaction.
The self-force was first studied by Lorentz and was later gener-
alized by Abraham. The derivation of the self-force is quite technical,
but let us discuss it within a simple context. Let us assume that the
charged particle consists of a charge distribution of small dimension
d (in the limiting case of a point particle, we can take this to be
zero). We will also assume that we are in a frame where the particle
is instantaneously at rest. As before, we will express quantities at the
instantaneous time (observation time) with a bar over them. Since
10.6 Self-force 363
the size of the charge distribution is small, in calculating the effect

that one element of the charge distribution produces on another, we
can Taylor expand the terms in powers of the size (note that both x
and ξ denote points inside the charge distribution whose dimension
is small). For example, we note that we can write

′ ′ dR(t′ ) 1 ′ 2 ′
2 d R(t )
R(t ) = R + (t − t) + (t − t)
dt′ t′ =t 2! dt′2 t′ =t
3

′
1 ′ 3 d R(t )
+ (t − t) + ···
3! dt′3 t′ =t
(t′ − t)2 ¯
= R + (t′ − t)(−v̄) + (−v̇)
2
(t′ − t)3 ¯ + ··· ,
+ (−v̈) (10.134)
6
where R(t′ ) = x−ξ(t′ ). Let us recall that the retarded time is defined
as
R(t′ )
t′ = t − . (10.135)
c
Using this as well as the fact that the particle is instantaneously at
rest (v̄ = 0), we obtain, from (10.134),
R2 (t′ ) ¯ R3 (t′ ) ¯
R(t′ ) = R − v̇ + v̈ + · · · . (10.136)
2c2 6c3
We are assuming here as before that Rc (and, therefore, t′ − t) is small
and, consequently, a Taylor expansion as in (10.136) is meaningful.
Furthermore, we keep expansions up to order R3 because, as we will
see shortly, that is sufficient for our purposes.
Equation (10.136) leads to
2 R2 (t′ ) 3 ′
¯ + R (t ) R · v̈
4 ′
¯ + R (t ) v̇
¯ · v̇
¯ + ··· ,
R2 (t′ ) = R − R · v̇
c2 3c3 4c4
(10.137)
which can be rearranged to the form
¯ ¯ ¯ · v̇
¯
2 ′ R · v̇ 2 R3 (t′ )R · v̈ R4 (t′ )v̇
R (t ) 1 + 2 ≈R + 3
+ 4
,
c 3c 4c
¯ − 2 2 R3 (t′ )R · v̈
1
¯ ¯ 2
¯ · v̇
1
′ R · v̇ R4 (t′ )v̇
or, R(t ) ≈ 1+ 2 R + + .
c 3c3 4c4
(10.138)
3
Iterating (10.138) and keeping terms to order R , we obtain
2 ¯ ¯
!
¯
R · v̇ ¯ 2 R(R · v̈)
3(R · v̇) ¯ R (v̇ · v̇)
′
R(t ) ≈ R 1 − 2
+ 4
+ 3
+ 4
.
2c 8c 6c 8c
(10.139)
Using this, we can rewrite (10.136) to leading orders as
2¯ 2 ¯ v̇
¯ 3¯
′ R v̇ R (R · v̇) R v̈
R(t ) = R − 2 + + 3 + ··· . (10.140)
2c 2c4 6c
In a similar manner, we can also Taylor expand (note that v̄ = 0)
′ 2
¯ + (t − t) v̈
v(t′ ) = v̄ + (t′ − t) v̇ ¯ + ···
2!
¯ ¯ v̇
¯ 2¯
Rv̇ R(R · v̇) R v̈
≈− + +
c 2c3 2c2
¯ v̇¯ ¯
R ¯ (R · v̇) Rv̈
= −v̇ + + ,
c 2c2 2c
¯ + (t′ − t)v̈
v̇(t′ ) ≈ v̇ ¯ = v̇¯ − R v̈.¯ (10.141)
c
As we have seen in (10.19), the electric field of a point charge
can be written as
" v̇ #
R − Rvc 1 − ( v 2
c ) R × R − Rv
c × c
E(x, t) = q 3
+ 3
.
(sR) c(sR) ′t =B(t)
(10.142)
Thus, using (10.139)-(10.141), we can expand
R(t′ ) · v(t′ )
s(t′ )R(t′ ) = R(t′ ) −
c
" 2 ¯ ¯
#
¯
R · v̇ ¯ 2 R(R · v̈)
(R · v̇) ¯ 3R (v̇ · v̇)
=R 1+ − − − + ··· ,
2c2 8c4 3c3 8c4
(10.143)
and, similarly,
2¯ 3¯ 2 ¯ v̇¯
R(t′ )v(t′ ) R v̇ R v̈ R (R · v̇)
R(t′ ) − =R+ 2 − 3 − 4
+ ··· ,
c 2c 3c 2c
2 ¯ ¯
v(t′ ) 2 R (v̇ · v̇)
1−( ) =1− 4
+ ··· . (10.144)
c c
10.6 Self-force 365
3
The reason for keeping terms up to order R in the expansion is now
clear. Since sR ∼ R and (sR)3 occurs in the denominator (in both
the terms), the dominant terms would come from terms up to order
3
R in the numerator.
Using (10.143) and (10.144), we now obtain
" 2 ¯ ¯
R− Rv
1 − ( vc )2 1 3(R · v̇) ¯ R (v̇ · v̇)
c
= 3 R 1− +
(sR)3 R 2c 2 8c 4
¯ 2 R(R · v̈)
15(R · v̇) ¯
+ +
8c4 c3
2 2
! 3
#
R 5R (R · ¯
v̇) R
¯
+v̇ − − 3 v̈ ¯ + ··· ,
2c2 4c4 3c
v̇
R× R − Rv × c 1 ¯
c
= R (R · ¯ − R(R · v̈)
v̇)
c(sR)3 cR3 c
2 ¯ ¯
!
R (v̇ · v̇) 3(R · v̇) ¯ 2
− −
2c2 2c2
2
! 3
#
¯
¯ −R + 2R (R · v̇)
+v̇
2
+
R ¯
v̈ + · · · .
c2 c
(10.145)
If we assume the charge density ρ(ξ) of the extended particle to

be spherically symmetric, then the electric field produced at x (inside
the particle) due to all the other charge elements of the particle would
be given by (see (10.142))
Z

Rv

v 2

R− c 1− c
E(x, t) = d3 ξ ρ(ξ) 
(sR)3
v̇ #
R× R − Rv
c × c
+ 3
. (10.146)
c(sR) ′ t =B(t)
Thus, the self-force acting on the particle is obtained to be

Z
Fself = d3 x ρ(x) E(x, t)
Z

Rv

v 2

R− c 1− c
= d3 x d3 ξ ρ(x)ρ(ξ) 
(sR)3
v̇ #
R× R − Rv
c × c
+ 3 . (10.147)
c(sR)
We can now use the expansions in (10.145) and note that, for spher-
ically symmetric charge distributions, the terms in the integrand
which are odd in R would vanish (recall that R = x − ξ). As a
result, we obtain
Z ¯ ¯ ¯
3 3 (R · v̇)R v̇ 2v̈
Fself = d x d ξ ρ(x) ρ(ξ) − 3 − +
2c2 R 2c2 R 3c3
¯ Z
2v̇ ρ(x) ρ(ξ) 2q 2 ¯
= − 2 d3 x d3 ξ + 3 v̈
3c R 3c
4U ¯ 2q 2 ¯
=− v̇ + 3 v̈. (10.148)
3c2 3c
Here, we have used symmetric integration in the intermediate steps
and have identified the self-energy of the system as
Z Z
1 3 3 ρ(x) ρ(ξ) 1 ρ(x) ρ(ξ)
U= d xd ξ = d3 x d3 ξ . (10.149)
2 R 2 |x − ξ|
In his studies of the forces acting on a charged particle, Lorentz

had already argued that, when a particle is accelerated, there must be
other forces acting on the particle besides the usual “Lorentz” force.
In fact, he had already studied as an additional force precisely the
second term on the right hand side of (10.148). This was further gen-
eralized by Abraham, following the works of Larmour, Heaviside and
others, and correspondingly, the self-force in (10.148) is also known
as the Abraham-Lorentz force. If there is an external force acting on
an electron, then, together with the self-force, the equation of motion
for an electron can be written as
4U 2e2
mI v̇ = Fext + Fself = Fext − v̇ + v̈,
3c2 3c3
2e2
or, (mI + mem ) v̇ = Fext + v̈,
3c3
1 2e2
or, v̇ = Fext + v̈, (10.150)
m 3mc3
10.6 Self-force 367
¯ = v̇, v̈
where we have identified, for simplicity, v̇ ¯ = v̈ and have
defined
4U
mem = , m = mI + mem . (10.151)
3c2
We can think of mI as the inertial mass of the electron while mem can
be thought of as the electromagnetic mass of the electron (charged
particle). Neither is individually observable. Experimentally, one can
measure only the observable mass m.
In deriving Eq. (10.148), we have neglected higher order terms
which would vanish in the limit that the particle has no structure.
However, in that limit, the self-energy of the electron (particle) in
(10.149) and, therefore, the electromagnetic mass diverges. In a quan-
tum theory, such a phenomenon is handled through renormalization.
Classically we can think of the electron not as a point particle, rather
as one with a structure of the size of about (experiments put an upper
bound of 10−17 cm on the size of the electron)
e2
re = ≈ 3 × 10−13 cm. (10.152)
mc2
This is also known as the Lorentz radius or the Thomson scattering
length. Consequently, the self-energy is finite. Furthermore, this
distance scale also defines a time scale
2re 2e2
τ= = ≈ 10−24 sec, (10.153)
3c 3mc3
where we have put in the factor of 23 in the definition of the time scale
for later convenience (see also the second term on the right hand side
of (10.150)). This time scale is tiny showing that the expansion used
is convergent. Moreover, since such a time scale is in the domain
of quantum mechanics, we recognize that we can, at best, think of
the classical equation in (10.150) as an approximate equation. If not,
Eq. (10.150) leads to conceptual problems. For, suppose there is no
external force present, namely, Fext = 0, then we see that
v̇ = τ v̈
t
or, v(t) = v(0) e τ + constant. (10.154)
Namely, we have a velocity that grows exponentially with time im-

plying that the particle self-accelerates to infinite velocities.
The problem with this run away solution can be fixed if we as-
sume that the self-force exists only in the presence of other external
forces and that the radiation loss due to the self-force is small com-
pared with the energy of the particle. We note that these assumptions
are quite reasonable. First of all, the self-force as we see from (10.148)
is proportional to the acceleration (as well as higher derivative terms)
and, consequently, cannot exist in the absence of an external force.
We also note that a charged particle of finite size cannot be in stable
equilibrium unless external forces are applied. Therefore, the first
assumption is quite reasonable. Since τ is very small, the energy loss
due to radiation in any finite amount of time can only be small. To
see this quantitatively, let us consider an one dimensional charged
oscillator in the presence of the self-force.
mẍ + mω02 x = mτ v̈,
or, ẍ + ω02 x = τ v̈, (10.155)
where ω0 is the natural frequency of the oscillator. This is a third
order equation in the time derivatives (recall that v = ẋ). Choosing
a solution of the form
x(t) = x(0) e−iωt , (10.156)
we obtain, from (10.155)
ω 3 τ − iω 2 + iω02 = 0. (10.157)
This can also be rewritten as
(ωτ )3 − i(ωτ )2 + i(ω0 τ )2 = 0. (10.158)
We note that the natural dimensionless variables in this equation are
ωτ and ω0 τ . Assuming that ω0 τ ≪ 1, we can obtain the solution to
the cubic equation in (10.158) perturbatively as
i
ωτ = ±ω0 τ − (ω0 τ )2 + O (ω0 τ )3 ,
2

ωτ = i + i(ω0 τ )2 + O (ω0 τ )3 . (10.159)
We can discard the last solution as unphysical since it leads to an
exponentially growing solution. The other two solutions, on the other
hand, lead to
1 2
x(t) = x(0) e∓iω0 t e− 2 ω0 τ t , t > 0. (10.160)
Both these solutions are exponentially damped. We recall that the
radiation component of the electric field is proportional to the accel-
eration (ẍ) which is proportional to ω02 and, consequently we expect
the radiation loss to be small. Furthermore, taking the Fourier trans-

formation, we find
Z∞
1
x(ω) = dt eiωt x(t)
2π
0
Z∞
x(0) i
dt ei(ω∓ω0 + 2 ω0 τ )t
2
=
2π
0
ix(0)
= . (10.161)
2π (ω ∓ ω0 ) + 2i ω02 τ
This shows that the oscillation of the charged particle no longer con-
sists of a sharp single frequency. The intensity of the oscillations is
obtained to be
|x(0)|2
|x(ω)|2 =
ω04 τ 2
. (10.162)
4π 2 (ω ∓ ω0 )2 + 4
This shows a resonant behavior which characterizes the broadening

of the line width due to radiation reaction (or the self-force).
1. The coordinates of a particle with charge q, moving in the x − y

plane, depend on time as
t2 t2
x = x0 e− a2 , y = y0 e− b2 ,
dP
where x0 , y0 , a, b are constants. Determine dΩ for the radiated
wave.
2. A particle moves along the z-axis with a time dependence
z = a cos ωt,
where a, ω are constants. Show that the angular distribution of

the average radiated power is given by
dP q 2 cβ 4 (4 + β 2 cos2 θ) sin2 θ
h i= 7 .
dΩ 32π 2 a2 (1 − β 2 cos2 θ) 2
3. Show that, for a charged particle with a linear acceleration

a (namely, velocity is parallel to acceleration), the “time cor-
rected” total radiated power is given by
2q 2 a2 1
P = .
3c (1 − β 2 )3
3
(What this means is that you should correct for the retarded
time and define
dP dP dt′
→ ,
dΩ dΩ dt
and integrate this to obtain the total “time corrected” power
radiated.)
4. For a real function, f (t), show from the definition of the Fourier
transformation
Z ∞
1
f (ω) = dt eiωt f (t),
2π −∞
that
f ∗ (−ω) = f (ω).
Using this, derive the relation used in this chapter that
Z ∞ Z ∞
2
dt |f (t)| = 4π dω |f (ω)|2 .
−∞ 0
5. Consider a transparent medium with an index of refraction n =

1.5 in the range of visible light. Calculate the angle for the
emission of Čerenkov radiation by an electron moving with a
speed 0.9c. Determine the number of photons of wavelength in
the interval 4000 − 6000 Angstroms, emitted per unit length of
the trajectory.
6. A classical relativistic theory of the electron (due to Dirac)

describes the electron motion by
dpµ µ µ
= fext + fself ,
dτ
where pµ = muµ with uµ representing the four velocity of the
particle. m, τ denote respectively the rest mass and the proper
µ µ
time of the electron. Similarly, fext , fself denote respectively the
relativistic generalization of the external force and the self-force.
From the fact that a relativistic force has to satisfy uµ f µ = 0

(discussed in chapter 12), show that
ν
µ 2e2 d2 pµ pµ dp dpν
fself = − 2 2 .
3mc3 dτ 2 m c dτ dτ
Chapter 11
Plasma
11.1 General features of a plasma
Conventionally, a partially or a fully ionized gas is called a plasma.

The ionosphere in our atmosphere and the ionized gas in a discharge
tube such as a diode provide examples of a plasma. However, one
can generalize the definition and think of a plasma as a medium
consisting of (free) positive and negative charges such that in any
arbitrary volume the total charge is zero. That is, a plasma defines
a charge neutral medium. If we consider a plasma of ionized gas,
say for example, then while both the electrons as well as the posi-
tively charged ions are free to move, by virtue of their large mass,
the positive charges (ions) do not move very much. Thus, we can
think of only the negatively charged electrons in a plasma to have
appreciable motion. In some sense, therefore, we can think of the
plasma as consisting of a large number of electrons moving freely in
a positively charged background. This seems very much like the free
electron theory of metals (in condensed matter physics). However,
there are essential differences. First, the number density of electrons
in a plasma (either in the laboratory or in nature) is much smaller
than that in a metal. As a result, a classical description of a plasma
leads to quite accurate results. Second, since the electron density is
so dilute in a plasma, the effects of collision can truly be neglected.
A plasma, therefore, appears to be a highly conducting medium.
However, as we will see, it has very different behavior from the con-
ductors that we have studied so far. A plasma in equilibrium develops
strong restoring forces when disturbed externally. For example, let us
suppose that we displace the electrons in a block of an infinite plasma
(for simplicity) by an infinitesimal distance x along the x-axis. This
would then lead to two charged surfaces, one positively charged on
the left and the other negatively charged on the right as shown in Fig.
11.1. Each of these surfaces will have equal, but opposite charge. For
373
374 11 Plasma
+±±±±±±±±−
+±±±±±±±±−
+±±±±±±±±−
+±±±±±±±±−
+±±±±±±±±−
+±±±±±±±±− x
Figure 11.1: The two charged surfaces which develop when the elec-
trons are displaced to the right along the x-axis.
example, if we assume the number density of electrons in the plasma

to be N (which will also represent the number density for positively
charged ions), then the surface charge density on the two surfaces will
have the magnitude N ex. (We are assuming that e > 0.) Each of
these surfaces will produce an electric field that can be calculated us-
ing the methods of electrostatics. In fact, we have already calculated
the electric field produced by an infinite charged surface and taking
over the results from (2.79) (for a single plane the field is derived in
(1.42)), we obtain that such a displacement will generate an electric
field (only) within the two surfaces of the form
E = 4πN ex êx . (11.1)
The motion of the electrons (within the two charged surfaces) along
the x-axis will now be subjected to a force leading to (e > 0)
mẍ = −eE = −4πN e2 x. (11.2)
We recognize that this is the equation for a harmonic oscillator with

a natural frequency given by
4πN e2
ωp2 = . (11.3)
m
This is known as the plasma frequency and this analysis shows that,
because of this displacement (disturbance), the plasma will begin to
oscillate as (this is the complex notation and the coordinate of the
particle will be given by the real part)
x(t) ∼ e−iωp t . (11.4)
Thus, we see that, in this simple case, the plasma will oscillate only
with the plasma frequency. Although this is not exactly the “plasma
oscillation” that one talks about in connection with a plasma (which
11.1 General features of a plasma 375
we will discuss next), this simple example illustrates how the plasma
tries to maintain its neutrality when disturbed slightly.
If the plasma is subject to a driving force with a given angu-
lar frequency, then it can oscillate with a frequency different from
the plasma frequency (11.3). For example, suppose we have a har-
monic electric field of the form (we are suppressing the coordinate
dependence for simplicity)
E(t) = E(0) e−iωt , (11.5)
acting on the plasma (with ω representing the frequency of the har-

monic field), then the equation for an electron in the plasma would
become
mẍ = −eE(t) = −eE(0) e−iωt . (11.6)
The solution, in this case, clearly would have the form

e e
x(t) = x(0) e−iωt = 2
E(0)e−iωt = E. (11.7)
mω mω 2
Namely, in this case the plasma will oscillate with the frequency of
the driving force. It also follows from Eq. (11.7) that we can obtain
the velocity of motion to be
ie
v=− E. (11.8)
mω
This is, in fact, very suggestive in that we see that the current asso-
ciated with the motion of a single electron can be written as
ie2 iωp2
j = −ev = E= E, (11.9)
mω 4πN ω
where we have used the definition of the plasma frequency in (11.3).
Relation (11.9) is very interesting in that it is reminiscent of the
Ohm’s law,
iN e2 iωp2
J = N j = −N ev = E= E = σ E, (11.10)
mω 4πω
except for the factor of “i” in the proportionality constant. This fac-
tor simply implies that the electric field and the current are not in
phase. In this sense, a plasma is somewhat like a dielectric medium
and, as we will see shortly, this phase difference has important con-
sequences.
376 11 Plasma
In a plasma, of course, there are always random thermal oscilla-

tions. When we can neglect thermal motions (because of their small
magnitude at low temperatures or because of averaging of the random
thermal motion), we can talk of a cold plasma. In such a case, the
effect due to the driving force is dominant and (11.9) should hold. If
the plasma is at a high temperature so that the thermal oscillations
are not negligible, then one must use a statistical description of the
plasma. Nonetheless, relations of the form (11.9) turn out to be fairly
accurate within such derivations.
Let us next analyze a simple example within the context of a
plasma at finite temperature. Let us consider a plasma in thermal
equilibrium at a temperature T . Let us introduce a single static
charged particle carrying a charge q into the plasma. This will gener-
ate a static electric field in the plasma and we would like to determine
this field. Since the electric field is static, in order to calculate the
field, we only need to determine the scalar potential Φ which will be
spherically symmetric.
In the presence of the electric field generated by the static charge,
the charged particles in the plasma will experience a force and, con-
sequently, will rearrange themselves so as to attain a configuration
of minimum (free) energy. We know that at a finite temperature T ,
the equilibrium density of particles is determined by the Boltzmann
E
factor e− kT , where E, k represent the energy of the particle and the
Boltzmann constant respectively. Thus, in the presence of the charge,
the electron and the ion densities will take the forms
eΦ
Ne = N e kT ,
eΦ
Ni = N e− kT , (11.11)
E
where N = e− kT denotes the equilibrium density of the charged
particles in the plasma before the introduction of the external charge.
The scalar potential (and, therefore, the electric field) away from the
location of the charged particle can be determined from Gauss’ law
(the first of the Maxwell’s equations), namely,
∇2 Φ = −4πρ,

1 d 2 d
or, r Φ = −4πe(Ni − Ne ),
r 2 dr dr

d2 Φ 2 dΦ eΦ
or, 2
+ = 8πN e sinh . (11.12)
dr r dr kT
11.2 Plasma oscillation 377
Here, we have used the spherical symmetry of the scalar potential.

Equation (11.12) is difficult to solve in closed form, in general.
However, let us assume that we have a plasma in equilibrium at a
very high temperature kT ≫ eΦ. In this case, we can approximate
the right hand side in (11.12) and write
d2 Φ 2 dΦ 8πN e2
+ − Φ = 0. (11.13)
dr 2 r dr kT
We recognize this as the spherical Bessel equation of order zero and
the solution that vanishes asymptotically has the form
C − rr
Φ(r) = e D, (11.14)
r
where C is a constant and we have defined the Debye length as
r s
kT kT
rD = = . (11.15)
8πN e2 2mωp2
The electric field is now easily obtained from
C(r + rD ) − rr
E = −∇ Φ = r̂ e D. (11.16)
r 2 rD
Thus, we see that the scalar potential as well as the electric field fall
off rapidly for r > rD . Namely, the charged particles of the plasma
will reorganize themselves so as to screen the external charge beyond
the Debye length. For this reason, the Debye length is also sometimes
referred to as the screening length. These examples illustrate that
when a plasma is disturbed, it tries to restore its charge neutrality.
11.2 Propagation of electromagnetic waves through a plasma
Let us next consider the question of propagation of electromagnetic

waves through a plasma. We assume that a harmonic electromagnetic
wave of frequency ω is incident on a plasma. Let the electron density
of the plasma in equilibrium be Neq which will also be the density of
positive ions in equilibrium. The incident electromagnetic wave will
set up a local fluctuation in the electron density (we are assuming
that the positive ions do not have appreciable motion) so that we can
write
Ne (x, t) = Neq + n(x, t), (11.17)

378 11 Plasma
where the fluctuation from the equilibrium value, n(x, t), is consid-
ered to be small. As we have already seen, there will be oscillations
in the plasma leading to a current density (see (11.9))
iNeq e2 iωp2
J = Ne (x, t) j ≈ E= E, (11.18)
mω 4πω
where we are neglecting terms quadratic in the fluctuations, an ap-
proximation which is also known as the linearized approximation
(namely, since the velocity v is already a fluctuation, n(x, t)v is
quadratic in the fluctuations).
The first of the Maxwell’s equations, in this case, takes the form
∇ · E = 4πe(Ni − Ne ) = −4πen(x, t) = 4πρ(x, t), (11.19)
where we have defined ρ(x, t) = −en(x, t) to correspond to the fluc-

tuation in the electron charge density. The other Maxwell’s equations
have the forms
∇ · B = 0,
iω
∇×E= B,
c
!
4π iω iω ωp2
∇×B= J− E=− 1− 2 E, (11.20)
c c c ω
where we have used (11.18). The last equation in (11.20) suggests

that we can define a permittivity for the plasma of the form
ωp2
ǫp = 1 − . (11.21)
ω2
It is worth noting that ǫp ≤ 1 as opposed to the case of a dielectric
for which ǫ ≥ 1.
There are now several interesting cases to be discussed. If the
frequency of the harmonic field coincides exactly with that of the
plasma, namely, if ω = ωp , then we have
ǫp = 0. (11.22)
In this case, the second and the fourth of Maxwell’s equations give
∇ · B = 0,
∇ × B = 0. (11.23)
Namely, the electron current, in this case, exactly cancels the dis-
placement current so that the curl of the magnetic field vanishes. In
the absence of sources, these two equations imply that the magnetic
field vanishes identically, namely,
B = 0. (11.24)
The other two of Maxwell’s equations take the forms
∇ · E = 4πρ(x, t),
∇ × E = 0. (11.25)
These are equations of the type in electrostatics implying that the

electric field can be expressed as the gradient of a scalar potential.
The difference is that here the fluctuations in the electron density
have a time dependence leading to a time dependence of the electric
field.
This is the case normally referred to as the “plasma oscillations”.
This can be seen more clearly as follows. The continuity equation
gives
∂ρ
+ ∇ · J = 0,
∂t
∂ρ
or, + iωp ρ = 0. (11.26)
∂t
Here, we have used Eqs. (11.18) as well as (11.19) (remember ω =
ωp ). Taking the time derivative one more time, we obtain
∂2ρ
+ ωp2 ρ = 0. (11.27)
∂t2
Namely, the density of electrons fluctuates in time with the plasma
frequency. This is a cooperative phenomenon in the sense that,
not one electron, but the plasma of electrons as a whole oscillates.
However, there is no traveling disturbance that is generated. This
should be contrasted with the transverse traveling wave solutions of
Maxwell’s equations for which
∇ · E = 0,
corresponding to the fact that there is no charge density. (It is worth

noting here that in a conducting medium, conductivity σ is real lead-
t
ing to a dissipative solution ρ(t) ∼ e− τ as we have seen in (8.55). In
380 11 Plasma
contrast in the case of a plasma the conductivity σ can be thought

of as purely imaginary as noted in (11.10) which is the reason for the
oscillatory solution.)
When ω 6= ωp , there can be traveling waves in the plasma. This
can be seen as follows. We note that when ω 6= ωp , then we can
truly associate the permittivity ǫp in (11.21) with the plasma. For
example, we see that for a harmonic field, the continuity equation
together with (11.18) leads to
iωp2
− iωρ = −∇ · J = − (∇ · E),
4πω
ωp2
or, ρ= (∇ · E). (11.28)
4πω 2
Using this in (11.19) as well as the definition in (11.21), we see that
we can write the Maxwell’s equation, in this case, as
∇ · (ǫp E) = 0,
∇ · B = 0,
iω
∇×E= B,
c
iǫp ω
∇×B=− E, (11.29)
c
where we have assumed µp = 1. Equations (11.29) coincide exactly
with (source free) Maxwell’s equations in a dielectric of permittiv-
ity ǫp . (Plasma behaves like an isotropic dielectric medium since ǫp
is a scalar.) It follows from (11.29) that both the electric and the
magnetic fields satisfy the wave equation
ǫp ω 2
∇2 E + E = 0,
c2
ǫp ω 2
∇2 B + B = 0. (11.30)
c2
These have traveling wave solutions of the form
E(x) ∼ eik·x , B(x) ∼ eik·x , (11.31)
where
ǫp ω 2
k 2 = k2 = ,
c2
or, c2 k 2 = ω 2 − ωp2 . (11.32)
We have used here the definition of the permittivity given in (11.21).

Relation (11.32) is interesting in the sense that it has exactly the
same form as the wave number that we have already seen in the case
of a wave guide. It implies that when ω > ωp , there is propagation of
the electromagnetic wave inside the plasma without any attenuation.
However, there cannot be any propagation of electromagnetic waves
inside the plasma when ω < ωp . Therefore, we see that ωp is the
analogue of the cut-off frequency in a wave guide. There is a difference
though. Unlike the wave guide, where there is loss of energy through
dissipation when ω < ωc , here the non-propagation does not imply
dissipation. Rather, this simply implies that the incident wave is
totally reflected in such a case. In fact, recalling that the index of
refraction for a medium is defined as
√
n = ǫµ,
we note that the index of refraction for a plasma (µp = 1),

s
√ ωp2
n p = ǫp = 1 − 2 , (11.33)
ω
becomes purely imaginary when ω < ωp . Taking the results of reflec-
tion from a dielectric at normal incidence (see (6.51)), for simplicity,
we see that, in the case of reflection of such a harmonic wave from a
plasma at normal incidence, we have
Er 1 − np
= . (11.34)
Ei 1 + np
Here we are assuming that the wave is traveling in vacuum before be-
ing reflected from the plasma. Note that since np is purely imaginary
for ω < ωp , the right hand side of (11.34) has the absolute magnitude
unity, implying that the coefficient of reflection is unity. Therefore,
the wave is totally reflected.
Thus, we see that an electromagnetic wave is totally reflected
from a plasma when ω < ωp . Using the definition of ωp in (11.3), this
condition is also sometimes expressed by saying that electromagnetic
waves will reflect from a plasma if the number density of electrons
will satisfy
mω 2
N> , (11.35)
4πe2
or simply if the plasma is overdense. This can happen if the density
of the plasma is high or the frequency of the harmonic wave is low.
382 11 Plasma
This is, in fact, the basic principle used in transmitting low frequency
radio waves. Let us recall that the ionosphere consists of a plasma
of electrons and positive ions where the ionization is a consequence
of radiations coming from the sun. As a result, low frequency radio
waves cannot penetrate the ionosphere and are reflected back, leading
to a transmission of such waves around the globe. The theory of
propagation of electromagnetic waves in the ionosphere is, however,
slightly more involved owing to the fact that the density of electrons
in the ionosphere is not uniform. Rather, it changes with the height
from the surface of the earth.
11.3 Motion of the positive ions
We have so far neglected the motion of the positive ions because

of their heavy mass. Under certain circumstances, however, their
motion becomes important as we will see later. Let us see how the
motion of the positive ions can be included into our analysis and how,
under the conditions that we have assumed, their contribution can
be neglected in the earlier analysis.
Let us recall from (11.6) and (11.8) that the motion of electrons
in a plasma driven by a harmonic electric field leads to
me v̇e = −eE,
ie
ve = − E. (11.36)
me ω
Similarly, the motion of the positive ions leads to
mi v̇i = eE,
ie
vi = E. (11.37)
mi ω
We have used the subscripts “e” and “i” to represent the respective
quantities associated with electrons and ions. We note now that
the total current density associated with the plasma (including the
contribution due to positive ions) takes the form

iN e2 1 1
J = N e (vi − ve ) = + E
ω mi me
i 2 2

= ωp,i + ωp,e E = σ E, (11.38)
4πω
11.4 Effect of a background magnetic field 383
where we have defined the plasma frequencies associated with the

electrons and the ions respectively as
2 4πN e2
ωp,e = ,
me
2 4πN e2 2
ωp,i = ≪ ωp,e . (11.39)
mi
Correspondingly, when the motion of the positive ions is taken into
account, the conductivity can be written as
i 2 2

σ= ωp,e + ωp,i . (11.40)
4πω
The fact that the conductivity σ is a scalar simply signifies that
the plasma behaves like an isotropic medium. If we now substitute
(11.40) into the Maxwell’s equations, we can derive, in this case, that
the permittivity of the plasma takes the form
1 2 2

ǫp = 1 − 2
ωp,e + ωp,i . (11.41)
ω
Namely, the permittivity also continues to be a scalar signifying that
the refractive index of the isotropic medium has the form
r
√ 1 2 2 .

np = ǫp = 1 − 2 ωp,e + ωp,i (11.42)
ω
We note now from Eq. (11.39) that
2 2
ωp,e ≫ ωp,i , (11.43)
since mi ≫ me (recall that a proton is about 2000 times heavier

than the electron and the positive ion will be at least as heavy as the
proton). Consequently, in our earlier analysis we can safely neglect
the motion of the positive ions. However, there may be situations
where the motion of the positive ions plays an important role and we
will discuss such an example later.
11.4 Effect of a background magnetic field
Let us next consider the behavior of a plasma in the background of a

constant magnetic field. This is quite important in the analysis of the
propagation of electromagnetic waves through the ionosphere. This
is because there is a magnetic field associated with earth and, con-
sequently, the plasma in the ionosphere is subjected to this constant
384 11 Plasma
background magnetic field. As we will see, this changes the nature

of the plasma as a medium.
Let us concentrate on the motion of the electron. In the presence
of a driving electric field which is harmonic and a constant magnetic
field, the equation for the electron takes the form
v
mv̇ = −e E + × B̄ ,
c
ie v
or, v = − E + × B̄ , (11.44)
mω c
where we have used the fact that the velocity has to be harmonic (with
the same frequency) since the driving electric field is. Here, we have
denoted the background magnetic field by B̄. (As a parenthetical
remark, let us note that a driving electromagnetic field can also lead
to a Lorentz force. However, since the velocity is already a fluctuation
as is the driving magnetic field, such a term is neglected under our
approximation of linearization. This is why we did not have a Lorentz
force in the earlier analysis.)
Since the velocity occurs on the right hand side of (11.44), solv-
ing for the velocity is a little more involved. To invert the relation,
let us define the cyclotron frequency associated with the background
magnetic field as
eB̄
Ω= . (11.45)
mc
In terms of this, the velocity can be expressed as

Ω ie
v + iv × =− E, (11.46)
ω mω

ie 1 i
or, v = − E − 2 (Ω · E) Ω − (E × Ω) ,
mω(1 − ( Ω 2
ω) )
ω ω
where Ω2 = Ω2 . The inversion in (11.46) is best done in matrix form.

We note that we can write (11.46) in component form as
ie
Pij vj = − Ei , (11.47)
mω
where
i
Pij = δij + ǫijk Ωk . (11.48)
ω
The inverse of this matrix can be determined in a straightforward

manner and corresponds to

−1 1 Ωi Ωj i
Pij = 2 δij − − ǫijk Ωk , (11.49)
1− Ω ω2 ω
ω
which leads to the solution of (11.47) in components to be

ie −1
vi = − P Ej
mω ij

ie 1 i
=− δij − Ω Ω
2 i j
− ǫ ijk k Ej
Ω (11.50)
mω(1 − ( Ω
ω )2) ω ω

ie e2 ie
=− δij − 2 2 2 B̄ i B̄ j − ǫijk B̄ k Ej ,
eB̄ 2
mω(1 − ( mcω ) ) m c ω mcω
2 2
where we have defined B̄ = B̄ and this can be compred with the
solution obtained in (11.46).
Relation (11.50) is interesting because it shows that even though
there is a relation between the components of the velocity and the
electric field, the proportionality constant is a tensor in the presence
of a background magnetic field. Consequently, in this case, the cur-
rent density takes the form
Ji = −N evi = σij Ej , (11.51)
where

iN e2 1 i
σij = δij − 2 Ωi Ωj − ǫijk Ωk
mω(1 − ( Ω 2
ω) )
ω ω

iωp2 1 i
= δij − 2 Ωi Ωj − ǫijk Ωk . (11.52)
4πω(1 − ( Ω 2
ω) )
ω ω
Substituting (11.51) and (11.52) into the Maxwell’s equations (as well
as using the continuity equation), we find that in the present case,
they take the forms (repeated indices are summed)
∇i (ǫp, ij Ej ) = 0,
∇i Bi = 0,
iω
(∇ × E)i = Bi ,
c
iω
(∇ × B)i = − ǫp, ij Ej , (11.53)
c
386 11 Plasma
where the permittivity also has a tensor structure given by

4πσij
ǫp,ij = δij −
iω

ωp2 Ωi Ωj i
= δij − 2 δij − − ǫijk Ωk . (11.54)
(ω − Ω2 ) ω2 ω
Thus, we see that in the presence of a magnetic field, the con-

ductivity as well as the permittivity become complex tensors. We
are already familiar with the complex nature of such quantities. The
tensor structure is new and simply signifies that the plasma ceases
to behave like an isotropic medium in the presence of a background
magnetic field. Namely, the properties of propagation depend on the
direction and the magnetic field is responsible for this anisotropy. We
note from (11.53) that, as is the case with harmonic fields, we need
to concentrate only on the last two equations. (The first two are
consequences of the last two.) Let us choose a traveling plane wave
solution of the form
Ei ∼ eik·x , Bi ∼ eik·x . (11.55)
Then, taking the curl of the third equation in (11.53) (and using the
fourth), we obtain
ω2
δij k 2 − ki kj Ej = 2 ǫp ij Ej ,
c

2 ki kj
or, ǫp ij − n δij − 2 Ej = 0, (11.56)
k
where we have defined the refractive index for the medium as

c2 k 2
n2 = . (11.57)
ω2
A nontrivial solution of (11.56) would exist only if the determi-
nant of the coefficient matrix vanishes. Let us analyze this a little
carefully. To simplify the analysis, let us assume that the magnetic
field B̄ lies along the z-axis. In this case, the only non-vanishing
component of the cyclotron frequency is given by Ωz = Ω3 = Ω.
Furthermore, let us define the dimensionless quantities
ωp2 Ω
X= ω2
, Y = ω, (11.58)
X XY
ǫ0 = 1 − X, ǫ1 = 1 − 1−Y 2 , ǫ2 = 1−Y 2 ,
so that the permittivity can be written in the simple form

 
ǫ1 iǫ2 0
 
ǫp ij =  −iǫ2 ǫ1 0 . (11.59)
0 0 ǫ0
We note that without any loss of generality, we can choose our coor-
dinate axes (x and y) such that the direction of propagation lies in
the y − z plane, namely,
k = k(0, sin θ, cos θ). (11.60)
With this, we can represent

 
n2 0 0
ki kj  
n2 δij − = 0 n2 cos2 θ −n2 sin θ cos θ  .
k2
0 −n2 sin θ cos θ n2 sin2 θ
(11.61)
Consequently, we are interested in the vanishing of the determinant

of the matrix

2 ki kj
ǫp ij − n δij − 2
k
 
ǫ1 − n 2 iǫ2 0
 
=  −iǫ2 ǫ1 − n2 cos2 θ n2 sin θ cos θ  .

0 n2 sin θ cos θ ǫ0 − n2 sin2 θ
(11.62)
Let us look at the solutions for the vanishing determinant in two

special cases. First, if θ = 0, namely, if the direction of propagation
is along the direction of the magnetic field, we see from (11.62) that
the vanishing of the determinant gives

ǫ0 (ǫ1 − n2 )2 − ǫ22 = ǫ0 n2 − ǫ1 − ǫ2 n2 − ǫ1 + ǫ2 = 0.
(11.63)
If ǫ0 = 0, the index of refraction is undetermined and we see from

(11.58) that this happens when ω = ωp . Namely, in this case, the
plasma oscillates unaffected by the presence of the magnetic field
(since the motion is parallel to the direction of the magnetic field).
388 11 Plasma
In fact, it is easy to see from (11.56) that, in this case, Ex = 0 = Ey

and only Ez is nonzero so that the electron motion is along the z-axis
and it does not feel the presence of the magnetic field. On the other
hand, if ω 6= ωp (namely, if ǫ0 6= 0), then we have
X XY X(1 ∓ Y )
n 2 = ǫ1 ± ǫ2 = 1 − ± =1−
1−Y2 1−Y2 1−Y2
X ωp2
=1− =1− , (11.64)
1±Y ω(ω ± Ω)
where we have used (11.58). Note that, in this case, it follows from
(11.56) that Ez = 0 and since the electric field is perpendicular to
the direction of propagation (and the magnetic field), the electrons
feel the effect of the magnetic field.
The other special case is when the direction of propagation is
perpendicular to the direction of the magnetic field, namely, θ = π2 .
In this case, we see from (11.62) that the vanishing of the determinant
leads to

(ǫ0 − n2 ) ǫ1 (ǫ1 − n2 ) − ǫ22 = n2 − ǫ0 n2 ǫ1 − ǫ21 + ǫ22 = 0.
(11.65)
The refractive index of the medium is now determined to be

ωp2
n 2 = ǫ0 = 1 − ,
ω2
ǫ21 − ǫ22 X X 2Y 2
or, n2 = =1− −
ǫ1 1 − Y 2 (1 − Y 2 )(1 − X − Y 2 )
2
ω 2 1 − ωp
X(1 − X) p ω 2
=1− 2
=1− 2 . (11.66)
1−X −Y ω − ωp − Ω2
2
The first case is interesting in that the refractive index is insensitive

to the magnetic field. This is understood from the fact that Eq.
(11.56), in this case, leads to Ex = 0 = Ey . Only Ez is nonzero so
that the motion of the electrons is along the direction of the magnetic
field. Correspondingly, they do not feel its effect. In the second case,
on the other hand, Ez = 0 and the electric fields are transverse to the
direction of the magnetic field and, consequently, the electrons do feel
its effect. Although one can solve the vanishing of the determinant
for an arbitrary angle θ, these two special cases show the variation of
the refractive index with the direction of propagation.
11.5 Faraday rotation 389
11.5 Faraday rotation
Let us go back to Eq. (11.64) and analyze the result in some more
detail. We see that when the direction of propagation is along the
direction of the magnetic field (which we have chosen to correspond
to the z-axis), Ez = 0 if ω 6= ωp . Let us denote the two eigenvalues
for the index of refraction, in this case, as
ωp2
n2± = ǫ1 ± ǫ2 = 1 − . (11.67)
ω(ω ± Ω)
The meaning of the two eigenvalues becomes clear once we recognize

that since Ez = 0, we can restrict ourselves to the study of the matrix
in (11.56) (or (11.62)) to the upper left 2 × 2 space. In this space, for
example for n = n+ , we have
!
2 kα kβ ǫ1 − n2+ iǫ2
Aαβ = ǫαβ − n+ δαβ − 2 =
k −iǫ2 ǫ1 − n2+
! !
ǫ1 − (ǫ1 + ǫ2 ) iǫ2 1 −i
= = −ǫ2 ,
−iǫ2 ǫ1 − (ǫ1 + ǫ2 ) i 1
(11.68)
where α, β = 1, 2. It is immediately clear from the structure of this

matrix that
! ! !
1 1 1
A = −2ǫ2 , A = 0. (11.69)
i i −i
On the other hand, for n = n− , we have

!
kα kβ 1 i
Aαβ = ǫαβ − n2− δαβ − 2 = ǫ2 . (11.70)
k −i 1
In this case, we find that

! ! !
1 1 1
A = 0, A = 2ǫ2 . (11.71)
i −i −i

1 1
The vectors and denote respectively basis vectors for right
−i i
circularly polarized and left circularly polarized waves.
390 11 Plasma
Let us next recall that right and left circularly polarized waves
traveling along the z-axis can be represented respectively as (see, for
example, (6.37))
ER = (x̂ − iŷ) E0 e−iωt+ikR z ,
EL = (x̂ + iŷ) E0 e−iωt+ikL z . (11.72)
It follows from this as well as Eqs. (11.69) and (11.71) that traveling
along the direction of the magnetic field in a plasma, the right and
the left circularly polarized waves will suffer different rotations of
phase – the right circularly polarized wave rotating with the index
of refraction n+ while the left circularly polarized wave rotates with
the index of refraction n− . The medium responds differently to right
and left circularly polarized waves. This phenomenon is commonly
known as the Faraday rotation.
We note that in vacuum, a linearly polarized wave can always
be written as a sum of a right and a left circularly polarized wave. If
such a wave, initially linearly polarized along the x-axis, is incident
along the z-axis (the direction of the magnetic field) on a plasma (at
z = 0), then traveling through the plasma, the planes of polarization
of the right and the left circularly polarized waves will rotate as
1
E= (ER + EL )
2
E0 −iωt h i
= e (x̂ − iŷ) eikR z + (x̂ + iŷ) eikL z , (11.73)
2
where
n+ ω n− ω
kR = , kL = . (11.74)
c c
As a result, in traveling through a certain distance z, the tilt in the
polarization will be given by
1 ω
ψ = (kR − kL )z = (n+ − n− )z. (11.75)
2 2c
If the harmonic frequency is high compared to the plasma frequency
as well as the cyclotron frequency, ω ≫ ωp , Ω, then we obtain from
Eqs. (11.67) that

ωp2 Ω
n± ≈ 1 − 2 1∓ . (11.76)
2ω ω
11.6 Alfvén waves 391
This leads to
ω ωp2 Ω
ψ= (n+ − n− )z ≈ z
2c 2cω 2
2πe3 B̄
= N z. (11.77)
m2 c2 ω 2
This shows that the tilt in the polarization planes is proportional to
the distance traveled in the plasma as well as to the number density
of electrons in the plasma. This has, of course, been derived assuming
that the electron density is a constant in the plasma. If the density
changes with distance, as is the case in the ionosphere, then the tilt
is obtained to be
Z z
2πe3 B̄
ψ= 2 2 2 dz ′ N (z ′ ). (11.78)
m c ω 0
In either case, it is clear that the number density of electrons in a
plasma (or the total number of electrons in a volume of unit height)
can be determined from a study of the tilt in the polarization planes.
11.6 Alfvén waves
As we have seen earlier, in the absence of a magnetic field the effect of

the positive ions can be safely neglected. In the presence of a magnetic
field, however, the positive ions play an important role under certain
circumstances which we would like to discuss.
Including the contributions of the positive ions to our analysis of
section 11.4 is quite easy. First, let us define the mass ratio between
the electron and the ion as
me
η= ≪ 1. (11.79)
mi
With this, we note that we can identify
eB̄
Ωi = = ηΩe ,
mi c
2 4πN e2 2
ωp,i = = ηωp,e ,
mi
2
ωp,i
Xi = = ηXe ,
ω2
Ωi
Yi = = ηYe . (11.80)
ω
392 11 Plasma
We recognize that the positive ions will also contribute to the conduc-
tivity tensor as well as to the permittivity. The qualitative structures
of these tensors will be the same and, if we choose the magnetic field
to be along the z-axis, we can write the permittivity tensor as in
(11.59)
 
ǫ1 iǫ2 0
 
ǫp,ij =  −iǫ2 ǫ1 0  , (11.81)
0 0 ǫ0
where
ǫ0 = 1 − Xe − Xi = 1 − (1 + η)Xe
≈ 1 − Xe ,
Xe Xi
ǫ1 = 1 − −
1 − Ye2 1 − Yi2
(1 + η)Xe (1 − ηYe2 )
=1−
(1 − Ye2 )(1 − η 2 Ye2 )
Xe (1 − ηYe2 )
≈1− ,
(1 − Ye2 )(1 − η 2 Ye2 )
Xe Ye Xi Yi
ǫ2 = −
1 − Ye2 1 − Yi 2
(1 − η 2 )Xe Ye
=
(1 − Ye2 )(1 − η 2 Ye2 )
Xe Ye
≈ . (11.82)
(1 − Ye )(1 − η 2 Ye2 )
2
We recognize that η ≪ 1 and it is clear from (11.82) that when

ω 2 ≫ ηΩ2e = Ωe Ωi (or equivalently, ηYe2 ≪ 1), we can neglect the
contributions from the positive ions. This is exactly like the analysis
before. However, at very low frequencies, ω ≪ Ωi = ηΩe , or equiva-
lently, when ηYe ≫ 1, the contributions from the positive ions play
an important role. We see from (11.82) that in this limit, we can
write
ǫ0 ≈ 1 − Xe ,
Xe (1 − ηYe2 )
ǫ1 ≈ 1 −
(1 − Ye2 )(1 − η 2 Ye2 )
11.6 Alfvén waves 393
Xe (−ηYe2 ) Xe
≈1− 2 2 2
=1+ ,
(−Ye )(−η Ye ) ηYe2
Xe Ye
ǫ2 ≈ 2
(1 − Ye )(1 − η 2 Ye2 )
Xe Ye Xe
≈ = 2 3. (11.83)
(−Ye2 )(−η 2 Ye2 ) η Ye
Since Ye ≫ 1, it follows that
ǫ1 ≫ ǫ2 , (11.84)
and, therefore, ǫ2 can be neglected in all our manipulations.
The relation (11.84) is significant in simplifying all the expres-
sions. For example, we can now write the matrix (11.62) as

2 ki kj
ǫp,ij − n δij − 2
k
 
ǫ1 − n 2 0 0
 
= 0 ǫ1 − n2 cos2 θ n2 sin θ cos θ  . (11.85)
0 n2 sin θ cos θ ǫ0 − n2 sin2 θ
The vanishing of the determinant of this matrix is now easily seen to
give
2
ωp,e
Xe
n 2 = ǫ1 ≈ 1 + = 1 + ,
ηYe2 ηΩ2e
ǫ1 ǫ0
n2 =
ǫ0 cos2 θ + ǫ1 sin2 θ
!
2
ωp,e 2 − ω2
ωp,e
≈ 1+ 2 . (11.86)
ηΩ2e 2 − (1 +
ωp,e
ωp,e
sin2 θ)
ηΩ2e
The electric fields corresponding to the two roots can be de-

termined from the form in (11.85). For n2 = ǫ1 , we note that
Ey = 0 = Ez while Ex is nontrivial. The electric field is trans-
verse to the magnetic field in this case. On the other hand, for the
second root, we see from (11.85) that Ex = 0 and
Ey n2 sin θ cos θ ǫ0
=− 2 2
= − cot θ. (11.87)
Ez ǫ1 − n cos θ ǫ1
In other words, in this case, the electric field has a component parallel
to the direction of the magnetic field in addition to a perpendicular
394 11 Plasma
component. Let us note from (11.50) that in this low frequency limit,
we can write for the first root (remember that the magnetic field is
along the z-axis),

ie ω2 iΩe e
ve,y ≈ − − 2 Ex = − Ex ,
me ω Ωe ω me Ω e

ie ω2 iΩi e
vi,y ≈ − 2 − Ex = − Ex . (11.88)
mi ω Ωi ω me Ω e
Namely, the component of the velocities perpendicular to the mag-

netic field is the same for both the electrons and the ions. (Note that,
for the positive ions, the magnetic interaction changes sign so that
the imaginary term in the inverse relation also changes sign, which is
why the velocities for the electron and the ion have the same sign.)
Similarly, for the second root, we obtain
e
ve,x ≈ Ey ,
me Ω e
e
vi,x ≈ Ey . (11.89)
me Ω e
Once again, we see that the velocity of the electrons and the ions
perpendicular to the direction of the magnetic field are the same. The
plasma, therefore, moves as a whole along the direction perpendicular
to the magnetic field.
These low frequency waves in a plasma are also known as the
Alfvén waves. We note, in particular that when ω ≪ Ωi and we have
2
further ωp,e ≫ ηΩ2e = Ωe Ωi (which is quite natural), the roots in
(11.86) take even simpler forms
2
ωp,e 2
ωp,e
n2 ≈ , . (11.90)
ηΩ2e ηΩ2e cos2 θ
Conventionally, the quantity
√
Ωe Ωi
vA = c, (11.91)
ωp,e
is also known as the Alfvén speed.
11.7 Collisions
We have so far talked about an idealized plasma where there is no

collision between the constituents. While this is the case for a very
11.7 Collisions 395
dilute plasma, in a realistic plasma such as in the ionosphere, there

are collisions that need to be taken into account. In a partially ion-
ized gas (or in the ionosphere), the electrons can undergo collisions
with charge neutral molecules. Charged particles can also “collide”
with one another, although the concept of “collision” is different in
such a case. The effect of the collisions can be introduced into the dy-
namical equations through a friction force. For example, the equation
of motion for the electron under the influence of an external electric
field, when collisions are taken into account, takes the form
mv̇ + mνv = −eE. (11.92)
Here, ν denotes the frequency of collisions which can be thought of

as the inverse of the average time interval of travel between collisions
for an electron.
For a harmonic driving force of frequency ω, we obtain from
(11.92)
− imωv + mνv = −eE,

ie
or, v = − E. (11.93)
m(ω + iν)
As a result, the electron current density can be written as
iN e2
J = −N ev = E = σE, (11.94)
m(ω + iν)
iN e2 iωp2 iωp2 (ω − iν)

σ= = = . (11.95)
m(ω + iν) 4π(ω + iν) 4π(ω 2 + ν 2 )
We note that this reduces to (11.10) when ν = 0. In the presence of

collisions, we see that the proportionality between the current density
and the electric field becomes complex which is the behavior of a
lossy dielectric. The real part of this proportionality constant can be
identified with the conductivity of the plasma and has the form
iωp2 (ω − iν) ωp2 ν

Re σ = Re = . (11.96)
4π(ω 2 + ν 2 ) 4π(ω 2 + ν 2 )
Substituting the form of the current density in (11.94) into the

Maxwell’s equations (and using (11.95)), we can determine the per-
396 11 Plasma
mittivity of such a plasma to be
4πσ ωp2 (ω − iν)

ǫp = 1 − =1−
iω ω(ω 2 + ν 2 )
ωp2 iωp2 ν
=1− + . (11.97)
(ω 2 + ν 2 ) ω(ω 2 + ν 2 )
The permittivity is complex signifying that propagation of electro-

magnetic waves in such a medium will be accompanied by attenua-
tion. The imaginary part of the permittivity, as is clear from (11.96)
and (11.97), is related to the conductivity. The analysis of the solu-
tions of the Maxwell’s equations can now be undertaken as we have
done earlier, but we will not go into the details of this.
1. i) For a laboratory plasma, the number density of electrons lies

between N = 1012 /cm3 − 1018 /cm3 . Determine ωp for such a
plasma. Calculate the same for the plasma in the ionosphere
at the F-level, where N ≃ 106 /cm3 .
ii) Determine the Debye length for a laboratory plasma at T =

2000◦ K.
2. Consider a monochromatic plane wave of frequency ω propa-

gating along the z-axis through a plasma. Assume that the
nontrivial components of the electric and the magnetic fields
are along the x-axis and the y-axis respectively. Calculate the
time averaged radiated power per unit area normal to the z-axis
when ω > ωp and ω < ωp .
3. A monochromatic plane wave in vacuum is incident normally

on the plane boundary of a semi-infinite uniform plasma. Find
the reflected and the transmitted waves, considering frequencies
above and below the plasma frequency. For what value of the
frequency is there a change of phase of π2 on reflection?
Chapter 12
Interaction of charged particles with

electromagnetic fields
12.1 Relativistic Lagrangian description
We have seen earlier that Maxwell’s equations are manifestly Lorentz

covariant. This is easily achieved by combining the electric and the
magnetic fields into a second rank anti-symmetric field strength ten-
sor of the form
Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ , µ, ν = 0, 1, 2, 3, (12.1)
where the four component vector potentials are defined as (see (6.151))
Aµ = (Φ, −A). (12.2)
It is easily seen from the definition in (12.1) that
1 ∂Ai
F0i = ∂0 Ai − ∂i A0 = − ∇i Φ = Ei ,
c ∂t
Fij = ∂i Aj − ∂j Ai = ∇i Aj − ∇j Ai
= −ǫijk (∇ × A)k = −ǫijk Bk , (12.3)
where we have used the fact that Ai = −(A)i , ∂0 = 1c ∂t

∂
and ∂i = ∇i .
Given the field strength tensor Fµν , we can define a Lorentz
invariant action for the free Maxwell theory of the form
Z Z
4 4 1 µλ νρ
S= d xL= d x − η η Fµν Fλρ . (12.4)
16π
In addition to being Lorentz invariant, this action is also invariant

under a gauge transformation of the form (see (6.166))
Aµ (x) → Aµ (x) + ∂µ α(x), (12.5)
397
398 12 Electromagnetic interactions
where α(x) is an arbitrary space-time dependent parameter of gauge

transformation. Under such a transformation,
Fµν → ∂µ (Aν + ∂ν α) − ∂ν (Aµ + ∂µ α) = ∂µ Aν − ∂ν Aµ

= Fµν . (12.6)
Namely, the field strengths are invariant under a gauge transforma-

tion, something that we already know. However, as a consequence
of this, it follows that the action in (12.4) is also invariant under the
gauge transformation (12.5).
We can derive the dynamical equations from the action in (12.4)
as the Euler-Lagrange equations. In this case, since the Lagrangian
density depends on the dynamical variable Aµ only through deriva-
tives, the Euler-Lagrange equation takes the form
∂L
∂µ = 0,
∂∂µ Aν
or, ∂µ F µν = 0. (12.7)
As we have seen earlier (see (6.157)), Eq. (12.7) gives only two of
Maxwell’s equations (in vacuum in the absence of sources), namely,
∇ · E = 0,
1 ∂E
∇×B= .
c ∂t
The other two equations are contained in the Bianchi identity satisfied
by the field strength tensor
∂µ Fνλ + ∂ν Fλµ + ∂λ Fµν = 0, (12.8)
which follows from the definition of the field strength tensors in (12.1).
We have just described Maxwell’s equations in the absence of
sources. Since Maxwell’s equations (with sources) are also Lorentz
covariant, we can try to introduce sources in a covariant manner as
well. The simplest case is, of course, to consider the interaction of a
charged particle with electromagnetic fields. Thus, we first need to
give a relativistic description of the motion of a free particle. This
is done in a simple manner as follows. First, let us consider a free
particle moving along a trajectory in the four dimensional space-time
manifold as shown in Fig. 12.1. Unlike the non-relativistic case, here
we cannot parameterize the trajectory with t which is not Lorentz
12.1 Relativistic Lagrangian description 399
invariant. Instead, we note that the invariant interval (length) in a

Minkowski manifold is defined as (recall that xµ = (ct, x))
ds2 = c2 dτ 2 = ηµν dxµ dxν = c2 dt2 − dx · dx. (12.9)
The parameter s and, therefore, the proper time τ are invariant
τ)
x µ(
b
Figure 12.1: Trajectory of a relativistic particle parameterized by τ .
under a Lorentz transformation. The trajectory can be labeled by

these parameters. We note from (12.9) that in the rest frame of the
particle where dx
dt = 0, the coordinate time can be identified with the
proper time. This will, of course, not be the case in other Lorentz
frames.
Parameterizing the trajectory of the particle as xµ (τ ), we note
that the free particle Newtonian equation can be generalized to
d2 xµ (τ )
m = 0, (12.10)
dτ 2
where m denotes the rest mass of the particle. This is manifestly
Lorentz covariant since m and τ are Lorentz invariant scalars and xµ
is a Lorentz vector. Introducing a relativistic four velocity associated
with the particle as
dxµ (τ )
uµ = = γ(c, v), (12.11)
dτ
where
dx 1 dt
v= , γ=q = , (12.12)
dt 1− v2 dτ
c2
we see that the four velocity transforms like a vector under a Lorentz
transformation and that

η µν uµ uν = γ 2 c2 − v2 = c2 , (12.13)
which follows from (12.9) (or the definitions in (12.11) and (12.12)).
Thus, the components of the four velocity are not all independent,
rather they are constrained by (12.13).
The free particle equation in (12.10) can now be written as
duµ
m = 0. (12.14)
dτ
We see that the relation (12.13) is consistent with the equations of
motion (12.14). Furthermore, the form of the equation in (12.14)
allows us to define a relativistic four momentum associated with the
particle as
v
pµ = muµ = γmc 1, = γmc (1, β) , (12.15)
c
so that the equation of motion, (12.14), can also be written as
dpµ
= 0. (12.16)
dτ
It follows now from (12.13) that
p2 = η µν pµ pν = m2 η µν uµ uν = m2 c2 . (12.17)
Recalling that pµ = ( Ec , p) (see (6.146)), we recognize this as the

Einstein’s relation for a relativistic particle. Furthermore, let us note
from Eq. (12.15) that we can now identify
Eβ
E = γmc2 , p= , (12.18)
c
which we have used earlier in connection with the method of virtual
photons (see (10.56)). In the non-relativistic limit, β ≪ 1 (or v ≪ c)
and we have
γ ≈ 1,
τ ≈ t,
uµ ≈ (c, v),
pµ ≈ m(c, v), (12.19)
so that Eq. (12.16) reduces in this limit to the familiar equation
dp
= 0.
dt
The constraints in (12.11) or (12.17) arise because the system

has a local gauge invariance, much like the Maxwell’s equations. To
see this, let us note that we can write the action for a massive rela-
tivistic particle as
Z Z
ds
S = mc ds = mc dλ
dλ
Z 1 Z
dxµ dxν 2
= mc dλ ηµν = dλ L, (12.20)
dλ dλ
where we are assuming that the trajectory of the particle is param-
eterized by λ and have used Eq. (12.9). It is easy to see that this
action is invariant under a local symmetry transformation. Namely,
under a reparameterization of the variable λ as
λ → ξ = ξ(λ), (12.21)
we have
dxµ dξ dxµ
→ . (12.22)
dλ dλ dξ
Consequently, under such a reparameterization, the action transforms
as
Z 1
dxµ dxν 2
S = mc dλ ηµν
dλ dλ
Z 2 !1
2
dξ dxµ dxν
→ mc dλ ηµν
dλ dξ dξ
Z 1
dξ dxµ dxν 2
= mc dλ ηµν
dλ dξ dξ
Z 1
dxµ dxν 2
= mc dξ ηµν = S. (12.23)
dξ dξ
This is an invariance under a local transformation much like the gauge
transformation in the case of Maxwell’s theory. Consequently, we can
choose a gauge and, in particular, we can chose the trajectory to be
parameterized by the proper time through the identification (which
is also known as a gauge choice)
λ = τ, (12.24)
which leads to
1
dxµ dxν 2 dτ
ηµν =c = c,
dλ dλ dλ λ=τ
λ=τ
µ ν 2
or, ηµν u u = c . (12.25)
With the gauge choice in (12.24), we note that the Lagrangian

µ
is a functional only of ẋµ = dx
dτ so that we can write the conjugate
momenta as
∂L mcẋµ
pµ = µ
= 1 = mẋµ = muµ , (12.26)
∂ ẋ (ẋν ẋν ) 2
where we have used (12.25). The Euler-Lagrange equation of motion
now follows to be
d ∂L dpµ
µ
= = 0, (12.27)
dτ ∂ ẋ dτ
which is the equation for the free particle, as we have seen in Eq.
(12.15). Here, we see that it can be derived from the principle of
minimum action in the Lagrangian framework (physically implying
that the path followed by a free particle between two points is the
shortest path). We also note from (12.26) that
p2 = η µν pµ pν = m2 c2 , (12.28)
which is the Einstein relation (12.18). The fact that not all the com-
ponents of the momenta are independent is a consequence of the
gauge invariance which the system possesses. (Let us note here par-
enthetically that the action in (12.20) is meaningful only for time-like
trajectories. For massless particles, an alternative form of the action
is more useful.)
So far, we have talked about a free relativistic particle. To
describe a relativistic particle subjected to a force, we can generalize
the dynamical equation (12.10) as
d 2 xµ
m = f µ,
dτ 2
dpµ
or, = f µ, (12.29)
dτ
where f µ represents the force four vector. From the fact that
dpµ duµ m d(uµ uµ )
uµ = muµ = = 0, (12.30)
dτ dτ 2 dτ
which follows from (12.9), we have
uµ f µ = 0. (12.31)
Namely, much like the four velocity and the four momentum, the
components of the relativistic force are not all independent. In fact,
explicitly we have from (12.31) that
uµ f µ = γ(cf 0 − v · f ) = 0,
v·f
or, f 0 = = β · f. (12.32)
c
We can determine the form of the relativistic force as follows.

First, we note that in the rest frame of the particle v = 0. In this
frame, we have
0
frest = 0, frest = F. (12.33)
If we know the force F in the rest frame of the particle, then we

can obtain its relativistic form by transforming to a general Lorentz
frame. The coordinate vectors in a frame moving with an instanta-
neous velocity v are related to those in the rest frame as

x′0 = γ x0 + β · x ,

′ β(β · x) β(β · x) 0
x = x− +γ + βx . (12.34)
β2 β2
The physical meaning of Eq. (12.34) is quite clear. Only the compo-
nent of the coordinates along the direction of the velocity of the frame
(and, of course, the time coordinate) transforms, while the compo-
nents perpendicular to the velocity of the frame do not. (Note that
the first parenthesis on the right hand side of the second relation in
(12.34) denotes the orthogonal component while the first term in the
second parenthesis describes the longitudinal component of the coor-
dinate with respect to β.) Using this, then, the general form of the
force in a frame with an instantaneous velocity v can be determined
from its form in the rest frame to be
f 0 = γβ · F,
β(β · F)
f = F + (γ − 1) . (12.35)
β2
It can be checked from the form of the components of the force in

(12.35) that
uµ f µ = u0 f 0 − u · f
= cγ 2 β · F − cγ(1 + (γ − 1)) β · F = 0. (12.36)
Let us observe that in evaluating the force in a general frame, quanti-
ties on the right hand side of (12.35) should be expressed in the new
frame where necessary.
Thus, we see that if we know the form of the force in the rest
frame of the particle, we can determine its general covariant form
through a Lorentz boost. With this, we can now describe the rela-
tivistic form of the electromagnetic force acting on a charged particle.
However, in the case of electromagnetic interactions, there is a sim-
pler way of deriving this. We note from our earlier studies that the
minimal electromagnetic interaction is linear in the (electromagnetic)
fields. As we have seen, the electromagnetic field strength is a sec-
ond rank anti-symmetric tensor. Thus, a natural candidate for a
relativistic force appears to be
f µ = α F µν uν , (12.37)
where α is a constant to be determined. Because of the anti-symmetry
of the field strength tensor, this form of the force automatically sat-
isfies the constraint (12.31), namely, uµ f µ = 0. Furthermore, the
constant α can be fixed by going to the rest frame of the particle,
where the force takes the form (in the rest frame γ = 1, ui = 0)
0
frest = α F 0i ui = 0,
i
frest = α F iµ uµ = α(F i0 u0 + F ij uj ) = α F i0 u0 ,
or, frest = αc E = F. (12.38)
Recalling that a charged particle at rest feels only an electrostatic
force of the form
F = qE,
we determine that
q
α= , (12.39)
c
where q represents the charge of the particle. As a result, the rela-
tivistic form of the (electromagnetic) force is determined to be
q
f µ = F µν uν , (12.40)
c
so that the dynamical equation takes the form

dpµ q
= f µ = F µν uν . (12.41)
dτ c
It is easy to see from (12.40) now that in a general frame, the com-
ponents of the force have the forms
q γq
f 0 = F 0i ui = v · E,
c c
q q
f i = F iµ uµ = F i0 u0 + F ij uj ,
c c

1
or, f = γq E + v × B , (12.42)
c
where we have used the definitions in (12.3). That this leads to the
usual Lorentz force can be seen by noting that
dp
= f,
dτ

dt dp 1
or, = γq E + v × B ,
dτ dt c

dp 1
or, =q E+ v×B . (12.43)
dt c
The form of the relativistic force can also be determined by the
conventional method through a Lorentz boost. For example, we know
that in the rest frame of the particle, a charged particle will only feel
the electrostatic force so that
0
frest = 0,
frest = F = qE′ , (12.44)
where we have designated the electric field in the rest frame by E′ .
It follows now from Eq. (12.35) that, in a frame moving with an
instantaneous velocity v, the components of the force will have the
forms
γq
f 0 = γβ · F = v · E′ ,
c
β(β · F)
f = F + (γ − 1)
β2

′ β(β · E′ ) β(β · E′ )
=q E − +γ . (12.45)
β2 β2
The fields, on the right hand side of (12.45), are the rest frame vari-
ables. They can be transformed to the new variables by noting that
the components of the electric fields parallel to the velocity of the
frame do not transform while the orthogonal components do so that
E′k = Ek ,

1
E′⊥ = γ E⊥ + v × B . (12.46)
c
Using these in (12.45), we obtain

γq
f0 = v · E,
c

β(β · E) v β(β · E)
f =q γ E− + ×B +γ
β2 c β2

1
= γq E + v × B , (12.47)
c
which is the same result as in Eq. (12.42).

We can now give a Lagrangian description of the interacting
theory as
Z
S = dλ L, (12.48)
where we have chosen to parameterize the trajectory with λ (later

identified with the proper time τ of the particle through a gauge
choice) and
1 q
L = mc (ηµν ẋµ ẋν ) 2 + Aµ (x(λ))ẋµ . (12.49)
c
Here, the first term is, of course, the Lagrangian for a free particle
that we have already studied. The second represents the minimal
interaction Lagrangian obtained in the standard manner,
Z
1
Sint = dx Aµ (x)j µ (x)
c
Z
q dxµ (λ)
= dx dλ Aµ (x) δ(x − x(λ))
c dλ
Z
q
= dλ Aµ (x(λ))ẋµ , (12.50)
c
where we have used the standard representation for the current den-
sity associated with a charged particle, namely,
Z
µ dxµ (λ)
j (x) = q dλ δ(x − x(λ)). (12.51)
dλ
The action in (12.48) is invariant under reparameterization and, if we
choose the gauge (12.24) (see also (12.26)), then we obtain
∂L q q
Πµ = µ
= mẋµ + Aµ = pµ + Aµ , (12.52)
∂ ẋ c c
which shows that the interaction is indeed that of minimal coupling.
Here we have identified
dxµ
pµ = mẋµ = m , (12.53)
dτ
which is also known as the kinematic momentum (or sometimes also
as the mechanical momentum) of the particle. The Euler-Lagrange
equation, in this case, leads to
d ∂L ∂L
− = 0,
dτ ∂ ẋµ ∂xµ
d q q
or, pµ + Aµ − (∂µ Aν )ẋν = 0. (12.54)
dτ c c
Noting that
dAµ
= (∂ ν Aµ )ẋν , (12.55)
dτ
the Euler-Lagrange equation in (12.54) takes the form
dpµ q q
= (∂ µ Aν − ∂ ν Aµ ) ẋν = F µν uν . (12.56)
dτ c c
This is indeed the dynamical equation that we have discussed in
(12.41).
We have, of course, neglected the dynamics of the electromag-
netic fields in this discussion. If we add the Maxwell term (12.4)
to the action, then, the complete set of coupled equations of motion
takes the form
dxµ
m = pµ ,
dτ
dpµ q
= F µν uν ,
dτ c
4π ν
∂µ F µν = j , (12.57)
c
where the current density has the form in (12.51) (with the identifi-
cation λ = τ ). We note here that sometimes the second equation in
(12.57) is also written as
dpµ q
m = F µν pν , (12.58)
dτ c
with the identification in (12.15).
12.2 Motion in a uniform electric field
Let us next work out the solutions for a few interacting systems. Let
us recall that the equations of motion (12.41) can be written as (recall
dt
that pµ = muµ = γm(c, v) and γ = dτ )
 
d  mc2 
q = qv · E,
dt 1− v2
c2
 

d  mv  1
q =q E+ v×B , (12.59)
dt 1− v
2 c
c2
where we have used the identifications in (12.42). Let us note some

general characteristics following from these equations. First, for a
static electric field, we note that we can write
E = −∇Φ,
so that the first equation in (12.59) takes the form

 
d  mc 2
q  = −q dx · ∇Φ = −q dΦ ,
dt 1− v
2 dt dt
c2
mc2
or, q + qΦ = constant. (12.60)
2
1 − vc2
This is known as the energy relation since in the non-relativistic limit,

it gives the energy conservation relation. (The time component of pµ
is related to the energy.) This equation also implies that when the
electric field vanishes, namely, E = 0 (or Φ = 0), the speed of the
particle v = |v| must be a constant. This would also be the case for
a static magnetic field which does not produce an electric field (only
a time dependent magnetic field can generate an electric field).
12.2 Motion in a uniform electric field 409
With these general observations, let us now examine the solu-

tions for the motion of a charged particle in the presence of a uniform
electric field. If we assume the electric field to be along the x-axis,
then we have
E = (E, 0, 0), B = 0,
where E is a constant and the fields can be generated by a scalar

potential of the form
Φ = −Ex, A = 0. (12.61)
In this case, the equations of motion in (12.59) can be trivially inte-

grated to give
mc2
q = qEx,
2
1 − vc2
mv
q = qEt, (12.62)
v2
1− c2
where we are assuming that the particle is at rest at t = 0 so that

the velocity is along the x-axis and, correspondingly, we have set the
constants of integrations to zero. The ratio of the pair of equations
in (12.62) determines
v ct
= . (12.63)
c x
Substituting this into the second equation in (12.62), we obtain
mc2 t
q x
= qEt,
c2 t2
1− x2
p mc2
or, x2 − c2 t2 = , (12.64)
qE
which can also be rewritten as
m2 c4
x2 − c2 t2 = ,
q2E 2
s
m2 c2
or, x = c + t2 . (12.65)
q2 E 2
The first form of the relation in (12.65) clearly shows that the motion
is hyperbolic in the presence of a constant acceleration. This has to
be contrasted with the non-relativistic case where
1
x = x0 + at2 ,
2
which describes a parabolic motion. The instantaneous velocity of

the particle can be obtained from (12.65) and leads to
dx ct
v= =q 22 . (12.66)
dt m c
+ t 2
q2 E 2
We see from this that, as t → ±∞, the magnitude of the velocity

approaches the speed of light, v → c. However, as t → 0, the particle
comes to rest (v → 0). Therefore, in the intermediate time interval
(−∞ < t ≤ 0), the particle seems to feel a deceleration. We also rec-
ognize that the motion will become non-relativistic when the electric
field is weak. In particular, if
qE
≪ 1, (12.67)
mc
mc
then, (in a time interval t ≪ qE ) we have from Eqs. (12.65) and
(12.66),
qE
v≈ t = at,
m
mc2 1 qE 2
x≈ + t
qE 2 m
1 2
= x0 + at , (12.68)
2
which is what we will expect. Here, we have introduced a to represent

the instantaneous acceleration.
In the derivation above, we assumed that the particle is initially
at rest. However, in the general case when the particle initially has a
momentum p(0) , we can solve the system of equations in the following
manner. First, let us note that the initial momentum and the electric
field define a plane where the particle motion is confined to. Without
loss of generality, we can choose this to be the x − y plane (recall that
we are assuming the electric field to be along the x-axis). Thus, our
12.2 Motion in a uniform electric field 411
equations in this case lead to

mc2
q = qEx,
2
1 − vc2
mv
q = p(0) + qEt . (12.69)
2
1 − vc2
By assumption, of course, the velocity has only x and y components.

From the second equation in (12.69), we obtain
2 2
2 2 v (0) v2
m c = p + qEt 1− 2 ,
c2 c
2
v2 2 2 (0)
2
(0)
or, m c + p + qEt = p + qEt ,
c2
2
v2 p(0) + qEt
or, = 2 ,
c2 m2 c2 + p(0) + qEt
v2 m2 c2
or, 1 − = 2 . (12.70)
c2 m2 c2 + p(0) + qEt
Substituting this into the first equation in (12.69), we obtain

m2 c4 2
2 q2 E 2 c2 2 2 (0)
x = 2 = 2 2 m c + p + qEt ,
1 − vc2 q E
q 2
c
or, x = m2 c2 + p(0) + qEt . (12.71)
qE
This, of course, reduces to Eq. (12.65) when the particle is initially
at rest, namely, p(0) = 0. We can now solve for the y coordinate of
the particle from (12.69), which yields
(0)
r
dy py v2
= 1− 2
dt m c
(0)
cpy
=p ,
m2 c2 + (p(0) )2 + 2qp(0) · Et + q 2 E 2 t2
!
(0)
cpy qE 2 t + p(0) · E
or, y = sinh−1 p .
qE E 2 (m2 c2 + (p(0) )2 ) − (p(0) · E)2
(12.72)
Here, we have used the standard formula from the table of integrals
(see, for example, Gradshteyn and Ryzhik, 2.261). We note from Eq.
(12.70) that as t → ±∞, the speed of the particle approaches the
speed of light as before. However, in the present case for the particle
moving in a plane, the motion cannot be characterized as hyperbolic.
12.3 Motion in a uniform magnetic field
Let us next consider the case of a charged particle moving in a uniform

magnetic field which we can take to be along the z-axis, namely, we
have
E = 0, B = Bẑ, (12.73)
where B is a constant. In this case, the dynamical equations, (12.59),

take the forms
 
d  mc 2
q  = 0,
dt 1− v2
c2
 
d  mv  qB
q = v × ẑ. (12.74)
dt 1− v
2 c
c2
It is clear from the first equation in (12.74) that v 2 is a constant and

we can write
1 E
γ=q = , (12.75)
1− v2 mc2
c2
where E represents the energy of the system (recall that the first
equation gives the energy relation and we are denoting energy by E
to avoid any confusion with the electric field). Furthermore, from
the second equation we see that vz is also a constant. It follows,
therefore, that
p
v⊥ = v 2 − vz2 , (12.76)
is a constant as well. Since the z component of the velocity is a

constant, we note that the solution for the z coordinate of the particle
is straightforward
z = z0 + vz t. (12.77)
12.3 Motion in a uniform magnetic field 413
The second equation can now be written out explicitly for the
x, y components of the velocity to give
dvx qB
= vy ,
dt γmc
dvy qB
=− vx . (12.78)
dt γmc
Recalling that vx2 + vy2 = v⊥2 is a constant, we can write the solutions
of Eq. (12.78) as

qB
vx = −v⊥ sin t ,
γmc

qB
vy = −v⊥ cos t . (12.79)
γmc
These can be integrated so that, together with (12.77), we have

γmcv⊥ qB
x= cos t ,
qB γmc

γmcv⊥ qB
y=− sin t ,
qB γmc
z = z0 + vz t. (12.80)
Here, we have chosen the constants of integration (as well as the

phases) such that at t = 0, the particle has coordinates x = γmcv
qB , y =
⊥
0, z = z0 and vx = 0, vy = −v⊥ , vz = constant. We note that the mo-

tion in the x − y plane is harmonic with a frequency
qB qBc
ω= = , (12.81)
γmc E
where we have used (12.75). In general, we see that the frequency

of motion depends on the energy of the particle. However, in the
non-relativistic limit, where γ ≈ 1, the expression for the frequency
becomes
qB
ω≈ , (12.82)
mc
which is independent of energy. This frequency is commonly known
as the cyclotron frequency and is important in the study of cyclotrons.
For a relativistic particle moving in a uniform magnetic field, however,
the relation is quite different and is of significance in the study of

synchrotrons. Let us also note that, for vz = 0, the motion is circular
in the x − y plane with the radius of the circle given by
v⊥ γmcv⊥ Ev⊥
R= = = , (12.83)
ω qB qBc
so that we can write the planar coordinates in (12.80) also as

qB
x = R cos ωt = R cos t ,
γmc

qB
y = −R sin ωt = −R sin t . (12.84)
γmc
12.4 Motion in uniform crossed electric and magnetic fields
Let us next consider the case of a charged particle moving in uniform

electric and magnetic fields that are orthogonal to each other. For
example, we may have
E = E x̂, B = Bẑ, (12.85)
so that
E · B = 0. (12.86)
We can, of course, solve the equations (12.59) as before. However, a

direct solution is much more involved. There is an alternative and
simpler method for analyzing the motion in this case that we describe
below.
Let us recall that the electromagnetic field strength tensor Fµν
is a second rank anti-symmetric tensor. We can construct from this
two independent quadratic Lorentz invariant scalars, namely
1
Fµν F µν = η µλ η νρ Fµν Fλρ , Fµν F̃ µν = ǫµνλρ Fµν Fλρ , (12.87)
2
where ǫµνλρ is the four dimensional Levi-Civita tensor. From the

definitions of the field strength tensors in (12.3), it follows that
Fµν F µν = 2F0i F 0i + Fij F ij = 2(B2 − E2 ). (12.88)

12.4 Motion in crossed fields 415
Furthermore, since (F̃ µν = 12 ǫµνλρ Fλρ is also known as the dual field
strength tensor, ǫ0ijk = ǫijk )
1 0ijk
F̃ 0i = ǫ Fjk = −Bi ,
2
F̃ ij = ǫij0k F0k = ǫijk Ek , (12.89)
it follows that
Fµν F̃ µν = 2F0i F̃ 0i + Fij F̃ ij = −4E · B. (12.90)
Thus, we see that E · B as well as (B2 − E2 ) are invariant under

Lorentz transformations so that they have the same value in any
Lorentz frame. As a result, when E · B = 0 (namely, when E and B
fields are orthogonal to each other), depending on whether
(B2 − E2 ) > 0, ⇒ |B| > |E|,

or, (B2 − E2 ) < 0, ⇒ |B| < |E|, (12.91)
we can go to a Lorentz frame where either E = 0 or B = 0 respec-

tively. (Note that if E · B 6= 0, we cannot go to a frame where either
B = 0 or E = 0.) Once we are in the frame where either the electric
or the magnetic field is zero, then the solution in that frame is exactly
the same as the ones we have found in the earlier two sections. To
come back to the solutions in the original Lorentz frame, we simply
have to make the appropriate inverse Lorentz transformation. Let us
recall that in going to a Lorentz frame moving with a velocity u, the
electric and the magnetic fields transform as
E′k = Ek ,
u
E′⊥ = γ(u)(E⊥ + × B⊥ ),
c
B′k = Bk ,
u
B′⊥ = γ(u)(B⊥ − × E⊥ ), (12.92)
c
where the “parallel” and the “perpendicular” decompositions are
with respect to the velocity u.
Let us first consider the case when |E| = E > B = |B| and
consider a frame moving with a velocity u such that
u E×B
= . (12.93)
c E2
Such a transformation (frame) is allowed since
u2 E 2 B 2 − (E · B)2 B2
= = < 1, (12.94)
c2 E4 E2
where we have used the fact that the electric and the magnetic fields
are orthogonal. In this case, therefore, we have
1 1
γ(u) = q =q . (12.95)
u2 B2
1− c2
1− E2
We note that such a velocity is orthogonal to both the electric and

the magnetic fields (u · E = 0 = u · B) and, consequently, in this case,
we have
Ek = 0 = Bk , ⇒ E′k = 0 = B′k
Furthermore, under such a transformation, we have (since there is no

component of the fields parallel to the velocity)

′ u (E × B) × B
E = γ(u) E + × B = γ(u) E +
c E2

(E · B)B − E(B2 )
= γ(u) E +
E2
r
B2
= 1 − 2 E = γ −1 (u) E,
E

′ u (E × B) × E
B = γ(u) B − × E = γ(u) B −
c E2

(E2 )B − (E · B)E
= γ(u) B −
E2
= γ(u)(B − B) = 0. (12.96)
Here, we have used the orthogonality of the electric and the magnetic
fields in the intermediate steps.
This shows that for orthogonal electric and magnetic fields, when
E > B, we can find a Lorentz frame where the magnetic field iden-
tically vanishes. Furthermore, the electric field, in this new frame, is
along the same direction as the original field, only scaled by a Lorentz
factor. The solution to this problem is, as before, unbounded motion
and is not very interesting.
12.4 Motion in crossed fields 417
Let us next consider the case when E < B and choose a Lorentz
transformation with
u E×B
= . (12.97)
c B2
In this case, we have
u2 E 2 B 2 − (E · B)2 E2
= = < 1, (12.98)
c2 B4 B2
so that it is an allowed transformation. For the present case,
1 1
γ(u) = q =q . (12.99)
u2 E2
1− c2 1− B2
We note here parenthetically that when E = B, the frame moves

with the speed of light and, therefore, is not meaningful.)
Once again the electric and the magnetic fields are orthogonal
to the velocity u and, consequently,
Ek = 0 = Bk .
Furthermore, we have

u (E × B) × B
E′ = γ(u) E + × B = γ(u) E +
c B2

(E · B)B − (B2 )E
= γ(u) E +
B2
= γ(u)(E − E) = 0,

′ u (E × B) × E
B = γ(u) B − × E = γ(u) B −
c B2

(E2 )B − (E · B)E
= γ(u) B −
B2
r
E2
= 1 − 2 B = γ −1 (u) B. (12.100)
B
Thus, in this case, we see that the electric field vanishes in the
new frame and the magnetic field is along the same direction as the
original field, but scaled by a field dependent Lorentz factor. For
simplicity of comparison with earlier results, let us choose E = E x̂

and B = B ẑ. In such a case, we see that
u E×B E
= 2
= − ŷ. (12.101)
c B B
Namely, the frame is moving along the y axis. If we assume that
the particle has no initial velocity along the z-axis, then following
the analysis of the last section, we can obtain the solutions which
would suggest that the particle will be moving in a circle in the plane
perpendicular to the magnetic field with a radius
γ(v ′ )mcv⊥
′ γ(v ′ )mcv⊥
′
R′ = = q . (12.102)
qB ′ qB 1 − B 2 E2
In fact, with initial conditions as before, we can write the solutions

to have the form (see (12.80))

qB ′
x′ = R′ cos t ′
,
γ(v ′ )mc
p
qB ′
y ′ = −R′ sin t ′
= − R′2 − x′2 . (12.103)
γ(v ′ )mc
The solution can now be transformed back into the original frame.
Since the frame is moving along the y-axis, the Lorentz transforma-
tions take the forms

′ β(u)
t = γ(u) t − y ,
c
x′ = x,
y ′ = γ(u) (y − β(u)ct) ,
z ′ = z, (12.104)
and this leads to

 q 
qB 1− E2
B2 E
x = x′ = R cos  qER
t− y ,
γ(v)mc(1 − ) Bc
γ(v)mc2
p
y ′ = γ(u)(y − β(u)ct) = − R2 − x2 ,
r
Ec E2 p
or, y= t − 1 − 2 R 2 − x2 , (12.105)
B B
12.5 Motion in a slowly varying magnetic field 419
where, as before,
γ(v)mcv⊥
R= .
qB
Here we have also used the fact that, under an inverse Lorentz trans-
formation,
R′ → R, v ′ → v, ′
v⊥ → v⊥ ,
and so on. In this case, we see that the motion corresponds to an el-
lipsoidal motion superimposed with a constant “drift” velocity along
the y axis. This is known as a trochoidal motion.
We note from the equations of motion (12.59), that when
v E×B u
= = , (12.106)
c B2 c
the Lorentz force identically vanishes (see also the first relation in
(12.100)), namely,

(E × B) × B
q E+ = q (E − E) = 0. (12.107)
B2
In this case, the particle would move along the initial trajectory com-
pletely undeflected by the presence of the fields, independent of its
mass. (Namely, f = 0 also implies that f 0 = v·f c = 0, see (12.32).)
This is a very important feature which is utilized in creating a veloc-
ity filter. Namely, if a number of particles are incident on a region
with crossed electric and magnetic fields (with E < B), then only
those particles that have the initial velocity coinciding with (12.106)
would travel undeflected. Correspondingly, one can choose different
electric and magnetic fields to select the desired particles with a given
velocity.
12.5 Motion in a slowly varying magnetic field
The fact that the magnetic force v × B introduces a curvature to

the trajectory of a charged particle is exploited profitably in many
physical situations. For example, this is quite useful in the study of
confinement of plasma as well as in astrophysical studies of plasma.
Let us recall that a charged particle in a uniform magnetic field,
has a circular path with a “drift” velocity along the direction of B
(normally taken to be along the z-axis). (Sometimes, one separates
out the circular motion to talk of a uniform motion of the “guiding
center”.) However, in many of these physical systems of interest the

magnetic field is not uniform. Although the magnetic field is static,
it varies with space, particularly longitudinally. A general problem
of this kind is, of course, very hard to solve. However, when the field
varies slowly, we can determine the behavior of such systems quite
well. Namely, if the field varies slowly within the radius of curvature
of the trajectory, we can think of the particle to be still executing
circular motion with a radius
γmcv⊥
R= , (12.108)
qB
where both v⊥ , B will now be functions of position (although slowly
varying).
If we assume that the magnetic field is everywhere parallel, then
without loss of generality, we can choose this to be along the z-axis,
B = B ẑ. However, in this case, it follows from the second equation
of Maxwell
∇ · B = 0,
that B = B(x, y) and that, since in general, the curl of the magnetic
field will not vanish, such a configuration must have an associated
steady current and we see that, in this case, the magnetic field can-
not depend on the (longitudinal) z coordinate. As we noted earlier,
in some physical situations, we do need a dependence of the magnetic
field on the longitudinal coordinate which can arise only if the mag-
netic field is not parallel everywhere. For confinement of plasma to a
small region in space, for example, we would expect the plasma not
to extend beyond a certain vertical and horizontal dimension. Under
the action of a magnetic field, as we have seen, the particle moves
in an orbit whose radius is determined by the magnetic field. Thus,
we see that with a suitable choice of a magnetic field, the plasma
can be easily confined to a given vertical dimension. Let us note
also that if we have a magnetic field that is converging along the z-
axis in some region, then the magnetic force acting on the charged
particles will be so as to force it into the interior of the region (see
Fig. 12.2). Therefore, there will be a component of the force along
the z-axis which would decelerate the “drift” velocity. As a result,
at some point along the horizontal (z) axis, the “drift” velocity will
vanish (the “guiding center” will come to rest) and then, the direc-
tion of the drift will change. This would lead to a containment of
the charged particle along the z-axis as well, as shown in Fig. 12.3.
This process is known as “mirroring” and is used to trap plasma by a
v×B
v×B
Figure 12.2: The Lorentz force directing the particle into the interior
of the region.
magnetic field by forming a “bottle” with one “mirror” point at each

end (namely, converging magnetic fields at both ends).
Figure 12.3: Confinement due to a magnetic field (magnetic bottle).
Keeping this qualitative picture in mind, let us consider a mag-

netic field with cylindrical symmetry, B = (Bρ , 0, Bz ). We assume
that the components of the magnetic field do not depend on the an-
gular coordinate. In this case, the second equation of Maxwell leads
to
1 ∂(ρBρ ) ∂Bz
∇·B= + = 0,
ρ ∂ρ ∂z
∂(ρBρ ) ∂Bz
or, = −ρ . (12.109)
∂ρ ∂z
To leading order, if we assume that ∂B∂z is independent of ρ, this

z
equation can be integrated and leads to

1 ∂Bz
ρBρ = − ρ2 ,
2 ∂z
1 ∂Bz
or, Bρ = − ρ . (12.110)
2 ∂z
The constant of integration is easily seen to vanish (with the assump-
tion that Bρ = 0 at ρ = 0). It is clear now from (12.110), that for
R ∂B
slowly varying fields (| B ∂z | ≪ 1), Bρ is much smaller than Bz within
a radius of the size of R. Consequently, we can approximate
B = |B| ≈ Bz , (12.111)
to write
1 ∂B
Bρ ≈ − ρ . (12.112)
2 ∂z
From the time component of the equations of motion, (12.59),
we note that in the absence of an electric field, v is constant. How-
ever, unlike the earlier case of motion in a uniform magnetic field,
here a non-vanishing Bρ leads to a magnetic force along the z-axis.
As a result, vz and, therefore, v⊥ will no longer be constant. In
fact, looking at the equation for the z component of the velocity (see
(12.59)) for a particle moving in an orbit of radius R, we have (γ is
a constant since v is)
d q
(γmvz ) = v⊥ Bρ ,
dt c
dvz 1 qRv⊥ ∂B 1 v 2 ∂B
or, =− =− ⊥
dt 2 γmc ∂z 2 B ∂z
2 (0)
1 v⊥ ∂B
≈− . (12.113)
2 B(0) ∂z
Here, we have used the fact that the velocity of the particle, in the x−
y plane, is given by v⊥ = −v⊥ φ̂ (see (12.79)), v⊥ = qRB
γmc (see (12.83))
as well as the relation in (12.112) for a particle moving in an orbit of
radius R. Furthermore, since the fields are slowly varying, we have
approximated the coefficient multiplying ∂B ∂z (which is already small)
by its value at z = 0 which can be thought of as the leading order
approximation. It now follows from (12.113) that (the assumption
here is that at t = 0, the particle is at z = 0 and that at t its
coordinate is z)
1 dvz2 dvz dz dvz dvz 1 v 2 (0) ∂B

= vz = = =− ⊥ ,
2 dz dz dt dz dt 2 B(0) ∂z
1 2 v 2 (0)
or, vz − vz2 (0) = − ⊥ (B(z) − B(0)),
2 2B(0)
2 (0)B(z)
v⊥
or, vz2 = v 2 − , (12.114)
B(0)
where we have used
vz2 (0) + v⊥
2
(0) = v 2 (0) = v 2 , (12.115)
since v is a constant (because there is no electric field). Furthermore,

using (12.115), we can rewrite the relation in (12.114) as
2 (0)B(z)
v⊥
v 2 − vz2 = v⊥
2
= ,
B(0)
2
v⊥ v 2 (0)
or, = ⊥ . (12.116)
B(z) B(0)
Relation (12.116) gives an adiabatic invariant associated with

this problem. Namely, it says that the ratio of the square of the
perpendicular velocity to the magnetic field
2
v⊥ qRv⊥
= , (12.117)
B(z) γmc
is an adiabatic constant of motion. Physically, what this means is

that an adiabatic invariant is not a true constant of motion because
of the variation in the fields. However, when the fields change slowly
(adiabatically), the change in an adiabatic invariant is even slower
than that of the field. We note from (12.117) that as B(z) increases,
vq ⊥ must also increase for this ratio to be constant. However, since v =
v⊥2 + v 2 is a constant, this also means that when B(z) increases, v
z z
must decrease, something that we had seen qualitatively earlier. The
magnetic flux associated with the particle motion (through a circle
of radius R) can now be seen to be an adiabatic invariant for the
problem as well. Namely,
γ 2 m2 c2 v⊥
2
πR2 B = π B
q2B 2
2
πγ 2 m2 c2 v⊥
=
q2 B(z)
2 (0)
πγ 2 m2 c2 v⊥
= , (12.118)
q2 B(0)
where we have used (12.108) and Eq. (12.118) follows by virtue of

(12.116). Similarly, we can define a magnetic moment associated with
the particle, which will also be an adiabatic invariant for the motion,
namely,
qω qωR2
µ = πR2 =
2π 2
2 2 (0)
γmc v⊥ γmc v⊥
= = . (12.119)
2 B(z) 2 B(0)
All the adiabatic invariants quantify the same qualitative be-

havior that we had discussed earlier. For example, from (12.117) we
see that as B(z) increases, v⊥ increases and, therefore, R decreases.
In this case, it follows from (12.114) that vz decreases. This shows
that by having a strong convergent magnetic field, a plasma can be
confined to a narrow region in space both vertically and horizontally.
The points z = z0 where vz vanishes are known as “mirror” points.
The plasma is turned back at this point and, consequently, one can
construct a magnetic bottle with two “mirror” points at the two ends.
The standard mechanism for confining plasmas is, therefore, to have
a region with stronger magnetic fields at the two ends and a weaker
field in the middle. In this case, the plasma is trapped oscillating
back and forth between the two “mirror” points. A similar natural
phenomenon is also observed in earth’s magnetic field. Because the
magnetic fields are strong at the poles and weaker at the center, pos-
itive and negatively charged ions in the atmosphere are trapped and
oscillate back and forth between the “Van Allen” belts.
12.6 Particles with spin and the anomalous magnetic moment
We have thus far considered the motion of particles carrying only

a charge. However, a particle may also have an intrinsic structure
12.6 Anomalous magnetic moment 425
leading to an intrinsic magnetic dipole moment µ, or an intrinsic

electric dipole moment d. In the presence of electric and magnetic
fields, these intrinsic moments lead to interaction energies of the forms
Emagnetic = µ · B, Eelectric = d · E. (12.120)
In a quantum theory, these moments are described in terms of the

spin of the particle as
g q f q
µ= S, d= S, (12.121)
2 mc 2 mc
where g = 2 and f = 0 for point charged particles as predicted by
Dirac’s theory and any deviation of g from this value (or a non-zero
value for (g − 2)) is known as the anomalous magnetic moment of
the particle. Quantum mechanically, spin is, of course, an opera-
tor. However, by Ehrenfest’s theorem, the expectation value of this
operator can be taken to correspond to the classical spin degrees of
freedom of a relativistic particle.
Just as under a Lorentz transformation electric and magnetic
fields get mixed, so do the electric and the magnetic dipole moments
under such a transformation. In fact, the electric and the magnetic
dipole moments (six in number like the components of the electro-
magnetic field strength tensor) can be combined into a second rank
anti-symmetric tensor Mµν = −Mνµ with
M0i = −di , Mij = −ǫijk µk , (12.122)
so that the additional interaction energy of the particle (with intrinsic

moments) in the presence of electric and magnetic fields takes the
relativistic form
1 µν 1
Eintrinsic = M Fµν = M 0i F0i + M ij Fij
2 2
1
= −M0i F0i + Mij Fij
2
1
= −(−di )Ei + (−ǫijk µk )(−ǫijm Bm )
2
= d · E + µ · B. (12.123)
The important thing to note from (12.123) is that this interaction

energy is invariant under a Lorentz transformation, much like the
rest mass of a particle. (Recall that energy normally transforms like
the time component of a four vector.) Therefore, we can obtain a
Lagrangian description of such an interacting system from (12.49) by

replacing

1 µν
mc → mc 1 + M Fµν = mc ∆, (12.124)
2mc2

1
∆= M µν Fµν . (12.125)
2mc2
Namely, we can consider a Lagrangian of the form
1 q
L = mc ∆(ηµν ẋµ ẋν ) 2 + Aµ (x)ẋµ , (12.126)
c
with such a nonminimal coupling. This Lagrangian leads to the
canonical momentum (in the gauge (12.24))
∂L q q
Πµ = = m∆ẋµ + Aµ = ∆pµ + Aµ . (12.127)
∂ ẋµ c c
Let us first consider the case when Mµν is a constant tensor.
In this case, the Euler-Lagrange equations following from the La-
grangian, in the gauge (12.24), lead to
dxµ
m = pµ = muµ ,
dτ
d ∂L ∂L
− = 0,
dτ ∂ ẋµ ∂xµ
dpµ 1 h q µν i
or, = F uν + m(∂ν ∆) c2 η µν − uµ uν
dτ ∆ c

1 q µν 1 λρ 2 µν µ ν

= F uν + 2 M Fλρ,ν c η − u u .
∆ c 2c
(12.128)
Here “comma” denotes a derivative. We note that the first term,

inside the square bracket, represents the usual minimal interaction
of a charged particle (although the coefficient is now modified by the
∆ term). The second term which represents a new force involving
derivatives of the field strength tensor, is manifestly transverse to uµ
as any relativistic force should be (see Eq. (12.31)). Second, we note
that for
1
M µν Fµν ≪ 1, (12.129)
2mc2
which would normally be the case, we can approximate
∆ ≈ 1, (12.130)
in which case, the dynamical equation for the momentum takes the
form

dpµ q µν 1 λρ 2 µν µ ν

≈ F uν + 2 M Fλρ,ν c η − u u . (12.131)
dτ c 2c
The set of equations in (12.128), however, is incomplete in the

sense that there is no dynamics for the spin variable. We do know
that a spin angular momentum precesses in the presence of a magnetic
field. However, the reason that we have no dynamics for the spin
variable is because our derivation of the equations has been from the
Lagrangian and in order to have the dynamical equations for the spin,
we should also have a dynamical Lagrangian for the new variables
representing the spin degrees of freedom. This can, of course, be
done. However, it is much simpler to obtain the equations for the
spin in an alternative manner. Let us note that since both the electric
and the magnetic dipole moments are proportional to spin, the tensor
Mµν describing spin must be highly constrained. Let us define a spin
four vector from the dual of Mµν as
1 µνλρ
sµ = ǫ pν Mλρ . (12.132)
2q
This is known as the Pauli-Lubanski spin variable. In the rest frame

of the particle pµ = (mc, 0, 0, 0) and then, it is easy to see that the
space component of sµ is proportional to the magnetic dipole moment
and, therefore, to spin. It is also clear from the definition in (12.132)
that
1
uµ s µ = pµ sµ = 0. (12.133)
m
Namely, much like the relativistic force, not all the components of sµ
are independent. Rather, we have
v·s
s0 = . (12.134)
c
In the rest frame of the particle, the spin four vector takes the form
sµrest = (0, S). (12.135)

For completeness, let us also note that we can invert the relation
(12.132) to write
q
Mµν = ǫµνλρ pλ sρ . (12.136)
2
It follows from this that
pµ Mµν = muµ Mµν = 0 = sµ Mµν . (12.137)
Let us note that in the rest frame of the particle, we can identify
the space components of the spin variable as the spin of the particle
(up to multiplicative factors). Furthermore, in the rest frame of the
particle, we know the equation for the spin in the presence of external
electromagnetic fields to correspond to

dS q g f
= S×B+ S×E . (12.138)
dt mc 2 2
From this, we can transform to any other Lorentz frame in a simple
manner. First, let us note that we can write
(S × B)i = −Fij Sj = Fij Sj = Fiν srest
ν ,
(S × E)i = ǫijk Sj F0k = F̃iν srest

ν , (12.139)
where we have defined the dual of the field strength tensor as
1
F̃ µν = ǫµνλρ Fλρ . (12.140)
2
With this, it is easy to see that the generalization of the equation for
the spin to any frame can be given by

dsµ q g ρ λµ uλ uµ f ρ λµ uλ uµ
= s Fλρ η − 2 + s F̃λρ η − 2
dτ mc 2 c 2 c
1 duλ µ
− s λ u . (12.141)
c2 dτ
It is easy to check that this equation reduces to (12.138) in the rest
frame of the particle. The last term in (12.141) vanishes in the rest
frame of the particle. However, it is essential in a relativistic gener-
alization in order to satisfy the constraint in (12.133). For example,
let us note from (12.141) that the terms in the square bracket are
orthogonal to uµ . Consequently, it follows that
dsµ duλ
uµ = −sλ ,
dτ dτ
d(uµ sµ )
or, = 0, (12.142)
dτ
which is consistent with (12.133) and holds only because of the pres-
ence of the last term in (12.141).
Equation (12.141) is still not in a simple form. It can be further
simplified using the dynamical equations for the particle in (12.128).
In fact, for a uniform field (and for ∆ ≈ 1), we can use (12.128) to
write the equation for the spin in any Lorentz frame as

dsµ q g µν 1 g
= F sν + 2 − 1 sλ F λρ uρ uµ
dτ mc 2 c 2

f µν 1 λρ µ
+ F̃ sν + 2 sλ F̃ uρ u . (12.143)
2 c
There are several things to observe from this equation. We note
that for a particle without an electric dipole moment (f = 0) moving
in a uniform magnetic field (F0i = 0), the equation for the spin takes
the form

dsµ q µν (g − 2) λρ µ
= gF sν + s λ F uρ u . (12.144)
dτ 2mc c2
Writing out the equations explicitly, we obtain,
ds gq γ 2 q(g − 2)
= s×B+ (v · (s × B)) v,
dτ 2mc 2mc3
ds0 1 d(v · s) γq(g − 2)
= = v · (s × B), (12.145)
dτ c dτ 2mc2
where we have made the identification in (12.134). The time rate
of variation of the longitudinal component of the spin can now be
calculated using (12.128) and the second relation in (12.145).
d(v̂ · s) 1 d(v · s) (v · s) dv
= − 3
v·
dτ v dτ v dτ
1 d(v · s)
=
v dτ
γq(g − 2)
= v̂ · (s × B),
2mc2
d(v̂ · s) q(g − 2)
or, = v̂ · (s × B). (12.146)
dt 2mc2
Here the second term on the right hand side in the first line van-
ishes because v · dv
dτ = 0 which follows from the second equation in
(12.128) for a uniform (constant) magnetic field. Equations (12.145)
and (12.146) show that when g = 2, the longitudinal component of

the spin (also known as the helicity) does not change with time, while
the spin precesses with the same frequency as the cyclotron frequency
qB
ω= .
γmc
However, when the particle has an anomalous magnetic moment and
(g − 2) 6= 0, then the longitudinal component of the spin changes
with time and, therefore, along the trajectory of the particle. Conse-
quently, the anomalous magnetic moment associated with a charged
particle can be experimentally determined and the current measure-
ments have it vanishing for a point particle up to an accuracy of about
10−10 . Furthermore, the experimental measurements are in excellent
agreement with the predictions of quantum electrodynamics.
Chapter 13
Scattering and diffraction of

electromagnetic waves
In classical mechanics as well as in quantum mechanics, we are famil-

iar with the phenomenon of scattering where a particle or a number
of particles incident on some center of force suffer deflection in their
trajectories because of the action of the force. Similarly, electromag-
netic waves can also undergo scattering. For example, we can imagine
a plane wave incident on a perfectly conducting sphere. The incident
fields would induce a surface current on the sphere which, in turn,
would generate electromagnetic fields and would lead to total fields
(incident plus scattered) which will be different from those associated
with the incident wave. So, conceptually the scattering set up in elec-
tromagnetism is quite similar to what we are used to. The meaningful
concept in such an experiment is again the cross section. However, in
the present context, the differential cross section is defined as (classi-
cally we have only electromagnetic waves and no particles) the ratio of
the average energy scattered per unit time per unit solid angle along
the direction (θ, φ) to the average incident energy per unit time per
unit area.
Average scattered energy/second/solid angle

σ(θ, φ) = . (13.1)
Average incident energy/second/unit area
We note that in a scattering experiment involving waves, there are

three length scales involved, namely, the size of the scattering source
d, the wave length λ of the incident wave and the distance R from
the scattering source where observations are made. The observations
are, of course, made far away from the scattering source so that R ≫
λ, d. When λ ≫ d, the phenomenon is commonly characterized as
scattering while for λ ≪ d, it is called diffraction. As we will see, the
techniques involved in the study of the two phenomena are different.
431
432 13 Scattering and diffraction
13.1 Scattering from a perfectly conducting sphere
It is clear that the problem of scattering corresponds to solving a

boundary value problem, much like what we have done in electrostat-
ics or magnetostatics. However, in the time dependent case, there are
very few problems that can be solved exactly. Scattering from a per-
fectly conducting sphere is one such example. In general, therefore,
we have to develop some approximate methods for dealing with such
problems. In this section, however, let us see how the boundary value
problem can be solved exactly in this case.
Let us assume that the scattering sphere has a radius d and that
a monochromatic plane wave of frequency ω is incident on it along
the z-axis. We assume that the center of the sphere coincides with
the origin of our coordinate system. First, let us recall some facts.
We know from the study of scattering (in quantum mechanics) that
a scalar plane wave, which is a solution of the Helmholtz equation,
can be expanded in terms of spherical harmonics as
eik·x = eikz = eikr cos θ
∞
X
= iℓ (2ℓ + 1) jℓ (kr)Pℓ (cos θ)
ℓ=0
∞
X p
= iℓ 4π(2ℓ + 1) jℓ (kr)Yℓ,0 (θ), (13.2)
ℓ=0
where jℓ (kr) denote spherical Bessel functions and we have used the
definition
r
(2ℓ + 1)
Yℓ,0 (θ) = Pℓ (cos θ).
4π
This shows that a scalar plane wave contains only waves with angular
momentum projection m = 0.
Let us next recall that a circularly polarized harmonic electro-
magnetic wave traveling along the z-axis in vacuum will have electric
and magnetic fields of the forms (factoring out the time dependence,
see also (11.72))
1
ER,L (x) = √ (x̂ ∓ iŷ) eikz ,
2
BR,L (x) = ẑ × ER,L (x) = ±i ER,L (x), (13.3)
where the first sign corresponds to a right circularly polarized wave
and the second to a left circularly polarized wave. (For simplicity, we
13.1 Scattering from a perfectly conducting sphere 433
are assuming fields of unit intensity.) Of course, the right and left
circularly polarized fields in (13.3) can be expressed as superpositions
of electric and magnetic multipole fields which we have discussed
earlier. Therefore, for the fields in (13.3), we can write in general
(see the discussion in sections 9.4 and 9.5 on multipole expansions,
in particular, see (9.116) and (9.117))
Xh
ER,L (x) = a∓ (ℓ, m)jℓ (kr)Yℓ,m
ℓ,m
i i
+ b∓ (ℓ, m)∇ × (jℓ (kr)Yℓ,m ) ,
k
Xh i
BR,L (x) = − a∓ (ℓ, m)∇ × (jℓ (kr)Yℓ,m )
k
ℓ,m
i
+ b∓ (ℓ, m)jℓ (kr)Yℓ,m . (13.4)
Here, we have used the well behaved spherical Bessel functions for the
expansion of the plane waves (and not the spherical Neumann func-
tion) since the plane wave is well behaved everywhere. The constants
a∓ (ℓ, m) and b∓ (ℓ, m) specify respectively the amounts of magnetic
multipole terms and the electric multipole terms present in the wave
corresponding to a given (ℓ, m). Furthermore, Yℓ,m (θ, φ) represent
the vector spherical harmonics defined in (9.114).
To determine the expansion coefficients a∓ (ℓ, m), b∓ (ℓ, m), we
need to understand some of the properties of the vector spherical
harmonics. We have already seen in (9.115) that
Z
dΩ Yℓ∗′ ,m′ (θ, φ) · Yℓ,m (θ, φ) = δℓℓ′ δmm′ . (13.5)
Let us next note that for any arbitrary radial function fℓ (r), we have
∇ × (fℓ (r)Yℓ,m (θ, φ)) = (∇fℓ (r)) × Yℓ,m + fℓ (r)∇ × Yℓ,m
dfℓ (r) fℓ (r)

= r̂ × Yℓ,m + p ∇ × LYℓ,m, (13.6)
dr ℓ(ℓ + 1)
where we have used the definition of the vector spherical harmonics

given in (9.114). Furthermore, using the relation (9.122) we obtain

2 ∂
∇ × L Yℓ,m = −i r∇ − ∇ 1 + r Yℓ,m
∂r
= −ir(∇2 Yℓ,m) + i(∇Yℓ,m)

ir 1
= ℓ(ℓ + 1)Yℓ,m + r̂ × LYℓ,m . (13.7)
r2 r
Here, we have used the fact that the spherical harmonics depend only
on the angular coordinates and are eigenstates of L2 (see Eq. (9.97))
and that

1 ∂ ∂ 1
∇2 = 2 r2 − 2 L2 . (13.8)
r ∂r ∂r r
We have also used the decomposition of the gradient given in (9.108).
Substituting all these relations into (13.6), we obtain
p r̂
∇ × (fℓ (r)Yℓ,m ) = i ℓ(ℓ + 1) fℓ (r)Yℓ,m
r

dfℓ (r) fℓ (r)
+ + r̂ × Yℓ,m . (13.9)
dr r
It follows now that
Z
dΩ Yℓ∗′ ,m′ · (∇ × (fℓ (r)Yℓ,m ))
Z
= dΩ Yℓ∗′ ,m′ L · (∇ × (fℓ (r)Yℓ,m ))
Z
1 dfℓ (r) fℓ (r)
= + dΩ Yℓ∗′ ,m′ L · (r × Yℓ,m )
r dr r
Z
1 dfℓ fℓ
= p + dΩ Yℓ′ ,m′ L · (r × L)Yℓ,m , (13.10)
r ℓ(ℓ + 1) dr r
where we have used (9.98). Furthermore, from the definition of the

angular momentum operator in (9.95), it follows that
L · (r × L) = ǫijk Li rj Lk
= −iǫijk ǫipq rp ∇q rj Lk
= −i (rj ∇k rj Lk − rk ∇j rj Lk )

= −i r · L + r 2 ∇ · L + r · L − ∇j (rj r · L)
= 0. (13.11)
Here we have used (9.98) as well as the fact that the defining relation
in (9.95) leads to
∇ · L = 0. (13.12)
As a consequence of (13.12), it follows now that

Z
dΩ Yℓ∗′ ,m′ · (∇ × (fℓ (r)Yℓ,m )) = 0. (13.13)
With these identities, we are now ready to determine the coef-

ficients of expansion in (13.4). Using (13.5) and (13.13), we obtain
from Eq. (13.4) that
Z
∗
a∓ (ℓ, m)jℓ (kr) = dΩ Yℓ,m · ER,L
Z
1
=p dΩ (LYℓ,m )∗ · (x̂ ∓ iŷ) eikz
2ℓ(ℓ + 1)
Z
1
=p dΩ (L± Yℓ,m )∗ eikz
2ℓ(ℓ + 1)
s Z
(ℓ ∓ m)(ℓ ± m + 1) ∗
= dΩ Yℓ,m±1 eikz , (13.14)
2ℓ(ℓ + 1)
where we have used the standard definition
L± = Lx ± iLy = L · (x̂ ± iŷ), (13.15)
as well as the action of L± on spherical harmonics. If we now use the

expansion of the plane wave in (13.2) as well as the orthonormality
of the spherical harmonics, we obtain
s
(ℓ ∓ m)(ℓ ± m + 1)
a∓ (ℓ, m)jℓ (kr) =
2ℓ(ℓ + 1)
X p Z
′ ∗
× iℓ 4π(2ℓ′ + 1)jℓ′ (kr) dΩ Yℓ,m±1 Yℓ′ ,0
ℓ′
s
(ℓ ∓ m)(ℓ ± m + 1)
=
ℓ(ℓ + 1)
X ′p
× iℓ 2π(2ℓ′ + 1)jℓ′ (kr) δℓ,ℓ′ δm,∓1
ℓ′
p
= iℓ 2π(2ℓ + 1) jℓ (kr)δm,∓1 ,
p
or, a∓ (ℓ, m) = iℓ 2π(2ℓ + 1) δm,∓1 . (13.16)
Similarly, using (13.3) and (13.16), we can determine

Z
∗
b∓ (ℓ, m)jℓ (kr) = dΩ Yℓ,m · BR,L
Z
∗
= ±i dΩ Yℓ,m · ER,L
= ±ia∓ (ℓ, m)jℓ (kr),

or, b∓ (ℓ, m) = ±ia∓ (ℓ, m)
p
= ±iℓ+1 2π(2ℓ + 1) δm,∓1 . (13.17)
Substituting Eqs. (13.16) and (13.17) into (13.4), we obtain
∞
X p
ER,L = iℓ 2π(2ℓ + 1)
ℓ=1

1
× jℓ (kr)Yℓ,∓1 ∓ ∇ × (jℓ (kr)Yℓ,∓1 ) ,
k
∞
X p
BR,L = iℓ 2π(2ℓ + 1)
ℓ=1

i
× − ∇ × (jℓ (kr)Yℓ,∓1 ) ± ijℓ (kr)Yℓ,∓1 . (13.18)
k
We see that since the projection of the angular momentum along
the z axis is unity for circularly polarized waves, only the m = ±1
azimuthal quantum numbers enter into the multipole expansion of the
electric and the magnetic fields (which is the reason ℓ = 0 is excluded
from the sum). This has to be contrasted with the expansion of the
scalar plane wave which involves only m = 0.
With these basics, let us now discuss the scattering of a plane
electromagnetic wave from a perfectly conducting sphere of radius d.
The incident wave can be right or left circularly polarized (or linearly
polarized which is a superposition of these two) and would scatter
due to the presence of the sphere. Therefore, at large distances away
from the scattering source, we expect outgoing spherical scattered
waves. We can, therefore, write the total electric and magnetic fields
to have the forms
(inc) (sc) (inc) (sc)
ER,L = ER,L + ER,L , BR,L = BR,L + BR,L , (13.19)
(inc) (inc)
where we can assume the incident plane waves ER,L , BR,L of unit
intensity to be given by (13.3) or (13.18). On the other hand, we
expect the scattered waves to have outgoing spherical wave forms at

large distances. Therefore, following our earlier discussions, we can
write
∞
(sc) 1 X ℓp
ER,L = i 2π(2ℓ + 1)
2
ℓ=1

(1) d∓ (ℓ) (1)
× c∓ (ℓ)hℓ (kr)Yℓ,∓1 ∓ ∇ × hℓ (kr)Yℓ,∓1 ,
k
∞
(sc) 1 X ℓp
BR,L = i 2π(2ℓ + 1)
2
ℓ=1

ic∓ (ℓ) (1) (1)
× − ∇ × hℓ (kr)Yℓ,∓1 ± id∓ (ℓ)hℓ (kr)Yℓ,∓1 ,
k
(13.20)
(1)
where hℓ (kr) is the spherical Hankel function of the first kind de-
fined in (9.91) which is spherically outgoing at large distances. The
fields in (13.20) have exactly the same forms as in (13.18) except for
the constant coefficients c∓ (ℓ), d∓ (ℓ) which will be determined from
the boundary conditions.
As we have already seen, there are no electric and magnetic
fields inside a perfect conductor and the conditions satisfied by the
electric and the magnetic fields on the boundary surface are given by
(see (7.4))
n̂ × E| = 0, n̂ · B| = 0. (13.21)
For a perfectly conducting sphere of radius d, the boundary is at r = d
(we are assuming that the origin of the coordinate system lies at the
center of the sphere) and the unit vector normal to the boundary is
the radial unit vector r̂. Thus, imposing the boundary condition on
the electric field, we obtain

(inc) (sc)
r̂ × ER,L + ER,L = 0. (13.22)
r=d
Let us recall from Eq. (13.9) that

dfℓ (r) fℓ (r)
r̂ × (∇ × (fℓ (r)Yℓ,m )) = + r̂ × (r̂ × Yℓ,m )
dr r

dfℓ (r) fℓ (r)
=− + Yℓ,m
dr r
1 d
=− (rfℓ (r)) Yℓ,m , (13.23)
r dr
where we have used the definition of the vector spherical harmonics

in (9.114) as well as the orthogonality relation (9.98). Furthermore,
we note that
Z
dΩ Yℓ∗′ ,m′ · (r̂ × Yℓ,m ) = 0, (13.24)
which follows from (13.11).

Using these identities, we obtain from (13.22)
Z
∗ (inc) (sc)
dΩ Yℓ,m · (r̂ × (ER,L + ER,L )) = 0,
r=d

∂ d∓ (ℓ) ∂ (1)
or, (rjℓ (kr)) + rhℓ (kr) = 0, (13.25)
∂r 2 ∂r r=d
where we have used the orthonormality of the vector spherical har-
monics in (9.115). Similarly, the boundary condition on the magnetic
field (see (13.21)) leads to
c∓ (ℓ) (1)
jℓ (kd) + hℓ (kd) = 0. (13.26)
2
Relations (13.25) and (13.26) determine the arbitrary coefficients
present in the definitions of the scattered waves, namely,
(2)
!
2jℓ (kd) hℓ (kd)
c∓ (ℓ) = − (1) = − (1) −1 ,
hℓ (kd) hℓ (kd)

∂
2 ∂r (rjℓ (kr))
d∓ (ℓ) = − (1)

∂
(rh (kr))
∂r ℓ r=d
" (2)
#
∂
(rhℓ (kr))
= − ∂r (1)
−1 , (13.27)
∂
∂r (rhℓ (kr)) r=d
where we have used the definition of the spherical Bessel functions in
(9.91).
(2) (1)
Let us recall that hℓ (kr) = (hℓ (kr))∗ . Therefore, we see that
the first term on the right hand side of each of the defining relations
in (13.27) has unit modulus. This allows us to define phase angles of
the forms
(2)
2iδℓ hℓ (kd)
e =− (1)
,
hℓ (kd)

∂ (2)
2iδℓ′ ∂r (rhℓ (kr))
e =− (1)
. (13.28)
∂
(rh (kr))
∂r ℓ r=d
We can think of δℓ , δℓ′ respectively as the magnetic and the electric

phase shifts in the ℓth multipole fields. This is completely parallel to
the analysis of scattering in quantum mechanics. In fact, in terms of
these phase shifts, we can write
c∓ (ℓ) = (e2iδℓ − 1) = 2ieiδℓ sin δℓ ,
′ ′
d∓ (ℓ) = (e2iδℓ − 1) = 2ieiδℓ sin δℓ′ . (13.29)
Correspondingly, the scattered electric and magnetic fields can be
written as
∞
1 X ℓp h
(sc) (1)
ER,L = i 2π(2ℓ + 1) (e2iδℓ − 1)hℓ (kr)Yℓ,∓1
2
ℓ=1
′
(e2iδℓ − 1) i
(1)
∓ ∇ × (hℓ (kr)Yℓ.∓1 ) ,
k
∞
1 X ℓp h i(e2iδℓ − 1)
(sc) (1)
BR,L = i 2π(2ℓ + 1) − ∇ × (hℓ (kr)Yℓ,∓1 )
2 k
ℓ=1
′
i
(1)
± i(e2iδℓ − 1)hℓ (kr)Yℓ,∓1 . (13.30)
Let us next consider the asymptotic forms of these fields for
large r, far away from the scattering source. In this case, we know
that
(1) eikr
hℓ (kr) → (−i)ℓ+1 . (13.31)
kr
It follows, therefore, from Eq. (13.9) as well as (13.31) that for large
r
(1) (−i)ℓ eikr
∇ × (hℓ (kr)Yℓ,∓1 ) → r̂ × Yℓ,∓1 . (13.32)
r
Substituting these into the expression for the scattered electric and
magnetic fields, we obtain that, for large r,
(sc) eikr
ER,L → f∓ (θ, φ),
r
(sc) (sc)
BR,L → r̂ × ER,L , (13.33)
∞ p
X 2π(2ℓ + 1)
f∓ (θ, φ) =
k
ℓ=1
h ′
i
× eiδℓ sin δℓ Yℓ,∓1 ∓ ieiδℓ sin δℓ′ r̂ × Yℓ,∓1 . (13.34)
In analogy with the discussion of scattering in quantum mechanics,

we can think of f∓ (θ, φ) as the vector scattering amplitude. We note
that the second relation in (13.33) is what we would expect in the far
(radiation) zone.
Let us note now that the average power radiated by the scattered
wave per unit solid angle through a large spherical surface of radius
R is given by
(sc)
dPR,L cR2
(sc) (sc)
= Re r̂ · ER,L × (BR,L )∗
dΩ 8π
cR2 (sc) 2
= |ER,L |
8π
c
= |f∓ (θ, φ)|2 . (13.35)
8π
We also note from the form of the incident wave (13.3) that it is a
plane wave traveling along the z-axis. Therefore, the average incident
power per unit area is given by
c
(inc) (inc)
IR,L = Re ẑ · ER,L × (BR,L )∗
8π
c (inc) c
= |ER,L |2 = . (13.36)
8π 8π
In this case, we obtain the differential cross section for scattering,
(13.1), to be
σR,L (θ, φ) = |f∓ (θ, φ)|2 . (13.37)
This clarifies the meaning of f∓ (θ, φ) as the (vector) scattering am-

plitude. The total scattering cross section can now be obtained by
integrating the differential cross section and, using the orthonormal-
ity relations for the spherical harmonics, we obtain
Z
total
σR,L = dΩ |f∓ (θ, φ)|2
∞
2π X
= 2
(2ℓ + 1) sin2 δℓ + sin2 δℓ′ . (13.38)
k
ℓ=1
This shows that the magnetic and the electric multipole fields con-
tribute incoherently to the total scattering cross section (there is no
cross term because of (13.24)) although, as we will see, there can be
interference terms present in the differential cross section.
With these general observations, let us calculate the differential

scattering cross section in the long wave length limit kd ≪ 1. This
would particularly be true for scattering by small particles. In this
case, from the definitions of the spherical Bessel functions we obtain
that
jℓ (kd) (kd)2ℓ+1
tan δℓ = →−
ηℓ (kd) (2ℓ + 1)[(2ℓ − 1)!!]2
(kd)2ℓ+1
or, δℓ ≈ − ,
(2ℓ + 1)[(2ℓ − 1)!!]2

∂
(rj (kr)) ℓ+1 (kd)2ℓ+1
ℓ
tan δℓ′ = ∂r →
∂
∂r (rηℓ (kr))
ℓ (2ℓ + 1)[(2ℓ − 1)!!]2
r=d
ℓ+1
or, δℓ′ ≈ − δℓ . (13.39)
ℓ
This shows that the higher angular momentum components of the
phase shifts fall off rapidly with ℓ in the long wave length limit. There-
fore, we can approximate the expression for the scattering amplitude
by keeping only the lowest order term corresponding to ℓ = 1. We
note from (13.39) that
(kd)3
δ1 ≈ − , δ1′ ≈ −2δ1 . (13.40)
3
Using this, the scattering amplitude in this approximation becomes
√
6π
f∓ (θ, φ) ≈ δ1 (Y1,∓1 ± 2i r̂ × Y1,∓1 ) . (13.41)
k
Furthermore, using the properties of the angular momentum opera-
tors and the spherical harmonics, we obtain
σR,L (θ, φ) = |f∓ (θ, φ)|2

6π
≈ |δ1 |2 |Y1,∓1 |2 + 4|Y1,∓1 |2
k2
∗

±4Im Y1,∓1 · (r̂ × Y1,∓1 )

2 4 5 2
= d (kd) (1 + cos θ) − cos θ . (13.42)
8
Here, we have used the definitions of the spherical harmonics in
(9.138) as well as L in (9.95). The cos θ term represents the inter-
ference between the magnetic and the electric dipole terms (ℓ = 1).
We note that the differential cross section is the same for both the
right and the left circularly polarized waves. Furthermore, it is pro-
portional to the fourth power of the frequency and has a peak in the
backward direction θ = π. Integrating this, we obtain
Z
10πd2
σR,L = dΩ |f∓ (θ, φ)|2 =
total
(kd)4 . (13.43)
3
The interference term does not contribute to the total cross section.
The dependence of the scattering cross section on the fourth power
of the frequency is a characteristic of dipole fields and is known as
Rayleigh’s law.
13.2 Kirchhoff’s approximation
We have already discussed briefly about the Kirchhoff’s representa-

tion in section 6.9. However, the discussion there was in terms of
the scalar potential. The basic observable fields in the Maxwell the-
ory are, on the other hand, vector fields, namely, the electric and the
magnetic fields. Thus, in this section we will generalize the discussion
of section 6.9 to vector fields. Let us, however, emphasize that the
components of the electric and magnetic fields can be thought of as
scalar functions (as far as the discussion of the Kirchhoff’s represen-
tation is concerned) and, as a result, we can, in principle, carry out
the discussion from the results already derived earlier. But, for com-
pleteness as well as continuity with the earlier section on diffraction,
we will discuss here Kirchhoff’s representation for vector fields.
Let us recall that if we Fourier transform the time variable (see
discussion in section 9.4), Maxwell’s equations in a dielectric medium
take the forms
4π
∇·E= ρ,
ǫ
∇ · B = 0,
iω
∇×E= B,
c
4πµ iǫµω
∇×B= J− E. (13.44)
c c
Defining
√
ǫµ ω
k= ,
c
13.2 Kirchhoff’s approximation 443
we note that (13.44) leads to

4iπµω
∇ × (∇ × E) − k 2 E = J,
c2
4πµ
∇ × (∇ × B) − k 2 B = ∇ × J. (13.45)
c
In the absence of sources, both the electric and the magnetic fields
are transverse and, consequently, (13.45) reduces to the Helmholtz
equations

∇2 + k 2 E(x, ω) = 0,

∇2 + k 2 B(x, ω) = 0, (13.46)
the solutions of which we have discussed in detail in section 9.4.

In the presence of sources, on the other hand, we see from (13.45)
that both the electric and the magnetic fields satisfy an equation of
the form
∇ × (∇ × V) − k 2 V = 4πf ,

or, ∂i ∂j − δij ∇2 + k 2 Vj = 4πfi , (13.47)
where Vi stands for the three components of either the electric or the
magnetic fields and fi the sources. (Repeated indices are assumed
to be summed.) To solve such an inhomogeneous equation, we will
make use of the method of Green’s functions. In this case, we note
that the Green’s function will be a tensor and from (13.47) it follows
that the equation satisfied by the Green’s function would have the
form

∂i ∂m − δim ∇2 + k 2 Gmj (x − x′ ) = 4πδij δ3 (x − x′ ). (13.48)
From the definition in (13.48) we see that the Green’s function is a

second rank symmetric tensor, namely,
Gij (x − x′ ) = Gji (x′ − x). (13.49)
Furthermore, the particular solution of an inhomogeneous Helmholtz

equation of the form (13.47) can now be written as
Z
Vi (x, ω) = d3 x′ Gij (x − x′ )fj (x′ ). (13.50)
The form of the Green’s function satisfying (13.48) can now be de-
termined along the lines discussed in section 6.8. First, we note that
the Green’s function for the scalar Helmholtz equation (9.87) satisfy-
ing the boundary conditions that at large distances it represents an
outgoing wave has the form
eikx
g(x) = , (13.51)
x
where x = |x|, namely,

∇2 + k 2 g(x) = −4πδ3 (x).
It can now be easily checked that the tensor Green’s function in

(13.48) is related to the scalar Green’s function as

′ 1
Gij (x − x ) = ∂i ∂j + δij g(x − x′ ). (13.52)
k2
The symmetry (13.49) of the Green’s function is manifest in the form

(13.52).
The Green’s identity (3.67) or (6.197), in this case, can be de-
rived as follows. Let us note that for an arbitrary vector Vi and a
second rank tensor Aij (not necessarily symmetric), we can write
Z h i
′ ′
d3 x′ Vi ∂i′ ∂j′ − δij ∇ 2 Ajk − Aik ∂i′ ∂j′ − δij ∇ 2 Vj
Z

= d3 x′ ∂i′ Vj ∂j′ Aik − ∂i′ Ajk + Ajk ∂i′ Vj − ∂j′ Vi
Z

= ds′ n′i Vj ∂j′ Aik − ∂i′ Ajk + Ajk ∂i′ Vj − ∂j′ Vi
Z

= ds′ − n′i Vj − n′j Vi ∂i′ Ajk + Ajk n′i ∂i′ Vj − ∂j′ Vi ,
(13.53)
where we have used Gauss’ theorem. Furthermore, n′ denotes the

unit vector normal to the surface S ′ . There are now two cases to
consider. If we identify
Vi (x′ ) = Ei (x′ , ω), Aij (x, x′ ) = Gij (x − x′ ), (13.54)
then, in a region within the bounding surface (free of sources), (13.53)

leads to
Z
1
Ei (x, ω) = − ds′ n′j Em − n′m Ej ∂j′ Gmi
4π

ik
− √ ǫjmℓ Gmi n′j Bℓ
ǫµ
Z
1 ′ ′
′ ik ′
=− ds ǫjℓm n × E m ∂j Gℓi − √ Bm ǫmjℓ nj Gℓi .
4π ǫµ
(13.55)
Here we have used (13.44) in a source free region as well as (13.48).

It is clear that (13.55) can determine uniquely the electric field when
the values of the tangential components (of the electric field) are
specified on a given surface provided the Green’s function satisfies
the boundary condition
ǫjℓmn′ℓ Gmp (x − x′ ) = 0. (13.56)
The Green’s function satisfying such a boundary condition is known

(e)
as the electric tensor Green’s function and is denoted by Gij . Choos-
ing such a boundary condition, we obtain from (13.55)
Z
1 (e)
Ei (x, ω) = − ds′ n′ × E(x′ ) m ǫmjℓ ∂j′ Gℓi (x−x′ ). (13.57)
4π
The magnetic field can now be determined from (13.44) to be

√
i ǫµ
Bi (x, ω) = − (∇ × E(x, ω))i
k
√ Z
i ǫµ
=− ds′ n′ × E(x′ ) j ǫjℓm ǫipq ∂ℓ′ ∂p′ G(e) ′
mq (x − x ).
4πk
(13.58)
The combination
ǫjℓmǫipq ∂ℓ′ ∂p′ G(e) ′

mq (x − x ), (13.59)
in the earlier equation is sometimes also referred to as the magnetic

tensor Green’s function (up to a multiplicative constant). Thus, we
see that the Green’s identity allows us to solve for the electric and
the magnetic fields in terms of the boundary values of the tangential
components of the electric field on a given surface and the electric
tensor Green’s function.
Similarly, identifying
Vi (x′ ) = Bi (x′ ), Aij (x, x′ ) = Gij (x − x′ ), (13.60)
in a source free region (within the bounding surface), we obtain from
(13.53),
Z
1
Bi (x, ω) = − ds′ n′j Bm − n′m Bj ∂j′ Gmi
4π
√
+i ǫµk ǫjmℓ Gmi n′j Eℓ
Z
1 √
=− ds′ ǫjℓm n′ × B m ∂j′ Gℓi + i ǫµk Em ǫmjℓ n′j Gℓi .
4π
(13.61)
We see once again that if we have the electric tensor Green’s function,
then the magnetic field can be solved uniquely in terms of the bound-
ary values of its tangential components specified on a given surface
as
Z
1 (e)
Bi (x, ω) = − ds′ n′ × B(x′ ) m ǫmjℓ ∂j′ Gℓi (x − x′ ).
4π
(13.62)
The electric field then follows from (13.44) to be
i
Ei (x, ω) = √ (∇ × B(x, ω))i
ǫµk
Z
i
= √ ds′ n′ × B(x′ ) j ǫjℓm ǫipq ∂ℓ′ ∂p′ G(e) ′
mq (x − x ).
4π ǫµk
(13.63)
This analysis makes it clear that the electric and the magnetic fields
can be determined uniquely if the values of the tangential compo-
nents of either the electric or the magnetic field are given on a given
boundary surface. However, specifying the tangential components of
both the electric as well as the magnetic fields on a boundary over-
specifies the system unless the tangent components of the electric
and the magnetic fields on the boundary are consistent. In such a
case, the electric field at any point within the region bounded by the
surface is given by (13.55), namely,
Z h
1
Ei (x, ω) = − ds′ ǫjℓm n′ × E m ∂j′ Gℓi
4π
ik i
− √ Bm ǫmjℓ n′j Gℓi . (13.64)
ǫµ
The relations obtained above are exact in the sense that there
has been no approximation used so far. However, determining the
boundary conditions in a given problem, (namely, determining the
tangential components of the electric and the magnetic fields on the
boundary) is in general difficult and it is here that approximations
creep in. In particular, Kirchhoff’s approximation uses the notions
from geometrical optics in estimating the tangential components of
the fields on a boundary surface. For example, let us assume that we
are considering diffraction of electromagnetic waves from a spherical
conducting surface. In this case, the surface of the sphere can be
divided into two parts, one that is illuminated by the incident wave
and the other that is in the shadow. Kirchhoff’s approximation con-
sists of assuming that in the illuminated part of the spherical surface,
we have already seen from our studies on reflection from a perfectly
conducting surface (see the discussion in sections 8.2 and 8.3) that

n × E = n × E(inc) + E(refl) = 0,

n × B = n × B(inc) + B(refl) = 2n × B(inc) . (13.65)
Using these we conclude that, in the illuminated part of the sphere,

we can write
(sc) (sc)
n × E(inc) = −n × EI , n × B(inc) = n × BI . (13.66)
On the other hand, in the shadow part of the spherical surface, Kirch-
hoff’s approximation assumes that there is no total field so that
(sc) (sc)
n × E(inc) = −n × EII , n × B(inc) = −n × BII . (13.67)
Here the subscripts I and II refer to the two regions of the surface of
the sphere.
Without going into too much technical details, let us indicate
how Kirchhoff’s approximation can be used to calculate the diffrac-
tion of a plane wave from a conducting sphere. Let us assume that
the plane wave is incident in vacuum (ǫ = 1 = µ) along the z-axis on a
perfectly conducting sphere of radius a. Thus, we identify k̂ = ẑ = n′ .
We assume that the origin of the coordinate system coincides with
the center of the sphere. Then, we can consider a point outside the
sphere to be contained in a region bounded by the surface of the con-
ducting sphere as well as the large spherical surface at infinity. With
a little bit of analysis, it can be shown that the surface integral over
the large sphere at infinity vanishes. Thus, we can write the scattered
electric field in terms of the fields on the surface of the conducting

sphere as
Z h
(sc) 1
Ei (x) =− ds′ ǫjℓm n′ × E(sc) ∂j′ Gℓi
4π m
i
(sc)
− ik Bm ǫmjℓ n′j Gℓi
Z h i
1
=− ds′ ǫjℓm k̂ × E(sc) ∂j′ Gℓi − i Bm
(sc)
ǫmjℓ kj Gℓi ,
4π m
(13.68)
where the surface integral is over the conducting surface. As we have

noted earlier, the surface integral can be divided into two parts, one
over the illuminated region and the other over the shadow region and
the boundary conditions are different for the two regions in Kirch-
hoff’s approximation. We see from (13.67) that we can write the
contribution from the shadow region as
Z h
(sc) 1
EII,i (x) = ds′ ǫjℓm k̂ × E(inc) ∂j′ Gℓi
4π m
i
(inc)
−iBm ǫmjℓ kj Gℓi . (13.69)
The shadow region has an interesting property that as long as

the observation point x does not lie inside the surface of the sphere,
the shadow integral can be evaluated over any surface bounded by
the diametric vertical plane (that separates the illuminated and the
shadow regions). This property can be easily proved using Gauss’
theorem, but the consequence of this is that the shadow integral can
be simplified and evaluated over the diametric plane. Here z = 0
and we can parameterize x′ = ρ = xêx + yêy . Furthermore, if the
observation point is very far away from the surface of the sphere,
namely, x ≫ x′ , then we can approximate
eikx −ik·x′ ′
g(x − x′ ) ≈ e = g(x) e−ik·x ,
x

ki kj ′
Gij (x − x ) ≈ − 2 + δij g(x) e−ik·x ,
′
(13.70)
k
where we have used the definition of the tensor Green’s function in

(13.52). Using this in (13.69), we obtain that at large distances away
from the conducting sphere, the contribution coming from the shadow
region has the form

Z h i
(sc) ik (inc) ′
EII,i (x) ≈− g(x) ds′ −Ei + k̂ × B(inc) e−ik·x
4π i
Z
ik (inc) ′
= g(x)Ei ds′ e−ik·x , (13.71)
2π
(inc)
where Ei represents the amplitude of the incident wave and we
have used the relation
k̂ × B(inc) = −E(inc) . (13.72)
The surface integral can now be done in a straightforward man-

ner
Z Z
′ −ik·x′ ′
ds e = ρdρdφ′ e−ikρ sin θ cos(φ−φ )
Z a
= ρdρ 2πJ0 (kρ sin θ)
0
aJ1 (ka sin θ)

= 2π , (13.73)
k sin θ
where we have used the formulae from the standard tables (see, for
example, Gradshteyn and Ryzhik, 6.5615 and 8.4111). Putting to-
gether all these, the contribution from the shadow region at large
distances can be written as
(sc) iaJ1 (ka sin θ) eikx (inc)

EII,i (x) ≈ Ei . (13.74)
sin θ x
The contribution from the illuminated region can also be cal-
culated in a similar manner. However, the integral, in this case, is
slightly involved (simply because it has to be carried out over the
hemisphere) and we will not go into the details of that.
Index
Čerenkov effect, 352 boundary value problem, 69

Brewster’s angle, 189
Abraham-Lorentz force, 366
absorption, 278 capacitance, 63
adiabatic invariant, 423 capacitor, 63
Alfvén speed, 394 parallel plate, 64
Alfvén wave, 391 spherical, 65
Ampere’s law, 151, 171 Cauchy problem, 82
generalization, 171 cavity resonator, 249
amperian loop, 152 characteristic curve, 86
angle of incidence, 187 characteristic point, 82
angle of reflection, 187
charge above a plane, 69, 110
angle of transmission, 187
charge inside a sphere, 112
angular frequency, 176
circular harmonics, 103
anomalous magnetic moment, 424,
cogradient, 207
430
collision, 394
antenna
conducting medium, 263
center-fed, 301
conductivity, 175, 225, 226
dipole, 301
conductor, 58
monopole, 301
good, 276
properties, 307
perfect, 276
attenuation factor, 246, 247
conservative field, 19
Bessel equation, 239 continuity equation, 144, 171
Bianchi identity, 398 contragradient, 207
Biot-Savart law, 146, 295 contravariant metric tensor, 204
boost, 204 contravariant vector, 203
boundary condition, 80, 163, 181, Coulomb potential, 5
226, 263 Coulomb gauge, 198
Cauchy, 82 Coulomb’s law, 1
Dirichlet, 87, 109 covariant metric tensor, 204
in dielectric, 128 covariant vector, 203
Neumann, 87, 114 cross section, 431
retarded, 214 differential, 431, 440
451
452
current, 142, 143, 323 energy density, 190

current density, 143 energy loss
cut-off frequency, 234, 236, 241, mechanical, 191
242 Euclidean space, 205
cyclotron frequency, 413, 430
cylindrical electret, 137 far zone, 290
farad, 64
D’Alembertian, 198, 207 Faraday rotation, 389
Debye length, 377 Faraday’s cage, 61
diamagnetic material, 157 Faraday’s law, 166
dielectric, 121 ferromagnetic material, 158
capacitor, 127 field
charge in, 125 asymptotic value, 341
constant, 125 field strength tensor, 209
linear, 122 four vector, 202
lossy, 395 four velocity, 399
diffraction, 431 Galilean transformation, 200
dipole, 46 gauge choice, 197, 401
continuous distribution, 50 gauge condition, 154
Hertzian, 292 gauge invariance, 153, 401
magnetic, 142 gauge transformation, 397
moment, 47 Gauss, 141
dipole approximation, 343 Gauss’ law, 9
Dirac delta function, 7 Gaussian surface, 13, 14, 16
dispersion, 179 Gaussian unit, 3
displacement current, 173 generating function
displacement field, 121, 124, 173 Legendre polynomial, 98
dominant mode, 236 Green’s function, 107, 443
double layer of charge, 55 Dirichlet, 110
dual field strength, 210 electric, 445
magnetic, 445
Einstein relation, 207 Neumann, 115
electric field, 5 retarded, 212
electric phase shift, 439 Green’s identity, 91, 218
electromagnetic interactions, 397 group velocity, 177
electromagnetic waves, 176 guiding center, 420
electromotive force, 167
electrostatic energy, 24, 65 Hankel function
electrostatic screening, 127 first kind, 336
electrostatics, 1 harmonic function, 90
elliptic equation, 87 general property, 90
emf, 167 helicity, 430
453
Helmholtz equation, 308 nature, 150

Helmholtz theorem, 19 of circular wire, 148
Huygen’s principle, 218, 222 of straight wire, 146
hyperbolic equation, 86 magnetic field intensity, 161
hysteresis, 158 magnetic induction, 141
magnetic monopole, 142, 150
image charge, 71 magnetic phase shift, 439
impedance, 278 magnetic susceptibility, 162
surface, 278 magnetization, 157
index of refraction, 352 magnetostatics, 141, 144
plasma, 381 laws of, 151
induced charge density, 73 Maxwell’s equations, 171, 174
inductance, 167 covariance, 208
inner product, 206 gauge invariance, 195
insulator, 58 plane wave solution, 175
invariant length, 204 method of images, 69, 132
method of virtual photon, 337
Kirchhoff’s approximation, 442
minimal coupling, 407
Kirchhoff’s representation, 218, 222
Minkowski space, 205
Laplace equation, 21, 87 mirror point, 424
Cartesian coordinate, 93 mirroring, 420
solution, 90 MKS unit, 3
Laplacian, 21 motion
Larmour formula, 346 crossed electric and magnetic
Legendre equation, 97 fields, 414
associated, 97 slowly varying magnetic field,
Legendre polynomial, 44, 97 419
associated, 97 uniform electric field, 408
orthonormality, 99 uniform magnetic field, 412
Levi-Civita tensor, 210 multipole expansion, 45, 155, 308
Lienard-Wiechert potential, 217, mutual inductance, 169
323, 324 near zone, 290
light-like vector, 205
linear acceleration, 346 Ohm’s law, 175, 226
linearized approximation, 378 optical fiber, 256
Lorentz force, 141, 175
Lorentz scalar, 206 parabolic equation, 88
Lorentz transformation, 200 paramagnetic material, 158
Lorenz gauge, 197 partial differential equation, 82
Pauli-Lubanski spin, 427
magnetic dipole moment, 156 penetration depth, 226, 277
magnetic field, 141 permeability, 162

Lectures On Electromagnetism PDF

Uploaded by

Copyright:

Available Formats

Lectures On Electromagnetism PDF

Uploaded by

Document Information

Original Title

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

Lectures On Electromagnetism PDF

Uploaded by

Copyright:

Available Formats

Lectures on

British Library Cataloguing-in-Publication Data

Printed in India, bookbinding made in Singapore.

Lakshmi - Lecs on Electromagnetism.pmd 1 2/22/2013, 5:00 PM

3.4.2 Dirichlet boundary condition . . . . . . . . . . 109

7.5 Wave impedance. . . . . . . . . . . . . . . . . . . . . 245

11.8 Selected problems . . . . . . . . . . . . . . . . . . . . 396

Over the years, I have taught a two semester graduate course on

1. P. C. Clemmow, “Introduction to Electromagnetic Theory”,

2. L. Eyges, “The Classical Electromagnetic Field”, Dover Publi-

3. J. Frenkel, “Princı́pios de Eletrodinâmica Clássica”, Editora da

4. W. Hauser, “Introduction to the Principles of Electromagnetism”,

Several of my colleagues at Rochester and at other universities,

Preface to the second edition

The modifications in this second edition of the book arose mainly

Electrostatics is the study of the properties of electric charge distri-

1.1 Coulomb’s law

1. is linearly proportional to the individual charges on each of the

2. varies inversely as the square of the distance between the two

Quantitatively, we can, therefore, write that the force experi-

Figure 1.1: Direction of the electrostatic force experienced by a charge

where r denotes the coordinate of the charge q in the new frame

k depends on the system of units used. In the CGS (Gaussian) units

1 esu = 1 statCoulomb = (10 c)−1 C = (3 × 109 )−1 C. (1.3)

In fact, let us suppose, 1 esu = α C. Then,

which determines α = (10 c)−1 and, therefore, 1 esu = (10c)−1 C. (We

This is known as the law of superposition of electric forces.

(in some units). If Coulomb’s law is exact, we should have µ = 0

1.2 Electric field

while, for a distribution of static charges, the electric field at r is

This quantity is inherently a characteristic of a given distribution of

presence of a distribution of charges, it would experience an electric

Integrating this over the volume containing the charge, we obtain

Figure 1.2: A graphical representation of the Dirac delta function

The one dimensional representations in (1.14) can be easily gener-

(δ(x − x′ ))∗ = δ(x − x′ ),

volume of the space on which it is defined). Consequently, it is clear

as it should be (see (1.6)). A similar derivation follows as well for the

1.3 Gauss’ law

Let us consider a continuous charge distribution given by the volume

Figure 1.3: A finite volume containing a continuous distribution of

From the defining relation in (1.11), we obtain the flux of the

where we have interchanged the orders of integration. The surface

where dΩ represents the solid angle subtended at r′ by the infinites-

Figure 1.4: An infinitesimal line element vector in polar coordinates.

Now, let us consider a surface (area) element in spherical coordinates

ds = r dθ θ̂ × r sin θ dφ φ̂ = r 2 sin θ dθ dφ r̂ = r 2 dΩ r̂. (1.23)

Figure 1.5: An infinitesimal surface (area) element vector in spherical

Here dΩ is the solid angle subtended by the surface element at the

dℓ = dr r̂ + r dθ θ̂ + r sin θ dφ φ̂, (1.25)

so that the area along the direction of r̂ is simply given by (here r is

ds = dℓθ × dℓφ = r dθ θ̂ × r sin θ dφ φ̂

where S represents the surface enclosing the volume V . Using this in

or, ∇ · E(r) = 4π ρ(r). (1.30)

This is the differential form of Gauss’ law. It is worth commenting

◮ Example (Modified Coulomb’s law). Let us consider the hypothetical example

Gauss’ law is quite useful in determining the electric field when

Figure 1.6: The dashed surface represents a cylindrical Gaussian sur-

or, |E(r)| 2πrh = 4π hλ,