Advanced Classical Electrodynamics Green Functions, Regularizations, Multipole Decompositions, Jentschura, WS, 2017

10514_9789813222847_tp.
indd 1 20/4/17 4:54 PM

This page intentionally left blank
10514_9789813222847_tp.indd 2 20/4/17 4:54 PM
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
Library of Congress Cataloging-in-Publication Data

Names: Jentschura, Ulrich D., author.
Title: Advanced classical electrodynamics : Green functions, regularizations, multipole decompositions /
by Ulrich D. Jentschura (Missouri University of Science and Technology, USA).
Description: Singapore ; Hackensack, NJ : World Scientific, [2017] |
Includes bibliographical references and index.
Identifiers: LCCN 2017011376| ISBN 9789813222847 (hardcover ; alk. paper) |
ISBN 9813222840 (hardcover ; alk. paper) | ISBN 9789813222854 (pbk. ; alk. paper) |
ISBN 9813222859 (pbk. ; alk. paper)
Subjects: LCSH: Electrodynamics--Textbooks.
Classification: LCC QC631 .J46 2017 | DDC 537.601/51--dc23
LC record available at https://lccn.loc.gov/2017011376
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.
Copyright © 2017 by World Scientific Publishing Co. Pte. Ltd.

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.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center,
Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from
the publisher.
Printed in Singapore
LerhFeng - 10514 - Advanced Classical Electrodynamics.indd 1 23-03-17 9:00:19 AM

April 17, 2017 11:16 ws-book961x669 BC: 10514 - Advanced Classical Electrodynamics 3rd Read jentschura page v
Dedicated to All Who Seek Beauty in Nature
v
April 17, 2017 11:16 ws-book961x669 BC: 10514 - Advanced Classical Electrodynamics 3rd Read jentschura page vii
Preface
Classical electrodynamics is a vast and open field. It would be hard to summarize

all that is known in field within a few hundred pages. Still, classical electrodynamics
has been the first theory to incorporate vector calculus into the realm of theoretical
physics, and notably, the first theory that was (later) shown to be Lorentz covariant,
upon a proper identification and grouping of the physical quantities into so-called
four-vectors. When Stokes set his theorem as an inspired problem in the Cambridge
mathematical tripos examinations, he had no idea how instrumental vector calculus
would later be in the development of physics. We shall attempt to summarize the
most important aspects of classical electrodynamics, with an emphasis on those
concepts that will be useful for further studies in quantum electrodynamics and
field theory. Emphasis will be laid on Green functions.
A general survey of the Maxwell equations is given in Chap. 1, with an em-
phasis on the physical meaning of gauge transformations of the scalar and vector
potentials. The Green functions of electrostatics and the concomitant multipole de-
compositions, in various coordinate systems, are discussed in Chap. 2. In particular,
analytic (multipole) decompositions are contrasted with eigenfunction expansions
of the Green functions of operators. The considerations are generalized to the
Green functions of electrodynamics in Chap. 4. In particular, the resolution of the
action-at-a-distance paradox is discussed, which otherwise plagues the theory of
electromagnetic interactions in the Coulomb gauge.
In Chap. 3, we discuss some paradigmatic calculations in electrostatics, with
an emphasis on the variational principle, and on series expansions of potentials.
The Laplace and Poisson equations are treated. The Green function of the
Helmholtz equations and the theory of harmonically oscillating sources are dis-
cussed in Chap. 5. The multipole decomposition of the radiation into electric and
magnetic multipoles, and longitudinal components of the fields, is discussed on
the basis of vector spherical harmonics, separating the fields into transverse and
longitudinal vector multipole components. We also discuss the calculation of the
Liénard–Wiechert potentials generated by a moving point charge, in Lorenz and
Coulomb gauges.
Electrodynamics in media should not be ignored in any textbook on classical
electrodynamics (see Chap. 6). In view of their illustrative power and technical
vii
April 17, 2017 11:16 ws-book961x669 BC: 10514 - Advanced Classical Electrodynamics 3rd Read jentschura page viii
viii Advanced Classical Electrodynamics
relevance, waveguides and the resonant eigenmodes of cavities (both rectangular as

well as cylindrical) are discussed in Chap. 7.
Finally, in Chap. 8, we discuss special relativity and quantum field theory, and
the connection to electrodynamics. Classical electrodynamics can be augmented in
two ways, first, supplying a manifestly relativistic interpretation, and second, adding
a quantum-field theoretical aspect. These concepts are introduced in Chap. 8.
Among other things, the magnetic force q⃗ v×B ⃗ is derived purely on the basis of the
relativistic Lorentz contraction of charge distributions. Also, the Lorentz transfor-
mation of electromagnetic fields is discussed on the basis of explicit calculations. It
is shown, e.g., that Gauss’s law in the laboratory frame transforms into a superpo-
sition of the Gauss’s law, expressed in terms of the coordinates of Lorentz-boosted
frame, and of the Ampere–Maxwell law. The quantum field-theoretical zero-point
energy is introduced in the calculation of the Casimir force between conducting
plates. Finally, the connection of electrodynamics and general relativity is high-
lighted on the basis of covariant derivatives.
The intended audience of this book is threefold. (i) First, the book is intended
for the advanced student who wishes to enhance his or her background knowledge on
an interesting subfield of theoretical physics, possibly supplementing a course on the
subject taught at a University. (ii) Second, we aim to reach the academic teacher
who wishes to give a one-semester course on electrodynamics, in a modern style.
Exercises represent challenges and serve as benchmarks for one’s own understanding
of the subject, but are designed not to represent insurmountable difficulties. The
use of SI mksA units in all derivations may enhance the applicability to different
subfields of physics where often, different unit systems are used. (iii) Last, but not
least, the book is also intended as a partial encyclopedia for the active researcher
who would like to look up the treatment of, e.g., angular momentum decompositions.
A relatively new development in the latter context is the solution of the problem
of calculating the Liénard–Wiechert potentials for a moving charge in Coulomb
gauge, where, as opposed to Lorentz gauge, the seemingly instantaneous character
of the Coulomb interaction has been an obstacle for a deeper understanding of the
mechanism of retardation in the past.
In writing this book, we acknowledge helpful conversations with Benedikt
J. Wundt and Barbara N. Hale, of Missouri University of Science and Technology,
and help in the proofreading of the manuscript by Amanda Wetzel and Chandra
M. Adhikari. This book project has been supported by the National Science Foun-
dation (Grant PHY–1403973).
Rolla, Missouri, January 2017

Ulrich D. Jentschura
Missouri University of Science and Technology
Department of Physics, 1315 North Pine Street
Rolla, Missouri 65409-0640, USA, [email protected]
April 17, 2017 11:16 ws-book961x669 BC: 10514 - Advanced Classical Electrodynamics 3rd Read jentschura page ix
Contents
Preface vii
1. Maxwell Equations 1
1.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Integral and Differential Forms of the Maxwell Equations . 1
1.1.2 Relation of the Differential to the Integral Form . . . . . . 4
1.1.3 Dirac δ Function . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.4 Green Function of the Laplacian . . . . . . . . . . . . . . . . 11
1.2 Potentials and Gauges . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.1 Transverse and Longitudinal Components of a Vector Field 11
1.2.2 Vector and Scalar Potentials . . . . . . . . . . . . . . . . . . 13
1.2.3 Lorenz Gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.2.4 “Gauge Always Shoots Twice” . . . . . . . . . . . . . . . . . 20
1.2.5 Coulomb Gauge . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3 Poynting Theorem and Maxwell Stress Tensor . . . . . . . . . . . . 23
1.3.1 Electric and Magnetic Field Energies . . . . . . . . . . . . . 23
1.3.2 Maxwell Stress Tensor . . . . . . . . . . . . . . . . . . . . . . 25
1.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2. Green Functions of Electrostatics 29

2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2 Green Function of the Harmonic Oscillator . . . . . . . . . . . . . . 29
2.3 Electrostatic Green Function and Spherical Coordinates . . . . . . 33
2.3.1 Poisson and Laplace Equations in Electrostatics . . . . . . 33
2.3.2 Laplace Equation in Spherical Coordinates . . . . . . . . . 34
2.3.3 Legendre Functions and Spherical Harmonics . . . . . . . . 37
2.3.4 Expansion of the Green Function in Spherical Coordinates 40
2.3.5 Multipole Expansion of Charge Distributions . . . . . . . . 45
2.3.6 Addition Theorem for Spherical Harmonics . . . . . . . . . 47
2.3.7 Multipole Expansion in Cartesian Coordinates . . . . . . . 48
ix
April 17, 2017 11:16 ws-book961x669 BC: 10514 - Advanced Classical Electrodynamics 3rd Read jentschura page x
x Advanced Classical Electrodynamics
2.3.8 Multipole Expansion in an External Field . . . . . . . . . . 50

2.4 Electrostatic Green Function and Cylindrical Coordinates . . . . . 52
2.4.1 Laplace Equation in Cylindrical Coordinates . . . . . . . . 52
2.4.2 Cylindrical Coordinates and Bessel Functions . . . . . . . . 53
2.4.3 Orthogonality Properties of Bessel Functions . . . . . . . . 56
2.4.4 Expansion of the Green Function in Cylindrical
Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.5 Electrostatic Green Function and Eigenfunction Expansions . . . . 63
2.5.1 Eigenfunction Expansions for the Green Function . . . . . 63
2.5.2 Application to Electrostatics in Spherical Coordinates . . . 65
2.6 Summary: Green Function of Electrostatics . . . . . . . . . . . . . . 67
2.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3. Paradigmatic Calculations in Electrostatics 71

3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.1.1 Differential Equations of Electrostatics . . . . . . . . . . . . 71
3.1.2 Boundary-Value Problems: One-Dimensional Analogy . . . 72
3.1.3 Green’s Theorem: Dirichlet and Neumann Green
Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.1.4 Boundary-Value Problems and Laplace Equation . . . . . . 76
3.2 Laplace Equation and Variational Calculations . . . . . . . . . . . . 78
3.2.1 Definition of the Functional Derivative . . . . . . . . . . . . 78
3.2.2 Variational Principle and Euler–Lagrange Equations . . . . 80
3.2.3 Variations with Constraints . . . . . . . . . . . . . . . . . . . 82
3.2.4 Second Functional Derivative and Hilbert Space . . . . . . 83
3.2.5 Variational Calculation of the Capacitance of Plates . . . . 84
3.2.6 Variational Calculation for Coaxial Cylinders . . . . . . . . 88
3.2.7 Exact Integration of the Capacitance of Coaxial Cylinders 91
3.3 Laplace Equation and Series Expansions . . . . . . . . . . . . . . . . 97
3.3.1 Coordinate Systems and Special Functions . . . . . . . . . . 97
3.3.2 Laplace Equation in a Rectangular Parallelepiped . . . . . 101
3.3.3 Laplace Equation in a Two-Dimensional Rectangle . . . . . 106
3.3.4 Boundary Conditions on the Finite Part of a Long Strip . 108
3.3.5 Cauchy’s Residue Theorem: A Small Digression . . . . . . 110
3.3.6 Boundary Conditions on the Infinite Part of a Long Strip . 114
3.3.7 Cylinders and Zeros of Bessel Functions . . . . . . . . . . . 119
3.4 Laplace Equation and Dirichlet Green Functions . . . . . . . . . . . 124
3.4.1 Dirichlet Green Function for Spherical Shells . . . . . . . . 124
3.4.2 Boundary Condition of Dipole Symmetry . . . . . . . . . . 125
3.4.3 Verification Using Series Expansion . . . . . . . . . . . . . . 126
3.5 Poisson Equation and Dirichlet Green Functions . . . . . . . . . . . 127
3.5.1 Source Terms with Boundary Conditions . . . . . . . . . . . 127
April 17, 2017 11:16 ws-book961x669 BC: 10514 - Advanced Classical Electrodynamics 3rd Read jentschura page xi
Contents xi
3.5.2 Sources and Fields in a Spherical Shell . . . . . . . . . . . . 127

3.5.3 Induced Charge Distributions on the Boundaries . . . . . . 130
3.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4. Green Functions of Electrodynamics 137

4.1 Green Function for the Wave Equation . . . . . . . . . . . . . . . . . 137
4.1.1 Integral Representation of the Green Function . . . . . . . 137
4.1.2 Why So Many Green Functions? . . . . . . . . . . . . . . . . 141
4.1.3 Retarded and Advanced Green Function . . . . . . . . . . . 142
4.1.4 Feynman Contour Green Function . . . . . . . . . . . . . . . 147
4.1.5 Summary: Green Functions of Electrodynamics . . . . . . . 153
4.2 Action-at-a-Distance and Coulomb Gauge . . . . . . . . . . . . . . . 154
4.2.1 Potentials and Sources . . . . . . . . . . . . . . . . . . . . . . 154
4.2.2 Cancellation of the Instantaneous Term . . . . . . . . . . . 156
4.2.3 Longitudinal Electric Field as a Retarded Integral . . . . . 158
4.2.4 Coulomb-Gauge Scalar Potential as a Retarded Integral . . 159
4.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
5. Paradigmatic Calculations in Electrodynamics 163

5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
5.1.1 General Considerations . . . . . . . . . . . . . . . . . . . . . 163
5.1.2 Wave Equation and Green Functions . . . . . . . . . . . . . 164
5.2 Helmholtz Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.2.1 Helmholtz Equation and Green Function . . . . . . . . . . . 165
5.2.2 Helmholtz Equation in Spherical Coordinates . . . . . . . . 167
5.2.3 Radiation Green Function . . . . . . . . . . . . . . . . . . . . 170
5.3 Localized Harmonically Oscillating Sources . . . . . . . . . . . . . . 174
5.3.1 Basic Formulas and Multipole Expansion . . . . . . . . . . 174
5.3.2 Asymptotic Limits of Dipole Radiation . . . . . . . . . . . . 177
5.3.3 Exact Expression for the Radiating Dipole . . . . . . . . . . 180
5.4 Tensor Green Function . . . . . . . . . . . . . . . . . . . . . . . . . . 181
5.4.1 Clebsch–Gordan Coefficients: Motivation . . . . . . . . . . . 181
5.4.2 Vector Additions and Vector Spherical Harmonics . . . . . 185
5.4.3 Scalar Helmholtz Green Function and Scalar Potential . . 189
5.4.4 Tensor Helmholtz Green Function and Vector Potential . . 189
5.5 Radiation and Angular Momenta . . . . . . . . . . . . . . . . . . . . 194
5.5.1 Radiated Electric and Magnetic Fields . . . . . . . . . . . . 194
5.5.2 Gauge Condition, Vector and Scalar Potentials . . . . . . . 195
5.5.3 Representations of the Vector Spherical Harmonics . . . . . 197
5.5.4 Poynting Vector of the Radiation . . . . . . . . . . . . . . . 199
5.5.5 Half-Wave Antenna . . . . . . . . . . . . . . . . . . . . . . . . 202
5.5.6 Long-Wavelength Limit of the Dipole Term . . . . . . . . . 208
April 17, 2017 11:16 ws-book961x669 BC: 10514 - Advanced Classical Electrodynamics 3rd Read jentschura page xii
xii Advanced Classical Electrodynamics
5.6 Potentials due to Moving Charges in Different Gauges . . . . . . . 209

5.6.1 Moving Charges and Lorenz Gauge . . . . . . . . . . . . . . 209
5.6.2 Liénard–Wiechert Potentials in Coulomb Gauge . . . . . . 211
5.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
6. Electrodynamics in Media 221

6.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
6.2 Microscopic and Macroscopic Equations . . . . . . . . . . . . . . . . 221
6.2.1 Macroscopic Equations and Measurements . . . . . . . . . . 221
6.2.2 Macroscopic Fields from Microscopic Properties . . . . . . 223
6.2.3 Macroscopic Averaging and Charge Density . . . . . . . . . 228
6.2.4 Macroscopic Averaging and Current Density . . . . . . . . 229
6.2.5 Phenomenological Maxwell Equations . . . . . . . . . . . . . 232
6.2.6 Parameters of the Multipole Expansion . . . . . . . . . . . . 233
6.3 Fourier Decomposition and Maxwell Equations in a Medium . . . . 234
6.4 Dielectric Permittivity: Various Examples . . . . . . . . . . . . . . . 238
6.4.1 Sellmeier Equation . . . . . . . . . . . . . . . . . . . . . . . . 238
6.4.2 Drude Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
6.4.3 Dielectric Permittivity and Atomic Polarizability . . . . . . 242
6.4.4 Dielectric Permittivity for Dense Materials . . . . . . . . . 244
6.5 Propagation of Plane Waves in a Medium . . . . . . . . . . . . . . . 247
6.5.1 Refractive Index and Group Velocity . . . . . . . . . . . . . 247
6.5.2 Wave Propagation and Method of Steepest Descent . . . . 249
6.6 Kramers–Kronig Relationships . . . . . . . . . . . . . . . . . . . . . . 254
6.6.1 Analyticity and the Kramers–Kronig Relationships . . . . . 254
6.6.2 Applications of the Kramers–Kronig Relationships . . . . . 256
6.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
7. Waveguides and Cavities 263

7.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
7.2 Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
7.2.1 General Formalism . . . . . . . . . . . . . . . . . . . . . . . . 264
7.2.2 Boundary Conditions at the Surface . . . . . . . . . . . . . 267
7.2.3 Modes in a Rectangular Waveguide . . . . . . . . . . . . . . 270
7.3 Resonant Cavities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
7.3.1 Resonant Cylindrical Cavities . . . . . . . . . . . . . . . . . 279
7.3.2 Resonant Rectangular Cavities . . . . . . . . . . . . . . . . . 287
7.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
8. Advanced Topics 295

8.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
8.2 Lorentz Transformations, Generators and Matrices . . . . . . . . . 297
April 17, 2017 11:16 ws-book961x669 BC: 10514 - Advanced Classical Electrodynamics 3rd Read jentschura page xiii
Contents xiii
8.2.1 Lorentz Boosts . . . . . . . . . . . . . . . . . . . . . . . . . . 297

8.2.2 Time Dilation and Lorentz Contraction . . . . . . . . . . . 298
8.2.3 Addition Theorem . . . . . . . . . . . . . . . . . . . . . . . . 299
8.2.4 Generators of the Lorentz Group . . . . . . . . . . . . . . . 300
8.2.5 Representations of the Lorentz Group . . . . . . . . . . . . 303
8.3 Relativistic Classical Field Theory . . . . . . . . . . . . . . . . . . . 309
8.3.1 Maxwell Tensor and Lorentz Transformations . . . . . . . . 309
8.3.2 Maxwell Stress–Energy Tensor . . . . . . . . . . . . . . . . . 313
8.3.3 Lorentz Transformation and Biot–Savart Law . . . . . . . . 315
8.3.4 Relativity and Magnetic Force . . . . . . . . . . . . . . . . . 316
8.3.5 Covariant Form of the Liénard–Wiechert Potentials . . . . 320
8.4 Towards Quantum Field Theory . . . . . . . . . . . . . . . . . . . . . 322
8.4.1 Casimir Effect and Quantum Electrodynamics . . . . . . . 322
8.4.2 Zero-Point Energy . . . . . . . . . . . . . . . . . . . . . . . . 324
8.4.3 Regularization and Renormalization . . . . . . . . . . . . . . 326
8.5 Classical Potentials and Renormalization . . . . . . . . . . . . . . . 331
8.5.1 Potential of a Uniformly Charged Plane . . . . . . . . . . . 331
8.5.2 Potential of a Uniformly Charged Long Wire . . . . . . . . 332
8.5.3 Charged Structures in 0.99 and 1.99 Dimensions . . . . . . 334
8.6 Open Problems in Classical Electromagnetic Theory . . . . . . . . 336
8.6.1 Abraham–Minkowski Controversy . . . . . . . . . . . . . . . 336
8.6.2 Relativistic Dynamics with Radiative Reaction . . . . . . . 338
8.6.3 Electrodynamics in General Relativity . . . . . . . . . . . . 342
8.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
Bibliography 351
Index 355
April 17, 2017 11:16 ws-book961x669 BC: 10514 - Advanced Classical Electrodynamics 3rd Read jentschura page 1
Chapter 1
Maxwell Equations
1.1 Basics
1.1.1 Integral and Differential Forms of the Maxwell Equations
There are T-shirts being sold which carry the enigmatic and deep writing: And God
said
1
⃗ ⋅ E⃗ =
∇ ρ, (1.1a)
0
⃗ ⋅B
∇ ⃗ = 0, (1.1b)
⃗ × E⃗ = − ∂ B
∇ ⃗, (1.1c)
∂t
⃗ ×B
∇ ⃗ = μ0 J⃗ + 1 ∂ E ⃗ (“T-shirt equations”) , (1.1d)
c2 ∂t
and there was light. God has spoken, but it generally takes a student a little while
to figure out and to interpret what He said. Still, it is instructive to consider the
“writing on the T-shirt” at the very beginning of this course. Please be reassured
that (i) the book is written while knowing the hardships of understanding electro-
dynamics very well, and that (ii) everyone needs to dwell on related questions for
quite some time before developing full mastery of the subject (hopefully). Classical
electrodynamics has a reputation for being a harder course than other courses in
the physics curriculum. The reason for this observation is that electrodynamics
combines, in quite a unique fashion, the intricacies of physics with those of mathe-
matics. Differential calculus in three and four dimensions, and mastery of Green
functions1 are prerequisites for an understanding of electrodynamics. In contrast
to analytic mechanics, quantum mechanics, relativistic quantum mechanics, and
thermodynamics, electrodynamics is a field theory (a classical field theory perhaps,
but still, a field theory).
We will not dwell on the philosophical question here regarding the success of the
description of nature by mathematics. Four observations, though, can be noted:
1
George Green (1793–1841)
1
2 Advanced Classical Electrodynamics
(i) Augustinus said that everyone who wishes to contribute to the understanding
of nature needs to be willing to overcome considerable personal and professional
obstacles to pursue his or her career; that was true then and it is still true today.
(ii) The discovery of laws of nature by means of mathematics was alternatingly seen
as an act of recognizing and interpreting the works of the almighty in this world
(especially, if science was carried through by pious monks); or the work of the devil
who is taking apart the mysteries of nature by evil means (especially, if science was
carried through by doubtful characters). The truth probably lies in the middle;
there is good science and bad science, in regards to both the methods used to gain
insight into natural phenomena as well as the consequences for mankind. We cannot
really separate the two. We may refer to the famous saga of the goat driven over
the newly constructed bridge at the Gotthard pass, his soul being sacrificed as the
first living being passing the newly established bridge. It takes some invention to
use the fruit of one’s quest of knowledge or achievement without facing the evil that
always surrounds us in the current world, whether we like it or not. (iii) Still, there
is nothing that has kept the curious characters of every generation from pursuing
science and the quest for knowledge, each one using their means. There is nothing
more gratifying and more adventurous than seeing the Laws of Nature unfold around
us. When these Laws are being explored by various techniques, such as high-
precision laser spectroscopy, or other means, we are happy to see the tiny wheels of
our imagination explain what we observe in the experiments. (iv) Finally, the self-
discipline required of any scientist serves, in the best possible sense of the word, as
a guide toward a deeper moral principle in all of us, and toward a factual approach
to the questions surrounding our life; it often solves problems that would otherwise
seem intractable.
It is useful, thus, to ponder the “T-shirt equations” (otherwise known as the
“Maxwell equations”) at the beginning of this book for two reasons: (i) in order to
clearly state that the subject of classical electrodynamics is nothing but the science
of solving the system of partial differential equations given on the above mentioned
T-shirt, and (ii) in order to illustrate right at the beginning how much physics
can be summarized by a set of compactly written differential equations. Let us
remember that electrodynamics covers the physics underlying the telephone as well
as personal computers, the servo-steering of cars and planes, the navigation using
GPS and radar, as well as radio and television. Only in the microworld, where
quantum fluctuations play a role, or in curved space-times, do we need to depart
from Eq. (1.1).
Indeed, electrodynamics is one of the subfields of physics where nothing can be
learned by any means except by constant practice. A famous proverb, probably
invented by some Englishman, states that “understanding is equal to getting used
to something”, but understanding electrodynamics is impossible without getting
used to it, and without constant practice. We recall the equations on the T-shirt,
Maxwell Equations 3
which are the Maxwell2 equations,
∇ ⃗ r , t) = 1 ρ(⃗
⃗ ⋅ E(⃗ r , t) , (1.2a)
0
∇ ⃗ r , t) = 0 ,
⃗ ⋅ B(⃗ (1.2b)
⃗ r , t) = − ∂ B(⃗
⃗ × E(⃗
∇ ⃗ r , t) , (1.2c)
∂t
⃗ × B(⃗
∇ ⃗ r , t) + 1 ∂ E(⃗
⃗ r , t) = μ0 J(⃗ ⃗ r , t) . (1.2d)
c2 ∂t
These equations are commonly referred to as Gauss’s law (1.2a), the law of the ab-
sence of magnetic monopoles (1.2b), Faraday’s law (1.2c), and the Ampere–Maxwell
law (1.2d) (we will consistently suppress the accent grave on the name Ampère in
this book). Here, metric (SI mksA) units are employed, E ⃗ is the electric field, B
⃗
is the magnetic induction field, μ0 is the vacuum permeability, 0 is the vacuum
permittivity, ρ is the charge density which measures the electric charge per test
volume, and ⃗j is the current density. In the SI mksA unit system, the vacuum
permittivity 0 and the vacuum permeability μ0 assume the numerical values
Vs 1
μ0 = 4π × 10−7 , 0 = . (1.3)
Am μ0 c 2
For the Maxwell equations, the use of SI mksA units implies that 0 and μ0 are
universally kept in all formulas, and that the mechanical units [meter (m), kilogram
(kg), and second (s)], are supplemented by the Ampere (A) as the unit of the current,
which in turn fixed the unit of charge to be 1 A × 1s = 1 C, with the Coulomb being
denoted as C. Often, in different subfields of physics, different unit systems are
used. E.g., in atomic units one would use 0 = 1/(4π), h ̵ = 1, and e2 = 1/c = α, where
α is the fine-structure constant, 0 is the vacuum permittivity, h ̵ is the reduced
Planck constant, and c is the speed of light. However, in natural units, one has
̵ = c = 0 = 1, and e2 = 4πα, where e is the elementary charge. Keeping all factors
h
of c, 0 (and h,̵ where applicable) in the formulas, we hope to arrive at a more
general treatment. Finally, the symbol ∇ ⃗ in Eq. (1.2) is the gradient operator. The
⃗ ⃗
quantity ∇ ⋅ E is the divergence of the electric field, and the quantity ∇ ⃗ × E⃗ is the
curl of the electric field. Also, ⋅ denotes the scalar product and
∂ ∂ ∂
⃗ = êx
∇ + êy + êz (1.4)
∂x ∂y ∂z
is the so-called gradient operator, where êx , êy , and êz are the unit vectors in
the x, y, and z directions. In writing Eq. (1.2), we have carefully observed that
all physical quantities in the Maxwell equations are functions of space and time.
Instead of Eq. (1.2), a reader may be more familiar with the integral form of the
2
James Clerk Maxwell (1831–1879)
Maxwell equations, which are familiar from undergraduate physics. These read
⃗ r, t) ⋅ dA⃗ = 1 ∫ ρ(⃗
∫ E(⃗ r, t) d3 r , (1.5a)
∂V 0 V
∫ ⃗ r, t) ⋅ dA⃗ = 0 ,
B(⃗ (1.5b)
∂V
⃗ s, t) ⋅ d⃗ ∂ ⃗ r, t) ⋅ dA⃗ ,
∮ E(⃗ s= − ∫ B(⃗ (1.5c)
∂A ∂t A
∮ ⃗ s, t) ⋅ d⃗
B(⃗ ⃗ r, t) ⋅ dA⃗ + 1 ∂ ∫ E(⃗
s = μ0 ∫ J(⃗ ⃗ r, t) ⋅ dA⃗ , (1.5d)
∂A A c2 ∂t A
with the same identifications as before (Gauss’s law, law of the absence of magnetic
monopoles, Faraday’s law, and the Ampere–Maxwell law). In these equations, V is
an arbitrary test volume, and ∂V is its surface, with the surface normal being ori-
ented outward. Also, A is an arbitrary (possibly curved) surface (a two-dimensional
manifold embedded in three-dimensional space), and the “surface” of the test area
A is denoted as ∂A. Specifically, ∂A is the closed loop (the curve) that constitutes
the outer edge of the area A and therefore is a one-dimensional manifold (line). The
sign convention is such that the following three vectors constitute a right-handed
system: (i) the line elements d⃗ s which are tangent to ∂A (vector 1), (ii) the vector
pointing from a line element to the inner region of the area (vector 2), (iii) and the
surface normal dA⃗ (vector 3). The quantity
r, t) d3 r = qencl (t)
∫ ρ(⃗ (1.6)
V
is the enclosed charge qencl (t) in the volume V , still a function of time, and the
quantity
⃗ r , t) ⋅ dA⃗ = Iarea (t)
∫ J(⃗ (1.7)
A
is the vector-summed total current Iarea (t) penetrating the area A as a function of
time.
1.1.2 Relation of the Differential to the Integral Form

We now investigate how to reconcile the differential form of the Maxwell equations
with the integral form, and we start with the first Maxwell equation. Indeed, we
start from the integral form Eq. (1.5a) of the first Maxwell equation,
⃗ r, t) ⋅ dA⃗ = 1 ∫ ρ(⃗
∫ E(⃗ r, t) d3 r . (1.8)
∂V 0 V
We consider a small, but not infinitesimally small, test volume V → Vδ , where
Vδ is the cubic volume extending over the intervals (x, x + δx), (y, y + δy), and
(z, z + δy), where we consider δx, δy, and δz as small quantities (see also Fig. 1.1).
The consideration of small test volume with side lengths δx, δy, and δz enables
us to perform a Taylor expansion, and to discard terms of higher order in the
displacements δx, δy, and δz.
Maxwell Equations 5
Let us consider the yz faces ∂Vδyz of the test volume Vδ in Fig. 1.1. The outward
surface normals are +êx and −êx . Sampling the electric field in the middle of the
test surface, the surface integral is approximated as
I≡∫ ⃗ r , t) ⋅ dA⃗ ≈ [Ex (x + δx, y + 1 δy, z + 1 δz, t)

E(⃗
∂Vδyz 2 2
− Ex (x, y + 12 δy, z + 12 δz, t)] δy δz . (1.9)
To first order in δx, we obtain

∂
I≈ Ex (x, y + 12 δy, z + 12 δz, t) δx δy δz . (1.10)
∂x
If we are interested only in the term proportional to δx δy δz, then we can replace
y + 12 δy → y, and z + 12 δz → z in the argument of the x component of the electric
field. Higher-order terms do not contribute to the order we are interested in. Thus,
∂
I= Ex (x, y, z, t) δx δy δz + higher-order terms . (1.11)
∂x
With this notion in mind, the left- and right-hand sides of Eq. (1.8) can then be
written as follows,
∫ ⃗ r , t) ⋅ dA⃗
E(⃗
∂Vδ
≈ [Ex (x + δx, y + 12 δy, z + 12 δz, t) − Ex (x, y + 12 δy, z + 12 δz, t)] δy δz

+ [Ey (x + 12 δx, y + δy, z + 12 δz, t) − Ey (x + 12 δx, y, z + 12 δz, t)] δx δz
+ [Ez (x + 12 δx, y + 12 δy, z + δz, t) − Ez (x + 12 δx, y + 12 δy, z, t)] δx δy , (1.12a)
and
1 1
r , t) d3 r ≈ ρ(x + 12 δx, y + 12 δy, z + 12 δz, t) δx δy δz .
∫ ρ(⃗ (1.12b)
0 Vδ 0
We now approximate as follows,
[Ex (x + δx, y + 12 δy, z + 12 δz, t) − Ex (x, y + 12 δy, z + 12 δz, t)]

∂
≈ [Ex (x + δx, y, z, t) − Ex (x, y, z, t)] ≈ Ex (x, y, z, t) δx , (1.13a)
∂x
[Ez (x + 12 δx, y + 12 δy, z + δz, t) − Ez (x + 12 δx, y + 12 δy, z, t)]
∂
≈ [Ey (x, y + δy, z, t) − Ey (x, y, z, t)] ≈ Ey (x, y, z, t) δy , (1.13b)
∂y
[Ez (x + 12 δx, y + 12 δy, z + δz, t) − Ez (x + 12 δx, y + 12 δy, z, t)]
∂
≈ [Ez (x, y, z + δz, t) − Ez (x, y, z, t)] ≈ Ez (x, y, z, t) δz . (1.13c)
∂z
Here, we ignore terms of higher order in the small quantities δx, δy and δz, which
would otherwise lead to terms quadratic in the small expansion parameters, such
as δx2 δy δz. Analogous approximations are made in Eq. (1.12b).
pp
ppp
pp
z + δz pp
pp
ppp
ppp
pp
pp
Test Volume Vδ pp
ppp
ppp
pppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppppp y + δy
ppppp
pppp ppppp
ppppp
z ppppp y
x x + δx
z
6 y

-x
Fig. 1.1 Test volume for the transformation of the first and second Maxwell equations
from integral to differential form.
Inserting Eq. (1.13) into (1.12), we see that the volume δx δy δz cancels, and so
∂ ∂ ∂ 1
Ex (x, y, z, t) + Ey (x, y, z, t) + Ez (x, y, z, t) = ρ(x, y, z, t) . (1.14)
∂x ∂y ∂z 0
This equation now needs to be transformed into vector form. For the position
vector, we write in this book
3
r⃗ = x êx + y êy + z êz = ∑ xi êi , (1.15)
i=1
identifying the components 1, 2, 3 with the x, y, z directions, respectively. The gra-

dient operator is a vector and reads, in Cartesian coordinates,
3
∂ ∂ ∂ ∂
⃗ = êx
∇ + êy + êz = ∑ êi . (1.16)
∂x ∂y ∂z i=1 ∂x i
We can thus write Eq. (1.14) as
⃗ r , t) = 1
⃗ ⋅ E(⃗
∇ ρ(⃗
r, t) , (1.17)
0
which is the differential form of the first Maxwell equation (1.2a). The transforma-
tion of the integral form of the second Maxwell equation (1.5b) into the differential
form (1.2b) is analogous to the above derivation for the first Maxwell equation.
We now turn our attention to the discussion of the third Maxwell equation (1.5c),
which involves the curl of the electric field. Its integral form is given as
⃗ s, t) ⋅ d⃗ ∂ ∂ ⃗ r, t) ⋅ dA⃗ ,
∮ E(⃗ s = − ΦB = − ∫ B(⃗ (1.18)
∂A ∂t ∂t A
Maxwell Equations 7
where ΦB is the magnetic flux. The closed-loop integral ∮∂A E(⃗ ⃗ s, t) ⋅ d⃗

s involves the
⃗
electric field E(⃗s, t) at the position s⃗ on the line integral. We assume that the test
area Azδ is oriented in the xy plane so that its surface normal is oriented along the
positive z axis. The test area extends over the interval (x, x + δx) in the x-direction
and over the interval (y, y + δy) in the y-direction, at an arbitrary z coordinate (see
Fig. 1.2).
We encircle the line integral about the infinitesimal test area Azδ → ∂Azδ in
the counterclockwise direction and pick up the contributions. It is permissible to
approximate the right-hand side of Eq. (1.18) as follows,
∮ ⃗ s, t) ⋅ d⃗
E(⃗ s ≈ Ex (x + 12 δx, y, z, t) δx + Ey (x + δx, y + 12 δy, z, t) δy
∂Azδ
− Ex (x + 12 δx, y + δy, z, t) δx − Ey (x, y + 12 δy, z, t) δy ,

(1.19a)
∂ ⃗ r , t) ⋅ dA⃗ ≈ − ∂ Bz (x + 1 δx, y + 1 δy, z, t) δx δy .
− ∫ B(⃗ 2 2
(1.19b)
∂t Aδ ∂t
By a Taylor expansion, keeping all terms up to second order in the small quantities
δx, δy, and δz, we obtain
∂ ∂ ∂
− Ex (x, y, z, t) δx δy + Ey (x, y, z, t) δx δy ≈ − Bz (x, y, z, t) δx δy . (1.20)
∂y ∂x ∂t
Third-order terms have been neglected on both sides of the equation. One may now
cancel the test area δx δy and write
∂ ∂ ⃗ z = − ∂ Bz ,
Ey − ⃗ × E)
Ex = (∇ (1.21)
∂x ∂y ∂t
where we suppress the space-time arguments of the fields. We identify ∂x ∂

Ey − ∂y
∂
Ex
as the z component of the curl of the electric field, thus obtaining the z-component
of Eq. (1.18). In general, the curl of a vector field is defined as
⎛ êx êy êz ⎞

⃗ × E = det ⎜ ⎟
⃗ ∂ ∂ ∂
∇ ⎜ ⎟. (1.22)
⎜ ∂x ∂y ∂z ⎟
⎝ Ex Ey Ez ⎠
With reference to Eq. (1.15) and (1.16), its components are given as
3 3
(∇ ⃗ i = ∑ ∑ ijk ∂ Ek ,
⃗ × E) i, j, k = 1, 2, 3 . (1.23)
j=1 k=1 ∂xj
Here, ijk is the totally antisymmetric Levi–Cività tensor, i.e., 123 = 1 and ijk is
equal to the sign of the permutation that transforms the triple (1, 2, 3) into (i, j, k).
If an even number of exchanges is required in order to transform (1, 2, 3) into (i, j, k),
then the sign of the permutation is 1, otherwise it is −1. So, 231 = 1, but 213 = −1.
Normal to the Area
ppppp
6
ppp
pp
ppp
pp
ppp
pp
pp
Test Area Aδ ppp y + δy
ppp
pp
ppp
pp
pp

z - y
x x + δx
z
6 y

-x
Fig. 1.2 Test area for the transformation of the third and fourth Maxwell equation from
integral to differential form.
If two of the indices are equal, such as in 122 = 0, the result is zero. We can also
write
3 3 3
∇ ⃗ = ∑ ∑ ∑ êi ijk ∂ Ek .
⃗ ×E (1.24)
i=1 j=1 k=1 ∂xj
In the above derivation leading to (1.21), we have assumed the test area ∂A is
oriented along the positive z axis. If we change the orientation of the test area Aδ ,
then we recover the x and y components of Faraday’s law
⃗ ×E
∇ ⃗ =−∂ B ⃗, (1.25)
∂t
where we suppress the space-time arguments of the electric and magnetic fields. The
transformation of the fourth Maxwell equation from integral to differential form is
analogous.
For a macroscopic volume V or area A, the validity of the integral and differ-
ential forms of the Maxwell equations can be verified based on a partition of the
macroscopic volumes and areas into small test volumes and areas, “glued” together
along their boundaries (colloquially speaking, interpreting the small test volumes
as “bricks in Legoland” from which the entire “Legoland” is being assembled).
One of the most paradigmatic applications of the formalism outlined above in-
volves the so-called continuity equation. The continuity equation is a simple state-
ment that says that the charge leaving a test volume V through currents can be
obtained in two alternative ways: (i) as the current density ⃗j summed over the
surface area of the test volume ∂V , and (ii) as the time derivative of the charge
density, summed/integrated over the test volume V . The integral form of the con-
tinuity equations is expressed as
∫ ⃗ r , t) ⋅ dA⃗ = − ∂ ∫ ρ(⃗
J(⃗ r , t) d3 r . (1.26)
∂V ∂t V
Maxwell Equations 9
Using the same test volume as given in Fig. 1.1, we obtain
[Jx (x + δx, y + 12 δy, z + 12 δz, t) − Jx (x, y + 12 δy, z + 12 δz, t)] δy δz

+ [Jy (x + 12 δx, y + δy, z + 12 δz, t) − Jy (x + 12 δx, y, z + 12 δz, t)] δx δz
+ [Jz (x + 12 δx, y + 12 δy, z + δz, t) − Jz (x + 12 δx, y + 12 δy, z, t)] δx δy
∂
=− ρ(x + 12 δx, y + 12 δy, z + 12 δz, t) δx δy δz . (1.27)
∂t
We have taken the differences at the midpoint of the edges of the test volume.
Observing that, to leading order, the midpoint can be replaced by the “lower corner”
of the test volume, one can replace the first term in Eq. (1.27) by Jx (x + δx, y, z, t) −
Jx (x, y, z, t). Then, expanding into a Taylor series and canceling the test volume
δx δy δz, we obtain
∂ ∂ ∂ ∂
Jx (x, y, z, t) + Jy (x, y, z, t) + Jz (x, y, z, t) = − ρ(x, y, z, t) . (1.28)
∂x ∂y ∂z ∂t
Finally, the differential form of the continuity equation is obtained,
∂
⃗ ⋅ J⃗ + ρ = 0 ,
∇ (1.29)
∂t
where again we suppress the space-time arguments of the current and charge density.
1.1.3 Dirac δ Function
Some readers of this book may be familiar with the Dirac-δ function based on other
sources, but it is still instructive to recall some basic facts. One may define the
(one-dimensional) Dirac delta function by the condition
b f (x) (a < x < b)
∫ f (x′ ) δ (x − x′ ) dx′ = { . (1.30)
a 0 (x < a or x > b)
Essentially, the Dirac-δ function extracts the value of the test function f at the
point x. If x = a or x = b, the integral is not well defined, but a common definition
is
b
∫ f (x′ ) δ (x − x′ ) dx′ = 1
2
f (x) for x=a or x = b. (1.31)
a
This definition is inspired by the fact that any representation of the Dirac-δ function
which is symmetric about the center yields the result (1.31), as explained in the
following.
The Dirac-δ function (or, more precisely, the Dirac-δ distribution) is very useful
in physics. A possible representation of the Dirac-δ function is
1 η
δ(ω) → Δ(η, ω) = , η → 0+ . (1.32)
π ω2 + η2
Fig. 1.3 Plot of the function Δ(η, ω) defined in Eq. (1.32), for small η and ω. The
emergence of the peak near ω = 0 is clearly discernible.
The limit is η → 0 is approached nonuniformly. The Δ function has the properties

∞
∫ dω Δ(η, ω) = 1 , (1.33a)
−∞
lim Δ(η, ω) = 0 (for ω ≠ 0), (1.33b)
η→0+
lim Δ(η, ω) = ∞ (for ω = 0), (1.33c)

η→0+
∞ ∞
∫ dω [ lim+ Δ(η, ω)] f (ω) = 0 , lim+ ∫ dω Δ(η, ω) f (ω) = f (0) . (1.33d)
−∞ η→0 η→0 −∞
The last two properties illustrate that one cannot take the limit η → 0 too early,
in the sense of non-uniform convergence. A graphical illustration can be found in
Fig. 1.3. In the limit η → 0+ , then, Δ(η, ω) → δ(ω), but the limit needs to be taken
after all integrals have been performed.
If the argument of a Dirac delta function is a function of x,
n −1
df
δ (f (x)) = ∑ (∣ ∣ ) δ(x − xi ) , (1.34)
i=1 dx x=xi
where the xi (i = 1, . . . , n) with g(xi ) = 0 denote simple zeros of g(x). The three-
dimensional delta function is equal to the product of three one-dimensional Dirac
delta functions,
δ (3) (⃗
r − r⃗′ ) = δ(x − x′ ) δ(y − y ′ ) δ(z − z ′ ) . (1.35)
Thus,
′ ′ ′
r′ ) δ (3) (⃗
∫ dx dy dz f (⃗ r − r⃗′ ) = f (⃗
r) (1.36)
V
if r⃗ is interior to the volume V and 0 otherwise. For multiple integrals such as
∫V (over the three-dimensional volume V ), we here follow the convention that
the integration domain is being indicated by a subscript of the integral sign. In
Maxwell Equations 11
particular, the integration domain then automatically defines the dimensionality of

the integral.
1.1.4 Green Function of the Laplacian

The Dirac-δ function enters the defining equation of the Green function correspond-
ing to the Laplacian operator,
⃗ 2 g(⃗
∇ ⃗ 2 g(⃗
r, r⃗′ ) = ∇ r − r⃗′ ) = δ (3) (⃗
r − r⃗′ ) , (1.37)
⃗ 2 is the Laplace operator, which acts on r⃗ (not r⃗′ ) unless stated otherwise.
where ∇
The desired solution is
1
r , r⃗′ ) = g(⃗
g(⃗ r − r⃗′ ) = − . (1.38)
r − r⃗′ ∣
4π ∣⃗
This result may be shown as follows. We recall Gauss’s theorem
∫ V⃗ (⃗ ⃗ ⋅ V⃗ (⃗
r ) ⋅ dA⃗ = ∫ ∇ r ) d3 r , (1.39)
∂V V
where V is an arbitrary volume and ∂V is its surface. We apply the theorem to the
vector field
1 r⃗ − r⃗′
V⃗ (⃗ ⃗
r) = ∇ = − , (1.40)
r − r⃗′ ∣
∣⃗ r − r⃗′ ∣3
∣⃗
and choose V as a sphere Sε of radius ε about the point r⃗′ . The surface of Sε is
denoted ∂Sε . Then,
1 ⃗ ⋅ V⃗ (⃗
∫ ⃗2 (
∇ ) d3 r = ∫ ∇ r ) d3 r = ∫ V⃗ (⃗
r) ⋅ dA⃗
Sε r − r⃗′ ∣
∣⃗ Sε ∂Sε
r⃗ − r⃗′ r⃗ − r⃗′
=∫ (− ) ⋅ r − r⃗′ ∣2 dΩ = −4 π .
∣⃗ (1.41)
Sε r − r⃗′ ∣3 ∣⃗
∣⃗ r − r⃗′ ∣
Because we can choose ε to be arbitrarily small, we conclude that
⃗2 ( 1
∇ ) = −4 π δ (3) (⃗
r − r⃗′ ) . (1.42)
r − r⃗′ ∣
∣⃗
and hence Eq. (1.37) is fulfilled. Because the Fourier3 transform (see Chaps. 4 and
6) of the Dirac-δ distribution is just a constant, the function G can be interpreted
as the inverse of the Laplacian operator.
1.2 Potentials and Gauges

1.2.1 Transverse and Longitudinal Components of a Vector Field
At any given point in time t, the electric and magnetic fields E(⃗ ⃗ r , t) and B(⃗
⃗ r , t)
represent vector fields. Central to the upcoming discussion is the fact that any
vector field can be separated into transverse and longitudinal components; we here
derive explicit formulas for this separation. Thus, let J⃗ (⃗
r , t) be a vector field which
3
Jean-Baptiste Joseph Fourier (1768–1830)
we would like to separate into transverse and longitudinal components. We seek a

separation
V⃗ (⃗
r , t) = V⃗⊥ (⃗
r, t) + V⃗∥ (⃗
r , t) , ⃗ ⋅ V⃗⊥ (⃗
∇ r , t) = 0 , ⃗ × V⃗∥ (⃗
∇ r , t) = 0 . (1.43)
It is not obvious how to do this. We first write a seemingly innocent vector identity,
∇ ⃗ × V⃗ (⃗
⃗ × (∇ r , t)) = ∇ ⃗ ⋅ V⃗ (⃗
⃗ (∇ ⃗ 2 V⃗ (⃗
r , t)) − ∇ r, t) . (1.44)
One may derive this identity on the basis of the Levi-Cività tensor ijk and the
identity
ijk km = δi δjm − δim δj . (1.45)
4
Here, the Einstein summation convention for repeated subscripts has been used.
(This convention implies that ai ai ≡ ∑ni=1 ai ai for repeated superscripts or sub-
scripts, with an applicable upper limit n for the summation which is derived from the
context in which the equation is written. In the current case, for three-dimensional
space, i, j, k, and m may assume values of 1, 2, 3; because k is being repeated, the
sum ∑3k=1 is implicitly understood in Eq. (1.45).)
We now rewrite Eq. (1.44),
⃗ 2 V⃗ (⃗
∇ r , t) = ∇ ⃗ ⋅ V⃗ (⃗
⃗ (∇ r, t)) − ∇ ⃗ × V⃗ (⃗
⃗ × (∇ r , t)) . (1.46)
We now appeal to Eqs. (1.36) and (1.37) for the definition of the Green function
r − r⃗′ ). If we assume that boundary contributions at infinity will vanish, then
g = g (⃗
we obtain (the second step involves a double partial integration)
V⃗ (⃗
r, t) = = ∫ d3 r′ ∇ r , r⃗′ ) V⃗ (⃗
⃗ 2 g (⃗ r′ , t) = ∫ d3 r′ g (⃗ ⃗ 2 V⃗ (⃗
r, r⃗′ ) ∇ r′ , t)
1 1
= − 3 ′
∫ d r [∇ ⃗ ′ ⋅ V⃗ (⃗
⃗ ′ (∇ ⃗ ′ × (∇
r′ , t)) − ∇ ⃗ ′ × V⃗ (⃗
r′ , t))]
4π r − r⃗′ ∣
∣⃗
1 ⃗ ′ (∇
∇ ⃗ ′ ⋅ V⃗ (⃗r′ , t)) 1 ⃗ ′ × (∇
∇ ⃗ ′ × V⃗ (⃗ r′ , t))
3 ′ 3 ′
= − +
4π ∫ 4π ∫
d r d r
r − r⃗′ ∣
∣⃗ r − r⃗′ ∣
∣⃗
1 1
= ∫ ⃗ ′ ⋅ V⃗ (⃗
d3 r′ (∇ r′ , t)) ∇⃗′
4π r − r⃗′ ∣
∣⃗
1 1
− ⃗′
d3 r′ ∇ × (∇⃗ ′ × V⃗ (⃗ r′ , t))
4π ∫ r − r⃗′ ∣
∣⃗
1 ⃗ ′ ⋅ V⃗ (⃗
∇ r′ , t) 3 ′ 1 ⃗ ′ × V⃗ (⃗
∇ r′ , t) 3 ′
= − ⃗∫
∇ d r + ⃗ ×∫
∇ d r . (1.47)
4π r − r⃗′ ∣
∣⃗ 4π r − r⃗′ ∣
∣⃗
Thus, the longitudinal and transverse parts of the current density, denoted as V⃗∥
and V⃗⊥ , respectively, are given as
1 ⃗ ⃗ ′ ⋅ V⃗ (⃗
∇ r′ , t)
V⃗∥ (⃗
r , t) = − ∇ ∫ d3 r′ , (1.48a)
4π ∣⃗
r − r⃗′ ∣
1 ⃗ ′ × V⃗ (⃗
∇ r′ , t)
V⃗⊥ (⃗
r , t) = ⃗ × ∫ d3 r′
∇ . (1.48b)
4π r − r⃗′ ∣
∣⃗
4
Albert Einstein (1879–1955)
These are highly non-local functions of the full current density V⃗ (⃗ r, t), in the sense
that in order to define V⃗∥ (⃗r, t) and V⃗⊥ (⃗
r , t), we have to sample values of the original
vector potential V⃗ (⃗
r , t) over wide areas of space.
Furthermore, if we define the scalar potential Ψ and the vector potential A⃗ as
1 ⃗ ′ ⃗ r′ , t)
3 ′ ∇ ⋅ V (⃗
Ψ (⃗
r , t) = − ∫ d r , (1.49a)
4π r − r⃗′ ∣
∣⃗
1 ⃗ ′ ⃗ r′ , t)
3 ′ ∇ × V (⃗
A⃗ (⃗
r , t) = ∫ d r , (1.49b)
4π r − r⃗′ ∣
∣⃗
then
V⃗∥ (⃗ ⃗ (⃗
r , t) = ∇Ψ r, t) , V⃗⊥ (⃗ ⃗ × A⃗ (⃗
r , t) = ∇ r, t) . (1.50)
Given a vector potential V⃗ , we thus have the explicit formulas that tell us how to
⃗ These potentials enable
calculate the scalar potential Ψ and the vector potential A.
us to write the longitudinal component of the vector field V⃗∥ as the gradient of the
scalar potential Ψ and the transverse component of the vector field V⃗⊥ as the curl
⃗
of a vector potential A.
1.2.2 Vector and Scalar Potentials

The contents of the current discussion are merely restricted to the definition of the
scalar and vector potentials, but their significance reaches far beyond, namely, to
quantum electrodynamics, which is a gauge theory, and even to the construction of
the standard model of particle interactions. Without the scalar and vector poten-
tials, and the concept of a covariant derivation, we would not have the tools at our
hand to analyze the gauge theories that define the standard model.
A detour with an illustrative example, regarding the action of the covariant
derivative operator on a quantum mechanical wave function, is thus in order. We
restrict the discussion to a time-independent configuration. Let us define a covariant
derivative operator Π ⃗ as follows,
⃗ ≡ p⃗ − e A(⃗
Π ⃗ r) , ̵∇
p⃗ = −ih ⃗, (1.51)
where e is the charge of the particle in question, A(⃗⃗ r) is the vector potential
(assumed here to be time-independent), h ̵ is Planck’s constant, and p⃗ is the mo-
mentum operator (in quantum mechanics), or the mechanical momentum of the
particle (in classical mechanics). The precise meaning of A⃗ will be clarified be-
low. The decisive property of the gauge potential is this: Let us assume that the
covariant derivative operator Π⃗ acts on a wave function ψ(⃗ r ). The addition of a
gradient
⃗ r) → A(⃗
A(⃗ ⃗ r) + ∇Λ(⃗
⃗ r) (1.52)
is called a gauge transformation. Gauge covariance now means the following: If we

simultaneously transform the wave function and the vector potential as follows,
i e r⃗ ⃗
r ) → ψ ′ (⃗
ψ(⃗ r) ≡ exp ( ̵ ∫ Λ(⃗ s) ⋅ ds) ψ(⃗
r) = U (⃗
r) ψ(⃗
r) , (1.53a)
h ⃗0
⃗ r ) → A⃗′ (⃗
A(⃗ ⃗ r ) + ∇Λ(⃗
r ) ≡ A(⃗ ⃗ r) , (1.53b)
then it is easy to show that

i e r⃗ ⃗ ⃗ r)) ψ(⃗
exp ( ̵ ∫ Λ(⃗ s) ⋅ ds) (⃗
p − e A(⃗ r)
h 0⃗
⃗ r) − e∇Λ(⃗ i e r⃗ ⃗
= (⃗
p − e A(⃗ ⃗ r)) exp ( ∫ Λ(⃗ s) ⋅ ds) ψ(⃗
r) . (1.54)
̵ 0⃗
h
When we define the prime to indicate the gauge transformed quantity, then this
identity takes a particularly sensible form,
r) (⃗
U (⃗ ⃗ r)) ψ(⃗
p − e A(⃗ p − e A⃗′ (⃗
r ) = (⃗ r )) U (⃗
r ) ψ(⃗
r) , (1.55a)
′
((⃗ ⃗ r )) ψ(⃗
p − e A(⃗ p − e A⃗′ (⃗
r)) = (⃗ r )) ψ ′ (⃗
r) , (1.55b)
and simply says that the gauge transform of the covariant derivative is equal to
the covariant derivative of the gauge transformed wave function. We shall see that
different vector potentials A⃗ produce equivalent observable electromagnetic field
configurations. In differential geometry [1–4], the vector potential A⃗ has the mean-
ing of a connection 1-form. Roughly speaking it measures how space is “bent”
for the charged particle in the presence of the vector potential, just like a light
trajectory is “bent” alongside a geodesic in the theory of general relativity when
traveling near heavy stars. Or, in differential geometry, the “bending” of the tra-
jectory is due to the curvature of the embedding manifold on which the particle
is moving. There is thus a deep philosophical notion behind the definition of the
covariant derivative: namely, that we can view the minimal coupling of a charged
particle wave function to the field as a consequence of the “bending” of space in the
presence of vector fields, for a particle that happens to carry charge, in such a way
that gauge freedom exists. In differential geometry, a gauge transformation simply
corresponds to a change in the coordinate system that we are using to describe
the curved manifold. In electrodynamics, a gauge transformation corresponds to
a local phase change of the wave function, leaving the probability density ∣ψ(⃗ r )∣2
unchanged, under a simultaneous change of the scalar and vector potentials, which
leave the physically observable fields unchanged. A gauge transformation thus is
a “coordinate transformation” involving the phases of the wave function, and the
scalar and vector potentials. The gauge freedom corresponds to the freedom of
choice of the coordinate system in differential geometry. Incidentally, the gauge
freedom is central to the construction of the current standard model of particle in-
⃗ r ) by a field that carries
teractions. In non-Abelian gauge theories, one replaces A(⃗
an additional (“color”) index, leading to a matrix-valued gauge transform factor.
These remarks are intended to motivate the central character of gauge transforma-
tions, and their significance within theoretical physics; their understanding is not
required for the following derivation.
Having illustrated the significance of scalar and vector potentials for quantum
theory and theoretical physics in general, let us finish the detour and return to
the case of classical electrodynamics. The gauge freedom allows us to define vector
and scalar potentials, and to choose a particular gauge that makes the calculation
easy for a particular source configuration (current and charges). Still, a lot can
be learned from gauge transformations on the classical level, and we shall briefly
dwell on related questions. In the following, we thus restrict our investigation to
the microscopic form of Maxwell’s equations. We start with the two source free
equations. First, we consider Gauss’ law for the magnetic induction, ∇ ⃗ ⋅B⃗ = 0. In
this case, according to the discussion in the Sec. 1.1.3, a vector potential A⃗ can be
defined so that the magnetic induction vector is given by
⃗ (⃗
B ⃗ × A⃗ (⃗
r , t) = ∇ r , t) . (1.56)
⃗ ×E
The vector potential is used in Faraday’s law ∇ ⃗ = 0 to yield
⃗ + ∂t B
⃗ (⃗ ∂ ⃗ ⃗ ⃗ (⃗ ∂
⃗ ×E
∇ r , t) + [∇ × A (⃗ ⃗ × [E
r , t)] = ∇ r , t) + A⃗ (⃗
r , t)] = ⃗
0. (1.57)
∂t ∂t
Note that, on the right-hand side, we use the zero vector 0⃗ instead of a simple
zero in order to be fully consistent. It follows that the curl of the vector field
r, t) + ∂t A⃗ (⃗
E⃗ (⃗ r , t) vanishes, which is why we can find a scalar potential Φ (⃗ r , t)
such that
⃗ (⃗ ∂
E r , t) + A⃗ (⃗ ⃗ (⃗
r, t) = −∇Φ r , t) (1.58)
∂t
and
⃗ (⃗ ∂
E ⃗ (⃗
r , t) = −∇Φ r, t) − A⃗ (⃗ r , t) , (1.59)
∂t
which expresses the electric field in terms of the scalar and vector potentials. The
fields constructed according to Eqs. (1.56) and (1.59) automatically fulfill the homo-
geneous Maxwell equations ∇⋅ ⃗ B ⃗ = 0 and ∇×
⃗ E ⃗=⃗
⃗ +∂t B 0. Given a field configuration
⃗ ⃗
B = B(⃗ r , t), we can obtain the vector potential A(⃗ ⃗ r , t) by the integral (1.49b), set-
ting V⃗ = B, ⃗ and the scalar potential Φ by Eq. (1.49a), setting V⃗ = −E ⃗ − ∂t A.
⃗
Gauss’s law for the electric field and the Ampere–Maxwell law couple the vector
and scalar potentials to their sources, which are the charge and current densities.
First we consider Gauss’ law and obtain
⃗ ⋅E⃗ = −∇⃗ 2 Φ (⃗ ∂ ⃗ ⃗ 1
∇ r , t) − ∇ r , t) = ρ (⃗
⋅ A (⃗ r , t) , (1.60)
∂t 0
while the Ampere–Maxwell law
⃗ (⃗ 1 ∂ ⃗
⃗ ×B
∇ r , t) = ∇ ⃗ × A⃗ (⃗
⃗ × [∇ r , t)] = 2 E r , t) + μ0 J⃗ (⃗
(⃗ r, t) (1.61)
c ∂t
becomes
1 ∂ ∂ ⃗
⃗ × [∇
∇ ⃗ × A⃗ (⃗
r , t)] + 2 [ A (⃗ r , t) + ∇Φ r , t)] = μ0 J⃗ (⃗
⃗ (⃗ r, t)
c ∂t ∂t
1 ∂ 1 ∂2
⇔ ∇ ⃗ ⋅ A⃗ (⃗
⃗ {∇ r , t) + 2 Φ (⃗ ⃗ 2 A⃗ (⃗
r , t)} − ∇ r , t) + 2 2 A⃗ (⃗ r, t) = μ0 J⃗ (⃗r, t) . (1.62)
c ∂t c ∂t
From Eqs. (1.60) and (1.62), we can conclude that the scalar and vector potentials
are coupled to the sources as follows, in a general gauge,
∂ 1
⃗ 2 Φ (⃗
−∇ r , t) − ⃗ ⋅ A⃗ (⃗
∇ r , t) = ρ (⃗ r, t) , (1.63a)
∂t 0
1 ∂ 1 ∂2
∇ ⃗ ⋅ A⃗ (⃗
⃗ {∇ r , t) + 2 Φ (⃗ ⃗ 2 A⃗ (⃗
r , t)} − ∇ r , t) + 2 2 A⃗ (⃗ r , t) = μ0 J⃗ (⃗
r, t) . (1.63b)
c ∂t c ∂t
These coupled second-order partial differential equations relate the potentials Φ and
A⃗ to the sources ρ and J.
⃗
1.2.3 Lorenz Gauge

We should precede the discussion of the Lorenz gauge with a brief remark on the
difference in the naming conventions between the Lorentz force and Lorentz trans-
formation on the one hand, and the Lorenz gauge on the other hand. Indeed, the
Lorentz force and the Lorentz transformation are associated with Lorentz5 in con-
trast to the Lorenz gauge, which is due to Lorenz.6 The two names are similar
but not the same.
At the beginning of the preceding Sec. 1.2.2, we briefly discussed the paradigm
of a gauge transformation, which is a local transformation of the vector potentials
⃗ r, t) → A(⃗
A(⃗ ⃗ r , t) + ∇Λ(⃗
⃗ r, t) (now, we assume a time-dependent field configuration).
Another interpretation, which is somewhat global, for a gauge transformation is as
follows: If one measures lengths or heights, one always has the freedom to choose
the zero point of the measurement. If one is to hang a picture on a wall, a conve-
nient zero point for the elevation measurement is the floor of the room, while, for
measuring the altitude of a mountain, a convenient reference point is the sea level.
A trivial gauge transformation which leaves the electric field invariant is simply
Φ(⃗r , t) → Φ(⃗
r , t) + C, where C is a constant potential, which of course does not affect
the gradient (“global” gauge transformation). This amounts to choosing a zero for
the measurement of the “altitude” of the potential. The local character of the gauge
transformation implies that we are free to choose the zeros for the measurements
of the potentials any way we want, and moreover, to choose these zero points dif-
ferently at every space-time point, provided the fields E(⃗ ⃗ r , t) and B(⃗
⃗ r , t) are left
invariant. This would amount to measuring (in absolute terms) the elevation of the
picture on the wall differently today and tomorrow, provided the difference of the
elevation measurements is left invariant. Let us now explore this concept in detail.
5
Hendrik Antoon Lorentz (1853–1928)
6
Ludvig Valentin Lorenz (1829–1891)
First, we observe that the addition of a term A⃗ (⃗

r , t) → A⃗ (⃗ ⃗ (⃗
r , t) + ∇Λ r , t) to the
vector potential does not change the magnetic induction field, because
⃗ × ∇Λ
∇ r , t) = ⃗
⃗ (⃗ 0. (1.64)
In this case we obtain the gauge transformed vector potential A (⃗
r , t), ⃗′
A⃗′ (⃗
r , t) = A⃗ (⃗ ⃗ (⃗
r, t) + ∇Λ r , t) . (1.65)
We now insert the equation A⃗ (⃗ r, t) = A⃗′ (⃗ ⃗ (⃗
r , t) − ∇Λ r , t) for the gauge trans-
formed vector potential into the formula Eq. (1.58) for the electric field,
⃗ (⃗ ∂ ∂
E r , t) = − A⃗′ (⃗
r, t) + ∇Λ ⃗ (⃗ ⃗ (⃗
r , t) − ∇Φ r , t)
∂t ∂t
∂ ∂
= − A⃗′ (⃗ ⃗ {Φ (⃗
r, t) − ∇ r, t) − Λ (⃗ r, t)} . (1.66)
∂t ∂t
Changing the vector potential will change the electric field unless the scalar potential
is also changed to the term that appears in curly brackets, which amounts to gauge
transform of the scalar potential,
∂
Φ′ (⃗ Λ (⃗
r , t) = Φ (⃗
r, t) −
r, t) . (1.67)
∂t
The electric field remains unchanged under the gauge transformation,
⃗ (⃗ ∂ ∂
E r, t) = − A⃗′ (⃗
r , t) − ∇Φ r , t) = − A⃗ (⃗
⃗ ′ (⃗ ⃗ (⃗
r , t) − ∇Φ r, t) . (1.68)
∂t ∂t
Equations (1.65) and (1.67) can be summarized as the gauge transform of the vector
and scalar potentials,
∂
A⃗′ (⃗
r , t) = A⃗ (⃗ ⃗ (⃗
r , t) + ∇Λ r, t) , Φ′ (⃗ Λ (⃗
r , t) = Φ (⃗
r , t) −
r , t) . (1.69)
∂t
The electric and magnetic fields, which are invariant under the gauge transforma-
tions, are said to be gauge invariant. In fact, they have to be gauge invariant as
they constitute physically observable field strengths.
We are now free to choose Λ so that the gauge transformed scalar and vector
potentials fulfill certain conditions. One possible requirement on the new potentials
is the Lorenz gauge condition,
1 ∂ ′
⃗ ⋅ A⃗′ (⃗
∇ r , t) + 2 Φ (⃗
r , t) = 0 . (1.70)
c ∂t
One may ask how, given A⃗ and Φ, Λ can be constructed to make the gauge trans-
formed vector and scalar potentials fulfill the gauge condition (1.70). The answer
is as follows. Given potentials A⃗ and Φ which do not satisfy the Lorenz gauge
condition, we have for the gauge-transformed potentials,
1 ∂ ′
⃗ ⋅ A⃗′ (⃗
∇ r , t) + 2 Φ (⃗ ⃗ ⋅ [A⃗ (⃗
r , t) = ∇ ⃗ (⃗
r , t) + ∇Λ r , t)]
c ∂t
1 ∂ ∂
+ 2 [Φ (⃗ r , t) − Λ (⃗ r, t)] = 0 , (1.71)
c ∂t ∂t
or
1 ∂ 1 ∂2
⃗ ⋅ A⃗ (⃗
∇ ⃗ 2 Λ (⃗
r , t) + ∇ r , t) + 2 Φ (⃗
r , t) − 2 2 Λ (⃗
r , t) = 0 . (1.72)
c ∂t c ∂t
Bringing now all terms with Λ to one side of the equation, we obtain a condition on
the gauge transform function Λ which mediates the transition to the Lorenz gauge,
1 ∂2 1 ∂
⃗2 −
(∇ ) Λ (⃗ ⃗ ⋅ A⃗ (⃗
r , t) = −∇ r , t) − 2 Φ (⃗
r, t) . (1.73)
c2 ∂t2 c ∂t
The gauge function Λ is required to satisfy the wave equation with the source
determined by the original potentials, i.e., by Φ and A⃗ (the ones before the gauge
transform). We shall see later that the differential operator (∇ ⃗ 2 − c−2 ∂t2 ) can be
inverted in much the same right as the operator ∇ ⃗ . This implies that it is always
2
possible to find an adequate solution Λ to Eq. (1.73), given Φ and A. ⃗ In other

words, we do not exclude any important solutions to the Maxwell equations if we
impose the Lorenz gauge condition.
A remark is in order: If the vector and scalar potentials A⃗ and Φ were to
fulfill the Lorenz condition in the first place, then the right-hand side of Eq. (1.73)
would vanish. The gauge transform function Λ has to fulfill an inhomogeneous wave
equation in which the source term is just equivalent to the “left-hand side of the
gauge condition Eq. (1.70)”. That means that if, accidentally, A⃗ and Φ were to fulfill
the Lorenz condition even before being gauge transformed, then we could choose for
Λ any function that fulfills the homogeneous wave equation, and still retain fields
that fulfill the Lorenz condition, after the gauge transformation. This illustrates
that the gauge condition itself does not yet determine the potentials uniquely.
Let us now couple the gauge transformed potentials to their sources. We recall
that according to Eqs. (1.63), Gauss’s law and the Ampere–Maxwell law relate the
sources to the potentials. In terms of the primed, gauge-transformed quantities, we
have
∂ 1
−∇⃗ 2 Φ′ (⃗
r , t) − ∇ ⃗ ⋅ A⃗′ (⃗
r , t) = ρ (⃗r , t) (1.74a)
∂t 0
for the coupling to the charge density, and
⃗ ⋅ A⃗′ (⃗ 1 ∂ ′ ⃗ 2 A⃗′ (⃗ 1 ∂2
⃗ {∇
∇ r, t) + 2 Φ (⃗ r , t)} −∇ r , t) + 2 2 A⃗′ (⃗ ⃗ r, t)
r, t) = μ0 J(⃗ (1.74b)
c ∂t c ∂t
,-- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -. - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -/
=0
for the coupling to the current density. If the primed potentials fulfill the Lorenz
gauge condition (1.70),
1 ∂ ′
⃗ ⋅ A⃗′ (⃗
∇ r , t) + 2 Φ (⃗
r , t) = 0 , (1.75)
c ∂t
the equations are uncoupled. For the second equation, this is immediately obvious,
and for the first equation, this becomes obvious as follows,
∂ 1 ∂ ∂ 1
⃗ 2 Φ′ (⃗
−∇ r, t) − ⃗ ⋅ A⃗′ (⃗
∇ ⃗ 2 Φ′ (⃗
r , t) = −∇ r , t) + 2 ( Φ) = ρ (⃗
r, t) . (1.76)
∂t c ∂t ∂t 0
The equations thus take the form of uncoupled, inhomogeneous wave equations, in
Lorenz gauge,
1 ∂2 1
( ⃗ 2 ) Φ′ (⃗
−∇ r , t) = ρ (⃗
r, t) , (1.77a)
c2 ∂t2 0
1 ∂2
( ⃗ 2 ) A⃗′ (⃗
−∇ r , t) = μ0 J⃗ (⃗
r , t) . (1.77b)
c2 ∂t2
Once these are solved, we can compute the electric and magnetic fields according
to Eqs. (1.59) and (1.56). In a source-free region of space-time, the scalar and
vector potentials thus satisfy homogeneous wave equations. This finding suggests
the existence of electromagnetic waves.
Let us add a remark on the relativistic formulation. It is customary to define
the “quabla” operator as
1 ∂2
◻= ⃗2 .
−∇ (1.78)
c2 ∂t2
⃗ which summarizes the scalar
Furthermore, the 4-vector potential is Aμ = (Φ, c A),
potential and the vector potential into a Lorentz-covariant 4-vector, denoted by a
Greek superscript μ. This means that if we have a Lorentz transformation that
transforms the space-time coordinates xμ = (c t, r⃗) according to
x′μ = Lμν xν (1.79)
then the 4-vector potentials, as seen from the moving observer, need to be trans-
formed in the same way,
A′μ = Lμν Aν , (1.80)
which clarifies the combined transformation properties of scalar and vector potential
under Lorentz transformations. Note that here, the primed quantities refer to those
seen in a different Lorentz frame (not the gauge transformed quantities). Then, in
̵ = c = 0 = 1 (Heaviside–Lorentz units, or just “natural units”), the two
units with h
equations (1.77a) and (1.77b) can be summarized into one equation, which takes a
particularly simple form, namely
1 μ
◻Aμ = J , (1.81)
0
where the components of the four-vector potential Aμ and the four-current density
J μ are
⃗ ,
Aμ = (Φ, cA) ⃗ .
J μ = (ρ, c−1 J) (1.82)
Greek indices μ, ν = 0, 1, 2, 3 describe space-time coordinates. Choosing potentials
which satisfy the Lorentz condition not only uncouples the equations for the po-
tentials but also yields equations which are invariant under the Lorentz/Poincaré
transformations, as is evident from Eq. (1.81). Both sides of Eq. (1.81) carry only
one Lorentz index.
1.2.4 “Gauge Always Shoots Twice”

In the preceding Sec. 1.2.3, we have already observed that the gauge condi-
tion (1.70), which we recall for convenience,
1 ∂ ′
⃗ ⋅ A⃗′ (⃗
∇ r , t) + 2 Φ (⃗
r , t) = 0 , (1.83)
c ∂t
does not determine Φ and A⃗ uniquely. Indeed, the gauge transformed potentials,
defined according to Eqs. (1.65) and (1.67),
∂
A⃗′ (⃗
r, t) = A⃗ (⃗ ⃗ (⃗
r , t) + ∇Λ r , t) , Φ′ (⃗
r , t) = Φ (⃗
r , t) −
Λ (⃗
r , t) , (1.84)
∂t
fulfill the Lorenz gauge condition if the Λ function obeys Eq. (1.73), which consti-
tutes an inhomogeneous wave equation. Once A⃗′ and Φ′ fulfill the Lorentz condition,
it is possible to do a second gauge transform, using a function Λ′ , and to still remain
within the family of potentials that obey the Lorenz gauge condition.
The Lorenz gauge condition is a first-order partial differential equation imposed
on the 4-vector potential. It is not an algebraic condition. We know that a partial
differential equation allows for many solutions. Therefore, Eq. (1.70) does not
determine the 4-vector potential uniquely, and even given the Lorentz condition,
we still have a certain freedom to choose our 4-vector potential within the “Lorenz
gauge family” without affecting the electric and magnetic fields. This is sometimes
expressed by saying that “gauge always shoots twice.”
Therefore, if A⃗′ and Φ′ satisfy the Lorentz condition (1.70) and Λ′ is yet another
solution of the homogeneous wave equation,
1 ∂2
⃗2 +
(−∇ ) Λ′ (⃗
r, t) = 0 , (1.85)
c ∂t2
then the second gauge transform of vector and scalar potentials reads
A⃗′′ (⃗
r , t) = A⃗′ (⃗ ⃗ ′ (⃗
r , t) + ∇Λ r , t) , (1.86a)
∂
Φ′′ (⃗r , t) = Φ′ (⃗r, t) − Λ′ (⃗ r , t) . (1.86b)
∂t
The second gauge transformation defines new vector and scalar potentials which
still satisfy the Lorentz condition and give the same E ⃗ and B ⃗ fields. The class of all
such combinations of vector and scalar potentials related by this restricted gauge
transformation belong to the Lorenz gauge.
1.2.5 Coulomb Gauge

Another gauge which is often used in radiation theory is the Coulomb, transverse,
or radiation gauge. The gauge condition is
⃗ r, t) = 0 ,
⃗ ⋅ A(⃗
∇ A⃗ (⃗
r, t) = A⃗⊥ (⃗
r , t) , (1.87)
where by A⃗⊥ we refer to the transverse component of the vector potential defined
according to Eq. (1.48b). In general, the vector and scalar potential are coupled to
the sources according to Eq. (1.63). In the radiation gauge, the scalar and vector
potentials thus satisfy the equations
1
⃗ 2 Φ (⃗
−∇ r , t) = ρ (⃗
r, t) , (1.88a)
0
1 ∂2 ⃗ 1 ∂
⃗ 2 A⃗ (⃗
−∇ r , t) + r , t) = μ0 J⃗ (⃗
A (⃗ ⃗ [ Φ (⃗
r , t) − 2 ∇ r , t)] , (1.88b)
2
c ∂t 2 c ∂t
respectively. Equation (1.88b) can be simplified further in terms of two simpler
equations.
We first observe that the left-hand side of Eq. (1.88b) is equal to its own trans-
verse part, because A⃗ (⃗ r, t) = A⃗⊥ (⃗
r , t) in Coulomb gauge [see Eq. (1.87)]; the trans-
verse character of a vector field is not changed by taking the Laplacian, as is imme-
diately obvious from Eq. (1.48a). Namely, for a general vector field V⃗ , if V⃗∥ vanishes
then so does ∇⃗ 2 V⃗∥ , and because of the uniqueness of the separation of a vector field
into longitudinal and transverse components, we then have ∇ ⃗ 2 V⃗ = ∇
⃗ 2 V⃗⊥ . Now we
analyze the right-hand side of Eq. (1.88b) in terms of longitudinal and transverse
components. First, we observe that the expression
1 ∂ 1 ∂
⃗ [ Φ (⃗
∇ ⃗ [ Φ (⃗
r , t)] = ( 2 ∇ r , t)])∣ (1.89)
c 2 ∂t c ∂t ∥
is a longitudinal vector field because the curl of a gradient vanishes,

1 ∂
⃗ ×(
∇ ∇ r , t)]) = ⃗
⃗ [ Φ (⃗ 0. (1.90)
c2 ∂t
The longitudinal component of Eq. (1.88b) is thus equivalent to
∂
J⃗∥ (⃗ ⃗ [ Φ (⃗
r , t) = 0 ∇ r , t)] , (1.91)
∂t
while for the transverse component of Eq. (1.88b), only the transverse component of
⃗ r, t) contributes on the right-hand side. So, we can establish
the current density J(⃗
the three equations couple the scalar and vector potentials to their sources in the
Coulomb or radiation gauge,
1
⃗ 2 Φ (⃗
−∇ r , t) = ρ (⃗
r , t) , (1.92a)
0
1 ∂2
( −∇⃗ 2 ) A⃗⊥ (⃗
r , t) = μ0 J⃗⊥ (⃗r, t) , (1.92b)
c2 ∂t2
∂
⃗ Φ (⃗
0 ∇ r , t) = J⃗∥ (⃗
r, t) . (1.92c)
∂t
Let us now consider, just like for the Lorenz gauge, a second gauge transformation
within the family of the radiation gauge, writing Λ′ as χ,
A⃗′′ (⃗
r , t) = A⃗′ (⃗ ⃗ (⃗
r, t) + ∇χ r, t) , (1.93a)
∂
Φ′′ (⃗r , t) = Φ′ (⃗r , t) − χ (⃗
r, t) . (1.93b)
∂t
The gauge condition imposed on the second transformed vector potential implies
that
⃗ ⋅ A⃗′′ (⃗
∇ r , t) = 0 , ⃗ 2 χ (⃗
∇ r , t) = 0 . (1.93c)
Thus, any function χ = χ(⃗ r , t) that fulfills ∇⃗ 2 χ (⃗r, t) = 0 describes a second gauge
transform of the vector and scalar potentials within the radiation gauge.
The radiation gauge is well suited for describing the propagation of radiation in
the absence of sources, i.e., in a region of space where there are no charges, and no
currents. If sources are absent, then ∇ ⃗ 2 Φ (⃗
r , t) = 0, and therefore Φ (⃗r , t) = 0, and
so
⃗ (⃗ ∂
E r , t) = − A⃗ (⃗ r , t) , (1.94a)
∂t
⃗ (⃗
B ⃗ × A⃗ (⃗
r , t) = ∇ r,t) . (1.94b)
Let us consider a typical vector potential describing a traveling electromagnetic

wave, of the form
⃗ r , t) = ê⃗ A0 cos(k⃗ ⋅ r⃗ − ωt) ,
A(⃗ (1.95)
kλ
where k⃗ is the wave vector, and ω = c ∣k∣.

⃗ The polarization vectors ê⃗ are two unit
kλ
⃗
vectors perpendicular to k (in consequence, we have λ = 1, 2). Then,
∇ ⃗ r, t) = −(k⃗ ⋅ ê⃗ ) A0 sin(k⃗ ⋅ r⃗ − ωt) = 0

⃗ ⋅ A(⃗ (1.96)
kλ
holds because the polarization vector is always perpendicular to the propagation

vector, k⃗ ⋅ êkλ
⃗ .
However, the gauge is also called the Coulomb gauge. This is because the
equation that couples the electrostatic potential to the source,
⃗ 2 Φ(⃗ 1
∇ r , t) = − ρ(⃗
r, t) , (1.97)
0
has the instantaneous, action-at-a-distance solution
1 1
r, t) = d3 r′ r′ , t) .
4π0 ∫
Φ(⃗ ρ(⃗ (1.98)
r − r⃗′ ∣
∣⃗
The similarity of this solution with the Coulomb law, i.e., the simple presence of
1
the factor ∣⃗r−⃗
r′ ∣
, implies the nomenclature of Coulomb gauge. Moreover, in a time-
independent situation, the dependence on t cancels, and the Poisson equation from
electrostatics is immediately recovered. However, for time-dependent problems, the
action-at-a-distance solution poses important questions. If r⃗′ is in the Andromeda
Galaxy and r⃗ is near Alpha Centauri, then we have an apparent problem with the
causality principle because the field is generated at the same point in time as the
source acts, namely, at time t. The task then is to show that the perceived action-at-
a-distance character of the potential does not imply that the electric fields generated
by the charges violate the causality principle (see Chap. 4.2).
1.3 Poynting Theorem and Maxwell Stress Tensor
1.3.1 Electric and Magnetic Field Energies

The introduction into basic properties of the electromagnetic field in the current
chapter would not be complete without a discussion of the field energy, or, energy
density stored in any nonvanishing electromagnetic field. Newton7 invented the
concept of a physical field, as an entity which exists at every point in space-time
even if the existence of the field is not always detected or observed by a probe.
In consequence, it is a natural question to ask if we can assign an “energy” to a
field configuration, and if yes, how the corresponding “momentum flow” of the field
could be described.
Imagine an electrically charged, heavy object in the middle of a coordinate
system, and an oppositely charged test object far apart. The initial configuration
has a nontrivial, nonvanishing electric field configuration. Then, the test object is
accelerated toward the center of the heavy charged object. At the moment when
they meet, the field vanishes because the system is effectively electrically neutral.
(In practice, of course, there is still a small dipole moment.) The test object has
gained some kinetic energy; the electric force has performed work on the test body.
However, the potential energy of the initial configuration is gone. We can either
ascribe the initial potential energy to the position of the charges in their potentials,
or ascribe it to the field configuration as a whole. The latter aspect will be the
center of discussion here.
We first derive a seemingly innocent vector identity,
1 ⃗ ⃗ ⃗ r , t)]
∇ ⋅ [E(⃗
r , t) × B(⃗
μ0
1 ⃗ ⃗ r, t)] + 1 B(⃗ ⃗ r , t) ⋅ ∇ ⃗ r , t)]
= − E(⃗ ⃗ × H(⃗
r, t) ⋅ [∇ ⃗ × [E(⃗
μ0 μ0
⃗ r , t) ⋅ [J⃗ (⃗ ∂ ⃗ 1 ⃗ ∂ ⃗
= −E(⃗ r , t) + 0 E(⃗ r, t)] + [B(r, t) ⋅ (− B(⃗ r , t))]
∂t μ0 ∂t
⃗ r , t) ⋅ J(⃗
= −E(⃗ ⃗ r , t) ⋅ ∂ E(⃗
⃗ r , t) − 0 E(⃗ ⃗ t) ⋅ ∂ B(⃗
⃗ r, t) − 1 B(r, ⃗ r, t)
∂t μ0 ∂t
= −E(⃗ ⃗ r , t) − ∂ [ 0 E(⃗
⃗ r , t) ⋅ J(⃗ ⃗ r, t) + 1 B(⃗
⃗ r , t) ⋅ E(⃗ ⃗ r , t) ⋅ B(⃗
⃗ r , t)] . (1.99)
∂t 2 2μ0
In order to interpret the terms on the right-hand side of this we first recall that if
the current density J⃗ (⃗
r , t) is to be nonvanishing, charges have to move. We can
write
⃗ r , t) = Δq
J(⃗ v̂, (1.100)
Δt ΔA
where Δq is the charge moving in the time interval Δt through the cross-sectional
area ΔA, in a direction given by the unit velocity vector v̂.
7
Sir Isaac Newton (1642–1726)
The mechanical work done on the moving charges per unit time and unit volume,
by the electric force acting on the charges due to the electric field, is
⃗ (⃗ Δq E⃗ (⃗ r, t) ⋅ ̂v (Δq E) ⃗ ⋅ Δ⃗
x
E r , t) ⋅ J⃗ (⃗
r, t) ≈ =
Δt ΔA Δt ΔV
(ΔF⃗elec ) ⋅ Δ⃗ x ΔW
= = . (1.101)
Δt ΔV Δt ΔV
We therefore identify
⃗ (⃗ ∂
E r , t) ⋅ J⃗ (⃗
r , t) = w(⃗ r , t) (1.102)
∂t
as the time derivative of the work density w(⃗ r , t), i.e., as the power density. Note
that this is the work done by the electric field on the charges, i.e., the work done
to accelerate the charges, not the work necessary to move the charges against the
electric field.
We can naturally identify
0 ⃗ 1 ⃗
u (⃗r , t) = [E(⃗ r, t)]2 + [B(⃗ r , t)]2 (1.103)
2 2μ0
as the energy density of the electromagnetic field. With these identifications,
Eq. (1.101) takes (almost) the form of a continuity equation,
1 ⃗ (⃗ ⃗ (⃗ ∂u (⃗ r, t) ∂w (⃗ r , t)
⃗ ⋅ [E
∇ r , t) × B r , t)] + + = 0. (1.104)
μ0 ∂t ∂t
We identify the Poynting vector S(⃗ ⃗ r , t) as
1 ⃗
S⃗ (⃗r , t) = E(⃗ ⃗ r , t) ,
r , t) × B(⃗ (1.105)
μ0
i.e., as the energy current density leaving a small test volume, where S⃗ has the
physical dimension of energy per time per area. We can write
∇ ⃗ r, t) + ∂u (⃗
⃗ ⋅ S(⃗
r, t) ∂w (⃗
+
r , t)
= 0. (1.106)
∂t ∂t
In view of the divergence theorem, this implies that
∫ ⃗ r , t) ⋅ dA⃗ = − ∫ E(⃗
S(⃗ ⃗ r , t) d3 r − ∫ ∂u (⃗
⃗ r , t) ⋅ J(⃗ r , t) 3
d r. (1.107)
∂V V V ∂t
The interpretation is clear: The rate at which energy leaves a region plus the
rate at which field energy is converted to mechanical energy is compensated by a
commensurate loss in field energy, i.e., equal to minus the rate at which the field
energy changes. Note that E(⃗ ⃗ r , t) ⋅ J(⃗
⃗ r , t) = ∂t w(⃗ r , t) is positive when work is done
on the charges by the field, i.e., when the work is done “in the direction” of the
electromagnetic fields. This work leads to a loss in the field energy density, or
otherwise to an increase in the kinetic energy of the charges inside the reference
volume. Work done on the charges against the direction of the electromagnetic
fields (“actual work”) leads to a negative value for ∂t w(⃗ r , t), and to a positive value
for the radiated energy ∫∂V S(⃗ ⃗ r , t) ⋅ dA⃗ per time.
For vanishing power density ∂t w(⃗ r , t) = 0, let us write the energy dissipation
equation (1.106) in a slightly different way,
⃗ r , t) = − ∂u (⃗
⃗ ⋅ S(⃗
∇
r , t)
,
∂w(⃗
r , t)
=0 (1.108)
∂t ∂t
and compare it to the continuity equation (1.29) (for the total current)
∇ ⃗ r , t) = − ∂ρ (⃗
⃗ ⋅ J(⃗ r , t)
. (1.109)
∂t
Here, J⃗ is the current distribution, and ρ is the volume charge density. We can
observe the analogy S⃗ ↔ J⃗ and u ↔ ρ. The continuity equation states that a loss
in the local charge distribution can only occur if charge leaves the area. The energy
dissipation equation states that a loss in the local field energy distribution is com-
pensated when field energy leaves the area (Poynting vector), or when performing
work on the local charges.
1.3.2 Maxwell Stress Tensor

From the above discussion, it is clear that the Poynting vector describes the flow of
electromagnetic field energy. Furthermore, the quantity S(⃗ ⃗ r, t)/c has the physical
dimension of an energy density. Energy has the dimension of force multiplied by
⃗ r , t)/∂t has the physical dimension of a force
length, and so the quantity (1/c2 )∂ S(⃗
density, because c t has the dimension of length. One may thus ask if it is possible
to formulate the force density acting on an ensemble of charges in terms of the
electromagnetic fields, and conjecture that the resulting formula might contain a
⃗ r, t)/∂t.
term proportional to (1/c2 )∂ S(⃗
We start from the Lorentz force on a charged particle,
F⃗ = q (E(⃗
⃗ r, t) + v⃗ × B(⃗
⃗ r , t)) . (1.110)
The force density, or force per unit volume, acting at point r⃗ and time t, is
ΔF⃗ Δq ⃗ Δq v⃗ ⃗
= E+ ×B. (1.111)
ΔV ΔV ΔV
Defining f⃗ as the force density, and identifying Δq/ΔV → ρ and Δq v⃗/ΔV → J,
⃗ in
the limit of a small reference volume ΔV , we have
f⃗(⃗
r , t) = ρ(⃗ ⃗ r , t) + J(⃗
r , t) E(⃗ ⃗ r , t) × B(⃗
⃗ r, t) . (1.112)
In the following, we suppress the space-time arguments (⃗ r, t) of the fields and
sources. We now use Gauss’s law and the Ampere–Maxwell law to express the
charge density ρ and the current density J⃗ in terms of the fields,
⃗
f⃗ = (0 ∇ ⃗ E
⃗ ⋅ E) ⃗ +( 1 ∇⃗ ×B⃗ − 0 ∂ E ) × B
⃗
μ0 ∂t
1 ⃗
⃗∇
= 0 E ⃗ ⋅ E⃗ + ⃗ ×B
⃗ × B)
(∇ ⃗ − 0 ∂ E × B⃗. (1.113)
μ0 ∂t
The time derivative of the Poynting vector may be written as

∂ ⃗ ⃗ ∂E⃗ ⃗ ⃗
(E × B) = ×B ⃗ +E⃗ × ∂B = ∂E × B ⃗ −E ⃗ × (∇ ⃗ ,
⃗ × E) (1.114)
∂t ∂t ∂t ∂t
where Faraday’s law (1.2c) has been used. Solving this equation for ∂t E ⃗ × B,
⃗ one
obtains
∂E⃗
×B ⃗ = ∂ (E⃗ × B)
⃗ +E ⃗ × (∇ ⃗ .
⃗ × E) (1.115)
∂t ∂t
Using this relation in Eq. (1.113), we have
f⃗ = 0 E
⃗ (∇ ⃗ + 1 (∇
⃗ ⋅ E) ⃗ ×B
⃗ × B) ⃗ − 0 ∂ (E ⃗ × B) ⃗ × (∇
⃗ − 0 E ⃗
⃗ × E)
μ0 ∂t
⃗ (∇
= 0 [E ⃗ − E⃗ × (∇
⃗ ⋅ E) ⃗
⃗ × E)]
1 ⃗ ⃗ −B ⃗ × (∇ ⃗ − 0 ∂ (E⃗ × B) ⃗ ,
+ [B (∇ ⃗ ⋅ B) ⃗ × B)] (1.116)
μ0 ∂t
where an “artificial zero” has been added in the form of the term ∇ ⃗ ⋅B⃗ = 0 in order
to render the (first part of the) resulting expression symmetric in E ↔ B. ⃗ ⃗
⃗
For any vector field V , one can write the identity
V⃗ × (∇
⃗ × V⃗ ) = êi ijk Vj km ∇ Vm = êi (δi δjm − δim δj ) Vj ∇ Vm
1
= Vj êi ∇i Vj − Vj ∇j êi Vi = ∇⃗ (V⃗ ⋅ V⃗ ) − (V⃗ ⋅ ∇)
⃗ V⃗ , (1.117)
2
where we carefully keep track of the fields on which the gradient operator acts (only
those to the right of the operator). The force density can be written as follows,
f⃗ = 0 [E⃗ (∇ ⃗ + (E
⃗ ⋅ E) ⃗ ⋅ ∇)
⃗ E]⃗ + 1 [B ⃗ (∇ ⃗ + (B
⃗ ⋅ B) ⃗ ⋅ ∇) ⃗
⃗ B]
μ0
1 ⃗2 1 ⃗2 ∂ ⃗ ⃗
+∇ ⃗ ( 0 E + B ) − 0 (E × B) . (1.118)
2 2μ0 ∂t
We are now in the position to define the stress tensor of the electromagnetic field,
with components ij ,
1 1 1
ij = 0 (Ei Ej − δij E⃗ 2 ) + (Bi Bj − δij B ⃗2) , (1.119)
2 μ0 2
or in dyadic notation,
⃗2 ⃗2
= 0 (E ⃗⊗E
2

⃗ − E 3×3 ) + 1 (B
μ0
⃗⊗B
2

⃗ − B 3×3 ) = ij êi ⊗ êj . (1.120)
Then,
1
⃗ ⋅ = ∇i ij êj = 0 êj (Ej ∇i Ei + Ei ∇i Ej ) +
∇ (Bj ∇i Bi + Bi ∇i Bj )
μ0
1 1 ⃗2
− êj ∇j ( 0 E⃗ 2 + B )
2 2μ0
= 0 (E ⃗ (∇ ⃗ + (E
⃗ ⋅ E) ⃗ ⋅ ∇)
⃗ E)⃗ + 1 (B ⃗ (∇ ⃗ + (B
⃗ ⋅ B) ⃗ ⋅ ∇) ⃗
⃗ B)
μ0
1 ⃗2 1 ⃗2
−∇ ⃗ ( 0 E + B ). (1.121)
2 2μ0
We can finally write Eq. (1.118) as follows,

∂ 1 ⃗ ⃗ 1 ∂ ⃗
f⃗ = ∇
⃗ ⋅ − 0 μ0 ⃗⋅ −
(E × B) = ∇ S. (1.122)
∂t μ0 c2 ∂t
Here, S⃗ is the Poynting vector, defined in Eq. (1.105). Equation (1.122) is dimen-
sionally correct; all components of the stress tensor have a physical dimension of
energy per volume, or force per area, which is the same as the physical dimen-
sion of pressure. The relativistic generalization of the stress tensor, the so-called
stress–energy tensor, is discussed in Sec. 8.3.2.
1.4 Exercises
● Exercise 1.1: Repeat the derivation the equivalence of the integral and differen-
tial forms of Maxwell equations for the absence of magnetic monopoles, i.e., show
that the following two equations are equivalent:
(i) ∇ ⃗ r, t) = 0 ,
⃗ ⋅ B(⃗ (ii) ∫ ⃗ r , t) ⋅ dA⃗ = 0 .
B(⃗ (1.123)
∂V
● Exercise 1.2: Show that the Kronecker symbol δij is a tensor under rotations.
Geometrically, this means that if two vectors a ⃗ and ⃗b undergo a rotation a
⃗→a⃗′ =
R⋅⃗ ⃗ ⃗′ ⃗
a, b → b = R⋅ b, then a ⃗
⃗ ⋅b = a ′ ⃗′
⃗ ⋅ b (the rotation matrix is denoted as R). Component-
wise, it means that the components of the tensor are invariant under rotations,
i.e., that
′
δij = Rik δk Rj
T
= δij . (1.124)
The sum over k and is implicitly assumed by the Einstein summation convention.
Show the equivalence of both statements, refer to a general form of a rotation
matrix, and relate Eq. (1.124) to a well-known property of rotation matrices.
● Exercise 1.3: Show that ijk is a tensor. Geometrically, this means that if three
vectors a⃗, ⃗b and c⃗ undergo a rotation, a ⃗→a ⃗, and ⃗b → ⃗b′ = R ⋅ ⃗b as well as
⃗′ = R ⋅ a
′
⃗ × (⃗b × c⃗) = a
c⃗ → c⃗ = R ⋅ c⃗, then a ⃗ ⋅ (⃗b × c⃗ ). Again, the rotation matrix is denoted
′ ′ ′
as R. Component-wise, it means that the components of the δ tensor are invariant

under rotations, i.e., that
′ijk = Ri mn Rmj
T T
Rnk = ijk . (1.125)
The sum over , m and n is implicitly assumed by the Einstein summation conven-
tion. Show the equivalence of both statements, and the invariance of the components
of the tensor under rotations.
● Exercise 1.4: Using your favorite graphical and numerical computer package
(Mathematica, Maple, Matlab, Axiom, etc.) investigate the properties of the
integral
π/2 π/2
I(ω0 ) = ∫ dω δ(ω − ω0 ) f (ω) → I(η, ω) = ∫ dω Δ(η, ω − ω0 ) f (ω) (1.126)
0 0
π
ω0 = , f (ω) = sin2 (ω) , (1.127)
2
where Δ is defined as in Eq. (1.32). Observe how I(η, ω) converges to I(ω0 ) = 1 as

η → 0.
● Exercise 1.5: Show Eq. (1.34) by first convincing yourself that δ[a (x − x0 )] =
δ(x−x0 )/a for a positive constant a. Then, expand the argument f (x) of δ(f (x)) in
terms of δ(f (x)) ≈ δ[f (x0 ) + f ′ (x0 ) (x − x0 )] and make an appropriate assumption
(which one?) about x0 .
● Exercise 1.6: Show that, because the Fourier transform of the Dirac-δ distri-
bution is just a constant, the function g = g(⃗ r − r⃗′ ) defined in Eq. (1.37) can be
interpreted as the inverse of the Laplacian operator. In particular, deliberate on
how you could formulate the quantity (1/∇ ⃗ 2 )G(⃗r , r⃗′ ) using the momentum-space
representation? (Consult Chaps. 4 and 6 for a systematic introduction to Fourier
transforms in space and time.)
● Exercise 1.7: Consider the scalar and vector potentials for a traveling wave [see
Eq. (1.95)],
⃗ r −ωt)
r , t) = 0 ,
Φ(⃗ ⃗ r , t) = ê⃗ A0 ei(k⋅⃗
A(⃗ , (1.128)
kλ
⃗ (We here anticipate the complex formalism for the description of
where ω = c ∣k∣.
fields; the real part of Eq. (1.128) is equivalent to Eq. (1.95).) (i) Show that the
scalar and vector potentials given in Eq. (1.128) fulfill both the Lorenz and Coulomb
gauge potentials. (ii) Consider the gauge function
⃗
r, t) = χ0 ei(k⋅⃗r−ωt) .
Λ(⃗ (1.129)
Show that Λ(⃗ r, t) fulfills the condition (1.73), and calculate the gauge transformed
scalar and vector potentials Φ′ and A⃗′ . Show that these give rise to the same fields
E⃗ and B.
⃗ Hint: You might obtain something like
⃗
Φ′ (⃗
r , t) = i ω χ0 ei(k⋅⃗r−ωt) , (1.130)
and
⃗
r , t) = i k⃗ χ0 ei(k⋅⃗r−ωt) + A(⃗
A⃗′ (⃗ ⃗ r, t) . (1.131)
● Exercise 1.8: Consider the scalar and vector potentials,
Φ(⃗ ⃗
r, t) = −E(t) ⋅ r⃗ , ⃗ r , t) = ⃗
A(⃗ 0. (1.132)
Show that under the gauge transformation
Λ(⃗ ⃗ ⋅ r⃗ ,
r, t) = A(t) (1.133)
the following gauge-transformed potentials are obtained,
Φ′ (⃗
r, t) = 0 , A⃗′ (⃗ ⃗ .
r , t) = A(t) (1.134)
You have transformed from a situation with a vanishing vector potential, to a gauge
with a vanishing scalar potential. Equation (1.132) is referred to as the length gauge,
and Eq. (1.134) is called the velocity gauge. A decisive feature is that E(t) ⃗ and
⃗
A(t) are time-dependent, but independent of the spatial coordinates.
● Exercise 1.9: Under which conditions can a gauge transformation be found which
implements the temporal gauge Φ′ (⃗ r, t) = 0? This is a difficult exercise.
Chapter 2
Green Functions of Electrostatics
2.1 Overview
In the current chapter, we shall encounter basic techniques involved in the calcula-
tion of so-called Green functions. These functions are central to theoretical physics.
They have many applications; in particular, Green functions connect a physical sig-
nal to its source, in space or time. We shall start by discussing the Green function
of the harmonic oscillator (see Sec. 2.2), which connects the position of a classical
particle bound in a harmonic potential, to a “source” term, which for mechanical
problems, is simply a force acting on the particle. The Green function is calculated
by Fourier transformation with respect to time, or alternatively by matching solu-
tions of the homogeneous equation near the cusp which defines the Green function.
Both techniques for calculating the Green functions will be used in the following
chapters of this book.
The “matching near the cusp” technique will be used in Secs. 2.3 and 2.4, for
the calculation of the electrostatic Green function in spherical and cylindrical co-
ordinates. Here, the connection of the “signal” (the electrostatic potential) to the
source (the charge density) is mediated in three-dimensional space as opposed to the
time domain (as for the harmonic oscillator). The calculation of the Green function
in spherical coordinates (see Sec. 2.3) directly leads to the multipole expansion of
electrostatics. Finally, in Sec. 2.5, we shall encounter a third, alternative method
for the calculation of a Green function, namely, a so-called eigenfunction expansion,
which is inspired by ideas originally developed within quantum mechanics.
2.2 Green Function of the Harmonic Oscillator
One of the most basic applications of the Green function formalism concerns the
harmonic oscillator, which is introduced here as an example for the basic paradigm
of the formalism. Indeed, from our elementary education on electric circuits, we
know that a series resonant circuit consisting of a coil with inductance L, a resistor
with resistance R, and a capacitor with capacitance C fulfills
dI Q
L + I R + = V (t) , (2.1)
dt C
29
where V (t) is the applied, possibly time-dependent voltage. If one uses the fact
that the current I through the coil is equal to the time derivative of the charge Q,
then the equation reads
d2 Q dQ Q
2
L+R + = V (t) . (2.2)
dt dt C
The mechanical analogue (damped harmonic oscillator) reads as follows,
d2 z dz
m2
+r +Dz = 0. (2.3)
dt dt
Here, D is the spring constant, r is the coefficient of the damping term (linear in
the velocity), m is the inertia term (kinetic energy, corresponds to the inductance
L for the electric circuit), and z is the time-dependent elevation (z coordinate) of
the test object attached to the spring.
Both Eqs. (2.2) and (2.3) can be brought into the following form by a trivial
scaling and reidentification of the parameters,
ẍ(t) + γ ẋ(t) + ω02 x(t) = f (t) , (2.4)
where the dot means differentiation with respect to the time. The Green function
g = g(t − t′ ) for this equation fulfills the equation
g̈ (t − t′ ) + γ ġ (t − t′ ) + ω02 g (t − t′ ) = δ (t − t′ ) , (2.5)
where δ is the Dirac-δ distribution and the dot again denotes the differentiation
with respect to t (not t′ ). Introducing the variable s ≡ t − t′ , we could write the
defining equation as
d2 d
2
g (s) + γ g (s) + ω02 g (s) = δ (s) . (2.6)
ds ds
The primary use of the Green function is as follows: It describes the response
of the system to a point-like (in time) perturbation, as described by a Dirac δ
function. The response of the system to an arbitrary perturbation can be obtained
by summing (integrating) the response of the system to the perturbation f (t′ )
weighted with the Green function g(t − t′ ),
x(t) = ∫ g(t − t′ ) f (t′ ) dt′ . (2.7)
In view of Eq. (2.5), x(t) defined according to Eq. (2.7) fulfills Eq. (2.4). The
relation (2.7) illustrates why the “natural” argument of the Green function is the
time difference s = t − t′ . We note that Eq. (2.5) does not define the Green func-
tion uniquely. Given a particular solution g, we can always add a solution of the
homogeneous equation
d2 d
h (s) + γ h (s) + ω02 h (s) = 0 , (2.8)
ds2 ds
and g + h will also be a Green function. This finding illustrates in very basic
terms that the Green function is only defined uniquely if it is supplied with suitable
Green Functions of Electrostatics 31
boundary conditions. For our problem, the general solution to the homogeneous
equation reads (s = t − t′ )
1 1√ 2
h(s) = A exp (− γ s) sin ( 4ω0 − γ 2 s)
2 2
1 1√ 2
+ B exp (− γ s) cos ( 4ω0 − γ 2 s) , (2.9)
2 2
where A and B are integration constants that can be adjusted to fulfill specific
boundary conditions.
Let us now consider the Fourier representation of the Green function. Indeed,
writing the Green function as a Fourier transform
∞
dω −iω(t−t′ )
g(t − t′ ) = ∫ e g(ω) , (2.10)
−∞ 2π
we note that g(t − t′ ) and g(ω) are different functions and have different physical
dimension; one of them is obtained from the other by integration over dω. One
might ask if this can be mathematically correct. The answer is that it can be
correct and unique if we interpret the symbols g to be “overloaded” in the sense of
an overloaded operator in “Fortran 90” or “C++” notation: The physical dimension
of the argument (time or frequency) defines the functional form to be used for g. It
is easy to show that
1 1
g(ω) = − =− , (2.11)
ω 2 + i γ ω − ω02 (ω − ω+ ) (ω − ω− )
where the poles are given by
1 √
ω± = (± 4 ω02 − γ 2 − i γ) . (2.12)
2
For light damping (4ω02 > γ 2 ), both of these poles ω± have a negative imaginary
part. In order to carry out the integration in Eq. (2.10), we now need to calculate
the residues
1 1
Res g(ω) = Res (− ) = ∓√ . (2.13)
ω=ω± ω=ω± ω2 + i γ ω − ω0
2
4 ω02 − γ 2
For t − t′ > 0, we have to close the contour in Eq. (2.10) in the lower half of the
complex plane because the expression exp[−iω(t − t′ )] is exponentially damped for
Im(ω) < 0. In that case, we pick up nonvanishing residues at ω = ω± and obtain
√
sin ( 12 4ω02 − γ 2 (t − t′ ))
′ ′ −γ(t−t′ )/2
g(t − t ) = 2 Θ(t − t ) e √ , (2.14)
4ω02 − γ 2
where Θ is the Heaviside step function, which is unity for nonnegative argument,
and zero otherwise. A nonvanishing result is obtained only for t > t′ . For t < t′ ,
the contour needs to be closed in the upper half of the complex plane and no
pole contributions are picked up. The Green function (2.14) is the retarded Green
function; it vanishes when t < t′ . Furthermore, it vanishes in the limit t − t′ → ∞.
An alternative way of constructing the Green function is as follows. We first

observe that the homogeneous and inhomogeneous Eqs. (2.8) and (2.6), which we
recall for convenience,
d2 d
g (s) + γ g (s) + ω02 g (s) = δ (s) , (2.15a)
ds2 ds
d2 d
2
h (s) + γ h (s) + ω02 h (s) = 0 , (2.15b)
ds ds
are actually equivalent to each other except for the immediate vicinity of s = 0,
because δ(s ≠ 0) = 0. Now, the general solution of the homogeneous equation is
given in Eq. (2.9). Hence, we use an ansatz of the form
1√ 2
g (s) = Θ(s) [A+ e−γ s/2 sin ( 4ω0 − γ 2 s)
2
1√ 2
+B+ e−γ s/2 cos ( 4ω0 − γ 2 s)]
2
1√ 2
+ Θ(−s) [A− e−γ s/2 sin ( 4ω0 − γ 2 s)
2
1√ 2
+B− e−γ s/2 cos ( 4ω0 − γ 2 s)] . (2.16)
2
The integration constants A± and B± can be determined by (i) boundary conditions
and (ii) integrating Eq. (2.6) in an infinitesimal interval about s = 0. For the
retarded Green function, we have to require that A− = B− = 0. This also follows
from the regularity requirement at s = −∞; the numerical value of the Green function
would otherwise diverge in that limit. Furthermore, as the Green function needs to
be continuous at s = 0, we have to impose the condition B+ = 0; the cosine would
otherwise induce a kink. After setting A− = B− = B+ = 0, the remaining parameter
A+ can be determined by integrating Eq. (2.15a) in an infinitesimal interval around
s = 0,
d2 d
1= ∫ δ (s) ds = ∫ [ 2
g (s) + γ g (s) + ω02 g (s)] ds
− − ds ds
d d
=g (s)∣ − g (s)∣ + γ [g() − g(−)] + ω02 g(0) + O(2 )
ds s= ds s=−
d d 1 √
= g (s)∣ − g (s)∣ + O() = ( A+ 4 ω02 − γ 2 ) − (0) + O() . (2.17)
ds s= ds s=− 2
We have assumed that the Green function itself is continuous while its derivative
may have a kink. The result thus reads
2
A+ = √ . (2.18)
4 ω02 − γ 2
Inserting this result into Eq. (2.16), we recover (2.14),
√
sin ( 12 4ω02 − γ 2 s)
g(s) = 2 Θ(s) e−γs/2 √ . (2.19)
4ω02 − γ 2
Fig. 2.1 Plot of the Green function g(s) given in Eqs. (2.14)
and (2.19) for γ = ω0 /10 and ω0 = 100.
A plot of the Green function for typical parameters is given in Fig. 2.1. We shall use
the “concatenation approach” for the construction of the Green function in future
derivations, related to electrodynamics.
2.3 Electrostatic Green Function and Spherical Coordinates
2.3.1 Poisson and Laplace Equations in Electrostatics

In generalizing the result of the previous section on damped harmonic oscillators
to electrostatics, we encounter the first example of a Green function relevant to
electrodynamics. Below, we formulate the Poisson equation of electrostatics as the
central inhomogeneous equation which gives rise to a Green function. In spherical
coordinates, the equations of electrostatics naturally split into angular and radial
parts. We shall find that the angular part of the equations is easily solved in terms of
so-called spherical harmonics, which constitute some of the most important special
functions of theoretical physics. For the radial equation, we shall first find the
solutions to the homogeneous equation. Near the turning point, the solutions to the
homogeneous equation will be matched such as to fulfill the defining inhomogeneous
equation of the Green function, in an approach similar to that used in Eqs. (2.16)–
(2.19). The calculation of the Green function for electrostatics is more difficult than
for the damped harmonic oscillator because it is manifestly three-dimensional.
In electrostatics (no time dependence), the electric field E(⃗⃗ r ) is obtained from
the potential Φ(⃗ ⃗
r ) as E(⃗ ⃗
r ) = −∇Φ(⃗
r ). If the charge distribution vanishes, ρ(⃗
r ) = 0,
then the potential fulfills the Laplace equation
⃗ 2 Φ(⃗
∇ r) = 0 (2.20)
r ), one obtains the Poisson
by virtue of Gauss’s law (1.2a). For nonvanishing ρ(⃗
equation
1
⃗ 2 Φ(⃗
∇ r ) = − ρ(⃗
r) . (2.21)
0
The Green function G of the Poisson equation therefore has to fulfill the defining
equation
1
⃗ 2 G(⃗
∇ r , r⃗′ ) = ∇⃗ 2 G(⃗r − r⃗′ ) = − δ (3) (⃗ r − r⃗′ ) , (2.22)
0
where ∇⃗ 2 is the Laplace operator, which acts on r⃗ (not r⃗′ ) unless stated otherwise.
The three-dimensional Dirac-δ function has been defined in Eq. (1.35). In view of
Eqs. (1.37), (1.38) and (1.42), the desired solution can immediately be written down
as
1
G(⃗ r , r⃗′ ) = G(⃗r − r⃗′ ) = . (2.23)
r − r⃗′ ∣
4π0 ∣⃗
The Green function provides a convenient way to evaluate integrals of the form
⃗ 2 Φ(⃗ 1
Φ(⃗r) = ∫ d3 r′ G(⃗
r , r⃗′ ) ρ(⃗
r′ ) , ∇ r ) = − ρ(⃗
r) , (2.24)
0
for extended charge distributions ρ(⃗ r′ ), which are solutions to the Poisson equa-
tion (2.21).
We here anticipate the central result of our discussion of the electrostatic Green
function in spherical coordinates, which is the multipole decomposition
1 1 ∞ 1 r<
r , r⃗′ ) =
G(⃗ = ∑ ∑ Y m (θ, ϕ) Y∗m (θ′ , ϕ′ ) . (2.25)
4π0 ∣⃗
r − r⃗ ∣ 0 =0 m=− 2 + 1 r>+1
′
Here, r< = min(r, r′ ), r> = max(r, r′ ), while θ and ϕ are the polar and azimuth angles
of the vector r⃗, and correspondingly for θ′ and ϕ′ . The Y m functions are spherical
harmonics, as defined in Eq. (2.53). We shall now work toward deriving Eq. (2.25),
using spherical coordinates. To this end, we first have to analyze the Laplace equa-
tion in spherical coordinates, before “concatenating” the expression (2.25), which
is a solution of the Poisson equation, from the solutions of the Laplace equation, at
the cusp r< = r> , in a way similar to the derivation outlined in Eqs. (2.16)–(2.19).
2.3.2 Laplace Equation in Spherical Coordinates

The Laplacian ∇ ⃗ 2 G(⃗
r , r⃗′ ) of the Green function G vanishes almost everywhere.
Indeed, it vanishes for all r⃗ except at the Dirac-δ source δ (3) (⃗
r − r⃗′ ). We thus have
⃗
to investigate the Laplacian operator ∇ in spherical coordinates, which are defined
2
as
x = r sin θ cos ϕ , y = r sin θ sin ϕ , z = r cos θ , (2.26a)
with
√ ⎛ z ⎞ ⎛ x ⎞
r= x2 + y 2 + z 2 , θ = arcsin √ , ϕ = arccos √ . (2.26b)
⎝ x +y ⎠
2 2 ⎝ x +y ⎠
2 2
One writes, by the chain rule,

∂ ∂r ∂ ∂θ ∂ ∂ϕ ∂
= + + , (2.27)
∂x ∂x ∂r ∂x ∂θ ∂x ∂ϕ
thereby expressing the partial derivative operator ∂/∂x as a linear combination of

∂/∂r, ∂/∂θ, and ∂/∂ϕ. By a tedious, but straightforward calculation, one then
obtains the conversion of the Laplacian operator in spherical coordinates,
2 2 2
⃗2 = ∂ + ∂ + ∂
∇
∂x2 ∂y 2 ∂z 2
∂2 2 ∂ 1 ∂ ∂ 1 ∂2
= 2+ + 2 sin θ + 2 2 . (2.28)
∂r r ∂r r sin θ ∂θ ∂θ r sin θ ∂ϕ2
(In Chap. 8, we shall explore Christoffel symbols which provide a more systematic
view on the curvilinear coordinate systems.)
In quantum mechanics, one often introduces the angular momentum operator
L ̵ r × ∇,
⃗ = r⃗ × p⃗ = −i h⃗ ̵∇
⃗ where p⃗ = −ih ⃗ is the momentum operator. This book is not
about quantum mechanics; however, the designation of the first subscript of the
spherical harmonic Ym actually is inspired by the angular momentum quantum
number. In the context of classical electrodynamics, it is thus helpful to introduce
a dimensionless version L ⃗ of the angular momentum operator as the vector product
⃗ = −i r⃗ × ∇
L ⃗, (2.29)
so that the Cartesian components of L ⃗ can be expressed in terms of the spherical
coordinates as
∂ ∂ ∂ ∂
Lx = − i (y −z ) = i (sin ϕ + cos ϕ cot θ ), (2.30a)
∂z ∂y ∂θ ∂ϕ
∂ ∂ ∂ ∂
Ly = − i (z − x ) = i (− cos ϕ + sin ϕ cot θ ), (2.30b)
∂x ∂z ∂θ ∂ϕ
∂ ∂ ∂
Lz = − i (x −y ) = −i . (2.30c)
∂y ∂x ∂ϕ
⃗ operator reads as
The square of the L
⃗2 = L
⃗⋅L
⃗ = −(⃗ 1 ∂ ∂ 1 ∂2
L ⃗ 2=−
r × ∇) sin θ − . (2.31)
sin θ ∂θ ∂θ sin2 θ ∂ϕ2
We can thus write the Laplacian operator as follows,
2
⃗2 = ∂ + 2 ∂ − 1 L
∇ ⃗2 . (2.32)
∂r2 r ∂r r2
This identity separates radial and angular variables in spherical coordinates. The
homogeneous (Laplace) equation in spherical coordinates becomes
⃗ 2 Φ (r, θ, ϕ) = 0 ,
∇ Φ(r, θ, ϕ) = R (r) Y (θ, ϕ) , (2.33)
where the latter equation constitutes an ansatz for the solution, obtained by sepa-
ration of variables. Inserting this ansatz into Eq. (2.32), one obtains
1 ∂ ∂ 1 ⃗ 2 Y (θ, ϕ) .
{ [r2 R(r)]} = L (2.34)
R(r) ∂r ∂r Y (θ, ϕ)
This equation must be satisfied for all (r, θ, ϕ), which are independent variables,
Equation (2.34) can only be satisfied if both sides are equal to a constant which we
write as ( + 1). Then,
⃗ 2 Y (θ, ϕ) = ( + 1) Y (θ, ϕ) ,
L (2.35)
and the radial part of the function fulfills
d 2 d
[r R(r)] − ( + 1)R(r) = 0 , R(r) = a r + b r−−1 , (2.36)
dr dr
where we indicate the power law solutions. Equations (2.35) and (2.36) imply that
if we can find eigenfunctions Y of the dimensionless angular momentum operator L ⃗2
satisfying (2.35), then we obtain solutions for the radial equation (2.36) as simple
power laws.
It is useful to observe that the angular equation (2.35) can be further separated
into polar and azimuthal parts by setting Y (θ, ϕ) = V (θ) W (ϕ),
1 ∂ ∂ 1 1 ∂2
sin θ (sin θ V (θ)) + sin2 θ ( + 1) V (θ) = − W (ϕ) .
V (θ) ∂θ ∂θ V (θ) W (ϕ) ∂ϕ2
(2.37)
Again, since θ and ϕ are independent variables, both sides must be equal to a
constant (the second separation constant) which we take to be −m2 and write
∂2
W (ϕ) = −m2 W (ϕ) , Wm (ϕ) = Am eimϕ + Bm e−imϕ . (2.38)
∂ϕ2
Here, Am and Bm are arbitrary integration constants, and m needs to be an integer
in order for the function W (ϕ) to be uniquely defined. So, the remaining equation
in the polar angle θ is given as
∂ ∂
sin θ (sin θ V (θ)) + (sin2 θ ( + 1) − m2 ) V (θ) = 0 . (2.39)
∂θ ∂θ
Substituting u = cos θ and defining F (u) = V (θ), we obtain the associated Legendre
equation,
d d m2
[(1 − u2 ) F (u)] + ( ( + 1) − ) F (u) = 0 . (2.40)
du du 1 − u2
It has two linearly independent solutions called the associated Legendre functions
∣m∣ ∣m∣
of the first, P (u), and second kind, Q (u). The general solution to the polar
equation reads as
F,m (u) = A,m Pm (u) + B,m Qm
(u) , − < m ≤ . (2.41)
For − < m ≤ , with = 0, 1, 2, . . ., the Pm (cos θ) are finite in the interval 0 ≤ θ ≤ π.
By contrast, the Qm (cos θ) diverge at u = cos θ = ±1 or θ = 0, π (i.e., on the “pole
caps” of the unit sphere). While the Qm (cos θ) are possible solutions when we can
manifestly restrict the range of allowed polar angles to ∣ cos θ∣ < 1, we must exclude
these solutions if we seek functions that converge to finite values on the entire unit
sphere. The functions Pm (u) and P−m (u) are not linearly independent, but rather,
directly proportional to each other,
−∣m∣ ( − ∣m∣)! ∣m∣
P (u) = (−1)∣m∣ P (u) . (2.42)
( + ∣m∣)!
Given a solution F,m (u), the integration constants A,m and B,m from Eq. (2.41)
∣m∣ −∣m∣
are thus not uniquely determined, because P (u) and P (u) are not linearly
independent. However,the form of the solution given in Eq. (2.41) is still general.
−∣m∣ ∣m∣
One can make the expansion unique by choosing between P (u) and P (u) de-
∣m∣
pending on the sign of m in the eimϕ phase term, i.e., ei∣m∣ϕ is multiplied by P (u),
−∣m∣
while e−i∣m∣ϕ is multiplied by P (u). This freedom of choice is incorporated into
the definition of the spherical harmonics to be discussed in Eq. (2.53).
We shall assume that the solution Φ = Φ(r, θ, ϕ) of the Laplace equation (2.33)
is well defined on the entire unit sphere and so, we will write the general solution
to the homogeneous equation as
∞
Φ(r, θ, ϕ) = ∑ ∑ (a,m r + b,m r−−1 ) Pm (cos θ) eimϕ , (2.43)
=0 m=−
with arbitrary constants a,m and b,m . We have absorbed the integration constants
A,m from Eq. (2.41) into the a,m and b,m that multiply the radial factors. From
Eq. (2.38), we might otherwise assume that the allowed range for m contains the en-
tire set of the positive and negative integers, and zero; however, the m in Eq. (2.38)
has to be the same as the one in Eq. (2.41), and thus have to fulfill the requirement
that − < m ≤ . For ∣m∣ > , the associated Legendre polynomials vanish because
the (∣m∣ + 1)th derivative annihilates P (cos θ) [see Eq. (2.47a)].
Based on Eq. (2.43), we may proceed to solve the inhomogeneous equation defin-
ing the Green function (2.22) by matching the solution regular at the origin (r )
with the solution regular at infinity (r−−1 ) near the turning point, where the Dirac-δ
function is nonvanishing, in analogy to the derivation outlined in Eqs. (2.16)–(2.19).
However, before we proceed to this endeavor, we first dwell a little on the Legen-
dre functions, and spherical harmonics, which constitute examples of the “special
functions” of mathematical physics.
2.3.3 Legendre Functions and Spherical Harmonics

Let us first discuss the Legendre functions of the first kind. In our problem, we
expect the functions to be well behaved for all ∣x∣ = ∣cos θ∣ ≤ 1. A special case is
found when ∣m∣ = 0. These solutions,
1 d
P0 (x) = P (x) , P (x) = [(x2 − 1)] , (2.44)
2 ! dx
are called Legendre polynomials. They fulfill the recursion relation
( + 1) P+1 (x) = (2 + 1) x P (x) − P−1 (x) . (2.45)
For non-integer , the Legendre polynomials are generalized to Legendre functions,

which in turn are expressible in terms of so-called hypergeometric functions. It is
useful to indicate the first few P (x) which read
P0 (x) = 1 , P1 (x) = x , P2 (x) = 3 2
2
x − 12 , (2.46a)
P3 (x) = 5
2
x − x,
3 3
2
P4 (x) = 35
8
x −
4 15 2
4
x + 38 . (2.46b)
The associated Legendre polynomials Pm (x) of degree and order m are related
to the Legendre polynomials by the relations,
∣m∣ ∣m∣ / 2d∣m∣ P (x)
P (x) = (1 − x2 ) , (2.47a)
dx∣m∣
−∣m∣ ( − ∣m∣)! ∣m∣
P (x) = (−1)∣m∣ P (x) , (2.47b)
( + ∣m∣)!
where the first relation applies to Pm (x) with m ≥ 0, and the second one, which we
have already encountered in Eq. (2.42), is to be used for m < 0. Because P (x) is a
polynomial of order , the associated Legendre polynomials are only non-zero when
≥ ∣m∣. There are orthogonality relations,
1 2 ( + m)!
′ (x) P (x) =
dx Pm m
∫ 2 + 1 ( − m)!
δ ′ . (2.48)
−1

Finally, we note that P0 (1) = 1, P0 (−1) = (−1) , and if m ≠ 0, Pm (±1) = 0. The
general symmetry relation of these functions for x → −x is
+m
Pm (−x) = (−1) Pm (x) . (2.49)
The irregular solutions are the Legendre functions of the second kind, which are
the functions Q (x) that have singularities at x = ±1. They are expressed as follows
(∣x∣ ≤ 1),
P (x) 1+x
Q (x) = ln ( ) − W−1 (x) , (2.50a)
2 1−x
−1
(n + s)! [ψ(n + 1) − ψ(s + 1)]
W−1 (x) = ∑ (x − 1)s , (2.50b)
s=0 2s (n − s)! (s!)2
W−1 (x) = 0 , W0 (x) = 1 , W1 (x) = 3
2
x, W2 (x) = 5
2
x2 − 23 . (2.50c)
Here, ψ(x) = d ln[Γ(x)]/dx is the logarithmic derivative of the Gamma function.
We note the logarithmic divergences of Q (x) for x → ±1. They fulfill a recursion
relation analogous to Eq. (2.45),
( + 1) Q+1 (x) = (2 + 1) x Q (x) − Q−1 (x) , (2.51)
and the associated Legendre functions (x)
Qm of degree and order m are obtained
by relations analogous to (2.47),
∣m∣ ∣m∣ / 2d∣m∣ Q (x)
Q (x) = (1 − x2 ) , (2.52a)
dx∣m∣
−∣m∣ ( − ∣m∣)! ∣m∣
Q (x) = (−1)∣m∣ Q (x) . (2.52b)
( + ∣m∣)!
The logarithmic singularity of the Legendre Q functions near the poles x = cos θ = ±1
implies that we can ignore them in the physically relevant solutions of the Laplace
equation.
In Eq. (2.43), we have already seen that the solutions to the angular equa-
tion (2.35) can be expressed as being proportional to Pm (cos θ) eimϕ . The normal-
ization is canonically chosen to define the spherical harmonics Ym (θ, ϕ) as
B
D 2 + 1 ( − m)! m
m D
Ym (θ, ϕ) = (−1) E P (cos θ) eimϕ . (2.53)
4π ( + m)!
We note the occurrence of the factor (−m)!/(+m)! under the square root. Together
with the prefactor in Eq. (2.47b), it ensures that the case of positive and negative
m are consistently defined. The spherical harmonics Ym (θ, ϕ) are solutions to
Eq. (2.35), which we recall for convenience,
⃗ 2 Ym (θ, ϕ) = ( + 1) Ym (θ, ϕ) ,
L (2.54)
and they have the following property,
⃗ z Ym (θ, ϕ) = −i ∂ Ym (θ, ϕ) = m Ym (θ, ϕ) ,
L (2.55)
∂ϕ
where Lz = −i∂/∂ϕ is the z component of the operator defined in Eq. (2.29) [see
also Eq. (2.30)]. Other useful properties are given by
m ∗
Y −m (θ, ϕ) = (−1) Ym (θ, ϕ) , (2.56a)

Y m (π − θ, 2π − ϕ) = (−1) Y −m (θ, ϕ) , (2.56b)

Y m (π − θ, π + ϕ) = (−1) Y m (θ, ϕ) . (2.56c)
Equation (2.56c) is the parity transformation, which is mediated by the replacement
θ → π − θ and ϕ → ϕ + π. Based on the convention (2.53), we can establish the
orthonormality conditions
2π π
∫ ∫ Y∗′ m′ (θ, ϕ) Ym (θ, ϕ) sin θ dθ dϕ = δ′ δmm′ , (2.57)
0 0
where the integral over the solid angle dΩ = sin θ dθ dϕ defines a scalar product.
We appeal to Eq. (2.48) for the derivation. The scalar product may also also be
written as ⟨Y′ m′ ∣Ym ⟩, if we use Dirac’s bra-ket notation, and we can interpret the
value of Ym (θ, ϕ) as the “matrix element”
Ym (θ, ϕ) = ⟨θ, ϕ∣Ym ⟩ , (2.58)
i.e., as the scalar product of the eigenvector ∣θ, ϕ⟩ of the position on the unit sphere,
and the eigenket ∣Ym ⟩ of the modulus square of the total angular momentum L ⃗ 2,
and the z component Lz . Furthermore, the integration over the solid angle in
Eq. (2.57) can be understood as a summation over the quantum numbers of the
eigenkets ∣θ, ϕ⟩ of the position operator. With this interpretation, Eq. (2.57) can
be written as
⨋ ⟨Y′ m′ ∣θ, ϕ⟩ ⟨θ, ϕ∣Ym ⟩ = δ′ δmm′ , (2.59)
θ,ϕ
The scalar product of two eigenkets ∣θ, ϕ⟩ and ∣θ′ , ϕ′ ⟩ simply is the Dirac-δ function
on the unit sphere,
1
⟨θ′ , ϕ′ ∣θ, ϕ⟩ =
δ (θ − θ′ ) δ (ϕ − ϕ′ ) . (2.60)
sin θ
These eigenkets also form a complete basis for functions defined on the unit sphere,
and we can thus write a completeness relation analogous to Eq. (2.59),
′ ′ ′ ′
∑ ⟨θ, ϕ∣Ym ⟩ ⟨Ym ∣θ , ϕ ⟩ = ⟨θ , ϕ ∣θ, ϕ⟩ . (2.61)
,m
The aim of the detour into Dirac’s bra-ket notation, comprising Eqs. (2.58)–(2.61),
has been to motivate the completeness condition
∞
∗ ′ ′ 1
∑ ∑ Y m (θ, ϕ) Y m (θ , ϕ ) = δ (θ − θ′ ) δ (ϕ − ϕ′ )
=0 m=− sin θ
= δ (ϕ − ϕ′ ) δ (cos θ − cos θ′ ) , (2.62)
which is equivalent to Eq. (2.61). We have used Eq. (1.34). For = 0, the spherical
−1/2
harmonic with m = 0 is only a constant, Y00 (θ, ϕ) = (4π) . For = 1, the spherical
harmonics read as follows,
√
3
Y10 (θ, ϕ) = cos θ , (2.63a)
4π
√ √
3 3
Y11 (θ, ϕ) = − sin θ eiϕ , Y1 −1 (θ, ϕ) = sin θ e−iϕ . (2.63b)
8π 8π
The spherical harmonics with = 2 are of second order in the trigonometric functions
sin θ and cos θ, and Y20 is proportional to P2 (cos θ) = 32 cos2 θ − 12 ,
√
5 3 1
Y20 (θ, ϕ) = ( cos2 θ − ) , (2.64a)
4π 2 2
√ √
15 15
Y21 (θ, ϕ) = − sin θ cos θ eiϕ , Y2 −1 (θ, ϕ) = sin θ cos θ e−iϕ ,
8π 8π
(2.64b)
√ √
15 15
Y22 (θ, ϕ) = sin2 θ e2iϕ , Y2 −2 (θ, ϕ) = sin2 θ e−2iϕ . (2.64c)
32π 32π
These functions are illustrated in Fig. 2.2.
2.3.4 Expansion of the Green Function in Spherical Coordinates

We have found the general solution to the Laplace equation in spherical coordinates,
given in Eq. (2.43),
∞
Φ(r, θ, ϕ) = ∑ ∑ (c,m r + d,m r−−1 ) Ym (θ, ϕ) , (2.65)
=0 m=−
= 0:
= 1:
= 2:
= 3:
Fig. 2.2 We illustrate the shape of the spherical harmonic functions Ym (θ, ϕ) for =
0, 1, 2, 3 by plotting f (θ, ϕ) = ∣Ym (θ, ϕ)∣2 as a function of the polar angle θ and of the
azimuth angle ϕ. The sequence of magnetic components is from m = − to m = .
with arbitrary constants c,m and d,m , which are related to the a,m and b,m used
in Eq. (2.43) by the prefactor given in Eq. (2.53). For convenience, we recall the
Poisson equation (2.22) fulfilled by the Green function, which reads as
1 (3)
⃗ 2 G(⃗
∇ r , r⃗′ ) = − r − r⃗′ ) .
δ (⃗ (2.66)
0
In order to solve this equation, we recall that the Dirac-δ function on the right-
hand side of Eq. (2.66) vanishes everywhere except for the immediate vicinity of
the region where r⃗ − r⃗′ . Hence, we can hope to construct a solution of Eq. (2.66) by
concatenating solutions of the homogeneous equation ∇ ⃗ 2 Φ = 0, given in Eq. (2.43),
using a method previously outlined in Eqs. (2.16)–(2.19).
However, before we engage in this endeavor, we should first clarify the explicit
form of the three-dimensional Dirac-δ function used in Eq. (2.66), in spherical
coordinates. In passing, we shall obtain an a posteriori justification for the form of

the Dirac-δ function on the unit sphere, given in Eq. (2.62). In order to express the
three-dimensional Dirac-δ function in spherical coordinates, we use the ansatz
δ (3) (⃗
r − r⃗′ ) = F (r, θ, ϕ) δ(r − r′ ) δ(θ − θ′ ) δ(ϕ − ϕ′ ) , (2.67)
which is justified because the coordinates r, θ, and ϕ describe the position uniquely,
just like x, y and z. We fix the function F by the condition
1 = ∫ d3 r δ (3) (⃗
r − r⃗′ ) = ∫ dr r2 dθ sin θ dϕ F (r, θ, ϕ) δ(r − r′ ) δ(θ − θ′ ) δ(ϕ − ϕ′ ) .
(2.68)
If this relation is to hold for all r′ , θ′ , and ϕ′ , then we must have
1
F (r, θ, ϕ) = 2 . (2.69)
r sin θ
Finally, the three-dimensional Dirac-δ function is expressed as
1
δ (3) (⃗
r − r⃗′ ) = 2 δ(r − r′ ) δ(θ − θ′ ) δ(ϕ − ϕ′ )
r sin θ
1
= 2 δ(r − r′ ) δ(cos θ − cos θ′ ) δ(ϕ − ϕ′ ) , (2.70)
r
where again, use is made of Eq. (1.34). We now re-examine the defining equation for
the Green function (2.22), which is equivalent to the Poisson equation (2.21) with a
“charge distribution” (in incorrect physical dimension) ρ(⃗ r ) = 0 δ (3) (⃗
r − r⃗′ ) located
′
at the point r⃗ . We recall that the Laplace and Poisson equations are defined in
Eqs. (2.21) and (2.20).
In spherical coordinates and with the convention
r − r⃗′ ) = G(⃗
Gr⃗′ (r, θ, ϕ) ≡ G(⃗ r , r⃗′ ) , (2.71)
we have
1
⃗ 2 G(⃗
∇ ⃗ 2 Gr⃗′ (r, θ, ϕ) = −
r − r⃗′ ) = ∇ δ(r − r′ ) δ(θ − θ′ ) δ(ϕ − ϕ′ ) . (2.72)
0 sin θ r2
The subscript r⃗′ on Gr⃗′ (r, θ, ϕ) is meant to indicate that the Green function G is
proportional to the electrostatic potential generated by a point charge located at
r⃗′ , measured at the point r⃗ whose spherical coordinates are given by r, θ, and ϕ.
Because the angular component of the Laplace operator takes a particularly simple
form when acting on a spherical harmonic, we use the following ansatz for the Green
function,
∞
Gr⃗′ (r, θ, ϕ) = ∑ ∑ fm (r; r′ , θ′ , ϕ′ ) Ym (θ, ϕ) . (2.73)
=0 m=−
Inserting this into the defining equation (2.22), using Eqs. (2.3.2) and (2.35), we
obtain
∂2 2 ∂ ( + 1)
∑( 2 + − ) fm (r; r′ , θ′ , ϕ′ )Ym (θ, ϕ)
m ∂r r ∂r r2
1
=− δ (r − r′ ) δ (cos θ − cos θ′ ) δ (ϕ − ϕ′ ) . (2.74)
0 r 2
One may wonder how it is possible that the right-hand side of this equation in-
cludes Dirac-δ functions in the angular variables while the left-hand side does not.
In fact, a further ansatz for the angular dependence of fm (r; r′ , θ′ , ϕ′ ), given in
Eq. (2.79), implies that the solution for fm (r; r′ , θ′ , ϕ′ ) includes a spherical har-
∗
monic Ym (θ′ , ϕ′ ). The completeness relation (2.62) implies that the summation
over and m generates a term which is equivalent to a Dirac-δ function in the
angular variables, on the left-hand side of Eq. (2.74), corresponding to the equiv-
alent term on the right-hand side. We thus multiply both sides of Eq. (2.74) by
Y∗′ m′ (θ, ϕ), and integrate over the solid angle
1 2π π 2π
∫ dΩ = ∫ d(cos θ) ∫ dϕ = ∫ dθ sin θ ∫ dϕ . (2.75)
−1 0 0 0
Using the orthonormality of the spherical harmonics, we find that
1 ∞ ∂ 2 ∂
∑ ∑ [( r − ( + 1)) fm (r; r′ , θ′ , ϕ′ )] δ ′ δm m′
r2 =0 m=− ∂r ∂r
q 1 π
=− ′ 2
δ (r − r′ ) ∫ d(cos θ) ∫ dϕ δ (ϕ − ϕ′ ) δ (cos θ − cos θ′ ) Y∗′ m′ (θ, ϕ) ,
0 r −1 0
(2.76)
where we use the operator identity
∂2 2 ∂ 1 ∂ 2 ∂
+ = r . (2.77)
∂r2 r ∂r r2 ∂r ∂r
After the θ and ϕ integrations and the use of the Kronecker symbols, we obtain
1 ∂ ∂ 1
[( r2 − ′ (′ + 1)) f′ m′ (r; r′ , θ′ , ϕ′ )] = − δ (r − r′ ) Y∗′ m′ (θ′ , ϕ′ ) .
r 2 ∂r ∂r 0 r ′ 2
(2.78)
It is suggestive to assume that the angular dependence of fm (r; r′ , θ′ , ϕ′ ) is given
∗
by Ym (θ′ , ϕ′ ). We now redefine ′ → and m′ → m and write
fm (r; r′ , θ′ , ϕ′ ) = G (r, r′ ) Ym
∗
(θ′ , ϕ′ ) . (2.79)
Our ansatz (2.73) now looks like
∞
r , r⃗′ ) = Gr⃗′ (r, θ, ϕ) = ∑ ∑ G (r, r′ ) Ym (θ, ϕ) Ym
G(⃗ ∗
(θ′ , ϕ′ ) , (2.80)
=0 m=−
where the radial equation that must be satisfied by the G thus reads
1 ∂ 2 ∂ 1
2
( r − ( + 1)) G (r, r′ ) = − δ (r − r′ ) . (2.81)
r ∂r ∂r 0 r 2
We now solve Eq. (2.81) by the method outlined previously in Eqs. (2.16)–(2.19),
obtaining solutions to the homogeneous equations for r < r′ and r > r′ . Finally, we
render G (r, r′ ) continuous at r = r′ , and satisfy the condition placed on G (r, r′ )
obtained by integrating Eq. (2.81) from r = r′ − ε to r = r′ + ε. In Sec. 2.3.2, we have
already seen that the solution to the homogeneous equation in r has the general
form
G (r, r′ ) = α r + β r−−1 , (2.82)
where α and β may depend on r′ . Assuming regularity of the Green function for
large and small arguments (boundary conditions), we choose the solutions as follows:
For r < r′ , we choose G (r, r′ ) = α r , and for r > r′ , we have G (r, r′ ) = β r−−1 .
Continuity at r = r′ requires that
⎧
⎪ r
⎪
⎪ a ′ +1 , r < r′
r< ⎪
⎪ r
⎪
G (r, r′ ) = a +1 =⎨ (2.83)
r> ⎪
⎪
⎪ r′
⎪
⎪ a r>r ′
⎪
⎩
,
r+1
where r< = min(r, r′ ), and r> = max(r, r′ ). The constant a needs to be determined.
Integrating Eq. (2.81) from r = r′ − ε to r = r′ + ε, after multiplying both sides by
r2 , we obtain
r ′ +ε
∂ ∂ 1 r +ε ′
∫′ (r2 G (r, r′ )) − ( + 1) G (r, r′ )] dr = − ∫

[ δ (r − r′ ) dr .
r −ε ∂r ∂r 0 r −ε
′
(2.84)
′
The second term on the left is continuous at r = r and drops out of the equation
in the limit → 0, and so
∂ ∂ 1
r2 G (r, r′ )∣ − r2 G (r, r′ )∣ =− , (2.85)
∂r ′
r=r +ε ∂r ′
r=r −ε 0
where we ignore higher-order terms in ε. The radial Green function G (r, r′ ) has to
be continuous at the cusp r = r′ , but the derivative may have a discontinuity. We
can thus approximate r2 ≈ r′2 in the prefactors in Eq. (2.85) and write
∂ ∂ 1
r′2 ( G (r, r′ )∣ − G (r, r′ )∣ )=− , (2.86)
∂r r=r ′ +ε ∂r r=r ′ −ε 0
where of course higher-order terms in ε have been neglected. For r = r′ + ε, r is a
little greater than r′ , so r = r> , and we have to use the functional form r−−1 for G ,
whereas for r = r′ − ε, we have r = r< , and we need to use the functional form r for
G . Thus, the condition (2.86) translates into the equation
∂ r′ ∂ r 1
r ′ 2 a ( +1
∣ − ′ +1
∣ )=− , (2.87)
∂r r r=r ′ ∂r r r=r ′ 0
and therefore
r′ r′−1 1
r′ 2 a [−( + 1) − ] = a (−( + 1) − ) = −a (2 + 1) = − . (2.88)
r′+2 r′ +1 0
We finally obtain the prefactor in Eq. (2.83) as
1
a = , (2.89)
0 (2 + 1)
and find that
1 r<
G (r, r′ ) = , r< = min(r, r′ ) , r> = max(r, r′ ) . (2.90)
0 (2 + 1) r>+1
(a) (b)
Fig. 2.3 Panel (a) has a plot of the function f (r1 , r2 ) = r< /r>2 as a function of r1 for
r2 = 0.2 (long dashes), r2 = 0.25 (medium dashes), r2 = 0.3 (short dashes) and r2 = 0.4
(solid line). The maximum is reached at r1 = r2 , with f (r1 , r1 ) = 1/r1 . This implies that
for r2 = 0.2, the maximum (kink) is at f = 5. In panel (b), we have a plot of the function
f (r1 , r2 ) = r< /r>2 in the range 0 < r1 < 0.3 and 0 < r2 < 0.3.
Our ansatz (2.73) with the intermediate result (2.80) and the result (2.90) finally
leads to the rather compact formula
1 ∞ 1 r<
G(⃗r , r⃗′ ) = Gr⃗′ (r, θ, ϕ) = ∑ ∑ Ym (θ, ϕ) Ym ∗
(θ′ , ϕ′ ) , (2.91)
0 =0 m=− 2 + 1 r>+1
confirming Eq. (2.25). On the other hand, we know that
1
r − r⃗′ ) = Gr⃗′ (r, θ, ϕ) =
G(⃗ . (2.92)
r − r⃗′ ∣
4π0 ∣⃗
Combining Eqs. (2.91) and (2.92), we can thus write the formula
∞
1
4π r< ∗
= ∑ ∑ Ym (θ, ϕ) Ym (θ′ , ϕ′ ) , (2.93)
r − r⃗′ ∣ =0 m=− 2 + 1 r>+1
∣⃗
which is the basis for the so-called multipole expansion in electrostatics. This is
a good point to include a specific remark. Namely, the terms in ascending order
of are canonically referred to as the monopole ( = 0), dipole ( = 1), quadrupole
( = 2), octupole ( = 3), and hexadecupole ( = 4) terms. The designation is
inspired by the numerals for the 0, 1, 4, 8 and 16-pole contributions. One may
easily convince oneself that in order to generate a 2 pole term, one has to form
a symmetric arrangement of positive and negative charges ±q, a linear distance d
apart, and a finite 2 pole is obtained upon letting d → 0, keeping the quantity qd
constant.
The case-by-case differentiation according to r< and r> for = 1 is illustrated in
Fig. 2.3. The kink at the turning point r = r′ is a characteristic property of radial
Green Functions.
2.3.5 Multipole Expansion of Charge Distributions

From the multipole expansion of the Green function [see Eq. (2.91)],
1 1 ∞ 1 r<
G(⃗r , r⃗′ ) = = ∑ ∑
∗
Ym (θ, ϕ) Ym (θ′ , ϕ′ ) , (2.94)
r − r⃗′ ∣ 0 =0 m=− 2 + 1 r>+1
4π0 ∣⃗
and the formula

r) = ∫ d3 r′ G(⃗
Φ(⃗ r , r⃗′ ) ρ(⃗
r′ ) , (2.95)
one can derive the multipole expansion of an electrostatic potential. If the charge
distribution is localized and we calculate the potential outside of the localized dis-
tribution (r > r′ ), then the relations r< = r′ and r> = r are simultaneously fulfilled,
and we may express Φ(⃗ r ) as
1 ∞ 1 Ym (θ, ϕ)
Φ (⃗
r) = ∑ ∑ (∫ d3 r′ ρ (⃗
r′ ) r′ Ym
∗
(θ′ , ϕ′ ) ) . (2.96)
0 =0 m=− 2 + 1 rl+1
The integral in round brackets in this expression defines the multipole moments qm
of the charge distribution,
∗
qm = ∫ d3 r ρ (⃗
r ) r Ym (θ, ϕ) . (2.97)
m ∗
Furthermore, because of the relationship Y −m (θ, ϕ) = (−1) Ym (θ, ϕ), it follows
that
m ∗
q −m = (−1) qm . (2.98)
The potential is given as a sum over the multipole moments of the localized charge
distribution, as follows,
1 ∞ qm Ym (θ, ϕ)
Φ (⃗
r) = ∑ ∑ . (2.99)
0 =0 m=− 2 + 1 r+1
The set of elements {q m ; m = −, − + 1, . . . , − 1, } form a spherical tensor of
rank . There are two common representations of tensors: the familiar rectilinear
representation (in Cartesian coordinates) and the spherical representation.
A discussion of two example cases for the multipole decomposition of charge dis-
tributions in spherical coordinates is now in order. The main use of the multipole
decomposition (2.99) lies in the calculation of potentials outside the generating
charge distributions which are centered around the origin; i.e., one assumes that
the coordinate r′ always is the lesser radial coordinate as compared to the obser-
vation point r. In consequence, results obtained using Eq. (2.99) are valid only for
r > R, where R is a radial variable that characterizes the extension of the charge
distribution.
As a first example (see Fig. 2.4), let us consider a charge distribution
√
′ Q ′ ′ ′ 4π
ρ(⃗r ) = 2 cos θ δ(r − a) , cos θ = Y10 (θ′ , ϕ′ ) , (2.100)
a 3
where we have identified the cosine as a spherical harmonic. This charge distribu-
tion describes a dipole-shaped charge distribution centered on a sphere. Because
of the orthogonality of the spherical harmonics, the only nonvanishing multipole
component is
√
3 ′ ′ ′ ∗ ′ ′ 4π
q10 = ∫ d r ρ (⃗
r ) r Y10 (θ , ϕ ) = Q a, (2.101)
3
Q
Fig. 2.4 r′ ) =
The left plot shows a sphere with a charge distribution ρ(⃗ cos θ ′ δ(r ′ −a),
a2
proportional to cos θ ′ . Positively charged areas are lighter than negatively charged areas.
Q
The right plot displays a sphere with a charge distribution ρ(⃗ r ′ ) = 2 sin2 θ ′ δ(r ′ − a).
a
which has the correct physical dimension (charge × length) of a dipole moment. In
terms of the multipole moments of the charge distribution, the potential for r > a
is found to be
√
1 ∞ qm Ym (θ, ϕ) 1 q10 Y10 (θ, ϕ) q10 3
Φ (⃗
r) = ∑ ∑ = = cos θ ,
0 =0 m=− 2 + 1 r +1 0 3 r 2 30 r 2 4π
(2.102)
and thus, in view of Eq. (2.101),
Qa
r) =
Φ(⃗ cos θ . (2.103)
30 r2
As a second example (see also Fig. 2.4), we consider the charge distribution
Q
r) =
ρ(⃗ sin2 θ δ(r − a) . (2.104)
a2
In this case, the potential involves a monopole and a quadrupole term. Calculating
the moments q00 and q20 , we may convince ourselves that the dipole moments
vanish, and we obtain the result as a sum of a monopole and a quadrupole term,
2Q Q a2
r) =
Φ(⃗ − (3 cos2 θ − 1) . (2.105)
3 0 r 15 0 r3
2.3.6 Addition Theorem for Spherical Harmonics

Suppose that in the expansion (2.93), which we recall for convenience,
∞
1
4π r< ∗
= ∑ ∑ Ym (θ, ϕ) Ym (θ′ , ϕ′ ) , (2.106)
r − r⃗′ ∣ =0 m=− 2 + 1 r>+1
∣⃗
we let r⃗′ = r′ êz , which is tantamount to letting θ′ = 0. Only the m = 0 terms will
appear in the sum because Ym (0, ϕ′ ) vanishes for m ≠ 0. This holds because the
associated Legendre polynomials Pm (cos θ = 1) vanish except for the case m = 0,
where P0 (cos θ) = P (cos θ). In this case, we use the expression for Ym (θ, ϕ). In
view of the equality P (1) = P0 (1) = 1, we have
∞
1 4π r< ∗
= ∑ Y0 (θ, ϕ) Y0 (0, ϕ′ )
r − r′ êz ∣ =0 2 + 1 r>+1
∣⃗
√ √
∞ ∞
4π r< 2 + 1 2 + 1 r<
= ∑ P (cos θ) P (1) = ∑ P (cos θ) .
=0 2 + 1 r>
+1 4π 4π +1
=0 r>
(2.107)
However, we also have the relation
1 1
′
=√ , (2.108)
∣⃗
r − r⃗ ∣ r + r − 2 r r′ cos θ
2 ′2
where θ is the polar angle of r⃗, which is equal to the angle ∠(⃗ r, r⃗′ ) between the
unit vector r̂ and r̂′ because r⃗′ = r′ êz . We can now use the derived relations in two
ways. First, we observe that we have chosen r⃗′ to lie along the z axis somewhat
arbitrarily. We can thus conclude that, more generally,
∞
1 1 r<
= √ = ∑ r , r⃗′ )) ,
P (cos ∠(⃗ (2.109)
r − r⃗′ ∣
∣⃗ ′2 ′
r + r − 2 r r cos ∠(⃗
2 ′
r, r⃗ ) =0 >
r +1
where cos ∠(⃗ r , r⃗′ ) = r̂ ⋅ r̂′ . Here, r̂ is the unit vector pointing in the direction of
r⃗. The angle γ connects the position vector locating the charge, r⃗′ , and the vector
locating the observation point, r⃗. Second, the original multipole decomposition
must still hold for the expression 1/∣⃗ r − r⃗′ ∣,
∞
1
4π r< ∗
′
=∑ ∑ Ym (θ, ϕ) Ym (θ′ , ϕ′ ) . (2.110)
∣⃗
r − r⃗ ∣ =0 m=− 2 + 1 r>+1
The equality of (2.109) and (2.110) has to hold for each component separately,
and we finally obtain the addition theorem for spherical harmonics, which reads

4π
P (cos ∠(r̂, r̂′ )) = ∗ ′ ′
∑ Ym (θ, ϕ) Ym (θ , ϕ ) . (2.111)
2 + 1 m=−
By expressing all direction cosines in spherical coordinates, we also obtain
cos ∠(r̂, r̂′ ) = r̂ ⋅ r̂′ = sin θ sin θ′ cos (ϕ − ϕ′ ) + cos θ cos θ′ . (2.112)
2.3.7 Multipole Expansion in Cartesian Coordinates

The multipole decomposition, in Cartesian coordinates, is based on the expansion
1 1 r⃗ ⋅ r⃗′ 3(⃗
r ⋅ r⃗′ )2 − r2 r′2
= + + + O(r−4 ) . (2.113)
r − r⃗′ ∣ r
∣⃗ r3 2r5
This leads to the following expansion for the potential,

1 3 ′ 1
r) =
Φ(⃗ ∫ d r r′ )
ρ(⃗
4π0 r − r⃗′ ∣
∣⃗
1 3 ′ 1 r⃗ ⋅ r⃗′ 3(⃗
r ⋅ r⃗′ )2 − r2 r′2
≈ ∫ d r ( + 3 + r′ )
+ . . . ) ρ(⃗
4π0 r r 2r5
Q r⃗ ⋅ p⃗ 1 Qij (3ri rj − δij r2 )

≈ + + + ... , (2.114)
4π0 r 4π0 r3 4π0 2r5
where the total charge Q, the dipole moment vector p⃗, and the components of the
quadrupole tensor Qij are given as
Q = ∫ d3 r′ ρ(⃗
r′ ) , (2.115a)
p⃗ = ∫ d3 r′ r⃗′ ρ(⃗
r′ ) , (2.115b)
Qij = ∫ d3 r′ (3ri′ rj′ − δij r⃗′2 ) ρ(⃗

r′ ) . (2.115c)
Let us focus on the dipole term, which is given by the term with = 1 in Eq. (2.99),
1 1
1 Ym (θ, ϕ)
Φ (⃗
r )∣=1 ∼ ∑ 2
(∫ d3 r′ r′ ρ (⃗
r′ ) Ym
∗
(θ′ , ϕ′ )) , (2.116)
0 m=−1 3 r
where we have already inserted the definition of the multipole moment according
to Eq. (2.97). The addition theorem (2.111) implies that for = 1,
1
4π
r, r⃗′ )) = r̂ ⋅ r̂′ =
P1 (cos ∠(⃗ ∗ ′ ′
∑ Y1m (θ, ϕ) Y1m (θ , ϕ ) . (2.117)
3 m=−1
Inserting this relationship into Eq. (2.116), one obtains

1 3
r )∣=1 ∼
Φ (⃗ 2
(∫ d3 r′ r̂ ⋅ r̂′ r′ ρ (⃗
r′ ))
3 0 r 4π
1 1 r⃗ ⋅ p⃗
∼ (∫ d3 r′ r⃗ ⋅ r⃗′ ρ (⃗
r′ )) = . (2.118)
4π0 r3 4π0 r3
Equations (2.116) and (2.118) express the conversion of the dipole term from spher-
ical to Cartesian coordinates.
It remains to clarify the relation of the components of the dipole moment vec-
tor, in the Cartesian versus the spherical representation. Indeed, in the spherical
representation, we find that
√ √
3 3
q1±1 = ∓ (p1 ∓ i p2 ) , q10 = p3 . (2.119)
8π 4π
We here appeal to the fact that the multipole moments are defined with the complex
conjugate of the spherical harmonic entering the integrand, according to Eq. (2.97).
The quadrupole term, by contrast, is given by the term with = 2 in Eq. (2.99),
1 2
1 Y2m (θ, ϕ)
Φ (⃗
r )∣=2 ∼ ∑ (∫ d3 r′ r′2 ρ (⃗
r′ ) Ym
∗
(θ′ , ϕ′ )) . (2.120)
0 m=−2 5 r3
The addition theorem implies that
2
1 4π
r , r⃗′ )) =
P2 (cos ∠(⃗ (3(r̂ ⋅ r̂′ )2 − 1) = ∗ ′ ′
∑ Y2m (θ, ϕ) Y2m (θ , ϕ ) . (2.121)
2 5 m=−2
Inserting this relationship into Eq. (2.120), one obtains

1 1 5 1
r )∣=2 ∼
Φ (⃗ (∫ d3 r′ r′2 ρ (⃗
r′ ) [3(r̂ ⋅ r̂′ )2 − 1])
0 5 4π 2
1 1
= (∫ d3 r′ r′2 ρ (⃗
r′ ) (3(⃗
r ⋅ r⃗′ )2 − r⃗ 2 r⃗′2 ))
4π0 2r5
1 Qij (3ri rj − δij r⃗ 2 )
= . (2.122)
4π0 2r5
Equations (2.120) and (2.122) variously express the quadrupole term from the
charge distribution, in spherical versus Cartesian coordinates.
Let us count. We have five components of the quadrupole tensor q2m with
m = −2, . . . , 2. The quadrupole tensor in Cartesian coordinates has the form
⎛ Q11 Q12 Q13 ⎞ ⎛ Q11 Q12 Q13 ⎞

(Qij ) = ⎜ Q21 Q22 Q32 ⎟ = ⎜ Q12 Q22 Q32 ⎟, (2.123)
⎝ Q31 Q32 Q33 ⎠ ⎝ Q13 Q2l −Q11 − Q22 ⎠
where we have used the symmetry Qij = Qji and the tracelessness of the quadrupole
tensor ∑3i=1 Qii = 0 which holds in view of ∑3i=1 δii = 3. Thus, because the quadrupole
moment of the charge distribution is a symmetric, traceless, second-rank tensor, it
is therefore specified by five parameters, as the second form in Eq. (2.123) suggests.
The simplest relationship occurs for q20 and Q33 . This simplicity is obtained
because the spherical tensors we are using have taken the z axis as a preferred axis,
Q33 = ∫ (3 z 2 − r2 ) ρ (⃗
r ) d3 r = ∫ (3 cos2 θ − 1) r2 ρ (⃗
r ) d3 r
√ √
16π 5
= ∫ (3 cos2 θ − 1) r2 ρ (⃗ r) d3 r
5 16π
√ √
π π
=4 ∫ 20Y (θ, ϕ) r 2
ρ (⃗
r ) d3
r = 4 q20 . (2.124)
5 5
2.3.8 Multipole Expansion in an External Field

Up to now, we have considered the multipole decomposition in spherical coordi-
nates, and we have considered the original charge distribution to be centered about
the point r⃗′ ≈ 0, with a distant observation point r⃗. An expansion of 1/∣⃗ r − r⃗′ ∣
into multipoles (spherical coordinates) then leads to the multipole decomposition.
The total charge of the charge distribution that generates the potential at r⃗ is the
monopole.
In order to switch the formalism of the multipole decomposition to Cartesian
coordinates, we consider an opposite situation, in some sense. A charge distribution
is centered about the point r⃗ = ⃗0, with the potential being generated by a distant
charge distribution that is centered about the point r⃗′ ≠ ⃗ 0. The distant charge
distribution generates a potential Φ(⃗r ) which interacts with the charge distribution
ρ(⃗r ). The potential energy of the configuration is given by the overlap integral
r ) Φext (⃗
W = ∫ ρ (⃗ r ) d3 r , (2.125)
1 3 ′ 1
Φext (⃗
r) = ∫ d r r′ ) ,
ρext (⃗ (2.126)
4π0 r − r⃗′ ∣
∣⃗
where ρext (⃗r′ ) is the “external” charge distribution, which is located at a large
distance from the origin, while the “probe” charge distribution ρ (⃗
r ) is located near
r⃗ ≈ ⃗
0. Hence, in particular,
3 3
∂ 2 Φext (⃗
r)
⃗ 2 Φext (⃗
∇ r )∣r⃗=0 = ∑ ∑ δij ∣ = 0, (2.127)
i=1 j=1 ∂ri ∂rj r⃗=0
because the potential Φext (⃗r ) is supposed to be due to charges external to the
volume in which ρ(⃗r) is nonzero.
Because the sources of the potential are external to the volume containing the
probe charge, the electric potential can be expanded in a Taylor series,
∂Φext (⃗
3
r′ ) 1 3 3 ∂ 2 Φext (⃗ r′ )
Φext (⃗
r ) = Φext (0) + ∑ ri ∣ + ∑ ∑ ri rj ∣ + ... .
i=1 ∂ri′ r⃗′ =⃗
0 2 i=1 j=1 ∂ri′ ∂rj′ r⃗′ =⃗
0
(2.128)
Expressing the gradient of the external potential in terms of the electric field,
⃗ ⃗ r′ )
∂Φ (⃗
E(0) = − ∣ , (2.129)
∂ri′ r⃗′ =⃗0
we obtain the expansion of the potential energy in Cartesian coordinates,
W = Φ(⃗
0) ∫ d3 r ρ (⃗ ⃗ ⃗
r ) − E(0) ⋅ ∫ d3 r r⃗ ρ (⃗
r)
1 3 3 ∂ 2 Φ (⃗ r′ )
+ ∑∑ ∣ ∫ d r ri rj ρ (⃗
3
r) + . . .
2 i=1 j=1 ∂ri′ ∂rj′ r⃗′ =⃗0
⃗ ⃗ 1 3 3 ∂ 2 Φ (⃗ r)
= Φ(⃗
0) Q − E(0) ⋅ p⃗ + ∑ ∑ ∣ Qij + . . . . (2.130)
6 i=1 j=1 ∂ri ∂rj r⃗=⃗0
The first term is a kind of average potential energy, the second is the interaction
of the electric dipole distribution with the applied electric field, and the third term
gives the interaction of the quadrupole moment of the charge distribution with
the gradient of the electric field. The formula for the total charge Q, the dipole
moment vector p⃗, and the components of the quadrupole tensor Qij , have been
given in Eq. (2.115). Equation (2.127) has been used in order to eliminate the term
with i = j in Eq. (2.115c).
2.4 Electrostatic Green Function and Cylindrical Coordinates
2.4.1 Laplace Equation in Cylindrical Coordinates

Up to now, we have analyzed the Green function in spherical coordinates, and ob-
tained a result which involves a case-by-case differentiation in terms of the mapping
(r, r′ ) → (r< = min(r, r′ ), r> = min(r, r′ )). It is very interesting to generalize the
formalism to cylindrical coordinates. In analogy with Eq. (2.26), let us first define
cylindrical coordinates as follows,
x = ρ cos ϕ , y = ρ sin ϕ , z = z, (2.131)
with the backtransformation [cf. Eq. (2.26b)] into Cartesian coordinates,
√ ⎛ x ⎞
ρ= x2 + y 2 , ϕ = arccos √ , z =z. (2.132)
⎝ x2 + y 2 ⎠
The Laplacian reads as
1 ∂ ∂ 1 ∂2 ∂2
⃗2 =
∇ ρ + 2 + . (2.133)
ρ ∂ρ ∂ρ ρ ∂ϕ2 ∂z 2
The homogeneous equation [cf. (2.33)] is
⃗ 2 Φ(ρ, ϕ, z) = 0 .
∇ (2.134)
As in the case of spherical coordinates, this equation is solved by a separation of
variables,
Φ (ρ, ϕ, z) = U (ρ) V (ϕ) W (z) . (2.135)
Substituting this ansatz into Eq. (2.133), we can separate all variable dependence:
1 1 d d 1 1 d2 1 d2
ρ U+ V + W = 0. (2.136)
U ρ dρ dρ V ρ2 dϕ2 W dz 2
The integration constants may be defined as
1 d2 1 d2
2
V = −m2 , W = k2 . (2.137)
V dϕ W dz 2
Thus, W (z) can be written as a linear combination of two solutions,
W (z) = Ak ekz + Bk e−kz , (2.138)
which are regular at z = −∞ and at z = +∞, respectively (provided k is positive).
However, if the z domain is confined to a finite interval, then k may be complex.
The dependence on the azimuthal angle is given by
V (ϕ) = Cm eimϕ + Dm e−imϕ . (2.139)
Again, m has to be an integer, because otherwise the function V (ϕ) is not uniquely
defined. Using Eq. (2.137) in Eq. (2.136), one may show that the equation fulfilled
by U is [see Eq. (3.176)]
d2 d
(ρ2 2
+ρ + k 2 ρ2 − m2 ) U (ρ) = 0 . (2.140)
dρ dρ
This equation is solved by
U (ρ) = am Jm (kρ) + bm Ym (kρ) , (2.141)
where the Jm and Ym are called Bessel functions of the first and second kind,
respectively.
The regular solution (at ρ = 0) is given by the Bessel J functions, while the
Bessel Y function (Neumann function) diverges for ρ → 0. The general solution to
Eq. (2.134) is thus given as
∞
Φ (ρ, ϕ, z) = ∑ ∑ [ak,m Jm (kρ) + bk,m Ym (kρ)] ekz eimϕ , (2.142)
k m=−∞
where the expansion coefficients are ak,m and bk,m . The sum over k entails all per-
missible values (including negative ones, as determined by the boundary conditions).
It would thus be redundant to indicate both terms exp(k z) and exp(−k z) in the
general solution; they are contained in the sum over k. Typically, the system fills
the region from ϕ = 0 to ϕ = 2π. As already stated, the continuity of V (ϕ) requires
that m be an integer. The values of k are determined by boundary conditions. If
we require the solutions to be periodic in z, then k must be purely imaginary. If
the solutions are exponential in z, then k is purely real. Also, the sum over k may
be discrete, or continuous, in which case one has to integrate over k.
If k is continuous and the allowed k values are the real numbers, and the solution
is required to be regular at the origin, then it can be written as
∞ ∞
Φ (ρ, ϕ, z) = ∑ ∫ dk (a(k, m) Jm (kρ) ekz eimϕ
m=−∞ 0
+ b(k, m) Jm (kρ) e−kz eimϕ ) , (2.143)

where the discrete subscripts become continuous functions of their arguments and
we have denoted the expansion coefficient for positive versus negative k by a and
b, respectively. This form will be used later in Eq. (2.184). However, before we
get to this point in our discussion, we shall first discuss a number of properties
of the Bessel functions, which belong to the most important special functions of
mathematical physics.
2.4.2 Cylindrical Coordinates and Bessel Functions

In the case of the representation of the Green function in the spherical basis, the
general solution to the homogeneous equation involves Legendre polynomials and
spherical harmonics [see Eq. (2.65)]. We recall that we explored properties of the
Legendre polynomials and spherical harmonics in Sec. 2.3.3, before discussing the
Green function in spherical coordinates, in Sec. 2.3.4. Let us attempt the same
procedure in the cylindrical basis, where the general solution to the homogeneous
equation involves Bessel1 and Neumann2 functions [see Eq. (2.142)].
Indeed, Bessel’s differential equation, which is fulfilled by the Bessel and Neu-
mann functions, reads [see Eq. (10.2.1) of Ref. [5]],
∂2 ∂
(z 2 2
+z + z 2 − ν 2 ) Jν (z) = 0 . (2.144)
∂z ∂z
Here, ν is not necessarily integer-valued (in contrast to m), and the argument z
may be complex. The Bessel J function may be defined via the series expansion
[see Eq. (10.2.2) of Ref. [5]],
∞
(−1)j z 2j+ν
Jν (z) = ∑ ( ) , z∈ . (2.145)
j=0 j! Γ(ν + j + 1) 2
which converges for all z ∈ . The Gamma function is defined as [see Eq. (5.2.1) of
Ref. [5]],
∞
Γ(x) = ∫ dt tx−1 e−t , x > 0. (2.146)
0
For complex argument, the Gamma function has to be defined by a complex contour
integral (see Ref. [6] for a comprehensive discussion). For integer argument, the
Gamma function has the property [see Eq. (5.4.1) of Ref. [5]],
Γ(m + 1) = m! , m∈ . (2.147)
The complementary solution Yν is defined as

1
Yν (z) = lim [ (Jα (z) cos(απ) − J−α (z))] , (2.148)
α→ν sin(απ)
where the limit is nontrivial if ν is an integer. The Neumann function Yν (Bessel
function of the second kind), which is sometimes denoted as Nν , is fundamentally
different from the spherical harmonic Ym . The notation Yν has found its way
into the literature and is currently adopted universally (including many current
computer algebra systems), and the symbol Nν is used only very rarely.
The recurrence relations for the J functions (analogous relations are fulfilled by
the Y ’s) read as follows [see Eq. (10.6.1) of Ref. [5]],
2ν
Jν−1 (z) + Jν+1 (z) = Jν (z) , (2.149a)
z
Jν−1 (z) − Jν+1 (z) = 2Jν′ (z) . (2.149b)
1
Friedrich Wilhelm Bessel (1784–1846)
2
Carl Gottfried Neumann (1832–1925)
The latter equation expresses the derivative of the Bessel function of order ν in
terms of Bessel functions of order ν − 1 and ν + 1. Alternatively, one has
ν ν
Jν′ (z) = Jν−1 (z) − Jν (z) = −Jν+1 (z) + Jν (z) . (2.150)
z z
For non-integer order ν and complex argument z with Re(z) > 0, an integral
representation of the Bessel J function is [see Eq. (10.9.6) of Ref. [5]]
π
dθ ∞ dθ
Jν (z) = ∫ cos [z sin(θ) − νθ] − sin(νπ) ∫ e−z sinh(θ)−ν θ , (2.151)
π 0 π
0
while under the same conditions, the Bessel Y function has the integral representa-
tion [see Eq. (10.9.7) of Ref. [5]]
π ∞
dθ dθ −z sinh(t) νt −νt
Yν (z) = ∫ sin [z sin(θ) − νθ] − ∫ e (e + e cos(νπ)) . (2.152)
π π
0 0
The asymptotic behavior for small argument of the Bessel J and Y functions is
given as
1 z ν Γ (ν) 2 ν
Jν (z) ∼ ( ) , Yν (z) ∼ − ( ) , ∣z∣ → 0 . (2.153)
Γ (ν + 1) 2 π z
The asymptotic formula is for ν > 0 for the Y function. For ν < 0, one has to
consider a second term,
cos(π ν) Γ(−ν) z ν Γ (ν) 2 ν
Yν (z) ∼ − ( ) − ( ) , ∣z∣ → 0 . (2.154)
π 2 π z
In the limit ν → 0, the asymptotic formula reduces to
2 z 2
ln ( ) + γE + O (z 2 ln(z)) ,
Y0 (z) ∼ ∣z∣ → 0 . (2.155)
π 2 π
Here, γE = 0.577 215 66 . . . is the Euler–Mascheroni constant. For ν = −n with
positive integer n, one has
(n − 1)! 2 n
Y−n (z) ∼ −(−1)n
( ) , ∣z∣ → 0 . (2.156)
π z
For large arguments x > 0, the asymptotic behavior of the Bessel functions is
given by a damped harmonic oscillation, with a nontrivial phase,
√
2 νπ π
Jν (x) ∼ cos (x − − ), x → ∞. (2.157a)
πx 2 4
√
2 νπ π
Yν (x) ∼ sin (x − − ), x → ∞. (2.157b)
πx 2 4
For intermediate values of x, the situation is more complicated. This is illustrated in
Fig. 2.5 for J0 (x) and J1 (x). Otherwise, it is helpful to know that the literature on
Bessel functions is abundant, with basic formulas being summarized in Refs. [7, 5]
and more intricate treatments in Refs. [8, 9].
J0x J1x
0.8
1.0
0.4
0.5
0 x
0 x 3 6 9 12
4 8 12
0.4
0.5
(a) (b)
Fig. 2.5 Plot of the first oscillations of the Bessel functions J0 (x) and J1 (x) [panels
(a) and (b), respectively]. For x → 0, one discerns that limx→0 J1 (x) = 0 and the linear
asymptotics for J1 (x) in the regime of small x.
2.4.3 Orthogonality Properties of Bessel Functions

While this is a textbook on physics, not on the theory of special functions, a cer-
tain familiarity with concepts related to the most important special functions of
mathematical physics is of universal importance for any theoretical physicist, and
the slight detour we are taking now will prove useful for a number of applications.
Indeed, of particular importance are the zeros of the Bessel and Neumann functions.
Let us denote the zeros of the Bessel and Neumann functions of integer order as
xmn and ymn , respectively. These fulfill the relations
Jm (xmn ) = 0 (2.158)
and Ym (ymn ) = 0, where xmn is the nth zero of the mth Bessel J function, and ymn is
the nth zero of the mth Neumann Y function. In particular, because Jm (x) ∝ xm
for small x, a trivial zero of Jm (x) for m ≥ 1 is x = 0. Furthermore, in view
of Eq. (2.157), there exists an infinite number of zeros of the Bessel functions on
the real axis. However, it is rather nontrivial to calculate these numerically, for
intermediate ranges of the argument. Selected numerical values for xmn are given
in Table 2.1. We also indicate numerical values for the zeros of derivative of the
Bessel functions, defined according to
′
Jm (xmn ) = 0 (2.159)
In computer algebra systems (e.g., Ref. [10]), one sometimes finds built-in algo-
rithms like BesselJZero or BesselYZero for the computation of zeros of Bessel
and Neumann functions; these can be used directly without recourse to any addi-
tional asymptotic formulas. Still, it is useful to know which asymptotic properties
determine the behavior of the roots of the Bessel functions in large order. A mod-
ern treatise on the intricate aspects of the calculation of Bessel functions, with
numerical examples in extreme parameter ranges, is given in [11].
Let now investigate whether the basis functions in our ansatz (2.142) for a
general solution of the Laplace equation in cylindrical coordinates can be interpreted
as orthogonal with respect to an appropriate integration measure. For a potential
Table 2.1 Numerical values for the nth zeros xmn of the Bessel
functions of order m, with Jm (x = xmn ) = 0, for m = 0, . . . , 5 and
n = 1, . . . , 5.
n m=0 m=1 m=2 m=3 m=4 m=5
1 2.405 3.832 5.135 6.379 7.588 8.771
2 5.520 7.016 8.417 9.761 11.065 12.339
3 8.654 10.173 11.620 13.015 14.373 15.700
4 11.792 13.323 14.796 16.224 17.616 18.980
5 14.931 16.470 17.960 19.409 20.827 22.218
Table 2.2 Numerical values for the nth zeros xmn of the deriva-
′
tives of the Bessel functions of order m, with Jm (x = xmn ) = 0,
for m = 0, . . . , 5 and n = 1, . . . , 5.
n m=0 m=1 m=2 m=3 m=4 m=5
1 3.832 1.841 3.054 4.201 5.317 6.415
2 7.016 5.331 6.706 8.015 9.282 10.520
3 10.174 8.536 9.969 11.346 12.682 13.987
4 13.324 11.706 13.170 14.585 13.540 17.313
5 16.471 14.864 16.348 17.789 14.970 20.576
regular on the cylinder axis, with ρ = 0, we must concentrate on the terms with J
functions. Furthermore, the zeros of the Bessel functions play a special role in the
orthogonality properties. Based on Eq. (2.144), for an arbitrary scale parameter λ,
it is straightforward to see that the function Jn (λ x) fulfills
d d
x [x Jn (λ x)] + (λ2 x2 − n2 ) Jn (λ x) = 0 . (2.160)
dx dx
Now, we assume that λ and μ are two distinct zeros of the Bessel function Jn (x),
Jn (λ) = Jn (μ) = 0 , λ ≠ μ, (2.161)
so that Jn (λ x) = Jn (μ x) = 0 for x = 1. Multiplying Eq. (2.160) by Jn (μ x)/x, we

obtain
d d λ2 x2 − n2
Jn (μ x) [x Jn (λ x)] + Jn (μ x) Jn (λ x) = 0 , (2.162)
dx dx x
and a trivial generalization is obtained by the replacement μ ↔ λ,
d d μ2 x2 − n2
Jn (λ x) [x Jn (μ x)] + Jn (μ x) Jn (λ x) = 0 . (2.163)
dx dx x
Furthermore, manipulating Eqs. (2.162) and (2.163), one can show that a few terms
cancel in the difference,
d d d d
Jn (μ x) [x Jn (λ x)] − Jn (λ x) [x Jn (μ x)]
dx dx dx dx
+ (λ2 − μ2 ) xJn (μ x) Jn (λ x) = 0 , (2.164)
which can be rewritten as
d d d d
[Jn (μ x) x Jn (λ x)] − [Jn (λ x) x Jn (μ x)]
dx dx dx dx
+ (λ2 − μ2 ) xJn (μ x) Jn (λ x) = 0 . (2.165)
Finally, one integrates over x ∈ (0, 1) to show the orthogonality
1
∫ dx x Jn (μ x) Jn (λ x) = 0 , λ ≠ μ, Jn (λ) = Jn (μ) = 0 , (2.166)
0
which concerns integrations up to specific zeros λ and μ, with λ ≠ μ.

The case λ = μ in Eq. (2.166) must be treated separately. First, we sidestep the
question and derive a helpful identity. From the recursion relations of the Bessel
function and its derivative, given in Eq. (2.150), one can conclude that, at a zero
x = λ of the Bessel function,
d
Jn (x)∣ = −Jn+1 (λ) . (2.167)
dx x=λ
In Eq. (2.165), one now sets μ = λ + , i.e., one slightly displaces one parameter from
the zero, integrates over x ∈ (0, 1) and obtains
1
[Jn ((λ + )) λJn′ (λ)] + (λ2 − (λ + )2 ) ∫ dx x Jn ((λ + ) x) Jn (λ x) = 0 . (2.168)
0
One then expands to first order in to obtain

1
λ [Jn′ (λ)] − 2 λ ∫
2 2
dx x [Jn (λ x)] = 0 , (2.169)
0
and thus shows that

1 1 2 2
∫ dx x [Jn (λ x)] =
[Jn+1 (λ)] . (2.170)
0 2
Equations (2.166) and (2.170) allow us to write
1 1 2
∫ dx x Jn (λ x) Jn (μ x) =
δλμ [Jn+1 (λ)] , (2.171)
0 2
where the Kronecker symbol is unity if the two (in this case, real rather than integer)
arguments are equal to each other and zero otherwise. Let now xmn = kmn a where
a is the radius of a cylinder. Then, a simple scaling of the integration variable in
Eq. (2.171) leads to the result (see Fig. 2.6)
a 1 2
∫ dρ ρ Jm (kmn′ ρ) Jm (kmn ρ) = δn′ n [a Jm+1 (kmn a)] . (2.172)
0 2
J0k0n x J1k1n x
1.0 0.5
0.25
0.5
0 x
0.25 0.5 0.75 1.0
0 x 0.25
0.25 0.5 0.75 1.0
0.5
0.5
(a) (b)
Fig. 2.6 Illustration of the property (2.172) for the Bessel functions fn (ρ) = J0 (k0n ρ)
and fn (ρ) = J1 (k1n ρ), with n = 1, 2 (long dashes are for n = 1, short dashes denote the
curves for n = 2). We set a = 1, so that k0n = x0n and k1n = x1n . As n is increased, the
number of oscillations increases, and the Bessel functions scaled with different x0n and
x1n are orthogonal to each other upon integration with respect to the measure given in
Eq. (2.172).
Now, the orthogonality and normalization has should be generalized for an inte-
grations over the real axis, x ∈ (0, ∞). One starts once more from Eq. (2.165), but
with the replacements λ → k, and μ → k ′ ,
d d d d
[Jn (k ′ x) x Jn (k x)] − [Jn (k x) x Jn (k ′ x)]
dx dx dx dx
+ (k 2 − k ′2 ) xJn (k x) Jn (k ′ x) = 0 . (2.173)
The well-known asymptotics of the Bessel function given in Eq. (2.157) imply that
ρ = ∞ is a zero of the Bessel function. Thus, integrating Eq. (2.173) within the
interval x ∈ (0, ∞), one easily shows that
∞
∫ dx x Jn (k x) Jn (k ′ x) = 0 , k ≠ k′ . (2.174)
0
The limit k → k ′ remains to be evaluated. The only region which can size-
ably contribute to the integral in this limit is the one for very large x; otherwise
an infinitesimal displacement k = k ′ + will lead to a vanishing integral due to
orthogonality. One writes the asymptotics (2.157) as an exponential,
√
1 2 (n + 12 ) π
Jn (k x) ∼ {exp [i (k x − )]
2 πkx 2
(n + 12 ) π
+ exp [−i (k x − )]} , x → ∞, (2.175)
2
and analogously for k replaced by k ′ . The only relevant terms in the integrand
x Jn (k x) Jn (k ′ x) in Eq. (2.174) is given by the following replacement,
1
x Jn (k x) Jn (k ′ x) → √ {exp (i (k − k ′ ) x) + exp (−i (k − k ′ ) x)} + . . . ,
2π k k′
(2.176)
where the ellipsis denotes terms proportional to exp (i (k + k ′ ) x) and

exp (−i (k + k ′ ) x). When integrated over the interval x ∈ (0, ∞), with the help of a
convergent factor exp(−η x), these terms give only finite corrections to the Dirac-δ
function term proportional to δ(k − k ′ ) which dominates in the limit k → k ′ . As a
last step, one symmetrize the the expression given in Eq. (2.176) for the interval
x ∈ (−∞, ∞) to show that
∞ 1 ∞ 1
k→k′ i(k−k′ ) x ′
∫ dx x Jn (k x) Jn (k ′ x) = ∫ dx √ ′ (e + e−i(k−k ) x )
0 2 −∞ 2π k k
1 1
= √ 2πδ(k − k ′ ) = δ(k − k ′ ) , (2.177)
2π k k ′ k
where we have used the fact that k = k ′ at the peak of the Dirac-δ function. Equa-
tions (2.174) and (2.177) conclude the proof of the orthogonality relation
∞ 1
∫ dρ ρ Jn (k ρ) Jn (k ′ ρ) =
δ (k − k ′ ) . (2.178)
0 k
For spherical Bessel functions, which are defined via the relation
√
π
j (x) = J+1/2 (x) , (2.179)
2x
the orthogonality relations are somewhat different. Identifying n = + 1/2 in
Eq. (2.178), one writes the relation
∞ √ √ √ 2
π π ′ π 1
∫ dx x2 ( J+1/2 (k x)) ( ′
J +1/2 (k x)) = ( ) δ(k − k ′ ) ,
0 2kx 2k x 2 k k′
(2.180)
which implies that
∞ π
∫ dx x2 j (k x) j (k ′ x) =
δ(k − k ′ ) . (2.181)
0 2 k k′
The above derivation has been lengthy but it serves to alert us to the beauty of the
theory of special functions, whose power we shall bring to bear on the calculation
of Green functions.
2.4.4 Expansion of the Green Function in Cylindrical Coordinates

We now repeat the analysis of Sec. 2.3.4 regarding the construction of a Green
function, but this time in cylindrical coordinates. The integration measure reads
d3 r = ρ dρ dϕ dz , (2.182)
and the Green function satisfies
1 1
⃗ 2 G (⃗
∇ r, r⃗′ ) = − δ (3) (⃗
r − r⃗′ ) = − δ (ρ − ρ′ ) δ (z − z ′ ) δ (ϕ − ϕ′ ) . (2.183)
0 0 ρ
The right-hand side of Eq. (2.183) vanishes almost everywhere, except for the
point r⃗ = r⃗′ . We refer to Eq. (2.143) and write the following general form for
r , r⃗′ ), leaving one set of variables (in this case z and z ′ ) in terms of an unknown
G (⃗
function, which we refer to as gm (k; z, z ′ ). Then,
1 ∞ im(ϕ−ϕ′ ) ∞
r , r⃗′ ) =
G (⃗ ∑ e
′ ′
∫ dk k gm (k; z, z ) Jm (k ρ) Jm (k ρ ) . (2.184)
2π m=−∞ 0
We now use the Laplace operator in cylindrical coordinates, given in Eq. (2.133),
⃗ 2 G (⃗ 1 ∂ ∂ 1 ∂2 ∂2 1 ∞ im(ϕ−ϕ′ )
∇ r , r⃗′ ) = ( ρ + 2 + ) ( ∑ e
ρ ∂ρ ∂ρ ρ ∂ϕ2 ∂z 2 2π m=−∞
∞
×∫ dk k gm (k; z, z ′ ) Jm (kρ) Jm (kρ′ ) dk) . (2.185)
0
In view of the equation fulfilled by the Bessel function, we can replace

1 ∂ ∂ 1 ∂2 m2 m2
ρ + 2 2
→ ( 2 − k 2 − 2 ) → −k 2 . (2.186)
ρ ∂ρ ∂ρ ρ ∂ϕ ρ ρ
Then,
1 ∞ im(ϕ−ϕ′ ) ∞
⃗ 2 G (⃗
∇ r, r⃗′ ) = ∑ e ∫ dk k Jm (kρ) Jm (kρ′ )
2π m=−∞ 0
∂2 1
) gm (k; z, z ′ ) = − δ (ρ − ρ′ ) δ (z − z ′ ) δ (ϕ − ϕ′ ) .
!
× (−k 2 + 2
∂z 0 ρ
(2.187)
We now apply the integral operator
2π ′ ∞
∫ dϕ e−im ϕ ∫ dρρ Jm′ (k ′ ρ) , (2.188)
0 0
and integrate both sides of Eq. (2.187) over dϕ and dρ. The left-hand side (LHS)
of Eq. (2.187) becomes
2π ∞ ∞
1 ′ ′
LHS = ∫ dϕ ∑ e−i(m −m)ϕ e−imϕ ∫ dk k Jm (kρ′ )
2π 0 m=−∞ 0
∞ ∂2
× (∫ dρ ρ Jm (kρ) Jm′ (k ′ ρ)) (−k 2 + ) gm (k; z, z ′ ) . (2.189)
0 ∂z 2
We now use the orthogonality condition
1 2π ′
∫ e−i(m −m)ϕ dϕ = δmm′ (2.190)
2π 0
to reduce the sum over m to a single term with m = m′ , and the integral (2.178),
∞ 1
∫ dρ ρ Jm (k ρ) Jm (k ′ ρ) = δ (k − k ′ ) (2.191)
0 k
to carry out the integration over k. Then, the left-hand side of Eq. (2.187) becomes
′ ′ ∞ 1 ∂2
LHS = e−im ϕ ∫ dk k Jm′ (kρ′ ) δ (k − k ′ ) (−k 2 + 2 ) gm′ (k; z, z ′ )
0 k ∂z
′ ′ ∂2
= e−im ϕ Jm′ (k ′ ρ′ ) (−k 2 + ) gm′ (k; z, z ′ ) . (2.192)
∂z 2
The right-hand side of Eq. (2.187) becomes
1 2π ∞ ′
RHS = − ∫ dϕ ∫ dρ δ (ρ − ρ′ ) δ (z − z ′ ) δ (ϕ − ϕ′ ) e−im ϕ Jm′ (k ′ ρ)
0 0 0
1 ′ ′
= − e−im ϕ Jm′ (k ′ ρ′ ) δ (z − z ′ ) . (2.193)
0
Equating LHS = RHS, we finally extract the differential equation for gm′ (k; z, z ′)
and substitute m′ → m,
∂2 1
( − k 2 ) gm (k; z, z ′ ) = − δ (z − z ′ ) . (2.194)
∂z 2 0
This equation is easy to solve by integrating over the kink at r = r′ . While the
function gm (k; z, z ′ ) is continuous, its derivative with respect to z should have a
discontinuity at z = z ′ .
As in Eq. (2.84), we now integrate both sides over the interval z ∈ (z ′ − ε, z ′ + ε)
and obtain
z ′ +ε ∂2 1 z +ε ′
1
∫ dz ( 2 − k 2 ) gm (k; z, z ′ ) = − ∫ δ (z − z ′ ) dz = − . (2.195)
z ′ −ε ∂z 0 z′ −ε 0
The left-hand side can be reformulated as
′
z ′ +ε z=z +ε
∂ ∂ ∂
∫ ( gm (k; z, z ′ )) dz = gm (k; z, z ′ )∣ . (2.196)
z ′ −ε ∂z ∂z ∂z z=z ′ −ε
The following ansatz

gm (k; z, z ′ ) = Cm,k exp(k z< − k z> ) , (2.197)
with z< = min(z, z ′ ), z> = max(z, z ′ ), is inspired by the general solution [Eq. (2.143)]
to the homogeneous equation. Differentiation leads to
′
z +ε
∂ ′ ∂ ′ ∂ ′ ′ 1
gm (z; z ′ )∣ = Cm,k [ek z { ′ e−k (z +ε) } − { ′ ek (z −ε) } e−k z ] = − ,
∂z ′
z −ε ∂z ∂z 0
(2.198)
and thus
′ ′ ′ ′
Cm,k (ek z (−k) e−k(z +ε ) − k ek(z −ε) e−k z ) = 1 . (2.199)
As ε → 0,
′ ′ 1 1
Cm,k (−2k ek z −k z ) = − , Cm,k = . (2.200)
0 20 k
We now insert the solution

1
gm (k; z, z ′ ) = exp [−k (z> − z<′ )] (2.201)
20 k
into the expansion (2.184) for the Green function and obtain
∞ ∞
1 im(ϕ−ϕ ) ′ 1
r , r⃗′ ) =
G (⃗ ∑ e ∫ dk k ( ) e−k(z> −z< ) Jm (kρ) Jm (kρ′ ) .
2π m=−∞ 0 2 0 k
(2.202)
The latter result can be simplified slightly,
∞ ∞
1 ) ′
r , r⃗′ ) =
G (⃗ ∑ eim(ϕ−ϕ ∫ dk e−k(z> −z< ) Jm (kρ) Jm (kρ′ )
4π 0 m=−∞ 0
1
= . (2.203)
r − r⃗′ ∣
4π0 ∣⃗
Multiplication by 4π0 leads to the identity
∞ ∞
1 im(ϕ−ϕ′ )
= ∑ e ∫ dk e−k(z> −z< ) Jm (kρ) Jm (kρ′ ) . (2.204)
r − r⃗′ ∣ m=−∞
∣⃗ 0
A comparison of Eqs. (2.203) and (2.204) with the corresponding expansions (2.91)
and (2.93) in spherical coordinates implies that in the spherical coordinate system,
we have to integrate over k in contrast to a summation over discrete values of and
m. The case differentiation is not carried out in the radial variables r< and r> , but
in the z> and z< coordinates. The dependence on z is such that the argument of the
exponential leads to exponential damping for both z, z ′ → ±∞, in agreement with
the boundary conditions imposed on the Green function.
2.5 Electrostatic Green Function and Eigenfunction Expansions
2.5.1 Eigenfunction Expansions for the Green Function

The calculation of Green functions may be mapped onto the inversion of linear
operators. This becomes clear if we take into account that the Dirac-δ function,
which appears in the defining equation for a typical Green function, is equal to a
constant in Fourier space. We consider the inversion of the linear operator Ξ,
Ξ = L − λ, (2.205)
where λ is a constant and L is a linear differential operator. For the Green function
of electrostatics, we would have L = −0 ∇ ⃗ 2 . We wish to solve the defining equa-
tion (2.22) for the Green function, which can be mapped onto the special case λ = 0
of the generalized equation
r , r⃗′ ) = δ (⃗
(L − λ) G (⃗ r − r⃗′ ) . (2.206)
∞
Let us suppose that we have a set {ψn (⃗
r )}n=0 of eigenfunctions that satisfy
L ψn (⃗
r ) = λn ψn (⃗
r) , (2.207)
with eigenvalues λn and with appropriate boundary conditions on the ψn (⃗ r ) so that

the eigenvalue equation for λn is properly defined.
A small detour for those readers who have already heard a lecture on quantum
mechanics is in order. If L is equal to the Hamiltonian for a quantum system,
then L = H has energy eigenvalues λn = En . The discreteness of the eigenvalues
for, e.g., the hydrogen atom, follows from the boundary condition for the bound
wave functions at infinity, i.e., from the condition that a bound state should be
normalizable. Otherwise, it would be possible to solve the Schrödinger equation for
any energy “eigenvalue”. The discreteness of the bound spectrum follows exclusively
from the normalization condition, which therefore is of preeminent importance.
∞
We assume that the {ψn (⃗ r )}n=0 provide a complete set of eigenfunctions, in the
sense of Eq. (2.207), so that
∗
∫ d r ψm (⃗
r) ψn (⃗
r) = δmn .
3
(2.208)
V
Completeness implies that
r ) ψn∗ (⃗
∑ ψn (⃗ r′ ) = δ (3) (⃗
r − r⃗′ ) , (2.209)
n
for r⃗, r⃗′ in V . The solution is
r ) ψn∗ (⃗
ψn (⃗ r′ )
G (λ, r⃗, r⃗′ ) = ∑ , (2.210)
n λn − λ
because
r ) ψn∗ (⃗
ψn (⃗ r′ )
(L − λ) G (λ, r⃗, r⃗′ ) = ∑(λn − λ) r) ψn∗ (⃗
= ∑ ψn (⃗ r′ ) = δ (3) (⃗
r − r⃗′ ) .
n λn − λ n
(2.211)
In the case λ = 0, we have
r ) ψn∗ (⃗
ψn (⃗ r′ )
G (λ, r⃗, r⃗′ ) = ∑ . (2.212)
n λn
This is a relation common to the Green function, because
L G(λ, r⃗, r⃗′ ) = ∑ ψn (⃗
r) ψn∗ (⃗
r′ ) = δ(⃗
r − r⃗′ ) . (2.213)
n
In practical cases, one may construct a Green function like so: One takes a dif-
ferential operator and formulates a discrete approximation to its spectrum and
eigenfunctions within a particular basis set of functions. Using the approximate
eigenstates, one can use Eq. (2.210) to calculate the Green function immediately.
One decisive feature of the Green function expansion in terms of eigenfunctions
is as follows. Let us reconsider Eqs. (2.91) and (2.202) for the Green function of
electrostatics (in spherical and cylindrical coordinates, respectively). In both cases,
we have had to implement the “cusp” of the Green function at equal arguments by
an explicit dependence on r> and r<′ , or z> and z<′ , respectively. In an eigenfunc-
tion decomposition, this cusp is naturally implemented in the sum over eigenstates
ψn (⃗r), leading to some sort of simplification. However, the sum over states leads
to an additional level of complexity in the calculation which we shall study in the
following.
2.5.2 Application to Electrostatics in Spherical Coordinates

We discuss the calculation of the Green function of electrostatics in terms of an
eigenfunction decomposition of solutions of a corresponding eigenvalue problem. In
order to apply the formalism developed in the previous Sec. 2.5.1, we bring the
defining Eq. (2.22) of the Green function into the standard form (2.206),
1 (3)
⃗ 2 G(⃗
∇ r , r⃗′ ) = − r − r⃗′ ) ,
δ (⃗ ⃗ 2 [−0 G(⃗
∇ r , r⃗′ )] = δ (3) (⃗
r − r⃗′ ) , (2.214)
0
so that
r , r⃗′ ) = δ (3) (⃗
(L − λ) G(⃗ r − r⃗′ ) , ⃗2 ,
L = −0 ∇ λ = 0. (2.215)
One parameterizes the eigenvalue as
λ = λk = 0 k 2 . (2.216)
The corresponding eigenvalue problem is
(L − λk ) ψkm (⃗ ⃗ 2 − 0 k 2 ) ψkm (⃗
r ) = (−0 ∇ r) = 0 . (2.217)
We will seek to find solutions of the form
⃗ 2 + k 2 ) ψkm (⃗
(∇ r) = 0 , ψkm (⃗
r ) = Rk (r) Ym (θ, ϕ) . (2.218)
The radial equation for Rk (r) reads
∂2 2 ∂ ( + 1)
2
Rk (r) + Rk (r) − Rk (r) + k 2 Rk (r) = 0. (2.219)
∂r r ∂r r2
Solutions regular at the origin are given by the spherical Bessel functions j (k r);
these vary as r for small r. These are defined according to Eq. (2.179),
√
π
Rk (r) = j (k r) , j (k r) = J 1 (k r) . (2.220)
2 k r + 2
A more thorough discussion on spherical Bessel functions follows in Sec. 5.2.2.
According to Eq. (2.181), the eigenfunctions satisfy the following orthogonality
condition
∞
⟨ψkm ∣ψk′ m′ ⟩ = ∫ dr r2 j (k r) j′ (k ′ r) ∫ dΩ Ym
∗
(θ, ϕ) Y′ m′ (θ, ϕ)
0
π
= δ (k − k ′ ) δ′ δmm′ . (2.221)
2 k k′
Of course, we could have written replaced k k ′ → k 2 in the prefactor, but the above
expression is explicitly symmetric. For a discrete set of eigenfunctions, Eq. (2.221)
becomes analogous to Eq. (2.208).
So, alternatively, the completeness relationship is given by
∞ ∞

2 k2
∑ ∑ ∫ dk j (k r) j (k r′ ) Ym (θ, ϕ) Ym
∗
(θ′ , ϕ′ ) = δ (3) (⃗
r − r⃗′ ) .
=0 m=− 0 π
(2.222)
For a discrete set of eigenfunctions, Eq. (2.222) is the analogue of Eq. (2.209). The
Green function thus has the form
∞ ∞
1 2 k2
r , r⃗′ ) = ∑ ∑ ∫
G(⃗ j (k r) j (k r′ ) Ym (θ, ϕ) Ym
dk ∗
(θ′ , ϕ′ ) ,
=0 m=− 0 λk π
(2.223)
where we have simply inserted the reciprocal of the eigenvalue according to
Eq. (2.212). It follows that
2 ∞ ∞
r , r⃗′ ) =
G(⃗ ∑ ∑ ∫ dk j (k r) j (k r′ ) Ym (θ, ϕ) Ym
∗
(θ′ , ϕ′ ) . (2.224)
π 0 =0 m=− 0
In this formula, the cusp is formulated in terms of the eigenfunction expansion, as
anticipated near the end of Sec. 2.5.1. The radial component of the Green function
is formulated in terms of r and r′ , not r< and r> . However, we now have to sum
over the eigenfunctions, i.e., integrate over k. Comparing with the familiar formula
1 1 ∞ 1 r<
r , r⃗′ ) =
G(⃗ = ∑ ∑
∗
Ym (θ, ϕ) Ym (θ′ , ϕ′ ) , (2.225)
r − r⃗′ ∣ 0 =0 m=− 2 + 1 r>+1
4π0 ∣⃗
we find that the cusp at equal argument is implemented in the formula
∞ π r<
∫ dk j (kr) j (kr′ ) = . (2.226)
0 2 (2 + 1) r>+1
We may perform a consistency check on the result (2.224), by considering a
point charge q0 located on the z axis at z = a. In order to describe a charge at the
origin, we shall consider the limit a → 0 near the end of the calculation. The charge
density for a point charge located at the point r⃗0 = a êz is given as
q0
ρ (⃗r ) = q0 δ (3) (⃗
r − r⃗0 ) = 2 δ (r − r0 ) δ (θ − θ0 ) δ (ϕ − ϕ0 )
a sin θ
q0
= 2 δ (r − a) δ (cos θ − 1) δ (ϕ − ϕ0 ) . (2.227)
a
We have used the fact that θ0 = 0 for a charge located at r⃗0 = a êz , with a > 0 (hence,
also, r0 = ∣⃗
r0 ∣ = a). In terms of the Green function, expanded in the spherical Bessel
functions and spherical harmonics, the potential is
1 3 ′ ρ(⃗ r′ ) ∞ 1 2π
Φ (⃗
r) = ∫ d r ′
=∫ dr′ r′2 ∫ d (cos θ′ ) ∫ dϕ′ G(⃗
r , r⃗′ ) ρ(⃗
r′ )
4π0 ∣⃗
r − r⃗ ∣ 0 −1 0
2
2q0 ∞ 1 2π r′
= ∫ dr′ ∫ d (cos θ′ ) ∫ dϕ′ δ (r′ − a) δ (cos θ′ − 1) δ (ϕ′ − ϕ0 ) ( )
π0 0 −1 0 a
∞ ∞
×∑ ∑ ∫ dk j (k r) j (k r′ ) Ym (θ, ϕ) Ym
∗
(θ′ , ϕ′ )
=0 m=− 0
2q0 ∞ ∞
∗
= ∑ ∑ ∫ dk j (k r) j (k a) Ym (θ, ϕ) Ym (θ0 , ϕ0 )
π0 =0 m=− 0
2q0 ∞ 2 + 1 ∞
= ∑ P (cos (∠(⃗
r , r⃗0 ))) ∫ dk j (k r) j (k a) . (2.228)
π0 =0 4π 0
The integration over ϕ′ leads to the replacement ϕ′ → ϕ0 in the argument of the

spherical harmonic. We have used the addition theorem (2.111). We now consider
the limit a → 0. Since j (0) = 0 for ≠ 0 and only the spherical Bessel function of
order zero survives, with unit value j0 (0) = 1, the only surviving term in the sum
over reads as
∞ ∞
a→0 2q0 1 q0
Φ (⃗r) = P0 (cos (∠(⃗r , r⃗0 ))) ∫ dk j (k r) = 2 ∫ dk j (k r) ,
π0 4π 0 2π 0 0
(2.229)
where we have used the fact that P0 (x) = 1. We thus have for a → 0,
a→0 q0 ∞ q0 ∞ sin (k r)
Φ (⃗
r) = 2 ∫ dk j0 (k r) = 2 ∫ dk
2π 0 0 2π 0 r 0 k
q0 ∞ dρ q0 π q0
2
= (∫ sin (ρ)) = 2 ( )= . (2.230)
2π 0 r 0 ρ 2π 0 r 2 4π0 r
This is equal to the familiar result for a point charge located at the origin.
2.6 Summary: Green Function of Electrostatics
The Green function of electrostatics, first encountered in Eq. (2.23),

⃗ 2 G(⃗ 1 1
∇ r , r⃗′ ) = − δ (3) (⃗
r − r⃗′ ) , r , r⃗′ ) =
G(⃗ , (2.231)
0 r − r⃗′ ∣
4π0 ∣⃗
has been the subject of our discussions. Let us summarize the three different repre-
sentations we have derived for the Green function of electrostatics. First, we have
the familiar decomposition in spherical coordinates, as derived in Sec. 2.3.4,
1 ∞ 1 r<
r , r⃗′ ) =
G(⃗ ∑ ∑
∗
Ym (θ, ϕ) Ym (θ′ , ϕ′ ) . (2.232)
0 =0 m=− 2 + 1 r>+1
It relies on a matching of the regular solutions at the origin (r = 0) and at infinity
(r = ∞). The corresponding representation in cylindrical coordinates is derived in
Sec. 2.4.4 and reads
∞ ∞
1 im(ϕ−ϕ′ )
G (⃗r ; r⃗′ ) = ∑ e ∫ dk e
−k(z> −z< )
Jm (kρ) Jm (kρ′ ) . (2.233)
4π 0 m=−∞ 0
It is based on a matching of the regular solutions at z = −∞ and z = +∞. The

decomposition into eigenfunctions of the Laplacian operator is discussed in Sec. 2.5.2
and leads to Eq. (2.224),
2 ∞ ∞
r , r⃗′ ) =
G(⃗ ′ ∗ ′ ′
∑ ∑ ∫ dk j (k r) j (k r ) Ym (θ, ϕ) Ym (θ , ϕ ) . (2.234)
π0 =0 m=− 0
The results (2.232) and (2.224) are matched using the result (2.226), which we recall
for convenience,
∞ π r<
∫ dk j (k r) j (k r′ ) = . (2.235)
0 2 (2 + 1) r>+1
This result implements the cusp of the Green function at equal argument.
2.7 Exercises
● Exercise 2.1: By directly inserting the result given in Eq. (2.14) into the left-hand
side of Eq. (2.5), verify the validity of the result given for the harmonic oscillator
Green function.
● Exercise 2.2: We investigate the harmonic oscillator using Green function tech-
niques and apply the Green function formalism to the harmonic oscillator. The
“displacement” x = x(t) of a damped, harmonic oscillator with unit mass m = 1
satisfies the equation
ẍ + γ ẋ + ω02 x = f (t) . (2.236)
(i) Obtain the Green function g = g(t − t′ ) for this equation. Note that it has to
fulfill the equation
g̈ (t − t′ ) + γ ġ (t − t′ ) + ω02 g (t − t′ ) = δ (t − t′ ) . (2.237)
(ii) Why is this Green function naturally obtained as the retarded Green function?
Why do you have to change the sign of the damping term in order to obtain the
advanced Green function? Consider the effect of time reversal on the equation of
motion, and provide an illustrative discussion.
(iii) In the case that the “force” is given by f (t) = f0 [Θ (t) + Θ (−t) exp (t/τ )], with
vanishing boundary conditions in the infinite past, x (−∞) = 0, and ẋ (−∞) = 0,
evaluate x (t).
(iv) Evaluate the work done by the driving force on the oscillator (per unit mass)
in the time interval from −∞ to +∞ as a function of ω0 , γ, and τ .
(v) Now let γ = 10 1
ω0 . Plot the work done by the driving force divided by the final
(potential) energy stored in the oscillator as a function of L = ln (ω0 τ ) from L = −4
to L = 4.
● Exercise 2.3: Starting from Eq. (2.30), derive Eq. (2.31) by applying the ope-
rators given in Eq. (2.30) to a test function.
● Exercise 2.4: Show Eq. (2.57) by using the explicit definitions of the spherical
harmonics in terms of Legendre polynomials given in Eqs. (2.44)–(2.47), as well
as Eq. (2.53), and the orthogonality properties of the Legendre polynomials [see
Eq. (2.48)].
● Exercise 2.5: Derive Eq. (2.77) by acting with it on a test function.
● Exercise 2.6: For the charge distribution given in Eq. (2.100), calculate the
potential in Eq. (2.103), i.e., fill the missing steps in the derivation.
● Exercise 2.7: Use the expansion (2.93) in order to generalize the potential (2.103)
for r < a, still assuming the charge distribution (2.100).
● Exercise 2.8: Consider the charge distribution
Q
r) =
ρ(⃗ sin θ δ(r − a) . (2.238)
a2
Evaluate all multipoles of the given charge distribution with angular momenta =
0, 1, 2 and write down the final potential. Hint: The potential should finally be
obtained as follows,
πQ π Q a2
r) =
Φ(⃗ − (3 cos2 θ − 1) . (2.239)
4 0 r 64 0 r3
● Exercise 2.9: For the charge distribution given in Eq. (2.104), carry out the
multipole decomposition explicitly and calculate the potential in Eq. (2.105).
● Exercise 2.10: Show Eq. (2.119) based on the explicit form of the spherical
harmonics for = 1, given in Eq. (2.63).
● Exercise 2.11: Show Eq. (2.147) based on Eq. (2.146).
● Exercise 2.12: Show how to obtain Eq. (2.150) from Eq. (2.149).
● Exercise 2.13: Write a computer program in the language of your choice, im-
plementing the Newton–Raphson method for finding the zeros of Bessel functions.
Choosing the starting values of the Newton–Raphson method as integer-valued,
verify the entries in Table 2.1.
● Exercise 2.14: In establishing Eq. (2.166), we have ignored the fact that we must
actually treat the case n = 0 separately; it requires special attention at the lower

limit of integration because Jn (0) = δn0 for n ∈ 0 . Fill this gap in our derivation.
Hint: Use the known fact that Jn′ (0) = 0.
● Exercise 2.15: Treat the special case n = 0 in Eq. (2.174), observing that the
slope of J0 (x) vanishes at x = 0.
● Exercise 2.16: Show the relation
1
δ (3) (⃗
r − r⃗′ ) = δ (ρ − ρ′ ) δ (z − z ′ ) δ (ϕ − ϕ′ ) (2.240)
ρ
using similar arguments to those in the derivation of Eq. (2.70).
● Exercise 2.17: Use the expansions (2.203) and conceivably (2.204) in order
to calculate the potential (in the upper half space z > 0) generated by a charge
distribution in a “ring of Saturn” geometry located in the xy plane,
ρ(⃗
a
Q

r ) = 2 a<ρ<b δ(z) . (2.241)

Here, a<ρ<b is the characteristic function of the interval ρ ∈ (a, b), i.e.,
a<ρ<b = Θ(ρ − a) − Θ(ρ − b) . (2.242)
Generalize your result for the lower half space, i.e., z < 0.
● Exercise 2.18: Generalize the ψkm to eigenfunctions of the free Schrödinger
Hamilton operator H = −∇ ⃗ 2 /(2μ) and establish their orthogonality properties.
Hint: A remark is in order. The function
ψkm (r, θ, ϕ) = j (k r) Ym (θ, ϕ) (2.243)
fulfills the Schrödinger equation
̵2 ∇
h ⃗2 p2
H0 ψkm (r, θ, ϕ) = (− ) ψkm (r, θ, ϕ) = ψkm (r, θ, ϕ) , ̵ k . (2.244)
p=h
2μ 2μ
In view of Eq. (2.178), the eigenfunctions satisfy the following orthogonality

condition,
∞
∗
⟨ψkm ∣ψk′ m′ ⟩ = ∫ dr r2 j (k r) j′ (k ′ r) ∫ dΩ Ym (θ, ϕ) Y′ m′ (θ, ϕ)
0
π
= δ (k − k ′ ) δ′ δmm′ . (2.245)
2 k k′
√
This is immediately obvious if we take the prefactor π/(2x) in the definition of
the spherical Bessel function into account.
● Exercise 2.19: Using Eq. (2.224), calculate the potential generated by a charge
distribution
q0
ρ(⃗r′ ) = 2 δ(r′ − a) (2.246)
a
where a is the radius of the sphere.
● Exercise 2.20: Using Eq. (2.224), reduce the potential generated by a charge
distribution without compact support, given in spherical coordinates,
q0
r′ ) = 3 exp(−λ r) cos θ′ ,
ρ(⃗ (2.247)
a
to a one-dimensional integral (a is a parameter). Choose convenient numerical val-
ues for the parameters q0 , a and λ, and evaluate the remaining integral numerically
for selected values of r and θ. Calculate the dipole moment of the charge distribu-
tion, and show analytically that it is equal to the long-range limit of the potential
just calculated, even if part of the charge distribution is always in the outer volume
r′ > r. Hint: Observe that cos θ′ is proportional to a spherical harmonic in the
primed coordinates, and use the result
∞ ( + 32 )n
∫ dx J+1/2 (x) rn e−λ r =
0 2+1/2 λn++3/2
1
× 2 F1 ( 14 (2n + 2 + 3), 14 (2n + 2 + 5), + 32 , −
),
λ2
(2.248)
where (a)n = Γ(a + n)/Γ(a) is the Pochhammer symbol. The complete (Gaussian)
hypergeometric function is denoted as 2 F1 . Part of the exercise is to look up the
definition of the hypergeometric function, and to become familiar with its numerical
properties.
Chapter 3
Paradigmatic Calculations in
Electrostatics
3.1 Overview
3.1.1 Differential Equations of Electrostatics

The previous chapter dealt with the Green functions of electrostatics, which fulfill
the differential equation (2.22), which we recall for convenience,
1
⃗ 2 G(⃗
∇ r − r⃗′ ) = − δ (3) (⃗
r − r⃗′ ) . (3.1)
0
Given a formula for G, one can calculate the electrostatic potential due to charge
r′ ) with the help of the formula (2.24), which we write as
distribution ρ(⃗
1 3 ′ ρ(⃗ r′ )
r) = ∫ d3 r′ G(⃗
Φ(⃗ r , r⃗′ ) ρ(⃗
r′ ) = ∫ d r . (3.2)
4π0 r − r⃗′ ∣
∣⃗
Given Φ, one can calculate the electric field according to Eq. (1.59),
⃗ r ) = −∇Φ(⃗
E(⃗ ⃗ r) , (3.3)
for the time-independent (static) case where the time derivative of the vector po-
tential A⃗ vanishes.
In principle, one might assume that formula (2.24) solves all physically relevant
problems in electrostatics. However, that is not the case. Let us look at the Poisson
equation
1
∇⃗ 2 Φ(⃗r) = − ρ(⃗ r) , (3.4)
0
and analyze its properties, being a differential equation. We know that the solu-
tions of differential equations are never uniquely defined without the specification of
boundary conditions. Let us assume that the charge distribution ρ(⃗ r′ ) in Eq. (3.2)
has no significant long-range tails, and that it is concentrated in the region around
the origin, r⃗′ ≈ 0. In this case, Φ(⃗ r ) vanishes for r = ∣⃗
r ∣ → ∞, as a potential pro-
portional to 1/r, commensurate with the leading (monopole) term in the multipole
expansion (2.99). The boundary condition implemented in Eq. (3.2) therefore reads
as
Φ(⃗r )∣r→∞ = 0 , (3.5)
and Φ is defined for all r⃗ ∈ .
71
Yet, this cannot be the whole story. E.g., one may ask how potentials should
be calculated in a volume V of space whose surface manifold ∂V , consisting of a
perfect conductor, is held at a constant potential Φ(⃗ r )∣r⃗∈∂V = Φ0 . In this case, if
there are no charges present inside the volume V , the simple solution Φ(⃗ r ) = Φ0 will
solve the differential equation (3.4) with ρ = 0. However, if the charge distribution
inside V is nonvanishing, then the boundary-value problem cannot be solved using
Eq. (3.2) because the boundary condition Φ(⃗ r )∣r⃗∈∂V = Φ0 is not fulfilled.
Let us also ask about the solution to a boundary-value problem given by a cube,
held at zero potential on five of the six side surfaces, with the remaining sixth
side surface held at potential Φ0 ≠ 0. In this case, even if there are no charges
present inside the cube, then the solution to the Poisson equation ∇ ⃗ 2 Φ(⃗
r) = 0
cannot be given by the simple ansatz Φ(⃗ r) = 0, because it does not fulfill the
boundary condition on the sixth side surface. Even more generally, one may ask
how to calculate the solution to a boundary-value problem
1
⃗ 2 Φ(⃗
∇ r) = − ρ(⃗ r) , Φ(⃗r)∣r⃗∈∂V = f (⃗ r) , (3.6)
0
with a nontrivial function f (⃗ r) describing the potential on the surface manifold.
In principle, all of the above questions can be traced to the availability of non-
trivial solutions to the Laplace equation (2.20)
⃗ 2 Φ(⃗
∇ r) = 0 , (3.7)
which can be added to the solutions of the Poisson equation (3.4), to fulfill the
boundary conditions. However, there are more sophisticated techniques available
in comparison to a simple guessing of the correct nontrivial solution of the Laplace
equation which will lead to a solution of a given boundary-value problem. We start
by temporarily falling back to a simpler one-dimensional model problem.
3.1.2 Boundary-Value Problems: One-Dimensional Analogy

⃗ 2 Φ(⃗
We investigate a one-dimensional analogue of the Laplace equation ∇ r) = 0,
∂2
f ′′ (x) =
f (x) = 0 . (3.8)
∂x2
It is known that the solution of a differential equation in one variable can be made
unique on the basis of two boundary conditions. As an example, Eq. (3.8) is solved
by any function of the form
f (x) = C + D x , (3.9)
with constant coefficients C and D. A differential equation for f = f (x) is an equa-
tion describing a function on a one-dimensional interval x ∈ [a, b]. The boundary
conditions are related to a zero-dimensional manifold of points, i.e., to the specific
points x = a and x = b. Let us consider a slightly more complex “boundary-value
problem”,
f ′′ (x) = s(x) , f (a) = sa , f (b) = sb , (3.10)
Paradigmatic Calculations in Electrostatics 73
where s(x) is a source term. We shall later see that this situation corresponds to
a so-called Dirichlet boundary-value problem. We are free to choose functions g(x)
with g ′′ (x) = 0 and add them to a particular solution f (x),
f (x) → f (x) + C + D x . (3.11)
The boundary conditions then read as follows,
f (a) = C + D a = sa , f (b) = C + D b = sb , (3.12)
and they fix the integration constants C and D uniquely, even if we have not used any
boundary conditions that involve the derivative f ′ (x). For a differential equation
in a single variable, the boundary of the integration interval is the set of points
[a, b] that define the interval a ≤ x ≤ b. Boundary conditions that fix the derivative
f ′ (x) of the function at the end points x = a and x = b determine f (x) up to an
additive constant. This situation corresponds to a so-called Neumann boundary-
value problem.
The integration interval in this case is one-dimensional, whereas the boundary
conditions are defined on a zero-dimensional “surface”, namely, at the end points of
the interval (a, b). We can now figure an analogy: Namely, for the three-dimensional
case, the integration interval is three-dimensional, and the boundary conditions
must be defined on a two-dimensional surface. Indeed, the time-independent Poisson
equation ∇ ⃗ 2 Φ (⃗
r ) = −ρ(⃗
r )/0 is a second-order differential equation whose solution
is defined for r⃗ ∈ V . Consequently, its boundary conditions are relevant for the
boundary ∂V of a three-dimensional volume V .
3.1.3 Green’s Theorem: Dirichlet and Neumann Green Functions

We have already stressed that the solution of the Poisson equation (3.4) is not
unique; it may be modified by the addition of a solution of the homogeneous
(Laplace) equation. This property can be taken advantage of in the analysis of
the Green function. Let us recall the defining differential Eq. (2.22),
1 1
∇⃗ 2 G(⃗
r , r⃗′ ) = − δ (3) (⃗ r − r⃗′ ) , G(⃗r , r⃗′ ) = , (3.13)
0 4π0 ∣⃗ r − r⃗′ ∣
as a suitable electrostatic Green function. Indeed, we may always add a solution
of the homogeneous equation, ∇ ⃗ 2 G(⃗
r , r⃗′ ) = 0. Green functions can be unique by
the specification of boundary conditions, much in the same way as potentials are
uniquely specified by differential equations and boundary conditions.
This statement needs to be illustrated. We have already encountered, in
Eq. (3.6), what we now identify as a Dirichlet boundary-value problem,
1
∇⃗ 2 Φ(⃗ r) = − ρ(⃗ r) , Φ(⃗r)∣r⃗∈∂V = f (⃗ r) . (3.14)
0
One specifies the charge distribution ρ(⃗ r) and the value f (⃗ r) on the boundary ∂V
of the volume V . A Neumann boundary-value problem is specified as
1
∇⃗ 2 Φ(⃗ r ) = − ρ(⃗ r) , ⃗ r)∣
∇Φ(⃗ = g⃗(⃗
r) , (3.15)
0 r⃗∈∂V
with a vector-valued function g⃗(⃗ r). This corresponds to a boundary condition for
the electric field instead of the potential on the surface ∂V . Actually, we could have
stated the Neumann boundary-value problem as
1
⃗ 2 Φ(⃗
∇ r)∣ =− r) ,
ρ(⃗ ⃗ r )∣
r ) ⋅ ∇Φ(⃗
n̂(⃗ = h(⃗
r) , (3.16)
r⃗∈V 0 r⃗∈∂V
r ) is the unit normal pointing away from the surface.

where n̂(⃗
We shall now attempt to relate the boundary conditions fulfilled by the potential,
to the boundary conditions fulfilled by the Green function. It is useful to consider
Green’s theorem in that context, which will enable us to accomplish the goal. On
our way to Green’s theorem, we first observe that the divergence theorem (Gauss’s
theorem) states that for any r⃗′ ∈ V ,
3 ⃗ ⃗ G (⃗ ⃗ r )]
∫ d r∇ ⋅ [Φ(⃗
r)∇ r , r⃗′ ) − G (⃗
r , r⃗′ ) ∇Φ(⃗
V
=∫ dA⃗ ⋅ [Φ(⃗ ⃗ (⃗
r)∇G r , r⃗′ ) − G (⃗ ⃗ r)] .
r, r⃗′ ) ∇Φ(⃗ (3.17)
∂V
By carrying out the differentation on the left-hand side explicitly, we see that
3 ⃗ ⃗ G (⃗ ⃗ r )]
∫ d r∇ ⋅ [Φ(⃗
r)∇ r , r⃗′ ) − G (⃗
r , r⃗′ ) ∇Φ(⃗
V
= ∫ d3 r [Φ(⃗ ⃗ 2 G (⃗
r) ∇ r , r⃗′ ) − G (⃗ ⃗ 2 Φ(⃗
r , r⃗′ ) ∇ r )]
V
1 (3) 1
=− ∫ d r δ (⃗
3
r − r⃗′ ) Φ(⃗ r , r⃗′ ) [− ρ (⃗
r) − ∫ d3 r G (⃗ r )]
0 V V 0
1 1
= − Φ(⃗r′ ) + ∫ d3 r G (⃗ r , r⃗′ ) ρ (⃗
r) (3.18)
0 0 V
where we assume that r⃗′ ∈ V . We solve Eq. (3.18) for Φ(⃗
r′ ),
r′ ) = ∫ d3 r G (⃗
Φ(⃗ r , r⃗′ ) ρ (⃗
r ) − 0 ∫ dA⃗ ⋅ [Φ(⃗ ⃗ (⃗
r) ∇G r , r⃗′ ) − G (⃗ ⃗ r )] .
r , r⃗′ ) ∇Φ(⃗
V ∂V
(3.19)
This formulation involves a volume term (convolution term with the charge distri-
bution) and a surface term (integral over ∂V ). It is known as Green’s theorem.
Equation (3.19) still does not solve the boundary-value problem. Moreover, the
right-hand side of Eq. (3.19) actually contains two terms, the evaluation of which
requires knowledge of both the values of the potential and of its gradient on the
surface ∂V . However, one can show that knowledge of either Φ(⃗ ⃗ r ) is
r) or of ∇Φ(⃗
sufficient in order to specify a unique boundary-value problem (in the latter case,
up to an additive constant).
At this stage, we recall that the Green function, (or just “Green’s function”
but without the “the”) is named after George Green, who was mentioned in this
book previously, and does not carry the Irish national color (which happens to be
green). Likewise, the Planck constant1 is denoted as h, ̵ and sometimes referred
1
Max Karl Ernst Ludwig Planck (1858–1947)
to as “Planck’s constant” (but in the latter case, without the “the” in front of the
word).
In order to evaluate appropriate solutions to a boundary-value problem of the
form Eq. (3.14), (3.15), or (3.16), it is useful to employ a Green function that
fulfills additional boundary conditions. The goal is to discard one of the terms on
the right-hand side of Eq. (3.19). The defining equation for the so-called Dirichlet
Green function GD is
1 (3)
⃗ 2 GD (⃗
∇ r , r⃗′ )∣r⃗, r⃗′ ∈V = − r − r⃗′ ) ,
δ (⃗ r , r⃗′ )∣r⃗∈V,⃗r′ ∈∂V = 0 .
GD (⃗ (3.20)
0
In this case, the solution to the boundary-value problem (3.14) becomes
r′ ) = ∫ d3 r GD (⃗
ΦD (⃗ r, r⃗′ ) ρ (⃗
r ) − 0 ∫ dA⃗ ⋅ ∇G
⃗ D (⃗
r , r⃗′ ) Φ(⃗
r)
V ∂V
r, r⃗′ ) ρ (⃗
= ∫ d3 r GD (⃗ r ) − 0 ∫ dA⃗ ⋅ ∇G
⃗ D (⃗
r , r⃗′ ) f (⃗
r) . (3.21)
V ∂V
The second integral requires the values of the potential on the boundary. We have
fulfilled our paradigm: The function GD depends only on the shape of the reference
volume V and can be used to integrate any given Dirichlet boundary-value problem
in the given volume. The solution is given by the general formula (3.21).
We now turn our attention to a Neumann boundary-value problem. In this case,
the gradient of the potential is given on the surface ∂V , as indicated in Eq. (3.15).
Let us write Green’s theorem (3.19) in a slightly different form,
r′ ) = ∫ d3 r G (⃗
Φ(⃗ r , r⃗′ ) ρ (⃗
r ) − 0 ∫ dA⃗ ⋅ ∇G
⃗ (⃗
r, r⃗′ ) Φ(⃗
r)
V ∂V
+ 0 ∫ dA⃗ ⋅ ∇Φ(⃗
⃗ r ) G (⃗
r , r⃗′ ) . (3.22)
∂V
Under Neumann boundary conditions, we do not know the potential Φ(⃗ r) itself on
the surface. So, we would like the second term on the right-hand side of Eq. (3.22)
to become irrelevant if the Green function G fulfills special properties, summarized
in the Neumann Green function GN . This is in analogy to the Dirichlet boundary-
value problem, where we had arranged GD so that the third term on the right-hand
side of (3.22) becomes irrelevant.
Now, in a typical case, we cannot simply assume that ∇G ⃗ (⃗r , r⃗′ ) vanishes on the
surface ∂V ; this would imply that all its three Cartesian components vanish, or in
other words, that the Green function G (⃗ r , r⃗′ ), as a function of r⃗, for an arbitrary
′
but fixed r⃗ , is an extremum on the surface ∂V in all three spatial directions. There
is a way out of this dilemma. It is instructive to observe that the second term on
the right-hand side of Eq. (3.22) involves the expression
dA⃗ ⋅ ∇G
⃗ (⃗
r , r⃗′ ) = dA (̂
n(⃗ ⃗ (⃗
r ) ⋅ ∇G r , r⃗′ )) . (3.23)
For reasons to be explained in the following, we shall assume that we can assign
a constant value to the projection of the gradient of the Neumann Green function
r , r⃗′ ) onto the surface normal ̂

GN (⃗ n(⃗
r ),
1 (3)
⃗ 2 GN (⃗
∇ r , r⃗′ )∣r⃗, r⃗′ ∈V = − r − r⃗′ ) ,
δ (⃗ (3.24a)
0
−1
1
⃗ N (⃗
∇G r, r⃗′ ) ⋅ ̂
n(⃗
r )∣r⃗∈V,⃗r′ ∈∂V = − = (−0 ∫ dA) , (3.24b)
0 Atotal ∂V
where Atotal is the total surface area of ∂V . We first convince ourselves that the
second term on the right-hand side of Eq. (3.22) assumes the form of an irrelevant
constant term, independent of the position on the surface,
∫ r ) dA⃗ ⋅ ∇G
Φ(⃗ ⃗ N (⃗
r , r⃗′ ) = ∫ dA ̂
n (⃗ ⃗ N (⃗
r ) ⋅ ∇G r , r⃗′ ) Φ(⃗
r)
∂V ∂V
r)
∫∂V dA Φ(⃗ ⟨Φ⟩
= − ≡− . (3.25)
0 ∫∂V dA 0
This corresponds to the average value ⟨Φ⟩ of the potential Φ, taken on the surface
∂V . With these assumptions on the Neumann Green function GN , Eq. (3.19)
becomes
r′ ) = ∫ d3 r GN (⃗
ΦN (⃗ r , r⃗′ ) ρ (⃗
r ) + ⟨Φ⟩ + 0 ∫ dA⃗ ⋅ ∇Φ(⃗
⃗ r) GN (⃗
r , r⃗′ )
V ∂V
r , r⃗′ ) ρ (⃗
= ∫ d3 r GN (⃗ r ) + ⟨Φ⟩ + 0 ∫ dA ̂
n(⃗
r ) ⋅ g⃗(⃗ r , r⃗′ )
r ) GN (⃗
V ∂V
r , r⃗′ ) ρ (⃗
= ∫ d3 r GN (⃗ r ) + ⟨Φ⟩ + 0 ∫ r , r⃗′ ) ,
r ) GN (⃗
dA h(⃗ (3.26)
V ∂V
where we have assumed Neumann boundary conditions as given in Eq. (3.15)

and (3.16). It becomes clear that in order write down a Neumann boundary-value
problem, it is sufficient to specify the component of the gradient of the potential
along the surface normal. Furthermore, it becomes clear that the solution to the
Neumann problem (3.26) involves a physically irrelevant constant term ⟨Φ⟩, which
can be chosen to be equal to the average value of the potential on the surface ∂V
if the condition (3.24) is fulfilled by the Green function.
3.1.4 Boundary-Value Problems and Laplace Equation

A special status must be attributed to electrostatic boundary-value problems where
the charge distribution vanishes, ρ(⃗ r ) = 0. In this case, the potential fulfills the
⃗
Laplace equation ∇ Φ(⃗2
r ) = 0 for r⃗ ∈ V . As already mentioned, this equation still
allows for nontrivial solutions specified by boundary conditions. The first way of
solving this kind of problem refers to Green’s theorem as discussed in the previous
section. In this case, the solution is formally given by setting ρ to zero in Eqs. (3.21)
and (3.26). For a Dirichlet problem, we have
r′ ) = −0 ∫
ΦD (⃗ dA⃗ ⋅ ∇G
⃗ D (⃗
r , r⃗′ ) Φ(⃗
r) , r) = 0 ,
ρ(⃗ (3.27)
∂V
and for a Neumann problem, the solution is given as
r ′ ) = 0 ∫
ΦN (⃗ dA⃗ ⋅ ∇Φ(⃗
⃗ r) GN (⃗
r , r⃗′ ) , r) = 0 .
ρ(⃗ (3.28)
∂V
However, the explicit calculation of the Dirichlet or Neumann Green functions

can be difficult for nontrivial geometries, and it is noteworthy that two alternative
methods exist which can be used in order to tackle boundary-value problems with
a vanishing charge distribution. These methods are based on two very fundamental
ideas which constitute recurrent themes throughout theoretical physics and there-
fore deserve special mention; the fundamental ideas are (i) the variational principle
and (ii) series expansions.
The electrostatic field energy E, in the absence of charges, is given by
0 ⃗ r ))2 .
E= d3 r (∇Φ(⃗
2 ∫V
(3.29)
⃗ r)]2 will lead to a concomitant

A slight change (“variation”) of the integrand [∇Φ(⃗
change in the field energy. As we shall see, variational calculus shows that the phys-
ically relevant, correct potential actually minimizes the integral E. The situation
bears a certain analogy to the shape of a soap film clamped within an irregularly
shaped closed wire loop: The wire acts as a boundary condition, and the soap film
minimizes its total surface area, which is tantamount to saying that its Gaussian
curvature vanishes locally [12]. The latter condition of vanishing local curvature,
translated into the variation of the electrostatic field energy, reads as
⃗ 2 Φ(⃗
∇ r) = 0 , (3.30)
which is, in fact, identical to the Laplace equation. Boundary conditions can be
implemented by the use of so-called Lagrange multipliers. In typical cases, a solution
to the boundary-value problem can be obtained through an explicit variation of
Φ(⃗r ). For example, one may assume for Φ, a particular functional form with free
parameters, and then find the particular set of expansion parameters that minimizes
the electrostatic field energy.
The other method is based on series expansions. Let us assume that we can
expand Φ(⃗ r) in terms of a series of mutually orthogonal functions, say, in terms of
a Fourier series,
r ) = ∑ an exp(ik⃗n ⋅ r⃗) ,
Φ(⃗ (3.31)
n
where the k⃗n are drawn from an “allowed” set of Fourier expansion coefficients. The
k⃗n are complex, so that
⃗ 2 exp(ik⃗n ⋅ r⃗) = 0 ,
∇ k⃗n2 = 0 . (3.32)
This condition does not imply that all components of k⃗n need to vanish; in fact,
the components of k⃗n can be complex-valued. However, it simply means that, say,
oscillatory solutions in the x and y directions must be accompanied by exponential
behavior in the z direction. Provided the surface ∂V is such that the Fourier com-
ponents remain orthogonal upon integration over the surface area, we can typically
project out the expansion coefficient an with the help of an integral proportional to
⃗
an ∝ ∫ r ) e−ikn ⋅⃗r .
dS Φ(⃗ (3.33)
∂V
Using this approach, given appropriate boundary conditions on ∂V , we can uniquely

determine the expansion coefficient an . The basic idea for the series expansion
approach to the solution of boundary-value problems in the context of the Laplace
equation can thus be given as follows: One writes a suitable expansion for the
potential Φ(⃗ r), hopefully in terms of mutually orthogonal functions (with respect
to a suitably defined scalar product, to be discussed in more detail in the following).
This renders the determination of the expansion coefficients unique. Then, one
projects out the expansion coefficients by evaluating overlap integrals like the one in
Eq. (3.33). Finally, all expansion coefficients are determined, and one can gradually
obtain better approximations to Φ(⃗ r ) by summing more terms of the (typically
infinite) series that describes the potential.
The main difficulties of the variational method and the series expansion method
are different; one difficulty lies in the implementation of structurally complex bound-
ary condition. In the case of the series expansion, the conditions imposed on
the expansion coefficients are defined uniquely only if the functions in the set are
orthogonal with respect to the integration measure (i.e., one must use a symmetry-
adapted basis). In the variational method, the form of the boundary conditions,
expressed in terms of the variational parameters, is easily tractable only if, again,
a symmetry-adapted basis is used.
In summary, we have three possibilities for solving Dirichlet problems in the
context of the Laplace equation. (i) The Dirichlet Green function method is very
convenient if we know the Dirichlet Green function. This function is uniquely
defined for a given reference volume, and incorporates the boundary conditions
naturally. (ii) The variational method is guaranteed to converge provided we use
a sufficiently large basis for our trial functions. (iii) The series expansion method
also is guaranteed to converge provided we use a complete set of functions, which
hopefully is orthogonal with respect to an integral measure. The latter two methods
will be illustrated by example calculations in the following Secs. 3.2 and 3.3, while
a nontrivial example for the conceptually difficult first method (Dirichlet Green
function) is relegated to Sec. 3.5.
3.2 Laplace Equation and Variational Calculations
3.2.1 Definition of the Functional Derivative
Let us first recall the formalism of multivariate functions, i.e., functions of many
variables. A function S of N arguments, denoted here as fi with i = 1, . . . , N , can
be written as
S = S(f1 , . . . , fN ) . (3.34)
If S just singles out a particular argument (say, of subscript j), then, for example,
∂S
S(f1 , . . . , fN ) = fj , = δij , (3.35)
∂fi
where δij is the Kronecker delta. For a function defined as the sum of the squares
of its (many) arguments, one has
N N
∂S
= F ′ (fi ) = 2 fi .
2
S = S(f1 , . . . , fN ) = ∑ F (fi ) = ∑ (fi ) , (3.36)
i=1 i=1 ∂fi
Let us now interpret the sum, multiplied by a “standard step size” Δx, as an
approximation to an integral,
N
2 2
∑ (fi ) Δx ≈ ∫ [f (x)] dx = S[f ] . (3.37)
i=1
Here, F is reinterpreted (with a slightly ambiguous notation) as a “function of the

function f .” A function of a function is commonly referred to as a “functional.”
Any function is specified by its values on an arbitrarily dense set of points,
which can be explained by the following correspondence (in a hopefully somewhat
self-explanatory notation),
△ △
{f1 , . . . , fN } = {f (x1 ), . . . , f (xN )} = f = f (x) . (3.38)
The generalization of Eq. (3.35) to the case of a continuous argument reads
respectively,
δS
S[f ] = ∫ f (x) δ(x − x′ ) dx′ = f (x′ ) , = δ(x − x′ ) , (3.39)
δf (x)
and for Eq. (3.36), the generalization reads as
δS
= F ′ (f (x)) = 2 f (x) .
2
S[f ] = ∫ dx F (f (x))2 dx = ∫ dx [f (x)] ,
δf (x)
(3.40)
This defines the functional derivative δS/δf (x) for a functional of the form
δS ∂F
S[f ] = ∫ F (f (x)) dx , = ∣ = F ′ (f (x)) . (3.41)
δf (x) ∂f f =f (x)
An intuitive understanding is gained if we consider a discretized approximation
to the integral and consider a “standard step size” of Δx = 1 for the integral.
Alternatively, we may investigate the variation
δS = S[f + δf ] − S[f ] = ∫ [F (f (x) + δf (x)) − F (f (x))] dx (3.42)
≈ ∫ F ′ (f (x)) δf (x)) dx = ⟨F ′ (f (x)) , δf (x)⟩ , (3.43)

and define the scalar product of functions, as the scalar product in the sense of
Hilbert space vectors, i.e. ⟨f, g⟩ ≡ ∫ dx f (x) g(x). Then, we can define the functional
derivative as the expression multiplying δf (x) under the integral sign, i.e.
δS ∂F
≡ ∣ = F ′ (f (x)) . (3.44)
δf (x) ∂f f =f (x)
A more systematic way of defining the functional derivative is as follows. Because
∂S/∂f (x) probes the function f (x) near x, we can formulate it as
δS F (f (x′ ) + δ(x′ − x)) − F (f (x′ ))
= lim ∫ dx′ . (3.45)
δf (x) →0 V
Of course, if F = F (f (x), f ′ (x)) is a function which depends not only on f itself,
but also, on the derivative f ′ , then we can calculate the variation as follows,
δS F (f (x′ ) + δ(x′ − x), f ′ (x′ ) + δ ′ (x′ − x)) − F (f (x′ ), f ′ (x′ )) ′
= lim ∫ dx ,
δf (x) →0 V
(3.46)
i.e., if the variation of f (x′ ) is given by the Dirac-δ term δ(x′ −x), then the variation
of f ′ (x′ ) is given by the term δ ′ (x′ − x). Below we shall carry out the variation
of a functional S[Φ], which is proportional to the electrostatic field energy, and we
shall find that the fulfillment of the Laplace equation is equivalent to the vanishing
of the variation. Here, Φ = Φ(⃗ r ) is the electrostatic potential.
3.2.2 Variational Principle and Euler–Lagrange Equations

In order to understand basic aspects of the variational technique, it is instructive
to consider the functional derivative of a functional S = S[ψ], where ψ = ψ(⃗ r)
replaces f as defined in Sec. 3.2.1 as the argument of the functional. So, ψ is to be
understood as a function of a vector-valued argument r⃗. More formally, we consider
the variation of a functional S,
S [ψ] = ∫ F (ψ (⃗ ⃗ (⃗
r ) , ∇ψ r)) d3 r . (3.47)
V
⃗ → ∇ψ+
The variation ψ → ψ+δψ implies the variation ∇ψ ⃗ ∇δψ,
⃗ ⃗ = ∇[δψ]
so that δ ∇ψ ⃗
⃗
is the variation of ∇ψ. Thus, the variation δS of S is found to be
⃗ + δ ∇ψ]
δS = S [ψ + δψ, ∇ψ ⃗ − S [ψ, ∇ψ]
⃗ , (3.48)
for small δψ. To first order in δψ, the variation δS reads as

∂F ∂F
δS = ∫ [ δψ (⃗ ⃗
r) + ∇δψ ] d3 r , (3.49)
V ∂ψ ⃗
∂ ∇ψ
where in Cartesian coordinates,
3
∂F ∂ ∂F
⃗ [δψ] ⋅
∇ ≡ ∑( δψ) . (3.50)
⃗
∂ ∇ψ i=1 ∂xi ∂(∂ψ/xi )
Using an integration by parts, one obtains

∂F ∂F ∂F
δS = ∫ d3 r (δψ (⃗
r) ⃗ ⋅ (δψ
+∇ ) − δψ(⃗ ⃗⋅
r) ∇ ). (3.51)
V ∂ψ ⃗
∂ ∇ψ ⃗
∂ ∇ψ
The divergence theorem implies that
∂F ∂F ∂F
δS = ∫ d3 [ ⃗⋅
−∇ ] δψ(⃗ r) (
r ) + ∫ dS δψ(⃗ ⋅̂
n) . (3.52)
V ∂ψ ⃗
∂ ∇ψ ∂V ⃗
∂ ∇ψ
Generally, the values of the function ψ (⃗r ) are uniquely specified on the boundary
∂V of the volume. Only those variations δψ (⃗ r) that vanish on the boundary are
permitted. If we could even vary the boundary conditions, then the variational
problem would be trivial: we would just choose the boundary conditions to be such
that F vanishes on the boundary (and perhaps everywhere), or even shrink the
volume V to zero. The functional S immediately vanishes and becomes optimal.
Under the assumption that δψ(⃗ r ) = 0 for r⃗ ∈ ∂V , the functional S has an
extremum when
3
∂F ∂ ∂F
r) [
δS = ∫ d3 r δψ(⃗ −∑ ] =0 (3.53)
V ∂ψ i=1 ∂xi ∂ (∂ψ/∂xi )
for all “permissible” variations δψ (⃗
r).
So, the requirement that the first-order variation vanishes implies that ψ (⃗
r)
must be a solution of the familiar Euler-Lagrange equation,
3
∂F ∂ ∂F ∂F
=∑ ⃗
=∇ . (3.54)
∂ψ i=1 ∂xi ∂ (∂ψ/∂xi ) ⃗
∂ ∇ψ(⃗r)
In order to determine whether the extremum is a minimum, maximum, or stationary
point, the second order term in δψ must be calculated.
The Euler–Lagrange equation (3.54) corresponds to a situation where the func-
tion ψ is a function of r⃗, that is, of three arguments x = x1 , y = x2 and z = x3 . E.g.,
for a functional of the form (3.29),
2E
S[Φ] = ⃗ r ))2 ,
= ∫ d3 r (∇Φ(⃗ (3.55)
0 V
the Euler–Lagrange equations can be verified to read

⃗ 2 Φ(⃗
∇ r) = 0 , (3.56)
which is simply equal to the Laplace equation.

Other cases also are of practical interest. For example, let us consider the
r (t)] = ∫ L(⃗
variation of S[⃗ r(t), r⃗˙ (t)) dt. Here, the argument function r⃗ = r⃗(t) has
three components but only one argument. This situation yields
3
∂L ∂ ∂L
=∑ . (3.57)
r(t) i=1 ∂t ∂ r⃗˙ (t)
∂⃗
For the classical action of a single particle moving under the influence of a potential
V = V (⃗
r ), the Lagrange function reads as
m ˙2
L(⃗r(t), r⃗˙ (t)) = r⃗ (t) − V (⃗
r (t)) . (3.58)
2
Under variation, the Euler–Lagrange equations then become the Newtonian equa-
tions of classical mechanics,
∂
⃗(t) = − V (⃗
m r¨ ⃗ (⃗
r ) = −∇V r (t)) . (3.59)
∂⃗r
3.2.3 Variations with Constraints

In an extension of the variational calculations, the function ψ (⃗r ) is required to
satisfy constraints in addition to the boundary conditions. Generally, these have an
integral form,
Cα [ψ] = ∫ Kα (ψ (⃗ ⃗ (⃗
r ) , ∇ψ r )) d3 r = Dα , (3.60)
V
where {Dα , α = 1, ..., N } is a set of N constants. This places restrictions on the
δψ (⃗
r) used in the variation of ψ (⃗ r) in the previous formalism. Instead of
S [ψ] = ∫ F (ψ (⃗ ⃗ (⃗
r ) , ∇ψ r)) d3 r , (3.61)
V
we now use the functional
N
S [{λα }α=1,...,N ; ψ] = ∫ F (ψ (⃗ ⃗ (⃗
r) , ∇ψ r )) d3 r + ∑ λα (Cα [ψ] − Dα ) . (3.62)
V α=1
Varying this functional with respect to the λα , which are normal arguments, we
obtain the N constraints back,
∂
S [{λα }α=1,...,N ; ψ] = Cα [ψ] − Dα = 0 . (3.63)
∂λα
Proceeding as before to set the δCα = 0, we have α = 1, 2, 3, . . . , N equations. Vari-
r ) leads to the modified variational equation,
ation with respect to ψ(⃗
3
∂F ∂ ∂F
−∑ + ∑ λα uα (⃗
r) = 0 , (3.64)
∂ψ i=1 ∂xi ∂ (∂ψ/∂xi ) α
where ψ is the function to be determined, and
3
∂Kα ∂ ∂Kα
uα (⃗
r) = −∑ (3.65)
∂ψ i=1 ∂x i ∂ (∂ψ/∂xi)
is a set of “vectors” in a function space. Let us mention a subtle point. The

variation of the functional (3.60) with respect to the parameters λα gives us back
the constraints, given in Eq. (3.63). If the variation is carried out to respect the
constraints, then we must demand that δψ (⃗ r) lie in the subspace orthogonal to the
uα (⃗
r ), in the sense that Cα [ψ+δψ] = Cα [ψ], which implies that d3 r uα (⃗
r ) δψ(⃗
r ) = 0.
The “expansion coefficients” λα are constants, called Lagrange multipliers, which
are determined such that the constraint equations are satisfied.
3.2.4 Second Functional Derivative and Hilbert Space

Let’s consider once more the variation of the functional
2
⃗ r )) ,
S[Φ] = ∫ d3 r (∇Φ(⃗ (3.66)
V
this time using an explicit calculation, including the second-order variation. We

calculate
2
⃗ r) + ∇δΦ(⃗
S[Φ + δΦ] = ∫ d3 r (∇Φ(⃗ ⃗ r ))
V
⃗ r))2 + 2 ∇Φ(⃗
= ∫ d3 r ((∇Φ(⃗ ⃗ r) ∇δΦ(⃗
⃗ ⃗
r) + (∇δΦ(⃗
2
r)) )
V
⃗ r))2 − 2 ∇
= ∫ d3 r ((∇Φ(⃗ ⃗ 2 Φ(⃗
r) δΦ(⃗
r ) + δΦ(⃗ ⃗ 2 ) δΦ(⃗
r ) (−∇ r )) . (3.67)
V
So, the following result actually is exact (no higher-order corrections),
⃗ 2 Φ(⃗
S[Φ + δΦ] − S[Φ] = −2 ∫ d3 r (∇ r )) δΦ(⃗ ⃗ 2 ) δΦ(⃗
r ) (−∇
r ) + ∫ d3 r δΦ(⃗ r ) , (3.68)
where we ignore boundary terms due to the obvious restrictions on the variation
δΦ(⃗r ), which is assumed to vanish on the boundary. The first- and second-order
variational derivatives read as follows,
δS ⃗ 2 Φ(⃗ δ2S ⃗2 .
= −2∇ r) , = −2δ (3) (⃗
r − r⃗′ ) ∇ (3.69)
r)
∂Φ(⃗ r′ )
r) ∂Φ(⃗
∂Φ(⃗
The first functional derivative should vanish at the extremum. This implies that
⃗ 2 Φ(⃗
∇ r) = 0 , (3.70)
at the minimum of the functional J, which is just the Laplace equation. How-
ever, we may ask, as well, how we intend to show that the second-order variation,
i.e., the second term on the right hand side of Eq. (3.68), is positive. The second
functional derivative is distribution-valued; it involves the Dirac-δ function (distri-
bution). First we need to define what the positivity of the operator −∇ ⃗ 2 could mean.
The operator −∇ ⃗ acts on a space of functions which, according to Chaps. 6 and 7
2
of Ref. [13], is infinite-dimensional and constitutes a so-called Hilbert space. The

determinant of an operator acting on an infinite-dimensional space is a so-called
Fredholm determinant; however, even the positivity of the Fredholm determinant
does not imply that the operator −∇ ⃗ 2 is positive. Namely, we can understand the
operators −∇ ⃗ in terms of its matrix representation, as it acts on a suitable set of
2
basis functions that span the Hilbert space. If we can show that all the eigenvalues
of −∇ ⃗ 2 corresponds
⃗ 2 in that basis are positive, then the matrix representation of −∇
to a positive definite matrix, and in that sense, we can understand the positivity of
the operator −∇ ⃗ 2 that occurs in the second functional derivative. It should be pos-
itive for the solution of the Laplace equation given in Eq. (3.70) to be a minimum
of the functional S[Φ] given in Eq. (3.66).
Let us consider a one-dimensional analogue in order to heuristically show that

⃗ 2 is a positive definite operator. On any finite interval {−L, L}, any function can
−∇
uniquely be decomposed into
∞
in
f (x) = ∑ an exp ( x) , (3.71)
n=−∞ L
⃗ 2 are the functions
in the sense of a Fourier decomposition. The eigenvectors of −∇
exp (i n x/L), because they fulfill
in n 2 in
⃗ 2 exp (
−∇ x) = ( ) exp ( x) . (3.72)
L L L
The eigenvalues are
n 2
( ) , n = −∞, . . . , ∞ . (3.73)
L
If an operator only has positive eigenvalues, then it is positive definite. Generalizing
from the one-dimensional example case to three dimensions, and letting L → ∞, we
can thus conclude that all eigenvalues of the operator −∇ ⃗ 2 are positive, which again
confirms that any solution of the Laplace equation indeed minimizes the functional
S given in Eq. (3.66).
3.2.5 Variational Calculation of the Capacitance of Plates

We have learned that variational calculus corresponds to the search for a solution
of an optimization problem in the infinite-dimensional Hilbert space of functions.
One may require the functions to fulfill additional conditions, such as being square-
integrable, or impose additional boundary-conditions, or other restrictions. Tem-
porarily, we simplify the problem a bit, and limit the function space to a finite
number of parameters. In the simplest case, we let our potential be Φ (⃗ r ) = w (β, r⃗),
where β is a single free parameter, and the class of functions Φ (⃗ r ) = w (β, r⃗) is
chosen to satisfy the boundary conditions on Φ (⃗r ) but is otherwise arbitrary.
The functional equation now is a function of only a vector of variables β, ⃗ and
we therefore replace square brackets by round brackets,
⃗ = ∫ d3 r F (⃗
S(β) ⃗ r⃗), ∇w(
r , w(β, ⃗ β,⃗ r⃗)) . (3.74)
V
⃗ r⃗) → Φ(⃗
If w(β, r ) is just a “trial function” for the electrostatic potential, then
⃗ = S(β)
E(β) ⃗ = 0 ∫ d3 r [∇w(
⃗ β, ⃗ r⃗)]2 . (3.75)
2
We now aim to find the extremum of S(β). ⃗ Let us suppose that there are boundary
conditions, which relate to the surfaces ∂V = ∂V1 ∪ ∂V2 of V ,
⃗ r⃗1 )∣
w(β, = 0, ⃗ r⃗2 )∣
w(β, = V0 , (3.76)
r⃗
1 ∈∂V1 r⃗
2 ∈∂V2
which can be implemented into our trial function w = w(β, ⃗ r⃗) via the use of La-
grange multipliers. In the presence of boundary conditions, some of the variational
⃗ constitute Lagrange multipliers. We then find the

parameters (components of β)
⃗ ⃗
extremum value of β, say βmin , for which
∂E
∣ = 0, i = 1, 2, . . . , n . (3.77)
∂βi β=
⃗ β⃗min
If the matrix of second derivatives with entries

∂2E
Mij = ∣ (3.78)
∂βi ∂βj β=
⃗ β⃗min
is positive definite (all eigenvalues of M are positive), then the type of the extremum
is as required, and w(β, ⃗ r⃗) is an approximation to Φ(⃗ r ).
The variational program should now be illustrated by way of an example. We
consider the potential between capacitor plates at z = 0 and z = d. The (large) plates
are held at zero potential and at V0 , respectively. We start from the following ansatz
for the potential inside the plates,
w(A, B, C, r⃗) = w(z) = A + B z + C z 2 . (3.79)
The energy E stored in the field assumes the role of a functional
0 2 3 0 z=d ∂w 2
E = E(A, B, C) = ∫ ∣ ⃗
∇w(A, B, C; ⃗
r )∣ d r = S ∫ ( ) dz
2 V 2 z=0 ∂z
0 S
= d (3B 2 + 6dBC + 4d2 C 2 ) . (3.80)
6
Here, S is the surface area of the capacitor plates. The boundary conditions are
w(z = 0) = A = 0 , w(z = d) = V0 = A + B d + C d2 . (3.81)
Including Lagrange multipliers, the functional thus is
0 S
E = E(A, B, C, λ1 , λ2 ) = d (3B 2 + 6 d B C + 4d2 C 2 )
6
+ λ1 A + λ2 (A + B d + C d2 − V0 ) . (3.82)
We now need to perform variations with respect to the five components of the vector
β⃗ = (A, B, C, λ1 , λ2 ) . (3.83)
The two equations
∂ ∂
E = 0, E = 0, (3.84)
∂λ1 ∂λ2
just give us the boundary conditions back. The other derivatives read
∂
E = λ1 + λ2 ,
∂A
∂
E = d [S 0 (B + C d) + λ2 ] ,
∂B
∂ d2
E= (S 0 (3B + 4C d) + 3λ2 ) . (3.85)
∂C 3
The solution of the system of equations

∂ ∂ ∂ ∂ ∂
E= E= E= E= E=0 (3.86)
∂A ∂B ∂C ∂λ1 ∂λ2
reads
V0 0 S V0
A = C = 0, B=, λ1 = −λ2 = . (3.87)
d d
Our trial function turned solution function thus becomes
V0 z
w(z) = wmin (z) = . (3.88)
d
The value of the energy at the extremum (minimum) becomes
0 z=d ∂w
min
2
1 0 S 2
Emin = S∫ ( ) dz = V . (3.89)
2 z=0 ∂z 2 d 0
Furthermore, from the equation
1
Emin = C V02 , (3.90)
2
we read off the capacitance
0 S
C= , (3.91)
d
which is of course a well-known result. In our example calculation, the exact result
is contained in the set of trial functions and is recovered by the variational calculus.
We still need to check that the extremum really is a minimum (not a maximum
or a saddle point) of the energy functional. In order to write the second-order
variation of the energy functional in compact form, we write
β1 = A , β2 = B , β3 = C , β4 = λ1 , β5 = λ2 . (3.92)
Furthermore,
5
∂E 1 5 5 ∂2E
δE = ∑ δβi + ∑ ∑ δβi δβj (3.93)
i=1 ∂βi 2 i=1 j=1 ∂βi ∂βj
up to the second order in the δβi . At the extremum values for the parameters given
in Eq. (3.87), we have
∂E 1 5 5 ∂2E
∣ = 0, δE = ∑∑ ∣ δβi δβj . (3.94)
∂βi β=
⃗ β⃗min 2 i=1 j=1 ∂βi ∂βj β=
⃗ β⃗min
Let us define the matrix M with components
⎛0 0 0 1 1⎞
⎜ 0 0 Sd 0 Sd2 0 d ⎟
∂2E ⎜ ⎟
Mij = ∣ =⎜ 2 4 3 2⎟
⎜ 0 0 Sd 3 0 Sd 0 d ⎟ . (3.95)
∂βi ∂βj β=
⃗ β⃗min ⎜ ⎟
⎜1 0 0 0 0⎟
⎝1 d d2 0 0 ⎠
According to a theorem from linear algebra, in order to establish that a matrix

is positive definite, it is sufficient to show that all its leading principal minors are
positive (Sec. 6.5 of Ref. [14], and Chap. 7 of Ref. [15]). The leading principal
minors are given as the determinants of the leading (k × k) submatrices of a given
matrix M , i.e., as
det (M1 ) = det (0) = 0 ,
0 0
det (M2 ) = ( ) = 0,
0 0 Sd
⎛0 0 0 ⎞
det (M3 ) = det ⎜ 0 0 Sd 0 Sd2 ⎟ = 0 ,
⎝ 0 0 Sd2 4 0 Sd3 ⎠
3
⎛0 0 0 1⎞
⎜ 0 0 Sd 0 Sd2 0 ⎟ 1 2 2 4
det (M4 ) = det ⎜ ⎟
⎜ 0 Sd2 4 Sd3 0 ⎟ = − 3 0 S d ,
⎜ 0 3 0 ⎟
⎝1 0 0 0⎠
⎛0 0 0 1 1⎞
⎜ 0 0 Sd 0 Sd2 0 d ⎟
⎜ ⎟ 1
det (M5 ) = det ⎜ 2 4 3 2⎟
⎜ 0 0 Sd 3 0 Sd 0 d ⎟ = 3 0 S d .
5
(3.96)
⎜ ⎟
⎜1 0 0 0 0⎟
⎝1 d d 2
0 0⎠
The fourth principal minor determinant is negative, indicating that M cannot be
positive definite: Namely, the product of the eigenvalues of M4 must be negative,
indicating the presence of at least one negative eigenvalue (of M4 ). A further
indication can be found by considering the case (setting aside physical units for the
moment) d = 1, S = 1, and considering the limit 0 → 0. Then,
⎛0 00 1 1⎞
⎜0 00 0 1⎟
⎜ ⎟
M →D=⎜
⎜0 00 0 1⎟
⎟, ⃗i = di w
Dw ⃗i 1w , i = 1, . . . , 5 . (3.97)
⎜ ⎟
⎜1 00 0 0⎟
⎝1 11 0 0⎠
(We do not consider the explicit form of the vectors w ⃗i .) The eigenvalues are
√ √
di = ± 2 ± 2 , i = 1, . . . , 4 , d5 = 0 . (3.98)
Two of these are negative, namely,
√ √ √ √
d2 = − 2 + 2 , d4 = − 2 − 2 , (3.99)
indicating again that D cannot be positive definite.
What happened here? We started from a minimization problem which, as we
showed in Sec. 3.2.4, concerns an operator which is manifestly positive definite,
and now we see that some eigenvalues of the matrix of second partial derivatives
are negative. We notice that the first three minors det (Mi ) with i = 1, 2, 3 are
nonnegative. These concern the variational parameters A, B, and C, which enter the
ansatz for the potential (3.79). The introduction of the Lagrange multipliers changes
the nature of the problem; in fact, it is easy to see that the term λ1 A in Eq. (3.82)
does not attain any global (or even local) minimum near the values attained by the
parameters in the solution (3.87). Or, in other words, one observes that the function
f (λ1 , A) = λ1 A has a hyperbolic structure as a function of its two arguments λ1
and A and therefore cannot obtain a minimum. The Lagrange multipliers are not
“physical” terms that need to be minimized; they are just included in order to
recover the boundary conditions.
If we use the fact that A = 0 by virtue of the boundary conditions, then the
“physical” portion of M can be identified as the submatrix of M consisting of the
second and third rows and columns, namely,
0 Sd 0 Sd2 1
Mphys = ( ), det(Mphys ) = (0 Sd2 )2 > 0 . (3.100)
0 Sd2 43 0 Sd3 3
All the leading principal minors of the matrix Mphys are positive; so it is indeed
positive definite.
3.2.6 Variational Calculation for Coaxial Cylinders

Our solution for the energy stored in a plane-plate capacitor was exact, because
the exact solution was contained in the space of the trial functions. By contrast,
the problem of two coaxial, long cylinders is not so trivial. For a very cylindrical
capacitor, one may calculate the potential using Gauss’s law. Here, we would like to
use this exactly soluble problem in order to demonstrate the utility of the variational
method for a case where the exact solution is not contained in the space of trial
functions.
Consider two coaxial cylinders, aligned along the z axis, with radii b and c,
extending from z = −L/2 to z = L/2. We use cylindrical coordinates and set
x = ρ cos ϕ , y = ρ sin ϕ , z = z. (3.101)
In a numerical example, we will later use the inner surface of the cylinder to lie at
b = 0.2 cm and the outer surface to be at c = 4b = 0.8 cm. Let x2 + y 2 = ρ2 and let us
assume that the z axis defines the symmetry axis of the cylinders. Because of the
symmetry of the problem, the potential should only depend on the radial variable
ρ. Our trial potential therefore has the form
w(A, B, C, ρ) = A + B ρ + C ρ2 . (3.102)
Our task is to estimate the potential between the plates, and eventually the capac-
itance per unit length. The inner surface is grounded, and the outer surface is held
at a potential V0 ,
w(A, B, C, b) = A + B b + C b2 = 0 , w(A, B, C, c) = A + B c + C c2 = V0 . (3.103)
We work in cylindrical coordinates, as defined in Eq. (3.101). In order to calculate

the energy functional, we need the gradient operator in cylindrical coordinates. It
is given by
∂Φ 1 ∂Φ ∂Φ
⃗ = êρ
∇ + êϕ + êz . (3.104)
∂ρ ρ ∂ϕ ∂z
The energy stored in the space in between the coaxial cylinders of radii b and c is
0
E= ∫ ⃗ r )]2 ρ dρ dϕ dz ,
[∇Φ(⃗ (3.105)
2 b≤ρ≤c
where ρdρ dϕ dz is the volume element in cylindrical coordinates. We pull out the
constant prefactor 0 /2 and a further factor 2πL from E; these do not affect the
minimization process. The result is the functional S,
0
E = 2πLS = 0 πL S , (3.106)
2
which can be expressed as
2π L/2 c
1 ∂w 1 ∂w ∂w
S= ∫ dϕ ∫ dz ∫ dρ ρ [êρ + êϕ + êz ]
2πL ∂ρ ρ ∂ϕ ∂z
0 −L/2 b
∂w 1 ∂w ∂w
× [êρ + êϕ + êz ]
∂ρ ρ ∂ϕ ∂z
2π L/2 c
1 ∂w 2
= ∫ dϕ ∫ dz ∫ dρ ( ) ρ
2πL ∂ρ
0 −L/2 b
1 4
= B 2 (c2 − b2 ) + BC (c3 − b3 ) + C 2 (c4 − b4 ) . (3.107)
2 3
The two boundary conditions at ρ = b and at ρ = c are added with the help of
Lagrange multipliers,
1 4
S(A, B, C, λ1 , λ2 ) = B 2 (c2 − b2 ) + BC (c3 − b3 ) + C 2 (c4 − b4 )
2 3
+ λ1 (A + B b + C b2 ) + λ2 (A + B c + C c2 − V0 ) . (3.108)
The vector β of variational parameters and the extremum condition read as follows,
∂S
β⃗ = (A, B, C, λ1 , λ2 ) , = 0, i = 1, . . . , 5 . (3.109)
∂βi
The system of equations
∂
S = λ1 + λ2 = 0 ,
∂A
∂ 4
S = B (c2 − b2 ) + C (c3 − b3 ) + λ1 b + λ2 c = 0 ,
∂B 3
∂ 4
S = B(c3 − b3 ) + 2 C (c4 − b4 ) + λ1 b2 + λ2 c2 = 0 ,
∂C 3
∂
S = A + b (B + b C) = 0 ,
∂λ1
∂
S = A + c (B + c C) − V0 = 0 , (3.110)
∂λ2
has the solution

b (b + 2 C) 1
A= V0 , B= V0 ,
b 2 − c2 c−b
1 2 b 2 + 4 b c + c2
C = 2 2 V0 , λ1 = −λ2 = − V0 . (3.111)
b −c 3 b 2 − c2
The solution w = wmin therefore reads
(b − ρ) (b + 2 c − ρ)
wmin (ρ) = V0 . (3.112)
b 2 − c2
The value of S at the minimum is
b 2 + 4 b c + c2 2
Smin = V0 . (3.113)
c2 − b 2
Restoring the prefactors, this translates into a field energy of
0 b 2 + 4 b c + c2 2
Emin = 2πL Smin = 0 πL V0 . (3.114)
2 c2 − b 2
Solving the equation
1
Emin = C V02 (3.115)
2
for the capacitance C, we obtain
2πL 0 b2 + 4 b c + c2
Cmin = . (3.116)
3 c2 − b 2
It remains to show that the optimum solution for the potential within the space of
our trial functions, wmin (ρ), approximates the exact solution
(b − ρ) (b + 2 c − ρ) ln(ρ/b)
wmin (ρ) = V0 ≈ wexact (ρ) = V0 . (3.117)
b 2 − c2 ln(c/b)
Furthermore, the approximate and exact formulas for the capacitance read
2πL 0 b2 + 4 b c + c2 2πL 0
Cmin = ≈ Cexact = . (3.118)
3 c2 − b 2 ln(c/b)
For b = 0.2 cm and c = 0.8 cm, the approximate and exact formulas for the capaci-
tance read
Cmin = 4.608 L 0 , Cexact = 4.532 L 0 , (3.119)
with a 1.6 % deviation, which is quite good for a rational approximation with three
parameters (see also Fig. 3.1).
Fig. 3.1 The variational solution wmin for the potential is shown for
b = 0.2 cm and c = 0.8 cm. The optimal approximation wmin (ρ) given
in Eq. (3.117) (dashed curve) approximates the exact solution quite
well. The latter is given by the solid curve, with V (ρ) = wexact (ρ).
3.2.7 Exact Integration of the Capacitance of Coaxial Cylinders
The variational calculation of the potential in between the capacitor plates led to
the polynomial approximation wmin (ρ) [see Eq. (3.117)] to the exact result
ln(ρ/b)
Φ(ρ) = V0 , (3.120)
ln(c/b)
which is valid in the limit of long cylinders, L → ∞, where again, L is the cylinder
length measured parallel to the symmetry axis. When L is finite, the exact result
for the coaxial cylinders is non-trivial. It is extremely instructive to perform an
explicit and exact integration over the charge distributions. The two cylindrical
surfaces are assumed to be located at distances ρ = b and ρ = c from the cylinder
axis. We assume a constant charge density of −σ0 on the inner cylinder at ρ = b and
a constant charge density of +σ0 b/c on the outer cylinder at ρ = c, in order to make
the arrangement electrostatically neutral. Let us briefly comment on the relation of
the constant surface charge and the equipotential surfaces used in our variational
problem, which was based on the additional assumption of a long cylinder L → ∞.
Namely, the constant surface charge density does not necessarily imply that the
cylindrical surfaces are equipotential surfaces. It is only in the limit of large L that
the equipotential surfaces will be surfaces of constant surface charge density. For
finite L, we here assume the constant surface charge density.
The solution can be obtained as follows. We have formulated the problem in
terms of the surface charge distributions on the two capacitor plates, not in terms
of the boundary conditions of the potential. Therefore, we may use the Poisson
equation (2.21) with the Green function G(⃗ r, r⃗′ ) = 1/(4π0 ∣⃗
r − r⃗′ ∣) [see Eq. (2.23)]
in order to integrate the potential,
⃗ 2 Φ(⃗ 1
r) = ∫ d3 r′ G(⃗
Φ(⃗ r , r⃗′ ) ρ(⃗
r′ ) , ∇ r ) = − ρ(⃗
r) . (3.121)
0
Here, ρ(⃗r′ ) is the volume charge density corresponding to the uniform surface charge
density on the coaxial cylinder plates. A major difficulty is encountered in how
to express the volume charge density ρ(⃗ r ) in terms of the surface charge density
r ) = ±σ0 (here, we assume it to be uniform). This problem can be formulated
σ(⃗
more generally. Let σ(⃗ r ) be a surface charge density on the boundary r⃗ ∈ ∂V defined
by an implicit equation F (⃗ r ) = 0. In our case, we would simply have F (⃗ r) = ρ − b
and F (⃗r) = ρ − c for the inner and outer cylindrical surfaces. How is ρ(⃗ r ) related to
r ) and F (⃗
σ(⃗ r )? This problem is of general interest and needs to be discussed.
The answer is as follows. On the surface defined by F (⃗ r ) = 0, we have dF (⃗r) = 0
when the argument r⃗ is varied within the surface according to r⃗ → r⃗ + d⃗ s. Thus,
dF (⃗
r ) = F (⃗
r + d⃗
s) − F (⃗
r) = d⃗ ⃗ (⃗
s ⋅ ∇F r) = 0 , (3.122)
where d⃗ s is an infinitesimal displacement within the surface, and r⃗ is a point on
the surface. Because d⃗ ⃗ (⃗
s is arbitrary, ∇F r) must be normal to the surface. Now,
let ξ ̂
n be the displacement of a point from the surface at r⃗, where ξ measures the
distance from the surface. Then,
d3 r = dA⃗ ⋅ (dξ n
⃗ ) = [dS (⃗
r) ̂
n] ⋅ (dξ ̂
n) . (3.123)
Here, ̂
n is the surface normal. If r⃗ is a point on the surface, then, using definition
⃗ operator, one writes
of the ∇
r′ + ξ u
∂F (⃗ ̂)
⃗ (⃗
û ⋅ ∇F r) = ∣ , (3.124)
∂ξ ξ=0
where u ̂=̂
̂ is an arbitrary unit vector. If u t were a tangential vector to the surface,
then we would have dF (⃗ r ) = 0 in the sense of Eq. (3.122), with d⃗ s =̂t dξ. We
conclude that the modulus of dF (⃗ r) = 0 becomes maximal (as a function of the
angle between ∇F⃗ (⃗ r) and ̂n) when ̂n is the surface normal. We now choose ̂ u=n̂
⃗ (⃗
to be the unit normal, parallel (not antiparallel) to ∇F r ), and write
∂F (⃗
r+ξ ̂
n) ∂F (ξ)
⃗ (⃗
n̂ ⋅ ∇F r) = ∣ = ∣ , (3.125)
∂ξ ξ=0 ∂ξ ξ=0
with an appropriately defined function F = F (ξ) = F (⃗
r+ξ ̂
n). So,
∂F (ξ) ⃗ (⃗
(∇F r))
∣ ⃗ (⃗
= n̂ ⋅ ∇F r) = ⃗ (⃗
⋅ ∇F ⃗ (⃗
r ) = ∣∇F r) ∣ . (3.126)
∂ξ ξ=0 ⃗ (⃗
∣∇F r) ∣
We have thus reduced our problem to a one-dimensional problem, replacing r⃗ → ξ.
Application of Eq. (1.34) then shows that
∂F (ξ)
δ(ξ) = δ(F (ξ)) ∣ ⃗ (ξ)∣ = δ(F (⃗
∣ = δ(F (ξ)) ∣∇F ⃗ (⃗
r )) ∣∇F r))∣ . (3.127)
∂ξ
Because ξ measures the distance from the surface, we finally obtain a suitable
expression for the volume charge density,
r ) = σ(⃗
ρ(⃗ r ) δ(ξ) = σ(⃗
r ) δ(F (⃗ ⃗ (⃗
r)) ∣∇F r)∣ . (3.128)
In our case, the two surfaces are simply given by
F (⃗
r) = ρ − b = 0 , F (⃗
r) = ρ − c = 0 . (3.129)
For the inner cylinder surface, we thus have
ρ(⃗ ⃗ − b)∣ .
r) = −σ0 δ (ρ − b) ∣∇(ρ (3.130)
In Cartesian coordinates, the gradient is expressed as
√
ρ = x2 + y 2 , ⃗ (⃗
∇F ⃗ − b) = êx √ x
r ) = ∇(ρ + êy √
x
, (3.131)
x +y
2 2 x + y2
2
and thus
⃗ − b)∣ = 1 ,
∣∇(ρ r ) = σ(⃗
ρ(⃗ r ) δ(ρ − b) = −σ0 δ(ρ − b) . (3.132)
r) = Φ(ρ, ϕ, z) in cylindrical coordinates. The source
We denote the potential as Φ(⃗
point r⃗′ is expressed as
r⃗′ = êx ρ′ cos ϕ′ + êy ρ′ sin ϕ′ + êz z ′ , (3.133)
Of special interest is the point r⃗ = êx ρ which is at a distance ρ from the z axis.
Then,
r⃗ − r⃗′ = êx (ρ − ρ′ cos ϕ′ ) + êy (−ρ′ sin ϕ′ ) − êz z ′ . (3.134)
This implies that
√
r − r⃗′ ∣ =
∣⃗ ρ2 + ρ′2 − 2ρ ρ′ cos ϕ′ + z ′2 . (3.135)
We thus need to calculate the following contribution to the potential from the inner
cylindrical surface at ρ = b, taking into account that the observation point r⃗ = êx ρ
has cylindrical coordinates ϕ = 0 and z = 0,
∞ 2π L/2
1 ′ ′ ′ ′ ′ 1
Φinner(ρ, 0, 0) = ∫ dρ ρ ∫ dϕ ∫ dz (−σ0 ) δ(ρ − b)
4π0 r − r⃗′ ∣
∣⃗
0 0 −L/2
2π L/2
σ0 ′ ′ b
= − ∫ dϕ ∫ dz √ 2 2
4π0
0
ρ + b − 2bρ cos ϕ′ + z ′2
−L/2
2π L/2
σ0 ′ ′ b
= − ∫ dϕ ∫ dz √ 2 2
2π0
0
ρ + b − 2bρ cos ϕ′ + z ′2
0
σ0 b
2π
√ z ′ =L/2
′ ′
= − ∫ dϕ [ln (2 (z + ρ2 + b2 − 2bρ cos ϕ′ + z ′2 ))]z′ =0 .
2π0
0
(3.136)
In principle, this expression already expresses the exact result for finite L in terms
of an integral, which has to be evaluated numerically. The full result Φ(ρ, 0, 0)
is obtained after adding the contributions from the inner and outer surfaces,
Φ(ρ, 0, 0) = Φinner(ρ, 0, 0) + Φouter (ρ, 0, 0), where Φouter (ρ, 0, 0) is obtained from
Φinner (ρ, 0, 0) by the replacement b → c and a sign change in σ0 . Then,
σ0 c
2π
√ z ′ =L/2
′ ′
Φ(ρ, 0, 0) = ∫ dϕ [ln (2 (z + ρ2 + c2 − 2cρ cos ϕ′ + z ′2 ))]z′ =0
2π0
0
σ0 b
2π
√ z ′ =L/2
′ ′ 2 + b2 − 2bρ cos ϕ′ + z ′2 ))]
− [ln (2 (z +
2π0 ∫
dϕ ρ . (3.137)
z ′ =0
0
It is difficult to express these integrals in closed analytic form; a numerical evaluation

of Eq. (3.137) for given parameters b, c and L is to be preferred.
Let us now consider the limit L → ∞ in Eq. (3.136), which corresponds to the
situation considered in our variational calculation described in Sec. 3.2.6. At the
upper limit, the logarithm in the integrand of Eq. (3.136) assumes the form
√ b2 + ρ2 − 2bρ cos ϕ′
z′ = L 1
ln (2 (z ′ + ρ2 + b2 − 2bρ cos ϕ′ + z ′2 )) = 2
ln(2L) ++ O( 3).
L2 L
(3.138)
The constant ln(2L) is independent of ρ and can be absorbed into an additive
constant potential Φ0 which can always be added to any electrostatic potential
without changing the physics. The term of order 1/L2 vanishes in the limit L → ∞,
which we are interested in here. We can thus replace, in the integrand of Eq. (3.136),
the logarithm by its value at the lower limit z ′ = 0,
√ z ′ =L/2 √
[ln (2 (z ′ +
ρ2 + b2 − 2bρ cos ϕ′ + z ′2 ))] ′ → − ln (2 ρ2 + b2 − 2bρ cos ϕ′ ) .
z =0
(3.139)
The remaining L-independent integral
σ0 b
2π
√
′
Φinner (ρ, 0, 0) = ∫ dϕ ln (2 ρ2 + b2 − 2bρ cos ϕ′ ) (3.140)
2π0
0
is still difficult; in particular, a direct evaluation leads to an elliptic function.

The integral (3.140) becomes easier if we first differentiate with respect to ρ,
then integrate over ϕ′ , and then, integrate back again with respect to ρ. This only
adds a physically irrelevant constant to the potential. In particular, we have
2π ′
∂ σ0 ′ b (2ρ − 2b cos ϕ )
Φ(ρ, 0, 0) = ∫ dϕ 2 . (3.141)
∂ρ 4π0 b + ρ2 − 2bρ cos ϕ′
0
A partial-fraction decomposition of the integrand leads to

b (2ρ − 2b cos ϕ′ ) b b (b2 − ρ2 )
′
= − . (3.142)
b + ρ − 2bρ cos ϕ
2 2 ρ ρ (b + ρ2 − 2bρ cos ϕ′ )
2
The integral of this expression with respect to ϕ′ is

bϕ′ 2b b+ρ ϕ′
− arctan ( tan ( )) . (3.143)
ρ ρ b−ρ 2
At face value, we have
ϕ′ =2π
b ϕ′ 2b b+ρ ϕ′ ? 2π b 2b
[ − arctan ( tan ( ))] = − [arctan (0) − arctan (0)] ,
ρ ρ b−ρ 2 ϕ′ =0 ρ ρ
(3.144)
where we simply insert the upper and lower integration limits; this would lead to
a value of 2π b/ρ for the integral. However, this result is incorrect. At ϕ′ = π, the
second term with the arctan function receives a jump by
ϕ′ =π+
b+ρ ϕ′ π π
[− arctan ( tan ( ))] = − − (+ ) = −π , b < ρ, (3.145)
b−ρ 2 ϕ′ =π− 2 2
(downward) because of the singularity of the tangent function. In order to define
the integral so that it becomes smooth at ϕ′ = π, we have to add a compensating
term in the form of a Heaviside Θ step function,
′ b b (b2 − ρ2 )
∫ dϕ ( − )
ρ ρ (b2 + ρ2 − 2bρ cos ϕ′ )
b ϕ′ 2b b+ρ ϕ′
= + [− arctan ( tan ( )) + π Θ(ϕ′ − π)] . (3.146)
ρ ρ b−ρ 2
Then,
2π b b (b2 − ρ2 ) 4πb
∫ dϕ′ ( − )= . (3.147)
0 ρ ρ (b + ρ2 − 2bρ cos ϕ′ )
2 ρ
The final result for the expression given in Eq. (3.141) therefore is
∂ σ0 4πb
Φinner (ρ, 0, 0) = , (3.148)
∂ρ 4π0 ρ
and so
σ0 b ρ
ln ( ) .
Φinner (ρ, 0, 0) = (3.149)
0 b
The argument of the logarithm has to be dimensionless for physical reasons; in
principle, one can ascertain that ln(ρ/C) is a valid integral for any constant length
C; our choice implements the condition Φ(ρ = b, 0, 0) = 0.
We recall that in view of Eq. (3.145), the result (3.149) is valid for ρ > b. On
the outer layer, we have a charge density +σ0 b/c in order to make the arrangement
neutral, and calculate its contribution to the potential.
At radial distance ρ, we have a corresponding integral which replaces Eq. (3.145)
by
ϕ′ =π+
c+ρ ϕ′ π π
[− arctan ( tan ( ))] = − (− ) = π , ρ < c, (3.150)
c−ρ 2 ϕ′ =π−
2 2
[the arctan(⋅) function jumps in the other direction here], so that

2π c c (c2 − ρ2 )
∫ dϕ′ ( − ) = 0, ρ < c. (3.151)
0 ρ ρ (c2 + ρ2 − 2cρ cos ϕ′ )
Hence,
⎧
⎪ (ρ < c)
⎪
⎪ 0
Φouter (ρ, 0, 0) = ⎨ σ0 b ρ . (3.152)
⎪
⎪ − ln ( ) (ρ > c)
⎪
⎩ 0 b
We thus verify that
⎧
⎪ σ0 b ρ
⎪
⎪ ln ( ) (b < ρ < c)
Φ(ρ, 0, 0) = Φinner (ρ, 0, 0) + Φouter (ρ, 0, 0) = ⎨ 0 b , (3.153)
⎪
⎪
⎪ 0 (ρ > c)
⎩
a result which could have been obtained by the application of the integral form
of Gauss’s law, given in Eq. (1.5a), to appropriate cylindrical surfaces inserted in
between the plates. In this case, only the inner cylinder matters as the “charge
inside” the Gaussian surface.
Using the result from Eq. (3.149) and solving the equation
σ0 b c Q c Q
V0 = Φ(c, 0, 0) − Φ(b, 0, 0) = ln ( ) = ln ( ) = , (3.154)
0 b 2πL0 b C
we obtain the capacitance as
2πL 0
C= , (3.155)
ln (c/b)
as already noted and used in Eq. (3.118). We recall that the area of the inner
cylinder is 2πb L, so that σ0 = Q/(2πb L). We thus verify Eqs. (3.117) and (3.118)
by an alternative calculation.
We have already stated that the approach based on Eq. (3.137) is easily gener-
alizable to cases where L is finite, e.g., within a numerical approach. One just has
to keep L finite at the places where we have taken the limit L → ∞ within the z
integration. Let us illustrate this procedure by calculating the first nonvanishing
correction term to the potential for large, but finite L. To this end, we keep the
term of relative order 1/L2 in Eq. (3.155) and carry out the integration over ϕ′ as
indicated. The same procedure has to be applied to the outer layer, i.e., we have
to replace b → c and (−σ0 ) → (+σ0 )b/c in the first line in Eq. (3.136). One finds
that the correction term of order 1/L2 to the potential (3.149) vanishes. Taking
into account the next order (1/L4 ), one obtains after some algebra
σ0 b ρ 6 σ0 b (b2 − c2 ) ρ2 Q ρ 3 Q (b2 − c2 ) ρ2
Φ(ρ, 0, 0) = ln ( ) + = ln ( ) + .
0 b L 4 0 2πL 0 b π L 5 0
(3.156)
This result leads to the following correction term for the capacitance,
2πL 0
C= . (3.157)
ln (c/b) − 6 (b2 − c2 )2 /L4
The denominator of the latter expression involves the difference of the potential
Φ(ρ, 0, 0) given in Eq. (3.156) for ρ = c and ρ = b.
3.3 Laplace Equation and Series Expansions
3.3.1 Coordinate Systems and Special Functions

After discussing the variational approach to the solution of the Laplace equation
⃗ 2 Φ(⃗
∇ r) = 0, we now turn our attention to the series expansion method. The basic
paradigm is that one writes down mutually orthogonal functions (with respect to a
particular integration measure); all of these fulfill the Laplace equation individually.
The expansion coefficients are fixed by the boundary conditions. The relevant
equations look fundamentally different in the Cartesian coordinate system (x, y, z),
in the spherical system (r, θ, ϕ), and in the cylindrical coordinates (ρ, ϕ, z).
Suppose we are given the problem of solving the Laplace equation in a system
with a given symmetry. In general, functions may be expanded in terms of a com-
plete set of other functions, suitably chosen so that the chosen “basis set” reflects
the defining properties of the function. Let us assume that we know that f (x) has
a period length L. Then, the (discrete) Fourier decomposition of f (x) is given by
∞
2π
f (x) = ∑ F (n) exp (i nx) , (3.158)
n=−∞ L
where
1 2π 2π
F (n) = ∫ dx f (x) exp (−i nx) . (3.159)
2π 0 L
The basis functions in this case are the exponentials
2π
exp (i nx) (3.160)
L
for integer n, all of which are periodic functions with period length L. Here, we will
encounter different sets of functions in which the solutions of the Laplace equation
are expanded.
We will also encounter the ∇ ⃗ 2 operator in different coordinate systems. Let us
suppose that we transform the gradient operator ∇ ⃗ from the Cartesian coordinate
system with coordinates x, y, and z, to a different coordinate system with coordinate
ξ, θ, and ϕ (here, ξ, θ, and ϕ can be any coordinates in any system, say, spherical,
in which case we would replace ξ → r, while in cylindrical coordinates, we would
replace ξ → ρ, θ → ϕ, and ϕ → z). Then, we can rewrite the differentials as follows,
∂ ∂ξ ∂ ∂θ ∂ ∂ϕ ∂
= + + , (3.161)
∂x ∂x ∂ξ ∂x ∂θ ∂x ∂ϕ
and likewise for ∂/∂y and ∂/∂z. Also, the unit vector in the ξ direction becomes
r −1 ∂⃗
∂⃗ r r −1 ∂x
∂⃗ ∂y ∂z
êξ = (∣ ∣) = (∣ ∣) ( êx + êy + êz ) . (3.162)
∂ξ ∂ξ ∂ξ ∂ξ ∂ξ ∂ξ
Inverting these relations, we obtain êx , êy , and êz as functions of êξ , êθ and êϕ . So,
we can rewrite the entire gradient operator ∇ ⃗ in terms of the “new” unit vectors
and “new” differential operators as

∂ ∂ ∂ ∂ ∂ ∂
⃗ = êx
∇ + êy + êz = êξ f1 (ξ, θ, ϕ; , , )
∂x ∂y ∂z ∂ξ ∂θ ∂ϕ
∂ ∂ ∂ ∂ ∂ ∂
+ êθ f2 (ξ, θ, ϕ; , , ) + êϕ f3 (ξ, θ, ϕ; , , ), (3.163)
∂ξ ∂θ ∂ϕ ∂ξ ∂θ ∂ϕ
where f1 , f2 and f3 are functions that may contain derivative operators. This
is a general algorithm for reformulating the gradient operator for a general non-
Cartesian coordinate system.
The calculation of ∇ ⃗ 2 then proceeds in a straightforward manner, but we have
to pay attention: the differentiations may incur additional terms because of the
intertwined base vectors êξ , êθ and êϕ . For example, the partial derivative ∂êξ /∂θ
does not necessarily vanish, nor does ∂êθ /∂ξ. Additional insight into the problem
can be gained through the definition of Christoffel symbols, see Ref. [16]. The
entire program has been carried through a long time ago. While respective formu-
las are contained in reference pages of other textbooks on related subjects (e.g.,
Refs. [17–19]), it is useful to summarize them, for the gradient operator (in Carte-
sian, cylindrical and spherical coordinates)
∂f ∂f ∂f
⃗ = êx
∇f + êy + êz
∂x ∂y ∂z
∂f 1 ∂f ∂f
= êρ + êϕ + êz
∂ρ ρ ∂ϕ ∂z
∂f 1 ∂f 1 ∂f
= êr + êθ + êϕ , (3.164a)
∂r r ∂θ r sin θ ∂ϕ
for the Laplacian operator,
∂2f ∂2f ∂2f
⃗2 =
∇ + +
∂x2 ∂y 2 ∂z 2
∂ 2 f 1 ∂f 1 ∂ 2 f ∂ 2f
= + + +
∂ρ2 ρ ∂ρ ρ2 ∂ϕ2 ∂z 2
∂ 2 f 2 ∂f 1 ∂ ∂f 1 ∂2f
= + + sin θ + , (3.164b)
∂r2 r ∂r r2 sin θ ∂θ ∂θ r2 sin2 θ ∂ϕ2
for the divergence,
∂Ax ∂Ay ∂Az
⃗ ⋅ A⃗ =
∇ + +
∂x ∂y ∂z
∂Aρ Aρ 1 ∂Aϕ ∂Az
= + + +
∂ρ ρ ρ ∂ϕ ∂z
∂Ar 2Ar 1 ∂Aθ cot θ Aθ 1 ∂Aϕ
= + + + + , (3.164c)
∂r r r ∂θ r r sin θ ∂ϕ
and the curl,

∂A ∂Ay ∂Ax ∂Az ∂Ay ∂Az
⃗ × A⃗ = êx ( z −
∇ ) + êy ( − ) + êz ( − )
∂y ∂z ∂z ∂x ∂x ∂x
1 ∂Az ∂Aϕ ∂Aρ ∂Az ∂Aϕ ∂Aρ
= êρ ( − ) + êϕ ( − ) + êz ( + Aϕ − )
ρ ∂ϕ ∂z ∂z ∂ρ ∂ρ ∂ϕ
1 ∂Aϕ cot θ Aϕ 1 ∂Aθ 1 ∂Ar ∂Aϕ Aϕ
= êr ( + − ) + êθ ( − − )
r ∂θ r r sin θ ∂ϕ r sin θ ∂ϕ ∂r r
∂Aθ Aθ 1 ∂Ar
+ êϕ ( + − ). (3.164d)
∂r r r ∂θ
In our investigations on the Laplace equation, let us start with Cartesian coor-
dinates, where
∂2 ∂2 ∂2
⃗ 2 Φ(x, y, z) = (
∇ + + ) Φ(x, y, z) = 0 . (3.165)
∂x2 ∂y 2 ∂z 2
The separation constant is the wave vector k⃗ = kx êx + ky êy + kz êz . The general
solution reads
⃗ ⃗
Φ(x, y, z) = ∑ [c(k)ek⋅⃗r + d(k)ek⋅⃗r {(ax + b)δkx ,0 + (cy + d)δky ,0 + (ez + f )δkz ,0 )}]
⃗ k=0
k⋅ ⃗
+ (a′ x + b′ ) (c′ y + d′ ) (e′ z + f ′ ) . (3.166)
The k⃗ vector may have complex rather than real components, and the summation
is over all permissible vectors k⃗ (as defined by the geometry). Note that if the
components of k⃗ are complex, then the condition k⃗ ⋅ k⃗ = 0 does not necessarily imply
⃗
that k⃗ = ⃗
0. One possibility for k⃗ = kx êx + ky êy + kz êz is to be an exponential ek⃗⋅⃗r
with kx,y = ±i ∣kx,y ∣ being purely imaginary and kz being real. This leads to an
exponential factor
⃗
ek⋅⃗r = ei(∣kx ∣ x+∣ky ∣ y) e±kz z . (3.167)
We thus have a sinusoidal dependence in x and y and an exponential in z. The

special terms in Eq. (3.166) are generated as polynomial solutions to the equation
f ′′ (x) = 0, in the directions where specific components of the k⃗ vector vanish.
The condition k⃗ ⋅ k⃗ = 0 implies that if kx and ky are given, then kz can be
determined (up to a sign). So, the summation over the possible values of k⃗ (a
discrete set of values, in a typical case) will in general be a double summation, not
a triple summation. Another possibility is kz = ±i∣kz ∣, with kx and ky being real.
Then,
exp(k⃗ ⋅ r⃗) = exp (kz x + ky y) exp (±i∣kz ∣z) , (3.168)
with a sinusoidal dependence in x and y, and an exponential in z.

Let us now switch to spherical coordinates. From Sec. 2.3.1, we recall that the
Laplace equation reads
1 ∂2 1 ∂ ∂ 1 ∂2
⃗ 2 Φ(r, θ, ϕ) = (
∇ r+ 2 sin θ + 2 ) Φ(r, θ, ϕ)
r ∂r 2 r sin θ ∂θ ∂θ r sin θ ∂ϕ2
1 ∂2 ⃗2
L 1 ∂2 ( + 1)
=(2
r − 2 ) Φ(r, θ, ϕ) = ( r− ) Φ(r, θ, ϕ) = 0 ,
r ∂r r r ∂r2 r2
(3.169)

with ∣m∣ ≤ , and ∈ 0 . The operator L ⃗ = −i r⃗ × ∇
⃗ has already been discussed in
Sec. 2.3.1. The general solution
∞
Φ(r, θ, ϕ) = ∑ ∑ (a m r + b m r−−1 ) Y,m (θ, ϕ) (3.170)
=0 m=−
involves the spherical harmonics,

2
⃗ 2 Y m (θ, ϕ) = ( + 1) Y,m (θ, ϕ) ,
L ⃗ 2 = − 1 ∂ sin θ ∂ − 1 ∂ , (3.171)
L
sin θ ∂θ ∂θ sin θ ∂ϕ2
whose explicit form we recall from Eq. (2.53),
1/2
2 + 1 ( − m)! ∣m∣
Y m (θ, ϕ) = (−1) ( m
) P (cos θ) eimϕ . (3.172)
4π ( + m)!
∣m∣
The associated Legendre polynomials P (cos θ) enter this expression. Using the
fact that
1 ∂2 1 ∂2
( r) r = ( + 1) r , r) r−−1 = − (− − 1) r−−1 = ( + 1) r−−1 ,
(
r ∂r2 r ∂r2
(3.173)
it becomes clear that all basis functions in the expansion (3.170) fulfill the Laplace
equation separately. In cylindrical coordinates, the Laplace equation reads
1 ∂ ∂ 1 ∂2 ∂2
⃗ 2 Φ(ρ, ϕ, z) = (
∇ ρ + 2 + ) Φ(ρ, θ, z) = 0 . (3.174)
ρ ∂ρ ∂ρ ρ ∂ϕ2 ∂z 2
With α, m = 0, ±1, ±2, . . . , the solution reads
Φ(ρ, ϕ, z) = ∑ (aαm Jm (αρ) + bαm Ym (αρ)) eimϕ e±αz . (3.175)

α,m
The α parameters may be complex. The basis functions of the expansion (3.175)
fulfill the Laplace equation provided
∂2 1 ∂ m2
( + − + α2 ) Jm (α ρ) = 0 . (3.176)
∂ρ2 ρ ∂ρ ρ2
Bessel’s differential equation [Eq. (2.144)] and other important properties of the
Bessel functions have been summarized in Sec. 2.4.2. The equivalence of Eqs. (3.176)
and (2.144) can be shown by explicit differentiation with respect to ρ. Bessel func-
tions have the orthogonality properties (2.171) and (2.178), which we recall for
convenience,
1 1 2
∫ dx x Jn (λ x) Jn (μ x) = δλμ [Jn+1 (λ)] , (3.177a)
0 2
∞ 1
∫ dρ ρ Jn (k ρ) Jn (k ′ ρ) = δ (k − k ′ ) , (3.177b)
0 k
where Jn (λ) = Jn (μ) = 0. This property ensures the orthogonality of the Bessel
functions with respect to the volume integral in cylindrical coordinates. Indeed,
the general idea is to expand the boundary condition to the solution of a Laplace
equation in cylindrical coordinates, into the basis functions of the solution later
used for the entire space, and then, to use the orthogonality properties of the Bessel
functions in order to project out specific components.
3.3.2 Laplace Equation in a Rectangular Parallelepiped

We have discussed solutions of the Laplace equation in Cartesian, spherical and
cylindrical coordinates. It is illustrative to consider a nontrivial example, and we
shall choose the case of a rectangular box, which corresponds to Cartesian coordi-
nates. The potential is specified on each of the six rectangular surfaces of the box.
A simplified version of this problem is given by the solutions to Laplace’s equation
in a rectangle, which is tantamount to assuming uniformity of the potential in the
z direction. The reference volume V is chosen as follows with the x and y direc-
tions being restricted to finite intervals, and the z direction extending from z = 0
to z = ∞,
0 ≤ x ≤ a, 0 ≤ y ≤ b, z ≥ 0. (3.178)
We assume that the charge density inside the volume V vanishes, so that the electro-
⃗ 2 Φ(⃗
static potential fulfills the Laplace equation ∇ r ) = 0. Furthermore, in the sense
of Dirichlet boundary conditions, the potential is held constant near the boundaries
in the x and y directions,
Φ(0, y, z) = Φ(a, y, z) = Φ(x, 0, z) = Φ(x, b, z) = Φ0 . (3.179)
The potential tends to a constant value at infinity,
Φ(x, y, z → ∞) = Φ0 , (3.180)
and in the xy plane (z = 0), we have a given, possibly non-constant, distribution for
the potential
Φ(x, y, 0) = Φ0 f (x, y) , (3.181)
where f (x, y) is a given function. This defines our problem. In the first step of the
calculation, one identifies the expansion (3.166) as the appropriate expansion for
Cartesian coordinates. So,
⃗ ⃗
Φ(x, y, z) = ∑ [c(k)ek⋅⃗r + d(k)ek⋅⃗r {(ax + b) δk1 ,0 + (cy + d) δk2 ,0 + (ez + f ) δk3 ,0 )}]
⃗2 =0
k
+ (ax + b) (cy + d) (ez + f ) . (3.182)
First, we note that any terms linear in x, y, or z in our approach given by Eq. (3.182)
do not satisfy the boundary conditions (3.179). If we eliminate the linear terms,
then the general solution can be written as a sum over the exponential terms alone,
and a physically irrelevant constant C0 ,
⃗
⃗ ek⋅⃗r + C0 ,
Φ(x, y, z) = ∑ c(k) where k⃗2 = kx2 + ky2 + kz2 = 0 . (3.183)
⃗2 =0
k
The overall constant term C0 is not eliminated (yet).

In the second step of the calculation, we observe that the deviation of the solution
from the constant C0 must vanish in the limit z → ∞. So, we take a solution which
is periodic in x and y and exponentially damped for large, positive z. This implies,
in particular, that
√
k⃗ = iκx êx + iκy êy + k3 êz , k3 = κ2x + κ2y . (3.184)
Then, the solution becomes

√
⃗ ei(κx x+κy y) e−
Φ(x, y, z) = ∑ c(k)
κ2x +κ2y z
+ C0 . (3.185)
⃗
k
In general, if (k1 , k2 ) is an admissible combination of wave vectors, then (−k1 , −k2 )

will also be admissible. We here restrict our approach, at first somewhat arbitrarily,
to the combinations
√
⃗ (eiκx x ± e−iκx x ) (eiκy y ± e−iκy y ) e−
Φ(x, y, z) = ∑ c± (k)
κ2x +κ2y z
+ C0 . (3.186)
⃗
k,±
In a third step of the calculation, we let (k1 , k2 ) = (u, v) where u and v are real
positive numbers. Then, our ansatz for the potential becomes
Φ(x, y, z) = ∑ c± (u, v) [eiux ± e−iux ] [eivy ± e−ivy ] exp (−[u2 + v 2 ]1/2 z) + C0 .

u,v,±
(3.187)
Finally, according to our assumption (3.179), we have
Φ(0, y, z) = Φ(a, y, z) = Φ(x, 0, z) = Φ(x, b, z) = Φ0 . (3.188)
We now force the periodic terms to vanish on the x and y boundaries and identify
C0 = Φ0 . We are now in the position to specify the admissible wave vectors even
more accurately, because they have to fulfill the condition that all terms except for
Fig. 3.2 Plot of the functions sin(πmx/a) for a = 1 in the interval

x ∈ (0, 1) with parameters/indices m = 1, 2, 3. The basis functions
sin(πmx/a) vanish at the boundaries and are mutually orthogonal
when integrated over 0 ≤ x ≤ a.
C0 = Φ0 vanish on the boundary. Let a u = mπ and b v = nπ where m, n ∈ . This

singles out the following solutions as admissible,
∞ ∞
1 πm πm
Φ(x, y, z) = ∑ ∑ d(m, n) [exp (i x) − exp (−i x)]
m=1 n=1 2i a a
1 πn πn
× [exp (i y) − exp (−i y)] + Φ0 , (3.189)
2i b b
where the d(m, n) represent expansion coefficients which will be determined later.
Then,
√
∞ ∞ ⎛
πm πn m 2 n 2⎞
Φ(x, y, z) = ∑ ∑ d(m, n) sin ( x) sin ( y) exp −πz ( ) + ( ) + Φ0 .
m=1 n=1 a b ⎝ a b ⎠
(3.190)
A plot of the first few basis functions sin(πmx/a) for a = 1 is given in Fig. 3.2.
In the last step, we recall that at z = 0, we have to satisfy the boundary condi-
tion Φ(x, y, 0) = Φ0 f (x, y). To this end, we note that at z = 0, the exponential
suppression factor in the z direction is just unity,
√ b
⎛ m 2 n 2 ⎞bbb
exp −πz ( ) + ( ) bbbb = 1. (3.191)
⎝ a b ⎠bbb
bz=0
The boundary condition Φ(x, y, 0) = Φ0 f (x, y) then reduces to

∞ ∞
πm πn
∑ ∑ d(m, n) sin ( x) sin ( y) = Φ0 {f (x, y) − 1} . (3.192)
m=1 n=1 a b
In order to determine the coefficients d(m, n), we now multiply both sides by
sin(πm′ x/a) sin(πn′ y/b) and integrate over x and y,
a b πm′ πn′
Φ0 ∫ dx ∫ dy [f (x, y) − 1] sin ( x) sin ( y)
0 0 a n
πm′ ⎞ ⎛ πn ′ ⎞
∞ ∞ a b
⎛ πm πn
= ∑ ∑ d(m, n) ∫ dx sin ( x) sin ( x) ⎜∫ dy sin ( y) sin ( y)⎟
m=1 n=1 ⎝ a a ⎠⎝ b b ⎠
0 0
∞ ∞ π π
a b
= ∑ ∑ d(m, n) ( ∫ dξ sin(mξ) sin(m′ ξ)) ( ∫ dχ sin(nχ) sin(n′ χ)) ,
m=1 n=1 π 0 π 0
(3.193)
π π
where ξ ≡ x and χ ≡ y. Using the orthogonality of the sine and cosine functions,
a b
π π π π
∫ sin(mξ) sin(m′ ξ) dξ = δmm′ , ∫ cos(nχ) cos(n′ χ) dχ = δnn′ ,
0 2 0 2
(3.194)
the right-hand side of Eq. (3.193) thus becomes
∞ ∞
a π b π ab
∑ ∑ d(m, n) ( δmm′ ) ( δnn′ ) = d(m′ , n′ ) . (3.195)
m=1 n=1 π 2 π 2 4
Renaming m′ → m, n′ → n, we can thus write a general expression for all expansion

coefficients d(m, n),
4 a b πm πn
d(m, n) = Φ0 ∫ dx ∫ dy [f (x, y) − 1] sin ( x) sin ( y) . (3.196)
ab 0 0 a b
We recall that the general solution for the potential Φ(x, y, z), in terms of the
d(m, n), is then given as
√
∞ ∞ ⎛
πm πn m 2 n 2⎞
Φ(x, y, z) = ∑ ∑ d(m, n) sin ( x) sin ( y) exp −πz ( ) + ( ) + Φ0 .
m=1 n=1 a b ⎝ a b ⎠
(3.197)
Our boundary value problem involves the boundary conditions Φ(0, y, z) =
Φ(a, y, z) = Φ(x, 0, z) = Φ(x, b, z) = Φ0 , with Φ(x, y, z → ∞) = 0 and Φ(x, y, 0) =
Φ0 f (x, y). We consider it for the choice
1 a 2 1 x x 2
f (x, y) = 1 + [ (x − ) − ] = 1 − + ( ) , 0 < x < a, 0 < y < b . (3.198)
a2 2 4 a a
This boundary condition depends only on x [see Fig. 3.3(a)]. The expansion coeffi-
cients d(m, n) in this case are given by
8 32
d(m, n) = − [1 − (−1)m ] [1 − (−1)n ] = − , (3.199)
m3 n π 4 (2 m + 1)3 (2 n + 1) π 4
(a) (b)
Fig. 3.3 Panel (a) illustrates the boundary condition (3.198) for a = b = 1. The function
G(x, y) is defined in Eq. (3.201). The right panel (b) displays the function Fz (x, y) =
Φ(x, y, z) − Φ0 for z = 5.5, for a = b = 1. Note the smaller scale of the ordinate axis.
Indeed, the departure from the boundary value assumed near x = 0, a and y = 0, b becomes
numerically smaller as we go up the z axis; observe the scale of the ordinate axis in
panel (b).
where m = 2m + 1 and n = 2n + 1. The full solution is then given as

∞ ∞
32 Φ0
Φ(x, y, z) = − ∑ ∑
m=0 n=0 (2 m + 1) (2 n + 1) π
3 4
π(2 m + 1) π(2 n + 1)
× sin ( x) sin ( y)
a b
√
⎛ 2m+1 2 2 n + 1 2⎞
× exp −πz ( ) +( ) + Φ0 . (3.200)
⎝ a b ⎠
For z = 0, we finally have to convince ourselves that the boundary condition is
fulfilled. To this end we investigate the function
G(x, y) = Φ(x, y, 0) − Φ0
π(2 m + 1) π(2 n + 1)
∞ ∞
sin ( x) sin ( y)
a b
= − ∑ ∑ 32 Φ0
m=0 n=0 (2 m + 1)3 (2 n + 1) π 4
x x 2
= Φ0 (− + ( ) ) 0 ≤ x ≤ a, 0 ≤ y ≤ b, (3.201)
a a
but still Φ(0, y, 0) = Φ(a, y, 0) = Φ(x, 0, 0) = Φ(x, b, 0) = Φ0 . We define Gn (x, y)
to be the function obtained from the defining formula (3.201) by truncating the
summations over m and n at n,
G(x, y) = lim Gn (x, y) , (3.202a)
n→∞
π(2 m + 1) π(2 n + 1)
n n
sin ( x) sin ( y)
a b
Gn (x, y) = ∑ ∑ 32 Φ0 . (3.202b)
m=0 n=0 (2 m + 1)3 (2 n + 1) π 4
(a) (b)
Fig. 3.4 Panel (a) illustrates how the partial sums over m and n approximate the solu-
tion (3.200), by plotting the function Gn (x, y) with n = 3, given by Eq. (3.202). The sum
is terminated at n = m = n = 3. Panel (b) displays a better approximation with the sum
being terminated at n = m = n = 20.
As evident from the left panel in Fig. 3.4, with both summations over m and n
truncated at n = 3, the boundary condition on the edges of the rectangle are fulfilled
by our solution (3.200), at x = 0, x = a, y = 0, y = b. However, as soon as we depart
from the lines with y = 0 and y = b, there is a discontinuity because the term
2
− xa + ( xa ) does not vanish for 0 ≤ x ≤ a and 0 ≤ x ≤ b. The approximation of this
discontinuity implies oscillations in the y direction for the truncated approximation,
as visible in Fig. 3.4(a) and Fig. 3.4(b).
It is interesting to consider what happens as we depart from the z = 0 plane and
go up to z axis, to arrive, say, at z = 5.5. This is illustrated in Fig. 3.3(b). The value
of the expression F (x, y, z) = Φ(x, y, z) − Φ0 becomes much smaller (of order 10−12 )
at z = 5.5 than at z = 0, where it amounts to G(x, y) = F (x, y, 0) and is of order
unity. Because of the smallness of the function G(x, y), the discontinuity near the
edges of the rectangle x = 0, x = a, y = 0, and y = b gradually disappears as we go
up the z axis. This is compatible with the boundary condition Φ(x, y, z → ∞) = Φ0 .
3.3.3 Laplace Equation in a Two-Dimensional Rectangle
A simplified version of the zero-charge-density problem is given by the solutions

to Laplace’s equation in a rectangle, which is tantamount to assuming uniformity
⃗ 2 Φ(⃗
of the potential in the z direction. In this case, the Laplace equation ∇ r) = 0
reduces to an equation with differential operators that only act in the x and y
directions. The technique used for the two-dimensional problem (for a rectangle)
thus is a specialization of the previously discussed problem for the rectangular
parallelepiped. Let us consider boundary conditions given on a two-dimensional
rectangle, with a potential Φ = Φ(x, y) that fulfills
∂2 ∂2
( 2
+ 2 ) Φ (x, y) = 0 , 0 < x < a, 0 < y < b. (3.203)
∂x ∂y
We take our boundary conditions to be
Φ (0, y) = Φ (a, y) = Φ (x, 0) = 0 , Φ (x, b) = f (x) , (3.204)
where f (x) is a given function. According to the above sections, we should search
for a series solution involving exponentials,
⃗ e±i∣k1 ∣x e±k2 y + C0 ,
Φ (x, y) = ∑ c(k) (3.205)
⃗
k
where each term in the series satisfies the Laplace equation in two dimensions and
so ∣k1 ∣ = ∣k2 ∣. In order to select the proper solution to these equations we refer to the
boundary conditions. Since Φ (x, b) is specified in the boundary we use a (complete)
set of orthogonal functions for the x dependence in the interval x ∈ (0, a). Since
Φ (0, y) = Φ (a, y) = 0, we use sine functions which are equal to zero at x = 0 and
x = a. The y dependence has to be adjusted so that the functions vanish at y = 0,
which singles out the hyperbolic sine (as opposed to the cosine) functions. It follows
that
πn πn
∞ exp ( y) − exp (− y)
nπ a a
Φ (x, y) = ∑ an sin ( x)
a πn πn
n=1 exp ( b) − exp (− b)
a a
nπy
∞ sinh ( )
nπx a
= ∑ an sin ( ) . (3.206)
a nπb
n=1 sinh ( )
a
The denominator term sinh (nπb/a) is a normalization. This ansatz automatically
fulfills
Φ (0, y) = Φ (a, y) = Φ (x, 0) = 0 . (3.207)
In order to also fulfill
Φ (x, b) = f (x) , (3.208)
we have to demand that

2 a nπx
an = ∫ f (x) sin ( ) dx . (3.209)
a 0 a
If we now specialize further to f (x) = Φ0 , then
2 a nπx 2Φ0 n
αn = ∫ Φ0 sin ( ) dx = {1 − (−1) } . (3.210)
a 0 a nπ
Finally, the potential is found to be
4Φ0 ∞ sin [(2n + 1) πx/a] sinh [(2n + 1) πy/a]
Φ (x, y) = ∑ , (3.211)
π n=0 2n + 1 sinh [(2n + 1) πb/a]
(a) (b)
(c) (d)
Fig. 3.5 Plot of the function Φ(x, y) as defined in Eq. (3.211) for Φ0 = 1 and a = b = 1.
We define Φn (x, y) to be equal to the series expansion given in Eq. (3.211), where the
sum is truncated at n. Panels (a), (b), (c) , and (d) show the truncated functions at
n = 5, n = 10, n = 20 and n = 80.
where the sum is over the odd integers 2n + 1, with n = 0, 1, 2, . . . . For 0 ≤ y < a,
this series and its first derivatives converge absolutely. A plot of the sum of the
terms up to n = 20 is shown in Fig. 3.5. Quite naturally, there are oscillations in
the (slow) convergence pattern near the discontinuities of Φ (x, y) which occur near
the points (0, b) and (a, b).
3.3.4 Boundary Conditions on the Finite Part of a Long Strip

We also consider a two-dimensional boundary problem, but this time, one of the
regions covered by the potential has an infinite extent, in contrast to the example
discussed in Sec. 3.3.3. The problem is given as
∂2 ∂2
( + ) Φ (x, y) = 0 , 0 < x < a → ∞, 0 < y < b. (3.212)
∂x2 ∂y 2
We define the boundary conditions to be
Φ (x, 0) = Φ (x, b) = 0 , Φ (0, y) = g (y) . (3.213)

For these boundary conditions, the appropriate choice is

∞
nπy nπx
Φ (x, y) = ∑ αn sin ( ) exp (− ). (3.214)
n=1 b b
The boundary condition Φ (x, 0) = Φ (x, b) = 0 is automatically implemented via the
sine function, and the boundary condition Φ (0, y) = g (y) implies that
∞
nπy
g (y) = ∑ αn sin ( ). (3.215)
n=1 b
In view of the orthogonality of the sine functions, we have
2 b nπy
αn = ∫ dy g (y) sin ( ). (3.216)
b 0 b
The coefficients bn depend on the particular choice of the boundary condition. If
we let
y
g (y) = Φ0 , (3.217)
b
then the expansion coefficients are given by
2Φ0 b nπy
αn = ∫ dy y sin ( ) (3.218)
b2 0 b
2Φ0 b b d nπy
= 2 ∫ dy y (− {cos ( )}) (3.219)
b nπ 0 dy b
b b
2 Φ0 nπy nπy
= [−y {cos ( ) }∣ − ∫ dy (− cos ( ))] (3.220)
bn π b 0 0 b
2Φ0 n Φ0
= − b cos(nπ) = −2 (−1) . (3.221)
anπ nπ
The potential thus is given as
2Φ0 ∞ (−1)n nπy nπx
Φ (x, y) = − ∑ sin ( ) exp (− ). (3.222)
π n=1 n b b
The sum over n can be performed by writing the entire expression in terms of
exponentials,
2Φ0 ∞ (−1)n n −n 1 πx n
Φ (x, y) = − ∑ [{eiπy/b } − {eiπy/b } ] {exp (− )}
π n=1 n 2i b
2Φ0 ∞ (−1)n v n
= − ∑ [(uv)n − ( ) ] ,
2iπ n=1 n u
u = eiπy/b , v = e−πx/b . (3.223)
We now use the formula

∞
(−1)n n
∑ x = − ln(1 + x) , (3.224)
n=1 n
Fig. 3.6 Plot of the function Φ(x, y) given in Eq. (3.225), for Φ0 = 1 and b = 1.
to obtain
Φ0 v Φ0 u+v
Φ (x, y) = − {ln (1 + ) − ln (1 + u v)} = − ln ( )
iπ u iπ u (1 + u v)
Φ0 eiπy/b + e−πx/b Φ0 1 + e−πx/b−iπy/b
=− ln ( iπy/b )=− ln ( )
iπ e (1 + e iπy/b−πx/b ) iπ 1 + eiπy/b−πx/b
Φ0 e−iπy/b + eπx/b Φ0 eiπy/b + eπx/b
=− ln ( iπy/b πx/b ) = ln ( −iπy/b πx/b ) . (3.225)
iπ e +e iπ e +e
Somewhat counter-intuitively, this result is purely real, not complex. This can be
seen as follows. If we consider the complex conjugate of the expression, the imagi-
nary unit in the prefactor changes sign, but the numerator and the denominator in
the argument of the logarithm are also exchanged. Inverting the argument of the
logarithm, one restores the original expression. A plot of the solution is found in
Fig. 3.6.
3.3.5 Cauchy’s Residue Theorem: A Small Digression

Helpful mathematical techniques are sometimes compared to the “toolbox” of a
theoretical physicist; within this analogy, one might identify the Cauchy residue
theorem with a “leatherman”. It has many physical applications; e.g., it becomes
important to the understanding of time ordering in electrodynamics. Another very
important application concerns the evaluation of integrals over an infinite domain
of integration, closing the integration contour along an appropriate curve in the
complex plane. This latter technique is important in the evaluation of certain inte-
grals occurring in Fourier transformations, which naturally arise in boundary-value
problems in electrostatics. Preparations for a graduate course in electrodynamics
should in principle include deep familiarity with the Cauchy theorem; yet experi-
ence shows that this preparation may be lacking. A detour on this subject therefore
is in order.
Let us start innocently, recalling that the Taylor expansions of a function f =
f (x) about the origin reads as
1 ′′ ∞
f (n)(0) n ∞
f (x) = f (0) + f ′ (0) x + f (0) x2 + ⋅ ⋅ ⋅ = ∑ x = ∑ an x2 . (3.226)
2! n=0 n! n=0
Here, the an with n = 0, 1, 2, . . . are constant coefficients, the coefficients of the

Taylor expansion. However, functions like
1
g(z) = (3.227)
z 4 sin(z)
obviously cannot be expanded in a Taylor series about z = 0, because the first
(leading) term is not a constant; the function diverges for z → 0. At best, we can
use the well-known expansion, sin(z) = z − z 3 /3! + z 5 /5! + . . . and write
1 1 1
g(z) = = 3 5
= 2
z 4 sin(z) z z z z4
z 4 (z − + + . . . ) z 5 (1 − + + ...)
3! 5! 3! 5!
1 1 7 31 z
=
+ + + + O(z 3 ) . (3.228)
z 5 6 z 3 360 z 15120
This expansion obviously does not start from the term of order zero, but from a
term proportional to z −5 . It is called a Laurent expansion. The general formula for
the Laurent expansion reads
∞
g(z) = ∑ an z n . (3.229)
n=−∞
In contrast to the Taylor expansion (3.226), the Laurent expansion starts at n = −∞

instead of n = 0. We can formulate a Laurent expansion about a point z0 , which
reads
∞
g(z) = ∑ bn (z − z0 )n . (3.230)
n=−∞
If all coefficients bn with n < 0 vanish, we say that g(z) is analytic about the point
z = z0 . If all an ’s with n < −M vanish, then we say that g has an M th order pole
at the origin.
Of preeminent importance is the integral
∞
I = ∮ g(z) dz = ∫ g(z) dz = ∫ ( ∑ an z n ) dz , (3.231)
C C n=−∞
where the contour C is an anticlockwise circle (i.e., in the mathematically positive

sense) of radius R around the origin. We shall see that only a very limited number
of an coefficients (in fact, just one of them) contributes to the integral. Let us
therefore consider the integral

1
J = ∮ dz , z = R exp(iθ) , θ ∈ (0, 2π) , (3.232)
C zm
where z = z(θ) thus describes an anticlockwise circle about the origin. Furthermore,
m is assumed to be integer. Now, for m ≠ 1, we have
2π 2π
−m
J=∫ dθ (iR) exp(iθ) [R exp(iθ)] = iR1−m ∫ dθ exp[−i (m − 1) θ]
0 0
θ=2π θ=2π
exp[−i (m − 1) θ] exp[−i (m − 1) θ]
= iR 1−m
[ ] = −R1−m [ ] = 0 . (3.233)
−(m − 1)i θ=0 (m − 1) θ=0
However, for m = 1, the correct result is obtained by first replacing m → 1 + and

doing a Taylor expansion,
1 θ=2π
1 − iθ θ=2π
lim J = − R− [lim exp(−iθ)] = −[ ]
m→1 →0 θ=0 θ=0
θ=2π
1 θ=2π
= − [ − i θ] = [i θ]θ=0 = 2π i . (3.234)
θ=0
Of course, this result could have been obtained by a direct calculation, without
letting m → 1, but the above derivation illustrates that the special case m = 1
actually follows from the more general case by a limiting process. For a function
g(z) that has a Laurent expansion of the form g(z) = ∑∞ n
n=−∞ an z about the point
z0 = 0, we therefore have
∞
I = ∮ dz f (z) = ∮ dz ( ∑ am z m ) = 2π i a−1 ≡ 2π Res f (z) , (3.235)
C C m=−∞ z=0
where the last term constitutes a definition of the residue at the point z = z0 = 0.
Only the coefficient a−1 contributes. We write the residue as
Res f (z) ≡ a−1 , ∮ dz f (z) = 2π i Res f (z) . (3.236)

z=0 C z=0
This result can be generalized easily, as follows. Let us assume that we have an
expansion about an arbitrary point z0 ,
∞
f (z) = ∑ bm (z − z0 )m . (3.237)
m=−∞
In view of the above considerations, a contour integral that encircles z0 in the

mathematically positive sense can be written as
∮ dz f (z) = 2π i b−1 = 2π i Res f (z) . (3.238)

C0 z=z0
Only the coefficient b−1 contributes. Closed contour integrals in regions where the
integrand is analytic, simply vanish. This enables us to generalize the result as
follows. For a contour C that encircles n residues located at the points z = zi with
i ∈ {1, . . . , n}, the result is

n
∮ dz f (z) = 2π ∑ Res f (z) . (3.239)
C z=zi
i=1
All the zi lie in the interior of the contour C. This is Cauchy’s residue theorem.
The calculation of residues and the application of this theorem should be illus-
trated on a few example cases. For functions that have a single pole, like
1 1
g(z) = ≈ , Res g(z) = 1 , (3.240)
sin z z − 3 z + . . .
1 3 z=0
the calculation of the residue is obvious. For functions that have poles of higher
order, it becomes more complicated. In Eq. (3.228), we have already derived the
result
1 7
Res ( 4 )= . (3.241)
z=0 z sin(z) 360
Let us also consider
cos(z) 1 z cos(z)
g(z) = 4
= ( 5) ( ). (3.242)
z sin(z) z sin(z)
We have split the function into a prefactor 1/z 5 and term which is unity for z → 0.
A simple Taylor expansion of numerator and denominator shows that
z cos(z) z2 z4
=1− − + ... (3.243)
sin(z) 3 45
and so, for z → 0,
1 z2 z4 1
g(z) ≈ (1 − − + . . .) , Res g(z) = − , (3.244)
z5 3 45 z=0 45
where the latter result is simply obtained by an expansion of the former, and reading
off the term that multiplies z −1 . The residue always is the factor multiplying 1/(z −
z0 ) in the Laurent expansion of F (z), independent of how many other terms of the
form (z − z0 )n , with positive or negative integer n, occur in the Laurent expansion
of g(z). One just has to expand g(z) to the required order and read off the term of
order (z − z0 )−1 .
Let us now investigate a generalization of this result, namely, the case where
f (z) has an M th order pole. For the example in Eq. (3.244), we have M = 5. We
are interested in finding out about the term of order (z − z0 )−1 in f (z). To this end,
one can follow the ad hoc procedure outlined above: We just expand numerator and
denominator to the required order and then read off the term of order (z − z0 )−1 .
However, it is also possible to follow a more systematic approach. If one multiplies
the terms {a−M (z−z0 )−M , a−M+1 (z−z0 )−M+1 , . . . , a−1 (z−z0 )−1 , . . . } by a factor (z−
z0 )M , then they transform into the set {a−M , a−M+1 (z−z0 ), . . . , a−1 (z−z0 )M−1 , . . . }.
Expressed differently,
a−M a−M+1 a−1
f (z) = + + ⋅⋅⋅ + + ⋅ ⋅ ⋅ + an (z − z0 )n + . . . , (3.245)
(z − z0 )M (z − z0 )M−1 (z − z0 )
(z − z0 )M f (z) = a−M + (z − z0 ) a−M+1 + ⋅ ⋅ ⋅ + (z − z0 )M−1 a−1 + . . . . (3.246)

In consequence, if we differentiate the expression (z−z0 ) f (z) a total of M −1 times
M
with respect to z, and set z = z0 after the differentiation, then the only contributing
term is the one proportional to a−1 . Terms of lower order are explicitly annihilated
by the differential operator, and terms of higher order give rise to positive powers
of (z − z0 ); these are annihilated if we set z = z0 after the differentiation. This leads
to the general formula, valid for an M th order pole,
⎡ M−1 ⎤
⎢d {(z − z0 ) f (z)} ⎥
M
1 ⎢ ⎥
Res f (z) = ⎢ ⎥ . (3.247)
z=z0 (M − 1)! ⎢⎢ dz M−1 ⎥
⎥
⎣ ⎦z=z0
A paradigmatic application of the above consideration is as follows. For an
integral such as
∞ η ∞ η
∫ dω 2 =∫ dω , (3.248)
−∞ ω +η 2 −∞ (ω + i η) (ω − i η)
we can close the integration contour either in the upper or the lower complex plane
because the modulus of the half-circle integral behaves as R−2+1 → 0 for R → ∞,
where ∣z∣ = R is the radius of the half circle along which the contour is being closed.
At the singularity, at ω = iη, we have
η η
≈ (3.249)
(ω + i η) (ω − i η) 2 i η (ω − i η)
and hence
η 1
Res = . (3.250)
ω=iη (ω + i η) (ω − i η) 2i
The result is
∞ η 1
∫ dω = 2π i = π. (3.251)
−∞ ω2 + η2 2i
Reversing the argument, for closing the contour in the lower half of the complex
plane, we have
∞ η 1
∫ dω ω 2 + η 2 = (−2π i) (− 2 i ) = π . (3.252)
−∞
The first minus sign is due to the reversed sense of revolution around the singularity.
The second minus sign is due to an additional sign reversal of the residue at ω = −iη.
3.3.6 Boundary Conditions on the Infinite Part of a Long Strip

We now consider again a long strip, as in Sec. 3.3.4, but this time, the boundary
condition is imposed on the infinite part of the long strip rather than the short
(finite) part. Our boundary conditions are thus given as
Φ (0, y) = Φ (x, 0) = 0 , Φ (x, b) = f (x) , 0 < x < a → ∞, 0 < y < b.
(3.253)
The boundary condition is given on the semi-infinite strip y = b. In this case, we

switch from a summation to an integration. The expansion functions are labeled by
a continuous variable, say k, for which the following relations are useful. In view of
the boundary condition Φ(0, y) = Φ(x, 0) = 0, we choose the ansatz
∞ sinh (k y)
Φ (x, y) = ∫ dk α (k) sin (k x) , (3.254)
0 sinh (k b)
so that
∞
Φ(x, b) = f (x) = ∫ dk α (k) sin (k x) . (3.255)
0
It is nontrivial to show that

2 ∞
α (k) = ∫ dx f (x) sin (k x) . (3.256)
π 0
First, one proves the following relation,
∞ ∞ 1 ik′ x −ik′ x 1 ikx −ikx
∫ dx sin(k ′ x) sin(kx) = ∫ dx [e −e ] [e − e ]
0 0 2i 2i
1 ∞
ik′ x ikx −ik′ x −ikx
=− ∫ [(e e + e e )
4 0
′ ′
− (eik x e−ikx + e−ik x e+ikx )] dx
2π 1 ∞ ′ ′
=− ∫ (eik x eikx − eik x e−ikx ) dx
4 2π −∞
π
= [δ(k ′ − k) − δ(k ′ + k)] . (3.257)
2
We verify that
2 ∞ 2 ∞ ∞
∫ dx f (x) sin(k x) = ∫ dk ′ α(k ′ ) [∫ dx sin (k ′ x) sin (k x)]
π 0 π 0 0
2 ∞ π
= ∫ dk ′ α(k ′ ) [δ(k ′ − k) + δ(k ′ + k)] dk = α(k ′ ) .
π 0 2
(3.258)
For the boundary condition Φ(x, b) = f (x), we now choose
f (x) = Φ0 . (3.259)
As we shall see later, this example can be used in order to illustrate the properties
of complex integrals. The Fourier-sine transform of f (x) = Φ0 is given by
2Φ0 ∞ 2Φ0 ∞
α(k) = ∫ dx sin(k x) = ∫ dx [eikx − e−ikx ] .
π 0 π2i 0
In the sense of Riemann, this improper integral is formally divergent. We thus
introduce the convergent factor exp(− x) in the integrand, where > 0 is real
and small. Under this assumption, we can assign a finite value to the expansion
coefficient α(k),
2 Φ0 b b
α (k) = lim+ lim [∫ e−(−ik)x dx − ∫ e−(+ik)x dx]
2πi →0 b→+∞ 0 0
2Φ0 1 1 2Φ0
= lim ( − )= . (3.260)
2πi →0 − ik + ik πk
Using this result for α(k), the potential can be written as a convergent integral,
2 Φ0 ∞ sin (k x) sinh (k y) Φ0 ∞ sin (k x) sinh (ky)
Φ (x, y) = ∫ dk = ∫ dk ,
π 0 k sinh (k b) π −∞ k sinh (k b)
(3.261)
where we notice the change in the integration limits in the second step. Integrals
from −∞ to ∞ are naturally prone to an evaluation by the method of residues,
as one can close the contour from +∞ to −∞ along a circle either above or below
the real axis, preferably along a large circle whose radius tends to infinity. We
shall evaluate the integral using the theory of residues. If the integrand falls off
sufficiently rapidly for large modulus of the argument, then the contribution of the
large circle closing the contour is zero. Closing the contour, one then picks up the
contributions from the pole terms only.
As it stands, the integral (3.261) has no poles along the real k axis. In partic-
ular, for k → 0, we have the finite limit sin(k x)/k → x. However, pole terms are
encountered as we use the formula
exp(ik x) − exp(−ik x)
sin (k x) = . (3.262)
2i
In performing the k integral, we thus have two alternatives for encircling the poles
on the real k axis. We can either shift the contour infinitesimally below the real k
axis, letting k = k −i and later let → 0+ , or we can shift the contour infinitesimally
above the k axis, letting k = k + i and later calculate the limit → 0+ . Yet, we have
to make a decision and then stick to it. In the end, both alternatives must lead
to equivalent results. The notion is that we later split the integrand into terms for
which we can close the contour along either side of the real axis. We write
Φ0 ∞ exp (ikx) − exp (−ikx) sinh (k y)
Φ (x, y) = ∫ dk
2πi −∞ k − i sinh (k b)
Φ0 exp (ikx) sinh (k y) Φ0 exp (−ikx) sinh (k y)
= ∫ dk − ∫ dk .
2πi C+ k − i sinh (k b) 2πi C− k − i sinh (k b)
(3.263)
The contour will be closed along a half-circle of large radius in the upper complex
half-plane (denoted as C+ in the first integral), which leads to exponential damping
for the term proportional to exp (ikx). The contour C− involves a large half-circle in
the lower complex half plane and suppresses the term exp (−ikx). The contour C−
is traversed in the mathematically negative direction, leading to an overall minus

sign for the residue from the contour closed in the lower half plane.
We now appeal to the discussion in Sec. 3.3.5 and attempt to identify the poles
in the integrand in Eq. (3.263). Notably, we shall use the method outlined in
Eqs. (3.248)–(3.252)] for the evaluation of relevant indefinite integrals. The poles
come from three locations. The first of these is (i) from the 1/(k −i) factor at k = i
with → 0+ . There are (ii) poles at kb = iπn from the sinh(k b) term (contour C+ )
and (iii) poles at kb = −iπn from the sinh(k b) term (contour C− ). The ensemble
of the poles thus is a equidistant string of singularities along the imaginary axis,

extending over all k b = iπm with m ∈ 0 . Cauchy’s residue theorem, as discussed
in Sec. 3.3.5, is thus brought to bear on the subject. The single pole of the first
category at k = i leads to the following residue
Φ0 exp (ikx) sinh (k y) Φ0 exp (i ⋅ ix) sinh (iy)

Res ( ) = lim+ ( )
k=i+ 2πi k − i sinh (k b) →0 2πi 1 sinh (ib)
Φ0 iy Φ0 y
= = . (3.264)
2πi ib 2πib
The contribution to the potential Φ(x, y) is
Φ0 exp (ikx) sinh (k y) Φ0 y

ΦI (x, y) = 2πi Res+ ( )= . (3.265)
k=i 2πi k − i sinh (k b) b
For the poles of the second category, we have kb = nπi and investigate the sinh(k b)
term. Indeed, we have sinh(k b) = 0 at k b = nπi for the integrand in the upper half

plane (n ∈ ). The hyperbolic sine can be expanded near the pole, as follows,
1 1
=
sinh (k b) sinh (inπ + (k b − inπ))
1
≈
d
sinh (inπ) + ( sinh(ξ)∣ ) (k b − inπ)
dξ ξ=inπ
1 1
= = . (3.266)
cosh(iπn) (k b − inπ) b (−1) (k − inπ/b)
n
This corresponds to a simple pole at k = nπi/b, with a coefficient as indicated. We

therefore have for the residue,
Φ0 exp (ikx) sinh (k y) Φ0 exp (i ⋅ inπx/b) nπiy

Res ( )= (−1)n sinh ( )
k=nπi/b 2πi k − i sinh (k b) 2πi b(nπi/b) b
Φ0 exp (−nπx/b) nπy
= (−1)n i sin ( )
2πi nπi b
= (−1)n sin ( ) . (3.267)
2πi nπ b
The contribution to the potential Φ(x, y) therefore is

exp (−nπx/b) nπy
ΦII,n (x, y) = Φ0 (−1)n sin ( ). (3.268)
nπ b
For the poles of the third category, at k b = −nπi with n ∈ , we expand
1 1 1
= ≈ , (3.269)
sinh (k b) sinh (−inπ + [k b + inπ]) b(−1)n [k + inπ/b]
finding a simple pole at k = nπi/b. We therefore have for the residue,
Φ0 exp (−ikx) sinh (k y) Φ0 exp (−nπx/b) nπiy
Res (− )= − (−1)n sinh (− )
k=−nπi/b 2πi k − i sinh (k b) 2πi b(−nπi/b) b
= − (−1)n sin ( ).
2πi nπ b
(3.270)
We encircle these poles in the clockwise direction, which is the mathematically
negative direction. The third category of residues therefore is
exp (−nπx/b) nπy
ΦIII,n (x, y) = Φ0 (−1)n sin ( ). (3.271)
nπ b
The potential is obtained from the sum over all the residues
∞
Φ (x, y) = ΦI (x, y) + ∑ (ΦII,n (x, y) + ΦIII,n (x, y))
n=1
Φ0 y 2Φ0 ∞ (−1)n nπx nπy
= + ∑ exp (− ) sin ( )
b π n=1 n b b
Φ0 y Φ0 ∞ (−1)n nπx nπy nπx nπy
= + ∑ {exp (− +i ) − exp (− −i )}
b πi n=1 n b b b b
Φ0 y Φ0 ∞ (−1)n+1 πx πy n
= + ∑ [exp (− − i )]
b πi n=1 n b b
Φ0 ∞ (−1)n+1 πx πy n
− ∑ [exp (− + i )]
πi n=1 n b b
Φ0 y Φ0 πx πy Φ0 πx πy
= + ln [1 + exp (− − i )] − ln [1 + exp (− + i )]
b πi b b πi b b
Φ0 y Φ0 ⎛ exp ( b ) + exp (−i b ) ⎞
πx πy
= + ln . (3.272)
b πi ⎝ exp ( πx b
) + exp (i πy
b
) ⎠
We recall that according to Eqs. (3.253) and (3.259), the boundary conditions ful-
filled by this solution are
Φ (0, y) = Φ (x, 0) = 0 , Φ (x, b) = Φ0 , 0 < x < ∞. (3.273)
Convergence at the point x = 0, y = b is nonuniform as is evident from Fig. 3.7. The

nonuniform convergence can be verified explicitly. On the one hand, we have for
Fig. 3.7 Plot of the function Φ(x, y) as defined in Eq. (3.272) for
Φ0 = 1 and a = 1.
the approach along the y axis,

⎧
⎪ ) ⎞⎫
⎪
⎪ Φ0 (b − ) Φ0 ⎛ 1 + exp (−iπ b− ⎪
lim+ Φ (0, b − ) = lim+ ⎨ + ln b
⎬
→0 →0 ⎪⎪ b πi ⎝ 1 + exp (iπ b ) ⎪
b−
⎠⎪
⎩ ⎭
⎧
⎪ ⎡ + (iπ b− ⎤⎫
) ⎥⎪
⎪ Φ0 ⎢ b− 1 exp ⎪
= lim+ ⎨Φ0 + ln ⎢exp (−iπ ) b
⎥⎬
→0 ⎪⎪ ⎢ + (iπ b−
) ⎥⎪
⎩
πi ⎣ b 1 exp b ⎦⎪
⎭
Φ0
= lim+ {Φ0 + (−iπ)} = 0 . (3.274)
→0 πi
The approach parallel to the x axis for y = b leads to
⎧
⎪ ) − 1 ⎞⎫
⎪
⎪ Φ0 b Φ0 ⎛ exp ( π ⎪ Φ0 y Φ0
lim+ Φ (, b) = lim+ ⎨ + ln b
⎬ = lim+ { + ln(1)} = Φ0 .
→0 →0 ⎪⎪ b πi ⎝ exp ( π
) − 1 ⎠⎪
⎪ →0 b πi
⎩ b ⎭
(3.275)
The potential is discontinuous at (x, y) = (0, b); the boundary condition itself is
discontinuous at this point. The discontinuity persists even though the result (3.272)
is expressible in closed analytic form. The sum of the result
Φ0 y Φ0 ⎛ exp ( b ) + exp (−i b ) ⎞

πx πy
Φ(x, y) = + ln (3.276)
b πi ⎝ exp ( πx
b
) + exp (i πy
b
) ⎠
and the result given in Eq. (3.225) reads as Φ0 y/b, which also is a solution of the
Laplace equation, and fulfills appropriate boundary conditions.
3.3.7 Cylinders and Zeros of Bessel Functions

In Secs. 2.4.2 and 2.4.3, we have discussed a number of useful properties of Bessel
functions. Equipped with this general knowledge on Bessel functions, we can now
turn our attention to the application of the general solution of the Laplace equation
in cylindrical coordinates, given in Eq. (2.142), to various example problems of

cylindrical symmetry.
We start with a grounded outer rim of a cylinder of radius a, i.e., Φ = 0 for
ρ = a. In this case, the solutions are Bessel functions of the first kind satisfying the
condition Jm (kmn ρ) = 0 for ρ = a. This implies that we must choose xmn = kmn a
to be a zero of the mth Bessel function. Furthermore, we must discard solutions
that involve Bessel Y functions as these are singular at the cylinder axis ρ = 0. This
requirement determines values of k = kmn , that is, kmn with n = 1, 2, . . ., where
xmn = kmn a is the nth root of the Bessel function of order m (see Table 2.1). The
general solution to the Laplace equation ∇ ⃗ 2 Φ(ρ, ϕ, z) = 0 with a vanishing potential
at ρ = a is a specialization of Eq. (2.142) and reads
∞ ∞
Φ(ρ, ϕ, z) = ∑ ∑ amn Jm (kmn ρ) eimϕ e±kmn z , Φ(ρ = a, ϕ, z) = 0 . (3.277)
m=−∞ n=1
The condition to be fulfilled by the allowed wave numbers kmn is
xmn
Jm (kmn a) = 0 , kmn = , (3.278)
a
where xmn is the nth zero of the mth-order Bessel function.
The following boundary conditions
Φ(ρ = a, ϕ, z) = 0 , Φ(ρ, ϕ, z = ±L/2) = Φ0 , (3.279)
describe a zero potential on the outer rim of the cylinder, and constant potential
on the endcaps. The potential is symmetric under z ↔ −z, and the ansatz (3.277)
thus is modified to read
∞ ∞
Φ(ρ, ϕ, z) = ∑ ∑ amn Jm (kmn ρ) eimϕ cosh(kmn z) . (3.280)
m=−∞ n=1
The area element on the cylinder endcaps is dA = dϕ dρ ρ with integration
domains ϕ ∈ (0, 2π) and ρ ∈ (0, a). Hence, the expansion coefficients are given
by the condition
a 2π
′
−im ϕ
∫ dρρ ∫ dϕ Φ(ρ, ϕ, ±L/2) e Jm′ (km′ n′ ρ)
0 0
a 2π
′
= Φ0 ∫ dρρ ∫ dϕ e−im ϕ Jm′ (km′ n′ ρ) . (3.281)
0 0
However, the determination of the expansion coefficients can be simplified some-
what, using the azimuthal symmetry of the potential. The only nonvanishing term
is the one with m = 0, and Eq. (3.280) is thus simplified to
∞
Φ(ρ, ϕ, z) = ∑ an J0 (k0n ρ) cosh(k0n z) . (3.282)
n=1
We now use the formula
a a
∫ dρ ρ J0 (ξ ρ) = J1 (ξ a) , (3.283)
0 ξ
(a) (b)
Fig. 3.8 Plot of the partial sums up to terms with n = 25 and n = 50 of the expres-
sion (3.288) which describes a zero potential on the outer rim of a cylinder, but constant
potential on the cylinder endcaps. The parameters are chosen as a = 1, L = 3 and
Φ0 = 2.3. Slow convergence (oscillatory behavior, so-called Gibbs phenomenon) is clearly
visible near the boundaries of the cylinder. Note that the potential Φ(ρ, z) has azimuthal
symmetry and is independent of ϕ. The plot of Φ(∣ρ∣, z) with −a < ρ < a corresponds to
a plot from one outer rim of the cylinder to the other.
which can be shown as follows. One first differentiates the expression x J1 (x) with
the help of Eq. (2.149b)
d x
[x J1 (x)] = J1 (x) + [J0 (x) − J2 (x)] . (3.284)
dx 2
The recursion relation (2.149a) leads to
2
J2 (x) = −J0 (x) +
J1 (x) . (3.285)
x
Plugging this relation into the right-hand side of (3.284), one has
d
[x J1 (x)] = x J0 (x) , (3.286)
dx
and a trivial scale transformation leads to (3.283). Finally, with the help of (3.283)
and the orthogonality condition (2.172), one verifies that
2Φ0 1
an = . (3.287)
k0n a cosh(k0n L/2) J1 (k0n a)
The solution to the boundary-value problem (3.279) thus reads as
2Φ0 ∞ 1 J0 (k0n ρ) cosh(k0n z)
Φ(ρ, ϕ, z) = ∑ . (3.288)
a n=1 kn0 J1 (k0n a) cosh(k0n L/2)
A plot of the partial sums of this expansion up to n = 5 and n = 50, is given in
Fig. 3.8.
Another question is how to modify the calculation when the potential vanishes on
the cylinder endcaps at z = ±L/2. In this case, we must choose linear combinations
of the functions exp(k z) that vanish at z = ±L/2. The only way to achieve this
goal is to choose k as a complex rather than real number. We thus form linear
combinations proportional to sine and cosine functions. Provided the potential is

regular at the cylinder axis ρ = 0, we again discard the Neumann functions and
write
∞ ∞
(1)
Φ(ρ, ϕ, z) = ∑ ∑ amn Jm (i kn ρ) e
imϕ
cos(kn(1) z)
m=−∞ n=0
∞ ∞
+ ∑ ∑ bmn Jm (i kn(2) ρ) ei m ϕ sin(kn(2) z) , (3.289a)
m=−∞ n=1
(2n + 1)π 2nπ
kn(1) = , kn(2) = , Φ (ρ, ϕ, z = ±L/2) = 0 . (3.289b)
L L
Observe that the sum over n goes through n = 0, 2, . . . , ∞ in the first sum, whereas
it starts from unity in the second sum. A convenient reformulation can be achieved
in terms of the modified Bessel functions [5, 7–9],
Im (x) = i−m Jm (i x) . (3.290)
The asymptotic behavior of the Bessel I functions is given as

1 x m
Im (x) ∼ Jm (x) ∼ ( ) , x → 0, (3.291a)
Γ (m + 1) 2
1
Im (x) ∼ 1/2
exp (x) , x → ∞. (3.291b)
(2πx)
The general solution (3.289) to the homogeneous equation becomes

∞ ∞
(1)
Φ(ρ, ϕ, z) = ∑ ∑ amn Im (kn ρ) e
imϕ
cos(kn(1) z)
m=−∞ n=1
∞ ∞
+ ∑ ∑ bmn Im (kn(2) ρ) eimϕ sin(kn(2) z) . (3.292)
m=−∞ n=0
The sine terms are asymmetric under z ↔ −z, whereas the cosine terms are
symmetric.
We choose as boundary conditions a zero potential on the endcaps of the cylin-
der, and constant potential on the outer rim,
Φ(ρ = a, ϕ, z) = Φ0 , Φ(ρ, ϕ, z = ±L/2) = 0 . (3.293)
The potential is symmetric under z ↔ −z. Hence,

∞ ∞
Φ(ρ, ϕ, z) = ∑ ∑ amn Im (kn(1) ρ) eimϕ cos(kn(1) z) , (3.294)
m=−∞ n=1
(1)
where we recall that kn = (2n + 1)π/L. The infinitesimal area element on the outer
rim of the cylinder is dA = a dϕ dz with the integration domains ϕ ∈ (0, 2π) and
z ∈ (−L/2, L/2). Hence, the expansion coefficients amn can be determined based on
(a) (b)
Fig. 3.9 Plot of the partial sums up to terms with n = 5 and n = 50 of the expres-
sion (3.298) which describes a zero potential on the cylinder endcaps, but constant poten-
tial on the outer rim of the cylinder. The parameters are a = 1, L = 3 and Φ0 = 2.3. Note
that the potenial Φ(ρ, z) has azimuthal symmetry and is independent of ϕ. Our plot of
Φ(∣ρ∣, z) with −a < ρ < a corresponds to a plot from one outer rim of the cylinder to the
other.
the conditions that

2π L/2 2π L/2
−im′ ϕ (1) ′ (1)
∫ dϕ ∫ dz Φ(a, ϕ, z) e cos(kn′ z) = Φ0 ∫ dϕ ∫ dz e−im ϕ cos(kn′ z) ,
0 −L/2 0 −L/2
(3.295a)
2π L/2 2π L/2
′ (2) ′ (2)
−im ϕ
∫ dϕ ∫ dz Φ(a, ϕ, z) e sin(kn′ z) = Φ0 ∫ dϕ ∫ dz e−im ϕ sin(kn′ z) .
0 −L/2 0 −L/2
(3.295b)
However, again, there is a more direct way in view of the simplicity of the boundary
condition which implies azimuthal symmetry. The only contributing term is the
one with m = 0, and we have
∞
(1)
Φ(ρ, ϕ, z) = ∑ an I0 (kn(1) ρ) cos(k0n z) . (3.296)
n=1
One verifies that

4(−1)n Φ0
an = (1)
. (3.297)
(2n + 1)π I0 (kn a)
The solution to the boundary-value problem (3.293) thus reads as (see Fig. 3.9)
(1)
4Φ0 ∞ (−1)n I0 (kn ρ)
Φ(ρ, ϕ, z) = ∑ cos(kn(1) z) . (3.298)
π n=0 (2n + 1) I0 (kn(1) a)
A few final remarks are in order. A general cylindrical problem with boundary
conditions on the cylinder boundaries (both outer rim as well as endcaps) can be
separated into one with a vanishing solution on the cylinder endcaps and one with
a vanishing solution on the outer cylinder boundaries. One adjusts the coefficients
in Eq. (3.277) so that the boundary conditions on the endcaps are fulfilled, and
then, one adjusts the coefficients in Eq. (3.289) so that the boundary conditions
on the outer rim of the cylinder are fulfilled. A general boundary value problem
on the cylinder thus involves both types of solutions and can be broken into pieces
according to the above analysis.
3.4 Laplace Equation and Dirichlet Green Functions
3.4.1 Dirichlet Green Function for Spherical Shells

In Sec. 3.2, we have considered approximate solutions to the Laplace equation using
the variational principle. In Sec. 3.3, the solution of the Laplace equation using
series expansions was treated. Now, we shall turn our attention to the third method
listed in Sec. 3.1.4, namely, the Dirichlet Green function. Let us consider grounded
concentric spheres, with radii a < b. As discussed in Sec. 3.1.3, the Dirichlet Green
function fulfills
1
∇⃗ 2 GD (⃗
r , r⃗′ )∣r⃗, r⃗′ ∈V = − δ (3) (⃗
r − r⃗′ ) , r , r⃗′ )∣r⃗∈V,⃗r′ ∈∂V = 0 .
GD (⃗ (3.299)
0
In order to illustrate the Green function techniques, we consider systems which lie
in a spherical region a ≤ r ≤ b bounded by two concentric conducting spherical shells
of radii a and b. In this case, we assume that the potentials (rather than the field
strengths) on the conductors are specified and the Dirichlet Green function (rather
than the Neumann Green Function) is required.
In view of the spherical symmetry, we follow the approach outlined in Sec. 2.3.2
and use the following ansatz for the Dirichlet Green function,
∞
r , r⃗′ ) = ∑ ∑ g (r, r′ ) Ym (θ, ϕ) Ym
GD (⃗ ∗
(θ′ , ϕ′ ) . (3.300)
=0 m=−
We proceed as in the case of the ordinary Green function, i.e., we exclude the
value at r = r′ and concatenate two solutions to the homogeneous equation, one for
a ≤ r < r′ and the other for r′ < r ≤ b. Just as in the case of the free Green function
(Sec. 2.3.2), we can combine the solution regular at the origin, and the one regular
at infinity, in such a way that they fulfill the homogeneous equation everywhere
expect at the cusp r = r′ . We then adjust the coefficients such as to ensure that
g (r, r′ ) = 0 at the surfaces of the conductor. A suitable combination is
r a2+1
g (r, r′ ) = α (r′ ) ( − ), a ≤ r < r′ , (3.301a)
r′ +1 +1
(r r′ )
⎛ (r r′ ) r′ ⎞
g (r, r′ ) = β (r′ ) − 2+1 + +1 , r′ < r ≤ b . (3.301b)
⎝ b r ⎠
For a → 0 and b → ∞, both of these equations reproduce the structure encountered
for the free Green function. The ansatz (3.301) is justified by the idea that the radial
terms of the free Green function are supplemented by a product of regular terms
(at the origin, or at infinity) of products of solutions of the homogeneous equations,
symmetric under r ↔ r′ . Furthermore, in view of the boundary conditions imposed
on the Green function, we need to have g (r, r′ ) = 0 at r = a and at r = b. Continuity
of g (r, r′ ) at r = r′ requires that
′ 2
1 a2+1 ′ ⎛ (r ) 1⎞
α (r′ ) ( − ) − β (r ) − + = 0. (3.302)
r′ (r′ )2+2 ⎝ b2+1 r′ ⎠
In order to satisfy the differential equation

1
⃗ 2 GD (⃗
∇ r , r⃗′ ) = − r − r⃗′ ) ,
δ(⃗ (3.303)
0
one needs to integrate Eq. (3.301) from r = r′ − to r = r′ + .
This procedure results in the following equation,
( + 1) a2+1 r′2 ( + 1) 1
α (r′ ) ( ′
+ ′
) + β (r′ ) ( 2+1 + )= . (3.304)
r r 2+2 b r′ 0 r ′
The two Eqs. (3.302) and (3.304) can be solved for α (r′ ) and β (r′ ),
1 1 − (r′ /b) 1 1 − (a/r′ )

2+1 2+1
1 1
α (r′ ) = − , β (r′ ) = − .
0 2 + 1 1 − (a/b)2+1 0 2 + 1 1 − (a/b)2+1
(3.305)
The overall result for the Green function is
1 ∞ (b − r>2+1 ) (r<2+1 − a2+1 )
2+1
′ ∗
r , r⃗′ ) =
GD (⃗ ∑ +1
′
∑ Ym (θ, ϕ) Ym (θ , ϕ ) .
0 =0 (2 + 1) (b2+1 − a2+1 ) (r r′ ) m = −
(3.306)
The Green function vanishes on the two concentric spheres at r = a and at r = b.
Two remarks are in order here. We have considered a system with an inner shell of
radius a and an outer shell of radius b, both held at a constant zero potential. If we
set a = 0, then this gives the Dirichlet Green function for the system confined to lie
inside a spherical shell of radius b. If we let b → ∞, then this is the Dirichlet Green
function for the region outside a spherical shell of radius a.
3.4.2 Boundary Condition of Dipole Symmetry
We now turn to the application of the Dirichlet Green function to a boundary-value

problem. We recall that, using the Dirichlet Green function and a suitable Dirichlet
boundary condition Φ(⃗ r ) for ∂V , and vanishing charge distribution, we can evaluate
r′ ) using Eq. (3.27), which we recall for convenience,
the potential Φ(⃗
r′ ) = −0 ∫
ΦD (⃗ dA⃗ ⋅ ∇G
⃗ D (⃗
r , r⃗′ ) ⋅ Φ (⃗
r) . (3.307)
∂V
Let us formulate a boundary condition which has a dipole structure on the inner
sphere, whereas the outer sphere remains grounded,
Φ (a, θ, ϕ) = fa (θ) = Φ0 cos θ , Φ (b, θ, ϕ) = fb (θ) = 0 . (3.308)
The task then is to find the electrostatic potential everywhere. Using the symmetry
property G (⃗
r , r⃗′ ) = G (⃗
r′ , r⃗) of the Green function, the solution can be written as
Φ (r, θ, ϕ) = a2 ǫ0 S dΩ′ Φ0 cos θ′ GD (⃗

r, r⃗′ )V
∂
. (3.309)
∂r′ r ′ =a
Using the orthogonality of the Legendre polynomials which are “hidden” in the
spherical harmonics, this reduces to
R
∂ ⎛ b − r r − a ⎞RRRR
3 3 ′3 3
a2
Φ (r, θ, ϕ) = R
∂r′ ⎝ (b3 − a3 ) (r r′ )2 ⎠RRRR ′
Φ0 cos θ
Rr =a
3
a b −r
2 3 3
= Φ0 cos θ 2 3 , (3.310)
r b − a3
which interpolates between zero (at r = b) and Φ0 cos θ (at r = a). The concept of
the multipole decomposition, outlined in Sec. 2.3.5, has been used with advantage
in the analysis.
3.4.3 Verification Using Series Expansion

In order to establish contact with the series expansion method (3.3), it is useful to
rederive the result given in Eq. (3.310) using the multipole decomposition, with a
suitable projection of the multipole components. Azimuthal symmetry implies that
in Eq. (3.170), we should only select the spherical harmonics with ℓ = 0. Alterna-
tively, in view of Eq. (2.53), we may formulate the angular part as being proportional
to the Legendre polynomial Pℓ (cos θ), to obtain
Φ (r, θ) = Q αℓ rℓ + βℓ r−ℓ−1 Pℓ (cos θ) , Φ (a, θ) = fa (θ) , Φ (b, θ) = fb (θ) .

∞
ℓ=0
(3.311)
A rather straightforward calculation shows that the αℓ and βℓ are given by
aℓ+1 Iℓ (a) − bℓ+1 Iℓ (b)
αℓ = , (3.312a)
a2ℓ+1 − b2ℓ+1
ℓ+1 b Iℓ (a) − a Iℓ (b)
ℓ ℓ
βℓ = − (a b) , (3.312b)
a 2ℓ+1 −b 2ℓ+1
where the overlap integrals Iℓ (a) involve the boundary conditions fa (θ) and fb (θ),
2ℓ + 1 π
Iℓ (a) = S fa (θ) Pℓ (cos θ) sin θ dθ , (3.313a)
2 0
2ℓ + 1 π
Iℓ (b) = S fb (θ) Pℓ (cos θ) sin θ dθ . (3.313b)
2 0
Using the boundary conditions (3.308), we obtain I (b) = 0 for all , and the only
nonvanishing α and β coefficients are found for = 1,
a2 a2 b 3
I1 (a) = Φ0 , α1 = Φ0
, β 1 = Φ0 3 3 . (3.314)
−b 3 a3 a −b
Plugging these coefficients into Eq. (3.311), we confirm that
a2 b 3 − r 3
Φ (r, θ, ϕ) = Φ0 cos θ , (3.315)
r 2 b 3 − a3
and thus, the validity of the result given in Eq. (3.310).
3.5 Poisson Equation and Dirichlet Green Functions
3.5.1 Source Terms with Boundary Conditions

In Sec. 3.1.3, we had developed the formalism for the treatment of boundary-value
problems (of the Dirichlet and Neumann type) with nontrivial source terms and
nontrivial boundary conditions. Given a Dirichlet boundary condition Φ(⃗r ) = f (⃗
r)
for r ∈ ∂V , we recall that the solution is given by Eq. (3.21),
r′ ) = ∫ d3 r GD (⃗
ΦD (⃗ r , r⃗′ ) ρ (⃗
r ) − 0 ∫ dS⃗ ⋅ ∇G
⃗ D (⃗
r , r⃗′ ) f (⃗
r) , (3.316)
V ∂V
where the defining relations for the Dirichlet Green function are [see Eq. (3.20)],
⃗ 2 GD (⃗ 1
∇ r , r⃗′ )∣r⃗, r⃗′ ∈V = − δ (3) (⃗
r − r⃗′ ) , r , r⃗′ )∣r⃗∈V,⃗r′ ∈∂V = 0 .
GD (⃗ (3.317)
0
r ) = 0, then the solution to the boundary-value
If the charge distribution vanishes, ρ(⃗
problem is given by the second term on the right-hand side of (3.316). By contrast,
if the potential is supposed to vanish on the boundary, then the solution is given
by the first term on the right-hand side of (3.316).
A hybrid approach can be taken to problems which involve both nontrivial
boundary conditions as well as nontrivial source terms. Namely, one can split
Dirichlet boundary-value problems into two parts, the first of which is given by the
charge distribution and needs to be integrated using the Dirichlet Green function,
and the second of which is given by the boundary condition and can be integrated
using a series expansion, using the variational principle or with the Dirichlet Green
function itself. In the following, we shall consider the first term on the right-hand
side of (3.316), and its integration using the Dirichlet Green function.
3.5.2 Sources and Fields in a Spherical Shell

We consider the Dirichlet Green function for concentric spheres with radii a < b [see
Eq. (3.306)],
1 ∞ (b − r>2+1 ) (r<2+1 − a2+1 )
2+1
∗
r , r⃗′ ) =
GD (⃗ ∑ +1 ∑ Ym (θ, ϕ) Ym (θ′ , ϕ′ ) .
0 =0 (2 + 1) (b 2+1 −a 2+1 ′
) (r r ) m = −
(3.318)
For concentric spheres with radii a < b, we consider a uniformly charged annulus
(ring) with total charge Q0 , located in the plane θ = π/2 (xy plane), i.e., stretched
between the two concentric spherical shells. This corresponds to a nontrivial source
term in Eq. (3.316); we recall that nontrivial boundary conditions have been treated
using two alternative approaches in Sec. 3.4. The potential in the interior region
between the concentric spheres, due to the uniformly charged annulus, can be cal-
culated with the help of the Dirichlet Green function (3.318), as follows,
r ) = ∫ d3 r′ GD (⃗
ΦD (⃗ r , r⃗′ ) ρ (⃗
r′ ) . (3.319)
V
The (constant) surface charge density on the ring (annulus) between the spheres is
Q0
σ0 = , (3.320)
π (b2 − a2 )
where Q0 is the total charge on the annulus [see also Fig. 3.10(a)]. The volume
charge density on the annulus therefore is
σ0
r′ ) = σ0 δ (z ′ ) = σ0 δ (r′ cos θ′ ) = ′ δ (cos θ′ ) ,
ρ (⃗ a ≤ r ≤ b, (3.321)
r
where the latter equality is valid for purposes of the integration over θ′ , in which
case r′ can be assumed to be a constant. The boundary conditions, which are
automatically implemented by the use of the Dirichlet Green function, read as
Φ(r = a, θ, ϕ) = Φ(r = b, θ, ϕ) = 0. We use Eq. (3.319) and write
Φ (r, θ, ϕ) = ∫ d3 r′ GD (⃗
r , r⃗′ ) ρ (⃗
r′ )
a≤r ′ ≤b
∞ (b2+1 − r2+1 ) (r2+1 − a2+1 )

1 3 ′ > <
= ∫ d r ∑
2+1 − a2+1 ) (r r′ )+1
0 a≤r′ ≤b =0 (2 + 1) (b

∗
× ∑ Ym (θ, ϕ) Ym (θ′ , ϕ′ ) σ0 δ (r′ cos θ′ )
m = −
∞ (b2+1 − r2+1 ) (r2+1 − a2+1 )

1 ′ ′2 ′ > <
= ∫ ′ dr r (2π) d cos θ ∑ +1
0 a≤r ≤b =0 (2 + 1) (b
2+1 −a 2+1 ) (r r′ )
√ 2
⎛ 2 + 1 ⎞ σ0
× P (cos θ) P (cos θ′ ) ′ δ (cos θ′ )
⎝ 4π ⎠ r
σ0 ∞ P (cos θ) P (0) b (b 2+1

− r>2+1 ) (r<2+1 − a2+1 )
= ∑ 2+1 ∫ dr′ r′ .
2 0 =0 b −a 2+1 a +1
(r r′ )
(3.322)
For the Legendre polynomials of zero argument, one can use the formula
√
π π Γ( 12 + 12 )
P (0) = = cos ( ) √ . (3.323)
Γ( 21 − 12 ) Γ(1 + 12 ) 2 π Γ(1 + 12 )
(a) (b) (c)
Fig. 3.10 Panel (a) shows the schematic arrangement consisting of the two concentric
spheres, with the annulus centered in the z = 0 plane. Panel (b) shows a plot of the po-
tential (3.326) in the plane of azimuth angle ϕ = 0, i.e., in the y = 0 plane. The potential
has a peak at z = 0, as outlined in the plot. Panel (c) shows the position of the annulus in
plot (b), confirming that the potential peaks where the annulus is. The problem is symmetric
with respect to the azimuth angle, i.e., symmetric with respect to rotations about the z axis.
The integral over r′ can be carried out for the first few by an explicit calculation,
where we use the general identity
b r b
∫ dr′ f (r< , r> ) = ∫ dr′ f (r′ , r) + ∫ dr′ f (r, r′ ) . (3.324)
a a r
The result for the first two novanishing terms (monopole and quadrupole) is
σ0 (a − r) (r − b)
Φ(r, θ, ϕ) =
0 4r
5σ0 b (b − r) r4 + a5 (b4 − r4 ) + a4 (r5 − b5 )
4
− (3 cos2 θ − 1) + . . . .
320 (b5 − a5 ) r3
(3.325)
The general result consists of the even multipoles,
σ0 ∞ π ( + ) Γ( 2 + 2 )
1 1 1
Φ(r, θ, ϕ) = ∑ cos ( ) √ 2 P (cos θ)
0 =0 2 π Γ(1 + 12 )
(b r)+1 (b r − b r ) + a2+1 (b+2 − r+2 ) + a+2 (r2+1 − b2+1 )

× .
r+1 (b2+1 − a2+1 ) (2 + − 2)
(3.326)
Odd are eliminated under the sum by the factor cos(π/2). The plot in Fig. 3.10
shows the sum of the first 40 multipoles in the expression (3.326) for the solution
of the charged annulus inside the two concentric, grounded, spherical shells. It is
given in the xz plane. The annulus lies in the plane z = 0 and is located where the
potential is large. As we cannot realistically plot the potential as a function of the
three coordinates x, y and z, we have chosen the x and z coordinates. The potential
is rotationally symmetric about the z axis.
3.5.3 Induced Charge Distributions on the Boundaries
In Sec. 3.5.2, we have considered a situation where a nonvanishing charge distribu-

tion leads to a nontrivial potential in the interior region between the two spherical
shells. The boundary condition for the potential, namely, the assumption that
Φ(⃗r ) = 0 for r = a and r = b, is automatically implemented through the use of the
Dirichlet Green function. Given the charge density on the annulus as in Eq. (3.321),
Dirichlet boundary conditions on the two concentric spheres actually imply that
compensating charge must be brought onto the two concentric spheres in order to
annihilate the electrostatic potential there.
The charge on the inner surface of the outer sphere can be calculated as follows
(r> = r, and r< = r′ ). We invoke a Gaussian cylinder surrounding the surface,
and take into account that the outer layer is grounded, i.e., held at zero potential.
Hence, the electric field outside of the concentric sphere vanishes. Furthermore, this
implies that the contribution to the surface integral (through the Gaussian cylinder)
comes exclusively from the “inner” region where the surface normal is −êr . So, there
isn’t a “factor 1/2 coming in from the other cap of the Gaussian cylinder” as one
might otherwise suggest. We simply have σ(⃗ r ) = −0 Er where Er is the normal
component of the electric field (which is pointing in the negative radial direction
for the outer sphere),
∂Φ (r, θ, ϕ)
Qb = ∫ r) = ∫
dΩ r2 σ(⃗ dΩ r2 (−0 Er ) = b2 0 ∫ dΩ ∣
r=b r=b ∂r r=b
∞
σ0 b2 1 P (cos θ) P (0)
= (2π) ∑ ∫ d cos θ 2+1
2 =0 −1 b − a2+1
∂ b (b2+1 − r2+1 ) (r′2+1 − a2+1 ) bbbb

× ∫ dr ′ ′
r bbb
∂r a (r r′ )
+1 bbb
br=b
′
σ0 b2 1 P0 (cos θ)P0 (0) ∂ b
′ b−r ′ r −a
= (2π) ∫ d cos θ ∫ dr r ∣
2 −1 b−a ∂r a r r′ r=b
σ0 b2 2 ∂ b−r b
= (2π) ( ∣ ) ∫ dr′ (r′ − a)
2 b − a ∂r r r=b a
σ0 b2 1 b b
= (2π) (2) (− 2 ) ∫ dr′ (r′ − a) = −πσ0 b (b − a) , (3.327)
2 b−a b a
which has the correct physical dimension. In the derivation, we have started from
the integral representation in the last line of Eq. (3.322). On the inner spherical
shell (r = a), we have σ(⃗r ) = 0 Er because the normal component of the electric
field is pointing in the positive radial direction. Furthermore, on the outer surface
of the inner sphere, we have r> = r′ and r< = r, and the surface charge therefore is
given by
∂Φ (r, θ, ϕ)
Qa = ∫ r) = ∫
dΩ r2 σ(⃗ dΩ r2 (+0 Er ) = −a2 0 ∫ dΩ ∣
r=a r=a ∂r r=a
∞
σ0 a2 1 P (cos θ) P (0)
= − (2π) ∑ ∫ d cos θ 2+1
2 =0 −1 b − a2+1
(b 2+1
− r′2+1 ) (r2+1 − a2+1 ) bbb
dr′ bbbb
∂ b
′
× ∫ r
∂r a (r r′ )
+1 bbb
br=a
1 P (cos θ) P (0) ′
σ0 a2 ′b−r r−a
∂ b
dr′ ∣
0 0
= − (2π) ∫ ( d cos θ) ∫ r ′
2 −1 b−a ∂r a r r r=b
σ0 a2 2 ∂ r−a b
= − (2π) ( ∣ ) ∫ (b − r′ ) dr′
2 b − a ∂r r r=a a
σ0 a2 1 a b
= − (2π) (2) ( 2 ) ∫ (b − r′ ) dr′ = −πσ0 a (b − a) . (3.328)
2 b−a a a
The two charges Qb and Qa both have the same sign but are manifestly different;
the ratio is b/a. This is in line with intuition because of the following consideration:
Namely, if the induced charge density on both concentric spheres were the same,
then the ratio would be equal to the ratio of the surface areas, i.e., b2 /a2 . However,
the average distance to the surface elements of the inner concentric sphere is smaller,
and because the potential goes with the inverse distance, the change in the ratio
from b2 /a2 to b/a is in line with intuition. In the derivation, we have used the
orthogonality of the Legendre polynomials in the form
1 1 1
∫ P (x) P (0) dx = P (0) ∫ P (x) dx = P (0) ∫ P (x) P0 (x) dx
−1 −1 −1
2 δ0
= P (0) = 2 δ0 , (3.329)
+1
where P0 (0) = P0 (x) = 1. The orthogonality property (2.48) has also been employed.
3.6 Exercises
● Exercise 3.1: For r⃗′ ∈ ∂V , Eq. (3.21) reduces to
r′ )∣
ΦD (⃗ =∫ dA⃗ ⋅ ∇G
⃗ D (⃗
r , r⃗′ ) f (⃗
r) . (3.330)
r⃗′ ∈∂V ∂V
By applying the divergence theorem to the right-hand side, show that the right-hand
r′ ), tantamount to the Dirichlet boundary condition.
side is equal to f (⃗
● Exercise 3.2: For r⃗′ ∈ ∂V , Eq. (3.26) can be written as
⃗ r⃗′ ΦN (⃗
∇ ⃗ r⃗′ GN (⃗
r′ )∣r⃗′ ∈∂V = ∫ d3 r∇ r , r⃗′ ) ρ (⃗
r ) + 0 ∫ ⃗ r⃗′ dA⃗ ⋅ ∇
(∇ ⃗ r⃗Φ(⃗ r, r⃗′ )) .
r)GN (⃗
V ∂V
(3.331)
Show that the right-hand side of Eq. (3.331) is indeed equal to g⃗(⃗ r′ ) is
r), where g⃗(⃗
defined as in Eq. (3.15), by applying the divergence theorem to the third term on
the right-hand side of Eq. (3.331).
● Exercise 3.3: For a functional of the form
S[f ] = ∫ dx F (f (x), f ′ (x)) , (3.332)
show that
δS ∂F d ∂F
= − (3.333)
δf (x) ∂f (x) dx ∂f ′ (x)
where you treat the partial differentiations as if f and f ′ were independent variables.
Hint: In carrying out the variations, you still need to take into account that a change
in f → f + δf implies a concomitant change in f ′ → f ′ + δf ′ .
● Exercise 3.4: Consider the three-dimensional generalization of Eq. (3.45),
δS r′ ) + δ (3) (⃗
F (f (⃗ r′ − r⃗)) − F (f (⃗
r′ ))
= lim ∫ d3 r′ . (3.334)
δf (⃗
r ) →0 V
Show that, for
1 3 ′ ⃗
S[f (⃗
r)] = ∫ d r [∇f r′ )]2 ,
(⃗ (3.335)
2 V
one has
δS
⃗ 2 f (⃗
= −∇ r) . (3.336)
δf (⃗
r)
● Exercise 3.5: For the functional
2
⃗ r )) ,
S[Φ] = ∫ d3 r (∇Φ(⃗ (3.337)
V
show that
δ2 S
= −2δ (3) (⃗ ⃗2 .
r − r⃗′ ) ∇ (3.338)
r′ )
r) ∂Φ(⃗
∂Φ(⃗
Verify the result in two ways, (i) with recourse to the definition (3.45), suitably gen-
eralized to a three-dimensional integral, (ii) by writing the first functional derivative
as an integral, and then reading off the second functional derivative “under the in-
tegral sign”, as in Eq. (3.44).
● Exercise 3.6: Carry out the generalization of Eqs. (3.71)–(3.73) to three dimen-
sions explicitly. For an infinite integration volume, consult Chaps. 4 and 6 for an
introduction to Fourier transforms.
● Exercise 3.7: Write the following ansatz for the potential energy stored in the
potential in a long coaxial cylinder,
c
0 ∂w 2
E= (2π) L ∫ dρ ρ ∣ ∣ , (3.339)
2 ∂ρ
b
where w = w(ρ) is your trial function for the potential, and L is the cylinder length.
This ansatz is inspired by Eq. (3.105) upon dropping the dependence on ϕ and z of
the potential, which is irrelevant in the limit of large L. By variation of this function
using the Euler–Lagrange equations, obtain the exact form given in Eq. (3.117).
● Exercise 3.8: Show Eq. (3.124) by expressing the unit vector ̂ n in Cartesian
coordinates, and using the definition (1.16) of the gradient operator.
● Exercise 3.9: Calculate the surface area of an ellipsoid given by the equation
r ) = (x2 + y 2 )/a2 + z 2 /c2 − 1 = 0
F (⃗ (3.340)
using Eq. (3.128).
● Exercise 3.10: Calculate the potential on the symmetry axis due to a uniform
charge density σ0 on the surface of an ellipsoid given by Eq. (3.340). Employ
Eq. (3.128).
● Exercise 3.11: Exercise: Verify the result (3.147) by a numerical integration,
choosing convenient numerical values of the parameters (proposal: b = 0.2 cm, ρ =
1.0 cm).
● Exercise 3.12: Treat the following boundary-value problem for a cube with
volume 0 < x < a, 0 < y < a, and 0 < z < a, whose upper side at z = a is at a potential
Φ0 , with grounded five other surfaces (zero potential on five of the six surfaces).
Hint: you may want to choose the ansatz
√
nπx mπy ⎛ nπ 2 mπ 2 ⎞
Φ(x, y, z) = ∑ Cnm sin ( ) sin ( ) sinh ( ) +( ) z . (3.341)
mn a a ⎝ a a ⎠
Now set Φ = Φ0 for z = a, resulting in the condition
a a
n′ πx m′ πy
∫ dx ∫ dy Φ(x, y, a) sin ( ) sin ( )
a a
0 0
a a
n′ πx m′ πy
= Φ0 ∫ dx ∫ dy sin ( ) sin ( ). (3.342)
a a
0 0
● Exercise 3.13: Repeat the calculation in Sec. 3.3.4 for the boundary condition
y 2
g (y) = Φ0 ( ) . (3.343)
b
● Exercise 3.14: Show that (maybe with the help of computer algebra)
1 d6 sin(z) 2
6
{z 7 ( 6 )}∣ = . (3.344)
6! dz z cos(z) z=0 15
Thus, calculate the residue

Res [tan(z) z −6 ] . (3.345)
z=0
● Exercise 3.15: Show that

1 1
Res ( 4
exp(z)) = , (3.346a)
z=0 z 6
1 1
Res ( exp(z)) = , (3.346b)
z=0 z5 24
1 exp(iz)
Res ( ) = 0, (3.346c)
z=0 z 5 sin(z)
1 1
Res ( 4
exp(z) cos(z)) = − , (3.346d)
z=0 z 3
1 1
Res (
exp(z) cos(z)) = − . (3.346e)
z5
z=0 6
● Exercise 3.16: Repeat the derivation outlined in Eqs. (3.264)–(3.272) upon an
alteration of the pole prescription in Eq. (3.263), i.e., 1/k → 1/(k + i) instead of
1/k → 1/(k − i).
● Exercise 3.17: Fill in the missing intermediate steps in the derivation of
Eq. (3.288) for the boundary conditions given in Eq. (3.279). Then, generalize
the treatment for the following boundary-value problem:
Φ(ρ = a, ϕ, z) = 0 , Φ(ρ, ϕ, z = ±L/2) = ±Φ0 , (3.347)
i.e., for an upper, positively charged endcap, and a lower, negatively charged endcap.
Eq. (3.298) for the boundary conditions given in Eq. (3.293). Then, generalize
the treatment for the following boundary-value problem:
2z
Φ(ρ = a, ϕ, z) = Φ0 , Φ(ρ, ϕ, z = ±L/2) = 0 , (3.348)
L
i.e., for grounded endcaps, and a nonvanishing potential on the outer rim of the
cylinder, which is odd under z ↔ −z.
● Exercise 3.19: Derive (3.304) based on (3.302) and ideas outlined in Eqs. (2.81)–
(2.90). Hint: Consider Eq. (2.86), in the form
∂ ∂ 1
r′2 ( G (r, r′ )∣ − G (r, r′ )∣ )=− , (3.349)
∂r r=r ′ +ε ∂r r=r ′ −ε 0
Eq. (3.310), starting from Eq. (3.308). In particular, clarify why the expression
⃗ D (⃗
∇G r , r⃗′ ) ⋅ dA⃗ can be replaced by [(∂/∂r)GD (⃗ r , r⃗′ )] dA. Generalize the solution
for the boundary-value problem
Φ (a, θ, ϕ) = fa (θ) = 0 , Φ (b, θ, ϕ) = fb (θ) = Φ0 cos θ . (3.350)
● Exercise 3.21: Fill in the missing steps in the derivation extending from
Eq. (3.311) to (3.315). Hint: Use the orthogonality relation (2.48) and Eq. (2.44).
● Exercise 3.22: Repeat the analysis in Secs. 3.4.2 and 3.4.3 for the boundary
conditions
Φ (a, θ, ϕ) = fa (θ) = Φ0 (3 cos2 θ − 1) , Φ (b, θ, ϕ) = fb (θ) = 0 . (3.351)
● Exercise 3.23: Derive Eq. (3.326), using as input all other formulas presented
in Sec. 3.5.1.
Chapter 4
Green Functions of Electrodynamics
4.1 Green Function for the Wave Equation
4.1.1 Integral Representation of the Green Function
In contrast to electrostatics, electrodynamics describes time-dependent sources, vec-

tor potentials and fields. Of prime importance are electromagnetic wave phenomena.
In consequence, it is natural to start the discussion of electrodynamics by consider-
ing the wave equation, first, in the context of a model problem. Namely, we assume
that a signal function ψ(⃗ r, t) is generated by sources F (⃗
r , t) and has the general
form
1 ∂2 1
( ⃗ 2 ) ψ (⃗
−∇ r , t) = F (⃗
r, t) . (4.1)
2
c ∂t 2 0
One approach to the solution of this equation is to work with the space-time Fourier
transforms of the fields and sources. The space-time functions are related to the
⃗ ω) and F (k,
corresponding functions in Fourier space, referred to as ψ(k, ⃗ ω),
d3 k dω ⃗
ψ(⃗r , t) = ∫ ∫ ψ(k, ω) exp (ik⃗ ⋅ r⃗ − iωt) , (4.2a)
(2π) 3 2π
d3 k dω ⃗ ω) exp (ik⃗ ⋅ r⃗ − iωt) .
F (⃗
r , t) = ∫ ∫ F (k, (4.2b)
(2π) 3 2π
Here, we denote the original function as well as its Fourier transform by the same
symbol, as already done in Sec. 2.2. The wave vector is k⃗ and the angular frequency
is denoted as ω. The Fourier backtransformation is given as
ψ(k, r, t) exp (−ik⃗ ⋅ r⃗ + iωt) ,

⃗ ω) = ∫ d3 r ∫ dt ψ(⃗ (4.3a)
F (k, r, t) exp (−ik⃗ ⋅ r⃗ + iωt) ,

⃗ ω) = ∫ d3 r ∫ dt F (⃗ (4.3b)
where we use the relationship

∞ dv −iu v
∫ e = δ(u) . (4.4)
−∞ 2π
137
The Fourier transform of the wave function fulfills the Fourier transformed version
of Eq. (4.1),
d3 k dω 1 ∂2 ⃗ r −iωt
⃗ ω) eik⋅⃗
∫ ∫ ⃗2 +
(−∇ ) ψ(k,
(2π) 3 2π c2 ∂t2
1 d3 k dω ⃗ r −iωt
⃗ ω) eik⋅⃗
=
∫ ∫ F (k, . (4.5)
0 (2π)3 2π
Because the differential operators only act on the exponential terms, the equation
is converted into an algebraic equation,
d3 k dω 2
⃗ 2 − ω ) ψ(k,
⃗ ω) − 1 F (k, ⃗ r −iωt
⃗ ω)} eik⋅⃗
∫ ∫ {( k = 0. (4.6)
(2π) 3 2π c 2 0
Since each Fourier component exp(ik⃗ ⋅ r⃗−iωt) is linearly independent, all the Fourier
coefficients must be zero. In other words, if a function vanishes, then its Fourier
transform vanishes, and vice versa. The solution of the algebraic equation fulfilled
by the Fourier transform therefore is
⃗
⃗ ω) = 1 F (k, ω) .
ψ(k, (4.7)
0 k⃗2 − (ω/c)2
The problem of calculating ψ(⃗ r, t) is now reduced to taking the inverse Fourier
transform of the expression F (k, ⃗ ω)/[k⃗2 − (ω/c)2 ]. There is at least one difficulty.
The integrand has singularities at ω = ±c ∣k∣, ⃗ which need to be handled when evalu-
ating the ω integral. The different choices for the contour will later be understood
in terms of the different choices for the boundary conditions imposed on the Green
function. In the case of electrostatics, the Green function G(⃗ r , r⃗′ ) = 1/(4π0 ∣⃗
r − r⃗′ ∣)
can be modified to fulfill different boundary conditions (in space), by adding to the
Green function a solution of the homogeneous equation. Likewise, here, we will
see that the Green function can fulfill the defining equation and still fulfill different
boundary conditions in space and time.
The Green function G(⃗ r − r⃗′ , t − t′ ) with its Fourier transform G(k, ⃗ ω) solves
Eq. (4.1) with the source term given by
d3 k dω ik⋅(⃗
⃗ ′ ′
r, t) = δ (3) (⃗
F (⃗ r − r⃗′ ) δ (t − t′ ) = ∫
∫ e r−⃗r )−iω(t−t ) , (4.8)
(2π) 3 2π
i.e., the Green function in coordinate space fulfills
1 ∂2 1 (3)
( ⃗ 2 ) G(⃗
−∇ r − r⃗′ , t − t′ ) = r − r⃗′ ) δ (t − t′ ) ,
δ (⃗ (4.9)
c2 ∂t2 0
where ∇⃗ = ∂/∂⃗r is a differential operator with respect to r⃗ (not r⃗′ ). In the case of
translation invariance, the dependence of the Green function on r⃗′ is absorbed in
the notation
r − r⃗′ , t − t′ ) ≡ G(⃗
G(⃗ r , t, r⃗′ , t′ ) , (4.10)
where the latter form explicitly indicates the dependence on the four independent
arguments. One can find the Fourier transform G(k, ⃗ ω) by applying the operator
Green Functions of Electrodynamics 139
⃗ 2 under the integral sign, to the Fourier decomposition. This oper-

(1/c2 )∂ 2 /∂t2 − ∇
ation is analogous to the calculation that leads to Eqs. (4.6) and (4.7), and results
in the identity
d3 k dω 2
⃗ 2 − ω ) G(k,
⃗ ω) − 1 } eik⋅(⃗
⃗ r−⃗r ′ )−iω(t−t′ )
∫ {( k = 0, (4.11)
(2π)3 2π c2 0
which must be valid for all Fourier components separately. In Fourier space, the
Green function therefore reads as
⃗ ω) = 1 1
G(k, . (4.12)
⃗
0 k − (ω/c)2
2
Furthermore, in terms of space-time coordinates, the Green function for the

wave equation is found by Fourier backtransformation
⃗ ′ ′
′ ′ 1 d3 k dω eik⋅(⃗r−⃗r )−iω(t−t )
r − r⃗ , t − t ) =
G(⃗ ∫ ∫ . (4.13)
0 (2π)3 2π k⃗ 2 − (ω/c)2
The integrand has singularities at k = ±ω/c.

As already anticipated in the discussion following Eq. (4.7), we must invoke
Cauchy’s residue theorem in evaluating the pole contributions in Eq. (4.13) (in the
ω integral), after a suitable choice of the integration contour. In view of the identity
1 c ⃗
ω − c ∣k∣ ⃗
ω + c ∣k∣ c 1 1
= ( − )= ( − ) , (4.14)
k⃗2 − ω 2 /c2 ⃗ ω 2 − (c k)
2 ∣k∣ ⃗ 2 ω 2 − (c k)
⃗ 2 ⃗ ω + c ∣k∣
2 ∣k∣ ⃗ ω − c ∣k∣
⃗
we write the Green function as follows,

∞ ⃗ ′ ∞
d3 k c ei k⋅(⃗r−⃗r ) dω −i ω(t−t′ ) 1 1
G(⃗ r′ , t−t′ ) = ∫
r −⃗ ∫ e ( − ) . (4.15)
(2π) 3 ⃗ 0
2∣k∣ 2π ω + c ∣ ⃗ ω − c ∣k∣
k∣ ⃗
−∞ −∞
The ω integration over the interval ω ∈ (−∞, +∞) cannot be properly defined in
the Riemannian sense; the integration contour has to be deformed into the complex
plane in order for the residue theorem to be applicable. The integrand has two
first-order poles along the real ω axis. We write G as follows,
⃗ ′
d3 k ⃗ PC (t − t′ , ∣k∣)
⃗ , ⃗ = c ei k⋅(⃗r−⃗r )
r − r⃗′ , t − t′ ) = ∫
G(⃗ f (k) f (k) , (4.16)
(2π)3 ⃗ 0
2∣k∣
where the following characteristic frequency integral needs to be studied,
∞
⃗ =∫ dω −i ω(t−t′ ) 1 1
PC (t − t′ , ∣k∣) e ( − )
2π ⃗ ω − c ∣k∣
ω + c ∣k∣ ⃗
−∞
dω −i ω(t−t′ ) 1 1
→∮ e ( − ). (4.17)
2π ⃗ ω − c ∣k∣
ω + c ∣k∣ ⃗
C
Im ω
Im ω
Contour CA
Contour CR
Re ω
Re ω
(a) (b)
Fig. 4.1 The ω integration contours CR and CA in the complex plane are illustrated.
Panel (a) shows the contour that gives the retarded Green function, while panel (b) leads
to the advanced Green function. The contour CR used in evaluating the integral in
Eq. (4.17) encircles the poles from above the real axis. For t < t′ , we have to close the
contour CR in the upper complex half plane, and the result for the integral (4.17) is zero;
the retarded Green function vanishes for t < t′ . By contrast, the contour CA is below the
real axis. So, if CA is used, then for t > t′ , the contour CA is completed below the real
axis and the advanced Green function vanishes.
Here, C is a contour that is infinitesimally displaced from the poles. We will discuss
its explicit form below; there is some freedom of choice. Indeed, the value of PC (t −
⃗ depends on the choice of the contour. Integrating ω over the interval (−∞, ∞),
t′ , ∣k∣)
one encounters poles at ω = −c∣k∣⃗ and ω = +c∣k∣.
⃗ The question then is how these poles
should be handled. Indeed, all reasonable ways of treating the poles on the initial
integration contour lead to Green functions that solve the defining Eq. (4.9). One
may, for example, treat the pole terms using a principal-value integration, which
is equivalent to the arithmetic mean of the results obtained using the “upper” and
“lower” contours, which encircle the poles infinitesimally above and below the real
axis. Alternatively, one may either encircle both of the poles infinitesimally above or
below the real axis, or one above and one below. All Green functions thus obtained
fulfill the defining equation (4.9). In order to apply the residue theorem, we must
close the contour C in Eq. (4.17). The poles have been infinitesimally displaced
from the real axis, and the contour initially extends from ω = −∞ to ω = ∞; we need
to close it alongside a semi-circle at ∣ω∣ = ∞, with exponential damping in the upper
or lower half plane to ensure the convergence of the integral. In our derivations, we
thus assume that the integral along the infinite semi-circle S at ∣ω∣ = ∞,
⃗ =∫ dω −i ω(t−t′ ) 1 1
QC (t − t′ , ∣k∣) e ( − ) (4.18)
2π ⃗ ω − c ∣k∣
ω + c ∣k∣ ⃗
S
vanishes because of the exponential suppression of the integrand for large ∣ω∣. As
we shall see, the way in which the contour has to be closed may depend on the sign
of t − t′ . Details are discussed in the following. Possible choices for the contour C
are given in Fig. 4.1 and Fig. 4.2.
4.1.2 Why So Many Green Functions?
In the following considerations, we shall encounter the retarded, advanced, and

Feynman1 Green functions. All of these solve the basic, defining equation for the
Green function, given in Eq. (4.9), which we recall for convenience,
1 ∂2 1 (3)
( ⃗ 2 ) G (⃗
−∇ r − r⃗′ , t − t′ ) = r − r⃗′ ) δ (t − t′ ) .
δ (⃗ (4.19)
c2 ∂t2 0
However, the Green functions differ in their causal behavior and in their analytic
structure.
In terms of a classical interpretation, let us first consider the case where the
Green function describes a string. The retarded Green function GR (x, t, x′ , t′ ) gives
the response of the string (initially at rest) to a “unit of momentum” (Dirac-δ
disturbance) applied to the string at a point in time t′ at a point x′ along the
string, with the response being recorded at space-time point (x, t) with t > t′ . The
advanced Green function GA (x, t, x′ , t′ ) gives the initial field configuration of the
string, designed such that a “unit of momentum” applied at (x′ , t′ ) causes it to
come to rest. The initial configuration is described at the space-time points (x, t)
with t < t′ . The advanced Green function “propagates into the past”, the retarded
Green function “propagates into the future”.
In electrodynamics, the retarded Green function describes outgoing waves [scalar
and vector potentials Φ(⃗ ⃗ r, t)], generated by the sources ρ(⃗
r , t) and A(⃗ r′ , t′ ) and
⃗ r′ , t′ ), with t > t′ , i.e., the electromagnetic potentials and fields are generated by
J(⃗
a unit disturbance (Dirac-δ function) at (⃗ r′ , t′ ), with the fields being zero for t < t′ .
The advanced Green function in electrodynamics describes incoming waves, i.e., it
describes the scalar and vector potentials Φ(⃗ ⃗ r, t)] which converge to a
r , t) and A(⃗
unit disturbance at (⃗ r , t ), with t < t , with the fields being zero for t > t′ .
′ ′ ′
The Feynman Green function was originally devised by Stueckelberg2 and

Feynman. The Feynman Green function essentially relies on a complex formalism, in
the sense of a Fourier transform of the source configuration. Positive-frequency com-
ponents of the source configuration are interpreted as sources for outgoing waves,
which propagate into the future, whereas negative-frequency components are inter-
preted as sinks of incoming waves, which propagate into the past. This enables one
to describe both field quanta creation as well as field quanta annihilation processes
with one and the same Green function; one thus summarizes a number of different
time orderings in the expansion of the time-dependent perturbation theory, into one
and the same Green function [20].
The necessity of introducing yet a third Green function, namely, the Feyn-
man Green function comes from field theory, notably, quantum electrodynamics.
Roughly speaking, in field theory, in order to describe processes which involve the
time evolution of the quantum fields, one needs to consider the so-called time-
1
Richard Phillips Feynman (1918–1988)
2
Count Ernst Carl Gerlach von Stueckelberg (1905–1984)
ordered T product of field operators [see Eqs. (3.124) and (3.125) of Ref. [20]],
⟨0 ∣T Aμ (x) Aν (x′ )∣ 0⟩ = Θ(t − t′) ⟨0 ∣Aμ (x) Aν (x′ )∣ 0⟩ + Θ(t′ − t) ⟨0 ∣Aμ (x′ ) Aν (x)∣ 0⟩ ,
(4.20)
where Θ is the Heaviside step function, x = (t, r⃗) is a space-time coordinate four-
vector, and Aμ (x) is a four-vector potential operator (μ, ν = 0, . . . , 3). The vacuum
state is denoted as ∣0⟩. The Green function
′
Gμν
F (x − x ) = g
μν
GF (x − x′ ) = −i ⟨0 ∣T Aμ (x) Aν (x′ )∣ 0⟩ (4.21)
contains both retarded [∝ Θ(t − t′ )] as well as advanced [∝ Θ(t′ − t)] components
and is proportional to the Feynman propagator GF . The space-time metric is
g μν = diag(1, −1, −1, −1).
A final point: The three Green functions (retarded, advanced, and Feynman)
differ by a solution of the homogeneous equation, i.e.,
1 ∂2
( ⃗ 2 ) (GF (⃗
−∇ r − r⃗′ , t − t′ ) − GR (⃗
r − r⃗′ , t − t′ )) = 0 , (4.22)
c2 ∂t2
and the same relation holds for GF − GA , and GR − GA . In our electrostatic analogy
the different Green functions would correspond to different solutions of the defining
equation (2.66) of the electrostatic Green function, which we recall as
1 (3)
⃗ 2 G(⃗
∇ r − r⃗′ ) = − r − r⃗′ ) ,
δ (⃗ (4.23)
0
with different boundary conditions, implemented for the “boundaries” of space-time
which can include the half-spaces t > t′ and t′ < t. Pursuing the analogy further,
one may remark that the Green function of electrostatics has the representation
−1/k⃗2 in wave number space, but in coordinate space, one may add a solution of
the homogeneous equation ∇ ⃗ 2 G(⃗
r − r⃗′ ) = 0, which leads to the Dirichlet and Neu-
mann Green functions. Analogous considerations are true for the Green function of
electrodynamics, with −1/k⃗2 being replaced by (0 (k⃗ 2 − (ω/c)2 ))−1 [see Eq. (4.12)].
4.1.3 Retarded and Advanced Green Function

We recall once more that according to Eqs. (4.16) and (4.17), the Green function is
given as
d3 k ⃗ PC (t − t′ , ∣k∣)
⃗ , ⃗ = c ⃗ ′
r − r⃗′ , t − t′ ) = ∫
G(⃗ f (k) f (k) ei k⋅(⃗r−⃗r ) . (4.24)
(2π)3 ⃗
2∣k∣0
We recall the following characteristic frequency integral, defined in Eq. (4.17),
⃗ = ∫ dω e−i ω(t−t′ ) ( 1
PC (t − t′ , ∣k∣) −
1
). (4.25)
C 2π ⃗ ω − c ∣k∣
ω + c ∣k∣ ⃗
The contour C consists two elements, the first being equal to the real axis, with an
appropriate choice regarding the poles of the integrand, the second being the semi-
circle, with infinite radius, in either the upper half complex ω plane or in the lower
complex ω half plane. Unlike the way in which the pole terms are encircled, the
choice of the semi-circle actually is determined by the requirement for the integral
to converge to a finite value.
Together with the choice of the integration contour for the pole terms, this
requirement determines the behavior of the Green function for positive and negative
time difference t − t′ . Let us consider the term in round brackets in the integrand
−2
of Eq. (4.25), which causes the integrand to vanish, on either semi-circle, as ∣ω∣
for large ∣ω∣ provided the exponential in the numerator is well behaved. In order to
analyze the behavior on the semi-circle, we set ω = Re(ω) + i Im(ω). The absolute
value of the exponential in the integrand of Eq. (4.25) is determined by the relation
∣exp [−i ω (t − t′ )]∣ = exp[Im(ω) (t − t′ )] . (4.26)
If t − t′ > 0 and Im(ω) → +∞, then this expression diverges. However, if t − t′ > 0
and Im(ω) → −∞, then this expression vanishes exponentially. Thus, for t > t′ , the
path used in evaluating (4.17) must be closed in the lower half of the complex ω
plane. Conversely, for t < t′ , the path must be closed along a semi-circle in the
upper half of the complex ω plane. The first path gives a nonvanishing contribution
for the retarded Green function (contour CR , see Fig. 4.1), whereas the second
yields a nonvanishing contribution for the advanced Green function (contour CA ).
For completeness, we should point out that two obvious further possibilities for
distorting the path exist which encircle the poles on opposite sides of the real axis;
these are not shown in Fig. 4.1.
Both integration paths outlined in Fig. 4.1 have their justification, and both
results for the Green function are valid solutions of the defining equation (4.9), i.e.,
they lead to solutions of the inhomogeneous differential equation
1 ∂2 1 (3)
( ⃗ 2 ) G (⃗
−∇ r − r⃗′ , t − t′ ) = r − r⃗′ ) δ (t − t′ ) .
δ (⃗ (4.27)
c2 ∂t2 0
The different solutions fulfill different boundary conditions, which is natural for a
second-order partial differential equation.
⃗ along the different contours differ in
The results obtained for PC (t − t′ , ∣k∣)
the boundary conditions fulfilled by the corresponding Green functions. For the
Dirichlet Green function in electrostatics, the boundary conditions are given along
two-dimensional submanifolds (surfaces) of three-dimensional space. The Dirichlet
Green function is required to vanish whenever one of the two arguments r⃗ or r⃗′ is on
the defining surface. For the time-dependent Green function, boundary conditions
are defined on submanifolds of four-dimensional space-time. Intuitively, one would
associate boundary conditions with limiting cases of the variables, e.g., in the infi-
nite past or the infinite future. In all cases considered, we have so far defined our
Green function to vanish in the infinite past or future, or, for large distances.
Theoretically, we could have added to any of our Green functions a solution
to the homogeneous equation to the Green function (e.g., a constant term) that
changes the behavior as t, t′ → ±∞ or ∣⃗ r′ ∣ → ∞. However, we choose not to do so.

r ∣, ∣⃗
Or, we could have added a solution of the homogeneous equation, proportional to a
plane and uniform wave cos(k⃗ ⋅ r⃗−ωt). This wave does not vanish for ∣t−t′ ∣ → ∞, and
r − r⃗′ ∣ → ∞. Indeed, the difference of the retarded and advanced Green functions is
∣⃗
just a solution of the homogeneous equation.
Let us now evaluate both the retarded as well as the advanced Green function,
explicitly. We consider the ω or angular frequency part of the integrand
⃗ = ∮ dω −i ω(t−t′ ) 1 1
PC (t − t′ , ∣k∣) e ( − ), (4.28)
2π ⃗ ω − c ∣k∣
ω + c ∣k∣ ⃗
closed
where C is the path of integration. If the path CR for the retarded Green function
in Fig. 4.1 is used and t < t′ , then we must close the integration path in the upper
half of the complex plane, and the result for PC is zero. However, if CR is used
and t > t′ , then the poles are encircled in the clockwise (mathematically negative)
direction, and we obtain
⎡ −i ω(t−t′ ) ⎞ ′ ⎤
⎢ ⎛ e−i ω(t−t ) ⎞⎥
⃗ = − 2πi Θ (t − t′ )
PCR (t − t′ , ∣k∣) ⎢ Res ⎛ e − ⎥
⎢ω=−ck ⎝ Res
⃗ ⎠ ω=+ck ⎝ ω − c ∣k∣ ⃗ ⎠⎥
2π ⎢ ω + c ∣k∣ ⎥
⎣ ⎦
′ ′
= − i Θ (t − t′ ) (e−i(−ck)(t−t ) − e−i(+ck)(t−t ) )
′ ′
= − i Θ (t − t′ ) (eick(t−t ) − e−ick(t−t ) )
= 2 Θ (t − t′ ) sin(c k (t − t′ )). (4.29)
⃗ is nonvanishing only for t > t′ ; this is manifest in the

The function PCR (t − t′ , ∣k∣)
step function. By looking at the integration contour, one might superficially assume
that the pole of the integrand is only half-encircled. However, the continuation of
the integration contour into the lower complex plane actually makes it a full pole.
The factor 1/(2π) of the ω integration thus is compensated by the factor 2π from
the residue theorem.
On account of Eqs. (4.16) and (4.29), we obtain the retarded Green function
GR ,
d3 k ⃗ PC (t − t′ , ∣k∣)
⃗
r − r⃗′ , t − t′ ) = ∫
GR (⃗ f (k)
(2π)3 R
⃗ ′
c d3 k ei k⋅(⃗r−⃗r )
= Θ(t − t′ ) ∫ sin(ck (t − t′ )) , (4.30)
0 (2π)3 ⃗
∣k∣
⃗ given in Eq. (4.16). Now we do the k inte-

where we recall the definition of f (k)
⃗
gration in spherical k space, assuming (without loss of generality) that r⃗ − r⃗′ points
along the êz direction [uk = cos(θk )]:

∞ 2π 1
c r − r⃗′ ∣ uk )
exp (i k∣⃗
GR = Θ(t − t′ ) ∫ dk k 2
∫ dϕ ∫ du sin(c k (t − t′ ))
0 (8π 3 )
k k
⃗
∣k∣
0 0 −1
c (2π) ∞ 1 r − r⃗′ ∣ uk )
exp (i k∣⃗
= Θ(t − t′ ) ∫ dk k 2 ∫ duk sin(c k (t − t′ ))
0 (8π ) 0
3 −1 ⃗
∣k∣
c ∞ r − r⃗′ ∣) − exp (− i k∣⃗
2 exp (i k ∣⃗ r − r⃗′ ∣)
= Θ(t − t′ ) ∫ dk k sin(ck (t − t′ ))
4π 2 0 0 r − r⃗′ ∣
i k 2 ∣⃗
c ∞ ′
= Θ(t − t′ ) ∫ dk [ei k ∣⃗r−⃗r ∣ sin(ck (t − t′ ))
r − r⃗′ ∣ 0
4π 2 0 i ∣⃗
′
+ei (−k) ∣⃗r−⃗r ∣ sin(c(−k) (t − t′ ))] . (4.31)
In view of the symmetry under k ↔ −k, we can extend the integration domain to
the entire real axis,
c ∞
GR = Θ(t − t′ ) ′ ∫ r − r⃗′ ∣) sin(ck(t − t′ ))
dk exp (i k ∣⃗
4π 2 0 i ∣⃗
r − r⃗ ∣ −∞
c ∞ 1 ′ ′
= Θ(t − t′ ) ′ ∫ r − r⃗′ ∣) [eick (t−t ) − e−ick (t−t ) ]
dk ( ) exp (i k ∣⃗
4π 2 0 i ∣⃗
r − r⃗ ∣ −∞ 2i
c ∞ ′ ′ ′
= Θ(t − t′ ) (− ′
)∫ dk ei k ∣⃗r−⃗r ∣ [eick (t−t ) − e−ick (t−t ) ]
8π 2 0 ∣⃗
r − r⃗ ∣ −∞
c ∞ ′ ′ ′ ′
= Θ(t − t′ ) (− ′
)∫ dk [ei k (∣⃗r−⃗r ∣+c(t−t )) − ei k (∣⃗r−⃗r ∣−c(t−t )) ] .
8π 2 0 ∣⃗
r − r⃗ ∣ −∞
(4.32)
The k integration is now trivial in view of the formula
∞
∫ dk eikx = 2π δ(x) . (4.33)
−∞
The retarded Green function thus is given as

⎧
⎪ ⎫
⎪
c 1 ′ ⎪ ⎪
r − r⃗′ , t − t′ ) = −
GR (⃗ Θ(t−t ) ⎨ δ(∣⃗
r −⃗
r ′
∣ + c (t−t ′
))−δ(∣⃗
r −⃗
r ′
∣ − c (t−t ′
)) ⎬,
4π0 ∣⃗ ′
r − r⃗ ∣ ⎪
⎪ ⎪
⎪
⎩ ⎭
(4.34)
which can be simplified to read
⎧
⎪ ⎫
⎪
c 1 ′ ⎪ ⎪
r − r⃗′ , t − t′ ) =
GR (⃗ Θ(t−t ) ⎨δ(∣⃗
r −⃗
r ′
∣ − c (t−t ′
))−δ(∣⃗ r −⃗r ′
∣ + c (t−t ′
))⎬.
4π0 ∣⃗ ′
r − r⃗ ∣ ⎪
⎪ ⎪
⎪
⎩ ⎭
(4.35)
One might otherwise argue that it should be possible to discard the last term in
curly brackets in Eq. (4.35), because the expression ∣⃗ r − r⃗′ ∣ + c (t − t′ ) cannot vanish
if t − t′ > 0, which in turn is necessary for the step function to be nonvanishing. One
might thus think that the Green function should be expressible as
c Θ(t − t′ )
r − r⃗′ , t − t′ ) ≈
GR (⃗ r − r⃗′ ∣ − c (t − t′ )) .
δ (∣⃗ (4.36)
r − r⃗′ ∣
4π 0 ∣⃗
In many cases, it is indeed possible to discard the second term in curly brackets in
Eq. (4.35), as done in Ref. [21]. However, as argued in Ref. [22], the approxima-
tion (4.36) can only be used if no partial integrations are to be carried out, i.e., if
we have manifestly t − t′ > 0 for all contributing source terms in a calculation. The
deeper reason is the following: In any representation of the Dirac-δ distribution,
e.g., formulated in terms of normalized Gaussians whose width tends to zero, the
step function will assume values of unity within the width of the representation
of the Dirac-δ function when ∣⃗ r − r⃗′ ∣ → 0+ and c (t − t′ ) → 0+ . In other words, for
any representation of the Dirac-δ distribution in terms of Gaussians with a small
width, the Dirac-δ function δ(∣⃗ r − r⃗′ ∣+c (t−t′ )) will start to “ramp up” already when
r − r⃗ ∣ + c (t − t ) assumes slightly positive, nonvanishing values (t − t′ > 0), which lie
∣⃗ ′ ′
in the domain where Θ(t − t′ ) > 0. One thus cannot discard any term in Eq. (4.35)
if one would like to carry out partial integrations (with respect to time or space)
correctly. Such partial integrations are useful in the investigation of the coupling of
the sources to the potentials and fields in classical electrodynamics [22].
It is sometimes useful to note that GR can be written as
c 1
r − r⃗′ , t − t′ ) =
GR (⃗ Θ(t − t′ ) δ(∣⃗
r − r⃗′ ∣ − c (t − t′ ))
r − r⃗′ ∣
4π0 ∣⃗
(−c) 1
+ Θ(t − t′ ) δ(∣⃗
r − r⃗′ ∣ − (−c) (t − t′ )) , (4.37)
r − r⃗′ ∣
4π0 ∣⃗
i.e., as the sum of two terms, the second of which is derived from the first by a
change in the sign of the speed of light. Likewise, the approximation
c
r − r⃗′ , t − t′ ) ≈
GR (⃗ Θ(t − t′ ) δ((⃗
r − r⃗′ )2 − c2 (t − t′ )2 ) (4.38)
2π0
is valid provided we can discard the second term in curly brackets in Eq. (4.35).
Indeed, if the only contributing zero of the argument of the Dirac-δ is the one at
r − r⃗′ ∣ − c (t − t′ ) = 0 ,
∣⃗ t − t′ > 0 , (4.39)
then in view of Eq. (1.34),
r − r⃗′ ∣ − c (t − t′ ))
δ(∣⃗
′ 2 ′ 2
r − r⃗ ) − c (t − t ) ) =
δ((⃗ 2
, t − t′ > 0 . (4.40)
r − r⃗′ ∣
2∣⃗
Under these assumptions, the approximations (4.38) and (4.36) remain valid.
The advanced Green function obtained using the path CA (see Fig. 4.1) is similar
to the retarded Green function but is nonvanishing only for t < t′ . Note that one
must repeat all the steps leading to the expression (4.35) for the retarded Green
Im ω
Contour CF
Re ω
Fig. 4.2 The Feynman contour CF corresponds to the time-ordered

product of photon field operators in field theory.
function but use the contour CA , rather than CR . The net result of this calculation
is that the advanced Green function is obtained from Eq. (4.34) by the substitution
t − t′ → t′ − t,
⎧
⎪ ⎫
⎪
c 1 ⎪ ⎪
GA (⃗ r − r⃗′ , t − t′ ) = Θ(t ′
−t) ⎨ δ(∣⃗
r −⃗
r ′
∣ + c (t−t ′
))−δ(∣⃗r −⃗ r ′
∣ − c (t−t ′
))⎬.
4π0 ∣⃗ ′
r − r⃗ ∣ ⎪
⎪ ⎪
⎪
⎩ ⎭
(4.41)
This can be rewritten as
⎧
⎪ ⎫
⎪
c 1 ⎪ ⎪
GA (⃗ r − r⃗′ , t − t′ ) = Θ(t ′
−t) ⎨ δ(∣⃗
r −⃗
r ′
∣ − c (t ′
−t))−δ(∣⃗ r −⃗ r ′
∣ + c (t ′
−t)) ⎬.
4π0 ∣⃗ ′
r − r⃗ ∣ ⎪
⎪ ⎪
⎪
⎩ ⎭
(4.42)
Under appropriate assumptions (manifestly t − t′ < 0, and no partial integrations)
we can approximate the advanced Green function as
c 1
GA (⃗ r − r⃗′ , t − t′ ) ≈ Θ (t′ − t) δ (∣⃗ r − r⃗′ ∣ + c (t − t′ )) , (4.43)
4π0 ∣⃗ r − r⃗′ ∣
or as
c
GA (⃗ r − r⃗′ , t − t′ ) ≈ Θ(t′ − t) δ((⃗ r − r⃗′ )2 − c2 (t − t′ )2 ) , (4.44)
2π0
where the latter form is valid if we neglect the overlap region of the Θ function
with the chosen representation of the Dirac-δ; this approximation is generally valid
unless partial integrations have to be taken [22].
4.1.4 Feynman Contour Green Function

We recall the definition of the PC function, which for the Feynman contour as given
in Fig. 4.2 reads
⃗ = ∮ dω e−i ω(t−t′ ) ( 1
PCF (t − t′ , ∣k∣) −
1
). (4.45)
2π ⃗ ⃗
ω + c ∣k∣ ω − c ∣k∣
C F
For t−t′ > 0, the Feynman integration contour has to be closed in the mathematically
⃗ is relevant. For t − t′ < 0, the contour is
negative sense, and the pole at ω = c ∣k∣
closed in the upper complex half plane, and the pole at ω = −c ∣k∣ ⃗ is relevant. It
is encircled in the mathematically positive sense. Alternatively, we could include
infinitesimal imaginary contributions into the denominators,
∞
′ ⃗ = ∫ dω e−i ω(t−t′ ) (
PCF (t − t , ∣k∣)
1
−
1
). (4.46)
2π ⃗ ⃗ + i
ω + c ∣k∣ − i ω − c ∣k∣
−∞
So,
′
−i ω(t−t ) ⎞
⃗ = − (−2πi) Θ (t − t′ ) Res ⎛ e
PCF (t − t′ , ∣k∣)
2π ⃗ ⎠
ω=+ck ⎝ ω − c ∣k∣
′
2πi ⎛ e−i ω(t−t ) ⎞
+ Θ (t′ − t) Res
2π ⃗ ⎠
ω=−ck ⎝ ω + c ∣k∣
′ ′
= i Θ (t − t′ ) e−ick(t−t ) + i Θ (t′ − t) eick(t−t ) . (4.47)
The second term equals the first under the replacement t − t′ → t′ − t. The Feynman
Green function GF is thus obtained as
⃗ ′
d3 k c ei k⋅(⃗r−⃗r ) ′ ′
r − r⃗′ , t − t′ ) = ∫
GF (⃗ (iΘ (t − t′ ) e−ick(t−t ) + iΘ (t′ − t) eick(t−t ) ) .
(2π) 3
20 ∣k∣⃗
(4.48)
Now we do the k integration in spherical k⃗ space, assuming, without loss of general-
ity, that r⃗ − r⃗′ is aligned along the êz direction [uk = cos(θk )]. The Feynman Green
function GF consists of two terms, the first of which reads
(1) c ∞ 2π 1 r − r⃗′ ∣ uk ) −ick(t−t′ )

exp (i k∣⃗
GF = i Θ(t − t′ ) ∫ dk k 2
∫ dϕ ∫ du e
2 0 (8π 3 ) 0 0
k
−1
k
⃗
∣k∣
c (2π) ∞ 1 r − r⃗′ ∣ uk ) −ick(t−t′ )
exp (i k∣⃗
= i Θ(t − t′ ) ∫ dk k 2
∫ du e
2 0 (8π 3 ) 0 −1
k
⃗
∣k∣
c ∞ r − r⃗′ ∣) − exp (− i k∣⃗
2 exp (i k ∣⃗ r − r⃗′ ∣) −ick(t−t′ )
= i Θ(t − t′ ) ∫ dk k e
8π 2 0 0 r − r⃗′ ∣
i k 2 ∣⃗
c ∞ ′ ′ ′ ′
= i Θ(t − t′ ) ∫ dk [ei k ∣⃗r−⃗r ∣−i ck (t−t ) − e−i k ∣⃗r−⃗r ∣−ick (t−t ) ]
r − r⃗′ ∣ 0
8π 2 0 i ∣⃗
c ∞
i k (∣⃗ r ′ ∣−c(t−t′ )+i ) ′ ′
= i Θ(t − t′ ) ′ ∫ dk [e
r−⃗
− e−i k (∣⃗r−⃗r ∣+c (t−t )−i )] .
0 i ∣⃗
r − ⃗
8π 2
r ∣ 0
(4.49)
In the last step, we have introduced infinitesimal convergent factors ±i, which
enables us to carry out the k integration. The limit → 0, after performing the
integrations, is understood here and in the following unless stated otherwise. We
finally obtain
(1) c 1
GF = i Θ(t − t′ ) (
r − r⃗′ ∣ −i (∣⃗
8π 2 0 i ∣⃗ r − r⃗′ ∣ − c(t − t′ ) + i )
1
− )
i r − r⃗′ ∣ + c(t − t′ ) − i )
(∣⃗
c 1 1
= i Θ(t − t′ ) ( + )
r − r⃗′ ∣ ∣⃗
8π 2 0 ∣⃗ r − r⃗′ ∣ − c(t − t′ ) + i ∣⃗
r − r⃗′ ∣ + c(t − t′ ) − i
c 1
= i Θ(t − t′ )
r − r⃗′ )2 − c2 (t − t′ )2 + i η
4π 2 0 (⃗
c 1
= − i Θ(t − t′ ) . (4.50)
4π 2 0 c2 (t − t′ )2 − (⃗
r − r⃗′ )2 − i η
Here η = 2 c (t−t′ ) > 0 is an infinitesimal quantity. The second term in the Feynman
Green function GF is analogously obtained as
(2) c ∞ 2π 1 r − r⃗′ ∣ uk ) ick(t−t′ )

exp (i k∣⃗
GF = i Θ (t′ − t) ∫ dk k 2
∫ dϕ ∫ du e
2 0 (8π 3 ) 0 0
k
−1
k
⃗
∣k∣
c ∞ r − r⃗′ ∣) − exp (− i k ∣⃗
exp (i k ∣⃗ r − r⃗′ ∣) ick(t−t′ )
= i Θ (t′ − t) ∫ dk k 2 ′
e
2
8π 0 0 i k ∣⃗
2 r − r⃗ ∣
c ∞ ′ ′ ′ ′
= i Θ (t′ − t) ′ ∫ dk [ei k (∣⃗r−⃗r ∣+c(t−t )+i ) − e−i k (∣⃗r−⃗r ∣−c (t−t )−i ) ]
8π 2 0 i ∣⃗
r − r⃗ ∣ 0
c 1 1
= i Θ (t′ − t) ( + )
r − r⃗′ ∣ ∣⃗
8π 2 0 ∣⃗ r − r⃗′ ∣ + c(t − t′ ) + i ∣⃗
r − r⃗′ ∣ − c(t − t′ ) − i
c 1
= − i Θ (t′ − t) , (4.51)
4π 2 0 c2 (t − t′ )2 − (⃗
r − r⃗′ )2 − i η ′
where η ′ = 2 c (t′ − t) > 0 in this case, because of the prefactor Θ (t′ − t). Adding
(1) (2)
the two contributions for GF = GF + GF and interpolating trivially for t = t′ , we
obtain the following result for the Feynman propagator,
c 1
r − r⃗′ , t − t′ ) = −i
GF (⃗ ′
. (4.52)
2 2 2 r − r⃗′ )2 − i
4π 0 c (t − t ) − (⃗
Here, we have substituted η → for the infinitesimal quantity in the denominator.

Two important alternative forms of this propagator are as follows. First, because
of the prescription
1 1
= (P.V.) + i π δ(g) , (4.53)
g − i g
we have
c 1
r − r⃗′ , t − t′ ) = − i
GF (⃗ (P.V.) 2
4π 2 0 c (t − t′ )2 − (⃗ r − r⃗′ )2
c
+ δ(c2 (t − t′ )2 − (⃗
r − r⃗′ )2 ) . (4.54)
4π0
Here, (P.V.) denotes the principal value. Another representation is
c 1 1
r − r⃗′ , t − t′ ) = −i
GF (⃗ ( − ).
r − r⃗′ ∣ c (t − t′ ) − ∣⃗
8π 2 0 ∣⃗ r − r⃗′ ∣ − i c (t − t′ ) + ∣⃗
r − r⃗′ ∣ + i
(4.55)
The poles are at
c (t − t′ ) = ∣⃗
r − r⃗′ ∣ + i , c (t − t′ ) = −∣⃗
r − r⃗′ ∣ − i . (4.56)
We consider the Fourier transformation with respect to τ = t − t′ of the Feynman

Green function for real rather than complex ω,
∞ 1 1 1
r − r⃗′ , ω) = − i ∫
GF (⃗ dτ eiωτ ′
( ′
− )
−∞ 0 ∣⃗
r − r⃗ ∣ τ − ∣⃗
8π 2 r − r⃗ ∣/c − i τ + ∣⃗r − r⃗′ ∣/c + i
⎧
⎪ ′ 1
⎪
⎪
⎪ −i 2π i eiω∣⃗r−⃗r ∣/c 2 (ω > 0) ′
⎪
⎪ 8π 0 ∣⃗ r − r⃗′ ∣ ei ω sgn(ω)∣⃗r−⃗r ∣/c
=⎨ = .
⎪
⎪ 1 4π0 ∣⃗r − r⃗′ ∣
⎪
⎪
⎪ −(−i) (−2π i) e −iω∣⃗r −⃗r ′ ∣/c
(ω < 0)
⎪
⎩ r − r⃗′ ∣
8π 2 0 ∣⃗
(4.57)
Here, sgn(x) is the sign function, which equals +1 for x > 0 and −1 for x < 0. For
complex ω, the Fourier integral would diverge either at τ = −∞ or at τ = +∞ and
could not be directly calculated. The result (4.57) can be understood as follows:
There are two poles in the τ integration, namely, at
r − r⃗′ ∣
∣⃗ r − r⃗′ ∣
∣⃗
τ= + i , τ =− − i . (4.58)
c c
If ω > 0, then the τ integral needs to be closed in the upper half plane, i.e., the pole
r − r⃗′ ∣/c + i contributes. Conversely, if ω < 0, then the τ integral needs to
at τ = ∣⃗
be closed in the lower half plane, i.e., the pole at τ = −∣⃗r − r⃗′ ∣/c − i contributes. For
real ω, we have in view of ω sgn(ω) = ∣ω∣,
′
′ ei ∣ω∣ ∣⃗r−⃗r ∣/c
GF (⃗
r − r⃗ , ω) = . (4.59)
4π0 ∣⃗ r − r⃗′ ∣
The question now is how to analytically continue the modulus function for mani-
festly complex ω. First of all, for real ω, we can write
√
ω sgn(ω) = ∣ω∣ = ω 2 , (4.60)
and this relation gives us a hint about how the analytic continuation needs to be
done. The boundary condition in frequency space translates into the condition that
the contribution of frequencies of very large complex modulus to the Feynman prop-
agator needs to vanish. This boundary condition can be implemented as follows.
We first supply an infinitesimal imaginary part under the square root,
√ √
ω2 → ω2 + i . (4.61)
√
This is compatible with the requirement that ω 2 = ∣ω∣ for real ω, no matter whether
we lay the branch cut of the square root along the positive or the negative real axis.
Then,
√ r − r⃗′ ∣
∣⃗ r − r⃗′ ∣
∣⃗
exp (i ω 2 + i ) exp (−b )
c c
r − r⃗′ , ω) =
GF (⃗ = , (4.62)
r − r⃗′ ∣
4π0 ∣⃗ r − r⃗′ ∣
4π0 ∣⃗
√ ! √ !
b = −i ω 2 + i , Re(b) ≥ 0 , Im ω 2 + i ≥ 0 . (4.63)
This condition is fulfilled if we choose the branch cut of the square root function
along the positive real axis. The argument of the square root function vanishes for
√
ω = ± exp (− iπ4
). If we define the branch cut of the square root function to be
along the positive real axis, then branch cuts, as a function of ω, extend to ω → ± ∞,
from the branch points, and the latter possibility is the one that allows us to use
the Feynman contour for the ω integration in the usual manner.
√
Having specified the infinitesimal imaginary part as ω 2 + i, and defining the
branch cut to be along the positive real axis, we may deform the Feynman contour
to the entire real axis, and we have the Fourier backtransformation,
√ r − r⃗′ ∣
∣⃗
∞
exp (i ω 2 + i − iωτ )
dω c
r − r⃗′ , τ ) = ∫
GF (⃗
−∞ 2π r − r⃗′ ∣
4π0 ∣⃗
∣r−
⃗ r⃗′ ∣ ∣r−
⃗ r⃗′ ∣
∞ c (i −i τ −η) ω 0 c (−i −i τ +η) ω
dω e dω e
=∫ +∫
r − r⃗′ ∣
2π 4π0 ∣⃗ r − r⃗′ ∣
2π 4π0 ∣⃗
0 −∞
⎛ ⎞
1 ⎜ 1 1 ⎟
= 2 ⎜− − ⎟
r − r⃗′ ∣ ⎜ ∣⃗
8π 0 ∣⃗ r − r⃗′ ∣ r − r⃗′ ∣
∣⃗ ⎟
⎝ i − i τ − η i + i τ − η ⎠
c c
1 ic ic
= ′
(− ′
− ′
)
8π 2 0 ∣⃗
r − r⃗ ∣ −∣⃗
r − r⃗ ∣ + c τ − i η −∣⃗
r − r⃗ ∣ − c τ − i η
c 1
= −i , (4.64)
2 2 2 r − r⃗′ )2 − i η
4π 0 c τ − (⃗
where η in the second line is a convergent factor. The limit η → 0 is again un-
derstood in intermediate steps after all integrations have been carried out; η is
an auxiliary infinitesimal parameter

√ which we could have labeled as . A priori, one
has in the limit → 0, simply ω 2 + i → ∣ω∣, but one can justify the occurrence of
the convergent factor based on the expansion
√ i
ω 2 + i = ∣ω∣ + + O(2 ) , (4.65)
2∣ω∣
which leads to an infinitesimal exponential suppression factor, in accordance with
Eq. (4.63).
Alternatively, for τ > 0, we can evaluate the Fourier backtransformation by eval-
uating the cut of the photon propagator, along the positive real axis, by bending the
Feynman contour into a segment Cup , infinitesimally above the real axis, extending
from zero to ω = ∞×exp(i), and a segment Cdown , extending from ω = ∞×exp(−i),
to the origin of the ω plane,
∣r−
⃗ r⃗′ ∣ ∣r−
⃗ r⃗′ ∣
′ dω ei ω c −iωτ − ω dω e−i ω c −iωτ − ω
GF (⃗
r − r⃗ , τ > 0) = ∫ − ∫
2π 4π0 ∣⃗r − r⃗′ ∣ 2π r − r⃗′ ∣
4π0 ∣⃗
Cup Cdown
∣r−
⃗ r⃗′ ∣ ∣r−
⃗ r⃗′ ∣
(i −i τ −) ω (−i −i τ −) ω
∞ dω e c ∞ dω e c
=∫ −∫
0 r − r⃗′ ∣
2π 4π0 ∣⃗ 0 r − r⃗′ ∣
2π 4π0 ∣⃗
1 ic ic
= (− + )
8π 2 0 ∣⃗
r − r⃗′ ∣ −∣⃗
r − r⃗′ ∣ + c τ − i ∣⃗
r − r⃗′ ∣ + c τ − i
c 1
= −i ′
4π 0 c τ − (⃗
2 2 2 r − r⃗ )2 − i sgn(τ )
c 1
= −i , (4.66)
2 2 2 r − r⃗′ )2 − i
4π 0 c τ − (⃗
where the last transformation holds because we have assumed τ > 0 in the first
place.
Let us also carry out a full Fourier transformation with respect to space and
time. With the definition of the Feynman propagator, we have
√ r − r⃗′ ∣
∣⃗
exp (i ω 2 + i )
c
r − r⃗′ , ω) =
GF (⃗ (4.67)
r − r⃗′ ∣
4π0 ∣⃗
and with ρ⃗ = r⃗ − r⃗′ ,
⃗ ′
⃗ ω) = ∫ d3 (⃗
GF (k, r − r⃗′ ) GF (⃗
r − r⃗′ , ω) e−i k⋅(⃗r−⃗r )
⃗
ρ, ω) e−i k⋅ρ⃗ ,
= ∫ d3 ρ GF (⃗ (4.68)
with two obvious variable substitutions. Assuming, without loss of generality, that
the k⃗ vector is parallel to the z axis, we can perform the calculation as follows,
⃗ ω) = ∫ d3 r GF (⃗ ⃗
GF (k, r , ω) e−i k⋅⃗r
√ r
∞ 1 ω 2 + i )
exp (i
= 2π ∫ dr r2 ∫ du c e−ik r u
0 −1 4π0 r
1 ∞ √ r sin(k r) 1 1
= ∫ dr exp (i ω 2 + i ) = . (4.69)
0 0 c k 0 k⃗ − ω /c2 − i
2 2
The poles of the Fourier transform of the Feynman Green function are thus seen to
⃗ − i and at ω = −c∣k∣
be at ω = c∣k∣ ⃗ + i, consistent with Fig. 4.2.
4.1.5 Summary: Green Functions of Electrodynamics

A summary of relevant formulas for the retarded, advanced, and Feynman Green
functions appears to be in order. All of these Green functions fulfill Eq. (4.9), which
reads
1 ∂2 1 (3)
( ⃗ 2 ) G(⃗
−∇ r − r⃗′ , t − t′ ) = r − r⃗′ ) δ (t − t′ ) .
δ (⃗ (4.70)
2
c ∂t 2 0
According to Eq. (4.42), the advanced Green function GA reads
⎧
⎪ ⎫
⎪
′ ′ c 1 ′ ⎪ ′ ′ ′ ′ ⎪
GA (⃗r − r⃗ , t − t ) = Θ(t −t) ⎨ δ(∣⃗
r −⃗
r ∣ − c (t −t))−δ(∣⃗
r −⃗
r ∣ + c (t −t)) ⎬.
r − r⃗′ ∣
4π0 ∣⃗ ⎪
⎪ ⎪
⎪
⎩ ⎭
(4.71)
Its Fourier transform with respect to time is
′
′ e−i ω ∣⃗r−⃗r ∣/c
GA (⃗
r − r⃗ , ω) = , (4.72)
4π0 ∣⃗r − r⃗′ ∣
and the full Fourier transform into frequency-wave-number space is
⃗ ω) = c 1 1 1 1
GA (k, ( − )= . (4.73)
⃗ 0
2∣k∣ ⃗ − i
ω + c∣k∣ ⃗ − i
ω − c∣k∣ 0 k⃗2 − ω 2 /c2 − i sgn(ω)
This result is immediately obvious from Fig. 4.1. The retarded Green Function GR
can be found in Eq. (4.35),
⎧
⎪ ⎫
⎪
c 1 ′ ⎪ ⎪
r − r⃗′ , t − t′ ) =
GR (⃗ Θ(t−t ) ⎨δ(∣⃗
r −⃗
r ′
∣ − c (t−t ′
))−δ(∣⃗
r −⃗
r ′
∣ + c (t−t ′
))⎬.
4π0 ∣⃗ ′
r − r⃗ ∣ ⎪
⎪ ⎪
⎪
⎩ ⎭
(4.74)
Its Fourier transform is
′
′ ei ω ∣⃗r−⃗r ∣/c
GR (⃗
r − r⃗ , ω) = . (4.75)
4π0 ∣⃗ r − r⃗′ ∣
The full Fourier transform into frequency-wave-number space is
⃗ ω) = c 1 1 1 1
GR (k, ( − )= . (4.76)
⃗ ⃗ ⃗
2∣k∣ 0 ω + c∣k∣ + i ω − c∣k∣ + i ⃗
0 k − ω /c + i sgn(ω)
2 2 2
The Feynman propagator GF [see Eq. (4.52)] reads

c 1
r − r⃗′ , t − t′ ) = −i
GF (⃗ , (4.77)
4π 2 0 c2 (t − t′ )2 − (⃗
r − r⃗′ )2 − i
and its Fourier transform is
√ r − r⃗′ ∣
∣⃗
exp (i ω 2 + i )
c √
r − r⃗′ , ω) =
GF (⃗ , Im ω 2 + i > 0 . (4.78)
r − r⃗′ ∣
4π0 ∣⃗
The latter condition clarifies how the branch cuts of the square root should be
defined. The full Fourier transform of the Feynman Green function is
⃗ ω) = c 1 1 1 1
GF (k, ( − )= . (4.79)
⃗ ⃗ ⃗
2∣k∣ 0 ω + c∣k∣ − i ω − c∣k∣ + i ⃗
0 k − ω /c2 − i
2 2
The Feynman Green function is important in field-theoretical calculations [20, 23].
4.2 Action-at-a-Distance and Coulomb Gauge
4.2.1 Potentials and Sources

It is generally acknowledged that some mystery surrounds the so-called action-at-a-
distance solution for the scalar potential which can be obtained in the Coulomb, or
radiation, gauge. Due to the gauge condition in radiation gauge, the divergence of
the vector potential vanishes, or, expressed differently, the vector potential is equal
to its own transverse component [see Eq. (1.87)],
⃗ ⋅ A⃗C (⃗
∇ r , t) = 0 , A⃗C (⃗
r, t) = A⃗C⊥ (⃗
r, t) . (4.80)
In the following, we distinguish the scalar and vector potentials by the subscripts
C for Coulomb gauge, and L for Lorenz gauge. In Coulomb gauge, the coupling to
the sources is governed by the following equations [see Eq. (1.92)],
1
⃗ 2 ΦC (⃗
∇ r , t) = − ρ (⃗
r , t) , (4.81a)
0
1 ∂2
( −∇ r , t) = μ0 J⃗⊥ (⃗
⃗ 2 ) A⃗C⊥ (⃗ r , t) , (4.81b)
c2 ∂t2
∂ ⃗
0 r , t) = J⃗∥ (⃗
∇ΦC (⃗ r, t) . (4.81c)
∂t
We recall that the longitudinal and transverse components of a general vector field
are analyzed in Eqs. (1.48a) and (1.48b). Equation (4.81a) couples the electrostatic
potential to the source; it has the instantaneous, action-at-a-distance solution given

in Eq. (1.98), which we recall for convenience,
1 3 ′ 1
ΦC (⃗
r , t) = ∫ d r r′ , t) + Φhom (⃗
ρ(⃗ r, t) , (4.82)
4π0 r − r⃗′ ∣
∣⃗
where Φhom (⃗r, t) is a solution to the homogeneous equation. The instantaneous
coupling of the electrostatic potential to the charge density gives rise to a number
of concerns, which we have already discussed near the end of Chap. 1.2.5. Our task
is to show that the instantaneous character of the solution (1.98) does not lead to
a contradiction with respect to the causality principle; an electromagnetic signal
cannot travel faster than light and the fields (as opposed to the potentials) should
be manifestly retarded.
The Lorenz condition for the scalar potential Φ and the vector potential A⃗ has
been given in Eq. (1.70),
1 ∂
⃗ ⋅ A⃗L (⃗
∇ r , t) + 2 ΦL (⃗
r , t) = 0 . (4.83)
c ∂t
In Lorenz gauge, the scalar and vector potentials are coupled to the sources by
inhomogeneous wave equations [see Eq. (1.77)],
1 ∂2 1
( ⃗ 2 ) ΦL (⃗
−∇ r , t) = ρ (⃗
r , t) , (4.84a)
2
c ∂t 2 0
1 ∂2
( ⃗ 2 ) A⃗L (⃗
−∇ r , t) = μ0 J⃗ (⃗
r , t) . (4.84b)
c2 ∂t2
We here provide two perspectives on the problem. The first perspective (see
Sec. 4.2.2) relies on the fact that the solution of Eq. (4.81a) might otherwise indicate
that the scalar potential in Coulomb gauge is non-retarded. However, once charges
move, currents are generated in view of the continuity equation. The seemingly in-
stantaneous scalar potential in Coulomb gauge is connected to the longitudinal part
of the current density by Eq. (4.81c). It is instructive to observe that Eq. (4.81c)
cannot alone provide the solution of the problem; its divergence simply is the time
derivative of Eq. (4.81a). However, Eq. (4.81c) shows that the situation is not so
easy: We cannot argue that the seemingly instantaneous Coulomb interaction auto-
matically gives rise to instantaneous fields; there is the additional condition (4.81c)
which has to be fulfilled by the scalar potential. We shall see that the vector po-
tential, in the Coulomb gauge, receives an additional contribution as compared to
the Lorenz gauge. The additional term corresponds to the longitudinal part of the
current density which has to be subtracted in order to obtain Eq. (4.81b). With
the help of Eq. (4.81c), we are finally able to show that the supplementary term in
the vector potential, in the Coulomb gauge, cancels the instantaneous contribution
to the electric field and leads to manifestly retarded expressions.
Expressed differently, the additional constraint (4.81c) implies that the action-
at-a-distance solution (4.82) is not universally a valid solution; it does not
automatically fulfill Eq. (4.81c). Indeed, we shall see that the homogeneous term
in Eq. (4.82) plays a crucial role in showing causality, together with Eq. (4.81c).
The second perspective (see Sec. 4.2.3) addresses the fact that the instantaneous
interaction integral can alternatively be written as a retarded integral, but with a
different source term, in accordance with arguments presented in Ref. [24]. Finally,
in Sec. 4.2.4 (third perspective), we find an explicitly retarded expression for the
Coulomb-gauge scalar potential, which fulfills both Eqs. (4.81a) as well as (4.81c).
4.2.2 Cancellation of the Instantaneous Term
According to Eq. (1.59), the electric field (in Coulomb gauge) is given as
⃗ r, t) = −∇Φ ∂
E(⃗ r , t) − A⃗C⊥ (⃗
⃗ C (⃗ r, t) , (4.85)
∂t
and therefore
∂
E⃗∥ (⃗ ⃗ C (⃗
r , t) = −∇Φ r , t) , ⃗⊥ (⃗
E r, t) = − A⃗C⊥ (⃗
r , t) . (4.86)
∂t
These components fulfill, explicitly, ∇ ⃗ ×E⃗∥ (⃗

r, t) = − (∇ ⃗ × ∇)
⃗ ΦC (⃗ ⃗ and
r , t) = 0,
⃗ ⃗ ⃗
r , t) = − ∂t ∇ ⋅ AC⊥ (⃗
∇ ⋅ E⊥ (⃗ ∂ ⃗
r, t) = 0. Therefore, the full longitudinal component
of the electric field is given by E⃗∥ (⃗ ⃗ C (⃗
r , t) = −∇Φ r , t), and the transverse component
in Coulomb gauge is purely given by the time derivative of the transverse vector
potential, without any “admixture” from the scalar potential. In Coulomb gauge,
the longitudinal component of the electric field is calculated as
⃗∥ (⃗ 1 1
E ⃗ C (⃗
r, t) = −∇Φ r, t) = − ⃗ ∫ d3 r′
∇ ρ(⃗ ⃗ hom (⃗
r′ , t) − ∇Φ r , t) , (4.87)
4π0 r − r⃗′ ∣
∣⃗
where Φhom is the solution to the homogeneous equation, adjusted so that

Eq. (4.81c) is fulfilled.
The retarded Green function (4.35) enters the solutions of the Lorenz-gauge
couplings given in Eqs. (4.84a) and (4.84b),
r , t) = ∫ d3 r′ dt′ GR (⃗
ΦL (⃗ r − r⃗′ , t − t′ ) ρ (⃗
r ′ , t′ ) , (4.88a)
1
A⃗L (⃗ r − r⃗′ , t − t′ ) J⃗ (⃗
r , t) = 2 ∫ d3 r′ dt′ GR (⃗ r ′ , t′ ) . (4.88b)
c
As has been stressed in Sec. 1.2.4, the potentials are not uniquely defined even
within the family of potentials that fulfill the Coulomb gauge condition; a gauge
re-transformation within the Coulomb gauge is possible according to Eq. (1.93). A
permissible way to proceed is to use the retarded Green function to solve Eq. (4.81b),
and to find the vector potential A⃗C (⃗r , t) = A⃗C⊥ (⃗
r , t). The solution A⃗C (⃗
r, t) can be
written in terms of the Lorenz-gauge expression A⃗L (⃗

r , t) and the supplementary
term A⃗S (⃗
r , t). We have,
1
A⃗C (⃗ r − r⃗′ , t − t′ ) J⃗⊥ (⃗
r , t) = 2 ∫ d3 r′ ∫ dt′ GR (⃗ r′ , t′ ) = A⃗L (⃗
r , t) + A⃗S (⃗
r , t) ,
c
(4.89a)
1
A⃗L (⃗
r , t) = 2 ∫ d3 r′ ∫ dt′ GR (⃗ ⃗ r ′ , t′ ) ,
r − r⃗′ , t − t′ ) J(⃗ (4.89b)
c
1
A⃗S (⃗ r − r⃗′ , t − t′ ) J⃗∥ (⃗
r , t) = − 2 ∫ d3 r′ ∫ dt′ GR (⃗ r ′ , t′ ) , (4.89c)
c
where we have used the identity J⃗⊥ (⃗ ⃗ r′ , t′ ) − J⃗∥ (⃗

r′ , t′ ) = J(⃗ r′ , t′ ). The supplementary
⃗
term AS (⃗
r , t) is relevant to the Coulomb gauge and reads
1
A⃗S (⃗ r − r⃗′ , t − t′ ) J⃗∥ (⃗
r, t) = − 2 ∫ d3 r′ ∫ dt′ GR (⃗ r ′ , t′ ) . (4.90)
c
The time derivative of the supplementary term A⃗S (⃗

r , t) contributes a supplementary
⃗
term ES to the electric field,
⃗S (⃗ ∂ 1 ∂
E r , t) = − A⃗S (⃗ r − r⃗′ , t − t′ )] J⃗∥ (⃗
r , t) = 2 ∫ d3 r′ ∫ dt′ [ GR (⃗ r ′ , t′ )
∂t c ∂t
1 ∂
=− 3 ′ ′
r − r⃗′ , t − t′ )] J⃗∥ (⃗
∫ d r ∫ dt [ ′ GR (⃗ r ′ , t′ )
c2 ∂t
1 ∂
= 2
3 ′ ′
r − r⃗′ , t − t′ ) [ ′ J⃗∥ (⃗
∫ d r ∫ dt GR (⃗ r′ , t′ )] , (4.91)
c ∂t
where we have first transformed ∂/∂t → −∂/∂t′ and then used integration by parts
to move the derivative on the current. Using Eq. (4.81c), this can be rewritten in
terms of the potential as
1 ∂2
E⃗S (⃗
r, t) = 0 ∫ d3 r′ ∫ dt′ GR (⃗ ⃗′ [
r − r⃗′ , t − t′ ) ∇ r′ , t′ )]
ΦC (⃗
c2 ∂t′ 2
1 ∂2 ′ ′
⃗ ∫ d3 r′ ∫ dt′ GR (⃗
= 0 ∇ r , t, r⃗′ , t′ ) [( ⃗ 2) + ∇
−∇ ⃗ 2 ] ΦC (⃗
r ′ , t′ )
c2 ∂t′ 2
1 ∂2
⃗ ∫ d3 r′ ∫ dt′ [(
= 0 ∇ ⃗ 2 ) GR (⃗
−∇ r , t, r⃗′ , t′ )] ΦC (⃗
r ′ , t′ )
c2 ∂t2
′
⃗ ∫ d3 r′ ∫ dt′ GR (⃗
+ 0 ∇ ⃗ 2 ΦC (⃗
r , t, r⃗′ , t′ ) ∇ r ′ , t′ ) . (4.92)
For the first term, we use partial integration twice, and we also take advantage of the
symmetry properties of the retarded Green function. With Eqs. (4.81a) and (4.9),
we find
⃗S (⃗ 1
E ⃗ ∫ d3 r′ ∫ dt′ δ (3) (⃗
r , t) = 0 ∇ r − r⃗′ ) δ(t − t′ ) ΦC (⃗ r ′ , t′ )
0
1
⃗ ∫ d3 r′ ∫ dt′ GR (⃗
− 0 ∇ r , t, r⃗′ , t′ ) ρ (⃗
r ′ , t′ )
0
⃗ C (⃗
= ∇Φ ⃗ ∫ d3 r′ ∫ dt′ GR (⃗
r, t) − ∇ r, t, r⃗′ , t′ ) ρ (⃗
r ′ , t′ )
⃗ C (⃗
= ∇Φ ⃗ L (⃗
r, t) − ∇Φ r , t) . (4.93)
In Coulomb gauge, the supplementary term E ⃗S = −∂t A⃗S due to the time derivative
of the supplementary vector potential cancels the gradient of the Coulomb gauge
scalar potential and adds the Lorenz gauge gradient of the scalar potential. Com-
paring the formulas for the electric field in Coulomb and Lorenz gauge, the following
identity follows immediately,
⃗C (⃗ ∂ ∂ ⃗
E r , t) = − ∇Φ r , t) − A⃗C (⃗
⃗ C (⃗ r , t) = −∇Φ ⃗ C (⃗ r , t) − (AL (⃗r , t) + A⃗S (⃗
r , t))
∂t ∂t
∂
= − ∇Φ r , t) − A⃗L (⃗
⃗ C (⃗ r , t) + (∇Φ⃗ C (⃗ ⃗ L (⃗
r , t) − ∇Φ r, t))
∂t
∂
= − ∇Φ r , t) − A⃗L (⃗
⃗ L (⃗ r , t) = E ⃗L (⃗r , t) . (4.94)
∂t
We have temporarily denoted the “Coulomb gauge” electric field as E⃗C (⃗ r, t) and
the “Lorenz gauge” electric field as E ⃗L (⃗ r , t), even if both are actually equal due to
gauge invariance, as shown.
Let us briefly summarize: In Coulomb gauge, there is an additional term A⃗S in
the vector potential which is generated by the negative of the longitudinal compo-
nent of the current density. The (negative of the) time derivative of the supplemen-
tary term in the vector potential yields an additional contribution to the electric
field, in Coulomb gauge. The additional term in the electric field can be transformed
into two parts, the first of which cancels the seemingly instantaneous electric field
contribution in Coulomb gauge, obtained from the Coulomb-gauge electric poten-
tial, and the second yields the same result (the retarded one) as the gradient of the
electric potential ΦL in Lorenz gauge. In the end, the action-at-a-distance integral
cancels, and the gauge invariance of the electric field is shown [Eq. (4.94)]. The
overall conclusion is that in Coulomb gauge, in view of the condition (4.81c), the
homogeneous solution Φhom in Eq. (4.82) has to be chosen so that the contribution
of the action-at-a-distance term to the electric field cancels.
4.2.3 Longitudinal Electric Field as a Retarded Integral

We recall once more Eq. (4.87), which for the longitudinal component of the electric
field reads as follows,
⃗∥ (⃗ 1 1
E ⃗ C (⃗
r, t) = −∇Φ r, t) = − ⃗ ∫ d3 r′
∇ ρ(⃗ ⃗ hom (⃗
r′ , t) − ∇Φ r , t) . (4.95)
4π0 r − r⃗′ ∣
∣⃗
The first term in this expression has an action-at-a-distance form, which could in
principle lead to a contradiction with respect to the causality principle, were it not
for the additional constraint (4.81c), which implies the necessity of adding a suitable
solution of the homogeneous equation. We recall that the decomposition of the elec-
tric field into longitudinal and transverse components is unique; an instantaneous
character of the longitudinal component also would have disastrous consequences.
We should thus investigate if, taking into account Eq. (4.81c), the longitudinal
component of the electric field can alternatively be written as a manifestly retarded
integral, for which we guess the form [24]
⃗∥ (⃗ 1 ∂
E r , t) = − ∫ d3 r′ ∫ dt′ GR (⃗ r − r⃗′ , t − t′ ) ( 2 ′ J⃗∥ (⃗ ⃗ ′ ρ(⃗
r ′ , t′ ) + ∇ r′ , t′ )) . (4.96)
c ∂t
In this expression, we use Eq. (4.81c) in order to substitute for J⃗∥ and Eq. (4.81a)
in order to substitute for ρ(⃗ r′ , t′ ),
⃗∥ (⃗ 1 ∂ ∂
E r , t) = − ∫ d3 r′ ∫ dt′ GR (⃗ ⃗′
r − r⃗′ , t − t′ )[ 2 ′ (0 ∇ r′ , t′ ))
ΦC (⃗
c ∂t ∂t′
+∇ ⃗ ′2 ΦC (⃗
⃗ ′ (−0 ∇ r′ , t′ ))]
1 ∂2
= − ∫ d3 r′ ∫ dt′ GR (⃗
r , t, r⃗′ , t′ ) ( −∇ ⃗ ′ ΦC (⃗
⃗ ′2 ) 0 ∇ r ′ , t′ ) . (4.97)
c2 ∂t′2
A double partial integration and use of Eq. (4.9) leads to the relation
⃗∥ (⃗
E r, t) = − ∫ d3 r′ ∫ dt′ δ (3) (⃗ ⃗ ′ ΦC (⃗
r − r⃗′ ) δ(t − t′ ) ∇ ⃗ C (⃗
r′ , t) = −∇Φ r , t) , (4.98)
which was to be shown.

In summary, we have demonstrated that the instantaneous integral for the longi-
tudinal part of the electric field (4.87) can be rewritten as an integral involving the
manifestly retarded Green function, with a nonstandard source term that does not
only involve the charge density but also the longitudinal part of the current density.
In the derivation, we have used Eq. (4.81c) which relates the charge density to the
longitudinal part of the current density in Coulomb gauge.
4.2.4 Coulomb-Gauge Scalar Potential as a Retarded Integral

The last step in the analysis of the Coulomb gauge entails the calculation of the
scalar potential, which is tantamount to finding an explicit expression for the ho-
mogeneous term Φhom in Eq. (4.82). We start from Eq. (4.96), which we recall,
⃗∥ (⃗ 1 ∂
E r − r⃗′ , t − t′ ) ( 2 ′ J⃗∥ (⃗
r , t) = − ∫ d3 r′ ∫ dt′ GR (⃗ ⃗ ′ ρ(⃗
r ′ , t′ ) + ∇ r′ , t′ )) . (4.99)
c ∂t
Here, in view of Eq. (1.48a), we can write the longitudinal component of the current
r′ , t′ ),
density as the gradient of a scalar field J (⃗
J⃗∥ (⃗ ⃗ ′ J (⃗
r ′ , t′ ) = ∇ r ′ , t′ ) . (4.100)
So, the longitudinal component of the electric field becomes

1 ∂
E⃗∥ (⃗
r , t) = − ∫ d3 r′ ∫ dt′ GR (⃗ ⃗′ (
r − r⃗′ , t − t′ ) ∇ r′ , t′ ) + ρ(⃗
J (⃗ r′ , t′ ))
c2 ∂t′
1 ∂
⃗ ′ GR (⃗
= ∫ d3 r′ ∫ dt′ ∇ r − r⃗′ , t − t′ ) ( r′ , t′ ) + ρ(⃗
J (⃗ r′ , t′ ))
c2 ∂t′
1 ∂
⃗ ∫ d3 r′ ∫ dt′ GR (⃗
= −∇ r − r⃗′ , t − t′ ) ( r′ , t′ ) + ρ(⃗
J (⃗ r′ , t′ )) . (4.101)
c2 ∂t′
⃗∥ (⃗
Because E ⃗ C (⃗
r, t) = −∇Φ r, t), a valid ansatz for the scalar potential is
1 ∂
r, t) = ∫ dt′ ∫ d3 r′ GR (⃗
ΦC (⃗ r − r⃗′ , t − t′ ) ( r′ , t′ ) + ρ(⃗
J (⃗ r′ , t′ )) , (4.102)
c2 ∂t′
where J is given in Eq. (4.100).
The decisive step is to show that our ansatz (4.102) fulfills Eq. (4.81c),
∂
0 r , t) = J⃗∥ (⃗
⃗ C (⃗
∇Φ r , t) . (4.103)
∂t
The proof is relatively straightforward. One first integrates by parts,
∂ ∂ 1 ∂
0 ⃗ C (⃗
∇Φ r , t) = 0 ∇ ⃗ ∫ dt′ ∫ d3 r′ GR (⃗
r − r⃗′ , t − t′ ) ( 2 ′ J (⃗
r′ , t′ ) + ρ(⃗
r′ , t′ ))
∂t ∂t c ∂t
= 0 ∫ dt′ ∫ d3 r′ GR (⃗
r − r⃗′ , t − t′ )
1 ∂2 ′ ∂ ′ ′ ′
×( ⃗ J (⃗
∇ ⃗ ρ(⃗
r ′ , t′ ) + ′ ∇ r , t )) . (4.104)
c2 ∂t′2 ∂t
Use of Eq. (4.102) and of the continuity equation leads to
1 ∂2 ⃗ ′ ′ ∂
⃗ C (⃗
0 ∂t ∇Φ r , t) = 0 ∫ dt′ ∫ d3 r′ GR (⃗
r − r⃗′ , t − t′ ) ( J∥ (⃗ ⃗′
r ,t ) + ∇ r′ , t′ ))
ρ(⃗
c2 ∂t′2 ∂t′
= 0 ∫ dt′ ∫ d3 r′ GR (⃗
r − r⃗′ , t − t′ )
1 ∂2 ⃗ ′ ′
×( J∥ (⃗
r ,t ) + ∇ ⃗ ′ ⋅ J⃗∥ (⃗
⃗ ′ (−∇ r′ , t′ ))) . (4.105)
c2 ∂t′2
This can be summarized as follows,
1 ∂2
⃗ C (⃗
0 ∂t ∇Φ r , t) = 0 ∫ dt′ ∫ d3 r′ GR (⃗
r − r⃗′ , t−t′ ) ( ⃗ ′2 ) J⃗∥ (⃗
−∇ r′ , t′ ). (4.106)
c2 ∂t′2
After a double partial integration, and use of Eq. (4.9), one can finally show that
Eq. (4.102) fulfills Eq. (4.103). Because Eq. (4.102) is manifestly retarded, we
have explicitly shown that the particular form of the scalar potential in Coulomb
gauge, given in Eq. (4.102), does not lead to a contradiction with respect to the
causality principle; the additional constraint (4.81c) ensures that the scalar potential
is manifestly retarded.
4.3 Exercises
● Exercise 4.1: Show that all contour choices CR , CA and CF shown in Figs. 4.1
and 4.2 lead to valid representations of Green functions that fulfill Eq. (4.9). Hint:
Apply the differential operator (1/c2 )∂ 2 /∂t2 − ∇ ⃗ 2 under the integral sign, invoking
a Fourier representation of the Green function.
● Exercise 4.2: Generalize the steps outlined in Eqs. (4.29)–(4.35), for the
advanced as opposed to the retarded Green function, using the contour CA given
in Fig. 4.1. Thus, verify the results given in Eqs. (4.41)–(4.44).
● Exercise 4.3: Generalize Eq. (4.66) to the case τ < 0. How would you have
to deform the integration contour in this case in order to identify the cut (in the
complex plane) of the photon propagator?
● Exercise 4.4: Pick some ψ(⃗ r2 + a2 )2 and ψ(⃗
r , t) = cos(ω0 t)/(⃗ r , t) = δ(t − t0 )/(⃗
r2 +
a ) and propagate these wave packets using (a) the retarded (b) the advanced,
2 2
and (c) the Feynman Green function. The dynamical equation is (4.1).
● Exercise 4.5: Show that the supplementary term in the vector potential, defined
in Eqs. (4.89a) and (4.90), can be written as a gradient vector and therefore does
not affect the result for the magnetic field, which thus is the same in Coulomb and
Lorenz gauges.
Chapter 5
Paradigmatic Calculations in
Electrodynamics
5.1 Overview
5.1.1 General Considerations

In electrodynamics as opposed to electrostatics, the emphasis is on the radiation
emitted by moving charge distributions, i.e., current distributions. In practically
important cases, the source oscillates at a specific frequency. Yet, the radiated
fields are needed at specific points in space, and one can assume that the radiation
is emitted continuously. In that case, it makes sense to transform the Green func-
tion to the mixed frequency-coordinate representation. In this representation, the
defining equation for the Green function becomes the so-called Helmholtz equation.
For classical fields, the radiation pattern from oscillating sources is described by the
retarded Green function, which has been given in Eq. (4.75) in the mixed frequency-
coordinate representation (see Sec. 4.1.5). In a typical case, the oscillating source
(antenna) is a dipole. However, higher-order multipole radiation can also be impor-
tant. It is thus imperative to expand the Green function of the Helmholtz equation
(which is equal to the retarded Green function in the mixed coordinate-frequency
representation) into multipoles. Appropriate integrations then lead to the relevant
formulas for the radiation pattern.
Finally, when calculating antenna problems, it is usually assumed that the an-
tenna (oscillating charge distribution) remains static during the emission process,
i.e., it does not move. In Secs. 2.3.4 and 2.3.5, we encountered the multipole de-
composition of the electrostatic Green function, which led to the multipole de-
composition of the electrostatic potential generated by a static charge distribution
[Eq. (2.99)]. The Helmholtz Green function fulfills a defining equation with a scalar
structure [see Eq. (5.19)]. In consequence, the angular momentum decomposition of
the Helmholtz Green function given in Eq. (5.72) involves the spherical harmonics,
which are scalar (not tensor) quantities. Indeed, in antenna problems, the equation
which relates the scalar potential to the oscillating charge distribution (5.77) retains
a scalar structure. By contrast, the equation which relates the current density to
the vector potential [Eq. (5.78)] has a vector structure, and the Green function,
strictly speaking, therefore is promoted to a tensor Green function. This is not
163
obvious from Eq. (5.78), but becomes more evident as one considers Eq. (5.158).
The tensor structure of the Helmholtz Green function necessitates an analysis of
the angular algebra which goes beyond the scalar structure of the scalar Helmholtz
Green function; the “spin” of the photon (the vector structure of the vector po-
tential) needs to be integrated into the formalism. This endeavor proceeds via the
introduction of vector spherical harmonics (Sec. 5.4.2).
By contrast, the emission of radiation by a moving point charge constitutes a
complementary problem where the charge distribution is trivial but the radiation
is due to the motion of the charge relative to the observer. The scalar and vector
potentials due to a moving point charge are otherwise known as the Liénard1 –
Wiechert2 potentials, which can be given in Lorenz and Coulomb gauges. It
is instructive to go through the alternative derivations because the final formulas
reveal a concrete realization of the cancellation mechanism originally described in
Sec. 4.2.
5.1.2 Wave Equation and Green Functions

We consider how general solutions to the wave equation can be obtained using
the retarded Green function. We integrate the source term directly (no partial
integrations) and may thus use the approximate simplified version of the retarded
Green function already given in Eq. (4.36). This is admissible because no additional
partial integrations need to be carried out. We recall the wave equation (4.1) with
sources, which has the general form
1 ∂2 1
( ⃗ 2 ) Ψ (⃗
−∇ r , t) = F (⃗
r , t) . (5.1)
c2 ∂t2 0
Here, Ψ is the signal generated by the source F . In the approximate form (4.36),
the retarded Green function is given by
c 1
r − r⃗′ , t − t′ ) ≈
GR (⃗ Θ(t − t′ ) δ(∣⃗
r − r⃗′ ∣ − c (t − t′ )) . (5.2)
r − r⃗′ ∣
4π0 ∣⃗
Using the retarded Green function and an arbitrary solution Ψhom (⃗
r , t) to the
homogeneous wave equation,
1 ∂2
( ⃗ 2 ) Ψhom (⃗
−∇ r , t) = 0 , (5.3)
c2 ∂t2
we obtain the general solution to the inhomogeneous wave equation as follows,
r , t) = Ψhom (⃗
Ψ (⃗ r − r⃗′ , t − t′ ) F (⃗
r , t) + ∫ GR (⃗ r′ , t′ ) d3 r′ dt′
1 t 1 ∣⃗
r − r⃗∣
= Ψhom (⃗
r , t) + ∫ d3 r′ ∫ dt′ δ (t′ − [t − r ′ , t′ )
]) F (⃗
4π0 Ê3 −∞ r − r⃗′ ∣
∣⃗ c
1 3 ′ 1 ′ r − r⃗′ ∣
∣⃗
= Ψhom (⃗
r , t) + ∫ d r F (⃗
r , t − ). (5.4)
4π0 r − r⃗′ ∣
∣⃗ c
1
Alfred–Marie Liénard (1869–1958)
2
Emil Johann Wiechert (1861–1928)
Paradigmatic Calculations in Electrodynamics 165
Here, the expression

r − r⃗′ ∣
∣⃗
tret = t − <t (5.5)
c
is the retarded time, which expresses the fact that the electromagnetic perturbation
propagates at the speed of light. Because we have tret < t for r⃗ ≠ r⃗′ , the integration
over t′ from −∞ to t always gives a nonvanishing result.
The same steps, applied to the advanced Green function, lead to
r , t) = Ψhom (⃗
Ψ (⃗ r − r⃗′ , t − t′ ) F (⃗
r, t) + ∫ GA (⃗ r′ , t′ ) d3 r′ dt′
1 3 ′ 1 ′ r − r⃗′ ∣
∣⃗
= Ψhom (⃗
r, t) + ∫ d r F (⃗
r , t + ), (5.6)
4π0 r − r⃗′ ∣
∣⃗ c
where
r − r⃗′ ∣
∣⃗
tadv = t + >t (5.7)
c
is the advanced time; this solution describes an electromagnetic signal (or a signal
of a different physical origin) that propagates from the future into the past.
5.2 Helmholtz Equation
5.2.1 Helmholtz Equation and Green Function

It is rather straightforward to apply the formalism developed in Sec. 5.1.2 to elec-
tromagnetic radiation phenomena. A simple radiating system consists of a localized
charge density ρ (⃗r , t) and a localized current density J⃗ (⃗
r , t). The system is consid-
ered to be localized if its dimensions are small compared to the wavelength of the
radiation. We consider radiating sources in a vacuum and begin with the equations
for the vector and scalar potentials in the Lorenz gauge [Eq. (1.77)],
1 ∂2 1
( ⃗ 2 ) Φ (⃗
−∇ r , t) = ρ (⃗
r , t) , (5.8)
2
c ∂t 2 0
1 ∂2
( ⃗ 2 ) A⃗ (⃗
−∇ r , t) = μ0 J⃗ (⃗
r , t) . (5.9)
c2 ∂t2
The second of these equations can be written as
1 ∂2 μ0 1 1 1 ⃗
( ⃗ 2 ) c A⃗ (⃗
−∇ r , t) = 2 J⃗ (⃗
r , t) = ( J (⃗
r , t)) . (5.10)
2
c ∂t 2 c c 0 c
The latter form is suggested by the relativistic formalism; indeed, one can order the
scalar and vector potentials, and the charge and current densities, into 4-vectors as
⃗ 1 ⃗
(Φ, c A) and (ρ, J) . (5.11)
c
The consistency of the physical units of the components of the first 4-vector can
be verified immediately, as follows: The electric field has the dimension of the
⃗ the magnetic induction field has the dimension of ∇×
expression ∇Φ, ⃗ A.⃗ For traveling
plane waves, one has ∣E∣⃗ ∼ c ∣B∣,

⃗ and so, in terms of physical units, the factor c in the
⃗
four-vector (Φ, c A) is explained. The charge density has units of [ρ] = [Q]/[r]3 =
⃗
[Q]/([t] [r]2 ) ([t]/[r]) = [J]/[c], which explains the units for the components of
the second 4-vector given in Eq. (5.11).
We integrate the wave equation as before,
r , t) + ∫ d3 r′ dt′ GR (⃗
r , t) = Φhom (⃗
Φ (⃗ r − r⃗′ , t − t′ ) ρ (⃗
r ′ , t′ ) (5.12a)
1
cA⃗ (⃗
r , t) = cA⃗hom (⃗ r − r⃗′ , t − t′ ) J⃗ (⃗
r , t) + ∫ d3 r′ dt′ GR (⃗ r ′ , t′ ) . (5.12b)
c
Here, Φhom and A⃗hom are solutions to the homogeneous wave equation. Using the
identity
1 r − r⃗′ ∣
∣⃗
r − r⃗∣ − c (t − t′ )) =
δ (∣⃗ δ (t − t′ − ), (5.13)
c c
we may carry out t′ integration, and find
1 1 r − r⃗′ ∣
∣⃗
r , t) = Φhom (⃗
Φ (⃗ r , t) + ∫ d3 r′ ′
r′ , t −
ρ (⃗ ), (5.14a)
4π0 ∣⃗
r − r⃗ ∣ c
1 1 r − r⃗′ ∣
∣⃗
cA⃗ (⃗
r , t) = cA⃗hom (⃗
r , t) + ∫ d3 r′ ⃗ (⃗
J r ′
, t − ). (5.14b)
4π0 c r − r⃗′ ∣
∣⃗ c
The latter equation can of course be rewritten as
1 r − r⃗′ ∣
∣⃗ 1
A⃗ (⃗
r , t) = A⃗hom (⃗
r , t) + ∫ d3 r′ [ ] J⃗ (⃗
r′ , t − ) , (5.15)
0 c 2 c r − r⃗′ ∣
4π∣⃗
,-- - - - -. - - - - -/
=μ0
and we see that the vacuum permeability μ0 naturally appears in the calculation.
The equation fulfilled by the retarded Green function,
1 ∂2 1
( ⃗ 2 ) GR (⃗
−∇ r − r⃗′ , t − t′ ) = δ (3) (⃗
r − r⃗′ ) δ(t − t′ ) , (5.16)
2
c ∂t 2 0
transforms to Fourier space as follows,
ω2 ⃗ 2 1
(− r − r⃗′ , ω) =
− ∇ ) GR (⃗ r − r⃗′ ) .
δ (⃗ (5.17)
c2 0
Setting k = ω/c, and defining
GR (k, r⃗ − r⃗′ ) ≡ GR (⃗
r − r⃗′ , ω = c k) , (5.18)
we obtain the (inhomogeneous) Helmholtz equation
1
(∇⃗ 2 + k 2 ) GR (k, r⃗ − r⃗′ ) = − δ (⃗ r − r⃗′ ) . (5.19)
0
The “natural parameter ” of the Helmholtz equation is the wave number k, not the
frequency ω. The frequency argument in the expression GR (⃗ r −⃗r′ , ω) is generated by
the Fourier transform of GR (⃗ r − r⃗′ , t − t′ ); it is thus natural to write ω in the second
argument slot. However, when k becomes a mere parameter of the (inhomogeneous)
Helmholtz equation (5.19), then it is natural to write it as the first argument of the
Green function. The Helmholtz equation is solved by the Green function given in
Eq. (4.75),
′
ei k ∣⃗r−⃗r ∣
GR (k, r⃗ − r⃗′ ) = . (5.20)
4π0 ∣⃗r − r⃗′ ∣
The electrostatic Green function G defined in Eq. (2.22) is recovered as the static
limit of the electrodynamic Green function,
1
r − r⃗′ , ω) = lim GR (k, r⃗ − r⃗′ ) =
lim GR (⃗ r , r⃗′ ) .
= G(⃗ (5.21)
ω→0 k→0 r − r⃗′ ∣
4π0 ∣⃗
In Fourier space, we have by convolution,
′
ω eiω∣⃗r−⃗r ∣/c
r , ω) = ∫ d r GR ( , r⃗ − r⃗′ ) ρ (⃗
Φ (⃗ 3 ′
r′ , ω) = ∫ d3 r′ r′ , ω) , (5.22a)
ρ (⃗
c 4π0 ∣⃗r − r⃗′ ∣
′
1 ω 1 eiω∣⃗r−⃗r ∣/c ⃗ ′
c A⃗ (⃗
r , ω) = ∫ d r GR ( , r⃗ − r⃗ ) J⃗ (⃗
3 ′ ′
r′ , ω) = ∫ d3 r′ J (⃗
r , ω) .
c c c 4π0 ∣⃗r − r⃗′ ∣
(5.22b)
This calculation identifies Eqs. (5.22a) and (5.22b) as the Fourier transforms of
Eqs. (5.12a) and (5.12b). Convolution in time is equivalent to multiplication in
frequency space.
5.2.2 Helmholtz Equation in Spherical Coordinates

The theory of Green functions of the wave equation is connected with the Helmholtz
equation (5.19), whose homogeneous form reads as follows,
⃗ 2 + k 2 ) Φ(r, θ, ϕ) = 0 .
(∇ (5.23)
We eventually aim to expand the Helmholtz Green function (5.20) into multipole
components, in the spherical basis. To this end, we first need to study the solution of
the homogeneous Helmholtz equation in the spherical basis, before connecting them
at the cusp, in order to calculate the multipole expansion of the Green function.
Therefore, we need to study the solutions of the homogeneous Helmholtz equation,
expressed in spherical coordinates r, θ, and ϕ,
⃗ 2 + k 2 ) Φ(r, θ, ϕ) = ( 1 ∂2 1 ∂ ∂ 1 ∂2
(∇ r + sin θ + + k 2 ) Φ(r, θ, ϕ)
r ∂r2 r2 sin θ ∂θ ∂θ r2 sin θ ∂ϕ2
∂2 2 ∂ ⃗2
L
=( 2+ − 2 + k 2 ) Φ(r, θ, ϕ) = 0 , (5.24)
∂r r ∂r r
⃗ 2 has been defined in Eq. (2.31). The
with ∣m∣ ≤ = 0, 1, 2, . . . . The operator L
general solution of the homogeneous Helmholtz equation can be written as
∞
Φ(r, θ, ϕ) = ∑ ∑ (a m j (kr) + b m y (kr)) Y m (θ, ϕ) , (5.25)
=0 m=−
where the separation constants a m and b m can take arbitrary values, and each
function in the set fulfills the Helmholtz equation, separately. The spherical Bessel
and Neumann functions are written as j and y . One uses lowercase letters instead
of the uppercase notation, which is used for the ordinary Bessel function J and Y
discussed in Sec. 2.4.2. The spherical Bessel functions have been mentioned very
briefly in Sec. 2.4.3 [see Eq. (2.179)]. We recall that they are defined as follows,
π 1/2 π 1/2
j (x) = ( ) J+1/2 (x) , y (x) = ( ) Y+1/2 (x) . (5.26)
2x 2x
The defining differential equation for spherical Bessel functions is given as
∂2 2 ∂ ( + 1)
( + − + 1) j (x) = 0 . (5.27)
∂x2 x ∂x x2
Spherical Bessel functions fulfill the following recursion relations,
2 + 1
j−1 (x) + j+1 (x) = j (x) , (5.28a)
x
j−1 (x) − ( + 1) j+1 (x) = (2 + 1) j′ (x) , (5.28b)

where j′ (x)
= ∂j (x)/∂x is the derivative with respect to the argument (not param-
eter). Alternatively, one can express the derivative as follows,
+1
j′ (x) = j−1 (x) −j (x) = − j+1 (x) + j (x)
x x
1 1
= (j+1 (x) + j−1 (x)) − j (x) . (5.29)
2 2x
By a straightforward generalization of formulas given in Sec. 2.4.2, we can establish
the following asymptotic behavior
x (2 + 1)!!
j (x) ∼ , y (x) ∼ − , x → 0, (5.30)
(2 + 1)!! x+1
1 π 1 π
j (x) ∼ sin (x − ) , y (x) ∼ − cos (x − ) , x → ∞. (5.31)
x 2 x 2
The double factorial fulfills the general relations (if is an integer)
(2 + 2)! 2+1 Γ( + 3/2)
(2 + 1)!! = (2 + 1)(2 − 1)(2 − 3) . . . 5 ⋅ 3 ⋅ 1 = = √ , (5.32)
2+1 ( + 1)! π
where the latter formula provides the generalization of the result (5.30) to non-
integer order .
In order for Eq. (5.25) to represent a general solution to the Helmholtz equation,
we have to demand that
∂2 2 ∂ ( + 1)
( + − + k 2 ) j (k r) = 0 , (5.33)
∂r2 r ∂r r2
which is in fact equivalent to Eq. (5.27), under the specialization x → k r.
Let us try to find a connection of the ordinary and spherical Bessel functions.
Indeed, from Bessel’s differential equation (2.144), we may infer that the function
Z = xα Jn (β xγ ) fulfills the relation
∂2 1 − 2α ∂ α2 − n2 γ 2
2
Z(x) + Z(x) + (β 2 γ 2 x2γ−2 + ) Z(x) = 0 . (5.34)
∂x x ∂x x2
For our case, we set γ = 1, α = − 12 , β = k, and n = + 1/2. Then,
∂ 2 Z 2 ∂Z
1
− ( + 12 )2
+ + (k 2 + 4
) Z = 0, (5.35)
∂x2 x ∂x x2
which is trivially equivalent to
√
∂ 2 Z 2 ∂Z ( + 1) π
2
+ + (k 2 − ) Z = 0, Z = Z(x) = J+1/2 (k x) . (5.36)
∂x x ∂x x2 2x
This is just the desired Eq. (5.33). The complementary solution y also fulfills the
defining equation. Let us verify some asymptotic properties by way of example. For
= 10, we have
x10
j=10 (x) ∼ , x → 0. (5.37)
13 749 310 575
This is illustrated in Fig. 5.1(a). For intermediate, finite values of x, there are
deviations from the asymptotic behavior [see Fig. 5.1(b)]. For large argument, we
have
1 10 π
j=10 (x) ∼ sin (x − ), x → ∞. (5.38)
x 2
This is verified in Fig. 5.1(c). The spherical Bessel functions can be expressed in
terms of trigonometric functions,
sin x sin x cos x 3 1 3
j0 (x) = , j1 (x) = 2 − , j2 (x) = ( 3 − ) sin x − 2 cos x ,
x x x x x x
(5.39a)
and we have for the Neumann functions,
cos x cos x sin x 3 1 3
y0 (x) = − , y1 (x) = − 2 − , y2 (x) = (− 3 + ) cos x − 2 sin x .
x x x x x x
(5.40a)
The Hankel function of the first and second kind are given as
(1) (2)
h (x) = j (x) + i y (x) , h (x) = j (x) − i y (x) . (5.41)
They can be written in terms of exp(ix) and powers of x. Based on Eqs. (5.39)
and (5.40), we can easily show that
(1) eix (1) eix eix (1) eix 3eix 3eix
h0 (x) = −i , h1 (x) = − −i 2 , h2 (x) = i − 2 − i 3 . (5.42)
x x x x x x
For large x,
(1) ei(x−π/2) +1 e
ix
h (x) → −i = (−i) x → ∞, (5.43)
x x
i.e., the Hankel functions differ by a complex phase in the limit x → +∞.

0 000 02
0 00
01
0 0004 0 00
0
0 0002 1 2 3 4
01 0 00
0 1 2 3 4 5 02 0 00
(a) (b) (c)
Fig. 5.1 Panel (a) illustrates the verification of Eq. (5.37), i.e., of the asymptotics of the
spherical Bessel function j10 (x) for x → 0. In panel (b), we verify Eq. (5.37) on a larger
scale; the oscillations of the Bessel function become evident. Finally, panel (c) verifies
Eq. (5.38), in a regime where the asymptotic behavior is approximated by a trigonometric
function, with a phase and a prefactor. In all panels, the solid curve is for the exact Bessel
function, whereas the dashed curve is the approximation.
5.2.3 Radiation Green Function

We have come across the inhomogeneous Helmholtz equation (5.19),
1
⃗ 2 + k 2 ) GR (k, r⃗ − r⃗′ ) = −
(∇ r − r⃗′ ) ,
δ (⃗ (5.44)
0
and the Helmholtz Green function given in Eqs. (4.75) and (5.20),
r − r⃗′ ∣)
exp (i k ∣⃗
GR (k, r⃗ − r⃗′ ) = . (5.45)
r − r⃗′ ∣
4π0 ∣⃗
⃗ = r⃗/r, it is instructive to independently verify that GR
Then, using the relation ∇r
fulfills the defining inhomogeneous Helmholtz equation. One needs to carefully keep
track of all singular terms in order not to miss the Dirac-δ function,
exp (ik r) exp (ik r) ik 1
⃗2
∇ ⃗ ⋅∇
=∇ ⃗ ⃗ ⋅ (exp (ik r)[ ∇r+
=∇ ⃗ ∇
⃗ ]) ,
r r r r
k⃗2 1 ik 1 1
= exp (ik r)(− ⃗ ⋅ ∇r+ik
∇r ⃗ ⃗ ⋅∇
∇r ⃗ ( )+ ∇⃗ ⋅ ∇r+ik
⃗ ⃗ ⋅∇
∇r ⃗ +∇⃗2 )
r r r r r
k⃗2 r⃗ r⃗ r⃗ −⃗r ik 2 r⃗ −⃗r 1

= exp (ik r)(− ⃗2 )
( ⋅ )+ik ⋅ ( 3 )+ ( )+ik ( ) ⋅ ( 3 )+ ∇
r r r r r r r r r r
−k 2 ik 2ik ik 1
= exp (ik r)( ⃗2 )
− 2 + 2 − 2 +∇
r r r r r
−k 2 ⃗ 2 1 k2
= exp (ik r)( + ∇ ) =− exp (ik r)−4πδ (3) (⃗ r′ ) .
r −⃗ (5.46)
r r r
Thus, the validity of Eq. (5.45) is verified once more.
The central result of the current derivation is the multipole expansion of the
radiation Green function, which is analogous to the electrostatic case discussed in
Sec. 2.3.5, but a little more involved. An appropriate ansatz for the Green function
in spherical coordinates reads as follows,

r − r⃗′ ∣) ∞
exp (i k ∣⃗ ∗
GR (k, r⃗ − r⃗′ ) = = ∑ ∑ G (k, r, r′ ) Ym (θ, ϕ) Ym (θ′ , ϕ′ ) .
r − r⃗′ ∣
4π0 ∣⃗ =0 m=−l
(5.47)
′
It is our task to find an appropriate expression for G (k, r, r ). We proceed as in
Sec. 2.3.5, by recalling that the operator ∇ ⃗ 2 + k 2 acts onto a test function of the
form R(r) Ym (r̂) as follows,
1 ∂ 2 ∂ ( + 1)
⃗ 2 + k 2 ) R(r) Ym (r̂) = (
(∇ r + k2 − ) R(r) Ym (r̂) , (5.48)
2
r ∂r ∂r r2
The Dirac-δ function is expanded as follows,
1
r − r⃗′ ) =
δ(⃗ δ(r − r′ ) δ(θ − θ′ ) δ(ϕ − ϕ′ ) . (5.49)
r2 sin θ
The completeness relation for the sum over spherical harmonics reads
∗ 1
∑ Ym (r̂) Ym (r̂) = δ(θ − θ′ ) δ(ϕ − ϕ′ ) . (5.50)
m sin θ
The radial Green function G (k, r, r′ ) as defined in Eq. (5.47) thus has to fulfill the
following relation,
∂ ∂ 1
(r2 G (k, r, r′ )) + (k 2 r2 − ( + 1)) G (k, r, r′ ) = − δ (r − r′ ) . (5.51)
∂r ∂r 0
Let us start with the “homogeneous domain” away from the peak of the Dirac-δ
function. We shall assume that the radial part of the Green function has the func-
tional form G (r, r′ ) = f (k r, k r′ ) and seek solutions to the homogeneous equation
∂ ∂
(r2 f (k r, k r′ )) + (k 2 r2 − ( + 1)) f (k r, k r′ ) = 0 . (5.52)
∂r ∂r
The general solution is a linear combination of the two spherical Bessel functions,
f (k r, k r′ ) = a (k r′ ) j (k r) + b (k r′ ) y (k r) . (5.53)
We now appeal to Eq. (5.26) for the behavior of the Bessel and Neumann functions
near the origin. Since the Green function is regular at r = 0, and in particular, for
r < r′ , we must choose the regular solution, i.e., the spherical Bessel function j , in
this domain,
f (k r, k r′ ) = G (k, r, r′ ) = a (k r′ ) j (k r) , r < r′ . (5.54)
′
For r > r , we certainly need a contribution of the Bessel y which is not regular at
the origin. However, if we wish to construct a Wronskian upon action of the radial
differential operator, then we need to add a contribution from j as well. A natural
assumption is to make an ansatz for the solution G to be a superposition of an
incoming and an outgoing wave,
π π
∝ exp [−i (kr − )] (incoming) , ∝ exp [i (kr − )] (outgoing) . (5.55)
2 2
Two considerations support the outgoing wave. The first is that the Green function
“propagates”, as it were, the point r⃗′ to the point r⃗. An incoming wave converging
to the point r⃗′ should be propagated to point r⃗. In an eigenfunction decomposition
of the Green function, we would use outgoing and incoming waves as given in
Eq. (5.55). These considerations are valid for r > r′ , in which case the wave is
outgoing at point r⃗. From the analogy with the Green function for electrostatics,
we also conjecture that the solution should fall off as r−1 for large r. We thus assume
that
f (k r, k r′ ) = G (k, r, r′ ) = c (k r′ ) [i j (k r) − y (k r)] , r > r′ . (5.56)
Then, for k r ≫ 1,
1 π
G (k, r, r′ ) = c (k r′ ) [i j (k r) − y (kr)] → c (k r′ )
exp [i (k r − )] .
kr 2
(5.57)
Continuity at r = r′ is automatically fulfilled if we write the radial component of the
Green functions as a product, supplementing for a (r′ ) and c (r′ ) the respective
“other” solution of the radial Helmholtz equation. The result of this operation is
G (r, r′ ) = a0 (k) j (k r< ) [i j (k r> ) − y (k r> )] , (5.58)
where a0 (k) is a prefactor which can only depend on k, and remains to be
determined. In order to convince ourselves of the continuity, we consider r′ to
be constant and vary r, starting at r = 0. As we increase r, the variable r assumes
the role of r< until r = r′ (cusp). At the cusp, it does not matter which identification
we make, because r and r′ are equal. For r > r′ , the identification of r< and r> is
different, but we have gone through the cusp. In some sense, the cusp thus ensures
the continuity.
The constant a0 is determined from the discontinuity in the derivative of
G (k, r, r′ ) at r = r′ . Note that G (k, r, r′ ) is continuous at r = r′ , but the derivative
d
G (r, r′ ) is discontinuous. Only the region
dr
r′ − ≤ r ≤ r′ + (5.59)
gives a non-zero contribution. Taking notice of the continuity of the Green function
and of the discontinuity of its derivative at the cusp, we have
r ′ +
∂ ∂
lim ∫ [ (r2 G (k, r, r′ )) + (k 2 r2 − ( + 1)) G ((k, r, r′ )] dr
→0 ∂r ∂r
r ′ −
′
1 r +
=− ∫′ δ (r − r′ ) dr . (5.60)
0 r −
This implies that the derivative of the radial component of the Green function must
have a discontinuity at the cusp, of magnitude [see Eq. (2.85)]
∂ ∂ 1
r2 G (r, r′ )∣ − r2 G (r, r′ )∣ =− . (5.61)
∂r ′
r=r +ε ∂r ′
r=r −ε 0
We can now identify the places where the differentiations take place, immediately to
the right and left of the cusp, based on the formulation of the radial Green function
in terms of r< and r> ,
′
r=r +
∂ 1
r2 a0 (k) j (k r< ) [ij (k r> ) − y (k r> )]∣ =− . (5.62)
∂r ′
r=r − 0
The differentiations are carried out as follows,

d
r2 a0 (k) {j (kr′ ) ( [ij (k r) − y (k r)]∣ )
dr r=r ′ +
d 1
−( j (kr)∣ ) [ij (k r′ ) − y (k r′ )]} = − . (5.63)
dr ′
r=r − 0
Finally, we evaluate the resulting expression in the limit → 0, which implies r = r′ ,

and obtain
d d 1
r2 a0 (k) (j (k r) [i j (kr) − y (kr)] − ( j (kr)) [i j (kr′ ) − y (k r′ )]) = − .
dr dr 0
(5.64)
The term proportional to the mixed product of Bessel functions and derivatives
cancels, and we have
d d 1
a0 (k) r2 (−j (k r) y (k r) + y (kr) j (k r)) = − . (5.65)
dr dr 0
From the differential equation (5.33) fulfilled by the j and y , we can easily derive
that
d 2 d d
[r (−j (k r) y (k r) + y (kr) j (k r))] = 0 , (5.66)
dr dr dr
and so the Wronskian −j y′ +y j′ can be evaluated for any argument. In particular,
we can use the form for the two spherical Bessel functions as r′ → 0,
⎧ ⎤⎫
⎪ (2 − 1)!! ⎡⎢ d (k r) ⎥⎪

⎪ − (k r) d (2 − 1)!! ⎪ 1
a0 (k) r2 ⎨ [ (− )] + (− ) ⎢ ⎥⎬ = − ,
⎪ ⎢ ⎥
(2 + 1)!! dr ⎣ dr (2 + 1)!! ⎦⎪
l+1 +1
⎪
⎩ (k r) (k r) ⎪
⎭
0
(5.67)
Carrying out the differentiation, we find that
⎡ ⎤
⎢
2 ⎢ (k r)

−( + 1) k 1 ⎛ k (k r)−1 ⎞⎥
a0 (k) r ⎢ ( ) + (− ) ⎥=−1 .
⎥ (5.68)
⎢ (2 + 1) (k r)+2 (k r)
+1
⎝ (2 + 1) ⎠⎥ 0
⎣ ⎦
Assembling all factors, we finally obtain the relation
−( + 1)k − k 1 1
a0 (k) r2 [ ] = a0 (k) (− ) = − . (5.69)
(k r)2 (2 + 1) k 0
The explicit expression for the overall coefficient thus reads as
k
a0 (k) = , (5.70)
0
and the final form for G (k, r, r′ ) is

1 1 (1)
G (k, r, r′ ) = k j (k r< ) [i j (k r> ) − y (k r> )] = i k j (k r< ) h (k r> ) ,
0 0
(5.71)
where the Hankel functions have been defined in Eq. (5.41). We conclude that
in spherical coordinates, the Green function has the following angular momentum
decomposition,
r − r⃗′ ∣)
exp (i k ∣⃗
GR (k, r⃗ − r⃗′ ) =
r − r⃗′ ∣
4π0 ∣⃗
ik ∞ (1) ∗ ′ ′
= ∑ ∑ j (k r< ) h (k r> ) Ym (θ, ϕ) Ym (θ , ϕ ) . (5.72)
0 =0 m=−
By contrast, we recall the angular momentum decomposition [Eqs. (2.25) and (2.91)]
of the Green function of electrostatics,
1 1 ∞ 1 r<
r , r⃗′ ) =
G(⃗ = ∑ ∑
∗
Ym (θ, ϕ) Ym (θ′ , ϕ′ ) . (5.73)
r − r⃗′ ∣ 0 =0 m=− 2 + 1 r>+1
4π0 ∣⃗
The matching is successful if
(1) 1 r<
lim GR (k, r⃗ − r⃗′ ) = G(⃗
r , r⃗′ ) , lim i k j (k r< ) h (k r> ) = . (5.74)
k→0 k→0 2 + 1 r>+1
The asymptotic relations (5.30) lead to the following expansion,
(1)
i k j (k r< ) h (k r> ) = i k j (k r< ) (j (k r< ) + iy (k r> ))
k→0
= − k j (k r< ) y (k r> )
k→0 (k r< ) (2 − 1)!! 1 r<
= −k (− ) = , (5.75)
(2 + 1)!! (kr> )+1 2 + 1 r>+1
confirming Eq. (5.74).
5.3 Localized Harmonically Oscillating Sources
5.3.1 Basic Formulas and Multipole Expansion

We consider sources which oscillate at a fixed angular frequency ω,
ρ (⃗
r , t) = ρ0 (⃗
r ) exp (−i ω t) , J⃗ (⃗
r, t) = J⃗0 (⃗
r ) exp (−i ω t) . (5.76)
The real part of these source functions and of the potentials are the physical pa-
rameters, but it is computationally useful to add the imaginary part, in order to be
able to restrict the discussion to a single Fourier component. In view of Eq. (5.22b),
the scalar and vector potentials generated by this current source are given as
ω
Φ(⃗r, t) = e−iωt ∫ d3 r′ GR ( , r⃗ − r⃗′ ) ρ0 (⃗
r′ ) , (5.77)
c
1 ω
A⃗ (⃗
r, t) = e−iωt ∫ d3 r′ 2 GR ( , r⃗ − r⃗′ ) J⃗0 (⃗ r′ ) . (5.78)
c c
It is natural to assume that the vector potentials and fields have the same harmonic
dependence,
A⃗ (⃗
r, t) = A⃗0 (⃗ ⃗ (⃗
r) exp (−i ω t) , B ⃗0 (⃗
r , t) = B r ) exp (−i ω t) ,
E⃗ (⃗ ⃗0 (⃗
r, t) = E r) exp (−i ω t) . (5.79)
The spatial variation of the vector potential is contained in the quantity A⃗0 (⃗
r)
which only depends on space (but not on time),
r − r⃗′ ∣/c) ⃗ ′
exp (i ω ∣⃗
A⃗0 (⃗
r ) = ∫ d3 r′ J0 (⃗
r ). (5.80)
r − r⃗′ ∣
4π0 c2 ∣⃗
The magnetic induction field is given by
⃗0 (⃗
B r) = ∇⃗ × A⃗0 (⃗r) . (5.81)
At the observation point, which is far away from the source, the current density
vanishes, J⃗0 (⃗
r) = ⃗
0, and the electric field is given by the Ampere–Maxwell law
⃗ ×B ⃗0 (⃗ 1 ∂ ⃗ ω ⃗
∇ r , t) = 2 E0 (⃗r , t) = −i 2 E 0 (⃗
r , t) . (5.82)
c ∂t c
The spatial dependence of the electric field thus reads as
⃗0 (⃗ ic ⃗ ⃗ ic ⃗ ⃗ × A⃗0 (⃗
E r) = (∇ × B0 (⃗
r)) = ∇ × (∇ r )) . (5.83)
ω ω
Equations (5.81) and (5.83) imply that the knowledge of the vector potential A⃗0 (⃗ r)
is sufficient for the calculation of both E ⃗0 (⃗
r) and B ⃗0 (⃗r ) in the source-free region; it
is not necessary to consider the scalar potential (5.77) at all.
We have already restricted our sources configurations (the domain where J⃗0 is
nonvanishing) to be localized with a characteristic dimension, d, satisfying
c λ
d≪ = , (5.84)
ω 2π
i.e., with a spatial dimension much less than the wavelength of the radiation. This
restriction only pertains to the dimension d of the sources configuration, not to
the distance r from the source itself. The exponential term in the integrand for
the vector potential suggests that the potential will have a drastically different
spatial dependence depending on the range of r⃗. The condition d ≪ r ≪ c/ω is
relevant to the near field or static zone, less than a wavelength away, while the
condition r ≫ c/ω characterizes the far field or radiation zone, many wavelengths
away. The vector potential, given by Eq. (5.78), can be evaluated using the angular
momentum decomposition (5.72) for the Helmholtz Green function, which we recall
for convenience,
ik ∞ (1)
GR (k, r⃗ − r⃗′ ) = ∗ ′ ′
∑ ∑ j (k r< ) h (k r< ) Ym (θ, ϕ) Ym (θ , ϕ ) . (5.85)
0 =0 m=−
In view of Eq. (5.78), the vector potential A⃗0 (⃗r ) can be decomposed into multipoles
as follows,
ik
A⃗0 (⃗ ∑ Ym (θ, ϕ) ∫ d r J⃗0 (⃗
3 ′ (1)
r) = r′ )j (k r< )h (k r> )Ym
∗
(θ′ , ϕ′ ) . (5.86)
0 c2 m
We here introduce the notation ∑m for the sum ∑∞

=0 ∑m=− ; it will be used in the
following. This expansion ignores, in some sense, the intrinsic angular momentum
of the current density J⃗0 (⃗r′ ). The current density J⃗0 (⃗
r′ ) transforms as a vector, just
like the spherical harmonics Ym (θ, ϕ). A more systematic expansion is found if one
adds the angular momentum inherent to the vector field to the angular momentum
of the spherical harmonic. As it stands, in order to obtain a valid expansion in
Eq. (5.86) and use the orthogonality properties of the spherical harmonics, one has
to expand each individual Cartesian component of the vector-valued current density
J⃗0 (⃗
r′ ) into spherical harmonics. This procedure mixes Cartesian and spherical
coordinate systems and is somewhat unsystematic. A better way will be described
in Sec. 5.4; but Eq. (5.86) is good enough for our purposes, for the time being. In
fact, we shall first discuss dipole radiation based on Eq. (5.86), before generalizing
to arbitrary multipole orders using the tensor Green function.
If the source current is localized near r′ ≈ 0, we can set r< = r′ and r> = r, and
the radiated vector potential reads
ik
A⃗0 (⃗ ∑ h (k r) Ym (θ, ϕ) ∫ d r J⃗0 (⃗
(1) 3 ′
r) = r′ ) j (k r′ ) Ym
∗
(θ′ , ϕ′ ) (5.87)
0 c2 m
If the wavelength emitted by the radiating source is much larger than the character-
istic length scale d of the charge distribution, then we have k d ≪ 1. Furthermore,
in this case, the argument k r′ of the Bessel function j (k r′ ) fulfills k r′ < k d ≪ 1,
and the spherical Bessel function can be approximated with its asymptotic form
near r′ = 0, that is, j (z) ∼ z /(2 + 1)!!. One then obtains
(1)
ik ∞ k h (k r)
A⃗0 (⃗
r) ≈ ∑ ∑ Ym (θ, ϕ) ∫ dr′ r′+2 ∫ dΩ′ J⃗0 (⃗r′ ) Ym
∗
(θ′ , ϕ′ ) .
0 c2 =0 m=− (2 + 1)!!
(5.88)
We define the near-field region to be the spatial region closer than a wavelength
away from the antenna, but still large compared to the spatial extent of the charge
distribution,
c (1) (2 − 1)!!
d<r≪ , kd ≪ kr ≪ 1, h (k r) ≈ +i y (k r) ≈ −i , (5.89a)
ω (k r)+1
1 ∞ Ym (θ, ϕ)
A⃗0 (⃗
r) ≈ ∑ ∑
′ ′+2
∫ dr r ∫ dΩ J⃗0 (⃗
′
r′ ) Ym
∗
(θ′ , ϕ′ ) .
0 c2 =0 m=− (2 + 1)!! r+1
(5.89b)
So, in the near-field region, the double factorial (2 − 1)!! cancels, and we have a
very compact expression for a long-wavelength emitter. In the far zone, still for a
long-wavelength emitter, we have in view of Eq. (5.43),
(1) ei(k r−π/2)
kd ≪ 1, kr ≫ 1, h (k r) ≈ , (5.90a)
ikr
∞
k ei(k r−π/2) Ym (θ, ϕ)
A⃗0 (⃗
r) ≈ ∑ ∑ ′ ′+2
∫ dr r ∫ dΩ J⃗0 (⃗
′
r′ ) Ym
∗
(θ′ , ϕ′ ) .
=0 m=− 0 c2 r (2 + 1)!!
(5.90b)
By dimensional analysis, the source integral is proportional to
∫ dΩ J⃗0 (⃗
′ ′+2 ′
∫ dr r r′ ) Ym
∗
(θ′ , ϕ′ ) ∼ j0 d+3 , (5.91)
where j0 is a characteristic scale of the current density. The expansion in multipoles
thus is seen to be an expansion in powers of the parameter (k d) , where k = ω c,
and can be evaluated term by term. The lowest nonvanishing term then gives the
dominant contribution. In evaluating the source integral (5.91), one has to multiply
each of the three spatial components of J⃗0 (⃗r′ ) by the spherical harmonic Ym
∗
(θ′ , ϕ′ )
′
and evaluate the overlap integral over the entire solid angle dΩ .
5.3.2 Asymptotic Limits of Dipole Radiation

The lowest nonvanishing contribution to electromagnetic radiation comes from an
oscillating dipole and is referred to as dipole radiation. By contrast, the integral
over J⃗0 (⃗ r′ ) for = 0 and m = 0 in Eq. (5.91) projects out only the spherically
symmetric part of the J⃗0 (r′ , θ′ , ϕ′ ),
⃗ = ∫ J⃗0 (⃗ ∗ 1
D r′ ) r′2 Y00 (θ′ , ϕ′ ) dr′ dΩ′ = √ ∫ J⃗0 (⃗
r′ ) d3 r′ , (5.92)
4π
If J⃗0 (⃗
r′ ) were a scalar, then this integral would naturally be referred as a monopole
radiation integral. Somehow, we have to relate the vector-valued space integral
over J⃗0 (⃗
r′ ) to the total oscillating dipole moment. Charge conservation helps. In
the mixed frequency-coordinate representation, the charge conservation condition
reads as
⃗ ⋅ J⃗0 (⃗
∇ r ) = − (−i ωρ0 (⃗
r )) = i ωρ0 (⃗
r) . (5.93)
This suggests the following trick in order to convert the integral over the current
distribution into an integral over the charge distribution: just multiply the current
density by a coordinate and calculate the divergence. The result consists of two
terms, one of which reproduces the current density, and the second yields the charge
density by virtue of current conservation. The integral over the total divergence
vanishes, and the charge density is obtained in the final result.
This program is implemented as follows. The integral over the current density
is converted to one over the charge density using the identity
3 3
∂
⃗ ⋅ (xm J⃗0 (⃗
∇ r )) = ∑ [xm J0,n (⃗ ⃗ ⋅ J⃗0 (⃗
r ) δn m + xm ∇
r )] = ∑ J0,n (⃗ r)
n=1 ∂xn n=1
= J0,m (⃗
r ) + i ω xm ρ0 (⃗
r) , (5.94)
r ) refers to the mth Cartesian component of J⃗0 (⃗
where J0,m (⃗ r ). So,
3
⃗ = √1 ∫ J⃗0 (⃗
D
ω
r′ ) d3 r′ = −i √ ∫ ∑ êm x′m ρ0 (⃗ r′ ) d3 r′
4π 4π m=1
ω ′ ω
= − i√ ∫ r⃗ ρ0 (⃗r′ ) d3 r′ = −i √ p⃗0 , p⃗0 = ∫ r⃗′ ρ0 (⃗r′ ) d3 r′ . (5.95)
4π 4π
This last integral is the electric dipole moment of the charge distribution, i.e., p⃗0 .
We recall Eq. (5.90); for the dipole case, this formula is exact for a localized, long-
(1)
wavelength source with k d ≪ 1, because the Hankel function h0 (k r) consists of
only one term [see Eq. (5.42). Then, the = 0 component amounts to
k exp (ik r) 1 1
A⃗0 (⃗
r) =
2
√ (∫ J⃗0 (⃗
r′ ) r′2 √ dr′ dΩ′ )
0 c kr 4π 4π
k exp (ik r) ω k p⃗0
= (−i p⃗0 ) = −i . (5.96)
4π0 c2 kr 4π 4π0 c
The magnetic induction generated by an oscillating electric dipole is
⃗0 (⃗ ⃗ × A⃗0 (⃗ k ⃗ exp (i k r) k p⃗0 ⃗ ( exp (i k r) )

B r) = ∇ r) = −i p0 (
∇ × [⃗ )] = i ×∇
4π0 c r 4π0 c r
k p⃗0 d exp (i k r) ⃗ k r⃗ exp (ik r) 1
=i ×( ) ∇r = −i ( × p⃗0 ) (ik − )
4π0 c dr r 4π0 c r r r
k 2 exp (ik r) 1
= (1 − ) (r̂ × p⃗0 ) , (5.97)
4π0 c r ik r
where r̂ = r⃗/r. The leading term, for large r, is the magnetic field in the radiation
zone,
⃗0 (⃗ k 2 exp (ik r)
B r) ∼ (r̂ × p⃗0 ) , k r → ∞. (5.98)
4π0 c r
We can use Eq. (5.83) in order to calculate the electric field,
ic2 c ik r⃗
E⃗0 (⃗
r) = (∇ ⃗0 (⃗
⃗ ×B r )) = i (∇ ⃗0 (⃗
⃗ ×B r )) = ⃗ × [( × p⃗0 ) f (r)] ,
∇
ω k 4π0 r
exp (ik r) 1
f (r) = (1 − ). (5.99)
r ik r
The leading term, for large r, is generated by the gradient operator pulling down a
factor k⃗ from the exponential, and can be written as
⃗0 (⃗ ik ⃗ r⃗ d ik r⃗ r⃗ ik exp (ik r)
E r) ≈ (∇ r) × ( × p⃗0 ) f (r) ≈ [ × ( × p⃗0 )]
4π0 r dr 4π0 r r r
k exp (ik r)
2
= [(r̂ × p⃗0 ) × r̂] . (5.100)
4π0 r
So, the electric field in the radiation zone is given as
⃗0 (⃗ k 2 exp (ik r)
E r) ∼ [(r̂ × p⃗0 ) × r̂] , k r → ∞. (5.101)
4π0 r
Together with the corresponding asymptotic formula for the magnetic field given in
Eq. (5.98), one infers that
⃗0 (⃗
E ⃗0 (⃗
r) ∼ c B r ) × r̂ , k r → ∞. (5.102)
In the radiation zone, the electric field, the magnetic field as well as the position
⃗0 form a right-handed system.
unit vector r̂ ∥ E⃗0 × B
It is very instructive to calculate the energy density radiated by the oscillating

dipole. To this end, we need the Poynting vector S(⃗ ⃗ r , t). From the electric and
magnetic fields E(⃗ ⃗ r , t) and B(⃗⃗ r, t) (which are functions of the spatial coordinates
and of time), the instantaneous Poynting vector S(⃗ ⃗ r, t) is calculated as follows,
⃗ r, t) = 1 E
S(⃗ ⃗ (⃗ ⃗ (⃗
r , t) × B r , t) . (5.103)
μ0
With the electromagnetic energy density
1 ⃗2 0 ⃗ 2
u(⃗r, t) = B (⃗ r , t) + E (⃗ r , t) , (5.104)
2μ0 2
which measures the amount of energy stored in the electromagnetic field per unit
volume, we have the Poynting theorem (1.106)
∇ ⃗ r , t) + ∂ u(⃗
⃗ ⋅ S(⃗ ⃗ r, t) ⋅ J⃗(⃗
r , t) + E(⃗ r, t) = 0 . (5.105)
∂t
Energy leaving the reference volume per unit time is obtained by integrating ∫V ∇ ⃗⋅
⃗
S(⃗ 3 ⃗ ⃗
r, t) ⋅ dA. The Poynting theorem states that energy leaving the
r , t)d r = ∫∂V S(⃗
reference volume, as described by the integral of the Poynting vector over the surface
normal, plus the work done on the charge in the reference volume by the electric
field, is accompanied by a loss in field energy within the reference volume. For
an oscillatory field described by a frequency component exp(−iωt), the physically
relevant quantity is the real part which is proportional to Re[exp(−iωt)] = cos(ω t).
The average value of cos(ω t) over a period of oscillation is just equal to 1/2. So, for
a component of angular frequency ω described by Eq. (5.79), the average Poynting
⃗ r)⟩ (over a period of oscillation) is given as
vector ⟨S(⃗
⃗ r)⟩ = 1 E⃗0 (⃗
⟨S(⃗ r) × B⃗ ∗ (⃗
0 r) . (5.106)
2μ0
The vector identity a ⃗ × (⃗b × c⃗) = ⃗b × (⃗ a ⋅ ⃗b) can be used to calculate this
a ⋅ c⃗) − c⃗ × (⃗
expression,
⃗ r)⟩ = 1 ⃗ ⃗ ∗ (⃗ 1 k2 k2 1
⟨S(⃗ E0 (⃗
r) × B 0 r) = [(r̂ × p⃗0 ) × r̂] × (r̂ × p⃗0 )
2μ0 2μ0 4π0 4π0 c r2
c k4 1 0 c k 4 1 2
= − (r̂ × ⃗
p 0 ) × [(r̂ × ⃗
p 0 ) × r̂] = r̂ (r̂ × p⃗0 )
2μ0 c (4π0 ) r
2 2 2 2 (4π0 ) r2
2
0 c k 4 1 0 c k 4 p⃗02 1
= p02 − (r̂ ⋅ p⃗0 )2 ) =
r̂ (⃗ r̂ (1 − cos θ2 ) , (5.107)
2 (4π0 ) r 2 2 2 (4π0 )2 r2
where cos θ = r̂ ⋅ p̂0 . The angular distribution of the average radiated power dPavg
per area dA is
dPavg (Ω) 0 c k 4 p⃗02 1
= r̂ ⋅ ⟨S(⃗r )⟩ = (1 − cos θ2 ) . (5.108)
dA 2 (4π0 )2 r2
The average intensity radiated parallel to the dipole vector is zero. The time aver-
aged power ⟨P ⟩ radiated by an oscillating electric dipole is obtained as
0 c k 4 p⃗02 1 c k 4 p⃗02
⟨P ⟩ = ∫ (1 − cos 2
θ) (r 2
dΩ) = . (5.109)
2 (4π0 )2 r2 3 4π0
The radiated power is proportional to the fourth power of the frequency of the
oscillation.
5.3.3 Exact Expression for the Radiating Dipole

Our exact result for the magnetic field generated by oscillating dipole (5.97) is given
as
⃗0 (⃗ k 2 exp (ik r) 1
B r) = (1 − ) (r̂ × p⃗0 ) . (5.110)
4π0 c r ik r
In Eq. (5.101), the electric field is obtained as
⃗0 (⃗ ic2 ⃗0 (⃗ k 2 exp (ik r)

E r) = ⃗ ×B
(∇ r )) ∼ [(r̂ × p⃗0 ) × r̂] , k r → ∞. (5.111)
ω 4π0 r
⃗0 (⃗
The exact expression for E r) remains to be derived. Certainly,
⃗0 (⃗ ic k 2 r⃗ exp (ik r) 1
E r) = ⃗ × ( × p⃗0 )
(∇ (1 − ))
k 4π0 c r r ikr
ik exp (ik r) 1
= ⃗ × (⃗
(∇ r × p⃗0 ) (1 − ))
4π0 r2 ikr
ik r⃗ d exp (ik r) 1
= ⃗ × ( × p⃗0 ) r
(∇r) ( (1 − ))
4π0 r dr r 2 ikr
ik exp (ik r) 1 ⃗ × (⃗
+ ( (1 − )) (∇ r × p⃗0 )) . (5.112)
4π0 r2 ik r
⃗ × (⃗b × c⃗) = ⃗b (⃗
According to the rule that a a ⋅ ⃗b),
a ⋅ c⃗) − c⃗ (⃗
⃗ × (⃗
(∇ r × p⃗0 )) = (∇ ⃗ ⋅ r⃗) = p⃗0 − 3 p⃗0 = −2⃗
⃗ ⋅ p⃗0 ) r⃗ − p⃗0 (∇ p0 .
⃗ operator acts on everything to the

We have taken into account the fact that the ∇
right. So,
⃗0 (⃗ ik r⃗ r⃗ d exp (ik r) 1
E r )= [ ×( × p⃗0 )] [r ( 2
(1 − ))]
4π0 r r dr r ikr
ik exp (ik r) 1
+ ( (1− p0 )
)) (−2⃗
4π0 r2 ikr
ik ik 3 3i ik exp (ik r) 1
= [r̂×(r̂× p⃗0 )]( − 2 − 3 ) exp (ik r)+ ( 2
(1− ))(−2⃗p0 )
4π0 r r kr 4π0 r ik r
ik ik ik 3 3i
= [r̂×(r̂× p⃗0 )] exp (ik r)+ [r̂×(r̂× p⃗0 )] (− 2 − 3 ) exp (ik r)
4π0 r 4π0 r kr
ik exp (ik r) 1
+ ( (1− )) (−2⃗
p0 ) . (5.113)
4π0 r2 ikr
Finally,
⃗0 (⃗ k2 exp (ik r) 1 3ik 3

E r) = [(r̂ × p⃗0 ) × r̂] + [r̂ (r̂ ⋅ p⃗0 − p⃗0 )] (− 2 + 3 ) exp (ik r)
4π0 r 4π0 r r
1 exp (ik r) 1
+ ( (ik − )) (−2⃗ p0 ) . (5.114)
4π0 r2 r
The first term is the leading term. The remaining terms have the structure of a
quadrupole term component when projected onto the dipole vector p⃗0 ,
⃗0 (⃗ k2 exp (ik r) 1 1 ik
E r) = [(r̂ × p⃗0 ) × r̂] + [3r̂ (r̂ ⋅ p⃗0 − 3⃗p0 )] ( 3 − 2 ) exp (ik r)
4π0 r 4π0 r r
1 ik 1
+ exp (ik r) (− 2 + 3 ) (+2⃗ p0 )
4π0 r r
k2 exp (ik r) 1 1 ik
= [(r̂ × p⃗0 ) × r̂] + [3r̂ (r̂ ⋅ p⃗0 − 3⃗p0 )] ( 3 − 2 ) exp (ik r)
4π0 r 4π0 r r
1 1 ik
+ (2⃗p0 ) ( 3 − 2 ) exp (ik r)
4π0 r r
k 2
exp (ik r) 1 1 ik
= [(r̂ × p⃗0 ) × r̂] + [3r̂ (r̂ ⋅ p⃗0 − p⃗0 )] ( 3 − 2 ) exp (ik r) .
4π0 r 4π0 r r
(5.115)
Finally, we have found the exact result for the electric field radiated from the dipole,
valid for all r,
⃗0 (⃗ k 2 exp (ik r) i k exp (ik r) 1

E r) = [(r̂ × p⃗0 ) × r̂] − (1 − ) [3r̂ (r̂ ⋅ p⃗0 − p⃗0 )] .
4π0 r 4π0 r2 ikr
(5.116)
In the near field, kr ≪ 1, we can replace exp (ik r) → 1 and find for the magnetic
field in the near zone,
⃗0 (⃗ ik
B r) ∼ (r̂ × p⃗0 ) , kr → 0, (5.117)
4π0 c r2
and for the electric field in the near zone,
⃗0 (⃗ 1
E r) ∼ [3r̂ (r̂ ⋅ p⃗0 − p⃗0 )] , kr → 0. (5.118)
4π0 r3
Even for dipole radiation, the derivation of the exact result (5.116) for the electric
field is a nontrivial exercise, and it is helpful to include all intermediate steps.
5.4 Tensor Green Function
5.4.1 Clebsch–Gordan Coefficients: Motivation

Angular momentum algebra is connected to the so-called Clebsch–Gordan or vector
addition coefficients. These coefficients are of rather universal applicability over
wide ranges of physical theory. One uses them in order to find the expansion
coefficients of a tensor of higher rank as it is composed out of elements of tensors of
lower rank, hence the name “vector” addition coefficients (while “tensor” addition
might otherwise be a more precise formulation).
One of the most important symmetries in physics concerns the group of rota-
tions, or, the special orthogonal group in three dimensions, called SO(3). We know
that the scalar product u ⃗ ⋅ v⃗ of two vectors is invariant under rotations. The vector
product u ⃗ × v⃗ transforms as a (pseudo-)vector itself. We recall that a pseudo-vector,
in contrast to a vector, conserves its sign under parity, u ⃗ × v⃗ → (−⃗
u) × (−⃗
v ). Vec-
tors do not mix with scalars under rotations. The vectors are of rank one, whereas
scalars are tensors of rank zero. Furthermore, from the tensor product of two vec-
tors, we can extract a third quantity which is a quadrupole tensor of rank two,
which also does not mix with tensors of different rank under rotations. From the
tensor product of two vectors, one may thus extract tensors of rank zero, one, and
two. The components of these “constructed” tensors are linear combinations of
products of the components of the vectors. The corresponding formalism, in the
spherical basis, involves the vector addition, or Clebsch–Gordan, coefficients.
The procedure is illustrated most effectively by way of an example. We consider
a second-rank tensor from the tensor product of two vectors,
⎛ ux vx ux vy ux vz ⎞
⃗ ⊗ v⃗ = ⎜ uy vx uy vy uy vz ⎟ .
=u (5.119)
⎝ uz vx uz vy uz vz ⎠
We can decompose as follows,
= ∣=0 + ∣=1 + ∣=2 , (5.120)
trc( )
∣=0 =
3
3×3 , trc( ) = ux vx + uy vy + uz vz , (5.121)
⎛ 0 1
2
(ux vy − uy vx ) − 12 (uz vx − ux vz ) ⎞
1 ⎜ ⎟
∣=1 = ( − T
) = ⎜ − 12 (ux vy − uy vx ) 0 1
(uy vz − uz vy ) ⎟ .
2 ⎜ 2 ⎟
⎝ 12 (uz vx − ux vz ) − 12 (uy vz − uz vy ) 0 ⎠
(5.122)
The ( = 0)-component is invariant under rotations. The entries of the matrix ∣=1
can be identified as the components of the vector product of u ⃗ and v⃗, with details
to be discussed below. Finally, the ( = 2)-component reads as
trc( )
∣=2 =
1
2
( + T) −
3
3×3 (5.123)
2u v − uy vy − uz vz ux vy + uy vx ux vz + uz vx
⎛ x x ⎞
⎜ 3 2 2 ⎟
⎜ ux vy + uy vx 2uy vy − ux vx − uz vz uy vz + uz vy ⎟
⎜ ⎟
=⎜ ⎟.
⎜ 2 3 2 ⎟
⎜ ⎟
⎜ ux vz + uz vx uy vz + uz vy 2uz vz − ux vx − uy vy ⎟
⎝ ⎠
2 2 3
As already anticipated, and with reference to Eq. (1.23), we can order the non-
vanishing components of the antisymmetric tensor ∣=1 into a vector, namely, the
⃗ and v⃗,
vector product of u
⎛ uy vz − uz vy ⎞
⃗ × v⃗ = ⎜ uz vx − ux vz ⎟ ,
u (⃗
u × v⃗)i = ijk uj vk , (5.124)
⎝ ux vy − uy vx ⎠
where a summation over j and k is understood by the Einstein summation conven-

tion [see Eq. (1.45)]. The fact that ∣=1 transforms as a (pseudo-)vector shows
that it is possible to construct a vector whose components themselves are products
of vector components. In Cartesian components, the “coupling coefficients” can
directly be read off from Eq. (5.124), in the sense that the components of a tensor
of rank one (of the vector product) are obtained by multiplying the products uj vk
of the components of the vectors u ⃗ and v⃗ by the coupling coefficient ijk .
However, the canonical formalism employs the spherical basis. We recall from
Eq. (2.30) that the z component of the angular momentum operator has a particu-
larly simple form in spherical coordinates [see Eq. (2.30c)],
∂ ∂ ∂
Lz = −i (x −y ) = −i . (5.125)
∂y ∂x ∂ϕ
The spherical basis of vector components is chosen to generate an explicit ϕ depen-
dence of the form exp(imϕ), i.e., it consists of eigenfunctions of Lz . In the spherical
basis, the components are denoted as x+1 , x0 and x−1 ; they read as follows,
√
1 1 4π
x+1 = − √ (x + i y) = − √ ∣⃗ r ∣ sin θ e =
iϕ
∣⃗
r ∣ Y11 (θ, ϕ) , (5.126a)
2 2 3
√
4π
x0 = z = ∣⃗r ∣ cos θ = ∣⃗
r ∣ Y10 (θ, ϕ) , (5.126b)
3
√
1 1 −iϕ 4π
x−1 = √ (x − i y) = √ ∣⃗ r ∣ sin θ e = r ∣ Y1−1 (θ, ϕ) ,
∣⃗ (5.126c)
2 2 3
√
where we have allowed ourselves the luxury to write r = ∣⃗ r ∣ = x2 + y 2 + z 2 explicitly.
The emergence of phase factors of the form exp(i m ϕ) with m = −1, 0, 1 is evident.
The spherical components are complemented by spherical basis vectors as follows,
1 1
e⃗+1 = − √ (êx + i êy ) , e⃗0 = êz , e⃗−1 = √ (êx − i êy ) . (5.127)
2 2
The coordinate vector can easily be expanded into the spherical basis,
1 1
r⃗ = ∑ xq e⃗∗q = ∑ (−1)q x−q e⃗q , e⃗∗q = (−1)q e⃗−q . (5.128)
q=−1 q=−1
The spherical basis vectors are normalized as follows,

e⃗q ⋅ e⃗q′ = (−1)q δq −q′ , e⃗q ⋅ e⃗∗q′ = δq q′ . (5.129)
The sum over q runs over the indices q = −1, 0, 1. For absolute clarity, we should
mention that the latter Kronecker symbol in Eq. (5.128) is to be understood as
δq,−q′ , i.e., q has to be equal to −q ′ in order for the Kronecker symbol to be equal
to unity rather than zero. Components can be extracted by calculating the scalar
product of the coordinate vector r⃗ with a spherical basis vector,
1 1
r⃗ ⋅ e⃗q = ∑ xq′ e⃗∗q′ ⋅ e⃗q = ∑ xq′ δq′ q = xq . (5.130)
q′ =−1 q′ =−1
The paradigm of the vector coupling, or Clebsch–Gordan, coefficients, is that

one obtains a tensor component of magnetic quantum number m for a tensor of
rank j by coupling two tensors of rank j1 and j2 as follows,
j1 j2
jm
v(jm) = ∑ ∑ Cj1 m1 j2 m2 u(j1 m1 ) u(j2 m2 ) . (5.131)
m1 =−j1 m2 =−j2
Here, u(j1 m1 ) and u(j2 m2 ) are two distinct vectors. In our example, we couple
two tensors of rank one (the vectors u ⃗ and v⃗ with spherical components u+1 , u0
and u−1 , as well as v+1 , v0 and v−1 ), to a tensor of rank one, which is the vector
product w ⃗=u⃗ × v⃗. So, we have j1 = j2 = j = 1 for our example, which is given by
Eq. (5.124) in Cartesian coordinates. The Clebsch–Gordan coefficients are tabu-
lated and nowadays implemented in most modern computer algebra systems. Using
tabulated values, one finds
1q
wq = ∑ C1q ′ 1q ′′ uq ′ vq ′′ , (5.132a)
q′ q′′
u+1 v0 − u0 v+1 i 1
w+1 = √ = [ √ ] {− √ [(⃗ u × v⃗)y ]} ,
u × v⃗)x + i (⃗ (5.132b)
2 2 2
u+1 v−1 − u−1 v+1 i
w0 = √ = [ √ ] (⃗
u × v⃗)z , (5.132c)
2 2
u0 v−1 − u−1 v0 i 1
w−1 = √ = [ √ ] { √ [(⃗ u × v⃗)y ]} ,
u × v⃗)x − i (⃗ (5.132d)
2 2 2
where the subscripts x, y, z denote the Cartesian components of the vector product,
i.e., (u × v)z = ux vy − uy vx and further by cyclic permutation. A comparison of
Eq. (5.124) with Eq. (5.126) shows that the components of w ⃗ are √
equal to those
obtained by writing u ⃗ × v⃗ in the spherical basis, up to a prefactor i/ 2, i.e.,
1
i
⃗ = ∑ (−1)q w−q e⃗q = √ (⃗
w u × v⃗) . (5.133)
q=−1 2
One can form two further linear combinations of the spherical basis vectors e⃗q
which are of interest,
√ 1 √ 1 1
e⃗q × e⃗q′ = i 2 ∑ C1q1q
1λ
⃗λ ,
′ e r⃗ = − 3 ∑ ∑ C1q1q00
⃗q′ .
′ xq e (5.134)
λ=−1 q=−1 q′ =−1
The first of these illustrates that the vector product of two basis vectors in the
spherical basis again is a vector; the second clarifies that the coordinate vector r⃗
actually is a scalar under rotations, obtained as the scalar combination (tensor of

rank zero, component number zero) composed out of the spherical coordinates xq
and the spherical basis vectors e⃗q . Of course, the vector composed of the components
xq is not a scalar. However, the scalar product of the vector composed of the xq with
the vector composed of the e⃗q constitutes a physical vector r⃗ which does not change
just upon a change of the reference frame [see Eq. (5.128)]. Upon rotation, this scalar
product is equal to the coordinate vector obtained using the new coordinates x′q , but
multiplied with the rotated basis vectors e⃗′q , which leaves the physical coordinate
vector r⃗ invariant. This corresponds to a passive interpretation of the rotation.
Let us now investigate the ( = 2)-component given in Eq. (5.123). It is sym-
metric and traceless and has five independent components. The counting works as
follows: We have five independent components, because the matrix is symmetric and
traceless. This leaves three off-diagonal and two diagonal components to be deter-
mined; the third component on the diagonal is fixed by the condition trc ( ∣=2 ) = 0.
A scalar ( = 0) has one component, a vector ( = 1) has three magnetic components
xm (which depend on ϕ as exp(imϕ) with m = −1, 0, 1), and a quadrupole tensor has
five magnetic components [which depend on ϕ as exp(imϕ) with m = −2, −1, 0, 1, 2].
The generalization calls for 2 + 1 magnetic components for a tensor of rank .
2q
Again, using tabulated values for Clebsch–Gordan coefficients of the form C1q ′ 1q ′′ ,
with q ′ , q ′′ = −1, 0, 1 and q = −2, −1, 0, 1, 2, we have
2q
tq = ∑ C1q ′ 1q ′′ uq ′ vq ′′ , (5.135a)
q′ q′′
u+1 v0 + u0 v+1
t+2 = u+1 v+1 , t+1 = √ , (5.135b)
2
3 u0 v0 − u⃗ ⋅ v⃗
t0 = √ , (5.135c)
6
u−1 v0 + u0 v−1
t−1 = √ , t−2 = u−1 v−1 , (5.135d)
2
where the spherical components u−1 , u0 and u+1 are defined in Eq. (5.126). If the
operators L ⃗ 2 address the vectors u
⃗ 1 and L ⃗ and v⃗ separately, then the functions tq
are eigenfunctions of the operator L⃗ 2 = (L
⃗1 + L
⃗ 2 )2 with eigenvalue L
⃗ 2 → ( + 1) = 6
and of the z component Lz = L1z + L2z with an eigenvalue q of Lz .
5.4.2 Vector Additions and Vector Spherical Harmonics
Vector spherical harmonics are obtained upon adding the “spin” of the photon,
namely, the spherical basis vectors which are used in the expansion of the vector
potential, to the spherical harmonics which represent the “orbital” angular momen-
tum of the photon. Photons (light particles) are spin-1 objects. The spherical basis
vectors are components of a tensor of rank one. To this tensor we add, vectori-
ally, the orbital angular momentum of the photon, as manifest in the spherical
harmonic. Hence, in view of Eq. (5.131), the vector spherical harmonic is given as
1
Y⃗jμ

(r̂) = ∑ jμ
∑ Cm 1q Ym (θ, ϕ) e⃗q . (5.136)
m=− q=−1
Because neither the total angular momentum quantum number j nor the orbital
angular momentum can be negative, the vector spherical harmonics with j = −1
and = −1 vanish; this observation comes in handy in regard to a number of
summations discussed in the following. From Eq. (5.127), we recall the basis vectors
in the spherical basis,
1 1
2 2
jμ
The Clebsch–Gordan coefficients Cm 1q assemble a tensor of angular symmetry jμ
from a spherical harmonic of angular symmetry m and a spherical basis vector of
angular symmetry 1q. We prefer the above notation for the vector spherical har-
monic; the magnetic projection μ may assume values from −j to j. The superscript
reminds us of the orbital momentum which was used in the construction of the
vector spherical harmonic. The spin of the photon (equal to one) is added to the
orbital angular momentum; hence the total angular momentum j can differ from
by at most unity; otherwise the vector spherical harmonic vanishes.
The spin operators of the photon are given by the matrices i (i = 1, 2, 3),
( k )ij = −i kij , (5.138)
where ijk is the Levi–Cività tensor [see Eq. (1.23)]. The explicit representation for
k = 1, 2, 3 reads as
⎛0 0 0 ⎞ ⎛ 0 0 i⎞ ⎛ 0 −i 0⎞
1 = ⎜ 0 0 −i ⎟ 2 = ⎜ 0 0 0 ⎟ , 3 = ⎜ i 0 0⎟ . (5.139)
⎝0 i 0 ⎠ ⎝ −i 0 0 ⎠ ⎝0 0 0⎠
The matrix given above is identified as the lower right 2×2 submatrix of 1 , and

also as the upper left 2 × 2 submatrix of 3 . These matrices and the corresponding
components of the angular momentum vector fulfill the algebraic relations
[ i , j ] = i ijk k , (5.140)
where the Einstein summation convention is used for the sum over k = 1, 2, 3 on the
right-hand side. The vector square of the matrices is
⎛2 0 0⎞
⃗ 2 = ⎜ 0 2 0 ⎟ = S (S + 1) 3×3 , S = 1, (5.141)
⎝0 0 2⎠
demonstrating that the photon is a spin-1 particle (with S = 1). The total angular
momentum operator of the photon is given by
J⃗ = L
⃗ 3×3 + ⃗ , (5.142)
where we are pedantic in multiplying the orbital angular momentum operator by the
three-dimensional unit matrix, implying that, say, the z component Lz A⃗ = Lz ⋅ A⃗
⃗ i.e., on all of the components of A⃗ separately. The z
acts on the entire vector A,
component of J acts on a vector-valued function as Jz V⃗ (θ, ϕ) = Lz V⃗ (θ, ϕ) + z ⋅
⃗
V⃗ (θ, ϕ). The vector spherical harmonics have the properties
J⃗2 Y⃗jμ

(θ, ϕ) = j(j + 1) Y⃗jμ

(θ, ϕ) , (5.143a)
⃗ 2 Y⃗ (θ, ϕ) = ( + 1) Y⃗ (θ, ϕ) ,
L (5.143b)
jμ jμ
Jz Y⃗jμ

(θ, ϕ) = μ Y⃗jμ

(θ, ϕ) . (5.143c)
The orthonormality relations are
′
∫ dΩ Y⃗jμ (θ, ϕ) ⋅ Y⃗j ′ μ′ (θ, ϕ) = δjj ′ δ′ δμμ′ ,
∗
(5.144a)
j
2j + 1
∑ ∫ dΩ Y⃗jμ (θ, ϕ) ⊗ Y⃗jμ (θ, ϕ) =
∗
3
3×3 , (5.144b)
μ=−j
where we assume that the vector spherical harmonic is nonvanishing, i.e., ∣j − ∣ ≤ 1.

In the second sum, j and are held constant; they just have to be the same for
both vector spherical harmonics but are not summed over. For given j, there are
three possible values of , namely, = j − 1, j, j + 1; summing over these, one obtains
a factor (2j + 1) instead of (2j + 1)/3 on the right-hand side.
There is also a completeness relation,
j
3×3 ,
+1
∑ Y⃗jμ (θ, ϕ) ⊗ Y⃗jμ (θ , ϕ ) = ∑ Ym (θ, ϕ) Ym (θ , ϕ )
∗ ′ ′ ∗ ′ ′
∑ (5.145)
j=−1 μ=−j m=−
which implies that

∞ j
3×3 .
+1
1
∑ Y⃗jμ (θ, ϕ) ⊗ Y⃗jμ (θ , ϕ ) =
∗ ′ ′
∑ ∑

δ(θ − θ′ ) δ(ϕ − ϕ′ ) (5.146)
=0 j=−1 μ=−j sin θ
The relation analogous to Eq. (2.56a) is
Y⃗jμ
∗
(θ, ϕ) = (−1)+1+j (−1)μ Y⃗j−μ (θ, ϕ) . (5.147)
Explicit representations are given as follows,
1
Y⃗jjμ (θ, ϕ) = √ ⃗ jμ (θ, ϕ) ,
LY (5.148a)
j(j + 1)
1
Y⃗jj−1
μ (θ, ϕ) = √
⃗ Yjμ (θ, ϕ)
(j r̂ + r ∇)
j (2j + 1)
1 ⃗ − j r̂) Yjμ (θ, ϕ) ,
= −√ (i r̂ × L (5.148b)
j (2j + 1)
1
Y⃗jj+1
μ (θ, ϕ) = − √
⃗ Yjμ (θ, ϕ)
((j + 1) r̂ − r ∇)
(j + 1) (2j + 1)
1 ⃗ + (j + 1) r̂) Yjμ (θ, ϕ) .
= −√ (i r̂ × L (5.148c)
(j + 1) (2j + 1)
Here, μ can take on the values μ = −j, . . . , j. Again, we emphasize that vector
spherical harmonics with ∣j − ∣ > 1 vanish.
The two equivalent representations of the vector spherical harmonics can recon-
ciled with each other on the basis of the operator identity
⃗ f (θ, ϕ) = −r ∇f
(i r̂ × L) ⃗ (θ, ϕ) , (5.149)
which is valid for any test function f that only depends on the angular variables.
We can write a representation of the vector spherical harmonics in terms of
Clebsch–Gordan coefficients. For example, we have in the case = j,
Y⃗jμ
j
(θ, ϕ) = Cjjμμ−1 11 e⃗+1 Yj μ−1 (θ, ϕ) + Cjμ
jμ
⃗0 Yj μ (θ, ϕ) + Cjjμμ+1 1 −1 e⃗−1 Yj μ+1 (θ, ϕ) .
10 e
(5.150)
Using known formulas for the Clebsch–Gordan coefficients, the vector spherical
harmonics with = j find the following representation,
B
D (j + μ) (j − μ + 1) μ
Y⃗jμ
j
(θ, ϕ) = − DE Yj,μ−1 (θ, ϕ) e⃗+1 + √ Yj,μ (θ, ϕ) e⃗0
2j(j + 1) j(j + 1)
B
D (j − μ) (j + μ + 1)
+D E Yj,μ+1 (θ, ϕ) e⃗−1 . (5.151a)
2j(j + 1)
For the case = j − 1, one has

B
D (j + μ − 1) (j + μ)
Y⃗jμ
j−1
(θ, ϕ) = D
E Yj−1,μ−1 (θ, ϕ) e⃗+1
2j(2j − 1)
B
D (j − μ) (j + μ)
+D
E Yj−1,μ (θ, ϕ) e⃗0
j(2j − 1)
B
D (j − μ − 1) (j − μ)
+D
E Yj−1,μ+1 (θ, ϕ) e⃗−1 . (5.151b)
2j(2j − 1)
Finally, for the case = j + 1, one has

B
D (j − μ + 1) (j − μ + 2)
Y⃗jμ
j+1
(θ, ϕ) = D
E Yj+1,μ−1 (θ, ϕ) e⃗+1
2(j + 1)(2j + 3)
B
D (j − μ + 1) (j + μ + 1)
−D
E Yj+1,μ (θ, ϕ) e⃗0
(j + 1)(2j + 3)
B
D (j + μ + 2) (j + μ + 1)
+D
E Yj+1,μ+1 (θ, ϕ) e⃗−1 . (5.151c)
2(j + 1)(2j + 3)
Here, one easily discerns the addition of the magnetic quantum number, of the
spherical harmonic and the spherical basis vector, to the total magnetic projection
of the vector spherical harmonic.
5.4.3 Scalar Helmholtz Green Function and Scalar Potential

We recall that the Helmholtz Green function is given as follows [see Eq. (5.20)],
r − r⃗′ ∣)
exp(ik ∣⃗
GR (k, r⃗ − r⃗′ ) = . (5.152)
r − r⃗′ ∣
4π0 ∣⃗
This Green function couples the scalar potential to the source, according to
Eq. (5.77),
ω
r , t) = e−i ω t Φ0 (⃗
Φ(⃗ r) , r ) = ∫ d3 r′ GR ( , r⃗ − r⃗′ ) ρ0 (⃗
Φ0 (⃗ r′ ) . (5.153)
c
For definiteness, we also recall the angular decomposition of the Green function for
the Helmholtz equation according to Eq. (5.72),
1 (1)
GR (k, r⃗ − r⃗′ ) = ∗ ′
∑ i k j (k r< ) h (k r> ) Ym (θ, ϕ) Ym (r̂ ) . (5.154)
0 ,m
Outside of the charge distribution, the scalar potential is thus given by
ik ∞ (1)
Φ0 (⃗
r) = ∑ ∑ pm h (k r) Ym (θ, ϕ) , (5.155)
0 =0 m=−
with
∗ k ∗
pm = ∫ d3 r j (k r) ρ(⃗
r ) Ym (θ, ϕ) ≈ r) Ym
d3 r r ρ(⃗ (θ, ϕ) .
(2 + 1)!! ∫
(5.156)
The pm generalize the multipole components of a static charge distribution, defined
according to Eq. (2.97), for a dynamical process, namely, the emission of radiation.
In the notation, we suppress their dependence on the wave number k. Indeed, for
k → 0 (zero-frequency radiation), the leading term in the expansion of pm according
to Eq. (5.156) is proportional to qm , a fact which is obvious from Eq. (5.30).
5.4.4 Tensor Helmholtz Green Function and Vector Potential

The tensor Helmholtz Green function is obtained from the Helmholtz Green function
by a multiplication with the unit matrix,
R(k, r⃗ − r⃗′ ) = 3×3 GR (k, r⃗ − r⃗′ ) . (5.157)
It enters the equation that couples the vector potential to the source, Eq. (5.78),
which we write as follows,
⃗ r , t) = e−iωt A⃗0 (⃗
A(⃗ r) , A⃗0 (⃗
c
1 ω
c

r ) = ∫ d3 r′ 2 R ( , r⃗ − r⃗′ ) ⋅ J⃗0 (⃗
r′ ) . (5.158)
Note the explicit matrix product of the tensor Green function and the current
density, which differentiates Eq. (5.158) from (5.78).
In order to proceed with the analysis of the tensor Green function, we first need
to write a tensorial decomposition. The unit matrix in Eq. (5.157), which describes
the spin of the photon, needs to be incorporated into the analysis. Of course, the
hope is that once we form the tensor product of all the vector spherical harmonics
pertaining to the same orbital angular momentum , we would somehow recover

∗
the angular structure in Eq. (5.72), namely, ∑m Ym (θ, ϕ) Ym (θ′ , ϕ′ ), multiplied

by the unit matrix 3×3 . Indeed, we recall Eq. (5.145),
j
3×3 .
+1
∑ Y⃗jμ (θ, ϕ) ⊗ Y⃗jμ (θ , ϕ ) = ∑ Ym (θ, ϕ) Ym (θ , ϕ )
∗ ′ ′ ∗ ′ ′
∑ (5.159)
j=−1 μ=−j m=−
For the case = 0, we recall the observation reported in the text following
Eq. (5.136). Essentially, in Eq. (5.145), for given , one sums over the possible
values of j, namely, j = − 1, , + 1, and then over the possible magnetic projections
μ, and obtains an angular structure which is familiar from Eq. (5.72). Here, the ten-
∗
sor product is the one which transforms the two vectors Ym (θ, ϕ) and Ym (θ′ , ϕ′ )
into a matrix; pedantically, one might otherwise have indicated the transpose of the
latter vector.
The angular decomposition of the tensor Green function (5.72) for the vector
Helmholtz equation can thus be given in terms of the vector spherical harmonics,
′
R(k, r⃗ − r⃗′ ) = exp4π(i k∣⃗r∣⃗r−−r⃗r⃗′ ∣ ∣) 3×3
0
∞ j j+1
ik
j (k r< ) h (k r> ) Y⃗jμ (θ, ϕ) ⊗ Y⃗jμ
(1) ∗ ′
= ∑ ∑ ∑

(θ , ϕ′ )
0 j=0 μ=−j =j−1
∞ j
ik
∑ ∑ (jj−1 (k r< ) hj−1 (k r> ) Y⃗jμ (θ, ϕ) ⊗ Y⃗jμ (θ , ϕ )
(1) j−1 j−1∗ ′ ′
=
0 j=0 μ=−j
+ jj (k r< ) hj (k r> ) Y⃗jμ (θ, ϕ) ⊗ Y⃗jμ
(1) j∗ ′
j
(θ , ϕ′ )
+ jj+1 (k r< ) hj+1 (k r> ) Y⃗jμ (θ, ϕ) ⊗ Y⃗jμ
(1) j+1∗ ′
j+1
(θ , ϕ′ )) . (5.160)
So, outside of the charge distribution, the vector potential is thus given by
A⃗0 (⃗
1

r ) = 2 ∫ d3 r′ R (k, r⃗ − r⃗′ ) ⋅ J⃗0 (⃗
c
r′ )
r − r⃗′ ∣)
exp (i k ∣⃗
= ∫ d3 r′
r − r⃗′ ∣
4π0 c2 ∣⃗
⃗ r′ )
3×3 ⋅ J0 (⃗
∞ j j+1
pjμ h (k r) Y⃗jμ
(1)
= i μ0 k ∑ ∑ ∑

(θ, ϕ) , (5.161)
j=0 μ=−j =j−1
with
k
pjμ = ∫ d3 r j (k r) J⃗0 (⃗
r ) ⋅ Y⃗jμ
∗
(θ, ϕ) ≈ ∫ d r r J⃗0 (⃗
3
r ) ⋅ Y⃗jμ
∗
(θ, ϕ) . (5.162)
(2 + 1)!!
The advantage of Eq. (5.161) over Eq. (5.87) lies in the fact that the vector structure
of the radiated vector potential is resolved and the spin of the photon is incorporated
into the formalism.
But we are not quite there yet. Namely, the decomposition (5.161) does not
clearly separate the longitudinal and transverse components to the electric and
magnetic fields generated by the vector potential. An alternative decomposition,
which accomplishes a separation into electric and magnetic multipole radiation,

reads as follows,
∞ j
R (k, r⃗ − r⃗′ ) = ik ∑ ⃗ (1) (k, r⃗> ) ⊗ M
∑ (M jμ
⃗ (0)∗ (k, r⃗< )
jμ
0 j=0 μ=−j
+N
(1) ⃗
⃗ (k, r⃗> ) ⊗ N (0)∗⃗ (k, r⃗> ) ⊗ L
⃗(1) (0)∗
jμ jμ (k, r
⃗< ) + L jμ jμ (k, r
⃗< )) . (5.163)
From the above discussion, it is clear that takes the role otherwise taken by j,
because we have added the spin of the photon to its orbital angular momentum. We
(K) (K) (K)
shall later see that the angular dependence of the functions Mjμ , Njμ , and Ljμ ,
is given by vector spherical harmonics with a total angular momentum number j.
The notation is to be explained in the following. For the magnetic (M ), electric
(N ), and longitudinal (L) multipole moments, we have
⃗ (K) (k, r⃗) = √ 1
M
(K)
fj (k r) LY⃗ jμ (θ, ϕ) , (5.164a)
jμ
j(j + 1)
⃗ (K) (k, r⃗) = i ∇
N ⃗ ×M ⃗ (K)(k, r⃗) , (5.164b)
jμ jμ
k
1
⃗ (K) (k, r⃗) = ∇
L jμ
⃗ (fj(K) (k r) Yjμ (θ, ϕ)) , (5.164c)
k
and conversely
⃗ (K) (k, r⃗) = − i ∇
M ⃗ (K) (k, r⃗) .
⃗ ×N (5.165)
jμ jμ
k
⃗ (K)
The definition of M 00 (k, r
⃗) needs to be clarified for Eq. (5.164). In principle, we
are dividing zero by zero √
because the application of the L⃗ operator to Y00 leads to
zero, but the prefactor 1/ j(j + 1) has a zero in the denominator. The solution is
to allow for an infinitesimal displacement of the angular momentum j → j + ε before
applying the L⃗ operator and then letting ε → 0 at the end of the calculation. This
clarifies that
M ⃗ (K) (k, r⃗) = ⃗
⃗ (K) (k, r⃗) = N 0. (5.166)
00 00
(K)
The fj (k r) with K = −1, 0, 1 are Bessel and Hankel functions, as follows,
(0) (1) (1) (−1) (2)
fj (k r) = jj (k r) , fj (k r) = hj (k r) , fj (k r) = hj (k r) . (5.167)
The vector multipole decomposition involves the following moments,
mjμ = ∫ d3 r′ J⃗0 (⃗ ⃗ (0)∗ (θ′ , ϕ′ ) ,
r′ ) ⋅ M jμ (5.168a)
njμ = ∫ d3 r′ J⃗0 (⃗ ⃗ (0)∗

r′ ) ⋅ N ′ ′
jμ (θ , ϕ ) , (5.168b)
ljμ = ∫ d3 r′ J⃗0 (⃗ ⃗ (0)∗ (θ′ , ϕ′ ) ,

r′ ) ⋅ L jμ (5.168c)
(K)
whose dependence on k is suppressed. (This dependence is due to the fj func-
⃗
tions defined in Eq. (5.167), which enter the M
(0)∗ ⃗ (0)∗
, N ⃗
, and L
(0)∗
, according
jμ jμ jμ
to Eq. (5.164).) We note that the curl operator in Eq. (5.168b) admixes Bessel
functions jj−1 and jj+1 to jj . The approximation
(0) (k r)j
fj (k r) = jj (k r) ≈ (5.169)
(2j + 1)!!
is otherwise applicable for a localized source, in much the same way as in Eqs. (5.156)
and (5.162). Finally, the decomposition of the vector potential reads as
∞ j
A⃗0 (⃗ ⃗ (1) (k, r⃗) + njμ N
r ) = ik μ0 ∑ ∑ (mjμ M jμ
⃗ (1) (k, r⃗) + ljμ L
jμ
⃗ (1) (k, r⃗)) .
jμ (5.170)
j=0 μ=−j
Here, the mjμ are the magnetic multipole moments, the njμ are the electric mul-
tipole moments, and the ljμ are longitudinal multipole moments which do not
contribute to the fields.
A (perhaps) more systematic expansion writes the M ⃗ (K) (k, r⃗), and
⃗ (K)(k, r⃗), N
jμ jμ
⃗ (K) (k, r⃗) in terms of the vector spherical harmonics with = j − 1, j, j + 1, but
L jμ
definite j,
⃗ (K) (k, r⃗) = f (K)(k r) Y⃗ j (θ, ϕ) ,
M (5.171a)
jμ j jμ
√ √
⃗ (K) (k, r⃗) = − j + 1 (K) j
(k r) Y⃗jj−1 (k r) Y⃗jj+1
(K)
N f μ (θ, ϕ) + f μ (θ, ϕ) ,
jμ
2j + 1 j−1 2j + 1 j+1
(5.171b)
√ √
⃗ (K) (k, r⃗) = j j + 1 (K)
(k r) Y⃗jj−1 (k r) Y⃗jj+1
(K)
L f μ (θ, ϕ) + f μ (θ, ϕ) .
jμ
2j + 1 j−1 2j + 1 j+1
(5.171c)
⃗ (K)
From Eq. (5.171b), one might think that N 00 (k, r
⃗) could incur a nonvanishing
contribution proportional to Y⃗00 (θ, ϕ), but the prefactor of this term vanishes.
1
Equation (5.171c) teaches us that the term proportional to Y⃗00 1

(θ, ϕ) is part of
the longitudinal component of the vector potential, which does not contribute to
the electric and magnetic fields. There is no such thing as a photon with vanishing
total angular momentum j = μ = 0. Yet another representation is as follows,
⃗ (K) (k, r⃗) = f (K) (k r) Y⃗ j (θ, ϕ) ,
M (5.172a)
jμ j jμ
⎧
⎪ d[kr fj(K) (k r)] √ ⎫
⎪
⃗ (K) (k, r⃗) = 1 ⎪
N ⎨ [ir̂ × ⃗ j (θ, ϕ)] − r̂ j(j + 1)f (K) (k r) Yjμ (θ, ϕ)⎪
Y ⎬,
jμ
kr ⎪
⎪ d(kr) jμ j ⎪
⎪
⎩ ⎭
(5.172b)
(K)
d[fj (k r)] √ 1 (K)
⃗ (K) (k, r⃗) = r̂
L Yjμ (θ, ϕ) − j(j + 1) f (k r) (ir̂ × Y⃗jμ
j
(θ, ϕ)) .
jμ
d(k r) kr j
(5.172c)
Let us now try to interpret the contributions M ⃗ (K) (k, r⃗), N

⃗ (K) (k, r⃗), and
jμ jμ
⃗ (K) (k, r⃗) physically. From Eq. (5.164c), we infer that L
L ⃗ (K)(k, r⃗) is the “longi-
jμ jμ
tudinal” solution constructed by taking the gradient of a scalar solution of the
Helmholtz equation. The magnetic and electric fields are obtained from the vector
potential, via the equations
⃗0 (⃗ ic2
B ⃗ × A⃗0 (⃗
r) = ∇ r) , ⃗0 (⃗
E r) = ∇ ⃗0 (⃗
⃗ ×B r) , (5.173)
ω
⃗ (K)
in the source-free region. Because L jμ (k, r
⃗) is a gradient of a scalar solution of
the Helmholtz equation, its curl vanishes. Furthermore, in view of Eq. (5.173), the
fields generated by the longitudinal components of the vector potential vanish. We
⃗ r, t) = −∇Φ(⃗
note that in view of the relation E(⃗ ⃗ r, t), the electric field can
⃗ r , t) − ∂t A(⃗
alternatively be calculated as
E⃗0 (⃗
r ) = −∇Φ r ) + iω A⃗0 (⃗
⃗ 0 (⃗ r) . (5.174)
We note that ⃗ (K)(k, r⃗)
L ⃗ 0 (⃗
is longitudinal, just like ∇Φ r ) (finally, it is just a gradi-
jμ
ent). We shall thus later have to show that in calculating E⃗0 (⃗ r ), the gradient of Φ0
given in Eq. (5.155) actually cancels against the time derivative of the longitudinal
contribution proportional to iω L ⃗ (K) (k, r⃗) from the time derivative of the vector
jμ
potential.
Hence, we have
E r ) = iω A⃗0⊥ (⃗
⃗0 (⃗ r) , (5.175a)
∞ j
A⃗0⊥ (⃗ ⃗ (1) (k, r⃗) + njμ N
⃗ (1) (k, r⃗)) ,
jμ (5.175b)
j=0 μ=−j
∞ j
⃗0 (⃗
E ⃗ (1) (k, r⃗) + njμ N
r ) = − k 2 c μ0 ∑ ∑ (mjμ M ⃗ (1) (k, r⃗)) . (5.175c)
jμ jμ
j=0 μ=−j
From Eq. (5.164a), we infer that M ⃗ jμ is the (normalized) elementary solution

consisting of a Bessel function times L Yjμ (θ, ϕ) ∝ Y⃗jμ
⃗ j
(θ, ϕ). It is (by construction)
purely transverse, because r̂ ⋅ Y⃗jμ (θ, ϕ) ∝ r⃗ ⋅ (⃗
j ⃗ Y⃗jμ (θ, ϕ) = 0. The general
r × ∇)
rationale is this: The M ⃗ terms are the magnetic multipoles. We have B ⃗ =∇ ⃗ × A⃗
and E ⃗∝∇ ⃗ The electric field E
⃗ × B. ⃗∝M ⃗ in this case is transverse, i.e., r̂ ⋅ E⃗ = 0.
This is the right characteristic of a magnetic multipole.
In view of Eq. (5.164b), we infer that N ⃗jμ is the solution constructed by the
⃗ ⃗
taking the curl of Mjμ . The Njμ terms are the electric multipoles. Forming the
curl of N ⃗ , one obtains an expression for B ⃗ which is proportional to M ⃗ jμ , which
evidently is transverse. In this particular case, one has r⃗ ⋅ B⃗ = 0. The general result
is as follows,
∞ j
⃗0 (⃗
B ⃗ (1) (k, r⃗) − njμ M
r ) = k 2 μ0 ∑ ∑ (mjμ N ⃗ (1) (k, r⃗)) , (5.175d)
jμ jμ
j=0 μ=−j
as will be shown below. A transverse magnetic field is characteristic of an electric

multipole. One observes, for the electric dipole, that the exact expression for the
magnetic field, given in Eq. (5.97), is transverse, while the exact expression for the
electric field, given in Eq. (5.116), is not transverse.
It is perhaps instructive to discuss the “construction principle” that leads to
Eq. (5.170), i.e., the rationale behind the transformation from Eq. (5.161) to (5.170).
For given j and μ, one first identifies the maximal longitudinal subcomponent and
⃗ (1) (k, r⃗). One then constructs the maximal transverse component whose
calls it L jμ
scalar product with r̂ vanishes, and identifies it as M ⃗ (1) (k, r⃗) = 0. The rest of the
jμ
component of the vector potential with given j and μ finally is the electric multipole,
⃗ (1) (k, r⃗) = 0.
called N jμ
A remark is in order. For the magnetic multipoles, the corresponding term in
A⃗0 (⃗
r) [the term containing M ⃗ (1) (k, r⃗)] is transverse, i.e., its scalar product with
jμ
r̂ vanishes. The electric field is parallel to the time derivative of the sum of the
non-longitudinal components of A⃗0 (⃗ r). The electric field involves the combination
mjμ M ⃗ (1) (k, r⃗) + njμ N
⃗ (1) (k, r⃗), so it is transverse for the magnetic multipoles. The
jμ jμ
magnetic induction field has the combination mjμ N ⃗ (1) (k, r⃗) − njμ M
⃗ (1) (k, r⃗), so it
jμ jμ
is transverse for the electric multipoles. This observation generalizes the behavior
found in Eqs. (5.97) and (5.116), for the exact expressions pertaining to the elec-
tric and magnetic fields radiated by a dipole (the magnetic field was found to be
transverse).
5.5 Radiation and Angular Momenta
5.5.1 Radiated Electric and Magnetic Fields

We shall now try to verify Eq. (5.175d) explicitly. The magnetic radiated field is
calculated as follows,
⃗0 (⃗
B ⃗ × A⃗0 (⃗
r) = ∇ r)
⃗ (k, r⃗) + njμ ∇

⃗ ×M
= i k μ0 ∑ (mjμ ∇
(1) ⃗ (k, r⃗) + ljμ ∇
⃗ ×N (1) ⃗ (k, r⃗))
⃗ ×L (1)
jμ jμ jμ
m
⃗ (1) (k, r⃗)) + njμ (i k M

= i k μ0 ∑ (mjμ (−i k N ⃗ (1) (k, r⃗)))
jμ jμ
jμ
⃗ (1) (k, r⃗) − njμ M

= k 2 μ0 ∑ (mjμ N ⃗ (1) (k, r⃗)) , (5.176)
jμ jμ
jμ
where we have used Eqs. (5.164) and (5.165). In order to calculate the electric field,
one observes that in a source-free region, the Ampere–Maxwell law implies that [see
also Eq. (6.87)]
⃗0 (⃗ iω ⃗ ⃗0 (⃗ ic2 ⃗0 (⃗
⃗ ×B
∇ r) = − 2 E 0 (⃗
r) ⇒ E r) = ⃗ ×B
∇ r) . (5.177)
c ω
Using Eqs. (5.164b) and (5.165), this is evaluated as
⃗0 (⃗ ic2 ⃗0 (⃗
E r) = ∇⃗ ×B r)
ω
ic2 ⃗ (1) (k, r⃗) − njμ ∇
⃗ ×N ⃗ (1) (k, r⃗))
⃗ ×M
= k 2 μ0 ( ) ∑ (mjμ ∇ jμ jμ
ω jμ
i k 2 μ0 c 2 ⃗ (1) (k, r⃗)) − njμ (−i k N
⃗ (1) (k, r⃗)))
= ∑ (mjμ (i k M jμ jμ
ω jμ
(ik) i k 2 μ0 c2 ⃗ (1) (k, r⃗) + njμ N

⃗ (1) (k, r⃗))
= ∑ (mjμ M jμ jμ
ck jμ
⃗ (k, r⃗) + njμ N

= − k 2 c μ0 ∑ (mjμ M
(1) ⃗ (k, r⃗)) , (1)
(5.178)
jμ jμ
jμ
confirming the result anticipated in Eq. (5.175c).
5.5.2 Gauge Condition, Vector and Scalar Potentials

We have defined the multipoles for the radiated scalar potential in Eq. (5.156); the
longitudinal components of the vector potential have been discussed in Eqs. (5.168c)
and (5.164c). In our discussion of the radiated electric field, we anticipated the
cancellation of the longitudinal contributions, in between the gradient of the scalar
potential and the time derivative of the vector potential, in Eqs. (5.174) and (5.175).
In consequence, we anticipate a simple relation between the pm coefficients from
Eq. (5.156) and the ljμ coefficients from Eq. (5.168c). The derivation of this relation
is necessary in order to establish the fulfillment of the Lorenz gauge condition and
will be discussed in the following. We remember that the equations used by us
in order to relate the sources to the potentials, namely, Eqs. (5.153) and (5.158),
precisely are the Lorenz gauge versions, equivalent to Eq. (1.77).
(K)
In the derivation, we shall use the representation (5.164c) for the Ljμ (k, r⃗)
functions. Furthermore, we have the charge conservation condition ∇ ⃗ r , t) =
⃗ ⋅ J(⃗
−∂t ρ(⃗ ⃗ ⃗
r, t) and so ∇ ⋅ J(⃗
r ) = iωρ(⃗
r ). A simple partial integration then leads to the
formulas
⃗ r′ ) ⋅ L
ljμ = ∫ d3 r′ J(⃗ ⃗ (0)∗
′ ′
jμ (θ , ϕ )
⃗ r′ ) ⋅ ∇ 1
= ∫ d3 r′ J(⃗ ⃗ ′ ( jj (k r′ ) Yjμ ∗
(θ′ , ϕ′ ))
k
= − ∫ d3 r′ [∇ ⃗ r′ )] 1 jj (k r′ ) Y ∗ (θ′ , ϕ′ )
⃗ ′ ⋅ J(⃗ jμ
k
1
= − ∫ d3 r′ i ω ρ(⃗ r′ ) jj (k r′ ) Yjμ (θ′ , ϕ′ )
k
ω 3 ′
= − i ∫ d r ρ(⃗ r ) jj (k r′ ) Yjμ (θ′ , ϕ′ ) = −i c pjμ .
′
(5.179)
k
With pjμ = ∫ d3 r′ ρ(⃗

r′ ) jj (k r′ ) Yjμ
∗
(θ′ , ϕ′ ), we thus verify the relation
ljμ = −ic pjμ . (5.180)
Let us recall the gauge condition (1.70) fulfilled by the potentials,
1 ∂Φ
⃗ ⋅ A⃗ +
∇ = 0. (5.181)
c2 ∂t
In view of the decompositions (5.155) and (5.170), we have for the scalar and vector
potentials,
ik ∞ j (1)
Φ0 (⃗
r) = ∑ ∑ pjμ hj (k r) Yjμ (θ, ϕ) , (5.182)
0 j=0 μ=−j
and
∞ j
A⃗0 (⃗ ⃗ (1) (k, r⃗) + njμ N
⃗ (1) (k, r⃗) + ljμ L
jμ
⃗ (1) (k, r⃗)) .
jμ (5.183)
j=0 μ=−j
In the mixed frequency-coordinate representation, the Lorenz gauge condition reads

as
iω
⃗ ⋅ A⃗0 (⃗
∇ r ) − 2 Φ0 (⃗
r) = 0 . (5.184)
c
Because the magnetic and electric multipole terms are divergence-free [see
Eq. (5.175b)], the Lorenz gauge condition is equivalent to
ik ∑ {μ0 ljμ ∇ ⃗ (1) (k, r⃗) + (−i ω) pjμ h(1) (k r) Yjμ (θ, ϕ)} = 0 ,
⃗ ⋅L (5.185)
jμ j
jμ 0 c2
which implies that
ik ∑ {ljμ ∇ ⃗ (1) (k, r⃗) − i kc pjμ h(1) (k r) Yjμ (θ, ϕ)} = 0 .

⃗ ⋅L (5.186)
jμ j
jμ
We use Eq. (5.164c), in the form
⃗ (1) (k, r⃗) = 1 ∇

⃗ (h(1)
L jμ j (k r) Yjμ (θ, ϕ)) , (5.187)
k
⃗ (1) (k, r⃗) = 1 ∇
⃗ ⋅L
∇ ⃗ 2 (h(1)
jμ j (k r) Yjμ (θ, ϕ))
k
(1)
= − k hj (k r) Yjμ (θ, ϕ) . (5.188)
(1)
Advantage has been taken of the fact that the function hj (k r) Yjμ (θ, ϕ) fulfills
⃗ 2 + k 2 )h(1)
the Helmholtz equation, i.e., (∇ j (k r) Yjμ (θ, ϕ) = 0. Inserting Eq. (5.187)
into (5.186), we see that the gauge condition is equivalent to the relation
(1)
∑ (ljμ + i c pjμ ) hjμ (k r) Yjμ (θ, ϕ) = 0 , (5.189)
jμ
which is fulfilled in view of Eq. (5.180).

Using Eq. (5.180), we should now be able to verify the cancellation of the lon-
gitudinal contributions to the electric field, in the steps leading from Eq. (5.174)
to (5.175a). To this end, we calculate
⃗0 (⃗
E r ) = − ∇Φ r ) + iω A⃗0 (⃗
⃗ 0 (⃗ r)
⎛ ik (1) ⎞
⃗
= −∇ ∑ pjμ hj (kr) Ym (r̂)
⎝ 0 jμ ⎠
⃗ (k, r⃗) + njμ N

+ i ω (ikμ0 ∑ {mjμ M
(1) ⃗ (k, r⃗) + ljμ L
⃗ (k, r⃗)})
(1) (1)
jμ jμ jμ
m
⃗ (1) (k, r⃗) + njμ N

= − k 2 μ0 c ∑ {mjμ M ⃗ r) ,
⃗ (1) (k, r⃗)} + E(⃗ (5.190)
jμ jμ
jμ
⃗ r)
where the first term is the result given in Eq. (5.175a) and the additional term E(⃗
must be shown to vanish,
⃗ r ) = − ik ∑ {pjμ ∇
E(⃗ ⃗ (hj(1) (kr) Yjμ (r̂)) + (− 0 ) iω(ikμ0 )ljμ L
⃗ (1) (k, r⃗)}
jμ
0 jμ ik
ik ⃗ (1)
= − ⃗ (h(1)
∑ {pjμ ∇ j (kr) Yjμ (r̂)) − ikc μ0 0 ljμ Ljμ (k, r
⃗)}
0 jμ
ik 2 i ⃗ (1) (k, r⃗)} = 0 ,

= − ∑ {(pjμ − ljμ ) L jμ (5.191)
0 jμ c
again in view of Eq. (5.180). We have shown that the longitudinal components of
the electric field cancel, as they should, in the source-free region. A few remarks
are in order. The factor [−0 /(ik)] in the first line is simply introduced in order
to cancel the first prefactor. Alternatively, one may observe that the extra term in
the electric field, from the longitudinal components of the vector potential, simply
⃗ (1) (k, r⃗), with the prefactor being fixed in Eq. (5.178).
reads as −k 2 c μ0 ∑jμ ljμ L jμ
5.5.3 Representations of the Vector Spherical Harmonics
In the following, we shall endeavor in rather difficult calculations. It is a good

moment to rest and to write down the most important representations of the vector
spherical harmonics, and of the multipole functions. We start from Eqs. (5.148a)
and (5.151a),
1
Y⃗jjμ (θ, ϕ) = √ ⃗ jμ (θ, ϕ) ,
LY
j(j + 1)
B
D (j + μ) (j − μ + 1) μ
= −D
E Yj,μ−1 (θ, ϕ) e⃗+1 + √ Yj,μ (θ, ϕ) e⃗0
2j(j + 1) j(j + 1)
B
D (j − μ) (j + μ + 1)
+D
E Yj,μ+1 (θ, ϕ) e⃗−1 . (5.192a)
2j(j + 1)
Equations (5.148b) and (5.151b) state that

1
Y⃗jj−1
μ (θ, ϕ) = √
⃗ Yjμ (θ, ϕ)
(j r̂ + r ∇)
j (2j + 1)
1 ⃗ − j r̂) Yjμ (θ, ϕ) ,
= −√ (i r̂ × L
j (2j + 1)
B
D (j + μ − 1) (j + μ)
=DE Yj−1,μ−1 (θ, ϕ) e⃗+1
2j(2j − 1)
B
D (j − μ) (j + μ)
+D
E Yj−1,μ (θ, ϕ) e⃗0
j(2j − 1)
B
D (j − μ − 1) (j − μ)
+D
E Yj−1,μ+1 (θ, ϕ) e⃗−1 , (5.192b)
2j(2j − 1)
Finally, we have from Eq. (5.148c) and (5.151c),

1
Y⃗jj+1
μ (θ, ϕ) = − √
⃗ Yjμ (θ, ϕ)
((j + 1) r̂ − r ∇)
(j + 1) (2j + 1)
1 ⃗ + (j + 1) r̂) Yjμ (θ, ϕ)
= −√ (i r̂ × L
(j + 1) (2j + 1)
B
D (j − μ + 1) (j − μ + 2)
=DE Yj+1,μ−1 (θ, ϕ) e⃗+1
2(j + 1)(2j + 3)
B
D (j − μ + 1) (j + μ + 1)
−D
E Yj+1,μ (θ, ϕ) e⃗0
(j + 1)(2j + 3)
B
D (j + μ + 2) (j + μ + 1)
+D
E Yj+1,μ+1 (θ, ϕ) e⃗−1 . (5.192c)
2(j + 1)(2j + 3)
The spherical harmonics are given by Eq. (2.53), which we recall for convenience,
1/2
2j + 1 (j − μ)! ∣μ∣
Yj μ (θ, ϕ) = (−1)μ ( ) P (cos θ) eiμϕ . (5.193)
4π (j + μ)!
We also summarize Eqs. (5.164), (5.171), and (5.172) as follows,
⃗ (K) (k, r⃗) = √ 1

M ⃗ (f (K) (k r) Yjμ (θ, ϕ))
L
jμ j
j(j + 1)
i
(k r) Y⃗jμ ⃗ (K) (k, r⃗) ,
(K) j ⃗ ×N
= fj (θ, ϕ) = − ∇ jμ (5.194a)
k
⃗ (K) (k, r⃗) = i ∇

N ⃗ ×M ⃗ (K) (k, r⃗) ,
jμ jμ
k
√ √
j + 1 (K) ⃗ j
(k r) Y⃗jj+1
j−1 (K)
= − f (k r) Yj μ (θ, ϕ) + f μ (θ, ϕ) ,
2j + 1 j−1 2j + 1 j+1
⎧ d[kr f (K) (k r)] √ ⎫

1 ⎪
⎪ j ⃗ j (K)
⎪
⎪
= ⎨ [ir̂ × Yjμ (θ, ϕ)] − r̂ j(j + 1)fj (k r) Yjμ (θ, ϕ)⎬ ,
kr ⎪
⎪ d(kr) ⎪
⎪
⎩ ⎭
(5.194b)
⃗ (K) (k, r⃗) = 1 ∇
L ⃗ (fj(K)(k r) Yjμ (θ, ϕ))
jμ
k
√ √
j j + 1 (K)
fj−1 (k r) Y⃗jj−1 (k r) Y⃗jj+1
(K)
= (θ, ϕ) + f μ (θ, ϕ)
2j + 1 μ
2j + 1 j+1
(K)
d[fj (k r)] √ 1 (K)
= r̂ Yjμ (θ, ϕ) − j(j + 1) f (k r) (ir̂ × Y⃗jμ
j
(θ, ϕ)) .
d(k r) kr j
(5.194c)
These results will be needed in the analysis of the Poynting vector.
5.5.4 Poynting Vector of the Radiation
We start from the relations (5.175c) and (5.175d),

∞ j
⃗0 (⃗
E ⃗ (1) (k, r⃗) + njμ N
r ) = −k 2 c μ0 ∑ ∑ (mjμ M ⃗ (1) (k, r⃗)) , (5.195)
jμ jμ
j=0 μ=−j
and
∞ j
⃗0 (⃗
B ⃗ (1) (k, r⃗) − njμ M
r ) = k 2 μ0 ∑ ∑ (mjμ N ⃗ (1) (k, r⃗)) . (5.196)
jμ jμ
j=0 μ=−j
The Poynting vector, which carries the physical dimension of radiated power per
area, is [see Eq. (5.106)]
⃗ r, t) = 1 E(⃗
S(⃗ ⃗ r , t) × B(⃗
⃗ r , t) , S⃗0 (⃗
r) =
1 ⃗
E0 (⃗ ⃗0∗ (⃗
r) × B r) . (5.197)
μ0 2μ0
From Eq. (5.164a), we recall the relation
⃗ (K) (k, r⃗) = f (K) (k r) Y⃗ j (θ, ϕ) ,
M (5.198)
jμ j jμ
which implies that

⃗ (1) (k, r⃗) = f (1) (k r) Y⃗ j (θ, ϕ) = h(1) (k r) Y⃗ j (θ, ϕ) .
M (5.199)
jμ j jμ j jμ
In view of the asymptotics (for large k r)
(1) ei k r−(j+1)iπ/2
hj (k r) → , (5.200)
kr
we have
i k r−(j+1)iπ/2 ikr
⃗ (1) (k, r⃗) = h(1) (k r) Y⃗ j (θ, ϕ) → e
M Y⃗ j (θ, ϕ) = (−i)j+1 e Y⃗ j (θ, ϕ) .
jμ j jμ jμ
kr k r jμ
(5.201)
From Eq. (5.164b), we infer that
√ √
⃗ (K) (k, r⃗) = − j + 1 (K) ⃗ j−1 (θ, ϕ) + j
(k r) Y⃗jj+1
(K)
N f (k r) Y f μ (θ, ϕ) .
jμ
2j + 1 j−1 jμ
2j + 1 j+1
(5.202)
This implies that for large r,
√ √
⃗ (1) (k, r⃗) = − j + 1 (1) ⃗ j−1 (θ, ϕ) + j
h (k r) Y⃗jj+1
(1)
N h (k r) Y μ (θ, ϕ)
jμ
2j + 1 j−1 jμ
2j + 1 j+1
√
j+1 ei k r ⃗ j−1
→ − (−i)(j−1)+1 Y (θ, ϕ)
2j + 1 kr jμ
√
j ei k r ⃗ j+1
+ (−i)(j+1)+1 Y (θ, ϕ)
2j + 1 kr jμ
√ √
ei k r ⎛ j + 1 ⃗ j−1 j ⃗ j+1 ⎞
→ − (−i)j Yj μ (θ, ϕ) + Y (θ, ϕ) . (5.203)
k r ⎝ 2j + 1 2j + 1 j μ ⎠
(1) (1)
We use the fact that hj−1 and hj+1 only differ by a phase (−i)2 = −1 in the long-
distance limit. From Eqs. (5.148b) and (5.148c), one may derive the relation
√ √
j + 1 ⃗ j−1 j ⃗ j+1
Yj μ (θ, ϕ) + Y (θ, ϕ) = −i (r̂ × Y⃗jjμ (θ, ϕ)) . (5.204)
2j + 1 2j + 1 j μ
⃗ (1) (k, r⃗) are thus given by

The long-range asymptotics of N jμ
√ √
⃗ (1) (k, r⃗) ei k r ⎛ j + 1 ⃗ j−1 j ⃗ j+1 ⎞
N = − (−i) j
Yj μ (θ, ϕ) + Yj μ (θ, ϕ)
jμ
k r ⎝ 2j + 1 2j + 1 ⎠
ei k r ei k r
= − (−i)j (−i) (r̂ × Y⃗jjμ (θ, ϕ)) = −(−i)j+1 (r̂ × Y⃗jjμ (θ, ϕ)) .
kr kr
(5.205)
We now use Eqs. (5.175c) and (5.175d) and the long-range asymptotic formulas,
ikr ikr
⃗ (1) (k, r⃗) → (−i)j+1 e
M Y⃗ j (θ, ϕ) , ⃗ (1) (k, r⃗) → −(−i)j+1 e
N (r̂ × Y⃗jjμ (θ, ϕ)) .
jμ
k r jμ jμ
kr
(5.206)
Hence,
⃗0 (⃗ ei k r ∞ j
E r) → −k 2 c μ0 ∑ ∑ (−i)
j+1
(mjμ Y⃗jμ
j
(θ, ϕ) − njμ (r̂ × Y⃗jjμ (θ, ϕ))) ,
k r j=0 μ=−j
(5.207)
and
⃗0 (⃗ ei k r ∞ j
B r) → −k 2 μ0 ∑ ∑ (−i)
j+1
(mjμ (r̂ × Y⃗jjμ (θ, ϕ)) + njμ Y⃗jμ
j
(θ, ϕ)) .
k r j=0 μ=−j
(5.208)
The Poynting vector evaluates to
1 ⃗
S⃗0 (⃗
r) = E0 (⃗ ⃗ ∗ (⃗
r) × B 0 r)
2μ0
1 k 4 c μ0 ∞ j
= ∑ ∑ (−i)
j+1
(mjμ Y⃗jμ
j
(θ, ϕ) − njμ (r̂ × Y⃗jjμ (θ, ϕ)))
2 (k r)2 j=0 μ=−j
∞ ′
j′ ′ ′
× ∑ ∑ ij +1 (m∗j ′ μ′ (r̂ × Y⃗jj′ μ∗′ (θ, ϕ)) + n∗j ′ μ′ Y⃗jj′ μ∗′ (θ, ϕ))
j ′ =0 μ′ =−j ′
′
k 2 c μ0 ∞ j ∞ j j ′ −j
= 2 ∑ ∑ ∑ ∑ i (mjμ Y⃗jμ
j
(θ, ϕ) − njμ (r̂ × Y⃗jjμ (θ, ϕ)))
2 r j=0 μ=−j j ′ =0 μ′ =−j ′
′ ′
× (m∗j ′ μ′ (r̂ × Y⃗jj′ μ∗′ (θ, ϕ)) + n∗j ′ μ′ Y⃗jj′ μ∗′ (θ, ϕ)) , (5.209)
which is a double-infinite sum. The total power radiated through a sphere of radius
r is
P = ∫ dΩ r2 r̂ ⋅ S⃗0 (⃗
r ) = T1 + T2 , (5.210)
and considerable simplifications occur in its calculation. Here,

k 2 c μ0 j ′ −j
′
T1 = ∫ dΩ r̂ ⋅ ∑ i (mjμ m∗j ′ μ′ Y⃗jμ
j
(θ, ϕ) × (r̂ × Y⃗jj′ μ∗′ (θ, ϕ))
2 jj ′ μμ′
′
− njμ n∗j ′ μ′ (r̂ × Y⃗jjμ (θ, ϕ)) × Y⃗jj′ μ∗′ (θ, ϕ)) ,
k 2 c μ0 j ′ −j
′
T2 = ∫ dΩ r̂ ⋅ ∑ i (mjμ n∗j ′ μ′ Y⃗jμ
j
(θ, ϕ) × Y⃗jj′ μ∗′ (θ, ϕ)
2 jj ′ μμ′
′
− njμ m∗j ′ μ′ (r̂ × Y⃗jjμ (θ, ϕ)) × (r̂ × Y⃗jj′ μ∗′ (θ, ϕ))) . (5.211)
Now the angular integrals have to be evaluated. We recall Eq. (5.148a),

1 1
Y⃗jjμ (θ, ϕ) = √ ⃗ jμ (θ, ϕ) ,
LY r̂ ⋅ Y⃗jjμ (θ, ϕ) = √ ⃗ jμ (θ, ϕ) = 0 ,
r̂ ⋅ LY
j(j + 1) j(j + 1)
(5.212)
⃗ = −i (⃗
where L ⃗ Hence,
r × ∇).
′
∫ dΩ r̂ ⋅ Y⃗jμ (θ, ϕ) × (r̂ × Y⃗j ′ μ′ (θ, ϕ))

j j∗
′ ′
= ∫ dΩ r̂ ⋅ [r̂ (Y⃗jμ
j
(θ, ϕ) ⋅ Y⃗jj′ μ∗′ (θ, ϕ)) − Y⃗jj′ μ∗′ (θ, ϕ) (r̂ ⋅ Y⃗jμ
j
(θ, ϕ))]
′ ′
= ∫ dΩ [Y⃗jμ
j
(θ, ϕ) ⋅ Y⃗jj′ μ∗′ (θ, ϕ) − (r̂ ⋅ Y⃗jj′ μ∗′ (θ, ϕ)) (r̂ ⋅ Y⃗jμ
j
(θ, ϕ))]
′
= ∫ dΩ Y⃗jμ
j
(θ, ϕ) ⋅ Y⃗jj′ μ∗′ (θ, ϕ) = δjj ′ δμμ′ , (5.213)
where we have used Eq. (5.212). Similarly, one shows that

′
∫ dΩ r̂ ⋅ [(r̂ × Y⃗j μ (θ, ϕ)) × Y⃗j ′ μ′ (θ, ϕ)] = −δjj ′ δμμ′ .
j j ∗
(5.214)
The two identities

′
∫ dΩ r̂ ⋅ (Y⃗jμ (θ, ϕ) × Y⃗j ′ μ′ (θ, ϕ)) = 0 ,

j j ∗
(5.215)
′
∫ dΩ r̂ ⋅ (r̂ × Y⃗j μ (θ, ϕ)) × (r̂ × Y⃗j ′ μ′ (θ, ϕ)) = 0 ,

j j ∗
(5.216)
are a little harder to show and are left as exercises. Armed with the results (5.213)–
(5.216), we can show that T2 = 0 and simplify Eq. (5.209) drastically,
k 2 c μ0
P= ∑ (∣mjμ ∣ + ∣njμ ∣ ) .
2 2
(5.217)
2 jμ
5.5.5 Half-Wave Antenna

Let us carry out the multipole decomposition explicitly for an example problem.
We consider a half-wave antenna with a current inside the antenna,
2πz −iωt
I = I0 cos ( )e , (5.218)
λ
where −λ/4 < z < λ/4. The length of the antenna therefore is exactly equal to half
the wavelength, hence the name. In order to invoke our formalism, we first need to
translate the current into a current density. A first guess might be
⃗ r ) = J⃗0 (⃗ 2πz
J(⃗ r ) e−iωt , J⃗0 (⃗
r ) = êz I0 cos ( ) δ(x) δ(y) , (5.219)
λ
because the antenna is oriented parallel to the z axis. However, this expression
leads to problems as we put in the formulas x = r sin θ cos ϕ and y = r sin θ sin ϕ
for the coordinates. How are we supposed to evaluate the integrals over θ and ϕ
if they enter the arguments of the Dirac-δ function in such an awkward way? We
therefore use the apparent radial symmetry of the problem, with respect to the z
axis, and write
2πr 1
J⃗0 (r, θ, ϕ) = êz I0 cos ( ) [2δ(θ) + 2δ(π − θ)] 2
(5.220)
λ 2π r sin(θ)
We use the convention [see Eq. (1.31)]
b 1 b 1
∫ f (a) ,
dx δ(x − a) f (x) = ∫ dx δ(x − b) f (x) = f (b) , (5.221)
a 2 a 2
for a Dirac-δ function “on the boundary”. One may thus calculate the integral over
the (partial) surface area of a sphere, centered at the origin, which encompasses the
antenna. The infinitesimal area element is
dA = R2 sin θ dθ dϕ . (5.222)
The surface integral, with the polar angle θ ∈ (0, ), samples points whose z coordi-
nate is just z = R, i.e., points in the vicinity of the “pole” of the “sampling sphere”.
It evaluates to
2π
∫ dA J⃗0 (R, θ, ϕ) = R ∫ dϕ J⃗0 (R, θ, ϕ)
2
d θ sin θ ∫
0 0
2πR 2π 1
= êz I0 cos ( )∫ dϕ ∫ d θ sin θ R2 [2δ(θ)]
λ 0 0 2π R2 sin(θ)
2πR 2πz
= êz I0 cos ( ) ∫ d θ [2δ(θ)] = êz I0 cos ( ). (5.223)
λ 0 λ
Assembling the current density from terms near the “North” and the “South” pole,
one has
2πr δ(θ) + δ(π − θ)
J⃗0 (r, θ, ϕ) = I0 cos ( ) Θ( 41 λ − r) êz . (5.224)
λ π r2 sin(θ)
In order to calculate the magnetic multipoles using Eq. (5.164a), one recalls that
according to Eq. (5.148a),
1
Y⃗jjμ (r̂) = √ ⃗ Yjμ (r̂) ,
L (5.225)
j(j + 1)
and so
1 1
êz ⋅ Y⃗jjμ (r̂) = √ Lz Yjμ (r̂) = √ μ Yjμ (r̂) . (5.226)
j(j + 1) j(j + 1)
Thus,
mjμ = ∫ d3 r J⃗0 (⃗ ⃗ (0)∗ (⃗

r) ⋅ M jμ r) = ∫ d3 r j0 (k r) J⃗0 (⃗
r ) ⋅ Y⃗jμ
j∗
(θ, ϕ)
δ(θ) + δ(π − θ)
= ∫ dϕ ∫ dθ sin θ ∫ dr r2 j0 (k r) I0 cos (k r) êz ⋅ Y⃗jμ
j∗
(θ, ϕ)
π r2 sin(θ)
I0
∫ dϕ ∫ dr j0 (k r) cos (k r) [êz ⋅ Y⃗jμ (0, ϕ) + êz ⋅ Y⃗jμ (π, ϕ)]
j∗ j∗
=
2π
I0 1 ∗ ∗
= ∫ dϕ ∫ dr j0 (k r) cos (k r) √ [μ Yjμ (0, ϕ) + μ Yjμ (π, ϕ)] .
2π j (j + 1)
(5.227)
From Eq. (2.53), one has
√
2j + 1
Yjμ (0, ϕ) = δμ0 ,
4π
√
2j + 1
Yjμ (0, ϕ) = (−1) δμ0
j
. (5.228)
4π
Both results are independent of ϕ, as they should be on the pole caps of the unit
sphere. The latter result follows from the former by the parity transformation
θ → π − θ, and ϕ → ϕ + π (parity transformation). In view of μ δμ0 = 0, it follows
that
mjμ = 0 , (5.229)
or in other words, all the magnetic multipoles vanish.
Now we turn our attention to the electric multipoles. From Eq. (5.148b), we
have
B
D (j − μ) (j + μ)
êz ⋅ Yj μ (θ, ϕ) = D
⃗ j−1 ∗ E Yj−1,μ (θ, ϕ) êz ⋅ e⃗0∗
j(2j − 1)
B
D (j − μ) (j + μ)
=DE Yj−1,μ (θ, ϕ) , (5.230)
j(2j − 1)
because e⃗q e⃗q∗′ = δq q′ . At the “North” pole of the unit sphere, one therefore has the
relation,
B
D (j − μ) (j + μ)
êz ⋅ Yj μ (0, ϕ) = D
⃗ j−1 ∗ E Yj−1,μ (0, ϕ)
j(2j − 1)
B √ √
D (j − μ) (j + μ) 2(j − 1) + 1 j
= D
E δμ0 = δμ0 , (5.231)
j(2j − 1) 4π 4π
while at the “South” pole of the unit sphere, one has

B
D (j − μ) (j + μ)
êz ⋅ Y⃗jj−1 ∗
(π, ϕ) = D
E Yj−1,μ (μ, ϕ)
μ
j(2j − 1)
B √
D (j − μ) (j + μ)
j−1 D 2(j − 1) + 1
= (−1) E δμ0
j(2j − 1) 4π
√
j
= (−1)j−1 δμ0 . (5.232)
4π
Let us also investigate, at the “North” pole of the unit sphere, using Eq. (5.148c)
B
D (j − μ + 1) (j + μ + 1)
êz ⋅ Y⃗j μ (0, ϕ) = − D
j+1 ∗ E Yj+1,μ (0, ϕ)
(j + 1)(2j + 3)
B √
D (j − μ + 1) (j + μ + 1) 2(j + 1) + 1
= − D
E δμ0
(j + 1)(2j + 3) 4π
√
j +1
= − δμ0 . (5.233)
4π
Conversely, at the “South” pole of the unit sphere, one has
B √
D (j − μ + 1) (j + μ + 1) j +1
êz ⋅ Y⃗j μ (π, ϕ) = −D
j+1 ∗ E Yj+1,μ (π, ϕ) = −(−1) δμ0 j−1
,
(j + 1)(2j + 3) 4π
(5.234)
where the intermediate steps are left as an exercise to the reader. From Eq. (5.164b),
we recall that
√ √
⃗ j + 1 (K) ⃗ j
(k r) Y⃗jj+1
(K) j−1 (K)
Njμ (k, r⃗) = f (k r) Yj μ (θ, ϕ) − f μ (θ, ϕ) .
2j + 1 j−1 2j + 1 j+1
(5.235)
We may now calculate the electric multipoles,
√
3 ⎛ j +1 ⃗
njμ = ∫ d3 r J⃗0 (⃗ ⃗ (0)∗ (⃗
r) ⋅ N r ) = ∫ d r jj−1 (k r) r) ⋅ Y⃗jμ
J (⃗ j−1∗
(θ, ϕ)
jμ
⎝ 2j + 1
√
− jj+1 (k r)
j ⃗ r ) ⋅ Y⃗ j+1∗ (θ, ϕ)⎞
J(⃗
2j + 1 jμ
⎠
√
I0 ⎛ j +1
= dϕ ∫ dr r2 cos(kr) jj−1 (k r) [êz ⋅ Y⃗jμ
j−1∗
(0, ϕ)
2π ∫ ⎝ 2j + 1
√
j ⎞
+ êz ⋅ Y⃗jμ (π, ϕ)] − jj+1 (k r)
j−1∗
[êz ⋅ Y⃗jμ
j+1∗
(0, ϕ) + êz ⋅ Y⃗jμ
j+1∗
(π, ϕ)] .
2j + 1 ⎠
(5.236)
With the help of the results (5.231), (5.232), (5.233) and (5.234), one simplifies this
expression to the form
√ √
I0 ⎛ j +1 j
njμ = ∫ dϕ ∫ dr cos(kr) jj−1 (k r) [1 + (−1) ] δμ0
j−1
2π ⎝ 2j + 1 4π
√ √
j ⎛ j + 1 ⎞⎞
− jj+1 (k r) − [1 + (−1) ] δμ0
j−1
2j + 1 ⎝ 4π ⎠⎠
B
D j (j + 1)
= I0 δμ0 D
E [1 + (−1)j−1 ] ∫ dr cos(kr) (jj−1 (k r) + jj+1 (k r)) .
4π (2j + 1)
(5.237)
Recalling the recursion relation for spherical Bessel functions, given in Eq. (5.28),
we find that the multipole moment is given as follows,
B
D j (j + 1) 2j + 1
njμ = I0 δμ0 DE [1 + (−1)j−1 ] ∫ dr cos(kr) jj (k r)
4π (2j + 1) kr
√
j (j + 1)(2j + 1) cos(kr)
= − I0 δμ0 [1 + (−1)j−1 ] ∫ dr (− ) jj (k r)
4π kr
√ λ/4
j (j + 1)(2j + 1)
= − I0 δμ0 [1 + (−1) ] ∫ dr y0 (k r) jj (k r)
j−1
(5.238)
4π
0
where y0 of course is the spherical Neumann function. The spherical Bessel and
Neumann functions fulfill the differential equation (5.27). Hence, one can show
that
x2
∫ dx f j (x) g j ′ (x) = [fj (x) gj′ ′ (x) − fj′ (x) gj ′ (x)] . (5.239)
j ′ (j ′ + 1) − j(j + 1)
Now we set fj = y0 and gj ′ = jj , and have
x2
∫ dx y0 (x) jj (x) = (y0 (x) jj′ (x) − y0′ (x) jj (x)) . (5.240)
j(j + 1)
For the case of interest, we have in view of k = 2π/λ, as well as y0 (π/2) = 0, and
y0′ (π/2) = 2/π,
λ/4 π/2
1
∫ dr y0 (k r) jj (k r) = ∫ dx y0 (x) jj (x)
k
0 0
1 (π/2)2 ′ π 1
= − y0 (π/2) jj (π/2) = − jj (π/2) .
k j(j + 1) 2k j(j + 1)
(5.241)
Hence,
√
π I0 2j + 1
njμ = δμ0 [1 + (−1)j−1 ] jj (π/2) . (5.242)
2k 4π j (j + 1)
The results (5.229) and (5.242) enter the relations (5.175c) and (5.175d), which in
our case simplify because mjμ = 0 [see Eq. (5.229)],
∞
⃗0 (⃗
E ⃗ (1) (k, r⃗) ,
r ) = − k 2 c μ0 ∑ nj0 N (5.243a)
j0
j=0
∞
⃗0 (⃗
B ⃗ (1) (k, r⃗) .
r ) = − k 2 μ0 ∑ nj0 M (5.243b)
j0
j=0
For the half-wave antenna, one can drop the mjμ terms and restrict the sum over
the njμ terms to those with μ = 0.
We now intend to use Eq. (5.217) in order to evaluate the radiated power in
the far field. It is instructive to rewrite Eq. (5.217) somewhat and to introduce the
factor
μ
μ0 c = √ = Z0 ≈ 377 Ω , (5.244)
0 μ0
which is commonly referred to as the vacuum impedance. The radiated power P

can be written as a product of the vacuum impedance and the square of the current
I0 , times some numerical factor,
k2 k2
P= Z0 ∑ {∣mjμ ∣2 + ∣njμ ∣2 } = Z0 ∑ ∣njμ ∣2
2 jμ 2 jμ
k2 π 2j + 1 2
= Z0 ∑ ∣nj 0 ∣2 = Z0 I02 ∑ [jj (π/2)] . (5.245)
2 j odd 8 j odd j (j + 1)
The sum
2j + 1 2
S= ∑ [jj (π/2)] = 0.246 986 (5.246)
j odd j (j + 1)
is rapidly converging as j is increased. The term with j = 1 is 0.246 384, implying

that dipole radiation accounts for 99.76 % of the radiated power. Equating the
radiated power with the “resistance to radiation emission”, one has
1 π
P= Rrad I02 , Rrad = Z0 S = 73.079 Ω . (5.247)
2 4
The physical interpretation is that the antenna acts just like a resistor, as it emits
energy in the form of outgoing radiation. The power absorbed by the antenna is
proportional to Z0 I02 . For given 0 , if the speed of light were infinitely fast, then
in view of μ0 0 = 1/c2 , we would have μ0 → 0 and Z0 → 0; it would be “easier” to
emit radiation, or in other words, the emitted electromagnetic waves would carry
less energy. With a grain of salt, we can remark that, because the impedance of the
vacuum has an appreciable numerical value of ≈ 377 Ω, radio communication works
well.
5.5.6 Long-Wavelength Limit of the Dipole Term
The somewhat elegant formalism we have discussed obscures the leading contribu-
tions to the dipole terms, and to other, higher-order multipole terms; we shall now
attempt to recover these terms, at least in the long-wavelength limit. In particular,
(1)
we attempt to show that the dipole term is part of the expression N1μ , and is
hidden in the contribution of jj−1 (kr) = j0 (kr). It remains to verify the prefactor.
Let us therefore investigate
njμ = ∫ d3 r J⃗0 (⃗ ⃗ (0)∗ (k, r⃗) .

r) ⋅ N jμ (5.248)
Now, in view of Eqs. (5.164a) and (5.164b), we have
⃗ (0) (k, r⃗) = − i ∇

N ⃗ (K) (k, r⃗)
⃗ ×M
jμ jμ
k
1 1
= −√ ⃗ × (⃗
∇ ⃗ jj (k r) Yjμ
r × ∇) ∗
(θ, ϕ) . (5.249)
j(j + 1) k
⃗ × (⃗
The expression ∇ ⃗ jj (k r) Yjμ (θ, ϕ) is tricky; one can simplify it but must
r × ∇)
remember that the leftmost gradient operator acts on everything that follows. The
result is that
∂
⃗ × (⃗
∇ ⃗ f (⃗
r × ∇) r) = [⃗
r∇ ⃗
⃗2 − ∇ r] f (⃗
r) , (5.250)
∂r
and so
⃗ (0) (k, r⃗) = − √ 1

N
1
[⃗
r∇ ⃗ ( ∂ r)] jj (k r) Yjμ
⃗2 − ∇ ∗
(θ, ϕ) . (5.251)
jμ
j(j + 1) k ∂r
One uses Eq. (5.250), the fact that jj (k r) Yjμ (θ, ϕ) satisfies the Helmholtz equation,
integration by parts, and the continuity equation, to write
1 1 ⃗ ( ∂ r)] jj (k r) Yjμ
njμ = − √ ∫ d3 r J⃗0 (⃗
r ) ⋅ [⃗
r∇⃗2 − ∇ ∗
(θ, ϕ)
j(j + 1) k ∂r
1 1 ∂
∫ d r J⃗0 (⃗ ⃗ ( r)] jj (k r) Yjμ
∗
= −√ 3
r ) ⋅ [−⃗
r k2 − ∇ (θ, ϕ)
j(j + 1) k ∂r
1 1 ∂
=√ ∫ r ⋅ J⃗0 (⃗
d3 r [k 2 jj (k r)⃗ ⃗ ⋅ J⃗0 (⃗
r ) − ( rjj (k r)) ∇ ∗
r)] Yjμ (θ, ϕ)
j(j + 1) k ∂r
1 ∂
∫ d r [k jj (k r) r⃗ ⋅ J⃗0 (⃗
∗
=√ 3
r ) − ( rjj (k r)) (ic ρ0 (⃗
r))] Yjμ (θ, ϕ) .
j(j + 1) ∂r
(5.252)
In the long-wavelength limit, i.e, small-k limit, the dominant term obviously is the
second one, as it carries no explicit factor k. Using the long-wavelength (small-k)
asymptotics of the Bessel function, we have

ic ∂ ∗
njμ = − √ ∫ d r ( ∂r rjj (k r)) Yjμ (θ, ϕ) ρ0 (⃗
3
r)
j(j + 1)
ic ∂ k j rj+1 ∗
≈ −√ ∫ d r(
3
) Yjμ (θ, ϕ) ρ0 (⃗
r)
j(j + 1) ∂r (2j + 1)!!
i kj c rj ∗
≈ −√ (j + 1) ∫ d3 r ( ) Yjμ (θ, ϕ) ρ0 (⃗
r)
j(j + 1) (2j + 1)!!
√
j +1 1 ∗
≈ − ik c
j
d3 r rj Yjμ (θ, ϕ) ρ0 (⃗
r)
j (2j + 1)!! ∫
√
ic j + 1 j (0)
≈ − k qjμ . (5.253)
(2j + 1)!! j
(0)
Here, the qjμ are the multipole moments defined in Eq. (2.97), as familiar from
electrostatics, but this time evaluated using the spatial part ρ0 (⃗
r ) of the oscillating
charge density ρ(⃗ r, t) = ρ0 (⃗
r ) exp(−iωt). Setting j = 1, one recovers the long-
wavelength limit of the electric-dipole term.
5.6 Potentials due to Moving Charges in Different Gauges
5.6.1 Moving Charges and Lorenz Gauge

The cancellation of the instantaneous interaction in the calculation of the electric
field in Lorenz gauge has already been discussed in Sec. 4.2, leaving the charge and
current distributions general. It is very instructive and nontrivial to verify the gen-
eral results on the basis of a concrete problem, namely, the potentials generated by
a moving point charge. In classical electrodynamics, all electric and magnetic fields
can in principle be described as a superposition of fields generated by infinitesi-
mal moving point charges, so that the specialization to a moving point charge does
not necessarily imply a loss of generality. Again, the intriguing problem is that
in Coulomb gauge (radiation gauge), the scalar potential Φ can be written as an
instantaneous integral over charges that are far away from the observation point. A
change in a charge distribution light years away would thus lead to an instantaneous
change in the scalar potential at an observation point on Earth. In order to address
this latter question, we here not only calculate the scalar potential, but also the
vector potential generated by a moving charge, in Coulomb gauge and in general
form. This problem has independently attracted some interest [25].
The representation (4.35) of the retarded Green function in coordinate space
contains two Dirac-δ functions. It is highly singular. Therefore, it is certainly
sensible to verify the conclusions reached thus far on the basis of a concrete problem
where the actual form of the retarded Green function needs to be used, and partial
integrations are required. We thus consider the potentials generated by a moving
point charge (Liénard–Wiechert potentials). The charge and current densities for
the moving point charge are given as
r , t) = q δ (3) (⃗
ρ (⃗ ⃗ (t)) ,
r−R (5.254a)
J⃗ (⃗
r , t) = q δ (3) (⃗ ⃗ (t)) [ d R
r−R ⃗ (t)] , (5.254b)
dt
⃗
where R(t) is the particle trajectory. If we keep only the first term in square brackets
in Eq. (4.35) and write [see Eq. (4.36)]
c Θ(t − t′ )
r, t, r⃗′ , t′ ) ≈
GR (⃗ r − r⃗′ ∣ − c (t − t′ )) ,
δ (∣⃗ (5.255)
r − r⃗′ ∣
4π 0 ∣⃗
then the integration of the Lorenz-gauge potentials [see Eqs. (4.88a) and (4.88b)] is
almost trivial,
r, t) = ∫ d3 r′ dt′ GR (⃗
ΦL (⃗ r , t, r⃗′ , t′ ) ρ (⃗
r ′ , t′ ) , (5.256a)
1
A⃗L (⃗ r , t, r⃗′ , t′ ) J⃗ (⃗
r, t) = 2 ∫ d3 r′ dt′ GR (⃗ r ′ , t′ ) . (5.256b)
c
After the integration over d3 r′ , one easily obtains
q ′ ′ 1 ∣⃗ ⃗ ′ )∣
r − R(t
ΦL (⃗
r, t) = ∫ dt Θ (t − t ) δ (t′ − (t − )) , (5.257a)
4π0 ∣⃗ ⃗ ′ )∣
r − R(t c
q ⃗˙ ′ )
R(t ∣⃗ ⃗ ′ )∣
r − R(t
A⃗L (⃗
r, t) = ∫ dt ′
Θ (t − t ′
) δ (t ′
− (t − )) . (5.257b)
4π0 c2 ∣⃗ ⃗ ′ )∣
r − R(t c
The δ function peaks at
∣⃗ ⃗ ′ )∣
r − R(t ∣⃗ ⃗ ret )∣
r − R(t
t′ = tret = t − =t− < t, (5.258)
c c
or
∣⃗ ⃗ ret )∣
r − R(t
tret = t − , c(t − tret )2 = (⃗ ⃗ ret )) ,
r − R(t
2
(5.259)
c
so that the step function is always unity at the point where the Dirac-δ peaks. That
means that all points on the trajectory of the particle for which the retardation
condition is fulfilled, contribute to the integrals. Let us assume that the Dirac-δ
peaks only once, namely, at t′ = tret . The integration over dt′ in Eq. (5.257) still
leads to a nontrivial Jacobian. Indeed, we have
d ′ ∣⃗ ⃗ ′ )∣
r − R(t ⃗ ′ ) r⃗ − R(t
1 dR(t ⃗ ′)
(t − t + ) = 1 − ⋅ , (5.260)
dt′ c c dt′ ∣⃗ ⃗ ′ )∣
r − R(t
at t′ = tret . So, the Liénard–Wiechert potentials in Lorenz gauge read

−1
q 1 ⎛ ⃗˙ ret ) r⃗ − R(t
R(t ⃗ ret ) ⎞
ΦL (⃗
r , t) = 1− ⋅ , (5.261a)
4π0 ∣⃗ ⃗ ret )∣ ⎝
r − R(t c ∣⃗ ⃗ ret )∣ ⎠
r − R(t
−1
q ⃗˙ ret ) ⎛
R(t R(t ⃗ ret ) ⎞
⃗˙ ret ) r⃗ − R(t
A⃗L (⃗
r , t) = 1− ⋅ . (5.261b)
4π0 c2 ∣⃗ ⃗ ret )∣ ⎝
r − R(t c ∣⃗ ⃗ ret )∣ ⎠
r − R(t
⃗ ret ) ˙ ⃗ ret )
With the definitions β⃗ ≡ β(t⃗ ret ) = R(t
c
, as well as n̂ ≡ ∣⃗rr⃗−−R(t
R(t
⃗ ret )∣ and R ≡ ∣⃗
r−
⃗
R(tret )∣, we can write the scalar and vector potentials in the familiar form,
q 1 q β⃗
ΦL (⃗
r , t) = , A⃗L (⃗
r , t) = , (5.262)
4π0 R (1 − β⃗ ⋅ n̂) 4π0 c R (1 − β⃗ ⋅ n̂)
2
where we keep SI mksA units.
5.6.2 Liénard–Wiechert Potentials in Coulomb Gauge

The calculation of the Liénard–Wiechert potentials in Coulomb gauge crucially relies
on a careful consideration of the longitudinal and transverse parts of the current
density of a point charge. We recall the formulas (5.254a) and (5.254b) in the form
r, t) = q δ (3) (⃗
ρ (⃗ ⃗ (t)) ,
r−R (5.263a)
J⃗ (⃗
r, t) = q δ (3) (⃗ ⃗ (t)) [ d R
r−R ⃗ (t)] = q δ (3) (⃗ ⃗ (t)) R(t)
r−R ⃗˙ . (5.263b)
dt
According to Eq. (1.48a), the longitudinal part of the current density can be
extracted as follows,
1 ⃗ ′ ⋅ J⃗ (⃗
∇ r′ , t)
J⃗∥ (⃗
r , t) = − ∇ ⃗ ∫ d3 r′ . (5.264)
4π r − r⃗′ ∣
∣⃗
The divergence of the current density is easily calculated:
⃗ ⋅ J⃗ (⃗
∇ ⃗˙
r , t) = q R(t) ⃗ (3) (⃗
⋅ ∇δ ⃗ (t)) .
r−R (5.265)
We now calculate an integral, which we call for convenience L (its gradient is pro-
portional to J⃗∥ )
∇ ⃗ r′ , t)
⃗ ′ ⋅ J(⃗ 1
L = ∫ d3 r′ = ∫ d3 r′ ⃗˙
q R(t) ⃗ ′ δ (3) (⃗
⋅∇ ⃗ (t))
r′ − R
∣⃗ ′
r − r⃗ ∣ r − r⃗′ ∣
∣⃗
⃗˙
= ∫ d3 r′ q R(t) ⃗ ′ 1 ) δ (3) (⃗
⋅ (−∇ ⃗
r′ − R(t))
r − r⃗′ ∣
∣⃗
⃗˙ q ⃗
= R(t) ⃗ ∫ d3 r′
⋅∇ δ (3) (⃗
r′ − R(t))
r − r⃗′ ∣
∣⃗
⃗˙ ⃗ 1
= q R(t) ⋅∇ . (5.266)
∣⃗ ⃗
r − R(t)∣
The longitudinal component J⃗∥ is finally calculated as follows,

1 1 ∇ ⃗ r′ , t)
⃗ ′ ⋅ J(⃗
J⃗∥ (⃗
r, t) = − ∇ ⃗L= − ⃗ ∫ d3 r′
∇
4π 4π r − r⃗′ ∣
∣⃗
q ⃗ ⃗˙ ⃗ 1 q ⃗˙ ⃗ ∇⃗ 1
= −
∇ (R(t) ⋅ ∇ ) = − (R(t) ⋅ ∇) .
4π ∣⃗ ⃗
r − R(t)∣ 4π ∣⃗ ⃗
r − R(t)∣
(5.267)
We can verify that J⃗∥ carries the entire divergence of J,
⃗
q ⃗˙ 1
⃗ ⋅ J⃗∥ (⃗
∇ r, t) = − ⃗∇
R(t) ⋅ ∇ ⃗2
4π ∣⃗ ⃗
r − R(t)∣
⃗˙
= q R(t) ⃗ δ (3) (⃗
⋅∇ ⃗
r − R(t)) ⃗ ⋅ J⃗ (⃗
=∇ r , t) . (5.268)
In principle, we could now simply calculate the transverse component J⃗⊥ (⃗ r , t)
as the difference of the total current density J(⃗ ⃗ r , t) and the longitudinal component
J⃗∥ (⃗
r, t), namely, J⃗⊥ (⃗ ⃗ r , t) − J⃗∥ (⃗
r , t) = J(⃗ r , t). However, it is more instructive to
calculate the transverse component according to Eq. (1.48b),
1 ⃗ ⃗ ′ × J⃗ (⃗
∇ r′ , t)
J⃗⊥ (⃗
r , t) = ∇ × ∫ d3 r′ . (5.269)
4π r − r⃗′ ∣
∣⃗
We calculate the vector-valued integral
⃗ ′ ⃗ r′ , t)
⃗ = ∫ d3 r′ ∇ × J (⃗
K = ∫ d3 r′
1
⃗ ′ × (q δ (3) (⃗
∇ r′ − R ⃗ (t)) R(t))
⃗˙
∣⃗ ′
r − r⃗ ∣ r − r⃗′ ∣
∣⃗
1 ⃗˙ ⃗ ′ δ (3) (⃗ ⃗
= − q ∫ d3 r′ R(t) × ∇ r′ − R(t))
r − r⃗′ ∣
∣⃗
⃗ ⃗˙ 1
= q ∫ d3 r′ δ (3) (⃗
r′ − R(t)) (R(t) ⃗′
×∇ )
r − r⃗′ ∣
∣⃗
⃗˙ 1
= − q R(t) ⃗
×∇ . (5.270)
∣⃗ ⃗
r − R(t)∣
The result for the transverse component of the current density therefore is
1 ⃗ ⃗ 1 ⃗˙ 1
J⃗⊥ (⃗
r , t) = ∇×K = ⃗ ×∇
q R(t) × ∇ ⃗ . (5.271)
4π 4π ∣⃗ ⃗
r − R(t)∣
The transverse component finally is found as follows,
1 ⃗˙ 1
J⃗⊥ (⃗
r , t) = ⃗ ×∇
q R(t) × ∇ ⃗
4π ∣⃗ ⃗
r − R(t)∣
q ⃗˙ 1 q ⃗˙ 1
= ⃗∇
R(t) ⋅ ∇ ⃗ − ⃗2
R(t) ∇
4π ∣⃗ ⃗
r − R(t)∣ 4π ∣⃗ ⃗
r − R(t)∣
q ⃗˙ 1 ⃗˙ ⃗
= ⃗∇
R(t) ⋅ ∇ ⃗ + q R(t) δ (3) (⃗
r − R(t)) = −J⃗∥ (⃗ ⃗ r , t) ,
r, t) + J(⃗
4π ∣⃗ ⃗
r − R(t)∣
(5.272)
verifying the identity J⃗∥ (⃗
r , t) + J⃗⊥ (⃗ ⃗ r , t).
r , t) = J(⃗
The results can be summarized as follows,
q ⃗˙ 1
J⃗∥ (⃗ ⃗ (− R(t)
r, t) = ∇ ⃗
⋅∇ ⃗ (⃗
) = ∇J r , t) , (5.273a)
4π ∣⃗ ⃗
r − R(t)∣
q ⃗˙ 1
J (⃗
r, t) = − ⃗
R(t) ⋅ ∇ , (5.273b)
4π ∣⃗ ⃗
r − R(t)∣
1 ⃗˙ 1
J⃗⊥ (⃗
r, t) = ⃗ ×∇
q R(t) × ∇ ⃗ , (5.273c)
4π ∣⃗ ⃗
r − R(t)∣
where we define the function J in Eq. (5.273b). The longitudinal and transverse
components of J⃗ enter the formulas for the scalar and vector potentials in Coulomb
gauge. We recall Eq. (4.102),
1 ∂
r, t) = ∫ dt′ ∫ d3 r′ GR (⃗
ΦC (⃗ r − r⃗′ , t − t′ ) (ρ(⃗
r ′ , t′ ) + r′ , t′ )) .
J (⃗ (5.274)
c2 ∂t′
As before, the retarded Green function may be used in the approximation [see
Eq. (4.36)],
c Θ(t − t′ )
r − r⃗′ , t − t′ ) ≈
GR (⃗ r − r⃗′ ∣ − c (t − t′ )) .
δ (∣⃗ (5.275)
r − r⃗′ ∣
4π 0 ∣⃗
So,
q ∂ ⃗˙ ′ 1
ΦC (⃗
r , t) = ΦL (⃗
r , t) − dt′ ∫ d3 r′ GR (⃗
r − r⃗′ , t − t′ ) ′ R(t ⃗′
)⋅∇
4π c2 ∫ ∂t ∣⃗′ ⃗ ′ )∣
r − R(t
= ΦL (⃗
r , t) + ΦS (⃗
r, t) , (5.276)
where ΦL is the Lorenz-gauge expression (4.88a). No attempt is made here to

simplify the expression found for the supplementary term ΦS . We can also write
the Coulomb-gauge Liénard–Wiechert vector potential as
1
A⃗C (⃗ r − r⃗′ , t − t′ ) J⃗⊥ (⃗
r , t) = 2 ∫ d3 r′ dt′ GR (⃗ r ′ , t′ )
c
1
= r − r⃗′ , t − t′ ) (J⃗ (⃗
d3 r′ dt′ GR (⃗ r′ , t′ ) − J⃗∥ (⃗
r′ , t′ ))
c2 ∫
q 1
= A⃗L (⃗
r , t) + 3 ′ ′ ⃗˙ ′ ) ⋅ ∇
r − r⃗′ , t − t′ ) (R(t
∫ d r dt GR (⃗ ⃗ ′ ) (∇
⃗′ )
4πc 2
∣⃗′ ⃗ ′ )∣
r − R(t
= A⃗L (⃗
r , t) + A⃗S (⃗
r , t) , (5.277)
where A⃗L (⃗
r , t) is the Lorenz-gauge expression, given in Eq. (4.88b). It is instructive
to write the supplementary term A⃗S (⃗ r, t) as a gradient, after an integration by parts
⃗ ′ → −∇
and a change ∇ ⃗ when acting on the Green function,
q 1
A⃗S (⃗ ⃗
r, t) = ∇ d3 r′ dt′ GR (⃗ ⃗˙ ′ ) ⋅ ∇
r − r⃗′ , t − t′ ) R(t ⃗′
4πc2 ∫
. (5.278)
∣⃗′ ⃗ ′ )∣
r − R(t
Evidently, the curl of A⃗S vanishes and there is no additional contribution to the
magnetic induction field. The supplementary electric field must necessarily vanish,
∂
E⃗S (⃗ r , t) − A⃗S (⃗
⃗ S (⃗
r, t) = − ∇Φ r , t)
∂t
q ∂ ⃗˙ ′ 1
= ⃗ R (⃗
dt′ ∫ d3 r′ ∇G r − r⃗′ , t − t′ ) ′ R(t ⃗′
)⋅∇
4π c2 ∫ ∂t ∣⃗′ ⃗ ′ )∣
r − R(t
q ∂ ⃗˙ ′ ) ⋅ ∇ ⃗′
⃗′ ∇ 1
− ∫ d3 r′ dt′ GR (⃗
r − r⃗′ , t − t′ ) R(t
4πc 2 ∂t ∣⃗′ ⃗ ′ )∣
r − R(t
q ′ 3 ′ ∂ ⃗˙ ′ ⃗ ′ ⃗ ′ 1
= − r − r⃗′ , t − t′ ) ′ R(t
∫ dt ∫ d r GR (⃗ )⋅∇ ∇
4π c2 ∂t ∣⃗′ ⃗ ′ )∣
r − R(t
q ∂ ⃗˙ ′ 1
3 ′ ′
∫ d r dt GR (⃗
+ r − r⃗′ , t − t′ ) ′ R(t )⋅∇ ⃗′
⃗′ ∇ = 0,
4πc2 ∂t ∣⃗ ⃗ ′ )∣
r′ − R(t
(5.279)
and it does, confirming the gauge invariance of the fields.
5.7 Exercises
● Exercise 5.1: Fill in the missing intermediate steps in Eq. (5.4). In particular,
explain why the Heaviside step function Θ(t − t′ ) does not figure in the final result.
● Exercise 5.2: Show that the function Z(x) = xα Jn (β xγ ) fulfills
∂ 2 Z 1 − 2α ∂Z α2 − n2 γ 2
2
+ + (β 2 γ 2 x2γ−2 + ) Z = 0. (5.280)
∂x x ∂x x2
Hint: Start from the relations
z 1/γ
Z(x) = xα Jn (β xγ ) , z = β xγ , x=( ) , (5.281)
β
z −α/γ z 1/γ
Jn (z) = x−α Z(x) = ( ) Z [( ) ] . (5.282)
β β
Then, insert these into Bessel’s differential equation which reads as
∂2 1 ∂ z 2 − n2 ∂2 1 ∂ z 2 − n2 z −α/γ z 1/γ
( + + )J n (z) = ( + + ){( ) Z [( ) ]} = 0 .
∂z 2 z ∂z z2 ∂z 2 z ∂z z2 β β
(5.283)
In particular, carry out all differentiations with respect to z and then replace
z → β xγ .
● Exercise 5.3: Tabulate the first few Hankel functions and convince yourself that
these can be written in terms of exp(iρ) and powers of ρ. Show that
(1) eiρ
h0 (ρ) = − i , (5.284a)
ρ
(1) eiρ eiρ
h1 (ρ) = − −i 2 , (5.284b)
ρ ρ
iρ iρ
(1) ie 3e 3ieiρ
h2 (ρ) = − 2 − 3 . (5.284c)
ρ ρ ρ
Show, by an explicit calculation for the three example cases above, that as ρ → +∞,
(1) ei(ρ−π/2) +1 e
iρ
ei(−π/2) = (e−iπ/2 ) = (−i) ,

h (ρ) → −i = (−i) , ρ → ∞,
ρ ρ
(5.285)
i.e., that the Hankel functions of different order only differ by a complex phase in
the limit ρ → +∞. Then, repeat the above exercise for the Hankel functions of the
(2) (2) (2)
second kind, h0 (ρ), h1 (ρ), h2 (ρ).
● Exercise 5.4: Verify (i) the property (5.140) and (ii) Eq. (5.141) with explicit
reference to the Levi–Cività tensor.
● Exercise 5.5: We saw that the action of the photon spin operator is given by

the matrices ( k )ij = −i ijk , where k is the kth Cartesian spin matrix. Verify that

you can alternatively write the spin operator as ⃗ = i 3×3 ×, where × is the vector
product. Hint: You first have to think about how to interpret the expression 3×3 ×.
● Exercise 5.6: Consider the two-dimensional Pauli matrices,
01 0 −i 1 0
σ1 = ( ), σ2 = ( ) σ3 = ( ). (5.286)
10 i 0 0 −1

Verify that the spin operator matrices i = σi /2 (with i = 1, 2, 3) fulfill an analogous

relation as compared to Eq. (5.140), namely,

j ] = i ijk
k ,
[ i, (5.287)
where the Einstein summation convention is used on the right-hand side [even if
the indices are all lower indices, see Eq. (1.45)].
● Exercise 5.7: Verify that M ⃗ (K) (k, r⃗), and L
⃗ (K) (k, r⃗), N ⃗ (K) (k, r⃗) fulfill the
jμ jμ jμ
Helmholtz equation
⃗ (K) (k, r⃗) = (∇
⃗ 2 + k2 ) M
(∇ ⃗ (K) (k, r⃗) = (∇
⃗ 2 + k2 ) N ⃗ (K) (k, r⃗) = 0 . (5.288)
⃗ 2 + k2 ) L
jμ jμ jμ
● Exercise 5.8: Show Eq. (5.149).

● Exercise 5.9: Verify the equivalence of the representations (5.164), (5.171),
and (5.172). In the proof, one uses Eqs. (5.29) and (5.148).
● Exercise 5.10: Show that Eq. (5.178) can alternatively be written as
⃗0 (⃗
E ⃗ (1) (k, r⃗) + nm N
r ) = −k 2 Z0 ∑ (mm M ⃗ (1) (k, r⃗)) . (5.289)
m m
m
√
where Z0 = μ0 /0 is the impedance of the vacuum, around 377 Ω.
● Exercise 5.11: We have encountered the Clebsch–Gordan coefficients. Show

that
1
s = ∑ C1q 00
′ 1q ′′ uq ′ vq ′′ = − √ ⃗ ⋅ v⃗
u (5.290)
′
qq ′′ 3
where the components uq and vq are given in the spherical basis, i.e., we would have
for the reference vector r⃗ = ∑ xq e⃗q∗ [see Eq. (5.126)],
1 1
x+1 = − √ (x + i y) , x0 = z , x−1 = √ (x − i y) , (5.291)
2 2
with [see Eq. (5.127)]
1 1
2 2
Which summation limits have to be used for the sums over q ′ and q ′′ ?
● Exercise 5.12: Consider the relations (5.148a) and (5.148c),
1
Y⃗jj−1
μ (θ, ϕ) = − √
⃗ − j r̂) Yjμ (θ, ϕ) ,
(i r̂ × L (5.293a)
j (2j + 1)
1
Y⃗jj+1
μ (θ, ϕ) = − √
⃗ + (j + 1) r̂) Yjμ (θ, ϕ) .
(i r̂ × L (5.293b)
(j + 1) (2j + 1)
With the help of these relations, and the recursion relations for the Bessel functions,
show that
√ √
⃗ j + 1 (K) ⃗ j
(k r) Y⃗jj+1
(K) j−1 (K)
Njμ (k, r⃗) = fj−1 (k r) Yj μ (θ, ϕ) − f μ (θ, ϕ) ,
2j + 1 2j + 1 j+1
⎧ d[kr f (K) (k r)] √ ⎫
1 ⎪
⎪ ⎪
⎪
[ir̂ × Y⃗jμ
j j (K)
= ⎨ (θ, ϕ)] − r̂ ( + 1)fj (k r) Yjμ (θ, ϕ)⎬ ,
kr ⎪
⎪ d(kr) ⎪
⎪
⎩ ⎭
(5.294)
and that
√ √
⃗ j ⃗ j + 1 (K)
(k r) Y⃗jj+1
(K) (K) j−1
Ljμ (k, r⃗) = fj−1 (k r) Yj μ (θ, ϕ) + f μ (θ, ϕ)
2j + 1 2j + 1 j+1
√ (K)
1 (K) d[f (k r)]
= j(j + 1) fj (k r) (ir̂ × Y⃗jμ
j
(θ, ϕ)) − r̂ Yjμ (θ, ϕ) .
kr d(k r)
(5.295)
√ √
j + 1 ⃗ j−1 j ⃗ j+1
Yj μ (θ, ϕ) + Y (θ, ϕ) = −i (r̂ × Y⃗jjμ (θ, ϕ)) . (5.296)
2j + 1 2j + 1 j μ
∂
⃗ × (⃗
∇ ⃗ f (⃗
r × ∇) r) = [⃗
r∇ ⃗
⃗2 − ∇ r] f (⃗
r) , (5.297)
∂r
for any test function f (⃗
r ).
● Exercise 5.14: We have shown that in the long-wavelength limit [see Eq. (5.253)],
√
ic j +1 j
njμ ≈ − k qjμ . (5.298)
(2j + 1)!! j
Complete the calculation of the corresponding electric dipole (j = 1) contribution
to A⃗0 (⃗
r ), based on the representation (5.170),
∞ j
A⃗0 (⃗ ⃗ (1) (k, r⃗) + njμ N
⃗ (1) (k, r⃗) + ljμ L
jμ
⃗ (1) (k, r⃗))
jμ
j=0 μ=−j
∞ j
⃗ (k, r⃗) ,
→ ∑ ∑ njμ N
(1)
(5.299)
jμ
j=0 μ=−j
and verify that you obtain the familiar form of the electric dipole term.
x2 ′ ′
∫ dx f (x) g′ (x) = ′ (′ + 1) − ( + 1) [f (x) g′ (x) − f (x) g′ (x)] , (5.300)
with the help of the differential equation (5.27),

∂2 2 ∂ ( + 1)
( 2
+ − − 1) f (x) = 0 , (5.301)
∂x x ∂x x2
where f can be j or y , or a superposition (spherical Hankel function).
● Exercise 5.16: This exercise summarizes important formulas and can be used
as a reference as well. We know that the coupling of the vector potential to the
current density reads as [see Eq. (5.158)],
r) = 2 ∫ d3 r′ R (k, r⃗ − r⃗′ ) ⋅ J⃗0 (⃗

A⃗0 (⃗
c
1

r′ ) . (5.302)
We had encountered three ways to expand the potential into multipole moments.
(i) The first expansion [Eq. (5.87)] reads as follows:
∞
R(⃗r − r⃗′ , k) = 1 ∑ 3×3 .

(1) ∗ ′ ′
∑ i k j (k r< ) h (k r> ) Ym (θ, ϕ) Ym (θ , ϕ )
0 =0 m=−
(5.303)
We define multipole moments as follows:
p⃗m = ∫ d3 r j (k r) J⃗0 (⃗ ∗
r) Ym (θ, ϕ) . (5.304)
The vector potential is obtained as follows,

∞
A⃗0 (⃗
(1)
r ) = i μ0 k ∑ ∑ p⃗m h (k r) Ym (θ, ϕ) . (5.305)
=0 m=−
(ii) The second expansion [Eq. (5.161)] involves the vector spherical harmonics,
∞ j j+1
R(⃗r −⃗r′ , k) = ik ∑ ∑ ∑ j (k r< ) h (k r> ) Y⃗jμ
(1)
(θ, ϕ)⊗Y⃗jμ
∗ ′
(θ , ϕ′ ) . (5.306)
0 j=0 μ=−j =j−1
The multipole moments read as follows [see Eq. (5.162)],
k
pjμ = ∫ d3 r j (k r) J⃗0 (⃗
r ) ⋅ Y⃗jμ
∗
(θ, ϕ) ≈ ∫ d r r J⃗0 (⃗
3
r ) ⋅ Y⃗jμ
∗
(θ, ϕ) . (5.307)
(2 + 1)!!
The vector potential is given as follows,
∞ j j+1
A⃗0 (⃗ pjμ h (k r) Y⃗jμ
(1)
r ) = i μ0 k ∑ ∑ ∑

(θ, ϕ) . (5.308)
j=0 μ=−j =j−1
(iii) The third expansion [Eq. (5.170)] involves the following formula for the
tensor Helmholtz Green function,
∞ j
R(k, r⃗ − r⃗′ ) = ik ∑ ⃗ (1) (k, r⃗> ) ⊗ M
∑ (M jμ
⃗ (0)∗ (k, r⃗< )
jμ
0 j=0 μ=−j
⃗ (1) (k, r⃗> ) ⊗ N

+N ⃗ (0)∗ (k, r⃗< ) + L
⃗ (1)(k, r⃗> ) ⊗ L
⃗ (0)∗ (k, r⃗< )) . (5.309)
jμ jμ jμ jμ
The multipole moments are [see Eqs. (5.168a), (5.168b) and (5.168c)]
mjμ = ∫ d3 r′ J⃗0 (⃗ ⃗ (0)∗

r′ ) ⋅ M ′ ′
jμ (θ , ϕ ) , (5.310a)
njμ = ∫ d3 r′ J⃗0 (⃗ ⃗ (0)∗ (θ′ , ϕ′ ) ,

r′ ) ⋅ N jμ (5.310b)
ljμ = ∫ d3 r′ J⃗0 (⃗ ⃗ (0)∗ (θ′ , ϕ′ ) .

r′ ) ⋅ L jμ (5.310c)
The vector potential is obtained as follows [see Eq. (5.170)],

∞ j
A⃗0 (⃗ ⃗ (1) (k, r⃗) + njμ N
⃗ (1) (k, r⃗) + ljμ L
jμ
⃗ (1) (k, r⃗)) .
jμ (5.311)
j=0 μ=−j
The magnetic, electric and longitudinal multipole functions read as follows [see
Eqs. (5.164a), (5.164b) and (5.164c)],
⃗ (K) (k, r⃗) = √ 1

M
(K)
fj (k r) LY⃗ jμ (θ, ϕ) , (5.312a)
jμ
j(j + 1)
⃗ (K) (k, r⃗) = i ∇
N ⃗ ×M ⃗ (K)(k, r⃗) , (5.312b)
jμ jμ
k
1
⃗ (K) (k, r⃗) = ∇
L jμ
⃗ (fj(K) (k r) Yjμ (θ, ϕ)) . (5.312c)
k
Treat the following current distributions in all three variants of the multipole
expansion.
(i) Assume that
I0 z2 r2
J⃗0 (⃗
r ) = 2 (3 2 − 1) exp (− 2 ) êz , (5.313)
a r a
and use all three multipole formalisms. You may replace the Bessel functions by
their leading asymptotic behavior for small argument. Hint: You might obtain
√
π π 3 2 3
n10 = − √ k a I0 ,
2 3
n30 = − k a I0 , (5.314)
25 6 25 7
π 3π
l10 = − √ k 2 a3 I0 , l30 = √ k 2 a3 I0 . (5.315)
25 3 50 7
Do all the magnetic multipoles vanish, i.e., mjμ = 0?
(ii) Assume that
I0 z2 r2
J⃗0 (⃗
r) = 2 (3 2 − 1) exp (− 2 ) e⃗+1 . (5.316)
a r a
Hint: You might obtain
π π
m21 = − √ k 2 a3 I0 , n11 = − √ k 2 a3 I0 , (5.317)
10 10 50 6
√
π 2 2 3 π
n31 = − k a I0 , l11 = √ k 2 a3 I0 , (5.318)
25 7 50 3
√
π 3 2 3
l31 = k a I0 . (5.319)
25 14
Do all the other multipoles vanish, i.e., mjμ = 0?
(iii) Now here is another current distribution,
I0 z y x r2
J⃗0 (⃗
r ) = 2 ( êx − êy ) exp (− 2 ) , (5.320)
a a a a a
which needs to be analyzed in terms of the electric and magnetic multipoles. Hint:
You might obtain
π
m20 = i √ k 2 a3 I0 . (5.321)
4 30
Do all the electric and longitudinal multipoles vanish?
● Exercise 5.17: Consider the equation defining the retarded time, tret = t − ∣⃗ r−
⃗ ret )∣/c, for uniform motion in only one dimension (along the x direction), i.e.,
R(t
⃗
for r⃗ = r êx , r > 0, and R(t) = v0 t êx . We thus consider only observation points r⃗
lying in the x axis, reducing the problem to one dimension.
(i) Solve the defining equation for tret , replacing all vector quantities by their x
components.
(ii) Calculate the scalar and vector potentials (Liénard–Wiechert potentials),
−1
q 1 ⎛ R(t ⃗ ret ) ⎞
⃗˙ ret ) r⃗ − R(t
Φ (⃗
r , t) = 1− ⋅ , (5.322a)
4π0 ∣⃗ ⃗ ret )∣ ⎝
r − R(t c ∣⃗ ⃗ ret )∣ ⎠
r − R(t
Φ (⃗
r, t) ⃗˙
A⃗ (⃗
r , t) = R(tret ) , (5.322b)
c2
and show that for the particular case, the final formulas are consistent with “no
retardation occurring”, because
q 1 q v⃗0
Φ (⃗
r, t) = , A⃗ (⃗
r , t) = . (5.323)
4π0 r − v0 t 4π0 c r − v0 t
2
The change in the interaction distance ∣⃗ ⃗ ret )∣ by the retardation is exactly

r − R(t
−1
⃗˙ ret )
R(t ⃗ ret )
r⃗−R(t
compensated by the Jacobian factor (1 − c
⋅ ∣⃗ ⃗ ret )∣ )
r −R(t
. Show your work and
all intermediate steps.
● Exercise 5.18: Investigate the scalar and vector potentials (5.323), but this time
for general r⃗, constant speed v⃗0 ,
v⃗0 R⃗ 0 ⋅ β⃗0 1 1
⃗
R(t) = v⃗0 , β0 = , = cos(θ0 ) = 1 − θ02 + θ02 + O(θ06 ) . (5.324)
c ⃗
∣R0 ∣ β0 2! 4!
Show that the first retardation correction for small θ0 reads as follows,
q 1 β 2 θ2
Φ (⃗
r, t) = (1 + 0 0 + O(θ04 )) . (5.325)
4π0 ∣⃗
r − v⃗0 t∣ 2
Hint: The formulas r − v0 tret = c (t − tret ), with tret = (c t − r)/(c − v0 ), may be
helpful. Then, calculate the term of order θ04 .
● Exercise 5.19: Verify that A⃗C (⃗ r , t) given in Eq. (5.277) is transverse, by an
explicit calculation. Hint: This may be explicitly verified in various ways; for
example, one may take the divergence of both sides of Eq. (5.277), then use the
⃗ R (⃗
identity ∇G ⃗ ′ GR (⃗
r − r⃗′ , t−t′ ) = −∇ r − r⃗′ , t−t′ ), integrate by parts under the integral
sign, and take into account that J⃗∥ carries the entire divergence of J, ⃗ according to
Eq. (5.268).
Chapter 6
Electrodynamics in Media
6.1 Overview
The purpose of this chapter is threefold. (i) We shall derive, based on a microscopic
model, the phenomenological Maxwell equations which involve the so-called dielec-
⃗ r , t) and the magnetic field H(⃗
tric displacement D(⃗ ⃗ r, t). This allows us to give
a definite physical interpretation of the dielectric displacement and reaction of the
material. We shall also study the frequency ranges over which the phenomenological
treatment is valid. (ii) We shall introduce Fourier transforms, and briefly discuss
the Maxwell equations in the mixed coordinate-frequency representation. This also
provides for an alternative viewpoint on the electromagnetic waves. (iii) We apply
the wisdom gathered to the frequency-dependent dielectric constant r (ω) of both
dilute materials as well as dense materials, and explore how causality dictates the
Kramers–Kronig relations which have to be fulfilled by the dielectric constant.
In particular, three examples of possible functional forms of dielectric permit-
tivities are discussed. First, based on a modification of the Sellmeier equation,
we shall study the connection of the atomic polarizability and the so-called Drude
model, which is relevant to the analysis of near-perfect conductors (see Sec. 6.4).
The dielectric permittivity is analyzed for dense materials, in Sec. 6.4.4, with con-
comitant necessary modifications to the functional form describing the dielectric
response function. Wave propagation in a dispersive medium (Sec. 6.5) is an
excellent testbed for the study of advanced mathematical techniques like the method
of steepest descent. Finally, the Kramers–Kronig relationships (Sec. 6.6) allow us
to connect the real (dispersive) and imaginary (absorptive) parts of the dielectric
response function, and add important tools in the study of the optical properties of
a medium.
6.2 Microscopic and Macroscopic Equations
6.2.1 Macroscopic Equations and Measurements

We consider the derivation of the macroscopic (phenomenological) Maxwell equa-
⃗ r, t) and the magnetic field
tions which involve the dielectric displacement D(⃗
221
⃗ r , t). To this end, it is instructive to first consider the response of materials

H(⃗
to electromagnetic radiation in different frequency ranges, as this determines the
physical conditions under which the phenomenological equations are valid. Let us
consider a light frequency range
c
ν∼ , λ ∼ a0 , (6.1)
a0
where ν is the frequency of the electromagnetic radiation, and λ the wavelength.
The speed of light is c = λ ν. The Bohr radius is
̵
h
a0 = ≈ 0.529 Å , (6.2)
α me c
where me is the electron mass, and λ is the wavelength. The wavelength range (6.1)
is commonly used for x-ray crystallography. This is because in a dense crystal
lattice, the distance among atoms is of the order of a Bohr radius, and therefore,
incoming x-rays are collectively scattered by crystal planes. In a Laue photograph,
a single crystalline sample is irradiated by a coherent x-ray source, which results
in defined maxima of the scattered intensity (corresponding to the bright spots
in the photograph). In a Debye–Scherrer photograph, a polycrystalline sample is
irradiated by x-rays. The crystalline grains are oriented in all spatial directions,
and therefore the Laue spots become concentric rings.
Note that Laue and Debye–Scherrer crystal diffraction could never have been
discovered on the basis of the macroscopic Maxwell equations alone. Rather, the
macroscopic Maxwell equations describe the perturbations to the average propaga-
tion velocity of light waves due to the presence of a medium, which is being felt
by the traveling light wave as a uniform medium. For that approximation to be
valid, the wavelength of the traveling wave must be much longer than the average
structural distances in the sample,
c
ν≪ , λ ≫ a0 . (6.3)
a0
Specifically, the Bragg condition n λ = 2 d sin θ governing the refracted waves from
the crystal planes could never have been discovered on the basis of the phenomeno-
logical Maxwell equations (in order to derive the Bragg condition, one investigates
the path difference of light rays emerging from crystal layers displaced by a dis-
tance d). The Bragg condition can be derived based on a consideration of the path
difference of light rays emitted from different crystal planes, with θ being the inci-
dent angle of the x-ray with respect to the crystal plane, or, equivalently the angle
of the normal to the crystal plane with the normal to the incident light “sheet”.
In the frequency range
c
ν≫ , λγ ≪ a0 , (6.4)
a0
the scattering takes place not from individual atoms, but from individual electrons
inside the sample. We can even treat the electrons as stationary, the reason being
that the momentum p of the incoming photon is much larger than the average
Electrodynamics in Media 223
momentum of a bound electron. According to the de Broglie formula p = h/λ, we

can write for the frequency range (6.4),
h hν h
p= = ≫ = 2παme c ∼ αme c . (6.5)
λ c a0
The classical “speed” of a bound electron is of the order of αc, where α ≈ 1/137
is the fine-structure constant, and so αme c is a typical electron momentum in the
bound state of an atom. In the frequency range (6.4), the photon momentum is
much larger than αme c, implying that the electron can be assumed to be stationary.
We thus expect to use the macroscopic equations and tabulated electric and
magnetic susceptibilities only if the electromagnetic fields have wavelengths which
are large compared to the characteristic lengths of the system. Specifically, we must
assume that simultaneously
λ ≫ a0 , λ ≫ ⟨d⟩ , λ ≫ rρ . (6.6)
Here, ⟨d⟩ is the average spacing of atoms in the medium (for dilute gases, the
corresponding frequency range is far in the infrared). Furthermore, rρ is a length
scale that characterizes density fluctuations inside the sample. For the frequency of
the perturbation (light frequency), this means that simultaneously,
c c c
ν≪ , ν≪ , ν≪ . (6.7)
a0 ⟨d⟩ rρ
We can then choose a distance R, subject to the condition
λ ≫ R ≫ a0 , ⟨d⟩ , rρ , (6.8)
and average the fields over a volume of dimension R3 . This averaging operation (spa-
tial average at constant time) leads to the macroscopic Maxwell equations formu-
⃗ r , t) and the dielectric displacement D(⃗
lated in terms of the magnetic field H(⃗ ⃗ r , t).
These quantities include the response of the medium to the external perturbations,
and an averaging of the fields over certain length scales.
6.2.2 Macroscopic Fields from Microscopic Properties

We consider the limit of low-frequency perturbations, characterized by the condi-
tions (6.6) and (6.7),
c c c c
λ ≫ R ≫ a0 , ⟨d⟩ , rρ , ν≪ ≪ , , , (6.9)
R a0 ⟨d⟩ rρ
where R3 measures the dimension of our sampling (or averaging) volume. The
dimension of the sampling volume is larger than the Bohr radius, larger than the
average interatomic distance, and larger than the distance scale rρ over which den-
sity fluctuations in the sample average out. Our classical approach is inspired by
the treatment described in Ref. [26] and provides for an intuitive interpretation of
the macroscopic Maxwell equations.
We consider a material with a typical density of ∼ 1023 atoms (or molecules) per
cubic centimeter, cm3 , i.e., one gram per mol. In a typical case, we can choose
100 Å < R < 1000 Å . (6.10)
In the volume R3 , there are approximately 106 atoms, and the charge constituents
(electrons, nuclei) can be treated as point charges. There is a subtle point. We
have previously classified the regime (6.9) as the regime of very-low-frequency light
waves. Indeed, the condition
c
λ ≫ a0 , ν≪ , (6.11)
a0
means that the wavelength is much longer than the typical size of an atom. This
condition is fulfilled by the frequencies corresponding to typical optical transitions of
atoms. These wavelengths typically are much larger than the Bohr radius. Namely,
we have for a typical atomic wavelength λ̄0 ,
a0 c α2 me c2
λ̄0 ∼ ∼ 1000 Å , ν ∼α = ̵ ∼ 3 × 1015 Hz . (6.12)
α a0 h
This is just in the range of atomic transition frequencies, or, in a classical picture,
equal to the typical frequency of electronic charge vibrations in a typical atom. In
the optical frequency range, the formalism to be outlined below is applicable: One
can define a meaningful, frequency-dependent dielectric permittivity (ω) and mag-
netic permeability μ(ω) for the optical frequency range. Of course, the formalism
remains valid for even longer-than-optical wavelengths, or, low-frequency driving of
the macroscopic sample.
In discussing the microscopic model to be outlined in the following, we might,
in principle, distribute and classify the nuclear and electronic charges at our will.
We denote the positions of the atomic nuclei by r⃗n and the relative positions of the
j electrons bound to the nth nucleus as r⃗jn , so that the absolute position of the jth
electron of the nth atom is at r⃗n + r⃗jn . Because almost all the mass of the atom
is in the nucleus, this approach is equivalent to defining r⃗n as the center of mass
of the atom, up to reduced-mass corrections of order me /mp ≈ 1/2000, where me is
the mass of the electron, and mp is the mass of the proton.
We shall be interested in the averages of the microscopic electric field, e⃗(⃗ r , t),
and of the microscopic magnetic induction field ⃗b(⃗ r , t), at the point r⃗ and at time
t. The vector r⃗ is understood to represent a position vector in a much larger region
of space (of the order of cm3 ) at which the macroscopic electric field E(⃗ ⃗ r , t), and
the macroscopic magnetic induction field B(⃗ ⃗ r, t) are set equal to the average of the
microscopic fields over the volume of dimension R3 . The fields shall be assumed
to be redefined in the indicated sense, for the remainder of the derivation of the
macroscopic Maxwell equations, i.e., within Sec. 6.2. With the definitions r⃗ − r⃗′ ≡ ρ⃗
and r⃗ − ρ⃗ = r⃗′ , we have
⃗ r , t) ≡ ⟨⃗
E(⃗ e(⃗r , t)⟩ = ∫ d3 ρ f (⃗ ρ, t) = ∫ d3 r′ f (⃗
r − ρ⃗) e⃗(⃗ r′ ) e⃗(⃗
r − r⃗′ , t) , (6.13a)
Ê 3 Ê 3
⃗ r , t) ≡ ⟨⃗b(⃗
B(⃗ r − ρ⃗) ⃗b(⃗
r , t)⟩ = ∫ d3 ρ f (⃗ r′ ) ⃗b(⃗
ρ, t) = ∫ d3 r′ f (⃗ r − r⃗′ , t) . (6.13b)
Ê 3 Ê 3
r′ ) is a normalized distribution function, which is used in order to average

Here f (⃗
the field over a volume of radius R. Both forms of the averaging operation outlined
in Eq. (6.13) will become useful in the following. The definite form of the normalized
distribution function f is not crucial, and it is convenient to use a simple normalized
Gaussian function
1 r′2
r′ ) =
f (⃗ exp (− ), 3 ′
r′ ) = 1 .
∫ d r f (⃗ (6.14)
π 3/2 R3 R2
A particular, important result is that the averaging of a Dirac-δ function reproduces
the sampling function itself,
⟨δ (3) (⃗ r − r⃗0 ) − ρ⃗) δ (3) (⃗

r − r⃗0 )⟩ = ∫ d3 ρ f ((⃗ ρ) = f (⃗
r − r⃗0 ) . (6.15)
This result is heavily used in the following. The function (6.14) has the advantage
that the distribution falls off smoothly at the edge of the volume. It is clear that the
gradient operator and the time differential operator commute with the averaging
operation,
⃗ ⋅ ⟨⃗
∇ e(⃗
r , t)⟩ = ∫ f (⃗ ⃗ ⋅ e⃗(⃗
r′ ) ∇ ⃗ ⋅ e⃗(⃗
r − r⃗′ , t) d3 r′ = ⟨∇ r , t)⟩ , (6.16)
∂ ∂ ∂
⟨⃗
e(⃗ r′ ) e⃗(⃗
r , t)⟩ = ∫ f (⃗ r − r⃗′ , t) d3 r′ = ⟨ e⃗(⃗
r , t)⟩ . (6.17)
∂t ∂t ∂t
The total charge density ρ(⃗
r, t) is the sum of a conduction band charge density
ρc (⃗
r , t), with
r, t) = ∑ qc δ (3) (⃗
ρc (⃗ r − r⃗c (t)) , (6.18)
c
and an atomic/ionic charge density ρa (⃗

r, t), where
r , t) = ∑ qjn δ (3) (⃗
ρa (⃗ r − r⃗n (t) − r⃗jn (t)) , (6.19)
nj
so that the total charge density ρ(⃗

r, t) is
r , t) = ρc (⃗
ρ(⃗ r , t) + ρa (⃗
r , t) . (6.20)
The charge and current densities mentioned above have the right physical dimension:
The Dirac-δ (3) has dimension of inverse length to the third power. The dimension
of J⃗ is found to be equal to the charge per time per area, which is the correct
dimension for a current density.
We reemphasize that we use a model in which all particles are described as
classical point particles moving on classical trajectories. (This model, of course,
has obvious limitations in describing a system of atoms, whose constituent particles
are known to be described by quantum mechanics. Still, it is instructive to consider
a classical model composed of point particles and to see how the averaging operation
leads to the phenomenological Maxwell equations.)
In Eq. (6.19), q1n = −Z e is the charge of the nth nucleus, where e is the elec-
tron charge and Z is the nuclear charge number. The variable j = 2, . . . , jmax (n)
enumerates the charges qjn = e of the electrons in the nth atom (we still keep the
subscript in order to identify the individual electrons). The current densities are
d⃗
rc (3)
J⃗c (⃗
r , t) = ∑ qc δ (⃗ r − r⃗c (t)) , (6.21)
c dt
rn d⃗
d⃗ rjn (3)
J⃗a (⃗
r , t) = ∑ qjn ( + ) δ (⃗ r − r⃗n (t) − r⃗jn (t)) , (6.22)
nj dt dt
⃗ r , t) = J⃗c (⃗
J(⃗ r , t) + J⃗a (⃗
r, t) . (6.23)
The positions r⃗n and r⃗jn still depend on time.
The following definitions and relations are relevant to the discussion (the electron
charge is denoted as e),
r , t) = ∑ qc δ (3) (⃗
ρ(⃗ r − r⃗c (t)) + ∑ qjn δ (3) (⃗
r − r⃗n (t) − r⃗jn (t)) , (6.24a)
c nj
⟨ρ(⃗
r , t)⟩ = ∑ qc f (⃗
r − r⃗c (t)) + ∑ qjn f (⃗
r − r⃗n (t) − r⃗jn (t)) , (6.24b)
c nj
q1n = − Z e , r⃗1n ≈ ⃗
0, (6.24c)
jmax (n) jmax (n)
∑ qjn = Z ′ e , ∑ qjn = ∑ qjn = (Z ′ − Z) e = Qn . (6.24d)
j=2 j j=1
Here, r1n is the position of the nucleus of the nth neutral atom or nth ion, Z ′ e is
the charge of the electrons in the nth neutral atom or nth ion, (Z ′ − Z) e is the total
charge of the nth neutral atom or nth ion, and ⟨r2 ⟩n is the charge radius of the
nth neutral atom or nth ion. For a neutral atom, we have Qn = 0. Unless indicated
otherwise, for the remainder of the derivation of the macroscopic Maxwell equations
(see Sec. 6.2), the sum over j will be interpreted as
jmax (n)
∑≡ ∑ , (6.25)
j j=1
where jmax (n) is the number of constituent charges in the nth atom or ion. The
(averaged) microscopic Maxwell equations are as follows,
1
⃗ ⋅ e⃗(⃗
⟨∇ r , t)⟩ = (3)
∑ qc ⟨δ (⃗ r − r⃗c )⟩
0 c
1 (3)
+ ∑ qjn ⟨δ (⃗ r − r⃗n − r⃗jn )⟩ , (6.26a)
0 jn
∂⃗
⃗ ⋅ ⃗b(⃗
⟨∇ r , t)⟩ = 0 , ⃗ × e⃗(⃗
⟨∇ r , t)⟩ + ⟨ r, t)⟩ = 0 ,
b(⃗ (6.26b)
∂t
⃗ × ⃗b(⃗
⟨∇ r, t)⟩ ∂ d⃗
rc (3)
− 0 ⟨ e⃗(⃗
r , t)⟩ = ∑ qc ⟨δ (⃗ r − r⃗c )⟩
μ0 ∂t c dt
rn d⃗
d⃗ rjn
+ ∑ qjn ( + ) ⟨δ (3) (⃗
r −⃗
rn −⃗
rjn )⟩ . (6.26c)
jn dt dt
Here, we have suppressed the explicit time dependence of the positions r⃗c (t) and
r⃗jn (t). Written in terms of the averaged fields, we can alternatively express the
inhomogeneous Maxwell equations as
∇ ⃗ r , t) = 1 ∑ qc f (⃗
⃗ ⋅ E(⃗ r − r⃗c ) +
1
∑ qjn f (⃗ r − r⃗n − r⃗jn ) ,
0 c 0 jn
(6.27a)
1 ⃗ ∂ ⃗ d⃗
rc
⃗
∇ × B(⃗
r , t) − 0 E(⃗ r , t) = ∑ qc f (⃗
r − r⃗c )
μ0 ∂t c dt
d⃗rn d⃗ rjn
+ ∑ qjn ( + r − r⃗n − r⃗jn ) . (6.27b)
) f (⃗
jn dt dt
Let us verify that the charge and current densities fulfill the continuity equation,
and start with ρa ,
∂ ∂
⟨ρa (⃗
r , t)⟩ = ∑ qjn f (⃗
r − r⃗n (t) − r⃗jn (t))
∂t jn ∂t
rn d⃗
d⃗ rjn
⃗ ⋅ ∑ qn (
= −∇ + ) f (⃗ ⃗ ⋅ ⟨J⃗a (⃗
r − r⃗n (t) − r⃗jn (t)) = −∇ r , t)⟩ .
n dt dt
(6.28)
Here, the three-dimensional chain rule has been used, for the function f (⃗ r − r⃗n (t) −
r⃗jn (t)), where f = f (⃗ x) is a function of a three-dimensional argument, while in turn
x⃗ = r⃗ − r⃗n (t) − r⃗jn (t) depends on the time t.
In order to understand the three-dimensional chain rule, let us first recall the
basic formulas for a Taylor expansion to arbitrary order in one dimension,
∂ 1 ∂2
g(x + η) = g(x) + η g(x) + η 2 2 g(x) + . . .
∂x 2! ∂x
∞
ηn ∂ n ∂
= ∑ [ n g(x)] = exp (η ) g(x) . (6.29)
n=0 n! ∂x ∂x
The generalization to three dimensions is as follows, and we give it in various dif-
ferent formulations,
1 ∂2
x + η⃗) = g(⃗
g(⃗ ⃗ x) +
x) + η⃗ ⋅ ∇g(⃗ ηα ηβ x)
g(⃗
2! ∂xα ∂xβ
1 ∂3
+ ηα ηβ ηγ x) + . . .
g(⃗
3! ∂xα ∂xβ ∂xγ
1
= g(⃗ ⃗ x) +
x) + η⃗ ⋅ ∇g(⃗ (⃗ ⃗ (⃗
η ⋅ ∇) ⃗ g(⃗
η ⋅ ∇) x)
2!
1 ⃗ (⃗ ⃗ (⃗ ⃗ g(⃗
+ (⃗η ⋅ ∇) η ⋅ ∇) η ⋅ ∇) x) + . . .
3!
∞
(⃗ ⃗ n
η ⋅ ∇)
= ∑ x) = exp (⃗
g(⃗ ⃗ g(⃗
η ⋅ ∇) x) . (6.30)
n=0 n!
Here, it is necessary to remember that the differential operators act only on the
argument vector x ⃗, not on the spatial displacement η⃗.
6.2.3 Macroscopic Averaging and Charge Density

In order to derive the phenomenological Maxwell equations, and relate them to the
microscopic quantities, one expands the charge and current densities in powers of
the displacements r⃗jn and its time derivative r⃗˙jn , assuming that these quantities
are smaller than the position vector r⃗n that describe the centers of the individual
atoms. The expansion of the charge density in powers of r⃗jn leads to
r , t)⟩ = ∑ qc f (⃗
⟨ρ(⃗ r − r⃗c ) + ∑ qjn f (⃗
r − r⃗n − r⃗jn )
c nj
= ∑ qc f (⃗
r − r⃗c ) + ∑ qjn f (⃗
r − r⃗n ) + ∑ qjn (−⃗ ⃗ f (⃗
rjn ⋅ ∇) r − r⃗n )
c nj nj
,-- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - . - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - /
=ρ0 (⃗
r,t)
+∑ 1
q (−⃗
rjn ⃗ (−⃗
⋅ ∇) ⃗ f (⃗
rjn ⋅ ∇) r − r⃗n ) . (6.31)
2 jn
nj
Here, we identify the free charge density ρ0 as the sum of the free charges in the
conduction band qc and the total ionic charges Qn = ∑j qjn . Then,
⎛ ⎞
⟨ρ(⃗
r, t)⟩ = ρ0 (⃗ ⃗ (⃗
r , t) − ∑ ∑ qjn r⃗jn ⋅ ∇f r − r⃗n ) + ∑ 12 qjn (⃗ ⃗ rjn ⋅ ∇)
rjn ⋅ ∇)(⃗ ⃗ f (⃗
r − r⃗n )
n ⎝ j ⎠ nj
⎛ ⎞
= ρ0 (⃗ ⃗ ⋅ ∑ p⃗n f (⃗
r , t) − ∇ r − r⃗n ) − ∑ 12 qjn r⃗jn (⃗ ⃗ f (⃗
rjn ⋅ ∇) r − r⃗n )
⎝n nj ⎠
= ρ0 (⃗ ⃗ ⋅ ∑ P⃗n f (⃗
r , t) − ∇ r − r⃗n ) . (6.32)
n
Here, the polarization of the nth atom, including the quadrupole term, is given as
P⃗n = p⃗n − ∑ 21 qjn r⃗jn (⃗ ⃗ ,
rjn ⋅ ∇) (6.33)
j
where the pure dipole contribution is p⃗n = ∑n qjn r⃗jn , and the second term includes
the quadrupole correction. The averaged form (6.26a) of Gauss’s law thus reads as
follows,
⃗ ⋅ e⃗(⃗
0 ⟨∇ r , t)⟩ = ρ0 (⃗ ⃗ ⋅ ∑ P⃗n f (⃗
r , t) − ∇ r − r⃗n ) , (6.34)
n
and it can be rewritten as
∇ ⃗ r, t) = ρ0 (⃗
⃗ ⋅ D(⃗ r , t) , (6.35)
provided we identify the dielectric displacement D(⃗ ⃗ r , t) as
D(⃗ ⃗ r , t) + P⃗ (⃗
⃗ r , t) = 0 E(⃗ r, t) = 0 ⟨⃗
e(⃗r , t)⟩ + ∑ P⃗n f (⃗
r − r⃗n ) . (6.36)
n
This equation identifies the dielectric displacement as being due to the induced
dipole moments of the constituent molecules, plus a correction proportional to the
projection of the quadrupole moment onto the gradient of the sampling function.
The projection of the quadrupole term carries a certain dependence on the shape
of the sampling function f . The pure dipole term p⃗n is averaged over the sampling
function f (⃗r − r⃗n ) and this average is not very dependent on the precise shape of
the sampling function, but the integral of the gradient may exhibit a significant
dependence. We can resolve this apparent contradiction in the following way: (i) In
a quantum mechanical treatment, the dependence on the shape of the sampling
function would be absorbed in a formulation of the scattering process under study.
One has to analyze the multipole components of the incoming and outgoing waves
and then write the multipole components of the atomic polarizabilities, in order
to describe the multipole components of the scattering process. The functional
form of the sampling function describes the structure of the atoms relevant for
the scattering process under study. In other words, the appropriate form of the
sampling function is determined by quantum mechanical considerations which are
beyond our classical model. (ii) The shape-independent pure dipole term, as we
shall see later, provides by far the dominant effect if the system is interrogated in
the infrared. At optical wavelengths, the pure dipole term only receives corrections
of relative order α, where α is the fine-structure constant. In most applications
described in the literature, one may thus restrict the discussion to the pure dipole
term.
6.2.4 Macroscopic Averaging and Current Density
The total current density is
⃗ r , t)⟩ = ∑ qc d⃗
rc rn d⃗
d⃗ rjn
⟨J(⃗ f (⃗
r − r⃗c ) + ∑ qjn ( + ) f (⃗
r − r⃗n − r⃗jn ) . (6.37)
c dt nj dt dt
An expansion to second order in the r⃗jn leads to
⃗ r , t)⟩ = ∑ qc d⃗
rc d⃗
rn
⟨J(⃗ f (⃗
r − r⃗c ) + ∑ qjn f (⃗
r − r⃗n )
c dt nj dt
d⃗
rjn rn d⃗
d⃗ rjn
+ ∑ qjn f (⃗
r − r⃗n ) + ∑ qjn ( + ) (−⃗ ⃗ (⃗
rjn ⋅ ∇)f r − r⃗n )
nj dt nj dt dt
d⃗
rn
+ ∑ 12 qjn (−⃗ ⃗
rjn ⋅ ∇)(−⃗ ⃗ (⃗
rjn ⋅ ∇)f r − r⃗n ) . (6.38)
nj dt
The first two terms on the right-hand side can be identified in terms of the free
current density J⃗0 ,
d⃗
rc d⃗
rn
J⃗0 (⃗
r , t) = ∑ qc f (⃗
r − r⃗c ) + ∑ Qn f (⃗
r − r⃗n ) , (6.39)
c dt n dt
where Qn = ∑j qjn . Furthermore, we identify p⃗n = ∑j qjn r⃗n and define
d⃗
rn d⃗
rjn
v⃗n = , v⃗jn = . (6.40)
dt dt
We can thus write the current density as

⃗ r , t)⟩ = J⃗0 (⃗ d⃗
pn
⟨J(⃗ r , t) + ∑ f (⃗ ⃗ (⃗
r − r⃗n ) − ∑ v⃗n ∑ qjn r⃗jn ⋅ ∇f r − r⃗n )
n dt n j
⃗ (⃗
− ∑ v⃗jn qjn r⃗jn ⋅ ∇f r − r⃗n ) + ∑ 12 qjn v⃗n (⃗ ⃗ rjn ⋅ ∇)f
rjn ⋅ ∇)(⃗ ⃗ (⃗r − r⃗n ) .
nj nj
(6.41)
We shall now transform the third term on the right-hand side,
T⃗ = − ∑ v⃗n ∑ qjn r⃗jn ⋅ ∇f
⃗ (⃗ ⃗ (⃗
r − r⃗n ) = − ∑ v⃗n p⃗n ⋅ ∇f r − r⃗n ) . (6.42)
n j n
⃗ and ⃗b which are independent of r⃗, we have ∇

For any a a × ⃗b) = a
⃗ × (⃗ ⃗ (⃗b ⋅ ∇)
⃗ − ⃗b (⃗ ⃗
a ⋅ ∇).
Thus,
⃗ × ∑(⃗
∇ r − r⃗n ) = ∑ [⃗
pn × v⃗n ) f (⃗ pn (⃗ ⃗ − v⃗n (⃗
vn ⋅ ∇) ⃗ f (⃗
pn ⋅ ∇)] r − r⃗n ) , (6.43)
n n
and so
T⃗ = − ∑ p⃗n (⃗ ⃗ (⃗
vn ⋅ ∇)f ⃗ × ∑(⃗
r − r⃗n ) + ∇ pn × v⃗n ) f (⃗
r − r⃗n ) . (6.44)
n n
The total current can thus be expressed as
⃗ r, t)⟩ = J⃗0 (⃗ d⃗
pn d⃗
rn
⟨J(⃗ r , t) + ∑ f (⃗
r − r⃗n ) + ∑ p⃗n (− ⃗ f (⃗
⋅ ∇) r − r⃗n )
n dt n dt
+∇ ⃗ × ∑(⃗ pn × v⃗n ) f (⃗ ⃗ (⃗
r − r⃗n ) − ∑ v⃗jn qjn r⃗jn ⋅ ∇f r − r⃗n )
n nj
+∑ 1
q v⃗ (⃗
rjn ⃗ rjn ⋅ ∇)f
⋅ ∇)(⃗ ⃗ (⃗r − r⃗n ) . (6.45)
2 jn n
nj
With the help of the chain rule, the second and third terms on the right-hand side
of Eq. (6.45) can conveniently be expressed in terms of the time derivative of the
dipole contribution to the polarization density, resulting in the expression
⃗ r , t)⟩ = J⃗0 (⃗ d
⟨J(⃗ r , t) + (∑ p⃗n f (⃗ ⃗ × ∑(⃗
r − r⃗n )) + ∇ pn × v⃗n ) f (⃗
r − r⃗n )
dt n n
⃗ (⃗
− ∑ v⃗jn qjn r⃗jn ⋅ ∇f r − r⃗n ) + ∑ 12 qjn v⃗n (⃗ ⃗ rjn ⋅ ∇)f
rjn ⋅ ∇)(⃗ ⃗ (⃗r − r⃗n ) .
nj nj
(6.46)
The third term on the right-hand side reminds us of the curl of an induced density
of magnetic moments, as will be further explored below. However, the fourth and
fifth terms on the right-hand side of Eq. (6.46),
U⃗ = − ∑ v⃗jn qjn r⃗jn ⋅ ∇f
⃗ (⃗
r − r⃗n ) + ∑ 1 qjn v⃗n (⃗ ⃗ rjn ⋅ ∇)f
rjn ⋅ ∇)(⃗ ⃗ (⃗r − r⃗n ) , (6.47)
2
nj nj
remain to be treated. Key to the further analysis is the following identity, which
⃗ appropriately,
allows us to rewrite U
⎛ ⎞ d
⃗ × ∑ 1 qjn (⃗
U =∇ rjn × v⃗jn ) f (⃗
r − r⃗n ) − ∑ 1 qjn r⃗jn (⃗ ⃗ (⃗
rjn ⋅ ∇)f r − r⃗n )
⎝ nj 2 ⎠ dt nj 2
⃗ × ∑ 1 qjn (⃗
−∇ ⃗ (⃗
rjn ⋅ ∇) rjn × v⃗n ) f (⃗
r − r⃗n ) . (6.48)
2
nj
The three terms on the right-hand side of Eq. (6.48) will eventually be identi-
fied as the curl of the dipole magnetization, the time derivative of the quadrupole
contribution to the polarization, and the curl of the quadrupole correction to the
magnetization, respectively. We start with the right-hand side of Eq. (6.48) and
rewrite the three terms with the help of the following identities. The first term is
transformed according to
⎛ ⎞
⃗ × ∑ 1 qjn (⃗
∇ rjn × v⃗jn ) = ∑ 12 qjn r⃗jn (⃗ vjn ⋅ ∇) ⃗ f (⃗ r − r⃗n )
⎝ nj 2 ⎠ nj
,-- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - .- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - /
=T1
− ∑ 12 qjn v⃗jn (⃗ ⃗ f (⃗
rjn ⋅ ∇) r − r⃗n ) , (6.49)
nj
while the second term is transformed as follows,

d
− ∑ 1 qjn r⃗jn (⃗ ⃗ (⃗
rjn ⋅ ∇)f r − r⃗n ) = − ∑ 12 qjn v⃗jn (⃗ ⃗ (⃗
rjn ⋅ ∇)f r − r⃗n )
dt nj 2 nj
− ∑ 12 qjn r⃗jn (⃗ ⃗ (⃗
vjn ⋅ ∇)f r − r⃗n ) − ∑ 12 qjn r⃗jn (⃗ ⃗
rjn ⋅ ∇)(−⃗ ⃗ (⃗
vn ⋅ ∇)f r − r⃗n ) . (6.50)
nj nj
,-- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - . - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - / ,-- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - .- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - /
=−T1 =T2
In the last term on the right-hand side of Eq. (6.50), we have two minus signs, one
overall minus sign out front and a second minus sign in the term (−⃗ ⃗ The
vn ⋅ ∇).
third term on the right-hand side of Eq. (6.48) is transformed according to
⃗ × ∑ 1 qjn (⃗
−∇ ⃗ (⃗
rjn ⋅ ∇) rjn × v⃗n ) f (⃗
r − r⃗n ) = − ∑ 12 qjn r⃗jn (⃗ ⃗ (⃗
rjn ⋅ ∇) ⃗ f (⃗
vn ⋅ ∇) r − r⃗n )
2
nj nj
,-- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -. - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -/
=−T2
+ ∑ 12 qjn v⃗n (⃗ ⃗ (⃗
rjn ⋅ ∇) ⃗ f (⃗
rjn ⋅ ∇) r − r⃗n ) ,
nj
(6.51)
which proves Eq. (6.48). Adding Eqs. (6.49)–(6.51), one obtains Eq. (6.48).
Using the identity Eq. (6.48) in Eq. (6.46), one obtains the following decompo-
sition for the current density,
d ⎛ ⎛ ⎞ ⎞
⟨J⃗(⃗
r, t)⟩ = J⃗0 (⃗
r , t) + ∑ p⃗n − ∑ 21 qjn r⃗jn (⃗ ⃗ f (⃗
rjn ⋅ ∇) r − r⃗n )
dt ⎝ n ⎝ j ⎠ ⎠
⎛ ⎞
⃗ × ∑ ∑ 1 qjn (⃗
+∇ rjn × v⃗jn ) f (⃗
r − r⃗n )
n ⎝ j ⎠
2
⎛ ⎛ ⎞ ⎞
⃗ × ∑ p⃗n − ∑ 1 qjn r⃗jn (⃗
+∇ ⃗ f (⃗
rjn ⋅ ∇) r − r⃗n ) × v⃗n , (6.52)
⎝n ⎝ j
2
⎠ ⎠
which can be rewritten as

⃗ r , t)⟩ = J⃗0 (⃗ d
⟨J(⃗ r , t) + (∑ P⃗n f (⃗ ⃗ × ∑m
r − r⃗n )) + ∇ ⃗ n f (⃗
r − r⃗n )
dt n n
⃗ × (∑ P⃗n f (⃗
+∇ r − r⃗n ) × v⃗n ) , (6.53)
n
where we have encountered the polarization density operator P⃗n in Eq. (6.33) above,
P⃗n = p⃗n − ∑ 21 qjn r⃗jn (⃗ ⃗ .
rjn ⋅ ∇) (6.54)
j
The partial time derivative acting on a field means that one leaves the observation
point r⃗ unchanged. We thus have the relation
∂ ⃗ ∂ d
P (⃗
r , t) = ∑ P⃗n f (⃗
r − r⃗n ) = ∑ P⃗n f (⃗
r − r⃗n (t)) . (6.55)
∂t ∂t n dt n
The magnetic moment of the nth atom is identified as follows,
1
⃗ n = ∑ qjn (⃗
m rjn × v⃗jn ) . (6.56)
2 j
This definition can be motivated in the following way. We first observe that the
magnetic moment of a circular ring current I of time period T , covering a circle of
radius R, is
M⃗ = I A êz = q πR2 êz . (6.57)
T
The surface normal to the area covered by the area A is assumed to be directed
along the positive z axis. The velocity of the atom is 2πR/T , and so the magnetic
moment can be expressed as
M⃗ = q πR2 êz = 1 qR 2πR êz = 1 qRv êz = 1 q R

⃗ × v⃗ . (6.58)
T 2 T 2 2
This justifies Eq. (6.56). We can finally identify the magnetization as
⃗ (⃗
M r, t) = ∑ m r − r⃗n ) + ∑ P⃗n f (⃗
⃗ n f (⃗ r − r⃗n ) × v⃗n , (6.59)
n n
where the second term is a “global” term due to “rotating” dipole moments.
6.2.5 Phenomenological Maxwell Equations

In view of Eqs. (6.52), (6.55), and (6.59), we have finally found a microscropic model
for the separation
⃗ r , t)⟩ = J⃗0 (⃗ ∂
⟨J(⃗ r , t) + P⃗ (⃗
r , t) + ∇ ⃗ (⃗
⃗ ×M r , t) . (6.60)
∂t
The averaged Ampere–Maxwell law (6.26c) thus reads as follows,
1 1 ∂
⃗ × ⃗b(⃗
⟨∇ r , t)⟩ − ⟨⃗
e(⃗ ⃗ r , t)⟩
r , t)⟩ = ⟨J(⃗
μ0 μ0 c2 ∂t
∂
= J⃗0 (⃗
r , t) + P⃗ (⃗ ⃗ (⃗
⃗ ×M
r , t) + ∇ r , t) . (6.61)
∂t
We may now define the dielectric displacement and the magnetic field,
⃗ r , t) = 1 ⟨⃗b(⃗
H(⃗ ⃗ (⃗
r, t)⟩ − M r , t) , ⃗ r , t) = 0 ⟨⃗
D(⃗ r , t)⟩ + P⃗ (⃗
e(⃗ r , t) , (6.62)
μ0
in which case the Ampere–Maxwell law reads as
∇ ⃗ r, t) − ∂ D(⃗
⃗ × H(⃗ ⃗ r , t) = J⃗0 (⃗
r , t) . (6.63)
∂t
Together with Eq. (6.35), this completes the derivation of the phenomenological
Maxwell equations.
6.2.6 Parameters of the Multipole Expansion

A final word should be supplemented, regarding the multipole expansion. Let us
recall the condition (6.8) under which our treatment is valid,
λ ≫ R ≫ a0 , ⟨d⟩, rρ , (6.64)
where a0 is the Bohr radius, ⟨d⟩ is a typical atomic separation distance, and rρ is a
length scale of density fluctuations. For our applications, we have
rjn ∣ ∼ a0 ,
∣⃗ (6.65)
because the Bohr radius gives the dimension of the atom, and
1
⃗ ∼
∇f f, (6.66)
R
in view of the fact that the gradient operator is distributed over a volume of di-
mension R. We recall that the sampling function f [defined in Eq. (6.14)] has a
physical dimension of inverse volume. Indeed, the extent to which the quadrupole
corrections and higher-order terms contribute to the averaged Maxwell equations
depends on the extent of the sampling volume. Our expansion is formulated in
terms of the expansion parameter, or merely expansion operator, r⃗jn ⋅ ∇. ⃗ One can
assign a parametric estimate to this operator by letting it act on a sampling func-
tion f ,
⃗ ∼ a0 f ≫ a0 f ,
r⃗jn ⋅ ∇f (6.67)
R λ
where we take into account the fact that R ≪ λ. The parametric estimate thus is
⃗ ∼ a0 ≫ a0 .
r⃗jn ⋅ ∇ (6.68)
R λ
This condition has to be taken with a grain of salt. For interactions (“interrogations
of the material”) with very long wavelengths, λ ≫ R, the condition is certainly
fulfilled. However, for interrogations at optical wavelengths, we can actually choose
the averaging dimension R of the sampling volume to be larger, but we cannot
choose it to be very much larger, than the (optical) wavelength. If we interrogate
the sample with an optical wavelength, then, for typical wavelengths,
a0 a0
∼ ∼ α. (6.69)
λ a0 /α
The condition
a0
λ∼ ≫ R ≫ a0 , 137 a0 ≫ R ≫ a0 , (6.70)
α
severely restricts the range of permissible values of R. This shows, in particular,
that the formulation in terms of the phenomenological Maxwell equations is justified
if one interrogates condensed matter (where ⟨d⟩ is of the order of a0 ) by optical
radiation. Furthermore, in this regime, the corrections to the dipole approximation
are of order α. The dipole approximation works even better at frequencies below
the optical transitions, i.e., in the infrared. In the optical regime, the expansion in
powers of the “expansion operator”
⃗ ∼ α,
r⃗jn ⋅ ∇ (6.71)
is an expansion in powers of the fine-structure constant α for interrogations with
optical wavelengths. Other examples for these relevant parametric estimates have
recently been discussed in a related context, in Ref. [27].
6.3 Fourier Decomposition and Maxwell Equations in a Medium
We have just encountered a classical model for the description of the dielectric
displacement D(⃗⃗ r, t), and the magnetic field H(⃗
⃗ r , t), in a medium. In reality, even
if the classical approach could be extended into the atomic (quantum) domain, it
would be impossible to follow all the trajectories r⃗n (t) and r⃗jn (t) of the constituent
particles in the sample.
One thus has to find a different way to describe the dielectric displacement
in terms of a phenomenologically inspired model. It is customary to assume a
functional relationship of the form
D(⃗ ⃗ r , t) + P⃗ (⃗
⃗ r , t) = 0 E(⃗ r, t) = ∫ dt′ (⃗ ⃗ r , t′ ) .
r − r⃗′ , t − t′ ) E(⃗ (6.72)
(All integrals extend of the maximum integration domain, namely, , if the range
of integration is not explicitly specified.) Here, (⃗ r − r⃗′ , t−t′ ) is the dielectric permit-
tivity of the material, which can be expressed in terms of the relative permittivity
r − r⃗′ , t − t′ ),
r (⃗
(⃗r − r⃗′ , t − t′ ) = 0 r (⃗
r − r⃗′ , t − t′ ) . (6.73)
In many cases, one may assume that the polarization is generated locally, i.e., that
r − r⃗′ , t − t′ ) = δ (3) (⃗
r (⃗ r − r⃗′ ) r (t − t′ ) , (6.74)
in which case the dielectric displacement can be given as a convolution integral,
⃗ r , t) = 0 ∫ dt′ r (t − t′ ) E(⃗
D(⃗ ⃗ r , t′ ) . (6.75)
The same is true for the permeability

r − r⃗′ , t − t′ ) = μ0 μr (⃗
μ(⃗ r − r⃗′ , t − t′ ) , (6.76)
which also can be approximated by a local function in many cases,

r − r⃗′ , t − t′ ) = δ (3) (⃗
μr (⃗ r − r⃗′ ) μr (t − t′ ) , (6.77)
so that
⃗ r, t) = μ0 ∫ dt′ μr (t − t′ ) H(⃗
B(⃗ ⃗ r , t′ ) . (6.78)
Furthermore, it is quite common for typical materials that

μr (t − t′ ) ≈ 1 , (6.79)
while the relative permittivity r (t−t′ ) appreciably differs from unity on time scales
t−t′ ∼ 1/ν, where ν is a transition frequency of the constituent atoms in the medium.
Formulas (6.72)–(6.79) are valid for isotropic materials, in which case the relative
permittivity and relative permeability are just scalar functions. For non-isotropic
materials, these quantities acquire a tensor structure, r → r,ij and μr → μr,ij ,
where i, j = 1, 2, 3. Off-diagonal elements in r,ij with i ≠ j describe the possibility
of a “sheared” induced polarization density, where the atomic dipoles align along
directions which deviate from the direction of the applied external electric field. We
shall contend ourselves, here, with the isotropic case and rather focus on the analytic
properties of the relative permittivity r in both the time and frequency domains.
Indeed, it will turn out to be very convenient to transform the entire formalism into
Fourier space. To this end, we shall briefly recall a few basic facts about Fourier
transformations, extending the previous discussions in Sec. 2.2 and 4.1.1.
The Fourier transform of the electric field into the mixed position-frequency
representation is
E(⃗ ⃗ (⃗
⃗ r, ω) ≡ ∫ dt eiω t E r , t) , (6.80)
with the Fourier backtransformation

⃗ r , t) = ∫ dω e−iω t E(⃗
E(⃗ ⃗ r , ω) . (6.81)
2π
In physics, as compared to some engineering disciplines, one usually distributes
the factor 2π in the Fourier integrals asymmetrically, with the integral over the
angular frequency having the factor 1/(2π). For a wave with oscillation period T ,
the frequency is just the reciprocal 1/T , and the angular frequency is ω = 2π/T .
The integration measure dω/(2π) therefore is nothing but dν, where ν = 1/T is the
frequency. The phase conventions in Eqs. (6.80) and (6.81) are inspired by the fact
that the application of the time derivative or “Hamilton” operator i∂t to the function
e−iω t yields just ω, which, multiplied by the unit of action h̵ (the reduced Planck
constant), yields the energy. As already emphasized in the text following Eq. (2.10),
⃗ r, t) and E(⃗
we recall that E(⃗ ⃗ r, ω) are two different functions, distinguished by the
physical dimension of the second argument.
Similarly, the space-time Fourier transform of the electric field is
⃗ r −ωt)
E( ⃗ ω) ≡ ∫ d3 r ∫ dt e−i(k⋅⃗
⃗ k, ⃗ (⃗
E r , t) , (6.82)
with the Fourier backtransformation

⃗ (⃗ d3 k dω i(k⋅⃗
⃗
⃗ ω) .
⃗ k,
E r , t) = ∫ ∫ e r−ω t) E( (6.83)
(2π) 3 2π
Here, too, the factor 2π are unevenly distributed. For a wave with wavelength
λ, the wave number is just the reciprocal 1/λ, and the angular wave number is
k = 2π/λ. The integration measure dk/(2π) is equal to d(1/λ), which is the number
of full spatial oscillations of the wave packet per unit of length, measured in inverse
meters in SI mksA units (a measure which also carries the name kayser). Formulas
analogous to Eqs. (6.80)–(6.83) hold for the Fourier transforms of the magnetic
induction field B, ⃗ as well as the dielectric displacement field D ⃗ and the magnetic
field H.⃗
Let us now reformulate Eq. (6.72) with the help of (6.73), under the assump-
tion (6.74),
∞
⃗ r ) = 0 ∫
D(⃗ ⃗ r , t − t′ ) .
dt′ r (t − t′ ) E(⃗ (6.84)
−∞
Then,
∞ ∞
⃗ r , ω) = 0 ∫
D(⃗ dt ei ω t ∫ ⃗ r , t′ )
dt′ r (t − t′ ) E(⃗
−∞ −∞
∞ ∞ ∞ dω ′ ′ ′
⃗ r , t′ )
= 0 ∫ dt ei ω t ∫ dt′ ∫ r (ω ′ ) e−i ω (t−t ) E(⃗
−∞ −∞ −∞ 2π
∞ ∞ ∞ dω ′ i (ω−ω′ ) t ′ ′
⃗ r , t′ )
= 0 ∫ dt ∫ dt′ ∫ e r (ω ′ ) ei ω t E(⃗
−∞ −∞ −∞ 2π
dω ′
∞
⃗ r , ω ′ ) = r (ω) E(⃗
⃗ r, ω) .
= 0 ∫ [2π δ(ω − ω ′ )] r (ω ′ ) E(⃗ (6.85)
−∞ 2π
This is a manifestation of the general wisdom that convolution in time simply

means multiplication in frequency (Fourier) space. The relative permittivity r (ω)
is dimensionless, while r (t − t′ ) has a dimension of inverse time.
From the representation (6.72), it is clear that the relative permittivity r takes
the role of a Green function, relating the electric field E⃗ to the induced polarization
density P⃗ which is part of the dielectric displacement D. ⃗ Causality dictates that
we should interpret r (t − t′ ) as a retarded rather than advanced Green function.
There is a great deal to learn from the analytic properties of r (ω), as we shall see
in the following.
It is very useful to summarize the phenomenological Maxwell equations in both
Fourier as well as coordinate space,
∇ ⃗ r, t) = ρ0 (⃗
⃗ ⋅ D(⃗ r , t) , (6.86a)
∇ ⃗ r, t) = 0 ,
⃗ ⋅ B(⃗ (6.86b)
⃗ r, t) = − ∂ B(⃗
⃗ × E(⃗
∇ ⃗ r , t) , (6.86c)
∂t
⃗ r, t) = J⃗0 (⃗ ∂ ⃗
⃗ × H(⃗
∇ r , t) + D(⃗ r , t) . (6.86d)
∂t
The second and third Maxwell equations have not changed. These are the homoge-
neous equations. The first and the fourth equation have been derived in Eqs. (6.35)
and (6.63). It is interesting to note that these equations, in SI mksA units, carry
no explicit prefactors like c, 0 and μ0 ; the prefactors are all unity. In mixed
coordinate-frequency space, one has
∇ ⃗ r , ω) = ρ0 (⃗
⃗ ⋅ D(⃗ r, ω) , ∇ ⃗ r, ω) = 0 ,
⃗ ⋅ B(⃗ (6.87a)
∇ ⃗ r , ω) = iω B(⃗
⃗ × E(⃗ ⃗ r, ω) , ⃗ r , ω) = J⃗0 (⃗
⃗ × H(⃗
∇ ⃗ r, ω) ,
r, ω) − iω D(⃗ (6.87b)
while in wave-number-frequency space, after a Fourier transformation with respect
to both space and time, one has
ik⃗ ⋅ D(
⃗ k, ⃗ ω) ,
⃗ ω) = ρ0 (k, k⃗ ⋅ B( ⃗ ω) = 0 ,
⃗ k, (6.88a)
k⃗ × E( ⃗ ω) = ω B(
⃗ k, ⃗ ω) ,
⃗ k, ik⃗ × H( ⃗ ω) = J⃗0 (k,
⃗ k, ⃗ ω) − iω D( ⃗ ω) .
⃗ k, (6.88b)
Central to the following analysis is the derivation of the wave equation in a
dielectric medium, in mixed coordinate-frequency space. We start from Eq. (6.87)
with J⃗0 = ⃗
0,
⃗ r, ω)] = 1 ⃗ r , ω)]
⃗ × [∇
∇ ⃗ × H(⃗ ⃗ × [∇
∇ ⃗ × B(⃗
μ0 μr (ω)
1 ⃗ r , ω)] = −iω ∇ ⃗ r, ω) .
= [−∇⃗ 2 B(⃗ ⃗ × D(⃗ (6.89)
μ0 μr (ω)
Furthermore,
⃗ r, ω) = 0 r (ω) ∇
⃗ × D(⃗
∇ ⃗ r , ω) = 0 r (ω) iω B(⃗
⃗ × E(⃗ ⃗ r , ω) , (6.90)
The two above equations could be summarized as
1 ⃗ r , ω) = iω ∇ ⃗ r, ω) ,
⃗ 2 B(⃗
∇ ⃗ × D(⃗ (6.91a)
μ0 μr (ω)
∇ ⃗ r , ω) = 0 r (ω) iω B(⃗
⃗ × D(⃗ ⃗ r , ω) . (6.91b)
One finally obtains the result
ω2 ⃗
⃗ 2 + r (ω) μr (ω)
(∇ r , ω) = ⃗
) B(⃗ 0. (6.92)
c2
Similarly, for the electric field, one has
ω2 ⃗
⃗ 2 + r (ω) μr (ω)
(∇ r , ω) = ⃗
) E(⃗ 0. (6.93)
c2
Equations (6.92) and (6.93) describe wave propagation in a medium. For μr (ω) = 1,
one may set
√
r (ω) = n(ω) + i κ(ω) , (6.94)
where the refractive index n(ω) describes diffraction, and κ(ω) is the absorption
coefficient.
6.4 Dielectric Permittivity: Various Examples
6.4.1 Sellmeier Equation

The propagation of waves in a medium depends on the relative magnetic perme-
ability and relative electric permittivity functions μr (ω) and r (ω) for the medium.
We will generally deal with systems in which μr (ω) can be approximated as having
the value unity. A general form often used to approximate the electric permittivity
is the Sellmeier equation [28], or, a generalization thereof,
Am
(ω) = r (ω) 0 , r (ω) = 1 + ∑ 2 − ω 2 − iγ ω
. (6.95)
m ωm m
Here, Am = Ω2m has the dimension of the square of a frequency. In its original form,
the Sellmeier equation describes the dielectric response of a transparent medium
such as a glass and contains two or three terms of the form given in Eq. (6.95),
in the sum over m. Furthermore, the original Sellmeier equation is formulated
without the damping terms proportional to γm in Eq. (6.95) and thus is capable
of describing damped oscillations. Hence, we refer to Eq. (6.95) as a generalized
Sellmeier equation, which is based on the assumption that the system behaves as a
set of resonances (cf. Sec. 2.2).
The real and imaginary parts of the dielectric permittivity given in Eq. (6.95)
are as follows,
(ωm
2
− ω 2 ) Am Am
Re r (ω) = 1 + ∑ ∼1−∑ , ω → ∞, (6.96)
m (ω 2 2 )2
− ωm + ω2 2
γm m ω2
and
ω γm Am γm Am
Im r (ω) = ∑ ∼∑ , ω → ∞. (6.97)
m (ω 2 2 )2
− ωm + ω2 2
γm m ω3
The real part is an even function of ω, the imaginary part is an odd function of ω.
6.4.2 Drude Model

A specialization of the Sellmeier equation to a single term is capable of describing
electrical conductors, either metals or plasmas. One strives to model the dielectric
response of a plasma, where electrons move about freely. One considers only one
resonance, with m = 0 and ω0 = 0, but nonvanishing width γ0 ≠ 0. This theory is
relevant to the oscillations of an electron gas in a near-perfect conductor, where the
“resonant excitation frequency” of the electrons is zero. The relaxation time τ0 is
defined as
1
τ0 = . (6.98)
γ0
We thus have
σ0 1
r (ω) = 1 − . (6.99)
0 ω (ω τ0 + i)
Here, σ0 is a characteristic conductivity of the plasma, and τ0 is the mean time

between electron collisions in the plasma. The Drude model gives us an excellent
opportunity to discuss the Green function G which determines the polarization
density. One defines
G(ω) = r (ω) − 1 , r (t − t′ ) = δ(t − t′ ) + G(t − t′ ) , (6.100a)
dω −iω(t−t′ )
G(t − t′ ) = ∫ e G(ω) . (6.100b)
2π
For a causal behavior, G(t − t′ ) vanishes for t − t′ < 0. The dielectric displacement
is obtained from the electric field as
∞ ∞
⃗ r , t) = 0 ∫
D(⃗ ⃗ r , t′ ) + 0 ∫
δ(t − t′ ) E(⃗ ⃗ r , t′ )
dt′ G(t − t′ ) E(⃗
−∞ −∞
∞
⃗ r , t) + 0 ∫
= 0 E(⃗ ⃗ r, t − τ ) .
dτ G(τ ) E(⃗ (6.101)
0
In frequency space, we have a simple multiplication,
⃗ r , ω) = 0 [1 + G(⃗
D(⃗ ⃗ r , ω) .
r , ω)] E(⃗ (6.102)
For small driving frequency, the polarization density P⃗ is obtained as
1 ⃗ r , ω) ≈ −σ0 1 E(⃗
P⃗ (⃗
r, ω) = 0 G(⃗ ⃗ r, ω) = −σ0
r , ω) E(⃗ E(⃗ ⃗ r , ω) . (6.103)
ω (ω τ0 + i) iω
We now refer to Eq. (6.60) and identify the polarization current density J⃗p as the
time derivative of the polarization density,
∂
r , t) = P⃗ (⃗
J⃗p (⃗ r , t) , J⃗p (⃗
r, ω) = −i ω P⃗ (⃗
r, ω) . (6.104)
∂t
The conductivity σ0 relates the induced (polarization) current flowing in the mate-
rial to the electric field,
J⃗p (⃗ r , ω) E⃗ (⃗
r , ω) = σ (⃗ r, ω) , (6.105)
which allows us to identify
σ (⃗
r , ω) = −iω 0 G(⃗
r , ω) . (6.106)
Here, σ0 has units of (C/(m s)) m/V = (C/(m s)) (C/J) = C /(Jms). By contrast,
2 2
0 has units of A s/(V m) = C2 /(J m). So, σ0 /0 has the physical dimension of a
frequency.
The Drude ansatz (6.99) for the relative permittivity has the right physical
dimension (unity). Furthermore, the low-freqency limit of the conductivity is found
as
1
lim σ (⃗r, ω) = lim (−0 iω) (− ) = σ0 , (6.107)
ω→0 ω→0 iω
which is fully consistent with the usual definition of the current density and con-
ductivity σ0 provided the current flowing in the material is identified with the
polarization current J⃗p (⃗
r, ω).
The Drude model also allows us to study the so-called plasma frequency of the
electron gas. The plasma frequency ωp is obtained from the relative permittivity,
ωp2 ≡ lim ω 2 [1 − r (ω)] . (6.108)
ω→∞
For the plasma model (6.99), one has
σ0
ωp2 = . (6.109)
0 τ0
We shall now try to use a microscopic theory of the plasma, and to find an alternative
representation, as follows,
NV e2 NV e2
ωp2 = , σ0 = τ0 . (6.110)
0 me me
Here, NV is the volume density of free electrons and me is the electron mass. A
bulk material of electrons is assumed to have a volume density NV , moving “freely”
relative to a jellium of ionized cores, by a collective distance xc . Let A be the (large)
cross-sectional area of the box, and V = A xc its volume. The induced charges of the
electrons, “sticking out” at either end of the box, are easily calculated as follows,
Q1 = NV e A xc , Q2 = −Q1 . (6.111)
We now apply Gauss’s theorem in integrated form, to the layer in between the
charged structures. The induced electric field obeys the equation ∮ E ⃗ind ⋅ dA⃗ =
−1
(0 ) ∫ d r ρ(⃗
3
r ), with
1 1
Eind A = − NV e A xc , Eind = − NV e xc . (6.112)
0 0
The sign follows from a geometric consideration, which is left as an exercise to the
reader. We now express the dielectric constant as follows,
Eext (ω) − Eind (ω)
r (ω) ≈ , (6.113)
Eext (ω)
where Eext (ω) is the external (driving) field. The sign in this equation can be justi-
fied by observing that the induced field is always directed against the driving field,
and thus opposite to the induced polarization density. The polarization density, by
contrast, has to be added to the expression 0 Eext in order to obtain the dielectric
displacement. Since the electrons in the bulk medium are free, we can approximate
d2 xc me
me ≈ e Eext (t) , −ω 2 xc (ω) ≈ Eext (ω) , (6.114)
dt2 e
where in the latter step we go into Fourier space. Finally, one has
me 1
−ω 2 xc (ω) + NV e xc (ω) ωp2
e 0 NV e2 1
r (ω) ≈ me = 1 − = 1 − . (6.115)
−ω 2 xc (ω) 0 me ω 2 ω2
e
This “plasma” model
ωp2
r (ω) = 1 − (6.116)
ω2
(a) (b)
Fig. 6.1 Plot of Re(r (ω)) and Im(r (ω)) for ωp = 1, ω0 = 2, and γ = 0.3, according to Eq. (6.117).
The functional shape of Re(r (ω)) is consistent with in-phase driving below resonance, with a
phase jump by π as one crosses the resonance. The deviation of Re(r (ω)) from unity at high ω
is proportional to 1/ω 2 .
only has a real part and thus ignores the possibility of damping. As we later see, it
does not fulfill the Kramers–Kronig relations, in view of the fact that r (ω) cannot
be real everywhere along the real ω-axis. Our model (6.99) thus is physically more
consistent than the “plasma” or “bulk” formula (6.116).
Infinitesimal damping can be studied by considering the limit τ0 → ∞ in
Eq. (6.99), while keeping the plasma frequency (6.109) constant. This leads to
a well-defined result for any generalized model with a finite resonance frequency
ω0 . We start from the Sellmeier equation (6.95), keep only the term with m = 0,
while refraining from making the additional assumption that ω0 = 0, and write (see
Fig. 6.1)
ωp2
r (ω) = 1 + , γ0 → 0 . (6.117)
ω02 − ω 2 − iωγ0
Using the prescription
1 1
= (P.V.) + iπ δ(ω0 − ω) , γ0 → 0 , (6.118)
ω0 − ω − iγ0 ω0 − ω
where (P.V.) denotes the principal value, one obtains
ωp2 1 iπ
= (P.V.) + [δ(ω0 − ω) + δ(ω0 + ω)] . (6.119)
ω02 − ω 2 − iωγ0 ω02 − ω 2 2ω0
For any finite ω0 , a model consistent with the Kramers–Kronig relations can thus
be written down as
1 iπ
r (ω) = 1 + (P.V.) + [δ(ω − ω0 ) + δ(ω + ω0 )] . (6.120)
ω02 −ω 2 2ω0
The limit ω0 → 0 cannot be taken, because it would lead to a divergence in the
second term. A nonvanishing damping term is essential for the internal consistency
of the model (6.99).
Finally, it is instructive to calculate the Fourier backtransform of Eq. (6.99),

σ0 1 σ0 dω −i ω (t−t′ ) 1
G(ω) = − , G(t − t′ ) = −
∫ e ,
0 ω (ω τ0 + i) 0 2π (ω + iη) (ω τ0 + i)
(6.121)
where we introduce an infinitesimal imaginary part iη in order to ensure causality.
The two residues are
′ 1 1
Res e−i ω (t−t ) = , (6.122a)
ω=−iη (ω + iη) (ω τ0 + i) i
′ 1 ′ 1
Res e−i ω (t−t ) = e−(t−t )/τ0 (− ) . (6.122b)
ω=−i/τ0 (ω + iη) (ω τ0 + i) i
The prefactor multiplying the residues is (−2πi), because the poles are encircled in
the mathematically negative (clockwise) direction, for t − t′ > 0. The end result is
σ0 t − t′
G(t − t′ ) = Θ(t − t′ ) [1 − exp (− )] . (6.123)
0 τ0
This quantity has dimension of frequency, as it should, because it is obtained as
the Fourier backtransform of a dimensionless quantity r (ω). For perfect electrical
conductors, τ0 → ∞, and
σ0
G (t − t′ ) → , τ0 → ∞ . (6.124)
0
For a realistic conductor and finite τ0 , we have G(t − t′ = 0) = 0, and the inte-
gral (6.101) converges.
6.4.3 Dielectric Permittivity and Atomic Polarizability

We have already seen that a very successful way of describing the relative per-
mittivity of a typical material consists in the analogy with a collection of har-
monic oscillators. One compares the functional form of the Green function of the
harmonic oscillator (2.11) with the Sellmeier equation (6.95). For dilute gases,
it is additionally possible to relate the given “collection of harmonic oscillators”
with the atomic properties of the gas atoms themselves, which are summarized in
the so-called dynamic atomic polarizability. This analogy will be explored in the
following.
By definition, the Fourier component p⃗n (ω) of a dipole moment of the nth atom
⃗ r , ω) at the
is related to its polarizability α(ω) and to the applied electric field E(⃗
coordinate of the atom, as follows (we assume spatially uniform fields),
⃗ r, ω) .
p⃗n (ω) = α(ω) E(⃗ (6.125)
We may thus relate the polarization density P⃗ (⃗
r , ω) to the volume density of atoms,
⃗ r, ω), as follows,
N /V , and the driving electric field E(⃗
N N
P⃗ (⃗
r , ω) = p⃗n = ⃗ r, ω) .
α(ω) E(⃗ (6.126)
V V
This implies that

NV N
r (ω) − 1 = α(ω) , NV = . (6.127)
0 V
In atomic physics, one writes the corresponding expression for α(ω) is
fn0
α(ω) = ∑ ̵ 2,
̵ ω − (hω) (6.128)
2
n En0 − i Γn h
where En0 are the atomic transition energies from the quantum mechanical ground
state (denoted by the subscript zero) to the nth excited state, and the Γn are the
energy widths of the transitions. The angular transition frequency ωn0 and the
damping factor γn of the transition are given as
En0 Γn
ωn0 = ̵ , γn0 = ̵ . (6.129)
h h
The factors fn0 in Eq. (6.128) are the so-called oscillator strengths of the transitions.
An alternative representation of the atomic polarizability thus is
fn0 1
α(ω) = ∑ ̵ 2 2 . (6.130)
n h ωn0 − ω − i γn ω
2
Written in this form, the analogy to the Green function (2.11) of the damped
harmonic oscillator is obvious,
1
g(ω) = 2 . (6.131)
ω0 − i γ ω − ω 2
Equation (6.130) represents the atomic polarizability as a sum over damped har-
monic oscillators. Restoring prefactors, Eq. (6.127) can be written as
NV fn0 1
r (ω) = 1 + α(ω) = 1 + ∑ ̵ 2 2 . (6.132)
0 n h ω n0 − ω 2 − iγ ω
n
A remark is in order. By convention, α(ω) = α=1 (ω) given in Eq. (6.128) is the
dipole (2=1 -pole) dynamic polarizability of the ground state ∣φ0 ⟩ of an atom. It is
the dominant term in the long-wavelength limit, and it describes the response of
the atom to incident dipole radiation. Incident quadrupole radiation ( = 2) will
dynamically (ω ≠ 0) induce a quadrupole moment in the atoms, while an incident
octupole radiation ( = 3) will do the same for an atomic octupole moment. Quan-
tum mechanically, it is useful to note at this stage that the 2 -multipole oscillator
strength is
2
() 4πe2
fn0 =2 En0 ∑ ∑ ∣⟨φn ∣∑ (ri ) Ym (θi , ϕi )∣ φ0 ⟩∣ , (6.133)
(2 + 1)2 m mn i
where we sum over the magnetic projections m of the spherical harmonic and over
the magnetic projections mn of the excited state ∣φn ⟩. The ground state is denoted
as ∣φ0 ⟩. Furthermore, ri is the radial coordinate of the ith electron, and Ym (θi , ϕi ) is
the spherical harmonic with the arguments being equal to the polar and azimuthal
angles of the position of the ith electron. The sum is over all electrons within
the atomic system; these are enumerated in the sum over i. The dipole oscillator
strength is recovered as
(=1)
fn0 = fn0 . (6.134)
There is a sum rule,
̵
h
∑ fn0 = Z e a0 Eh , a0 = Eh = α2 me c2 .
2 2
, (6.135)
n α me c
Here, a0 is the Bohr radius, Eh is the Hartree energy, α is the fine-structure constant,
Z is the number of electrons in the atom, and e is the elementary charge. Let us
check the consistency of physical units. The oscillator strength fn0 is measured in
(Cm)2 J, while 0 carries a unit of VAms or VCm . Hence,
NV NV fn0 1 m V (Cm)2 JC
α(ω) ∼ 2
∼ J= = 1, (6.136)
0 0 En0 m3 C J2 V
as it should be.
6.4.4 Dielectric Permittivity for Dense Materials

Up to this point, we have ignored the backreaction of the induced electric field,
inside the sample, onto the orientation of the dipole moments. This approximation
is justified if the relative permittivity of the sample does not significantly differ
from unity, or in other words, if the sample is dilute. Indeed, for a dilute gas, the
dielectric constant may be expressed as [Eq. (6.132)],
NV
r (ω) = 1 + α(ω) , (6.137)
0
because the backreaction of the induced polarization field onto the atoms in the
sample can be ignored. From Eqs. (6.33) and (6.36), we recall that, within the
dipole approximation, the polarization is given as
P⃗ (⃗
r , t) = ∑ p⃗n f (⃗
r − r⃗n (t)) , (6.138)
n
where f constitutes the test function for the averaging volume V over which the
macroscopic Maxwell equations are determined. The distribution function f is nor-
malized according to Eq. (6.14), namely, ∫V d3 r f (⃗
r) = 1. Thus, f has dimension of
inverse volume. It is advantageous to choose the reference volume V , as a rectangu-
lar volume of dimensions Δx, Δy, and Δz. We choose Δz to be the distance of the
two charges that represent the dipole (see Fig. 6.2). If the dipoles are distributed
uniformly in the test volume, then
N
P⃗ = ⟨⃗
p⟩ = NV ⟨⃗
p⟩ , (6.139)
V
where ⟨⃗p⟩ is the average dipole moment. Let us suppose that the dipoles are oriented
in the z direction. Then, a cut through the sample, in the xy plane, with N atoms
in the volume, will result in a surface charge density (see also Fig. 6.2),
q N
σ=N = (qΔz) = NV ∣⟨⃗p⟩∣ = ∣P⃗ ∣ . (6.140)
Δx Δy Δx Δy Δz
Δz
σpol (θ) = −σ cos(θ)
a

E
θ
1 σpol (θ) cos(θ)
dE = − dS
P 4π0 a2
Fig. 6.2 Illustration of Eq. (6.143) and (6.144).
If the surface is tilted, then the appropriate modification is
σ = ∣P⃗ ∣ cos θ . (6.141)
The sign of σ has to be determined based on additional considerations.

We recall that for a uniform medium, the polarization
P⃗ = 0 (r − 1) E
⃗ (6.142)
is aligned and parallel to the applied electric field. Let us draw a sphere of radius
a about the point r⃗ in the dense sample. Let us assume, furthermore, that the
polarization points into the +z direction, and the applied electric field also points
along +z. We cut through the dipole layers in the middle of the dipoles, separating
the dipoles, and ignore the outer portion of the sphere as well as the rest of the bulk
medium in the following. Then, the surface charge density at an angle θ with respect
to the symmetry axis of the polarization (the +z axis is parallel, not anti-parallel,
to P⃗ ) is
σpol (θ) = −∣P⃗ ∣ cos θ , (6.143)
where θ is the polar angle with respect to the +z axis (see also Fig. 6.2). Let us
assume that the dipoles are oriented so that the positive charges are to the top and
the negative ones to the bottom. The polarization vector points upward. We use
the center of the sphere as an anchor point. This center is below the upper surface
of the sphere. Cutting through the sample, positive charge is left on the upper side
of the sphere, i.e., on the outer rim of the cut-out sphere. At the same time, θ = 0.
What counts for us is the negative charge on the lower side of the cut. This justifies
the negative sign in Eq. (6.143).
Now, imagine that we cut out a sphere of radius a about our reference point
r⃗. The z component of the field due to the infinitesimal surface charge element of
radius a, generated at the reference point r⃗, amounts to
1 σpol (θ) cos θ 1 +∣P⃗ ∣ cos2 θ
dEz = − 2
dS = dS , (6.144)
4π0 a 4π0 a2
because the positive charges at the top of the sphere lead to an electric field that
points downward. The field generated by the polarized constituents at the point r⃗
is
1 ∣P⃗ ∣ cos2 θ ∣P⃗ ∣ 1 4πa2 ∣P⃗ ∣
Ez = ∫ dS = = , (6.145a)
4π 0 a2 4π a2 3 0 3 0
⃗
⃗pol = Ez êz = P .
E (6.145b)
30
The total field of the remaining dipoles within the sphere of radius R is
⃗i = 1 pi ⋅ r⃗i ) r⃗i p⃗i

3(⃗
∑E ∑( − 3) (6.146)
i 4π0 i ri5 ri
where i enumerates the molecules, and r⃗i is the direction vector from the location
of the ith molecule to r⃗. Provided the entire sample is z polarized, we may average
out the field according to the prescription
1
3(⃗pi ⋅ r⃗i ) r⃗i → 3êz pi zi2 → × 3êz pi ri2 , (6.147)
3
where in the last step, we have performed the angular average zi2 → ri2 /3. In
summary,
⃗i = ⃗
∑E 0. (6.148)
i
The polarization charge is exclusively given by the integral (6.145). Letting the
radius of the sphere become large, we see that the result (6.145) holds universally.
The local field at the point r⃗ is thus found as the sum of the external field E ⃗
and the field E⃗pol due to the polarized constituents,
⃗
⃗loc = E
E ⃗+E ⃗+ P .
⃗pol = E (6.149)
3 0
Using Eq. (6.142), we obtain
E ⃗ = r (ω) + 2 E
⃗loc = E⃗ + 1 0 (r − 1) E ⃗. (6.150)
3 0 3
Hence, we have two relations,
P⃗ = NV α(ω) E ⃗loc = NV α(ω) r (ω) + 2 E

⃗, (6.151)
3
P⃗ = 0 (r (ω) − 1) E⃗, (6.152)
the second of which is just the definition (6.142). Eliminating P⃗ , then E,

⃗ one
obtains
r (ω) − 1 NV α(ω)
= , (6.153)
r (ω) + 2 3 0
which is the Clausius-Mosotti equation. It is easy to show that
dr (r (ω) − 1) (r (ω) + 2)
= , (6.154)
dNV 3 NV
which is left as an exercise to the reader.
6.5 Propagation of Plane Waves in a Medium
6.5.1 Refractive Index and Group Velocity

The wave equation in a medium has been given in Eqs. (6.92) and (6.93). For a
⃗ the dispersion is
plane wave with angular frequency ω and wave vector k,
ω2
k 2 = r (ω) , (6.155)
c2
⃗ and in general, r (ω) is complex rather than real. In consequence, the
where k = ∣k∣
wave number k = 2π/λ can be complex. The relation between the wave number and
the angular frequency is given by k = Re k + i Im k, with
ω ω
Re k = n (ω) , Im k = κ (ω) . (6.156)
c c
From the dispersion relation (6.155), one gets
ω2 ω2
k 2 = r (ω)
= [Re r (ω) + i Im r (ω)]
c2 c2
2
2 ω ω2
= [n (ω) + i κ (ω)] = [n 2
(ω) − κ 2
(ω) + 2 i n (ω) κ (ω)] .
c2 c2
(6.157)
This can be summarized as follows,
Re r (ω) = n2 (ω) − κ2 (ω) , (6.158a)
Im r (ω) = 2 n (ω) κ (ω) . (6.158b)
Alternatively, one may write
√
r (ω) = n (ω) + i κ (ω) . (6.159)
We note that in all our models and√calculation, √ r (ω) has a positive imaginary and
a positive real part. We identity r (ω) = ∣r (ω) ∣ exp[ 2i arg(r (ω))] with the
“obvious” branch of the square root, consistent with the branch cut of the square
root being along the negative real axis.
Just as for the relative permittivity r (ω), the real part of the index of refraction
is an even function of ω and the imaginary part is an odd function of ω,
n(ω) = n(−ω) , κ(ω) = −κ(−ω) . (6.160)
Using these properties, a wave traveling in the +x direction can be described by the
expression
∞ dω ⃗
⃗ t) = ∫
E(x, E0 (ω) exp [i (Re k(ω) + i Im k(ω)) x − ωt]
−∞ 2π
∞ dω ⃗ ω ω
=∫ E0 (ω) exp [i (n(ω) x − c t)] exp (− κ(ω) x) . (6.161)
−∞ 2π c c
This wave packet is the Fourier backtransformation of a wave packet composed of
plane waves, each of which fulfills the wave equation
∂2 ω2 ⃗ ω ω
( 2
+ r (ω) 2
) E0 (ω) exp (i n(ω) x) exp (− κ(ω) x) = 0 , (6.162)
∂x c c c
which is a specialization of Eq. (6.93) to one spatial dimension, with μr (ω) = 1.
⃗0 (ω)∣ has a single, narrow peak of
We restrict our analysis to the case in which ∣E
width Δω at ω0 . Here, Δω is the characteristic width of the wave packet in angular
frequency space. The peak will be considered narrow if
Δω dn(ω0 )
≪ 1. (6.163)
n(ω0 ) dω0
⃗
For a wave composed of a narrow range of frequencies, the wave vector k(ω) can
be expanded about the central frequency, to obtain an approximate expression for
the propagation of the wave. In our example, the central frequency is ω0 , and
ω ω0 dn(ω0 ) ω − ω0
n(ω) = n(ω0 ) + [n(ω0 ) + ω0 ] + O(ω − ω0 )2 . (6.164)
c c dω0 c
While a similar expression holds for ω κ(ω)/c, we here set κ(ω) = κ(ω0 ), assuming
that the damping is not very strong, i.e., that κ(ω) ≪ n(ω). Working to first order
in ω − ω0 , we find
⃗ t) = exp [i ω0 (n(ω0 )x − c t)] exp (− ω0 κ(ω0 ) x)

E(x,
c c
∞
dω ⃗ ω − ω0 dn(ω0 )
×∫ E0 (ω) exp (i {[n(ω0 ) + ω0 ] x − c t}) . (6.165)
2π c dω0
−∞
The interpretation is as follows: The first factor identifies a term that oscillates at
frequency ω0 and propagates at the phase velocity c/n(ω0 ). The second term is an
integral (envelope function) that travels with the group velocity vg (ω0 ), where
ω dRe k 1 ⎛ dn(ω) ⎞ 1
Re k = n(ω) , ∣ = n(ω0 ) + ω0 ∣ = ,
c dω ω=ω0 c ⎝ dω ω=ω0 ⎠ vg (ω0 )
−1
dn(ω0 )
vg (ω0 ) = c [n(ω0 ) + ω0 ] . (6.166)
dω0
⃗ t) as
One can write E(x,
⃗ t) = exp [i ω0 (n(ω0 )x − c t)] exp [− ω0 κ(ω0 ) x] F⃗ (x − vg (ω0 ) t) ,

E(x,
c c
∞ dω ω − ω0
F⃗ (x − vg (ω0 ) t) = ∫ E⃗0 (ω) exp (i {x − vg (ω0 ) t}) , (6.167)
−∞ 2π vg (ω0 )
where the first factor describes a damped wave traveling at the phase velocity,
while F⃗ (x − vg (ω0 ) t) is an (undamped) wave traveling at the group velocity. The
“broad” wave packet has an envelope function F⃗ traveling with speed vg (ω0 ), while
the underlying wave has the phase velocity c/n(ω0 ).
6.5.2 Wave Propagation and Method of Steepest Descent

The evaluation of Fourier backtransform integrals of the type (6.161) can be rather
complicated and in many cases, not analytically feasible if the functional form of
n(ω) or κ(ω) is sufficiently complex. However, in some cases, it can be helpful
to resort to a method of integration otherwise known as the method of stationary
phase, or method of steepest descent, which can be applied with good effect to
integrals of the form (6.161).
We will ask the following question. Let u (x, t) be a wave propagating in a
medium with a permittivity r (ω) and let a signal u(x, t) be written as the Fourier
backtransform of a sum of waves, propagating in the +x and/or −x directions, as
follows,
∞
dω −iωt ω n (ω) x ω n (ω) x
u (x, t) = ∫ e [A (ω) exp (i ) + B (ω) exp (−i )] ,
−∞ 2π c c
(6.168)
where for reference we temporarily set
√
n (ω) ≡ r (ω) , (6.169)
unifying the refractive index n(ω) and the absorption coefficient κ(ω) into a single
symbol.
Because all traveling waves have to pass any point of the x axis at some time, we
can conjecture that knowledge of the signal u(x, t) at a given point, but for all time,
should enable us to calculate the signal for all (x, t). Our knowledge of the signal at
the given point, say, the origin, but for all time, enables us to calculate the Fourier
transform of the signal. Fourier backtransformation to space-time coordinates then
enables us to calculate u(x, t), as desired. Yet, in the Fourier backtransformation
integrals, one may ask if contributions from times t′ > t might influence the signal at
(x, t), arguing that Fourier integrals stretch from −∞ to +∞. This question and the
eventual solution, as well as the evaluation of the resulting integral by the method
of steepest descent, will be discussed in the following.
We consider a situation where u (0, t) and ∂u (x, t) /∂x∣x=0 are given for all t,
and attempt to calculate the signal for all (x, t). First, from Eq. (6.168), we find
that
dω ∞
u (0, t) = ∫ . e−iωt [A (ω) + B (ω)]
(6.170)
2π−∞
Fourier transformation of u (0, t) enables us to calculate the sum of the Fourier

amplitudes,
∞ ′ ∞ ∞ ′ dω
∫ ei ω t
u (0, t) dt = ∫ ∫ e−i(ω−ω )t [A (ω) + B (ω)] dt
−∞ −∞ −∞ 2π
∞
=∫ [A (ω) + B (ω)] δ(ω − ω ′ ) dω = A (ω ′ ) + B (ω ′ ) .
−∞
(6.171)
Similarly, differentiating Eq. (6.168) with respect to time, one finds that
∂ ∞ ω n (ω) ω n (ω) dω
u (x, t)∣ =∫ e−iωt [i A (ω) − i B (ω)] . (6.172)
∂x x=0 −∞ c c 2π
Fourier transformation of the spatial derivative of u (x, t) at the origin thus gives
us the difference of the Fourier amplitudes,
∞ ′ ∂ ∞ dω ω n (ω)
∫ dt eiω t u (x, t)∣ = ∫ [A (ω) − B (ω)] i δ(ω − ω ′ ) (6.173)
−∞ ∂x x=0 −∞ 2π c
and so
c ∞ ′ ∂
−i ∫ eiω t u (x, t)∣ dt = A (ω ′ ) − B (ω ′ ) . (6.174)
ω ′ n (ω ′ ) −∞ ∂x x=0
Solving Eqs. (6.171) and (6.174) for A (ω ′ ) and B (ω ′ ), and setting ∂x u(0, t) ≡
∂
u (x, t)∣ , one obtains
∂x x=0
1 ∞ ′ c
A (ω ′ ) = ∫ eiω t {u (0, t) − i ′ ∂x u (0, t)} dt , (6.175a)
2 −∞ ω n (ω ′ )
1 ∞ ′ c
B (ω ′ ) = ∫ eiω t {u (0, t) + i ′ ∂x u (0, t)} dt . (6.175b)
2 −∞ ω n (ω ′ )
As anticipated, the Fourier expansion coefficients A (ω ′ ) and B (ω ′ ) can be recon-
structed from integrals over the signal and its derivative, evaluated at x = 0 but
observed over all time. Under the interchange ω ↔ ω ′ and t ↔ t′ , these formulas
can be used directly in Eq. (6.168),
∞ ∞
1 dω ′ −iω (t−t′ ) ω n (ω) x ′ ic ∂x u (0, t′ )
u (x, t) = (i ) {u (0, ) − }
2 ∫ 2π ∫
dt e exp t
c ω n (ω)
−∞ −∞
∞ ∞
1 dω ′ −iω (t−t′ ) ω n (ω) x ic ∂x u (0, t′ )
+ ∫ ∫ dt e exp (−i ) {u (0, t′ ) + }.
2 2π c ω n (ω)
−∞ −∞
(6.176)
One might ask the following question: Under the integral signs on the right-hand
side, we integrate over the entire range of t′ , i.e., t′ ∈ (−∞, ∞). However, on the
left-hand side, we have u(x, t). The suspicion would be that the signal at t′ > t as
it enters the integrand on the right-hand side might influence the left-hand side in
a non-causal way. In order to ensure causality, we have to slightly deform the inte-
gration over ω into a contour C, which is conveniently chosen to lie infinitesimally
above the real axis, just like for the retarded Green function (see Sec. 4.1.3). We
thus write
∞
1 ′ dω −i ω (t−t′ ) ω n (ω) x ′ ic ∂x u (0, t′ )
u (x, t) = ∫ dt ∫ e exp (i ) {u (0, t ) − }
2 2π c ω n (ω)
−∞ C
∞
1 ′ dω −i ω (t−t′ ) ω n (ω) x ′ ic ∂x u (0, t′ )
+ (−i ) {u (0, ) + }.
2∫ ∫ 2π
dt e exp t
c ω n (ω)
−∞ C
(6.177)
We now take the observed signal at x = 0 to be a single Dirac-δ peak at t = 0,
∂u (x, t)
u (0, t) = u0 δ (t) , ∣ = 0. (6.178)
∂x x=0
These boundary conditions will provide identical waves traveling in the +x and
−x directions. Since the Fourier transform of a Dirac-δ function is a constant in
frequency space, this acts as a “white light” source. The generated signal is, by
virtue of Eq. (6.177),
∞
1 dω −iω(t−t′ ) ω n (ω) x ω n (ω) x
u (x, t) = ∫ dt′ ∫ e [exp (i ) + exp (−i )] u (0, t′ )
2 2π c c
−∞ C
dω −iωt ωn (ω) x
= u0 ∫ e cos ( ). (6.179)
2π c
C
One can split the signal into waves traveling in either the +x or −x direction,
u0 dω n (ω) x u0 dω n (ω) x
u (x, t) = ∫ exp [−iω (t − )] + ∫ exp [−iω (t + )]
2 2π c 2 2π c
C C
= u+ (x, t) + u− (x, t) . (6.180)
We investigate the wave traveling in the +x direction,

1 dω n (ω) x
u+ (x, t) = u0 ∫ exp [−iω (t − )] , (6.181a)
2 C 2π c
1 ∞ dω n (ω) x
u+ (x, t) = u0 ∫ exp [−iω (t − )]
2 0 2π c
1 0 dω n (ω) x
+ u0 ∫ exp [−iω (t − )]
2 −∞ 2π c
1 ∞ dω n (ω) x
= u0 ∫ exp [−iω (t − )] + c.c. . (6.181b)
2 0 2π c
We have used the fact that n (ω) = n∗ (−ω), and divided the integration along the
contour C into one which extends from −∞ to 0 and another one from 0 to +∞
(ignoring the infinitesimal displacement off the real axis). Thus,
∞ dω n (ω) x
u+ (x, t) = u0 Re ∫ exp [−iω (t − )] . (6.182)
0 2π c
The integrand contains an exponential with a complicated structure. Typically,
integrals of this type are very well suited for the application of the method of steepest
descent. One looks for the points in the complex ω plane where the “phase”, i.e.,
the argument of the exponential, is stationary. In our case, this would imply that
one defines
n (ω) x df (ω) dn(ω) x
f (ω) = ωt − ω , = t − (ω + n (ω)) . (6.183)
c ω dω c
This equation determines the saddle point in the integral over ω, via the condition
df (ω)
∣ = 0. (6.184)
dω ω=ωc
Referring to Eq. (6.166), we establish that the saddle point occurs at
x = vg (ωc ) t , (6.185)
which is the point where the wave signal is expected to travel, according to the
interpretation of the group velocity.
One therefore approximates
∞
dω
u+ (x, t) = u0 Re ∫ exp [−i f (ω)]
2π
0
∞
dω ⎛ d2 f (ω) ⎞
≈ u0 Re exp [−i f (ωc )] ∫ exp −i ∣ (ω − ωc )2 . (6.186)
2π ⎝ dω 2
ω=ωc ⎠
0
The spirit of the method of steepest descent dictates to choose in the complex
plane, a contour which passes through the saddle point, and leads to the most rapid
decrease in the magnitude of the integrand. One then extends this contour to ∞, in
both directions, to evaluate the Gaussian integral, under the assumption that the
integrand decreases sufficiently rapidly to make this extension possible. In other
words, one heads down to the valley of the integrand as soon as possible. This may
imply that one has to distort the path of integration in the complex plane, in order
to “hit” the saddle point at the “right” angle. For example, if
d2 f (ω)
∣ > 0, (6.187)
dω 2 ω=ωc
then one would set
π 1 i
ω − ωc = ω(w) = exp (−i ) w = ( √ − √ ) w . (6.188)
4 2 2
Extending the integration contour to the interval w ∈ (−∞, ∞), one obtains
π ∞ 1
J = exp (−i ) ∫ exp (− f ′′ (ωc ) w2 ) dw
4 0 2
√
π ∞ 1 ′′ 2π
−iπ/4
≈ exp (−i ) ∫ dw exp (− f (ωc ) w ) = e
2
′′
, (6.189)
4 −∞ 2 f (ωc )
and thus
√
−iπ/4 2π
u+ (x, t) ≈ u0 Re exp [−i f (ωc )] e , (6.190)
f ′′ (ω c)
where the primary phase factor is
x
exp [−i f (ωc )] = exp [−i ωc (t − )] . (6.191)
vg (ωc )
One of the simplest possible applications concerns the plasma model without
damping term,
⎧ √
⎪
⎪ ω − ωp , (ω > ωp )
1/2 2 2
ωp2 ⎪
n (ω) ≈ (1 − 2 ) , ω n (ω) = ⎨ √ . (6.192)
ω ⎪
⎪
⎪ i ω 2 − ω2 , (ω > ω)
⎩ p p
With these observations, the function can be written as

⎡ √ 2 ⎤
u0 ∞ ⎛ ⎢ ω − ωp2 x ⎥⎞
u+ (x, t) = Re ∫ dω exp −i ω ⎢ ⎢ t− ⎥
2π ωp ⎝ ⎢ ω2 c⎥ ⎥⎠
⎣ ⎦
⎡ √ ⎤
u0 ω p ⎛ ⎢ ω − ωp x ⎥⎞
2 2
+ Re ∫ ⎢
dω exp −ω ⎢it + ⎥ .
c⎥
(6.193)
2π 0 ⎝ ⎢ ω2 ⎥⎠
⎣ ⎦
We thus estimate
⎡ √ 2 ⎤
u0 ∞ ⎛ ⎢ ω − ωp2 x ⎥⎞ ∞
u+ (x, t) ≈ Re ∫ exp −i ω ⎢ − ⎥ dω = u0 Re ∫ e−i f (ω) dω
⎝ ⎢t ⎥
c ⎥⎠
2π ωp ⎢ ω 2 2π ωp
⎣ ⎦
(6.194)
by the method of steepest descent. We have defined the complex phase of the
integrand as
√
⎛ ω 2 − ωp2 x ⎞
f (ω) = ω t − . (6.195)
⎝ ω2 c⎠
The critical value ω = ωc is determined as follows,
∂f ωp c t √ ωp
∣ = 0, ω = ωc = √ . f (ωc ) = (c t)2 − x2 , (6.196)
∂ω ω=ωc (c t) − x
2 2 c
3/2
∂2f ((ct)2 − x2 ) 1 ′′
f ′′ (ωc ) = ∣ = , f (ω) ≈ f (ωc ) + f (ωc ) (ω − ωc )2 .
∂ω 2 ω=ωc c x2 ωp 2
(6.197)
We have assumed ct > x in writing these expressions, which implies, in particular,

that f ′′ (ωc ) > 0. Hence, we may use Eqs. (6.188) and (6.189) with the result
√
∞ 1 ′′ ∞ 1 ′′ 2π
dω e−i 2 f (ωc ) (ω−ωc ) ≈ e−iπ/4 ∫ dw e− 2 f (ωc ) w = e−iπ/4
2 2
J=∫ .
ωp −∞ f ′′ (ωc )
(6.198)
The final result for the signal is
√ √
u0 π 2π u0 2π π
u+ (x, t) ≈ Re {exp [−i (f (ωc ) + )] }= cos (f (ωc ) + )
2π 4 f ′′ (ωc ) 2π f ′′ (ωc ) 4
−1/2
u0 √ ⎛ ((ct) − x ) ⎞
2 2 3/2 √ ωp π
= 2π ⎜ ⎟ cos ( (c t)2 − x2 + )
2π ⎝ c x2 ωp ⎠ c 4
√ √ √
u0 c ωp x ⎛ (c t)2 − x2 π⎞
=√ cos ωp + . (6.199)
2π ((ct)2 − x2 )3/4 ⎝ c 4⎠
This result fulfills the condition that u(x = 0, t) = u0 δ(t) = 0 for t > 0, in that it
vanishes at x = 0. In the “acausal” region x > c t, the value of the critical phase
f (ωc ) becomes imaginary, and the contribution of the saddle point is exponentially
suppressed. Hence, the result given in Eq. (6.199) for the “causal” region c t > x is
the complete result obtained by the steepest descent method.
6.6 Kramers–Kronig Relationships
6.6.1 Analyticity and the Kramers–Kronig Relationships

We recall once more the basic relations relevant for the dielectric constant and the
electric field, summarized in Eqs. (6.100)–(6.102),
r (ω) = 1 + G(ω) , r (t − t′ ) = δ(t − t′ ) + G(t − t′ ) , (6.200a)

⃗ r , ω) = 0 [1 + G(ω)] E(⃗
D(⃗ ⃗ r , ω) , (6.200b)
∞
⃗ r , t) = 0 E(⃗
D(⃗ ⃗ r , t) + 0 ∫ ⃗ r, t − τ ) .
dτ G(τ ) E(⃗ (6.200c)
0
Here, G(t − t′ ) is a retarded Green function, which implies that it vanishes for
t − t′ < 0. Furthermore, it means that G(ω) cannot have poles in the upper half of
the complex plane. We can thus formulate the Fourier transform as
∞
r (ω) − 1 = ∫ dτ G (τ ) exp (iωτ ) . (6.201)
0
⃗ and E
Since D ⃗ are real, G (τ ) must be real, and we have
∗
r (ω) = r (−ω) . (6.202)
Because r (ω) − 1 is an analytic function in the upper half complex plane, we

have by Cauchy’s residue theorem (see Sec. 3.3.5),
′
1 ′ r (ω ) − 1
r (ω) = 1 + ∮ dω ′
, Im ω > 0 , (6.203)
2πi C ω −ω
with the imaginary part of all points on the curve C restricted to be greater than
or equal to zero. The path of integration is closed at infinite radius, in the upper
complex half plane, which ensures that the only pole of the integrand in Eq. (6.203),
inside the contour of integration, occurs at ω ′ = ω, with Im ω > 0. Because r (ω)
can be interpreted as a retarded Green function, it does not contribute any further
pole terms.
For initially real ω, we can enforce the position of the pole to be in the upper
half plane, by adding a small imaginary part to ω, replacing ω in Eq. (6.203) by
ω + iδ. Then,
1 ∞ r (ω ′ ) − 1
r (ω) = 1 + lim ∫ dω ′ ′ , Im ω = 0 , (6.204)
2πi δ→0 −∞ ω − (ω + iδ)
where we assume that both the real as well as the imaginary parts of r (ω) vanish
sufficiently rapidly for ∣ω∣ → ∞ so that the contribution from the closing half-circle
can be neglected. When drawing the integration contour, it is immediately obvious
that the integral taken below the pole equals the principal part of the integral plus
πi times the residue at ω ′ = ω. This is encompassed in the formula,
1 1
= (P.V.) ′ + iπ δ(ω ′ − ω) . (6.205)
ω′ − ω − iδ ω −ω
Therefore,
1 ∞ (ω ′ ) − 1
dω ′ + iπ (r (ω) − 1))
r
r (ω) = 1 + ((P.V.) ∫
2πi −∞ ω′ − ω
1 ∞ (ω ′ ) − 1 1
dω ′ + (r (ω) − 1)
r
=1+ (P.V.) ∫ ′
2πi −∞ ω −ω 2
1 1 1 ∞ (ω ′ ) − 1
r
= r (ω) + + (P.V.) ∫ . (6.206)
2 2 2πi −∞ ω′ − ω
Taking the real and imaginary parts of this equation, one obtains
1 ∞ Im (ω ′ )
dω ′ ,
r
Re r (ω) = 1 + (P.V.) ∫ (6.207a)
π −∞ ω′ − ω
1 ∞ Re (ω ′ ) − 1
dω ′ .
r
Im r (ω) = − (P.V.) ∫ (6.207b)
π −∞ ω′ − ω
Using Re r (ω) = Re r (−ω) and Im r (ω) = −Im r (−ω), this can be written as
1 0 Im (ω ′ ) ∞ Im (ω ′ )
′
dω ′ ]
r r
Re r (ω) = 1 + (P.V.) [∫ dω + ∫
π −∞ ω′ − ω 0 ω′ − ω
1 ∞ Im (−ω ′ ) ∞ Im (ω ′ )
dω ′ + ∫ dω ′ ]
r r
=1+ (P.V.) [∫ ′
π 0 −ω − ω 0 ω′ − ω
1 ∞ Im (ω ′ ) ∞ Im (ω ′ )
′
dω ′ ]
r r
=1+ (P.V.) [∫ dω + ∫
π 0 ω′ + ω 0 ω′ − ω
2 ∞ ω ′ Im (ω ′ )
dω ′ .
r
=1+ (P.V.) ∫ (6.208a)
π 0 ω ′2 − ω 2
For the imaginary part, we have
1 0 Re (ω ′ ) − 1 ∞ Re (ω ′ ) − 1
′
dω ′ ]
r r
Im r (ω) = − (P.V.) [∫ dω + ∫
π −∞ ω′ − ω 0 ω′ − ω
1 ∞ Re (−ω ′ ) − 1 ∞ Re (ω ′ ) − 1
′
dω ′ ]
r r
= − (P.V.) [∫ dω + ∫
π 0 −ω ′ − ω 0 ω′ − ω
1 ∞ Re (ω ′ ) − 1 ∞ Re (ω ′ ) − 1
dω ′ + ∫ dω ′ ]
r r
= − (P.V.) [− ∫ ′
π 0 ω +ω 0 ω′ − ω
2ω ∞ Re (ω ′ ) − 1
dω ′ .
r
= − (P.V.) ∫ (6.208b)
π 0 ω ′2 − ω 2
Equations (6.208a) and (6.208b) are the classic Kramers–Kronig relations between
the real and imaginary parts of the electric permittivity. Similar relations hold for
the frequency response function for any causal system. They are very useful in
quantum mechanics (particularly in scattering theory), circuit analysis in electrical
engineering, etc.
6.6.2 Applications of the Kramers–Kronig Relationships
It would be impossible to give a complete overview of the many applications of the

Kramers–Kronig relationships in physics. We need to restrict ourselves to a few
examples. For example, it is possible to use known values of r (ω), in support of
the calculation of unknown values. Suppose we have measured or calculated the
imaginary part of the index of refraction. This may happen by a measurement of
the absorption of the system at given angular frequency. Also, suppose that the
real part of r (ω) is known, but only for a single value ω = ω1 . We are seeking the
real part Re r (ω) for given ω. In view of Eq. (6.208), r (ω1 ) satisfies
2 ∞ ω ′ Im (ω ′ )
dω ′ .
r
Re r (ω1 ) = 1 + (P.V.) ∫ (6.209)
π 0 ω ′2 − ω12
It follows that
2 ∞ 1 1
Re r (ω) − Re r (ω1 ) = (P.V.) ∫ ω ′ Im r (ω ′ ) ( ′2 − ′2 ) dω ′
π 0 ω −ω 2 ω − ω12
2 ∞ ω ′ Im r (ω ′ )
(ω 2 − ω12 ) (P.V.) ∫
= dω ′ .
π 0 (ω ′2 − ω 2 ) (ω ′2 − ω12 )
(6.210)
This form for r (ω) reduces the integral’s dependence on the values of Im r (ω ′ ) at
large ω ′ , because of the ω ′−4 dependence for large ω ′ . This is known as a “subtracted
dispersion relation”. The process of subtraction can be repeated for each known
value of r (ω).
Other applications concern so-called sum rules. We noted that, at high fre-
quency, the expression r (ω) − 1 behaves as ω −2 + O (ω −3 ) [see Eq. (6.96)]. Physi-
cally, at high frequencies, the binding forces for the electrons become negligible and
the system responds as a plasma. With this interpretation, we define the plasma
frequency for the system as [see also Eq. (6.108)]
ωp2 ≡ lim [ω 2 (1 − r (ω))] . (6.211)
ω→∞
This definition of the plasma frequency includes the contributions of the responses
of all charged particles in the system. In the sense of Eq. (6.95), the sources of
the resonances include optically active phonons, plasmons, electronic excitations,
electronic ionizations, and others.
For a functional form of the kind introduced in Eq. (6.95),
Am
r (ω) = 1 + ∑ 2 − ω 2 − iγ ω
, (6.212)
m ωm m
one has
ωp2 = ∑ Am (6.213)
m
by direct inspection of the limit. Using the Kramers–Kronig relation, it is possible

to derive a more universal integral representation of ωp2 as defined in Eq. (6.211).
Namely, if Im r (ω) varies as ω −3 + O (ω −4 ) for high frequencies [see Eq. (6.97)],
then Eq. (6.208) gives
ωp2 = lim [ω 2 (1 − Re r (ω)) − ω 2 i Im r (ω)] = lim [ω 2 (1 − Re r (ω))]
ω→∞ ω→∞
′ ′
2 ∞ ω Im r (ω )
= lim [−ω 2 (P.V.) ∫ dω ′ ]
ω→∞ π 0 ω ′2 − ω 2
2 ∞ 1
= lim [− (P.V.) ∫ dω ′ ′2 2 ω ′ Im r (ω ′ )]
ω→∞ π 0 [ω /ω − 1]
,-- - - - - - - - - - - - - - -. - - - - - - - - - - - - - - - /
→−1
2 ∞
= ∫ dω ω Im r (ω) . (6.214)
π 0
In typical cases, the imaginary part itself has no singularities along the integration
contour, and we can dispose of the principal value prescription, as done in the last
step.
One can derive other sum rules. Let us assume, again, that the real part of
the dielectric function goes as Re r (ω) − 1 ∼ −ωp2 /ω 2 + O (ω −4 ) for large ω whereas
the imaginary part goes as Im r (ω) ∼ O(ω −3 ) for large ω [see Eq. (6.97)]. The
asymptotic relationship between the real and imaginary parts of r (ω) can be used
to take advantage of the asymptotic properties in the range of large ω, say, ω ≫
ωL > ωp . We write the Kramers–Kronig relation as
2 ωL Re [r (ω ′ ) − 1] ∞ ′
′ Re [r (ω ) − 1]
Im r (ω) = [(P.V.) ∫ dω ′ 2 + ∫ dω 2 ].
πω 0 1 − (ω ′ /ω) ωL 1 − (ω ′ /ω)
(6.215)
Again, we assume that ωL is sufficiently large that the asymptotic form for the
real part of r (ω), namely, Re r (ω) ≈ 1 − ωp2 /ω 2 , provides a valid approximation.
Hence,
2 ωL Re [ (ω ′ ) − 1]
dω ′
r
Im r (ω) ≈ (P.V.) ∫ 2
πω 0 1 − (ω ′ /ω)
2 ∞ 1 ωp2
+ (P.V.) ∫ [− + O (ω ′−4 )] dω ′
πω ωL 1 − (ω ′ /ω)
2
ω ′2
2 ωL ωp2
≈ [(P.V.) ∫ Re [r (ω ′ ) − 1] dω ′ − + O (ω −2 )] . (6.216)
πω 0 ωL
One may now reason as follows. On the one hand, we have derived an approximation
for the coefficient of order ω −1 of Im r (ω), for large ω. On the other hand, we have
by assumption Im r (ω) ∼ O (ω −3 ) for large ω. It follows that the term in square
brackets in the last line of Eq. (6.216) must vanish, asymptotically, as
ωL ωp2
(P.V.) ∫ Re [r (ω ′ ) − 1] dω ′ = + O (ω −2 ) . (6.217)
0 ωL
For large ωL , we must therefore have the property
1 ωL ωp2
(P.V.) ∫ Re r (ω ′ ) dω ′ = 1 + 2 + O(ωL
−3
). (6.218)
ωL 0 ωL
This is sometimes called a superconvergence relation.
6.7 Exercises
● Exercise 6.1: Show that the charge density (6.20) and the current density (6.21)
fulfill the continuity equation.
● Exercise 6.2: Generalize the derivation (6.28) to the conduction band electrons.
● Exercise 6.3: Consider a field configuration with a single Fourier mode,

⃗ r , t) = E(⃗
E(⃗ ⃗ r) exp(−iωt) , ⃗ r, ω ′ ) = E(⃗
E(⃗ ⃗ r ) 2π δ(ω ′ − ω) , (6.219a)
⃗ r , t) = B(⃗
B(⃗ ⃗ r) exp(−iωt) , ⃗ r , ω ′ ) = B(⃗
B(⃗ ⃗ r) 2π δ(ω ′ − ω) . (6.219b)
Show that the dielectric displacement, in direct space (coordinates and time) is
related to r (ω) as a proportionality factor in
⃗ r , t) = 0 r (ω) E(⃗
D(⃗ ⃗ r) exp(−iωt) , (6.220)
where ω is the carrier frequency. Show that for waves composed of a single Fourier
mode, the wave equations (7.5a) and (7.5b) are obtained.
● Exercise 6.4: Derive Eq. (6.93) from the phenomenological Maxwell equations
in coordinate-frequency space, given in Eq. (6.87).
● Exercise 6.5: Consider a plane wave traveling in free space with (ω) = 0 and
μ(ω) = μ0 ,
⃗ ⃗
⃗ r , t) = E
E(⃗ ⃗0 eik⋅r−iωt , ⃗ r , t) = B
B(⃗ ⃗0 eik⋅r−iωt , (6.221a)
⃗ ⃗
⃗ r) = E
E(⃗ ⃗0 e ik⋅r
, ⃗ r) = B
B(⃗ ⃗0 e ik⋅r
, r , t) = 0 ,
ρ(⃗ ⃗ r, t) = ⃗
J(⃗ 0. (6.221b)
Show that the Maxwell equations become
k⃗ ⋅ E⃗0 =0 , k⃗ ⋅ B
⃗0 = 0 , (6.222a)
i k⃗ × E ⃗0 = 0 ,
⃗0 − iω B ⃗0 + i ω E
i k⃗ × B ⃗0 = ⃗
0. (6.222b)
c2
Assume that the magnetic field amplitude is given as
⃗
⃗0 = k × E
B ⃗0 . (6.223)
ω
Show that the Ampere-Maxwell law becomes
⃗0 = − ω E⃗0 ,
k⃗ × B (6.224)
c2
where we have used the relation k⃗2 = ω 2 /c2 . With reference to the considerations
preceding Eq. (5.106), calculate the time-averaged Poynting vector as
1 ⃗ ⃗
S⃗ = ⃗ ∗ = k ∣E⃗0 ∣2 .
E0 × B (6.225)
0
2μ0 2 ω μ0
● Exercise 6.6: Derive (6.112) based on a geometric consideration.
● Exercise 6.7: Derive (6.116) from Eq. (6.114), observing that you can obtain
the polarization density as NV e xc (ω), and expressing the dielectric displacement
appropriately.
● Exercise 6.8: Derive (6.143) based on a geometric consideration. This task is
a little subtle, because one must carefully consider the limit Δz → 0, i.e., the limit
of small dipoles, which corresponds to the continuum limit of the dipole density.
First, carefully consider what happens when you displace the separating plane from
the mid-dipole position in Fig. 6.2.
● Exercise 6.9: Derive Eq. (6.154).

● Exercise 6.10: Use the steepest descent method for the Gamma function, to
obtain Stirling’s approximation to the factorial. Then, study the literature on the
utility of the steepest descent method for the evaluation of Bessel functions.
● Exercise 6.11: Consider a model with a simple resonance,
α α 2αω0
r (ω) = + ≈1+ 2 . (6.226)
ω0 − ω − i 2 γ ω0 + ω + i 2 γ
1 1 ω0 − ω 2 − i γ ω
Calculate the limit of the imaginary part as γ → 0, express it in terms of Dirac-δ
functions, and use the Kramers–Kronig relationships to reproduce the real part.
● Exercise 6.12: Practice the calculation of principal-value integrals. Remember
that a principal value integration can be done in many ways, in particular, using
the following alternative possibilities. (i) One does the integral analytically, then
uses the upper and lower boundaries of the integration interval, after “throwing
away” the imaginary part. (ii) If x is the integration variable and x = a is the
location of the singularity, then one integrates up to x = a − , and starts again from
x = a + , observing that the the divergent term in 1/ cancels. (iii) One uses a
complex contour above and immediately below the singularities on the real axis,
and takes the mean. The different sense of revolving around the singularity then
means that its pole contribution cancels.
Employing all three prescriptions, show that
1 3
(P.V.) ∫ = ln(2) ,
dx (6.227a)
x−1
0
31 1
(P.V.) ∫ dx 2 = ln(2) , (6.227b)
0 x −1 2
∞ 1
(P.V.) ∫ dx 2 = 0. (6.227c)
−∞ x −1
Using your knowledge, show that the Drude model (6.99) satisfies the Kramers–
Kronig relationships (6.208a) and (6.208b).
● Exercise 6.13: Use the result derived in Eq. (6.214),
2 ∞
ωp2 = ∫ dω ω Im r (ω) , (6.228)
π 0
and consider the imaginary part (6.97),
ω γm Am
Im r (ω) = ∑ 2 . (6.229)
m (ω 2 − ωm
2 ) + ω2 γ 2
m
By contour integration, show that

2 ∞ ω 2 γm Am
ωp2 = ∑∫ dω = ∑ Am . (6.230)
π m 0 2 − ω 2 )2 + (γ ω)2
(ωm m
m
Finally, simplify the result for a dilute gas of atoms, expressing the result with the
help of the sum rule (6.135) for the oscillator strengths.
● Exercise 6.14: Derive a relation similar to Eq. (6.210) for the imaginary part of
the relative permittivity.
● Exercise 6.15: Explain why is it a good idea to model the relative permittivity
of a solid (dense material) according to the ansatz [see Eq. (6.95)]
r (ω) − 1 Am
=∑ 2 . (6.231)
r (ω) + 2 m ωm − ω 2 − iγm ω
Study the application of this ansatz to α-quartz, according to Ref. [27], and try to
find other suitable approximations for different materials, according to data pub-
lished in Ref. [29].
Chapter 7
Waveguides and Cavities
7.1 Overview
When we are talking about a laser diode, we often say that it is “excited in the
TE1,0 mode”, which is the laser mode with the lowest possible frequency of the laser
cavity—hence we can be sure that no other laser modes are excited, and the laser
light is monochromatic. An illustration of the precise meaning of this statement
is the subject of the current chapter. We shall try to understand why the spec-
trum of electromagnetic waves in cavities and waveguides is discrete and, in typical
cases, classified according to transverse electric (TE) and transverse magnetic (TM)
modes. When we say “discrete” we mean that we can assign “reference numbers”
to the waves, counting the number of nodes of the electric or magnetic fields in
different directions of space.
Along the same lines, when one analyzes the spectrum of atoms, say, the hy-
drogen spectrum, one may classify the “matter waves” of the bound electron ac-
cording to the number of nodes in the (radial) wave functions [17, 18]. Quantum
mechanics entails the assumption that the trajectories of point particles are in fact
smeared out and constitute matter waves. The transition from classical to wave
mechanics is analogous to the transition from ray to wave optics (were it not for
the additional postulate regarding the collapse of the wave function upon measure-
ment, see Refs. [17, 18]). Yet, the analysis of electromagnetic waves in cavities and
waveguides teaches us that the spectrum of waves can be discrete, and is a good
preparatory exercise for quantum mechanics.
A few more explanatory words are in order; these may not be fully discernible for
the novice but are still included here for better reference. You may have heard that
the electromagnetic field itself needs to be quantized at some point, leading to the
theory of quantum electrodynamics [20]. This latter quantization works differently
as compared to the quantum matter waves; namely, one assumes that the energy
̵
stored in the wave-like excitations is discrete and comes in “packages” of size hω,
where h ̵ is the reduced quantum unit of action (Planck’s constant divided by 2π),
and ω is the angular frequency. This amounts to a quantization of the electric and
magnetic fields themselves; the fields “wiggle” at every given point in space-time.
263
The wave character of the electromagnetic excitations is taken for granted in this
context; in fact, if we could determine the value of the electric and magnetic fields
simultaneously with absolute precision at any given space-time point, then we would
not need to quantize the field. Even the classical electromagnetic theory allows for
the presence of waves. The quantization of the electromagnetic field therefore is
called “second quantization”; quantum electrodynamics is a “second-quantized”
theory.
From this point of view, we should add that, what we do in the current chapter
is assume that the fields in the cavities and waveguides are sufficiently strong so
that we do not need to quantize them; the “wiggling” is a small effect and negligibly
affects strong macroscopic fields. We thus treat the problems on the level of “first
quantization” which in the case of electrodynamics, is just the classical wave theory
of light.
Before we set forth in this endeavor, let us indicate a few more paradigmatic
applications of waveguides and cavities, to illustrate their practical importance. In-
deed, in the generation of radar waves, the so-called klystron has been instrumental.
So-called X band radar has a typical frequency of 7175 MHz, corresponding to a
wavelength of roughly 4.2 cm; it has been used in Doppler tracking of the Cassini
spacecraft in superior conjunction on its way to Saturn [30]. Cavities also find
applications in accelerators; the basic idea here is that the electric field in the fun-
damental mode in a cylindrical cavity has oscillations in the direction parallel to
the symmetry axis of the cavity; this direction in turn can be aligned with the beam
direction [31].
Last, but certainly not least, the study of the field modes inside a cavity is
a necessary preparatory exercise for a quantum electrodynamic (yes!) calculation
of the attraction between perfectly conducting plates (see Sec. 8.4). Two large
parallel plates a small distance apart form an approximate “cavity” whose modes
are distinct from what would be expected in free space. In the fully quantized
theory, one associates a so-called “zero-point” energy with every possible excitation
of the electromagnetic field. The difference of the so-called “zero-point” energy
of the modes inside the cavity, compared to the modes in free space, results in
a distance-dependent energy shift. This shift leads to a force between the plates
known as the Casimir effect, as discussed in more detail in Sec. 8.4.
7.2 Waveguides
7.2.1 General Formalism
Let a waveguide be oriented along the z axis, with the (hollow) structure being
cylindrical or rectangular. For a wave propagating along the positive z direction,
we shall assume that
⃗ r ) = E(x,
E(⃗ ⃗ y) eikz , ⃗ r) = B(x,
B(⃗ ⃗ y) eikz . (7.1)
Waveguides and Cavities 265
Let us see if we can express the Ex and Ey components as functions of Ez only (for
the electric field), and likewise for the magnetic field. There are no sources inside
the conductor, but we shall assume the material inside the waveguide is isotropic
with electric permittivity
(ω) = r (ω)0 ≡ r 0 , (7.2)
and magnetic permeability,
μ(ω) = μr (ω) μ0 ≡ μr μ0 . (7.3)
Without boundary conditions, the speed of the propagating wave in the medium
√ ⃗ = E(⃗
⃗ r, ω)
is c/ r μr . We work in mixed position-frequency space, with fields E
and B⃗ = B(⃗
⃗ r , ω). The Maxwell equations yield, with = r 0 and μ = μr μ0 , in a
source-free region,
⃗ r , t) = − ∂ B(⃗
⃗ × E(⃗
∇ ⃗ r , t) ⇒ ∇ ⃗ r , ω) = iω B(⃗
⃗ × E(⃗ ⃗ r , ω) , (7.4a)
∂t
⃗ r , t) = ∂ D(⃗
⃗ × H(⃗
∇ ⃗ r, t) ⇒ ∇ ⃗ r, ω) = −i ω r μr E(⃗
⃗ × B(⃗ ⃗ r, ω) . (7.4b)
∂t c2
Inserting the curl of the first equation into the second, or vice versa, one obtains
the following wave equations,
ω2 ⃗
⃗ 2 + r μr
(∇ r , ω) = ⃗
) B(⃗ 0, (7.5a)
c2
ω2 ⃗
⃗ 2 + r μr
(∇ r , ω) = ⃗
) E(⃗ 0. (7.5b)
c2
From now on, we restrict the discussion to a single Fourier mode,
⃗ r , t) = E(⃗
E(⃗ ⃗ r ) exp(−iωt) , ⃗ r , ω ′ ) = E(⃗
E(⃗ ⃗ r ) 2π δ(ω ′ − ω) , (7.6a)
⃗ r , t) = B(⃗
B(⃗ ⃗ r ) exp(−iωt) , ⃗ r , ω ′ ) = B(⃗
B(⃗ ⃗ r ) 2π δ(ω ′ − ω) . (7.6b)
With Eq. (7.1), the wave equations become
ω2 ⃗ y) = ⃗
⃗ 2∥ + r μr
(∇ − k 2 ) E(x, 0, (7.7a)
c2
ω2 ⃗ y) = ⃗
⃗ 2∥ + r μr
(∇ − k 2 ) B(x, 0, (7.7b)
c2
where
∂2 ∂2 ∂ ∂
⃗ 2∥ =
∇ + , ⃗ ∥ = êx
∇ + êy . (7.8)
∂x2 ∂y 2 ∂x ∂y
We shall denote the “in-plane” component of a vector (i.e., the component parallel
to, or “inside”, the xy plane) by the subscript ∥. We can decompose the fields into
z components, and in-plane components,
⃗ − êz Ez = (êz × E)
E⃗∥ = E ⃗ × êz = −êz × (êz × E⃗∥ ) , (7.9a)
B ⃗ − êz Bz = (êz × B) × êz = −êz × (êz × B
⃗∥ = B ⃗∥ ) . (7.9b)
The êz ×(êz ×. . .)-operation thus amounts to a simple minus sign for in-plane vectors.
Let us try to take this a little further and decompose Faraday’s law into parallel
and transverse components. We start from the relation
⃗ = (êz ∇z + ∇
⃗ ×E
∇ ⃗ ∥ ) × (êz Ez + E ⃗ r) .
⃗∥ ) = i ω B(⃗ (7.10)
Then, in view of êz × êz = ⃗
0, we have
⃗ = êz ∇z × E
⃗ ×E
∇ ⃗∥ − êz × ∇ ⃗ ∥ Ez + ∇ ⃗∥ × E ⃗∥ = iω B ⃗∥ + iω êz Bz , (7.11)
,-- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -. - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -/ ,-- - - - - -. - - - - - -/ ,-- - - - - -. - - - - - -/
in-plane z-oriented in-plane z-oriented
⃗ = êz ∇z × B
⃗ ×B
∇ ⃗∥ − êz × ∇ ⃗ ∥ Bz + ∇ ⃗∥ × B ⃗∥ = − i ω r μr E⃗∥ − i ω r μr êz Ez .
,-- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - .-- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - / ,-- - - - - -. - - - - - - / c2 c2
,-- - - - - - - - - - - - - -. - - - - - - - - - - - - - - / ,-- - - - - - - - - - - - - - - - - - -. - - - - - - - - - - - - - - - - - - -/
in-plane z-oriented in-plane z-oriented
(7.12)
The in-plane component of Eq. (7.11) is
⃗∥ − êz × ∇
êz × ∇z E ⃗∥ .
⃗ ∥ Ez = i ω B (7.13)
We now apply the (êz × . . .)-operation to both sides and in view of Eq. (7.9), we
obtain
⃗∥ + iω (êz × B
∇z E ⃗∥ ) = ∇
⃗ ∥ Ez . (7.14)
Finally, in view of Eq. (7.1), we may replace ∇z → i k and write
i k E⃗∥ = ∇
⃗ ∥ Ez − iω(êz × B∥ ) . (7.15)
Analogously, the in-plane component of Eq. (7.12) reads
⃗∥ − êz × ∇ ω
êz × ∇z B ⃗ ∥ Bz = −i r μr E⃗∥ . (7.16)
c2
Replacing ∇z by i k, we have
⃗∥ = 1 (êz × ∇
êz × B ⃗ ∥ Bz ) −
ω r μr ⃗
E∥ . (7.17)
ik c2 k
Applying the (êz × . . .)-operation to both sides of Eq. (7.16),
∇z B⃗∥ − i ω r μr (êz × E⃗∥ ) = ∇⃗ ∥ Bz . (7.18)

c2
Using Eq. (7.1) and replacing ∇z → i k, one obtains
⃗∥ − i ω r μr (êz × E
ik B ⃗∥ ) = ∇⃗ ∥ Bz . (7.19)
c2
Also, since the divergence of both electric and magnetic fields vanishes,
∇ ⃗∥ + ∇z Ez = 0 ,
⃗∥ ⋅ E ⃗∥ + ∇z Bz = 0 .
⃗∥ ⋅ B
∇ (7.20)
⃗∥ and B
Finally, we solve for E ⃗∥ if Ez and Bz are known (and not both are zero).
Combining Eqs. (7.15) and (7.17), we may eliminate B⃗∥ and write
ω ω 2 r μr ⃗
ik E⃗∥ = ∇
⃗ ∥ Ez − ⃗ ∥ Bz ) + i
(êz × ∇ E∥ . (7.21)
k c2 k
From the last equation, multiplying by i k, we find that

ω2 ⃗∥ = i k ∇
( r μr − k 2 ) E ⃗ ∥ Ez − i ω (êz × ∇
⃗ ∥ Bz ) . (7.22)
c2
We can finally solve for E⃗∥ ,
⃗∥ = i 1
E ⃗ ∥ Ez − ω (êz × ∇
(k ∇ ⃗ ∥ )Bz ) . (7.23)
ω 2 r μr /c2 − k 2
In Eq. (7.19), we can replace E ⃗∥ with the result just obtained,
⃗∥ = i 1 ω r μr
B ⃗ ∥ Bz +
(k ∇ ⃗ ∥ ) Ez ) .
(êz × ∇ (7.24)
ω 2 r μr /c2 − k 2 c2
We have discussed waves which manifestly travel in the positive z direction. For
waves traveling in the opposite direction, one just changes k to −k. We should also
point out that in vacuum, we have r → 1 and μr → 1, so that the denominator
ω 2 r μr /c2 − k 2 vanishes. However, in this case, the z components of the electric
and magnetic fields also vanish; we have assumed that the wave propagates in the
z direction, and the magnetic and electric fields are strictly transverse in vacuum.
Indeed, we shall soon realize that the dispersion relation for electromagnetic waves
⃗
in a waveguide is different from the dispersion relation in vacuum, i.e., ω ≠ c ∣k∣,
even if r = μr = 1.
Equations (7.23) and (7.24) are valid for the general propagation of waves in a
medium with permittivity r and permeability μr , not necessarily for a waveguide.
The waveguide aspect comes in through the boundary conditions, which will be
discussed next.
7.2.2 Boundary Conditions at the Surface

Let us investigate special boundary conditions for the case of a perfect electric
conductor; the walls of a waveguide are typically idealized as conducting walls.
Inside the conductor, the electric field is zero, because any conceivably entering field
is immediately compensated by a rearrangement of conducting electrons. Hence,
we have the boundary condition
⃗ = 0,
n̂ × E∣ or E∥,S = 0 , (7.25)
S
where n̂ is the surface normal and the subscript S denotes the evaluation on a point
(x, y, z) that satisfies the equation F (x, y, z) = 0 defining the surface of the conduc-
tor (see Sec. 3.2.7). We use the notation E∥,S in order to denote the component of
E⃗ parallel and inside the surface.
There is a further boundary condition, which concerns the magnetic field. First,
we observe that in principle, there is a no obstacle against a static magnetic field en-
tering a perfect conductor. However, let us suppose that a time varying, oscillating
field enters the conductor. Then, in view of Faraday’s law,
⃗ ×E
∇ ⃗+ ∂ B ⃗ = 0, (7.26)
∂t
b Relative Permeability μr
Relative Permittivity r
Fig. 7.1 Picture of the rectangular waveguide.
one would generate a time-varying electric field in the conductor, which in turn
cannot exist. Specifically, let us suppose that n̂ is the surface normal of the conduc-
tor. The magnetic field is an oscillating field by assumption. If the projection of
⃗ were nonzero, we would generate a
the magnetic field on the surface normal n̂ ⋅ B∣ S
time-varying electric field inside the plane defining the surface. In order to see this
⃗ B
more clearly, let us assume that the surface normal is the z axis. If ∂t B∣∣ ⃗ ∝ êz Bz ,
⃗∥ × E
then ∇ ⃗∥ also is directed along the z axis; the latter quantity, however, has to
vanish. Hence, we need another boundary condition,
⃗ = 0,
n̂ ⋅ B∣ or B⊥,S = 0 . (7.27)
S
However, a nonvanishing and oscillating component of the magnetic field, in the

boundary plane, may exist in the vicinity of the conductor. By Faraday’s law,
this oscillating magnetic field gives rise to an oscillating electric field, perpendicular
to the surface, and this is not in contradiction to the boundary condition for the
electric field.
If we assume the waveguide to be aligned along the z axis, then for a rectangu-
lar waveguide (see Fig. 7.1), the boundaries are the xy, xz and yz planes. There
are two kinds of propagating electromagnetic modes for which the boundary condi-
tions (7.25) and (7.27) are fulfilled. These are called the transverse magnetic (TM)
and transverse electric (TE) modes and we shall discuss them in detail. The basis of
our considerations lies in Eqs. (7.23) and (7.24), which imply that any fields inside
the waveguide can be decomposed into those generated for a vanishing field Ez = 0
(in which case E ⃗ = E⃗∥ , and the electric field is transverse), and those generated for
a vanishing field Bz = 0 (in which case B ⃗=B ⃗∥ , and the magnetic field is transverse).
We shall discuss boundary conditions for both TE and TM modes in the follow-
ing. The first and most straightforward case concerns the TM mode,
[TM defining property:] Bz = 0 everywhere , ⃗=B
B ⃗∥ . (7.28)
Here, “transverse” is to be understood as perpendicular with respect to the beam

direction, i.e., perpendicular to the z axis. For TM modes, we also have to require
that according to Eq. (7.25), the component of the electric field tangent to the
interface (Ez ) must vanish at the surface, i.e., Ez,S = 0, giving us
[TM conditions:] Bz = 0 everywhere , Ez ∣S = 0 . (7.29)
Furthermore, for TM modes, we have according to Eq. (7.24),

⃗∥ ∝ êz × ∇
B ⃗ ∥ Ez . (7.30)
At the boundary, since Ez,S = 0 by assumption, the vector ∇ ⃗ ∥ Ez (just like the entire
⃗ z ) must be normal to the surface. Forming the vector product of ∇
gradient ∇E ⃗ ∥ Ez
with êz at the surface, we thus get a vector lying inside the surface, i.e., we have
shown that B ⃗∥ must necessarily be parallel to the surface in its immediate vicinity.
We shall verify this observation below, for the explicit formulas we shall derive for
the TM modes. For now, in view of Eqs. (7.23), (7.24) and (7.29), we summarize
the relations for the transverse magnetic (TM) modes,
[TM wave:] Bz = 0 everywhere and Ez ∣S = 0 , (7.31a)
⃗∥ = i 1
E k∇⃗ ∥ Ez , (7.31b)
r μr ω 2 /c2 − k 2
⃗∥ = i 1
B ⃗ ∥ Ez ) ,
ω r μr (êz × ∇ (7.31c)
r μr ω 2 /c2 − k 2
⃗ 2∥ + r μr ω2
0 = (∇ − k 2 ) Ez (x, y) . (7.31d)
c2
For TE modes, we have different relations,
[TE defining property:] Ez = 0 everywhere , ⃗=E

E ⃗∥ , (7.32)
and the boundary condition (7.25) for the electric field is automatically fulfilled.
In view of Eq. (7.24), we must have B ⃗∥ ∝ ∇⃗ ∥ Bz . Thus, the normal component
⃗∥ =
n̂ ⋅ B ⃗ can alternatively be obtained by projecting ∇
n̂ ⋅ B ⃗ ∥ Bz onto the surface
normal. The corresponding condition on Bz is
⃗∥ ∣ ∝ (n̂ ⋅ ∇
⃗ ∝ n̂ ⋅ B ∂Bz
0 = n̂ ⋅ B∣ ⃗ ∥ ) Bz ∣ = (n̂ ⋅ ∇)
⃗ Bz ∣ ≡ ∣ . (7.33)
S S S S ∂n S
The latter expression merely is a definition motivated by the relation
∂Bz (⃗
r + ξ n̂)
∣ ⃗ z = n̂ ⋅ ∇
= n̂ ⋅ ∇B ⃗ ∥ Bz . (7.34)
∂ξ ξ=0
We summarize that for a TM mode,

∂Bz
[TE conditions:] Ez = 0 everywhere , ∣ = 0. (7.35)
∂n S
In view of Eqs. (7.23), (7.24) and (7.35), we find for the transverse electric (TE)
modes,
∂Bz
[TE wave:] Ez = 0 everywhere and ∣ = 0, (7.36a)
∂n S
⃗∥ = i 1
B ⃗ ∥ Bz ,
k∇ (7.36b)
r μr ω 2 /c2 − k 2
⃗∥ = i 1 ⃗ ∥ Bz )] ,
E [−ω (êz × ∇ (7.36c)
r μr ω 2 /c2 − k 2
ω2
⃗ 2∥ + r μr
0 = (∇ − k 2 ) Bz (x, y) . (7.36d)
c2
The boundary conditions give rise to eigenvalues of k (dependent on ω) for which the
propagation is allowed. These eigenvalues may be continuous. Since the boundary
conditions for Ez and Bz are different, the eigenvalues are also different for TE as
opposed to TM modes. The allowed TE and TM waves (and the TEM wave, if it
exists) provide a complete set of waves from which one can construct an arbitrary
electromagnetic disturbance in the waveguide or cavity.
7.2.3 Modes in a Rectangular Waveguide

We aim to determine the TE modes in a rectangular waveguide with edge length a
in the x direction and b in the y direction (with a > b), so that 0 < x < a and 0 < y < b
(see Fig. 7.1). So, E⃗∥ will be found from Bz alone, according to the relation (7.36c),
⃗∥ = i 1 ⃗ ∥ )Bz ]
E [−ω(êz × ∇
r μr ω 2 /c2 − k 2
ω k ω ⃗∥ .
= − êz × (i ⃗ ∥ Bz ) = − êz × B
∇ (7.37)
k r μr ω 2 /c2 − k 2 k
This, in particular, means that E⃗∥ and B⃗∥ are still perpendicular to each other.
The general solution of the wave equation (7.7b) for Bz is
⃗ ⃗
Bz (x, y) = C1 e+ik⋅⃗r∥ + C2 e−ik⋅⃗r∥ , r⃗∥ = êx x + êy y . (7.38)
The form for Bz (x, y) which is non-zero when x = y = 0 is
Bz (x, y) = B0 cos(kx x) cos(ky y) . (7.39)
⃗ ∥ ) Bz ∣S = 0, for transverse
We recall the boundary condition (7.35), namely, (n̂ ⋅ ∇
magnetic (TM) modes in component form,
∂
B0 cos(kx x) cos(ky y)∣ = 0, (7.40a)
∂x x=0,a
∂
B0 cos(kx x) cos(ky y)∣ = 0. (7.40b)
∂y y=0,b
Any presence of a sine function in these two equations would lead to a nonvanishing
derivative unless B0 vanishes and thus the whole term. This justifies, a posteriori,
our ansatz with a cosine. The boundary condition (n̂ ⋅ ∇ ⃗ ∥ ) Bz ∣S = 0 leads to the
following condition on the components of the wave vector,
mπ
sin(kx a) = sin(kx 0) = 0 , kx = , m = 0, 1, 2, . . . , (7.41a)
a
nπ
sin(ky b) = sin(ky 0) = 0 , ky = , n = 0, 1, 2, . . . . (7.41b)
b
This means that for given m and n, a dispersion relation can be established between
ω and k,
mπx nπy
Bz (x, y) = B0,mn cos ( ) cos ( ), (7.42)
a b
mπ 2 nπ 2 ω2
( ) + ( ) = r μr 2 − k 2 , m, n = 0, 1, 2, 3, . . . , (7.43)
a b c
√
ω2 mπ 2 nπ 2
k = r μr 2 − ( ) −( ) . (7.44)
c a b
Propagation in the (m, n) mode cannot proceed unless the solution for k is real,
1/2 1/2
c mπ 2 nπ 2 πc m 2 n 2
ω > ωmn ≡ √ [( ) +( ) ] =√ [( ) + ( ) ] . (7.45)
r μr a b r μr a b
For ω < ωmin , the wave number k becomes purely imaginary, k = i ∣k∣, and the
wave, propagating in the z direction, being proportional to exp(−∣k∣ z), becomes
evanescent. Moreover, ωmn is the m and n-dependent waveguide angular frequency.
The minimum frequency is obtained for
π c
ωmin = ωmin;TE = √ , m = 1, n = 0, a > b. (7.46)
r μr a
In view of the occurrence of the factor c/a, the minimum frequency for propagation
in the waveguide is commensurate with the inverse time it takes light to travel the
spatial dimension of the waveguide. For a non-trivial solution, m and n cannot both
be zero. When the condition (7.45) is fulfilled, real rather than imaginary solutions
can be found for the wave vector.
Let us summarize the field configuration for a TEmn mode, and start with the
z component,
mπx nπy
Bz (x, y) = Bz,mn (x, y) = B0,mn cos ( ) cos ( ), Ez (x, y) = 0 , (7.47a)
a b
The in-plane component of the B ⃗ field reads as follows,
−1
⃗∥ (m, n) = ik mπ 2 nπ 2
B ⃗ ∥ Bz,mn ei (k z−ω t) = −i k [(
∇ ) + ( ) ] B0,mn
ω 2 r μr /c2 − k 2 a b
mπ mπx nπy nπ mπx nπy
× [êx sin ( ) cos ( ) + êy cos ( ) sin ( )] ei(k z−ω t) ,
a a b b a b
(7.47b)
while the electric field is given by

2 2 −1
ω ⃗∥ (m, n)] = −i ω [( mπ ) + ( nπ ) ] B0,mn
E⃗∥ (m, n) = − [êz × B
k a b
nπ mπx nπy mπ mπx nπy
× [êx cos ( ) sin ( ) − êy sin ( ) cos ( )] ei(k z−ω t) .
b a b a a b
(7.47c)
The modulus of the wave vector k is given by Eq. (7.44). It is instructive to indicate
the solution for the fundamental TE1,0 mode with m = 1 and n = 0 separately,
πx
Bz (x, y) = B0 cos ( ) ei (k z−ω t) , (7.48a)
a
B⃗∥ (m = 1, n = 0) = − i ka B0 êx sin ( πx ) ei (k z−ω t) , (7.48b)
π a
E⃗∥ (m = 1, n = 0) = i ωa B0 êy sin ( πx ) ei (k z−ω t) . (7.48c)
π a
Transverse Electric Mode TE1,0

E
Bz (x, y)

B and B
E
Fig. 7.2 Four plots illustrating the transverse electric mode TE1,0 . The leftmost upper plot
shows Bz (x, y) as a function of x and y. As the mode is transverse, Ez (x, y) vanishes. The
rightmost upper plot shows, as a vector plot, E ⃗∥ (x, y) in a plane of constant z, at a particular
time of the sinusoidal oscillation. All vector arrows oscillate sinusoidally. The left lower plot
shows B ⃗∥ (x, y) in a plane of constant z, at a particular time of the sinusoidal oscillation. It is
⃗ ∥ )Bz vanishes at the surface, as is required for all TE modes. In the
clearly visible that (n̂ ⋅ ∇
right lower plot, we illustrate the perpendicular character of the two fields.

E
Bz (x, y)

B and B
E
Fig. 7.3 Same as Fig. 7.2, for the transverse electric mode TE1,1 .
The dispersion relation for the fundamental mode is

√ B
ω 2 π 2 √ ωD 2
k = k10 = r μr ( ) − ( ) = r μr D E1 − 1 ( πc ) . (7.48d)
c a c r μr ωa
We recall that propagation happens only for
π c
ω 2 > ωmin;TE = √ . (7.48e)
r μr a
Note the 90○ phase difference between Bx , By and Bz arising from the −i = e−iπ/2
factor. The in-plane components B ⃗∥ and E⃗∥ are 180○ out of phase. Graphical
representations of the first TE modes can be found in Figs. 7.2, 7.3, 7.4 and 7.5.
The dispersion relation (7.44) can trivially be rewritten as
√
r μr √ 2
1/2
πc m 2 n 2
k= ω − ωmn ,
2 ωmn = √ [( ) + ( ) ] , (7.49)
c r μr a b
where ωmn is the cutoff frequency for the mode. The dispersion relations in
Figs. 7.6, 7.7 and 7.8 represent the functional relationships ω = ω(k) for the first
few modes of a typical waveguide.
It is often convenient to choose the dimensions of the waveguide so that at the
operating frequency, only the lowest mode TE1,0 can occur. Since the wave number,

E
Bz (x, y)

B and B
E
Fig. 7.4 Same as Figs. 7.2 and 7.3, for the transverse electric mode TE2,1 .

E
Bz (x, y)

B and B
E
Fig. 7.5 Same as Figs. 7.2, 7.4, and 7.5, but for the transverse electric mode TE2,2 .
Fig. 7.6 Dispersion relations for the first few modes of a rectangular waveguide, for
all modes with m ≤ 2 and n ≤ 2. The dimensions of the waveguide are a = 2.3 cm and
b = 0.71 cm. The relative permittivity of the waveguide medium is r = 1.2, and the
permeability is μr = 1.0, independent of the angular frequency ω. The abscissa gives
the inverse wavelength 1/λ in units of inverse meter (m), where 1/λ = k/(2π), and the
ordinate axis gives the light frequency f = ω/(2π) in units of cycles per second, which
is equal to Hertz (Hz). We note that the SI mksA unit of the angular frequency ω is
radians per second. A full oscillation period per second corresponds to one cycle per
second, or 2π radians per second (rad/s).
Fig. 7.7 Same as Fig. 7.6, but for a region of medium wave number, where the modes
√
of the waveguide approach the free dispersion relation ω = ck/ r μr .
kmn , is always less than the “free space” value, the wavelength in the waveguide
is always larger than the free space wavelength. Surprisingly, this means that the
phase velocity in free space is always greater than the free-space value, and equal
to ω/kmn .
To supplement Eq. (7.47), let us also give the formulas for the TM modes,
mπx nπy
Ez (x, y) = E0,mn sin ( ) sin ( ) ei (k z−ωt) , Bz (x, y) = 0 . (7.50a)
a b
Fig. 7.8 Same as Figs. 7.6 and 7.7, but for even larger wave numbers. The deviation
√
from the dispersion relation without waveguide, ω = ck/ r μr is hardly discernible.
The in-plane component of the electric field is
−1
k mπ 2 nπ 2
E⃗∥ (m, n) = i ⃗ ∥ Ez = ik [(
∇ ) + ( ) ] E0,mn
r μr ω 2 /c2 − k 2 a b
mπ mπx nπy nπ mπx nπy
cos (
×[êx ) sin ( ) + êy sin ( ) cos ( )] ei(k z−ωt) ,
a a b b a b
(7.50b)
⃗
while the same component for the B field reads as follows,
2 2 −1
B ⃗∥ ) = i r μr ω [( mπ ) + ( nπ ) ]
⃗∥ (m, n) = ω r μr (êz × E E0,mn
kc 2 c 2 a b
nπ mπx nπy mπ mπx nπy
×[−êx sin ( ) cos ( )+êy cos ( ) sin ( )] ei(k z−ωt) .
b a b a a b
(7.50c)
Illustrations are provided in Figs. 7.9, 7.10 and 7.11. The dispersion relation for
TM modes is just the same as for TE modes, but the available values of m and n
are different,
mπ 2 nπ 2 ω2
( ) + ( ) = r μr 2 − k 2 , m, n = 1, 2, 3, . . . . (7.51)
a b c
The z component of the electric field is Ez (x, y) = 0 at x = 0 and x = a, and at y = 0

and y = b. In the TM modes, if n = 0 or if m = 0, then we have Ez = 0, as a conse-
quence of the properties of the sine versus the cosine function. So, the m = 1, n = 0
mode is not available for TM waves because the fields simply vanish. The lowest
Transverse Magnetic Mode TM1,1

E
Bz (x, y)

B and B
E
Fig. 7.9 Four plots illustrating the transverse magnetic mode TM1,1 . The leftmost upper plot
shows Ez (x, y) as a function of x and y. As the mode is transverse magnetic, Bz (x, y) vanishes.
The rightmost upper plot shows, as a vector plot, E ⃗∥ (x, y) in a plane of constant z, at a particular
time of the sinusoidal oscillation. All vector arrows oscillate sinusoidally. The left lower plot shows
B⃗∥ (x, y) in a plane of constant z, at a particular time of the sinusoidal oscillation. The magnetic
field lines form “closed loops” in the xy plane, as they do for all TM modes. In the right lower
plot, the perpendicular character of the two fields is illustrated.
possible mode is n = m = 1 with

1/2
πc 1 2 1 2 π c
ωmin;TM = ω11 =√ [( ) + ( ) ] > ωmin;TE = ω10 = √ . (7.52)
r μr a b r μr a
Thus, the transverse electric mode TE10 provides the smallest available frequency
for propagation in the waveguide. For a > b, for frequencies
ωmin;TE < ω < ωmin;TM , (7.53)
only one mode is available for wave propagation. The phase velocity is given by
√
ω c k 2 + (ωmn /c)2 c ω
vp = = √ =√ √
k r μr k r μr ω 2 − ωmn
2
−1/2
c ω2
=√ (1 − mn ) > c, (7.54)
r μr ω2

E
Bz (x, y)

B and B
E
Fig. 7.10 Four plots illustrating the transverse magnetic mode TM2,1 , otherwise the same as
Fig. 7.9.
where the last inequality holds for r = μr = 1 and ωmn ≠ 0. The group velocity is
given by
dω c d c2 k 1
vg = =√ [(ck)2 + ωmn2
]1/2 = √ √
dk r μr dk r μr (ck)2 + ωmn
2
√ 1/2
c ω 2 − ωmn
2 c ω2
=√ =√ (1 − mn ) → 0, ω → ωmn , (7.55)
r μr ω r μr ω2
where we indicate a few possible, alternative forms. The latter limit indicates
the possibility of “slow light” because the group velocity is the speed at which
a
√ light pulse (or, “light pulse train”) travels. Furthermore, the functional form
ω 2 − ωmn
2 /ω is consistent with the group velocity always remaining smaller than
the speed of light. The index n(ω) of refraction can be determined from
√
ω ω 2 − ωmn
2
k = n(ω) = , (7.56a)
c c
√
ω2 1 c
n(ω) = 1 − mn =√ < 1, (7.56b)
ω2 r μr vp
where the latter inequality holds for vp > c.

E
Bz (x, y)

B and B
E
Fig. 7.11 Four plots illustrating the transverse magnetic mode TM2,2 , otherwise the same as
Figs. 7.10 and 7.11.
7.3 Resonant Cavities
7.3.1 Resonant Cylindrical Cavities
In a waveguide, modes are characterized by two discrete quantum numbers m and

n, and a continuous quantum number k (or ω). The oscillations in the xy plane are
quantized. In a cylindrical cavity (as opposed to a waveguide), the oscillations in
the third direction (z direction) are also quantized, and one ends up with a fully
discrete set of modes. In some sense, the waves are “bound” into the cavities,
and the additional quantization conditions induced by the cavity correspond to
the “bound states” in atoms, which are also discrete. The resonant modes are
standing waves, not traveling waves as in the z-oriented waveguide, and resemble
the “resonant modes” of, say, a drum or the string of a violin. In a waveguide, the
allowed frequencies are of the form ω = ωmn (k), where m and n are discrete numbers
(integers) but k is a continuous variable. The cavity adds one more boundary
condition, in the z direction, because both the lower as well as the upper end are
covered by a perfect conductor. This implies a further quantization condition, and
we anticipate that the allowed cavity mode frequencies are of the form ωmnp , with
three integer subscripts.
We start the discussion with the TM modes and use cylindrical coordinates ρ,
ϕ, and z. Let the cavity extend from z = 0 to z = d and from ρ = 0 (symmetry axis)
to ρ = R. The first step is to separate the spatial and time dependence of the fields,
just as we did for the waveguide,
⃗ r , t) = E(⃗
E(⃗ ⃗ r ) e−iωt , ⃗ r , t) = B(⃗
B(⃗ ⃗ r ) e−iωt . (7.57)
Let us assume first that the magnetic field is transverse, i.e., that only the electric
field has a z component,
pπz pπz
Ez = f (ρ) g(ϕ) h(z) = f (ρ) ei m ϕ [C sin ( ) + D cos ( )] . (7.58)
d d
The Helmholtz equation requires that
ω2 1 ∂ ∂ 1 ∂2 ∂2 ω2
⃗ 2 + r μr
(∇ ) Ez = ( ρ + 2 + 2 + r μr 2 ) Ez
c 2 ρ ∂ρ ∂ρ ρ ∂ϕ 2 ∂z c
1 ∂ ∂ 1 ∂2 pπ 2 ω2
=[ ρ + 2 + (− ( ) + r μ r )] Ez
ρ ∂ρ ∂ρ ρ ∂ϕ2 d c2
,-- - - - - - - - .- - - - - - - - - / ,-- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -. - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -/
=−m2 /ρ2 =γp2
∂2 1 ∂ m2
=( 2
+ − 2 + γp2 ) Ez = 0 . (7.59)
∂ρ ρ ∂ρ ρ
The parameter γp is defined as
ω2 pπ 2
γp2 = r μr − ( ) . (7.60)
c2 d
We recall the defining differential equation for the ordinary Bessel function
Eq. (2.144),
∂2 1 ∂ n2
( + − + 1) Jn (ρ) = 0 . (7.61)
∂ρ2 ρ ∂ρ ρ2
In the ansatz (7.58) for the TM modes, we can thus set f (ρ) = Jm (γp ρ),
pπz pπz
[TM:] Ez = E0 Jm (γp ρ) ei m ϕ [C sin ( ) + D cos ( )] , (7.62)
d d
while, of course, Bz = 0 for the TM mode. The constants C and D have to be
determined from the boundary conditions. The gradient operator, in spherical
coordinates, has the representation
∂ 1 ∂ ∂
⃗ = êρ
∇ + êϕ + êz ⃗ ∥ + êz ∇z .
=∇ (7.63)
∂ρ ρ ∂ϕ ∂z
In general, because the z axis is still a valid symmetry axis of the problem, we can
apply Eq. (7.14) to read as follows,
⃗∥ + iω (êz × B
∇z E ⃗∥ ) = ∇
⃗ ∥ Ez . (7.64)
Furthermore, we recall Eq. (7.18), which for Bz = 0 (TM mode) reads as
∇z B⃗∥ − i ω r μr (êz × E⃗∥ ) = 0 . (7.65)

c2
In a waveguide, we could replace ∇z E⃗∥ = ik E ⃗∥ and ∇z B ⃗∥ = ik B⃗∥ without altering
ikz
the z dependence, which was exclusively manifest in the e prefactor. We could
⃗∥ , after replacing ∇z B
thus solve Eq. (7.65) for B ⃗∥ → i k B ⃗∥ , and insert the result in
Eq. (7.64), finally obtaining E ⃗∥ as a function of Ez .
For the cavity, E ⃗∥ and Ez will have different sin(p π z/d) and cos(p π z/d)
dependences. and we have to integrate Eq. (7.65) with respect to z before us-
ing the relation in Eq. (7.64). For an electric field of the form (7.62), whose z
dependence is sinusoidal or cosinusoidal, we have the relations
2 2
⃗∥ = − ( π p ) ∫ E
∇z E ⃗∥ = − ( π p ) ∫∫
⃗∥ dz , E ⃗∥ dz dz .
E (7.66)
d d
A double integration with respect to dz thus is equivalent to a multiplication by
−(d/(π p))2 .
We can integrate Eq. (7.65) with respect to z, and solve ⃗∥ , with the result
for B
⃗∥ = i ω r μr (êz × ∫ E
B ⃗∥ dz) , (7.67)
c2
and insert the result in the z-integrated form of Eq. (7.64),
E ⃗∥ dz) = ∫ ∇
⃗∥ + i ω (êz × ∫ B ⃗ ∥ Ez dz . (7.68)
So, now, inserting (7.67) into (7.68), we have
⃗∥ + i ω (êz × i ω r μr (êz × ∫∫ E
E ⃗∥ dzdz)) = ∫ ∇
⃗ ∥ Ez dz . (7.69)
c2
We now use Eq. (7.9) for the in-plane component E ⃗∥ ,
2
⃗∥ − i2 ω r μr ∫∫ E
E ⃗∥ dz dz = ∫ ∇
⃗ ∥ Ez dz . (7.70)
c2
Using Eq. (7.66), we can transform the double integration with respect to z into a
multiplication by −(d/(pπ))2 , and so
2 2
⃗∥ [1 − ω r μr ( d ) ] = ∫ ∇
E ⃗ ∥ Ez dz . (7.71)
c2 pπ
We now pull out a prefactor,
d 2 ⃗ pπ 2 ω 2
( ⃗ ∥ Ez dz ,
) E∥ [( ) − 2 r μr ] = ∫ ∇ (7.72)
pπ d c
so that with Eq. (7.60),
pπ −2 2 ⃗
−( ⃗ ∥ Ez dz .
) γp E∥ = ∫ ∇ (7.73)
d
It now becomes clear why in the ansatz given by Eq. (7.62), we need to choose the
term with the cosine as opposed to the sine. Namely, the in-plane field E ⃗∥ needs to
vanish inside the endcap surfaces at z = 0 and z = d. According to Eq. (7.73), E ⃗∥
is given by an integral of Ez with respect to z; the in-plane gradient operator ∇ ⃗∥
[Eq. (7.63)] does not play a role in this consideration. The sine term in Eq. (7.62)
would lead to a term proportional to cos(pπz/d) in E⃗∥ , which would lie inside
the endcap surfaces, leading to a contradiction with our assumption of a perfect
conductor. Hence, we choose the cosine term in Eq. (7.62), which upon integrating
⃗∥ ,
with respect to z leads to following result for E
2
⃗∥ = − ( pπ ) ( d ) 1 E0 ∇
E ⃗ ∥ Jm (γp ρ) sin (
pπz imϕ −iωt
)e e . (7.74)
d pπ γp2 d
Here, E0 is an overall multiplicative constant, and we recall Eq. (7.63) for the
definition of ∇ ⃗∥ =
⃗ ∥ . In view of Eq. (7.67), which we recall for convenience, B
⃗ ⃗
i c2 r μr (êz × ∫ E∥ dz), the final result for B∥ is
ω
⃗∥ = E0 (i ω r μr ) 1 (êz × ∇
B ⃗ ∥ Jm (γp ρ)) cos (
pπz imϕ −iωt
)e e . (7.75)
c 2 2
γp d
We have used the result [see Eq. (7.63)] Finally, we summarize the result for TM
modes in a cylindrical cavity,
pπz imϕ −iωt
Ez = E0 Jm (γp ρ) cos ( )e e , Bz = 0 , (7.76a)
d
⃗∥ = − pπ 1 E0 ∇
E ⃗ ∥ Jm (γp ρ) sin (
pπz imϕ −iωt
)e e , (7.76b)
d γp2 d
⃗∥ = E0 (i ω r μr ) 1 (êz × ∇
B ⃗ ∥ Jm (γp ρ)) cos (
pπz imϕ −iωt
)e e . (7.76c)
c2 γp2 d
Here, we suppress the arguments of the electric field in our notation, i.e., we under-
⃗∥ = E
stand that Ez = Ez (ρ, ϕ, z, t), E ⃗∥ (ρ, ϕ, z, t), and B
⃗∥ = B
⃗∥ (ρ, ϕ, z, t).
There is one more point to clarify. Namely, Ez becomes parallel to the outer
surfaces of the cylinder near ρ = R, where R is the cylinder radius. Hence, we need
to require that
Jm (γp R) = 0 , γp R = xmn , Jm (xmn ) = 0 , m = 0, 1, 2, . . . , n = 1, 2, 3, . . . ,

(7.76d)
where xmn is the nth zero of the Bessel function of order m [see Eq. (2.158)]. So,
because of the condition that Ez must vanish on the outer rim of the cylinder, we
must postulate that
! xmn
γp = γmn = γmnp = , n = 1, 2, 3, . . . . (7.76e)
R
This condition gives us an equation which needs to be fulfilled by the “quantum

numbers” m, n, and p,
ω2 pπ 2 xmn 2
γp2 = γmnp
2
= r μ r − ( ) = ( ) , (7.76f)
c2 d R
√
c pπ 2 xmn 2
ω = ωmnp = √ ( ) +( ) , (7.76g)
r μr d R
m = 0, 1, 2, . . . , n = 1, 2, 3, . . . , p = 0, 1, 2, . . . . (7.76h)
The quantum numbers are m (azimuthal), n (radial) and p (longitudinal). Indeed,
the quantum number n start from n = 1 because we must integrate at least up to the
first node of the Bessel function, m starts from zero because the magnetic quantum
number may vanish, and p also can be zero because Ez is proportional to cos(pπz/d),
not sin(pπz/d), so the fields do not all vanish if we set p = 0. The quantum number
n counts the nodes of the radial wave function and acts, in some sense, as the
principal quantum number (the number of nodes actually is n − 1 for given n).
The quantum number m acts as a projection quantum number characterizing the
“angular momentum” about the quantization axis, Lz = −i∂/∂ϕ [see Eq. (2.30c)].
The quantum number p characterizes the number of oscillations in the z direction.
The fundamental TM mode in a cylindrical cavity is the 010 mode (m = 0, n = 1,
and p = 0). The numerical value of the first zero of the Bessel function of order
m = 0 is x01 = 2.404 825 557 . . . (see the discussion in Sec. 2.4.3 and the numerical
values in Table 2.1). It has the lowest frequency available,
c x01 c 2.404 825
ω010 = √ ≈√ . (7.77)
r μr R r μr R
As will be shown in the following, this frequency typically is less than the cor-
responding minimum frequency for the lowest TE mode, and so the fundamental
mode for a cylindrical cavity is TM rather than TE. The ground state is a radially
symmetric state (no ϕ dependence) with quantum numbers m = 0, n = 1, and p = 0.
The TM states with m = 0 and p = 0 are characterized by the field configurations
(“wave functions”),
Ez = E0 J0 (γ0n ρ) e−iω0n0 t , ⃗∥ = 0 ,
E
⃗∥ = E0 iωr μr (êz × ∇
B ⃗ ∥ J0 (γ0n ρ)) e−iω0n0 t , Bz = 0 . (7.78)
2 c2
γ0n
Of course, J0 (γ0n R) = 0. The wave function with n = 1 has one node, with n = 2,
two nodes, etc., much like in quantum mechanics. This is illustrated in Fig. 7.12
for the ground state. The electric and magnetic fields of the ground state with
m = p = 0 and n = 1 are shown in Fig. 7.13. In view of Eq. (7.78), the oscillations of
the magnetic and electric fields are 90○ out of phase for the standing wave. Because
of the longitudinal electric field, TM modes are useful for accelerator cavities in
storage rings.
Fig. 7.12 Illustration of the first few radial wave functions for the cylindrical cavity.
(a) (b)
Fig. 7.13 Illustration of the electric field for the TM ground state of the cylindrical cavity
(left plot), and of the magnetic field (right plot). The electric field has a z component
(and no transverse component), while the magnetic field is transverse and circulating
around the symmetry axis.
We now switch to the TE modes. Our ansatz for the z components of the electric
and magnetic fields of a TE mode is
pπz imϕ −iωt
[TE:] Bz = B0 Jm (γ p ρ) sin ( )e e , Ez = 0 , (7.79)
d
The γ p are assigned differently as compared to the γp for the TM modes. The
normal component of the B ⃗ field has to vanish on the outer rim of the cylinder, in
view of the boundary condition (7.27). Now, Eq. (7.63) mandates that the gradient
operator, in cylindrical coordinates, has a component in the radial direction which

is just given by the partial derivative with respect to ρ. We thus have to require
that
xmn ′
γ p = γ mnp = , Jm (xmn ) = 0 , (7.80)
R
where we appeal to Eq. (2.159). For TE modes, in view of Eqs. (7.14) and (7.18),
we have in general,
⃗∥ + i ω (êz × B
∇z E ⃗∥ ) = ∇
⃗ ∥ Ez = ⃗
0, (7.81a)
∇z B⃗∥ − i ω r μr (êz × E⃗∥ ) = ∇
⃗ ∥ Bz . (7.81b)
c2
We integrate the latter equation with respect to z and obtain
B⃗∥ − i ω r μr (êz × ∫ E ⃗∥ dz) = ∫ ∇ ⃗ ∥ Bz dz . (7.82)

c2
From Eq. (7.81a), we have upon integration with respect to z,
⃗∥ = −i ω (êz × ∫ B
E ⃗∥ dz) . (7.83)
Inserting Eq. (7.83) into (7.82), we have the following relation,

2
B⃗∥ − ω r μr (êz × (êz × ∫ B
⃗∥ dz)) = ∫ ∇
⃗ ∥ Bz dz . (7.84)
c2
Using Eq. (7.9), this becomes
2
⃗∥ + ω r μr ∫∫ B
B ⃗∥ dzdz = ∫ ∇ ⃗ ∥ Bz dz . (7.85)
c2
Within the ansatz (7.79), a double integration with respect to z is equivalent to a
multiplication by −(d/(pπ))2 [see also Eq. (7.66)], and so
2 2
⃗∥ [1 − ω r μr ( d ) ] = ∫ ∇
B ⃗ ∥ Bz dz . (7.86)
c2 pπ
We can rewrite this as
d 2 ⃗ pπ 2 ω 2 ⃗ ∥ Bz dz .
( ) B ∥ [( ) − 2 r μr ] = ∫ ∇ (7.87)
pπ d c
Because the z component of the magnetic field given in (7.79) has to fulfill the
Helmholtz equation (7.59) separately, we require that
ω2 pπ 2
γ p2 = 2
r μr − ( ) . (7.88)
c d
Hence,
pπ −2 2 ⃗
−() γ p B∥ = ∫ ∇ ⃗ ∥ Bz dz . (7.89)
d
We can then trivially integrate the ansatz (7.79) with respect to z, to simplify the
right-hand side,
pπ −2 2 ⃗ pπ ⃗ ∥ Jm (γ p ρ) cos ( pπz ) eimϕ e−iωt .
−( ) γ p B∥ = − ( ) B0 ∇ (7.90)
d d d
⃗∥ , we obtain
Solving for B
⃗∥ = B0 ( pπ ) 1 ∇
B ⃗ ∥ Jm (γ p ρ) cos (
pπz imϕ −iωt
)e e . (7.91)
d γ p2 d
The boundary condition is fulfilled if
ω2 pπ 2 xmn 2
γ p2 = r μ r − ( ) = ( ) . (7.92)
c2 d R
Solving for ω and setting ω = ωmnp , we have
√
c pπ 2 xmn 2
ωmnp = √ ( ) +( ) . (7.93)
r μr d R
⃗∥ from Eq. (7.81a) after z integration,
Finally, we can determine E
⃗∥ = − iω (êz × ∫ B
E ⃗∥ dz) ,
1 pπz imϕ −iωt

= − iω B0 ⃗ ∥ Jm (γ p ρ)) sin (
(êz × ∇ )e e . (7.94)
2
γp d
It now becomes clear why in the ansatz (7.79), we need to choose the term with
the sine as opposed to the cosine. Just as for the TM modes, the requirement is
that the in-plane field E ⃗∥ vanish inside the endcap surfaces at z = 0 and z = d. A
cosine term in Eq. (7.79) would lead to a cosine term in Eq. (7.94), leading to an
inconsistency inside the endcap surfaces at z = 0 and z = d.
On the other hand, the sine term in Eq. (7.79) [as opposed to the cosine term
in Eq. (7.62)] may lead to a problem in the case p = 0, because in this case, the
“generating” z component of the magnetic field (7.79) and the in-plane component
of the electric field (7.94) vanish altogether. Specifically,
pπz imϕ −iωt
Bz = B0 Jm (γ p ρ) sin ( )e e = 0, (p = 0) , (7.95)
d
and
E⃗∥ = −iω B0 1 êz × ∇⃗ ∥ Jm (γ p ρ) sin (
pπz imϕ −iωt ⃗
)e e = 0, (p = 0) . (7.96)
γ p2 d
However, one might counter argue that the in-plane component (7.91) of the mag-
netic field does not vanish for p = 0,
⃗∥ = B0 ( pπ ) 1 ∇
pπz imϕ −iωt
)e e ≠ 0, (p = 0) , (7.97)
2
d γp d
and so the mode with p = 0 might still exist as a nontrivial field configuration. The
final blow to the idea of a p = 0 comes when we consider that we cannot fulfill
Eq. (7.18),
∇z B⃗∥ − i ω r μr (êz × E⃗∥ ) = ∇
⃗ ∥ Bz , (7.98)
c2
because Bz = 0, and E⃗∥ = ⃗
0, but ∇z B ⃗∥ ≠ 0. This means that the mode with p = 0 is
excluded for TE modes, in contrast to TM modes.
We summarize the result for TE modes in a cylindrical cavity,

pπz imϕ −iωt
Bz = B0 Jm (γ p ρ) sin ( )e e , Ez = 0 , (7.99a)
d
⃗∥ = − iω B0 1 êz × ∇
E ⃗ ∥ Jm (γ p ρ) sin (
pπz imϕ −iωt
)e e , (7.99b)
γ p2 d
⃗∥ = B0 ( pπ ) 1 ∇
pπz imϕ −iωt
)e e , (7.99c)
d γ p2 d
where the radial quantization is taken into account by the formulas,
ω2 pπ 2 xmn 2
γ p2 = γ mnp
2
=2
r μr − ( ) = ( ) , (7.99d)
c d R
√
c pπ 2 xmn 2 ′
ω = ωmnp = √ ( ) +( ) , Jm (xmn ) = 0 . (7.99e)
r μr d R
According to Eq. (7.80), xmn is the nth zero of the derivative of the Bessel function
of order m. Just like for TM modes, the quantum numbers are m (azimuthal), n
(radial) and p (longitudinal). One might think that the fundamental TM mode in
a cylindrical cavity is the 011 mode (m = 0, n = 1, and p = 1). The reasoning is as
follows: The lowest-order Bessel function has m = 0, the first zero of its derivative
is at n = 1, and we have to have at least p = 1 because the magnetic field Bz is
otherwise vanishing. The numerical value is easily found as x01 = 3.831 705 970 . . . .
However, according to Table 2.2, the first zero of the Bessel function of order m = 1
actually occurs before the first zero of J0 (x), namely, at x11 = 1.841 183 781 . . . .
Hence, the corresponding angular frequency for the fundamental TE mode is
√ √
c π 2 x11 2 c π 2 1.841 183 2
ω 111 = √ ( ) +( ) ≈√ ( ) +( ) . (7.100)
r μr d R r μr d R
TE modes are useful for giving a transverse deflection to a beam in an accelerator,
but are not much use for providing acceleration.
7.3.2 Resonant Rectangular Cavities

In order to describe the resonant TE and TM eigenmodes of a rectangular cavity,
it is advantageous to start from the vector potential, i.e., from an ansatz where
⃗ r , t) = A(⃗
A(⃗ ⃗ r) e−iωt , ∇ ⃗ r) = 0 ,
⃗ ⋅ A(⃗ r, t) = 0 ,
Φ(⃗ (7.101)
so that
⃗ r , t) = ∇
B(⃗ ⃗ × A(⃗ ⃗ r , t) = − ∂ A(⃗
⃗ r, t) ,
E(⃗ ⃗ r , t) . (7.102)
∂t
We shall first derive a wave equation for the vector potential, and start from the
Ampere–Maxwell law in the absence of source terms,
⃗ r, ω) = −iω D(⃗
⃗ × H(⃗
∇ ⃗ r, ω) . (7.103)
In a material with relative permittivity r and relative permeability μr , one has

⃗ r , ω) = 1 ⃗ r , ω) = 1 ⃗ r , ω)
H(⃗ B(⃗ ⃗ × A(⃗
∇ (7.104)
μr (ω) μ0 μr (ω) μ0
and
⃗ r , ω) = r (ω) 0 E(⃗
D(⃗ ⃗ r , ω) = i r 0 ω A(⃗
⃗ r, ω) . (7.105)
Inserting Eqs. (7.104) and (7.105) into (7.103), one obtains
1 ⃗ r , ω) ,
⃗ r, ω)) = r (ω) 0 ω 2 A(⃗
⃗ × (∇
∇ ⃗ × A(⃗ (7.106)
μr (ω)μ0
⃗ ⋅ A⃗ = 0 [see Eq. (7.101)] results in
which for ∇
⃗ × (∇
∇ ⃗ r, ω)) = ∇
⃗ × A(⃗ ⃗ (∇ ⃗ r , ω)) − ∇
⃗ ⋅ A(⃗ ⃗ r, ω)
⃗ 2 A(⃗
ω2 ⃗
⃗ r , ω) = r (ω)μr (ω)
⃗ 2 A(⃗
= −∇ A(⃗
r, ω) . (7.107)
c2
We thus obtain the wave equation for the vector potential,
ω2 ⃗
⃗ 2 + r (ω)μr (ω)
(∇ ) A(⃗
r, ω) = 0 . (7.108)
c2
We write the vector k⃗ as follows,
kx = k sin θ cos ϕ , ky = k sin θ sin ϕ , kz = k cos θ , (7.109a)
k⃗ = kx êx + ky êy + kz êz . (7.109b)
For the later discussion, it is instructive to remember that the angles θ and ϕ belong
⃗ not to the coordinate vector r⃗. We now define the two polarization vectors for
to k,
TE and TM modes,
ˆk,TE
⃗ = sin ϕ êx − cos ϕ êy , (7.110a)
ˆk,TM
⃗ = − cos θ cos ϕ êx − cos θ sin ϕ êy + sin θ êz . (7.110b)
k⃗ ⋅ ˆk,TE
⃗ = k⃗ ⋅ ˆk,TM
⃗ =0 ˆk,TM
⃗ × ˆk,TE
⃗
⃗ k∣
= k̂ = k/∣ ⃗. (7.110c)
We write the vector potential for the mode functions as
A⃗k,λ
⃗ (⃗r , t) = Ak,λ,x
⃗ (⃗
r, t) êx + Ak,λ,y
⃗ (⃗
r , t) êy + Ak,λ,z
⃗ (⃗
r, t) êz , (7.111)
where λ can be TE or TM, and
√
8
Ak,λ,x
⃗ = A0 ⃗ cos(kx x) sin(ky y) sin(kz z) e−i ω t , (7.112a)
V k,λ,x
√
8
Ak,λ,y
⃗ = A0 ⃗ sin(kx x) cos(ky y) sin(kz z) e−i ω t , (7.112b)
V k,λ,y
√
8
Ak,λ,z
⃗ = A0 ⃗ sin(kx x) sin(ky y) cos(kz z) e−i ω t . (7.112c)
V k,λ,z
Here, V = Lx Ly Lz , and A0 is a global amplitude normalized so that
∫ d r ∣A⃗k,λ
2
⃗ (⃗r, t)∣ = A20 .
3
(7.113)
V
The polarization vector for the mode with wave vector k⃗ and polarization λ is
⃗ = k,λ,x
ˆk,λ ⃗ êx + k,λ,y
⃗ êy + k,λ,z
⃗ êz ; (7.114)
it can describe a TE or TM mode, according to Eq. (7.110). The components of

the wave vectors for the discrete modes are given as follows,
π mπ nπ π mπ nπ
kx = , ky = , kz = , k⃗mn = êx + êy + êz . (7.115)
Lx Ly Lz Lx Ly Lz
where , m, n are integers. The electric field corresponding to the polarization
λ = TE is transverse, i.e., its z component vanishes, while the magnetic induction
field corresponding to the polarization λ = TM is transverse, i.e., its z component
vanishes. The boundary conditions for the electric field and the boundaries of the
rectangular cavity are fulfilled if Eq. (7.115) is fulfilled.
The transversality condition (7.101) for the vector potential is verified as follows,
√
8 ⃗
⃗ ⋅ A⃗k,λ
∇ ⃗ (⃗r ) = − A0 ⃗ ) sin(kx x) sin(ky y) sin(kz z) = 0 ,
(k ⋅ ˆk,λ (7.116)
V
which is fulfilled if k⃗ ⋅ ˆk,λ
⃗ = 0.
The wave equation (7.108), applied to the vector potential (7.112), results in the
condition
ω2 ⃗
(−k⃗2 + r (ω)μr (ω) ) A(⃗
r , ω) = 0 , (7.117)
c2
and thus in component form,
r μr ω 2
kx2 + ky2 + kz2 = , (7.118)
c2
where we suppress the ω argument of r and μr . Together with Eq. (7.115), this
leads to the “quantization condition”
B
D π 2 2
nπ 2
ω = ωmn = √
c D
E( ) +(
mπ
) +( ) ,
r μr Lx Ly Lz
= 0, 1, 2, . . . , m = 0, 1, 2, . . . , n = 0, 1, 2, . . . . (7.119)
In the rectangular waveguide, we have no continuous symmetry axis. Still, three

quantum numbers and the polarization λ suffice to characterize all possible modes.
Let us try to identify the fundamental mode. In order for the vector potential in
Eq. (7.112) to be nonvanishing, two of the “quantum numbers” , m, and n must be
nonvanishing. Otherwise, the entire vector potential in Eq. (7.112) vanishes. First,
let Ly and Lz be the longest edges of the rectangular cavity. Then, the angular
frequency of the fundamental mode is
B
D π 2 2
ω011 = √
c D
E( ) + ( π ) . (7.120)
r μr Ly Lz
Expressed differently, the fundamental mode carries the quantum numbers = 0,
m = 1 and n = 1, with frequency ω011 given in Eq. (7.120). For the fundamental
mode, we thus have kx = π/Lx = 0 and hence the azimuth angle of the k⃗ vector is
ϕ = 90○ = π/2. According to Eq. (7.110), the two polarization vectors are given by
ˆk⃗011 ,TE = êx , ˆk⃗011 ,TM = − cos θ êy + sin θ êz , (7.121a)
kz 1/Lz
cos θ = √ =√ . (7.121b)
ky2 + kz2 (1/Ly )2 + (1/Lz )2
For = 0, m = 1 and n = 1, the only nonvanishing component of A⃗ is the x component
Ak⃗011 ,TE,x ≠ 0 [see Eq. (7.112)]. One may combine Eqs. (7.115) with Eq. (7.112) and
observe that sin(kx x) = 0. Furthermore, because only ˆk⃗011 ,TE has a nonvanishing
x component, the fundamental mode is TE if the longest edges are Ly and Lz .
Second, let Lx and Ly be the longest edges of the rectangular cavity. In this
case, the angular frequency of the fundamental mode is
B
D π 2 2
ω110 = √
c D
E( ) + ( π ) . (7.122)
r μr Lx Ly
In this case, the polar angle of the k⃗ vector is θ = 90○ . According to Eq. (7.110),
the two relevant polarization vectors are
ˆk⃗110 ,TE = sin ϕ êx − cos ϕ êy , êk⃗110 ,TM = êz , (7.123a)
kx 1/Lx
cos ϕ = √ =√ . (7.123b)
kx + ky
2 2 (1/Lx)2 + (1/Ly )2
For = 1, m = 1 and n = 0, the only nonvanishing component of A⃗ is the z component
Ak⃗110 ,TE,z ≠ 0. Furthermore, because only ˆk⃗110 ,TM has a nonvanishing z component,
the fundamental mode is a TM mode if the shortest edge is Lz .
Let us try to verify once more, the explicit fulfillment of the boundary conditions
by the electric and magnetic fields generated by the vector potentials indicated in
Eq. (7.112). We recall that, in view of Eq. (7.102), we have E ⃗ = −∂ A/∂t
⃗ ⃗ The
= iω A.
transversal (normal) component of the electric field does not need to vanish at the
boundaries. Indeed we have, e.g., for the planes with z = 0 or at z = Lz ,
√
8
Ek,λ,z
⃗ (x, y, L = 0) = Ek,λ,z
⃗ (x, y, L = Lz ) = −iω A0 ⃗ sin(kx x) sin(ky y) .
V k,λ,z
(7.124)
However, the mode functions (7.112) are such that there is a cosine function if
the Cartesian component of the wave vector in the argument of the trigonometric
function is the same Cartesian component as the vector potential component itself.
The presence of the sine functions implies that the tangential component of the
vector potential (the component lying inside the boundary planes) vanishes for all
three boundaries,
Ak,λ,x
⃗ (x, y, z =0)= Ak,λ,x
⃗ (x, y, z =Lz )=Ak,λ,x
⃗ (x, y =0, z)=Ak,λ,x
⃗ (x, y =Ly , z)=0 ,
(7.125a)
Ak,λ,y
⃗ (x, y, z =0)= Ak,λ,y
⃗ (x, y, z =Lz )=Ak,λ,y
⃗ (x=0, y, z)=Ak,λ,y
⃗ (x=Lx , y, z)=0 ,
(7.125b)
Ak,λ,z
⃗ (x=0, y, z)= Ak,λ,z
⃗ (x=L, y, z)=Ak,λ,z
⃗ (x, y =0, z)=Ak,λ,z
⃗ (x, y =Ly , z)=0 .
(7.125c)
⃗ ⃗ ⃗
In view of E = −∂ A/∂t = iω A, the conditions (7.125) are exactly equivalent to the
boundary condition for the electric field, given in Eq. (7.25), namely, E ⃗∥,S = 0.
The next step is to look at the B ⃗ field. An explicit calculation of the curl of the
vector potential given in Eq. (7.112) leads to the following result for B ⃗=∇ ⃗
⃗ × A,
√
8
Bk,λ,x
⃗ = A0 (k,λ,z
⃗ ky − k,λ,y
⃗ kz ) sin(kx x) cos(ky y) cos(kz z) e−i ω t , (7.126a)
V
√
8
Bk,λ,y
⃗ = A0 (k,λ,x
⃗ kz − k,λ,z
⃗ kx ) cos(kx x) sin(ky y) cos(kz z) e−i ω t , (7.126b)
V
√
8
Bk,λ,z
⃗ = A0 (k,λ,y
⃗ kx − k,λ,x
⃗ ky ) cos(kx x) cos(ky y) sin(kz z) e−i ω t . (7.126c)
V
Here, the sine term depends on the same coordinate as the component of the B ⃗
field; it means that the component in the direction of the normal to the boundary
surface vanishes on the boundaries, fulfilling Eq. (7.27).
In our considerations leading up to Eqs. (7.121) and (7.123), we considered
the fundamental mode for the cases of the shortest edge length being in the x
and z directions. We found that we have a special situation where one of the mode
functions given in Eq. (7.112) vanishes, and we only have one polarization available.
In this case, one of the quantum numbers , m, n is zero. Let us try to generalize
the consideration somewhat. If either , m, n is zero, then according to Eqs. (7.115)
and (7.112), only one of the components of the vector potential A⃗ “survives”; the
others vanish. The only valid polarization vector points into the same Cartesian
direction as the vanishing component of k. ⃗ If = 0, then we need to have ˆ⃗ = êx ,
k,λ
⃗ = êz . This can be verified on the examples given in Eqs. (7.121)
and if n = 0, then ˆk,λ
and (7.123) and generalizes to any mode with one vanishing quantum number.
7.4 Exercises
● Exercise 7.1: Reproduce the derivation from Eq. (7.1) to (7.24) in your own
words, conceivably letting r = μr = 1, but being aware of the fact that ω ≠ c k for
waves in the waveguide.
● Exercise 7.2: Start from the dispersion relation (7.44) for electromagnetic waves
in a rectangular waveguide,
√
r μr √ 2
k= ω − ωmn2 , (7.127)
c
1/2
πc m 2 n 2
ωmn = √ [( ) + ( ) ] . (7.128)
r μr a b
For each mode, the kmn varies with frequency ω > ωmn . The ωmn is the cutoff
frequency for the mode. Question: How can you generate “slow light” (as far as
the group velocity is concerned) in a waveguide? For which frequencies ω does the
group velocity become zero?
● Exercise 7.3: Tabulate the minimum frequencies ωmn for the TE and TM modes
of a rectangular waveguide of dimension a = 2.3 cm and b = 1.7 cm. Perform the same
task for the TE and TM modes with m ≤ 2 and n ≤ 2. Then, plot the dispersion
relation for a rectangular waveguide and try to reproduce the findings of Figs. 7.6–
7.8 regarding the small-frequency and large-frequency asymptotics. Give the results
to an accuracy of at least 6 decimals.
● Exercise 7.4: Tabulate the first few energy eigenvalues Emnp = h ̵ ωmnp in the
spectrum of a perfectly conducting cylindrical cavity with R = 2.3 cm and height
d = 1.9 cm. Consider the cases m = 0, 1, n = 1, 2, 3, and p = 0, 1, 2. Give the results
to an accuracy of at least 6 decimals.
● Exercise 7.5: Try to devise dimensions of a cylindrical cavity adapted to generate
radiation whose frequency is equal to that of X band radar at 7175 MHz.
● Exercise 7.6: We have determined the waves of a rectangular waveguide, as well
as the modes of cylindrical and rectangular cavities. Try to determine the waves of
a cylindrical waveguide. Remember that you only need to devise an ansatz for the
z component of the fields; the rest follows.
Hint: You might use the following ansatz for the z component of the electric
field, for TM modes,
xmn
Ez = E0 Jm ( ρ) eimϕ e−iωt , (7.129)
R
where xmn is defined in Eq. (2.158). Note that Ez must vanish on the boundaries
of the cylindrical waveguide. For TE modes, you might choose
xmn
B z = B 0 Jm ( ρ) eimϕ e−iωt , (7.130)
R
where xmn is defined in Eq. (2.159). Note that Bz must vanish on the boundaries
of the cylindrical waveguide.
● Exercise 7.7: Show that (7.101) fulfills both Coulomb gauge as well as Lorenz
gauge conditions.
● Exercise 7.8: Convince yourself that you can write
⃗ ∥ Jm (γp ρ) = êρ Jm
∇ ′
(γp ρ) . (7.131)
Also, show that

⃗ ∥ Jm (γp ρ) = (êz × êρ ) Jm
êz × ∇ ′ ′
(γp ρ) = êϕ Jm (γp ρ) (7.132)
and same for γp → γ p . With these results at hand, try to decompose the formulas
given in Eqs. (7.76) and (7.99) into components parallel to êρ and êϕ .
● Exercise 7.9: Adjust the constant A0 in Eq. (7.113) so that the field energy
̵ in
stored in the fundamental mode is equal to the energy of a single photon, hω,
contrast to the normalization condition (7.113).
Chapter 8
Advanced Topics
8.1 Overview
In the current chapter, we shall attempt to cover a few advanced topics, such as
Lorentz transformations of the electromagnetic fields and the transition to quantum
field theory. It is instructive to put this endeavor into context.
When the Maxwell equations were discovered, and the formalism of (special)
relativity was developed, it was observed that the Maxwell theory was in fact com-
patible with the formalism of special relativity. Thus, it was not necessary to add
any “relativistic corrections” to these equations, as opposed to the equations of clas-
sical mechanics, which require numerous relativistic correction terms. How exactly
do the electromagnetic fields transform under the action of the Lorentz group?
How do we transform, e.g., the Ampere–Maxwell law into a moving coordinate
system?
We shall try to approach this problem carefully, by first recalling a few
basic facts about Lorentz transformations, then developing some basic aspects of
the representation theory of the Lorentz group, before applying this formalism to
the Maxwell theory. In the latter step, we shall assign a specific behavior to the
potentials and fields under Lorentz transformations, and understand the concept
of a four-vector, which groups certain entities into a vector with four components,
all of which transform jointly under Lorentz transformations. An example is the
⃗ composed of charge and current density, which transforms as a
four-vector (ρ, c−1 J)
joint four-vector under a Lorentz transformation, i.e., the charge density transforms
into the current density and vice versa. This endeavor will comprise Chaps. 8.2
and 8.3.
Up to this current chapter, all subjects treated in this book required only the
classical theory as input. In Sec. 8.4, we shall attempt to “build a bridge” toward
the quantum theory of the electromagnetic field, known as quantum electrodynam-
ics (QED). In principle, quantum electrodynamics is a relativistic quantum field
theory. Here, the word “relativistic” means that the field operators, and all the
couplings, are written down such as to be compatible with Lorentz invariance [20].
In Sec. 8.2, we shall see that Lorentz and spinor indices qualify themselves as
295
indices which transform under specific representations of the Lorentz group. QED
describes processes in which virtual quanta are created and annihilated, and par-
ticles (and antiparticles) are annihilated and created. Furthermore, the important
entities (scalar and vector potentials, and fields) are functions of space and time.
By contrast, classical electrodynamics is a field theory because all potentials (scalar
and vector) as well as the fields are functions of space-time, but classical electrody-
namics does not constitute a quantum theory, where quanta of the electromagnetic
radiation (photons) are created and annihilated.
In Sec. 8.4, we attempt to approach the concept of a quantum field theory by
treating the field quantization. We shall see that it is possible to calculate the
Casimir attractive force between plates in a semi-quantized formalism where we
simply assign a nonvanishing zero-point energy to all harmonic modes of the elec-
tromagnetic field. This formalism avoids a full quantization of the electromagnetic
field as well as any mention of relativity while allowing us to study a truly quan-
tum effect, namely, the attractive force between perfectly conducting plates due
to the shift in the zero-point energy of the modes in the space in between them.
Specifically, some otherwise available electromagnetic oscillation modes are missing
from the spectrum of available oscillator modes under the influence of the boundary
conditions; these missing modes generate a “lack of zero-point energy” which leads
to an attractive force. The closer the two plates are, the greater is the influence of
the quantization, and the more “zero-point energy is missing.” Hence, the system
can lower its energy by moving the plates closer, and this generates the Casimir
force.
Two of the most important concepts in quantum field theory are the regular-
ization and renormalization. We shall discuss these in Sec. 8.4, in a “nutshell’.
So-called loop integrals in quantum field theory require regularization and renor-
malization in higher orders. An aura of mystery and paradox still overshines these
concepts. It is less well known that regularization and renormalization can also be
applied with advantage, to the calculation of classical, electrostatic potentials (see
Sec. 8.5). Divergences typically result when the expression 1/∣⃗ r − r⃗′ ∣ is integrated
′
over source configurations [charge densities ρ(⃗ r )] with an infinite extent. In this
case, the naturally occurring divergences can be regularized, and the potential can
be renormalized, using concepts borrowed from field theory. The calculation gives
us an opportunity to introduce dimensional regularization, which is an important
concept in modern field theory, and the concept of a renormalization condition. Af-
ter these preparations, a reader of this book should be well prepared to study further
literature on quantum electrodynamics, or indeed, quantum field theory [20].
Finally, in Sec. 8.6, we study open problems in the field, such as the complete
understanding of the radiative reaction problem, which is still a subject of active
research, and the connection of relativistic electrodynamics and general relativity.
In the latter endeavor, we need to generalize the gradient vector, and the Laplacian,
to their respective forms relevant to curved space-times.
Advanced Topics 297
8.2 Lorentz Transformations, Generators and Matrices
8.2.1 Lorentz Boosts

We shall begin the endeavor by first recalling a few basic facts about Lorentz trans-
formations. A Lorentz boost is a Lorentz transformation which describes the trans-
formation of time and space into moving frames of reference. When combined with
rotations of space, Lorentz boosts describe all possible coordinate transformations of
space and time compatible with special relativity. In SI mksA units, the most basic
Lorentz transformation for the transformation from the unprimed frame (laboratory
frame) into the moving (primed) frame reads as
v
t′ = γ t − 2 γ x , x′ = γ x − v γ t , y′ = y , z′ = z . (8.1)
c
Here, we assume that the moving frame has a uniform velocity v in the positive x
direction (so-called Lorentz boost in the x direction). The Lorentz factor is
1
γ=√ . (8.2)
1 − v 2 /c2
The transformation goes from the laboratory frame to the moving frame; the primed
quantities are those measured in the moving frame. It is customary to define the
four-vector (c t, x, y, z) of space-time coordinates so that all entries have the same
physical dimension (corresponding to a spatial coordinate), i.e.,
v v
ct′ = γ ct − (γ ) x , x′ = γ x − (γ ) (ct) , y′ = y , z′ = z . (8.3)
c c
It is interesting to verify that the backtransformation from the moving to the rest
frame is achieved by replacing v → −v. That means that
v v
ct = γ ct′ + (γ ) x′ , x = γ x′ + (γ ) ct′ , y = y′ , z = z′ . (8.4)
c c
We shall now generalize the formalism to a Lorentz boost into a frame that moves
with velocity v⃗,
v v⃗
ct′ = γ ct − (γ ) (v̂ ⋅ r⃗) = γ (ct − ⋅ r⃗) , (8.5a)
c c
⃗
v ⋅ ⃗
r v
r⃗′ = r⃗ + (γ − 1) 2 v⃗ − (γ ) v̂ (ct) , (8.5b)
v⃗ c
where v̂ = v⃗/v. The quantity (ct′ )2 − r⃗′ 2 = (ct)2 − r⃗ 2 is invariant under all Lorentz
boosts.
The transformation property (8.5) describes how time and space transform into
each other when we change the frame of reference. It is based on the fact that time
and space are not independent physical quantities under Lorentz transformations,
but in fact, transform into each other as we change the frame of reference. Indeed,
(ct, r⃗) constitutes a so-called four-vector. Just as much as the quantities (c t, r⃗) con-
stitute a four-vector, there is also a four-vector composed of energy and momentum
(E, c p⃗). The transformation properties of all Lorentz four-vectors have to be the
same; otherwise we could not call them Lorentz vectors. We thus conclude that
v
E ′ = γ E − (γ ) v̂ ⋅ (c⃗
p) = γ (E − v⃗ ⋅ p⃗) , (8.6a)
c
v⃗ ⋅ (c⃗ p) v
p′ = c⃗
c⃗ p + (γ − 1) v⃗ − (γ ) v̂ E. (8.6b)
v⃗ 2 c
For a massless quantum particle, we can relate the energy to the angular frequency
̵
(E = hω) and the momentum to the wave vector (p = h ̵ k). Quite generally, the
⃗
entities (ω, c k) constitute a four-vector for any electromagnetic or massless matter
wave, and so we have
v ⃗ = γ (ω − v⃗ ⋅ k)
⃗ ,
ω ′ = γ ω − (γ ) v̂ ⋅ (ck) (8.7a)
c
v⃗ ⋅ (ck) ⃗ v
c k⃗′ = c k⃗ + (γ − 1) v⃗ − (γ ) v̂ ω . (8.7b)
v⃗ 2 c
We are now in the position to investigate the consequences of Lorentz transforma-
tions regarding basic phenomenology of special relativity.
8.2.2 Time Dilation and Lorentz Contraction

Time dilation and Lorentz contraction are basic relativistic phenomena which need
to be discussed. The derivation is somewhat nontrivial because one has to carefully
distinguish between the two observers, the moving observer and the one at rest.
The question is who sees which event at which time.
Let us start with time dilation. We investigate one event which is at the origin
in both the moving as well as the rest frame,
x1 = 0 , t1 = 0 , x′1 = 0 , t′1 = 0 . (8.8)
Now, let us assume that the particle moves at the same speed as the moving frame,
namely v, and therefore remains at rest in the moving frame. So, if we look some
time later (time t′2 ), the position in the moving frame is x′2 = x′1 = 0. How much
time τ has elapsed for the moving observer? Well, in the rest frame, the particle
is at position x2 = v t at time t, and the second event has the following space-time
coordinates,
x2 = v t , t2 = t , x′2 = 0 , t′2 = τ . (8.9)
For consistency, we verify that the relation
x′2 = γ x2 − v γ t2 = γ v t − v γ t = 0 (8.10)
is trivially fulfilled. The time t′2 , which is the proper time elapsed in the moving
coordinate system, can be expressed in terms of the unprimed quantities t2 and x2 ,
by virtue of the Lorentz transformation,
√
′ v v v2 v2 1
t2 = τ = γ t2 − 2 γ x2 = γ t − 2 γ v t = γ (1 − 2 ) t = 1 − 2 t = t . (8.11)
c c c c γ
Advanced Topics 299
Note that the equation

1
τ=
t≤t (8.12)
γ
implies that the time that has passed in the frame of the moving observer is less
than or equal to the time that has elapsed in the rest frame (τ ≤ t). This is called
time dilation and experimentally confirmed, e.g., by the enhancement of the muon
lifetime at accelerator storage rings. The time τ is also called the proper time that
elapses in the rest frame of the moving particle.
Length contraction works as follows. We first have to carefully consider what a
length measurement actually means. It means that we have two events, one, where
we measure the position of the left end of a moving slab, and another one, where
the position of the right end of a moving slab is measured. We assume that the first
event (left end of moving slab) happens at the following space-time coordinates,
x1 = x< , t1 = 0 , x′1 = x′< , t′1 = 0 . (8.13)
The measurement of the right end of the slab happens at the same time t2 = 0 in the
rest frame. It results in a position measurement at x′2 = x′> which is not necesarily
happening at t′2 = 0. We have
v v
x2 = x> , t2 = 0 , x′2 = x′> , t′2 = γ t2 − 2 γ x2 = − 2 γ x> . (8.14)
c c
Then,
x′> − x′< = x′2 − x′1 = γ (x2 − x1 ) − v γ (t2 − t1 ) = γ (x> − x< ) , (8.15)
or
1 ′ 1
Δx = x> − x< = (x> − x′< ) = Δx′ ≤ Δx′ . (8.16)
γ γ
The length Δx seen in the rest frame is smaller or equal to the length Δx′ in the
moving frame; this is the result of Lorentz contraction.
8.2.3 Addition Theorem

From nonrelativistic intuition, we would say that if a particle moves with velocity
u in the positive x direction, then the velocity will be measured as u′ = u − v from
the point of view of an observer who moves with uniform velocity v in the positive
x direction. One simply subtracts the relative velocity v. However, this is not the
case anymore if we consider special relativity. The particle velocities are u = dx/dt
and u′ = dx′ /dt′ , respectively, and the Lorentz transformation tells us how the
infinitesimal changes in the positions of the particles transform into each other. We
have, according to Eq. (8.3),
v
t′ = γ t − 2 γ t , x′ = γ x − v γ t , y′ = y , z′ = z . (8.17)
c
The differential form of the Lorentz transformation is
v
dx′ = γ dx − v γ dt , dt′ = γ dt − 2 γ dx . (8.18)
c
The velocity addition theorem follows quite immediately, dividing dx′ by dt′ ,
dx′ γ dx − v γ dt γu−γv u−v
u′ =
′
= v = v = uv . (8.19)
dt γ dt − 2 γ dx γ − 2 γ u 1 − 2
c c c
The reverse transformation is obtained by inverting v → −v, as with the ordinary
Lorentz transformation (8.4),
u′ + v
u= . (8.20)
u′ v
1+ 2
c
From this representation, we see that the magnitude of u can never exceed that of
c if the velocity in the moving frame fulfills −c < u′ < c. This is because u grows
monotonically with u′ and v, and it reaches its terminal velocity u = c when either
u′ or v (or both) reach a magnitude of c. Lorentz transformations single out the
velocity of light as a limiting velocity.
When the particle velocity u⃗ in the rest frame and the velocity of the moving
⃗′ can be calculated as
frame v⃗ are not collinear, then the vector-valued velocity u
v⃗ ⋅ d⃗x v
x + (γ − 1)
d⃗ v⃗ − (γ ) v⃗ (cdt)
x′
d⃗ v⃗ 2 c
⃗′ =
u = (8.21)
dt v
γ c dt − (γ ) (v̂ ⋅ d⃗ x)
c
v⃗ ⋅ u⃗ v⃗ ⋅ u
⃗
u⃗ + (γ − 1) 2 v⃗ − γ v⃗ u ⃗ + (γ − 1) 2 v⃗ − γ⃗
v
= v⃗ = v . (8.22)
u⃗ ⋅ v⃗ v⃗ ⋅ u
⃗
γ −γ 2 γ (1 − 2 )
c c
If u⃗ is parallel to v⃗, then we can replace v⃗ ⋅ u⃗/c2 → 1 and have
v⃗ ⋅ u⃗
⃗ + (γ − 1)
u v⃗ − γ⃗
v u⃗ ∥ v⃗ γ (⃗
u − v⃗) u⃗ − v⃗
⃗ =
u ′ v2 → =
v⃗ ⋅ u
⃗ v⃗ ⋅ u⃗ vu , (8.23)
γ (1 − 2 ) γ (1 − 2 ) 1 − c2
c c
which is the same as the vector-valued version of the addition theorem (8.19).
8.2.4 Generators of the Lorentz Group

Up to now, we have written the Lorentz transformations of various quantities in
terms of separate transformations for the “time-like” and “space-like” components.
The transformations are linear, which calls for a matrix representation. It is cus-
tomary to summarize the Lorentz vectors (four-vectors) which we have encountered
so far, into an explicit vector form, with components being labeled as μ = 0, 1, 2, 3,
and μ = 0 being reserved for the “time-like” component. This will eventually lead
to a unification of the transformation properties of time and space under a common
matrix representation. For the space-time position vector, we would write
xμ = (c t, r⃗) , x0 = c t , x1 = x , x2 = y , x3 = z . (8.24)
Advanced Topics 301
In the physics literature, one has almost universally adopted a notation with μ =
0, 1, 2, 3, which corresponds to “C conventions” instead of “Fortran conventions” for
the labeling of the vector or array structure (we recall that array indices canonically
start from zero for programs written in C, while they start from one in the case of
a Fortran program). In the natural unit system, one would set h ̵ = c = 0 = 1 and
then the components xμ of the relativistically covariant four-vector vector would
be the same as those of the spatial vector xi . Here, because we use SI mksA
units, we need to supply an extra factor c in x0 . The convention is that xμ is
understood as a four-component object which, if μ (Greek index) attains the value
μ = 0, is equal to c t, while for μ = i (Latin index), we have xi = (⃗ r )i , equal to
the ith component of the position vector r⃗. In the following, we adopt the general
convention that Greek superscripts and subscripts denote a Lorentz index that
assumes the values μ = 0, 1, 2, 3, whereas Latin superscripts and subscripts denotes
spatial indices i = 1, 2, 3. The distinction between contravariant (upper index) and
covariant (lower index) components in relativity becomes necessary because of the
presence of a nontrivial metric, to be discussed below.
The four-vectors we have encountered so far are
xμ = (c t, r⃗) , P μ = (E, c⃗
p) , ⃗ ,
Kμ = (ω, ck)
⃗ ,
Aμ = (Φ, cA) ⃗ .
J μ = (ρ, c−1 J) (8.25)
We use calligraphic symbols in order to denote those four-vectors where we introduce

a factor c in the spatial components; for the vector potential, the spatial components
are thus Ai = c Ai . In a unit system with c = 1, such as the natural unit system,
one may universally denote the four-vectors and its spatial components by the same
symbol, but in the SI mksA unit system, this would lead to an ambiguity for the
spatial components Ai = Ai /c.
Under Lorentz transformations (with matrix representation Λμ ν ), the four-
vectors transform as follows,
x′μ = Λμ ν xν , p′μ = Λμ ν pν , k ′μ = Λμ ν k ν , (8.26)
etc., where the sum over ν = 0, 1, 2, 3 is implicitly assumed by the Einstein sum-
mation convention, and Λ is the Lorentz transformation. We now have to find
the explicit matrix representation of Λ. The Lorentz transformation (8.3) can be
written in matrix form,
′
⎞ ⎛ γ −γ c
v
⎛ct 0 0⎞ ⎛ct⎞
⎜ x′ ⎟ ⎜ −γ v
γ 0 0⎟ ⎜ ⎟
⎜ ⎟=⎜ c ⎟⎜x⎟.
⎜ y′ ⎟ ⎜ 0 0⎟ ⎜
⎟ ⎟ (8.27)
⎜ ⎟ ⎜ 0 1 ⎜y ⎟
⎝ z′ ⎠ ⎝ 0 0 0 1⎠ ⎝ z ⎠
The Lorentz boost can be interpreted as a hyperbolic rotation,
v
γ = cosh ρ , γ = sinh ρ , cosh2 ρ − sinh2 ρ = 1 , (8.28)
c
where ρ is the rapidity. We can convince ourselves by explicit calculation that
(ct′ )2 − x′2 − y ′2 − z ′2 = [(ct) cosh(ρ) − x sinh(ρ)]

2
− [x cosh(ρ) − (ct) sinh(ρ)] − y ′2 − z ′2 = (ct)2 − x2 − y 2 − z 2 .

2
(8.29)
This amounts to the conservation of a scalar product with a negative weight being
assigned to the spatial components, i.e., with regard to a metric with the signature
(+, −, −, −). If we define the matrix representation of the elementary Lorentz boost
Λμ ν and of the metric gμν , as follows,
⎛ cosh ρ − sinh ρ 0 0⎞ ⎛1 0 0 0 ⎞
⎜ − sinh ρ cosh ρ 0 0⎟ ⎜ 0 −1 0 0 ⎟
(Λμ ν ) = ⎜
⎜ 0
⎟, (gμν ) = (g μν ) = ⎜ ⎟,
1 0⎟ ⎜ 0 0 −1 0 ⎟
(8.30)
⎜ 0 ⎟ ⎜ ⎟
⎝ 0 0 0 1⎠ ⎝0 0 0 −1 ⎠
then the following quantity (“scalar product”) remains invariant under Lorentz
boosts,
(c t′ )2 − r⃗′2 = gμν x′μ x′ν = gμν Λμ ρ xρ Λν δ xδ = gρδ xρ xδ = (c t)2 − r⃗ 2 . (8.31)
Because xρ and xδ are arbitrary, we can establish that

μ
gμν Λμ ρ Λν δ = (ΛT )ρ gμν Λν δ = gρδ , g = ΛT g Λ . (8.32)
The latter equation expresses the defining property of a Lorentz transformation in

matrix form. One can show that any matrix Λ that fulfills Eq. (8.32) actually is a
valid Lorentz transformation, composed of a rotation of space followed by a Lorentz
boost. Because the signature of the conserved metric is (+, −, −, −), we speak of the
group SO(1, 3). Furthermore, the connected component of the Lorentz group (the
one with determinant equal to +1 instead of −1) is all composed of Lorentz boosts
and spatial rotations.
If aμ is a Lorentz vector of contravariant components, then the vector aμ of
covariant components transforms as follows under a Lorentz transformation,
ν
aμ = gμν aν , a′μ = gμρ a′ρ = gμρ Λρ δ aδ = gμρ Λρ δ g δν aν = ((Λ−1 )T )μ aν , (8.33)
where we use Eq. (8.32). So, the vector aμ is transformed with the transpose of
the inverse Lorentz matrix. For pure rotation matrices, the transpose is equal to
the inverse, so that the distinction between covariant and contravariant components
becomes unnecessary. Under Lorentz transformations, the scalar product of any of
the quantities listed in Eq. (8.25) is a scalar, i.e., if aμ and bμ are two vectors, then
a′ ⋅ b′ = a′μ b′μ = gμν a′μ b′ν = gμν Λμ ρ Λμ δ aρ bδ = gρδ aρ bδ = aμ bμ = a ⋅ b . (8.34)
So, the scalar product a ⋅ b transforms under the elementary scalar representation of
the group SO(1, 3); it is a Lorentz scalar and a conserved quantity under Lorentz
transformations.
Advanced Topics 303
8.2.5 Representations of the Lorentz Group

We have seen that the scalar product aμ bμ of two space-time vectors aμ and
bμ is conserved under Lorentz boosts, which describe the transformation from
a rest frame into a moving frame, without any spatial rotation. We may gen-
eralize this concept and define the Lorentz transformations as the group of lin-
ear space-time transformation that leave the scalar product invariant, i.e., as the
group SO(1, 3). From the above discussion, it is evident that the identity trans-
formation Λμ ν = δ μ ν is a member of the Lorentz group, with unit determinant.
The parity transformation changes the sign of the spatial coordinates and has the
matrix representation diag(1, −1, −1, −1). Its determinant is equal to −1. It does not
belong to the connected component of the unit matrix. The connected component
of the identity transformation is called the proper Lorentz group. Furthermore, we
would like the direction of time to be preserved under the Lorentz transformation,
which defines the orthochronous Lorentz group. In physics, we are thus primarily
interested in the proper orthochronous subgroup of the Lorentz group, although
the important discrete transformations of time inversion [time inversion carries the
matrix representation diag(−1, 1, 1, 1)] and parity also play an important role.
The space-time representation of the Lorentz group defined above plays an egre-
gious role in physics. Indeed, proper orthochronous Lorentz transformations are
products of spatial rotations and Lorentz boosts. Let Λ1 and Λ2 be two Lorentz
transformations. We find a suitable representation D of the Lorentz group if
D(Λ1 Λ2 ) = D(Λ1 ) D(Λ2 ) . (8.35)
A trivial representation of the Lorentz group is obtained when we simply put

D(Λ1 ) = . In the following, starting from spatial rotations, we shall develop
the concept of a generator of spatial rotations and Lorentz transformations, as well
as the defining structure equations of the Lie group which generates the Lorentz
transformations upon exponentiation. Having obtained the structure constants of
the Lie group, we shall indicate a second representation of the Lorentz group, based
on Dirac matrices, which describes spin-1/2 particles. However, before we do so, we
need to go back once more and investigate the orbital and spin angular momenta,
previously introduced in Secs. 2.3.2 and 5.4.2, in a different light.
Let us start with the matrix representations of a rotation of two-dimensional
space by a rotation angle θ. In the following, we shall denote a matrix by a double-
lined symbol such as
for a rotation matrix or for the matrix representation of
a Lorentz transformation. The two-dimensional rotation matrix reads as
=(
cos ϕ sin ϕ
− sin ϕ cos ϕ
)≈(
10
01
)+(
0 ϕ
−ϕ 0
). (8.36)
For a small rotation angle ϕ, we have

x′ x 0 1 x x + ϕ δy
( )≈( )+ϕ×( )⋅( ) = ( ), δxi = ϕ ij xj , (8.37)
y′ y −1 0 y y − ϕ δx
,-- - - - - .-- - - - - /
Å
=
where 12 = −21 = 1 is an antisymmetric tensor, and the Einstein summation con-
vention has been used. Geometrically, Eq. (8.37) finds a natural interpretation as
a rotation by an angle −ϕ, which corresponds to the “passive” interpretation of
the rotation, where the coordinates are rotated by the angle −ϕ, so that the trans-
formed and original coordinates, under a transformation of the coordinate system
by an angle +ϕ, represent the same vector in real space.
Throughout this derivation, we denote the Cartesian components of the coordi-
nate vector r⃗ as xi , not as ri , so that the Cartesian representation reads as
r⃗ = ê1 x1 + ê2 x2 + ê3 x3 = ê1 x + ê2 y + ê3 z . (8.38)
The notation is not completely free of small idiosyncrasies; the problem with the
notation ri for the Cartesian components is that the symbol r = ∣⃗ r ∣ also is used for
i
the modulus of the coordinate vector. Thus, r could very well be misunderstood
as the ith power of ∣⃗
r ∣, whereas the risk of a misunderstanding is somewhat reduced
for the Cartesian component pi , which is hard to misunderstand as ∣⃗ p∣i , for reasons
that have to do with general human intuition and cannot easily be quantified. In
any case, Eq. (8.38) defines our notation uniquely.
A scalar function transforms as f ′ (⃗r′ ) = f (⃗
r) where r⃗′ = ⋅ r⃗. Here,
= exp(ϕ ) ≈ 1 + ϕ , −1 = exp(−ϕ ) ≈ 1 − ϕ , (8.39)
for infinitesimal ϕ. Then, at the same coordinate r⃗ as for the original argument,
but with the transformed function f ′ , we have
f ′ (⃗ −1 ⋅ r⃗) = f (⃗r − ϕ ⋅ r⃗) = f (ri − ϕ ij rj )
r) = f ( (8.40)
∂ ∂
= f (⃗
r) + ϕ ij xi j
f (⃗
r) = f (⃗
r) + i ϕ (−i ij xi j ) f (⃗
r) = f (⃗
r) + i ϕ L̂ f (⃗
r) .
∂x ∂x
We have used the antisymmetry of the tensor and have defined the operator L̂,
∂
L̂ = −i ij xi . (8.41)
∂xj
The angular momentum operator L̂ only has one component in two dimensions.
Let us define, as a generalization of the matrix
given above, three matrices
k
with k = 1, 2, 3, which are given as follows (in component form) [see Eq. (5.138)]
(k )ij = −ikij , 123 = 1 , (8.42)
where ijk is antisymmetric upon interchange of any two arguments. These matrices
are identical to the ones indicated in Eq. (5.139), the only difference being the
different convention for the Cartesian coordinate indices which are indicated as
superscripts rather than subscripts. The matrix
given above is identified as the

upper right 2 × 2 submatrix of . The matrices and the corresponding components
3
of the angular momentum vector fulfill the algebraic relations

[ i , j ] = iijk k , [Li , Lj ] = i ijk Lk , (8.43)
Advanced Topics 305
the first of which was already mentioned in Eq. (5.140). The tensor both enters the

explicit form of the matrices but also the structure constants of the Lie algebra.
The rotation matrix is given as
= exp (i ∑ ϕk k ) = exp (i ϕ⃗ ⋅ ⃗ ) , ⋅ r⃗ ,
3
r⃗′ = (8.44)
k=1

where ⃗ is the vector of matrices, and ϕ⃗ is a vector of rotation angles about the
three axes. Then,
ij
[exp (iϕ⃗ ⋅ ⃗ )] ≈ [ + i ϕ⃗ ⋅ ⃗ ] ≈ δ ij + ϕk kij .
ij
(8.45)
For small rotation angles ϕ⃗ = n̂ δϕ, the components of a vector v⃗ thus transform as
follows,

[exp (iϕ⃗ ⋅ ⃗ )]
ij
v j ≈ (δ ij + ϕk kij ) v j = v i − (ϕ⃗ × v⃗) ,
i
(8.46)
which is just the representation of an infinitesimal rotation by an angle ϕ, ⃗ in the
case of a passive interpretation of the rotation. That means that the coordinate
system turns by an angle ϕ,⃗ and the coordinate vector r⃗ → r⃗′ needs to turn by an
⃗ with the net result that r⃗′ and r⃗ represent the same vector in real space.
angle −ϕ,
Let us briefly investigate if the transformation v⃗ → v⃗′ = v⃗ − (ϕ⃗ × v⃗) under
i
infinitesimal rotations is compatible with the transformation of a vector product.

Indeed, the vector product w⃗=u ⃗ × v⃗ transforms into w ⃗′ = u⃗′ × v⃗′ with
⃗′ = u⃗′ × v⃗′ = (⃗
w u − ϕ⃗ × u
⃗) × (⃗
v − ϕ⃗ × v⃗)
≈u
⃗ × v⃗ − (ϕ⃗ × u
⃗) × v⃗ − u
⃗ × (ϕ⃗ × v⃗) = u⃗ × v⃗ − ϕ⃗ × (⃗
u × v⃗) = w
⃗ − ϕ⃗ × w⃗,
(8.47)
where we expand to first order in ϕ, confirming the consistency. We have used the
identity
(ϕ⃗ × u
⃗) × v⃗ + u
⃗ × (ϕ⃗ × v⃗) = ϕ⃗ × (⃗u × v⃗) . (8.48)
The generalization of (8.40) to three-dimensional rotations is
f ′ (⃗ −1 ⋅ r⃗) = f (xi − ϕk kij xj ) ≈ f (⃗r) − ϕk kij xj ∂x∂ i f (⃗r)

r) ≈ f (
∂ ⃗ f (⃗
= f (⃗
r) + i ϕk (−i kij xi
) f (⃗
r) = f (⃗
r ) + i ϕ⃗ ⋅ L r) , (8.49)
∂xj
where we take notice of the interchange i ↔ j in the third step of the transformation.
The angular momentum operator L ⃗ is defined in Eq. (2.29). For finite rotations, we
summarize the transformation law of a vector r⃗ and of a scalar function as
⃗ ⋅ r⃗ ,
r⃗′ = exp(iϕ⃗ ⋅ S) f ′ (⃗ ⃗ f (⃗
r ) = exp(i ϕ⃗ ⋅ L) r) . (8.50)
The question now is how to generalize the above formalism to Lorentz transfor-
mations. The answer can be given as follows. As mentioned above, all connected or
orthochronous Lorentz transformations can be generated by rotations and boosts.
In analogy to rotations in quantum mechanics, we want to investigate the form
of infinitesimal Lorentz transformations. According to Eq. (8.50), the generators
of spatial rotations are the components of the angular momentum operator [see
Eqs. (2.29) and (8.41)]. If we are to generalize the spatial rotations to rotations of
space-time, then it becomes clear that we need to find a suitable Lorentz-covariant
generalization of the angular momentum operator. This generalization is given by
generators Lμν with μ, ν = 0, 1, 2, 3, where
∂ ∂
Lμν = i (xμ ∂ ν − xν ∂ μ ) = i (xμ − xν ). (8.51)
∂xν ∂xμ
On the occasion, we define the derivatives with respect to covariant and contravari-
ant components of space-time as
∂ ∂
xμ = (c t, r⃗) , xμ = (c t, −⃗
r) , ∂μ ≡ , ∂μ ≡ . (8.52)
∂xμ ∂xμ
For spatial indices i, j, k = 1, 2, 3, the generators are connected to the angular mo-
mentum by the relation
1 ik k ∂ ∂
Li = L , Lk = −i (xk
− x k ) , (8.53)
2 ∂x ∂x
because X i is the Cartesian component, not Xi . The generalization of the ma-
trices defined in Eq. (5.138) is given by
(αβ )μν = gμα gν β − gμβ gν α . (8.54)

Here, we interpret ( αβ )μν to be the element with Lorentz indices μ, ν of the
generator matrix
αβ . That is to say, a priori we have a total of 16 = 4 × 4

matrices αβ , each of which have 16 components ( αβ )μν . However, because of

the antisymmetry αβ = − βα , only 6 matrices are actually nonvanishing and
different from each other. Both the Lorentz indices of the matrices as well as the
component indices can be lowered and raised with the metric, i.e.,
( αβ )μ ν = gμα gνβ − gμβ gνα . (8.55)
With generalized rotation angles ω = −ω , infinitesimal Lorentz transformations
αβ βα
read as
1
1
Λμ ν = g μ ν + ω αβ ( αβ )μ ν = g μ ν + (ω μ ν − ω ν μ ) = g μ ν + ω μ ν .
2 2
(8.56)
For antisymmetric rotation angles ω αβ = −ω βα , we have, in view of Eq. (8.54),

1 αβ
ω (M μν )αβ = ω μν . (8.57)
2
The matrix with elements ω μν (μ, ν = 0, 1, 2, 3) has the structure of a general, anti-
symmetric matrix with zeros on the diagonal. It has six independent nonvanishing

elements corresponding to the six generator matrices αβ . They fulfill the algebraic
relations:
[μν , κλ] = gμλ νκ + gνκ μλ − gμκ νλ − gνλ μκ , (8.58)
[Lμν , Lκλ ] = i (g μλ Lνκ + g νκ Lμλ − g μκ Lνλ − g νλ Lμκ ) . (8.59)

Advanced Topics 307
These commutation relations define the Lie algebra of the group and are funda-
mental. Exponentiating a matrix representation of the Lie group yields a matrix
representation of the Lorentz group. The generalization of Eq. (8.49) to finite
Lorentz transformations is
1
2
1

f ′ (x) = f (Λ−1 x) = f (xμ − ω αβ ( αβ )μ ν xν ) = f (x) − ω αβ ( αβ )μ ν xν ∂μ f (X)
2
= f (x) − ω α β xβ ∂α f (x) = f (x) + ω αβ xα ∂β f (x)
i i
= f (x) − ω αβ (i (xα ∂β − xα ∂β )) f (x) = f (x) − ω αβ Lαβ f (x) . (8.60)
2 2
We have the following Lorentz transformations for the four-vector X and the trans-
formed scalar function f ′ , expressed as a function of the old coordinates,
1
x′μ = exp ( ω αβ (
2
αβ )μν ) xν , i
f ′ (x) = exp (− ω αβ Lαβ ) f (x) .
2
(8.61)
Here, the expression exp [ 12 ω αβ (Mαβ )μ ν ] needs to be interpreted as follows: One

first multiplies the generalized rotation angles ω αβ with the generator matrices
Mαβ , with the components of Mαβ taken to be the Lorentz components of upper
index μ and lower index ν. One then exponentiates the resultant matrix with
components 12 ω αβ (Mαβ )μ ν and obtains a Lorentz transformation matrix which
can be multiplied with a Lorentz vector component xν , to obtain the transformed
vector component x′μ .
The phase conventions in Eq. (8.61) differ from those used in Eq. (8.50), and
are not uniform. (One can bring them into a more uniform appearance under the
replacement Lμν → iM μν .) In order to understand this problem deeper, let us
consider the specialization of a Lorentz transformation to the case of a rotation
about the z axis, i.e.,
ω 12 = −ω 21 = −ϕ , ω 13 = ω 23 = ω 0i = 0 , (8.62)
for i = 1, 2, 3. In this case,
μ μ
⎡⎛ 0 0 0 0 ⎞⎤ ⎡⎛ 0 0 0 0 ⎞⎤
⎢ ⎥ ⎢ ⎥
⎢⎜ 0 0 ϕ 0⎟ ⎥ ⎢⎜ 0 cos ϕ sin ϕ 0 ⎟⎥
X=
1 αβ
ω ( αβ )μ ν ⎢
= ⎢⎜
⎢⎜
⎟⎥
⎟⎥
−ϕ 0 0 ⎟⎥
, [exp(X)]μ ν
⎢⎜
= ⎢⎜ ⎟
⎢⎜ 0 − sin ϕ cos ϕ 0 ⎟
⎥
⎥
⎟⎥
.
2 ⎢⎜ 0 ⎥ ⎢ ⎥
⎢⎝ 0 0 0 0 ⎠⎥ ⎢⎝ 0 0 0 0 ⎠⎥
⎣ ⎦ ν ⎣ ⎦ ν
(8.63)
We consider the spatial part of the matrix X, in the xy plane. This is the submatrix
of X with μ, ν = 1, 2, corresponding to the second and third rows and columns of
the matrix X defined in Eq. (8.63), which is equal to the matrix defined in
Eq. (8.37). We conclude that the specialization of the transformation law (8.61), to
spatial generators with α, β = 1, 2, 3, leads to the same transformation law for the
spatial components of a four-vector as the transformation law (8.44), but one has
to be careful with the sign in the identification of the rotation angle.
Let us also investigate the expression

i i i
Y = − ω αβ Lαβ = − ω 12 L12 − ω 21 L21 = −iω 12 L12
2 2 2
∂ ∂
= iϕ L12 = iϕ (ix1 ∂2 − ix2 ∂1 ) = iϕ (−ix1 2 + ix2 1 ) = i ϕ Lz . (8.64)
∂x ∂x
The correspondence to the formula (8.50) is thus established; the expression
− 2i ω αβ Lαβ in Eq. (8.61) replaces iϕ⃗ ⋅ L⃗ in Eq. (8.50).
In order to complement the discussion, let us now try to find at least one further
nontrivial matrix representation of the Lorentz group. To this end, we define the
Dirac γ μ matrices as follows,
γ0 = β = (
2×2 0 ⃗
0 σ
0 − 2×2 ) , γ⃗ = (
−⃗σ0
). (8.65)
For i = 1, 2, 3, γ i has a structure with two 2 × 2 Pauli matrices on the off-diagonal.

These Dirac matrices are 4 × 4 matrices, and the Pauli σ k matrices with k = 1, 2, 3
are given as
01 0 −i 1 0
σ1 = ( ), σ2 = ( ), σ3 = ( ). (8.66)
10 i 0 0 −1
They fulfill the commutator relations
[ 12 σ i , 12 σ j ] = i ijk 12 σ k . (8.67)
These commutators look similar to those for the Cartesian components of the an-
gular momentum given in Eq. (8.43). These relations suggest that the matrices 12 σ⃗
might have something to do with angular momentum, or intrinsic angular momen-
tum (spin). Indeed, we can easily convince ourselves that the Σμν matrices, defined
as
i
Σμν = [γ μ , γ ν ] , (8.68)
2
fulfill
σk 0
Σi = ijk Σjk , Σk = ( ). (8.69)
0 σk
The relation in Eq. (8.69) is basically the same as Eq. (8.53) only with the spin
instead of the angular momentum operators. The spin matrices 12 Σμν fulfill the
same algebraic relations for generators of the Lorentz group, in full analogy to the
angular momenta Lμν [see Eq. (8.59)],
[ 12 Σμν , 12 Σκλ ] = i (g μλ 21 Σνκ + g νκ 12 Σμλ − g μκ 21 Σνλ − g νλ 12 Σμκ ) . (8.70)
Given generators ωμν , a matrix representation of the Lorentz group is thus given by
the second term in Eq. (8.61), with Lμν replaced by 12 Σμν . So, we obtain a spinor
representation of the Lorentz group element Λ as follows,
i
S(Λ) = exp (− ω αβ Σαβ ) . (8.71)
4
Advanced Topics 309
This is a 4 × 4 matrix, which “lives” in some internal, spin space of the particle, not
in space-time. The element ab of S(Λ), for infinitesimal Lorentz transformations,
reads as
i
S(Λ)ab ≈ δab − ω αβ (Σαβ )ab . (8.72)
4
Here, a, b = 1, 2, 3, 4 are spinor indices; they live in the “internal” world of the
particle.
8.3 Relativistic Classical Field Theory
8.3.1 Maxwell Tensor and Lorentz Transformations

One may have noticed that the list of four-vectors given in Eq. (8.25) contains
the electromagnetic scalar and vector potentials, but not the field strengths. The
reason for this curious phenomenon is to be explained. Namely, the field strengths
are actually obtained as derivatives of the potentials with respect to the space-
time coordinates and therefore transform as a second-rank tensor under Lorentz
transformations, and not as a four-vector.
Let us define the antisymmetric electromagnetic field-strength tensor (obviously
of second rank) as
F μν = ∂ μ Aν − ∂ ν Aμ , ⃗ ,
Aμ = (Φ, c A) (8.73)
where we recall (8.25). In the current derivation, we suppress the space-time
arguments (⃗r , t) of the fields. Under Lorentz transformations, a second-rank tensor
transforms as follows
F ′μν = Λμ ρ Λν δ F ρδ . (8.74)
This transformation property is different from that of the scalar and vector poten-
tial. We take notice of the explicit form of the components F i0 and write
∂ 1 ∂
F i0 = ∂ i A0 − ∂ 0 Ai = −∂i Φ − c ∂ 0 Ai = − i
Φ− (cAi ) = E i , (8.75)
∂x c ∂t
while F 0i = −F i0 = −E i , where i = 1, 2, 3 (spatial index). The component F 23 reads
∂ z ∂ y ⃗ x = −c B x .
F 23 = ∂ 2 A3 − ∂ 3 A2 = c (− A + ⃗ × A)
A ) = −c(∇ (8.76)
∂y ∂z
A more general form is
F i0 = E i , F 0i = −E i , F ij = −c ijk B k . (8.77)
⃗ has the same physical dimension as
In SI mksA units, the electric field strength E
⃗
c B. We can show the second equation in (8.77) as follows,
1 mi ijk jk c
c mi B i = − F = −c mi ijk ∂ j Ak = − (δ j δ mk − δ k δ mj ) ∂ j Ak
2 2
= c (∂ m A − ∂ Am ) = F m = −F m . (8.78)
We have used the relation im ijk = δ j δ mk − δ k δ mj .
The field strengths can be reconstructed from the components of the field-
strength tensor as
1 1
E i = F i0 , c B i = − ijk F jk = ijk F jk , (8.79)
2 2
where 123 = 1 and 123 = −1. The second equation in (8.79) can be shown as follows,
1 ijk 1 1
B i = ijk ∂j Ak = (∂j Ak − ∂k Aj ) = − ijk (∂ j Ak − ∂ k Aj ) = − ijk F jk ,
2 2 2c
(8.80)
The structure of the field-strength tensor in all detail, in a component-wise repre-
sentation, thus reads as follows,
μν
⎡⎛ 0 −E x −E y −E z ⎞⎤
⎢ ⎥
⎢⎜ E x 0 −cB z cB y ⎟⎥
⎢⎜ ⎟⎥
F μν = ⎢⎜ y x ⎟⎥ . (8.81)
⎢⎜ E cB z 0 −cB ⎟⎥
⎢ ⎥
⎢⎝ E z −cB y cB x ⎠⎥
⎣ 0 ⎦
With the metric gμν , we can lower the indices to obtain Fμν = gμρ gνδ F ρδ ,
⎡⎛ 0 E x E y E z ⎞⎤
⎢ ⎥
⎢⎜ −E x 0 −cB z cB y ⎟⎥
⎢⎜ ⎟ ⎥
Fμν = ⎢⎜ x ⎟⎥ . (8.82)
⎢⎜ −E y cB z 0 −cB ⎟ ⎥
⎢ ⎥
⎢⎝ −E z −cB y cB x 0 ⎠⎥
⎣ ⎦μν
The Lagrangian density (in short: the Lagrangian) of classical electrodynamics is a
Lorentz scalar and reads as
0 ⃗ 2 2 ⃗ 2 0
L= (E − c B ) = − F μν Fμν . (8.83)
2 4
Together with currents, we have
0 ⃗ .
L = − F μν Fμν − J μ Aμ , J μ = (ρ, c−1 J) (8.84)
4
Appealing to the formalism introduced in Sec. 3.2.2, the variation of the four-vector
potential then leads to the equation
δS = ∫ d4 x [L(A + δA) − L(A)]
0 1 1 ν
= − ∫ d x [− (−2∂μ F δAν + 2∂ν F δAμ ) − J δAν ]
4 μν μν
4 4 0
0 1
= − ∫ d4 x [ ∂μ F μν δAν − J ν δAν ]
4 0
0 1 !
= − ∫ d4 x (∂μ F μν − J ν ) δAν = 0 . (8.85)
4 0
By virtue of the arbitrariness of the variation of the four-vector potential, the Euler–
Lagrange equations are obtained,
∂L ∂ ∂L 1 ν
= , ∂μ F μν = J . (8.86)
∂Aν ∂xμ ∂(∂μ Aν ) 0
Advanced Topics 311
The Cartesian components of the equation ∂μ F μν = J ν /0 read as follows,

∂ x ∂ y ∂ z 1
ν=0∶ E + E + E = ρ, (8.87a)
∂x ∂y ∂z 0
1 ∂ x ∂ z ∂ 1 x
ν=1∶ − E +c B − c By = J , (8.87b)
c ∂t ∂y ∂z 0 c
1 ∂ y ∂ ∂ z 1 y
ν=2∶ − E + c Bx − c B = J , (8.87c)
c ∂t ∂z ∂x 0 c
1 ∂ z ∂ y ∂ x 1 z
ν=3∶ − E +c B −c B = J . (8.87d)
c ∂t ∂x ∂y 0 c
With μ0 = 1/(0 c2 ), these equations can be summarized, in vector form, as the
inhomogeneous Maxwell equations,
∇ ⃗ = 1 ρ,
⃗ ⋅E ⃗− 1 ∂ E
⃗ ×B
∇ ⃗ = μ0 J⃗ . (8.88)
0 c2 ∂t
The question is how to get the homogeneous Maxwell equations. First of all, we
define the dual field-strength tensor F̃μν as
1
F̃μν = μνρδ Fρδ . (8.89)
2
It reads, in components,
μν
⎡⎛ 0 −cB x −cB y −cB z ⎞⎤
⎢ ⎥
⎢⎜ cB x 0 E z −E y ⎟ ⎥
⎢ ⎟⎥
F̃μν = ⎢⎜
⎜ ⎟⎥ . (8.90)
⎢⎜ cB −E
y z
0 E ⎟⎥
x
⎢ ⎥
⎢⎝ cB z E y −E x 0 ⎠⎥
⎣ ⎦
The homogeneous Maxwell equations, which cannot be derived from the variation
of the action S = ∫ d4 x L, are given by
∂μ F̃μν = μνρδ ∂ν Fρδ = 0 . (8.91)
In Cartesian components, these equations read
∂ ∂ ∂
ν=0∶ (cB x ) + (cB y ) + (cB z ) = 0 , (8.92a)
∂x ∂y ∂z
1 ∂ ∂ y ∂ z
ν=1∶ − (cB x ) + E − E = 0, (8.92b)
c ∂t ∂z ∂y
1 ∂ ∂ z ∂ x
ν=2∶ − (cB y ) + E − E = 0, (8.92c)
c ∂t ∂x ∂z
1 ∂ ∂ x ∂ y
ν=3∶ − (cB z ) + E − E = 0. (8.92d)
c ∂t ∂y ∂x
In vector notation, one can summarize them into the homogeneous Maxwell
equations,
∇ ⃗ = 0,
⃗ ⋅B ⃗ ×E
∇ ⃗+ ∂ B⃗ = 0. (8.93)
∂t
The Lorentz transformation of the elements of the field-strength tensor has already
been indicated in Eq. (8.74), in symbolic form, as F ′μν = Λμ ρ Λν δ F ρδ . We now
need to clarify how this translates into the corresponding transformations of the
field strengths E ⃗ and B.
⃗
A little algebra reveals that the transformations intertwine the electric and mag-
netic fields under a Lorentz boost into an inertial frame moving with velocity v⃗, and
are given as follows,
⃗
E⃗ ′ = γ (E ⃗ + (1 − γ) E ⋅ v⃗ v⃗ ,
⃗ + v⃗ × B) (8.94a)
v2
⃗
B⃗ ′ = γ (B ⃗ + (1 − γ) B ⋅ v⃗ v⃗ .
⃗ − 1 v⃗ × E) (8.94b)
c 2 v2
This transformation generalizes the transformation laws given in Eqs. (8.5), (8.6)
and (8.7), to the case of a second-rank tensor. The transformation from the moving
to the rest frame of the electric and magnetic induction fields is given by Eq. (8.94)
under the replacement v⃗ → −⃗ v,
⃗′
⃗ = γ (E
E ⃗ ′ ) + (1 − γ) E ⋅ v⃗ v⃗ ,
⃗ ′ − v⃗ × B (8.95a)
v2
⃗′
⃗ = γ (B
B ⃗ ′ + 1 v⃗ × E⃗ ′ ) + (1 − γ) B ⋅ v⃗ v⃗ . (8.95b)
c2 v2
It is easy to check by explicit calculation that the following two quantities are
Lorentz invariant,
F ′μν Fμν
′
= 2 (c2 B ⃗ ′2 ) = 2 (c2 B
⃗ ′2 − E ⃗ 2 ) = F μν Fμν ,
⃗2 − E (8.96a)
F̃′μν Fμν
′ ⃗′ ⋅ B
= − 4c E ⃗ ′ = −4c E
⃗⋅B
⃗ = F̃μν Fμν . (8.96b)
Current and charge density transform into the moving frame as follows,
J⃗ ⋅ v⃗
J⃗′ = J⃗ − γ ρ v⃗ + (γ − 1) 2 v⃗ , (8.97a)
v
1
ρ′ = γ (ρ − 2 J⃗ ⋅ v⃗) , (8.97b)
c
and back,
J⃗′ ⋅ v⃗
J⃗ = J⃗′ + γ ρ′ v⃗ + (γ − 1) 2 v⃗ , (8.98a)
v
1
ρ = γ (ρ′ + 2 J⃗′ ⋅ v⃗) . (8.98b)
c
Finally, let us briefly comment on a somewhat subtle point. According to
Eq. (8.81), the field-strength tensor F μν is composed of elements which constitute
vectors (electric field) and pseudo-vectors (magnetic field). Under parity, a pseudo-
vector does not change sign while a vector does. Still, the Lorentz transformation
of the tensor is given by Eq. (8.74). How can we intuitively understand why F μν is
composed of entries which transform differently under parity? In order to answer
Advanced Topics 313
this question, let us step back and consider the velocity four-vector β μ = (γ, γ⃗
v),
the relativistic momentum m γ v⃗, and the proper time τ . The Lorentz force (1.110)
reads, relativistically,
d μ d
Fμ =
p = (m β μ ) = q F μν βμ . (8.99)
dτ dτ
The spatial components of this equation read as follows,
d
F⃗ = (mγ⃗v) = γ q (E⃗ + v⃗ × B)
⃗ . (8.100)
dτ
Identifying the infinitesimal laboratory frame time interval as dt = γ dτ , one finds
d ⃗ + v⃗ × B)
⃗ .
v ) = q (E
(mγ⃗ (8.101)
dt
The Lorentz force F μ transforms as a vector. In order for the expression F μν βμ
to also transform as a vector, we consider the case of spatial index μ = i = 1, 2, 3
and we must investigate the cases ν = 0 and ν = j = 1, 2, 3. For ν = 0, we have
β 0 → β 0 under parity (no sign change), and so we conclude that F i0 must be the
component of a vector. For ν = j, we have β j → −β j under parity (sign change), and
so, for F i to transform as a vector, we conclude that F ij must be the component
of a pseudo-vector. Indeed, an inspection of the field-strength tensor reveals that
the entries F ij constitute a pseudo-vector: namely, the components of the magnetic
field.
8.3.2 Maxwell Stress–Energy Tensor

We have discussed the relativistically covariant field-strength tensor, which allowed
us to identify the transformation properties of the electric and magnetic fields
under Lorentz transformations. One ensuing question then is how to generalize
the Maxwell stress tensor , discussed in Sec. 1.3.2, to the Maxwell stress–energy
tensor, i.e., into covariant form, in which case the tensor also includes a reference
to the field energy (hence the addition to its name, where the “stress” becomes
“stress–energy”). Indeed, the tensor
1 1
T μν = − (F μα F ν α − g μν Fαβ F αβ ) (8.102)
μ0 4
has the structure
⎛ 2 (0 E⃗ + B ⃗ 2 /μ0 ) Sx /c Sy /c Sz /c ⎞
1 2
⎜ Sx /c − xx − xy − xz ⎟
T μν =⎜
⎜
⎟,
− yz ⎟
(8.103)
⎜ Sy /c − yx − yy ⎟
⎝ Sz /c − zx − zy − zz ⎠
where S⃗ is the Poynting vector, defined in Eq. (1.105), i.e., S⃗ = (E
⃗ × B)
⃗ /μ0 , and
is the Maxwell stress tensor, defined in Eq. (1.120),
E⃗ 2 ⃗2
⃗⊗E
= 0 (E ⃗−
2
3×3 ) + μ1 ⃗⊗B
(B ⃗− B
2
3×3 ) = ij êi ⊗ êj . (8.104)
0
We had previously written the equation (1.118) for the force density,
1 ∂ ⃗
f⃗ = ∇
⃗⋅
S. − (8.105)
c2 ∂t
This equation finds a natural expression in terms of the components T μν of the
stress–energy tensor, given in Eq. (8.103),
∂ ∂ Si ∂ ∂ ∂
fi = ji
− = − j T ji − 0 T 0i = − μ T μi , (8.106)
∂xj ∂x0 c ∂x ∂x ∂x
where i, j = 1, 2, 3 (Latin indices are spatial), whereas Greek indices are space-time
indices (μ, ν = 0, 1, 2, 3).
Relativistically, one should interpret the force density as
∂ ∂ ⃗
f⃗ = P⃗ = P(⃗r , t) , (8.107)
∂t ∂t
where P⃗ is the relativistic momentum density [here, the differentiation is with regard
to t, not τ , as suggested by Eq. (8.101)], and it takes place at constant r⃗. Promoting
the momentum density to a four vector, one identifies the energy density P 0 = w/c,
where w is the electromagnetic energy density.
In the sense of special relativity, the force density f⃗ should be generalized to a
four-vector f μ , which implies that we should find a physical interpretation for f 0 .
We write for the zeroth component,
∂ μ0 1 ∂ 00 ∂
f0 = − μ
T = − T − i T i0
∂x c ∂t ∂x
1 ∂ 0 ⃗ 2 B ⃗2 ∂ Si 1 ∂ 1
= − ( E + )− i =− u− ∇⃗ ⋅ S⃗ , (8.108)
c ∂t 2 2μ0 ∂x c c ∂t c
where
0 ⃗ 2 B⃗2
u= E + (8.109)
2 2μ0
is the field-energy density. We should thus identify
∂ w ΔW ∂w ΔP ⃗ ⋅ J⃗ ,
f0 = , w= , = =E (8.110)
∂t c ΔV ∂t ΔV
as the work done per unit volume on the charges, by the electric field (here, ΔW
denotes the work dissipated into the mechanical motion of the particles, and ΔP
the power). The equation f 0 = −∂T μ0 /∂xμ thus is equivalent to
∂ ∂
⃗ ⋅ S⃗ + u + w = 0 ,
∇ (8.111)
∂t ∂t
which is just the Poynting theorem (1.106). Finally, the relativistic generalization
of Eq. (1.122) is found as
∂ μν
fν = − T . (8.112)
∂xμ
Advanced Topics 315
In the absence of material forces (f μ = 0), one can interpret this equation as
follows. The explicit separation into time and spatial derivatives
∂ T 0ν ∂
+ i T iν = 0 , (8.113)
∂t c ∂x
is reminiscent of the continuity equation,
∂ ∂
ρ + i Ji = 0 . (8.114)
∂t ∂x
So, for given ν, the vector with components T iν can be interpreted as the current
for the “charge density” T 0ν . In some sense, as much as J⃗ measures the flow of the
charge, T iν measures the flow of the field momentum density T 0ν .
If ν = 0, then the “charge density” T 00 is equal to the energy density of the elec-
tromagnetic field, while the “current density” with components T i0 is the Poynting
vector or, the momentum density of the field times c in the ith Cartesian direction.
Indeed, from Eq. (8.105), we learn that the momentum density of the field is S/c ⃗ 2.
If ν = j with j = 1, 2, 3, then the “charge density” T is the momentum density in
0j
the jth Cartesian direction, while the “current density” with components T ij is the
vector êi T ij = −êi ij , i.e., the (negative of the) jth column vector of the Maxwell
stress tensor. Thus, we can say that the jth column vector of the Maxwell stress
tensor measures the flow of the jth component of the field momentum.
8.3.3 Lorentz Transformation and Biot–Savart Law

Magnetic and electric fields transform into each other by a Lorentz transformation.
Arguably, magnetic-field effects are the only relativistic effects that can easily be
seen in a laboratory and remind us of the enormous strength of the electromagnetic
interaction, which over long distances is suppressed by the neutrality of macroscopic
objects. We start from Eq. (8.95b),
⃗′
⃗ = γ (B
B ⃗ ′ + 1 v⃗ × E⃗ ′ ) + (1 − γ) B ⋅ v⃗ v⃗ . (8.115)
c2 v2
Let us consider a moving charged object and aim to calculate its magnetic field.
We might use the Liénard–Wiechert potentials and calculate the curl of the vector
potential (5.261b). However, in a conductor (stationary current distribution), the
electrons move slowly, and it is otherwise possible to find a suitable approximation
to the magnetic field generated by the moving charge, as follows. We observe that
in a comoving frame, by definition, the moving object is at rest, and therefore it
does not generate a magnetic field, i.e., B ⃗′ = ⃗ 0. Then, in view of Eq. (8.115),
⃗=γ v⃗ ⃗′ ,
B ×E (8.116)
c2
where E ⃗ ′ is the electric field strength in the moving frame. We investigate the
infinitesimal magnetic field dB ⃗ corresponding to the infinitesimal electric field dE⃗ ′ ,
where dE⃗ is the electric field generated by an infinitesimal charge element,
′
′
⃗ = γ v⃗ × dE⃗ ′ , ⃗ ′ = dq r⃗ − r⃗ = 1 r⃗ − r⃗′ ′ ′ 3 ′
dB dE ′
ρ (⃗
r )d r . (8.117)
c 2 4π0 ∣⃗
r − r⃗ ∣ 3 r − r⃗′ ∣3
4π0 ∣⃗
The coordinate r⃗′ is given in the moving frame; the element of charge dq = ρ(⃗ r′ ) d3 r′
′
is located at r⃗ . However, we shall ignore the transformation of the coordinates,
assuming that r⃗ and r⃗′ are equal in the rest frame of the particle and in the labora-
tory frame, effectively performing our calculations in the limit γ → 1. Within this
approximation, the moving charge generates a magnetic field of the form
⃗≈ 1 r⃗ − r⃗′ μ0 ⃗ ′ r⃗ − r⃗′
dB γ ρ′ (⃗
r′ ) (⃗
v× ′
) d3 r′ = r )×
(J(⃗ ) d3 r′ . (8.118)
4π0 c 2 ∣⃗
r − r⃗ ∣ 3 4π r − r⃗′ ∣3
∣⃗
This is the Biot–Savart law for a point charge. Here, we have identified the current
density corresponding to a charge density ρ(⃗r′ ) as
⃗ r′ ) = γ ρ′ (⃗
J(⃗ r′ ) v⃗ . (8.119)
While the factor γ is beyond the order of approximation considered in our derivation,
in view of approximations carried out previously, it is perhaps interesting to point
out that it can easily be understood. We consider the Lorentz contraction of the
charge density seen in the moving frame, ρ′ (⃗r′ ) = dq/(d0 dA), into the laboratory
frame, where 0 is supposed to be directed parallel to the velocity v⃗, and dA is an
infinitesimal cross-sectional area element. Then, 0 is Lorentz-contracted, acquires
a factor 1/γ, and we have Eq. (8.119). Alternatively, this follows from Eq. (8.98a),
setting J⃗′ = ⃗
0. Integrating, we find the Biot–Savart law,
′
B(⃗ ⃗ r′ ) × r⃗ − r⃗ ) .
⃗ r ) = μ0 ∫ d3 r′ (J(⃗ (8.120)
4π r − r⃗′ ∣3
∣⃗
This is the Biot–Savart law for extended, stationary current distributions. Again,
magnetic fields are the only relativistic effects we are going to see in our everyday
life, probably. They exist because the electromagnetic force is that much stronger
than the gravitational force; the former is being compensated on macroscopic scales
by the presence of unlike and compensating charges.
8.3.4 Relativity and Magnetic Force
As we have just seen, it is possible to derive the Biot–Savart law, and thus, the mag-
netic fields generated by moving charges, on the basis of the Lorentz transformation.
This concerns the magnetic field, not the magnetic (Lorentz) force. Likewise, it is
instructive to investigate whether or not the magnetic term in the Lorentz force can
be obtained as a relativistic effect. To this end, we investigate the Lorentz contrac-
tion of moving charges in a conductor, where the test charge moving alongside the
conductor would see a net electric field because the positive and negative charges in
the conductor move at different speeds and experience a different degree of Lorentz
contraction. The net result is that the conductor is electrically charged from the
point of view of the moving charge, and a net electric force results. It turns out to
be a little nontrivial to formalize this general idea, and this is why we present the
Advanced Topics 317
v
I

v−
q x
Fig. 8.1 We investigate the geometry of the wire as seen from the frame of reference of the
moving charge q. A current current I ′ flows to the right. The electrons in the wire flow to
the left at velocity v−′ . The positively charged ions are stationary and move to the left at
the same velocity as the reference frame, namely, at velocity v. By the addition theorem,
the drift velocity of the electrons in the moving frame is equal to v−′ = (v + vd )/(1 + v vd /c2 ),
where vd is the drift velocity of the electrons in the rest frame of the wire.
derivation in some detail. We also recall that the Lorentz force cannot be derived
from the Maxwell equations alone.
Let us assume that a particle is moving alongside and parallel to a conductor,
with its trajectory parallel to the conductor. The conductor is an infinite wire along
the x axis. Inside the wire, negative charges move with a drift velocity vd to the left.
A charged particle moves at velocity v to the right, i.e., in the positive x direction,
at a distance R from the conductor, and parallel to the conductor. The particle in
its own rest frame is described by primed quantities like I ′ , which is the current
seen in the rest frame of the moving particle. The positively charged particles are at
rest with respect to the wire (see Fig. 8.1). From the rest frame of the particle, the
positive charges inside the wire move to the left with velocity v, i.e., with the same
velocity as the entire conducting wire moves to the left. The negatively charged
particles move to the left, with velocity v−′ (as seen in the frame of the particle),
where v−′ is the sum of the velocity v of the conducting wire and the drift velocity
vd of the electrons, the latter being determined in the laboratory frame.
The velocity of the negative charges moving to the left, seen in the reference
frame of the moving particle at distance R, is given by the addition theorem (8.19)
as
v + vd β + βd
v−′ = v vd , β−′ = , (8.121)
1+ 2 1 + β βd
c
where we define the scaled velocity β = v/c (and analogously for all other velocities).
Let the linear charge density inside the conductor be denoted as λ+ and λ− for
the positive and negative charges, respectively. The charge densities seen in the
reference frame of the moving particles are denoted by a prime. The total charge
density λ seen in the laboratory frame and λ′ seen by the moving particle are given
as
λ = λ+ + λ− = 0 , λ′ = λ′+ + λ′− ≠ 0 , (8.122)
where the first equality holds because the conductor remains electrically neutral
in its own rest frame (laboratory frame). Due to Lorentz contraction, the moving
observer (at velocity v) sees the positively charged particles as being contracted by
1
λ′+ = √ λ+ , (8.123)
1 − v 2 /c2
where we assume that v is the velocity of the particle moving alongside the wire. We
assume that the positively charged atomic nuclei are stationary with respect to the
wire. Now, the moving observer sees the negative linear charge density enhanced
by a factor
1
λ′− = √ (λ− )0 , (8.124)
1 − v−′2 /c2
where v−′ is the drift velocity of the negative moving charges with respect to the
particle flying by [see Eq. (8.121)]. In Eq. (8.124), the quantity (λ− )0 is the charge
density of the electrons at rest. The charge density λ− seen in the laboratory frame,
which is identical to the rest frame of the conductor, is different from the one in the
rest frame of the moving particle and also different from the one in the rest frame
of the positive charges in the conducting wire. Namely, λ− reads as
1 √
λ− = √ (λ− )0 , (λ− )0 = 1 − βd2 λ− , (8.125)
1 − vd2 /c2
where βd = vd /c. Hence,
√ √ √
′ 1 − βd2 1 − βd2 1 − βd2
λ− = √ λ− = B λ− = (1 + β βd ) √ λ−
1 − β−′2 D β + β 2
(1 + β β )
2
− (β + β )
2
D
E1 − ( d
)
d d
1 + β βd
√
1 − βd2 1 + β βd
= (1 + β βd ) √ λ− = √ λ− . (8.126)
(1 − β ) (1 − βd )
2 2
1 − β2
So, relating the charge densities λ′+ and λ′− in the rest frame of the moving particle
to the charge densities λ+ and λ− in the rest frame of the conductor, we have
1 1 + β βd β βd λ−
λ′ = λ′+ + λ′− = √ λ+ + √ λ− = √
1−β 2 1−β 2 1 − β2
1 v 1 vI vI
= −√ 2
(−vd λ− ) = − √ 2
= −γ 2 , (8.127)
1−β c
2 1−β c
2 c
Advanced Topics 319
where γ = (1 − β 2 )−1/2 is the relativistic Lorentz factor. We have used Eq. (8.122).
For the moving particle at velocity v, the conductor behaves as if it were a long,
electrically charged wire with linear charge density λ′ = λ′+ + λ′− . Note that the
current I = −λ− vd is calculated to be moving to the right, and it is calculated
in the laboratory frame (we would assign a negative value to vd as the electrons
move to the left). We must now relate the current I seen in the rest frame of the
conductor, to the current I ′ in the rest frame of the moving particle. We calculate
the magnitude of I ′ by relating the current I ′ seen by the moving particle, to the
current I in the laboratory frame. We define I ′ to be the current moving to the
right. If v−′ is the modulus of the velocity and λ′− is the charge per unit length,
defined to be negative, then
⎛ 1 ⎞ ⎛ 1 + β βd ⎞ ′
−I ′ = λ′+ v + λ′− v−′ = √ λ+ v + √ λ− v
⎝ 1−β 2 ⎠ ⎝ 1 − β2 ⎠ −
1 ⎛ 1 + β βd ⎞ v + vd
= −√ λ− v + √ ( ) λ−
1 − β2 ⎝ 1 − β 2 ⎠ 1 + β βd
λ− vd I
=√ = −√ . (8.128)
1 − β2 1 − β2
Hence, based on Eq. (8.127), we have
v I′
λ′ = − (8.129)
c2
in the rest frame of the particle; the result is exact to all orders in the relativistic
corrections. The infinitesimal potential generated by a line element dx on the
conductor reads
1 λ′
dΦ′ = √ dx , (8.130)
4π0 x2 + R2
where R is the transverse distance from the particle’s trajectory to the wire. The
electric field acting in the rest frame of the particle reads
1 ∞ ∂ λ′ λ′
Ez′ = ∫ dx (− )√ = . (8.131)
4π0 −∞ ∂R x2 + R2 2π0 R
Here, R denotes the distance from the wire, and we implicitly assume that the test
particle is located above the wire; a positive value of λ implies that the particle is
pushed upward. If the test particle were below the wire, then the direction of Ez′ is
reversed. So, the electric force on the test particle is
d ′ d λ′ q v ′ q μ0 v I ′ q μ0 v I
Fz′ = q Ez′ = pz = γ pz = q =− 2
I = − = −γ ,
dτ dt 2π0 R 2π0 R c 2πR 2πR
(8.132)
which is directed in the negative z direction, i.e., toward the conducting wire. Here,
dτ = dt/γ is the proper time interval. We note that p′z = pz for the momentum
component because the relativistic Lorentz boost is in the x direction. Our result is
in agreement with intuition, because the negative particles move to the right faster
than the positively charged particles. The drift velocity is added to the relative
velocity of the wire; the negative charges experience a more pronounced Lorentz
contraction, and the test charge q of the particle (which we assume to be positive)
is attracted to the conducting wire because of the negative net charge of the latter.
In the rest frame, thus, we have
d q μ0 v I
Fz = pz = − . (8.133)
dt 2πR
Let us briefly show that this result is compatible with the usual formulation of the
Lorentz force. The magnetic field can be calculated from Ampere’s law as follows
(in the rest frame of the particle). We reduce all vector-valued quantities to their
moduli and consider a straightforward application of the Ampere–Maxwell law in
⃗ ⋅ dA⃗ = μ0 I. This leads to the formulas
steady state, where its integral form is ∮ B
μ0 I q μ0 v I
2πRB = μ0 I , B= , F = qvB = . (8.134)
2πR 2πR
The direction of the force (toward the conducting wire) is in line with the right-
hand rule applied to the magnetic force as obtained for the current configuration,
as implied by the formula F⃗ = q⃗ ⃗ The current to be used in Ampere’s law
v × B.
⃗ ⃗
∮ B ⋅ dA = μ0 I points into the positive x direction because the negatively charged
electrons inside the conductor move to the left, and we are investigating a positively
charged test particle above the conductor. Then, the magnetic induction field above
the conductor is directed toward us, and we recall that the particle is moving to
the right. So, the right-hand rule for the vector product dictates that the magnetic
force is pointing in the downward direction, toward the conductor.
8.3.5 Covariant Form of the Liénard–Wiechert Potentials
We have discussed the Liénard–Wiechert potentials in Sec. 5.6.1, in a manifestly

noncovariant form. For relativistic moving objects, it is instructive to express the
potentials in covariant form, implying that we shall encounter the proper time of
the particle. Due to time dilation, the proper time interval dτ for a particle moving
at velocity v is given as [see Eq. (8.12)]
−1/2
1 v2
dτ = dt , γ = (1 − ) , dτ < dt , (8.135)
γ c2
where dτ is the time interval measured in the rest frame of the particle, and dt is
the time interval in the laboratory frame. The velocity four-vector is
v⃗ v⃗ v⃗ 2
β μ = (γ, γ ) , β0 = γ , β⃗ = γ , β μ βμ = γ 2 (1 − ( ) ) = 1 . (8.136)
c c c
Advanced Topics 321
We recall the manifestly non-covariant form of the Liénard–Wiechert potentials as

given in Eqs. (5.261a) and (5.261b),
−1
q 1 ⎛ ⃗˙ ret ) r⃗ − R(t
R(t ⃗ ret ) ⎞
Φ (⃗
r , t) = 1− ⋅ , (8.137a)
4π0 ∣⃗ ⃗ ret )∣ ⎝
r − R(t c ∣⃗ ⃗ ret )∣ ⎠
r − R(t
−1
q ⃗˙ ret ) ⎛
R(t ⃗˙ ret ) r⃗ − R(t
R(t ⃗ ret ) ⎞
A⃗ (⃗
r , t) = 1− ⋅ . (8.137b)
4π0 c2 ∣⃗ ⃗ ret )∣ ⎝
r − R(t c ∣⃗ ⃗ ret )∣ ⎠
r − R(t
These potentials are still given in a not completely covariant form, in the sense that
the quantities in them carry spatial, not Lorentz, indices. We therefore explore
the question how the Liénard–Wiechert potentials qualify themselves as manifestly
covariant formulas. In this endeavor, we start with the scalar potential, which we
rewrite as follows,
γq 1
Φ (⃗
r, t) = . (8.138)
4π0 ⎛ ⃗˙ ret ) ⎞
R(t
γ ∣⃗ ⃗ ret )∣ − γ
r − R(t ⋅ (⃗ ⃗ ret ))
r − R(t
⎝ c ⎠
The retarded time fulfills the relations,

∣⃗ ⃗ ret )∣
r − R(t
tret = t − < t, c(t − tret ) = ∣⃗ ⃗ ret )∣ .
r − R(t (8.139)
c
We are now in the position to order the space-time coordinate of the particle at the
retarded time, when the radiation is emitted, into a four-vector form,
⎛ ⃗˙ ret ) ⎞
R(t
⃗ ret )) ,
Rμ (tret ) = (c tret , R(t β μ (tret ) = γ, γ . (8.140)
⎝ c ⎠
The space-time coordinate where the potentials are observed is written as xμ =

(c t, r⃗). With β 0 = γ, we can thus write Φ (⃗
r, t) as
γq 1
Φ (⃗
r , t) =
4π0 ⃗
˙
⎛ R(tret ) ⎞
γ c (t − tret ) − γ ⋅ (⃗ ⃗ ret ))
r − R(t
⎝ c ⎠
q β0
= . (8.141)
4π0 β μ (tret ) (xμ − Rμ (tret ))
Repeating the same steps for the vector potential, we obtain the formulas
q β 0 (tret )
Φ (⃗
r , t) = , (8.142a)
4π0 β μ (tret ) (xμ − Rμ (tret ))
q ⃗ ret )
β(t
c A⃗ (⃗
r , t) = , (8.142b)
4π0 β (tret ) (xμ − Rμ (tret ))
μ
⃗
or, using the ⋅ notation for the four-vector scalar product and ordering Aμ = (Φ, c A)
into a four-vector, we obtain
q β μ (t)
Aμ (X) = ∣ . (8.143)
4π0 β(t) ⋅ (x − R(t)) t=tret
This result is given in SI mksA units.
8.4 Towards Quantum Field Theory
8.4.1 Casimir Effect and Quantum Electrodynamics

In the preceding sections, we have approached the relativistic quantum field theory
regime from the relativistic side; now, we approach it from the quantum side. To
this end, it is instructive to realize that fluctuations in both the macroscopic as
well as the quantum world can have important observable consequences. Seamen
know this effect; ships in the ocean traveling at constant speed, and close to each
other, are known to attract each other because the waves are broken in between
the ships. In other words, the wave motion is attenuated between the ships. By
contrast, no such attenuation occurs on the respective “outward” sides of the ships,
i.e., on the sides not opposing each other. Therefore, the waves exert forces on the
ships; these are attenuated in between but not attenuated from outside, and the two
ships attract each other. Therefore ships on the open sea have to keep a minimum
distance from each other; and that includes ships traveling in convoys. The Casimir
attraction of perfectly conducting plates is of the same physical origin.
Our brief overview of the effect gives us an opportunity to look at field quan-
tization, which is necessary in order to describe the effect. The most important
paradigm in the derivation is the “fundamental conceptual equation” of quantum
electrodynamics:
Physical Observable = Bare Observable∣reg. − Counter Term∣reg.
= lim ∫ (Bare Observable∣reg.−int. − Counter Term∣reg.−int. ) (8.144)

regulator→0
The meaning of these expressions is as follows: (i) Bare Observable is an an expres-

sion describing the physical observable, as derived from first principles, invoking the
formalism of quantized fields. In quantum field theory, one would use the “normal
Feynman rules of QED” (see Ref. [20]). (ii) The Bare Observable∣reg. is a regular-
ized version of the expression for the bare observable, where the regulator respects
basic symmetries of the physical problem, as discussed in greater detail in the fol-
lowing. (iii) A Counter Term is is a quantity which needs to be absorbed into a
parameter of the theory, i.e. interpreted as a physically unobservable contribution to
an energy, or to a scattering amplitude, which needs to be subtracted in order to ob-
tain a physically sensible answer. (iv) In the regularized version Counter Term∣reg.
of the counter term, one needs to choose a regularization which is compatible with
Advanced Topics 323
the one used for the bare observable. (v) The Bare Observable∣reg.−int. is the inte-
grated form of the regularized bare observable. Without specifying the variables
which need to be integrated over, we denote the regularized integrand by the symbol
“reg.-int.”. (vi) Of course, the Counter Term∣reg.−int. : is the regularized integrand
for the counter-term. (vii) Finally, as we let the regulator go to zero, limregulator→0 ,
the regularization is removed after all other operations (including integrations) have
been carried out. Only this sequence ensures that no spurious finite contributions
remain in the final expressions, which could otherwise lead to inconsistent results.
The basic, fundamental Eq. (8.144) is illustrated in the current section. The
physical situation is illustrated in Fig. 8.2. Two perfectly conducting plates are a
distance R apart, and the space in between them is empty. However, there may
still be a nonvanishing interaction, i.e. a nonvanishing (attractive) force between
the plates, because the vacuum (the “nothing”) is influenced by the presence of
the plates. In other words, the discretization of the electromagnetic field modes,
as implied by the presence of the plates, generates a small residual effect, which
leads to an attractive interaction of the plates. All terms listed in the fundamental
Eq. (8.144) are thus important.
We need a counter term. The reason is that the plates are only felt as residual
effects, as a change of the zero-point energy in view of the presence of the plates, as
opposed to a situation where the plates are absent. However, the zero-point energy
in the absence of the plates cannot be physically observable and therefore, must be
subtracted. The approach consists in calculating the sum of the zero-point energies
of the discretized modes in the presence of the plates, minus the sum of the zero-
point energies of all the modes in the absence of the plates. The total zero-point
energy in the presence of the plates should be interpreted as the unrenormalized
(“bare”) quantity, whereas the total zero-point energy in the absence of the plates
is the counter term.
We need a regulator. The reason is that both the sum of the zero-point en-
ergies of the electromagnetic modes in the presence as well as in the absence of
the plates are highly divergent quantities. From physical considerations, only the
infrared region of long wavelengths is influenced by the quantization. The ultraviolet
region does not receive any corrections. The divergence of both the unrenormalized
quantity as well as of the counter term can be remedied in a physically sensible way
by the introduction of an ultraviolet regulator, i.e. by the introduction of a quan-
tity which suppresses the divergence induced by the high-frequency electromagnetic
modes. This ultraviolet regulator should respect basic symmetries of the problem.
In the current case, it means that this ultraviolet regulator should respect, e.g., all
geometric symmetry of the total arrangement in the limit of large plates, where the
edge length L in Fig. 8.2 tends to infinity.
We should be careful about the order of integrations. The regularization param-
eter which induces an exponential damping of the contribution from the ultraviolet
modes, should be kept up until the very end of the calculation. Denoting the
.............................................................................
.................. .............
............. ...........
........... .........
......... ........
........ ........
........ .......
....... .
........................................
...........
....... ........ L
...... ......
.....
..... .....
..... ....
.... ....
.... .... ....
.... ....
.....
..... .....
...... .....
....... ......
.......... ........
..........................................
x
6
.......................... ....................................
......... ...... .......
- z ....
....
......
..... .....
....
....
.....
.....
....
.....
....
....
....
....
?
...
... ... ...
...
... ... ...
..
... ..
....
....
..... .
...
..
...
....
y ......
......... .
.......................
..
..
.......
L

-
a
Fig. 8.2 The panel shows a Casimir box composed of two large
plates of side length L, a distance R apart, with L ≫ R. Some of
the discrete modes of the electromagnetic field (standing waves with
vanishing electric-field amplitude on the plates) are sketched.
regularization parameter as η, we should be careful not to perform the limit η → 0

before all other integrations are done.
Our consideration starts with the calculation of a formal expression for the sum
of zero-point energies of the electromagnetic modes between the plates. We shall
interpret every mode of the electromagnetic field as a harmonic oscillator, which
is n-fold excited when there are n photons present in the mode, and we simply
take this concept for granted in the current derivation. The derivation relies in an
absolutely essential way on the quantization of the electromagnetic field.
8.4.2 Zero-Point Energy

In order to describe the vacuum fluctuations of the electromagnetic field of angular
frequency mode ω, we draw an analogy to the zero-point energy of a harmonic
oscillator of frequency ω. The Hamiltonian in this case is
p2 1
H= + mω 2 q 2 , (8.145)
2m 2
where p is the momentum operator, and m is the mass. Annihilation and creation
operators may be defined as follows,
√ √
1 x x0 p + 1 x x0 p
a= ( +i ̵ ) , a = ( −i ̵ ) . (8.146)
2 x0 h 2 x0 h
The length scale x0 is given by
√
̵
h
x0 = . (8.147)
mω
Advanced Topics 325
The Hamiltonian may thus be rewritten as
̵ (a+ a + 1 ) ,
H = hω (8.148)
2
where a+ a is a number operator that counts the number oscillation quanta. Its spec-
trum consists of nonnegative integers. The zero-point energy (of the fundamental
mode) can thus be inferred as
e0 = 1 ̵ .
hω (8.149)
2
Normally, this zero-point energy has no physically observable effect. However, as

shown by Casimir and Polder [32], a small change in the total zero-point energies
of a collection of oscillators (in this case, modes of the electromagnetic field), be-
tween two parallel perfect mirrors, has a small effect which results in an attractive
interaction between the plates.
We consider two square perfectly conducting metal plates (of dimension L × L,
see Fig. 8.2), which define planes parallel to the xy plane, a distance Δz = R apart.
The plates are supposed to act as perfect mirrors over the entire frequency range
of possible photon modes, which is, of course, an idealization. Furthermore, we
assume that the plate dimension L is much larger than the plate distance, L ≫ R.
We can now explore the connection to the discussion of the fundamental modes
of the rectangular cavity, with L = Lx = Ly being the “long” sides of the rectangle,
and Lz = R being the separation of the plates. In this case, we have seen in
Eq. (7.123) that the fundamental mode is a TM mode, i.e., there is only one mode
with n = nz = 0, but two modes for all other nonzero integer n.
According to Eq. (7.115), we need to have
π
kz = nz , nz = 0, 1, 2, . . . . (8.150)
R
For the x and y components of the photon momentum vector, no such discretization
is indicated, and we have
π
kx = nx ,
L
π
ky = ny ,
L
nx , ny ∈ .
(8.151)
In the limit of large L, summations over nx and ny can therefore be replaced by

integrations,
∞ ∞ ∞ ∞ ∞ ∞
L 2 L 2
∑ ∑ → ( ) ∫ dkx ∫ dky → ( ) ∫ dkx ∫ dky , (8.152)
nx =0 ny =0 π 2π
0 0 −∞ −∞
where we assume that the integral operator is applied to a function symmetric under
kx ↔ −kx and ky ↔ −ky .
According to Eq. (8.149), the zero-point energy of a harmonic oscillator of fre-
̵
quency ω is e0 = 12 hω. We may thus calculate the sum of the zero-point energies
of the photon modes in the space confined by the two plates as follows, keeping
track of the summation limits and the photon polarizations λ. Denoting by λmax
the maximum number of polarization modes available for a given electromagnetic

⃗ one has
mode with wave vector k,
∞ ∞ ∞ λmax ∞ ∞ ∞ λmax
1̵ ⃗
Ebare = ∑ ∑ ∑ ∑ e0 = ∑ ∑ ∑ ∑ h c ∣knx ,ny ,nz ∣
nx =−∞ ny =−∞ nz =0 λ=1 nx =−∞ ny =−∞ nz =0 λ=1 2
√
̵c
h ∞ dk
x
∞ dk
y
∞ λmax
n2 π 2
= 2
L ∫ ∫ ∑ ∑ (kx )2 + (ky )2 + z 2
2 −∞ 2π −∞ 2π nz =0 λ=1 R
√
̵ c L2
h ∞ λmax
n2 π 2
= ∫
2
d k∥ ∑ ∑ k⃗∥2 + z 2 . (8.153)
8π 2 Ê 2
nz =0 λ=1 R
The specification “bare” means that it refers to an unrenormalized, unregularized
quantity which may be divergent (and actually is divergent in our case). The
notation d2 k∥ is chosen for the integral dkx dky . Our considerations from Sec. 7.3.2
imply that λmax = 1 for nz = 0, while λmax = 2 for nz ∈
(in the latter case, we
have both TE as well as TM modes available). It is interesting to observe that
the different number of allowed polarizations for nz = 0 as opposed to nz ≥ 1 finds
a natural mathematical complement in terms of the Euler–Maclaurin summation
formula, as discussed in the following. For now, we obtain
√
̵ c L2
h ⎛⃗ ∞
n2 π 2 ⎞
Ebare = ∫ 2 d k∥ ∣k∥ ∣ + 2 ∑
2
k⃗∥2 + z 2 . (8.154)
8π 2 Ê ⎝ nz =1 R ⎠
The zero-point energy of any particular harmonic oscillator mode is finite, but there
is an infinite number of such modes. So, one might have expected that the total
zero-point energy of the entire system should be a highly divergent quantity. This
is confirmed by Eq. (8.154). Since the quantity on the right-hand side is highly
divergent, the equality only holds formally.
8.4.3 Regularization and Renormalization
We must now transform Ebare → Ebare ∣reg. in the sense of Eq. (8.144), by the intro-
duction of a suitable regularization. Specifically, we introduce a regularization by
replacing the moduli k∥ = ∣k⃗∥ ∣ of the photon wave vectors,
k∥ → hη (k∥ ) ≡ k∥ fη (k∥ ) , (8.155a)
where the regulator fη (k∥ ) fulfills the following conditions, fη (k∥ ) ≈ 1 for k∥ ∼ R−1 ,
and fη (k∥ ) → 0 for k∥ ≫ R−1 , i.e. the regulator suppresses the contribution of those
modes for which the wave length is much smaller than the plate distance. Also, the
function should preserve basic symmetries of problem. One possibility is
fη (k∥ ) = exp (−η k∥ ) , (8.155b)

Advanced Topics 327
where η ≪ R is a free parameter; the limit η → 0 is taken after all other operations
are carried out. We then have
⎡ √ ⎤
̵ c L2
h ⎢ ∞ ⎛ 2 2 ⎞⎥
Ebare ∣reg. = 2 ⎢ (k ) + 2 + nz π ⎥.
∫ 2 ∥ ⎢ η ∑
R2 ⎠⎥
d k h 2 h k (8.156)
8π 2 Ê ⎢
⎣
∥
n=1
η
⎝ ∥
⎥
⎦
This quantity is Bare Observable∣reg. in the sense of Eq. (8.144), and it becomes
obvious how to interpret the “regularized integrand” in our case.
However, the introduction of a regularization is not sufficient; we also need
to perform a subtraction (counter term). Namely, we know that the zero-point
energy of the photon modes without the plates present is physically unobservable.
Therefore, we have to subtract the energy of these modes, duly regularized in the
same manner as the energy of the modes in the modified vacuum. The difference of
these two terms then gives the physically observable energy shift, i.e., the Casimir
energy, in the sense of Eq. (8.144). The counter-term is given by the vacuum energy,
with the boundary conditions removed, i.e., transforming the discrete sum over the
allowed modes in Eq. (8.154) into an integral,
√
̵ c L2
h ∞ 2 2
C= 2 2 + nz π ,
∫ d k ∫ dn k
4π 2 Ê2 ∥
0
z ∥
R2
√
̵ c L2
h ∞ ⎛ n2 π 2 ⎞
C∣reg. = 2
∫ 2 d k∥ ∫ dnz hη k∥2 + z 2 . (8.157)
4π 2 Ê 0 ⎝ R ⎠
We, therefore, obtain the physically observable, renormalized zero-point energy shift
Eren (R) as a function of the distance R of the plates as Eren (R) = Ebare ∣reg. − C∣reg. ,
⎡ √
̵ c L2
h ∞ ⎢1 ∞ ⎛ n2 π 2 ⎞
Eren (R) = ∫ dk∥ k∥ ⎢ hη (k∥ ) + ∑ hη k∥2 + z 2
⎢2 ⎝ R ⎠
2π 0 ⎢ nz =1
⎣
√ ⎤
∞ ⎛ n2z π 2 ⎞⎥
⎥,
−∫ k∥ +
2
R 2 ⎠⎥
dnz hη (8.158)
0 ⎝ ⎥
⎦
where the integral over the azimuth angle has been carried out. We now substitute
π√ R2 k∥2
k∥ = u, u= . (8.159)
R π2
This implies the relations
dk∥ π π2 π√ π√
= √ , k∥ dk∥ = du , hη (k∥ ) = u fη ( u) , (8.160)
du 2 R u 2R2 R R
so that
√
⎛ n2 π 2 ⎞ π √ π√
hη k∥2 + = u + n 2 f (
η u + n2 ) . (8.161)
⎝ R2 ⎠ R R
The substitution k∥ → u, applied to Eq. (8.158), leads to

Eren (R) ̵ c π2 ∞
h 1√ π√ ∞ √
π√
2
= lim 3 ∫ du [ u f η ( u) + ∑ u + n2z fη ( u + n2z )
L η→0 4R 2 R nz =1 R
0
∞ √ π√
−∫ dnz u + n2z fη ( u + n2z )]
0 R
̵ c π2 1
h ∞ ∞
= lim 3
[ Fη (0) + ∑ Fη (nz ) − ∫ dnz Fη (nz )] , (8.162)
η→0 4 R 2 nz =1 0
where we exchange the order of the u- and nz -integrations for the last term, and
define the function Fη (nz ) by
∞ √ π√ ∞ √ π√
Fη (nz ) = ∫ du u + n2z fη ( u + n2z ) = ∫ dv v fη ( v) , (8.163)
0 R 2 R nz
where the latter form is given by the trivial substitution v = u + n2z . The exchange
of the order of integration is possible because the regularization has led to integrals
which are absolutely convergent. This aspect is very important as it illustrates the
necessity of keeping all regulators finite up until the very last steps of the calculation.
The expression in square brackets in the last line of Eq. (8.162) has the struc-
ture of a discrete sum minus a corresponding integral. The prefactor 12 in front
of the first term finds a natural, geometrical interpretation. Quite fortunately, the
Euler–Maclaurin formula comes to the rescue in our quest of evaluating the tiny dif-
ference between the sum and the integral in the last line of Eq. (8.162). The Euler–
Maclaurin summation formula is given in many textbooks [see, e.g., Eqs. (2.01) and
(2.02) on p. 285 of [9]]. In its full form, the Euler–Maclaurin formula reads (see
Fig. 8.3)
M−1 M
1
2
f (N ) + 12 f (M ) + ∑ f (n) − ∫ dn f (n)
n=N +1 N
q
B2j
=∑ [f (2j−1) (M ) − f (2j−1) (N )] + Rq , (8.164a)
j=1 (2j)!
where the remainder term Rq is
1 M
Rq = − ∫ dx B2q (x − [[x]]) f (2q) (x) . (8.164b)
(2q)! N
Here, [[x]] is the integral part of x, i.e., the largest integer m satisfying m ≤ x, and
Bk (x) is a Bernoulli polynomial defined by the generating function [see Eq. (1.06)
on p. 281 of Ref. [9]]
∞
t exp(xt) tm
= ∑ Bm (x) , ∣t∣ < 2π . (8.165)
exp(t) − 1 m=0 m!
The Bm = Bm (0) are the Bernoulli numbers, for which we can indicate some example
cases,
1 1 1 1
B1 = − , B2 = , B3 = 0 , B4 = − , B5 = 0 , B6 = ,.... (8.166)
2 6 30 42
Advanced Topics 329
f (x) 6
s
s s
s s
s
s s
s s s s s
-x
N N +1 N +2 N +3 N +4 N +5 M −1 M
Fig. 8.3 Illustration of Eq. (8.164).
Geometrically, it is clear that the expression 12 f (N ) + 12 f (M ) + ∑M−1

n=N +1 f (n) ap-
M
proximates the integral ∫N dx f (x) according to the trapezoid rule. The remainder
term obtained by forming the difference is then expressed, by the Euler–Maclaurin
formula, as a sum over expressions containing higher-order derivatives of the func-
tion at the boundaries of the domain of integration. In practical applications, the
assumption is that the remainder term Rq on the right-hand side of Eq. (8.164b)
tends to zero as q → ∞, and that the first few terms on the right-hand side of
Eq. (8.164a) yield increasingly better approximations to the remainder term given
by the difference of the sum and the integral over n.
For our case [see Eq. (8.162)], we identify N → 0, M → ∞, f → Fη and n →
∞
nz . The convergence of the integral ∫0 dnz Fη (nz ) is ensured by the exponential
damping at large nz given by the regulator. The same exponential suppression of
the contribution from large nz is responsible for the fact that all derivatives of the
function Fη (nz ) vanish in the limit nz → ∞. We can thus rewrite the Eq. (8.162)
considering only the derivatives of the function Fη (nz ) at nz = 0,
∞ ∞ ∞
1
1
F (0) +
2 η ∑ Fη (nz ) − ∫ dnz Fη (nz ) = − ∑ B2j Fη(2j−1) (0)
nz =1 0 j=1 (2j)!
1 1 ′′′
B2 Fη′ (0) − B4 Fη (0) − . . . .
= −
2! 4!
(8.167)
We have expressed the renormalized vacuum energy shift as
Eren (R) ̵ c π2
h 1 1 ′′′
= lim [− B2 Fη′ (0) − B4 Fη (0) − . . . ] . (8.168)
L2 η→0 4 R3 2! 4!
The evaluation of the derivatives now proceeds using the representation for Fη (nz )
given in Eq. (8.163), and we recall here for convenience this definition as well as the
definition of fη (k) given in Eq. (8.155b),
∞ √ π√
Fη (nz ) = ∫ dv v fη ( v) , fη (k) = exp(−η k). (8.169)
n2z R
The differentiation with respect to n2z can easily be reformulated as a differentiation

with respect to nz ,
∂Fη (nz ) π 1 ∂Fη (nz )
= −nz fη ( nz ) = , (8.170)
∂n2z R 2nz ∂nz
and so
∂Fη (nz ) π πη
= −2n2z fη ( nz ) = −2n2z exp (− nz ) . (8.171)
∂nz R R
This means that
π
Fη′ (nz ) = − 2n2z exp (−η nz ) , (8.172a)
R
′′ 2n2z πη πη
Fη (nz ) = (−4nz + ) exp (− nz ) , (8.172b)
R R
′′′ 8nz πη 2n2z π 2 η 2 πη
Fη (nz ) = (−4 + − ) exp (− nz ) , (8.172c)
R R2 R
′′′′ 12πη 12nz π 2 η 2 2n2z π 3 η 3 πη
Fη (nz ) = ( − 2
+ 3
) exp (− nz ) , (8.172d)
R R R R
′′′′′ 24π 2 η 2 16nz π 3 η 3 2n2z π 4 η 4 πη
Fη (nz ) = (− 2
+ 3
− 4
) exp (− nz ) . (8.172e)
R R R R
We have performed all integrations and differentiations with a finite cutoff η. The
actual zero-point energy as well as the counter term have been regularized, and the
regularization parameter η can now be removed at the very end of the calculation.
Namely, we can easily read off the results
′′ ′′′
lim Fη′ (0) = lim Fη (0) = 0 , lim Fη (0) = −4 , lim Fη(j≥4) (0) = 0 .
η→0 η→0 η→0 η→0
(8.173)
All higher derivatives vanish because further powers of η are generated by the dif-
ferentiation of the exponential function. Finally, after removing the regularization,
we obtain the Casimir energy shift as
Eren (R) ̵ c π2 1
h ′′′ ̵ c π2 1
h 1 h̵ c π2
= − B 4 lim Fη (0) = − (− ) (−4) = − . (8.174)
L2 4 R3 4! η→0 4 R3 24 30 720 R3
The energy shift is negative, corresponding to an attractive interaction, which con-
firms the intuitive picture regarding the attractive force between ships on the open
sea. The Casimir force per area (Casimir pressure) on the plate at z = R is conse-
quently negative (attractive) and amounts to
Fren (R) ∂ Eren (R) ̵ c π2
h
2
=− 2
=− . (8.175)
L ∂R L 240 R4
Our brief investigation of the Casimir effect familiarizes us with the ideas of reg-
ularization and renormalization, which are extremely important in quantum field
theory.
Advanced Topics 331
8.5 Classical Potentials and Renormalization
8.5.1 Potential of a Uniformly Charged Plane

We attempt to understand the concepts of regularization and renormalization on
the level of the classical theory. To this end, we consider a uniform plane sheet
of charge, with a surface charge density σ0 , lying in the z = 0 plane. The goal is
to determine the value of the potential at r⃗ = x êz + y êy + z êz with the condition
that the potential is to vanish at the surface. The differential surface area for this
problem is dS ′ = dx′ dy ′ . This can be used in Eq. (2.24),
1 ρ(⃗r′ ) 3 ′ 1 σ0
r) =
Φ(⃗ ∫ ′
d r = ∫ dx′ dy ′ , (8.176)
4π0 ∣⃗
r − r⃗ ∣ 4π0 r − r⃗′ ∣
∣⃗
to obtain the required integral
Φ (x, y, z) = Φ (z)
1 ∞ ∞ σ0
= ∫ ∫ √ dx′ dy ′ + Φ0
4π0 −∞ −∞ (x′ )2 + (y ′ )2 + z 2
σ0 ∞ 1
= ∫ √ 2πρ dρ + Φ0 . (8.177)
4π0 0 ρ2 + z 2
This integral is divergent. It even diverges linearly for large R. In order to handle
this difficulty, we note that the potential is independent of x and y. We assume that
the surface has a finite, but large radius R centered on the z axis. The radius R
acts as a cutoff for the xy integration, and it respects the symmetry of the problem.
The regularized potential is
σ0 R ρ
Φ (z) = ∫ √ 2 dρ + Φ0 (R)
20 0 ρ + z2
σ0 √ 2
= [ R + z 2 − ∣z∣] + Φ0 (R) . (8.178)
20
Here, we have used the result
∂ √ 2 1 2ρ ρ
ρ + z2 = √ =√ . (8.179)
∂ρ 2 ρ2 + z 2 ρ + z2
2
We have assumed that the counter term Φ0 (R) is a function of R only, and does
not depend on z. It, therefore, is physically unobservable. We can now fix a
“renormalization condition” by requiring the potential to acquire a specific value at
a specific point, because we know that only potential differences matter. We choose
the condition that the the potential be zero at z = 0. This requires that Φ0 (R) =
− 2
σ0
0
R. We can finally remove the regularization at the end of the calculation, and
let R go to infinity,
⎡ √ ⎤
σ0 ⎢ 2 ⎥
⎢R 1 + z − ∣z∣ − R⎥ = − σ0 ∣z∣ .
Φ (z) = lim ⎢ ⎥ (8.180)
R→∞ 20 ⎢ R2 ⎥ 20
⎣ ⎦
We have used the fact that

√ z2 z2
lim R2 + z 2 − R = lim R (1 + 2
) − R = lim = 0. (8.181)
R→∞ R→∞ 2R R→∞ 2R
The resulting electric field is directed along the z axis, as expected. The electric
field is
⎧
⎪
σ0
(z > 0)
⎪
⎪
⎪
êz
⃗ ⃗ =⎨ 2
E(z) = −∇Φ 0
. (8.182)
⎪
⎪ −êz
σ0
(z < 0)
⎪
⎪
⎩ 20
This behavior is consistent with intuition. We can unify the two above formulas as
⃗ σ0
E(z) = êz sgn(z) , (8.183)
20
where the sign function is defined as sgn(z) = 1 for z > 0 and sgn(z) = −1 for
z < 0. The value sgn(0) is fixed to be zero by convention. The derivative of the sign
function is
d sgn(z)
= 2 δ(z) . (8.184)
dz
We can thus verify that
⃗ ∂ σ0 1 1
⃗ ⋅ E(z)
∇ = Ez (z) = 2 δ(z) = σ0 δ(z) = ρ(x, y, z) . (8.185)
∂z 20 0 0
We convince ourselves that the integral for the electric field strength is convergent,
∞ ∞
σ0 1 z − z′ σ0 1 z
Ez (z) = ∫ ′ ′
dx′ dy ′ = ∫ 2 √ 2πρ dρ
4π0 ∣⃗
r − r⃗ ∣ ∣⃗
2 r − r⃗ ∣ 4π0 ρ +z 2
ρ2 + z 2
0 0
∞ ρ2 =∞
σ0 z ρ σ0 z 2 σ0 z
= ∫ dρ = [− 2 ] = . (8.186)
20 (ρ2 + z 2 )3/2 40 (ρ + z 2 )1/2 ρ2 =0 20 ∣z∣
0
We thus reproduce the previously obtained result.
8.5.2 Potential of a Uniformly Charged Long Wire

Let us consider an infinitely long, charged wire that extends along the x axis, and
let us try to evaluate the electrostatic potential a distance z away from the wire.
By symmetry, the potential can only be a function of z. We start from Eq. (2.24),
1 ρ(⃗r′ ) 3 ′
r) =
Φ(⃗ ∫ d r . (8.187)
4π0 r − r⃗′ ∣
∣⃗
The charge density ρ which corresponds to a long charged rod along the x-axis,
reads as
r′ ) = λ δ(y ′ ) δ(z ′ ) ,
ρ(⃗ (8.188)
Advanced Topics 333
where λ is the charge per unit length along the wire. The potential is to be evaluated
at the point r⃗ = (0, 0, z), where it reads as follows,
λ ∞ 1 λ ∞ 1
Φ(z) = ∫ √ dx′ = ∫ √ dx′ . (8.189)
4π0 −∞ z 2 + (x′ )2 2π0 0 z 2 + (x′ )2
Observe that the integration limits have changed. This expression is formally
infinite, but the divergence at the upper limit of the integration domain is only
logarithmic. First we regularize this integral by introducing an upper cutoff for the
x′ -integration, which, again, respects the symmetry of the problem,
λ R 1
Φ(z) = ∫ √ 2 dx′ , (8.190)
2π0 0 z + (x′ )2
where R is assumed to be large (at the end of the calculation, we let R → ∞). Be-
cause the integral diverges, we have to subtract a counter term. We cannot impose
the renormalization condition Φ(z = 0) = 0 because the regularized integral would
otherwise be divergent at zero. Fixing the potential to be zero at an intermediate
point z = z0 , we can devise the following counter term,
λ R 1
C= ∫ √ 2 dx′ (8.191)
2π0 0 ′
z0 + (x )2
which also is infinite in the limit R → ∞ and independent of z. In view of the result
′ 1 √
∫ dx √ 2 = ln(z + z 2 + x′2 ) , (8.192)
z + x′2
the renormalized expression,
λ R⎛ 1 1 ⎞ ′
Φ(z) = lim ∫ √ −√ dx
2π0 R→∞ 0 ⎝ z + (x ) 2 ′ 2 z0 + (x ) ⎠
2 ′ 2
λ √ √
= lim (ln(R + R2 + z 2 ) − ln(R + R2 + z02 ) − ln(∣z∣) + ln(∣z0 ∣))
2π0 R→∞
λ ∣z∣
= − ln ( ) (8.193)
2π0 ∣z0 ∣
is finite in the limit R → ∞ (removal of the cutoff). By differentiation, we obtain
∂ λ ∂ ∣z∣ λ
Ez (z) = − Φ(z) = ln ( )= , (8.194)
∂z 2π0 ∂z ∣z0 ∣ 2π0 z
which is positive for z > 0 and negative for z < 0. The integral defining the field
strength is
λ ∞ z λz 1 λ
Ez (z) = ∫ dx′ = = . (8.195)
′
4π0 −∞ (z + (x ) )
2 2 3/2 2π0 ∣z∣2 2π0 z
This confirms the result based on the evaluation of the potential. In field theory, a
renormalization condition formulated at a finite, but nonvanishing, z0 is otherwise
known as an “intermediate renormalization.”
8.5.3 Charged Structures in 0.99 and 1.99 Dimensions

Let us try to generalize the above approach. A charged, long wire is a one-
dimensional structure, whereas a charged plane sheet is a two-dimensional structure.
We investigate the question of whether it is possible to perform the regularization
of the above two examples by investigating charged structures in 0.99 and 1.99 di-
mensions, i.e., by slightly displacing the dimension of the structure from an integer
dimension, using so-called dimensional regularization.
We still operate in three spatial dimensions, i.e., the Coulomb law is unchanged.
Only the dimensionality of the charged structure is changed. We first need the
volume ΩD of the D-dimensional unit sphere. The matching of the result for integer
dimensions to the result for non-integer dimensions is done as follows. On the one

hand, if we integrate all variables which span D one by one, for a “Gaussian bell”
function, then we obtain
∞ dk D √ D
dD k −k⃗2 −k2 π 1 π D/2
∫ D e = [ ∫ e ] = ( ) = = . (8.196)
Ê (2π)D −∞ 2π 2π 2D π D/2 (2π)D
On the other hand, if we single out the integral over the angular variable ∫ dΩd ,
then we have
dD r −⃗r2 1 ∞ Γ(D/2)
dr rD−1 e−r =
2
∫ D e = [∫ dΩD ] ∫ [∫ dΩD ] ,
Ê (2π) D (2π)D 0 2(2π)D
(8.197)
∞
where we recall that the Gamma function is defined as Γ(x) = ∫0 tx−1 exp(−t) dt,
for x > 0. So, we have
2 π D/2
ΩD = ∫ dΩD = . (8.198)
Γ(D/2)
This result takes the following values,
⎧
⎪ 2 (D = 1)
⎪
⎪
ΩD = ⎨ 2π (D = 2) . (8.199)
⎪
⎪
⎪
⎩ 4π (D = 3)
One may ask about the origin of the result ΩD=1 = 2. The reason is that we attempt
to replace an integral over the entire D-dimensional space ∫ÊD dD r with an integral
∞
of the form proportional to ∫0 dr rD−1 over a radially symmetric function. Then,

in one dimension, for a function f (x) = f (−x), with x ∈ , we have a factor two
because the regimes x ∈ (−∞, 0) and x ∈ (0, ∞) have to be treated separately.
For a radially symmetric integrand f (⃗r) = f (r), one has
∞
∫ dD r f (⃗
r ) = ΩD ∫ dr rD−1 f (r) , (8.200)
Ê D 0
in a general dimension D (where r = ∣⃗r ∣ and the function f is defined accordingly,
depending on the scalar or vector character of its argument). In a non-integer
dimension D, slightly displaced from the integer n by ε,
D = n −ε, (8.201)
Advanced Topics 335
one has to restore the physical dimensionality of the integrand by the introduction
of a reference length scale which we denote by z0 and which is not necessarily equal
to the reference length scale z0 used for the renormalization of the charged long wire
used above. (For n = 1 and n = 2, and ε = 0.01, we verify the title of the current
section.) The appropriate replacement which preserves the physical dimension, is
rD → rn−ε z0ε = rD z0ε . (8.202)
We denote the uniform charge density in n dimensions as Ξ. For one dimension, we
have Ξ = λ, while Ξ = σ in two dimensions. Finally,
1 ∞ Ξ
Φ(z) = ΩD ∫ dr rD−1 z0ε √ , D = n −ε. (8.203)
4πε0 0 r2 + z 2
We now consider the integral (z0 > 0)
∞ 1 1
J(D) = ΩD ∫ dr rD−1 z0ε √ = π 2 (D−1) Γ ( 12 (1 − D)) z0ε ∣z∣D−1 , (8.204)
0 r2 + z 2
which is strictly valid only for D < 1, because we otherwise have a divergence at
the upper limit of integration. However, we courageously assume the validity of
Eq. (8.204) even for general D ≥ 1 by analytic continuation.
For D = 1 − ε, an expansion for “almost integer D = 1” results in
2 ∣z∣
J(D = 1 − ε) = π −ε/2 (z0 ) ∣z∣−ε Γ ( 12 ε) =
ε
− γE − ln(π) − 2 ln ( ) + O(ε) . (8.205)
ε z0
In the expansion for small ε, the “replica trick”
xε = exp [ε ln(x)] = 1 + ε ln(x) + O(ε2 ) (8.206)
and the formula
1
− γE + O(ε)
Γ(ε) = (8.207)
ε
prove useful. Here, γE is the Euler–Mascheroni constant γE = 0.57712 . . . . The
complete potential is thus obtained as (D = 1 − ε)
λ 1 λ λ ∣z∣
Φ(z) = − (γE + ln(π)) − ln ( ) + O(ε)
2πε0 ε 4πε0 2πε0 z0
λ 1 λ ∣z∣ √ γE /2
= − ln ( πe ) + O(ε) . (8.208)
2πε0 ε 2πε0 z0
We have evaluated the integral (8.208) without having a concrete renormaliza-
tion condition at hand. In the so-called MS (minimal subtraction) renormalization
scheme, one now simply throws away the divergent term of order 1/ε and writes
λ ∣z∣ z0
Φ(0, 0, z) = − ln ( ) , z̃0 = √ . (8.209)
2πε0 z̃0 π eγE /2
A posteriori, i.e., after evaluating the integral, we see that the MS scheme corre-
sponds to a renormalization condition where the potential is required to be zero
at a distance z̃0 from the wire, where z̃0 is not equal to the originally introduced
parameter z0 , but differs only by a multiplicative constant.
In the modified minimal subtraction MS scheme (“MS-bar scheme”), we would
also throw away the constant terms in Eq. (8.208), i.e., those with the Euler–
Mascheroni constant and the logarithm of π. In this case, the meaning of z0 is
the same as the one used in the calculation described in Sec. 8.5.2. In any case, if
one uses the MS or MS scheme, one has to work out the renormalization condition
fulfilled by the calculated potential a posteriori. In field theory, one then has to
adjust the renormalization scale z0 that determines the “coupling constant” to the
appropriate scale, using a multiplicative renormalization. This is beyond our scope
here; however, the principle is clear.
For D = 2 − ε, the expansion is
1 1 √ √
J(D = 2 − ε) = π 2 − 2 ε (z0 ) (∣z∣)
ε 1−ε
Γ ( 12 (ε − 1)) =
π ∣z∣ (−2 π) = −2π ∣z∣ + O(ε) .
(8.210)
It is characteristic of dimensional regularization that only logarithmic divergences
give rise to divergent terms in the dimensional analytic continuation variable ε.
Consequently, expanding the potential (8.204) using the result (8.203) about ε = 0
for D = 2 − ε, one recovers the result
σ
Φ (z) = − ∣z∣ + O(ε) , D = 2 − ε, (8.211)
2ε0
where we have used the fact that Ξ has the meaning of the surface charge density
σ in two dimensions. This is the result originally derived in Eq. (8.180).
8.6 Open Problems in Classical Electromagnetic Theory
8.6.1 Abraham–Minkowski Controversy

We shall conclude this book with a brief digression on physical problems which con-
stitute (at least partially) open problems and leave room for further investigation.
One of the most basic open questions in classical electrodynamics, which still has
not been conclusively answered, concerns the proper definition of the Maxwell stress
tensor (see Sec. 1.3.2) in a dielectric. It is commonly known as the Abraham1 –
Minkowski2 controversy. The classic literature references are [33, 34]. Let us try
to shine some light onto the discrepancy between the two formulations. We recall
from Eq. (1.122) that the force density acting on an electromagnetic medium can
be written as [see Eq. (1.122)]
∂ ⃗
f⃗ = ∇
⃗⋅ − S, ⃗ ×B
S⃗ = 0 E ⃗, (8.212)
∂t
1
Max Abraham (1875–1922)
2
Hermann Minkowski (1864–1909)
Advanced Topics 337
where we recall that f⃗ denotes the force density, and the components of the stress
tensor are [see Eq. (1.119)]
1 1 1
ij δij E⃗ 2 ) +
= 0 (Ei Ej − ⃗2) .
(Bi Bj − δij B (8.213)
2 μ0 2
The scaled Poynting vector S⃗ differs from S⃗ used in Eq. (1.122) by a prefactor 1/c2 ;
the scaled form allows for simpler prefactors in the generalization to the case of
dielectric. Indeed, one might imagine two possible generalizations of the Poynting
vector,
1 ⃗ ⃗
⃗ ×B
S⃗ → S⃗M = D ⃗, S⃗ → S⃗A = 2 E ×H, (8.214)
c
both of which coincide with the original form S⃗ = 0 E⃗×B ⃗ in vacuum, i.e., for
r = μr = 1. Of course, the superscripts M and A refer to Minkowski and Abraham.
The Minkowski form of the stress tensor is
1 ⃗ ⃗ ⃗ ⃗
ij = Ei Dj + Hi Bj −
M
(E ⋅ D + H ⋅ B) δij . (8.215)
2
It is not symmetric under the interchange i ↔ j, but one can show that
∂ ⃗M
f⃗ = ∇
⃗⋅S , M
− (8.216)
∂t
which carries a formal resemblance with Eq. (8.212). The formal resemblance
of Eqs. (8.216) and (8.212) is a primary argument in favor of the Minkowski
formulation.
Let us now dwell on Abraham’s alternative formalism. First, we mention
that, based on the phenomenological Maxwell equations (6.86), one may derive the
formula,
∇ ⃗⋅ ∂ D
⃗ ⋅ J⃗0 − E
⃗ ⋅ S⃗A = −E ⃗⋅ ∂B
⃗ −H ⃗, (8.217)
∂t ∂t
which carries a formal resemblance with Eq. (1.106) and would suggest that the
Abraham form of the field momentum should be used. Note, however, that we
cannot simply define the energy density as
⃗ ⋅B
u=H ⃗+E
⃗⋅D
⃗, (8.218)
because the time derivative contains mixed terms,
⃗ ⋅B
∂t u = ∂t H ⃗ +H
⃗ ⋅ ∂t B
⃗ + ∂t E
⃗⋅D
⃗ + E⃗ ⋅ ∂t D
⃗. (8.219)
In this case, the Abraham form of the force law is
∂ ⃗A
f⃗A = ∇
⃗⋅ A
− S , (8.220)
∂t
where f⃗A is the Abraham force density, which differs from the Lorentz force density
f⃗. For an isotropic and dispersionless medium, one can show that [35]
1 ∂ ⃗ ⃗
f⃗A = f⃗ + 2 (r − 1) E ×H. (8.221)
c ∂t
The additional term in comparison to the Lorentz force is sometimes called the
Abraham force. Despite considerable effort, a final word on the subject matter
seems to be lacking.
√
Quantum mechanically, for a refractive index n = r , one can show that the
Minkowski momentum density corresponds to a photon momentum
̵ω
nh
pM = , (8.222)
c
while the Abraham version reads as
̵ω
h
pA = . (8.223)
nc
A relatively recent proposal concerning the detection of the Abraham force with a
succession of short optical pulses has been given in Ref. [36]. The controversy still
is an active field of research, even if the Abraham formulation is generally preferred
in the literature [19, 37].
8.6.2 Relativistic Dynamics with Radiative Reaction
When a charged particle is accelerated, it radiates off electromagnetic fields and,

therefore, field energy. The backreaction of the electromagnetic field onto the emit-
ting particle is known as the radiative reaction. One might think that it would
be very difficult to find a formula that connects the backreaction of the radiation
onto the emitting particle. However, by equating the energy loss of the particle
with the radiated power, one is actually led to a very intuitive ansatz for the back-
reaction force F⃗rad . In the nonrelativistic approximation, the derivation is quite
straightforward. Indeed, the problem continues to attract considerable attention,
even after a number of physicists have already spent considerable time and effort
analyzing its intricacies (for a necessarily incomplete list of literature references, see
Refs. [38–48]).
In the nonrelativistic approximation, the power dissipated by an accelerated
particle of elementary charge e, dissipated into the radiation, is given by the Larmor
formula
2 2αh̵ 2 α λ̄e
Prad = m τrad v⃗˙ , τrad = = = 6.26 × 10−24 s . (8.224)
3 m c2 3c
⃗ = v⃗˙ is the acceleration, and τrad is the characteristic emission time for the
Here, a
radiation. Note that the radiation damping time τrad depends on the mass of the
radiating particle. In Eq. (8.224), the numerical result for τrad is valid provided m
denotes the electron mass. In the nonrelativistic limit, Eq. (8.224) was originally
derived by Lorentz [49].
We do not derive Eq. (8.224) here, but we can motivate the equation by matching
the time-averaged (over an oscillation period), radiated power of a radiating dipole
Advanced Topics 339
given in Eq. (5.109) with the momentous kinetic energy of the oscillating dipole,
2c k ⟨⃗
p02 ⟩ 2 ω ⟨⃗
p02 ⟩
4 4
c k 4 p⃗02
⟨Prad ⟩ = → = 3
3 4π0 3 4π0 3c 4π0
2 e2 ω ⟨⃗
r 2⟩
4
2 e2 v⃗˙ 2 2 α λ̄e ˙ 2
→ 3
→ 3
=m v⃗ = m τrad v⃗˙ 2 . (8.225)
3c 4π0 3 c 4π0 3c
The justification for the replacements is as follows: (i) The maximum dipole mo-
ment p⃗02 is equal to twice the time-averaged dipole moment ⟨⃗ p02 ⟩, (ii) the average
dipole moment is related to position as ⟨⃗ p02 ⟩ = e2 ⟨⃗r 2 ⟩. In the sense of a derivative
in Fourier space, we can then replace in (iii) the expression ω → i ∂t , to obtain
the replacement ω 4 ⟨⃗ r 2⟩ → a⃗2 , where a
⃗ = v⃗˙ = r¨⃗ is the acceleration, whose mod-
ulus does not depend on time for a harmonic oscillator. Assuming, then, that
the relation is valid at any given time, i.e., for a momentous acceleration, we
arrive at the formula for the radiated power P . The fine-structure constant is
̵ 0 c) ≈ 1/137.036.
α = e2 /(4π h
The expression (8.224) for Prad is valid for the energy radiated off from the
particle. By energy conservation, the radiated power has to be compensated by the
action of the radiative-reaction force F⃗rad , so that
t2 t2 t2
∫ F⃗rad ⋅ v⃗ dt = − ∫ Prad dt = − ∫ m τrad v⃗˙ ⋅ v⃗˙ dt . (8.226)
t1 t1 t1
Provided boundary terms can be neglected, one obtains the result

t2
∫ (F⃗rad − m τrad v¨⃗) ⋅ v⃗˙ dt = 0 , F⃗rad = m τrad v¨⃗ . (8.227)
t1
The radiative backreaction force F⃗rad depends on the time derivative of the acceler-
ation, and the solution of the corresponding equation of motion therefore requires
an additional boundary condition. Indeed, the equation of motion
F⃗ = m v⃗˙ = F⃗ext + F⃗rad = F⃗ext + m τrad v¨⃗ , (8.228)
for the particle involves the external force F⃗ext and the backreaction force F⃗rad . For
a vanishing force F⃗ext = ⃗
0, one has a runaway solution,
F⃗ = m a
⃗(t) = m τrad a
⃗˙ (t) , (8.229a)
a ⃗0 exp(t/τrad ) ,
⃗(t) = a v⃗(t) = a
⃗0 τrad exp(t/τrad ) + v⃗0 , (8.229b)
v⃗(t) = a
⃗0 τrad
2
exp(t/τrad ) + v⃗0 t + r⃗0 . (8.229c)
If we postulate that the kinetic energy of the particle needs to remain finite forever,
then we can eliminate the runaway solution, but this comes at a price: One can
show [47] that the solutions which preserve the finiteness of the energy suffer from
problems of causality, namely, the particle would accelerate before the external
force is turned on. One can eliminate the runaway solution in a different manner,
by replacing F⃗rad → m a
⃗(t), to write
d
F⃗ = m a
⃗ = F⃗ext + τrad ( F⃗ext ) , (8.230)
dt
in which case the right-hand side remains zero in the case of a vanishing external
force F⃗ext . This is equivalent, within the nonrelativistic approximation, to the
transition from the Abraham–Lorentz to the Landau–Lifshitz formulation of the
backreaction force, which will be discussed in the following.
The first steps toward a relativistic formulation of backreaction were performed
by Dirac [40]. One uses the four-velocity β μ and the proper time τ ,
1/2
v⃗ v2
β μ = (γ, γ ) , dτ = (1 − ) dt . (8.231)
c c2
The Lorentz–Abraham–Dirac equation is obtained [48] as a natural generalization
of Eq. (8.227) to the relativistic domain,
dβμ d2 β ν
m = qFμν β ν + τrad (gμν − βμ βν ) . (8.232)
dτ dτ 2
The replacement which leads to the “Landau–Lifshitz” form (8.230) corresponds to
the replacement
d2 β ν d dβ ν d q νρ
≈ → F βρ , (8.233)
dτ 2 dτ dτ dτ m
where the latter form corresponds to the zeroth-order approximation (without the
radiative reaction term). Indeed, in Chap. 76 of Ref. [50], one finds the Landau–
Lifshitz equation
dβμ d
m = q Fμν β ν + τrad (gμν − βμ βν ) [q F μρ βρ ] . (8.234)
dτ dτ
This equation has no runaway solutions, in contrast to Eq. (8.232).
The Lorentz–Abraham–Dirac equation (8.232) is sometimes written as follows,
dβμ d2 βμ dβ ν dβν μ
m = qFμν β ν + τrad ( 2 + β ), (8.235)
dτ dτ dτ dτ
which corresponds to an integration by parts of the terms in Eq. (8.232). The
“Landau–Lifshitz” version of this equation [44, 46, 51] reads as follows,
dβμ q q2
m = q Fμν β ν + τrad [ β α (∂α Fμν ) β ν + 2 Fμν F νρ βρ
dτ m m
q2
+ (F να βα ) (Fνβ β β ) βμ ] . (8.236)
m2
Advanced Topics 341
This equation is sometimes given in a slightly different form in the literature. E.g.,
in comparison to Eq. (2) of Ref. [46], one replaces q → −e and takes notice of
the fact that in the second term in square brackets in Eq. (8.236), the author of
Ref. [46] interchanges the order of indices and uses the antisymmetry of the field
strength tensor. In Refs. [44, 51], it is argued that on the critical manifold, i.e., on
the manifold of solutions of the Landau–Lifshitz equation which fulfill “reasonable”
boundary conditions and which are, therefore, unstable against any small perturba-
tions which would otherwise lead to runaway solution, the Landau–Lifshitz version
of the radiative reaction describes the physical trajectories correctly. Furthermore,
it is argued that the trajectory described by the Landau–Lifshitz equation repre-
sents an “unstable fixed point” of the manifold described by the trajectories under
a renormalization group analysis, and that any modifications beyond the Landau–
Lifshitz equation are small for any realistic value of the field strength. However, in
Ref. [47], it is argued that the “truncation” of the Abraham–Lorentz–Dirac equation
in the sense of Landau and Lifshitz can lead to physically nonsensical trajectories
near the onset of time-dependent perturbations, and this statement is illustrated
on a number of example cases.
Alternatively, one can point out that, for reasons of principle, the problem can-
not be solved on the level of quantum mechanics and requires the added power
of quantum electrodynamics, where the “photon recoil”, i.e., the backreaction of
radiation emission onto the energy and the momentum of the emitting particle, is
being taken into account at all interaction vertices without further approximations.
However, for an electron moving in a strong, external, classical field, it would be im-
possible to describe all interactions to all orders, using the formalism of QED, and
therefore, the classical models are still useful. Lorentz [49] traced the mechanism
of the backreaction to the fact that a spatially extended object emitting radiation
actually cannot fulfill Newton’s third law, but one has to take into account the
backreaction of the emitted radiation onto itself. Within the nonrelativistic ap-
proximation, for a spherical radiation emitting object of radius R, the result for the
self-reaction force F⃗self can be expressed as follows,
mτrad 2R
F⃗self = 2
[⃗
v (t − ) − v⃗(t)] ,
2R c
2R 2R ˙ 1 2R 2 ¨
v⃗ (t − ) = v⃗(t) − v⃗(t) + ( ) v⃗(t) + . . . ,
c c 2 c
mcτrad
Fself = − ⃗ + mτrad a
a ⃗˙ ,
R
m c τrad
⃗ = F⃗self + F⃗ext = −
ma ⃗˙ + F⃗ext ,
⃗ + mτrad a
a
R
m c τrad
(m + ⃗ = F⃗ext + mτrad a
)a ⃗˙ = F⃗ext + F⃗rad . (8.237)
R
This equation looks suspiciously similar to Eq. (8.228), but with a “renormalized”
mass
m c τrad
m→m+ . (8.238)
R
In some sense, taking the backreaction into account, we find a “dynamical” correc-
tion to the electron’s mass, which anticipates the mass renormalization inherent to
the field-theoretical formulation of QED.
One can also attempt to include the spin precession of the electron in the
external field, which results in the Bargmann–Michel–Telegdi equation described in
Ref. [52]. A further development is the Barut–Zanghi equation derived in Ref. [53].
By applying the Landau–Lifshitz prescription to the Barut–Zanghi equation, which
eliminates runaway solutions to the Lorentz–Dirac equation, the authors of Ref. [48]
arrive at a variant of the Bargmann–Michel–Telegdi equation which does not have
runaway solutions. Indeed, the authors of Ref. [48] argue that, at least for spin-
polarized orbits in a constant magnetic field, the orbits of their magnetically inter-
acting electron theory (with spin) decay as physically expected. They also point
out that it should be possible to test the equations using a small (few MeV) electron
synchrotron or a muon storage ring.
We should point out some open problems. Indeed, in Ref. [48], no attempt is
made to rigorously derive the proposed “Landau–Lifshitz form” of the Bargmann–
Michel–Telegdi equation, or, to analyze its properties on the critical manifold, as
done for the “Landau–Lifshitz form” of the Lorentz–Abraham–Dirac equation (with-
out spin) in Refs. [44, 51]. This constitutes an open problem. The next step would
then be to include the “classical spin” in the equation of motion of an electron accel-
erated by a laser field; this would otherwise generalize the considerations reported
in Ref. [46].
8.6.3 Electrodynamics in General Relativity
According to classical nonrelativistic physics, the coordinates assigned to an event

by different observers are connected by Galilei transformations. Time, in par-
ticular, is absolute. By contrast, according to special relativity, time and space
transform into each other, but still, in flat space, once the relative velocity of two
observers is known, no matter how far they are apart, they can uniformly trans-
form the space and time coordinates assigned to an event by a global Lorentz
transformation. In special relativity, time no longer is an absolute quantity. In
general relativity, the situation again is different: Time and space, locally, have
a Lorentzian geometry, i.e., locally, it is always possible to choose the space-time
metric as g μν = diag(1, −1, −1, −1). However, it is not possible to extend this local
reference frame throughout the Universe. Space-time is curved, and one gives up,
in particular, the Euclidean postulate that two (locally) parallel curves never can
cross each other at infinity.
Advanced Topics 343
From differential geometry, we know that Christoffel symbols enter the definition
of the covariant derivative, which is a generalized derivative adapted to curved
geometries. For a position-dependent metric g μν (x), the covariant components
Aμ (x) are connected to the contravariant components Aμ (x) of a vector field Aμ (x)
by the relations
xμ = g μν (x) xν , xμ = g μν (x) xν ,
Aμ (x) = g μν (x) Aν (x) , Aμ (x) = g μν (x) Aν (x) . (8.239)
From now on, we suppress the dependence of the metric on the coordinates and
just write gμν for g μν (x) for the curved-space metric, in order to distinguish it from
the flat-space metric given in Eq. (8.30). We also employ units with h ̵ = c = 0 = 1,
anticipating the natural unit system used in high-energy physics.
Let us briefly motivate the definition of the covariant derivative. First of all, we
recall that a vector A⃗ actually is independent of a coordinate system. A local basis
consists of the set of unit vectors e⃗μ ,
A⃗ = Aμ e⃗μ , e⃗μ ⋅ e⃗ν = gμν , e⃗μ ⋅ e⃗ν = g μ ν = δ μ ν . (8.240)
Because the basis vectors e⃗μ are non-constant, an additional term emerges as one
calculates the derivative of the vector A⃗ with respect to a coordinate,
∂ ⃗
A = ∂ν A⃗ = (∂ν Aμ ) e⃗μ + Aμ (∂ν e⃗μ ) = (∂ν Aμ ) e⃗μ + Aμ e⃗λ g λρ (⃗
eρ ⋅ ∂ν e⃗μ )
∂xν
= (∂ν Aμ ) e⃗μ + Aμ e⃗λ g λρ Γρ μ ν = (∂ν Aμ ) e⃗μ + Aμ e⃗λ Γλ μ ν , (8.241)
where Γρ μ ν = e⃗ρ ⋅ ∂ν e⃗μ is a Christoffel symbol, and the Einstein summation conven-
tion has been used (all the Greek indices are summed from zero to three). We now
swap the dummy indices (summation) in the second term according to (λ ↔ μ) and
identify the components of the covariant derivative,
∂ν A⃗ = e⃗μ [(∂ν Aμ ) + Γμ λ ν Aλ ] , ∇ν Aμ = ∂ν Aμ + Γμ λ ν Aλ . (8.242)
While the Christoffel symbols Γμ λ ν are not tensors, one can still raise the first index
with the metric,
∂⃗
eμ
Γρμν = e⃗ρ ⋅ , Γβ μν = gβρ Γρμν . (8.243)
∂xν
A rather straightforward calculation shows that
1 ∂gβμ ∂gβν ∂gμν
Γβμν = ( + − ). (8.244)
2 ∂xν ∂xμ ∂xβ
By inspection, one can easily derive the symmetry properties
Γβμν = Γβνμ , Γβ μν = Γβ νμ . (8.245)

The covariant derivative of a vector field Aμ (x) has the properties,

∂
∇μ Aν = ∂μ Aν + Γν μβ Aβ , ∂μ ≡ , (8.246a)
∂xμ
∇μ Aν = ∂μ Aν − Γβ μν Aβ , (8.246b)
g βν ∇μ Aν = ∇μ (g βν Aν ) = ∇β Aν , (8.246c)
i.e., the covariant derivative transforms as a tensor (unlike the partial derivative),
and its indices can therefore be raised and lowered with the metric, both before as
well as after the covariant differentiation. The Riemann curvature tensor Rρ σμν is
of rank four and can be expressed in terms of the Christoffel symbols,
Rρ σμν = ∂μ Γρ νσ − ∂ν Γρ μσ + Γρ μλ Γλ νσ − Γρ νλ Γλ μσ . (8.247)
For the Ricci tensor Rμν , one contracts the first and the third index of the Riemann
curvature tensor,
Rμν = Rα σαν , R = gμν Rμν = Rμ μ . (8.248)
The scalar curvature R of space-time is obtained as the contraction of the Ricci

tensor. The Einstein–Hilbert action SEH is given by
√ δSEH 1
SEH = ∫ d4 x − det g R , αβ
= Rαβ − gαβ R , (8.249)
δg 2
and the variational derivative of the action with respect to the metric tensor leads to
the condition Rαβ − 12 g αβ R = 0, which is the Einstein–Hilbert equation for a region
of the Universe, in which there is no matter and no other fields which give rise
to a nonvanishing energy-momentum tensor. The variational condition δSEH = 0
is equivalent to saying that the global integral over the curvature of space-time R
acts like a soap film whose surface tension integral is being minimized by the shape
assumed by space-time geometry, which is four-dimensional but might be embedded
in a higher-dimensional manifold which would defy our intuitive understanding. The
√
integration measure ∫ d4 x − det g is Lorentz-invariant. The coupling of the space-
time curvature to the matter density in the Universe can be written in terms of a
sum of the Einstein–Hilbert action SEH and the matter action SM , where the latter
can be expressed in terms of the matter Lagrangian LM ,
1 √
SGR = SEH + SM , SM = ∫ d4 x − det g LM , (8.250a)
16π G
√
2 δSM 2 ∂( − det g LM ) ∂LM
Tμν = − √ = − √ = −2 μν + gμν LM .
− det g δg μν
− det g ∂g μν
∂g
(8.250b)
The latter functional derivative acts as a definition of the energy-momentum tensor
Tμν . The form given in Eq. (8.250b) is valid if the matter Lagrangian does not
Advanced Topics 345
depend on derivatives of the metric. Examples of the energy-momentum tensor

include a collection of particles,
δ (4) (x − x(τa ))
T μν = ∑ ∫ dτa √ ma vaμ vaν . (8.251)
a − det g
Here, the vaμ are the four-vector components of the velocity va of particle a, and τa
is the proper time of the trajectory of particle a. For an electromagnetic field, the
stress–energy tensor is given by Eq. (8.102),
1
T μν = − (F μα F ν α − g μν F αβ Fαβ ) , (8.252)
4
where we recall that 0 = μ0 = 1 for the current discussion. Under the inclusion of
the energy-momentum tensor, the variational equations read as follows,
1
Rαβ − R gαβ = 8πG T αβ . (8.253)
2
These equations have all been derived under a variation of the action with respect
to the metric. They still do not clarify how the Maxwell equations need to be
generalized to curved space-time, and this will be explored next.
It is not hard to guess that for a free electromagnetic field, one simply has
√
to use the Lorentz-invariant measure ∫ d4 x − det g in the action integral for the
electromagnetic field,
√ 1
SEM = ∫ d4 x − det g (− F μν Fμν ) . (8.254)
4
The field-strength tensor, in the curved space-time, is given as follows,
∇μ Aν − ∇ν Aμ = ∂μ Aν − ∂ν Aμ . (8.255)
Here, we have used the definition (8.246) of the covariant derivative and the sym-
metry properties (8.245) of the Christoffel symbols. Variation with respect to the
metric leads to the energy-momentum tensor already given in Eq. (8.252), modulo
some total divergence. However, the variational principle also dictates that the
action SEM should be invariant under a variation of the four-vector field Aμ . The
corresponding Euler–Lagrange equation gives rise to the condition
∂ √
(F μν − det g) = 0 , ∇β F αβ = 0 , (8.256)
∂xβ
where ∇β is the covariant derivative. The covariant derivative of the field-strength
tensor can be expressed in terms of Christoffel symbols,
∇μ Fαβ = ∂μ Fαβ − Γρ αμ Fρβ − Γρ βμ Fαρ . (8.257)
If we consider a source term, then Eq. (8.254) gets modified to read
√ 1
SEM = δ ∫ d4 x − det g (− F μν Fμν − J μ δAμ )
4
δSEM ∂ √ √
= (F μν
− det g) − J μ
− det g = 0 . (8.258)
δAν ∂xν
In component form, we have

1 ∂ √
√ (F μν − det g) = J β , ∇β F μν = J β , (8.259)
− det g ∂xβ
where ∇β is the covariant derivative. The equations (8.259) are the generaliza-
tions of the inhomogeneous Maxwell equations, ∂μ F μν = J μ . For the homogeneous
Maxwell equations, we had defined in Eq. (8.89), for a flat space-time, the tensor
F̃μν = 12 μνρδ Fρδ , where μνρδ is the totally antisymmetric tensor. The homoge-
neous equations were obtained as ∂μ F̃μν = 0.
Obviously, in curved space-time, one would replace partial derivatives ∂β with
covariant derivatives ∇β to obtain the following equation, which is the appropriate
generalization to curved space-time,
αβγδ ∇β Fγδ = 0 . (8.260)
Any cyclic permutation of the indices βγδ leaves the result invariant, and therefore
we can reformulate the condition (8.260) as follows,
1
αβγδ ∇β Fγδ = αβγδ (∇β Fγδ + ∇γ Fδβ + ∇δ Fβγ ) = 0 . (8.261)
3
The value of the index α is arbitrary, and so we have
∇β Fγδ + ∇γ Fδβ + ∇δ Fβγ = ∂β Fγδ + ∂γ Fδβ + ∂δ Fβγ = 0 , (8.262)
which is otherwise known as the Bianchi identity. We recall that the covariant
derivative of the Maxwell tensor has been defined in Eq. (8.257).
A very interesting problem concerns the question whether a freely falling electron
(in the gravitational field of a star, say) emits radiation or not. On the one hand,
one can argue that any accelerated electron should radiate, a statement which would
in principle pertain to gravitational interactions. On the other hand, one can argue
that gravitational interactions are special, in the sense that any massive particle
follows a geodesic, described by the equation
d2 xμ dxρ dxσ
+ Γ μ
ρσ = 0, (8.263)
d2 τ dτ dτ
i.e., it follows a “straight line” in the curved space-time, and thus, it is highly
questionable if a straightforward generalization of the concept of the “radiating
electron” is applicable to gravitational interactions. In the literature, it is sometimes
said that a “particle on a geodesic does not radiate”, keeping in mind that a massive
body follows a geodesic trajectory described by Eq. (8.263).
The last word on this issue still needs to be spoken. In order to treat this
problem systematically, one has to formulate the problem in such a way that the
Liénard–Wiechert potentials are calculated, using the formalism of curved space-
time, for a particle moving along a specific trajectory. One thus needs to solve the
equations that couple the potentials to the sources in curved space-time, as given
in Eq. (8.259). Or, expressed differently, one has to calculate the Green functions
Advanced Topics 347
of the electromagnetic theory in curved space-time [54]. A particularly interesting

case is given by the stationary (time-independent) Schwarzschild metric,
g μν = diag (eν , −eλ , −r2 , −r2 sin2 θ) , (8.264)
2GMP
eν = e−λ = 1 − , (8.265)
r
where the Schwarzschild radius is rs = 2G MP /c2 , and G is Newton’s gravitational
constant. Furthermore, MP is the mass of the planet under consideration. If MP is
the mass of the Earth, then we have rs ≈ 0.0089 m. For a gravitational center and a
given trajectory of a freely falling body in the Schwarzschild metric, the scalar and
vector potentials for the pointlike particle should be expressed as a function of the
space-time coordinates along the geodesic. Using the curved-space Green functions
mentioned above, one should be able to calculate the scalar and vector potentials,
and calculate the Poynting vector which describes the dissipation of energy through
radiation. A systematic way of treating this problem has been outlined in Ref. [54];
the conclusion seems to be that a particle on an (almost) stable orbit (almost)
does not radiate. We recall that, in view of the perihelion precession (of Mercury
and other planets), there actually are no stable orbits in general relativity in the
Schwarzschild metric. Also, in Ref. [54], one reaches the conclusion that a particle
on a “crash” orbit toward the center of the gravitating object (almost) does not
radiate. However, a particle on a hyperbolic orbit seems to give off a substantial
amount of radiation. In general, this is a very difficult problem, both conceptually
and technically, and it may well be worth the effort to recalculate the formulas
presented in Ref. [54].
8.7 Exercises
● Exercise 8.1: Consider the Lagrangian (in classical mechanics)

1
L= m v⃗ 2 − V (∣⃗
r ∣) , (8.266)
2
for a single particle, and its variation under rotations characterized by a vector of
rotation angles ϕ.⃗ Under the assumption that L is independent of the rotation
⃗ show that the conserved quantity, in the sense of Noether’s theorem, is
angles ϕ,
the angular momentum.
● Exercise 8.2: Show the commutation relations for the orbital angular momentum
operators (8.43),
[Li , Lj ] = i ijk Lk , (8.267)
by acting on a general test function f = f (⃗
r).
● Exercise 8.3: Show the identity (8.48),
(ϕ⃗ × u
⃗) × v⃗ + u
⃗ × (ϕ⃗ × v⃗) = ϕ⃗ × (⃗
u × v⃗) , (8.268)
with the help of the tensor.
● Exercise 8.4: Derive the Biot–Savart law from the Lorenz-gauge coupling of the
vector potential to the current, in steady state,
∂2 ⃗ r , t) = μ0 J(⃗
⃗ r, t) ⃗ r) = −μ0 J(⃗
⃗ r) .
( ⃗ 2 ) A(⃗
−∇ ⇒ ⃗ 2 A(⃗
∇ (8.269)
∂t2
Hint: Write an “action-at-a-distance” solution to Eq. (8.269) (which is not problem-
atic because we considering a steady state) and take the curl of the vector potential.
● Exercise 8.5: Start from the relativistic transformation of the magnetic field,
⃗
⃗ ′ = γ (B
B ⃗ − 1 (⃗ ⃗ + (1 − γ) B ⋅ v⃗ v⃗ .
v × E)) (8.270)
c 2 v2
These formulas are valid for the transformation from the rest frame to the moving
(primed) frame. Recall that, for a Lorentz boost along the z axis, the backtrans-
formation of z and t reads as
v
x = x′ . y = y′ , z = γz ′ + γ vx t′ , t = γt′ + γ 2 z ′ . (8.271)
c
Derive an expression for the following partial derivative, keeping t′ constant,
∂ ∂z ∂ ∂t ∂ ∂ ∂
∣ = ∣ + ∣ =F +G , (8.272)
∂z ′ t′ ∂z ′ t′ ∂z ∂z ′ t′ ∂t ∂z ∂t
where the variable which is being kept constant is indicated as a subscript and you
calculate the terms F and G. Show that for v⃗ = v êz , i.e., under a boost in the z
direction,
v v
Bx′ = γ (Bx + 2 Ey ) , By′ = γ (By + 2 Ex ) , Bz′ = Bz . (8.273)
c c
Furthermore, show that
′
∂Bx′ ∂By ∂Bz′ ∂Bx ∂By ∂Bz v γ ∂Bz ⃗ ]
+ + = + + + 2 [ ⃗ × E)
+ (∇ (8.274)
∂x′ ∂y ′ ∂z ′ ∂x ∂y ∂z c ∂t z
where (∇ ⃗ × E)⃗ = ∂Ex /∂y − ∂Ey /∂x is the z component of the curl of the electric
z
field. Show that the expression given in Eq. (8.274) vanishes and interpret your
result in terms of the absence of magnetic monopoles in the primed and unprimed
coordinate systems, and in terms of (perhaps, if applicable) the Faraday law.
● Exercise 8.6: Start from the relativistic transformation of the electric field and
charge density,
⃗
⃗ ′ = γ (E
E ⃗ + (1 − γ) E ⋅ v⃗ v⃗ ,
⃗ + v⃗ × B) 1
ρ′ = γ (ρ − 2 J⃗ ⋅ v⃗) . (8.275)
v2 c
These formulas are valid for the transformation from the rest frame to the moving
(primed) frame.
Set v⃗ = vx êx and write explicitly the components Ex′ , Ey′ and Ez′ as a function
of the unprimed (rest-frame) components of the fields Ex , Ey and Ez and Bx , By
and Bz . Recall that, for a Lorentz boost along the x axis, the backtransformation
of x and t reads as
vx
x = γx′ + γ vx t′ , y = y′ , z = z′ , t = γt′ + γ 2 x′ . (8.276)
c
Advanced Topics 349
Derive an expression for the following partial derivative, keeping t′ constant,

∂ ∂x ∂ ∂t ∂ ∂ ∂
′
∣ = ′
∣ + ′
∣ =K +L , (8.277)
∂x t′ ∂x t′ ∂x ∂x t′ ∂t ∂x ∂t
where the variable which is being kept constant is indicated as a subscript and you
calculate the terms K and L.
In the moving frame, Gauss’s law reads as follows,
∂ ′ ∂ ∂ 1
′
Ex + ′ Ey′ + ′ Ez′ = ρ′ . (8.278)
∂x ∂y ∂z 0
Again, assuming v⃗ = vx êx , use the results for the components of E⃗ ′ derived pre-
viously, and Eq. (8.277), to rewrite the left-hand side of Eq. (8.278) in terms of
the unprimed components of the fields. Then, transform ρ′ on the right-hand side
of Eq. (8.278) according to Eq. (8.275), again assuming v⃗ = vx êx . Show that, in
the unprimed (rest) frame, Eq. (8.278) is equivalent to a linear superposition of the
“unprimed” Gauss’s law in the rest frame, and of the x-component of the Ampere–
Maxwell law in the rest frame.
● Exercise 8.7: We had encountered in Eq. (8.203) the following integral for
potential generated by a uniformly charged, D = n − ε dimensional structure,
1 ∞ Ξ Ξ
Φ(z) = ΩD ∫ dr rD−1 z0ε √ = J(D) , D = n −ε, (8.279)
4πε0 0 r2 + z 2 4πε0
Here, J(D) is defined as
∞ 1 2 π D/2
J(D) = ΩD ∫ dr rD−1 z0ε √ , ΩD = . (8.280)
0 r2 + z 2 Γ(D/2)
(a.) Show that
1
J(D) = π 2 (D−1) Γ ( 12 (1 − D)) z0ε ∣z∣D−1 . (8.281)
You may use the result
∞ rD−1 Γ( 21 (1 − D))Γ( 21 D)
∫ dr √ = √ . (8.282)
0 1 + r2 2 π
(b.) Investigate a “zero-dimensional uniformly charged structure” (a point
charge). To this end, first, show that, for D = 0 − ε,
1 1+ε 1
J(D = 0 − ε) = π − 2 (1+ε) z0ε z −1−ε Γ ( ) = + O(ε) . (8.283)
2 z
Give a heuristic argument why you identify the uniform charge density Ξ ≡ q in zero
dimensions, where q is the value of the point charge. Finally, show that
q
Φ(z) = , D = 0 −ε, (8.284)
4πε0 z
and interpret your result in terms of the well-known potential generated by a point
charge.
● Exercise 8.8: Show Eq. (8.217).

● Exercise 8.9: Investigate the Abraham–Minkowski controversy discussed in
Sec. 8.6.1 on the basis of the microscopic model introduced in Sec. 6.2.
● Exercise 8.10: Form your own opinion regarding the material presented in
Sec. 8.6.2. Study the literature and attempt to derive an equation for the description
of radiative reaction free from the problems of runaway solutions. Attempt to
reconcile the approach with quantum electrodynamics.
● Exercise 8.11: Study the literature on Christoffel symbols. Show Eq. (8.244). As
a next step, attempt to generalize the approach to the gradient, curl and Laplacian
in spherical and cylindrical coordinates, with reference to the discussion in Sec. 2.3.2
[see Ref. [16]]. The covariant derivatives in Eq. (8.260) act on vectors according to
∇α V β = ∂α V β + Γβ αμ V μ (8.285)
and on tensors as follows,
∇α F βγ = ∂α F βγ + Γβ αμ F μγ − Γγ αρ F βρ . (8.286)
How would the Coulomb law change around a massive black hole? Use Eq. (8.259),
specialized to the Schwarzschild geometry.
Bibliography
[1] M. Spivak, Calculus on Manifolds: A modern approach to classical theorems of ad-

vanced calculus. Addison-Wesley, Reading, MA (1992).
[2] L. Parker and S. M. Christensen, MathTensor: A System for Doing Tensor Analysis
by Computer. Addison-Wesley, Reading, MA (1994).
[3] B. O’Neill, Elementary Differential Geometry, 2nd edn. Academic Press, New York
(1997).
[4] M. Spivak, A Comprehensive Introduction to Differential Geometry, Vols. 1–5, 3rd
edn. Publish or Perish, Houston, TX (1999).
[5] F. W. J. Olver, D. W. Lozier, R. F. Boisvert and C. W. Clark, NIST Handbook of
Mathematical Functions. Cambridge University Press, Cambridge (2010).
[6] H. Bateman, Higher Transcendental Functions, Vol. 1. McGraw-Hill, New York
(1953).
[7] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions, 10th edn.
National Bureau of Standards, Washington, D. C. (1972).
[8] G. N. Watson, A Treatise on the Theory of Bessel Functions. Cambridge University
Press, Cambridge, UK (1922).
[9] F. W. J. Olver, Asymptotics and Special Functions. Academic Press, New York, NY
(1974).
[10] S. Wolfram, Mathematica-A System for Doing Mathematics by Computer. Addison-
Wesley, Reading, MA (1988).
[11] U. D. Jentschura and E. Lötstedt, Numerical Calculation of Bessel, Hankel and Airy
Functions, Comput. Phys. Commun. 183, pp. 506–519 (2012).
[12] S. Hildebrandt and A. J. Tromba, Mathematics of Optimal Form. W. H. Freeman,
New York (1984).
[13] R. Courant and D. Hilbert, Methods of Mathematical Physics—Volume 1. Inter-
science, New York (1966).
[14] G. Strang, Introduction to Linear Algebra, 4th edn. Wellesley–Cambridge Press,
Wellesley, MA (2009).
[15] C. D. Meyer, Matrix Analysis and Applied Linear Algebra. Society for Industrial and
Applied Mathematics, Philadelphia, PA (2001).
[16] D. A. Clarke, A Primer on Tensor Calculus, Saint Mary’s University, Halifax, Nova
Scotia, Canada (unpublished) (2011).
[17] C. Cohen-Tannoudji, B. Diu and F. Laloë, Quantum Mechanics (Volume 1), 1st edn.
J. Wiley & Sons, New York (1978a).
351
[18] C. Cohen-Tannoudji, B. Diu and F. Laloë, Quantum Mechanics (Volume 2), 1st edn.
J. Wiley & Sons, New York (1978b).
[19] J. D. Jackson, Classical Electrodynamics, 3rd edn. J. Wiley & Sons, New York, NY
(1998).
[20] C. Itzykson and J. B. Zuber, Quantum Field Theory. McGraw-Hill, New York (1980).
[21] J. Schwinger, L. L. DeRaad, K. A. Milton and W.-Y. Tsai, Classical Electrodynamics.
Perseus, Reading, MA (1998).
[22] B. J. Wundt and U. D. Jentschura, Sources, Potentials and Fields in Lorenz
and Coulomb Gauge: Cancellation of Instantaneous Interactions for Moving Point
Charges, Ann. Phys. (N.Y.) 327, pp. 1217–1230 (2012).
[23] U. D. Jentschura, Precision theory of atoms: Quantum electrodynamics at work.
World Scientific, Singapore (scheduled for 2017).
[24] F. Rohrlich, Causality, the Coulomb field, and Newtons law of gravitation, Am. J.
Phys. 70, pp. 411–414 (2002).
[25] J.-J. Labarthe, The vector potential of a moving charge in the Coulomb gauge, Eur.
J. Phys. 70, pp. L31–L32 (1999).
[26] G. Russakoff, A Derivation of the Macroscopic Maxwell Equations, Am. J. Phys. 38,
pp. 1188–1195 (1970).
[27] G. L ach, M. DeKieviet and U. D. Jentschura, Multipole Effects in Atom–Surface
Interactions: A Theoretical Study with an Application to He–α-quartz, Phys. Rev.
A 81, p. 052507 (2010).
[28] W. Sellmeier, Zur Erklärung der abnormen Farbenfolge im Spectrum einiger Sub-
stanzen, Annalen der Physik und Chemie 219, pp. 272–282 (1871).
[29] E. D. Palik, Handbook of Optical Constants of Solids. Academic Press, San Diego
(1985).
[30] B. Bertotti, L. Iess and P. Tortura, A test of general relativity using radio links with
the Cassini spacecraft, Nature (London) 425, pp. 374–376 (2003).
[31] A. Wolski, Theory of Electromagnetic Fields, Part II: Standing Waves, CERN Accel-
erator School, Specialised Course on Radiofrequency for Accelerators (unpublished)
(2010).
[32] H. B. G. Casimir and D. Polder, The Influence of Radiation on the London-van-der-
Waals Forces, Phys. Rev. 73, pp. 360–372 (1948).
[33] H. Minkowski, Die Grundgleichungen für die elektromagnetischen Vorgänge in be-
wegten Körpern, Nachr. Ges. Wiss. Göttingen, Math.–Phys. Kl. 1, 53–111 (1908).
[34] M. Abraham, Zur Elektrodynamik bewegter Körper, Rend. Circ. Math. Palermo 28,
1–28; Sull’Elettrodinamica di Minkowski, ibid. 30, 33–46 (1910).
[35] P. W. Milonni and R. W. Boyd, Momentum of Light in a Dielectric Medium, Adv.
Opt. Photonics 2, pp. 519–533 (2010).
[36] I. Brevik and S. A. Ellingsen, Detection of the Abraham force with a succession of
short optical pulses, Phys. Rev. A 86, p. 025801 (2012).
[37] L. D. Landau and E. M. Lifshitz, The Classical Theory of Fields, Volume 2 of the
Course on Theoretical Physics. Pergamon Press, Oxford, UK (1960).
[38] M. Abraham, Prinzipien der Dynamik des Elektrons, Ann. Phys. (Leipzig) 315, pp.
105–179 (1902).
[39] M. Abraham, Zur Theorie der Strahlung und des Strahlungsdruckes, Ann. Phys.
(Leipzig) 319, pp. 236–287 (1904).
Bibliography 353
[40] P. A. M. Dirac, Classical theory of radiating electrons, Proc. Roy. Soc. London, Ser.
A 167, p. 148 (1938).
[41] B. S. DeWitt and R. W. Brehme, Radiation Damping in a Gravitational Field, Ann.
Phys. (N.Y.) 9, p. 220259 (1960).
[42] J. M. Hobbs, A Vierbein Formalim of Radiation Damping, Ann. Phys. (N.Y.) 47,
pp. 141–165 (1968).
[43] H. Levine, E. J. Moniz and D. H. Sharp, Motion of extended charges in classical
electrodynamics, Am. J. Phys. 45, pp. 75–78 (1977).
[44] H. Spohn, The critical manifold of the Lorentz-Dirac equation, Europhys. Lett. 50,
pp. 287–292 (2000).
[45] R. Medina, Radiation reaction of a classical quasi-rigid extended particle, J. Phys. A
39, pp. 3801–3816 (2006).
[46] A. di Piazza, Exact Solution of the Landau–Lifshitz Equation in a Plane Wave, Lett.
Math. Phys. 83, pp. 305–313 (2008).
[47] D. J. Griffiths, T. C. Proctor and F. D. Schroeter, Abraham–Lorentz versus Landau–
Lifshitz, Am. J. Phys. 78, pp. 391–402 (2010).
[48] A. Kar and S. G. Rajeev, On the relativistic classical motion of a radiating spinning
particle in a magnetic field, Ann. Phys. (N.Y.) 326, pp. 958–967 (2011).
[49] H. A. Lorentz, Electromagnetic phenomena in a system moving with any velocity
smaller than that of light, Kon. Akad. Weten. Amsterdam. Proc. 6, pp. 809–831
(1904).
[50] L. D. Landau and E. M. Lifshitz, Quantum Mechanics, Volume 3 of the Course on
Theoretical Physics. Pergamon Press, Oxford, UK (1958).
[51] H. Spohn, Dynamics of Charged Particles and their Radiation Field. Cambridge Uni-
versity Press, Cambridge, England (2004).
[52] V. Bargmann, L. Michel and V. I. Telegdi, Precession of the Polarization of Particles
Moving in a Homogeneous Magnetic Field, Phys. Rev. Lett. 2, pp. 435–436 (1959).
[53] A. O. Barut and N. Zanghi, Classical Model of the Dirac Electron, Phys. Rev. Lett.
52, pp. 2009–2012 (1984).
[54] R. A. Breuer, Gravitational Perturbation Theory and Synchrotron Radiation (Lecture
Notes in Physics Vol. 44). Springer, Heidelberg (1975).
Index
Absence of Magnetic Monopoles, 3 Covariant Derivative, 13

Addition Theorem
Spherical Harmonics, 48 Dirac Delta Function, 9, 170, 203
Velocities, 299 Dirichlet Boundary–Value Problem, 73,
Advanced Green Function, 140, 153 131
Ampere–Maxwell Law, viii, 3, 4, 15, 18, Dirichlet Green Function, 75, 124, 127
175, 287, 295, 320, 349 Dispersion Relation, 273, 276
Bare Quantities, 322 Eigenfunction, 63

Bare Quantity, 322 Einstein summation convention, 12
Bargmann–Michel–Telegdi Equation, 342 Einstein–Hilbert Action, 344
Barut–Zanghi Equation, 342 Einstein–Hilbert Equation, 344
Bernoulli Polynomial, 328 Euler–Lagrange Equation, 81, 310
Bessel Function, 54, 191, 206 Euler–Maclaurin Formula, 326
First Kind, 53 Euler–Mascheroni Constant, 336
Second Kind, 53, 54, 206
Spherical, 60, 70, 168, 206 Faraday’s Law, 3, 4, 8, 15, 26, 266–268,
Bessel Functions, 101 348
Bessel’s Differential Equation, 54 Feynman Green Function, 141
Biot–Savart Law, 348 FeynmanGreen Function, 153
Boundary–Value Problem Field–Strength Tensor, 309, 312, 313
Dirichlet, 73, 131 First Quantization, 264
Neumann, 73 Fourier Series, 77, 97
Fourier Transform, 11, 31
Casimir Effect, 322 Fourier Transformation, 137
Casimir Energy, 330 Fourier–Sine Transform, 115
Cauchy’s Residue Theorem, 110, 113
Cavity Gamma Function, 54, 334
Cylindrical, 279 Gauge Dependence, 17
Rectangular, 287 Gauss’s Law, 3, 4, 15, 96, 349
Christoffel Symbol, 343 Gaussian Bell, 334
Clebsch–Gordan Coefficient, 181, 216 Green’s Function, 45, 60, 66, 68, 138, 161
Coulomb Gauge, 20, 157, 161, 211 Advanced, 140, 153, 165
Coulomb’s Law, 22, 350 Dirichlet, 75, 124, 127
Counter Term, 322 Feynman, 141, 153
355
Harmonic Oscillator, 30 Minimal Subtraction, 335

Helmholtz, 218 Multipole Decomposition, 34, 48
Neumann, 75 Multipole Expansion, 45, 48
Retarded, 31, 32, 140, 153, 165, 213
Green’s Theorem, 74 Neumann Boundary–Value Problem, 73
Neumann Function, 54
Hamiltonian, 324 Spherical, 206
Hankel Function, 174, 191, 215 Neumann Green Function, 75
First Kind, 169 Newton–Raphson Method, 69
Second Kind, 169 Newtonian Equations, 82
Harmonic Oscillator, 29
Heaviside Step Function, 31, 95, 214 Pauli Matrix, 215
Helmholtz Equation, 166, 208 Poisson Equation, 33, 71
Helmholtz Green Function, 189, 218 Poynting Vector, 24, 200, 347
Hilbert Space, 80, 83
Quabla Operator, 19
Jacobian, 220 Quantization, 283
Quantum Electrodynamics, 322
Kronecker Delta, 27, 58, 79, 184
Radar, 292
Lagragian, 347 Regularization, 322
Lagrange Function, 82 Renormalization, 322
Lagrange Multiplier, 82, 84, 85, 89 Retarded Green Function, 140, 153
Lagrangian, 344 Ricci Tensor, 344
Landau–Lifshitz Equation, 340 Riemann Curvature Tensor, 344
Laplace Equation, 33, 72, 99
Two–Dimensional, 106 Schwarzschild Metric, 347, 350
Laplacian, 34, 52 Second Quantization, 264
Laplacian Operator, 11 Slow Light, 278
Legendre Equation, 36 Spherical Bessel Function, 168, 206
Legendre Function Spherical Harmonic, 39
First Kind, 36 Spin Operator, 186
Second Kind, 36 Surface Charge Density, 92
Legendre Functions
First Kind, 37 Temporal Gauge, 28
Legendre Polynomial, 68, 126, 131 Time Dilation, 298
Legendre Polynomials, 37, 100 Transverse Electric Mode
Levi–Cività Tensor, 7 Cavity, 283, 288
Liénard–Wiechert Potential, 210 Graphs, 272
Liénard–Wiechert Potentials, 321 Waveguide, 263
Light Pulse, 278 Transverse Magnetic Mode
Lorentz Contraction, 298 Cavity, 283, 288
Lorentz Transformation, 19, 295 Graphs, 277, 284
Lorentz–Abraham–Dirac Equation, 340 Waveguide, 263, 268
Lorenz Gauge, 16, 17, 157, 196, 348
Vector Field
Matter Waves, 263, 298 Longitudinal Component, 11
Maxwell Equations, 1, 265, 311 Transverse Component, 11
Maxwell Stress Tensor, 26, 313
Maxwell Stress–Energy Tensor, 313 Wave Equation, 164

Advanced Classical Electrodynamics Green Functions, Regularizations, Multipole Decompositions, Jentschura, WS, 2017

Uploaded by

Copyright:

Available Formats

Advanced Classical Electrodynamics Green Functions, Regularizations, Multipole Decompositions, Jentschura, WS, 2017

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Advanced Classical Electrodynamics Green Functions, Regularizations, Multipole Decompositions, Jentschura, WS, 2017

Uploaded by

Copyright:

Available Formats

10514_9789813222847_tp.

indd 1 20/4/17 4:54 PM

Library of Congress Cataloging-in-Publication Data

British Library Cataloguing-in-Publication Data

Copyright © 2017 by World Scientific Publishing Co. Pte. Ltd.

LerhFeng - 10514 - Advanced Classical Electrodynamics.indd 1 23-03-17 9:00:19 AM

Dedicated to All Who Seek Beauty in Nature

Classical electrodynamics is a vast and open ﬁeld. It would be hard to summarize

viii Advanced Classical Electrodynamics

relevance, waveguides and the resonant eigenmodes of cavities (both rectangular as

Rolla, Missouri, January 2017

2. Green Functions of Electrostatics 29

x Advanced Classical Electrodynamics

2.3.8 Multipole Expansion in an External Field . . . . . . . . . . 50

3. Paradigmatic Calculations in Electrostatics 71

3.5.2 Sources and Fields in a Spherical Shell . . . . . . . . . . . . 127

4. Green Functions of Electrodynamics 137

5. Paradigmatic Calculations in Electrodynamics 163

xii Advanced Classical Electrodynamics

5.6 Potentials due to Moving Charges in Diﬀerent Gauges . . . . . . . 209

6. Electrodynamics in Media 221

7. Waveguides and Cavities 263

8. Advanced Topics 295

8.2.1 Lorentz Boosts . . . . . . . . . . . . . . . . . . . . . . . . . . 297

1.1.1 Integral and Diﬀerential Forms of the Maxwell Equations

2 Advanced Classical Electrodynamics

which are the Maxwell2 equations,

4 Advanced Classical Electrodynamics

1.1.2 Relation of the Diﬀerential to the Integral Form

I≡∫ ⃗ r , t) ⋅ dA⃗ ≈ [Ex (x + δx, y + 1 δy, z + 1 δz, t)

− Ex (x, y + 12 δy, z + 12 δz, t)] δy δz . (1.9)

To ﬁrst order in δx, we obtain

≈ [Ex (x + δx, y + 12 δy, z + 12 δz, t) − Ex (x, y + 12 δy, z + 12 δz, t)] δy δz

[Ex (x + δx, y + 12 δy, z + 12 δz, t) − Ex (x, y + 12 δy, z + 12 δz, t)]

6 Advanced Classical Electrodynamics

identifying the components 1, 2, 3 with the x, y, z directions, respectively. The gra-

We can thus write Eq. (1.14) as

where ΦB is the magnetic ﬂux. The closed-loop integral ∮∂A E(⃗ ⃗ s, t) ⋅ d⃗

− Ex (x + 12 δx, y + δy, z, t) δx − Ey (x, y + 12 δy, z, t) δy ,

where we suppress the space-time arguments of the ﬁelds. We identify ∂x ∂

⎛ êx êy êz ⎞

8 Advanced Classical Electrodynamics

Normal to the Area

Using the same test volume as given in Fig. 1.1, we obtain

[Jx (x + δx, y + 12 δy, z + 12 δz, t) − Jx (x, y + 12 δy, z + 12 δz, t)] δy δz

1.1.3 Dirac δ Function

10 Advanced Classical Electrodynamics

The limit is η → 0 is approached nonuniformly. The Δ function has the properties

lim Δ(η, ω) = ∞ (for ω = 0), (1.33c)

particular, the integration domain then automatically deﬁnes the dimensionality of

1.1.4 Green Function of the Laplacian

1.2 Potentials and Gauges

12 Advanced Classical Electrodynamics

we would like to separate into transverse and longitudinal components. We seek a

1.2.2 Vector and Scalar Potentials

14 Advanced Classical Electrodynamics

is called a gauge transformation. Gauge covariance now means the following: If we

then it is easy to show that

16 Advanced Classical Electrodynamics

1.2.3 Lorenz Gauge

∫′ (r2 G (r, r′ )) − ( + 1) G (r, r′ )] dr = − ∫