Classical

ElectrDdynam ics
Julian Schwinger • Lester L. DeRaad, Jr.

Kimball A. Milton • W u-yang Tsai

A d v a n c e d Bo o k P r o g r a m

CRC Press
Taylor & Francis Group
Boca Raton London New York

CRC Press is an imprint of the

Taylor & Francis Group, an informa business
The Advanced Book Program

First published 1998 by Westview Press

Published 2018 by CRC Press

Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742

CRC Press is an imprint of the Taylor & Francis Group, an informa business

Copyright © 1998 by Clarice Schwinger, Lester DeRaad, Jr., Kimball A.

Milton, and Wu-yang Tsai

No claim to original U.S. Government works

This book contains information obtained from authentic and highly regarded sources.
Reasonable efforts have been made to publish reliable data and information, but the
author and publisher cannot assume responsibility for the validity o f all materials
or the consequences of their use. The authors and publishers have attempted to trace
the copyright holders of all material reproduced in this publication and apologize to
copyright holders if permission to publish in this form has not been obtained. If any
copyright material has not been acknowledged please write and let us know so we
may rectify in any future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted,
reproduced, transmitted, or utilized in any form by any electronic, mechanical,
or other means, now known or hereafter invented, including photocopying,
microfilming, and recording, or in any information storage or retrieval system,
without written permission from the publishers.

Trademark Notice: Product or corporate names may be trademarks or registered

trademarks, and are used only for identification and explanation without intent to
infringe.

Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com

and the CRC Press Web site at

http ://www. crcpress. com

Library of Congress Catalog Card Number: 98-86259

ISBN 13: 978-0-7382-0056-9 (hbk)

Cover design by Suzanne Heiser

 T o our children,
Elizabeth, Lester III
Ysabel, Madeleine, Camille
Tenno
Contents
Preface xvij

Reader’s Guide xix

1 Maxwell’s Equations 1
1.1 E lectrostatics....................................................................................... 1
1.2 Inference of Maxwell’s E q u a tion s..................................................... 8
1.3 D iscussion........................................... 12
1.4 Problems for Chapter 1 .................................................................... 13

2 Magnetic Charge I IT
2.1 A Very Brief History of Magnetic C h a r g e ..................................... 19
2.2 Problems for Chapter 2 .................................................................... 19

3 Conservation Laws 21
3.1 Conservation of Energy .................................................................... 21
3.2 Conservation of M om en tu m .............................................................. 22
3.3 Conservation of Angular Momentum. Virial Theorem ................ 24
3.4 Conservation Laws and the Speed of L ig h t..................................... 25
3.5 Problems for Chapter 3 .................................................................... 29

4 Macroscopic Electrodynamics 33
4.1 Force on an Atom ............................................................................. 33
4.2 Force on a Macroscopic B o d y ........................................................... 38
4.3 Macroscopic Electrodynam ics..................... 40
4.4 Problems for Chapter 4 .................................................................... 42

5 Simple Model for Constitutive Relations 45

5.1 C o n d u c tiv ity ....................................................................................... 45
5.2 Dielectric C on stan t............................................................................. 47
5.3 Plasm a................................................................................................... 51
5.4 Polar M olecules................................................................................... 53
5.5 Clausius-Mossotti E q u a tio n .............................................................. 57
5.6 Problems for Chapter 5 .................................................................... 59

6 Magnetic Properties of Matter 63

6.1 Canonical Equations of Motion in Electromagnetic Fields . . . . 63
6.2 Diamagnetism....................................................................................... 65

v
vi CONTENTS

6.3 Paramagnetism.................................................................................... 67
6.4 Ferromagnetism.................................................................................... 69
6.5 Problems for Chapter 6 .................................................................... 72

7 Macroscopic Energy and Momentum 75

7.1 General D iscussion.............................................................................. 75
7.2 Nondispersive M ediu m ....................................................................... 77
7.3 Dispersive M ed iu m .............................................................................. 77
7.4 Problems for Chapter 7 .................................................................... 80

8 Review of Action Principles 85

8.1 Lagrangian V ie w p o in t....................................................................... 86
8.2 Hamiltonian Viewpoint .................................................................... 87
8.3 A Third V iew p oin t.............................................................................. 89
8.4 Invariance and Conservation L a w s .................................................. 90
8.5 Nonconservation Laws. The Virial T h e o re m .................................. 94
8.6 Problems for Chapter 8 .................................................................... 95

9 Action Principle for Electrodynamics 97

9.1 Action of Particle in Field .................................................................. 97
9.2 Electrodynamic A c tio n ....................................................................... 98
9.3 E nergy...................................................................................................... 100
9.4 Momentum and Angular Momentum C onservation.........................102
9.5 Gauge Invariance and the Conservation of Charge .........................105
9.6 Gauge Invariance and Local Conservation L a w s ............................... 107
9.7 Problems for Chapter 9 .......................................................... 110

10 Einsteinian Relativity 111

10.1 Relativistic M odification....................................................................... I l l
10.2 Lorentz Transformations....................................................................... 114
10.3 Transformation of F ield s....................................................................... 118
10.4 Problems for Chapter 1 0 ....................................................................... 122

11 Stationary Principles for Electrostatics 125

11.1 Stationary Principles for the E n e r g y ..................................................126
11.1.1 The Scalar Field F o r m ..............................................................127
11.1.2 The Vector Field F orm ..............................................................127
11.2 Force on Dielectrics............... 129
11.3 Boundary Conditions . ...........................................................................132
11.4 C o n d u c to r s .............................................................................................134
11.5 Problems for Chapter 1 1 ....................................................................... 135

12 Introduction to Green’s Functions 137

12.1 Reciprocity R e la tio n .............................................. ... . ................... 138
12.2 Problems for Chapter 1 2 ....................................................................... 139
CONTENTS vü

13 Electrostatics in Free Space 141

13.1 2 + 1 D im en sion s................................................................................ 143
13.2 Problems for Chapter 1 3 ....................................................................... 146

14 Semi-Infinite Dielectric 147

14.1 Green’s Function for Charge Outside D ie le ctric............................147
14.2 Derivation in Terms of Bound C h a r g e ........................................... 150
14.3 Green’s Function for Charge Within Dielectric ............................153
14.4 Full Green’s Function and Image C harge........................................ 155
14.5 Problems for Chapter 1 4 ........................................................... . . . 156

15 Application of Green’s Function 157

15.1 Force between Charge and D ielectric................................................. 157
15.2 Infinite Conducting P l a t e .................................................................... 161
15.3 Problems for Chapter 1 5 ....................................................................... 163

16 Bessel Functions 165

16.1 Delta Functions and C om pleteness.................................................... 170
16.2 Problems for Chapter 1 6 ....................................................................... 174

17 Parallel Conducting Plates 177

17.1 Reduced Green’s F unction.................................................................... 177
17.2 Induced C h a rge...................................................................................... 179
17.3 E nergy...................................................................................................... 180
17.4 F orce............................................................................................... 182
17.5 Im ages........................... 183
17.6 Linear L a t t ic e s ...................................................................................... 185
17.7 Periodic Green’s F u n ctio n .................................................................... 186
17.8 Problems for Chapter 17......................................................... 189

18 Modified Bessel Functions 193

18.1 More Bessel F u n ction s........................................... 197
18.2 Problems for Chapter 1 8 .......................................................................203

19 Cylindrical Conductors 205

19.1 R e c ta n g le ................................................................................................205
19.2 Isosceles Right Angle Triangle..............................................................207
19.3 Equilateral Triangle .............................................................................208
19.4 Circle ......................................................................................................215
19.5 Circle and S e p t u m ................................................................................ 225
19.6 Problems for Chapter 1 9 ........................... 228

20 Spherical Harmonics 231

20.1 Solutions to Laplace’s E quation............... 231
20.2 Spherical H a rm on ics........................................... ... . . . ................ 233
20.3 Orthonormality C on d ition ..................... 236
20.4 Legendre’s P o ly n o m ia ls .......................................................................238
viii CONTENTS

20.5 Problems for Chapter 20 ................................................................... 238

21 Coulomb’s Potential 243

21.1 Legendre’s P o ly n o m ia ls ....................................................................... 244
21.2 Infinitesimal R otation s.......................................................................... 246
21.3 Spherical Bessel Functions....................................................................249
21.4 Problems for Chapter 2 1 ....................................................................... 253

22 Multipoles 257
22.1 Problems for Chapter 22 .................................................................... 261

23 Conducting and Dielectric Spheres 265

23.1 Interior of Conducting SphericalS h e l l ................................................265
23.1.1 Bessel Function Representation.............................................. 267
23.2 Exterior of Conducting Sphere ....................................................... 271
23.3 Conducting Plate and HemisphericalB o s s ..........................................274
23.4 Dielectric Sphere ................................................. 276
23.4.1 Interior of sp h e re ....................................................................... 278
23.5 Problems for Chapter 23 .................................................................... 280

24 Dielectrics and Conductors 283

24.1 Variational P rin cip le ................................................................ 283
24.2 Restricted Forms of the Variational P r in c ip le ..................................285
24.3 Introduction of Additional Conductor .............................................. 286
24.4 Alternate Variational P r in c ip le ...........................................................288
24.5 Green’s F u n ctio n ................................................................................... 289
24.6 Capacitance........................................................... . . ...................... 290
24.7 Problems for Chapter 24 .................................................................... 293

25 Modes and Variations 295

25.1 A Comparison M e th o d ............................................................. 301
25.2 Iteration ...................................................................................................304
25.3 E xam ple...................................................................................................311
25.4 Problems for Chapter 25 . . ..................................... 312

26 Magnetostatics 313
26.1 Variational P rin cip le ............................................................................. 313
26.2 Boundary Conditions............................................................................. 315
26.3 Vector Potential ................................................................................... 316
26.4 Problems for Chapter 2 6 ....................................................................... 318

27 Macroscopic Current Distributions 319

27.1 Magnetic Energy. Coefficients of Inductance............................ 322
27.2 Problems for Chapter 27 .................................................................. 323
CONTENTS IX

28 Magnetic Multipoles 325

28.1 Magnetic Dipole M o m e n t ....................................................................325
28.2 Rotating Charged Spherical S h e ll........................................................ 327
28.3 Problems for Chapter 28 .................................................................... 329

29 Magnetic Scalar Potential 331

29.1 Problems for Chapter 29 .................................................................... 334

30 Magnetic Charge II 337

30.1 Problems for Chapter 30 .................................................................... 339

31 Retarded Green’s Function 341

31.1 Potentials and G a u g e s ..........................................................................341
31.2 Green’s Function in the Lorentz G auge........................................... 343
31.3 Problems for Chapter 3 1 .......................................................................346

32 Radiation— Field Point of View 351

32.1 Asymptotic Potentials and F ie ld s ....................................................... 351
32.2 Angular Distribution of RadiatedP o w e r ............................................ 353
32.3 Radiation by an Accelerated Charged P article..................................354
32.4 Dipole R a d ia tio n ................................................................................... 355
32.5 Potentials in Radiation G auge............................................................. 357
32.6 Problems for Chapter 32 .................................................................... 359

33 Radiation— Source Point of View 361

33.1 Conservation of Energy .......................................................................361
33.2 Dipole R a d ia tio n ................................................................................... 362
33.3 Hamiltonian............................................................................................ 363
33.4 Problems for Chapter 33 .................................................................... 366

34 Models of Antennas 367

34.1 Simplified M o d e l ................................................................................... 367
34.2 Center-fed an ten n a................................................................................ 371
34.3 Problems for Chapter 34..................................................................... 372

35 Spectral Distribution of Radiation 375

35.1 Spectral and Angular D istrib u tio n .................................................... 375
35.2 Spectral Distribution for Dipole R adiation........................................379
35.3 Damped Harmonic M otion ....................................................................380
35.4 Problems for Chapter 35 .................................................................... 383

36 Power Spectrum and Cerenkov Radiation 385

36.1 Macroscopic Power S p e ctru m ..............................................................385
36.2 Cerenkov Radiation .............................................................................386
36.3 Problems for Chapter 36 .................................................................... 389
X CONTENTS

37 C on stan t A cce le ra tio n an d Im p u lse 391

37.1 Radiation by a Uniformly Accelerated P a r t i c le ............................ 391
37.2 Radiation by Impulsive S c a tte r in g ........................................ ... 396
37.3 Problems for Chapter 37 ................................................................. ... 398

38 S y n ch ro tro n R a d ia tio n I 401

38.1 Motion of a Charged Particle in a Homogeneous Magnetic Field . 401
38.2 Spectrum of Synchrotron Radiation .....................................................403
38.3 Total Power Emitted into the mth H a r m o n ic .................. 406
38.4 Total Radiated P o w e r .......................................................................... 408
38.5 Problems for Chapter 3 8 ....................................................................... 410

39 S y n ch ro tro n R a d ia tio n II— P ola riza tion 413

39.1 Problems for Chapter 3 9 ................................................................. ... 415

40 S y n ch ro tro n R a d ia tio n III— H igh E nergies 417

40.1 Range of Important H arm on ics............... 417
40.2 Asymptotic Form for J 2m(2m ).............................................................. 418
40.3 Spectral D istribution........................ 419
40.4 Angular D istribution.............................................. 422
40.5 Qualitative Description .......................................................................424
40.6 Problems for Chapter 40 .............................................. 426

41 P ro p a g a tio n in a D ie le ctric M e d iu m 427

41.1 Equations for the Normal Modes ........................................................427
41.2 Reflection and Refraction: Jl P ola rization ........................................ 431
41.3 Reflection and Refraction: || Polarization ........................................ 436
41.4 Total Internal R eflection.............................................. 440
41.5 Energy C onservation................................................. 441
41.6 Problems for Chapter 4 1 .......................................................................443

42 R e fle ctio n b y an Im p e rfe ct C o n d u c to r 445

42.1 Problems for Chapter 42 .................................................................... 448

43 C y lin d rica l C oord in a tes 449

43.1 2 + 1 Dimensional Decomposition of Green’s Function . . . . . . 449
43.2 Three Dimensional Fourier Representation . . ................................451
43.3 Hankel F unctions..................... 452
43.4 Problems for Chapter 43 .................................................................... 455

44 W avegu ides 459

44.1 E and H m o d e s ................................. 459
44.2 Boundary Conditions............................................................................. 461
44.3 M o d e s ......................................................................................................464
44.4 Problems for Chapter 44 ........................................ 466
CONTENTS xi

45 Scattering by Small Obstacles 471

45.1 Thomson S c a tte r in g .............................................................................471
45.2 Scattering by a Bound Charge .......................................................... 473
45.3 Scattering by a Dielectric S p h e r e ....................................................... 475
45.4 Radiation Dam ping............................................................................. 476
45.5 Problems for Chapter 45 .................................................................... 478

46 Partial-Wave Analysis of Scattering 479

46.1 Mode D ecom position.............................................................................479
46.2 Interior of conducting sphere ............................................................. 481
46.3 Spherical Hankel F u n c tio n s .................................................................483
46.4 S catterin g............................................................................................... 484
46.5 Problems for Chapter 46 .................................................................... 490

47 Diffraction I 491
47.1 Diffracted Electric F i e l d .......................................................................491
47.2 Diffraction by a Circular A p e r t u r e ................................................. 495
47.3 Diffraction by a Slit ........................................... 499
47.4 Diffraction by a Straight E dge............................................................. 503
47.5 Problems for Chapter 47...................................................................... 507

48 Diffraction II 509
48.1 Approximate S o lu t io n ..........................................................................510
48.2 Exact Solution for Current .................................................................512
48.3 Exact Diffraction Cross S e c tio n ................................................. . . . 5 1 6
48.4 Field Near Edge ................................................................................... 517
48.5 Problems for Chapter 48 .................................................................... 520

49 Babinet’s Principle 523

49.1 Problems for Chapter 49 .................................................................... 525

50 General Scattering 527

50.1 Integral E quation................................................................................... 527
50.2 Optical T h e o r e m ................................................................................... 529
50.3 Born A pproxim ation .............................................................................531
50.4 Problems for Chapter 50 .................................................................... 532

51 Dispersion Relations for theSusceptibility 539

51.1 Problems for Chapter 5 1 .......................................................................543

52 Charged Particle Energy Loss 545

52.1 General Expression................................................................................ 545
52.2 Evaluation in Terms of Spectral Functions..................................... 547
52.3 High Energy L i m it ................................................................................ 551
52.4 Energy Loss by a Magnetic M o n o p o le ........................................... 552
52.5 Problems for Chapter 52 .................................................................... 554
xii CONTENTS

A Units 555

B Bibliography 561

Index 563
List of Figures

1.1 A surface S bounding a volume V used in computing the electric

flux............................................... 4
1.2 Geometrical definition of solid angle............................................. 5
1.3 Topology if e¡, is inside (a) or outside (b) the surface S .............. 5

3.1 Electric and magnetic fields for an electromagnetic pulse propa­

gating with velocity v ...................................................................... 29

5.1 The real and imaginary parts of the conductivity, as givenin (5.14). 48
5.2 Singularities of the integrand in (5.33) in the complexa;plane. . 50

6.1 Solutions to (6.58) for temperatures above and below the Curie
temperature. The solid curves are the plots of the right side of
(6.58) for T < T C and T > Tc , respectively, while the dashed line
is the left side of that equation........................................................... 71

8.1 A possible path from initial state to final state................................ 85

8.2 8u> Xr is perpendicular to 8<jj and r, and represents an infinitesi­
mal rotation of r about the 8w axis.................................................. 92

9.1 Rigid coordinate displacement...............................................................103

10.1 The transformation (10.43), (10.44) carries us to a new coordinate

frame moving relative to the original one with a velocity —v along
the z-axis...................................................................................................117
10.2 Motion of original frame relative to the new frame............................118

11.1 Dielectric slab immersed in an inhomogeneous electric field. The

forces F shown are those on each surface due to (11.46), (11.50). 131
11.2 Boundary between two regions with different dielectric constants. 132
11.3 Dielectric between parallel plates, as discussed in Problem 11.5. . 136

13.1 Contour used to evaluate (13.22).......................................................... 144

14.1 Geometry of semi-infinite dielectric region.............................. 147

xiii
XIV LIST OF FIGURES

14.2 The primary field, P, and secondary field, S', contributing to

Green’s function (14.26).......................................................................151

15.1 Image charge for a grounded conducting plane................................ 162

17.1 Geometry of grounded, parallel, conducting plates..........................177

17.2 Positions of image charges for two parallel conductors. . . . . . . 185
17.3 Infinite conducting sheet, described in cylindrical coordinates by
(j> = 0,7r. The coordinate z is perpendicular to the page. A unit
point charge is located at the point ( //, f t , z1) ..................................190

19.1 Cylindrical region created by intersecting perpendicular conduct­

ing planes........................ 205
19.2 Bisection of a cylinder with square cross section................................ 207
19.3 Cross section of equilateral triangular cylinder................................... 209
19.4 Reflected equilateral triangle................................................................. 210
19.5 Repeated reflection of equilateral triangle........................................... 210
19.6 Cylinder with septum .......................................................................... 225

22.1 Geometry of field point and source point for a bounded charge
distribution............................................................................................... 257
22.2 Two interacting charge distributions.................................................... 260

23.1 Sphere bisected by plane. Shown are the locations of the physical
charge at r' and the image charges at r', r^, and ....................... 275
23.2 Geometry of dielectric sphere with source point outside................ 276

24.1 Conducting surfaces embedded in a dielectric medium...................284

24.2 Introduction of an uncharged conductor into a region where there
is no free charge density............... 287
24.3 Finite dielectric bounded by conductors.............................................. 292
24.4 Parallel plate capacitor........................................................................... 293

26.1 Discontinuity in permeability across a plane interface....................... 315

27.1 Magnetic field produced by a current-carrying wire...........................319

27.2 Force between parallel, current-carrying wires.................................... 320

29.1 Closed current loop................................................................................. 331

29.2 Topology of curve C relative to current loop............................. 333

30.1 Sphere surrounding a point magnetic charge.......................................338

30.2 Circle surrounding string........................................................................ 339
30.3 Contour for line integral in Problem 2................................................. 340

32.1 Electric and magnetic fields for a wave propagating in the direc­
tion n ................................................................................................... . 353
LIST OF FIGURES xv

32.2 Orientation o f acceleration at emission time to direction of ob­

servation. . ................ ......................................................................... 355

34.1 Geometry of linear antenna.................................................................368

34.2 Radiation pattern produced by a short antenna, / < A. In this
and the following figures axial symmetry about the z-axis is to
be understood.......................................................................................369
34.3 Radiation from an antenna of intermediate length, A < / < 2A.
This diagram is meant to be understood schematically only. Ac­
tually, the side lobes are far smaller than indicated........................ 370
34.4 Radiation pattern produced by center-fed antenna for kl = 7r.
This is called a half-wave antenna because / = A /2............ . . . . 373
34.5 Radiation pattern produced by center-fed antenna for kl = hit. . 373

35.1 Lorentzian line shape for the energy radiated per unit frequency
range...................................................................................................... 382

36.1 Cone of Cerenkov light produced at a definite wavelength by a

particle traveling with a velocity v, v greater than the speed of
light in the medium.............................................................................388

37.1 Simple model of a linear accelerator................................................. 393

37.2 Spectral distribution of energy radiated by accelerator in Fig.37.1.394
37.3 Impulsive scattering.............................................................................396

38.1 Precession of velocity vector around a constant magnetic field. . 402

38.2 Diagram of charged particle moving in a circle in a magneticfield. 403

39.1 Polarization vectors for synchrotron radiation................................ 414

40.1 Change of contour used in evaluating (40.19)................................. 420

40.2 Stationary phase contour for evaluation of (40.24)......................... 421
40.3 Sketch of power emitted into mth harmonic as a function of m.
What is actually plotted is for /3 = 0.99. In this
case mc == 356....................................................................................... 422
40.4 Angular distribution of synchrotron radiation for m = 100 and
¡3 = 0.99, from (38.42)......................................................................... 424
40.5 Directionality of synchrotron radiation............................................ 425

41.1 Interpretation of direct and reflected waves. . .................................. 434

41.2 Incident, reflected, and transmitted propagation vectors.............. 434
41.3 Relation between field orientations for incident and reflectedwaves.439
41.4 Parallel dielectric backed by conductor............................................ 443
41.5 Three parallel dielectrics.................................................................... 444

43.1 Contour for evaluating (43.20)......................................................... 452

xvi LIST OF FIGURES

44.1 Cylindrical waveguide with arbitrarily shaped surface S ................461

44.2 Cross section of waveguide consisting of inner and outer conductor.463

45.1 Scattering of a wave polarized in the scattering plane.................... 473

47.1 Plane wave incident on an aperture. Here, the z direction is out

of the p a g e ...........................................................................................492
47.2 Image charge induced by a point charge adjacent to a perfectly
conducting plane......................................................................................492
47.3 Geometry of diffraction by a circular aperture....................................495
47.4 Behavior of 2J\(z)/z and the differential cross section (normal­
ized to unity at 0 = 0) given by (47.29) as a function of z — kaO. 499
47.5 Geometry for diffraction by a slit..........................................................501
47.6 Differential cross section for diffraction by a wide slit, a X . . 502
47.7 Magnitude o f the electric field resulting from diffraction by a
straight edge.............................................. 504
47.8 Diffraction by a slit as seen on a nearby screen.................................. 506
47.9 Diffraction pattern far away from a slit............................................... 506

48.1 Geometry of diffraction by a perfectly conducting straight edge.

The 2 axis is perpendicular to the page............................................ 509
48.2 Diffraction by a straight edge at a fixed angle 0................................. 512
48.3 Cylindrical coordinate system for diffraction by a straight edge. . 518
48.4 Geometry of a slit....................................................................................519

49.1 Superposition of solutions for Ez polarization.................................... 523

49.2 Electric fields for x > 0, for the Ez polarization................................. 524
49.3 Superposition of solutions for the Bz polarization................. 524
49.4 Magnetic fields for x > 0, for the Bz polarization.............................. 525
49.5 Equivalence between dual diffraction problems...................................525

50.1 Scattering of a scalar field <f> by an object O. . ............................... 536

52.1 Contour used to re-express (52.13)..................................................... 548

52.2 Energy loss by a charged particle, incorporating (52.23) and (52.45).
Here x = pz, where p is the density of the medium, and the ab­
scissa is /?7 = ( v /c ) ( l —(v /c )2)-1 / 2. We have taken representative
values for the parameters, Kc/ue = 100, w\¡u¡* = 30, and the
damping constant 7 = 0.......................................................................552
Preface

This book is an outgrowth of lectures that Julian Schwinger gave at UCLA in

1976, when, for the first time in many years, he was asked to teach Classical Elec­
trodynamics. Of course, Schwinger had contributed greatly to the development
of the subject, partly through his profound work on the theory of waveguides
and microwave cavities at the MIT Radiation Laboratory during World War
II. Only a small part of that material was ever published, in Discontinuities in
Waveguides.1 After the war, Schwinger played a key role in the perfection of
the synchrotron, and, especially, was largely responsible for the theory of syn­
chrotron radiation.2 His first course at Harvard in 1946, Applied Science 33,
was on advanced applications of electromagnetic theory, particularly to wave­
guides. (We have incorporated four problems from the final exam to that course
in the present volume.) Schwinger’s discoveries in classical electrodynamics led
directly to his solution of the difficulties in quantum electrodynamics a few years
later.
As his former graduate students at Harvard, and postdoctoral fellows work­
ing with him, we attended the UCLA lectures and found them so novel and
exciting that we proposed turning them into a textbook. Schwinger agreed,
and cooperated to the extent of supplying us with his detailed notes. By 1979
a typed manuscript existed, and a publication contract was signed. At that
point, Schwinger began to read the manuscript carefully, and not finding his
voice there, began an extensive process of revision. That revision, of the first
half of the manuscript, continued until 1984. He used the partially revised
manuscript as the basis for his Classical Electrodynamics course he taught at
UCLA again in 1983. We had gone our separate ways long before that point,
so the project lay dormant for a decade.
Shortly after Schwinger’s death in 1994, one of us (KAM) began teaching
Classical Electrodynamics, and used the opportunity to begin completion of the
manuscript using the extant materials. The present volume is the result of that
effort. We offer it now, not merely as a homage to a great physicist and teacher,
but as a vital approach to a fundamental, and still not closed, subject. We

1J. Schwinger and D. Saxon, Discontinuity in Waveguides (Gordon and Breach, New York,
1968).
2 “ On Radiation by Electrons in a Betatron,” unpublished (1945) [transcribed by M. A.
Furman, LBNL-39088, July 1996]; “On the Classical Radiation of Accelerated Electrons,”
Phys. Rev. 75, 1912 (1949).

XVII
have retained our original organization of the manuscript, and not the few, very
long chapters of Schwinger’s later revision, because we feel that this makes the
contents more accessible to the student and teacher. We have tried to retain
Schwinger’s inimitable lecturing style in this book, in which everything flows
inexorably from what has gone before. However, as an aid to the reader, we
have included a Reader’s Guide, which identifies major themes in each chapter,
suggests a possible path through the book, and identifies topics for inclusion or
exclusion from a given course, depending upon circumstances.
We dedicate this book to the memory of Julian Schwinger. We are indebted
to Clarice Schwinger for her gracious permission to pursue this project, and
for her encouragement. We are also grateful to Professor Michael Strauss, who
attended Schwinger’s 1983 course, and made available his notes, and the corre­
sponding book manuscript at that time. Further materials have been obtained
from the Julian Schwinger Papers (Collection 371), Department of Special Col­
lections, University Research Library, University of California, Los Angeles.
Many students at the University of Oklahoma contributed by finding innumer­
able errors in the typescript, so we thank them as well.
Finally, we acknowledge the Alfred P. Sloan Foundation and the U. S. Na­
tional Science Foundation for partial financial support during the early stages
of this project, and the U. S. Department of Energy for partial support of KAM
during the completion of this project.

Lester L. DeRaad, Jr. Tustin,CA

Kimball A . Milton Norman,OK
Wu-yang Tsai Torrance,CA
December 1997

X V lll
Reader’s Guide

8+ 4(/ ; 8M
ther included, in its original incarnation, two chapters, now Chapters 51 and
52, which were based on lectures Schwinger gave, around the same time, on the
energy loss by electrically and magnetically charged particles when they pass
through matter. This latter work was evidently based on the excitement gener­
ated by the purported discovery of magnetic charge in cosmic ray experiments,3
evidence that was subsequently interpreted in terms of a conventional but rare
nuclear fragmentation event.4 The manuscript prepared during the 1977-79
period remains the core of the present volume.
However, as noted, Schwinger undertook extensive revisions beginning in
1979, and continuing at least through 1983. Indeed, the UCLA archives contain
an extensive revision of the book through electrostatics, corresponding to the
material in the present Chapters 1-25. In the interest of preserving the more
spontaneous flavor of the original lectures, we have retained most of the origi­
nal material, but incorporated new material at the appropriate points. A few
chapters were not in the original version and are based closely on the 1979-83
revision; they are Chapter 10, “Einsteinian Relativity,” Chapter 18, “Modi­
fied Bessel Functions,” Chapter 19, “Cylindrical Conductors,” and Chapter 25,
“Modes and Variations.” Schwinger ceased work on the project in 1984, but
he did present new material on radiation in his UCLA lectures in the previous
year, and that new material now appears in Chapter 44, “Waveguides,” and
Chapter 46, “ Partial-Wave Analysis of Scattering.” Of course, the present au­
thors have incorporated numerous improvements and additional material. Most
significant is the inclusion of a great many problems, of a wide range of diffi­
culty, some of which had been prepared by us in the late 1970s, but many of
which are new. The problems constitute an integral part of the text; indeed
many important concepts appear only in the problems, and an instructor might
feel it appropriate to base lectures on those topics. Examples include covariant
notation, introduced in Chapter 10, and vector spherical harmonics, introduced
in Chapter 50.
In view o f this long history, the reader will not be surprised to detect a certain
variation in the level of the material. This seems to us entirely appropriate.
3P. B. Price, E. K. Shirk, W . Z. Osborne, and L. S. Pinsky, Phys. Rev. Lett 35, 487 (1975).
4L. W . Alvarez, in Proceedings of the 1975 International Symposium on Lepton and Photon
Interactions at High Energies, Stanford, August 21-27, 1975, ed. W . T. Kirk, p. 967.

XIX
Some o f the topics are quite elementary, but there are many which are more
advanced, and which therefore could be omitted in a first course. Indeed, it is
the authors’ experience that there is far more material here than can be covered
in a typical two-semester graduate electrodynamics course. It is hoped that the
more advanced material will be of utility to practicing researchers, and indeed,
the authors have found the material contained herein extremely valuable in their
own research in electromagnetic theory and quantum field theory.
The balance o f this Guide will sketch major subjects covered in each chapter,
and make suggestions for material that is essential, and inessential, for a first
course or reading.
C h a p ter 1. The central point of this chapter is a hueristic derivation of
Maxwell’s equations starting from Coulomb’s law together with the imposition
of Galilean invariance. This is not to be taken as a rigorous derivation, especially
because the correct relativity group is that of Einstein. Essential.
C h a p ter 2. Here we show how Maxwell’s equations are modified if magnetic
charge is present. Inessential.
C h a p ter 3. We discuss the conservation of energy, momentum, and an­
gular momentum, thereby establishing the consistency of Maxwell’s equations.
Essential.
C h a p ter 4. This chapter is devoted to the inference of the macroscopic
Maxwell equations. Essential.
C h a p ter 5. Here we present simple classical models for the electric perme­
ability and conductivity, and discuss the Clausius-Mossotti equation. Essential.
C h a p ter 6. We give models for the magnetic properties of matter here, and
introduce the vector potential. Essential.
C h a p ter 7. We treat the somewhat subtle issues of how to construct the
energy and momentum for macroscopic electrodynamics. Essential.
C h a p ter 8. Mechanical action principles are the subject here. Although
these should be familiar, the viewpoint is somewhat novel and crucial in the
following. Essential.
C h a p ter 9. Here we present the action principle for electrodynamics. Es­
sential, except for Section 9.6.
C h a p ter 10. We discuss the modification of the particle Lagrangian re­
quired by Einstein’s relativity, as well as the usual relativistic kinematics and
field transformations. Probably only Section 10.1 is essential.
C h a p ter 11. Here the study of electrostatics begins. We show that the
general action principle implies the stationary principles for electrostatics. Es­
sential.
C h a p ter 12. We introduce Green’s functions. Essential.
C h a p ter 13. We discuss Green’s function for free space. Essential.
C h a p ter 14. We derive Green’s function for a semi-infinite dielectric. Only
Sections 14.1 and 14.4 are essential.
C h a p ter 15. Here, we use the previous Green’s function to compute the
force between a point charge and a dielectric slab. Inessential.
C h a p ter 16. Now, we introduce Bessel functions. As they are used through­
out the following, this is essential.

xx
 C h a p ter 17. We give the derivation of Green’s function for parallel con­
ducting plates, along with an extensive discussion of image charges in this case.
Only Section 17.1 is essential.
C h a p ter 18. Now, we introduce modified Bessel functions. This material
is very useful, but perhaps inessential for a first reading.
C h a p ter 19. We discuss cylindrical conductors, of rectangular, triangular,
and circular cross section, and derive useful mathematical results. Inessential,
C h a p ter 20. Now we introduce spherical harmonics. Essential,
C h a p ter 21. Legendre’s polynomials and spherical Bessel functions are the
principal topics here. Only the beginning through Section 21.1, and perhaps
Section 21.3, are essential.
C h a p te r 22. The general multipole expansion for the energy of interac­
tion of two arbitrary bounded charge distributions is the subject. At least the
beginning of this chapter is essential.
C h a p ter 23. We consider Green’s functions for conducting and dielectric
spheres. The first parts of Sections 23.1 and 23.4 seem essential.
C h a p ter 24. We give a general treatment of electrostatics in the presence of
dielectrics and conductors . We prove Thomson’s theorem, and give the general
expression for capacitance. At least the latter, given in Section 24.6, is essential.
C h a p ter 25. Variational methods for estimating eigenvalues are the sub­
jects treated here. Elegant, but inessential.
C h a p ter 26. We now introduce the variational principle for magnetostatics.
Essential.
C h a p ter 27. The energy of interaction of steady currents, and inductance,
are the subjects here. Essential.
C h a p ter 28. Here the topic in magnetic dipoles. Essential.
C h a p ter 29. We introduce the magnetic scalar potential. Inessential.
C h a p ter 30. Here, the subject is the “string” required for the definition of
the vector potential when magnetic charge is present. Inessential.
C h a p ter 31. Now the treatment of radiation begins, with the retarded
Green’s function and the Lienard-Wiechert potentials. Essential.
C h a p ter 32. We discuss the asymptotic fields, and dipole radiation. Es­
sential.
C h a p ter 33. Here we study radiation from the point of view of energy
transferred from the source, and derive the Darwin-Breit Hamiltonian for the
charges. Inessential.
C h a p ter 34. We discuss radiation from simple models of antennas. Inessen­
tial.
C h a p ter 35. We derive the general formula for the spectral distribution of
radiation. Essential.
C h a p ter 36. Now, we obtain a formula for the power spectrum, and apply
it to Cerenkov radiation. Essential.
C h a p ter 37. We discuss radiation in the extreme situations of constant
acceleration, and of impulsive scattering. In the problems, we give general
formulas for the energy radiated by a particle undergoing arbitrary acceleration
for a finite amount of time. Inessential.

xxi
 C h a p ter 38. Synchrotron radiation is the subject here. Essential.
C h a p ter 39. We derive the power spectrum for the different polarization
states in synchrotron radiation. Inessential.
C h a p ter 40. The high energy regime for synchrotron radiation is the sub­
ject here. Sections 40.1 and 40.5 are essential.
C h a p ter 41. We derive Snell’s law, and the reflection and transmission
coefficients for light incident on a plane dielectric interface. Mostly essential.
C h a p ter 42. We treat reflection by a conductor having finite conductivity
here. Inessential.
C h a p ter 43. Here we perform the 2 + 1 dimensional break up of Green’s
function for Helmholtz’ equation leading to Hankel functions. Inessential.
C h a p ter 44. Here we give a general treatment of cylindrical waveguides,
harking back to Chapters 19 and 25. Inessential.
C h a p ter 45. We present simple models of scattering, including Thomson
scattering. Essential.
C h a p ter 46. We give here a partial-wave approach to scattering. Although
this is very useful for quantum mechanics, it is here inessential.
C h a p ter 47. The elementary theory of diffraction is the subject. At least
Sections 47.1 and 47.2 are essential.
C h a p ter 48. Here we show that the diffraction of a normally incident plane
wave on a straight edge is exactly solvable. Inessential.
C h a p ter 49. We derive Babinet’s principle in the case of a plane with
an aperture or slit to give the diffraction for the opposite polarization state.
Inessential.
C h a p ter 50. The general formulation of scattering, the Born approxima­
tion, and the optical theorem are the subjects. Essential.
C h a p ter 51. We derive general dispersion relations for the dielectric con­
stant from causality requirements. Inessential.
C h a p ter 52. We give a general treatment of energy loss by an electrically
or magnetically charged particle traversing matter. Inessential.

XXII
Chapter 1

Maxwell’s Equations

The teaching of electromagnetic theory is something like that of American His­

tory in school; you get it again and again. Well, this is the end of the line.
Here is where we put it all together, and yet, not quite, since it is still classical
electrodynamics and the final goal is quantum electrodynamics. This preoccu­
pation reflects the all-pervasive nature of electromagnetism, with implications
ranging from the farthest galaxies to the interiors of the fundamental parti­
cles. In particular, the properties of ordinary matter, including those properties
classified as chemical and biological, depend only on electromagnetic forces, in
conjunction with the microscopic laws of quantum mechanics.

1.1 Electrostatics
Our intention is to move toward the general picture as quickly as possible, start­
ing with a review of electrostatics. We take for granted the phenomenology of
electric charge, including the Coulomb law of force between charges of dimen­
sions that are small in comparison with their separation. This is expressed by
the interaction energy, E ) of a system of such charges in otherwise empty space,
a vacuum:
&ct&b
E I v a-
Tab
a*b
where ea is the charge o f the ath particle while

rab = |ra-r*| (1.2)

is the separation between the ath and 6th particles. (Throughout this book
we use the Gaussian system of units. Connection with the SI units will be
given in Appendix A.) As we shall see, this starting point, the Coulomb energy
(1.1), summarizes all the experimental facts of electrostatics. The energy of
interaction of an individual charge with the rest of the system can be emphasized

1
2 CHAPTER 1. M AXW ELL’S EQUATIONS

by rewriting (1.1) as

(1.3)
a bjia a

where we have introduced the electrostatic potential at the location of the ath
charge that is due to all the other charges,

This is an action-at-a-distance point of view, in which the charge at a given

point interacts with charges at other, distant points. Another approach, which
generalizes and transcends action at a distance, employs the field concept (due
to Faraday), a field being a local quantity, defined at every point of space.
We take a first step in this direction by considering the potential as a field,
which is defined everywhere, not just where the point charges are located. This
generalized potential function, or simply the potential, <j>{r), is

« r> = E j r r ^ ¡ ' <1 5 >

where we now treat every charge on an equal footing, which means that in (1.5)
we sum over all charges e¿. In terms of this potential, which is different from
<j)a, the energy E can be written as

16
( - )
a a

The last part of (1.6) is not to be understood numerically, but rather as an

injunction to remove those terms in the first sum that refer to a single particle.
In other words, we remove “self-action,” leaving the mutual interactions between
particles. The field concept naturally leads to self-action.
The notion of force is derived from that of energy, as we can see by consid­
ering the work done as a result of a spatial displacement. If we displace the ath
charge by an amount <$ra, the energy changes by an amount

6E = ( V aE) •6va = —F a •6ra, (1.7)

where F a is the force acting on the ath point charge. Comparing this with the
energy expression (1.1) we find the force on the ath particle to be

Fa = - V a J 2 — = -V «e „ E — = -V «e.*(r«). (1-8)
fe r‘ 4 f e r“

In the last form, we have substituted <£(ra) for <j>ay so it would appear that an
extra self-action contribution has been introduced. To see that this is not true,
we first argue physically that the difference between <£(ra) and <f>a is independent
1.1. ELECTROSTATICS 3

of position, and so self-action does not contribute to the force. Mathematically,

what is this additional, unwanted, term? It is the negative gradient of the
self-energy:
—V e a (1.9)
r- 1*0 r-+r0

Can we make sense of this? We could define the limit here by arbitrarily adding
a displacement vector e of fixed direction to r a and letting its length approach
zero:
r = ra + c, € —►0, (1.10)
but at the cost o f picking out a particular direction. In order to remove the
most blatant aspect of this directional dependence, let us also approach r a from
the opposite direction,
r = ra - c , € —►0, (1.11)

and average over the two possibilities, so that the additional term (1.9) becomes

21 ( 1 .12)
a2

More elaborate limiting procedures, such as an average over all directions, can
be used, but the simple procedure of (1.12) suffices. Therefore, we can employ
^(r) in (1.8), with the implicit use of the two-sided limit, (1.12), to calculate
the force.
With the force given in terms of the gradient of a field (the potential), the
electric field E can now be defined by

E (r) = - V * ( r ) , (1.13)

so that the force on a point charge ea located at r a is

Fa = eaE(ra). (1.14)

The electric field E so introduced is a function calculable at r in terms of the

point charges located at r&,

e M = E '* ¡ 7 T ^ ¡ 3 - ( ' - 15>

As such, it remains an action-at-a-distance description, whereas, for many pur­

poses, it would be much more convenient to be able to completely characterize
the electric field by local properties. Such local statements will lead to differen­
tial equations, which, of course, must be supplemented by boundary conditions.
From its definition as the negative gradient of the potential, (1.13), the
electric field has zero curl:

V x E (r) = —V xV ^(r) = 0. (1.16)

4 CHAPTER 1. M AXW ELL’S EQUATIONS

Figure 1.1: A surface S bounding a volume V used in computing the electric

flux.

Besides the curl, the other elementary differential operation that can be applied
to a vector field is the divergence. To find V •E, we consider a related integral
statement. The integral of the normal component of E over a closed surface S
bounding a volume V is the electric flux (see Fig. 1.1):

r — Tb 1
j ) d S .E (r ) = ] C C4/ d S ‘ | r - r 6||r — rj|2 p dilb- (1.17)

Here, dS is an area element, directed normal to the surface, and d i is an

element of solid angle, which is defined in the following manner. The element
of area perpendicular to the line from the 6th charge is (see Fig. 1.2)
r - r&
dS (1.18)
| r -n | ’
which, when divided by the square of the distance from the 6th charge gives the
solid angle subtended by dS as seen from the 6th charge. There are now
two possible situations: either e¡> is inside, or it is outside the closed surface S',
as shown in Fig. 1.3. Correspondingly, the integral over all solid angles in the
two cases is
4ir if e& is inside S,
j> diift (1.19)
0 if e¡, is outside S.
Hence, the electric flux through a closed surface S is proportional to the enclosed
charge:
dS *E(r) = ^ 47re&. ( 1.20)
1 b in V
1.1. ELECTROSTATICS 5

Figure 1.2: Geometrical definition of solid angle.

a) b)

Figure 1.3: Topology if is inside (a) or outside (b) the surface S.

6 CHAPTER 1. M AXW ELL’S EQUATIONS

This is the theorem of Carl Friedrich Gauss (1777-1855).

With our aim of deriving local statements in mind, we generalize the idea
of point charges to that of a continuous distribution of charge, as measured by
p(r), the volume density of charge at the point r. Then, the total charge in a
volume V is obtained by integrating the charge density over that region:

ej = f (dr)p(r). (1.21)
6iny Jv

[Throughout this book we use the following notation for the element of volume:

(dr) = d x d y d z ] (1.22)

For point charges, the charge density is zero except at the location of the charges,

K r ) = ^ 2 ebS(r - r 6), (1.23)

b

where the three-dimensional (Dirac) 6 function is defined by

f , , v c/ v f 0 if Yb is outside V
r —r . ) = ( l i f r i i g i n s l d e V (1.24)

Then, the flux statement (1.20) becomes

4tt f (d r ) p(r) = / d S - E ( r ) = [ ( d r ) V - E ( r ) , (1.25)

Jv Js Jv

by use of the divergence theorem relating surface and volume integrals. (See
Problem 1.2.) Since (1.25) is true for an arbitrary volume V , the integrands of
the volume integrals must be equal, so we obtain the equation satisfied by the
divergence of E,
V •E (r) == 4irp(r). (1.26)
These differential equations for the curl and divergence of E, (1.16) and
(1.26), respectively, completely characterize E when appropriate boundary con-
ditions are imposed. It is evident from (1.15) that, for a localized charge dis­
tribution, the magnitude of the electric field becomes vanishingly small with
increasing distance from the collection of charges:

|E| —> 0 as r —* oo. (1.27)

One can also specify how rapidly this occurs. But it is remarkable that the weak
boundary condition (1.27) already implies a unique solution to the differential
equations (1.16) and (1.26). To show this, we suppose that Ei and E 2 are two
such solutions. The difference, E = Ei — E 2 satisfies

V * £ = 0, V x£ = 0 everywhere, (1.28)
\E\ —►0 as r —* 00, (1.29)
1.1. ELECTROSTATICS 7

from which we must prove that £ = 0. The identity

V x ( V x f ) = V ( V - f ) - V 2f , (1.30)

combined with the vanishing o f V x £ and V •£ , implies that

- V 2£ = 0. (1.31)

Let the single function £ (r) be any Cartesian component of the vector field
£; it obeys
—V 2£ (r) = 0. (1.32)
We present this as the everywhere valid statement

0 = - £ V 2£ = - V . (£ V £ ) + ( V £ ) 2, (1.33)

or
( V £ ) 2 — V 2-^£2 = 0. (1.34)
z
Now we integrate this over the interior volume V (R ) of a sphere of radius R
centered about an arbitrary point, which we take as the origin. The integral of
the second term in (1.34) is turned into an integral over the surface S(R) of the
sphere by means of the divergence theorem,

-/ (dr)V.(vi£2) = - / dS .V ^£2 = — á d S --£ 2 (1.35)

JV(R) ¿ J S( R) ¿ J S( R) db 8R T '

Using the relation between an element of area and an element of solid angle,
dS = R 2dily we can present this surface integral in terms of the average value
o f £ 2 over the surface o f the sphere,

(1.36)

And so the integral of (1.34) is

(1.37)

The decisive step now is to divide by the area 4irR2, and then integrate (1.37)
over R from 0 to oo:

f JnR)w <VÍ>’+1 =»' O '38»

which finally incorporates the boundary condition (1.29), that £ vanishes at all
infinitely remote points. Everything on the left side of (1.38) is non-negative, yet
it all adds up to zero. Accordingly, every individual contribution must be zero.
This tells us quite explicitly that £ = 0 at the origin, which is anywhere, and,
consistently, that V £ = 0 everywhere, or, that £ is a constant, which is required
8 CHAPTER 1. M AXW ELL’S EQUATIONS

to be zero by the boundary condition. This being true of any component, we

conclude that the vector 5 = 0. This completes our proof of the “uniqueness
theorem” of electrostatics, that the differential equations (1.16) and (1.26) have
a unique solution when the boundary condition (1.27) is imposed. (See Problem
1.3.)
From the Coulomb energy, we have thus derived the equations of electrostat­
ics:

V * E = 4np, ^ p = 0,
Q
V x E = 0, — E = 0, (1.39)

where the time independence has been made explicit. We are now going to
remove the restriction to static conditions by letting the charges move in a
particularly simple way. The equations of electromagnetism that emerge from
this discussion will then be accepted as applicable to more general motions, as
justified by various tests of internal consistency.

1.2 Inference of Maxwell’s Equations

We introduce time dependence in the simplest way by assuming that all charges
are in uniform motion with a common velocity v as produced by transforming
a static arrangement of charges to a coordinate system moving with velocity
—v. (We insist that the same physics applies in the two situations.) At first we
will take |v| to be very small in comparison with a critical speed c, which will
be identified with the speed of light. To catch up with the moving charges, one
would have to move with their velocity, v. Accordingly, the time derivative in
the co-moving coordinate system, in which the charges are at rest, is the sum
of explicit time dependent and coordinate dependent contributions,

(1.40)

so, in going from the static system to the uniformly moving system, we make
the replacement
d d d f .
S ~ 5 = « + v -V ' (1'41)
The equation for the constancy of the charge density in (1.39) becomes, in
the moving system

• = f b § - & + T -v * <i 4 2 >

or, since v is constant,
^ + V . ( v p ) = 0. (1.43)

We recognize here a particular example of the charge flux vector or the (electric)
current density j,
j = pv. (1.44)
1.2. INFERENCE OF M AXW ELL’S EQUATIONS 9

The relation between charge density and current density,

j ^ ( M ) + V .j(r ,t) = 0, (1.45)

is the general statement of the conservation of charge. Conservation demands

that the rate o f decrease of the charge within an arbitrary volume V must equal
the rate at which the charge flows out of the bounding surface S', that is

- l j v (d r)p (r,t) = j í d S .j( r ,í ) = ^ ( d r ) V •j( r ,i ) . (1.46)

Since V is arbitrary, the local conservation law, (1.45), follows. We also note
that the expression for the current density, (1.44), continues to be valid even
when v is dependent upon position, v —> v (r ,t). (See Problem 1.4.)
We can perform a similar transformation on the equation for the electric
field d'E/dt = 0; namely,

° = 5 E = f + <’ " V >E - O '47»

Making use of a vector identity, together with (1.26) and (1.44), (v is constant),

V x ( v x E ) = v ( V - E ) - ( v . V )E (1.48)
= v 47rp — (v •V )E
■= 4 7 r j - ( v V ) E , (1.49)

we find an equation relating E to the current density,

0 = ^ + 4*j-Vx(vxE). (1.50)

[Notice that by taking the divergence of (1.50) we recover the local charge conser­
vation equation (1.45), so that the conservation of charge is not an independent
statement.] The quantity v x E represents a new phenomenon combining the
effects of motion with those of electric charge. To describe this new, induced
effect, we define the magnetic induction1 B by

v x E = cB, (1.51)

where c is a constant having the dimensions of velocity (which will turn out to
be the speed of light). Expressed in terms of the magnetic field, (1.50) becomes
an equation determining the curl of B,
^ 1 d„ 4x . ^
V X B = - — E + — j. (1.52)
COt C
Next, we naturally ask for the divergence of B . According to the definition,
(1.51), we have

V •B = V ♦ Q x e ) = - ( ^ x v ) . E = ~ . ( V x E ) = 0, (1.53)

1We will usually call B the magnetic field, but see Chapter 4.
10 CHAPTER 1. M AXW ELL’S EQUATIONS

or
V •B = 0. (1.54)
Moreover, in the co-moving coordinate system where the charges are at rest—
static— the magnetic field should also not change in time:

S B = ¿ B + ( W ) B = 0, (1.55)

which becomes, when we use the identity in (1.48) as well as (1.54),

¥ = V x (v x B), (1.56)

consistent with V •B = 0.
What do we do now? We need one experimental fact. Light is an electro­
magnetic oscillation. The evidence for this is overwhelming. As examples, we
note that electric and magnetic fields are known to influence the emission, prop­
agation, and absorption of light; and that radio and infrared waves, which differ
only in wavelength from visible light, are emitted by electric charge oscillations.
What must be done so that this fact is built into the equations we are inferring?
The existence of electromagnetic waves means that the equations determining
the electric field have solutions of the form

E ~ i ( z — ct), (1.57)

where c is the speed of the waves. Such waves, propagating in the z direction,
satisfy the second-order differential equation

<92
(1.58)
d z ^ ~ C2 5 i2 E;
for an arbitrary direction of propagation, the corresponding wave equation is

V’ E = i ^ E . (1.59)

More precisely, we require that this equation should hold far from the charges
that produce the field. The left side of this equation can be written as [cf. (1.30)]

V 2E = - V x ( V x E), (1.60)

since V » E = 0 outside the charge distribution, while, by means of (1.52) and

(1.56), the right side becomes (j is zero outside the charge distribution)

? i ? E = j s v * B = J v x 'v * ( v x B ) i ' <i m >

This shows that the desired differential equation will hold if

E = - - xB. (1.62)
c
1.2. INFERENCE OF M AXW ELL’S EQUATIONS 11

But this cannot be a completely correct statement, since then v —> 0 would
require E —* 0. No electrostatics! However, all that is really necessary is that
the curl of this tentative identification be valid:

V x E = —V X i (1.63)

or, if we use (1.56),

1 fi
V x E = -~ B . (1.64)
c at
This is consistent with electrostatics since it generalizes V X E = 0 to the time-
dependent situation. The fact that V x E = 0 has been used before to derive
V •B = 0 is consistent here since the error is now seen to be of order (t>/c)2.
[See (1.53).]
Collecting the above relations, you will recognize that we have arrived at
Maxwell’s equations,

_ _ 1 9_ 47r .
V xB = E + — j, V •E = 47rp,
c at c
- V x E = — B, V « B = 0. (1.65)
c at
These equations of electromagnetism, as local, differential field equations, are
no longer restricted to the initial assumption of a common small velocity for all
charges.
To complete the dynamical picture we ask: What replaces (1.14) to describe
the force on an electric charge, when that charge moves with some velocity v
in given electric and magnetic fields E and B ? We consider two coordinate
systems, one in which the particle is at rest (co-moving coordinate system) and
one in which it moves at velocity v. Suppose in the latter coordinate system,
the electric and magnetic fields are given by E and B, respectively. In the
co-moving frame, the force on the particle is

F = eEeff, (1.66)

where Eeff is the electric field in this frame. In transforming to the co-moving
frame, all the other charges— those responsible for E and B — have been given
an additional counter velocity —v. We then infer from (1.62) that (v /c ) x B
has the character of an additional electric field in the co-moving frame. Hence,
the suggested Eeff is
Eeff = E H— XB, (1.67)
c
leading to the force law, due to Hendrick Antoon Lorentz (1835-1928),

F = e (E + ^ x B ) . (1.68)

These results, Maxwell’s equations, (1.65), and the Lorentz force law, (1.68),
have not been derived, but inferred from a special circumstance. We will adopt
12 CHAPTER 1. M A XW ELL’S EQUATIONS

these equations as describing the electromagnetic fields produced by, and acting
on, charges possessing arbitrary velocities, although the above discussion does
allow room for additional terms if v/c is no longer small. The fact that no
such terms are actually required is part of the implication of the special theory
o f relativity (see problem 1.6). We will prefer, instead, to show the physical
consistency of the equations as they stand (see Chapter 3).

1.3 Discussion
We have arrived at the Maxwell-Lorentz electrodynamics by combining three
ingredients: the laws of electrostatics; the Galileo-Newton principle of relativity
(charges at rest, and charges with a common velocity viewed by a co-moving
observer, are physically indistinguishable); and the existence of electromagnetic
waves that travel in a vacuum at the speed c. The historical line of development
was otherwise. Until the beginning of the nineteenth century, electricity and
magnetism were unrelated phenomena. The discovery in 1820 by Hans Christian
Oersted (1777-1851) that an electric current influences a magnet— creates a
magnetic field— is formulated, for stationary currents, in the field equation

^ « 4tt.
V x B = — j. (1.69)
c
The symbol c that appears in this equation is the ratio of electromagnetic and
electrostatic units of electricity (see Appendix A). Then, in 1831, Michael Fara­
day (1791-1867) discovered that relative motion of a wire and a magnet induces
a voltage in the wire— creates an electric field. Such is the content of

-V X E = Í ¿ B, (1.70)

which extends the magnetostatic relation

V - B = 0, (1.71)

that expresses the empirical absence of single magnetic poles. Finally, in 1864,
James Clerk Maxwell (1831-1879) recognized that the restriction to stationary
currents in (1.69), as expressed by V •j = 0, was removed in

_ Air. 1 d_
V xB = t j + -s e, (1.72)

when joined to the electrostatic equation

V •E = Airp. (1.73)

The deduction of the existence of electromagnetic waves that travel at the speed
c, in remarkable numerical agreement with the speed of light, was confirmed in
1867 by Heinrich Rudolf Hertz (1857-1894). It was the conflict between the
1.4. PROBLEMS FOR CHAPTER 1 13

existence of this absolute speed c and the relativity concept of Newtonian me­
chanics that set the stage for Einsteinian relativity. Already at the age of 16,
Albert Einstein (1879-1955) had recognized this paradox: To a co-moving New­
tonian observer, light waves should oscillate in space, but not move; however,
Maxwell’s equations admit no such solutions. Einsteinian relativity is an out­
growth of Maxwellian electrodynamics, not the other way about. That is the
spirit in which electrodynamics is developed as a self-contained subject in this
book.

1.4 Problems for Chapter 1

1. Verify the following identities explicitly:

A x ( B x C ) + B x ( C x A ) + C x (AxB) = 0,

V x ( A x B ) = A x ( V x B ) - B x ( V x A ) - ( A x V ) x B + (BxV)xA,

V •(AAxB) = A(B • V x A - A * V x B ) + A x B * V A .

2. Verify, using Cartesian coordinates, the divergence theorem,

V.E
Í dS *E,
where V is the volume contained within the closed surface 5, dS being
the surface element in the direction of the outward normal, and Stokes’
theorem,

[ dS •(VxE) = <f dl.E,

Js Jc
where C is the closed boundary of the open surface 5, and d\ is the
tangentially directed line element. The sense of the line integration is
given by the right hand rule. [That is, if the contour C is traversed in the
sense of the fingers of the right hand, the thumb points in the sense of the
orientation of the surface.]

3. This question has to do with the uniqueness theorem which follows from
(1.37).

(a) Directly from that equation, what assumption about |£(r)|, |r| —►'oo,
will produce the conclusion that £ = 0 everywhere?
(b) How would it work out if one had integrated this equation from R = 0
to oo, without dividing by R21
(c) How fast would ( £ 2) r have to fall off with R so that we could conclude
£ = 0 everywhere by simply taking R —►oo in (1.37)?
14 CHAPTER 1. M AXW ELL’S EQUATIONS

4. For an arbitrarily moving charge, the charge and current densities are

p (r,t) = eS(r - R (t)), j ( r ,t ) = - R (t)),

where R (t) is the position vector of the charged particle. Verify the state­
ment of conservation of charge,

0¡P(r’t) + V.j(r,t) = 0.

5. In a region where no charges are present, the potential satisfies Laplace’s

equation,
V 2<¿ = 0 .
Such a function is called harmonic. Show that in a region where the poten-
tial is harmonic, the potential nowhere assumes a maximum or minimum
value. Use this result to give another proof of the uniqueness theorem of
electrostatics proved in Section 1.1.

6. In this chapter we “derived” Maxwell’s equations by exploiting approxi­

mate Galilean invariance. However, we cannot push Galilean invariance
further, since it is not valid in 0 ( v 2/ c 2). The correct relativity is that of
Einstein. Verify that Maxwell’s equations are invariant under the trans­
formations of Einstein’s special relativity, as follows. Consider a Lorentz
transformation corresponding to a boost in the x direction, which on the
space-time coordinates is defined by

x'0 = i(x o - /Sail),

x\ = 7 (xi - flxo),
x'i — X2,

x'3 = x 3.

Here *o = ct, * 1 x, *2 = V, x 3 = z, and

? = -, 7 = (1 - P 2)~ 1/2,

v being the relative velocity of the two coordinate frames. We can regard
the four quantities Xp, p = 0, 1, 2, 3, as forming a four-vector. The four-
current jo = c/9, ji, i = 1 , 2 , 3 , constructed from the electric charge
and current densities, transforms by the same law:

j'o = rO'o - An).

j'i = y (h - Pío),
32 = h ,
is = is-
1.4. PROBLEMS FOR CHAPTER 1 15

On the other hand, the electric and magnetic field vectors are components
of a four-tensor, and so they have a more complicated transformation law.
Consider a boost by an arbitrary velocity v. Then the components of the
electric and magnetic fields in the direction of v do not change, while the
components in directions perpendicular to v are entangled:

Ei| = E,|, E1 = T ( e + ^ x b ) x ,

B' x = 7 ( B ~ xe )x .

For v = (v,0,0) verify explicitly that if Maxwell’s equations hold in the

unprimed frame, they hold in the primed frame as well, no matter how
near v may approach c. This was essentially the path by which Lorentz
and Poincare derived the transformation equations (but not the physics)
of special relativity. A more complete treatment of Einsteinian relativity
will be given in Chapter 10.

7. A charge e moves in the vacuum under the influence of uniform fields E

and B. Assume that E •B = 0 and that v •B = 0. At what velocity does
the charge move without acceleration? What is its speed when |E| = |B|?
Chapter 2

Magnetic Charge I

Our discussion in Chapter 1 contains a certain implicit assumption, When it

came to (1.62),
E = --X B , ( 2 . 1)
c
with its implication that static electric charges produce no electric field, we
knew better than to accept this and altered it to

VxE = - V x ( ^ x B ) , (2.2)

thereby admitting, for v = 0 , a static electric field, one obeying V x E = 0 .

Why then did we earlier accept without question the relation (1.51),

B = —xE, (2.3)
c

with its implication that all magnetic fields are due to the motion of electric
charges? This is the (1820) hypothesis of André Marie Ampere (1775-1836).
But is it true? An affirmative response is conventional, but the mathemati­
cal development allows a more general possibility. Again, all that was really
required in the above was the curl relation

V xB = V x (^ xe), (2.4)

admitting the possibility, for v = 0, of a static magnetic field obeying V X B = 0,

one that has its origin in magnetic charge. If pm is the density of such magnetic
charge, the analogy with electrostatics suggests that

V •B = 47rpm. (2.5)

The implication of (2.5) is that a further source of magnetic fields, other than
moving electric charges, could exist in magnetic charge. Whether this possibility
is realized in nature still awaits experimental confirmation.

17
18 CHAPTER 2. MAGNETIC CHARGE I

Further changes in Maxwell’s equations are required if magnetic charge ex­

ists. Since then V •B ^ 0, the co-moving time derivative of B becomes [cf.
(1-48)]

° = + (v •V )B = ^ B - V x ( v x B ) + v ( V . B). (2.6)

Then, using (2.5) and the magnetic current density j m, defined as

jm = PmV, (2.7)

together with (2.2), we obtain the following modified Maxwell equation

—V X E = ~ B + — j m. (2.8)
c at c

Notice that (2.8) implies the conservation of magnetic charge:

d
+ V •jm = 0. (2.9)

The complete set of Maxwell’s equations, when magnetic charge is present,

now reads

Id 47T.
VxB = E + -Je, V * E = 47TPe,
c at c
—V x B = ~ B + - j m, V 'B = 4irpm , (2.10)
c at c

where we have consistently used the subscript e to denote densities for electric
charge. Observe that these equations are invariant in form under the replace­
ments (duality transformation)

Pe * pm > E —» B, je ►j m,
Pm * ~Pe> B ¥ B, j m ►“ je* (2-11)

The generalized Lorentz force law, suggested by this symmetry, is

F = e(E + ^ x B )+ i(B -^ x E ), ( 2 .1 2 )

for a particle carrying both electric and magnetic charge, e and g ) respectively.
Although from time to time there have been spectacular reports of the dis­
covery of magnetic charge (Price, 1975; Cabrera, 1982), these “discoveries” were
never replicated, and serious objections were raised in each instance. Neverthe­
less, there are strong theoretical reasons to believe that magnetic charge exists
in nature, and may have played an important role in the development of the uni­
verse. Searches for magnetic charge continue at the present time, emphasizing
that electromagnetism is very far from being a closed subject.
2.1. A VERY BRIEF HISTORY OF MAGNETIC CHARGE 19

2.1 A Very Brief History of Magnetic Charge

It is said that Peregrinus in 1269 observed that magnets (lodestones) always have
two poles, which he called north and south. This was elevated to a “hypothesis”
by Ampere in the early 19th Century. The first theoretical calculation of the
motion of a charged particle in the presence of a single magnetic pole was
performed by Poincare in 1896 to explain recent observations. A few years
later, Thomson showed that a static system consisting of a magnetic pole and
an electric charge possessed an angular momentum— see Problem 3.8. It was
Dirac in 1931 who showed that magnetic charge was consistent with quantum
mechanics only if electric and magnetic charges were quantized: For a system
consisting of a pure magnetic charge g and a pure electric charge e, eg had to
be an integral (or half-integral) multiple of he. Many people have contributed
to the theory of magnetic charge subsequently; notable is the work of Schwinger
in the 1960s and 1970s, especially his concept of dyons, particles which carry
both electric and magnetic charge.
Many searches, both terrestrial and cosmic, have been carried out to find
magnetic monopoles in nature, but, so far, to no avail. Worth mentioning is the
induction technique of Luis Alvarez, et al. Positive reports were given by Price
in 1975 [cited in the Reader’s Guide] and by Bias Cabrera in 1982.1 These,
however, were never confirmed, and are no longer believed to offer any evidence
for magnetic charge, even by their authors.
However, modern unified theories of fundamental interactions typically imply
the existence of magnetic monopoles, or of dyons, often at extremely high mass
scales (~ 1016 GeV), but perhaps at nearly accessible energies (~ 10 TeV).
Moreover, there appears to be no reason why an elementary monopole or dyon
of the Dirac-Schwinger type could not exist. So experimental searches continue.

2.2 Problems for Chapter 2

1. Write Maxwell’s equations with magnetic charge in terms of

F = E -f iB , i = \/~T,
and related combinations of charge and current. Verify that these equa­
tions retain their form under the transformation illustrated by

F -* e -^ F ,
where <j>is an arbitrary constant. Express this as a transformation of E, B,
and the charge-current quantities. What is the geometric interpretation?
What is the particular form of this transformation when <j) = 7t/ 2 ?

2. Suppose every charged particle carried electric and magnetic charge in the
universal ratio gk/^k = A. Is there another way of looking at this situation
in which we would be unaware of magnetic charge?

1B. Cabrera, Phys. Rev. Lett. 48, 1378 (1982).

Chapter 3

Conservation Laws

In order to check the physical consistency of the above set of equations governing
Maxwell-Lorentz electrodynamics [(2.10) and (2.12) or (1.65) and (1.68)], we
examine the action of, and reaction on, the sources of the electromagnetic fields.
To be precise, we ask whether there is a correct balance in the exchange of
energy, momentum, and angular momentum between the charged particles and
the electromagnetic fields. As we shall see, the Maxwell-Lorentz system as it
stands implies the conservation of these mechanical properties, no matter how
rapidly the charges are moving.

3.1 Conservation of Energy

We start with a consideration of the rate at which work is done on the particles,
that is, the rate of energy transfer, or the power absorbed by the particles. For
one particle, we know that the rate at which work is done on it is

(3.1)

where we have used the Lorentz force law, (2.12), and the expressions for the
currents, (1.44) and (2.7), for a point particle. We interpret this equation as
meaning, even for general current distributions, that j e •E + j m •B is the rate of
energy transfer from the field to the particles, per unit volume. Then through
elimination of the currents by use of Maxwell’s equations, (2.10), this rate can
be rewritten as

(3.2)

The general form of any local conservation law, (1.45) or (1.46), suggests the
following interpretations:

21
22 CHAPTER 3. CONSERVATION LAWS

1. In the absence of charges ( je = j m = 0 ), this is the local energy conserva­

tion law
d E 2 + B 2 + v >JLE x B _ Q ( 3 .3 )
dt 87r 47t
We label the two objects appearing here as

E2 + B2
energy density = U = (3.4)
8 tt 5

energy flux vector = S = — E x B . (3.5)

47r
[The latter is usually called the Poynting vector, after John Henry Poynt-
ing (1852-1914).]

2. In the presence of charges, the relation (3.2) is

-f V •S -|-je •E 4* jm •B = 0, (3.6)

which, if we integrate over an arbitrary volume V, bounded by a surface

S, becomes

Jt j {dt) U + j i dS «S + j f (dr) (je •E + j m . B ) = 0. (3.7)

The three terms here are identified, respectively, as the rate of change
of the electromagnetic field energy within the volume, the rate of flow of
electromagnetic energy out of the volume, and the rate of transfer of elec­
tromagnetic energy to the charged particles. Thus, (3.6) gives a complete
description of energy conservation.

3.2 Conservation of Momentum

Next we consider the force on a particle, (2.12), as the rate of change of mo­
mentum,

F = e ( e + ~ x b ) + 0 (b —~ x e )

= J(,dv) (/oeE + - j ex B + pmB ~J»t» x e )

= J {d t)f, (3.8)

where f is the force density. Removing reference to the (generalized) charge

and current densities by use of Maxwell’s equations, (2.10), we rewrite the force
density f as

f = ¿ [E (V .E ) + B (V .B )]
3.2. CONSERVATION OF MOMENTUM 23

+i P[(-ilE
wD
+VxB) xB+Ex(4 Ib- VXE)]
1
<39>
= + t -[-E x ( V x E) + E(V «E) — B x ( V x B ) + B ( V ♦B)].
ut 4irc 4ir
The quadratic structure in E occurring here is

E2
— E x ( V x E ) + E(V •E) = - V — + (E. V)E + E(V.E)
¿i

= V . ( - 4 + E e ), (3.10)

which introduces dyadic notation, including the unit dyadic 1 , with components

(3n>
where Ski is the Kronecker 6 symbol. (See Problem 3.1.) The analogous result
holds for B. Accordingly, the force density is

d ExB / E2+ B2 EE + B B \
(3.12)
dt 47re \ 87t 4w )

We interpret this equation physically by identifying

ExB
momentum density = G = (3.13)
47rc
and
E2+ B2 EE + BB
momentum flux (stress tensor) = T = (3.14)
Sir 47T
When f = 0, we obtain the local statement of the conservation of momentum
of the electromagnetic field. A full account of momentum balance is contained
in
J -G + V ♦T + f = 0. (3.15)

The volume integral of this equation for electromagnetic momentum is inter­

preted analogously to the energy result, ( 3 .7 ).
The components of the stress tensor are given by

EkEi + BkBj
Tki = h\U - (3.16)
4ir
Notice that the stress tensor is symmetrical, Tki = T}fc, which, as we shall see
in the next section, is required in order to obtain a local conservation law for
angular momentum. The trace of T , the sum of the diagonal elements T**, is
simply the energy density, (3.4),

T rT = £ > * * = [/. (3.17)

k
24 CHAPTER 3. CONSERVATION LAWS

We also note that the Poynting vector, (3.5), is proportional to the momentum
density,
S = c2G, (3.18)
which has the structure of

energy density x velocity = c2(mass density x velocity). (3.19)

This is the first indication of the relativistic connection between energy and
mass, E = me2.

3.3 Conservation of Angular Momentum. Virial

Theorem
Having discussed momentum, we now turn to angular momentum. We will use
tensor notation to write (3.15) in component form,
A
^ G k + V iT lk + f k = 0, (3.20)

where we have also used the summation convention: Whenever an index is

repeated, a sum over all values of that index is assumed,
3
cljbj ~ —sl•b. (3.21)
¿=i
The rate of change o f angular momentum is the torque r , which, for one particle,
is
r = rx F = (3.22)

where the volume-integrated form is no longer restricted to a single particle. The

torque density, the moment of the force density, can be written in component
form as
(rXf)¿ — £ijk%jfk) (3.23)
where we have introduced the totally antisymmetric (Levi-Civita) symbol e ^ ,
which changes sign under any interchange of two indices,

^ijk — €jik — ^kji — Qkj — (3.24)

and is normalized by e \23 = 1. In particular, then, it vanishes if any two indices

are equal, e \\2 = 0, for example. The torque density may be obtained by first
taking the moment of the force density equation (3.20),

^ X j G t + V ,(x jT lk) - Tjk + Xjfk = 0, (3.25)

where we have noted that

V¡Xj — 6jj. (3.26)
3.4. CONSERVATION LAWS AND THE SPEED OF LIGHT 25

When we now multiply (3.25) with e,-,-* and sum over repeated indices, we find
that the terms involving spatial derivatives can be written as a divergence:

k%j G k ) “I" /(Qj>k%j T/fc)“1”U j k X j f k — 0* (3.27)

This final step is justified only because Tki is symmetrical (thus this symmetry is
required for the existence of a local conservation law of angular momentum). We
therefore icKvn.áfy the following electromagnetic angular momentum quantities:

angular momentum density = J — r x G (3.28)

angular momentum flux tensor = /C, ICij = CjkiXkTu. (3.29)

The interpretation of (3.27) as a local account of angular momentum conserva­

tion for fields and particles proceeds as before. (See Problem 3.5.)
Another important application of (3.25) results if we set j = k and sum.
With the aid of (3.17) this gives

¿ ( r . G ) + V . ( T t ) —tf + r . f = 0, (3.30)
at
which we call the electromagnetic virial theorem, in analogy with the mechanical
virial theorem of Rudolf Clausius (1822-1888). (See Chapter 8 .)

3.4 Conservation Laws and the Speed of Light

In this section, we restrict our attention to electromagnetic fields in domains
free of charged particles, specifically, moving, finite regions occupied by elec­
tromagnetic fields, which we will refer to as electromagnetic pulses. The total
electromagnetic energy of such a pulse is constant in time:

[ (dr) ! - [ / = - / (dr) V •S = 0, (3.31)

Jpulse ^ Jpulse

inasmuch as the resulting surface integral, conducted over an enclosing surface

on which all fields vanish, equals zero. Similar considerations apply to the total
electromagnetic linear and angular momentum,

4 p = / (* )¿G = - / ( * ) V . T = 0, (3.32)
at Jpulse 01 Jpulse

¿J = / (dr)¿(rX G ) = - / ( d r ) V •(—T x r ) = 0. (3.33)

^ Jpulse ^ Jpulse

With an eye toward relativity, we consider the space and time moments of
( 3 .6 ) and (3.15), respectively, combined as a single vector statement:

°=« (su+v-s)-A(sc‘+v'r“)' (3.34)

26 CHAPTER 3. CONSERVATION LAWS

outside the charge and current distributions. Exploiting the connection between
S and G [(3.18)], we can rewrite (3.34) as a local conservation law, much as the
equality of Tjk and Tkj lead to the conservation of angular momentum:

— { v U - c 2tG ) + V . ( S r - c 2¿T) = 0. (3.35)

When (3.35) is integrated over a volume enclosing the electromagnetic pulse,

the surface term does not contribute, and we find

■j- f (dr) (r U — c2tG ) = 0. (3.36)

dt Jpulse

The volume integral of the momentum density is the total momentum P,

G = P, (3.37)

which as noted in (3.32) is constant in time. Consequently, we can rewrite (3.36)

as
i f (dr)rU = c 2P , (3.38)
at Jpulse
where the integral here provides an energy weighting of the position vector, at
each instant of time,

/ (dr) rU (r, t) — E (r)E(t), (3.39)

«/pulse

where, as in (3.31), the energy E is

E= f (dr)U. (3.40)
Jpulse
Thus the motion of this energy-centroid vector is governed by

!!« * (< ) = p, (3«)

which is to say that the center of energy, (r)£ (t), moves with constant velocity,

|<r>j*(<) = v js , (3.42)

the total momentum being that velocity multiplied by a mass,

m = E/c2. (3.43)

The application of the virial theorem, (3.30), to an electromagnetic pulse

supplies another velocity. We infer that

£
E. (3.44)
dt
3.4. CONSERVATION LAWS AND THE SPEED OF LIGHT 27

By introducing a momentum weighting for the position vector,

/ (dr) r •G (r, t) = (r)P(<). P , (3.45)

J pulse

we deduce that the center of momentum moves with velocity

(3.46)

«S.

II
which is constant in the direction of the momentum,

(3.47)

>

II
ft,
We combine (3.47) with (3.41) to yield

vP •v e = c2. (3.48)

If the flow of energy and momentum takes place in a single direction, it would
be reasonable to expect that these mechanical properties are being transported
with a common velocity,
V p = v P = v, (3.49)
which then has a definite magnitude,

v •v = c2, v = c, (3.50)

which supplies the physical identification of c as the speed of light. Of course,

this identification was an input to our inference of Maxwell’s equations. We
here recover it from a consideration of energy and momentum, thus indicating
the consistency o f Maxwell’s equations. The relation between the momentum
and the energy of this electromagnetic pulse is then

E = v P, P = 4 V> (3.51)
cl
so we learn that p
E - Pc, v = c—, (3.52)

which results express the mechanical properties of a localized electromagnetic

pulse carrying both energy and momentum at the speed of light, in the direction
of the momentum.
There is another, somewhat more direct, mechanical proof that electromag­
netic pulses propagate at speed c. When no charges or currents are present, the
local equation of energy conservation, (3.3), implies

(r 2 - cH2) = 0, (3.53)

which can be rewritten, using (3.18), as

28 CHAPTER 3. CONSERVATION LAWS

Integrating this over all space and using the idea of energy and momentum
weighting to define averages, as before, we obtain

^ [(r 2)je(t) - c 2t2] E = 2c2 [(r)p(t) •P - t E ] . (3.55)

According to (3.47), the combination appearing on the right is a constant of

the motion, which we can put equal to zero by identifying the coordinate origin
with (r)p at t = 0. The time integral of this equation is then

( v2) e = (ct ) 2 -f constant, (3.56)

which implies, for large times,

« r a)* (< )) 1 / a ~ c t ; (3.57)

the center of energy of the pulse moves away from the origin at the speed of
light.
What are the fields doing to enforce the conditions (3.51) of simple mechan­
ical flow in a single direction? The relation between momentum and energy,
(3.52), E = |P|c, can be expressed in terms of the fields as

E2+ B2 ExB
(3.58)
87T 47r

where the volume integrations are extended over the pulse. Now, a sum of
vectors of given magnitudes is of maximum magnitude when all those vectors
are parallel, which is to say here that

f(d r ) < J (d r ) |ExB|, (3.59)

where equality holds only when E x B everywhere points in the same direction,
that of the pulse’s total momentum or velocity. On the other hand, we note the
inequality,

(ExB)2 = £ 2B 2 -(E.B)2
2
/ E2 + B2
V 2
+ (E ♦B)2

where the equality holds only if both E •B = 0 and E 2 = B 2. So we deduce

the opposite inequality to (3.59),

/( d r ) | E x B | < j (dr) E 2 ± -?--. (3.61)

Comparing (3.59) and (3.61), we see that both equalities must hold, so that

E •B = 0, E 2 = B 2, (3.62)
3.5. PROBLEMS FOR CHAPTER 3 29

Figure 3.1: Electric and magnetic fields for an electromagnetic pulse propagating
with velocity v.

and E x B is unidirectional, pointing in the direction of propagation. Accord­

ingly, the electric and magnetic fields in a unidirectional pulse are, everywhere
within the pulse, of equal magnitude, mutually perpendicular, and perpendicu­
lar to the direction of motion of the pulse. (See Fig. 3.1.) These are the familiar
properties of the electromagnetic fields of a light wave, which are here derived
without recourse to explicit solutions to Maxwell’s equations.

3.5 Problems for Chapter 3

1. The unit dyadic 1 is defined in terms of orthogonal unit vectors i, j, k by

1 = i i + j j + kk.

Verify that (A is an arbitrary vector)

A •1 = A, 1 •A = A, 1-1 = 1.

Repeat, using components, i.e., A * B = Expand the following

products of vectors with dyadics:

A •(BC), (AB).C, Ax(BC), (AB)xC.

2. Let A (r ) and B (r) be vector fields. Show that

V •(AB) = (A •V)B + B ( V .A).

Let A(r) further be an arbitrary scalar function. Simplify

V-(AAB), V-(AAxB).

3. An infinitesimal rotation is described by its effect on an arbitrary vector

V by
¿V = <5u>xV,
30 CHAPTER 3. CONSERVATION LAWS

where the direction o f Su> points in the direction of the rotation, and has
magnitude equal to the (infinitesimal) amount of the rotation. Check that

5(V2) = 0.

The statement that, if B and C are vectors, so is BxC, is expressed by

Í(BXC) = 6 B X C + B x 6 C = ^ x ( B x C ) .

Verify directly the resulting relation among the arbitrary vectors.

4. Verify the following relations for the electromagnetic stress tensor:

(a)
T tT = Tkk = U,

0 >)
T rT 2 = TkiTik = W 2 - 2(cG )2 > U2,

and

(c)
detT = —U[U2 — (cG )2].

Here the summation convention is employed, and the trace and determi­
nant refer to T thought of as a 3 x 3 matrix.

5. Show that the angular momentum conservation law derived in Section 3.3
can be written as
J p 7 + V-JC + rxf = 0,

where the angular momentum density is

J = rxG,

and the angular momentum flux tensor is

K = — Txr,

the cross product referring to the second vector index of T.

6. What if (r)p(O) ^ 0 in (3.55)? Show that the integral of that equation

can be interpreted in analogy with a group of particles that, at time t = 0 ,
are set off with various positions and velocities, thereafter to move with
those constant velocities,

r(<) = r ( 0 ) + vt.

Square and average this position vector, and upon comparison with the
solution of (3.55), identify (r(0) •v) and v.
3.5. PROBLEMS FOR CHAPTER 3 31

7. As in Problem 2.1, let

F = E -H B , F* = E — ¿B.

Identify the scalar

8^ - F’
the vector
1
tF* XF,
&7TÍ
and the dyadic
— (FF* + F*F).

What happens to these quantities if F is replaced by e“ *^F, 4> being a

constant?

8. Electric charge e is located at the fixed point |R. Magnetic charge g is

stationed at the fixed point —|R. What is the momentum density at the
arbitrary point r? Verify that it is divergenceless by writing it as a curl.
Evaluate the electromagnetic angular momentum, the integrated moment
of the momentum density. Recognize that it is a gradient with respect
to R . Continue the evaluation to discover that it depends only on the
direction of R , not its magnitude. This is the naive, semiclassical basis
for the charge quantization condition of Dirac,

n t,
eg — —he.
y 2
Chapter 4

Macroscopic
Electrodynamics

4.1 Force on an Atom

The Maxwell-Lorentz system of equations, (1.65) and (1.68), provides a micro­
scopic description of electromagnetic phenomena, at the classical level, ranging
from the simplest two-particle system to the detailed behavior of all particles in
a macroscopic system. However, for the latter case, we usually do not require
such a complete description, since our measurements involve macroscopic quan­
tities which are only indirectly related to the microscopic behavior of individual
atoms. We must develop a theory that is directly applicable to the macroscopic
situation with only an implicit reference back to the detailed characterization of
the system. The resulting macroscopic electrodynamics is a phenomenological
theory, by which is meant a theory that operates at the level of the phenom­
ena being correlated and predicted, while maintaining the possibility of contact
with a more fundamental theory— here, microscopic electrodynamics— that op­
erates at a deeper level. That contact exists to the extent that the macroscopic
measurements can be considered to be averages, over very many atoms, of the
results of hypothetical microscopic measurements.
To begin, we consider an atom, an electrically neutral assembly of point
charges,
£ e« = 0> (4-1)
a
that are bound together in a small region. We want to study the response of
such a system to external electric and magnetic fields that vary only slightly
over the spatial extent of that system. We will first concentrate our attention
on the net force on the system at a given time, the sum of the forces on its
constituents, ( 1 .6 8 ),

F= £ [e«E(rJ + ea^ X B ( r j] . (4.2)

33
34 CHAPTER 4. MACROSCOPIC ELECTRODYNAMICS

Since the system is small, all the charges are near the center of mass of the
charge distribution, which lies at the position R . (For the purposes of the fol­
lowing expansion we could let R represent an arbitrary point inside the charge
distribution; the use of the center of mass allows us to separate intrinsic proper­
ties from those due to the motion of the atom as a whole.) We can then expand
the electric and magnetic fields about this reference point,

E(r0) = E(R) + [(ra — R) •V]E(R) + . . . , (4.3)

and likewise for B, in which the subsequent terms are considered negligible.
Here V means the gradient with respect to R . Now, the total force on the
atom, (4.2), can be rewritten in terms of this expansion as

F = ^ e aj E ( R ) + ^ e 0 [(ra - R ) . V ] E ( R ) + xB(R)

+ X > « ^ x [ ( r a - R ) * V ] B ( R) + . . . . (4.4)
a

The first term here is zero because of the neutrality of the system, (4.1). In the
second term, we identify the electric dipole moment, d,

d= ^ ea(ra - R ) = ^ e° r a, (4.5)
a a

(which is independent of R ), while in the third, we recognize its time derivative,

5 > v« = £ d - (4.6)

Momentarily setting aside the fourth term, we find the force on the system
to be
F = (d •V )E (R ) + i ( ^ d ) X B (R ) + . . . . (4.7)

For the second term here, we can transfer the time derivative,

í ( | d ) XB(R) = i ¿ [ d x B ( R ) ] - id X ( | + V . v ) B(R), (4.8)

where V = dR/dt. Using (1.64) for <9B/3t, and rewriting the resulting double
cross product according to

d X ( V x E(R)) + (d •V)E(R) = V(d •E(R)), (4.9)

we can present (4.7) as

F = V[d -E(R)] - !(V •V)dxB(R) + ~ [ d x B ( R ) ] + .... (4.10)

c c at
4.1. FORCE ON AN ATOM 35

Recalling that force is the time rate of change of momentum, we see that
( l / c ) d x B introduces a redefinition of the momentum of the system.
We now return to the fourth term of (4.4), which would seem to correspond
to a small effect, since for atomic systems, va/c < 1. A rearrangement of it is

£ e« v x - R ) •V 1B (R ) = E 7 V x Kr <* - R ) •V !B (R )
a a

+ £ ^ ( v . - V ) x [ ( r a-R ).V ]B (R )> (4-H)

where, recalling the definition of the electric dipole moment (4.5), we can express
the first term on the right side as

i v x ( d - V )B (R ). (4.12)

Combining this contribution with the second term on the right side of (4.10),
and using (1.54), we obtain

-[(d * V )V — (V •V )d ]x B (R ) = i [ ( d x V ) x V ] x B ( R )
c c

-(d x V ).B (R ) (4.13)

c

Collecting the various results to this point, we can now rewrite the total force
on the atom, (4.4), as

F = V [d •E(R)] + V -(d X V ) •B (R ) £ -d x B (R ) + Fb , (4.14)

dt
where Fjg represents the second term on the right side of (4.11),

FB = £ — (va - V ) X [(ra - R) •V ]B (R ), (4.15)

which can be rearranged as follows:

F b = Jt £ ~ ( ra - R) x [(ra - R) •V ]B (R )

- £ T ( ro - R ) x - v )• B (R )

- £ T (r“ - R ) x [(r“ - R ) * ¿ i B (R )- (4.16)

We must now recall the restricted nature of this description: The electric and
magnetic fields change only slightly over the dimensions of the system. The first
of the three terms on the right side of (4.16) is a small correction to what is
36 CHAPTER 4. MACROSCOPIC ELECTRODYNAMICS

already present in (4.14) as J j-[(l/c )d x B (R )], and is therefore to be neglected.

Furthermore, the last term o f (4.16), which is well approximated by

£ e«(ra —R)X [(r„ - R) •V] V x E (R ), (4.17)

a

is of the same order o f magnitude as the omitted terms in the expansion (4.4),
and is therefore also to be neglected. An average of the initial form of F # ,
(4.15), with the single remaining contribution of (4.16), the second line there,
now gives

Fb = \ £ 7 (v “ - v ) ;* Kr « - R ) •V 1R (R )
a

- 5 E 7 ( r a - R ) x [ ( Va- v ) . V ] B ( R )
a

= i ^ ^ - { [ ( r a- R ) x ( v a-V )]x V }x B (R ). (4.18)

a

What has finally emerged here is the magnetic dipole moment of the system,

**- i
^ = ¿ E e<*(ro - R ) x (v“ - v )’ (4-19>
a
so (4.18) is equal to (¿i is constant in space)

F b = i ^ x V ) x B ( R ) = V [/i •B (R )], (4.20)

where we have used (1.54). (It is, of course, the similarity of this structure to
V [ d « E ( R ) ] that justifies the identification of as the magnetic analogue of
d.) We also recognize that a contribution of this form already appears in the
second term on the right side of (4.14), bearing the information that a moving
electric dipole also acts as a magnetic dipole. The comparison of the two effects,
characterized by £ d x V and \i, is that of the typical speeds of the relatively
heavy atoms, |V|, and of the light electrons, |va|, in the interior of atoms,

|V| < |va|< c. (4.21)

Accordingly, we neglect the motional effects of the atoms, and finally write
(4.14) as
F = V[d •E(R) + fi •B(R)] + ^ ( i d x B ( R ) J . (4.22)

In the absence of time variation, what remains is a force associated with the
respective potential energies o f a given electric dipole in an electric field,

-d*E (4.23)

and o f a given magnetic dipole in a magnetic field,

—/Li *B. (4.24)

4.1. FORCE ON AN ATOM 37

The energy interpretation does more than supply the force components as
negative gradients with respect to position coordinates. It also produces torques
as negative gradients with respect to angles. Take the example of a magnetic
dipole ¡i in the presence of a magnetic field B. If 0 is the angle between /¿ and
B, the magnetic potential energy is
-|¿z||B|cos0. (4.25)
The implied internal torque, that is, the torque on this individual dipole, and
not the moment of the force on the dipole, is then

— (H|B|coe0) = -H |B |8in0, (4.26)

(the reference point of this torque is at the position of the dipole), which can
be represented by a vector perpendicular to the plane formed by fi and B,
r = pxB. (4.27)
We shall now derive this vectorial result directly, along with its electric
counterpart; for simplicity, additional time derivative terms are omitted. The
torque, the moment of the force about the center of the charge distribution at
R is
r = ^ ( r a - R ) X ( e aE (r J + i eav a X B ( r j ) . (4.28)
a ' '
The part proportional to the electric field is, when we neglect the variation of
E over the system, the electric torque
te = dxE(R), (4.29)
as expected in analogy with (4.27). In deriving the magnetic torque, we first
make the unimportant change, v a —►v a — V , using (4.21), and then transfer
the time derivative to get

tb = X > a - R ) x ( i Ca( va —V ) X B ( R ) )

- £ ( v‘ - v ) x - e a(ra - R ) x B ( R )

1
f « r<* " R ) x [ ( v « - V ) X B ( R ) ] - ( va - V ) x [ ( r „ - R ) x B ( R ) ] }
2 z—' c
a
= pxB(R) (4.30)
where in the second line we have omitted the —7 JjB = V x E contribution as
negligible in comparison with r E . [See Problem 4.2 for a justification of (4.30).]
In the third line, we averaged the two preceding forms, and then used the first
identity in Problem 1.1. Putting all this together, we find the torque on the
system is given by
t = d x E + v X B, (4.31)
so that, as with the force, the result can be expressed in terms of the electric
and magnetic dipole moments, d and f,1.
38 CHAPTER 4. MACROSCOPIC ELECTRODYNAMICS

4.2 Force on a Macroscopic Body

To this point, we have considered the response of a small system, an atom, for
example, to external electric and magnetic fields, which vary smoothly over the
system. Macroscopic materials are made up of large numbers of atoms. What
is the total force on such a piece of material? We must sum up all the forces
on the individual atoms. To the extent that the forces on the atoms vary but
slightly from one atom to another, the summation can be replaced by a volume
integration, weighted by the atomic density, n(r), the number of atoms per unit
volume at the macroscopic point r:

F = J (dr) n(r) d x ( V x E ) + ( d . V ) E + a* x ( V x B) + . V)B

4 O®) (4.32)

Notice that we have rewritten (4.22) with the aid of the identities

V ( d •E) = d x ( V x E) + (d •V )E , (4.33)
V ( m - B ) = m x ( V x B) + ( / x * V) B . (4.34)

First a word about d and \i in these expressions. In the single atom formula
(4.22), the derivatives act only on E and B, which is reflected in (4.32). For
a many-atom system, the dipole moments could well vary from one location to
another and so have macroscopic spatial dependence. Accordingly, d (r) and
fjt(r) are the average dipole moments at the point r. We now define the electric
polarization, P , and the magnetization, M , by

P (r ,t) = n (r )d (r ,i), (4.35)

and
M ( r , i ) = n (r)fi(r,t), (4.36)

respectively. The resulting macroscopic form of the total force at time t is

P(<) = J ( d r ) P M M V x E M l + tPtr^.V jE tr,*)

+ M (r,i)x [V x B (r,t)] + [M (r, t ) . V ]B (r, t )

+ 4 f*P 0M )xB (r,i)) (4.37)

dt \c

(Here, the distinction between | B and J¿B has been dropped, because the
difference is of order of the small atomic velocity V , which is averaged to zero
in any case.)
4.2. FORCE ON A MACROSCOPIC BODY 39

We proceed to simplify this in various ways. First, we use one of Maxwell’s

equations to obtain

P x (V x E) + | ( i r x B ) = ( I | p ) x B , (4.38)

and then we use the identity

V ( M •B ). = M X ( V X B ) + (M •V )B + B x ( V X M ) + (B •V ) M , (4.39)

which is a generalization of (4.34). All subsequent steps involve the statement

that the integral is extended over a volume that includes the whole body, so
that, on the bounding surface of that volume, n(r) = 0. This means that in
performing partial integrations through the use of the divergence theorem, the
surface integrals vanish. In effect, then,

(P •V )E - ( V •P )E , (4.40)

and similarly, using V •B = 0, (4.39) yields

M X ('V X B ) + ( M •V ) B (V xM )xB . (4.41)

The immediate result is

p=/ (dr) [ - ( V . P ) E + i ^ P j XB + ( V X M ) X B . (4.42)

The comparison of this with the microscopic description of the force on charge
and current densities, ( 3 .8 ) for zero magnetic charge, suggests the definition of
an effective charge density, peff, and an effective current density, j eff, as

Peff(r,t) = - V . p ( r , f ) , (4.43)

je ff(M ) = ^ P ( M ) + c V x M (r ,< ). (4.44)

Notice that these effective densities satisfy the equation of charge conservation,

¿P e ff + V . j eff = 0. (4.45)

It is left to the reader to verify (Problem 4.3) that the total torque, r , on the
body, the sum over all atoms of the external torques:

^ ext = J ( dr )nr X d x ( V x E ) + (d •V ) E + f t x ( V x B ) + (/*• V )B

+ I l ( d x B) (4.46)
40 CHAPTER 4. MACROSCOPIC ELECTRODYNAMICS

and of the internal torques:

'T’int — J (dr) n ( d x E -f /x X B), (4.47)

is properly reproduced as the integrated moment of the effective force density,

„ 1 .
T = P effE + ~ J e f f X B (4.48)
C

4.3 Macroscopic Electrodynamics

Now we construct a phenomenological macroscopic electrodynamics. And what
is that? Nothing more than the form in which electrodynamics first arose, in
the pre-atomic period, when only the properties of bulk matter were involved.
But the challenge here is to derive the phenomenological theory from the mi­
croscopic Maxwell-Lorentz description. Both theories will employ concepts that
are abstracted from the kinds of measurements that are appropriate to their
level of description. The microscopic regime is characterized by rapid space-
time variations unlike the macroscopic one, which is characterized by scales
large compared to those of atoms. Laboratory instruments, being large, di­
rectly measure average quantities. Macroscopic fields are thus defined in terms
of averages over space and time intervals, V and T, large on the atomic scale
but small compared to typical macroscopic intervals. We adopt the convention
that lower-case letters, like f ( r ,t ) , represent microscopic quantities while capi­
tal letters, like F (r ,t), represent the corresponding macroscopic quantities. The
connection between the two is

^ (r, *) = j , Jtdi' y j v (dr'} + r'>1 + O = 7 (M )- (4-49)

This is a linear relation, in the sense that

T T h = ñ +h, J f = A /, (4.50)

where A is a constant. From this follows the connection between derivatives of

microscopic and macroscopic quantities, that is, that the averaged derivative of
a function is the derivative of the average:

V/(r,<) = V / ( M ) . (4.51)

The microscopic charge distribution 's composed of two parts. That which
is confined to atoms is called bound charge. When the remaining, “free,” mi­
croscopic charge distributions are appropriately averaged, we obtain the macro­
scopic densities
P — Pivee J — Jfree* (4.52)
4.3. MACROSCOPIC ELECTRODYNAMICS 41

electric magnetic charge current

field field density density

e b Pfree + Abound Jfree 4" Jbound

E B p - V .p J + ^P + cVxM

Table 4.1: Connection between microscopic and macroscopic quantities

What is the macroscopic role of the bound charge distributions? It must be

related to the effective charge and current densities given in terms of the polar­
ization and the magnetization by (4.43) and (4.44),

peff = - V . P , (4.53)

jes = ¿ P + c V x M . (4.54)

As we have seen in the preceding section, these densities are examples of macro-
scopically measured quantities, disclosed by slowly varying electric and magnetic
fields. The physical measurements necessary for the definitions of /?eff and j efr,
since they employ slowly varying fields, should correspond to the mathematical
process of averaging involved in the definitions of pbound and jb ou n d , so we have
the identifications

Abound — P efí

jb ou n d = Jeff- ( 4 . 55)

In view o f (4.45), these two forms of macroscopic charge are separately con­
served.
The correspondence between microscopic and macroscopic quantities is given
by Table 4.1: The microscopic Maxwell equations now read

^ Xb ~ “•TTT©H (jfree ”1” Jbound), V •e = 47r(/?free "I" Abound))

c ot c
1 Í)
—V Xe = —— b, v . b = 0. (4.56)
c at
These are averaged to yield the macroscopic equations,

V xB = ( j + ^ P + cV x m ) , V •E = 4ir(p — V •P),

-V xE = -¿ B , V •B = 0, (4.57)
c ox
which can be cast into the form of the microscopic equations if we define the
displacement, D ,
D = E H- 47tP, (4.58)
42 CHAPTER 4. MACROSCOPIC ELECTRODYNAMICS

and the magnetic field, H,

H = B - 4ttM (4.59)
(recall that B is properly called the magnetic induction). The final form of the
historical, macroscopic Maxwell equations is

1 3 4?r,
V xH =
„ _ _ _
D H------J,
:
V * D = 47r/>,
- - t v

c ot c
1 r)
- V x E = - — B, V *B = 0. (4.60)
c at
Note that the macroscopic charge is conserved,

V . j + J ^ = °, (4.61)

which follows from the first pair of equations. As microscopically smooth distri­
butions, the density and flux of free charge will serve to measure the macroscopic
fields E and B. That is exhibited in the expression for the force on a macroscopic
charge distribution,
F = J (dr) (p E + ^ J x b ) . (4.62)

[If bound charge is present, there is an additional contribution to the force

coming from (4.42).]
For a complete description of the system, we require further relations be­
tween D , E, P , and J, expressing how material bodies respond to electric fields.
Similar remarks hold for H, B, and M . These constitutive relations depend
on the characteristics of the particular material under consideration. Simple
classical models— which are not qualitatively misleading— will be considered in
the following two chapters.

4.4 Problems for Chapter 4

1. Find the total charge and the dipole moment of the charge density

p(r) = —d •V<S(r).

2. Justify the approximation leading to the final form of r in (4.30). In

particular, show that the total time derivative omitted in going from the
first to the second line of (4.30) leads to

where the “canonical momentum” p a is defined by

Pa = ma\a + ^-A (ra),

4.4. PROBLEMS FOR CHAPTER 4 43

where the vector potential A for a constant magnetic field B is

A = - i r xB , V x A = B.
z

3. By summing the torque on an individual charge,

r a = raX (e „E (r a) + e„ X B (ra)) ,

first, over the charges in an individual atom, and thereby obtaining ex­
pressions in terms of d a and fjtai the dipole moments of the atom, and then
over the atoms making up a macroscopic body, obtain the result that

ext 4 “ 'T’ int)

where the external and internal torques are given by (4.46) and (4.47),
respectively. Then, verify that the torque acting on a macroscopic object
in electric and magnetic fields is given in terms of peff and j eff according
to (4.48).
Chapter 5

Simple Model for

Constitutive Relations

5.1 Conductivity
We start by considering a simple model of a metal in which the current is linearly
related to the electric field. The model is to be considered as suggestive only
but it does lead to a qualitative understanding of the important phenomena of
conduction. Of course, an accurate description requires quantum mechanics.
First consider a free electric charge (an electron) moving under the influence
of an external electric field, and subject to collisions with the atoms of the
substance. The electric field accelerates the charge, and the collisions slow it
down. Our model represents the effects of the collisions by a frictional force
that is proportional— and opposed— to the velocity. The equation of motion for
the particle, having charge e and mass m, is

m ^ -v (t) = —7717V(t) + eE(f), 7 > 0, (5.1)

at
or
| v (<) = - 7 v (<) + ¿ E ( í). (5.2)
(The variation of the electric field with position is ignored here— the velocities of
interest are of very small magnitude compared with c.) The frictional constant
7 is given a physical interpretation by considering the situation for E = 0 :

^ v ( t ) = —7 v (i), v (i) = v 0e 1t\ (5.3)

any initial velocity decreases exponentially in time, due to collisions with atoms,
with I / 7 supplying the characteristic decay time. The general solution to (5.2)
is found by first rewriting it as

Jt P * v ( 0 ] = ^ E ( i ) , (5-4)

45
46 CHAPTER 5. SIMPLE MODEL FOR CONSTITUTIVE RELATIONS

and then integrating from t* = —oo (a time before any field has been applied),
to t (the time o f observation),

(5.5)
J—00

Notice that the response of the system under the action of an external electric
field is nonlocal in time. (That is, the velocity at a given time depends on the
electric field at earlier times.) The main contribution to the integral comes from
the region of time differences that are of the order of I / 7 .
The current density for a single charge is proportional to its velocity. If n /
is the (constant) density of (free) conduction electrons, then J is

(5.6)

For the particular example of a constant electric field, this reduces to

(5.7)
my

which is a statement of Ohm’s law (Georg Simon Ohm, 1787-1854), a being the
static conductivity. [This, of course, can be more directly obtained by looking
for the static solution of (5.2).] A more general situation arises when the electric
field exhibits harmonic variation (i.e., has a definite frequency),

(5.8)

or in terms of the complex amplitude E( cj),

E (t) = Re(E(u>)e-¿ü;t). (5.9)

Now, the current density, (5.6), becomes

— E (w )e-iu,t (5.10)
m 7 — iu>

Here displayed is a complex amplitude for J(t), J(u>). In terms of it, the complex
conductivity, <r(o;), is defined by

J(w) = cr(a>)E(a>) (5.11)

where, from (5.10)

(5.12)
5.2. DIELECTRIC CONSTANT 47

For qj = 0, we regain <r(0) = a, the static conductivity given in (5.7).

The conductivity is a function of ¿w, which means that complex conjugation
is equivalent to changing the sign of co:

a ( c j ) * = cr( —u;). (5.13)

This says that the real and imaginary parts of a(w),

a(oj) = — — \ f 1?-- - Re <r(u;) + 2lm<r(u;), (5.14)

are, respectively, even and odd functions of u;, as shown in Fig. 5.1. Finally, we
note that the integral of the conductivity over all frequencies is

rife*
r to * (» ) = ^ 2 fd a , , r °_ * L
7-00 rn Jo m V o 1 + ^2
_ rife*
(5.15)
m
which is called a “sum rule.” The significant feature of this sum rule is that the
right hand side is independent of the frictional force constant, so that we could
use it to determine n/ experimentally. What underlies this is the simplicity of
the response to an electric field pulse that is localized at time t = 0. Without
time to act, the frictional forces are effectively absent. (See Problem 5.1.)

5.2 Dielectric Constant

We now modify the above model in order to discuss bound charge by including
an additional binding force term in (5.1). We will take as the simplest model
of such binding a harmonic oscillator force, which turns out, for the most part,
to give qualitatively correct results. That is, we will adopt, taking the origin to
be the center of the force,
u o _ ar _v
m — v = —uiloqT — m yv -f eE, v = —, (5.16)
dt at
as the new equation of motion. Here u>o is the natural (angular) frequency of
the electron bound in the atom, while 7 is a damping constant, primarily due
to electromagnetic radiation. (More about this in Chapter 35.)
For a harmonic time dependence of the driving electric field, (5.9), the above
force equation becomes

w 1 + u «r + 7 J tr = ¿ Re (E (w)e_ib,t) • (5 1 7 )
This implies that the steady-state solution for the position vector will also ex­
hibit harmonic time variation, that is,

E (u )e~ iut '

r (0 = m (5.18)
—w2 + Uq — iyu> J
48 CHAPTER 5. SIMPLE MODEL FOR CONSTITUTIVE RELATIONS

0
co

Figure 5.1: The real and imaginary parts of the conductivity, as given in (5.14).
5.2. DIELECTRIC CONSTANT 49

Under the usual circumstance of 7 <C the amplitude of the induced oscilla­
tion becomes very large for u> = u>o, the condition of resonance.
It is now immediate to calculate the polarization (4.35) in terms of the
induced electric dipole moment and the density of bound electrons, n&,

P = n^er, ( 5 .1 9 )

or, explicitly in terms of the electric field,

nbel
p (t) = Re
m —u>2 + u)q — iyu
= R e [ x e ( w ) E ( w ) e _ u *'t ], (5.20)

where Xe is the (frequency-dependent) electric susceptibility,

nbe¿ 1
Xe(w) = (5.21)
m —u)2 — iu>7 + coq ’

which satisfies
Xe(w ) = X e ( - w ) * . (5.22)
The static susceptibility, the real value when u> = 0, is
z>2
Xe = > 0. (5.23)

What is the order of magnitude of ;\;e? Since u o is identified with a characteristic

atomic frequency,
w0 ~ j ( 5 .2 4 )

where v and / are representative atomic speeds and distances, respectively, we

can estimate Xe in terms of atomic quantities as

nbe2
(5.25)
mu)2

Since a typical value of the electron’s kinetic energy is of the same order of
magnitude as a typical Coulomb potential energy (this is the virial theorem—
see Chapter 8 ),
p2
mv 2 ~ —e , (5.26)

the value of the electric susceptibility is of order

nbe2 3
0 ~ ribl , (5.27)

the number of bound electrons in an atomic volume, For dense matter, where
the atoms are tightly packed, we may have, therefore,

Xe ~ 1. (5.28)
50 CHAPTER 5. SIMPLE MODEL FOR CONSTITUTIVE RELATIONS

Figure 5.2: Singularities o f the integrand in (5.33) in the complex ui plane.

The displacement field, (4.58), is

D (t) = E(<) + 4irP(t)

= R e[(l + 4irx.(«))E(w)e-,,-t] , (5.29)

which defines the frequency-dependent dielectric constant, or the permittivity,

e(a>), through the relation

e(u>) = 1 -b 47rxe(^) = c (-w )* . (5.30)

Therefore, there is a linear relation between the field amplitudes,

D(w) = c(w)E(w). (5.31)

In particular, the real, static dielectric constant e is

e = l + 47rxc>l, (5.32)

where, as noted, the excess over unity can be significant in dense substances.
For example, the static dielectric constant for mica is about 6 , and for water,
about 80.
We can also derive a sum rule for x c(u>), by considering the integral

rue- —VjJ
du(-iu)Xe(w) = - du>- "Ó2 = Wo - Q t ) •
r ■ f
(u + h ) 2 ~ u f
(5.33)
For the following discussion, we assume that y < 2u>o, so that u>02 > 0. (This
restriction is actually unnecessary— see Problem 5.2.) Now think of a; as a
complex variable, and note that the integrand has singularities at

w = - | 7 ± <*>(,, (5.34)

which lie in the lower half plane (see Fig. 5.2). Accordingly, the integrand is
everywhere regular in the upper half plane, and the integral over a path that is
closed by a large semicircle in the upper half plane equals zero. The integral of
5.3. PLASMA 51

interest, (5.33), is therefore the negative of that over the semicircle of arbitrarily
large radius, which is
n¿e2 f , —i e2
------ / duj-— = TTUb— . (5.35)
m Jn u m
Thus, the desired sum rule, independent of 7 and u)0, is

f du) (-i v ) X e ( v ) = 7T
m
(5.36)

In retrospect, the same method could have been applied to the conductivity
integral, (5.15), using (5.12), which integral is obtained from that in (5.33) by
putting (júq = 0. Physically, we should have expected this result on the basis of
the a- relation, (5.15), since the two quantities e and a are just parts of a whole.
The two phenomena being discussed are just the free electron and bound electron
contributions to the total current, which have the following form for a definite
frequency u>,

J + ^ -P -► Re ([<r(w) - iu>Xe(u)]E(w)e *wt) , (5.37)

so that the sum rule corresponding to this total current is proportional to the
total electron density rib + n / ,

e2
/ du[a(u) - *u>Xe(w)] = (nt + nb)it— , (5.38)
J- o o m

which expresses the fact that, in the response to an electric pulse localized at
time zero, only the inertia of the electrons matter— frictional and binding forces
have no time in which to act.

5.3 Plasma
Let us combine the results of the preceding two sections by considering the mo­
tion of free charge in a conducting dielectric material, for which the conduction
current is
J = crE = j D . (5.39)

First suppose that both a and e are taken to be independent of frequency,

an approximation which is valid for low frequencies. Then the local charge
conservation equation,

J^> + V . j = 0, (5.40)

becomes
d <T
52 CHAPTER 5. SIMPLE MODEL FOR CONSTITUTIVE RELATIONS

In the interior of a homogeneous substance, the use of (4.60), V * D = 47rp,

produces the differential equation

d or
dtP + - 4vp = 0. (5.42)

The solution to this equation, corresponding to an initial charge density p(r, 0),
is
p(v,t) = p ( r , 0 ) e - ^ > (5.43)

implying that the charge disappears from the interior of the conducting body
at a rate measured by
47T0"
(5.44)
e
The charge in the interior of the volume eventually all migrates to the surface.
(See Problem 5.3.)
The use of the static conductivity here assumes that this decay rate, 7 ', is
small compared to the frictional constant 7 , 7 ' <C 7 , or, using (5.7), that

47r n fe 2
< 7 (5.45)
e m

This must be satisfied by some combination of small density of free charge and
a high coefficient of friction. But what of the opposite situation, where there
is a high density of free charge and little friction, as encountered in a plasma?
For that, it may be clearer to return to the equation of motion (5.2), presented
as the differential equation

^ í ! e = !V£!d (5.46)
m me

for the current

J = n /ev . (5.47)

Now we can use the charge conservation equation, in the form

G H ¿ , + v . ( | + 7) a = o, (5.48)

to get, when we again use V ♦D = 47rp,

( d2 d 2\
(5.49)
W + ', M + V P = 0-

where the plasma frequency is defined by

2 47rn/e2
wp = (5.50)
me
5.4. POLAR MOLECULES 53

(Note that the usual form of the plasma frequency has e = 1.) In general,
the solutions to (5.49) are damped oscillations. If the rate of change is small
compared to the scale set by 7 , the equation (5.49) simplifies to

(;§+7,)p=0, 7>~ ~ j < 7 , (5-51)

which will be recognized as the previous result (5.42), including the restriction
(5.45). This limit corresponds to exponential decay. But in the other limit,
where change occurs rapidly relative to 7 , it is the 7 term that can be approxi­
mately neglected,

<5-52)

and the charge oscillates at the angular frequency u)p.

The appearance of e in this plasma frequency is a reminder of a restriction
still in force, that u>p be small compared to u>0, the characteristic atomic fre­
quency, near which the frequency dependence of e(o;) can no longer be ignored.
There is an extreme plasma circumstance in which ojq is not relevant. Let the
physical conditions be such that all atoms are completely ionized, removing the
distinction between free and bound charge. Then the entire charge density can
be viewed as the result of polarization,

/> = - v . p , (5.53)

arising from the displacement r of the electrons relative to the oppositely charged
heavy ions,
P = ne r. (5.54)

That displacement changes in time, responding to the electric field as

d2r
(5.55)
m ~dt? ~ eE’
or
d2 ne“
^ 2 P = -----E
(5.56)
at1 m '
From this, and the field equation

V •E = 4 ir(—V •P ), (5.57)

we derive (5.52), describing the plasma oscillations, with e = 1.

5.4 Polar Molecules

In the above model for the electric susceptibility, leading to (5.21), the dipole
moments were induced by the applied electric field. But what about permanent
electric dipole moments? Do individual atoms possess such static properties?
54 CHAPTER 5. SIMPLE MODEL FOR CONSTITUTIVE RELATIONS

With the exception of atomic hydrogen where the orbital motion respects a
preferred direction in space (as in the classical elliptical orbits), atomic electric
dipole moments change direction in space so rapidly in response to the fast
electronic motion that no average effect survives. But things are different with
molecules, specifically those of a polar nature. In the example of H+ C1“ , the
hydrogenic electron is transferred to form the chlorine ion, and a dipole moment
is associated with the relative motion of the heavy ions. Other examples of polar
molecules associated with familiar substances are H2 O, SO 2 , NH3 , and CH 3 CI.
For such molecules, in isolation, it is not misleading to think of a permanent
electric dipole moment that changes its spatial orientation only in response to
the slow rotation o f the molecule.
But molecules are not ordinarily isolated; they exist in an environment in
which other molecules collide with them at a rate determined by the temperature
of the substance. The effect of these collisions is to remove any particular spatial
orientation o f the dipole moments; it still requires an electric field to provide
a preferred direction. But now there is a competition between the organizing
effect of the electric field, with its preference for lower values of the energy,

£ = - d •E = —|d||E| cos 0, (5.58)

and the disorganizing effect o f the ambient temperature T. For a static field,
the net balance of that competition is expressed by the Boltzmann factor, which
gives the probability of finding a configuration of energy E ,

e ~ E /kT ^ (5.59)

where Ar, the constant of Ludwig Boltzmann (1844-1906) has the value

k = 1.381 x 10“ 16erg/K. (5.60)

In thermal equilibrium, the fraction of dipole moments that are directed within
the solid angle
dSl = sin 0 d0 d<j) (5.61)

is proportional to the product of dQ with the Boltzmann factor:

l^ d -E /fcT (5.62)
Z 4ir

where the choice of the normalization constant, the so-called “partition func­
tion”
Z = f — eA ' ^ kT (5.63)
J 4tt
ensures that the totality of such fractions equals unity. Consequently, the aver­
age dipole moment is
5.4. POLAR MOLECULES 55

Except for very low temperatures or very high fields, the condition |d| |E|
kT holds true, so that we may expand the exponential Boltzmann factor in the
partition function Z:

In this uniformly weighted average over all directions, where d and —d appear
with equal weight, the average of d is zero, and that of the square of some
component, say d2 z , is the same as any other component, so that

(4) = ( 4 ) = ( 4 ) = |(d)2. (5.66)

The leading contribution to (5.64) is produced by

d 1 d2 p
(5.67)
0E 3 (kT )2 ’

so that in this approximation (d )r and E are linearly related,

(d )T « (5-68)

What kind of limit does our weak field condition impose on the magnitude of
the electric field? If we recall that room temperature, T = 300K, corresponds
to kT « eV, and typical atomic dimensions set the scale for the dipole
moment, |d| 1 0 ” 8 e cm (where e represents the magnitude of the charge on
the electron), then |d| |E| C kT requires

|E| <C 3 x 106 volts/cm, (5.69)

which is indeed a large electric field. (For the average dipole moment when
this approximation is not valid, see Problem 5.4.) We note that one of the
largest observed moments, that of potassium chloride, (foci = 2.14 x 10~8 ecm,
is slightly more than twice the value that was used in the estimate of (5.69).
With a density of nmoi polar molecules per unit volume, the (weak field)
contribution to the polarization is

(5.70)
P - n m o l 3fcTE ’
and the complete static susceptibility becomes

d2
Xe — Xe,atom “h nmo\ (5.71)

where Xe,atom is the atomic susceptibility due to the induced dipole moments
of the atoms [see (5.23)]. This result is accurate for a “polar gas,” where the
densities are low. Note that the first term in (5.71) is independent of T, while
56 CHAPTER 5. SIMPLE MODEL FOR CONSTITUTIVE RELATIONS

the second is proportional to 1 /T , allowing us to separate the two contributions

experimentally, and thereby measure the dipole moment of a polar molecule.
The expression (5.71) is known as the Langevin-Debye equation, which has
been of great importance in interpreting molecular structure.
The Boltzmann factor describes the situation of a static electric field. But
suppose the electric field is vibrating, or rotating, with a definite frequency.
Owing to the large inertia of the molecules, it takes a significant time for the
dipole moment to readjust or relax into the configuration demanded by the new
direction of the electric field. With increasing frequency the ability to readjust
decreases and eventually the dipole moment ceases to follow the variations o f the
electric field. Experiments indicate that this occurs already for radio frequencies;
at infrared and visible light frequencies, the dipole moments are effectively inert.
A simple quantitative version of this picture combines two ideas. First, in
the absence of the electric field, any net dipole moment relaxes to zero with a
characteristic decay rate, or inverse relaxation time, 7 , and in the presence of
a static field, the average moment is given by (5.68). These are united in the
differential equation

(5.72)

The steady-state solution, when

E(t) = Re (E {u )e^ ) , (5.73)

is directly verified to be

' d2 7
(d )(t) = Re (5.74)
S kTy — iu
m

The implication for the static situation, u = 0, and for the high-frequency
limit, u) 7 , are as already described. In particular, (5.74) expresses the fact
that as uj increases, (d) decreases eventually like y/u. In general, the static
susceptibility contribution of the polar molecules is multiplied by

1
(5.75)
1 — iiOT ’

where
r = - (5.76)
7
is the relaxation time of the electric dipole moments. The molecular effect
disappears at relatively low frequencies, leaving only the atomic contribution,
since the atomic frequency, u>0, is much greater than the molecular frequency,
7-
5.5. CL A USIUS-MOSSOTTI EQUATION 57

5.5 Clausius-Mossotti Equation

One further point needs discussion. In our atomic model for the dielectric
constant in Section 5.2, we used the following equation for the electronic motion,
(5.16):
d2 2 d
m — r = —mw0r — m y — r + eE. (5.77)

Here the driving force has been taken to be eE where the macroscopic field is
E( r, f ) = e (r,¿). The same assumption for the driving field was made for the
alignment of polar molecules in Section 5.4. This is incorrect since e includes the
field of the atom (or molecule) itself, the effects of which are already represented
in the harmonic restoring force. The correct driving field is the field acting on the
electron due to all the other atoms (we use the word atom to stand, generically,
for either atom or molecule),

^driving = ^driving — e e atom = E ^atom j ( 5 .7 8 )

where now the overbar represents a spatial average over a volume V which
contains exactly one atom, that average volume per atom being the inverse of
the density,

n=y- (5-79)
[In (5.78) we assume that no significant contribution is produced by neighboring
atoms, so that Edriving does not differ appreciably from edriving? and also that
the average over the atom is already sufficiently representative of a macroscopic
average that the field E can be introduced. What we are doing should not be
expected to apply for a strongly polar liquid or solid where the forces produced
by neighboring molecules could be the dominant effect.] The field due to the
atom in which the electron is located is

e atom (5.80)
I*- r a l ’
where the summation extends over all charges in the atom. We can calculate
eatom by averaging over a sphere (of radius a and volume V = 47ra3 / 3 ) which
is large enough to include the atom; the negative of the average field can be
written as
(5.81)
-* = -b
An individual term here,

ea(ra) = - V a Í (dr) r ~ V >, (5.82)

Jv lr a — r l
is the electric field, at the point r a within the sphere, arising from a uniform
charge distribution within the sphere of density ea/V. As such, it obeys the
differential equation (1.26)
. Cfl __ 4tt ea . v
V a . e 0 = 47T— = V a • — — (ra - R), (5.83)
58 CHAPTER 5. SIMPLE MODEL FOR CONSTITUTIVE RELATIONS

where R is the position vector of the center o f the sphere. The last form merely
uses (with subscript a) the fact that

V •(r — R ) = V •r = 3. (5.84)

And now the spherical symmetry of the situation, telling us that the electric
field at a point must be directed along the line from the center, immediately
yields (see also Problem 5.5)

4?r
e a(ra) = -g-nea(r 0 - R ), (5.85)

and then
_____ 47t 4tt _ 47
e atom = — - R ) = — nd = — P . (5.86)

We conclude that the correct driving field (subject to the caveats mentioned) is

47T
^driving =: E H P• (5.87)

To appreciate what effect this has on our earlier results, let use denote by
what was previously, and incorrectly, called the susceptibility. For static fields,
then,
d2
xé = nb + nmol (5.88)
rrnuÁ 3k T ’
and now we have
(5.89)

or
(5.90)

Prom this form, or the equivalent versions

Xe , _ Xe
(5.91)
1 m Ú 2 .v ^ e’ ^e 1 _ i2_ / >
1 t" 3 1 3

it is clear that the earlier identification of Xe with x*e ls valid only when x e < l ,
as in substances of low density. Yet another way of presenting matters, one that
introduces the dielectric constant according to (5.32)

4irxe = e - 1 , (5.92)

namely
e — 1 _ 47T ,
(5.93)
7+2 - T Xe’
is known, from its historical origins, as the Clausius-Mossotti formula [O. F.
Mossotti (1850), R. Clausius (1879); the years cited indicate the dates of signif­
icant publication].
5.6. PROBLEMS FOR CHAPTER 5 59

When one turns to fields of definite frequency, the names change, the relation

e(w) - 1 47T ,
(5.94)
Í H ! = T * '* " *

being called that of Lorenz and Lorentz [L. Lorenz (1869) and H. A. Lorentz
(1880)]. For the particular example of a nonpolar substance, where [(5.21)]

n&e2/m
X » (5.95)
u;$ —ÜJ2 — iujy ’

the formula (5.91),

^ ^ 47rx'e(a;)
(5.96)

immediately gives
nje2/m
X e (w ) = (5.97)
WÍ?1 —
—a; — «20^7
>

with the shifted resonant frequency, o>i, being given by

47Tn&e2
Wi = wñ - = <^o> (5.98)
m 6+2

where e is the static dielectric constant given by 1 + 47TXe(0). Clearly the sum
rule (5.36) remains unchanged since it is independent of uo.
We conclude with an experimental example of a situation in which the modi­
fication of the driving field is both significant and accurate. The static dielectric
constant of nonpolar nitrogen gas has been measured at low and high densities.
At the density 0.06604 g /cm 3, the observed value of e — 1 is 0.03109. If e — 1
were equal to 47r%g, which is proportional to the density, the value predicted for
a density of 0.5780 g /cm 3— slightly more than half the density of water— would
be e — 1 = 0.2721. The measured value is 9% higher: 0.29633. If, however, one
uses the Clausius-Mossotti relation (5.93), it is (e — l ) / ( e + 2) that is propor­
tional to the density. Now the value of e — 1 predicted for the larger density is
0.2959, which falls short of the measurement by only 0.1%.

5.6 Problems for Chapter 5

1. Use ( 5 . 2 ) to find the related equation for the conduction current J =
n /ev. Solve this equation for E (t) = Eo^(t) if 3(t < 0) = 0 . What is J
immediately after t = 0? Connect this with ( 5 .1 5 ) .

2. Evaluate the integral in (5.33) in the case that 7 > 2u>o, and, thereby,
establish the sum rule (5.36) in general.

3. Show that the charge that disappears from the interior of a conductor
according to (5.43) appears on the surface of that conductor.
60 CHAPTER 5. SIMPLE MODEL FOR CONSTITUTIVE RELATIONS

4. Evaluate Z in (5.63) without the weak field approximation. Then deduce

(d )y . What happens in the limit |d||E| >► kT l

5. Prove that the solution of (5.83) is (5.85), directly by the application of

Gauss’ law.

6. Another way to evaluate (5.81) is to compute the integral

X (<fr)
k —r«|
= 47T
1
2a
2 1 2
6 r*

where the volume integral is over a sphere of radius a which contains the
point r a, and the origin is taken to be the center of the sphere. [Hint:
Take r a to lie in the z direction.] Upon taking the gradient in (5.81), the
result in (5.86) follows.

7. Verify the numerical improvement stated in the paragraph at the end of

Section 5.5.

8. On the basis of the formula derived for the dielectric constant, (5.32)
and (5.23), estimate as accurately as you can, based on simple physical
arguments, the value of e for air and water, in the regime u>moi <
^atom- What about the regime probed by capacitive measurements, cv <C
u>moi, where, for water, e « 80? How closely do your estimates agree with
the observed values? Discuss possible sources of error in your estimate.
How are your results changed if the Clausius-Mossotti equation (5.93) is
used instead?

9. A way of determining the sign of the charge carriers in a conductor is by

means of the Hall effect. A magnetic field B is applied perpendicular to the
direction of current flow in a conductor, and as a consequence a transverse
voltage drop appears across the conductor. If d is the transverse length
of the conductor, and v is the average drift speed of the charge carriers,
show that the voltage, in magnitude, is
v
V = - dB.
c

What is the sense of this potential drop?

10 . Consider the Hall effect discussed in the previous problem. If the mag­
netic field (perpendicular to the current flow) is in the 2 direction, the
diplacement current is neglected, and the resistivity l /<7 is very small,
show that there exist waves of the form exp[i(kz — ut)] in E and J, with
the dispersion relation
Be
U) = k2
2wne
These are helicon waves, which provide a means of measuring 1/ne.
5.6. PROBLEMS FOR CHAPTER 5 61

11. (a) Show that Ohm’s law for a neutral conducting fluid moving with
velocity v is
J = <r ( e + -^ x b ) ,

and thereby derive from Maxwell’s equations

dB „ . c2
—— = V x ( v x B ) + -— V 2B.
at 7 47TCT
(Argue that the displacement current, ignored here, gives only v2/c2
corrections.)
(b) For a fluid at rest this means that B satisfies the diffusion equation

dB _ c 2
V 2B.
dt Aircr
If B varies over a characteristic distance L, what is the characteristic
time r for the decay of the field? Estimate r for the earth’s core,
where L ~ 106 m, <7 ~ 1017 esu.
(c) For times short compared to the diffusion time r, B satisfies

dB „ ,
= Vx(vxB ).

Show that this means that the magnetic flux through any closed loop
moving with the local fluid velocity is constant in time.

12. Show that a compressible, nonviscous, perfectly conducting fluid in a mag­

netic field is governed by

w + V .(pv) = 0

(conservation of fluid),

d 1
P q,-V + p(v •V ) v = - V p - — B X (V X B )

(Newton’s law, where p is the pressure), and

Í = V X (.X B )

(conservation of magnetic flux).

Linearize these equations by substituting

B = B 0 + B i( r, t ) ,
P = po + p i ( r , t ) ,

v = v^r.f),
62 CHAPTER 5. SIMPLE MODEL FOR CONSTITUTIVE RELATIONS

where quantities with 1 subscripts are supposed to be small. Show that

Vi satisfies

d2
O P Vi - s 2V ( V •V i ) + VA X ( V X [ V X ( v i X v a )]) = 0,

where s is the sound speed,

s2

and va is the Alfvén velocity,

Bo
"VA ri
y jw p

If v i ( , r , t ) represents a plane wave propagating parallel to Bo,

Vl(r,i) = vic<k k||B0)

show that there are two modes:

(a) A longitudinal wave, v i ||k, with phase velocity s. (This is an ordi­

nary sound wave).
(b) A transverse wave, Vi _L k, with phase velocity v a . (This is the
Alfvén wave.)

What are the modes and wave velocities if k ± Bo?

13. Show that for a collisionless plasma, for which the charge density satisfies

the electric field satisfies the same equation provided B = 0 . (W hy is

this consistent?) Derive the linearized equation then satisfied by the fluid
velocity v(r,t).
Chapter 6

Magnetic Properties of
M atter

6.1 Canonical Equations of Motion in Electro­

magnetic Fields
Since magnetic effects are more subtle than electric ones, it is helpful to first
develop the formalism describing a charge moving in the presence of electric and
magnetic fields. We start with the equation of motion for such a charge,

( 6 . 1)

Although, unlike an electric field, a magnetic field does no work on a charge,

there is a magnetic term in the energy, because the act of turning on a magnetic
field produces an electric field, according to Faraday’s law, (1.64). To see this, it
is convenient to recast ( 6 .1 ) so that potentials appear instead of field strengths.
Since the magnetic field has zero divergence,

V - B = 0, ( 6. 2)

we may introduce a vector potential, A , such that

B = V x A. (6.3)

When we substitute (6.3) into Faraday’s law

(6.4)

we observe that the quantity E + ^ J~A is irrotational,

(6.5)

63
64 CHAPTER 6. MAGNETIC PROPERTIES OF MATTER

which means that a corresponding scalar potential (j) exists,

E + ~ A = -V<j>. ( 6 .6 )
c ot

This is the generalization of (1.13) to incorporate time dependence, In terms of

the potentials A and <j>) the equation of motion (6.1) reads

d _ , e d . e . Ax
m —-v = -e V < ^ -----— A H— v X (V X A ), (6.7)
at cot c

which can be rewritten as

d cd t-t ( i ^
( 6 .8 )
ms y = - . ; s A _ e V AJ'

since ( V operates only on A ),

v X (V X A ) = V ( v •A ) - (v •V ) A , (6.9)

and Q I

Qj A (r, t) + (v •V ) A ( r , t) = — A (r(t), t). ( 6 . 10 )

This suggests that we define the canonical momentum of the particle by

p = mv + - A , (6.11)
c

which supplements the kinetic momentum mv by the potential momentum f A .

In terms of p the equation of motion reads

ip = -v (* -iv .A ). (6.12)

Note that if the potentials are spatially constant, p is conserved.

Next, let us derive the corresponding equation for the energy of the system.
Taking the scalar product of v with (6.1), we obtain

jt { l ™ 2) = eE•'v = - eé + + 4 (* - r A) ’ (613)
where we have used ( 6 .6 ), as well as (6.10) for <j> instead of A . [In (6.13), v is
an implicit function of t, so d/dt does not act on it.] Rewriting this equation as

jt ( l ™ 2+e4) = it (e^ " r •A ) ’ (614)

we see that the energy, E ,

E — —mv 2 + e(¡), (6.15)
6.2. DIAMAGNETISM 65

is conserved if there is no explicit time dependence in the potentials. Since we

have shifted our emphasis from v to p, it is convenient to rewrite E in terms of
this new variable,

We now summarize the system of equations we have derived,

_ d _ 1 / e \
—- ( p -----A ) , (6.17)
~ dt* mV c )
d / e
—V 1e<p-----v •A (6.18)
dtP ~
±E = (6.19)
dt ¿ ( e<>“ í v , A )
Observe that this set has the form of the equations of motion in the Hamil­
tonian formulation of mechanics. If the Hamiltonian function is H {q)p^ t), the
Hamiltonian equations of motion are

é a = z H- <6'20)
where, in the case we are considering here, the variables are identified as

q -> r , p - + p, ( 6 .2 1 )

while the Hamiltonian is identified with (6.16).

6.2 Diamagnetism
We now apply this formalism to a simple mutually interacting system (an atom)
immersed in a homogeneous magnetic field, B, for which the vector potential
can be taken to be
A = ^BXr. (6.22)

The energy of this system, as a simple generalization of (6.16), is

(6.23)

where the last term, U ) represents the potential energy of atomic forces that
keep the atom together. The energy can be rewritten as

( B x r 0f , (6.24)

where the first term is the ordinary expression for the energy of the atom. The
second term in (6.24) involves the intrinsic magnetic moment of the atom, /x0,
defined by
= X ^ 9 ^ r _ra><Po’ (6-25)
66 CHAPTER 6. MAGNETIC PROPERTIES OF MATTER

when r„ is the position of the ath particle relative to the center of mass of the
atom. This is very similar to the definition of the magnetic moment given in
(4.19),

v = Yl^ñ;raXVa’ (6-26)
but is not the same since p ^ mv. To see that m is the actual magnetic moment,
which is required to satisfy [see (4.24)]

dE
(6.27)
dB ~
we note that the change in the energy, 6&E, caused by a change in the magnetic
field, <SB, is

5b E = ^ Pa ~ i .
? Xro .lirax5B
ma 2c

= - ^C^ra><Va^*¿B = (6.28)

The actual field-dependent magnetic moment can then be read off from (6.28)
to be given by (6.26) or explicitly by

¡■ra x ( B X r a), (6.29)

4 mac:

which can be derived from (6.26) by use of ( 6 . 1 1 ), the second term being an
induced effect. [Note that this argument provides a demonstration that the
intrinsic magnetic moment Mo is independent of the magnetic field B.]
We first consider an atom which has zero intrinsic magnetic moment,

Mo = 0 . (6.30)

This will be true for a spherically symmetric system or, since for a single particle

Mo ~ r X p — l — angular momentum, (6.31)

for a system with zero total angular momentum as long as all the particles
involved have the same e/m ratio (which is very nearly true for an atom). For
such a spherically symmetric system, which has no preferred direction, we see
that
(rr) = - r 2 l , (6.32)

and hence
( r X ( B x r ) ) = ([r2B — (r *B)r]) = - r 2B, (6.33)
o
so that the time average of m is

■ r2B (6.34)
**— E 6 m aC2 a
6.3. PARAMAGNETISM 67

Note that the induced magnetic moment is directed oppositely to the magnetic
field. If n is the density of such atoms, the magnetization (4.36) is

where the last step is justified since —Xm <C 1. The relative sizes of Xm and Xe>
from (5.23), are estimated by the ratio

(6.39)

if we use typical atomic dimensions r and speeds v. The effect we have discussed
here is called diamagnetism, which is universally exhibited by all matter.
That the induced magnetization in atoms is opposite in sign to the inducing
field is a microscopic example of what is called Lenz’s law [Heinrich Lenz (1804-
1865)]. Bodies for which this effect is the dominant one are called diamagnetic.
They are repelled by regions of strong magnetic field, where the energy is in­
creased. (See Chapter 26.) The characterization of diamagnetism, /i < 1 , is seen
to be analogous to the usual situation for dielectrics, e > 1 , since the parallel
relationship between magnetic and electric quantities is H = ^-B and D = eE.
We close this section by remarking that the simple formula (6.38) suggests
that the ratio of susceptibility to density for a given substance is independent of
temperature. This is almost universally valid experimentally. In the example of
water, where the mass density is practically constant, the measured susceptibil­
ity is close to —0.7 x 10"6. The major exception is the metal— with the largest
diamagnetic susceptibility— bismuth. Its susceptibility at room temperature is
about —1.3 x 10“ 6, but decreases significantly with rising temperature. Here
the quantum insights of the modern theory of metals are indispensable.

6.3 Paramagnetism
What happens if there is a permanent intrinsic moment,

Mo # 0? (6.40)
68 CHAPTER 6. MAGNETIC PROPERTIES OF M ATTER

This situation is analogous to that of permanent electric dipole moments, dis­

cussed in Section 5.4. Due to thermal motion, the average magnetization will be
zero if there is no external field. In the presence of a magnetic field, we obtain
the thermal averaged magnetic moment from (5.64) by replacing d by and
E by B. The obvious high temperature limit, from (5.68), is

M r = (6.41)

corresponding to the magnetization

m = J£| b . (6.42)
3 kT
This is appropriate to the weak field circumstance

fi0B < kT. (6.43)

Inasmuch as the typical magnitudes of ¡io/d are of order v/c ~ 1 0 ~ 2- 1 0 - 3 , the

upper limit to JB, at room temperature, for (6.42) to be valid is in the range
of millions of gauss, or hundreds of Teslas. Note that unlike in diamagnetism,
the magnetization here is parallel to the magnetic field. The permeability is [cf.
(6.38)]
// = Xm — n (6.44)
1_A'itt) ' + 4’ n w f > i - 3kT '
1 ^ n 3kT
since, again, the magnetization is small. Substances with positive magnetic sus­
ceptibilities are called paramagnetic. For this class of materials, the permeability
is greater than one. The simple models indicate that the ratio of paramagnetic
to diamagnetic susceptibilities is of the order

Xra,para mv
100 at room temperature, (6.45)
Xm,dia Hr

where mv2 is related to the magnitude of energies in the atom. The estimate in
(6.45) is in general agreement with the observation that paramagnetic gaseous
oxygen at standard pressure and room temperature has a positive susceptibility
about one fifth the susceptibility of water, although the molecular density of the
oxygen is less than a thousandth of that of water. The susceptibilities of para­
magnetic substances are still so small compared with unity (for liquid oxygen,
Xm = 3 x 10~4) that the approximation of neglecting the distinction between
B and H in (6.44) is well justified. The inverse dependence on temperature
displayed there was discovered experimentally by Pierre Curie (1859-1906).
Again, we have persisted in an error. In the electric case (see Section 5.5),
the correct driving electric field, Edriving was obtained by removing the field
of the atom itself. Exactly the same arguments apply here. The effect is one
of no practical importance but is conceptually significant. Analogously to the
previous arguments, we define the driving magnetic field by

B driving — B batom* (6.46)

6 .4 . F E R R O M A G N E T IS M 69

Recalling from Section 1.2 that the magnetic field, which is present in a coordi­
nate system moving relative to a static electric field, is

B = ~ X E , ( 6 .4 7 )

we can approximate the atomic magnetic field by

t>atom := ^ ] ~~~~X e a, ( 6 .4 8 )
a

where now e a is an individual term in the second sum in (5.80). The average
value of this field is, according to (5.85) [e^ = —ea(ra) there],
47T ,
" 3
-'atom
= E t * - = x: t * V
4ir 1 r-A ea
<±7T 87r 87r ,
= ------- > — r aX v a = — n\i — — M, (6.49)
T V ^ c T T
where we recall (6.26). Hence, the driving magnetic field is

« 871V „ 47T
Bdiriving (6.50)

which is analogous to the electric case, (5.87). Since the magnetic susceptibility
is defined by
M = XmH, (6.51)
the correct driving field is negligibly different from B ,

/ 4 tt
* driving
=( 1 +T Xm) H « B , ( 6 .5 2 )

because %m < 1 .

6.4 Ferromagnetism
The history o f magnetism did not begin with the phenomena of paramagnetism
and diamagnetism, which were first recognized by Faraday in 1845. The ancients
were familiar with the remarkable properties of Magnesian stone, the iron oxide
Fe3 Ü4 . The term ferromagnetism refers to the property of such substances,
primarily members of the iron group, of exhibiting permanent magnetization.
A simple model of this effect was introduced by Pierre Weiss (1865-1940), who
effectively postulated that the driving magnetic field within ferromagnets is not
(6.50), but rather
B driving = H + A M , ( 6 .5 3 )

where A 1 . In terms of Bdriving we wish to calculate the thermal average of

the intrinsic magnetic moment, ( ^ o)t - Rather than use a classical distribution
70 CHAPTER 6. MAGNETIC PROPERTIES OF MATTER

(but see Problem 6.3), it is simpler and more accurate quantum mechanically
to suppose that the atomic magnetic moment is either lined up parallel or
anti-parallel to Bdriving, which defines the 2 axis. Since the interaction energies,
for the two possibilities, are

A^O *® driving — T/^O-^driving > (6.54)

the Boltzmann weighting of states yields

(6.55)

* = + AM). (6.56)

The resulting magnetization has magnitude

M = nfio tanh t ^ ( H + AM). (6.57)

Ki
The possible existence of a magnetization in the absence of the field H is
implied by the equation

(6.58)

in which
Tc ^ A . (6.59)
k
In Fig. 6.1 there is plotted the left side of this equation, and also the right side
with examples of the two situations T > Tc and T < T C. For T > T C the curves
corresponding to the two sides of (6.58) intersect only at M = 0; there is no
magnetization. But, for T < Tc, there is also an intersection at a positive value
of M/rtfio that is less than unity; a permanent magnetization can exist. The
critical temperature Tc above which no permanent magnetization is possible is
called the Curie temperature. Its value is of the order Tc ~ 1000 K. If T <C Tc
all the magnetic moments are lined up,

M = n/i0, (6.60)

which corresponds to a permanent magnet.

Above the Curie temperature, an external field is required to produce mag­
netization. Suppose we are sufficiently above the Curie temperature so we are
allowed to take x in (6.56) as a small quantity. Then (6.57) reduces to

implying the magnetic susceptibility

 6.4. FERROMAGNETISM 71
tanh[(M/n\i,)(T/r)]

M/n]^

Figure 6.1: Solutions to (6.58) for temperatures above and below the Curie
temperature. The solid curves are the plots of the right side of (6.58) for T < T C
and T > Tc, respectively, while the dashed line is the left side of that equation.
72 CHAPTER 6. MAGNETIC PROPERTIES OF MATTER

which is valid providing

T — Tc fi0H
(6.63)
Tc * kTc *
The result (6.62), which is called the Curie-Weiss law, exhibits the characteristic
paramagnetic behavior seen in (6.44).
Finally, we estimate the phenomenological parameter A, by inserting fio ~
er(v/c) into (6.59):

kTc ~ A ( ~ ) 2 ~ ( nr3)- (6.64)

In a solid at ordinary densities, the product of n (n ~ 1023 cm” 3) with r 3

( 1 0 ""8 cm)3, the fraction of the solid occupied by the atoms, is of the order of
1/10. And a tenth of the atomic energy magnitude e2/ r is not far from the
value kTc• We conclude that
A~ (6.65)

which is the inverse of the typical factor that relates magnetic energy to electric
energy. The clear suggestion is that the underlying mechanism of ferromag­
netism is not magnetic, but electrical, in origin. The quantum theory of ferro­
magnetism initiated by Werner Heisenberg (1901-1976) vindicates this conclu­
sion.

6.5 Problems for Chapter 6

1. (a) Verify that a uniform electric field is described by

<j>~ —E •r, A = 0 (E = constant).

(b) Show that for such a uniform field acting on a system of charged
particles {ea}, the electric dipole moment is given by the analogue of
(6.27),
dE
<9E “
where E is the energy.
(c) Taking this differential equation as defining d generally, find the E
dependence of E in the case that d = do + #E.

2. What electromagnetic fields do the following potentials describe? Here

the fields are functions of r = («, y ) z) and t.

(a)
<j>= 0 , As — A z = 0, Ay = K x, (K — constant)

(b)
0 = o, Ax = Ay = 0, Az - Kt
6.5. PROBLEMS FOR CHAPTER 6 73

(c)
<j) = 0, A = a(a •r), (a = constant)

(d)
a •r
, A = 0; r > 0

(e)
= 0- r > 0.

3. Suppose we used the classical canonical (Boltzmann) distribution instead

of the quantum mechanical one treated in Section 6.4 for ferromagnetism,

J
that is,
( a*o ) t = ^ dSlti0 exp [//0 •( H + A M )/¿ r ],

where the partition function is

Z= J do exp[A*o •( H + XM)/kT].

Repeat the analysis given in Section 6.4 for this distribution and discuss
similarities and differences, particularly in the low and high temperature
limits.

4. Generalize (5.16) by including a time-dependent magnetic field B. Find

the equivalent of (5.18) in the case of a weak B. Specify the meaning of
iveak. Show that the polarization vector is linearly related to the electric
field,
P(w) = Xs(w) *E(w),
and find the electric susceptibility dyadic Xe(u )i whenever the magnetic
field is constant in time.

5. A charge e is bound harmonically to a center that is moving nonrelativis-

tically (v/c < 1 ) along the trajectory R (t). Show that the equation of
relative motion, including a damping force, is

d2 2 dr d2 _ ( 1 dH _ 1 dr
m _ r = _ mWor_ m 7 _ _ m _ R + e ^E + _ _ x B + - - x B

For E = 0 , and R (t) a circular motion with constant angular velocity fi,
show that to first order in Q there is an effective electric field acting on the
charge. What would thus be the polarization of a dielectric body rotating
in an external constant magnetic field?

6. A circular cylinder carries an axially symmetric charge density, and rotates

around its axis of symmetry. Write the resulting current as a curl, and by
inspection infer the induced magnetic field.
74 CHAPTER 6. MAGNETIC PROPERTIES OF M ATTER

7. Use the two preceding problems to find the magnetic susceptibility of a

circular cylinder with axially symmetric dielectric constant, that is rotat­
ing around its axis n in a constant magnetic field B = Bn. Can you
confirm the general rule M = - ^ X P ? For a homogeneous cylinder of
radius a and length / find the total magnetization.
Chapter 7

Macroscopic Energy and

Momentum

7.1 General Discussion

We now turn to the consideration of the distribution of energy and momentum of
electromagnetic fields within material media, following closely the development
of Chapter 3 (however, here we will assume no magnetic charge is present). We
will base our discussion on the macroscopic form of Maxwell’s equations, (4.60).
Accordingly, the rate at which the electric field does work on the free charges is

c 1 d '
J •E = E « (7.1)
4 ^ V X H — 4ñ d t B

If we add to this the parallel equation, appropriate to the absence of magnetic

charge,
e „ „ 1 5
0 = H• (7.2)
'47TV X 4v d t B
we obtain the suggestive form

J . E = - V . (¿E X H ) - ¿ ( E •¿ D + H •| B ) . (7.3)

[Recall that if there were free magnetic currents, (7.2) would represent the work
done on the magnetic charges.] Our aim is to write this result as a local energy
conservation law. We immediately identify, from the divergence term, the energy
flux or Poynting’s vector, S, to be

S = ¿E X H , (7.4)

which has the same form as that of the microscopic flux, (3.5), except that here
B is replaced by H. More intractable is the identification of the last term in

75
76 CHAPTER 7. MACROSCOPIC ENERGY AND MOMENTUM

(7.3). To what extent is it the negative time derivative of an energy density,

—dU/dt? If there does exist some quantity U such that

— ¡J ± — ( E.^ D + H (7.5)
dt 4w v

we would have a local statement of energy conservation,

— U + V - S + J . E = 0. (7.6)
at

Similarly, we consider the rate at which momentum is transferred to the

charges, or equivalently, the force density, f,

f = />E + -J x B . (7.7)
c

The relevant pair of equations here is

pE + i j x B = -i-(V .D )E + T f v x H - - ¿ ü ) XB, (7.8)

c 47t 47t \ c at J

0 = ¿ (V .B )H + ¿ ( v x E+ I | b) x D. (7.9)

which is added to yield

fE + ijx B = - ¿ ( ¿ D xb) - ¿ [DiV £ i - V . ( D E ) ]

-¿ [B iV /f.-V -iB H )]. (7.10)

From the time derivative, we identify the momentum density, G, to be

G = — DxB, (7.11)
47TC 7

which is analogous to (3.13) except the correct electric field here is D. If a local
momentum conservation law holds, the last two terms of (7.10)should be the
divergence of a tensor, representing the flow of momentum,

■ ^ [D i'V E i - V •(DE) + - V . (BH)] l V ♦T. (7.12)

Then, a local law of momentum conservation would hold,

7.2. NONDISPERSIVE MEDIUM 77

7.2 Nondispersive Medium

We cannot proceed further without specific assumptions about the properties
of the material medium. The simplest hypothesis is that of a homogeneous,
isotropic, nondispersive medium,

D = eE, B = /iH, (7.14)

where e and ¡jl are constants. This is not an unrealistic situation for many
substances over a sufficiently limited frequency range. For this case, the energy
density and the stress tensor exist, and have the following forms:

„ _ e g 2 + /zff2 m _ , e g 2 + /zff2 eEE + fiKB.

8» 8ir 4T ’ ( ' ’
while we recall that the energy flux and the momentum density are given by

S = -^ -E x H , G = -^ -E xH = ^S . (7.16)
47r 47tc c¿
It is interesting that we can transform these expressions, as well as Maxwell’s
equations, to look like those in vacuum, by redefining the fields, the charges,
and the speed of light as follows:

E? = y/iE, H' = y/JlH, c '= ^ = , p' = ^ p' (717)

(See Problem 7.1.) The ratio of c to cf is the index of refraction for the medium,

4 = n = ^/eji. (718)
c
By this transformation we see that the speed of propagation of electromagnetic
waves in the medium is c'; for propagation in a definite direction, the transcrip­
tion from the vacuum statement (see Section 3.4) that Ex and B ' are mutually
perpendicular and equal in magnitude is

eE 2 = i i H V E •H = 0. (7.19)

In the usual situation at low frequencies, where / i « 1, c > 1, the speed of light
in the medium, c', is less than that in vacuum, c, and the electric field is smaller
than the magnetic field, by the same factor:

c' = |E| = ! 5 i , n > 1. (7.20)

n n

7.3 Dispersive Medium

The naive way to incorporate the effects of a frequency dependence in the dielec­
tric constant (a dispersive medium) is to modify (7.15) and (7.18) by replacing
78 CHAPTER 7. MACROSCOPIC ENERGY AND MOMENTUM

the constants e and // with frequency dependent functions, e(u>) and /i(u>), which
for the simple model leading to (5.97), are

47rne2
// = !, e(w) = 1 + (7.21)
m u>2 —a?2 — iu>7

However, this replacement cannot be correct, because if we consider a frequency

large compared to wi, u u>i, the dielectric constant becomes less than unity,

e(w) (7.22)

where u>p is the plasma frequency, (5.50) with e = 1. Substituting this into
(7.18), we see that the speed of light in the medium, c', is larger than the speed
of light in the vacuum: c' = £ > c- If d is the speed of energy flow, as is
true for a nondispersive medium, this is impossible, since it manifestly violates
causality. (Causality is the concept that information cannot travel faster than
the speed of light in vacuum.)
We must now return to our starting point and carefully re-examine energy
flow. A realistic electromagnetic wave is characterized by a finite spatial extent,
so correspondingly contains a finite range of frequencies. This is relevant be­
cause the identification of the energy density according to (7.5) involves a time
integration, the building up of the field. But a field with a definite frequency
has no such transient behavior; a range of frequencies is required. Since the
properties of the medium are frequency dependent, the spread in frequency has
very important consequences, no matter how small the spread is.
We begin again by representing the time behavior of the electric field as a
superposition of frequencies (Fourier transform)

m = Jf—c du E(co)e —iu)t dw E(u>)*el< (7.23)

- r
where we have used the fact that E (t) is a real function, which implies that

E(-u>)* = E(w). (7.24)

Likewise, the displacement vector can be written as

/
oo
du> D (u )e~ iwt, (7.25)
-OO
where [see (5.31)]
D(ci) = c(w)E(w). (7.26)
Again, since D (t) is real, the dielectric constant obeys

e(-w )* = e(w), (7.27)

just as we saw in the model (7.21)— see (5.30). For simplicity, we will assume
e(u>) is real (that is, we stay away from the absorption regions) so that it is an
7.3. DISPERSIVE MEDIUM 79

even function o f lo. Since ordinary materials are non-magnetic, /z = 1, we will

take
B = H (7.28)
With this preparation we now return to our general discussion of energy flow,
in particular, to the purported equality (7.5). Since (7.28) holds, the magnetic
term is simple:
d H2 ,
<7-29>
The electric part can be written in terms of the product of two Fourier integrals,
(7.25) and the second form of (7.23) with u —►u/:

-If .A n J dw du1 eiw tE (—u ') ♦c(w)E(u ) ( - k j )e ~ iu,t

47T dt

J du dw'e~t(u’~u’ ^(-»)[w t(w ) —u/e(o/)]E(u>) •E (—u ').

(7.30)

In the last form the symmetry u> <-►—u/ has been made manifest. This can be
easily seen to be a time derivative by inserting the factor (u> — u/)/(u> — u /) so
that (7.30) becomes

5_JL
[ é j E(o.) ■E (-u .'). (7.31)
dt 87t J w —u>*

[A check of this result is to note that if we let e(a>) = e, we recover the result
for the nondispersive energy density, (e/87r)(E(¿))2.] This provides a general
expression for the electric part of the energy density.
Now suppose that the turning on of the field takes place so slowly that only
a very small band of frequencies about a central frequency & occurs (the precise
relation between turning-on time and band width is not required here— but see
Problem 7.5). Now with u « uj' « U or —a7, we have, since c(—w) = c(u;),

L)e(w) —u/e( u/)

(7.32)
UJ — U)f duj M w)]L =Sr

On the other hand, for cj & the above quantity is simply e(a7). However,
this contribution to the energy density oscillates in time with the frequency 267
and will not contribute significantly when averaged over one or more periods.
In addition, the spatial dependence will also be rapidly oscillating, suppressing
the contribution in spatial averages. In the sense of such averages, then, the
final form for the energy density in this situation is inferred from (7.29) and
(7.31) to be
H2 E2 d
(we(w)) (7.33)
u = s ^ + 8;r duj
To find the speed at which energy is propagated in an electromagnetic pulse,
we proceed as in Section 3.4. In regions without charges, p, J = 0, local energy
80 CHAPTER 7. MACROSCOPIC ENERGY AND MOMENTUM

conservation implies
U + V * S = 0, (7.34)
3t
which, when integrated over the pulse, implies

f J ( d r ) r U = J (d r )S , (7.35)

if we discard the surface term. Recalling the notion of energy weightings, we

write the magnitude of (7.35) as

E2
(dr) E x H , (7.36)
8ir

where [(3.39) and (3.42)]

v s = j t (r)B. (7.37)

Since we are considering a nearly monochromatic pulse, with the effects of dis­
persion adequately summarized in (7.32), the relations given at the end of Sec­
tion 7.2 between E and H are still true, at frequency U:

IP H2
|ExH| = E H = E2 (7.38)
v?’

We therefore conclude from (7.36) that the speed of propagation of energy at

frequency a; is
2y/e
|v b | = (7.39)
[^r(we) + e]
(For the situation when jx ^ 1, see Problem 7.3.)
For the specific model where the dielectric constant is given by (7.22),

[a>e(u>)] = 1 + > 1. (7.40)

du
The corresponding speed of energy flow is

|vfi| = y jl - W 2/ui2C< C, W > U)p. (7.41)

The speed o f propagation o f energy is less than the speed of light in vacuum, c,
in accord with the general idea that information cannot travel faster than c.

7.4 Problems for Chapter 7

1. Verify that Maxwell’s equations transform to the corresponding equations
in vacuum (where D f = E ', B ' = H') under the transformation (7.17)
when e and fi are constant.
7.4. PROBLEMS FOR CHAPTER 7 81

2. Show that (7.39) can be presented as

|ve| = c

3. Repeat the discussion in Section 7.3 when e(u>) = 1 and /¿(u>) ^ 1. What
happens if both e and \i are frequency dependent functions?

4. Use the model for the dielectric constant, (7.21), for |u/2 — u/2|>> tuy} so
that 6(u/) is real, to calculate the speed of energy flow, v # , from (7.39) in
the two cases oj < and w > uq. In both cases, show that the speed of
energy flow is less than c. (While this conclusion is not altered, the story
is more complicated when \u> — uq| ~ y. Treatment of this question by
Sommerfeld and Brillouin can be found in L. Brillouin, Wave Propagation
and Group Velocity, Academic Press, New York, 1960. See Problems 7
and 8 below.)

5. Imagine an electric field E(t) is slowly built up from a value of zero in the
distant past to a value Re (E(u/)e~Ut'i) in the present. Let T be the time
scale over which this turning on of the field occurs. Compute the band
width of frequencies, Ac*/, occuring in the Fourier transform of E (t).

6. A metal may be thought of, roughly, as a plasma in which electrons are

free to move. The dielectric constant is therefore approximately given by

u>2

The metal becomes transparent when the frequency w is large enough

so that n = yfe is real. Estimate the frequency at which copper (Nc =
8.5 x 1022 electrons/cm3) becomes transparent.

7. In Section 7.3, we considered the electric part of the energy density in

the absence of absorption. We will here consider absorption within the
framework o f the dielectric model discussed in Section 5.2. There, a simple
damping model is employed, and, it is clear that this damping will remove
a certain amount of power per unit volume, V.

(a) Calculate V and write it in the form of (7.30).

(b) We might expect that (7.5) needs to be modified in the presence of
absorption to read

4it (E- s D+H- | B)

Determine the electric part of U.
(c) Find the narrow band analog of (7.33).
 CHAPTER 7. MACROSCOPIC ENERGY AND MOMENTUM

8. Use the above formulation to calculate a propagation speed in the presence

of absorption. In particular, for the oscillator model of Section 5.2, derive
the speed of energy flow found by Brillouin (suitably corrected),

Ren
vE = c ------ ,

where

1+
( W2- W2)2 + W V

9. (a) The Pauli exclusion principle asserts that only one electron can oc­
cupy a given state. The number of states per element of phase space
is given by
0 (<fr)(dp)
ft3
where h is Planck’s constant, and the factor of 2 arises from the two
possible spin orientations of the electron. At zero temperature the
lowest energy states are completely filled. Show that the energy of
the highest occupied state, the Fermi energy, is

where h = ~ = 1.05 x 10"“27 erg-s, and n is the number density.

(b) In a metal, the conduction electrons are essentially free. Let n(r) —no
be the (small) deviation from a uniform electron concentration. The
chemical potential
H = EF( r)-e<£(r),
where <f> is the electrostatic potential, is constant is space in equilib­
rium. Show then that (j) satisfies

V 2<j>= A2<¿,

and compute A in terms of no. What value does A have for copper?
Show that the spherically symmetric solution of this equation is the
screened Coulomb potential,
e -A r

</Kr ) = « - —
r •

(c) From the analysis of (b), we see that placing an external charge dis­
tribution pext in a metal leads to an induced electronic charge density
pe. Derive the following relation between the Fourier transforms,
7.4. PROBLEMS FOR CHAPTER 7 83

and hence show that this situation corresponds to a wavevector-

dependent dielectric constant,

(d) An alternative derivation of this last result follows by noting that

*ftext(k)
e(k) =
m
<!> being the total potential, and 0ext that due to the external charge
distribution. From (b) show that

4irpext(k) 4X/Jext(k)
@ex\ ( k ) — m =
k2 k 2 + A2 ’
from which the result follows.
Chapter 8

Review of Action Principles

We have already mentioned the Hamiltonian formulation for the particle equa­
tions of motion [see (6.17)-(6.19)]. We now want to show that the whole system
of particles and fields is a mechanical system derivable from a Hamilton action
principle.
We start by reviewing and generalizing the Lagrange-Hamilton principle
for a single particle. The action, W 12 , is defined as the time integral of the
Lagrangian, L, where the integration extends from an initial configuration or
state at time t% to a final state at time 1 1 :
ti
dt L ( 8 . 1)
2

The integral refers to any path, any line of time development, from the initial
to the final state, as shown in Fig. 8.1. The actual time evolution of the system
is selected by the principle of stationary action: In response to infinitesimal
variations of the integration path, the action W 12 is stationary— does not have
a corresponding infinitesimal change— for variations about the correct path,
provided the initial and final configurations are held fixed,

SW12 = 0. (8.2)

Figure 8.1: A possible path from initial state to final state.

85
86 CHAPTER 8. REVIEW OF ACTION PRINCIPLES

This means that, if we allow infinitesimal changes at the initial and final times,
including alterations of those times, the only contribution to SW\2 then comes
from the endpoint variations, or

6W 12 = G i - G2, (8.3)

where G a, a = 1 or 2, is a function depending on dynamical variables only at

time ta. In the following, we will consider three different realizations of the
action principle, where, for simplicity, we will restrict our attention to a single
particle.

8.1 Lagrangian Viewpoint

The nonrelativistic motion of a particle of mass m moving in a potential F (r, t)
is described by the Lagrangian

í=H£)í~nM)' (8-4)
Here, the independent variables are r and t } so that two kinds of variations can
be considered. First, a particular motion is altered infinitesimally, that is, the
path is changed by an amount <$r:

r(<) —►r(i) + 6r(<). (8.5)

Second, the final and initial times can be altered infinitesimally, by St\ and St2 ,
respectively. It is more convenient, however, to think of these time displacements
as produced by a continuous variation of the time parameter, St(t),

t - > t + St(t), 8.6

so chosen that, at the endpoints,

St(t1) = Stu St(t2) = St2 (8.7)

The corresponding change in the time differential is

) d< , ( 8 . 8)

which implies the transformation of the time derivative,

± ( dSt\ d_
(8.9)
dt \ dt ) dt

Because of this redefinition of the time variable, the limits of integration in the
action,

(8.10)
8.2. HAMILTONIAN VIEWPOINT 87

are not changed, the time displacement being produced through 6t(t) subject

}
to (8.7). The resulting variation in the action is now

Sii v
dr d dSt
SW12 = ¿r.V F -
™dt * d tSr dt

r_ £ _ _ 1 (d r
+ Sr *
[ m*2r v v -I* 6t —m —
2 V dt

where, in the last form, we have shifted the time derivatives in order to isolate
<$r and St.
Because <$r and St are independent variations, the principle of stationary
action implies that the actual motion is governed by

d2
v = - w’ ( 8 . 12)

LmUj +yJ
A ri
dt 2
(d r y i
= -v
dt ’
(8.13)

while the total time derivative gives the change at the endpoints,

G = p •Sr — ESt, (8.14)

with

momentum = p — m ~ -, energy — E — - m (8.15)

* dt ¿i

Therefore, we have derived Newton’s second law [the equation of motion in

second-order form], (8.12), and, for a static potential, dV/dt = 0, the conser­
vation of energy, (8.13). The significance of (8.14) will be discussed later in
Section 8.4.

8.2 Hamiltonian Viewpoint

Using the above definition of the momentum, we can rewrite the Lagrangian as

L = p -^ -H (r ,p ,t ), (8.16)

where we have introduced the Hamiltonian

88 CHAPTER 8. REVIEW OF ACTION PRINCIPLES

We are here to regard r, p, and i as independent variables in

W i2 = j \ p ' d v - d t H ] . (8.18)

The change in the action, when r, p, and t are all varied, is

dc . dH . dv 8H dSt
6W 12 = j f dt [|>

= 1 -isr)
(8.19)

The action principle then implies

dr dH _ p
(8 .2 0 )
dt dp ra’
dp dH
( 8. 21)
dt dr
dH dH
(8 .22)
dt dt ’
G = p •<$r — H6t. (8.23)

In contrast with the Lagrangian differential equations of motion, which involve

second derivatives, these Hamiltonian equations contain only first derivatives;
they are called first-order equations. They describe the same physical system,
because when (8.20) is substituted into (8.21), we recover the Lagrangian-
Newtonian equation (8.12). Furthermore, if we insert (8.20) into the Hamil­
tonian (8.17), we identify H with E. The third equation (8.22) is then identical
with (8.13). We also note the equivalence of the two versions of G.
But probably the most direct way of seeing that the same physical system
is involved comes by writing the Lagrangian in the Hamiltonian viewpoint as

The result of varying p in the stationary action principle is to produce

dr
(8.25)
p = m di-
But, if we accept this as the definition of p, the corresponding term in L disap­
pears and we explicitly regain the Lagrangian description. We are justified in
completely omitting the last term on the right side of (8.24), despite its depen­
dence on the variables r and t ) because of its quadratic structure. Its explicit
contribution to SL is
_1_ (( p _ m _dr\j . (^ p _ m _d S r + r o _dr _dSt\
j , (8.26)
m
8.3. A THIRD VIEWPOINT 89

and the equation supplied by the stationary action principle for p variations,
(8.25), also guarantees that there is no contribution here to the results of r and
t variations.

8.3 A Third Viewpoint

Here we take r, p, and the velocity, v, as independent variables, so that the
Lagrangian is written in the form

i = p * ( < ^ ~ v) + = (8-27)

where
H ( r ,p ,v ,t ) = p . v - imt;2 + V(r ,t). (8.28)

The variation of the action is now

i
SW12 = [p •dr — H dt]

X dr ^ d dH 8H 8H
,ñr - 3 í + p - dx

i
dt dt
dp 8H \
— (p •Sr — HSt) — Sr <
at dt ^ dr )
8H \ (dH_ dH
+ 5p ¿V • + St (8.29)
dp) ov V * dt

so that the action principle implies

dp dH
= -W , (8.30)
dt dr
dv dH _
v, (8.31)
dt dp ’

0 _ _dH
_ = _ p+raV) (8.32)
dH dH
(8.33)
dt dt ’
G p * <5r — HSt. (8.34)

Notice that there is no equation of motion for v since dv/dt does not occur in
the Lagrangian, nor is it multiplied by a time derivative. Consequently, (8.32)
refers to a single time and is an equation of constraint.
From this third approach, we have the option of returning to either of the
other two viewpoints by imposing an appropriate restriction. Thus, if we write
90 CHAPTER 8. REVIEW OF ACTION PRINCIPLES

(8.28) as

H ( r , p , v , t ) = § ^ + V (r ’ t ) ~ 2 ~ ( P — m v)2> (8-35)

and we adopt
V = —p (8.36)
m
as the definition of v, we recover the Hamiltonian description, (8.16) and (8.17).
Alternatively, we can present the Lagrangian (8.27) as

_ m ( dr\2 , ( dr \ m ( dr \2
L = 2{di) + ■ (8'37)

Then, if we adopt the following as definitions,

v = ^ , p = mv, (8.38)

the resultant form of L is that of the Lagrangian viewpoint, (8.4). It might

seem that only the definition v = dr/dt, inserted in (8.37), suffices to regain the
Lagrangian description. But then the next to last term in (8.37) would give the
following additional contribution to ¿L, associated with the variation Sr:

(p — m v) •— (Sr. (8.39)
dt
In the next chapter, where the action formulation of electrodynamics is con­
sidered, we will see the advantage of adopting this third approach, which is
characterized by the introduction of additional variables, similar to v, for which
there are no equations of motion.

8.4 Invariance and Conservation Laws

There is more content to the principle of stationary action than equations of
motion. Suppose one considers a variation such that

SW12 = 0, (8.40)

independently of the choice of initial and final times. We say that the action,
which is left unchanged, is invariant under this alteration of path. Then the
stationary action principle (8.3) asserts that

SW12 = Gi - G 2 = 0, (8.41)

or, there is a quantity G (t) that has the same value for any choice of time t\ it
is conserved in time. A differential statement of that is

j f G {t) = 0. (8.42)
8.4. INVARIANCE AND CONSERVATION LAWS 91

The G functions, which are usually referred to as generators, express the inter­
relation between conservation laws and invariances of the system.
Invariance implies conservation, and vice versa. A more precise statement is
the following:

If there is a conservation law, the action is stationary under an in­

finitesimal transformation in an appropriate variable.

The converse of this statement is also true.

If the action W is invariant under an infinitesimal transformation

(that is, SW = 0), then there is a corresponding conservation law.

This is the celebrated theorem proved by Amalie Emmy Noether (1882-1935).

Here are some examples. Suppose the Hamiltonian of (8.16) does not depend
explicitly on time, or

W 12 = J [p •dr — H (r, p)dt].

Then the variation (which as a rigid displacement in time, amounts to a shift

in the time origin)
St = constant (8.44)
will give SW\2 = 0 [see the first line of (8.19), with Sr = 0, Sp = 0, dSt/dt = 0,
dH/dt = 0]. The conclusion is that G in (8.23), which here is just

Gt = -H S t, (8.45)

is a conserved quantity, or that

(8.46)

This inference, that the Hamiltonian—the energy— is conserved, if there is no

explicit time dependence in H , is already present in (8.22). But now a more
general principle is at work.
Next, consider an infinitesimal, rigid rotation, one that maintains the lengths
and scalar products o f all vectors. Written explicitly for the position vector r,
it is
¿r = í w X r , (8.47)
where the constant vector Suj gives the direction and magnitude of the rotation
(see Fig. 8.2). Now specialize (8.17) to

(8.48)

where r = |r|, a rotationally invariant structure. Then

W 12 = \p‘ d r — H dt] (8.49)

92 CHAPTER 8. REVIEW OF ACTION PRINCIPLES

Figure 8.2: S u X r is perpendicular to Sio and r, and represents an infinitesimal

rotation of r about the 8u axis.

is also invariant under the rigid rotation, implying the conservation of

G$u> = p •Sr = <$u> •r Xp. (8.50)

This is the conservation of angular momentum,

(8.51)
i-3

h3

©
X
u
II

II

Of course, this is also contained within the equation of motion,

dL
— T = —rX W = - r X r~9V
— := 0, (8.52)
at dr
since V depends only on |r|.
Conservation of linear momentum appears analogously when there is invari­
ance under a rigid translation. For a single particle, (8.21) tells us immediately
that p is conserved if V is a constant, say zero. Then, indeed, the action

(8.53)
Wu=i H '- f ? "
is invariant under the displacement

Sy = 6e = constant, (8.54)

and
Gse = p *8e (8.55)
is conserved. But the general principle acts just as easily for, say, a system of
two particles, a and 6, with Hamiltonian

H = j L + l L + v ( r a - v b). (8.56)
2m a 2mb
This Hamiltonian and the associated action
8.4. INVARIANCE AND CONSERVATION LAWS 93

are invariant under the rigid translation

6ra = 6rb = 6e, ( 8 . 58 )

with the implication that

Gse = Pa • + p b •Srb = (p a + p b) •Se (8.59)

is conserved. This is the conservation of the total linear momentum,

P = Pa+p&, ^ P = 0. (8.60)

Something a bit more general appears when we consider a rigid translation

that grows linearly in time:

6ra = 6rb — 6 v t y (8.61)

using the example of two particles. This gives each particle the common addi­
tional velocity 6v, and therefore must also change their momenta,

6pa = maSv, Spb — mb6v. (8.62)

The response of the action (8.57) to this variation is

SW12 = i i(p a + Pb) •Sv dt + Sv . (madra + mjoirj) - (p a + Pi) •6v dt]

l
d[(mar a + m jrj) •¿v ]. (8.63)

The action is not invariant; its variation has end-point contributions. But there
is still a conservation law, not of G = P •Svt, but of N •Sv, where

N = Pt - (mar„ + nun). (8.64)

Written in terms of the center-of-mass position vector

„ mar a + mi r h
R = -------—-------, M = ma+nib, (8.65)

the statement of conservation of

N = Pi - M R, ( 8 . 66)

namely
dN dR
0 = P - M (8.67)
dt dt ’
is the familiar fact that the center of mass of an isolated system moves at the
constant velocity given by the ratio of the total momentum to the total mass of
that system.
94 CHAPTER 8. REVIEW OF ACTION PRINCIPLES

8.5 Nonconservation Laws. The Virial Theorem

The action principle also supplies useful nonconservation laws, Consider, for
constant ¿A,
£r = ¿Ar, £p = —¿Ap, (8.68)
which leaves p •dr invariant,

¿(p •dr) = (—<$Ap) •dr + p •(<$Adr) = 0. (8.69)

But the response of the Hamiltonian

H = T(p) + V ( r), T(p) = (8.70)

is given by the noninvariant form

8H = SX(—2T + r •W ) . (8-71)

Therefore we have, for an arbitrary time interval, for the variation of the action
(8.18),

/*! /»1 j
SW12 = j d t [ 6 X ( 2 T - r .V V ) ] = G 1 - G 2 = j dt— ( p . « A r ) (8.72)

or, the theorem

d
—r - p = 2 T - r . W . (8.73)

This is an example of the mechanical virial theorem to which we referred at the

end of Section 3.3.
For the particular situation of the Coulomb potential between charges, V =
constant/r, where
i
r« W = r — V = -V , (8.74)
dr
the virial theorem asserts that

^ ( r . p ) = 2T + K (8.75)

We apply this to a bound system produced by a force of attraction. On taking

the time average of (8.75) the time derivative term disappears. That is because,
over an arbitrarily long time interval r = t\ —t 2) the value of r •p(ti) can differ
by only a finite amount from r *p(^2), and

r . p ( t i ) - r » p ( < 2) iQ
r *p (8.76)

as t —►oo. The conclusion.

2T = —V, (8.77)
8.6. PROBLEMS FOR CHAPTER 8 95

has been used qualitatively in Section 5.2.

Here is one more example of a nonconservation law: Consider the variations

6r = SX-,
r

(8.78)

Again p •dr is invariant:

and the change of the Hamiltonian (8.70) is now

+ -* W . (8.80)
mró r

The resulting theorem, for V = U(r), is

(8.81)

which, when applied to the Coulomb potential, gives the bound-state time av­
erage relation

.(8.82)

This relation is significant in hydrogen fine-structure calculations.

8.6 Problems for Chapter 8

1. Suppose the system consists of N particles interacting through a pairwise
potential V (ra—r&). Write down the Lagrangian and obtain the equations
of motion. What is the Hamiltonian, i7 (ra,Pa)? Show that energy and
total momentum are conserved. What is required for angular momentum
to be conserved?

2. For a free relativistic particle of rest mass mo, the energy is

Use this as the Hamiltonian H ) and from the Lagrangian

determine the relationship between the velocity v = dr/dt and the momen­
tum. Compute the energy in terms of the velocity. Write the Lagrangian
in terms o f v.
96 CHAPTER 8. REVIEW OF ACTION PRINCIPLES

3. Consider a particle bound by a potential of the form

V = arh.

Derive the time-aver aged virial theorem relating T to V. What is the

smallest value of b for which a bound state can occur?
Chapter 9

Action Principle for

Electrodynamics

9.1 Action of Particle in Field

It was stated in our review of mechanical action principles that the third view­
point, which employs the variables r, p, and v, was particularly convenient for
describing electromagnetic forces on charged particles. With the explicit, and
linear, appearance of v in what plays the role of the potential function in (6.12),
we begin to see the basis for that remark. Indeed, we have only to consult (8.27)
to find the appropriate Lagrangian:

L= p • + ^m v2 - e<f>+ A. (9.1)

To recapitulate, the equations resulting from variations of p, r, and v are,

respectively,
dr
= v, (9.2)
dt

- p = -.V (j)----- V « (9.3)

c
p = mv + - A . (9.4)

We can now move to either the Lagrangian or the Hamiltonian formulation.

For the first, we simply adopt v = dr/dt as a definition (but see the discussion
in Sec. 8.3) and get
_ 1 ( d r\ 2 . e dr . _\
2 \dt J cd t v '
Alternatively, we use (9.4) to define

v (9.6)

97
98 CHAPTER 9. ACTION PRINCIPLE FOR ELECTRODYNAMICS

and find
dr
(9.7)
p - d ¡ ~ H’
1 / e \2
(9.8)
* = 2s ( » - ; a) + e *-
Here we make contact with the energy considerations of Sec. 6.1; in particular,
H coincides with the form of the energy given in (6.16).

9.2 Electrodynamic Action

The electromagnetic field is a mechanical system. It contributes its variables
to the action, to the Lagrangian of the whole system of charges and fields. In
contrast with the point charges, the field is distributed in space. Its Lagrangian
should therefore be, not a summation over discrete points, but an integration
over all spatial volume elements,

¿field = J (dr) £ fleid; (9.9)

this introduces the Lagrange function, or Lagrangian density, C. The total

Lagrangian must be the sum of the particle part, (9.1), and the field part, (9.9),
where the latter must be chosen so as to give the Maxwell equations (1.65):
— « 1 47r .
V xB = E + — j, V - E = 4tt/>,
c at c
1 f)
-V xE = B, V •B = 0. (9.10)
c ut
The homogeneous equations here are equivalent to (6.6) and (6.3),

i « A = - E -V *, (9.11)

B = VxA. (9.12)

Thus, we recognize that A(r,t), E(r, t), in analogy with r(t), p(t), obey equa­
tions of motion while <¡>(v, t), B(r,¿), as analogues of v(t), do not. There are
enough clues here to give the structure of £fieid> apart from an overall factor.
The anticipated complete Lagrangian for microscopic electrodynamics is

'dtra ^ , 1
L = - v a ) + - m av 2 - e oi6(ra) + ^ v a . A ( r a)

+ E•( ) - B - V x A + ^ (B 2 - E 2) .(9.13)
47T r ; f A - v *
The terms that are summed in (9.13) describe the behavior of charged parti­
cles under the influence of the fields, while the terms that are integrated describe
the field behavior. The independent variables are

r a (*)> v a(<)> Pa(<)> íK M ), A(r,í), E (r ,i), B( r ,i ), t. (9.14)

9.2. ELECTRODYNAMIC ACTION 99

We now look at the response of the Lagrangian to variations in each of these

variables separately, starting with the particle part:

6L = j t (6ra - p a) + 8ra - - V aea (<j>(ra) - ^ • A ( O ) ] ,

(9.15)
6L = 8va • [-P a + mav a + y A ( r j ] , (9.16)

Spa SL = S p ,- ( ^ - t. ) - (9.17)

The stationary action principle now implies the equations of motion

^ = - e aV a (¿(ra) - ^ . A ( r a) ) , (9.18)

mav a = Pa - ~ A ( r a), (9.19)

dr a
(9.20)
ya = ~dP

which are the known results, (9.2)-(9.4).

The real work now lies in deriving the equations of motion for the fields.
In order to cast all the field-dependent terms into integral form, we introduce
charge and current densities,

P (r,t) = ^ e 0<$( r -r a(f)),

a

j(M ) = X ) eaV°(<)5(r - r a W ) ’ (9.21)

so that

5 3 [~ ea<K*a) + y v a •A ( r j ] = J (d r ) ' + - j •A (9.22)

c

The volume integrals extend over sufficiently large regions to contain all the
fields of interest. Consequently, we can integrate by parts and ignore the sur­
face terms. The responses of the Lagrangian (9.13) to field variations, and the
corresponding equations of motion deduced from the action principle are

6<j>: 6L= (dr) 6<¡>(V •E - 4%p), (9.23)

V *E = 47rp, (9.24)

8A : 6L - ÍA .E
4 ttc (
1 f /J \ CA i l9E 4jr- ~
(9.25)
100 CHAPTER 9. ACTION PRINCIPLE FOR ELECTRODYNAMICS

1 9_ 47T.
V xB = (9.26)
~cdiE + T J’

6L =
¿ /<*>“ •( (9.27)

1 d
E = (9.28)
- < I M A - V *-
SL = ¿ / ( * ) » • ( ■ - V x A + B), (9.29)

B = V XA. (9.30)

We therefore recover Maxwell’s equations, two of which are implicit in the con­
struction of E and B in terms of potentials. By making a time variation of the
action [variations due to the time dependence of the fields vanish by virtue of
the stationary action principle— that is, they are already subsumed in (9.23)-
(9.30)],

St : (9.31)

we identify the Hamiltonian of the system to be

H = f ( Pa “ — A (r a)) *V<» ~ \ m*vl +

a *• ■

+¿ / (dr) [e *v ^ + b *V x A + ^ 2 - b2) - (9.32)

which is a constant of the motion, dH/dt = 0. The generators are inferred from
the total time derivative terms in (9.15), (9.25), and (9.31),

SW\2 = C?i — g 2, (9.33)

to be
G = ^ i r a .pa E -6 A -H 6 t. (9.34)
a

9.3 Energy
Notice that the total Lagrangian (9.13) can be presented as

t =I> a
<935>
where the Hamiltonian is given by (9.32). The narrower, Hamiltonian, descrip­
tion is reached by eliminating all variables that do not obey equations of motion,
and, correspondingly, do not appear in G. Those “superfluous” variables are
the v a and the fields (¡> and B, which are eliminated by using (9.19), (9.24), and
9.3. ENERGY 101

(9.30), the equations without time derivatives, resulting, first, in the intermedi­
ate form
r£2 + £2
H = E ( ¿ (p* - ~ A(r<o)2 + + / (dr) 87T - p<t>

(9.36)
The first term here is the energy of the particles moving in the field [particle
energy— see (9.8)], so we might call the second term the field energy. The
ambiguity of these terms (whether the potential energy of particles is attributed
to them or to the fields, or to both) is evident from the existence of a simpler
form of the Hamiltonian

E2+ B2
7 A (r a)) + j ( d*) (9.37)
8?r

where we have used the equivalence of the two terms involving given in (9.22).
This apparently startling result suggests that the scalar potential has disap­
peared from the dynamical description. But, in fact, it has not. If we vary the
Lagrangian (9.35), where H is given by (9.37), with respect to E we find

ít=-¿/<iir»sE-GlA+E)=o- <9-38>
Do we conclude that + E = 0? That would be true if the <$E(r,¿) were
arbitrary. They are not; E is subject to the restriction— the constraint— (9.24),
which means that any change in E must obey

(9.39)
^0
>

tí

0
II

The proper conclusion is that the vector multiplying <SE in (9.38) is the gradient
of a scalar function, just as in (9.28),

~ A + E = -V ¿, (9.40)
c ot

for that leads to

SL = —E J (dr) ( V •8E)<j> = 0, (9.41)

as required.
The fact that the energy is conserved,

dH A
(9.42)
-3T = ° '
where
(9.43)
a J
102 CHAPTER 9. ACTION PRINCIPLE FOR ELECTRODYNAMICS

is a simple sum of particle kinetic energy and integrated field energy density,
can be verified directly by taking the time derivative of (9.36). The time rate
of change of the particle energy was computed in (6.14),

^ Y ( \ m*V* + e« ¿ ( r « ) ) = Y , ^ - 7 - v « * A (r a)) • (9-44)

a ' ' a

We can compute the time derivative of the field energy by using the equation
of energy conservation, (3.7),

J (d r ) U = - J (d r ) j •E, (9.45)

to be

= - /< * ) N í* - ? - s a]

a N x
(9.46)

Here we have used (9.28), and have noted that

= 0 (9.47)
dr

by charge conservation. Observe that (9.44) and (9.46) are equal in magnitude
and opposite in sign, so that their sum is zero. This proves the statement of
energy conservation (9.42).

9.4 Momentum and Angular Momentum Con­

servation
The action principle not only provides us with the field equations, particle equa­
tions of motion, and expressions for the energy, but also with the generators
(9.34). The generators provide a connection between conservation laws and in­
variances of the action (recall Section 8.4). Here we will further illustrate this
connection by deriving momentum and angular momentum conservation from
the invariance of the action under rigid coordinate translations and rotations,
respectively. [In a similar way we could derive energy conservation, (9.42), from
the invariance under time displacements— see also Section 9.6.]
Under an infinitesimal rigid coordinate displacement, <$€, a given point which
is described by r in the old coordinate system is described by r + Se in the new
9.4. MOMENTUM AND ANGULAR MOMENTUM CONSERVATION 103

one. (See Fig. 9.1.) The response of the particle term in (9.34) is simple:
Se • Y2a f° r the field Parf >we require the change, 6A , of the vector potential
induced by the rigid coordinate displacement. The value of a field T at a
physical point P is unchanged under such a displacement, so that if r and r + <$€
are the coordinates of P in the two frames, there are corresponding functions F
and F such that
F {P ) = F ( r) = F ( r + Se), (9.48)
that is, the new function F of the new coordinate equals the old function F
of the old coordinate. The change in the function F at the same coordinate is
given by
F (r) = F (r) + 6F(r), (9.49)
so that
SF(r) = F (r - Se) - F (r) = - 8 e •V F { r), (9.50)
for a rigid translation (not a rotation).
As an example, consider the charge density

/>(r ) = ^ 2 ea S ( r - r a). (9.51)

a

If the positions of all the particles, the r a, are displaced by ¿c, the charge density
changes to
p(r) + 6p(r) = ^ 2 ea6(r - r a - 6e), (9.52)

where
5(r - ra —Se ) = 6(r — ra) —6 e •V r 6 ( r — r a), (9.53)
104 CHAPTER 9. ACTION PRINCIPLE FOR ELECTRODYNAMICS

and therefore
<5p(r) = -<5e.Vp(r), (9.54)
in agreement with (9.50).
So the field part o f G in (9.34) is

- /<*) 'SA = ¿ /(*> •V ) A

= - J X I e«f a •*(>■*)+ ¿ y < * ) (E x B ).i« ,(9 5 5 )

where the last rearrangement makes use of (9.24) and (9.30), and the vector
identity
íex(VxA) = V(¿e.A)-(¿€. V)A. (9.56)
Including the particle part from (9.34) we find the generator corresponding to
a rigid coordinate displacement can be written as

G = 6 e- P, (9.57)

where

P = E(
a
p« - 7 A (r a)) + ¿ / ( * ) E X B = 2 > 0v a +
J a J
f (dr) G , (9.58)

with G the momentum density, (3.13). Since the action is invariant under a
rigid displacement,

0 = 6W = Gi - G2 = (P i - P 2) •«r, (9.59)

we see that
P i = P 2, (9.60)
that is, the total momentum, P , is conserved. This, of course, can also be
verified by explicit calculation:

£ ’ 1. „
J(dr)±-EXB = - J ( d r ) p E + - J X B
dt c

= - E e« (E(rfl)+JvaXB(rjj, (9.61)

which restates (3.15), from which the constancy of P follows from (6.1).
Similar arguments can be carried out for a rigid rotation for which the change
in the coordinate vector is
6r = Su>Xr, (9.62)

with 6u constant. The corresponding change in a vector function is

A (r -f Sr) = A( r) + 6u> X A (r) (9.63)

9.5. GAUGE INVARIANCE AND THE CONSERVATION OF CHARGE 105

since a vector transforms in the same way as r, so the new function at the initial
numerical values of the coordinates is

A (r ) = A (r ) — (<Sr •V ) A (r) + Su X A (r). (9.64)

(See Problem 9.2.) The change in the vector potential is

5A = -(tfr •V ) A + <5u>X A . (9.65)

The generator can now be written in the form

G= (9.66)

where the total angular momentum, J, is found to be (see Problem 9.4)

J = ^ r aXmav a + J (dr) r X ^ j ^ E x B ^ , (9.67)

which again is a constant of the motion.

9,5 Gauge Invariance and the Conservation of

Charge
An electromagnetic system possesses a conservation law, that of electric charge,
which has no place in the usual mechanical framework. It is connected to a
further invariance of the electromagnetic fields— the potentials are not uniquely
defined in that if we let
1 fi
A - + A + VA, A, (9.68)

the electric and magnetic fields defined by (9.28) and (9.30) remain unaltered,
for an arbitrary function A. This is called gauge invariance; the corresponding
substitution (9.68) is a gauge transformation. [The term has its origin in a now
obsolete theory (1918) of Hermann Weyl (1885-1955).]
This invariance of the action must imply a corresponding conservation law.
To determine what is conserved, we compute the change in the Lagrangian,
(9.13), explicitly. Trivially, the field part of L remains unchanged. In considering
the change of the particle part, we recognize that (9.68) is incomplete; since v is
a physical quantity, p —(e /c )A must be invariant under a gauge transformation,
which will only be true if (9.68) is supplemented by

p —►p + - VA. (9.69)

c
Under the transformation (9.68) and (9.69), the Lagrangian becomes

t - r s i + £n.[ •*T VA- (x ^ - v") + T S A+ T ’,*-VAJ

106 CHAPTER 9. ACTION PRINCIPLE FOR ELECTRODYNAMICS

(± X + * ± .V > )
\dt dt J
a
d
(9.70)
dtW’
where
w = J 2 ~ x (Va’ t)- (9.71)

What is the physical consequence of adding a total time derivative to a La-

grangian? It does not change the equations of motion, so the system is unal­
tered. Since the entire change is in the end point behavior,

W 12 = W \2 + ( w i - w 2 ) y (9.72)

the whole effect is a redefinition of the generators, G,

G = G -f 6w. (9.73)

This alteration reflects the fact that the Lagrangian itself is ambiguous up to a
total time derivative term.
To ascertain the implication of gauge invariance, we rewrite the change in
the Lagrangian given in the first line of (9.70) by use of (9.20),

! / < * ) [ , | A + j.V A , (9.74)

and apply this result to an infinitesimal gauge transformation, A —►<$A. The

change in the action is then

s w 12 = G 6M - G ex, - J ' d t j ( d r ) U x (J ¡p + V •j ) , (9.75)

with the generator being

Gía = j ( d r ) - c p6X. (9.76)

In view of the arbitrary nature of 6\(r, t ), the stationary action principle now
demands that, at every point,

s / . + v . j = o, (9.77)

that is, gauge invariance implies local charge conservation. (O f course, this
same result follows from Maxwell’s equations.) Then, the special situation
6X = constant, where 6A = 8<j) = 0, and W\2 is certainly invariant, implies
a conservation law, that of
Gsx = -6 X Q , (9.78)
in which
= /(* )/ (9.79)

is the conserved total charge.

9.6. GAUGE INVARIANCE AND LOCAL CONSERVATION LAWS 107

9.6 Gauge Invariance and Local Conservation

Laws
We have just derived the local conservation law of electric charge. Electric
charge is a property carried only by the particles, not by the electromagnetic
field. In contrast, the mechanical properties of energy, linear, and angular mo­
mentum are attributes of both particles and fields. For these we have conser­
vation laws of total quantities. What about local conservation laws? Early in
this development (Chapter 3) we produced local non-conservation laws; they
concentrated on the fields and characterized the charged particles as sources (or
sinks) of field mechanical properties. It is natural to ask for a more even-handed
treatment of both charges and fields. We shall supply it, in the framework of a
particular example. The property of gauge invariance will be both a valuable
guide, and an aid to simplifying the calculations.
The time displacement of a complete physical system identifies its total en­
ergy. This suggests that time displacement of a part of the system provides
energetic information about that portion. The ultimate limit of this spatial
subdivision, a local description, should appear in response to an (infinitesimal)
time displacement that varies arbitrarily in space as well as in time, <ft(r,¿).
Now we need a clue. How do fields, and potentials, respond to such coordi­
nate-dependent displacements? This is where the freedom of gauge transfor­
mations enters: The change of the vector and scalar potentials, by V A (r,¿),
—(l/c)(d/dt)Á (r,t), respectively, serves as a model for the potentials them­
selves. The advantage here is that the response of the scalar A(r,t) to the time
displacement can be reasonably taken to be
(A + 6X)(r,t + 6t) = A(r , t ) , (9.80)
or
<$A(r,t) = (9.81)
Then we derive

{(VA) = - « ¿ ( V A ) + ( - I | a) cV í í, (9.82)

(9.83)

which is immediately generalized to

Q
SA = - S t ~ A + <¡>cV6t, (9.84)
at
f) f)
6$ = - 5t — ¿í , (9.85)

or, equivalently,
8A. — cStlij “I- V(<^c<5t), (9.86)

S<j> = --^-{<j>c8t). (9.87)

c at
108 CHAPTER 9. ACTION PRINCIPLE FOR ELECTRODYNAMICS

In the latter form we recognize a gauge transformation, produced by the scalar

<j>c8t, which will not contribute to the changes of field strengths. Accordingly,
for that calculation we have, effectively, 6A = cStE, 6<j> = 0 , leading to

(9.88)

the last line employs the field equation V x E = —(l/c )(3 B /3 t ).

In the following we adopt a viewpoint in which such homogeneous field equa­
tions are accepted as consequences of the definition of the fields in terms of
potentials. That permits the field Lagrange function (9.13) to be simplified:

Afield = L (£ 2 _ B 2). (9.89)

Then we can apply the field variation (9.88) directly, and get

¿ ¿ fie ld = S t^ C ñ M ~ ^ E 2 § ¡ St ~ ¿ E X B *

= -— (« ¿ a d d ) - ¿ ( # 2 + - ¿ E x B •VSt. (9.90)

Before commenting on these last, not unfamiliar, field structures, we turn to

the charged particles and put them on a somewhat similar footing in terms of
a continuous, rather than a discrete, description.
We therefore present the Lagrangian of the charges in (9.13) in terms of a
corresponding Lagrange function,

(9.91)

where
(9.92)
a
and

the latter adopts the Lagrangian viewpoint, with v a = dra/dt accepted as a

definition. Then, the effect of the time displacement on the variables r a(t),
taken as

(ra + 6va)(t + St) = r 0(t), (9.94)

Sra(t) = -S t(r a,t ) v a(t), (9.95)

implies the velocity variation

(9.96)
9.6. GAUGE INVARIANCE AND LOCAL CONSERVATION LAWS 109

the last step exhibits both the explicit and the implicit dependences of 8t(ra,t)
on t. In computing the variation of <j>(ra, t ) } for example, we combine the po­
tential variation given in (9.85) with the effect of 8r a:

t) = - M ~ ~ Stv« * ~ ( 9 -97)
and, similarly,
A 1
<$A(ra(t), t) == —8t— A + <j)cV8t — 8tva •V aA = —6 t ~ A + <j>cV8t. (9.98)

The total effect of these variations on Ca is thus

SCa = - S t ^ C a + S(r - r 0(<)) (~ m avl - ^ -A •v a + ea<fj ^ S t + v a •VSt^ ,

(9.99)
or
SCa = - J t (StCa) - S(V - ra(t))E a ( ^ 6t + Va . V S tj , (9.100)

where we see the kinetic energy of the charged particle,

Ea = ^ mav2
a. (9.101)

We have retained the particle symbol d/dt to the last, but now, being firmly back
in the field, space-time viewpoint, it should be written as d/dt, referring to all
t dependence, with r being held fixed. The union of these various contributions
to the variation of the total Lagrange function is
p\ r\
¿A ot = — U tot-Q ^fo — S tot • V f t , (9.102)

where, from (9.90) and (9.100),

Utot = - ^ ( E 2 + B 2) + '5 2 s{r - v a(t))E a (9.103)

a

and
S tot= ¿ E x B + I ^ ¿ (r - r «(0 )^ a V ai (9.104)
a
are physically transparent forms for the total energy density and total energy
flux vector.
To focus on what is new in this development, we ignore boundary effects in
the stationary action principle, by setting the otherwise arbitrary 8t(r, t) equal
to zero at t\ and 12 . Then, through partial integration, we conclude that

SW12 = j f ‘ dt J (dr) St (Jj¡Utot + V •Stot ) = 0 , (9.105)

from which follows the local statement of total energy conservation,

¿ C /t o t + V. S to t = 0 , (9.106)

which generalizes (3.3).

110 CHAPTER 9. ACTION PRINCIPLE FOR ELECTRODYNAMICS

9.7 Problems for Chapter 9

1. Consider the Lagrangian of a particle in a given electromagnetic field,
(9.1):

( d~v \ 1 g
L ( r ,p ,v ,t ) = p* ( "¿i - v ) + 2 mv2 ~ e,K M ) + - v * A (r ,i).

(a) Reduce to Lagrangian form and then derive the equations of motion.
(b) Reduce to Hamiltonian form and then derive the equations of motion.
(c) Show the equivalence between a) and b).

2. Verify the transformation laws under rigid rotations for scalars and vec­
tors,
6S(r) = - ( ¿ w x r ) •V S (r),

SV(r) = - ( i w x r ) •V V (r ) + 6w X V (r),
by considering the transformations of the charge and current densities
(9.21).

3. Derive the behavior of a vector field V (r ) under infinitesimal rotations by

considering the example of the gradient of a scalar field S(r).

4. Fill in the details to derive (9.67).

5. Verify directly the local conservation law obeyed by i/tot and Stot, (9.106).

6. By considering <Sr(r,tf), analogous to 6t(r,t) in Section 9.6, in its effect on

Afield, derive the field momentum density and stress dyadic.
Chapter 10

Einsteinian Relativity

10.1 Relativistic Modification

After the discussion in the previous section one might well ask whether a unified
dynamics of charges and fields has now been attained. The answer is no— there
is still a major flaw. An electromagnetic pulse is a mechanical system that
travels at the speed of light, carrying a mass proportional to the total energy
content, as in (3.43),
E
m (10.1)

In contrast, the masses of the charged particles are fixed quantities that have
no reference to the particle’s state of motion and its associated energy. Another
way of expressing this lack of mechanical unity between fields and particles
comes from the physically evident expression for the total momentum density
(Problem 9.6)

Gtot = t^-E x B + ^ S ( r - r„(i))mava. (10.2)

The relation
G t ° t = ^2 S tot, ( 1 0 .3 )

which is valid for the field terms, does not hold for the particle contribution
[see (9.104)]. Could it be that Newtonian mechanics is mistaken, and that the
correct expressions for particle inertia and energy do satisfy m = E/c2? We now
follow this unifying suggestion— that the relation between inertia and energy,
which the electromagnetic field has disclosed, is, in fact, universally valid.
Consider a single particle in the absence of applied electromagnetic fields.
What we are proposing is that the connection between momentum and velocity
is actually

p = ;§v- (10-4)

ill
112 CHAPTER 10. EINSTEINIAN RELATIVITY

To this we add the relation of Hamiltonian mechanics,

dE
(10.5)
V _ 5p’
and deduce that
c2p •dp = E dE, ( 1 0 .6 )
which is integrated to
E 2 = c2p2 + constant. (10.7)
We already know (Section 3.4) that the added constant is zero for an electromag­
netic pulse, moving at the speed c. What is its value for an ordinary particle?
The energy (10.7) is smallest for p = 0, when the particle is at rest. Then we
write
p = 0: E = ra0c2, (10.8)

where mo is the mass appropriate to zero velocity— the rest mass. Therefore,
we have in general,

E 2 = p2c2 + m%c4, E = \Jp2c2 + m%c4. (10.9)

From (10.4), we have

p2c2 = E 2- j ( 10. 10)

implying that the energy and momentum are explicitly

„2
rriQC m0v
E (10.11)
— V 2/ c 2 ’ y j \ — V 2/ c 2

The last momentum construction exhibits the relation to, and the limitation
of, the initial Newtonian formulation of particle mechanics. For speeds small
in comparison to that of light, |v| <C c, the momentum is p = mov and the
particle inertia is constant. This is the domain of Newtonian mechanics. But
even here something is different, as we see from the energy derived from the
approximation
v ( 10. 12)
-<1
c
namely,
E « moc2 + im o v 2. (10.13)
¿i
In addition to the Newtonian kinetic energy ^mo^2 there is a constant, the
rest energy moc2, displacing the Newtonian origin of energy. The same thing
appears in the momentum form of E, (10.9), as

p < m 0c 2 :
9
moc 2 +
P‘
E (10.14)
2 mo
10.1. RELATIVISTIC MODIFICATION 113

For speeds approaching the speed of light we enter a new physical domain,
one where the speed of light is an impassable barrier. This we can see from the
particle velocity, exhibited as

p (10.15)
v= c
(p2 + m 2c 2 )V 2 ’

it is such that
v<c, (10.16)

with the equality sign occurring only for rao = 0. As for the last conclusion, it
is not unreasonable that a system, such as an electromagnetic pulse, which can
never be at rest, has no rest mass.
Now we must reconstruct the Lagrangian-Hamiltonian dynamics of particles.
If we omit the potential from the Lagrangian of (8.27), we have

1
L= P + 2 mv
2
(10.17)

Clearly, the Newtonian term \mv2 must be replaced by a new function of v,

L(v):

L=P*(Í _V)+I(V)’ (1018)

that will reproduce the new forms. We can find L (v) by using the velocity
construction of the energy,

E = p •v — L(v), (10.19)

and of the momentum. It is

m0v
o -s r
moc*
L(v) ( l _ * ; 2 / c2 ) l/2 (1 - „2/ c2 ) 1 / 2
= —rriQC ( 10 .20)

In the Newtonian regime, the L(v) reproduces the original form to within a
constant,
- < 1 : L(v) = —m 0c 2 + - m 0t;2 . (1 0 .2 1 )
c ¿

The consistency of the action principle is verified on noting the consequence of

a v variation:
9L(v) m0v
P= ( 10 .22)
<9v ^ \ - v 2/<?
The elimination of v produces the Hamiltonian version

L = H = c(p 2 + mgC2) 1/2 = E. (10.23)

114 CHAPTER 10. EINSTEINIAN RELATIVITY

10.2 Lorentz Transformations

We shall find it especially rewarding to use this new particle dynamics in re­
examining a subject previously discussed in the Newtonian framework. The
topic is the coordinate translation that grows linearly in time, (8.61), or equiv­
alently, the introduction of a new coordinate system with a constant relative
velocity. Here we consider only a single particle. The displacement

<$r(f) = 6 v t , (10.24)

combined with the momentum change

íp (t) = J Sv, (10.25)

induces the following alteration of the action element dt L :

6rjP[dt L] = <$r,p[p •dr—H dt] = p •5vdt-f ~-<Sv •dr—p *6vdt = Hd >

(10.26)
where we have used (10.6), or dH/dp = p c2/H. At the analogous Newtonian
stage, (8.63), m (= mo) appeared in place of i f / c 2, and we concluded that
the action was not invariant, but changed by a differential. Now a totally new
situation presents itself. If we also vary the time by

8t = ^ 6 v 'T , (10.27)

the additional contribution, —H d(6t)) will cancel (10.26), and the action is
invariant under the combined space and time transformations

6r = 6 v t, 6t = ^ 6 v r. (10.28)

In view of the invariance of the action, the implied conservation law should now
follow directly. Indeed

G = p.«5r — HSt = i v N , (10.29)

where ( H = E )
N = p (10.30)
c1
is conserved:
dN E dr
(10.31)
dt ^ c 2 dt
and we have recovered our starting point, m = E/c2.
But, from this initial dynamical modification of Newtonian dynamics has
now emerged a change in Newtonian kinematics: The absolute distinction be­
tween time and space has been removed. That is emphasized by the fact that
10.2. LORENTZ TRANSFORMATIONS 115

neither r 2 nor (ct ) 2 is left unchanged by the transformation (10.28), whereas

the difference r 2 — (ct)2 is invariant:

S[r2 - (ct)2] = 2[r •6\ t - 16v •r] = 0. (10.32)

The physical meaning o f this invariance appears on considering an electromag­

netic pulse, that, at time t = 0 , is emitted from the origin, r = 0 . Moving at
the speed o f light, c, the pulse, at time t ) will have travelled the distance r = ct,
so that
r 2 - cH2 = 0. (10.33)

The fact that an observer in uniform relative motion will assign different values
to the elapsed time, and to the distance traversed, but agree that (10.33) is still
valid, means that he also measures the speed of light as c. This is Einsteinian
relativity.
One might object that it could all be true only for infinitesimal transforma­
tions. But, from infinitesimal transformations, finite transformations grow. We
make this explicit by letting Sv point along the 2 -axis, so that

6x = 0, Sy = 0, 6z = — ct, Set = — z. (10.34)

c c

Let us regard this infinitesimal transformation as the result of changing a pa­

rameter 0 by the infinitesimal amount

(10.35)

The implied differential equations in the variable 0 are

dx(9) _ dy(9) _ d z(8 ) _ , d d ($ )

z{9). (10.36)
~dT - ° ’ ~dT - cm ’ ~dT

From the latter we derive

d2z(0) d2ct($)
= z(0), = ct(e), (10.37)
dd2 de2

which are solved by the hyperbolic functions, cosh 6 and sinh 9. The explicit
solutions o f these equations that obey the initial conditions

z( 0 ) = z, ct( 0 ) = ct,

% («)=<*, £ «> > = » (10.38)

are

z(0) = z cosh 0 ct sinh 0,

ct(0) = 2 sinh 0 -|- ct cosh 0 . (10.39)
116 CHAPTER 10. EINSTEINIAN RELATIVITY

Physical interpretation is facilitated by focusing on tanh0, the ratio of sinh#

and cosh 0, which cannot exceed unity in magnitude. We now write

ta n h 0 = ^ , (10.40)

which reduces to (10.35) for infinitesimal values of these parameters. Then, the
constructions

cosh 0 - ( ! _ v2 / c 2 ) l /2 ’ Slnh 6 ( ! _ vl/c 2)U 2 (10.41)

satisfy (10.40) as well as the hyperbolic relation

cosh2 0 —sinh2 0 = 1 . (10.42)

If we distinguish the transformed values of the coordinates by a prime, the

transformation equations read

(1 <* + ” <)’

1 ~ (1 - V2/C2)l/2 (* + C2Z) ’ (10.43)

along with
x' = x, y' = y. (10.44)
We see that the point with coordinates

(10.45)
o

-Si
II

II

II
1

is represented by the transformed coordinates

(10.46)
II

II

II
O

It is the origin of the new reference frame which therefore moves with velocity
—v relative to the initial one. See Fig. 10.1.
To see that r 2 —(ct ) 2 is left invariant by these finite transformations, it helps
to present (10.43) as
1/2
_ ( 1 + v /c\
z* + ct' > (z + ct),
1/2
_ fl-v / c \
z' - ct' 1 (z — ct), (10.47)
\1 + v/ c)
for it is immediately apparent, on multiplication, that

* '2 - (ct')2 = z 2 - ( c t ) 2 (10.48)

and then [(10.44)]

r - {ct')2 = r 2 - ( c t ) 2. (10.49)
10.2. LORENTZ TRANSFORMATIONS 117

Figure 10.1: The transformation (10.43), (10.44) carries us to a new coordinate

frame moving relative to the original one with a velocity —v along the z-axis.

The space-time transformations of the new kinematics are called Lorentz

transformations, although it was Albert Einstein (1879-1955) who, in 1905,
first understood their significance as describing the full physical equivalence of
reference frames in uniform relative motion. As an aspect of that equivalence,
we note the following. The original reference frame moves with velocity + v
relative to the new one (Fig. 10.2):

# = 0, y —0, z= 0 (10.50)
implies
x 1= 0, y( = 0 , z' — vtf. (10.51)
Then, should not the transformation that produces the unprimed coordinates
from the primed ones be of precisely the same form as (10.43), (10.44) but with
the sign of v reversed? Indeed it is, as is evident on rewriting (10.47) as
1/2
Í 1 — v/c
zA d
\1 + v/c

z — ct ( * '- 0 , (10.52)

together with x = x f, y = yf. More generally, suppose the coordinate transfor­

mation z ) t —* z'yt*, produced by relative velocity —v\ along the common'z-z*
axis, is followed by the transformation z'¡t* —►^//,t,/, produced by a relative
velocity —i/2 along the common z'-z" axis. Is the net result a transformation
z ,t zn¡i*1 that is produced by some relative velocity —v? Yes. It suffices to
consider just one of the pairs of equations analogous to (10.47), say

2' +c,,= ( t^ ) VÍ(" + cí)' (10'53)

and, similarly

*"+c‘" = ( r ^ ) 1/,(/+c,')' <I0'54)

118 CHAPTER 10. EINSTEINIAN RELATIVITY

Figure 10.2: Motion of original frame relative to the new frame.

(the other set emerges by the systematic substitution of —c for c) which imme­
diately yields

(10.55)

with

(10.56)

For any of the square root factors, the variation of the value of the appropriate
v/c from — 1 to - f l changes the square root from 0 to oo; it is a positive number,
and the product of two positive numbers is again a positive number. In other
words, no succession of Lorentz transformations can produce a net transforma­
tion with |v| > c. The specific value of v in (10.56) is identified by writing this
relation as
1 + v/c _ 1 + v\v2/c2 + (vi -f v2)/c
(10.57)
1 — v/c 1 + v\v2/c2 — (vi + v2)/c'

or
V\ + v2
(10.58)
1 + v\v2/c2 ‘

Simple addition of the velocities occurs only in the Newtonian regime, \v\¡2\<C c.

10.3 Transformation of Fields

We cannot end this chapter without showing that the kinematical space-time
transformations of (10.43), (10.44) do indeed produce a dynamical unification
of charged particles and electromagnetic fields. That requires a study of the
behavior o f fields and potentials under the infinitesimal Lorentz transformations.
That has already been touched on in Problem 1.6, but we prefer to apply our
recently developed methods here, beginning with the analogue of (9.81) for
10.3. TRANSFORMATION OF FIELDS 119

Lorentz transformations (10.28):

<5A(r, t) — - ^Svt •V + ^ S v •r A ( r , t) = - 0 CoorA(r, t). (10.59)

Differentiation supplies the model for potential variations,

<$A(r,¿) = - ¿ CoorA(r,¿) + ^5v^(r,¿),

S<¡>{v,t) = ~<5Coor^(r, 0 + A (r,¿), (10.60)

which are alternatively presented as (see Problem 4)

6A = xB + i« v r E + V Svt •A + -<$v •

S<¡> = Svt •E ■ - ( —Svt •A + -*#v •r<£ J . (10.61)

c at \ c )

The use o f the latter simplifies the calculation of the field variations; they emerge
as

SE = -S coorE - i t fv x B ,
c
SB = -¿ c o o r B + it fv x E . (10.62)
c
Then further differentiation in accordance with the field equations,
1
', = 4 i V ' E'

(10.63)

yields

Sp = SqooyP “b ^ V#j>
Sj = -¿coorj + Svp. (10.64)

Notice that the Lorentz transformation properties of j / c , p are the same as those
for A , <j>. If we introduce a new symbol ‘S’ that involves both changes of fields
(5) and of coordinates (<$Coor)> then its meaning as applied to any field F (r ,t) is

*S’F (r, t) = (F + SF)(r + fir, 1 4 - St) - F (r, t) = SF(r, t) + ScoorF (r , t). (10.65)

Note that the use of Scoor in (10.59) is consistent with this, since A(r,t) is a
scalar, ‘<$’A(r, t) = 0.
We must now examine the response of the various parts of the total Lagrange
function. First consider
¿ fie ld = ¿ ( £ 2 ~ B 2 ). ( 10 . 66 )
120 CHAPTER 10. EINSTEINIAN RELATIVITY

It is immediately apparent that the contributions of the last terms in the trans­
formation equations (10.62) just cancel and

^ A fie ld — " ~ ^ c o o r A fie ld - (10.67)

Then we consider the interaction contribution to the Lagrange function [see

(9.22)]
C¡nt = -p<t>+ i j •A. (10.68)
Again, the last terms in the transformation (10.60) and (10.64) have no net
effect, and
^£int = "“ ^coor^int* (10.69)
Finally, we come to the Lagrange function of the individual particles. For one
particle, with the Lagrangian description adopted (v = dr/dt)y the Lagrangian
( 1 0 .20 ) is
/ t;2 \ 1/2
La = - m 0ac2 ( 1 - J . (10.70)

In contrast with the procedure of (9.91), our introduction of the Lagrange func­
tion now is dictated by the impossibility of maintaining a common time for
particles at different spatial points— that Newtonian concept has disappeared
in the world of Einsteinian relativistic kinematics. Accordingly, we give each
particle its individual time coordinate, and present its contribution to the action
as
j dta La - J ( - m 0ac)[{c<Uay - ( d r a)^ = J dt(dr)Ca, (10.71)

where

£ 0( M ) = - m 0ac j 6(r - r a)S(t - ta)[(cdt a ) 2 - (dra)2]1^2. (10.72)

[Unlike (9.93), interaction terms are not included here.] The last integral is
extended over the trajectory of the particle, with r a varying as a function of
ta, or, better, with r a and t a given as functions of some parameter that is
not changed by Lorentz transformations. Apart from a sign change and the
considerations of infinitesimals, the space-time structure of the square root has
just the invariant form in (10.32). Then, in response to

6ra = Svtay Sta = - < $ v r a, (10.73)

c
the delta function product becomes

6(r — ra —Svta)6(t—ta —-^6v*ra) = <$(r - 6 v t - r a)6(t—~6v •r —ta), (10.74)

with the last form following from the delta function property, and the resulting
change is just
^coor[^(r ra)^(^ ” ^a)]* (10.75)
10.3. TRANSFORMATION OF FIELDS 121

We have now verified, for every individual constituent of £ tot, that

$£tot — """^coor-Ctot) (10.76)

so that the total variation in the sense of (10.65) is then zero. More generally
this can be expressed as follows: On introducing the transformed particle and
field variables associated with the transformed coordinates,

C‘ (r',t') = C (r ,t); (10.77)

the Lagrange function of the system of interacting charges and fields is invariant
under the Lorentz transformations of Einsteinian relativity.
The action
W i2 = ^ dt{dr)C (10.78)

is also Lorentz invariant because the space-time, four dimensional, element of

volume has that property. To prove this it suffices to examine the Jacobian
determinant of the transformation (10.39), for example,

cosh2 0 —sinh2 0 = 1. (10.79)

We shall not repeat the discussion of the various conservation laws in the light
of these relativistic modifications. It should be sufficiently evident that such ex­
pressions as the total energy density and energy flux vector, (9.103) and (9.104),
will be regained with Ea replaced by the relativistic energy ( 1 0 . 1 1 ). And, cer­
tainly (10.3) is now satisfied!
Finally, we supply the finite version of the Lorentz transformations for the
various fields. Proceeding in analogy with (10.36), we write (10.60), for a
Lorentz transformation along the z axis, as

dAx {6) _ n dAy(0) n dAz{6) _ ^ d m _

(10.80)
dd ’ dO ~ ’ dO ~ n ’ ’ dO
where d denotes the total change in the sense of (10.65). The finite transforma­
tion equations are then

r ',i ') = Ax (r,t),

A'yWyt') = A y (r,t),

K ir 'X ) = H *(r,<)) .

= ( 1 _ „ , ‘/ c ¥ / , ( « ' . < ) + > < M ) ) . (10.81)

The same forms apply with A , <¡>replaced by j /c , p. The transformations given

in (10.81) identify these quantities as four-vectors (c/. Problems 1.6 and 10.10).
The electric and magnetic field equations supplied by (10.62) are

dE,(0) . dBm
dO ’ dd
122 CHAPTER 10. EINSTEINIAN RELATIVITY

dEx(0) dB,
By(0),
dO - i r = «-W '
dEy{0) dBx {6)
= -E y {9 ), (10.82)
d$ dO
and therefore

E'z(r',t') = E ,(r ,t), B'z(r ',t ’ ) = B *(r,i),

W O = ( i _ ' ¿ 2) 1 / 2 Í ^ ¿) + ^ ( M ) ] ,

s y(r'> 0 = (i •
ti2/c2)i/2[g y(r ’ *) + (10-83)

and

Ey(r , t ) — ^ _ v 2j c 2y l 2 [Ey(v>t) c Bx (r,t)],

B 'ÁV>>*0 = (1 - t;2/ c 2) l /2 *) " ~ ^ y (r ’ *)]• (10-84)

The invariance, under finite transformations, of E 2 —B 2 is readily apparent, as

is the invariance property

E '(r; , t') •B '(r ', t') = E(r, i) •B (r, t). (10.85)

10.4 Problems for Chapter 10

1. The relativistic modification of the action principle involves the replace­
ment

in L (r,p , v, t) in Problem 9.1. Use the Lagrangian viewpoint and find the
equations of motion.

2. Repeat the above problem using the Hamiltonian viewpoint.

3. Find the explicit form of 9 in terms of v/c as a logarithm. How does the
composition law of velocities appear in terms of 01

4. Verify (10.61) and use the results to derive the statements o f (10.62).

5. Supply the proof o f (10.64)

6. Given the infinitesimal coordinate variations

10A. PROBLEMS FOR CHAPTER 10 123

what is
« (V ), « (£ )?

Do you recognize another four-dimensional vector?

7. Use these variations of derivatives, and the total variations of fields defined
in (10.65), to confirm that Maxwell’s equations are maintained under the
combined variations.

8. Show that the finite Lorentz transformations have the vectorial form [7 =
( l - i ^ / c 2) - 1/ 2]

72 1
r' = r + 7—------r w * r + yvt,
1 + 7 cJ

t' = 7 ( t + I v . r ) .

Check the invariance of r 2 — (ct)2.

9. Derive the expressions analogous to (9.103) and (9.104) for relativistic

particles.

10. As in Problem 1.6, let Greek indices like range over // = 0 ,1 ,2 ,3 and
let repeated indices be summed. We have identified (<^,A) as a four-
vector [see (10.81)], which we will denote by A**. Problem 10.6 identifies
another four-vector that we may call d^ defined by do = d/d(ct), d% = V¿,
i = 1, 2, 3. Also note that the invariant scalar product of two four-vectors
is
ApB** = A •B — A qB q.
Here, upper index quantities ( “contravariant vectors” ) are related to lower
index quantities ( “covariant vectors” ) by, for example,

A^ = ( - A o , A ) = ^ M „

where g is the metric tensor, having only diagonal elements, g00 = —1 ,

g0t = 0, gli = . In terms of the two four-vectors A^ and <9^, we may
define the field strength tensor by

F¡ív — dfiAu dj/Ap,

and its so-called dual *F„„ by

Here, €pPap is the four-dimensional Levi-Civita symbol with

¿0123 = -hi.
124 CHAPTER 10. EINSTEINIAN RELATIVITY

(a) Identify the components of FyU and * FyU with the fields E and B.
(b) Show that by performing a vector transformation on each vector in­
dex, as in (10.81), FyV transforms in a manner consistent with (10.83)
and (10.84).
(c) Write E 2 — B 2 and E •B in terms of Lorentz invariant combinations
of Fpj, and *FyV,
(d) Write Maxwell’s equations (1.65) in terms of F ^ , *FyU) dy and j** =

11. In covariant notation, the Lagrange function for the electromagnetic field
interacting with a prescribed current = (/>,j) is (we set c = 1 for
convenience)

C= T - dyA „) + + 8 ^ “ = 0.
4ir
The action is obtained from the Lagrange function by integrating over all
space and time, W = f(d x )£ , where (dx) = dt(dr). In £ the vector poten­
tial, A[¿, and the field strength tensor, F ^ , are regarded as independent
variables.

(a) What equations are obtained by requiring that W be stationary un­

der independent variations in Ay and FyVl
(b) Consider a coordinate displacement 6xy. A scalar field <j) changes
under such a displacement by 8<j){x) = <j>(x — Sx) — <j>{x) = —8xpdv<j).
Because £ is invariant under a gauge transformation, Ay —►A y+dyX,
while FyU —►Fypy where A is an arbitrary function of spacetime,
conclude that Ay must respond to a coordinate displacement by
8Ay, = —8x\dxAy — A xdy8x\.
(c) Now consider a source-free region, where jy = 0. Assume now that
Fyv = dyAv — dpAy. Show that 8W is zero under a rigid coordinate
displacement, 8x\ = constant, provided the fields vanish outside the
space-time region in question.
(d) Now suppose 8x\ ^ constant. Show that

6W = [ (dx)dy8xptfiU,

where t is the energy-momentum, or stress, tensor,

4**'“ ' = F » XF ¥a - i ¡ T F ^ F a0.

(e) Use the action principle to show that t^v is conserved, dyt^v = 0.
(f) Verify that t00 is the energy density, t 0t is the energy flux vector (or
the momentum density), and that t^ is the three-dimensional stress
dyadic discussed in Chapter 3.
(g) What is the trace of What is the significance of that result?
Chapter 11

Stationary Principles for

Electrostatics

An overview of electrodynamics— the mechanics of charged particles and elec­

tromagnetic fields in interaction— is now before us. In this, and in the following
several chapters, we specialize to more restricted situations— by considering
prescribed distributions of charges and currents, without inquiring how those
distributions were established and maintained; and by discussing only arrange­
ments in which there is no time dependence in charge and field quantities— a
static regime. We do this both because such situations are of physical inter­
est, and because of the opportunity offered to develop general mathematical
methods in simpler contexts.
Let us look back at the Lagrangian (9.13) and isolate those terms that make
reference to the electromagnetic field, including the interaction with charges.
With the further omission of the time derivative terms, what remains (apart
from a minus sign) is the energy of the field in the presence of prescribed charges
and currents: [see also the Hamiltonian (9.32)]

E — E e \e c “b -É^mag) (li.i)

with
Eelec = J (d r ) ¿ (e • + ± E 2) ( 11.2 )

and
= J (*) [-ij •A + ¿ (b •V x A - ÍB>) (11.3)

In the latter, j is restricted by

V - j = 0, (11.4)
as the time independent version of local charge conservation. The two energy
expressions are entirely independent. We will consider only the situation gov­
erned by ( 1 1 .2 ) in the next several chapters, and will defer consideration of
magnetostatics, governed by (11.3), until Chapter 26.

125
126 CHAPTER 11. STATIONARY PRINCIPLES FOR ELECTROSTATICS

11.1 Stationary Principles for the Energy

In this chapter we consider only the electrostatic energy (omitting the subscript)
1 „ 1
E = + — E . V<f> + — E 2 (11.5)
4ir 8w

Derived as it is, by a specialization of the action, this energy must also have
the property of being stationary for variations about the solutions of the field
equations. Indeed, independent variations of <^(r) and E (r) gives

SE = /( d r ) \p6<i> + -^-E •VS<f> + -j-6 E •(V<j> + E) , (11.6)

J L 47T A lt J
which may be simplified by use of the identity
E •V6<f> = V . (E 64) - ( V •E (11. 7)

J
to read
SE = (dr) 64> (/> - ^ ^ ) + ¿ Í E •(V<¿ + E) ( 11 .8)

where we have ignored the surface term since we assume the integral extends
over the entire region where the fields are nonzero. (In particular, we explic­
itly assume that (j) retains a zero value at infinity: S# = 0.) The stationary
requirement on the energy Ü7, SE = 0, now implies the two basic equations of
electrostatics,
V •E = 4ttp, E = - V < ¿ . (11.9)
Of course, the connection between E and <t> implies
V x E = 0. (11.10)
Having seen this, we can immediately modify the energy expression, (11.5),
to incorporate the effects of a dielectric medium,
eE -V<t> eE2
E = ( 1 1 .1 1 )
Air 8tt
The validity of this form is indicated by noting that if it is required to be
stationary under variations in (j) and D = eE, we recover the equations of
electrostatics in a dielectric. That is, the variation in the energy,

SE = J(dr) [ i ^ - i - V . D ) + — ¿ D . ( V ¿ + E) = o, (11.12)
47T
implies
V . D = 4ttp, E = -V <¿. (11.13)
(As before, in writing the variation in the form of (11.12), we have integrated
by parts and omitted the surface term.)
The discussion of action forms has accustomed us to the idea of restricting
the choice of variables by adopting one or more of the (field) equations as defini­
tions. Here, we can use two restrictive versions of the above energy functional,
(li.ii).
11.1. STATIONARY PRINCIPLES FOR THE ENERGY 127

11.1.1 The Scalar Field Form

We adopt, as the definition o f E, its construction in terms of the scalar potential,

E = -V < ¿, (11.14)

the curl of which is zero. The energy, (11.11), as a functional of the potential,

m =J (*) [p4>- ¿ eW)2

is then
' r 1 \2
(11.15)

The requirement that E be stationary under the variation of <j>yields Maxwell’s

equation
V •D = 47T/9, (11.16)
where D is defined by
D = -cV (¡). (11.17)
An infinitesimal variation in the scale of the function <£(r),

6<j>(v) = 6\<j>{r), (11.18)

for which
6E[4] = 6 \ J (d r ) [ p < ¿ - ¿ e E 2 = 0 , (11.19)

supplies information about E , the value of E[</>] for the correct field. It is

E = J (d r )^ e E \ ( 1 1 .2 0 )

which is indeed the total energy of the system, according to (7.15).

The energy functional, (11.15), contains yet further information. At the
stationary point <¡>o, E[<j>o] = E is an absolute maximum, as is seen by making
a finite variation in the potential, A<^,

E[4>0 + A<¿] = E - J.(dr) ^ ( V A ^ ) 2 < E, ( 1 1 .2 1 )

since the linear term in A<^ vanishes by the stationary principle. The correct
energy, given by the physical <^o, is a maximum of the functional (11.15). Eval­
uating the energy functional for an arbitrary potential bounds the energy from
below,
E > E[<j>]. ( 1 1 .2 2 )

11.1.2 The Vector Field Form

Now we use the vector D as the independent variable, but restrict its choice by
insisting on the validity of the field equation

V ♦D = 47rp. (11.23)
128 CHAPTER 11. STATIONARY PRINCIPLES FOR ELECTROSTATICS

Then, if we replace eE by D in the expression for the energy, ( 1 1 .1 1 ), and

integrate by parts on the D •V<^ term, we obtain

D2
E[ D] = j (dr) (11.24)
Sttc 5

which is a functional of D . How does the stationary principle work here? The
variation of (11.24) is
6E = /(d r )^ D .5 D , (11.25)

where the displacement vector is varied subject to the restriction that (11.23)
be satisfied:
V •(D + 5D) = 47rp, (11.26)

or
V •<$D = 0. (11.27)

Therefore, any variation in D must be a curl,

ÍD e V xíA (11.28)

where 8A is an arbitrary, infinitesimal vector, enabling us to write the variation

in the energy as
SE = J ( d r ) ^ - . V x 6 A . (11.29)

An integration by parts then implies the irrotational property of E,

V x E - 0. (11.30)

A second way of getting this result is based on the replacement of the local
restriction (11.27) by the equivalent integral statement

0 = J (d r)l^ < t> (r)V -6 D (r) = -J (d v )l^ 'V < l> .6 D , (11.31)

where <j>(r) is an arbitrary scalar function. Then the comparison of

SE[T>\ = / (dr) 1 - E . 6D = 0 (11.32)

J 4ir

with the restriction (11.31) yields the condition E = —V<^, which is equivalent
to (11.30).
The advantage to this functional form of the energy, (11.24), is seen by again
considering finite variations (called A D ). If Do is the correct physical solution,
then i£[Do] = E is an absolute minimum,

E [D 0 + A D ] (11.33)
87T6
11.2. FORCE ON DIELECTRICS 129

(since the linear term in A D is zero due to the stationary principle). Therefore,
the correct energy is the minimum value of (11.24) while an arbitrary D [com­
patible with (11.23)] will give an upper bound to E . These bounds, (11.21)
and (11.33),
E[4>\ < E < £[D ], (11.34)
are useful for finding approximate solutions when exact solutions are difficult or
impossible to obtain.

11.2 Force on Dielectrics

First, we present an essentially trivial example of the effect on the energy of a
finite change in dielectric constant. It requires that e be independent of position.
Let us consider the scalar field form (11.15) and write

= o. (11.35)

The consequence for the energy is

E = ^ E 0, (11.36)

where
E0 = J (d r ) |¿0/, — i - ( V ¿ 0) 2 (11.37)

will be recognized as the energy expression for e = 1, the vacuum. The same
energy relation follows immediately from (11.24), inasmuch as D retains its
meaning in order to satisfy (11.23). That the presence of a uniform dielectric
medium reduces the energy by a factor of e is familiar in elementary presenta­
tions of electrostatics as the corresponding reduction in the strength of the force
between charges.
Now we turn to a position-dependent dielectric constant, e(r), and examine
the effect of an infinitesimal alteration,

e(r) —> e(r) + <$e(r). (11.38)

This will change the fields, <j>(r) or D (r), but these changes have no first order
effect on the energy, which is stationary. Accordingly, the only contribution to
the energy change is produced by the explicit appearance of e(r) in the respective
energy expressions. Using (11.15), we get

SeE 2
6E{<t>] = - J ( d v ) 6
- ^ £ = -J (d v ) (11.39)
8 tt

Equivalently, from the second form, (11.24), the first order variation in the
energy is
8eE 2
6E[D] = - J ( d r ) ¿ ( y ) 6e = - j ( d r ) (11.40)
8 ir ’
130 CHAPTER 11. STATIONARY PRINCIPLES FOR ELECTROSTATICS

in agreement with (11.39).

An example is provided by the infinitesimal displacement of an uncharged
inhomogeneous dielectric. If the material is displaced by S r , the new dielectric
constant at r is the old dielectric constant at r — <Sr:

e(r) -+ e(r) -f <$e(r) = e(r - <$r) = e(r) - Sr •V e(r), (11.41)

or
<Se(r) = —Sr - V e(r). (11.42)
Then, the change in energy is

f E2
SE = S r- / (dr) ( V c ) — = - F •Sr. (11.43)

We therefore identify the force, F, on the dielectric due to the inhomogeneity

of the medium to be
F (11.44)

This result, (11.44), also shows the continued relevance of the electric field part
of the stress tensor, (7.15) with H = 0,

£^_£E E
(11.45)
8 7T 4ir ’
for spatially inhomogeneous dielectrics. Since the stress tensor (dyadic) de­
scribes the outward flow of momentum across a directed unit area, the total
force on a body bounded by a closed surface S, the net flow of momentum into
that body, is
F
Í dS •T = V •T, (11.46)

where the volume integral extends over the body. The divergence of the stress
tensor is

E2 1 V D e _(D .V )E
V •T = ( V e ) - ---- b — Di'VEi — (11.47)
8 tt 4it 4 ít 47T

Since for electrostatics,

DCVEi - ( D • V ) E = D x ( V x E ) = 0, (11.48)

and when no free charge is present,

V •D = 4wp = 0, (11.49)

the divergence reduces to

E2
v - T = <V e % ' (11.50)

so that the force calculated from (11.46) is identical with (11.44).

11.2. FORCE ON DIELECTRICS 131

F F

E 2 small E 2 large

Ve ^7

Figure 11.1: Dielectric slab immersed in an inhomogeneous electric field. The

forces F shown are those on each surface due to (11.46), (11.50).

As an application of the above result, consider a slab of dielectric material,

with e = constant, immersed in an inhomogeneous electric field. The gradient
of £ arises from the discontinuity of the dielectric constant between vacuum and
medium. The situation might be described by Fig. 11.1. If E 2 is small on the
left and large on the right, the dielectric material will be pulled to the right, that
is, into the region of strong field. [Note that this depends on the slab having
e > 1 so that the directions of V e are as shown in the figure.]
Incidentally, if there is a distribution of free charge in the body, an additional
term survives in (11.47), —pE, which leads to the expected force contribution

(11.51)

The same result emerges from the energy form (11.15) if we introduce the ana­
logue of (11.42) for the charge density,

(11.52)

In using the vector field form (11.24), we must recognize that the displacement
of the charge distribution requires a corresponding displacement of the field D,
in order to preserve (11.23):

/<*)
Then, the identity

—(<5r •V)D + <5rV. D = V x (irxD ) (11.54)

and use of both electrostatic field equations then gives the anticipated result
(11.51).
132 CHAPTER 11. STATIONARY PRINCIPLES FOR ELECTROSTATICS

e2 ) Vi ci, Vi

Figure 11.2: Boundary between two regions with different dielectric constants.

11.3 Boundary Conditions

An arrangement of objects with different values of e that are in contact with each
other provides the simplest example of an inhomogeneous dielectric constant.
Imagine we have two volumes, V\ and V2 , with dielectric constants c\ and €2 ,
respectively, sharing a common surface S', as shown in Fig. 11.2. Because of this
discontinuity of e on the surface, when the energy expression ( 1 1 .1 1 ) is varied
and expressed in the form ( 1 1 .1 2 ), there is an additional contribution from the
surface term previously omitted. In the interior of both V\ and V2 the same
arguments as those given in Section 11.1 still apply, so that we have the same
field equations,
V •D = Airp] V x E = 0. (11.55)
The additional surface term, which is now the total variation in the energy, is

ÍE = / y_ ( * ) V . ( ^ ) + / Vj( * ) V . ( § « )

= Is dS (ni * + n 2 * l w ^ 2) ’ (11.56)

where S is the common surface and n i and 112 are the oppositely directed
outward unit normal vectors for V\ and V2 , respectively:

ni = - n 2. (11.57)

At the atomic level, the transition between two macroscopically different

substances is reasonably smooth. And certainly the averaged microscopic field
11.3. BOUNDARY CONDITIONS 133

that is E (r) remains finite in that transition region. In the static regime now
under consideration, where E (r) = —V<^(r), this implies continuity of the scalar
potential across the contact surface, or

^ 1 = ^ 2, (11.58)

at a common point on the surface S. The variation of the energy, (11.56), at

the stationary point is therefore

6 E = [ dS-— [n2 •(D 2 — Ti\)}8<j> = 0 (11.59)

Js 4 tt

or, since 8<j) is arbitrary,

n 2 •(D 2 — D i) = 0. (11.60)
The normal component of D is continuous across the boundary between the two
media because of our implicit assumption that there is no free surface charge
density (cr = 0 ).
There is, however, the possibility of forming a distribution of free charge
along the contact surface S. Then the volume integral involving the charge
density p in ( 1 1 . 1 1 ) is supplemented by a surface integral referring to the sur­
face charge density a (which symbol is not likely to be confused with that for
conductivity in these static circumstances), namely

J dS ar<f>. (11.61)

Thus, there is a further term in the variation of the energy (11.11),

J dScrS<f>. (11.62)

The resulting generalization of (11.59) is

SE = J d 5 ^ [ n 2 . ( D 2 - D i ) + 47rir]^ = 0 ) (11.63)

so correspondingly the stationary energy principle now asserts that

n 2 •(D i - D 2) = 47rcr (11.64)

is the general boundary condition on the normal component of D. We may

regard (11.64) as the surface version of the volume statement

V . D = 4ttp. (11.65)

Likewise, there is a surface statement corresponding to V x E = 0 . This

follows from the continuity of <£, (11.58), which implies that of the tangential
derivative (that is, the component of the gradient parallel to the surface):

(n 2X V)(<^i — f c ) = 0 ( 11 . 66)
134 CHAPTER 11. STATIONARY PRINCIPLES FOR ELECTROSTATICS

or
n2x ( E i - E 2) = 0. (11.67)
Equation (11.67) states that the tangential component of the electric field is
continuous. This result can also be derived directly from the vector field form
by considering the surface terms that follow from (11.29).

11.4 Conductors
The surface of a conducting body provides an important example of these bound­
ary conditions. Within a conductor electric current flows in response to the
presence of an electric field. Since in the static situation there is no flow of
charge, E = 0 everywhere inside the conductor; the interior of a conductor is a
region of constant potential </>. The continuity of the tangential component of
E, (11.67), then implies
nxE = 0 (11.68)
on the surface of the conductor. Indeed, if it were otherwise, charge would
flow on the surface, which must also be a region of constant potential, of value
equal to that in the interior. Moreover, since E and D vanish inside, the other
boundary condition, (11.64), implies

n *D = 47rcr (11.69)

just outside the conductor, where n is the outward normal to the surface, and
again cr is the surface charge density.
We use these properties of the field to examine the electrical forces exerted
on the conducting surface, as described by the stress dyadic (11.45)

__ „ eE 2 eEE
T — 1 ----------------- (11.70)
8 tt 4tt ’
where e is the dielectric constant of the medium surrounding the conductor.
The force acting on a unit element of area, with its outwardly directed normal
vector n, is
eE 2 n •D
—n •T E, (11.71)
‘ ■ s r n + 47T
where
eE 2 = e(nxE)2 + i ( n - D ) 2 (11.72)
is a helpful decomposition into tangential and normal field components. We
now recognize that
—n •T = + <rE, (11.73)
which force has no tangential components; there is only a normal component of
force per unit area:

/ v ™ / x 2tt(t2 cr 27r
(—n) •T •( - n ) = --------------n •D = (11.74)
v ' e e
11.5. PROBLEMS FOR CHAPTER 11 135

The negative sign attached to this normal force per unit area tells us that it is
not a pressure but a traction or tension, drawing the surface element into the
region occupied by the electric field.
Here is a simple application of this result. Two identical conducting plates of
cross sectional area A and negligible thickness are placed in parallel proximity,
just short of total contact. Equal and opposite charges are bought up and
placed on the respective conductors, resulting in the final charges Q and —Q.
In view of the almost complete cancellation of the electric fields produced by the
opposite charges, essentially no work is performed during the act of charging up
the plates. Over most of the area of each plate, the surface charge density is
uniform, of magnitude
a — (11.75)
A'
Therefore, the total force on each plate, drawing it toward its partner, has the
strength
2tt
F = (11.76)
e
where, again, e is the dielectric constant of the medium in which the plates are
inserted (only the material between the plates is physically significant).
Now holding one plate fixed, move the other away to a distance a, which
distance is very small on the linear scale provided by A 1/2. The work required,
the energy of the resulting parallel plate configuration, is just F a , or

1 Q2
(11.77)
2 C ’

where
cA
C = (11.78)
4ira
is identified as the capacitance of the electric capacitor. We will discuss capac­
itance in general in Sec. 24.6.

11.5 Problems for Chapter 11

1. An uncharged pith ball is placed in the vicinity of a positive charge. Which
way will the pith ball move? What if the charge is negative? Get your
answer ( 1 ) by using the electrostatic stationary principle, and ( 2 ) by direct
physical argument.

2. Consider the boundary conditions between two different media with differ­
ent dielectric constants. Prove that the continuity of the potential implies
that the tangential derivative is also continuous, hence that the tangential
component of the electric field is continuous.

3. Derive (11.67) from the vector field form of the energy functional (11.24).
136 CHAPTER 11. STATIONARY PRINCIPLES FOR ELECTROSTATICS

+<r

e> 1 e= 1

Figure 11.3: Dielectric between parallel plates, as discussed in Problem 11.5.

4. Derive (11.44) from the force given in (4.42), namely

F = J (d r ) [ - ( V •P )E ].

(Hint: Integrate by parts and use V x E = 0.)

5. Consider two parallel nonconducting plates with uniform charge densities

-¡-a and —a. As shown in Fig. 11.3, half the region between the plates
is filled with a dielectric with dielectric constant e > 1 , and half is filled
with vacuum (c = 1). Ignoring fringing effects at the ends of the plates,
calculate the electric field E in both the dielectric and the vacuum. Is
there a force on the dielectric? Which way does it point? (Does it matter
if the dielectric extends in the region outside the plates?) What happens
if the plates are conductors, so the charge is free to redistribute itself on
them?

6. By applying a scale transformation <¡>{v) —►A0(r) in E[<j>], (11.15), derive

the stationary, scale-independent energy form

1 [J(dr)p<¡>}2
4f(dr)e(V<f>y/8n'
Verify that this is indeed stationary.

7. Recall that for homogeneous e, <¡> = <¡>q/c. Now use E[</)\ and E[T>] to
find bounds for the energy of a charge distribution p(r) in the dielectric
medium e(r) where e is slowly varying:
|Ve(r)| |V<ft(r)| |V¿„(r)|
e(r) <j>(r) ~ <^0(r)
Here <j>o is the solution to Poisson’s equation in the absence of the dielec­
tric,
~ v 2d>o = 4irp,
and the bounds should be expressed in terms of integrals involving e,
V<j)o, and Ve.
Chapter 12

Introduction to Green’s
Functions

The differential equation governing the scalar potential of a given charge distri­
bution in a dielectric medium, the combination of

V •D = 47T/?,
D = eE, E = -V <¿, (12.1)

is
- V . ( e V < ¿ ) = 47rp. (12.2)

In the special situation of the vacuum, e = 1 (or, effectively, for a spatially

homogeneous dielectric constant), this becomes

- V 2«5 = 4irp, (12.3)

the equation of Simeon Denis Poisson (1781-1840); in regions devoid of charge,

it reduces to the equation of Pierre Simon Laplace (1749-1827),

V 2(j) = 0. (12.4)

Our task is to find the solution of (12.2) for <^, for a given charge distribution.
Since (12.2) is a linear differential equation relating p and <^, the potential at a
point r is the additive contribution of all individual charge elements (dr,)/?(r/):

<j>{v) = y ( d r ') G ( r , r X r ') . (12.5)

It is evident that G { r, r') is the potential at r arising from a unit point charge
at r', so that it satisfies the differential equation

- V . [f(r)V G (r, r')] = 4irS(r - r'). (12.6)

137
138 CHAPTER 12. INTRODUCTION TO GREEN’S FUNCTIONS

Equation (12.5) expresses the fact that the potential due to a charge distribution
is simply the sum of the contributions of each of the charges. G (r, r') is an
example of a class of functions introduced by George Green (1793-1841). Once
we have Green’s function, the solution for any charge distribution in a given
dielectric arrangement is a matter of integration.

12.1 Reciprocity Relation

The most striking property of Green’s function is the reciprocity between the
position of the unit charge and the point at which the potential is evaluated.
This is the symmetry
G (r 1 r/) = G (r ',r ). (12.7)
A first derivation of this follows from a consideration of the energy of the system.
From the stationary property of the energy functional (11.15) we notice that an
arbitrary change in the charge density, 6p(r), produces the following change in
the energy
6E = j(dv)6p(v)<f>(ry, (12.8)

only the explicit dependence of E on p contributes. Of course, this energy

relation merely restates the significance of <^(r) as the interaction energy of a
unit point charge at r with the given charge distribution. But, its importance
for us here is in its emphasis that the potential function, and thereby Green’s
function, is uniquely determined by a knowledge of the energy for an arbitrary
charge distribution.
Now, from (11.19), (11.20) we recognize that the energy of the physical
configuration can be written in two alternative forms:

E = J ( d r ) e ^ = \j(dr)p<f>. (12.9)

[See also (11.24).] If we use the second of these expressions together with (12.5),

E = \ J (d r ) p(r)(j>(r)

= ¿ / ( drMdr 0 ^ O G ( r >r' M r')

= i / ( ^ ( ^ V W G i r '.r H r ') (12.10)

(where the last form has used r —» r'), we see that only the symmetrical part
of G (r ,r /) contributes to the energy. This proves (12.7) because the energy
completely determines Green’s function. In other words, the reciprocity relation
states that Green’s function is the interaction energy of a pair of unit charges at
specified points, and is a property of that pair of points, not of their individual
labels. [Note that (12.8) and (12.10) imply the construction (12.5).]
12 .2 . PROBLEMS FOR CHAPTER 12 139

A clumsier, but more conventional proof of the symmetry of Green’s function

uses the differential equation that it obeys. The equations satisfied by the
Green’s function due to point charges at r ’ and r", respectively, are

- V •[f(r)VG(r, r')j = 4ir6(r - r'), (12.11)

- V •[e(r)VG(r, r")] = 4irS(r - r"). (12.12)

We now multiply ( 1 2 . 1 1 ) by G (r ,r ") and (12.12) by G (r,r') and subtract the

resulting equations:

47r[6(r - r')G(r', r") - 8 (r - r")G(r", r')]

= V •[e(r){G(r,r')VG(r,r") - G(r,r")VG(r,r')}], (12.13)

in which we have also applied the following identity for three arbitrary scalar
functions,

<f>V •(AVVO - V’V . (AV<¿) = V . [A(<¿VV> - 1¡>V4>)], (12.14)

which is a slight generalization of what is known as Green’s second identity. On

integrating (12.13) over a volume that is bounded by a remote surface S, we
learn that

G (r', r") — G (r", r')

= h j s d S ' W r ^ G (r ’ r') V G (r, r") - G(r, r")V G (r , r ')}]. (12.15)

Let us stop for a moment and appreciate that this same volume integration,
applied to the differential equation ( 1 2 .6 ), gives

1 = ¿ i d s ' £ (r ){_ V G (r> r') } - (12-16)

If we suppose, for definiteness, that the remote surface S is a sphere of radius

R drawn in the vacuum (e = 1), it is clear that the magnitude of the surface
area, AirR?, is balanced by the magnitude of the electric field derived from the
potential l / i í , appropriate to unit charge. The same elements are present on
the right-hand side of (12.15), with the additional factors of G( r, r') or G( r, r"),
which are of order 1 /R. Accordingly, the surface integral in (12.15) certainly
vanishes in the limit R —►oo, and again we conclude that Green’s function
depends symmetrically upon its two variables.

12.2 Problems for Chapter 12

1. Derive (12.9) from the equations of electrostatics directly.

2. Verify the identity (12.14).

140 CHAPTER 12. INTRODUCTION TO GREEN’S FUNCTIONS

3. Show that if (j) is a solution to Poisson’s equation,

—V 20 = 47rp

and G is the corresponding Green’s function

- V 2G(r,r,) = 47r5(r-r')>

<j> is expressed in terms of the charge density and the boundary values by

- jf ^ •[G(r, r ') W ( r ') - W V'G(r, r')]

), r in V,
+ Jv (d r')G (v,v')p (v') = ^
r outside V,

where S is the closed surface bounding the volume V , and dS' points in
the direction of the inward normal. If <t> assumes specified values on 5,
and G vanishes there, then inside V

m = f (dr') G(r, r')p(r') + [V'G(r,

Chapter 13

Electrostatics in Free Space

The simplest electrostatic situation is for the vacuum, c = 1. The differential

equation for the potential is then Poisson’s equation,

- V 2^(r) = 47rp(r), (13.1)

so that the corresponding Green’s function equation is

- V2G(r, r') = 47T<5(r - r'). (13.2)

The solution to (13.2) is the well-known Coulomb potential, since it is the

potential at r produced by a unit point source at r':

G (r ' r' , = ¡T T 7 Í " (1 3 3 )

We will now derive (13.3) from the differential equation (13.2). In order
to do this, we require an integral representation for the delta function, the
charge density of a unit point charge. The relevant properties are that it vanish
everywhere except at a single point, while possessing a unit integrated value. In
one spatial dimension, these properties appear as

6(x — x') = 0 if x ^ x\ (1^-4)

while 8{x — x f) is so singular at x = a?' that

/
oo
dz6(x-z') = 1. (13.5)
-CO

No conventional function can satisfy these requirements. They can, however,

be realized in a limit. Consider, for example, the function
1 6
6e(x — x f) (e > 0). (13.6)
7T(x — x f) 2 4 - e2 ’

141
142 CHAPTER 13. ELECTROSTATICS IN FREE SPACE

First, the limiting values of the function for e —>■+ 0 are

lim 6e( x - x ' ) = { ° ’ X1ÍX>; (13.7)

e-++0 v [ CX), X = X , v 7

while the value of the integral of the function over the whole range of x is unity,
independent of the choice of e:

/
oo
i

-oo
d(a *')z
ir (x — a?')2 + e2 = 1,
•\L dt■_
OO t 2 + 1
(13.8)

which uses the substitution x — x' = et.

To arrive at a more useful form o f the function, we observe that
1 1 poo
' ) — Re—------------
6t( x - x c1 ‘ ' — = R e—
1 / dk eik(-x~x'+ie\ (13.9)
7T X — X' + 1C 7T J q

or
6<(x - x 1) = r 7T (13.10)
J —oo

Can we now set e equal to zero? No; the resulting integral would not exist—
it would not converge at infinity. But we can think of giving e an arbitrarily
small positive value, in order to make the integral mathematically meaningful,
without significantly altering thereby the physical properties being represented.
After all, when we speak of a point charge, we mean no more than one of size
that is very small on the scale set by all the other significant lengths in the given
physical situation. In the following we shall keep in mind the necessary presence
of the convergence factor, exp(—e|Ar|), but not write it explicitly. Accordingly,
the desired expression for the delta function, in one dimension, is

6 (x -x ') = ^ . e ik(-x- x'\ (13.11)

J —oo

The delta function in three dimensions is now realized by

<5(r — r') = 6(x — x /) 6(y — yf) 6(z — zf). (13.12)

Indeed, on integrating over a volume V,

J (d r ) 6(r — r') = J dx 6(x — x') J d y 6(y — yf) J d z 6(z — z/) ) (13.13)

the possible outcomes for the right-hand side are just zero and unity, the latter
being realized only if the range of ^-integration includes x', that of ¿ /-in te g r a tio n
includes t/, and the range of ^-integration includes z'— which is just to say that
the point r' is within the volume V. The use of the integral representation
(13.11) for the three individual delta functions, with the integration variables
labeled kX) ky) and kz, respectively, then produces the three-dimensional repre­
sentation
/ i (d
* k ) ¿fc . ( r - r ;)
6(r —r')
J (2 * :
)3 (13.14)
13.1. 2 + 1 DIMENSIONS 143

We now employ the above representation of the delta function to solve (13.2)
for Green’s function. This can be very simply accomplished if we note that
V e ,k .( r - r ') = i k e * k .( r - r ') ) ( 1 3 .15)

or, effectively,
V — *k, (13.16)
so that we can read off a solution to (13.2),

/
fdk'l e*k * (r~r‘)
(m7)
We can verify that this is in fact the known result, (13.3), by using spherical
coordinates for k, with r — r; pointing along the 2 -axis,

k •(r — r ;) = kRcos 0, i j = | r — r, |, (dk) = k2dk 27rd(cos0), (13.18)

so that
k2dk 27rd(cos 0) e ikRcos0
G (r ,r ') = 47t J
8 tt3 k2

-7TJof
1
/
,
ikR
1
dk----- AkR — e —ikR^

2 1 f°° sin#
7TR Jo x dx = R'
(13.19)

The convergence of the final integral here shows that only values of k of
order 1 /R are significant, which also applies to the individual components of k.
Accordingly, the implicit convergence factors will indeed effectively equal unity
if the individual value of e are restricted by e <C R- This is entirely consistent
with the physical context in which a point charge has a linear extent small in
comparison to the distance R.

13.1 2 + 1 Dimensions
In the above, we have treated all three directions of space on an equal footing.
However, we need not do this. We can separate out one direction, say that of 2 ,
and treat it differently. The reason this is useful is because there are geometries
in which physically interesting quantities vary in only a single direction. Singling
out the 2 direction, we can write Green’s function, (13.17), as

dkXdky ilknix-x'j + kjy- V)] [ dk*

G(r,r' (2 tt)3 J 2irk2+k2+k2 (13.20)

Adopting the nomenclature that the two-dimensional space of x and y is trans­

verse (i_) to the selected direction, we write the first part of (13.20) as

f (dkj.) ¿kx * (r-Q .

J (27T)2e
(13.21)
144 CHAPTER 13. ELECTROSTATICS IN FREE SPACE

Figure 13.1: Contour used to evaluate (13.22).

The remaining integration over kz (k‘j_ = k% + ky and kx > 0)

fOO dkz
(13.22)
J-oo 2* k± kz 2k± +

is performed by doing a contour integration as indicated in Fig. 13.1, which

indicates the simple poles occurring in the integrand of (13.22) at kz = diik±)
and with the contour being closed by infinitely remote semicircles which give no
contribution. We have therefore recast Green’s function into the form

G (r ,r ') = 4* J i ^ e ^ ' ^ ' ^ g i z ^ 1; ^ ) , (13.23)

where, for free space, the “reduced” Green’s function is

g{z,z'; (13.24)

The form (13.23) applies to any problem which is translationally invariant in x

and y but not necessarily in z. The representation is particularly adapted to
the situation in which the dielectric constant is a function only of z. For the
case at hand, where Green’s function satisfies (13.2), we can easily derive the
differential equation for g ( z ,z , \k±) as follows:

(d k -i) t-|Cx •( r - r ') x . i l ' g{z,z'-, k±)

- V 2G = 4 7T /
J ( 2 tt) 2 dz 2

= 4 tt [ •(r~r’ ^ 6(z - z') (13.25)

J (2^r)2
which implies
g(z,z'-,kx ) = 6( z - z ' ) . (13.26)

Of course, we could also have derived this equation directly from the left-hand
side of (13.22), which is only to say that the latter expression is an obvious so­
lution of the one-dimensional differential equation. Can one also arrive directly
13.1. 2 + 1 DIMENSIONS 145

at the form o f the solution given in (13.24) by solving the differential equation
(13.26)? Yes.
This differential equation is solved by noting that for z ^ z ',

2
kX f - g (z ,z ']k x ) = 0 , (13.27)
dz2
while the derivative of g at z = zf is discontinuous. Indeed, if we merely accept
that g is bounded when 2 is in the infinitesimal neighborhood of integration
of (13.26) over the interval between z 1 — 0 and z' -f 0 yields

d z'+0
-^ -g (z ,z ';k ±) =1. (13.28)
OZ z 1- 0

Then, the continuity of g at 2 = z' follows. To prove that last statement, we

multiply (13.26) by z, and present the result as

+ k\zg = z'S(z - z'), (13.29)

’ dz z Tzg
and integrate between z' — 0 and z' + 0. The result,
z’+ 0 z'+ 0

+ 0 = Z (13.30)
z'-O z'-O

taken together with (13.28), proves that g is continuous.

The two independent solutions of the differential equation (13.27) are given
by exp(=L&j_;z). The choice between them is dictated by the physical requirement
that g be bounded as z recedes to infinity in either direction:
( A (z,)e k±z, 2 > / ,
(13.31)
{ B (z')ek±z , z < z ',
where A(z') and B (z f) are independent of z. Imposing continuity of g at 2 = z\
as well discontinuity of the derivative there, according to (13.28), gives the
system of equations
(13.32)
-i¿
>4

o'
KÍ

H
H
1

II
1

A {z')kx e~k^z' + B (z’ )k± ek^z' = 1, (13.33)

which when solved yields
1 1
A— 1 pfc-L2' R — 1 r - k j . z '
(13.34)
A 2 k± ’ 2k±
which implies (13.24).
In summary, we have found two alternative expressions for Green’s function
in empty space:
1 ( rfkx ) ¿ k x -C r -r 'U 1 c k ± \ z -z'\
f (13.35)
k - r '| J (2x)2 2k±
We shall have more to say in Chapter 16 about the equivalence of these two
forms.
146 CHAPTER 13. ELECTROSTATICS IN FREE SPACE

13.2 Problems for Chapter 13

1. Consider
6v (x - x') = A e - ^ - ^ 3/ 2^

as another model of the delta function, in the limit rj —+ 0. Find the

normalization factor A and verify the counterpart of (13.7). Write Sr)(x —
x ') in a form analogous to (13.10).

2. Show that
6e(x) - - x - ^ 6e(x)

is also a model of the delta function. Can you support otherwise the
inference that
- x - ^ - 6(x) = S(x)?
ax
3. Express
6(x 2 — a2), a> 0
in terms of 6(x ± a). What can you say about S (f(x )), where / is a real
function with a simple zero at x = a?

4. Perform the integral in (13.8) and the last integral in (13.19) by contour
integration.
Chapter 14

Semi-Infinite Dielectric

14.1 Green’s Function for Charge Outside Di­

electric
We now apply the above representation, (13.23), to find Green’s function for the
simplest situation involving a nonuniform dielectric constant, that of a body of
uniform dielectric constant occupying the semi-infinite region z < 0, while the
region z > 0 is a vacuum— see Fig. 14.1. The inhomogeneity in e(r) is limited
to a discontinuity across the surface z = 0. This physical arrangement has the
required translational invariance in the transverse plane, that is, in x and y.
The Green’s function equation, (12.6),

V •[e(z)V G {r, r')] = 4*6(r - r'), (14.1)

becomes, in the two regions,

z > 0: - V 2G = 4ir<5(r-r'), (14.2)

z < 0: - e V 2G = 47r6 ( r - r /). (14.3)

To solve these equations, we must impose appropriate boundary conditions.

Across the interface between the vacuum and the medium, the normal compo-

Figure 14.1: Geometry of semi-infinite dielectric region.

147
148 CHAPTER 14. SEMI-INFINITE DIELECTRIC

nent of the displacement vector D must be continuous [see (11.60)],

(14.4)

Of course, G must be continuous across the interface because it represents the

potential o f a point charge:

G( r ,r ') = C?(r ,r ') (14.5)

z=+0 z= -Q

The reduced Green’s function, g(z, zf;k±), introduced in the 2 + 1 dimensional

representation (13.23),

G(r, r') = 4 * J ^ e ikx ' (r- r')x iK*> kj.), (14.6)

satisfies the differential equations [see (13.26)]

Z>0: (“ ¿3 = 6(Z ~ Z')’ (14.7)

(14.8)
Z<0: C(“ ^2 ^ _
subject to the boundary conditions

= 9\m
z =.
= + O’
(14.9)

(14.10)
€Tz9 Z——Q = d~z9 z—-\-0

In the following we will first solve this problem by assuming that z' > 0
(that is, the unit charge lies in the vacuum and not in the dielectric). [For
the converse situation see Problem 14.1 and (14.55).] The solutions to (14.7),
(14.8), (14.9), and (14.10) can be expressed in terms of the solutions, ek±z and
e~k±z, of the corresponding homogeneous equation. The forms of the solution
in the three regions are as follows:

z < 0: g = A ekxZ, (14.11)

z > z ': g = Be~k±z, (14.12)
0 < z < z ': g — C ekj-Z + De~k±z, (14.13)

where the single exponentials in (14.11) and (14.12) are required by the bound­
ary condition that g remain finite for \z\ —* oo. The boundary conditions at
z = 0, (14.9) and (14.10), require

a = c + d, (14.14)
ek± A = kx ( C - D ) , (14.15)
1 4 .1 . G R E E N ’S F U N C T I O N F O R C H A R G E O U T S I D E D I E L E C T R I C 149

from which we infer

C = ^ 4 , D = - ~
€ y -A. (14.16)

Just as in the situation mentioned at the end of the last chapter [see (13.28)], the
singularity in the differential equation requires, at z — z*, that g be continuous,
while —dg/dz be discontinuous,

*=*'+°
z= z'-0 U> (14.17)
z—z1-j-0
= i, (14.18)
z —z' —O

which imply, explicitly,

B e~ k-L2' - ('Cek±z' + D e-*-*-*') = 0, (14.19)

k ± B e~ k±z' + k± (C ek±z' - De~kj-Z') = 1 . (14.20)

The elimination of B and D between (14.19) and (14.20) immediately gives

C = — e —k ± z f (14.21)
2k±

and then, from (14.16), we obtain

1 -fcjy e — 1 1 0-k±z'
A = D = (14.22)
€ -J“ 1 2k± e + 1 2kj_

Finally, (14.19) supplies

B = —Í _.________ (14.23)
e + l 2k x. 2 JbjL

Inserting these coefficients back into (14.11), (14.12), and (14.13), we find for
the reduced Green’s function, g, in the two media,

1 -k ,\ z -z e- 1
z >0 : g = >-k ± (z+ z‘ )
(14.24)
2 k± L’ e+l'
n:
z < 0 q — -------------
1 2 e - k ± ( z ' —z)
y 2k± e + l

,-k ± \ z -z ‘ \_ £___ - e - k j . ( z ' - z )

(14.25)
2 kL e+ l

or, taken together,

z '> 0 : ^ , z '; ^ ) = (14.26)

Of course, if we set e = 1 , we recover the vacuum result, (13.24), as we must.

150 CHAPTER 14. SEMI-INFINITE DIELECTRIC

It is helpful to analyze the Green’s function we have found in terms of pri­

mary and secondary fields. The primary field results from the point charge at
z' and so is represented by the Green’s function (13.24). The secondary field is
due to the bound charge built up on the interface and is given by
e —1 1 e - k ± z ' e -k±\ z\
9s = - (14.27)
e + 1 2 k±,
The situation is illustrated by Fig. 14.2, for e > 1.

14.2 Derivation in Terms of Bound Charge

Let us derive this result (14.26) more directly. Recall that macroscopic electro­
statics begins by recognizing that the charge bound in atoms contributes the
polarization charge density

-V <fo) ~ 1
P efí — Pbound — — V • P (14.28)
•( Air
or
<fo) ~ 1 e (r )~ l
P efi — —V • D •V { « * ) - n (14.29)
47re(r) e(r) P -
V <r) )'
which introduces the average free charge density p. The problem of finding the
potential that is produced in a dielectric medium by the free charge density is
therefore equivalent to determining the potential that is produced in the vacuum
by the total charge density

P t o t ( r ) = />(r) + ^eff(r) = ¡ ^ ( r ) + ¿ e(r )E (r ) ' ' (14.30)

If e were completely uniform, the total charge density, and the potential it
produces, would be reduced below the vacuum values by the factor e, as already
given in (11.35).
In the situation before us, e(r) is discontinuous across the surface 2 = 0, so
that

dz7{r) = ( 1 _ l ) S^ ’ (14.31)
as one verifies by integrating over an interval that includes z = 0. Note that the
multiplier o f this derivative in (14.30), e(r)Ez(r), is continuous at z = 0 and can
be evaluated as Ez at 2 = -1-0, the vacuum side of the boundary. Accordingly,
the bound charge density has a surface charge component,

Abound = % M r x ) , (14.32)

where the surface charge density of bound charge is

<r(r±) = ^ ^ i Ez(rx ,z = +0). (14.33)

14.2. DERIVATION IN TERMS OF BOUND CHARGE 151

Figure 14.2: The primary field, P, and secondary field, S', contributing to
Green’s function (14.26).
152 CHAPTER 14. SEMI-INFINITE DIELECTRIC

Green’s function is the potential of a unit point charge,

p(r) = ¿(r - r'), (14.34)

and therefore the multiplier of p in (14.30) is l/e (r '), which is unity for the
present arrangement (z 1 > 0). Accordingly, G for z* > 0 is the potential
produced in the vacuum by the combination of the point and surface charge
distributions:

/» t o t (r ) = <5(r-r') + 5(2)<T(rx), (14.35)

1 €-
<r(rx) = G(r,r') (14.36)
47T e ¿=+0

That potential is given by

g (r ix )
G(r,r') (14.37)
+ J s dSl tr -r ix l’
where the element of area on the boundary surface S, which is the interface at
z\ = 0 , is
dSi = (dr1±). (14.38)
This construction of G in terms of itself (through a) is an integral equa­
tion. That we are able to solve it with ease is directly attributable to the
two-dimensional translational invariance embodied in the representation (14.6),
along with the one for l/|r — r7|, (13.35), which we now write as

r h\ 4*J = ^ e,kx'(r~ r')^ o ( * - * ,;*L),

« 7 o (* -* ';* x ) = ¿ e - fc- , W l (14.39)

Let us multiply the terms of (14.37) by exp(—¿kx t j l ) and integrate over rx-
We encounter the delta function

¿ ( k x - k ,x ) = y ’ ^ l e - ¿k-L- ^ e ikx * ^ , (14.40)

which produces (apart from a common factor of 47r)

e -* x • z' ; k±) = e_ikj- •r'±g0(z - kx )

+ 9o(z]kx ) J (á r ix )e_íkx *riJ-°'(r ix);(14.41)

where, in turn, from (14.36),

_d_
J (dr ix )e _ ik i *I‘1-Lcr(r ix ) = e_,kj- *r^- g(zi,z'-,kx )
dz\ z i= + 0
(14.42)
14.3. GREEN’S FUNCTION FOR CHARGE WITHIN DIELECTRIC 153

Thus, the one-dimensional reduction of the integral equation (14.37) is the al­
gebraic relation

g(z, z'\¿j.) = g0(z - z'; k±) + g 0(z; ( - g ~ ) s ( z i , z'; *±)

*i=+0
(14.43)
Now we have only to compute the negative 2 - derivative at z = + 0 ,

1 _ fc, zi 1 e —I f d
— e k±z + - , (14.44)
z' ,k ± ^z = + o 2 c z= + 0

to learn that
~ ^ g (z ,z ';k ±) ____ (14.45)
z= + 0 €+1

Therefore, the solution to (14.43) is

g(z, z'\fcx ) = 9o(z - z'; k±) - ^ -j-yg0(z; k±)e kxZ', (14.46)

which is just the result stated in (14.26).

14.3 Green’s Function for Charge Within Di­

electric
In the preceding discussion we chose to return to fundamentals [(14.28)] and
then encountered an integral equation [(14.37)] that was easily reduced to a one­
dimensional statement for the 2 -f 1 -dimensional geometry under consideration.
It is, then, not altogether surprising that, in the pursuit of the same outcome,
one can also avoid any reference to an integral equation. To this end, let us
consider the differential equation (14.1), but without specializing to the two
regions z > 0 and z < 0 , as in (14.2) and (14.3). The left-hand side of (14.1) is
also expressed by
-< W V > G - ^ £ > A 0 , (14.47)

and, after dividing by e(z), we can write

—V 3G(r,r') = i X S (t- 0 + (I¿y ) (-1 ) 0(ry>. (14.48)

Then, we have only to insert the representation (14.6) to obtain

( ~ & +kl) (¿ ¿ > ) ( - ¿ ) «(*■ ‘ J.

(14.49)
154 CHAPTER 14. SEMI-INFINITE DIELECTRIC

Now if we specialize to the homogeneous dielectric in contact with the vacuum,

this becomes [(14.31)]

(-¿2 + * i) *x) = 6(z -

(14.50)

The above equation describes the ^-dependence of the potential produced in the
vacuum by the combination of two localized charge distributions: one at z = z',
the other at z = 0. Expressed in terms of the vacuum function go, which obeys

(-^ 2 + fci) - z'\ kx) = S(z - z'), (14.51)

that potential is given by

g(z, z'; k±) = go(z - z1; kx)~j~pr + 9o(z; kx )~— - ( ) g ( z i , z'\k± )

e Kz ) ■ e \ ozi J 2i= + 0
(14.52)
For zf > 0, e(zf) = 1, this is (14.43).
Having left the choice of z' open in (14.52), we can now proceed to the
arrangement in which the unit point charge is embedded in the dielectric: z' < 0 .
With c(z') = e, we have

—^ 9 ( z ,z '- ,k x ) , (14.53)
z=+0

and therefore
(14.54)

The outcome is

z’ < 0 : g {z ,z ']k x ) = ^ 9o{z - z ';k x ) + ^ ^ j g o ( z - , k x ) e

— I 1 c -k ± \ z-z'\ , I f ____ * 1 _ c -k±\ z\ - k ± \z'\

~ e 2k± e c+ 12ix
(14.55)

It was really unnecessary to do the last calculation, however. This situation

in which the point charge is in the dielectric medium can be derived from the
one with the charge in the vacuum. First note that the use of vacuum, e = 1,
is a matter of convenience. What is physically significant is the ratio of the two
dielectric constants. To change all dielectric constants by a common constant
factor only modifies the potential— Green’s function— by the inverse of that
factor. [Again, (11.35) is an example.] Now, begin with Green’s function for
14.4. FULL GREEN’S FUNCTION AND IMAGE CHARGE 155

zf > 0 and replace e by 1/e. That converts the region containing the point
charge into the one with the larger dielectric constant, in the ratio e : 1. On
restoring the value unity to the smaller dielectric constant— raising the overall
scale by the factor e— Green’s function becomes divided by e. And finally, to
have the region of higher dielectric constant in the half-space z < 0 , one replaces
all ^-coordinates by their negatives. Doing this in (14.26) yields (14.55).

14.4 Full Green’s Function and Image Charge

A slightly different way of presenting these results leads to the full three-
dimensional Green’s function, G (r,r'). Let us return to (14.24) and (14.25),
referring to z‘ > 0 , and write

z > 0 : g{z,z'\kx) = ga{z - z'\kx) - \g 0( z - z 7), (14.56)

f T 1

z<0: g (z ,z ')k x ) = ~ ^ - j g 0(z - z';k± ), (14.57)

where we have introduced zf, the image point of z* y defined by

F = - * '. (14.58)

Because (14.56) and (14.57) involve only the vacuum form of the one-dimensional
function, we can interpret (14.56) and (14.57) as follows. For 2 > 0, the Green’s
function appears to describe the potential due to two point charges, one, of
strength unity, at r' = (#', y\ ¿'), and another, the image charge, of strength
— at the image point r' = (x'^y^z1). For z < 0, only one point charge
appears, of strength located at r'. In either medium, the total effective
charge is the same. With this interpretation, the full Green’s function may be
written down immediately, {z! > 0 )

1 _ € —1 1
z > 0: G( r ,r ') = (14.59)
|r —r'| e+ 1 |r - F| ’
2 1
z < 0: G(r,r') = (14.60)
e + 1 |r —r'| ’

The related forms for z' < 0 , as produced, for example, by the substitution
£ —*•1 /e followed by the procedure in the preceding paragraph, are

1 e - 1 1
z < 0 : G(r,r')=i 1 ■+ -• (14.61)
e |r —r'| e e -f 1 |r — r'|
2 1
z > 0: G(r,r') = (14.62)
e + 1 I r - r 'l ’

One sees directly, particularly on comparing the z < 0, z* > 0 and 2 > 0, z* < 0
forms, that the symmetry of Green’s function is realized in these results.
Applications of these Green’s functions will be treated in the next chapter.
156 CHAPTER 14. SEMI-INFINITE DIELECTRIC

14.5 Problems for Chapter 14

1. Consider the half-infinite dielectric region discussed in this chapter. Re­
peat the calculation for z' < 0 by the discontinuity method which led to
(14.24) and (14.25), and reproduce the result stated in (14.55). Give the
primary and secondary field picture. Check the symmetry of the Green’s
function (both G and g) for all z and z1.

2. A homogeneous dielectric medium, with dielectric constant e, occupies

not only the region 2 < 0 as discussed in this chapter, but also the region
z > a > 0. A unit point charge is stationed in the vacuum between the
dielectrics, a > z* > 0. Use any method to find the function g in the three
regions. What happens as a —» oo, and as a —» 0?

3. Now suppose a dielectric slab occupies the region 0 < z < a. Assume the
unit point charge is in the vacuum region z 1 > a. Compute the reduced
Green’s function in this case. Can you derive from this result additional
information about the arrangement of Problem 14.2?
Chapter 15

Application of Green’s
Function

15.1 Force between Charge and Dielectric

Armed with Green’s function for the semi-infinite dielectric arrangement, we
can now answer mechanical questions about interaction energy and force. The
total energy of a charge distribution in the presence of the dielectric medium,
( 12. 10),

E=Z\ J (dr)(dv>) />(r)G(r>rX r')> (151)

includes the mutual interactions of the charges. We are not interested in this
but rather in the change of the energy due to the introduction of the dielectric.
To calculate this change, we let Go be Green’s function in vacuum,

Co(r-r') = ¡ ^ , (15-2)

while G is Green’s function in the presence of the dielectric, as found above.

Therefore, the interaction energy between the charge distribution and the di­
electric is
#int = ^ J(dr)(dr')p(r)[G(r, r') - <?0(r, r ') M r')- (15.3)

Evaluating this for a point charge at position ro, with z q > 0,

p(v) = e<S(r - r0), (15-4)

and making use of (14.59), we find the energy of interaction to be

= y[G(r,r') - G„(r,r')]r,r-=r„ = (15-5)

Is this a physically meaningful result? If e > 1, E < 0 so that there is a force of

attraction pulling the dielectric toward the charge and into the region of higher

157
158 CHAPTER 15. APPLICATION OF GREEN’S FUNCTION

fields, in agreement with the earlier discussion o f Section 11.2. The magnitude
of this force is
1 e-1
9Eint _ e - 1 e2
\p\
1 t + i e \e \
(15.6)
d ( - z 0) ~ € + 1 4zfi ( 2z 0)2
which can be interpreted as the force between the charge and the image charge.
The field point of view, as opposed to that of action-at-a-distance, provides
an alternate derivation of this result. To calculate the force on the dielectric,
we calculate the normal component of the flow of momentum into the dielectric.
In terms of the stress tensor, this force is

F —— (15.7)

where the integration is over a surface just outside the dielectric, at 2 = + 0 .

Correspondingly, we use the vacuum form of TZZ) (3.14),

Tzz=h{E2 ~2E2)=h{E2±~e2)■ (15-8)

Since Green’s function is the potential of a unit point charge, the electric field
is
r - r0 1 r - r0
E (r) = —V e G (r ,r 0) = e (15.9)
|r — r 0|3 e + l | r —ró|3.
according to the Green’s function form for z > 0 in (14.59). Then, on writing

/ > = | ( r - r 0)x|, (15.10)

the field components on the surface are found to be

2 (r - r0)x
z = +0 : Ex — e
C + l(p* + * 2 ) 3 / 2 ’
2e z0
Em —e (15.11)
6 + l ( p 2 +*2)3/2»

giving us this expression for the force:

e2 / o \2 2

(ih )c
€ Zi
27tp dp
87r (,p2 + 4 ) 3
■- 1 - ( t - 1 ) _ c - i
(15.12)
(* + 1 ) 3 e + 1 4z% ’

where we have introduced the variable t = p2/zq, in agreement with (15.6).

But let us not hurry on without examining, in the context of this simple
example, the connection between these two modes of computation. We know
that, in regions of homogeneous dielectric constant, the stress dyadic (11.45)
obeys the electrostatic specialization of (7.13),

V •T + p E = 0 . (15.13)
15.1. FORCE BETWEEN CHARGE AND DIELECTRIC 159

Now integrate the ^-component of this equation over the entire volume of the
semi-infinite region z > 0 to get (infinitely remote surfaces do not contribute):

+0) = - J'(dr) pEz e (15.14)

d~zeG{ r ’ r°)
The electric field required here is (15.9). The force on the dielectric body is now
found as the negative of that acting on the charge e, as produced by the latter’s
own Coulomb field (which force we know to be zero), and by the Coulomb field
of the image charge. This derivation is in the spirit of the first such calculation,
but lacks the reference to the energy.
As far as the field on the surface z = +0 is concerned, the construction given
in (15.9) could as well apply everywhere. If we accept this fiction, the volume
integral of the ¿-component of (15.13), extended over the semi-infinite region
z < + 0 , gives

F —— J dSTzz(z — + 0 ) — J (dr) pimageEz — —t

£+ 1
-^ (r .r o )
(15.15)
comprising the zero self-force of the image charge, and the force exerted on the
image charge by the charge e. Quite correct.
But suppose we want to use the physical field produced in the dielectric
region by the point charge in the vacuum. First we move across the boundary
by writing

F=~j dS[Tzz(z = + 0 ) - Tzz(z = - 0 )] - J dSTzz(z = - 0 ). (15.16)

Then, integration of the ¿-component of (15.13) over the region z < —0 relates
the last integral above to the force on the free charge within the dielectric.
There isn’t any! That effectively reduces the calculation to the one already
carried out in (15.12) or (15.14). Nevertheless, let us evaluate F by means of
the first part of (15.16), involving the discontinuity of Tzz across the interface
between dielectric and vacuum:

F=~J dS[Tzz(z = + 0) - T zz(z = - 0)]. (15.17)

Written in terms of the electric field in the vacuum (z = +0), we have

Tzz(z = - 0 ) = ± [ e E l - (15.18)

and then

Tzz(z = +0) - Tzz(z = - 0 ) = ~ ( e - l ) [ E l + -l E l } . (15.19)

Incidentally, the same expression for the force emerges from (11.44), with its
¿-component presented as
160 CHAPTER 15. APPLICATION OF GREEN’S FUNCTION

which involves the continuous transverse electric field and normal component
of D . With these field components given their z = +0 values, the resulting
^-integration over the discontinuity region yields

F = J dsL(e - l ) [ E l + -l E % (15.21)

in agreement with (15.17), (15.19). The insertion of the fields displayed in

(15.11) then produces

F dp p 2tt €Zo + P
( P 2 + Zo
2) 3
€- l e2 f ° ° € + 1 + (t — 1 )
dt (15.22)
(e + 1 ) 2 2z l J 0 (t + 1 ) 3 ’

which is indeed equal to (15.12), in consequence of the null value of the integral
appearing in both (15.12) and (15.22),

i J í( í T ¡ j 3 = 0i <15'23>

as can be seen from the substitution t —* l/t. This property just expresses the
zero value possessed by the surface integral of Tzz(z = —0).
The version given in (15.17) directs attention to the physical interpretation
of F as the force on the bound charge that is localized on the interface between
the two regions. As such, the force should also equal

F = J dS<r Ez(z = 0). (15.24)

But what is the value of Ez(z = 0)? One way to answer that question has
us returning to (14.31), which describes how a surface distribution of charge
emerges in the limit where e(z) is discontinuous. If we regress to the continuous
description, through the replacement [the other factors in <r, (14.33), and the
reference to a particular point on the surface, are understood]

0 -0 e ^/ > (¿¿i) &(*>' <i5-25>

and use the continuity of Dz, we find this integral to be

1 1
Dz (z = 0) = Dz{z = 0). (15.26)
< *) 2
That identifies Ez(z = 0) with

Ez(z = 0) = i ( l + I ) Dz(z = 0) = l [ E z(z = +0) + Ez(z = -0 )]. (15.27)

15.2. INFINITE CONDUCTING PLATE 161

According to the form given for Ez(z = +0) in (15.11), this is

z0
E ,(z = 0) = (15.28)
V + * o 2)3/2’
which is just the field of the point charge. But we could have anticipated that;
the self-force of the surface charge distribution is zero. The above field is now
combined with (14.33), (15.11),

e c—1 zp
(15.29)
2 tt e + 1 ( p 2 + Z q ) 3/ 2

to produce

f e2 e - 1 f°° z2
zo 6 -1
F = j d S a E 2(z = 0) = — — J^ d pp 2*
(p2 + *o ) 3 6 + 1 ( 2 Z q) 2
(15.30)
Six derivations of the same quantity? Time to move on. Yes, but there is one
little question about the surface charge distribution that has not been answered.
What is the total amount of that charge? Obviously, one can compute it by
integration— and so we shall, for a particular example. But a more immediate
answer comes from Green’s function (z' > 0 ) as displayed in (14.59), when this is
regarded as the potential produced in the vacuum by the unit point charge and
the surface charge. At very large distances, on the scale set by z1, so that the
finite dimensions of the surface charge distribution are negligible, the behavior
of G in both half-spaces is simply

G( r ,r ') .
0 - 7TÍ)? (15.31)

the total surface charge equals the image charge.

15.2 Infinite Conducting Plate

The apparently unphysical process of letting c become arbitrarily large, in fact,
produces a new physical situation, one in which the dielectric body acts like
a conductor. Indeed, we see from (14.60) that G(z! > 0) then vanishes in the
region z < 0; the latter has become a region of constant— zero— potential. In
the limit that 6 —►oo, (14.59) and (14.60) become

z > 0,
G (r ,r ') = | ^ Q \*~*V (15.32)
2 < 0,

which is obviously Green’s function for a grounded conductor. For z > 0,

Green’s function can be interpreted as the potential of a unit point charge at
r' = ( 0 , 0 , z') and an image charge of strength —1 at r' = ( 0 , 0 , —zf), as shown
in Fig. 15.1. The image charge is exactly opposite to the charge placed in the
162 CHAPTER 15. APPLICATION OF GREEN’S FUNCTION

vacuum

+1

Figure 15.1: Image charge for a grounded conducting plane.

vacuum, and G vanishes on the surface, all of its points being at the same
distance from the equal and opposite charges.
For such a unit point charge, we now calculate the free surface charge density,
cr, induced on the conductor. We know, from (11.69), that

4*-<r(rx) = #,(r)|,=+0 = - ¿ G(r»r0 (15.33)

which is just the e —+ oo limit of what is displayed in (14.36). Alternative and

equivalent forms for Green’s function are, for z > 0, [recall (14.39)]

G(r,r') =
r -r r |r —r;

i
(dkx ) ,*k-L •( r - r 1
( 2 ’r) 2 '
(15.34)
This is the e —» oo limit of (14.26). The outcome for the form of a computed
from the first form can be read oif from (15.29) with e —» oo, e = 1, and Zq = zf:
1 2z*
(15.35)
4w (p2 + z /2)3/2 *
The second form gives

f (<lkx)
J (2*)2
ikj. . f r - r 'l | r - k j . z ' (15.36)

which is also available from the e oo limit of (14.45). The equivalence of

(15.35) and (15.36) will be exploited in the next chapter.
For now, let us check that in both cases the total induced charge on the
surface of the conductor is —1, the strength of the image charge. The first form,
(15.35) , yields for the total charge

g,nd = j d s * = j^pip (-¿ ) ^ +*'„)3/>

1 f°° 1
(15.37)
- ~ 2 j 0 d\ t + 1) 3/3 - - 1*
15.3. PROBLEMS FOR CHAPTER 15 163

In using the second form, (15.36), one encounters the two-dimensional delta
function [dS = (drj.)], as already noted in (14.40),

/ *(r-r'U _ ( 15.38)
J y

That gives
Qind = ~ J (dk± )5(kx ) e - k^ ' = - 1 . (15.39)

In another exercise of this kind we give two derivations of the force pulling the
conductor toward the unit point charge, using the force per unit area presented
in ( 1 1 .7 4 ), but with e = 1 and omitting the minus sign:

F =
J dS 27TCT2. (15.40)

We need not trouble with the calculation that employs (15.35) for cr; it is just
what is exhibited in (15.30), with e —*•oo, e = 1, and z 0 replaced by z':

1
F = (15.41)
( 2z' ) 2

The calculation that uses (15.36) begins with

F = 2ir J (drj.) J e ¿ (k ~ k 'U * , (15.42)

in which it is convenient to write —kj_ as the integration variable for the second
cr. Performing the rj_ integration with the aid of (15.38) we get the same result,

J? f (tflcx) (tflc'x) / o _ \ 2 _
C /1 . k/ V
F ~ 2 J (2ir)2 (2tt)2 ^ 5^kj-

- / =j f = (¿F- <15-43>

15.3 Problems for Chapter 15

1. Look at the answer to Problem 14.2 in the limit e —►oo and bring it into
the form, for z > z ', 0 < (z, zf) < a,

sinh k(a — z) sinh kz 1

g — ---- -— -— ------ .
k sinh ka
What is it for z < z,el Show directly that this satisfies the correct differ­
ential equation with the proper boundary conditions.

2. Prove the identity (15.23) and show that it expresses the fact that the last
integral in (15.16) vanishes.
Chapter 16

Bessel Functions

Useful mathematical identities can be obtained if we solve a physical problem

by using different representations for Green’s function. In particular, through
the consideration of situations where physical quantities vary only in a single
direction, we learn of the properties of the important class of functions called
Bessel functions. An illustration of this was encountered in the last chapter,
where we obtained two forms for the surface charge density, (15.35) and (15.36).
If we let p = |(r — r')x| and introduce polar coordinates for k x, that is, the
magnitude k = |kx|, and the angle 0 , which may be taken to be the angle
between kx and (r — r ')x , we have for the exponent in (15.36)

kx •(r —r ')x = kpzosfj), (16*1)

and for the element of integration

(dkx) = kdkd<j>. (16.2)

Accordingly, the identity may be written as (for 2 > 0)

2z f k dk d<j> ikpcQB¿ kz
cP2 + z 2)3/2 J ( 2 tt) 2
(16.3)

The (j) integral in (16.3) is defined as the Bessel function of zeroth order, Jo,

^ ikpcam* = Mkp)\ (16.4)

it is the first member of the class of functions that are named for Friedrich
Wilhelm Bessel (1784-1846), although the infinite series that represents these
functions had appeared more than half a century earlier, in a work by Leonard
Euler (1707-1783). Now expressed in terms of this function, the equivalence
(16.3) appears as
z r°°
(p2 T z 2)3'/~2 = Jo k d k M k p ) e ~ k\ Z > 0. (16.5)

165
166 CHAPTER 16. BESSEL FUNCTIONS

Since this result was obtained by equating 2 derivatives [c/. (15.33)], it may be
immediately integrated to yield
1 r°°
■. / dk J0( k p )e -kM. (16.6)
Jo V ' '
(The integration constant vanishes since both sides go to zero as \z\ —> 00 .)
Note that (16.6) may be directly derived from the equality of the two forms of
the Coulomb potential in (13.35), which with r' = 0 reads

i = 4tt / ' rx d -c-*l*l. (16.7)

r J ( 2 tt) 2 2k v '

Here in (16.6) we have recognized two different forms for the potential of a
unit point charge at the origin, which satisfies Laplace’s equation,

V 2(j> = 0 , (16.8)

except at the origin. Actually, the integrand of the right hand side of (16.6) is
also a solution of Laplace’s equation, that is, for each positive value of k,

V 2 [J o (M e ~ * W] = 0, (16.9)

provided z / 0. This is most easily proved by returning to (16.7) and noting

that

z± 0 : V 2 ^eikj- ' = ( - k i + fc2)eik-L , r ± e ~ kW = 0. (16.10)

Integration over the direction angle of k ¿ according to (16.4) then produces

(16.9).
We can convert (16.9) into the differential equation that J 0 obeys by writing
V 2, the Laplacian differential operator, in terms of circular cylindrical coordi­
nates. Let us first recall that in using orthogonal curvilinear coordinates, ui, t¿2 ,
U3 , such that the distance represented by the infinitesimal coordinate change
duk (A; = 1,2,3) is hk(u)duk) the element of volume becomes

h\du\ h,2du2 h^du^ = h du\ du2 du^) (16.11)

while the element of area perpendicular to the u\ direction, for example, is of

magnitude
Ji2du2 hzdu%. (16.12)
Then, the translation into curvilinear coordinates of the divergence theorem,

j ( d r ) V 2<f>(r) = j dS.V<j>(v), (16.13)

is as indicated by
 167

Rewriting the latter surface integral in terms of a curvilinear coordinate volume

integral (with due attention to the sense of the normals) yields the desired
Laplacian form (see Problem 16.1)

1 3 d
(16.15)
duk
fc=i

In the example of circular cylindrical coordinates, where infinitesimal dis­

tances in the three orthogonal directions are produced by dp, pd<j>, dz, and
h — p, the Laplacian is

\d_ f d_\ £_
(16.16)
p dp v dp) p2 d<j>2 dz 2 ’

leading to the reduced form for (16.9),

'1 d_
+ k 2 J0(kp) = 0, (16.17)
.P dp

or, with t — kp,

1 d
+ 1 Jo(t) = Jo(t) = 0 . (16.18)
t dt dt2 t dt~^

Inserting here the integral representation of the Bessel function, (16.4),

J0(t) = j * * ^Leitcos*, (16.19)

and multiplying by t, we have

f 2* 7T~\~t c°s 2 ^ * cos <j>+ ile*tcos^ = / 2,r^ ^ - [ i s i n ^ e itcos^] = 0 , (16.20)

Jo 27T J q 27r d<f>

thereby again proving (16.9), and establishing the equation satisfied by the
Bessel function of order zero, (16.18).
To exploit the more general solution of Laplace’s equation exhibited in
(16.10), we take

—— g i k p C O S ( f > __ exp iii(e^ + e-^)l , (16.21)

introduce the symbol u = ie1^} and then encounter the generating function for
the Bessel functions of integer order,
oo
(16.22)
m = —oo
168 CHAPTER 16. BESSEL FUNCTIONS

which serves as a definition of all the Bessel functions of integer order m. Since
the generating function is invariant under the substitution

1
u (16.23)
u’
that is,

E umJm( t ) = ]T (-1 Y (-1 (16.24)

we learn that Bessel functions of positive and negative integer orders are related
by
= (-1 (16.25)
Now let us regard the generating function as the product of two exponentials,
with arguments proportional to u and —1/w, respectively. The insertion of the
appropriate infinite power series for these exponentials,

E
( i t ) m+n um+n
(m + n)! ■E n\
(16.26)

which summations extend over all non-negative integer values of m + n and n,

then supplies the infinite power series representation for Jm,

( i * ) m+2n
Jm(t) = E C " 1)” (m + n)! n! ’ (16.27)
n—0

that (to within a constant factor) Euler first encountered in 1764.

By writing (16.22) in terms of 0,
oo
eitcos4 _ r e i,n*Jm(i), (16.28)

and using the orthogonality condition

J0
[
21T
<l> im<}> _ ^
-« m m ',
(16.29)

we obtain an integral representation for Jm,

Am j (4\ — f d<LJ ( t cos <f>-m<j>) (16.30)

m[)~J0 2 ?re

which contains the result for Jo, (16.4). Incidentally, changing the sign of m,
and replacing <f> by 27t — <j>, leaves the integral in (16.30) intact, or

= imJm(t), (16.31)
 169

which is the content of (16.25).

Since (16.30) is the Fourier coefficient of etkpcos^y we see from (16.10) that
another set of solutions to Laplace’s equation is

eim+Jm(kp)e~kW. (16.32)

The Laplacian, (16.16), acting on this solution yields the differential equation
satisfied by Jm,

'\d_ ( d_
J<m{kp) — 0 , (16.33)
p dp \ dp

or
d2 Id m2
Jm{t) — 0 . (16.34)
H? + t l t ~ 7 r +

One can, of course, verify that Jm(t)y as represented by the integral (16.30),
obeys this differential equation. See Problems 16.4 and 16.5.
Another way to present the differential equation uses the differential operator
relation (an operand is understood)

t- i l 2 ± tH 2 - 1 + 1 (16.35)
dt dt 2t ’

and its square

(16, 6)

The latter leads to the following differential equation for f^ J m ^ ) :

~ + l) = 0. (16.37)

This version has the advantage of suggesting more clearly the behavior of the
Bessel functions at extreme values of t. For sufficiently small values of t y such
that unity is negligible in comparison with (m 2 — 1/4) f t 2y the differential equa­
tion indicates that ~ ¿lml+ i/2? in accordance with the leading term of
(16.27),
( k ) m
m> 0 : Jm{t) = + (16.38)

And, at sufficiently large values of t , such that (m 2 - 1 /4 )/t2 is negligible com­

pared to unity, t l/2Jm(t) should be a trigonometric function of t. Indeed, the
leading asymptotic term for Jm(t) is

(^ ) cos + 1/ 2 ) 0 . (16.39)
170 CHAPTER 16. BESSEL FUNCTIONS

16.1 Delta Functions and Completeness

We have spoken of Coulomb’s potential as a solution of Laplace’s equation,
which it is when the position of the charge is excluded. If all of space is con­
sidered, however, we are dealing with a solution of Poisson’s inhomogeneous
equation, where the delta function appears as the charge density. That brings
us to the question: How is the delta function represented in curvilinear coordi­
nates? To answer this, we refer to the basic integration property

Jv (dv) 6(r - r') = { J; - (16-40)

and recall that [(16.11)]

(dr) = h(u) du\du2dus. (16.41)

Accordingly, we must have

5( ui - « () S(u2 - u'2) S(u3 - « 3 ).

6(r - r') = T7 - t < (16.42)
h(u)

This is illustrated in circular cylindrical coordinates, where

(dr) = dp pd(j> dz) (16.43)

by the construction

<$(r — r') = —S(p —pf) 8(<j) — <j>f) 6(z — z'). (16.44)

A particular example of a curvilinear coordinate delta function appears when

we return to (16.6), or

dk J0(k p )e -k\z^ (16.45)

and apply the Laplacian differential operator, without excluding z = 0. On

recalling from (13.24) and (13.26) that

/ d2 o\
( - 5 ? + * ) - • * - = '<*>■ <1M6)

we arrive at
1 1 r°°
- V 2- = 4t dkkJo(kp). (16.47)
r 27T J0
What can we say about the latter integral? A glance at (16.5) shows that

TO if p ^ 0
-2ttÍJ0 d k k J o (k p )= lim
+o 27T (p2 -{- z2)3/2 \ oo if p = 0
(16.48)
16.1. DELTA FUNCTIONS AND COMPLETENESS 171

To these limiting results we append the value of the polar coordinate surface
integral [this is (15.37) again]

f°° 1 2 i r°° i
I dpp2*2^{p*+z*y/> = 2 Jo dt(t + 1)3/2 = 1 ’ (16-49)
independent of the particular choice oí z > 0 . Here, then, is the realization, in
polar coordinates, of S(x)S(y)) as required to complete the structure of <S(r) in

dkkJo(kp) = - 8{p)-? -. (16.50)

p Z7T
Notice that there is no reference here to <\>\that angle is indeterminate at p = 0.
[One can regard the factor 1/ (27r) as the average of 6(<j) — <j)f) over all values of
<j)f\it also helps to interpret S(p) as the limit of 8(p —pf) as p' approaches zero.
See also (16.70).]
The characteristics of any delta function— say the one-dimensional S(x —
x f)— are expressed in terms of an arbitrary function f ( x ) by the integration
property
/ dx'6(x — * 0 / ( * ') = /(*)• (16.51)
J —oo
Then, the insertion of the integral construction of the delta function

8(x — x') = J (16.52)

- o o 27T

produces
f°° dk ikx
™mL y ° ° d x 'e -ikx' f ( x ') . (16.53)

This exhibits the completeness of the set of functions exp (ikx) in which k ranges
continuously from —oo to oo; any function of x can be constructed as a linear
combination of that set, with the coefficients as displayed in (16.53). Such
integral representations of functions were introduced by Jean Baptiste Joseph
Fourier (1768-1830), although the particular complex form appearing in (16.53)
can be attributed to the independent investigations of Augustin Louis Cauchy
(1789-1857). The analogous two- and three-dimensional completeness state­
ments are similar consequences of the appropriate delta function constructions.
We now find the polar coordinate version of this completeness in two dimensions.
The starting point is the two-dimensional analog of (13.14) and (16.52),

/ ^ eiki ’ ( r " r ') A = - r' ^ = p

- ^p ~ p ' w ~ <16-54)

where we have introduced polar coordinates r¿(/>, (f>) and r ¿ (¿ /, </>'). Corre­
spondingly, if we use polar coordinates for kj.(Ar,a), (16.54) becomes
172 CHAPTER 16. BESSEL FUNCTIONS

We next expand the exponentials here by use of the generating function ex­
pression, (16.28) [with <¡) —> <j) — a] together with its complex conjugate [with
<j) —►ft — a] and perform the a integration by means of (16.29):

27r da
Y ^ i mé m^ j m{kp)
1 27T
.m . m'

771,771'
OO
= Y eim^Jm{kp)e~im*' Jm{kp'). (16.56)

The result of these operations is the identity

/ "o¿ir
—E
/»OO I JI oo -|
eim,t,Jm{kp)e-im*' Jm(kp') = U ( p - pf)6{4>- # ) . (16.57)
Jo P

Again this is a completeness statement, this time for the functions eim<^Jm(&p),
where k ranges continuously from 0 to oo, and m assumes all integer values:
0, ±1, ± 2 , --- That is made explicit on multiplying (16.57) by an arbitrary
function / ( / / , <^'), and by the area element pf dp' d<¡>\ followed by integration:

fM )= l
r°° b(jb

Jo 27r
Y
JE
eim*J™(kp) /
Jo
r°°
p'dp'
Jo
r 2lx .
Jm{k p ')f(p ',<!>')■

(16.58)
We may now easily isolate the individual p and <t> dependencies of (16.57).
If we multiply (16.57) by ) and integrate over <^, we select a particular
value of m, according to (16.29), so that we obtain

f°°
/ kdk Jm{kp)Jm{kpt) = 1-6(p - p'). (16.59)
Jo P
This states that Jm(kp) is a complete set of functions of p for any integer m:

f(p) = /
Jo
k dk Jmi^kp)
í
,Jo
p'dp' Jm{kp')f(p') (16.60)

Putting this information back into (16.57), we determine the completeness re­
lation for the functions etmi
oo
Y — = 8{<j> - (16.61)

and correspondingly
16.1. DELTA FUNCTIONS AND COMPLETENESS 173

which is the Fourier series expansion for /(</>), that is, the statement of com­
pleteness, over a 27T interval, of the set of functions exp(¿m<^), m integral.
Perhaps we should remark here that the validity of (16.61) depends on the
restricted 2ir range o f the variables (j) and <^'. But the restriction can be lifted
if the right-hand side of (16.61) is made periodic in its variable to match that
property o f the left-hand side. With this unlimited extension of <j) — ft now
called x y that is displayed by
1 oo oo
_ £ e<" » = X 8 {x -2 * v ), (16.63)

and, indeed, when x is restricted to the interval between —2ir and 27r, only v = 0
can contribute to the right-hand side. It’s interesting to check the correctness of
(16.63) by introducing the integral representation for the delta function, (16.52).
That produces, for the right-hand side of (16.63),
oo

/
dk ikx —i2iruk
e > (16.64)
-oo

which can equal the summation on the left-hand side of (16.63) only if

X e~i27tuk = 5^ S (k -m ). (16.65)

After appropriate redefinitions (k —» x/2ir, m —* v , v —►—m), this becomes just

the original relation (16.63).
A further property of the Bessel functions may be obtained by expanding
e«kx •( r - r ') x using (16.28), and integrating over a [as we did in (16.56)]:

r2ir j 00
/ ^ e ik x ' (r_r'U = X Jm{kp'). (16.66)

On the other hand, we could also have specified on as the angle between the two
vectors kj_ and (r — r /)±. That would identify the left-hand side of (16.66) as
the Bessel function of zeroth order, (16.4),

C ^ eifc|(r-r')x|coS« _ J0(fc|(r _ r'Jxl), (16.67)

Jo
where _______________________ _
|(r — r')j. |= \Jp2 + p'2 — 2ppf cos(<j) — </>') = P (16.68)
(read P as capital rho). Therefore we have derived the addition theorem for the
Bessel functions of integer order:
oo
J0( k P ) = X eim*Jm(k p )e-im*' Jm(kp'). (16.69)
174 CHAPTER 16. BESSEL FUNCTIONS

Now, if we return to the completeness relation (16.57), and recall the integral
(16.50), the implication of (16.69) is that

U ( p - p l)8(<t>-t') = ± 6 ( P ) ± - , (16.70)

which is a quantitative version of the remark that coincident points have zero
spatial separation.

16.2 Problems for Chapter 16

1. Fill in the steps leading to (16.15).

2. Using the divergence theorem, show that the divergence in orthogonal

curvilinear coordinates is

v'v=í ? ¿ ( ¿ 4

Use Stokes’ theorem [Problem 1.2] to show that the curl is

where z, j, and k are in cyclic order.

3. Derive the differential equation (16.34) satisfied by Jm, directly from the
generating function (16.22).

4. Use the integral representation of Jm, (16.30), to prove the recurrence

relations

2 ~ (16.71)
TTl
2 = Jm -l(t) + Jm+1(t). (16.72)

5. Using the results of the previous problem, show that

d m —1 d m . _ ..
dt t ~Jt + T ) Jm(<)
= Jm(t)

and from this derive the differential equation satisfied by Jm.

6. Using the integral representation of Jm, (16.30), for m > 0, develop the
power series for Jm, (16.27).
16.2. PROBLEMS FOR CHAPTER 16 175

7. Bessel’s equation of order i/,

u" -f - u f + ( l - u = 0,
# \ x¿)
possesses two solutions. One of the them is Bessel’s function of the first
kind, Ju(x), which is regular at x = 0. Show that a series solution of
Bessel’s equation yields, up to a multiplicative constant
2m
( - 1 )”
*<■>-( D ' m
E= 0a i rv ! + ^ + l ) (I)
where T(^) is the gamma function, which generalizes the factorial,

r(* + i) = *r(*), r(n + l) = n!.

Note that the series expansion of Ju(x) agrees with (16.27) if v — n, an
integer. Show that this series converges uniformly and absolutely for all
finite x , but possesses a branch line for v integer.

8. If v is not an integer, Ju(x) and «/_„(#) are independent solutions to

Bessel’s equation. However, from (16.25),
Jn(x) - ( - l ) n J - n(x).
Prove this directly from the series representation. (Note that l/T(—k) = 0
for & = 0 ,1 ,2 ,__ ) In this case we need a second solution, called Bessel’s
function of the second kind (or the Neumann function)
cos utt Ju(x) — J - U(x)
Nu(x)
sin v'K
Show that
1 (d M x ) n 9J-u(x)
Nn = lim Nu( -(-i)
u—+n dv v —n ^

is a linearly independent solution.

9. Show that the recursion relations in Problem 16.3 hold for non-integral
order. [Use the series representation.]

10. By substitution, show that the asymptotic solution to Bessel’s equation

has the form ( x —►oo)

u(x) = w \ ( x ) x ~ 1/2 cos — —Í/7T— ^ J -\-W 2( x ) x ~ 1/2 sin

S ^x — 7r — —

where

Wi{x) ~ a o ^ i - i r c a n ^ + t o ^ C - i r c a n + i * " 2^ 1
n —0 n —0
oo oo
w2(x) ~ 60 ^ ( - l ) nC2„ * - 2n - a o X ; ( - l ) nc 2n+ia:- 2n- 1)
n=0 n=0
176 CHAPTER 16. BESSEL FUNCTIONS

with ao, bo arbitrary. Compute cn ,

_ (4t/2 — l)(4i/2 — 32) •••(4u2 — (2n — l ) 2)

co = l, cn = --------------------------— -------------------------- , n = 1 , 2,....

Note that the series terminates when v = n + 1/2 and then these asymp­
totic relations become exact. Jv(x ) is represented by the above with
a0 = ( 2 / 7r)1/ 2, 60 = 0 , and Nu by the above with a 0 = 0, 60 = ( 2 / 7r)1/ 2.

11. Coulomb’s potential in circular cylindrical coordinates is given by the two

alternative representations in (16.6). Insert the integral representation for
Jo, (16.4), and integrate over k. Express in simplest terms the integral
you have thereby evaluated.
Chapter IT

Parallel Conducting Plates

17.1 Reduced Green’s Function

Having developed some mathematical machinery, let us now turn to another
essentially one-dimensional problem, that of the potential due to a point charge
between two parallel grounded conducting plates, as illustrated in Fig. 17.1.
Green’s function— the potential of a unit point charge— is defined by the differ­
ential equation
—V 2G(r, r') = 47n5(r —r'), (17.1)

together with the boundary conditions

G= 0 at 2 = 0 , a. (17.2)

Since the geometry depends only on the 2 coordinate, Green’s function can be
written in the (2 + l)-dimensional form (13.23),

<¡>= 0

z = 0 z —a

Figure 17.1: Geometry of grounded, parallel, conducting plates.

177
178 CHAPTER 17. PARALLEL CONDUCTING PLATES

where the reduced Green’s function g (z,z']k ± ) satisfies (13.26),

(-^ 2 + ¿ l ) g(z, z'; kx ) = S(z - z'), (17.4)

subject to the boundary conditions

g (z ,z f]k±) = Q at 2 = 0, a. (17.5)

Our first response is in the spirit of the straightforward development used

in Chapters 13 and 14, based on solving the homogeneous equation appropriate
to 2 ^ 2 ', (13.27), with the connections provided by the continuity of g and
the discontinuity condition of (13.28). As before, the basic solutions of the
homogeneous equation are still of the form e±kz (with k = |kj_|); the linear
combinations that satisfy the boundary conditions (17.5) are expressed in terms
of the hyperbolic function sinhar.

0 < 2 < z* : g = Asinh& 2 , (17.6)

a > 2 > 2' : g = B sinh k(a — 2 ). (17.7)

The constants A and B are to be determined by the conditions on g in the

neighborhood of 2 = zf [recall (14.17) and (14.18)],

z = z '+ 0
g continuous, -S -g 1, (17.8)
oz z —z ' —Q

which leads to the equations

A sinh £ 2 ' = 5 sinh k(a — 2 '), (17.9)

kB cosh k(a — 2 ') -f kA cosh kz‘ = 1. (17.10)

It is convenient to satisfy (17.9) by letting

A = Csinh k(a — 2 '), B = Csinh kzf, (17.11)

which, when substituted into (17.10) yields

kC sinh ka = 1 . (17.12)

The reduced Green’s function is thus found to be

, . _ sinh sinh fc(a — z>)

(17.13)
’ ’ fcsinhfca ’
where 2 > ( 2 <) is the greater (lesser) of 2 and 2 '. Note that the reciprocity
condition, (12.7), is satisfied because g(z,z']k) is symmetrical in 2 and 2 '.
It is worth noting what happens as the two plates are withdrawn from the
neighborhood of the points specified by 2 and 2 '. This means that both kz<
17.2. INDUCED CHARGE 179

and k(a — z>). and, of course, ka, are considered to become very large. Then
every sinh» appearing in (17.13) is dominated by |e®, and

g(z, k) ~ 'I, ( 1 7.14)

the vacuum form displayed in (13.24). Alternatively, we can withdraw just the
right-hand wall, for example, as accomplished by having ka become very large
without reference to z and z'. The outcome

g ( z , z ' ] k ) ~ y-sinhkz< e~kz> = _ JLe- * ( '+ ,,)| (17 .15 )

k 2k 2k
is just what is contained in (15.34), appropriate to a single conducting plate at
z = 0.
With a knowledge of Green’s function, we can now answer questions about
the charge induced on the conducting plates, and about the interaction energy
and force between the plates and the point charge.

17.2 Induced Charge

One application of this result lies in calculating the charge densities on the
conducting plates induced by a unit point charge at z*. According to (11.69),
these charge densities are
c\
z = 0: 4iut = E z = —^ G , (17.16)
C\
z = a: Ancr = —Ez = ^ G . (17.17)

As in Chapter 15, it is simplest to calculate the total charge induced on each

plate:

Q in d (* = 0) = - ± 4 i r j ( d r ± ) j r'U ¿ * ( * . *)
z= 0
(17.18)

Qiad(z = a) = ¿ 4 » J (drj.) J ^ ^ ^ e’k'L*(r-r,)j- z'\k)

(17.19)
We have seen such integrals previously [see (15.39) and (15.38)]. The spatial
integrations over rx yields (27r)2<5(kx), while the subsequent k± integration sets
fc = 0 so that ^
g(z,z'; 0) = -z < (a - z > ) . (17.20)

The total induced charges are therefore

180 C H A P T E R 17. P A R A L L E L C O N D U C T IN G P L A T E S

with the total induced charge on both plates being —1, of course. The total
induced charge is divided between the two plates in inverse proportion to their
distances from the point charge:

z'Qind(z = 0 ) = (a - z')Qmd(z = o). (17.23)

17.3 Energy
The interaction energy between the conducting plates and a point charge e, at
the location ro, is given by [cf. (15.5)]

Emt = i e 2[G(r,r') - G0(r, r')]r,r

f (dk±) sinh kzo sinh k(a — Zq) 1
= e227r
J (2 * k sinh ka 2k
(17.24)

_ 1 2 f°° ju rcosh kci n cosh k(a — 2zo)

JQ
ip
(17.25)
int ~ 2 6 dk [sin h ka ~ ¡iühJ^
Then, if we introduce the quantities

x — e —ka , %z 0
a = 11 -------- . (17.26)

where a ranges between + 1 and —1 , this reads

2 f l , X a + x ~ a - 2x
a x --------------------- . (17.27)
l-x 2

One situation is particularly simple: a = 0 or zq — -|a; the point charge is

equidistant between the two plates. Then, we have

1 e2 f 1 1 e2 ,
z0 — 2 ° ' ----- / dx--------- = -------In 2 . (17.28)
a Jo 1+ x a

To proceed with the general situation, we introduce the power series expan­
sion for the denominator in (17.27), and get

Ei
■ 2 x ) J 2 x 2n
n=0

- I4 Vaf_ l_ 1 2
(17.29)
n
¿— i I J - 1±°L
+ n + lT L n+ 1J'
But before presenting this summation in terms o f a known transcendental func­
tion, we use it as it stands to isolate the leading terms for the two situations,
1 —a <C 1 (^o < «) and l + <* <C 1 (a —zq •< a), which position the point charge
17.3. ENERGY 181

close to one of the conducting plates. To focus on the circumstance zo/a <C 1,
we first rewrite the summation (without the factor 1/4) as

1
n + (z0/a) n + 1 — (zo/a) n + l_

1 1
1 V + (17.30)
- izo/a) + 2 ^ n + (zo/a) n — (zo/a) nj ’
n= 1

which makes clear that the two leading terms are

1 9 00 1

(17.31)
O T + J( ? ) n—1
The summation encountered here is an example of the zeta function introduced
by Georg Friedrich Bernhard Riemann (1826-1866),

00 1
( ( ') = £ ? ! <17'32>
n= l

the relevant numerical value is

C(3) = 1 .2 0 2 .... (17.33)

Thus, we arrive at the interaction energy

«,««: (17.34)

where the first term is recognized as that associated with the nearby conductor
[(15.5) with e = 00], thereby identifying the next term as the leading contri­
bution of the second, distant conductor (more about this later). The analo­
gous result for the other situation, a — zo <C a, requires only the substitution
z0 —> a - zq .
The gamma function, first studied by Euler, is conveniently defined in the
manner of Karl Wilhelm Theodor Weierstrass (1815-1897), by stating its recip­
rocal as an infinite product:

1
(17.35)
r(<) = n

here, 7 is Euler’s constant,

N 1 1
7 = lim ----- In iV = 0.5772. (17.36)
N -+ 00
Ln = l

Of more immediate interest to us, however, is the related logarithmic derivative

function, the digamma function,

1
iKO = ^ lnFW = - 7 - \ (17.37)
n+ 1 nJ
182 CHAPTER 17. PARALLEL CONDUCTING PLATES

Indeed, on introducing the relation

(17.38)
t) =1-
we can present (17.37) as

(17.39)
W n=0n \xn + t n + l 'j

which allows us to evaluate (17.29) generally as

e>2 r
-^int — ^ ( — 77— ) + 2t (17.40)
4a

The comparison of this form with the particular example of (17.28), correspond­
ing to a = 0 , supplies the value

T
^ \ 2 J = “ 7 — 2 In 2; (17.41)

this can be checked directly through the following evaluation of the summation:

i
lim x 2n ( ------------------- r ) = lim ( - In % ^ In - = 2 In 2 ,
V
\ n -f- 1 / 2 nn--f f 1i y/ x -+ i\\xx ^ 1l —
—xx xx2 1 — x 2J
n=0
(17.42)
which uses the series expansion of the logarithm,
oo n
in (i+ » ) = - ^ ( - i r ^ - . (17.43)
n=l

17.4 Force
The force exerted on the point charge is given by the negative derivative of the
interaction energy with respect to the coordinate Expressed in terms of the
Z q .

derivative of the digamma function,

i oo 1
(17.44)
nr
= 0i ( n + *)2 ’

that force is

+ a —a
)]
1 1
■4>'

(17.45)
17.5. IMAGES 183

As one would expect, this force vanishes when the charge is equidistant between
the plates,Z =
q One also expects that a displacement from the equilibrium
^ a .

point should produce a larger force of attraction toward the nearer conducting
plate, resulting in a net force that acts to increase that displacement— the equi­
librium at the midpoint is unstable. That is made explicit on expanding (17.45)
for small values of a,

M < 1 : ^ * ¿ « W 2) = ¿ [ - ^ " ( l / 2) ( z q - a/2), (17.46)

where (see Problem 17.2)

1 00 1
- ^ " ( V 2 ) = £¿ -? ,i(T„ +T1 7/ 2™ (17.47)
)3 = 7« 3>'

although the qualitatively important point here is that the latter is a positive
number; the force is proportional to a small displacement from the equilibrium
point, and acts in the same sense.
The explicit expression for the force,

e2
OO o

4. y ' -------------- ----------------- (17.48)

( 2*0 + 2 nay ¿ ¿ ( 2 ( a - 2 0) + 2 na)2 ’

clearly suggests an interpretation in terms of an equivalent distribution of point

charges— image charges. It is not possible, however, just from a knowledge
of the force to assign uniquely the values and locations of these charges. For
example, a positive contribution to the force on charge e could equally well
arise from an image charge e located to the left, or from an image charge —e
positioned to its right. And, one would be unaware of the possible compensating
force contributions of identical charges arrayed equidistantly on opposite sides
of charge e. Accordingly, having thus been led to the question of an image
charge description, we make a direct attack on this problem.

17.5 Images
Let us look back at Green’s function in the physical region to the right of a
single conducting plate, as it is expressed by (15.34),

* > 0: g(z, z'; *) = L [e- fcl'-* 'l - e-*(*+''>] . (17.49)

This function vanishes at z = 0, satisfying the boundary condition for a con­

ductor at zero potential, and it obeys

z>0: ^ - ■ ^ + k2Sj g ( z , z ' ; k ) = S ( z - z l), (17.50)

184 CHAPTER 17. PARALLEL CONDUCTING PLATES

which makes explicit the unit charge at the point z' > 0. Now, we get a bounded
extrapolation of (17.49) into a fictitious vacuum region 2 < 0 by writing g as

d(z , k) = ¿ [e-*l*-*'l - e -* l'+ ''l] , (17.51)

which function obeys the differential equation

^ -^ 2 + fc2) 9(z ’ z>’ = s (z - z') - s (z + z'); (17.52)

the image charge appears here on the same footing as the physical charge.
The effect of the coordinate reflection 2 —» —2 is to reverse the sign of the
right-hand side of (17.52). Then the function </, being determined by the right-
hand side of its differential equation, will also reverse sign. Accordingly, the
continuous function g necessarily vanishes at z = 0. Here is the recognition
that symmetry considerations in extended regions can occasionally be applied
to satisfy boundary conditions. With this lesson in mind we turn back to the
pair of conducting plates.
In this circumstance the function g must vanish a,t z = a, as well as at
z = 0. Let us begin at the latter point and follow g to its zero at z — a. Now we
extrapolate beyond z = a, where the continuous function g will start out with
reversed sign. Suppose that the course of the function from z = a to 2 = 2a
duplicates that from z = a to z — 0, but with the opposite sign. Then the
function g will vanish at 2 = 2a. If we continue in the same way, with another
sign reversal at this zero, the behavior of g from z = 2 a to z = 3a duplicates
exactly that from z = 0 to z = a. In short, we are led to extrapolate the
function over the entire range of z %—oo to oo, by imposing the requirement of
periodicity, with period 2a. That periodicity of g must be matched by a like
periodicity of the point charges, so that the initial unit positive charge at zf is
supplemented by unit positive charges at z* ± 2a, z* ± 4 a ,___Doing the same
thing with the unit negative image charge at —z1 then replaces (17.50) with

{ - ’0^2 + k 2) g ( z , z , ; k ) = 1*>(Z z* - 2va) - 6(z + z* - 2va)\. (17.53)

Fine, but does it work? Will this charge structure reverse sign under reflec­
tion, both at z = 0: ¿ —* —z , and at z = a: z —a^> —(z —a)? Without question
for 2 = 0: The replacement v —» —v leaves the charge structure intact. And
then we recognize that ¿ —* —2 + 2 a differs only in the additional displacement of
2a, which is just the periodicity interval of the charges. The solution of (17.53)
will indeed obey the boundary conditions for the pair of conducting plates. See
Fig. 17.2.
But before we construct that solution, let us look back at the force and
energy in light of the now known pattern of image charges. For the circum­
stance of charge e stationed at , the image charges comprise charges e at
z q

the positions 4 * 2 z/a, v = ± 1 , ± 2 , . . . , and charges —e located at — -f 2 va,

z q Z q
17.6. LINEAR LATTICES 185

Figure 17.2: Positions of image charges for two parallel conductors.

v — 0, ±1, ± 2 , __ The image charges of the same sign as the physical charge
are disposed symmetrically about the latter, resulting in no net force. The dis­
tances between the physical charge and the image charges of opposite sign that
stand on its right, and on its left, are 2(a — Zo), 2(a — Zq) -f 2a, . . . , and 2zo,
2zq -f- 2a, . . . , respectively. Inasmuch as forces of attraction from the right and
left are counted as positive and negative, respectively, the total force is precisely
as given in (17.48). Turning to the interaction energy given in (17.29), we first
rewrite it as

£int = JE ^ ena
1 00

n= l L
2
2 e2
2na — 2zo
e2
2 (n — l)a + 2z$m
(17.54)

Then we can recognize in the successive terms of the summation the electrostatic
energy of interaction between the physical charge and the image charges; first,
those of like sign, and then the oppositely signed charges, positioned to the
right and to the left, respectively. Very good, but why the additional factor of
1/2? [This question might well have been raised at our first encounter with the
interaction energy that was interpreted in terms of an image charge, (15.5).]
The electrostatic energy of the physical point charge can be identified with
the work performed in assembling successive infinitesimal multiples, ¿Ae, of
the final charge, with A increasing from zero to unity. At a stage where the
assembled charge is Ae, the image charges that it induces are also reduced by
the fraction A of their final values, and so also is the electrostatic potential of the
image charges. If <j) is the final value of that potential, the total work performed
in raising A from 0 to 1 is

d\e\<l>=le<f>; (17.55)

just one half of the electrostatic energy associated with the final charges, re­
garded as independently assignable.

17.6 Linear Lattices

The latter remark directs us to the solution of a different physical problem.
Consider a one-dimensional lattice of charges —e, with uniform separation 2 a,
186 CHAPTER 17. PARALLEL CONDUCTING PLATES

and a similar lattice composed of charges -f-e, which is displaced to the right
(along the 2 -axis), relative to the first lattice, by the distance 0 < 2zq < 2a.
In particular, for z0 = |a, the complete one-dimensional lattice is composed of
alternatingly signed charges with uniform spacing a. What is the electrostatic
energy of any individual charge? The charge pattern here is just that of a
physical charge e and its images in the parallel conducting plates, as shown
Fig. 17.2. Accordingly, the electrostatic energy per unit charge is twice that
given in (17.40); in the example with all charge spacings equal to a, the energy
per charge, twice that presented in (17.28), is —(e 2 /a)21n2.

17.7 Periodic Green’s Function

We shall now solve the differential equation (17.53) with the aid of the relation
(16.63). First we rewrite the delta function by supplying the variables with a
constant multiplicative factor in accordance with the simple relation

k6(kx) = 6(x), k > 0, (17.56)

which has already been used without comment in connection with (16.65). Thus
we have

Y ] [8{z —z* —2va) — ¿(2 + 2 ; — 2 va)]

00

= 1 £—00 K ^ - « - H
v~
+

_
1 00
\ s [ imw(z —z^/a ÍTmr(z-\-z, ) l a 1
(17.57)
~ 2a ^ i J’
which, in turn, can be rearranged as

lA r n 7r / n 7r / 2 ^ . nw . nir ,
- > c o s — (2 — 2 ) — c o s — (2 + 2 ) = - > sin — 2 sin — 2 . (17.58)
a n= 1 L a a J a {
n —1
a a

Should we now return to the physical range of the coordinates 2 and 2 ', so
that
—a < 2 — zf < a, 0 < 2 + 2 ' < 2a, (17.59)

only the physical charge contributes to (17.57) and this relation becomes
0 00
- ^ s i n — 2 sin — 2 ' = S(z — 2 '). (17.60)
n= 1

This is a statement of completeness of the functions sin rnrz/a, n = 1 ,2 ,...,

over the interval between 0 and a. That is, for any function f(z) defined over
17.7. PERIODIC GREEN’S FUNCTION 187

this interval,

. UTTZ
/(* ) = £ sin ----- (17.61)
a
n=l

which is a (sine) Fourier series.

With the right-hand side of (17.53) replaced by (17.58), the solution of the
differential equation is evident:

g (z ,z ';k )= - £ (17.62)

And this function obviously obeys the boundary conditions, inasmuch as all
sin(mr/a)z vanish at z = 0 and z = a. Or is it really that cut and dried? Let’s
construct the Fourier series for the function 1 — (z/a):
oo
Z r -^ • n7r
1 - - = > sin — z (17.63)
a " a
n= 1
where the latter integral is evaluated as
a mr , / ' a . mr . a
— cos — z 1 ------ sin — 2 (17.64)
mr a \ a (rnr)2 a mr
The result is
2 rnr
— sin — z. (17.65)
> - : - E mr a
n=1
Certainly both sides of this equation vanish for z = a. But suppose we put
* = 0?
In order to refute the possible inference that the sine Fourier series applies
only to functions that vanish at 2 = 0 and a, let us evaluate explicitly the
right-hand side of (17.65):

£ — sin ^ - z = I m i £ I f e * - / * ) ” = Im ^ ln , 1 . , (17.66)
a T r^ n V / tt i _ eW<« ’ v >
n=1 n=l

which, in turn, is [0 < z < a]

0—i(nz/2a)
Im—In .«72 = l - i . (17.67)
7r _ 2 s i n ( 7 r z /2 a ) a
The point here is that, while individual terms of a sine Fourier series vanish as
z —* 0, the series as a whole can converge to a nonzero value in this limit. Thus,
suppose we were to assume that, with sufficiently small 2 , sin(n7r/a)z « (mr/a)z,
for all values of n that contribute significantly to the summation in (17.65). Then
the latter would be approximated by
188 CHAPTER 17. PARALLEL CONDUCTING PLATES

which does not exist. We are being told that unlimitedly large values of n do
contribute, contradicting the assumption involved in replacing sm{nK/a)z by
(inir/a)z. But if the dominant contribution comes from very large values of n,
the summation can effectively be replaced by an integral,

^ 2 . n'K 2 f°° dn . uk 2 f°° dt , v

> — sin — z « — / — sin — z = — / — sini, (17.69)
n 7r a 7T JQ n a % J0 t

yielding a finite limit as 2 —►0. As to its specific value [already quoted in

(13.19)— see also Problem 13.4], note that, through the option of redefining the
integration variable, we have, for arbitrary a?,

dt 2 [°° dt
— sin(xT) = —€(x) I — sini, (17.70)
t n Jo t

where
/ x / 1, x > 0
(17.71)

Now we differentiate with respect to x and observe that the derivative of the
step function e(x) is 2 6(x), reflecting the value of the discontinuity at x = 0 :

9 i 00 9 f ° ° rli
—/ dt cos(xt) = 26(x)— I — sint, (17.72)
ft Jo TCJo t
where the left-hand side is recognized to be

1 f°°
- / dtelxt = 2 S(x); (17.73)
^ J —oo

as expected, we learn that

2 [°°d t .
— / — smz = 1 . (17.74)
* Jo t

It is now clear that g(z, z'\ k) will indeed vanish at z = 0, say, if the infinite
series that results from the replacement of sin(n7r/a)^ by (nw/a)z is a convergent
one. This being a question referring to very large values of n, it suffices to make
the test for k = 0. Then, with zja < 1, we get just the sum given in (17.65),

UK z(a — z')
- < 1: g (z} 0) = * I T — sin — J = (17.75)
a n'K "a
n=1

Yes, g(z , z'\ k) does vanish at z = 0 and at z = a. Incidentally, if one looks at

the k = 0 limit of (17.13),

z< ( a - z > )
g (z,z’ )0) = (17.76)
17.8. PROBLEMS FOR CHAPTER 17. 189

it is evident that (17.75) holds, not only for 2 <C a, but for all 2 < z*. More
generally, the equivalence of the two forms for g (z, z'\ Ar), (17.13) and (17.62),

sinh kz< sinhAr(a — z>) _ 2 sin(n 7r^/a) sm^rntz1/a)

(17.77)
fe sinh Ara a (mr! a)2 + k2
n= 1 v ' '

can be checked by carrying out the integrations involved in representing the first
version as a Fourier series in 2 , or in z*. [See Problem 17.4]

17.8 Problems for Chapter 17.

1. A semi-infinite dielectric slab, with dielectric constant e, fills all space with
2 < 0. A perfect conducting plate occupies the plane z = a > 0. Calculate
the reduced Green’s function, g(z, z'] k±), for 0 < z^z* < a, and the full
Green’s function G(r, r') in the same region.

z —a

Geometry of parallel dielectric and conducting surfaces

2. Prove that
1
E (n + 1 / 2 )*
= (2 * - l)C (s ),

and hence establish (17.47).

3. We now wish to investigate the properties of ^(^, ^), (17.13), considered

as a function of a complex variable k. First notice that g is even in k and
is finite at k = 0. The behavior of g when the real part of k is large and
positive is given by (17.14). This limiting form is the reduced Green’s
function for empty space, (13.24), which is evidently bounded (in fact,
goes to zero) as Re& —* oo. The singularities of g (z, z'\ k) occur on the
imaginary axis where
sinh ka = 0
190 CHAPTER 17. PARALLEL CONDUCTING PLATES

(p, •!>,*)

Figure 17.3: Infinite conducting sheet, described in cylindrical coordinates by

<j> = 0, 7T. The coordinate z is perpendicular to the page. A unit point charge is
located at the point (//, ft, z').

that is, at the points where

ka = inn, n = ±1 , ± 2, ± 3 , . . . .

By examining the behavior of g in the neighborhood of these singularities,

where g has simple poles, and noting that apart from these poles, g is a
bounded, analytic function of Ar, express g entirely as a sum over these
pole contributions,

Finally, by combining the contributions for n and —n, obtain the result
(17.62).

4. Alternatively, show the equivalence of the two forms of g in (17.77) by

expanding the left-hand side in a sine Fourier series as suggested.

5. Consider an infinite conducting sheet, with a unit point charge above it.
Use circular cylindrical coordinates, so that the point charge is located
at ( p '.f t ,^ ) , as shown in Fig. 17.3. Find Green’s function by using ap­
propriate representations of S(z — z') and 6(<f> — ft), Get the differential
equation for the radial functions g(p, //) . Now consider the potential of
an infinite line charge of unit density,

and the associated radial functions y(p,p/). Construct the latter from the
differential equation and boundary conditions. Show that the resulting
function can be expressed in terms of the complex quantities
17.8. PROBLEMS FOR CHAPTER 17. 191

C -C *
r = 2 Reln-^—

6. Do the same for a semi-infinite conducting sheet, defined by <j>= 0. Under

the circumstance where <f> and <f>' are small positive angles while p and p'
are large, in the sense that \p —p'\ <C p, p', the physical situation is effec­
tively that o f Problem 5 (the edge is far away). Show the correspondence.
Chapter 18

Modified Bessel Functions

An eigenfunction expansion for Green’s function, G (r,r'), for the geometry

shown in Fig. 17.1 can now be obtained by substituting the corresponding form
for the reduced Green’s function, (17.62), into (13.23). The Bessel function of
zeroth order, Jq, (16.67), is introduced upon performing the angular integration
associated with k x , so that Green’s function becomes

G (r ,r ') = 2 [ dk k J0( k P ) ~ Y (18.1)

where
P = \ ( r - r')j. |= [p2 + pn - 2pp' cos(<j> - <j>')]lf2. (18.2)
A complete eigenfunction decomposition of G could be obtained by using the
addition theorem (16.69). In (18.1), we encounter a new type of Bessel func­
tion, ATo, the modified Bessel function of zeroth order, defined by the integral
representation
Ko{Ap ) = J ~ dk k ^ Q . (18.3)

It suffices to make the change o f variable kP = s to recognize that the integral

depends only on the combination AP = t :

K ° W = Jo dss^ p - ( 18-4)

In terms of this new function, the Green’s function in the region between two
parallel plates is

G (r, rO = \ £ sin ( ™ z) sin ( ^ ' ) I<0 ( ™ p ) . (18.5)

n—1

If we had not performed the angular integration in (18.1), we would have

encountered the modified Bessel function as
, v f (dkx) eikjL
PTo(AP) = 2 i r j ( 2^)2 A:2 + A2 ' ^18'^

193
194 CHAPTER 18. MODIFIED BESSEL FUNCTIONS

Then, we see immediately that Ko is a two-dimensional Green’s function,

( - V l + X2) K 0(XP) = 2irtf((r - r')x )- (18-7)

As always, that statement comprises two bits of information. The first one is
the homogeneous equation applicable for rx / , which is

p > 0: ( - V i + X2) K 0(XP) = + * 2) Ko(XP) = 0, (18.8)

or

i>0: ( - 7 5 ‘ 5 + I) ír»(í ) = 0 <18'9>

Notice that the latter equation differs from the one obeyed by Jo(0> (Id. 18),
only in a sign change, such as would be produced by the substitution t —►
it. Accordingly, Ko(t) is also sometimes called a Bessel function of imaginary
argument. Incidentally, the relationship between the two kinds of differential
equation is emphasized on using the construction (18.3),

|Jo(kP)
( - é ¿ Fé + * ) w p) = t ’ ( - p dp p f p + ^ j
pOO
= / d kkJ 0(kP), (18.10)
Jo
which is
p ( P ) = 0, P > o, (18.11)

according to (16.50).
The second bit of information, that carried by the inhomogeneous term in
(18.7), is extracted on integrating the latter equation [or (18.10), (18.11)] over
an arbitrarily small circle surrounding the point r'± , using the two-dimensional
form of the divergence theorem:

XP < 1 : -2 ir P -jp K o (X P ) = 2 it. (18.12)

We have assumed that the integral of K q is arbitrarily small, which is justified

by the implication of (18.12),

¿<1: K o ( t ) « In i -f constant. (18.13)

The value of the constant appearing here, along with other useful results,
can be obtained from alternative representations of K q. For these we return to
(18.6) and evaluate this rotationally invariant integral by using the direction of
(r — r ')x as the a?-axis:
eikxP
K »(AP) = i / (18.14)
dkmdk*k* + k¡ + W
 195

We now have the options of integrating first with respect to kx, using the integral
of (13.22), which results in (ky = k)

K0 ^ , (18.15)
V F T I7
or, of applying a particular example of the cited integral to carry out the inte­
gration with respect to ky (here k = kx)

1 Í 00 JkP
K 0(XP) = - j dk (18.16)
yjk2 4- A2 ‘
Equivalent versions of these two forms are
e-vs'+t* pc
d9 e —t co s h 6
(18.17)
K° V = l * 7y/s2
37 + ?
t2
(with the latter produced by s = tsinh#), and

[°° cos ss
COS f°°
Ko(t) = / ds-—======= = / <¿0cos(¿sinh0), (18.18)
Jo y s 2 + 12 Jo
respectively.
We apply the exponential version (18.17), for small t , by introducing a tran­
sitional value of the integration variable s, S', such that < < S < 1 . Then the
integral can be divided into two parts, with appropriate approximations, as

I<o(t) » [ S ds- t = L = + r d s^ -. (18.19)

Jo Vs¿ + Js s
The first of these two parts is equal to

In + V s2 + 1 2) . 25 (18.20)
nT ’

while the second one, rearranged by partial integration, is

OO poo poo
In s e s -f / dsln se~ —In 5 + / d s ln s e ” 5. (18.21)
5 Js Jo
This gives
t < 1 : K q(í ) » In j +
/ ds In s e *, (18.22)

which is indeed of the form (18.13).

The integral that appears above can be recognized as something known if
we look at the following [perhaps more familiar than that appearing in (17.35)]
representation of the gamma function,

dss* i e (18.23)
196 CHAPTER 18. MODIFIED BESSEL FUNCTIONS

or, more precisely, its derivative:

1 pOO
T'(t) = — T(t) = / ds ln s s <- 1 e- *. (18.24)
dt Jo

We now see that the integral appearing in (18.22) is r '(l), which is also ^(1),
inasmuch as T(l) ■= 1. Then on consulting (17.39), we learn that r '(l) is just
the negative of Euler’s constant 7 :

t < 1: K 0(t) » ln(2/t) - 7 = ln (l/t) + 0.1159 . . . . (18.25)

The exponential representation (18.17) is also particularly useful in finding

the leading asymptotic form of Ko(t) for t 1. In this circumstance, contri­
butions to the integral have already begun to decrease significantly when the
integration variable attains values s ~ t 1/2 <C which leads to the asymptotic
approximation

1
K 0(t) I d s-ex p (18.26)
Jo t
or
(18.27)

Of course, apart from the specific numerical factor, this asymptotic form is
immediately apparent in the version of the differential equation for Ko(t) that
is analogous (m = 0) to (16.37),

(18.28)

An application of the asymptotic behavior of Ko(t) appears on returning

to the three-dimensional Green’s function (18.5) and considering the situation
P a, where the distance between r and r' in the transverse plane is large
compared with the separation of the conducting plates. Then, of the sequence
of decreasing exponentials, exp(—nir(P/a)), it is the first one that dominates,

(18.29)
irP /a
P > a, G

The parallel conducting plates suppress most effectively the Coulomb field of
the point charge at large distances, on the scale set by the distance a. That
is qualitatively clear in the picture wherein the conductors are replaced by the
infinite set of image charges; the potentials of the two lattices composed of
positive and negative charges, respectively, will cancel almost completely at
large distances.
The image charge picture also makes evident that G(r, r') approaches Green’s
function for the vacuum— Coulomb’s potential— as |r — r'| becomes very small
on the scale set by a (except when the unit charge is quite close to one of the
18.1. MORE BESSEL FUNCTIONS 197

conductors). Indeed we can recognize in (17.24) a statement about the approach

to that limit:
| r - r '| < a : G (r ,r ') « + 2Pint, (18.30)

where Eint is the energy of interaction, with the conductors, of a unit charge
that is stationed at a point r « r'.

18.1 More Bessel Functions

First, let us summarize what we have learned about (2 + l)-dimensional repre­
sentations of Coulomb’s potential. Let’s begin with (13.17), presented as

00 dkz
Ir - r 'l 2 tt k\+k2
poo pOO
1
Jo(kP)e,ikx Z

=- dkk (18.31)
* JO J -o c k2 + k2

in the latter version k = k± and Z = z — z '. Now we have the options of

integrating first over kz , using (13.22):

1 _ 1
dkJ0( k P ) e - kW , (18.32)
|r-r'| “ ,/ W T Z 2

[this is (16.6)] or, of integrating first over Ar, using (18.3):

f
1 _ 1
dkzK 0{\kz \P)e i k x Z dkI<0(k P ) c o s k Z ) (18.33)
V P 2+ Z2 ~ *
where k replaces kz .
Observe that in one representation an exponentially decreasing function of
Z is combined with (for sufficiently large kP) an oscillating function of P, while
in the other representation it is an oscillating function of Z that combines with
(for sufficiently large k P ) an exponentially decreasing function of P. The neces­
sity for this mixing in the functional character of the multiplicative constituents
stems from the significance of l/|r — r'|, r ^ r', as a solution of Laplace’s equa­
tion:
/ d2 1 d d2 \ 1
+ p o p + dZ2) V F 2 + Z 2 ~ ° ' (18.34)
To the extent that the first derivative term is relatively negligible (P large), the
two second derivatives must be of opposite sign, requiring that a convex function
of one variable be combined with a concave function of the other variable.
Perhaps we should also point to the close relationship, indeed, equivalence of
(18.18) and (18.33). First notice that the statement of completeness in (16.53)
reduces, for an even function of x, to
■ pOO
2 f°°
/(# ) = — / dk cos kx / dx1cos kx1/ ( x 7) (18.35)
ft J o Jo
198 CHAPTER 18. MODIFIED BESSEL FUNCTIONS

Now regard (18.33), with Z playing the role of x, as such a Fourier integral
representation o f (P 2 + ¿J2) - 1/ 2 and conclude that
r°° i
K 0(kP) = J d Z 'z o s k Z '- — = = ; (18.36)

the substitution = s, kP = t then yields (18.18). In a similar application o f

the completeness of the Bessel functions J0(kP), (16.60), we infer from (18.32),
written as (Z > 0)

i— .r
-kZ
dk k Jo(kP)- (18.37)
+ Z 2 Jo k ’
that
- kZ

= [ d p ' p ' j 0( k p ' ) - - l (18.38)

Jo W P F + Z2’
or
(18.39)

The latter formula lends itself to the derivation of additional mathematical

relations. For example, one can integrate (18.39) with respect to t from 0 to oo,
using the integral

/ dt-
1

y/s2 + 12
= sinh
■(?)
2T
In — ,
s
T >^, (18.40)

and the null value recorded in (16.48),

poo
/ dssJo(s) = 0, (18.41)
Jo
to arrive at [a convergence factor, exp(—es), e —♦ + 0 , is understood]

f°° 1
/ dssln - Jo(s) = 1. (18.42)
Jo s
Or, we might multiply (18.39) by cos At before integrating with respect to t from
0 to oo. Then the following version of (18.18),

i At
Ko(Xs) = f dt (18.43)
Jo
yields
poo poo 1

J ds s J0(s) K o(Xs) = J ds s J0(Xs) I<o(s) — ^ (18.44)

where the alternative integral is produced by suitable redefinitions of variables.

[This formula is attributed to Oliver Heaviside (1850-1925).] Note the con­
nection between (18.42) and (18.44) that is provided by the initial behavior of
18.1. MORE BESSEL FUNCTIONS 199

K o , as presented in (18.25). But enough mathematical doodling. Let’s ask this

question: The addition theorem relating Jo(kP) to the Jm(kp) and Jm(kpf) y
(16.69), is of obvious physical importance. Does K o(k P ), as it enters (18.5), for
example, also possess an addition theorem?
To set the stage for our affirmative response, we first present an alternative
derivation of the addition theorem for Jo(kP). And for that we need relations
among the Bessel functions. Let’s begin with another version of (16.30),
p2ir j
/ p eim“ ei k i , r i = i meim*Jm(kp), (18.45)
Jo ^
in which a and <j) are the polar angles of kx and rx , respectively. (One returns
to the earlier form by using a — <^, or its negative, as the integration variable
while noting that the resulting integral is independent of the algebraic sign of
m.) Now observe that

h ( l ± % ) e'li =V k- ± . <«»•«»
from which follows

¿ ( ¿ ± % ) (imeim*Jm(kP)) = *m± 1 ei(m±1 )^Jm± i ( M - (18.47)

Although this is the preferred form for our present purposes, it is also possible to
remove the reference to <j> and present these just as relations among the Bessel
functions. [See Problem 18.1.] The two differential operators appearing here
behave as inverses when acting on the functions exp(im<¡>)Jm(kp):

I ( A ±4 1. (18.48)
ik \dx d y ) J [ ik \dx d y )_

It is in this sense that we begin with Jo(kp) and derive from it the following
constructions for both positive and negative values of m:

^ J m (kp )- ik (^d x + i d y ) J0(kp). (18.49)

Now let us consider

Jo(kP) = J 0(fc|(r — r,)x|) = e" ^ • V j - j 0( M . (18.50)

where the latter form uses a symbolic expression of a Taylor series expansion in
powers o f r^. It’s time to recognize that (18.48), or

(18,51)

which is a statement of the differential equation [see (16.33)] obeyed by the

functions exp(im^Jm^kp), asserts that (i/k)'V± behaves like a unit vector,
200 CHAPTER 18. MODIFIED BESSEL FUNCTIONS

when Jo(kp) is the operand. We can also assign to this unit vector a polar angle
4>, in the sense that

el® = cos<I> + ¿sin<I> = 7 - ( V (18.52)

k \ax ayJ
In consequence, we have from (16.28)
00
e - r '± • V _ l = • (i/fc )V x eikp' cos($>-<j>') = Jm(kp'),
m = — 00
(18.53)
where, with Jo(kp) made explicit,

imeim*J 0(kp) = i~m l ( A + iA)j Jo(kp) = eim^Jm(kp), (18.54)

so that (18.53) is the addition theorem (16.69),

J0( k P ) = ^ eim^Jm(kp)e~im^ 'jm(kpl). (18.55)

And so, with this example before us, we now write

K 0(kP) = K 0(k\(r - r')x|) = * V ± K 0(kp). (18.56)

But we see immediately that matters are somewhat different here. First, the
function Ko(kP) has a singularity at P = 0 [(18.25)] and the power series
expansion cannot converge outside the domain |rj_ | = / / < / ? . This fits in
well, however, with another, related difference. Unlike Jo(kp), Ko(kp) obeys a
homogeneous differential equation only if p = 0 is excluded. And that brings us
to the fact that the homogeneous differential equation obeyed by Ko [(18.8)] is
not the same as the one for Jo. Indeed, now we have, from (18.7),

¿V i - 1, (18.57)

and it is (—l/fc)V j_ that behaves like a unit vector.

That is one way to recognize that the Bessel functions entering the addi­
tion theorem of Ko must be those of imaginary argument, the ones that obey
the same homogeneous differential equation that governs ATo, corresponding to
(18.57). The real function Im(t) is conventionally defined by [see (16.25)]

) = Im(t) = J -m(0 , (18.58)

where either sign of i can be adopted on the left-hand side. Indeed, we get the
form we want by replacing t with —it in (16.28), thereby arriving at
oo
e *cos* = V eim4,Im(t)

OO

= Io(t) + 2 5 3 cos m<b (18.59)

m= 1
18.1. MORE BESSEL FUNCTIONS 201

which also supplies the integral representation

Im(t) = f — cos (18.60)

Jo *
We also note the appropriate modification of (18.49), as produced by the sub­
stitution k —+ —ik:

lid d \ lm
+ h { k p )- (18'61)
Accordingly, this time we have
oo
e - r ^ * V _ L _ g fcrl • ( - l / f c ) V i ekp‘ co s($ -< ¿') __ e*m(^~^ ) / m (& //),

(18.62)
where
d .d \
e*'* (18.63)
dx *dy ) ’
and the outcome is

¡* K m(k p )e-im*' Im(kp'), (18.64)

where we have defined

K 0(kp). (18.65)

This definition is analogous to (18.61), but differs in the additional sign factor
(—l ) m. That flaw is accepted in order to achieve the positiveness of both Im(t >
0 ) and K m(t > 0 ), which property of the Jm is evident in the power series
inferred from (16.27),
oo /J ,\ m + 2 n

Jm(t) = Y , ( [ M (18.66)
“ (m + n)! n!

To discuss the positivity of the K m(t)t as defined by (18.65), we return to

the initial definition of K q} (18.6), which is now presented as

Ko(tp) = 2* J (2, , , + t, ■ (18.67)

The effect of the differential operator in (18.65) on the exponential function

appearing here is given by

1 ( d
+ RíkX ,r j- = R<ki *rj-. (18.68)
-]fc(*i + i* i)
k V dx '*)
202 CHAPTER 18. MODIFIED BESSEL FUNCTIONS

It is clear that integration over the polar angle of k'± [k'x -|- iky = k'e101] will
produce the factor of exp( 2m^) that stands on the left-hand side of (18.65).
Accordingly, we are free to specialize to <f> = 0, which is to say, x = /?, y = 0,
and get

JKp
+ k12 + k2
0 dkl eK p
2 tt k'2 + k'2 + k2
1 e - V k’y + k2P
(18.69)
J k '2 + k2

which uses (13.22), or, with k'y = fcsinh#,

1 r°°
K m(kp) = - dO (cosh 0 + sinh $ y ne ~ kPcoshe. (18.70)

We are thus led to the integral representation

poo
K m( t ) = / dO cosh m.0 e~*cosh6, (18.71)
Jo
which is indeed positive, and also displays the property

= K m(t). (18.72)

We present one application of the Ko addition theorem that is related to

the integral (18.44). First, let’s use the integral representation of Ko(t) given
in (18.17) to evaluate
p oo pOO pOO poo
J dttI<o(t) = J dsj = y dse~s = 1 . (18.73)

Now observe from (18.64) that (ft = 0)

J * ^ K 0(kP) = K 0(kp)I0(kpf), P>P, (18.74)

from which follows

poo
J ™ dkkKo(kp)Io(kp') = J ^ l ° ° d k k IU {k P ) = ] ' * (18.75)

the latter integral is

i i i
Jo 2 tt p2 p'2- pp>cos <j>~ [(p2 p'2)2 pp1)2]1/2~ p2- p'2'
+ 2 + - (2 ( 1 8 -7 6 )
18.2. PROBLEMS FOR CHAPTER 18 203

With a change of variables, this result appears as

d s s I 0(Xs)Ko(s) = A < 1, (18.77)

which is just what emerges from the second form of (18.44) after A —►iX.
Finally, we comment on the asymptotic form of K m(t) for any given m and
sufficiently large t. Approximations analogous to those in (18.26) (9 ~ s/t)
gives the leading term from (18.71),

/ i m W ~ / ° ° ^ e~ t(1+#3/2) = t:>1, (18,78)

independently of m. And, a related treatment of the integral representation for
(18.60), yields

/ „ , ( < ) - r ^ e t(1- ^ 2) = 4 = e<> * » ! ’ (18-79)

Jo n y/2irt

which, of course, coincides with the implication of (16.39) for i~mJrn(it).

18.2 Problems for Chapter 18

1. Derive the recursion relations in Problem 16.4 from (18.47)

2 . Show that, for m > 0, the differential operator in (18.49) acting on Jo(kp)
satisfies
JL
m

(kp'£t)-
-A Y Jm<f> jp
ik dx ^ %d y )
where F is a differential operator constructed from kp and d/dkp.

3. The modified Bessel functions 7^, K v satisfy the differential equation

u" + - u ' - ( l + u = 0.
X y X J

Determine the leading asymptotic behavior, given in (18.79) and (18.78),

directly from this differential equation. Because

/j/(<c) — 1 <7|/(23?),
K u(x) = ^ íi/+ 1 (J í/ (¿x ) -f iNv{ix)),

show that this asymptotic behavior is consistent with that found in Prob­
lem 16.10, which holds for complex x ) \arg#| < 7r.

4. From (18.61) and (18.65) derive recursion relations similar to those in

Problem 16.4 for Im and (—l ) m7im.

5. Show that K m(t) defined by (18.71) continues to obey the appropriate

differential equation for nonintegral values of ra.
Chapter 19

Cylindrical Conductors

19.1 Rectangle
In a step beyond the parallel conducting plates of Chapter 17, we consider
another set of plates that intersect the first set at right angles, thereby producing
a cylindrical region with a rectangular cross section. The latter is displayed in
Fig. 19.1 along with a convenient coordinate system. Again, we seek Green’s
function for a vacuum region,

Now we begin with both (17.60)

(19.2)

Figure 19.1: Cylindrical region created by intersecting perpendicular conducting

planes.

205
206 CHAPTER 19. CYLINDRICAL CONDUCTORS

and
c 2 tti'k . mir ,
- y )= £ X u sin~ y sin~ y >
v — ' .

°{y (19.3)
m=1
combining them into the two-dimensional delta function

S(x - x')6(y - y') = j/)- (19.4)

lm

The complete set of functions

rmr .
h m{x ,y ) = ^ j\ s m l^ x y j \ s m — y, /,rn = 1 ,2 ,3 ,..., (19.5)

possesses the orthonormality property:

pOL rh
pb
I dx I dy<j>lm(x,y)<f>,'iml( x } y) = 6w 6mm,. (19.6)
Jo
to Jo
The related construction of Green’s function,

G (r ,r ') = 4ir'^2<t>im(x,y)<t>im(x' ,y')gim (z,z'), (19.7)

lm

then supplies the one-dimensional differential equation

52
+ 7lm) 9lm{z, Z') = S(Z - z'), (19.8)
dz 2

in which
/7T
Tim = (19.9)

We know the solution of (19.8) for a cylinder of unlimited length [see (13.24)],

9 im (z,z') — 1 ep-Tlml*-*'!
= ------ (19.10)
2tí„

When |z — is large, on the scale set by the greater of the rectangular dimen­
sions, it is the / = m — 1 term that dominates (19.7):

1 -yulz-z
G (r ,r ') ~ ^ u i x y y W n i x 1,!/) (19.11)
2th
Suppose, now, that conducting walls at z = 0 and c truncate the infinite
cylinder, thereby encasing the unit point charge in a rectangular conducting
enclosure or cavity. The solution of (19.8) under such circumstances is given in
Chapter 17. But there is a more symmetrical procedure here; begin with the
three-dimensional delta function construction

<5(r - r') = y^ im n (r)0 ;m n (rO , (19.12)

Imn
19.2. ISOSCELES RIGHT ANGLE TRIANGLE 207

y
T

Figure 19.2: Bisection of a cylinder with square cross section.

where the complete set of functions

*"»„<r) = VüsmT* V 6Sm~ y v csmT

. Í2 . hr Í2 . m ix Í2 . hr
(19.13)

obeys
J (^r ) ) — b\\i8mmi8nni. (19.14)

Green’s function is then given by

G (r ,r ') = 4tt^ ^ t l l^ m n í r ') ^ (19.15)

Imn "y ,m n

in which

o fh r \ 2 ( m ir \ 2 ( n ix \ 2 /ir\2 /ir\2 /ir\2

= +(— j +(T

19.2 Isosceles Right Angle Triangle

The situation in which the rectangular cross section of the cylinder is a square
(b = a) has only one special feature, the symmetry

Itm = ( J ) ( /2 + ™2) = 7ml- (19.17)

Yet this is at the heart of the possibility of constructing a new cylindrical shape
by applying the sort of symmetry consideration developed in Chapter 17. As a
glance at Fig. 19.2 indicates, Green’s function for a cylindrical conductor with
the cross section of an isosceles right angle triangle will be produced if we satisfy
208 CHAPTER 19. CYLINDRICAL CONDUCTORS

the boundary condition of vanishing on the diagonal line y = x. And that will
be accomplished by introducing an image charge through reflection in this line.
If the positive charge is stationed at the point with coordinates x ' , y' , the
negative image charge produced by the interchange of the x- and y-axes is found
at yf, x * . Thus the two-dimensional charge structure is

8(x - x')6(y - ] / ) - 6(x - y')(y - x') =

E
. hr . mir . Í7T . . m7T , /7T , m7T .
sin — x sin -----y sm — x sin -----y — sm — y sin ----- x
a a a a a a
lm
- y')> (19.18)
/< m

where now

2 . / tt . mw . m7T . /7T
<t>lm(x,y) — sin — x sin -----y - - s m -----x sm — y (19.19)
a a a a a

a function that reverses sign on interchanging / and m, and on interchanging x

and y. With l < m, these functions constitute a complete orthonormal set over
the area of the isosceles triangle. And the significance of the symmetry (19.17)
is now seen; both parts of (19.19) are associated with the same value of 7 , which
is therefore assigned to the complete function,

id 2 d2 \
- 2 + ^ 2) = y?m<l>im(x,y), (19.20)

where (/ < m)
(19.21)

19.3 Equilateral Triangle

Some of the geometry of an equilateral triangle is displayed in Fig. 19.3; a is
the common length of the sides, the length of the perpendicular from any apex
to the opposite side is
. it \/3
n — a sin — = ——a, (19.22)
3 2
and the radius of the inscribed circle equals

1 » 1 1 .
r = -a ta n — = — -=a = - a . (19.23)
2 6 2i/3 3

We employ two coordinate systems in the transverse plane perpendicular to

the axis of the triangular cylinder. One is the conventional rectangular frame—
coordinates x and y— with its origin placed at the center of the inscribed circle.
The second is a trilinear coordinate system based on the three unit vectors
19.3. EQUILATERAL TRIANGLE 209

ei

Figure 19.3: Cross section of equilateral triangular cylinder.

directed from the center of the inscribed circle to the three apexes. These three
vectors are presented in terms of the unit orthogonal vectors i and j by

y/3. 1. y/S. 1. ______

(19.24)
2 1 2J’ 63 2 1 2J’
where
3

1 > = °- (19.25)
a—\
One notes that
3 ... ... 3
= 2 (U + U) = 2 l x ’ (19.26)

from which the trilinear coordinate representation of rj. is derived as

2 3
r x = l x t x = - J 3 e a/xa, (19.27)
<2=1

where
Ma = e a • r j . . ( 1 9 .2 8 )

These coordinates, which obey

3
£ > « = 0, ( 1 9 .2 9 )
a—\

are given individually by

\ /3 1 \ /3 1
/¿i = 2/, = — Y x " 2 V) ~ ~ Y X ~ 2y' ( 19-30)

The trilinear coordinates of a point are the perpendicular distances from the
origin to the three lines, drawn through the point, parallel to the sides of the
triangle. A coordinate is negative if the associated line lies between the origin
and the related side. Thus, the three sides of the triangle are represented in
trilinear coordinates as /iX = —r, ^2 = —r, and ^3 = —r, respectively.
210 CHAPTER 19. CYLINDRICAL CONDUCTORS

Figure 19.4: Reflected equilateral triangle.

The potential that is Green’s function can be extrapolated in a way similar

to that of the parallel plates, by reflection in the three sides of the equilateral
triangle. It is thereby initially defined in a larger region (see Fig. 19.4), one that
is an inverted triangle of side 2a. The repetition of this process for the bigger
triangle produces an enlarged copy of the initial triangle, with side 4a [Fig.
19.5]. Having been generated by two reflections, the values of Green’s function
within the three triangles, of side a, that are produced by a displacement of 2h
in each of the directions e a, a = 1,2,3, duplicate those in the initial triangle.
In short, the indefinite repetition of this reflection process supplies a function
defined over the entire two-dimensional space, a function that is periodic in each
of the three directions, with period 2h.
We now construct a set of point charges with the same periodicity. A first
impulse might be to locate them at positions along the three trilinear directions
19.3. EQUILATERAL TRIANGLE 211

according to
3
r'± + 2 h Y , » ae a, va = 0, ± 1 ,. (19.31)
a=l

That would be redundant, however, in view of the relations given in (19.25).

Indeed,
3
y ; ^ e a = (^1 ~ ^ 3 )e i + (i>2 - ^ 3)^2 = (^2 ~ ^ i)e 2 + (^3 — ^ i ) e 3
a=l

= (^ 3 - ^ 2 )e 3 + (^1 - ^ 2 )e i, ( 1 9 .3 2 )

which means that any pair of trilinear basis vectors will do (no surprise in a
two-dimensional space). In actuality, then, we extend <5(rj_ — r±)> the initial
point charge density (in the transverse plane) to

y 6(r± - r'x - 2 h(is2e 2 + ^ e a ))

^ 2 ) ^ 3 = — OO

(<fltx)c<kx . (r-r')x ^ e - i 2hv2k2e-i^hu3k3 ^ (19.33)

J (2ir

which introduces some of the trilinear coordinates of k x ,

¿ a = ea *kx, = 0. (19.34)

All of these coordinates are employed in writing

2 _r
(19.35)
k± •(r - r')x = kx •lx •(r - r')x = ñ X ) ka^ a ~

The i>2 and 1/3 summations in (19.33) are evaluated as [(16.63) and (17.56)]

1 00
—i2hvzkz
^ E
-i2hV2^2
(2tt) v3zz—oo

1 w TT
= (2*)i E
V 7 / 2 = — OO
E
/ 3 = — OO
(19.36)

which tells us that the ka have the discrete values

3

*« = £*«, J > = o. (19.37)

a=l

Let us also record the discrete values of

2 r ^ l9 2
=iE*J-i(s) E«- (19.38)
212 CHAPTER 19. CYLINDRICAL CONDUCTORS

Now we must compute the integral

J (d k ± ) S(k2 - *h/h)6(ka - xla/h), (19.39)

which is
J dk2dkz6(k2 — Trl2/h)8{kz — xh/h) = -^=, (19.40)

according to the value of the Jacobian determinant [Karl Gustav Jacob Jacobi
(1804-1851)] that relates the integration variables k2, k3 and kX) ky:

d(k2lk3) _ —a/3 /2 —1/2 V3

(19.41)
d(kx ,ky) ~ VZ/2 - 1 / 2 2 '

The outcome for (19.33) is [recall that a = 2h/y/Z]

Y s(r x - r'x - 2h(v2e 2 + v3e z))

V^Vz

=—
3ha Y
exp
.27r ^
a
i
exp
a
(19.42)

in which the /-summation extends over all integral values of the la that satisfy
the restrictive condition (19.37). It is worth noting here that a displacement of
rj_, for example, by 2Ae&, 6 = 1,2,3, alters the trilinear coordinates j±a by

/¿a —►Va + e a •e b2A, (19.43)

where, through inspection of (19.24) or otherwise,

_ 3 1 / a= b: 1 ( .
ea . e j - 2 ¿a6 2 _ \a#6:-l/2' ( 1944)

Accordingly, any exponential function of the (ia in (19.42) changes by the factor

.27r
exp /»2fc+ £ ( - / « / » ) = exp(i27r/6) = 1. (19.45)
l Zh
a?b

The next step supplements the lattice o f positive charges with a negative
charge distribution that is designed to satisfy the specific boundary conditions on
the three sides of the triangle. For that purpose we first observe that the image
of r'± in the triangle side perpendicular to ei has the rectangular coordinates

x' = y' = - y ' - 2r, (19.46)

the latter being more transparently written as

y' + r = - ( j / + r). (19.47)

19.3. EQUILATERAL TRIANGLE 213

And more generally, the image in the side perpendicular to ej is

r'x = r'x - 2e&(e¡, •rx + r), (19.48)

which appears in trilinear coordinates as

K = K + Vh + r ~ 3M / 4 + 9 = | “ ^ I ; ^ ’ (19.49)

where c designates the trilinear axis that is neither a nor 6.

Let us select a particular exponential function of the fi'a in (19.42), supply
it, for convenience, with the numerical factor exp(/27r/i/3), and then subtract
from it the result of replacing //*a by using the example of reflection in the
side perpendicular to e i. This yields the combination (recall 3r = h)

2tt Oír
exp + ^3/^3 ” hh) — exp * + h ¿4 + — hh)
(19.50)
which, from its construction, vanishes when y!x = —r, for that is the line in
which reflection takes place. (See also Problem 19.1.) This property, being
independent of the particular choice of the /a, continues to hold when the indices
of the la are cyclically permuted: 1 —^ 2, 2 —^ 3, 3 —^ 1, thereby producing

9 7T .27T
exp —*3A — exp
^3^2 + “ hh) *3¿ (h u í + ¿3/4 + ^1/4 ~ ^A)
(19.51)
and
27T Oír
exp 3^03^1 + il /^2 + *2/4 - *3*0 exp ^ ^2/4 feft)
(19.52)
Furthermore, the sum of the three pairs, which of course also vanishes for
fi\ = —r, responds in a simple way to cyclic permutations of the ¡Ja indices.
One can check that the substitution //2 —► ¿4 —* M reproduces
the set of six terms, multiplied by the common factor

2TT . 2TT . ’ 27T,, . .

exp = exp = exp - * y ( / 3 - / i )
to"“

tT
1

1
1
s.

HJ
1-----

H!

the equivalence of these forms follows from the relations

(/1 - h ) - (/2 ~ h ) = - 3 / 2, (l2 - h ) - (/3 - h ) = —3/2. (19.54)

And the implication of this property is that as the complete structure of six
terms vanishes for = —r, so does it vanish for fi2 = —r, and for /4 = —r; it
obeys the boundary conditions.
In an extension of what could be done without comment in (19.18), we
remark that the six term structure is either unaltered or reverses sign under
the six operations of cyclic permutation of the la (three in number, including
 214 CHAPTER 19. CYLINDRICAL CONDUCTORS

the identity), and anticyclic permutations combined with sign reversal (three in
number), as illustrated by /i —►—/ i , ¡2 —* —/3 > h —» —h • These operations,
applied to an exponential function of the fia in (19.42), produce an analogous
six term structure, thereby presenting us with the final generalization of the
initial transverse delta function:

6( * x - r'x) -*• (19.55)

I
where

(3fca)1/3*»(/0
27T 27r
= exp + hP2 + ht*3 — hh) — exp
~ 23fi(^l^ 1 ^ 2 ^ 3 ~~
” 27T 27r
+ exp — exp
23 ^ 2 ^ 3 ~~ ^2^ 3ft(^2^i ^ 2 ^ 3 ~ ^2
2 tt 27t
+ exp i7rr{hpi + I1P2 4- ht*3 — hh) — exp
oh
(19.56)

The /-summation is extended over all sets of the la that are distinct with respect
to cyclic and anticyclic-reflection permutations. But one must not overlook the
fact that vanishes if any la = 0 (this is verified in Problem 19.2); the
possible values of la are ±1, ± 2 , ___Such are complete, and orthonormal
over the equilateral triangle, in the sense [analogous to (16.29)] that

/ (drx)<Mrx)’V /'(rx) = h>- (19.57)

Inasmuch as the functions <^¡(rj_) have their origin [(19.33)] in the exponen­
tials exp[?’k_L •rx], they obey the differential equation

-Vx<£( = 7? <t>u (19.58)

where [(19.38)]
(19.59)

is indeed a quantity associated with the set /, being unaltered by cyclic and
anticyclic-reflection permutations of the la. The smallest value of 7 2 is realized
for the /-set indicated by { 1 , 1 , —2 },

7? > ( x ) 2- ( 19-6°)

The associated function, multiplied by i [which leaves intact its contribution to

(19.55)] is the real function

2 . 2w,
<MaO (19.61)
(3 h 7 a =l
19.4. CIRCLE 215

Its complete symmetry in the three trilinear coordinates is consistent with the
general observation [(19.53)] that <f>](fj), with any two la equal, is unchanged by
cyclic permutations of the /ia. [Note that the two other possibilities, multipli­
cation by exp(±¿27r/3), are not options for a real function.]
It has been remarked that the </>i(fi) are normalized functions in the sense
of (19.57). We rise (partly) to the challenge of verifying this for the simplest
function, <¡)q(/¿). That presents us with the integral

+ X ! sin + sin +
a^b '

_ . 27r/ x . 2tt .
+ 2 sin — (/ii + r ) sin — (/i2 + r ) j. (19.62)

The second version exploits the threefold rotational symmetry of the equilateral
triangle; the three integrals involving fxa are independent of the particular choice
of direction, and the six integrals containing fia and fib are the same for all
pairs of directions. Now the latter trigonometric functions within braces can be
presented as
1 1 47T/ x 27T 27 r ,
2 - 2 cos ■y (a41 + r) “ cos y (A41 + r ) + COS — { m - /12), (19.63)
which also uses the equivalence under integration of —fi\ —/i2 = fi% and fi\. It is
left to the reader [Problem 19.3] to prove the null value that the area integrals
of these cosine functions possess. Then the outcome is just

(19.64)

which finally involves the area of the equilateral triangle. (See Chapter 25 for a
simpler treatment.)

19.4 Circle
Parallel plate, rectangle, triangle— we come to the circle, which is to say, the
circular cylinder. Let it be of radius a, with its axis identified as the 2 -axis of
a circular cylindrical coordinate system. The potential of a unit point charge
in vacuum within a conducting cylinder at zero potential— Green’s function— is
the sum of the Coulomb potential of the point charge and the potential of the
surface charge distribution:

G(r,r') dSi
F Lí i + / s | r- rx|’
ron S : G(r,r') = 0. (19.65)
216 CHAPTER 19. CYLINDRICAL CONDUCTORS

We make use o f the Coulomb potential construction (18.33) and the addition
theorem o f (18.64) in writing
-I -| pOO 00

-Í— = - / dkzeik«<*-*'> V e<m^ - ^ ') / m(|Jb,|p< ) i f m(|*,|p> ). (19.66)

|r “ r I » -/-oo

Then the integration over the cylindrical surface in (19.65), where

dS\ = dz\ a d(j) i, (19.67)

introduces the following measure of the surface charge density,

/ OO /*27T

dz\ a I d<j>ie~tk‘ z' <r(ri). (19.68)

■ oo •/ 0
With this notation, the potential produced by the surface charge density is
(* = N )
-I poo 00
- / dlfe.e**" V eim^ Im(kp)K m(ka)<rm(kz). (19.69)
* -/ -o o m“ oo

The boundary condition on Green’s function requires that the surface charge
potential annul the point charge potential on the surface S (p ~ a, z : —oo —►oo,
<j) : 0 —►27r). In view of the completeness of the functions exp{ikzz) exp(¿m<^)
over this domain, that null statement must apply to every value of kz and m (a
null function is uniquely represented by zero coefficients). Thus, we have

0 = e - ik‘ M'e - im+'lm(kf/)Km(ka) + Im(ka)K m(ka)crm(kz), (19.70)

or
<rm(kz) = (19.71)
Im(k(i)
The result is presented, in the notation

/ OO rib 1 °°

¿ />';*), (19.72)
■ °° Til— — OO

where the reduced Green’s function is

9m{p) p >&) — Im(kp<) K m(kp>) - (19.73)

-tm(^íí)
Another derivation employs the differential equation
1
—V 2G (r ,r /) = 4tt6(v — r') = 4tt- 6 ( p — p')S(</) — ^ ¿ ( z — z'). (19.74)

The introduction of the construction (19.72) then yields the radial differential
equation satisfied by gm,

L IA f A ^ i + fe2 H— 2" 9m(p, p f ] k) = - 6 ( p — p ') . (19.75)

. pdp X dp)
19.4. CIRCLE 217

It is to be solved under the general requirements of finiteness and continuity

for gm; the discontinuity of the derivative of gm implied by the delta function
(multiply by p before integrating):

-p 'S -9 m (P ,p ';k ) = 1; (19.76)

dp n—o' —0

and the boundary condition

gm{a,p '] k) = 0. (19.77)

The homogeneous differential equation that applies for p / pf identifies the

solutions as Bessel functions of order m and imaginary argument, the functions
Im(kp) and K m(kp). First, we recognize, from the small t behavior of Ko(t)
[(18.25)] and the general construction of (18.65), that the K m{t) are singular at
t = 0. Accordingly, for p < / / , a domain that includes p — 0, only Im(kp) is
acceptable:
p < p' : gm(p, p'\ k) = A Im(kp). (19.78)
As for p > / /, where both Im and K m can enter, the requirement that gm vanish
at p — a picks out a specific linear combination [we see it in (19.73)]:

Km(ka)
P > p' • 9m(p,p'\k) = B Rm(kp) Imikp^j (19.79)
Im(k(l)

Continuity at p = p' is satisfied by writing

A = C [ffm( V ) - Im(kpr) ^ ^

B = C Im(kp‘ ). (19.80)

And, finally, the discontinuity condition (19.76) tells us that

CkpV<rn(kp')l'm(kp') - Im(kp')K'm(kp')\ = h (19.81)

where the prime on the function denotes a derivative, for example,

(19.82)

We meet here an example of the Wronskian [Hoené Wronski (1778-1853)],

characterizing two independent solutions of a differential equation. It is suffi­
cient to write that equation as

£ (19.83)
dt

from which it follows that

£ = 0, (19.84)
dt
218 C H A P T E R 19. C Y L IN D R IC A L C O N D U C T O R S

or
t [ K m (t)I'm ( t ) - I m ( t ) K ‘m ( t ) ] = 1; (19.85)
the value of the constant on the right hand side is supplied by an inspection of
the asymptotic forms of the functions, (18.78), (18.79). Therefore C is unity.
When the forms for p < pf and p > pf are united we regain (19.73).
Incidentally, the Wronskian appears explicitly in verifying that the surface
charge density described by (19.71) is what it should be in terms of the normal
component of the electric field:

(19.86)
,T(ri)=¿ ¿ G(t't') p - a
The introduction of the representation (19.72) for G, along with the definition
of crm{kz), (19.68), converts the above relation into

Gm{kz) — C (19.87)
P~a

Now, according to (19.73),

Itn jk pf)
a-i-9m {p,p']k) = Im{kp') ka
dp P~a Irn{ka) ’
(19.88)
on applying the Wronskian relation (19.85). The result for am{kz) reproduces
(19.71).
This reference to the surface density of induced charge naturally raises the
question o f its total amount. That is

/
oo p2 tt
dzi a / d<j>i <r(ri) = <ro(0) = -1 , (19.89)
-oo J0
according to (19.68) and (19.71), along with the fact that I o ( t ) is unity at t = 0.
We ask another question: What is the potential generated outside the radius
p = a by the combination of the internal point charge and the surface charge
distributions? A glance back at the forms appropriate to p > a will show
that the significant combination of the two constituents is just what appears
in (19.70), with the common factor K m{ka) replaced by K m(kp). In short, the
potential for all p > a is zero; the world outside the cylinder is unaware of the
balanced distribution of charge within.
Now let us turn back to the parallel conducting plates in order to develop
a useful correspondence with the circular cylinder. We recall from (17.13) that
the reduced Green’s function for the former pase is
sinh kz< sinh k(a — z>)
9 (z,z'\k) (19.90)
k sinh ka
or, equivalently,

g (z) z*; k) — —sinh kz< e kz> — sinh kz> (19.91)

k sinh ka
19.4. CIRCLE 219

Notice the close analogy between the pairs sinht, e_< and 7m(A), K m(t)> re­
spectively. There is also an analogous Wronskian relation, appropriate to the
simpler differential equation,

sinhJ —sinht-7 -e~* = 1. (19.92)

at at

All this leads us to search for an analogue of the alternative form of g {z, z'\ fc)
that is presented in (17.77). We stress that the statement of equivalence in the
latter is an identity, holding for all values of k, real and complex. As shown in
Problem 17.3, both sides of this equation have the same singularity structure,
poles along the imaginary axis at k = ±i(n7r/a) (n a positive integer), with the
same residues, and vanish asymptotically as k —►oo. The proof o f equivalence
of the two sides of (17.77) follows from the observation that, as a function of the
complex variable Ar, the difference between the two sides is everywhere regular
and bounded, which identifies it as a constant, according to the theorem of
Cauchy, ascribed to Joseph Liouville (1809-1882), said constant being zero in
virtue of the asymptotic behavior.
With this analogy before us, we now look back at gm{p, 7/; Ar), (19.73), and
recognize that it must be given equivalently as the sum over all the pole terms
that are associated with the zeros of Im(ka). Again, as with sinh Ara, these occur
for imaginary values of k [recall (18.58)]:

ka = dtiymn > Jmisimn) — 0, Tl ~ 1 , 2 , . . . , (19.93)

in which n numbers the successive positive zeros of Jm{i). In order, however,

not to rely entirely on analogy, we supply an explicit proof that the zeros of
Im(ka) occur only for imaginary values of k. Let us consider complex values of
k and begin with the differential equations obeyed by Im(kp) and its complex
conjugate, Im(k*p):

1 d d o m 21
~ — ImiJ^p) — 9?
p idp
~ pidp k2
' 7 j
Id d l*2 m2
7m(Ar>) = 0. (19.94)
P d>P dp

Now, as a generalization of the Wronskian relation, we multiply each equation

by the other function, along with a factor of /?, and subtract the two, finding
that

L . ’ Im(k *p )p ^ -Im(kp) - Im(k p )p -L lm(k*p)

dp
= (k2 - k * 2)plm(k*p)lm(kp). (19.95)

Then we integrate over /?, from 0 to a, under the assumption that

7m(A?a) — 0, Im(k a) — O? (19.96)

220 CHAPTER 19. CYLINDRICAL CONDUCTORS

these being mutually complex conjugate statements. That yields

(k2 - k*2) f dpp\Im(kp)\2 = 0, (19.97)

Jo
which permits only one inference, k2 — k*2 = 0, or

¿fe* = ¿fe; ¿fe* = -¿fe. (19.98)

The option k* = k is not realized, however, for Im of real positive argument is

everywhere positive— it has no real zeros except possibly for zero, as may be
seen from the power series expression (18.66). Thus, k is imaginary, as stated
in (19.93).
With k in the infinitesimal neighborhood of ±¿7mn/^, we have

Im(ka) ~ [ka - (±iymn)] I,m(±iymn)- (19.99)

Then we refer to the Wronskian relation (19.85), for Im(t) = 0, to learn that

1 1
-^-m(d:¿7mn) = ~7~' r/ / i • , \* (19.100)
Tran)

Finally, with due attention to the various powers of i [(18.58)], we arrive at

j , . / i t i\ J m jjm n p/a )J m {7m n P /a ) 1___________^

k ~ ± i y mn/a: gm( p , p , k ) ~ [J^(ymn))2 ± iy mn ka T iymn ’
(19.101)
where the sum of the two complex conjugate pole contributions results in the
following multiple for the Bessel function factor:

(19.102)
k2a2 + 7mn'

The sum over all such pairs of poles then yields the desired construction

/ / I\ P m n (p )P m n (p ')
*»(/>, p ;* ) = ! , mn/a2 -
k2 + J2 (19.103)

in which
y/% J m { y m n p / a )
(19.104)
p / \

mAP) ~ a J U 7mn) ’

The latter function obeys the Bessel differential equation

(19.105)

and the boundary condition

Pm n (^) —■0. (19.106)
19.4. CIRCLE 221

On applying the differential operator in (19.75) to the construction of (19.103),

the denominator is cancelled, leaving us with

1
-S (p - p ' ) = Y l Pmn(p)Pmn(p'), (19.107)
P n= l

the statement of completeness of the functions Pmn{p) over the interval from
p zz 0 to p = a. These functions are also orthonormal in the sense that

Í d p p P m n ( p ) P rnn , ( p ) = fifin' • (19.108)

J O

Although this property is quite analogous to those encountered in earlier ex­

amples of orthonormality, we provide a proof, one that begins with the explicit
statement of completeness,
oo „a
f(p ) = ^ > m » ( p ) / dp'p'Pmn{p ')f{p '). (19.109)
n= 1 do
We choose f(p ) to be Pmn(p) and present the result as

^ ^ Pm n ( P ) i Í d p ' p ' Pm n ( p ' ) P m n > ( p f ) - Sn = o. (19.110)

Uo
Then we assert that there is no such linear relationship among the P m n ( p ) ) ex­
cept the one with completely null coefficients, which indeed yields the orthonor­
mality relation (19.108). To provide a basis for that assertion, consider such a
null combination of N different P m n ( p ) functions, where N can be arbitrarily
large, but is finite in that we can identify a particular P m n ( p ) as possessing the
largest value of 7 mn in the set of N such numbers. Thus, we accept that the
members of the set can be ordered in the sense of increasing values of 7 , and
thereby labeled from 1 to A . Now the functions Pmn(/>) obey the differential
equation given in (19.105). Let us apply, to the sum of N terms, the differ­
ential operator associated with the largest value of 7 . That will remove the
TVth term and multiply all the other coefficients by positive numbers. Proceed
similarly for the ( N — l)th term in the sum of N — 1 terms, and so on until one
reaches a statement containing only N = 1, which declares its coefficient to be
zero. Then read back along the list of relations with 2, 3, .. .terms to conclude
that all coefficients are zero— nontrivial linear relations do not exist. Of course,
the orthonormality property can also be derived using the differential equation.
That is left to the reader in Problems 19.4, 19.5.
What are the values of 7 mn? The starting point is the differential equation
(16.37),

( ¿ + 1 - m-2 ~ — ) = 0, (19.111)

and the asymptotic behavior for large t , (16.39),

Jm(t) = cos (t - (m + 1 / 2 ) - J . (19.112)

222 CHAPTER 19. CYLINDRICAL CONDUCTORS

We use the latter as a guide to writing, generally,

/ / X 1/ 2
i y j Jm(t) = A (t)c o s $ (t). (19.113)

Introducing this into the differential equation, and setting the coefficients of
cos 4> and sin $ equal to zero, yields

^ _ ( ^ + ( , _ 2 ! ! l l i Z i ) A = 0,

2A'<$>' + A S " = 0. (19.114)

The latter equation, multiplied by A, then supplies the information:

j t {A 2& ) = 0, A 3$ ' = l, (19.115)

where the stated value of the integration constant is provided by the asymptotic
form,
t \m\ : A (t) ~ 1, <&(t) ~ t — ( m 4- (19.116)

We now use the relation of (19.115), along with

m 2 _ 1 ah

^ i2 = 1 - - W ± + T ’ (19-117)
in a sequence of approximations that begins with (19.116) as the first approxi­
mation. For the second approximation we use A « 1 and t \m\ to get

(19.118)

and
<!>(< ) « t - ( m + l ) ! + l . (19.119)

Then (19.115) supplies the second approximation to A:

(19.120)

To go on to the third approximation, note that

A" 3m2 - |
_____ ______________ÍÉ_
(19.121)
A t4
and then

1 m2 - i 1 (m2 - l ) 2 3 m2 - j
(19.122)
t2 8 t4 + 4‘ t4
19.4. CIRCLE 223

leading to

. l\ 7r 1 m2 — 4 1 (m 2 — 4 ) 2 1 m2 — 4
* ( ‘ ) M<- ( r o + 2 ) 2 + 2 — f ^ + u — t^ L - 4 - i r L+- - <1£U23>

The roots of the Bessel function Jm(¿), the zeros of cos<$(tf), are inferred
from
/ 1\
$ ( 7) = ( n _ I ) n = 1,2....... (19.124)

Accordingly, the successive approximations to $>(t) provide approximations to

7 (m and n understood) as follows,

1 ^
1 : 7 = + 2n —
2) ^ 2 ’

n i ^ m2 - \
2 : 7 = + 2n — (19.125)
2; ^ 2 m + 2n — \ ’

3 : 7 = + 2n —
2]^ 2 ' m + 2n — \
m'
2_ 1
1 ___ 4 7 ( TXl — - 7 I — 6
37T3 (m -f 2n — 1/2)3

The smallest value of 7 appears for m = 0, n = 1. The values produced by

the successive approximations are

1: = 2.3562,
*47T 1
(19.126)
2: T + 6^ = 2-4092’
3tt J _ 31
= 2.4031.
4 67T 162tt3
Compared to the actual value

701 = 2.4048255577, (19.127)

these are in error by —2.0%, 0.18%, and —0.07%, respectively. Turning to

m = 0 , n = 2 we get
7tt
1: = 5.4978,

2: X + T - = 5.5205, (19.128)
4 147r
3: — + —---------- = 5.52004,
4 14x 2058tt3
which are much more accurate approximations as compared with

702 = 5.5200781103, (19.129)

224 CHAPTER 19. CYLINDRICAL CONDUCTORS

which represents errors of —0.4%, 0.008%, and —0.0007%, respectively. As a

last numerical example, consider m = 1, n = 1, where
57T
1: = 3.9270,
T
57r 3
2: = 3.8315, (19.130)
T 10tt
57T 3 3
3: = 3.8319.
107r 2507T3
Relative to
7n = 3.8317059702, (19.131)
these are in error by 2.5%, —0.005%, and 0.005%, respectively.
We have met two statements of completeness for functions of the variable />,
(16.59) and (19.107):

1 f°°
-6 (p -p ') = / d k k J m(kp)Jm(kp')
P Jo

2 Jm(ymnP/Q^JmiSfrnnP I a) (19.132)
= £ a2 [^ m (T m n )]2

The first of these is valid for all / ? , / / > 0; the second one applies with 0 < p, p' <
a. Does the latter yield the first version in the limit a —> oo? Indeed. We begin
by remarking that, for any appreciable value of 7mn/u> which becomes k in
the limit, ymn is arbitrarily large. That, in turn, implies correspondingly large
values of the integer n, which is to say that the range of values— the spectrum—
of 7mn/u —►k is effectively a continuum. In view of these very large values of n,
the first approximation of (19.125) suffices, and the asymptotic form (19.112)
can be applied in evaluating J^(7mn)> where

1\ 7r
7mn = n7T + m (19.133)
2 /2

Accordingly, we have

(19.134)

and the summation in (19.132) becomes

^ ^ ~7mn JmniCímnP/ a)Jm{7mnP / a)* (19.135)

n

Now we have only to recognize [(19.133)] that w/a is the interval between suc­
cessive n-values of 7 m„ / a , for then, with the replacements

J in iL ^ k, — —*•dk, (19.136)
a a
we realize the integral form of (19.132).
19.5. CIRCLE AND SEPTUM 225

Figure 19.6: Cylinder with septum.

19.5 Circle and Septum

To illustrate a class of cylindrical conductors related to the circular cylinder, we
introduce into the latter a radial septum, a conducting partition of negligible
thickness, as illustrated in Fig. 19.6. Now Green’s function must vanish at
4> = 0, the top face of the partition, and at <j) = 27T, the bottom face. That kind
of situation is already present in parallel plates, and we have only to transcribe
(17.60) with 2 , z' —* <^, ft and a —» 2tt (we also change n into ^), to arrive at the
proper replacement for 6(<j) — ft):

1
8(<j> - <j>') - - y ' s i n ^ s i n ^ ' . (19.137)
7T z ' Z Z
v—l

Indeed, the function sinm<^, which vanish at (j> = 0, will also vanish at <¡> =
provided

2m = v = 1 , 2 , 3 , 4 , . . . , or m = i t / = 1, 1, 2,.... (19.138)

Thus, in addition to integral values of m, excluding zero, we need integer plus

1/2 values, beginning with m = 1/2.
The appropriate modification of the Green’s function construction (19.72) is

/
OO J jL 1

z_eikt(z-z )__ ^ sin m<l)sin m ftgm(pyp'; fc), (19.139)

■°° m—u/2

and the differential equation obeyed by gm is just (19.75), with m assuming

the range of values in (19.138). It is to be emphasized here that in writing, for
example,
( V l + k2)eim*Jm(kp), (19.140)
there is no inherent requirement that m be an integer— that is a matter of the
boundary conditions. Which is to say that the constructions of gm for the circle
and septum, whether in terms of Bessel functions of imaginary argument or,
alternatively, by means of Bessel functions and their zeros, have exactly the
226 CHAPTER 19. CYLINDRICAL CONDUCTORS

same appearance as for the unobstructed circle, only the implicit reference to
the spectrum of m indicating the change. That directs our attention to what is
new, the values m = 1/2, 3/2, . . . .
As we know, the differential equation for Jm(t) can be presented as

£_ m2 — 1/4
+ 1 - ) = 0. (19.141)
dt2 ¡5

A striking consequence emerges for m = 1/2; we immediately get exact solu­

tions: sintf, cost. The latter possibility is rejected in order to achieve finiteness
of at t = 0,

sini=V^rfj
/ 2 \ 1//2 / 2\
< /i/2 (0 = (^ J cos(< — ít/2 ), (19.142)

where the numerical factor is chosen to conform with the here everywhere valid
asymptotic form (19.112). All the roots of this Bessel function are given exactly
by
yn = n7r, n = l , 2 , . . . , (19.143)
which is also displayed in the three identical statements of (19.125) for m = 1/2.
Beginning with this simple form for J 1 / 2 , we now proceed to construct J3/ 2 ,
J 5/ 2 , ___ In doing this, we apply (18.47), which continues to hold for non­
integral values of m. As we have mentioned, exp(im(j>)Jrn(kp) is a solution
of the differential equation in (19.140), irrespective of the choice of m, and
any derivative of this function is also a solution of the differential equation.
Furthermore, the combination of derivatives employed in (18.47) is such as to
introduce the additional exp(±i<j>) dependence:

(dp
±+-± e*** A ± £ A j (19.144)
dx dy pd<j>)

In using the relation (18.47), we shall find it convenient to write

eim* = (e**)m = , (19.145)

while noting that

( ± iy^ = ^ iy^ + ± iy) = (19.146)

Accordingly, on choosing the upper sign of i, we get

-<■+ w i ( s + ^ + <19-147>
where, in view o f (19.144) and (19.145),

í d_ _0\ Jm(kp) _ x + iy d Jm(kp)

(19.148)
\ dx 1d y ) pm p dp pm
19.5. CIRCLE AND SEPTUM 227

This produces the relation

1 d Jm(t) Jm-fi {t)

(19.149)
~ t d t t™ ~
Its repeated use then yields (/ is a positive integer)

Jm+l CO 1 d \ Jm(t)
(19.150)
tm+l \ t dt J
A tm

We apply the latter, with m = 1/2, to get

1 d\ sin t
w o - d y v ^ - i i (19.151)

This general construction is illustrated by

1/2
<^3/2(0 = ( ^ ) ( - c o s < + “T " ) ~ ( ^ ) cos( * - ’r)> (19.152)

which also displays the asymptotic form. For arbitrary /, the function yj2ir/t x
cos(t — is multiplied by a polynomial in l/ t} of even degree, that begins
with unity, while (2'jrft)1^2 sin(t — (m -f ^ )~ ) is multiplied by a odd polynomial
in 1/t.
We can give the construction (19.151) another form by observing that

sint _ 1 f 1
dfi eZfit (19.153)
~ T ~ 2 j_ 1

Then we write
1 d sint
dfi n e
t dt t

(19.154)

The repetition of this operation produces

(19.155)

as we can check by doing it once more: / —»•/ + 1. As a result we see that

2V 2 . i+ x/2 1
1 (- d2 \ ’ s m t
228 CHAPTER 19. CYLINDRICAL CONDUCTORS

The equivalence of the two forms for / = 1 tells us that

( d2 2 d \ sint
(19.157)
\ w + t 7 t + 1 ) ~ r - 0-

More generally, the connection between J1+ 3/2 and Jj+ 1/2 that is inferred from
(19.156),
Ji+3/2(t) _ 1 A ^ + 1/2(0 ( .
ti+3/2 — 2/ + 2 \ dt2) f '+ !/2 ’ 119.15»;

combines with the implication of (19.150),

Jl+3/2(t) _ 1 d J/+ 1 / 2 CO
<1+3/2 ~ t dt tl+l/2 ’ (19.159)

to supply the differential equation

(d 2 21 + 2 d W , + i / 2(<)
(19.160)
\dt2 t d t' ) <1+ 1/2 - u -

This is, of course, a version of the Bessel differential equation for m = / + 1/2.
One can rewrite it by noting the symbolic relations

Id d l
dt^ dt t’

ti— t ~i - ( i - i y - H i . * ( * + 1) (19.161)
dt2 t) dt2 t dt t2

They lead to the differential equation

d2 2d /(/ + 1 )
f " 1/2 J/+ i/2(¿) — 0; (19.162)
dt2 t dt t2

we shall meet it again in Chapter 21.

19.6 Problems for Chapter 19

1. Show, explicitly, that the combination (19.50) vanishes on the line of re­
flection, fii = —r.

2. Prove that <¡>i{p) defined by (19.56) vanishes if any la = 0.

3. Prove that the area integrals of the cosine functions in (19.63) vanish.

4. Show that

d_ d d
pJm(kp)-7£-Jm(k p) pJmfó1
dp
= (k2 - k,2)pJm(kp)Jm(k'p).
19.6. PROBLEMS FOR CHAPTER 19 229

Use this to verify that

n 7^ n : Í dp p Pmn(p)Pmnt(p) = 0 .
Jo

5. Employ the first relation in Problem 4 to derive

■J- P2JL(kP)kJm(kP) ~ PJm(kp)-LpJ'm{kp) = 2kp[Jm(kp)]2.

Use this to verify that

Í dpp[Prnn(p)]2 1.
Jo

6. Again employ the first relation of Problem 4 , with kf = k*, to prove that
the roots of Jm are real.

7. Check the Wronskian relation between / m(tf) and K m(t), (19.85), using
the known t —> 0 behavior for m = 0 . Then, for m > 0 , use the known
t 0 behavior of Im(t) to deduce that of K m(t).

8. Apply an image method to find Green’s function for an infinite conducting

cylinder with cross section in the form of a semicircle of radius a. Get an
approximation to the smallest value of 7 in this geometry [the correct
value is 3.8317 ...].

9. A point charge e is situated on the axis of an infinite conducting circular

cylinder of radius a. Prove that the interaction energy with the cylinder
is
E'int
it a J0 Io(t)
or, equivalently (think of the Wronskian)

l e 2 f° ° di
Eint~ n a J0 U oW )2’

10. Find Green’s function for a unit point charge in vacuum outside an infinite
conducting circular cylinder of radius a.

11. Having introduced the Wronskian for K m and 7m,

W (K m>Im) = K m(t)I'm{t) - Im{t)K'm{i),

in (19.85), we can consider the corresponding quantity for the solutions of

Bessel’s equation of order v,

W( J V, NV) = Jv{t)N'v{t) - Nv{t)J'v{t),

where Nv was introduced in Problem 16.8.

230 CHAPTER 19. CYLINDRICAL CONDUCTORS

(a) From the differential equation satisfied by Jv and Nv [see Problem

16.7], show that
t W { Ju, Np) = constant.

(b) Show that the constant here is 2/7T.

12. Calculate Green’s function for Laplace’s equation,

—V 2G(r,r') = 47n$(r —r')

in the interior of a hollow conducting circular cylinder of radius a and

length L with conducting endcaps. Introduce a Fourier series in the di­
rection of the axis. Give an image interpretation of the result. Compute
the charge density on the surface.

13. Show that the orthonormality relations established in Problems 4 and

5 hold for nonintegral m. What further is required to prove that any
function f ( p) defined on the interval 0 < p < a can be expanded in a
“Fourier-Bessel” series,

f ( p) = (?„„£ ) ,
n= 1
Chapter 20

Spherical Harmonics

20.1 Solutions to Laplace’s Equation

The fundamental solution to Laplace’s equation in unbounded space is 1/r,

V 2i = 0, r > 0. (20.1)

In terms of this solution, we can generate a large number of others. For example,
taking a to be a constant vector,

V 2(a •V ) —= 0, (20.2)
r
we find
-1 a •r
a •V - y»3
(20.3)
r
is also a solution for r ^ 0. Continuing this operation, we see that

(«, •V )(a2 •V ) i = 3 (a ,.r )(«i . r ) - ( , , . « i y (20.4)

is yet a third solution. This process can be repeated an indefinite number of

times, to yield the following solution to Laplace’s equation,

[(ai •V )(a2 . V ) ... (a, •V )]i = ^ / / ( r ) , (20.5)

where / / ( r) is a homogeneous function of r of degree /. We also observe that

/ / ( r) itself is a solution to Laplace’s equation. [This follows from the inversion
theorem, that if <^(r) is a solution to Laplace’s equation, so is ^ ( r / r 2). See
Problem 20.1.]
Thus, our attention is directed to solutions that are homogeneous polyno­
mials of degree /. The above construction provides examples for / = 0,1,2,
which are summarized in Table 20.1. [Here we denote ( x, y, z) by (®i, « 3 ).]

231
232 CHAPTER 20. SPHERICAL HARMONICS

l fl number of independent
solutions
0 1 i
1 X\y X2) X3 3
2 3xmXn 5

Table 20.1: Polynomial solutions of Laplace’s equation.

Why are there only five independent solutions for / = 2? A symmetrical tensor
has six independent components but because of the constraint that the tensor
satisfies Laplace’s equation, it must be traceless, leaving but five independent
components.
The general polynomial of degree / can be constructed from the monomials

k\ + k2 + k3 = /. (20.6)

How many of these monomials are there? To answer this, we first ask the
analogous question in two dimensions: how many monomials of the form

(20.7)

are there with k\ + k2 = n? The answer to this question is simple since if k\

goes from 0 to n, ^ must go from n to 0, giving n - f 1 possibilities. Thus to
answer our three-dimensional question, we first assign a definite value to k3,

kt + k2 = / - k3. (20.8)

The number of monomials with this value of k3 is

/ - Jf es+l , (20.9)

so the number of homogeneous polynomials of degree / is

* i
£ ( / - f c 3 + l) = - ( / + / ) ( / + 2 ) . (20.10)
fc3=0

From this set of polynomials, we wish to find those combinations which are
solutions to Laplace’s equation. Since V 2 acting on a homogeneous polynomial
of degree l produces a homogeneous polynomial of degree / — 2, of which there

i ( f - 2 + l ) ( Z - 2 + 2) = i / ( / - l ) (20.11)

independent ones, there are - / ( / — 1) restrictions on the polynomials, that is,

the number of independent solutions to Laplace’s equation of degree l is

i(/+ l)(/ + 2 ) - i / ( / - l ) = 2/+l. ( 20 . 12 )

20.2. SPHERICAL HARMONICS 233

For the cases / = 0,1,2, this agrees with what we found above. The solutions
we find in this way are called solid harmonics, Y¡(r). To emphasize the fact
that they are homogeneous polynomials of degree /, the solid harmonic may be
written in terms of a surface (or spherical) harmonic, Y /(r/r):

Y,(r) = r'Y, ( Í ) (20.13)

- ¡¿ r tf ( ; ) ■ (20-M)

where the latter form, also a solid harmonic, results from inversion and is the
solution constructed in (20.5).

20.2 Spherical Harmonics

Our next task is to devise a systematic and convenient way to generate the
spherical harmonics as functions of the spherical angles 9 and <j>. We first ask
under what condition is the polynomial,

(a •r)', (20.15)

a solution to Laplace’s equation? Since

V (a •r)l = /(a •r )i” 1a, (20.16)

we see that the Laplacian acting on this polynomial is

V 2(a •r )/ = /(/ - l ) ( a . r )/“ 2a2, (20.17)

which is Laplace’s equation if a2 is zero (necessarily, a must then be complex).

A convenient way to write a2 is

a2 = (ai — ia2)(ai 4 ia2) 4- <*!, (20.18)

(again, note that the components a¿ are complex), suggesting that the condition
that a2 be zero can be automatically satisfied if we write

ai +«*2 = É-,
ü\ id2 — 5

*3 = í + í - , (20.19)

where are two arbitrary complex numbers. Then we have

r 1, . . x 4 iy 1/ . . x — iy z
* --= 2^ai ~ + 2 (ai + + a3r

= ^ [-£ + sin Oe'4, + sin Oe- '4, + cos 6] , (20.20)

234 CHAPTER 20. SPHERICAL HARMONICS

and the polynomial (20.15),

(a*r)* = rl , (20.21)

can be rewritten in terms of as

( 20.22)

where, in the last line, we have employed a convenient form of a Taylor ex­
pansion. In this way, we have constructed, from polynomials of degree /, 2/ -b 1
independent functions which are solutions to Laplace’s equation, the coefficients
of the powers of in the expansion

£¡—m
(l + m)!
(sin 6)
r l , y/(l + m)! (/ - m)! Y 0 ~ m)-
l—m
(cos2 0 — 1)*
(20.23)
dcosO 2Ul

All we need is a normalization factor in order to define the spherical har­

monics. Employing the notation for the monomials

á +mt -
= V’ln (20.24)
a/ ( / + m)\ (l - m)\

we obtain the generating function for the spherical harmonics, Y/m(0,<£),

( a « r )'
(20.25)
/! r' E

where, according to (20.23),

d ' 1 m (cos2 6 — 1)(

(sin $)~m
(/ — m)\ dcosO _ 2*/!
(20.26)
20.2. SPHERICAL HARMONICS 235

in which —/ < m < /, with / = 0 ,1 ,2 ,.... An alternative form can be derived

by noting that the left hand side of (20.25) is unaltered by the transformation
[see ( 2 0 .20 )]
í+ < -í-, 0 ^ -0 , (20.27)
which implies that the spherical harmonics must remain unchanged under the
substitutions
m —►—ra, 0 —* —0, <t>—►—<j). (20.28)
In this way, we learn that

Y im (M ) - *íf- m ( - 0 , - ¿ ) , (20.29)

or, using the explicit form (20.26), we obtain the alternate version

(a jl\ _ . /2 / + 1 / ( / TTl)! ■|i+m (cos 2 0 - 1 )'

( a\n
Yim{ 0 A ) - \ i W (/ + m ), e ( - sin^)2 '/! d cos 9
(20.30)
Sometimes it is convenient to separate Y/m into its 9 and <j> dependences,
im<f>
Y/m( M ) = -3 = e im (0 ) (20.31)
V 2x

where

2/ + 1 (/ ± m)! d ,:f:m (cos 2 0 — l ) 7

0 /m W = ( ± sin 9) (20.32)
(J T m)! dcos6_ 2,l\

The functions Yim(0,</>) are called surface spherical harmonics, or simply

spherical harmonics, a term introduced by Lord Kelvin, William Thomson
(1824-1907). Here are the explicit forms of the harmonics of degree / = 0,1,2:

1= 0 : Too = - ? = , (20.33)
V4?r

3 x + iy
1= 1: Yn = sin
8it

Yio = L i - cos 9, (20.34)

4 tt r V 4 tt

3 x — iy 3
Y i,-i = sin 0e

f i 15
b I(x + iy)2 _ Q L sini 0e2i*
1= 2 : Y22
~ V 32 ir 3 2 tt

15 z(x + iy) _ 15
Y2i = - —\l — cos 0 sin 0 e*^,
236 CHAPTER 20. SPHERICAL HARMONICS

(20.35)

20.3 Orthonormality Condition

The particular factors that occur in the definition of Y/m are such as to make
the spherical harmonics an orthonormal set of functions. To see this, consider
the product of two generating functions, with parameters a and a*, integrated
over all angles:

with
díl = sin 0 dO d<j). (20.37)
This integral can only contain rotationally invariant combinations, that is, it has
to be a function of scalars constructed from a and a*. Since a2 = a*2 = 0, the
only such scalar is a •a*. Therefore, there must be an equal number of factors
of a and a*, which means that the integral (20.36) is zero unless / = we have

J dSl (a* •£)* (a* = Sil'C l(a * - a)1. (20.38)

To calculate C\ we consider a particular form of a:

a = (l,* ,0 )) a* = ( 1 , —¿, 0 ). (20.39)

The quantities appearing in (20.38) are then

a* •- = sin$ e - * + , a •- = sinfle’ *, a* •a = 2, (20.40)

r r
implying that the integral, (20.38), for / = is
pTT p 2 tT
C i2l = / sin 9d0 d</>(sin^e” ^ ) i(sin^e^)i
Jo Jo

= 2ir J d(cos#)(l — cos2 0)1 = 4 ir J dx (1 — x 2)1

ljrm 2
~ 4 (2 / + 1 ) ! ’
(20.41)

The final integral in (20.41) is evaluated cis follows. Defining

h = / d x ( l - x 2)1, I0 = 1 , (20.42)
Jo
20.3. ORTHONORMALITY CONDITION 237

we integrate by parts, for / > 0 , to derive the recursion formula

I, = 2 //,_ ! - 211,, (20.43)

which implies that

21
h = -h - 1
2/ + 1
21 2 1 - 2 2 1 - 4
■ Io
2/ + 1 2/ - 1 2/ - 3
( 2 1/!)2
{21+ 1 )!’ (20.44)

the last form being valid for / > 0 . Alternatively, we could evaluate // in terms
of the beta function, B {m ) n),

Il~\j ***(! +*)*

1 —x
= 221 f d tt'il-t)1 t =
Jo
= 22' 5 ( / + l , / + l )
/!/!
= 2 2'- (20.45)
( 2 / + 1 )! ’
where we have noted that for integer m and n,

B ( m ,n ) = i * d t f n- 1( l - t ) n~ l = -^ )!(W 1. ) ) ! . (20.46)
Jo (m + n - 1 ) !

We have therefore learned that

(a* •r/r)1 (a ^ r/r)^ _
/
2l
du­ = Si,i 4 tt (a * , a)'
ll i'\ ( 2/ + 1)!

= ¿ É düYrm{ 0 , m m'(O A ),
m = -l J
(20.47)

where we have used the generating function (20.25). What we now must do is
extract the coefficient of fr°m (a* *a )i>which is achieved as follows:

2 '( a * .a ) ' ( 4 £+ + e < £ - ) 2'

( 2 / + 1 )! ( 2 / + 1 )!

_ ^ \ ' _______ (^ 0 * _______ I f * f c \l—m

{21 + 1 )! (/ + m )!(/ - m)\K^ +>

1 1
21 j ^ (20.48)
m ——l
238 CHAPTER 20. SPHERICAL HARMONICS

where we have used (20.19), (20.24), and the binomial expansion. Compar­
ing this with (20.47), we obtain the orthonormality condition for the spherical

J
harmonics:
dnY,*m(0, ct>)Y,,m,(0, <j>) = 6w 6mm,. ( 2 0 .4 9 )

When Y,m is separated as in (20.31), the orthonormality condition reads

/ sin6d0Qim(6)OVm(0) = 6u<. (20.50)

Jo

20.4 Legendre’s Polynomials

A few special cases of 0 /m can be easily extracted from (20.32):

, m _ t/(2 7 T Ó !(s in 0 y
©i. - \ j— ------- j Ü T ’ (20.51)

e«(0) = ( - 1 )'e,,_,(#), (20.52)

d 1 1 (cos 2 0 — I )1
© how =
d(cos 0)J 2 7/!

2/ -f* 1
P/( cos 6). (20.53)

Occurring in (20.53) is Legendre’s polynomial of order /,

1 (cos 2 6 — 1 )/
Pi (cos 0) = (20.54)
_d( cos 0)_ 2 */!

named for Adrien Marie Legendre (1752-1833), which is so normalized that

P /(l) = 1. (20.55)

According to (20.50), Legendre’s polynomials satisfy the following orthonormal­

ity condition:

/
! 9
ci(cos 0)Pj (cos 0)Pj/(cos 0) = —— -Su». (20.56)
-i 2/ + 1
We will begin to explore the physical significance of Legendre’s polynomials in
the next chapter.

20.5 Problems for Chapter 20

1. Assume that <^(r) is a solution of Laplace’s equation. Show that for r > 0,

is also a solution.
20.5. PROBLEMS FOR CHAPTER 20 239

2. Check that
0
£+ — V>lm = V (l ~ m)(l + m + IJV’i m + l ,

= \/(i + m )(/ - m +
<7?+
Note that these are related by £+ 'ip¡m —* Show that

3 .1 3 V/ vi . 3 ,
a ? - c o t # 7 8 ? J ( a " r)

rot4 ¿) ( a - r)' = f + ^ T ( a ' r )' ’

d
r ‘* ~
which are related by £+ 0 —>•—0, <j> —<j). The generating function
(20.25) shows that these are equivalent to

e* ~ cot4 ^ ) y'm = V ( ^ - » » ) ( / + m + l)y,TO+i,

e -^ ( - A - c o t 4 ^ ) Ylm = W + ” * )(/- m + l ) y , m_ i.

These two equations are related by 0 —* —6, <j) —* — Y/m —►Y/_m.

3. Rewrite the conclusion of the previous problem for 0 /m(0). Derive the
differential equations

- (m + 1) c o t^ ) ( ^ - m c o t ^ j 0 /m = [/(/ + 1 ) - m(m + 1 )] 0 (TO,

( l i é ~ ( m ~ l ) c o t^ ) ( - J j j ~ m c o tfl) 0 /m = [/(/ + 1 ) - rn(m - l)] 0 /m.

4. Prove
y*m = ( - l ) mVi-m,
by noting that (a •r)* is obtained from (a •r) by the replacement £+ —*
, £ - —►—*£J.. Show the consistency of this result with the conclusion
of Problem 2 .

5. Assume ai, a 2 , a 3 are all null vectors. Then

J dQ (ai •r)/l(a2 •r)h (sL3 *r)h = C(a2 •a3)ni(a3 •ai)n3(ai •a2)n\

Find the condition on /i, /2, /3 such that

J dQ, YilTni Y/2m2 Y/3)m3 ^ 0 .

This is equivalent to

Y?imiY¡2m2 = ^ (coefficien ts )Y/ 3m3-

240 CHAPTER 20. SPHERICAL HARMONICS

6. An example of a null vector is

a = (—¿cos a, —¿sin a, 1 ),

that is,
a •r = —i(x cos a + y sin a) + z.

Show that, for this a,

r
i'2w

£(*-r)v””=
and, in particular, that
P7T ^
Pi(cos 0) = / — (cos 0 — ¿ sin 6 cos a )/ .
Jo ^

7. Apply the last formula to evaluate, for |¿| < 1 ,

oo
Y j ' P i j c o s 9)
1=0

and arrive at a known result. Would you expect that P/(cos 0) is also given
by
i*IT J
I — (cos# — ¿sinocos a)
JO *
In any event, again work out ]T^ 0 t lPi(cos0), using the latter formula.

8. Apply the integral representation of Jo, and the result in Problem 6 , to

show that ^

P/(cos0) = ( cos# — s in 0 ~ ) J0(t)

\ d tJ t=o
Verify this for / = 0,1,2. Now let 9 = x/l and, for fixed #, consider the
limit / —» oo. You should get

lim Pi (cos y ) = Jo(#),

l— +oo \ l/

which is often used in the approximate form

0 « 1 , / » 1 : P i( cqbO ) ~ J 0(W ).

9. For what geometrical reason does one expect an asymptotic connection

between spherical and cylindrical coordinate functions? Use this insight
in comparing the differential equations for Pi(cos9) and Jo(^) t° arrive
again at the result of the previous problem.
20.5. PROBLEMS FOR CHAPTER 20 241

10. (¿jj) - is a solution of Laplace’s equation provided r > 0. (W hy?) It is

homogeneous in r of degree —/ — 1 , and is independent of the angle <j>.
Therefore,

Find the constant C/. Check the result for / = 0,1,2.

11. This is a generalization of Problem 8 . Show that

f c V s T r M i ) = < - 1)m j" w '

12. Show that

COS0) = e~kl Jo{kp).
1=0

13. This is a generalization of Problem 10. Indicate the analogous reasoning

that leads to

+%) (é)
d_
dx

which also holds for m < 0 , in the sense that

~i
(d _ . a \ d 2 _ f d f j_
\ d x ^ ldy) dz2 ~~ 9yJ 1

understood as an operator statement acting on l /r . Find C\m.

Chapter 21

Coulomb’s Potential

The motivation for constructing the solid harmonics was that they formed,
in terms of homogeneous functions, a particular set of solutions to Laplace’s
equation. Since these harmonics are functions of the spherical angles 9 and
<t>, Laplace’s equation should be expressed in spherical coordinates, where the
Laplacian has the form

1 r 1 a (. &\ i d2
V2
r 2 dr \ d r) r2 sin 9 dO \ m dO) sin2 0 d(j>2
(2 1 .1 )

(This may be immediately inferred from (16.15) by noting that the distances
corresponding to infinitesimal changes of the spherical coordinates r, 9, and
<j> are dr, r d9, and rsin9d(j), respectively.) Thus, since the solid harmonics,
(20.13) and (20.14),

y¡m(r)= {//_! } b m ( M ) , (21.2)

are solutions to
V 2y,m(r) = 0, r ¿ 0, (21.3)

and
(21.4)
Tr { r~ ]~l } - i(/ + 1 } { r ~ l~ 1 } ’

the differential equation satisfied by the spherical harmonics, Y/m, is

i d2
+ + 1(1+ 1 ) Yim(0,<f>) = 0 . (21.5)
sin2 0 d<f>2

When the 9 and <j> dependence of Y¡,n is separated as in (20.31), the differential
equation for 0 ;m is

1 d
sin 9 89
©im(0) = O. (21.6)
sin2 9

243
244 CHAPTER 21. COULOMB’S POTENTIAL

21.1 Legendre’s Polynomials

The fundamental solution of Laplace’s equation is Coulomb’s potential, (13.3),
for r ^ r', which, written in spherical coordinates, is
1_________________ 1__________
(21.7)
|r —r'| y V 2 + r /2 — 2 rr' cos y ’

where y is the angle between r and r', explicitly,

cos y = cos 9 cos 0f 4* sin 0 sin 9' cos(<j> — f t ) . (21.8)

We now expand (21.7) as

___________ 1___________
(polynomial of degree l in cos 7 ),
V r l + r< — 2 r>r< cos 7
(21-9)
where r> (r<) is the greater (lesser) of r and r*. (The recognition that the square
root is also the absolute value r>|l — (r</r>)e*7 | makes it evident that this
expansion converges.) The polynomial of degree / appearing here is a solution
to (21.5), and so must be a linear combination of Y¡m(0, ft)'s, —/ < m < L On
the other hand, as we will show below, this is just Legendre’s polynomial in
cos 7 , (20.54), that is

1
(21 .10)
| r-r'|

(In fact, this is how Legendre first introduced his polynomial in 1784.) For
7 = 0 , this expansion is trivially

(2U1)
which supplies the normalization condition

Pi( 1) = 1, ( 2 1 .1 2 )

as is required for Legendre’s polynomials [see (20.55)].

The demonstration that the polynomial introduced here coincides with Leg­
endre’s polynomial follows immediately if we take r' to lie on the 2 -axis, so
that 7 and 9 are identical. Then, with r< = r, we recognize in (21.10) that
rlP¡(cos9) is a solution to Laplace’s equation, a solid harmonic of degree /, in
which the surface harmonic factor is independent of <£. In short, Pj(cos 9), as de­
fined in ( 2 1 . 10 ), is proportional to 0 j o ( 0 )> an < ^ virtue of the property ( 2 1 .1 2 ),
is identical with what is defined in (20.53).
A more direct proof of this identity, and further properties of the spherical
harmonics, follow if we expand Coulomb’s potential
1
G( r,rO = (21.13)
21.1. L E G E N D R E ’S P O L Y N O M I A L S 245

which satisfies the inhomogeneous Green’s function equation

—V 2 G ( r, r') = 4ir«$(r - r#), (21.14)

in terms of the solutions to the homogeneous Laplace’s equation, (21.3). In

spherical coordinates, the delta function is [see (16.42)]

¿(r - r') = 1 % - r ') ¿ ^ ( 0 - (21.15)

while the Laplacian is given by (21.1). For r < r', (21.10) shows that the
solution to (21.14) can be expanded in powers of r, [cf. (21.2)]

r < r' : G = ’^ A i mrlYim{0,<j>), (21.16)

lm

while for r > rf the expansion is in terms of powers of 1 /r ,

r > r' : G= jB|mr - , - 1 Y;m(0, <^). (21.17)

lm

(This guarantees that G is bounded both at r = 0 and as r —» oo.) The

expansion coefficients, A\m and B\m, depending on r', 0' and <£', are to be
determined by the conditions on Green’s function near the source:

G is continuous at r = r'; (21.18)

and
r'+0 i
—r = 4w~— 6(0 - e')6U - (21.19)
2d~rG r'-0 Sm$

These two conditions imply, respectively,

1
r' Aim — r,i+ i Blm, ( 2 1 .20 )

(l + l ) ± j B lm + l r ',+1Aim Yim{0, </>) = 4 * ^ -¿ 8 (0 - 0')6{<f> - (2 1 .2 1 )

£ sin u

If we write
Aim — 1 Cr/m» Elm = T Clm) ( 2 1 .2 2 )
( 2 1 .20 ) is satisfied automatically, while ( 2 1 .2 1 ) reads

£ ( 2 / + 1 ) C i m Y l m (0,4>) = - ¿0- ( 2 1 ‘23)

The use of the orthonormality condition, (20.49), now yields

47T
Cim = ¿ T T T ^ O ^ ' ) . (21.24)
246 CHAPTER 21. COULOMB’S POTENTIAL

By substituting this into (21.23) we obtain the completeness statement for the
spherical harmonics,

£ r hB( í , ^ ( í ' , í O = <!>'), ( 2 1 .2 5 )

r—
lm
- sin (7

which allows us to expand any function of 9 and <j) in terms of spherical har­
monics. We therefore have obtained such an expansion for Green’s function

1 - ■■ Affr
G (r’ r '} = = £ T + T ^ T V fn fr m m W ,* ')- (21-26)

Comparing this with the alternative representation, (21.10), we obtain the re­
lation
4 tt l l
P,(C0S7) = ^ E ^ M ) ^ ', ^ ) - (21-27)
m

The relation (21.27) is called the addition theorem for spherical harmonics.
Let us finally show explicitly that this function of cos 7 actually is Legen-
dre’s polynomial, (20.54). As noted above, this is easily done by considering a
particular coordinate system, where

9' = 0 => 7 = 9. (21.28)

From (20.32), we learn that

F¡m( 0/ ,^ /) oc (sin 0,) ,m|, (21.29)

implying [see (20.53)]

y|m(0 , ^ ) = (21.30)

so that only the m = 0 term contributes to the right-hand side of (21.27), which
is therefore, by (20.53),

^ v ^ p'(OT9)v^r =fl(cos9)' (21-31)

Thus we have proved that the function of cos 7 occurring in (21.10) is indeed
Legendre’s polynomial.

21.2 Infinitesimal Rotations

So far we have not explicitly used the differential equation satisfied by the
spherical harmonics, or by 0 /m(0), (21.6). We confine ourselves to m = 0 in
noting that the introduction of the variable

¡1 — cos 9 (21.32)
21.2. INFINITESIMAL ROTATIONS 247

provides the following form for the differential equation obeyed by Legendre’s
polynomial:

P M = o. (21.33)

One easily checks this equation for the first few polynomials,

P0(fi) = 1, Pl(fl) = fi,

P Á ») = \ { ^ 2 ~ 1), PM = |(5/ts ~ 3fi). (21.34)

While the relevance of Laplace’s equation cannot be minimized, the differ­

ential properties of the harmonics are fundamentally statements about their
responses to infinitesimal rotations of the coordinate system, or, equivalently,
of the vectors r and a. An infinitesimal rotation is described by a vector 6u>,
specifying the axis and angle of rotation. The common rotation of r and a is
represented by the infinitesimal changes

Sr = 6u>Xr, <$a = <$u>Xa, (21.35)

or by equivalent infinitesimal variations of the parameters that specify these

vectors: 9 and <f>for r (r is invariant), £+ and £_ for a. We exhibit these for the
three orthogonal axes of rotation as (see Problem 21.2)

(69,6<j), <$£+,<$£_) = 6u)x (—sin^>, —cot 9 cos <j>, — —¡j¿£+)

¿i z
+ ¿wy(cos <j>, - cot 0 sin<¿, , i ^ + ) + 6wz( 0 , 1, - | t £ + , i¿£_).(21.36)

We shall express the induced changes in Yim(9,(f)) and 'i/jim by a vector ro­
tational operator R , which is defined initially by its action on r:

ir = (S w .R )r ; R = rxV . (21.37)

Correspondingly, we write

8Yim(9,<t>) = ( ¿ u ; . R ) y /m ( M ) , Him = ( S u , (21.38)

Thus, for the relatively simple situation of a rotation about the 2 -axis, given by
the coefficient of 6u>Zi we have, respectively, from (21.36),

<2L39>
For the rotations about the other axes, we see that

Rx = - s i n < ^ - c o s ( ¿ c o t 0 ^ , ií* = - i » + , (21.40)

and

Ry = c o s ~ sin^c o t ^ = \ (-* - ^ + u 4 : ) ’ (21-41)

248 CHAPTER 21. COULOMB’S POTENTIAL

although more convenient combinations are

d
Rx ± iRy = e*’ * ( ± i “ cot # J ^ ) > R * ± iR y = - t £ (21.42)
T dt±

Let’s begin with the direct effect of these operators on

4>lm = (21.43)
■\/{l + m )!(i — m)!

For example, we have

£± - (f ^ (21.44)

from which follows

Rz^lm = -imV'/m- (21.45)
Next, we learn that

(R x i i R y )ipi m — ®v u ± m ) ( / m + • (21.46)

We can also apply these two operations in either order:

(R x “F i R y ) ( R x i iRy)'tfiim = ~~(l di m)(l tn + 1 (21.47)

If we take half the sum of these, from which RxRy and R yR x cancel, we get

{R l + Ry)il>im = - [ / ( / + 1) - m2]4w (21.48)

Then, in view of (21.45), there emerges

R 2V»;m = - / ( / + l ) V w (21.49)

We have already made use of the fact that, as a scalar, a •r is unaltered by

a common rotation of the two vectors. But now we point to the inference that
the right-hand side of (20.25) is therefore unaltered by such a rotation. And
then it takes only a glance at the addition theorem (21.27), where P / ( c o s y ),
cos 7 = (r /r ) •(r '/r ') , is surely a rotational invariant, to conclude that the
and behave in the same manner under rotations. The immediate conclusion
is that every statement about the effect of R on t/>/m has its precise counterpart
in the action of R on Y/m. (We may relate Y and Y* by the result of Problem
20.4.) Thus, we infer from (21.45) that

d
Oj¡Yi™(0> <t>) = imYim(0, <t>), (21.50)

which is certainly true. Then (21.46) leads analogously to (we divide by i)

** ( ± ¿ + ¿ c o t * A ) Ylm(e,<f>) = V ( l T m ) ( l ± m + l)Yl¡m±1(0,4>).
(21.51)
21.3. SPHERICAL BESSEL FUNCTIONS 249

And it will be clear that the analogue o f (21.49), still written compactly as

[R 2 + /(/ + l)]y ;m( M ) = 0, (21.52)

can be no other than (21.5). [Note that the result (21.51) is derived by equivalent
means, without the geometrical interpretation, in Problem 20.2.]
On making the (^-dependence explicit in (21.51), we are left with

( ± ^ j - m co tfl) @im(6) = \ /(/ T m )(l ± m + l ) 0 f >m± i(0 ), (21.53)

where it should be noted that

-m c o t 0 = ± (sin 0 )±m-^ (sin 0 ):Fm. (21.54)

do do
Immediate consequences appear if we set m = ± / in (21.53). Then, the right-
hand side vanishes, which, on referring to (21.54), tells us that Qjt±i are pro­
portional to sin* 0 [see (20.52), (20.51)]. Then one can start with these extreme
values of m and work up or down in m to construct all the other 0 /m( 0) func­
tions. Specifically, we rewrite (21.53) with the upper sign as

- d (sin 0)~m&im (9) = yj(l - m ) ( l + m + l)(sin 0 )- m - 1 0 ;)m+i(0).

C° S (21.55)
Then the fc-fold repetition of this operation gives

(/ — m)\ (/ -f ra -f Ar)!
(sin 0 )-m- * 0 ,,m+*(0). (21.56)
(/ —m —k)\ (/ 4- m)\
If we here set m = where [(20.51)]

(sin *)'©,,_,(*) = ^ y ^ ( —1 y(C

Os2Jn1)<, (21.57)

and replace the positive integer k by / + m, we get

l+m
(cos 2 9 — 1)(
e,„(«) = ( 3^ ) (21.58)
2*11 ’

in agreement with the lower sign choice in (20.32). Working down from m = /
produces the equivalent version.

21.3 Spherical Bessel Functions

We return to the discussion of completeness, in a manner that extends to three
dimensions what has been studied for two dimensions in Chapter 16. The start­
ing point is
<$(r - o = / w ,»k ♦( r - r ')
(21.59)
( 2 tt) 3
250 CHAPTER 21. COULOMB’S POTENTIAL

which directs attention to an individual exponential,

g*k •r _ cos 8 (21.60)

the latter form appearing when the direction of k is adopted as the 2 -axis. Now,
for functions that depend only on the angle 0 , the completeness statement of
(21.25) reduces to [recall (20.31) and (20.53)]
oo fi pir
f(0 ) = V ) ( 2 / + l ) f i ( c o s 0 ) - / de( sin 0' Pi (cos 0f)/(O') (21.61)

The application of this expansion to (21.60) yields

gikr cos $ _ J 2 (2 l + l)P;(cos 0)ilj¡(kr), (21.62)

/=0

where
i'M k r) = \ J x dp P,(p)eikr* . (21.63)

To explain the presence of the factor il in this definition we must recall

the nature of solid harmonics as being homogeneous in the components of the
vector r. In particular, on using the following notation to relate solid and surface
harmonics,
y,m(r) = r 'y im( M ) , (21.64)
we have the simple relations,

Y im (-r) = ( - l ) 'y , m(r),

Ylm( i r - e , * + 4>) = ( - l ) 'Y Im( M ) , (21.65)

which includes
P /(—cos#) = (—l /P /( c o s 0). ( 2 1 .6 6 )
This tells us what happens when the complex conjugation of the right-hand
side of (21.63) is combined with the transformation ¡i —> —fi. The appearance
of a factor of (—1)* is compensated by the response of il. In short, the function
ji(k r) defined by (21.63) is a real function.
The introduction of the Legendre polynomial construction (20.54) into this
definition yields

d ^V - i y
M *) = r ' l j ' d\i d¡S l
dp J 2 ' /! 2 ' /!
(21.67)
the latter being produced by /-fold partial integration. [This can be turned into
a differential formula. See Problem 21.3.] But we recognize this! According to
(19.156), we have

h {t) = ( ^ ) Ji+i/2(t), ( 21 .68)

21.3. SPHERICAL BESSEL FUNCTIONS 251

which accounts for the name “spherical Bessel function” that is applied to j¡.
We note the translation into this notation of (19.158), (19.159):

3i+1 ft) 1 d2 \ ji(t) _ 1 d ji(t)

(21.69)
t'+l 21 + 2 \ dt2) t' td t tl '

As for the differential version of (19.162), or

<P_ 2 d_ *(* + 1 )
ji(t) = 0 , (21.70)
dt2 + td t + t2

we have only to remark that

(V 2 + k2)eik •r = 0; (21.71)

the spherical coordinate version of this equation, valid for each / in the expansion
of (21.62) is [see (21.1) and (21.5)]

L A . f. ji(kr) = 0 , (21.72)
r2 dr \ dr J

which is the content of (21.70).

The statement of the expansion (21.62) in an arbitrary coordinate frame is,
using the addition theorem, (21.27),
oo l
elk ’ r = ^ 2 ^ 2 4-iri,j¡(kr)Y,m(d,<j>)Y,Zt (a,f3), (21.73)
1=0 m ——l

where a and (3 are the spherical coordinate angles for the vector k. The complex
conjugate version is, with r —» r',
oo /
g—*k = £ Y2 (21.74)
/=0 m——l

Now we carry out a solid angle integration over the directions of k,

(dk) = k2dkdtth) (21.75)

using the orthonormality of the Y/m(a?,/?), to arrive at

Jdihw«_ e. k .re- , w = Í £ ¿ ^ r ) ¿ K ^ 0 £ y ím( M ) O O ')•

(21.76)
Accordingly, the spherical coordinate transcription of (21.59) is

Q oo /»oo Í
= - £ / dkk2 j ^ n i k r ' ) Y , Ylm{eA)YL(0',<t>')- ( 2 1 .7 7 )
m tl /
252 CHAPTER 21. COULOMB’S POTENTIAL

Then, if we multiply by a particular and integrate over solid angle

in the primed coordinates, what emerges is

1 2 f°°
-=*£(r — r') = — / dkk2 jj(k r)ji(k rf) ) (21.78)
r n Jo

the completeness property of the radial functions jj(k r), 0 < k < oo. It is not
a new statement, for on introducing ( 2 1 .68 ) we get

1 I°°
-8 (r -r ')- d k k J l+1/2(kr)Ji+1i 2(kr'), (21.79)
r Jo

the completeness statement for the Bessel functions Jm, with m = / 4 - 1/2 [cf.
(16.59)]. Incidentally, putting (21.78) back into (21.77), yields the completeness
relation for the spherical harmonics, (21.25).
The reverse side of the coin of completeness carries the image of orthonor­
mality. We look to (21.77) for the basic functions,

í°° / o \ l' 2 / 2 \ 1/2 V

8{v - 1 1) = Jo dk k2 ( - J M kr)Y lm(e, <j>) (jJ ¿l(*f 0 Y í m ( O ') ,

(21.80)
which combines summation over l and m with integration over k. The corre­
sponding orthonormality statement is contained in the values of
♦
r / 2 \ 1/2 i Y ‘A 1/2 1
(-) ii(* r )y /TO(tf,^) -J ji'(k'r)Yiimi(0, <j>)
[0
2 f°°
= 8>w8mm' ~ d r r 2 ji(k r)j¡(k 'r)
n Jo

= 6¡l'6mmi - ^ ( f c ~ ^)> (21.81)

where the radial integral is just (21.78), with the substitutions k -»• r r -* k
r' k'.
Let us not hurry on without noting the form taken by Coulomb’s potential
(13.17):

1
w i * (r-r')
|r^r'| ( 2 tt) 3 P e
v° ° s 9 ,oo
2 ^ (2 / + l)P /(co s 7 )--- / d kj](kr) ji(kr'), (21.82)
l-Q * JO

which produces the identification, from ( 2 1 . 10 ),

poo
2 1 r<
/ dk j¡(k r)j¡(k r') (21.83)
Jo 2/ + l r(+1 ’
21.4. PROBLEMS FOR CHAPTER 21 253

or, equivalently,

2 r°° 1
A< 1: - jf dtj,(X t) j,{t) = — A'. (21.84)

This is illustrated for / = 0 by (see Problem 21.3)

POO -t pOO
n i
1,00 dtii / 1 \

—— / - 75- sin Af sin t = — / d ( 7 ) [cos(l + X)t — cos(l — A)t]

it A Jq t ttA jo \f /
1 r
= ¿ [ ( 1 + A) J Hf
ysin(l + A ) t - ( l - A ) j f
r°° di
ysin(l-A)*

= = <2 1 8 5 >

21.4 Problems for Chapter 21

1. Another proof of the addition theorem, (21.27), proceeds as follows. By
considering r< = r, argue that rlPi(cosy) is a solution of Laplace’s equa­
tion that is homogeneous of degree / in the components of r. As such,
P /(cos 7 ) must be a linear combination of the 2 / + 1 spherical harmonics
Y/m(05<^). A similar consideration of the situation r' = r< tells us that
P/( cos 7 ) is also a linear combination of the spherical harmonics Y/m(0 ', ft).
Then, show that because cos 7 depends only on the difference of the az­
imuthal angles and ft we can write

/
Pi( cos7) = £ aimYlm{ e j ) Y ; m{e'
m——l

in terms of 2/ + 1 real coefficients a/m. Now set 0', ft equal to 0, <£,

respectively, so that 7 = 0 , and deduce by integrating over all angles 0
and <j>¡ that
/
4tt = a,m.
m——l
Then, multiply the above expansion for P/(cos 7 ) by its complex conjugate,
and integrate over 0 , <¡) and over 0', ft separately, to deduce

(± 0 1 _ V «2
2/+ 1 - 2 ^
m——l

Infer from these two relations that

4ir
a,m = 2/TT’
hence proving (21.27).
254 CHAPTER 21. COULOMB’S POTENTIAL

2. Verify the rotations of the coordinates 0, <f>, £+, and in (21.36), and
thereby verify the rotation operators in (21.39)-(21.41).

3. Starting from the integral representation for the spherical Bessel function
ji(t) given in (21.67), show, by repeatedly integrating by parts so as to
eliminate the (1 — p 2)1 factor, that

[this is (19.155)], and from this give explicit forms for ji for / = 0, .. .3.

4. The expansion

F (x ,t) = J 2 tIpi(x), i*i < i,

\/l — 2 xt + 12 /=o
is usually referred to as the generating function for Legendre’s polynomi­
als. From it all the properties of these polynomials may be derived. In
particular,

(a) by differentiating F ( x yt) with respect to t , derive the recurrence re­

lation for the Legendre polynomials,

(2/ + l)xP t(x) = (/ + l)P i+ i(x) + IP>_ !(* ), / = 1,2,3,...,

(b) and by differentiating with respect to x , derive the differential equa­

tion satisfied by P\\

(1 - x*)P {\x) - 2xP {{x) + /(/ + 1 )Pi{x) = 0.

5. As with the spherical Bessel function of the first kind, the spherical Neu­
mann function may be defined in terms of cylinder functions of half-integer
order by

»»(*) = ] f ^ Nl+ 1 / 2 (*)-

Using the result of Problem 16.10, compute no and n\ in terms of sines

and cosines.

6. Using the results derived in Problem 19.13, derive the orthonormalization

condition for spherical Bessel functions on a finite interval, 0 < / > < « ,

dpp2j¡ (y¡n j, [j/+l (Tin)]2 ,

where jjn is the nth zero of j¡.

7. Recognize that the completeness properties of jo(kp) on an infinite and on

a finite interval [recall (19.107)] are equivalent to previously established
results, in terms of sine functions and imaginary exponentials.
21 A. PROBLEMS FOR CHAPTER 21 255

8. Prove that

kJ m {k p )J m {k p ) k J m+ l (k p )J m + l (k p )

A k ( J m + l { k p ) J m { k p ) -f- Jm ( k p ) J m + \ ( k p ) ) •
d k [ p 4* p f

Use this result to show that if the Jm{kp)1 k : 0 —» oo are complete, so also
are the Jm+i(&p)> i.e., the completeness of all the J/+ i/ 2 (&p), / = 1 , 2 , .. . ,
follows from that of the J\p¿{kp). [You might also want to think about a
convergence factor or something equivalent.]
Chapter 22

Multipoles

In terms of the above discussion of spherical harmonics, we now make a general

analysis of the potential, due to a given charge distribution, p(r'), outside of
that distribution. [See Fig. 22.1.] The potential is given by (1.5),

H r) = J(dr') | r ^ r/] ’ C22-1)

where, for convenience, we will choose the origin to lie within the charge dis­
tribution. If r is large compared to the characteristic dimensions of the charge
distribution, we may expand Coulomb’s potential as follows:

1 1
—r' •v ! + l(r' •V )2- + ...
| r-r'| r r 2 r
1
+ + 1-lr »(3rV - lr '2) *r + ( 22 . 2)
r r 2 r
so that the potential, in its leading behavior for large distances, has the form

#/ \ e r •d 1 1
(22.3)
^(r ) = ¡: + — + 2 ^ r ' q< r + ....

•r

•r

Figure 22.1: Geometry of field point and source point for a bounded charge
distribution.

257
258 CHAPTER 22. MULTIPOLES

Occurring here are the first three moments of the charge distribution,

e = j(d v ')p (v '), (22.4)

d = y W ) r V ( r'), (22.5)

q= / ( d r O ( 3 r V - l r ' 2M r '), ( 2 2 .6 )

which are the total charge, the dipole moment vector, and the quadrupole mo­
ment dyadic, respectively.
Using this potential, we can now calculate the interaction energy of the
charge distribution with an additional point charge e\ located at a point r lying
far outside the charge distribution:

E = ei^(r) = + d• -f - e i -^-r •q •r + ---- (22.7)

We may alternatively interpret (22.7) as the interaction energy of the various

moments of the charge distribution with the field produced by e\ at the origin,
that is
E = 6<f> — d •E + —■V •q •E -f-. . . (22.8)
6
where

(22.9)

E = (22.10)

(We have seen this form of the dipole interaction energy before, in Section
4.1.) Note that the trace of q is zero, YLila = 0* This is a starting point
for considering the interaction of one charge distribution with another charge
distribution. For example, if one had a dipole di rather than a charge, e\)
interacting with a charge distribution which had only a dipole moment, d 2 , the
interaction energy deduced from ( 2 2 .8 ) would be

3r •d x r •d 2 — di •d 2r 2
E — —d 2 • ( 22. 11)
ip5

which is the dipole-dipole interaction.

Although this approach could obviously be continued indefinitely, it rapidly
becomes unwieldy for higher multipoles. A systematic approach can be based
on the use of spherical harmonics. Outside a charge distribution (r > r'), the
potential ( 2 2 .1 ) can be expanded in spherical harmonics according to (21.26);
that is, the expansion

4?r
( 22. 12)
ir 1 lm
27+TYrm( e ' , n
 259

implies

¿(*) = £lm ^ \ / w í í Ylm{9>

v
(22-13)
Here, the multipole moments, p/m, are defined by

Pim = J (d v ')r 'l^ ^ i YCm{0'A ')p {r'). (22.14)

The connection with the previous definition, (22.4) and (22.5), is, for example,
given by [see (20.34)]

(22.15)
05
II
o
o

(22.16)
<51
II
1

1
p\o = d Z} (22.17)

P i-i — ~^=(dx + id y ). (22.18)

Now we return to the consideration of the energy of interaction of a charge

distribution, p{r), with an external potential, <j){v):

E = J(dr)p(r)<f>(r). (22.19)

Since the potential is produced by sources outside of the charge distribution, it

can be expanded in terms of spherical harmonics,

^(r) = ^ r ' y , m( 9 » X ^ I ^ m) ( 2 2 .20 )

lm V

(j)]m being the expansion coefficients. Inserting this multipole expansion for the
potential back into (22.19) and using the definition (22.14) for the multipole
moments, we obtain the simple expression for the energy of interaction

E = '£pL<f>im, ( 2 2 .2 1 )
lm

generalizing ( 2 2 .8 ).
Rather than expressing the interaction energy in the unsymmetrical form
( 2 2 .2 1 ), let us formulate the energy in terms of the interaction of the charge
multipole moments of each distribution; that is, we seek a generalization of the
dipole-dipole interaction, (22.11). If we let ri and r 2 be measured from points
within pi and p2, respectively, while r measures the distance between these two
origins, as illustrated in Fig. 22.2, the interaction energy can be written as

E= /( d r i ) ( d r 2) f i ^ ^ 4 . (22.22)
J |r + r i —r 2|
260 CHAPTER 22. MULTIPOLES

Figure 22.2: Two interacting charge distributions.

Since the two charge distributions are non-overlapping, we can expand the de­
nominator occurring here in a double Taylor series:

1 (ri •V )'i (—r 2 •V )'* 1

(22.23)
|r+ ri - r 2| = £ /2! r
hh
We already know that, for r > r',

(22.24)
p h i = = E

or, if we take the complex conjugate and equate powers of r',

(22.25)
m

Further, recall the generating function for the spherical harmonics, (20.25),

^IT " = r/' £ V ^ T T (22‘26)

which is valid for a 2 = 0. Thus it is permissible to replace a by a gradient,

a —►V , (22.27)

as long as the derivatives act on 1 /r ,

aa _ * V 3i = 0 , r > 0. (22.28)
r
In this way, a comparison of (22.25) and (22.26) gives the identity

= (-l)V fc n (22.29)

where 'ipim is now regarded as a differential operator, constructed according to

(20.24) from V . (We might recall from Section 21.2 that and transform
22.1. PROBLEMS FOR CHAPTER 22 261

the same way under rotations.) Using (22.26) twice with the above replacement,
we obtain

(r i . v y x (- r 2 . v ) M , ,9
h\ h\ r ~ (~ 1} ^ ____

x ^ 2 \ j2¡x ^2)

X'0 /imi '0 /2^2 ~ • (22.30)

r

According to the definition of -0/m, (20.24), the product of two of these functions
is
V^imi = ^Iii2mim2 ^i+^,w i+m 2 ) (22.31)
where
1/2
(U + h + mi -b m2)! (h + h - rni - m2)!'
C\i l^rri\m>2 — (22.32)
.(¿i + mi)! (/i - mi)! (/2 + m2)! (¿2 - m2)!.

Then we evaluate the derivative structure in (22.30) by means of (22.29):

h+h ,
^/í+íajmi+ma —( 1)
2(/1 + /a) + r 11 h
(22.33)
Combining (22.22), (22.23), (22.30), (22.31), and (22.33), taking the complex
conjugate, and identifying p/m, (22.14), we find for the energy of interaction

47T l 1/2 l
E = Cl1l2mim2 /l + /2+1
y ( - 1 ) '1 .2(Zi + / 2) + 1J — r«

x (/> 1 ) l i f m (^ ) ^ )(/^ 2 )/a * W 3 *

(22.34)

If we were to set /^ = /2 = 1 we would rederive the dipole-dipole interaction,

(22.11). However, this is a completely general result for the interaction between
two arbitrary non-overlapping charge distributions, of a remarkably simple and
compact form.

22.1 Problems for Chapter 22

1. Derive the explicit connection between the quadrupole moment q defined
by (22.6) and p2m defined by (22.14).

2. Check that, indeed, ( 2 2 .2 1 ) generalizes ( 2 2 .8 ) by showing

E = Y , = e<£ — d •E + - V •q •E + ----
262 CHAPTER 22. MULTIPOLES

3. Derive (22.11) from (22.34).

4. Compute the thermal averaged interaction energy between two dipoles, of

dipole moment d\ and c?2 , separated by a distance r, in the high tempera­
ture limit. Express the result in terms of the average electric polarizabili­
ties, defined by
(d)'r = aE.

[Answer:

This result holds whether the dipoles are permanent or induced. This is
the high-temperature, or classical, limit of the van der Waals, or Casimir-
Polder interaction between molecules.]

5. Develop directly the expression for the energy of interaction, analogous to

(22.11), of two electric quadrupoles. Show that your result is consistent
with the general expression (22.34).

6. The solid harmonic is related to the spherical harmonic by

Yim(r) =

since it is a homogeneous polynomial of degree /. Similarly, we may define

a “solid” Legendre polynomial by

P¡(r •r') = r1P¡(cos y)r'1,

which is homogeneous of degree / in r and r'. The addition theorem reads

where (check this)

where the latter notation means the remaining terms have increasing pow-
ers of r 2r/2. Now replace r by V (with an operand 1 /r understood) and
arrive at

Check this for m = 0.

7. Consider
22.1. PROBLEMS FOR CHAPTER 22 263

and conclude that

(2Q! 1
tfm (V )- = (-1 )' V,m(r),
/! 2' r2i+1
from which the result (22.29) follows. Compare with the result of Problems
20.10 and 20.13.
Chapter 23

Conducting and Dielectric

Spheres

23.1 Interior of Conducting Spherical Shell

The spherical harmonics are useful in solving problems possessing spherical sym­
metry. In this chapter, we will solve a few such problems. First, we wish to
find Green’s function inside a hollow conducting sphere of radius a, which is
grounded, that is, the potential is zero on its surface. As usual, the Green’s
function equation is
—V 2 G (r, r') = 4ir<5(r - r'). (23.1)

The solution must be expressible in terms of spherical harmonics as

r < r' : G = £ r lYlm(9, , (23.2)

lm

r* < r < a : G = a2/+i 1 ^/m(0, ^)#/m> (23.3)

lm ^

where we have imposed the boundary conditions that

G — finite at r = 0 , (23.4)

and
G = 0 at r = a. (23.5)

To determine the expansion coefficients, Ajm and Bim, we use the equations for
the continuity of G, (21.18),

r"A ,m = ( ^ r r ~ ^ 2i+ r ) B,m’ (23-6)

265
266 CHAPTER 23. CONDUCTING AND DIELECTRIC SPHERES

and for the discontinuity of -§¿G, (21.19), [this uses (21.25)]

y ' | 7 /! +± !1 + !r
lr'l+1 \ „ , .,., ,
1 1 r" L ~
a 2 f+r ) Blm + lr A lm Ylm{6 A ) = 4 » £ y hl, ( M ) y ; ( i ' 1A
lm 1 V /m
(23.7)
at r = r; . Solving (23.6) by introducing C\m defined by

A,m = (p r p r - ^ 2¡+ r ) Clm> (23.8)

S/m = r'lClm, (23.9)

we find, from (23.7),

47T
C,m = Y L iO 'A ')- (23.10)
2 /+ 1
Therefore, Green’s function is

o - E ( ^ r - ^ ) « ~ T ) . P 3 11)

where we have used the addition theorem (21.27), y being the angle between r
and r', (21.8). Noticing that

rlr/l _ a rl
— > a > r, (23.12)
a^+T “ 7 (a 2 / r ' ) /+1 ’ r

we can perform the summation in (23.11) by use of (21.10):

a 1
G= (23.13)
|r-r'| r' |r — r'|

where r' locates the so-called image point,

r (23.14)

which, of course, lies outside the sphere. Thus we have achieved for the sphere
the analog of the image solution given for the conducting plane in (15.32).
What is the induced charge density on the inside surface of the sphere? This
charge density is proportional to the radial electric field, according to (11.69),

47TCT = —Er = — G , (23.15)

or r= a

since the normal is inward, and so in the negative radial direction. Differenti­
ating (23.11) with respect to r = r>, we obtain

4™ = - £ ( 2/ + l ) ¿ Pi (cos 7 ). (23.16)
i a
23.1. INTERIOR OF CONDUCTING SPHERICAL SHELL 267

Alternatively, we could use the image charge form of Green’s function, (23.13),
to derive
_________ 1 - (r '/ a f_________
(23.17)
[1 — 2 (r '/a ) cos 7 -f (r'/a)2]3/2

which indicates that as r' —* a, the only significant charge buildup is near 7 = 0 .
The total charge can be computed from (23.16) by use of the orthonormality
condition, (20.56), which, for /' = 0 , implies

/ sin y dy Pi(cos y) = 2Siq. (23.18)

Therefore, only the / = 0 term in (23.16) contributes to the total charge:

J a2 sin 7 dy d<j>
47ra2
= -l, (23.19)

which is the expected result. Of course, it is possible to use (23.17) to compute

the total charge, but that is more elaborate. (See Problems 23.1 and 23.2.)

23.1.1 Bessel Function Representation

Another approach to the interior of the sphere exploits the completeness of the
spherical Bessel functions over the finite range between r = 0 and r = a, as
inferred from (19.132),

1 c/ A r ' 2 r 1,2Ji+i/2Í7inr/a)r' 1/2J¡+i/2(7inr'/a)

---------------- i a - / y , V l /,<T,„ ) p ----------------
n=1

2 ji ( 7 /n r/a)j¡ (jin r'/a)

(23.20)
- E : y ¡(7 ,n )?

where now the y/n are the roots of j ¡ :

jl(7ln) = 0. (23.21)

Exhibited in (23.20) are the radial functions that are orthonormal in the sense
of the integral
/ d rr2R i„(r)R i„i(r) = Snn>, (23.22)
Jo
namely
11/2 ji(7tnr/a) (23.23)
JÍ (7>n)
This orthonormality property can be verified easily for / = 0, jo(t-) =
1 sini,
where (23.21) reads

sin Yon = 0 , Ton^O, (23.24)

268 CHAPTER 23. CONDUCTING AND DIELECTRIC SPHERES

or
7 on = m r, n = l , 2,.... (23.25)
Then we have

ñon(r)=(á) ( - 1 ) " ; sin T r

(23.26)

and the integral in (23.22) becomes

fa i 9 2 / • W7rr . n'7rr
/ dr r2“ ^ (—1 ) ^ s i n ----- sin -------
Jo a3 r2 a a

= 2 (—l ) n“ n / dt sin mrt sin (23.27)

Jo
The orthonormal radial functions are combined with the orthonormal spher­
ical harmonics:
Íímn(r ) — ^ím(^) ^)-R/n(^)> (23.28)
to produce the three-dimensional set of complete,

* ( * - * 0 = £ * » m » ( r ) t f mn(r ')) (23.29)

Imn

orthonormal,
I (dr) mn (r)<£i7m 'n ' (r) — fim m ' fifin' j (23.30)

functions that obey the differential equation

(23.31)
(V 2 + y?n/a2)<j),mn(r) = 0.
Consequently, the solution of the differential equation for Green’s function,
G {r, r'), is given by

(r)<f>*mn(r')
G (r ,r ') = 47
Imn tL / « 2

= £ ( 2 '+ i w = o s r ) 5 i # M 2 . (23.32)
Jn 7' " /a

Comparison with (23.11) then supplies the relation

v2. Rm(r)R*n(r') 1 , ( 1 ^ \
(23.33)
¿ í t,2
„ /« 2 ~ 2/+r< U 1+1 o2,+i y

presenting equivalent forms for the solution of the radial equation

fdr.2 +
+ H
r dr! _ E ±
r 2 i r ,U
1 ^ (r, r')
r; =-
V - r Or j, (23.34)
23.1. INTERIOR OF CONDUCTING SPHERICAL SHELL 269

with
gi(a.,r) = 0 . (23.35)

Approximations for the roots of the spherical Bessel functions are obtained
from (19.125) by writing m a s i + | ,

7,„ * (I + 2 „ ) f - £ $ ± 2 - 1) * f t (23.36)

as illustrated by

(23.37)
3tt3 (n + 1/2 ) 3

The approximation to 711 produced in this way is 4.4938, which is in excess by

0.009% of the correct number,

711 = 4.4934. (23.38)

Incidentally, as in the discussion of (19.132), the leading term in (23.36) is used

to convert the summation of (23.33) into an integral, in the limit a —> 00 , thereby
recovering what is displayed in (21.83).
Some useful statements about the 7 /n are obtained from (23.33) through a
general process that is most simply illustrated by setting r' = r and integrating
(r2dr) from 0 to a. The normalization of the R]n then produces

A a2 _ 1 2/l r21 \ a? (I 1 \
39)
^ 7 ,2„ 2/ -h 1 70 dVr ( r a2, +1) 2/+1V2 2/ + 3 J ’ ( '

or
V 'J_-- I01
-v?
1
(23.40)

For / = 0, 7 on = nir, this yields a familiar summation,

^ 1 1
(23.41)
2—1 (mr)2 6 ’

which is usually derived from the pole expansion of cott,

00 1 1 00 1
cot (23.42)
^2 i — n7r f ^ ^2 ( n7r) 2 __ f2 ’
n = —00 n= 1 v 7

the comparison o f the t terms in the small t expansion of the two sides yields
(23.41).
The second in the unlimited sequence o f such statements considers the square
of (23.33), integrated over both r and r'. Here, the orthonormality of the R¡n
270 CHAPTER 23. CONDUCTING AND DIELECTRIC SPHERES

yields

00 4 1 Ca Cr
r' 2r/2/
^ = 2w ^ l drr,l dr' r'+i a2* * 1
1 1 ra 0 / / r \ 2/ +
2
i\ 2

= 2W Í W ü T > l i r r { 1 - U )• (23A3)
where the symmetry in r and r' permits one to simply double the contribution
of the integrals for r* < r. The outcome of this integration is given by

1
V 1 = 1 1 (23.44)
¿^-s7 (4„ 2 (Í 2 /+ 3)2 2 /+ 5 ’
2

which is illustrated, for / = 0 , by

^ 1 _ 1
(23.45)
~r[ ( n7r) 4 90 ’

as also follows from the t3 terms o f (23.42).

One naturally asks for a generalization of the expansion (23.42) to / ^ 0. It
is given by the pole expansion of j¡(t)/ ji(t) which, for / = 0, is co tt —l/t. First,
we need to notice that j\(t), as a numerical multiple of t~ l/2J¡+1 / 2 O0 >behaves
as tl for small t. That gives the pole expansion in terms of the roots of j¡, yin ,
as
m = i 1 (23.46)
m t
The required power series for ji(t) can be derived from (19.151) and (21.68),

sin t
M *) = ~T
^2/+ 2fc
( 2 / + 1 + 2k)\
x 7 Jk=0
00
(l+ k )\ 2k (23.47)
= (2<)' ¿ ( - 1)kfc! ( 2 / + 1 + 2 *)!
0

or, writing out the first few terms,

/! 1 1
M<) = ( 2 Í)' 1 - 1 V t4 + . . . (23.48)
( 2 Í + 1 )! 2 21+ 3 + S (21 + 3)(21+ 5)

Then the expansion produced by differentiating the logarithm of j¡(t) is

m = ____________ <3 , (23.49)

ji(t) t 2 /+ 3 (2/ + 3) 2 (2/ + 5) ’
23.2, EXTERIOR OF CONDUCTING SPHERE 271

from which we regain (23.40) and (23.44) by examining the t and t3 terms in
the power series expansion of (23.46). We leave it to the reader to carry out the
next step and derive the sum of the inverse sixth powers of the y¡n [Problem
23.3].
One use of these summations,
oo 1
(23.50)
n= 1 l¡n
comes from the evident inequality

rW (23.51)
’ N r
or
1
7/i > (23.52)
(<r,(r)) l / 2r ’
these are lower bounds to the smallest of the 7 ’s for a given /.
As an example, consider (23.41), (23.45), and the next member of the se­
quence,
00 1 1
S ( ^ = 945' <23'53>
n= 1 v 7
which yield, successively,

7T > 2.4495; 3.0801; 3.1326; . . . , (23.54)

rapidly converging from below to the limiting value. One can also improve the
last member of such a sequence by an approximate extrapolation to the limit
that employs it and one or more of the preceding numbers. [See Problem 23.4.]
We defer the discussion of another application of such sums, one that focuses
on the second smallest 7 value. [See Chapter 25.]

23.2 Exterior of Conducting Sphere

Now let us use the surface charge density in setting up the conditions to be
satisfied for a unit charge at the point r; exterior to a conducting sphere S of
radius a, where the origin of coordinates is taken to be the center of the sphere:

* (n )
dSi
G(r'r,)- l v h i + l |r-rj|’
G {r, r') = 0 , r on S. (23.55)

Implicit in this is the restriction to the external region, r, r* > a.

The individual potentials that appear here are

»<
00 7
Alt
a
1
(23.56)
|r —r'| ¿ ¿ « 2 7 T T m= —l
272 CHAPTER 23. CONDUCTING AND DIELECTRIC SPHERES

and (r > a)

Js dSl\ r - r i| = 1 (23.57)
1 A| 1=0 m

where
«■im = J dS\Y*m(6\, ^i)<r(n). (23.58)

The requirement that G, the sum of these components, vanishes for all 6 and </>
at r = a leads immediately to

(23.59)

and we have learned that

00 / n2¡+l \ i
G(r,rO = ¿ K - y + j-J ^ P / Í cosy) (23.60)
i=o v r< ' r>
which is the analog of (23.11).
We move toward an equivalent form by writing this as
1 -/<
G (r ’ r') = IT— J] " £E ¿ ¡ T P'( cos7)> (23.61)
1=0

where
-/ u
r = — < a. (23.62)

Then, with the introduction of the vector r', with magnitude r' and the direction
of r', presents Green’s function in the image form
1 a 1
G (r,r') = (23.63)
|r —r'| r' |r —r'| ’
the same as (23.13).
With the information contained in (23.59), that

‘Um = - ^ TTYim( 0 ' , n (23.64)

we can construct the surface charge density. According to the definition (23.58)
and the completeness of the spherical harmonics, it is [dSi = a2dQ]

<r(v) = ] T y ím(M )¿ < 7 im = - ¿ r ¿ ( 2 / + l ) H ^ l)- (23.65)

Im 1=0

One should notice, however, that an aspect of the surface charge distribution,
the total induced charge, is given directly by croo:

Qind — d s cr(r ) = y/iircroo — (23.66)

23.2. EXTERIOR OF CONDUCTING SPHERE 273

it is just the image charge displayed in (23.63).

An alternative approach to the charge induced on the conducting surface is
provided by the normal component of the electric field,

(23.67)
= ¿ ( - | G(ry))
The introduction of the Green’s function form (23.60) immediately reproduces
(23.65). Now let’s use the image version, (23.63), which we write as

[r/2 — 2 rr' cos 7 -f r 2] 1/2 [(r r '/a ) 2 — 2 rr'COS7 -f a2] 1/2 (^3-68)

thereby making quite transparent that G vanishes at r = a. This gives

l a 1 — (a /r ' ) 2
(23.69)
47ra2 r' [1 — 2 (a /r ') cos 7 + (a /r ' ) 2]3/2 ’

and we note particularly the ratios of the surface charge density to its average
value, ( —a / r ') / 47ra2, at the point nearest the unit charge:

14 .1
7= 0: (23.70)
(i-^ r

and at the point farthest from the unit charge:

7 = tt : w < 1. (23.71)
(i + f ) 5

Of course, we also have

1
— id S a -(r) = ^ f ¿¿(cost ) rj— = i. (23.72)
ind J
Qind 2 J_x [1 - 2 (a /r /) cos 7 + (a /r ' ) 2]3/2

The equivalence of the two surface density forms, as presented by (t = a /r ')

1 _ f2 00

' < 1 : [ i - 2„ r + , r » = s <2' + ( 2 3 -73)

is especially interesting in the limit t 1. For fi < 1, the limiting value of the
left-hand side is zero [as illustrated by (23.71)], whereas with /jl = 1 [see (23.70)]
the limit is infinite. And it is the content of (23.72) that, independently of the
value of t , half of the /i-integral of the left-hand side, from — 1 to 1 , is equal to
one. In short, we have learned that

(23.74)
274 CHAPTER 23. CONDUCTING AND DIELECTRIC SPHERES

where unity in the argument of the delta function is to be understood as 1 — 0 .

This can be recognized as the completeness relation (21.25), rewritten with the
aid of the addition theorem (21.27),
°° 9/ -4- 1 1
E ^ r p ,(cos ? ) = — (23.75)
sin#A e ~ W i t - *'),
1=0
where the right-hand side is expressed in terms of 7 as

= (23.76)
S in 7 Z7T Z7T

As has been done so often before, we use G to determine the energy of

interaction between a charge e, stationed at ro, and the conducting sphere:

Eiint = Le2[
2 G ( r , r ' ) - G 0(r,r')]
1
2 r0 r0 - (a 2/ r 0)

(23.77)
2 ” r l - a 2‘
This is just the Coulomb energy between the charge e and the image charge
—(a/ro)e, multiplied by the characteristic factor of 1/2. [Recall Section 17.6.]
The magnitude of the force of attraction between charge and conductor,
Q ar0
F = ^int = e2
Eint
dr0 ( r 2 - a 2)2

ell —(a/ ro)el

(23.78)
[r 0 - ( a 2/ ro )]2 ’

is that between charge e and its image charge.

23.3 Conducting Plate and Hemispherical Boss

Let the conducting sphere be bisected by a plane conducting sheet of unlimited
extent, thereby producing the situation of a hemispherical boss raised over a
plane surface. We choose the coordinate system, with its origin at the center of
the sphere, such that the conducting sheet lies in the x-z plane. The range of
spherical coordinates for this exterior problem is: r > a , 7r > # > 0 , 7r > < ^ > 0 .
Now, G is required to vanish at r = a, the surface of the boss, and at 4> = 0, 7r,
the surface of the sheet. As for the latter requirement, we already know that it
implies the substitution [(17.60), with z , z1 —* (j), f t ; a —> 7r; n —►m]

2 00
6(<j) — (j)1) —►— sin m<j) sin m<¡)1
7T
m= 1
^
00 i
(23.79)
= E s
23.3. CONDUCTING PLATE AND HEMISPHERICAL BOSS 275

Figure 23.1: Sphere bisected by plane. Shown are the locations of the physical
charge at r' and the image charges at r', , and rfy .

which exhibits the equivalent image charge structure at reversed values of <j>') or
of y'. Accordingly, if images in the plane y = 0 are indicated by the subscript
y, Green’s function for this external situation is (see Fig. 23.1)

1 1 a / 1
G( r ,r ') = (23.80)
k - r '| lr ~ ryl r>

As an application, let charge e be placed on the positive y-axis, at distance

ro > a from the origin. The image charges are also located on the y-axis, their
charges and y-coordinates being: —e, —ro; —e(a/ro), a 2/ro; e(a/ ro)> —a2/ro.
The magnitude of the attractive force between charge e and the conductors is,
then,

a/r0 a/r0
F = e2 (23.81)
_(2 r 0) 2 [ro - {a2/r0)]2 [r0 + (a 2 / r 0)]2.

One notes that as ro approaches a the first and third terms of F tend to cancel,
leaving just the force associated with the sphere, whereas, in the limit of large
ro/a, it is the first term, associated with the conducting plane, that dominates,

a/r0 < 1 X (23.82)

■+ 4 - r
( 2 r0)‘

It should also be observed that, while the sum of the second and third
terms in (23.81) is positive, so that the force always exceeds that produced by
the conducting plane alone, a similar remark about the first and third terms,
and the force produced by the conducting sphere alone, requires that ro be
sufficiently large. Indeed, the contrary circumstance,

a/r0 ^ 1
(23.83)
[r0 + (a 2/ r 0)]2 > ( 2 r 0) 2 ’

or (x = ro/a )
4x3
> 1 (23.84)
(x 2 + 1 ) 2
276 CHAPTER 23. CONDUCTING AND DIELECTRIC SPHERES

Figure 23.2: Geometry of dielectric sphere with source point outside.

occurs in the range

1 < x < 3.383; (23.85)

here the effect of the conducting plate is to reduce the attractive force that the
conducting sphere alone produces. The largest reduction occurs for ro/a = 2.297
where it is 8.7%.

23.4 Dielectric Sphere

To move beyond the consideration of conductors as examples of the use of spher­
ical harmonics in solving Green’s function problems with spherical symmetry,
we consider a dielectric sphere of radius a, with a unit point charge outside, as
shown in Fig. 23.2. In this case, the Green’s function equation is

r > a: —V 2G(r, r') = An8(v — r'), (23.86)

r < a: - V 2 G (r ,r ') = 0, (23.87)

where we will take e to be a constant in the interior of the sphere, r < a. The
boundary conditions at r = a are, from (11.58) and (11.60),

G is continuous, (23.88)

and
(23.89)
~€T r G

The conditions on G at r = rf are as given in (21.18) and (21.19),

G is continuous, (23.90)

and
r = r ; +0
= 47r-T^7:<5(0 - 0 ' W ~ *')• (23.91)
r = r '-0 SlIid
23.4. DIELECTRIC SPHERE 277

As is familiar by now, the solution in the three regions has the form

r < a: g = £ > ' y /m(M M im , (23.92)

lm

r > r* : G = 5 > “ ' ' ly ft»(0 ,* )A m , (23.93)

lm

a < r < rf : G = ^ ( r ' S t a + r - ,- 1 Q m)Y(m( M ) . (23.94)

lm

It is very easy to find the expansion coefficients by use of (23.88)-(23.91):

4 ^ ( 0 0 (23.95)
Aim
l(e + 1 ) + 1
,-,-r
Blm (23.96)
2/+1
( e - 1 )/ 4-KY*m(6', <f>') a 2,+1
C]m (23.97)
l(e + 1 ) + 1 2 1+ 1 r ' ,+ 1 ’

4
Dim (23.98)
]m+ 21 + 1 r •
Green’s function, outside the sphere, is therefore found to be
CO
(e - 1 )/ a21* 1
- E l(e + 1 ) + 1 rl+l r,l+l Pl{ cos 7 ).
1
r,r' > a : G( r, r') (23.99)
|r - r'|

We now ask what is the leading behavior of this potential when the separa­
tion between the point charge and the sphere is large compared to the radius
of the sphere, r' a. Since the Ith term in the sum behaves as ( a /r ') i+1, only
small values of / contribute. The leading contribution arises from / = 1,

1 /N 4 e — 1 a3
r a: G (r ,r ) ~ ------- - 7 ---------- cos 7 . (23.100)
v ’ ' | r-r'| e+ 2 rV 2 1

Since 7 is the angle between r and r',

r *r'
cos 7 = , (23.101)
rr
this asymptotic form o f Green’s function can be rewritten as

G (r ,r O = Í - T - 7 7 + 4 *d, (23.102)
|r —r'| ró

the two terms of which have simple physical interpretations. The first term
is due to the point charge while the second is the potential arising from the
induced electric dipole moment of the sphere [cf. (22.3)]. The latter is identified
from (23.100) to be
(23.103)
278 CHAPTER 23. CONDUCTING AND DIELECTRIC SPHERES

where —r '/ r /3 is interpreted as the electric field, Eo, at the center of the sphere
(in the absence of the dielectric) produced by the unit point charge. Since this
electric field is essentially constant over the sphere, we recognize that the electric
dipole moment induced in a dielectric sphere of radius a by a constant electric
field Eo is
d = ^a 3 E0. (23.104)
e+ 2
Finally, we write the expression for Green’s function inside the sphere:

r < a } 00 rl 2/ + 1
: G( r ,r ') = £ Pi (cos 7 ). (23.105)
r' > a r ,,+ 1 l(e + 1 ) + 1
' ;=o

Again, in the situation in which the point charge is located far from the sphere,
r' a, low values of / predominate:

I 3 r«r'
G ~ ^ 2 r E 0 = i —r •E, (23.106)
r' e + 2 r '3

where we identify the electric field in the dielectric as the negative gradient of
G, so that the field E in the dielectric,

E = 7 T 2 Eo’ (23-107)

is less than the applied field Eo if e > 1. This is equivalent to (5.87), (5.93)— See
Problem 23.7.

23.4.1 Interior of sphere

We illustrate a slightly different method of computing Green’s function for the
circumstance when the point charge is inside the dielectric sphere, rf < a. We
recall that if all space were filled with a medium of dielectric constant e, the
Coulomb potential would be reduced by the factor e. Then, if we define the
reduced Green’s function by

G (r ,r ') = Y ^ 9l(r,r')P i( cos 7 ), (23.108)

1=0

we recognize that surface charge density on the sphere induced by that Coulomb
potential at r> = a is proportional to (r< = r')*, and so we have the forms for
the reduced Green’s functions in the two regions:

1 r Ufi
V
rf < a, r < a : 9i(r,r') = 7 - f i j + Pi
e r! e a /+1
2 1

1 r"
rf < a, r > a : gi(r, r') = rj (23.109)
e r H-l •
23A. DIELECTRIC SPHERE 279

Here pi and 77 are numerical coefficients to be determined. Now we impose the

boundary conditions at r = a: the continuity of the potential that is Green’s
function, or
\ + p i = ru (23.110)
and (in the absence of a surface distribution of free charge) the continuity of
the radial component of D , or

e ( /+ 1 - I p i ) = (/ + l)n - (23.111)

Combining these equations, we obtain

( e - l ) ( / + l) _ 2J+1
(23.112)
P, ~ ( c + l ) / + l ’ T,_£(e+1)/+1’
and so the reduced Green’s function is
1 ( e - !)(/+ !) rl \
r' < a, r < a : <//(r, r') = - r <
r '* 1 (c +1)1 + 1 a«+* ) ’

/ < a, r > a : g,(r, r') = + 1 J +1 • (23-113)

For comparison, let us write down the reduced Green’s function for the
previously considered situation in which the point charge is exterior to the
sphere:

r' > r > a: #/(r, rf) ( r< . (‘ - D ' ° WIN| J _

V<
a,
(>: +1)1 + 1
21 + 1 rl
rf > a, r < a: g i ( r , rf) (23.114)
(c -+-!)/ + ! r '^ 1'
One recognizes here the symmetry in r and r1 of g\{r) r ,) ) in particular, that
between the two situations r < a, r' > a, and r > a, r' < a; it is the radial aspect
of the symmetry of G (r ,r ') in r and r'. As an illustration of that symmetry
consider r = 0 , r' > a, where all gi vanish except

5o(r = 0 y > a ) = i (23.115)

r
Thus, according to (23.108), the potential, at the center of the dielectric sphere,
that is produced by a unit point charge at distance rf > a is just 1 /r ', indepen­
dently of the value of the dielectric constant. We can understand that through
the symmetry of Green’s function, which asserts that what we have just consid­
ered is also the potential produced at r' > a by a unit charge at the center of
the sphere. But Gauss’ theorem, in the form

/ dS •D = 47T Pi (23.116)

assures us that the radially directed electric field at distance r', in the vacuum
surrounding a spherically symmetrical arrangement containing a unit charge, is
indeed just that derived from the potential 1 /r '.
280 CHAPTER 23. CONDUCTING AND DIELECTRIC SPHERES

23.5 Problems for Chapter 23

1. Derive (23.17).

2. Use (23.17) to compute the total charge on the sphere.

3. By carrying out the expansion in (23.49) to order t5, find an expression

for the sum ]T}n y¡^6. As an example, extablish (23.53).

4. Verify the lower bounds for 1r shown in (23.54) and carry out the improve­
ment referred to at the end of Section 23.1.1.

5. Construct Green’s function, from its differential equation, for the region
exterior to a conducting sphere. [That is, use the method given in Section
23.1.] Give the image interpretation of the result. What is the physical
significance of the two leading terms when one point is very far from the
sphere?

6. Calculate Green’s function for a dielectric sphere when the point charge
is inside the sphere, r' < a, using the direct discontinuity method given
in Section 23.4.

7. Recall our discussion of the Clausius-Mossotti equation in Chapter 5.

Show that Edriving discussed there may be identified with E o , and derive
the formula for the induced dipole moment, (23.104), from the considera­
tion in Chapter 5.

8. Consider the limit in which e —» oo, so that the dielectric sphere discussed
in Section 23.4 may be regarded as a conductor. Give the form of the
Green’s function in that situation for r' > a by taking the e —* oo limit
of (23.99) and (23.105). Show that indeed the interior of the sphere is an
equipotential region, but not one of zero potential. Also, show that in this
case there is zero charge on the conductor. Show that by adding a suitable
charge distribution to the sphere one can recover the situation considered
in Section 23.2.

9. Consider a dielectric cavity, that is, a spherical region with 6 = 1 embed­

ded in an otherwise uniform dielectric medium. As in the case of plane
surfaces, this is not a new situation. Green’s function can be obtained,
first, by replacing e by 1 , and then increasing the scale of all dielectric
constants by the factor of e, which decreases G by the same factor. Obtain
the reduced Green’s function g\ in the interior of the sphere r < a for r'
either inside or outside the sphere. In the latter case, by letting rf a,
show that the field within the cavity is uniform,

e = 5 T T e <°>'
where E(0) is the field that the unit charge would produce at the origin
if there were no cavity in the dielectric medium. Also, discuss the limit
e —* oo.
23.5. PROBLEMS FOR CHAPTER 23 281

10. What is the statement of completeness for the functions 0 /m(0), as inferred
from that of the Y¡m(0, <¡>)1 Consider a conducting sphere that is bisected
by an infinitely thin conducting partition, as discussed in Section 23.3.
Find Green’s function for the interior of the conducting hemisphere.
Chapter 24

Dielectrics and Conductors

24.1 Variational Principle

In Chapter 11, we investigated the stationary properties of the electrostatic
energy when only dielectrics are present, that is, we had a stationary principle,

6E = 0 , (24.1)

where [see ( 1 1 . 1 1 )]

E = / ( * ) [* ¿ + ¿ ( e . v * + ! e 2) (24.2)

We now wish to generalize this situation to include conductors as well. The new
feature here is the existence of surface charges on various conductors implying
an additional contribution to the energy:

E = + ¿ ( e .v * + Ie’) ] + ¿ / ( ís ^ , (24.3)

where the volume integral extends over all space exterior to the conductors and
the surface integral is over all of the conductors, cr being the surface charge
density. (See Fig. 24.1.) This energy functional is to be supplemented by the
condition that the total charge on each conductor,

Qi - j dSior, ¿ = 1 , 2 , . . . , n, (24.4)

is fixed. The electrostatic problem is completely specified by the location of the

conductors and dielectrics, the free volume charge density, p, and the charge on
each conductor, Qi. Note, in particular, that the surface charge density, cr, is
to be determined dynamically. The change of the energy under variations of <£,

283
284 CHAPTER 24. DIELECTRICS AND CONDUCTORS

conductor conductor
Si S2

dielectric

Figure 24.1: Conducting surfaces embedded in a dielectric medium.

E, and a is

8E = J (d v ) ^p8<f> + ^ (5E •V<{> + E •V8ij> + E ♦¿E)] dSi[8a<j)-\-<T8<j)],

(24.5)
which is subject to the condition that Qi be constant, that is

dSi 8a = 0. (24.6)

We rewrite the T7 8<f>term by means of an integration by parts, which makes use

of the identity
D /D \ V-D
= . (24.7)

The implied surface integral here cannot be discarded since now there are contri­
butions arising from the surfaces of the conductors. If we let n¿ be the outward
normal on the ¿th conductor, this surface term is

J (d r )V - ( £ « * ) = - ■ £ (24.8)
t

The variation in the energy, (24.5), now reads

8E = [ (dr) \p8<¡> + ^ -8 E •V<¿ - 8 <¡>—A— + J _ d . 6 E

J L 47T 4tt 4?r _

+V f dSi +8 a<j>+<j8<
j>. (24.9)

The requirement that the energy be stationary under independent variations in

<j) and E then implies, in the interior of the dielectric,

8<j> : V •D = 47rp, (24.10)

<SE : E = —T7<^, (24.11)
24.2. RESTRICTED FORMS OF THE VARIATIONAL PRINCIPLE 285

while just outside the surface of the conductors,

8<j) : n * D = 47rcr. (24.12)

Finally, the variation in the surface charge density requires

6E = ? / dSi6a<t> = 0 (24.13)

which is subject to the restriction (24.6), implying that each conductor is an

equipotential surface,
<j) = constant on Si = <¡>i. (24.14)
Thus, the stationary action principle, based on the energy functional (24.3),
yields all the physical laws governing electrostatics in the presence of conductors
and dielectrics.

24.2 Restricted Forms of the Variational Prin­

ciple
As in Chapter 11, there are two restricted forms of the variational principle we
may discuss. In the first, we take the electric field as being defined by

E = V(f>, (24.15)

so that the energy functional becomes

E[4>, V] = J (dr) [ri - ¿ ( W ) 2] + £ J dSi v<j>. (24.16)

The independent variables are <j> and ít, the latter of which is subject to the
condition (24.4). For the second form, D is regarded as an independent variable,
subject to the condition

V •D = 47rp, inside dielectric, (24.17)

while cr is determined by

n •D = 47tí7 , on surface Si. (24.18)

To rewrite the energy as a functional of D only, we integrate by parts on the

D •V<^ term in (24.3) and use (24.17) and (24.18) to obtain

E[D] = / ( * ) ¿ ^ . (24-19)

while the subsidiary condition (24.4) becomes

(24.20)
=/ dS,si r
286 CHAPTER 24. DIELECTRICS AND CONDUCTORS

In (24.19), we identify the integrand as the energy density of the field.

Let us now verify that the second restricted form of the variational principle
correctly describes the electrostatic situation under consideration. For a finite
change in D , D —» D + A D , the change in the energy functional (24.19) is

i £ = / < * > ¿ T -A D + / ( * ) ¿ ^ 2 1 1 , (24.21)

while the constraints read

V •A D = 0, (24.22)

J dSin- A H = o. (24.23)

The stationary condition requires that the integral linear in A D in (24.21)

vanishes. To incorporate the constraint (24.22), we add to (24.21) the volume
integral

0 = / < * > T T v •i D = - E / « - /(* > 1 7 'iD (24'24>

i

where <¡>(r) is an arbitrary function. Likewise, to incorporate the constraint

(24.23), we add to (24.21) a sum of surface integrals,

r—^ f n •A D
0 = ^ 2 <f>i I dSi —— — (24.25)
i J

where <j>i is an arbitrary constant. In the resulting form of A E , the variations

A D can be regarded as independent, so that the stationary principle implies,
in the volume,
j = E = -V<f>, (24.26)

while, on the surfaces,

(¡> = (f>i. (24.27)
In this way, we recover the full set of equations for electrostatics. Moreover,
(24.21) also tells us that, in going from the correct field configuration to any
other, A E > 0 , that is, the physical field minimizes the energy functional
(24.19). This is a statement of Thomson’s Theorem: The charges on the surfaces
of conductors always readjust themselves in such a way that each conductor be­
comes an equipotential surface and the total energy of the system is a minimum.

24.3 Introduction of Additional Conductor

We now consider a region of space with dielectric constant e(r) bounded by
an array of conductors into which we introduce an uncharged conductor at a
location where there is no free charge density. We are interested in the change
24.3. INTRODUCTION OF ADDITIONAL CONDUCTOR 287

(a) (b)

Figure 24.2: Introduction of an uncharged conductor into a region where there

is no free charge density.

of energy in going from the initial configuration (a) to the final configuration
(b). (In the following, the subscript 0 refers to the introduced conductor.) See
Fig. 24.2. The energy for (a) is

Ea (24.28)

where V is the volume exterior to the conductors and the charge on the ith
conductor is
J dSi = Qi • (24.29)

For (b) the energy is

E t= [ (dr) £ * -, (24.30)
Jv-Vo bTre
where now the volume occupied by conductor 0 (Vo) is also excluded, and the
charges on the conductors are

Jds“iir”0" (24.31)

I dS^ L=»• (24.32)

The energy, for case (a), satisfies the following inequality,

Ea (24.33)

Although D a is not the correct field for (b), it is an allowable trial function
to use in the energy functional, (24.19), because it satisfies all the necessary
conditions:
V -D „ = 47rp, (24.34)

/
n «D a
dSi = Q i, (24.35)
47T

/ dSo
n *D a V -D a
= 0, (24.36)
47T 4w
288 CHAPTER 24. DIELECTRICS AND CONDUCTORS

since, by hypothesis, the region Vo originally had no charge (p = 0). According

to Thomson’s Theorem, the correct field yields a minimum value of the energy
functional,
(24.37)

implying, upon comparison with (24.33),

Ea > Eb (24.38)

which states that the introduction of an uncharged conductor into a charge-free

region lowers the energy of the system.

24.4 Alternate Variational Principle

In the first restricted version of the variational principle, (24.16), the charges on
the conductors, (24.4), are specified. For some purposes, it is more convenient to
regard the potentials, fa, on the surfaces of the conductors as specified, rather
than the charges, Q¿. For simplicity we will assume that there is no volume
charge density, p = 0. In order to obtain a new form of the energy functional,
we note that the stationary property of (24.16) under the replacement

<j) —*•A<j>, (24.39)

for A infinitesimally different from unity, implies

0 = ^ x 1= 2 J (dr) ( " ¿ ) (W)2 + E / 'dSi a*' (24-40)

Consequently, the energy functional is

(24.41)

which becomes, for the actual field values on the surfaces, <j>=

(24.42)

Therefore, we obtain another energy functional by combining (24.42) and (24.16),

E = Qi<i>i - J 2 j dS<v * + j ( d l) ( 2 4 -4 3 )
% l

or, using (24.4),

e [<i>, <a = T j j d 5 '*<T( ^ - 4 ) + J (< fr) ¿ ( v ^ ) 2 - (24.44)

24.5. GREEN’S FUNCTION 289

Here we regard (j> and a to be the variables while fa is specified [note that here
we impose no subsidiary restriction on a]. Under variations in (j> and a the
energy changes by

SE = £ J'dSi[6ff(4i <t64>] + J (d r ) . (V t y ), (24.45)

i

which becomes

SE = ^ 2 J ¿Si ^8a(fa -</>)- (t8<¡) + -f j (dr) - 6<j), (24.46)

by identifying D and integrating by parts. The stationary principle, 6E = 0,

implies, from the volume part of (24.46),

8<t>: V . D = 0, (24.47)

and from the surface part,

8<¡> : n*D=47Tcr, (24.48)

8a : (¡> = fa. (24.49)

These are the correct equations of electrostatics when there is no volume charge
density.

24.5 Green’s Function

In our study of electrostatics, we have found Green’s functions to be of great
use. We will here introduce Green’s function in the presence of conductors that
are grounded, that is <j>i = 0 ; the corresponding differential equation is

- V . [e(r)V G (r, r')] = 4ir8(r - r'), (24.50)

with the boundary condition

G (r ,r /) = 0 fo rro n S ¿. (24.51)

We will show that this Green’s function can be used to solve the electrostatics
problem in which the potentials on the conductors are specified. We wish to
consider a situation for which the free charge density is zero,

V •D = - V . (eV<j>) = 4ttp = 0, (24.52)

while the potential, fa on each conducting surface, 5 ¿, is constant,

= fa on Si. (24.53)

If we multiply (24.50) by <^(r) and (24.52) by G (r,r') and subtract, we obtain

- f a r )'V . [e(r)V G (r, r')] + G(r, r ')V . (c(r )V fa r )) = 4ir8(r - r')far). (24.54)
290 CHAPTER 24. DIELECTRICS AND CONDUCTORS

Since the left-hand side of (24.54) is a divergence,

V •[G(r, r')e(r) V</>(r) — <£(r)e(r)VG(r, r')] = 47r6 (r —r')^ (r), (24.55)

when we integrate over the entire volume, V, exterior to all of the conductors,
we obtain an integral over a surface S which is made up of all the surfaces of
the individual conductors, S¿,

47r0 (r/) = j dS * [G(r, r')e(r) V<¿>(r) — <^>(r)e(r)VG(r, r')]

= - y ; J dSi[G(r,v')e(r)ni *V4>(r) - <^(r)e(r)n¿. V G (r, r')].

(24.56)

The negative sign occurs since dS is directed out of the volume V, and so into
the conductors, while n* is the outward normal for the ¿th conductor. Deleted
here is the surface at infinity for which

dS ~ R2,

G < I | V G | < -^ ,

\V<j>\<±, (24.57)

so that the corresponding contribution goes to zero as the volume gets arbitrarily
large. Now imposing the boundary conditions, (24.51) and (24.53), we obtain
the desired expression for the potential,

r) = / dSi ' V ' G (r -r')- (24.58)

i
where we have interchanged the roles of r and r' and used the symmetry property
of G, (12.7). (See Problem 24.2.) Therefore, if we know G (the potential due
to a point charge with zero potential on the conductors), we can calculate the
potential, <¿>(r), due to arbitrarily specified potentials on the conductors. [This
result, and its generalization to the situation when the free charge is not zero,
was anticipated in Problem 12.3.]

24.6 Capacitance
Once we know the potential, we can compute the surface charge density on the
zth conductor by using

= ¿ n*•

= - 1 > / dS'i 1 7 ! ? ( “ <•v Xni •v ' W r ’ r,)- (24-59)

24.6. CAPACITANCE 291

The total charge on Si is therefore

Q i= j dSi * = - £ h J dSi •V )(n ' . V ') 6 '(r, r'). (24.60)

i

Occurring here are the coefficients of capacitance, C¿¿, defined by

Cii = -J dSidS'j * V )(nJ •V ')G (r , r'), (24.61)

which are purely geometrical quantities that are symmetric in i and j,

Cij = Cji. (24.62)

The total charge on the ¿th conductor is thus simply written as

Qi = Y 2 Ci^ i - (24.63)
j

The energy of a system of charged conductors can be expressed in terms of

the coefficients of capacitance by means of (24.42),

E = \ n Q i*i = \ E • (24-64)
i ij

There is a consistency check between this expression and the variational prin­
ciple which employs (24.44). Suppose we vary the potential on conductor i by
an amount 6(pi. Such a change induces variations in a and <p but the resulting
change in the energy from these induced variations is of second order due to the
stationary principle. So the first order variation in the energy arises only from
the explicit variation of

6E = J dSi <t6<f>i = 64iQi, (24.65)

or,

i s = « ‘. (24.66)

This result is in agreement with that obtained from (24.64),

BE
Q fi = ' E ' c n *i = Qi > (24-67)
j

where we make use of (24.63).

Suppose the system consists of a finite region bounded by conducting sur­
faces, as in Fig. 24.3, that is, there is no surface at infinity. The total charge
292 CHAPTER 24. DIELECTRICS AND CONDUCTORS

induced on such a system of grounded conductors by a unit point charge at any

interior point, r', is

“ ¿ 1 C / ¿5 ic(r)n i . V G ( r , r /) = J (d r) ( - V ) . [f(r )V G (r,r')j = - 1 ,

(24.68)
where we have used the first line of (24.59), with <j) replaced by G, as well as the
differential equation satisfied by Green’s function, (24.50). This implies that
the coefficients of capacitance, C¿¿, (24.61), satisfy

E C« = - / dSi T r K •'V ') E / dS<! ? ( ■ * •■ ' W . r')

i i

= - J •V ') ( l ) = 0 , (24.69)

that is, the sum of all the coefficients of capacitance referring to a given con­
ductor vanishes,
(24.70)
* 3

Consequently, the total charge on the conductors is zero when there is no volume
charge present:
J 2 Q í = J 2 C í^ j = 0. (24.71)
i ij
Furthermore, for this system, only relative values of the potential are significant.
If we were to add a common constant to all potentials, all charges would remain
the same:
Qi = + constant) = (24.72)
3 3

As a simple example, consider a closed system bounded by only two conduc­

tors. In this case, in order to satisfy (24.62) and (24.70), we must have

C n — —C 21 = —C 12 — C 22 = C ) (24.73)
24.7. PROBLEMS FOR CHAPTER 24 293

Figure 24.4: Parallel plate capacitor.

where C is called the capacitance of the system. The charges on the two con­
ductors are
Qi = -Q 2 = C (4 1 - fa ) = C V (24.74)
where V is the potential difference between the two conductors, while the energy
is
= \ Y2 w a + i = \ c y 2 -
e (24-75)
n
As an application of these ideas, consider a capacitor constructed from two
parallel conducting plates of area A. The separation of the plates, a, is assumed
to be small compared to the transverse extent of the plates, a <C V A , the
approximate Green’s function therefore being that of two infinite plates as shown
in Fig. 24.4 [cf. (17.13)]. The material between the plates is characterized by
a dielectric constant, e . The above discussion applies to this situation so that
the system has a capacitance (7,

C = C\\
/ dSdS' (47r)2 (24.76)

The first surface integral here was previously evaluated in (17.18) and (17.21):

(24.77)
z= 0

where we have now included the presence of c in (24.50). The remaining surface
integral is trivial,

C = Í i í Íg = t a (2478)
yielding the well-known result for a parallel plate capacitor, derived in Section
11.4.

24.7 Problems for Chapter 24

1. Use the stress tensor to evaluate, in terms of the surface charge density, the
force per unit area on the surface of a conductor placed in an electric field.
294 CHAPTER 24. DIELECTRICS AND CONDUCTORS

Check the result by considering the energy of a parallel plate capacitor

with prescribed charges. Reverse the last argument to find the capacitance
of a spherical capacitor.

2. Prove the symmetry of Green’s function for a system of conductors em­

bedded in a dielectric medium.

3. How is (24.58) modified if the free charge density p is not zero?

4. Use (24.61) and (24.73) to calculate the capacitance of a cylindrical capac­

itor of inner radius a, outer radius 6 , and overall length L, L a, 6 , filled
with material of dielectric constant e. Calculate G (r ,r ') following the
differential equation (discontinuity) approach discussed in Section 19.4,
beginning with (19.74).

5. The inner surface of a sphere of radius a is kept at a potential ^ a(0,<^).

Use the result (24.56) to show that the interior potential, when no internal
charges are present, is

<fa{6'A Q
*,(r'#''6)=¿ ( ' - © O Í " ' l “ " '[ i2 4 r 2 - 2 ar cos 7 ]3/2 ’

where 7 is the angle between the direction specified by 0 , <^, and that
specified by 0', <^',

cos 7 = cos 6 cos 01 4- sin 9 sin O' cos(<j> —

6. Prove that the electrostatic potential at any point in charge-free space is

equal to the average of the potential on any spherical surface surrounding
that point. How is this consistent with the result of the previous problem?
Chapter 25

Modes and Variations

The interior of a conducting sphere, and the interiors of conducting cylin­

ders, truncated at both ends by conducting sheets, are examples of conduct­
ing cavities— finite regions within which electric fields are contained. Green’s
function for the volume V bounded by the surface S is characterized by

—V •[e(r)V G (r, r')] = 47r£(r —r'),

r on S : G (r ,r ') = 0 , (25.1)

as is appropriate to the presence of a dielectric medium in V. We have already

provided examples (for e = 1 ) of the construction of G [for example, see (19.15)],

G ( r y ) = 4 ^ ^ (r )f ( r /), (25.2)
a
where the 4^ (r ) obey

- V •[e(r)V ^ a(r)] = 7 %<l>a(r), (25.3)

and constitute a complete:

] C ^ « ( r ) C ( r ') = % “ *')> (25.4)

a
orthonormal:
1 ( * ) f t W M r) = ^ ' (25.5)

set of functions.
The functions <f>a(r) describe the various possible field configurations, or
modes, such that the charge density is proportional to the potential that it
produces,
4npa(r) = 7 l<j>a(r). (25.6)
The coefficients of proportionality,

= 71, (25.7)

295
296 CHAPTER 25. MODES AND VARIATIONS

are called eigenvalues, and constitute a fundamental aspect of the modes. We

shall now describe an approximation method for the estimation of the lowest
eigenvalues and the associated modes.
The starting point is the stationary energy expression for a given charge
density in the interior of a volume V, subject to the boundary condition

r on S : <f>(r) = 0 , (25.8)

namely, [(11.15)]

m = jT (dr) [/*r)*(r) - ^ ( V * ( r ) ) 2 (25.9)

We review the effect of an infinitesimal variation,

6E[<f>] = J (dr)ty(r) p(r) + ¿ V . (e(r)V0(r)) - f dS •«¿(r)^V tf(r),-

(25.10)
noting that 8E[<j)\ = 0 when <p(r) obeys the appropriate differential equation,
and when 6<f>(r) = 0 on 5, corresponding to the imposition of the boundary
condition (25.8). And, for finite variations we have, corresponding to the minus
sign of the quadratic term in (25.9),

E [<t>) < ^ J ( * ) ( * > ( r ) G ( r , r > ( r ' ) . (25.11)

It is also occasionally useful to add a surface integral to (25.9) [corresponding

to the surface integral in (24.16)],

j ) dS<j)(r ) ^ ^ n * V<£(r), (25.12)

where n is the outwardly directed normal (dS = n dS). The additional terms
thereby produced in SE[</)] are

£ds V ¿ + ¿ ¿ ¿ (n .V ¿ )], (25.13)

the first of which cancels the surface integral term of (25.10). Accordingly, this
E[<j>] is stationary for infinitesimal variations about a solution of the differential
equation for </> that obeys the boundary condition, without resort to surface
restrictions on the variations. One does, however, lose the general inequality of
(25.11), for finite variations.
For our immediate purposes, however, we want to retain that inequality, into
which we introduce two modifications. First, let us supply <j>with a scale factor,
AC
^(r) —+ K^(r), (25.14)
which is to be chosen to maximize E[<¡>\. Now,

E[K<t>] = k J{dv)p<j> - k 2 J ( d r ) ^ ( v ^ ) 2 (25.15)

 297

vanishes at /c = 0 , and initially rises linearly with increasing k until the negative
quadratic term dominates, thus producing a maximum at the value of k given
by
J(dr)p<j) - 2nJ{dr) ^ ( V <£)2 = 0 . (25.16)

Thus, (25.11) yields the scale independent inequality (divided by 27t)

}{d v )]{v \ f ^ J r , r > ( r ') . (25.17)

Then we subtract a common term from both sides

~ X J{dt)pl£/<*)'(*V,r)f^1G i r . r O - ^ r - r ' ) P(r'),

25.18)
where A remains to be chosen. Its role becomes more apparent on recognizing
;nizing
that
¿ G ( r , r') - j«5(r - r') = ^ <¿«(r)<¿«(r') 1 1 (25.19)
L7S
Of all the Xa = 7 ^ there is a smallest one, Ai. Let A = Ai. Then the
summation on the right-hand side of (25.19) begins with the mode (or modes)
with the next smallest value of Aa , A2 , where
, A 1 1
A« > A2 : t----- T - < 0. (25.20)
Aa a 1
In consequence, the right-hand side of (25.18) is

E |/(*w.| ( ¿ - ^ <». (25.21)

Xa>\x
the equality sign holding only if p is orthogonal to all the <j)a with Xa > X\. The
inequality (25.18) now states that

f(d r )p 2
^1 < J (dr)e(V<f>)2 (25.22)
[J(dr)p<t>r

which supplies an upper limit to A i, one that involves two arbitrary functions [</>,
of course, must obey the boundary condition]. For a given <f>, we can choose p to
minimize the right-hand side of (25.22). There is a general inequality, analogous
to that for a pair of vectors,

\^J(dv)p4> < J (dv)p2 J(dv)<t>2, (25.23)

the equality sign applying only when p is a constant multiple of <j)—just the
situation described in (25.6). This gives
f (dr)e('V<j>)2
(25.24)
lS f(d r W
298 CHAPTER 25. MODES AND VARIATIONS

where <j) is to be chosen so as to minimize the right-hand side. That will be

accomplished when </>, which is a constant multiple of /?, is orthogonal to all the
(f)(x, Xa > Ai, and therefore is a mode function with eigenvalue Ai.
Note that if p were chosen to be a constant multiple of —V •(eV<^), rather
than of (j)) we should generally get a larger value in (25.22), which is the in­
equality
/(dr)[V .(e V ¿ )]2 > f(dv)e(V<l>)\
(25.25)
f ( d r)e(V<¿)2 ' /( dv)<p
the equality sign appears only when <j>obeys the differential equation of (25.3),
and is therefore a mode function. Perhaps we should also inject here the more
general observation that if

f(dr)e(V<j>)2
m = (25.26)
f(dr)<f>2
is required to be stationary, [6A = 0], so that

6 | y (dr) e(V<f>)2 - A J (dr) <¿2J = 2J (dr) 6<f>[-V •(eV4>) - \4>] = 0, (25.27)

one infers that (j) is a mode function having the stationary value of A as its
eigenvalue. It is only the lowest eigenvalue, Ai, for which the stationary value
is a minimum.
As a prelude to another application of the inequality (25.18), let us, for a
given choose p according to

47rp = - V •(eV<^) = D(j). (25.28)

Then we have satisfied the differential equation connecting (j) with p, and the
equality sign in (25.18) applies. Indeed, one can verify directly that both sides
of this relation now coincide with
1
J (dr)e(V<j>)2 (25.29)
(4 tt) 2

In the next step we add some infinitesimal Sp to the p of (25.28). That would
require an infinitesimal change of (j) to maintain (25.28). However, the stationary
property tells us that ignoring the need for such a 8<j) only introduces an error
of second order. In short, (25.18), with the equality sign, continues to hold on
introducing the infinitesimal change of p) or

¿ J ( dv) 6p<t>- \ J (dr) 6pp

= J ( * ) ( * ') 6p(r) J ^ G ( r, r') - i ¿ ( r - r') P(r0, (25.30)

which also incorporates the relation

l- j(dr)p<¡> = J(dr)^(V4>)2. (25.31)

 299

[The use of the latter is unnecessary if one returns to the E[<j>] of (25.9), as
already noted in Chapter 12.]
Now we choose 6p to be an infinitesimal multiple of

A'k{D(¡> — (25.32)

Omitting that infinitesimal factor, we then get

¿ J ( d r ) 6p* = f ( d r ) < W ) 2 - X1 J ( d r ) <

j>2 (25.33)

and

J ( d r ) 6pp = J (dr)(D<f>)2 - Ax J ( d r ) e(V<j>)2

= A 2 / ( * ) <¿2 + / ( ^ y - [ / (^ ) W ) 2- j (d r) <t>2 ,(25.34)

where
2 f (d r)(D </))2 /( d r ) e ( V ^ ) 212
>0, (25.35)
S(dr)4>2 /(dr)¿2 .
according to the inequality (25.25). So within a factor of f (dr) <j>2, the left-hand
side of (25.30), multiplied by A, is thus given by

(\[<j>] - A i)(A - A[<¿]) - A 2, (25 .3 6 )

which employs the notation (25.26).

We encounter, on the right-hand side of (25.30), integrals of the type (the
infinitesimal factor in 6p is again omitted)

J(dr)6p<j>a = J(dr)(D<j> - Ax<f>)4>a

= J(dr)4(D-M)*a

= (A« - Ai) J (dr) d><f>a (25.37)

and

4ttJ ( d r ) f ap = J ( d r ) f aD<f>

= j{dr)(D<j>a)*cl>

= \a J ( d r ) r a<i>, (25.38)

in which the partial integrations employed to shift £) = —V *eV from one

function in an integral to the other are validated by the boundary condition
300 CHAPTER 25. MODES AND VARIATIONS

that (j) and the <t>a obey. The result for the right-hand side of (25.30), also
multiplied by A, is

I r 2
Y (Aa - Ai)(A - Aa), (25.39)
A<*>A2 ^

where the presence of the factor \a — Ai removes any contribution associated

with the lowest eigenvalue. Then, if we choose A to be a lower limit to A2 ,
A2 ^ ^ 2 j and note that
Aa > A 2, A2 — A<* < 0, (25.40)
we are assured that (25.39) is a negative quantity (< 0). That statement, in
conjunction with (25.36) for A = A2 produces the inequality

(A [$ - Ai )(A 2 - A[<£]) < A 2. (25.41)

The useful aspect of this inequality emerges when A[0], an upper limit to
Ai, and A2, a lower limit to A2, are both sufficiently close to their respective
referents that
A2 - \[</>] > 0. (25.42)
Then the inequality supplies a positive upper bound for A[<^] — Ai, which is a
lower bound to Ai. Indeed, Ai is now bracketed as

A[<¿]> A, > A M - ( 25. 43)

We have used the stationary property of the energy as a guide to these

results. But, once seen, they can be given somewhat shorter derivations, as
suggested by the structures of (25.21) and (25.39). Instead of Green’s function,
we employ the completeness of the mode functions <f>ot(r), writing the arbitrary
(j)(r) as the linear combination

^(r) = ^ C a^a(r), (25.44)

a

which we also subject to term by term differentiation,

(D - A)^(r) = Y C - * )* «(')• (25-45)

a

The complex conjugates of the expansions of these real functions are required
as well. First, let A = Ai and then use the orthonormality of the <j>a to evaluate

/ ( d r ) ^ ( £ ) - A 1 )« i= Y |C'c|2 (A« - Ai) > 0; (25.46)

J A«>A1

this is the inequality

A[<¿] > Ax. (25.47)
25.1. A COMPARISON METHOD 301

Then consider the product of the expansions, (25.45), referring to A = Ai and

A — A2, and get

[ (d r)(D - X íM D - A 2)<¿ = ] £ |C'«|2( A « - A 1) ( A a - A 2) > 0, (25 .4 8 )

J Aa > A2

which is

J (dr) (D<t>)2 - (Ax + A2) J (dr) e(V<¡>)2 + A, A2 J (dr) > 0, (25.49)

or

A 2 + (A[<^])2 — (Ai -f A2 )A[<£) + Ai A2 — (A[<£) —Ai)(A[<^] —A2) -f A 2 > 0; (25.50)

this is the inequality (25.41).

The usefulness of the lower bound to Ai depends on the availability of a
reasonably good lower bound to A2, one that lies significantly above Ai. We
have already alluded to a possibility in Chapter 23, the knowledge, preferably
for some subset of modes, of summations of the type

1 1 1 1
r = 1 , 2 ,. > — + > =F + (25.51)
Ai Ai Ar2

The last statement above recognizes that the inequality is maintained if Ai is

replaced by a larger value, an upper limit, Ai. And, if the final relation were
solved for A2 as an equation, what would emerge is necessarily a smaller number
than the true A2; it is a lower limit, A2. We shall illustrate this procedure shortly.

25.1 A Comparison Method

We have remarked on the stationary property of A[<£], (25.26). In this discussion
the boundary condition was respected by the infinitesimal variation 6<j>. But we
can remove that restriction by proceeding as in (25.9), (25.12), which replaces
A[<^] with

(25.52)
w = ----------------- ü m w -----------------

Now we want to examine what happens to A«, the eigenvalue associated with
the (unnormalized) mode function <^a , when the boundary surface S is altered
infinitesimally by a displacement 6n(r) along the normal to the original surface.
According to the stationary property in which the boundary condition need not
be maintained, it suffices to extrapolate the original mode function into the new
region, or terminate it within the initial region. Thus we have

6J (dr) t(V<j>af zz j, dSSne

d4a \ 2
dn J ’
(25.53)
302 CHAPTER 25. MODES AND VARIATIONS

where
d . _
(25.54)

the contribution of gradients tangential to the surface being of second order,

and
5 j> dSe<j>an . V 4>a = j> dS 8n e ( ^ ~ ^ j , (25.55)

whereas
6 Jv (dr ) = °> (25.56)

inasmuch as the change here is of second order. That gives

cA <fs dS 6n e{d<j)aldn)2
(25.57)
l v (dr)<f>l

as the net variation: Any outward displacement generally lowers the value of
Aa ; any inward displacement generally increases it.
One application of (25.57) refers to situations where Xa has a known de­
pendence on geometry. Then the normalization of the mode function (¡>a— the
requirement that the denominator in (25.57) be unity— can be achieved in terms
of the surface behavior of the mode function. This is illustrated by a sphere, of
radius a, where [(23.28),which refers to e = 1]

<¡>lmn{V) = Ylm{0A)Rln{r) (25.58)

and
Rin(r) = Cinji(yinr/a), ji(yin) = 0; (25.59)
here Cin is the normalization constant left unspecified by the radial differential
equation. In this example we know that A/n = 7 ?n/a2 varies inversely with a2.
Hence, on choosing Sn = 6a, (25.57) reads

- 2 — 7(n = - a 2 6a[Ci„7i„j,'(7;n)]2, (25.60)

a

/ O X 1/2 1
(25.61)
C,n Va3/ j[ (yin) '
in agreement with (23.23).
Before pressing on let’s use (25.57) to fill a small gap in Section 19.3, where
we promised a simpler verification of the normalization constant in <¡>o{n)) the
mode function of lowest eigenvalue for an equilateral triangle. That function is
(e = 1)

OjJT . 2ir .
4>o{fj) sin — {y + r)
h
= C ^ s in — (//<, + r ) = C

a—1
. 2* f 3 1/ 2 1 \ . 2 tt f 3 1/ 2 1 V

+ smT V 2~X~ 2 y + r ) + SmT \ ~ T X ~ 2 y + r)

, (25.62)
25.1. A COMPARISON METHOD 303

according to (19.61) and (19.30). The dimensions of the triangle are specified
alternatively by the height /¿, the radius of the inscribed circle r = |/i, or the
base a = (2 /3 1f 2)h. Thus, the lowest eigenvalue is given equivalently as [(19.60)]

A (25.63)

In the two-dimensional version of (25.57) the surface integral is replaced by

a line integral

dx 8n (25.64)

where the latter form exploits the threefold symmetry of the triangular sides. It
is clear from the geometry that the uniform displacement Sn needed to maintain
the triangular shape is
8n — 8r — ~ 8h. (25.65)
o
Now, the derivative appearing in (25.64) is

d<¡>\ 1 27T ( 3 1/2 h 1 2* f 3l/2 h\

—cos — -----— x + — +
dy y = —r 2 h \ 2 2 2 h v 2
2w
4- cos — x (25.66)
a

which immediately yields

= A1 |«C'2. (25.67)
y= -r ¿

Then (25.64) gives

- 2 ^ A j = - S h X i l a C 2, (25.68)
h 2
and
C - 2 (25.69)
( 3 / i a ) 1/ 2 ’

as stated in (19.61).
We turn from these quantitative uses of (25.57) to its qualitative side. An
outward displacement of the boundary of a cavity decreases the eigenvalue of any
mode. If the given region can be transformed in this way into another one, with
known properties, and a desired mode identified among those of the modified
region, the latter eigenvalue provides a lower bound to the eigenvalue of interest.
As a first example, consider a sphere of radius [to avoid confusion] 77, and the
enclosing cube of side 277. The mode functions of the sphere that have / = 0 are
spherically symmetrical [Ybo = ( 47T)""1/ 2], and we look for their counterparts in
304 CHAPTER 25. MODES AND VARIATIONS

the modes of the cube that are symmetrical in the three orthogonal directions
defined by the cube. Accordingly, we should have [(19.16)]

Aon > 3 n = 1,2,..., (25.70)

or
31/ 2 n7T
TOn > 0 . 866 ^ , (25.71)
~ ~ R
which is true, for [(23.25)] Ton = mr/R.
With cylindrical conductors, attention focuses on the two-dimensional cross
section. Let’s compare a circle of radius R with an enclosing equilateral triangle,
for which R is the radius of the inscribed circle, and with an enclosing square, of
side equal to 2R. The modes of the circle with m = 0 are axially symmetrical;
we look for the first few counterparts among the real mode functions of the
equilateral triangle, specifically those with l\ = h, say, where [(19.59)]

(25.72)
7? = 5 © 2 ( ' ? + ' ? + 4 ' ?>= i( S ' 0 !

Thus, we expect that

27T 2.094 47r 4.189
(25.73)
701 > 3R ~ R ’ 702 > 3R R ’
and indeed [(19.127), (19.129)] 70 1 R = 2.405, 702 R = 5.520. The analogous
consideration for the somewhat more closely fitting square gives

9 /717T \ 2
(25.74)
7on > 2 ( 2 r ) ’
and
7o i# > 2 " 1 / 27t = 2.221, jo iR > 21/2?r = 4.443. (25.75)
A lower limit to 711 is provided by the first example of an unsymmetrical mode
of the square, [/ = 1 , m = 2, with a = b in (19.9)]
2 / 7T\ 2 5 / 7T\ 2
(25.76)
* ■ > ( £ ) + © - ? (5 )'
and
51/2
JuR > j ^ — 3.512, (25.77)

to be compared with [(19.131)] j n R = 3.832.

25.2 Iteration
A (real) mode function, say one with the smallest eigenvalue, Ai, obeys the
boundary condition, and the differential equation

- V 2^i(r) = A^i(r), (25.78)

25.2. ITERATION 305

where, for simplicity, we restrict this discussion to e = 1 . Suppose one has picked
some initial approximation, <£i°^(r), which satisfies the boundary condition but
not the differential equation. Is there a procedure for systematically improving
that initial choice? We begin our affirmative response by defining <^^(r) as the
solution of the differential equation

- V ^ / V ) = <^,0)(r) (25.79)

that satisfies the boundary condition. It is the first step in a process of iteration
wherein an approximation is introduced on the right-hand side of (25.78) and the
equation solved (with a change of scale) to produce an improved approximation.
The general statement of the iteration process is

- V 20 (i "+ 1 )(r) = <¿(i "}(r). (25.80)

We use the nth iterate in the stationary expression (25.26) to define

aW = A [ ^ ] = ^ | I _ 1 1 > Ai. (25.81)

This has involved the rearrangement

y (d r )(V < ¿ i n)) 2 = J (d r ) (—) V 24 nV i n) = J ( d r ) / 1n _ 1 V in), (25.82)

as justified by the boundary condition, and introduces the useful notation

[m -f n] = J (d r )^ m^ n\ (25.83)

The stated dependence on only the sum of the indices follows from the relations

J (dr)//"Vin) =/ (dr)4m>(-)V2*in+1>=j (*)(-)V2*<mV<n+1)

= y >(d r )^ m-1)^ n+1). (25.84)

Now we demonstrate that the iteration process does converge to the correct
eigenvalue and mode function, provided that the initial choice <j>^ is not or­
thogonal to the desired mode function. We begin by applying the inequality
- ^(n+ 1)
(25.25) with ,<j>
[2 n] [2 n -f 1 ]
(25.85)
[ 2 n + l ] “ [2n + 2]’
Then we use the inequality, analogous to (25.23), that refers to vector fields,

vy>] < J w v t f f w w y (25.86)

306 CHAPTER 25. MODES AND VARIATIONS

The choice <j>= <p\n \ ip = <^n+1-* yields

[2n]2 < [ 2 n - l][2 n + 1] (25.87)

or
[2n — 1] ^ __[2n]
(25.88)
[2 n] [2 n + 1 ] ‘

The combination of the two inequalities, (25.85) and (25.88), is expressed by

A( » } > A(n + l/ 2) > A(n + 1) > A ^ ( 2 5 .8 9 )

where
A(n+1 / 2) _ i l nJ (25.90)
1 [2 n + l] v '
is indeed obtained from (25.81) by the formal substitution n —» n + Thus,
the sequence of approximations to Ai is monotonically decreasing, but cannot
be less than Ai— it approaches a limit, fi > Ai. Preparatory to a discussion
of that limit, let us remark that one solution of the iteration equation given in
(25.80) can be displayed with the aid of Green’s function:

# +1)(r) = J (dr')¿G(r,r')<¿(r V ) . (25.91)

On applying the inequality (25.23) we then learn that

(i"+1)(r))2 < f(dr>) Air

G{ r ,r ') [2 r (25.92)

The only question about the integral involving the square of Green’s function is
its existence for a small region that includes the point r' = r. In such a region,
however, Green’s function is dominated by Coulomb’s potential, and the integral
of |r' — r |“ 2 does exist. Accordingly, (25.92) presents us with a bound of the
form
|<¿i"+ 1 )(r)| < C (r)[2n]1/2. (25.93)
Now consider the following series, for positive /?,

Fi(r) = f > " « ^ +1)(r).

n=0
(25.94)

We can say that

|*i(r)| < ¿ ^ | / ” +1)(r)l < (25.95)

n=0 n=0

and so the function defined by the infinite summation will exist if the ratio of
successive terms in this upper bound approaches a limit less than unity,

J
1
1¡m P lim [A (n + l/ 2 )A( n + l) ] 1/2 < 1 , (25.96)
0
n- .00 y [2n]
25.2. ITERATION 307

which is to say, provided

(25.97)
The function F i(r) obeys the differential equation

—V 2Fi(r) = 2 2 / r 4 n)(r) = m ( t ) + 4 % ) . (25.98)

n=0
Let us multiply it by the mode function <^i(r), and integrate, to get

J ( d v ) M r ) 4 ° \ r ) = f (dr) <fii(—V 2 — /3)Fi

> j { d v ) F l { - V 2 -P)4>i

= (Ax-/?) J (dr) (25.99)

Here is where we prove that /i = Ai, provided the initial choice is not
orthogonal to the actual mode function <t>\— the left-hand side of (25.99) does
not vanish. Assume the converse, that /j, > X\. Then, according to (25.97), the
function F i(r) exists for /? = Ai. But with that value of /? the right-hand side of
(25.99) does vanish. The contradiction shows that /i = Ai. And we know that
A[</>] = Ai can only be achieved with = (j>\. Perhaps the following afterthought
will be helpful. One can verify, with the aid of the completeness of the mode
functions, that the solution of the differential equation (25.98) for F\ is

F i(r) = 2 2 / (dr')Mr')<f>\i V ) . (25.100)

Here, clearly displayed, is the first singularity that appears with increasing /?,
at ¡3 = Ai, unless this term is missing because is orthogonal to <j>\. In the
latter circumstance the first singularity— and the number to which the sequence
of approximants converges— appears at A2 unless___
How rapidly does the approximation sequence approach Ai? To answer that
let’s look at the inequality (25.49), with (j) = <^n+1^:

[2n] — (Ai -f- ^ 2 )!/^ -f" 1] ^ i A2 [^^ “h 2] ]> 0, (25.101)

or
x(n + l / 2)^(n + l)
A ) " T 1 /^ A ) nT1' - (Ax + A2) A (1n + 1 ) + A xA2 > 0, (25.102)
which we present as
A(xn + 1 / 2 ) - Ax ^ A2
(25.103)
A(1n + 1 ) - Ax ~ A(! n + 1 ) '

Another such inequality is produced by replacing one (^-function in (25.48) by

D(f). The effect in (25.101) is to change 2n into 2n — 1, and that yields

A(xn ) - A x
(25.104)
A(n+X/2) _ Ai - a (»+1/2)-
308 CHAPTER 25. MODES AND VARIATIONS

The product of these two inequalities then states that

A(jn) - At A2
^(n +1) _ ^ — ^ ( n + l / 2 )^ (n + l) * (25.105)

The last result shows that the error of the (n + l)th iteration is smaller than
that of the nth iteration by a number that (for A2 = A2) approaches (A1 /A 2 )2 =
( 7 1 / 7 2 )4 as the iteration proceeds. Thus, the larger the ratio A2 /A 1 , the more
rapid is the convergence. We must point out here that the second eigenvalue re­
ferred to in this convergence criterion can well exceed the second smallest eigen­
value of the cavity. This occurs in the presence of spatial symmetry properties
that permit the decomposition of the modes into different symmetry classes. If
the initial function <j>^ possesses the particular symmetry that is characteristic
of <¡>\ so also will the successive approximations <j>^\ every member of this se­
quence is automatically orthogonal to modes of other symmetry classes, and the
relevant second eigenvalue is that of a mode having the same symmetry as <j>\.
Furthermore, in consequence of this orthonormality, the method under develop­
ment applies to the lowest eigenvalue of each symmetry class, independently.
These remarks are illustrated by the circular cylinder. A real mode function
has the angular dependence cosm<¡> or sinm<^, and modes associated with dif­
ferent m values are orthogonal; each value of m is independent. Accordingly,
in discussing the lowest eigenvalue of the m = 0 class, A01 , the relevant sec­
ond eigenvalue is not the actual next larger one, An, such that (An/ Aoi )2 =
(3.832/2.405)4 = 6.445, but rather it is A02 , with (A02/A 01)2 = (5.520/2.405)4 =
27.75.
As we have already noted, the inequality (25.41) provides a lower limit to
Ai if a reasonably accurate A2 is available. The evaluation of A 2 [(25.35)] for
<j) = ( f is

[2n - 2] / [2n — 1]\ 2

A2
[2n] V [2n] )
( a (” - 1/2) - A(xn)) A$n), (25.106)

and we learn from (25.43) how Ai is bounded at the nth iteration stage:
—1/2) \(n)
A(!n) > Ai > A<"> - ^ (25.107)
a 2/ a ^ - i k
the lower limit also follows from (25.102), with n —> n —1. As always, the corre­
sponding statement at the (n -f l/2 )th stage is produced by formal substitution.
We now want to recognize how the iteration process can be used to optimize
the bounds within which Ai is confined. Suppose that iteration has advanced
so far that (25.105) can be presented as

A ^ -A x
(25.108)
A(” +1) - Ai
25.2. ITERATION 309

which employs the optimum choice, A2 = A2, for that is what the process selects
(we return to this point shortly). This asymptotic ratio of iteration errors shows
that the approach to the limit is as

(25.109)

Now we use

(25.110)

to learn the asymptotic behavior of the lower limit given in (25.107),

A A 2n A2 - A i
A<” > * * , + « , ( £ ) - « ( £
A2 —Ai
2n
A A2 — A2 ( Ai
= A\ — -------- ; — CL (25.111)
A2 — Ai \A2

Thus, if one can merely pick A2 such that

A2 > ^2)> (25.112)

the error of the lower limit at any (sufficiently advanced) stage is less than that
of the upper limit. And one can improve matters, for the asymptotic ratio

A(n - l / 2 ) _ A(n) ^ A2

A(n) _ a(» + 1/2) “ Ai (25.113)

provides an internally generated estimate of A2 that can be used for A2. The
closer the latter is to A2, the more rapidly will the successive lower limits con­
verge, according to (25.111). [As we shall see in a moment, with A2 = A2, the
error of A ^ is dominated by a multiple of (Ai/As)2n.] And the latter equation
also shows that too large a choice for A2 will betray itself—then the “lower
limits” will converge from above!
Before illustrating all this in a reasonably favorable situation, we shall supply
additional assurances concerning the convergence of the iteration process. We
begin by representing 4 in terms of the complete set of <j>ai

(25.114)

Then the first iteration, the solution of (25.79), is

Ca
\ r) = T ^ « ( r ). (25.115)
310 CHAPTER 25. MODES AND VARIATIONS

and, in general,
Ca
A(n)W = £ -M * )- (25.116)
(Aq.)’

Notice that the evaluation of the integral defining [m + n], (25.83), confirms its
dependence on the sum of the indices,

Cl
\m + n\ = 'y 77—. a , (25.117)
v ( A« ) m+n
Now we have
\ («) _ [ 2 n — 1] _ ^ a { C a f A” ) 2
(25.118)
1 " [2 n ] " E „ ( C « / A » ) 2

or
\(n) _ \ I Y 2a (^a ~ A l ) ( C a / A £ ) 2
(25.119)
1 " 1+ Y , « { c a/ K Y •
As the iteration proceeds, the summations in the latter version will be dominated
increasingly by the leading terms:

,(») .. V , (*a - A i ) ( C 2/ A 5 ) 2 / a , ) 2"

1 , + — w w — - A, + " U ) ' <25 120)
where
a = (A2- A x) ( g ) - (25.121)

which is indeed of the form (25.109), with A2 entering automatically as the

optimum choice of A2.
It is also worth noting what would happen to A ^ if one managed to choose
A.2 = A2-

,(») ,(») A < r 1/2)- A ^ _ A 2 - A r i / 2 ) . (w)

(25.122)
Al " Al (a2/ aW ) - i " a2 - a <"> Al •

Now we need to carry (25.120) a step farther,

(A2 - AQiCVAS ) 2 + (A3 - Ai)(C 3 /Ag ) 2

A<"> Aj +
(C i/A y)2 + (C 2/A5)2
2n
A iV " a2 (X i V n , ,
Ai + a
A2 ; A2 - A i VA2 ' (£) (25.123)

in which
&= (A3 — M){C^/C\)2. (25.124)
We find that
. A3 — A2
(Sf-
A(in) Al — ------ ;— 0 (25.125)
A2 — Ai
and thus the correct localization of A2 is signaled by a marked increase (to the
extent that A3 > A2) in the convergence rate o f the lower limit.
25.3. EXAMPLE 311

25.3 Example
Consider the parallel plate geometry, 0 < 2 < a, for which the lowest mode
function is sin with corresponding eigenvalue 7 2 = 7r2/a 2. First consider the
variational bound (25.24). We must choose a trial function which vanishes at
z = 0, a. The simplest example is

<t>\0)(^) = z (a - z), (25.126)

SO

AÍ°> =
XTdz [&(*(<» - z))]2
Jo* d zz2(a — z )2
_ 1 fp dt(l — 4< + At2)
~ a2 fg dt(t 2 - 213 + t4)

_ l i _ l £ . (25.127)
Q?“ -h.
U30 “
5

indeed, \/T0 = 3.162 is only 0.7% bigger than it. To get the next iterated bound,
we have to solve (25.79), or

d2 ,(i) , .
(25.128)

the solution which vanishes at z = 0, a is

(25.129)
^ = T 2 z 4 - r 3 + T2z -
So, the second iterate is
A(1) - M (25.130)
1 - w
where

1» = f i , *<«V « = * «1 - < ) ( £ - £ + ¿ ) = ^ L a \ (25.131)

and

[2] = £ d , « < ," ) ’ = »• ¡ I * ( n ■- J + 4 ) 1‘ = 5 ^ 5 » * . (“ • ! » )

which gives
(i) _ 306 1 (i)
3.14181
(25.133)
aí
31 a2 ’ 7l
only 0.007% too high. As for the lower bound, let us simply take A2 = ^2 =
(27r/a)2, and then set n = 1 on the right-hand side of (25.107):
AC1/ 2) A(l)
\(i) _ \W A1 “~ Ai (25.134)
Ai - Ai - 7 rT T 7 T 7 (I) 7’
(27r)2/ a 2A\
312 CHAPTER 25. MODES AND VARIATIONS

\(1 / 2 ) _ [0] - 1 6 8 1 (25.135)

1 ~ [1] ~ 17 a 2 ’

S° ,(D _ 9.8672 (d 3.14121

(25.136)
Al 71 ~ ~ ~
only 0 .0 1 % below the correct value.

25.4 Problems for Chapter 25

1. Derive the approximate expression for the lower bound for the lowest
eigenvalue, (25.125).

2. Consider a circle of radius R , for which the lowest eigenvalue corresponds

to yo\R = 2.40483. Using the trial function <f>^ — R — p, where p is
the radial coordinate, compute the corresponding upper bound 7 i°\ Do
better by solving the differential equation (25.79) for <j¿[ \ and thereby
find the improved upper bound 7 ^ . If you assume the lower bound to
the second eigenvalue to be given by j _2 = 5, compute the lower bound
to the first eigenvalue, or 7 ^ . How is that estimate changed if the exact
second eigenvalue 702 = 5.520 is used instead?

3. Carry out similar estimates for the lowest eigenvalue of a spherical cavity.
Chapter 26

Magnetostatics

26.1 Variational Principle

We now return to the general action principle of electrodynamics, (9.13), be­
fore the specialization to electrostatics. Recall at the beginning of Chapter 11
we isolated the terms referring to the electromagnetic field, and omitted time
derivative terms. In this way we obtained a separation of static electric and
magnetic energies, given by (11.2) and (11.3). The former was the basis for our
ensuing discussion of electrostatics. The second describes magnetostatics, which
is the subject of our investigation here. Notice that this separation is possible
only because of the condition
¿ A = °; (26.1)

otherwise, electric and magnetic effects are interrelated. Analogously to our

incorporation of dielectrics in electrostatics (see Chapter 11), we here pass to
a macroscopic description of fields in permeable media. The energy, (11.3), for
these circumstances, becomes

1, ' 1
£[A,B] = - J(dr) —J •A + ~—
c 4tt/jl )]■
-B. V x A + - B 2 (26.2)

where is the permeability of the medium. We now have to check that the sta­
tionary principle applied to this form of the energy yields the correct equations
of magnetostatics. We are to regard A and B as the independent variables, so
the variation of the energy is

SE= - Ij.tfA- T ® .Vx<$A + -i-5B.(B- V x A ) (26.3)

c 47r \i 4irfi

From the coefficient of 6B, we obtain

B = VxA, (26.4)

313
314 CHAPTER 26. MAGNETOSTATICS

which is equivalent to
V •B = 0. (26.5)
By making use o f the identity

® - V x 5 A = V . (<5AX®) + 5 A ‘ (V X ^ ) ’ (26.6)

and discarding the implied surface integral, we find from the vanishing of the
coefficient of <5A ,
4w
V x H — — J, (26.7)
c
a consequence of which is that only steady currents occur here:
V. J = 0. (26.8)
As appropriate to macroscopic media, we have introduced the magnetic field,

(26.9)

Thus we have recovered Maxwell’s equations in the static limit, (26.5) and
(26.7).
As in electrostatics, there is a restricted version of the stationary principle
for the energy. If we take
B = Vx A (26.10)
as the definition of B, the expression for the energy becomes

E [ A] = - J (dr) -J •A — — ( V x A ) 2 (26.11)
c 8tiV '
Regarding this as stationary under variations in A, we derive the equation
satisfied by the vector potential,

Vx Q v x a ) = (26.12)

which coincides with (26.7).

Proceeding in a manner parallel to the corresponding discussion in elec­
trostatics (see Section 11.2), we consider a change in the permeability, 8/jl{r).
Because of the stationary property, the only first order variation in the energy
arises from the explicit appearance of /i in (26.11):

6E = - / ( d r ) ^ i - ( V x A ) 2 = - / (dr) ^ H 2, (26.13)

which is the analog of (11.39). In particular, by considering <S/i to arise from a

displacement of the material, we infer the force on the (inhomogeneous) perme­
able medium to be [cf. (11.44)]

F = - / ( d r ) ^ V M; (26.14)

a diamagnetic material, with fi < 1, is repelled from a region of stronger mag­

netic field.
26.2. BOUNDARY CONDITIONS 315

Figure 26.1: Discontinuity in permeability across a plane interface.

26.2 Boundary Conditions

The simplest example of an inhomogeneous magnetic material occurs when //
is discontinuous across an interface. First we consider the boundary conditions
that B and A must satisfy across the interface. The fact that B is the curl of
A implies that the tangential component of A , At must be continuous across
the boundary, in order that B* be finite:

niXAi + 112X A 2 = 0, (26.15)

or,
m x ( A i - A 2) = 0, (26.16)
where ni ( 112) is the outward normal to Vi (V2 ) so that 112 = —n i. The relation,
(26.15), is true for all points on the surface. Thus, when we take the divergence
of this expression, in which only tangential components of V occur, we find

ni •B i -f n 2 •B 2 = 0, (26.17)

or
n i . ( B ! - B 2 ) = 0, (26 .1 8 )

that is, the normal component of B is continuous across the boundary. [We
may regard this as a surface version of V •B = 0.]
If we include the possibility that there is a surface current, K, on the bound­
ary between V\ and V2 , we must amend the energy expression, (26.2), to read

- B - V X A + Í B 2) j - J d S - K - A . (26.19)
E
= - /< * > [?J •A -f-
47Xfi

In our previous discussion of the variation in the energy, we discarded the surface
integral [see (26.6)]; this is no longer permissible because of the presence of the
boundary. Consequently, there is a new contribution to the variation of the
energy arising from the occurrence of the interface,

6E = ¿ A 2X ^ ) - t d S ^ K 'S A
<a '*Í7
316 CHAPTER 26. MAGNETOSTATICS

=
74/ JC| ÓAiXHi
/ a6 ni •----- ---------- b n2 •
47T

ihX
6A2XH2

H j f n2x H 2 , 1 ^
1 T^
;------------- K •0A
47T C
£A

= - / £/S<5A. ----- ------------ 1-------- ------------ 1----- IV (26.20)

7 47r 47T c

Here we have used the fact that A t must be continuous,

¿ A i t = S A 2t = ¿ A t . (26.21)

We then conclude from the stationary principle on the surface that

4-tt
m x H i + n 2x H 2 + — K = 0, (26.22)
c
or
4-tt
ni X (H i - H 2) + — K = 0. (26.23)

When no surface current is present, K = 0, this states that the tangential

component of H is continuous. If, in addition, we have fi2 [idealized as
fj>2 —» 00 ; we might call this a perfect magnetic conductor (see Chapter 49)], the
magnetic field in medium 2 goes to zero,

H 2 = — B 2 -H. 0, (26.24)
P2

so that H i is normal to the surface,

ni x H i = 0. (26.25)

This is the same condition satisfied by the electric field at the surface of a
conductor.

26.3 Vector Potential

The fundamental equation of magnetostatics is (26.12),

V x Q v x a ) = Í p í, (26.26)

which reduces for vacuum (/! = 1) to

A*r
V ( V . A ) - V 2A = — J. (26.27)
c

This equation can be simplified by using the fact that it is invariant under a
gauge transformation [see (9.68)],

A —►A -J- VA. (26.28)

26.3. VECTOR POTENTIAL 317

Because of this gauge freedom, we can usually choose some particular gauge to
simplify the problem at hand. In the present situation, the convenient choice of
gauge is one for which
V •A = 0, (26.29)

called the radiation, Coulomb, or transverse gauge. To show that it is always

possible to choose this gauge, suppose we start with a vector potential, Ao,
which does not satisfy this condition,

V •A 0 0. (26.30)

It is possible to make a gauge transformation,

Ao —>•A = Ao 4- VA, (26.31)

such that (26.29) is satisfied, that is

V . ( A o + VA) = 0, (26.32)

for then A is a solution to Poisson’s equation,

—V 2A = V - A 0. (26.33)

In the radiation gauge, (26.27) becomes

—v 2a = i l l , (26.34)
c

the solution of which is

(26.35)
* ' > - ; / < * ' ) I,1- rV

in precise analogy to the solution o f the electrostatic problem,

—V 2</>= Air p. (26.36)

As a consistency check, we verify explicitly that (26.35) satisfies the radiation

gauge condition (26.29),

V -A (r) = i J < * ') ( V ^ j .J (r')

= - f ( d r ' ) , 1 J 1•J (r') = 0 (26.37)

cj |r — r' I

where we have used (26.8), and the fact that the current distribution is localized.
318 CHAPTER 26. MAGNETOSTATICS

Once we have the vector potential, we can compute the magnetic induction'
B:

B = V X A = i v x [ (dr1) , J ^ /,
c J |r — r|

= iy<<ít')J(r')Xj¡í f I F . (26.38)

For a point charge moving with velocity v, the electric current is

J(r) = ev¿(r — R ), (26.39)

where R is the position of the particle, which produces the magnetic field

v e(r-R )
(26.40)
c |r — R|3

This has the form

B = —x E , (26.41)
c
which was our starting point for introducing magnetic fields in Chapter 1. [Of
course, we have now transcended the domain of magnetostatics since dA/dt ^ 0.
However, since (l/c)(dA/dt) is of order v 2/c2y (26.40) is the correct magnetic
field to first order in v/c. For the general case, see Chapter 31.]
In the following three brief chapters, we will develop some applications of
magnetostatics. Many situations in magnetostatics can be attacked by evident
variations on the techniques developed for electrostatics.

26.4 Problems for Chapter 26

1. In 1935 F. and H. London proposed that a superconductor is characterized
by c

j+ j ¿ í A=v*
Show that this leads to (/i = 1)

V 2B = ^ B ,

which implies the Meissner effect, that a uniform magnetic field cannot
exist inside a superconductor. is called the London penetration depth,
and has the typical value of 10""5 cm.
Chapter 27

Macroscopic Current
Distributions

The simplest example of a macroscopic current is that which flows in a long

straight wire. We will take the wire to lie along the z axis, carrying a current
I that flows in the + z direction. We will let the direction of current flow
be denoted, generally, by n. See Fig. 27.1. We wish to find the magnetic
induction B produced by this current. Since B is independent of 2 , without loss
of generality we evaluate it at 2 = 0. For a wire with negligible cross section,
any volume integral involving the current density becomes a line integral

J (dr) J ( r ' ) . . . = J dz' dS' n J ( r ' ) . . .

The expression for the magnetic induction, (26.38), a distance p from the wire,
is then reduced to

(27.2)

t I

Figure 27.1: Magnetic field produced by a current-carrying wire.

319
320 CHAPTER 27. MACROSCOPIC CURRENT DISTRIBUTIONS

Figure 27.2: Force between parallel, current-carrying wires.

For a long wire of length 2L, L^> p, the integral occurring here is

fL t 1 [L dz' ( L \
(27.3)
J-L \/p
\fpi2 + zf2 Jo \Zz'2 + p2 \ P )
the gradient of which is
, L \ Vp 2
V 2 In — b constant = —2-----— — p. (27.4)
P ) P P
where p is a unit vector in the radial direction. The magnetic field produced by
this wire is therefore
B = — (n x p ), (27.5)
cp

which is concentric with the wire, and in the sense given by the right hand rule,
that is, if the thumb of the right hand points in the direction of current flow n,
the fingers curl in the sense of B.
The force exerted on a current distribution by a magnetic field is [see (7.7)]

F = ^J(dr)3xB, (27.6)

which, when specialized to current flowing in a long straight wire, becomes

F= j d z^n xB. (27.7)

One possibility is that the magnetic field is produced by a second, parallel,

current-carrying wire, as shown in Fig. 27.2. Of course, the total force acting
on I 2 due to the magnetic field produced by I\ is unbounded. The quantity of
interest is the force per unit length on I^)
force h 2ii ( ^
-n jX B -------- n 2x ( n i X p ) . (27.8)
length c c cp
For parallel flowing currents, ni = n 2, so
force __ 2 / i / 2 p
(27.9)
length c2 p ’
 321

that is, the force is attractive. If the currents flow in opposite senses, the force
is repulsive.

where the energy is given by (26.11), with \i = 1. Equation (26.11) is quite

analogous to the electrostatic energy, (11.15), except for the overall sign, which
implies that the sense of attraction or repulsion is reversed when we go from
static charge distributions to steady current flows. By integrating by parts
on the ( V x A ) term and then using the differential equation (26.12), we may
rewrite the magnetostatic energy as

(27.11)

[Notice that (27.11) is gauge invariant, since under a gauge transformation,

A -> A + VA (27.12)

the energy does not change,

(27.13)

since we can integrate by parts and use (26.8).] Introducing the explicit form
for A , (26.35), we can write the energy in terms of the current density alone,

(27.14)

which is analogous to the electrostatic result, (12.10), or (1.1), except that its
sign is opposite. For the case of two current distributions,

J(r) = Ji(r) + J2(r), (27.15)

the energy expression contains self energies as well as the mutual interaction
energy. We are here interested only in the latter, which is

(27.16)

as it is the sole term that contributes to the force, (27.10). For straight wires,
this becomes
(27.17)

For parallel wires with currents flowing in the same sense,

ni .112 = 1, |r - r'| = p1 + (z - z')2 (27.18)

322 CHAPTER 27. MACROSCOPIC CURRENT DISTRIBUTIONS

the integration over zf is evaluated as in (27.3), so that

E = —-^p-2 ^ln — + constant ) / dz

0 /'
— ^■2 ^ln — + constant^ 2L, (27.19)

where we have used the restriction L/p 1. The force can now be calculated
from (27.10), or, since E depends only on /?,

f = - | A £|a (27.20)

The force per unit length is therefore

F_ _ 2h h p
(27.21)
2L ~ ' c2 p’

which is our previous result, (27.9).

27.1 Magnetic Energy. Coefficients of Induc­

tance
We have mentioned repeatedly the difference in sign between magnetostatic
energy and electrostatic energy. This might cause some confusion. In particular,
we might recall the form of the magnetic energy density given in (7.15),

= (27.22)

If the volume integral of this is evaluated in terms of prescribed currents, using

(26.38) and (26.7), we obtain a current-current interaction of the form (27.14),
but with the opposite sign. What is going on?
Equation (27.22) is correct, but not relevant to the situation o f statics, where
the current has no time dependence. The difference is attributable to the energy
required to build up the magnetic field, which is a dynamical consideration. If
we are interested in the force between steady currents, the energy (27.16) is
the relevant quantity to derive the force through (27.10). [The reader is also
referred to the independent derivation of the magnetostatic energy presented,
in the context of radiation, in Section 33.3.]
For a system of current-carrying wires, with currents / ¿ n t h e coefficients of
inductance Lij are defined in terms of the energy by
27.2. PROBLEMS FOR CHAPTER 27 323

analogously to the way the coefficients of capacitance were defined in (24.64).

We can express Lij in terms of a geometrical integral, for wires of negligible
cross section, by introducing local coordinates along the wires, so that

J (r ) = X ) 7¿n ¿(&)«5(r -u)> (27.24)

where rj_¿ is perpendicular to n¿(£¿). The coefficients of inductance are then

given by

=h¡ d£idt ;3 (27.25)

a formula discovered by Franz Neumann in 1845.

For closed filamentary current loops, we can also write the energy (27.11) as

1
(27.26)
2c

in terms of the magnetic flux penetrating the ith loop,

$,■= [ dS •B. (27.27)

JSi

Thus that flux is given by

$i = c J 2 L i jI j, (27.28)
3
analogous to (24.63), which is the basis of elementary circuit analysis.1
Examples of calculation of the inductance for simple systems are given in
the problems.

27.2 Problems for Chapter 27

1. Do the integral in (27.3) exactly, and derive the approximate result given
there.

2. Derive (27.11) from (26.11) by scaling A ,

A —►AA , A = constant,

and then using the stationary property of E[A].

1For example, Faraday’s law, (1.70), gives the emf in the ¿th circuit as S% = j> d ( i • E =
However, such circuit analysis goes beyond statics, dealing as
it does with alternating currents. In this case, as noted above, the energy associated with an
inductor L is positive, E¿, = ^ L I 2, and thus an L C oscillator, for example, can be understood
in terms of the energy interchange between that in the inductor, and that in the capacitor,
E c = | Q 2 /C.
324 CHAPTER 27. MACROSCOPIC CURRENT DISTRIBUTIONS

3. Calculate the mutual inductance M = 2L \2 of a pair of parallel wires of

length / separated by a distance a. Do the integral exactly, and examine
the limit / >> a, where

4. Calculate the self-inductance of a straight wire of length / and radius b if

/ ^> 6. Do so using an argument originally due to Maxwell: Consider the
self-inductance of the wire as arising from the mutual inductance of each
pair of filaments that make up the wire, so that

where (¿**±1 ,2) are the cross-sectional area elements of the two filaments,
and pi 2 is the distance between the two filaments. The integral over In p
may be carried out by noticing that —2 In p is the potential of a unit point
charge at the origin in two dimensions (two-dimensional Green’s function),
so that

is the potential due to a uniform charge distribution. But the latter must
satisfy the two-dimensional Poisson’s equation [cf. (16.16)], so for p < b,

<t>= A + B In p — 1rp2,

By doing the integral at the center of the wire, determine the constants,
and thereby find the self-inductance

5. Calculate the self-inductance of a two-wire transmission line which consists

of parallel wires of great length / and of radius 6, separated by a distance
a, traversed by equal currents in opposite directions. The self-inductance
L is given by the energy
e = - \ l i\

where I is the common current.

6. Show that the self-inductance of a circular loop of wire of radius r and

cross-sectional radius b is (6 <C r)
Chapter 28

Magnetic Multipoles

28.1 Magnetic Dipole Moment

We now direct our attention to the magnetic field produced by a confined current
distribution. If we wish to evaluate the vector potential far outside the current
distribution, |r| ^ extent of the region of current flow (where the origin of the
coordinate system is located in the current distribution), we may use the first
two terms in the expansion (22.2),

1 __ 1 r*r'
(28.1)
|r —rx| r ^ r3

in the expression for A , (26.35). The resulting expansion for the vector potential
is then

A (r) = ^ ( * ,)iJ(r/) + y ( d r')r'(iJ(r))+..., (28.2)

which is analogous to the expansion for <£, (22.3). From current conservation
for steady currents,
V •J (r) = 0, (28.3)

the first term of (28.2) vanishes for a confined current distribution:

0 = J (d r ) r V •J = J ’(dr) [ V . (Jr) - J] = - j ( d r ) J, (28.4)

so that there is no 1 /r term in the expansion of A (r). Physically, there is no

Coulomb-like potential for magnetism, when magnetic charge is not present.
The leading term in the vector potential expansion is therefore

A( r) = -“ / (d r')r .r 'J (r ') + ---- (28.5)

r° c J

325
326 CHAPTER 28. MAGNETIC MULTIPOLES

To evaluate this integral, we again use (28.3) and consider the integral

3^ /»
^ ^ 1 (dr) (#? Ji "h Jj) j (28.6)
Ar= l J

or, in a dyadic notation,

Í (dr) [rJ -F Jr] = 0. (28.7)

Using this fact, we make the following rearrangement:

J (dr' ) r •r'J = - J (d r ') r •[(r'J + J r') -f (r'J — Jr')]

= 1-J ( d r ’ ) r . ( v ' 3 - 3 v >)

= lj(d r ')(r 'x 3 )x r . (28.8)

The leading term o f the vector potential now becomes

K( \ ^Xr (28.9)
AW = — >

where \jl is the magnetic dipole moment, defined by

(28.10)
**= ¿ / ( dr)r><J(r)'
[For a point charge, the current density is given by (26.39), so the magnetic
dipole moment is
(28.11)

which obviously generalizes to a system of point charges, in agreement with

(6.26).]
The leading contribution to the magnetic field can now be calculated from
this vector potential,

B = Vx

= 4 *„¿(r)-v (í^ ), (28.12)

since
- vV i ---
= - Q> (28.13)
r ró
and consequently

0 = MX ( v x ¿ ) = v ( ^ ) - O . . V ) i . (28.14)
28.2. ROTATING CHARGED SPHERICAL SHELL 327

The delta function term in B is necessary in order to satisfy V •B = 0:

V - B = 4tt(/z - V)<$(r) + V 2(/x •V ) - = 0. (28.15)

r
For r > 0, this magnetic field, (28.12), has the same form as that of the electric
field produced by an electric dipole moment, which is contained in (22.11), that
is,
3rn •r — r2/x
B(r) = (28.16)

In general, this is only the leading contribution, since there are higher multipoles.
We will not, however, explore these further here.

28.2 Rotating Charged Spherical Shell

An example for which the dipole expression is exact beyond a certain distance
results if a charge e is distributed uniformly over a spherical shell of radius a
that is rotating with angular velocity a>. If we choose the origin to be at the
center of the sphere, the velocity of a point r' on the surface is

V - (j o X r;. (28.17)

The current density is

J = />v, (28.18)
where here the charge density is entirely concentrated on the surface,

J (dv')p. . . = j ) dS' a . . . , (28.19)

where the surface charge density is constant,

<28-2°)
The expression for the vector potential, (26.35), becomes
e wXr '
A(r ) = \ f d S ? (28.21)
Aira2 |r — r'

where r may be either inside or outside the sphere.

We first calculate A inside the sphere, that is, for |r| < a = |r'|. Recalling
the expansion, (21.10),
-j °° i
] 7 r 7 i = E ¿ + r Pi(COS7) ’ (28-22)
1 1 /=0
and noting that r; occurring in the numerator of (28.21) is related to Yim, we
see that the surface integral selects only / = 1:

j>dS'Y?mYlm, ^ ^ll^mm1• (28.23)

328 CHAPTER 28. MAGNETIC MULTIPOLES

Therefore, only the / = 1 term in (28.22) will contribute to the integral (28.21),

1 j=i r r « r ' 1
- — r -r (28.24)
| r-r'|

yielding for the vector potential,

= < 2 8 -2 5 >

Using spherical symmetry, we easily evaluate the integral over the dyadic to be

j ) dS' r V = j ) dS* ^ l r /2 = ^a247ra2l . (28.26)

Therefore, the vector potential inside the sphere is

(28.27)
A = ¿ (“ xr)s5B*r’
where, using the result of (6.22), we identify the magnetic field B as

2e
(28.28)

which is uniform inside the sphere.

We now calculate A outside the sphere, where |r| > a = |r; |, and the appro­
priate expansion is
1 ° ° i
i= l CL V •Y* 1
-|---------- n r f l f*( c° s 77;) = , (28.29)
r _ r1r = rL . -rw

since, as before, only / = 1 contributes to the surface integral. The calculation

proceeds as above except for a factor of a3 / r 3 with the result

ea 1 w X r _ /xXr
(28.30)
3c r 3 r3 ’
which, upon comparison with (28.9), allows us to identify the magnetic dipole
moment as
_2
ea*
LL — ---- U?. (28.31)
^ 3c
The magnetic field B is then given by (28.16).
Notice that B is discontinuous across the spherical shell because there is a
surface current density. The values of B just outside and just inside the surface
(n = outward normal = r /a) are

3(/x •n )n — ¿í ea 2 3(u •n )n — u>

T — a “f- 0 . B — 3 q 3 ) (28.32)
T ná 3c nó
2e
r = a —0 : B _ = - — u>, (28.33)
3ca
28.3. PROBLEMS FOR CHAPTER 28 329

so the discontinuity in B is

B+ — B _ — — [(u: •n )n — u>} — — n x ( n X w ) . (28.34)

ca ca

Now recall that in vacuum (B = H), the normal component of B is continuous

while the tangential component is discontinuous if there is a surface current.
Written in the notation above, these boundary conditions [(26.18) and (26.22)]
read

n •(B + — B _ ) = 0, (28.35)
Anr
n x ( B + - B _ ) = — K. (28.36)
c

Obviously, (28.35) is satisfied. From (28.36) and (28.34), we calculate the surface
current density,

K = T-nX [— nx(nxw)l = -^-wXn, (28.37)

4w lea yJ 47ra v 7

in agreement with the direct result (r7 = an)

K = (t v = (u;Xn)a = - ^ w X n . (28.38)
47ra2 47ra

28.3 Problems for Chapter 28

1. Derive (28.31) directly from (28.10).

2. Show that a perfectly conducting sphere of radius a placed in a constant

magnetic field B acquires a magnetic moment,
Chapter 29

Magnetic Scalar Potential

We now return to the macroscopic situation with a steady current flowing in a

permeable medium characterized by a constant /i, so that the vector potential
is
A(r) = ñ|(*')jÍ & l , (29.1)

in the Coulomb gauge. In particular, consider a current / flowing in a closed

loop, as shown in Fig. 29.1, so that the volume integral is to be replaced by a
line integral,
J (dr')J •••= / d r 'I .. (29.2)

where dv' is a directed line element tangential to the wire, in the direction of
the current flow. Now the vector potential, (29.1), becomes

<»*>
which implies for the magnetic induction,

B= civ * /F = 7 f <2£U>

dr'

Figure 29.1: Closed current loop.

331
332 CHAPTER 29. MAGNETIC SCALAR POTENTIAL

or, for the magnetic field,

H = - / V : ■■— 7 Xdr' = - ¿ dv' X V ' 1 . (29.5)

c j |r —r'l c j |r-r'|
It is convenient to rewrite H in terms of a surface integral instead of a line
integral. We make use of Stokes’ theorem for a vector field, V , which reads

cf drf •V = / d S '. ( V 'x V ) , (29.6)

Jc Js
where S is any surface which has the contour C as its boundary, where the
orientation of the surface is given by the right hand rule. If we replace

V -^ V x a (29.7)

where a is an arbitrary constant vector, Stokes’ theorem becomes

<f dr' x V •a = / d S '. V 'x ( V x a )

Jc Js
= J dS' •[(a •V ')V —aV ' •V ]. (29.8)

The identity
a x ( V 'X V ) = V '( a ♦V ) - (a •V ') V , (29.9)
allows us to rewrite (29.8) as

<( dr'XV •a = / dS1 •[V ^ V •a) —a x (V 'x V ) —aV' •V]. (29.10)

Jc Js
Furthermore, if e v e r y w h e r e on the surface S, the vector field satisfies

V / . V = 0, and V ' x V = 0, (29.11)

(29.10) reduces to
/ d r 'x V = / ( ¿ S ' . V ') V . (29.12)
Jc Js
We will apply this result to rewrite (29.5) for which

1
V = V' -V -, (29.13)
|r - r'| r —r n )

which satisfies the conditions (29.11) as long as r / r', that is, at points outside
the wire. We therefore find
H = —V ^ m, (29.14)
where the magnetic scalar potential, <j)m) is

(29.15)
c Js , |r - r'l
 333

Figure 29.2: Topology of curve C relative to current loop.

where S' is a surface bounded by the current loop. According to (1.17), the
surface integral,

ls< dS1- |r-r'|3

1
dS' •V' — — = Q, (29.16)
|r-r'|
is the solid angle subtended by the current loop at the observation point, so
that
<j>m = - - Í 1 . (29.17)
c
Therefore, a scalar potential for H exists for points not on the wire, consistent
with the Maxwell equation, (26.7),

V x H = 0. (29.18)

However, this scalar potential is not single-valued. We consider the integral

of H around a closed path, C, which does not touch the wire,

j ) dr •H = — dr •V<^m = - 0 m<|>. (29.19)

The naive anticipation is that (29.19) would be zero. An alternative calculation

of this quantity can be made using Stokes’ theorem:

f dr.H = J dS ♦(V x H ) = j dS-^ -3

± ^ J , YES,
= { 0, NO,
(29.20)

where YES means the wire passes once through the surface S, bounded by
the path C, (the ± sign refers to the relative orientations of dS and J) while
NO means the wire does not pass through the surface. Some examples of this
are supplied by the illustrations in Fig. 29.2. Therefore, contrary to our naive
expectation,
YES,
(29.21)
0, NO,
334 CHAPTER 29. MAGNETIC SCALAR POTENTIAL

or
f ± 4 tr, YES,
(29.22)
I 0, NO,

which means that <¡>m is a multivalued function, as required by the fact that
V x H is not zero everywhere. The discontinuity found in (29.22) corresponds
to the fact that when one crosses the surface S' defined by the current loop and
used in the evaluation of the solid angle in Q, there is a change of 4ir.
Very far away from the current loop, the solid angle subtended by it is, if
we assume that the points of S' are localized in the vicinity of the loop

ÍÍ = S' = J d S 1, (29.23)

so the corresponding magnetic field is

H = -V (29.24)

which upon comparison with (28.12) identifies the magnetic moment of the
current loop to be

A* = I s . (29.25)

We obtain the same result if we use the definition of the magnetic moment,
(28.10),

\x — [ ( d r ' ) r ' x J (r') = ~ ¿ r ' x d r ' = - S , (29.26)

2c J c2J c

where we have used (29.10) to evaluate the line integral, which is also obvious
geometrically.

29.1 Problems for Chapter 29

1. Use (29.17) to calculate the magnetic field produced by a circular current
loop along the symmetry axis perpendicular to the loop.

2. Using either (29.17) or (26.35), compute the magnetic field produced by

a circular current loop of radius a at an arbitrary point. Express the
answer in terms of complete elliptic integrals. In cylindrical coordinates,
the results are

I 2z 1
H0 = —
C p \/{a + p)2 + z 2
I 2 a2 _ ,,2 _ 2
H; ) + 7T
m (a - pL) 2 +, z 2
29.1. PROBLEMS FOR CHAPTER 29 335

where the complete elliptic integrals of the first and second kind are, re­
spectively,

pn/2 i
m = / w h L2, 2 ,
Jo y\ — k2sm¿ ^
pn/2 / ... ..
E{k) = / difty 1 — k 2 sin2 -0,
Jo
and
p = 4a?
(a + p)2 + z2 '

3. By taking the large distance limit of the result found in Problem 2, de­
termine the magnetic dipole moment of a circular current loop. Compare
the result with (29.26).
Chapter 30

Magnetic Charge II

In the previous chapters, we have considered the magnetic fields produced by

steady currents with some attention to the attendant vector potential. As we
have indicated at various points, an alternative source of a static magnetic field
would be static magnetic charge, if such exists. We would here like to consider
a few consequences for the vector potential corresponding to such a magnetic
field.
Let a magnetic charge, g , be located at the origin so that the magnetic field
satisfies
V . B = 4 w0 «(r), (30.1)
which has the solution
= <3» 2)
Away from the origin, B is divergenceless,

V . B = 0, (30.3)

so we would once again expect B to be the curl of a vector potential,

B = V x A. (30.4)

However, this cannot be true everywhere since

j ) d S ' V x A = J (d r ) V « V x A = 0, (30.5)

while (30.1) implies for a closed surface surrounding the magnetic charge

/ d S -B = J (dr) V •B = 47rg. (30.6)

We now want to find a vector potential that satisfies (30.4) almost everywhere.
The simplest possibility is that this equation fails to hold on a line, which we
may take to be the + z axis. We apply Stokes’ theorem in the form

l A •dr = - J B •dS, (30.7)

337
338 CHAPTER 30. MAGNETIC CHARGE II

Figure 30.1: Sphere surrounding a point magnetic charge.

where C is a circle of constant S on a sphere of radius r about the origin and

S is the lower portion of the spherical surface bounded by C. (See Fig. 30.1.)
Equation (30.7) holds since (30.4) is true everywhere on S. [The minus sign
appears because we use the outward normal to the surface S .] The surface
integral follows trivially from (30.2),

J B •dS = -~-27rr2(l + cos#). (30.8)

An obvious solution of (30.7) is then

(30.9)
►
II
•e-

where
g 1 + cos 6
A <t>~ • UQ ♦ (30.10)
r sin
The structure of the singularity on the 0 axis is now isolated by taking the limit
0 - + 0,
6 dr * A = f d S '. V x A = -4 * 0 , (30.11)
Je Js 1
where C' is an infinitesimal circle about the z axis and S' is the enclosed area,
as shown in Fig. 30.2. Since (30.11) shows that ( V x A ) ¿ has the singular­
ity —4irg6(x)S(y) on the -fz axis, we conclude that the magnetic field can be
expressed everywhere by

B = V x A -f 4irg6(x)6(y)r)(z)z, (30.12)

where rj(z) is the step function,

(3013)
This result can be confirmed by noting that B has the correct divergence,

V . B = 0 + 47T3 <5(ic)<5(y) 6 (z), (30.14)

30.1. PROBLEMS FOR CHAPTER 30 339
z

C'

z = 0

Figure 30.2: Circle surrounding string.

or, alternatively, that (30.6) is consistent with (30.5). The vector potential
(30.9) is an example of a class of potentials that yield the correct magnetic field
except for a one-dimensional set of points, a curve. On this curve, called a
string, A is singular, whereas the magnetic field is regular, being the curl of A
plus a compensating singularity on the string.

30.1 Problems for Chapter 30

l. More generally, a string can point in a fixed direction ñ. Then

g nXr
A = ----------- -— .
r v — n •r

Show that B = V x A almost everywhere, and give the generalization of

(30.12).

2. Consider the vector potential given in spherical coordinates by

A (r, 0, (p) = —cos 0V<^.

Using both geometrical and analytical arguments, determine the corre­

sponding magnetic field. Compute the line integral

j ) dr •A ,

where C is a circle about the z axis as shown in Fig. 30.3. What is the
limit of this line integral as 0 —►0, 0 7r?
 CHAPTER 30. MAGNETIC CHARGE II

Figure 30.3: Contour for line integral in Problem 2.

Chapter 31

Retarded Green’s Function

31.1 Potentials and Gauges

In the previous chapters, we have primarily confined ourselves to the discus­
sion of electrostatics and magnetostatics. We will now study in general how
time-dependent electromagnetic fields are produced by arbitrary charges and
currents. In vacuum, we recall that Maxwell's equations are [see (1.65)]

^ „ 1 <9 4 tt.
V x B = 7c at
*7 ® + — (31.1)
c j,
V •E = 4717?, (31.2)

-V x E = ~ B , (31.3)
c ot
V •B = 0, (31.4)

where p is the charge, and j is the current density, and we have assumed that
no magnetic charge is present. Notice that the local conservation law
c\

v-j + d tp = 0’ (3L5)
is not an independent statement, but is derivable from (31.1) and (31.2).
To solve Maxwell's equations, we first recognize that the last two equations,
(31.3) and (31.4), make no reference to charge or current, and they can be
identically satisfied by introducing potentials through the definitions

B = V x A, (31.6)

E = _1JLA - V ¿ . (31.T)

As we have observed previously, in Section 9.5, the potentials A and (j) are not
uniquely defined. Since the magnetic field is the curl of A , it is unchanged when
a gradient is added to A ,
A A -f VA (31.8)
342 CHAPTER 31. RETARDED GREEN’S FUNCTION

where A is an arbitrary function. In order that this new choice of vector potential
not alter the electric field, (31.7), it is necessary to simultaneously replace the
scalar potential by

+ <31-9>
This new set of potentials, (31.8) and (31.9), is as acceptable as the original
one since only the fields B and E are physically measurable quantities. This
arbitrariness in the choice of potentials is called the gauge freedom of the theory,
while the corresponding transformations are called gauge transformations. In
the following, we will exploit this freedom in the process of solving the differential
equations for the potentials.
Upon substituting the constructions of B and E in terms of potentials, (31.6)
and (31.7), into the first set of Maxwell’s equations, we find, from (31.1),

_ /_ AN 1 d ( _ , 1 d \ 4 tt.
V x (V x A ) = - - - - ^ A j + T J, (31.10)

1 d2 \ . _ /_ . 1 d A 4tt.
- (v 2- ? » J a = - v (v -a + ; » ^ + t j’ <3111>

and, from (31.2),

1 Í)
— - ^ ( V •A) = 4np. (31.12)

This is a pair of coupled second order differential equations for A and <£, which
may be simplified by utilizing the gauge freedom in defining the potentials. The
two most convenient and common choices of gauge are discussed below.

1. The radiation gauge (or Coulomb gauge) is defined by the condition

V •A = 0. (31.13)

That we can always make this choice was shown in Section 26.3. In this
gauge, (31.11) and (31.12) reduce to

—V 2<¿> = 47rp, (31.14)

__o » 47t . 1 9 .v
- □ a = - j - ; 5 í (V « , (31.15)

where
m
a 2¿ = ------------
1 92 (31.16)
c2 dt2
is the d’Alembertian. (Jean d’Alembert’s Traite de Dynamique was pub­
lished in 1758.) The equation for </>, (31.14), is just the same as that in
electrostatics (hence the origin of the term “Coulomb gauge” ) so that </>
is, in principle, known. The structure on the right hand side of (31.15) is
31.2. GREEN’S FUNCTION IN THE LORENTZ GAUGE 343

proportional to an effective current, the second term of which is present

in order that it be divergenceless:

V
47t at
where the last equality follows from charge conservation, (31.5). This
relation also entails the consistency of the choice of the radiation gauge in
that if we set V •A equal to zero at one time, it remains zero for all time,
since
J_d_
- D 2( V - A ) j - V (31.18)
47t dt
2. The Lorentz gauge condition is a relation between vector and scalar po­
tentials,
1

V * A + - J ^ = °- (31.19)
In this gauge, the equations for A and t}> have the symmetrical form,

- □ 2A = — j, (31.20)

- □ 2<¿ = 4wp. (31.21)

The consistency of this gauge choice again follows from the fact that charge
is conserved,

- D2(v -A + ^ ) = T ( v -j + t ) = ° - <3L22>

31.2 Green’s Function in the Lorentz Gauge

In the following, we will solve the differential equations, (31.20) and (31.21), for
the potentials in the Lorentz gauge. Since the potentials are linearly related to
their sources, they may be expressed in terms of a Green’s function,

<f)(r,t) = J (d r ') dt'G (r - r \ t - t')p (r',t')} (31.23)

A (r ,t ) = J'(dr') dt' G (r - r ',t - 0 “ j ( r', 0 * (31.24)

This Green’s function, G (r —r', t —t'), is a function only of relative positions and
times because of translational invariance in unbounded space. Since <¡) satisfies
(31.21), this Green’s function obeys the differential equation

- □ 2G (r - r' , t - t ' ) = 4ir6( r - r ' ) 8( t - t ' ) , (31.25)

which is a four-dimensional generalization of the three-dimensional Green’s func­

tion equation we studied in electrostatics,

—V 2G (r — r') = 4ir6(r — r'). (31.26)

344 CHAPTER 31. RETARDED GREEN’S FUNCTION

To solve (31.25), we will analyze its time dependence by making use of the
exponential representations (Fourier transforms in time)

= r (31.27)
J—OO ^

/
°0 J
r _ r'), (31.28)
-oo ^

where G^ satisfies the three-dimensional differential equation,

- ( v 2 + ^ P ) G OJ( r - r ,) = 4 T r 6 ( r - r ') . (31.29)

In the static limit, u —> 0, (31.29) reduces to (31.26), the solution of which is
Coulomb’s potential, (13.3):

G a ,= o (r -r ,) = 1— [— r (31.30)
|r — r'|

Since Geo depends only o n r - r ' , we may set r' = 0 , without loss of generality
in the following discussion. Also, since we are now looking for a spherically
symmetrical solution for Gw, it is natural to use a spherical coordinate system
in which the Laplacian here reduces to

V2 (31.31)

Therefore, for r > 0, we wish to solve the homogeneous equation

1 . — ( r 2± U)
+ GU r) = 0, (31.32)
r 2 dr \ dr

subject to the boundary condition that there is a point charge at the origin. The
consequence of this requirement is most conveniently extracted by integrating
(31.29) over a sphere S of vanishing radius r 0 about the origin,

4* = ~ j (dr)V •( V G U) = - d S V rGu = -4 w r 2-^ G u (31.33)

or
= 1. (31.34)
~ r 2T rG“ V

[We have noted that the u>2/c2 term in the differential equation does not con­
tribute to the integral since

(31.35)
c¿r G w ~ -r,
— as r 0,
31.2. GREEN’S FUNCTION IN THE LORENTZ GAUGE 345

which has vanishing volume integral as ro —►0.] To solve (31.32), we introduce

defined by
Geo = - g U)) (31.36)
r
which satisfies the differential equation

( d ^ + 7 T jff* (r ) = 0, for r > 0, (31.37)

where we have used

dn -
2 d (31.38)
ri G uj — ^~T~9u ~~ Qun
dr dr
1 d ( 2 d \ _ 1 d2
(31.39)
r 2 dr \ dr / r dr2^u

The independent solutions of (31.37) have the form

gu ~ e±lW/ c, (31.40)

and the corresponding forms for G& are

Gu (r) = £ e±ia,r/ c, for r > 0. (31.41)

r
For either choice of + or — sign, the constant C is determined by the boundary
condition (31.34) to be
C = 1. (31.42)
Therefore, we have two fundamental solutions to (31.29),

G ^r - r') = _ í _ c±M r-r'|/c (31.43)

|r —r|

while from (31.28) we now obtain the space-time form of the Green’s functions,

/
oo

-oo 2 tt | r-r'|

= ¡7 r 7 ¡s ( ± ; l r - t'l - ( < - i ') ) . (31.44)

What is implied by the use of the + or — sign in (31.44)? The choice of the +
sign leads to the retarded Green’s function,

Gret(r — r ',t — t>) = |^ ± - ^ 6 (± | r - r'| - (t - * ')) , (31.45)

which is nonvanishing when

< = * ' + - |r-r'|. (31.46)

c
346 CHAPTER 31. RETARDED GREEN’S FUNCTION

This means that the signal propagates with the speed of light c from the source
(at time tf) to the observer (at time t); the effect occurs later than the cause.
If we pick the — sign, we obtain the advanced Green’s function

Gadvtr - r', t - 1') = (^r“ r'l + (* “ * ')) - (31-47)

which is nonzero when

t = tf- -|r —r;|, (31.48)
c
the signal arriving at the observer before it is emitted by the source. Since
the latter is not in accordance with elementary ideas of causality, we adopt
the retarded Green’s function as the solution which satisfies the correct time
boundary condition. (Actually, both retarded and advanced Green’s functions
are useful in physics.) We can now obtain explicit expressions for the potentials
by substituting (31.45) into (31.23) and (31.24),

<f>(r,t) = (31.49)

A (r ,t) = (31.50)

Integrating over tf, we obtain the so-called retarded or Lienard-Wiechert poten­

tials (published in 1898 and 1900, respectively),

<l>(r,t) (31.51)
= / ( * ')

A (r ,t) (31.52)
= /<*'>
These results are elementary generalizations of the potentials for electrostatics
and magnetostatics, but now reflecting the finite propagation speed of light.

31.3 Problems for Chapter 31

1. A particle with charge e moves along the z axis with constant speed v. Its
coordinates are
*(<) = 0 , y(t) = 0 , z(t) = vt.
Construct the potentials, <j) and A , in the Lorentz gauge by solving the
differential equations (31.20) and (31.21), noting that the only variables
are #, 2/, and z — vt. The result is

y ■■ 6 ......... , A =
(z - vt )2 + (l - ^ ) ( x 2 + J/2) C

(A particle in uniform motion is not very different from a particle at rest.)

31.3. PROBLEMS FOR CHAPTER 31 347

2. What are the electric and magnetic fields implied by Problem 1?

3. Obtain the results of the preceding two problems by performing a Lorentz

transformation of the Coulomb potential from a frame in which a charge
e is at rest, to one in which it is moving with velocity v. Use the trans­
formation of potentials, (10.81), and of the fields, (10.83), (10.84).

4. From Maxwell’s equations, without introducing potentials, show that the

electric and magnetic fields satisfy the inhomogeneous wave equations

-0 ’E = 4 ,(-V ,-^ | j),

- □ 2B = — V x j .
C

Extend this result to magnetic charges and currents.

5. The charge and current densities of a point charge are given by

{ j } (r,í) = e{v(í)}á(r" rW)-

From the expressions (31.49) and (31.50) for the retarded potentials, derive

^ ^ k - r ( i ,) | - [ r - r ( </) ] ‘ 2^ i ’

A (M ) =

where the retarded time satisfies

t* = t -
KOI

These are often called the Lienard-Wiechert potentials.

6. From Problem 5 compute E and B. Express the answers as the sum

of a “velocity” part (involving v only, and asymptotic to 1 / r 2, and an
“acceleration” part (proportional to v and asymptotic to 1/r). Only the
latter is significant for radiation.

7. A point charge in n Euclidean dimensions corresponds to Green’s function

G^n\ which satisfies the differential equation

- ( ¿ ¿ k /)G(n)(X i>•••.*«) =
\k= 1
•••«(*»)•

The solution to this equation can be written as the Fourier transform

G^n \ x u . . . , x n ) = / f [ ^ -e x p n 1 ,2 •
J m=l ¿1T \ m=l /
348 CHAPTER 31. RETARDED GREEN’S FUNCTION

Evaluate this, by making use o f the exponential representation

Express the answer as a function of R 2 = ^ m = i x m• Verify, as special

cases,
3) — „..1 , Q (4 ) — ---1---
“ 4 wR' 4v*Rr
8. If we integrate the above Green’s function over one coordinate, we obtain
Green’s function in one lower dimension,

/
oo

•OO

Check this explicitly for n = 4, and then in general, from the explicit
answer in the previous problem.

9. The 4-dimensional Euclidean Green’s function in Problem 7 satisfies

4 ( ± y n=
and is explicitly
1 1
G =

By making the complex replacement

X4 —> ict = lim e * ^ 2“ 6^ci,

e-*+o

show that = iG satisfies the differential equation

- ( v 2 - ¿ ¿ i ) D + (r » o = ¿(r )^(c<).

where r = (# 1 , a?2 >« 3 ). Starting from the solution in Problem 7, show that

(a)
1
D+ = 1
47T2 r 2 — (ct ) 2 + «e ++o
and
(b)

where the two terms here are the retarded and advanced Green’s
functions, respectively, apart from an overall factor of 1 / 47T.
31.3. PROBLEMS FOR CHAPTER 31 349

(c) From the Fourier representation

(dk)d 4 e’'(k-r+*4®4)
f (dk)dk
G
~J ( 2 »r)) 4 k 2 + k2 ’

make the replacements (complex rotations)

jb4 -H. e— ( - ^)

to obtain
(dk)(dw /c) e*(k *r-a,<)
f (dk)(d
J ( 2V)4 k2 - 7T - e— + 0

(d) Show that

s-«|k|c|<|
r °°
° ^ 1 _____ 1
7-0 0 c 2 ir k 2 - K - ie

and consequently

Cr t\ - i f W i ¿(k»,-cikiitn
^ 7 ( 2 7T)3 2 |k|e

(e) Show that D+ given in part (d) satisfies

and
V2 - ^ | O + (r,«) = 0

for t ^ tf. Consequently, £)+ obeys the correct differential equation,

(f) By integrating over k, obtain the alternative representation

f°
1 duj t°
e i\u>\r/ c

D.
(t' , ) = c L 2w( 4ixr

Note that the Fourier transform of D+ is a function of |a; |, while the

retarded and advanced Green’s functions are respectively functions
of u> and —co. The replacement |u>| —►—\uj\leads to a function called
£)_. The factor exp(i\u)\r/c) characterizes an outgoing wave.
(g) From (f), directly derive the result of part (b).
Chapter 32

Radiation— Field Point of

View

32.1 Asymptotic Potentials and Fields

As we have seen, the distinction between static electric and magnetic fields
and those produced by time-varying charges and currents is that, in the latter
case, we must take into account the finite propagation speed of light. The fact
that the time of emission is different from the time of detection is the basis
for the existence of electromagnetic radiation, as we will now see. Sufficiently
near the source, retardation effects can be neglected. That is, if and j do not
change appreciably over a time scale of |r — r'|/c, the time of emission, t ', can be
effectively replaced by the time of detection, t, in (31.51) and (31.52), which is to
say that the potentials are not very different from those occurring in statics. On
the contrary, far away from the source, retardation effects become important.
Choosing the origin of coordinates to lie inside the charge distribution, having
characteristic length a, we may use the expansion, for r a,

|r — r'| == \fr 2 — 2r •r' -f r /2 = r — n •r' + O , (32.1)

to derive the asymptotic form of the potentials in a Lorentz gauge [cf. (31.51)
and (31.52)],

<¿(r, t) « i J {d r ')p (r', t — £ + h i ♦r ') , (32.2)

A(r,<) « 1 y (d r')h (r '.i - V

- + h i - r ') , (32.3)

where
n = - (32.4)

351
352 CHAPTER 32. RADIATION— FIELD POINT OF VIEW

is the unit vector in the direction toward the observation point. In the above
equations, the n * r ' term in the expansion of l/|r — r'| has been deleted since
it gives rise to a 1 / r 2 term in the potential, while it has been retained in the
expression for the time of emission, t\

P « t — - -f ~n •r t = tr . (32.5)
c c
The last term in tr reflects the finite amount of time it takes radiation to propa­
gate across the source, which can be significant if the source distribution changes
rapidly, or, more precisely, when a typical frequency of oscillation of the source
distribution is of order c/a.
The fields at large distances can now be calculated by substituting (32.2)
and (32.3) into (31.6) and (31.7), and by using the evaluation

r 1 n i l . /x d_
V - + -n * - + — n x (n x r ) l/(* r )
c c c cr dt r
n d
(32.6)
c dt
We see that because of the appearance of r in the time dependences, the fields
behave as 1 / r rather than the behavior 1 / r 2 characteristic of statics, and in
particular, for r a, the field strengths are

B (r,<) « - ^ n X I J (oirO— jC r',^ ), (32.7)

EM ) « “ / (dr' ) ^ ( r '><-) “ l l J (32,8)

These two terms in (32.8) can be further combined by using the local charge
conservation condition (31.5),

r '. M = - V / -j(r ',< r) + (32.9)

where we have used the fact that the divergence operator in (31.5) acts only on
the spatial arguments of j, while V 7 in (32.9) also differentiates the r' depen­
dence of tr , where from (32.5) V 't r = n /c . The first integral in (32.8) may then
be simplified through the use of

J (d r ')V '.j = 0, (32.10)

since the charge distribution is bounded, and the remaining terms involving the
time derivative of j can be combined by means of the identity

n (n •V ) - V = n x ( n x V ) (32.11)

to read
■ \n 1 f , , ,v 1 d , ‘
E(r,<) = nX l^ -X - J ( d r ) - —3(r ,tr) = -n x B (r ,i). (32.12)
32.2. ANGULAR DISTRIBUTION OF RADIATED POW ER 353

Figure 32.1: Electric and magnetic fields for a wave propagating in the direction
n.

We observe that, far from the source distribution, E and B are perpendicular
to each other,
E = — nxB, B = nxE, (32.13)
are perpendicular to the direction of propagation n, and have equal magnitude,

E 2 = B 2. (32.14)

These are the same characteristics seen in Section 3.4, where we considered the
propagation of electromagnetic waves along a single direction, in terms of the
flow of energy and momentum. (See Fig. 32.1.)

32.2 Angular Distribution of Radiated Power

Next we ask at what rate does this time-varying charge and current distribution
radiate energy. The amount of energy flowing across a unit area per unit time
is expressed by Poynting’s vector, (3.5),

S= ExB, (32.15)
47T

which points in the direction of propagation n. Substituting the asymptotic

expressions for the fields, (32.7) and (32.12), into (32.15), we may write the rate
of energy radiated per unit area in terms of the current distribution:

n •S = — (nxE)*B = — B 2 (32.16)
47r 47T

Rather than the energy crossing an element of area dS, we would instead like
the energy radiated into a solid angle dfi,

dñ = (32.17)

since the latter measure is independent of how far away the observer is from the
source. Therefore, the amount of energy radiated per unit time (the power) per
354 CHAPTER 32. RADIATION— FIELD POINT OF VIEW

unit solid angle in the direction n is

dP
n X J (dr') j(r',U) (32.18)
dll 47TC3 dt

while the total power radiated is obtained by integrating this over all solid
angles,

p=Jda(§)- <3219>
This finite energy flow at large distances is a consequence of the 1f r behavior of
the fields, which, in turn, arises from the time variation of the current density.

32.3 Radiation by an Accelerated Charged Par

tide
Let us first apply the above general result, (32.18), to the simple example of
a particle, with charge e, moving with a velocity v, small compared with the
speed of light, v / c <C l . If R (t') is the position of the charged particle, the
corresponding current density is

j(r ', t') = e v (t') 6{r' - R (*')). (32.20)

For this situation, the time of emission, (32.5), of the radiation may be approx­
imated by

t r = t - - + - n . R ( í ') « Í - - = t e, for — <C 1, (32.21)

c c c c
since |R(¿')| ls bounded by v times a characteristic time. Therefore, the integral
in (32.18) is immediately evaluated to be

/ w ( * 0 jV , M = e^ ( ^ ) > (32.22)

implying that radiation is produced whenever a charged particle is accelerated.

(However, see Chapter 37.) The angular distribution of the radiated power is
then given by, from (32.18) and (32.22),

dP_
díl

(32.23)

where 0 is the angle between the direction of observation n and the direction
of the acceleration at the emission time te. (See Fig. 32.2.) Evidently there is
32A. DIPOLE RADIATION 355

Figure 32.2: Orientation of acceleration at emission time to direction of obser­

vation.

no radiation emitted along the direction of the acceleration. By employing the

angular integral

<32 24)
we obtain the total radiated power:

2e 2 f dv\ 2
P for - < 1 , (32.25)
3 ^ [~dt J c

which is called the Larmor formula.

32.4 Dipole Radiation

Next, we generalize the above discussion to a system consisting of many charged
particles. For a small system in which particles are all moving with low ve­
locities, the time of emission, tej does not vary significantly over the current
distribution. Consequently, the integral in (32.18) becomes

/ (dr/ ) j ( r, »<«)» (32.26)

which is evaluated by means of an identity, derived from current conservation:

o= j (dr)r [V ' j + §¡p\ ~ J (dr) [v *(jr) - j +rJ^

= -J (dr) j ( r ,t ) + J ( d r ) r p ( r ,t ). (32.27)

Recalling the definition of the electric dipole moment,

d (i) = J ( d r ) r p ( r ,t ), (32.28)
356 CHAPTER 32. RADIATION— FIELD POINT OF VIEW

we recognize (32.26) as the second time derivative of d (/). Therefore, the angular
distribution is given by

f ^ ¿ (nXa)2’ for! < 1 > (32'29)

while the total power is
P * ¿ ( d ) 2, (32.30)

where we have used a dot to denote time differentiation. Radiation described

by these formulae is called electric dipole radiation. For a single charged particle
•* dy
d = er, d = e— , (32.31)

and (32.23) and (32.25) are recovered, as expected. However, for a system of n
charged particles, the electric dipole moment is
n
d = ^ e ar a, (32.32)
a= 1
so the power radiated is not additive, but exhibits interference effects:

dñ = w (nx¿ eaiSa) • (32-33)

Let us now make a better approximation by keeping the v/c correction arising
from the ^n*r ' term in the retarded time. The integral over the current in
(32.18) now becomes

J ( d r ' ) j ( r ' , t e + ^ n .r') » d(fe) + J (d r'^ n .r '^ -jir'.te ) + •••

= d (fe )+ “ j ( d r ' ) n - |l(r'j + jr ') + |(r' j - jr ')

= d (ie )~ ^ J ( d r ' ) n x r'xj(r',*e)

2c

= d ( t e) - nX//(ie), (32.34)

where /x is the magnetic dipole moment, (28.10). [Here we have neglected the
contribution due to
¿ ( r 'j + j r ') (32.35)

since this is an electric quadrupole moment effect,

J (*')[<)* +*;*] =-J ( * > ;.; v'.j

(32.36)
=í J & W i»
32.5. POTENTIALS IN RADIATION GAUGE 357

from ( 2 2 .6 ), since the unit dyadic evidently does not contribute to the radiated
power because n X n = 0.] The angular distribution of the radiated power is
therefore more accurately given by
dP 1 , ...,0
d ñ = 4 í ? [“ x < d ~ n x , ‘ )1
= [(nxd)3 + (nxjj)2 + 2n >(dx/i)], (32.37)

where the last term represents interference between d and fx, which, since it
depends linearly on n, does not contribute to the total radiated power:

P = ¿ [ ( d )2 + ( ü n (32.38)

The behavior of the fields at large distances can be obtained by substituting

(32.34) into (32.7) and (32.12),

B ~ -4r-nx(d-nx/i), (32.39)
c1 r
E ------n x B ~ - " n x ( / ¿ + n x d ) . (32.40)

Note that these results are invariant under the replacements E —> B and B —►
—E together with d /x and /x —►—d, which is a manifestation of the symmetry
discussed in Chapter 2. We here have an indication of the connection between
the directions of the electric field for electric and magnetic dipole radiation.

32.5 Potentials in Radiation Gauge

To this point, we have discussed radiation by use of the Lorentz gauge. However,
we do have the freedom to choose an arbitrary gauge without affecting the
physical results. As an illustration, let us here consider the radiation gauge,
which exhibits a certain physical simplicity. In this gauge, the potentials satisfy
the differential equations (31.14) and (31.15), that is

- V 2<t> = 4irp, (32.41)

- D’ A = T ( j - ¿ v S*)' (32 42)

while the vector potential is subject to the gauge condition (31.13),

V •A = 0. (32.43)

The electric and magnetic fields are obtained from these potentials by the rela­
tions (31.6) and (31.7). In order to make contact with what has gone before, we
consider the fields at large distances, which are those of interest for radiation.
The solution of (32.41) is the Coulomb potential,

4 (r ,t) = J (d v ')^ y ^ . (32.44)

358 CHAPTER 32. RADIATION— FIELD POINT OF VIEW

which, asymptotically, behaves as

<j>~ - , (32.45)
r
with e the total charge. Since the gradient of this is inversely proportional to
the square of the distance,
-V<j> ~ ^ r , (32.46)

we can neglect the scalar potential in computing the radiation fields, which
decrease only as 1/r. Therefore, the vector potential alone determines the ra­
diation fields:
1 <9A
E (32.47)
c dt ’
B = V x A. (32.48)

We also note that the gauge condition (32.43) enforces the transversality of these
fields. That is, as a consequence of V •A = 0, we recover the scalar Maxwell
equations outside the sources,

V •B = 0 and V - E = 0, (32.49)

which, by virtue of (32.6), supplies the relations

n*B = 0 and n * E = 0, (32.50)

while E _L B follows immediately from (32.47) and (32.48).

To solve (32.42) for the vector potential, we first write the solution to (32.41)
symbolically as

¿ = _ V 2 4^ > (32.51)

the time derivative of which is

d 1 d 1 . _ .
(32.52)
m * = - v = 4 ” a / = v ’ 4” v -J'
Consequently, we may rewrite (32.42) as

9 4 tt / VV\ .
(32.53)
- DA= c l1- V » ) - j’
which makes the radiation gauge condition (32.43) transparent. The solution
to (32.53) may be obtained from that of (31.20) by applying the operator

VV
1 - (32.54)
V2
to (31.52):

A ( r ,i) = ( l - ^ J ) J (d r ') (32.55)

—r'|
32.6. PROBLEMS FOR CHAPTER 32 359

At large distances, by making use o f (32.1) and (32.6), we have eifectively the
replacement

v - - ” s <32-56>
so that the operator 1 — W / V 2 can be replaced by

VV ( _ s JL) ( _üJL)
_ V _
1 - c dt) \ c dt)
(32.57)
V2 (_ “ A )2
V c dt)

Notice that this symbolic notation is convenient when 1 /V 2 can be computed

simply. By making use of (32.11) and (32.1), we obtain the asymptotic form of
the vector potential, in the radiation gauge, to be

A ( r ,t) ~ ( 1 - n n ) ' ^ J ( d r 1) j(r ',< r)

= -n X n X ^ y \ d r ') j ( r ', t r) (32.58)

The resulting electric and magnetic fields arq precisely the same as those found
in the Lorentz gauge, (32.7) and (32.12).

32.6 Problems for Chapter 32

1. A particle, of charge e and mass m, moves with speed v, v/c <C 1, in a
uniform magnetic field B. Suppose the motion is confined to the plane
perpendicular to B. Calculate the power radiated P in terms of B and v,
and show that
„ dE
dt ~ 7 ’
where E is the energy of the particle, and find 7 . Since then
E (i) = E( 0)e - 7 *

I / 7 is the mean lifetime of the motion. For an electron, find I / 7 in seconds

for a magnetic field of 10 4 gauss.

2. A nonrelativistic particle of charge e and mass m moves in a Hooke’s

law potential (a linear oscillator) with natural frequency u>q. Again find
P, the power radiated. Recall that for such motion, the time-aver aged
kinetic and potential energy satisfy

t = v = \e .
2
Show then that the power radiated, averaged over one cycle is

and find 7 . Compute I / 7 in seconds when cjo is 1015 sec 1 (a characteristic

atomic frequency, corresponding to visible light).
360 CHAPTER 32. RADIATION— FIELD POINT OF VIEW

3. An electron of charge e and mass m moves in a circular orbit under

Coulomb forces produced by a proton. The average potential energy is
related to the total energy by

Suppose, as it radiates, the electron continues to move on a circle, and

calculate the power radiated, and thereby —dE/dt, as a function of E
(the relation is no longer linear). Integrate this result, and find how long
it takes for the energy to change from E 2 to E\. In a finite time the
electron reaches the center, so calculate hów long it takes the electron to
hit the proton if it starts from an initial radius of

^initial — 10 cm.

(This instability was one of the reasons for the discovery of quantum me­
chanics.)

4. Derive the alternative form for the angular distribution of radiated power,
(32.18)

dP _ 1
dD 4irc
Chapter 33

Radiation— Source Point of

View

33.1 Conservation of Energy

Having examined the radiation fields, we turn our attention to an examination
of the source of the radiated energy. Energy and momentum are transferred
from the charges to the electromagnetic field; the rate at which the current does
work on the field is

d / e2+ b 2
—j . E = + V (33.1)
at 8w ( ¿ exb)'
which is the local statement of energy conservation, (3.6). When (33.1) is inte­
grated over a large volume enclosing the charge and current distributions, the
conservation of total energy follows:

/ < *> ( - j •E >= s / < *> sLg¡r1 + / JS- ¿ E x B = é E + p ’ (33-2>

or stated in words, the rate at which the charged particles transfer energy to
the electromagnetic field is equal to the sum of the rate of increase of the total
electromagnetic energy, E , in the volume, and the rate of flow of energy, P, out
of the surface bounding the volume. Equation (33.2) gives us an alternative way
of calculating the radiated power, P, by computing the rate at which energy is
transferred to the fields,
J (d * ) (—j *E), (33.3)

and discarding total time derivative terms, which are not associated with ra­
diation. From this point of view, we need to know the electric field inside the
current distribution, in contrast to the previous discussion, in which we com­
puted the radiated power by evaluating the fields far from the source.

361
362 CHAPTER 33. RADIATION—SOURCE POINT OF VIEW

33.2 Dipole Radiation

To illustrate this method, we again consider dipole radiation produced by a
small charge distribution, for which the nonrelativistic approximation is valid,
that is, all particle speeds are small compared to the speed of light, v/c <C 1 .
In this limit, we will require an expression for the field E accurate to order
1 / c 3, which means, from the definition (31.7), that the scalar potential <j> must
be expanded in powers of 1 /c up to order 1 /c3, while the vector potential A
need only be expanded up to order 1 / c 2. In the Lorentz gauge, the appropriate
expansion of the scalar potential (31.51) is

where we have omitted the second term on the right-hand side of the first line
because the total charge e is conserved,

f ( dr,)í p = 7te = 0- (33'5)

Similarly expanding the vector potential (31.52), we find

A(r,<) « i / ( * ' ) j | ^ i - ¿ / ( * ' ) | j ( r ' . * ) + - .

where we have used (32.27) and (32.28) for the electric dipole moment d. The
contribution of the 1 /c 3 term in (j) to E can be simplified as follows:

-V ¿ /( d r 'X r -r y J L r t t ',» )

(1 ) (33.7)

The expression for the energy transfer, (33.3), becomes, in this approximation,
3 3 .3 . H A M IL T O N IA N 363

(33.8)

(33.9)

(33.10)

where we have again used (32.27) and have set aside total time derivative terms,
which do not contribute to the radiation. The power radiated, (33.10), is the
same as that found in the preceding chapter, given by (32.30).

33.3 Hamiltonian
As a byproduct of this source approach we may identify the total electromagnetic
energy of the system, to order 1/c2, by comparing (33.2) with the total time
derivative terms in (33.8),

+ ¿ / i d r ) ( d r ') ( * ) ) lr - r'l + •••> (33.11)

where the first two terms have the form of the electrostatic and magnetostatic
field energies. However, the sign of the current-current interaction is opposite to
that of (27.14). The resolution of this apparent discrepancy requires the third
term in (33.11), which we rewrite by means of the local charge conservation
condition, (31.5):
364 C H A P T E R 33. R A D IA T IO N — S O U R C E P O IN T O F V I E W

which, when combined with the first two terms in (33.11) yields

t) , [(r-rQ .j(r,<)][(r-rQ .j(r',<)]

+ ¿ /< * > < * ') { « • I _./|Q

(33.13)
For point charges, the charge and current densities are

P(r,t) = ^ 2 e aS ( r - r a(t)), (33.14)

a
j( r ,t ) = ' ¿ T e av a6 ( r - r a(t) ) , (33.15)
a

so the terms in (33.13) referring to the mutual interaction of the particles are
Va'Vt, t ( r ai, . V a ) ( r a b ' V b )
Afield = I2J^2 tr ab
t + ¿4 c 2 Y^ , e « e i I 3 (33.16)
aj£b a^b
. ah ' ab

where
= ra - r6. (33.17)
To this we must add the particle kinetic energy, from (10.11),

¿particle = ^ m a C2 ^ ( l - ) - 1^

V—s f\ 2 ^ va \ va -t (33.18)
« E 2 m“,'« + 8 mV + - ’ ^2 < 1 ’
a ' '

in order to obtain the total energy of the system:

^ to ta l := -Afield “b -^particle* (33.19)

We now wish to describe this mechanical system in Hamiltonian language.

We recall [from (8.16)] that the Hamiltonian is related to the Lagrangian by

H = ]T)pa*v<. - £({*»}, { n } ) , (33.20)

a

where the canonical momentum p a is defined by

Pa = (33.21)
dva
To determine the Lagrangian, we substitute (33.21) into (33.20),

fí = E »WnM2^ . Vi„ í({ri){vi))| (33.22)

3 3 .3 . H A M IL T O N IA N 365

where H is given in terms of v by (33.19). Equating terms on either side of

( 3 3 .2 2 ) that are independent of v, quadratic in v, and quartic in v, we find for
the Lagrangian
r 2 , 1 Va \ , 1 V ' f v a.vb (rai • va)(ra6 • v6) V
L = £ {2 ’maV° + 8'710 V2 ) + 4 c ¡ 5a,b eaC4 ( — + ---------- Z ---------- )
aibb
1 €qg& (33.23)
a^tb

Note that the particle term in (33.23) agrees with the expansion of (10.20). The
canonical momenta are given by (33.21),

Pa = m ax a +
, ^1r n a —vl va
2 ~ + ^ „2 ^ 2 e ae i
V» raj(rat •vt)
(33.24)
bjia f*ab ' ab

in terms of which the Hamiltonian of the system is

jj 1 v~> e a^b — ^ 'r‘2I
p 11 Pi \
H = n Y
2 tí' rab 2m,a 8m % c2 )
a^tb
e ae h
—
4^2 Y
4c2 / ti im a ^ b L
Pa * P b
r ab
+ (rab ’ P a ) ( * a f P t ) (33.25)
1ab
a^bb

where we have consistently kept terms up to order 1/c2. This form of the
Hamiltonian is appropriate for small systems and has application in both atomic
and nuclear physics. It is called the Darwin Hamiltonian when it is applied
classically, and the Breit Hamiltonian when it is applied quantum mechanically
(with an accompanying re-expression in terms of Dirac matrices).
From the general discussion given in Chapter 9, the canonical momentum
can be expressed in terms of the vector potential

p0 = x a- l / a + ^ A ( r . , < ) , (33.26)
V 1 - v l /c 2 C

where we have used (10.11) in generalizing (9.19). Upon comparison with

(33.24), we obtain an explicit form of the vector potential, A (ra,t), which,
when generalized to an arbitrary position, reads
vj__ . [(r —r¡,)»v¡,](r —r¡>)
+ | r -r ¡,|3
>-n|
Vi
=£ (33.27)

We notice that this vector potential satisfies the radiation gauge condition
(31.13):

vj •V
|r-n|
iv 6 .V (V 2| r -r 6|) = 0, (33.28)
366 C H A P T E R 33. R A D IA T IO N — S O U R C E P O IN T O F V IE W

since
V 2|r —rj| = (33.29)
|r-rt|’
and, as required, it also satisfies the differential equation (31.15) to order 1 /c 2:

- □ 2A k - V 2A
4»v,«(r - r.) + j V ( v , . V jT T ^ j) ]
- Eb ?c
4 ir .. 1 9 #/ x
(33.30)

Furthermore, if we rewrite the last term of (33.25) in terms of A, (33.27), we

obtain

X ) e“Va ’ J (<fr)j(r,<).A(r,<), (33.31)

which has the form of the magnetostatic field energy, (27.11), which is therefore
correctly given by (33.11). As we commented in Chapter 27, because of the
negative sign in (33.31), “like” currents attract each other.

33.4 Problems for Chapter 33

1. Verify the Lagrangian (33.23) and the Hamiltonian (33.25).
Chapter 34

Models of Antennas

34.1 Simplified Model

We have been discussing a small system, in which the time delay effects are not
great. For an example of the opposite situation, let us consider an oversimplified
model of an antenna. In this model we have a wire of length l, and of negligible
cross section, carrying a current density flowing in the z direction:

Jz = I S (x )6 (y ) sin u t , ~ ^ < z < ^ , (34.1)

which has the property

V .j = — J = i ° for (34.2)
dz z \ # 0 for z = ± t.

From the local charge conservation condition (31.5),

<343)
we see that (3 4 .2 ) implies that there is an oscillating charge density at both
ends of the antenna. In any realistic model, Jz will depend on 2 . We lack this
dependence since our model assumes that the antenna is fed at every point along
its length. Even though this model is oversimplified, it possesses many of the
significant characteristics of a real antenna. To compute the power radiated,
(32.18), we evaluate the integral
' , 1 T .> A
/
d ( r
+ - n .r j
1
( * ') - d i Jl (r c

LÜ ,J/2 (' r 1 , A
= - I / d z' COSU) 1t ------ h - z cos# 1
c ,J—1/2 \, c c J
W/2 / r\ /'u>zf ü
= -/ / d z' COSCJ t ---- COS ---- cos 6
C ,J - l /2 \ c) \< c

367
368 C H A P T E R 34. M ODELS OF ANTENN AS

Figure 34.1: Geometry of linear antenna.

2~7 cosa; (t — £) sin cos9)

(34.4)
c

where, as indicated in Fig. 34.1, 6 denotes the angle between the direction of
observation and the antenna. The angular distribution of the radiated power,
at the observation time t, is then

d p (t) _ J_ „ 42 cos2 a; (t —§) sin2 (ffcosfl)

(34.5)
-2
dil ire cos2 0
which, when averaged over one cycle of oscillation, becomes
dP 12 sin2 0 sin2 cos 9 )
(34.6)
dQ 2i tc cos2 9
To rewrite (34.6) in terms of more convenient parameters, we recognize that,
far from the antenna, the fields oscillate periodically both in space and in time:

(34.7)

so we identify the relation between the frequency v (Hertz) and wavelength A

to be
\ v — c, (34.8)
or
u 2ir
(34.9)
7 “ T ‘
Therefore the radiation of the system is characterized by two parameters: the
length of the antenna, /, and the wavelength of the radiation, A. The combina­
tion that appears in the expression for the power, (34.6), is
ÜJ l wl
(34.10)
75 = T ’
in terms of which the angular distribution is
dP _ 12 sin2 0 sin2 ( y cos 6)
(34.11)
dQ 2irc cos2 0
3 4 .1 . S IM P L IF IE D M O D E L 369

Figure 34.2: Radiation pattern produced by a short antenna, / < A. In this and
the following figures axial symmetry about the 2 -axis is to be understood.

In particular, in the direction perpendicular to the antenna, 9 = j , the radiated

power is proportional to the square of the length of the antenna,

dP_ 12 / irl\ 2
(34.12)
^ g=x/2 2'7TC \ \ J

To appreciate the characteristic features of this radiation, we will consider the

application of this general formula, (34.11), to three special circumstances:
1. A > /.
For a short antenna, / < A, the approximation

sin2 (y cos 0)
(34.13)
cos2 0
holds true for all angles. The resulting radiation pattern may be alterna­
tively derived from the dipole radiation formula, (32.29), which is appro­
priate to a small system:
2
dP _ 12 nl
sin2 0. (34.14)
df2 ~ 27tc T
2. A > /.
When y < 7r, the argument of the factor sin2 (y-cos0) goes from 0 to
something less than w when the angle 9 varies from “ to 0. Therefore,
the only angles at which the power radiated vanishes are 0 and 1r, so the
radiation pattern has a single lobe. (See Fig. 34.2.)
3. A < / < 2 A.
When 7T< y < 27Tthere is an additional zero in the radiated power at the
angle 9 — cos""1 ( j ) . Consequently the radiation pattern exhibits both a
main lobe and two side lobes. See Fig. 34.3. Evidently, as //A increases,
more and more side lobes appear.
370 C H A P T E R 34. M ODELS OF AN TEN N AS

Figure 34.3: Radiation from an antenna of intermediate length, A < / < 2A.
This diagram is meant to be understood schematically only. Actually, the side
lobes are far smaller than indicated.

The total power radiated by this antenna may be obtained by integrating

(34.11) over all angles:

f
I2 sin2 0 sin2 ( y cos 6)
27rsin 9 dO
27TC cos2 0
p ,*/2 sin2 ( j - s in x )
= — / cos x dx cos X
C J —7T/ 2 sin2 x
212 sin2 z
1- (34.15)
c ( tt/ / A ) 2 J

where we have made the successive changes of variables,

7T 7r/ .
* = y sinx- (34.16)

If y- 1, the second term in the square brackets in (34.15) is negligible com­

pared with the first for the significant values of z y so we find
„ 2/2 7rZ /°° sin2 z 7r2 /2 /
P ^ -------- r (34.17)
c Al dZ = c A'
where we have used the integral
poo
sin2 z P

l =i
_

(34.18)
z Jo t 2
Here we observe that the total radiated power, (34.17), increases linearly with /,
while the power radiated in the direction perpendicular to the antenna, (3 4 .1 2 ),
is proportional to /2; that is, as the length of the antenna increases, a larger and
larger fraction of the radiated power is concentrated near 0 = 7r/ 2 .
To see how much energy is radiated into a very small angular range near
X = 0 (or 6 = 7r/ 2 ), we consider the power radiated into the main lobe by a
long antenna
0<X< j <1. (34.19)
3 4 .2 . C E N T E R -F E D A N T E N N A 371

Following the same procedure used to obtain the total power radiated, (34.15),
we find for the total power radiated into the main lobe

(34.20)

The fraction of the energy radiated into the main lobe is obtained by taking the
ratio of (34.20) to (34.17):

(34.21)

where the infinite series is derived by integrating by parts repeatedly. Over

90% of the power is radiated into the main lobe, which has angular width
A//, implying that the radiation from the antenna is highly directional, a fea­
ture characteristic of large systems. In contrast, small systems, for which the
dipole approximation is valid, typically have angular distributions proportional
to sin2 9.

34.2 Center-Fed Antenna

The previous model is greatly oversimplified, of course, and the results (see
Problem 34.2) greatly de-emphasize the importance of radiation into the side
lobes. Therefore, we turn to the consideration of what might seem to be a
somewhat more realistic model, that of a cen ter fed, linear antenna , described
by
\z\ < 1 / 2, (34.22)

where now we have introduced the wavenumber k = co /c = 27r/A. Now, when

we repeat the steps in (34.4) we find
372 C H A P T E R 34. M ODELS OF ANTENN AS

= kl J d zf |sin k ^'(cos 9 —1) + — 'j — sin k ^z'(cos 9 + 1 ) — -^

x coso;
coso; (t — £)
(?cosd)
21 kl
cos cos ■ (34.23)
sin2 $ I V2 J 2

The power radiated into a given solid angle is then given by (32.18), or

dP 1 4/2 cos2 ui kl \ kl
12
sin2 6 cos ( — cos 9 J —cos — (34.24)
dQ, 4 it c sin4 6

which, when averaged over one cycle, gives

dP 1 kl A
— cos 9
«I2
—cos —
COS (34.25)
dO, 27tc sin2 9 2 J 2 J

Notice for a short antenna, A ^> /, we recover the dipole formula:

dP __ (J/2 ) 2 / tt/
y ) sin2 $, (34.26)
dQ 27TC

which may alternatively be derived from the dipole radiation formula, (32.29).
See Problem 34.1.
The angular distribution of the power radiated, d P /d Q ,) vanishes whenever

47rn\
n = 1,2, — (34.27)
IF J

Thus side lobes appear in the radiation pattern whenever kl passes through a
multiple of 27r. Plots of d P /d Q , are given in Figures 34.4 and 34.5. It will be
noted that now as //A —►oo, substantial energy is radiated into the extreme
side lobes, increasingly near the z axis, the antenna direction. See Problem 34.5.
This is very similar to the radiation pattern produced by impulsive scattering-
see Chapter 37. Therefore, such an antenna is most useful in the half- or full-
wave regime.

34.3 Problems for Chapter 34.

1. Derive (34.14) and (34.26) from the dipole radiation formula, (32.29).

2. Using Mathematica, Maple, or any other computer program of your choice,

plot accurately the radiation pattern produced by the simplified antenna
for / = A/2, / = 3A/2, l = 5A, and / = 13A/2. Calculate the fraction
of power in the main lobe and check how closely the limit ( 3 4 .2 1 ) is ap­
proached.
3 4 .3 . P R O B L E M S F O R C H A P T E R 34. 373

Figure 34.4: Radiation pattern produced by center-fed antenna for kl = 7r. This
is called a half-wave antenna because / = A/2.

Figure 34.5: Radiation pattern produced by center-fed antenna for kl = 57r.

374 C H A P T E R 34. MODELS OF AN TEN N AS

3. Verify the result (34.21) by computing a sufficient number of terms in the

series, and by direct numerical integration. Give an error estimate for
your answer.
4. Obtain formulas for the angular distribution of radiated power for a half­
wave center fed antenna, kl = w, and for a full-wave antenna. Plot the
latter, and compare with Figure 34.4.
5. Consider (34.25) for kl 1. Compute the total power radiated in this
limit, and compare with the power radiated in the two extreme side lobes,
corresponding to n = [ //A ] and n = 1 in (34.27), where [x] denotes the
largest integer less than or equal to x .
6. A straight wire of negligible thickness and infinite length carries a current
that varies in time and with distance 2 along the wire, according to the
relation
I(z\ t) = Iqcos(kz ut).
—

Prove that this current does not radiate if the propagation constant k
exceeds k , the intrinsic wavenumber of the external medium, supposed to
be uniform and infinitely extended. Verify that radiation does occur if
k < & , and that the time-averaged power radiated per unit length of the
wire is given by

Note that this agrees with (34.17) if k = 0.

Chapter 35

Spectral Distribution of
Radiation

35.1 Spectral and Angular Distribution

In the previous chapters, we discussed the angular distribution of the radiation
produced by a time varying charge and current distribution. Here we will turn
our attention to the spectral characteristics of this radiation, that is, its depen­
dence on frequency, or wavelength. To investigate this dependence, we return
to our starting point, the potentials in the Lorentz gauge, given by (31.49) and
(31.50),
¿ ( i | r - r '| - ( < - * ') )
= J (dr1) d ------- F v ]------- p v >* )' (35.1)

A (M ) = J(dv') dt (35.2)
•r'| c
In deriving these results, we had used the spectral representation [cf. (31.44)]

6 !|r _r/| _ ( < _ * ' ) ) = ¡ ° ° — eiw[lr-«-/l/c-(i-i')] (35.3)

c J J-OO 2 tt

If we now reinsert (35.3) into (35.1) and (35.2), and carry out the i' integration
by introducing the temporal Fourier transform

f dt e 7(0 = /(*),
/ A(jJt (35.4)

du -icot
l-oo ( 7(w) = /(<').
2 tt
(35.5)

we obtain the Fourier transformed versions of (35.1) and (35.2):

/
piw\Y—r |¡c
^ ú id r—r 'l/ c
(35.6)

375
376 C H A P T E R 35. S P E C T R A L D IS T R IB U T IO N O F R A D IA T IO N

/
i w \ r - r'|/c i
(35.7)

We observe that if f(t') is a real function of t' , its Fourier transform, /(u>),
satisfies the condition
/(«)* = / ( - " ) ; (35.8)
consequently,
/ ( « ) * /( « ) = |/(^)|2 = /(-« )/(« ) (35.9)
is a real positive number, symmetric under the interchange w —►—u>. This
implies that the algebraic sign of u is not significant, since only its magnitude
enters into physical quantities.
Let us focus our attention on the radiation fields, far from the sources.
Following the procedure given in Section 32.1, in particular, using the expansion
(32.1) for |r —r'| in (35.6) and (35.7), we obtain the asymptotic expression for
the potentials, in terms of spatial Fourier transforms,
Ju>r/c p
<¿(r,w) ~ (dr')e-iWn' r /cp(r',u), (35.10)
iu r/ c p i
A(r,w) - - —
e / ( d r O e - ^ '- ' / ^ j ^ » , (35.11)

where n = r/r. Evidently, the effectiveness of radiation with a given wavelength

and direction of propagation depends upon the Fourier analysis of the time and
spatial dependences of the charges and currents. In the exponential, the term
u)j\ •r'/c, which is of the order of the ratio of the size of the system to the
wavelength of the radiation, is significant for all but small systems. We call

k = ~n (35.12)

the propagation vector, in terms of which the potentials are written as Fourier
transforms in space and time,

(35.13)

where, for example,

p( k,w) = J ( d v ' ) e - i k ' r' p ( r ' , U ). (35.14)

The corresponding field strengths can now be computed from the time Fourier
transforms of (31.6) and (31.7):
ÍÜ
E(r,u>) = i—A(r,w) —V^(r,u>), (35.15)
c
B(r,u>) = V xA (r,w ), (35.16)
3 5 .1 . S P E C T R A L A N D A N G U L A R D IS T R IB U T IO N 377

where we have used the effective replacement

d_
—iu , (35.17)
di

because

/
OO Q

(35.18)

provided that .F(r, i) —» 0 as t —» -boo. Physically, the time boundary conditions

state that in the infinite past and the infinite future, nothing is happening: what
is significant for our observation takes place in a finite time interval only. The
measure of significant variation in r is the wavelength, A = 27rc/o;, as indicated
by the effective replacement for the gradients in (35.15) and (35.16),
V ik (35.19)

[recall (32.6)]. Consequently, the asymptotic forms of the electric and magnetic
fields are
u) eiu>r! c 1 u eiiürí c
E(r,w) ~ - j ( k , w ) - i - —— np(k,u>), (35.20)
e iu>r/c i
B(r,u>) ^ i ---------- nX-j(k,u>). (35.21)
c r c

By using the Fourier transformed version of the local charge conservation con­
dition, (31.5),
<jüp(k,u>) = k •j(k,u>), (35.22)
we may rewrite the second term of (35.20) as
¿ Ju r/c • Ju>r/c i
-----------nk •j(k,o;) = -------------- nn* -j(k,a;). (35.23)
c r c r c

The electric field now becomes

e iu r/ c i
E(r,o;) ^ ------- i —( l —n n) •—j(k,cj)
r c c
e iu>r/c w
---------- i —n X nX ^j(k,w )
r c
—nxB(r,u>), (35.24)

which reconfirms (32.13).

Before proceeding, we remark that the relation between the two terms in
(35.20) can also be obtained by using the Lorentz gauge condition; this is not
surprising since the consistency of the Lorentz gauge depends upon current
conservation [see (31.22)]. The Fourier transform of (31.19) reads

V *A (r, w) —i — <t>{c, u ) = 0, (35.25)

378 CHAPTER 35. SPECTRAL DISTRIBUTION OF RADIATION

which becomes, upon using the asymptotic replacement (35.19),

ik[n •A(r, u>) —<^(r, u;)] = 0, (35.26)

<^(r, a;) ~ n •A(r, u>), (35.27)

so from (35.15) the reduction (35.24) follows.

The instantaneous flux of energy, at a particular time t , is given by Poynting’s
vector
S(r,<) = ^E (r,<)xB (r,<), (35.28)
47r
so the total radiated energy crossing a unit area of surface normal to S is

/
OO r poo 1*00 1.
dtS(v,t) = — / dt — E(r,o))*eia,ixB(r,i)
-oo 47T J —oo J —oq ^
oo J
/

= n¿ / _ „ 5 í |B(r’" ,l>’ (35'29)

using (35.24), and the fact that n •B = 0. The energy flows in the direction of
n and the energy radiated per unit area perpendicular to this direction is

dt n •S du |B(r,a>) (35.30)

where we have used the symmetry property (35.9). As before, it is more useful
to consider the total energy radiated into the solid angle dFl [see (32.17)],

/
oo r f°°
dt(n*S)r2dO, = ffir2— / dw |B(r,a;)|
-oo 47T2 J o
dE d2E
= dCl EEdft / (35.31)
dQ Jo du)dQ'

where [cf. Problem 35.1]

d2E or
é M = 3 Í v | n * j ( k ' " )l

UT
- ip(k »w)r (35.32)
47T2 C

is the general expression for the spectral distribution, the energy radiated per
unit frequency per unit solid angle in the direction of observation n. This
equation is the analog of (32.18) for dP/dD.
3 5 .2 . S P E C T R A L D IS T R IB U T IO N F O R D IP O L E R A D IA T IO N 379

35.2 Spectral Distribution for Dipole Radiation

As an application of the general result, (35.32), we consider a small system with
typical length a much smaller than the reduced wavelength of the radiation,
that is, A = A/27T a. Then, the exponential factor in the Fourier transforms
in (35.10) and (35.11) can be approximated by unity, since

^•n-r' = ^-n.r' < 1, (35.33)

whence the Fourier transform of the current becomes that of the time derivative
of the electric dipole moment, according to (32.27) and (32.28):

j(k,w) = j dt
J u t
d_
dt
d(<) = —ia)d(w). (35.34)

Then, the spectral distribution becomes

d2E
|nxd(w)| = ^ Sin2 ^|d(W)| , (35.35)
~
3
dudD 4 tt2 c 3

where 0 is the angle between the observation direction and the direction of
d(o>). For the small system discussed here, the only reference to the direction
of observation, n, occurs as a multiplicative factor, implying the sin# behavior
exhibited above, characteristic of dipole radiation. For larger systems, n also
enters in the exponential so that the angular distribution could be completely
different. The total energy radiated per unit frequency range can be obtained
by integrating (35.35) over all angles,

él - /V d%E (35.36)
du> ~ J d u d il

where we have used (32.24).

Suppose we further specialize to a single point charge in nonrelativistic mo­
tion, corresponding to the current density

j(r, t ) = ev(<)6 (r - r(<)), (35.37)

in which case, of course,

d (t) = ev(t), (35.38)
of which the Fourier transform is

d(u>) = ev(cj). (35.39)

The energy radiated per unit frequency interval, d E / d u j, is

dE 2 o, / \i2 ^ , x,o
(35.40)
380 C H A P T E R 35. S P E C T R A L D IS T R IB U T IO N O F R A D IA T IO N

where v(w) is the Fourier transform of v(t):

(35.41)

The total energy radiated,

„ I 00 d E ; e2 2| .
(35.42)
2 2

can also be obtained from the Larmor formula, (32.25):

(35.43)

as expected. In the above, we have used (35.41), and the theorem

(35.44)

Thus we see that we can calculate the total energy radiated by a small system
through the use either of the spectral distribution, (35.36), or of the Larmor for­
mula for the power, (32.30). The equivalence of these two descriptions is demon­
strated generally in Problem 35.1, where the spectral distribution d 2E/du>dQ,,
(35.32), is derived directly from the angular distribution of radiated power,
d P ( t ) /d S l , (32.18).

35.3 Damped Harmonic Motion

As a further simple application of the spectral distribution, consider a model in
which a charged particle undergoes damped motion in a Hooke’s law potential,

r = -wgr - yr, (35.45)

where, due to the radiation produced by the accelerating charged particle, there
is a damping force represented by —y r , which we will assume to be small:

(35.46)

This model is often taken as an oversimplified description of a bound electron

inside an atom. (Recall Section 5.2.) Given the initial conditions that at t = 0,
the particle has a displacement a from the force center and has zero velocity,
the solution to (35.45) when (35.46) holds is approximately

r (t) fa a cos wqí e 7t/ 2, for t > 0 (35.47)

3 5 .3 . D A M P E D H A R M O N IC M O T IO N 381

which exhibits the fact that many oscillations are necessary before significant
damping occurs. The velocity of the particle,
v(f) « —a w0 sin wote ytl2, for t > 0, (35.48)
has the Fourier transform
POO

v(w) = / dt (—auo)ew( sinwof e-7 */2

Jo
Poo

_g*(w —W o+*7/2)t e i(u>+<ju0 + i y / 2 ) t

Jo
( ____ L__________I____ A
OJpa
2 \u> —u>o + iy/2
i y /2 u>
u>+ u>>00 -1- /0 )J ’
+ iy/2 (35.49)
Without loss of generality, we may assume u > 0 here, in which case the two
terms are very different. Only the first denominator can be small, implying
that radiation is predominantly emitted with frequencies u> « u>0 . Therefore, we
approximate the square of the magnitude of (3 5 .4 9 ) by
wna
2„2
1
|v(u;)|2 (35.50)
4 (a; —u>o)2 + 72 / 4 ’
implying for the energy radiated per unit frequency range, (35.40),
dE_ 2 e2 ,2^02 „2 1
for ' w0. (35.51)
dio 3 7TC3 ^ 4 (u; —o>0) 2 + 7 2/ 4 ’

The behavior of [(a; —loo) 2 + 7 2 / 4 ]- 1 is plotted in Fig. 35.1, which exhibits

what is called the Lorentzian line shape. Since when |o; — c^o| = 7/2, the
intensity is half that at u>p, where the intensity is maximum, 7 is the width of
the spectrum at half-maximum intensity. In the limit of negligible damping,
7 —>0, only one frequency, cjp, is emitted, with infinite intensity. In general, a
range of frequencies is emitted, with 7 being a measure of the sharpness of the
spectrum. That is, if the system decays slowly ( 7 small), the emission line is
very narrow, while if it decays rapidly ( 7 large), it is very broad.
To demonstrate that no significant amount of energy is radiated outside of
this sharp peak, we evaluate the total energy radiated in the peak by integrating
(35.51) over all frequencies:

F/rad
Jo
to dw
2_e* 2( r duj
3 7TC3 (to - up)2 + 72/4
2 e 2 ^o « 2 2 tt
____
(35.52)
3 7rc3W° 4 y ’

where we have used the integral

/
/ OO 27r
/
OO
du) dx
(w - w o )2 + 7 2 / 4
-2a;o/7 (t2 /4)(1 + x2) -col + X 2 7 *
(35.53)
382 C H A P T E R 35. S P E C T R A L D IS T R IB U T IO N O F R A D IA T IO N

CO,,

Figure 35.1: Lorentzian line shape for the energy radiated per unit frequency
range.

We wish to compare this radiated energy to the original energy of the oscillator.
Since the latter is given by

^initial — ’ (35.54)

the radiated energy, (35.52), can be rewritten as

£ ra d = i n i t i a l ( |
jL d (35.55)
m e3 7

Assuming that the Lorentzian peak adequately accounts for the energy radiated
and that there are no other forms of energy dissipation, we learn, from the
conservation of energy,
2
7
3 0
2
* (35.56)

Since this is the same result found in Problem 32.2 where we calculated the
power radiated by the oscillator and identified 7 from

P = - E = energy of oscillator, (35.57)

we conclude that (35.51) is an adequate representation of the energy spectrum.

Besides thus demonstrating that most of the radiated energy is contained within
the peak, we have also established self-consistency in that the damping of the os­
cillator is shown to arise from the reaction of the radiation back on the radiating
system.
3 5 .4 . P R O B L E M S F O R C H A P T E R 35 383

35.4 Problems for Chapter 35

1. Derive the expression (35.32) for the spectral distribution directly from
the power spectrum ^ ( t ) given in Problem 32.4.
2. Suppose 7 is not small so that (35.45) must be solved exactly. Calculate
d E /d u j , E ra,a without approximation and compare the latter to (35.52).

3. Repeat the analysis in Section 35.1 in the Coulomb gauge. In particular

show that the asymptotic vector potential is
e ‘iior/c ^
A (r,w )-------------[j(k,w) “ nc/i(k,w)].
r c

Using current conservation show that

e iu>r/c j
A(r,a;) ~ --------- n x (jx n ).
r c

From this derive the formulas for B and E, (35.21) and (35.24).
4. Derive a formula for the collisional broadening of spectral lines by com­
puting the spectrum of a weakly damped oscillator that vibrates freely for
only a finite time T. Assume that the probability of an oscillation time in
excess of a particular value decreases exponentially,

and perform a statistical average.

Chapter 36

Power Spectrum and

Cerenkov Radiation

36.1 Macroscopic Power Spectrum

In the previous chapter, we derived a general expression for the spectral dis­
tribution, (35.32), which accounts for the radiation from all times, t —* —oo to
t —* -f o o . Here we wish to obtain a generalization applicable to a limited epoch.
In so doing, we must note that time and frequency are complementary; that is,
a time interval of many periods is required in order to identify a corresponding
frequency. We first rewrite (35.32) in the space-time form:

d2E
dudCl
= ~ 2c3 jnX J dt e,wij(k,<) • JnX J dt' el(vt‘j(k,i') , (36.1)

and focus our attention on the part involving time integrations:

J dtdt'e Íü;(í <;\j(k,i)* j(k, <')

= j d T d r e - iuT3 ( k , T + T/2)*i(k,T-T/2), (36.2)

where we have introduced the average time and the time difference,

T= i ( t -M7), t = t — t\ d td t' = d T d r . (36.3)

¿i

From the exponential structure of (36.2), we infer that the important range of
r that contributes to the integral is of order 1 /u;, thus setting the time scale
for the emission of radiation. This microscopic time scale may be much smaller
than macroscopic time intervals; for example, for visible light, r ~ 1 0 “ 15 sec.
The time T is then interpreted as the average (macroscopic) time of emission,

385
386 C H A P T E R 36. P O W E R S P E C T R U M A N D C E R E N K O V R A D IA T IO N

which can be specified only to within a time of order r. Substituting (36.2) into
(36.1), we write
d2 E _ [ d2P ( T )
(36.4)
dojdD J dcudO, ’
from which we infer the power spectrum at time T,

£
d2P ( T ) _
dr e“ nxlj(k,T + r/2)« n x £ j(k ,T —r / 2 )
dcodSl 47r2c
(36.5)
or, alternatively, using the second form of (35.32),

d2P ( T ) ÜJ*
r dr e " , w ( ij(k , T + r/2)* . lj(k , T - r/2)
dwdQ, 47T2 C J-o o l c c

- p ( k , r + r /2 ) V ( k ,T - r / 2 ) | . (36.6)

36.2 Cerenkov Radiation

As an application of (36.6), we consider the radiation produced by a charged
particle moving with constant velocity v, for which the charge and current
densities are

p ( r ,t ) = e 6 (r -v t),

j(r,i) = e v S (r - v<). (36.7)

The Fourier transforms in (36.6) are trivially evaluated:

f ( d r ) e ±iujn *r/ c / ^ ^ T^ \
e ± i u n • v ( T ± r /2 ) /c
(36.8)
J {dv) Uj(r,T±r/2)J

When these are substituted into (36.6), the T dependence disappears, as ex­
pected, and we obtain

d2P e¿u;n • v r /c
d r e - iuTe 2
dujdCl

uP_
4ir2c
e2 ( ^ - l ) J°° dr e-*u,T(1-n ’ v/ c)
u>2e 2
(^2 _ 1 ) 2tt5 ( u> ( 1 - ^ cos0)) , (36.9)
4r2c

where 0 is the angle between the direction of observation, n, and the velocity of
the particle, v. The 6 function implies that there is no radiation, since
v
—cos 6 < 1. (36.10)
c
3 6 .2 . C E R E N K O V R A D IA T IO N 387

This is the familiar result that a charged particle moving with a constant velocity
in vacuum does not radiate.
However, if it were possible that

- > 1, (36.11)
c
the argument of the delta function could vanish, and radiation would be emitted
by the charged particle. Is there any way of effectively satisfying (36.11)? In a
medium, light can move with a speed, c', less than c, and correspondingly the
speed of a particle can be greater than c'. Now does the particle radiate? Recall
that the macroscopic Maxwell’s equations, (4.60), for a medium with dielectric
constant e and magnetic permeability //, can be put into vacuum form (1.65),
by the redefinitions (recall Section 7.2)

E' = V^E, H' = y/J IH , c' = -£ = , ft = -L>, J' = 4=J. (36.12)

y/ejl Ve V6

Therefore, the power radiated when a charged particle is moving with constant
velocity in a nonmagnetic medium {¡i = 1 ) of index of refraction n = y/e can be
obtained immediately from (36.9) by the substitutions e e/ n and c —* c/n:

(i)! - 0 2,5 (“ 0 - (Í7Í) c“ e) ) ■ (3613)

Thus, indeed there is radiation if the condition
TIV
— cos 0 = 1, (36.14)
c
or
cos 0 ~ —— < 1, (36.15)
nv
is satisfied. Here we see that, for a charged particle moving with a constant
velocity inside a medium characterized by an index of refraction n > 1 , electro­
magnetic radiation can be emitted if the criterion

V > - (36.16)
n
is satisfied. Such a medium can be easily found for fast particles. The radiation
is emitted on a cone described by (36.15) [see Fig. 36.1], and because of its unique
characteristics, is especially suited for determining the velocities of relativistic
charged particles. This phenomenon is called Cerenkov radiation. We emphasize
that the condition (36.15) can only be satisfied when n > 1. Because the index
of refraction depends on frequency, that is, media are dispersive, this means that
the condition (36.16) can only hold for a finite range of frequencies (typically, in
the optical region). Moreover, this dispersion implies that different frequencies
are emitted at different angles. Cerenkov radiation is commonly seen in water­
moderated nuclear reactors as blue light surrounding the core.
388 C H A P T E R 36. P O W E R S P E C T R U M A N D C E R E N K O V R A D IA T IO N

Figure 36.1: Cone of Cerenkov light produced at a definite wavelength by a

particle traveling with a velocity v, v greater than the speed of light in the
medium.

Historically, this radiation was first observed by Marie Curie in 1910, studied
deliberately by L. Mallet in the 1920s, but only definitively explored by P. A.
Cerenkov from 1934 through 1938. The theoretical explanation was given by
I. M. Frank and I. Tamm in 1937. For a complete account of the history and
application of the Cerenkov effect through the 1950s see J. V. Jelley, C eren k ov
Radiation and its A p p lica tion s , Pergamon, New York, 1958.
The frequency spectrum of the radiated power can be obtained by integrating
(36.13) over all angles,

±
d = f da*£-
du J dtodCl

2 e 2 n 2v 2
nc c2 = H )
c2 \ C
"2~;”x~2 . if »(w) > v
n 2(u))v¿ J
(36.17)

and the total radiated power is

dE
P
dt

(36.18)

where E is the energy of the particle. In practice, it is more convenient to

consider the energy lost per unit distance traveled by the particle, since this is
what can be directly measured by a Cerenkov counter:

dE_
(36.19)
dz

In (36.18) and (36.19) it is understood that the u> integration extends only over
the range where n(u>) > c / v . Finally we note that detection is a quantum
3 6 .3 . P R O B L E M S F O R C H A P T E R 36 389

process, involving photons. The energy of a photon of frequency v = u>/27r

is h v — h u , where h = 2irh is Planck’s constant. Therefore, the number of
photons emitted per unit length is

(36.20)

where the fine structure constant a is defined by

_ e2 1
(36.21)
h e* 137*
To obtain an order of magnitude estimate, we suppose that over a certain range
of cv, Au)j the 1 —c2/ n 2v 2 factor is of order 1:
dN Au> a 1
------ ot------ ~ . (36.22)
dz c A 137V
where we have noted that, typically, the range of wavelengths is of the same
order of magnitude as the wavelengths themselves. Equation (36.22) implies
that, roughly speaking, in a distance of 137 wavelengths, one photon is emitted.
In the visible spectrum, where A ~ lO” 5 cm, about 103 photons/cm are emitted.
(A more accurate estimate is 102 photons/cm.)

36.3 Problems for Chapter 36

1. Cerenkov light of a given wavelength is emitted on a cone of half-angle 0C.
Show that for small 6 C,

where ¡3c is the velocity of the particle moving in a medium with index of
refraction n.
2. Show that the number of photons produced per unit path length of a
particle with charge Z e and per unit energy interval of the photons, d E }
is
d2N _ a Z 2 . 2 _ a 2Z 2 f _ 1 \
dEdz h e Sm c romec2 \ f32 n 2( E ) ) ’

where the classical electron radius r0 = e 2J m ec2 . For Z = 1, show that

d2N
-- j - « 370 sin2 9c( E ) e V 1 cm'"1.
dEdz

Equivalently, show that

d2N 2'w a Z 2 ( 1 \
dzd\~ A2 V1 $ 2n \ A)/*
390 C H A P T E R 36. P O W E R S P E C T R U M A N D C E R E N K O V R A D IA T IO N

3. Cerenkov detectors are often used in high-energy experiments to distin­

guish particles. Modern “ring-imaging” Cerenkov detectors can, for ex­
ample, distinguish electrons from pions with momenta below 5 Gev/c.
Discuss why such a detector cannot discriminate between particles of dif­
ferent mass if the momenta are too high, and make an estimate of the
discrimination limit for e/ 7r, 7r / K , K / p , assuming sin2 0C can be measured
to 0 .1 % accuracy.
4. An electric charge e moves faster than the speed of light in a medium
characterized by e = 1, /¿ > 1. Find the energy radiated per unit length.
Repeat for a magnetic charge g and a medium with e > 1, p = 1.
5. (a) Show that the power emitted at macroscopic time T at the frequency
co can be expressed as

du
=
*cj_„ [ (dr)(dr^
J'
p - p
A ' “ |r-r'|
I j ( t ,r + i ) . I j ( , , r _ : ) _ P (r , r + i ) , ( , , r - i ) -

(b) If c/ lo is large compared to a typical length characterizing the source,

we can expand the above sine function. Thus derive the dipole radi­
ation formula (32.30). [As in (33.8)—(33.10), you can omit total time
derivatives.]
(c) For a system with d = 0, derive the next leading approximations for
the radiated power:

Magnetic dipole: = --r-(/i ) 2 [see (32.38)],

2 1
P
3 có
Electric quadruple: P = — ^

6. Consider Maxwell’s equations with both electric (pe,j e) and magnetic

{ p m ) 5m ) charges. Derive second order differential equations for E and
B. Show that
E — j-—A e V X A m,
c at
__ _ . 1 d . __ .
B — ^m TTTAm -f- V X A e ,
c at
and exhibit the differential equations for these potentials in the Lorentz
gauge, and in the radiation gauge.
7. Solve for the above potentials in some gauge, and find the asymptotic radi­
ation field. Now what is the relationship between E and B? Construct the
spectral-angular distribution of the radiated power. How does it change
when electric { £ ) and magnetic { M ) quantities are redefined according to
£ —►8 cos <j> + M sin <¡>) M —* M cos $ —£ sin <j>?
Chapter 37

Constant Acceleration and

Impulsive Scattering

In this chapter we will discuss radiation by a uniformly accelerated particle,

and by one which undergoes a sudden change in velocity, an inpulse. Although
it might, at first glance, appear that these two extreme situations are rather
unrelated, we will see, particularly in the problems, that the considerations for
radiation are closely connected.

37.1 Radiation by a Uniformly Accelerated Par­

ticle
Next, let us examine the characteristics of the radiation emitted by a uniformly
accelerated charged particle. For simplicity we will assume that the particle is
nonrelativistic (which remains valid only for a finite length of time), but the con­
clusions we will draw are often thought to be independent of that simplification.
[See Problem 37.3.] The equations of motion of such a particle are

r(i) = t 2, (37.1)
v(t) = at, (37.2)

in terms of which we construct the current density

j(r,i) = ev(i)«5(r - r(t)). (37.3)

The power spectrum, at time T, (36.5), becomes

391
392 C H A P T E R 37. C O N S T A N T A C C E L E R A T IO N A N D IM P U L S E

= ^ ^ 3 (n X a ) 2 J dT exP -* WT ( l ~ *V(T) ) ] P 12 “ (T/ 2)2]

= lirc^ n><a^ - n *v/c)) + j£ "(w (l - n -v /c ))| ,

(37.4)

where we have identified the second derivative of the ¿-function according to

f d r r 2e ~ iTX = -2 ir 6 "( x ) . (37.5)

J —OO

For u) > 0, the argument of the delta functions in (37.4) never vanishes,

^ 2 = °, for w > 0 . (37.6)

The power, (37.4), is nonzero only for w = 0, which corresponds to static fields.
Such fields do not correspond to radiation, and a uniformly accelerated charged
particle does not radiate. On the other hand, the total radiated power can be
computed from the Larmor formula, (32.25):

p=lSw2 =^ a2^0’ <37-7)

which seems to indicate that a uniformly accelerated charge does radiate. How
can we reconcile the above two seemingly contradictory results, (37.6) and
(37.7)? Actually there is no logical contradiction, because a constant power
radiated would appear to correspond to a power spectrum which consists only
of zero frequency, which, to reiterate, does not represent radiation. The Larmor
formula thus is not applicable to this situation. (But, see later.)
It is instructive to consider this radiation process from the source point
of view, discussed in Section 33.2. Remember that, there, we obtained the
following expression for the rate at which charges in a small, nonrelativistic
system do work on the electromagnetic field [cf. (33.9)],

-/( * ) <37'8>
where we identified E as the electromagnetic field energy while the remaining
term is the power radiated. To obtain the Larmor formula, we neglected a
further total time derivative. If we now use the above expression for the power
radiated,

p = - h i '~á = - \ ‘i v ' ii' <37'9)

since the dipole moment of a point charge is

d = er, (37.10)
3 7 .1 . R A D IA T IO N B Y A U N IF O R M L Y A C C E L E R A T E D P A R T IC L E 393

E = constant

v (0) v(T )

t = 0 t = T

Figure 37.1: Simple model of a linear accelerator.

we immediately see that there is no radiation produced by a uniformly acceler­

ated charged particle.
Having persuaded you that there is no radiation produced by uniform accel­
eration, we now enlarge the picture and recognize that radiation is associated
with the whole history of a process, not just a particular period of time. What
we have assumed so far is a situation, in which, for all time, the particle un­
dergoes uniform acceleration, which, if nothing else, violates our nonrelativistic
treatment since eventually the velocity of the particle will become comparable
with the speed of light. The essential point to recognize is that uniform accel­
eration for all time is an idealization of the realistic situation in which uniform
acceleration is only experienced for a finite time interval.
As an example, we consider the acceleration of a charged particle by a linear
accelerator. Suppose originally the particle moves with a small constant speed
v(0). At a particular time t = 0, it enters a region where a uniform electric field
is applied, causing the particle to undergo a constant acceleration a. After a
period of time T, the particle is ejected from the accelerator with a velocity
v(T) = v(0) + aT, (37.11)
which is still assumed to be small compared to the speed of light. This acceler-
ation process is represented in Fig. 37.1. The spectral distribution of the energy
radiated by this accelerated particle is given by (35.40), which can be rewritten
in the form
dJE _ 2 e M JL_
IvH I2, (37.12)
duj 3 C3 7Tu>2
since, from (35.41),
v(u>) = —¿u>v(u>), (37.13)
and
v(u>) = —iw v (u j). (37.14)
The quantity v(u>) can be calculated by noting that since the acceleration is
discontinuous at t = 0 and t = T, the derivative of a has an impulse at these
two times:

v(w) dt eiwtv ( t )
394 C H A P T E R 37. C O N S T A N T A C C E L E R A T IO N A N D IM P U L S E

Figure 37.2: Spectral distribution of energy radiated by accelerator in Fig. 37.1.

/
OO

dt e‘wt[a<5(í) —a<5(í - T )]

•CO

= a (1 - etVr)

= —2 ¿a eMl'T/2 sin . (37.15)

The square of the magnitude of (37.15) is

|v(cj)p = 4u siu
1
iu T , (37.16)
¿4

so that the spectral distribution of radiated energy is

dE 2 e2 1 4a2 . 2 1
T~ = * - 3 ----- 2~sm 2o (37.17)
du> 3 cd 7T u)1

which is nonzero in general. Of course, there is no contradiction between this

result and (37.6), since we are no longer talking about uniform acceleration for
all time. We also observe that the contributions from the changes of acceleration
at t = 0 and t = T interfere with each other. This is incompatible with locality
in time, that is, we cannot specify at which time the radiation is emitted.
To see what are the important frequencies being radiated, we plot the fre­
quency distribution, (37.17), as a function of u in Fig. 37.2. The significant
range of frequencies is set by the time of acceleration, T,

u - 1 /T, (37.18)
37.1. RADIATION B Y A UNIFORMLY ACCELERATED PARTICLE 395

that is, as T becomes longer, dE/du> is more and more concentrated at low values
of a;. In the limit when T —►oo, we recover the situation of uniform acceleration
discussed above. Moreover, we see again that the emission spectrum is not time
analyzable, since the relation (37.18) implies that it takes a time of the order
of the whole process to determine the frequency. The total energy radiated is
obtained from (37.17) by integrating over all frequencies,

. dE 2cM a2 f du . 2 loT
Er&d —r sin
JQ U du) 3 C 3 7T Jo (jü¿ 2
2 e2 2 /2 f°° dx . 2
3c^a T \7T Jo ^ Smi
-2 e- a * T (37.19)
3e3 ’

where we have used the integral (34.18). From (37.19), we observe that the
total energy radiated per unit time is

E yeló. 2 2
(37.20)
3^ a

which is precisely that obtained from the Larmor formula, (37.7), which refers
only to uniform acceleration.
Finally let us calculate, from the source point of view, the radiated power
using (37.9),

2 e2 ..
P
3 ^ V *V

| $ a .[v (0 ) 6 (t)-v (T )6 (t-T )], (37.21)

which incorrectly attributes the radiation entirely to the beginning and the end
o f the acceleration process. However, the total radiated energy,

f?rad = Í d tp
J —oo
2 e2

2 e2 2
(37.22)
= 3 c3 *1 T ’

is the same as (37.19).

Thus we have seen that the question of whether there is radiation produced
by a uniformly accelerated charge is only properly answered by taking into
account the beginning and ending of the acceleration. Facetiously, we may
say that a uniformly accelerated charge radiates because it is not uniformly
accelerated.
396 CHAPTER 37. CONSTANT ACCELERATION AND IMPULSE

t = 0

Figure 37.3: Impulsive scattering.

37.2 Radiation by Impulsive Scattering

As an idealization of a scattering process, consider one in which a charged par­
ticle abruptly changes its velocity from a constant value V2 to another constant
value v i. (See Fig. 37.3.) We will calculate the radiation produced by this
accelerated particle, bearing in mind that this description can be realistic only
for radiation of sufficiently low frequencies since long wavelengths cannot probe
the detailed character of the particle motion. The charge and current densities
before and after the deflection act, which is assumed to take place at t = 0, are
p = e£(r - \2t), j = ev2<S(r —v 2f), for t < 0, (37.23)

p = e6(r - vi<), (37.24)

o
«2

A
>
>

^1

U
II

The spectral distribution for the radiated energy can be computed by substi­
tuting (37.23) and (37.24) into (35.32), where we encounter the integral

j(k , u ) = J (dr) dt eiwte~iwn •*/«j(r> t)

rOO
= ev2 / d u ,'a,t(1“ n , V c>
/ —oo «-/ 0
f Vi v2
= iL(
u \^1 —n ♦Vj/ c 1 —n •v 2/ c /
) . (37.25)

Here we have used the effective evaluations

POO •

L d“ ,u - v (37.26)

L Jt‘M=- ly (37.27)

since physical quantities cannot depend on what transpired at infinitely remote

times. As a consistency check of (37.26) and (37.27), we can calculate the
corresponding integral on p directly,

p(k, w) = *— ( ------- ------ 7- - ------ ------ ] , (37.28)

w \ l —n>\x/c \—vl'\ 2/cJ '
7 -
v
37.2. RADIATION B Y IMPULSIVE SCATTERING 397

or by means of the charge conservation condition, (35.22),

/>(k,w) = — •j(k, w). (37.29)

UJ

Indeed, if we take the scalar product of (37.25) with n /c , we have (37.28):

.e f jn .v , _ jn .v 2 \ = .c ( 1___________ 1 _ \
w \1 —n^Vj /c 1 —n * v 2/ c y ^l —n - V j / c 1 —n * v 2/ c / ‘
(37.30)
We now immediately obtain the spectral distribution from (35.32),

d2£ / v i____________ V2 \
(37.31)
d u ;d f2 47r2 c 3 n \1 —n -V j/ c 1 —n » v 2/ c /

Since (37.31) is independent of the frequency, the implied total radiated energy
is unbounded. This unphysical result is due to the idealization that the scat­
tering occurs instantaneously, that is, our assumption that the particle changes
its velocity abruptly. Realistically, this change occurs over some finite period of
time, T, so that our result, (37.31) holds only when u <C 1/T . In the nonrela-
tivistic limit, where ¡vi^/c] <C 1, (37.31) reduces to

d2E ____
-|nx(vj —v2)| = ^ 2^ ( v i - v 2) sin 0, (37.32)
dudQ ~ 47T2 c :

where 9 is the angle between vi —V2 and n. Integrating this over all angles, we
obtain the energy radiated at the frequency u

dE_ 2e^
•(vi - V 2) 2, (37.33)
du> 3 7rc:

which can also be easily obtained from (35.40), since the Fourier transform of
the derivative here is simply

/
OO poo
dt eZUJt\(t) = / dt elwt(vi — V2 )^(t) = v i —V2 - (37.34)
-oo J—oo
Because either denominator in (37.31) has the structure

1 —- n . v¿ = 1 — — cos 9 i , (37.35)
c c

where is either Vi or V2 , and $i is the angle between n and v¿, we see that the
radiation in the ultrarelativistic limit (|v¿| ~ c) is preferentially emitted near
the direction of the velocity of the particle (the forward direction), either before
or after the scattering act. On the other hand, this behavior is softened by
the numerator factor |nXv¿| = t;¿sin which forbids radiation in the exactly
forward direction, = 0. In either region of significant radiation, characterized
398 CHAPTER 37. CONSTANT ACCELERATION AND IMPULSE

by Oí -C 1, only one term in (37.31) makes a major contribution (in the following
we drop the subscript ;):

/ n X v /c \2 J . ___________

V I - fe o s fe o s 6») 2
462

( i- f + f ) 2 ~ ( i- £ + * 2) 2
f 4/02, if y/1 - (v/c)2 <C 0 < 1,
(37.36)
X 0, if 0 = 0 .

Therefore, the maximum intensity occurs at

* = V I - («/<02 = ^ (37-37)

(which we have implicitly assumed to be small compared to the scattering angle),

and most of the radiation is emitted in a small angular range near this angle. As
the particle moves faster, the peaking of the radiation is more pronounced. This
is a fundamental difference between radiation produced by relativistic (v ~ c)
and nonrelativistic (v <C c) particles. We will see this again as we now turn to
a discussion of synchrotron radiation.

37.3 Problems for Chapter 37

1. By considering the Lienard-Wiechert fields found in Problem 31.5, demon­
strate that no radiation zone is formed by a uniformly accelerated particle.

2. Consider a particle that undergoes nonuniform motion for only a finite

period of time, t\ < t < Otherwise, we have

_ Í R i + v i(i - ti), t < <1 ,

^' \ R 2 + v 2(t - < 2), t > h -

(a) Write j(k,u>) and p(k,a;) as integrals over the time interval t\ < t <
t2-
(b) Show that charge conservation, (35.22), is satisfied.
(c) Derive the general formula for dE/dil as an integral over the time
interval t\ < t < ¿2 ,

d E _ _ _ f _ y*2 , 1 id . n X v (Q V
dCl 47re3 Jtl 1 — n •v (t )/c \di 1 — n •v (t)/ c)

This can be viewed as a generalization of (32.23).

37.3. PROBLEMS FOR CHAPTER 37 399

(d) Integrate over Q in the above general formula for dE/dil and thereby
derive the general formula for the total energy radiated,

where /3 = v(t)/c. This can be viewed as a generalization of (32.25).

3. Apply the results of the previous problem to treat the radiation by a rel­
ativistic, uniformly accelerated1 particle, undergoing so-called hyperbolic
motion, for a finite time interval,
_ cr . - cr ^
z — a cosh — , ct — asinh — , —tq < r < to.
a a
(a) By considering short times (or r « 0), make connection with the
nonrelativistic formulas and identify the parameter a.
(b) Show that
-Í® . = i l L
du> dil n #2_ 0 tt2c i c V c/J
in the Tq —►oo limit. Use the second integral representation in (18.18)
and note that
K t (x) = -K '0(z)
[cf- (18.71)].
(c) For 6 / 0, 7T (n *z = cos#) and tq —> oo, show that

dE_ _ 3e2 1
dO, 32a sin3 6

(d) Show that the total energy radiated is

jp _ 4 g2 Pm

where /3m is the maximum value of v/c (= |v(±ro)|/c).

(e) Comment on the following quotation from Pauli:
Hyperbolic motion thus constitutes a special case, for which
there is no formation of a wave zone nor any corresponding
radiation. (Radiation, on the other hand, does occur when
two uniform, rectilinear, motions are connected by a “por­
tion” of hyperbolic motion.) [W. Pauli, Theory of Relativity
(Pergamon, Oxford, 1958), p. 93]
1According to Pauli, cited in the following, “In relativistic kinematics one will naturally
describe as ‘uniformly accelerated’ a motion for which the acceleration in a system, moving
with the medium or particle, is always of the same magnitude. The system is a different one
at each instant; for one and the same Galilean system, the acceleration of such a motion is
not constant in time.”
400 CHAPTER 37. CONSTANT ACCELERATION AND IMPULSE

4. Consider a particle undergoing a constant acceleration g for a finite time,

—¿o < t < to, during which time its velocity changes from — (= —0 ¿o) to
-j-v 0 - [Here, the “constant acceleration” means dPv/dkt 2 = constant, so the
hyperbolic motion described in Problem 3 is not being considered. Hence,
we require that \g\to < c.]

(a) Let g be equal to the acceleration at t = 0 for Problem 3 and find a

relation between g and a.
(b) By means of the result of Problem 2(d), calculate the total energy
radiated. Express it as a function of (3m ( = v q / c ) .
(c) Show that the nonrelativistic limit of this result agrees with the cor­
responding limit of Problem 3(d).
Chapter 38

Synchrotron Radiation I

A charged particle, moving in a circular orbit, undergoes centripetal accelera­

tion and therefore radiates electromagnetic energy. The physical circumstances
under which this process occurs arise in synchrotrons, in which charged particles
are guided in a circle by external magnetic fields. In this and the following two
chapters, we will explore the characteristics of this synchrotron radiation. We
begin by considering some kinematics.

38.1 Motion of a Charged Particle in a Homo­

geneous Magnetic Field
If E is the energy of the charged particle and p is its momentum, then the
(relativistic) equations of motion of the particle in a magnetic field B are
dp e ^
= -vxB, (38.1)
at c
dE A
(38.2)
at ~ ° ’
where the momentum is related to the velocity, v, by
__ rav _ E
p ™ ^ (38.3)

and the energy is

me 2
E = (38.4)

Since the magnetic force is always perpendicular to the direction of motion,

no work is done on the particle, and consequently the energy of the particle is
conserved, as stated by (38.2). This fact, together with (38.3), enables us to
rewrite (38.1) in the form
E dv e _
-v x B , (38.5)
c 2 dt c

401
402 CHAPTER 38. SYNCHROTRON RADIATION I

Figure 38.1: Precession of velocity vector around a constant magnetic field.

or
(38.6)

This implies a constant deflection of the velocity vector, that is, v precesses
with angular velocity u?o>
6C
« o = —-g B , (38.7)

about the direction of ( —e x ) the magnetic field. (See Fig. 38.1.) The angular
speed of this precession is the Larmor frequency,

u>o = (38.8)

which reduces, in the nonrelativistic limit, to the cyclotron frequency,

(38.9)
me ’

The component of the velocity parallel to the magnetic field, according to

(38.6), is constant. In practice, we constrain the motion such that v is per­
pendicular to B. Then the particle moves with angular speed u>o in a circle of
radius R ,
v _ /3E
(38.10)
u>o \e\B'
The relation between the momentum of the particle and the radius of the circular
orbit,
BR, (38.11)
cz c
supplies us with a practical method of measuring the momentum of a relativistic
particle. When the information thus obtained is coupled with that derived
38.2. SPECTRUM OF SYNCHROTRON RADIATION 403

Figure 38.2: Diagram of charged particle moving in a circle in a magnetic field.

from a Cerenkov counter, which measures the speed of the particle according to
(36.14), we can determine the mass of the particle:

m — ------------ p (38.12)

38.2 Spectrum of Synchrotron Radiation

We now proceed to calculate the radiation emitted by a charged particle moving
in a circle. For a point particle, the charge and current densities are given by

/?(r, t) = e 6 ( r - r ( t ) ) y (38.13)
j ( r ,t ) = e v ( t ) 6 ( r (38.14)

where r(t) is the position vector of the particle at time t , and v(tf) is its velocity.
Substituting (38.13) and (38.14) into (36.6), we obtain the spectrum of the
power, emitted at time T, into the element of solid angle in the direction n,

/
d?P(T) _ w V
oo

dr e~ ’ lv (T + r /2 ).v (T -r /2 )-l
dudil 4tt2c
-OO
x exp (38.15)
{ ¿ ^ n . [r(T + r/2) - r(T - r/2)]} .
This description in terms of an average macroscopic time can be important, for
then it is possible to consider the effect of a slow alteration in the parameters
describing the motion.
It is convenient to choose the coordinate system such that the particle is
moving in a circle of radius R about the origin in the xy plane, and the magnetic
field B is directed along the z direction. Also, without loss of generality, we
choose the observation direction, n, to lie in the xz plane, making an angle 6
with the -j-z axis, as shown in Fig. 38.2. Then we have, for a convenient choice
of initial conditions,

n = (sin 0 , 0 , cos#), (38.16)

x(t) = Rcosu>ot} (38.17)
404 CHAPTER 38. SYNCHROTRON RADIATION I

y(i) = Rsinu)0t , (38.18)

z(t) = 0, (38.19)

where wo and Ü are given by (38.8) and (38.10), respectively, The corresponding
velocity is given by

Vgf(t) = —i;sinu>of, (38.20)

vy(t) = v coseno (38.21)
vz(t) = 0 , (38.22)

where the speed v satisfies

v — Rw 0. (38.23)
Using these explicit representations for r (t) and v(t), we may simplify the inte-
grand of (38.15) by means of the following:

~ v ( T + r /2 ) •v (T — r /2 ) = /?2 cos o>0t (38.24)

(a fact which is apparent geometrically), and

n * r ( T ± r / 2 ) = sin6 x ( T ± r/2) = RsinOcosu>o(T ± r /2 ), (38.25)

C± t(u ;/c )n • r (T ± r /2 ) = e ± « ( W c)^ sin 0 cosu/o( T ± t /2 )
oo
= Y j ( ± i)me±imWo(-T±T/2)Jm ( - f l s i n f l ) . (38.26)
m = —oo

In the last equation, we have used the generating function for the Bessel func­
tions of integer order, (16.28),
oo
eizcos* = imeim*Jm(z). (38.27)

With these evaluations, we obtain the power spectrum, (38.15), in the form

d2P (T ) w2 e2 [°° , _ iurra2

= 4^~c L - O J
O
Tt
OO OO

x J2 S imeimUo(T+T/2)Jm ( - f l s i n f l )
m = —oo m ,— — oo

e - im'Wo(-T- TM j m' (^ fls in fl) . (38.28)

Now we recall that the emission time T represents an average over many periods
of the motion, since many oscillations are required to identify a frequency. Here,
it is sufficient to consider the average of (38.28) over one period, by using the
relation
{e¿(m -m > oT )oneperiod = 5mmi (38.29)
38.2. SPECTRUM OF SYNCHROTRON RADIATION 405

After this averaging procedure, (38.28) becomes

d2P 'u
' 2° 2 r°°
2e2 r
diüdíí = COSa' 0 T ' 11
OO
x £ eim“°r [/m 0 fls in 0 )] . (38.30)

Incidentally, we remark that this result, (38.30), can also be obtained by com­
bining the two exponentials directly as follows:
i(u i/ c )n ♦[r (T + r /2 ) - r ( T - t/ 2)]^
(e one period
— / /c)Rsin 0 ( - 2 sin u>0T sin ^ o r ) \
\c /one period
f2 u R . . u 0r
= Jo ------ sin 6 sin — —
\ c 2 )
gimtd0r (38.31)
= £ J m ^ s in ^ l
m = —oo ' '

Here we have used the integral representation of J0, (16.4),

j°(a:)=C (38.32)

and the addition theorem for the Bessel functions, (16.69),

oo
•fo + /»'2 - 2/9/9' cos(<¿ - «¿')) = X) J™(kp)eim4'Jm(kp,) e - im4’',

(38.33)
with p = / / = R, k = (u>/c) sin 0 , and </>— 0 ' = u/0t .
Exploiting the exponential representation for cosu >0 t ,

cosw0r = - [e®"oT + e ,a,°T] , (38.34)

we can easily perform the r integration in (38.30) by means of the integral

f dT e~tUTetmu<sT = 2 ir6(u> - mw0).

Consequently, the spectrum is discrete in that only harmonics of the Larmor

(38.35)

frequency, (38.8), are radiated. That is, the emitted frequencies are multiples
o f u>o, u> = mw0, where the integer m is positive since u > 0. Thus the power
spectrum can be written as a sum of the contributions of each harmonic,

d2P (T ) ^ £/ ,d P m(T)
(38.36)
dudO,
m= 1
406 CHAPTER 38. SYNCHROTRON RADIATION I

where the power radiated into the mth harmonic is

dPm m2u le 2 \0i / t 2 2 v a'

(38.37)
dfi “ 2ttc 2 1 m+1+ m- lj m
the argument of the Bessel functions being

---- sin 0 = mP sin 6. (38.38)

c
In (38.36) the T dependence could arise from any slow variation in the parame­
ters ujq and /?. We may further simplify (38.37) by using the following recurrence
relations for Bessel’s functions, found in Problem 16.4,

Jfn-l(z ) — 2 J m (¿), (38.39)

Orn
+ Jm+l(z) = ---- Jm(z), (38.40)
z
which imply that

\ + (J— O '] - = Vm? + (38.41)

Therefore, the angular distribution for the power radiated into the mth harmonic
is
dPm Jm(m/3 sin 0) 1 2
(m/? sin 0)] + (38.42)
dQ ¡3 tan 9

38.3 Total Power Emitted into the mth Har­

monic
To obtain the total power radiated into the mth harmonic, we could integrate
(38.42) over all angles. However, we find it simpler to return to the general
expression (38.15) and perform the integration over the angles first, since there
all the angular dependence is in the exponential factor. If we define s to be

s = r(T + r / 2 ) - r(T - r / 2 ), (38.43)

|s| = \fR? + R2 — 2B? cosc^o^ = 2/2| sin (38.44)

¿i
the angular integration we encounter in this way is

J dO = 2ttJ * sin XdX e¿(Wc)Ncosx:

,_sinw|s|/c
- 4’ ^ R 7 T
_ 27rcs i n ( g f f i s i n ^ )
(38.45)
wR sin
38.3. TOTAL POW ER EMITTED INTO THE M T H HARMONIC 407

The resulting frequency distribution of the radiated power is, from (38.15),

lo e“2 /’°° , s in (^ fts in ^ )

f dr e
dP iu,T,n*
cos 0 (/?2 w t — 1 ) -----V c. ? '
doj 27r R J — c
sin
02

R J_c
= — — r dre~WTf(w0T). (38.46)
~ 2 tt

The challenge now is to carry out the r integration. We observe that /(u>or) is
a periodic function,

/(w 0r) = / ( wqt + 2 n 7r), n = integer, (38.47)

so that it may be represented by a Fourier series,

oo
/(«or)= eimW° Tfm, (38.48)

where the Fourier coefficient f m is given by

/m = (38.49)

Inserting the representation (38.48) into (38.46), and carrying out the then
trivial r integration, (38.35), we obtain the spectrum of the radiated power:

dP x°°A
- muJo)Pm, (38.50)
m= 1
where the total power radiated in the mth harmonic is expressed as a Fourier
coefficient,

sin ^2 m/? sin ^

= ^mw0 J 7£~e *m4,(l32 cos<t> - 1)

e2 f'd * , 0 ,2 • a * ') sin ( 2 ^ sin f )

= -—mtüo I — cos m<f) I p — 1 — 2 p sin — 1 ----- ------- j ------ —.
R Jo n \ 2/ sin f
(38.51)

Not surprisingly, this can be rewritten in terms of Bessel functions. Starting

from the integral representation (16.30), equivalent to (38.27),

imJm( z ) = r d A e-im ,eizCost ( 3 8 .5 2)

J — 7T 2 tt

we make the substitution <j> <j>— 7r/2, leading to

pTT 11
Jm(z) = I — cos(zsin<^ — m<j>). (38.53)
Jo *
408 CHAPTER 38. SYNCHROTRON RADIATION I

Breaking this integral at n/2 and substituting <f>—*•ir —(j>for the range tt/2 to
we finally find
7T,

I*tt/2
Jm{z) = / — [cos(;zsin<£ — m<j>) + cos(^sin<^ -f m<f> — m 7r)]. (38.54)
Jo *
This supplies the following integral representation for the Bessel functions of
even order:
f n/2
J2m{z) = / — [cos(;zsin <¡) — 2 m<^) + cos(zsin <¡) -f 2 m<j))]
Jo ft
r d<j> ( . <f>\ ,
= / — cos ( zsm — 1cosrrnp, (38.55)
Jo ft \ */

where <j>—►^ in the second line. The integral in (38.51) can now be expressed
in terms of the derivative and the integral of J2m >which are represented by

t/ / \ r d * . * ( . <a
(38.56)
- -J ~ sm2 cosm(Psm l *sm 2 y ’
and
/■* rw ¿<t> sin ( * sin I)
/ dzJ2m{z) = / — cos m<¿-------r -r ------. (38.57)
7o Vo f sin §
Therefore, the total power radiated into the mth harmonic is

p2 m(3
Pm = 2^2mPm^) “ (1 — ft2)JOI dzJ2m(z) (38.58)

38.4 Total Radiated Power

To find the total radiated power, we could sum (38.58) over all m. However,
as before, it is much simpler to return to an earlier stage, this time to (38.46),
and first perform the frequency integration, before doing the r integration. An
intermediate form for the radiated power is

f°° , dP 1 f°° . dP
= / OW— = - / d u -r -
Jo du 2 du

-~ eiRL r
2 ir J _ 0
f3 2 COS U qT

sm 7 j U o t
— 1 1l í°°
- /
2 J-oo
1
du u e~lWT —
2*
p ^ fis in ^ _ e-i2 *R Sm

=R 4 Lr /?2 cos <f>— 1

d<¡>
sm •
S' \4> — 2/3sin — S’ ( <¡>+ 2j3 sin ^

(38.59)
38.4. TOTAL RADIATED POW ER 409

where we have made use o f the representation

/
OO J p CO
du> u e~iuX = i — / dw e~iuX = 2irtf'(A). (38.60)
•CO dX J — OO

Equation (38.59) can be evaluated in a straightforward manner. However, it is

simpler to combine the delta functions there by means of

2/3sin | 8* ^ -f 2/3 sin ^0 - 6' - 2/3 sin 0 j = J dX 8"{y), (38.61)

where
y = <j>+ 2A/?sin^,
(38.62)
so that we may rewrite the total power as

(38.63)
lit y -00 1 + A pcosf

Here the y integration can be performed by integrating by parts twice and noting
that the support of the delta function occurs at </>= 0 :

d2 / /32 cos </) — 1 1 d ¡32 cos <j) — 1

+ A /?co sf y 1 -f A/? cos 4 ^ \ 1 + A/? cos ^ d<j) 1 -f A/? cos ^ 0=0
y=o

1 d2 ( (32 cos </>— 1

(1 + A/?)2 d p \ l + A/?cos|
0= 0

£+¥&$■ <*«>
2
(1 + A/?)3
The remaining A integral can now be easily evaluated, yielding [see Problem
38.1]

/?2 —\
P = u>0/3
/ > J(T T W (- 2 + 8 1 + A/?
e2 2 /?3
= oTTw 0
3 i? ( I - / ? 2)2
2 e2 ~/ E
= (— 2 (38.65)
3 /£ Vme2
which is the exact result for the total radiated power.
In the nonrelativistic limit, /? <C 1, (38.65) reduces to the Larmor formula,
(32.25),

Pn .k . = z~jiUoP3

- i t (tt\ (38.66)
~ 3 c3 \dt ) ’
410 CHAPTER 38. SYNCHROTRON RADIATION I

because, for circular motion, the centripetal acceleration is

dv
= - = «*. (38.67)
dt

For a relativistic particle, the power radiated is larger by the factor (E /m c 2)4,
signifying that high energy charged particles moving in circular orbits emit
substantial synchrotron radiation. In one period, T = 27r/u>o, the total energy
radiated is

A£=pt=t ^G ¿)4' (38 68)

In practical units, if we express energy in electron-volts (eV), and the radius
of the accelerator in meters, then for /? « 1 , the energy loss per cycle by an
electron is
A £(k eV ) = 88.5^ , (38.69)

where
1 GeV = 106 keV = 109 eV. (38.70)
Inserting typical numbers for an electron synchrotron, R = 10 meters and E =
10 GeV, we find for the energy loss per cycle
A E = 88.5 MeV, (38.71)
which is quite substantial. For this reason electron synchrotrons are impractical
for energies greater than ~ 10 GeV. [Roughly twice this energy is radiated
from LEP, where E = 55 GeV and R = 4.25 km.] On the other hand, the
radiation from any existing or projected proton synchrotron is quite negligible,
being smaller, for the same energy, than electron synchrotron radiation by the
factor (me/m p )4 ~ 10“ 13. [For the LHC, where E — 7 TeV and R = 4.3 km,
the energy loss per cycle is only about 4 keV.]

38.5 Problems for Chapter 38

1. Carry out the A integration by which (38.65) is obtained from (38.64).

2. Provide an alternative derivation of the total power emitted in synchrotron

radiation, (38.65), by proceeding as follows: Starting from (38.15) let
n . ( r - r ' ) = -A | r -r '|

where
r = r(T + r / 2 ), r' = r ( T - r / 2 ) ,
thereby obtaining the frequency distribution
’2- 2 r°° ^ v (T H- t /2 ) •v (T — r /2 )
^ = ^ r dTe— (:
du 2 ir e J _ 0O \

xf d \ e - i(-uteW r~r'l,
38.5. PROBLEMS FOR CHAPTER 38 411

whence, by integration over w, the total power is obtained,

p= jrL ix -0 s" ( t + A
|r- r'|/c)-
Specialize this general result to circular motion, and let ujot = <^, to obtain
(38.63) and hence (38.65). (In this way, the second derivative of the 8
function emerges automatically.)

3. Verify the numerical result for the energy loss per cycle, (38.69). (1 esu
unit of potential = 300 volts— see Appendix A.)

4. What strength of magnetic field (in Tesla) is required to keep a relativistic

electron of energy E (in GeV) moving in a circle of radius R (in meters)?

5. An electron is raised to energy E (in GeV), while moving in a circle of

radius R (meters). Then the power input is cut off. Assuming that the
radius is maintained constant, how long (in milliseconds) will it take for
the energy to drop to half its maximum value?

6. Derive the total power radiated, (38.65), from the result of Problem 37.2(d).
Chapter 39

Synchrotron Radiation
II— Polarization

The polarization state of an electromagnetic wave is determined by the direction

of its electric field of which there are two independent possibilities, each normal
to the direction of propagation, n. The results obtained in the previous chapter
referred to the sum of powers radiated into both polarization states. We now
investigate the polarization characteristics of synchrotron radiation. One way in
which this information is practically useful lies in studying astrophysical objects;
for example, the Crab Nebula is inferred to emit synchrotron radiation, because
of the unique polarization characteristics of the latter. In order to calculate
the power radiated into each polarization state, we return to the basic formula
(36.5). However, we recall that the latter equation was obtained from (35.31)
in which energy flow was expressed in terms of |B|. It is desirable to shift the
emphasis from |B| to |E|, since it is the electric field that is measured when
polarization is determined. We do this by noting that since E and B are related
by (35.24), (36.5) can be equivalently written as
d2P (T )
dujdQ 47r2c f dr e~ nX ( n X - j ( k ,T + r / 2 ) * ) ]

-nX ( n X 1- j./( k ,T - r /2 ) (39.1)

where the terms in the square brackets are proportional to the electric field
strength.
To isolate the effect of each polarization state, we look at the two components
of E, in the plane perpendicular to n, separately. We choose the coordinate
system (which is naturally set by the synchrotron) as before, with n given
by (38.16). We define two polarization orientations as follows: For “parallel
polarization” the electric field is in the direction of e||, which lies in the orbital
plane and is perpendicular to n, that is, ey points in the + y direction,
e || = ( 0 , 1 , 0 ). (39.2)

413
414 CHAPTER 39. SYNCHROTRON RADIATION II— POLARIZATION

Figure 39.1: Polarization vectors for synchrotron radiation.

The “perpendicular polarization” vector, ex , is perpendicular to both e|| and

n, with e||, e x , and n forming a right-handed system of unit vectors as shown
in Fig. 39.1:
ex = (—cos 0,0, sin 0). (39.3)

It is sufficient to compute the radiation produced in a single polarization

state since the result of the previous section then supplies the power radiated
in the other state. It is simpler to consider the e|| polarization. The partial
intensity with this polarization arises from the product of y components of the
vectors in square brackets in (39.1). Since n lies in the xz plane, this component
is
[~ n x (n Xj)]y = U - n (n •j)]„ = j y , (39.4)

and the resulting contribution to the power spectrum radiated in the parallel
polarization state is

(Sr),, =4^ ')<*')e*'"'1)"

x ^ V (r >T, + r / 2 ) ^ y ( r' . 7 ’ - 'r / 2 ) . (39.5)

For synchrotron radiation, the current density is described by (38.14), so that

(39.5) reduces to

( d_P(Ty\ 0> e _ í ^ —¿WT t(w /c)n • [ r (T + r /2 ) —r ( T —t / 2 )]

\ d wdQJi W c J _ J Te

x ^2vy ( T + r / 2 ) v y ( T - T / 2 ) . (39.6)

We now can evaluate (39.6) by following closely the procedure given in the
previous chapter. Therefore, instead of the factor

¿ v ( T + t / 2) •v (T - t / 2) - \ = p 2 cos w0r - 1 (39.7)

39.1. PROBLEMS FOR CHAPTER 39 415

appearing in (38.15), we now encounter, because the position vector and velocity
are given by (38.17)—(38.19) and (38.20)-(38.22), respectively,

\ v y(T + r/2)vy (T - r /2 ) = 0 2 cosu 0(T + t /2 ) cosw 0(T - r /2 )

02
= — (cos 2u>oT + costa r). (39.8)

The time average of (39.6) involves not only (38.31) but also the new evaluation
[most easily done starting with (38.26)],

(cos 2u>oT e*(‘*'/e)n •M T + r ^ - K T - r ^ ) ] ^ ^

= - £ eimu°TJm+i ( ”^ s i n ^ ) Jm-i ( - ^ - s i n f l) . (39.9)

m——oo \ / \ /

Therefore, instead of the combination

00 2
(/?2 cosw0r - 1 ) ^ eim" ir [<7m ^ ñ s i n ^ j , (39.10)
m ——00

appearing in (38.30), we have here, for parallel polarization, the term

1 00
-/? 2 eima,oT[ - J m+1 Jm _ 1 + cosw ot J £ ]. (3 9 .1 1 )

The r integration now leads to the power radiated into the mth harmonic being
proportional to

y ( ■'” « - 7 „ +1

(39.12)
where we have used the recurrence relation (38.39). Thus, the two terms in
(38.42) represent, respectively, the radiation in the states characterized by ey
and e± :

dPm
sin6))2, (39.13)
dQ,

wo e2 3 2 'Jm(m 0 sin O)"

- — —p m (39.14)
2ir R H 0 tan 6

39.1 Problems for Chapter 39

1. Fill in the details leading to (39.9).
416 CHAPTER 39. SYNCHROTRON RADIATION II— POLARIZATION

2. Compute the power radiated into the mth harmonic for a given polar­
ization, (Pm)|| and (Pm)j_, starting froin (39.1). Note carefully that ||
polarization means that only the <j> component of j enters, so that

[ - n X (n X j)] •<£ = j •<£.

One form of the answer is

2 r ¡imp
(-Pm)li = — mu>o/? 12 /?J2m( 2 m/?) + ¡3 J^ dzJ2m(z)

r 2m>3 dz
2m

p 2m p j p2m p
(Pm)x = 0 2 m/? / - ? 2m W - / dzJ2m(z)
Jo z Jo

[Hint: In integrating over <¡) it is convenient to perform a coordinate rota­

tion about the x axis. That is, if

n = (x ) t/, z) = (sinocos </>, sin0 sin<j>, cos#),

replace y z, z —y, x —+ x. In this way the exponent becomes

independent of </>, and the integrals are greatly simplified.]

3. Compute the total radiated power for a given polarization, and P±.
The result is
6 + B2 2 - B2
= P±= P D
8 8
so nonrelativistically, and ultrarelativisitically, respectively, the ratio of
power in the two polarizations is

¿ = 3 (/?<!), = 7 (/?«!).
P±

[Hint: Do the (j) integral, as described in Problem 2, then the frequency

integral, followed by the r integral.]
Chapter 40

Synchrotron Radiation
III— High Energies

40.1 Range of Important Harmonics

For most practical applications, we are interested in the frequency spectrum of
the power radiated by high energy electrons. In the limit /? —►1 the power
radiated into the mth harmonic, (38.58), is approximately

e2
Pm « W0— 2mJ'2m(2m), (40.1)

which becomes for large harmonic number, m^> 1 ,

q1/6 e2
Pm ---- r(2/3)u>0— m 1/3, (40.2)
7r rt

where we have used the asymptotic form of •72m(2m),

ol/6
^2m(2 m) ~ — r(2/3)m"2/3, m > 1, (40.3)

which is already good to within 15% for m = 1. We will derive (40.3) in

the following section. The total power radiated is thus roughly estimated by
summing (40.2) over all harmonics:

oo mc

P= £ Pm ~ £ ml/3> (40-4)
m=l m=l

where we have noted that the continuing increase in the power radiated into
higher and higher harmonics must break down for sufficiently large m, since the
total power radiated, (38.65), is finite. Consequently, we have cut off the sum

417
418 CHAPTER 40. SYNCHROTRON RADIATION III— HIGH ENERGIES

at mc, the critical harmonic number. In terms of this cutoff, the total power
radiated is
P ~ m*/3 , (40.5)
which, on the other hand, is, by (38.65), proportional to (E/mc2)4:

— (40.6)
me2 Jv
P

Therefore, we conclude that the order of magnitude of rac is

(40.7)

Roughly speaking, then, the maximum frequency of radiation is of the order

3
E
(40.8)
mc 2

which implies the shortest (reduced) wavelength radiated is about

'm e21 V
E
-J R, (40.9)

where we have used the relations

c c
ÜJ = —, and ujq « (40.10)
A R
For an electron, the shortest wavelength radiated is approximately

R( m)
^c(A) (40.11)
£ 3 ( Ge V) '

where one Angstrom, Á = 10~ 8 cm, is a characteristic x-ray wavelength. Thus,

the synchrotron radiation produced by a high energy electron is characterized
by very large harmonic numbers; and consequently one gets visible, ultraviolet,
and x-ray radiation from a typical accelerator.

40.2 Asymptotic Form for

We now furnish a derivation of the asymptotic formula for J U 2 m ), (40.3).
Starting from the integral representation for the Bessel function [see (38.53)]
f W 11
J2m(¿0 = / — cos(¿sin<£ — 2m<¡>), (40.12)
Jo n
we obtain /•7T J1
J'2m{z) — ~ / — sin $ sin(z sin <£ — 2m<j>), (40.13)
Jo *
40.3. SPECTRAL DISTRIBUTION 419

which, when we set z — 2 m, becomes

J2m(2 ra) = — f
— sin<f)sm[2m(sm<f) — <j))].
Jo ft
(40.14)

If 2m is large, the integrand of (40.14) oscillates rapidly as varies, leading to

destructive interference except near <j>= 0. (This is the basis of stationary phase
or saddle point evaluations.) The main contribution comes from that value of
<j) satisfying the stationary condition

d
— (sin - 0, (40.15)

which implies, as expected,

cos <j) = 1 , or <j) = 0 . (40.16)

Therefore, making use of the approximations, for </><C 1,

<f>3
sm — —, (40.17)
sin<^ « <^, (40.18)

and noting the range of integration can be extended to infinity with negligible
error, we obtain

4 (2» " ) ------Im

= -I m t 'f h - *

3 \ 2/3 r(2 /3 )c_,W3

— —Im
mJ 37r
31/ 6 r(2/3)
for m 1, (40.19)
2ir m 2/3 ’
which is just (40.3). In the above evaluation, we have used Cauchy’s theorem to
perform a change of contour, as shown in Fig. 40.1, and have used the definition
of the gamma function (18.23):

(40.20)

Notice that (40.19) is valid for m either integer or half-integer.

40.3 Spectral Distribution

We now want to improve upon the qualitative discussion of the high energy
power spectrum given in Section 40.1. As we have previously remarked, the
420 CHAPTER 40. SYNCHROTRON RADIATION III— HIGH ENERGIES
<j> plane:

Figure 40.1: Change of contour used in evaluating (40.19).

approximation (40.2) breaks down for sufficiently large m. This can be traced
to the fact that, in (38.58), /? is not exactly equal to 1. Consequently, both terms
there contribute but we will concentrate on the first one as it contains all the
essential characteristics of the radiation. [See Problem 40.2.] Thus we seek an
asymptotic expression for J 2m( 2 m/?), starting from the integral representation
(40.13):

J2m(2m/3) = — [ — sin sin 2m(/? sin — <j>). (40.21)

Jo *
As before, in the limit when /? —» 1, m 1, the main contribution comes from
the region near <j> = 0. Therefore, we may expand the integrand in (40.21) as
follows:

sin <j) sin 2 m(/3 sin (j) — <¡>) « (j) sin 2 m -<t>
9 3!
1
= (j) sin ( 2 m - 4 ( 1 - f}) -

>sm m ( ( l - / 3 2 ) ¿ + i ¿ 3

= —\/l — /32x sin mi(1 - / ? 2)3/2 ( x + i x 3

(40.22)

where we have introduced the change of scale

<j>= \Jl - (P x. (40.23)

As a result, in this limit, (40.21) can be approximated by

Jím(2mP)~ (1 “ P2)J ~xs^n( m(! _ /?2)3/2 (* +

_ (1 — /^2) Im f°° d x x e im( 1-0 3)3/ 2Oc+*3/3)
(40.24)
7T Jo
40.3. SPECTRAL DISTRIBUTION 421

Figure 40.2: Stationary phase contour for evaluation of (40.24).

For m fixed and /? approaching unity in such a way that m( 1 —/?2)3/ 2 <C 1, the
significant contribution to (40.24) comes from the region where x is large, and
(40.24) reduces to (40.19):

JLVm/3) ~ (1 - /?2) sin ( y (1 - /?2 ) 3/ V )

=T sin(?*’)■ (4°'25)
where all reference to the speed of the particle has disappeared. However, for
sufficiently large m, the parameter ra (l—/?2) 3/2 becomes large, and the integrand
undergoes rapid oscillations in x except near the stationary points, which satisfy

A 1 + x 2 = 0; (40.26)
dx

that is, the stationary phase points are located at

x = ± i. (40.27)

By extending the region of integration from —oo to +oo, we evaluate (40.24)

asymptotically by following the standard procedure of the saddle point method
(or the method of steepest descents). We deform the contour of integration so
that it passes through the stationary point x = i, because then the dominant
contribution comes from the vicinity of that point. (See Fig. 40.2.) In the
neighborhood of x = z, we let
x = i + £, (40.28)
where £ is real, to take advantage of the saddle point character. For small £

X+ l* 3 = + ^ + 5(* + £)3 w * (jj + £2) > (40.29)

so that the exponential factor in (40.24) becomes

e-fm (l-/5 2)3/ 2e-m (l-/? 2)3/ 3í : (40.30)

422 CHAPTER 40. SYNCHROTRON RADIATION III— HIGH ENERGIES

Figure 40.3: Sketch of power emitted into mth harmonic as a function of m.

What is actually plotted is 2mJf2m(2mf3) for ¡3 = 0.99. In this case mc = 356.

which falls off exponentially on both sides of x = i. The resulting Gaussian

integral in (40.24) leads to the following asymptotic form:

J 'm(2m/?) ~ m( 1 - /?2 ) 3/2 > 1. (40.31)

z y 7rm
Thus, for very large harmonic numbers, the power spectrum decreases expo­
nentially in contrast to the behavior for smaller values of m where it increases
like m 1/ 3. The transition between these two regimes occurs near the critical
harmonic number, mc , for which

mc( l - £ 2) 3/2 = l, (40.32)

or
m« = ( l - ^ ) - ^ = ( - ¿ ) 3 , (40.33)

supporting our previous estimate, (40.7). The bulk of the radiation is emitted
with harmonic numbers near mc. The qualitative shape of the spectrum is
shown in Fig. 40.3.

40.4 Angular Distribution

We now go back one more stage and examine the angular distribution in the
high energy limit. For the radiation in the plane of the orbit (6 = 7r / 2 ), (38.42)
40.4. ANGULAR DISTRIBUTION 423

becomes

^ <4°-34>
If we further let /? —> 1 and mc ^> m^> 1, we find from (40.3) and (40.2),

1o l / 6 T2
dPm CJo
m — r(2/3)(2/m )2/ 3
2w d9 27r R

.2/3 (40.35)

Here we compare Pm, which increases as ra1/ 3, with dPm/dOy which behaves as
m2/ 3, from which it is evident that the radiation, for large harmonic numbers,
is confined to a small angular range around 0 = 7r/2 of width

AO ~ m “ 1/3. (40.36)

Since most of the radiation is emitted with harmonic numbers in the neighbor­
hood of

mc (40.37)

the radiation is concentrated in an angular range about the plane of the electron
orbit of the order
2
A0 ~ —— = \/\ —/32. (40.38)
E
A plot of the distribution in 0 is given in Fig. 40.4.
This radiation, concentrated around 0 = 7r / 2 , is predominantly polarized in
the plane of the orbit, according to (39.13) and (39.14). More precisely, it can be
shown (see Problem 39.3) that in the ultrarelativistic limit the ratio of the power
radiated with parallel polarization to that with perpendicular polarization is

P\
7, for / ? « 1, (40.39)
P±

while in the nonrelativistic limit,

P\
3, for ¡3 <C 1. (40.40)
P.L

That is, for any value of /?, the radiation is strongly polarized, the degree of po­
larization increasing with the speed of the charged particle. This characteristic
distinguishes synchrotron radiation from thermal radiation, and was, for exam­
ple, the clue to understanding the origin of the nonthermal radiation emitted
by the Crab Nebula.
424 CHAPTER 40. SYNCHROTRON RADIATION III— HIGH ENERGIES

Figure 40.4: Angular distribution of synchrotron radiation for m = 100 and

f3 = 0.99, from (38.42).

40.5 Qualitative Description

One might ask whether there is a more direct way of seeing that, for syn­
chrotron radiation produced by a high energy charged particle, the character­
istic frequency emitted is so much larger than the Larmor frequency, as stated
by (40.8), and that the angular distribution is so strongly peaked in the for­
ward direction, as suggested by (40.38). We here wish to emphasize the simple
physics behind these striking features. For a point charge, the current density,
j(r,¿), is given by (38.14), so that the vector potential, in the Lorentz gauge,
can be computed from (31.52) to be [see Problem 31.5]

A(r ,t) = M O (40.41)

lr — r ( 0 1 [ i — * v (0 ] '

where r (tf) is the position vector of the particle at the emission time t', while t
is the detection time, which are related by (31.46),

t —t' + —|r —r(t')|, (40.42)

c

and n is the direction o f observation

r - r(t')
n= (40.43)
| r - r ( t ') r
40.5. QUALITATIVE DESCRIPTION 425

Figure 40.5: Directionality of synchrotron radiation.

If we let <j> be the angle between n and v, the factor

1 _ 1
(40.44)
1 _ JLL2L _ 1 - ¡3 cos <f>’
C

for ¡3 —*• 1, is dominated by the small angle region, so we may approximate it

by

l-/?(l-^/2+...) * 1(1 - p + 4,2) ■ (4(L45)

As we saw in greater detail for impulsive scattering [see (37.37)], the radiation
is therefore concentrated in a narrow angular range near the forward direction

me 2
<t>~ x/ 1 - /?2 = (40.46)
HT'
which is just the behavior found in the preceding section.
Next, we seek a qualitative understanding of the characteristic frequencies
radiated by a high energy charged particle moving in a circle with frequency
cj0. Because of the strong directionality of the emitted radiation as expressed
by (40.46), the radiation detected at a particular point only arises from a small
portion of the orbit, or equivalently, is only emitted during a time interval
small compared with the period of revolution, 2ir/u)o. (See Fig. 40.5.) This
effective emission time interval is of order (27r/u>o)<^, so the important frequencies
radiated are ~ (emission time) - 1 ~ (u>o/27r)(l/</>). That is, the smaller the time
interval involved in the emission, the higher the frequency emitted. Therefore
a typical frequency emitted is
ÍÜQ E
—7 7 = ----- 0 ^ 0- (40.47)
a/ 1 - ff2 me2

However, this is not what the observer sees since detection time intervals are not
equal to emission time intervals. To see the connection between these intervals,
the Doppler effect [Christian Doppler (1803-1853)], we recall that these two
times are related by (40.42), which implies for the respective time intervals

dtf — d t -----n •[—v( t f)dtf] (40.48)

426 CHAPTER 40. SYNCHROTRON RADIATION III— HIGH ENERGIES

dt = ^1 — ~n • dt' ~ (1 —/32)dtl, /? « 1, <t>~ \/\—ft2. (40.49)

This is the origin of the denominator factor appearing in (40.41). From

(40.49), we see that the time interval of detection is shorter than that of emission
by a factor 1 —/?2, implying that the detection frequency, u>d, is higher than the
emission frequency by a factor (1 — ft2) " 1:

1 1 E
Ud = Wo ( (40.50)
( l - £ 2) ~ e (1 — /32)3/2 K m c <

which is the characteristic frequency found in (40.33).

The theory of radiation from electrons in betatrons was worked out by
Schwinger in 1945, although not published until 1949 [J. Schwinger, Phys. Rev.
70, 798 (1949)]; an earlier unpublished manuscript by Schwinger has now been
transcribed by M. Furman [LBNL-39088/CBP Note-179]. Independently, D.
Ivanenko and A. A. Sokolov worked out the spectral and angular properties of
the radiation [D. Ivanenko and A. A. Sokolov, Dokl. Akad. Nauk SSSR [Sov.
Phys. Dokl.] 59, 1551 (1948)]. For the history of the development of this sub­
ject and many further details, with a particularly Russian perspective, see A.
A. Sokolov and I. M. Ternov, Synchrotron Radiation (Akademie-Verlag, Berlin;
Pergamon Press, Oxford, 1968).

40.6 Problems for Chapter 40

1. Evaluate the power radiated into the mth harmonic, (38.58), for m = 1,
/? —►1 , and compare with the corresponding value obtained from the
asymptotic formula (40.3).
2. Improve on the discussion: of the high energy limit in Section 40.3 by
considering the contribution from the integral in (38.58).
3. For an ultrarelativistic electron, the radiation is concentrated near the or­
bital plane, 0 = 7t/ 2 , so the quantity e = 1 —0 1 sin2 6 becomes very small.
Show that asymptotically the following relation holds for the angular dis­
tribution for the power radiated into the mth harmonic:

in the limit m —►oo, e —►0. Here we encounter the integrals

( r 3/2)l
# 1 /3 (0 = J dtC
OS + j ) >
To poo / ^3\
# 2/ 3 ( 0 = — J dt t sin yzt + - j ) ,

where z = ( 3 < / 2 )2/ 3.

Chapter 41

Propagation in a Dielectric
Medium

41.1 Equations for the Normal Modes

We now turn from the mechanisms by which electromagnetic radiation (or light)
is produced to how it propagates in a material medium, described at a macro­
scopic level. Quite quickly, we shall specialize to the case of a dielectric, with
inhomogeneity in a single direction only. We begin by restating Maxwell’s equa­
tions for a macroscopic medium,

V x H = lf>+ — J, (41.1)
c c
— V x E = -B, (41.2)
c
V •D = 47T/0, (41.3)
V •B = 0. (41.4)
Before proceeding, we note that here the scalar equations, (41.3), (41.4), are
not independent of the first set, (41.1), (41.2), since taking the divergence of
the latter, and making use of the current conservation condition, (31.5), we find

0 = J ^ (V •D — 47iy>), (41.5)

0=^(V.B). (41.6)

Thus, excluding statics, the divergence equations, (41.3), (41.4), are subsumed
within the curl equations, (41.1), (41.2).
To concentrate our attention on light of a definite frequency cj, we introduce
the Fourier transform

/ d teiutF ( r , t ) = F ( r,w ), (41.7)

427
428 CHAPTER 41. PROPAGATION IN A DIELECTRIC MEDIUM

which amounts to the replacement in (41.1) and (41.2) of

—iu), (41.8)
dt
yielding for the vector equations

tco,
V x H = ----- D +
IÜJ
—V X E = B. (41.9)

Specifically, we assume a linear, dispersive medium, for which

B = /(H, D = eE, (41.10)

where //(r,u>) and e(r,u>) are complex functions. We will further assume

K r ,w ) « l, ' (41.11)
since a ferromagnet cannot follow the rapid oscillations of the electromagnetic
field in a light wave. It will be sufficient for our purposes to suppose that spatial
variation of e occurs in the z direction only,

e(r>W) = £ (* ,«). (41.12)

Since the material is then translationally invariant in x and t/, we introduce

corresponding spatial Fourier transforms:

J d x d y e ~ i(k*x+k»y)F (r,w ) = J (dr±)e~ik-L' rxF(r,u>) = F(z,h±,u>).

(41;13)
Letting n be the unit vector pointing in the 2 direction, we replace the gradient
operator by
f) d
V = Vj_ + n — ik± + n — . (41.14)

It is natural to project Maxwell’s equations, (41.9), onto spaces parallel to,

and perpendicular to, n. The z components of (41.9) become, using a highly
redundant notation,

, _ V tU /C A '1 T
—ik j..( n x B j_ ) = ------- Ez -\------ Jz ,
c c
2Ll)
.(nxEx) = B z, (41.15)

while the ± (*, y) components appear as

41.1. EQUATIONS FOR THE NORMAL MODES 429

For w 5¿0, (41.15) determines Ez and Bz in terms of Ex and B x :

Bz = ——k x •( n x E x ),
U)
c A ir
Ez = — k± •( n x B ± ) - i — Jz . (41.17)
UJC WC

Inserting these expressions into the J_ set, (41.16), we have the following equa­
tions for the J. components:

dn ÍU€ „ C, , „ . 47T , ,^
— B x -------- n x E x + « k x —kx * (n x E x ) = ------ n x J x , (41.18)
oz c w c
d ^ iu> ^ c , / ^ x ,47rT
^ -E x H-----n x B x - « k x — k x * ( n x B x ) = k x — Jz , (41.19)
OZ C (jJ€ we

which mix B x = (B X)By) with n X E x = (—Ey ,E x), and Ex = (E XyEy) with

n x B x = (—By ,B x). Next, as an alternative form of the same equations, we
take the cross product o f the above equations with n,

d iwe c 47T
■^nxB i + Ex + i n x k x - k x - ( n x E x ) = — J x , (41.20)
oz c to c
d m zw , , c _ . _ v , A tt T / i< \
— n x E x ------ B x — m x k x — kx •( n x B x ) = n x k ± — Jz , (41.21)
oz c we we

where we have noted that for any vector V x in the xy plane, —n X ( n X V ± ) =

V x - Now, we project these vector equations on kx- From (41.18) and (41.21)
we find a system of equations relating kx •B x and kx •(n x E x ):

d , „ iwe . ^ k\c 2 \ 47r

_ k x .B x - — (1 kx-(nxEx) = --- kx-(nxJx),
(41.22)
d, . „ . ito, _
— kx • n X E x -------kx - B x = 0, (41.23)
oz c

and, from (41.19) and (41.20), a system relating kx - Ex and kx •( n x B x ) :

^ - k x -E x + — ( l - ^ ~ ) kx •(n X B x ) = k \ — Jz , (41.24)
oz c \ to2e J toe

| - k x - ( n x B x ) + — kx- Ex = — kx -Jx (41.25)

OZ c c

The vectors n and kx define a plane, called the plane of incidence, so that the
system (41.22) and (41.23) relates the component of B x in this plane to the
component Ex perpendicular to this plane, and vice versa for system (41.24),
(41.25). That is, if we take kx in the x direction, we have equations governing Bx
and Ey, and By and EX1 respectively. These perpendicular components belong
together physically [see Section 7.2 and also the relativistic transformations
in (10.83) and (10.84)]. By combining (41.22) with (41.23), and (41.24) with
430 CHAPTER 41. PROPAGATION IN A DIELECTRIC MEDIUM

(41.25), we convert these systems of first order differential equations into second
order ones:

82 u>2e , „ iu> 47T, , , x

(41.26)
/ V

+ = -f- * í kj_*(nxEx) = ~ — — kj_ •(nxJx),

dz2
d_ I d . . __ v
7 ^ lc x .(n xB x)
dz ( i? ~ ^ ) k x *( n x B j -)
iu> , 9 4ttt d ( 4ir, , \
— ---- k± —-J z + tt-( — kx •Jx ) j (41.27)
c we dz \ ec J

where we must remember that e is a function of z. From these components,

perpendicular to the plane of incidence, those in the plane of incidence can
be obtained from (41.23) and (41.25), and the longitudinal components (the
^-components) from (41.17).
When 6 is a constant function of 2 , the differential operators in these last
two equations are the same, namely

d2 u>2c 2
(41.28)
+ ~c? ~ k ± '

We further assume that e is real. Then, depending on the sign of u>2e /c2— , the
solutions to the corresponding homogeneous differential equations are different:

“ 2---- &2 < 0: exP ( ^ (41.29)

U)2€
2---- &x > 0 : exp ±i (41.30)

The first possibility is essentially that discussed in electrostatics, in that there is

no wave propagation, and so we will not consider it further here. [See Chapter
14.] For the second possibility, we do have propagating waves. In terms of

k2 -k i, (41.31)

we construct a plane wave by combining the imaginary exponential structure in

(41.30) with the transverse spatial dependence etk±

e*kjL • r ± e ikzz __ e ¿k • r
(41.32)

which represents a plane wave moving in the direction k, since the phase is
constant on a plane perpendicular to k. The wavelength, A, is defined as the
distance over which the phase advances by 27T, so

27T 1
|k|A = 2tt, or |k| = (41.33)
A5
41.2. REFLECTION AND REFRACTION: JL POLARIZATION 431

|k| is called the wavenumber, k the propagation vector, and A the reduced
wavelength. Including the time factor e“ *wt, we find the dependence of a plane
wave on space and time to be
«(k • r-w t)
(41.34)

The surface of constant phase advances with time so that

. dr
(41.35)
k '3 i = " ;
the phase speed thus is
v — — = o>A = v\. (41.36)
|k|

From (41.31) we see that the square of the wavenumber is

\k\2 = k2
± +k2
0= ^ , (41.37)

so that the phase speed is

c c
(41.38)
y/l n’
n — y/c, (41.39)

as we saw in Section 7 .2 .

41.2 Reflection and Refraction: X Polarization

We now specialize the above discussion to a situation in which the inhomogeneity
in the dielectric constant is due to a plane interface between two dielectric
substances:

z > 0: e(z) = 62,

z < 0: c(z) = e\. (41.40)

Since the relation between field and source is linear, the field produced by a
prescribed current can be expressed in terms of an appropriate Green*s function.
Green’s function corresponding to (41.26) satisfies

- +£ - ‘ i ) »<*. *') = «(* - *')■

which as we noted above refers to the component of Ex perpendicular to the

plane of incidence, defined by kx and n, and so we will call this the _L polariza­
tion. Weimagine that the source lies in medium 2, that is z1> 0 . The boundary
conditions at the interfaceare that kx * (n x E x ) andkx *Bx becontinuous,
which impose the conditions that g and, by (41.23), dg/dz be continuous at
432 CHAPTER 41. PROPAGATION IN A DIELECTRIC MEDIUM

z = 0. For 2 —» ± o o we must have outgoing waves, since the source is localized

at z'. The solutions to (41.41) in the three regions have the forms

(41.42)
z’ > z > 0 : ti k t 2 z
g = B eik“ z + Ce~ik‘ *z (41.43)
z< 0: g = D e~ik‘ lZ, (41.44)

where the subscripts 1 and 2 refer to the value of kZ) as defined by (41.31), in
medium 1 and 2, respectively. Note that these equations are very similar to
those considered in electrostatics; in fact, when in both dielectrics

(41.45)

we make the replacement

(41.46)

in (41.42)-(41.44), and thereby recover the same system in (14.11)—(14.13),

except that, here, the wavenumbers are different in the two media.
We now determine the coefficients A through D by satisfying the boundary
conditions stated above as well as the discontinuity at z = z1. At z = 0,
continuity of g yields the relation

B + C=D (41.47)

while continuity of dg/dz requires

B -C = -~ D . (41.48)
kZ2

The combination of these supplies

(41.49)

(Note that if e\ = 62 , these coefficients are simply B — 0 and C = D, which

expresses the obvious fact that the entire wave is transmitted, with no reflection,
since there is no interface.) The presence of the source term in (41.41) imposes
the conditions that, at 2 = zf>

g is continuous, and
* Q • Z — z'-\-Q

(41.50)
41.2. REFLECTION AND REFRACTION: ± POLARIZATION 433

Consequently, there are two additional equations for the constants,

Aeik“ *' = B e * ’ **'+ Ce~ ik‘ *z' , (41.51)

- A e ik‘ + Beik‘ *z' - C e~ik^ z' = — . (41.52)

Equations (41.49) and (41.51)-(41.52) can be solved to yield, successively,

1 Jkyz' (41.53)
C =
2k2 ’
k2 - h i
B = (41.54)
k2 + k\ 2k2
2k2 i cikzz'
D = (41.55)
k2 ■+■k\ 2k2

A = a*' i J _ e-ik 3z' (41.56)

k2 + ki 2k2 T 2k2

where we have simplified the notation by dropping all the z subscripts on the
k ’s. Combining these results, we may now write down the Green’s function in
the two regions,

z > 0: g(z, z') = *'■ + (41.57)

2AJ2 ¿K2
z < 0: g (z,z') = t-!r - e - ik' zeik*z', (41.58)
2k2
in terms of the reflection and transmission coefficients,

&2 — k\
(41.59)
k2 + fa ’
2k2
t = (41.60)
¿2 + &1
The two terms in (41.57) have a simple physical interpretation. The first refers
to the direct wave from the source, since \z — z*\ is (the ¿-projection of) the
distance from the source, while the second represents the wave reflected from
the interface, since ¿ + ¿ ; is (the ¿-projection of) the distance from the source to
the interface and back to the observation point ¿. [See Fig. 41.1.] The following
algebraic identities relating t and r,

t = 1 + r, (41.61)
M = Jfe2' ( l - r ) , (41.62)

restate the continuity of g and g' at z = 0, which in turn are consequences of

the continuity of the tangential components of E and B across the interface.
Moreover, on combining (41.61) and (41.62) to give

kit2 = *2(1 - ^2), (41.63)

434 CHAPTER 41. PROPAGATION IN A DIELECTRIC MEDIUM

Figure 41.1: Interpretation of direct and reflected waves.

k±

Figure 41.2: Incident, reflected, and transmitted propagation vectors.

we have a statement of conservation of energy, as we will show in Section 41.5.

We are here primarily interested in how a plane interface reflects and trans­
mits light, irrespective of how the radiation is produced. Accordingly, for regions
to the left of the source, we regroup the factors of Green’s function suggestively:

0 < z < z' : g(z, z') = (e~ik*z + reik*z) ( a*‘ ) - (41-64)

z < 0: g (z,z') = te~ lklZ ) • (41.65)

The common factors in parentheses refer to how the wave is produced, which
is not of interest to us at the moment. The other remaining factors describe
the reflection and transmission of a plane wave: unit amplitude incident on the
interface, amplitude r reflected, amplitude t transmitted.
We recall that Green’s function, g (z) zf) ) describes only the ^-dependence of
the wave propagation; the x and y dependence is contained in the factor e*kj- *rJ-
[recall (41.32)]. Since the transverse momenta are unaltered, while the longitu­
dinal momenta (kz) change from kz 2 to kz\ as the wave crosses the interface,
we have the geometrical picture shown in Fig. 41.2 for the incident, reflected,
and transmitted propagation vectors. Under reflection kz merely changes sign,
so the angle of incidence equals the angle of reflection. The relation between
41.2. REFLECTION AND REFRACTION: ± POLARIZATION 435

the wavenumber and the index of refraction is, from (41.37),

|k| = — , (41.66)
c
with n = yfl. The continuity of kx at the interface,

|k2 |sin02 = |ki|sin0 i, (41.67)

then implies Snell’s law of refraction (discovered in 1621),

n\ sin#i = n 2 sin (41.68)

relating the angle of incidence 62 to the angle of refraction .

We may rewrite the reflection and transmission coefficients given in (41.59)
and (41.60) in terms of the angles 9\ and $2 by using

(41.69)
i k x r cot9 '
to obtain
_ sin(#i — 62)
(41.70)
sin(0i + 02) ’
and
2 sin 0 i cos 02
(41.71)
sin(0i + 02)
but the wavenumber form is usually preferable. This is particularly evident
for normal incidence, where 6\ = 02 = 0 and (41.70)-(41.71) is indeterminate.
On the other hand the wavenumber form, (41.59)—(41.60), is well defined, since
when = 0 , kz = which yields
ni - n2
(41.72)
ni + n2 ’
2n2
(41.73)
ni + n2'
Also from the wavenumber form it is easy to see that if n\/n2 —» 00 , the
coefficients become
r = —1, t = 0; (41.74)
this corresponds to the situation of total reflection by a perfect conductor. We
verify this last statement by recalling the plane wave part of the Green’s function
between the source and the interface, (41.64):

0 < z < zf : g oc e~ik2tZ + reik2tZ. (41.75)

When (41.74) is satisfied, g vanishes at z = 0,

0 o c (l + r) = O, (41.76)

which expresses the vanishing of the tangential component of E on the surface,

characteristic of a perfect conductor.
436 CHAPTER 41. PROPAGATION IN A DIELECTRIC MEDIUM

41.3 Reflection and Refraction: || Polarization

We now turn to a consideration of the second polarization state, in which Ex
lies in the plane of incidence and B x is perpendicular to that plane. We will
call this the || polarization in contrast to the _L case discussed above.
Green’s function in this situation, which relates kx •(n X B x ) to its source,
satisfies, according to (41.27),

(41.77)

1
1
dz (\el

II
8*1

-
where, again,
" 2
k \ ^ e -k \ . (41.78)

With the plane interface between the two dielectric media located at 2 = 0, the
form of the solution in the three regions, when z‘ > 0, is

z > z' : g = Aeik2Z, (41.79)

0 < z < z' : g = B eik3Z + C e~ik2Z, (41.80)
z < 0: g = D e~iklZ, (41.81)

where, as before, the 2 subscript is suppressed. Here, the boundary conditions

at 2 = 0 are the continuity of g (since B x is continuous) and the continuity of
jg ' [since E x, which is related to g by (41.25), is continuous], which furnish the
relations

B + C = £>, (41.82)

B - C = - — h-£> (41.83)
£1 h

implying

c“Ki+Hs.,a (41.84)

1
B , . 1- ^ 10 . (41.85)
2 V £i h ,

At z = z', g is continuous, while from (41.77)

z —z'+O
\_d_
= 1, (41.86)
€2 d z* z= z'~ 0

so
B eik*z' + C e~ik2Z' = Aeik2*', (41.87)

and
ik2 ( - A e ik2Z> + B eik2Z' - C e~ik' z') = e2- (41.88)
41.3. REFLECTION AND REFRACTION: || POLARIZATION 437

The solution to this system of equations is

^ikzz1 (41.89)
C
2h6 ’
1 É2. kl •
_____ £i k2 ik2z'
B (41.90)
i + '

D (41.91)
l + f f f t 2*2
i _ e_2_ÍLL *
£i k2 c ¿fc2^; i ^2 —jk2z '
A (41.92)
1 + iajp. 2A;2 2*2
£l «2

These are as in the other polarization, as given by (41.53)-(41.56), except that

in the amplitudes multiplying the exponentials, where we had kz , we now have
kz/e. Green’s function is thus

z > 0 : g = — eik^\Z~Z'\ 1 r iílp * fc 2 (^ + ^ /) (4 1 .9 3 )

y 2k2 + 2 fc2 6

. - i kiz gik^z1
z < 0 : g — t ——e (4 1 .9 4 )
2k2

Green’s functions for the two polarizations, J_ and ||, are given by (41.57)-
(41.58) and (41.93)—(41.94), respectively, with the corresponding reflection and
transmission coefficients given by

_L mode mode
ki - * 2 . * iA i ~ k2/e2
^1 + ^2 ' ki/ei + k2/c2 '
2Ar2 2 * 2/62
(41.95)
*1 + *2) * iA i + * 2/^2

In both cases the algebraic relation (41.61) holds true,

t = 1 + r, (41.96)

expressing the continuity of Ej_ and of B x in the _L and ||polarizations, respec­

tively, but the statement of energy conservation is different in the two situations
[recall (41.63)]:

-L mode || mode

k it2 — * 2 (1 — r2), — t 2 = — (1 — r2), (41.97)

¿1 ^2

where, again, the second relation is obtained from the first by the substitution
kz ►kzf 6 .
It is particularly interesting to ask when the reflection coefficient can vanish.
For the _L polarization, r = 0 only when kz\ = kZ2, or 61 = 62 , that is, in the
438 CHAPTER 41. PROPAGATION IN A DIELECTRIC MEDIUM

absence of an interface. However, for the || polarization, another solution to

r = 0 exists. The vanishing of the reflection coefficient requires

kzi _ kz 2
(41.98)
^1 ^2

or, since
*2 lo21 k\
(41.99)
e2 c2 e e2 ’
the condition becomes

("D i (41.100)

The new possibility for r vanishing occurs when

- = . / í i ± í i |kx|. (41.101)
c V eie2
Geometrically, since
|kx| = ~ n sin 0, n = y/l, (41.102)
c
(41.101) is satisfied when the angle of incidence, 02 equals 0#, where

£1__
sin Ob = (41.103)
+ ^2 *
or,
tan Ob = — . (41.104)
n2
At the incident angle 0b , the || mode is completely transmitted, so that the
reflected wave is completely polarized perpendicular to the plane of incidence.
The phenomenon was discovered by Sir David Brewster about the year 1800,
and in consequence 0b is called Brewster’s angle.
To express r and t again in terms of the angle of incidence O2 and the angle
of refraction 0 i, we use the geometrical relations

fc* = |ki|cot 0, nsintf = —|k¿|, (41.105)

to write
kz
(41.106)
M 2
Consequently, for the || mode, the reflection and transmission coefficients can
be given as

tan (0i — 02)

(41.107)
tan(0i + 02) ’
2 sin 02 cos 02
(41.108)
sin(0i + 0 2 ) cos(0i — 02)
41.3. REFLECTION AND REFRACTION: || POLARIZATION 439

B B

Incident:
k Reflected:

Figure 41.3: Relation between field orientations for incident and reflected waves.

Here we observe, from (41.107), that Brewster’s angle occurs when the sum of
the angle of incidence and the angle of refraction forms a right angle,

0i + 02 = */2. (41.109)

For normal incidence ($i = 0 2 = 0 ) , as before it is convenient to return to

(41.95) and employ
kz uj
(41.110)
e nc
For the two polarizations we have [recall (41.72) and (41.73)]

± mode mode
n\ - n2 n\ — n2
r = — , r = --------------------
n\ + n 2 ni + n 2

t= 2 n 2 . (41.111)
ni + n2
< = 2 n i .
ni 4- n 2

But for normal incidence there can be no physical difference between the two
polarizations, since the plane of incidence is not defined. To reconcile the two
forms in (41.111), we recognize that the reflection and transmission coefficients
for the ||mode refer to B, while those for the Jl mode refer to E [see (41.26) and
(41.27)]. Under reflection, E reverses its sense relative to B, since the direction
of propagation is reversed. [See Fig. 41.3.] So the two forms for the reflection
coefficient in (41.111) are physically equivalent. To reach the same conclusion
for the transmission coefficients, we recall from Section 7.2 that the magnitudes
of the electric and magnetic fields in the plane wave are connected by

e|E|2 = |B|2, (41.112)

which relates transmission coefficients referring to electric and magnetic fields

by
(41.113)
V^I ni
which explains the ratio of the two forms for t in (41.111).
440 CHAPTER 41. PROPAGATION IN A DIELECTRIC MEDIUM

41.4 Total Internal Reflection

We now return to Snell’s law, (41.68),

n\ sin#i = ri2 sin 02 - (41.114)

Suppose we consider a plane wave going from a region of greater dielectric
constant to a region with a smaller dielectric constant, that is

m < n 2 => sin#i > sin 02 . (41.115)

For this arrangement, there are incident angles satisfying

n2 sin 02 > ^i> (41.116)

for which Snell’s law can be satisfied only if

sin 0i > 1. (41.117)

What does this mean physically? It is helpful to consider the angle of transition
into this new regime, where
n2sin02 = ni, (41.118)
or
sin 0i = 1. (41.119)
At exactly this angle, the transmitted wave travels along the interface since

(41.120)
-Sé

o
11
II

*4

2 becomes greater than so that

k^ = ] / ^ ~ kl (41.121)

becomes imaginary,
kli < 0 or kz i = i/c, (41.122)
where k is real. The propagation of the wave in medium 1 is now described by
gjfc.ikl _ e- KU\t (41.123)

a decreasing exponential: the field penetrates only a short distance into region
1, and no energy is transmitted into that region, it being all reflected. This last
fact follows from the form of the reflection coefficients,
kZ2 —
1: r = (41.124)
kZ2 + i’

r — (41.125)
kz2U 2-iK i/ei
fc*2/£2 + * « l A l ’
both of which have unit magnitude,
|r|2 = 1. (41.126)
In the next section, we will prove that this implies that all the energy is reflected.
We have here the situation of total (internal) reflection.
41.5. ENERGY CONSERVATION 441

41.5 Energy Conservation

We now turn to a general consideration of energy flow across the plane interface.
Recall Poynting’s vector, (7.4), which has as the component normal to the
interface
S, = f n . ( E x H ) , (41.127)
47T
where here
H = B. (41.128)
From this the total energy per unit area flowing in the + z direction is computed
to be

J dtSz(r ,t) = j •E(r, w)* X B (r, w)

/>OO i

= -7 - 2 Re / ^ n .E (r ,« )* x B (r ,w ), (41.129)
4 7T J o 27r

We next integrate over the xy plane by introducing the Fourier transform in the
kx space:

J d xd yd tS 2 = ¿ R e j ~ ^ j l | ^ n . E ( * , k x , w ) * x B ( * ,k x ,« ) .
(41.130)
The integrand in (41.130) states that the flow of energy in the + z direction per
unit frequency interval and per unit kx volume is

^ -R e [n •E (z, kx ,w )* x B (z ,k j., w )]. (41.131)

Z7T

We adopt a coordinate system such that kx is in the x-direction so that the

y-axis is perpendicular to the plane of incidence. Then for ± polarization, the
flow of energy in the —2 direction is

Re[—n •E* X B] = ¿ -R e E t Bx . (41.132)

Z7T Z7T y

From the relation between Bx and Ey , (41.23),

Bx — i 7 T“ Ey , (41.133)
UJ oz

we see that (41.132) is

— ReE*yi — Ey. (41.134)

Now we use (41.65) for the transmitted field (z < 0):

Ey = te~iklZ,
.d _ . _
l — Ey — k\Ey^ (41.135)
442 CHAPTER 41. PROPAGATION IN A DIELECTRIC MEDIUM

so the energy flowing in medium 1 away from the interface is proportional to

r\

Re E *i— Ey oc Re(&i|i|2). (41.136)

For the case of total reflection, k\ is purely imaginary, so no energy is transmitted

into region 1 as stated in the preceding section.
In region 2, 2 > 0, (41.64) expresses the incident and reflected fields as

Ey = e~ik*z + reik*z , (41.137)

so that A

i-^ E y = k2 (e~ik2Z - reikiZ) . (41.138)

Just outside the interface, at 2 = 0+, where

Ey = 1 + r>

i - E y = k2( l - r ) , (41.139)

the flow of energy toward the interface is given by

■^— R e E li^ -E y oc Re M l - |r|2 + r* - r). (41.140)

2ttw y oz
Since we are considering an incident plane wave, with real kZ) this becomes

M l - M a). (41.141)

The quantities (41.136) (at z = 0—) and (41.141) (at z — 0 + ) therefore must
be identical for two reasons:

1. The tangential components of E, B must be continuous, since there are no

surface currents. [For k\ real, we derived this equality earlier in (41.63).]

2. Energy is conserved, since there is no source of energy on the surface.

Thus, if Ari is real, the statement of energy conservation is

M*|2 = *2(1 - M 2)> (41-142)

while if k\ is imaginary, there is no flow of energy in region 1 and

0 = M l - M 2) (41.143)

or
|r|2 = 1, (41.144)
which is the situation of total reflection considered in Section 41.4.
For || polarization, the flow of energy is given in terms of

—n « E * x B = —E*By , (41.145)
41.6. PROBLEMS FOR CHAPTER 41 443

^2 perfect electric conductor

raC- i k 2( z + a / 2 )

e ik2( z + a / 2 )

2 = —a/2 z = 0

Figure 41.4: Parallel dielectric backed by conductor.

where, using (41.25) in the absence of currents,

Ex = - ^ ^ - B y . (41.146)
lüJ€ OZ
Thus the flow of energy is proportional to

Re[—B*EX] oc (41.147)

with the same constant of proportionality as before. Since now the transmission
and reflection coefficients, and the Green’s function, refer to By) we may obtain
the result by simply replacing
(41.148)

in the previous form. And so, again, the ||form of (41.97) expresses the conser­
vation of energy when k\ is real; generally, for this polarization, the conservation
of energy for a wave reflected and transmitted by an interface between plane
dielectrics is stated by

R e f — 1<|2) = — ( l - | r | 2). (41.149)

\£i J £2

41.6 Problems for Chapter 41

1. Verify (41.107) and (41.108).

2. Two parallel dielectric media are backed by a perfect electric conductor

as shown in Fig. 41.4. (A perfect electric conductor is characterized by
the fact that the tangential component of E vanishes on the surface.) A
source, to the left of the first interface, initiates an incident plane wave
described by
e ik2( z + a / 2 )

with _L polarization. (The phase is so chosen as to make the exponential

equal 1 at the first interface.) The reflected wave is represented in terms
of
rae-ik 2(z+a./2)
444 CHAPTER 41. PROPAGATION IN A DIELECTRIC MEDIUM

¿2 Cl ¿2

r e - i k 2( z + a / 2)

teik2(z-a/2)
eifc2(*+a/2)

z = —a/2 z = a/2

Figure 41.5: Three parallel dielectrics.

Find the reflection coefficient ra. What is the form of the statement of
energy conservation here?

3. Consider the same situation given in Problem 2 except that the perfect
electric conductor is replaced by a perfect magnetic conductor, for which
the tangential component of B vanishes on the surface. (In the coordinate
system given in Section 41.5, this means that dEy/dz = 0 on the magnetic
conductor for _L polarization.) Calculate the reflection coefficient for this
situation, r&. What is the form of energy conservation here?

4. Consider two parallel dielectric interfaces as shown in Fig. 41.5. A plane

wave, incident form the left, given by
eik2( z + a / 2)

gives rise to a reflected wave,

r e - i k 2( z + a / 2)^

and a transmitted wave,

l eik2(z-a/2)'

Calculate the reflection and transmission coefficients, r and t . Again, what

form does energy conservation take here?

5. Compare the results of Problem 4 with those of Problems 2 and 3, and

deduce the relations

r = l(r a+ n)

t = l(p j - r„).

These relate the statements of energy conservation in the different situa­

tions. Why do these relations hold?

6. Show that a wave incident normally on a plane boundary between two

media is transmitted without reflection if the ratio of dielectric constant
to magnetic permeability is the same for the two media.
Chapter 42

Reflection by an Imperfect
Conductor

We here briefly consider the reflection and transmission of electromagnetic ra­

diation by a conductor. A conductor is characterized by Ohm’s law, (5.11),

Jcond = <rE, (42.1)

where J COnd is the conduction current and cr is the conductivity. The effect of
this current can be incorporated into the previous discussion by noting that
Maxwell’s equation for the curl of H now becomes

__ ___ 18 ^ 47t 4 tt
V x H — D H ------Jcond H------J> (42.2)
C C/t c c
where J is a source, external to the conductor, for E and H. For fields of
a definite frequency, that is, whose time dependence is given by e~lU)t, this
equation becomes

_ „ iu ( .4 tt \ _ 47t ,
V x H — ---(6 + %— cr ) E H--- J. (42.3)
c \ U) J c

This shows that the conductor can be described by an effective dielectric con­
stant,
47T
e 4- i — <r, (42.4)
u>
which is complex, expressing the dissipative nature of the conductor. (This
was implicit in the considerations in Section 5.2.) According to (41.31), the
component of the propagation vector normal to the surface of a planar conductor
is then also complex:

n i/2

)
.47T 2
kz e+ i— cr - k± (42.5)
(jj

445
446 CHAPTER 42. REFLECTION B Y AN IMPERFECT CONDUCTOR

The simplest situation is that of a good conductor, where the conduction

current dominates the displacement current, that is

47TCT
(42.6)
u>
In the following we will assume that a is real, which can be approximately
valid only for sufficiently low frequencies, as (5.12) states. Since there is wave
propagation outside the conductor, k\ < u>2/c2, and (42.5) is, approximately,

U)2 .47T LÜ /4ira 1 + i

kz a = (42.7)
C2 U) CV ~ " v r
or
« l y 1, (42.8)

where
6= 1. (42.9)
v27ro)(r
The z dependence of the propagation of a plane wave through the conductor is
then given by
e ik z \z\ _ e -\z\/6e i\z\/6 (42.10)

The first factor on the right side of (42.10) represents the exponential attenua­
tion of the wave; the amplitude is reduced by a factor of e_1 in the distance 6.
This characteristic distance 6 is called the skin depth. As indicated by (42.1),
the current in a conductor is confined to a region within a distance ~ S of the
surface, a distance which becomes smaller as the frequency increases. For a good
conductor characterized by (42.6), the ratio of the skin depth to the reduced
wavelength of the impinging radiation, A « c/u>, is very small:

- « \/uj/2'K(t < 1. (42.11)

Thus, only radiation of long wavelengths (low frequencies) penetrates the con­
ductor appreciably.
The discussion of transmission and reflection given in the previous chapter
remains essentially unchanged except for the replacement of kz by (42.8) inside
the conductor. Therefore, for _L polarization, the reflection coefficient for a plane
wave impinging from a dielectric (labeled 2) onto a flat conductor (labeled 1)
is, from (41.59) (again the z subscript is suppressed),

&2 — &i _ ki
r —[1 — 2^(1 *)L (42.12)
k\
^ “
&2 &2 + ” 1^

since
> f c 2. (42.13)
 447

For a perfect conductor, a —►oo, 6 0, and

r = -1 (42.14)

as we saw earlier in (41.74). The fractional amount of power reflected by a good

conductor is
|r|2 « 1 - 2k2S, (42.15)
so that 2&2<$ is the relative power entering the conductor. For a perfect conductor
there is total reflection, with no energy absorbed by the conductor.
Another way of deriving this result, (42.15), is to return to the equation for
E x, (41.19),

d uj ic
— E _l + i —n X B i -------kj_[kx • (n X B j.)l = (external currents), (42.16)
oz c we
in which the effects of conduction currents are incorporated in the replacement

.47rcr
e —» e + i ----- , (42.17)
U)
which has a magnitude much greater than that of e ~ 1 inside a good conductor.
Under such circumstances, the third term in (42.16) is negligible, implying,
inside the conductor, r\

^ -E x + t-n x B j. « 0. (42.18)
OZ c
The derivative here is proportional to E x , since the electric field in the conductor
attenuates exponentially according to (42.10),

¿ E x = l^Ex, (42.19)

which implies that inside a good conductor,

„ 1 — i ujS ^
E± ~ — ---- ~ n x B x , (42.20)

relating the tangential components of the electric and magnetic fields. This re­
lation is valid for an arbitrarily shaped surface as long as the radius of curvature
of the surface is much larger than the skin depth 8. For a perfect conductor, this
supplies the boundary condition that the tangential component of the electric
field vanishes on the surface,
Ex = 0. (42.21)
The energy flow in the —2 direction just inside the conductor is now seen to
be proportional to [recall (41.131)]

- R e n - E ’ XB = R « i y ^ ( n x B i ) •( n x B j

- w5 ir I2 (42.22)
- 2 7 |B±I '
448 CHAPTER 42. REFLECTION B Y AN IMPERFECT CONDUCTOR

To find the fractional amount of energy absorbed by a good conductor, we

approximately evaluate B ¿ as though S = 0 (a perfect conductor), and compare
(42.22) at 2 = 0— to the incident flux at z = 0 -f. The latter is constructed in
terms of the fields (normalized so that the amplitude of the incident magnetic
field is unity), for the two independent polarization states,

-L: Bx = e~ik** + eik*z,

F — U (e.~ikiz _ Jk3z\ (42.23)
hy ~ ck2 [C 6 >’
II : By = e - ik*M+ eitaM,
Es = - S h . (e-ik 2* _ eik3z\ (42.24)
we2
Here, we have used (41.133) and (41.146) to relate Bx to Ey) and Ex to By)
respectively. The incident energy flux arises from the terms proportional to

Incident flux oc Re { - 7 “ limine |2, “ -|A/inc|2

( Ck2 L)€2

=i { i |B
'("=0)|1'S;|S!'(2=0)|i}' <42-25)
The tangential magnetic field is continuous since, for an imperfect conductor,
there is no surface current. Consequently, to determine the fractional amount
of power absorbed, we take the ratio of (42.22) to (42.25):

i¿8/2c
= 2k2S, (42.26)
U > /4 c &2

u>8/2c
(42.27)
ck2/4u)€2

Of course, (42.26) agrees with (42.15). [See also Problem 1.]

42.1 Problems for Chapter 42

1. Derive the analog of (42.15) for || polarization.

2. Show that the rate of power loss per unit area by an electromagnetic wave
impinging on a good conductor is

^ oss _ 1 lzr\2
dS 4<r<5|K| ’
where K is the effective surface current,

K = J.
Chapter 43

Cylindrical Coordinates

43.1 2+1 Dimensional Decomposition of Green’s

Function
A waveguide, in which conducting walls force a wave to propagate in a single
direction (say along the £ axis), motivates a consideration of a 2 -f 1 dimensional
break-up of Green’s function for a wave propagating in vacuum with a definite
frequency u>. This is analogous to our earlier discussion in electrostatics. (See
Chapters 13-19.) Furthermore, by deriving alternative expressions for Green’s
function, we are able to obtain additional mathematical properties of Bessel’s
functions.
The retarded Green’s function satisfying (31.29), the inhomogeneous Helm­
holtz equation,

- ( V2 + 0 ) G“'(r >r/) = 4 ir 5 ( r - r ') , (43.1)

is given by (31.43), that is

eiu\r-r'\/c
Gu ( r , r ' ) = |r-lr /| ■ (43.2)

In deriving this expression we treated all directions on the same footing; if

instead we single out the z axis, by introducing a Fourier transform in the Jl
(transverse) space,

G«(r, r') = 4* j ^ e ik- *<'-''S(*, *'), (43-3)

(43.1) is satisfied providing the reduced Green’s function g {z ,z ¡) obeys

+ k± ~ ( 7 ) ] 9^z ' z>’) = s (z ~ *')• (43-4)

449
450 CHAPTER 43. CYLINDRICAL COORDINATES

We have seen this differential equation before; it is (41.41) with e = 1, and its
solution is given by (41.57) with r = 0:

1 ikz\z—z>\ (43.5)
g{*, *') 2 kz

where, for w > 0,

if * > |kx|,
k, (43.6)
*V *i - if 7 < |kxl-

In statics (uj = 0) the second possibility in (43.6) occurs. [See (13.24).] Without
loss of generality, we set r' = 0, and define k = u>/c > 0. Then, comparing
(43.2) with (43.3) and (43.5), we obtain the identity

pikr ieW k2~k1M

—
r
= 4 tr
I (2^)2 2y/k2 - k { ’
(43.7)

where the square root is defined by (43.6). It is convenient to introduce a

cylindrical coordinate system, with

r = (p ,0 ,z), r = \/z2 + p2,

k x =(\,<t>), (43.8)

from which follow

g i k x T j L e i X p COS (f>

(dk± ) = XdXd<¡>. (43.9)

Making use of the integral representation for the Bessel function Jo, (16.4),

J * ’ f ^ iV cos* = /o(Ap), (43.10)

we find, from (43.7), the identity

p ik y / p 2+ Z 2 fO O eW k 2- \ 2\z\
= = t / A dX Jq(Xp ) —■ = = = = -. (43.11)
yjp 2 + *2 Jo Vfc2 - A2

In the static limit (k = 0, y/k2 — X2 —» ¿A), we recover the previously derived

result (16.6),
1
pOO

/ dAJo(Ap)e-Al*l. (43.12)
Jo
43.2. THREE DIMENSIONAL FOURIER REPRESENTATION 451

43.2 Three Dimensional Fourier Representation

An alternative representation for Green’s function can be obtained by taking
the three dimensional Fourier transform of (43.1), which immediately leads to
the formal solution

(dk) e¿k*(r- r')

Gw( r ,r (43.13)
(27r)3 k 2 —lo2/c2 '

The above expression is not well defined since the integrand has a singularity
wherever |k| = w/c, corresponding to real wave propagation. To assign meaning
to this expression, we must specify the boundary conditions. Recall that the
retarded and advanced Green’s functions have the explicit forms given in (31.43),

G ret = |— j|e*w | r-r,| /c, ( 4 3 .1 4 )

G adv = _ L j . e-H r-r'| /e. (43.15)

Mathematically, both signs of i are possible; a consideration of physical condi­

tions is necessary to select the appropriate solution. If we let the wavenumber
be complex, which is the situation occurring in a conductor [for example, see
(42.8)]:
W U\ IU>2 ^ „
------►------1----- , Ui > 0, (43.16)
c c c
the retarded and advanced Green’s functions become, respectively,

G ret ->• l - e iu' Rlce - u*Rl\ (43.17)

Gadv -> I e-< "i* /«e "al*/c> (43.18)

R
where R = |r — r'|. As R —> oo, Gadv increases without bound, while the
physically correct solution, Gret, goes to zero. Thus the device for defining
Gret in terms of the representation (43.13), which builds in the boundedness
requirement, is to replace u> there by a> —►w + i t ) e —> +0:

(dk) eik
Gret(v,r') = 4 * J g (43.19)
)3 j^2 _ ^v+iey
^+0
We can quickly check the validity of this result by explicitly evaluating the
integral using spherical coordinates [recall the spherical average of a plane wave
seen, for example, in (38.45)]:

47vk2dk sin kR
2
(2 tr)3 kR ~
452 CHAPTER 43. CYLINDRICAL COORDINATES

Figure 43.1: Contour for evaluating (43.20).

cos kR
IL
w RdR
jL
J[°°
0
dk
C

1 l__d_ r°° éJkR

dk 2’ (43.20)
■x R d R J _x

since the integrand is even in k. Equation (43.20) can be evaluated by expressing

it in terms of a contour integral in the complex k plane, as shown in Fig. 43.1.
For both u) > 0 and u < 0, we close the contour in the upper half plane, since
the infinite semicircle then makes no contribution (because R > 0). Then by
the residue theorem, we find, because only the pole at k = u)/c-\- ie is enclosed,
11 J JwR/c 1

= = r <43-21>
which is the retarded Green’s function given in (43.14).

43.3 Hankel Functions

We have thus established the equivalence between two representations for ,
namely (43.2) and (43.19),
0i k • r

J
}_eiwr/c _ 47r (dk)
(43.22)
r W £->+0
Note that if we now perform the integral on kZ) and use the integral represen­
tation for Jo, (43.10), we rederive the identity (43.11). [See Problem 1.] If,
instead, we rewrite (43.22) in the form of a 1 + 2 dimensional decomposition,

1 ,iwr/c f°° dkz f (rfkx )

r ”7 - 0 0 2 X J (2 w )2 ¿ 2 + k 2 _ (H ± il) 2 ’
(43.23)

and first integrate over the angle associated with kj_, we find
1 ,iu>r/c _ 4jr f°° Xd\ Jo(X p)
(43.24)
r 2x Jo 2ir X2 _ Í(íü±i¿)2 - ife2
43.3. HANKEL FUNCTIONS 453

which, in the static limit, u> = 0, we have already seen in (18.31).

If we substitute the representation (43.24) into the Green’s function equa­
tion, (43.1), and use the Laplacian in cylindrical coordinates, (16.16), we imme­
diately see that the A integral obeys Bessel’s equation

id 2 Id Jo(Xp)
dXX = 0, (43.25)
\dp2 + p d p + K A2 — K2

for p ^ 0. We may also obtain this result explicitly from the differential equation
satisfied by Jo, (16.17), since

( d2 i d 2\ r dXX Jo(*p)
W + pdp + K ) J 0 A2 — K2
1
l dAAA2 - « 2
(—A2 + K2)J0(Xp) = 0, for p # 0, (43.26)

where we have used (16.50),

[ ° ° dXXJo(Xp) = -S (p). (43.27)

Jo P
With the exception of the origin, here is another solution of Bessel’s equation,
which is called the zeroth order Hankel function of the first kind, defined by

Jo(Ap)
f XdX
A2 _ [ ( » ± i e ) 2
(43.28)

When oj/ c < \kz\, the square root in (43.28) becomes imaginary, — w2/ c 2,
and in this case, the Hankel function becomes the modified Bessel function,
(18.3),
in
Hn1^ ( \ i—r — k2 , Ko (43.29)
k* - ^ P

In terms of the Hankel function defined by (43.28), the equality (43.24) reads

\^iu xr/ c

(43.30)

The representation (43.11) has its p dependence expressed in terms of Jo(A/>),

which is a slowly damped oscillatory function, while the associated 2 depen­
dence is either oscillatory or exponentially damped, depending on whether the
square root is real or imaginary. On the other hand, the form (43.30) represents
the function in terms of oscillatory functions of 2 , while the corresponding p
454 CHAPTER 43. CYLINDRICAL COORDINATES

dependence is either oscillatory or exponentially decreasing. Let us extract the

Hankel function in (43.30) by taking the Fourier transform in 2 :

r°° eiky/p*+z* , ___ v

j ^ d z - j = = e - " > ' = i*H ^ p) , (43.31)

where k = u/c. Then by setting kz = 0, we obtain another integral representa­

tion for the Hankel function,

poo eiky/p2+z2
Í7tH0 (kp) = / dz . . :♦ (43.32)
J-00 \/p2 + z2

We will now use this representation to find the asymptotic form of the Hankel
function for kp ^ 1. As we will justify a posteriori, the main contribution of
the integral comes from small values of z jp ) so that we may expand the square
root in the exponential as

p2 + z2 (43.33)

and approximate (43.32) by

1 f°°
Í7rHQ1\kp) ~ - e ikp / d z e - kz2/2ip
P J-00
ir/4) (43.34)

Here the Gaussian integral is evaluated as

y ” d z e - kz^ 2ip = tt^ . (43.35)

The consistency of our approximation hinges on the fact that the dominant
contribution to the integral, (43.35), comes from small values of z/p, for which

< 1. (43.36)

As an application of (43.34), we may obtain the asymptotic behavior of J q by

taking the imaginary part of (43.28) for kz = 0 and using the representation
given in Problem 43.2a:

Jo(kp) = Re H ^\kp) ~ cos { k p - 7r/4). (43.37)

This is a result, derived in an alternative way in Problem 43.2, which we first

saw in (16.39).
43.4. PROBLEMS FOR CHAPTER 43 455

As with the previous Bessel functions, we can ask whether H ^ ^ k P ), with P

given by (18.2), possesses an addition theorem. We might expect an affirmative
answer in view of (43.29) and (18.64), the addition theorem for K q. Since
satisfies the same differential equation as J0, we can follow the development in
Section 18.1, leading to the addition theorem for J0> (18.55). In particular, we
have the Taylor series expansion [cf. (18.50)]

H ^ \ k P ) = e~r ± * V j -H ^ (k p ), (43.38)

where, similarly to the K o discussion, p' < p because H ^ \ k P ) has a singularity

at P = 0. The introduction of the polar angle $ , (18.52), remains unchanged
as does the expansion of exp(—r 'x •V i ) , (18.53). Consequently, we have
oo

H tf\ k P ) = Y Jm{kp')eim<s>H (0l\kp). (43.39)

771 — — OO

This becomes the addition theorem for H^ (recall p' < p)

oo

H ¡1\ k P ) = Y eim* H £ \ k p )e -im* 'j m(kp’ ), (43.40)

with the introduction of the mth order Hankel functions of the first kind,
-i zfcm
l(d _ .d_
eim4,H£ )(k p ) = imeim* H £ \ k p ) = H ^ ik p ), (43.41)
^ k \dx 1dy
for m > 0 (m < 0), respectively. Notice that since the whole construction is
identical to the Jm construction, Hm^ satisfies the same differential equation,
(16.33), and the same recurrence relations, (16.71) and (16.72), as does Jm.

43.4 Problems for Chapter 43

1. Derive (43.11) from (43.22), as suggested.

2. (a) Using the representation

l i ml m— r- = lim —r--—r- = 7rS(x)

e-+0 X — l€ e~f0 X ¿ + €¿

show that R e H ^ ^ ) = Jo(t)-

(b) Use the result of Problem 2a to derive

sin y jt 2 + x 2
M t) = \ f dx- -------- -
T* J Oo y/t2 + x 2
2 [°°
— I dO sin(t cosh 0)
TCJo
_ 2 f° sin ty
dy
~ *Ji \Jy2 - 1 ’
456 CHAPTER 43. CYLINDRICAL COORDINATES

and hence derive the asymptotic form

Jo(t) ~ s J ^ c o s (t - tt/4 ), (t > 1).

Reconcile the third form of J0(t) above with the fact that Jo(0) = 1.

3. (a) Recalling Problem 18.2, show that

= Jm(kp).

(b) Show that

Im H ^ ik p ) = Nm(kp).

4. In general, Hankel functions of the first and second kind are defined by

= M x ) + iN „(x)
H l2\ x ) = M x ) - iNv(x).

Show from Problem 16.10 that these have the leading asymptotic behav­
iors as x —> oo:

H il>2){x) ~ ( — ^ 1 e±i^ - v*t2- * l A\

Thus, H W correspond to outgoing or incoming cylindrical (or spher­

ical) waves, respectively.

5. Consider unbounded two-dimensional space. Find Green’s function for

Helmholtz’ equation

(V 2 + k2)Gk{ r — r') = —47r5(r — r')

in three ways. (Why is Gk translationally invariant?)

(a) By directly solving the equation in polar coordinates, noticing that

there is no preferred direction, and selecting the solution of the ho­
mogeneous equation that is singular, ~ —2 In |r — r'|, as |r —r'l —►
0, and represents, at large distances, an outgoing wave. [Answer:
M r/T^^lr —r'|).]
(b) Expand Green’s function in terms of eigenfunctions, in polar coordi­
nates, p, <j). [Prove that the normalized eigenfunctions are

where k is a continuous variable.]

43.4. PROBLEMS FOR CHAPTER 43 457

(c) Write, in polar coordinates,

00 1
G *(r-r')= ^ eim(*~4'>Prn(p,p'),
m ——oo

and solve the resulting differential equation for pm, by using the
discontinuity method. Note that the boundary condition for large p
is to be an outgoing wave, so the appropriate behavior there is given
by Hm\kp).

6. Show that the relation between the modified Bessel function K m, and the
Hankel function H ^ } is

K m(z) = y eiirm/2H £ \ iz), -7T < argz < | .

Use (43.41) and (18.65). This result is already given in Problem 18.3.
Chapter 44

Waveguides

The theory of microwave cavities and their application to radar technology was
largely developed at the MIT Radiation Laboratory during World War II. A key
player in that development was Julian Schwinger. In this chapter we consider
some o f the elementary aspects of electromagnetic waveguides, which we will
see is closely tied to the electrostatic mode theory discussed in Chapter 25.

44.1 E and H modes

In this chapter we want to introduce the theory of waveguides, especially relevant
for microwave radiation. We start from the analysis in Chapter 41, where now
we write (41.17) and (41.18), (41.19) as

Bz -V-nxE, (44.1)
ic ^ _ .4tt t
Ez --- V •n x B — i — Jz (44.2)
LUC UJC
VjJ A irm
Ex = ------ ( 1 _l H— r V x V x * n x B - i — V ± J Z, (44.3)
dz c \ u)¿e ) uc
zcoe / c2 \ „ 47T
---- (
I jl + i - V x V x •n x E -------- nxJ, (44.4)
c \ u ¿e J c

where we have used the gradient operator V j. in place of ¿kj_, that is, we
are undoing the transverse Fourier transform. Now in the transform space the
two independent transverse vectors are kj_ and n x k i , so correspondingly, the
trasverse fields Ex and B x can be written in terms of two scalar functions each,

Ex = - V ± V ‘ - V x n l / " , (44.5)
Bx = V x n / '- V x / " , (44.6)

and then, since

n x (V x n ) = V - n n * V = V x, (44.7)

459
460 CHAPTER 44. WAVEGUIDES

the transversely rotated fields are

nxE± = VxnF' - Vj.7", (44.8)

nxBj. = V ± I ' + V x n I". (44.9)

When we insert these constructions into (44.3) we obtain, for the Vj_ and the
V x n components,

4 - V = — ( l + - ¿ - V i ) / ' + «— X (44.10)
OZ C \ L )¿ 6 J LÚC

± v " = * - 1" (44.11)

dz c

We proceed similarly for the dB±/dz equation, (44.4), where we introduce

similar components for J :

nxJ = V x n J ' - V XJ", (44.12)

which are determined by

V x - n x J = - V i J ", (44.13)
- V . J j . = V\J'. (44.14)

The resulting equations are

A ;- = ¿ íü v _ 1l r , (44.15)
dzz e e
d_
-I" = i— ( 1 + 4 - V i ) v " - — J". (44.16)
;
dz' c \ u¡¿e J c

We thus have four coupled differential equations for the four functions I",
V ' , V " . Once those functions are found, the longitudinal fields can be found
from (44.1) and (44.2):

Et = - i — Jz , (44.17)
we ue
Bz = - i ^ 7 \ V " . (44.18)

Because there is no coupling between the sets V* and / 7/, V n we have two sets
of independent modes, characterized by the values of Ez and Bz , respectively.

1. If I n, V " — 0, and hence Bz = 0, Ez ^ 0, these are called the E modes

(or “transverse magnetic,” TM, modes).

2. If V 1 = 0, and hence Ez = 0 (outside the current distribution), Bz ^ 0,

these are called the H modes (or “transverse electric,” TE, modes).
44.2. BOUNDARY CONDITIONS 461

Figure 44.1: Cylindrical waveguide with arbitrarily shaped surface S.

(These correspond to what were referred to as || and _L polarization, respec­

tively, in Chapter 41.) For these two modes, the electric and magnetic fields
are given simply in terms of V7, and V" , respectively, outside the current
distribution, as

E modes: Ex = - V x V ' ,

Sh
ii

$
nxB = V i / ' , Bz = 0 , (44.19)
H modes: BX = - V x / " ,
S3
II
o

nxE = —V x V " , Bz = - * - V \ V ". (44.20)

U)

44.2 Boundary Conditions

When we discussed reflection and refraction in Chapter 41, the transverse char­
acter of the plane waves was exp(ikx * r i), which meant that we could make the
replacement Vj|_ —* —k\. Now, however, we have in mind a 2 + 1 dimensional
situation, translationally invariant in the z direction, but with a perfectly con­
ducting boundary surface S in the transverse directions, as sketched in Figure
44.1. The boundary condition on the surface S is that the tangential component
of E must vanish, or Ez = 0 and the circumferential component of Ex must
vanish. Thus, for the E modes this means that on the surface,

E modes: = 0, V f = constant on S. (44.21)

For the H modes, Ez is already zero, so only the tangential component of Ex

is required to vanish, or
i / . V L V " = 0 on S, (44.22)
where u is the outward normal to the cylindrical boundary.
The various modes are characterized by functions satisfying

( V i + ■y1)<j> = 0, (44.23)

where the eigenvalue —7 2 of the transverse Laplacian operator is the general­

ization of —&2 in the case of unbounded transverse space. The E modes are
462 CHAPTER 44. WAVEGUIDES

characterized by a set of functions { 0 a} which satisfy the condition 7 ^<pa = 0 ,

or

<f)a = 0 on S, 73 0 (44.24)
<pa = constant on 5, 7^ = 0. (44.25)

The H modes are characterized by a set of functions {ip a) satisfying

— ^ = 0 on 5. (44.26)

Thus, for the E modes we write the eigenfunction expansion

V '(r x ,z ) = 'Y^4>a{vx)V^{z), (44.27)

a

I '{v x ,z ) = ^2<f>a(rx)I'a(z), (44.28)

a
while the H modes have the expansion

V "(v ± ,z ) = ' £ /M r ± ) Va (z ), (44.29)

a
I " ( r ± ,z ) = 5 > « ( r i.) /" (* ). (44.30)

If y l = 0 , then the E modes have both Ez = 0 and Bz = 0, and, since both

electric and magnetic fields are transverse in this case, we call it merely a T
mode (or a TEM mode). Since such a mode is both an E and an H mode, we
must have
E = V jl<£ = V XnV>, (44.31)
which indeed implies the required Laplacian equation for 0,

± </>= 0.
V2 (44.32)

The boundary conditions for the T modes are that <j> is constant on 5, or

d<¡) dip
= 0, (44.33)
Os dv
where ds is a tangent vector to the surface. If we write the differential relation
between <j> and ip in Cartesian coordinates,

d , d . d , d
(44.34)
r J = d i* ' s i* =
we see that these are just the Riemann conditions satisfied by the real and the
imaginary parts of an analytic function; that is, because
44.2. BOUNDARY CONDITIONS 463

Figure 44.2: Cross section of waveguide consisting of inner and outer conductor.

we see that
<j>4- = F (x 4- iy) (44.36)

is an analytic function.
How general is the existence of T modes? Because they are characterized
by a harmonic function (that is, a solution of Laplace’s equation), if the region
enclosed by S is simply connected, 4> = constant, and hence there is no electric
field. The proof of this assertion follows from the divergence theorem:

J d S (V j » 2 = j d S V x .{<t>Vx <t>) = £ d s ( t - j l * ) , (44.37)

where E is the cross-sectional area of the waveguide, and C is the circumference.

Because <j> is constant (= <j)c) on the boundary, we conclude

/ d S ( V x <f>f
Js
<f)c f dSV2
x <f>= 0, (44.38)

which implies that Vj_<£ = 0 , (j) = (j)c everywhere in the waveguide.

However, the above conclusion does not hold if the bounding surface is dis­
connected, as shown in Figure 44.2. In this case, <j) can take on different constant
values on the inner and outer conductors. The most familiar example of this
geometry, which permits T modes, is that of a coaxial cable.
Let us consider two simple examples of T modes.

1. For parallel plates, parallel to the x axis, the analytic function is merely
a linear one,

(j) 4- = F (x 4- iy) = —i(x 4- iy)V, or = Vy, ^ = —V x. (44.39)

Note that ó is constant on the surface and

dll’ d
- = - _ v * = o, (44.40)

as should be the case.

464 CHAPTER 44. WAVEGUIDES

2. For a coaxial cable lying along the z axis, we take

<j>+ iij) = C ln(a; + iy ) = C In pelB, (44.41)

where 0 is the azimuthal angle. Thus

<j>= C\np, ip = C0. (44.42)

Again, the boundary condition dij>/dv = 0 is satisfied:

(44.43)
d v ~ dp ~

44.3 Modes
Let us summarize where we are. There are three types of modes: E, H (with
the eigenvalue 7 ^ 0 ) and T (with 7 = 0 ). The modes are characterized by the
equations

( V i + 7 2) { $ } = 0. (44.44)

For each mode, labeled by a, the ¿-dependence is given by the functions / ' ,
(E) or V¡¡ (H), which satisfy the equations

OVLa _ ico ( 7a ° 2 ^ dl'a iU)€ .

E: (44.45)
dz = t ( ] u)2e j ) C dz =
dV ” . - *±T" ÍU)€ yll
H: (44.46)
II

dz c a’ C ( * - u 2e ) a'

Note that if 7 ^ = 0, these two sets of equations are identical, that is, the E and
H modes reduce to the common T mode, where

dVg _ ÍÜJ día %U)€

T: Va, (44.47)
~ d z ~ c ' dz
which implies the second-order equation

(44.48)

the solutions of which are e±tkz with k = Uyfe/c, reflecting the expected speed
of light in the medium, c/y/e.
Let us henceforth confine ourselves to a situation where there are no T
modes. Consider a rectangular cavity, with sides a (along the x axis) and b
(along the y axis). For the E modes, where </> must vanish on the boundaries,
we have
m'KX mry
<f>mn oc sin ^ sin - y - , m ,n = 1 ,2 ,3 ,— (44.49)
44.3. MODES 465

Correspondingly, the eigenvalues are

o /m7r\ 2 /ri7r\ 2 9
(44.50)
Tmn - ( ~ ) + ( t ) - 7lu

a relation noted in Chapter 19. In contrast, for the H modes, the normal
derivative must vanish, so

. m'Kx niry ~ ,
Wmn cos cos , m, n = 0 , 1 , 2 , . . . (m, n not both = 0). (44.51)
a o

In this case the eigenvalue has a smaller bound,

9 /m7r \ 2 /n7T\ 2 9
Tmn = J + ( - j - J > Tio, when a > b. (44.52)

The 2 dependence, for either mode, is characterized by the same differential

equation,

( ¿ + k , - /' ) { r } = 0’ <44-53>
for which the solutions are

e±ik‘ z, kz = \ A 2 - 7 2, Jb = - V e > 7 . (44.54)

c

We see here the appearance of a cutoff wavenumber; if k2 < 7'2 there is no

propagation. To have any propagation at all, we must have

k > -, (44.55)
a

or, in terms of the wavelength A = 27r/fc,

A < 2 a, (44.56)

where a is the larger of the two sides of the rectangular cavity.

Let us look at the longest wavelength mode, Hio- The corresponding mode
function is ^ io oc cos( 7raj/a). Of course, Ez = 0, and because

Ex = VxXnV>, (44.57)

Ex = 0 as well, leaving only

_ . 7TX
Ev oc sin — . (44.58)
a
On the other hand,
„ __9 , nx
Bz oc V I ip oc cos — , (44.59)
a
while
d d TTX
By oc — = 0, Bx oc — oc sin — . (44.60)
y dy ox a
466 CHAPTER 44. WAVEGUIDES

Of course, the above relation between E and B follows directly from Maxwell’s
equation
V xE = ~ B . (44.61)

Now let us consider a circular guide of radius a. The boundary conditions

are

E: <j) = 0 when p = a, (44.62)

H: = 0 when p — a. (44.63)
dp
The solutions to the mode equations are of the general form

= (44.64)

where the boundary conditions are implemented by

E: Jm( 7 'a ) = 0 , (44.65)

H: = 0. (44.66)
The values of the lowest few zeroes for the E modes have been given already in
Chapter 19, see (19.127), (19.129), (19.131), for example:

7 ^ 0 .= 2.405, (44.67)
7 o2a = 5.520, (44.68)
y'n a = 3.832. (44.69)

For the H modes, m = 0 provides nothing new, because [see (16.71)]

J i(x) = (44.70)

so 7o„ = Tin) so
7^ 0 = 3.832. (44.71)
A new situation emerges, however, for m = 1 . We could estimate the zeroes in
this case by using the method described in Section 19.4; a better approximation
is given in Problem 44.3. The result is

7 n a — 1-8412. (44.72)
Thus the lowest mode for a circular waveguide is the H mode with m = n = 1,
H u. The cutoff wavelength for such a waveguide is Ac = 3.4125a.

44.4 Problems for Chapter 44

1. In regions where there are no currents, show that for E modes,

E = -y-rVx ( v x - / ' n ] ,
e(z) V « /
B = Vx(J'n),
44.4. PRO BLEM S FO R C H A P T E R 44 467

whereas H-mode fields are given by

E = -Vx(K"n),
B = V x ( V x — V "n ) .

2. This problem refers to a coaxial waveguide. Let the inner radius be 6 , the
outer radius a. Assume a —b <C a, and find the cutoff wavenumbers 7 ',
7 " for the E and H modes. [Hint: You do not need Bessel functions. Look
at the differential equations for the mode functions. Which mode has the
lowest cutoff wavenumber (other than the T mode)?]
3. This problem illustrates an approximate method for finding the lowest
cutoff wavelength for a circular waveguide, that is, for finding the first
zero of J [.
(a) Derive the following formula for the eigenvalue from the differential
equation satisfied by the mode function ip from (25.26):

I o dtt (ít'f’ W Y -f rn¿

(7 a ) 2
fo d tt ( i/> ( t ) ) 2

Notice that if this is varied with respect the ip we recover the differ­
ential equation

d2 Id / X2 m2
ip = 0.
5 ? + r s + (T a ) - - F

(b) We obtain an estimate for 7 by choosing a trial function which sat­

isfies the appropriate boundary conditions,

V'(o) = o, v-/(i) = o.
(Where do these boundary conditions come from?) [Actually, the
estimate so obtained is an upper bound to 7 . See Chapter 25.] The
function ip(t) = t ( 1 —t 2/ 3) satisfies these boundary conditions; show
that if it is inserted into the formula for 7 it gives
7"a « 1.842265.

(c) Show that if, in the spirit of Section 25.2, we now write the differential
equation as
1£ 1_
'tp — —(7 a ) 2ip « —(7 a)2
d t2 t dt t2

we can approximately solve for ip¡ up to an overall factor, as

' 0 « 2 2 t - 9 t 3 + f 5.
468 CHAPTER 44. WAVEGUIDES

Put this in the formula for 7 to find

7 "a = 1.841198,

in remarkable agreement with the exact answer, 1.841184.

4. Show generally that in source-free regions the electric and magnetic fields
can be represented by two Hertz vectors II and II* ,

e = vx(vxn) + —vxn*,
C
ÍüJ€
b = -----vxn
c
+ vx(vxn*).
5. Show that the fields in a cylindrical waveguide are represented by Hertz
vectors where

n - Ez
“ fcf - ew2/c2 ’
n* Bz
n &f — ew2/c 2

What is the connection between II, II* and the functions l7, V7, V nl

6. A resonant cavity consists of a perfectly conducting circular cylinder with

conducting endcaps. Calculate the resonant frequencies in terms of Bessel
functions, and the explicit form of the fields for the lowest H and E modes.

7. A perfectly conducting cavity consists of a region of empty space V bound­

ing by a conducting surface S . Consider the radiation gauge. Define mode
functions A a , which satisfy

V X (V X A«(r)) = A„(r), V . A «(r) = 0,

with the boundary condition

n X A a(r) = 0 on 5,

where n is the normal to S. If we use real mode functions, show that the
following orthonormality condition holds,

f (dr) A a(r) • A ^ r) = 8ap.

Jv
Show that these mode functions are complete, for divergenceless functions,

£ A « ( r ) A a(rO = [15(r —r')]x,

a
where the latter notation means that

V •[15(r — r')]x = 0 .
44.4. PROBLEMS FOR CHAPTER 44 469

8. Now let the vector potential be given in terms of the current by a Green’s
dyadic G ,
A (r,w ) = / ( d r / G ^ r 'j w ) ♦-J (r ',w ).
Jv c
Show that, in vacuum,

V X (V X ) - 0 ) j G (r ,r ';w ) = 47r[16(r - r')]x-

9. Prove that
/ \ . v -' A a(r )A 0 (r')
G (r, r ;o>) = 4 tr > ,2 ’ ,\ 2 •
“ ¿ 2 _ ( w / c )2

Show that, with u —>u> + Oi, the resulting Green’s dyadic is the retarded
function,

0, t - t ' < 0,
G (r,t;
,t] r ' , 0 = | 4 x ^ 0, A a (r )A a (r/)^-sin[fcac ( f - t ') ] , t - t ' > 0.

10. A current J(r, i) is turned on and eventually off. Demonstrate that the
total amount of electromagnetic energy created in the cavity is

A E = /( d r ) dt (dr') •^ -G (r, ^ r ',^ ) •- J ( r , , / /).

J C uv C

11. Show that A E can also be presented as

i p 2
A E = 27T^ / (dr) dt A a(r)etkoiet ♦J(r, t) .
IJ

12. A resonant cavity, composed of a homogeneous dielectric medium enclosed

by perfectly conducting walls, allows an infinite number of electromagnetic
oscillation modes. Prove that the wavenumber associated with a mode is
given by

2 _ /(d r ) ( V x E ) 2
or
~ /(d r ) E 2
2 _ /(d r ) ( V x H ) 2
" /(d r ) H 2

Verify that the first expression is stationary with respect to variations that
do not violate the boundary condition, while the second form is stationary
with respect to unrestricted variations.

13. A resonant cavity is constructed from a uniform waveguide by placing

perfectly conducting walls at the planes z = 0 and z = L. Determine the
470 CHAPTER 44. WAVEGUIDES

characteristic wavenumbers of the cavity in terms of the cutoff wavenum­

bers of the various guide modes. Prove that the skin effect dissipation Q
for a mode of this cavity is related to the Q of the guide by

1
H mode :
V cavity Q guide
1 1
E mode :
Q cavity Qguide

where 6 is the skin depth, and k is the guide propagation constant for
the mode in question. [For an oscillator, Q is defined as the ratio of the
resonant frequency to the dissipation constant, Q = ujq/j . Here, this is
27T times the average energy stored in the cavity divided by the power lost
per cycle.]

14. (a) A waveguide is filled with a homogeneous dielectric of permittivity

€\ in the half space z < 0 , and a homogeneous dielectric of permit­
tivity 62 in the half space z > 0. Prove that the decomposition into
independent modes is still valid. Discuss the effect of the dielectric
discontinuity on the modes.
(b) A dielectric medium of permittivity 62 and length L is inserted into a
waveguide between two other media of permittivity c\ and 63, respec­
tively. Determine under what condition waves will be transmitted
without reflection from medium 1 into medium 3.
Chapter 45

Scattering by Small
Obstacles

When a plane electromagnetic wave interacts with a macroscopic dielectric, we

talk of reflection and refraction. For the interaction of the wave with a small
object, the appropriate word is scattering. Here, small refers to the size of the
object in comparison to the wavelength of the radiation. In this chapter we
will consider the elementary theory of scattering— in the following, a systematic
approach will be given.

45.1 Thomson Scattering

As a first illustration, we consider a single charge e, of mass m, which accelerates
due to an applied electric field such as is present in a light wave:

dv
(45.1)
mS = eE"
(as long as |v/c| <C 1). An accelerated charge radiates, which gives rise to
the “scattered” radiation; the power radiated is given by the Larmor formula,
(32.25),

<*•»
This scattered power is to be compared to the incoming power per unit area of
the plane wave, which is, in vacuum, given by the magnitude of the Poynting
vector,
|S| = ¿ | E X B | = ¿| E | 2. (45.3)

The ratio of these two quantities defines the cross section,

Fscatt
= (45.4)
|s| ’
(T

471
472 CHAPTER 45. SCATTERING B Y SMALL OBSTACLES

which is the effective area presented by the obstacle to the wave. For the
situation considered here, the cross section is

cr (45.5)

the so-called Thomson cross section. For an electron, this cross section is con­
veniently expressed in terms of the length, ro,

g 2

r 0 = — = 2 .8 1 7 9 x l(T 13cm, (45.6)

me¿
which, for historical reasons only, goes by the name “classical radius of the
electron.” Numerically, the Thomson cross section for the electron is about

a « - X 10 24cm 2 (45.7)
o
What is the angular distribution of the radiation scattered by the charge?
We recall the nonrelativistic expression (32.23),

d-Pscatt
(45.8)
d il

which gives the power radiated in the direction n per unit solid angle. Using
(45.1), we rewrite this as

dPsc 1
r(n X E )2, (45.9)
dfi 4tt m 2 c3
or, introducing e as a unit vector along the direction of E,

dPes catt 1
E 2[l — ( n « e ) 2]. (45.10)
dfi 47Tm 2 c3

There are two distinct possibilities here: If e is perpendicular to the scattering

plane defined by n and the initial direction of propagation, then n * e = 0. If
e lies in that plane, then |n*e| = sin0, where, as Fig. 45.1 shows, 9 is the
scattering angle. Therefore, for the two choices of e, the factor in the scattered
power is
1, e L scattering plane,
1 — (n •e (45.11)
cos2 0 , e ||scattering plane.

Notice that there is no scattering at 0 = n/2 for the second choice, which just
re-expresses the fact that there is no radiation emitted in the direction of the
acceleration. If the incoming radiation is unpolarized, we have the average of
the two possibilities given in (45.11), so that

1 + cos2 9
1 — (n •e ) 2 (45.12)
2
45.2. SCATTERING B Y A BOUND CHARGE 473

Figure 45.1: Scattering of a wave polarized in the scattering plane.

The differential cross section, the effective area for scattering into a given ele­
ment of solid angle, is defined by

d(T d P SCQ.ttl d il
(45.13)
dñ= fsj ’
For an unpolarized wave, the differential Thomson cross section is then

dcr / e2 \ 2 1 + cos2 6
(45.14)
dñ = \ ^ J 2 *

The total Thomson cross section, (45.5), is recovered when (45.14) is integrated
over all solid angles,
a= [ ^ d S l. (45.15)
J dS2

45.2 Scattering by a Bound Charge

The preceding section described scattering by a free particle. We now suppose,
instead, that the “electron” is bound by a damped harmonic oscillator force. As
we discussed in Section 5.2, this interaction is characterized by a natural fre­
quency, u>o, and a damping constant, 7 . The corresponding equation of motion
of the electron is given by (5.16):

ip j» »
m — r + mu or 4- m y — = eE. (45.16)
at¿ at
Suppose the electric field is due to an incoming light wave of a definite frequency
w,
E (t) = Re Ee- *"* = \ (V e~iut + E V 'wt) , (45.17)
it
so the deviation of the particle from the center of the harmonic force is given
by (5.18):
474 CHAPTER 45. SCATTERING B Y SMALL OBSTACLES

The corresponding radiated power,

2 e2 s2
-Pscatt —3 (45.19)

involves
e„ w2 Ee~iut
r = Re 9 9 (45.20)
m cuq —u)¿ — %yu)

We are interested, not in the instantaneous scattered power, but rather in its
time average over one cycle. For a typical time varying quantity,

A (t) = Re A e ~ iwt = U A e ~ iu,t + A*eiut), (45.21)

Z

the time average of its square is

(A(t))* = t\A \\ (45.22)

Thus, the time averaged power scattered is, from (45.19) and (45.20),

2 e2 1 e2 |EP
P sca tt — ^ o ~ (45.23)
scatt 3 c3 2 m 2 U (u>2 — w2) 2 + ( t ^ ) 2 ’

while the average incident energy flux is

(45.24)

both expressed in terms of the complex amplitude E. The resulting cross section
is
_ 87t Í e2 or
(45.25)
a 3 \ me 2 - w2) 2 + ( 7 U>)2 ‘
This reduces to the Thomson cross section for a free particle when both u>o and
7 are zero. More realistically, the above cross section approaches the Thomson
limit,
*-*<■> T h , (45.26)

when the frequency of light is large compared to the natural frequencies of the
bound system,
u > » a ; 0,7. (45.27)

Physically, in this limit, the time scale is set by the period of the incident light;
if this is very short, the electron does not have time to be influenced by internal
forces so it behaves as though it were free.
In the opposite limit of very low frequencies,

w < w0) (45.28)

45.3. SCATTERING B Y A DIELECTRIC SPHERE 475

the cross section (45.25) approaches

(45.29)

which is called the Rayleigh cross section. In this domain, the highest frequencies
are scattered the most strongly. This frequency behavior is presumably one of
the reasons the sky is blue.
We have the situation of resonant scattering when uj = u>o, at which point
the cross section becomes
(45.30)

For small damping, u>o 7 , (45.30) is much larger than the Thomson cross
section. Of course, 7 cannot be zero; energy is necessarily dissipated because, if
for no other reason, the particle radiates due to its acceleration (recall Section
35.3). We will return to this issue shortly.

45.3 Scattering by a Dielectric Sphere

We do not require a microscopic description in order to calculate the scattering
of electromagnetic waves by small objects. The scattering by such an object
may be calculated from the dipole radiation formula (32.30),

(45.31)

As an example, recall from (23.104) that a dielectric sphere, of radius a, in a

static electric field E acquires a dipole moment

(45.32)

This formula is applicable to a radiation field whenever E varies slowly on the

scale set by the radius of the sphere, that is, if

A> a (45.33)

where A = c/u. Then the radiated power is

(45.34)

so, by (45.24), the total cross section for scattering is

(45.35)

Again we see the uj4 dependence of the cross section characteristic of the scat­
tering of long wavelength radiation.
476 CHAPTER 45. SCATTERING B Y SMALL OBSTACLES

45.4 Radiation Damping

We now return to the oscillator and consider the damping due to radiation. From
(33.9), the radiation reaction force on the charged particle can be identified from
P = —F •v to be

F=|^v. (45.36)

For a definite frequency, this becomes

2 e2 2
“ 3^ v = ~ m> v , (45.37)

where here yr is the radiative part of the dissipation, which must be present
in order that energy be conserved. It coincides with the previously determined
damping constant for radiation by a harmonic oscillator, (35.56), as long as the
characteristic radiated frequencies are near the resonant frequency u>q. Recog­
nizing that there may be other forms of dissipation (for example, collisions), we
write the total dissipation constant as

2 e2
7 = x — 3 W2 + 7d = 7r + 7d . ( 4 5 .3 8 )
ó mc°

The energy removed from the incident electromagnetic field feeds both j r and
jd . The rate of energy transfer from this field is, according to (45.18),

(eE •v) = eR e(E e-‘wt) •Re f — ^ ). (45.39)

\mu>5 —u>2 — tju y y

Generically, the time average of such a product is given by

ReA (t)R eB (t) = - (Ae-™* + A*eiUfi) - (Be-™* + B*eiwt)

z z
= \(AB* + A* B) = eA*B, (45.40)
z

implying the power transferred to the oscillator,

e2 1 , —VjJ
( e E « v ) = — -| E | 2 R e ^2-------2----- •— * (45.41)
m2 uJq —u>¿ — «70 ;

Consequently, the total power removed from the incident field (not just scat­
tered), is
yu>
(45.42)
(u,2 - W2) 2 + (Tu;)2 ’
which must exceed the scattered power, (45.23),

Ptot ^ P scatt • (45.43)

45.4. RADIATION DAMPING 477

The corresponding total cross section is

PtOt _ e2 yu>2
(45.44)
|S| atot * me {u l - w2) 2 + (y u )2

If we use (45.38),

(45.45)

we can rewrite (45.44) as

_ A 3 2 7r7
(45.46)
a to t ^ 2 ° (u >1 — a ;2 ) 2 + (7 a ;)2 '

At resonance, uj = u>o> the total cross section becomes

0-tot (^ 0) = 471-- (45.47)

and therefore if we write c / lüq = Ao, and note that

^ < 1, ( 4 5 .4 8 )
7 ~

we have the inequality

0 -t o t ( ^ o ) < 67 tA q . ( 4 5 .4 9 )

We now break up Otot into two pieces corresponding to the two channels for
energy loss, given in (45.38),

0-tot — 0"scatt “h 0"diss) (45.50)

where the scattering cross section is identified as

0"scatt — 47T-C
2 (a)l - u>2) 2 -f- ( 7 a;)2
__ 87t / e2 V LO
(45.51)
3 \mc2) (w2 “ ^ 2) 2 + ( j v ) 2

[which is the same result found in (45.25)], while the dissipation cross section is

7r7d
0rdiss = 47T--C (45.52)
2 (u>q —oj2)2 + (Tw) 2

If we ignore 7 , and consider the high frequency limit u u>o, the scattering
cross section becomes
1 0rTh
0-scatt — 0”Th (45.53)
1 + ( 7 /w ) 2 l + [(2/3)(e 2/m c 2 ) ( l /A )]2 '
478 CHAPTER 45. SCATTERING B Y SMALL OBSTACLES

If X e2/me2, the correction to the cross section is very small. But if we were
to take literally the limit o f very short wavelengths,

A< = r0, (45.54)

mcz

the cross section would behave as

8 tt 2
tfscatt ~ — - — ” 2 = ^7r^2> (45.55)
( l r07)

which has the same form as the upper bound (45.49). This limit is completely
unbelievable, however, since quantum effects become significant already at the
considerably larger distance

h e*
= 137- 2 (45.56)
mc me *

45.5 Problems for Chapter 45

1. Calculate the scattering cross section by a small, perfectly conducting
sphere of radius a «< Notice that the result does not coincide with the
e —►oo limit of (45.35), for the latter refers only to electric dipole radiation,
while for a perfect conductor magnetic dipole radiation is important as
well. See Problems 28.2 and 46.3.
Chapter 46

Partial-Wave Analysis of
Scattering

A general approach to scattering problems is the so-called method of partial

waves, in which there is a decomposition into modes of a definite angular mo­
mentum /. This technique is most commonly seen in a description of quantum-
mechanical scattering, but we will see that the concept proves useful in the
classical domain of electromagnetic scattering as well.

46.1 Mode Decomposition

We apply the formalism of Chapter 44 to the case of scattering. We begin by
making a 2 + 1 dimensional break up for a spherical geometry, where the trans­
verse directions are represented by the spherical angles 0, <t>) and the longitudinal
one by the radial coordinate r. The Maxwell equations

V X B = —i —eE + — J, —V x E = —i —B, (46.1)

c c c
are projected into transverse and longitudinal parts by taking the cross product,
and the dot product, with r, respectively. Thus, for example, we easily see
r\
r x ( V x B ) = V (r *B) - B - r ~ - B , (46.2)
or
or, since this is purely transverse,

r X ( V x B ) = V ± ( r . B ) - B j _ - r — Bj_. (46.3)

If we then divide by r, the transverse parts of the two Maxwell equations above
are
_ „ id 1\ _ vjj r „ 47r r ,
1V ±B r — l —— |— ) Bj_ — ------c- x E H-------- x J i , (46.4)
\or rJ c r c r

479
480 CHAPTER 46. PARTIAL-WAVE ANALYSIS OF SCATTERING

„ IU v
Ex = -----xB. (46.5)
c r

The longitudinal components are simplified according to

i r . V x B = i r x V * B = r x V « -B = — V x - «B = — V •- xB, (46.6)
r
to give

~ r ^ . 4tt t
—V i •- x B — ----- cEr -\----- J r, (46.7)
i

r c c
__ r „ iu ^
- V x •- x E = — B r . (46.8)
r c
We can combine these equations to obtain

iue i' c2 „ \ 47T r

1i + - t V x V x • x E x -------- xJ, (46.9)
c '^ w2e y c r
iu
c
(ix+ _ _ V x V x )
As we did in the cylindrical geometry o f Chapter 44, we eliminate the trans­
verse electric and magnetic fields in favor o f the scalar functions V and I :

Ex = - V x V " - V i X - r , (46.11)

Bx = V x X - Z '- V x /" , (46.12)

r
and write
-XJ = V xX -J ' -V x J ". (46.13)
r r
Now using
— V x = V x - --------V x , (46.14)
or or r
we obtain the following equations for the modes:

^ i 9 ¿w A c2 o\ w i4w _
E modes: -F = _ (^1 + _ V i J / + — X, (46.15)

I -;' = " V _ H r , (46.16)

or c c
1-. ,-,2 r/ ¿47T _
= ----- V i / ' --------- Jr, (46.17)
we we
H modes: - Í j " = H i A + - ¿ - V i ) V " - — J", (46.18)
or c \ J c
®_v" - '— V (46.19)
dr ~ c ’
Br = 3 V lV ". (46.20)
W
46.2. INTERIOR OF CONDUCTING SPHERE 481

The characteristic functions of the transverse Laplacian operator,

1 _ i_ J l ( . a d i d2
vl = sin 9 89 \Sm°d9
+
sin2 0 d<f)2m
(46.21)

are the spherical harmonics,

(46.22)

So we write, for example,

V'(r,e,<t>) = J2Yi™(0><f>)VL(r), (46.23)

lm

and similarly for V , V ", I". Then the mode equations become

^2 ¡a _l i\\ ¿4 ir
E modes: = 7 (l , + (4624>

(46.25)
¿ U ) = 7 t W - 7 4 ( ') -
C2 / (/ + 1 )
H modes: ~ ~ ~ f l ---- " (46-26)
dr ‘ v ' c V w

(46.27)

Let us see what the differential equations are when we are outside the sources.
Let us set J (r) = 0 and take e to be a constant. Then we can turn the above
coupled system of first-order equations into the second-order one,

or / ( '+ ! )
+ T 6' (46.28)
dr2

46.2 Interior of conducting sphere

In the interior of a perfectly conducting sphere of radius a, filled with material
of dielectric constant 6 , the boundary condition that must be satisfied is

E x = 0 for r = a. (46.29)

This implies that

(46.30)
is

'S'
is

'’S'

o'
o

II
II
u

II
II

or

H modes: V L (r = a) = 0 , (46.31)

E modes: dp —0 (46.32)
d ¿I,m r = a
u ~
482 CHAPTER 46. PARTIAL-WAVE ANALYSIS OF SCATTERING

The solutions of the differential equation (46.28) are rji(k r), where j¡ is the
spherical Bessel function, and k = u 2e/c2. For the two modes, then, the fre­
quencies are determined by
H mode: ji(ka) = 0, (46.33)
d
E mode: — (rjA k r)) = 0. (46.34)
dr
Let us make quick estimates of the characteristic frequencies. First, in the
spirit of Chapter 19, we recall the asymptotic behavior
cos(t —(/ + 1)7t/2)
3l{i) ~ oo, (46.35)

so the H mode zeroes occur approximately where t = 7 ",

T " « ( / + l)| + (2 n -l)| , n = 1 ,2 ,3 ,..., (46.36)

where we are never concerned with l = 0 because then V^Yoo = 0. Thus, the
lowest H mode has
k"a = 7'/=1}n=1* y = 4.71. (46.37)

And, for E modes, where it is the zero of ^ tj\(t) that is sought, we have ap­
proximate zeroes at

Í = (/ + l)^ + w r, n = 0 , 1 ,2 ,.. (46.38)

so the lowest mode is

(46.39)
CO

k'a — 7 / - i >n=0
4^
II

II

Let us improve on these estimates using the methods of Chapter 25. The
eigenvalue functional corresponding to the differential equation

*(/ + !) = 0 (46.40)
r2

is

(46.41)

The small t behavior of the solution is u ~ t1+1. For the H modes, consider the
test function, for / = 1 ,
u = t2{ l - t 2). (46.42)
When this is inserted into the functional (46.40) we obtain the upper bound
7 " = 4.74, which is worse than our previous estimate. (The exact answer is
4.49341.) For the E mode we can try

(46.43)

for this has vanishing derivative at t = 1. This does give a considerably improved
bound, 7 ' = 2.7514, compared to the exact root 2.74371. In Problem 46.1 these
bounds are improved by iteration.
46.3. SPHERICAL HANKEL FUNCTIONS 483

46.3 Spherical Hankel Functions

The fundamental solution to Helmholtz’ equation,

(V 2 + k2)<j> = 0, k2 = ^ j , (46.44)

is the Green’s function

e «*|r—r'|
G 0(r - r') = (46.45)
|r — r'| ’
as long as r / r'. For r < r' we can write this as a Taylor expansion,

e«*|r—r'| _ r . V ' e’ fcr'

------------------------- ----- g 1 V ---------------_ (46.46)
|r — r'| r'

Now the operand here itself satisfies Helmholtz’ equation,

(V ' 2 + fc2) — 1 = 0, r' > 0, (46.47)

so, in the spirit of Chapter 18, we can regard zV'/Ar as a symbolic unit vector
when acting on etkr /r'. Thus, we rewrite the Taylor expansion (46.46) as

p«A;|r—r'| A k r'
e _ eikr*(iV /k)C___ (46.48)
r —r

Now recall the expansion of the plane wave in terms of spherical Bessel functions,
(21.62), or (21.73),

OO l / \
eik •:' = 4 ttJ 2 E Y *m 0 ) Ytm ( T i'ji(kr). (46.49)
J=0 m ——l ' '

When this is used to express the operator exp[z¿r •¿V//k] we have

e ¿fc|r-r'|
rpf = T , 4’ ‘'M tr)Y ^ (S) r lm (iv ') (46.60)
|r• - r 'l ~ Im
What is the latter symbolic operator? We recall the generating function for the
spherical harmonics, (20.25),

i = <46-51>
(a

lm

where a is a null vector, a •a = 0. It is easy to see that

( a . i v y eikr> _ ( a . f r ') ' f ] _ ± \ ' e ^ '

(46.52)
/! /! r' dr' J r'
484 CHAPTER 46. PARTIAL-WAVE ANALYSIS OF SCATTERING

Thus, we recognize, when acting on etkr'/r', that

/ i .\ / r ^i
V/m
( H =G) r"n”(?) (G¿) ■ <46 53>
and so we have the expansion
e ik\r-r'\ J gikr1

lr '
(46.54)
Let us define a set of functions by

(46.55)

these functions, which are defined analogously to the spherical Bessel functions
ji (see Problem 21.3), are called the spherical Hankel functions. As with the
cylinder functions we have the decomposition into real and imaginary
parts,
h i ( i ) = ji ( t ) + irn(t), (46.56)
the imaginary part being the spherical Neumann function. Thus we have de­
duced the addition theorem

/¿0(&|r - r'|) = 5 ^ (2 / + ^)Pi(^osy)ji(kr< )hi(kr> ), (46.57)

i
(7 being the angle between r and r') the real part of which is

io(*|r - r'|) = Smk ^ T

_ * = J p 2/ + 1)^ (cos7)Ji(k r < )ji(k r > ), (46.58)

the spherical analog to the addition theorem for the cylinder functions, (16.69).
We might also note the leading asymptotic behavior of the spherical Bessel
function:
Jt «(t —(/ + 1 ) tt/ 2 )
h,(t) ~ ( -* ) — = * » 1, (46.59)
it t
the real part of which coincides with (46.35).

46.4 Scattering
We set the stage for our discussion by recalling the result from the previous
chapter for the scattering by a dielectric sphere. In particular, if we take the
limit e —►oo of (45.35) we would expect to recover the situation of a perfectly
conducting sphere, which limit gives

87r a 6
(46.60)
46.4. SCATTERING 485

However, the well-known result for the scattering cross section by a perfectly
conducting sphere is that of Gustav Mie and Peter Debye (1908-1909),
IOtt a 6
cr = (46.61)

What is the reason for this discrepancy? It is simply that in this case there is also
magnetic dipole scattering, which gives the additional contribution 27ra6/3A4.
(See Problem 45.1). So this poses a nice problem for our general formulation.
Consider a circularly polarized electromagnetic wave propagating in the 2
direction, for which

Em= eikz, Ey = ieikz, k = — = \. (46.62)

A A

The time dependence, e~ZU)t, is implicit. The corresponding magnetic field is

B = z x E , or
Bx — —Eyi By — Ex • (46.63)
Of course, Ez = Bz = 0. We wish to calculate the scattering from a spherical
conducting shell, so it is natural to use spherical coordinates,

Er = l . E = Í ± Í Ü , ‘ *«. (46.64)

(46.65)
r r
We express this in terms of spherical harmonics and spherical Bessel functions
as follows: First we recall the expansion of the plane wave, (21.62),

etkz _ ^P(2/ + \) i]j^k^P^cosO), (46.66)

i=o
so, using the addition theorem for the spherical harmonics, we have

(* + iy)eikz = J ( ~ - + i A - ' j 2 2 4t t (k)y,m (r) (46.67)

kj_=0

We note that
kx d f(k )
m 0, (46.68)
dkx kx=0 k dk kx= 0
so in taking the derivative in (46.67) we can neglect its action on ji(kr). In
working out

á T + ' Jdkv
-V -™ ® (46'69)
we are looking for stuctures like (kx + iky)/k = sin Oe1^ in Y/m(k), so that picks
out m = 1 ; from (20.30) we therefore have only

¡21+1 i /+i
k,-ik, ( d \ ‘" (r-iy^+iy

(46.70)
486 CHAPTER 46. PARTIAL-WAVE ANALYSIS OF SCATTERING

where the fi = 1 condition comes from setting kx = ky — 0 after the differenti­

ation. Thus the required derivative is

121 + 1
(
1
1(1 + 1). (46.71)
ky 4ir
Thus, the radial electric and magnetic fields are

Er = iBr = — ¿ - ^ - ^(21 + 1)1(1 + l)Yn ( f ).

13 (46.72)
i=i kr
Now we return to the formalism o f Section 46.1 and write, for e — 1, using
(46.17),
Er = - ¿ V i / ' , (46.73)

where the effect of Vj_ on a spherical harmonic is to divide by r 2 and multiply

by /(/ + 1 ). Therefore,

T - ¿ V W T / i ( r ) ^ / + ^ r j¡(k r ), (46.74)

which represents an incoming wave. Also, since Br = —(z'/&)V 2 V7' , (46.20), we

have also
(46.75)
II
1

The boundary conditions are that

Ej. = - V XV" - V x X - V " = 0 at r = a, (46.76)

r
from which follows
Vf = 0 , V " = 0 at r = a. (46.77)
Because we further have (46.16),

I - / ' = ikV\ (46.78)

or
we have
d .
— I = 0 at r = a. (46.79)
or
Let us consider the H modes, where in order to satisfy the boundary condi­
tion we must have not only the incident wave, but the scattered wave as well:

21+ 1 ji(ka)
V n = —i ^ V+7rilYn(r) rji(k r) — rhi(kr) (46.80)
Ki + 1 ) hi(ka)
i=i
hi represents the outgoing spherical wave. Let us express the coefficient of the
scattered wave,
jl(ka) _ ji(ka)
(46.81)
hi(ka) ji(ka) + ini(ka) ’
46.4. SCATTERING 487

in terms of an angle 6¡f} defined by

ji(ka)
tan 8" (46.82)
ni(ka) ’

so that
ji(ka) //e i6i
—i sin 6f/ (46.83)
hi(ka)
The scattered wave, the second term in (46.80), is then expressed asymptotically
by the large t behavior given in (46.59):

ikr °° ___ / o/ i i
Kc ~ - i — E (i)y sin 6¡'eis" . (46.84)

Similarly, for the E modes, the mode function which satisfies the appropriate
boundary condition, and therefore contains both the incident and scattered
waves, is

2/ + 1 (rji(kr)Y\r=a 1
E V ^ i'Y n (r) rji(kr) — rhi(kr) (46.85)
i=i ¡(1 + 1 ) (rA,(A:r)),|r=<1J '

As before, we define an angle in terms of the coefficient of the scattered wave,

= *“ *». (46-86)

and then, exactly as before, we find

ikr 00 / 2 / 4- 1
4 ~ Kc ~ — E (f ) y j y f j j sin • (46.87)

The transverse electric field is now given by

Ex = - V ± V ' - V x x V ' , (46.88)

which consists of an incident part and a scattered part. The radial component
of the Poynting vector is proportional to

Sr = - •S oc R e - •E* X B = |E± |2, (46.89)

r r
while the incident flux, which is in the z direction, is proportional to

^c(inc) OC |Marinej 4" l^yincj — 2 . (46.90)

The total scattered power is proportional to

2
Pec oc r 2 / dft|E± |2 = r 2 J dfi|V xI/ , + V i X ^ r (46.91)
488 CHAPTER 46. PARTIAL-WAVE ANALYSIS OF SCATTERING

where by partial integration we may show

j < m v ± v '* • Vi x ^v" = o, (46 .92 )

and further

J d n ^ x V ± V " * - ^ X V X V " = J d Q V ± V "* ' V XV ". (46 .93 )

B y another partial integration we obtain

Psc oc r 2 J dQ [V,/* ( - V i ) V r/ + V " * { - V \ ) V " ] , (46 .94 )

in which, for each angular m om en tu m m od e, we m ay now replace the transverse

Laplacian b y its eigenvalue —/( / + l ) / r 2. Using the above expressions for Vjc
and in term s o f the phases 6{ and 6{'f and exp loitin g the orth og on a lity o f the
spherical harm onics, we get the follow in g expression for the scattered pow er,

1 00
psc OC J 2 y ' / 4 ,r ( 2 J + l)(s in 2 ó'i + sin2 6 " ). (46 .95 )

T h e cross section is obtain ed by dividing this b y |Exinc|2 = 2 :

9 _ 00
a = T 2 ' H i 21 + W 85» 12 s! + sin2 8" )- (46 .96 )
K 1=1
N ow we want to look at tw o special cases, where ka = a/\ is either m uch
sm aller, or m uch larger, than one. In the form er case, where ka <C l , the
dom inant term is / = 1, for which

. /iX d sin t t
(46 .97 )
■»<'> = n t < < 1 ’
/fv d cost 1
(46 .98 )
" , ( , ) = d« 1 * ( >’ , < i '
so
tan 8 " « 8 " « — i t 3 = —i ( & a ) 3 . (46 .99 )
o 0
Sim ilarly,
w _ (< V 3 )' _ 2 3
(46.100)
W l -(_ !/< )/“ Z ’

°r 2
«í = ¡ ( k a f . (46.101)

If we add together the contributions o f the E m odes and the H m odes, we find
for the cross section in the long wavelength lim it

2 tt„ [ / % , A 2 /1 n3\ 21 10 a6 ,
(46.102)
17 ~ W 3 (3 (U>) + ( f (fca) J = T F '
46.4. SCATTERING 489

the result stated in (46.61).

In the opposite limit, ka 1, we use the asymptotic forms of the spherical
Bessel functions,

j t(t) ~ sin(* - ^ V 2) _ cos ft — * 0 + l ) / 2 ) ^ t > (46.103)

í i
„ ( <) ~ Sin(‘ - ’r ( l + 1 ) / 2 ) = - 0 0 8 ( 1 O l , (46,104)

Thus, for the H modes, we have

t»n f f = . TO =$■ 6}' = — I ka — , (46.105)

COs(í —7TÍ/2)

and, for the E modes, we have

tan 6; = - 8in(|- ’r(i, + ; )/ )2> => s; = - (k a - l2±lT) = S? +

' cos(^ - 7r(/ + 1 ) / 2 ) ’ * V 2 2 /
(46.106)
So in the short wavelength limit, the cross section is formally divergent:

OO

a = ¿ ( 2/ + ^ (co s 2 8" + sin2 si ) = (46.107)

In reality this is an erroneous conclusion, because the asymptotic expansion is

only valid when t l. We can estimate the effect by cutting off the sum at a
value of L » 1 such that L 2 ~ ¿2, as we can see from the differential equation,

+ = (46 108)

Therefore,

<T~ pr ¿(2/ + !) ~ % £ dl 21 = f L 2 ~ 27ra2. (46.109)

This result is not unexpected. In the short wavelength regime we should see
just the cross sectional area presented to the beam by the sphere, 7Ta2. The
extra factor of 2 comes from the fact that we have a shadow behind the sphere.
The sphere must throw enough energy forward (with opposite sign) to cancel
the incident light beam. Thus we get

(7TCt )scattered ”1“ ( tTU )shadow = 27TCZ . ( 4 6 . 110 )

This idea is explored in greater detail in Problem 50.10.

490 CHAPTER 46. PARTIAL-WAVE ANALYSIS OF SCATTERING

46.5 Problems for Chapter 46

1. Use the iteration method discussed in Sections 25.2 and 25.3 to improve
on the estimates for the characteristic frequencies given in Section 46.2.

2. In Chapter 51, we will define the scattering amplitude / by

e •Esc ~
r
where e is the polarization vector for the scattered wave. Obtain an
expression for the scattering amplitude in terms of Legendre’s function,
P/(cos0), and the phase shifts 6¡ and 6'/. Show that the differential cross
section is given by

to = IWI2-
Show that if we sum over final polarizations and integrate over all angles
of scattering, we obtain the total cross section (46.96). Show that the
optical theorem is satisfied,

3. (a) From the discontinuity conditions (26.22) [for the surface current K]
and (11.64) [for the surface charge density a] show the conservation
of charge,
V_l •K — iujar = 0.

(b) Calculate the leading contributions to K and a in the long wavelength

limit, ka <C 1 , and, from these, calculate the electric and magnetic
dipole moments, d and /x, used in Problem 45.1. The results are

d — u Einc(r = 0),

/i. — “ CL Binc(r — 0).

4. Derive an expression for the differential cross section for scattering by a

perfectly conducting sphere, dcr/dQ,. Show that in the long wavelength
limit it becomes
da 5
a2(ka)A - ( 1 + cos2 0) — cos 0 ,
d£l 8
which becomes (46.61) when integrated over all scattering angles.

5. The analysis of Section 46.4 considers a particular circularly polarized

incident wave, E = ( l ,i ,0 ) e ikzy B = (—i, 1, ti)elkz. Carry out the same
analysis for the other possibility, E = (1, —i, 0)elkz, B = (z, 1 , 0)elkz.
Chapter 47

Diffraction I

We now turn to the consideration of the scattering of electromagnetic radia­

tion by objects large compared to the wavelength. We begin with a scatterer,
consisting of a perfectly conducting screen, with a (macroscopic) hole in it. As
we shall see, the predictions of geometrical optics are altered because of the
finite wavelength of the radiation. This phenomenon is called diffraction. We
first derive a general expression for the diffracted electric field in terms of its
boundary values. Although it is an exact expression, we will apply it to three
examples of near forward scattering.

47.1 Diffracted Electric Field

Away from the sources, in vacuum (e = \i = 1), the pertinent Maxwell equations
are

V xB = ÍÉ , V •E = 0,
c
—V x E = Í b . (47.1)
c

A consequence of these equations is that E satisfies the wave equation,

-V x ( V x E) = - V ( V . E ) + V2E = ^ E , (47.2)

or

(v ! - ? ^ ) e=o' <47-3)
Specifically, we consider a plane polarized wave normally incident on a perfectly
conducting screen, with an aperture. We choose the coordinate system (see Fig.
47.1) such that x is in the direction of propagation and z is along the direction
of polarization of the incident wave. From our choice of coordinate system,
far from the source, the incident electric field has only a ^-component. For

491
492 CHAPTER 47. DIFFRACTION I

z
X

Figure 47.1: Plane wave incident on an aperture. Here, the z direction is out of
the page.

•r' = ( - x ' , y ' , z ' ) •r' = (* ', z')

x = 0

Figure 47.2: Image charge induced by a point charge adjacent to a perfectly

conducting plane.

radiation of a particular frequency u (with k = a;/c), the wave equation for Ez ,

for example, becomes
(V 2 + P ) E Z = 0. (47.4)

We are interested in finding the electric field to the right of the screen (x > 0),
subject to the boundary condition that the tangential electric field vanishes on
the screen (since it is a perfect conductor).
The solution to the differential equation (47.4) can be expressed in terms of
Green’s function, for x > 0, which satisfies (31.29),

(V 2 + k2)G ( r, r') = —r'), (47.5)

subject to the boundary condition that G = 0 on the surface x = 0. This

Green’s function refers to the field of a unit point charge in the presence of a
perfect conductor lying in the entire x = 0 plane, as shown in Fig. 47.2. The
solution in the region to the right of the conducting plane is the desired Green’s
function, which may be found by the method of images (c/. Chapter 15). If the
coordinates of the source point r' are (a?', $/, ^'), the image point is located at

r' = ( - « ' . jA z '), (47.6)

47.1. DIFFRACTED ELECTRIC FIELD 493

in terms of which the solution to (47.5) is

e ¿fc| r-r,| e ik |r-r'|

G (r ,r ') (47.7)
|r — r'| |r — r7!

This is a generalization of the electrostatic Green’s function (15.32).

To express the electric field in terms of this Green’s function, we first multiply
(47.4) by G (r ,r ') and (47.5) by Ez(r). Then subtracting the two resulting
expressions, and interchanging r and r', we obtain, (dropping the 2 subscript
on Ez for simplicity)

4 7 r£ (r') 5 (r - r') = - E ( r ')(V /2 + k2)G {r', r) + G (r', r )(V /2 + k2)E (r')

= V ' •[G(r', r)V '£ (r') - ^ (rO V 'G ^',^]. (47.8)

The electric field to the right of the screen can now be obtained in terms of its
boundary values by integrating (47.8) over the volume to the right of the plane
{x* > 0) and using the divergence theorem:

AnE{v) = — Í dS' G (r' ' r ) ¿ £<r' ) - E (r' ) ¿ G (r,' r)

Js<
screen+aperture

+ [ dS' (47.9)
Js0

The first integral extends over the entire x' = 0 plane, the minus sign appearing
because the outward normal to the volume is in the —a?' direction. The second
integral is over a surface which can be taken to be a hemisphere with radius
tending toward infinity. Far from the aperture, which acts as the source of the
scattered wave, the leading behavior of the electric field, an outgoing spherical
wave, is the same as that of the Green’s function:

pikr‘ p ikr'

G (r', * ) - * — , m ~ — (47.10)

Then, the r' derivative can be effectively replaced by

ik, (47.11)
drf

so that the integral in (47.9) over the hemisphere at infinity vanishes. Physically,
this follows from the fact that the source of the electric field is not at infinitely
remote points, but at the aperture.
Now we incorporate the boundary conditions that G = 0 on the entire x' = 0
surface, while E = 0 on the surface x f = 0 except for the aperture (since we are
talking about Ez here). Thus the electric field to the right of the screen is

4irE(r) = f dS1 (47.12)

Jai
aperture
494 CHAPTER 47. DIFFRACTION I

The derivative o f Green’s function can be evaluated from (47.7) by noting that,
for the first term,

¿ =- h <4713>
since |r' — r| depends on x f — x y while in the second term,

A - A (47.14)
dx' dx ’

since |r' — r| depends on x* + x. Therefore at x ' = 0, the two terms contribute

equally. Of particular interest is the electric field far away from the aperture
(kr 1, r r'), for which we may use the expansion (32.1) to write

1 c«felr-r'| ^ l eikre-ik n 'F (47.15)

|r —r'| r ’

where n is the unit vector in the direction of observation, r. If we note that

= cos 0, (47.16)
ox r
the normal derivative of the Green’s function is seen to be

" G ( r ', r ) - ——- cos *r . (47.17)

x'-Q r

Therefore, far from the aperture, the electric field is

ik pikr Í i
E (r )~ -f-c o s 0 — / dS' e~tkn' r E { t'). (47.18)

This expression is valid for any shape of the aperture. This form holds for
Ey as well as EZ) but for Ex one must integrate over the whole plane. In the
following sections, we will apply this result to discuss the diffraction by a circular
hole, a slit, and a straight edge. We will do this in the approximation that the
wavelength of the radiation is small compared to the dimensions of the aperture.
Consequently, the wave travels mostly forward, with small angular deviation.
We may further reasonably suppose that only Ez is present, and that it has
very little ^-dependence,

3E
Ex & 0, Ey « 0 , =► “a f » 0. (47.19)

In the same approximation, we may assume that the field in the hole is just
that of the incident wave,
-^hole ^ E/'mC) (47.20)
which is valid except near the edges. Since only relative amplitudes enter into
our discussion, we take the normally incident wave to have unit amplitude,
-S’inc := 1.
47.2. DIFFRACTION B Y A CIRCULAR APERTURE 495
n

Figure 47.3: Geometry of diffraction by a circular aperture.

47.2 Diffraction by a Circular Aperture

We here specialize the aperture to be of circular shape, having a radius a, with
A < a . Using polar coordinates with the origin at the center of the hole, we can
then express the electric field to the right of the hole as (cos# « 1 )

ikr pa p 2 tt j ,
E ~ - i k ----- / pdp _ z e-ikpsine c o s t ^ (47.21)
r Jo Jo 27t

where # is the angle between n and the ar-axis, and <j> is the angle between r'
and the projection of n on the plane defined by the hole (see Fig. 47.3). Because
the integration over <j) is identified as Jo, according to (16.4), the electric field
is proportional to

pa pkasm.9
Jq p d p j a{k p s m 6 )= Jo zd zJ o(z). (47.22)

The remaining ^-integration can be performed with the aid of Bessel’s equation
of zeroth order, (16.18),

d_
Z -M z ) + zJ0(z) = 0, (47.23)
dz

that is,
fz d
j d z z J 0(z) = - z — J0(z) = zJi(z), (47.24)

where we have used the recurrence relation

(47.25)
~ d z Ja('Z^ = J l^ ’

which follows from (38.39) and (16.25). Thus the relative diffracted electric field
is
eikr a
E ~ —ik ----- — :—r J i (Arasin#), (47.26)
r Arsin# v
496 CHAPTER 47. DIFFRACTION I

or, consistently making the small angle approximation, sin 9 « 9,

(47.27)
r had

The ratio of diffracted to incident energy fluxes is

\EI2 _ l r, 3v2 ( Ji{ka6)\ 2

(47.28)
l^ind2 > \ ka$ ) ’

which, when multiplied by r2, yields the differential cross section,

- (b n W ( Jl ( ka$) (47.29)
df2 \ kad

We anticipate that the total cross section, in the small wavelength limit,

k a ^ 1, or (47.30)

will be
ira (47.31)

since this is just the geometrical area of the circular hole, and we are in the
domain of “geometrical” optics. This can be shown explicitly by integrating
(47.29) over all solid angles. The element of solid angle is approximately

d£l = 2ir sin 0 dO £¿2ir0d9, (47.32)

since physically the wave is mostly scattered near the forward direction. Math­
ematically, this last fact is evident because the important values of 0 are those
for which the argument of the Bessel function is of order one, that is,

0~ = - < 1. (47.33)
ka a

The total cross section is, therefore, approximately,

f w/2

•r a 2 ’ l

(47.34)

The expectation, (47.31), will be borne out if

(47.35)

is true.
47.2. DIFFRACTION B Y A CIRCULAR APERTURE 497

To prove (47.35), we first need an integral representation for J\. This can be
achieved by starting from the addition theorem for Bessel’s functions, (16.69),

OO

Jq( V z2 + z'2 - ^zz' cos 4) = , (47.36)

171 — — OO

setting z = z',

Jo ^2zsin ^ ) = '¿T [Jm{z)]2elm,t,, (47.37)

' ' Til — — OO

and extracting the m = 1 harmonic:

/ t o e~**Jo ( 2zsin I ) = [ M z) ? - (47.38)

We next insert this representation into the integral in (47.35). In order to be

able to interchange the order of integration, we introduce a vanishingly small
lower limit for z, zq —* 0, and obtain

(47.39)

where z and x are related by

2z (47.40)

Because of the orthonormality properties of the functions e*m<^, (16.29), we are

only interested in the coefficient of in the Fourier series expansion of the x
integral. Since Jq( x <C 1) « 1, the x integral is

/2zq \sin ^/2|

— Jo(x)
x
« In-----p----- + constant.
2 Zq s in |
(47.41)

By means of the following expansion,

1 1, 4
In • x In
sin ■ 2 (1 - e - ‘> ) ( l - é * )

= In 2 + i [ ( e - ^ + + ...) + (e* + + ...)

(47.42)

we obtain the coefficient of e*^ to be 1/2, verifying the identity (47.35).

Let us turn to a discussion of the angular distribution given by (47.29). For
such small angles that kaO < 1, we need to know the behavior of J\(z) for small
498 CHAPTER 47. DIFFRACTION I

z, which is easily inferred from the integral representation, (16.30),

Jx(z) = ^ 7 f d 4 e iMCO‘ + e -i+

1 I - - +
« ----- 6 d ó iz ------ ------- e 9
2m J 2
= 2’ for z < 1» (47.43)

which, of course, follows from the series representation (16.27). Therefore, very
near to the forward direction, the differential cross section is
da ^ 1
(47.44)
dñ * 4
Comparing this with the total cross section, (47.31), we see that, since a/A >> 1,
the diffraction is mostly forward, consistent with our assumption.
For scattering angles such that z = kaO ^ 1, we make use of the asymptotic
behavior of J\(z), which is given in (16.39) and used as a starting point for the
asymptotic analysis of the zeros of the Bessel functions in Sec. 19.4. A direct
derivation can be obtained by using the integral representation, (16.30), with
(j) = f t — 7r/2:

J\{z) = ¿ )= f cos^ sin ^ ' — ft). (47.45)

J 27T J o it

For z 1, the main contribution to (47.45) comes from the region near the
stationary phase point, ft « 7t/ 2 ; therefore, we reintroduce <j> = ft — 7t/ 2 to
obtain

/
7r/2
— c o s(z COS </) — <j) — 7t/ 2 )
■7T /2 ^
ss R,e / * /2 fí^eí ( * - '/2)e-< ^ a/3i (47.46)
J —n/2 ^
where we have expanded 2 cos —<j) about the stationary phase point, (j) « —l/z.
Evaluating the Gaussian integral, we find (for 2 > i )

7i(z) ~
ir y iz

= ^
¡2
_
,
cos(z _ _ _ _ ) .
7T 7TS ,
( 4 7 .4 7 )

Alternatively, we can obtain this same result by using the recurrence relation
(47.25) and the asymptotic form for J qj (43.37),
47.3. DIFFRACTION B Y A SLIT 499

Z=kaQ

Figure 47.4: Behavior of 2J\(z)/z and the differential cross section (normalized
to unity at $ = 0) given by (47.29) as a function of z = kaO.

Z\ (asymptotic) Z\ (exact)
¥ + ? = 3.927 3.832
= 7.069 7.016
¿2L + JUL = 10.210 10.172

Table 47.1: Table of the first three zeroes of Ji, compared with the estimate
from the asymptotic form (47.47).

This asymptotic formula is already a good approximation for relatively small z .

The behavior of 2J\{z)/z) as suggested by (47.43) and (47.47), is given in
Fig. 47.4, as well as the corresponding differential cross section. Here we see
quite clearly a diffraction pattern, with minima occurring at the zeroes of J\(z)y
denoted by z\. These zeroes can be found approximately from the asymptotic
form (47.47). A comparison of these with the exact values is shown in Table
47.1 for the first three zeroes.

47.3 Diffraction by a Slit

We now consider another example of diffraction, in which the aperture is a slit
running along the entire 2 -axis. Because of this symmetry in 2 , when the electric
field has no initial 2 -dependence, it acquires none subsequently. Therefore, we
have an essentially two-dimensional problem, for which the preceding analysis
applies, leading to (47.12): [Note that we cannot use (47.15) because the slit is
500 CHAPTER 47. DIFFRACTION I

infinitely long.]

E (x, y) = ■£- J ' ^dtf d z 'E ( y ')-^ G (r ',r ). (47.49)

(The justification for omitting the contribution of the surface at infinity is given
in Problem 47.1.) Green’s function is given by (47.7), or explicitly,

eiH(*'-*) 2+ (r'± -r ±)2)1/2 e* M ( * #- * ) a+ ( r i - ? A ) a]1/a

G (r ’ ^ = [(*' - z )2 + (r'x - r x )2]1/2 “ [ ( ^ - z ) 2 + (r^ _ r x )2]i/2> (47,5°)

where

r± = (x,y), (47.51)
r± = (~ x ,y ). (47.52)

The z* integral here is identified as the Hankel function, (43.32),

oo eiky/p2+ ( z ’- z ) 2

/ - oo
dz1
\Jp2 + (z1 - z )2
(47.53)

(The Hankel function typically emerges in two-dimensional problems.) Thus

the electric field to the right of the slit is

E (x, y) = \ J H E (y') ( - ¿ ) (ky/x* + ( y - y > ) * ) • (47.54)

We will be concerned with the diffracted field far away from the slit, for which
we can use the asymptotic form (43.34),

ikix2+ (y -y 'f]1/2
(k^/x2 + (y - j/02) (kx > 1). (47.55)
ink [;x2 + (y — y')2]1/4 ’

To represent diffraction by a slit of width a, we adopt the coordinate system

shown in Fig. 47.5. For a finite slit, since a < a ? , as well as

y a < p = y j x 2 + y2> (47.56)

we make the following approximation:

y/x2 + (y - y')2 « y jp 2 - 2yy' ~ - -y' = p-sin0y', (47.57)

where 0 is the angle of diffraction. Correspondingly, the asymptotic form of the

Hankel function becomes

(ky/x* + ( y - y > ) * ) ~ ] f ^ eÍkPe~ÍkSkíey‘ - (47.58)

47.3. DIFFRACTION B Y A SLIT 501

Figure 47.5: Geometry for diffraction by a slit.

[We see here again a remnant o f the factor

ikn •r; (47.59)

characteristic of radiation fields. See, for example, (47.17).] When the wave­
length is small compared to the slit, a A, we may again approximate the field
in the slit by the incident field,

E ( y ' ) * E inc(y') = l. (47.60)

We also recognize that the radiation is predominately forward, implying

eikp = - i k - e ikp
ox p
« —ikeikp. (47.61)

Thus, for x a ;> A, the field, (47.54), may then be approximated by

p af 2
E (x, y)

, n ~ ( ik)cikp2 sin ^ sine\ (47.62)

V 2irkp' ' Arsiné?

representing an outgoing cylindrical wave.

The differential cross section per unit length of the slit is determined by

(47.63)

which, upon use of the asymptotic field, implies

da _ 2 / sin (4 f sin#)
dO nk l sin#

„ ± { « n m
«5 y
(47.64)
irk \ 0 ) '
502 CHAPTER 47. DIFFRACTION I

0 n 2n 3n
kae/2

Figure 47.6: Differential cross section for diffraction by a wide slit, a A.

The latter holds for 0 <C 1, which is the only region of validity for our result.
The resulting diffraction pattern is shown in Fig. 47.6, where the zeroes occur
at equally spaced points:

ka6 = n27T, n = ±1, ±2, (47.65)

At 6 = 0, the differential cross section is

(47.66)

In the short wavelength limit, we anticipate that the total cross section per
unit length is given by the width of the slit. In fact, for ka 1, most of the
contribution arises from values of 0 near zero, so that we indeed have

cr

(47.67)

where we have used the integral (34.18).

As a check of consistency, we now turn to the limit a —►oo, for which the
slit disappears and the incident plane wave should propagate undisturbed. The
field far away from the “slit” is given by (47.54) and (47.55),

1/2
E (* ,¡,)= Í /V W )(— ) f\iirk
— )1 (47.68)
[*3 + ( y - y 0 2]1/4
47.4. DIFFRACTION B Y A STRAIGHT EDGE 503

If we put E (y') = 1 for all y', we must recover the incident plane wave, E (x , y) =
etkx. We expect that yf ~ y gives the major contribution to the integral, since
the field should just advance with constant phase in y, thereby allowing us to
use
\Jx2 4- (y - y ')2 ~ % + - - ■— (47.69)
¿X

Inserting this approximation into (47.68), we have the following asymptotic

evaluation (x A):

\ 1/2
( ^ } Jkx i k ( y - y ' ) 2/2x
\ srikx) 6 ~e

-j S * Ubj'
i k x e i k ( y - y ' ) 2/2x

/ Pi k ( y - y ’ ) 2/2x __ e ik x
dy'e (47.70)
=\ & -£
as is expected. Here we note, as a check of the approximation |y —y!\ <C that
the significant contributions to the Gaussian integral come from the values of y
satisfying
\ v-if\
(47.71)
x

47.4 Diffraction by a Straight Edge

Finally, we consider the diffraction produced by a semi-infinite plane conductor,
lying in the region defined by x = 0 and y < 0. The field produced by such
a half plane conductor can be again described by (47.68), where, as a first
approximation, we take

E {y ) w Einc — 1, y >0,
E (yf) « 0 , y' < 0. (47.72)

Using the approximation (47.69), we arrive at the expression [cf. (47.70)]

E (x, y) ~ ) f ^ e ikx H e-***-*')9' 3* , (47.73)

which is valid for

x A and |y — yf\<^x. (47.74)
It is convenient to shift the origin of the y' integration:
504 CHAPTER 47. DIFFRACTION I

Figure 47.7: Magnitude of the electric field resulting from diffraction by a

straight edge.

= <47J5>
For sufficiently large $/,
y > V 5*, (47.76)
the second term in (47.75) may be neglected, andthe wave travels undisturbed:

E ( x , y ) ~ eik*. (47.77)

On the other hand, for y = 0 the integral in (47.75) is

= >, (47.78)

and, in consequence, the amplitude of the wave there is reduced by half:

E (x, 0 ) ~ i e ifc‘ . (47.79)

Far below the edge,

y < 0, |j/| > V 5 Í, (47.80)
the diffracted field vanishes,
E {x ,y ) ~ 0. (47.81)
To summarize the above results, we present a picture of the magnitude of the
field strength as a function of y in Fig. 47.7. The region over which the intensity
varies significantly has width
Ay V%x, (47.82)
47.4. DIFFRACTION B Y A STRAIGHT EDGE 505

a distance small compared to x,

< 1. (47.83)
x y x
A quite different limit of diffraction by a straight edge occurs when both x
and y are large, while the diffraction angle 6 is fixed:

— = tan 6. (47.84)
x
We anticipate that the dominant contribution to the scattered field comes from
the region near the edge, where yf is small, and in consequence [see (47.57)]

\/x2 + { y - y ' ) 2 « p - y f sin 6, (47.85)

where
p = \ A 2 -f y2. (47.86)
Using this approximation in (47.68), we obtain, for 6 <C l ,

i eikp
(47.87)
27rkp 6

We see the appearance of a cylindrical wave, originating from the edge, x / = 0,

yf = 0. The corresponding differential scattering cross section per unit length is

d* I W 1 1 (47.88)
<lf IE...A1
[In the next chapter we will provide an exact treatment of this diffraction prob­
lem.]
The above discussion of diffraction by an edge provides a clarification of our
earlier consideration of diffraction by a slit. If we make observations by use of
a screen sufficiently close to the slit, so that
a \/X«, (47.89)
we see a geometrical image of the slit modulated by edge diffraction (see Fig.
47.8). However, if we move the screen so far back that

\f\x >> a, (47.90)

the two patterns overlap to form the diffraction pattern derived in (47.64), see
Fig. 47.9. Mathematically, the origin of the requirement (47.90) can be traced
to the fact that the expansion (47.57) is only valid when terms of order a2/p
are negligible, that is
ka2 1 a2
506 CHAPTER 47. DIFFRACTION I

Figure 47.8: Diffraction by a slit as seen on a nearby screen.

(k/nx)y

Figure 47.9: Diffraction pattern far away from a slit.

47.5. PROBLEMS FOR CHAPTER 47. 507

47.5 Problems for Chapter 47.

1. Verify, in the situation of diffraction by a slit, that it is permissible to
drop the contribution of the surface at infinity from the expression for the
electric field given by (47.49).

2. Show that (47.75) can be represented in terms of Fresnel integrals

C (x ) = J dt cos > S(x) = dt sin ( J * 2) ,

as

£ < * ,v) = «“ ■ {j Vs(C(

+ + (Vs»))}
*'s

3. In the same approximation, give expressions for the electric field produced
by a slit, and verify the results shown in Figs. 47.8 and 47.9.

4. Show that when the slit disappears (a —►oo), (47.54) identically reduces
to eikx.

5. A perfectly conducting sheet has two infinitely long slits, of width a, that
are separated by a distance / a. An electromagnetic field of wavelength
A, polarized parallel to the slit edges, falls normally on the sheet. Using
a reasonable physical assumption that does not involve the magnitude
of a/A, and the formalism applied in the text to one slit, compare the
differential transmission cross section (per unit length) for the two slits
to that of one slit. What are the new features of the diffraction pattern,
particularly for //A >> 1? How does this pattern alter if there are a very
large number of equidistant slits?

6. Extend the argument in Section 47.2 to show that

Chapter 48

Diffraction II

We now adopt a more physical approach to diffraction in which the currents that
give rise to the scattered wave are made explicit. We will reconsider diffraction
by a semi-infinite metal conductor, for which this method is capable of giving an
exact solution. The geometry of the situation is as given in Fig. 48.1. We con­
sider E to possess only a ^-component (z subscript suppressed), and decompose
the electric field into incident and scattered parts,

E = E'mc + Esc&tt • (48.1)

We assume the incident field is a normalized plane wave with frequency u> = Arc,

£inc = eik*. (48.2)

The scattered field arises from the induced current that flows on the metal
plate, which by symmetry can have only a ^-component, J = J^, and has no
dependence on 2 ,
r\
a ;M * ,y ) = o, (48.3)

E—oi ----- ► ^ o--------- ►

-
X
<— E = 0 on conductor

x' = 0
Figure 48.1: Geometry of diffraction by a perfectly conducting straight edge.
The z axis is perpendicular to the page.

509
510 CHAPTER 48. DIFFRACTION II

in conformity with the properties of the incident electric field. Consequently,

there is no charge density and the scattered electric field is expressed by

/
ik|r-r'| i
(dr7) - - p| - J ( r ;), (48.4)

which follows from (35.7) and (35.15). We consider an infinitesimally thin con­
ductor for which we may reduce the three-dimensional integral in (48.4) to a
two-dimensional one by introducing the surface current,

J dx' Jz = K . (48.5)

Further, by identifying the integral with the representation of the Hankel

function (43.32), we may express the result as an integration over y7 alone:

E = Einc + ik J dy' ir iH ^ (k^/x* + (y - j / ) 2) ^ K (y '). (48.6)

For the semi-infinite perfect conductor being considered here, (48.6) is subject
to the boundary condition

E = 0 for x = 0, y < 0. (48.7)

48.1 Approximate Solution

Before finding an exact solution to (48.6), we will first solve this equation by an
approximate treatment, based on the fact that, on the conductor and far from
the edge, the conducting sheet appears to be infinite. There, the ircident wave
is totally reflected,
771 _
it/z —e ikx—e~ikx,
By - - eikx - e~ikx, (48.8)

which we will assume hold for all x < o, y < 0, and that the fields vanish for
x > 0, y < 0. From these forms for the electric and magnetic fields, we can find
the induced current through the use of Maxwell’s equation

V x B = - É + — J, (48.9)
c c
which becomes here
— Jz = dx By - - É z. (48.10)
c c
By integrating (48.10) across the conducting surface from just to the left to just
to the right,
47T
By (x = +0) — By(x — —0) = — K> (48.11)
48.1. APPROXIM ATE SOL UTION 511

[this is (26.22)], and noting that, in our approximation, (48.8),

3/ < 0: By(x = + 0) = 0, By(x = - 0) = - 2 , (48.12)

we find the current appropriate to an infinite conducting sheet to be

K = (48.13)
2 -it'

Then from (48.6), the electric field everywhere is approximately given by

E = eik x- ^ J ° d y'H (0l) (k^/x2 + (y - j/')2) . (48.14)

If we further use the asymptotic form (47.55) for the Hankel function, together
with the approximation that when \x\ is large, only small values of \y — t/| are
significant,

V * 2 + (y - y ')2 « M + — ^ , (48.15)

the expression for the electric field (48.14), becomes

’ rO
d y 'é Hy-y'YI 2k|
ik\x\
E ~ eikx - [ (48.16)
2iri\x\ J—oo

With the substitution y1 — y ►y', the integral in (48.16) has the form

f~y k y l2/2\x\
/ dy'e' (48.17)
J — OO

When y > 0 and far away from the edge, y the integral (48.17) is
negligible, and the wave propagates undisturbed,

E ~ é ikx (48.18)

On the other hand, sufficiently below the edge, y < 0, \y\ \/A]xJ, the integral
is ri\x\/k, and

E ^ e ikx — e ik\x\
_ ( 0, x > 0,
- y eikx _ e-ikx^ x < 0 i (48.19)

reproducing the boundary condition (48.8). On line with the edge, y = 0, the
electric field is
E ~ e ik x- (48.20)

These results are identical with those found earlier in Section 47.4, but now a
physical picture has been provided: The total electric field is the sum of the
incoming field plus the field produced by the currents induced in the metal.
512 CHAPTER 48. DIFFRACTION II

Figure 48.2: Diffraction by a straight edge at a fixed angle 0.

We next turn to the limit of large x and y where the angle of diffraction 0 is
fixed, as shown in Fig. 48.2. In this limit, the dominant contribution at (x, y) is
the sum of the incident wave plus the wave scattered by the edge. In terms of
the description employing currents, this is given approximately by (48.14), with

y/x2 + (y - y ')2 ta p - y' sin 6. (48.21)

As in (47.87), the scattered field is

k p e —ik sin 8y'

® 1 ,ikp_ 1
(48.22)
2irk y/p sin 6 ’

leading to a differential cross section per unit length (valid only for * < i )

da 1 1
(48.23)
dO 2tt& 92 ’
in agreement with (47.88).

48.2 Exact Solution for Current

We now seek an exact solution to (48.6) subject to the boundary condition
(48.7). This is an integral equation since K (y f) and E (y) are interrelated func­
tions. To find the surface current, we consider the electric field on the x = 0

/
plane,
OO 1
H ^o1}(*l» - y' I) - W ). (48-24)
-OO c
In order to solve this equation, we must recognize that an incident plane wave
is an over-idealization, one that can be removed by introducing an exponential
cutoff in y:
e —» 4-0. (48.25)
£W (* = 0 ) = l - e - ‘ l*l,
Note in (48.24) that we have two conditions:
48.2. E XAC T SOLUTION FOR CURRENT 513

1. K (y f) = 0 if yf > 0 (since there is no conductor there),

2. E {y) = 0 if y < 0 (since the conductor is perfect).

We have introduced an infinite range of integration in (48.24) in order to employ
Fourier transforms:
POO

d ye~tiyE {y),

II
-O O
('OO

dy' e~'iy>K ( j/). (48.26)

*<« = ( -O O

The Fourier transform of the incident field on the surface is

p OO

/ d ye~ ltye~ «13/1 _ * i * (48.27)

J — 00 C -ie C + ie ’
which, as expected, becomes 27r<5(£) as c —►0. Furthermore, we require the
Fourier transform of the Hankel function. Starting from its integral representa­
tion, (43.32), using the three-dimensional Fourier transform of Green’s function,
(43.19), and then integrating over 2 , we have
fO O Jky/(y-y,y+z‘2
m H (0 \ k \ y -y '\ ) = dz . ^
J -0 0 y ( y - y'y + z2
. f°° j [ dkxdkydkz eiky(y-y')eikzz
=4’ L diJ-----
(2^r)3 kl + kl + kl - ( ^ ) :
_ ^ dkxdky
I dkxa e<My-y')
*J l2V)2 *2 + * 2 - ( “ * * )
(48.28)

Employing the simple contour integral

r°° dk£ i U)
k= (48.29)
7-00 2tt + 2 ^/(k +i e y - k f

(where the cuts are chosen not to cross the real k axis) we arrive at the following
representation for the Hankel function,
¿ky(y-y')
wiH^\k\y - j/|) = 2wi Í ^ - —7= (48.30)
J- ~ / ( k 4- ^ ) 2 — fcy

which supplies the desired Fourier transform

oo o

/ (48.31)

We now find for the Fourier transform of the integral equation (48.24)

E(0 = - J— r + -7^—r ~ nk (48.32)

C -te C+ ^ (k + i e y - O c
514 CHAPTER 48. DIFFRACTION II

where, if we make explicit the regions in which the integrands are nonzero,
/* o o

E (0 = / d y e -^ E (y ), (48.33)
Jo

K (Q = / dy' e -* * ' R t f ) . (48.34)

J — OO

If these integrals exist for real C> they will also exist for complex values of (.
Anticipating that E (y), K (y ) fall off like e~e\y\ as \y\ —* oo, we see that
E (Q exists for ImC < e,
K(C) exists for ImC > —e. (48.35)
We will call the half planes ImC < e, ImC > —e the lower half plane (LHP) and
upper half plane (UHP), respectively. It is essential to observe that the UHP
and the LHP overlap in a strip,
—e < ImC < (48.36)
Our physical requirements of boundedness ensure that
E (() is regular in the LHP,
Ar(C) is regular in the UHP. (48.37)
In order to examine clearly the analytic properties of (48.32), we multiply it by
y/k -f ie - C":
r.-----;----- 7n/>\ . y/k "p ÍC —C .y/k Í€ —C , km )
y j k -j- Í € — C E ( C ) ~ — í ------ -- ------ :-----------h l ------1—;— ;------------- 27Tk - . (48.38)
C — ie C+ y/k + ie -f C
The factors in (48.38) can be chosen to be regular in the following regions,

y/k + ie - C : LHP,
y/k + ie + C UHP,
y/k + ÍC — C
LHP,

y/k + ie - C
, . . - e...........*
< ImC < . . .e. (48.39)
C + te V '
The last combination can be written as the sum of terms regular in the LHP
and the UHP, respectively,
y/k + ie — C _ y/k -f ie —C — y/k + 2ie y/k -f 2ie
(48.40)
C+ ¿c C + ie C + ie

Thus we can reorganize (48.38) into parts that are regular in the LHP and in
the UHP:
n r - ----- 7nw\ , .y/k + i e - C .Vk + i e - { - V k + 2ie
y k + ie — QE(Q + i -----:---- :---------
(-ie C + ie
.y/k + 2ic km )
2irk (48.41)
C + l€ s/k + Í€ + C
48.2. E X AC T SOLUTION FOR CURRENT 515

The right hand side of (48.41) is regular in the UHP, the left hand side in the
LHP. Since the two functions are regular in a common region [the strip (48.36)],
they may be analytically continued into a function regular for all
We will now show that this function vanishes at infinity, so that it vanishes
everywhere. To this end, we examine (48.34) in the limit C —» oo, for which
only the behavior of K (y ') for y' —» 0 is significant. Because there is no intrinsic
length scale in this limit the current must behave as a power of yf,

~ y' - -0 , (48.42)

where, in order that the integral (48.34) exist,

a < 1. (48.43)

The behavior of the corresponding Fourier transform of the current for large £
is therefore

K (0 ~ f e - « * ' ( - 0 T a<V ~ T jL - - 0, < -oo. (48.44)

J—oo S>

Similarly, since E {y) = 0 if y < 0, the continuity of E requires the following

power law for E near the edge,

E {y )~ y p, y — +0, /? > 0, (48.45)

which is equivalent to the asymptotic statement

f°° 1
E ( 0 = Jo e - ^ E ( y ) dy ~ ^ - 0, <- oo. (48.46)

Hence, the function represented by the analytic continuation of either side of

(48.41) vanishes at infinity, and thus, by Cauchy’s theorem (mistakenly at­
tributed to Liouville), is zero everywhere. Thus we have solutions for K(C) and
£ « ):

— K<r\ - { V k + ie + ( (48.47)
C C + *e

(48.48)
~ ^ >
iV k 1
E (Q = i ( - r - L - - —1 __ (48.49)
VC + *e C -ie C 4" ic "\/k + ic — C

(48.50)
~ £3/2’ 1^1 ^ ^'

Comparison with (48.44) and (48.46) determines the powers

1
(48.51)
2’
516 CHAPTER 48. DIFFRACTION II

implying the following spatial dependences,

(48.52)
* (V ) ~ - M
E {y) ~ \ftj, (48.53)

35 |í/|> V —* 0- We will supply a physical discussion of the meaning of the

behavior near the edge in Section 48.4.

48.3 Exact Diffraction Cross Section

The exact current and field in the plane x = 0 are given by (48.47) and (48.49).
By rewriting the expression for the surface current in the form

1 i 1 ___ 1 y/k + C + if — V k
(48.54)
ch ^ 2n \ C + ie ' y/k C+

we can identify the first term as the Fourier transform of

y < 0, (48.55)
2tr \ 0, y > 0,

which is the current, (48.13), we used in the first approximate solution to this
problem, corresponding to the neglect of edge effects. [Here, (48.55) includes
the exponential cutoff.] The second term in (48.54) thus gives the correction
that must be added in order to obtain the exact current. By considering the
behavior of (48.54) for £ small (which, as we will see below, corresponds to small
diffraction angles), we note that the first term is singular as £ —►0, while the
second is finite. Thus, the first approximation is valid for small angles, as we
have previously asserted.
The asymptotic scattered field at fixed angle 0 follows by use of (47.58) in
(48.6):

k cikp
27rip
(48.56)
where we have used the Fourier transform (48.34) with

C = fcsin0. (48.57)

Then from the solution (48.47) for the surface current, we find for the exact
asymptotic scattered field

1 cik<> + sin 9 (48.58)

-É^scatt ^
2irkp sin 9
48.4. FIELD NEAR EDGE 517

As we anticipated, it agrees with the first approximation (48.22) when 9 <C l .

The corresponding exact differential cross section per unit length is

da __ 1 1 + sin 0
(48.59)
dO 27rk sin2 9 ’

generalizing (48.23). We recognize that the new factor in (48.58), \A + sin0, is

present in order to enforce the boundary condition that the electric field vanish
at 0 = -§ >
Finally, we wish to make contact with the other method, in which the field in
the “aperture,” not the current on the conducting plate is employed. That is, we
use (47.54), relating the scattered field to the field in the aperture, and (47.58),
the asymptotic form of the Hankel function, to give the scattered electric field
for finite diffraction angle 0, |0| < f ,

I L P ikp poo

E sctt ~ V Jo ^ e~‘ *Sin^ ( 2 / ) ’ (48.60)

where E (y) is the exact electric field in the aperture. The integral in (48.60) is
the Fourier transform (48.33),

/ d y e - ik™ °yE (y) = E (( = ksin6) = - i - ^ - = J = = = , (48.61)

Jo k sin 0 v 1 — sm 9

according to (48.49). The resulting scattered electric field coincides with (48.58).

48.4 Field Near Edge

Here we wish to examine the form of the field near the edge, (48.53), from a
different point of view. We may alternatively derive this result by solving the
wave equation, (47.4),

(v 2 + ¿ )E = °> (48-62)
near the edge. Since there the field is rapidly varying over a distance small
compared to the wavelength, we can omit the 1/A2 term. Thus our problem is
the electrostatic one of finding the field near the edge of a plane conductor. Using
the cylindrical coordinate system shown in Fig. 48.3, we write the Laplacian as

/ Id d 1 <92 \
(48.63)
\ p d ^ p dp + p¡ d 9 ^ ) ’

since there is no z dependence. By separating variables (c)'2/()02 —►—m2), we

find the characteristic solutions to Laplace’s equation to be

E Pm■ (48.64)
518 CHAPTER 48. DIFFRACTION II

Figure 48.3: Cylindrical coordinate system for diffraction by a straight edge.

The boundary conditions that the field must vanish on the conducting plane,

_ . 7T 37 T
E = 0 at 6 and — , (48.65)

imply that the characteristic solutions are

E ~ sin m(0 + 7r/2)pm, (48.66)

with
sin27rm = 0. (48.67)

The smallest value of m consistent with (48.67) is

1
m=-, (48.68)

giving the electric field

E = C^/psm - { 0 + » /2 ), (48.69)

where C is a constant. [Note that the solution (48.58) exhibits this same be­
havior, since -y/l + sin6 = \/2sin |(0 + x/2 ).] For x = 0, y > 0 (6 = ir/2), the
electric field is
E(j>, ^ ) = C ^ p = Cy/y, (48.70)

which is (48.53). Another way of writing (48.69) is

E = C lm [v £ e * (,+ ' /a)

i6
= C lm W p e

= C \ m ^ i(x + iy )- (48.71)

We note that a general solution to the two-dimensional Laplace s equation,

( * L + * L ) e = o, (48.72)
\ dx2 dy2)
48.4. FIELD NEAR EDGE 519

a_
2

Figure 48.4: Geometry of a slit.

has the form

£? = I m /(* + iy), (48.73)
where / is a locally regular function. Equation (48.71) has this property away
from the origin, and satisfies the appropriate boundary condition,

E~0 for a: = 0, y < 0, (48.74)

since there
E = C \m yji(x 4- iy) C = 0. (48.75)
Next, we apply the above ideas to the situation of a slit, as shown in Fig.
48.4. When the slit is small compared to the wavelength,

a < A, (48.76)

the solution to Laplace’s equation becomes relevant here. Since the appropriate
boundary conditions are

E = 0 v hen a? = 0, |y| > —, (48.77)

the electric field is found by generalizing our previous result, (48.71),

E = C lm [i (x -f i(y + a/2)) i (x -f i(y — a /2 ))]1^2 . (48.78)

Explicitly, we note that on the surface x = 0, this reduces to

E = Clm-y/V2 — (a/2 ) 2
= C Im *\ /a2/4 — y2
\v\ > f. (48.79)
C \fa2/4 — y2, |y| < f .

To determine the constant C for the diffraction problem, we consider the

fields far from the slit,
1*1 > *, M. (48.80)
520 CHAPTER 48. DIFFRACTION II

where the static result, (48.78), becomes

E ~ Clmi\x\ = C\x\. (48.81)

In order to reproduce the static limit in the absence of the slit (a —►0), we
superimpose on this field another solution to Laplace’s equation, —C#, which
satisfies the same boundary conditions, (48.77), so that we have the following
asymptotic solution,

E ~ —C x + C\x\
_ r o, x > o,
(48.82)
~~ \ — 2C x ) x < 0 .

Far from the surface, we can no longer neglect k2 = 1/A2 in (48.62). Conse­
quently, the approximate diffracted wave, the static limit of which is (48.82),
is
2c
E = — r-sinfc#, x < 0. (48.83)
K
This is exact for zero aperture, a = 0, since it represents a standing wave due
to total reflection by the conducting plane. The associated magnetic field,

d
ikBy = —— Ez = 2 C co sk x t (48.84)

does not vanish on the conducting plane,

ikBy (x = 0) = 2C, (48.85)

thereby determining the constant C in terms of the unperturbed magnetic field

(that is, in terms of the magnetic field present when the slit is absent). Thus
from (48.79), the electric field in the aperture, \y\ < a/2, is determined by the
magnetic field at x = 0,
2i* --- -..- ------ --
E ,(x = 0) « — By{x = 0 ) ^ 0 7 4 - 1/*, (48.86)

for a narrow slit satisfying (48.76). The magnitude of the electric field in the
aperture is small compared to the magnetic field there,

\EZ\~ ka\By \<C \By\, (48.87)

and would be zero if (48.83) were exact. Equation (48.86) is the basis for treating
the diffraction by a narrow slit.

48.5 Problems for Chapter 48

1. Use the development in the last section to calculate the diffraction pattern
produced by a narrow slit, a/A <C 1.
48.5. PROBLEMS FOR CHAPTER 48 521

2. In this chapter, we obtained the exact electric field diffracted by a straight

edge. Use this result to obtain the limiting result for the regime

discussed in Section 47.4.

3. A slot of finite length / and very small width a is cut in a perfectly conduct­
ing sheet. An electromagnetic field of wavelength A, polarized parallel to
the long axis of the slot ( 2 -axis), falls normally on the sheet. Find the an­
gular (0, <j)) distribution of the transmitted radiation in an approximation
valid for //A 1. Compare the 0 dependence with that of the uniformly
excited antenna discussed in Section 34.1; explain the difference in the <j>
distribution.
Chapter 49

Babinet’s Principle

In the preceding chapter we considered the diffraction of a plane wave with a

particular polarization, in which the electric field has only a z component. We
will show here how we can obtain results for the other polarization, in which
the magnetic field has only a ^-component, without additional effort.
We first recognize that because of the linearity of Maxwell’s equations we can
write the solution to our original problem, in which the incident wave approaches
the screen from the left, as the superposition of the solutions of two problems, in
each of which the incident wave approaches the screen from both the left and the
right, as illustrated by Fig. 49.1. The arrows indicate the sense of propagation,
while the signs denote the sense of Ez. The electric field corresponding to the
first term here is odd under reflection in the x = 0 plane, so it satisfies the
boundary condition
E = 0 at x = 0. (49.1)

That is, the entire x = 0 surface can be considered to be a perfect electric

conductor. The electric field in the second term is even under reflection,

E (x) = £ ( - « ) . (49.2)

Ez: ___ , _ i
~ 2 + 3
(+ ) (+ ) (-) (+ ) (+ )

Figure 49.1: Superposition of solutions for Ez polarization.

523
524 CHAPTER 49. BASIN ETS PRINCIPLE

I
I

Ez: _ l
~ 2

Figure 49.2: Electric fields for x > 0, for the Ez polarization.

Figure 49.3: Superposition of solutions for the Bz polarization.

In this case we have two regimes:

y < 0 : Ez = 0 at x = 0,
BE
y > 0: = 0 =* By = 0 at x = 0, (49.3)

where we have used the continuity of the derivative for y > 0. Thus the second
term corresponds to a perfect magnetic conductor (for which the tangential
magnetic field vanishes) for y > 0 and to a perfect electric conductor for y < 0 .
The situation for x > 0 is shown in Fig. 49.2, where a solid line represents a
perfect electric conductor, and a dashed line a perfect magnetic conductor.
We now consider another situation (the “dual” of the above), in which the
incoming wave has polarization Bz , and strikes a conducting plane x = 0 , y > 0 .
The decomposition into two terms may be made as before, as shown in Fig. 49.3,
where the + , — signs now refer to the sense of Bz . In the first term here, Bz is
even, so at x = 0 , y < 0 ,
r\
— Bz = 0 ^ E y = Q. (49.4)

Correspondingly, the entire x = 0 plane acts as a perfect electric conductor. In

the second term, Bz is odd under reflection in the x = 0 plane, so we may, as
in (49.3), describe that situation in two regions,

y < 0 : Bz = 0 at x = 0,
49.1. PROBLEMS FOR CHAPTER 49 525

i
i
I

Figure 49.4: Magnetic fields for x > 0, for the Bz polarization.

Ez Bz

Figure 49.5: Equivalence between dual diffraction problems.

y > 0 : Ey = 0 at x = 0, (49.5)

since Bz is continuous for y < 0. Thus the x = 0 plane acts as a perfect electric
conductor for y > 0, and as a perfect magnetic conductor for y < 0. So for this
polarization the situation for x > 0 appears as shown in Fig. 49.4. For both the
above circumstances, reflection from the infinite conducting planes corresponds
to exactly forward scattering, and can therefore be ignored when we calculate
the scattering cross section. The remaining contributions differ only in that
electric and magnetic quantities are interchanged,

E i- £,
B -E . (49.6)

Since Maxwell’s equations are invariant under this transformation (recall Chap­
ter 2 ), the cross sections for these two problems are identical.
An immediate generalization of the above arguments shows that, for normal
incidence, the problems (E Zy aperture, screen) and (Bz , screen, aperture) cor­
respond to the same cross sections, but with orthogonal polarizations. See Fig.
49.5. This is an example of what is called Babinet’s principle.

49.1 Problems for Chapter 49

1. Compute the cross section for the scattering of a normally incident plane
526 CHAPTER 49. BABINET’S PRINCIPLE

polarized electromagnetic wave by a circular disk, in the spirit of Section

47.2, and thereby verify Babinet’s principle.
Chapter 50

General Scattering

50.1 Integral Equation

Previously, for simplicity, we have considered special geometries and polariza­
tions. (The formalism of Chapter 46 is completely general, but applied there in
only simple situations.) Here we wish to give an outline of a method for treat­
ing more complicated problems. We start from Maxwell’s equations in vacuum,
which for radiation of a definite frequency u>, are in part (k = u>/c)

V x B = - i k E H------J, (50.1)
c
V x E = ikB, (50.2)

which can be combined to yield

V x ( V x E ) = k2E + ik — J. (50.3)
c

The divergence of (50.1) is

ik V •E = J, (50.4)

which also follows from current conservation. Correspondingly, (50.3) can be

written as

—(V2 + k 2)E = ( J + ¿ V(V.J)), (50.5)

which is the inhomogeneous wave equation satisfied by E. (It is equivalent to

the result given in Problem 31.4.)
For an incoming wave satisfying

(V 2 + fc2) E i „ c = 0 , (50.6)

527
528 CHAPTER 50. GENERAL SCATTERING

we can immediately write down the solution to (50.5) in terms of the retarded
Green’s function,

E(r) = Einc(r) + ¿fc J ( d r ' ) y ^ ± (j(r') + ¿ V ' ( V ' . J(r'))) , (50.7)

or, if we integrate by parts,

E (r) = Einc(r) + ik ( l + ¿ w ) •J ( d r O ^ ^ i j ( r ' ) , (50.8)

assuming a confined current distribution. This is the three-dimensional gener­

alization of (48.4). Alternatively it may be derived from the radiation gauge
potential, (32.55).
We now consider the scattering due to a perfectly conducting surface S',
which imposes the boundary condition

Etang = 0 on 5. (50.9)

Because the conductor is perfect, only a surface current K is present. When we

apply the boundary condition (50.9) to (50.8), we obtain an integral equation
for the surface current, which uniquely determines K . Once the surface current
is known, the scattered field is determined from (50.8). This is a generalization
of the procedure we followed in Chapter 48.
In terms of the surface current, we may easily write down the scattered field
at a large distance from the surface S :
ikv r “i

Escatt ~ ik - — ( 1 — nn) • / (dr')e~tka ’ r' - J ( r ') , (50.10)

V J c
where we have used the asymptotic form of the Green’s function, (47.15), with
n being the direction of the scattered wave, and

V e ikr = i k n e ikr. (50.11)

We assume an incident plane wave of unit amplitude with polarization eo and

propagation vector ko,
Einc = eoeiko- r . (50.12)
We are interested in the scattered wave with polarization e and propagation
vector k = kn. Since the polarization is transverse,

e *n = 0, (50.13)

we have ^
e * Escatt ~ ~ ~ f > (50.14)
r
where / is called the scattering amplitude, and depends on the emission direction
and the polarization,

/ = ike • f (dr,)e~ik#r - J ( r 7) = — e* J(k). (50.15)

J c c
50.2. OPTICAL THEOREM 529

Note that the integral here is the three-dimensional Fourier transform of ~J(r/).
The differential cross section for scattering is immediately given by the absolute
square of / ,
^ = r 2 |Escatt- e |2 = |/|2. (50.16)

50.2 Optical Theorem

There is a very general theorem that relates the total cross section for scattering
to the forward scattering amplitude, the optical theorem. We will derive this
theorem under restricted conditions. If we write the differential cross section
for a particular polarization, (50.16), explicitly,

da
= k2 |e« J (dr')e~ik' r'~3(r) (50.17)
dQ,
the total cross section is obtained by summing over polarizations and scattering
angles,

<7 = J d Ü ^ k 2J{dv)-c 3*{v)ei^ ' r .BB. J { dv,) e - ^ ’ r'-c 3{r'). (50.18)

Since e is perpendicular to n, we have the completeness relation,

y^ee + nn = 1 , (50.19)

which can be used to write the sum over polarizations in (50.18) as

y ee = 1 —nn —» i + p V V , (50.20)
e

where the latter form is assumed to act on e*k *(r" r/). The angular integral in
(50.18) is easily found to be [cf. (38.45)]

f „ sin&lr —r'l . _ e*fclr r'\

(50.21)
J**' ( , = 4’ nE¡íbTJ = teImí¡r r ? í'
and in consequence the total cross section is

Air
* = T \mk2
'
J{dv){dv')-r{v). ( l + p V V j
f 1 / 1 \ e**|r-r#|
-j—
i
^ . - J ^ ') . (50.22)

We next use the boundary condition (50.9) together with the fact that J is
confined to the surface and therefore has only a tangential component to infer
that
-J * •E = 0. (50.23)
c
530 CHAPTER 50. GENERAL SCATTERING

Consequently, if we make use of the general expression for the field, (50.8), we
deduce the identity

0 = J (d r )h * * E

= J ( d r ) i j* •Einc

eik\r-r'\
+ ik J \ d r)(d r')-3 * (r)
1 +¿W | r-r'|
(50.24)

Therefore, the total cross section, (50.22), is simply

« - y Im J (dr)^J* •E inc

(50.25)
= T Im [ik / ( dr)E>*nc * \3 •

Since Einc is simply a plane wave, (50.12), this is

ik j{d Y )e ~ iVa *re 0 . i j ( r ) , (50.26)

cr = — Im /(forward scattering), (50.27)

where we recall the expression (50.15) for the scattering amplitude. This result is
the “optical theorem:” the total cross section is proportional to the imaginary
part of the forward scattering amplitude. Note that this theorem is here a
consequence of the conservation of energy; the result follows from

j ( d r ) 3 •E = 0, (50.28)

which just says that the induced currents do no work in a perfect conductor.
As an indication that this theorem is actually much more general than the
derivation that we have supplied, let us return to the scattering of radiation by
an electron bound by a harmonic oscillator potential. The forward scattering
amplitude, (50.15), is, in this circumstance

/ = ikeo • -ev(a>) = k2eeo •r(u>), (50.29)

c
since ko *r = 0, where [compare (45.18)]

e E(u>)
r(«) (50.30)
mu)Q — to2 — iu>y ’

Since the incident field has unit amplitude, eo *E = 1 , (50.29) becomes

(50.31)
m uIq — w2 — iuiy ’
50.3. BORN APPROXIMATION 531

implying
47r ________ u 27
(50.32)
k m ^m e ( wq — w2)2 + (wy)

which is the total cross section given by (45.44).

50.3 Born Approximation for Scattering by Di­

electric
As a final example we consider scattering by a dielectric. For radiation of a
definite frequency, Maxwell’s equation involving the displacement current is,
when no free current is present,

V x B = -ik e E = - i k E - ik(e - 1)E. (50.33)

Comparison with (50.1) now identifies the current effective in scattering to be

- ik U - 1)E = — J. (50.34)
c
Inserting this in the formula (50.8) for the electric field, we obtain the following
integral equation determining the electric field in the dielectric,

/ i \ r eik\r-r'\ / r /\ __ i
E (r) = Einc(r) + ik (l + p V V j . J (dr’ ) ^ _ ^ ( - i k ) { E(rO-
(50.35)
Once E is known in the dielectric, the scattering amplitude, describing the field
far from the dielectric, can then be found from (50.15),

/ = tibe. J(d r ')e -ik ' r>( - i k ) E(r'). (50.36)

When the dielectric constant is near unity,

|e — 1| <C 1, (50.37)

we can employ a simple approximation based on the fact that inside the dielec­
tric, E is then roughly equal to the incident field,

E « Ei„c = e 0e,k° *r . (50.38)

Under this approximation ( “the Born approximation” ), the scattering amplitude

is
f = k2 [(d r ) ~ 1 e¿(k° - k>•re . e0, (50.39)
J 47T

which is expressed as the Fourier transform of the electric susceptibility,

/ = k2e •e 0x (k 0 — k), (50.40)

532 CHAPTER 50. GENERAL SCATTERING

corresponding to the differential cross section

4 ^ = ( e * e 0) 2 fc4 |x(k0 - k ) | 2 . (50.41)

As a simple example, consider a uniform uncharged dielectric sphere of radius

a, small compared to the wavelength, 1. The differential cross section is
then
dct
s0)2k4 (50.42)
dñ = (e '
which, when averaged over incident polarizations and summed over final polar­
izations, agrees with (45.35) for e near unity.
For a plasma, the high frequency behavior of the susceptibility, from (5.21),
is
ne 2
(50.43)
m u2
The corresponding differential cross section, in this approximation, is expressed
in terms of the Fourier transform of the number density,

da
( e - e 0) 5 (dr)n(v)e i ( k 0 - k ) * r
(50.44)
díl

Note that for a single particle, for which n(r) = <S(r — R ), we obtain the Thom­
son cross section, given by (45.14) [cf. (45.10)], when the sum over final polar­
izations is performed.

50.4 Problems for Chapter 50

1. Derive (50.8) from (32.55).

2. Consider an infinitely long, perfectly conducting cylinder with radius a

and axis coinciding with the 2 axis. An incident plane wave, propagating
in the —x direction and polarized in the z direction,

scatters from the cylinder. Because the cylinder is perfectly conducting,

a surface current K is present with the physically obvious property
oo

K = iK{(¡>) = z Yj K meim*,

where the latter form is the Fourier decomposition of K{(¡>).

(a) Using (50.8), calculate the electric field everywhere (p > a) in terms
of the coefficients K m.
(b) Apply the boundary conditions (50.9) and determine K m.
50.4. PROBLEMS FOR CHAPTER 50 533

(c) Determine expressions for B for p > a and check the consistency of
K m by means of the boundary condition (26.22).

3. (a) Derive the integral equation for B (r) equivalent to (50.8).

(b) Consider scattering from the cylinder of the previous problem by an
incident plane wave polarized in the y direction so that the incident
magnetic field is
B in e = - z e ~ ikx.

Here, the surface current will have the form

oo

K = 4> K <neim*-

Calculate the magnetic field everywhere (p > a) in terms of the

coefficients K m and the p behavior of Hm\kp)>
(c) Determine K m by means of (26.22).
(d) Check this result by verifying (50.9).

4. (a) Show that in the region exterior to a bounded current distribution,

the electric and magnetic fields can be expressed as

E = V x ( V X r ü i ) + — V x r n 2,
c
B = - y V x r l l i + V x ( V x r n 2),

where the scalar functions Hi and n 2 are called Debye potentials.

This is a specialization of Problem 44.4.
(b) Show that an alternative way of writing the result of part (a) in
empty space is

1C
E(r) = £ / ,( r ) X im(Q) + - V x 3 ,(r)X ,m(Q)
u>
lm L

BW = £ g ,(r )X lm(Q) - - V x / , ( r ) X , m(fi)

U)
lm

where X /m is a vector spherical harmonic,

X/m(Q) = - - ^ = = L y /m(fi),

L being the angular momentum operator,

L = t X t V .
i

(The / summation starts at / = 1.)

534 CHAPTER 50. GENERAL SCATTERING

(c) Establish the following properties o f the vector spherical harmonics

J da = sw &mm' j

£ |Xjm(íí)|2 =

J d Q [f(r ')X llml(H )Y > [V X g (r)X ,m(Ll)] = 0,

J dil [ V ' x f ( r ' ) X , lml(Il)]* •[V X</(r)Xjm(ii)]

1
rr' 1 ? Tr{r9{r)) + l(l + 1) f ( r'y a (r ) mm* j

and for an arbitrary vector function V,

1 [d
[ i - r + i(/ + l ) ] y , dQ X ;m(i2 ).V (r )

= J dÚ [VxX*m (i2)]•V XV(r),

d ,J dnx?m( n ) .v ( r) = J dnxrm( n ) .( v 2v(r))
= -J dfiXrm (fi).[Vx(VxV( *))],
where

Dl rd r2 r2 '

5. Electromagnetic Green’s dyadics are defined by the system of equations

V xr(r,r') = f^-$(r,r'),

- V x § ( r , r ' ) = *—r ( r ,r ') + iw l5(r - r').

c
In terms of
r' = r + I6(r - r0,
show that

r'(r,r') = £ [/,(r,r')Xlm(i2) + X g , ( r y ) X lm( ^ ) ] ,

lm
50.4. PROBLEMS FOR CHAPTER 50 535

where (use the results o f the previous problem)

91 = 9h

fi =
r*
CD , + k 2)g, = i-c j dfl" X,*m(Q ") •[V "x li5 (r " — r)],

(D, + k2)f, = - J dQ" X*m(Q ") •[ V " x ( V " x l 6 (r" — r))].

Show that the solution to these equations are

9i(r, v') = - i ^ V ' x G , ( r , r ')X ?m(Q),

= ^ F ,( r ,r ') X ? m(fi),

where F\ and G\ are scalar Green’s function satisfying

( Di + k2)A ¡(r , r') = - r').

Then show

r (r ,r O = £ { * 2 F K r y ) X ím(fi)Xrm(íí')
lm

- V x [G,(r, r ') x , m( n ) x ;m(i2')] X ^ ' }

+ - r') ^ X !m(fi)Xrm(i 20 - 16(v - r')-

lm

6. What are the boundary conditions imposed on Fj and Gj by a perfectly

conducting surface? Compute F\ and G¡ in the region exterior to a per­
fectly conducting spherical shell.

7. How is the formalism of Problem 5 modified in the presence of a spatially

varying dielectric constant, e(r)?

8. Show that Green’s function for Helmholtz’ equation in empty space,

(V 2 + k2)G k(r ,r ') = - 47r«5(r - r'),

may be written in the form

G k( r ,r ') = 4ttik '2 2 j,(k r < )h ,(k r> )Yhn(n)Y ¿n{ f f ) .

lm
536 CHAPTER 50. GENERAL SCATTERING

Figure 50.1: Scattering of a scalar field <j) by an object O.

9. A scalar plane wave, e*k #r, is scattered by an object O, as shown in Fig.

50.1. Because of the scattering, the wave field is a superposition of the
plane wave plus the scattered wave <j>SC)

<j>(r) = eik ' r + <j>sc(r).

Consider the volume V to the right of the imaginary plane S (O is to the

left of S). In this region <j>sc satisfies

(V 2 + k2)4>sc( r) = 0.

In terms of Green’s function of Helmholtz’ equation in unbounded space,

derive, for r in V }
Í rlS '
< M r) = j s 4 7 ’ {«isc(r,) v ,G !(r >^,) “ G 'ir.rO V '^ scir')} .

This expresses the scattered wave in V in terms of its values, and normal
gradients, on S. [It is necessary to show there is no contribution from
the hemisphere “at infinity” also bounding V .] Now make the “Fraun­
hofer approximation” that the point of observation r is far away from the
portions of S that contribute significantly to this integral:
ptkv . .
r > r', so G (r, r') « ----- e~lk *r , k; = k~.
r r
Then writing

^sc(r) = ~— / ( M ) >
r
where / is the scattering amplitude, show that

/(0, <t>) = -L J dS '. {^ sc(r/)V 'e -‘k' *'' - e"ik' ’ r'V ^ sc(r0 } .

10. Apply the result of the previous problem to calculate the scattering am­
plitude for a sphere in the “black” approximation, where

4 = eik ‘ , + 4tc = 0
50.4. PROBLEMS FOR CHAPTER 50 537

directly behind the sphere (the “shadow” ) and

<^sc := 0

everywhere else. Express the answer as a Bessel function. Compare the

cross section with (47.29). Why is there a similarity?
Chapter 51

Dispersion Relations for the

Susceptibility

We obtained a dispersion relation for the dielectric constant, or for the electric
susceptibility, that is, an expression for e or x as a function of cj, in Chapter 5
[see (5.21), for example], by assuming a particular model for the motion of the
charges in the medium. Here we will, starting from general physical principles,
derive the form of the dispersion relation satisfied by e(u>) or %(o>), of which
(5.21) is a particular example. For simplicity, we will assume the medium to be
nonmagnetic, that is
¡ i n i or B « H , (51.1)

which is the usual situation. In a linear, isotropic, dielectric medium, the dis­
placement, the polarization, and the electric field are related by (4.58) and
(5.31),
D = E + 47tP = eE, (51.2)

which implies that

P = = %E, (51.3)

where e, or x, is a complex function of u>.

Such a relation between P and E must be consistent with the principle
of causality, which here implies that the effects produced by E (£) occur only
at a later time. Bearing this in mind, we may write the most general linear,
time-translationally invariant, causal relation between P and E as

P (t) = i ° ° dt' r](t - - t')E (0 > (51.4)

J —oo

which is valid for most materials for sufficiently weak fields. The unit step
function, 7](t —¿'),

<m -5>

539
540CHAPTER 51. DISPERSION RELATIONS FOR THE SUSCEPTIBILITY

is introduced to satisfy the causality requirement; it has the integral represen-

tation

■+ 0 . (51.6)
v -f* ie
The function f ( t ) may be represented by a Fourier integral

m = /~ (51.7)

which, for f ( t ) real, implies for the Fourier transform

/ > ') = / h a (51.8)

Substituting the representation (51.6) into (51.4), and using the Fourier trans­
form (51.7), we identify

dw' 1
E(w). (51.9)
2w u> — o>' + ie / ( « ' )

Consequently, causality demands that the following mathematical representa­

tion for the susceptibility is always possible:

x(w) = i M z I = i r *aL W ) (51.10)

M ’ 4 tt J_m 2w u + ie- u '
In order to simplify this relation, we note that f ( t — t'), defined in (51.4), is
arbitrary for t —t' < 0. A convenient assumption is that the function is odd,

= -f{t' -t). (51.11)

This, together with (51.8), means that we can write

f ( u ') = ^ 'w '^ (w '). (51.12)

where $ (u /) is a real even function. Consequently, an alternative expression for

(51.10) is
r°° d j cy2$ (u /)
(51.13)
Jq 27r u'2 -(u> + ie)2 ‘
There is a physical requirement which will impose a further condition on
A s a charged particle moves through a medium, it loses energy to the
atoms, not the other way around. The origin of this energy loss is that the
electric field, E, produced by the particle, does work on the induced currents,
J jP , the total work per unit volume being

♦E > 0. (51.14)
 541

If the Fourier representation for the electric field is introduced, (51.14) becomes

I d p(,)4 •£ •p <-“ >

doo
-o o

-/—cj 2**
|E(o;)|2 > 0. (51.15)

On the other hand, since |E(u>)|2 is even in (51.15) can also be written

duj
f 2ir
( — jw ) x ( w ) | E ( w ) | 2 > 0. (51.16)

The sum of these two equations yields

/
CO I POO J

— lo Ime(u>)|E(u>)|2 = / — u>Ime(u;)|E(u>)|2 > 0, (51.17)

-oo 27r Jo w
where we have used the analog of (51.8) for e(u;). Therefore, since this holds for
any electric field, E(u>), we obtain the positivity condition

Ime(w) > 0 for u> > 0. (51.18)

When this condition is applied to (51.13), we find, for u > 0,

_ f°° du)' u/2$(u/')

0 a
poo J /
/ —— u / 2<I>(u/)7r<$(o / 2 — J 2)
Jo 2 tt

= v4$ (w ) ~
> 0- (51.19)

In deriving (51.19), we have used the identity

I m — -— = 7t 6 ( ). (51.20)
x — it
x

An additional property of the dielectric constant may be inferred by con­

sidering the high-frequency limit, where uj is much larger than a characteristic
excitation frequency of the atoms making up the medium, a;atomic- In this sit­
uation, the effect of binding on the electrons can be neglected, leaving us with
the response of free electrons to the applied electric field. This is given by the
high frequency limit of (5.21),

ne
x (w ) OJ COatomic- (51.21)
mu2
(O f course, we assume that Tiw <C me2 so that the electron remains nonrela-
tivistic.) On the other hand, the dispersion relation (51.13) becomes, in this
limit,
, , 1 f ° ° dw’ /2 . , „
(51.22)
*<"> = “ l j 7 " *<" >■ U> W a to m ic ,
542CHAPTER 51. DISPERSION RELATIONS FOR THE SUSCEPTIBILITY

if the integral exists. In this way, we arrive at a sum rule for 4>:

f° ° du>' /2 _ ne2
/ — w'2 $(w ') = -----. (51.23)
Jo 2 tt m

Consequently, it is convenient to define p(u>') by

(a/ 2 ne 2
(51.24)

which has the following properties,

p(u') > 0 , w' > 0 , (51.25)

and pOO
/ p(u/)cfu/ = 1 . (51.26)
Jo
The dispersion relation, (51.13), now appears as

, , ne 2 r , , P ("')
X{0J) = m y/0 du a;'2
,2 / -f, tc
• )\2j * (51.27)
— (a;

As a simple example, recall the situation when the system has only one
excitation frequency, wo, with a damping constant 7 . This is the Fermi model,
which we considered in Chapter 5. In particular, we found there the dielectric
constant to be [see (5.21)]

V2
e(u>) — 1 + 2 2 • ’ (51.28)
Uq — — iyw

where the plasma frequency of the medium is defined by

9 47rne2
u 2 = ---------. (51.29)
F m
The spectral weight function p(u/) in (51.27) can be obtained from the imaginary
part of e:
a /27
(51.30)
,j'2 wg) 2 + (w'7 )2 '
For small 7 ( t < wo), this exhibits the structure of the Lorentzian line shape.
[Recall Section 35.3.]
It will be useful in the following to have a dispersion relation for the inverse
of e(u>) as well. Its form is suggested by the Fermi model. The reciprocal of
(51.28) is

1 O>0 — iljJJ — W 2
e(w) (u l + w2) - i'a>7 - u>2
1 _________ 1
(51.31)
p (w§ + w2) - JU>7 - w2 ’
51.1. PROBLEMS FOR CHAPTER 51 543

which differs from (51.28) by the sign of the second term, and the replacement
of cjq by Wq -f u>2. Therefore, we anticipate that l/e(u>) satisfies a dispersion
relation of the form

(51.32)
c(w) P Jo U)/2- (w + «02
where the spectral function <?(u/) satisfies the conditions

g(w') > 0, (51.33)

pOO

I q(u>f)du' = 1 . (51.34)
Jo
To show that (51.32) is the correct dispersion relation for l/e(u>), we may pro­
ceed in the same manner as we did in deriving the dispersion relation for e(a>).
We apply the causality condition to write a general relationship between P and
D , leading to a dispersion relation for The positivity of the spectral
function, (51.33), follows from the physical requirement (51.14), expressed in
terms of |D|2, while the sum rule (51.34), results from the expression for l/e(u>)
at high frequencies, u u>atomic> which follows from (51.21)

1
(51.35)
e(w)

51.1 Problems for Chapter 51

1. Derive the form o f the dispersion relation (51.32) by following the steps
outlined above.

2. Show that the sum rule (5.36) follows from the dispersion relation (51.13)
and the sum rule (51.23).
Chapter 52

Charged Particle Energy

Loss

We will now apply the general dispersion relations given in the previous chapter
for the dielectric constant to discuss the energy loss by a charged particle when it
passes through a dielectric medium. The interaction between a charged particle
and a medium is described by the macroscopic Maxwell equations, (4.60). As
we remarked before, (4.60) can be transformed, formally, into the vacuum equa­
tions, (1.65), by the redefinitions given in Section 7.2. It is simplest, therefore,
to start with a vacuum description and then transform to that of a medium.

52.1 General Expression

For a purely electrically charged particle, the rate at which the charge does work
on the medium is
dE_ ‘ d(¡> l. <9A '
(52.1)
dt pe at cJe dt ’
according to (9.44). In the Lorentz gauge, the potentials are related to the
charge and current densities by (31.23) and (31.24), in terms of Green’s function,
G (r — r', t —t'). Therefore, the rate of energy loss by the particle can be rewritten
as

J
dE_
(dr)(drf)dt/ ^je(r, t) •7¡¿G(r — r ',t — t') • ^ je(r', t')
dt

- pe(r ’ t ) ~diG(v - r' . t - t ' ) p e (r' , t') (52.2)

Occurring here is the retarded Green’s function given by (31.28) and (43.19):

(dk) du> e*tk *

G (v -r ',t-t' (52.3)
(2tt)3 27T ^2 _

545
546 CHAPTER 52. CHARGED PARTICLE ENERGY LOSS

The charge and current distributions for a single particle of charge Ze are
given by

pe(r ,i) = ZeS(r - v t),

j e(r ,f) = Z ev6(r - vt), (52.4)

where v is the velocity of the particle which we will approximate by a constant.

Substituting (52.3) and (52.4) into (52.2), and carrying out the now trivial P
integration, we obtain

(dS\ , 2 f («&)*, 2 *i(k.v-u) (52.5)

U J « - 4' (Ze) JW ¡ ? T 7 S ¡ f
Although this expression vanishes, as it must since it refers to the vacuum, from
it we can easily obtain the result for a particle traversing a medium by making
the substitutions,

? “ * ¿2 > (52‘6)
leading to the following formula for energy loss,

(dk)duj —iu
27r6(k •v —cj),
medium (27r)4 k 2 — a>2
(52.7)
where now e(u>) supplies the necessary imaginary part in the denominator. The
angular integration here is trivial:

/: d{cos0)8{kv cosO —u>) = — r)(kv — |u;|). (52.8)

We see here an intimation of Cerenkov radiation [recall Chapter 36]. This step
function supplies the lower limit for the |k| integration. As for the upper limit,
we note that momentum transfers must be limited in this description, since
we have ignored recoil. Large momentum transfer events must be dealt with
differently, as collisions with individual electrons (¿-rays). We therefore cut off
the |k| integration at some characteristic momentum transfer K , reflecting the
boundary between different ways of measuring energy loss. The energy loss rate
is therefore

dE
dt

(52.9)

where we have assumed that the important values of u which contribute to the
integral satisfy K 2 eu>2/ c 2. The energy loss per unit distance traveled, which
52.2. EVALUATION IN TERMS OF SPECTRAL FUNCTIONS 547

is more accessible experimentally, is

_dE_ _ (Z e)2 '°°

dz 27rc2 £ < M - * 0 ( l - In ^ (7 f | , e i„ ) )

_ {Z e ? /■ °° / C2 \ / i 2 W2
r Im / duiu ( 1 ----- 5- 7 —7 I In ---- ;------- 3------ - , (52.10)
7TC
» 2
Jo V w2 ( l - £ c ( W)) v '
where we have used the reality condition, (51.8), for e(u>). Note that if e(u>) is
real, this implies the formula (36.19) for Cerenkov radiation.

52.2 Evaluation in Terms of Spectral Functions

The remaining task is to express the above u> integral in terms of the dispersion
relations for the dielectric constant given in Chapter 51. To do this, we first
decompose (52.10) into two pieces:

_ ¿ E = (Z e £
(52.11)
dz 7rc2
where

I<2v2
pOO

I = Im 1 diOLO |
f i - 4 - ' I In (52.12)
- 2 (1 - £ ) ’
poo 1 _!¿
I I = Im 1 dujuj j 1 c2 (52.13)
1 In 1 y2 e'
1 ~

The imaginary part in (52.12) arises entirely from the imaginary part o f —l/e(o;),
which, from (51.32), is proportional to the spectral function g(u>),

(52.14)
V i(w )) '

If we define an average frequency u)e by

poo
f°° 1 1
/ du g(w)ln — = In — , (52.15)
Jo w2 w2

the integral (52.12) becomes

ir 2 ( °2 \ i K'2y2 (52.16)
J ~ 2 ^ { ^ ) l n We2 ( i - $ y

Here, we is an effective atomic excitation energy produced by an electrically

charged particle, which, in the Fermi model, (51.31), is [see Problem 52.1]

u e = wo-\/<K0 ). (52.17)
548 CHAPTER 52. CHARGED PARTICLE ENERGY LOSS

if 7 < w0, wp.

The remaining integral (52.13) is most conveniently evaluated by replacing
the integration along the positive real axis by a contour integral in the complex
w-plane, as shown in Fig. 52.1. Since l/e(w ) is regular in the first quadrant, the
integration along the contour C\ is equivalent to that along C 2 + C 3 . Further­
more, we notice that along the imaginary axis, w = «V, the dielectric constant
is real, which follows from the dispersion relation (51.27) [or from (51.32)]:

e(iv) = 1 + <4 JQ 5 < e(0 ). (52.18)

This last inequality shows that whenever

v2 ... . v 1
(52.19)
? t(0)<1' or
there is no contribution to I I from the integration along the imaginary axis, C 2 .
This means that we may replace the integral I I by the contour integral along
the quarter circle, C 3 , where |w| —►00 . Since, in this circumstance, |w| is large
compared to atomic frequencies, we have there

e(u0 « 1 - (52.20)

Consequently, we may approximate the logarithm in the integrand by

1 V2 v2 1
In 1 ~ ^ (52.21)
1- ~ u2 c2 1 1
“
V2
P-

which results in

f ( c2 \ ( w? V2
= Im / dujio I 1
~ v 2) ( J c -

= (im
l . ) ■ rt-
(52.22)
52.2. EVALUATION IN TERMS OF SPECTRAL FUNCTIONS 549

Therefore, under the condition (52.19), we obtain the energy lost by the particle
per unit distance traveled,

dE _ 1 u>2(Z e )2 i I<2v2 v2 v 1
for (52.23)
d z~ 2 c2J

When the velocity of the particle is so large that (52.19) no longer holds,

v 1
(52.24)
c >

the integration along the imaginary axis C 2 , is no longer zero, since the argument
of the logarithm, 1 — p-e(u>) goes through zero. Now, there is an additional
contribution arising from C 2'.

dwu) (52.25)

where u = iv {u > 0). The value of is, vVl for which 1 — e(u>) vanishes can be
determined by the dispersion integral

w2 r dU' j ! t o _ = 1 _ i ! (52.26)
p Jo w' 2 + *'a C2

The two extreme situations are

V
”2 1
(52.27)

L)„
4 - . 1 (52.28)
1

Equation (52.25) can be rewritten in terms of vv,

dE\ _ (Z e)2
-Im In------- ---------------- - .
) add _ ™ 2 Jo \c2 e(tv)J (-1 ) (2rc(tV) - 1)
(52.29)
To determine the branch of the logarithm involved here, we return to the general
spectral representation for e(u;), (51.27), and let u> approach the imaginary axis
as follows:
u -► vei(*l2- 6\ (52.30)

implying for the spectral denominator

u/ 2 - w2 J 2 - ( v é ^ ! 2- ^ ) 2 = a/ 2 + v2 — i6, 6 ^ + 0. (52.31)
550 CHAPTER 52. CHARGED PARTICLE ENERGY LOSS

The imaginary part of e(a;), in this limit, is positive and tending toward zero,
which implies that

1
Im I In — Im I In
1
1—
vJc
< (l-^Ree)-z^Im e

= Imln ■ = 7T. (52.32)

The additional term now may be expressed as follows:

( - £ ) = ( | ^ r '^ [ , - 4 _ u, ; r ^ _ « g L
V dz ) add 2v JO C2 PJ0 U i '2 + V 2

_{Ze)2 i ( 1 n2 \ 2 2 r J / , „ « '2 + ^ l
2v2
_ 1 h>¡{Ze)2
fii
r, u>2
(52.33)
u ve \ c2J up.

where u>2 is defined by (52.15), while w2e is given by the analogous equation,

f d u j'q ^ ) In
a/ 2 + V?,
= In ■ 2 •
w.
(52.34)

Therefore, by adding (52.33) to (52.23), we obtain the energy lost by a fast

charged particle to be

_ d E _ l a>2
p {Z e)2 I<2v2
L
In ■ . 2X
dz 4) c2 V c2) “ l
fo r (52.35)
c Ve(Ó)
A somewhat more convenient form of (52.35) can be obtained by expressing
ln l/u ;2e in terms of ln l/o ;2. We first observe that, from (52.26) and (52.34),

d , 2 f°° , / / /x 1 d 2
■ o T ~2\ = 1/
d( v2/c2)
l n u > v e
J/o d w
v 7j a
q ( u -T S —
J2 + v2o d(v2/c2)
2 'T i \ U v

I v2 dv2
_ 1 ~ p- (52.36)
>
U)2
p
d(ti2 / c 2)

which implies that

d
v2A V2
i___ ___ ____ I

H 1^

___ ___ V
(52.37)
jS

d(v2/c2)
I

2
} c2) ( jj
P

Integrating this, we obtain the identity

ff vv2 /c
/c 2 /n/2\ u2 1 1 1 —”

(5238)
52.3. HIGH ENERGY LIMIT 551

Therefore, an alternative form for the energy loss per unit distance for v/c >
V y / f f l is

dE 1 u l{Z e )2 K 2v2
In _ _ ü ! _ (52.39)
’ dz ~ 2 v2 S) c2 JiK0) U K 2
where, in this form, we recognize that the last term is the added contribution
when v/c becomes larger than l/^ /e ( 0 ).

52.3 High Energy Limit

The energy lost by a ultrarelativistic charged particle passing through a dielec­
tric medium is given by (52.35) where i/2 is given by (52.28). For a sufficiently
large speed, vv is large compared to the characteristic frequencies that determine
the dielectric constant, and in consequence
1 1 1
In — » / du>fq(u') In — = In — . (52.40)
U)*
wve J/A
0 ' l/¿ V*

The energy lost in this ultrarelativistic limit is therefore

dE u 2 (Z e)2 K 2c2 v
In as — 1. (52.41)
dz c

Notice that this means that the energy loss saturates at a constant value not
much above the minimum ionization loss rate. This phenomenon, often called
the density effect, was first pointed out by Fermi in 1941.
Let us see, in the Fermi model, how this constant value (the Fermi plateau) is
approached. From the definition of i/Vy the value of v for which l/e(iv) = t>2 / c 2,
we obtain from (51.31),

(52.42)
Wq€(0) + yvv + v2 ’
because e(0 ) = 1 + w2¡w\, or, neglecting 7 in comparison to « 0 ,

7 (52.43)

From this explicit expression for v„ , we can then perform the v'2/c2 integration
in (52.39):
1/21

woe(0)

_ ^ 1/2
2 u€ u)p 1
(52.44)
552 CHAPTER 52. CHARGED PARTICLE ENERGY LOSS

Figure 52.2: Energy loss by a charged particle, incorporating (52.23) and

(52.45). Here x = pz, where p is the density of the medium, and the ab­
scissa is /3y = {v/c)( 1 — {y/c)2)~ 1! 2. We have taken representative values for
the parameters, Kc/u>e = 100, u>2¡w 2 = 30, and the damping constant 7 = 0.

where we have used the Fermi model expression for u>e, (52.17), which implies
1 — l/e (0 ) = cOp/u)2, and retained terms through order (1 —v2/c2). Substituting
(52.44) into (52.39), we obtain the energy loss per unit distance, in this model,

(52.45)
This approach to the Fermi plateau [deviation proportional to (1 —v2/c2) 112] is
not unrealistic.
We conclude this section by sketching the result for the energy loss by a
charged particle passing through matter in Fig. 52.2. This figure should be
taken as a qualitative indication only. For detailed discussion of the comparison
with experiment, and full references to the literature, the reader is referred to
The Review of Particle Physics, Particle Data Group [Physical Review D 54,
132 (1996)].

52.4 Energy Loss by a Magnetic Monopole

We now discuss the energy loss by a purely magnetically charged particle (a
magnetic monopole) of strength g , when it passes through a linear, isotropic,
non-magnetic medium. [This calculation would be relevant for an experimental
discovery of a cosmic magnetic monopole, such as that claimed by Buford Price
52A. ENERGY LOSS B Y A MAGNETIC MONOPOLE 553

in 1975.] The formula can be immediately obtained from that given above
by exploiting the symmetry of Maxwell’s equations discussed in Chapter 2.
Specifically, we have the replacements, in vacuum,

je *jm> Pe * Pm (or Ze g); E B, (52.46)

while, in going from vacuum to a non-magnetic medium, we have the further

substitutions
Jm *Jm > #, and c ■ (52.47)
v^'
The net effect of these substitutions is obtained by multiplying (52.10) by
e(u)g2/(Z e)2, yielding

dE K 2v2
(52.48)
dz 7TC2 J o V V ) W

for the rate o f energy loss by a magnetically charged particle.

Following the same decomposition given in Section 52.2, we find in place of
(52.23)
2„2
dE _ 1 K 2v2 V 1
In - 1 for - < ------ , (52.49)
dz 2 c2 c v W
(1 " $)
where, because of the appearance of e(w), instead of l/e(w ), in (52.48), we
encounter, from the dispersion relation (51.27),

f ’ 1
du p(ui) In — = In — .
1
(52.50)

Comparing (52.49) with (52.23), we see that the energy loss by a magnetically
charged particle is approximately obtained from that of an electrically charged
particle by the substitution
Ze g
(52.51)
V c )

provided that
2 2
ÜJe
w U) m • (52.52)
In the Fermi model, we recognize that

We » C O o \ /e (0 ) > VQ. (52.53)

The general validity of this inequality can be established as follows. From (51.27)
and (51.32), we easily infer that

/ pK ) j
(52.54)
W>2 + Z/2 ^

and
= i - » l [ / ^ < 1 (52.55)
w) p Jo u>'2 + v 2 ’
554 CHAPTER 52. CHARGED PARTICLE ENERGY LOSS

from which follow the integral statements for L walom¡c,

rL
J dv2(e — 1 ) = Wp J
du1p(u')\n = w^ln (52.56)

(52.57)
in view o f (52.54). Therefore, the desired inequality is proved,

« e > Wm . (52.58)
Just as in the electrically charged particle situation, there is here an addi­
tional contribution when ^ > —1—

2 „2
dE _ lWp</ , * 2t>2
In----- ;------- rr — 1 + (52.59)
dz 2 c2
. wm ( ! - v ^Add’
where, instead o f (52.25), we have

( dE\ g2 ¥ fioo r 1-al

( —— ) = — r-Im / dww e(w) - In- (52.60)
V ^ /a d d « 3 Jo L 1 -¿(w )'
The imaginary part arises from = ju; < vv, where vv satisfies

- = e(iuv) = 1 + wp2 j Q
C (52.61)

We can simplify (52.60) to obtain a form for the energy loss analogous to (52.39),
2 n2
dE _ 1 v*g- tfv i f « v Jf c 2\ v 2
v:
dz 2 c2
¿v.» W « i ] ’
V 1
for - > —= = , (52.62)
e.

C \/e(0)
[For the approach to the limiting high energy value, see Problem 52.4.]

52.5 Problems for Chapter 52

1. Derive the formula (52.17) for the effective excitation energy in the Fermi
model. Similarly, derive (52.53), that is, um « uo.

2. Fill in the steps leading to (52.44) and (52.45).

3. Supply the steps necessary to derive (52.62).

4. Find the approach to the high energy limiting value of the energy loss
rate for a magnetically charged particle, in the Fermi model, in analogy
to (52.45).
Appendix A

Units

This book has been written entirely in the Gaussian system of units, which is
most convenient for theoretical purposes. However, for practical uses, and for
engineering applications, it is essential to make contact with the SI (Systeme
International) units, in which the meter, kilogram, second, and ampere are the
fundamental units. Fortunately, it is very easy to transform equations written
in the Gaussian system to the SI system. Here we will give that transformation,
as well as explain the Gaussian units. Thus, evaluations may be performed in
either system, or by using dimensionless quantities, such as the fine structure
constant.
The macroscopic Maxwell equations, (4.60), in Gaussian units, are

V •D = 4717?, V * B = 0,

(A .l)

where

D = E + 47rP,
H = B - 4ttM. (A.2)

In addition, the Lorentz force law is

(A.3)

In contrast, the equations in the SI units are

V •D = p, V B = 0,

V X “ = ¿ D + J, - V X E = | B, (A.4)

where

D = e0E 4* P ,
H = —B -M . (A.5)
Ho

555
556 APPENDIX A. UNITS

Here the constants Co and / íq have defined values,

eofio = c " 2,
Ho = 47T x 10~7, (A .6)

where the speed of light is now defined to be exactly

c = 299 792 458 m /s. (A.7)

In addition, the SI Lorentz force law is

F = e(E + v x B ) . (A .8)

To convert from the equations in Gaussian form to those in SI form, we

multiply each quantity in the former system by a suitable constant:

D —*• k¡) D, E -*• kEE,

H -> kHH, B —* kEB,
P ^ kPP , M —>•knjNl,
P -*■ kpp, J -»• kjJ.

Maxwell’s equations, together with the relation between D , E, and P , and

between H, B, and M , determine the ratios of all the k’s; the Lorentz force law
determines the overall scale. The results are

kj) kE = V 4 ttco ,

kfí \Z4npo,
1
kp k j — kp
\ /4 7 T 6 o ’

k]M (A.10)

So this means that all that is necessary to convert a formula in Gaussian units
into the corresponding formula in SI units is to make the scaling of the electro­
magnetic quantities by the factors given in (A. 10).
To find the actual values of electromagnetic quantities in the two systems,
such as the charge on the electron, it is further necessary to convert the length
and mass units,
x —» Ai a?, m —* A2m, ( Al l )
where here, of course, Ai = 102 and A2 = 103. Thus, for example, an energy
scales by the factor A2A2, and hence the Coulomb potential of a electron at the
origin is, in the two systems, related by

p2 p2
£G _ \2 \ esi
(A.1'2)
ro 1 ^Treorsi’
 557

Quantity SI Unit Corresponding number of Gaussian units

E V /m 3 X 10-4
<t> V 1
300 o
e C 3 x 109
p C /m 3 3 x 103
i A 3 x 109
j A /m 2 3 x 105
p C /m 2 3 x 105
D C /m 2 12tt x 105
cr m ho/m 9 x 109 s_1
C F 9 x 1011 cm
B T 104 gauss
H A-turn/m 4tt x 10-3 oersted
M A-turn/m

CO
o
r*H
1
L H i x 1 0 -11

Table A .l: Relation between SI electromagnetic units and Gaussian units. Here
3 is an approximation for 2.99792458.

where the subscripts now indicate the system of units. Thus

eo= es' V ^ \ ¡¥ ^ 2
= esiVlO-WIoVTo7
= 1.602 x 1(T 19 x 2.998 x 109 = 4.803 x 10- l o esu, (A.13)

where the name of the electrostatic unit is no longer of much significance. In

this way we can easily work out the equivalence between SI units and their
Gaussian counterparts, which are given in the Table.1
Finally, we turn to extraction of numbers from electromagnetic formulas,
since this sometimes leads to confusion. We illustrate three different methods.

• The formula in Gaussian units may be evaluated directly.

• The formula may be transformed to SI units, using (A .10), and then eval­
uated using familiar SI quantities.

• The formula may be evaluated by eliminating electromagnetic quantities

in favor of dimensionless numbers, and then the resulting formula is com­
puted with the aid of dimensional analysis. This method, which may
1It might be noted that the Gaussian system is an amalgamation of the earlier cgs electro­
static and electromagnetic systems of units. Here esu units are used for electrical quantities
and emu units are used for magnetic. Inductance sits on the fence: We use the esu unit, but
some authors use the emu unit, which differs by a factor of (3 x 1010) - 2 . A .single factor of c
gives the ratio of emu to esu units of charge. The other commonly used set of electrical units
is the Heaviside-Lorentz system, which is “rationalized” like the SI system, the units differing,
therefore, from the Gaussian units by various powers of 47r. See the reference at the end of
this Appendix for more detail.
558 APPENDIX A. UNITS

seem less direct, has the virtue that one need not keep track of factors of
c, which is merely a conversion factor between time and space intervals,
and is naturally set equal to unity in relativistic calculations.

We illustrate these three methods by evaluating the following formula, derived

in Problem 32.1, for the mean lifetime of the orbital motion of an electron,
having charge e and mass m, moving in a uniform magnetic field B, whose
motion decays due to the radiation emitted:

_ 3m3c5
(A.14)
T = 4e45 2 '

Suppose B = 104 gauss. Then, in Gaussian units, where m = 9.11 x 10“ 28 g,

c = 3.00 x 1010 cm /s, and e = 4.80 x 10” 10 esu, this evaluates to r = 2.58 s.
On the other hand, we may convert the formula first to SI:

3ireom3c3
(A.15)
e4B 2

Now m = 9.11 x 10"31 kg, c = 3.00 x 108 m /s, and e = 1.60 x 10“ 19 C, so for
a magnetic field of IT, we obtain, again, r = 2.58 s.
The dimensionless method consists, first, in replacing the electric charge by
the fine structure constant,
e2 1
(A.16)
a ~ h e * 137’
and the magnetic field is to be expressed in terms of its ratio to the characteristic
(or critical) field strength,

m2c3
Bc = = 4.41 x 1013gauss. (A.17)
eh
Thus our formula becomes

3 ( B e\ 2 h
r
4 \B J am ec2 ’
(A.18)

so, because h = 6.58 x 10~22 MeV s, and mec2 = 0.511 MeV, we get the same
result, r = 2.58 s. Although this latter method may seem unnatural because it
introduces quantum quantities into a classical calculation, it has the virtue of
allowing the powers of h and c to be determined at the end of the calculation
by elementary considerations of dimensional analysis. Here because me2 has
dimensions of energy, exactly one factor of h must be present in the numerator
to obtain a quantity having the dimensions of time.
This advantage is lost if one uses a purely classical method of eliminating e
and J3, for example, by eliminating the former in favor of the electron mass in
terms of the classical electron radius, (45.6),

e2
ro = — - = 2.818 x 10” 13cm, (A.19)
mcz
 559

and expressing B in terms of the cyclotron frequency for a field B q of 1 gauss,

w0 = — - = 1.759 x 107rad/s, (A .20)

me
so that our formula becomes

(A.21)

This, of course, gives the same answer for B = 104 gauss, namely 2.58 s, but
now in terms of two dimensional quantities, ro and Wo, instead of the one, m,
in the atomic system. The latter is obviously preferable.
For a rather complete discussion of the vexing issue of electromagnetic units,
the reader is referred to Francis B. Silsbee, Systems of Electrical Unitsy National
Bureau of Standards Monograph 56 (U. S. Government Printing Office, Wash­
ington, 1962).
Appendix B

Bibliography

The following general textbooks on electromagnetic theory have been found

useful by the authors in writing this book. All of these books are an essential
part of a physicist’s library.

• J. D. Jackson, Classical Electrodynamics, John Wiley and Sons, New York,

2nd Edition, 1975.

• L. D. Landau and E. M. Lifshitz, Classical Theory of Fields, Pergamon

Press, Oxford, 4th Edition, 1979.

• L. D. Landau and E. M. Lifshitz, Electrodynamics of Continuous Media,

Pergamon Press, Oxford, 1984.

• W. Pauli, Theory of Relativity, Pergamon Press, Oxford, 1958.

• A. Sommerfeld, Electrodynamics, Academic Press, New York, 1964.

• J. Stratton, Electromagnetic Theory, McGraw-Hill, New York, 1941.

561
Index

action principle 85 spherical Hankel functions 484

for electrodynamics 98
addition theorem Babinet’s principle 523
Bessel functions 173 Bessel, Friedrich Wilhelm 165
Hankel functions 455 Bessel functions 165
modified Bessel functions 201 half-integer order 225
spherical Bessel function 484 modified 193
spherical Hankel functions 484 noninteger order 175
spherical harmonics 246 spherical 250
Ampere, Andre Marie 17 zeroes 221, 312, 499
Angstrom 418 beta function 237
angular distribution 354, 378, 386 black approximation 489, 536
center-fed antenna 372 Boltzmann, Ludwig 54
Cerenkov radiation 387 Born approximation 531
impulsive scattering 397 boundary conditions
radiated power 354, 386 displacement 133
nonrelativistic particle 354 electric field 6, 134
relativistic particle 398 magnetic field 316
simple antenna 368 magnetic induction 315
spectral distribution 378 potential 133
synchrotron radiation 406 vector potential 315
high energy 423 Breit Hamiltonian 365
polarization 415 Brewster, Sir David 438
angular momentum Brewster’s angle 438
electromagnetic field 25, 30 Brillouin, L. 81, 82
operator 533
static electric and magnetic charge Cabrera, B. 18, 19
19, 31 canonical momentum 42, 64, 113,
antenna 364
center-fed 371 capacitance
simple 367 coefficients of 291
asymptotic behavior cylinder 294
Bessel functions 169, 175, 221 parallel plate 135, 293
Hankel functions 456 sphere 294
J'2m{2m) 419 Casimir-Polder interaction 262
modified Bessel functions 203 Cauchy, Augustin Louis 171, 219,
spherical Bessel function 482 515

563
564 IN D E X

cavity, resonant 206, 295, 468, 469 straight edge 505, 517
Cerenkov, P.A. 388 Thomson scattering 472
Cerenkov radiation 386 two slits 507
charge density 6 Curie, Marie 388
bound 40 Curie, Pierre 68
effective 41 Curie temperature 70
induced 162, 179 Curie-Weiss law 72
magnetic 17, 337, 552 current density
surface 133 effective 41
charge quantization 19, 31 electric 8
classical radius of the electron 389, magnetic 18, 552
472 cutoff wavenumber 465
Clausius-Mossotti equation 58 cyclotron frequency 402
Clausius, Rudolf 58 cylindrical wave 501
collisonal broadening 383
completeness relation d’Alembert, Jean 342
Bessel functions 172, 221, 224 d’Alembertian 342
Fourier 171, 172 Darwin Hamiltonian 365
Fourier-Bessel 172, 230 Debye, Peter 485
Legendre’s polynomials 250 Debye potentials 533
sine functions 186 delta function 6, 141, 170, 186, 225
spherical Bessel functions 252, delta rays 546
267 density
spherical harmonics 246 angular momentum 25
conductivity 46 charge 6
conductor energy 22
force on 134, 163 force 22
imperfect 445 momentum 23, 76
conservation law density effect 551
and invariance 90 diamagnetism 65
angular momentum 25, 105 dielectric constant 50
charge 9, 105 differential cross section 473, 529
energy 22, 109 circular aperture 496
magnetic charge 18 circular disk 525
magnetic flux 61 slit 501
momentum 23, 104, 110 slot 521
Coulomb energy 1 spherical shell 490
Coulomb force law 1 straight edge 505, 517
Coulomb’s potential 141, 243, 257 Thomson scattering 473
critical harmonic number 418, 422 two slits 507
cross section 471 differential equations
circular aperture 496 Bessel functions 169
partial-wave expansion 488 Hankel functions 455
slit 501 Legendre’s polynomials 247
slot 521 modified Bessel functions 203
spherical shell 485 spherical harmonics 243
IN D E X 565

0 /m 239, 243 Faraday, Michael 2, 12

diffraction 491, 509 Faraday’s law 12, 63
circular aperture 495 Fermi, Enrico 551
slit 499 Fermi model 542
slot 521 Fermi plateau 551
straight edge 503, 509 ferromagnetism 69
two slits 507 field concept 2
digamma function 181 field variation 103, 119, 124
dipole-dipole interaction 258 fine structure constant 389, 558
dipole moment flux
electric 34, 258 angular momentum 25
induced 278, 329, 490 electric 4
intrinsic magnetic 65 energy 22, 75
magnetic 36, 66, 326 magnetic 323, 338
potential energy 36 momentum 23, 77
torque 37 force
Dirac, P. A. M. 19, 31 electric 2
dispersive medium 77 Lorentz 11, 18
dispersion relation 49, 59, 539 on conductor 134
displacement 41 on dielectric 130, 158
divergence theorem 13 on macroscopic body 39
Doppler, Christian 425 on magnetic media 314
Doppler effect 425 radiation reaction 476
two long straight wires 320
duality 18, 357, 524, 553
Fourier, Jean Baptiste Joseph 171
Fourier series 172, 187
edge fields 517
Fourier transform 78, 171, 344, 375
eigenvalues
four-vector 14, 123
variational methods 295
Frank, I.M. 388
Einstein, Albert 13, 117 Fraunhofer approximation 536
electromagnetic pulse 25, 79, 82 frequency spectrum 378, 386, 390
electrostatics 1, 125 Cerenkov radiation 388
elliptic integrals 335 constant acceleration 394, 399
energy damped harmonic motion 381
bounds 129 dipole radiation 379
Coulomb 1 synchrotron radiation 408
density 22, 77 high energy 417, 422
electrostatic 125 Furman, M. 426
for dielectric 126
interaction 157, 180, 258, 261, gamma function 175, 181, 195
274, 321 gauge
loss 545 Coulomb 317, 342
magnetic 125, 313, 322 invariance 105, 107, 342
relativistic 112 Lorentz 343
Euler, Leonard 165 radiation 317, 342
Euler’s constant 181 transformation 105, 341
566 IN D E X

transverse 317 spherical 484

Gauss, Carl Friedrich 6 Heaviside, Oliver 198
Gauss’s theorem 4, 279 Heisenberg, Werner 72
generating function helicon waves 60
Bessel functions 167 Helmholtz’ equation 344, 483, 535
Legendre polynomials 244, 254 Hertz, Heinrich Rudolf 12
modified Bessel functions 200 Hertz vectors 468
spherical harmonics 234 hyperbolic motion 399
generator 90
Green, George 138 image charge 155,161,183, 266, 272,
Green’s 275
dyadics 469, 534 impulsive scattering 396
function 137 index of refraction 77
advanced 346 inductance
circle 215 coefficients of 322
circle and septum 225 mutual, parallel wires 324
dielectric sphere 276 self
equilateral triangle 208 circular loop 324
Euclidean 347 long straight wire 324
free space 141 transmission line 324
grounded conducting plate 161, infinitesimal transformations
179 fields 107, 119
Helmholtz’ equation 483, 535 Lorentz 114
hemispherical boss 274 rotation 29, 104, 246
hollow conducting sphere 265 translation 102
isosceles right angle triangle inversion theorem 231
207 iteration method 304
parallel conducting plates 177 Ivanenko, D. 426
parallel polarization 437
perpendicular polarization 433 Jacobi, Karl Gustav Jacob 212
rectangular cavity 206 Jelly, J. V. 388
rectangular cylinder 205
retarded 345 Kelvin, Lord [WilliamThomson] 235
2 + 1 143 Kronecker delta symbol 23
for dielectric 148
wave equation 348 Lagrangian 86
second identity 139 density 98
Langevin-Debye equation 56
Hall effect 60 Laplace, Pierre Simon 137
Hamiltonian formulation 65, 87 Laplace’s equation 14, 137, 166, 231
Hamiltonian Laplacian
Breit-Darwin 365 curvilinear coordinates 167
electrodynamics 100 cylindrical coordinates 167
mechanical 87, 91 spherical coordinates 243
Hankel function 452, 455, 456, 457 Larmor formula 355
Fourier transform 513 Larmor frequency 402
INDEX 567

lattice, linear 185 Neumann, Franz 323

Legendre, Adrien Marie 238 Neumann functions 175
Legendre’s polynomials 238, 244 Noether, Amalie Emmy 91
Lenz, Heinrich 67 Noether’s theorem 91
Lenz’s law 67
Levi-Civita symbol 24, 123 Oersted, Hans Christian 12
Lienard-Wiechert potentials 346, 347 Ohm, Georg Simon 46
Liouville, Joseph 219, 515 Ohm’s law 46
London, F. and H. 318 optical theorem 490, 530
Lorentz, H. A. 11, 59
force law 11 paramagnetism 67
transformation 14, 116 partial waves 479
Lorentzian line shape 381 partition function 54
Lorenz-Lorentz equation 59 Pauli, W. 399
Lorenz, L. 59 Peregrinus 19
permeability 67
magnetic charge 17, 19, 337, 390, permittivity 50
552 Planck’s constant 82, 389, 558
magnetic field 9, 42 Plasma frequency 52, 542
circular current loop 334 Poincare, Henri 15, 19
current loop 333 Poisson, Simeon Denis 137
long straight wire 320 Poisson’s equation 136, 137
rotating charged sphere 327 polarization, electric 38
magnetic induction 9 polar molecules 53
magnetic moment 36, 66, 326 potential 2
intrinsic 65 Coulomb 141, 243
magnetization 38 in presence of conductors 290
magnetostatics 313 magnetic scalar 332
Mallet, L. 388 scalar 64, 341
Maxwell, James Clerk 12, 324 vector 63, 341
Maxwell’s equations 11, 18 Power spectrum 386
macroscopic 42 Cerenkov radiation 387
Meissner effect 318 synchrotron radiation 406
metric tensor 123 Poynting, John Henry 22
Mie, Gustav 485 Poynting’s vector 22, 75
Mossotti, O. F. 58 Price, Buford 18, 19, 552
modes 295, 459 propagation vector 376
eigenvalue bounds 297, 300
scattering 479 Q 470
momentum quadrupole moment 258
field 23, 77 quadrupole radiation 390
relativistic 112
total density 111, 121 radiation
multipole moments damping 359, 360, 382
electric 259 electric dipole 356, 363
magnetic 325 electric quadrupole 390
568 INDEX

fields 352 Sokolov, A. A. 426

magnetic dipole 357, 390 solid harmonics 233
potentials 351, 359 Sommerfeld, Arnold 81
synchrotron 401 spectral distribution 378
total power 354, 411 Cerenkov radiation 387
Cerenkov radiation 388 constant acceleration 393
constant acceleration 395, 399, dipole radiation 379
400 impulsive scattering 397
nonrelativistic particle 355 power 386
relativistic particle 399 synchrotron radiation 406
synchrotron radiation 409 spherical Bessel functions 250
polarization 416 zeroes 267, 269
Rayleigh cross section 475 spherical harmonics 234
reciprocity 138 step function 338, 539
recurrence relations Stokes’ theorem 13, 332
Bessel functions 174 stress tensor 23, 30, 77, 130, 158
Legendre’s polynomials 254 string 339
modified Bessel functions 203 summation convention 24
spherical harmonics 239 sum rule
reduced wavelength 379, 430 conductivity 47
reflection coefficient 433, 435, 437 electric susceptibility 51, 542
relativity 1/e 543
Einsteinian 13, 111 susceptibility
Galileo-Newton 8, 12 electric 49, 539
resonance 381, 475 magnetic 67
resonant cavity 468, 469 synchrotron radiation
resonant scattering 475 high energy 417
rest mass 112 polarization 413
Riemann, Georg Friedrich Bernhard power 401
181
right hand rule Tamm, I. 388
magnetic induction 320 Ternov, I. M. 426
Stokes’ theorem 13, 332 third viewpoint (action principle) 89
Thomson, J. J. 19
scattering Thomson scattering 471
amplitude 490, 528, 536 Thomson’s theorem 288
black sphere 536 Thomson, William [Lord Kelvin] 235
bound charge 473 torque on macroscopic body 40
circular disk 525 total internal reflection 440
dielectric sphere 475, 532 transformation
perfectly conducting spherical 119
shell 478, 485 duality 18
Thomson 471 electromagnetic field 15, 122
Schwinger, J. 19, 426, 459 gauge 105, 341
skin depth 446, 470 Lorentz 14, 115
Snell’s law 435 media to vacuum 77, 387, 546
INDEX 569

potentials 121
rotation 29, 104, 110
translation 102
velocity 118
transmission coefficient 433, 437
trilinear coordinates 208

uniqueness theorem of electrostatics

8, 13, 14
units 555

van der Waals interaction 262

variational principle 126, 283, 296,
313
vector spherical harmonics 533
virial theorem 25, 94

wave equation 10, 348, 491

waveguides 459
wavenumber 371
Weierstrass, Karl Wilhelm Theodor
181
Weiss, Pierre 69
Weyl, Hermann 105
Wronskian 217, 229
Wronski, Hoene 217

zeta function 181