Vector Formulas a-+(b xc) =b-(c x a) =c- (ax b) a x (bX c) = (a- ob — (a- be (a x b)-(c x d) = (a+ e)(b- d) — (a+ d)(b- ce) Vx VWy=0 v-(Vxa)=0 vx (V xa) =V(V-a)- Va Ve (ya) =a-Vyt Vea Vx (pa) = Vp xatuVxa V(a-b) = (a+ V)b + (b+ V)a + aX (V x b) + b x (V X a) V- (ax b) =b-(V x a) —a-(V Xb) V x (a x b) = a(V-b) — b(V- a) + (b- V)a — (a+ Vb If x is the coordinate of a point with respect to some origin, with magnitude r = |x|, m = x/ris a unit radial vector, and f(r) is a well-behaved function of r, then Vex=3 Vxx=0 Vein = 25+ or (a+ V)nf(r) = 1a [a — n(@a-n)] + n(a-n) V x [nf] = 0 of or V(x-a) =a+x(V-a) + i(L X a) 1 where L => (x x V) is the angular-momentum operator.Theorems from Vector Calculus In the following ¢, , and A are well-behaved scalar or vector functions, V is a three-dimensional volume with volume element d°x, S is a closed two- dimensional surface bounding V, with area element da and unit outward normal mat da. J, VeAd’x = I A-nda (Divergence theorem) [ Vu d’x = [ ym da v s [vx aae= [ax Ada v s [ (OV? + Vb VW) d?x = I on- Vip da (Green’s first identity) [ (dw — WV) dx = I (@¥y — ¥¥$)+nda — (Green’s theorem) In the following S is an open surface and C is the contour bounding it, with line element dl. The normal n to S is defined by the right-hand-screw rule in relation to the sense of the line integral around C. [ (V x A)-nda = ¢. A-dl (Stokes’s theorem) [a x Verda = 4 watClassical ElectrodynamicsClassical Electrodynamics Third Edition John David Jackson Professor Emeritus of Physics, University of California, Berkeley JOHN WILEY & Sons, INC.This book was set in 10 on 12 Times Ten by UG and printed and bound by Hamilton Printing Company. This book is printed on acid-free paper. ©) ‘The paper in this book was manufactured by a mill whose forest management programs include sustained yield harvesting of its timberlands, Sustained yield harvesting principles ensure that the numbers of trees cut each year does not exceed the amount of new growth. Copyright © 1999 John David Jackson. Alll rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise, except as permitted under Sections 107 or 108 of the 1976 United States Copyright Act, without either the prior written permission of the Publisher, or authorization through payment of the appropriate per-copy fee to the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923, (978) 750-8400, fax (978) 750-4470. Requests to the Publisher for permission should be addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, (201) 748-6011, fax (201) 748-6008, E-Mail: [email protected]. To order books or for customer service please, call 1(800)-CALL-WILEY (225-5945). Library of Congress Cataloging-in-Publication Data Jackson, John David, 1925~ Classical electrodynamics / John David Jackson. Pp. cm. Includes index. ISBN 0-471-30932-X (cloth : alk. paper) 1, Electrodynamics. I. Title. QC631.53 1998 537.6—de21 97-46873 cP Printed in the United States of America 109To the memory of my father, Walter David JacksonPreface It has been 36 years since the appearance of the first edition of this book, and 23 years since the second. Such intervals may be appropriate for a subject whose fundamental basis was completely established theoretically 134 years ago by Maxwell and experimentally 110 years ago by Hertz. Still, there are changes in emphasis and applications. This third edition attempts to address both without any significant increase in size. Inevitably, some topics present in the second edition had to be eliminated to make room for new material. One major omission is the chapter on plasma physics, although some pieces appear elsewhere. Read- ers who miss particular topics may, I hope, be able to avail themselves of the second edition. The most visible change is the use of SI units in the first 10 chapters. Gaussian units are retained in the later chapters, since such units seem more suited to relativity and relativistic electrodynamics than SI. As a reminder of the sys- tem of units being employed, the running head on each left-hand page carries “$I” or “—G” depending on the chapter. My tardy adoption of the universally accepted SI system is a recognition that almost all undergraduate physics texts, as well as engineering books at all levels, employ SI units throughout. For many years Ed Purcell and I had a pact to support each other in the use of Gaussian units. Now I have betrayed him! Al- though this book is formally dedicated to the memory of my father, I dedicate this third edition informally to the memory of Edward Mills Purcell (1912-1997), a marvelous physicist with deep understanding, a great teacher, and a wonderful man. Because of the increasing use of personal computers to supplement analytical work or to attack problems not amenable to analytic solution, I have included some new sections on the principles of some numerical techniques for electro- statics and magnetostatics, as well as some elementary problems. Instructors may use their ingenuity to create more challenging ones. The aim is to provide an understanding of such methods before blindly using canned software or even Mathematica or Maple. There has been some rearrangement of topics—Faraday’s law and quasi- static fields are now in Chapter 5 with magnetostatics, permitting a more logical discussion of energy and inductances. Another major change is the consolidation of the discussion of radiation by charge-current sources, in both elementary and exact multipole forms, in Chapter 9. All the applications to scattering and dif- fraction are in Chapter 10. The principles of optical fibers and dielectric waveguides are discussed in two new sections in Chapter 8. In Chapter 13 the treatment of energy loss has been shortened and strengthened. Because of the increasing importance of synchro- tron radiation as a research tool, the discussion in Chapter 14 has been aug- mented by a detailed section on the physics of wigglers and undulators for syn- chroton light sources. There is new material in Chapter 16 on radiation reaction and models of classical charged particles, as well as changed emphasis. There is much tweaking by small amounts throughout. I hope the reader will viiPreface not notice, or will notice only greater clarity. To mention but a few minor addi- tions: estimating self-inductances, Poynting’s theorem in lossy materials, polar- ization potentials (Hertz vectors), Goos-Hanchen effect, attenuation in optical fibers, London penetration depth in superconductors. And more problems, of course! Over 110 new problems, a 40% increase, all aimed at educating, not discouraging. In preparing this new edition and making corrections, I have benefited from questions, suggestions, criticism, and advice from many students, colleagues, and newfound friends. I am in debt to all. Particular thanks for help in various ways go to Myron Bander, David F. Bartlett, Robert N. Cahn, John Cooper, John L. Gammel, David J. Griffiths, Leroy T. Kerth, Kwang J. Kim, Norman M. Kroll, Michael A. Lee, Harry J. Lipkin, William Mendoza, Gerald A. Miller, William A. Newcomb, Ivan Otero, Alan M. Portis, Fritz Rohrlich, Wayne M. Saslow, Chris Schmid, Kevin E. Schmidt, and George H. Trilling. J. David Jackson Berkeley, California, 1998, 2001Preface to the Second Edition In the thirteen years since the appearance of the first edition, my interest in classical electromagnetism has waxed and waned, but never fallen to zero. The subject is ever fresh. There are always important new applications and examples. ‘The present edition reflects two efforts on my part: the refinement and improve- ment of material already in the first edition; the addition of new topics (and the omission of a few). The major purposes and emphasis are the same, but there are extensive changes and additions. A major augmentation is the “Introduction and Survey” at the beginning. Topics such as the present experimental limits on the mass of the photon and the status of linear superposition are treated there. The aim is to provide a survey of those basics that are often assumed to be well known when ‘one writes down the Maxwell equations and begins to solve specific examples. Other major changes in the first half of the book include a new treatment of the derivation of the equations of macroscopic electromagnetism from the micro- scopic description; a discussion of symmetry properties of mechanical and elec- tromagnetic quantities; sections on magnetic monopoles and the quantization condition of Dirac; Stokes’s polarization parameters; a unified discussion of the frequency dispersion characteristics of dielectrics, conductors, and plasmas; a dis- cussion of causality and the Kramers-Kronig dispersion relations; a simplified, but still extensive, version of the classic Sommerfeld—Brillouin problem of the arrival of a signal in a dispersive medium (recently verified experimentally); an unusual example of a resonant cavity; the normal-mode expansion of an arbitrary field in a wave guide; and related discussions of sources in a guide or cavity and the transmission and reflection coefficients of flat obstacles in wave guides. Chapter 9, on simple radiating systems and diffraction, has been enlarged to include scattering at long wavelengths (the blue sky, for example) and the optical theorem. The sections on scalar and vectorial diffraction have been improved. Chapters 11 and 12, on special relativity, have been rewritten almost com- pletely. The old pseudo-Euclidean metric with x, = ict has been replaced by g*” (with g® = +1, g = —1,i = 1, 2, 3). The change of metric necessitated a complete revision and thus permitted substitution of modern experiments and concerns about the experimental basis of the special theory for the time-honored aberration of starlight and the Michelson—Morley experiment. Other aspects have been modernized, too. The extensive treatment of relativistic kinematics of the first edition has been relegated to the problems. In its stead is a discussion of the Lagrangian for the electromagnetic fields, the canonical and symmetric stress-energy tensor, and the Proca Lagrangian for massive photons. Significant alterations in the remaining chapters include a new section on transition radiation, a completely revised (and much more satisfactory) semi- classical treatment of radiation emitted in collisions that stresses momentum transfer instead of impact parameter, and a better derivation of the coupling of multipole fields to their sources. The collection of formulas and page references to special functions on the front and back flyleaves is a much requested addition. Of the 278 problems, 117 (more than 40 per cent) are new.X Preface to the Second Edition The one area that remains almost completely unchanged is the chapter on magnetohydrodynamics and plasma physics. I regret this. But the book obviously has grown tremendously, and there are available many books devoted exclusively to the subject of plasmas or magnetohydrodynamics. Of minor note is the change from Maxwell’s equations and a Green’s func- tion to the Maxwell equations and a Green function. The latter boggles some minds, but is in conformity with other usage (Bessel function, for example). It is still Green’s theorem, however, because that’s whose theorem it is. Work on this edition began in earnest during the first half of 1970 on the occasion of a sabbatical leave spent at Clare Hall and the Cavendish Laboratory in Cambridge. I am grateful to the University of California for the leave and indebted to N. F. Mott for welcoming me as a visitor to the Cavendish Laboratory and to R. J. Eden and A. B. Pippard for my appointment as a Visiting Fellow of Clare Hall. Tangible and intangible evidence at the Cavendish of Maxwell, Ray- leigh and Thomson provided inspiration for my task; the stimulation of everyday activities there provided necessary diversion. This new edition has benefited from questions, suggestions, comments and criticism from many students, colleagues, and strangers. Among those to whom T owe some specific debt of gratitude are A. M. Bincer, L. S. Brown, R. W. Brown, E. U. Condon, H. H. Denman, S. Deser, A. J. Dragt, V. L. Fitch, M. B. Halpern, A. Hobson, J. P. Hurley, D. L. Judd, L. T. Kerth, E. Marx, M. Nauenberg, A. B. Pippard, A. M. Portis, R. K. Sachs, W. M. Saslow, R. Schleif, V. L. Telegdi, T. Tredon, E. P. Tryon, V. F. Weisskopf, and Dudley Williams. Especially helpful were D. G. Boulware, R. N. Cahn, Leverett Davis, Jr., K. Gottfried, C. K. Gra- ham, E. M. Purcell, and E. H. Wichmann. I send my thanks and fraternal greet- ings to all of these people, to the other readers who have written to me, and the countless students who have struggled with the problems (and sometimes written asking for solutions to be dispatched before some deadline!). To my mind, the book is better than ever. May each reader benefit and enjoy! J.D. Jackson Berkeley, California, 1974Preface to the First Edition Classical electromagnetic theory, together with classical and quantum mechanics, forms the core of present-day theoretical training for undergraduate and grad- uate physicists. A thorough grounding in these subjects is a requirement for more advanced or specialized training. Typically the undergraduate program in electricity and magnetism involves two or perhaps three semesters beyond elementary physics, with the emphasis on the fundamental laws, laboratory verification and elaboration of their con- sequences, circuit analysis, simple wave phenomena, and radiation. The mathe- matical tools utilized include vector calculus, ordinary differential equations with constant coefficients, Fourier series, and perhaps Fourier or Laplace transforms, partial differential equations, Legendre polynomials, and Bessel functions. As a general rule, a two-semester course in electromagnetic theory is given to beginning graduate students. It is for such a course that my book is designed. My aim in teaching a graduate course in electromagnetism is at least threefold. The first aim is to present the basic subject matter as a coherent whole, with emphasis on the unity of electric and magnetic phenomena, both in their physical basis and in the mode of mathematical description. The second, concurrent aim is to develop and utilize a number of topics in mathematical physics which are useful in both electromagnetic theory and wave mechanics. These include Green’s theorems and Green’s functions, orthonormal expansions, spherical har- monics, cylindrical and spherical Bessel functions. A third and perhaps most important purpose is the presentation of new material, especially on the inter- action of relativistic charged particles with electromagnetic fields. In this last area personal preferences and prejudices enter strongly. My choice of topics is gov- erned by what I feel is important and useful for students interested in theoretical physics, experimental nuclear and high-energy physics, and that as yet ill-defined field of plasma physics. The book begins in the traditional manner with electrostatics. The first six chapters are devoted to the development of Maxwell’s theory of electromagne- tism. Much of the necessary mathematical apparatus is constructed along the way, especially in Chapter 2 and 3, where boundary-value problems are discussed thoroughly. The treatment is initially in terms of the electric field E and the magnetic induction B, with the derived macroscopic quantities, D and H, intro- duced by suitable averaging over ensembles of atoms or molecules. In the dis- cussion of dielectrics, simple classical models for atomic polarizability are de- scribed, but for magnetic materials no such attempt to made. Partly this omission was a question of space, but truly classical models of magnetic susceptibility are not possible. Furthermore, elucidation of the interesting phenomenon of ferro- magnetism needs almost a book in itself. The next three chapters (7-9) illustrate various electromagnetic phenomena, mostly of a macroscopic sort. Plane waves in different media, including plasmas as well as dispersion and the propagation of pulses, are treated in Chapter 7. The discussion of wave guides and cavities in Chapter 8 is developed for systems of arbitrary cross section, and the problems of attenuation in guides and the Q ofPreface to the First Edition a cavity are handled in a very general way which emphasizes the physical pro- cesses involved. The elementary theory of multipole radiation from a localized source and diffraction occupy Chapter 9. Since the simple scalar theory of dif- fraction is covered in many optics textbooks, as well as undergraduate books on electricity and magnetism, I have presented an improved, although still approx- imate, theory of diffraction based on vector rather than scalar Green’s theorems. The subject of magnetohydrodynamics and plasmas receives increasingly more attention from physicists and astrophysicists. Chapter 10 represents a sur- vey of this complex field with an introduction to the main physical ideas involved. The first nine or ten chapters constitute the basic material of classical elec- tricity and magnetism. A graduate student in physics may be expected to have been exposed to much of this material, perhaps at a somewhat lower level, as an undergraduate. But he obtains a more mature view of it, understands it more deeply, and gains a considerable technical ability in analytic methods of solution when he studies the subject at the level of this book. He is then prepared to go on to more advanced topics. The advanced topics presented here are predomi- nantly those involving the interaction of charged particles with each other and with electromagnetic fields, especially when moving relativistically. The special theory of relativity had its origins in classical electrodynamics. And even after almost 60 years, classical electrodynamics still impresses and de- lights as a beautiful example of the covariance of physical laws under Lorentz transformations. The special theory of relativity is discussed in Chapter 11, where all the necessary formal apparatus is developed, various kinematic consequences are explored, and the covariance of electrodynamics is established. The next chapter is devoted to relativistic particle kinematics and dynamics. Although the dynamics of charged particles in electromagnetic fields can properly be consid- ered electrodynamics, the reader may wonder whether such things as kinematic transformations of collision problems can. My reply is that these examples occur naturally once one has established the four-vector character of a particle’s mo- mentum and energy, that they serve as useful practice in manipulating Lorentz transformations, and that the end results are valuable and often hard to find elsewhere. Chapter 13 on collisions between charged particles emphasizes energy loss and scattering and develops concepts of use in later chapters. Here for the first time in the book I use semiclassical arguments based on the uncertainty principle to obtain approximate quantum-mechanical expressions for energy loss, etc., from the classical results. This approach, so fruitful in the hands of Niels Bohr and E. J. Williams, allows one to see clearly how and when quantum-mechanical effects enter to modify classical considerations. The important subject of emission of radiation by accelerated point charges is discussed in detail in Chapters 14 and 15. Relativistic effects are stressed, and expressions for the frequency and angular dependence of the emitted radiation are developed in sufficient generality for all applications. The examples treated range from synchrotron radiation to bremsstrahlung and radiative beta processes. Cherenkov radiation and the Weizsicker—Williams method of virtual quanta are also discussed. In the atomic and nuclear collision processes semiclassical argu- ments are again employed to obtain approximate quantum-mechanical results. I lay considerable stress on this point because I feel that it is important for the student to see that radiative effects such as bremsstrahlung are almost entirelyPreface to the First Edition 1 classical in nature, even though involving small-scale collisions. A student who meets bremsstrahlung for the first time as an example of a calculation in quantum field theory will not understand its physical basis. Multipole fields form the subject matter of Chapter 16. The expansion of scalar and vector fields in spherical waves is developed from first principles with no restrictions as to the relative dimensions of source and wavelength. Then the properties of electric and magnetic multipole radiation fields are considered. Once the connection to the multiple moments of the source has been made, examples of atomic and nuclear multipole radiation are discussed, as well as a macroscopic source whose dimensions are comparable to a wavelength. The scat- tering of a plane electromagnetic wave by a spherical object is treated in some detail in order to illustrate a boundary-value problem with vector spherical waves. In the last chapter the difficult problem of radiative reaction is discussed. The treatment is physical, rather than mathematical, with the emphasis on delim- iting the areas where approximate radiative corrections are adequate and on finding where and why existing theories fail. The original Abraham-Lorentz the- ory of the self-force is presented, as well as more recent classical considerations. The book ends with an appendix on units and dimensions and a bibliography. In the appendix I have attempted to show the logical steps involved in setting up a system of units, without haranguing the reader as to the obvious virtues of my choice of units. I have provided two tables which I hope will be useful, one for converting equations and symbols and the other for converting a given quantity of something from so many Gaussian units to so many mks units, and vice versa. The bibliography lists books which I think the reader may find pertinent and useful for reference or additional study. These books are referred to by author’s name in the reading lists at the end of each chapter. This book is the outgrowth of a graduate course in classical electrodynamics which I have taught off and on over the past eleven years, at both the University of Illinois and McGill University. I wish to thank my colleagues and students at both institutions for countless helpful remarks and discussions. Special mention must be made of Professor P. R. Wallace of McGill, who gave me the opportunity and encouragement to teach what was then a rather unorthodox course in elec- tromagnetism, and Professors H. W. Wyld and G. Ascoli of Illinois, who have been particularly free with many helpful suggestions on the treatment of various topics. My thanks are also extended to Dr. A. N. Kaufman for reading and com- menting on a preliminary version of the manuscript, and to Mr. G. L. Kane for his zealous help in preparing the index. J.D. Jackson Urbana, Illinois, January, 1962Contents Introduction and Survey I I1 Maxwell Equations in Vacuum, Fields, and Sources 2 12 Inverse Square Law, or the Mass of the Photon 5 13 Linear Superposition 9 14 Maxwell Equations in Macroscopic Media. 13 Is Boundary Conditions at Interfaces Between Different Media 16 16 Some Remarks on Idealizations in Electromagnetism 19 References and Suggested Reading 22 Chapter 1 / Introduction to Electrostatics 24 ib) Coulomb’s Law 24 1.2 Electric Field 24 1.3 Gauss’s Law 27 1.4 Differential Form of Gauss’s Law 28 1.5 Another Equation of Electrostatics and the Scalar Potential 29 1.6 Surface Distributions of Charges and Dipoles and Discontinuities in the Electric Field and Potential 31 1.7 Poisson and Laplace Equations 34 1.8 Green’s Theorem 35 1.9 — Uniqueness of the Solution with Dirichlet or Neumann Boundary Conditions 37 1.10 Formal Solution of Electrostatic Boundary-Value Problem with Green Function 38 1.11 Electrostatic Potential Energy and Energy Density; Capacitance 40 1.12 Variational Approach to the Solution of the Laplace and Poisson Equations 43 1.13 Relaxation Method for Two-Dimensional Electrostatic Problems 47 References and Suggested Reading 50 Problems 50 Chapter 2 / Boundary-Value Problems in Electrostatics: I 57 2.1. Method of Images 57 2.2 Point Charge in the Presence of a Grounded Conducting Sphere 58 2.3 Point Charge in the Presence of a Charged, Insulated, Conducting Sphere 60 24 Point Charge Near a Conducting Sphere at Fixed Potential 61 2.5 Conducting Sphere in a Uniform Electric Field by Method ofImages 62 2.6 Green Function for the Sphere; General Solution for the Potential 64 2.7 Conducting Sphere with Hemispheres at Different Potentials 65 xv18 Introduction and Survey not contribute to the integrals on the left in (1.13) and (1.14). Only the top and bottom contribute. If the top and the bottom are parallel, tangent to the surface, and of area Aa, then the left-hand integral in (1.13) is $ D-nda=(D,-D,)-n Aa and similarly for (1.14). If the charge density p is singular at the interface so as to produce an idealized surface charge density c, then the integral on the right in (1.13) is By = | pare o Aa Thus the normal components of D and B on either side of the boundary surface are related according to (D, - D,)-n =o (1.17) (B, — B,)-n = 0 (18) In words, we say that the normal component of B is continuous and the discon- tinuity of the normal component of D at any point is equal to the surface charge density at that point. In an analogous manner the infinitesimal Stokesian loop can be used to de- termine the discontinuities of the tangential components of E and H. If the short arms of the contour C in Fig. 1.4 are of negligible length and each long arm is parallel to the surface and has length A/, then the left-hand integral of (1.16) is $ Bedi = (tx n) (EB) al and similarly for the left-hand side of (1.15). The right-hand side of (1.16) vanishes because B/at is finite at the surface and the area of the loop is zero as the length of the short sides goes to zero. The right-hand side of (1.15) does not vanish, however, if there is an idealized surface current density K flowing exactly on the boundary surface. In such circumstances the integral on the right of (1.15) is [ [2+ 2] -1do=K-eas S' a The second term in the integral vanishes by the same argument that was just given. The tangential components of E and H on either side of the boundary are therefore related by n x (E, - E,) =0 (1.19) n x (H, — H,) = K (1.20) In (1.20) it is understood that the surface current K has only components parallel to the surface at every point. The tangential component of E across an interface is continuous, while the tangential component of His discontinuous by an amount whose magnitude is equal to the magnitude of the surface current density and whose direction is parallel to K x n. The discontinuity equations (1.17)-(1.20) are useful in solving the MaxwellSect. 1.6 Some Remarks on Idealizations in Electromagnetism 19 equations in different regions and then connecting the solutions to obtain the fields throughout all space. 1.6 Some Remarks on Idealizations in Electromagnetism In the preceding section we made use of the idea of surface distributions of charge and current. These are obviously mathematical idealizations that do not exist in the physical world. There are other abstractions that occur throughout electro- magnetism. In electrostatics, for example, we speak of holding objects at a fixed potential with respect to some zero of potential usually called “ground.” The relations of such idealizations to the real world is perhaps worthy of a little dis- cussion, even though to the experienced hand most will seem obvious. First we consider the question of maintaining some conducting object at a fixed electrostatic potential with respect to some reference value. Implicit is the idea that the means does not significantly disturb the desired configuration of charges and fields. To maintain an object at fixed potential it is necessary, at least from time to time, to have a conducting path or its equivalent from the object to a source of charge far away (‘‘at infinity”) so that as other charged or uncharged objects are brought in the vicinity, charge can flow to or from the object, always maintaining its potential at the desired value. Although more sophisticated means are possible, metallic wires are commonly used to make the conducting path. Intuitively we expect small wires to be less perturbing than large ones. The reason is as follows: Since the quantity of electricity on any given portion of a wire at a given potential diminishes indefinitely when the diameter of the wire is indefi- nitely diminished, the distribution of electricity on bodies of considerable dimensions will not be sensibly affected by the introduction of very fine metallic wires into the field, such as are used to form electrical con- nexions between these bodies and the earth, an electrical machine, or an electrometer.* The electric field in the immediate neighborhood of the thin wire is very large, of course. However, at distances away of the order of the size of the “bodies of considerable dimensions” the effects can be made small. An important historical illustration of Maxwell’s words is given by the work of Henry Cavendish 200 years ago. By experiments done in a converted stable of his father’s house, using Leyden jars as his sources of charge, thin wires as conductors, and suspending the objects in the room, Cavendish measured the amounts of charge on cylinders, discs, etc., held at fixed potential and compared them to the charge on a sphere (the same sphere shown in Fig. I.1) at the same potential. His values of capaci- tance, so measured, are accurate to a few per cent. For example, he found the ratio of the capacitance of a sphere to that of a thin circular disc of the same radius was 1.57. The theoretical value is 77/2. There is a practical limit to the use of finer and finer wires. The charge per unit length decreases only logarithmically [as the reciprocal of In(d/a), where a *J.C. Maxwell, A Treatise on Electricity and Magnetism, Dover, New York, 1954 reprint of the 3rd edition (1891), Vol. 1, p. 96.20 Introduction and Survey is the mean radius of the wire and d is a typical distance of the wire from some conducting surface]. To minimize the perturbation of the system below some level, it is necessary to resort to other means to maintain potentials, comparison methods using beams of charged particles intermittently, for example. When a conducting object is said to be grounded, it is assumed to be con- nected by a very fine conducting filament to a remote reservoir of charge that serves as the common zero of potential. Objects held at fixed potentials are sim- ilarly connected to one side of a voltage source, such as a battery, the other side of which is connected to the common “ground.” Then, when initially electrified objects are moved relative to one another in such a way that their distributions of electricity are altered, but their potentials remain fixed, the appropriate amounts of charge flow from or to the remote reservoir, assumed to have an inexhaustible supply. The idea of grounding something is a well-defined concept in electrostatics, where time is not a factor, but for oscillating fields the finite speed of propagation blurs the concept. In other words, stray inductive and ca- pacitive effects can enter significantly. Great care is then necessary to ensure a “good ground.” Another idealization in macroscopic electromagnetism is the idea of a surface charge density or a surface current density. The physical reality is that the charge or current is confined to the immediate neighborhood of the surface. If this region has thickness small compared to the length scale of interest, we may approximate the reality by the idealization of a region of infinitesimal thickness and speak of a surface distribution. Two different limits need to be distinguished. One is the limit in which the “surface” distribution is confined to a region near the surface that is macroscopically small, but microscopically large. An example is the pen- etration of time-varying fields into a very good, but not perfect, conductor, de- scribed in Section 8.1. It is found that the fields are confined to a thickness 6, called the skin depth, and that for high enough frequencies and good enough conductivities 6 can be macroscopically very small. It is then appropriate to in- tegrate the current density J over the direction perpendicular to the surface to obtain an effective surface current density K.y. The other limit is truly microscopic and is set by quantum-mechanical effects in the atomic structure of materials. Consider, for instance, the distribution of excess charge of a conducting body in electrostatics. It is well known that this charge lies entirely on the surface of a conductor. We then speak of a surface charge density o. There is no electric field inside the conductor, but there is, in accord with (1.17), a normal component of electric field just outside the surface. At the microscopic level the charge is not exactly at the surface and the field does not change discontinuously. The most elementary considerations would in- dicate that the transition region is a few atomic diameters in extent. The ions in a metal can be thought of as relatively immobile and localized to 1 angstrom or better; the lighter electrons are less constrained. The results of model cal- culations* are shown in Fig. L5. They come from a solution of the quantum- mechanical many-electron problem in which the ions of the conductor are approximated by a continuous constant charge density for x < 0. The electron density (r, = 5) is roughly appropriate to copper and the heavier alkali metals. *N. D. Lang and W. Kohn, Phys. Rev. B1, 4555 (1970); B3, 1215 (1971); V. E. Kenner, R. E. Allen, and W. M. Saslow, Phys. Lett. 38A, 255 (1972).Sect. 6 Some Remarks on Idealizations in Electromagnetism 21 Eg x (10-8cm) Figure LS Distribution of excess charge at the surface of a conductor and of the normal component of the electric field. The ions of the solid are confined to x <0 and are approximated by a constant continuous charge distribution through which the electrons move. The bulk of the excess charge is confined to within +2 A of the “surface.” ‘The excess electronic charge is seen to be confined to a region within +2 A of the “surface” of the ionic distribution. The electric field rises smoothly over this region to its value of o “outside” the conductor. For macroscopic situations where 10~° m is a negligible distance, we can idealize the charge density and electric field behavior as p(x) = o(x) and E,,(x) = 76(x)/€o, corresponding to a truly surface density and a step-function jump of the field. We see that the theoretical treatment of classical electromagnetism involves several idealizations, some of them technical and some physical. The subject of electrostatics, discussed in the first chapters of the book, developed as an exper- imental science of macroscopic electrical phenomena, as did virtually all other aspects of electromagnetism. The extension of these macroscopic laws, even for charges and currents in vacuum, to the microscopic domain was for the most part an unjustified extrapolation. Earlier in this introduction we discussed some of the limits to this extrapolation. The point to be made here is the following. With hindsight we know that many aspects of the laws of classical electromagnetism apply well into the atomic domain provided the sources are treated quantum mechanically, that the averaging of electromagnetic quantities over volumes con- taining large numbers of molecules so smooths the rapid fluctuations that static applied fields induce static average responses in matter, and that excess charge is on the surface of a conductor in a macroscopic sense. Thus Coulomb’s and Ampére’s macroscopic observations and our mathematical abstractions from them have a wider applicability than might be supposed by a supercautious phys-22 Introduction and Survey icist. The absence for air of significant electric or magnetic susceptibility certainly simplifies matters! References and Suggested Reading The history of electricity and magnetism is in large measure the history of science itself. We have already cited Whittaker’s two volumes, the first covering the period up to 1900, as well as the shorter account emphasizing optics in Born and Wolf. Another readable account, with perceptive discussion of the original experiments, is N. Feather, Electricity and Matter, University Press, Edinburgh (1968). The experimental tests of the inverse square nature of Coulomb’s law or, in modern language, the mass of the photon, are reviewed by I. Yu. Kobzarev and L. B. Okun’, Usp. Fiz. Nauk 95, 131 (1968) [transl., Sov. Phys. Usp. 11, 338 (1968).] and A. S. Goldhaber and M. M. Nieto, Rev. Mod. Phys. 43, 277 (1971). An accessible treatment of the gauge principle in the construction of field theories, building on classical electrodynamics and ordinary quantum mechanics, can be found in I.J.R. Aitchison and A. J. G. Hey, Gauge Theories in Particle Physics, 2nd ed., Adam Hilger, Bristol (1989). Suggested reading on the topic of the macroscopic Maxwell equations and their der- vation from the microscopic equations can be found at the end of Chapter 6. The basic physics of dielectrics, ferroelectrics, and magnetic materials can be found in numerous books on solid-state physics, for example, Ashcroft and Mermin Beam Kittel Wert and Thomson Wooten The second of these is aimed at electrical engineers and stresses practical topics like semiconductors. The last one is mainly on optical properties. The need for spatial non- locality in treating the surface impedance of metals (the anomalous skin effect) is discussed in several places by A. B. Pippard, Advances in Electronics and Electron Physics, Vol. VI, ed. L. Marton, Academic Press, New York (1954), pp. 1-45; Reports on Progress in Physics, Vol. XXIII, pp. 176-266 (1960); The Dynamics of Conduction Electrons, Gordon and Breach, New York (1965). The concept of a wave-vector and frequency-dependent dielectric constant ¢(k, «) is developed by Kittel, Advanced Topic D. D. Pines, Elementary Excitations in Solids, W. A. Benjamin, New York (1963), Chapters 3 and 4. F. Stern, Solid State Physics, Vol. 15, eds. F. Seitz and D. Turnbull, Academic Press, New York, pp. 299-408. The field of nonlinear optics is now nearly 40 years old. Beginnings and introductions can be found in J. A. Giordmaine, Phys. Today 22(1), 38 (1969). N. Bloembergen, Am. J. Phys. 35, 989 (1967).References 23 Nonlinear optical phenomena and applications are discussed in R. L. Sutherland, Handbook of Nonlinear Optics, Marcel Dekker, New York (1966). Some texts and monographs on the subject are R. W. Boyd, Nonlinear Optics, Academic Press, New York (1990). M. Schubert and B. Wilhelmi, Nonlinear Optics and Quantum Electronics, Wiley, New York (1986). Y.R. Shen, The Principles of Nonlinear Optics, Wiley, New York (1984).CHAPTER 1 Introduction to Electrostatics ‘We begin our discussion of electrodynamics with the subject of electrostatics— phenomena involving time-independent distributions of charge and fields. For most readers this material is in the nature of a review. In this chapter especially we do not elaborate significantly. We introduce concepts and definitions that are important for later discussion and present some essential mathematical appara- tus. In subsequent chapters the mathematical techniques are developed and applied. One point of physics should be mentioned. Historically, electrostatics devel- oped as a science of macroscopic phenomena. As indicated at the end of the Introduction, such idealizations as point charges or electric fields at a point must be viewed as mathematical constructs that permit a description of the phenomena at the macroscopic level, but that may fail to have meaning microscopically. 1.1 Coulomb’s Law All of electrostatics stems from the quantitative statement of Coulomb’s law concerning the force acting between charged bodies at rest with respect to each other. Coulomb, in an impressive series of experiments, showed experimentally that the force between two small charged bodies separated in air a distance large compared to their dimensions varies directly as the magnitude of each charge, varies inversely as the square of the distance between them, is directed along the line joining the charges, and is attractive if the bodies are oppositely charged and repulsive if the bodies have the same type of charge. Furthermore it was shown experimentally that the total force produced on one small charged body by a number of the other small charged bodies placed around it is the vector sum of the individual two-body forces of Coulomb. Strictly speak- ing, Coulomb’s conclusions apply to charges in vacuum or in media of negligible susceptibility. We defer consideration of charges in dielectrics to Chapter 4. 1.2 Electric Field Although the thing that eventually gets measured is a force, it is useful to intro- duce a concept one step removed from the forces, the concept of an electric field due to some array of charged bodies. At the moment, the electric field can beSect. 1.2 Electric Field 25 defined as the force per unit charge acting at a given point. It is a vector function of position, denoted by E. One must be careful in its definition, however. It is not necessarily the force that one would observe by placing one unit of charge on a pith ball and placing it in position. The reason is that one unit of charge may be so large that its presence alters appreciably the field configuration of the array. Consequently one must use a limiting process whereby the ratio of the force on the small test body to the charge on it is measured for smaller and smaller amounts of charge.* Experimentally, this ratio and the direction of the force will become constant as the amount of test charge is made smaller and smaller. These limiting values of magnitude and direction define the magnitude and direction of the electric field E at the point in question. In symbols we may write F=qE (1.1) where F is the force, E the electric field, and q the charge. In this equation it is assumed that the charge q is located at a point, and the force and the electric field are evaluated at that point. Coulomb’s law can be written down similarly. If F is the force on a point charge qj, located at x,, due to another point charge qz, located at x), then Coulomb’s law is x — Xx F = kqiqo hoe? (1.2) Note that q; and q> are algebraic quantities, which can be positive or negative. The constant of proportionality k depends on the system of units used. The electric field at the point x due to a point charge q, at the point x, can be obtained directly: Xi E(x) = kay Zone (1.3) as indicated in Fig. 1.1. The constant k differs in different systems of units.’ In electrostatic units (esu), k = 1 and unit charge is chosen as that charge that exerts a force of one dyne on an equal point charge located one centimeter away. The esu unit of charge is called the statcoulomb, and the electric field is measured in statvolts per centimeter. In the SI system, which we employ here, k = (477€))~! = 10°7c2, where €, ~ 8.854 X 107" farad per meter (F/m) is called the permittivity of free space. The SI unit of charge is the coulomb (C), and the electric field is measured in volts per meter (V/m). One coulomb (1 C) produces an electric field Figure 1.1 +The discreteness of electric charge (see Section 1.1) means that this mathematical limit is impossible to realize physically. This is an example of a mathematical idealization in macroscopic electrostatics. "The question of units is discussed in detail in the Appendix.26 Chapter 1 Introduction to Electrostatics—SI of approximately 8.9874 x 10° V/m (8.9874 GV/m) at a distance of 1 meter. One electron (q ~ 1.602 x 10°! C) produces a field of approximately 1.44 x 107° Vim (1.44 nV/m) at 1 meter. The experimentally observed linear superposition of forces due to many charges means that we may write the electric field at x due to a system of point charges q;, located at x;, i = 1, 2,...,n, as the vector sum: x-X Ew) = 3S ce) 4m, "Ix = x)) If the charges are so small and so numerous that they can be described by a charge density p(x’) [if Aq is the charge in a small volume Ax Ay Az at the point x’, then Aq = p(x’) Ax Ay Az], the sum is replaced by an integral: 1 | x-x' E(x) = —— | ox’) 3 ax 15 0) = Fe J oe) pos as) where d°x' = dx' dy’ dz’ is a three-dimensional volume element at x’. At this point it is worthwhile to introduce the Dirac delta function. In one dimension, the delta function, written 6(x —a), is a mathematically improper function having the properties: 1. &(x — a) = Oforx +a, and 2. J 6(x — a) dx = 1 if the region of integration includes x = a, and is zero otherwise. The delta function can be given an intuitive, but nonrigorous, meaning as the limit of a peaked curve such as a Gaussian that becomes narrower and narrower, but higher and higher, in such a way that the area under the curve is always constant. L. Schwartz’s theory of distributions is a comprehensive rigorous mathematical approach to delta functions and their manipulations.* From the definitions above it is evident that, for an arbitrary function f(x), 3. S f(x) 8x — a) dx = f(a). The integral of f(x) times the derivative of a delta function is simply understood if the delta function is thought of as a well-behaved, but sharply peaked, function. Thus the definition is 4. J f(&) 8x ~ a) dx = —f'(a) where a prime denotes differentiation with respect to the argument. If the delta function has as argument a function f(x) of the independent variable x, it can be transformed according to the rule, 5. af) = a(x — x) where f(x) is assumed to have only simple zeros, located at x = x). In more than one dimension, we merely take products of delta functions in each dimension. In three dimensions, for example, with Cartesian coordinates, 6. (x — X) = B(x, =X) B(x» — Xp) A(x — X) *A useful, rigorous account of the Dirac delta funetion is given by Lighthill. See also Dennery and Krzywicki (Section 111.13). (Full references for items cited in the text or footnotes by italicized author only will be found in the Bibliography.)Sect. 1.3 Gauss’s Law 27 is a function that vanishes everywhere except at x = X, and is such that 1 if AV contains x = X - By = a Lu 5x ~ X) ax {i if AV does not contain x = X Note that a delta function has the dimensions of an inverse volume in whatever number of dimensions the space has. A discrete set of point charges can be described with a charge density by means of delta functions. For example, p(x) = 5 qi (x — X,) (1.6) represents a distribution of n point charges q;, located at the points x,. Substitution of this charge density (1.6) into (1.5) and integration, using the properties of the delta function, yields the discrete sum (1.4). 1.3 Gauss’s Law The integral (1.5) is not always the most suitable form for the evaluation of electric fields. There is another integral result, called Gauss’s law, which is some- times more useful and furthermore leads to a differential equation for E(x). To obtain Gauss’s law we first consider a point charge q and a closed surface S, as shown in Fig. 1.2. Let r be the distance from the charge to a point on the surface, n be the outwardly directed unit normal to the surface at that point, da be an s 7 qg outside S q inside $ Figure 1.2 Gauss’s law. The normal component of electric field is integrated over the closed surface S. If the charge is inside (outside) 5, the total solid angle subtended at the charge by the inner side of the surface is 4 (zero).28 Chapter 1 Introduction to Electrostatics—SI element of surface area. If the electric field E at the point on the surface due to the charge q makes an angle @ with the unit normal, then the normal component of E times the area element is: aL (1.7) -nda= Bemdam ar Since E is directed along the line from the surface element to the charge q, cos @ da = r? dQ, where dQ is the element of solid angle subtended by da at the position of the charge. Therefore q E-nda=-— dQ 1.8) ne bre, c If we now integrate the normal component of E over the whole surface, it is easy to see that § Benda = {1% — ifq lies inside $ as) s 0 if q lies outside S This result is Gauss’s law for a single point charge. For a discrete set of charges, it is immediately apparent that 1 E-nda=— A cae) fe-nde=+S q 2.10) where the sum is over only those charges inside the surface S. For a continuous charge density p(x), Gauss’s law becomes: 1 $ Enda =~ [ p(x) dx (1.11) where V is the volume enclosed by S. Equation (1.11) is one of the basic equations of electrostatics. Note that it depends upon the inverse square law for the force between charges, the central nature of the force, and the linear superposition of the effects of different charges. Clearly, then, Gauss’s law holds for Newtonian gravitational force fields, with matter density replacing charge density. It is interesting to note that, even before the experiments of Cavendish and Coulomb, Priestley, taking up an observation of Franklin that charge seemed to reside on the outside, but not the inside, of a metal cup, reasoned by analogy with Newton’s law of universal gravitation that the electrostatic force must obey an inverse square law with distance. The present status of the inverse square law is discussed in Section 1.2. 1.4 Differential Form of Gauss’s Law Gauss’s law can be thought of as being an integral formulation of the law of electrostatics. We can obtain a differential form (ie., a differential equation) bySect. 1.5 Another Equation of Electrostatics and the Scalar Potential 29 using the divergence theorem. The divergence theorem states that for any well- behaved vector field A(x) defined within a volume V surrounded by the closed surface S the relation $ A-nda =| VAdx s v holds between the volume integral of the divergence of A and the surface integral of the outwardly directed normal component of A. The equation in fact can be used as the definition of the divergence (see Stratton, p. 4). To apply the divergence theorem we consider the integral relation expressed in Gauss’s theorem: 1 § E-nda=+{ p(x) dx s & Jv Now the divergence theorem allows us to write this as I, (V +E — ple) d°x = 0 (1.12) for an arbitrary volume V. We can, in the usual way, put the integrand equal to zero to obtain V-E = ple, (1.13) which is the differential form of Gauss’s law of electrostatics. This equation can itself be used to solve problems in electrostatics. However, it is often simpler to deal with scalar rather then vector functions of position, and then to derive the vector quantities at the end if necessary (see below). 1.5 Another Equation of Electrostatics and the Scalar Potential The single equation (1.13) is not enough to specify completely the three com- ponents of the electric field E(x). Perhaps some readers know that a vector field can be specified almost* completely if its divergence and curl are given every- where in space. Thus we look for an equation specifying curl E as a function of position. Such an equation, namely, VxE=0 (1.14) follows directly from our generalized Coulomb’s law (1.5): Be) = ze | ee) SS The vector factor in the integrand, viewed as a function of x, is the negative gradient of the scalar 1/|x — x’ |: “Up to the gradient of a scalar function that satisfies the Laplace equation. See Section 1.9 on uniqueness.30 Chapter | Introduction to Electrostatics—SI Since the gradient operation involves x, but not the integration variable x’, it can be taken outside the integral sign. Then the field can be written -1 P(X’) aye E(x) = ra Vv k-x] ax (1.15) Since the curl of the gradient of any well-behaved scalar function of position vanishes (V x Vy = 0, for all J), (1.14) follows immediately from (1.15). Note that V x E = 0 depends on the central nature of the force between charges, and on the fact that the force is a function of relative distances only, but does not depend on the inverse square nature. In (1.15) the electric field (a vector) is derived from a scalar by the gradient operation. Since one function of position is easier to deal with than three, it is worthwhile concentrating on the scalar function and giving it a name. Conse- quently we define the scalar potential ®(x) by the equation: E=-vo (1.16) Then (1.15) shows that the scalar potential is given in terms of the charge density by p(x’) , (x) = ae Row? (1.17) where the integration is over all charges in the universe, and ® is arbitrary only to the extent that a constant can be added to the right-hand side of (1.17). The scalar potential has a physical interpretation when we consider the work done on a test charge q in transporting it from one point (A) to another point (B) in the presence of an electric field E(x), as shown in Fig, 1.3. The force acting on the charge at any point is F=qE so that the work done in moving the charge from A to B is 2 B w=-f F-dl=—4 f E-dl (1.18) A A ‘The minus sign appears because we are calculating the work done on the charge against the action of the field. With definition (1.16) the work can be written B B weafe vo-d=a{, d® = q(®z — ®,) (1.19) Figure 1.3Sect. 1.6 Surface Distributions of Charges and Dipoles 31 which shows that g® can be interpreted as the potential energy of the test charge in the electrostatic field. From (1.18) and (1.19) it can be seen that the line integral of the electric field between two points is independent of the path and is the negative of the potential difference between the points: B [. E-dl = ~(®, - %,) (1.20) This follows directly, of course, from definition (1.16). If the path is closed, the line integral is zero, $ E-dl=0 (1.21) a result that can also be obtained directly from Coulomb’s law. Then application of Stokes’s theorem [if A(x) is a well-behaved vector field, S is an arbitrary open surface, and C is the closed curve bounding S, es Ue where dl is a line element of C, n is the normal to S, and the path C is traversed in a right-hand screw sense relative to n] leads immediately back to V x E = 0. 1.6 Surface Distributions of Charges and Dipoles and Discontinuities in the Electric Field and Potential One of the common problems in electrostatics is the determination of electric field or potential due to a given surface distribution of charges. Gauss’s law (1.11) allows us to write down a partial result directly. If a surface S, with a unit normal n directed from side 1 to side 2 of the surface, has a surface-charge density of o(x) (measured in coulombs per square meter) and electric fields E, and E) on either side of the surface, as shown in Fig. 1.4, then Gauss’s law tells us imme- diately that (E, — E,) + = ofey (1.22) This does not determine E, and E, unless there are no other sources of field and the geometry and form of o are especially simple. All that (1.22) says is that there Figure 1.4 Discontinuity in the normal component of electric field across a surface layer of charge.32 Chapter 1 Introduction to Electrostatics—SI is a discontinuity of o/e, in the normal component of electric field in crossing a surface with a surface-charge density , the crossing being made in the direction of n. The tangential component of electric field can be shown to be continuous across a boundary surface by using (1.21) for the line integral of E around a closed path. It is only necessary to take a rectangular path with negligible ends and one side on either side of the boundary. An expression for the potential (hence the field, by differentiation) at any point in space (not just at the surface) can be obtained from (1.17) by replacing p dx by oda: -t ff @) yy oa = [, 7 de (1.23) For volume or surface distributions of charge, the potential is everywhere con- tinuous, even within the charge distribution. This can be shown from (1.23) or from the fact that E is bounded, even though discontinuous across a surface distribution of charge. With point or line charges, or dipole layers, the potential is no longer continuous, as will be seen immediately. Another problem of interest is the potential due to a dipole-layer distribution on a surface S. A dipole layer can be imagined as being formed by letting the surface § have a surface-charge density o(x) on it, and another surface S’, lying close to S, have an equal and opposite surface-charge density on it at neighboring points, as shown in Fig. 1.5. The dipole-layer distribution of strength D(x) is formed by letting S’ approach infinitesimally close to S while the surface-charge density a(x) becomes infinite in such a manner that the product of o(x) and the local separation d(x) of S and S’ approaches the limit D(x): lim (x) d(x) = D(x) d(x)>0 The direction of the dipole moment of the layer is normal to the surface S and in the direction going from negative to positive charge. To find the potential due to a dipole layer we can consider a single dipole and then superpose a surface density of them, or we can obtain the same result by performing mathematically the limiting process described in words above on the surface-density expression (1.23). The first way is perhaps simpler, but the second gives useful practice in vector calculus. Consequently we proceed with 8 s a(x) d(x) \s Figure 1.5 Limiting process involved in 8’ "creating a dipole layer.Sect. 1.6 Surface Distributions of Charges and Dipoles 33 Figure 1.6 Dipole-layer geometry. the limiting process. With n, the unit normal to the surface S, directed away from S', as shown in Fig. 1.6, the potential due to the two close surfaces is ) 1 + nd| da 4rre, Js: a re Js a(x’) (x) For small d we can expand |x — x’ + nd|~!. Consider the general expression |x + al-', where Ja] << |x|, We write a Taylor series expansion in three dimensions: 1 oli y.y(t)y... t+al ox ey In this way we find that as d — 0 the potential becomes (1.24) 1 . , 0) =f dem ¥( In passing we note that the integrand in (1.24) is the potential of a point dipole with dipole moment p = n D da’. The potential at x caused by a dipole p at x’ is 1 p-(x-x’) pi@—x) (1.25) °0) = tre k- Equation (1.24) has a simple geometrical interpretation. We note that 6 da’ _ 00s 0da’ ag where dQ. is the element of solid angle subtended at the observation point by the area element da’, as indicated in Fig. 1.7. Note that dQ has a positive sign if @ is Figure 1.7. The potential at P due to the dipole layer D on the area element da’ is just the negative product of D and the solid angle element dQ subtended by da’ at P.34 Chapter 1 Introduction to Electrostatics—SI an acute angle (i.e., when the observation point views the “inner” side of the dipole layer). The potential can be written: 1 , (x) = re i: D(x') do (1.26) For a constant surface-dipole-moment density D, the potential is just the product of the moment divided by 4re, and the solid angle subtended at the observation point by the surface, regardless of its shape. There is a discontinuity in potential in crossing a double layer. This can be seen by letting the observation point come infinitesimally close to the double layer. The double layer is now imagined to consist of two parts, one being a small disc directly under the observation point. The disc is sufficiently small that it is sensibly flat and has constant surface-dipole-moment density D. Evidently the total potential can be obtained by linear superposition of the potential of the disc and that of the remainder. From (1.26) it is clear that the potential of the disc alone has a discontinuity of D/e, in crossing from the inner to the outer side, being —D/2e, on the inner side and +D/2e, on the outer. The potential of the remainder alone, with its hole where the disc fits in, is continuous across the plane of the hole. Consequently the total potential jump in crossing the sur- face is: @, — ®, = Die (1.27) This result is analogous to (1.22) for the discontinuity of electric field in crossing a surface-charge density. Equation (1.27) can be interpreted “physically” as a potential drop occurring “inside” the dipole layer; it can be calculated as the product of the field between the two layers of surface charge times the separation before the limit is taken. 1.7 Poisson and Laplace Equations In Sections 1.4 and 1.5 it was shown that the behavior of an electrostatic field can be described by the two differential equations: V-E= ple (1.13) and VxE=0 (1.14) the latter equation being equivalent to the statement that E is the gradient of a scalar function, the scalar potential ®: E=-vo (1.16) Equations (1.13) and (1.16) can be combined into one partial differential equation for the single function (x): Vb = —pley (1.28) This equation is called the Poisson equation. In regions of space that lack a charge density, the scalar potential satisfies the Laplace equation: Vo =0 (1.29)Sect.1.8 Green's Theorem 35 We already have a solution for the scalar potential in expression (1.17): 1 fp) os @(x) = Tee ee eG (x) Gre, J ix—x']“* (117) To verify directly that this does indeed satisfy the Poisson equation (1.28), we operate with the Laplacian on both sides. Because it turns out that the resulting integrand is singular, we invoke a limiting procedure. Define the “a-potential” ®,(x) by 1 , . p(x’) wx’ dmc) Vex Pra The actual potential (1.17) is then the limit of the “‘a-potential” as a > 0. Taking the Laplacian of the “epoca gives 1 a? oar "he px ee ae where r = |x — The square-bracketed expression is the negative Laplacian of Vr + a®. It is well-behaved everywhere for nonvanishing a, but as a tends to zero it becomes infinite at r = 0 and vanishes for r # 0. It has a volume integral equal to 47 for arbitrary a. For the purposes of integration, divide space into two regions by a sphere of fixed radius R centered on x. Choose P such that p(x’) changes little over the interior of the sphere, and imagine a much smaller than R and tending toward zero. If p(x‘) is such that (1.17) exists, the contribution to the integral (1.30) from the exterior of the sphere will vanish like a” as a > 0. We thus need consider only the contribution from inside the sphere. With a Taylor series expansion of the well-behaved p(x’) around x’ = x, one finds 1(* 3a P Vb,(x) = — 7 [ era [oe + Get: ®,(x) Pdr + O(@) = Direct integration yields > 1 V°O,(x) = —= p(x) (1 + O(@IR?)) + O(a”, a’log a) Vp + +++ ‘0 In the limit a — 0, we obtain the Poisson equation (1.28). The singular nature of the Laplacian of 1/r can be exhibited formally in terms of a Dirac delta function. Since V*(1/r) = 0 for r # 0 and its volume integral is 47, we can write the formal equation, V°(I/r) = —4778(x) or, more generally, = —478(x — x’) (1.31) 1.8 Green’s Theorem If electrostatic problems always involved localized discrete or continuous distri- butions of charge with no boundary surfaces, the general solution (1.17) would36 Chapter 1 Introduction to Electrostatics—SI be the most convenient and straightforward solution to any problem. There would be no need of the Poisson or Laplace equation. In actual fact, of course, many, if not most, of the problems of electrostatics involve finite regions of space, with or without charge inside, and with prescribed boundary conditions on the bounding surfaces. These boundary conditions may be simulated by an appro- priate distribution of charges outside the region of interest (perhaps at infinity), but (1.17) becomes inconvenient as a means of calculating the potential, except in simple cases (e.g., method of images). To handle the boundary conditions it is necessary to develop some new math- ematical tools, namely, the identities or theorems due to George Green (1824). These follow as simple applications of the divergence theorem. The divergence theorem: [v-aar=$ a-nda v s applies to any well-behaved vector field A defined in the volume V bounded by the closed surface S. Let A = Vis, where ¢ and ip are arbitrary scalar fields. Now Vs (GVW) = 6 VN + Vb+ Vb (1.32) and aoe oVe-n= > on (1.33) where /dn is the normal derivative at the surface S (directed outward from inside the volume V). When (1.32) and (1.33) are substituted into the divergence the- orem, there results Green’s first identity: a [ (6 Vs + Vb Vy) dx = $ o ae da (1.34) If we write down (1.34) again with ¢ and y interchanged, and then subtract it from (1.34), the V+ Vy terms cancel, and we obtain Green’s second identity or Green’s theorem: 2 24) dix = oy y ob Love WV) d’x $ [+2 #2 da (1.35) The Poisson differential equation for the potential can be converted into an integral equation if we choose a particular y, namely 1/R = 1/|x — x’|, where x is the observation point and x’ is the integration variable. Further, we put ¢ = ®, the scalar potential, and make use of Vd = —p/ey. From (1.31) we know that V(I/R) = —478(x — x’), so that (1.35) becomes 1 1\_1a® i [-+706 dx —x) +R ots')| x! = ¢ [ e (3) - 22 da’ Tf the point x lies within the volume V, we obtain: 1 fe) py, 1G [2 9 (1) ae ®0) = Tra ly Ro + aa Ss [Ran Pan \R) | 03Sect. 1.9 Uniqueness of the Solution with Dirichlet or Neumann Boundary Conditions 37 If x lies outside the surface S, the left-hand side of (1.36) is zero.* [Note that this is consistent with the interpretation of the surface integral as being the potential due to a surface-charge density « = €y ¢b/an' and a dipole layer D = —e,. The discontinuities in electric field and potential (1.22) and (1.27) across the surface then lead to zero field and zero potential outside the volume V.] Two remarks are in order about result (1.36). First, if the surface S goes to infinity and the electric field on S falls off faster than R~', then the surface integral vanishes and (1.36) reduces to the familiar result (1.17). Second, for a charge- free volume, the potential anywhere inside the volume (a solution of the Laplace equation) is expressed in (1.36) in terms of the potential and its normal derivative only on the surface of the volume. This rather surprising result is not a solution to a boundary-value problem, but only an integral statement, since the arbitrary specification of both @ and ab/an (Cauchy boundary conditions) is an overspe- cification of the problem. This is discussed in detail in the next sections, where techniques yielding solutions for appropriate boundary conditions are developed using Green’s theorem (1.35). 1.9 Uniqueness of the Solution with Dirichlet or Neumann Boundary Conditions What boundary conditions are appropriate for the Poisson (or Laplace) equation to ensure that a unique and well-behaved (i.e., physically reasonable) solution will exist inside the bounded region? Physical experience leads us to believe that specification of the potential on a closed surface (e.g., a system of conductors held at different potentials) defines a unique potential problem. This is called a Dirichlet problem, or Dirichlet boundary conditions. Similarly it is plausible that specification of the electric field (normal derivative of the potential) everywhere on the surface (corresponding to a given surface-charge density) also defines a unique problem. Specification of the normal derivative is known as the Neumann boundary condition. We now proceed to prove these expectations by means of Green’s first identity (1.34). We want to show the uniqueness of the solution of the Poisson equation, V°@ = —p/eo, inside a volume V subject to either Dirichlet or Neumann boundary conditions on the closed bounding surface S. We suppose, to the contrary, that there exist two solutions ®, and ®, satisfying the same boundary conditions. Let U=,- 9, (1.37) Then V°U = 0 inside V, and U = 0 or aU/an = 0 on S for Dirichlet and Neumann boundary conditions, respectively. From Green’s first identity (1.34), with @ = = U, we find i (U VU + VU + VU) dx = f. ve da (1.38) “The reader may complain that (1.36) has been obtained in an illegal fashion since 1/|x — x" | is not well-behaved inside the volume V. Rigor can be restored by using a limiting process, as in the pre- ceding section, or by excluding a small sphere around the offending point, x = x’. The result is still (1.36).38 Chapter 1 Introduction to Electrostatics—SI With the specified properties of U, this reduces (for both types of boundary condition) to: | |VUP a°x = 0 v which implies VU = 0. Consequently, inside V, U is constant. For Dirichlet boundary conditions, U = 0 on S so that, inside V, &; = ®, and the solution is unique. Similarly, for Neumann boundary conditions, the solution is unique, apart from an unimportant arbitrary additive constant. From the right-hand side of (1.38) it is evident that there is also a unique solution to a problem with mixed boundary conditions (i.e., Dirichlet over part of the surface S, and Neumann over the remaining part). It should be clear that a solution to the Poisson equation with both ® and ac/an specified arbitrarily on a closed boundary (Cauchy boundary conditions) does not exist, since there are unique solutions for Dirichlet and Neumann con- ditions separately and these will in general not be consistent. This can be verified with (1.36). With arbitrary values of © and a¢/an inserted on the right-hand side, it can be shown that the values of &(x) and V(x) as x approaches the surface are in general inconsistent with the assumed boundary values. The question of whether Cauchy boundary conditions on an open surface define a unique elec- trostatic problem requires more discussion than is warranted here. The reader may refer to Morse and Feshbach (Section 6.2, pp. 692-706) or to Sommerfeld (Partial Differential Equations in Physics, Chapter Il) for a detailed discussion of these questions. The conclusion is that electrostatic problems are specified only by Dirichlet or Neumann boundary conditions on a closed surface (part or all of which may be at infinity, of course). 1.10 Formal Solution of Electrostatic Boundary- Value Problem with Green Function The solution of the Poisson or Laplace equation in a finite volume V with either Dirichlet or Neumann boundary conditions on the bounding surface S can be obtained by means of Green’s theorem (1.35) and so-called Green functions. In obtaining result (1.36)—not a solution—we chose the function to be 1/|x — x’|, it being the potential of a unit point source, satisfying the equation: 1 v?{ ——__ } = -4n8(x — x’ 1.31 (5 = zi) 75(x — x’) (131) The function 1/|x — x’| is only one of a class of functions depending on the variables x and x’, and called Green functions, which satisfy (1.31). In general, V?G(x, x') = —48(x — x’) (1.39) where G(x, x’) = + F(x, x’) (1.40) Ix x’] with the function F satisfying the Laplace equation inside the volume V: V?F(x, x!) = 0 (1.41)Sect. 1.10 Formal Solution of Electrostatic Boundary-Value Problem with Green Function 39 In facing the problem of satisfying the prescribed boundary conditions on & or a@/an, we can find the key by considering result (1.36). As has been pointed out already, this is not a solution satisfying the correct type of boundary condi- tions because both ® and a@/an appear in the surface integral. It is at best an integral relation for ©. With the generalized concept of a Green function and its additional freedom [via the function F(x, x’)], there arises the possibility that we can use Green’s theorem with y= G(x, x’) and choose F(x, x') to eliminate one or the other of the two surface integrals, obtaining a result that involves only Dirichlet or Neumann boundary conditions. Of course, if the necessary G(x, x’) depended in detail on the exact form of the boundary conditions, the method would have little generality. As will be seen immediately, this is not required, and G(x, x’) satisfies rather simple boundary conditions on S. With Green’s theorem (1.35), @ = ®, = G(x, x’), and the specified prop- erties of G (1.39), it is simple to obtain the generalization of (1.36): (x) = i I plx')G(x, x") ax" 1 Mb ; tah, [61 x’) = = Ox’) aa | da’ The freedom available in the definition of G (1.40) means that we can make the surface integral depend only on the chosen type of boundary conditions. Thus, for Dirichlet boundary conditions we demand: Gp(x, x’) = 0 forx' on S (1.43) Then the first term in the surface integral in (1.42) vanishes and the solution is . all , " dix! — i¢ 1») Gp aa x) = Fe J, POG x’) d TP, PO) Gr da’ (1.44) For Neumann boundary conditions we must be more careful. The obvious choice of boundary condition on G(x, x’) seems to be dGy mn’ (x,x')=0 forx’ on S since that makes the second term in the surface integral in (1.42) vanish, as de- sired. But an application of Gauss’s theorem to (1.39) shows that aG $ = da! = — 40 s on’ Consequently the simplest allowable boundary condition on Gy is dGy ee , aan x!) = =F for x’ on S (1.45) where S is the total area of the boundary surface. Then the solution is rt I : oe $ ab ; - — Px! +— 9 — ’ (x) = (®)s + tra ly A(x')Gu(x, x’) dx’ + FP 5; Gy da’ (1.46) where (@)s is the average value of the potential over the whole surface. The customary Neumann problem is the so-called exterior problem in which the vol-40 Chapter 1 Introduction to Electrostatics—SI ume V is bounded by two surfaces, one closed and finite, the other at infinity. Then the surface area S is infinite; the boundary condition (1.45) becomes ho- mogeneous; the average value (®), vanishes. We note that the Green functions satisfy simple boundary conditions (1.43) or (1.45) which do not depend on the detailed form of the Dirichlet (or Neumann) boundary values. Even so, it is often rather involved (if not impossible) to de- termine G(x, x’) because of its dependence on the shape of the surface S. We will encounter such problems in Chapters 2 and 3. The mathematical symmetry property G(x, x’) = G(x', x) can be proved for the Green functions satisfying the Dirichlet boundary condition (1.43) by means of Green’s theorem with = G(x, y) and y= G(x’, y), where y is the integration variable. Since the Green function, as a function of one of its variables, is a potential due to a unit point source, the symmetry merely represents the physical interchangeability of the source and the observation points. For Neumann boundary conditions the symmetry is not automatic, but can be imposed as a separate requirement.* As a final, important remark we note the physical meaning of F(x, x')/477€. It is a solution of the Laplace equation inside V and so represents the potential of a system of charges external to the volume V. It can be thought of as the potential due to an external distribution of charges chosen to satisfy the homo- geneous boundary conditions of zero potential (or zero normal derivative) on the surface S when combined with the potential of a point charge at the source point x’. Since the potential at a point x on the surface due to the point charge depends on the position of the source point, the external distribution of charge F(x, x') must also depend on the “parameter” x'. From this point of view, we see that the method of images (to be discussed in Chapter 2) is a physical equivalent of the determination of the appropriate F(x, x’) to satisfy the bound- ary conditions (1.43) or (1.45). For the Dirichlet problem with conductors, F(x, x')/47ey can also be interpreted as the potential due to the surface-charge distribution induced on the conductors by the presence of a point charge at the source point x’. 1.11 Electrostatic Potential Energy and Energy Density; Capacitance In Section 1.5 it was shown that the product of the scalar potential and the charge of a point object could be interpreted as potential energy. More precisely, if a point charge q, is brought from infinity to a point x, in a region of localized electric fields described by the scalar potential (which vanishes at infinity), the work done on the charge (and hence its potential energy) is given by W; = g(x) (1.47) The potential ® can be viewed as produced by an array of (n — 1) charges qj = 1,2,...,n — 1) at positions x,. Then (1.48) *See K.-J. Kim and J. D. Jackson, Am. J. Phys. 61, (12) 1144-1146 (1993).Sect. 1.11 Electrostatic Potential Energy and Energy Density; Capacitance 41 so that the potential energy of the charge q; is s ee f (1.49) Ix = =a The toral potential energy of all the charges due to all the forces acting between them is: i 44 vo 4m) A [x — x1 x; (1.50) as can be seen most easily by adding each charge in succession. A more symmetric form can be written by summing over i and j unrestricted, and then dividing by 2: (1.51) It is understood that i = j terms (infinite “self-energy” terms) are omitted in the double sum. For a continuous charge distribution [or, in general, using the Dirac delta functions (1.6)] the potential energy takes the form: af [eh dot” > _ (1.52) Another expression, equivalent to (1.52), can be obtained by noting that one of the integrals in (1.52) is just the scalar potential (1.17). Therefore © Bre We ; / p(x) P(x) dx (1.53) Equations (1.51), (1.52), and (1.53) express the electrostatic potential energy in terms of the positions of the charges and so emphasize the interactions between charges via Coulomb forces. An alternative, and very fruitful, approach is to emphasize the electric field and to interpret the energy as being stored in the electric field surrounding the charges. To obtain this latter form, we make use of the Poisson equation to eliminate the charge density from (1.53): We =e OVO dix Integration by parts leads to the result: = of |VOP dx = 2 | [EP dx (1.54) 2 2 where the integration is over all space. In (1.54) all explicit reference to charges has gone, and the energy is expressed as an integral of the square of the electric field over all space. This leads naturally to the identification of the integrand as an energy density w: w=2 2 IEP (1.55) This expression for energy density is wsaively reasonable, since regions of high fields “must” contain considerable energy. There is perhaps one puzzling thing about (1.55). The energy density is pos-42 Chapter 1 Introduction to Electrostatics—SI itive definite. Consequently its volume integral is necessarily nonnegative. This seems to contradict our impression from (1.51) that the potential energy of two charges of opposite sign is negative. The reason for this apparent contradiction is that (1.54) and (1.55) contain “self-energy” contributions to the energy density, whereas the double sum in (1.51) does not. To illustrate this, consider two point charges q, and q, located at x, and x3, as in Fig. 1.8. The electric field at the point P with coordinate x is 1 a&—m), 1 ax * Ges [x =x, * Fre, [x — x) so that the energy density (1.55) is Gi @ aX ~ Xi) + (X = X2) u +2 1.56 Ix =x 0 [x = x2)? Ix — xP [x — x} co? 32m ew = Clearly the first two terms are “self-energy” contributions. To show that the third term gives the proper result for the interaction potential energy we integrate over all space: 142 (x = &) lone) J |x — xP [x — Win = Bx (1.57) A change of integration variable to p = (x — x,)/|x, — x»| yields 1 N92 1 fe-(@+n) a | eee 7 aa! ale 3 ap (1.58) Va Ae, |x, — x] where n is a unit vector in the direction (x, — x,). Using the fact that (p + n)/ |p + n> = —V,(1/|p + n|), the dimensionless integral can easily be shown to have the value 47, so that the interaction energy reduces to the expected value. Forces acting between charged bodies can be obtained by calculating the change in the total electrostatic energy of the system under small virtual displace- ments. Examples of this are discussed in the problems. Care must be taken to exhibit the energy in a form showing clearly the factors that vary with a change in configuration and those that are kept constant. As a simple illustration we calculate the force per unit area on the surface of a conductor with a surface-charge density (x). In the immediate neighbor- hood of the surface the energy density is we S JEP = 07/2€ (1.59) If we now imagine a small outward displacement Ax of an elemental area Aa of the conducting surface, the electrostatic energy decreases by an amount that is the product of energy density w and the excluded volume Ax Aa: AW = ~0? Aa Ax/2e, (1.60) 1 0 Figure 18Sect. 1.12 Variational Approach to the Solution of the Laplace and Poisson Equations 43 This means that there is an outward force per unit area equal to o7/2e) = w at the surface of the conductor. This result is normally derived by taking the product of the surface-charge density and the electric field, with care taken to eliminate the electric field due to the element of surface-charge density itself. For a system of n conductors, each with potential V; and total charge Q; (i = 1, 2,..., n) in otherwise empty space, the electrostatic potential energy can be expressed in terms of the potentials alone and certain geometrical quan- tities called coefficients of capacity. For a given configuration of the conductors, the linear functional dependence of the potential on the charge density implies that the potential of the ith conductor can be written as V; >, Pu, G=1,2,...,”) FE where the p,; depend on the geometry of the conductors. These n equations can be inverted to yield the charge on the ith conductor in terms of all the potentials: O—> GVy 12) 1) (1.61) The coefficients C,, are called capacities or capacitances while the C,, i # j, are called coefficients of induction. The capacitance of a conductor is therefore the total charge on the conductor when it is maintained at unit potential, all other conductors being held at zero potential. Sometimes the capacitance of a system of conductors is also defined. For example, the capacitance of two conductors carrying equal and opposite charges in the presence of other grounded conduc- tors is defined as the ratio of the charge on one conductor to the potential dif- ference between them. The equations (1.61) can be used to express this capaci- tance in terms of the coefficients C;. The potential energy (1.53) for the system of conductors is w-!3 av.-5 353 cy, (1.62) 251 2A The expression of the energy in terms of the potentials V; and the C;j, or in terms of the charges Q, and the coefficients p,, permits the application of variational methods to obtain approximate values of capacitances. It can be shown, based on the technique of the next section (see Problems 1.17 and 1.18), that there are variational principles giving upper and lower bounds on C;;. The principles permit estimation with known error of the capacitances of relatively involved configu- rations of conductors. High-speed computational techniques permit the use of elaborate trial functions involving several parameters. It must be remarked, how- ever, that the need for a Green function satisfying Dirichlet boundary conditions in the lower bound makes the error estimate nontrivial. Further consideration of this technique for calculating capacitances is left to the problems at the end of this and subsequent chapters. 1.12 Variational Approach to the Solution of the Laplace and Poisson Equations Variational methods play prominent roles in many areas of classical and quantum physics. They provide formal techniques for the derivation of “equations of mo-44 Chapter 1 Introduction to Electrostatics—SI tion” and also practical methods for obtaining approximate, but often accurate, solutions to problems not amenable to other approaches. Estimates of resonant frequencies of acoustic resonators and energy eigenvalues of atomic systems come readily to mind. The far-reaching concept that physical systems in equilibrium have minimal energy content is generalized to the consideration of energy-like functionals. As an example, consider the functional Ty] = sf V+ Vip dx — , gy dex (1.63) where the function ¢(x) is well-behaved inside the volume V and on its surface S (which may consist of several separate surfaces), and g(x) is a specified “source” function without singularities within V. We now examine the first-order change in the functional when we change > y + 5ip, where the modification 6yp(x) is infinitesimal within V. The difference 81 = I[w + dy] — I [is al = [ V+ VS) d?x — i gop d?x +++ (1.64) The neglected term is semipositive definite and is second order in Sip. Use of Green’s first identity with 6 = dy and y= w yields a a= iE [-Vu — g] dy dx + ¢ oy da (1.65) Provided dys = 0 on the boundary surface S (so that the surface integral vanishes), the first-order change in I[¥] vanishes if yp(x) satisfies Vu=—-g (1.66) Recalling that the neglected term in (1.64) is semipositive definite, we see that T[y] is a stationary minimum if # satisfies a Poisson-like equation within the volume V and the departures 6i vanish on the boundary. With y— ® and g—> . pléo, the minimization of the functional yields the “equation of motion” of the electrostatic potential in the presence of a charge density and Dirichlet boundary conditions (® given on S and so 5@ = 0 there). The derivation of the Poisson equation from the variational functional is the formal aspect. Equally important, the stationary nature of the extremum of /[y] permits a practical approach to an approximate solution for (x). We choose a flexible “trial” function (x) = A'V(x, a, B,...) that depends on a normalization constant A and some number of other parameters, «, 6,..., and is constructed to satisfy the given boundary conditions on the surface S. The function V may be a sum of terms with the parameters as coefficients, or a single function of several parameters; it should be chosen with some eye toward the expected form of the solution. (Intuition plays a role here!) Calculation of /[,] gives the func- tion, I(A, a, B, . . .). We now vary the parameters to locate the extremum (actually a minimum) of /(A, a, ,...). With the optimum parameters, the trial solution is the best possible approximation to the true solution with the particular func- tional form chosen. For the Laplace equation, the normalization constant is de- termined by the Dirichlet boundary values of . For the Poisson equation, it is determined by the source strength g(x), as well as the boundary values on S. A different functional is necessary for Neumann boundary conditions. Sup-Sect. 1.12 Variational Approach to the Solution of the Laplace and Poisson Equations 45 pose that the boundary conditions on y are specified by ay/an|s = f(s), where s locates a point on the surface S. The appropriate functional is 1 Ty] = af Vy Vu dex — [ giid?x — bt da (1.67) The same steps as before with — y + dy lead to the first-order difference in functionals, =f wy 3 oy 81 fi Vy — g] dp d3x + $ (2 f(s) ) dy da (1.68) The requirement that 6/ vanish independent of dy implies Vw = —g within Vand a = f(s)onS (1.69) Again the functional is a stationary minimum for satisfying (1.69). Approximate solutions can be found by the use of trial functions that satisfy the Neumann boundary conditions, just as described above for Dirichlet boundary conditions. As a simple application to the Poisson equation, consider the two-dimen- sional problem of a hollow circular cylinder of unit radius centered on the z-axis, with an interior source density g(x) = g(p), azimuthally symmetric and inde- pendent of z. The potential vanishes at p = 1. The “equation of motion” for w (a function of p alone) in polar coordinates is laf a) | 12 ( 2) - g(0) (1.70) For trial functions we consider finite polynomials in powers of (1 — p) and p. A three-parameter function of the first type is W, = a(1 — p) + Bl — p)’ + v(1 = py? (1.71) This choice might seem natural because it automatically builds in the boundary condition at p = 1, but it contains a flaw that makes it a less accurate represen- tation of y than the power series in p. The reason is that, if the source density g is well behaved and finite at the origin, Gauss’s law shows that has a maximum or minimum there with vanishing slope. The requirements at both the origin and p = 1 are met by a three-parameter trial function in powers of p: WV, = ap’ + Bp’ + yp'— (a+ B+ y) (1.72) We expect this trial function in general to be a better approximation to than ‘W, for the same number of variational parameters. [We could, of course, impose the constraint, a; + 28; + 3y; = 0 on (1.71) to get the proper behavior at the origin, but that would reduce the number of parameters from three to two.] The functional integral (1.63) for V, is easily shown to be 2 wy =|ber +S apttays ip In — E 4 ae) 2 . +7 Byt v| — [ew + eB + esy] where e,, = Jo g(p)(p” — 1) pdp.46 Chapter 1 Introduction to Electrostatics—ST The integral for V, has the same form as (1.73), but different coefficients. As described above, we seek an extremum of (1.73) by setting the partial deriv- atives with respect to the parameters a, B, and y equal to zero. The three coupled algebraic linear equations yield the “best” values, a = 225e, — 420e; + 210e, 4: B= —420e, + 2 es; — 420e, (1.74) y = 210e, — 420e; + a es These values can be inserted into (1.73) to give I[Wz]min as a not very illuminating function of the e,,. One would then find that the “kinetic” (first) bracket was equal to half the “potential” (second) bracket and opposite in sign, a character- istic of the extremum. To go further we must specify g(p). The results for the best trial functions ‘V, and W are shown in Fig. 1.9 for the source density, &(p) = —5(1 — p) + 10%p°(1 — p)> (1.75) The choice of source is arbitrary and is chosen to give a potential that is not quite featureless. The “best” parameters for VW are a = 2.915, 8 = —7.031, and y = 3.642. The variational integral has the value, [[V>]min = —1.5817, compared to I [Wlexact = —1.6017. The fractional error is 1.3%. Note that the trial function V, fails rather badly for p < 0.3 because it does 0.7)—— T T T J Charge density SF (arvitray scale) \. op) L L L L L 00 01 02 03 04 O05 06 O07 O08 09 10 p—> Figure 1.9 Comparison of the exact solution #(p) (solid curve) with two variational approximations for the potential, V, (dotted curve) and VW, (dashed curve). The charge density (1.75) is indicated by the dash-dot curve (arbitrary scale). 0.3 L L LSect. 1.13 Relaxation Method for Two-Dimensional Electrostatic Problems 47 not respect the vanishing slope at p = 0. Nonetheless, it gives I[V]min = —1-5136, which is somewhat, but not greatly, worse than V, (5.5% error). The insensitivity of J[] to errors in the trial function illustrates both a strength and a weakness of the variational method. If the principle is used to estimate eigenvalues (related to the value of /[‘¥]), it does well. Used as a method of estimating a solution wy ~ W, it can fail badly, at least in parts of the configuration space. The reader will recognize from (1.70) that a polynomial source density leads to an exact polynomial solution for yf, but the idea here is to illustrate the vari- ational method, not to demonstrate a class of explicit solutions. Further illustra- tion is left to the problems at the end of this and later chapters. 1.13 Relaxation Method for Two-Dimensional Electrostatic Problems The relaxation method is an iterative numerical scheme (sometimes called iter- ative finite difference method) for the solution of the Laplace or Poisson equation in two dimensions. Here we present only its basic ideas and its connection with the variational method. First we consider the Laplace equation with Dirichlet boundary conditions within a two-dimensional region S with a boundary contour C. We imagine the region S spanned by a square lattice with lattice spacing h (and the boundary contour C approximated by a step-like boundary linking lat- tice sites along C). The independent variables are the integers (i, j) specifying the sites; the dependent variables are the trial values of the potential (i, j) at each site. The potential values on the boundary sites are assumed given. To establish the variational nature of the method and to specify the iterative scheme, we imagine the functional integral /[y] over S as a sum over small do- mains of area h’, as shown in Fig. 1.10a. We consider the neighboring trial values of the potential as fixed, while the value at the center of the subarea is a varia- tional quantity to be optimized. The spacing is small enough to permit us to approximate the derivatives in, say, the northeast quarter of the subarea by oy) 1 yy. ow) 1a and similarly for the other three quarters. The functional integral over the north- east quarter is ie r aw\? — (au\? == oe) eee Ine 2 Jo or ax. ay ji (1.76) —Lr——O—O The complete integral over the whole (shaded) subarea is evidently 1 T= 41 — Uy)” + (ho — Ye)? + (ho — Ys)” + (ho — tw] (1.77) Minimizing this integral with respect to Yo gives the optimum value, 1 (o)oprimum = 3 (s+ te + hs + Yow) (1.78)48 Chapter 1 Introduction to Electrostatics—SI Ve waiver onc | vs @ oe Figure 1.10 (a) Enlargement of one of the subareas in the functional integral (shaded). The trial values of the potential at the neighboring sites are labeled Ux, Us, Ue, and Yow, while the value at the center of the subarea is yp. (b) One possible iteration is to replace the trial values at the lattice sites (0) with the average of the values at the surrounding sites (x). The integral is minimized if i is equal to the average of the values at the “cross” points. Now consider the whole functional integral, that is, the sum of the integrals over all the subareas. We guess a set of (i, j) initially and approximate the functional integral [i] by the sum of terms of the form of (1.77). Then we go over the lattice and replace half the values, indicated by the circles in Fig. 1.10b, by the average of the points (crosses) around them. The new set of trial values y(i, j) will evidently minimize /[] more than the original set of values; the new set will be closer to the correct solution. Actually, there is no need to do the averaging for only half the points—that was just a replication for half of the subareas of the process for Fig. 1.10a. There are many improvements that can be made. One significant one con- cerns the type of averaging. We could have taken the average of the values at the corners of the large square in Fig. 1.10a instead of the ‘‘cross” values. Or we could take some linear combination of the two. It can be shown (see Problem 1,22) by Taylor series expansion of any well-behaved function F(x, y) that a particular weighted average, 4 1 a —LhLrmrm—eN (1.79) where the “cross” and “square” averages are (F(x, Ye = : [FQ@cth, y) + F(x, y+h) + F(x—h, y) + F(x, y-h)] (1.80a) (F(x, y)). = ; [FQ@c+h, y+h) + FOet+h, yh) ( 1.80b) + F(x-h, y+h) + F(eh, y—h)] yields Fx y))) = Fle y) + = RVR + z ht VAVF) + O(N) (1.81)Sect. 1.13 Relaxation Method for Two-Dimensional Electrostatic Problems 49 In (1.81) the Laplacians of F are evaluated at (x, y). If F(x, y) is a solution of the Laplace equation, the weighted averaging over the eight adjacent lattice sites in (1.79) gives F at the center with corrections only of order A®, Instead of (1.78), which is the same as (1.80a), a better iteration scheme uses tipew(é, j) = ((ab(i, j))) + O(h®). With either the “cross” or “square” averaging separately, the error is O(h*). The increase in accuracy with (()) is at the expense of twice as much computation for each lattice site, but for the same accuracy, far fewer lattice sites are needed: ((N)) = O((N)**), where ((N)) is the number of sites needed with ((i)) and (N) is the corresponding number with the “cross” or “square” average. Equation (1.81) has an added advantage in application to the Poisson equa- tion, Vs = —g. The terms of order h? and h* can be expressed directly in terms of the specified charge density and the simplest approximation for its Laplacian. It is easy to show that the new value of the trial function at (i, /) is generated by : : reali) = (Ui) +E eid +E (eli Die + OO) (1.82) where (g)- is the “cross” average of g, according to (1.80a). A basic procedure for the iterative numerical solution of the Laplace or Poisson equation in two dimensions with Dirichlet boundary conditions is as follows: 1. A square lattice spacing h is chosen and the lattice sites, including the sites on the boundary, are labeled in some manner [which we denote here as (i, /)]. 2. The values of the potential at the boundary sites are entered in a table of the potential at alll sites. 3. A guess is made for the values, called ®,,4(i, /), at all interior sites. A constant value everywhere is easiest. These are added to the table or array of “start- ing” values. 4. The first iteration cycle begins by systematically going over the lattice sites, one by one, and computing ((P(i, ))) with (1.79) or one of the averages in (1.80). This quantity (or (1.82) for the Poisson equation) is entered as ®,.w(i, j) in a table of “new” values of the potential at each site. Note that the sites next to the boundary benefit from the known boundary values, and so their (()) values are likely initially to be closer to the ultimate values of the potential than those for sites deep in the interior. With each iteration, the accuracy works its way from the boundaries into the interior. 5. Once all interior sites have been processed, the set of P,y4(i, j) is replaced by the set of ®,.y(i, j), and the iteration cycle begins again. 6. Iterations continue until some desired level of accuracy is achieved. For ex- ample, one might continue iterations until the absolute value of the differ- ence of old and new values is less than some preassigned value at every interior site. The scheme just outlined is called Jacobian iteration. It requires two arrays of values of the potential at the lattice sites during each iteration. A better scheme, called Gauss-Seidel iteration, employs a trivial change: one replaces ®ga(i, /) with ®,.(i, /) as soon as the latter is determined. This means that during an iteration one benefits immediately from the improved values. Typically, at any given site, ((®)) is made up half of old values and half of new ones, depending50 Chapter 1 Introduction to Electrostatics—SI on the path over the lattice. There are many other improvements possible— consult Press et al., Numerical Recipes, or some of the references cited at the end of the chapter. The relaxation method is also applicable to magnetic field prob- lems, as described briefly in Section 5.14. References and Suggested Reading On the mathematical side, the subject of delta functions is treated simply but rigor- ously by Lighthill Dennery and Kryzwicki For a discussion of different types of partial differential equations and the appropriate boundary conditions for each type, see Morse and Feshbach, Chapter 6 Sommerfeld, Partial Differential Equations in Physics, Chapter II Courant and Hilbert, Vol. II, Chapters I-VI The general theory of Green functions is treated in detail by Friedman, Chapter 3 Morse and Feshbach, Chapter 7 The general theory of electrostatics is discussed extensively in many of the older books. Notable, in spite of some old-fashioned notation, are Maxwell, Vol. 1, Chapters II and IV. Jeans, Chapters II, VI, VII Kellogg Of more recent books, mention may be made of the treatment of the general theory by Stratton, Chapter III, and parts of Chapter IT. Readers interested in variational methods applied to electromagnetic problems can consult Cairo and Kahan Collin, Chapter 4 Sadiku, Chapter 4 and Polya and Szego for elegant and powerful mathematical techniques. The classic references to relaxation methods are the two books by R. V. Southwell: Relaxation Methods in Engineering Science, Oxford University Press, Oxford (1940). Relaxation Methods in Theoretical Physics, Oxford University Press, Oxford (1946). Physicists will be more comfortable with the second volume, but much basic material is in the first. More modern references on relaxation and other numerical methods are Sadiku Zhou Problems 1.1 Use Gauss’s theorem [and (1.21) if necessary] to prove the following: (a) Any excess charge placed on a conductor must lie entirely on its surface. (A conductor by definition contains charges capable of moving freely under the action of applied electric fields.)13 14 eS 1.6 Ch.1 Problems 51 (b) A closed, hollow conductor shields its interior from fields due to charges out- side, but does not shield its exterior from the fields due to charges placed inside it. (©) The electric field at the surface of a conductor is normal to the surface and has a magnitude of, where o is the charge density per unit area on the surface. The Dirac delta function in three dimensions can be taken as the improper limit as @ > 0 of the Gaussian function D(a; x, y, 2) = (2a)? a? ev| 35 oe ye | Consider a general orthogonal coordinate system specified by the surfaces u = constant, v = constant, w = constant, with length elements du/U, du/V, dw/W in the three perpendicular directions. Show that &(x — x’) = 5(u — w’) 6 — v') 6(w — w’)- UVW by considering the limit of the Gaussian above. Note that as a > 0 only the infin- itesimal length element need be used for the distance between the points in the exponent. Using Dirac delta functions in the appropriate coordinates, express the following charge distributions as three-dimensional charge densities p(x). (a) In spherical coordinates, a charge Q uniformly distributed over a spherical shell of radius R. (b) In cylindrical coordinates, a charge A per unit length uniformly distributed over a cylindrical surface of radius b. (©) In cylindrical coordinates, a charge Q spread uniformly over a flat circular disc of negligible thickness and radius R. (a) The same as part (c), but using spherical coordinates. Each of three charged spheres of radius a, one conducting, one having a uniform charge density within its volume, and one having a spherically symmetric charge density that varies radially as r" (n > —3), has a total charge Q. Use Gauss’s theorem to obtain the electric fields both inside and outside each sphere. Sketch the behavior of the fields as a function of radius for the first two spheres, and for the third with n= —2, +2. The time-averaged potential of a neutral hydrogen atom is given by where q is the magnitude of the electronic charge, and a! = ag/2, do being the Bohr radius. Find the distribution of charge (both continuous and discrete) that will give this potential and interpret your result physically. A simple capacitor is a device formed by two insulated conductors adjacent to each other. If equal and opposite charges are placed on the conductors, there will be a certain difference of potential between them. The ratio of the magnitude of the charge on one conductor to the magnitude of the potential difference is called the capacitance (in SI units it is measured in farads). Using Gauss’s law, calculate the capacitance of (a) two large, flat, conducting sheets of area A, separated by a small distance d; (b) two concentric conducting spheres with radii a, b (b > a); (©) two concentric conducting cylinders of length L, large compared to their radii a,b (b >a).52 Chapter 1 1.7 19 1.10 111 1.12 1.13 1.14 Introduction to Electrostatics—SI (a) What is the inner diameter of the outer conductor in an air-filled coaxial cable whose center conductor is a cylindrical wire of diameter 1 mm and whose capacitance is 3 x 10" Fim? 3 x 10-!? Fim? Two long, cylindrical conductors of radii a, and a are parallel and separated by a distance d, which is large compared with either radius. Show that the capacitance per unit length is given approximately by a a C= ram ‘) where a is the geometrical mean of the two radii. Approximately what gauge wire (state diameter in millimeters) would be nec- essary to make a two-wire transmission line with a capacitance of 1.2 x 107"! Fim if the separation of the wires was 0.5 cm? 1.5 cm? 5.0 cm? (a) For the three capacitor geometries in Problem 1.6 calculate the total electro- static energy and express it alternatively in terms of the equal and opposite charges Q and —@Q placed on the conductors and the potential difference between them. (b) Sketch the energy density of the electrostatic field in each case as a function of the appropriate linear coordinate. Calculate the attractive force between conductors in the parallel plate capacitor (Problem 1.6a) and the parallel cylinder capacitor (Problem 1.7) for (a) fixed charges on each conductor; (b) fixed potential difference between conductors. Prove the mean value theorem: For charge-free space the value of the electrostatic potential at any point is equal to the average of the potential over the surface of any sphere centered on that point. Use Gauss’s theorem to prove that at the surface of a curved charged conductor, the normal derivative of the electric field is given by eee Ean R,” Re where R, and R; are the principal radii of curvature of the surface. Prove Green’s reciprocation theorem: If ® is the potential due to a volume-charge density p within a volume V and a surface-charge density « on the conducting surface S bounding the volume V, while ©” is the potential due to another charge distribution p' and o’, then [ow s+ f oo a=f poax+ f oo da Vv s Vv s Two infinite grounded parallel conducting planes are separated by a distance d. A point charge q is placed between the planes. Use the reciprocation theorem of Green to prove that the total induced charge on one of the planes is equal to (4) times the fractional perpendicular distance of the point charge from the other plane (Hint: As your comparison electrostatic problem with the same surfaces choose one whose charge densities and potential are known and simple.) Consider the electrostatic Green functions of Section 1.10 for Dirichlet and Neumann boundary conditions on the surface $ bounding the volume V. Apply Green’s theorem (1.35) with integration variable y and = G(x, y), ¥ = G(x', y), with V°, G(z, y) = —478(y — z). Find an expression for the difference [G(x, x’) — G(x’, x)] in terms of an integral over the boundary surface S. (a) For Dirichlet boundary conditions on the potential and the associated bound- ary condition on the Green function, show that Gp(x, x’ must be symmetric in x and x'.1.15 1.16 1.17 1.18 Ch.1 Problems 53 (b) For Neumann boundary conditions, use the boundary condition (1.45) for G(x, x’) to show that Gy(x, x) is not symmetric in general, but that Ga(x, x’) — F(x) is symmetric in x and x’, where Fs) = £f, Gabe.) da, (c) Show that the addition of F(x) to the Green function does not affect the po- tential (x), See problem 3.26 for an example of the Neumann Green function. Prove Thomson's theorem: If a number of surfaces are fixed in position and a given total charge is placed on each surface, then the electrostatic energy in the region bounded by the surfaces is an absolute minimum when the charges are placed so that every surface is an equipotential, as happens when they are conductors. Prove the following theorem: If a number of conducting surfaces are fixed in po- sition with a given total charge on each, the introduction of an uncharged, insulated conductor into the region bounded by the surfaces lowers the electrostatic energy. A volume V in vacuum is bounded by a surface S consisting of several separate conducting surfaces S,. One conductor is held at unit potential and all the other conductors at zero potential. (a) Show that the capacitance of the one conductor is given by Caaf word's 7 where ®(x) is the solution for the potential. (b) Show that the true capacitance C is always less than or equal to the quantity CY] = 6 [ [VHP dx where is any trial function satisfying the boundary conditions on the con- ductors. This is a variational principle for the capacitance that yields an upper bound. Consider the configuration of conductors of Problem 1.17, with all conductors ex- cept S; held at zero potential. (a) Show that the potential «b(x) anywhere in the volume V and on any of the surfaces S; can be written (x) = = é. o4(x')G(x, x’) da” where o;(x’) is the surface charge density on S, and G(x, x’) is the Green function potential for a point charge in the presence of all the surfaces that are held at zero potential (but with S, absent). Show also that the electrostatic energy is = Ww 87rE f da f da! 04(x)G(x, x’)o;(x') where the integrals are only over the surface S;. (b) Show that the variational expression f, da £ da'o(x)G(x, x’)o(x’) z 4relf o(x) aa] with an arbitrary integrable function o(x) defined on Sj, is stationary for small variations of « away from ;. Use Thomson’s theorem to prove that the54° Chapter 1 119 1.20 122 Introduction to Electrostatics—SI reciprocal of C~'[a] gives a lower bound to the true capacitance of the con ductor S,. For the cylindrical capacitor of Problem 1.6c, evaluate the variational upper bound of Problem 1.17b with the naive trial function, ¥;(p) = (b — p)(b — a). Compare the variational result with the exact result for b/a = 1.5, 2, 3. Explain the trend of your results in terms of the functional form of ‘V,. An improved trial function is treated by Collin (pp. 275-277). In estimating the capacitance of a given configuration of conductors, comparison with known capacitances is often helpful. Consider two configurations of n conduc- tors in which the (n — 1) conductors held at zero potential are the same, but the one conductor whose capacitance we wish to know is different. In particular, let the conductor in one configuration have a closed surface S, and in the other configu- ration have surface Sj, with S; totally inside S,. (a) Use the extremum principle of Section 1.12 and the variational principle of Problem 1.17 to prove that the capacitance C’ of the conductor with surface Sj is less than or equal to the capacitance C of the conductor with surface S, that encloses S/. (b) Set upper and lower limits for the capacitance of a conducting cube of side a. Compare your limits and also their average with the numerical value, C = 0,655(477€9a). (©) By how much do you estimate the capacitance per unit length of the two-wire system of Problem 1.7 will change (larger? smaller?) if one of the wires is replaced by a wire of square cross section whose side is equal to its diameter? dimensional potential problem consists of a unit square area (0 = x <1, = 1) bounded by “surfaces” held at zero potential. Over the entire square there is a uniform charge density of unit strength (per unit length in z). (a) Apply the variational principle (1.63) for the Poisson equation with the “vari- ational” trial function W(x, y) = A-x(1 — x) -y(1 — y) to determine the best value of the constant A. [I use quotation marks around variational because there are no parameters to vary except the overall scale.] (b) The exact (albeit series) solution for this problem is [see Problems 2.15 and 2.16] S sinl@m + Dax) {i _ cont ee ah 16 P(x, y) = AreePlx, ) = 3 cosh[(2m + 1)7/2] Ho (m+ ip For y = 0.25 and y = 0.5, plot and compare the simple variational solution of part a with the exact solution as functions of x. Two-dimensional relaxation calculations commonly use sites on a square lattice with spacing Ax = Ay = h, and label the sites by (i, j), where i, j are integers and x, = ih + Xo, y; = jh + yo. The value of the potential at (i, j) can be approximated by the average of the values at neighboring sites. [Recall the relevant theorem about harmonic functions.) But what average? (a) If F(x, y) is a well-behaved function in the neighborhood of the origin, but not necessarily harmonic, by explicit Taylor series expansions, show that the “cross” sum S. = F(h, 0) + FO, h) + F(-h, 0) + FOO, -A) can be expressed as Sc = 4F(0, 0) + W2N2F + = (Face + Fyyyy) + O(F°)(b) Ch.1 Problems 55 Similarly, show that the “square” sum, Ss = Fh, h) + F(-h, h) + F(-h, h) + F(h, ~h) can be expressed as hs he Ss = 4F(O, 0) + PVF — 5 (Fesee + Byys) + > VWF) + OCH") Here Fee is the fourth partial derivative of F with respect to x, evaluated at x=0,y = 0, etc. If VF = 0, the averages S./4 and S,/4 each give the value of F(0, 0), correct to order A? inclusive. Note that an improvement can be ob- tained by forming the “improved” average, ((F(O, 0))) = F(O, 0) + = WPF + z VWF) + O(h°) «ROO, 09) = 4 [s. +4 where If VF = 0, then § gives F(0, 0), correct to order h* inclusive. For Poisson's equation, the charge density and its lowest order Laplacian can be inserted for the same accuracy. A transmission line consists of a long straight conductor with a hollow square region in its interior, with a square conductor of one-quarter the area of the hollow region centered in the empty space, with edges parallel to the inner sides of outer con- ductor. If the conductors are raised to different potentials, the potential and electric field in the space between them exhibit an eightfold symmetry; the basic unit is sketched in the accompanying figure. The efficacy of the relaxation method in de- termining the properties of the transmission line can be illustrated by a simple calculation. @ (b) © Using only the four interior points indicated in the figure, write down the relaxation equation for each point for the “cross” and the “improved” aver- aging schemes (defined in Problem 1.22) if the inner conductor has = V and the outer has © = 0. By performing either the relaxation iterati process or solving the set of algebraic equations for each scheme, find esti- mates for the potential at each of the four points for the two schemes. From the results of part a make the best estimate (or estimates) you can for the capacitance per unit length of the transmission line. (Optional) Using your favorite computational tools, repeat the relaxation cal- culation with half the lattice spacing (21 interior points) and compare. Answer: , = 48.87 V, ©, = 47.18 V, ©; = 38.34 V, @, = 19.81 V and C = 10.23 € F/m [from an accurate numerical calculation]. Problem 1.2356 Chapter 1 Introduction to Electrostatics—SI 1.24 Consider solution of the two-dimensional Poisson equation problem of Problem 1.21, a unit square with zero potential on the boundary and a constant unit charge density in the interior, by the technique of relaxation. Choose h = 0.25 so that there are nine interior sites. Use symmetry to reduce the number of needed sites to three, at (0.25, 0.25), (0.5, 0.25), and (0.5, 0.5). With so few sites, it is easy to do the iterations with a block of paper and a pocket calculator, but suit yourself. (a) Use the “improved grid” averaging of Problem 1.22 and the simple (Jacobian) iteration scheme, starting with 4zre = 1.0 at all three interior sites. Do at least six iterations, preferably eight or ten. (b) Repeat the iteration procedure with the same starting values, but using Gauss— Seidel iteration. (c) Graph the two sets of results of each iteration versus iteration number and compare with the exact values, 47€,(0.25, 0.25) = 0.5691, 477€(0.5, 0.25) = 0.7205, 477€)P(0.5, 0.5) = 0.9258. Comment on rate of convergence and final accuracy.CHAPTER 2 Boundary-Value Problems in Electrostatics: I Many problems in electrostatics involve boundary surfaces on which either the potential or the surface-charge density is specified. The formal solution of such problems was presented in Section 1.10, using the method of Green functions. In practical situations (or even rather idealized approximations to practical sit- uations) the discovery of the correct Green function is sometimes easy and some- times not. Consequently a number of approaches to electrostatic boundary-value problems have been developed, some of which are only remotely connected to the Green function method. In this chapter we will examine three of these special techniques: (1) the method of images, which is closely related to the use of Green functions; (2) expansion in orthogonal functions, an approach directly through the differential equation and rather remote from the direct construction of a Green function; (3) an introduction to finite element analysis (FEA), a broad class of numerical methods. A major omission is the use of complex-variable techniques, including conformal mapping, for the treatment of two-dimensional problems. The topic is important, but lack of space and the existence of self- contained discussions elsewhere accounts for its absence. The interested reader may consult the references cited at the end of the chapter. 2.1 Method of Images The method of images concerns itself with the problem of one or more point charges in the presence of boundary surfaces, for example, conductors either grounded or held at fixed potentials. Under favorable conditions it is possible to infer from the geometry of the situation that a small number of suitably placed charges of appropriate magnitudes, external to the region of interest, can simu- late the required boundary conditions. These charges are called image charges, and the replacement of the actual problem with boundaries by an enlarged region with image charges but not boundaries is called the method of images. The image charges must be external to the volume of interest, since their potentials must be solutions of the Laplace equation inside the volume; the “particular integral” (i.e., solution of the Poisson equation) is provided by the sum of the potentials of the charges inside the volume. A simple example is a point charge located in front of an infinite plane con- ductor at zero potential, as shown in Fig. 2.1. It is clear that this is equivalent to the problem of the original charge and an equal and opposite charge located at the mirror-image point behind the plane defined by the position of the conductor. 5758 Chapter 2 Boundary-Value Problems in Electrostatics: I—ST Figure 2.1 Solution by method of images. The original potential problem is on the left, the equivalent-image problem on the right. 2.2. Point Charge in the Presence of a Grounded Conducting Sphere As an illustration of the method of images we consider the problem illustrated in Fig. 2.2 of a point charge q located at y relative to the origin, around which is centered a grounded conducting sphere of radius a. We seek the potential (x) such that (|x| = a) = 0. By symmetry it is evident that the image charge q’ (assuming that only one image is needed) will lie on the ray from the origin to the charge q. If we consider the charge g outside the sphere, the image position y’ will lie inside the sphere. The potential due to the charges g and q’ is: (2.1) We now must try to choose q’ and |y’| such that this potential vanishes at |x| = a. If nis a unit vector in the direction x, and n’ a unit vector in the direction y, then q4mey , _q'lAney 9) Tem = yal] * ow = 907] (2.2) If x is factored out of the first term and y' out of the second, the potential at x = a becomes: gle, : q' lame, y aln—= a (x =a) = (23) Figure 2.2 Conducting sphere of radius a, with charge q and image charge q’.Sect. 2.2 Point Charge in the Presence of a Grounded Conducting Sphere 59 From the form of (2.3) it will be seen that the choices: Gq. Ye 7 a a make (x = a) = 0, for all possible values of n+ n'. Hence the magnitude and position of the image charge are raf oe 24) q y qd De y 7 We note that, as the charge q is brought closer to the sphere, the image charge grows in magnitude and moves out from the center of the sphere. When q is just outside the surface of the sphere, the image charge is equal and opposite in magnitude and lies just beneath the surface. Now that the image charge has been found, we can return to the original problem of a charge q outside a grounded conducting sphere and consider various effects. The actual charge density induced on the surface of the sphere can be calculated from the normal derivative of ® at the surface: 2 yy 7 m3 @ a yoy where y is the angle between x and y. This charge density in units of —q/4ma? is shown plotted in Fig. 2.3 as a function of y for two values of y/a. The concentra- (2.5) z 2 co Figure 2.3 Surface-charge density o induced on the grounded sphere of radius a as a result of the presence of a point charge q located a distance y away from the center of the sphere. a is plotted in units of —q/4za* as a function of the angular position -y away from the radius to the charge for y = 2a, 4a. Inset shows lines of force for y = 2a.60 Chapter 2 Boundary-Value Problems in Electrostatics: I—SI dF = (07 /2eq)da Figure 2.4 tion of charge in the direction of the point charge q is evident, especially for yla = 2. It is easy to show by direct integration that the total induced charge on the sphere is equal to the magnitude of the image charge, as it must be, according to Gauss’s law. The force acting on the charge q can be calculated in different ways. One (the easiest) way is to write down immediately the force between the charge q and the image charge q'. The distance between them is y — y’ = y(1 — a’/y’). Hence the attractive force, according to Coulomb’s law, is: 3 / eee ane IF = are a (}) ( y oS For large separations the force is an inverse cube law, but close to the sphere it is proportional to the inverse square of the distance away from the surface of the sphere. The alternative method for obtaining the force is to calculate the total force acting on the surface of the sphere. The force on each element of area da is (o7/2¢,) da, where ais given by (2.5), as indicated in Fig, 2.4. But from symmetry it is clear that only the component parallel to the radius vector from the center of the sphere to q contributes to the total force. Hence the total force acting on the sphere (equal and opposite to the force acting on q) is given by the integral: 2 2 g a e fi cos y Fj =~“ (4) (1-5) | ——“%* Fl aed (') (: #) ( @ 2a ) a 14+ 5 -* cosy Integration immediately yields (2.6). The whole discussion has been based on the understanding that the point charge q is outside the sphere. Actually, the results apply equally for the charge q inside the sphere. The only change necessary is in the surface-charge density (2.5), where the normal derivative out of the conductor is now radially inward, implying a change in sign. The reader may transcribe all the formulas, remem- bering that now y = a. The angular distributions of surface charge are similar to those of Fig. 2.3, but the total induced surface charge is evidently equal to —q, independent of y. 2.3 Point Charge in the Presence of a Charged, Insulated, Conducting Sphere In the preceding section we considered the problem of a point charge q near a grounded sphere and saw that a surface-charge density was induced on theSect. 2.4 Point Charge Near a Conducting Sphere at Fixed Potential 61 sphere. This charge was of total amount g' = —ag/y, and was distributed over the surface in such a way as to be in equilibrium under all forces acting. If we wish to consider the problem of an insulated conducting sphere with total charge Q in the presence of a point charge q, we can build up the solution for the potential by linear superposition. In an operational sense, we can imagine that we start with the grounded conducting sphere (with its charge q' distributed over its surface). We then disconnect the ground wire and add to the sphere an amount of charge (Q — q'). This brings the total charge on the sphere up to Q. To find the potential we merely note that the added charge (Q — q’) will dis- tribute itself uniformly over the surface, since the electrostatic forces due to the point charge q are already balanced by the charge q'. Hence the potential due to the added charge (Q — q') will be the same as if a point charge of that mag- nitude were at the origin, at least for points outside the sphere. The potential is the superposition of (2.1) and the potential of a point charge (Q — q') at the origin: a Qt+-q 1 aq y oo ne ; ®@) = Fe z Bl (2.8) WX By y The force acting on the charge q can be written down directly from Coulomb’s law. It is directed along the radius vector to q and has the magnitude: 4 [ - «or 2 . (2.9) © Ame) y? yr ay fy In the limit of y >> a, the force reduces to the usual Coulomb’s law for two small charged bodies. But close to the sphere the force is modified because of the induced charge distribution on the surface of the sphere. Figure 2.5 shows the force as a function of distance for various ratios of Q/q. The force is expressed in units of q?/47re,y’; positive (negative) values correspond to a repulsion (at- traction). If the sphere is charged oppositely to q, or is uncharged, the force is attractive at all distances. Even if the charge Q is the same sign as q, however, the force becomes attractive at very close distances. In the limit of Q >> q, the point of zero force (unstable equilibrium point) is very close to the sphere, namely, at y ~ a(1 + }Vq/Q). Note that the asymptotic value of the force is attained as soon as the charge q is more than a few radii away from the sphere. This example exhibits a general property that explains why an excess of charge on the surface does not immediately leave the surface because of mutual repulsion of the individual charges. As soon as an element of charge is removed from the surface, the image force tends to attract it back. If sufficient work is done, of course, charge can be removed from the surface to infinity. The work function of a metal is in large part just the work done against the attractive image force to remove an electron from the surface. 2.4 Point Charge Near a Conducting Sphere at Fixed Potential Another problem that can be discussed easily is that of a point charge near a conducting sphere held at a fixed potential V. The potential is the same as for62 Chapter 2 Boundary-Value Problems in Electrostatics: I—SI Figure 2.5 The force on a point charge q due to an insulated, conducting sphere of radius a carrying a total charge Q. Positive values mean a repulsion, negative an attraction. The asymptotic dependence of the force has been divided out. 4r¢,Fy/q? is plotted versus y/a for Q/q = —1, 0, 1, 3. Regardless of the value of Q, the force is always attractive at close distances because of the induced surface charge. the charged sphere, except that the charge (Q — q') at the center is replaced by a charge (Va). This can be seen from (2.8), since at |x| = a the first two terms cancel and the last term will be equal to V as required. Thus the potential is i q aq Va ox) = -_-%@_| , “4 2.10 ©) = Tres Ix— y| | ¢ [x] ou K-35 The force on the charge q due to the sphere at fixed potential is q Gaya Yi F= 2] va -——” _|% 211 y [v Ame (y? — 7 y eS For corresponding values of 47r¢)Va/q and Q/q this force is very similar to that of the charged sphere, shown in Fig. 2.5, although the approach to the asymptotic value (Vaq/y”) is more gradual. For Va >> g, the unstable equilibrium point has the equivalent location y ~ a(1 + 4V/q/4zre)Va). 2.5 Conducting Sphere in a Uniform Electric Field by Method of Images Asa final example of the method of images we consider a conducting sphere of radius a in a uniform electric field Ey. A uniform field can be thought of as beingSect. 2.5 Conducting Sphere in a Uniform Electric Field by Method of Images 63 produced by appropriate positive and negative charges at infinity. For example, if there are two charges +Q, located at positions z = +R, as shown in Fig. 2.6a, then in a region near the origin whose dimensions are very small compared to R there is an approximately constant electric field Ey ~ 20/47€)R? parallel to the zaxis. In the limit as R, Q > ~, with Q/R? constant, this approximation becomes exact. If now a conducting sphere of radius a is placed at the origin, the potential will be that due to the charges +Q at +R and their images +Qa/R at z = Fa? /R: Ae OlArre, _ OlAre, (P + R? + 2rR cos)? (r? + R? — 2rR cos 6)!” (2.12) : aQl4rey + aQi4rre, og + — i og R(r+ E+ eco) a(t + S20 cosa where ® has been expressed in terms of the spherical coordinates of the obser- vation point. In the first two terms R is much larger than r by assumption. Hence we can expand the radicals after factoring out R?. Similarly, in the third and fourth terms, we can factor out r? and then expand. The result is: 20 20 a 1 = 22 5 cos + 22% cosa) +--+ 2.13 a R RP ey where the omitted terms vanish in the limit R > ©. In that limit 20/47e)R? becomes the applied uniform field, so that the potential is 3 &= -(r . ) cos 6 (2.14) The first term (—E,z) is, of course, just the potential of a uniform field Ey which could have been written down directly instead of the first two terms in (2.12). ) Figure 2.6 Conducting sphere in a uniform electric field by the method of images.64 Chapter 2 Boundary-Value Problems in Electrostatics: I—SI The second is the potential due to the induced surface-charge density or, equiv- alently, the image charges. Note that the image charges form a dipole of strength D = QalR X 2a@?/R = 47€) Eoa®. The induced surface-charge density is a a= ~e | = Zea cos (2.15) We note that the surface integral of this charge density vanishes, so that there is no difference between a grounded and an insulated sphere. 2.6 Green Function for the Sphere; General Solution for the Potential In preceding sections the problem of a conducting sphere in the presence of a point charge was discussed by the method of images. As mentioned in Section 1.10, the potential due to a unit source and its image (or images), chosen to satisfy homogeneous boundary conditions, is just the Green function (1.43 or 1.45) ap- propriate for Dirichlet or Neumann boundary conditions. In G(x, x‘) the variable x’ refers to the location P’ of the unit source, while the variable x is the point P at which the potential is being evaluated. These coordinates and the sphere are shown in Fig. 2.7. For Dirichlet boundary conditions on the sphere of radius a the Green function defined via (1.39) for a unit source and its image is given by (2.1) with q > 479 and relations (2.4). Transforming variables appropriately, we obtain the Green function: G(x, x!) = (2.16) Ix— x] Figure 2.7Sect. 2.7 Conducting Sphere with Hemispheres at Different Potentials 65 In terms of spherical coordinates this can be written: 1 1 (Qe? + x? — 2x2" cosy)? (fe G(x, x') = 1 gt a — 2xx' cos 7) (2.17) where y is the angle between x and x’. The symmetry in the variables x and x’ is obvious in the form (2.17), as is the condition that G = 0 if either x or x’ is on the surface of the sphere. For solution (1.44) of the Poisson equation we need not only G, but also aG/an'. Remembering that n’ is the unit normal outward from the volume of interest (i.e., inward along x' toward the origin), we have aG _ (a Gs) an a(x? + a — 2ax cos y)>” (2.18) [Note that this is essentially the induced surface-charge density (2.5).] Hence the solution of the Laplace equation outside a sphere with the potential specified on its surface is, according to (1.44), a(x? — a’) G? Fe = 2ax cos? (2.19) 1 (x) = LE] oe, 0’, 6’) where dQ is the element of solid angle at the point (a, 6’, ') and cos y = cos @ cos 6" + sin @ sin 6’ cos(é — #'). For the interior problem, the normal derivative is radially outward, so that the sign of aG/an’ is opposite to (2.18). This is equivalent to replacing the factor (x? — a?) by (a? — x?) in (2.19). For a problem with a charge distribution, we must add to (2.19) the appropriate integral in (1.44), with the Green function (2.17). 2.7 Conducting Sphere with Hemispheres at Different Potentials As an example of the solution (2.19) for the potential outside a sphere with prescribed values of potential on its surface, we consider the conducting sphere of radius a made up of two hemispherical shells separated by a small insulating ring. The hemispheres are kept at different potentials. It will suffice to consider the potentials as +V, since arbitrary potentials can be handled by superposition of the solution for a sphere at fixed potential over its whole surface. The insu- lating ring lies in the z = 0 plane, as shown in Fig. 2.8, with the upper (lower) hemisphere at potential +V (—V). Figure 2.866 Chapter 2 Boundary-Value Problems in Electrostatics: ISI From (2.19) the solution for ®(x, 6, 6) is given by the integral: vip 1 (x, 6, 6) = ai ao{f (cos 6’) / a(x? — a’) [. Ce i} (@ + x? — 2ax cos y)*? By a suitable change of variables in the second integral (0 > 7 — 0’, 6’ > ¢' + 7), this can be cast in the form: (2.20) Va(x? — a?) 4 ~(@ + x2 + 2ax cos y)*?] Qn 1 P(x, 0, 6) = I do! I d(cos 6')[(@ + x? — 2ax cos y) *? (2.21) Because of the complicated dependence of cos yon the angles (6’, ') and (0, 4), equation (2.21) cannot in general be integrated in closed form. As a special case we consider the potential on the positive z axis. Then cos y = cos 6’, since @ = 0. The integration is elementary, and the potential can be shown to be 2 @ ®(z) = yp. . S44 (2.22) zwVz+a At z =a, this reduces to @ = V as required, while at large distances it goes asymptotically as @ = 3Va?/2z*. In the absence of a closed expression for the integrals in (2.21), we can ex- pand the denominator in power series and integrate term by term. Factoring out (@ + x?) from each denominator, we obtain Vale? -@) P7 eal, de’ [, d(cos 6’)[(1 — 2ecos y) 3? — (1 + 2acos y)3?] (2.23) where a = ax/(a? + x2). We observe that in the expansion of the radicals only odd powers of acos y will appear: [( — 2acos y)~3? — (1 + 2acos y)*?] = 6acos y + 350° cos? y+-+- (2.24) It is now necessary to integrate odd powers of cos y over d¢' d(cos 6’): Qn 1 f dg’ [ d(cos 6") cos y = cos 8 on 1 7 (2.25) I dg’ I (cos 8") cosy = 7 cos (3 ~ cos?d) If (2.24) and (2.25) are inserted into (2.23), the potential becomes 3Va? (x3? — & 35 ax? B(x, 6, 6) = oe (9) cos rE + ae (3 — cos’6) + | (2.26)Sect. 2.8 Orthogonal Functions and Expansions 67 We note that only odd powers of cos @ appear, as required by the symmetry of the problem. If the expansion parameter is (a’/x’), rather than a”, the series takes on the form: 2 2 B(x, 6, 6) = x [oso 7 2 G cos'@ — 5 cos ) + | (2.27) For large values of x/a this expansion converges rapidly and so is a useful rep- resentation for the potential. Even for x/a = 5, the second term in the series is only of the order of 2%. It is easily verified that, for cos @ = 1, expression (2.27) agrees with the expansion of (2.22) for the potential on the axis. [The particular choice of angular factors in (2.27) is dictated by the definitions of the Legendre polynomials. The two factors are, in fact, P\(cos #) and P3(cos 6), and the expan- sion of the potential is one in Legendre polynomials of odd order. We establish this in a systematic fashion in Section 3.3.] Further consideration of both the exterior and interior problem of the two hemispheres is found in Problem 2.22. 2.8 Orthogonal Functions and Expansions The representation of solutions of potential problems (or any mathematical phys- ics problem) by expansions in orthogonal functions forms a powerful technique that can be used in a large class of problems. The particular orthogonal set chosen depends on the symmetries or near symmetries involved. To recall the general properties of orthogonal functions and expansions in terms of them, we consider an interval (a, b) in a variable £ with a set of real or complex functions U,,(é), n=1,2,..., square integrable and orthogonal on the interval (a, b). The ortho- gonality condition on the functions U,,(€) is expressed by » ], Un(EUn(é) dé=0, men (2.28) If n = m, the integral is nonzero. We assume that the functions are normalized so that the integral is unity. Then the functions are said to be orthonormal, and they satisfy » [ Un(EU nl) dE = Sim (2.29) An arbitrary function f(£), square integrable on the interval (a, b), can be expanded in a series of the orthonormal functions U,,(&). If the number of terms in the series is finite (say N), N FE) > a,U,(E) (2.30) ms then we can ask for the “best” choice of coefficients a, so that we get the “best” representation of the function f(é). If “best” is defined as minimizing the mean square error My: -b N 2 My = [ f@) — Y aU,(} dé (2.31) nt68 Chapter 2 Boundary-Value Problems in Electrostatics: ISI it is easy to show that the coefficients are given by » 4a, = i UN(E)F(E) dé (2.32) where the orthonormality condition (2.29) has been used. This is the standard result for the coefficients in an orthonormal function expansion. If the number of terms N in series (2.30) is taken larger and larger, we in- tuitively expect that our series representation of f(é) is “better” and “better.” Our intuition will be correct provided the set of orthonormal functions is com- plete, completeness being defined by the requirement that there exist a finite number No such that for N > No the mean square error My can be made smaller than any arbitrarily small positive quantity. Then the series representation = a,U,(€) = Fé) (2.33) with a, given by (2.32) is said to converge in the mean to f(é). Physicists generally leave the difficult job of proving completeness of a given set of functions to the mathematicians. All orthonormal sets of functions normally occurring in math- ematical physics have been proven to be complete. Series (2.33) can be rewritten with the explicit form (2.32) for the coef- ficients a,,: bf FE) = [ {5 venue} F(E') dé (2.34) Since this represents any function f(€) on the interval (a, b), it is clear that the sum of bilinear terms U%(é')U,(é) must exist only in the neighborhood of é' = & In fact, it must be true that 2 Un(E)U,AE) = 8(E" — €) (2.35) This is the so-called completeness or closure relation. It is analogous to the or- thonormality condition (2.29), except that the roles of the continuous variable and the discrete index n have been interchanged. The most famous orthogonal functions are the sines and cosines, an expan- sion in terms of them being a Fourier series. If the interval in x is (—a/2, a/2), the orthonormal functions are Ap : (=) E ez) = sin : = cos| —— a a a a where m is a non-negative integer and for m = 0 the cosine function is 1/V/a. The series equivalent to (2.33) is customarily written in the form: f(x) = 4A + > [40 eos( 2) a Be sin( 2) | (2.36) where a “a By =2 mic) sin( 22) dx net 2 (ee 2amx aie i fe) eo ) . (237)Sect. 2.8 Orthogonal Functions and Expansions 69 Tf the interval spanned by the orthonormal set has more than one dimension, formulas (2.28)—(2.33) have obvious generalizations. Suppose that the space is two-dimensional, and the variable £ ranges over the interval (a, b) while the variable 7 has the interval (c, d). The orthonormal functions in each dimension are U,(€) and V,,(7). Then the expansion of an arbitrary function f(, 7) is £E 0 = = DX Gam U EV nm) (2.38) where im = [a [ef anun(E)Vn(MF(E (2:39) If the interval (a, b) becomes infinite, the set of orthogonal functions U,,(€) may become a continuum of functions, rather than a denumerable set. Then the Kronecker delta symbol in (2.29) becomes a Dirac delta function. An important example is the Fourier integral. Start with the orthonormal set of complex exponentials, U(x) = i(Qmmxia) 2.40 (x) vac (2.40) m = 0, £1, +2,..., on the interval (—a/2, a/2), with the expansion: 1 ¢ = A,geleameia) 2.41 $e) = 574 3) Ant em) where al _ etna Fy") dx! (2.42) ~al2 Do] _am= = ak (2.43) Am > ie A(k) The resulting expansion, equivalent to (2.41), is the Fourier integral, fx) = 1 E A(ke** dk (2.44) where AW) = Fe J} e F(x) dx (2.45) The orthogonality condition is 7 ef dy = 8(k — k’) (2.46) On70 Chapter 2. Boundary-Value Problems in Electrostatics: I—SI while the completeness relation is Lf? gece’ 7 HeO-*) dk = B(x — x! 247, lene (x — x') (2.47) These last integrals serve as convenient representations of a delta function. We note in (2.44)-(2.47) the complete equivalence of the two continuous variables x and k. 2.9 Separation of Variables; Laplace Equation in Rectangular Coordinates The partial differential equations of mathematical physics are often solved con- veniently by a method called separation of variables. In the process, one often generates orthogonal sets of functions that are useful in their own right. Equa- tions involving the three-dimensional Laplacian operator are known to be sep- arable in eleven different coordinate systems (see Morse and Feshbach, pp. 509, 655). We discuss only three of these in any detail—rectangular, spherical, and cylindrical—beginning with the simplest, rectangular coordinates. The Laplace equation in rectangular coordinates is PO Pb Fo _ ax? ay? az? A solution of this partial differential equation can be found in terms of three ordinary differential equations, all of the same form, by the assumption that the potential can be represented by a product of three functions, one for each coordinate: (2.48) (x, y, z) = X@)Y(y)Z(z) (2.49) Substitution into (2.48) and division of the result by (2.49) yields 1 @x 1 @y 1 @Z Enon) Se or Xx) de? * ¥(y) dy?" Zz) dz where total derivatives have replaced partial derivatives, since each term involves a function of one variable only. If (2.50) is to hold for arbitrary values of the independent coordinates, each of the three terms must be separately constant: Lex _ (2.50) 7 (2.51) where e+e If we arbitrarily choose a” and A? to be positive, then the solutions of the three ordinary differential equations (2.51) are e“**, ec“, e* "=, The potential (2.49) can thus be built up from the product solutions: @ = erinigtiye" VTP (2.52)Sect. 2.9 Separation of Variables; Laplace Equation in Rectangular Coordinates 71 At this stage @ and are completely arbitrary. Consequently (2.52), by linear superposition, represents a very large class of solutions to the Laplace equation. To determine a and B it is necessary to impose specific boundary conditions on the potential. As an example, consider a rectangular box, located as shown in Fig. 2.9, with dimensions (a, b, c) in the (x, y, z) directions. All surfaces of the box are kept at zero potential, except the surface z = c, which is at a potential V(x, y). It is required to find the potential everywhere inside the box. Starting with the requirement that ® = 0 for x = 0, y = 0, z = 0, it is easy to see that the required forms of X, Y, Z are X = sinax Y = sin By (2.53) Z = sinh(Va? + Bz) To have ® = 0 atx = aand y = b, we must have aa = nz and Bb = mz. With the definitions, (2.54) we can write the partial potential ®,,,, satisfying all the boundary conditions except one, Dim = Sin(@,x) SiN(Bn,Y) SiNNYmZ) (2.55) ‘The potential can be expanded in terms of these ®,,,, with initially arbitrary coefficients (to be chosen to satisfy the final boundary condition): B(x, y, 2) = SY Anm Sin(@,X) Sin(Bny) SiNH(Yp»Z) (2.56) nme There remains only the boundary condition @ = V(x, y) at z = Vox, y) = > Aum Sin(ct,) Sin(B,,Y) Sinh(YumC) (2.57) Figure 2.9 Hollow, rectangular box with five sides at zero potential, while the sixth (z = c) has the specified potential ® = V(x, y).72° Chapter 2 Boundary-Value Problems in Electrostatics: I—SI This is just a double Fourier series for the function V(x, y). Consequently the coefficients A,,,, are given by: 4 “4b Aum = te dx hi dy V(x, y) sin(a,x) sin(B,.Y) (2.58) If the rectangular box has potentials different from zero on all six sides, the required solution for the potential inside the box can be obtained by a linear superposition of six solutions, one for each side, equivalent to (2.56) and (2.58). The problem of the solution of the Poisson equation, that is, the potential inside the box with a charge distribution inside, as well as prescribed boundary condi- tions on the surface, requires the construction of the appropriate Green function, according to (1.43) and (1.44). Discussion of this topic will be deferred until we have treated the Laplace equation in spherical and cylindrical coordinates. For the moment, we merely note that the solution given by (2.56) and (2.58) is equiv- alent to the surface integral in the Green function solution (1.44). 2.10 A Two-Dimensional Potential Problem; Summation of a Fourier Series We now consider briefly the solution by separation of variables of the two- dimensional Laplace equation in Cartesian coordinates. By two-dimensional problems we mean those in which the potential can be assumed to be indepen- dent of one of the coordinates, say, z. This is usually only an approximation, but may hold true to high accuracy, as in a long uniform transmission line. If the potential is independent of z, the basic solutions of the previous section reduce to the products eclaxgtay where a is any real or complex constant. The imposition of boundary conditions on the potential will determine what values of a are permitted and the form of the linear superposition of different solutions required. A simple problem that can be used to demonstrate the separation of variables technique and also to establish connection with the use of complex variables is indicated in Fig. 2.10. The potential in the region, 0 = x = a, y = 0, is desired, subject to the boundary conditions that ® = 0 at x = 0 and x = a, while ® = V at y = 0 for 0 = x =a and © > 0 for large y. Inspection of the basic solutions shows that @ is real and that, to have the potential vanish at x = 0 and x = a for all y and as y — , the proper linear combinations are e~®” sin(ax) with a@ = nz/a, The linear combination of solutions satisfying the boundary conditions on three of the four boundary surfaces is thus P(x, y) = > A, exp(—n7ry/a) sin(nmx/a) (2.59) nt The coefficients A,, are determined by the requirement that ® = V for y = 0 =x =a. As discussed in Section 2.8, the Fourier coefficients are A, = 2 f ®(x, 0) sin(rax/a) dx (2.60)Sect. 2.10 A Two-Dimensional Potential Problem; Summation of a Fourier Series. 73 Figure 2.10 Two-dimensional potential problem. With ®(x, 0) = V, one finds A, = AV {1 forn odd an |0 — forn even The potential (x, y) is therefore determined to be a(x, y) = YY | exp(—nayla) sin(nax/a) (2.61) T nod Nt For small values of y many terms in the series are necessary to give an accurate approximation, but for y = a/7 it is evident that only the first few terms are appreciable. The potential rapidly approaches its asymptotic form given by the first term, ®(x, y) > + exp(—zyla) sin(arx/a) (2.62) Parenthetically, we remark that this general behavior is characteristic of all boundary-value problems of this type, independently of whether (x, 0) is a constant, provided the first term in the series is nonvanishing. The coefficient A; (2.60) will be different, but the smooth behavior in x of the asymptotic solution sets in for y = a, regardless of the complexities of (x, 0). This is shown quan- titatively for the present example in Fig. 2.11 where the potential along the two dashed lines, y/a = 0.1, 0.5, of Fig. 2.10 is plotted. The solid curves are the exact potential, the dotted, the first term (2.62). Close to the boundary (y/a = 0.1) the curves differ appreciably, but for y/a = 0.5 the asymptotic form is already an excellent approximation.74° Chapter 2 Boundary-Value Problems in Electrostatics: I—SI 10 08 06 oy) 04 0 1 ala Figure 2.11 Potentials at y/a = 0.1, 0.5 (along the dashed lines of Fig. 2.10) as functions of x/a. The solid curves are the exact solution; the dashed curves are the first term in the series solution (2.61). ‘There are many Fourier series that can be summed to give an answer in closed form. The series in (2.61) is one of them. We proceed as follows. Observing that sin @ = Im(e’”), where Im stands for the imaginary part, we see that (2.61) can be written as av 1 oe, y) = 2m >, = ernionsro With the definition, Z this can be put in the suggestive form, elintnecrty) (2.63) Vv P(x, y) =m >= 7 n odd At this point we can perhaps recall the expansion,* Inf + Z) = Z-43Z7+4Z7-4GZ4 +--+ *Alternatively, we observe that (didZ)(27-1Z"In) = BioZ" = U(L ~ Z). Integration then gives Z"In = -In(1 ~ Z).Sect. 2.11 Fields and Charge Densities in Two-Dimensional Corners and Along Edges 75 Evidently, and 2Vv ee = —In|1 . (x, 9) = = im (3 = 2) (2.64) Since the imaginary part of a logarithm is equal to the phase of its argument, we consider 1+Z_(+Z2)0-Z*)_1-|ZP+2iImz 1-Z [1 - ZP [1 - Zp The phase of the argument of the logarithm is thus tan~'[2 Im Z/(1 — |ZP)]. With the explicit form (2.63) of Z substituted, it is found that the potential becomes 2 D(x, y) = a tan“! (2.65) The branch of the tangent curve corresponds to the angle lying between 0 and w/2. The infinite series (2.61) has been transformed into the explicit closed form (2.65). The reader may verify that the boundary conditions are satisfied and that the asymptotic form (2.62) emerges in a simple manner. The potential (2.64) with Z given by (2.63) is obviously related to functions of a complex variable. This connection is a direct consequence of the fact that the real or the imaginary part of an analytic function satisfies the Laplace equa- tion in two dimensions as a result of the Cauchy—Riemann equations. As men- tioned at the beginning of the chapter, we omit discussion of the complex-variable technique, not because it is unimportant but for lack of space and because completely adequate discussions exist elsewhere. Some of these sources are listed at the end of the chapter. The methods of summation of Fourier series, with many examples, are described in Collin (Appendix A.6). 2.11 Fields and Charge Densities in Two-Dimensional Corners and Along Edges Tn many practical situations conducting surfaces come together in a way that can be approximated, on the small scale at least, as the intersection of two planes. The edges of the box shown in Fig. 2.9 are one example, the corners at x = 0, y = 0 and x = a, y = 0 in Fig. 2.10 another. It is useful therefore to have an understanding of how the potential fields, and the surface-charge densities be- have in the neighborhood of such sharp “corners” or edges. To be able to look at them closely enough to have the behavior of the fields determined in functional form solely by the properties of the “corner” and not by the details of the overall configuration, we assume that the “corners” are infinitely sharp.76 Chapter 2 Boundary-Value Problems in Electrostatics: I—SI The general situation in two dimensions is shown in Fig. 2.12. Two conducting planes intersect at an angle 8. The planes are assumed to be held at potential V. Remote from the origin and not shown in the figure are other conductors or possibly configurations of charges that specify the potential problem uniquely. Since we are interested in the functional behavior of the fields, etc. near the origin, but not in the absolute magnitudes, we leave the “far away” behavior unspecified as much as possible. The geometry of Fig. 2.12 suggests use of polar rather than Cartesian coor- dinates. In terms of the polar coordinates (p, #), the Laplace equation in two dimensions is la aD 1 ab =-—(p—)+555=0 2.66) pop (+ ) pad? CS Using the separation of variables approach, we substitute P(p, 6) = R(pyW(d) This leads, upon multiplication by p7/®, to pd (dR epee Rdp\? dp Since the two terms are separately functions of p and ¢ respectively, each one must be constant: (2.67) pd (dR) _ ld __ Rdp (o #) “pag” (2.68) The solutions to these equations are R(p) = ap’ + bp* 2.69 W(o) = A cos(vp) + B sin(vd) (2.69) For the special circumstance of v = 0, the solutions are R(p) = do + bo In p V(b) = Ao + Bob } om Figure 2.12 Intersection of two conducting planes defining a corner in two dimensions with opening angle B.Sect. 2.11 Fields and Charge Densities in Two-Dimensional Corners and Along Edges 77 These are the building blocks with which we construct the potential by linear superposition. Although not central to our present purpose, we note the general solution of the Laplace equation in two dimensions when the full azimuthal range is per- mitted as, for example, for the potential between two cylindrical surfaces, p = a and p = b, on which the potential is given as a function of ¢. If there is no restriction on @, it is necessary that v be a positive or negative integer or zero to ensure that the potential is single-valued. Furthermore, for v = 0, the constant Bo in (2.70) must vanish for the same reason. The general solution is therefore of the form, P(p, 6) = ay + bo np + Sy a,p” sin(nd + a) nal (2.71) +S dsp" sin(nd + B,) If the origin is included in the volume in which there is no charge, all the b,, are zero. Only a constant and positive powers of p appear. If the origin is excluded, the ,, can be different from zero. In particular, the logarithmic term is equivalent to a line charge on the axis with charge density per unit length A = —27€ bo, as is well known. For the situation of Fig. 2.12 the azimuthal angle is restricted to the range 0 <= $= B. The boundary conditions are that © = V for all p = 0 when ¢ = 0 and # = £. This requires that bp = By = 0 in (2.70) and b = 0 and A = O in (2.69). Furthermore, it requires that v be chosen to make sin(vB) = 0. Hence ya ma1,2... B and the general solution becomes O(p. 6) =V + > agp*® sin(madip) (2.72) ml The still undetermined coefficients a, depend on the potential remote from the corner at p = 0. Since the series involves positive powers of p”®, for small enough p only the first term in the series will be important.* Thus, near p = 0, the po- tential is approximately ®(p, &) = V + ayp™® sin(/B) (2.73) The electric field components are ab za, =~ Pp E,(p, 6) = ip (=!) sin(arqp/B) (2.74) Ea(p, 6) = Ty p"™ cos( pip) *Here we make a necessary assumption about the remote boundary conditions, namely, that they are such that the coefficient a, is not zero. Ordinarily this is of no concern, but special symmetries might make a, or even dz, etc., vanish. These unusual examples must be treated separately.78 Chapter 2 Boundary-Value Problems in Electrostatics: I—SI The surface-charge densities at 6 = 0 and = Bare equal and are approximately a(p) = &Eg(p, 0) = ~SE pio (2.75) The components of the field and the surface-charge density near p = 0 all vary with distance as p\“”~1. This dependence on p is shown for some special cases in Fig. 2.13. For a very deep corner (small 8) the power of p becomes very large. Essentially no charge accumulates in such a corner. For 8 = 7 (a flat surface), the field quantities become independent of p, as is intuitively obvious. When B > 7, the two-dimensional corner becomes an edge and the field and the surface- charge density become singular as p —> 0. For B = 27 (the edge of a thin sheet) the singularity is as p-“. This is still integrable so that the charge within a finite distance from the edge is finite, but it implies that field strengths become very large at the edges of conducting sheets (or, in fact, for any configuration where B> 7). The preceding two-dimensional electrostatic considerations apply to many three-dimensional situations, even with time-varying fields. If the edge is a sharp edge of finite length, as the edge of a cube away from a corner, then sufficiently close to the edge the variation of the potential along the edge can be ignored. The two-dimensional considerations apply, although the coefficient a in (2.75) may vary with distance along the edge. Similarly, the electrostatic arguments are valid even for time-varying fields. The point here is that with time dependence another length enters, namely, the wavelength. Provided one is concerned with distances away from the edge that are small compared to a wavelength, as well as other relevant distances, the behavior of the fields reduces to electrostatic or magnetostatic behavior. In the diffraction of microwaves by a hole in a thin conducting sheet, for example, the fields are singular as p~ as p > 0, where p is the distance from the boundary of the hole, and this fact must be taken into account in any exact solution of the diffraction problem. The singular behavior of the fields near sharp edges is the reason for the effectiveness of lightning rods. In the idealized situation discussed here the field strength increases without limit as p > 0, but for a thin sheet of thickness d with a smoothly rounded edge it can be inferred that the field strength at the surface will be proportional to d~". For small enough d this can be very large. In ab- solute vacuum such field strengths are possible; in air, however, electrical break- down and a discharge will occur if the field strength exceeds a certain value (depending on the exact shape of the electrode, its proximity to the other elec- Figure 2.13 Variation of the surface-charge density (and the electric field) with distance p from the “corner” or edge for opening angles B = 7/4, 7/2, 7, 37/2, and 27.Sect. 2.12 Introduction to Finite Element Analysis for Electrostatics 79 trodes, etc., but greater than about 2.5 X 10° V/m for air at normal temperature and pressure (NTP), sometimes by a factor of 4). In thunderstorms, with large potential differences between the ground and the thunderclouds, a grounded sharp conducting edge, or better, a point (see Section 3.4), will have breakdown occur around it first and will then provide one end of the jagged conducting path through the air along which the lightning discharge travels. 2.12 Introduction to Finite Element Analysis for Electrostatics Finite clement analysis (FEA) encompasses a variety of numerical approaches for the solution of boundary-value problems in physics and engineering. Here we sketch only an introduction to the essential ideas, using Galerkin’s method for two-dimensional electrostatics as an illustration. The generalization to three dimensions is mentioned briefly at the end. The reader who wishes a deeper introduction may consult Binns, Lawrenson, and Trowbridge, Ida and Bastos, Sadiku, Strang, or Zhou. Consider the Poisson equation, V7 = —g in a two-dimensional region R, with Dirichlet boundary conditions on the boundary curve C. We construct the vanishing integral, ie [6 Vu + gd] dx dy = 0 (2.76) where $(x, y) is a test function specified for the moment only as piecewise con- tinuous in R and vanishing on C. Use of Green’s first identity on the first term above leads to Ee [Vb- Vu — gh] dx dy = 0 (2.77) The surface integral vanishes because # vanishes on C. Galerkin’s method con- sists first of approximating the desired solution (x, y) by a finite expansion in terms of a set of localized, linearly independent functions, #,(x, y), with support only in a finite neighborhood of x = x;, y = y;. For definiteness, we imagine the region R spanned by a square lattice with lattice spacing h. Then a possible choice for $4(x, y) is, Gylx, y) = (A - [x = x|A) — ly — yj") (2.78) for |x — x, >py, where p = V(@™+y*) and po = VG% + Yo)] the leading term in the potential is _ 4A (Xayo)(2Y) oo, om ep Interpret. A point charge is placed a distance d > R from the center of an equally charged, isolated, conducting sphere of radius R. (a) Inside of what distance from the surface of the sphere is the point charge attracted rather than repelled by the charged sphere? (b) What is the limiting value of the force of attraction when the point charge is located a distance a (= d — R) from the surface of the sphere, if a << R? (©) What are the results for parts a and b if the charge on the sphere is twice (half) as large as the point charge, but still the same sign? [Answers: (a) d/R — 1 = 0.6178, (b) F = —@i(16 76a"), i.e., image force, (c) for Q = 2g, dIR — 1 = 0.4276; for Q = q/2, d/R — 1 = 0.8823. The answer for part b is the same.] (a) Show that the work done to remove the charge q from a distance r > a to infinity against the force, Eq. (2.6), of a grounded conducting sphere is ga We ore —@) Relate this result to the electrostatic potential, Eq. (2.3), and the energy dis- cussion of Section 1.11. (b) Repeat the calculation of the work done to remove the charge q against the force, Eq. (2.9), of an isolated charged conducting sphere. Show that the work done is 1 We Ame lx Relate the work to the electrostatic potential, Eq. (2.8), and the energy dis- cussion of Section 1.11. @) wor The electrostatic problem of a point charge q outside an isolated, charged con- ducting sphere is equivalent to that of three charges, the original and two others, one located at the center of the sphere and another (“the image charge”) inside the now imaginary sphere, on the line joining the center and the original charge.2.7 Ch.2 Problems 87 If the point charge and sphere are replaced by two conducting spheres of radii r, and r,, carrying total charges Q, and Q,, respectively, with centers separated by a distance d > r, + r,, there is an equivalence with an infinite set of charges within each sphere, one at the center and a set of images along the line joining the centers The charges and their locations can be determined iteratively, starting with a charge qa(1) at the center of the first sphere and q,(1) correspondingly for the second sphere. The charge q,(1) has its image q,(2) within the first sphere and vice versa. Then the image charge within the first sphere induces another image within the second sphere, and so on. The sum of all the charges within each sphere must be scaled to be equal to Q, or Qy. The electrostatic potential outside the spheres, the force between the spheres, ete. can be found by summing the contributions from all the charges. (a) Show that the charges and their positions are determined iteratively by the relations, Gi) = —MaQG — WldsG- 1), XV) = vali -— Dali) Gli) = ~rogd — Wdej — 1), Xo) = rildG-— 1), dal) for j = 2,3,4,..., with d,(1) = d,(1) = d, and x,(1) = x,(1) = 0. (b) Find the image charges and their locations as well as the potentials on the spheres and force between them by means of a suitable computer program. [In computing the potential on each sphere, evaluate it in different places: e.g., in the equatorial plane and at the pole opposite the other sphere. This permits a check on the equipotential of the conductor and on the accuracy of computation.) ~ xa) d— x,(i) (©) Asan example, show that for two equally charged spheres of the same radius R, the force between them when almost in contact is 0.6189 times the value that would be obtained if all the charge on each sphere were concentrated at its center. Show numerically and by explicit summation of the series that the capacitance of two identical conducting spheres in contact is C/47repR = 1.3863 - - [= In 4]. Reference: J. A. Soules, Am. J. Phys. 58, 1195 (1990). Consider a potential problem in the half-space defined by z = 0, with Dirichlet boundary conditions on the plane z = 0 (and at infinity). (a) Write down the appropriate Green function G(x, x’). (b) If the potential on the plane z = 0 is specified to be & = V inside a circle of radius a centered at the origin, and @ = 0 outside that circle, find an integral expression for the potential at the point P specified in terms of cylindrical coordinates (p, , z). (©) Show that, along the axis of the circle (p = 0), the potential is given by @=Vi1-—— ( Va + 2) (a) Show that at large distances (p” + z? >> a’) the potential can be expanded in a power series in (p? + 2)", and that the leading terms are Vaz _[,__ 3a _ 27 Gee ae Ap? + + Gora tay 8p" + 2’)? Verify that the results of parts c and d are consistent with each other in their common range of validity.88 Chapter 2 28 29 2.10 241 Boundary-Value Problems in Electrostatics: I—SI A two-dimensional potential problem is defined by two straight parallel line charges separated by a distance R with equal and opposite linear charge densities A and —A. (a) Show by direct construction that the surface of constant potential V is a cir- cular cylinder (circle in the transverse dimensions) and find the coordinates of the axis of the cylinder and its radius in terms of R, A, and V. (b) Use the results of part a to show that the capacitance per unit length C of two right-circular cylindrical conductors, with radii a and b, separated by a distance d>arb,is 208 cosh“! ( £ (©) Verify that the result for C agrees with the answer in Problem 1.7 in the appropriate limit and determine the next nonvanishing order correction in powers of a/d and bid. (d) Repeat the calculation of the capacitance per unit length for two cylinders inside each other (d < |b — al). Check the result for concentric cylinders (d= 0). An insulated, spherical, conducting shell of radius a is in a uniform electric field Eo. If the sphere is cut into two hemispheres by a plane perpendicular to the field, find the force required to prevent the hemispheres from separating (a) _ if the shell is uncharged; (b) _ if the total charge on the shell is Q. A large parallel plate capacitor is made up of two plane conducting sheets with separation D, one of which has a small hemispherical boss of radius a on its inner surface (D >> a). The conductor with the boss is kept at zero potential, and the other conductor is at a potential such that far from the boss the electric field between the plates is Ey. (a) Calculate the surface-charge densities at an arbitrary point on the plane and on the boss, and sketch their behavior as a function of distance (or angle). (b) Show that the total charge on the boss has the magnitude 3r¢E a”. (©) If, instead of the other conducting sheet at a different potential, a point charge q is placed directly above the hemispherical boss at a distance d from its center, show that the charge induced on the boss is 7 | eee ¢ 4f Were | A line charge with linear charge density 7 is placed parallel to, and a distance R away from, the axis of a conducting cylinder of radius b held at fixed voltage such that the potential vanishes at infinity. Find (a) the magnitude and position of the image charge(s); (b) the potential at any point (expressed in polar coordinates with the origin at the axis of the cylinder and the direction from the origin to the line charge as the x axis), including the asymptotic form far from the cylinder; (©) the induced surface-charge density, and plot it as a function of angle for Rib = 2, 4 in units of 7/27b; (a) the force per unit length on the line charge.2.12 2.13 2.14 2.15 Ch.2 Problems 89 Starting with the series solution (2.71) for the two-dimensional potential problem with the potential specified on the surface of a cylinder of radius b, evaluate the coefficients formally, substitute them into the series, and sum it to obtain the po- tential inside the cylinder in the form of Poisson’s integral: 2 Pap 0 PF +p — pose — 4 ? 96.6)= + [6,6 What modification is necessary if the potential is desired in the region of space bounded by the cylinder and infinity? (a) Two halves of a long hollow conducting cylinder of inner radius b are sepa- rated by small lengthwise gaps on each side, and are kept at different poten- tials V, and V>, Show that the potential inside is given by |_|. B(p, ) = neh + tan (ee cos ) where ¢ is measured from a plane perpendicular to the plane through the gap. (b) Calculate the surface-charge density on each half of the cylinder. A variant of the preceding two-dimensional problem is a long hollow conducting cylinder of radius b that is divided into equal quarters, alternate segments being held at potential +V and —V. (a) Solve by means of the series solution (2.71) and show that the potential inside the cylinder is _4V & (p\"” sinf(4n + 2)01 Pe, 0-23 (6) mn +1 (b) Sum the series and show that 2p 09, 4) = ton (= 138) op (c) Sketch the field lines and equipotentials. (a) Show that the Green function G(x, y;.x', y’) appropriate for Dirichlet bound- ary conditions for a square two-dimensional region, 0 = x = 1,0 fa(y. y") sin(n mx) sin(n mx’) where g,(y, y’) satisfies : (2 -') S10. ¥') = —4r—(y’ = y) and guy, 0) = g4(0, 1) = 0 (b) Taking for g,(y, y’) appropriate linear combinations of sinh(n7y’) and cosh(n7y’) in the two regions, y’ < y and y’ > y, in accord with the boundary conditions and the discontinuity in slope required by the source delta function, show that the explicit form of G is Gls vex y) =8> Fann Snes) sin(rn’) sinh(nry.) sinhora(1 — ys)] where y.(y.) is the smaller (larger) of y and y’.90 Chapter 2 2.16 2.17 2.18 Boundary-Value Problems in Electrostatics: I—SI A two-dimensional potential exists on a unit square area (0 = x = 1,0 = y = 1) bounded by “surfaces” held at zero potential. Over the entire square there is a uniform charge density of unit strength (per unit length in z). Using the Green function of Problem 2.15, show that the solution can be written as $ sinl@m + ax] [, _ cosh[Qm + Naty =D) io Qm+1)p cosh{(2m + 1) 7/2] (a) Construct the free-space Green function G(x, y; x’, y’) for two-dimensional electrostatics by integrating 1/R with respect to (z’ — z) between the limits +Z, where Z is taken to be very large. Show that apart from an inessential constant, the Green function can be written alternately as CGy xy). NG 4) = —In[p? + p? — 2pp' cos(d — #’)] (b) Show explicitly by separation of variables in polar coordinates that the Green function can be expressed as a Fourier series in the azimuthal coordinate, 4 eae io - : Gaz > em gulp, 2") where the radial Green functions satisfy ta( ia a a(p— p' +3( Se) (cee ap’ p p Note that g,,(p, p’) for fixed p is a different linear combination of the solutions of the homogeneous radial equation (2.68) for p’ < p and for p’ > p, with a discontinuity of slope at p’ = p determined by the source delta function. (©) Complete the solution and show that the free-space Green function has the expansion G(p, 4; p', #’) = —In(p2) + 2 > i (2) cosfen(d — ')] where p_(p.) is the smaller (larger) of p and p’. (a) By finding appropriate solutions of the radial equation in part b of Problem 2.17, find the Green function for the interior Dirichlet problem of a cylinder of radius b [g,,(p, p’ = b) = 0. See (1.40)]. First find the series expansion akin to the free-space Green function of Problem 2.17. Then show that it can be written in closed form as G- n| +p | b°(p* + p’? — 2pp' cos( — 4')) 2 92\(? — p) + BP lp — pt o=n[2 jest natal et) 2 |p — e'| (b) Show that the solution of the Laplace equation with the potential given as B(b, #) on the cylinder can be expressed as Poisson’s integral of Problem 2.12. (©) What changes are necessary for the Green function for the exterior problem (b < p< ®), for both the Fourier expansion and the closed form? [Note that the exterior Green function is not rigorously correct because it does not vanish