0% found this document useful (0 votes)
287 views745 pages

Mathematical Physics - E.butkov

Mathematical Physics Textbook

Uploaded by

SS1989RF
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
287 views745 pages

Mathematical Physics - E.butkov

Mathematical Physics Textbook

Uploaded by

SS1989RF
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
You are on page 1/ 745
MATHEMATICAL PHYSICS EUGENE BUTKOV St. John’s University, New York Se é 3g ee *o SryoeS ADDISON-WESLEY PUBLISHING COMPANY Realing, Massachusetts - Menle Pork, California - London - Syéney - Manila oes oe OM. Sept, “eRRS Stove’ WORLD STUDENT SERIES FIRST PRINTING 1973 A complete and unabridged reprint of the original American textbook, this World Student Series edition may be sok! only in those countries to which it is con- signed by Addison-Wesley or its authorized (rade distributors. [t may not be re-exported from the country to which it has been consigned, and it may not be sold in the United States of Americaor its possessions. Copyright © 1968 by Addison-Wesley Publishing Company, inc. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, with- out the prior written permission of the publisher. Original edition published ‘in the United, States of America. Published simultaneously in Canada. Philippines copyright 1968 Library of Congress Catalog Card Number: 68- 11391, PREFACE During the past decade we have witnessed a remarkable increase iri the number of students seeking higher education as well as the development of many new colleges and universities. The inevitable nonuniformity of conditions present in different institutions necessitates considerable variety in purpose, general approach, and the level of instruction in any given discipline. This has naturally contributed to the proliferation of texts on almost any topic, and the subject of mathematical pliysics is no exception. There is a number of texts in this field, and some of them are undoubtedly of outstanding quality. Nevertheless, many teachers often feel that none of the existing textsis properly suited, for one reason or another, for their particular courses. More important, students sometimes complain that they have difficulties studying the subject from texts of unquestionable merit. This is not as surprising as it sounds: Some texts have an encyclopedic character, with the material arranged in a different order from the way it is usually taught; others become too much involved in complex mathematical analysis, preempting the available space from practical examples; still others cover a very wide variety of topics with utmost brevity, leaving the student to struggle with a number of difficult questions of theoretical nature. True enough, a well-prepared and bright student should be able to find his way through most of such difficulties. A less-gifted student may, however, find il very difficult to grasp and absorb the multitude of new concepts strewn across an ad- vanced text. ‘Under these circumstances, it seems desirable to give more stress to the peda- gogical side of a text to make it more readable to the student and more suitable for independent study. Hopefully, the present work represents a step in this direction. It has several features designed to conform to the path an average student may conceivably follow in acquiring the knowledge of the subject. First, the inductive approach is used in each chapter throughout the book. Fol- Towing the fundamentals of modern physics, the text is almost entirely devoted to linear problems, but the unifying concepts of linear space are fully developed rather Jate in the book after the student is exposed to a number of practical mathematical techniques. Also, almost every chapter starts with an example or discussion of elementary nature, with subject matter that is probably familiar to the reader. The introduction of new concepts is made against 2 familiar background and is later extended to more sophisticated situations. A typical example is Chapter %, where the basic aspects of partial differential equations are illustrated using the “elementary functions” exclusively. Another facet of this trend is the repeated use of the harmonic oscillator and the stretched string as physical models: no v vi PRETACE attempt is made to solve as many problems for the student as possible, but rather to show how various methods can be used to the same end within a familiar phys- ical context. In the process of learning, students inevitably pose a number of questions necessary to clarify the material under scrutiny. While most of these questions naturally belong to classroom discussion, it is certainly beneficial to attempt to anticipate some of them in a text. The Remarks and many footnotes are designed to contribute to this goal. The author hopes they answer some questions in the mind of the student as well as suggest some new ones, stimulating an interest in further inquiry. A number of cross-references serves a similar purpose, inviting the reader to make multipte use of various sections of the book. The absence of numbered formulas is intentional: if the student bothers to look into the indicated section or page, he should not simply check that the quoted formula “is indeed there,” but, rather, glance through the text and recall its origin and meaning. The question of mathematical rigor is quite important in the subject treated here, although it is sometimes controversial, It is the author's opinion that a theo- retical physicist should know where ‘he stands, whether he is proving his own deduc- tions, quoting somebody else’s proof, or just offering a reasonable conjecture. Consequently, he should be trained in this direction, and the texts should be written in this spirit. On the other hand, it would be unwise to overload every student with mathematics for two main reasons: first, because of the limitations of time in the classroom and the space in a text, and second, because physicists are apt to change their mathematical postulates as soon as experimental physics lends support to such suggestions. The reader can find examples of the latter philosophy in Chapters 4 and 6 of the text. Whether the author was able to follow these principles is left to the judgment of users of this book. Each chapter is supplied with its share of problems proportional to the time presumed to be allotted to its study. The student may find some of the problems rather difficult since they require more comprehension of the material rather than sheer technique. ‘To balance this, a variety of hints and explanations are often supplied. Answers are not given because many problems contain the answer in their formulation; the remaining ones may be used to test the ability of the student for independent work. The exercises within the text can be used as problems to test the students’ manipulative skills. For many of the methods of instruction of mathematical physics presented in this book, the author is indebted to his own teachers at the University of British Columbia and McGill University. The encouragement of his colleagues and students at St. John’s University and Hunter College of the City University of New York is grcatly appreciated. Also, the author wishes to thank Mrs. Ludmilla Verenicin and Miss Anne Marie Nowom for their help in the preparation of the manuscript. Palo Alto, Calif. EB. August 1966 Chapter 2 21 22 23 24 25 2.6 27 28 29 2.10 21f 212 2AZ 2.14 215 CONTENTS Vectors, Matrices, and Coordinates Introduction . Vectors in Cartesian Coordinate Systems. Changes of Axes, Rotation Matrices 2. Repeated Rotations. Matrix Multiplication. 2 2, Skew Cartesian Systems. Matricesin General... Scalar and Vector Fields Vector Fields in Plane. .. Vector Fields in Space . Cu ear Coordinates Functions of a Complex Variable Complex Numbers Basic Algebra and Geometry of Complex Numbers De Moivre Formula and the Calculation of Roots . Complex Functions. Euler's Formula... Applications of Euler's Formula... Multivatued Functions and Riemann Surfaces Analytic Functions, Cauchy Theorem Other Integral Theorems. Cauchy Integral Formula Complex Sequences and Series : Taylor and Laurent Series - Zeros and Singularities. . “The Residue Theorem and its Applications | Conformal Mapping by Analytic Functions . Complex Sphere and Point at Tefnity woe ee Integral Representations. woe Linear Differential Equations of Second Order Genera! Introduction, The Wronskian General Solution of The Homogeneous Equation The Nonhomogencous Equation. Variation of Constants . Power Scrics Solutions . The Frobenius Method . soe Some other Methods of Solution . vi ann nH 20 26 34 44 45 48 49 Ss 54 58 62. 7 7B 83 97 102 104 123 125 126 128 130 147 vi Chapter 4 4l 42 43 4A 45 46 aT 48 Chapter 5 SA 5.2 33 54 35 56 5.7 58 59 5.10 Chapter 6 6.1 62 6.3 64 65 6.6 6.7 6.8 6.9 6.10 Chapter 7 Td CONTENTS. Fourier Series Trigonometric Series 5 6 6 se Definition of Fourier Series Examples of Fourier Series Parity Properties, and Cosine series Complex Form of Fourier Series . Pointwise Convergence of Fourier Series Convergence in the Mean. Applications of Fourier Series . The Laplace Transformation Operational Calculus The Laplace Integral Basie Properties of Laplace Transform The Inversion Problem . ‘The Rational Fraction Decomposi The Convolution Theorem . Additional Properties of Laplace Transform Periodic Functions. Rectification . ‘The Mellin Inversion Integral . Applications of Laplace Transforms . n Concepts of the Theory of Distributions Strongly Peaked Functions and The Dirac Delta Function Delta Sequences . ‘The 3-Caleulus . Representations of Delta Functions Applications of The 5-Calculus Weak Convergence . Correspondence of Functions and Distributions | Properties of Dist Sequences and Series.of Distributions {Distributions in N dimensions . Fourier Transforms Representations of a Function 7.2 Examples of Fourier Transformations 7.3 Properties of Fourier Transforms . 7.4 Fourier Integral Theorem : 715. Fourier Transforms of Distributions.» 7.6 Fourier Sine and Cosine Transforms 7.4 Applications of Fourier Transforms. The Principe ‘of Causality . 154 135, 157 161 165 167 168 172 179 180 184 187 189 194 206 210 22h 223 226 229 232 236 240 245 250 257 260 262 266 269 27h 273 276 Chapter 8 al B2 83 84 a5 8.6 87 88 a9 Chapter 9 O11 92 93 a4 95 96 OT 98 99 9.10 OAL Chapter 10 10.t 10.2 10.3 10.4 10.5 10.6 10.7 10.8 10.9 Chapter 11 wl 1.2 11.3 114 11.5 116 17 HB CONTENTS Partial Differential Equations The Stretched String. Wave Equation The Mettiod of Separation of Variables. Laplace and Poisson Equations The Diffusion Equation : Use of Fourier and Laplace ‘Transforms ‘The Method of Eigenfunction Expansions and Finite Transforms ‘Continuous Eigenvalue Spectrum . . Vibrations of a Membrane. Degeneracy . Propagation of Sound. Heimboltz Equation Special Functions Cylindrical and Spherical Coordinates The Common Boundary-Value Problems The Sturm-Liouville Problem . Self-Adjoint Operators . Legendre Polynomials Fourier-Legendre Series Bessel Functions Associated Legendre Functions and Spherical Harmonies - Sphericat Bessel Functions . Neumann Functions : Modified Bessel Functions - Fini --Dimensional Linear Spaces Oscillations of Systems with Two Degrees of Freedom Normal Coordinates and Linear Transformations Vector Spaces, Bases, Coordinates Lincar Operators, Matrices, Inverses . Changes of Basis . {nner Product. Orthogenality. Unit ry Operators . The Metric. Generalized Orthogonality Eigenvatue Problems. Diagonatization Simultaneous Diagonalization.. Infinite-Dimensional Vector Spaces Spaces of Functions : The Postulates of Quantum Mechanics ‘The Harmonic Osc so Matrix Representations of Lincar Operators Algchraic Methods of Solution : Bases with Generalized Orthogonality Streiched String with a Discrete Mass in the Middle Applications of Figenfunctions ix 287 298 295 27 299 304 308 313 319 332, 34 337 MO 42 350 355 372 3at 388 494 40s ait 49 aA 433 ay? al a3 451 463 467 47t 476 483 488 492 495 % CONTENTS Chapter 12 Green’s Functions 121 Introduction. 2. rr) 122 Green's Function for the Starm-Liouville Operator... . «508 12.3 Series Expansions for G(xlf) . 2... 7 ee ee SIA 12.4 Green’s Functions in Two Dimensions . 5 2 5 1 1). 520 12.5 Se ee SB 12.6 Green's Functions with Refiection Properties 2. . 0... - 527 12.7 Green's Functions for Boundary Conditions . 9... . . . 331 12.8 The Green's Function Method . 2 2 1 1 we eee 86 12.9 A Case of Continuous Spectrum 2. 2. ee SB Chapter 13 Variational Methods 13.1. The Brachistochrone Problem. 2 2. 2 2. 2.) e - 588 13.2 The Euler-Lagrange Equation. 2... . . . . © ss 554 13.3. Hamilton's Principle. coe eee 580 13.4 Problems invotving Sturm-Liouville Operators... . . . 562 13.5 The Rayleigh Ritz Method So 585 13.6 onal Problems with Constraints. | | 5...) (36? 137 v ional Formulation of Eigenvalue Problems . . . - . . $73 13.8 Variational Problems in Many Dimensions . ... 37 13.9 Formulation of Eigenvalue Problems by The Ratio Method . |. 581 Chapter 14. Traveling Waves, Radiation, Scattering, 14.1 Motion of Infinite Stretched String... 2 2 we ee + 589 14.2 Propagation of Initial Conditions... co ee 582 143. Semiinfnite String. Use of Symmetry Properties . |... 5 595 144 Energy and Power Flow ina Stretched String... 1... 599 14.5 Generation of Waves ina Stretched String. . . . . . . - 603 14.6 Radiation of Sound from a Pulsating Sphere... . 2. + GIT 14.7 The Retarded Potential. coe ee AD 148 Traveling Waves in Nonhomogeneous Media... . .. @24 14.9 Scattering Amplitudes and Phase Shifts. Los OB 14.10. Scattering in Three Dimensions. Partial Wave Analysis... 5. 633 Chapter 15. Perturbation Methods 15.1 ‘Introduction . So 152 The Born Approximation - . Lo ee ee AT 153. Perturbation of Eigenvalue Problems|.) 5 ss. ss 680 15.4 First-Order Rayleigh-Schrddinger Theory... . . . + - 653 15.5. The Second-Order Nondegenerate Theoty . . . - 5 1 + + 658 15.6 The Case of Degenerate Eigenvalues. 2 - . Chapter 16 Tensors 16.1 Introduction. - ee eee TL 16.2 ‘Two-Dimensional Stresses Soe ee ke ee ee OD 16.3 16.4 16.5 16.6 16.7 16.8 16.9 16.10 16.11 Cartesian Tensors 5 1. ee Algebra of Cartesian Tensors | | Kronecker and Levi-Civita Tensors. Pseudotensors CONTENTS Derivatives of Tensors. Strain Tensor and Hooke’s Law . Tensors in Skew Cartesian Frames, Covariant and Contravariant Representations . General Tensors. . . Algebra of General Tensors. Relative Tensors. The Covariant Derivative . . Calculus of General Tensots . . Index . xi 676 681 684 687 696 700 705 TUL 7S Ry CHAPTER 1} VECTORS, MATRICES, AND COORDINATES 1.1 INTRODUCTION To be able to follow this text without undue difficulties, the reader is expected to have adequate preparation in mathematics and physics. This involves a good working knowledge of advanced calculus, a basic course in differential equations, and a basic course in undergraduate algebra. Rudimentary knowledge of complex numbers, matrices, and Fourier scries is very desirable but not indispensable. As for the subjects in physics, the reader should have completed the standard undergraduate. training in mechanics, thermodynamics, electromagnetism, and atomic physics. Despite these prerequisites, a need is often recognized for reviewing some of the preparatory material at the beginning of a text. Let us follow this custom and devote some time to the subject of vector analysis which has a bearing, in more than one way, on the material developed in this text, Of course, such @ review must be brief and we must omit all the details, in particular those involving mathematical proofs. The reader is referred to standard textbooks in advanced calculus and vector analysis* for a full discussion. On the other hand, we hope to draw attention to some interesting points not always emphasized in commonly used texts. 1.2 VECTORS IN CARTESIAN COORDINATE SYSTEMS In many elementary tcatbooks a vector is defined as a quantity characterized by magnitude and direction, We shalt see in Chapter 40 that vectors are much more general than this, but it is fair to say that the concept of vectors was first intro- duced into mathematics (by physicists) to represent “quantities with direction,” e.g., displacement, velocity, force, ete. Doubtless, they are the simplest and most familiar kinds of vectors. As we well know, quantities with direction can be graphically represented by arrows and are subject to two basic operations: a) multiplication by a scalar,t 6) addition. These operations are illustrated in Fig. 1.1. * For example, A. E. Taylor, Advanced Calculus; T. M. Apostol, Mathematical Analysis; W. Kaplan, Advanced Calculus. + Until we are ready to discuss complex vectors (Chapler 10) we shall assume that scalars are reat numbers. 1 2 VECTORS, MATRICES, AND COORDINATES 1.2 ee a" ta In many cases we can plot various vectors from a ‘single point, the origin. Then each vector can be characterized by the coordinates of its “tip.” Various coordinate systems are possible but the cartesian coordinate systems are the most convenient. The reason is very simple and very deep: The cartesian coordinates | of a pointcan serve as the components of the corresponding vector at the same time. ‘Thi is itlustrated in Fig. 1.2, where orthogonal cartesian systems, in plane and in space, are selected. Note that the three-dimensional system is “right-handed” ;* in general, we shall use right-handed systems in this book. Figure 1.4 Figure 1.2 We can now associate with a vector u (in space)a set of three scalars (ite, ty, te)s such that au will correspond to (Aux, Ay, Aue) and u -+ ¥ will correspond to (ug + de, Uy + by, Ue + 04). Note that no suck relations hold, in general, if a vector is characterized by other types of coordinates, ¢.g., spherical or cylindrical. In addition, orthogonal cartesian coordinates result in very simple formulas for other commen quantities associated with vectors, such as a) length (rhagnitude) of a vector: Nol =u = (ud + uh ay'?, b) projections of a vector on coordinate axes:f tz = wcos (u, i), uy = ucos (U, j), u, = u cos (u, k), * Rotation of the x-axis by 90° to coincide with the y-axis appears counterclockwise for all observers with z > 0. + Standard notation is used: The symbols i, j, k are unit vectors in x-, ¥-, and z-directions, respectively. The symbot (u, ¥} stands for the angle between the directions given by wand ¥, 12 VECTORS IN CARTESIAN COORDINATE SYSTEMS = 3 ) projection of a vector on an arbitrary direc- = tion defined by vector s (Fig. 1.3): OP =u, = wcosy = uy 60s (8,1) + uy €05 (6, J} + 1, 608 (6,10, d) scalar product (dot product) of two vectors: (0+) = Wwc08 (HY) = de + ydy + ids, ©) vector product (cross product): . , Figure 1.3 Wa X 1] = (pe — Had) Cade — WE + (Udy — MYODK. The important distinctive feature of the cross product is that fu X ¥] + [¥ X uw], namely, it is not commutative; rather, it is anicommutative: fe xX vy} = -[v Xo} Remark. Apart from its important physical applications, the cross product of two vectors leads us to the concept of “oriented area.” The magnitude of [u X ¥}, namely no [sin (a, 1), is equal Lo the atea of the parallelogram formed by wand ¥. The direction of [u X ¥) can serve to distinguish the “positive side” of the paraflelogram from its “negative side.” Figure 1.4 shows two views of the same parallelogram illustrating, this idea, * [u xv] Positive side Figure 1.6 Closely related to this property is the concept of a right-handed triple of vectors. ,Any three vectors u, v, and w, foken in this order, are said to form a right- handed (or positive) triple if the so-called #riple product iw X J) is posilive.* This happens if w is on the same side of the plane defined by the vectors u and ¥, as illustrated in Fig. 1.5. It is not hard to verify that ({w x ¥]- ¥) represents, in this case, the volume V of the parallelepiped formed by the vectors u, ¥, and w. * These vectors form a left-handed (negative) triple if (tu X v1 w)< 0. 4 VECTORS, MATRICES, AND COORDINATES 13 Exercise. Show that v = \(a X 1) -W)] under any circumstances. Show also that the sign of the triple product is unchanged under cyclic permutation of u, ¥, and w, that is, {la X v]- 9) = (lw Xu} y) = (ly X wu). Figure 1.5 13. CHANGES OF AXES, ROTATION MATRICES We have seen that a given vector wis associated with a set of three numbers, namely its components,* with respect to some orthogonal cartesian system. However, it is clear that if the system of axes is changed, the components change as well. Let us study these changes. Consider, for vectors in a plane, a change in the system of axes which is pro- duced by a rotation by the angle 6, as illustrated in Fig. 1.6. The old system is (x, y) and the new system is (x’, y’). Since u = uzi + u,j, the x’-component of u is the sum of projections of vectors wei and u,j on the x-axis, and similarly for the p’-component.{ From the diagram we see that this yields jub = upcos@ + tysing, fy = —uzsin@ + ay cos. It is instructive to note that the angle between the x‘- and y-axes is (x/2 — 6) while the angle between the y’- and x-axes is (x/2 + 8). In view of sino = cos (5 ~ 2) / ~sin ¢ = vos (3 40): we see that all four coefficients in the above equations represent cosines of the angles between the respective axes, Let us how turn to the three-dimensional case. Figure 1.7 represents two orthogonal cartesian systems, both right-handed, centered at O. It is intuitively clear that the primed system can be obtained from the unprimed one by the mo- tion of a “rigid body about a fixed point.” In fact, it is shown in almost any text- book on mechanics} that such a motion can be reduced to a rotation about some axis (Euler’s theorem). and i * Instead of “components,” the lerm “coordinates of a vector” is often used (sex also Section 10,3). + For example, Goldstein, Classical Mechanics, Section 4.6. 13 CHANGES OF AXES. ROTATION MATRICES 5 Write a = ust + uj + uk, collect contributions to ui from the three vectors w.i, uj, and u,k, and obtain i, = uy cos (i', i) + u, cos (¥, §) -F u.cos @, k), where i’ is, of course, the unit vector in the x’-direction. Note that the cosines involved are the directional cosines of the x'-direction with respect to the unprimed system or, for that matter, the dot products of i” with f, j, and k. Figure 1.7 It is clear that similar formulas can be written for uf, and ui. At this stage, however, it is very convenient to switch to a different notation: Instead of writing (tz, yy te), let Us write (15, ue, 43) and similarly (4, 4, w4) for (x, af, 2). More- over, denote by cnn the angle between the mth primed axis and the nth unprimed axis (three such angles are marked on Fig. 1.6) and by dna the corresponding cosine (that is, @mn = COSamn). This new notation permits us to write the transformation formulas in an easily memorized pattern: wy = ayitt + ayaa + A133, 4 = Gait + agaitg + aastts, Wh = Gait + dgaits + agate, or, if desired, in the compact form thn 3 D Gantt, = 1, 2,3). ‘+ From this analysis we conclude that the new components (u/,, w), 1%) can be obtained from the old components (1;, tv, va) with the help of adne coellicients. 6 VECTORS, MATRICES, AND COORDINATES 13 These nine coefficients, arranged in the self-explanatory pattern below are said to form a matrix* We shall denote matrices by capital letters. Columns Ist 2nd 3d Ist And > Rows ard Matrix A has three rows and three columns; the individual coefficients dmx are referred to as matrix elements, or entries. It is customary to use the first subscript ~ (min our case) to label the row and the second one to label the column; thus the matrix element aj: should be located at the intersection of the kth row with the ith column. The set of elements in a given row is very often called a row vector and the set of elements in a column, a column vector. This nomenclature is justified by the fact that any three numbers can be treated as components of some vector in space. However, at this stage it is worthwhile to make a digression and establish a geo- metric interpretation for the column vectors of 4. The reader should pay par- ticular attention to the argument because of its general significance. Let us imagine a unit vector u. Suppose the unprimed system was oriented in such a way that u was along the x-axis. Then the components of w are (1, 0, 0) and u actually coincides with the vector i, If the coordinate system is now rotated, the new components of u are given by uy = aul + 420 + 4130 = air, ail + 4200 + aoa = aor, & El uh = agi! + 4320 + 4330 = aor. We sec that the first column vector of matrix A is composed of the new components of vector u., In other words, we can say that the new components of i are (G11, 421, 431) and we can write f= ayyi' + aos’ + aaik’. Similar statements relate j and k to the second and third column vectors of A. Note that in this discussion the unit vectors J, j,k assume a role independent of their respective coordinate axes. The axes are rotated but the vectors i, j,k stay in place and are then referred to the rotated system of axes. ' * More precisely, a 3. 3 matrix is formed. The reader can easily construct-an analogous 2% 2 matrix to account for two-dimensional rotations. 13 (CHANGES OF AXES. ROTATION MATRICES 7 Exercise. Establish the geometrical meaning of the row vectors of matrix A representing. a rotation. The definitions introduced above allow the computation of (ui, #2, 5) from (4, 412, us) by the following cule: To obtain uf, take the dot product of the kth row of matrix A with the vector u, as given by the triple (1, 2, us). Since in this process cach row of the matrix is “dotted” with (u1, w2, 3), We may regard it as some kind of multiplication of a vector by a matrix. In fact, this operation is commonly known as vector-matrix multiplication and is visually exhibited as shown: a1 412 13 my cH G21 pz G25 I ta = oe) 931 a2 Ang uy ty Matrix A Column vector uw — Column vector u’ As we see, the old components are arranged in a cohumn which we shall denote by u. This column vector is multiplied by the matrix A and this results in another column vector, denoted by a’. The multiplication means, of course: Form the dot product of the first row of A with u for the first component of u'; then form the dot product of the second row of A with u to get u’, and similarly for us. The entite procedure is symbolically written as Au = wf. Remark, Note that the set (1, 2, #3), arranged ina column, has xof been denoted simply by u but rather by a new symbol «.* The point is that in the context of our problem, both w and w’ represent the same vector u, but with respect to different systems of axes, We must think of uw and u’ as two different representations of u, and the matrix A shows us how to switch from one representation to another, Before we discuss further topics involving matrices, let us record the fact that our matrix A is not just a collection of nine arbitrary scalars. Its matrix ele- ments are interdependent and possess the following properties. a) The columns of A are orthogonal to each other, namely, 11012 + A21a2 + agiaa2 = 0, Fy 2019 + Ay2023 + ay2M95 = 0, 413411 + A23da4 + 439031 = 0. This property follows from the fact thal the columns of 4 are representations (in. the new system) of the vectors i, j, and k and these vectors are mutually orthogonal. * The symbol should not be confused with ful, the magnitude of vector u, which is also denoted by # (p. 2). 8 VECTORS, MATRICES, AND COORDINATES 14 b) The columns of 4 have unit magnitude, namely, ai, + hy + ahi = 2 2 2 a2 + a22 + age a}y + 033 + a9 = I, 1, 1, because i, j, and k are unit vectors. ) The rows of A are also mutually orthogonal and have unit magnitude, This is verified by establishing the geometrical meaning of the sow vectors of A.* Matrices satisfying these three properties are called orthogonal, We may then conclude that the matrices representing rotations of orthogonal cartesian systems in space are orthogonal matrices. Remark, There are orthogonal matrices which do not represent rotations. Rotation matrices have an additional property: their determinant (i.¢., the determinant of the equations on p. 5) is equal to 4-1. Otherwise, orthogonal matrices can have the deter- minant equal to —1. The point is that a rotation must yield a right-handed triple (Wf, 4) of unit vectors since (i, j,k) is a right-handed triple.t 1.4 REPEATED ROTATIONS. MATRIX MULTIPLICATION ‘The matrix notation introduced in the preceding sections is particularly useful when we are faced with repeated changes of coordinate axes. In addition to the primed and unprimed systems related by a matrix A, let there be a third double- primed system of axes (x”, y", 2!) and let it be related to the primed system by a matrix B: bit baa bis B=| ba boa bes bs: baz bas Evidently, the system (x”, y’”, 2") can be related directly to the system (x, y, z) through some matrix C and our task is to evaluate the matrix elements Cpa in terms of matrix elements dan and ban. We have ‘ tf = Brite + Brats + Oras, bast + Booty + bests, Bayh + buat + Bsatlss ug f PA and , w= ayy + yatta + aisua, Gays + Gyaite + oats, i ty = Agatty + Asatte + dgalta, ub * Htwill be shown in Chapter 10 that any NX N matrix which satisfies (a) and (bo) must also satisfy (¢). +See Problem 4 at the end of this chapter. 14 REPEATED ROTATIONS, MATRIX MULTIPLICATION 9 from which it follows that = Gra + byzan + brsasiuy + Grsare + br2a2a + bisag2)ea + Guiaig + Byzaa3 + Biaaaz)us, . 1 = (boiay1 + Drager + bastas dus + (berda2 + baeGea + basago}te + (621013 + be2deg + Bogays)ua, Wy = (bards, + baade, + basagy)ur + (barare + byzdez + bazaya)iea + (631413 + bs2az3 + B33a53)e5- The maze of matrix elements above becomes quite manageable if we observe that every coelficient associated with », is a dot product of some row of matrix B and some column of matrix A. A closer look at these relationships leads us to the following statement: The element can of matrix C is obtained by taking the dot product of the mth row of matrix B with the nth cohann of matrix A. Now, if we record our relationship in the vector-matrix symbolic notation ws Au = Bt, = CH, then we are naturally led to the relation uw" = Cu = B(Au). It seems reasonable now to define the product of two matrices, like B and A, to be equal to a third matrix, say C, so that the above relationship may also i * be written as’ a? = Cu = (BA. In this sense we write Cin Ciz Cis bit bia das Min 42 413 21 C22 Cas |= | bor dae bas |-| Ger Gon aon 31 Caz C33 531 baa baa 431 39 aan or, symbolically, CaBA given that the matrix elements of C are defined by the rule quoted above. Having introduced the notion of matrix multiplication, we are naturally interested in determining whether it has the same properties as the mulliplication of ordinary numbers (scalars). A simple check shows that the associative Jaw holds: If we multiply three matrices 4, B, and C in that order, then this can be done in two ways: ABC = (ABYC = A(BC) {where it is understood that the operation in parentheses is performed first). * The difference is, of course, that in B(Ax) the column vector w is first multiplied by A, producing another column vector which is, in turn, multiplied by B. In (BA)« the matrices are being multiplied first, resulting in a new matrix which acts on x. 10 VECTORS, MATRICES, AND COORDINATES 14 Exercise. Verify this statement. [Hint: If AB = D, then the elements of D are given by 2 dren = Dy Amid in: fi Develop the matrix elements of (AB)C and A(BC) in this fashion and verify that they are the same.] On the other hand, matrix multiplication is not commutative: AB ¥ BA, and, furthermore, there is no simple relation, in general, between AB and BA. This noncommutativity feature precludes the possibility of defining “matrix division.”* However, it is possible to talk about the inverse of a matrix and this concept arises naturally in our discussion of rotations. Indeed, if we rotate our orthogonal cartesian system of axes, the new coordinates of vector u are obtained from the vector-matrix equation w= An. Suppose now that we rotate the axes back to their original position. The new components of vector v are given by (1, ta, #g) and the old ones are given by (u4, ub, 9); these components must be related by some matrix B: u= Bu, Combining these relations we obtain u= B(Au) = (BAU. Therefore, the matrix BA must transform the components (a1, ta, ug) into them- selves, It is easy to see that this task is accomplished by the so-called unit matrix (“identity matrix”) Exercise. Show that if 2 = (BA)u is to bold for an arbitrary vector u, then BA must necessarily be of the above form. In other words, the identity matrix is unique. It is now customary to call B the inverse of matrix A and to denote it by symbol AW" 50 that we have A714 = F. Since we could have performed our rotations in reverse order, it is not hard to see that AB = J as well, that is, AT1A =) AAm* and A = B~'. While two rotation matrices may not commute, a rotation matrix always commutes with its inverse. * If we write 4/B = X the question would arise whether we imply A = BXor A = XB. + See Section 10.4 for a general statement to that effect. 15 SKEW CARTESIAN SYSTEMS. MATRICES IN GENERAL i It may now be of interest to relate the elements of matrix B to those of matrix A. To- obtain bn», we should, in principle, solve the equations on p. 5 for tty, tz, ua. However, in the case of roiations, we have a much simpler method at our disposal. Let us write the matrix equation BA = J in detail: Oy, biz Ora iy G12 G13 loo ber bea bas f-| 21 azz a2g |=] O 1 0 Bs, baz bay 431 G32 a3 9 To get the first row of J we must take the dot products of the first row vector of B with each of the column vectors of A. However, we know that the latter are just the vectors I, j, k in new representation. We sec that the first row vector of B is orthogonal to J and k and its dot product with iis unity, Consequently, it could be nothing else but the vector i (in new.representation, of course), and we conclude that by) = 414, O12 = gy, and by3 = ag). Repeat this argument for the other rows of # and deduce that the rows of B are the columns of A and vice versa. This is also expressed by the formula Bn = Gam Any two matrices A and B satisfying these conditions are called transposes of each other and are denoted by B = AT and A = BT, While it is not, in general, true that the inverse and transpose of a matrix are identical, this rule holds for rotation matrices and is very useful. 1.5 SKEW CARTESIAN SYSTEMS. MATRICES IN GENERAL If the coordinate axes in a cartesian system form angles other than 90°, we have a skew cartesian system, Figure 1.8 shows two such systems, one in plane and one in space, along with the decomposition of a vector into its respective components, Figure 1.8 x 12 VECTORS, MATRICES, AND COORDINATES 15 Skew systems are specified by the angles between the axes (one in plane, three in space) which may vary between 0° and 180°, As before, &, §, and k will be used to denote unit vectors in the direction of axes. Note that we can still talk about right-handed systems in space, where the vectors i, j, k (in that order!) form a right-handed triple. The vectors are added and multiplied by scalars according to the same rules that were stated before. However, the length of a vector is now given by a different formula. For instance, for a plane vector we have from Fig. 1.8(a), by the cosine theorem, l hat” so that fot = Wid + ud + 2uguy, cos 6. In general, the dot product is no longer given by the sum of the products of com- ponents, but by a more complicated formula. As a matter of fact we shall even introduce a new name for it and call it the inner product of two vectors which is then defined by @y= The reason is that we would like (to retain the name dot product to mean the sum of the products of components of two vectors, regardless of whether the axes are orthogonal or skew.* * "The derivation of a formula for inner product in a skew system is greatly facilitated by its distributive property, namely, @-@ +) = @- y+ GW). Indeed, from Fig. 1.9, no matter which coordinate system is used, we see that (u-@ +) However, MN - MP + PN; since MP is the projection of v on u, we have ful - (4P) = fu} - vf cos (u, ¥), and similarly for PN, establishing the result. Note that this argument is also valid for vectors in space. Now we can writet for two vectors in a plane u = ui -+ aj and v= vad + vy; (uv) = (tak Bal) (ead Od) + Gayd vat) + Crd Dad = Udy + Upby COP $ Myre COS H+ Uy. = u2 + ub — Duigiy cos (e — §) = ub + uy + 2uzH, cose + fy] - cos (a, ¥). - GAN). : [fv +] cos @ = Note that this formula reduces to the usual dot product when ¢ = 90°. | * With this distinction, the dot product becomes an algebraic concept (referring to the ' components) while inner product is a geometrical concept, independent of the coordinate system. The two are identical provided the system is orthogonal. FSince (a-¥) = (vw), the second distributive law ((u + +9) = GW) +O #) is uriviat. LS SKEW CARTESIAN SYSTEMS. MATRICES IN GENERAL 13 There is no difficulty now in establishing other formulas for skew systems, in plane or in space. We shall not go into these details but rather consider another important question: the transformation of coordinates from an orthogonal to a skew system of axes. Consider, for instance, Fig. 1.10; here a skew x’y’-system with unit vectors i and j’ is superimposed on an orthogonal xp-system with unit vectors i and j. A given vector u can be represented either as u — ui + u,j or as Figure 1.9 Figure 1.10 Although i’ = 1, the components wu, and #, are not equal; rather, we have = OF = 00 - FB up — My tan y. Also, u, = OF = ny sec. We see that the new components (uz, ui) are linearly related to the old components (t,, a) and this relationship can be represented by means of vector-matrix multiplication: cA 1 -tany te This can be written symbolically as Ww = Aa, where w’ stands for the column vector (ui, ui), and u stands for the column vector (Uz, ty). The obvious difference from the previous cases is that the matrix A is no longer orthogonal. Its inverse A~? is readily calculated by solving for t, ty in terms of i, uf, and it reads 1 siny 0 cosy Note that it is no longer the transpose of A. 14 VECTORS, MATRICES, AND COORDINATES 1.6 It is still true that the columns of A are the old unit vectors in new representa- tion (Fig. 1.11), that is, i Li +0-7, j= —tany-¥ + secr-j’ The rows of matrix A do not have a simple geometrical interpretation, but the columns of matrix A—? do have one.* From the above analysis we may conjecture that, in general, a change from one set of unit vectors to another in a plane or in space involves a linear relationship between the old and new components of a vector, expressible by a vector-matrix multiplication uw = Au (Aisa 2 X 2 or a3 X 3 matrix). We shall consider this problem in detail in Chapter 10. For the time being we shall mention the fact that not every matrix A can represent such a relationship. Consider, for instance, the following hypothetical relation between the new and old coordinates of some vector u: ue = du, — Quy, Wy = Dee — ty. If we attempt to solve these two equations for u, and , we find that they have #o solution if ul and uy, are arbitrary. We say that the matrix 4-2 2-1 Az does not possess an inverse; such matrices are called singular matrices, It is not difficult to see that Figure 1.11 in this example the pair uZ, u, cannot possibly repre- sent an arbitrary vector in plane: Our equations imply u; = 2m), so that there is only one independent component instead of the two required for a plane. Remark. It is of interest to note that if = 2uj is actually satisfied, our system of equations has an infizity of solutions for uz and ny (since two equations reduce to a singte one), Furthermore, if an additional requirement 2, — 2u, is imposed, the system has a unique solution (it — 4e3 ty — 4ul). All these features should be remembered since they are important in physical applications. 1.6 SCALAR AND VECTOR FIELDS So far we have been discussing constant vectors, but we can also contemplate vectors which depend on one or more variable parameters. Fhe simplest example ! is, perhaps, a position vector which depends on time f. In a fixed coofdinate systein, this is equivalent to saying that its components are functions of time * If A were orthogonal, the rows of A would be identical with columns of A~! (see pp. LI and 440). 1.6 . SCALAR AND VECTOR FIELDS 15 and we write w= uf) = u(Q-i+ wy) + ut) k, Such vectors can be differentiated with respect to the variable f according to the definition 4 # ate) = tim L490 — o), Ano at With u(/) and u(t + Ad) expressed in terms of their components, it is trivial to deduce that 2uy-% dig 5 +24 ding so that the operation of differentiation of a vector is reduced to differentiation of its components. While vectors depending on time are widely used in mechanics of particles, we shall be more interested in another type of variable vectors: those depending : on space coordinates (x, y, 2)-* Such vectors are said to form vector fields and can be denoted as follows: . = U(x, yy Z) = uals Ys Zi yl, Ys ZI bax, ¥, 2K. Common examples are electric and magnetic fields in space, velocity field of a fluid in motion, and others. The simplest kind of such a field is probably the so-called gradient field{ which can be derived from a single scalar function (x, », z), usually referred to as a scalar field. Familiar cases of scalar fields include the temperature distribution in a solid body, density of a nonhomogeneous medium, clectrostatic potential, etc. A scalar field gives rise to numerous other quantities through its various partial derivatives. Jn particular, let us concentrate our attention on a) the total differential ag ax Be gy 4 28 do = Sede + So dy + 3 a, and b) the directional derivativet de _ ag dx , de dy , dg dz. ds ax ds ' ay ds | az ds * These vectors may also depend on time, but we shall be mostly interested in instanta- neous relationships, where ¢ has some fixed value, t This is also called conservative field or potential field, Cee { Rate of change of ¢ per unit length in some particular direction characterized, 3yy fy fem the clement of are ds of some curve. See, e.g., Apostol, p. 104. ie 16 VECTORS, MATRICES, AND COORDINATES 1.6 ‘The expressions on the right-hand side of the equations in (a} and (b) have the appearance of a dot product. It is convenient to define the gradient of a scalar field (x, y, z) by the vector ay, -%; ae grade = Seat eit Sek Then we can write # = (grad ¢- 80), dp = (grad e- ds) and where ds = dxi+ dyj+dzk represents infinitesimal displacement in some direction and ax diy So = i+ Ex is the unit vector in the specified direction * Since every differentiable scalar field generates a gradient field, it is natural to ask whether any given vector field u = u(x, y,z) may not be the gradient of some scalar g. The answer is negative and this becomes clear as we examine the basic properties of gradient fields. In this survey we shall need certain assumptions regarding the differentiability of various functions and analytic properties of curves and surfaces involved in vector analysis. We shall mention these assump- tions as we need them. In many cases they can be relaxed and the results can be generalized, but we shall confine ourselves to the common situations encountered in physics. A curve in space is called smooth if it can be represented by x= x), ys Wh 7 = 2), where x(#), y(1), and z(t) have continuous derivatives with respect to the parameter ¢ (for a curve in a plane, simply set z = 0), Smooth curves possess tangents at all points and a (vector) line element ds can be defined at any point. The smooth- ness also guarantees the existence of line integrals} This last property is trivially extended to piecewise smooth curves; ic., those consisting of a fitite number of smooth parts. We shall assume that all curves considered by us are piecewise smooth. / Regarding the differentiability of various functions, we must remember the following definitions and statements: the interior of a sphere of arbitrary radius ¢ (usually thought to be small) centered at some point M(x, y, z) is called a neigh borhood of this point (in a plane, replace “sphere” by “circle”). Ifa set of points — : * Observe that dx/ds = cos (So, i}, etc., are the directional cosines of the ‘direction defined by ds or by S9. . + We shall assume that all integrals are Riemann integrals which are adequate for our purposes. For example, see Apostot p. 276. {A more precise term is an eneighborhood. 16 SCALAR AND VECTOR FIELDS = 17 is such that it contains some neighborhood of every one of its points, then it is called an open set. For instance, the interior of a cube is an open set; we can always draw a small sphere about each interior point which will lie entirely within the cube. However, the cube with boundary points included is no longer an open set. The reason these concepts are needed is that partial derivatives of a function in space are defined by a limiting process that is ticd to a neighborhood. We must make sure a region is an open set before we can say that /(x, y, z) is differentiable in this region. In addition, we shall be mostly interested in connected open sets, or domains. ‘These are open sets any two points of which can be connected by a polygon, ie., a curve which is formed by a finite number of connected straight-line segments. From now on we shall assume that all our piecewise smooth curves lie in domains where the functions under consideration (scalar fields and components of vector fields) possess continuous first-order partial derivatives. Let us now return to the properties of gradient fields. Suppose that u = grad (x, y, z) and consider the following integral between points Mx, YonZ0) and —-N(x4, ¥15 Zi) taken along some curve C: L (u-ds) = L@ e+ 3 dy + ean). Along Along ¢ Using the parameter # as the variable of integration, we have Nn t. de) = ag dx | dy dy | ay de [* de _ f (ds) -f. (e a+ 3y det az di dt dt = en) — olta), Along C where f) and #, are the values of the parameter 1 corresponding to points M and N. We see that the integral is simply the difference of values of g(x, y, z) at points Nand M and, therefore, is independent of the choice of curve C. Conversely, if the integral fr (u- ds) is independent of path,* then keeping Mf fixed and treating W as a variable point, ' we can define a function oles 2) = FP Gu ds) mf" Gan do + ty dy + 4 * From now on, we shall occasionally use the term “path” to indicate a piecewise smooth curve, 18 VECTORS, MATRICES, AND COORDINATES 1.6 Path C, Mt 2 N Path Cy Path C Path ue ‘ N (a) ) Figure 1.12 It is now simple to show that grad ¢ = u. For instance, + Azw.2) oe +88, 7,2) — 968952) = (OOP Gas de + ay dy + oe de) Key) 24 Bey. a fT dy Htae2) and the statement uz — dy/dz follows from the fundamental theorem of integral calculus. ‘We have then established the following theorem: The necessary and sufficient condition that u = grad ¢ is the independence of path of the integral { (u- ds). An alternative way of stating this result follows [rom consideration of the integral ° § a -as) over a simple closed path C, called the circulation of vector u around C. By simple closed path we mean a closed path which does not intersect itself * The following theorem holds: The circulation of u vanishes for an arbitrary simple closed path C (in a domain D) if and only if the integral [%, (a - ds) is inde- pendent of path (in D). Indeed, let C (Fig. 1.122) be a simple closed path. Choose two arbitrary points M and N on C and write (weds) = f (as) +f" (ads) = f" (ads) — f” (wd), ¢ Mw Nv Mf Me Along Cy Along Cs Alone Ch Along Cz If the integral fi (u - ds) is independent of path, the right-hand side vanishes and the cireulation is zero. Conversely, if two paths C, and Cz connecting two points (Fig. 1.12b) do not intersect (in space), a simple closed path can be formed from them and the above equation holds. If the feft-hand side is zero, so is the right-hand side, yielding the independence of path. If C; and Cy intersect a finite number off times, the proof is obtained by splitting the closed path into a finite number of simple * This property permits us to assign the direction of integration around the curve, char- acterized by the vector ds, in a unique fashion. Lé SCALAR AND VECTOR FIELDS 19 closed paths. {n the rather excepti an infinite number of times, a Ii case to the preceding one.* al case when Cy and Cz cross each other ing process can be invoked reducing. this Remark, Within the context of the above discussion, it is emphasized that by a(x, y, z) we mean a well-defined single-calued functiont over the entire domain and nothing short of this requirement wilt suffice. In many treatmentst of the magnetostatic field H, one introduces the so-called scalar magnetic potential X so that H = gradX and yet the circulation of H does not vanish over some contours. However, in all such cases it is impossible to define X uniquely over the entire contour (for those contours for which it is possible, the circulation of H does indeed vanish). It should now be clear that many vector fields do not fall into the category of gradient fields since it is easy to construct a vector u for which the integral J(u+ ds) will actually depend on path. It is perhaps even easier to sketch some such fields, a task facilitated by the introduction of the concept of field lines. These field lines are curves with tangents directed along the vector field u at every point. For instance, Fig. 1.13 shows the velocity field (in the plane) of a fluid rotating around a circular obstacle. In this case the field lines are the tra- jectories along which the particles of fluid actually move. Figure 1.13 Figure 1.14 It is evident that the circulation of the velocity vector u around any one of the circles in Fig. 1.13 cannot be zero (the product u - ds has the same sign at each point of the circle). Consequently, the above field cannot be a gradient field. ‘The velocity field of a fluid is, perhaps, the best starting point for investigation of other types of vector fields because it naturally jeads us to another fundamental concept: the flux of a vector field. Consider the element dS of a surface S (Fig. 1.14). Just as in the case of curves, we shalt deal only with piecewise smooth surfaces, i.e., those consisting of * The details may be found in O. D. Kellogg, Foundations of Potential Theory. { A multivalued function is not one function, but a collection of severai different functions. For example, Reitz and Milford, Foundations of Electromagnetic Theory, Section 8.8. 20 VECTORS, MATRICES, AND COORDINATES 7 a finite number of smooth portions. By a smooth surface we mean a surface representable by x= xg. y= Weg 72 = 27,9), where p and g are independent parameters and the functions x, y, and z have con- tinuous first partials with respect to p and q in the domain under consideration. Smooth surfaces possess tangential planes at all points and can be oriented; that is, we can distinguish between the positive side and the negative side of the surface. We shalt also assume that our piecewise smooth surfaces are constructed in such a way that they are oriented.* It is customary to represent surface elements dS by vectors dS of magnitudes that are directed along the positive normal to the sur- face, as illustrated in Fig. 1.14. Suppose that the vector field u represents the velocity of a moving fluid. Jt can be secn that the inner product (u- dS) repre- sents the amount of fluid passing through dS per unit time. Indeed, the particles of fluid crossing dS at time ¢ will occupy the face ABCD of the shown parallelepiped at time ¢+ dt and all particles of fluid which have crossed dS between ¢ and t + dé wilt be located at + dt inside the parallelepiped. Consequently, the amount of fluid passing through dS in the interval dt is given by the volume of the parallelepiped, equal to 5+ |ul-dicos@ = (u: dS) dt. Divide by at and obtain the desired statement. By analogy with these observations, we define, in generat, the flux of a vector field u through a surface S by the surface integral & = ffu- ds). In this formula, S can be either an open or a closed surface. A very familiar case of the latter is found in Gauss" theorem in electrostatics. 1.7 VECTOR FIELDS IN PLANE According to the material of the preceding section, integrals representing circula- tion and flux are important in the study of vector fields. For vectors in a plane, the circulation integral has the form $, (u- ds) = $ (ie dx + tty d,). Integrals of this type can be analyzed by means of Green's theorem: If C }s a (piece- wise smooth) simple closed curve in a simply connected domain D and if Poy) * For details, consult, eg., Kaplan, p. 260 et seq. . ~ + Unless stated otherwise, it is conventional to take the direction of integration in such integrals, Le., the orientation of ds, as countercfockwise. 17 VECTOR FIELDS IN PLANE. 21 Simply connected domain Doubly connected domain Figure 115. and Q(x, ») have continous first partials in D, then fear om-{f (2% | where S is the area bounded by C. ‘The importance of the requirement that Cis a simple closed curve (see p. 18) lies in the fact that we can distinguish the interior of the curve from its exterior by the following rule: As we proceed along the curve in the direction of ds we designate the region appearing on our right as the exterior and that on our left as the interior. If the curve crosses itself such a formulation leads to contradiction as should be obvious by considering a curve in the shape of a “figure eight." A domain in a plane is said to be simply connected if every simple closed curve in it has its interior inside the domain as well. Figuratively speaking, a domain is simply connected if it has no “holes” (Fig. 1.15). Without going into mathematical details we shall sketch a possible method of proving Green’s theorem which greatly facilitates its physical interpretations. First of all, note that P(x, y) and Q(x, y} can always be treated as the components ux, y) and u(x, y) of some vector field, and we shall adopt, for convenience, this identification. Let us now divide the area S into a network of meshes, as illustrated in Fig, 1.16(a). Taking the integrais f(u- ds) around each mesh in Figure 1.16 22 VECTORS, MATRICES, AND COORDINATES 17 counterclockwise direction we can easily deduce that §, Ga) = Ef way. melon (The contribution from a common boundary between two meshes cancels out because of opposite orientations of vectors ds for each mesh; this leaves only the contributions from the pieces of C) Furthermore, multiplying and dividing each term in the sum by the area AS of each mesh, we obtain fo- ds) = rfey 8) as. AW anonties Suppose now that the number of meshes is increased to infinity and that each mesh “shrinks to a point” so that AS ~> 0, If the limit lira £6 08 i as—o = {%y) exists and is independent of the shape of AS,* then the sum reduces to an integral and we have ¢, {u- ds) = fff yds. s Therefore, it remains for us to cvaluate the function f(x, y). A typical mesh is shown in Fig, 1.16(b); it need not be rectangular since the arguments presented below are valid for an arbitrary shape. If #, and uy have continuous partials, then we can write ° (udu & Gade + Me) (-ot (2), o~ con = ade + (G8) 0+ (F)0-» with the approximations being within the first order in [x — é| and |y — n)-f Here P(t, 9) is the fixed point to which AS ultimately shrinks and M(x, y) is an arbitrary point on the boundary of the mesh, Writing now f ude) = fudx t+ $y dy, : , : + Except that the largest diameter of AS must approach zero; the mesh should not become infinitesimally thia while retaining finite length. + By the theorem on existence of tolal differential, guaranteed by the continuity of partial derivatives. 7 VECTOR FIELDS IN PLANE = 23 we see that the following six integrals will be needed: fax, ¢ dy, f xdy, f va, ¢ x dx, f vay. The first two are zero, the second two are +-AS and —AS, respectively, and the last two are zero. As a result, we have me (Olly atte ¢ (ude (4H — Se) as. flo, ¥) = Kien LEG) _ Sy _ Bee, a il I Consequently, AS Ox ay where the stipulation that the partials are to be calculated at P(é, 9) can be omitted since P is now an arbitrary point within C. We have then the result Z (te de + uy dy) = f ff (tt 2) as, which is simply Green’s theorem in our notation. With regard to a vector field u = uci + uyj, the function f(x, y) is known as the cur? of u so that, by definition, £0) #. curl b= lim We have then evaluated the expression for curl u in (orthogonal) cartesian co- ordinates in plane: Ouy = du, curl uw = ‘ox oy Remark, Attention is drawn to the fact that, fora vector field in a plane, curl u is essen- tially a scalar* and not a vector. The point is that by vectors we must mean quantities expressible as ai + bj and curl is definitely not this type, whether or not we introduce the third axis. A vector field u which has zero curl at some point is called an irrotational field (at that point). If u is irrotational in a simply connected domain, then by Green's theorem, it is a conservative field (geadient field) in this domain, ie., it has zero circulation. The converse has to be worded rather carefully: If u is a gradient field (namely, u = grad y), then it is irrotational provided ¢ has continuous second- order partial derivatives. *In a more elaborate nomenclature, curl u is called pseudoscalar due to ils peculiar property of changing sign if the x- and y-axes are interchanged. See Section 16.5. 24 = VECTORS, MATRICES, AND COORDINATES 17 Figure 1.17 Figare 1.18 Example. The magnetic induction field (B-field) due to an infinite current- carrying wire is known to be (outside the wire)* B= B25, (KSA unit). a? In the xy-plane (as shown jn Fig. 1.17}, = —Bsine= —H Ym _y, B= —Bsin@ = — FE Oe B, = Boos 9 = Hol % dn EP This field happens to be irrotational everywhere except at the origin. Therefore Fc (B- ds) = Oif C does not encircle the origin, but not otherwise. A function x(x, y) may be found such that B = grad x in a simply connected domain D, but this can be done only if the domain does not contain the origin. The B-field inside the wire is known to be = tol : B30 where Ris the radius of the wire. Here Be = —(#ol/2+R*)y and By = (stol/2xR7)x. ‘The field is not irrotational and cannot be represented as grad p anywhere. FS, Let us now turn our attention to the concept of flux for vectors in plane. The obvious analog to the three-dimensional case is the integral . a= f (ung) ds =f (u-dn) . : ’ taken over a curve C (not necessarily closed) with ny being the wif normal to the curve and dn = nods. This is illustrated in Fig. 1.18 > * go is the unit vector that is tangential to the circle drawn around the axis of the wire, Le. VECTOR FIELDS IN PLANE (25 Exercise. For a flow of fluid ina plane, relate the integral & above to the amount of fluid crossing the curve C per unit time. Specify physical units of ali the quantities used. In many applications, the flux through a closed curve is involved. Evidently if ds = dxi + dy}, then* dn = dyi — xj and f, (w+ dn) = $, (ts dy ~ tty dx). Setting P = — 1, Q = tty, this integral can be transformed by means of Green’s theorem so that fo m= [Ger ae) acy provided, of course, that the partials are continuous everywhere inside C. This relationship is usually referred to as the divergence theorem (in a plane) and is written as, f,, (ada) aes au, diva = ive 4 a where is another function derivable from a vector field and is known as the divergence of u. While the above derivation is straightforward, it docs not reveal the geometric (or physical) meaning of divu. It is instructive to invoke the technique used in Green's theorem: Dividing the area S into a network of meshes, we find that it is not hard to establish that oan - DLL Mas, Alt meshes because the vectors du at the common boundary of two adjacent meshes are op- positely directed. This observation permits us to define the divergence of u as flux out of an infinitesimal area (per unit area), namely, cen = him Seca) , divu = fim, AS Exercise. Derive the formula div u = du2/dx + du,/dy starting from the above defini- tion and using the arguments analogous to those for curlu. Spell out the conditions required in the derivation. * It is a standard convention that for closed curves in a plane (and closed surfaces in space) the positive normal is the outward-pointing normal. 26 VECTORS, MATRICES, AND COORDINATES 18 Remark. From our definitions of curl # and div u itis seen that both represent a new kind of derivative, namely, a derivative with respect to infinitesimal area rather than infinitesi- mal displacement: curly = in SE, diva = i, Feu), It may be of interest to mention that the gradient of a scalar field y can be represented in a similar fashion; i.c., the following statement holds: ‘eo dn as grad yp = lim £ 450 Another interesting observation is that cur! « can be related to an integral involving da and div a to an integrat involving ds. Indeed, the identities (u> dn) = [u X ds], (a: ds) = [dn X ul, are not hard to verily provided we treat the cross product of two veciors in plane as a scalar, which is the logical thing to do.* Consequently, the following statements also hold: Fla x 4) = fldn x cutu = fim RZ, AS divu = lim aged asa Vector fields which have zero divergence are called solenoidal fields. They are very common in physics. For instance, the electrostatic field is solenoidal in the absence of charged matter; the magnetic induction field is solencidal everywhere. 1.8 VECTOR FIELDS IN SPACE We would now like to extend the analysis of the last section to vectors in space. We shall start with the flux of a vector field a through a closed surface S, becaiase this integral is, perhaps, easiest to handle. It reads a= ffoas) = ff weds. + uy dS, + te dS), where dS,, dS,, and dS, are projections of vector dS on the coordinate axes. Integrdls of this type can be handled by Gauss’ theorem: If S is a (piecewise smooth) closed orientable surface contained, along with its interior, in @ domain D and if L(x, y, z), M(x, y, 2), and N(x, y, 2) have continuous first partials in D, then OL , aM , aN $f as + Mads, + NaS) “fff (+4424) ”, ! where V is the volume bounded by S. * A typical cross product of two vectors in physics is the torque I" = [r X F] which, for vectors in a plane, is completely described by magnitude and sign (clockwise or counter- clockwise). See the remark on p. 23 which may lead to a conjecture that the cross product in a plane is a pseudoscalar (and, indeed, it is) 18 VECTOR FIELDS IN SPACE 27 Figure 1,19 Gauss’ theorem can be restated in vector notation, by identifying ZL, M, and N with the components of a vector field u = wei + 1,j + uk. The concept of divergence of u, au, diva = Fe 4 Sey ou, is teadily introduced so that Gauss’ theorem can be rewritten as ff 48) = fffeiva-av and is often referred to as the divergence theorem. The deduction of this theorem can be done again by a method analogous to that in the preceding section. In this case we cut the volume V into small pieces, say, rectangular blocks, one of which is shown in Fig. 1.19. If we calculate the flux of u through each block and add alll the results, we must obtain the flux through the outer boundary S; just as before, the flux of u through an interface between two blocks must appear in the sum twice, but with the opposite sign because of the changed direction of the “outward normal” (see the footnote on p. 25). Consequently, fp dS) = 5 fe® 9S) ay am blocks introducing, for an obvious purpose, the volume AV of each block. We now define the divergence of u by means of divu = tim Hen). avao Al Gauss’ theorem can now be deduced by ironing out the mathematical de- tails and developing the formula for divu in cartesian coordinates. We shall 28 VECTORS, MATRICES, AND COORDINATES 18 give a simplified version of this from the consideration of a rectangular block AV (Fig. 1.19). It is not hard to show that the flux of u through a rectangle such as ABCD, is given* within the first order in Ay and 4z by Pancp = (us) Ay Az, where (1t,)p, is calculated at the center P of the rectangle. Indeed, for any point within ABCD we have ‘Ou, dus ue ue + (88 pO D+ a7) pF and the last two terms yield zero when integrated over the area ABCD. Consequently, the flux of u through the face A‘B'C'D? is to be determined by the value of uz at the center of this face and is given by [woe + (te | Ay az. x Jp Similarly, the flux through the opposite face is - [ure - ae) 4 Ay Az. There is a minus sign in front because we now want the positive normal in the negative x-direction. Adding these two fluxes we obtain ou 3x) p Ox ae oF The fluxes through the other four faces are obtained in a similar way leading to the statement about the flux through a small paralielepiped AV: cw (Me 4 Oty ae tov = (4 + ay + “ur Av. ‘The expression for divergence now readily follows and concludes our analysis. Let us now consider the question of circulation of a vector in space. Such an integral has the form ¢ (u-ds) = (ide + ty dy + us de), a and we shall assume that C is a (piecewise smooth) simple closed curve in space. It nced not be a plane curve but we shafl assume that it can be spanned by a (piece- wise smooth) orientable surface 5; that is, C can serve as the boundary for such a * Assuming a positive normal in the positive x-direction. 18 VECTOR FIELDS IN SPACE = -29 Figure 1.20 surface, This situation js iWustrated in Fig. 1.20{a). It is a standard convention to define the positive direction of circulation around a surface element dS so that it forms the “right-hand screw system” with the positive normal to the surface* The positive direction of circulation around the boundary curve C is now chosen so that it coincides with such direction for the adjacent surface clement (Fig. 1.20a). When all this is done, we can quote Stokes’ theorem; If S is a (piece- wise smooth) orientable surface spanning a (piecewise smooth) simple closed curve C, mutually oriented as described, then $e», 2) di + M(x, y, 2) dy + M(x, y,2) 2] -[[(@-3 2) yaa + (22) crac +(e ae axa 5 provided L, M, and N have continuous partials in a domain containing S and C. As before, we can treat L, M, and N as the components of a vector field u. By the now familiar technique, we divide S into a network of (curved) meshes (Fig. 1.20b) and claim where AS is the area of a given mesh. As before, we calculate Eads ds) Fu, dx + ty dy + uw, dz), fim FAM eT ty OY 4830 ay AS Considering a small mesh about a point P(E, 7, 7) we have, for each point Af(x, y, 2) on its boundary T (Fig. I.2ta), Ou, (dar & de + a) @- pF ( oe) po~ a+ — ete. * A rule well known from the study of magnetic fields. 30° VECTORS, MATRICES, AND COORDINATES 1s (@) Case I () Case 11 Figure 1.21 ‘We shall need integrals like ¢, dx, 4, x dx, $, ydx, ¢, zdx, and similar ones. ‘These integrals can be evaluated by projecting our mesh on the coordinate planes. Suppose thal we want to obtain the integrals of the type Sr f(x, y)dx and $rI(% y) dy. Since the points M and M’ (see Fig. 1.21) have the same x and y, these integrals reduce to $r- f(x, pdx and Sr f(x, ») dy except for the sign* which depends on the orientation of the mesh: §. Se yax = tf. SG pdx, ete. ‘The plus sign is for Case I (Fig. !.21a), where the point M’ describes I’ in counter- clockwise direction, the minus sign is for Case 1 (Fig. 1.21b), with M’ going clockwise.f In particular, note that : $d = ¢ xd = 0 pyar = gb yar FAS’, while where AS’ is the area bounded by I”, ie., the area of the projection of the mesh * The symbol fy. by itself is meant to indicate counterclockwise integration! {see the footnote on p. 20). + The motion of M’ cannot be chosen at will because it is determined by the motion of 4 in the positive direction of circulation around the mesh. 18 VECTOR FIELDS IN SPACE 3. on the xp-plane. Since the mesh is small, we have (Fig. 1.21) AS’ & AS'cos (9, k) in Case I, AS’ = —AS cos (ng, k}_ in Case Il, (where no is the unit normal to AS). In either case, vax = —AScos (to, k). Similarly, we can deduce ¢, xdy = +AS'cos (no, k) (this integral is also needed). Other integrals are evaluated in an analogous fashion. In particular, $; 2 dx and g', x dz require projection on the xz-plane, and so on. The net result of this calculation reads Ou, ai ny a Ps ficw- ayacas| (ts — | c08 (noi) + (See Ot 205 (3) ‘au, du, | + ( -&, 008 (Mo, k) Introducing the curl of a vector in space (which is now a vector, in distinction to the plane case} by means of the relation . -ds) lim £0248) asso AS = (curl ano), we have the result a ( ‘tt. duy\, . fatty aus curl = Sie el (Be i oe) * which also gives rise to the statement that circulation of u around an infinitesimal oriented area described by dS is equal to (curlu-dS). Thus Stokes’ theorem can now be-written in the compact form:* te (w- ds) = freon u- dS). As in the case of vectors in a plane, the vector fields satisfying diva = 0 are called solenoidal and those satisfying curl u = 0 are called irrotational. The concept of irrotational field is closely related to the concept of conservative field | {a = grad y) but these two fields should not be identified because of topological complications, In particular, if a field is irrotational in a domain D, if docs not’ * The shape of the area element is irrelevant; because of this we can identify dS, in dS = dS,\ + dSyj + aS.k with dy dz, ete. 32 VECTORS, MATRICES, AND COORDINATES 18 { folfow that its circulation about an arbitrary simple closed curve in Dis zero. The point is that we may not be able to invoke Stokes’ theorem because we may not be able to construct a suitable spanning surface S which would lie inside D. Example. Consider a tightly wound coif of current-carrying wire in the shape of a torus, The B-field inside the torus is irrotational; indeed, cur} B satisfies the equation* curl = pod + conto OE (MKSA units) and there is no current density J and no displace- ment current in the interior of the torus; we assume a dc-situation. However, the B-field at the center of each turn of the coil is known to be B = nol yielding a circulation along the central cirele C of the torus: §, ds) = wont - 2nR x 0. It is easy to see that any surface spanning C must necessarily extend outside the torus and cut through the windings where curl B # 0. Since the quantities divu and curlu in a plane have been defined as area derivatives (p. 26) and since div u in space has been defined as a volume derivative (p. 27), it may be conjectured that curl u in space is also reducible to a volume derivative. This is in fact true and the formula reads = im FES X48 curl t = fim PT avo where the surface integral is over the boundary enclosing the volume AV. We shall sketch the proof of this relation by considering AV in the form of a rectangular parallelepiped as shown in Fig. 1.22. The contributions to #1dS X ul from the top face involve only u,i and #,j and yield, in the usual notation, fees x me axay (oe 2 ¥) 5 w+) i. Top face where P is the center of the parallelepiped. The bottom face will have dS with the direction reversed and [fas x oy -axay (x — 2 Im n+ oe 4. Bottom inca * One of Maxwell's equations; see, e.g., Reilz and Milford, pp. 296-297. 18 VECTOR FIELDS IN SPACE = 333. Adding these, we obtain au) 5 (Nd aya az) i — (az), | y Az. Contributions from the other four faces are treated similarly, establishing the result, Remark, The gradient of a scalar field g can also be represented as. a volume derivative, namely dg = fi Sheds | grado = tim ay We shall conclude this scction by mentioning some quantities obtained from the repeated operations involving gradient, divergence, and curl. First of all observe the identity curl grad y = 0 representing the statement that a conservative field is always frrotational (pro- vided, of course, that the second-order partials of g are continuous, as mentioned on p. 23). The second operation of similar type yields the definition of the Laplace differential operator V7 ot simply the Laplacian,* Vp = div grad g, with the well-known expression in cartesian coordinates 2. Ca Vex ax? + ay? + i Both of these operations have their counterparts for plane vectors. There are also two operations which are only possible for vectors in space: div curl u and curlcurlu. Straightforward calculation in cartesian coordinates yields div curlu= 0. A more sophisticated argument of some interest is to take a closed surface Sin a domain D where curl « is defined (and its components have continuous partials). Split Sinto two parts, 5 and Sa, by a curve C, as illustrated in Fig. 1.23. By Stokes’ theorem, f, (u- ds) = ff(curta-asy) = ~ ff teort wd); 3 Ss * The symbol V? is related to the so-called “nabla” or “del” operator a a a "ae tay + Bae capable of representing gradient, divergence, and curl by means of the notations grad ¢ = Ve, diva =s (Vu), and curl a= [VX ul], which are sometimes quite handy. ed VECTORS, MATRICES, AND COORDINATES 19 the minus sign arises from the “wrong” oricntation of S; with respect to C. Consequently, (curl u-dS,) + f[(eurl u- dS2) = ff (curl a- dS) = 0 [fonts 4 foe crf for any closed surface in D. Applying the divergence theorem, we obtain, f[faivcunu-dy =o. v Then, since this is true for an arbitrary volume Vin D, it follows that div curl u = 0. Regarding the operation curl curl u, we can decive the following expression for the cartesian coordinates: curleurlu = grad divu — V’u, where the symbol Vu stands for the operation V?ud + V7uyj + Vuk. This formula, however, is not valid for other coordinate systems.* Exercise. Verily the above formula using the expressions for divergence, gradient, and curl in cartesian coordinates. 19 CURVILINEAR COORDINATES Sometimes it is more convenient to use coordinate systems other than cartesian. In general, a point in space can be described by three parameters which we will denote by [, 7,2. A well-known example is given by Spherical coordinates 7,6,¢, as shown in Fig, 1.24, along with the usual cartesian coordingtes x, y, z * Of course, it is possible to define the operation ¥?u by (grad divu — curl curly) in any coordinate system, but this is not done since V?u does not reduce to the operation “div grad” applied to components of u. 19 CURVILINEAR COORDINATES 35 which are related to 7, 6, 6 by x=rsinécos¢, y=rsinésin¢, = rosé. In general, x, y,z can be thought as being functions of 1, m, 2: x= xm, y = ylm,n), = 2hm,n). We shall assume that at least within some domain D in space, these functions have continuous derivatives and can also be solved for /, m, a: meurve t m= m0, y,2) Hx 2). MO) n as Fs 2). Observe that this implies that the Jacobian ay, y, 2) aG, m,n) oO Figure 1.25 docs not vanish." Let us now choose a particular point Mf with cartesian coordi- ’ tales (£, 4, ¢), it can also be denoted as A4(A, p, v), in terms of the coordinates iin,n. He we keep m = = const and m = y = const and change /, then we obtain a (smooth) curve passing through M which may be called the Feurve. Similarly, we can define m-curve and n-curve. T) shown in Fig. 1.25. Further- more, we can introduce unit vectors Ip, mo, and mg along the tangents to these curves (pointed in the direction of increasing /,m,1). This establishes a local system of axes. For convenience, the labels /, m,n are chosen so that (Io, to, No) form a right-handed triple. Our Iocal systems possess, in general, the following features which distinguish them from the system formed by cartesian unit vectors {, j, and k. {. The axes may not be orthogonal; moreover, the angles between the axes may change from one point to another. 2. The orientation of Io, mto, No (with respect to i,j,k) may change from one point to another, even if the angles between the axes remain the same. 3, The physical meaning of parameters J, 1,2 may not be the length, and df, dm, dn need not be identical with the elements ds of arc in the respective directions. Let us investigate properties (1), (2), and (3). We can always think of point M as being defined by a position vector + = xi-+ yj + 2k. Treating x, y, z as * See, ¢.2., Kaplan, pp. 31 ef seq. 36 © VECTORS, MATRICES, AND COORDINATES 19 functions of f, m,n we can write r= x(im, ni + (maj + 20m, rk. Changing r by dr amounts to changing x, y, z by dx, dy, dz which is, in turn, caused by changing f, m, a by df, din, da. We have the following general relations: a 4 % am 4 dx = tt ag et a = 4 Pam 4 2 dy= at ag tt + 5g = 4 Zam 4% dz= Het 5m! + an If we move along the /-curve, then da = da = 0, and dr becomes* (elon = (Be 14 dy-5+ de Wan = (Sit Bit F a. This defines the derivative of r with respect to the parameter J: ox, , ay, , a = ait ait ae By its very meaning, or/af is a vector along the direction of ly. Therefore, Io can be expressed as tp = oral _ (ax/ani_ + (ey/a)i + (z/ak__ lor/all xa? + Gy/al? + (@z/al)? The quantity ~ (2) @ f (zy = & + Ma) * Na has a simple geometric interpretation: The length of elementary arc ds produced when only / changes is given by ds = |(@t)m.nf = rdf. In a similar fashion we deduce ng = Granite (ov/am + @z/amk, ( ay ( a ry ( ay , ny = Gan + Capsan + @2/amk, 4, (#) + (zy {xy * We use the nolation familiar from thermodynamics: (dr)m,. means such dr where mr and mare kept constant. 19 CURVILINEAR COORDINATES 37, Suppose now that the triple In, mto, no is an orthogonal triple. Then we must have the relations ax ax, ay ay , az az a am * at amt at am ~ % ot These relations are satisfied for most coordinate systems employed in physics. In particular, this is true for spherical and cylindrical coordinate systems, as can be easily verified. This analysis clarifies feature (1) of local systems of axes. Regarding the orientation of local axes, note that it docs indeed vary from point to point for spherical and cytindrical coordinate systems. In fact, this is the characteristic Properly of curvilinear coordinate systems, as opposed to cartesian ones. Fealure (3) is also iNtustrated by spherical and cylindrical coordinates where some of the parameters J m,n represent angles rather than lengths, As a general rule, the elementary displacement dr decomposed along the local system of axes will read or a= Sd + © don + SE dn = hadl- Sy + bm din mtg + fg t+ Mo. Let us assume that the local system is orthogonal;* then the element of arc is given by a simple formula, ds = {del = Vibd® + am + de. For instance, for spherical coordinates, A, = 1, Ay > r, hg = resin é, and ds = Va de® Ere sin® 6de?. We shall conclude our survey of curvilinear coordinates by the derivation of formulas for common differential operations in vector calculus. In order to express grad ¢ in terms of new axes and new variables, we could start with grad p = then use de _ ap a, ap am ax = af ax t am ax t and also express |, j,k in terms of Ig, mo, Mo. A quicker way is to ulilize the statement =de= a4 oe (grado dt) = do = SF dl + SE dim + 2 dn * The associated coordinates are then called orthogonal coordinates, 38 VECTORS, MATRICES, AND COORDINATES 19 {b) Figure 1.26 and rewsite it in the form (grad pdr) = (i 9) heal + be SE) hed + (t 36h, dn from which it follows immediately* that 1 a L a L a grad ¢ = 5G Selo + 5 5m Mot 5, 5a The calculation of divergence can be also carried out starting from the general definition diva = tim HU), avso AV Without loss of generality, AV can be taken as a volume element with the sides along the /, m-, and n-curves (Fig. 1.26a). In general, the flux through an ele- mentary area oriented in the Eo-direction is given by t+ hin din - fg dit. As we subtract the fluxes through the areas M'N’P'Q’ and MNPQ we must hot forget that not only 1, but also Ay, and Ay, are functions of f,m, and xf By an argument similar to that on pp. 27-28 (we give here a simplified version), we deduce that the net outward flux through these two faces is 3 (ahha) al dem da. Adding the analogous contributions from the other four faces and dividing by the volume of our volume clement (which is fd hy dra hy dn), we oblain immediately . diva = EL? (a shatn) + 2 (Unlinks) + 2 (ugha J}. Hy lat ™ one” ane : * Recall that de is arbitrary; therefore, by setting dm = dn = 0 we obtain (grad ei: = (/A)Ge/2N, ete. We tacitly assume that the coordinates are orthogonal. } In-cartesian coordinates fi = fn — fn = 1 19 CURVILINEAR COORDINATES 39 Example, In spherical coordinates we identify J, m,n with r, 8, ¢, in that order. Then hy = hy = 1, lim = lig = 7, hy = hy = sin @, and 1 fa pa. a. a div w= asia g ae 17 sin Bu) + sp (rsin @ue) + 3 tus} This may be simplified and, if desirable, ultimately reduced to 1 au rt divu= a yt 244 cot 6 Lang | tut ine op r The curl of w can be deduced from the circulation of u around the facts of the very same volume element. For instance, the face MNPQ yields (Fig. 1.26b) P Ae (8A) +f (de) = Cn der — (nha ddnra = 32 Cun) dd, Pwd f way which is combined to form —(tadin EPG + Clmbim dn)aiw 1 2 ~ £m) din dt a @ (curl w)t fin dit hy dn = (2, Gruln) — 55 Cental} dm dn. This determines the Lcomponent of curl w, The complete formula reads a a cula = 7 (2 (nln) — 2 unbal lo + 5K | Rem — Scuba} mo +7 [Btu 1 2 ho} no. Finally, the expression for the Laplacian V2 is obtained by combining the formulas for gradient and divergence: Rnlin Oy fir ap 8 (htky, ag }. = div grad u = ila {5 ("ae th i) 4 3, (tt ie) + 2( Tin 2) For instance, in the spherical system this reads (after trivial simplification) ve = BEAM) + tng & (sin 0% +e, P ar F sin 6 36 FF sin? @ ag? 40 = VECTORS, MATRICES, AND COORDINATES BIBLIOGRAPHY Apostot, T. M., Mathematical Analysis, Reading, Mass.: Addison-Wesley Publishing Co., 1957. Gowpsrein, H., Classical Mechanics, Reading, Mass.: Addison-Wesley Publishing Co., 1959, Kaptan, W., Advanced Calculus, Reading, Mass.: Addison-Wesley Publishing Co., 1957. Kettoce, O. D., Foundation of Potential Theory, Berlin: Springer Verlag Obg., 1929. Tavtor, A. E., Advanced Calculus, Boston: Ginn & Company, 1955. Retrz, J. R. and E. J. Mitrorp, Foundations of Electromagnetic Theory, Reading, Mass. : Addison-Wesley Publishing Co., 1960. PROBLEMS 1. Let two vectors in a plane, uw; and uz, be defined by the polar coordinates of their tips: @1, 71) and @2,r2). If ug = ur + uz is defined by (8a, ¢3), show how @3 and 3 are related to 0), 02, 1, and rz. . A triple vector product of three vectors is defined by the expression [u X [¥ X w]}- Show that for any three vectors the following identity holds: [u x Wy x wh) + Uy X Ow ul] + fH x fo X I = 0. » Hint: Use the vector identity {a x [b x e]] = bla -c) — ea -b). ‘The above formula, known as the Jacobi identity, appears in a variety of contexts in physics and mathematics. 3. Consider the following three vectors in space given by their coordinates uf bP, VEEP w-h-BD a) Verify that these vectors are unit vectors, orthogonal to each other, and form a right-handed triple, if ordered as above. b) Construct the rotation matrix transforming the old components of a vector (namiely those with respect to 4, j,k) 10 the new ones (with respect to u, ¥, #7). ©) Evaluate, by vector-matrix multiplication, the new coordinates of the vectors 2(0, 3, 2), b(—L, 4, —3), and (2, —2, —2). Can you give a geometrical interpre- tation of the peculiar behavior of vector ¢? 4, 8) Show that the triple product (p. 3) of vectors ule, w2, ay), v(t1, 42, 23), and we(w1, 2, ws) can be expressed by the determinant = (ux vw). ~ a ™ PROBLEMS 4. b) Using this, prove that if a 3 x 3 matrix is orthogonal, then its determinant can have only two values, either +1 or — 1. ©) Consider the matrices 1 0 0 va $ 0 A=| 0 1 oO], B= $0 -v3;2 o |, o 0-1 0 0 “1 -1 0 6 2wiI/s —3Vi0 4 c=/ 0-1 of], p= : -p va |, oo 1 g 4 0 and indicate which ones represent rotations. meaning of others, Compare, in general, the ith matrix element of AA with that of BA, for 3 X 3 matrices A and B. Construct two noncommuting 3 X 3 matrices of your choice, i.e., such that AB ¥ BA. According to the discussion on p. 4, the matrix Also, describe the geometrical cos@ sin® —siné cosé@ represents a rotation of axes in planc. Show that sin 26 cos 28 cos 34 sin 38 A? = AA= and A® = AAA = —sin 28 cos 26 —sin 30 cos 3 and give the geometrical interpretation of this result, Show that the matrix does not represent a rotation of axes, Give a geometrical interpretation of matrix B. [Hint: Draw the old and the new coordinate axes, as well as the straight line y = xtan (§/2).] A2 ‘VECTORS, MATRICES, AND COORDINATES 8. Find the inverses of the following matrices by solving the equations B.A, = 7 or otherwise: 2-10 4 3 a AL= aouidl, #& 0 -UWVE v2 -3 04 wis -Vi6 ~V2/6 -3 -§4F 1 3 2 As=| -} @ 4 |, 4e= 1 ~1 o |, a 48 2 o Comment on the cases Az, Aa, and Ay. y 9. Let (x', »”) be the coordinates of a point in a skew y cartesian system in plane. Let the x'- and y’-axes make angles a and 6, respectively, with the x-axis (Fig. 1.27). Show that the equation of a circle with radius R and the center at the origin reads x? 4 y® + 2x’y' cos (8 — a) = RB? x 10, Show that the vector v = 21+ § — 6k cannot be G expressed as a linear combination of the vectors * mp =it+j+ 2k, w= 3-5, ws = Atk Figure 1.27 Show that the vector w = —2j— 3k can be expressed in this fashion, in more than one way. Give the algebraic explanation of these facts. Also give a geometrical interpretation. 1f, Evaluate the following integrals around the circle x? + y? = 1; use Green’s theorem if it is convenient: a) S(u- ds), where u = (Qy? — 3x")i + xy — 29), by) $ (2x? — y3) dx + (x? + yay. c) $(r-dn), where ¥ = (x? + y?)i — 2xyj (dn is defined on p. 24). 12, Let F(x, y) — x? — y?, Evaluate a) §8B (grad F + ds) atong the curve y = x?, b $ Fas around the circle x? + y? = 1. Here @F/dn is the directional deriva- 2 five of F along the outer normal and ds = |ds|. 13. Show that the vector field u = yzi + zxj + xyk is both irrotational and solenoidal. Find g such that grad y = u. Can you find a vector field A such that curl A = u? 14, Prove the following identities for scalar fields f, ¢ and vector fields u, v in space: a) grad (fe) = ferad y + e grad f, , ° yb) curl (fa) = fcurlu + [ered f X ul, ©} div fu X y] = (v- curla) — (a curly). 15. PROBLEMS = 43 Using divergence and Stokes’ theorems, if it is convenient, calculate the following integrals, a) $s (a- dS), where u the origin, : b) sv dS), where v = x‘ + y5} + 25k and S is the sphere as in (@), O Hs (edydz + ydz dx + z dx dy), where S is the sphere as in (a), d) $v (u- ds), where w= —3yh + 3xj + k and Dis the circte x? 4 y? — 1 lying in the plane z = 2. + ¥3j + 2% and S$ is the sphere of radius R about . A flat disk rotates about the axis norma! to its plane and passing through its center. Show that the velocity vector y of any point on the disk satisfies the equation curly = 2a, where w is the angular velocity vector. . Consider a conducting medium with variable charge density p(x, y, z) and variable current density J(x, y, 2). Let V be an arbitrary volume within the medium bounded by a (piecewise smooth) closed surface 5. Considering the total amount of charge inside V and the amount entering it per unit time through the surface S, deduce that 4 f[focsnar = - Ff o0sn2-49) ; 5 With the help of the divergence theorem, deduce the so-called equation of continuity diva + Bao, |. Using the techniques employed in Sections 1.7 and 1.8, outline the possible proofs of the following statements: lim avo AV rad y = lim, Fed" a plane), grad p = (in space). ase AS . Evaluate the quantities f,, Ay, As (See p. 36) for the cylindrical coordinate system. Using appropriate formutas from Section 1.9, write the expressions for gradient, divergence, curl, and the Laplacian ia cylindrical coordinates. CHAPTER 2 FUNCTIONS OF A COMPLEX VARIABLE 2.4 COMPLEX NUMBERS In the course of study of roots of algebraic equations and in particular the cubic equation, it has been found convenient to introduce the concept of a number whose square is equal to —1, By a well-established tradition, this number is denoted by #, and we write i? = —Iand i = /—T. If we allow ito be multiplied by real numbers, we obtain the so-called émaginary numbers* of the form bi (where bis reat), If the usual rules of multiplication are extended to imaginary numbers, then we must conclude that the products of imaginary numbers are real numbers; moreover, their squares are negative real numbers. For instance, GBA 4) = 3-4? = (-12(-D = 2, (~5i)? = (—5)*? = —25. If imaginary numbers are adjoined to real numbers, we have a system within which we can perform multiplication and division (except by zero, of course). We say that such a system is closed under multiplication and division, However, ‘our system is not closed under addition and subtraction.f To eliminate this de- ficiency, so-called complex numbers are introduced. These are numbers which are most often written in the form a+bi (a,b = real numbers) and are assumed to obcy appropriate algebraic rules. As will be shown below, the system of complex numbers is closed under addition, subtraction, multi plica- tion, and division plus the “extraction of roots” operation. In short, it has all the desirable algebraic characteristics and represents an extension of the real number system. The study of complex numbers is invaluable for every physicist because the description of physical laws is much more complicated without them. * Imaginary numbers are also called pure imaginary numbers to stress the distinction from the more general case of complex numbers. The name originated from the belief that imaginary numbers, as well as complex numbers, do not represent directly observable quantities in nature. While this point of view is now mostly abandoned, the original nomenclature still exists. + The system is not closed under the operation of extraction of the square root either; for example, V/ is neither real nor (pure) imaginary. 44 22 BASIC ALGEBRA AND GEOMETRY OF COMPLEX NUMBERS 45 2.2 BASIC ALGEBRA AND GEOMETRY OF COMPLEX NUMBERS Ifcomplex numbers are written in the usual form a ++ ih (or a + bi) then the usual algebraic operations with them are defined as follows. 1, Addition: (ay + iby) + (@2 + ib2) = (ai + a2) + i, + 2). 2. Multiplication: (ay + ib\)+ G2 + tbe) = (ayay — bibo) + iaiby + aob1) The second rule is easy to follow if we recognize that the expressions a + ib are multiplied i in the same manner as binomials, using the distributive and associ laws, and i? is replaced by — 1. Complex numbers of the form a + i0 are tacitly identified with real numbers since they obey the same algebraic rules and are generally indistinguishable from cach other.* Complex numbers of the form 0 -+ ib are thea (pure) imaginary numbers, It is customary to write simply a + #0 = a and 0+ ib = ib. Sub- traction of complex numbers can be defined as fnverse addition so that if (a1 + 181) — (@2 + iby) = x + iy, then ay + iby = (x + iy) + (a2 t ibe) from which it follows thatt x=a,—4, and y= b,— by. An alternative is to form the negative of a complex number, —(a + ib) = (~ 1a + 6) = (—1 + OMG + ih) = ~a— ib, and reduce the subtraction to addition. The rule for division can be similarly deduced by inverting the multiplication. A shortcut method is given by the following technique: atit _ (ac + bd) + chid ct ead? be ~ ad) ac+ bd, be —ad ere ape (c? + d? = 0). Itis readily seen that the divisor can be any complex number except zero (eamely the number 0 + 10, which is unique and is written simply 0). * Ina more rigorous language, “the subset of complex numbers of the form a + iO is isomorphic to the set of reat numbers under the correspondence a + i0 © a.” t ILis tacitly postulated that x1 + ayy = x2 + pa if'and only ix) = x2 and yi = ya. 46 FUNCTIONS OF A COMPLEX VARIABLE 22 Remarks 1. The addition of complex numbers obeys the same rule as the addition of vectors in plane, provided a and 4 are identified.with components of a vector. Note, however, that the multiplication of complex nunibers differs from the formation of dot and cross products of vectors. 2. The use of the symbol i and the related binomial @ + ib is conventional, but not indispensable. It is possible to define a complex number as a pair of real numbers, {@,A), obeying certain peculiar rules, e,, the multiplication.can be defined by (@1, bia, b2) = (@1a2 — bib2, arba + arab), and so on.; It should be clear that the form a + i is just # representation of a complex umber, _It is customary to represent complex numbers by points in the so-called com- plex plane, or Argand diagram (Fig. 2.1). If we denote the complex number x + iy by, 2 single symbol 2 and write z = x + iy, then to each z there cor- responds 4 point in the comptex plane with the abscissa x and the ordinate y. This idea also leads us to the irigonometric representation of a complex number: Z= 7 (cos@ + isin 8), 4 Beraxtiy where A= /x2-F yp? and tan@ = y/x. In ia this representation r is unique (positive square = root) but @ is not. A common convention isto | demand thatt = Re a —w<@Sn, Real axis along with the standard rule of quadrants, Figure 2.1 namely, @ < Oif yp < 0. The foflowing nomenclature and notation will be widely used: If z= x4 iy = (cos + isin @ then is the real part of z, is the imaginary part of z, is the modulus of z, also known as the magnitude or absolute value of z, @ is the argument of z, also called the polar angle or phase.t ‘The number x — ip is called the complex conjugate of the number z = x + iy and vice versa. We shall denote it by 2*. We can say that z and z* reptesent (on the complex plane) the reflections of each other with respect to the real axis. + Another commonly used convention is 0 < @ < 2n. $A more precise name for 8 would be the “principat value of the argument of 2” (see p. 57). See | | 22 BASIC ALGEBRA AND GEOMETRY OF COMPLEX NUMBERS = 47 Remarks 1. The quantity zz* is always a nonnegative real number equal to [2|? or to [2*[? (which are the same), 2. The quantity z + z* is always a real number, equal to 2 Re z or to 2 Re z* (which are the same). 3. The rules (21 + z2)* = 2f +23 and (c1z2)" = 242% are evident and should be remembered. Figure 2.2 Because complex numbers obey the same addition rule that applies to vectors in a plane, they can be added graphically by the parallelogram rule (Fig. 2.2a). Conversely, vectors in a plane can be represented by complex numbers. The scalar product of two such vectors can be obtained by the rule (21-22) = Re (2]z5) = Re (2123), where it is understood that z, and z2 are vectors corresponding to complex num- bers z; and z respectively. The vector product can be obtained in a similar fashion: fer X za] = Im (2322) = ~—Im G28). Exercise. Verify the validity of the above rules for scalar and vector products. In the theory of complex variables, the expression |z1 — za] is often used. According ‘to Fig, 2.2(b) this quantity (modulus of the complex number z; — zz) is equal to the distance between the points z) and zg in the complex plane. It follows that the statement [2 — zol < (which often occurs in proofs of various theorems) means geometrically that point z is within the circle of radius R drawn around the point zg as a center (i.¢., z is in the R-neighborhood of zo; see p. 16). The foltowing two inequalities are easily proved from geometrical considerations: We fan + zal S leat + [zal (A side of a triangle is less than or cqual to the sum of the other two sides.) 2 fza — zef > [zal — zall. (The difference of two sides of a triangle is tess than or equal to the third side.) \ 48 FUNCTIONS OF A COMPLEX VARIABLE 23 Remark, 1 should be emphasized that inequalities can cxist only among the moduli of complex numbers, not among the complex numbers themselves. A complex number cannot be greater or smailer than another complex number. Also, there are no positice or negative complex numbers. 23 DE MOIVRE FORMULA AND THE CALCULATION OF ROOTS While addition and subtraction of complex numbers are most casily performed in their cartesian form z = x + éy, multiplication and division are easier in trigonometric form, If z, = ry(cos #, + isin @)) and zz = ro(cos #2 + isin G2), then elementary calculation shows that 2122 = Fyro{cos (6) + G2) + isin (@; + 82) with the provision that if #; + 62 happens to be greater than x, or Jess than or equal to —z, then the amount 21 should be added or subtracted to fulfill the condition ~+ < (6; + 62) 0) to which the positive imaginary semiaxis is added. The negative imaginary semiaxis is not included. The second branch, which has no special name, maps the z-plane onto the left half-plane (Re w < 0) plus the negative imaginary semiaxis. Except for z = 0, no other point on the w-plane (image plane) is duplicated by both mappings. Also observe another important feature of the two branches. Each branch taken separately is discontinuous on the negative real semiaxis. The meaning of this is as follows: The points ee at BY where 8 is a small positive number, are very close to each other. However, their images under the principal branch mapping, namely Siler = 2D and © fy(za) = eT HH), are very far from each other. On the other hard, note that the image of zz under the mapping f(z), namely, Soleo) — ete 248/21, is very close to the point f:(z1). It appears that the continuity of mapping can be preserved if we switch branches as we cross the negative real semiaxis. To give this idea a more precise meaning we must define the concept of con- tinuous function of a complex variable. Let w = f(z) be defined in some neigh- borhood (see pp. 47 and 16) of point zp and let f(zo) = Wo. We say that f(z) is continuous at zq if* f(z) wo whenever z— Zo in the sense that given 5 > 0 {arbitrarily smafl), the inequality |/(2) — wol’< 8 holds whenever |z — zo| <€ holds, for sufficiently small ¢. It is readily shownf that if w = u(x, y) + HG, y), then the continuity of w implies the continuity of u(x, p) and u(x, y) and vice versa. * Also written as lim..1, £(2) = f(z0). } For example, see Kaplan, p. 495. 2.6 MULTIVALUED FUNCTIONS AND RIEMANN SURFACES 55 Riemann proposed an ingenious device to represent both branches by means of a single continuous mapping: Imagine two separate z-planes cut along the negative real semiaxis from “minus infinity” to zero. Imagine that the planes are superimposed on each other but retain their separate identity in the manner of two sheets of paper laid on top of each other. Now suppose that the second quadrant of the upper sheet is joined along the cut to the fourth quadrant of the lower sheet to form a continuous surface (Fig. 2.6). It is now possible to start acurve C in the third quadrant of the upper sheet, go around the origin, and cross the negative real semiaxis into the third quadrant of the lower sheet in a con- tinuous motion (remaining on the surface). The curve can be continued on the lower sheet around the origin into the second’ quadrant of the lower sheet. Two edges joined here, Lower sheet Figure 2.6 Figure 2.7 Now imagine the second quadrant of the lower sheet joined to the third quadrant of the upper sheet along the same cut (independently of the first joint and actually disregarding its existence). The curve C can then be continued onto the upper sheet and may return to the sfarting point, This process of cutting and cross- Joining two planes leads to the formation of a Riemann surface which is thought of as a single continuous surface formed of two Riemann sheets (Fig. 2.7). An important remark is now in order: The line between the second quadrant of the upper sheet and the third quadrant of the lower sheet is to be considered as distinct from the line between the second quadrant of the lower sheet and the third quadrant of the upper one. This is where the paper model fails us. According to this model the negative real semiaxis appears as the fine where all four edges of our cuts meet. However, the Riemann surface has no such property; there are two real negative semiaxes on the Riemann surface just as there are two real posi- tive semiaxes. The mapping f(z) = \/Z may help io visualize this: The principal branch maps the upper Riemann sheet (negative real semiaxis excluded) onto the region Re w > 0 of the w-plane. The line joining the second upper with the third lower quadrants is also mapped by the principal branch onto the positive imaginary semiaxis, The lower Riemann sheet (negative reat semiaxis excluded) is mapped by the second branch onto the region Re w < 0. The line joining the second fower with the third upper quadrants is mapped (by the second branch) onto the negative imaginary semiaxis. In this fashion the entire Riemann surface. 56 FUNCTIONS OF A COMPLEX VARIABLE 2.6 is mapped one-to-one onto the w-plane (z = 0 is mapped onto w = 0; this particular correspondence, strictly speaking, belongs to neither branch since the polar angle @ is not defined for z = 0). The splitting of a multivalued function into branches is arbitrary to a great extent, For instance, define the following two functions which also may be treated as branches of f(z) = V2: Jr eft for O<@ cos z tanz : cosh 2 = $(e* + e7"), sinh z = }e* — e77), ' sinh z 1 tanh z = Sogha’ cothz = Ganz’ . * If the so-called point at infinity (Section 2.14) is taken into account, then the mapping f(@) = V7 also has two branch points, . . 26 MULTIVALUED FUNCTIONS AND RIEMANN SURFACES 57 Alt these functions are periodic: sin z and cos z have a (primitive) period 2x, tan z has a (primitive) period x, e', sinh z, and cosh z have a (primitive) period 2ni. A score of familiar formulas can be established, for instance, sin (21 + 22) = sin z, cos zg +- cos 2, sin za, sin z1 — sin zp = 2cos Zit2o.. m+ ze 5-3 sin +) ete. 2 2 Also note that coshz = cos(iz) — sinhz = —/sin (iz). It is worthwhile to mention that {sin z] and |cos z| are by no means bounded by unity, for instance, Isin 2if = 3.24. The logarithmic function is defined as the inverse of exponential function. Solving e” = z = re“ for w, we obtain the general solution w= loge} i+ ne — (n = integer). This function is multivalued: Its principal branch is usually denoted by w = log z and is defined as logz = logr+i@ (-¥ <8 < xh The entire multivalued function is referred to as w = Logz = logz + inn. ‘These formulas are often written with the help of the argument of z function which is also multivalued, the principal branch being agz=@ (-w<@<7) and the entire function reading Arg z = arg z+ 2x. Thus we may write logz = log|z| + fargz, Logz = log |z] + /Argz. The functions Arg z and Log z require a Riemann surface consi many Rietnann sheets. The definition of inverse trigonometric and hyperbolic functions now easily follows. All are multivalued: Arccosz = i Log(z + Vz? — 1), x Are sinz = 3 ~ Arccosz, ing of infinitely 1 i-z Arctanz — y;Log 743° Arsinh z = Log (z + Wz? + 1), Arcosh z = Log (z + Vz — 1), Artanhz = 4 Log $2. 58 | FUNCTIONS OF A COMPLEX VARIABLE 27 Their principal branches are denoted by arc cos z, etc.* The general power fisction 2° is defined by fom aloes, For rational a(a.= p/q, where p,q are integers and q x 0) this function coincides with ¥2? (g¢ > 0) or with (V2*)-? | < 0) and has g branches. For irrational a the power function is infinitely multivatued. 2.7 ANALYTIC FUNCTIONS. CAUCHY THEOREM In this section we shall discuss the subject of calculus of functions of a complex variable. The basic concept of the continuity of a complex function has already been presented, and it is not ifficult to verify that the sum, product, and quotient {except for division by zero) of two continuous functions is continuous, A con- tinuous function of a continuous function is also continuous.t Let C be a piecewise smooth curve in a complex plane. If f(z) is continuous on C, then the complex integral d: [foe can be defined and expressed in terms of real integrals by putting f@) = ule, y) + wl, y) and de = dx + idy; this yields fayde = [ (udx —vdy) + if @dx + udy), Cc Cc Cc where the real integrals Jo (u dx — vdy) and fo (v dx + udy) are known to exist. Curve C may be open or closed but the direction of integration must be specified in cither case. The reversal of this direction results in the change of sign of the integral. Complex integrals are, therefore, reducible to curvilinear real integrals and possess the following properties: [, Wer + a@)de =f, ferds +f, ete) de Ww GC Cc Cc f kf@dz =k , fizjdz — (k = complex constant), ' @ f, fa) dz = f, fo dz + i, fea, e where C is decomposed into two curves, C, and Co. The absolute value of an integral can be estimated by the formula | I fey «| < ML, where M = max[f(z)| on C, and L is the length of C. * Another widespread notation is arc sin z = sin7! z, arsinh z = sinh~ 'z, and so on. { For example, Kaplan, p. 496. {In the sense of Ricmann; see, ¢.2., Courant, Vol. 1, p. 133, and Kaplan, p. 299. 27 ANALYTIC FUNCTIONS. CAUCHY THEOREM 59 As our next concept we shall define the derivative of a complex function: Changing z into z + Az (with complex Az), we obtain f(z + 4z) and we can write™ fz), I) = 2 fe) = jim SEAM As in the case of real functions, this limit may or may not exist. It may be empha- ~ sized that in the above formula Az may approach zero in an arbitrary fashion, that is, z + 42 may approach z along any curve or by any sequence, This rather stringent requirement implies that f(z) must indeed be “well behaved” at point z in order to be differentiable. Function f(z) is said to be analytic (regular, or holomorphic) at point z if it possesses a derivative at z and at all points of some neighborhood of z (small but finite). This additional requirement results in many desirable properties of analytic functions, such as the existence of derivatives of all orders. The theory of functions of a complex variable deals essentially with analytic functions. Mere existence of a derivative at all points of a neighborhood may be shown to imply that the derivative is continuous.t Also, it is a simple matter to verity (by the same technique as for real variables) that the derivatives of complex functions obey the usual rules: a Hors +0) — 0 4 Oe, ) Lovin) = Mt 4 Py, @) a # where w= wr) and f= #2), @) 4 nz" (4 = integer), 3) and so on. The differentials of complex functions are defined in the same way as for real functions: If w = f(z), then dw = f"(2) dz If we set f(z) = w = u(x, y) + io(x, y), then the definition of the derivative can be rewritten as Je) = tim WA Ax y+ Ay) — wx, 9) + ole + Ox y+ Ay) — 08»), Aso Ax | iAy Ay0 The limiting value on the right-hand side must be the same for the arbitrary ap- proach Az — 0. In particular, set Az = Ax (approach along the real axis); then Pea Be i®. * See p. 54 (including the footnote) for the definition of a limit. t See, e.g., Knopp, Theory of Functions, Vol. 1, p. 63. 60 FUNCTIONS OF A COMPLEX VARIABLE 27 Alternatively, set Az = Ay (approach along the imaginary axis); then ee Lo-Fi It follows that for a differentiable function w = u + iv we must have au av au oe ox ay’ aye These are the Cauchy-Riemann conditions; they follow directly from the definition of the derivative. If, further, f(z) is analytic, then f’(z) must be continuous which, in turn,* implies that the partial derivatives of w and v are continuous. The inverse theorem also holds: If u(x, y) and ofx, y) have continuous first partial derivatives satisfying Cauchy-Riemann conditions in some neighborhood : of z, then f(z) = # + 1 is analytic at z. ao Integrals of analytic functions possess some very important properties. Per- haps the most fundamental one is expressed by the Cauchy theorem: Uf f(2) és analytic in a simply connected domain D, and C is a (piecewise smooth) simple closed curve in D, then $e f@) dz = 0. Proof. Write the integral as $f a = gua — ody) + if Ode + udy). Analyticity of f(z) implies continuity of gartial derivatives of w and v and Green's theorem (p. 26) is applicable. However, then the Cauchy-Riemann conditions imply dv au fad oan =ff (34% dx dy = 0, ai fou+uay=ff 2 dx dy = 0, 8 and the theorem follows. There is a converse of the Cauchy theorem, known as the Morera theorem:t If f(z) is continuous in a domain D and if $f(2) dz = 0 for every simple closed path in D with its interior also in D, then f(2) is analytic in D. rt It is not difficult to see that the Cauchy theorem is true for multiply/connected domains provided the interior of the simple closed path C is also inside the domain * See, e.g., Kaplan, p. 510. } Knopp, Theory of Functions, Vol. 1, p. 66. 27 ANALYTIC FUNCTIONS, CAUCHY THEOREM 61 (ie, the path does not encircle a hole; see Fig. £.15), Similar extensions hold for related integra! theorems that will be quoted later, The vanishing of a contour integral (an integral around a simple closed path) is closely related to the independence of path of an integral. In fact, the considera- tions of Section 1.6 can be applicd easily to complex integrals, leading to the statement: If ¥ f(z) dz = 0 for every simple closed path, then the integral ff fevar is independent of path (between zp and z). Suppose now that the point zo is fixed. If the integral [°, f(¢} dt is inde- pendent of path, then it must represent a function of z. This function is then a primitive function of f(z) (or an indefinite integral of f(2)) as follows from the Sundamental theorem of integral calculus: If f(2) is analytic in a simply connected domain D, then the function Fe =[" Aor is also analytic in D and f(z) = (d/dz)F(2). Proof. Since f(z) is analytic, the integral is independent of path and is therefore a function of z, In the expression ay F@)=U+iV= (de — vay) + if" (ode + udy), (4 a,¥o? 0.m0 both integrals are independent of path (by Green’s theorem and the Cauchy- Riemann conditions). It also follows that* w_, wy ax” ay , wi, My x By so that and v satisfy the Cauchy-Riemann conditions as well. Therefore F(Z) is analytic, Morcover, dF _ au, av a Bt an tigg tet w= fe), and the theorem follows. Any two primitive functions must differ by a (complex) constant; this follows from the fact that/”(z) = Oimplies /(z) = const (integrate au/ax = 0, au/ay = 0, elc.). * For example, Kaplan, p. 244. 62 FUNCTIONS OF A COMPLEX VARIABLE 28 2.8 OTHER INTEGRAL THEOREMS. CAUCHY INTEGRAL FORMULA It should be emphasized that all conditions stated in the Cauchy theorem must be checked before applications. Consider, for instance, the integral = 1 = T= fie (a = const). Is this integral zero or not? Generally speaking, f(z) = 1/(z — a) is an analytic function but it fails to be analytic at one (and only one) point, namely, z= a. The function is not even defined at this point and thus cannot possess 2 derivative. . Let the curve C involved in the definition of J be a simple closed curve. Then, if the point z = a is outside the curve, the Cauchy theorem holds and J = 0. If it is inside, the Cauchy theorem cannot be applied. In fact, the integral is not equal to zero, as demonstrated by the following considerations: If Cis a circle of radius R centered at z = a, then the integral is easily evaluated by setting z=a-+ Re®, In this case dz = iRe” do and r= [ide = dvi. It is not difficult to show that the same result is true for any simple closed path C,, which encircles point z = a. Suppose that C, is entirely inside the circle C (Fig. 2.8). Then a thin channel made up of curves By and Bz can be constructed to connect the interior of C1 with the exterior of C and the Cauchy theorem can be applied to the shaded region; a domain D can be constructed so that the shaded region is within it. The integral over C, is clockwise. As the sides By and Ba of the channel are allowed to approach each other, the integrals of {@ = 1/ — @) along B, and B, will (in the limit) cancel out, leaving us with the statement ia c ¢, Se)dz + ¢ flay dz = 0. c ° cs Counterclackwise ——Clackwise Reversing tHe direction of the second integration to > counterclockwise, we obtain A Ah 5 8; an a ee 2 fod = 4, 1@d Re Figure 2.8 (with both directions counterclockwise). : If Cis entirely within C1, the proof is similar. If C and C, intersect, the proof is even simpler. Exercise. Produce a proof of the discussed statement given that C and Cj intersect at two points. | 2.8 OTHER THEOREMS. CAUCHY FORMULA 63 If the integrat J is evaluated around a closed path which is not simple, its valuc is not necessarily 2x, In cases of practical interest it will be equal to n2zi, where 1 is the number of times the path encircles the point z = a counterclockwise less the number of times it encircles the point z = a clockwise. Of course, it should be understood that the integral $f(2) dz may happen to be zero even if the Cauchy theorem does not apply. For instance, calculate the integral t= $eige where 7 is a positive integer not equal to unity and the contour is a citcle of radius Raround z = a. Using z = a + Re®, obtain +n os sf pRingt—08 gg = PEN san —0 1 This result evidently holds for any closed path encircling z = a. In both of these examples, the possibility of the point z = a being exactly on the path of integration has been avoided, and for a good reason; such integrals cannot actually be defined. Whenever this situation occurs in practical problems, the path must be deformed to avoid the troublesome point. How this is to be done depends on the nature of the problem.* Function f(z) in the Cauchy theorem must, of course, be single valued. It may be a particular branch of a multiple-valued function, but then care should be taken that this particular branch is analytic. Consider, for instance, the integral Fines VF along the nit circfe about the origin. First of all, the branch of the (double-valued) function 1/2 must be specified. Suppose it is the principal branch. Then § Vids =f eRe ao = — 45, The Cauchy theorem is not applicable because f(2) is not analytic within the circle lz] = 1. The points where it fails to be analytic are along the real axis from x= —1 to x = 0 where f(z) is not even continuous. Note also that although £(z) is continuous at z = 0, it is not analytic at that point either. Consider now the same integral ¢ Vide #4-21=1 * One such example is given in Section 12.9. 64s FUNCTIONS OF A COMPLEX VARIABLE 28 taken around the point z = —2 (Fig. 2.9). If the principal branch is involved in the integration, the Cauchy theorem is not applicable. However, split V2 into the following two branches (as on p. 56): fz = VF eh? if O<0<5, Branch A: ‘i Vea Vr eiltael if we 8S 0. Branch B: (v3 = Vrelltti if OC OSs, ~ V2 = vrei? if -x<6<¢0. Now the branch cut is along the positive real semiaxis, and each branch js analytic within (and on) the circle |z + 2| = 1 and the Cauchy theorem may be applicd. Figure 2.9 Figure 2.10 The Cauchy theorem can be generalized in more than one way, Observe that the interior of 2 simple closed path is evidently a simply connected domain with the path serving as its boundary. A (finite) collection of nonintersecting simple closed paths may form a multiply connected domain, as illustrated in Fig. 2.10: The paths Cy, C2, and C3 form the boundary of the (triply connected) domain S. With respect to this domain S, the contours Cy, Cx, and Cg are oriented in the following sense (see p. 21): The positive direction of circulation around C1, Ca, and C, is such that the domain S appears to the feft. This implies, in our example, a counterclockwise direction for C; but a clockwise direction for Cz and C3. ‘These concepts lead (o the so-called Cauchy theorem for multiply connected domains: If S is a multiply-connected domain whose boundary B consists of a finite number of simple closed paths, then (2) dz = 0, §, 40 provided f(z) is analytic in S and on B and the integration is carried out in,a positive direction over all parts of B. The proof is constructed without difficulty by using the technique of channels (p. 62). In this (and other) integral theorems the analyticity of f(z) on @ contour fas well as in its interior) is demanded. ‘This is because the Cauchy theorem requires 28 OTHER THEOREMS. CAUCHY FORMULA 65 the contour to be located within some simply connected domain D. Actually, this condition can be relaxed and the following boundary integral theorem holds: If f(2) is analytic in the interior S of a simple closed path C and f(z) is continuous on C* then $c f(z) dz = 0. The proof is based on the construction of a contour C’ which is in $ and arbitrarily close to C; the Cauchy theorem is valid for C’. The continuity of £(z) is then used to show that the integral over C’ must approach the integral over C as C’ approaches C. The Cauchy theorem can be used to deduce many other properties of integrals, the basic one being the Cauchy integral formula: If f(2) is analytic inside and on C and if the point z = a is in the interior of C, then Me 5, BG = ifla). Proof. Construct a circle C’ about z = a with an arbitrarily small radius R such that the circle is within C (Fig. 2.11). The integral over C is evidently the same as the integral over C’ (use the channel technique). Rewrite this integral: ade _ £ fa)— fla) $08 - | MaMa no$ 5 f2)— Ka) pO da + fla\axi by a previous result. The remaining integral on the right-hand side must be inde- pendent of the radius R (because other terms are independent of R). It can also be estimated that $ LO= LO 4.) < ¢ LO = LON a,c Mage = 16M ro ~ Ser "= OR 7 zoe = al where Mf — max 1f(z) — f(@[ on the circle C’. If R is chosen sufficiently small, then M can be made arbitrarily small [by continuity of f(z) at the point z = a}. Therefore the quantity ¢ $O= fe) 4) zZ—a@ can bé shown to be less than any positive number, no matter how small. This is possible only if £2) — fla) dz =0, cz a and the theorem follows. Figure 2.11 * To be precise, provided a point on C is approached from the interior. 66 FUNCTIONS OF A COMPLEX VARIABLE 29 The Cauchy integral formula reveals a remarkable property of analytic fune- tions: If the values of a function ace specified along a closed contour, then its value at an arbitrary point inside the contour is already predetermined. To emphasize this, replace a in the formula by z (variable) and denote the dummy variable by ¢: I $fod, JO) = FGF Fz Jn this formula, the integral is a Function of the (variable) parameter z and can be differentiated with respect to z. The following theorem (Leibmitz rule) follows from the corresponding, properties of real integrals:* d — § ahh 2) £$ pode f op ot (contour C.must be a simple closed curve in the usual sense, namely, it should not extend to infinity, otherwise the convergence questions must be considered). Application of the Leibnitz rule to the Cauchy integral theorem yields an inte- gral expression for the derivative: Ye) _ 1g LOK | dz Imi l 2 Repetition of this process gives us the nth derivative Pf2)_ mf MOG. dn ~ di} (— 2H 2.9 COMPLEX SEQUENCES AND SERIES No serious study of analytic functions is possible without their representation in the form of series, and we now turn our attention to consideration of this aspect of the theory. Our first task would be to consider complex sequences. ‘An infinite sequence of complex numbers {2} = (21, 22,--} is said to converge to the (complex) limit z, provided ; len — 21 <€ for sufficiently large 7; ¢ is, of course, an arbitrarily small positive number. Con- vergence of complex sequences is reducible to that of real sequences by the fol- lowing fundamental theorem: The sequence {2,} converges to z = x + iy if and oniy if Re Zp converges to x and Im z, converges to y. Proof. \f Rezm = X,— x and Im zy = yay, then : len = [Gn = 2) iG — PS ben — 31 + [rn WL < (6/2) + /2) = € * See Kaplan, p. 289. 29 COMPLEX SEQUENCES AND SERIES or for sufficiently large n. Conversely, if zn — 2] < ¢, then zy is within the circle of radius ¢ about the point z. This implies In - xl ce and [pp — yl Zk fat * See, e.g., Knopp, Theory and Application of Infinite Series, Section 10. 68 FUNCTIONS OF A COMPLEX VARTADLE 29 is a convergent sequence. Denoting S$ = lim S,, we customarily write If the sequence of partial sums does not converge, then the series is said to be divergent. It is to be emphasized that, under certain circumstances, divergent series can be given definite meanings and such series are widely used in applica- tions.* However, the theorems derived for convergent sequences should not be indiscriminately applied to divergent sequences. A series is said to be absolutely convergent if the (real) series of moduli E lal nol is a convergent series. An absolutely convergent series is convergent (the proof is trivial). In most cases onc proves convergence of a complex series by establishing that jt is absolutely convergent. The following tests are most common. Comparison test. Mf |zn| & for all n sufficiently large, and k > 1, then ¥°z, diverges. Root test. Hf Y]z,] < k < | for all n sufficiently large, then 272, converges absolutely, and if [za] 2 & > 1 for all sufficiently large, then 3'z, diverges. Prools of these theorems are similar to those for real series; they are based on the inequalities for absolute values which are also true for complex numbers. Divergence of a series can often be quickly established by the ath term test: If z, fails to converge to zero then the series }0z, diverges. If necessary, the question of convergence of a complex series can always be re- duced to that of two real series by the basic theorem: The series 127, = Fo(%n + iva) converges to S = P + iQ if and only if 2x, converges to P and Syn converges to Q. For instance, series which are convergent but not absolutely convergent can be treated in this fashion. Complex series can be addéd and subtracted provided they are convergent. They can be multiplied only if they are absolutely convergent, the product being also an absolutely convergent series; if the series are not absolutely convergent, then we are faced with the problem of arranging the product series. * For example, see Section 6.4. {See Knopp, Theory and Application of Infinite Series, Section 45. 29 COMPLEX SEQUENCES AND SERIES 69 Terms of a complex series may depend on a complex variable z. Most common serics are power serics, for instance, Ssltz24+274274.-.. In many cases such series will converge only if z is confined to a certain region. The above series converges absolutely, by ratio test, provided [z| < 1. This series diverges, by ratio test, if |z| > 1. The ratio test is inconctusive if |z| = 1, but then the xth-term test shows that the series diverges. It is seen that the above power series converges absolutely for all points inside a circle of radius R = 1 called the radius of convergence. The concept of the radius of convergence can be applied to every power series, Indeed, if a power series is convergent on a circle of some radius 7, then it is absolutely convergent everywhere inside this circle* (by comparison test). The problem js then to find the upper bound of r which is the sought radius of convergence. Exercise, Show that the series 1 — 32-4 92? ~ 2729 + Bizt — +++ has a radius of convergence equal to $, while the series L424 22? + 33 4 Aled 4 --- has a radius of convergence equal to zero; the point z = 0 is the only value for which the series converges, If the upper bound described above does not exist, then the series converges absolutely for all values of z and is said to have an infinite radius of convergence, For example, 2 2 zt (hatha tat Partial sums of a power series represent a sequence of polynomials in z. Sequences of other functions can also be considered. The sequence {f,(2)} of functions defined in a region R {z belongs to R) is said to converge to a limit func- tion f(z) in R, provided lim fue) = fl2y for each z in R. For instance, the partial sums of the series Draltr¢ 27424. aed * This statement is known as Abel's theorem. See Kaplan, p. 350. 70. FUNCTIONS OF A COMPLEX VARIABLE 29 form a sequence of functions (polynomials) = oF I= Vz, kao and this sequence converges to the function f(z) = 1/(1 — z) in the (open) region (z| < 1 because at fal z¢ 2p $ r= <3 and . zt tim y 0 for |z| < L. For this reason the function f(z) = 1/(1 — z) is said to be the sum of the above series (for |z| < 1 only!): Frad-y) ld N, holds simultaneously for all 2 in R. In plain terms: Let us suppose we desire a certain accuracy ¢ for our ap- proximation. For some particular z, the tenth function in the sequence may suffice (ten terms of a series, if we are talking about partial sums). But for another z, the tenth function may be inadequate because the speed of convergence is slower. In general, we may need to go farther and farther along the sequence as we proceed to points where the convergence gets increasingly worse. Uniform convergence sets the end to this process. The convergence can be no worse than a certain specified degree and the Nth term will guarantee a certain accuracy for the entire region. Example. The sequence of partial sums of the series m0 ‘ i is convergent for |z| <1 but it is not uniformly convergent; the convergence becomes increasingly worse as |z} > I. However, in the region-|z| < & where k <1, the convergence is uniform, It may be “bad” at z = k, but once M is 2,10 TAYLOR AND LAURENT SERTES Ww found such that 2 0 for alln > Nand z = &, then the same value of W will hold for all other 2 with bl sk. The series of functions is called uniformly convergent (in a region R) if the sequence of its partial sums is uniformly convergent (in that region). Uniform convergence of series is most commonly established by the Weierstrass M-test: The series f(z) of Functions is uniformly convergent in a region R if there exists a series of positive constants M, such that Wi@| 1/R’ = Ra, and the statement follows. Therefore we conclude thatif Ra > Xj, then the series 42 D en(z — a)" nate will converge within the annulus Ry < |z — al < Ro. It can, of course, happen that Re < Ri, in which case our series will diverge everywhere. The following theorem can now be derived: Every function f(z) analytic in an annulus Ri<|z-al< Re con be expanded in a series of positive and negative powers of (2 — a), namely & . f@= Di ene - ay. This series, known as the Laurent series, is unique for a given annulus, and the coefficients ¢c, can be obtained from fe)d m= ie 1G — ati’ where I’ is a circle of radius R such that Ry < R < Re. Figure 213° Proof. Contract the radius Re slightly and expand the radius R, slightly to-obtain an annular region, with point z inside, to which the generalized Cauchy formuta is applicable (Fig. 2.13):* fora, Vf Ma. mr—z | dyn, f—z oat The first integral can be treated as in the derivation of the Faylor series: 1 KO _ Oz Oa 1 OS ta Lp Ane erat = y C—O ae Lg Gta , tid, Foz @ — ati e- Geo Term-by-term integration is permissible by uniform convergence. * We shall occasionally use the convenient symbols g and g¢ to indicate clockwise or counterclockwise direction of integration. 2.10 TAYLOR AND LAURENT SERIES 75 The second integral is treated by expanding I/(¢ — z) in a somewhat different geometric series: Lot i ~~ = =o Poe" (Foe i= Gs aera ~~ Eye a ae which is convergent by ratio test. Then id for _ _ 1g fox Wit, Foz” TT, 2 2 1 1 m = 2 @septi HP, Hone - 0 a. Replace m by —(+ + 1) (n must be negative) and rewrite the above as al foe. F eo bg fayag Ii Fr, Foe ni Fn, & — apr” Finally, note that the integrals a §, 4, (8 =0,1,2,..) _fodg 1 (@ — at and @ = -1,-2,~3,...) may be just as well evaluated over a common citcle I’, concentric with I, and I's, and lying within the annulus Ry < R < Ry, To prove the uniqueness, assume that an expansion te S@) = a. en(z — a)" exists and is valid in the annulus Ry < jz — a] < Re. Choose an arbitrary integer k, multiply both sides of this expression by (2 — a)~*—!, and integrate around a circle T about z = a, lying within the annulus. Then fad dz . All integrals on the right-hand side will vanish except one, for which n = k, and whose value is 2x7. Therefore ¢ fe) de = aay, , @— ae which completes the proof. 76 FUNCTIONS OF A COMPLEX VARIABLE 2.10 The part of the Laurent series consisting of positive powers of (z — a) is called the regular part, it resembles the Taylor series but it should be emphasized that (he nth coefficient cannot be associated, in general, with f(a) because the Jatter may not exist. In most applications, f(z) is not analytic at z= a. The other part, consisting of negative powers, is called the principal part. Either part (or both) may terminate or be identically zero, Of course, if the principal part is identically zero, then f(z) is analytic at z = a, and the Laurent series is identical with the Taylor series. Remark. The Laurent series is unique onty for a specified annulus. In general, a func- tion f(z) may possess two or more enlirely different Laurent series about a given point, valid for different (nonoverlapping) regions. For instance, 1 _! 2 Ba. Aicyrtitete tet O |al. Write If [z| > |al, then Ja/z] < 1, and we can expand Loe E (vir T= fz fn 2.10 TAYLOR AND LAURENT SERIES 77 Therefore Ls dal > lel. This is the desired Laurent series. The function f(z) can be expanded by this method about any point z = 5: Indeed, write ee ee | $0 = a> ED TEL FTE (b #4). Then, either =-—!_ Fe 57 @— 6)" =~ GDR & Dime @= by (lz — b| < fa — dp, or = ye ey _ _ fa\= Xe (hh > Ja ~ bp. Example 2. Rational fraction decomposition 1 9-5 -@piR ER The roots of the denominator are a = i, b = 2 (simple and distinct). There- fore f(z) fails to be analytic only at z = fand z = 2 and should possess a Taylor series about z = 0 valid for |z| < 1 ({i] = 1) and two Laurent series about z = 0 valid for 1 < |z| < 2and {z| > 2. To obtain these three series, we use the identities 2 Q4iz+ l= @- 2-2 1 | | 1 £0 erHE Sym THI (; - =): Suppose that the Laurent series valid for 1 < fz| < 2 is desired. The function 1/@ ~ 2) should then be expanded in Taylor series about z — 0 (Example 1), This series is, in particular, vatid for ! < Jz] <2. The function t/(z — i) should be expanded in the Laurent series about z = 0 valid for jz| > 1 (Example 1). This series is also vatid for < Jz < 2. If these two series are subtracted, we may obtain (multiplying by 1/(2 — i) a sevies for f(z) valid for 1 < jz} <2 which is the desired Laurent series, and Example 3, Differentiation. f(z) = 1/(z ~ 192. The method applied in Example 2 fails here because of the double root of the denominator. Among the alternative methods the simplest one is, perhaps, to observe that it i A(t). zZ-tpo a@\i—z 7% ~~ FUNCTIONS OF A COMPLEX VARIABLE 2410 The seri 3 1 Scrics boetteteteo= Det (zp < fan can be diflcrentiated term by term within the circle of convergence. Therefore —! Ho Py. ¥ » Gop Tt tPF = Lat be (al < 1). Example 4. Integration. f(z) = log (1 + z) = log [1 + z] + éarg (I + 2). This is the principat branch of the {muttivalucd) logarithmic function, The branch line extends from “minus infinity” to minus one and log (1 + z) is analytic within the circle |z| = 1. We know that ‘d i BOBUt OK pgs Therefore we may expand le 2atoet 2-24 te = dCs" da a in an arbitrary fashion: 1. f(2) remains bounded, that is, [/(2)] < B for a fixed B. 2. f@) is not bounded and |f(2)| approaches infinity, namely, if(z)| > M (any M) for |z ~ af < € (some ©). 3. Neither of the two cases above; in plain terms, f(z) oscillates in a “wild” manner. Examples of these three types of singularities (at z = ) are Case l, f(z) = sin z/z, Case 2, S(2) = Vsin z, Case 3, f(z) = et? The first case turns out to be trivial because then the limit lim,_, f(z} must exist, and if the function f(z) is defined at z = a by f(a) = fim, ., f(z), then it must be regular at z = a as well. Remark. The formuta f(z) — sin z/z does not define, in a rigorous sence, the value of the function at z = 0. The extended formula sin 2/2, 2 #0, yo =| 2-0 does define the function f{2) at z = 0 (and elsewhere). To prove the above statement concerning Case 1, observe that f(z) is analytic in an annulus p < |z — a] < R within a neighborhood of z = a. By the Cauchy theorem, for any point z within this annulus (Fig. 2.14), we have fz) = LhMat, 1 gh Ma, Qi §—2 | Ini Sy ft 2 rb We shall show that the second integral must be zero for 3 all p. If this is 80, then f(z) must approach the limit (LP) which will prove two things at once: (a) lima f(2)y exists, (b) if f(a) is defined by this limit then the rede- fined function f(z} is analytic at z = a as well. Figure 2.14 80 FUNCTIONS OF A COMPLEX VARIABLE 21 To achieve this result, write ¢ — z = ($ — a) — (¢ — a) and observe that IG - 2) - @- a 2k - 4] - ft - 4 Then, for a fixed 2, 1g Mae 2xi Jy § ~ Zz > as e—al —p? ~ [zal —p ‘The integral must be independent of p because of the analyticity of the integrand. Since it is less than an arbitrary positive number (for sufficiently small p) it must be equal to zero. The proof is now complete. Because of the described property, isolated singularities of the first type are called removable singularities. In practice, if a function is defined by a formula which fails for some isolated point z = a, then the formuta is tacitly replaced by the corresponding limit. In this sense, the functions S@) = sinz/z, g@)= er", AZ) = I/z — cotz are analytic at z = 0. The second type of isolated singularity, when [f(z)| 7 oo as z— a, is called a pole. Since the singularity is isolated, there must exist a Laurent series $e fa) = Dente ~ a" valid for 0 < |z — al < R (for some R). If the principal part terminates, ie, if the Laurent series is of the form += fay~ EX ele — ay", then f(z) has a pole of order mat z = a. Conversely, if f(z) has a pole at z = 4, it must have the Laurent series (for 0 < |z — a] < R) of the above form as seen from the following argument: Consider the function gz) = I/f@)- Unless f(z) = 0 there must exist a neighborhood of z = a where f(z) has no zeros. In this neighborhood, g(z) is analytic and lg@>0 as za (because [/(z) + oz). Therefore g(z) has a zero at z = a. This zero miust be of some definite order m, namely, ee)= YA a) OX < R). 241 ZEROS AND SINGULARITIES 8l Rewrite this as ae) = @ — 0)" © KG ~ a = aH, where 6, = Bay, and 6) 0. Now Ati. 3@) ~ @ — a ¥@) Since y(z) is analytic at z = a and does not vanish there, it follows that 1/p(2) must be analytic as well and possess a Taylor series at z = a. Then f@) = lose e n= ch ~ a). fiz) ea wes a y However, this is evidently a Laurent series with a terminating principal part so that the argument is complete. Example. The function f(z) = cscz = 1/sinz has the Laurent series valid for O<|eb 0 nor approaches infinity, for an arbitrary manner of approach: For instance, if z approaches zero along the negative real semiaxis, then |f(z}| > 0; if it approaches zero along the positive real semiaxis, then |f(z)| > 90; if it approaches zero along the imaginary axis, then | (2) remains constant but arg f(z) oscillates, and so on. 82 FUNCTIONS OF A COMPLEX VARIABLE 2.41 In fact it is not difficult to show that even in an arbitrarily small neighborhood of an essential singularity, a function f(z} assumes values arbitrarily close to any desired complex number (Weierstrass-Casorati theorem}. Also, an even more explicit statement can be proved, known as Picard’s theorem. In an arbitrarily small neighborhood of an essential singularity, a function f(z) assumes infinitely many times every complex value except, perhaps, one particular value Remark. It should be emphasized that an infinite principal part in the Laurent series implies essential singularity only if the series is valid “up to the singular point”:* 1 I i fO- Gaap* aa tet does not mean that f(z) has essential singularity at z = |. The point is that the above series converges only if |z — t| > 1. [actually represents the function 1 fa) = 3242 in the annulus 1 < |z — 1] < R (for any R). This function evidently has a simple pole at z = 1. Apart from isolated singularities, complex functions can have other types of breakdown of analyticity. The most common case is that of a branch point. Con- sider, for instance, the function f(z) = /z. For every point except the origin it is possible to construct a neighborhood and find a branch of \/z that will be analytic in this neighborhood. This branch may not be the principal one or its complementary, called the second branch in Section 2.6; rather, it may be some- thing like branch A in the terminology on p. 56. For the origin, this is not possible because any neighborhood of z — 0 will contain a portion of the branch tine (whatever it may be) where the sclected branch is discontinuous. Simitar state- ments can be made about other multivalued functions and their branches. It is evident that there can be no Taylor or Laurent series valid for the region Q < jz — al'< R (for some R) about a branch point z = a, However, Laurent series valid for R, < |z — al < Rz are sometimes possible. Example VETT This series is valid for |z| > 1 and represents a branch of the function yz? — I which is analytic in this region. The branch line joins, in this case, two branch points z = +1 and z = —1 and does not extend to infinity (Fig. 2.15). Replacing * Namely, for all points in a neighborhood |z — al < e except z = @. | 2.12 THE RESIDUE THEOREM AND ITS APPLICATIONS 33 zby ( — 1), we obtain a Laurent series centered about the branch point z = |. The Jatter will converge for |z — I[ > 2. Another type of singular behavior of an analytic function occurs when it possesses an infinite number of isolated singularities converging to some fimit point. Consider, for instance, Im 1 1 IQ) = er = Say" The denominator has simple zeros whenever =1 tl ne ! Branch the zea he (a = +1,42,.. Figure 2.45 The function f(z) has simple poles at these points and the sequence of these poles converges toward the origin. The origin cannot be called an isolated singularity because every neighborhood of it contains at feast one pole (actually an infinite number of then). 2.12 THE RESIDUE THEOREM AND ITS APPLICATIONS. Let f(z) be analytic in some neighborhood of z = a except, perhaps, at z = a itself [in other words f(z) is either analytic at z = a or has an isolated singularity there]. Let C be a simple closed path lying in this neighborhood and surrounding z= a, Then the integral Res fla) = oa $ Sle) dz is independent of the choice of C and is called the residue of function at the point z=, Evidently, if f(z) is analytic at z = @ (the point z = a is then called a regular point), the residue is zero. If z = a is an isolated singularity, then the residue may or may not be zero. Examples 1. f(z) = 1/2; the residue at z = 2. f@) = 1/27; the residue at z = 0 is equal to zero. is equal to unity. According to the formula on p. 74, the residue is seen to be identical with the coefficient ¢_, in the Laurent series +o S@) = ee calz — a)", which is valid for 0 < {z — al < R {for some R). The residues of a function at its isolated singularities find their application in the evaluation of integrals, complex or real, The basis for these applications is 84 FUNCTIONS OF A COMPLEX VARIABLE 2.12 the residue theorem: if f(z) is analytic on and inside a closed contour C except for @ finite number of isolated singularities at z = 1,42, ...,4,, which ate all located inside C, then ¢, fQ) dz = Ui xy Res f(a). Cc kel ‘This theorem is proved by the technique of cutting the channels between the con- tour Cand small circles Cy, Cy, ... around each singularity (Fig. 2.16). Figure 2.16 Figure 2.17 There is a variety of practical methods for quick evatuation of residues: FIRST METHOD. From the definition, Res fla) = 54; ¢ fa) az (using a suitably chosen contour C). This method is rarely used, but may be valu- able if the primitive function of f(z) is known and has a branch point at z = a. Example. (2) = t/r, F@) = Logz. Any branch of Log z may be chosen, bal in order to preserve the relationship dF(z) &! J@ = F@) = the closed contour must be disconnected and the appropriate limiting process must be applied. For instance (Fig. 2.17), a $ 1g tin | % = tim {log (4) — tog (B)} = 2ni. cz AoA SR OZ BoA Here the principal branch was used, which possesses a discontinuity 2n/ on the negative real semiaxis. 212 ‘THE RESIDUE THEOREM AND TTS APPLICATIONS — 85. SECOND METHOD. For a simple pole at 2 = a the following formula holds: Res f(a) = lim (2 — a)f(z). The limit involved is often obtained by simple substitution or the use of well- known Jimiting values. Example.’ f(z) = tan z/z?. Then sinz 1 = lim — 0 COSZ " tan se Res f(0) = lim z=, ee THIRD METHOD. For a pole of order m at z = a the following formula holds: grt Res f(a) = me ir lim ( I@- array} Example. f(z) = e/z*. Then Res (0) = dtin {& (4) = jlime* ‘ FOURTH METHOD. A common case of a simple pole is when /(2) has the form o(z) 42) = we’ where o(a) # 0-and ¥(2) has a simple zero at z = a. In this case Res f(a) ~ £3. Note that ifz = a is a simple zero of o(2), then ¥’(@) cannot vanish. Exanple. f(@) = e/sinz. Then Res foy= | at cos 212-0 FIFTH METHOD. Expand f(z) in the Laurent series and pick out the residue. This method is valuable if f(z) is a product of functions with known Laurent series. The series for f(z) is then obtained by multiplication. In practice, the coefficient c_, can be picked out by inspection. Example. {(Q) = et/(2 + DG — Yt The residue at z — 1 is desired. In the first step, transfer the pole to the origin: z-l=0, r=e+1, Then 86 FUNCTIONS OF A COMPLEX VARIABLE 2.12 In the second step, expand e and 1/( + w): uu Pot, e attw+ aw +e eee (all w), 1 11 1 eo 2 stathiten Sb SS -$t (ot <3). In the third step, evaluate (by inspection) the coefficient with « from the product of the above two series: ifoiié 3 (BT In the fourth step, evaluate the residuc: 4 1 ff ne sn=(G-R-% Prove the validity of the third method given above. [Hint: Represent f(2) as F@ = eP@/Z- a” and use the formula from p. 66.} Exercis The residue theorem can be applied to the evaluation of a wide variety of definite integrals, real or complex. Some of the most frequently used procedures are shown in the several examples which follow. Example 1. Consider the real integral 2 1 ' =f, Toapesey pe lel). This integral can be converted into a contour integral in complex plane by selting z= e, Then = cose = 3 (2+ 2) iz 2 Zz ‘ and : I= $ _ a, cz — py — pz) where C is the unit circle in the z-plane. The integrand has two poles: at z— p and z = I/p. If |p] < 4, the pole z = p is inside the contour, while the pole z ='I/pis outside. Only the residue at z = p is necded; it is equal to t ‘ fl therefore (pl <1). 2.12 THE RESIDUE THEOREM AND ITS APPLICATIONS a7 If |p| > 1, the needed residue is at 2 — 1/p and it is equal to yielding al 1 2 Padip git oT le. Note that both results can be combined into -poy lxd., while the integral is not defined for |p] = 1. 2 This method can be used for integrals of the type fq" R(cos 6, sin 8) d6, where R(cos 8, sin 8) is a rational function of cos @ and sin 8. Example 2, Consider the real integral dx 7 dx =f) wia- ial, eae > The integral [jy dx/(x? + a) can be treated as a portion of the complex integral Fc dz/(z? + a?) evaluated over the contour C shown in Fig. 2.18. Indeed (set z = xon the real axis): +R dz f dx f az cmb a Sp Ha SoH ae Let us estimate the integral over the semicircle Ce when R is very large: Write ee 2a” 21+ at/et If |z| = R is very large, then ja?/z?| = a?/R? is small and {1 +- a7/z?| is almost equal to unity (Fig. 2.19). To be precise, observe that [1 + a?/z?| > 4 for Im =plane rmicircle Ce Im 22 il +R Re Figure 2.18 Figure 2.19 88 FUNCTIONS OF A COMPLEX VARIABLE 2.42 R > ay/7, and consequently 1 Teapey < 2 (for R > aV2). This implies l 2 5 Tere eT avV/2). Now, employ the estimate (p. 58) f dz Qn Cr? + a R < xR max <= te eral dz lim sai = 0 Rae Jeg Z? + a Observe now that the integral ¢ dz/(z? + a”) is independent of the radius K (so long as R is greater than a) because the only singularity of the integrand within C is at z = ai and, by the residue theorem, a . . lor fate = Dwi Res flai) ~ Ini z= ~ 5 (for all € such that R > a). Consequently, if we let R > oo, we have +R dz . ax . a fete tim aia iin | ta which reduces to ar : a~ L. wpe Gr Exercise. Show that ifa < 0, then $00 , dx = 7. ’ no +a a [Hint; No new calculations need be done, just some logical deductions ] The above procedure can be applied to integrals of the type f * Po) dx, : -» OG)” i where P(x) and Q(x) are polynomials in x and (a) Q(x) should have no real zeros* * If Q(x) has reat zeros, then the integra! is not defined (see p. 63), unless particular modifications are made (e.g., p- 111). 212 THE RESIDUE THEOREM AND ITS APPLICATIONS 89 and (b) the degree of Q(x) must exceed the degree of P(x) by at least 2 (otherwise the integral over the semicircle Cy may not tend lo zero). For such integrals it is true that 40 POD gy = Oni f. 0G)” = ani Res, where 374. Res is the sum of the residues of the integrand in the upper half-plane. ‘This statement is a special case of the following theorem: Uf $(2) is continuous for [2] > Ro (Some Ro) and |zf(z)| > 0 uniformly as |z| + «, then im l2l=R fe) a= Proof. az| 2nR Fran ¢l = Uf 420 8] pm rere Uniform convergence of |z/(2)} means that |2f(2)| < (for arbitrarily small whenever |z| > R (that is, independently of the manner in which z approaches infinity), Then Ms nasore| <2 and the theorem follows. The conditions of the theorem are satisfied by the function fz) = 3 (with deg @ > deg P + 2) because (a) all zeros of Q must be within some fixed circle about the origin and (b) the condition |zf(z)| < ¢ for |z| > k (for some k) can be satisfied, Remark. The method described above can be extended to certain integrals of the type ” 9) di f FO) dx, where f(x) Is an even function of x, Example 3. Consider the real integral ” cos x dx [ xe at (a> 0). Note, first of ali, that Tn 4 feosxds. © 2 Se X* Fat The replacement of x by z will not help in this case because cos z is not “well behaved” in the upper half-plane; it is not bounded. However, the function e* is bomded in the upper half-plane because e -= ee and |e] = 1 (all real x) 90 FUNCTIONS OF A COMPLEX VARIABLE 242 while Je] < 1 (all nonnegative y). For this reason, consider the complex integral i= edz +8 6 dx + et de cE a Jp PHA Scyz bat evaluated over the contour shown in Fig. 2.20. Observe that . et dz fim | ~~ nom Jen 2? + a =0 {as in Example 2). Also, J = ni Res flai) = xi so that 49g 40 40. { e“ dx cosxdx ) . sin x dx oa COS * OX = ==e7 Pram], eta! J. ere ae @>0. Since the right-hand side is real, it follows that 40 cosxdx Ime r-f, Bee are a>). Remark. The statement to. sin x dx = x2 fat 0 could have been made immediately on the grounds of symmetry. However, if it were not teue, we would have obtained the value of this integra! as well from the imaginary part of 2r? 4 Res. Example 4. Consider the real integral r= f ae Se pal ft Sin xX 2t-w X Since sin z is not “well behaved” on the upper half-plane, we shall try to evaluate thé comptex integral | fe A new problem now arises: the integrand has a pole on the real axis. Note that this pole is not caused by sin x but rather by cos x, which has been added to form the complex integral. . As before, observe that 242 THE RESIDUE THEOREM AND ITS APPLICATIONS — 91 cr eRe +R Figure 2,20 Figure 2.21 Te be able to apply the residue theorem we must avoid the pole at the origin in some ion, Let us suppose we do this by means of a semi sadius r in the upper hal/-plane, as shown in Fig. 2.21. Then we ea of a small write The reason the chosen contour C is helpful in the evaluation of our Teal integral is as follows. Note that (by continuity of sin Xx/x} ee o (in either order), we obtain * sin x ; i . if = dx — wi. x Therefore* I >> »|2 # i via ; * Of course, this integral can be evaluated by much more elementary methods. The pur- pose of the above analysis is to iflustrate the techniques of the residue calculus on a simple example rather than to obtain this particular result. . 242 THE RESIDUE THEOREM AND ITS APPLICATIONS — 93 Example 5. We shall now evaluate the integral of Example 4 by a somewhat different method. We write ” sin tf sin a taf eae bf stay Lf Hazy, ox Zo 2h. Here we treat our integral as a complex integral over a path (real axis) which is open (50 far). Since sin 2/z is continuous at z = 0, we may deform the contour as shown in Fig. 2.22 and claim that ratio [ S102 | loz rot Now set ——+Re sinz = bee? — e* z= 5" — e7¥4), The problem is to evaluate . ev = lim saz and rea tor 2 For /,, close the contour as usual (Fig. 2.23a), Show that the integral over Cr. approaches zcro, and deduce that fe [Sass on z For /2, close the contour through the fower haff-plane as shown in Fig, 2.23(b). Now [e~“| is bounded in the lower half-plane and the integral over Cp approaches zero. On the other hand, note that (a) there is a contribution from the pole at the origin and (b) the clockwise integration introduces a change in the sign. When this is taken into account, we obtain f s- de = —Ini Res f(0) = —2ni. 1 Im Tm Ce a —-_+ Re Figure 2.23 94 FUNCTIONS OF A COMPLEX VARIABLE 2.12 Combining these results, we deduce that 40 sinx 1 1 E SA de = 55h — fe) = 30+ 28) = & in conformity with the previous result. Remark |. The integral over the semicircle C, in Example 4 has been shown to yield the value —zi in the limit ry > 0. This is just one-half the value oblained by integration over the fult circle. One may prove a general theorem to this effect: Let f(z) be analytic at z= a, Consider the integral [Woe nna taken from z1 = @ + re to z2 = a+ re along the circle \z — al =r (Fig, 2.24). Then lim 1, = aéffa) ro Figure 2.24 where a = 02 — 6; + 2nm (choose nso that la] < 2x). Exercise. Prove this theorem using @ technique similar to that in Example 4. Remark 2. The statement that the integral of eit/z over Cr in Example 4 vanishes in the limit R —+ © is also a special case of a more general result, known as Jordan’s lerama: if f(2) concerges uniformly to zero whenever 2 approaches infrity, then tim f ftzye™ dr Ram ICR 0 where ) is any positive nwnber and Cx is the upper half of the circle The term “uniform convergence” as z—» © means that |/{z)] < « whenever |z| > M (for some M) no matter what the phase of z is, Example 6, The integral reads +7 or e rf cosh xx (ow on. St follows that J; is bounded by, say 2e'""+””, and this approaches zero since lal < x. A similar argument holds for 2. Collecting our results, we obtain 12 2m =IntlatAtia= te so that = 1 —— = a, ~ cos (ajay ~ 3 96 FUNCTIONS OF A COMPLEX VARIABLE 212 Cfixample TD) This example deals with the extension of previous techniques to contours involving branch lines of multivalued functions. Consider the real integral “es ref Me @ 0 and x —> 00. In the field of complex numbers the general power function z*~* (p. 58) implies ZIT) = Ge MEogs o [zfpmteile Ares os [af bella Met ake? The branch which reduces to the real function x7~* (for x > 0) is the principal branch, with k = 0. It is convenient to work with the branch for which O< Argz < 2r. This branch coincides with the principal branch for Im z > O and with the branch k = 1 for lmz < 0. Then our real integral £ coincides* with the integral of this branch along the upper edge of the branch cut (see Fig. 2.26). Close the contour as shown in the figure. The integral over the lower edge of the branch cut is then ° ‘ - [etree bee _ rons f wt dx . [ rer ¥= ne lo T+x Therefore, ede golds co, V2 Seg 12 Ii Res f(—1) = [LE — oe Now, for [z} = R, roa ce i +z | ~ 2m R79 (as Ro) and for [el » 7, [ od dz! cp bbe The residue at z = —1 is e@—*; therefore vw Dart = 2x7 30 {as r > 0). jeter * In the limit ¢—+ 0,7 = 0, and R-> =, of course. 2.13 CONFORMAL MAPPING BY ANALYTIC FUNCTIONS 97 Figure 2.27 Remark. Itis also possible to operate with the standard Principal branch —r < Argz

You might also like