PHYSICS | Eugene ButkovMATHEMATICAL PHYSICS EUGENE BUTKOV St. John’s University, New York ADDISON-WESLEY PUBLISHING COMPANY Reading, Massachusetts - Menlo Park, California - London Sydney - Manilap20 Wey s g z g Dxpye STupE® WORLD STUDENT SERIES EDITION FIRST PRINTING 1973 ‘A complete and unabridged reprint of the original American textbook, this World Student Series edition may be sold only in those countries to which it is con- signed by Addison-Wesley or its authorized trade distributors. It 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 America or 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- ut 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 in 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 physics is no exception. There is a aumber of texts in this field, and some of them are undoubtedly of outstanding quality. Nevertheless, many teachers often feel that none of the existing texts is 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 it 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- lowing 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 late 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 a familiar background and is later extended to more sophisticated situations. A typical example is Chapter 8, 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 PREFACE 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 multiple 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 greatly 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 1 11 12 13 14 1.5 1.6 17 1.8 19 Chapter 2 24 2.2 23 24 2.5 26 27 28 2.9 2.10 241 2.12 2.13 214 215 Chapter 3 34 ake mae 3.4 35 3.6 CONTENTS Vectors, Matrices, and Coordinates Introduction . Vectors in Cartesian Coordinate Systems Changes of Axes. Rotation Matrices Repeated Rotations. Matrix Multiplication . ‘Skew Cartesian Systems. Matrices in General Scalar and Vector Fields Vector Fields in Plane . Vector Fields in Space . Curvilinear 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 Multivalued 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 Infinity Integral Representations Linear Differential Equations of Second Order General Introduction. The Wronskian : General Solution of The Homogeneous Equation The Nonhomogeneous Equation. Variation of Constants . Power Series Solutions . The Frobenius Method . : Some other Methods of Solution . vii 45 48 49 51 54 58 62 1 78 83 97 102 104 123 125 126 128 130 147Vi CONTENTS Chapter 4 4.1 4.2 43 44 45 4.6 47 48 Chapter 5 5.1 5.2 53 5.4 5.5 5.6 5.7 5.8 5.9 5.10 Chapter 6 6.1 62 63 64 65 6.6 67 68 69 6.10 Chapter 7 WW 12 73 14 15 16 17 Fourier Series Trigonometric Series Definition of Fourier Series Examples of Fourier Series Parity Properties. Sine 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 Basic Properties of Laplace Transform The Inversion Problem . ‘The Rational Fraction Decomposition The Convolution Theorem. . Additional Properties of Laplace Transform Periodic Functions. Rectification . The Mellin Inversion Integral . ae Applications of Laplace Transforms . Concepts of the Theory of Distributions Strongly Peaked Functions and The Dirac Delta Function Delta Sequences . The 6-Calculus Representations of Delta Functions Applications of The é-Calculus Weak Convergence... Correspondence of Functions and Distributions. Properties of Distributions . Sequences and Series of Distributions Distributions in N dimensions . Fourier Transforms Representations of a Function Examples of Fourier Transformations Properties of Fourier Transforms . Fourier Integral Theorem... Fourier Transforms of Distributions . Fourier Sine and Cosine Transforms . Applications of Fourier Transforms. The Principle of Causality - 154 155 157 161 165 167 168 172 179 180 184 187 189 194 200 204 206 210 221 223 226 229 232 236 240 245 250 257 260 262 266 269 211 273 276Chapter 8 8.1 8.2 8.3 84 8.5 8.6 8.7 88 89 Chapter 9 9.1 9.2 9.3 94 9.5 9.6 97 98 99 9.10 9.11 Chapter 10 10.1 10.2 10.3 10.4 10.5 10.6 10.7 10.8 10.9 ‘Chapter 11 iui 11.2 113 14 Ws 11.6 17 11.8 CONTENTS Partial Differential Equations The Stretched String. Wave Equation The Method 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. Helmholtz 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 : 7 Associated Legendre Functions and Spherical Harmonics . Spherical Bessel Functions . ‘Neumann Functions ' Modified Bessel Functions . Finite-Dimensional Linear Spaces Oscillations of Systems with Two Degrees of Freedom Normal Coordinates and Linear Transformations Vector Spaces, Bases, Coordinates Linear Operators, Matrices, Inverses . Changes of Basis Inner Product. Orthogonality. Unitary Operators . The Metric. Generalized Orthogonality . Eigenvalue Problems. Diagonalization Simultaneous Diagonalization. . . ~ Infinite-Dimensional Vector Spaces Spaces of Functions . The Postulates of Quantum Mechanics _ The Harmonic Oscillator Matrix Representations of Linear Operators. Algebraic Methods of Solution Bases with Generalized Orthogonality Stretched String with a Discrete Mass in the Middle Applications of Eigenfunctions 1x 287 291 295 297 299 304 308 313 319 332 334 337 340 342 350 355 372 381 388 394, 405 4u1 419 424 433 437 441 443 451 463 467 471 476 483 488 492 495X% CONTENTS Chapter 12 12.1 12.2 12.3 12.4 12.5 12.6 12.7 12.8 12.9 Chapter 13 13.1 13.2 13.3 13.4 13.5 13.6 aaa 13.8 13.9 ‘Chapter 14 14.1 142 143 144 14.5 146 147 148 149 14.10 ‘Chapter 15 15.1 15.2 15.3 15.4 15.5 15.6 Chapter 16 16.1 16.2 Green's Functions Introduction . Green's Function for the Sturm: Series Expansions for G(x|£) Green's Functions in Two Dimensions Green's Functions for Initial Conditions . Green's Functions with Reflection Properties Green’s Functions for Boundary Conditions The Green's Function Method ‘A Case of Continuous Spectrum jouville Operator Variational Methods The Brachistochrone Problem . The Euler-Lagrange Equation . Hamilton's Principle . Problems involving Sturm-Liouville Operators. The Rayleigh-Ritz Method Variational Problems with Constraints Variational Formulation of Eigenvalue Problems Variational Problems in Many Dimensions . Formulation of Eigenvalue Problems by The Ratio ‘Method Traveling Waves, Radiation, Scattering Motion of Infinite Stretched String Propagation of Initial Conditions . ‘ Semi-infinite String. Use of Symmetry Properties. Energy and Power Flow in a Stretched String Generation of Waves in a Stretched String Radiation of Sound from a Pulsating Sphere The Retarded Potential : Traveling Waves in Nonhomogeneous Media Scattering Amplitudes and Phase Shifts. Scattering in Three Dimensions. Partial Wave Analysis Perturbation Methods Introduction . . The Born Approximation . Perturbation of Eigenvalue Problems . First-Order Rayleigh-Schrédinger Theory The Second-Order Nondegenerate Theory The Case of Degenerate Eigenvalues . Tensors Introduction . : : ‘Two-Dimensional Stresses . 503 514 520 523 $27 531 536 543 553 554 560 562 565 567 573 577 581 589 592 595 599 603 611 619 624 628 633 647 650 653 658 665 671 67216.3 16.4 16.5 16.6 16.7 16.8 16.9 16.10 16.11 CONTENTS Cartesian Tensors : Algebra of Cartesian Tensors zi Kronecker and Levi-Civita Tensors. Pseudotensors Derivatives of Tensors. Strain Tensor and Hooke’s Law . Tensors in Skew Cartesian Frames. Covariant and Contravariant Representations . General Tensors . EE ie Algebra of General Tensors. Rel The Covariant Derivative Calculus of General Tensors e Tensors . PR ceeeceecce 676 681 684 687 696 700 705 m1 ms RI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 series 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 a 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 textbooks a vector is defined as a quantity characterized by magnitude and direction. We shall see in Chapter 10 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, etc. 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 _b) 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 (Chapter 10) we shall assume that scalars are real numbers. 1J Vb bana MANIC Es, AND) COORDINATES 12 a 4 In many cases we can plot various vectors from a’single point, the origin. ‘Then euch 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 point can serve as the components of the corresponding vector at the same time. This is illustrated 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 1.1 Figure 1.2 We can now associate with 2 vector u (in space) a set of three scalars (14, Uy, Us), such that du will correspond to (dts, My, Muz) and u-+ v will correspond to (uz + Ve, Uy + by Uz + v,). Note that no such relations hold, in general, if a vector is characterized by other types of coordinates, e.g., spherical or cylindrical. In addition, orthogonal cartesian coordinates result in very simple formulas for other common quantities associated with vectors, such as a) length (magnitude) of a vector: ul = uw = G2 + uj + ui)", 'b) projections of a vector on coordinate axes: uz = ucos(u,i), uy = ucos(u,j), us = ucos (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-, y-, and z-directions, respectively. The symbol (u, v) stands for the angle between the directions given by wand v,eae VECTORS IN CARTESIAN COORDINATE SYSTEMS. c) ©) projection of a vector on an arbitrary direc- tion defined by vector s (Fig. 1.3): OP = u, = ucosy = uz C05 (5,4) + uy COs (5, j) + us cos (s, k), d) scalar product (dot product) of two vectors: (uv) = wv cos (u,v) = usvs + uy + Us, €) vector product (cross product): x Figure 1.3 (u X v] = (ube — uavy)i + (ude — Uads)j + (Usdy — Uyrs)k. The important distinctive feature of the cross product is that [u X v] ¥ [v X ul], namely, it is not commutative; rather, it is anticommutative: fu X vy) = —[v Xu}. 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 v}, namely ue |sin (u, v)|, is equal to the area of the parallelogram formed by u and v. The direction of [u X v] can serve to distinguish the “positive side” of the parallelogram from its “negative side.” Figure 1.4 shows two views of the same parallelogram illustrating this idea, “ [uxv] Positive side ‘Negative side Figure 1.4 Closely related to this property is the concept of a right-handed triple of vectors. Any three vectors u, v, and w, taken in this order, are said to form a right- handed (or positive) triple if the so-called triple product (fu X v]-w) is positive* 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 ((u X ¥] - w) represents, in this case, the volume V of the parallelepiped formed by the vectors u, v, and w. * These vectors form a left-handed (negative) triple if (lu X v] + W)< 0.4 VECTORS, MATRICES, AND COORDINATES 13 Show that Vv = \(fu x vw) under any circumstances. Show also that the sign of the triple product is unchanged under cyclic permutation of u, v, and w, that is, (tu Xx v)-w) = (lw X u)-¥) = (ly X w)-u). Figure 1.5 1.3 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 = u,i + uj, the x’-component of u is the sum of projections of vectors usi and uyj on the x/-axis, and similarly for the y’-component. From the diagram we sce that this yields ul, = u,cos@ + uysin@, wl, = —uz sin @ + wy cos 6. It is instructive to note that the angle between the x’= and y-axes is (x/2 — 0) while the angle between the y’- and x-axes is (7/2 + 6). In view of sind = 00s (5 - ) —sin @ = cos G + a): we see that all four coefficients in the above equations represent cosines of the angles between the respective axes. Let us now 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 * Instead of “components,” the term “coordinates of a vector” is often used (see also Section 10.3). + For example, Goldstein, Classical Mechanics, Section 4.6,13 CHANGES OF AXES. ROTATION MATRICES 5 Write u = ui + uj + uk, collect contributions to wz from the three vectors uzi, uyj, and uk, and obtain uw, = uz cos (i’, i) + uy cos (i’, j) + uz cos (i’, 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 i, j, and k. Figure 1.7 It is clear that similar formulas can be written for uj and u;. At this stage, however, it is very convenient to switch to a different notation: Instead of writing (uz, Uy, Uz), let us write (u4, U2, wg) and similarly (u',, w, u's) for (ut, uf, uf). More- over, denote by amn the angle between the mth primed axis and the nth unprimed axis (three such angles are marked on Fig. 1.6) and by dn, the corresponding cosine (that is, Gn = COS mn). This new notation permits us to write the transformation formulas in an easily memorized pattern: uy = arity + ayotte + arsits, Wz = Gait + Goat, + azsits, us = Ggity + ag2¥2 + assis, or, if desired, in the compact form From this analysis we conclude that the new components (u', w2, #4) can be obtained from the old components (wi, 12, uz) with the help of nine coefficients.6 VECTORS, MATRICES, AND COORDINATES 3 These nine coefficients, arranged in the self-explanatory pattern below are said to form a matrix.* We shall denote matrices by capital letters. Columns a2 ay3 | Ist a20 23 | 2nd pRows a32 a33 | 3rd Matrix A has three rows and three columns; the individual coefficients an, are referred to as matrix elements, or entries. It is customary to use the first subscript (m in our case) to label the row and the second one to label the column; thus the matrix element a; 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 colwmn 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 A. 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 u 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 + 4120 + 4430 = au, Ub = dail + a220 + G230 = aoi, uy = 311 + 320 + a330 = agi. We see 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 (@11, @21, @31) and we can write i= ay’ + aoij’ + azik’. Similar statements relate j and k to the second and third column vectors of A. Note that in this discussion the unit vectors i, 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 X 3 matrix is formed. The reader can easily construct an analogous 2 X 2 matrix to account for two-dimensional rotations.13 CHANGES OF AXES. ROTATION MATRICES a Exercise. Establish the geometrical meaning of the row vectors of matrix A representing a rotation. The definitions introduced above allow the computation of (w, u4, w4) from (ux, ua, ua) by the following rule: To obtain uf, take the dot product of the kth row of matrix A with the vector u, as given by the triple (uy, ua, us). Since in this process each row of the matrix is “dotted” with (u1, wa, v3), 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: 411 a2 a3 uy uh 42, G92 Gog | Ua 7 ue 431 432 33 us us Matrix A Column vector u Column vector u’ As we see, the old components are arranged in a column which we shall denote by u. This column vector is multiplied by the matrix A and this results in another column vector, denoted by u’. 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 w’, and similarly for u. The entire procedure is symbolically written as Au=u. Remark. Note that the set (1, u2, us), arranged in a column, has not been denoted simply by u but rather by a new symbol u.* The point is that in the context of our problem, both w and 1’ represent the same vector u, but with respect to different systems of axes. ‘We must think of u and 1’ 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, 411412 + 421422 + 431432 = 0, 412013 + a22de3 + a32d33 = 0, 413011 + G23a21 + a33@31 = 0. This property follows from the fact that the columns of A are representations (in the new system) of the vectors i, j, and k and these vectors are mutually orthogonal. * The symbol u should not be confused with |ul, the magnitude of vector u, which is also denoted by u (p. 2).8 VECTORS, MATRICES, AND COORDINATES 14 b) The columns of 4 have unit magnitude, namely, ai, + ah + ah, = 1, ais + ads + ae = 1, ats + ais + ads = 1, because i, j, and k are unit vectors. c) The rows of A are also mutually orthogonal and have unit magnitude. This is verified by establishing the geometrical meaning of the row 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 (j.e., the determinant of the equations on p. 5) is equal to +1. Otherwise, orthogonal matrices can have the deter- minant equal to —1. The point is that a rotation must yield a right-handed triple (i,j, K’) 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: 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 Cyn in terms of matrix elements dyn aNd Brn. We have uy = Brau + biotte + bisuss, uy = bewuis+ beoud + bosus, Us = bath + dgatts + basus, and uy = ayy + Gyatle + aysts, ua = aot, + Goats + Goats, Uy = 31, + azale + asgus, * It will be shown in Chapter 10 that any N X N matrix which satisfies (a) and (b) must also satisfy (c). + See Problem 4 at the end of this chapter.14 REPEATED ROTATIONS. MATRIX MULTIPLICATION 9 from which it follows that wy = (brrais + Bi2der + bisasi)ur + (Birai2 + bi2de2 + bisase)u2 + (611413 + bizaes + b13@33)us, uy = (barair + border + besasi)ur + (barai2 + ba2de2 + besase)u2 + (b2ia13 + beeaes + be3as3)us, ug = (b3iai1 + bsage1 + b3sas1)ur + (bsidi2 + 632022 + b33a32)u2 + (bsiais + bs2aes + baadas)us- The maze of matrix elements above becomes quite manageable if we observe that every coefficient associated with u, 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 Cmn of matrix C is obtained by taking the dot product of the mth row of matrix B with the nth column of matrix A. Now, if we record our relationship in the vector-matrix symbolic notation Ce ee ers 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 : . be written as’ Ww’ = Cu = (BA. In this sense we write Ci Cuz Cis bir biz bis M1 42 a3 Ca1 C22 Cos |=| bei bee bes |*| ai 22 aes €31 C32 C33 bs1 B32 bss 931 432 G33 or, symbolically, ela 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 multiplication of ordinary numbers (scalars). A simple check shows that the associative law holds: If we multiply three matrices A, B, and C in that order, then this can be done in two ways: ABC = (AB)C = A(BC) (where it is understood that the operation in parentheses is performed first). * The difference is, of course, that in B(Au) the column vector u is first multiplied by A, producing another column vector which is, in turn, multiplied by B. In (BA)u the matrices are being multiplied first, resulting in a new matrix which acts on u.10 VECTORS, MATRICES, AND COORDINATES 14 Exercise. Verify this statement. [Hint: If AB = D, then the elements of D are given by 3 don = Do dmibine ro Develop the matrix elements of (4B)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 ul = Au. Suppose now that we rotate the axes back to their original position. The new components of vector u are given by (11, #2, ug) and the old ones are given by (u4, ub, U4); these components must be related by some matrix B: u= Bu. Combining these relations we obtain = B(Au) = (BA)u. Therefore, the matrix BA must transform the components (11, U2, us) into them- selves. It is easy to see that this task is accomplished by the sc-called unit matrix (“identity matrix”) SN 0 oon 0 1 0 - oo Exercise. Show that if w = (BA)u is to hold 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 A~" 50 that we have A~'A = I. Since we could have performed our rotations in reverse order, it is not hard to see that AB = J as well, that is, A~'A = AA~* and A = B~!. While two rotation matrices may not commute, a rotation matrix always commutes with its inverse.t * If we write 4/B = X the question would arise whether we imply A = BX or A = XB. See Section 10.4 for a general statement to that effect.ae SKEW CARTESIAN SYSTEMS. MATRICES IN GENERAL 7 It may now be of interest to relate the elements of matrix B to those of matrix A. To obtain Bm, we should, in principle, solve the equations on p. 5 for Uy, U2, us. However, in the case of rotations, we have a much simpler method at our disposal. Let us write the matrix equation BA = J in detail: bir bia bis 411 412 13 100 bo: bee bas |*| aa1 G22 aes | =| 0 1 0 bs1 bs2 bss 431 432 a33 ool 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 see that the first row vector of B is orthogonal to j and k and its dot product with i is unity. Consequently, it could be nothing else but the vector i (in new representation, of course), and we conclude that by, = 44, by2 = Gay, and by3 = ag). Repeat this argument for the other rows of B and deduce that the rows of B are the columns of A and vice versa. This is also expressed by the formula bmn = 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, i, j, 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, ul? = u2 + uj — usw, cos (w — 6) = ul + uf + 2u-u, cos @ that cae [ul = Vu? + uz + 2u,u, cos $. 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, (a @ +) = @-y) + ew). Indeed, from Fig. 1.9, no matter which coordinate system is used, we see that @- w+) - (MN). However, MN = MP + PN; since MP is the projection of v on u, we have |u| - QP) = |u| |v| -cos (u,v), 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 = usi + uj and v = va + oy): + 608 (u, ¥). u|- |v + wl cos @ = (> ¥) = Quek val) + (uct ogi) + (uy val) + (yi: evi) = Ups + Uv, COS > + Ups COS + Uy. 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. } Since (u-v) = (v-u), the second distributive law ((u + ¥)-w) = (u-w) + (v-w) is trivial.ae SKEW CARTESIAN SYSTEMS. MATRICES IN GENERAL eg 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 xy-system with unit vectors i and j. A given vector u can be represented either as u = u,i + u,j or as u= uli’ + uj’, ua” oO Figure 1.9 Figure 1.10 Although i’ = i, the components ut and u, are not equal; rather, we have u, = 00' = 00 — 00 = u, — uy tan. Also, uy = OP = uy sec 7. We see that the new components (uf, u,) are linearly related to the old components (Uz, ty) and this relationship can be represented by means of vector-matrix multiplication: uy 1 -tany Uy uy | | 0 secy ele This can be written symbolically as w= Au where w’ stands for the column vector (u:, uw), and w stands for the column vector (uz, uy). 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 wz, uy in terms of u%, ut, and it reads 1 siny A= 0 cos ¥ 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, +0-j, f= —tany-i’ + secy-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 u’ = Au (A is a 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 w: Ue, = dug — Quy uy = Qty — Uy. If we attempt to solve these two equations for uz and uy we find that they have no solution if u, and u’, are arbitrary. We say that the matrix 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 w;, ui, cannot possibly repre- sent an arbitrary vector in plane: Our equations imply u; = 2u, 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 u; = 2u, is actually satisfied, our system of equations has an infinity of solutions for uz and u, (since two equations reduce to a single one). Furthermore, if an additional requirement uz = 2u, is imposed, the system has a unique solution (uz = 4u'; ty = 4uj). Alll 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. The simplest example is, perhaps, a position vector which depends on time ¢. In a fixed coordinate system, 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. 11 and 440).1.6 SCALAR AND VECTOR FIELDS 15 and we write ws ut) = ue() + lt) § + melt) “ke Such vectors can be differentiated with respect to the variable 1 according to the definition d — fm Ue +40 (1), Ae eo ae ae With u(1) and u(t + Af) expressed in terms of their components, it is trivial to deduce that ede dey dus HO eige eae ge 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, z).* Such vectors are said to form vector fields and can be denoted as follows: w= u(x, y, 2) = we(X, Ys Zi + myx, Ys ZI) + uel, Y, ZK. 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, y, 2), 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, electrostatic potential, etc. A scalar field gives rise to numerous other quantities through its various partial derivatives. In particular, let us concentrate our attention on a) the total differential = % oe oe de = Sede + Seedy + 52 de, and b) the directional derivativet de _ de dx , d0 dy, 36 de, a ~ ax d& + ay a + oz * These vectors may also depend on time, but we shall be mostly interested in instanta- neous relationships, where ¢ has some fixed value. } This is also called conservative field or potential field. t Rate of change of g per unit length in some particular direction characterized, say, by the element of arc ds of some curve. See, e.g., Apostol, p. 104,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 o(x, y, 2) by the vector a = 98; 4 9°; 4 grad e = Sei + eit 5k. Then we can write de dp = (grad p-ds) and SP = (grad @- 0), where ds = dxi-+ dyj + dzk represents infinitesimal displacement in some direction and so = Fi 4 Bi 4 Sx 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), y=, z= 2, where x(1), y(i), and 2(1) have continuous derivatives with respect to the parameter 1 (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.t This last property is trivially extended to piecewise smooth curves; i.e., those consisting of a finite number of smooth parts. We shall assume that ali 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”). If a set of points * Observe that dx/ds = cos (50, i), etc., are the directional cosines of the direction defined by ds or by So + We shall assume that all integrals are Riemann integrals which are adequate for our purposes. For example, see Apostol p. 276. {A more precise term is an neighborhood.1.6 SCALAR AND VECTOR FIELDS 7 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 tied to a neighborhood. We must make sure a region is an open set before we can say that f(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 = grad oC, ¥, 2) and consider the following integral between points M(Xo¥os26) and NX, YZ), taken along some curve C: L (u- ds) = ft (¢ ax +9 e ay + $a): Aloe C Oe Using the parameter ¢ as the variable of integration, we have N ‘4 i -dsy =| (20 4 4 de ay, ae de’ a= [4 _ a i w a =f, ( ate Et eS di t= Hh) — eos sioatc where fo and f, are the values of the parameter ¢ 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 M fixed and treating N as a variable point, we can define a function 0 952) =f" (weds) = [OP (ued + ty dy + te de). * From now on, we shall occasionally use the term “path” to indicate a piecewise smooth curve.18 VECTORS, MATRICES, AND COORDINATES 16 Path C; M, ty N Path Cy Path C; Path C M : N @) (b) Figure 1.12 It is now simple to show that grad y = u. For instance, (e4azu.2) el + Ax, 2) — 9662) = [ONY (ue de + uy dy + ue de) e.u.2) (etaeu2) = fae esn.2) and the statement u, = d/az 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 from consideration of the integral § ay 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 S¥, (w ds) is inde- pendent of path (in D). Indeed, let C (Fig. 1.12a) be a simple closed path. Choose two arbitrary points M and N on C and write (u-ds) = [™ (w-ds) + [™ (eds) = [” (weds) — [” -c), e ae E 7 7 Along Cr Along C2 Along Cy Along C2 If the integral {if (u- ds) is independent of path, the right-hand side vanishes and the circulation 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 left-hand side is zero, so is the right-hand side, yielding the independence of path, If Cy and C2 intersect a finite number of 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.1.6 SCALAR AND VECTOR FIELDS 19 closed paths. In the rather exceptional case when C, and C2 cross each other an infinite number of times, a limiting process can be invoked reducing this case to the preceding one.* Remark. Within the context of the above discussion, it is emphasized that by g(x, y, z) we mean a well-defined single-valued functiont over the entire domain and nothing short of this requirement will suffice. In many treatmentst of the magnetostatic field H, ‘one introduces the so-called scalar magnetic potential X so that H = grad X 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 : “7 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 leads 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 shall 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 several different functions. t For example, Reitz and Milford, Foundations of Electromagnetic Theory, Section 8.8.” VECTORS, MATRICES, AND COORDINATES LB w fluite number of smooth portions, By a smooth surface we mean a surface toprerentuble by x= (7,9, Y= VW(P9), 27,9), where p and q 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 shall 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. It can be seen 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 ¢ + df and all particles of fluid which have crossed dS between ¢ and 1 + dt will be located at 1+ df 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 dS -|ul -dtcos @ = (u- dS) dt. Divide by dt and obtain the desired statement. By analogy with these observations, we define, in general, the flux of a vector field u through a surface S by the surface integral & = ff(u- 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 fu ds) = § (ued + tyd,). Integrals of this type can be analyzed by means of Green's theorem: If C is a (piece- wise smooth) simple closed curve in a simply connected domain D and if P(x, y) * For details, consult, e.g., Kaplan, p. 260 er seq. + Unless stated otherwise, it is conventional to take the direction of integration in such integrals, i.e., the orientation of ds, as counterclockwise.17 VECTOR FIELDS IN PLANE ri Simply connected domain Doubly connected domain Figure 1.15 and Q(x, y) have continuous first partials in D, then feac+ oar If (2-F)as where S is the area bounded by C. The importance of the requirement that C is 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 u(x, y) and uy(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 integrals [(u-ds) around each mesh in eS] 4 NM a AV Ny of [Pelt we _ SS 7 Figure 1.1622. VECTORS, MATRICES, AND COORDINATES 17 counterclockwise direction we can easily deduce that Fwd) = Ef (wa, mites (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 $ cua) = LEO Mas, All ‘meshes 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 . 7) lim £& asso AS = fx, y) exists and is independent of the shape of AS,* then the sum reduces to an integral and we have ¢ (u- ds) = |[ fos yds. ‘$ Therefore, it remains for us to evaluate 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 u, and wu, have continuous partials, then we can write (uz) (udp + (), (-)+ (ue ,o- 1 and ‘au, wo + 3) «-9+ (#4) 0-9 with the approximations being within the first order in |x — ¢| and |y — |-f Here P(, ») 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 § (us) = f ted + $ 4 dy, (ay) * Except that the largest diameter of AS must approach zero: the mesh should not become infinitesimally thin while retaining finite length. } By the theorem on existence of total differential, guaranteed by the continuity of partial derivatives.17 VECTOR FIELDS IN PLANE 23 we see that the following six integrals will be needed: ga, $a pxa, $ vax, fp xae, gry. The first two are zero, the second two are +AS and —AS, respectively, and the last two are zero. As a result, we have nw (Atty _ aus $ (wa) = x ayo Consequently, = tim F024) _ Bly _ due Sy y) = im, AS ee ay where the stipulation that the partials are to be calculated at P(é, 7) can be omitted since P is now an arbitrary point within C. We have then the result $ ued + yay) = IG@-s)« which is simply Green’s theorem in our notation. With regard to a vector field u = uzi + uyj, the function f(x, y) is known as the curl of u so that, by definition, ead): ‘We have then evaluated the expression for curlu in (orthogonal) cartesian co- ordinates in plane: Quy us Cu oye curlu = ox ~ ay Remark. Attention is drawn to the fact that, for a 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 + 4j and curl u 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 wis irrotational in a simply connected domain, then by Green’s theorem, it is a conservative field (gradient field) in this domain, ic., it has zero circulation. The converse has to be worded rather carefully: If u is a gradient field (namely, u = grad ¢), then it is irrotational provided ¢ has continuous second- order partial derivatives. * In a more elaborate nomenclature, curl u is called pseudoscalar due to its 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 Figure 1.18 Example. The magnetic induction field (B-field) due to an infinite current- carrying wire is known to be (outside the wire)* B= gol, 5) (MKSA units). In the xy-plane (as shown in Fig. 1.17), =—Bsing= —MY_ _ mol _y eos oe ee = abel x. By = Boosd = 5 wey This field happens to be irrotational everywhere except at the origin. Therefore $c (B- ds) = Oif C does not encircle the origin, but not otherwise. A function x(x, )) 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 eee Bo aR where Ris the radius of the wire. Here B, = —(uol/2eR*)y and By = (ol /2xR?)x. The field is not irrotational and cannot be represented as grad y anywhere. 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 & =f (uno) ds =f (a dn) taken over a curve C (not necessarily closed) with no being the unit normal to the curve and dn = no ds. This is illustrated in Fig. 1.18. * so is the unit vector that is tangential to the circle drawn around the axis of the wire.17 - VECTOR FIELDS IN PLANE 25 Exercise. For a flow of fluid in a plane, relate the integral © above to the amount of fluid crossing the curve C per unit time. Specify physical units of all the quantities used. In many applications, the flux through a closed curve is involved. Evidently if ds = dxi + dyj, then* dn = dyi — dxj and ¢. (u-dn) = ¢. (uz dy — uy dx). Setting P = —uy, Q = uz, this integral can be transformed by means of Green’s theorem so that = ff (4 fo fl + we) dx dy 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 (adn) = [| div u- dS, s where ' Otte | Oty div = 3 +5, is another function derivable from a vector field and is known as the divergence of u. While the above derivation is straightforward, it does not reveal the geometric (or physical) meaning of div u. 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 «day = yo Flue da) fw da) = AS, ‘a mites because the vectors dn 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, divu = lim 2S. sno AS Exercise. Derive the formula div u = duz/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 u and div u it is seen that both represent a new kind of derivative, namely, a derivative with respect to infinitesimal area rather than infinitesi- mal displacement: diva = tim £04), curlu = lim : asso AS aso It may be of interest to mention that the gradient of a scalar field y can be represented in a similar fashion; i.e., the following statement holds: . £e-dn dee i . Brace = yso AS Another interesting observation is that curl u can be related to an integral involving dn and div u to an integral involving ds. Indeed, the identities (u-dn) = (u X ds}, (u-'ds) = [dn X ul, are not hard to verify provided we treat the cross product of two vectors in plane as a scalar, which is the logical thing to do.* Consequently, the following statements also hold: _ yn Fld xu) Su X ds), curlu = in AS hae divu Ase 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 solenoidal 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 u through a closed surface S, because this integral is, perhaps, easiest to handle. It reads = fea) = ff (ods + u, dS, + u,dS.), where dS, dS,, and dS, are projections of vector dS on the coordinate axes. Integrals of this type can be handled by Gauss’ theorem: If S is a (piecewise smooth) closed orientable surface contained, along with its interior, in a domain D and if L(x, y, 2), M(x, y, 2), and N(x, y, 2) have continuous first partials in D, then aL , aM , aN ff was. + MdS, + NdS.) “ff (e+ Me a) av, s where V is the volume bounded by S. * A typical cross product of two vectors in physics is the torque IP = [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 ina 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 L, M, and N with the components of a vector field u = wei + uj + uk. The concept of divergence of u, : uz , Oy , dus divu= 52+ ‘ay * Oz is readily introduced so that Gauss’ theorem can be rewritten as (u- dS) = f[[divu- dv fof 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 all 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, ff ww- 48) = 5 HO) gy al bieke introducing, for an obvious purpose, the volume AV of each block. We now define the divergence of u by means of : . (u- dS) divu= tim 2-48), wee yo AV 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 Az by anc = (uz)p Ay Az, where (uz)p, is calculated at the center P of the rectangle. Indeed, for any point within ABCD we have up 2 (us)p + (), o- + (), @- 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 u, at the center of this face and is given by [ude + (), 2 ay as Similarly, the flux through the opposite face is -[oor~(@), oe 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 au (#), ‘Ax Ay Az. The fluxes through the other four faces are obtained in a similar way leading to the statement about the flux through a smalll parallelepiped AV: ce (Mz 4 My 5 dus yy = @& ae ay oe ae 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 $. (4 ds) = fue dx + uy dy + us de), and we shall assume that C is a (piecewise smooth) simple closed curve in space. It need not be a plane curve but we shalll assume that it can be spanned by a (piece- wise smooth) orientable surface S; 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 is illustrated 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 element (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) dx'+ M(x, y,z) dy + N(x, y, z) dz] -{f{er- 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 fu ds) = DEG B as, All meshes where AS is the area of a given mesh. As before, we calculate F(u-ds)_ 4, Suz dx + uy dy + uz dz) F (WAS) _ jim Le Ge + My dy + te dz), AS Asoo lim As—0 Considering a small mesh about a point P(&, n, ¢) we have, for each point M(x, y, z) on its boundary I (Fig. 1.21a), udu = de + (82) ce 9+ (28), 0-94 (M2) G0, ete oy * A rule well known from the study of magnetic fields.30 VECTORS, MATRICES, AND COORDINATES 18 (@) Case I (b) Case IT Figure 1.21 We shall need integrals like f, dx, f, x dx, ¢, ydx, ¢, zdx, and similar ones. These integrals can be evaluated by projecting our mesh on the coordinate planes. Suppose that we want to obtain the integrals of the type Srf(x, y) dx and $rf(% y) dy. Since the points M and M’ (see Fig. 1.21) have the same x and y, these integrals reduce to Sr f(x, y)dx and $r-f(x, y) dy except for the sign* which depends on the orientation of the mesh: ¢, Sy y)dx = + ¢. fx, y) dx, ete. The plus sign is for Case I (Fig. 1.21a), where the point M’ describes I” in counter- clockwise direction, the minus sign is for Case II (Fig. 1.21b), with M’ going clockwise.t In particular, note that ¢.dx =f xdx = 0 while . $9 ax = gp vax = FAS’, where AS’ is the area bounded by I’, i.e., the area of the projection of the mesh * The symbol fr, 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 M in the positive direction of circulation around the mesh.18 VECTOR FIELDS IN SPACE 31 on the xy-plane. Since the mesh is small, we have (Fig. 1.21) AS’ & AS cos (mo,k) in Case, AS’ & —AS cos (Mo, k)_ in Case II, (where no is the unit normal to AS). In either case, grax = —AScos (no, k). r Similarly, we can deduce ¢, x dy = +AS-os (no, k) . (this integral is also needed). Other integrals are evaluated in an analogous fashion. In particular, $y z dx and gy x dz require projection on the xz-plane, and so on. The net result of this calculation reads + ds) = Ou at iy + (uz _ dus i fo ds) as (2 -3 £05 0m) + (8 Fie) p £08 (Ro i) + & ue) 0s (no, k) Introducing the cur] of a vector in space (which is now a vector, in distinction to the plane case) by means of the relation . £(u-ds) tim £@-4) ~ Curt u- no), ie ase 7 we have the result tua (Ole _ aun), (Ate _ auc), , (ty _ aus) Cul Mis oye @ arnt Nar ear) ts ox soy) 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:* 4, (u- ds) = fftoma- as). As in the case of vectors in a plane, the vector fields satisfying div u = 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 (u = 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, it does not * The shape of the area element is irrelevant; because of this we can identify dS. in aS = dS.i + dS,j + dS.k with dy dz, etc.32 VECTORS, MATRICES, AND COORDINATES 18 follow that its circulation about an arbitrary simple closed curve in D is 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 coil of current-carrying wire in the shape of a torus. The B-field inside the torus is irrotational; indeed, curl B satisfies the equation* curlB = pod + cou (MKSA units) and there is no current density J and no displace- ment current in the interior of the torus; we assume a de-situation. However, the B-field at the center of each turn of the coil is known to be B = pon! yielding a circulation along the central circle C of the torus: $B: ds) = wont - 2nR x 0. Figure 1.22 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 curlu in space is also reducible to a volume derivative. This is in fact true and the formula reads . dS X ul = ti He x a, oan aT 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 #f{dS X u] from the top face involve only wzi and u,j and yield, in the usual notation, Jftes x ae acay((u + 8%) i (ut +88) i. Top face where P is the center of the parallelepiped. The bottom face will have dS with the direction reversed and du, dz\ . du, dz\ | [fs x ae —sear ((u- 35 9),3~ (w+ 522), pena face * One of Maxwell’s equations; see, e.g., Reitz and Milford, pp. 296-297.18 VECTOR FIELDS IN SPACE 33 Adding these, we obtain Ou Ona az) p Oz) il cen, 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 Hoas = lim 2 grad yp in ‘AV We shall conclude this section by mentioning some quantities obtained from the repeated operations involving gradient, divergence, and curl. First of all observe the identity curl grad g = 0 representing the statement that a conservative field is always irrotational (pro- vided, of course, that the second-order partials of ¢ are continuous, as mentioned on p. 23). The second operation of similar type yields the definition of the Laplace differential operator ¥? or simply the Laplacian,* ¥2p = div grad ¢, with the well-known expression in cartesian coordinates y= Oe Po ont +% a+ at 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 divcurlu=0. A more sophisticated argument of some interest is to take a closed surface S in a domain D where curl u is defined (and its components have continuous partials), Split S into two parts, S, and Sa, by a curve C, as illustrated in Fig. 1.23. By Stokes’ theorem, fu as) TS 7 eo * The symbol V? is related to the so-called “‘nabla” or “‘del” operator a a a x t5ay tae v capable of representing gradient, divergence, and curl by means of the notations grad y = Vy, divu = (V-u), and curlu=[V X ul}, which are sometimes quite handy.34 VECTORS, MATRICES, AND COORDINATES 19 the minus sign arises from the “wrong” orientation of Sz with respect to C. Consequently, (curl u- dS.) + [[ (curl u-dS2) = ¢f (curl u- dS) = 0 [foes 1 ler 2 ff (curtn for any closed surface in D. Applying the divergence theorem, we obtain, divcurlu- dV = 0. I) iv curl w Then, since this is true for an arbitrary volume V in D, it follows that div curl u = 0. Regarding the operation curl curl u, we can derive the following expression for the cartesian coordinates: curl curlu = grad divu — V°u, where the symbol V2u stands for the operation V2ui + V2mj + Vuk. This formula, however, is not valid for other coordinate systems.* Exercise. Verify the above formula using the expressions for divergence, gradient, and curl in cartesian coordinates. Figure 1.23 1.9 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 J, m,n. A well-known example is given by spherical coordinates r, 0, g, as shown in Fig. 1.24, along with the usual cartesian coordinates x, y, z * Of course, it is possible to define the operation V2u by (grad div u — curl curl u) in any coordinate system, but this is not done since V2u does not reduce to the operation ‘div grad” applied to components of u.19 CURVILINEAR COORDINATES 35 which are related to r, 8, 6 by x=rsin@cos¢, y=rsindsing, z= rcosé. In general, x, y, z can be thought as being functions of J, m, n: x= x(mn), y= y(mn), z= 2(hmyn). We shall assume that at least within some domain D in space, these functions have continuous derivatives and can also be solved for /, m,n: meurve 1=xy,2, Mu) m = m(x, y, Z), n= n(x, y, 2). 4 Observe that this implies that the Jacobian = 2% » 2) 7 cl (7, m,n) oO Figure 1.25 does not vanish.* Let us now choose a particular point M with cartesian coordi- nates (£, n, ¢). it can also be denoted as M(A, u,v), in terms of the coordinates I,m,n. If we keep m = y = const and n = y = const and change J, then we obtain a (smooth) curve passing through M which may be called the /-curve. Similarly, we can define m-curve and n-curve. This is shown in Fig. 1.25. Further- more, we can introduce unit vectors 1p, mo, and mo along the tangents to these curves (pointed in the direction of increasing /, m,n). This establishes a local system of axes. For convenience, the labels /, m,n are chosen so that (Io, Mo, No) form a right-handed triple. Our local systems possess, in general, the following features which distinguish them from the system formed by cartesian unit vectors i, j, and k. 1. The axes may not be orthogonal; moreover, the angles between the axes may change from one point to another. 2. The orientation of 1p, mo, mo (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 /, m,n may not be the length, and dl, 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 r = xi + yj + zk. Treating x,y,z as * See, e.g., Kaplan, pp. 31 et seq.{0 VICTORS, MATRICES, AND COORDINATES 19 functions of /, m,n we can write r= x(I,m, n)i + y(l,m, nj + 2(1, m, n)k. Changing r by dr amounts to changing x, y, z by dx, dy, dz which is, in turn, caused by changing /, m, n by dl, dm, dn. We have the following general relations: dx = a+ Xam +2 an, dy = 2 at + % dm + ° a, dz = Fat Zam + Zn, If we move along the -curve, then dm = da = 0, and dr becomes* (d8)\mn = (dx 1 + dy jE Wn = (i +3j + 3x) a This defines the derivative of r with respect to the parameter /: Or _ (dt)mn _ Ox, | dy, aa ~ ait aii ez + 5k By its very meaning, ar/al is a vector along the direction of Io. Therefore, lo can be expressed as _ or/al_ _ _(ax/alji + (@y/adj + (0z/aDk_, ° © or/all (ax/aly? + (ay/aly? + (z/aly= The quantity a l(ax\: (2y (#) iS (%) +a) + Xai has a simple geometric interpretation: The length of elementary arc ds produced when only I changes is given by ds = |(dt)mn| = hdl. In a similar fashion we deduce — (x/am)i+ (ay/am)j + (ez/am)k, , _ [[ax\? , (ay)? , (az\*, 2 hn m= (3) + (2)'+ (2) 9 = x/andi + (@y/ami + (e/a, 4, (#y a (ys 7 (y- Tig * WNaa) + Man) * Nan * We use the notation familiar from thermodynamics: (dt)m,, Means such dr where m and 7 are kept constant.19 CURVILINEAR COORDINATES 37 Suppose now that the triple Io, mo, no is an orthogonal triple. Then we must have the relations ax ax , ay ay , dz az ‘a am + af amt al am—% 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 does indeed vary from point to point for spherical and cylindrical coordinate systems. In fact, this is the characteristic property of curvilinear coordinate systems, as opposed to cartesian ones. Feature (3) is also illustrated by spherical and cylindrical coordinates where some of the parameters /, m,n represent angles rather than lengths. As a general rule, the elementary displacement dr decomposed along the local system of axes will read dr or or or i : ae + mdm + 5, dn = idl lo + him dm mo + thy dn - no. Let us assume that the local system is orthogonal;* then the element of arc is given by a simple formula, ds = |de| = Whi dl? + 13 dm? + 12 dn’. For instance, for spherical coordinates, h, = 1, hy = r, hy = rsin 6, and ds = Vdr® + r* do” + r? sin? 6.dp”. 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 a ax ae. , a +aitak grad g = then use 9¢ _ ae al , d9 am | dy on ax ~ af ax * am ox t Sn ax ott and also express i, j,k in terms of Io, mo,m. A quicker way is to utilize the statement (grad y- de) = de 20 a 4 2 am 4. 98 af + 3m am + an * The associated coordinates are then called orthogonal coordinates.MB VICTORS, MATRICES, AND COORDINATES 19 () Figure 1.26 The calculation of divergence can be also carried out starting from the general definition divu = tim 2028). ai 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 Ip-direction is given by uy + him dm + hy dn. As we subtract the fluxes through the areas M’N’P’Q' and MNPQ we must not forget that not only w;, but also Am and hy, are functions of /, m, and n.t 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 a a (Uthmhn) al dm da. Adding the analogous contributions from the other four faces and dividing by the volume of our volume element (which is fi dl hm dm lin dn), we obtain immediately a 1 divu = Fae 3 (ual) ton o (umbn ht) ton Zeutitn * Recall that dr is arbitrary; therefore, by setting dm = dn = 0 we obtain (grad ¢): = (1/h)@¢/AN, etc. We tacitly assume that the coordinates are orthogonal. + In cartesian coordinates fy = htm = hy = 1.19 CURVILINEAR COORDINATES 39 Example. In spherical coordinates we identify /, m,n with r, 6, $, in that order. Then hy = hy = 1, hm = ho = 7, hyn = hy = rsin 6, and divu = snd ee sin @u,y) + J crsin ou) + & Lew} This may be simplified and, if desirable, ultimately reduced to see ous 2) 1 due , cote Te Gus Ott ort rt Ur oo tare rsind o¢ The curl of u can be deduced from the circulation of u around the faces of the very same volume element. For instance, the face MNPQ yields (Fig. 1.26) P Mt a f (u- ds) +f, (us ds) = (unhn dw — (Unbin dn)atg = 3,7 Unlin) dn den, [way cus which is combined to form —(Umlm dm) pg + (Umm dm) un a = 5 Umlim) dnd, a a (cutl u): hm dim fy dt = {es (aha) = 2 (mn dm dn. This determines the component of curl u. The complete formula reads ic tin) — 2 (eimbin curl w= 51 32 (en) an Umtin){ to tik fe (eh) ~ 2 ¢ah3} mo a a + hi {2 cena) = om aro} No. Finally, the expression for the Laplacian V? is obtained by combining the formulas for gradient and divergence: oe 1 ‘imlin 2¢\ , 9 (lint de) , 2 (hihm d¢\) , Vie = div gradu = FG fas hi 2) + om ( Tim ie) . a( Tn 22} For instance, in the spherical system this reads (after trivial simplification) 1 a (2a 1a 1 ea 299 geeeeapecee Bese te 2 on 2¢r ae) + sind 00 (sin : 4) + ain? 6 ag?40 VECTORS, MATRICES, AND COORDINATES BIBLIOGRAPHY Apostot, T. M., Mathematical Analysis, Reading, Mass.: Addison-Wesley Publishing Co., 1957. Gotpstetn, H., Classical Mechanics, Reading, Mass.: Addison-Wesley Publishing Co., 1959. Kapian, W., Advanced Calculus, Reading, Mass.: Addison-Wesley Publishing Co., 1957. Kettoae, O. D., Foundation of Potential Theory, Berlin: Springer Verlag Ohg., 1929. Taytor, A. E., Advanced Calculus, Boston: Ginn & Company, 1955. Rerrz, J. R. and F. J. Mirorp, Foundations of Electromagnetic Theory, Reading, Mass.: Addison-Wesley Publishing Co., 1960. PROBLEMS 1. Let two vectors in a plane, ux and ua, be defined by the polar coordinates of their tips: (81,71) and (@2,r2). If uz = ui + uz is defined by (@3, r3), show how 63 and rg are related to 01,02, r1, and r2. 2. A triple vector product of three vectors is defined by the expression [u X [vy X w]. Show that for any three vectors the following identity holds: [u x lv x w]] + [v x Iw x ul] + [w x [fu x vI] = 0. Hint: Use the vector identity [a X [b X ]] = ba) — c(a-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 0G -3 DP, WEED, w-h-$9- 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 (namely those with respect to i, j, k) to the new ones (with respect to u, v, w). c) Evaluate, by vector-matrix multiplication, the new coordinates of the vectors a(0, 3, 2), b(—1, 4, —3), and ¢(2; —2, —2). Can you give a geometrical interpre- tation of the peculiar behavior of vector ¢? 4. a) Show that the triple product (p. 3) of vectors u(w1, u2, 43), v(v1, 02,03), and w(w1, W2, ws) can be expressed by the determinant det = (fu X vy) ).PROBLEMS: 41 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. c) Consider the matrices 100, v3/2 4 0 A= Of 0s 4 -Vv3/2 Oe 0 0 0 0 -1 -1 0 0 23/5 -3V3/10 $ c= o-1 oj, D= z ete vo/2 |. O01 2 4 0 and indicate which ones represent rotations. Also, describe the geometrical meaning of others. . Compare, in general, the jth matrix element of 4B with that of BA, for 3 X 3matrices A and B, Construct two noncommuting 3 X 3 matrices of your choice, ie., such that AB x BA. . According to the discussion on p. 4, the matrix cos@ sin@ —sin@ cosé cos 26 sin 20 —sin 28 cos 26 cos 36 sin 30 and A? = AAA = —sin 38 cos 36 and give the geometrical interpretation of this result. . Show that the matrix cos sing sing —cos¢ 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).)42. VECTORS, MATRICES, AND COORDINATES 8. Find the inverses of the following matrices by solving the equations B.A, = I or otherwise: 2-10 4 3 3 A= eee Az 0 -1/v2 1/2 |, 35014: 22/3 -V2/6 —V2/6 -3 -3 % 1 Ky ri ee eee ae 1 -1 oe 4% 2 0 1 Comment on the cases A, As, and Ag. ; 9. Let (x’, ’) be the coordinates of a point in a skew a cartesian system in plane. Let the x’ and y’-axes make angles @ and 8, 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’? + y'? + 2x’y’ cos (B — a) = R?. a x 10. Show that the vector v = 2i + j — 6k cannot be a expressed as a linear combination of the vectors . up=itj+2k, ue =3i-j, us = 2+k 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. 11. Evaluate the following integrals around the circle x? + y? = 1; use Green’s theorem it is convenient: a) $ (u- ds), where u = (2y? — 3x?y)i + (xy — x9)j, b) £(2x? — y8)dx + (3 + y3)dy, c) £(v- dn), where v = (x? + y?)i — 2xyj (dn is defined on p. 24). 12. Let F(x, y) = x? — »?. Evaluate a) J&B) (grad F - ds) along the curve y = x°, ») 4 eas around the circle x? + y? = 1. Here aF/an is the directional deriva- in 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 g = u. Can you find a vector field A such that curl A = u? 14. Prove the following identities for scalar fields f, y and vector fields u, v in space: a) grad (fp) = ferad y + grad f, yb) curl (fa) = fcurlu + [grad f X ul], ¢) div w x v] = (v- curlu) — (a- curly).PROBLEMS 43 15, Using divergence and Stokes’ theorems, if it is convenient, calculate the following integrals. a) ffs (u- dS), where u the origin, b) ffs (v- dS), where v = x°i + yj + 25k and S is the sphere as in (a), ©) ffs (x dy dz + ydzdx + zdx dy), where S is the sphere as in (a), 4) $r (w- ds), where u = —3yi + 3xj + k and Tis the circle x? + y? = 1 lying in the plane z = 2. 16. A flat disk rotates about the axis normal to its plane and passing through its center. Show that the velocity vector v of any point on the disk satisfies the equation + y3j + 2%k and S is the sphere of radius R about curly = 2w, where w is the angular velocity vector. 17. Consider a conducting medium with variable charge density p(x, y, z) and variable current density J(x, y, z). Let V be an arbitrary volume within the medium bounded by a (piecewise smooth) closed surface S. Considering the total amount of charge inside V and the amount entering it per unit time through the surface S, deduce that £ [[foss2av ~~ £f oen9-as) ; i With the help of the divergence theorem, deduce the so-called equation of continuity : op dvd + 37 = 0. 18. Using the techniques employed in Sections 1.7 and 1.8, outline the possible proofs of the following statements: grady = lim = (ina plane), grad = lim Hoas asso AS Dea aas tna 19. Evaluate the quantities f,, hy, he (see p. 36) for the cylindrical coordinate system. Using appropriate formulas from Section 1.9, write the expressions for gradient, divergence, curl, and the Laplacian in cylindrical coordinates.CHAPTER 2 FUNCTIONS OF A COMPLEX VARIABLE 2.1 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 i, and we write 7? = —1 andi = \/—1. If we allow i to be multiplied by real numbers, we obtain the so-called imaginary numbers* of the form bi (where bis real). 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, Gi(—4/ = BY-4)? = (— 12-1) = 12, (-5i? = (-5)27? = -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.t 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 obey appropriate algebraic rules. As will be shown below, the system of complex numbers is closed under addition, subtraction, multiplica- 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/s neither real nor (pure) imaginary. 44ae BASIC ALGEBRA AND GEOMETRY OF COMPLEX NUMBERS 45 2.2 BASIC ALGEBRA AND GEOMETRY OF COMPLEX NUMBERS If complex numbers are written in the usual form a + ib (or a + bi) then the usual algebraic operations with them are defined as follows. 1. Addition: {ay + ibi) + (az + ibe) = (a + a2) + (br + 2). 2. Multiplication: (a, + iby) @2 + ibe) = (@ya2 — b1b2) + i(aib2 + obi). The second rule is easy to follow if we recognize that the expressions a + ib are multiplied in the same manner as binomials, using the distributive and associative 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 each other.* Complex numbers of the form 0 + ib are then (pure) imaginary numbers. It is customary to write simply a+ i0 = a and 0+ ib = ib. Sub- traction of complex numbers can be defined as inverse addition so that if (a, + bi) — (az + ibs) = x + iy, then a, + iby = (x + iy) + (az + ibe) from which it follows thatt x=a1—4, and y= b,— be. An alternative is to form the negative of a complex number, —(@ + ib) = (—1)@ + ib) = (-1 + 0a + id) = —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: at th at be— id (ac + bd) + ibe — ad) . i i. + a (24+ @ #0). It is readily seen that the divisor can be any complex number except zero (namely the number 0 + 10, which is unique and is written simply 0). * In a more rigorous language, “the subset of complex numbers of the form a + i0 is isomorphic to the set of real numbers under the correspondence a + i0 © a.” 4 Its tacitly postulated that x1 + iyi = x2 + iya if and only if x1 = x2 and y= yo.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 b are identified. with components of a vector. Note, however, that the multiplication of complex numibers differs from the formation of dot and cross products of vectors. 2. The use of the symbol i and the related binomial a + ib is conventional, but not indispensable. It is possible to define a complex number as a pair of real numbers, (aA), obeying certain peculiar rules, e.g., the multiplication can be defined by (@1, b1)(a2, b2) = (a1a2 — bib2, a1b2 + a2b.), and so on.: It should be clear that the form a + ib is just a representation of a complex number. 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 bya single symbol z and write z = x + iy, then to each z there cor- responds"4 point in the complex plane with the abscissa x and the ordinate y. This idea also leads us to the trigonometric representation of a complex number: z= r(cos 0 + isin 6), Im Pyrextiy where A= \/x? +p? and tand = y/x. In this repgesentation r is unique (positive square root) but @ is not. A common convention is to demand that} Imaginary axis —r<0kn, Real axis along with the standard rule of quadrants, Figure 2.1 namely, @ < Oif y <0. The following nomenclature and notation will be widely used: If z=x+iy = r(cosé + isin) then x = Rez is the real part of z, y = Imz_ is the dmaginary part of z, r= |z| __ is the modulus of z, also known as the magnitude or absolute value of z, 6 is the argument of z, also called the polar angle or phase.t The number x — iy is called the complex conjugate of the number z = x + iy and vice versa. We shall denote it by z*. We can say that z and z* represent (on the complex plane) the reflections of each other with respect to the real axis. + Another commonly used convention is 0 < @ < 2m. + A more precise name for 8 would be the “principal value of the argument of 2” (see p. 57).pI 3 BASIC ALGEBRA AND GEOMETRY OF COMPLEX NUMBERS 47 Remarks 1. The quantity z2* is always a nonnegative real number equal to. are the same). ? (which 2. The quantity.z + 2* is always a real number, equal to 2 Re z or to 2 Re 2* (which are the same). 3. The rules (z; + z2)* = zt + z% and (ziz2)* = ztz2 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 (1-22) = Re (ziz2) = Re (2123), where it is understood that z; and z2 are vectors corresponding to complex num- bers z, and z2 respectively. The vector product can be obtained in a similar fashion: [21 X 22] = Im (2422) = —Im (2123). Exercise. Verify the validity of the above rules for scalar and vector products. In the theory of complex variables, the expression |z1 — z2| is often used. According to Fig. 2.2(b) this quantity (modulus of the complex number z1 — 22) is equal to the distance between the points z; and z2 in the complex plane. It follows that the statement |z — zo| < R (which often occurs in proofs of various theorems) means geometrically that point z is within the circle of radius R drawn around the point zo as a center (i.e., z is in the R-neighborhood of zo; see p. 16). The following two inequalities are easily proved from geometrical considerations: I [za + zal < eal + [el (A side of a triangle is less than or equal to the sum of the other two sides.) 2. [za ~ 20] > [lea — [eall- (The difference of two sides of a triangle is less than or equal to the third side.)AB FUNCTIONS OF A COMPLEX VARIABLE 7 Remark. It should be emphasized that inequalities can exist only among the moduli of complex numbers, not among the complex numbers themselves. A complex number cannot be greater or smaller than another complex number. Also, there are no positive or negative complex numbers. 2.3 DE MOIVRE FORMULA AND THE CALCULATION OF ROOTS While addition and subtraction of complex numbers are most easily performed in their cartesian form z = x + iy, multiplication and division are easier in trigonometric form. If z1 = r1(cos 61 + isin 0) and zz = re(cos 02 + isin 42), then elementary calculation shows that Z12Z2 = rira[cos (81 + 62) + isin @, + 62)] with the provision that if 6; + 2 happens to be greater than x, or less than or equal to —z, then the amount 2x should be added or subtracted to fulfill the condition —x < (0; + 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 8) (—e+8) 2 = eff and zp) = ef where 6 is a small positive number, are very close to each other. However, their images under the principal branch mapping, namely filer) = 12-92) and fy(zq) = e282), are very far from each other. On the other hand, note that the image of z» under the mapping f2(z), namely, falas) = eitx/24812, 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 zo and let f(zo) = wo. We say that f(z) is continuous at zo if* f(z) wo whenever z —> zo in the sense that given 8 > 0 (arbitrarily small), the inequality |f(z) — wo] < 8 holds whenever |z — zo| <€ holds, for sufficiently small ¢. It is readily shownt that if w = u(x, y) + iv(x, »), then the continuity of w implies the continuity of u(x, y) and v(x, y) and vice versa. * Also written as lim:_,:9 f(z) = f(Zo)- + For example, see Kaplan, p. 495.2.6 MULTIVALUED FUNCTIONS AND RIEMANN SURFACES ae 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 a curve 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. 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 starting 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 line 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 to 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 real semiaxis excluded) is mapped by the second branch onto the region Re w < 0. The line joining the second lower with the third upper quadrants is mapped (by the second branch) onto the negative imaginary semiaxis, In this fashion the entire Riemann surface$6 FUNCTIONS OF A COMPLEX VARIABLE 2.6 is mapped one-to-one onto the w-plane (z = 0 is mapped onto w = 0; this particular correspondence, strictly speaking, belongs to neither branch since the polar angle @ is not defined for z = 0). ‘The splitting of a multivalued function into branches is arbitrary to a great extent. For instance, define the following two functions which also may be treated as branches of f(z) = V2: Vr etal? for O<0<7, Vr ell@H20/2) for 4 << 0, W/rell@t20/21 for O< 9 0) or with (W/2")" (g < 0) and has q branches. For irrational a the power function is infinitely multivalued. 2.7 ANALYTIC FUNCTIONS. CAUCHY THEOREM In this section we shall discuss the subject of calculus of functions of a complex variable. The basic concept of the continuity of a complex function has already been presented, and it is not difficult to verify that the sum, product, and quotient (except for division by zero) of two continuous functions is continuous. A con- tinuous function of a continuous function is also continuous.f Let C be a piecewise smooth curve in a complex plane. If f(z) is continuous on C, then the complex integral [fara lc can be defined and expressed in terms of real integrals by putting S@ = ulx,y) + v(x y) and dz = dx + idy; this yields [,f@d = [ude — ody) + if (dr + udy), where the real integrals fo (u dx — v dy) and Jo (v dx + u dy) are known to exist.$ Curve C may be open or closed but the direction of integration must be specified in either 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: [,G@+ e@)d& = fleyde + fale) az, @ - Af(z)dz =k i flz)dz — (k = complex constant), Q) [foe =f, ; fleyde +f fla)de, 6) where C is decomposed into two curves, C, and C2. The absolute value of an integral can be estimated by the formula |f, Ae | < ME, le where M = max |f(z)| on C, and L is the length of C. * Another widespread notation is arc sin z = sin=! z, arsinh z = sinh~! z, and so on. + For example, Kaplan, p. 496. } In the sense of Riemann; see, e.g., 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 + Az) and we can write* S@) = Efe) = jim L242) — £0). 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 + Az 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 verify (by the same technique as for real variables) that the derivatives of complex functions obey the usual rules: Zon + me) = St 4 ee, @ E (oye) = avs + ee Was (2) & - * HE, where w = w(g) and f= $0), @) Ze = n2"-! — (n = integer), (4) 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’(z) dz. If we set f(z) = w = u(x, y) + io(x, y), then the definition of the derivative can be rewritten as “) = Ym Wt Amy + Ay) — ule y)) + ilo + Ax, y + Ay) — v(x, y)), 107 ax +i Ay ayo 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 cee £@ = + 5g * See p. 54 (including the footnote) for the definition of a limit. t See, e.g., Knopp, Theory of Functions, Vol. 1, p. 65.60 FUNCTIONS OF A COMPLEX VARIABLE Pag Alternatively, set Az = i Ay (approach along the imaginary axis); then ; ou EG) aay lay It follows that for a differentiable function w = u + iv we must have ou av, du _ av. Petey ey woe These are the Cauchy-Riemann conditions; they follow ditectly 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 u and v are continuous. The inverse theorem also holds: If u(x, y) and v(x, y) have continuous first partial derivatives satisfying Cauchy-Riemann conditions in some neighborhood of z, then f(z) = u + iv is analytic at z. Integrals of analytic functions possess some very important properties. Per- haps the most fundamental one is expressed by the Cauchy theorem: If f(z) is analytic in a simply connected domain D, and C is a (piecewise smooth) simple closed curve in D, then $, fz) dz = 0. Proof. Write the integral as ¢ f@)az = 4, (udx — vdy) + if (@dx + udy). c c c Analyticity of (2) implies continuity of partial derivatives of u and v and Green's theorem (p. 20) is applicable. However, then the Cauchy-Riemann conditions imply $ was — ody) = “fCR+3 22) dy = 0, fetes aan ff (- Been 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(z) dz = 0 for every simple closed path in D with its interior also in D, then f(2) is analytic in D. It is not difficult to see that the Cauchy theorem is true for multiply connected domains provided the interior of the simple closed path Cis 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. 1.15). Similar extensions hold for related integral 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 applied easily to complex integrals, leading to the statement: If $ f(z) dz = 0 for every simple closed path, then the integral [soar is independent of path (between zo and 2). Suppose now that the point zo is fixed. If the integral f°, f(t) 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(z)) as follows from the fundamental theorem of integral calculus: If f(z) is analytic in a simply connected domain D, then the function Fe) = [seat 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 (eu) Fa) = U+ i = [2 (ude — vay) + fe? (wart udy), 40 fe 0.0) cz both integrals are independent of path (by Green’s theorem and the Cauchy- Riemann conditions). It also follows that* i ee ee ee ee, tee ee oy so that u and v satisfy the Cauchy-Riemann conditions as well. Therefore F(z) is analytic. Moreover, qdF_ au, av nr tig -ut i= fo, and the theorem follows. Any two primitive functions must differ by a (complex) constant; this follows from the fact that f"(z) = Oimplies f(z) = const (integrate du/ax = 0, au/ay = 0, ete.). * For example, Kaplan, p. 244.62 FUNCTIONS OF A COMPLEX VARIABLE 2.8 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= gt dz (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 a derivative. Let the curve C involved in the definition of I be a simple closed curve. Then, if the point z = a is outside the curve, the Cauchy theorem holds and I = 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 C is 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 I= [via = dni. It is not difficult to show that the same result is true for any simple closed path C1, 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 B, and Be can be constructed to connect the interior of C, 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 By ofthe channel are allowed to approach each other, the integrals of £2) = 1/(z — a) along By and Bz will (in the limit) cancel out, leaving us with the statement ¢, Sf) dz + $0, flz) dz = 0. Counterclockwise Clockwise Reversing the direction of the second integration to counterclockwise, we obtain $ Se) de = $0, fle) de Figure 2.8 (with both directions counterclockwise). If Cis entirely within C,, 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 Cy intersect at two points.28 OTHER THEOREMS. CAUCHY FORMULA 63 If the integral J is evaluated around a closed path which is not simple, its value is not necessarily 2i. In cases of practical interest it will be equal to n2xi, where n 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 g f(z) dz may happen to be zero even if the Cauchy theorem does not apply. For instance, calculate the integral 1 r= fete where is a positive integer not equal to unity and the contour is a circle of radius Raround z = a. Using z = a + Re*, obtain +e - J -/ Recreate ds = i 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 fe along the unit circle about the origin. First of all, the branch of the (double-valued) function /z must be specified. Suppose it is the principal branch. Then f Vide =" elie dn = 1. The Cauchy theorem is not applicable because f(z) is not analytic within the circle lz| = 1. The points where it fails to be analytic are along the real axis from x = —I1 to x = 0 where f(z) is not even continuous. Note also that although F(Z is continuous at z = 0, it is not analytic at that point either. Consider now the same integral oe ee * One such example is given in Section 12.9.64 FUNCTIONS OF A COMPLEX VARIABLE 2.8 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 ¥/2 into the following two branches (as on p. 56): (\/z = v/r ef? if O<0 K for all n sufficiently large, and k > 1, then Dizn diverges. Root test. If Y[zq| < k <1 for all n sufficiently large, then Xz, converges absolutely, and if *Y[zm] > k > 1 for all sufficiently large, then Dz, diverges. Proofs 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 nth term test: If z, fails to converge to zero then the series Sz, 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 52m = C(Xn + in) converges to S = P + iQ if and only if xq converges to P and Cyn converges to Q. For instance, series which are: convergent but not absolutely convergent can be treated in this fashion. Complex series can be added 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.t * For example, see Section 6.4. t See Knopp, Theory and Application of Infinite Series, Section 45.ri ‘COMPLEX SEQUENCES AND SERIES 69 Terms of a complex series may depend on a complex variable z. Most common series are power series, for instance, l+zt27424- 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 inconclusive if |z| = 1, but then the nth-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 r, then it is absolutely convergent everywhere inside this circle* (by comparison test). The problem is then to find the upper bound of r which is the sought radius of convergence. Exercise. Show that the series 1 — 324 92? — 2723 + Bizt — --+ has a radius of convergence equal to 4, while the series Lz + 2tz? + 3123 + atzA +++ 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, ak zs Te ttt a Partial sums of a power series represent a sequence of polynomials in z. Sequences of other functions can also be considered. The sequence {f,(z)} 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 fa(z) = f(2) for each z in R. For instance, the partial sums of the series Saige te eye mo * This statement is known as Abel’s theorem. See Kaplan, p. 350.70 FUNCTIONS OF A COMPLEX VARIABLE ea form a sequence of functions (polynomials) : faz) = D2, & and this sequence converges to the function f(z) = 1/(1 — 2) in the (open) region |z| < 1 because gt mi Sl) = Vp zt 22 pop = and ntl z lim 0 for |2| <1. For this reason the function f(z) = 1/(1 — 2) is said to be the sum of the above series (for |z| < 1 only!): =1i-2 (<2). m0 When representing a function by a sequence of other functions, it is necessary to know how well a certain function f(z) is approximated by the nth term of a sequence {f,(z)}. This question leads to the definition of a uniform convergence: The sequence {f,(z)} is said to converge uniformly to f(z) in a region R provided that the inequality \fl2) — f@ <6 which is satisfied if n > N, holds simultaneously for all z in R. In plain terms: Let is 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 is convergent for |z| < 1 but it is not uniformly convergent; the convergence becomes increasingly worse as |z| > 1. However, in the region |z| < k where k <1, the convergence is uniform. It may be “bad” at z = k, but once N is2.10 TAYLOR AND LAURENT SERIES 7 found such that “ for all n > N and z = k, then the same value of N will hold for all other z with 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 © fn(z) of functions is uniformly convergent in a region R if there exists a series of positive constants M, such that \fa@| < Mn for allzin R and the series 2M, is convergent. The proof follows from the comparison test and the fact that the M,, are independent of z. Several important theorems can be established for a uniformly convergent series as follows. Continuity theorem. The sum of a uniformly convergent series of continuous functions is a continuous function. Integrability theorem. A uniformly convergent series of continuous functions can be integrated term by term. Differentiability theorem. A uniformly convergent series can be differentiated term by term, provided all terms have continuous derivatives and the re- sulting series is uniformly convergent. All these theorems can be proved by the same methods as those used for real variables.* One can also show that the sums and products of uniformly con- vergent series are uniformly convergent (within the same region, of course). From the above results one can deduce the Weierstrass theorem. If the terms of the series I f,(z) are analytic inside and on a simple closed curve C and the series converges uniformly on C, then its sum is an analytic function (inside and on C) and the series may be differentiated or integrated any number of times. 2.10 TAYLOR AND LAURENT SERIES Consider a power series of powers (z — a)" where a is a fixed complex number: co + ex(z — a) + co(z — a)? + e9(z — a)? + °-- If this series converges for some value zo # a (for zo = a the series always converges), then it is absolutely convergent everywhere in the interior of the circle * For these methods see Kaplan, pp. 345-348.72° FUNCTIONS OF A COMPLEX VARIABLE 2.10 of radius |zp — a| = Ro about the point a (by comparison test). Moreover, it will be uniformly convergent within a circle of radius R less than Ro (by the Weierstrass M-test). It follows that the above power series represents, within a circle R (at least), a complex function $0) = Zee — a. By the Weierstrass theorem, this function must be analytic within the circle. Summarizing these results, we can state that every power series with a nonzero radius of convergence represents a regular function in some neighborhood of the point z = a. Such series may be added and multiplied (where the neighborhoods overlap) and differentiated and integrated any number of times. The converse statement is also true: Every function S( analytic at z = a can be expanded in a power series f@) = = en(z — a)" r valid in some neighborhood of point a. This series, known as the Taylor series, is unique, and the coefficients cy can be obtained from the formula el fe) Figure 2.12 Proof. Let f(z) be analytic within a circle C about the point a and let z be inside C (Fig. 2.12). It is then always possible to construct a circle T such that I is inside C and the point z is inside I. This is necessary to ensure that f(z) is analytic on I’ and the Cauchy formula can be applied: Sf) 5, ate 2h The quantity 1/(¢ — z) may now be expanded by means of a geometric series: nao \S — @, This is allowable because the series converges (by ratio test). Then fey = ¢ AO ro $s F AOE =" arin e — 2% tao (F — att2.10 TAYLOR AND LAURENT SERIES rf The series appearing in the integrand, viewed as a function of §, is uniformly convergent on I’ and, by the Weierstrass theorem, can be integrated term by term: f= Ee - ot fp LO ae. nao T(t — ayett In view of the formula (p. 66) nt Fa = TL) we obtain f= ¥ EPM@e - oy" Zoi It remains for us to show that this power series is unique. Indeed, if there is a series with undetermined coefficients cn, f@) = Li ealz — a)", n= which represents an analytic function in some neighborhood of point a; then it is uniformly convergent within and on a circle [inside the neighborhood. Differen- tiating the series n times and setting z = a, we obtain 1 ny n= LOO, which completes the proof. Power series can be generalized to contain negative powers of (z — a) to read + XD enlz — a)”. Such series may be split into two parts: os = a Cn(z — a)” and me xy . z x, (Zz — ay" and the original series will converge provided both parts converge. The series of positive powers converges inside a circle of some radius Ry about the point z = a. The series of negative powers will, in general, converge outside a circle of some radius Ry about z = a. To see this, denote so that the series of negative powers of (2 — a) becomes cca re: ear74 FUNCTIONS OF A COMPLEX VARIABLE 2.10 Unless this series happens to have zero radius of convergence it will, in general, converge within a circle of radius R’ about the origin. But || < R’ implies |z— al > 1/R’ = Ra, and the statement follows. Therefore we conclude that if Rp > R,, then the series + Dez — a)” will converge within the annulus Ri < |z—a| < Ro. It can, of course, happen that Re < Rj, 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 < Ry can be expanded in a series of positive and negative powers of (2 — a), namely ts a: f= Veale — a". This series, known as the Laurent series, is unique for a given annulus, and the coefficients c, can be obtained from 1 ¢ fede, * Oxi In @ — ati where I is a circle of radius R such that Ry < R < Ro. Figure 2.13 Proof. Contract the radius Re slightly and expand the radius Rj slightly to. obtain an annular region, with point z inside, to which the generalized Cauchy formula is applicable (Fig. 2.13):* Loa 4 1 gf fat, 2f—2z' Mitr, F—z f= 4 The first integral can be treated as in the derivation of the Taylor series: Lg Mot _ 1 f§ ME-o yp _ Se_ yi dG Soa ai -ah4 d= Ze - V5; Qi Sr, fz “GC —aett Tt) ( — ayeri Term-by-term integration is permissible by uniform convergence. * We shall occasionally use the convenient symbols ¢ and § to indicate clockwise or counterclockwise direction of integration.2.10 TAYLOR AND LAURENT SERIES 75 The second integral is treated by expanding 1/(¢ — z) in a somewhat different geometric series: 1 1 s (= a)” tae a Pw TSeSae= a" © pinto @ = art which is convergent by ratio test. Then fod _ _l@gy foe nf—2 oni i —— ere Lf, FUME = ay" ds. m=0 oe Im Replace m by —(n + 1) (m must be negative) and rewrite the above as feyat _ Ee- rig Sey dt Win FZ Ga Finally, note that the integrals ¢ LOU = 01,2...) 1 ( ae and d (oo may be just as well evaluated over a common circle I’, concentric with T and T'2, and lying within the annulus Ry < R < Re. To prove the uniqueness, assume that an expansion +e fe) = Dente — a" exists and is valid in the annulus R; < |z — a| < Re. Choose an arbitrary integer k, multiply both sides of this expression by (z — a)~*~1, and integrate around a circle T about z = a, lying within the annulus. Then $ fad _ & 4 dz : Ir (2 — aye Op @ — ati All integrals on the right-hand side will vanish except one, for which n = k, and whose value is 2ri. Therefore ¢ Oe = epri, Z — ayert 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 the nth coefficient cannot be associated, in general, with f(a) because the latter 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 only for a specified annulus. In general, a func- tion f(z) may possess two or more entirely different Laurent series about a given point, valid for different (nonoverlapping) regions. For instance, 1 2 —2) 1 A aptitet 2 thee, O lal. Write f@) = = If |z| > |a|, then |a/z| < 1, and we can expand ar E(Y cm2.10 TAYLOR AND LAURENT SERIES tae Therefore ee =Lsn (> la). n=0 This is the desired Laurent series. The function f(z) can be expanded by this method about any point z = 6: Indeed, write 1 1 $0 Fa" @=H-@-H tas) OM Then, either eee eS ee I9-~ GHG” @- DAG (2-6 <|e— Bp, er © (a— 5)" f= assy (lz — b| > la — 4). Example 2. Rational fraction decomposition. 1 JO- 2 = O4 FH The roots of the denominator are a = i, b = 2 (simple and distinct). There- fore f(z) fails to be analytic only at z = i and 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-Q+iz2+2%=@-)E-D and 1 f2)= GoNE=D =~ TR Suppose that the Laurent series valid for 1 < |z| < 2 is desired. The function 1/(z — 2) should then be expanded in Taylor series about z = 0 (Example 1). This series is, in particular, valid for 1 < |z| < 2. The function 1/(z — i) should be expanded in the Laurent series about z = 0 valid for |z| > 1 (Example 1). This series is also valid for 1 < |z| < 2. If these two series are subtracted, we may obtain (multiplying by 1/(2 — i) a series for f(z) valid for 1 < |z| < 2 which is the desired Laurent series. Example 3. Differentiation. f(z) = 1/(z — 1)?. 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 canis ae) @- 1p a\i—z8 FUNCTIONS OF A COMPLEX VARIABLE at The series (a <1) =0 can be differentiated term by term within the circle of convergence. Therefore ec Lh 2t3h tes Vit ie (<0). Example 4. Integration. f(z) = log(1 + z) = log |1 + 2| + iarg (1 + 2). This is the principal branch of the (multivalued) logarithmic function. The branch line extends from “minus infinity” to minus one and log (I + 2) is analytic within the circle |z] = 1 We know that 1 142° d qe +2)= Therefore we may expand Wdta)slozt+7—-F4A—---= VC" (A M (any M) for |z — a| < € (some 6). 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 = 0) are Casel, f(z) = sin z/z, Case 2, f(2) = I/sinz, Case 3, f(@) =e". 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) = lim, f(z), then it must be regular at z = aas well. Remark. The formula f(z) = sin z/z does not define, in a rigorous sense, the value of the function at z = 0. The extended formula (sin z/z, z #0, ro ~{ 1 0. does define the function f(z) 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 fol fisds A gy fide a re—z } lity fz We shalll show that the second integral must be zero for all p. If this is so, then f(z) must approach the limit 1 GMO, mi cca lim fle) = which will prove two things at once: (a) lim... (2) exists, (b) if f(a) is defined by this limit then the rede- fined function f(z) is analytic at z = a as well.80 FUNCTIONS OF A COMPLEX VARIABLE 2.11 To achieve this result, write ¢ — z = (¢ — a) — (z — a) and observe that I@ - a) — @ — @| 2 |z — al — |¢ — a] = |[z - | — pp. Then, for a fixed z, 1 f Aya] 2ni Jy $= 1 B = Bo. is Boal —p ~ al —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 formula is tacitly replaced by the corresponding limit. In this sense, the functions f(2) = sinz/z, gz) = e"*"*, (2) = 1/z — cotz are analytic at z = 0. The second type of isolated singularity, when |f(z)| > 0 as z— a, is called a pole. Since the singularity is isolated, there must exist a Laurent series f= XE ele - a" valid for 0 < |z — al < R (for some R). If the principal part terminates, ie., if the Laurent series is of the form + f= E ole — ay then f(z) has a pole of order m at z = a. Conversely, if f(z) has a pole at z = a, it must have the Laurent series (for 0 < |z — al < R) of the above form as seen from the following argument: Consider the function a2) = 1/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 le@@|70 as za (because |f(2)| > o). Therefore g(z) has a zero at z = a. This zero must be of some definite order m, namely, s@= > be- a" OX zal Goa TS Since ¥(2) is analytic at z = a and does not vanish there, it follows that 1/4(z) must be analytic as well and possess a Taylor series at z = a. Then f2) = ———~ = Cr However, this is evidently a Laurent series with a terminating principal part so that the argument is complete. Example. The function f(z) = csc z = 1/sin z has the Laurent series valid for O0 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)| —> 00; if it approaches zero along the imaginary axis, then | f(2)| remains constant but arg f(z) oscillates, and so on.82 FUNCTIONS OF A COMPLEX VARIABLE ae 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 rd’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”:* a 1 7 A f= Go tenis tenet is that the above 1| > 1. It actually represents the function does not mean that f(z) has essential singularity at z = 1. The poi series converges only if |2 1 eee ee fe) in the annulus 1 < |z — 1| 1 and represents a branch of the function /z? — 1 which is analytic in this region. The branch line joins, in this case, two branch points z = +1 andz = —1 and does not extend to infinity (Fig. 2.15). Replacing * Namely, for all points in a neighborhood |z — al < ¢ except zcee THE RESIDUE THEOREM AND ITS APPLICATIONS 83 z by (z — 1), we obtain a Laurent series centered about the branch point z = 1. The latter will converge for |z — 1| > 2. Another type of singular behavior of an analytic function occurs when it possesses an infinite number of isolated singularities converging to some limit point. Consider, for instance, 1 1 S@) = 0802 = Sa" The denominator has simple zeros whenever Branch line (n = +1, +2,...). Figure 2.15 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 least one pole (actually an infinite number of them). 2.12 THE RESIDUE THEOREM AND ITS APPLICATIONS Let f(z) be analytic in some neighborhood of z = a except, perhaps, at z= @ 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 f(a) = a ¢ fz) dz is independent of the choice of C and is called the residue of function at the point z= a. Evidently, if f(z) is analytic at z = a (the point z = a is then called a regular point), the residue is zero. If z = @ is an isolated singularity, then the residue may or may not be zero. Examples 1. f(2) = 1/2; the residue at z = 0 is equal to unity. 2. f(2) = 1/z*; the residue at z = 0 is equal to zero. According to the formula on p. 74, the residue is seen to be identical with the coefficient c_, in the Laurent series +0 SO) = Lentz — 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 a finite number of isolated singularities at z = a1, a2, ... , dy, which are all located inside C, then §, J@)de = 2xi ¥ Res lar). 6 a This theorem is proved by the technique of cutting the channels between the con- tour C and small circles C,, C2, ... around each singularity (Fig. 2.16). Re Figure 2.16 Figure 2.17 There is a variety of practical methods for quick evaluation of residues: FIRST METHOD. From the definition, Res fla) = 4 ¢ fade (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. f(2) = 1/2, F(2) to preserve the relationship Log z. Any branch of Log z may be chosen, but in order aF() , f@)= Fe = the closed contour must be disconnected and the appropriate limiting process must be applied. For instance (Fig. 2.17), A $ hae = lim tim (log (A) — log (B)} = 2t. loz poz Rea BoA Here the principal branch was used, which possesses a discontinuity 2ri on the negative real semiaxis.2.12 THE RESIDUE THEOREM AND ITS APPLICATIONS 85 SECOND METHOD. For a simple pole at z = a the following formula holds: Res f(a) = lim (z — a) f(z). The limit involved is often obtained by simple substitution or the use of well- known limiting values. Example.’ f(z) = tan z/z2. Then sinz 1 iim 290 Z COSZ tan z Res f(0) = lim 20 THIRD METHOD. For a pole of order m at z = a the following formula holds: Res f(a) = x im {= = Esta — a)" f2y- Example. f(z) = e/24. Then a 1 Res f(0) = 3 zim ae A = Gili e* ae FOURTH METHOD. A common case of a simple pole is when f(z) has the form eZ), f2) = 4G’ where g(a) x 0 and ¥(z) has a simple zero at z = a. In this case Res f(a) = ge. Note that if z = ais a simple zero of y(2), then y’(a) cannot vanish. Example. f(z) = e/sin z. Then Re 10) 5. cos z 2| oF 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, f(z) = e#/(e + 3(z — 1). The residue at z = 1 is desired. In the first step, transfer the pole to the origin: z-l=u zeo+l. Then ef f=86 FUNCTIONS OF A COMPLEX VARIABLE. 2.12 In the second step, expand e“ and 1/(3 + 2 2 MeL wth wot Sul 4 (allw), 1 tae 1 eo, he thee e | (lel <3). In the third step, evaluate (by inspection) the coefficient with w* from the product of the above two series: In the fourth step, evaluate the residue: Exercise. Prove the validity of the third method given above. (Hint: Represent /(2) as I@) = e@/@ — a)” and use the formula from p. 66.) 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 1 Toapeossy pt “lO. This integral can be converted into a contour integral in complex plane by setting z= e. Then dz 1 i wa cxo-}(e+!) and PM = pa)’ where C is the unit circle in the z-plane. The integrand has two poles: at z = p and z = I/p. If |p| < |, the pole z = p is inside the contour, while the pole z = I/pis outside. Only the residue at z = p is needed; it is equal to therefore (lal < 1.2.12 THE RESIDUE THEOREM AND ITS APPLICATIONS 87 If |p| > 1, the needed residue is at z = 1/p and it is equal to yielding (lp| > 1). Note that both results can be combined into (al # D, while the integral is not defined for |p| = 1. ‘i This method can be used for integrals of the type Jo” R(cos 6, sin 6) dé, where R(cos @, sin 6) is a rational function of cos @ and sin 6. Example 2, Consider the real integral te oy [7 wan inf wea @>% The integral [*y dx/(x? + a?) can be treated as a portion of the complex integral $c dz/(z? + a*) evaluated over the contour C shown in Fig. 2.18. Indeed (set z = x on the real axis): dz ae dz Let us estimate the integral over the semicircle Cp when R is very large: Write _!_.1_3 _. Py@ Pitaee If |z| = R is very large, then |a*/z?| = a/R? is small and || + a?/z*| is almost equal to unity (Fig. 2.19). To be precise, observe that |1 + a?/z?| > 4 for z-plane r+ Re Re Figure 2.18 Figure 2.1988 FUNCTIONS OF A COMPLEX VARIABLE 2.12 R > ay/2, and consequently 1 . Tre <2. (for R > av2). This implies (for R > av/2). 2 a+ SR Now, employ the estimate (p. 58) [2 ca z+ a S rR max < aR & = i + a R Then . dz Hal a Bea ~% Observe now that the integral fc dz/(z? + a*) is independent of the radius R (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, . —— = 2ni Res flai) = anit = 5 (for all C such that R > a). Consequently, if we let R 00, we have +R $ -# = tim [ ax y im f dz cP +a peJg Pte) Rm Jege te which reduces to r_ [*_ ax i- [ata @>o Exercise. Show that if a < 0, then : f os me X82 + a? a [Hint: No new calculations need be done, just some logical deductions.] The above procedure can be applied to integrals of the type P(x) dx, Cc Ox) where P(x) and Q(x) are polynomials in x and (a) Q(x) should have no real zeros* * If Q(x) has real zeros, then the integral is not defined (see p. 63), unless particular modifications are made (e., p. 111).2.12 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 Cg may not tend to zero). For such integrals it is true that ic FO ax = 2ni DY Res, s + where 54 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: If f(z) is continuous for Jz| > Ro (some Ro) and |zf(z)| > 0 uniformly as |z| > co, then lim Siz) dz Rood ek Proof di 2nR Fogle =|f,,_ aH] $ may son Uniform convergence of |zf(z)| means that |zf(z)| < (for arbitrarily small ¢) whenever |z| > R (that is, independently of the manner in which z approaches infinity). Then if f@) ae| < 2xe, and the theorem follows. The conditions of the theorem are satisfied by the function fl) = ae (with deg Q > 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 i * Ax) dx, 0 where f(x) is an even function of x. Example 3. Consider the real integral Note, first of all, that cos x dx “J. +a" 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 bounded in the upper half-plane because e* = e~Ye"* and |e*| = 1 (all real x)90 FUNCTIONS OF A COMPLEX VARIABLE mie while |e-¥| < I (all nonnegative y), For this reason, consider the complex integral etd: _ [eax etd coPe a lp wtat|,.2¢ a J= evaluated over the contour shown in Fig. 2.20. Observe that : eit dz tie iL Boca J = 2ni Res f(ai) = wi TG (as in Example 2). Also, ai so that E eM dx cos x de pine de i [cree "Ja Pa? a® (@ > 0). Since the right-hand side is real, it follows that : 1-[ Remark. The statement could have been made immediately on the grounds of symmetry. However, if it were not true, we would have obtained the value of this integral as well from the imaginary part of 2ri ° 4 Res. Example 4. Consider the real integral As before, observe that +° 1 sin x : if me Since sin z is not “well behaved” on the upper half-plane, we shall try to evaluate the complex integral dz. Zz 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.fae THE RESIDUE THEOREM AND ITS APPLICATIONS 91 Im In Cr Cr Cr Rea tRRe xR i ee Figure 2.20 Figure 2.21 To be able to apply the residue theorem we must avoid the pole at the origin in some fashion. Let us suppose we do this by means of a semicircle of a small radius r in the upper half-plane, as shown in Fig. 2.21. Then we can write is —T ie iz ee jie | es a é ee fia [Sat [ Serf Sat dz. The reason the chosen contour C is helpful in the evaluation of our real integral is as follows. Note that (by continuity of sin x/x) * sin ~ sin * sin * si f IDX ax = tim f x dx + tim f nS dx + tim f nO ax. 0 x rola Xx 190 roar x Now, since evidently fe (aoe i in x/x) dx 0, rao dar it follows that * sin ~ sin ** sin [ SOX dx = lim f “2* dx + lim SOX ax. see ro Je X ro dtr x We can now see that the imaginary parts of the first and the third integrals on the right-hand side of the relation ta fat te iz +E ie i e e e e e Ser a = dx sa dx dz fs Gs Re. J. ee will give us the desired result in the limit where r—> 0 and R— oo, provided we manage to calculate the other three integrals in the formula. Let us estimate the integral over Cp. The method of Example 2 fails here; however, we can perform integration by parts:* e Ca. cn iz * From the results of Section 2.7 it is trivial to verify that integration by parts applies to complex integrals as well as to real ones.92 FUNCTIONS OF A COMPLEX VARIABLE 2.12 If R— ~, both terms on the right-hand side approach zero (cos R is bounded because R is real), Next, we evaluate the integral over C: ef lte*-1, fa Ree Since e* is continuous at z = 0 and is equal to unity there, it follows that le* — I] 0,R> «© zero (no poles of e“*/z within the contour). Taking the limits ither order), we obtain de sinx dx — wi. =f sind gy =F ee Therefore* * Of course, this integral can be evaluated by much more elementary methods. The pur- pose of the above analysis is to illustrate the techniques of the residue calculus on a simple example rather than to obtain this particular result.ae 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 i oo ee ra fac tae ae Geeta eeaeeeaesage Here we treat our integral as a complex integral over a path (real axis) which is open (so far). Since sin z/z is continuous at z = 0, we may deform the contour as shown in Fig. 2.22 and claim that . T= tim | S22 a, | rao dor z Now set ee oes dee Sao), i" Figure 2.22 The problem is to evaluate =lim | “dz and fy =lim rao Jor Z 130 JC For 1;, close the contour as usual (Fig. 2.23a). Show that the integral over Ce approaches zero, and deduce that I For Ia, close the contour through the lower half-plane as shown in Fig. 2.23(b). Now |e~| is bounded in the lower half-plane and the integral over C’p 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 i ~ dz = —2ri Res f(0) = —2r1. Im Im G a Re CR 4 Re aw (b) Figure 2.2394 FUNCTIONS OF A COMPLEX VARIABLE ie Combining these results, we deduce that +0 i 1 1 [ Se dx = 35 (h — be) = 50 + 2m) = + in conformity with the previous result. Remark 1. The integral over the semicircle C, in Example 4 has been shown to yield the value —i in the limit r + 0. This is just one-half the value obtained by integration over the full circle. One may prove a general theorem to this effect: Let f(z) be analytic at z = a, Consider the integral Le [ ** (2) dz Z2 ,zZ-a taken from z1 = a + re to z2 = a + re along the circle = r (Fig, 2.24), Then le- lim I, = ai f(a) 0 Figure 2.24 where a = 92 — 61 + 2mm (choose » so that la] < 2r). Exercise. Prove this theorem using a technique similar to that in Example 4. ‘Remark 2. The statement that the integral of e'*/z over Ce in Example 4 vanishes in the limit R —+ © is also a special case of a more general result, known as Jordan’s lemma: If f(z) converges uniformly to zero whenever z approaches infinity, then lim S2)e™* dz = 0, Roe Jog where d is any positive number and Cx is the upper half of the circle |, The term “uniform convergence” as z—> © means that |/(2)| < « whenever |2| > M (for some M) no matter what the phase of z is. Example 6. The integral reads (-1 oo. It follows that J, is bounded by, say 2e~**®®, and this approaches zero since |a| 0) is the principal branch, with k = 0. It is convenient to work with the branch for which 0 0 and with the branch k = 1 for Imz <0. Then our real integral I 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 ° 2 ree ears 0 I+z o I+x dz [ ote THz tle, t+z— Therefore, 2ni Res f(—1) = [l — enemy gf Now, for [zh = R, Zo dz ar fg oe ee Gtk) 27" dz| pa @ [at war) = In 30 (asr—0). ‘The residue at z = —1 is e@~*; therefore ee — ei d2* ~ sin ow * In the limit « — 0, r — 0, and R > &, of course.2.13 CONFORMAL MAPPING BY ANALYTIC FUNCTIONS = 97 Im Im ae >Re Poe >Re zo 7 Figure 2.26 Figure 2.27 Remark. Itis also possible to operate with the standard principal branch —w < Argz Re Figure 2.29 w-plane zplane Note that the function f(z) = z? maps half the z-plane onto the entire w-plane. If the points on the real z-axis are included then the real positive semiaxis in the w-plane serves as a double image (for a single image it is necessary to exclude either the real negative or the real positive semiaxis in the z-plane). It is, perhaps, most convenient to say that w = z? maps the open upper half- plane (the region Im z > 0) onto the open region consisting of the w-plane with the real positive semiaxis removed. This is usually referred to as the “w-plane with a cut along the positive real semiaxis.” Now consider a point z == Zo in the z-plane and a smooth curve C passing through it. Let f(z) be analytic at z = zo and let f(z) map point zp onto wo and the curve C onto curve I’ passing through it (Fig. 2.30). Also consider a point z, on C, close to zo, and its image w; on T. Denote z — Zo = Az, w1 — wo = Aw. By definition of the derivative, ae Jim 3% = FG). Therefore Jim aay = God Geometrically, the left-hand side can be interpreted as the magnification of an (infinitesimal) arc of curve C as the curve C is transformed into the curve I. This magnification is the same for all curves passing through Zo. Now consider the angles @9 and ¢o which the curves C and T make with the real axis. We have ¢0 — 99 = lim [arg Aw — arg Az] = lim (xs ax) = arg f(Zo) = nia Pay az Geometrically, this means that the curve C is rotated through an angle a when it is transformed into I. This angle of rotation is the same for all curves passing through zo. Observe, however, that these two geometrical statements lose their meaning at the points zo where f"(z9) = 0. The properties described above are usually referred to as the conformal mapping properties. Customarily the term “conformal mapping” is defined by:2.13 CONFORMAL MAPPING BY ANALYTIC FUNCTIONS 101 Im Im y ic i ia! 2 4 — z-plane “ w-plane wal Re Re Figure 2.30 a) Invariance of angles: An angle at zo formed by two curves remains unchanged (in magnitude and orientation). b) Invariance of infinitesimal circles: An infinitesimal circle around zo retains its shape; it differs from a circle by higher order infinitesimal than its radius. Our analysis leads then to the following theorem: If a function f(z) is analytic at 29 and the derivative f’(zo) does not vanish, then the mapping z—> f(z) is con- formal at zo. ‘Remarks 1. The statement f”(zo) # 0 guarantees the existence of the inverse mapping because if = f(z) and z = g(w), then g’(wo) = 1/f’(zo) provided f"(zo) ¥ 0. 2. The mapping f(z) = z*, which reflects the z-plane in the real axis, preserves the angles and maps circles into circles but it is not considered conformal because the orienta- tion of the angles is reversed (Fig. 2.31). Im Im * G t v a LY TT Ty >Re +Re Conformal mapping Im Gy Im v C Re +Re n Figure 231 Mapping f(z)=2* r102 FUNCTIONS OF A COMPLEX VARIABLE 2.14 2.14 COMPLEX SPHERE AND POINT AT INFINITY Many concepts of the theory of complex variables are greatly simplified through the introduction of the so-called point at infinity. This is done with the help of the stereographic projection between the complex plane and the complex sphere which is defined as follows: Construct a sphere of radius R (for convenience, R may be taken as 4) such that the complex plane is tangential to it at the origin, as shown in Fig. 2.32. The point P on the sphere opposite to the origin (called the north pole for convenience) is used as the “eye” of stereographic projection. Lines through P are drawn which inter- sect both the sphere and the plane per- : 7 mitting a mapping of point z on the plane image of onto ite point ¢ on the sphere (Fig. 2.32), "™*sinary XS In this fashion the entire complex plane is mapped onto the complex sphere (or Riemann sphere). Curves and regions in the z-plane are mapped onto curves and regions on the ¢-sphere. Note that the point P itself has no counterpart on the z-plane. Nevertheless, it has been found convenient to adjoin to the z-plane an extra point, known as the point at infinity, in such a way that a curve passing through P on the ¢-sphere is defined as approaching the point at in- Figure 2.32 finity in the z-plane. The concept of the point at infinity is very useful, particularly in the analysis of mappings. The following statements of properties can be verified without much difficulty. 1, Circles in plane are mapped onto circles on the sphere which do not pass through P. . Straight lines in plane are mapped onto circles on the sphere which do pass through P. 3. Maps of intersecting straight lines have two common points on the {-sphere, one of which is P. North pole P Image of real axis, The origin 0 N 4. Maps of parallel straight lines have only the point P in common and they have common tangent at P. 5. The exterior of a circle |z| = R with R >> 1 is mapped onto the interior of a small spherical cap around point P. As R— o the cap “shrinks to P.” These and similar properties give rise to a series of definitions applied to the so-called extended complex plane, i., a complex plane to which the point at infinity is adjoined.2.14 COMPLEX SPHERE AND POINT AT INFINITY 103 Examples a) The mapping w = 1/z maps zo = 0 onto wo = e (point at infinity adjoined to w-plane) and vice versa. The rigorous meaning of this is, of course, that if a sequence of points in the z-plane converges to zo = 0, then a correspond- ing sequence of points on the w-sphere converges to its north pole. b) The region |z| > R is a neighborhood of the point at infinity. The importance of the complex sphere is greatly enhanced by the fact that if two curves intersect in the z-plane at an angle Y, then their images on the sphere will intersect at the same angle. In fact, the stereographic projection is conformal. (The proof is not difficult, but will be omitted here.) This permits the definition of the angle at infinity which two curves make if they recede to infinity in the z-plane. This angle is defined to be the angle that their images on the sphere make at the point P. Now the following theorem can be stated. Theorem. The mapping w = 1/z is conformal at the origin z = 0. Observe that the function f(z) = 1/z is not defined at z = 0 but the mapping w = 1/2 is defined (by the subterfuge of the sphere). The conformality does not follow from the analyticity, but rather from the relationships on the complex sphere. Corollary. The mapping w = 1/z is conformal at infinity (despite the fact that f”(z) approaches zero as z recedes to infinity). The concept of the point at infinity is closely interwoven with the study of singularities of analytic functions. The very notion of analyticity can be extended to the point at infinity by the following device: A function f(2) is defined to be analytic at infinity if the function az) = f(l/z) is analytic at z = 0. Moreover, it is possible to introduce the concept of a pole at infinity, branch point at infinity, etc., through the corresponding behavior of g(z) at the origin. In this connection, the function f(z) = e, which has no zeros and no singularities in the entire complex plane, turns out to possess an essential singularity at infinity. Other functions which have no singularities (e.g., all polynomials in z) are also found to have a breakdown of analyticity at infinity. Exercise. Show that a polynomial P(z) of degree n has a pole of nth order at infinity. In fact, a survey of familiar functions reveals the fact that functions which are analytic at infinity possess at least one singularity elsewhere, i.e. for some finite value of z. The natural conjecture is that there may not be a “perfectly analytic” function. This problem has actually been resolved and is embodied in the Liowville theorem. The only function f(z) which is analytic in the entire com- plex plane and the point at infinity is the constant function f(z) = const.104 FUNCTIONS OF A COMPLEX VARIABLE 2.15 In conclusion, it may be mentioned that in some texts the term complex plane is tacitly assumed to mean the extended complex plane, with the point at infinity included. Certain theorems may then be more conveniently stated. However, one should never forget that while there is a point at infinity, there is still no such thing as a complex number “infinity,” in the sense that it should possess the algebraic properties shared by other complex numbers. 2.15 INTEGRAL REPRESENTATIONS It is a very common occurrence that certain functions are represented and even defined by integrals, with constant or variable limits. For instance, consider the real function, the derivative of which is equal to sin x/x. It is known that the indefinite integral of sin x/x is not expressible in terms of elementary functions* in finite form. Therefore, there is no other alternative but to define such functions by means of an integral and it is customary to write Si (x) -[ ant a, where Si (x) is a new function, called the sine-integral. It is, of course, but one of many primitive functions of sin x/x (i.e. the one which vanishes at the origin) the others being obtainable by adding an arbitrary constant to Si (x). If we want to extend the definition of Si (x) to complex variables, it can be done in a trivial way. We write Si (2) -[ Ear, where f is a complex variable and the integral is now a curvilinear integral in the complex {-plane over some path connecting the points ¢ = 0 and ¢ = z. It does not matter which particular path is chosen since the function f(t) = sin ¢/¢ is analytic for all (finite) valuest of ¢ and the integral is independent of path. A different situation is encountered in the so-called cosine integral, defined by 42 Ci(y) = -{ COS E ae This function is easily seen to be a primitive function of cos x/x. The choice of limits is conventional but the main feature is that the integral diverges at x = 0. While it is still possible to extend the definition to complex variables by writing ete aw--f oon as, * That is, algebraic, exponential, trigonometric, and their inverse functions. +t See the discussion on pp. 79-80.oat INTEGRAL REPRESENTATIONS 105 the question arises as to the specification of the path of integration. Presumably, the upper limit implies an asymptotic approach to the real axis.* Three such paths are illustrated in Fig. 2.33. The integrals taken along C; and C2 are easily shown to be identical, but the integral along C; will differ by 2ni times the residue of the integrand at ¢ = 0. This means essentially that Ci (z) is a multivalued function, the integrals along different paths yielding different branches. Two of these branches are characterized by the paths C; and Cs (or other paths equivalent to them). Other branches can be obtained by circling the origin n times before proceeding to “plus infinity.” Im a Re Pole, Figure 2.33 Remark, The above analysis reveals that Ci (x) for real negative x should be defined by one of the branches of complex function Ci (z) because the réal integral = cost [a is divergent for x < 0 due to the behavior of the integrand at = 0. It has been found, however, that this divergent integral (like many similar others) can be given a definite meaning by the prescription car coue Eaten € os Uf a+ | al known as the Cauchy principal value (or simply the principal value) and is often denoted by - cos £ t re is customary to define Ci (x) for x < 0 by Because of this result, it — cw --f ee at (<0). Principal values of integrals occur very often in physical applications; the complex version of this definitiont will be given shortly. * As usual, we require Ci (z) to reduce to Ci (x) as z becomes real and positive. The method of approaching infinity cannot be modified because cos z has essential singularity at infinity (see p. 109). + The integral —£."* (cos /£) dé is still a branch of Ci (2), although not one of those described previously.106 FUNCTIONS OF A COMPLEX VARIABLE 2.15 A second type of representation of functions by integrals occurs when the limits of integration are fixed but the integrand depends on a parameter. A well- known example of this kind is the integral f oo dy = VE : 2 occurring in the kinetic theory of gases and many other branches of physics.* It may be viewed as an integral representation of the function I(k) = (/#/2)A— "7. This point of view is, as a matter of fact, utilized in practice when the function IQ) is differentiatedt to yield other important integrals, like dQ) _ _ Vr \-3)2 _ 2h? one aah =—] xe dx, @IQX) _ 3025/2 ft os 7 Ce! “s* = 7 xe ee eee In this particular example the function (A) represented by the integral is a familiar one, but the same idea can be used to generate a variety of new functions. Con- sider, for instance, the integral Sin) =f te'dt (m= 1,2,3,...), E which is equal to n! as is not difficult to prove. The above integral representation of n! suggests the extension of the notion of factorial. We may define the factorial function TI(x) by the integral T(x) = if * Fe dt, where x is not necessarily a positive integer. The integral converges at infinity for all values of x, but we must demand that x > —1 for the integral to converge at f= 0. Consequently, the function I(x) is defined for all (real) x > —1 by the above integral representation. Since I(n) = n!, we have M@+1)=@+DNM@ (= 1,2,3,..). This formula holds for all values of x(x > —1). Indeed, integrating by parts, we obtain i Pte dy = — +e aoe (x + Ifem* dt. 0 oo * For the evaluation of this integral, see Kaplan, p. 218. + Differentiation of improper integrals should be justified by appropriate theorems. See e.g., Kaplan, p. 379.sea INTEGRAL REPRESENTATIONS 107 Since lim F#e* = 0 (alll x), te lim # +e 0 (x>-1), 0 it follows that fo ereta = (+ vf Fe—' dt 0 0 or Ux+ 1) =(@+DIQ) («> -D. In particular, 11(0) = 1. Thisis the actual reason for the commonly used convention oO! The recursion formula I(x + 1) = (x + 1)H(x) permits calculation of I(x) for any x (x > —1) provided a table of II(x) is compiled for the interval 0 < x <1. Moreover, it permits an extension of the definition of I1(x) into the region x < —1. Example. Calculate 11(4), (—4), 1(—3). The first value can be obtained directly from the definition mG) = [° te at @) = [ore by the substitution ¢ = x?. Then m1) = af xe dx = 2 0 by a previous result. The value of II(— }) can also be obtained directly in the same manner. But it can be, alternatively, obtained from the recursion formula. If we set x = —3, then M1) = 3(—4); m(-% therefore The value of 1l(—3) cannot be obtained directly (the integral diverges) but it may be defined via the recursion formula. If we set x = —$, then M(—4) = (-)m(-9, yielding 1(—4) = —2Vr. ‘An important feature of II(x) is that it approaches infinity as x approaches —1 from the right. The conclusion that II(x) has infinite discontinuities at x equal to a negative integer follows from this fact and the recursion formula. The graph of I(x) is sketched in Fig. 2.34.108 FUNCTIONS OF A COMPLEX VARIABLE ae While the extension of the concept of the factorial is widely used in applied mathematics, it is not usually accomplished by means of the factorial function T(x) but rather by means of the related gamma function defined by T(x) = Hx — 1). Evidently P(x) satisfies the relations T(x + 1) = TQ) (real x), T(x) = (x— 1)! (x = positive integer), and has the integral representation Te) -f[ foe dt (x > 0). The behavior of P(x) a8 a function of x is easily obtained from Fig. 2.34 just by shifting the origin to the point x = —1,y = 0. 1) Figure 2.34 Integral representations of the type given for factorial and gamma functions can be extended to complex integrals as well. A complex function g(z) may be defined by a definite integral . ate) = ["" fe, 3) ds, Along C where the endpoints s; and sz as well as the path C between them is prescribed. In many cases the points s and s coincide and the path is a closed contour. From the theory of analytic functions, it follows that the path of integration can be deformed within certain limits (as much as the Cauchy theorem and its consequences permit this).