BASIC COMPLEX ANALYSIS JERROLD E. MARSDEN California Institute of MICHAEL J. HOFFMAN California State University, Los Angeles New YorkAcquisitions Editor: Richard J. Bonacci Marketing Manager: Kimberly Manzt Project Editor: Penelope Hull Text Designer: Jerrold F, Marsden Cover Designer: Michael Minchillo Illustrator: Fine Line Studio Production and Mlustration Coardinator: Suson Wein Compositor: Jerrold F. Marsden Manufacturer: RR Donnelley & Sons Company Library of Congress Cataloging-in-Publication Data Marsden, Jerrold E. cea eta ree oem ne Hoffman. — 3rd ed. Includes index ISBN 0-7167-2877-X (EAN: 9780716728771) 1. Functions of camplex variables. 1 Hoffman, Michael J. 11. Title. QA331.7.M36 1999 5151.9-de21 98-41459 cr © 1999, 1987, 1973 by W. H, Freeman and Company. All rights reserved. No pan of this book may be reproduced by any mechanical, photographic, or electronic process, or in the form of phonographic recording, nor may it be stored in a retrieval system, transmitted, or otherwise copied for public or private use without written permission from the publisher. Printed in the United States of America Fifth printing W. H. Freeman and Company 41 Madison Avenne, New York, NY 10010 Houndmills. Besingstake RG21 6XS, EnglandContents Preface vii 1 Analytic Functions 1.1 Introduction to Complex Numbers 60.2.0... 0000 eee 12 ee a ee - 13 14 1.5 Basie Properties of Anelytic Functions . es 1.6 Differentiation of the Elementary Functions 2 Cauchy’s Theorem 95 Si Comomrinege 00 95 2.2 Cauchy's Theorem—A First Look . . - 11 2.3 A Closer Look at Cauchy’s Theorem . . . AS 123 2.4 Cauchy's Integral Formula. ........------ ere 44 2.5 Maximum Modulus Theorem and Harmonic Functions ....... - 164 3 Series Representation of Analytic Functions 3.1 Convergent Series of Analytic Functions 3.2 Power Series and Taylor’s Theorem . 3.3 Laurent Series and Classification of Singularities. 2... eee ee eee eee eee 222 4 Calculus of Residues 4.1 4.2 i 4.3 Evaluation of Definite Integrals 4.4 Evaluation of Infinite Series and Partiel-Fraction Expansions .. 1... - 2-0 eee eee eee 304 5 Conformal Mappings 319 5.1 Basic Theory of Conformal Mappings ......-...2--0-55 319 5.2 Fractions! Linear and Schwarz-Christoffel Transformations ....-----.....-+-+ 3275.3 Applications of Conformal Mappings to Leplace’s Equation, Heat Conduction, Electrostatics, and Hydrodynamics .--...........-2.-5 345 6 Further Development of the Theory 365 6.1 Analytic Continuation and Elementary 7 Asymptotic Methods 409 7.1 Infinite Products and the Gamma Function .............. 409 7.2 Asymptotic Expansions and the Method of Steepest Descent... 2... 2-2. --eeeeeeeeee 427 7.3 Stirling's Formula and Bessel Functions ..-.-........64. 446 8 Laplace Transform and Applications 457 8.1 Basic Properties of Laplace Transforms ..........--.... 457 8.2 Complex Inversion Formula ......-----.--------005 47 8.3 Application of Laplace Transforms to Ordinary Differentie!l Equations .............-2..0005 476 Answers to Odd-Numbered Exercises 481 Index 506Preface This text is intended for undergraduates in mathematics, the physical sciences, and engineering who are taking complex analysis for the first time. Two years of calculus, up through calculus of several variables and Green’s theorem, are adequate preparation for the course. The text contains same references to linear algebra and basic facts sbout ¢-§ analysis, but the extent to which they are emphasized can be adjusted by the instructor depending on the background and needs of the class. The book has a generous number of examples, exercises, and applications. We have made a special! effort to motivate students by making the book readable for self-study and have provided plenty of material to help students gain an intu- itive understanding of the subject. Our arrangement enables application-oriented ‘students to skip the more technical parts without sacrificing an understanding of the main theoretical points. Applications include electric potentials, heat conduc- tion, hydrodynamics (studied with the aid of harmonic functions and conformal mappings), Laplace transforms, asymptotic expansions, the Gamma function, and Bessel functions. The core of Chapters 1 to 6 can be taught in a one-semester course for math- ematics majors. In applied mathematics courses, if some of the technical parts of Chapter 2 and parts of Chapter 6 are omitted, then parts of Chapters 7 and 8 can be covered in one semester. It is healthy for mathematics majors to see as many of the applications as possible, for they are an integral part of the cultural and historical heritage of mathematics. Symbols The symbols used in this text are, for the most part, standard. The set of real numbers is denoted R, while C denotes the set of complex numbers. “If” stands for “if and only if” (except in definitions, where we write only “if”). The end of a proof is marked IM, the end of the proof of a lemma in the middle of a proof of a theorem is marked v and occasionally, the end of an example in the text is marked @. The notation Ja, }[ represents the open interval consisting of all real numbers = satisfying a < x < b. This is to avoid confusion with the ordered pair notation (a,b). The notation f : AC C + C means that the mapping f maps the domain A, which is a subset of C, into C, and we write z+ f(z) for the effect of f on the point z € A. Occasionally, = is used to mean “implies”. The set theoretic difference of the sets A and B is denoted by A\B, while their union and intersection are denoted by AUB and ANB. The definitions, theorems, propositions, lemmas, and examples are numbered consecutively for easy cross reference; for example, Definition 6.2.3 refers to the third item in Section 6.2. vilPreface Classic texts Despite the large numbers of texts written in recent years, some of the alder classics remain the best. A few that are worth looking at are A. Hurwitz and R. Courant, Vorlesungen tiber allgemeine Funktionentheorie und elliptische Funktionen (Berlin: Julius Springer, 1925); E. T. Whittaker and G. N. Watson, A Course of Modcrn Analysis, Fourth Edition (London: Cambridge University Press, 1927); E. T. Titchmarch, The Theory of Functions, 2d ed. (NewYork: Oxford University Press, 1939, reprinted 1985); and K. Knopp, Theary of Functions (New York: Dover, 1947). The reader who wishes further information on various of the more advanced topics can profitably consult E. Hille, Analytic Function Theory, 2 volumes, (Boston: Ginn, 1959); L. V. Ablifors, Complex Analysis (New York: McGraw-Hill, 1966); W. Rudin, Real and Complez Analysis (New York: McGraw- Hill, 1969); and P. M. Morse and H. Feshbach, Methods of Theoretical Physics (New York: McGraw-Hill, 1953). Some additional references are given throughout the text, ‘The modern treatinent of complex analysis did not evolve rapidly or smoothly. The numerous creators of this area of mathematics traveled over many rough roads and encountered many blind alleys before the superior routes were found. An ap- preciation of the history of mathematics and its intimate connection to the physical sciences is important to every student's education. We recommend tooking at M. Klein’s Afathematical Thought from Ancient to Modern Times (London: Oxford University Press, 1972). Third edition The third edition features an Instructor's Supplement as well as @ Student Guide. Answers to the odd numbered problems are in the back of the book and exercises with solutions in the Student Guide are marked with a bullet (°) in the text. We have streanilined a number of features in the text, such as the treatment of Cauchy’s Theorem. We have substantially rewritten Chapter 4 on the evaluation of integrals, making the treatment less encyclopedic. An Internet Supplement is available free from ltup://www.whfreeman.com/ (look in the mathematics section) or from http://cds.caltech.edu/“mansden/ (look under “books”). It contains ad- ditional information for those who want to delve into some topics in a little more depth. Acknowledgments We are grateful to the many readers who supplied correc- tions and comments for this edition. There are too many to be thanked individ- ually, but we would like especially to mention (more or less chronologically) M. Buchner (who lielped significantly with the First Edition), C. Risk, P. Roeder, W. Barker. G. Hill, J. Seitz, J. Bradowski, H. O. Cordes, M. Choi, W. T. Stallings, E. Green. R. Itis, N. Starr, D. Fowler, L. L. Campbell, D. Goldschmidt, T. Kato, J. Mesirov. P. Kenshaft, K. L. Teo, G. Bergmann, J. Harrison and C. Daniels, Finally, ay thenk Barbara Marsden for her accurate typsetting of this new edition.Chapter 1 Analytic Functions In this chapter the basic ideas about complex numbers and nualytic functions are introduced. The organization of the text is analogous to that of an clementary calculus textbook, which begins by introducing R, the sct of real numbers, and functions f(z) of a real variable z. Que then studies the theory and practice of differentiation and integration of functions of a real variable. Similarly, in complex analysis we begin by introducing C, the set of cumplex numbers z. We then study functions f(z) of a complex variable z, which are differentiable in a complex sense; these are called analytic functions. ‘The analogy between real and complex variables is, however, a tittle deceptive, because complex analysis is a surprisingly richer thcory; a lot more can be said about an analytic function than about a differentiable function of a rea) variable, as will be fully developed in subsequent chapters. In addition to becoming familinr with the theory, the student should strive to gain facility with the standard (or elementary”) functions such as polynomials, e*, log 2, sin 2—as in calculus. These functions arc studied in §1.3 and appear frequently throughout the text. 1.1 Introduction to Complex Numbers ‘The following discussion will assume some familiarity with the main propertics of real numbers. The real number system resulted from the search for a system (au abstract set together with certain rules) that included the rationals but that also provided solutions to such polynomial equations as x? — 2 = 0. Historical Perspective Historically, a similar consideration gave rise to an ex- tension of the real numbers. As early as the sixteenth century, Geroniino Cardano considered quadratic (and cubic) equations such us z? + 2x +2 = 0, which is sat- isfied by no real number x. The quadratic formula (—6 4: Vi? — 4ac) /2a yields “formal” expressions for the two solutions of the equation az* + br+c¢ = 0. But this2 Chapter 1 Analytic Functions formula may involve square roots of negative numbers; for example, 1 V=I for the equation z? + 27 +2 = 0. Cardano noticed that if these “complex numbers” were treated as ordinary numbers with the added rule that /—T- /—T = —1, they did indeed solve the equations. The important expression 1 is now given the widely accepted designation i = J-1. (An alternative convention is followed by many electrical engincars, who prefer the symbol j = /—1 since they wih to reserve the symbol i for electric current.) However, in the past it was felt that. no meaning could actually be assigned to such expressions, which were therefore termed “imaginary.” Gradually, especially ‘85 a result of the work of Leonhard Euler in the eighteenth contury, these imaginary quantities came to play an important role. For example, Buler’s formula e” = cos@ + ésin@ revealed the existence of a profound relationship between complex numbers and the trigonometric functions. The rule e¥( +8) ~ P gi was found to summarize the rules for expanding sine and cosine of u sum of two angles in a neat way, and this result alone indicated that some meaning should be attached to these “imaginary” numbers. However, it took nearly three hundred years until the work of Casper Wessel (ca. 1797), Jean Robert Argand (1806), Karl Friedrich Gauss (1831), Sir William R. Ramilton (1837), and others, when “imaginary” numbers were recognized as legit- imate mathematical objects, and it was realized that there is nothing “imaginary” about them at all (although this term is still used). The complex analysis that is the subject of this book was developed in the ninetcenth century, mainly by Augustin Cauchy (1789-1857). Later his theory was made more rigorous and extended by such mathematicians as Peter Dirichlet (1805-1859), Karl Weierstrass (1815-1897), and Gcorg Friedrich Bernhard Riemann (1826-1866). ‘The search for a method to describe heat conduction influenced the developnent of the theory, which has found many uses outside mathematics. Subsequent chap- tons will discuss some of these applications to problems in physics and engineering, such as hydrodynamics and electrostatics. The theory also has mathematical ap- plications to problems that at first do not seem to involve complex numbers. For example, the proof that “sintz « [enh or that, 20 yo-1 a —_dr = ——., fo lta sin(am) (where 0 < a < 1), or that t doe fy atsind Vat—1" may be difficult or, in some cases, impossible using elementary calculus, but these identities can be readily proved using the techniques of complex variables.§1.1 Introduction to Complex Numbers 3 ‘The Complex Number System Coniplex analysis has become an indispensable and standard tool of the working mathematician, physicist, and engineer. Neglect of it can prove to be a severe handicap in most, areas of research and application involving mathematical ideas and techniques. The first objective of this section will be to define complex numbers and to show that the usual algebraic manipulations hold. To begin, recall that the zy plane, denoted by R?, consists of all ordered pairs (x,y) of real numbers. Definition 1.1.1 The system of compler numbers, denoted C, is the set R? together with the usual rules of vector addition and scalar multiplication by a real number a, namely, (tin) + (tate) = (=r + 22m +42) a(z,y) = (az,ay) and wrth the operation of complez multiplication, defined by (21,41)(@2,¥2) = (212 — tive, T192 + 172). ‘We will need to explain where this strange rule of multiplication comes from! Rather than using (z, y) to represent a complex number, we will find it more conve- nient to return to more standard notation as follows. Let us identify real numbers z with points on the x axis; thus x and (2,0) stand for the same point (£,0) in R?. The y axis will be called the imaginary azis, and the unit point (0,1) will be denoted i. Thus, by definition, i = (0,1). Then (uy) =24+y8 because the right side of the equation stands for (z,0) + y(0,1) = (2,0) + 0.) = (2.9). Using y = (y,0) and Definition 1.1.1 of complex multiplication, we get iy = (0,1)(y,0) = (0-y- 1-0,y-1+0-0) = (0,y) = (0,1) = vi, so we can also write (x,y) = z+ ty. A single symbol such as z = ¢ + ib is generally used to indicate a complex number. The notation z € C means that z belongs to the set of complex numbers. Note that # =4-4=(0,1)- (0,1) = (0-0-1-1,(1-040-1)) = (-1,0) = -2, 80 we do have the property we want:4 Chapter 1 Analytic Functions If we remember this equation, then the rule for multiplication of complex numbers is also easy to remember and motivate: (atibe+id) = actiad + ike + bd = (ac — bd) + ifad + be). For example, 2 + 3i is the complex number (2,3), and (2 + 3i)(1 ~ 48) = 2 — 193? + 31 - Si = 14 - 53 is another way of saying that (2,3)(1, ~4) = (2-1 3(—4),3- 2+ 2(—4)) = (14, -5). ‘The renson for using the expression @ + bi is twofold. First, it is conventional. Second, the rule i? = —1 is easier to use than the rule (2, b)(c, d) = (ac—bd, bc-+-ad), although both rules produce the same result. Becanse multiplication of real numbers is associative, commutative, and dis- tributive, it is reasonable to expect that multiplication of complex numbers is also; that is, for all complex numbers 2, w, and s we bave (zw)s = 2(ws), zw=wz, and 2(w +s) = 2w+z8. Let us verify the firet of these properties; the others can be similarly verified. Let 2 = 0+ ib,w=c-+id, and s=e+if. Then 2w = (ac—bd) + i(be-+ad), 50 (zw)s = e(ac ~ bd) — (be + ad) + ife(be + ad) + f(ac — bd)}. Similarly, (a+bi)|(ce - df) + i(cf + de)] ace — df) ~ Uef + de) + ifa(of + de) + b(ce - df). Comparing these expressions and accepting the usual properties of real numbers, we conchide that (zw)s = 2(we). Thus we can write, without ambiguity, an expression like :" = 2-...-z (v limes). Note that a+ ib = c+ id means a = meuns in R) and that 0 stands for 0+i0 a@=Oandb=0. in what sense are these complex numbers an extension of the reals? We have already said Uat if a is real we also write a to stand for a + Of = (a,0). In other words, the reals R are ideutified with the z axis in C = R*; we are thus regarding ‘the real numbers as those complex numbers a + bi for which & = 0. If, in the expression a + bi, the term a = 0, we call bi = 04 bi a pure imaginary number. Ins the expression a + bi we say that a is the real part and b is the imaginary part. This is sometimes written Rez = a, Imz = 6, where z = a + bi. Note that Res and lm : are always real numbers (see Figure 1.1.1). 2(ws) d b = d (since this is what equality .0). Thus a +ib = 0 means that both§1.1 Introduction to Complex Numbers 5 Figure 1.1.1: The geometry of complex numbers. Algebraic Properties The complex numbers obey all the algebraic rules that ordinary real numbers do. For example, it will be shown in the follawing discussion that multiplicative inverses exist for nonzero elements. This means that if z 4 0, then there is « (complex) number 2 such that 22 = 1, and we write z= 271. We can write this expression unambiguously (in other words, z’ is uniquely determined), because if zz” = 1 as well, then 2’ = 2!-1 = 2'(2z”) = (2'z)z" = 1-2" =z", and so 2” = z/, To show that z’ exists, suppose that z = a + ib # 0. Then at least one of a # 0,6 ¢ 0 holds, and so a? +b? # 0. To find z’, we set 2! = a! + Vi. The condition zz’ = 1 imposes conditions that will enable us to compute a’ and U. Computing the product gives zz’ = (aa’ — bY/) + (ab’ +a'b)i. The linear equations aa’ — bb! = 1 and al’ + a'b = 0 cam be solved for a’ and U giving a’ = a/(a? + #7) and & = —b/(a? + 6), since a? + U? #0. Thus for z = a + ib £0, we may write aut . +R a+ Having fond this candidate for 2~' it is now a straightforward, albeit tedious, computation to check that it works. If z and w are complex numbers with w ¥ 0, then the symbol z/w means 2w7"; we call 2/w the quotient of z by w. Thus 2~' = 1/z. To compute z~*, the following series of equations is common and is a useful way to remember the preceding formula for 2~': 1 a-tb anth oi oe at (@tije-%) +P @+P ote”6 Chapter J Analytic Functions In short, all the usual algebraic rules for manipulating real numbers, fractions, polynomials, and so on, hold for complex numbers. Formally, the system of complex numbers is en example of a field. The crucial rules for a field, stated bere for reference, are Addition rules @) zt+w=wtez (i) z+ (w4s) = (z+) +s (ii) z40=2 (iv) 2 + (-z) =0 Multiplication rules (i) zw = wz (ii) (zw)s = 2(ws) (ii) lz =z (iv) 2(27') = 1 for 2 #0 Distributive law 2(w + 8) = w+ zs In summary, we have Theorem 1.1.2 The complex numbers C form a field. The student is cautioned that we generally do not define inequalities like z < w, for complex z and w. If one requires the usual ordering properties for reals to hold, then suck an ordering is impossible for complex numbers.’ Thus in this text the notation z < w will be avoided unless z and w happen to be real. Roots of Quadratic Equations As mentioned previously, one of the reasons. for using complex numbers is to enable us to take square roots of negative real numbers. That this can, in fact, be done for all complex numbers is verified in the next proposition. Proposition 1.1.3 Let z € C. Then there exists a complex number w € C such that w* =z. (Notice that —w also satisfies this equation.) "Thin statement can be proved as follows. Suppose that such an ordering exists. Thon aither £2 ort <0. Suppose that i > 0. Then -4 > 0, #0 —1 > 0, which is absurd, Alternatively, suppose that i <0. Then —# > 0, so (—i)(—d) 2 0, Umt is, —1 > 0, agnin absurd. If > = a+ ib and w = c+ id, we could say that z 0 andp=-lifs<0. © ‘The formula for square roots developed in this proof is worth summarizing explicitly. Namely, the two (compler) square roots of a + ib are given by Va+ ib = +(a + ufi), where a and 6 are given by the displayed formula preceding this one and where =196b>0 undp =-i fb <0. From the expressions for a and f we can conclude three things:8 Chapter 1 Analytic Functions 1, The square roots of a complex number are real if and only if the complex number is real and positive. 2. The square roots of a complex number are purely imaginary if and only if the complex number is real and negative. 3. ‘The two square roots of a number coincide if and only if the complex number is vero. {The student should check these conclusions.) ‘We can easily check that the quadratic equation az? + 6z + ¢ = 0 for complex numbers a, 0, c hes solutions 2 = (-b + V0? — 4ac) /2a, where now the square root denotes the two square roots just constructed. Worked Examples Example 1.1.4 Prove that 1/i = -i and that 1/(i + 1) = (1 —#)/2. Solution First, since (144)(1-#) =141=2 Example 1.1.5 Find the real and imaginary parts of (z +2)/(z ~ 1) mhere 2 = rtiy. Solution We start by writing the fraction in terms of the real and imaginary parts of z and “rationalizing the «lenoiinator”. Namely, z4+2_ (e+2)riy = (w+ 2)+ey (e#-1)-y 2-1) (c-l)tiy @-l)+iy (e-1)-yy (@+2)(e- 1) +9? + ily(z — 1) - Wz +2)) (z-1P +9? 7 Hence, +z-2+y (2-1 +y? <3 ea Example 1.1.6 Solve the equation z* +i = 0 for z.§1.1 Introduction to Complex Numbers 9 Solution Let z? = w. Then the equation becomes w? + ¢ = 0. Now we use the fonnnin Va + ib = £(a+ i) we developed for taking square roots. Letting a = 0 and b = —1, we get. i wet (3 aa =i) : Consider the equation 2? = (1 — i)/-V2. Using the same formula for square roots, bnt now letting @ = 1/72 and b = —1/V3, wo obtain the two solutions .o+(EP2 GB) i}. Froin the second possible value for w we obtain two further solutions: -as(EG2 FB) In the next section, de Moivre’s formula. will be developed, which will enable us to find the nth root of any complex number rather simply. Example 1.1.7 Prove that, for complex numbers 2 and w, Ro(z +w) =Rez+Rew and In(z + w) = Imz + bow. Solution Let z = 2 + iy and w = a+ ib. Then z+ w = (x +0)+i(y +0), 50 Re(z + w) = 2 +a = Rez + Rew. Similarly, Im(z + w) = y +b =lmz + bow. Exercises 1. Express the following complex numbers in the form @ + ib: (a) (2438) + (448) 2438 ©) Ter = ltt 2. Express the following complex numbers in the form a + bi: (a) (2 + 3i)(4 + i) (v) (8 + Gi)? &) i+Chapter 1 Analytic Functions ©) (+4) 3. Find the solutions to 2? = 3 — 4. 4, Find the solutions to the following equations: (a) (z+1? =3+4i (b) z4-i=0 5. Find the real and imaginary ports of the following, where z = x + iy: ws 1 () sea 6. Find the real and imaginary parts of the following, where z = x + ty: z+1 () <3 () 7. Is it true that Re(zw) = (Re2z)(Rew)? 8. Ifa is real and z is complex, prove that Re(az) = a Rez and Im(az) = a Imz. Generally, show that Re : C — R is a real linear map; that is, Re(oz + bu) = a Rez + b Rew for a,b real and z, w complex. 9.° Show that Re(iz) = —Im(z) and that Im(iz) = Re(z) for any complex ssumber 2. 10. (a) Fix a complex number z = 2 + iy and consider the linear mapping ¢. : R? — R? (that is, of C ~ C) defined by ¢.(w) = 2-1 (that is, multiplic by z). Prove that the matrix of ¢, in the standard basis (1,0), (0,1) of R? is given by cy v2). (b) Show that $.,25 = $2; 0 ¢as- 11. Assuming that they work for real numbers, show that the nine rules given for a field also work for complex numbers. 12. Using only the axioms for a field, give a formal proof (including all details) for the following:§1.1 Introduction to Complex Numbers pb 1,1 _atz ©) Stat ae 13.* Let (x — iy)/(z + ty) = a + ib. Prove that a? +6? = 1. 14. Prove the binomial theorem for complex numbors; that is, letting 2,w be complex numbers and n be a positive integer, (z+-w)" = 2" + () Oy + (:) ay (9) wu, where oo tr) rin =r Use induction on n. 15. Show that z is real if and only if Rez = z. 16. Prove that, for each integer k, oh os Latte ts 1,09 = 7. Show how this result gives a formula for for all n by writing n = 4k+ 9,0 < js3. 17. Simplify the following: (@) G+ (b) (-#)-? 18, Simplify the following: (a) G-a" ti ©) 1 19. Simplify the following: () Vi4+vi (b) ViFe (or Ww 20. Show Una the following rules uniquely determine complex multiplication on C=R?: (a) (2: + 22)w = 21 + 290 (b) e122 = za i (A) 21 (2223) = (2122)28 (c) Uf 2) and 2 are real, 21 - zz is the usual product of real numbers. =-l12 Chapter 1 Analytic Functions 1.2 Properties of Complex Numbers It is important vo be able to visualize mathematical concepts and to develop geomet- ric intuition—an ability especially valuable in complex analysis. In this section we define and give a geometric interpretation for several concepts: the absolute ualue, argument, polar representation, and complez conjugate of a. complex number. Addition of Complex Numbers In the preceding section a complex number was defined to be a point in the plane R?. Thus, a complex number may be thought of geometrically as a (two-dimensional) vector and pictured as an arrow from the origin to the point. in R? given by the complex number (see Figure 1.2.1). Innaginary axis « y-oxds s=(ab)easth an — ine Read ais = wis Figure 1.2.1: Vector representation of complex numbers. Because the points (z,0) € R? correspond to real numbers, the horizontal or x axis is called the real azis, Similarly, the vertical axis (the y axis) is called the imaginary azis, because points on it have the form iy = (0,y) for y real. As we already saw in Figure 1.1.1, the addition of complex numbers can be pictured as addition of vectors (an explicit example is given in Figure 1.2.2). Polar Representation of Complex Numbers To understand the geometric meaning of multiplying two complex numbens, we will write them in what is called polur coordinate form. Recall that. the length of the vector (a, b) = a+-ib is defined to be Va? +0. Suppose the vector makes an angle @ with Ube positive direction of the real axis, where 0 < @ < 2r (see Figure 1.2.3). ‘Thus, tan@ = b/a. Since @ = rcos@ and b= rsin@, we have 2+bi = rcos8 + (rsin6)i = r(cas@ + ising). This way of writing the complex number is called the polar coordinate represen- tation. The lenath of the vector z = (a,6) = a+ ib is denoted |z| and is called the norm, or modulus. or absolute value of 2. The angle @ is called the argument of the complex numbez and is denoted 0 = arg z.§1.2 Properties of Complex Numbers 13 (3 +21) + (144) 4 45H Figure 1.2.2: Addition of complex numbers. Figure 1.2.3: Polar coordinate representation of complex munbers. Af we restrict @ to the interval 0 < @ < 27, then cach nonzero complex number as an unambiguously defined argument. (We learn this in trigonometry.) How- ever, it is cloar that we can add integral multiples of 2 to @ and still obtain the same complex number. In fact, we shall find it convenient to be flexible in our requirements for the values that 0 is to assume. For exansplo, we could equally well allow the range of @ to be —4 < 8 < w. Such an interval must always be spécificd or be clearly understood. Once an interval of length 2a is specified, then for each z # 0, a unique @ is determined that lies within that specified interval. It is clear that any @ € R can be brought into our specified interval by tbe addition of some (positive or negative) integral multiple of 27. For these reasons it is sometimes best to think of arg z as the set of possible values of the angle. If @ is one possible value, then so is 0-+ 2 for any integer n, and we can sometimes think of arg z as {0+2rn | n is an integer}. Specification of a particular range for the angle is known as choosing a branch of the argument,“4 Chapter I Analytic Functions Multiplication of Complex Numbers The polar representation of complex numbers helps us understand the geometric meaning of the product of two complex numbers. Let 2) = r)(cos@; + isin@)) and 22 = r2(cos 62 + isin 62). Then 4z2 = 1172[(cosA; - cos G2 — sin & - sin A2)] + i{(cosd, - sin 82 +cos Hg - sin 4;)] = rirafeos(@, + 2) + ésin(O, + 62)], by the addition formula for the sine and cosine functions used in trigonometry. ‘Thus, we have proven Proposition 1.2.1 For any comples numbers z; and 22, lz1z2] = [21] +122] and arg(212z) = argz, + argzz (mod 2n). In other words, the product of two complex numbers is the complex number that has a length equal to the product of the lengths of the two complex numbers and an argument equa! to the sum of the arguments of those numbers. This is the basic geometric representation of complex multiplication (see Figure 1.2.4). Figure 1.2.4: Multiplication of complex numbers. The second equality in Proposition 1.2.1 means that the sets of possible values for the left and right sides are the same, that is, that the two sides can be made to agree by the addition of the appropriate multiple of 2x to one side. If a particular branch is desired and arg 2; + arg 22 lies outside the interval that we specify, we should adjust it by a multiple of 27 to bring it within that interval. For example, if our interval is (0,2x{ and 2: = —1 and 22 = —i, then arg 2) = and argz2 = 34/2 (sce Figure 1.2.5), but 2122 = #, 90 arg(2)z2) = 7/2, and arg 2) -+arg 2 = 5430/2 = 2m + 2/2. We can obtain the correct answer by subtracting 2x to bring it within the interval [0, 2x{. Multiplication of complex numbers can be analyzed in another useful way. Let 2 € Cand define #, : C + C by y(w) = wz; that is, y, is the map “multiplication§1.2 Properties of Complex Numbers 16 Figure 1.2.5: Multiplication of the complex numbers -1 and —i. hy 2”. By Proposition 1.2.1, the effect of this map is to rotate a complex number through an angle equal to argz in the counterclockwise direction and to stretch its length by the factor |z|. For example, ¥; (multiplication by #) rotates complex numbens by 1/2 in the counterclockwise direction (see Figure 1.2.6). y vw) Figure 1.2.6: Multiplication by ¢. ‘The map y, is a linear transformation on the plane, in the sense that Welw + pra) = Avbe(un) + Hype(wa),16 Chapter 1 Analytic Functions where A, 4 are real numbers and w;, wz are complex numbers. Any linear transfor- mation of the plane to itself cau be represented by a matrix, as we learn in lincar algebra. If z = a + ib = (a,6), then the matrix of q. is (3 =) (3 2) (5) -(Caet) De Moivre’s Formula The formula we derived for multiplication, using the Polur coordinate representation, provides more than geometric intuition. We can use it to obtain a formula for the nth power of a complex number. This formula can then be used to find the nth roots of any complex number. since {see Exercise 10, §1.1). Proposition 1.2.2 (De Moivre’s Formula) If z = r(cos + isin9) and n is a positive integer, then 2” =r"(cosnd + isin nd). Proof By Proposition 1.2.1, 2 = r|ous(0 + 0) + isin(@ +-6)] = r?(cos28 + sin 20). Multiplying agnin by = gives 2=z-27 =r-r*[cos(26 + 0) + ésin(20 + 6)] = r*(cos30 + isin 30). This procedure may be coutinued by induction to obtain the desired result for any integern. @ Let w he a complex number; that is, let w € C. Using de Moivre’s formula will help us solve the equation 2" = m for 2 when w is given. Suppose that w = r(cosd + isin8) and 2 = p(cosy +fsiny). Then de Moivre's formula gives 2” = p"(cosny +isin ny). It follows that p" = r = |w| by uniquencs of the polar representation and ny = 0 + k(2x), where k is some integer. Thus a gi fen(2-+tas) sian (2 «Har. Each value of k = 0,1,... ,n —1 gives a different valuc of z. Any other valne of k merely repeats one of the values of z corresponding to k = 0,1,2,...,n— 1. Thus there are exactly n nth roots of a (nonzero) complex number.§1.2 Properties of Complex Numbers 7 An example will help illustrate how to use this theory. Consider the problem of finding the three solutions to the equation z* = 1 = 1(cos0+-ésin 0). The preceding formula gives them as follows: kon |. kan 2 2008 —— + isin ——, 3 3 where k = 0,1,2. In other words, the solutions are Wi 1 Ws oe i ee) ‘This procedure for finding roots is summarized as follows. Corollary 1.2.3 Let w be a nonzero compler number with polar representation w=r(cosé + isin). Then the nth roots of w are given by the n complex numbers @ . ak a {8 | Ink a= 07 oon (2 4 2t) + inn ($4 2) k=0,1,....n-1. As a special case of this formula we note that the n roots of 1 (that is, the nth roots of unity) are 1 and n — 1 other points equally spaced around the unit circle, as illustrated in Figure 1.2.7 for the case n = 8. Figure 1.2.7: The eighth roots of unity. Complex Conjugation Subsequent chapters will include many references to the simple idea of conjugation, which is defined as follows: If z = a + ib, then z, the complex conjugate of z, is defined by 7 = a ~ ib. Complex conjugation can be pictured geometrically as reflection in the real axis (see Figure 1.2.8). ‘The next proposition summarizes the main properties of complex conjugation.18 Chapter 1 Analytic Functions Figure 1.2.8: Complex conjugation, Proposition 1.2.4 The following properties hald for complex numbers: @) z47=24+7. (ii) Zz? = 22. (iii) 272! = 2/2 for 2 #0. (iv) 22 = Je[* and hence if 2 #0, we have 2~! = 2/|z/?. (v) 2 = #f and only if z is real. (vi) Re z = (z + 2)/2 and Im z = (z — 2)/28. (vil) B= 2. Proof (i) Let z= a+ tband let 2 =a! +i’. Then 2+ 2 =atal+i{b+U), and so Z4+2 =(a+a')-i(b4+ 0) =a—ih+a’ —if =z47. (B) Let 2 =a + tb and let 2’ =a! + iW. Then aa = (aa! — WH) + ial + 2°) = (aa' — bb") — i(ab! + a’b). On the other hand, 22/ = (a — ib)(a’ — il/) = (aa' —bi/) — ial’ +a'b). (iii) By (ii) we have 22/2! = 22/2 = 2. Hence, 2/2! = z/Z. (iv) 22 = (a+ tha - db) = 0? +P = [2/?. (v) a+ tb = a~ tb, then th = —ib, and 90 b=0. (vi) This assertion is clear by the definition of 2.§1.2 Properties of Complex Numbers 19 (vii) This assertion is also clear by the definition of complex conjugation. © The absolute value of 2 complex number |2| = |a + ib| = Vo? +P, which is the usual Euclidean length of the vector representing the complex number, has already been defined. From Proposition 1.2.4(iv), note that |z| is also given by \:? = 22. The absolute value of a complex number is encountered throughout complex analysis; Lhe following propertics of the absolute value are quite basic. Proposition 1.2.5 (i) |zz/| = [el-|z'l- (ii) Uf 2 £0, then |z/2/| = lel/le'|- (iii) -f2| < Rez < fal and —|z| 0. The other inequality asserted in (iii) is similarly proved. (iv) If z = a4 db, then Z = o — ib, and we clearly have |2| = Vat +o! = Vat (—ay = [2 (v) By Proposition 1.2.4(iv), +2? = (+2\e+z) (z+2'\(2+7/) zaps + E422. 8 But 277 is the conjugate of 2’ (Why?), so by Proposition 1.2.4(vi) and (iii) in this proof, lz)? +2"? + 2Rez’z < fel? + |e"? + 2e'al = Jel? + le’ P + elle’). But this equals (|z{ + |z’|)?, so we get our result. (vi) By applying (v) to 2’ and z— 2/ we get le] = fe + (2-2) S lel + le- 30 |z—2'| > |z|—|z'|. By interchanging the roles of z and z’, we similarly get Jz— 2] 2 [eI lzl = -(lel — |e"]), which is what we originally claimed. (vii) This inequality is less evident and the proof of it. requires a slight wathematical trick (see Exercise 22 for a different proof). Let us suppose that not all the wy = 0 (or else the result is clear). Let * ® “ v= Dla t= ohm, s= Doan. and c= s/t. hed kod k=l Now consider the sum ® Dike - evel? k=l which is > 0 and equals 8 ® vt|et—cS tte -2) awe = v+|e*t-2Recs kel kel 2 . = v4 HE one, Since ¢ is real and si = |s[? is real, 2+ Us) — 2(\s[?/t) = v — |s|?/t > 0. Hence |s{? < ut, which is the desired resul It.§1.2 Properties of Complex Numbers 21 Worked Examples Example 1.2.6 Solve 2® = 1 for z. Solution Since 1 = cosk2n +isink2a when k equals any integer, Corollary 1.2.3 gives kan iar, cos = + tein = k=0,1,2, x w 1 -1 i -1 i . 1 See = -1, ee. -i, Se -e. De RNA AOR These may be pictured as points evenly spaced on the circle in the complex plane (sec Figure 1.2.10). Figure 1.2.10: The eight 8th roots of unity. Example 1.2.7 Show that G+] _ 3-7)? [Gees ~ @=Gi)" Solution The point here is that it is not necessary first to work out (3 +7i)?/(S+ Gi) if we simply use the properties of complex conjugation, namely, 2? = (z)? and 2/2 = 2/2. Thus we obtain Gray _ G7)’ _ @-70? ES] . . @+6) (8+6) (8-6i)” Example 1.2.8 Show that the maximum absolute value of z? +1 on the unit disk lz] <1 #2.22 Chapter 1 Analytic Functions Solution By the triangle inequality, [z? + 1] < |z?| +1 = |z?+1<1241=2, since |z| < 1 thus |e? + 1] doos not exceed 2 on the disk. Since the value 2 is achieved at z= 1, the maximum is 2. Example 1.2.9 Express cus30 in terms of cos and sin@ using de Moivre's for- mula. Solution De Moivre’s formula for r = 1 and n = 3 gives the identity (cos0 + isin)? = cos30 + isin 30. The left side of this equation, when expanded (see Exorcise 14 of §1.1), becomes com @ + i3cas? @sin @ — 3cos@ sin? 0 — isin? @. By equating real and imaginary parts, we get 60538 = cos* 0 — 3casOsin? 6 and the additional formula sin 30 = —sin? @ + 3cos? @sind. Example 1.2.10 Write the equation of a straight line, of a circle, and of an ellipse ‘using complex notation. Solution The straight line is most conveniently expressed in parametric form: z=a+bt,a,b€C,t € R, which represents a line in the direction of b and passing through the point a. The circle can be expressed as |z — a| = r (radius r, center a). ‘The ellipse can be expressed as |z — d| + |z +d] = 2a; the foci are located at +d and the semimajor axis equals a. ‘These equations, in which |-| is interpreted as length, coincide with the geometric definitions of these loci. Exercises 1, Solve the following equations: (a) 2 -2=0 (b) 24 +i=0 2. Solve the following equations: (a) +8=0 {b) 3 -4=0§1.2 Properties of Complex Numbers 23 3. What is the complex conjngate of (3 + 8i)4/(1 + i)!°? 4. What is the complex conjugate of (8 — 2i)!°/(4 + Gi)5? 5. Express cos Sz and sin $z in terms of cosz and sin x. 6. Express cos 6x and sin Gz in terms of cosa and sin z. 7. Find the absolute value of [i(2 + 3i)(5 — 2i)]/(-2 — i). 8. Find the absolute value of (2 — 3i)?/(8 + Gi)”. 9.* Let w be an nth root of unity, w # 1. Show that 1+w+w?+...4w"! =0. 10. Show that the roots of a polynomis! with real cocflicients occur in conjugate pairs. 11. Ifa,b € C, prove the parallelogram identity: |a—b[?-+|a-+b|? = 2{|al?-+|b/?). 12, Interpret the identity in Exercise 11 geometrically. 13. When docs equality hold in the triangle inequality |z + 22 + -.. + zl < lea] + |22] +... + l2n[? Interpret your result geometrically. 14. Assuming either |z| = 1 or |w| = 1 and Zw # 1, prove that z-w 1- zw 15. Does z? = |2{?? If s0, prove this equality. If not, for what z is it truc? 16.* Letting 2 = x + iy, prove that: [z|+-|y| < V3jzl. 17° eee te Prove that |z2/| = |2|)2'| by evaluating each side. 18. Prove the following: (a) arg? = — arg z(mod 2r) (b) arg(z/m) = arg z — arg w(mod 2m) (c) 2} =0 if and only if 2 =0 19. What is the equation of the circle with radius 3 and center 8 + $i in complex notation? 20. Using the formula 2) = 2/J2[?, show how to construct z~* geometrically. 21. Describe the set of all z such that Im (z +5) = 0.24 Chapter 1 Analytic Functions 22.° Prove Lagrange’s identity: a 2 " n [ram = (= taf) (= mf) — LD bead, - zyme! rast 1 = ke Deduce the Cauchy-Schwarz ineqnulity from your proof. 23.* Given a € C, find the maximum of |2" + al for those 2 with |z| <1. 24. Compute the least upper bound (that is, stpremum) of the sct of all real numbers of the form Re (iz? + 1) such that |2| <2. 25.° Prove Lagrange’'s trigonometric identity: is a 1 + cos@ + 00820 +... + cosnd = i 4 Sine 43) 8 2 2win ¥ (Assume that sin(0/2) # 0.) 26. Suppose that the complex numbers 2;, 22, 23 satisfy the equation ara _ anes mm ey Prove that (22 — =| = |zs — zi] = |22 — al. Hint: Argue geometrically, interpreting the meaning of cach statement. 27. Give a necessary nud sufficient. condition for (2) 21, 22:23 to lic on a straight fine. (b) 21, 22,23, 24 to lie ou a straight line or a circle, 28, Prove the identity (sin) (sm 32). (in @S98) = 5. Hint: The given product cnn he written as 1/2"-? times the product of the nouzero roots of the polynominl (1 — 2)" — 1. 29, Let w be an nth root of unity, w # 1. Evaluate ] +2 +3w?+...+nw™-", 30. Show that the correspondence of the complex number z = a + bi with the matrix (5 = t. noted in the text preceding Proposition 1.2.2 has the following properties: (a) aw = Pte. (b) Pete = Be + Ber§1.3 Some Elementary Functions 25 wu=(4 7, (d) Ate = das if d is real. (c) ve = (¥-)f (the transposed matrix), () days = (He). (g) z is real if and only if w. = (vz)! (h) {z| = 1 if and only if y, is an orthogonal matrix. 1.3 Some Elementary Functions We learn about the trigonometric fimctions sine and cosine, as well as the exponen- tial function and the logarithmic function in elementary calculus. Recall that the trigonometric functions may be defitied in terms of the ratios of sides of a right- angled triangle. The definition of “angle” may be extended to include any real value, and thus cos@ and sin@ become real-valued functions of the real variable 8. It is a basic fact that cos and sin@ are differentiable, with derivatives given by d(cos®)/d0 = —sin@ and d(sin0)/d0 = cos8. Alternatively, oo and sin @ can be defined by their power series: x csr = icc The proof of convergence of these series can be found in Chapter 3 and in many calculus texts.? Alternatively, sin z can be defined as the unique solution f(z) to the differentinl equation f(x) + f(z) = 0 satisfying (0) = 0, f'(0) = 1; and cosz: can be defined as the unique solution to f"(z) + f(z) = 0, f(0) = 1,/°(0) = 0 (again, sce a calculus text for proofs). Exponential Function The exponential function, denoted e*, may he defined as the unique solution to the differential equation f’(x) = f(z), subject to the initial condition that f(0) = 1; one has to show that a unique solution exists. The exponential function can also be defined by its power scries: est faltrt att: We accept from calculus the fact that ¢* is a positive, strictly increasing function of x. Therefore, for y > 0,logy can be defined as the inverse function of e*; that 2An example is J. Marsden and A. Weinstein, Coleulus, Second Edition (New York: Springer- ‘Vering, 1985), Chapter 12.26 Chapter ! Analytic Functions is, cl = y. Another approach that is often used in calculus books is to begin by defining logn = f Zat for y > 0 and then to define * as the inverse function of logy. (Many calculus books write In y for the logarithm to the base e. As in most advanced mathematics, throughout this book we will write logy for Iny.) In this section these functions will be extended to the complex plane. In other words, the functions sin z,cosz,e7, and log z will be defined for complex z, and their restrictions to the real line will be the usual sin z,cosz,¢*. and logz. The extension to complex numbers should be natural in the sense that many of the familiar properties of sin, cos,exp, and log are retained. We first extend the exponential function. We know from calculus that for real it, e* can be represented by its Maclaurin series: ni a Pele o¢ motos... e Witatyt ‘Thus, it would be reasonable to define e'¥ by Gu) , Gy)? 1+ TT a Sole Hee for y € R. Of courne, this definition is nat quite legitimate, as convergence of series in C has not yet been discussed. Chapter 3 will show that this series does indeed represent a well-defined complex number for each y, but for the moment the series is used informally as the basis for the definition that follows, which will be precise. A slight rearrangement of the series (using Exercise 16, §1.1) shows that which we recognize as being cosy + isin y. Thus we define eo = cosy +isiny. So fur, we have defined e* for z aloug both the real and imaginary axcs. How do we define e* = e*4¥7 We desire our extension of the exponential to retain the familiar properties, and among these is the law of exponents: e*** = c*-¢°, This requirement forces us to define e=+¥ = c*-e'¥, This can be stated in a formal definition. Definition 1.3.1 If z= x + iy, then e* is defined by c*(cusy + isiny). Note that if z is real (that is, if y = 0), this definition agrees with the usual exponential function e7. The student is cautioned that we arc not, at this stage,§1.3 Some Elementary Functions 27 fully justified in thinking of ¢* as “e raised to the “power” of z,” since we have not yet, for example, established Jaws of exponents for complex numbers. ‘There is another, again purely formal, reason for defining e'¥ = cosy + isiny. If we write e'¥ = f(y) + ig(y), we note that since we want e° = 1, we should have f(0) = 1, and g(0) = 0. If the exponential function is to have the familiar differentiation properties, we will need if = f(y) + ig'y), so when y = 0 we got f’(0) = 0,9'(0) = 1. Differentiating again gives us we = J") +ig"ty)- Comparing this equation with e” = f(y) + ig(y), we conclude that f(y) + f(y) = 0, f(0) = 1, and f'(0) = 0. Therefore, f(y) = cosy by the definition of cosy in terms of differential eqnations. Similarly, we find that g”(y)+9(y) = 0,9(0) = 0,9’(0) = 1, and hence g(y) = siny. Thus, we would obtain eY = cosy + isin y as in Definition 1.3.1. Some of the important properties of e* are summarized in the following propo- sition. To state it, we recall the definition of a periodic fuoction. A function 1: > Cis called periodic if there exists a w € C (called a period) such that S(z4w) = f(z) for all z EC. Proposition 1.3.2 (i) et” = ee” for all zw EC. (ii) e* is never zero. (tii) If x is real, then e* > 1 when x > 0 and0 0. {iv) Using |z2"| = [el]2’| (see Proposition 1.2.5) and the fncts that e* > 0 and cos? y + sin? y = 1, we get let] = [ere = le*lle™| = e*lcosy +isiny| = e*. (v) By definition, c*/? = cos(x/2) + isin(z/2) = i. The proofs of the other formulas are similar. (vi) Suppose that ¢7*” = e* for all z € C. Setting z = 0, we get e” = 1. If w= s+, then, using (iv), e” = 1 implies that e* = 1, so s = 0. Hence any period is of the form li, for some t € R. Suppose that « = 1, that. is, that cost +isint = 1. Then cost = 1, sint = 0; thus, ¢ = 2mm for some integer n. (vii) © = 1, as we have soon, and €™ = 1 because e* is periodic, by (vi). Conversely, c* = 1 implies that e***’ = e*’ for all 2’; so by (vi). 2 = 2nni for some integer n. How can we picture ¢#? Since e¥ = (cosy,siny), it moves along the unit circle in a counterclockwise direction as y goes from 0 to 2x. It reaches i at y = 7/2, —1 at z,—1 at 37/2, and 1 again at 2z. Thus, e'? is the point on the unit circle with argument y (sec Figure 1.3.1). Note that in expouential form, the polar representation of a complex number becomes z= [2|ellorm=) which is sometimes abbreviated to z = re, ‘Trigonometric Functions Next we wish to extend the definitions of cosine and sine to the complex plane. The extension of the exponential to the complex plane suggests a way to extend the definitions of sie and cosine. We have ei! = cosy + isiny, and e~¥ = cosy — isin y, which implies that fv — ewe a ea ied z ad oS ‘TAnother proof utilizing the definition of c* in toruy of differontin! equations is wm follows. Recall that ¢* ix Uw: unique solution to f’(z) = f(z) with e” = 1 (x real). Since e* is continuous ‘and is never sero, it nuust he strictly positive. Heuce (e*)' = e* is always positive and consequently & is strictly increasing. Thus for a > 0,¢* > 1. Similarly, for z <0, we have e” <1.§1.3 Some Elementary Functions 29 Figure 1.3.1: Points on the unit circle. But since e'* is now defined for any z € C, we are led to formulate the following definition. Definition 1.3.3 The complex sine and cosine functions are defined by Jor any complex number z. Again, if z is real, these definitions agree with the usual definitions of sine and cosine learned in clementary calculus. ‘The next proposition lists some of the properties of the sine and cosine functions that have now been defined over the whole of C and not merely on R. Proposition 1.3.4 (i) sin? z + cos? 2 = 1. (i) sin(z + w) = sin z-cosw + cos 2 -sinw and c0s(z + w) = cosz- cosw — sin z-sinw. Again the student is cautioned that these formulas, although plausible, must be proved, since at this stage we know their validity only when w and z are real. Proof Using the definitions, we have el gi a et genit\? a, 2 et 24a 424% rr a sin? z + cos? z =1,30 Chapter 1 Analytic Functions which proves (i). To prove (ii), write sinz-cosw + cosz-sinw ct ent civ yeni which, using ee” = ¢f(*+) and noting cancellations between the two terns, simplifies to gilt) — gnilete) — gilete) _ a ¥ a ‘The student can similarly check the addition formula for cus(z +10). il In addition to cus z and sin z, we cau define tan z = (sin z)/(cosz) when cos z # 0, and similarly obtain the other trigonometric functions. iste) gi(=tw) _ 9-i(s+w) = = siin(z + w). Logarithm Function We now define the logarithm in a way that agrees with the usual definition of log x when « is real and positive. In the real case we can view the logarithm as the inverse of the exponential (that is, log = y is the solution of e” =z). When we allow z to range over C, we must be more careful, because the exponential is periodic and thus cannot have a unique inverse. Furthermore, the exponential is never zero, 80 we cannot expect to be able to define the logarithm at. zero. Thus, we must he careful in our choice of the domain in C on which we can define the logarithm. The next proposition indicates how this may be done. Proposition 1.3.5 Let Ay, denote the set of complex numbers x + iy such that Yo Sy < yo + 20; symbolically, Ay = {z+iy|zER and y Sy < yt 2x}. Then e* maps Ay, ina one-to-one manner onto the set C\{0}. Recall that 4 map is one-to-one when the map takes every two distinct points to two distinct points; in other words, two distinct points never get mapped to the same point. A map is onto a sct B when every point of B is the imnge of some point under the mapping. The notation C\{0} means the whole plane C minus the point 0; that is, the plane with the origin removed. Proof Ife" = e®, then e'-*? = 1, 80 2) — 22 = 2min for some integer 1, by Proposition 1.3.2. But because 2; and zz both lic in Ay, where the difference between the imaginary parts of any points is less thon 2x, we must have 2) = 22. This argument shows that e* is one-to-one. Let w € C with w #0. We claim the equation ¢* = w has a solution z in Ay,. The equation ¢*+¥ = w is equivalent to the two equations ¢* = |w| and e” = w/}w|. (Why?) The solution of the first equation is z = log tol, where “log” is the ordinary logarithm (with base e) defined§1.3 Some Elementary Functions 31 on the positive part of the res! axis. The second equation has infinitely many solutions y, each differing by integral multiples of 2x. but exactly one of these is the interval [yo, yo + 27]. This y is merely arg, where the specified range for the ang function is [yo, yo + 2n[. Thus e* is onto C\{0}. ‘The sets defined in this proposition are shown in Figure 1.3.2. Here e* maps the horizontal strip between yoi and (yp + 2x)i one-to-aue onto C\{0}. (The notation 2+ {(z) is used to indicate that z is sont to f(z) under the mapping f.) Figure 1.3.2: e* as a one-to-one function onto C\{0}. In the proof of Proposition 1.3.5 an explicit expression was derived for the inverse of e* restricted to the strip yo < Imz < yo + 2x, and this expression is stated formally in the following definition. Definition 1.8.6 The function log : C\{0} — C, with range yo < Imlogz < Yo +2n, is defined by log z = log|z| + iargz, where arg z tokes values in the interval [yo, yo+2x| and log |2| és the usual logarithm of the positive real number |2|. This function is sometimes referred to as the “branch of the logarithm function lying in {z + iy | yo S y < yo + 2z}.” But we must remember that the function logz is well defined only when we specify an interval of length 2m in which urg z takes its values, that is, when a specific branch is chosen. For example, suppose that the specified interval for the argument is {0,2x|- Then log(1 + i) = log V2 + ix/4. However, if the specificd interval is fx, 3x], then log(1 + #) = log V2 + i9x/4. Any particular branch of the logarithm defined in this way tndergoes a sudden jump as z moves across the ray arg z = yo. To avoid this jumping, one can restrict the domain to yo < y < yo + 27. This idea will be important in §1.6.32 Chapter 1 Analytic Functions Proposition 1.3.7 The logarithin log z is the inverse of e* in the following sense: For any branch of lugz, we have el6* = z, and if we choose the branch lying in Yo S¥ < yo + 2x, then log(e*) = z for z = z+ ty and yo < y < yo +27. Proof Since log z = log |z| + iarg z, we have ole = clonlelpiangs Couversely, suppose that z = x + iy and yy S y < yo + 2a. By definition, loge* = log e®| + farge®. But |e*] = c* and arge® = y by our choice of branch. Thus, loge® =loge* +iy=a+iy=z. OF ‘The logarithm defined on C\{0} behaves the same way with respect to products as the logarithm restricted to the positive part of the real axis. Proposition 1.3.8 If 21,22 € oe then log(z)22) = log 2, + log 22 (up to the addition of integral multiples of 27, Proof By definition, log 222 = log [12a] + farg(2122), where an interval (yo, Yo + 2a{ has been chosen for the values of the arg function. We know that log |z122) = tog |es[lz2] = log |z1] + log |zo] and arg(2)22) = arg 21 + arg 22 (up to integral multi- ples of 24). Thus log 2; zz = (log |z;|+é arg 21) + (log |z2| + i arg 22) = log 2 +log z2 (up to integral multiples of 277). To illustrate this proposition, let us find log{(—1 - é)(1 — é)}, where the range for the arg function is chosen as, for instance, (0, 2x|. Thus, logl(—1 — )(1 — )] = log(-2) = log? + x. Ou the other hand, log(—1 - i) = log V2 + i5x/4 and log(1 — i} = log v2 + #7m/4. Thus, log(—1 - 7} + log(1 — i) = log 2 + #3% = (log 2 + mi) + 273, 80 in this case, when z) = —1—# and zo = 1 —i. log 2122 differs from log 21 + log zo by 2ni. ‘The basic property in Proposition 1.3.8 can help one remember the definition of log z by writing log z = log(re) = logr + loge = log|z| +éargz. Complex Powers We arc now in a position to define the expression a® where a,b € C and a # O (read “a raised to the power of 8°). Of course, however we define a*. the definition should reduce to the usual one in which a and 6 are real numbers. Notice that a can also be written c!&* by Proposition 1.3.7. Thus, if 6 is an integer, we have a = (¢!%*)> = e's, ‘This last equality holds since if n is an integer and 2 is any complex number, (*)" = ¢* ...e* = e"* hy Proposition 1.3.2(i). Thus we are led to formulate the following definition.§1.3 Some Elementary Functions 33 Definition 1.3.9 Let a,b € C with a #0. Then a! is defined to be c?™8*; it is understood that some interval {yo, yo + 2{ (that is, some branch of log) has been chosen within which the arg function takes its values. it is important to understand precisely what this definition involves. Note cs- pecially that in general logs is “multiple-valied”; that is, logz can he assigned many different values because different intervals [yo + 2x[ can be choseu. This is not surprising, for if b = 1/9, where q is an integer, then our previous work with de Moivre’s formula would lead us to expect that a? is one of the qth roots of a and thus should have q distinct valucs. The following theorem elucidates this point. Proposition 1.3.10 Let a,b € C,a # 0. Then a” is single-valued (that is, the value of a dos not depend on the choice of branch for log) if and only if b is an integer. If b is a real, rutional number, and if b = p/q is in its lowest terms (in other words, if p and q have no common factor), then a> has exuctly q distinct talues, namely, the g roots of a”. If b is real and irrational or if b has a nonzero imaginary part, then a> has infinitely many values. When ab has distinct values, these values differ by factors of the form 2", Proof Chovse sonic interval, for example, [0,2x[, for the values of the arg fime- tion. Let log z be the corresponding branch of the logarithm. Lf we were to choose any other branch of the log function, we would obtain loga + 2mni rather than log a, for some integer n. Thus a® = chloeet2enbi .. ghloga , .2enbi where the valuc of ni depends on the branch of logarithm (that is, on the interval chosen for the values of the arg function). By Proposition 1.3.2. e?*""' remains the same for different valucs of 2 if and only if b is an integer. Similurly, e2°""?/9 has q distinct values if pand g have no common factor. If b ix irrational, and if ¢?*" = ¢?*™, it follows that c2#4)—™) = 1 and hence U(r — m) is an integer; since: b is irrational, this implies that n —m = 0. Thus if b is irrational, ¢?*"™ has infinitely many distinct values. If b is of the form x + iy.y # 0. then e274 = @-24nv . e2iz, which also has infinitely many distinct values. @ To repeat: When we write c#€*, it is understood that sume branch of log has been chosen, and accordingly e!°5° has a single well-defined value. But as we change the branch of log, we get values for c*”S* that differ by factors of ¢?*!®, This is what we mean when we say that a! = ¢°'s4 is “inultiple-valucd”. An cxample should inake this clear. Let a = 1+ and fet b be some real irrational number. Then the infinitely many different possible values of a® are given by (1 4) = eto tiptzemd) . Ming V84ix/442emi) _ (qhles Va+ibe 14) gh2eni as n takes on all integral valucs (corresponding to different choices of the branch). For instance, if we used the branch corresponding to [—z, a[ or [U.2x[. we would sein = 0. Some general properties of a! are fonnd in the exercises at the end of this section, but we are now interested in the special case when 6 is of the: form 1/n, because this gives the nth root.34 Chapter 1 Analytic Functions ‘The nth Root Function We know that 7/ has exactly n values for z # 0. To make it a specific function we single out a branch of log as described in the Preceding paragraphs. Definition 1.3.11 The nth root function is defined by z= 2hln x fonz)in for a specific choice of branch of log z; with this choice, o/z = ef8#)/" is called a branch of the nth root function. ‘The next proposition verifies a familiar property of root functions. Proposition 1.3.12 The function YZ so defined is an nth root of 2; that is, ( 92)" =z. It is obtained as follows. If z= re™, then f= oro, where 0 is choscn so that it lies within a particular interval corresponding to the branch choice. As we add multiples of 2x to 0, we run through the n nth roots of z. On the right-hand side, {/F is the usual positive real nth root of the positive real number r. Proof By definition, 9/2 = c{82)/", But logz = logr +18, so eftonad/n = gllonrd/n . gi8/n — gfpi9/n_ ‘The assertion is then clear. The reader should now take the time to become convinced that this way of describing the n. nth roots of z is the same as thal described in Corollary 1.2.3, Geometry of the Elementary Functions To further understand the functions 2", 2.e?, and log z, we shall consider the geometric interpretation of each in the remainder of this section. Let us begin with the power function 2” and let n = 2. We know that 2? has longth |z/? and argument 2argz. Thus the map z + 2? squares lengths and doubles aryuments (see Figure 1.3.3). From this doubling of angles it follows that the power function z? maps the first quadrant to the whole upper half plane (see Figure 1.3.4). Similarly, the upper half plane is mapped to the whole plane. Now consider the square root function 2 = y7c'®/2, Suppose that we choose a branch by using the interval 0 < @ < 2x. Then 0 < 0/2 < x, so VZ will always lie in the upper half plane, and the anyles thus are cut in half. The situation is similar to that involving the exponential function in that z+ «/2 is the inverse for z+ 2? when the latter is restricted to a region on which it is one-to-one. in like manner, if we choose the branch —x < 6 < x, we have —1/2 < 0/2 < 1/2, 50 «/Z takes its values in the right half plane instead of the upper half plane. (Generally, any “halfFigure 1.3.4: Effect of the squaring function on the first quadrant. plane” could be used—see Figure 1.3.5.) If we choose a specific branch of /Z, we also choose which of the two possible square roots we shall obtain. Various geometric statements can be made concerning the map z +> z* that also give information about the inverse, z ++ /Z. For example, a circle of radius r described hy the set of points re!,0 < 0 < 2z, is mapped to re, a circle of radius 17; as ro“ moves once around the first circle, the image point moves twice around (see Figure 1.3.6). ‘The inverse map does the opposite: as z moves along Che aaa ea er tice Domains on which z + e* and z ++ log z are inverscs have already beon discussed (sce Figure 1.3.2). Note that the lines y = constant, described by the points x + ty as 2 varies, are mapped by the function z+ e* to points ee, which is a ray with argument y. As z ranges from —0o to -+50, the image point on the ray goes from 0 out to infinity (see Figure 1.3.7). Similarly, the vertical tine 2 = constant is mapped to a circle of radius e*. If we restrict y to an interval of length 2x, the image circle is described ance, but if y is unrestricted, the image circle is described infinitely many times as y ranges from36 Chapter 1 Analytic Functions Figure 1.3.5: Squaring function and its inverse. y ——., ey Figure 1.3.6: Effect of the squaring function on a circle of radius r. ~00 to +00. The logarithm, being the inverse of e*, maps points in the opposite direction to ¢*, as shown in Figure 1.3.7. Because of the special nature of the striplike in Figures (.3.2 and 1.3.7 (on them e* is one-to-one) and because of the periodicity of e*, thesc regions deserve a name. They are usually called period strips of e*. Worked Examples Example 1.3.13 Find the real and imaginary parts of exp(e*). (It is common to use expw as another way of writing e”.) Solution Let z =x + iy; then e* = c* cosy + ée*siny. Thus, expe* = e *¥[cos(e* siny) + isin(e* sin y)).Figure 1.3.7: Geometry of e* and log z. Therefore, Re (expe®) = (e°" ™¥) cos(e*siny) and Im (oxpe*) = (e* ©”) sin(c*siny). Example 1.3.14 Find all the values of i. Solution = etlest = gillon le (én /Hmn)it — (g-2eM)e—a/? = g-BH(nt1/9), Co ee ee n=0,+1,42,.... Example 1.3.15 Solve cosz = 4 for z. Solution We know that z, = £(x/3 + 2nn), where 7 is an integer, solves the equation cos2z = 3; we shall show that z,,7 = 0,:1...., are the only solutions; that. is, there are no solutions off the real axis. We are given een 1 oe oo) ‘Therefore, e* — e +1 = 0, and so by the quadratic formula, e* = } + V3i/2. Hence iz = log(3 + V3i/2) = + log(3 + V3i/2), since } + V3i/2 and 3 ~ V3i/2 are checked to be reciprocals of one another. We thus obtain z= ate (3 +#,) - i (log + Fit 2mni) = + (F + 2mm). Example 1.3.16 Consider the mapping z+ sinz. Show that lines parallel to the real azis are mapped to ellipses and that lines parollel to the imaginary axis are mapped to hyperbolas.38 Chapter 1 Analytic Functions Solution Using Proposition 1.3.4 (also see Example 1.3.14), we get sinz = sin(x+ ty) = sinxcos(iy) + sin(iy) cosz = sinzooshy+¢tsinhycosx where S4eT and sinhy = 272" Suppose that y = yp is constant; if we write sin z = u + iv, then we have cosh y = a 2 v ote ada. since sin? x + cus? = 1. This is an ellipse. Similarly, if 2 = xo is constant, from cosh? y — sinh” y = 1 we obtain uw v sin? xq cos? 9 =1, which is a hyperbola. Exercises 1. Express in the form a + bi: (a) et (b) sin(i +4) 2. Express in the form a+ bi: (@) & (b) cos(2-+ 34) 3. Solve (a) cosz = $+ i (b) cos 2 = 4 4. Solve (a) snz=343 (b) sinz = 4 5. Find all the values of (2) log (b) logé§1.3 Some Elementary Functions 39 6. Find all the values of (a) los(-#) (b) log(d +4) 7. Find all the values of () (- b) (+a 8. Find all the values of (@) (-1) (b) 2 9. For what values of z is (e) = cf? 10. Let /- denote the particular square root defined by vr(cosé + isin 6) = r"/*|cos(4/2) + isin(/2)},0 < @ < 2r; the other square root is 11/2 {cos|(0 + 2)/2] + isin{(@ + 2n)/2}}. For what values of z docs the equation V2? = z hold? 11.* Along which rays through the origin (a ray is determined by arg z = constant) does lim, soo |o*} exist? 12. Prove the identity 1 14iz\ 7 z=tan [me (H28 . 13. Simplify e*”,¢, and e'/*, where z= x + iy. For e'/* we specify that 2 #0. 14, Examine the behavior of eV as z — -too and the behavior of e+#¥ as yo to. 15.* Prove that sin(—2) = —sin z; cos(—2) = cos; sin(x/2— z) = cosz. 16. Define sinh and cosh on all of C by sinhz = (¢* — e7*)/2 and coshe = (e +e7*)/2. Prove that (a) cosh? z — sinh? z = 1 (b) sinh(2, + 22) = sinh 2; cosh z2 + cosh 2; sink 2 (c) cosh(2; + 22) = cash 2; cosh 22 + sinh z; sinh 20 Chapter I Analytic Functions (d) sink(x + iy) = sinh zoosy + icosh zsiny (c) cosb(z: + ty) = cosh zeosy +isinhrsiny 17.° Use the equation sin z = sin z cosh y+tsinh y cos where z = x-++iy to prove that [sinh y] < [sin 2] < | cosh yl. 18. Lf 6 ix real, prove that fa| = fal’. 19. Is it true that [a] = fal! for all a,b € C? 20. (a) For complex numbers a, b,«, prove that a%a* = a!*¢, using a fixed branch of log. (b) Show that. (ab)* = a°l if we choose branches so that log(ab) = loga + Jog b (with no extra 2mni). 21. Using polar coordinates, show that z++ 2+ 1/z maps the circle |z] = 1 to the interval [2,2] on the 2 axis. 22. (a) The nap z++ 23 maps the first quadrant onto what? (b) Discuss the geometry of z~ 9/2 as was done in the text for /z. 23.° ‘The map z ++ 1/z takes the exterior of the unit circle to the interior (ex- cluding zero) and vice versa. To what are lines arg z = constant mapped? 24. What are the images of vertical and horizontal lines under z ++ cos 2? 25. Under what conditions docs loga’ = bloga for complex numbers a,b? (Use the branch of log with -z <8 < a.) 26. (a) Show that under the map z ++ 2%, lines parallel to the real axis are mapped to parabolas. (b) Show that under (a branch of) z + V2, lines parallel to the real axis are mapped to hyperbolas. 27. Show that the n nth roots of unity are t,w,w?,w4,...,w"!, where w = eniin, 28. Show that the trigononictric identities can he deduced if e472) = eM! . gi? is assuined. 29.° Show that sinz = 0 iif z = kz.k =0,41,+2,. 30. Show that the sine and cosine are periodic with minimum period 27; that is, that (a) siu(z + 2n) = sin z for all z. (b) cos(z +27) = cos z for all z. (c) sin(z +w) = sinz for all z implies w = 2am for some integer n.§1.4 Continuous Functions 4] (€) cas(z +) = cos z for all z implies w = 2am for some integer n. 31. Find the maxintum of |cos2| ou the square 0 < Rez < 2n.0< kz ¢ 2n. 32. Show that log z = 0 iff z = 1, using the branch with -w < argz <7. 33. Compute the following quantitics numerically to two significant figures: (a) e* 24646 () log(1.2- 3.07) (ce) sin(8.1i — 3.2) 34.° Show that the function sin z maps the strip —7/2 < Rez < 4/2 onto the set C\{z | Luz = 0 and |Re2| > 1}. ‘35.° Discuss the inverse functions sin~' z and cos~' z. For example, is sin z one- to-oue on the set defined by 0 < Rez < 27? 1.4 Continuous Functions In this section and the next, the fundamental notions of continuity and differen- tiability for complex-valued functions of a complex variable will be analyzed. The results are similar to those learned in the calculus of functions of real variables. These sections will be concerned mostly with the underlying theory, which is ap- plied to the elementary functions in §1.6. Since C is R? with the extra structure of complex multiplication, many geometric concepts can be translated from R? into complex notation. This has already been donc for the absolute value, |z{, which is the same as the norm, or length, of z regarded as a vector in R?. Furthermore, we will usc calculus for functions of two variables in the study of functious of a complex variable. Open Sets We will need the notivn of an open set. A sct A C C = R? is called open when, for each point zp in A, there is a ronl number € > 0 such that z € A whenever Jz — zo <¢. Sce Figure 1.4.1. ‘The value of ¢ may depend on 29; a5 29 gels close to the “edge” of A,¢ gets smaller. Intuitively, a set is open if it does not contain any of its “boundary” or “edge” points. For a uumber r > 0. the r neighborhood or r disk around a point zo in C is defined to be the set D(zo:r) = {z € C | |z - zo] 0, the disk A = {z € C | |z — ta| < r} is itself open. A deleted r neighborhood is an r ncighborhood whose center point has been removed. Thus a deleted r-neighborhood hes the form D(z;r)\{20}, which stands for the set D(29;r) minus the singleton set, {zo}. See Figure 1. A neighborhood of a point 2 is, by definition, a set containing sume r disk around zo. Notice that a set A is open iff for each zo in A, there is anr neighborhood of zq wholly contained in A. The basic properties of upen sets are collected in the next propusition.Figure 1.4.1: Open set. (a) Figure 1.4.2: (a) r-Neighborhood. _(b) Deleted r-neighborhood. Proposition 1.4.1 (i) € iy open. (i) The empty set @ is open. (iit) The union of any collection of open subsets of C és open. {iv) The intersection of any finite collection of open subsets of € is open. Proof The first two assertions hold almost by definition; the first because any € will work for any point 29, and the second because there are no points for which ‘we are required to find such an ¢. The reader is asked to supply proofs of the last two in Exercises 19 and 20 at the end of this section. @ Mappings, Limits, and Continuity Let A be a subset of C. Recall that a mapping f : A — C is an assignment of a specific point f(z) in C to each point zin A. The set A is called the domain of f, and we say f is defined on A. ‘When the domain and the range (the set of values f assumes) are both subsets of C, as here, we speak of f as a complex function of a complex variable. Alternatively, we can think of f as a map f : A C R? — R?; then f is called§1.4 Continuous Functions 43 a vector-valued function of two real variables. For f : A C C — C, we can let z=a-+iy = (x,y) and define u(x,y) = Re f(z) and u(z,y) = Im f(z). Then u and v are the components of f thought of as a vector function. Hence we may uniquely write f(x + iy) = u(z,y)+tv(z.y), where u and w are real-valued functions defined on A. Next we consider the idea of limit in the sctting of complex numbers. Definition 1.4.2 Let f be defined on a set containing sume deleted r neighborhood of zy. We say that f has the limit a as 2 -+ zy and wrile Jim f(e)=«. when, for every ¢ > 0, there is a5 > 0 such that for all z € D(zo;r) satisfying 2 fz and |z — 29| <6, we have [f(z) — al 0. There is a 5 > 0 such that 0 < |z — z| < 6 implies that |f(z) — a] < € and |f(z) — o| < €. Choose such a point z £ zy (because J is defined in a deleted neighborhood of 29). Then, by the triangle inequality, fa — 8] < lo — f(z)| + f(z) ~ 4 < 2c, a contradiction. Thusa=6. Proposition 1.4.4 Iflim,—., f(z) = 0 and lim,—s, 9(z) = 6, then (i) lim. gL f(z) + 9{2)] = 0 +8. (ii) lim, —s9/f(z)9(2)] = ab. (iii) lim, [f(2)/9(2)] = a/b if b #0. Proof Only assertion (ii) will be proved here. The proof of assertion (i) is easy, and proof of assertion (iii) is slightly nore challenging, but the reader can get. the necessary clues from the corresponding real-variable case. To prove assertion (ii), we write (F(z)o(z) — ab] < [Ff (2)9{z) — F(z)b| + [f(2)b— ab] (triangle inequality) = [F(2)Mo(z) - 41 + 1/(z) - alld] (factoring).44 Chapter 1 Analytic Functions To estimate each term. we choose 6, > 0 so that 0 < [z — zo| < 4) implies that Is(z) — a] < 1, and thus [f(z)| < [a] + 1, since [f(z) — a] > f(z) — lal. by Proposition 1.2.5(vi). Given ¢ > 0, select positive numbers 6, and 3 so that 0 < |z— 20] < d2 implies f(z) — al < €/2((b] + 1) and 0 < |z — 20] < ds implies lo(z) - | < €/2(la] +1). Let 6 be the smallest of 5), 52, 53. If 0 < 2 — 20f < 6, we have Uy (e)at2) — ab] < LfC2) lle) ~ 81+ Ls) — ot bt : : 0, there is a 4; > 0 such that |h(w) - h(a)| < ¢ whenever jw — a] < 6; and aS > O such that {f(z) — al < 4, whenever 0 < |z ~ zo| < 5. Therefore we get {h( f(z) — h(a)] < « whenever 0 < |z— zl < 6, which establishes (i). A proof of (ii) follows from (i) aud is requested in Exercise 22 at the end of this section. @ Sequences The concept of convergent sequences of complex uumbers is analogous to that for sequences of real numbers studied in calenlus. A sequeace ~.n = 1,2.3.... of points of C converges to < if and only if far every « > 9. there = an imeger N such that n > V implies iz, - =i <¢. The Seat cf = eorene is expressed as§1.4 Continuous Functions 45 Limits of sequences have the same properties, obtained by the same proofs, as limits of fanctions. For example, the limit is unique if it exists; and if zy — z9 and Ww, — wo, then (i) Zn + Wp 29 + wo. (ii) Zt + otto. Gil) 2n/ty 7 20/109 (if wo and wp are not: 0). Als0. 2y — 29 iff Re z, -+ Re zg and Imz, — Im zp. A proof of this for functions is requested in Exercise 2 at the end of this section. A sequence 2, is called » Cauchy sequence if for every ¢ > 0 there is an integer NV such that |z,, — Zm| < whenever both x > N and m > N. A hasic property of real numbers, which we will accept without proof, is that every Cauchy sequence in R converges. More precisely, if {an }92., is a Cauchy sequence of real unmbers, then there is a real number zo such that limps 2% = to. This is equivalent to the completeness of the real munber system.’ From the fact that 2, — zp iff Re 2, — Rezo and Im z, — lm zp, we can conchide that every Cauchy sequence in converges. This is a technical point, but is nseful in convergence proofs, a5 we shall see in Chapter 3. Jt should be noted that a link exists between sequences and continuity; namely, Sf: AC €— Cis continous iff for every convergent sequence z» — 2 of points in A (that is, z, € A and 29 € A), we have f(z,) — f(z). The student is requested to prove this in Exercise 18 at tho end of this section. Closed Sets A subsct F of C is said to be closed if its complement, C\F = {2€C|z¢ F}, is open. By taking complements and using Proposition 1.4.1, one discovers the following properties of closed sets. Proposition 1.4.7 (i) The empty set is closed. (ii) C és closed. (iii) The intersection of any collection of closed subsets of C is closed. (iv) The union of any finite collection of closed subsets of C is closed. Closed and open sets are important for their relationships to continuous func- tions and to sequences and for other constructions we will sec later. Proposition 1.4.8 A set F CC is closed iff whenever 2), 22, 73,... 18 @ sequence of points in F’ such that w = lity —oe Zn enists, then w € F. Seo, for cxamplo, J. Marsden and M. Hoffman, Slomentary Classical Analysis, Second Edition (New York: W. H. Freeman and Company, 1993).46 Chapter 1 Analytic Functions Proof Suppose F is closed and 2 is a sequence of points in F. If D(w;r) is any disk around w, then by the definition of convergence, z, is in D(w;r) for large enougit n. Thus, D(w;r) cannot be contained in the complement of F. Since that complement is open, w must not be in the complement of F. Therefore, it must be in F. If F is not closed, then the complement is not open. In other words, there point w in C\F such that no neighborhood of w is contained in C\F. In particular, ‘we may pick points z, in FM.D(w;1/n); this yields a convergent sequence of points of F whose limit is not in F. Proposition 1.4.9 if f:C— C, the following are equivalent: (i) f is continuovs. (ii) The inverse image of every closed set is closed. (iii) The inverse image of every open set is open. Proof ‘To show that (i) implies (ii), suppose f is continous and F is closed. Let 21,22, 23,--. be a sequence of points in f—'(F) and suppose that 2, -+ w.5 Since f is continuous, f(zn) — {(w). But the points f(z_) are in the closed set F, and 50 f(w) is also in F. That is, w is in -"(7). Proposition 1.4.8 shows that f-'(F) is closed. ‘To show that (ii) implies (iii), let U be open. Then J = C\U is closed. If (ii) holds, then f-?(F) is closed. Therefore, C\f-"(F) = J-'(C\F) = J-"(U) is open. implies (3), fix 2 and let « > 0. Then zo is a member of the open set f~!(D(f(za);«)). Hence there is a 6 > 0 with D(z0i6) ¢ fF" (D(f(z0}i4))- This says precisely that |{(z) — f(2o)| < ¢ whenever |z — 2o| < 6. We thus get exactly the inequality needed to establish continuity. © ‘To handle continuity on a subset of C, it is convenient to introduce the notion of relatively open and closed sets. If A C C, a subset B of A is called open relative to A if B = ANU for some open set U. It is said to be closed relative to A if B= ANF for some closed set F. This leads to the following proposition, whose proof is left to the reader. Proposition 1.4.10 If [: AC, the following are equivalent: (i) f #8 continuous. (ii) The inverse image of every closed set is closed relative to A. 3) eld Deitel cil teri de e AO ee te ee fo Sereno$1.4 Continuous Functions a7 Cannected Sets This subsection and the ncxt study two important classes of sets which occupy to some exteut the place in the theory of complex variables held by intervals and by closed bounded intervals in the theory of functions of a real variable. These are the connected sets and the compact sets. A connected set should be one that “consists of one piece”. This may be approuched from a positive point of view—“Any point can be connected to any other"—or from a negative point of view—“The set cannot be split into two parts”. This leads to two possible definitions. Definition 1.4.11 A set C C C is path-connected if for every pair of points a,b in C there is a continuous map 7: |0,1)—+ C with 7(0) =a and (1) =b. We call 7 4 path joining a and b. Onc cau often casily tell if a set is path-connected, as is shown in Figure 1.4.3. The negative point of view suggests a slightly diferent definition. Figure 1.4.3: Regions in (a) and (b) are connected while the region in (c) is not. Definition 1.4.12 4 set CC C is not connected (see Figure 1.4.4) if there are open sets U and V such that (@) CcuuV i) CNU ¢ @ ond CNV ZB (ii) (CA) N(CAV) =e Af a set fails to be “not connected", it is called connected. ‘The notions of relatively open and closed sets allow this to be rephrased in terms of subsets of C. Since the intersection of C with U is thc same as its intersection with the complement of V, the sct CNU is both open and closed relative to C, as is CNV. This proves the next result. Proposition 1.4.13 A set C is connected if and only if the only subsets of C that are both open and closed relative to C are the emply set and C itself.48 Chapter 1 Analytic Functions Figure 1.4.4: The set C is not connected. ‘The next two propositions give the relationship hetween the two definitions. The two notions are vot in general equivalent, but they are for open sels. The proof of this last assertion (given below iu Proposition 1.4.15) illustrates a fairly typical way of using the notion of connectivity. Onc shows that a certain property holds everywhere in C by showing that. the set of places where it holds is not empty and is both relatively open and relatively closed. Proposition 1.4.14 A path-connecied set is connected. Proof Suppose C is a path-conuected set and D is a nonempty subset of C that is both open and closed relative to C. If C # D, thore is a point 2; in D and a point 22 in C\D. Let 7: [a,b] — C he a continuous path joining z; to z. Let B=7"(D). Then B is a subset of the interval {a,0], since 7 is continuous. (See Proposition 1.4.10.) Since a is in 8, B is not empty. and [a,4]\8 is not empty since it contains b. This argument shows that it is sufficient Lo prove the theorem for the case of an interval [a, 6}. We thus need to establish that intervals on the real line are connected. A proof uses the least upper bound property (or some other characterization of the fact that the system of real numbers is complete). Let x = sup B (that is, the least npper bound of B). We find (hat is in B since B is closed. Since B is open there is a neighborhood of x contained in B (note that x # b, since b is in fa, ]\B). Thus, for some € > fl, the point + is in B. Thus z cannot be the least upper bound. This contradiction shows that such a sct B cannot exist. © A connected set need not be path-connected,* but if it is open it must be. In fact, more is truc. A standard example is given by letting C be the union of the graph of y = sin 1/z, where 2 > 0, and the lino segment —1 < y < 1,2 =0. This set is contected but not path-connected.§1.4 Continuous Functions 49 Proposition 1.4.15 If C is an open connected set and a and b are in C, then there is a differentiable path + : [0,1] —+ C urith (0) = a and (1) = 6.7 Proof Let a be in C. If zq is in C, then since C is open, there is an € > 0 such that the disk D(29;«) is contained in C. By combining a path from a to 2 with one from zy to z that stays in this disk, we sce that zq can be connected to a by a differentiable path if and only if the same is true for every point z in D(za:¢). This shows that both the sets A= {z€C| 2 can be comected to a by a differentiable path} B= {z € C| z cannot be so connected to a} are open. Since C is connected, either A or B must be empty. Obviously it must. be B. See Figure 1.45. Figure 1.4.5: An open connected set is patl-connected. Because of the importance of open connected sets, they are often designated by @ specia) term. Although the usnge is not completely standard in the literature, the words region and domain are often used. In this text these terns will be used synonymously to mean an open connected subsct of C. The reader should be careful to check the meanings when these words are encountered in other texts. ‘The notion of connected sets will be of use to us several times. One observation is that a continuous function caunot break apart a connected set. Proposition 1.4.16 If f is a continuous function defined on a connected set C, then the tmage set f(C) is also connected. ‘7Dilforontinbility of -y means that each component of 7 is differentinble in the usual sense of umvariahle calculus.50 Chapter 1 Analytic Functions Proof If U and V arc open sets that disconnect f(C), then f-'(U) and f-"(V) are open sets disconnecting C. Be careful. This proposition works in the opposite direction from the one about open and closed sets. For continuous functions, the inverse images of open sets are open and the inverse images of closed sets are closed. But it is the direct images that are guaranteed to be connected and not the inverse images of connected sets. (Can you think of an example?) The same sort of thing will happen with the class of sets studied in the next subsection, the compact sets. Compact Sets The next special class of sets we introduce is that of the compact sets. These will turn out to be those subsets K of C that are bounded in the sense that there is a number Af such that |z| < Af for every z in and that are closed. Onc of the nice properties of such scts is that every sequence of points in the sct. mmust have a subsequence which converges Lo some point in the set. For example, the sequence 1,4, 3,1, $,1,7,2,2,... of points in JO, 2[ bas the subsequence 1,4, 2,..., which converges Lo the point 0, which is not in the open interval ]0, 2[ but is in the closed interval {0,2}. Note that in the claimed property, the sequence itself is not asserted to converge. All that is claimed is that some subsequence does; the example shows that this is necessary. As often happens in mathematics, the study consists of three parts: (i) An easily recognized characterization: closed and bounded (ii) A property we want: the existence of convergent subsequences (iii) A technical definition useful in proofs and problems In the case at hand, Lhe technical definition involves the relatiouship between compactness and open sets. A collection of open sets U, for a in some index set A is called a cover (or an open cover) of a sct K if KC is contained in their union: K C UseaUa- For example, the collection of all open disks of radius 2 is an open cover of C: U, = D(z;2) CC Use D(z; 2). Tt may be, as here, that the covering process has been wasteful, using more sets than neaded. In that case we may wse only some of the sets and talk of a subcover, for example, C C UnmezD(n + mi;2), where Z denotes the set. of integers. Definition 1.4.17 A set K is compact if every open cover of K has a finile subcover. That is, if U, is any collection of open sets whose uniuu contains K, then there is a finite subcollection Us, ,Uo,,--- Un, such that K C Ug, Ug, U...UUa,- Proposition 1.4,18 The following conditions are equivalent for a subset K of C (or of R):§1.4 Continuous Functions Bt (i) K in closed and bounded. (ii) Buery sequence of points in K has a subsequence which converges to some point in K. (ili) K is compact. This proposition requires a deeper study of the completeness properties of the real numbers than it is necessary for us to go into here, so the proof is omitted. It may be found in most advanced calculus or analysis texts. It is easy to see why (i) is necessary for (ii) and (iii). If K is not bounded we can select 2; in K and then successively choose 2 with |z2] > |z:| +1 and, in general, 2, with len] > |zn--1|-+1. This gives a sequence with no convergent subscquence. The open disks D(0;n),n = 1,2,3,..., would be an apen cover with no finite subcover. If K isa sct in C that is not closed, tlien there is a point win C\X and a sequence 21) 22)--. Of points in K that converges to w. Since the sequence converges, w is the only possible limit of a sitbsequence, so uo subsequence can converge to a point of K, The sets {z such that |z — w| > 1/n} for n = 1,2,3,... form an open cover of K with no finite subcover. The utility of the technical Definition 1.4.17 is illustrated in the following results. Proposition 1.4.19 If K is a compact sct and f is a continuous function defined on K, then the image set {(K) is also compact. Proof If U, is an open cover of {(K), then the sets f-"(U,) form an open cover of A’. Selection of a finite subcover gives KC J Wa,)U--. US" (Yon) so that {(K)C Up, U...UUa,, Theorem 1.4.20 (Extreme Value Theorem) if K is a compact set and f : K +R is continuous, then f attains finite mazimum and minimum values. Proof The image f(KC) is compact, hence closed and bounded. Since it is bounded, the numbers M = sup{f(z) | z € K} and m = inf{f(z) | z € K} are finite. Sinco J(K) is closed, m and M are included in f(K). Another illustration of the use of compactness is given by the following lemma, which asserts that the distance from a compact set to a closed set is positive. That is. there must be a definite gap between the two sets. Lemma 1.4.21 (Distance Lemma) Suppose K is compact, C is closed, and KN C=. Then the distance d(K,C) from K lo C is greater than 0. That is, there ts a number p > 0 such that |z — w| > p whenever z is in K and w is in C. See, for example, J. Maruden and M, Hoffman, Elementary Classical Analysis, Second Edition (New Yorke W. H. Frocinan and Company, 1993).52 Chapter 1 Analytic Functions Proof The complement of C, namely the set U = C\C, is an open set and K CU, 30 that each point z in K is the center of some disk D(z; p(z)) C U. The collection of smaller disks D(z; p(z)/2) also covers K, and by compactness there is a finite mumber of disks, which we denote by Dy = D(z: p{z1.)/2)k = 1,2,3,...,N that cover IC. (Seo Figure 1.4.6.) Let py = p(z.)/2 and p = min(p1, p2,- pw). If z is in K and w is in C, then z is in Dy for some k. and so fz — 24] < px- But fw — zx] > plzx) = 2pr- Thus, [2 — 10] > px > p. a Figure 1.4.6: The distance between a closed set C and e compact set Kis greater than zero. Uniform Continuity Remember that a function is said to be continuous ov a set K if it is continuows at cach point of K. This is called a local property since it is defined in terms of the hebavior of the function at or near each point and can be determined for each point. by looking only near the point and not at the whole ‘sot at once. This is in contrast to global properties of a function. which depend on its behavior on the whole sct. ‘Aun example of a global property is boundedness. Saying that a function f is bounded by some number Af on a sct K is an assertion that depends on the whole set at once. If the function ix continuous it is certainly bounded near each point, but that would not automatically say that i is bounded on the whole set. For example, the function f(z) = 1/x is continuous on the open interval }0, I[ but is certainly not bounded there. We have seen that if a function f is continous on a compact set K, then it is bounded ou K and in fact the bounds are attained. ‘Thus, compactness of IX allowed us to carry the local boundedness near each point given by continuity over to the whole set. Compactness often can be used to make such a shift from a local property to a global one. The following is a global version of the notion of continuity. Definition 1.4.22 A function f : A — C (or R) is uniformly continuous on A41.4 Continuous Functions 53 uf for every choice of ¢ > 0 there is a 5 > 0 such that |f(s) — f(t)| < € whenever s and t are in A and |s - | <6. Notice that the difference between this and the definition of ordinary continuity ts that now the choice of 6 can be made so that the same 6 will work everywhere in the set A. Obviously, uniformly continuous functions are continuous. On a compact set the apposite is true as well, Proposition 1.4.23 A continuous function on a compact set is uniformly contin- uous. Proof Suppose f is a continuous function on a compact set 1, and let ¢ > 0. For vach point ¢ in K, there is a number 6(l) such that [f(s) — f(t)] < «/2 whenever s—4€ < d(t). The open sets D(t;4(t)/2) cover K, so by compactness there are a finite number of points t),f2,... ,t such that the gets Dz = D(ty:6(ts)/2) cover K. Let = 5(te)/2 and set § equal to the minimum of 61 iy. Hs] < 5, then ¢ isin Dy for sume k, and so |t~ty| < d,. Thus |/(t) ~ f(tx)| <¢/2. But also, |s — te] = [s— £46 tel S Js — 4] + b— te] SO + de < Ste) and so [f(s) — f(te)| $€/2. Thus (40s) - F@) = UF (6) - Fee) + Se) - FOI S Mss) - Lte)l + te) - () < e/2 +¢/2 =e. We have produced a single 6 that works evorywhere in K, and so f is uniformly continuous, Path-Covering Lemma The notion of uniform continuity is a very powerful one that will be uscful to us several times. We use it first in conjunction with the Distance Lemma and some of the properties of compact sets to establish a useful geometric lemma about curves in open subsets of the complex plane. This lemma will be useful later in the text, particularly for studying integrals aloug such curves. It says that the curve can be covered by a finite number of disks centered aloug the curve in such a way that each disk is contained in the open set and each contains the centers of both the preceding and the succeeding disks along the curve, (See Figure 1.4.7.) Lemma 1.4.24 (Path-Covering Lemma) Suppose 7 : [a,0] —+ G is a contin- nous path from the interval [a,b] into an open subsct G of C. Then there are a auniber p > 0 and a subdivision of the interval a = tg < ly < ty <...< tn =b such that (i) D(t);0) CG for alk i) a(t) € D(y(to):0) fora st Sty (ili) 4(4) € Deite): a) for te-1 SOS bear (iv) y(t) € D(r(te)sp) for ta StS byFigure 1.4.7: A continuous path in an open set can be covered by a Ginite number of well-overlapping disks. Proof Since + is continuous and the closed interval fa, 4] is compact, the image curve K = 7(la.}}) is compact. By the Distance Lemma 1.4.21 there is a number p such that each point on the curve is a distance at least p from the complement of G. Therefore, D(y(t);n) ¢ G for every t in [a,b]. Also, since +y is continuous on the compact set: [a. 6], it is uniformly continuous. and there is a number 6 > 0 such that fy(t) — 2(5)] < p whenever |s— ¢| < 5. Thus if the subdivision is chosen fine enough so that 4; — t.—-1 <6 for all k = 1,2,3, NW, then the conclusions of the theorem hold. Riemann Sphere and Point at Infinity For some purposes it is convenient to introduce: a point oo in addition to the points z € C. One must be careful in doing 50, Since it can lead to confusion and abuse of the symbol oo. But with care it can be uscful, and we certainly want to be able to talk intelligently about infinite imits and limits at infinity. Tn contrast to the real line, to which -+-oc and —oo can be added, we have only one 00 for C. The reason is that C has no natural ordering as R does. Formally we add a symbol oc to C to obtain the extended complex plane, C, and define operations with oo by the rules z+o0 = 2-00 = ato = co-oc = z = = oy for z € C. Notice that some things are not defined: 00/00, - 00,00 — 00, and #0 forth are indeterminate forms for essentially the same reasons that they are in the calculus of real numbers. We also define appropriate limit concepts: provided z 40 3888§1.4 Continuous Functions 55 lim—eo f(z) = zo means: For any € > 0, there is an R > 0 such that |f(z) - zal <¢ whenever |z| > R. Tim, 2p f(z) = 00 means: For any R > 0, there is a 5 > 0 such that |f(2)| > R whenever |z — 20] < 6. For sequences: litn,—os 2n = 00 means: For any R > OQ, there is an N > 0 such that |z,| > R whenever n > N. Thus a point z € C is “close to 00” when it lies outside a large circle. This type of closcness can be pictured geometrically by means of the Riemann sphere shown in Figure 1.4.8. By the method of stereographic projection illustrated in this figure, 4 point 2’ on the sphere is associated with cach point z in C. Exactly oue point on the sphere S has been omitted—the “north” pole. We assign oo in C to the north pole of S. We see geometrically that z is close to oo if and only if the corresponding points nre close on the Riemann sphere in the usual sensc of closeness in RS. Proof of this is requested in Exercise 24. Figure 1.4.8: Riemann sphere. The Riemann sphere S represents a convenient geometric picture of the extended plane € = CU {oo}. The sphere docs point up one fact about the extended plane that is sometimes useful in further theory. Since S is a closed hounded subsct of R®, it is compact. Therefore every sequence in it bas a convergent subsequence. Since stereographic projection makes convergence on S coincide with convergence of the sequence of corresponding points in C, the same is true there. That is, € is compact. Every sequence of points in € must have a subsequence convergent in €. Caution: Since the convergence is in the extended plane, the limit might be 00, in that case we would normally say that the limit does not exist. Basically we have thrown in the point at infinity as another available limit so that sequences that. did not formerly have a limit now have onc. The sphere can be used both to help visualize and to meke precise some notions about the behavior of functions “at infinity” that we will mect in future chapters.56 Chapter 1 Analytic Functions Worked Examples Example 1.4.25 Where is the function 42241 1)= continuous? Solution Since sums. products, and quoticuts of contimons functions are contin- uous except where the denominator is 0, this function is continuous on the whole plane except at the cube roots of —1. In other words, this function is continuous on the set C\ for, ef i/#, 1}. Example 1.4.26 Show that the set {z | Rez > 0} #s open. Solution A proof cau be based on the following properties of complex numbers (sce Exercise 1): Ifw €C, then () [Rew < jul (8) [Ima < ful Gil) feo] < | Rew| + | Lm Let U = {z | Rez > 0} and let zo be in U. We claim that the disk D(zo;Re 2) lies in U. To see this, let z be in this disk. Then |Rez — Rez | = |Re(z — 20)| < |z—zo| < Rezg, and so Rez > O and z isin U. Thus, D(z0;Re zo) is @ neighborhood af zp that is contained in U. Since this can be done for any point zo which is in U, the set U is open. See Figure 1.4.9, 4 Figure 1.4.9: Open right half-plane.11-4 Continuous Functions 87 Example 1.4.27 Prove the following statement: Let Ac C be an open sct and 2) € A, and suppose that D, = {z such that |z — zq] r such that D(2o;p) C A- Solution We know from the Extreme Value Theorem 1.4.20 that, a continous real-valued function on a closed bounded set in € attains its maximum and min- imum at some poiut of the set. For z in D, let f(z) = inf{|z —w| such that «© € C\A}. (Here “inf” means the greatest lower bound.) In other words. f(z) is the distance from 2 to the complement of A. Since A is open, f(z) > 0 for each = in D,. We can also verify that f is continuous. Thus f assumes its minimum at some point 2; in D,. Let p = f(z:) +1, and check that this p hes the desired properties. See Figure 1.4.10. Figure 1.4.10: A closed disk in an open set may be enlarged. A eet Example 1.4.28 Find lim; co eee et) +1 Solution tim 32414227 -241 _ pg 34227? — 2S 42°F 3 apo et sao 142-8 using lim,—o.2z~! =0 and the basic propertics of limits. Exercises 1. Show that if w € C. thenChapter 1 Analytic Functions {a) [Rew] < [wl (h) [Lmew} < fo} {c) |v] < [Rew] + |Imw} 2.* (a) Show that [Re 2) — Rezo] < lax — 22| < [Rez — Rezo] + |Imz, — Im29| for any two complex numbers 2; and 22. (b) If (2) = u(z,y) + 0(z,y), show that aim J(2) = Jim u(z,y) +# Jim o(z,y) a iw exists if both limits on the right of the equation exist. Conversely, if the limit on the left. exists, slow that both limits on the right exist as well and equality holds. Show that f(z) is continuous iff u and v are. 3. Prove: If f is continuous and f(z) # 0, there is a neighborhood of z on which f is #0. 4, If 29 € €, show that the set {zo} is closed. Prove: The complement of a finite number of points is an open set. ‘Use the fact that a function is continuous if and ouly if the inverse image of every open set is open to show that a composition of two continuous functions is continuous. 7. Show that {(z) = is continuous. e 8 Show that f(2) = |2| is continuous. 9. What is the lurgest set on which the function f(z) = 1/(1—e*) is continuous? 10. Prove or find 2 counterexample if false: If lim,—., f(z) = a, ht is defined at the points f(z), and limy—» ii(w) = ¢, then lim.—., h({(z)) = ¢. (Hint: Could we have h(a) ¥ ¢?] 11, For what z does the sequence z, = nz" converge? 12." Define f : C + C by setting f(0) = 0 and by setting f(r[cos 0-+isin0]) = sind ifr > 0. Show that f is discontinuous at 0 but is continuous everywhere else. 13. For each of the following sets, state (i) whether or not it is open and (ii) whether or uot it is closed. (a) {2 such that [2] <1} {b) {z]0<|z| <1}21.4 Continuous Functions 59 15. Ww. 18. 19. (c) fz] 1 ¢ Rez <2} |. For each of the following sets, state (i) whether or not it is open aud (ii) whether or not it is closed. (a) {2 |unz > 2} (b) {21 <2} <2} (c) {2-1 < Rez <2} For each of the following sets, state (i) whether or not it is connected and (ii) whether or not it is compuct, (a) {2|1 < lz} <2} (b) {z such that |z| <3 and |Rez| > 1} (c) {z such that |Re2z| <1} (a) {z-such that |Rez| > 1} For each of the following sets, state (i) whether or not it is connected and (ii) whether or not it is compact. (a) {z|1 1} If ACC and f :C 4, show that C\f-(A) = f-(C\A). Show that f : A C C - C is continuous if and only if z, — 29 in A implies that f(2n) > f(z0)- Show that the union of any collection of open subsets of C is open. 20. Show that the intersection of any finite coltection of open subsets of C is open. 21. 22. 23. 24. Give un example to show that the statement in Exercise 20 is false if the word “Ginite” is omitted. Prove part (ii) of Proposition 1.4.6 by using part (i). Show that if Jz] > 1, then limn—so(2"/n) = 00. Introduce the chordal metric p on € by setting p(z1,22) = d(zj, 2) where 24 and 2 are the corresponding points ou the Riemann sphere aud d is the ugual distance between points in R°. (8) Show that z, — z in C if and only if p(zq,2) 0. (u) Show that 2, — oc if and only if a(zq,00) + 0. (c) If f(z) = (az + )/(cz +d) and ad ~ be # 0, show that f is continuous at 00.60 Chapter 1 Analytic Functions 1.5 Basic Properties of Analytic Functions Although continuity is an important concept, its importance in complex analysis is overshadowed by that of the complex derivative. There are several approaches to the theory of complex differentiation. We shall begin by defining the derivative as the limit of difference quotients in the same spirit as in calculus. Many properties of the derivative including useful computation rules follow from the properties of limits just as they do in calculus. However, there are some surprising and beautiful results special to the complex theory. Several different words are used to describe fumctions that are differentiable in the complex sense, for example. “regular”, “holomorphic”, and “analytic”. We will use the term “analytic” since it is used in calculus to describe functions for which the Taylor series couverges to the valuc of the function. An elegunt result of complex analysis justifies this choice of language. Ludeed, we will see in Chapter 3 that. in sharp distinction from the case of a single real variable, the assumption that a function is differentiable in the sense of complex variables guarantees the validity of the Taylor expansion of that function. Definition 1.5.1 Let f: A —C where ACC is an open set. The function f is said lo be differentiable (in the complex sense) at z € A if m £(2) = f(e0) at 2-2 exists, This limit is denoted by f'(z0), or sometimes by (df/d2)(20). Thus, f"(20) és @ complex number. The function f is said to be analytic on A if f is compler- differentiable at each 2 € A. The word “holomorphic”, which is sometimes used, is synonymous with the word “analytic.” The phrase “analytic at x" means f is analytic on @ neighborhood of zy. Note that the quotient £2) ~ f(z) 2% is undefined at z = zo, and this is the primary reason why deleted neighborhoods were used in the definition of init. Although the definition of the derivative J"(z9) is similar to that of the usual derivative of a function of a real varialile and they share many similar properties, the complex case is much richer. Note also that in the definition of f“(zo). we are dividing by the comple number z— zo and the special nature of division by complex numbers is n key consideration. The limit as z + zo is taken for an arbitrary = approaching 29 but not along any particular direction, The existence of f implies a great deal about f. Tt will be proven in §2.4 that if J’ exists, then all the derivatives of f exist (that is, f”, the (complex) derivative of f’, exists, and so on). This is in contrast to the case of a function g(x) of the real variable x, in which g'(z) can exist without the existence of g!(z).§1.5 Basic Properties 61 The analysis of what are called the Cauchy-Riemann equations in Theorem 1.5.8 will show how the complex derivative of f is related to the usual partial derivatives of f as # function of the real variables (x,y) and will supply a useful criterion for determining the existence of f'(zq). As in elementary calculus, continuity of f does uot imply diffcrentiability; for cxample, f(z) = |2| is continuons but is not differentiable (see Exercise 10 at the end of this section). However, as in one-variable calculus, @ differentiable function must be continuous. Proposition 1.5.2 If {'(zo) exists, then f is continuous at 2p. Proof By the sum rule for limits, we need only show that im Ue) Sea)] = But Jim U2) ~ sea) = Jim [LEL= Le — sy], which, by the product rule for limits, equals f’(zo)-0=0. The usual rules of calculus-—the product rule, the quotient rule, the chain rule, and the inverse function rule—can be used when differentiating analytic functions. ‘We now explore these rules in detail. Proposition 1.5.3 Suppose that f and g are analytic on A, where AC C is an open set. Then i) af +09 a on A and (af + bg)'(z) = af'(z) + bg'(z) for any complex numbers a and (ii) fg is analytic on A and (f9)'(z) = J'(z)9(2) + S(2)9'(2)- (iii) If g(z) #0 for all z € A, then f/g is analytic on A and LY’ (0) = £@9(2) - 92) (© wear (iv) Any polynomial ag + a)2 +...+ 0,2" is analytic on all of C with derivative @; + 2agz +... + naz", (v) Any rational function Og +412 +..-+On2" bo thz +... + bm2z™ is analytic on the open sct consisting of all z except those (at most, m) points where the denominator is zero. (Sex Review Exercise 24 for Chapter 1.)62 Chapter 1 Analytic Functions Proof The proofs of (i), (ii), and (iii) are all similar to the proofs of the corre- sponding results found in calculus. The procedure can be illustrated with a proof of (ii). Applying the limit theorems and the fact that lim,—., f(z) = {(z0) (Propo- sition 1.5.2), we get lim £(2)9(2) - F(z0)9(20) ene z—% = tim [£l2)9(2) = Fle)ole0) , f(2)9(e0) - f(e0)o(e0) rt z2-% zm = fn [e282 + ny [LDL 2-2 = $(20)a'(20) + J’(2)9(20)- To prove (iv) we must first show that f’ = 0 if f is constant. This is immediate from the definition of derivative because f(z) — f(zo) = 0. It is equally ensy to prove that dz/dz = 1. Then, using (ii), we can prove that a ge chet lam fon dee tai 2-2 232 An general, we see by induction that dz"/dz = nz"-'. Then (iv) follows from this and (i), and (v) follows from (iv) and (iii). For example, Set +ar- 2) =224+8 SF dz\z+1) 0 (z+1)?" ‘The student will also recall tht one of the most important rules for differenti- ation is the chain rule, or “function of a function” rule. To illustrate, é [(25 + 1)9] = 10(2* + 1)° - 32? = 3027(2 +1). This procedure for differentiating should be familiar; it is justified by the next result. Theorem 1.5.4 (Chain Rule) Let f : A and g: B+ C be analytic (A,B are open sets) and let f(A) CB. Then gof : A— C defined by (gof)(z) = 9(J(2)) is analytic and and Sto nie) = 9M) “st.§1.5 Basic Properties 63 The basic idea of the proof of this theorem is that if w = f(z) aud wo = (zo), then (F(2)) ~ a (2a)) _ glu) ~ lwo) | (2) - (0) | 2-m w— wo zZ- a and if we let z -> 29, we also have w — two, and the right side of the preceding equation thus becomes g'(wp)"(2). The trouble is that even if z # zo, we could have uw = wo. Because of this possibility, we give a more careful proof. (Although the chain rule here can be deduced from the chain rule for the usual derivative for functions of several variables—see the proof of 1.5.8—a separate proof is instruc- tive.) Proof Let wo = {(zo), and define, for w € B, 1) = Led = ao) _ or) if w 9 wo and h(wo) = 0. Since g/(wo) exists, / is continuous. Since the composite of continuous functions ix continuous, sim WU) = be(twa) = From the definition of k and letting w = f(z). we get (g0f)(z)—9(un) = [R(f(2)) + 9(19))Lf(z) — wo]. Note that this still holds if f(z) = woFor 2 # 29, we get (oe De) (90 12) esta + gm LL, As 2 — 2g, the right side of the equation converges to [0 + '(wo)] - L/’(zo)], so the theorem is proved. An argument similar to the one just given proves a slightly different version of the chain rule. Namely, if :Ja,0{— C is differentiable, we can differentinte the curve o(t) = f(7(t)) and obtain o(¢) = f’((2))-7/(¢). Here +y (t) is the derivative of yan a function Ja, b[—+ R?; that is, if y(t) = (x(t), y(t)), then /(t) = (2/(),9(4)) = a(t) + iy). ‘We now use the chain rule to prove a complex version of the following theorem from calculus: A function whose derivative is identically 0 must be constant. The Tesult illustrates the importance of regions, or open connected sets, in which we may, by Proposition 1.4.15, connect any two points by a differentiable path. Proposition 1.5.5 Let A C € be open and connected and let f : A + € be analytic. If f'(z) =0 on A, then f is constant on A.64 Chapter 1 Analytic Functions Proof Let 21,22 € A. We want to show that {(21) = f(z2). Let 7(t) be a path Joining 2, to zz. By the chain rule, df(4(t))/dt = f'(r(t)) -(t) = 0, since f* = 0. "Thus if f = u+-4v, we have du(7(t))/dt = 0 and du(y(t))/dt = 0. From calculus, we know this implies that u(7(t)) and v(7(¢)) are constant functions of t. Comparing the values at t = a and t = 6 gives us f(z1) = f(z). Clearly, connecteduess is needed because if A consisted of two disjoint pieces, we could let f = I on one piece and f = 0 on the other, Then f(z) would equal 0 but. f would not be constant on A. Conformal Maps The existence of the complex derivative f’ places severe but very useful restrictions on f. The first of these restrictions will be briefly discussed here. Another restriction will be mentioned when the Cauchy-Riemann equations are analyzed in Theorem 1.5.8. It will be shown that “infinitesimally” uear a point 2» at which f'(zo) # 0, f is a rotation through the angle arg f"(29) and a magnification by the factor |{*(z0). ‘The term “infinitesimally” is defined more precisely below, but intuitively it means that locally / is approximately a rotation together with a magnification (see Figure 1.5.1). If f"(zo) = 0, the structure of f is more complicated. (This point will be studied further in Chapter 6.) Y fe Figure 1.5.1: Conformal at zo. Definition 1.5.6 A map f : A — C is called conformal at 29 if there exist a0 € [0,2n[ and an r > 0 such that for any curve y(t) that is differentiable at 1 =0, for which y(t) € A and (0) = 29, and that satisfies (0) # 0, the curve o(t) = S(r(t)) is differentiable at t = 0 and, setting u = o'(0) and v = 7(0), we have |u| = rlv| and argu = argu + @(mod2z). A map is called conformal when it is conformal at every point. ‘Thus a conformal inap merely rotates and stretches tangent vectors to curves. ‘This is the precise meaning of “infinitesimal” as previously used. It should be noted§1.5 Basic Properties 65 that a conformal map preserves angles between intersecting curves. (By definition, the angle between two curves is the angle between their tangent vectors— see Figure 1.5.2). Figure 1.5.2: Preservation of angles by a conformal map. Theorem 1.5.7 (Conformal Mapping Theorem) iff : A — C és enalylic and if S'(z0) # 0, then f is conformal at zp with @ = arg f'(zo) andr = |f"(zo)|. fulfilling Definition 1.5.6. ‘The proof of this theorem is remarkably simple. Proof Using the preceding notation and the chain rule, we get u = o'(0) = f'(z0)- 1'(0) = f'G0)-v. Thus argu = arg f'(z0) + argu(mod2n) and [ul = 1f"(z0)|- |v. as required. ‘The point of this proof is that the tangent vector v to auy curve is multiplied by a fixed complex number, namely, f’(zo), no matter in which direction v is pointing. This is because, in the definition of f’(2o), lim,z, is “taken through all possible directions” as z — 29. Cauchy-Riemann Equations Recall that if f : AC C = R? — R? and if F(z,u) = (u(z,y), u(z,y)) = u(x,y) + 0(z,y), then the Jacobian matriz of f is of partial derivatives given by as the matrix of es ay oy & Or ay at each point (x,y). We shall relate these partial derivatives to the complex dcriva- live. From the point of view of real voriables, f is called differentiable with Df(zy) = Pele66 Chapter 1 Analytic Functions derivative the matrix Df (zo, yo) at (xa, yo) iff for any ¢ > 0, there is a 5 > 0 such that |(z,y) — (za, ¥0)| < é implies Ifz,u) - (20,40) — Df (zo, 40)[(2.¥) — (70-¥0)]1 < el(z,¥) — (20:40)! where Df(0, 0) - [(t.¥) — (t0,¥0)] moans the matrix Df(zo,yo) applied to the (column) vector (3:2) and |tw| stands for the length of a vector w. Here are some facts from the calculus of several variables that we shall use.” If f is differentiable, then the usual partials Ju/@z, @u/Oy,v/8z, and Gv/Oy exist and Df (zo.¥0) is given by the Jacobian matrix. The expression Df (29, yo){t) represents the derivative of f in the direction w. If the purtials exist and are continuous, then J is differentiable. Generally, then, differentiability is a bit stronger than existence of the individual partials. The main result connecting the partial derivatives and analyticity is stated in the next theorem. Theorem 1.5.8 (Cauchy-Riemann Theorem) Suppose A is an open set in C and f : AC © — C is a given function. Then f'(2o) exists if and only if f is differentiable in the sense of real variables and at (x9, yo) = zo, the functions u,v Satisfy (called the Cauchy-Riemann equations). Thus, if Bu/Be, 8u/Oy,80/Ax, and Bv/By exist, are continuous on A, and sale isfy the Cauchy-Réemann equations, then f is analytic on A. Uf f'(zo) does exist, then Ou, du Of _ 8 du_10f TAA) s pa Be @ Be = Oy | Oy a Uy Proof Let us first show that if f'(z9) exists, then u and v satisfy the Canchy- Riemann equations. In the lianit $a) = fig 12)= Sea), 0 ZO let us take the specinl case that z= r+ iyo. Then L(2)~ fo) __ (2,40) + éu(z, yo) — (20, yo) — ¢v(z0, vo) z-2 n— tq u(eiyo) — u(zoro) , , u(x 0) — v(t0 Yo) =— 2 na iq : Proofs of the following statements are not inchided here but can be found in any advanced calculuy text, such ax J. Marsden and M. Hoffman, Eicmentary Classical Analysis, Second Edition (New York: W. H. Freeman and Company, 1993), Ch. 6.§1.6 Basic Properties a As x — zo, the left, side of the equation converges to the limit f’(zo). Thus both the real and imaginary parts of the right side must converge to a limit (see Exercise 2 of §1.4). From the definition of partial derivatives, this limit is (8u/8x)(xo, Yo) + i(v/Az)(xo, yo). Thus f"(z9) = Bu/Az + 18v/Azx evaluated at (20, ya)- Next let z = a9 + iy. Then we similarly have £(2)- fa) _ azo) + te(z0, 9) — u(z0, ye) — fe(z0, v0) 2% ~ vo) = Yoru) —uleo,m) , v(za,¥) ~ v(z0, 40) iy vo) 156: As y — yo, we get By comparing real and imaginary parts of thesc equations, we derive the Cauchy- Riemann equations as well as the two formulas for f’(ze). Another argument. for this direction of the proof and one for the opposite im- plication may be based on the matrix representation for complex multiplication developed in Exercise 10 of §1.1. Lemma 1.5.9 A matriz ab ed represents, under matrix multiplication, multiplication by a complex number iff a = d andb=~c. The complex number in question is a + ic = d — ib. Proof First, let us consider multiplication by the complex number e-tic. It sends 2 + ty to (a + ic)(z + iy) = ax — cy + i(ay + cz), which is the same as @ -c =\_{o-a@ eoa@ y cxt+ay }* Conversely, let us suppose that (22)(g)-neem for a complex number z = a +48. Then we get az+by=ar—fy and cx+dy=ay+ fx68 Chapter 1 Analytic Functions for all x,y. This implies (setting x = 1,y = 0, then z = 0,y = 1) thata=a,b= —f,e= B, and d =a, and so the proof is complete. We can now complete the proof of Theorem 1.5.8. From the definition of f’, the statement that f'(ze) exists is equivalent to the following statement: For any ¢ > 0, there is a 6 > 0 such that 0 < [z — 2| < 5 implies [f() - S20) — £'(20)(z — 20)| < «lz — zal. First let us suppose that f‘(z9) exists. By definition, Df(xo, yo) is the unique matrix with the property that for any « > 0 there is a 6 > 0 such that, setting 2 = (z,y) and 29 = (zo, yo), 0 < |z — 20] < 6 implies M4(z) — (20) — DF (z0)(2 - 20)| < €lz - 2al- If we compare this equation with the preceding onc and recall that mulliplication by a complex number is a linear map, we conclude that f is differentiable in the sense of real variables and that the matrix D(z) represents multiplication by the complex number f’(zo). Thus. applying the lemma to the matrix ae du Ditsy=| 2 ov az ay with a = Gu/dz,b = du/dy,c = Gv/Az,d = dv/dy, we have a = d,b = —c, which are the Cauchy-Riemann equations. Conversely, if the Cauchy-Riemann equations hold, Df (zo) represents uniltipli- cation by a complex number (by the lemma) and then, as above, the definition of differentiability in the sense of real variubles reduces to that for the complex derivative. The formula for f’(z) follows from the Iast statement of the lemma. ‘We can also express the Cauchy-Riemann equations in terms of polar coordi- nates, but care must be exercised because the change of coordinates defiued by r= 2? +y? and @ = arg(z + iy) is a differentiable change only if @ is restricted to the open interval JO, 2x| or any other open interval of length 27 and if the origin (r = 0) is omitted. Without such a restriction @ is discontinuous, because it jumps by 2m on crossing the z axis. Using Oz/8r = cos @,Ay/Ar = sind, we see that the Cauchy-Riemann equations are equivalent to saying that du 190 Oy -1d0 or +80 ar or 80 ou a region contained in a region such as those shown in Figure 1.5.3. Here we are employing standard abuse of notation by writing u(r, 0) = u(rcos@,rsin 6). (For a more precise statement, see Exercise 12 at the end of this section.)215 Basic Properties 69 Figure 1.5.3: Two regions of validity of polar coordinates. Inverse Functions A basic result of real analysis is the Inverse Function The- orem: A continuously differentiable function is one-to-one and onto an open sct and has a differentiable inverse in some neighborhood of a point where the Jacobian cterminant of the derivative matriz is not 0. We will give proof here of the com- plex counterpart of this result, which assumes that the derivative f’ is continuous aud depends on the corresponding theorem for functions of real varinbles. After we have proved Cauchy's Theorem in Chapter 2, we will sec that the continuity of f’ is automatic, aud in Chapter 6 we will prove the theorem in another way that docs not depend on the real-variable theorem. The proof given here, however. ulustrates the relationship between real and complex variables and the relevance of the Cauchy-Ricwann equations. Theorem 1.5.10 (Inverse Function Theorem) Let f : A -» C be analytic with f° continuous) and assume that J'(z9) #0. Then there exists a neighborhood U of 29 and a neighborhood V of f(zo) such that f : U — V is a bijection (that is. is one-to-one and onto) and its inverse function f- is analylic with derivative green by a 1 ar eo) = Fy where w= f(z). The student is cautioned that application of the Inverse Function Thourem allows ‘ane only to conclude the existence of a local inverse for f. For example, Iet us consider f(z) = 2? defined on A = C\{0}. Then f'(z) = 22 #0 at each point of A. The Inverse Function Theorem says that f has a unique local analytic inverse, which is, in fact. merely some branch of the square root function. But f is not ‘one-to-one on all of A, since, for example, f(1) = f(—1). Thns f will be one-to-one only within sufficiently small neighborhoods surrounding each point. ‘To prove this theorem. let us recall the statement for real variables in two dimensions.70 Chapter 1 Analytic Functions Theorem 1.5.11 (Real-Variable Inverse Function Theorem) If f : Ac R? + R? is continuously differentiable and Df(zy,yo) has a nonzero determinant, then there are neighborhoods U of (9,40) and V of {(x0,%o) such that f:U + V isa bijection, f-! : V — U is differentiable, and DI-'(F(z,y)) = [D(z (this is the inverse of the matrix of partials). The proof of this theorem may be found in advanced calculus texts. See, for in- stance, J. Marsden and M. Hoffman, Elementary Classical Analysis, Second Edition (New York: W. H. Freeman and Company, 1993), Ch. 7. Accepting this statement and assuming that f’ in Theorem 1.5.10 is continuous, we can complete the proof. Proof of Theorem 1.5.10 For analytic functions such as f(z), we have seen that the matrix of partial derivatives is Be Bu) / du ~o8 [mw \_[ a o PE \ oy on |~| ae ou | Ox Oy or ox which has determinant du\? | (av\? (3) +(%) -rer since f’(2) = 9u/Az-+i18v/8z. All these functions are to be evaluated at the point (20,¥o) = ze. Now f'(2a) # 0, 80 Det Df(xo,yo) = If‘(20)|? # 0. Thus the real- variable Inverse Function Theorem applies. By the Cauchy-Riemann Thoorem 1.5.8 we need only verify that the entries of (Df(x,y)]~' satisfy the Cauchy-Riemann equations and give ({-1)’ as stated. As we have just seen, ou Df= Se Be 2 oy and the inverse of this matrix is oe oy e| az oz§1.5 Basic Properties 7 ‘Thus if we write [-"(x, y) = ¢(x,y) + is(z, y), then, comparing a oy ds ay DI')= fle le with the inverse matrix for Df, we get a 1 ae Oz” Det Dfay Det Df ar ast -dy_ ae de ~ Det Df Oz ~ Det Df dy’ which are evaluated at f(xo, yo). Similarly, eee Oe oe dy DeDypaz “ay ~ DaDyoy ‘Thns the Cauchy-Riemann equations hold for ¢ and s since they hold for u and v. Therefore, f~' is complex-diflerentiable. From the Canchy-Riemann Theorem we see that at the point f(z), “ty = 1 = - 7) __? Unt = Fee = De Dy ) real ~ Fey ™ ‘The real and imaginary parts of an analytic function must satisfy the Cauchy- Riemann equations. Manipulation of these equations leads directly to another very important property, which we now isolate. A twice continuously differentiable function u : A — R defined an an open set A is called harmonic if fu Pu Veu= aa? a= 0. The expression Vu is called the Laplacian of u and is one of the most. basic operations in mathematics and physics. Harmonic functions play a fundamental role in the physical examples discussed later in Chapters 5 and 8. For the moment. let us study them from the mathematical poiut of view. For Vu = 0 to make sense, the function u must be twice differentiable. In Chapter 3 an analytic function will be shown to be infinitely differentiable. Thus its real and imaginary parts are infinitely differentiable. Let us accept (or assume) these properties here. In particular, the second partial derivatives are continuous, and so a standard result of calculus says that the mixed partials are equal. The Cauchy-Riemann equations may then be uxed to show that the functions are harmonic. Proposition 1.5.12 /f f is analytic on an open set A and f = u + iv (thot is, if u = Ref and v= Im f), then u and v are harmonic on A.72 Chapter 1 Analytic Functions Proof We use the Cauchy-Riemann equations, 9u/8z = @u/dy and du/ay = —@v/8x. Differentiating the first equation with respect vo x and the second equa- tion with respect to y, we get Gu Fo yg Fu ao Os * Oady ay? ~~ bya" As above, the second partials are symmetric because they are continuous: ae oy azdy Aydx" Therefore, adding the equations in the preceding display gives us The equation for wv is proved in the same way. © If u and » aro real-valued functions defined on an open subset A of C such that the complex-valued function f = u + iv is analytic ou A, we sat that w and v are harmonic conjugates on A. For example, ule,y) =2?—y? and ulz,y) = 2ry arc harmonic conjugates of cach other since they are the real and imaginary parts of f(z) = 2”. These functions are linked geometrically as well as algebraically. The level curves of u and v passing through any nonzero point intersect at right angles at that. point, which is illustrated in Pigure 1.5.4. The next Proposition asserts that this occurs generally for the real and imaginary parts of an analytic function at Points where its derivative is not 0. The proof uses the Cauchy-Ricmann equations to show that the dot product of the tangent vectors (or the normul vectors) is 0. Proposition 1.5.13 Let u and v be harmonic conjugates on a region A. Suppose that the equations u(z,y) =constaut =c, and v(zx,y) = constant = c define smooth curves. Then these curves intersect orthogonally (see Figure 1.5.4). ‘We shall accept from calculus the fact that (x,y) = ¢, defines a smooth curve if the gradient. grad u(x, y) = (8u/8x, du/Ay) = (u/x) + i(u/Ay) is nonzere for and y satisfying u(x.) = c.. (The student should be aware of this fact even though it is a technical puint that does not play a major role in concrete examples.) It. is also true that the vector grad u is perpendicular ta that curve (sce Figure 1.5.5). ‘This perpendiculurity property can be explained as follows. If (x(t), y(t)) is the curve, then u(x(t),y(t)) = cr, a constant, so tute). w(0)] =0,31.6 Basic Properties B Figure 1.5.4: Harmonic conjugates: u = 2? — y?,v = 2zy, f(z) = 22. 20.9) gradu ute, y) «= constant Figure 1.5.5: Gradients are orthogonal to level sets. and thus by the chain rule, ou Fal) + 3 v= 0. That is, Ou du’ (3.3) -eoven=o. Proof of Proposition 1.5.13 By the above remarks, it suffices to show that grad u and grad v are perpendicular. Their inner product is ou dv du dv gradu: gad = 57-5 + ay a which is zero by the Cauchy-Riemann equations.“a Chaptor 1 Analytic Functions This orthogonality property of harmonic conjugates has an important physical interpretation, which will be used in Chapter 5. Another way to see why this prop- erty should hold is to consider conforma) maps and the Inverse Function Theorem. ‘This is illustrated in Figure 1.5.6. mee uz, y) =¢, po fas.) =, g Figure 1.5.6: Since f and f~' arc analytic, they are conformal and so preserve orthogonality. If f = u+ivis analytic and f’(zo) # 0, then f~) is annlytic on a neighborhood V of too = f(zo) and (f~")(2up) #£ O by the Inverse Function Theorem. Tfwo = ci-+ice, then the curves u(z,y) = ¢) and v(z,y) = o2 are the images of the vertical and horizontal lines through wa under the mapping f~!. They should cross at right angles since f~! is conformal by Theorem 1.5.7. Proposition 1.5.12 says thet the rea! part of an analytic function is harmonic. A natura) question is the opposite onc: Is every harmonic function the real part of an analytic function? More precisely: Given a harmonic function u op a set A, need there be a harmonic conjugate v such that f = u + iv is analytic ou A? ‘The full answer is a little tricky and depends on the nature of the set A. However, the answer is simpler if we confine ourselves to small neighborhoods. Indeed, the property of being harmonic is what is called a local property. The function w is harmonic on a set if V2u = 0 holds at cach point of that set. Therefore, it makes sense to study this property in a neighborhood of any point. Proposition 1.5.14 [fu is a twice continuously differentiable harmonic function on an open set A and z € A, then there is some neighborhood of 2 on which u is the real part of an analytic function. In other words, there exist an r > 0 and a function v defined on the open disk D(zo;r) such that u and v are harmonic conjugates on D(zo:r). In fact, D(zy;r) may be taken to be the largest disk centered at 29 and contained in A. A direct proof of this is outlined in Exercise 32. A different proof of a slightly stronger result will be given in Chupter 2. Since the Cauchy-Riemann equations must hold, v is uniquely determined up to the addition of a constant. These equations may be used as a method for finding v when w is given.§1.5 Basic Properties 8 Worked Examples Example 1.5.15 Where is the function f(2) = (25 + 2z + 1)/(z5 +1) analytic? Compute the derivative. Solution By Proposition 1.5.3(iii) f is analytic on the set A = {z € Clz+1 £0}; that is, f is analytic on the whole plane except the cube roots of ~1 = c™, namely, the points e*4/S,¢*#, and e®**/3_ By the formula for differeutiating a quotient, the derivative is fee (2 +132? +2) ~ (29 + 22 +1682") _ (2-424) (+P * (41 Example 1.5.16 Consider f(z) = 25 +1. Study the infinitesimal behavior of f at =i. Solution We use the Conformal Mapping Theorem 1.5.7. In this case f"(2o) = 3# = —3. Thus f rotates locally by * = arg(-3) and multiplies lengths by 3 = \f"(z0)|. More precisely, if ¢ is any curve through zo = i, tbe image curve will, at Ff (20), have its tangent vector rotated by 7 and stretched by a factor 3. Example 1.5.17 Show that f(z) = 2 is not analytic. Solution Let f(z) = u(x, y) + ie(z,y) = 2 - iy where 2 = (x,y) = 2+ iy. Thus, u(z,y) = 2, v(z,y) = -y. But Gu/Az = 1 and Ov/ay = -1 and hence 8u/dz £ 8v/Ay, so the Cauchy-Ricmaun equations do uot hold. Therefore, f(z) = 2 cannot be analytic, by the Cauchy-Riemaun Thoorem 1.5.8. Example 1.5.18 We know by Proposition 1.5.3 that f(z) = z* +1 is analytic. Verify the Cauchy-Ricmann equations for this function. Solution If f(z) = u(x,y) + iu(x,y) when z = (a,y) = 2 + iy, then in this case u(x,y) = 25 — 32y? +1 and u(z,y) = 327y — y*. Therefore, 8u/Ax = 327 ~ 3y?, 0u/y = —Gry,9v/z = Gry, and Gu/By = 3? — 3y?, from which we see that Ou/Oz = Sv/Oy and Gu/dy = -dv/Az. Example 1.5.19 Let A be an open subset of C and A® = {z | 2 € A}. Suppose f is analytic on A, and define a function g on A* by g(z) = f(2). Show that g is analytic on A*. Solution If f(z) = u(z,y) + év(z,y), then g(z) = F(Z) = u(z,—y) — iv(x,—y). We check the Cauchy-Riemann equations for g as follows: a a au 2 ime a = aaen) = gals. a)= ae = a = Fbve.-wl= Fame)a Chapter 1 Analytic Functions -2 Fives) = Zute,a)~ - Z| > - 7 Ms.—y) = -Z1-tz, -l=-2 jz (img). Since the Cauchy-Riemann equations hold and g is differentiable in the sense of real variables (why?), it is analytic on A” by the Cauchy-Ricmann Theorem 1.5.8. (One could also solve this exercise by direct appeal to the definition of complex differentiability.) Example 1.5.20 Find the harmonic conjugates of the following harmonic func- tions on C: (a) u(z,y) = 27 - 7? (b) u(z,y) = sin zeosh y ‘The reader might recuguize 27 — y? and sinxoashy as the real parts of z” and sin 2,2 = +iy. From this observation it follows that the conjugates, up to addition of constants, are 2zy and sinh ycosz. (We shui sce in the next section Uhat sin z is annlytic.) It is instructive, however, to solve the problem directly using the Cauchy- Riemann equations, because the student might not, always recognize an appropriate analytic f(z) by inspection. Solution To solve (a), suppose that # is a harmonic conjugate of v. By the Cauchy-Ricmann equations, ov au Ov _ du eH BR Therefore, v = 2yz + g1(y) and v = 2zy + go(z). Hence 91(y) = g(x) = constant, and so v(x, y) = 2yx-+ constant. To find the harmonic conjugate v for part (b), we use the Cauchy-Ricmanu equations again and write ov du ov . _ ou Be 7 By Tz and Bo = Be = emzonshy. ‘The first oquation implies that v = cosxsinh y + g:(y) and the second equation implics that v = cosasinh y + go(x). Hence g:{y) = ga(x) = constant. Therefore u(z.y) = cos zsinh y+ constant. Example 1.5.21 Suppose f is an analytic function on a region (an open connected set) A and that |f(z)] is constant on A. Show that f is constant on A.§1.5 Basic Properties 7 Solution We use the Cauchy-Riemann equatious to show that f"(z) = 0 every- where in A. Let f = u+iv. Then [f[? = u? + v? = c is constant. If e = 0, then [f(2)| = 0 and so f(z) = 0 for all z in A. Ic # 0, we take derivatives of u? +0? = ¢ with respect to z and y to obtain ou au ou an dug, + up =0 and dugy tee =o. By the Canchy-Riemaun equations these become wet yay and vt Ho. Or Oy az ay As a system of equations fur the two unknowns u/Oz and du/dy, the matrix of coefficients has determinant u? + v? = ¢, which is not 0. Thus the only solution Gu/Or = Gu/dy = 0 at all points of A. Therefore f'(z) = 9u/dx + i(Gv/dz) =0 everywhere in A. Since A is connected, f is constant (by Proposition 1.5.5). Exercises 1. Determine the sets on which the following functions are analytic, and compute their derivatives: (@) @+1P @)2+t 1 \% @ (<4) (a) a (2 = 1)(2? +2) 2. Determine the sets on which the following functions are analytic, and compute their derivatives: (a) 3274-7245 (b) (22 + 3)* 32-1 Os. 3. On what sets are the following functions analytic? Compute the derivative for each. (a) 2",n being an integer (positive or negative) (b) ca (z+ 1/z)? (c) 2/(2" — 2),n being a positive integer8 Chapter 1 Analytic Functions 4, Por 7 :]a.b[—+ C differentiable and f : A + C analytic with 4(Ja,[) ¢ A, prove that o = 07 is differentiable with o/(t) = f*((2)) +7 (t) by imitating the proof of the chain rule (1.5.4). 5. Study the infinitesimal behavior of the following functions at the indicated points: (2) (2) = 243,20 =344i (b) F(z) = 2° +32, 2 =0 2+zt1 () {@)= == - 4 =0 6. Study the infinitesimal behavior of the following functions at the indicated points: (2) f(z) = 22 + 5,2 = 5468 (b) f(z) = 24 442,29 <3 (©) f(2) =e - 1.2058 7. Prove that df-" /dw = 1/f'(2) where w = f(z) by differentiating f-?(f(z)) = 2, using the chain rule. Assume that f-' is defined aud is analytic. 8.° Use the Inverse Function Theorem to show that if f : A — C is analytic and J'(2) £0 for all z € A, then f maps opeu scts in A to open sets. 9. Verify the Cauchy-Riemann equations for the function f(z) = z? +3z + 2. 10. Prove that f(z) = |2| is not analytic. 11.° Show, by changing variables, that the Cauchy-Riemann equations in terms of polar coordinates become 12, Perform the computation in Exercise 11 by the following procedure. Let f be defined on the open set AC € (that is, f : A CC — C) and suppose that S(z) = u(z) + iv(2). Lot T :]0,2x[xR* + R®, whore Rt = {z € R|z > 0}, be given by T(0,r) = (rcox0,r'sin 9). Thus T is one-to-one and onto the set R*\{(x,0) | z > 0}. Define (0,7) = v(rcos8,rsin8) and 5(0,r) = v(rcosd,rsin8). Show that {a} T is continuously differentiable and bas a continuously differentiable in- verse.{1.5 Basic Properties 2 (b) fis analytic on A\{z-+iy | y = 0,2 > 0} if and only if (@,%) : T—(A) + R? is differentiable and we have 180 4, Bb _ 1% or a8 or 36 on T-"(A). 13. Define the symbol Of /02 by ag _isag 1 22) az 2 (Z tay)" Show that the Cuuchy-Riemun equations are equivalent to 8f/92 = 0. Note: It is sometimes said, because of this result, that analytic functions are not, functions of Z but of z alone. This statement can be made more precise as follows. Given f(x,y), write z = }(z +2) and y = (1/2i)(z — 2). Then f becomes a function of z and and the chain rule gives af 1 (3h a rg = 14, Define the symbol 2f/82 by Of _ 1 (8 4 LaF a iy (a) Show that if f is analytic, then /' = 3f/dz. (b) If f(z) = 2, show that Of/0z = 1 and 8f/8z = 0. (c) If f(z) = 2, show that 8f/dz =0 and 8f/82 = 1. (4) Show that the symbols 0/42 and 8/03 obcy the sum, product, and scalar multiple rules for derivatives. (e) Show that the expression 37", Mo ¢am2"2” is an analytic function of 2 if and only if dam =0 whenever m # 0. 15, la roa hee pens oie tetova and that Re f(z) =3 for all z in D. Show that f is constant on D. 16.* (a) Let f(z) = u(z,y) + iv(z,y) be an analytic function defined on a con- nected open set A. If au(x,y) + bv(z,y) = c in A, where a,b,c are real constants not all 0, prove that f(z) is constant in A. (b) Is the result obtained in (a) still valid if 2, b,¢ are complex constants? 17. Suppose f is anelyticon the set A = {z | Rez > 1} and that 8u/dz-+dv/ay = Oon A. Show that there are a real constant ¢ and a complex constant d such that f(z) = —icz +d on A.80 Chapter | Analytic Functions 18. Let f(z) = z°/|z|* if 2 # 0 and f(0) = 0. (a) Show that f(z)/z does not have a limit ag z ~~» 0. (b) If u = Ref and v = Im f, show that u(z,0) = z,v(0,y) = y,u(0,y) = u(x,0) =0. (c) Conclude that the purtials of 2, exist and that the Cauchy-Riemann equations hold but that (0) does uot exist. Does this conclusion con- tradict the Cauchy-Riemann Theorem? (a) Repeat exercise (c), letting f = 1 on the x and y axes and 0 elsewhere. (e) Repeat exercise (c), letting f(z) = /lzy]- 19. Let f(z) = (2 +1)/(z = 1). (a) Where is f analytic? (b) Is f conformal at 2 = 0? (c) What are the images of the a and y axes under f? (2) At what angle do these images intersect? 20. Let f be an analytic functiou ou um open connected set A aud suppose that s*)(z) (the n + Ist derivative) exists and is zero on A. Show that f is a polynomial of degree < n. 21. On what set is u(z, ¥) = Re(2/(z* — 1)) harmonic? 22.° Verify directly tbat the real and imaginary parts of f(z) = 24 are harmonic. 23. On what sets are each of the following functions harmonic? (a) u(x, y) = Ln(2? + 3z +1) r-l1 (0) Mew) = pyre 24. Ou what sets are each of the following functions harmouic? (u) u(x,y) = Im(z + 1/2) y (b) u(z,y) = Goa 25. Let f : A+ C be analytic and let w : B — B he harmonic with f(A) ¢ B. Show that wo f : A — R ix harmonic. 26. If u is harmonic, show that, in terms of polar coordinates, Hint: Use the Cauchy-Riemann equations in polar form (Exercise 11).§1.6 Differentiation of the Elementary Functions 81 27. (a) Show that u(x, y) = ¢* cosy is harmonic on C. (b) Find a harmonic conjugate »(z,y) for u on C such that w(0,0) = 0. (c) Show thut f(z) = e* is analytic on C. 28. Show that u(z,y) = 25—3sy? is harmonic on C and find a harmonic conjugate v such that v(0,0) = 2. 29. If u(z) is harmonic and v(z) is harmonic, then are the following harmonic? (2) ufv(z),0) (b) u(z)- v(2) (©) u(z) + o(z) 30. Consider the function f(z) = 1/2. Draw the contours u = Re f = constant and v = km f = constant. How do they intersect? Is it always true that grad tis parallel to the curve v = constant? 31. Let u have contimous second purtials on an open set A and let ?u/dz? + Fu/dy? =0. Let f = Gu/Az — idu/By. Show that f is analytic. 32. Suppose u is a twice continuously differentiable real-valued harmonic function ‘on a disk D(zo:r) centered at zo = to + iyo. For (x1,41) € D(z0;7), show that the equation stern) ae [” Fetensddy— f (ewe ve defines a harmonic conjugate for u on D(z9;r) with v(z0, 40) =< 1.6 Differentiation of the Elementary Functions Exponential Function and Logarithm This section discusses differentiability properties of the clementary functions introduced in §1.3 and we will begin with the exponential function and its inverse, the logarithm. Proposition 1.6.1 The map f :C + C,z++ ¢®, is analytic on C and de* aut: Proof By definition, f(z) = e*(cosy + isiny), 60 the real and imaginary parts are u(z,y) = e“ cosy and v(z,y) = e*siny. These are O° (infinitely often dif- ferentialsle) functions, so f is differentiable in the sonse of real variables. To show82 Chapter 1 Analytic Functions that f is analytic, we must verify the Cauchy-Riemann equations. To do so, we first compute the partial derivatives oy ay On: oy ov . ov z& =c*siny ay = c* cosy. ‘Thus, 8u/8z = dv/Ay and u/@y = —du/Az, so by the Cauchy-Riemanu Theorem 1.5.8, f is analytic. Finally, af _ du, av & 32 * fae A function that is defined and analytic on the whole pline C is called entire. Thus, f(z) = e? is an entire function. Using the differentiation rules (Proposition 1.5.3) as in clemeutary calculus, we can differentiate e* in combination with various other functions. For instance, = + L is entire because z ++ 2? and w ++ e¥ are analytic, so by, the chain rule, +e” is analytic. By the chain rule and the sum rule, (d/dz)(e"" +1) =2ze". We recall that logz : C\{0} — C is an inverse for e* when e* is restricted to a period strip {x + iy | yo S y < yo + 2a}. However, for differeutiability of logz we must restrict logz to a set that is smaller than C\{0}. The reason is simple: logz = log|z| + iargz for, say, 0 < argz < 2x. But the arg function is discontinuous; it jumps by 2x as we cross the positive real axis. If we remove these points, then we are excluding the usual positive reals on which we want log z defined. Therefore, it is convenient to use the brauch —x < argz <7. Then an appropriate set on which log z is analytic is given as follows. Proposition 1.6.2 Let A be the open set that is C minus the negative real aris including zero (that is, C\{x + iy | x <0 and y = 0}). Define a branch of log on Aby =e (cosy +isiny)=c7. logz =log|z|+iargz —a 0,y = 0}. We will use the principal branch unless otherwise stated. ox Figure 1.6.1: Domain of log z. When using logz in compositions, we must be careful to stay in the domain of log. For example, consider f(z) = logz* using the principal branch of log. This function is analytic on the set A = {2 | z #0 and argz # +:n/2} by the following reasoning. Proposition 1.5.3 shows that z” is analytic on all of C. The image of A under the map z ++ 2? is precisely C\{z + iy | x < 0,y =0} (Why?), which is theae ! Figure 1.6.2: Composing the squaring and the log functions. set on which the principal branch of log is defined and analytic. By the chain rule, 2+-+ Jog 2? is then analytic on A (see Figure 1.6.2). The function f(z) = log 2? also illustrates that caution must be exercised in manipulating logarithms. Consider the two functions Jog 2” and 2logz. If we consider all possible values of the logaritlun, then the twe collections arc the same. If, however, we pick a particular branch, for example, the principal branch, and use it for both, then the two are not always the same. For example, if z= —1-+#, then argz = 32/4 and logz? = log 2 — mi/2, while 2logz = log2 + 3xi/2. The function log z? is analytic on the plane with the imaginary axis deleted. while 2log z is analytic on the plane with the negative real axis deleted. The Trigonometric Functions Now that we have established properties of ¢*, the differentiation of the sine and cosine fuuctions follows readily. Proposition 1.6.3 The sine and cosine functions sinz end cos are entire func- tions with derivatives given by a. a Gimzacose and 7 cosz = —sinz. Proof By definition, sin z = (e* —e~**)/2i; using the sum rule and the chain rule and the fact that the exponential function is entire, we conclude that sin z is entire and that d(sin 2)/dz = {ie* — (-ie~®)} /2i = (c= + e7#*)/2 = cosz. Similarly, a i Hose = Egle +e) = He -e*) = ~ ple eH) = ~sinz. . We can also discuss siu~' z and cos~'z in somewhat the same way that we discussed log 2 (which is exp-! z with appropriate domains and ranges). These functions are analyzed in Exercise 6 at the end of this section.$1.6 Differentiation of the Elementary Functions 8 The Power Function Let a and b be complex numbers. Recall that ab = eb!s¢, whieb is, in general, multivalued according to the different branches for log that ‘=r choose. We now consider the functions z +» z® and z +» a?. Although these functions appear to be similar, their properties of analyticity are quite different. The case in which 6 is an integer was covered in Proposition 1.5.3. The sitnation ior gencral b is slightly more complicated. Proposition 1.6.4 (i) For any choice of branch for the log function, the func- tion z+ a is entire and has derivative z ++ (log a)a?. ii) Choose a branch of the log function—for example, the principal branch. Then the function z + z* ix analytic on the domain of the branch of log chosen and the derivative is z+ bz'-", Proof (i) a® = e%8¢, By the chain rule this function is analytic on C with derivative (log a}et "52 = (loga)u*(loga is mercly # constant). tii) 2 = !%*, This function is analytic on the domain of logz, since w+ oo is entire. By the chain rule, d b b at = webloss = 22h aes 2" (That this equals 6:'~! follows from Exercise 20 at the end of §1.3.) @ IT} is a non-negative integer we know that z* is entire (with derivative b2'~). In general, however, z* is anslytic only on the domain of log z. Let us emphasize what is stated in (ii). If we choose the principal branch of log 2, which has domain C\{z + iy | y = 0 and x < 0} and range Rx] - =,7{ ‘Why?), then 2 ++ 2? is analytic on C\{x + iy | y = 0,2: < 0} (see Figure 1.6.3). We could also choose the branch of log that has domain C\{x + ix | x > 0} and range Cx | - 77/4, 7/4{; then z+ 2° would be analytic on C\{z + iz | x > O}. The nth Root Function One of the nth roots of z is given by 2'/*, for a choice of branch of log z. The other roots are given by the other choices of branches, as in 71.3. The principal branch is the one that is usually used. Thus, from Proposition 1.6.4(ji), we get the following as a special case. Proposition 1.6.5 The function z ++ 2!/" = 4/z is analytic on the domain of log = (for example, the principal branch) and has derivative (3) =r" As with log2z, we must exercise care with the functions z ++ 24,24 9/2 when composing with other fimctions to be sure we slay in the domain of analyticity. The procedure is illustrated for the square root function in Worked Example 1.6.8.86 Chapter 1 Analytic Functions Figure 1.6.3: Regions of analyticity for z+ 2%. Worked Examples Example 1.6.6 Differentiate the following functions, giving the appropriate region on which the functions are analytic: (a) &* () sin(e*) (©) e*/(@? +3) (@) ver FT (e) cosz () 1 -1) (2) log(e* +) Soluti (a) The function e* is entire, so by the chain rule, z + e*” is entire. The chain rule also tells us that the derivative at z is ee” (b) Both z + ef and w + sinw are entire, so by the chain rule z + sine* is entire and the derivative at z is (cose*)e*. (c) The z + @ is entire and the map z + 1/(=? + 3) is analytic on Cleve, Hence z ++ ¢*/(z? +3) is analytic on C\{-V3i} and has deriva- cease eu Bz et? - 2243) 2rs ry” eryL6 Differentiation of the Elementary Functions 87 (d) Choose the branch of the function w ++ \/w that is analytic on C\{z + iy | vy = 0,z <0}. Then we must choose the region A such that if z € A, then «* +1 is not both real and < 0. Notice that ¢* is real iff y = Im2 = nz for some integer n (Why?). When n is even, ¢* is positive; when n is odd, e* is negative. Here a‘| = ¢%, where x = Rez and e* > 1 iff x > 0. Therefore if we define A = C\{z + iy | x > 0,y = (2n + 1)a,n an integer } (as in Figure 1.6.4), then e* +1 is real and < 0 iff z ¢ A. Siuce e* +1 is entire, it is certainly analytic on A. By the chain rule, Ve™ +1 is analytic on A with derivative at 2 given by (e* + 1)-/(e*)/2. y + Figure 1.6.4: Region of analyticity of /e* +1. te) Since z = z+ iy, by Proposition 1.3.4, cos = cos(z — iy) = cosxcas(-iy) — sinxsin(—iy) = coszcoshy + isinzsinhy, so u(z,y) = coszcoshy and »(z,¥) = sinzsinhy. Thus du/dz = — sin zoosh y,8v/dy = sinzeoshy. If cos 2 were analytic, Su/9z would equal v/By, which would occur iff sinz = 0 (that is, if 2 = 0, or if + = an,n = £1,£2,...). Thus, there is not an open (nonempty) set A on which z + cosz is analytic. ff) By Proposition 1.5.3(iii) and the fact that z+ e* — 1 is entire, we conclude tut z+ 1/(e* — 1) is analytic on the set on which c* — 1 # 0; namely, the set A=C\{z = 2nni | n =0,41,42,...}. The derivative at z is —e*/(e* — 1)?. (g) Since (the principal branch of) the log is defined and is analytic on the same region as the square root, namely, A = C\{z + iy | y = 0,2 < O}, we can use the results of (d). By the chain rule and the results of (d), z+ log(e* + 1) is analytic on the region depicted in Figure 1.6.4. Example 1.6.7 Verify directly that after mapping by the function c*, the angles wtween lines parallel to the coordinate axes are prescrued.88 Chapter 1 Analytic Functions Solution The live determined by y = yo is mapped to the ray {c+-éztan yo | x > O}, and the line determined by x = 20 is mapped to the circle |z| = c74; see Figure 1.3.7. The angle between any such ray and the tangent vector to the circle al the point of contact of the ray and the circle is 7/2. Thus angles are preserved. This is consistent with the Conformal Mapping Theorem 1.5.7. Example 1.6.8 Show that a branch of the function w+ Vie can be defined in such a way that z + /27 —1 is analytic in the region shaded in Figure 1.6.5, and using the nolations of that figure, show that Jz2—1 = Vrirge"+6)/?, where 0<0, <2n, 7 <0 0}; thus z+ ¥z—Tis analytic on the rogion C\{x + iy | y = 0,2 > 1}. For the map 2 Yz +1, we may choose defined and anulytic on the region C\{z-+4y | y = O and x < 0}; therefore, 2+ Vz is analytic on C\{z + iy | y = 0,2 < -1}. Thus z+ yz—1veFT is analytic on C\{x + iy | y = 0, [2] 2 1} with the appropriate branches of / as indicated. With these branches we have ve-1= rel? and Vz+1 = Vroeil?, so Vz—1VzFl = Argel +2, Since z + J/z—1V¥z41 is analytic on C\lz tiv | y = 0,f] > > 1}, it should correspond to some branch of the square31.6 Differentiation of the Elementary Functions root function in the map z++ V2? —T. To see which one, uote that z € C\{z + ty | y = 0,2] > 1} iff 22-1 € C\{z + iy | y = 0,2 > 0} (Why?), so if we take the squure root defined and analytic on C\ {z+ iy | y = 0,2 > 0}, then z+ Vz? — 1 is analytic on C\{x + ty | y = 0, |x] > 1}. The symbol «/- bas been used here to mean different branches of the square root function, the particular branch being clear from the context. Thus, ane normally thinks of y* together with a choice of branch. Exercises 1. Differentiate and give the appropriate region of analyticity for each of the following: (a) 242 (b) fz (c) sin z/cosz (@) ew (241) 2. Differentinte and give the appropriate region of analyticity for each of the following: @" a (b) log(z +1) (200 @) vz )* # 3. Determine whether the following complex limits exist aud find their value if they do: (@) tim,» = (b) tin,» = 4. Determine whetlier the following complex limits exist and find their value if they do: (0) tam, a 8% (b) tim, 24 1 5. Is it true that |sin2| < 1 for all z€ C?90 Chapter 1 Analytic Functions 6.* Solve sin z = w and show how to choose a domain and thus how to pick a particular branch of sin~'z so that it is analytic on the domain. Give the derivative of this branch of sin~' z; see Exercise 35 at the end of §1.3. Let f(z) = 1/(1 = 2); is it coutinuous on the interior of the unit circle? Let u(z,y) and v(2,y) be real-valued functions on an open set A C R? = C and snppose that they satisfy the Cauchy-Riemann equations on A. Show that ex us(z,y) = [u(z,y)]? - [v(e,y))? and vi(z,y) = 2u(z, y)o(z.y) satisfy the Cauchy-Riemann equations on A aud that the functions ‘W) cosu(z,y) and v2(z,y) = eM(*¥) sin u(x, y) ue(2,y) = also satisfy the Cauchy-Riemann equations on A. Can you do this without performing any computations? 9. Find the region of analyticity und the derivative of each of the following functions: z @ a4 (b) eta) 10." Find the region of analyticity and the derivutive of each of the following functions: (a) VB=1 (b) sin Yz 11. Find the minimum of |e*"| for those z with Jz] < 1. 12. Prove Proposition 1.6.5 using the method of the first proof of Proposition 1.6.2. 13.° Where is z+ 2*" analytic? z+ 22*? 14, Define a branch of \/1 + «/z and show that it is analytic.LR Review Exercises for Chapter 1 91 Review Exercises for Chapter 1 2 fe eo 10. ue. . Compute the following quantities: (a) e! (b) log(L+#) (c) siné_ (4) 3¢ (e) etet-) For what values of z is log 2” = 2logz if the principal branch of the logarithm is used on both sides of the equation? . Find the eighth roots of i. Find all numbers z such thnt 2? = 144. Solve cos z = V3 for z. . Solve sin z = V3 for z. Describe geometrically the set of points z € C satisfying (a) [z+] =|z-3] () [2-1] = 3[2-2] Describe geometrically the set of points z € C satisfying (a) [2-1 =|z+1] () [2-1] = 2I2] . Differentiate the following expressions on appropriate regions: (a) 248 Os (¢) exp(z* - 1) (4) sin(log 2*) On what set is 2? — 2 analytic? Compute the derivative. Describe the sets on which the following functions are analytic and compute their derivatives: (2) eM 1 () a —sinz)? (c) ao a for a real 12.* Repeat Review Exercise 11 for the following functions: (a) er() foraec92 Chapter 1 Analytic Functions sinz © =. 13. Can a single-valued (analytic) branch of log z be defined on the following sets? (a) {z|1<|z| <2} (b) {z| Re z > 0} (c) {| Re z > Im z} 14.* Show that the map z+ z+ 1/z maps the circle {z such that |2| = c} onto the ellipse described by {w =u+tiv | u= (c+ 1/c)00s6,v = (c— 1/c)sin8,0 < 0 < 2x}. Can we allow ¢ = 1? 15. Lot f be analytic on A. Define g : A — C by g(z) = J(2). When is g analytic? 16. Find the real and imaginary parts of f(z) = 2° and verify directly that they satisfy the Cunchy-Riemann equations. 17. Let f(x + iy) = (z? + 2y) + i(a? + y?). At what points docs f"(zp) oxist? 18. Let f: AC C +C be analytic on an open set A. Let AY = {2|2 € A}. (a) Describe A* geometrically. (b) Define g: A” — C by g(z) = [f(2)F- Show that g is analytic. 19.° Suppose that f : A C C — C is analytic on the open connected set A and that f(z) is real for all z € A. Show that f is constant. 20. Prove the Canchy-Riemann equations as follows. Let f : AC C — C be differentiable at 29 = 29 + iyo. Let gi(t) = t+ iyo and go(t) = zo +i. Apply the chain mule to fog, and f 092 to prove the result. 21. Let f(z) be analytic in the disk {z—1| <1. Suppose that f"(z) = 1/z, f(1) = 0. Prove that f(z) = log 2. 22.° Use the Inverse Function Theorem to prove the following result. Let f : ACCC be anslytic (where A is open and connected) and suppose that $(A) ¢ {z such that |z| = 3}. Then f is constant. 23. Prove that im Oth) = 28 _ nt eh ad for any zo € C. 24. (a) Ifa polynomial p(z) = ap +2 +-..+@n2" bas a root c, Lhen show that we can write p(z) = (z - c)h(z), where h(z) is a polynomial of degree n— |. (Use tlivision of polynomials to show that 2 —¢ divides p(z).){LR Review Exercises for Chapter 1 93 (b) Use part (a) to show that p can have no more than # roots. (c) When is ¢ a root of both p(z) and p’(z)? 25.° On what set is the function z +» z* analytic? Compute its derivative. 26. Let g be analytic on the open set A. Let B= {2 € A | g(z) #0}. Show that B is open and that 1/9 is analytic on B. 27, Find and plot all solutions of z* = —8i. 28. Let f: AC © +€ be anulytic on the open set A and let f"(29) # 0 for every 29 € A, Show that {Re f(z) | z € A} C R is open. 29,° Show that u(z,y) = 2° — 3zy?,u(z,y) = 3x%y — y* satisfy the Cauchy- Riemann equations. Comment on the result. 30. Prove that the following functions arc continuous at z = 0: af (Rez*)7/\zP 240 to) stay { Ree yieeh axe (b) (@) = Fel 31. At what points 2 are the following functions differentiable? (2) J@) =P (b) f(z) =y- ix 32.° Use de Moivre's theorem to find the sum sin z + sin 2z +... -+sinnz. 33. For the function u(x, y) = y* — 327y, {a) Show that w is harmonic (see Proposition 1.5.12). (b) Determine a conjugate function v(x, y) such that u + év is analytic. 34. Consider the function w(z) = 1/2. Draw the level curves u = Re(w(z)) = constant. Discuss. 35. Determine the four different values of 2 that are mapped to unity by the function w(z) = 2, 36. Suppose that f(z) is analytic and satisfies the condition |f(z)? - 1] <1 ina region 9, Show that either Re f(z) > 0 or Re f(z) < 0 throughout 2. 37." Suppose that f : C + C is continuous and that f(z) = f(22) for all z € C. Show that f is constant on C. 38. Suppose that f: C+ C is entire and that f(2z) = 2f(z) for all 2 € C. Show thnt there is a constant ¢ such that f(z) = ez for all 2. (You might want to use Exercise 37.)94 Chapter 1 Analytic FunctionsChapter 2 Cauchy’s Theorem An attractive feature of complex analysis is that it is based on a few simple, yet pwerful theorems from which most of the results of the subject follow. Foremost emong these theorems is a remarkable result called Cauchy’s Theorem, which is one of the keys to the development of the rest of the subject and its applications. 2.1 Contour Integrals Definitions and Basic Properties To state Cauchy's Theorem, we first define ~satour integrals and study their basic propertics. Let ft: [2,8] CR — C he a complex-valued function of one real variable and let . and he its real and imaginary parts; that is, set h(t) = u(t) + iv(t). Suppose, tor the sake of simplicity, that u and v are continuous. Definc the integral of h to Se the complex number f A(e)at = f (thie +i He v(t)de, where the integrals of u and v have their usval meaning from single-variable calcnlus. ‘We want to extend this definition to integrals of functions along curves in C. To accomplish this, we will need a few definitions. A continuous curve or contour in Cis, by definition, a continuous map 7 : [a,b] + C. The curve is called piecewise C! if we can divide up the interval (a, }) into finitely many subintervals 2 = @g 1) D : rt ar a b a b a b He) 1% ) ® @ Figure 2.1.1: Curves in the complex plane C. (a) Smooth curve. (h) Piecewise C? curve. (¢) Discontinuons curve. expression [12 [ seme > [CO toorroe is called the integral of J along. ‘The dofinition is analogous to the following definition of a line integral from vector calenlus: Let P(z,y) and Q(z, y) be real-valued functions of x and y and let. 7 be a curve. Define i i; P(z,y)dz + Az, = P(x(t), xO) + QAz(2), v(t) [reads cena =F [" [retour G + acetone] a where 7(t) = («(t),x(t)). The two definitions are related as follows. Proposition 2.1.2 If f(z) =u(z,y) + te(x,y), then f f= [ [u(z, y)dx — u(x, y)dy] + [ [u(z,y)dy + vfz, vex]. om ee T Proof According to the definition we must. work out f(7(t))7/(t). We do this as follows: LOM) [ufa(t), v(t)) + ivfx(t), 9())] - 2’) + y(O) [ula(e), v2)" — v(2(t), we))y'(O) + fole(), y())2"(t) + u(x). w()y'()- Integrating both sides over {a;,0;;1] with respect to ¢ and using definition 2.1.1 then gives the desired result.72.1 Contour Integrals 97 The formula in this proposition can easily be remembered by formally writing fle)dz = (w+ tv)(de + idy) = udz — vdy + i(vde + udy). For a curve 7: [c,4] + C, we define the opposite curve, ~ : [a,b] ~ C, by setting (—7)(t) = 7(@+6—1). The curve —7 is thus +y traversed in the opposite zense (see Figure 2.1.2). yy Figure 2.1.2: Opposite curve. We also want to define the join or sum or union y +72 of two curves 7; and 2. Intuitively, we want to join them at their endpoints to make a single curve (sec Figure 2.1.3). Precisely, suppose that 7 : [a,b] + € and that 7» : [b,c] + C, with 1b) = y2(d). Define 41 +2 : [a,¢e] + C by ult) if t€ (a,b) nm +mo-{ nit) if te lad. y nets Figure 2.1.3: Join of two curves, Clearly, if 7, and 7 are piccewise smooth, then so is 71 +72. If the intervals ja,0) and [6,¢] for “n, and -7z are uot of this special form (the frst interval ends where98 Chapter 2 Cauchy’s Theorem the second begins), then the formula is a little more complicated, but the special form will suffice for this text. The general suin 7, +... + Yn is defined similarly. The next proposition gives some properties of the integral that follow from the dcfinitions given in this section. The student is asked to prove them in Exercise 6 at the end of this section. Proposition 2.1.3 For (continuous) functions f,g, complex constants ¢,,¢2, and piecewise C! curves 7,11172, the following hold: @) flarraneaa [tra fo o [ t=-f7 © ee Of course, more general statements (that follow from the preceding) conld be fue B laf and Dam 2 bt ‘To compute specific examples it is sometimes convenient to use the formula in Proposition 2.1.2. However, it may be that we are not given 7 95 a map but are told only that it is, for example, “the straight-line joining 0 to i+ 1” or “the unit circle traversed counterclockwise”. To use the definition, we need to choose some explicit map 7(t) that describes this geomotrically given curve. Obviously, the same geometric curve can be described in different ways, so the question arises whether the integral Sy f is independent of that description. ‘To answer this question, we use the following definition. Definition 2.1.4 Let_y : [a,b] -+ C be a piecewise smooth curve. A piccewise smooth curve 7 : [4.6] — C is called a reparametrization of 7 i there is a C* function a : [a.6] — [a,b] with a’(t) > 0,a(@) = a, and a(b) = b such that at) = Halt) (sce Figure 2.1.4). The conditions a'(t) > 0 (hence a is increasing), a(a) = @, and a(b) = 4 imply that 7 traverses the curve in the same sense as + does. This is the precise meaning of the statenient that y and ¥ represent the same (oriented) geometric curve. Also. the points in |@, 6] at which 7/ does not exist correspond under a to the points of [a,0) at which -/ does not exist. (This is because a has a strictly increasing C' inverse.)§2.1 Contour Integrals 99 Figure 2.1.4: Repsrametrization. Proposition 2.1.5 If 7 is @ reparametrization of 7, then i for any continuous { defined on an open set containing the image of y = image of ¥. Proof We can, by breaking up {a,0] into subintervals, assume that 7 is C!. By definition [}- [torre By the nin rule, (0) = dt) = Aa) /dt = (a(t). Lat «= at) be a new variable, so that s = 4 when t =a and s = when t [rortoa = frome Za = iE SHS) ()ds. Changing variables in « complex integral (here, from ¢ to 4) is justified by applying the usual real-varisbles rule to its real and imaginary parts. This proposition “justifies” the use of any parametrization y that describes a given oriented geometric curve to evaluate an integral.’ As an example, let us evaluate [| zdz, where 7 is the straight-line from z = 0 to z = 1 +i (see Figure 2.1.5). ‘Strictly speaking, this statement is not quite correct, since two maps with the same image need not be reparametrizations of one another. However, they are reparametrizatious if we ignore points where 7/(¢) = 0. The proposition can be generalized to cover this situution as well, but the complications that result froin generalizing it to caver this case have been omitted to simplify the exposition,100 Chapter 2 Cauchy’s Theorem Figure 2.1.5: Curve from 0 to 1+ i. We choose the curve 77 : [0,1] + C, defined by 7(t) = t+ it. OF course, when we write x in f, xdz, we mean the function that gives the real part of any complex mumber (that is, f(z) = 2 = Rez). Thus, [a2 [ortorytoe. Hence, [eu= [a tijd tt An orientation is often described by saying that “y goes from 2; to 22.” However. if 7 is a closed curve, wilh 2; = 22, we need a different prescription. When solving examples, where the curves are always easy to visualize, the student should assume that a closed curve 7 is traversed in the counterclockwise direction unless advised to the contrary. From calculus, recall that the arc length of a curve 7 : [a,b] — C is defined by Wa) = f I let = i va? + yi (ePat. Are length, too, is independent of the purametrization, by a similar proof to that of Proposition 2.1.5. The reader sliould be familiar with the fact that the arc length of the unit circle is 2x, the perimeter of the unit square is 4, $0 on. The next result: gives an important way o estimate integrals. Proposition 2.1.6 Let f be continuous on an open sct A and let + be a piecewise C! curve in A. If there is a constant M > 0 such that |{(z)| < Af for all points =321 Contour Integrals 101 on (that is, for all z of the form (1) for some t), then | [4] sane. More generally, we have fas f ile wher the latter integral is defined by [mas = f ely lee. Proof For a complex-valued function g(t) on [a, 0], we have Re if olt)dt = f Reg(t)dt since fe o(t)dt = fo u(t)de +4 f° v(t)de if g(t) = u(t) + iv(t). Let us use this fact to prove that, ff ane s [woo ‘We learn in calculus how to prove this for g that are real-valued, but here g is complez-valued.) For onr proof, we let f° g(t}dt = re” for fixed r and 8, where r 20, 60 that r= ef? oft)de = f° oe" g(t)dt. Thus, r=Rer=Re f 7 g(t)dt = i Ro(e~!g(t))dt. By Proposition 1.2.3(4ii), Re(e~g(2)) < Je-g(2)| = Ig(t)|, since [e-| = 1. Thus, JE Re(e-g(t) ae < SF lolt)lat, so ars foto. | f g(t)de = [venir Using this and |z2"| = [zll2’[, we get If Since [f(7(t))| < Af, the preceding expression is bonnded by Af [*/(t)Idt = My). ©102 Chapter 2 Cauchy’s Theorem This proposition provides a basic tool that we shull use in subsequent proofs to estimate the size of integrals. The student might try to prove this result directly in terms of the expression [1= [ceva +6 { (way + ude) to be convinced that the result is not altogether trivial. Fundamental Theorem of Calculus for Contour Integrals The Fimdamen- tal Thcorem of Calculus is a basic fact in the calculus of real-valued functions. It says that, the integral of the derivative of # function is the difference of the values of the function at the endpoints of the interval of integration and that the indcfinite integral of a function is an antiderivative for the function. Both of these assertions have important analogues for complex path integrals. Theorem 2.1.7 (Fundamental Theorem of Calculus—Contour Integrals) Suppose that +y : [0,1] + € is a piecewise smooth curve and that F is a function defined and analytic on an open set G containing y. Assume F" is continuous (later we will sce that this is redundant). Then [ Fe(a)de = F(t) - Fl). In particular, if (0) = (1), then f P'(2)dz =0. Proof The chain rule and the definition of the path integral will be used to reduce the problem to the standard Fundamental Theorem of Calculus for real-valued functions of a real variable. Let g,u, and v be defined by F(t) = a(#) = ult) +iv(t). Break the parameter interval into pieces on which + is smooth, so, by the chain rule, F(r(t))7 (t) = 9'(t) = u'(t) + iv").+2.1 Contour Integrals 103 We apply the Fundamental Theorem of Calculus on each subinterval and get a celescoping sum: nol oe t))7'(@) = (Yr (ae . EL wom £ veo] un = SY hulesss) - ules) + fo(ersi) - vfes)] a= = [u(an) + #v(an)] — [e(a9) + iv(ao)) = F(y(1))- P(r). Using this result can save a lot of effort in working examples. For instance, consider f, 2°dz where 7 is the portion of the ellipse 2? + 4y” = 1 that joins z= 1 to z= 4/2. To evaluate the integral we note that 2° = }(dz*/dz). so [2m 8f°-()()-Qor-8 Notice that we did not even need to parametrize the curve! By applying the Fun- damental Theorem of Calculus, we would have obtained the snme answer for any curve joining these two points. We will investigate the independence of the value of an integral from the particular path used in the next subsection. The Fundamental Theorem of Calculus has many applications and ramifications, one of which is the following proof of a property of open connected sets, which first appeared as Proposition 1.5.5. It is the analogue in the complex domain of the following principle so useful in calculus: A function whose derivative is identically is constant. Corollary 2.1.8 If f is a function defined and analytic on an open connected set GCC, and if f(z) =0 for every point z in G, then f is constant on G. i Fi(2)d2 Proof Fix a point 29 in G and suppose that z is any other point in G. By Proposition 1.4.15 there is a smooth path + from 2 to z in G. By Theorem 2.1.7, S(2)— F(a) = f, (dG = 0. Therefore f(z) = f(z0). The value of f at any point of G is thus the Same ay its valuc at 2p. That is, f is constant on G. Path Independence of Integrals The idea that an indefinite integral is an antiderivative does not carry over directly to the complex domain. Whet should we mean by the integral between two points? There arc many possible paths. The connection comes up in the study of one of the central questions we will study in this chapter: Under what conditions is the value of an integral independent af which path is selected between the two points? Consider the following two examples.104 Chapter 2 Cauchy's Theorem Example 1 Let zo = 1 and z; = —1 aud let f(z) = 322. Then F(z) = 2 is an antiderivative for f everywhere in the complex plane. Therefore, by the Fundamental Theorem of Calculus, no matter whut path y we take from 29 to 21 we will have f, f{(z)dz = F(z)) - F(zo) = 1°—(~1)® = 2. The value of the integral does not depend on the particular path selected, but only on the function and the two endpoints. @ Example 2 Again let 2 = 1 and z, = —1, but now take f(z) = 1/z. Let he the upper balf of the unit circle from 1 to —1. Then 7y is parametrized hy 7(t) = c! for 0 <14.< x. Thus, S(z)dz —Hinktdt = wi. fa [* restorers Now let 2 be the lower half of the unit circle from 1 to —1. Then 72 is parametrized by y2(t) =e for 0 $1 (ii) by picking two points 29 and 2 along a closed curve and thinking of it as amade up of two curves from one poiut to the other and then back to the first. The construction is illustrated in Figure 2.1.6, aud the computation runs us follows: [tene= [says [ some= f stents f seve Thns, the integral along the closed loop I is 0 if and only if the integrals along the paths Yo and 7 arc equal. y Figure 2.1.6: yw —71 =P. ‘The implication (iii) => (3) follows from the Fundamental Theorem. The value of the integral is F(z;) — F(z) regardless of which path is selected. To show that (i) = (iii), we will attempt to use an integral ending at < to define the value of the antiderivative at z. Let zo be any point fixed in G, and let z be any other point in G. Since G is open and connected, it is path-connected, and by Proposition 1.4.15 there is at least one smooth path in G from 2 to z. Let 7 be any such path and set F(z) =f f(()d¢. This defines a function F on G in a nonambiguous way since (i) says that the value F(z) depends only on z and not on the particular path selected so long as it stays in G. (OF course, it nlso depends on 2o, but that is fixed for the entire discussion.) We say that F is well defined. Our remaining task is to check that F is differentiable and that F’ = f. This computation is illustrated in Figure 2.1.7. Let € > 0. Since G is open and f is coutinuons at z, there is a number 5 > 0 such that the disk D(z;5) C G and [{(¢) — {()| < ¢ whenever |¢ —2| < 6. Suppose fw —2| <6. Connect z to w hy a straight-line segment p. Then all of p lies in106 Chapter 2 Cauchy’s Theorem Figure 2.1.7: Defining an antiderivative by an integral. D{z;6) and Fu) — F(2) = f foe rf Fo)ae = [ SoMa. wnz fw — 2] — Upflcras ~ He) J, tael fo 2} 2 fw -2] elength(p) _ (t) = cost + isint,0 <¢ < 4/2. Hence, [tes ly I f “Recntenhiteat = f tat = $ fi [Re(r2(2)) ns (tat = f idt =4 [remorse = [°-@-na=- [mecnrina= f"oa=o. 3 3 W “2 [Oo omsteont sine + toostat a [(emcotsint)-sint —e infsint) cost Oo x/2 +i [ [-e™¢ sin(sint) - sin + ®€ cos(sin) «cos t}dt o =" eos(cint) [3/7 + fe" siu(sint) = emitted108 Chapter 2 Cauchy’s Theorem Example 2.1.11 Let -y be the upper half of the unit circle described counterclock- wise. Show that 0. Then a= Gay is the derivative of an analytic function, so by the Fundamental Theorem 2.1.7, [@-a-ae=0. , Second, let n < —2. Then agnin (e-ayr = Zea, which is analytic on the set A = C\{a}. (Note that this formula fails if n = —1.) Since + lies in A, the Fundamental Theorum again shows that f(z — a)"dz = 0. Finally, let n = —1. We proceed directly and parametrize 7 by +(0) = re" + 0,0 < @ < 2x (see Figure 2.1.8).32.1 Contour Integrals 109 Figure 2.1.8: Paramctrization of the circle of radius r and center a. By the chain rule, -/(8) = ric, so 1 1 i. fal [ne-[ wie a0 = [ida = 2. in summary, 0 né-1 [ie-erae= {Oni ne=-1- This is a useful formula and we shall have occasion to use it later. Exercises 1. Evaluate the following: (n) [v4e, where 7s the union of the line segments joining 0 toi and then oi? ) [sin 22d, where isthe line segment joining § +1 to —i y ©) [ ze"dz, where is the unit circle 7 2, Evaluate the following: (a) [ed where +y is the union of the line segments joining 0 to i and then 7 toit+2110 Chapter 2 Cauchy’s Theorem (b) [22 +224 9d, were is the stnighttne segment joining 1 to 2+i ©) f Ay where 7 is the circle af radius 2 centered at 1 traveled quce counterclockwise 3. Bvatuate f (1/2), where 7 is the circle of radius 1 centered at 2 traveled once counterclockwise. 4. Evaluate f ode, where ix the curve in Bxercae 3, . Does Ref f yde} = [Ros as? 6. Prove Proposition 2.1.3. 7. Evaluate the following integrals: (a) 2s, where isthe unit circle travereed once in & counterclockwise direction (6) [et vP)ts, where i he etraighttine fom @ tof T 8.° Evaluate [# along two paths joining (0,0) to (1,1) as follows: v (6) 7 is the straight-line joining (0,0) to (1,1). (b) 7 Is the broken line joining (0,0) to (1,0), then joining (1,0) to (1, 1). In view of your answers to (a) and (b) and the Fundamental Theorem, could # be the derivative of any analytic function F(z)? 9. Find a number Af such that where 7 is the upper half of the unit circle. 10.* Let C be the arc of the circle |z| = 2 that lies in the first quadrant. Show that loses 1. Evaluate the following:42.2 Cauchy’s Theorem—A First Look lll LS Li Lon fn (b) fa, where + is the curve given by 4(t) =e! sin®1,0 << /2. dz 2 12, Let y be a closed curve lying entirely in the set C\{z | Rez < 0}. Show that [oteds=o. l 13. Evaluate f esinstd, where 7 is the unit circle. : 4. Give some conditions ona closed carve 7 that will guarantee that (1/2)dz =0. a 15, Let 7 be the unit circle. Prove that If 16.* Show that the arc length [(7) of a curve 7 is unchanged if 7 is reparametrized. 2.2 Cauchy’s Theorem—A First Look One form of Cauchy's Theorem states that ify is a simple closed curve (the word -simple” meaning that -y intersects itself only at its endpoints) and if f is analytic on and inside 7, then [1=o f ‘This remarkable theorem lies at the heart of complex analysis, and this section is devoted to its proof (see Figure 2.2.1). If the function f is not analytic on the whole region inside -y, then the integral may or may not be 0. For example, let 7 be the unit circle and f(z) = 1/z. Then f is analytic at all points except z = 0, and indeed the integral is not zero. In fact, [t= by Worked Example 2.1.12. On the other hand, if f(z) = 1/27, then f is still analytic at all points except z = 0, but now the integral is 0. This value of 0 results not from Cauchy’s Theorem: f is not analytic everywhere inside ~—but rather from the fact that f hus an antiderivative on C\{0} namely, f is the derivative of ~1/z. Mare generally, the Path Independence Theorem 2.1.9, shows that Cauchy’s Theorem is valid if there is an antiderivative of f. This is made explicit in Theorem. 22.5.112 Chapter 2 Cauchy’s Theorem Fnalytic hore "As inside of y Figure 2.2.1: Cauchy's Theorem: Lf =0. Green’s Theorem Our proof of Canchy’s Theorem uses Green’s Theorem from vector calculus, which states that, for continously differentiable functions P(x, y) and Q(z,u), aQ aP [Pesnds+oeney= ff Bees Feu] eae Recall that if 7: [0,8] + C,4(t) = (x(t), w(t)), then we define the line integrals by i Pla,y)de = f Pew) "0d [a= faco.vovioe. , /, In Green's Theorem, A represents the “inside” of 7, 7 is traversed in a counter- clockwise direction, and P and Q are sufficiently smooth— class C’ is sufficient, but we shall take a closer look at this later. Green's Theorem is a fundamental result from multivariable calculus that ev- ery student should know. Recall that the basic idea of the proof is really quite simple: We use the technique of evaluation of multiple integrals by iterated inve- gration (which in turn is related to equality of mixed partial derivatives) and the Fundamental Theorem of Calculus.” Part of the job of developing a context for Green’s Theorem is to make precise nolions like the inside of +y. Intuitively, the meaning of “inside” should be clear. ‘We shall come back to issues like this as we proceed. See a calculus text. such ns J. Marsden and A. Weinsicia, Calculus 111 (New York: Springes- Verlag, 1985), 908-911, o J. Maniten and A. ‘Tromba, Vector Caleutus, Fourth Edition (New York: W. H, Frocaan and Company, 1996}, §8.1, for a proof of Green's Thearcxa,§2.2 Cauchy’s Theorem—A First Look 113 Preliminary Version of Cauchy's Theorem The following statement is pre- liminary in the sense that it assumes that a satisfactory context for Green’s Theorem has been developed. We will come back to this puint and make things more precise in due course. Theorem 2.2.1 Suppose that f is analytic, with the derivative f' continuous on and inside a simple closed curve y. Then [t-o Proof Setting f = u + iv, we have = dz = iv) (dx +i: [1 = [1 f (u+iv)(de + idy) 7 [lode vay) +i [ (udy + 00m, ie | 7 By applying Green's Theorem to each integral, we get [eff 3 - Fe] acay +s ff [FE - $e] aca. Both terms are zero by the Cauchy-Ricmann equations. A closer look at the proof of Canchy’s Theorem given in §2.3 shows that one need not assume that f’ is continuous. Amazingly, the continuity of J" follows automatically, but this is not obvious. We also need to do some additional chores, such as eliminating the assumption that the curve is simple. In many cases, the assumption of simplicity of the curve can be avoided by viewing the path as being made up of two or more simple pieces. In Figure 2.2.2, the “figure eight” can be treated as two simple loops. We will discuss this in §2.3 as well. Here is a simple example of Cauchy's Theorem. Let 7 be the unit square and J(2) = sin(e**). Thicn f is analytic on and inside y (in fact, f is entire), 50 f, f = 0. Deformation Theorem It is important to be able to study functions that are not analytic on the entire inside of y and whose integral therefore might not be zero. For example, f(z) = 1/z fails to he analytic at z = 0, and f, f = 2mi where 7 is the unit circle. (The point z = 0 is called a singularity of f.) To study such functions it is important to be able to replace f, f by J, f, where 7 is a less complicated curve (say, a circle). The strategy is that if might be easier to evaluate. The Procedure that allows us to pass from -y to 7 1s based on Cauchy's Theorem and is as follows.4 Chapter 2 Cauchy's Theorem Figure 2.2.2: Treating a nonsimple curve as made up of several simple loops. ‘Theorem 2.2.2 (Preliminary Version of the Deformation Theorem) Let f be analytic on @ region A and let 1 be a simple closed curve in A. Suppose that 7 can be continuously deformed to another simple closed curve 7 without passing outside the region A. (We say that y is homotopic to { in A.) (The precise def- inition of “homotopic” is given in $2.3, and the assumption the curves are simple will be eliminated.) Then Ha os ‘The Deformation Theorem is illustrated in Figure 2.2.3. Figure 2.2.3: Deformation theorem. Note that f need not be analytic inside -7, so Cauchy's Theorem does not impir that the integrals in the preceding equation are zero. It is implicit in the stutemes:32.2 Cauchy's Theorem—A First Look 115 of the Deformation Theorem that both + and 7 are traversed in a counterclockwise airection. As with Cauchy's Theorem itself, we will come back to this theorem in she next section and have a closer, more careful look at the notion of “deformation” 4s well as a more careful look at the following proof. Proof Consider Figure 2.2.4, in which a curve 7 is drawn joining 7 to 7; we assume such a curve can in fact be drawn (it can in all “practical” examples). set a new curve consisting of 7, then -y, then —7, and then —%o, in that order. Yc Curve begins and ends here Figure 2.2.4: Curve used to prove the Deformation Theorem. The inside of this curve is the shaded region in Figure 2.2.4. The function f is, by assumption, analytic on this region. so Cauchy’s Theorem gives of fale fr [s- Ls [of Thus, f, f = f, f as required. Strictly speaking, this new curve is not a simple closed curve, but such an oljjection can be taken care of by drawing two parallel vopies of yo and taking the limit as these copies converge together. Simply Connected Regions A region A C C is called simply connected if A ty connected and every closed curve 7 in A can be deformed in A to some constant curve 7(l) = zp € A: wealsosay that 7 is homotopic to a point or is contractible to @ point. Intuitively, a region is simply connected when it bas no holes; this is because a curve thet loops around a bole cannot be shrunk down to a point in A without leaving A (see Figure 2.2.5). Therefore, the domain on which a function like f(z) = 1/z, which bas a singularity, is analytic is not simply connected. ‘We can rewrite Cauchy's Theorem in terms of simply connected regions as16 Chapter 2 Cauchy's e@@e © Figure 2.2.5: (u) Simply counected region. (b). (c) Non-simply connected regious. Theorem 2.2.3 (Cauchy’s Theorem for a Simply Connected Region) If f is analytic on a simply connected region G and + is a closed curve in G, then Independence of Path and Antiderivatives In the Path Independence Theo- rem 2.1.9 we saw how to relate the vanishing of integrals along closed curves to path independence of integrals between points and to the existence of antiderivatives on regions. We can exploit these ideas in the present context. Proposition 2.2.4 Suppose that f is analytic on a simply connected region A. ‘Thon for any two curves 7 and “42 joining two points zy and 2 in A (as in Figure 2.2.6), we have ieee Figure 2.2.6: Independence of path.42.2 Cauchy’s Theorem—A First Look uz Proof ® Consider the closed curve y = ¥2 — 11. By Cauchy's Theorem, on fs=fs-fs and therefore, asrequircd. @ Just as in Theorem 2.1.9, we also get the existence of an antiderivative for f on the region. Theorem 2.2.5 (Antiderivative Theorem) Let { be defined and analytic on a simply connected region A. Then there is an analytic function F defined on A that 1s unique up to an additive constant, such that F"(2) = f(z) for all z in A. We call F the antiderivative of f on A. Proof The existence of the antiderivative follows from the Path Indepondence Theorem 2.1.9. (Strictly speaking. we should get rid of the assumption of simple curves first.) The uniqueness assertion means that if Fo is auy other such function, then Fo(z) = F(z) +C for some constant C. This follows because the regiou A is connected and (Fo ~ FY(2) = Fa(2) - F(z) = §(2) - J) = 0 for all z in A. Thus, Fy — F is constant on A by Corollary 2.1.8. More on the Logarithm If A is not simply connected, the conclusions of the preceding proposition need not bold. For example, if A = C\{0) and f(z) = 1/2, there is no F defined on all of A with FY = f sinco the integral of f around the unit circle ig not zero (see Example 2.1.12). In some sense, F ought to be the Jogarithm, but we cannot define this in a consistent way on all of A. However, on any simply connected region not containing 0 we can find such an F as the following Proposition shows. Proposition 2.2.6 (Existence of Logarithms) Lct A be a simply connected re- gion and assume that 0 ¢ A. Then there is an analytic function F(z), unique up to the addition of multiples of 2xi, such that e"@) = z, In uhis proof we assume that 7 is a simple closed curve. Wa will show in the next section that this is not necessary.us Chapter 2 Cauchy's Theorem Proof By the Antidcrivative Theorem 2.2.5, there is an analytic function F with F(z) = 1/z on A. Fix a point 29 € A. Then 2p lies in the domain of some branch ‘of the log function defined in §1.6. If we adjust F by adding a constant so that F(zo) = log zo, then at zn, c7*) = 29. We now want to show that ¢"() = 2 is true on all of A. To do this, we let g(z) = o°*)/z. Then, since 0 ¢ A,g is analytic on A, and since F’(z) = I/z, g()= Fea =0. Thus g is constant on A. But g = 1 at zg, so g is 1 on all of A. Therefore, eF@) = = on all of A. For uniqueness, let F and G be functions analytic on A and suppose that ¢*) = zand eS) = z. Then ef2)-Gl) = 1, so at a fixed zp, F(z) — Gz) = 2mni for some integer n. But J*(z) = 1/2 = G'(z), 50 we have d(F -G)/dz = 0, from which we conclude (from the fact that a function with zero derivative on a connected region is a constant) that F-G=2nnion allof A. We write F(z) =logz and call such a choice of F a branch of the logarithm function on A. Clearly, this procedure generalizes that in §1.6 and we get the usual log as defined in that section if A is C minus 0 and the negative real axis. Note that this A is simply connected. However, the A in this proposition can be more complicated, as depicted in Figure 2.2.7. Figure 2.2.7: A possible domain for the log function.12.2 Cauchy's Theorem—A First Look 119 Worked Examples Example 2.2.7 Evaluate the following integrals: : | edz, where + is the perimeter of the unit square of ipetas, where is the unit circle Ly 11 [ ifede, where ts the cite 3+ 0,00 ¢ 2x : yf sds, where 9 is the segment joining 1 + i to 2 : Solution To solve (a), notice that e* is entire; thus, by Cauchy's Theorem, #, edz = 0, since 7 is a simple closed curve. Alternatively, e* is the derivative e®, and since 7 is closed we can apply the Path Independence Theorem 2.1.9. To evaluate the integral in (b), note that 1/z? is defined and analytic on C\{0} and is the derivative of —1/z, which is defined and analytic on C\{0}. By the Path independence Theorem 2.1.9 and the fact that the unit circle lies in C\{0}, we have # U/2?)dz = 0. Alternatively, we can use Worked Example 2.1.12 for our solution. Next, we solve (c). The circle y = 3 + c,0 < @ < 2x, does not pass through ' or include 0 in its interior. Hence 1/z is analytic on +y and the interior of 7, so by Cauchy’s Theorem, f,(1/z)dz = 0. An alternative but less direct solution is the fullowing. The region {x+-éy | z > 0} is simply connected and 1/z is analytic ou it. Therefore, by Propasition 2.2.6, 1/2 is the derivative of some analytic function F(z) one of the branches of log z) and thus, since + is closed, the Path Independence Theorem gives f,(1/2)dz = 0. Finally, to evaluate the integral in (d), note that 2? is entire and is the derivative of 23/3, which is also entire. By Theorem 2.1.7, [een Pp (+85 2 fy ee eta Example 2.2.8 Use the Deformation Theorem to argue informally that if 7 is a simple closed curve (not necessarily a circle) containing 0, then [peat ip Solution The inside of -y contains 0, so we can find an r > 0 such that the circle * of radius r and centered at 0 lies entirely inside y. Our intuition tells us that we can deform 7 to 7 without passing through 0 (that is, by staying in the region A = C\{0} of analyticity of 1/2; see Figure 2.2.8). Therefore, the Deformation Theorem and the calculation in Worked Example 2.1.12 give the required answer: [ies [yea peed120 Chapter 2 Cauchy’s Theorem se r IN | Figure 2.2.8: Deformation of 7 to the circle 7. Example 2.2.9 Outline a proof of the following extension of the Deformation The- orem: Suppose that ,... ,%m are nonoverlapping simple closed curves and that ~ és a simple closed curve with J analytic on the region betucen-y and-1,-.- 5% (see Figure 2.2.9). Then [-DL + Figure 2.2.9: Generalized Deformation Theorem.12.2 Couchy’s Theorem—A First Look 121 Solution Draw curves}, J2,-.- In joining 7 to 11...- . In. Pespectively, as shown au Figure 2.2.10(left). Let » deuote the curve drawn in Figure 2.2.10(right). The it aide of pis a region of analyticity of J, so J, f = 0. But p consists of 7, —71,—721-- n- and each 7 traversed twice in opposite directions, so the contributions from. “hese last portions cancel. Thus, on fief seo4f re fs-D fs Figure 2.2.10: Path used to prove the Gencralized Deformation Theorem. Example 2.2.10 Let f(z) be analytic on a simply connected region A, except pos- sibly not analytic at 2 € A. Suppose, however, that f is bounded in absolute value near 29. Show that, for any simple closed curve *y containing 20, ff = 0. Solution Let ¢ > 0 and small enough so thut the circle 4, of radius ¢ and center zy lies inside 7. By the Deformation Thcorem, ie Re By asvumption there bs a coustant. M with [f(z)| < Af near 29. Thus, if ¢ is small enough to mnke this estimate valid, then [A-l Asm Ee holds for all small cnongh positive ¢. Letting ¢ approach 0, we conclude that f=0 ,122 Chapter 2 Cauchy’s Theorem Exercises 1. Evaluate the following integrals: (®) [2+ 9d, where isthe upper half ofthe wat circle : () [02+ 9)s, where is the nit circle : (©) ff etlede, where isa circle of radivn 3 centred nt 53-41 s (@) [ couf3 + 1/(2 — 3)]d2, where 7 is a unit square with comers at.0,1, 1+, and i 2. Let 7 be a simple closed curve containing 0. Argue informally that 3. Let f be entire. Evaluate [F teatret for k an integer, k > 1. 4.° Discuss the validity of the formula log z = logr + 49 for log on the region 4 shown in Figure 2.2.7. 5. For what simple closed curves +7 does the equation dz Ue ay 6. Evaluate [ (2 —(1/2)}dz, where +y is the straight-line path from I to i. 5 hold? 7. Does Cauchy's Theorem hold separately for the real and imaginary parts of J7 If so, prove that it does; if not, give a counterexample. 8.° Let 7, be the circle of radius 1 and let 72 be the circle of radius 2 (traversed counterclockwise and centered at the origin). Show that [ de [ dz ly 2(Z7 +10) Sag 25(2? +10)"32.3 A Closer Look at Cauchy’s Theorem 123 9. Evaluate [ Vide, where 4 is the upper half of the unit circle: first, directly, and second, using the Fundamental Theorem 2.1.7. 10.* Evaluate [V2 =T de, where is» circle of radius } centered at 0. 7 H. Evaluate [ 227 - 152 +30 23 = 1027 + 322 — 32 where 7 is the circle |z| = 3. Hint: Use partial fractions; one root of the denominator is 2 = 2. 2.3. A Closer Look at Cauchy’s Theorem {n this section we take another look at sonic of the issues that were treated in- fonnally in the previous section. The strategy is to start by carefully examining Cauchy's Theorem for a rectangle and then to use the theorem in this special case. together with subdivision arguments, Lo build up vo more general regions in a sys- tematic way. Recall that the basic theme of Cauchy’s Theorem is that #f a function is analytic everywhere inside @ closed contour, then its integral around that contour must be ». The principal goal of this section is to give a proof of a form of the theorem known as a homotopy version of Cauchy’s Theorem. This approach exteuds and sharpens the idea presented in the preceding section of the continuous deformation of a curve. The primary objective will be the precise formulation and proof of deformation theorems which say, roughly, that if a curve is continuously deformed through a region in which a function is analytic, then the integral along the curve sloes not change. The reader will also notice that in this section references are made not to “simple closed curves” but only to “closed curves.” Cauchy's Theorem for a Rectangle We begin with a careful statement of Cauchy's Theorem in this case. Theorem 2.3.1 (Cauchy’s Theorem for a Rectangle) Suppose R is a rectan- gular path with sides parallel to the azes and that f is a function defined and analytic on an open set G containing R und its interior. Then fy f =0. ‘There are several methods to prove Cauchy's Theorem for a rectangle. One way, which fits the spirit of the previous section, is to prove a strong version of Green's Theorem for rectangles". Another technique, the one that we follow, is a bisection technique due to Edouard Goursat in 1884. It was Goursat® who first noticed that 4B, Ackor, The missing link, Mathematical Intelligencer, 18 (1996), 4 9. 5 Acta Mathematina, 4 (1884), 197 200 and Transactions of the Ameriom Mathematical Soci- ety, 1 (1900), 14-16,124 Chapter 2 Cauchy's Theorem cone does not need to assume that the derivative of f is coutinuous. Surprisingly, this follows automatically, which is a rather different. situation than that for real functions of several variables. Besides these techniques, there have been many other proofs of Cauchy's The- orem. For example, Pringslicim® nsos triangles rather than rectangles, which has some advantages. Cauchy's original proof (for which the assumptions of coutimiity of the derivative were not made clear), bad the content of Green's Theorem ini- plicit in the argument—in fact Green did not formulate Green's Theorem as such until about 1830, whereas Cauchy presented his theorem in 1825.7 There are also interesting proofs based on “homology” given by Ahifors.* Local Version of Cauchy’s Theorem Before proving Cauchy’s Theorem for a rectangle, we indicate how it can already be used lo prove a limited but still important and more gencrat case of Cauchy's Theorem. Theorem 2.3.2 (Cauchy's Theorem for a Disk) Suppose thal f : D + C ix analytic on a disk D = D(zo3p) CC. Then (i) f has an antiderivative on D: that is, there is u function F : D + C that is ‘analytic on D and that satisfies F(z) = f(z) for all z in D. (ii) IFT in any closed curve in D, then f.. f =0. From the discussion in §2.1 on the path independence of integrats (see Theorem 2.1.9), we know that (i) and (ii) are equivalent in the sense that whichever we establish first, the other will follow readily from it. Our problem is how to obtain either one of them. In the proof of the Path Independence Theorem 2.1.9, it was shown that (ii) follows ensily from (i), and the construction of an antiderivative to get (i) was facilitated by the path independence of integrals. The strategy for proceeding is quite interesting. 1. Prove (ii) directly for the very special case in which T is the boundary of a rectangle. 2. Show that this limited version of path independence is enough to carry out @ construction of an antiderivative similar to that in the proof of the Path Independence Theorem. 3. With (i) ee in its full generality follows as in the Path Independence Theor Transactions of the American Mathematical Society, 2 (1902) Tio his Mémoire sur les intégrales définies prises entre: des limites imaginazres. SL, Abifars, Complex Analysis. Second Edition (Now York: McGraw-Hill, 1966).2.3.A Closer Look at Cauchy’s Theorem 125 ‘roof of Cauchy’s Theorem for a Rectangle A subtle technical point worth ‘peating: Care must be taken because we do not know in advance that the deriva- ve of f is continuous. In fact, we will use Cauchy's Theorem itself to eventually rove that f’ is automatically continuous. Now lel’s get down to the proof. Let P be the perimeter of R and A the length of its dingonal. Divide the xtangle I into four congruent smaller rectangles RM, R@, A), and RO if ach is oriented in the counterclockwise direction, then cancellation along common Ages leaves fee tout fC fot ae Isls i f +l Pa +[fua4l ne zee must be at leust one of the rectangles for which | frum J| > U1 Sy J}. Call this of R abrectangle I. Notice that the perimeter and diagonal of 2, are ialf those Figure 2.3.1). y Figure 2.3.1: Bisection procedure. Now repeat this bisection process, obtaining a sequence A), Rz, Rs..-. of smaller nd smaller rectangles that have the following properties: oh teal Rant (i) Pevimeter(2q) = 3: perimeter(R) = = j2-2e A126 Chapter 2 Cauchy's Theorem (Gi) Diagonsl(R) = J dingonal(R) = > (cee Figure 2.32) Figure 2.3.2: Goursat’s repeated bisection process for the proof of Cauchy's Theo- rem for a rectangle. Since these rectangles are nested one within another and have diagonals tending to 0, they must shrink down to a single point we. To be precise, let z,, be the upper Icft-hand corner of Ry. If m > n, then |2n — 2m < diagonal (R,) = A/2", and thus {z,} forms a Cauchy sequence that must converge to some point two. If z is any point on the rectangle Rn, then since ull 2, with k > n are within Ry, z can be no farther from wp than the length of the diagonal of R,,. That is, |z—wo| < 4/2" for z in R,. From (i) we see that | fp fl <4"| fy, /|- To obtain a sufficiently good estimate on the right side of this inequality, we use the differentiability of f at the point wo. For e > 0, there is a number 6 > 0 such that, fi 2) fled ~ F"(un)| 0, we must have | f, {| = 0 and so fpf = 0, as iesired. Back to Cauchy’s Theorem on a Disk For most of the rest of this section, curve” means “piecewise C’ curve.” However, at one point in the technical devel- opment it will become i important to drop this piecewise C! restriction aud consider «omtinuous curves.? We can now carry out the second step of the proof of Cuuchy’s Theorem for a disk (Theorem 2.3.2). Since the function f is analytic on the disk D = D(z0;p), the result for a rectangle just proved shows that the integral of f is 0 around any rectangle in D. This is enough to carry out a construction of an antiderivative for £ very much like that done in the proof of the Path Independence Theorem 2.1.9 and thus to establish part (i) of the theorem. We will again define the antiderivative F(z) as an integral from z to 29. Hawever, we do not yet know that such an integral is path independent. Instead we will specify a particular choice of path and usc the new information available— the auulyticity of f and the geometry of the situation together with the rectangular case of Cauchy’s Theorem —to show that we get an antiderivative. For the duration of this proof we will use the notation (a, 6)) to denote the polygonal path proceeding from a point a to a point b in two segments, first parallel to the x axis, then parallel to the y axis, as in Figure 2.3.3. ~The technical treatment of Integration over continuous curves fs given in the Internet Supple- ment,128 Chapter 2 Cauchy’s Theorem Figure 2.3.3: The path (a,b). If the point 6 is in a disk 1(a; 8) centered at a, then the path (a, )) is contained in that disk, Thns, for z € D, we may define a function F(z) by F(:)= i, sony 1 ‘We want to show that F’(z) = f(z). To do this we need to show that tim = Fw)= Fle) woe wz = fiz). Fixing z € D and ¢ > 0, we use the fact that D is open and f is coutinuous on D to choose 6 > 0 small enongh that D{z;5) C D and |f(z) — f(é)| < « for € € D(z;6). If w € D(z;6), then the path (z, w)) is contained in D(z;4) and ence in D. ‘The paths (zy, z)) and (zo, w)) arc also contained in D, and these three paths fit together in a nice way with a rectangular path F also contained in D and having one comer at z; see Figure 2.3.4. We can write, for the two cases in Figure 2.3.4, if nay OH [Hows i, any OE rh sony OH By the Cauchy theorem for a rectangle, [p f(é)d£ = 0, so the preceding equation becomes F(e)+ aL voy SOE = Fl Neither side of the right triangle defined hy z,w)) can be any longer than it:$2.3 A Closer Look at Cauchy’s Theorem 129 Figure 2.3.4: Two possible configurations for R, 29, z, and w. hypotenuse, which has length |z — wl, s0 length ((z,t))) < 22 ~ w], and thus F(w) - F(z) waz L fw - 2] ie jw= 2] fh, Hee — Hey — | (=,0)) Foy ORF i ay we fle) - seed 2.0) -f0| = a lw - 2] Baal Ie Ur) - se)Nae| = aigthonitee) 5 sz joe Qfeo — 2] = 2e. AN A ‘ F(w) - F(z) os HAO) and therefore F"(z) = f(z), as desired. This establishes part (i) of the theorem. ‘Since f has an antiderivative defined everywhere on D and 7 is a closed curve in D, we have ff = 0 by the Path Independence Theorem 2.1.9. This establishes part (ii) of the theorem and so the proof of Cauchy's Theorem in a disk (Theorem 2.3.2) is now complete. Deleted Neighborhoods For technical reasons that will be apparent in §2.4, it will be useful to have the following variant of Cauchy's Theorem for a rectangle. Lemma 2.3.3 Suppose that R is a rectangular path with sides parallel to the axes, that f is a function defined on an open set G containing I and its interior, and130 Chapter 2 Cauchy’s Theorem that f is analytic on G except of some fized point 2, in G which is not on the path rete that et x, the function f satisfies lim..,(z — ») f(z) = 0. Then fpf =0. Notice that the limit condition in this lemma holds under any of the following three situations: (i) If f is bounded in a deleted neighborhood of 21 (ii) If f is continuous on G (ii) lim, _.., f(z) exists Proof If 2; is outside R, then the situation is really just that of the Cauchy theorein for a rectangle (Theorem 2.3.1), so we may assume that 2, is in the interior of R. For ¢ > 0, there is a number & > 0 such that [z— zi[[f(z)] < € whenever Jz — 2:| < 6. Choose 5 small enough to do this and so that the square S of side length 6 centered at 2; lies cntirely within R. Then everywhere along S we have If(2) < ¢/|z — 21). Now divide R into nine subrectangles by extending the sides of S, as shown in Figure 2.3.5. Figure 2.3.5: Construction of S and subdivision of R for the proof of Lemma 2.3.3. By Cauchy's Theorem for a rectangle, the integrals of f around all eight of the subrectangles other than S are 0, 0 ff = fp f. But along $ we have « sm VON< Fay Sa F since |z — 2;| 2 6/2 along S. Thus, [fa] s emer eon Ss§2.3.A Closer Look at Cauchy’s Theorem 131 Therefore, [Jn fl= fof < 8 for every €> 0 ‘Thus we must have | fy, f| = 0 and 30 fpf =0, as desired. If we strengthen the assumption on f and assume that it is continuous at 21, then we can drop the stipulation that z; not be on the path R. Lemma 2.3.4 Suppose that R is a rectangular path with sides parallel to the azes, that f is a function defined and continuous on an open sect G containing Rand Ppa tna aad ced te bitelaaeel inG. Then fpf =0. The only rea) problem is to make sure thut the integral is well behaved if the rectangle Ft happens to pass through 2;. In that case, the subdivision is a little different, but the estimates are simpler. Proof Again let ¢ > 0. We may choose 5 so that |f(z) - f(z)| < ¢ whenever -2—2| <4. Adding a constant to f, we may assume f(z;) = 0. If z) is not on R. then Lemma 2.3.3 applies. If it is on R, ict S be half a square of side 6, and subdivide Rt as shown in Figure 2.3.6. Figure 2.3.6: What happens if z; es on R. By Cauchy’s Theorem for rectangles, the integrals of f around all five of the jes other than S are 0, 80 fy f = ff. Along S we have |f(z)| < Thus, if we also require § < 1/3, we get | fs] tenet (sye = a8 ¥2/6. If Ras is the rectangle with corners at (8-14 tj—1)s (Seety—1)s (9e2t3)s (8e—-16t5), zen the whole rectangle is mapped into the disk Dag = D(H (6x—1,ty—1);p) that contained in G. Let [3 be the closed curve described by H(i%;) oriented by aking Ry; to be oriented in the counterclockwise direction.The image of each edge F each of the subsquares Ryj enters as part of two of the closed curves Ij and ith opposite orientation, except those subsquares along the outer edge where t or is O or 1. Notice that these edges actually piece together to make up the curves ait) and 2,(s). If we sum the integrals around all the loops I',;, all the edges used xice will cancel out and leave only ee j=tk=) ee Figure 2.3.15). Hl f).4) His.) Hoe tv) “igure 2.3.15: Cancellation of edges of the subsquares in the proof of the deforma- ion theorems. Since [yy is a closed curve lying entirely within the disk D,; on which the unction f is analytic, Cauchy's Theorem for a disk implics that each integral in he sum on the left is 0, so the right side is also 0. Thus, o=frefs-fs-[140 Chapter 2 Cauchy’s Theorem ie ea eile Up to this point the proofs for the fixec-endpoint case aud the closed-curve case have beeu the same. Now they diverge a bit. For the fixed-endpoint case, Ao(s) = H(s.0) = z for all », and Ay(s) = Hi(6,1) = 2, for all 3. Both nre constant curves, that is, single points, so fy. f = hf =0. * For the closcd-curve case, Ap and Ay ure the same curve: do(s) = H(s,0) = A(s,1) = da(s) for all s, 0 fy, f = J,, J. In cither caso, the equation f, f+ JJ = Sy ft hf becomes ff = J, J, which is exactly what we want. ‘The proof just given is actually not quite valid. The difficulty may appear to be an wuintcresting technical subtlety, but it is crucial nevertheless. The finction H bas been assumed to be continuans, but no assumption was made about differen- tinbility. Thus the curves 7.(!) and A;(s) are continuous but need not be piecewise C!. Unfortunmtcly, all our theory about: contour integrals is based on piecewise C’ curves. Thus, the integrals appearing previously do not necessarily make sense. They would, and everything would be all right. if all the curves in question were Piecewise C’. Therefore we make one more provisional definition and assutuption. ‘That is, Definition 2.3.13 A homotopy H : (0,1| x [0.1] ~ G ix called smooth if the intermediate curves 7,(t) are piecewise C' functions of t for cach s and the cross curves ),(#) are piereurise C functions of s for each t. Assuming that the homotopies in the Deformation Theore:zn are smooth, all the curves in the preceding proof are piecewise C', the integrals all make sense; and the proof is valid. With this additional assumption in place, we will refer vo the theorem as the Smooth Deformation Theorem. The technical discussion of Low tv relax the smoothness assumption is given in the Internet Supplement. ‘Thearem 2.3.14 (Homotopy Form of Cauchy’s Theorem) Let f be anulyti: on a region G. Let be a closed curve in G which is homotopic to a point in G Then [s-o Proof The curve + is homotopic in G to a constant curve A(t) = 24 for all ¢. Therefore, f,f=J,f=0. We can also prove Cauchy’s Theorcin for a simply connccted region in this context (see Theorem 2.2.3). Indeed, every closed curve 7 in G is homotopic to point in G, so the result follows from the homotopy form of Cauchy's Theorem.-23 A Closer Look at Cauchy’s Theorem 141 Old Results in the New Setting In the Path Independence Theorem 2.1.9 =» saw that the existence of antiderivatives is closely tied to path independence of =tegrals and to the vanishing of integrals around closed curves. Notice that the Acitiderivative Theorem 2.2.5, and Tlicorem 2.2.6 on the existence of logurithms, are valid with the sharper meaning of simply connected, ns discussed in this section. The Deformation Theorem, Cauchy’s Theorem, and al] these consoquences were wved from the conclusions of Cauchy's Theorem for a rectangle. From Theorem . we see thet all these conclusions remuin valid if we assume merely that f = contimous on G aud analytic on G\{z1} for some fixed z, in G. In §2.4 it ill be shown that this assumption implies that f is analytic on G so that such a smakening of the hypotheses of the theorems is only apparent. But it is necessary #7 the logical development of the theory. Worked Examples Example 2.3.15 Let A be the region bounded by the 2 axis and the curve o(0) = 2e.0 < 0 < x, where R > 0 is fired. Let f(z) = e* /(2R — 2)". Show that for sack closed curve y in A, ff =0. Solution First observe that f fails to be analytic only when z = 2f and hence f = analytic on A, since 2 lies outside A (see Figure 2.3.16). 4 i 1 Figure 2.3.16; Convex region, We claim that A is simply connected. That any two points in A can be joined bv a straight-line lying in A (that is, that A is convex) is obvious gcometrically and also is a simple matter to check (which the student should do). Hence A is simply counected, by Corollary 2.3.11. By Cauchy's Theorem, Sf = 0 for any closed curve in A. Example 2.3.16 Let A = {z €C|1< 2| < 4}. First intuitively, then procisely, show that A is not simply connected. Also show precisely that the circles |2| = 2 and |z| = 3 are homotopic in A.142 Chapter 2 Cauchy’s Thearem Solution Intuitively, the circle |z| = 2 cannot be contracted continuously to a point without passing over the hole in A; that is, the set {z € € such that J2] < 1}. Precisely, the function 1/z is analytic on A, and if A were simply connected, then we would have f (1/2)dz = 0 for any closed curve in A. But if we let (t) = 2e#,0 0 is the radius and the parameter range is 0 < t < 2en, has index n with respect to 2, while the circle y(t) = 29 + re7#, where again 0 St < 2zn, has index —n.146 Chapter 2 Cauchy’s Theorem Qo Index of circle ‘Yand y have the same index with respect to =, Figure 2.4.2: Index = the number of times that zp is encircled. (ii) Uf 20 does not lie on either ¥ or and if ¥ andy are homotopic in C\{z0). then TF z0) = 17520). Since homotopies can sometimes be awkwurd to deal with directly, it is custom- ary merely to give an intuitive geometric argument that J(-7; zp) has a certain value. but again the student should be prepared to give a complete proof when called for (see Worked Example 2.4.12 at the end of this section). The next result provides a check that the index I(7; 29) is always an integer. This should be the case if Definition 2.4.1 actually represents the ideas illustrated in the figures. Theorem 2.4.3 Let [a,b] — C be a (piecewise C*) closed curve and 29 a point not on 7; then I(7;z0) is an integer. = ¥(s) att) = 0 wen At points where the integrand is continuous, the Fundamental Theorem of Calculus gives Proof Let -_7) v= a(t) - 20° The right-hand side is the “logarithmic derivative” of 7(t) — 2. Motivated by this observation and using the product rule for derivatives, we can write the preceding display as fob) - a] =032.4 Cauchy’s Integral Formula uaz at points where g/(t) exists, Thus, e~9[y(t) — za] is piecewise constant on {a, }}. But e~9()[-(t) — zo] is coutinuous and therefore must be constamt on [2,6]. This Punstant value is 7 #)Fy(a) — 20), so we get o-%)f4(b) — za] = e-#°)[y(a) — za}. But (b) = (a), x0 e739) = o- 906). On the other band, 9(a) = 0; hence e~%) = 1. Thus g(b) = 2nni for an integer n, and the theorem follows. The index, (752) = (1/2n4) [, d¢/(¢ — 2), is a continuous function of z as long as = does not cross 7. (Why?) But we have just scen that if 7 is a closed curve, shen its value must be an integer. Thus it aust stay constant except when z crosses she curve, Caution: This need not be truc if 7 is not a closed curve. The inside of a closed curve + is defined by {z | I(y;2) # 0}; this definition agroos with the intuitive ideas illustrated in Figure 2.4.1. Derivation of Cauchy's Integral Formula Cauchy's Theorem will now be “sed to derive a useful formula relating the value of an analytic function at z toa certain integral. Theorem 2.4.4 (Cauchy’s Integral Formula) ' Let f be analytic on a region A. let be @ closed curve in A that is homotopic to a point, and let 2 € A be a punt not ony. Then S20) Hai) = ge This formula is often applied when + is a simple closed curve and zo is inside 7. Then 1(7; 20) = 1, so the formula becomes The preceding formula is remarkable, for it says that the values of J on + completely determine the values of f inside y. In other words, the value of f is determined by its “boundary valucs.” Proof The proof makes a clever use of the analyticity of f and the techni- cal strengthening of Cauchy's Theorem for which we laid the groundwork in the strengthened Cauchy's Theorem 2.3.5 for a disk, in which the function was allowed ‘©Cauchy’s Integral Formula can be strengthened ly requiring only that f be continuous on 7 and analytic inside 7. This change makes little difference in solving most. examples, For the proof of the strengthened theorem, the methods givun in the Internet Supplement for Chapter 2 and an approximation argument may be used. Soc aleo E. Hille, Analytic Punction Theory, Volume 1 :Boston: Ginn and Company, 1958.)48 Chapter 2 Cauchy’s Theorem to be merely continuous and not necessarily analytic at oue point. (See also Worked Bxuunple 2.2.10 and the remarks following Proposition 2.3.14.) Let S2)- Fo), wo-{ iO Seog f't) ifz=2 Then g is analytic except perbaps at 29, and it is continuous at z9 since f is differentiable there. Thus fg = 0, 50 v= [oan [ fhe [fe Lede =s00 f and therefore the — follows. at = 2nif(z0)1 (7370): Cauchy's Integral Formula 2.4.4 is extremely useful for camputations. For ex- ample, if + is the unit circle, we can immedintely calculate ee [fae= tie = ni. ly % In Cauchy's Lutegral Formula, we simply choose f(z) = e* and 2% = 0. Note that. in Cauchy's Integral Formula, it is f and not the integrand f(2)/(z - zp) that is analytic on A; the integrand is analytic only on A\{zo}, so we cannot use Cauchy’s Theorem to conclude that the integral is zero—in fact, the integral is usually nonzero. Integrals of Cauchy Type Cauchy's Integral Formula is a special and powerfur fornula for the value of f at zy. We will now use it to show that all the higher derivatives of f also exist. The central trick in the proof is an ides that is ofteu useful. Lf we start assuming only that we know the values of a function along « curve, then we can consider integrals along the curve as defining a new function called an integral of Cauchy type. Theorem 2.4.5 (Differentiability of Cauchy-Type Integrals) Suppose 1s ¢ curve in C and g #s a continuous function defined along the image ‘y({a, #}). Set el) G@)= Qat f C- (2% Then G is analytic on C\-y({a, b)); in fact, G is infinitely differentiable, with the kt derivative given by ote) = | Gen Qty b= 2d.:2.4 Cauchy’s Integral Formula 149 ‘The formula for the derivative can be remembered hy “differentiating with re- spect lo z under the integral sign”: W@ ae = Zot 2" maa fy OZ The formal proof justifies this procedure and appears at the cud of this section. Existence of Higher Derivatives Using integrals of Cauchy type, we can show zaat a differentiable function of one complex variable is actually infinitely differen- ~sable and at the same time give a formula for all the derivatives. Theorem 2.4.6 (Cauchy Integral Formula for Derivatives) Let f be analytic =a region A. Then all the derivatines of { exist on A. Furthermore, for 2 in A -nd ¥ any closed curve homotopic to a point in A with 29 not ony, we hance hy) = IO ke F209) 145 20) = Le jays 1,2.3,..., here f) denotes the kth derivative of f. Proof Since A is open and 2g is not on 7, we can find a smal) circle yp ccntered #1 29 with interior in A and such tht 7 does not cut acress jo. (See Figure 2.4.3.) Far 2 in A and not on 7, define IQ Ole) = s2)- ais) = 55 f Pa. This is an integral of Cauchy type. so il is infinitely differentiable ou A\y, and Wf _s0) om f c-y* mil Pow wa tensa ss [ae are also integrals of Cauchy type, so they are infinitely differentiable near zp. As snentioned earlier, the index is constant except when z crosses the curve. In par- cicular, it is constant inside 4. Thus G(")(z) = f(zq)J(7; 20). Combining this with the preceding fornula for G(*)(z) gives the desired result. Sz) = Oni Cauchy's Inequalities and Liouville’s Theorem We continue developing the ~onsoquences of the Cauchy Theorem with an important set of inequalities for the derivatives of an analytic function and its consequences.150 Chapter 2 Cauchy’s Theorem Figure 2.4.3: Let 7 be a circle centered at 29 and small enough that it docs not meel 7. ‘Theorem 2.4.7 (Cauchy's Inequalities) Let J be analytic on a region A and let + be a circle with radius R and center zo that lies in A. Assume that the disk {z such that |z — z0| < R} also lies in A. Suppose that [{(z)| faster than a constant times k!/R*. We can use Cauchy's Inequalities to derive tas following surprising result: The only bounded entire functions are constants.424 Cauchy’s Integral Formula 161 Theorem 2.4.8 (Liouville’s Theorem}? Jf f is entire and there is a constant 34 euch that |f(2)| $M for all z €C, then f is constant. Proof The Cauchy Inequalities 2.4.7 with k = 1 show that for any z € C, the s+quality |{"(za)| < A¢/R holds. Holding zg fixed and letting — 00, we conclude reat [f’(zo)| = 0 and therefore that f’(z9) = 0. This is true for every zo in C, so f constant. This ix agnin a quite different property than any that could possibly hold for zuctions of a real variable. Certainly, there are many nonconstant bounded smooth S:uctious of a real variable, such as f(z) = sin. Fundamental Theorem of Algebra Next we shall prove a result that appears +» he clomentary and that the student has, in the past, probably taken for granted. sugebraically, the theorem is quite difficult.'? However, there is a simple proof that xes Liouville’s theorem. Theorem 2.4.9 ( Fundamental Theorem of Algebra) Let ay,a1,... ,@n be a collection of n +1 compler numbers and suppose that n > 1 and ay #0. Let pi) S Oo taiz+...+4,2". Then there exists a point 29 E C such that p(zo) = 0. Proof Suppose that p(zo) # 0 for all z €C. Then f(z) = 1/p(2) is entire. Now ) and hence f(z) is not constant (because an # 0), 50 it suffices, by Liouville’s “keorein, to show that f(z) is bounded. To do so, we first show that p(z) — 00 as z — 00, or equivalently, that f(z) + 0 4= 2 + 00. In other words, we prove that, given M > 0, there is a number A > 0 sich that [2] > i implics [p(z)] > Af. From p(z) = 09 +012 +... + an2" we have s42)] 2 feall2)"*—laol — laste] —.--—[an-all2|"~. (We set an2" = p(z)— 20— 212 — <7 2n-12""" and apply the triangle inequality.) Let a = |ap|+|ai|-+---+[an—11- We} = bet (Intl — 2 Ys] > 1, then lanza ) 2 [el'oao! -0). Let K = max{1, (Af + a)/laq|}; then, if jz] > 1, we have [p(z)| > M. Thus if |z| > 1, we have 1/p(z)| < 1/M. But on the set of z for which |2| < K, the function 1/p(z) is bounded in absolute value because it is continuous. If this ™ According to E. T. Whittaker and G. N. Watson, A Course of Modern Analysis, Fourth Edition (London: Cambridge University Press, 1927), p. 105, Liouville’s thoorem is incorructly ributed to Liouville by Borchardt (whom others copied), who heard it in Liowville's lectures in Tt in dae to Cauchy, in Comptes Rendus, 19 (1844), 1377-1378, although it may have been known to Gauss earlier (veo the uxt footnote). "211 was first proved by Karl Lvicdrich Gauss in his doctoral thesis in 1799. The present proof appears to be ementially due to Gauss ax well (Comm. Soc. Gott, 3 (1816), 59-64),152 Chapter 2 Cauchy’s Theorem bound for 1/p(z) is denoted by L, then on C we have 1/|p(z)| < max(1/M, L), so If (2)| is bounded on C. By Review Exercise 24 ut the end of Chapter |, the polynomial p can have no. more than n roots. It follows by repeated factoring that p has exactly n roots if they are counted according to their multiplicity. Another arguinent for showing that f(z) + 0 as z — co that is a little simpler but accepts the validity of various limit theorems is ax follows: 1 {0) = Dayan pts yz" Oy + On—1(1/z) + an—2(1/2?) +... + aa(1/z")° Letting = — 00, we get 0 aee0e ero im, f(z) = since oy #0. The existence of higher derivatives, when combined with that of antiderivative- (Theorem 2.2.5), has some important consequences. Two of these are a partial converse to Cauchy's Theorem and an extension of the existence of logarithms. Morera’s Theorem Cauchy's Theorem says roughly that the integral of an an- alytic function around « closed curve is 0. Morera’s Theorem says that if we know the function ix continous we can get the converse implication. Theorem 2.4.10 (Morera’s Thearem) Let f be continuous on a region A, ona suppose that [ f = 0 for every closed curve in A. Then f is analytic on A, anc {=F for some analytic function F on A. Proof The existence of the antiderivative follows from the vanishing of integra around closed curves and the Path Independence Theorem 2.1.9. The antiderivatin, P is certainly analytic (its derivative is f). Therefore, by the Cauchy Integra! Formula for Derivatives 2.4.6, it is infinitely differentiable. lu particular, F” = 7 cxists, In applying Morera’s Theorem, one often wishes only Lo show that f is aualyr ov a region. If the region is not simply connected, { might not have an antiderivative on the whole region. But to show differentiability near a point one may restrict attention to a small neighborhood of he point and to special curves if convenient ‘This idea is illustrated in the following corollary and Worked Examples 2.4.16 anc 24.17. Corollary 2.4.11 Let J be continuous on a region A and analytic on A\{29) jer a point 2 € A. Then f ix analytic on A.§24 Cauchy's Integral Formula 153 Proof To show analyticity at, z), we may restrict attention to a small disk D{z,¢) C A. I 7 is any closed curve in this disk. then f, f= 0, by the strengthened Cauchy theorem for a disk (2.3.5). Thus Morera’s Theorem 2.4.10 implies that. f is analytic on this disk. We alrendy know it is analytic on the rest of A. More on Logarithms In Theorem 2.2.6, we used the existence of antiderivatives to obtain logarithms on simply counected regious not containing tle origin. With the existence of higher derivatives we can get « more gencral version of this result. Proposition 2.4.12 (Logarithms of Functions) Let f(z) be a function that is analytic and never 0 on a simply connected region A. Then there ia a function (2) analytic on A and unique up to the addition of a constant multiple of 2xt such that 32) = f(z) for all z in A. Proof In effect, we are looking for a logarithm for f(z). Lf g(z) is to be such a function. we must have f’(z) = e9)q/(z) = f(z)y'(z). Since f(z) is never 0, this says that g'(z) = f"(z)/f(z). Thus, g must be an antiderivative for f'/f on the connected open set A. The difference between two such antiderivatives would have derivative 0 and so be constant on A. Since the now function would still have to be a logarithm for f(z), that. constant would be an integer multiple of 2xi. It remains to show that there is such # fiction. Since f” exists, f’ is analytic on A. Sinoc f(z) is never 0, the quotient f*(2)/f(z) is analytic on the simply connected region A. From the Autiderivative Theorem sev Theorem 2.2.5), there is a function g : A + € such that g/(z) = f"(z)/S(z) for ull z in A. Fix zo in A. By adjusting our antiderivative by adding a constant, we way assume that 9(z9) is any convenicut choice of n value of log(f(zo)). Let qs) = em) /f(2). Then We) = IOLA) _ Hey fle) MOF) 9 (42) U@)P Since h'(z) is identically 0 on the counected open set A, the function h must be constant on A. But h(29) = e!°8(/(+0))/ (29) = 1. Therefore, orf Hla) = h(z) = (29) = 1 for all z in A. Thus, e+) = f(z) on A, as required. Tf we apply this resnlt to the function f(z) = 2 on a simply connected region not containing 0, we recover our earlier results ou logarithms (soc Theorem 2.2.6). Technical Proof of Theorem 2.4.5 We will prove Theorem 2.4.5 with a some what weaker assumption on g than continuity. Alt we assume is that the function is bounded and integrable along +y. We call such functions admissible. First we use several facts from advanced calculus that were developed in §1.4. The image curve ¥ is 8 compact sct since it is a continuous image of a closed bounded interval. Tf 2p is not on +, then by the Distance Lemma 1.4.21, it lies at a positive distance154 Figure 2.4.4: A point zo not on # curve 7 is a positive distance from +. & from it. If we let 7 = 6/2 aud U be the 9 disk around zy, then z € U and ¢ on > implies that |z—¢| > 9 since |z—¢| 2 IC — zal —[zo— 2 2n— 2q (see Figure 2.4.4). We begin with the case k = 1. We want to show sin aS — aa Lee) =* ‘The expression in brackets may be written G(z) - Geo) _ gos = =) 9(6) a6, “Ont wal oF G- Qt 2—2% I (G — 20)*(C- 2) where we have used the identity — ts es) - ee eS a-m\C-z C-2) (6-20)? (6 -2)*(C-2) Let the 7 neighborhood U of 29 be constructed as previously described and let M be the maximum of g on 7. Then |(¢ — z9)*(¢—z)| > 97-9 = 1°, 50 we have the estimate |9(6)/I(¢ — 20)?(¢ — 2) < M@n7 (a fixed constant independent of ¢ on - and 2,2) € U). Thus, 2-% a6) \< _ yur f erepeca| Seat. This expression approaches 0 as z — 29, so the limit is 0, as we wanted. ‘To prove the general case, we proceed by induction on &. Suppose the theorer. is known to hold for all admissible functions and all values of & from 1 to n — i We want to prove that it works for k =n. We phrase the induction bypothesis ths way since we will apply it not only to g, but also to 9(¢)/{¢ — zo), which ix alsx§2.4 Cauchy’s Integral Formula 155 bounded and integrable along 7. We know that G can be differentiated n —1 times on C\y and that GON) = a f £% a. Let 29 € le Using the identity i . z-% Gy =F * €-2""-a) © C-2"C =)" we obtain GE (2) — GEN (29) . oa! e Fee nner ae eer €-ay «| a 0) ae) | eo We can conclude from this equation that G"-) is continuous at z9, for che — reason. By applying the induction hypothesis to 9(¢)/(¢ — 20), we see c laa as aualytic as a function of z on the set C\7([o,4]) and thus is continuous in z. Therefore, 0) a6) [ Haya) ~ of =a as z— 2g. If the distance from 29 t0 +7 is 2n, if |g(2)| < Af on +, and if |z —2al <9, we have 9(6) . |, C=C Gaara |< pe tO where I(7) is oe Bence 9(¢) fy (C — 2)°(6 — 20) as z -+ 2, and therefore Gl'~) is continuous on C\y((a, d})- From the equation for G"—))(z) — Gl" )(zq), we obtain GUMs) — G-Me9) 2-20 _ (lt ato) tat (e= 79) Uf erieat [ee wor] (e-1)! 90) Oni [ C=) 4|--0 FS156 Chapter 2 Cauchy’s Theorem By applying the induction hypothesis to 9(¢)/(¢ — 20), we see that the first tenn on the right side of the preceding equation converges to (n- forte = f__9) I C= 20) agri as z ~+ . We have already shown that G-) is continuous on C\7(la, 0), and this fact applied to 9(C)/(¢ — 20), instead of to 9(¢), implies that 9) _, Sto Lasttca> | Gay as z 7.29. Thus we have shown that as z —+ zo, GIs) — GO M(29) zz at ae g, = 9(6) [cae SY | ee at a(6) al aore This concludes the induction and thus proves the theorem. Worked Examples Example 2.4.13 Consider the curve 7 defined by y(t) = (cost, 3sint),0 << 4z Show that 1(30) =2. Solution The strategy is to show that + is homotopic in C\{0} to a circle 7 that is centered around the origin and that is traversed twice in the counterclockwis- direction (that is, 7(t) = e“,0 < t < 4x). Once this is done, by Proposition 2.4.2. 1(59) = 1740) = 2 A suitable homotopy is H(t, s) = cost+i(3—2s) sin; note that H is continuous. H(t,0) = (1) and H(E,1) = 7(t), and is never zero (sce Figure 2.4.5). Example 2.4.14 Evaluate [Ree ot [Bas z Lz where 7 is the unit circle.i24 Cauchy's Integral Formula 157 Figure 2.4.5: Homotopy of (1) = (cost, 3sint) to 7(t) = (cost, sint). Solution The circle + is contractible to a point in the region in which cosz is analytic, since in fact cosz is entire. Therefore, we can apply Cauchy's Integral Formula 2.4.4, observing that J(+7;0) = 1, to obtain L=cs0= Sas, dni, 2 = [ ee = Oi. ln 8 By the Cauchy Integral Formula for Derivatives 2.4.6, we ave 0) = 1 f a8 sit) = 55 | res that is, Example 2.4.15 This example, which deals with analytic functions defined by in- tegrals, generalizes Theorem 2.4.5. Let {(z,w) be @ continuous function of z,w for158 Chapter 2 Cauchy’s Theorem ee For cach w ony assume that f is analytic in 2. F(2)= f Slevw)dw. Show that F és analytic and Pa)= LF Le wjaw, where 8 /z denotes the derivative of f with respect to z with w held fixed. Solution Let z € A. Let yp be a circle in A around zg whose interior also lies in A. For z inside yo, L(G) F(z.) = L,, C2 ade by Cauchy's Integral Formula 2.4.4. Thus Fe) = alle ow), de Next we claim that we may invert the oo cttaagtion, thus obtaining rnd [Seal an sh | Fx This procedure is justifiable because the integrand is continuous and when writ- ten out in terms of real integrals has the form [ff wonasaes [f° esse ‘We know from advanced calculus that this order can be interchanged (Fubini’s theorem)*5, Thus, Fe=aq f Fou, re FO ig 162) 5, FO = sal quart ial, [eee _ 16.) = ie Gogo = [% 2 (¢,a)dao, 13S¢e, for instance, J. Maniden and M. Hoffman, Blementary Classical Analysis, Second Editice (New York: W. H. Frooman and Company, 1993), Chapter 9.22.4 Cauchy's Integral Formula 159 again by Cauchy's Integral Formula. Since 29 is arbitrary we obtain the desired xsoult. Remark The function f should be analytic in z but it needs to only be aitegrable in the w variuble, as is evident from the preceding proof; we merely need an adequate hypothesis to justify interchanging the order of integration. Example 2.4.16 Prove the following assertion: Suppose that f is continuous on a gion A and that for cach 2y in A there is a disk D = D(29;p) such that f, f =0 Jur enery rectangular path R in D with sides parallel to the ares. Show that f is sualytic on A (sce Figure 2.4.6). Figure 2.4.6: If fy, f = 0, then f is analytic. Solution Let z be in A. The vanishing of f, f for rectangles in D was the conclusion of Cauchy's Theorem 2.3.2 for a rectangle and the tool used iu the construction of the antiderivative for f in the proof of Cauchy’s Theorem for a slisk. Thus the antiderivative exists on 1) (not necessarily on all of A at once). Analyticity on D follows as the proof of Morera’s Theorem 2.4.10, so f is analytic uear 2y. Since z was an arbitrary point in A, f is analytic on A. Example 2.4.17 Prove the following: Suppose A is a region that intersects the veal aris and that f is a function continuous ou A and analytic on A\R. Then f 1s analytic on A. Solution We know f is anslytic everywhere in A except on the real axis, so suppose z) € R. Since A is open there is a disk D = D(z;p) C A. Let Rbea rectangular path in this disk with sides parallel to the axes. If R does not touch or cross the real axis, then f, f = 0 by Canchy’s Theorem. If it does cross, as in Figure 2.4.7, then fr f = Sr, f+ Jn, J+ where i and Re are rectangles with one160 Chapter 2 Cauchy’s Theorem Figure 2.4.7: Construction used to show fy, f = 0. edge on the axis. (The edges on the axis are traversed in opposite directions and 80 cancel out.) ‘Thus it is enough Lo show that f, f = 0 when R is a rectangle with one side on the real axis, as in Figure 2.4.8. Let a and 6 be the ends of the edge on the axis and note that b-o 0. Since f is continuous, it is uniformly continuous on the compact set composed of 2 and its interior, so there is a > O such that |f(zi) — f(z2)| <¢ whenever |z; — zo] < 6 and 2 and zp are in this set. We may also choose 5 to be less than €. Let M be the maximum of |f(z)| on R and its interior, and let S be another rectangle the same as R except that the edge on the axis ix moved a distance 6 from the axis. Then with the notation of Figure 2.4.8, fe-L = LeeLee Le La tele Lala s sue +| fe) 1+ sn w < + 6M s 2m +f Ute)- He + Bide S 26M +(b-a) < (2M +p).02.4 Cauchy’s Integral Formula 161 ¥ Figure 2.4.8: The rectangte pulled away slightly from the real axis, Since this holds for every ¢ > 0, we must have fp f - fg f = 0: But J, f = 0 by Cauchy's Thearem since S does not cross the axis and jies entirely within a region in which F ix known to be analytic. Thus fy, f = 0. We have shown that the conditions of the preceding Worked Example apply, #0 f is analytic on A. Exercises 1, Evaluate the following integrals: 2 f 2 dz, where + is a circle of radius 2, centered ut. 0 z-1 (b) [3 Faz, where + is the unit circle 2, Evaluate the following integrals: 2. (a) [Fe Rite where 7 ia circle of radius 2, contered at 0 5 (by f Aas, where 7 i the unit circle 7 3. Let f be entire and assume that |f(z)| < M|z|" for large |z|, for « constant M, for some integer n. Show that f is a polynomial of degree < n. 4." Let f be analytic “inside and on” a simple closed curve 7. Suppose that J =0 on 7. Show that f =0 inside +. 5. Evalunte the following integrals:162 Chapter 2 Cauchy’s Theorem , where + is the square with vertices —1 — i,1—i,1+i,-1+44 we of o [5S 502 42, where 7 is the unit circle G. Let f be analytic on a region A and let 7 be a closed curve in A. For any 29 € A not on +, shaw that Lé ase ae [ oe Cau you think of a way to generalize this result? 7. Suppose that f(z) is analytic on the set |z| <1 and that it satisfios the inequality |f(z)| < 1. What estimate can be made about |J"(0)|? 8.° Suppose that f is entire and that lim... {(z)/z = 0. Prove that f is constant. 9. Prove that if is a circle, 7(t) = 20 +re"*,0 1 on the whole complex planc. Prove that J is constant. 18.° Docs . = 0? Does = ater 2? oo lpi 3 19. Evaluate () 2 de = 0? ee he-tfu2 2-2 dz © | aarm 20. Prove that for closed curves 11,772: (1320) = —1(ni 20) (mn +925 20) = L(niz0) + L212). Interpret these results geometrically. 21.° Let. f be analytic inside and on the circle 7 : |z — 2o| = R. Prove that fel ie) =i) HSS? - re ae | [e=ayezay~ ap" for z;, 29 inside -y.164 Chapter 2 Cauchy's Theorem 2.5 Maximum Modulus Theorem and Harmonic Functions One of the most powerful consequences of the Cauchy Integra) Formula is the Mazimum Modulus Theorem, also called the Maximum Modulus Principle. it states that if f is a nonconstant analytic function on a region A, then |f] cannot Lave a local maximum anywhere inside A—it can attain a maximum ouly on the boundary of A. This theorem and the Cauchy Integral Fonmnla will be used to develop sume of the important properties of harmonic functions. Maximum Modulus Theorem The ceutral idea of the Maximum Modulus Principle can perhaps best be stated as follows: If an analytic function has a local maximum (of its absolute value) al a point, then it must be coustant near that point. A preliminary version of the theorem follows. Theorem 2.5.1 (Maximum Modulus Principle—Local Version) Let f be analytic on a region A and suppose that |f| has a relative maximum at 29 € A. (That és, |f(z)| < |f(z0)| for all z in some neighborhood of 29.) Then f i constant in some neighborhood of 29. The proof rests on a striking cotisequence of the Cauchy Integral Formula: The value of an analytic function at the center of a circle is the average of its valurs around the circle. All this will be made precise shortly, but the local version of the principle follows essentially because the average of a function cannot be greater than or equal to the values of the function unless they are all equal. We develoy this idea in preparation for proving the Maximum Modulus Theorem. Theorem 2.5.2 (Mean Value Property) Let f be analytic inside und on a c1-- le of radius r and center 2y (that is. analytic on a region containing the circle an< its interior). Then Fea) = fc Sle + rei*)a0, (251 Proof By Cauchy's utegral Formula 2.4.4, Heo) = 5; [ LOhae, where 9(0) = 29 + re"9,0 <0 < 2x. However, by definition of the integral, [ie af Lear) gg tp Heo+re yo,«25 Mazimum Modulus Theorem and Harmonic Functions 165 Res > worth noting that as long as we are integrating all the way around the circle, 5 dues nol: matter through what range of 2x the angle gocs. A change of variable “bows that, for example, fig £ (20 + re) dd = f*, f(zo + rel) dd. The Mcan Value Property will now be used to establish the local version of the Maximum Modulus Principle. The idea is that if f(z) is at least as grent as all sae other values of f near zo and also equal to the average of those values around ssuall circles centered at 29, then |f(z)| must be constant near zg. Once we know czat |f| is constant, it follows from the Cauchy-Riemann cquations that f is itself + stant. Proof of Theorem 2.5.1 Suppose that f is analytic and has a relative maximum 3 20 s0 that [f(z)| < [f(z0)| on some disk Do = D(zoiro). We want to shaw that = [f(zo)| on Do, so suppose instead that there is a point z, in Dp where Grit inoquallty holds: |f(21)| < Lf (e0)|- Let 21 = 20+ re! with r < ro. Since f is catinuous, there are positive numbers ¢ and 6 snch that UF (20 + re™)| < [f(z0)| - 6 whenever |@ — al < ¢. Equivalently, [fle + ret] < [f(za)| - shenever |4| < €. We now obtain a contradiction by using the Mean Value Property and considering scparately that part of the integral over the circle where we know the function is smaller (see Figure 2.5.1) Figure 2.5.1: Construction for the proof of the Maximum Modulus Principle—tocal version. |¢| < € gives a part of the circle where |f| is known to be smaller.166 Chapter 2 Cauchy’s Theorem To carry this out, write 2 f lots) Fe fF Meat rete yaa] [teow retorntes Ef saat rete yag + x [ ” feat relerorag] tf [Fleo-+ rater] abt f° [pen + rele) db \F(z0)| w A In the first and third inegrals the integrand is no greater than |{(z0)| and the interval length is — ¢. Thus each of these integrals is no more than |f(zo)|(m —¢). In the middle integral, the interval length is 2e and the integrand is less than U(z0)| - 4. Hence this integral is no more than ([f(zo)| — 6)2«. Putting these together gives UfCeoM < Sel colle — 0) + (Uf Ccadl —B)2e + UNCzallen ~)] ie, [Fol < fo) - =. ‘This impossibility shows that there can be no such point z in Do with ({(2)| < {f(20)|- The only remaining possibility is that |f(z)| = |{(zo)| for all z in Do. ‘Thus || is constant on Do. Use of the Cauchy-Riemann equations as in Worked Example 1.5.21 shows that the function f itself must be constaut. This is exactly what we wanted. The loca! version of the Maximum Modulus Principle says that an analytic fanction cannot have « local maximum point unless it is constant near that point ‘We will soe in Chapter 6 that more is true. A function analytic on an open connected set cannot have a local maximum anywhere in that sct unless it is constant on the whole set. ‘We now turn our attention here to a somewhat different global version of thi: principle. We investigate absolute maxima, thot is, the largest value |f(z)| takes anywhere in the set. We shall show that this can be found ouly on the edge or boundery of the set. In §1.4, we saw that a real-valued function continuous on 3 closed bounded set actually attains a finile maximum hut that it might not if the set fails to be closed or bounded.32.5 Maximum Modulus Theorem and Harmonic Functions 167 Closure and Boundary The intuition in §1.4 was that a set is closed if it con- tains all its boundary points and open if it contains none of them. Thus if we start h a set A and add to it any of its boundary points which happen Lo be missing we should obtain a closed set containing A. This is true, but there are some technical problems. One is that we really have no definition yet for “boundary.” Definition 2.5.3 The closure of a set AC C, denoted by A or by cl(A), consists of A together with the limit points of all convergent sequences of points of A. ‘This produces the desired result, the smallest closed set containing A. Proposition 2.5.4 If ACC, then @) Acei(A). (ii) A és closed if and only if A = cl(A). titi) AC C and C is closed, then cl(A) C C. (iv) el(A) is closed. Proof The first assertion is immediate from the definition. Tle basic tool for the Temainder is Proposition 1.4.8, which states that a set is closed if and only if it contains the limits of all convergent sequences of its points. If we Ict limit(A) = {w | there is a sequence of points in A convergent to w}, then A C limit(A), since constant sequences certainly converge. The closure was defined by cl(A) = AU limit(A), so we actually have cl(A) = limit(A). But Propo- sition 14.8 says exactly that A is closed if and only if limit(A) C A, so (ii) is established. It also shows that if C is closed and A C C, then limit(A) C C, so we have (iii). The only remaining gup is to show that cl(A) is actually closed. To do this we noed only show that cl(A) = cl(cl(A)) that is, limit(A) = limit(limit(A)). Since limit(A) C limit(limit(A)) automatically, it remains to show that limit(limit(A)) € Timit(A). Suppose 21, 22,23,--- i8 a sequence of points in limit(A) such that lintty soo Zn = w. We want to show that w is in limit(A). Each 2, is in limit(A), so there arc points w, in A with |w_ — zn} < 1/2. This forces limy—oo Wn = tt, 90 w € limit(A), as desired. @ The boundary of a set A is the set of points on the “edge” of A. If w is in the boundary, we should be able to approach it through A and through the complement of A. This lends to the following definition. Definition 2.5.5 The boundary of a set ACC is defined by bd(A) = el(A) Nel(C\ A). It is not hard to sec that cl(A) = AUbd(A). (Sce Worked Example 2.5.16.)168 Chapter 2 Cauchy's Theorem Global Maximum Modulus Principle Now we arc ready for the promisod global version of the Maximum Modulus Principle. ‘Theorem 2.5.6 (Maximum Modulus Principle) Let A be an open, connected. bounded set in € and suppose f : c\(A) — € is analytic on A and continuous on cl(A). Then |f| has a finite maximum value on cA) which is attained at some point on the boundary of A. If it is also attained in the interior of A, then f must be constant on cl(A). This theorem states that tbe maximum of f occurs on the boundary of A and that if that maximum is attained ou A itsclf, then f must be constant. This is a very striking remlt and is certainly a very special property of analytic functions. The values of |f| inside a region A must be smaller than the largest value of |/] on the boundary of A. One must exercise some care. For example, the Maximum Modulus Principle in this form need not be true if A is not bounded. In such a case the function need not be bounded on A even if it is on bd(A). (See Exercise 3.) In applications of this theorem, A will ofveu be the inside of a simple closed curve 4. so el(A) will be AU and bd(A) will be 7. It is reasonably clear that if A is bounded, so is its closure. If || < B for all = in A and 21,22,23,... is a sequence in A converging to mw, then |zy| converges to |u]. so Jur] < B. Thus el(A) = limit (A) is also bounded by B. From the Extreme Value Theorem 1.4.20, we know that a continuous real-valued function on a closed bounded set attains a maximum on that set. It follows that if M! = sup{{f(z)] such that z € cl(A)}, then Ad’ = [f(a)] for some a € cl(A). Proof of the Maximum Modulus Principle Since the real-valued function I/1 is continnons on the closed bounded set cl(A), the Extreme Value Theorem says thal, i attains a finite maximum value Af al some point in el(A). If it is not attained in the interior, then it umst be attained somewhere on the boundary since it is altained somewhere. We next show that if it is attained in the interior, then J must be cowstant on A. By continuity f will be constant on all of cl(A), so Af is attained on the boundary as well in this case also. Suppose there is a point a in A at which |f(a)| = Af. Define subsets of A by Ar ={2€ Al s(z)=f(a)} and Az =A\cl(Ai). If z is in A but wot in Ap. then it must be in cl(A,). Choose a sequence in A; converging to z. Since f is continuous on cl(A) and bas value f(a) at each point in the sequence, we also hnve f(z) = @ and z € Ay. Thns AC A) U Az. Since Ar C cl(A;), we certainly have A, M Az = @. The set Ag is the intersection of the open sets A and C \ cl(Aj). so it is open. Finally, if 29 € Az, then [f(za)| = [f(a)| = AY ‘Thus, [f| has a local mucin at zo and must be constantly equal to f(a) on disk centered at zo. Since A is open, we can take the radius small enough so that this disk is contained in A and hence in A,. This shows that A; is open. If the set= A; and Ap were both nouempty, they would disconect the connected set A. Since a € Aj, we must have Ap = @ and A = A). Thus f(z) = f(a) for all z in A and hence also in cl(A) as claimed.*25 Mazimum Modulus Theorem and Harmonic Functions 169 Schwarz Lemma The next theorem is an exampl: of au application of the Max- usum Modulus Theorem. This result is not one of the most basic results of the tevory, but it further indicates the type of severe restrictions that analyticity im- pases. This result will be quite useful in Chapter 5. Lemma 2.5.7 (Schwarz Lemma) Let f be analytic on the open unit disk A = iz SCI fz] <1} with f(0) =0 and |f(2)] <1 for each z in A. Then {f'(0)| <1 and [f(2)] < |e| for each z in A. If|f'(0)| =1 or if there is a point zp other than vn A with |f(z9)| = |zol, then there és a constant ¢ with cl = 1 and f(z) = cz for lz in A. Proof Let 9(2) = f(2)/z if z #0 and g(0) = f"(0). The function g is analytic an A because it is continuous ou A and analytic on A\{O} (see Corollary 2.4.11 to Morera's Theorem). Let A, = {z such that [z| m for (2,y) € A. (ii) If u(x, y) = m for some (z, y) € A, then wu is constant. ‘The Mininmm Principle for Harmonic Functions may be deduced by applying the Maximum Principle to —u. Dirichlet Problem for the Disk and Poisson’s Formula There is a very important problem common to mathematics, physics, and eugincering called the Dirichlet Problem. it is this: Let A be an open bounded region and let tg be a given coutinuous function on bd(A). Find @ real-valued function u on cl(A) that ir continuous on cl(A) and harmonic on A and that equals up on bd(A). There ure (reasonably difficult) theorems stating that if the boundary bd(A) is “sufficiently smooth,” then there always is a solution . However, we can ensily show that the solution is always unique. Theorem 2.5.12 (Uniqueness for the Dirichlet Problem) The solution to th- Dirichlet Problem is unique (assuming that there is a solution). Proof Let u and & be two solutions. Let @ = u- di. Then @ is harmonic and @=0 on bd(A). We must show that ¢ = 0. By the maximum principle for lurmonic functions, d(z,y) <0 inside A. Simi- larly, from the minimum principle, (x,y) >00n A. Thusd=0. @ ‘We want to find the solution to the Dirichlet Problem for the case where the region is an open disk. To do so we derive a formula that explicitly expresses the values of the solution in terms of its values on the boundary of the disk.32.5 Maximum Modulus Theorem and Harmonic Functions 173 Theorem 2.5.13 (Poisson's Formula) Assume that u is defined and continuous .n the closed disk {z such that |2| p, we conclude that for s > p. [u(se!)(o? — 9?)] /[s? + 0? — 2epcos(¢ — 9)] is a continuous function of ¢ and @ and hence (with ,¢ fixed) is uniformly con- tinuous on the compact set, 0 < @ < 2n,(r + )/2 < 8 0, show that there exist az € A and a¢€ A such that |z — 2] < ¢,|€ — zo] <€, and UN > Uo) LF) < Lf@)- Hint: Use the Maxinum Modulus Theorem. Prove Hadamard's Three-circle Theorem: Let f be analytic on a region containing the sct # in Figure 2.5.4. Let R= {z |r < |z| <1r3} and assume O 0 1e)={ 2? forz<0 is differentiable, but f(z) = 2z|. Thus, the second derivative does not exist at 0. Even if all the derivatives exist, the Taylor scries might not converge to the function. The function ale forz #0 1) = { 0 forz =0 'Sea, for exnmple, H. Cartan, Elementary Theory of Analytic Functions of One or Several Complex Variables (Reading, Mass: Addison-Wesley, 1963 and New York: Dover Publications, 1995). 183184 Chapter 3 Series Representation of Analytic Functions is an example. Using induction one can check that f/)(0) exists for all k (in the real-variabte sense) and f'*)(Q) = 0. Here all coefficients of the Taylor series at 0 are 0, so the resulting series is zero, which does not cqual f(z) in any nontrivial interval around 0. A tice thing about complex analysis is that neither of these difficulties arises. ‘This reinforces the fact that assuming the existence of a complex derivative is much stronger than assuming the existence of a real derivative. We discovered in Chapter 2 that as soon as the first derivative exists on a region, all the higher ones must also. We will find in this chapter that the sccoud difficulty also disappears. If f is analytic on a region A and 2p is in A, then the Tuylor series of f centered at = automatically converges to f on the largest open disk centered at z and contained in A. ‘The reader is probably familiar with the geometric series Se =lettP elt. oe ane which is valid provided |t| < 1. We wilt show in the first section that this work: just as well for complex as for real numbers, In §3.2 we will use it to expant the integrand in the Cauchy Integral Formula as an infinite series, integrate thie series term by term, and use the Cauchy Integral Formula for Derivatives 2.4.6 v recognize the resulting coefficients as the correct ones for a Taylor series. Building on the preparation in §3.1. in §3.3 we investigate the series represex- tation of a function analytic on a deleted uvighborhood, that is, a fuuction with az isolated singularity. The resulting series, called the Laurent series, yields valuubs information abont the bchavior of functions near singularities, and this belavior = the key to the subject of residues and its subsequent applications. 3.1 Convergent Series of Analytic Functions ‘We shall use the Cauchy Integral Fornmla 2.4.4 to determine when the limit uf » convergent sequence or series of analytic functions is an analytic function and whe. the derivative (or integral) of the limit is the limit of the derivative (or integra: of the terms in the sequence or series. The basic type of convergence studied < this chapter is uniform convergence; the Weierstrass Af Test is a basic tool uset determine such convergence. In §3.2 we shall be especially interested in the sper.a case of power series, but we should be aware that some important functions sc convergout series that are not power scries. such as Lhe Riemann zeta function »- Worked Example 3.1.15). The proofs of the first few results are slightly technical, and since they a analogous to the case of real series, they appear at the end of the section. Convergence of Sequences and Series We begin with some basic defnit:x and tenninology.§3.1 Convergent Series of Analytic Functions 185 Definition 3.1.1 A sequence z,,n = 1,2,3,... of compler numbers is said to converge to a complet number 20 tf, for each « > 0, there is an integer N such that n>N_ implies that |r — %| <6. Convergence of zm to zy is denoted by 2n + 29. An infinite series "3... ax. of compler numbers is said to converge to S, and we write Sans k=1 Vf the sequence of partial sums defined by sq = S:t_, ay connerges to S. ‘The limit of a convergent sequence is unique; that is, a sequence can converge to only one point zy. (This and other properties of limits were discussed in §2.4.) A sequence 2, converges iff it is a Cauchy sequence, in other words, if, for each «> 0, there is an N such that n,2 > N implies that |Z, — 2m] 0 there is an N’ such that n> N implies that |z, — zn+p| < ¢ for every integer p = 0,1.2,....) This property of C = R? follows from the corresponding property of R, and we shall accept it from advanced calculus. Corresponding statements for the series S772, a, can be made if we consider the sequence of partial sums Sq = Dopo, Ox. Since Susp — 8n = Dpores Gk, the Cauchy criterion for sequences becomes the Cauchy criterion for series: SCZ, ae converges iff, for each € > 0, there is an N such that n > N implies nip Y a ant 1. (ii) Comparison test: If S72, bx converges and 0 < ay < by, then De, a4 converges; if Dye diverges and 0 < cr S dy, then SE, dy diverges. (iii) p-series test: 3°, n~P converyes if p > 1 and diverges to 00 (that is, the partial sums increase without bound) if p < 1. (iv) Ratio test: Suppose thet lim n,|2at*| eit ad tity ls than Thee $n converges absolutely. If the limit is strictly greater than 1, the series diverges. If the limit equals 1, the test is inconclusive. (v) Root test: Suppose thot tirtn—co(|an|)'/ exists and is strictly less than 1 Then 5%, dn converges absolutely. If the limit is strictly greater than 1, th: series diverges; if the limit equals 1, the test is inconclusive. ‘There are a few other tests that we shall occasionally call upon from calculzs such as the alternating series test and the integral test. We assume the reader wt review them as the need arises. Uniform Convergence Suppose that f, : A — C is a sequence of functions 2: defined on the set A. The soqnence is said to converge pointwise iff, for eacr 2 € A, the sequence f,,(z) converges. The limit defines a new function f(z) on 4 A more important kind of convergence is called uniform convergence and is defian as follows. Definition 8.1.4 A sequence fa : A C of functions defined on a set A ws sax’ to converge uniformly to a function f if, for each € > 0, there is an N som- that n > N implies that |f,(z) — f(z)| < ¢ for all z € A. This is written “f, — 7 uniformly on A.” A series 5°, x(z) is said to converge pointwise if the corresponding p= tial sums n(z) = Dhar gx(z) converge pointwise. A series Of", ge(z) is saxd converge uniformly iff %,(z) converges uniformly.§3.1 Convergent Series of Analytic Functions 187 Evidently, uniform convergence implies pointwise convergence. The difference between uniform and pointwise convergence is as follows. For pointwise conver- gence, given € > 0, the N required is allowed to vary from point to point, whereas for uniform convergence we must be able to find a single N that. works for all z. Iu is difficult to draw the graph of a complex valued function of a complex variable, since it would require four rea] dimensions, but the corresponding notions for real-valued functions are instructive to illustrate. The geometric meaning of uniform convergence is shown in Figure 3.1.1. ¥ Figure 3.1.1: Uniform convergence on an interval (a, 4}. If € > 0, then for large enough n, the graph y = f(z) nmst stay inside the “c- tube” around the graph of f (the tube’s width is measured in the vertical direction). ‘The concept of uniformity depends not only on the functions involved but also on the set on which we are working. Convergence might be uniform on onc set but not ou a larger set. The following example illustrates this point. The sequence of functions f,(x) = 2" converges pointwise to the zero function f(x) = 0 for x in the half-open interval (0,1[, but the convergence is not uniform. The function value 7" takes much longer to get close to 0 for x clase to 1 than for z close to 0: by taking 2 close enough to 1, we need arbitrarily large values of n. The convergence is uniform on any closed subinterval {0,r] with r <1. Since the worst case is at x =r, whatever n works there also works for all smaller z. Soe Figure 3.1.2. ‘Theorem 3.1.5 (Cauchy Criterion) (i) A sequence fx(z) converges uniformly on A iff, for cach « > 0, there is an NN such thai n > N implies that |fn(z) - fa+p(2)| < € for all z € A and all p=1,2,3)...- (ii) A series D3, ge(2) converges uniformly on A iff, for each « > 0, there is an188 Chapter 3 Series Representation of Analytic Functions Figure 3.1.2: Convergence of z® to 0 is not unifonn on {| 0 < 2 < 1}. N such that n > N implies that n+p Ss ate] << ont) for all z € A ond oll p = 1,2,.... The next result states a basic property of uniform convergence. Proposition 3.1.6 If the sequence f,, consists of continuous functions defined A and if fy — { uniformly, then f is continuous on A. Similarly, if the functors 9x(z) are continuous and g(z) = 37%, 9x(z) converges uniformly on A, then gc continuous on A. Propositions 3.1.5 and 3.1.6 are proved at the end of this section. Thus, a uniform limit of continuous functions is continuous. If the convergens= is not uniform, then the limit might be discontinuous. For example, Ict -1 forz<-1/n Sola) = 4 nx for -Ijn 0 such that both of the following conditions hold. (i) lon (2) S Mu for allz€ A Then >, ga converges absolutely and uniformly on A. Proof Since J Ma converges, for any ¢ > 0 there is an N such that n> implies 1722, Mx << for all p = 1,2,3,.... (Absolute value hars are not needed because M, > 0.) Thus, n > N’ implies = DY az) < = lou(z) < x M M,, converges by comparison to the convergent geometric series 2 9r*. Therefore, our series converges uniformly on Ay by the Weierstrass M Test. It converges pointwise ou the sot A = {2 € € | |2| < 1} since each z in A is in A, for r close enough to 1. (See Figure 3.1.4). Figure 3.1.4: Region of convergence of }>(2"/n): uniformly on A,, pointwise on A. ‘This series docs not, however, converge uniformly on A. Indeed, if it did, 3" z”/n would converge uniformly on {0, 1[. Suppose that this were true. Then for any € > U there would be an NV such that 2 > N would imply that amt grt ante atagit tnep <* for all x € [0,1[ aud p =0,1,2,.... But the harmonic-type scries. 1 1 weet diverges to infinity (that is, the partial sums — 00), so we can choose p such that 1 1 Nteot Wap? Next, we choose x so close to 1 that 2+? > 1/2. Then aN alte wan {2 1 Wet wap?? (F+--+ wig) D643.1 Convergent Series of Analytic Functions 191 which is a contradiction. However, note that 9(z) is still continuous on A because it is continuous at each z, since each z lies iu some A, on which we do have uniform convergence. @ Series of Analytic Functions The next result, formulated by Karl Weierstrass in approximately 1860, is one of the main theorems concerning the convergence of analytic functions. Theorem 3.1.8 (Analytic Convergence Theorem) (i) Let A be an open set in C and let f,, be @ sequence of analytic functions defined on A. If fa — £ uniformly on every clased disk contained in A, then f is analytic. Further- more, f', — f' pointwise on A and uniformly on every closed disk in A (see Figure 9.1.5). (ii) If ox is @ sequence of analytic functions defined on an open set A in C and Hz) = TZ ge(2) converges uniformly on every closed disk in A, then g és analytic on A and g/{z) = Fetes sh(=) pointwise on A and also uniformly on every closed disk contained in Closed disk « (21 is - zl Sel Figure 3.1.5: Uniform convergence on closed disks in the set A. This theorem reveals yet another remarkable property of analytic functions that is not shared by functions of a real variable (compare §2.4). Uniform convergence usually is not sufficient to justify differentiation of a serics term by term, but for analytic functions it is sufficient. ‘The proof of the Analytic Convergence Theorem 3.1.8 depends on Morera’s Theorem and Cauchy's Integral Formula, which were studied in §2.4. To prepare for this proof let us first analyze a result concerning integration of sequences and series, Proposition 3.1.9 Let 7: [a,b] + A be a curne in a region A and let fn be a sequence of continuous functions defined on (fa, }) which converges uniformly to192 Chapter 3 Series Representation of Analytic Functions Ff on y([e,b]). Then [fe Similarly, if O&, 9n(z) converges uniformly on +, then we can interchange infinite gums and integrals: [ Youn) = x [ gn(z)dz. nel Proof The function f is coutinuous by Proposition 3.1.6 and so is i Given ¢ > 0, we can choose A such that n > N implies that |f,(z) — f(z) < = all zon 7. Then, by Proposition 2.1.6, [[- [o]=|[m-a]s [ust semes ff, 90 f,f = 0. Thus by Morera’s Theorem f is analytic on D{z:r). ‘We must still show that f, — f’ uniformly on closed disks. To do this we use Cauchy's Integral Formula for Derivatives 2.4.6. Let B = {z such that |z— zo} < r+ be a closed disk in A. We can draw a circle + of radius p > r centered ut zy the- contains B entirely in its interior (sce Worked Example 1.4.27 and Figure 3.1.6,. For any z € B, Lf eo 1 1G ey= an [ Ee ws rer= sn | ax by tbe Cauchy Integral Formula. By hypothesis, f, + f uniformly on the closa= disk {z such that |z— zal $ p}, which lies entirely in A. Then, given c > 0, we pick NV such that n > N implies that [fq(z) — f(2)| N implies that fa) — £0) < ¢ on 7. Note that ined-sen=|s5 [ BOO Sener ae raees eri B,|\¢-2| > p—r. Hence n > N implies that Wfalz)- FEN S 5 x ‘CaF ty) = wor Since pan rand oman tt ae aspen op he de result. Applying the Analytic Convergence Theorem 3.1.8 repeatedly we sec that the kth derivatives #" converge to f\#) uniformly on closed disks in A. Notice also that this theorem does not assume uniform convergence on all of A. For example, SOR 29/n on A = {z such that |2| <1} converges uniformly on the sets A, = {2 such that |z| 2"/n is analytic on A and that the derivative is $>2"~', which also converges on A. However, as that example demonstrated, we do have pointwise but not. uniform convergence on A; convergence is uniform only on each closed subdisk in A. Technical Proofs Now we provide the missing proofs. Proof of Proposition 3.1.2 By the Cauchy Criterion 3.1.5, given ¢ > 0 there is an N such that n > N implies nee DY bl 1. we have convergence if |r] < 1 and divergence if |r] > 1. Obviously, 729 r” diverges if |r| = 1, since r™ does not converge to zero. (i) The partial suns of the series °°, b, form a Cauchy sequence and thus the partial sums of the series }°;_, a also form a Cauchy sequence, siner for any k and p we have ax + Qiy2 + +--+ Oktp S be + begs +--+ Ontp- Hence S72. ax converges. A positive series can diverge anly to +-00, so given M > 0, we can find ko such that k > ko implies ¢) + cz +... +e, > AM Therefore, for k > ko,d: +2 +... +d > M, 50 D3-, di also diverges to oc. (iii) First suppose that p < 1; in this case 1/n” > 1/n for all n = 1,2,.... Therefore, by (ii), O52, 1/n? will diverge if 77, 1/n diverges. We now recall the proof of this from calculus: If s, = 1/1+1/2+...+1/k, then s, is a strictly increasing sequence of positive real numbers. Write gu a8 follows: 1 tt tl on = 145 4(545) + (E4545 +8) tet (Gogteté TN BRT oe 1 1 1 1 k 2 149+(3)+(Z)+--+(2) 145 Hence s, can be made arbitrarily large if & is made sufficiently large; thus Tea L/n diverges. Now suppose that p > 1. If we let watedede.ed oie 2We can alvo prove (iii) hy using the integral test for positive series (see any calculus text ‘The demonstration given here alsa proves the Cauchy condensation test: Let San be a series of positive terms with on¢1 San. Then Jaq converges if 3°32, 2403 convarges (see G. J. Porte An alternative to the integral tert for infinite saries, Am. Math. Monthly, 79 (1972), 634).§3.1 Convergent Series of Analytic Functions 195 then s, is an increasing sequence of positive real numbers. On the other hand, Spar = e+($+ +3)+(% te Deets 3) 12 4 abot (gigi) sh ae oy 7 1 1 - 7 patter pa tt gee < pT (Why?) Thus the sequence {sj} is bounded from above by 1/(1 ~ 1/2°-7); hence 7&2, 1/n? converges. tiv) Suppose that tim |n#t =r <1. Choose r’ such that r N implies | N, we have ler! < r'fax-a| < (r'Plox-2l S++ $ (F')*“ lanl. Bags ed eae nt series with ratio r’. If See tim, =r>1, choose 1 such that 1 N implies that eat >’, Hence favypl > (r’Plew|, and so litty—oo jan] = 00, whereas the limit would have to be zero if the sum converged (see Exercise 10). Thus, Der ae diverges. To see that the test fails if lim i, |e =1, consider the two series 1414144... and 2%, }/n? for p> 1. Tn both cases ett] —1, bus the Sst sores diverges and the second convert lim nso (v) Suppose that lintn-eo [an|"/" =r < 1. Choose 1’ such that r <1’ <1and N such that n > N implies that |anl!/* 1, choose 1 N implies that |an]'/" > r’ or, in other words, that lanl > (7)". Hence lim, co |an| = 00. Therefore, Sf ay diverges. ‘To show that the test fails when lim, —oc |an|"/" = 1, we use these limits from calculus: yun 1\"e tim, (2) =1 and tin (3) =1 nae: \n nmoo \ x! (take logarithms and use L'Hépital’s rule to show that (logz)/z — 0 os 2 00). But D2, 1/n diverges and °°. 1/n? converges. Proof of 3.1.5 (Cauchy Criterion) (i) First we prove the “if part. Let f(z) =limp—os fa(z), which exists because for each 2, fa(z) is a Canchy scquence. We wish to show that f, + f uniformly on A. Given € > 0, choose N such that [fn(z) — fnop(2)l < €/2 for n > N and all p > 1. The first step is to show that for any z and any n > N,|fn(z) — S(2)| < €. For z € A, choose p large enough so that Mfn+p(2) — f(2)| < ¢/2, which is possible by pointwise convergence. Then, by the triangle inequality, Maz) — S(2M S Ufn(2) — fas (2)| + Mfntn(z) — £2) < ¢/2 + f2 =. (Notice that although p depends on z, N docs not.) Conversely, if fn + { uniformly. given ¢ > 0 choose N such that n > NV implies |fn(z) — f(2)| < €/2 for all 2. Sincen+p>N, Unl2) — foe (2) $ Maz) — S(2) + F2) - Sntpl=) < 6/24 €/2 =e. (i) By applying (i) to the partial sums, we deduce (ii). Ml Proof of Proposition 3.1.6 It suffices to prove the assertion for sequences (Why?). We wish to show that for z9 € A, given ¢ > 0, there is a § > 0 such thet [s—zo] < 6 implies that [J(2)— (20) < ¢. Choose N such that |fy(2)—J(2)| <¢ 3 for all z € A. Since fy ix continuous, there is a 6 > 0 such that [fiy(2) — fre(zo)! < «/3iif fz — 2o| < 6. Thus, 1£(2) — F201 SLF(2) — See) + [fre(2) — Srv(z0)l + [fie (0) - S(20)1 < ¢3+efstes=e Of Note that: in the last step we need au A’ that is independent of z to conclud: that. both [fxv(z) — f(z)| < ¢/3 and [fn(20) — L(20)| < ¢/3- Worked Examples Example 3.1.10 Show that the sequence of functions x(x) = sin(z/n) conuere:s uniformly lo the constant function f(x) =0 for x in the interval (0,7).