Springer ‘New York Berlin Heidelberg Barcelona Hong Kong London Milan Paris Singapore Tokyo Undergraduate Texts in Mathematics ations 8. Axler FW. Gehring KA. RibetTheodore W. Gamelin Complex Analysis With 184 Ilustrations“Theodore W. Gamelin Department of Mathematics UCLA, Box 951555 Los Angeles, CA 90095-1555 USA, aitorial Board S. Axler EM, Gehring KA. Ribet Mathematics Department Mathematics Department Mathematics Department San Francisee State East Hall, University of California University University of Michigan at Berkeley san Francisco, CA 94132. Ann Arbor, MI48109 Berkeley, CA 947203640 USA Usa USA Mathematics Subject Classification (2000); 2801, 3001 Abrary of Congress Cataloging-in-Publication Data Gamelin, Theodore W. “Complex analysis /‘Theodore W. Gamelin, pcm, — (Undergraduate texts in mathematics) Ineludes bibliographical references and index. ISBN 0-387-95002-1 (alk, paper) ~ ISBN 0-387-85069-9 (softcover : alk. paper) 1. Mathematical analysis. 2. Functions of complex variables. 1. Tite Series. (4300 .625 2000 S15—de21 0.041905 Printed on acid-free paper. © 2001 Springer-Verlag New York, Inc. All rights reserved. This work may not be translated or copied in whole or in part ‘without the weitten permission ofthe publisher (Springer-Verlag New York, Inc, 175, Finth Avenuc, New York, NY 10010, USA), except for brief excerpts in connection with reviews oF scholarly analysis, Use in connection with any form of information stor- lage and retrieval, electronic adaptation, computer software, or by similar or dissim- lar methodology now known or hereafter developed is forbidden. ‘The ute of general descriptive names, rade names, rademarks, ete, in this publica- tion, even if the former ate not especially identified, is not to be taken as a sign hat Such names, as understood by the Trade Marks and Merchandise Marks Act, may accordingly be used freely by anyone Production managed by Timothy Taylor; manufacturing supervised by Erica Bresler. ‘Typeset by MacroTeX, Savoy, IL. Printed and bound by RR. Donnelley and Sons, Harrisonburg, VA. Printed in the United States of Amer! ge7es4321 ISBN 0-387-95095.1 spin 10772447 (hardoover) ISBN 0.387-95009.9 SPIN 10770762 (softcover) Springer-Verlag New York Berlin Heidelberg 1 member of BertelemannSpringer SciencorBusiness Media GmbH To the many wonderful students at Perugia who took my course, and to the others there who contributed to an enjoyable experiencePreface This book provides an introduction to complex analysis for students with some familiarity with complex numbers from high school. Students should be familiar with the Cartesian representation of complex numbers and with the algebra of complex numbers, that is, they should know that ® = —1. A familiarity with multivariable calculus is also required, but here the funda- ‘mental ideas are reviewed. In fact, complex analysis provides a good trai ing ground for multivariable calculus. It allows students to consolidate their understanding of parametrized curves, vangent vectors, arc length, gradients, line integrals, independence of path, and Green’s theorem. The ideas surrounding independence of path are particularly difficult for stu- dents in calculus, and they are not absorbed by most students until they are seen again in other courses. ‘The book consists of sixteen chapters, which are divided into three parts. ‘The first part, Chapters I-VI, includes basic material covered in all un- dergraduate courses. With the exception of a few sections, this material is much the same as that covered in Cauchy's lectures, except that the em- phasis on viewing functions as mappings reflects Riemann’s influence. The second part. Chapters VIII-XI, bridges the nineteenth and the twentieth centuries. About half this material would be covered in a typical under- graduate course, depending upon the taste and pace of the instructor. The ‘material on the Poisson integral is of iaterest to electrical engineers, while the material on hyperbolic geometry is of interest to pure mathematicians and also to high school mathematics teachers. The third part, Chapters XIL-XVI, consists of a careful selection of special topies that illustrate the scope and power of complex analysis methods. These topics include Julia sets and the Mandelbrot set, Dirichlet series and the prime number theo- rem, and the uniformization theorem for Riemann surfaces. The final five chapters serve also to complete the coverage of all beckground necessary for passing PhD qualifying exams in complex analysis Note to the instructor ‘There is a glut of complex analysis textbooks on the market. It is a beaut ful subject, so beautiful that a large number of experts have been moved toiit Preface ‘write their own accounts of the area. In spite of the plethora of textbooks, T have never found an introduction to complex analysis that is complevely suitable for my own teaching style and audiences. ‘The students in each of my various audiences have begun the course with 1a wide range of backgrounds. Teaching to students with disparate back- «grounds and preparations has posed a major teaching challenge. I respond by including early some topics that can be treated in an elementary way and yet are usually new and capture the imagination of students with already ‘some background in complex analysis. For example, the stereographic pro- jection appears early, the Riemann surface of the square root function is ‘explained early at an intuitive level, and both conformality and fractional linear transformations are treated relatively early. Exercises range from the very simple to the quite challenging, in all chapters. Some of the exercises that appear early in the book can form the basis for an introduction to a more advanced topic, which can be tossed out to the more sophisticated students, Thus for instance the basis is laid for introducing students to the spherical metric already in the first chapter, though the topic is not ‘taken up seriously until much later, in connection with Marty’s theorem in Chapter XIL ‘The second problem addressed by the book has to do with flexibi of use. There are many routes through complex analysis, and many structors hold strong opinions concerning the optimal route. I address this problem by laying out the material so as to allow for substantial flexibil- ity in the ordering of topics. ‘The instructor can defer many topics (for instance, the stereographic projection, or conformality, or fractional linear transformations) in order to reach Cauchy's theorem and power series rel- atively early, and then return to the omitted topics later, time permitting. There is also flexibility with respect to adjusting the course to under- graduate students or to beginning graduate students. ‘The bulk of the book was written with undergraduate students in mind, and T have used various preliminary course notes for Chapters I-XI at the undergraduate level. By adjusting the level of the lectures and the pace I have found the course notes for all sixteen chapters appropriate for a first-year graduate course sequence. ‘One of my colleagues wrote in commenting upon the syllabus of our undergraduate complex analysis course that “fractional powers should be postponed to the end of the course as they are very difficult for the stu- dents.” My philosophy is just the reverse. If a concept is important but difficult, I prefer to introduce it early and then return to it several times, in ‘order to give students time to absorb the idea. For example, the idea of a branch of a multivalued analytic function is very difficult for students, yet it is a central issue in complex analysis. I start early with a light introduction to the square root function. The logarithm function follows soon, followed by phase factors in connection with fractional powers. ‘The basic idea is returned to several times throughout the course, as in the applications of Preface x residue theory to evaluate integrals. I find that by this time most students are reasonably comfortable with the idea. ‘A solid core for the one-semester undergraduate course is as follows: Chapter I Chapter IT Sections TIL-5 Sections IV.1-6 Sections V.1-7 Sections VI1-4 Sections VIL1-4 Sections VIILI-2 Sections IX.1-2 Seetions X.1-2 Sections XL1-2 To reach power series faster I would recommend postponing 1.3, 116-7, TIL4-5, and going light on Riemann surfaces. Sections 16-7 and TIL45 should be picked up again before starting Chapter IX. Which additional sections to cover depends on the pace ofthe instructor and the level of the students. My own preference is to add more contour Integration (Sections VIL5 and VIL8) and hyperbolic geometry (Section 1X3) to the syllabus, and then to do something more with conformal map- ping, as the Schwarz reflection principle (Section X.3), time permitting. To gain time, T mention some topics (as trigonometric and hyperbolic fune- tions) only briefly in class. Students learn this material as well by reading and doing assigned exercises. Finishing with Sections XI.1-2 closes the ci- cle and provides a good review at the end of the term, while at the same time it points to a fundamental and nontrivial theorem (the Riemann map- ping theorem). Note to the student You are about to enter a fascinating and wonderful world. Complex analysis is a beautiful subject, filled with broad avenues and narrow backstreets leading to intellectual excitement. Before you traverse this terrain, let me provide you with some tips and some warnings, designed to make your journey more pleasant and profitable. Above all, give some thought to strategies for study and learning. ‘This is easier if you are aware of the difference between the “what,” the “how,” ‘and the “why,” (as Halmos calls them). The “what” consists of defini- tions, statements of theorems, and formulae. Determine which are most important and memorize them, at least in slogan form if not precisely. Just fs one maintains in memory the landmark years 1066, 1453, and 1776 as markers in the continuum of history, so should you maintain in memory the definition of analytic function, the Cauchy-Riemann equations, and the residue formula. The simplest of the exercises are essentially restatements of “what.”x Preface ‘The “how” consists in being able to apply the formulae and techniques to solve problems, as to show that a function Is analytic by checking the Cauchy-Riemann equations, or to determine whether a polynomial has a zero in a certain region by applying the argument principle, or to evaluate fa definite integral by contour integration. Before determining “how” you must know “what.” Many of the exercises are “how” problems. Working these exercises and discussing them with other students and the instructor fre an important part of the learning process. ‘The “why” consists in understanding why a theorem is true or why a technique works. This understanding can be arrived at in many different ‘ways and at various levels. There are several things you can do to under- stand why a result is true. Try it out on some special cases. Make a short synopsis of the proof. See where each hypothesis is used in the proof. Try proving it after altering or removing one of the hypotheses. Analyze the proof to determine which ingredients are absolutely essential and to deter- mine its depth and level of difficulty. The slogan form of the Jordan curve theorem is that “every closed curve has an inside and an outside” (Section VIII.7). What is the level of difficulty of this theorem? Can you come up with a direct proof? ‘Try it Finally, be aware that there is a language of formal mathematics that is related to but different from common English. We all know what “near” means in common English. In the language of formal mathematics the ‘word carries with it a specific measure of distance or proximity, which is, traditionally quantified by © > 0 or a “for every neighborhood” statement. Look also for words like “eventually,” “smooth,” and “local.” Prepare to absorb not only new facts and ideas but also a different language. Develop- ing some understanding of the language is not easy ~ it is part of growing up and becoming mathematically sophisticated. Acknowledgments This book stems primarily from courses I gave in complex analysis at the Interuniversity Summer School at Perugia (Italy). Each course was based on a series of exercises, for which I developed a computer bank. Gradually I deposited written versions of my lectures in the computer bank. When I finally decided to expand the material to book form, I also used notes based ‘on lectures presented over the years at several places, including UCLA, Brown University, Valencia (Spain), and long ago at the university at La, Plata (Argentina). I have enjoyed teaching this material. I learned a lot, both about the subject matter and about teaching, through my students. T would like to thank the many students who contributed, knowingly or unknowingly, to this book, ‘The origins of many of the mathematical ideas have been lost in the thickets of the history of mathematics. Let me mention the source for one item. ‘The treatment of the parabolic case of the uniformization theorem follows a line of proof due to D. Marshall, and I am grateful for his sharing Preface xi his work. As far as I know, everything else is covered by the bibliography, and I apologize for any omissions. Each time I reread a segment of the book manuscript I found various ‘mathematical blunders, grammatical infringements, and stylistic traves- ties, Undoubtedly mistakes have persisted into the printed book. I would appreciate receiving your email about any egregious errors you come across, together with your comments about any passages you perceive to be par- ticularly dense or unenlightening. My email address, while I am around, is ‘twg@math.uclaedu. I thank you, dear reader, in advance. Julie Honig and Mary Edwards helped with the preparation of class notes, that were used for parts of the book, and for this I thank them. Finally, I am happy to acknowledge the skilled assistance of the publishing staff, who turned my doodles into figures and otherwise facilitated publication of the book. T.W. Gamelin Pacific Palisades February 2001Contents Preface vil Introduction xvii FIRST PART Chapter I The Complex Plane and Efementary Functions 1. Complex Numbers 1 2. Polar Representation 5 3, Stereographic Projection 11 4. The Square and Square Root Functions 15 5. The Exponential Function 19 6. The Logarithm Function 21 7. Power Funetions and Phase Factors 24 8. ‘Trigonometric and Hyperbolic Functions 29) Chapter Il Analytic Functions 33 1. Review of Basic Analysis 33 2. Analytic Functions 42 3. The Cauchy-Riemann Equations 46 4. Inverse Mappings and the Jacobian 51 5. Harmonic Functions 54 6. Conformal Mappings 58 7. Fractional Linear Transformations 63 Chapter IIT Line Integrals and Harmonic Functions 70 1. Line Integrals and Green’s Theorem 70 2. Independence of Path 76 3. Harmonic Conjugates 83 4. The Mean Value Property 85 5. ‘The Maximum Principle 87 6. Applications to Fluid Dynamics 90 7. Other Applications to Physics 97Contents Chapter IV Complex Integration and Analyticity 102 Po one eeere Pieters Q Beane a enogeeee® eoseee Complex Line Integrals 102 Fundamental Theorem of Calculus for Analytic Functions 107 Cauchy's Theorem 110 ‘The Cauchy Integral Formula 113 Liouville’s Theorem 117 Morera’s Theorem — 119 Goursat’s Theorem 123 Complex Notation and Pompeiu’s Formula 124 apter V Power Series 130 Infinite Series 130 Sequences and Series of Functions 133 Power Series 138 Power Series Expansion of an Analytic Function 144 Power Series Expansion at Infinity 149 ‘Manipulation of Power Series 151 ‘The Zeros of an Analytic Function 154 Analytic Continuation 158 .pter VI Laurent Series and Isolated Singularities 165 ‘The Laurent Decomposition 165 Isolated Singularities of an Analytic Function 171 Isolated Singularity at Infinity 178 Partial Fractions Decomposition 179 Periodic Functions 182 Fourier Series 186 apter VIT ‘The Residue Calculus 195 The Residue Theorem 195 Integrals Featuring Rational Functions 199 Integrals of Trigonometric Functions 203, Integrands with Branch Points 206 Fractional Residues 209 Principal Values 212 Jordan’s Lemma 216 Exterior Domains 219 SECOND PART Chapter VIII The Logarithiic Integral 224 ‘The Argument Principle 224 Rouché’s Theorem 229 Hurwitz’s Theorem 231 Open Mapping and Inverse Function Theorems 232 Critical Points 236 Winding Numbers 242 Contents 7 8 ‘The Jump Theorem for Cauchy Integrals 246 Simply Connected Domains 252 Chapter IX The Schwarz Lemma and Hyperbolic Geometry 260 1 2 3. ‘The Schwarz Lemma 260 Conformal Self-Maps of the Unit Disk 263 Hyperbolic Geometry 266 Chapter X Harmonic Functions and the Reflection Principle 274 1 2 3 The Poisson Integral Formula 274 Characterization of Harmonic Functions 280 ‘The Schwarz Reflection Principle 282 Chapter XI Conformal Mapping 289 1. Mappings to the Unit Disk and Upper HalfPlane 289 2. The Riemann Mapping Theorem 204 3. The Schwarz-Christoffel Formula 296 4. Return to Fluid Dynamics 304 5. Compactness of Families of Functions 306 6. Proof of the Riemann Mapping Theorem 311 ‘THIRD PART Chapter XII Compact Families of Meromorphic Functions. 315, L 3, 4 thay 1 2. 3, 4 Marty's Theorem 315 ‘Theorems of Montel and Picard 320 Julia Sets 324 Connectedness of Julia Sets 333 ‘The Mandelbrot Set. 338 ;pter XIIT Approximation Theorems 342 Runge’s Theorem 342 ‘The Mittag-Leffler Theorem 348 Infinite Products 352 ‘The Weierstrass Product Theorem 358 Chapter XIV Some Special Functions 361 ‘The Gamma Function 361 Laplace Transforms 365, ‘The Zeta Function 370 Dirichlet Series 376 ‘The Prime Number Theorem 382 Chapter XV_The Disichlet Problem 390 1 Green’s Formulae 390 Subhiarmonie Functions 394 Compactness of Families of Harmonic Functions 398 The Perron Method 402 ‘The Riemann Mapping Theorem Revisited 406xvi Contents 6. Green's Function for Domains with Analytic Boundary 407 7. Green's Funetion for General Domains 413, Chapter XVI Riemann Surfaces 418 Abstract Flemann Surfaces 418 Harmonic Functions on a Riemann Surface 426 Green’s Function of a Surface 429 ‘Symmetry of Green’s Function 434 Bipolar Green’s Function 436 ‘The Uniformization Theorem 438 Covering Surfaces 441 Hints and Solutions for Selected Exercises 447 References 469 List of Symbols 471 Index 473 Introduction Complex analysis is a splendid realm within the world of mathematics, ‘unmatched for its beauty and power. It has varifold elegant and often: times unexpected applications to virtually every part of mathematics. It is broadly applicable beyond mathematics, and in particular it provides powerful tools for the sciences and engineering, Already in the eighteenth century Euler discovered the connection be- ‘tween trigonometric functions and exponential functions through complex analysis. (It was he who invented the notation e.) However, it was not until the nineteenth century that the foundations of complex analysis were laid. Among the many mathematicians and scientists who contributed, there are three who stand out as having influenced decisively the course of development of complex analysis. The first is A. Cauchy (1789-1857), who developed the theory systematically along the lines we shall follow, with the complex integral calculus, Cauchy’s theorem, and the Cauchy inte- gral formula playing fundamental roles. ‘The other two are K. Weierstrass (1815-1897) and B. Riemann (1826-1866), who appeared on the mathe- ‘matical scene about the middle of the nineteenth century. Weierstrass developed the theory from a starting point of convergent power series, and this approach led towards more formal algebraic developments, Riemann contributed a more geometric point of view. His ideas had a tremendous impact not only on complex analysis but upon mathematics as a whole, ‘though his views took hold only gradually. In addition to the standard undergraduate material, we shall follow sev- eral strands and obtain several poster theorems, which together with the ‘more elementary material cover what might be called the “complex anal- ysis canon,” the part of complex analysis included in the syllabus of the typical PhD qualifying exam. One of the strands we shall follow culminates in the prime number theo- rem. Already Euler in the eighteenth century had written down an infinite product for the zeta function, connecting the prime numbers to complex analysis. In the 1830's Dirichlet used variants of the zeta function to prove the existence of infinitely many primes in arithmetic progressions. Riemantwilt Introduction did fundamental work connecting the zeta function to the distribution of, prime numbers. And finally just before the close of the nineteenth century J. Hadamard and C.J. de la Vallée Poussin independently proved the prime number theorem using techniques of complex analysis Another strand we shall follow is the conformal mapping of domains in the plane and more generally of Riemann surfaces. We shall aim at ‘wo poster results: the Riemann mapping theorem and the uniformization theorem for Riemann surfaces. The definitive version of the Riemann map- ping theorem, which one finds in all complex analysis textbooks today, was proved by W. Osgood in 1900. The uniformization theorem for Riemann surfaces was proved independently in 1907 by P. Koebe and H. Poincaré, thereby solving Hilbert’s 22nd problem from his famous address to the International Mathematical Congress in 1900, ‘The first quarter of the twentieth century was one of rapid development. of the foundations of complex analysis. P. Montel put his finger on the notion of compactness in spaces of meromorphic functions and developed the theory of normal families. P. Fatou and G. Julia used Montel’s theorem in their seminal work around 1914-1921 on complex iteration theory. On another front, O. Perron developed in 1923 a powerful method for solving the Dirichlet problem. By the end of the first quarter of the twentieth century, the complex anal- ysis canon had been established, and neatly all the main results constituting the undergraduate and first-year graduate courses in complex analysis had been obtained. Nevertheless, throughout the twentieth century there has been much exciting progress on the frontiers of research in complex anal- ysis, and meanwhile proofs of the most difficult foundational results have been gradually simplified and clarified. While the complex analysis canon has remained relatively static, the developments at the frontier have led to new perspectives and shifting emphases. For instance, the current re- search interest in dynamical systems and the advent of computer graphics contributed to elevating the work of Fatou and Julia to a more prominent position. What lies before you is the distillation of the essential, the useful, and the beautiful, from twe centuries of labor. Enjoy! I The Complex Plane and Elementary Functions In this chapter we set the scene and introduce some of the main charac- ters. We begin with the three representations of complex numbers: the Cartesian representation, the polar representation, and the spherical rep- resentation. Then we introduce the basic functions encountered in complex analysis; the exponential function, the logarithm function, power functions, and trigonometric functions. We view several concrete functions w = f(z) as mappings from the z-plane to the u-plane, and we consider the problem of describing the inverse functions. 1. Complex Numbers A.complex number is an expression of the form z = x+y, where z and are real numbers. The component is called the real part of z, and y is the imaginary part of 2. We will denote these by 2 = Rez, y = Imz The set of complex numbers forms the complex plane, which we denote by C. We denote the set of real numbers by R, and we think of the real mumbers as being a subset of the complex plane, consisting of the complex numbers with imaginary part equal to zero. ‘The correspondence z=Etiy — (x,y) is a one-to-one correspondence between complex numbers and points (or vectors) in the Euclidean plane R?. The real numbers correspond to the axis in the Euclidean plane. The complex numbers of the form iy are called purely imaginary numbers. ‘They form the imaginary axis iR2 1 The Complex Plane and Elementary Functions tween Harty g exty in the complex plane, which corresponds to the y-axis in the Euclidean plane. We add complex numbers by adding their real and imaginary parts: (+ iy) + (u4iv) = (@+u) +i(y +0), ‘Thus Re(z + w) = Re(2) + Re(w), and Im(z + w) = Im(z) + Im(w) for z,w € C. The addition of complex numbers corresponds to the usual ‘componentwise addition in the Euclidean plane ‘The modulus of a complex number 2 = z+ iy is the length y/27 + ¥? of the corresponding vector (x,y) in the Euclidean plane. The modulus of 2 is also called the absolute value of z, and it is denoted by || bel = Vere ‘The triangle inequality for vectors in the plane takes the form, letul < kelt+hw, zwee By applying the triangle inequality to z = (zw) + w, we obtain |z| < 2 — w+ fo). Subteacting ltl, we obtain a very useful inequality, (4) le-ul > l-lw, wee. ‘Complex numbers can be multiplied, and this is the feature that distin- guishes the complex plane € from the Euclidean plane R?, Formally, the ‘multiplication is defined by (e+ ty)(u + iv) = 2u—yo+ i(zv + yu). One can check directly from this definition that the usual laws of algebra hold for complex multiplication: (vz2)z3 = z1(z222), (associative law) aim = tat (commutative law) ai(zz +s) = 2129 +212. (distributive law) With respect to algebraic operations, complex numbers behave the same as real numbers. Algebraic manipulations are performed on complex numbers using the usual laws of algebra, together with the special rule i = —1 1. Complex Numbers Every complex number 2 # 0 has a multiplicative inverse 1/2, which is given explicitly by LL oetiy 2 yy +vec, 240. ‘Thus for instance, the multiplicative inverse of # is 1/1 ‘The complex conjugate of a complex number z= + iy is defined to be Z=x~ iy, Geometrically, 2 is the reflection of = in the z-axis. If we reflect twice, we return to 2, 2 2eC. Some other useful properties of complex conjugation are TFG = 740, zwec, FO = 20, z,weC, lel = [h, zEC, le? = 23, 2€C. Each of these identities can be verified easily using the definition of 2 and |2|. The last formula above allows us to express 1/2 in terms of the complex conjugate 2: Az = 2/le, reCz #0, ‘The real and imaginary parts of z can be recovered from z and Z, by Rez = (2+3)/2 zeG, Imz = (2~2)/2i, zeC From |zt|? = (2w)(#0) = (22)(wiB) = |2[2|w)?, we obtain also [ew] = l2llwl, z,wec. A complex polynomial of degree n > 0 is a function of the form Pz) = Gn2™ + a_az" 1 +---+aiz+a9, 2EC,4 I. The Complex Plane and Elementary Functions ‘where dg,... aq are complex numbers, and a, # 0. A key property of the ‘complex numbers, not enjoyed by the real numbers, is that any polynomial ‘with complex coefficients can be factored as a product of linear factors. Fundamental Theorem of Algebra. Every complex polynomial p(2) of degree n > 1 has a factorization ple) = ez ~ 21) + (z 4), where the 2's are distinct and m, > 1. This factorization is unique, up to permutation of the factors. ‘We will not prove tais theorem nowt, but we will give several proofs later. Some remarks are in order. ‘The uniqueness of the factorization is easy to establish. The points say. +4 are uniquely characterized as the roots of p(2), or the zeros of p(z). These are the points where p(z) = 0. The integer my is character ‘ized as the unique integer m with the property that p(2) can be factored as (2 — 2,)"q(z) where q(2) is a polynomial satisfying q(=,) # 0. For the proof of the existence of the factorization, one proceeds by indue- tion on the degree n of the polynomial. ‘The crux of the matter is to find ‘point 2; such that p(z1) = 0. With a root 2; in hand, one easily factors 2(2) as a product (2 ~ 21)g(2), where q(2) is a polynomial of degree n— 1 (Gee the exercises.) The induction hypothesis allows one to factor q(z) as a product of linear factors, and this yields the factorization of p(z). Thus the fundamental theorem of algebra is equivalent to the statement that every complex polynomial of degree n > 1 has a zero. Example. The polynomial p(z) = 2? +1 with real coefficients cannot be factored as a product of linear polynomials with real coefficients, since it does not have any real roots. However, the complex polynomial p(z) = 2? +1 has the factorization P41 = (2-248), corresponding to the two complex roots +i of 2? +1 Exercises for I.1 1. Identify and sketch the set of points satisfying: (@) |z-1-d)=1 (f) 0 0 @k-u< bl () [2a + fe +a <2 2. Verify from the definitions each of the identities (a) 7Fw = 2+ w@, (b) BU = 20, (c) |2| = [2], (d) |2/? = 22. Draw sketches to illustrate (a) and (c) 2. Polar Representation 5 3 10, nL Show that the equation |2|?~2 Re(a2)+ al? centered at a with radius p. Show that |2| < |Rez|+|Imz|, and sketch the set of points for which equality holds. ? represents a circle Show that |Re | < |z| and |Im2| < |2|. Show that lz +wl? = [2/7 +|wi? +2 Re(za. Use this to prove the triangle inequality |z + w| < lz] + [wl For fixed a € €, show that lz ~ al/|l — a2] = 1 if |2| 1a 40. = 1 and Fix p > 0,9 #1, and fix 9.21 € C. Show thatthe st of = satisfying lz ~ 20] = plz — 2] is a circle. Sketch it for p = 3 and p= 2 =O and 2; = 1. What happens when p = 1? Let p(z) be a polynomial of degree n > 1 and let zo € C. Show that, there is a polynomial hz) of degree n 1 such that p(z) = (=~ 20)h(2) +p(29). In particular, if p29) = 0, then p(z) = (2 —29)h(2) |. Find the polynomial h(z) in the preceding exercise for the following choices of p(z) and 20: (a) p(z) and 2 = —1, (c) p(2) Let q(z) be a polynomial of degree m > 1. Show that any polyno- ‘ial p(z) can be expressed in the form A(z) = h(z)a(z) +r(z), where h(z) and r(z) are polynomials and the degree of the remain- der r(2) is strictly less than m. Hint. Proceed by induction on the degree of p(z). The resulting method is called the division algorithm, Find the polynomials A(z) and r(z) in the preceding exercise for p(2) = 2" and 9(2) = 2-1. ? and 29 = i, (b) p(2) Slt ete? e+ 2" and 29 Paste -1. 2. Polar Representation Any point (x,y) # (0,0) in the plane can be described by polar coordi- nates r and 0, where r= /2? +y? and 0 is the angle subtended by (¢,y) and the z-axis. ‘The angle @ is determined only up to adding an ntegral ‘multiple of 2n. The Cartesian coordinates x,y are recovered from the polar coordinates 7,9 by paren {porane6 1. The Complex Plane and Elementary Functions If we write the polar representation in complex notation, we obtain ea) z (cos +isind). Here r= |2| is the modulus of z. We define the argument of z to be the angle 8, and we write ztiy @ = age. ‘Thus arg is a multivalued function, defined for > 4 0. The principal value of arg 2, denoted by Arg, is specified rather arbitrarly to be the value of 8 that satisfies —x <@ 1 Hint. First consider the case b > 0, and then reduce the general ‘case to this case by a rotation. 4, For which n is # an nth root of unity? 5. For n> 1, show that (a) Lt 2botbertah—= (1-2-2), #1, 1, sin(n + $0 2+ psing/2 6. Fix n> 1. Show that the nth roots of unity wo... 4-1 satisfy: (a) ( ~ wo)(2 =e) (wnat) = 2" — 1, (b) wy tos tent = 9, (¢) @o--wWy—a = (5 Se 0 1sksn @ Sot ={ no k=n 7. Fix R> Land n > 1,m>0. Show that (b) 14 608 8 + 608 29 + +--+ 008 nd a RO Sketch the set where equality holds. Hint. See (1.1) 3. Stereographic Projection n 8. Show that cos2# = cos? — sin®@ and sin2@ = 2cossin@ using de Moivre’s formulae. Find formulae for cos 49 and sin 40 in terms of cos and sin 8. 3. Stereographic Projection ‘The extended complex plane is the complex plane together with the point at infinity. We denote the extended complex plane by C*, so that C* = CU {oo}. One way to visualize the extended complex plane is through stereographic projection. This is a function, or map, from the unit sphere in three-dimensional Euclidean space R° to the extended complex plane, which is defined as follows. If P = (X,Y, Z) is any point of the unit sphere other than the north pole V = (0,0,1), we draw a straight line through N and P, and we define the stereographic projection of P to be the point 2 = 2+ iy ~ (x,4,0) where the straight line meets the coordinate plane Z = 0. The stereographic projection of the north pole V is defined to be co, the point at infinity. An explicit formula for the stereographie projection is derived as follows, We represent the line through P and N parametrically by N+ t(P — N), —o0 <¢ 1/2 of C. ‘Suppose (x,y,0) is the spherical prejection of (X,¥,Z). Show that the product of the distances from th> north pole N to (X,¥,Z) and from N to (c,y,0) is 2. What is the situation when (X,Y, Z) lies on the equator of the sphere? |. We define the chordal distance d{(z,w) between two points 2,w € C* to be the length of the straight line segment joining the points P and Q on the unit sphere whose stereographic projections are = and w, respectively. (a) Show that the chordal distance is a metric, that is, it is symmetric, d(z,w) = d(w,2); it satisfies the triangle inequality d(z,w) < d(z,¢) + d(C, 2); and d(z,w) = 0 if and only if 2 =w. (b) Show that the chordal distance from = to w is given by ae — wh Jit ee t+ poe” (c) What is d(z,00)? Remark, The expression for d(z,w) shows that infinitesimal are length corresponding to the chordal metric is given by d{z,w) = zwee. 2ds dole) = yap where ds = |d2| is the usual Euclidean infinitesimal arc length. ‘The infinitesimal are length do(2) determines another metric, the spherical metric (2,1), on the extended complex plane. See Section IX.3. 4. The Square and Square Root Functions 15 7. Consider the sphere of radius } in (X,Y, Z)-space, resting on the (2.¥.0}plane with south pole tthe origin (00.0) and nore poe bt (0,0,1)- We define a streographiepreeetion ofthe sphere cate the complex plane ns before, that caresponding pole (XYZ) and 2 ~ (29,0) i onthe same Ine through the orth pole. Find the equation for + =-2-4iy terms of XY. and the equations for X1Y-2 in terms of =, What i the coreaponding formals for the slr dace? pte, tics te stunner (@-1)"=4 4. The Square and Square Root Functions Real-valued functions of a real variable can be visualized by graphing them in the plane B®. The graph of a complex-valued function f(z) of a complex variable 2 requires four (real) dimensions. Thus some techniques other than graphing in R* must be developed for visualizing and understanding functions of a complex variable. One technique isto graph the modulus of the function |f(2)| as a surface in three-dimensional space R®. Another is to graph separately the real and imaginary parts of f(2) in B®. We describe a different technique for gaining insight into the behavior of the function f(z). We create two planes, a 2-plane for the domain space and a w-plane for the range space. We then view f(z) as a mapping from the z-plane to the w-plane, and we analyze how various geometric configurations in the z-plane are mapped by w = f(z) to the w-plane. Which geometric configurations in the 2-plane to consider depends very much on the specific function f(2) To ilustrate how this method works, wwe consider the simplest nontrivial function, the square function w plane plane From the polar decomposition w = 2? = re, we have (4) lw = LEP, (42) argw = 2argz, Equation (4.1) shows that the circle |2| = ro in the z-plane is mapped to the circle |w| = 13 in the w-plane. As z = roe"? moves around the in the positive direction at constant angular velocity, the image w moves around the image circle, in the same direction but at double the16 1 The Complex Plane and Elementary Functions ‘plane plane angular velocity. As = makes one complete loop, the image w makes two ‘complete loops around the image circle. Equation (4.2) shows that a ray {argz = 6o} issuing from the origin in the z-plane is mapped to a ray in the u-plane of twice the angle. As = traverses the ray from the origin to co at constant speed, the value w traverses the image ray from 0 to co, starting slowly and increasing its speed. The positive real axis in the 2-plane, which is a ray with angle O, is mapped to the positive real axis in the w-plane by the usual rule 2 + 2? ‘As z traverses the ray {argz = 7/4}, the image w traverses the positive imaginary axis, and as z traverses the positive imaginary axis, the image w traverses the negative axis. As the rays in the 2-plane sweep out the first quadrant, the image rays in the w-plane sweep out the upper half-plane, and as the rays in the 2-plane sweep out the second quadrant, the image rays in the w-plane sweep out the lower half-plane. Eventually, we reach the ray along the negative real axis in the 2-plane, which is mapped again to the positive real axis in the w-plane, and as we continue, the behavior is repeated in the lower half of the 2-plane. Now we tum to the problem of finding an inverse function for w = 2 Every point w 0 is hit by exactly two values of =, the two square roots £10. In order to define an inverse function, we must restrict the domain in the z-plane so that values w are hit by only one 2. There are many ways ‘of doing this, and we proceed somewhat arbitrarily as follows. Note that as rays sweep out the open right half of the z-plane, with the angle of the ray increasing from ~7/2 to 7/2, the image rays under w= z? sweep out the entire w-plane except for the negative axis, with the angle of the ray increasing from —z to x. This leads us to draw a slit, or branch cut, in the w-plane along the negative axis from —oo to 0, and to define the inverse function on the slit plane C\(—o,0). Every value w in the slit plane is the image of exactly two 2-values, one in the (open) right half-plane {Rez > 0}. the other in the left half-plane {Rez <0}. Thus there are two possibilities for defining a (continuous) inverse function on the slit plane. We refer to each determination of the inverse function as a branch of the inverse. One branch fx(w) of the inverse function is defined by declaring that f,(w) is the value = such that Re > 0 and 2? = w. Then fi(w) maps 2 i ? | 4. The Square an Saunre Root Functions w& the slit plane C\(—co, 0} onto the right half-plane {Re= > 0}, and it forms an inverse for z* on that half-plane. To specify f,(w) explicitly, express *? where ¢ lies in the range —x <<, and then filw) = ype", wap, meyer ‘The function fy(w) is clled the principal branch of /. It is expressed in terms of the principal branch of the argument function as, filw) = w[M7KAE/2, wy € C\(-00,0) [As w approaches a point ~r on the negative real axis (~20,0) from above, the values f(t) approach the value iy on the positive imaginary axis. We express this by writing f,(—r + i0) = iy, Similarly, as w approaches —r from below, the values fi(w) approach the value ~iy on the negative imaginary axis, that is, fi(—r ~ i0) = ~iyF. The braneh cut (—00,0] in the w-plane can be regarded as having two edges, and the function f,(w) extends continuously to each edge. The top edge, labeled * +" in the figure, is mapped to the positive imaginary axis by fi(w), and the bottom edge, labeled “-", is mapped to the negative imaginary axis by f(w) =) = vi (principal branch) ee We use the other value of ii to define a second branch fa(w) of the inverse function v/s. For this we use a second copy of the w-plane, as in the figure. On this sheet the second branch of /i is defined by fa(w) =fi(w). This branch maps the slit plane onto the left half-plane {Rez < 0}. As w approaches a point —r on the negative axis (~c0,0) from above, the values fa(w) approach the value ~iy on the negative imaginary axis,18 1 The Complex Plane and Elementary Functions and as w approaches —r from below, the values fa(w) approach the value iyF on the positive imaginary axis. Again we think of the elit as having two edges, though on this sheet the top edge is mapped to the negative imaginary axis and the bottom edge is mapped to the positive imaginary axis. Further, we have Sul—r +10) = ivr = fal — 10), fi(—r — 10) = ~ivF = fa(—r + 10), This leads us to the idea of constructing a surface to represent the inverse function by gluing together the edges where the functions f,(w) and fi(w) coincide. We glue the top edge ofthe branch cut en the sheet corresponding to fi(w) to the bottom edge of the branch cut on the sheet corresponding to fa(w), and similarly for the remaining two edges, to obtain a two-sheeted surface. Since the values of f,(w) and fo(w) coincide on the edges we have slued together, they determine a function f(w) defined on the two-sheeted surface, with values in the z-plane that move continuously with w. Since each sheet of the surface is a copy of the slit w-plane, we may think of the sheets as “lying over" the w-plane. Each w € C\{0} corresponds to exactly two points on the surface. The function f(w) on the surface represents the multivalued function /@ in the sense that the values of Jz are precisely the values assumed by f(w) at the points of the surface lying ‘The surface we have constructed is called the Riemann surface of y/w. The surface is essentially a sphere with two punctures corresponding to O and 0. One way to see this is to note that the function f(w) maps the surface one-to-one onto the z-plane punctured at 0. Another way to see this is to deform the surface by prying open each sheet at the slit, opening it to ‘a hemisphere, and then joining the two hemispheres along the slit edges to form a sphere with two punctures corresponding to the endpoints 0 and 00 of the slits Exercises for L4 1. Sketch each curve in the 2-plane, and sketch its image under w ()lz-=1 0 (y= (e) =z?-1, r>0 (b)z=1 @y=2+1 (y=Va 40 2. Sketch the image of each curve in the preceding problem under the principal branch of w = v2, and also sketch, on the same grid but 5. The Exponential Function 19 in a different color, the image of each curve under the other branch of Vz. 3. (a) Give a brief description of the function 2+ w = 2°, considered fas a mapping from the z-plane to the w-plane. (Deseribe what happens to w as z traverses a ray emanating from the origin, and as z traverses a circle centered at the origin.) (b) Make branch cuts and define explicitly three branches of the inverse mapping. (c) Describe the construction of the Riemann surface of 21/6 4. Deseribe how to construct the Riemann surfaces for the following functions: (a) w = ='/%, (b) w = VZ=1, (c) w = (2-1)"/5, Remark. To describe the Riemann surface of a multivalued function, begin ‘with one sheet for each branch of the function, make branch cuts ‘0 that the branches are defined continuously on each sheet, and identify each edge of a cut on one sheet to another appropriate edge 0 that the function values match up continuously. 5. The Exponential Function We extend the definition of the exponential function to all complex nurm- bers 2 by defining ef = cosy tietsiny, 2=2tiye. Since e' = cosy + isiny, this is equivalent to iy. ‘This identity is simply the polar representation of e*, (6) lel = (52) arge = y. If = is real (y = 0), the definition of e* agroes with the usual exponential funetion e. If 2 Is Imaginary (x = 0), the definition agrees with the definition of e" given in Section 2.2 2 The Complex Plane and Elementary Functions A fundamental property of the exponential function is that it is periodic. ‘The complex number A is a period of the function f(z) if f(z-+) = f(z) for all z for which f(z) and f(z +A) are defined. ‘The function (2) is periodic if it has a nonzero period. Since sinz and cosy are periodic functions with period 2z, the funetion e* is periodie with period 2: zeC, In fact, 2rik is a period of e* for any integer k Another fundamental property of the exponential function is the addi tion formula (63) ete = ee", zwel. To check this, let 2 = r+ iy and w= u+ iv. Then where we have used the addition formulae for e* and e, From the addition formula (5.3) we have e*e~* = e? = 1. Consequently, the inverse of e* is e*, je =e, re, ‘To understand the exporential function better, we view w — ef as a mapping from the z-plane to the w-plane. If we restrict the exponential function to the real line #, we obtain the usual exponential function 2 -= €, ~00 < z < 00, which maps the real line R to the positive real axis (0.00). The equation (5.2) shows that an arbitrary horizontal line 2+ iyo, =00 < 2 < 00, is mapped to the curve ee", ~o0 <1 < 00, which is 0 ray issuing from the origin at angle yo. If we move the horizontal line up, the angle subtended by the ray increases, and the image ray is rotated in the pesitive (counterclockwise) direction, As we move the horizontal line upwards from the a-axis at yp = 0 to height yo — 7/2, the image rays sicep out the fst quadrant in tho w-plane. ‘The horizontal line at height tm = 7/2 is mapped to the positive imaginary axis, the horizontal line of height yo = x is mapped to the negative real axis, and when we reach the horizontal line of height yp = 2x, the image rays have swept out the full tw-plane and returned to the postive real axis. The picture then repeats itself periodically. Bach point in the w-plane, except w =O, is ht infinitely often, by a sequence of -values spaced at equal intervals of length 27 along a vertical line. While the images of horizontal lines are rays issuing from the origi, the images of vertical lines are circles centered at the origin. ‘The equation (5.1) shows that the image of the vertical line 2» + iy, ~00 < y < 00, isa circle in the w-plane of radius e8. As = traverses the vertical lin, the value 1 ‘wraps infinitely often around the circle, completing one turn each time y= Imz increases by 2 6, The Logarithm Function a horzoea ines) > rays etcl lines —> cices Exercises for 1.5 1. Calculate and plot e* for the following points 2: (a) 0 (©) eG-1/3_— (@) ifm, m= 12,3, (b)i+1 (4) 37m (me -1), 2,3, 2, Sketch each of the following figures and its image under the expo- nential map w = e*. Indicate the images of horizontal and vertical lines in your sketch, {a) the vertical strip 0 < Rez <1, (b) the horizontal strip 51/3 0 and e™™™ = 1 ‘On the other hand, suppose that w =u + iv is an arbitrary complex number such that e = =. ‘Then the polar representation of z is z = re”, where r= [2] = e%. Thus u = log|2|. and v is a value of arg, so that Arg 2 +2rm for some integer m. Recall that the principal value Argz of arg is the value @ satisfying 0, (b) the half-disk 2] <1, Rez >0, (c) the unit cirele |2|=1, (@) the slit annulus. yé < |2| < {e) the horizontal line y =e, (6) the vertical ire x =e. 3. Define explicitly 4 continuous branch of log 2 in the complex plane slit along the negative imaginary axis, C\(0, —ioo). 4, How would you rake a branch cut to define a single-valued branch of the function log(2 +i — 1)? How about log(z ~ 29)? 7. Power Functions end Phase Factors Let a be an arbitrary complex number. We define the power function 2% to be the multivalued function w= eles, 240, ‘Thus the values of 2* are given by = etlbelel+é Arges 2nim] ea toneg2riom m= 0,41, 42, ‘The various values of 2% are obtained from the principal value e*/°#* by multiplying by the integral powers (e2"!4)™ of e2"'2. If a is itself an integer, then e?"'* = 1, and the function 2° is single-valued, the usual power function. If a = 1/n for some integer n, then the integral powers 27 /" of e211” are exactly the nth roots of unity, and the values of =!" fre the m nth roots af » fisenssed earlier (Seetian 9) Example. The values of i are given by eiloet go Argi-2em /—-2em, m= 0,1, 42, The values form a two-tailed sequence of positive real numbers, accumu lating at 0 and at +90. Similarly, the values of i~* are given by Argi-i)-2k _ gw/%en2k, = 0,41, 42,. efest-9 7. Power Functions and Phase Factors % Danger! If we multiply the values of # by those of it, we obtain infinitely many values e2", co 1. Draw branch cuts so that the function ean be defined continuously off the branch cuts. Describe the Riemann surface of the function, Consider the branch of the function /=(2— I)(= + 1)° that is por itive at z= 2. Draw branch cuts so that this branch of the function ‘can be defined continuously off the branch cuts. Describe the Rie- ‘mann surface of the function. To what value at 2 = 2 does this branch return if it is continued continuously once counterclockwise around the circle {|2| = 2}? Consider the branch of the funetion /=(2 — I) > 1)(2—1) that is positive at z= 2. Draw branch cuts so that this branch of the function can be defined continuously off the branch cuts. Describe the Riemann surface of the function. To what value at z = 2 does this branch return if itis continued continuously once counterclock- wise around the circle {I=| = 2}? Find the branch points of {25-1 and describe the Riemann surface of the function. 8. Trigonometric and Hyperbolic Functions 2 8, Trigonometric and Hyperbolic Functions If we solve the equations 9 = cos0 + ising, = cos — ising for cos @ and sin 8, we obtain sind : We ‘This motivates us to extend the definition of cos and sin2 to complex numbers 2 by dttett cosz = S48", zee, 4 e nz = SE" cee. 2 © “This definition agrees with the usual definition when = is real. Evidently, cos is an even function, cos(=2) = cons, FEC, while sin z is an odd function, sin(—2) = sine, 2 €C. [As functions of a complex variable, cos and sin z are periodic, with pe- lod 2m, cos(z+2n) = cosz, 2 C, sin: +2n) = sinz, 2€C. After some algebraic manipulation, one checks (Exercise 1) that the addi- tion formulae for cos > and sin remain valid, cos(z + w) = coszeosu—sinzsinw, 2,weC, sinzeosw +coszsinw, — z,w €C. sin(z +) If we substitute w = —= in the addition formula for cosine, we obtain the familiar identity costztsin?z = 1, 2€C. ‘We shall see, in fact, that any reasonable identity that holds for analytic functions of a real variable, such as cos and sin-z, also holds when the: functions are extended to be funetions of a complex variable. This will be a20 1 The Complex Plane and Elementary Functions special ca ofthe principle of permanence of functional equations, proved in Chapter V. “s ‘The hyperbli functions cosh ~ (e*+6°*)/2 and ih, tre alo extended othe complex plane in the obvious way, by (ee #)/2 ze, zeC. period 2xi, cosh(z +2mi) = coshz, 2 EC, sinh(z+2ni) = sinhz, ze. Evidently, cosh 2 is an even function and sinh > is an odd function. There are addition formulae for cosh = and sinh 2, derived easily from the addition formulae for cos and sin z (Exercise 1). ‘When viewed as functions of a complex variable, the trigonometric and the hyperbolic functions exhibit a close relationship. They are obtained from each other by rotating the domain space by 7/2, cosh(iz) = cosz, —_cas(iz) = cosh 2, sinh(iz) = isinz, —sin(iz) = ésinh 2. If we use these equations and the addition formula, sin(x + iy) = sinxcos(iy) + coszsin(iy), we obtain the Cartesian representation for sin z, sinz = sinzcoshy+icosrsinhy, 2 =2+iyeC. Thus [sin 2? = sin®.zcosh? y + cos? x sinh? y Using cos? x + sin?x = 1 and cosh? y = +sinh?y, we obtain [sin 2|? = sin? + sinh? y, From this formula it is clear where the zeros of sin z are located; sin z = 0 only when sinz = 0 and sinhy = 0, and this oceurs only on the real axis y= 0, at the usual zeros 0,7, +2z,... of sinzz. Similarly, the only zeros Of cos 2 are the usual zeros of cos on the real axis (Exercise 2). Other trigonometric and hyperbolic functions are defined by the usual formulae, such as tanz = ‘Thus tan z and tanh 2 are odd functions, and tanh(éz) = étan z. Exercises a1 ‘The inverse trigonometric functions are multivalued functions, which can be expressed in terms of the logarithm function. Suppose w = sin” z, that ‘Then ce” —2izet® —1 = 0, This is a quadratic equation in e", which can be solved by the usual quadratic formula. ‘The solutions are given by = ~tlog (ie Vi) ‘This identity is to be understood as a set identity, in the sense that w sat- isfes sin w — = ifand only if wis one of the values of —ilog (iz + VT ="). ‘To obtain a genuine function, we must restrict the domain and specify the branch. One way to do this is to draw two branch cuts, from —co to —1 and from +1 to +00 along the real axis, and to specify the branch of v= 2? that is positive on the interval (1,1). With this branch of ¥I= 2%, we obtain a continuous branch ~iLog (is + Vi 22) of sin" = Exercises for L8 1. Establish the following addition formulae: (a) cos(2 + w) = cos cos w — sin zsin w, (b) sin(z + w) = sin zcosw + cos zsinw, (©) cosh(z + w) = cosh z cosh w + sinh = sinh w, (@) sinh(2 +.w) = sinh z cosh w + cosh zsinhw, 2. Show that |cosz|? = cos? + sinh y, where z = 2 + iy. Find all zeros and periods of cos z. 3. Find all zeros and periods of cosh > and sinh z. tants = 4 tog (LAH an = 38 ae)” where both sides of the identity are to be interpreted as subsets of the complex plane. In other words, show that tanw =z if and only if 2iw is one of the values of the logarithm featured on the right. 4. Show that Let $ denote the two slits along the imaginary axis in the complex plane, one running from i to +i00, the other from —i to ~ice. Show that (1+ i2)/(1~ iz) lies on the negative real axis (—co,0] if and2 ont €8, Show tin the pineal bath thats = tag (1 ae(i) mapa the fe lane C\$ one-to-one onto the horizontal strip {—= < imw <7 Describe the Rian sie fran"? *. Show tnt (= w £ ViPP=T, Show tha ctw = -tteg [vt Vie=T] where both sides of the identity are to be interpreted as subsets of the complex plane, Set w = cos and ¢ = Show tn he vo sip [fe] <2 i mapped by th an tee Sle eee a a end wie (oat) etl Si he verse function is the branch of sin~'z = —iLog (iz +V1—2 First show that the function 1 — 2? on the slit plane omits the nega- ase star eet a eee II Analytic Functions In this chapter we take up the complex differential calculus. After review- ing some basic analysis in Section 1, we introduce complex derivatives and analytic functions in Section 2 and we show that the rules for complex differ- entiation are the same as the usual rules for differentiation. In Section 3 we characterize analytic functions in terms of the Cauchy-Riemann equations. In Sections 4 and 5 we give several applications of the Cauchy-Riemann, ‘equations, to inverses of analytic functions and to harmonic functions. In Section 6 we discuss conformality, which is a direct consequence of complex dlifferentiability. We close in Section 7 with a discussion of fractional linear transformations, which form an important class of analytic functions. 1. Review of Basic Analysis, We begin by reviewing the background material in analysis that will (even- tually) be called upon, and we say something about the language of formal mathematics. For the most part, we will not phrase our arguments com- pletely formally, though any bilingual person will be able to translate easily to the language of formal mathematics in such a way that our development becomes completely rigorous. Since the complex derivative is defined as a limit, we require some back- ground material on limits and continuity. To be able to define and work with analytic funetions, we also require some basic topological concepts, including open and closed sets, and domains. The confident reader may pass directly to the definitions of complex derivative and analytic function in the next section, and refer back to the material inthis section only when needed. We begin with the notion of a convergent sequence. For this we have two definitions. Informal Definition. A sequence {sq} converges to s if the sequence eventually lies in any disk centered at & 33uM II Analytic Functions The language of formal mathematics serves 10 quantify this statement and make it precise. The “small disk” is traditionally given radius e > 0. ‘That “s, lies in the disk” means that |sq — s| 1 such that the event occurs for n > N. ‘Thus the trarslation of the definition of convergent sequence to the language of formal mathematics is as follows, [ben the formed Formal Definition. A sequence of complex numbers {s,,} converges tos if for any € > 0, there is an integer N > 1 such that [Sq —s| <€ for all nen, If {5n} converges to 8, we write Sq —+ 5, or lim sy = s. Some examples of convergent sequences that appear frequently are 1 (1) dim 5-0 0 0 take NV to be an integer satisfying N > 1/e!/?. Then for n > N we have nP > NP > ie, and 1/n® < ¢._ To prove (1.2) formally, we would take NV to be an integer satisfying N > (loge)/(log|s|)- To prove (1.8) formally, we resort to a trick. Let tp = (ni — 1. We estimate t, from the binomial expansion n(n n= (14tq)" = 1+ nty + +> 2 th ‘This yields ty < 2/(n—1), Thus |tn| = | /7—1| <= whenever 2/(n—1) < ©, that is, for n > 142/e?, For the formal definition we can take N to be any integer satisfying N > 1+ 2/e? ‘We give some definitions and state some theorems, without proofs, that ‘we will be using. A sequence of complex numbers {sq} is said to be bounded if there is some number R > 0 such that |sq| < R for all n. In other words, the sequence is bounded if itis contained in some disk. Theorem. A convergent sequence is bounded. Further, if {sq} and {tn} are sequences of complex numbers such that sn —+ 8 and ty —t, then (a) 8 + th 8 +t, (b) ata st, (6) Sn/tn — s/t, provided that t #0. 1. Review of Basie Analysis 35 ‘Thus the limit of a sum is the sum of the limits, the limit of a product, is the product of the limits, and the limit of a quotient is the quotient of the limits, provided that the denominator is not 0. Example. We can use these rules to evaluate the Ii pression of the form it of @ rational ex- tim 3024 2n=1 nt Sn? —dn +8 5" ‘As a preliminary trick, we divide numerator and denominator by the lead- Ing power, and rewrite the expression as 3+ (2/n) — 1/n? B= Un) Bin Since 1/m +0 and 1/n? —+0, the sum and product statements show that the numerator converges to 3 and the denominator converges to 5. By the qiotient statement, the quotient then converges to § ‘The most useful criteria for convergence of sequences of real and complex numbers are gathered in the next several theorems. ‘The first criterion is sometimes called the in-between theorem. ‘Theorem. If rq < sq Sty, and ifr,» L and ty —» L, then 5, — E. A sequence of real numbers {sq} is said to be monotone increasing if sa41 > mq for all n, monotone decreasing if sa41 S, then 5, > t for only finitely many indices m, while if 1 t for infinitely many indices n. It is easy to see that any such S is unique. The existence of such an S can be deduced from the preceding theorem. In fact, the existence is equivalent: to the completeness axiom of the real numbers.36 TI Analytic Functions ‘A “lower limit” of the sequence {sq}, denoted by liminf sq, is defined similarly. It satisfies liminf s, =~ limsup(—sn). ‘The sequence {s,,} converges if and only if its lim sup and lim inf are finite and equal. Example. The sequence {(-1)"}2o = {+1, converge. Its upper and lower limits are limsup (~1)" = +1, limit (—1)" so Iy...} does not For complex sequences, the following simple criterion is used very often ‘Theorem. A sequence {sx} of complex numbers converges if and only if the corresponding sequences of real and imaginary parts of the s,'s con- verge. We define a sequence of complex numbers (s,} to be a Cauchy se- quence if the differences — sm tend to 0 a8 m and m tend to oo. In the language of formal mathematics, this means that for any © > 0, there exists V > 1 such that |&n ~ Sm| << if m,n > N. The following theorem is an equivalent form of the completeness axiom. Itis important because it provides a means of determining whether a sequence is convergent without producing explicitly the limit of the sequence. ‘Theorem. A sequence of complex numbers coaverges if and only if itis a Cauchy sequence. We say that a complex-valued function f(z) has limit L as 2 tends to 2 if the values f(2) are near L whenever 2 is near zp, 2 # z. The formal definition is that f(z) has limit L as z tends to zo if for any € > 0, there is 6 > 0 such that [f(2) — Z| < © whenever = in the domain of f(2) satisfies 0 < [2 ~ zo] <6. In this case we write an 10) = 1 cx (0) Lae + tis imply unerdod that there ae pints in the domain of f(z) that are arbitrarily close to 29 and different from zo. ‘The definition can be rephrased in terms of convergent sequences. Lemma. The complex-valued function f(2) hes limit L as 2 —» zy if and only if f(z) — L for any sequence {2} in the domain of f(z) such that sn #20 and 29 — Zo From the theorem on limits, we obtain easily the following. 1, Review of Basie Analysis 37 Theorem. If a function has a limit at 29, then the function is bounded near zo. Further, if f(2) + L and (2) > M as 2 — 2, then as 2 + 2 we have (a) f(2) + 92) + b+ M, (b) f(z)g(z) > LM, (c) f(z)/g(z) + L/M, provided that M #0. We say that f(z) is continuous at 29 if f(z) > f(z0) as 2 + 2. A continuous function is a function that is continuous at each point of its domain. The preceding theorem shows that sums and products of, ‘continuous functions are continuous, and so are quotients, provided that the denominator is not zero. Further, the composition of continuous functions Js continuous. Example. Any constant function is continuous. The coordinate function f(2) = 7 is continuous. ‘Thus any polynomial function p(2) = 492" + “+412 + a9 is continuous. Any rational function p(2)/4(2) is continuous wherever the denominator q(2) is not zero. ‘A useful strategy for showing that (2) is continuous at 2o is to obtain an estimate of the form |f(z) ~ f(zo)| < Olz ~ 20| for z near zo. This guarantees that |f(2) — f(20)| <€ whenever |2~ z9| < €/C, so that we can take 5 = 0, So that itis an open set. Other examples of open sets38 TT Analytic Functions described by strict inequalities are the open sector {8 < arg z < 61}, the ‘open horizontal strip {1 < Im < 11}, the open annulus {r < |s| < #}, and the punctured disk {0 < |2| c} or {f(2) < ¢}, where f(2) is a continuous real-valued function, are closed. Thus for instance the closed upper half- plane, consisting of points = such that Re(2) > 0, is a closed set. The boundary of a set E consists of points 2 such that every disk centered at z contains both points in E and points not in E. Thus a set is closed if it contains its boundary, and a set is open if it does not include any of its boundary points. For example, the boundary of the closed disk {\z ~ zol PPT (b) lim p= (@) im =0, ec 2. For which values of = is the sequence {2"}%2., bounded? For which values of 2 does the sequence converge to 7 3. Show that {n"2"} converges only for 5 Nt! 4 Show that Jit NEW AI40 10. uM. 2 II Analytic Functions ‘Show that the sequence 1 1 1 eae —logn, n> belt tat +2 log, 21 is decreasing, while the sequence a, = by —1/n is increasing. Show ‘that the sequences both converge to the same limit +. Show that 4<7< 2. Remark. The limit of the sequence is called Buler’s constant. It is not known whether Euler's constant. is a rational number or an irrational number. For a complex number a, we define the binomial coefficient “a ‘choose n” by (s)=1 (2) a(a~1) Ba oneny aoa Show he flloing (2) For ach a, the sequence () is bounded 0 (0) itantenit tea 1 (0) Har #0,1,2,..., then (31) /( : (0) tea < ayo¢ 1 sie (,",) ()]ecanse (9 tea > 1 and ont anise te |(,",)] < (°)| for n large. Define 9 = 0, and define by induction an, = 22 +4 for n > 0. Show that 2 —}. Hint. Show that the sequence is bounded and ‘monotone, and that any limit satisfies 2 =2? + 4 Show that if sq —+ 5, then |¢q ~ sq-1| 0 Plot ench sequence and determine its timint and li sup. (8) b= 1424 (-1)" (0) 55 = (-nJ* (ER fixed) At what points are the following functions continuous? Justify your answer. (a) 2, (b) 2/l2l (€) =/lzhs (@) 2°/leF* [At what points does the function Arg have a limit? Where is Argz continuous? Justify your answer. Let A(z) be the restriction of the function Arg2 to the lower half- plane {Im <0}. At what points does f(z) have a limit? What is the limit? (©) & = sin(zn/4) (@) b= 2" Exercises a 13. For which complex values of a does the principal value of 2% have a limit as = tends to 0? Justify your answer. 14, Let h(t) be a continuous complex-valued function on the unit inver- val (0, 1], and consider H(2) f "HO a Where is H(z) defined? Where is H(2) continuous? Justify yous answer, Hint. Use the fat that i [f(t)—alf)| <= for 0- 1}, (c) the exterior of the closed unit disk in the plane, {|z| > 1}, (d) the plane with the open unit interval removed, C\(0,1),(e) the plane with the closed unit interval removed, C\(0, 1), () the semidisk {Jz| < 1, Im(z) > 0}, (e) the complex plane C. 16. Show that the slit plane C\(~oo,0) is star-shaped but not convex. Show that the slit plane C\|~1,1] is not star-shaped. Show that a punctured disk is not star-shaped, 17. Show that a set is convex if and only ifit is star-shaped with respect, to each of its points. 18. Show that the following are equivalent for an open subset U' of the ‘complex plane. (a) Any two points of U can be joined by a path consisting of straight line segments parallel to the coordinate axes. (b) Any continuously differentiable function h(z,y) on U such that Wh = 0 is constant (c) IfV and W are disjoint open subsets of U such that U = VUW, then either U = V or U = W. Remark. In the context of topological spaces, this latter property is taken as the definition of connectedness. 19. Give a proof of the fundamental theorem of algebra along the fol- lowing lines. Show that if p(z) is a nonconstant polynomial, then [p(z)] attains its minimum at some point 2 € C. Assume that the minimum is attained at 2 = 0, and that p(z) = 1442" 4 +=, where m > 1 and a #0. Contradict the minimality by showing that |P(ce)| <1 for an appropriate choice of 8p,2 II Analytic Funetions 2. Analytic Functions If the development inthis section has a familia ring, it should. ‘The basic definitions and rules for the complex derivative are exactly the same as those forthe usual derivative in elementary calculus. The only difference is that multiplication and division are now performed with complex numbers instead of real numbers. "A complex-valued fnetion f(z) is differentiable at zo ifthe difference quotients oy $0) = fe) have limit as = —+ 2p. The limit i denoted by (20), or by (2), and we refer to it as the complex derivative of f(z) at 2p. Thus p Le) f(z0) (22) Ze = se) Example. A constant function f(z) = c has derivative f"(zo) = 0 at any ppoint 29. In this case the difference quotients (2.1) are all zero, so that the limit is also 0. It is often useful to write the difference quotient (2.1) in the form flz0 + Bz) ~ flz0) ‘Az . so that 2 — 2» is replaced by Az. The formula for the complex derivative becomes ea Mea) = fi, Heat Ae)— fla) Occasionally we use 2 instead of zy in the expression (2.4) (2.3) Example. The power function f(z) = 2" has derivative f'(z) = m2™-! In this case the binomial expansion (er Aeym = es mam tae + MM Damage)? 4 oo + (Be) vields fle + Az) fle) = mnt = PAs one + (Az)™ which has limit m2™-" as Az — 0. ‘Example. The function f(z) = 7 is not differentiable at any point 2. In this case the difference quotient (2.3) becomes (@FBa) -2) /Az = Beyaz 2. Analytic Functions 43 If Az = ¢ is real, then this difference quotient is equal to 1, whereas if Az = ie ie imaginary, then the difference quotient is equal to —1. Thus the difference quotients do not have a limit as Az — 0. ‘The various properties of the complex derivative can be developed in exactly the same way as the properties of the usual derivative. ‘Theorem. If f(2) is differentiable at 29, then f(2) is continuous at zo ‘This follows from the sum and product rules for limits. We write J) = a+ (4 fy (=f), ~ #0) Since the difference quotient tends to f"(co) as z —+ zo, and z ~ 29 tends to 0 as z — 2, the preduct on the right tends to 0, and consequently, Se) = flea) a8 2 20 ‘The complex derivative satisfies the usual rules for differentiating sums, products, and quotients. The rules are (2.5) (f(z) = ef'(2). 26) F+aVe) = s@)+0'e), 7) (sayz) = slel(e) + Fee), 28) faye) = LOS) gs) xo. we Here we are assuming that f(2) and g() are differentiable at 2, and that € is any complex constant. The conclusion is that ef(z), f(z) + 9(2) F(2)a(2), and, provided that g(z) # 0, also f(z)/g(z) are all differentiable at z and satisfy the rules (2.5) to (2.8) listed abave. ‘The proofs depend on the theorems for limits of sums, products, and quotients. For instance, to ‘stablish the product rule (2.7) we begin with the usual trick and rewrite the difference quotient [(fg)(z + Az) ~ ({9)(2)]/Az for the product as gerade) , f+ A= 16), We now take a limit as Oa apply he uk fr Hits of su and products, and we obtain (2.7) "The identities (25) and (2,6 expres the fct that complex difeentiation isa linear operation. Note that (25) iva consequence of the product rule (2.7) and the fact thatthe derivative of «constant function cs 0 "octablsh the Mdetity (28), isis toetablih the spl identity (29) (say) = -9'2)/al2?, ale) #0, ‘and then to apply the product rule, (e+ Az)4 I Analytic Funetions Bxample. Any polynomial Plz) = nz" + aqaa2" t+ tanz bag has @ complex derivative, which is given by the usual formula, P(e) = mag2" E+ (n= Taqaae? eee ba, ‘This follows from the linearity rules (2.5) and (2.6), since 2” is dilferen- iable with derivative m2""!. Further, any rational funetion p(2)/q(2) is rentiable at all points = except for the (finitely many) zeros of a(2) ‘The chain rule is also valid for the complex derivative. We give a careful statement and proof. ‘Theorem (Chain Rule). Suppose that g(2) is differentiable at zo, and suppose that f(w) is differentiable at wy = 9(29). Then the composition (fog)(z) = flg(2)) is differentiable at 29 and (2.10) (f09)'(z0) = F'(a(20))9"(20). weWe Sef) A useful mnemonic device for remembering the chain rule is of dw dz dw a’ where we have written w = g(z). Danger! We regard f on the right-hand side as a function of w, and we regard f on the left-hand side as the function J{g(z)) of z, ‘Lhe mnemonic device can be justified by the proot, which involves multiplying and dividing by Aw. The proof goes as follows. We consider two cases. For the first case, we assume that 9/(2o) # 0. ‘Then 9(2) # 9(zo) for 0 < |2 ~ 20| < ¢, so we are justified in writing £(g(2)) = Flglz0)) _ Fig(z)) = F(9(20)) 9(2) = 9(20) => 20 g(2) ~ 9(20) zm Since g(z) is differentiable at zo, itis continuous at 20, that is, (2) — 9(zo) fas + 2. Consequently, Sla(2)) = flgl20)) (2) — 9(20) as + zo. Thus we can pass to the limit in (2.11), and we obtain (2.10) ean) = #920) 2 Analytic Functions 6 For the second case, we assume that g'(z9) =0. Since f(w) is differen- tiable at wp, the difference quotients (f(w) — f(wo)/(w ~ wo) are bounded near wo, say for some constant C and 0 < |w ~ wo] <¢. Hence | f(9(2)) ~ fa(zo))| < Clg(z) — 9(20)| for = near =p, and consequently, | fla(=)) ~ F(a(z0)) (2) ~ 920) {z= z9 2 7 [ 0 (b) zy + 3x7y—y? (d)e*—¥" cos(2ey) (F) x/(a? + 9°) 2. Show that if v is a harmonic conjugate for u, then —u is a harmonic conjugate for v. 3, Define u(z) = Im(1/22) for z #0, and set u(0) = 0. (a) Show that all partial derivatives of u with respect to x exist at all points of the plane C, as do all partial derivative of u with respect to y. Pu Pu (b) Show that 28 + =, (€) Show that 1 is not harmonic on . Pu (d) Show that: Bxdy 4. Show that if h(2) Ba complex-valued harmonic funetion (solution of Laplace's equation) such that 2h() i alo harmonic, then (2) is analytic does not exist at (0,0) 5. Show that Laplace's equation in polar coordinates is a 100, 1 0u a Or * aR 0.58 II Analytic Functions 6. Show using Laplace's equation in polar coordinates that log|2| is harmonic on the punctured plane C\{0}. 7. Show that log|z| has no conjugate harmonic function on the pune- tured plane C\{0}, though it does have a conjugate harmonie func- tion on the slit plane C\(—20, 0). 8. Show using Laplace's equation in polar coordinates that u(re'®) = Glogr is harmonic. Use the polar form of the Cauchy-Riemann equations (Exercise 3.8) to find a harmonic conjugate v for u. What is the analytic function u-+ iv? 6. Conformal Mappings Let (t) = z(t) + iy(t), 0 < t <1, be a smooth parameterized curve terminating at 29 = (0). We refer to ue) = (0) YO) = jim as the tangent vector to the curve 7 at zp. It is the complex representation of the usual tangent vector. We define the angle between two curves at 29 to be the angle between their tangent vectors at 29, = 2 +iy(0) Theorem. If7(t),0 1, the function 2" multiplies angles at the origin by n, ‘0 it is not conformal at z = 0. The following theorem shows that. 2" is conformal et any point 2 other than 0. ‘Theorem. If f(z) is analytic at 2) and f'(20) # 0, then f(z) is conformal at 29 Let 7o and 71 be two curves terminating at zo with nonzero tangents. By the preceding theorem, the tangents to the curves 9°79 and go" are obtained by multiplying the respective tangents to yo and 71 by #0) ‘Thus the arguments of both tangents are increased by the same angle, namely the argument of f"(zo). Consequently, the angle between them is preservedCy HL Analytic Functions There is a converse to this theorem, to the effect that conformal map- pings are analytic. Though the result is elementary, we postpone it to Section IV.8. (But see Exercise 9.) Example. The function w = 2? maps the right half plane {Re > 0} con- formally onto the slit plane C\(~0e, 0). For any fixed 4, 0 < Op < 7/2, it raps the sector {|arg2| < @} conformally onto the sector {J ars 2] < 260} of twice the aperture Example. Fix @, 0 < 0o < x. If 0 . Determine where f(2) is conformal and where it is not conformal. 2, Sketch the families of level curves of u and v for f(z) = Log= = u-+ iv. Relate your sketch to one of the figures in this section 3, Sketch the families of level curves of u and v for the functions f = u-biv given by (a) f(z) =e, (b) f(z) = e%*, where a is complex. Determine where f(z) is conformal and where it is not conformal.6 11 Analytic Punetions Find a conformal map of the horizontal strip {-A < Imz < A} onto the right half-plane {Rew > 0}. Hint. Recall the discussion of the exponential function, or refer to the preceding problem, Find a conformal map of the wedge {—B < argz < B} onto the right half-plane {Rew > 0}. Assume 0 < B 1, f(s) maps the circle {l2| =r) onto an ellipse, and that f(z) maps the cirele {|z| = 1/r} conto the same ellipse. Show that f(2) is one-to-one on the exteriot domain D = {\z| > 1}. Determine the image of D under f(2) Sketch the images under f(z) ofthe circles {|z| =r} for r > 1, and sketch also the images of the parts of the rays {argz = 3} lying in D. For the function f(z) = z+ 1/2 = u+ iv, sketch the families of level curves of u and v. Determine the images under f(z) of the top half of the unit disk, the bottom half of the unit disk, the part of ‘the upper half-plane outside the unit disk, and the part of the lower half-plane outside the unit disk. Hint. Start by locating the images of the curves where u = 0, where v = 0, and where v = 1. Note that the level curves are symmetric with respect to the real and imaginary axes, and they are invariant under the inversion 2 1/2 in the unit circle, Consider f(z) = z+ e'4/2, where 0 < a < x. Determine where J(2) is conformal and where it is not conformal. Sketch the images under f(2) of the unit circle {|z| = 1} and the intervals (~o0, ~1] and [+1, +00) on the real axis. Show that w = f(z) maps {lz| > 1} conformally onto the complement of a slit in the w-plane. ‘Sketch roughly the images of the segments of rays outside the unit circle {arg 2 — 6, |z| > 1} under f(z). At what angles do they meet the slit, and at what angles do they approach co? Let f = wiv be a continuously differentiable complex-valued func- tion on a domain D such that the Jacobian matrix of f does not vanish at any point of D. Show that if f maps orthogonal curves to orthogonal curves, then either f or f is analytic, with nonvanishing derivative. 7. Fractional Linear ‘Transformations 63 7. Fractional Linear Transformations A fractional linear transformation is a function of the form aztd ay w = fe) = SH, where a, 6, ¢, d are complex constants satisfying ad — be #0. Fractional Tinear transformations are also called M@bius transformations. Since (2) = ada be SO) = iar the condition ad ~ be 0 simply guarantees that f(2) is not constant. Tf we multiply each of the parameters a,b,c,d in (7-1) by the same nonzero constant, we obtain the same function. Thus different choices of the parameters may lead to the same fractional linear transformation ‘Example. A function of the form f(z) = az +0, where a # 0, is called an affine transformation. These are the fractional linear transformations of the form (7.1) with c= 0. Special cases are the translations z > z+b and the dilations 2 ++ az Example. The fractional linear transformation f(2) = 1/2 is called an inversion. It is convenient to regard a fractional linear transformation as a map from the extended complex plane C* = CU {co} to itself. If f(2) 1s affine, wwe define f(c0) = co. Otherwise, f(2) has the form (7.1) where ¢ # 0, and wwe define f(—d/c) = oo and Foo) = jm, fle) = ‘Thus translations and dilations map oo to co, while the inversion 2» 1/z interchanges 0 and oo. ‘The inverse of a fractional linear transformation is again a fractional linear transformation. ‘To see this, we solve (7.1) for =, to obtain ‘The condition on the coefficients is satisfied, since (—d)(—a)--be = ad—be # 0, oF alternatively, since the function 2 = 2(2) is not constant. This shows that each fractional linear transformation is a one-to-one function from the extended complex plane onto itself, “The composition of two fractional linear transformations is again a frac- tional linear transformation. To see this, suppose f(2) = (az +b)/cz +4) and g(z) = (az + 8)/(92 +), and compute _ allaz +S)/(2+A) +6 _ (aa bby)z+ad-+ 65 I) = Cast Bieta) ed ~ (carde+ hres6a II Analytic Functions Since the composition f © g cannot be constant, the condition on the pa- rameters is met, and fog is a fractional linear transformation. ‘Note that the composition corresponds to matrix multiplication, ab) (a B) _ (aatby a8-+b5 cd) \y 6) = catdy cB +d6 The condition ad ~ be # 0 on the parameters is simply the condition that the matrix associated with the fractional liar transformation has nonzero determinant, that is, that the matrix is invertible. ‘The fact that matrix multiplication corresponds to composition can be reformulated in the language of group theory. If we assign to each 2 x 2 invertible matrix the corresponding fractional linear transformation, we obtain what is called a “group homomorpaism.” from the group of 2 x 2 invertible matrices with complex entries onto the group of fractional linear transformations with operation composition. A fractional linear transformation depends on four complex parameters. One of these can be adjusted without changing the transformation, for instance by multiplying all the parameters by the same nonzero constant. ‘That leaves three parameters to be specified. The next theorem shows that there are three independent complex parameters that describe fractional linear transformations uniquely, namely, the images of any three prescribed points, ‘Theorem. Given any three distinct points 2,21, 22 in the extended com- plex plane, and given any three distinct values wo, 1, ws in the extended ‘complex plane, there is a unique fractional linear transformation w = (2) such that w(z9) = wo, w(z1) = wy, and w(22) = we. To establish the existence assertion, it suffices to show that any three distinct points can be mapped by a fractional linear transformation to 0, 1, and co. Indeed, if f maps 29, 21,22 respectively to 0,1, 90, and g maps ‘wo, ws 2 respectively to 0, 1,90, then the composition gf, of f followed by the inverse of g, maps 29, 21,72 to wp, w:,w2. If now none of the points 2, 21,22 is 00, a transformation mapping them to 0, 1,00 is given explicitly by (72) w= f(z) 1} together with oo, Since i is mapped to 0, which is inside the unit crcl, the image of the upper half-plane is the inside, that is, itis the open unit disk {[w| < 1} (6) The image of the horizontal line through i is a circle passing through O and 1, and it lies inside the unit disk, soit must be the cirele centered at 4} of radius 4 The image of any horizontal line is a circle through w(oo) = 1, and it is ‘orthogonal to the real line (the image of the imaginary axis). These images ‘of the horizontal lines form a pencil of circles as sketched in the figure. The images of vertical lines are circles through w(s0) = 1. Since the real axis is, the image of the imaginary axis, these circles must be tangent to the real axis at 1. ‘The images of the vertical lines are also sketched in the figure. Note that the images of the horizontal and vertical lines are orthogonal to each other. Exercises for IL7 1. Compute explicitly the fractional linear transformations determined by the following correspondences of triples:6s 10. 1. II Analytic Functions (a) (144,20) 0,00,8- 1) (e)_(1,2,00) + (0,1,00) (b) (0.1.00) ++ (1,142) (6) (0,00,8) + (0,1, 00) (©) (2,1 + 4,2) (0,1, 00) (a) (0,1,00) + (0, 00,8) @) (24,2) 0 -24,0,1421) ) 4-1) 00,1) . Consider the fractional linear transformation in Exercise 1a above, which maps 1 +i to 0, 2 to oo, and 0 to i—1. Without referring to ‘an explicit formula, determine the image of the circle {|2—1] = 1}, the image of the disk {|z — 1] <1}, and the image of the real axis, Consider the fractional linear transformation that maps 1 to i, 0 to 1 +i, and -1 to 1. Determine the image of the unit circle {|2| = 1}, the image of the open unit disk {Jz| <1}, and the image of the imaginary axis. Ilustrate with a sketch. Consider the fractional linear transformation that maps —1 to ~i, 1 to 2i, and i to 0. Determine the image of the unit circle {|2| = 1}, the image of the open unit disk {\z| <1}, and the image of the interval [=1, +1] on the real axis. Ilustrate with a sketch. What is the image of the horizontal line through i under the frac- tional linear transformation that interchanges 0 and 1 and maps —1 to 144? Mustrate with a sketch. Show that the image of a straight line under the inversion 2 +» 1/2 is a straight line or circle, depending on whether the line passes through the origin. Show that the fractional linear transformation f(2) = (az-+8)/(cz+ @) is the identity mapping = if and only if b= c= 0 and a=d £0. Show that any fractional linear transformation can be represented in the form f(z) = (az +8)/(cz +d), where ad ~ be = 1. Is this, representation unique? Show that the fractional linear trensformations thet are real on the real axis are precisely those that can be expressed in the form (az +8)/(cz +d), where a, b, ¢, and d are real ‘Suppose the fractional linear transformation (az+b)/(cz-+d) maps R to R, and ad— bc = 1, Show that a, 6, ¢, and d are real or they are all pure imaginary. ‘Two maps f and g are conjugate if there is h such that g = ho Font, Here the conjugating map f is assumed to be one-to-one, ‘with appropriate domain and range. We can think of f and g as the “same” map, after the change of variable w = h(2). A point. 29 isa fixed point of f if f(za) = 20. Show the following. (a) If f Exercises 6 2 is conjugate to g, then g is conjugate to f. (b) If fy is conjugate to fa and fa to fs, then fi is conjugate to fy. (c) If f is conjugate to 9, then fo fis conjugate to go, and more generally, the m-fold composition fo---o f (m times) is conjugate to g0---0g (m times) (d) If f and g are conjugate, then the conjugating function h maps the fixed points of f to the fixed points of g. In particular, f and g have the same number of fixed points Classify the conjugacy classes of fractional linear transformations by establishing the following (a) A fractional linear transformation that is not the identity has tither 1 or 2 fixed point, that is, points satisfying f(zo (b) Ifa fractional linear transformation f(z) has two fixed points, then it is conjugate to the dilation z ++ az with a # 0,a #1, that is, there is a fractional linear transformation h(z) such that A(J(2)) = ah(2). Isa unique? Hint. Consider a fractional linear transformation that maps the fixed points to 0 and oo. (c) Ifa fractional linear transformation f(z) has exactly one fixed point, then it is conjugate to the translation ¢ ++ ¢+1. In other words, there is a fractional linear transformation A(z) such that AUF(1(O)) = C+ 1, oF equivalently, such that h(/(3)) = Az) +1." Hint. Consider a fractional linear transformation that maps the fixed point to oo.Tl Line Integrals and Harmonic Functions In Sections 1 and 2 we review multivariable integral caleulus in order to prepare for complex integration in the next chapter. The salient features are Green's theorem and independence of path for line integrals. In Sec- tion 3 we introduce harmonic functions, and in Sections 4 and 5 we discuss ‘the mean value property and the maximum principle for harmonic func- tions. Sections 6 and 7 include various applications to physics. ‘The student may proceed directly to complex integration in the next chapter after pag- ing through the review of multivariable calculus in Sections 1 and 2 and reading about harmonic conjugates in Section 3. 1. Line Integrals and Green’s Theorem Line integrals play an important role in complex analysis. In this section and the next we review line integrals in the plane, without filling in all the details. We begin by saying something about paths and curves. ‘A path in the plane from A to B is a continuous function t+ >(¢) on some parameter interval a <¢ by (2@),¥(@) = (cosd.sind), 0 @ 0 in the curve 7, while dr <0 on ice + decreases on the curve), on 310) =(cos 4, sin Gol a0, ‘A domain D has piecewise smooth boundary if the boundary of D can be decomposed into a finite number of smooth curves meeting only at endpoints. By “smooth” we usually mean “continuously differentiable,” though in applications the curves making up the boundary will usually be straight line segments or arcs of circles. We denote the boundary of D by D. For purposes of integration, the orientation of 9D is chosen so that D lies on the left of a curve in @D as we traverse the boundary curve in the positive direction, that is, as the parameter value increases, 1, Line Integrals and Green's Theorem 2 Example. To evaluate fj, zy dz, where D is the quarter-dsk in the first quadrant, we divide the integral into three pices, on 09) a ce eo loo /, fio) too) where 7 is the quarter-ircle in the preceding example, and the other two paths are straight line segments, ‘The integral along the horizonval interval on the z-axis is 0, because zy = 0 there. ‘The integral along the vertical interval on the y-axis is 0, because dr — 0 there. (To soe this, either parametrize the line segment explicitly, o go back to the defiition (1.1) and observe that each of the 2,'s is 0.) Using the result of the preceding example, we find that the value of the integral around Dis ~$ A very useful tool for evaluating line integrals is provided by Green’s theorem, which converts a line integral around the boundary of a domain to an area integral over the domain. Green's Theorem. Let D be a bounded domain in the plane whose boundary OD consists ofa finite number of disjoint piecewise smooth closed curves. Let P and Q be continuously differentiable functions oa DU aD. Then (a) [Pe + Qdy = I@ = Peed, Example. We again evaluate pp zydr, where D isthe quarter-disk in the first quadrant, this time using Green’s theorem. In this case, P(x, y) = zy and Q(z, y) = 0, so (1.4) becomes [278 =~ [[,eaedy = - [[ roonoraran = [cmos far = 00h, as before, ‘Since Green’s theorem is of fundamental importance, we provide a sketch Of the ideas behind the derivation of the formula (1.4). One basic idea is to cut the domain into little curvilinear triangular pieces and treat each piece separately. Another is to reduce the double integral over a triangle4 TIL Line Integrals and Harmonic Punetions ‘to a line integral by applying the fundamental theorem of calculus in one variable with the other variable as a parameter. For convenience, we break the proof-sketch into three steps. ‘The first step is to establish the formula (1.4) for the triangle T with vertices at (0,0), (1,0), and (0,1). We must establish the two identities [pre =~ [fen [au = ff Baoan Consider just the first identity here, We represent the double integral as an iterated integral and use the fundamental theorem of calculus, to obtai ff eeear= [LL “EE alas = [rea 2) de~ [f reeves ‘The sum on the right we recognize as — [yy Pdr, after we parametrize separately the three sides of 27. Indeed, the line integral of Paz along the vertical edge of T is 0, since dr = 0 there; the line integral of Pdr along the bottom edge of Tis [j P(z,O)das and the line integral of Paz back along the hypotenuse of T is — {i P(x,1~2}dz, where we have used the parameterization y = 1 — 2: XK ~~ differentiable tangle ‘The second step of the proof is to establish the formula for any domain D that can be obtained from the triangle T by a change of variables. (See Exercise 7.) ‘The final step in the proof, for an arbitrary domain D, involves triangu- lating D, that is, cutting D into small triangular pieces, each of which can be obtained from the triangle T by a change of variables. Green’s theorem is applied to each triangular piece, and the results are added. ‘The surm of the area integrals over the triangular pieces is the area integral over D. The boundary integrals over the sides of the triangular pieces inside D cancel in pairs, since each curvilinear triangle side is traversed twice, once in each, direction, and the opposing directions cancel. The boundary integrals over the curvilinear triangle sides in OD add up to the integral over OD. ‘angulation Exercises Ey Note that we will be using Green’s theorem only for relatively simple do- mains, those whose boundaries consist of straight line segments and circular arcs, for which Green's theorem can be established relatively easily. Exercises for 1.1 1. Evaluate f, y’dr + 2%dy along the following paths + from (0,0) to (2,4): (a) the arc of the parabola y = 2%: (b) the horizontal interval from (0,0) to (2,0), followed by the vertical interval from (2,0) to (2,4); (6) the vertical interval from (0,0) to (0,4), followed by the horizontal interval from (0,4) to (2,4). 2. Evaluate [, rydr both directly and using Green's theorem, where ‘vis the boundary of the square with vertices at (0,0), (1,0), (1,1), and (0,1) 3, Evaluate [yp 2°dy both directly and using Green's theorem, where Dis the quatter-disk in the first quadrant bounded by the unit circle and the two coordinate axes 4. Evaluate {, ydr both directly and using Green's theore is the semicircle in the upper half-plane from R to ~R. 5. Show that fy dy is the area of D, while Jy. ydz is minus the ‘area of D. , where 7 6. Show that if P and Q are continuous complex-valued functions on curve 7, then ry = f Bs f aie (2=2+%y) is analytic for w C\y. Express F'(w) as a line integral over 7) 7. Show that the formula in Green's theorem is invariant under coor- dinate changes, in the sense that ifthe theorem holds for a bounded domain J with piecewise smooth boundary, and if F(z.y) is @ smooth function that maps UT one-to-one onto another such do- main V and that maps the boundary of’ one-to-one smoothly onto the boundary of V, then Green's theorem hols for V. Hint. Fist, note the change of variable formulae fr line and ate integral, given i ~ f wor)(%ar « % [Pa = [wor (Gide » San), Jf rasan = [faery det Je dr dy,6 TI_Line Integrals and Harmonic Funetions (E(2,y),n(z,y)). and where Jp is the Jacobian where Fe et —@P/dy. The summand Inatrix of F. Use these formulae, with R ‘[ Qanis treated similaely. 8, Prove Green's theorem for the rectangle defined by zp < 2 <2) and yo ,,(t), tea St < te, lie in the same disk Ay, in D. This can be done on account of the continuity of 74(t) in sand t. Then “,_, is “sufficiently near” to “js, for each j, and the rolling wave argument above shows that the integrals of Pd + Qdy over the two paths are the same, A slight variation of the argument establishes a deformation theorem for closed paths in D. These are paths in D that start and end at the same point. When we deform closed paths, we allow the starting point to move also, Theorem. Let D be a domain, and let 79(t) and 71(t), a exact => closed. For star-shaped domains we have shown that Independent of path <=> exact <=+ closed. We have also shown that if Pdr + Qdy is a closed differential, then a deformation in the path from A to B does not change the value of the integral of Pdz + Qdy along the path. Exercises for 111.2 1. Determine whether each of the following line integrals is indepen- dent of path. If it is, find a function h such that dh = Pdr + Qdy. Irie isnot, find a closed path around which the integral isnot zero. (a) rdz+ ydy, (b) x2dr + y°dy, (c) yd + 2dy, (d) ydz — dy. 2 Show that the differential minke + a M4000). is closed, Show that itis not independent of path on any annulus centered at 0 3. Suppose that P and @ are smooth functions on the annulus {a < [z| < 6} that satisfy @P/dy = AQ/dx. Show directly using Green's theorem that f.,_, Pdx + Qdy is independent of the radius r, for a(¢), 0 <# <1, in the annulus {a < |e (0)] is constant. Then start by finding a subdivision 0 = to < t) <-:- < ty =1 such that arg(t) has a continuous determination on each interval t).1 <¢ < ty 3, Harmonie Conjugates 83 7. Show that if 0 and oo lie in different connected components of the complement €*\D of D in the extended complex plane, then there is a closed path 7 in D such that [,d0 #0. Hint. The hypothesis means that there are 6 > 0 and a bounded subset E of C\D such that 0 € E, and every point of E has distance at least 56 from every point of C\D not in B. Lay down a grid of squares in the plane with side length 6, and let F be the union of the closed squares in the grid that meet E or that border on a square meeting E. Show that OF isa finite union of closed paths in D, and that Jyp dé = 2r. 3. Harmonic Conjugates The basis for application of Green’s theorem to harmonic functions is the following important observation. Lemma. IFu(z,y) is harmonic, then the differential au, , Ou (a) Hye + Fe is closed. Indeed, for this differential the condition (2.1) for P Ou/dx becomes ~9*u/dy? = 0%u/Ax?, which is equivalent to Laplace's equation, ‘Now suppose that u(, y) is harmonic on a star-shaped domain D. If we apply the theorem in Section 2 to the differential given in (3.1), we obtain ‘a smooth function v(z,y) such that ou, , ou (32) de = ~Fde + Fay ‘The equation (3.2) is equivalent to av due Ge ~ By’ By B" which are the Cauchy-Riemann equations, ‘Thus u + iv is ana have established the following theorem. and we ‘Theorem. Any harmonic function u(2,y) on a star-shaped domain D (as a disk or rectangle) has & harmonic conjugate function v(x, y) on D. By (3.2), the harmonic conjugate v(z,y) is given explicitly up to an additive constant by a Bou Ou (3) By = f° Bae + Bay,8 TIL Line Integrals and Harmonic Functions where A is fixed, and the integral is independent of path in D. If D is a disk, and we take the path from A to B to be a vertical interval followed by a horizontal interval, we obtain the formula (5.3) derived in Section IL, Example. To find a harmonic conjugate v(2) for u = log |z| on the star- shaped domair C\(~o0,0), we express u in the form g(a? + y? u(x) = 5 lost2? +9"), and we compuce au = 52 Pap Equation (3.2) becomes ‘This leads to the identity ame = [oats + stan 2¢ (0,0), since the prineipal branch Arg. is the unique harmonic conjugate of log [2] on C\(~20,0), normalized to vanish at z = 1 Exercises for IIL 1. For each ofthe following harmonic functions u, find du, find dv, and find v, the conjugate harmonic function of u (2) way) =2—y (b) ulay) = 28 — Bay? (4) u(z,y) = Pte 2 Show that a complex-valued function (2) on a star-shaped do- ‘main D is harmonic if and only if A(z) = f(z) + 9(z), where f(2) and (2) are analytie on D. 3. Let D = {a < |2| <0}\(—b,~a), an annulus slit along the negative real axis. Show that any harmonie funetion on D has a harmonic conjugate on D. Suggestion. Fix ¢ between a and b, and define v(2) explicitly as a line integral along the path consisting of the straight line from ¢ to |2| followed by the circular arc from |2| to 2. Or map the slit annulus to a rectangle by w = Log 2. 4. Let u(z) be harmonic on the annulus {a < |2| 0, and the constant is u(29). This establishes (4.1). We say that a continuous function h(z) on a domain D has the mean value property if for each point 2 € D, h(2p) is the average of its values over any small circle centered at 29, In the language of formal mathematies, this means that for any =) € D, there is © > 0 such that Aza) = fo (20+ re! Our theorem above has a simple restatement: Harmonic functions have the mean value property. We will show in Chapter X that the converse js true, that any continuous function on D with the mean value property Js harmonic. This is rather remarkable, since the hypothesis requires ouly continuity but no differentiability of the function, Exercises for IL. 1, Let f(z) be a continuous function on a domain D. Show that if (2) has the mean value property with respect to circles, as defined above, then f() has the mean value property with respect to disks, that is, if 3) € D and Do is a disk centered at zp with area A and contained in D, then £20) = % f if, F(z) de dy, 2. Derive (4.2) from the polar form of the Cauchy-Riemann equations (Exercise IL3.8). 5. The Maximum Principle 87 3. A funetion f(t) on an interval J = (a,b) has the mean value prop- st+t) _ fis)+s() (32) = 2231, ae 4, Formulate the mean value property for a function on a domain in R°, and show that any harmonic function has the mean value property. Hint, For Ae B® and r > 0, let By be the ball of radius r centered at A, with volume element dr, and let AB, be its boundary sphere, ‘with area element da and unit outward normal vector n. Apply the Gauss divergence theorem Sh? ado = fff, v-Par Vu. 5. The Maximum Principle ‘The strict maximum principle asserts that if a real-valued harmonic fune- tion attains its maximum on a domain D, then it is constant, Strict Maximum Principle (Real Version). Let u(2) be a real-valued harmonic function on a domain D such that w(z) 0) and continuous, the integral (5.1) can be zero only if the integrand is zero. Thus u(zi + re) = u(e1) = M for 0-< 0 <2n and 0 1,r > 0, and A= pel. What is the maximum modulus of 2" + A over the disk {|z| 1 has a zero, by applying the maximum principle to 1/p(z) on a disk of large radius. Let f(z) be an analytic function on a domain D that has no zeros on D. (a) Show thet if |f(2)| attains its minimum on D, then f(z) is constant. (b) Show that if D is bounded, and if f(z) extends con- tinuously to the boundary OD of D, then |f(z)| attains its minimum on OD. Let f(z) be a bounded analytic function on the right half-plane. Suppose that f(z) extends continuously to the imaginary axis and satisfies |f(iy)| © M for all points iy on the imaginary axis. Show that | f(=)| 0 small, consider (z + 1)~*f(z) on a large semidisk. Let f(z) be a bounded analytic function on the right half-plane. ‘Suppose that lim sup |f(z)| < M as z approaches any point of the imaginary axis. Show that |f(2)| < M for all = in the half-plane. ‘Remark. This is a technical improvement on the preceding exercise for students who cen deal with a limsup (see Section V.1) Let f(z) be a bourded analytic function on the open unit disk D. Suppose there are a finite number of points on the boundary such that f(z) extends continuously to the arcs of @D separating the points and satisfies [f(e'*)| 0 on 2D. Show that u(z) >0 on D. Hint. Suppose 0 € OD, and consider functions of the form u(2) + plog|2| on DM {\z| > 6} Let D be a bounded domain, and let 2) € AD. Let u(z) be a harmonic function on D that extends continuously to each boundary point of D except possibly 9. Suppose that u(z) is bounded below0 TIT Line Integrals and Harmonic Functions on D, and that u(2) > 0 for all z € AD, z # 29. Show that u(z) > 0 on D. 11, Let E be @ bounded set of integer lattice points in the complex plane. A point m_+ni of B is an interior point of F if its four immediate neighbors m_+1+ni, m+ni-+i belong to E. Otherwise, m+ni is a boundary point of E, A function on E is harmonic If its value at any interior point of B is the average of its values at the four immediate neighbors. Show that a harmonic function on a bounded set of lattice points attains its maximum modulus on the boundary of the set. 6. Applications to Fluid Dynamics ‘We consider a fluid flow in a domain D in the plane. We think of the fluid as a collection of particles that move in the plane as time evolves. We associate with the particle at the point = its velocity vector V(z)- The direction of V(z) is the direction the particle is moving, and the magnitude [V(2)| is its speed. We make the following assumptions on the flow. 1. The flow is independent of time, that is, the velocity vector field ‘V(2) does not change with time. 2, There are no sources or sinks in D; that is, no fluid is created or destroyed in D. 3. The flow is incompressible; that is, the density of the fluid is the same at each point of D, and it does not change with time. 4. The flow is “irrotational”; that is, there is no circulation of fluid around small circles centered in D. ‘We will return shortly te explain this fourth condition. First we define “flux” across a curve and “circulation” around a closed curve, Let 7 be a curve in the plane, We denote the unit tangent vector to 7 by ¢ and the unit normal vector to 7 by n. If 7 is parameterized by arc length s, 9(s) = ((s),(3)), then the unit tangent and normal vectors are given respectively by dy _ de ~ \as* as ‘The normal component of V to the curve + is then V-n. If the flow is parallel to 7, then V -n = 0, and no fluid crosses. The maximum flow per unit length is obtained when 7 is orthogonal to V. We define the flux of the fluid flow across y to be the integral of the normal component of V with respect to arc length, puseneary = [Vina 6. Applications to Fluid Dynamics By Except for a constant involving units ané the density, the flux across y can be regarded as the amount of fluid crossing 7. If we express V = (P,Q) in terms of its components P and Q, we obtain an expression for the flux across 7 as a line integral: tay vinds = [rlten oftes = [pas — ose If 7 is a closed curve, we define the cireulation of the fluid flow around + to be the integral of the tangential component of V with respect to are length: Gieulation around y =f V-tds ‘This can also be expressed in terms of the components P and Q of V as a line integral, (2) [vse = [vlan + often = [rae + aan Let ye = 20 +e" be a small circle around a fixed point z € D. From (6.2) and Green's theorem, the amount of fluid circulating around 2 is piven by fvewe ff Gao ‘The ivottionality ofthe Suid flow (the fourth condition above) means ‘Ey ha has integrals or oral salle > 0. This occur i and onl ithe integrand is ar, thot or _ 9 oy Oe Thus the rotational of the fow on D means tat (6.3) holds on D. "The mathemati formulation ofthe second and hid conditions that che ne fow of ud aros the boundary of any small ile centered at point 2 of Dis sero, Using Green's theorem again, we btn on f Vente = [ray —aie~ ff (2 + )aee (6)2 INL Line Integrals and Harmonic Functions We conclude as before that ar, ag de * dy ‘Thus the second and third conditions on the flow can be reinterpreted to assert that (6.4) holds on D. ‘The conditions (6.3) and (6.4) can be expressed directly in terms of the velocity V(2), a8 (6.4) 0. VxV=0, VV=0. ‘This is the form of the equations that is usually most recognizable to phys ical scientists. It generalizes to three dimensions. Now, the condition (6.3) is just that the differential Pde+@Qdy be closed. ‘On any disk in D, we can then find a smooth function @ suck that d = Paz + Qdy, that is, In terms ofthe vector field V. this isthe sameas Vo = V. The function is called the potential function of V. It is unique, up to adding a constant. In terms of, the condition (6.4) is that #6 , #e Ox? © Oy that is, that the potential function ¢ is harmonic. On any disk in D, @ ‘then has a conjugate harmonic function 1, so that f(z) = (2) + i¥(2) is analytic. The function f(z) is called the complex velocity potential of the flow. It is also unique up to adding a constant. Note that while > and y may be defined on any disk in D, neither ¢ nor y need be defined onall of D ‘The velocity vector field V(z) is expressed in verms of the complex ve- locity potential (2) = (2) + 86(2), in complex notation, by 6, (2% 4 os ty 7 oe 7 ae FG). ‘Thus the speed of the fluid particles is given by |V(z)| = |f"(z)|- [As we saw in Section 6, the level curves of 6 and of are orthogonal to each other. Since the level curves of ¢ are also orthogonal to V¢ = V, the level curves of are tangent to V. ‘Thus the Ruid particles fow along the level curves of We define the streamlines of the fluid flow to be the level curves of y, that is, the curves {v ~ ¢}, for c constant. ‘The streamlines of the Aow represent the paths ofthe fluid particles, ‘The funetion p(2) i called the Stream function of the fd flow. ‘The stream function 9 ean be used to calculate the flax ofthe ud flow across a curve 7. Indeed, the Ru across 7 is given withthe help of the Vie) 6. Applications to Fluid Dynamics 98 A a [vinis = [2a - Bite = | Bay « Bae = fue ‘Thus the flux of the flow across a curve + is equal to the increase of the stream function y(2) along 7. - KE =O, Dot)edery Example. The simplest flow in the plane is a constant flow V = (a,), where a and 8 are constants. The velocity potential @ satisfying Vo = is given, up to an additive constant, by o(z,y) = ax + By. The stream function ¢) is the harmonic conjugate of 6, given up to an additive constant by W(t.) = ay — G2. The complex velocity potential of the flow is (2) = (a-is)z. example. Consider the vector field on the punctured plane C\{0} defined by ve lw where u, is the unit vector in the radial direction. We can express viz) a ee ee me basen VG) trees Boe ee ant el lator oft Nowa ele) oe cee ee tad IT Pete complsveety peteatd 6 t= ce dn eo th fn fhe Bow acon scl enter Ce ei er iciy mop ae usando tad oy vends = foeuerde = [an = 24 IIL Line Integrals and Harmonic Functions This coincides with the increase of the stream function arg around the circle. ‘The origin is a source for the fluid flow. The speed of the fluid particles is given by |V(z)| = |f"(2)| = 1/l2| = 1/r. ‘The fluid particles emanate from the origin and follow along rays at continually diminishing speeds. In addition to the four conditions above, there is one further important condition to be placed on a fluid flow velocity V(2), which is a boundary condition. 5. If no fluid is injected or extracted through a boundary curve, then the velocity vector field V(z) is parallel to the boundary along that ‘Thus the boundary curve should be a streamline of the flow, and the stream function (2) should be constant on the curve. “nie ae Aq ~ ty De Cn wt % sowce a0 Example. We consider a fluid flow in the upper half-plane with a source or sink at 0. The fluid is injected or extracted from the upper half plane fat a constant rate at the origin. The stream function ~(z) should be ‘ harmonic function in the upper half-plane that is constant on each of the boundary intervals (00,0) and (0, +90). Such a function is given by W(z) = Cargz, where C is a constant. ‘The corresponding complex velocity potential function is f(=) = Clog z, and the velocity vector field is V(2) = FE) = 1/2 = 2/2|*. The Mus of the fluid flow entering or exiting at the origin is the increase of (=) along a small semicircle in the upper half-plane centered at 0, which is tC. This determines the constant C. If C> 0, we have a source at 0, and if C <0, we have a sink at 0. One way to gain insight into a fluid flow is to map D conformally onto a domain for which the corresponding flow is simpler to understand. Flows are preserved by conformal maps, in the following sense. If h: D— U is ‘@ one-to-one analytie function (conformal map) from D onto a domain U, and if fo(w) = do(w) + tvo(w) is a complex velocity potential for a flow ‘on U, then the composition f(2) = fo(/(2)) is analytic, hence the complex velocity potential for a flow on D. One of the simplest flows to understand is the constant horizontal flow ‘on the upper half-plane H = {Im 2 > 0}, for which no fluid enters or leaves, sucrose the bounding real line R. A complex velocity potential for the flow is 6. Applications to Fluid Dynamics 95 fo(2) = =. The streamlines of the flow are horizontal lines in H, for which Jim(z) is constant. The boundary B of His a streamline, corresponding to Inn(z) =0. Suppose now + is a curve in the plane extending to infinity in both dizections, and suppose D is the domain lying on one side of 7. Suppose we wish onda ud ox in D such that no fd enters or exis through ‘Recording to the boundary condition, 7 should be a streamline of the flo ‘The problem can be solved by Snding a conformal map h : D—> Hl that maps 20 to 0, In this case the cemplex velocity potential f(2) = fo((z))is simply f(z) itself, and the vector velocity for the Auid flow is V(=) = h’(2) ‘The stream function is uz) = Im(h(z)), and this is zero on. Thus 7 is a streamline, and the boundary condition is satisfied. Note that any real multiple Chi(2) is also a complex velocity potential for a fluid flow that satisfies the boundary conditions and has the same streamlines. It corresponds to multiplying the velocity veetor field by C. 4, oe = » —— a SS a Example. Consider a fluid flow at « comer, represented by a sector D = {0 < atgz 0, 22 + y? = 1} of the disk, and at ~1 on the edge of the bottom half {y <0, 2? +y? = 1} of the disk Solution. We recall that the argument function Arg w has a similar behav- ior in the upper half-plane H, in that itis constant on each of the intervals (-00,0) and (0, +00) of the boundary of H. We map the upper half-plane to the unit disk by a fractional linear transformation so that the interval (00,0) corresponds to the lower semicircular edge of the disk and (0, +0) to the upper semicircular edge. The transformation is given explicitly by wri wees Since Log w(2) is analytic on the disk, its imaginary part Argw(2) is har- monic, and it attains the values 0 on the top edge of the disk and = on the bottom edge. Thus the solution is given by us) = 1~ 2 Argu(e) = 2[arg(a +2) ~ Ara(t —2)) 7. Other Applications to Physies 29 ‘This solution can be expressed as u(z) = (2n)(y + ¥), where y and w are the angles represented In the figure. 7) Now we turn to electrostatics. Here it is usual to start with the electric force field E on D, defined so that E(A) is the force exerted on a unit, charge if it were placed at the point A. The work done on a unit charge in moving it along a path 7 in D is then given by the integral f, E-tds, ‘where tis the unit tangent to the curve, and ds is are length. The field B is assumed to ke irtotational: V x E = 0. This is equivalent, at least locally, to assuming shat J E-t ds is independent of path, ‘Thus E = Vé for some function ¢, which is called the electric field potential. The work done ‘on a unit cherge in moving it along a path + from A to B is then given by work done [tas = $B) - (A) If there are no sources or sinks in D, then V-E = 0, and this is equivalent to Laplace's equation Ag = 0. Thus the electri field potential ¢ is harmonic in D. Example. By a “line charge” we mean a uniform distribution of charge along a line (infinite straight wire) in three-dimensional space. We assume that the line is perpendicular to the (,y)-plane. The electric field is then independent of the z-direction, and it is sufficient to deseribe it in the (2,u)-plane. The electrie field corresponding to a line charge through the origin is e=hy, up to a constant multiple depending upon units, where us is the unit vector in the radial direction. The potential function associated with a line charge at Ois @ = logr, since Vo = B. Note that ¢ is harmonic; however, 6 has no single-valued conjugate harmonic function. ‘The equipotential lines for E are circles centered at the origin. Through each such circle {|2| =r} there is a positive outward flux given by Bends = [Beds = [a9 = 2 har hair lo in Is a source for the electric field B. The origin is a100 TIT Line Integrals and Harmonic Functions Exercise. Find the electric fleld E and the potential function in a cir- ‘cular ring whose inner edge is grounded (¢ = 0) and whose outer edge is conducting (@ = constant) Solution. The problem is invariant under rotations, so we try a potential solution that is invariant under rotations, say ¢ = alogr +b, The constant ‘values of the potential on the inner and outer circles determine the con- stants a and 6, The electric field is given by E = Vo = (a/r)ur, which is the electric field of a line charge. As in the case of fluid flow, problems in electrostaties and in heat flow that are effectively two-dimensional and can be analyzed using complex analysis and conjugate harmonic functions are very special. A typical problem in three dimensions involves a vector field F that satisfies the conditions VxF = 0 (Fis irrotational), V-F = 0 (no sources or sinks in D). Again, the first condition is equivalent to fF - dx being independent of path, and we say that F is a conservative field. In this case F = Vo for some potential function ¢, at least locally. In terms of the potential function 6, the second condition becomes Ad = 0, and ¢ is harmonic: ‘Thus harmonic functions play an important role, as do the Dirichlet and Neumann problems. However, in three-dimensional problems the harmonic functions cannot be analyzed as the real parts of analytic functions, and conformal mapping techniques are not available. Exercises for 111.7 1. Find the steady-state heat distribution u(z,y) in a laminar plate corresponding to the half-disk {27 + y? <1, y > O}, where the semicircular top edge is held at temperature 7; and the lower edge (1,1) is held at temperature T;. Find and sketch the isothermal curves for the heat distribution. Hint. Consider the steady-state heat distribution for the full unt disk with the top held at tempera- ture T; and the bottom at temperature T3, where Ty = (Ty +Ts)/2. 2. Find the potential function (2, y) for the electric field for a con- ducting laminar plate corresponding to the quarter disk (1? + y? < 1,2 > 0,y > 0}, where the two edges on the coordinate axes are grounded (that is, = 0 on the edges), and the semicircular edge is held at constant potential Vi. Find and sketch the equipoten- tial ines and the lines of force for the eleetrie field. Hint. Use the conformal map ¢ = =? and the solution to the preceding exercise. 3. Find the potential function 4.7, y) for the electric field for a conduct- ing lamulnar plate corresponding to the unit disk where the boundary Exercises 101 quarter-circles in each quadrant are held at constant voltages Vi, Va, Vs, and Vj. Hint. Map the disk to the upper half-plane by w = w(2) ‘and consider potential functions of the form Arg(w ~ a). Find the steady-state heat distribution in a laminar plate corre- sponding to the vertical half-strip {|x| < 1/2, y > 0}, where the vertical sides at = 7/2 are held at temperature Tp = 0 and the bottom edge (—x/2,/2) on the real axis is held at temperature T, = 100. Make a rough sketch of the isothermal curves and the lines of heat flow. Hint. Use w = sinz to map the strip to the upper half-plane, and make use of harmonic functions of the form Arg(w — a). Find the steady-state heat distribution in a laminar plate corre- sponding to the vertical half-strip {|x| < 1/2, y > 0}, where the side = —7/2 is held at constant temperature Tp, the side x = 1/2 is held at constant temperature T;, and the bottom edge (—/2,7/2) ‘on the real axis is insulated; that is, no heat passes through the bottom edge, so the gradient Vu of the solution u(x, y) is parallel there to the z-axis. Hint. Try linear functions plus constants Find the steady-state heat distribution in a larainar plate corre- sponding to the upper half-plane {y > 0}, where the interval (—1,1) is insulated, the interval (~90,1) is held at temperature To, and the interval (1,00) is held at temperature T;. Make a rough sketch of theisothermal curves and the lines of heat flow. Hint. Use the so- lution te the precesling exercise and an appropriate conformal map. ‘The gravitational field near the surface of the earth is approximately ‘constant, of the form F = ck, where k is the unit veetor in the 2- direction in (2, y,2)-space and the surface of the earth is represented by the plane where z = 0 (the flat earth theory). Show that F is conservative, and find a potential funetion ¢ for F. Show that the inverse square force field F = u,/r® on B® is com servative. Find the potential function ¢ for F, and show that is harmonic. |. For n> 3, show that the function 1/r"-? is harmonic on R"\ {0} ind the vector field F that has this function as its potential.IV Complex Integration and Analyticity In this chapter we take up the complex integral calculus. In Section 1 we introduce complex line integrals, and in Section 2 we develop the complex integral calculus, emphasizing the analogy with the usual one-variable in- tegral calculus. In Section 3 we lay the cornerstone of the complex integral calculus, which is Cauchy's theorem. The version we prove is an immedi- ate consequence of Green's theorem. In Section 4 we derive the Cauchy Integral formula and use it to show that analytic functions have analytie derivatives. Each of the final four sections features a “named” theorem, In Section 5 we prove Liouville's theorem. In Section 6 we give a version of Morera’s theorem that provides a useful criterion for determining whether continuous function is analytic. Sections 7 and 8, on Goursat’s theorem and the Pompeiu formula, can be omitted at first reading. 1. Complex Line Integrals For complex analysis it is convenient to define dz = dr + idy. According to this notation, if h(z) is a complex-valued function on a curve 7, then (ay [roe = [roe + [nerdy Suppose 7 is parameterized by t > 2(0) = x(t) + iy(t), a 0, we show that Fo cgoglt 208 = {y anes ni, man104 IV Complex Integration and Analyticity ‘We parametrize the circle |z~ 29| = R, with the usual positive orientation, by 2(@) = 20 + Re, and we calculate that dz = iRe'¥d9. Thus the integral becomes [ eryminetan = ures [omen The integral on the right is 0 unless m = —1, in which case it is 2x. This vyields the formula. a4 Re A NN In complex analysis tis customary to denote the infinitesimal arc length ds by [de |ds| = ds = Vda + (iy. ‘This means that if a curve 7 is parameterized by 2(¢) = x(t) + iy(t), then as rena = facras = fneen| (2) + (2) a In particular, the length of + is [ laz| “The notation can be justified by considering the sums approximating these integrals. For the subdivision of the parameter interval used earlier, the usual sum used in multivariable calculus to approximate J, h(x.) ds is dt. a) *\a LV(@)-@y [resnds = Snenudy enna) + tse In complex notation this becomes (4) [reriae = Saale — 5) In particular, the sum approximating the length L of 7 is given by (5) Le Slay Example. The parameterization 2(9) = 29 + Re’ of the citcle [2 ~ zo] = R ‘can be used to derive the expression for the infinitesimal arc length for 1. Complex Line Integrals 105 the circle in terms of the central angle @. In this case, 2(0) = Reos8, so that dz = —Rsin@d9, and similarly dy = Rcos@d8. Thus |d2| = ds /(dz}"+ (dy)? = Redd. ‘The length of the circle is {2" ldz| = 2nR, as usual. ‘Theorem. Suppose 7 is a piecewise smooth curve. If h(2) is a continuous function on 7, then [reel < [imei Further, iy has length L, and [h(=)| < M on 7, then | rceyae an | foo (1.6) < ML. ‘The estimate (1.6) is the triangle inequality for integrals. It will be used frequently and without reference. The estimate (1.7) is called the ML-estimate. It may or may not be referred to, These estimates both follow from the corresponding estimates for the approximating sums. The triangle inequality (1.6) for integrals follows from the triangle inequality for the approximating Riemann sums, [Daten -29] < Dimenleser ak in view of 1.2) and (1.4), The MZ-estimate (1) follows from the etimate [Sheer 2) 1, m>1, n20. if a < Fag BO 1}, with the usual counterclockwise 2. Fundamental Theorem of Calculus for Analytic Functions 107 '8. Suppose the continuous function f (e!) on the unit circle satisfies s(e®)| SM and | fy, S(2)dz| = 20M. Show that f(z) = cf for some constant ¢ with modulus |o| = M. 9. Suppose h(z) is a continuous function on a curve +. Show that = [Phe wee, Hw) is analytic on the complement of >, and find H’(w). 2. Fundamental Theorem of Calculus for Analytic Functions Let f(z) be a continuous function on a domain D. A function F(z) on D is a (complex) primitive for f(z) if F(z) is analytic and F"(z) = f(z). The following theorem is the analogue of the first statement of the fundamental theorem of calculus. Theorem (Part I). If f(z) is continuous on a domain D, and if F(z) is a primitive for f(z), then I Slz)dz = F(B) ~ F(A), where the integral can be taken over any path in D from A to B. ‘This formula follows from the corresponding formula for line integrals. In this ease, we have so that Bo OF OF [lr [ses Eu [predesian = [roa This theorem provides a powerful tool for evaluating definite integrals. ‘The problem of evaluating {,° f(z)dz is reduced to that of finding an ana- lytic function F(2) whose derivative is f(2). F(B)~ F(A) Example. To integrate z? from 1 to i, we observe that 9/3 is a primitive for 2, and then we proceed as we would to evaluate an ordinary garden- variety integral, f a: = =|" a+y?108, IV. Complex Integration and Analyticty ‘This coincides with the result that was obtained in the preceding section by parameterizing the straight line from 0 to 1+i, Note, though, that the integral is independent of path. Example. The function 1/2 does not have an analytic primitive defined ‘on any domain containing the unit circle. To evaluate the integral of dz/2 around the unit circle we can still use the primitive F(z) = Log: in the slit plane C\(~20,0), by taking the integral over a path counterclockwise ‘around the unit circle starting at a point ~1 — Oi just below the slit and ending at a point ~1 +01 just above the slit. Then = in — i(-n) = 2m, which coincides with the result obtained earlier by parameterizing the unit circle. ‘The analogue of the second statement in the fundamental theorem of calculus is that every analytic function on a star-shaped domain has a primitive, which can be defined as an indefinite complex integral ‘Theorem (Part II). Let D be a star-shaped domain, and let f(2) be analytic on D. Then f(z) has a primitive on D, and the primitive is unique up to adding a constant. A primitive for (2) is given explicitly by Fe) = [How =D, where 2» is any fixed point of D, and where the integral can be taken along any path in D from 2 to =. ‘To see this, we write f = u+ in’ as usual, and we consider the differen- tial udx — vdy. Since f is analytic, the Cauchy-Riemann equations yield u/Ay = -Au/Ax, and the differential is closed. By the fundamental the- lorem of calculus for line integrals (Section TII.2), the differential is exact fon D, that is, there is a continuously differentiable function U on D such Exercises ro that aU = ude ody. ther word ststen ow ow hey Fs The ousting show that U sn fe eo continu diferente nother pption ofthe Caly Riemann uations pd ey, BY _ ou om GF * OF ~ Or By Hence Ui harmonic. Since Dis tc shaped thr i conjugate barmont function V for on none On en naira ae wv wae C= Ft Ge ig ten and O(2 is a primitive for 2). 1G) another primitive for (2), then Gp has derive ooo Go Gp ncomtant an He oo Part {abot the primithe FU) 6) Oa) a Fhe ae ferme Pace Not inptictlar tet integral of analytic function in star shped d- isin aefdapendet of uty Thies nt tue or every acon ke ft-mive shows that the differential dz/z does not have a primitive on the punctured plane C\ {0}. Exercises for IV.2 1, Bvaluate the following integrals, fora path >that travels from —xi torn in the right half-plane, and also for a path 7 from mi to = in the lef half-plane (a) [ eae wy fea (©) fyeoszdz —(d) fysinhedz 2. Using an appropriate primitive, evaluate [, 1/zds for a path » that travels from —7i to x1 inthe right half plane, and also for path from ~zi to zi in the left halfplane, For each path give a precise definition of the primitive used to evaluate the integral 3. Show that if m # —1, then 2" has a primitive on C\{0}. 4. Let D = C\(-oo, 1}, and consider the branch of V2?=T on D that is positive on the interval (1,0). (a) Show that 2+ ¥2?=1 omits the negative real axis, that is, the range of the funetion on D does not include any values in the interval (~20,0] on the real axis. (b)110 IV. Complex Integration and Analyticity Show that Log(z + V2? —1) is a primitive for 1/v2? = 1 on D. (c) Evaluate where 7 is the path from —2i to +2i in D counterclockwise around the circle |z| = 2. (d) Evaluate the integral above in the case + is the entire circle |s| = 2, oriented counterclockwise. (Note that the primitive is discontinuous at 2 = —2.) 5, Show that an analytic function f(=) has a primitive in D if and only: if f, fl2)dz = 0 for every closed path + in D. 3. Cauchy’s Theorem We begin with a smooth complex-valued function f(z) =u + iv, and we express S(2)dz = (u+iv)(de + idy) = (u+ivjde + (-v + indy ‘The condition that f(2)dz be a closed differential is a a Fete) = let, Taking the real and imaginary parts, we see that this is equivalent to av au _ ou by ey which are the Cauchy-Riemann equations for w and v. Thus we obtain the following theorem, which is the original form of Morera’s theorem, ‘Theorem. A continuously differentiable function f(2) on D is analytic if and only if the differential f(2)d2 is closed. From Green's theorem (Section IIL1.1) we obtain immediately the fol- lowing far-reaching theorem. ‘Theorem (Cauchy's Theorem). Let D be a bounded domain with piecewise smooth boundary. If f(z) is an analytic function on D that extends smoothly to 8D, then f(e)dz = 0. Joo ise mm Sn > Cx Example. If we apply Cauchy's theorem to a function f(2) analytic on the annulus D = {r < |2| < R}, we obtain o= [soa = Fiegl4 - f_ fede Note the change in sign, due to the fact that the inside boundary circle is traversed in the negative (clockwise) direction according to its orientation as a boundary curve for D. Thus fi gfte =f ferae. fan ster This is in accord with the discussion in Section IIL.2, since the differential J(e)dz is closed, and since the outside circle of the annulus can be con- tinuously deformed to the inside circle of the annulus through a family of closed curves (intermediate circles). Exercises for IV.3 1. By integrating e~*"/? around a rectangle with vertices +R, it + R, and sending R to 00, show that Lf eenetids ae? wcteen el. dt , 0 by aya) = (re Ze"), dala) (a) Show that Ha(z) = 2"2"+--- is a polynomial of degree n that is even when nis even and odd when mis edd. (b) By integrating the function e-aeyty2 a cowie 2 (o) around a rectangle with vertices +R, it + R and sending R to oo, show that 27H (2), Fae | tolade de = (ibn), -90 <<.m IV. Complex Integration and Analyticity Hint. Use the identity from Exercise 1, and also justify and use the identity 1a mtv? rae enter? (c) Show that, 6%, ~ 224 + (2n + 1)dy = 0. (4) Using [ On¢mdz = f Pndmdr and (c), show that: [onto ventayaz ngm. Remark. This shows that the d,'s form an orthogonal sys tem of eigenfunctions tor the (normalized) Fourier transform operator F with eigenvalues +1 and -ti. Thus F extends to a unitary operator on square-integrable functions. Further, F# is the identity operator, and the inverse Fourier transform is, given by (F~*f)(x) = (FA)(—2), 3. Let f(2) =e +012 +--+ €n2" be a polynomial (a) Ifthe cy’s are real, show that [sore < af ere ir Hint. For the first inequality, apply Cauchy's theorem to the function f(z)? separately on the top half and the bottom half of the unit disk. (b) If the ¢4’s are complex, show that pa)? ae _ ae [ite ‘de < af" Is (e*)[? 5 = lee (c) Establish the following variant of Hilbert’s inequality, that yr | ca Sel. with strict inequality unless the complex numbers ¢o,.++ sen are all zero. Hint. Start by evaluating p f(x)*dr. 4, Prove that a polynomial in z without zeros is constant (the fun- damental theorem of algebra) using Cauchy’s theorem, along the following lines. If P(2) is a polynomial that is not a constant, write P(2) = P(0) +2Q(2), divide by zP(z), and integrate around a large circle, This will lead to a contradiction if P(z) has no zeros. 4. The Cauchy Integral Formula ns 5. Suppose that D is a bounded domain with piecewise smooth bound- ary, and that f(2) is analytie on DUAD. Show that Area(D) 2B LON 2 2 Tangti(@Dy Show that this estimate is sharp, and that in fact there exist D and F(2) for which equality holds. Hint. Consider Jgp[2~ f(2)\de, and ‘use Exercise d in Section 1 6. Suppose f(z) is continuous in the closed disk {|2| < R} and analytic on the open disk {\2| < R}. Show that f,\__ f(z)d= = 0. Hint Approximate f(z) uniformly by J-(2) = f(") 4, The Cauchy Integral Formula Integral representation formulae are powerful tools for studying functions. One application of an integral representation is to estimate the size of the funetion being represented. Another is to obtain formulae for derivatives, by differentiating under the integral sign. The prototype of the integral representation is provided by the Cauchy integral formula, representing an analytic function. The integral representation will allow us to show that all the derivatives of an analytic function are analytic. It will also allow us to obtain power series expansions for analytic functions. ‘Theorem (Cauchy Integral Formula). Let D be a bounded domain with piecewise smooth boundary. If f(2) is analytic on D, and f(2) extends Smoothly tothe boundary of D, then (aay Haw, 26D. EI lov w—2 To establish the formula, fix a point z € D, let © > 0 be small, and consider the domain D. = D\{|w ~ z| < ¢} obtained from D by punching ‘out a disk centered at z of radius ©. The boundary @D. is the union of AD and the circle {|w— 2| = ¢}, oriented clockwise. Since f(2)/(w — analytic for w € De, Cauchy’s theorem yields i: Reversing the orientation of the circle to counterclockwise produces a sign change, and we obtain Foye t hee =14 IV. Complex Integration and Analyticity Writing w = z+ ee", dw = iee'®u@, and di [resend =f am pw 2 ing by 2ni, we obtain By the mean value property of harmonic functions, the integral on the left-hand side coincides with (:), and the formula is established. QS «= © We remark that without the mean value property at our disposal, we could complete the argument by observing that the integral on the left is the average value of f(w) on the circle centered at z of radius e. Since {J (w) is continuous at z, these averages tend to f(z) es € + 0. This can be Justified with complete rigor by writing [rere 8 = 90 + [erect ra. and showing by a direct estimate that the latter integral tends to 0 as 230. If we differentiate under the integral sign and use aoa mt wae ¢ ‘we obtain integral formulae for the derivatives f0)(2) of f(2) ‘Theorem. Let D be a bounded domain with piecewise smooth boundary. If f(2) is an analytic function on D that extends smoothly to the boundary of D, then f(z) has complex derivatives ofall orders on D, which are given by se) = Bf Opt, 2D, mao. ‘The case m = 0 of (4.2) is the Cauchy integral formula (4.1). Thongh it is not really necessary, we give separately the argument for the case ™ ‘This special case already includes all the ideas for the proof of the general Using (4.1) and the identity 1 1 az w-Gtay wv we Faas) 4. The Cauchy Integral Formula ns wwe express the difference quotient approximating f"(2) as fe+ds-f) Aft fp _ fw) 4 Vf Se he acl eee Bri Jap wz 1 1 a te SOD a As Az tends to 0, the integrand converges to f(w)/(w— 2)®, uniformly for w € OD. Hence the integrals converge, and we obtain in the limit that 4) = 2 f Seo) . sae I, Eats, sep, which is the case m = 1 of (4.2). The general case of (4.2) is proved by induction on m. We assume that f("-)(z) is complex differentiable and the formula holds for f("-")(z). ‘We must show that f("-1)(z) is complex differentiable and the formula holds for f(z). Using the binomial expansion m(m 1) 2 (w~ 2A)" = (w—2)"—m(w-2)"1A2+! (w—2"9(2)2+- and simplifying, we obtain —e e+ Bay where the dots indicate terms with powers of Az up to (Az)". The integral formula for f(z) then yields the expression CM" [so age OH Jae for the difference quotient (f™-0(z + Az) ~ f™-0(2)) /Az. Again the integrand converges as Az — 0, uniformly for w € @D, and we can pass to the limit to conclude that f(z) exists and is given by (42). Since each of the suecessive complex derivatives of f(2) is complex differ- ‘entiable, each is continuous, and thus each is analytic. ‘The hypothesis that the domain has smooth boundary is irrelevant for determining analyticity, as we can restrict the function to an appropriate small disk. ‘Thus we have proved the following. Corollary. If f(2) is analytic on @ domain D, then f(z) is infinitely dif ferentiable, and the successive complex derivatives f'(2), f(z), .., are all ‘analytic on D.nie IV Complex Integration and Analyticity Example. The Cauchy integral formula for 2? yields f 2 de = onic tea 1 = on ‘Example. The Cauchy integral formula for the derivative of 2?sin > yields Poin: , ori a f, coast = Be Pans) Example. Consider the integral fu which does not have the form given in the Cauchy integral formula. By applying Cauchy’s theorem to the domain D- obtained by excising two small disks centered at 0 and 1 from the disk {|2| < 2}, we can express the integral above as the sum of two integrals, each of which can be evaluated by the Cauchy formula ni(4rcosn) = —4n%. 2ni(e — 2) ‘These integrals can also be handled by residue theory, as we shall soon see. Oo Exercises for IV.4 1. Evaluate the following integrals, using the Cauchy integral formula: n20 © faz, -o0 tesa de . © fuses = TF a: jelar 2°(2? ~ Ader 5. Liowvilles Theorem nt od Sa wf De 2. Show that a harmonic function is C*, that is, a harmonie function has partial derivatives of all orders. ans 202 3. Use the Cauchy integral formula to derive the mean value property of harmonic functions, that o aw ua) = [ulcer net) 2, seed, whenever u(2)s harmonte ina domain D and the closed disk 2 — | Sp is contained in D. 4. Let D be a bounded domain with smooth boundary @D, and let 20 € D. Using the Cauchy integral formula, show that there is a constant C such that LFG20)I $C supflf(2)| : = € OD} for any function f(z) analytic on DU@D. By applying this estimate to fiz)", taking nth roots, and letting n — oc, show that the estimate holds with C = 1.’ Remark. This provides an alternative proof of the maximum principle for analytic functions. 5. Liouville’s Theorem Suppose thet f(z) is analytic on the closed disk {|z — zo < p}, that is, it fs enalytic on some domain containing the closed disk: By the Cauchy integral formula for f(z), mag) = fee FC) = ei Jessen E20) ‘We parametrize the boundary circle by z = 2 + pe", dz = ipet@d#. Then 1 f@) 4, | fet Bri (zz) perm 2? 0) and we obtain 2 teovanenel ES $20) The obvious estimate [rma] < Leo oe now leads to the following version of the Cauchy estimates. 9 aeus IV. Complex Integration and Analyticity ‘Theorem (Cauchy Estimates). Suppose f(z) is analytic for |2~2o| < p- Ife) = M fore ~29| =p, then [po] < Note that this estimate scales correctly with respect to M, in the sense ‘that if we multiply f(2) by a positive constant, the estimate is multiplied by the same constant. It also scales correctly with respect to p, in the sense that if we dilate the disk by a factor ¢ > 0, then both the mth derivative ‘of the dilated function and the factor 1/p" are multiplied by 1/e". ‘The ‘estimate is invariant under translations. In effect it would have sufficed to check the estimate in the special case of analytic functions on the unit disk that are bounded in modulus by one on the unit circle. In this special case the estimate asserts that all derivatives of the function are bounded in modulus by one at the origin. As an application of the Cauchy estimates, we prove the following. ‘Theorem (Liouville’s Theorem). Let {(2) be an analytic function on the complex plane. If f(z) is bounded, then f(z) is constant. Indeed, suppose |f(z)] < M for all € C. We can apply the Cauchy ‘estimate to a disk centered at any 29, of any radius p, to obtain reo s 5 Letting p tend to +00, we obtain f(z) F(z) is constant. ‘We define an entire function to be a function that is analytic on tho entire complex plane. ‘The polynomials aq2" +--+ 12 + a9 are entire functions. ‘The transcendental functions e*, cos 2, sinz, cosh 2, sinh = are ‘also entire. Any linear combination of entire functions is entire, and any product of entire functions is entire. Examples of functions that are not entire are 1/2, log, and In terms of entire functions, Liouville’s theorem has a succinct statement: A bounded entire function is constant [As a test of the strength of Liouville's theorem, we apply it to give yet another proof (one of hundreds of proofs) of the fundamental theorem ‘of algebra, that every polynomial in z of degree n > 1 has a zero. (See Section 1.1.) The proof is by contradiction. Suppose p(z) = 2"-+an—12"" 1+ + ay is a polynomial with no complex root. Then 1/p(2) is an entire function. Since 0. Since this is true for all 29, 6. Morera's Theorem 9 tends to 1 as 2 —+ 00, p(z) ~+ 00 and 1/p(2) +0 as z ~ 20. Consequently, 1/p(2) is bounded. By Liouville’s theorem, 1/p(z) is constant. Since the constant cannot be 0, we have a contradiction. Our supposition is false, and p(z) must have a zero. Exercises for IV.5 1. Show that if w is a harmonic function on C that is bounded above, then u is constant. Hint. Express w as the real part of an analytic function, and exponentiate. 2. Show that if f(z) is an entire function, and there is a nonempty. such that f(2) does not attain any values in the disk, then f(z) is constant, 3. A function /(2) on the complex plane is doubly periodic if there are two periods wy and w of f(z) that do not lie on the same line through the origin (that is, do and w: are linearly independent, over the reals, and f(z +p) = f(z + wi) = f(2) for all complex numbers 2). Prove that the only entire functions that are doubly periodic are the constants. 4, Suppose that f(z) is an entire function such that f(2)/2" is bounded for || > R. Show that f(2) is a polynomial of degree at most n. ‘What can be said if f(2)/2” is bounded on the entire complex plane? 5. Show that if V(2) is the velocity vector field for a fluid flow in the entire complex plane, and if the speed |V(2)| is bounded, then V(z) is a constant flow 6. Morera’s Theorem What Moreis did was to observe that the differential f(z)dz is closed if and only if f(z) is analytic. The following more precise variant of this observation is often referred to ac “Morera’s theorem.” It has a number of useful applications. ‘Theorem (Morera’s Theorem). Let f(z) be a continuous function on a domain D. If Jy, f(2)d2 = 0 for every closed rectangle R contained in D with sides paralel to the coordinate axes, then f(z) is analytic on D. The power of Morera’s theorem resides in the fact that no hypothesis is made concerning the smoothness of f(2). Only continuity is required.120 IV. Complex Integration and Analyticity 2+a2 ~ ‘To prove the theorem, we can assume that D is a disk with center 29 Define FE = [OK red, ‘where the path of integration runs along a horizontal line and then a vertical line, as indicated in the figure. We could as well define F'(2) using the path starting from zo along a vertical ine followed by a horizontal line. ‘The hypothesis guarantees that these two paths yield the same integral, as the difference is the integral of f(2)dz over the boundary of a rectangle. Now wwe differentiate F(s) by hand. We have sea FOe+a2) ~ Fle) =f f0)ac, where the path of integration is the path from = to 2+ Az that follows a horizontal line and then a vertial line, asin the figure. Here we have again used the fact that the integral along the boundary of a rectangle is zero, as indicated in the figure. To deal with the integral on the right, we use the trick of adding and subtracting f(2) from the integrand. Since = is fixed, the value f(2) is constant for the integration, and we obtain Fls+ Az) ~ Fle) were sy [aes [sana = feoass ["U@-Henas Now, the length of the contour from 2 to 2 + Az is at most 2A2|, If we divide by Az and use the ML-estimate on the last integral, we obtain Fle+A2)~ Fe) 7 = fl) < 2M, [Aa H(z). Let R be a closed rectangle in D. By Cauchy’s theorem, we have W{t2)dz = 0. lor Consequently, 6 ff [reads = 0 I, Jon Parameterization of the sides of OR converts the inside integral to a sum of four garden-variety integrals of a continuous function, and we can inter- change the order of integration. This yields o- fff ne2aeas = [Hea ‘The hypotheses of Morera’s theorem are met, and H(z) is analytic. ‘As another typical application, we prove the following useful result. ‘Theorem. Suppose that f(2) is a continuous function on a domain D that is analytic on D\R, that is, on the part of D not lying on the real axis. Then f(z) ie analytic on D Let R be a closed rectangle contained in D, with sides parallel to the coor- dinate axes. To prove the theorem, it suffices to show that ae f(z)dz = 0 ‘There are three cases to consider. Ifthe closed rectangle R does not meet the real axis R, then f(z) is analytic on R, so the integral is zero by ‘Cauchy's theorem, For the second case, suppose that F has one edge that lies on the real axis, say the lower edge of Ris an interval (a,8] on the real axis. For ¢ > 0 small, let Re be the closed rectangle in the upper half-plane consisting of 2 € R such that Im > e. By Cauchy's theorem, Jon, £(2)d2 =0. We claim that (61) Jp. Here f stents12 IV Complex Integration and Analyticity as £0. Indeed the integral along the bottom edge of R, has the form [rerea, and since f(t+ ie) converges to f(t) as © ~ 0, uniformly for a <¢ 0. Remark. An entire function satisfying such a growth restriction is called an entire function of Bnite type. 7. Goursat’s Theorem 18 3. Let A(t) be a continuous function on a subinterval |a,)] of (0,00). Show that the Fourier transform (2), defined as above, is bounded in the lower half-plane. 4. Let -y be a smooth curve in the plane R?, and let D be a domain in the complex plane, and let P(z,y,¢) and Q(z,y,¢) be continuous complex-valued functions defined for (x,y) ony and ¢ € D. Suppose that the functions depend analytically on ¢ for each fixed (z,y) on > Show that 66) = [ Peancrde + Qe. nOdy is analytic on D. 7. Goursat’s Theorem We have defined f(z) to be analytic on D if the complex derivative f"(z) exists at each point of D and further, f’(2) is a continuous function of z. Goursat's theorem asserts that the requirement that j’(z) be continuous is redundant. ‘Theorem (Goursat’s Theorem). If f(2) is a complex-valued function on a domain D such that = tim L2)= Leo) F(20) = jim exists at each point 29 of D, then f(z) is analytic on D, Goursat’s theorem is as useless as it is aesthetically pleasing. In appli- cations it has always turned out that if one can show the existence of a complex derivative at each point, then one can see with little more effort that the complex derivative is continuous. Nevertheless, the idea of the proof has proved to be very useful in other contexts. ‘The proof is based on Morera’s theorem. Let R be a closed rectangle in D, We subdivide R into four equal subrectangles. Since the integral of f(z) around OR is the sum of the integrals of f(z) around the four subrectangles, there is at least one of the subrectangles, call it Ri, for J (2)ds| whieh 1 LP peayae Jory 4 \ft y Now subdivide Ry into four equal subrectangles and repeat the procedure. This yields a nested sequence of rectangles {Fn} such that f,.104 Alf, r«| i Toa. f(z)de124 IV Complex Integration and Analyticity Since the R,'s are decreasing and have diameters tending to 0, the Rx's ‘converge to some point 2 € D. Since f(2) is differentiable at zo, we have ‘an estimate of the form | £02) - Flo) are where 5, — 0 as n — co. Let L be the length of OR. Then the length of Rr, is L/2", For z belonging to R,, we have the estimate #2) ~ F(0) ~ £'(20)(z ~ 20)| < enlz — 20] S 2enL/2". From the ML-estimate and Cauchy’s theorem, we obtain F'0)| < en z€ Rn, F(z) dz | fp) = e0)— sane sas S (enh /2")-(L/2") = 2749/4. Hence [nro < Since én + 0 as n — oo, we must have [feu =o bn By Morera’s theorem, f(z) is analytic. Se)de| < 20%, Ry 1 a pot Exercises for IV.7 1, Find an application for Goursat’s theorem in whieh it is not patently clear by other means that the function in question is analytic 8. Complex Notation and Pompeiu’s Formula Many results in complex analysis can be expressed very simply in terms of the first-order differential operators aot [ aa a a 8 Complex Notation and Pompeiv's Formula 125 ‘We may think of Me as an average of the derivatives of f(z) in the x and the iy directions, of _ ilar , af) a 3% * aan) When we derived the Cauchy-Riemann equations in Chapter IT, we found two expressions for the derivative of an analytic function (2), ey = 2 and fe) = iL = OF fe= 2% ma fe = a = If we take the average ofthese two expressions for f(z), we obtain oF (2) sa=Z. again provided that (2) is anal "To understand the operator 4/02, we write f = u+iv and compute of _ 1fdu ov) | ifau } Lou _ du) , f/du | ov alae ~ ol * alae + a If we equate the real and imaginary parts of the right-hand side to zero, ‘we obtain the Cauchy-Riemann equations for u and v. Thus the equation of oH is equivalent to the Cauchy-Riemann equations for « and v. Equation (8.2) is referred to as the complex form of the Cauchy-Riemann equations ‘We summarize our observations. (82) o ‘Theorem. Let f(z) be a continuously differentiable function on a do- main D. Then f(z) is analytic if and only if f(z) satisfies the complex form (8.2) of the Cauchy-Riemann equations. If f(z) is analytic, then the derivative of f(z) is given by (8.1) We list some rules for operating with 8/02 and 0/82. Since these are both first-order differential operators with constant coefficients, they are linear, tor +b) = Klos + bo) and they satisfy the Leibniz rule, af a oy , 2 Bue = +0, Zum = rot ‘The z-derivative and the Z-derivative are related to each other by af _ of _ OF oe -&126 IV Complex Integration and Analyticity “These formulae can be verted by writing f = u-+iv, f = ui, and carrying out the calculations. The Taylor series expansion of a smooth function f(2) at 2 Ubrough only the ines terms, given by He) = $20) + (zoe — 2) + 2 (20) F=B) + 0 (12 a"), where the oh” term is a remainder term bounded by Cz — zo|?. The complex Tiplor expansion can be derived fom the usual Taylor expansion by substituting the definitions and ealeulating. It ean alo be derived by observing that it suffices to check the formula for the three funtion 1, ia, and =m ‘As tn pplication of Taylor's formula, we drive the complex form ofthe formula forthe tangent vector toa curve Let (2) be a smooth function, and let7(t) be a smooth curve terminating at (0) ~ sy. From the linear Taylor eerie approximation we have seo) -se10y = Zeanire 20)+ 22 (26) (77 =%) +0 (int) — sal?) Dividing by and taking the limit ast 0, we obtain " 8 H ay eT 3) rent) = Zeer + Lea) We use this formula to show that conformal maps are analytic Theorem. Let f(2) be a continuously diferentable funtion on a do- iain D. Suppose that the gradient of j{2) does not vanish at ny point of D, and that f(z) is conformal. Then f(z) is analytic on D, and f(z) # 0 on D. To prove this, fix a point 2» € D, and consider the straight line >(t) = zotte?,0 3) imust be independent of @. However this occurs only when 24/08 = 0 Hence f(:) aalytic on D. ‘The gradient condition tnples thet (2) 5/0: 4 0, andthe theorem i established. Nove that this theoreit sould OF (25)e-28 pe ecnle 8. Complex Notation and Pompeiu's Fermula wr fas well be proved using x and y derivatives rather than z and = derivatives. However, the complex notation makes the proof more transparent For the case of complex line integras, Green's theorem yields the follow- ing formula, which can be regarded ¢s an extension of Cauchy's theorem to arbitrary smooth functions. ‘Theorem. If D is a bounded domait in the complex plane with piecewise smooth boundary, and if g(2) is a smooth function on DUAD, then es [ote = 8 ff Bae "To see this we replace dz by dr + idy and apply Green’s formula: [tes [iote = ff (QE 22) arty = ff ae Note that if g(2) is analytic on D, then 39/02 = 0 on D, so the integral ‘over D vanishes, and we obtain Cauciy’s theorem, Cauchy's integral formula can also be extended to apply to arbitrary smooth functions. ‘Theorem (Pompeiu’s Formula). Suppose D is a bounded domain with piecewise smooth boundary. If g(2) is a smooth complex-valued function on DU@D, then a2) ae ‘ff 7 1 es) ou) = sf, Pompein’s formula is established by the same argument as was used in Section 4 to establish Cauchy's itegal formula, except thet now he correction term appears in the ealeultion. Le D, be the domain obtained from D by punching oun disk centered at w of radius €- We apply the Complex version (84) of Green's theorem to the funtion g(2)/(e--1). For this, note that a(ad) M1 2()- ML ser, so that by (8.4), anf, Shee aff Mtoe ‘The singularity of 1/(2 ~ w) at = =v is absolutely integrable: hoger ate = [U [brea = <0 ydedy, we.128 IV. Complex Integration and Analyticity Hence the area integral in (8.6) over D. tends to the (improper) area inte- gral over D as ¢ 0, The boundary integral in (8.6) has the form TE we parametrize the ctce |x — | = 2, we obtain for the integral on the right [fF ooes eer aa, ‘This tends to 2rig(w) as © —+ 0, since 9(z) is continuous at w. Thus if we let ¢ —» 0 in (8.6), we obtain [2 2nig(w) = aff5 which is equivalent to (8.5). ‘The formula (8.5) is also known as the Cauchy-Green formula, since it is proved using Green's theorem. ‘The formula can be regarded as Cauchy's integral formula with a correction term added to account for the fact that (2) may not be analytic. If g(z) is analytic on D, it reduces to the Cauchy integral formula (4.1) for 9(2). bxercises for IV.8 1, Show from the definition that a a a a7 2 Compute (as? + bes + 22). Use the result to determine where az? +625 + c2? is complex-differentiable and where itis analytic (See Problem IL2.3.) 3. Show that the Jacobian of a smooth function f is given by jor? arp? 4+ = laz| ~|as| 4, Show that P&P ae an * BF ‘202 Deduce the following, for a smooth complex-valued function h. (a) his harmonic if and only if 3%h/202 = (b) fis harmonic if and only if h/02 is analytie. (0) his harmonie if and only if 3b/@2 is conjugate-analytic. Exercises 129 10. (d) If his harmonic, then any mth order partial derivative of h is a linear combination of @™h/02" and d”*h/32". With dz = dr ~idy, show for a smooth function f(z) that Show that if D is a domain with smooth boundary, and if f(z) and g(2) are analytic on DU OD, then [seas = 2 ff piv aca ‘Compare this formula with Exercise 1.4. ". Show that the Taylor series expansion at 2» = Oof asmooth function (2), through the quadratie terms, is given by f= $10) + Leo): + Laz 1 ey + aa 24 &F qs 3] ga? +25 peel? + FaOF| +O (| *) Establish the following version of the chain rule for smooth complex- valued functions w = w(z) and h = Aw) a ah Ow | dh ow pyltow) = Se + Saar a ah ow zltow) = Fe + Saas Show with the aid of the preceding exercise that if both h(w) and tw(z) are analytic, then (ho w)(z) is analytic, and (ho w)'(z) = H(wo(2) (2). Let g(z) be @ continuously differentiable function on the complex plane that is zero outside of some compact set. Show that so) = 2 [2 Remark. If we integrate this formally by parts, we obtain ato) = 2 ff ove 5 (Ay) aren Thus the “distribution derivative” of 1/(n(z ~ w)) with respect to = is the point mass at w (“Dirac delta-function”), in the sense that it is equal to 0 away from w, and it is infinite at w in such a way that its integral (total mass) is equal to 1 drdy, we.Vv Power Series In this chapter we show that the analytic functions are exactly the functions ‘that can be expanded in a convergent power series about any point. Since ower series can be treated very much as polynomials, this provides a Powerful tool for dealing with analytic functions. In Sections 1 and 2 we review infinite series and series of functions. Sections 3 through 6 contain the basic material on power series. In Section 7 we use power series to show that the zeros of an analytic function are isolated, This leads to the uniqueness principle for analytic functions. Section 8 contains a formal definition of analytic continuation, which can be omitted at first reading, 1. Infinite Series In this section we review some basic material on infinite series of complex numbers. In the next section we provide some background material on sequences and series of functions. ‘The reader may wish to skip to Seetion 3 and refer to tae background sections only when necessary. A series SF ga of complex numbers is said to converge to S if the sequence of partial sums {S,}, defined by Sy = a9 +--+ a, converges to S. For notation, the sum 5 of the series is denoted by Soy ax, oF simply by See. Any stateizent concerning series can be reinterpreted as a statement about sequences, by phrasing the statement in terms of the sequence of partial sums ofthe series. For instance, we know that if sy —» s and tn — t, then s,:+ty —~ s+, Ifwe apply this statement to the partial sums of series, we conclude that if Sax = A and Sob, = B, then (ax +b,) = A+B Similarly, since sn + s implies cs, — cs, we deduce that if a, = A, then Seay = cd. Thus taking limits of sequences is a linear operation, and this implies that summing series is also a linear operation. If the ra’s are positive real numbers, then the partial sums Sp = ro + ++ form a monotone increasing sequence. Since a monotone sequence of real numbers converges if and only ifit is bounded, we see that a series of 130 1. Infinite Series 131 positive numbers converges if and only if the partial sums are bounded. If 0 < ax < re, and if the partial sums of Sr are bounded, then the partial sums of yay are bounded by the same bound. ‘This observation leads to the following convergence test for series with positive terms. ‘Theorem (Comparison Test). If0 < ay < ra, and if Sr converges, then Say convergeo, and Sax < Sore ‘The terms of a series can be recovered from the partial sums, by ae = Sk ~ Se-1- Suppose the series Sax converges to S. Then Sq converges to Sas k— 00, and also S,_1 converges to S as k + co. Hence a, +0 as ke + co. This simple necessary condition provides a useful screening test for convergence. ‘Theorem. If So a4 converges, then a, +0 as k — 00. Example. The most important series for us will be the geometric series Dio 2%. The kth partial sum of the geometric series is given by A lezt Pega d 2#L ‘To see this, we multiply and divide the sum by 1 ~ 2, and we note that the ‘numerator telescopes: Se (1-2) = Deere If [2] <2, then #1 0 and S, + 1/(1 = 2) as k + 00. Hence aah gH Tools On the other hand, if |2| > 1, then the kth term 2* does not converge to 0, so that the series does not converge. The series Say is said to converge absolutely if S>|ax| converges ‘Thus for a series of positive terms, convergence and absolure convergence ave (he sane. ‘There are convergent series that are not absolutely eouver- gent. (See Exercises 4 and 5.) However, every absolutely convergent series is convergent. Theorem. If Dax converges absolutely, then Sax converges, and aap The proof is ensy, modulo a little trick, which is to express Reay as a difference Reay = (Reag + lax|) ~lax|- Since |Reax| < Jax|, we have132 V Power Series 2. Sequences and Series of Functions 133 0 < Reax + lax| < 2axl. Hence S-(Reay + lag!) is a series of non negative real numbers. Since its partial sume are bounded, by 25° asl, ns the series converges. Now > Rear is the difference of two convergent. se- HE ries, $(Reag + lag|) and SS lai|, and so SReax converges. Similarly, & Dimas converges, and consequently Sax converges. For each N we have TEM paul < kno aul. Letting N’ ~~ co, we obtain the estimate (1:1). converges. Hint, Show that the partial sums of the series satisfy Sy < Sp S4> 56> 4. Show that the series ye E Example. The geometric series converges absolutely when |2| <1, and eee retlyglylqjiji k<1 +E-gtete-dtata converges to 35/2, where $ is the sum of the series in Exercise 4. (Ie tums out that 5 =log2.) Hint. Organize the terms in the series in Exercise 4 in groups of four, and relate it to the groups of three in the above series. This leads also to a useful estimate for the difference between the partial ‘sums of the geometric series and the full sum, From ls, 1 6 Show that Yay diverges while S -7o ayy converses 7. Show that the series Soa, converges if and only if TE" ay tends to 0 as m,n — oo. Remark. This is the Cauchy criterion for series. pr NA 2. Sequences and Series of Functions 1, (Harmonic Series) Show that i eae Let {f;} be a sequence of complex-valued functions defined on some set E: oa We say that the sequence {f;} converges pointwise on E if for each point & = ben. x € E the sequence of complex numbers {f;(zr)} converges. The limit f(x) ca of {f,(2)} is then a complex-valued function on E. Dedluce that the series 5" does not converge. Hint. Use the esti- mate ar rt he 2. Show that if p < 1, then the series 57 Exercise 1 and the comparison test. 3. Show that if p > 1, then the series SUZ, 1/” converges to S, where bE Example. The sequence of functions f,/x) = 23, 0 < x <1, converges pointwise on the unit interval to the function f(x) defined on the unit a interval by f(x) =0 for 0 400 It turns out that with uniform instead of pointwise convergence the two ‘theorems we desire are valid: (1) a uniform limit of continuods functions js continuous, and (2) an integral of a uniform limit is the limit of the integrals. We state the theorems more precisely. ‘Theorem. Let {;} be a sequence of complex-valued functions defined on a subset E of the complex plane. IFeach f, is continuous on E, and if {fs} converges uniformly to f on E, then f is continuous on E. 2. Sequences and Series of Functions 135, ‘Theorem. Let 7 be a piecewise smooth curve in the complex plane. If {J} is a sequence of continuous complex-valued functions on 7, and if {f,} converges uniformly to f ony, then f) f;(2)d2 converges to J, f(z)dz. ‘The first theorem above on the continuity of a uniform limit has a stan- dard formal proof, which we omit. The second theorem above on the limit of the integrals is an easy consequence of the M/L-estimate, Indeed, sup- pose { f;} converges uniformly to f on 7. Let 2, be the worst-case estimator for fy = f on 7, 30 that |f, — | < ej on 7, and let L be the length of > Then the ML-estimate gives [sere [ reas and this tends to 0, since the J's converge uniformly to f. Hence f fydz tends to f f dz. Now we turn to series of functions. Let $7 9,() be a series of complex- valued functions defined on a set E. The patil sums of the series are the functions < SL, Suz) = YO a9jl2) = golz) + 91(z) +--+ + gaz) fe ‘We say that the series converges pointwise on E if the sequence of partial ‘sums converges pointwise on F, and the series converges uniformly on E if the sequence of partial sums converges uniformly on E. The following criterion for uniform convergence ofa series of functions is extremely useful In fact, itis the only test for uniform convergence of series that we will ever need. ‘Theorem (Weierstrass M-Test). Suppose My > 0 and >My con- verges. Ifge(2) are complex-valued functions on a set E such that |ge(2)| < My for all x € EB, then > 9x(x) converges uniformly on E. The proof is straightforward. For each fixed z, the estimate for gx(2) shows that the series 3° gi(r) is absolutely convergent, and Y>gx(=)] Mj. By the theorem in Section 1, the series 7 g,(2) converges to some complex number g(z), and by (1.1), |g(x)| < 5 lge(x)| <]Me. The same estimate, applied to the tail of the series, shows that =| > ate] so Mi |oc2)- ssc If we set ey = EEanyy Mu, then &y — 0 as n —> 00, and the estimate shows that the partial sums 5, (2) converge uniformly on E to 9(z)136 V. Power Series Example. For the geometric series which tends to +00 as 2 — 1. Hence the partial sums do not converge uniformly for |2| < 1. However, suppose we fix a radius r <1, Define Mx = 7. Then Jo Me converges, and |2*| < Ms for |2| 1, the sequence of mth derivatives {4 (2)} converges uniformly to f'")(2) for |z ~ 20] s—r, and so MO=I0| «to (-2) (s—r)er Exercises 137 From the ML-estimate, we obtain, & zo Since py +0 as k + oo, we obtain uniform convergence cf the mth deriva- tives for |z — zo] 1, the sequence of mth derivatives {f{"(=)} converges normally to f)(z) on D. Exercises for V.2 1. Show that fi(z) = 2*/(k-+2*) converges uniformly to 0 on [0, 20) Hint. Determine the worst-case estimator &, by calculus. 2. Show that ge(z) = 2*/(1 + 2*) converges pointwise on (0,00) but not uniformly. What is the limit function? On which subsets of (0.c0) does the sequence converge uniformly? 3. Show that fe(2) = 2*/k converges uniformly for |2] <1. Show that J{(2) does not converge uniformly for |z| < 1. What can be said about the uniform convergence of £,(2)? a 4 Show that > 2," converges uniformly for—s0 < 9 < toe. 5, For which real numbers x does S> toy converge? at ite converges uni- 5 1 6. Show that for each series > ach € > 0, the series > formly for 2 > 14 7. Let an be a bounded sequence of complex numbers. Show that for each ¢ > 0, the series S7*,ann™* converges uniformly for Rez > 1 +2. Here we choose the principal branch of n~=138 V_ Power Series at > converges unlformly for |z| <1 8 Show that 5725 converges uniformly for 2] < 2 9, Show that 55 does not converge uniform for 3 <1 10. Show that if a sequence of functions {fe(z)} converges uniformly on B) for 1 0 such that |z—uw| > 6 for all € D and w € C\D. Show that E can be covered by a finite number of closed disks contained in D. Hint. Consider all closed disks with centers at points (m +ni)6/10 and radius 6/10 that meet E. 12. Let f(z) be analytic on a domain D, and suppose |f(z)| < M for all Z€ D. Show that for each 5 > 0 and m > 1, |f((z)| < miM/6™ for all € D whose distance from 9D is at ieast 5. Use this to show that if {fx(z)} is a sequence of analytic functions on D that converges uniformly to f(=) on D, then foreach m the derivatives (2) converge uniformly to f(z) on each subset of D at a positive distance from 2D. 3. Power Series ‘A power series (centered at 29) is a series of the form Sf ay(z — zo). By making a change of variable w = 2 — 29, we can always reduce to the ‘case of power series centered at z =0. The, main result on convergence of power series is the following, ‘Theorem. Let Taxz* be a power series. Then there is R, 0< R-< +00, such that ax" converges absolutely if |z| < R, and 3.a42* does not converge if |s\ > R. For each fixed r satisfying r < R, the series 5 axz* converges uniformly for |2] a,2*. The radius of convergence depends only on the tail of the series. If we alter a finite 3. Power Series rao) number of coefficients of the series, the radius of convergence remains the For the general case of a power series ST ax(2 ~ zo), the domain of convergence is a disk |z ~ zo| < R. The series diverges if |2 — z| > R, and anything can happen when |z ~ 20] = For the proof of the theorem, note first that if the sequence |ag|r* is bounded for some value r = ro, then it is bounded for all r satisfying O R, then there is a sequence of terms with |ag,|r" — +00. In the borderline case r= R, anything can happen. The sequence lo,|R* might bbe bounded and it might not. If |2| > R, then the terms axz* do not tend to 0, so that the series does not converge. On the other hand, suppose r < R. Choose s such that 1 0. If 2] z* has radius of convergence R = 1 The series does not converge on the boundary circle |z| = 1, since the terms do not tend to 0. Example. The power series > 2*/k® converges uniformly for |2| <1. This follows from the Weierstrass i/-test, with majorants My = 1/k?, On the other hand, ifr > 1, then r*/K? ~» 00 as k +00. Thus the series does not converge for |z| > 1, and the radius of convergence of the series is R= 1 Example. The series, (a) becomes a geometric series if we set w = ~2%/2, The series converges precisely when |w| <1, that is, when |z?| < 2. The radius of convergence is thus = V2. ‘The series converges to 1/(1 ~ w) = 2/(2 = 2).140 V. Power Series Example. The series 57 k¥2* has radius of convergence R = 0. It converges only for = = 0, since B¥r¥ — +00 fog all r > 0. Example. The series JD k-#2* has radius of convergence R = +00. It converges for all z ‘The partial sums of power series are polynomials in 2, and in particular they are analytic functions. From the convergence theorem of Section 2, ‘we obtain the following, ‘Theorem. Suppose Sa,2* is a power series with radius of convergence R > 0, Then the function J) = Doane bl 7 eventve ally, say for k > N. Then jay| > riaxs| for k > N, and lawlr® 2 Jawsilr™*? > Jawsair’*? > Hence the sequence fa,|r* is bounded. From the definition of R we have 1 < R, and since r < Lis arbitrary, we also have L < R. Suppose next that s > L. Then Jax /ay1| N Then lax| < slaxss| for k > N, and lavls™ < Jassals** < laysals*? <2 V_ Power Series Hence the terms ax2* do not converge to 0 for |2| > s, so that the series does not converge, and s > R. Since s > L is arbitrary, we also have 1 R. We conclude that L Example. For the series 5 k2*, the ratio test gives k a oe a1 ane Hence the radius of convergence is R Example, For the series 524, the ratio test gives | _ (e+pt Ost) AE Hence the radius of convergence is R= +00. = k+l $00, ‘The second formula is based on the root test. ‘Theorem. If {/[az) has a limit as k + 00, either finite or +00, then the radius of convergence of Ya,2* is given by (3.3) R= Ir > 1/lim Yfar], then ¢/fag[r > 1 eventually, so that laxir* > 1 ‘eventually, the terms of the series 3 ax2* do not converge to 0 for |2| = r and r > R. On the other hand, if» < 1/lim ¢/fay| then {/ayir <1 even- tually, so that laglr* < 1 eventually, the sequence [ay|r* <1 is bounded, and from the definition of R we have r < R. It follows that (3.3) holds. Example. For the series > kz*, the root test gives R= 1/lim Vk = 1. ‘There is a more general form of the formula (3.3), called the Cauchy- Hadamard formula, that gives the radius of convergence for any power series in terms of a lim sup. Recall (Section II.1) that the lim sup of a sequence {sq} is characterized as the number S, —oo < S < +00, with the property that if ¢ > S, then only finitely many terms of the sequence satisfy s, > t, while if ¢ < S, then infinitely many terms of the sequence satisfy 5, > t. If the sequence sq has a limit, then the lim sup of the sequence coincides with the limit. However, every sequence has a lim sup. The Cauchy-Hadamard formula is obtained simply by replacing the limit in (3.3) by a lim sup, 1 G4) imp Vind Bxercises 143 "The proof is identical to the proof given above, except that the character- izing property of the lim sup is used. Example, We return to the seres (3.1) treated earlier. This ean be ex: pressed as a power series )>axz*, where ay = (—1)*/?/2*/? if k is even, and ay = 0 if kis odd. ‘Ths ¢/far] is 1/2 if kis even and 0 if kis odd ‘The lim sup of this sequence is 1/V2. By the Cauchy-Hadamard formula, the radius of convergence of the series is V2, as before Exercises for V.3 1, Find the radius of convergence of the following power series: ole Olaae © Lime ms & a = t,t = writ — LAE mL toswtret & & & = = a = ast ©) ek © ae OD 2. Determine for which z the following series converge. @ Ye-pF © YL aMe-I™ | Ye o a = we wre" Hy Fe- i 7 a So wat at 3" the radius of convergence of the following series. sade a eet, 0) haste tebe aet eet 4. Show that the function defined by f(z) = > 2" is analytic on the open unit disk {Jz| <1}. Show that [f(P2)| — +00 as 7 1 whenever 2 is a root of unity. Remark. Thus f(2) does not extend ‘analytically to any larger open set than the open unit disk. '5. What functions are represented by the following power series? (@) Skt (b) DE, Reema V_ Power Series 6. Show the series 2 acz*, the differentiated series ST kazz*~!, and the integrated series > p24, = al have te sme radi of eon- vergence. 7. Consider the series Y e+ yy Use the Cauchy-Hadamard formula to find the radius of convergence ofthe series. What happens when the ratio testis applied? Evaluate explicitly the sum of the series 8. Write out a proof of the Cauchy-Hadamard formula (3.4) 4. Power Series Expansion of an Analytic Function We have seen that power series expansions 3 ay(2— 29)" are analytic inside the disk of convergence {2 29| < R}. It is an important and far-reaching fact that conversely, any function analytic on a disk can be expanded in a power series that converges on the disk. ‘Theorem. Suppose that f(z) is analytic for |z ~ 20] 0, ra and where the power series has radius of convergence R > p. For any fixed r,0 p. The formula (4.2) for ay coincides with formula (3.2), which was obtained by differentiating the power series k times and substituting 2 = 29, The estimates (4.4) are another version of the Cauchy estimates already derived in Section IV.A. Recall that they are obtained by applying the M L-estimate to the integral in (43) Di ' M — as . Example. The exponential function e* has power series Sax", where Hence Since e is entire, the adius of convergence of the power series is R= +00. Similarly, the entire functions sin 2 and cos z have power series expansions with infinite radius of convergence, given by (ayisen 2 sing - Ser = 2-54 ; ven ie cose = So OR We state for emphssis two results that are immediate consequences of the theorem. The first is that an analytic function on a disk is completely determined by its value and the values of its derivatives at the center of the disk. This is because the power series representing f(2) is determined by the derivatives of f(z) at the center zp via the formula (4.2) for the power series coefficients.146 V._ Power Series Corollary. Suppose that f(z) and g(2) are analytic for |z ~ 29| 0, then f(z) = 9(s) for |e 20) ax(z ~ 5)* of the function (Log.2)/(z ~ 1) about z = 5 is R= 5. While the function has an apparent singularity at z = 1, in fact it extends analytically to that point. Its power series representation about : = 1 can be obtained by dividing the power series representation of Log by — 1, Log: 1 z-1 1 1 (2-1) + 5(2- 1) == + 2 Since the function tends to +00 as z tends to 0 from along the positive real axis, the function cannot be extended analytically to 2 = 0. Thus 2 = 0 is ‘a genuine singularity of the function, of a type called a “branch point,” and the radius of the largest disk centered at 2 = 5 to which (Log2)/(z— 1) extends analytically is R = 5. Exercises for V4 1, Find the radius of convergence of the power series for the following functions, expanded about the indicated point, 1 (2) Ayr abouts (@) Log, about ¢=14+24, (ze about, ep, about =a, (©) Sanz sbout z= 0, () SE. about = 24 2. Show that the radius of convergence of the power series expansion of (2 = 1)/(22 = 1) about = =2 is VF 3, Find the power series expansion of Log 2 about the point = = i~2. Show that the radius of convergence of the series is R= V5. Explain why this does not contradict the discontinuity of Log z at 2 = ~2.8 10. 11. . Expand Log(1 + iz) and Log(1 — iz) in power series about = V Power Series Suppose f(z) is analytic at z = 0 and satisfies f(2) = 2+ f(2P. ‘What i the radius of convergence of the power series expansion of f(z) about z = 0? . Deduce the identity e!* = cosz + isinz from the power series expansions. Find the power series expansions of cosh z and sinh z about > What are the radii of convergence of the series? . Find the power series expansion of the principal branch Tan“!(2) of the inverse tangent function about 2 = 0. What is the radius of convergence of the series? Hint. Find it by integrating its derivative (a geometric series) term by term, 0. By comparing power series expansions (see the preceding exercise), establish the identity Lt ie Im Tan™ (See Exercise 5 in Section L8.) |. Let a be real, and consider the branch of * that is real and positive fon (0,00). Expand 2* in a power series about » = 1. What is the radius of convergence ofthe series? Write down the series explicit Recall that for a complex number a, the binomial coefficient ‘a: choose n” is defined by ()=5 and (2) =e ne. Find the radius of convergence of the binomial series ‘Show that the binomial series represents the principal branch of the function (1+ z)*. For which a does the binomial series reduce toa polynomial? For fixed n > 0, define the function Jq(2) by the power series (ant 2 Inte) =D mat IE 5. Power Series Expansion at Infinity 149 Show that Jn(2) isan entire function. Show that w = Jn(2) satisfies the ditterential equation votes Remark. This is Bessel’s differential equation, and Jq(2) is Bessel's function of order n. 12, Suppose that the analytic function f(z) has power series expansion Yan". Show that if f(2) is an even function, then a, = 0 for n ‘odd: Show that if f(z) is an odd funetion, then 2, = 0 for n even. 13, Prove the following version of L’Hospital’s rule. If f(2) and g(2) are analytic, f(20) = 920) =0, and g(2) is not identically zero, then 22) < tim £2 A 9G) ~ 8 9G)" in the sense that either both limits are finite and equal, or both limits are infinite 14, Let f be a continuous function on the unit circle 1’ = {\z| = 1}. Show that f can be approximated uniformly on T by a sequence of polynomials in 2 if and only if f has an extension F that is continuous on the closed disk {\z| < 1} and analytic on the interior {|2| <1). Hint. To approximate such an F, consider dilates F,(2) = Fr). 5. Power Series Expansion at Infinity We say that the function f(z) is analytic at 2 = co if the function g(w) = ‘{(1/1w) is analytic at w = 0. Thus we make a change of variable w = 1/2, = = L/w, and we study the behavior of f(z) at 2 = oo by studying the behavior of a(w) at w = 0. If f(z) is analytic at oo, then g(w) = f(1/w) has a power series expansion centered at w = 0, 9) = Donut = byt biwt bu? burt, ful 1/p, and for any r > 1/p it converges uniformly for |2] > r150 V_ Power Series + converges uniformly convertes ~ don't kw diverges A formula for the coefficients can be obtained by multiplying the series by 2” and integrating term by term around the circle |z| = r. We have Ji fee f_(Daet) ae hf otde = aetna Thus the coefficient by of 1/2* is given by 1 b= yf feet td, ko ‘This formula should not be memorized. It will be superseded by the formula for the coefficients of a Laurent expansion, to be derived in Chapter VI. Example. If n > 0, the function f(z) = 1/2" is analytic at oo, since g(w) = w" is analytic at w = 0. Example. Consider the function f(z) = 1/(2? + 1). In this ease, 1 wv se) = 10) = aa = Tet Since g(w) is analytic at w = 0, f(2) is analytic at oo. The power series for ‘a(w) is obtained by expanding 1/(1 + w2) in a geometric series, to obtain g(w) = w? S(-1)hw?* + twist. ‘This expansion can also be obtained by expressing, 1 aa Tee > Five and expanding 1/(1 + 1/2) in a geometric series, 6. Manipulation of Power Series 11 Exercises for V.5 1. Expand the following functions in power series about co: 1 area je (@) zsinh(1/2) (@) 2. Suppose f(z) is analytic at oo, with series expansion (6.1). With ‘the notation f(c) = by and j"(o0) = b1, show that, $"(oo) = Jim, zif(2) — f(0e)} 3. Suppose f(2) is analytic at oo, with series expansion (5.1). Let 2 > 0 be the smallest number such that /(2) extends to be analytic for |2| > 0. Show that the series (5.1) converges absolutely for |2| > and diverges for |2| <<. 4, Let E be a bounded subset of the complex plane C over which area integrals can be defined, and set sou) = [ff S%, weee, where z = r+ iy. Show that f(w) is analytic at oo, and find 1 formula for the coefficients of the power series of f(w) at oo in descending powers of w. Hint. Use a geometric series expansion. Determine explicitly the function f(w) defined in Exercise 4, in the case that E = {|u| < 1} is the unit disk. Hint. There are two formulae for f(w), one valid for |w| > 1 and the other for |w| <1 Be sure they agree for |u| = 1. 6. Manipulation of Power Series Power series are easy to work with. For all practical purposes, power series can be treated like polynomials. We have already seen that power series can be differentiated term by term and that they can be integrated term by term. Power series can also be added and multiplied, just like polynomials. Suppose, for instance, that f(z) and g(2) are analytic at 0, with power series representations Je) = Yoanss, ale) = Seine ‘Then the power series of the sum f(z) + (2) is obtained by simply adding coefficients, $12) +92) = Solan + bedet. &152 V_ Power Series If is a complex constant, tae power series of ef(2) is obtained by multi- plying coefficients by c, els) = Yronzt ‘The formula for the coefficients of the power series of the product f(2)9(2) is more complicated, though it is the same as for products of polynomials. Ie Seal2) = Des, ca) then the coefficients cy are given by (61) ayby + aya t+ ardk-1 tag, 20. This can be justified as follows. The partial sums fa(2) = SC7_9 4)2/ and Ga(2) = Sho ax2* are sequences of polynomials that converge uniformly to f(z) and g(2), respectively, in some disk centered at 0 as n — oo. Conse ‘quently, fa(z)g_(2) converges uniformly to f(2)9(2) on the disk. It follows (Section 2) that the derivatives of f,(2)gq(2) converge to the corresponding, derivatives of f(z)g(z), and by formula (4.2), the coefficient of 2* in the polynomial fa(2)gn(z) converges to the power series coefficient of 2* for J(z)g(z). Now the coefficients of fa(2)gq(z) are obtained by multiplying the polynomials and gathering terins, Sul2)dn(2) = (+) (&--) = aGbo + (arby + aobi)= ++ +(anbo + an-abs +--+ aobn)2" + O(2"*"), where here the notation O(2") is used for terms involving powers 2* for k > m. (This is consistent vith our earlier use of the “big-oh” notation. See Exercise VI2.6.) For k m: 7a) + OF) ee +0) + (5 Fsow ) +(e) = 1h dats oe), woe sinz ob » la, 54 6 Se = (2-218 coun) (1s bey Sots 008) 1a, 2.5 7) = a4 gee Ge +(e), ‘The end result can be checked by differentiating tan z five times. Note that tan z is an odd function, so that only odd terms appear in the power Exercises for V.6 1. Calculate the terms through order seven of the power series expan- about 2 = 0 of the function 1/cos2. 2. Calculate the terms through order five of the power series expansion about 2 =0 of the function 2/sin=. 3. Show that e Lats, 34 Ms, Tee 49? 737 t 8" 50154 V_ Power Series Show that the general term of the power series is given by an = [EF " » nee What is the radius of convergence of the series? 4. Define the Bernoulli numbers B,, by Sootle/2) = 1- Bi - Bi Explain why there are no odd terms in this series. What is the radius of convergence of the series? Find the first five Bernoulli numbers. 5. Define the Buler numbers E, by war 7 at = What is the radius of convergence of the series? Show that Ey =0 for n odd. Find the first four nonzero Euler numbers. 6. Show that the coefficients of a power series “depend continuously” ‘on the funetion they represent, in the following sense. If {Jm(2)} is a sequence of analytic functions that converges uniformly to f(2) for jz| 0, we have ay.m — ay a8 m — oo. 7. The Zeros of an Analytic Function Let f(z) be analytic at zo, and suppose that f(z9) = 0 but f(z) is not identically zero. We say that (2) has a zero of order N at 29 if f(2o $20) = = $"M (a9) = 0, while f""(z9) # 0. In view of the for- ‘mula (4.2) for the power series coefficients, this occurs if and only if the power series expansion of f(z) has the form S(2) = an (2 ~ 20)" + angale — 20) too where ay #0. We can factor out the term (2— 29)" from the power series and write (7a) F(z) = (2-20) A(z), where h(z) is analytic at zp and A(z) = ay # 0. Conversely, if there is 1 factorization (7.1) where h(2) is analytic at 20 and h(20) # 0, then the 7. The Zeros of an Analytic Function 155 leading term in the power series for f(z) is h(z9)(z ~ 20)", and f(z) has a zero of order N at 2. ‘A zero of order one is called a simple zero, and a zero of order two is called a double zero. Example. The zeros of sinz are at the points nt, 00 0 such that |z ~ 29] > p for all points z € E other than 29. In ‘other words, zo is an isolated point of E if zo is at a positive distance from E\{z0}. If E is a set such that each point of B is an isolated point of E, wwe say that the “points of E sre isolated.”156 V__ Power Series Example. If E = |-1,0]U{1/n : n> 1}, then each of the points 1/n is an isolated point of E, while no point of the interval {—1,0] is an isolated point of E. 1 -! ° 2 1 nonisolate points ‘ole potas ‘Theorem. If D is a domain, and f(2) is an analytic function on D that is not identically zero, then the zeros of f(2) are isolated. ‘The proof of this theorem breaks into two parts, an observation about the local behavior of an analytic function near a zer0, and a connected- ness argument that depends on D being a domain. We begin with the connectedness argument, which is deceptively subtle. Let U be the set of all = € D such that f(z) = 0 for all m > 0. If z») € U, then the power series expansion f(z) = Lay(z — 2)" has fag = f'®(29)/F! = 0 for all k > 0. Hence f(2) = 0 for = belonging to a disk centered at zp. The points of this disk all belong to U. This shows that U is an open set. On the other hand, if 2 € D\U, then f'*)(29) #0 for some k. Therefore, f)(2) # 0 for z in some disk centered at 2, and this disk is contained in D\U, so D\U is also open. Since D is connected, either U = D or U is empty. If U = D, then f(z) = 0 for all z € D, contrary to our hypothesis. Hence U is empty. ‘Thus we conclude from the connectedness argument that each zero of f(z) has finite order. The closing argument is easier. If zp is a 2er0 of f(z), say of order N, we can factor f(z) = (z— 20)" A(z), where A(z) is analytic at zo and h(2o) # 0. Then for p > 0 sufficiently small, we have h(2) # 0 for |2— 20] < p, and consequently |f(z)| # 0 for 0 < |z—z0| < p. Thus f(z) has distance at least p from any other zero of f(z), and the zeros of j(2) are isolated. By applying the preceding theorem to f(z) -(z), we obtain immediately the following important result, which is also referred to as the identity principle. ‘Theorem (Uniqueness Principle). If f(z) and g(2) are analytic on a domain D, and if f(2) = g(2) for 2 belonging to a set that has a nonisolated point, then f(z) = g(2) for all z € D. Example. Once we know that sinz and cos are entire functions that satisfy sin? +cos?.r = 1 for all real numbers 2, then necessarily sin? = + cos? 2 = 1 for all complex numbers 2. ‘This follows from the uniqueness principle, applied to f(z) = sin? + cos® 2 and 9(2) = 1 Exercises 187 ‘The uniqueness principle has a natural extension to functions of two complex variables, which is sometimes referred to as the principle of per- manence of functional equations, ‘Theorem. Let D be a domain, and let E be a subset of D that has a nonisolated point. Let F(z,w) be a function defined for z, w € D such that F(z,w) is analytic in 2 for each fixed w € D and analytic in w for each fixed z € D. If F(z,w) = 0 whenever z and w both belong to E, then F(2,w) =0 for all 2,w € D. ‘This follows from two applications of the uniqueness principle, one for each variable. First fix 29 € E. Then F(zo,tw) is analytic for w € D and vanishes for w € E. By the uniqueness principle, F(zo,w) = 0 for all we D. Now fix w € D. We have shown that F(zo,w) =0 for all z9 € E. By the uniqueness principle, F(2,1) = 0 for all = € D. Example. As a typical application, we derive the addition formula for the exponential function, assuming that e*!* = e*et for s and t real. The function F(2,w) = e*** — ee" is an entire function of each variable for fixed values of the other variable, and it vanishes when both the variables are real. By the permanence principle, it then vanishes for all values of = and w. Thus e*" = e*e" Exercises for V.7 1, Find the zeros and orders of zeros of the following functions. 241 @sH @en1 wt+s (hy) sinh? 2 + cosh? z (9 #sinz (i) E282 (principal value) 2. Determine which of the functions in the preceding exercise are an- alytic at oo, and determine the orders of any zeros at oo. 3. Show that the zeros of sin z and tan are all simple. 4, Show that cos(z + w) = cos2 cosw —sinz sinw, assuming the corresponding identity for z and w real 5. Show that r ‘where we take the principal branch of the square root. Compare the result to Exercise 1V.3.1, Hint. Show that the integral is analytic evtsretge = Ee, wet, Rez > 0,158, V. Power Series Jn 2 and w, and evaluate it for 2 = z > 0 and w real by making ‘a change of variable and using the known value V for z = 1 and 6. Suppose f(2) is analytic on a domain D and =) € D. Show that if F(z) = 0 for m > 1, then f(z) is constant on D. 7. Show that if u(z,y) is @ harmonic function on a domain D such that all the partial derivatives of u(z,y) vanish at the same point of D, then u(z, y) is constant on D. 8, With the convention that the function that is identically zero has a zero of infinite order at each point, show that if f(=) and g(=) have zeros of order n and m respectively at zo, then f(z) + 9(2) has a zero of order k > min(n,m). Show that strict inequality can occur here, but that equality holds whenever m # n. 9. Show that ifthe analytic function f(2) has a zero of order N at 59, then f(z) = 9(z) for some function g(z) analytic near zo and satisfying o'(zo) # 0. 10. Show that if f(z) is a continuous function on a domain D such that sey" is analytic on D for some integer N, then (2) is analytic on 11. Show that if f(2) is a nonconstant analytic function on a domain D, then the image under f(z) of any open set is open. Remark. This is the open mapping theorem for analytic functions. The proof is easy when f'(z) # 0, since the Jacobian of f(z) coincides with If'(2)P. Use Exercise 9 to deal with the points where "(2) is zero. 12. Show that the open mapping theorem for analytic functions implies the maximum principle for analytic functions. 13, Let fn(2) be a sequence of analytie functions on a domain D such that f(D) C D, and suppose that fa(z) converges normally to f(2) on D. Show that either f(D) cD, or else f(D) consists of a single point in AD. 14, Aset F's discrete if every point of Fis isolated. Show that a closed discrete subset of a domain D either is finite or can be arranged in a sequence {2} that accumulates only on {20} UAD. 8. Analytic Continuation In Chapter I we analyzed the branches of the functions ¥/2 and log = by following the values of the functions along curves in the complex plane. By 8. Analytic Continuation 159 this method we saw that these functions cannot be extended continuously to C\{0}. We constructed Riemann surfaces to which the functions do extend continuously, by defining branches of the functions on separate sheets and pasting these sheets together. In this section we develop more formally the idea of tracking an analytic function along a path, The power series expansion of an analytic function about a point contains complete information about the function near the point. Rather than track only the values of the analytic function, we will track the power series expansicns of the function about points of the path as we move along the path, We begin by observing that the radius of convergence of a power series depends continuously on the center of the ‘expansion. Lemma. Suppose D is a disk, f(2) is analytic on D, and R(21) is the radius of convergence of the power series expansion of f(z) about a point 2 € D. Then (a) [R(a)-R(e)| < |ar-al ee D. We use the characterization of R(z1) as the radius of the largest disk centered at 21 to which f(z) extends analytically. Thus f(z) does not extend analytically to any disk containing {|z ~ 21| < R(z1)}, and conse- quently R(22) < R(2:) + |22 —2:|- Interchanging 2 and za, we obtain also R(z1) < Rlca) + [22 ~ 2i|- These two inequalities yield (8.1). ‘Now we start with a power series J aq(2—z0)" that represents a function. f(z) near 29, We are interested in the behavior of f(z) only near 20, and iwe say that the power series represents the “germ” of f(z) at z0- Let 7(¢), a (a). We say that f(z) is analytically continuable along 7 if for each t there is a convergent power series Vane such that fa(z) is the power series representing f(z) at zo, and such that when s is near t, then f.(2) = fe(2) for = in the intersection of the disks ‘of convergence. By the uniqueness principle, the series f(2) determines uniquely each of the series f,(z) for s near t. Tt follows that the series fo(z) is uniquely determined by fa(2). (Otherwise, we could define ty to bbe the infimum of the parameter values t such that fe(z) is not uniquely determined by fa(2), and we would soon have a contradiction to the local Uniqueness assertion at to.) We refer to fo(z) as the analytic continua- tion of /(z) along 7, where we regard fi(2) either as a power series or as aan analytic function defined near 7(6). Since the coefficients a(t) in (8.2) are given by aq(s) = f{")(7(s))/m! for s near t, the coefficients depend continuously on the parameter t. The preceding lemma shows that the (8.2) OY, z= < nO.160 radius of convergence of the power series (8.2) also depends continuously on the parameter t. We summarize and give some examples, ‘Theorem. Suppose f(z) can be continued analytically along the path >(¢), @ 0 the coefficient an(t) of the series (8.2) depends continuously on t, and the radius of convergence of the series (8.2) depends continuously on t Example. Suppose f(z) is analytic ina domain D. ‘Then f(z) has an analytic continuation along any path in D. Simply define f(z) to be the power series expansion of f(2) about 7(t) Example. Suppose f(z) is the principal branch of /2, with series expan- $e) = 143-9 ets about 2 = 1. Let 7(t) = e, 0 < ¢ < 2n, be the closed path around the unit circle starting at (0) = 1. ‘Then f() has an analytic continuation along 7, and the power series f,(=) in (8.2) is given explicitly by wa ert see file) = ot? (ee 8 Thus the analytic continuation of f(2) aroand the circle is 1 e-1yPte-, 1 Sar(2) = -1-5(2-1) + which is just the other branch of /Z. Now suppose {(z) is analytic at 2, and suppose that 9(0),a (b) along which f(z) has an analytic continuation fy(2). The radius of convergence R(t) of the power series (8.2) varies continuously with ¢. Hence there is 6 > 0 such that R(t) > 6 for all hatch Lemma. Let f, 7, and 6 be as above. Ifa(t), a 0 such that Fy, > 6 for all s and t. We can choose, then, parameter values 0 = 89 < 81 <--- < sq = 1 such that |7,,(t) — 7s, (8) <5 for @. 1-46 for some 8 > 0. Suppose further that the series has radius of convergence R = 1. Siow that f(2) does not extend analytically to any point of the unit circle. Remark. Such a sequence with large gaps between successive nonzero terms is called a lacunary sequence. This result is the Hadamard gap theorem. There is a slick proof. If f(z) extends analytically across consider g(w) = f(w"™(1 + t4}/2), where m is « large integer. Show that the power series for 9(w) has radius of convergence r > 1, and that this implies that the power series of f(z) converges for 1 p and at oc. If we normalize the decompesition s0 that f,(co) = 0, then the decomposition is unique If f(2) is already analytic for |2- zo| <0, the Laurent decomposition becoines the trivial decomposition f(z) = fo(2). with fa(z) = 0. IF f(2) is already analytic for |z ~ zo| > p and vanishes at oo, the Laurent decompo- sition is the trivial decomposition f(z) = fi(2), with fo(2) = 0. 165,168 VI Laurent Series and Isolated Singularities ‘The uniqueness of the decomposition follows from Liouville's theorem by the following argument. Suppose that f(2) = go(z) + 9:(2) is another decomposition with the properties of the theorem. Then, (1.2) 90(2) ~ folz) = filz)-mlz), p< |z- 20) <0 Define (2) to be equal to go(z) ~ fo(2) in the disk {|2 ~ zo] < a}, and equal to fi(2) —9i(2) in the exterior domain {|2~2o| > p}- These domains ‘overlap in the annulus {p < |2~ 29] n is analytic for |z — zo] > r and tends to 0 as z — oo. Thus we obtain the decomposition f(2) = fo(2) + fue) forr < [2 —2a| < s. Technically, this decomposition depends on r and s. However, the uniqueness assertion already established shows that the decomposition i independent ofr and s, so that fo(z) and f,(z) are defined for p < |z — zo| < @ and have the properties ofthe theorem, 1, ‘The Laurent Decomposition 67 Example. The function f(z) = 1/(z ~1)(-2) has three Laurent decom positions centered at 0. One represents the function in the punctured disk {0 < [2] <1}, one in the annulus {1 < |2| <2}, and one in the exterior domain {2 <|2| < co}. Since the function is already analytic in the disk {\2| <1), the decomposition in the punctured disk is given by fo(z) = f(=) and fi(2) =0. Since the function is analytic at oo and vanishes there, its Laurent decomposition with respect to the exterior domain {2 < |z| < so} is given by f(z) = fi(2) and fo(z) = 0. The only nontrivial Laurent de- composition is with respect to the annulus {1 < |z| <2}. To see what it is, we consider the partial fractions decomposition 1 1 1 @=NE-9 > 2 ‘The summand fo(z) = 1/(=—2) is analytic for |2| < 2, and the summand file) = =1/(z — 1) is analytic for [2] > 1 and it vanishes at oo. ‘Thus this Partial fractions decomposition coincides with the Laurent. decomposition ‘with respect to the annulus. Example. The function 1/sinz has a Laurent decomposition with respect to each annulus {na <2] <(n+1)x}, for n =0,1,2,.... We will see how to obtain the Laurent decompositions in the next section, Suppose now f(2) = fo(z) + fi(z) is the Laurent decomposition for a function analytic for p < |z ~ 29] 0168 VI_ Laurent Series and Isolated Singularities ‘This series converges absolutely, and for any r > p it converges uniformly for |z — 29| > r. If we add the two series, we obtain a two-tailed expansion for F(2), (1.3) that converges absolutely. and that converges uniformly for r < |2—zo| $s. "The series (1.3) is called the Laurent series expansion of f(z) with respect to the annulus p< |2 ~ 2ol <0. ‘To obtain a formula for the coefficients in the expansion, we divide f(z) by (= — 2)" in (1.8) and integrate around the circle {|2 — z9| = Tr} Since the series converges uniformly on the circle, we ean interchange the summation and integration. ‘The result is f aye Dale she =e foie sin'de ‘The integral of (2 ~ 20)" is 2vi if m = 1, otherwise zero, so all the terms in the series disappear except one, and the series reduces to 2iaq. Thus a4) a, = f(z) dz, co 0, where ay = Example. The function f(z) = 1/(2 ~ 1)(z ~ 2) can also be expanded in a Laurent series centered at z = 1, convergent in the punctured disk {0<|z~1] <1}, Again we rely upon a geometric series, -Ye-pe e-uea (2-1) -(2-18-(2-)8~ + O r} to which f,(z) ex- tends analytically. Thus the largest open domain on which the full Laurent series (1.3) converges is the largest open annular set centered at: 2 con- taining the annulus {p < |z ~ zo| <0} to which (2) extends analytically. This annular set might extend to 29 oF to co, to be a punctured disk or & full disk, or a punctured complex plane or the full complex plane. Exercise. Consider the Laurent series for f(z) = (2? ~ x®)/sinz that is centered at 0 and that converges for |2| = 1. What is the largest open set ‘on which the series converges? Solution. Since sin z has a simple zero at x, the function (sin2)/(z ~ x) extends to be analytic and nonzero at 2 = x. Hence (2?~72)/ sn z extends to be analytic at z = x. Similarly, it extends to be analytic at 2 = —7 (We say that the singularities at z= +m are “removable.”) The function tends to oo at 2 = 0 and at z = 2x. Thus the largest annular domain containing the circle {|2| = 1} to which the function extends analytically10 ‘VI Laurent Series and Isolated Singularities 2. Isolated Singularities of an Analytic Function im is the punctured disk {0 < |2| < 2}. This is then the largest open set on 7, Show that a harmonic funetion won an annulus {A <|2| < B} has ‘which the series converges ‘unique expansion ulre®) = Sant” cos(n) + S> bar” sin(nd) + logr, se Eo Exercises for VIL which is uniformly convergent on each circle in the annulus. Show 1. Find all posible Laurent expansions centered at 0 ofthe following that foreach r,A' 0, given in al, Exercise Vil, and define J-»(2) = (=1)"Jq(2). For fixed w € C, Hint. Use » decomposition of the form u = Re f + ¢log|z|, where establish the Laurent series expansion {fis analytic on the annulus. fe elesia =o i2-1/2)] = YD Jnlw)z", 0< 21 <00. 2. Isolated Singulari of an Analytic Function . A point 29 is an isolated singularity of f(z) if f(z) is analytic in some From the coeficient formula (1.4), deduce that igularity iti (14), punctured disk {0 < |2—zo| 0 be small. Using the Ml-estimate to estimate the integral in (1.4), we obtain 2m) If n <0, the right-hand side tends to 0 as r —+ 0. We conclude that a, for n <0, and consequently the singularity at 29 is removable. ‘The isolated singularity of f(2) at 2 is defined to be a pole if there is, N > 0 buch that ay 40 but ay = 0 for all k < —N. The integer N is the order of the pole. In this case the Laarent series (2.1) becomes M 2. Isolated Singularities of an Analytic Function 13 is called the principal part of f(2) at the pole zo. ‘The principal part P() coincides with the summand f,(2) in the Laurent decomposition (2) = Jol) +fi(z) given in the preceding section. The bad behavior of f(2) at 29 is incorporated into P(z), in the sense that (=) ~ P(2) is analytic at = A pole of order one is called a simple pole, and a pole of order two is called a double pole. Thus 1/z has a simple pole at = 0, and 1/(2—i)? has a double pole at 2 = i. Theorem. Let zo be an isolated singularity of f(z). Then 29 is a pole of J{=) of order N if and only if f(z) = 9(2)/(z— 20), where g(2) is analytic ‘at 29 and g(29) #0. The proof is straightforward. Suppose that f(2) has a pole of order N at zp, and that f(z) has the above Laurent series. Then the power series ay -+a—y44(229)-+0_n42(2—29)?+--- converges toa function g(2) that is analytic at zo and satisfies (2) = 9(2)/(2 ~ 20)", and further, 9(=o) a_y £0. Conversely, if 9(2) is analytic at 29 and satisfies g(zo) #0, then f(z) = a(z)/(2~20) has Laurent series with leading term g(0)/(z—zo)", So that f(2) has a pole of order NV at zo. ‘Theorem. Let z» be an isolated singularity of f(z). Then zy is a pole of F(z) of order N if and only if 1/f(2) is analytic at 2 and has a zero of order N. Again the proof is easy. Suppose f(2) has a pole of order N at zo. Let 9(2) = (2 — 29)" f(z) be as above. Since a(2) # 0, the function h(z) 1/a(2) is analytic at 29 and satisiesh() # 0. Thus 1/f(2) = (220) h(2) has a zero of order NV at 29. This argument is also reversible. If 1/f(2) hhas a zero of order NV at 29, then 1/(z) = (2 — 20)" A(z) for some analytic function A(z) satisfying h(2) # 0, and then g(z) = 1/h() is analytic and nonzero at 2,80 f(z) = 9(2)/(z — 20) has a pole of order N at Zp. Example. The function 1/sinz has poles at each of the zeros of sin 2 ‘Since the zeros of sin z are simple, they are simple poles for 1/ sin z, Exercise. Consider the Laurent series expansion for 1/sin = that converges on the cirele {|z| ~ 4}. Find the coefficients ap, a1, a2, and as of 1, 1/2, 1/2, and 1/2°, respectively. Determine the largest open set on which the series converges Solution. The only zeros of sin are at the integral multiples of x. These are then the only singularities of 1/sin2, and they are all simple poles ‘The largest open annular set containing the circle and to which 1/sin 2 ‘extends analytically is then the annulus {x <|2| < 2}. This annulus is then the largest open set on which the Laurent series converges. From the174 VI Laurent Series and Isolated Singularities expansion sin z = 7 + O(=3) near z= 0, we see that gh = b+ amine peut 2 =0, end I/sinz — it cay at = =O. Silly, om the ‘expansion sin 2 = —(z — ©) + O((z — x)*) at z = 7, we see that 1 near z= x, and 1/sinz + 1/(2—2) is analytic at z = x. By the same token, 1/sinz + 1/(2 +2) is analytic at 2 = —x. We conclude that if + analytic Ale) then fol) = 1/sinz— fi(z) is analytic for [z| < 2n. Thus 1/sin z = fo(=)+ filz) is the Laurent decomposition of 1/sinz. ‘To obtain the negative powers of the Lanreut expansion, we expand fi(2) in a series of descending powers of z, using the geometric series expansion. This is done most easily by combining the first two summands, vee Al) = Note that all the even powers of z disappear, as they must, since 1/sin zis ‘an odd function. We read off the coefficients a = 1 and ag = 2r? We say that a function f(z) is meromorphic on a domain D if f(2) is analytic on D except possibly at isolated singularities, each of which is, fa pole. Sums and products of meromorphic functions are meromorphic. Quotients of meromorphic functions are meromorphic, provided that the denominator is not identically zero. Note also that if there are infinitely many poles of f(2) in D, then we can arrange them in a sequence that ‘accumulates only at the boundary of D. Otherwise, there would be a point of accumulation in D of the poles of f(2), and this point would not be an isolated singularity of f(z). Example, The function 1/sin is meromorphic on the entire complex plane. As another example, let R(=) be any rational function. We can express R(z) a8 a quotient of polynomials in the form ( eam = @aaye Goa)” where the G's and 2s are all distinct, Evidently, (2) has a zero of order rm; at each ¢, and a pole of order nj at each z). Thus R(2) is meromorphic on the entire complex plane. We add one more useful characterization of poles. 2 Isolated Singularities of an Analytic Function 15 ‘Theorem. Let 29 be an isolated singularity of f(z). Then 2 is a pole if and only if |f(2)| + 00 as 2 —> 29. One direction of the theorem is vial. If f(z) has a pole of order NV at 2p, then 9(2) = (= ~ 0)” f(2) is anslytic and nonzero at 29, so that If) = [z= 201 |g(2)| > 20 as z —+ zo. For the converse, we use Riemann's theorem on removable singularities. Suppose |f(2)| —+ 00 as 2 ~+ 5. Then f(2) #£0 for 2 near zo, 0 that A(2) = 1/f(2) is analytic in some punctured neighborhood of zo, Further, h(2) —- 0 as 2—+ 29. By Riemann’s theorem, h(2) extends to be analytic at 2, and moreover, h(zo) = 0. If N is the order of the zero of ‘n(z) at zo, then f(z) = 1/f(2) has a pole of order N at 29 ‘The isolated singularity of f(2) at 2 is defined to be an essential sin- gularity if, #0 for infinitely many k <0. ‘Thus an isolated singularity that is neither removable nor a pole is declared to be essential Example. The Laurent expansion of e!/* at z =0 is given by veeyety tt yta nett gat Gate. reo. Since infinitely many negative powers of = appear in the expansion, the Isolated singularity at = 0 is essential. That the singularity is essential can also be seen from the behavior of e'/* as = —= 0. Since e!/* —+ +00 as 2 > 0 tends to 0, the singularity is not removable. And siace e¥/*) hhas unit modulus, the modulus of e'/* does not tend to +00 as 2 tends to 0 along the imaginary axis, and the singularity is not a pole. The only remaining possibility is that the singularity is essential At an essential singularity, the values of f(z) cluster towards the extire complex plane. That is the content of the following theorem. ‘Theorem (Casorati-Weierstrass Theorem). Suppose 2 is an essen- tial isolated singularity of f(z). Then for every complex number wo, there is a sequence z —+ 29 such that f(z.) —» wo. We argue the contrapositive, Suppose that there is some complex num- ber up that is not a limit of values of f(2) as above. Then there is some small ¢ > 0 such that |f(z) ~ wo| > ¢ for all 2 near zo. Hence (2) = 1/(f(2) ~ wo) is bounded near zo. By Riemann’s theorem, h(z) has ‘a removable singularity at 29. Hence h(2) = (2 ~ 20) (2) for some N > 0 and some analytic function 9(2) satisfying g(20) # 0. Thus f(z) ~ wi 1/h(z) = (2-29) (1/9(2)), where 1/9(2) is analytic at 29. If N = 0, f(z) extends to be analytic at 29, while if V > 0, f(2) has a pole of order NV at 2p. This establishes the theorem.176 VI_ Laurent Series and Isolated Singularities Later we will prove Picard’s theorem, that if f(2) is an analytie func~ tion with an essential isolated singularity a: zo, then for all complex num bers wo with possibly one exception, there i a Sequence 2, —+ 2p such that S(é) = wo. The function f(2) = e'/*, which omits the value w = 0, shows that we must allow for the exceptional point. Exercises for VI2 1. Find the isolated singularities of the following functions, and de- termine whether they are removable, essential, or poles. Determine the order of any pole, and find the principal part at each pole. (a) 2f(2 1? (@) tanz = SE (@) tor (1-4) 1 Log z (@) #sin(2) wy eS 0 a ( ever 2. Find the radius of convergence of the power series for the following functions, expanded about the indicated point. (@ ZA, about = 34%, (6) Za, about z= () ota: tbows= (a) S555, about 8. Consider the function f(z) = tanz in the annulus {8 < |2| < 4}. Let (2) = fol) + fil2) be the Laurent decomposition of f(z). 80 that fo(2) is analytic for |z| <4, and fi(2) is analytic for |z| > 3 ‘and vanishes at 00. (a) Obtain an explicit expression for fi(z). (b) Write down the series expansion for fi(z), and determine the largest domain on which it converges. (¢) Obtain the coefficients faq, @1, and a of the power series expansion of fo(2). (d) What is the radius of convergence of the power series expansion for f(z)? 4, Suppose (2) is meromorphic on the disk {|2| < s}, with only a finite number of poles in the disk. Show that the Laurent decompesition of f(z) with respect to the annulus {s ~= <|z| < s} has the form J(2)= fol2) + fale), where fi(2) is the sum of the principal parts of 2) its poles 5. By estimating the coefficients of the Laurent series, prove that if 29 is an isolated singularity of f, and if (2—z0) f(z) — 0 as = — zo, then 29 is removable. Give a second proo’ based on Morera’s theorem, 6. Show that if f(2) is continuous on a domain D, and if f(2)* is analytic on D, then (2) is analytic on D. Exercises 7 7. Show that if 29 is an isolated singulacity of f(z), and if (2~20)* f(z) is bounded near 29, then 29 is either removable or a pole of order at most. N. 8. A meromorphic function f at zo is said to have order Nat zo if J(2) = (= — 29)" 92) for some analytic function g at zo such that ‘9(20) #0. The order of the function 0 is defined to be +20. Show that (a) order( fg, 20) = order( f,20) + order(9, 20), (b) order(1/, 29) = —order(f.z0), (c) order(f +9, 20) > min{order(f, 0), order(9, 20)} Show that strict inequality can occur in (c), but that equality holds in (c) whenever f and g have different orders at 29 9. Recall that “f(2) = O(h(2)) as 2 — 2” means that there is a constant C such that |f(2)| < CjA(z)| for z near 29. Show that if 2 is an isolated singularity of an analytic function f(2), and it J(2) = O((2—20)") as = — 2p, then the Laurent cocficents of f(z) are 0 for k < m, that is, the Laurent series of f(z) has the form: H(z) = aag(2 — 20)" + Oma ~ 20) + Remark. This shows that the use of the notation O(2”) in Sec- tion V.6 is consistent 10. Show that if f(2) and g(2) are analytic functions that both have the same order N’> 0 at 29, then tim £2) _ £°%20) 25% gl2) ~ o7(e0) 11. Suppose f(z) = Saxz* is analytic for |2| < R, and suppose that J(2) extends to be meromorphic for |z| < R +, with only one pole z on the circle [2] = R. Show that ax/axy1 — 29 a8 k oo. 12. Show that if 29 is an isolated singularity of f(2) that is not remow- able, then 2p is an essential singularity for ef), 13. Let $ be a sequence converging to a point =) € C, and let f(z) be analytic on some disk centered at 25 except possibly at the points of S and at zo. Show that either f(z) extends to be meromorphic on some neighborhood of zo, or else for any complex number L there isa sequence {wy} such that w, + 2 and f(w;) > L- 14, Suppose w (re) is harmonic on the punctured disk {0 0 is such that r®u(re) — 0 as r + 0. Show that a, = 0 = bn for ns-a178 VI Laurent Series and Isolated Singularities 15, Suppose u(z) is harmonic on the punctured disk {0 < |2| < p} Show that if wie), Jog(1/i21) as 2 0, then u(z) extends to be harmonie at 0. What can you say if you know only that |u(z)| < Clog(1/lz|) for some fixed constant C and 0< |2| 0 such that f(z) is analytic for |2| > R. Thus f(2) has an isolated singularity at oo if and only if g(w) = f(1/w) has an isolated singularity at w = 0. We classify the isolated singularity of f(2) at 00 according to the isolated singularity of g(w) at w =0. Suppose that f(2) has a Laurent series expansion Hz) = YO be2*, | > ‘The singularity of f(z) at co is removable if by = 0 for all k > 0, in which case f(z) is analytic at oo. The singularity of f(z) at oo is essential if by #0 for infinitely many k > 0. For fixed N’ > 1, f(z) has a pole of order N at oo if by #0 while by = 0 for k > N. Suppose f(z) has a pole of order N at oc. The Laurent series expansion of f(z) becomes Fle) = bu bye thet byt ht, (eID R, where by # 0. We define the principal part of f(2) at oo to be the polynomial Plz) = bya® tye be thet by ‘The inclusion ¢f the constant bp in the principal part is a matter of conve- nience. It guarantees that f(2) ~ P(2) is not only analytic at co but also vanishes there. Example. Any polynomial of degree N > 1 has a pole of order 1V at oo. ‘The principal part of the polynomial coincides with the polynomial itself. Example. The function e = 1+2+22/21+--- has an essential singularity at co. 4. Partial Fractions Decomposition 179 Exercises for VIS 1. Consider the functions in Exe:cise 1 of Section 2 above. Determine ‘which have isolated singularities at oo, and classify them. 2, Suppose that f(2) is an entire function that is not a polynomial. What kind of singularity can f(2) have at oo? 3. Show that if f(z) is a noncorstant entire function, then e/(2) has an essential singularity at z = co. 4. Show that each branch of the following functions is meromorphic at co, and obtain the series expansion for each branch at oo. (a) (2? =)? (b) (5 = 1) (Oy 4. Partial Fractions Decomposition Proceeding in analogy with our earlier definition, we say that a functi f(z) is meromorphic on a domain D in the extended complex plane C* if (2) is analytic on D except possibly at isolated singularities, each of which is @ pole. Again, sums and products of meromorphic functions are mero- ‘morphic. Quotients of meromorphic functions are meromorphic, provided that the denominator is not identically zero. ‘Any rational function is meromorphic on the extended complex plane C*, including at oo. We aim to establish the converse. ‘Theorem. A meromorphic function on the extended complex plane C* is rational. ‘To see this, note frst that a meromorphie function f(2) on the extended complex plane can have only a finite number of poles. Otherwise, they would accumulate at a point thet would not be an isolated singularity of f(z). If f(z) is analytic at oo, we define P,,(z) to be the constant function Foo). Otherwise, #(2) has a pole at oo and we define Pu(z) to be the Principal part of f(z) at co. In any event, Pa(2) is a polynomial, ond f(z) — Pao(z) + 0 as 2 = 00. Let 21,... ,2m be the poles of f(z) in the finive complex plane C, and let F(z} be the principal part of f(2) at ze Te has the form AG) = at Gap tt Goa and in particular, P,(z) is analytic at oo and vanishes there. Consider the function 9(2) = f(z) — Prol2)— > Py(2). jt180 VI_ Laurent Series and Isolated Singularities Since f(z) ~ P,(2) is analytic at ze, and each P,(z) is analytic at zx for 3 # k, g(2) is analytic at each 24. Hence g(z) is an entire function, and Further, g(z) — 0 as 2 + co. By Liouville’s theorem, g(+) is identically zero. Thus (4) F(z) = Pool2)+ Do Pi(2), which shows in particular that f(2) is a rational function. The decomposition (4-1) is called the partial fractions decomposition Of the rational function f(z). As a byproduct of the proof we obtain the following, ‘Theorem. Every rational function has a partial fractions decomposition, expressing it as the sum of a polynomial in = and its principal parts at each of its poles in the finite complex plane. Suppose that p(2) and q(2) are polynomials. If the degree of q(2) is strictly less than the degree of p(2), then p(z)/q(2) ~ 0 as 2 —+ oo. Thus (2) = p(2)/a(2) is analytic at co and vanishes there, and the principal part Pao(2) is zero. Formula (4.1) expresses f(2) as the sum of the principal parts at each ofits finite poles. Example. The function 1/(2? ~ 1) is analytic at oo and vanishes there, ‘and it has poles at +1. The partial fractions decomposition is 1 riooid Qe and these summands are the principal parts at +1 and —1, respectively. For arbitrary polynomials p(=) and q(z), we can use the division algo- rithm to find the principal part P,.(z) of p(2)/q(2) at co. The division algorithm is a procedure that produces in a finite number of steps polyno- mials P,,(z) and r(2) that satis (42) plz) € Palz)ale) +r(2), deg r(2) < deg a(2) It proceeds as follows. We assume that 4(2) is 8 monic polynomial of degree rm, #0 that q(z) — 2" +--+. We start with a polynomial p(2) of degree 12> m, say plz) = e92"-+-- For the frst step we kill the top coefficient of (=) by defining py(z) = pz) —co2"-"q(z). Thus py[2) has degree m a," in this annulus, and this yields the exponential series for f(2) ‘Theorem. Suppose f(2) 1s analytic on the half-plane {Im(z) > a}, and f(z) is periodic with period 1. IF f(2) is bounded as Im(z) -+ +00, then (2) can be expanded in an absolutely convergent series of exponentials fl) = ae, mlz) > a. =F ‘The series converges uniformly on any smaller half-plane {Im(z) > 00}, where a9 > a. Inchis case the change of variable w = e2*"* converts f(z) to an analytic function g(1) on the punctured disk 0 < |u| < e~®*°. The hypothesis on f(2) implies that g(w) is bounded as w — 0. By Riemann’s theorem on removable singularities, g(w) extends to be analytic at 0. Hence g(w) has « power series expansion ate) = Vaww', wl p Tf. and un are two different periods of f(2), then wy ~ w is @ nonzero period, so jw, ~we| > p. Since any bounded subset of the complex plane can be covered by a finite number of disks of radius p/2, and each of these ‘contains a: most one period, we conclude that any bounded subset of the complex plane contains only finitely many periods. (‘The periods form a “discrete subgroup” of the complex numbers.) Let L be a straight line through 0 that contains a nonzero period, and Jet wr be a nonzero period on L that is closest to 0. Then there cannot be a period between two consecutive integral multiples kur and (k + Tx of 2, of by subtracting we would obtain a period on L closer to 0 than wr ‘Ths the periods on L are precisely the integral multiples of wy Tt inay occur that all the periods of f(2) lie on the same straight line through 0. Otherwise, from among the periods not on the line L, choose ‘period up that is closest to the line segment [0,0]. We claim that all periods of #(2) have the form mi + nicy for integers m and n. We argue a follows. Let P be the closed parallelogram with vertices 0, w1, w2, and w + Ww ‘Thus P consists of precisely the complex numbers of the form sux + tie where 0 0. Suppose that there are A,C > 0 such that [fiz + iy)| < Ce*™ for y > 0. Show that Fl) = YO ane, oan where the series converges uniformly in each half-plane {y ><}, for fixed © > 0. 5. Suppose that +1 are periods of a nonzero doubly periodie function f(z), and suppose that there are no periods w of f(2) satisfying 0 < |u| < 1. How many periods of f(z) lie on the unit circle?186 VI_ Laurent Series and Isolated Singularities Describe the possibilities, and sketch the set of periods for each possibility. 6. We say that w and wz generate the periods of a doubly periodic function if the periods of the funetion are precisely the complex numbers of the form mu; + nw, where m and n are integers. Show that if ws and wp generate the periods of a doubly periodic function f(z), and if Ar and Az are complex numbers, then 4; and dz generate the periods of f(z) if and only if there is 2x2 matrix A with integer entries and with determinant 1 such that A(w, 02) = (1,32). 7. Let and w» be two complex numbers that do not lie on the samme line through 0. Let k > 3. Show that the series = 1 rng = (i + a) converges uniformly on any bounded subset of the complex plane to a doubly periodic meromorphic function f(z), whose periods are. generated by w and w. Strategy. Show that the number of periods in any annulus {N < |s| < N+ 1} is bounded by CN for some constant C. 6. Fourier Series A complex Fourier series is a two-tailed series of the form (6.1) SO cnet? = «+ te ne te pe Hey tere” + ene? +- Laurent expansions are intimately related to Fourier series. If the Laurent series 4s the Fourier series expansion of f(re'®), regarded as a function of 8. The Foutled coefficients of the expansion are the coefficients cy = ayr ‘Suppose the series (6.1) converges uniformly to a function f (¢ fle) = Yo ge®. 6. Fourier Series 187 ‘We can capture the coefficients of the series by multiplying by the exponen- tial function e~*? and integrating with respect to the probability measure 49/2. The orthogonality relations for the exponential functions, * ngunedd fe 5=k (62) f on {i jth, then yield [serene = 3 og [tet = a. This leads us to define the Fourier coefficients of any piecewise contin- uous function (or any integrable function) f(e") to be (63) a= [seen De and we associate to f(e "e noo ck cye'*? the Fourier series of f (e'®). However, we face now two big problems. Does the Fourier series ‘of f (e") converge? And if so, to what? Example. Define f (c!) to be -1 for —7 <0 <0, and +1 for 0<8< >. ‘The Fourier coeficent op is the average value of fe”), which is 0. If k #0, then usd ed eed a=-f Bs fe Re = ~ oak |, = -Lp-cent- nts = Lp-c-ot = gpg h-(CD*- (t+) = Se] This is 0 if & is even and 2/vik if k is odd. Thus the complex Fourier series of F (€) is 1 Ea lore (Re To(g)ee(e)ers If we combine the terms for =k, we obtain sine functions, and the series becomes (65) (e) ~ $ (snes snan + fan s0 :)188 VI_ Laurent Series and Isolated Singularities, ‘The terms of the series are all zero at. @ = 0 and at @ = +x. We will soon see that the series converges to f (¢"9) for ( < [6 <7. ‘Theorem. If f (¢) is piecewise continuous (or more generally, square- integrable), with Fourier series f (e'®) ~ Sexe", then for m,n > 0 we hhave 5) > tee + [ if (e' 5s oof [irene This is established by writing 1(€%)- > acl = (s(¢8)- ce) (7e5- y ae) =n om ‘multiplying out the product on the right, and integrating both sides from =r tox. Lhe integrals featuring the produets of exponentials for j # & drop out, on account of the orthogonality relations (6.2). What remains oon the right after we integrate is [Lent Z-ra [eyed Lo [Teme E + Dio In view of formula (6.3), we recognize the second and third integrals ap- pearing here as cy and G respectively. Thus the sum is equal to fen? 2-Diar-Liot+ Dik If we cancel the two sums over j and move J2|eal? to the other side, we obtain (6.6). ‘The identity (6,6) shows that the partial sums of the series JD lox|? are bounded. Consequently, the series J” |cx|? converges, and from (6.6) we obtain the following estimate 6. Fourier Series 189 ‘Theorem (Bessel’s Inequality). If f (e) is plecewise continuous (or ‘more generally, square-integrable), with Fourier series f (e®) ~ Sexe'*®, then 67 Yat s fers. we mm From Bessel’s inequality it follows in particular that cy —> 0 as k — +00, With this observation in hand, we are now ready to state and prove our main theorem on pointwise convergence of Fourier series. ‘Theorem. Suppose f (e') is piecewise continuous (or square integrable), with Fourier series fe) ~ S-ege". IF f (e®) is differentiable at 60, then the Fourier series of f (2) converges to f(e™’) at 6 = 6, ne = Siac — tm Se ‘Though Fourier series had been studied intensively for well over a cen- tury, this ingenious proof was discovered only relatively recently, by P. Chernoff in 1980. ‘We consider first the special case in which @ = 0 and e = 1. Define a(e) = [f (e) ~ f(0)] / (e - 1). The differentiability of f (e) at 6 = 0 implies that ge) has a limit as @ + 0. Consequently, 9 (e") is also piecewise continuous. Denote the Fourier coefficients of g(e") by bx, 80 that g (c!) ~ Sobket. Bessel's inequality for 9 (e!) shows that bx —+ 0 as k too. Now we compute the cx’s in terms of the by’s. Since f (e! 9 (c®) (e# — 1) + f(1), we have a= fateryie? near «scp f ‘When we express these integrals as Fourier coefficients of 9 (e#), we obtain = bua — be if k #0, and co = 6-1 ~ by + f(1). Hence the series cx telescopes, and we obtain ino 10 oe Se = 10) + SF bebe) = $0) tomer Bes which tends to f(1) a8 m,n — +00. This proves the theorem when fy = 0. The case when 0p is arbitrary is reduced to the above special case by a change of variable. Consider the function h(e®) = f (e(@*0)) ~ Lace’, which is piecewise continuous and which is differentiable at @190 VI_ Laurent Series and Isolated Singularities The Fourier coefcent ay of h(e# an = [Us otra) med g(otoy rhein fe aot ‘Thus the Fourier series off (eM) evaluated at 0 ~ Oy is S7cxe!* — Pan, which is the same as the Fourier series of h (e!) evaluated at @ = 0. We have shown that the latter converges to h(1) = fe). This completes the proof. Example. Consider the series in (6.5). By the convergence theorem, the series converges to +1 for 0 < 9 < mand to ~1 for —n <0 <0. The ‘convergence theorem does not give any information about the points @ = 0 and @ = |cx| converges, by com- parison with 351/K2. By the Weierstrass M-test the series Saye"? then converges uniformly in 0. By the pointwise convergence theorem proved above, the sum of the series is f (e"). We have proved the following ‘Theorem. Suppose f (e'*) is a twice continuously differentiable function of 8. Then the Fourier series of f(e) converges to J (e'®) uniformly in 8. Example. Define 9(9) = @ — 2n96?, —n < @ < m. We check that a—n) = a(n), 9'(—n) = g(x), and "(—x) = g"(x). Consequently, if we set f (e) = 9(0), we obtain a twice continuously differentiable function of. By the theorem, the Fourier series of f (e"®) converges uniformly in 0. The explicit calculation of the Fourier series is left to the exercises (Bxercise 4). Example. The Fourier series in (6.5) does not converge uniformly, since the sum is not continuous. Exercises for V6 1. Consider the continuous function f(e) = [8], —x < 0-< x. Find the complex Fourier series of f (e) and show that it can be ex- pressed as a cosine series. Sketch the graphs of the first three par- tial sums of the cosine series. Discuss the convergence of the series. Does it converge uniformly? Partial answer. The cosine series is w= E$(oa04 Howie Lente) 2. Let fle") = @, —x < 0 such that [eel < Crl¥! for —o0 < k < ov 12, Using the convergence theorem for Fourier series, prove that every continuous function on the unit circle in the complex plane can be approximated uniformly there by trigonometric polynomials, that is, by nite linear combinations of exponentials e, ~20 < k <0. Strategy. First approximate f (e) by a smooth function. 13, Let D be a domain bounded by a smooth boundary curve of length 2x. We parametrize the boundary of D by are length s, so the boundary is given by a smooth periodic function 7(s), 0 < s < 2r. Let 5> exe"** be the Fourier series of ys). (a) Show that > K2lox|? = 1. Hint. Apply Parseval’s identity to 7/(s) and use /9/(5)| = 1 for a curve parameterized by arc length. (b) Show that the area of D is xD Mlex|?. Hint. Use Exercise IV.L4. (c) Show that the area of D is < x, with equality if and only if Disa disk. Remark. This proves the isoperimetric theorem: Among all smooth closed curves of @ jiven length, the curve that surrounds the largest area is a circle. 14, Show that for any choice of complex numbers 54, —m < k oo. Hence which is (2.1). "The same technique can be used to evaluate integrals of the form Se 22) laa where P(z) and Q(z) are polynomials, and Q(z) has no zeros on the real axis, For convergence of the integral, we require that. (23) deg Q(z) > deg P(z) + 2. i 2 .e boundary ‘The integral is evaluated by integrating P(z)/Q(z) around the bot cof a half-disk in the upper half-plane, as above, and letting the radius tend to co. This yields the formula °° Plz) PQ) (24) Jaa - Dee [GG s) “Summed over the poles z) of P(z)/Q(z) in the upper half-plane. 2. Integrals Featuring Rational Functions 2o1 The same contour can be used to evaluate the integrals of rational func- tions times trigonometric functions. The typical integral has the form = P(r) [aay ene where the polynomials P(z) and Q(z) have real coefficients and satisfy (2.3). To use the semicircular contour, we cannot use the function cos(az), because it behaves too badly in the upper half-plane. (As we have seen, the cosine function is essentially a hyperbolic cosine on the imaginary axis, which grows exponentially fast.) To obtain a tractable integral, we resort to a trick. ‘The trick is to substitute e* for the cosine function in the contour integral, and to recover the cosine integral at the end by taking real parts. Since |e***)| — e-¥, the exponential function e is bounded by 1 in the upper half-plane, le] <1, Im(z)>0. Example. We show by contour integration that (3) [22a on a>0. T+a Again we let Dp be the half-disk in the upper half-plane bounded by the interval [—R, R] on the real axis and the semicircular contour I'g of radius R in the upper half-plane. This t'me we integrate the function e'*/(1 + 22) cover the boundary of Dp. The function has only one pole in the upper half-plane, a simple pole at i, with residue calculated by Rule 3 to be ‘Thus Since [e| <1 in the upper half plane the ML-estimate yields I,.ies4l § Asin we have [fe - fapg 1+? Passing to the limit as R — oo, we obtain202 VIL The Residue Caleulus Now we take the real parts of the integral, and we obtain (2.5). Note that if we take the imaginary part of the integral, we obtain [age - ‘which is no surprise, since the integrand is an odd function, Exercises for VIL2 1. Show using residue theory that ode [ixta-b oo Remark, Check the result by evaluating the integral directly, using ‘the arctangent function, 2. Show using residue theory that leeton - Remark. Check the result by differentiating the formula in the pre- ceding exercise with respect to the parameter, 3, Show using residue theory that ade on |. (@ +I? 2 Remark. Check the result by combining the preceding two exercises. 4 Vtg ste ths tht [AE = 6, Ung raie tonya that [EG ae = sg 6, show tt [" ora area = i 7. Show that i et (c= « 8. Show that [tape oe" 3 Integrals of Trigonometric Functions 203 9. Show that [a 10. Show that ™ cos(ax) evlalb | FHP ¢ 00 1. Show that [lage - Ee [35 -5) summed over the zeros z, of Q(2) in the upper half-plane. 3. Integrals of ‘Trigonometrie Functions ‘We have seer how complex contour integrals along a curve can be converted to garden-variety integrals by parameterizing the curve. Some definite integrals can be evaluated through the reverse process, by converting them to complex contour integrals and using the residue theorem. ‘To illustrate this, we show how the integral (3.1) [ = . a>l, fp a+cosd can be evaluated using residue calculus. The iden is to convert this to a ‘complex contour integral around the unit circle. The usual parameteriza- tion z = ef gives dz = ie¥?d9 = iz dB, and we have205 VIL The Residue Calculus ‘The trigonometric functions are easily expressible in terms of z on the unit circle. For cos@ and sin@ on the unit circle we have eee staf eee 9 ei =1z snd = a = If we substitute the expression for cos# into the integral (3.1), we obtain cr Hof 1 te tf a: fp atcsd Siyyargetijs) iz ify? +2az ‘The poles of the integrand are the two zeros of 2? +2az +1, which are =a + Va? =I. Only one of these roots is inside the unit circle, at 29 = . a+ cosd 4, Show using residue theory that Lae - LT sin ~ 5. Show using residue theory that oe [LT Brest ae ~ b OST <1 Remark. The integrand is the Poisson kernel (Section X.1). 6. By expanding both sides of the identity af wo 4 Defy wreoss ~ FaF=1 in a power series at 00, show that Df coat (omy a ff coed = a, ko 7. Show using residue theory that a do 2nw Treat = Gaye WECM Specify carefully the branch of the power function. Check h 1e power function your answer by differentiating the integral of 1/(w + cos) with respect to the parameter w,206 VIL The Residue Caleulus 4, Integrands with Branch Points wy [pty git icons suas 0 Wey ttaee 0 We consider the branch of the function 2°/(1 + 2)? defined on the slit pie hy = (42) fe) = Gey We regard the slit [0, +00) as having a top edge and a bottom edge, and we extend the function by continuity to each edge of the slit, soit is defined by (4.2) with @ = 0 on the top edge, and by (4.2) with @= 2x on the bottom edge. The values on the bottom edge are obtained from those on the top edge by multiplying by the phase factor e?* pare”, 0<0< 2 seyhote contour For ¢ > 0 small and R > 0 large, we consider the keyhole domain D consisting of 2 in the slit plane C\[0,00) satisfying © < |2| < R. The function f(2) has one pole in D, a double pole at z = —1. For the residue, Rule 2 gives nee] = ‘The residue theorem yields (43) [, fle)de = —2niae™ The integral around AD breaks into the sum of four integrals, from ¢ to R. “along the top edge of the slit, around the circular contour Pz of radius R Exercises 207 in the counterclockwise direction, from R back to ¢ along the bottom edge: of the slit, and around the circular contour 7. of radius € in the clockwise direction, Rye < goen [ wept + [toes f Tot + [soe For the integrals over Tye and 7., the ML-estimate gives [wiaee| < pipe Since -1 2, show by integrating around the keyhole contour that St rales t)(atm—2) a = Gna ising) 7 mS OSE Remark. The result can be obtained also by integrating the formula in Exercise 1 by parts. (log 2)? By integrating a branch of ny around the eyhole con- tour, show that = log 2 (loga)? = (log dy? L fraery = a= mbm aFd Using residue theory, show that ™ stlog sy, = Senlra)—asPew(re) 1 og f wea = sin™(ra) , ci 0 small, we consider the integral [, seva. where C, is the arc-of the circle {|z ~ zo] = ¢} subtended by a sector of aperture a. If a = 2r, then C. is the full circle, and the integeal is equal to 2ni Res{f(2),2|. In general, there is no exact formula for the integral. However, if the isolated singularity of f(2) is a simple pole, then we can calculate the limit of the integrals as e — O. Le ‘Theorem (Fractional Residue Theorem). If zy is a simple pole of F(z), and C; is an are of the circle {\z ~ 29] = ¢} of angle a, then (1) ling i $(2)dz = ai Res| f(z), 20] ‘Tosee this, express f(z) = A/(2—20)+g(2), where Ais the residue of f(2) at zo and g(2) is analytic at 29. Parameterizing the circle by 2 = zo +2e", and supposing 6 < @ <6 +a on the ate, we calculate Ad: lo. Since g(z) is bounded near zp and the length of C, is at most 2r¢, the MLcestimate shows that Jo, g(z)de tends to 0-as ¢ — 0, Consequently, Jc, fle)dz tends to ai a3 <0, which is (5.1). = 14 [ae = oa by210 VIL The Residue Calculus Note that the angle a is taken to be negative if the arc of the circle is traversed in the negative direction, that is, in the clockwise direction. To illustrate how the fractional residue theorem is used, we show that (52) We choose the branch of log that is real on the positive real axis (the principal branch of log z), and we integrate f(2) = (log 2)/(2? 1) around the boundary aD of a domain D that isa haif-disk of radius R in the upper half-plane, with indentations of radii ¢ and 6 at the singularities of f(z) at —1 and 0, respectively. Since the zeros of log and 2? 1 at = 1 are both simple, they cancel each other out, so that f(2) is analytic at 2 = 1 and no indentation is required there. Since f(z) is analytic on D, Cauchy's theorem yields (63) I, fede Now the integral around 3D breaks into the sum of six integrals, from 6 to R along the positive real axis, around tke semicircular contour I of radius R in the counterclockwise direction, from —R to —1—¢ along the negative real axis, around a semicircular contour C; of radius ¢ in the clockwise direction around -1, from -1 + to ~6 along the negative real axis, and finally around a semicircular contour 75 of radius 6 in the clockwise direction around 0. For the integrals over I'g and 7s, the ML-estimate gives Vio RE py _ loeR Bayh ~ Re Jog 5? <8 6 ~ Blog. Exercises an If we let R00 and 5 —+ 0, these two contributions disappear, and (5.3) becomes og lal + i oe fo “ear loge + mi Ms: , +f Gites [F . ‘The integral around C; is handled by the fractional residue formula. The angle a is ~n, since C, is traversed halfway around the circle in the negative (5.4) direction. The function f(z) has a simple pole at z = —1, with residue computed by Rule 3 to be logz log z| _ix wa] - | fle) = -2 ‘Thus the factional residue formula (5:1) becomes log ix inf, Bede = m8) = ‘If we take the real part of (5.4) and let ¢ — 0, we then obtain logs log gp , 2-1" +[% zt - > = 0 After a change in variable x -+ —z in the second integral, this becomes (5.2). Exercises for VIL5 1. Use the keyhole contour indented on the lower edge of the axis at = 1 to show that © logr an? f Bey = Toeesltray? OS 25h 2. Show using residue theory that © sin(az) ne Ce se- (=e), a>. Hint. Replace sin(az) by e**, and integrate around the boundary of a half-disk indented at 2 =0 3. Show using residue theory that [r qitithegte 2 av | co HUH? — atzA) 4 4. Show using residue theory that [Stea2 VIL The Residue Caleulus 5. By integrating (e*%* ~ 1)/2? over appropriate indented contours and using Cauchy's theorem, show that 6, By integrating a branch of (log 2)/(z* — 1) around the boundary of an indented sector of aperture 27/3, show that f log a, = 4 lo z a Remark. See also Exercise 4.10. 6, Principal Values An integral f° f(2)dr is absolutely convergent if the (proper or im- proper) integral | f(x)|dz is finite, The integral is absolutely divergent if JP|f(2)\de = +00. There is essentially only one way to assign a value to’an absolutely convergent integral, while there may not be an obvious way to assign a value to an absolutely divergent integral. This is analogous to the dichotomy between absolutely and conditionally convergent series. Every rearrangement of an absolutely convergent series converges to the same value, while the rearrangements of a conditionally convergent series can converge to just about anything Example. The integral is absolutely divergent, since [7 (1/\r|)dr = +00. One natural way to as- sign a value to the integral is to take a limit as ¢ — 0 of integrals over the two intervals [-1,~¢) and |e, +1}, which ate obtained by excising a sym- metric interval centered at the singularity 0, ‘The negative contributions for x <0 cancel out the positive contributions for z > 0, and we obtain what is called the “principal value” of the integral, ig See sia Observe, though, that if we excise an asymmetric interval, say frot to ce, and take a limit, instead of 0 we obtain ra( [= [i= me which can be any value at all, 6. Principal Values 213 Suppose that f(x) is continuous for a <2,< zy and for 2 <2 0 small and R > 0 large, we consider the indented half-disk D in the upper half-plane, consisting of in the upper half-plane satisfying |z| < R and |z ~ 1] ><. The function f(z) has one pole in D, a simple pole at z = e?"*/%. For the residue, Rule 4 gives eonilt he rsd thee els 9 [fo - mm 5: The integral around OD breaks into the sum of four integrals, from ~R to 1 ¢, around a semicircular contour C. centered at 1 of radius = in the clockwise direction, from 1+¢ to R, and around the semicircular contour Ppa4 VIL_ The Residue Caleulus of radius R in the counterclockwise direction, wy [atte (ff. fe no ‘The ML-estimate gives {one Spa ~ me which tends to 0 as R + oo. We combine (6.2) and (6.3), and we pass to the limit as R + co, to obtain (fo + fee f. From Rule 4, the residue of f(2) at 2 =1 is If we pass to the limit in (6.3) as © + 0, we then obtain ~ 1 e : wf gigte- He -R oh If we add i/3 to both sides, we obtain our asserted identity. Tr a 7 R Principal value integrals are important in harmonic analysis. One of the most important integral operators in analysis is the singular integral ‘operator called the Hilbert transform, defined by (Hut) = pv fo? BO ds, -20 2 and ay < ag < +++ < am. By integrating around the boundary of an indented half-disk in the upper half-plane, show that 0 1 one can integrate a branch of 2°1/(z* ~ 1) around fa sector of aperture 2r/b, indented at z= 1 and at 2 = e%*¥/6, 6. By integrating a branch of (log 2)/(2*—1) around an indented sector of aperture 27/b, show that for b > 1, loge = param YL 7. Suppose that P(2) and Q(2) are polynomials, deg Q(2) deg P(2) + 2, and the zeros of Q(z) on the real axis are all simple, Show that w fae = Ee 3G-5] Dee [g5- 4) dz = ~F cot(x/b)216 VIL The Residue Calculus ‘summed over the poles 2; of P(z)/Q(2) in the open upper half-plane and the poles xy of P(z)/Q(z) on the real axis. Remark. In other words, the principal value of the integral is 2ri times the sum of the residues in the upper half-plane, where we count the poles on the real axis as being half in and half out of the upper half-plane. 7. Jordan's Lemma Another class of absolutely divergent integrals is made up of integrals of the form PQ) inzdz, — [” P@ cosede, © Oz) co Oz) where P(2) and Q(z) are polynomials satisfying deg Q(z) = deg Plz) + 1 ‘To evaluate such an integral we would replace the sine or cosine function by et and integrate over the boundary of the usual half-disk in the upper half-plane. If deg Q(z) > deg P(z) +2, then the ML-estimate shows that the integral over the semicircular piece Ip of the boundary tends to 0 as R— oo. If deg Q(z) = deg P(z) +1, the integral over Pg still tends to 0, but this is not so obvious. It is a consequence of the following estimate. Lemma (Jordan’s Lemma). If, is the semicircular contour 2(8) Re, 0<0 2/r, 0<0 oo, in the limit we obtain (7.4) Exercises for VILT 1. Show that [itil = +0 Hint: Show that the area under the mth arch of | sin z|/2 is ~ 1/m. 2 Show that ® sine a am, | pete ~ 3° 3, Evaluate the limits ssin(az) tim fae, 00 R for some R. The residue theorem is valid also for exterior domains, though the residue formula must take into account the point at co. ‘Theorem. Let D bean exterior domain with piecewise smooth boundary. Suppose that f(z) is analytic on DU AD, except for a finite number of isolated singularities =1,-.. ,2m in D, and let a_ be the coefficient of 1/2 in the Laurent expansion {(2) = >a,2* that converges for |z| > R. Then (e1) [to = -2niay + xi) Res|f(2),2)- Dy =a @ iF Here the boundary OD is oriented as usual, so that D lies on its left as @D is traversed in the positive direction. The formula is proved by applying ‘the usual residue theorem to the bounded domain Dp consisting of = € D such that |2| < R. This yields [toes [toe = [soa - ari Res f(2),33] If we substitute the Laurent series expansion for f(z) into the integral over the circle {|2| = R} and integrate term by term, we obtain 2ria_ for the integral. This is simpy the integral formula for the Laurent coeficient a1. In any event, we move this summand to the right-hand side, picking up & minus sign, and we obtain (8.1).20 VIL The Residue Calculus Example. We show using contour Integration that eT fee = Lyars Consider the function 1/,/2(1 ==), which has two analytic branches on the slit plane C\(0,1}. The branches are analytic at oo, and the Laurent expansions are obtained in terms of a binomial series: gaa 30-3)” st b + (-1/2\(-1/2) + sfbe dere] mn 1a) beth rf etn hat psc heap ef Ee ieee nip rae ar ee sc oaameras eee ee esas eee ee ce eae aera weasel aches er ae eel fia ec poe agua ee TT ean ae ee ps ee eres eee eee Res | f(z), 0] = -a We integrate f(z) around the dogbone contour [', (or dumbbell con- tour), traversing the top edge of the slit from © to 1—«, a circle 7, centered at 1 of radius e, the bottom edge of the slit from 1 © to, and a circle C- centered at 0 of radius ©. If we apply the residue formula (8.1) to the exterior domain lying outside the dogbone contour, we obtain (82) ff Sle)de = BHI Res[f(2), 00] = 2ri(-1) = 28 ‘The integrals over the circles 7, and C, are bounded by the ML-estimate IL flz)ds| < 2.2ne ~ VE, fo | a ew VE. [ sea] < Zp-2e ~ ve Since the values of f(2) on the bottom edge of the slit are minus those on the top edge, while the direction is reversed, the integral along the bottom ‘edge is equal to the integral along the top edge. Thus we obtain a fists ~ 2 rahe de 8, Exterior Domains 2a as © 0. If we combine this with (8.2), we obtain the desired identity. c r, ogbone contour Note that it was not necessary to determine the choice of sign in the Laurent expansion of f(z). If we had carried along the + sign, we would have arrived at a value for the integral of x. Since the integrand is, positive, the value + must be the correct choice. Working backwards, ‘we see that our choice of the + sign in the Laurent expansion of f(z) was correct: ‘Suppose now that f(z) is analytic for |z| > R, with Laurent expansion f2)= Done" | We define the residue of f(2) at oo to be (83) Res f(2), 00} = If Dp is the exterior domain {|2| > R}, this definition is equivalent to fle)dz = 2ni Res|f(=), 00} lope ‘The orientation of the circle {|z| = R) with respect to Dr is clockwise, and this accounts for the minus sign. With this definition of residue at oe, formula (8.1) becomes (8.4) [toe = 2ri Res|f(z),00] + 2xt Res | f(2)-2)] Thus the residue theorem for exterior domains is the same as for bounded domains, except that we must take the residue at oo into account.2m VII The Residue Caleulus Note that the function f(z) can be analytic at oo and still have a nonzero residue at co. For instance, the function 1/2 is analytic at oo, while ne [Eee] = = There is ¢ mathematical explanation for this state of affairs, The expla nation i that inorder to havea coordinate fee dentin of residue, the residue should noe be associated withthe fanction (2) bat withthe dit ferential f(2)ds- Eventhough 72) may be analytic ace, the diferent Fle}de need not be analyte at co, ae can be soen by making the sual change of variable w = 1/2. (See Exercise 13.) Exercises for VIL8 1. Evaluate the residue at oo of the following functions, z eat (a) Os 2 (e) 2"e?, n= 0,41, 1 gn Sar Osa Note. There are two possibilities for (f), one for each branch of the square root. 2. Show by integrating around the dogbone contour that 35m Lge 3, Fix an integer n, positive or negative. Determine for which complex values of the parameter a the integral L converges, and evaluate it. 4. Show that r vine ita de = x(V8-1) 5. Show that the sum of the residues of a rational function on the extended complex plane is equal to zero. 6. Find the residue of =/(z?+1) at each pole in the extended complex plane, and check that the sum of the residues is zero. 7. Find the sum of the residues of (324 + 22 + 1)/(825 +52? +2) at its poles in the (finite) complex plane. Exercises 223 8 10, u. 12, Fix n > 1 and & > 0. Find the residue of 2*/(2" ~ 1) at oo by expanding 1/12" ~ 1) in a Laurent series. Find the residue of 24/(2" — 1) at ecch finite pole, and verify that the sum of all the residues is 2er0. Show that if f(2) is analytic at oo, then Reslf(z),o0] = ~ Jim, 2(f(2) ~ (co) Let D be an exterior domain. Suppose that f(z) is analytic on DuaD i at oo. Show that LO yg 2m Sop C2 f(z) fle), ED. If f(2) is not integrable at co, we define the principal value ev {sods = im, wv [" see Show that is, Ima>0, py = fo, Ima=0, . . Thus the logarithmic negel measures the change of log f(z) along the curve 7. It can be used to count. zeros and poles of meromorphic functions. The following theorem is one version of the argument principle. ‘Theorem. Let D be a bounded domain with piecewise smooth boundary @D, and let f(z) be a meromorphic function on D that extends to be analytic on OD, such that f(z) #0 on AD. Then 2) tf £@) ail Fart = % 224 1. ‘The Argument Principle 225 where No is the number of zeros of f(z) in D and Noo is the number of poles of f(2) in D, counting multiplicities. ‘The formula (1.2) is simply the residue theorem for f"(2)/f(2)- This function is analytic on DUD except possibly at the zeros and poles of f(z) in D. Let 2 be a zero or pole of f(z), and let N be the order of f() fat Zp, that is, N’is the order of the zero ify is a zero, and Nis minus the order of the pole if 2 is a pole. Then H(z) = (2-=0)" al), where 9(2) is analytic at 2» and g(29) #0. So £2) _ N(e=20)"""4l2) , (= 2% (2) (2 = 20) Nalz) @ a2) ‘Thus f’(z)/f(2) has a simple pole at 2, with residue N. If now we sum the N's over the zeros and poles, we find that the sum of the residues of F(2)/f(z) in D is No — Noo. ‘Thus (1.2) follows from the residue theorem. Now we look more carefully at the logarithmic integral (1.1), which we express in the form Hy [tee se) = ge [atonitoi+ gf dare ‘The differential dlog|{(2)| is exact. If we parametrize the curve + by a(t) = x(t) + iy(t), a St (a)) differ by an integral226 VII The Logarithmic Integral multiple of 2x, so that the increase in the argument of f(=) around 7 is an integral multiple of 2r. Example. The increase in the argument of f(z) = (2 — 26)" counterelock- wise around the circle [2 ~ 29| = p is determined as follows. We parame- trize the circle by 7(¢) = 20 + pe", 0(t)) = pe, ‘and a continuous determination for the argument of /(7(¢)) is given by arg f(o(t)) = nt, 0 (¢)) along the curve is —m, no matter how erratically the curve behaves in the lower < half-plane. 1. The Argument Prirciple 27 fo) 7a) Foray Example. We show how to use the argument prineiple to find the number (of zeros of the polynomial p(2) = 2° +924 +2°+2s-+4 in the first quadrant {0 < argz < x/2} and in the second quadrant {x/2 < arg <7}. First note that p(2) > 0 for ~co 0, and it follows from the estimates 4 +2° +27 >0 for -1 <2 <0 and 9x‘ +2°+2r>0 for z <1. Since there are no real zeros of p(z), and since the coefficients of plz) are real, the zeros come in complex conjugate pairs, and there are three zeros in the upper half-plane. To determine the number of zeros in the first quadrant, we estimate the increase of argp(z) around the boundary of the quarter disk Dr of a large radius R, consisting of the points 2 in the first quadrant satisfying |2| < R. Since the increase in the argument of p(2) around Dz is 2x times the number of zeros in Dp, any reasonable approximation will yield the exact value. We break @Dp into three paths. Along the real axis from 0 to R we have p(z) > 0, so that the argument is constant, and the increase is zero, Along the quarter-circle Ig defined by |z| = Rand 0 < argz < x/2, the term 2° dominates, and argp(z) ~ Garg. ‘The increase in argp(z) along Ux is thus approximately 6(n/2) = 3r. ‘To determine the increas> in arg p(z) down the imaginary axis from iR to 0, wwe substitute z = iy and we follow the values of pliy) = —y® + 9y* +4 + i(—y* + 2y) as y decreases from B to 0. At the initial point of this segment we have Rep(iR) ~ —R® and Imp(iR) ~ —R®, so that p(iR) lies in the third quadrant and has argument approximately —x. At the terminal point we have p(0) = 4 on the positive real axis. To see how arg p(iy) behaves, we determine where the curve p(iy) crosses the z-axis, The curve crosses the real axis for parameter values y that satisfy —y* + 2y = 0. The solutions of this cubic are y = 0 and y = +V2. There is only one crossing point corresponding to y > 9, at y= V2. The crossing point is at p (iv2) = 82, which ies on the positive real axis. Thus p(iy) remains in the lower half plane as the parameter y decreases from y = R to y = V2, and it hits the positive real axis at y= V2. Thus we see that the increase in the argument of pliy) as y decreases from R to v3 is approximately x. Since p(y) is in the upper half-plane for y between /2 and 0, and since the starting and ending points for this segment of the curve are on the positive real axis, the Increase in the argument of pliy) as y decreases from V2 to 0 is zero. Thus the total increase of arg p(z) around Dp is approximately Sn + x = 4,228 VIIL_ The Logarithmic Integral hence it is exactly 4x. By the argument principle, p(z) has two zeros in the first quadrant. We have seen that there are no zeros on the imaginary axis, so the remaining zero of p(2) in the upper half-plane lies in the second quadrant. Exercises for VITL1 1, Show that 24 +22? —z +1 Las exactly one root in each quadrant, 2. Find the number of zeros of the polynomial p(z) = 24+ 294427 + 3: +2 in each quadrant. 3. Find the number of zeros of the polynomial p(2) = 2° + 424.423 + 22? 42-45 im the first quadcant (Re= >0, Im2 > 0}. 4. Find the number of zeros of the polynomial p(=) = 2° +225—22 + 243 In the right half-plane. 5, For a fixed real number a, find the number of zeros of 24 + 29+ 42? +az+3 satisfying Re <0. (Your answer depends on a.) 6. For a fixed real number a, find the number of solutions of 2° +2:3— 2242 =a satisfying Rez > 0, 1. For a fixed complex number A, show that if m and n are large Integers, then the equation e* = 2 + has exactly m +n solutions in the horizontal strip {—2nim < Im < 2xin} 8. Show that if Red > 1, then tie equation e* = 2+ has exactly one solution in the left half-plane. 9. Show that if f(z) is analytic in a domain D, and if 7 is a closed curve in D such that the values of f(z) on > lie in the slit plane 2, Rouché's Theorem 229 C\(—ce, 0}, then the increase in the argument of f(z) around + is 2. Rouché’s Theorem ‘There is a general principle to the effect that the number of zeros of an analytic function on a domain does not change if we make a small change in the function. In other words, the number of zeros of an analytic function remains constant under small perturbations. One version of this principle is as follows. ‘Theorem (Rouché’s Theorem). Let D be a bounded domain with piecewise smooth boundary OD. Let f(2) and h(=) be analytic on DU@D. If |h(2)| < |f(2)| for 2 € OD, then f(z) and f(z) + h(z) have the same number of zeros in D, counting: multiplicities. ‘The hypothesis implies that f(z) #0 on aD, and that f(21-+ (2) #0 on OD. From f(2) + h(z) = f(2)|L + h(2)/f(2)}, we obtain arg( f(z) + (2) = arg(f(z)) +a hey (21) af fe) +M=)) = angle) +e (1+ 322) Since {h(2}/f(2)| <1, the values of 1+ A(=)/{(2) lie in uo right halk plane, and the increase of arg(1 + A(z)/(2)) around a closed boundary curve is 0, From (2.1) we see that arg(f(z) + A(z)) and arg f(2) have the same increase around D. By the argument principle, the functions have the same numberof zeros in D. Example. We use Rouché’s theorem to find the number of zeros of the polynomial p(z) = 2° +924 + 23+ 22+4 that lie inside the urit circle. ‘To apply Rouché’s theorem, we seek to express p(2) in the form p(z) = BIG + litde, where the function “BIG” dominates “little” on the unit circle, and where itis apparent how many 2eros “BIG” has inside the unit circle Our choice for “BIG” in this case is f(z) = 924, which has four zeros inside the unit circle, all at the origin. The function “little” is h(z) = 2°-+23422-+4, which satisfies |h(=)] < |f(2)| for |2| = 1. By Rouché’s theorem, p(2) = f(2)+h(2) also has four zeros inside the unit circle. Example. We wish to find all solutions of the equation e* = 1+ 22 that satisfy |2| < 1. One obvious solution is z = 0, To determine whether there are other solutions, we apply Rouché’s theorem to count the number of zeros of e* — 1 ~ 22 in the unit disk. In this case, the term f(z) = 22 is the dominant term on the unit circle, We take h(2) = ef ~ 1, and we210 VIN ‘The Logarithmic Integral estimate he) = |e@ 11 = 2 2 le? (ef wey = feta = fer Fe Fa] 1. For [2| large, the term 2” domi- nates. We take f(z) = 2" and A(z) = ay12"-" +--+ a9. If Ris large, we have [f(2)| < [f(2)| for |2| = R. Consequently, p(z) = f(z) + f(z) has the same number of zeros in the disk |2| < Ras f(2) = 2", which is m. Exercises for VITL2 1. Show that 22° + 6: — 1 has one root in the interval 0< x < 1 and four roots in the annulus {1 < |2| <2}. 2. How many roots does 2° +25 ~ 8:9 4.22 +1 have between the circles {l2| = 1} and {|2| = 2}? 3. Show that if m and n are positive integers, then the polynomial we) = 1t24 54-4204 2 mi has exactly n zeros in the unit disk, 4, Fix a complex number such that |A| <1, For n> 1, show that (2—1)"eF — 2 has n zeros satisfying |2 — 1] < 1 and no other zeros in the right half-plane. Determine the multiplicity of the zeros, For a fixed A satisfying |A| < 1, show that (z ~ 1)"et + M2 +1)" hhas n zeros in the right half-plane, which are all simple if A # 0. 6. Let p(z) = 2° +924 + 29 +22 +4 be the polynomial treated in the example in this section. (a) Determine which quadrants contain the four zeros of p(2) that lie inside the unit circle. (b) Determine which quadrants contain the two zer0s of p(2) that lie outside the unit circle. (e) Show that the two zeros of p(2) that lie outside the unit circle satisfy {2 + 3i] < 1/10} 7. Let f(z) and g(2) be analytic functions on the bounded domain D that extend continuously to OD and satisfy |f(2) + 9(2)| < [f(2)] + lg(z)| on AD. Show that f(z) and g(2) have the same number of 3. Hurwitz’s Theorem 231 zeros in D, counting multiplicity. Remark, ‘This is a variant of Rouché's theorem, in which the hypotheses are symmetric in f(2) and g(z). Rouché's theorem is obtained by setting A(z) = ~ f(z) — .9(2). For the solution of the exercise, see Exercise 9 in the preceding section. 8, Let D bea bounded domain, and let f(2) and h(2) be meromorphic functions on D that extend to be analytic on D. Suppose that |h(2)] <|4(2)| on AD. Show by example that f(z) and f(2)+ h(2) can have different numbers of zeros on D. What can be said about J(z) and f(2) + h(2)? Prove your assertion. 9. Let f(z) be a continuously differentiable function on a domain D. Suppose that for all complex constants a and b, the increase in the argument of f(2)+az+6 around any small circle in D on whieh J(2)+az +640 is nonnegative. Show that f(2) is analytic. 3. Hurwitz’s Theorem ‘The argument principle (logarithmic integral) provides a tool for studying, the behavior of the zeros of a convergent sequence of analytic funetions. ‘The zeros of the funetions in the sequence converge to the zeros of the limit function, in a sense made precise by the following theorem, Theorem (Hurwits’s Theorem). Suppose {fx(z}} isa sequence of an- alvtie functions on a domain D that converges normally on D to f(2), and suppose that f(z) has a zero of order N at z. Then there exists p > 0 such that for k large, f(z) has exactly N zeros in the disk {|z — zo] < p}, counting multiplicity, and these zeros converge to 29 as k ~ oo. For the proof, let p > 0 be sufficiently small so that the closed disk {|2 — 20] 0 such that |f(z)| > 4 on the circle |z — zo| = p. Since fe() ‘converges uniformly to f(z) for |z—zo| < p, for k large we have | fx(2)| > 6/2 for |2 —29| = p, and further, £,(2)/fu(z) converges uniformly to f'(2)/F(2) for |z — zo| = p. Hence the integrals converge, 1 Hy 1 fie) Bi Siem sstay 2) BS. ste FE) Now, the integral on the let is the number Nj of zeros of fe(z) in the disk {\2— 20] 0, so the zeros of f(z) must accumulate at zo. de232 VILL The Logarithmic Integral An alternative proof of Hurwita’s theorem can be based on Rouché's theorem, applied to the “little” function fi(2)— f(z) and the “big” function f(z). See Exercise 2. ‘We say that a function is univalent on a domain D if it is analytic and ‘one-to-one on D. In other words, the univalent functions on D are the ‘conformal maps of D to other domains. If we apply the preceding theorem to univalent functions, where NV = 1, we obtain the following, which is sometimes also referred to as Hurwitz’s theorem, ‘Theorem. Suppose {fx(2)} is a sequence of univalent functions on a do- ‘main D that converges normally on D to a function f(z). Then either f(z) is univalent or f(z) is constant. Indeed, suppose the limit function f(z) is not constant. Suppose = and Go satisfy f(co) = F(Co) = uo. Then zo and Gp are zeros of finite order for {(2) ~ wo. By the preceding theorem, there are sequences 2 —> 29 and Ge Go stich that f(z) ~ wo = 0, and fe(Ge) ~ wo = 0. Since fy(z) is tunivalent, we have 24 = Ck, and in the limit 9 = Go. Thus f(2) is univalent. Example. For each k > 1, the fonction f,(z) = 2/K is univalent on the complex plane. The sequence {f,(2)} converges normally to the constant function f(z) = 0. Thus a normal limit of univalent functions need not be univalent, and both alternatives of the theorem occur. Exercises for VIILS 1. Let {fx(z)} be a sequence of analytic functions on D that converges normally to f(z), and suppose that f(=) has a zero of order Nat 39 € D. Use Rouché's theorem to show that there exists p > 0 such that for k large, f,(2) has exactly N zeros counting multiplicity on the disk {lz ~ 291 = 0, and f(2) attains the value 20 'm times at 29 if 2 is a pole of f(z) of order m. ‘The number of times that f(2) attains a value w» on D is obtained by adding the number of times it attains the value wp at the various points of D. Example. The polynomial 2 + 1 attains the value w 2 =0, and it attains the value w = co m times at z= 00. sm times at Example. A polynomial f(z) of degree m > 1 attains each value w € C* m times on C*, always counting multiplicity. In fact, for fixed finite w, f(2)~w is a polynomial of degree mn, which has m zeros counting multiplicity. Since f(z) has a pole of order m at oo, it attains the value w = oe m times at We can use the logarithmic integral to study the dependence on w of the numberof times that an analytic or meromorphic function attains the value w. ‘The technique is the same as that used in the preceding section, except that instead of a sequence of functions we treat functions that depend continuously on a parameter. The basic idea is as follows Let f(z) be a nonconstant analytic function on a domain D. Let zy € D. ‘wo = f (20), and assume that f(z) ~ wo has a zero of order m at zo. Since the 2eros of f(z)w are isolated, we can choose p > Oso that f(2)—wo #0 for 0 < |z — z9| < p. Let 6 > 0 satisfy |f(z) ~ wo| > 6 for |z — zo| = p. The integral v 1 f(z) 1 4a) Nw) = LO, w—wsl cd (1) NC) = FE I nten Taye wml < is then define, and it depends analytically oa w. Since N(u) i the umber ‘of zeros of f(z) — w in the disk {|= ~ 29| 0 be chosen such that |f(z) — wo| > 6 for |z—29| =p. Then for each w such that jw —wol <6, there is a unique z satisfying |2 ~ 2|

where ,... 29 ar the zeros and poles of f(2), and m, isthe order of f(z) at 2. Let f(z) be a meromorphic function on the complex plane that is doubly periodic. Suppose that the zeros and poles of f(z) are at the points 2,... ,2, and at their translates by periods of f(2), and Exercises 235 suppose no z; is a translate by a period of another 2. Let my be the order of f(z) at 2). Show that >mjz, is a period of f(z). Hint. Integrate zf'(z)/f(2) around the boundary of the fundamental par- allelogram P constructed in Section VIS. 3. Let { fx(z)} be a sequence of analytic functions on a domain D that converges normally to f(z). Suppose that f(z) attains each value w ‘at most m times (counting multiplicity) in D. Show that either f(2) is constant, or f(z) attains each value w at most m times in D. 4, Let f(z) be an analytic function on the open unit disk D = {|2| <1}. Suppose there is an annulus / = {r < |2| < 1} such that the restriction of f(z) to U is one-to-one. Show that f(z) is one-to-one on D. 5. Let f(z) = p(z)/q(z) be a rational function, where p(z) and q(2) are polynomials that are relatively prime (no common zeros). We define the degree of f(z) to be the larger of the degrees of p(2) and q(2). Denote the degree of f(z) by d. (a) Show that each value w eC, w# f(co), is assumed d times by f(z) on C. (b) Show that J (2) attains each value w € C* d times on C* (as always, counting multiplicity). 6. Let f(z) be a meromorphic function on the complex plane, and suppose there is an integer m such that f~"(w) has at most m points for all w € C. Show that f(2) is a rational function, 7. Let F(z,w) be a continuous function of z and w that depends an- alytically on z for each fixed w, and let F,(z,w) denote the de- tivative of F(z,w) with respect to z. Suppose F(z, wo) = 0, and Fi(z0,10) # 0. Choose p such that F(2, 109) # 0 for 0 < |2~z9) < p. (a) Show that there exists 6 > 0 such that if |w —wo| < 6, there is a unique 2 = g(w) satisfying |2 ~ z9| < p and F(z,w) = 0. (b) Show that te) = CFG.) a 0) = ef Fa, tw wol <8 (€) Suppose further that F(z,.) is analytic in w for each fixed =, and let Fa(z,w) denote the derivative of F(z, w) with respect to w. Show that g(w) is analytic, and gw) = —Fa(g(w),w)/Fi(g(w),w). (€) Derive the inverse function theorem given in this section, to- gether with the usual formula for the derivative of the inverse function, asa corollary of (a), (b), and (¢).236 VIII The Logarithmic Integral Remark. This is the implicit function theorem for analytic functions. Note that a specific formula is given for the function o(w) defined implicitly by F(g(w), 8, Let D be a bounded domain, and let f(2) be a continuous function on DUD that is analytic on D. Show that (f(D)) © f(OD), that is, the boundary of the open set. f(D) is contained in the image under f(2) of the boundary of D. 5. Critical Points Tet f(z) be a nonconstant analytic function on a domain D. A point =o is called a critical point of f(2) if f"(s9) — 0. The value (20) ~ wo is called acritical value of f(2). We define the order of the critical point zo to be the order of 2er0 of #"(z) at 29. Since the eritical points are the zeros of the nonconstant analytic function f"(z), critical points are isolated. Our aim is to understand the behavior of an analytic function near a critical point, and to understand the behavior of the inverse function near the corresponding critical value. Example. The function f(z) = 2" +c has a critical point of order m—1 at 2 = 0, with critical value c. ‘The inverse function for f(z) is given by the m branches of (w ~ ¢)!/". Suppose zo is a critical point of f(2) of order m —1 with critical value ‘wo. Then f(z) — wo has a zero of order m at 29, and we can proceed as before by considering the logarithmic integral (4.1). This time the constant value of N(w) is m. Thus for p > 0 small and for w near wo, there are m points 21(tt),.-. .2m(w), repeated according to multiplicity, such that f(z) attains the value w in the disk {\z ~ 20] < p} precisely at the points 2(w),... ,2m(t)- We wish to see how the points z;(w) depend on w. To do this, we will use a simple procedure, based only on the existence of an analytic inverse for an analytic function at a noneritical point, The procedure is to make a change of variable in order to show that the behavior of f(2) near 29 is effectively the same as the behavior of the analytic function ¢” at the critical point ¢ = 0. We proceed as follows. ae an aL Since f(z) uy has order m at zo, we can factor f(z)—w = (2~20)"A(=)- where A(2) is analytic at zo, and (zo) #0. Near zo we define an analytic 5. Critical Points 2ar branch of A(z)", and then for o(=) = (2 — zo)h(z)!/" we have Sl) wo = 92)", Sle) = 9(2)" + wo Moreover, (:) analyte at so and has simple zero there, 0 (24) #0 tnd 2) inonetoone nea ip, Thus J) ls reprecnted asthe compan of three functions, the unvaleatfneiong() followed bythe map ¢>> flowed by a translation € "> € sy. Since we understand the power function ¢ = (andthe trensation w= € + wy very well and since (2) isunivalet near 2p, we can dra from this aca pctue ofthe beevor of fe). For instance, the st where i teal consist of stright ine Segments passing through 0, dividing the plane near ¢ = Onto 2m sectors of aperture =/m, on which Im(¢") is alternately positive and negative. The txt hore f(2) “uy seal athe mage under 4 (¢) of thee ne egrets tnd ence const of canals curves through. They dive the plane ner» ito 2m pleslice domains with vertex at zo and aperture z/m,on which m(f(2)~ ty) i aernately postive and negative A point w near wo, w 4 wo, has m distinet preimages 21(w),...,2m(W). They are the m branches of (w ~ wo)"/" composed with g"i(¢). If we make a branch cut and consider the principal branch (w — wo)!/"" on the slit disk {lw — wo] < 5}\(wo ~ 6, wo}, the other branches are of the form 29/™(y —~ ug)", and the preimages of w are given by the composition of g-'(¢) and these branches, - 200) = 97* (e297 —w)"), jem In particular, for w # wo, the preimages =;(w) are distinct, and each =)(w) depends analytically on w. The values of 24(t) at the “top edge” of the coincide with the values af z;+1(w) at the “bottom edge” of the sli, s0 that 2(w) is continued analytically to =,.,(w) when w follows a circle around 9 in the positive direction. In the same way that we constructed & Riemann surface for the inverse function of 6", we can construct a Riemann surface for the inverse function 2(w) of w = f(z). We create m copies of the slit disk, and we define z(w) to be =,(w) on the jth copy. Then we glue238 VILL ‘The Logarithmic Integral Sti > together the edges of the branch cuts so that the function (w) becomes continuous on the surface. Now consider the function (2 21(w)) ++ (= zal) = 2" + am aw)2"* + ay (w)2 + a0(w), where ag(w) = (—1)"21(t) + 2m(w), am-s(te) = ~ > ay(u) If we continue ax(w) around a circle centered at arp, the values of the (ws are permuted and we return to the same value for ay(w). ‘Thus u(t) is analytic in the punctured disk 0 < r—wol < ¢. Evidently, ox(t) is bounded as w — wo. By Riemann’s theorem on removable singularities, ag(w) is analytic at wy. In fact, the value ay(up) is just the coefficient of 2* in the expansion of (z — zo)” in powers of z, Thus we have found a polynomial equation A) Fam alw)™ 1 tan(w)z baa(w) = 0, with analytic coefficients ap(w),... ,am-i(0}, that reduces to the equa- tion (z ~ Z9)!" = 0 at w = wp and that has solutions given precisely by ai(w),--+ .2m(w), ‘We summarize our results as follows. Theorem. Let f(z) be analytic at 2p. Suppose 2 is a critical point of order m1 for f(z), with cttical value (29) = up. Let p > 0 satisfy F(2) # wo for 0 < |7 ~ 70] 0 satisfy |f(z) — wol > 6 for >. Then for each w such that 0 < |w ~ wol < 6, the equation F(>) = w has exactly m distinct solutions =1(w),... ,2m(te) in the disk {ls ~ 20] < p). The functions zi(w).-.. .2m(t) can be chosen to depend analytically on w in the slit disk {lw — wol < 8}\(up ~ 5,wo). They form the m branches of an analytic function = 2(w) on an m-sheeted Riemann le ~ 201 Exercises 239 surface over the punctured disk {0 < |w ~ wol < 6}. The z,(w)'s are the solutions of a polynomial equation (5.1) with coefficients a,(w) that are analytic for |w ~ wo| < 6, We have dealt with the solutions of the equation f(=) — w = 0 near a solution point (29,9). A similar state of affairs holds for the solutions of fan equation F(2,w) = 0, where F(z,w) depends analytically on both 2 and w in a neighborhood of a solution point (29,1), though in this case the solution set is a finite union of surfaces passing through (29,1). This is the simplest example of an analytic variety, which is defined to be the set of common zeros of a family of analytic functions on some domain in complex n-space. Exercises for VIS 1, Find the critical points and critical values of f(z) = 2-+1/z. Sketch the curves where f(2) is real. Sketch the regions where Im f(z) > 0 and where Im f(z) <0. 2, Suppose g(z) is analytic at z = 0, with power series 9(2) = 2+i=4+ O(z*).. Sketch and label the curves passing through = = 0 where Reg(z) = 2 and where Im (z) = 0. 3, Find the critical points and critical values of f(z) = =? +1. Sketch the set of points such that |f(2)] <1, and locate the critical points of f(z) on the sketch. 4. Suppose that f(z) is analytic at zo. Show that if the set of = such that Ref(z) = Ref(zo) consists of just one curve passing through zo, then J"(20) #0. Show also that ifthe set of = such that L/(2)| = |f(20)| consists of just one curve passing through zo, then S'(2) £0. 5. How many critical points, counting multiplicity, does a polynomial of degree m have in the complex plane? Justify your answer. 6, Find and plot the eritical points and critical values of f(z) = 2? +1 and of its iterates f(f(z2)) = (2?+1)?+1 and f(f(f(2)))- Suggestion. Use the chain rule. 7. Let f(z) be a polynomial of degree m. How many (finite) critical points does the V-fold iterate fo---0 f (N times) have? Describe them in terms of the critical points of (2). 8. We define a pole 29 of f(z) to be a critical point of f(z) of order k if sy is a critical point of 1/f(2) of order k. We define z — 00 to be a critical point of /(=) of order & if w ~ 0 is a critical point of g(w) = f(1/w) of order k. Show that with this definition, a point240 10. nL. 12 13, ua, VIII The Logarithmic Integral 2 € C" is a critical point of order k for a meromorphie function (=) if and only if there are open sets U containing 29 and V containing wp = f(z) such that each w € V, w # wo, has exactly k +1 preimages in U. Remark. We say that f(z) is a (k + 1)-sheeted covering of f-"(V\{wo}) VU over V\{wo} Show that a polynomial of degree m, regarded as a meromorphic function on C*, has a eritieal point of order m ~ 1 at 29 = co, Locate the critical points and critical values in the extended complex plane of the polynomial f(2) = 24 — 22%. Determine the order of each critical point. Sketch the set of points > such that Im f(z) > 0. Show that if f is a rational function, and if g is a fractional linear transformation, then f and go f have the same critical points in the extended complex plane C*, What can be said about the critical values of g of? What can be said about the critical points and critical values of f 0.9? Let f(2) = p(2)/a(2) be a rational function of degree d, so that p(2) and 4(2) are relatively prime, and d is the larger of the degrees of, p(2) and g(2). (See Exercise 4.5.) Show that f(z) has 2d—2 critical points, counting multiplicity, in the extended complex plane C* Hint. If deg # deg. then the numberof critical points of (2) in the finite plane C is deg(qp! — 4'p) = degp + degq ~ 1, while the ‘order of the critical point at oo is | deg p ~ deg — 1. Show that the set of solution points (w,2) of the equation 2? — 2(costw)2+1 = 0 consists of the graphs of two entire functions = (w) and 22(w) of w. Specify the entire functions, and determine where their graphs meet. Remark. The solution set forms a reducible ‘one-dimensional analytic variety in C2 Let ao(w),--. .am—1(t) be analytic in a neighborhood of w = 0 and vanish at w = 0. Consider the monic polynomial in z whose coefficients are analytic functions in w, Pl2,w) = 2 +am-a(w)2""1 +---+ag(t), rol <6. Suppose that for each fixed w, 0 < |u| < 6, there are m distinct solutions of P(z, ) = 0. (a) Show that the m roots of the equation P(:,w) = 0 deter- mine analytic functions =(w),...,%(w) in the slit disk {lvl < 6\(-6.0]. Hint. Use the implicit function theorem (Exercise 4.7). (b) Glue together branch cuts to form an m-sheeted Riemann sur- face over the punctured disk on which the branches =,(w) de- termine a continuous (single-valued, analytic) function. Exercises 2a (6) Suppose that the indices are arranged so that for some fixed k, 1 k =: (2— z4(ww)) determines a polynomial in 2 whose coefficients are an. alytie functions of w for |1| < 6. Show further that the poly- nomial Q(z, we) is an irreducible factor of P(=, w), and that all irreducible factors of P(z,t) arise from subsets of the 2,(1)'s in this way. (a) Show that if f(2) is an analytic function that has a zero of order m at z = 0, and 2i(w),...,2m(t) are the solutions of the equation w = f(z), then the polynomial P(z,w) = (2 — 21(t))-+-(2 ~ 2 (t)) is irreducible. Consider monic polynomials in z of the form Plz,w) = 2+ am a(we)2" b+ sb ag(w), where the functions ap(w),-.- ,@m-1(w) are defined and meromor- phic in some disk centered at w = 0. Let Pa(z,1) and Py(z,w) be two such polynomials, and consider the following algorithm. Us ing the division algorithm, find polynomials Ag(z,w) and P,(z,w) such that Po(z,w) = Aa(z,w)Pi(2,.) + Py(z,w) and the degree of ‘P,(z,w) is less than the degree of P,(2,1)- Continue in this fashion, finding polynomials A,1(2,) and P)1(z,w) such that, Py-a(zw) = Ayal, w)P;(2,w) + Prsilz,w) and deg P,.1(z, 1) < deg P,(z,w), until eventually we reach Pesa(2st) =O, Peaa(2,) = Ae(z,w)Pe(2,u). Let D(z,w) be the monic polynomial in z obtained by dividing Pa(z,w) by the coefficient of the highest power of =: (a) Show that D(z,1) is the greatest common divisor of F(z.) and P;(2,w), in the sense that D(z, w) divides both Pi(z,) and P,(2,.), and each polynomial that divides both Fy(z.1) and P,(z,1) also divides D(z, 1). (b) Show that there are polynomials A(=,1) and B(z,w) such that D= AP, + BP, (6) Show that if Py(z,w) and Pi(z,w) are relatively prime (that is, D(z,w) = 1), then there is ¢ > 0 such that for each fixed w, 0-< [w < ¢, the polynomials Py(z,w) and Pi(z,w) have no (4) Show that any polynomial P(z,1) as above can be factored as a product of irreducible polynomials, and the factorization is unique up to the order of the factors and multiplication of a factor by a meromorphic function in w.m2 VIII The Logarithmic Integral (©) Show that if the coefficients of P(z,w) are analytic at w = 0, then the coefficients of each of the irredueible factors of P(2,w) are analytie at w = 0. (f) Show that if P(z, w) is irreducible, then there is ¢ > 0 such that for each fixed w, 0 < \tl <, the roots of P(z,.) are distinct. (2) Show that the results of Exercise 14(a)-(4) hold without the supposition that the solutions of P(z,1) =O are distinct. 6. Winding Numbers Let y(t), @ < # 0 so small that every point of T has distance at least 45 from any point of C\D, and we divide the complex plane into atid of closed squares of side length 6. (See the figure.) Let be the unton of all the (closed) squares in the grid that contain a point with distance less than 6 from T, and let U = K\OK. Then U is a domain with boundary @U = AK consisting of the sides of squares in K that are adjacent to a ‘square in the grid not in K’. By following along these sides, we sce that OU i: a finite union of closed curves consisting of consecutive sides of squares. ‘Two boundary curves may intersect at a point where two squares in U are kitty comer from each other, but these intersections can be eliminated by adjoining to U a smaller square with vertex in the corner. Finally, we adjoin to U the squares from any component of C*\U' that does not meet CAD. Then each component of C*\U meets C™\D. Green’s theorem is proved (completely rigorously) for domains built in this way from contiguous squares, so that the Cauchy integral theorem isry VIIL_ The Logarithmic Integral at our disposal. If z € P, we have 1 I) Omi Jou C—2 ff) = If we integrate around 7 and then interchange the order of integration, we obtain [roe = 3 E Ufe ‘The winding number W(7,¢) is constant on each connected component of C\P, and it is zero for ¢ € C\D. Since each component of C\U meets C\D, it is zero for ¢ € C\U, hence also for ¢ € AU. Hence the integral is also zero. ‘As an analogue of the Cauchy integral formula we have the following. ae] 1046 =~; [, wonoroa ‘Theorem. Let f(z) be analytic on a domain D, and let +y be a closed path in D with trace P = 7([a,8)). IF W(>,¢) = 0 for all ¢ ¢ C\D, then al LO) ae ‘The proof is simple. We apply the preceding version of the Cauchy integral theorem to the analytic function g(2) = (f(=) ~ f(20))/(2 ~ 20. From f, 9{2) dz =0, we obtain lyf oni J, = = Wy, z0)f(z0), 20 € D\T. (@) 1 f (20) a = 5 [ he = woes) Winding numbers canbe defined for a inte collection of closed curves, by simply adding the winding numbers for the varius cares In the col Ieetion. (See the exereises for Section 8.) ‘Thus fr instance, i Dis & Bounded domain vith piecewise smnoth boundary, we define the winding number W(OD,¢) of AD around (- OD to be the sum of the winding tumbers of the sarious component curves of @D, appropriately oriented Fom Covehy's theorem andthe Cauchy integral reprenentation formula, Exercises 245 wwe then have wn,.g = al 1 ced, $ lo ¢¢Duap. Thus the winding number of 8D around points ¢ outside of DU aD is 0, and the winding number jumps to +1 as ¢ erosses over AD into D. This is a special case of the jump theorem for Cauchy integrals treated in the next section. Exercises for VIIL6 1. Sketch the closed path >(¢) = e sin(2#), 0 << 2x, and determine the winding number 1V(7,¢) for each point ¢ not on the path, 2. Sketch the closed path 7(t) = e-™* cost, 0 < 1 < 2m, and determine the winding number 1V(7,¢) for each point ¢ not on the path. 3. Let f(z) be analytic on an open set containing a closed path 7, and suppose f(2) # 0 on 7. Show that the inerease in arg f(2) around is 2nW(F 070) 4. Let D bea domain, and suppose 2 and 2; lie in the same connected component of C\D. (a) Show that the incresse in the argument of f(2) = (2 ~ 29)(z ~ 21) around any closed curve in D is an even multiple of 2x. (b) Show that (z — z9)(z ~ 21) has an analytic square root in D. (c) Show by example that (2 — z9)(2 — 21) does not necessarily have an analytic cube root in D. 5. Show that if-7 is a closed curve in the complex plane, with trace P, and if 25 ¢ I, then Laaare 6. Let + be a closed path in a domain D such that W(7,¢) = 0 for all ¢ ¢ D. Suppose that f(2) is analytic on D except possibly at a finite number of isolated singularities 21,... ,2 € D\T.. Show that n>2. [se ds — Qri SW 24) Reef, 24) Hint. Consider the Laurent decomposition at each 2x, and use Ex: ercise 5. 7. Evaluate 1 dz mila"246 VIIL_ The Logarithmic Integral ‘where 7 is the closed path indicated in the figure. Hint. Either use Exercise 6, or proceed directly with partial fractions. 8. Let (t) and a(t), a < t < b, be closed paths. (a) Show that if € C does not lie on the straight line segment between 7(t) and a(t), for a 1, and f(z) Las z 50. Show that f(2) # 0 for |2| <1. 10. Let K be a nonempty closed bounded subset of the complex plane, and let f(z) be a continuous complex-valued function on the com- plex plane that is analytic on C\K and at oo. Show that every value attained by f(z) on C* = CU {oo} is attained by f(z) somewhere on K, that is, f(C*) = f(K). 7. The Jump Theorem for Cauchy Integrals Suppose 7 is a piecewise smooth curve with trace T, and let f(z) be a continuous function on I. We define the Cauchy integral of (2) along to be the function F(¢) defined off I’ by -2/ (7) FO = safe We have seen that the Cauchy integral F(¢) is analytic on C\P, and F(¢) vanishes at oo a, Cec. Example. If7 is a closed curve parameterizing the (positively oriented) boundary T of a domain D, and if f(2) is analytic on D and acros then the Cauchy integral F(¢) of f(2) around » is given by FQ, GED. Ro (Eko. ‘This follows from Cauchy's theorem if € C\(DUP), and from the Cauchy wogral formula if ¢€ D. 7. The Jump Theorem for Cauchy Integrals 27 Example. If isa closed path, the Cauchy integral of the constant function f(z) = 1 around 7 is the winding number F(¢) = W(7,¢) of y around ¢. Now suppose that the continuous function f(z) on I’ is analytic at some point 29 € F. We suppose that there is a small disk U centered at. 2 such that f(z) is analytic on U U OU and such that 7 passes through U in a piecewise smooth subare 7p traveling from 2, € AU to another point 22 € AU, so that 7\7p does not enter U, while the trace Ty of 70 divides U into two connected open domains. As we traverse 79 in the positive direction, one of these domains U7, lies on the right side of 7p, the other U_ on the left side. We break the integral for F(¢) into two pieces Lis s FO = gale ‘The second integral G(¢) is analytic on U, while the first integral Fo(¢) is analytic on U\Eo. Let 4 denote the arc of OU from 2, to 22 that lies in the boundary of Uz, and let 7 denote the arc of U from 2 to 2 that lies in the boundary of U_, as in the next figure. We define e+ if Brae = nora. nO=gl Pare, cev. By Cauchy’s theorem, 1 f(z) 1 f(z) y, Lf e- bf Ba com and consequently, F(¢) = F,(¢) for ¢ € Uy. Similarly, we define r= Gf Marco, con and then from Cauchy’s theorem we obtain F(¢) = F_(¢) for ¢ € U_. Ifwe integrate along the arc 74 of @U from =; to 22, then backwards along the are of AU from 22 to =, we obtain the integral around the boundary 3U. ‘This and Cauchy's theorem yield rO-nO- Gf =F, CEU We have proved the following Theorem (Jump Theorem for Cauchy Integrals). Suppose that I is smooth curve that passes through zo, and f(z) is a continuous function oa’ that is analytic at 29. Let U be a small disk containing 2) such that J(2) is analytic on U and such that I divides U into the two components Us. ‘and U_ as above. Then there are analytic functions F,(C) and F_(¢) on U satisfying FAQ) - FQ) = £0), CEU.8 VT The Logarithmic Integral and such that the Cauchy integral F(¢) of f(z), defined by (7.1), satisfies FO, EUs, FO = 3) {EG Geu- Thus as ¢ crosses P trom right to left ftom U, to U-), the values of the Cauchy integral F(¢) jump by $(6), FO = FOO ev. Example. In the first example above, where +7 parameterizes the bound- ary P of a domain D and f(2) is analytic on D and across I, the Cauchy integral of f(z) jumps from 0 outside P to f(¢) inside T. Now consider the special case where j(z) = 1 and 7 is a closed curve. In this case, the Cauchy integral (7.1) is the winding number W(7,<), and ‘we are led to the following jump theorem. ‘Theorem (Jump Theorem for Winding Numbers). Let 7 bea closed path with trace, and let 29 = (to) €T. Suppose that »(t) is continuously differentiable for t near to, with 7'(to) # 0. Let U be a small disk centered at 29 that is divided into two components Us by a segment yo on 7 as above, and suppose that ~/\7o does not enter U. Then as ¢ € U crosses T from right to left (from U,. to U_), the winding number W(y,¢) jumps by +1, We) = WG) th EUW, Ge EUs ‘This does not follow immediately from the preceding jump theorem, since the path 7\7o is not necessarily smooth, However, the winding number can be expressed as a dz mid, where G(¢) is the increase of the argument of = ~¢ along 7\7o. If we apply the jump theorem to the Cauchy integral along 7o, and we observe that G(¢) is analytic on U and hence does not jump, we obtain the jump em for the winding number. Word) +60), 7. The Jump Theorem for Cauchy Integrals 249 Recall (Section III) that a closed path Tin the complex plane is sim- ple if it is parameterized by continuous function 7(t), a < t < b, such that 7(s) ¢ y(t) for a r-_ Show further that, fol6) = F-(6) and fil¢) = —F, (6). Remark. The formula f(z) = fol2) + fu(z) reflects the jump theorem for the Cauchy integral of (2) around circles |2| =r. 5. Let 7 be a piecewise smooth curve, and let F(¢) be the Cauchy integral (7.1) of a continuous function f(z) on 7. Show that if 9(2) is a smooth function on the complex plane that is zero off some bounded set, then [ocrteaas = 2 ff Brea ay Hint, Recall Pompeiu’s formula (Section IV.8). 6. Determine whether the point = lies inside or outside. Explain. 7. Asimple arc I’ in C is the image of a continuous one-to-one func- tion 7(f) from a closed interval {a,6] to the complex plane. Show that a simple are Tin C has @ connected complement, that is, C\T is connected. You may use the Tietze extension theorem, that a con- ‘tinuous real-valued function on a closed subset of the complex plane can be extended to a continuous real-valued funccion on the entire complex plane. Hint. Suppose zo belongs to a bounded component: of C\P. Find a continuous determination h(s) of log(= — 30) on P, extend A(z) to @ continuous function on C*, and define f(2) = 20 ‘on the component of C\P containing 29, and f(:) = eM) on the remainder of C*\P. Consider the increase in the argument of f(=) around circles centered at zo 8. Prove the Jordan curve theorem fora simple closed curve by fling in the following proof outline. (a) Show that each component of C\I has boundary [. Hint. For 70 = 7(to) € T, apply the preceding exercise to the simple arc T(r), where I is a small open parameter Interval conealn- ing to. (b) Prove the Jordan curve theorem in the ease where contains a straight line segment.252 VIN The Logarithmic Integral (c) Show that for any 29 = vito) € T, any small disk Do con- taining 2, and any component U of C\P, there are points 2, = (ty) and =2 = 9(t2) such that the image of the parameter segment between fy and t2 is contained in Dp and such that 21 and 27 can be joined by a broken line segment in U1 Do. (a) With notation as in (b), let be the simple closed curve ob- tained by replacing the segment of y in Dy between zo and =; by the broken line segment in U' Do between them, and let 7 be the simple close! curve in Dy obtained by following the segment of 7 in Dp from 29 to 21 and returning to 9 along the broken line segment. Show that W(r,¢) = 0 and W(7,¢) = W(o,<) for CE CAL, ¢ ¢ Do. {e) Using (b) and (4), show that C\ has at least two components and that W(y,¢) = 41 for ¢ in each bounded component of ev. (£) By taking U in (c) to be @ bounded component of C\P, show that W(7,¢) =0 for ¢ in any other component of C\T 8, Simply Connected Domains A domain in the plane is ‘simply connected” if it has no “holes.” Disks and rectangles are simply connected. More generally, star-shaped domains are simply connected. Annuli, punctured disks, and the punctured plane have “holes” and are not simply connected. Our aim in this section is to make more precise the notion of “simple connectedness.” Let (t), a (0) = 9(). Suppose that 7 can be deformed continuously 10 a potnt in D. Then there is a continuous family of closed paths ,, 0 < <1, such that 9 = 7, %1 is the constant path at zo, and each path 7 starts and ends at 2 Suppose that 7 is deformed to a point by closed paths a4, 0< 8 < } Thus oo(t) = (0) for 0 0 such that J2—w| > 6 for all 2 ¢ E and we X\E. rm DROWN connected ot comected For domains in the plane many conditions can be formulated that are equivalent to simple connectivity. We give several that. are useful for our purposes. Theorem. The following properties are equivalent, for a domain D in the complex plane: (i) D ts simply connected, every closed differential on D is exact, i) for each % € C\D, there is an analytic branch of log(z— 29) defined on D, (iv) each closed curve + in D has winding number Wy, 2) = 0 about all points 2 € C\D, (v) the complement of D in the extended complex plane” is connected. In practice, the condition (v) is usually the easiest to check for a given domain D. It is usually easy to spot any set E in C\D that has positive distance from the rest of C\D. We think of such sets E as representing, holes in D. Example. If D is an annulus {r < |2| < s}, we can take E to be the closed disk {|z| s}, every point of which has distance at least s —r from £. Consequently, C\D is not connected, and D is not simply connected. The set {|z| (i). Proof that (i) —> (ii): This follows from our discussion of deformation of paths in Section IIL2. Let Pdz + Qdy be a closed differential on D. ‘According to the discussion in Section IIL2, f\ Pdr + Qdy = J, Pdz + Qdy for any closed paths 7 and o in D such’ that 7 can be deformed continuously to 0. Since the integral over a one-point path is zero, the simple connectivity of D implies that f, Pax + Qdy = 0 for any closed Pdx + Qdy is independent of path, hence exact Proof that (ii) => (ili): If every closed differential on D js exact, then in particular for fixed 29 € C\D the differential d arg(z — zo) is exact. We write d arg(z ~ zq) —dh(2) for some smooth function A(2), normalized so that (1) is a value of arg(s1 ~ 29) for some fixed point 2 < D. ‘Then hz) is a continuous determination of arg(z ~ 20) in D, and log|2 ~ zo] + éh(z) is an analytic branch of log(2 — 2) in D. Proof that (iii) <> (iv): If there isan analytic branch f(z) of log(~z0), then malt 9 for any closed path 7 in D. Conversely, if W(7,20) = 0 for any closed path 7 in D, then the analytic differential d=/(z — 29) is independent of path, and there is an analytic function f(+) such that df = dz/(z ~ 20). ‘An analytic branch of log(= — =9) is obtained by adding an appropriate constant to f(2). Proof that (v) —> (iv): Let 7 be a closed path in D. The winding number W(7,¢) is a continuous integer-valued function on C\D that is zero for large ¢. If we set it equal to 0 for ¢ = 00, we obtain a continuous integer- valued function on C*\D that vanishes at 0. Since C*\D is connected, the function is identically zero. Hence W(,¢) = 0 for ¢ € C\D. Proof that (iv) —> (v): If C"\D is not connected, then there are a closed bounded subset E of C\D and 5 > 0 such that |z ~ w| > 45 for every point 2 ¢ E and every point w ¢ F = C\(DUB). Fix a point 2 € E, and cover the complex plane by a grid of squares of side length 5 ‘0 that 2 lies in the interior of one of the squares. Let K be the union of all the (closed) squares in the grid that contain a point of distance at most 6 from E. Then K is a finite union of closed squares, K is disjoint from F, and the boundary 2K of K does not contain points of either E or F. Thus 8K is a finite union of straight line segments contained in D. Further, if we denote by KC, the squares from the grid in K, then fever - Df, Wr, 20) = 2%) = 2m,256 VIIL_ The Logarithmic Integral since the only summand that is not zero is the integral over the boundary of the square containing 9. Now, 9K is a union of a finite number of closed paths in D. Thus there is a closed path in D obtained by following consecutively sides of squares in the grid, for which f,d arg(2 ~ 29) #0. Consequently, W(1, 20) # 0. oe Proof that (iv) => (i): Our proof is based on the “northeast corner” argument. We can deform any closed 7 in D to a path that follows the sides of a grid of squares of small side length 6 > 0. ‘Thus we fix 6, and we consider closed paths in D that are obtained by following consecutively sides of squares in the grid, so that the path can be represented as a sequence of sides, each with a direction. We proceed by induction on the number of utive sides in the path. Thus we assume that 7 is represented by a sequence of sides of the grid, and that any closed path formed by a shorter sequence of sides in the grid can be deformed in D to a point. We must show that 7 can be deformed in D to a point. ‘Suppose first that the closed path -y intersects itself, so that + is obtained by traversing successively two shorter closed paths, each of which starts and terminates at a vertex 29 of the grid. By our induction hypothesis, each of the shorter paths can be deformed in D to a point. By the lemma, each of, the paths can be deformed in D to the point 2» by means of closed paths that begin and end at zp. By traversing successively the paths in the two deformations, we cau then deform the path 7 In D to the point 2. We may suppose, then, that the closed path + does not intersect itself. In the path consider the vertices furthest to the right, for which Re(z) is the largest, and among these let 2x be the highest vertex, for which Im(z) is the largest. Reversing the direction of the path if necessary, we can assume that the path follows the horizontal edge from 2p ~ 8 to 2xE where it makes a right turn and proceeds down to wwe = 2x — im, then ‘makes another right turn and proceeds to wye ~6, as in the figure. Let R be the open rectangle with vertices 2p — 4, 2B, wwe, and wye — 4. Since W(7.6) = 0 for ¢ in the unbounded component of the complement of 3, we have W’(7,¢) = 0 just to the right of the segment from zs to wwe. By the Jump theorem for the winding number (Section 3), we have W(,¢) Tor ¢ to the left of the vertical segment, hence for all ¢ € R. It follows that R is contained in D. Further, by continuity we have W(7,¢) = —1 at all points of the left vertical side of R that are not on 7, so that none of the points of the boundary of R belong to C\D, and AR is also contained Exercises 287 in D. Let o be the path obtained from 7 by replacing the segments of + consisting of the three edges of the rectangle R by the vertical edge of R from 2xp — 5 to wye ~ 6. Then 7 can be deformed to a in D, simply by deforming the three sides of the rectangle to the fourth side. Since is shorter than -y, o can be deformed in D to a point. Thus + can also be deformed in D to a point. This completes the proof of the implication and of the theorem. exe" 8 fp eye = NE corner ik rb wre We In Chapter XI we will prove the Riemann mapping theorem, which as- serts that every simply connected domain in the plane except the plane itself can be mapped conformally onto the open unit disk. What we will actually establish in Section XI.3 is that if D # © and if (ii) holds, then there is a conformal map of D onto the open unit disk. From this it follows easily that a domain D satisfying il) is simply connected (see Exercise 7) tnd also that D has a connected complement. Thus the proof of the Rie- ‘mann mapping theorem will provide us with an alternative method to show that (iii) implies (3) and (). The proof given above has the advantage of being “elementary” in nature, whereas the proof of the Riemann mapping theorem depends on the arguably nonelementary notion of compactness of families of analytic functions. Exercises for VITLS 1. Which of the following domains in © are simply connected? Justify your answers. (a) D = {Imz > 0}\(0,%), the upper half-plane with € vertical slit from 0 to i. (b) D = {Im2 > O}\[i,2i), the upper half-plane with a vertical slit from é to 2i. (c) D = C\[0,-+00), the complex plane slit along the positive real axis. (d) D = C\(-1. 1, the complex plane with an interval deleted. 2. Show that a domain D in the extended complex plane C* = CU{oo} is simply connected if and only ifits complement C*\D is connected. Hint. If D 4 C*, move a point in the complement of D to co. If D=C*, first deform a given closed path to one that does not cover the sphere, then deform it to a point by pulling along ares of great circles. 3. Which of the following domains in C* are simply connected? Justify your answers. (a) D 1,1] the extended complex plane with,258 10. u. VIII The Logarithmic Integral an interval deleted, (b) D = C*\{—1,0,1), the thrice-punctured sphere, Show that a domain D in the complex plane is simply connected if and only if any analytic function f(2) on D that does not vanish at any point of D has an analytic logarithm on D. Hint. If f(z) #0 on D, consider the function * £(w) oye [Law = J, Foe) Show that a domain D is simply connected ifand only ifany analytic function f(z) on D that does not vanish at any point of D has an analytic square root on D. Show that this occurs if and only if for any point 29 ¢ D the function z— zp has an analytic square root on D. Show that a domain D is simply connected if and only if each con- tinuous function f(z) on D that does not vanish at any point ot D hhas a continuous logarithm on D. Let E be a closed connected subset of the extended complex plane C*. Show that each connected component of C*\E is simply con- nected. Show that simple connectivity is a “topological property,” that is, if U and V are domains, and y is a continuous map of U onto V such that "is also continuous, then U’ is simply connected if and only if V is simply connected, |. Suppose that f(=) is analytic on a domain D, and f"(2) has no zeros on D. Suppose also that f(D) is simply connected, and that there isa branch o(w) of f-" that is analytic at wp = f(¢o) and that can be continued analytically along any path in f(D) starting at wo. Show that f(2) is one-to-one on D. We define an integral I-cycle in D to be an expression ofthe form o = Sky75, where 71,-.- im are closed paths in D and ky,... km are integers. We define the winding number of « about ¢ to be W(o,¢) = SkjW(y.6), ¢ € C\D. Show that if h(C) is a continuous integer-valued function on C*\D such that (20) = 0, then there is an integral I-eyele @ on D such that W(a,<) = h(6) for all © C\D. An integral L-cycle o is homologous to zero in D if W(0,<) = 0 ‘whenever ¢ ¢ D. Let U be a bounded domain whose boundary con- sists of a finite number of piecewise smooth closed curves 71,--- +m: oriented positively with respect to U, such that U together with its Exercises 250, 2 boundary is contained in D. Show that the L-eycle AU = 3, is homologous to zere in D. Let D be a domain in C such that C*\D consists of m +1 disjoint. closed connected sets. Show that there are m piecewise smooth. closed curves 71,-.- 7% such that every integral I-cycle @ can be expressed uniquely in the form a = 09 + 57 y7j, where the ky’s are Integers and a9 is homologous to zero in D. Remark. The yj°s form. a homology basis for D.IX The Schwarz Lemma and Hyperbolic Geometry This short chapter is devoted to the Schwarz lemma, which is a simple consequence of the power series expansion and the maximum principle. ‘The Schwarz Jemma is proved in Section 1, and it is used in Section 2 to determine the conformal self-maps of the unit disk. In Section 2 we formulate the Schwarz lemma to be invariant under the conformal self-maps of the unit disk, thereby obtaining Pick’s lemma, This leads in Section 3 to the hyperbolic metric and hyperbolic geometry of the unit disk. 1. The Schwarz Lemma ‘The Schwarz lemma is easy to prove, yet it has far-reaching consequences, ‘Theorem (Schwarz Lemma). Let f(2) be analytic for |2| <1. Suppose {f(2)| <1 for all|2| <1, and §(0) = 0. Then (1a) WEI < be Further, if equality holds in (1.1) at some point 29 # 0, then f(2) some constant of unit modulus. lel< a, dz for For the proof, we factor f(z) = zg(z), where g(2) is analytic, and we apply the maximum principle to (2). Let r <1. If |z| =r, then [g(s)| = \f(2)\/r <1/r. By the maximum principle, |g(2)| < 1/r forall = satisfying [2 |f(0)| Hint. Let v(z) = (2 ~ 29)/(1 ~ 702), which is ¢ fractional linear transformation mapping the unit disk onto itself, and show that 1G) < WED".262 IX. The Schwarz Lemma and Hyperbolic Geometry 3. Suppose that f(s} is analytic for z| < 1, and suppose that 1 < |f(2)] 1/M. Hint. For the second assertion, consider U(/(2)), where v(t) is 4 fractional linear transformation mapping 1 to 0 and the circle {|w| = M} to the unit circle, Or use Exercise 2. 4. Suppose that f(z) is analytic for |2| <1 and satisfies f(0) = 0 and Re f(z) < 1. (a) Show that |f(2)] < 2Jz|/(1 ~ |z|). Hint. Consider the composition of f(z) and the fractional linear transformation ‘mapping the half-plane {Rew < 1} onto the unit disk. (b) Show that |f"(0)| <2. (c) For fixed zo with 0 < |z9| < 1, determine for which functions f(z) there is equality in (a). (d) Determine for which funetions f(2) there is equality in (b). (e) By sealing the estimates in (a) and (b), obtain sharp estimates for |g(2)| and |g'(0)| for functions (2) analytic for |2| < R and satisfying 9(0) = 0 and Reg(z) , then |/(2)] > (r—|2|)/(1 ~ rl2) for |2| O and and show that the maximum is 2r, attained only for dz |=. Show that any conformal self-map of the upper half-plane has the form aztb {> oae Imz>0, where a,),cyd are real numbers satisfying ad~be = 1. When do two such coefficient choices for a,b,c,d determine the same conformal self-map of the upper half-plane? 6. Show that the conformal maps of the upper half-plane onto the open unit disk are of the form Ima>0, 0 in the hyperbolic metric by (62) hyperbolic length of = 2 f {dz} 2 ‘The factor 2 is a harmless factor, which is often omitted. (It adjusts the metric so that its curvature is 1.) The identity (3.1) shows that f 07 has the same hyperbolic length as for any conformal self-map f(z) of D. ‘Thus hyperbolic lengths are invariant under conformal self-maps of D. We define the hyperbolic distance p(z9,21) from z to 2; to be the infimum (greatest lower bound) of the hyperbolic lengths of all piecewise smooth curves in D from zo to 21. Since conformal self-maps of D preserve the hyperbalic lengths of curves, they also preserve hyperbolic distances; that is, for any conformal self-map w= f(z) of D, p(f 0), F(21)) = (20.7), 20.21 € D. ‘Theorem. For any two distinct points z9, in the open unit disk D, there is a unique shortest curve in D from 2 to 2 in the hyperbolic metric, namely, the arc of the circle passing through z and 2 that is orthogonal to the unit circle. The paths of shortest hyperbolic length between points are called hy- perbolic geodesics. The hyperbolic geodesics play the role that straight lines play in the Euclidean geometry of the plane. They satisfy all the ax- joms of Euclidean geometry except the parallel axiom (that through each point not on a given line there passes a unique stiaight line Uough the point and parallel to the given line) nyperboic geodesics For a proof of the theorem, let w = f(z) be a conformal self-map of D such that, f(z0) = 0. By multiplying by a unimodular constant, we can arrange that f(=:) =r > 0. Since f(2) preserves hyperbolic lengths, and268 IX. The Schwarz Lemma and Hyperbolic Geometry since (2) maps circles orthogonal to the unit circle onto circles orthogonal to the unit cizele, it suffices to show that the straight line segment from 0 to r is a unique path of shortest hyperbolic length from 0 to r. For this, let y(t) = x(t) + iy(t), 0 < t < 1, be a piecewise smooth path in D from 0 tor, Then a(¢) = Re(7(t)) = 2(#) defines a path in D from 0 to r along the real axis, and lal * lied f' iat < f [im ~ [iSee < fhe < [te If y(t) £0 for some t, then (0) > [| and the fist inequality above is strict. In this eave, the path a(2) on the rel axis is strictly shorter than the path 4(¢). Further, if a(t) is decreasing on some interval, we could reduce the integral by deleting a parameter interval over which a(@) starts tnd ends atthe same valve. We conclude that the integral is « miniraum txactly when 7(2) is teal and nondecreasing, in which ease the path is the straight line segment from 0 tor. "We turn now to an important reinterpretation of Pick’s lemma, dz] ‘Theorem. Every analytic function w = f(2) from the open unit disk D to itself is a contraction mapping with respect to the hyperbolic metric p, (3.3) p(F(20),f(41)) < plzz), 2 €D. Further, there is strict inequality for all points zo, 21 € D, z9 # 21, unless {J (2) is a conformal self-map of B, in which case there is equality for all 20.21 €D. ‘To see this, let be the geodesic from zo to 21. Then fo7 is a curve from ‘F(zo) to f(z1). Pick’s lemma and the definition of the hyperbolic metric yield Pf (20), f(2)) Mw af ldwl_ gf Fee rn If f(2) is not a conformal self-map of D, there is strict inequality in Pick’s Iemma, and we obtain strict inequality in this estimate, hence in (3.3) The hyperbolic distance from 0 to 2 can be computed explicitly. It is dt He 1 (28) je2f' 4 = | « = . ee of rae {lati der ‘This shows that the hyperbolic distance from 0 to z tends to +e when = tends to the boundary of the unit disk ‘A geodesic triangle is an area bounded by three hyperbolic geodesics. Since the hyperbolic geodesics and the angles between them are preserved Q es by conformal self-maps of , we can map any geodesic triangle to a triangle with vertex at 0 and with the same angles between sides. For a geodesic triangle with vertex at 0, two of the sides are radial segments, and the third is an arc of a circle lying inside the Euclidean triangle with the two radii as sides. From this representation we see that the sum of the angles of any geodesic triangle is strictly less than x, which is the sum of the angles of the corresponding Euclidean triangle. In connection with complex analysis, we have now been in contact with three spaces with strikingly different geometries. The first space is the complex plane C with the usual Euclidean metric |dz|. In the Euclidean plane, the geodesics are straight lines, and the sum of the angles of a geodesic triangle is exactly equal to =. The second space is the open unit disk D with the hyperbolic metric 2|dz|/(1 — 2/*). For the hyperbolic disk, the geodesics are arcs of circles orthogonal to the unit circle, and the sum of angles of a geodesic triangle is strictly less than 7. The third space is the extended complex plane C* = CU {o0} with the spherical metric, which can be introduced in a manner completely analo- gous to the hyperbolic metric. Recall (Section 1.3) that the chordal metric induced on C by the Euclidean metric of the sphere via the stereographic projection is given explicitly by chordal distance from 2 to w 2iz—w) Vie PVT I lw? ‘The infinitesimal form of this metric is 2d=|/(1+/2/*). If is a path in C*, its length in the spherical metric is dz} | hol spherical length of 7 = 2 Mf =2 at ” [1+ [2/7 1+h@P This is the length of the corresponding path on the unit sphere in RS ‘The distance from z; to 2 in the spherical metric is defined to be the infimum of the spherical lengths of the paths joining 21 to 2. Since the chordal metric is invariant under rotations of the sphere, s0 is the spherical metric, and consequently, the lengths of paths and the distances between points in the spherical metric are invariant under rotations. It is not difficult to show that the geodesics in the spherical metric correspond to great circles on the sphere, and the sum of the angles of a geodesic trlangle is strictly greater than r.270 IX. The Schwarz Lemma and Hyperbolic Geometry Each of these three spaces is homogeneous, in the sense that: any pre scribed point can be transported to any other by an “isometry.” Uhus for each of these spaces, any scalar quantity that is invariant under isometries is constant. It turns out that a notion of “scalar curvature” can be asso ated to each of the spaces (see the exercises), and the curvature is invariant under isometries, so that in each case the curvature is constant, The Eu- clidean plane has constant zero curvature, the sphere has constant positive curvature, and the hyperbolic disk has constant negative curvature. The curvature can be related to the area and the sum of angles of geodesic tri- angles (Gauss-Bonnet formula). We summarize these properties in tabular form, Geometry Euclidean Spherical “Hyperbolic 2ide| Infinitesimal length id fee Oriented isometries lz rotations conformal sel:aps Curvature 0 “1 “1 Geodesies lines great cles —_citles unit circle Antes of triangle > 0 consists of all = € D such that p(z,z) < p. (a) Show that the hyperbolic disk centered at 0 of radius p is a Euclidean disk of radius r = (e ~1)/(c? +1). (b) Show that any hyperbolic disk is a Euclidean disk. 3, Denote by e(=,p) and r(2,p) the Euclidean center and Buclidean radius of the hyperbolic disk centered at z of hyperbolic radius p. {a) For fixed p, show that r(z,)/(1—|21) tends to a constant A > 0 as [2) 1. (b) For fixed p, show that |2~e(3.)|/r(z,p) tends toa constant B,0< B <1, as |2| 1. Exercises m 4. Show that the circumference of a hyperbolic disk of radius p is 2nsinhp. Hint. Show first that the hyperbolic circumference of a Euclidean disk of radius r centered at 0 is 4rr/il — r?), 5. We define the hyperboli area of a subset E of D to be amit Show that the hyperbolic area is invariant under conformal self maps of D. Show that the hyperbolic area of a hyperbolic disk of radius p is given by aefeahp ~ 1) = nets Boh v0) 6. Establish the following, for the spherical metric. (a) The circumfer- ence of a spherical disk of radius p is 2rsin p, 0 < p o. Show that the horizontal strip S hyperbolic metrie /2 0 this extension is r* which is analytic. If k < 0 this extension is r-Fet#? = 2(-®), which Is conjugate-analytic hence harmonic. Proceeding by linearity, we see that the trigonometric polynomial A (e!) = S7t._y axe" has the (unique) ynometric monomial e**. A harmonic extension hoe ae oma 1, The Poisson Integral Formula 275 harmonic extension (re) We capture the coefficient am, by multiplying h (e™) by e~'"® and integrat- ing, The orthogonality relations for complex exponentials (Section VL6) viele am fre em Substituting this expression into the formula fr h (rel), we obtain ire) = So ([" nyetyenieee) ptige ire) = SO (five Je se) le menor] # In order to simplify this expression, we introduce the Poisson kernel function defined by uy pa = So Meo For each fied p < 1, this series converges uniformly or r < p and ~n < 8-< zr, by the Weierstrass M-test, since then [Mle < glM. In terms of the Poisson kernel, the formula for the harmonic extension h (re!®) becomes (1.28) here) = [" nor) Ro-oE, re ev. If. we make a change of variable y ~ @ —y end use the 2=-periodicity of P,(6), we obtain an alternative form of (1.23), (2) ive) = [a(n Se, ree. Now we look at the Poisson kernel function P,(@) more closely. Setting 2 = re and setting j = —k when k <0 in (1.1), we obtain PA) = 14 oaks Ow ‘Summing these two geometric series, we obtain (2.3) PQ) = 1276 X Harmonic Functions and the Reflection Principle Putting this over the common denominator fs? = @—2)(1—2) = 147? 2rc0s0, ‘we obtain pera Ter arews’ 77 EP a4) P(0) Since the Poisson kernel is 27-periodie, we focus on its behavior on the period interval —s < @ < . From the Poisson integral representation formula (1.2b) for the constant function h = 1, we have (1s) [riof =1 |, ae ‘Three other properties of the Poisson Kernel, which follow from the formula (1.4), are as) P,(0)>0, -n 20m, az PA-0)= PO, —n< 0m (1.8) P,(6) is increasing for —r < @ <0 and decreasing for 0 < @ 0, we obtain from (1.8) the following key property of the Poisson kernel (19) for fixed 6 >0, max{ P-(0) : 6 < [0] 0. Let g (e#) = Saxe" be a trigonometric polynomial such that |h(€#) —9(€%)| < & for all e € aD. Then fi(z) ~ 9(2)] < e for all 2€D. Now, g(re”) = Cagrle'# attains the boundary values 9 (e) continuously on 8D. Hence the composite function equal to h (e”) on D and f (e') on D can be approximated to within « by a continuous function Consequently, the values h(z) cluster to within of h (ei®) as z € D tends toe € OD. Since « > 0 is arbitrary, h(z) tends to h (e#) as z € D tends to el? € AD. ‘The second proof depends only on properties (1.5). (1.6), and (1,9) of the Poisson kernel, Again let ¢ > 0. Choose M > 0 such that |h (et) | 0 so small that |h (8) — h(e)| 0, 30 A(z) +h (ef#) ns 2 8, Exercises 279 Exercises for X.1 1. Show that the Fourier coefficients of P,(6) are ce = riM, 00 < k < co. (Here we fix r, 0 0, and let h (Re) be a continuous function on the circle {\21 = R}. Show that the function ; = Ra iy = [ =— 2" i L B47? WReelo—o) harmonic on the disk {|z| < R} and has boundary values h (Re ‘on the boundary circle de n(Rere) $2, alk 3. Suppose that f(z) = u(z) + iv(z) is analytic for |z| < 1 and that wt0) eens cob costa ot ad dk (haf Show ra ae y= ff wey SE sto, en Remark This i he Sehware formula, exresing an nate tic rs Dundee i al Pr 4. Let {fa(2) = mn(2) + ima(z)} be a sequence of analytic functions on the open unit disk D such that up(z) extends continuously to JD, tun (e®) converges uniformly on 9D to u (eH), and uq(0) converges Show that fa(2) converges normally on D to an analytic function F(z) whose real part is (2). Hint. Use the Schwarz formula, 5. Let h (e) bea piecewise continuous function (or an integrable func- tion) on the unit circle. Show that the Poisson integral h(=) tends to h{¢) as 2 © D tends to any point ¢ of the unit circle at which (eis continuous. 6. A function f(z), = € D, is said to have radial limit L at ¢ ¢ 0D if ‘F(r6) + Las r increases to 1. Let h (e%*) be a piecewise continuous Tunction on the unit circle. Show that (2) has a radial limit at each € AB, equal to the average of the limits of h (e") at ¢ from each side. 7. For each t > 0, define the kernel function C;(s) on the real line by ta zo wc sco aoa Cs) = For h(¢) a bounded piecewise continuous function on the real line define 4 function f(s + it) on the open upper half-plane Ht by ie+it) = f Cis On e)dg, s+ite HL280 XX. Harmonic Functions and the Reflection Principle (a) Sketch the graph of Ci(s) for t= 1, t= 0.1, and t= 001. (b) Show that Crs) > Oand Jo, Cy(s) ds ~ 1. (Thus Cys) isa prob- ability measire) (c) Show that for each 8 > 0, fgg) Cela) ds — 0 as t 0. (Thus C,(3)ds is an approximate ident) {d) Show that f(s + it isa bounded harmonic function in the upper half-plane (6) Show that (s+ it) —» M&) as sit © Ht tends to gy, whenever A(€) is continuous at £o. Remark. The kernel function C;(s) is the. Poisson kernel for the upper half-plane 8, Let Ae) be a continuous complex-valued function on 2D, and let 12) be its Poisson integral (a) Show that le = mt fo he") zeD, we Siw = nt [Were zeD. Hint, Use the identity PO) = 1+ Tose * Tee" ‘and justify differentiating under the integral sign. (b) Show that if |a (e!)| 0, define h(:) = e712, Show that if pis a zero of the Bessel function Jp(2), then the value of A(z) at any point is its average over the circle of radius r = 1 centered at the point, h(a) -f h(e tel) Use the Schldmilch formula (Exercise VL13). jdu(zo.0), wo zeC, 4. Suppose that r1,r2 > 0 are such that rz/r; is a quotient of two positive zeros of the Bessel function Jo(z). Show that there is a continuous function on the complex plane such that the averages of the function over the circles of radius r; and rp centered at any point z coincide with the value of the function at 2, yet the function is not harmonic, Use the fact (easily derived from the differential equation in Section V4) that Jo(z) has zer0s on the positive real axis. Remark. The condition is sharp. Ifr2/ry is not the quotient of tsvo positive zeros of the Bessel function Jo(2), then any continuous function on the complex plane that has the mean value property for circles of radius r) and ra is harmonic, by a theorem of L. Zaleman, 5. Suppose that ry,r2 > 0 are such that ra/ry is a zero of the Bessel function J,(z). Show that there isa continuous function f(z) on the complex plane such that [, f(z)dz = 0 forall circles in the complex plane of radius ry and 7, yet f(2) is not analytic. Suggestion. Proceed as in Exercises 3 and 4, and use the fact that Jy(2) has zeros on the positive real axis. Again it turns out that the condition is sharp. 3. The Schwarz Reflection Principle Suppose a given function is analytic on a domain that abuts on one side of an analytic curve. The Schwarz reflection principle asserts that under certain conditions, the analytic function extends analytically across the curve, and further, there is a formula for the extension. The formula reflects the function analytically to the mirror image of the domain, and the nub of the problem is to establish analyticity on the axis of reflection. At its core, the Schwarz reflection principle is a theorem about harmonic functions. We begin by reflecting harmonic functions across the real line. 3. The Schwarz Reftection Principle 283 First note that if u(2) is harmonic in a domain D, then u*(z) = u(2) is harmonic on the reflected domain D* = {2 : z € D}. Indeed, if u(z,) has the mean value property on circles, then so does u(z,—y). Another way to see this is to note that the change of variables (x,y) =» (x,-y) does not change the Laplacian operator . ‘Theorem. Let D be a domain that is symmetric with respect to the real axis, and Jet D+ = Dn {Imz > 0} be the part of D in the open upper half-plane. Let u(z) be a real-valued harmonic function on D* such that u(z) +0 as 2 € D* tends to any point of DMR. Then u(z) extends to be harmonic on D, and the extension satisfies (3.1) uz) = -ulz), 2D. For the proof, note that D is the disjoint union of D* = D/{Imz > 0}, its reflection D- = D1 {Imz <0}, and DAR. We extend u(z) to D by defining u(z) = —u(2) for 2 € D~, and setting u(z) = 0 for z € DAR. Then u(2) is continuous on D, u(z) is harmonic on D* and on D™, and (3.1) holds. We claim that the extended function has the mean value property. Fix zo € D. If z9 isin D* or in D~, then u(z) has the mean value property for small disks centered at 29, since u(2) is harmonic near 29. If 29 € DOR, Ue by (3-1) the average value of u(2) over the top half ofa circle centered at zo cancels the average over the bottom half of the circle, and the average of u(z) over the circle is 0, which coincides with u(zo)- Thus u(z) has the ‘mean value property on D, and u(z) is harmonic on D. (Nos CT oe ‘There is a corresponding result for analytic functions. First note that if f(z) is analytic on a domain in the upper half-plane, then (2) = (2) is analytic on the reflected domain in the lower half-plane. This can be checked using the Cauchy-Riemann equations or a power series expansion. ‘We can think of g(2) as the composition of an anticonformal map 2 2, @ conformal map f(2) (except where f'(2) = 0), and an anticonformal map w+ a; hence it is conformal.24 X Harmonic Functions and the Reflection Principle Theorem. Let D be a domain that is symmetric with respect to the real axis, and let D* = D0 {liz > 0}. Let f(z) be an analytic function on D* such that Im f(2) +0 as 2 © D* tends to DOR. Then f(z) extends to be analytic on D, and the extension satisfies (3.2) J@) = FB, zed. For the proof, write f(z) = u(2) +in(2) in D*. By the preceding theorem, v(2) extends to be harmonic on D and satisfies v(2) = —v(2), = € D. Fix a point to € DOR, and let Dy be a disk in D centered at x9. Since v(2) is harmonic on Dp, it has a harmonic conjugate on Do, and consequently f(2) extends to be analytic on Dp. The function f(z) is also analytic on Do, and it coincides with f(2) on the real axis, so by the uniqueness principle it coincides with f(2) in Do. Hence (3.2) holds for 2 € Do. If we now extend J(2) to D~ = Driflmz <0} by setting f(z) = (2), 2 € D~, the extended ‘F(2) is analytic on D~ and coincides with the analytic continuation of f(2) across DAR from D*, and so is analytic on all of D. Further, it satisfies (82) for all 2 € D, ‘The above theorem is somewhat easier to prove if it is assumed that f(2) extends continuously from D* to DVR and is real-valued there. In this case, formula (3.2) defines a continuous extension of f(z) to all of D, which is analytic on D~, and we can use Morern's theorem to show that it is also analytic across DOR. We define a curve 7 to be an analytic curve if every point of 7 has an ‘open neighborhood U for which there is a conformal map ¢ + 2(¢) of a disk D centered on the real line R onto U, such that the image of DMR coincides with U7, We also refer to such a as an analytie are. The conjugation Gr ¢ induces a map z+ 2* of U onto itself, by 2(¢)" = =(¢). The map is an “involution” in the sense that the map followed by itself is the identity, 27 = 2, Further, 2" = if and only if 2 €. The map ¢—~ ¢ interchanges the top half and the bottom half of D\R, which are the two components of D\R, so the map z+ 2" interchanges the two components of U'\7. We refer to these two components as the neighborhoods of the sides of 7, ‘and we refer to the map 2 + <* as the reflection across 7. Since the map G+ Cis anticonformal, and the composition of an anticonformal map and conformal maps is anticonformal, the reflection 2 2* is anticonformal cao (<\ ~~ ee ‘The reflection 2 — 2* is unique, in the following sense. If there is an ‘open neighborhood of a point 2» € 7°U and an anticonformal map = + 2* 3. The Schwarz Reflection Principle 285 that fixes each point of + in the neighborhood, Indeed, the map zr (3*)" is the composition of two anticontormal maps, 80 it is conformal, hence analytic (Section V.8). Since it fixes points of 7, the ‘uniqueness principle shows that it is the identity map. Thus (2* and 2° = (:4)"* = (2**)" = 2. We can extend the involution 2 + 2* to a neighborhood of the entire analytic arc 7, by covering with small coordinate sets U and using the uniqueness principle to see that the involutions agree on overlapping U's ‘Example. The map =(¢) = (¢~i)/(C+#) sends the extended real line onto ‘the unit circle. It induces a reflection across the unit circle, z= C-aC +8) = CHDAT-H = 1/2 = 2/2. From this formula we see that the reflected point 2° lies on the same ray through the origin as z. The reflection z +» 2* maps the circle centered at 0 of radias rto that of radius 1/r, sending re! to er lane Thlane a A more general version of the Schwarz reflection principle permits us to reflect, harmonic and analytic functions across analytic curves. The idea is that if w = f(z) ts defined and analytic on one side of an analytic arc , ‘and ifthe values w tend to another analytic arc ¢ as z tends to 7, then f(z) extends anzlytically across +, and the extension satisfies f(2") = f(2)" in a neighborhood of 7. (Here we use the same notation for both reflections.) It is a straightforward matter to justify the general version by coordinatizing both analytic arcs and reducing to the case of reflection across the real line.286 X Harmonic Functions and the Reflection Principle Rather than state a precise theorem, we derive a version of the principle ‘that wil be used later. Let D be a domain. An analytic are 7 C OD is a free analytic bound- ary arc of D if every point of 7 is contained in a disk U such that U'\y has two components, one contained in D and the other disjoint from D. Example. Suppose D is obtained from the open disk {|| < 2} by excising a sequence of disjoint closed disks from the top half accumulating on the interval [-1, 1], as in the figure. Then the boundary circle of each excised disk is a free analytic boundary are of D, but the analytic are (—1,1) is not a free analytic boundary are ‘Theorem. Let D be a domain, and let 7 be a free analytic boundary arc of D. Let f(z) be analytic on D. If |f(z)| > 1 as z © D tends to 7, then f(z) extends to be analytic in a neighborhood of 7, and the extension satisfies (33) fe") = FH in a neighborhood of », where z ++ =* is the reflection across ~ For the proof, fix point 20 € +, and consider a conformal map ¢ + =(¢) of a disk Dp centered on R onto an open set containing = s0 that Dp 0 corresponds to U7. Then 9(¢) = log f(2(¢)) is defined and analytic on fone side of Dy MR. Further, Reg(¢) = log f(2(G))| + 0 as ¢ + Do R. By the Schwarz reflection principle, (6) extends to be analytic on Dp, as does f(2(6)) =e). Consequently, f(2) extends analytically to U, and the extension is unique. To see that the extension satisfies (3.3), we argue ‘as follows. The function (=) = 1/f(s*) is the composition of f(z) and the two anticonformal maps 2 ++ 2" and w r+ 1/15. ‘Thus itis conformal wherever f"(z) #0. By the equivalence of analyticity and conformality (ection 1V.8), h(z) is analytic wherever "(2) # 0, and it is continuous, hence analytic also at the isolated points where '(z) = 0. Further, if 27, then 2” = 2, and [f(2)| = 1, 0 A(z) = f(z) on. By the uniqueness principle, h(:) = f(2) everywhere, and f(2) satisfies (3.3) Exercises os Exercises for X.3 1. Show that the reflection in the circle {|2 ~ zo] = R} is given by 2° = 29 + 2 — 20)/|2 ~ 20)”. 2. Show that a reflection in a circle maps circles in the plane to circles, 3. What happens to angles between curves when they are reflected in an analytic arc? 4, Suppose the curve 7 passing through 0 is the graph of a function y= A(z) that can be expressed as a convergent. power series h(x) = SE pane’, —r 0 and some constant ¢, where 29 is the center of the circle. 7. Show that if f(2) is meromorphie for |2| <1, and [f(z)| = 1 as |z| 11, then f(2) is a rational funetion. Show further that f(2) is the quotient of two finite Blaschke products. (For the definition of finite Blaschke product, soe Exercise IX.2.3.) 8, The modulus of an annulus {a < |2 ~ z9| < 6} is defined to be (1/27)log(b/a). (a) Show that any conformal map from one annulus centered on the origin to another such annulus extends to a conformal self-map of the punctured plane. (bb) Show that there is a conformal map of one annulus onto another if and only if the annul have the same moduli. (c) Show that any conformal self-map of the annulus {a < |z| < 6} is either a rotation z > e'¥z or a rotation followed by the inversion z+ abjz. 9. Let 7 be an analytic curve passing through the origin and tangent to the imaginary axis at 0. Suppose that 7(t) = it + bot? + bst! + 1-6 to the angle between the straight line segments at angles 0 and 8 XI Conformal Mapping In this chapter we will be concerned with conformal maps from domains onto the open unit disk, One of our goals is the celebrated Riemann map- ping theorem: Any simply connected domain in the complex plane, except the entire complex plane itself, can be mapped conformally onto the open unit disk We begin in Section 1 by reviewing and enlarging our reper- toire of conformal maps onto the open unit disk, or equivalently, onto the upper half-plane, In Section 2 we state and discuss the Riemann map- ping theorem. Before embarking on the proof, we give some applications to the conformal mapping of polygons in Section 3 and to fluid dynamics in Section 4. In Section 5 we develop some prerequisite material concern- ing compactness of families of analytic functions, which is at a deeper level than the analysis used up to this point. The proof of the Riemann mapping ‘theorem follows in Section 6, 1, Mappings to the Unit Disk and Upper Half-Plane Recall that a conformal map of a domain D onto a domain V is an ‘analytic funetion (2) from D to V that is one-to-one and onto. The composition of two conformal maps is a conformal map, and the inverse of a coniormal map is a conformal map. In this section we seek to find conformal maps from various domains D onto the open unit disk D. Such a map 2) is never unique, as we can compose it with any conformal self- map of D. However, if y(2) is any other conformal map of D onto D, then 9 = Woe is a conformal self-map of D for which 9 oy = v. Thus once fone conformal map ¢ of D onto D is known, the others are precisely the: maps of the form v= go, where g is a conformal self-map of D. ‘We have seen in Section IX.2 that the conformal self-maps of the open unit disk have the form gl) =» zeD, 250290 XI Conformal Mapping where al < 1 and |Aj = 1. The conformal seltmmaps of D are determined by three real parameters, namely, the real and imaginary parts of a, and the argument of the unimodular constant A. Consequently, a conformal map of D onto D is uniquely determined by specifying three real parameters. Specifying the image of a fixed point of D counts as two real parame- ters, corresponding to real and imaginary parts. Specifying the image of a boundary point of D counts as one real parameter. The fractional linear transformation w = (2 ~ é)/(2 +) maps the open upper half-plane Hf onto the open unit disk D. The problem of finding conformal map of a given domain onto D is equivalent to finding one onto HI, We can go back and forth between D and H by composing with the above map and its inverse 2 = i(1 +w)/(1 ~ w). The complex analyst does not distinguish between the half-plane and the unit disk, and works in whichever space the calculations are simpler. We turn to three specific classes of domains: sectors, strips, and lunes. Sectors. A sector can be mapped onto a half-plane with the aid of the power function 2%, for an appropriate choice of 3. Any sector with vertex at 0 can be rotated by the map z+ Az, |X| = I, to a sector of the form D= {0 ew eT Under this map, 2 = —ioo is mapped to w = +1 and 2 = +ico is mapped to Vertical lines in the strip in the z-plane are rotated to horizontal lines in the ¢-plane, which are carried by the exponential map to rays in the Eplane. These are carried to circles in the unit disk in the w- plane passing through +1. Horizontal lines in the strip are mapped to the ‘orthogonal trejectories ofthe circles passing through +1. These orthogonal trajectories are also ares of circles, passing from the bottom half ofthe unit circle to the top half. Note that the solution map for this problem is not unique, and different maps will give different images of horizontal and vertical lines. However, ‘once the images of —ico and +ioo are known, then the vertical lines in the strip will be mapped to atcs of circles through these two image points, and horizontal lines in the strip will be mapped to the ares of circles that are their orthogonal trajectories, Exercise. Find a conformal map of the vertical strip (-1 < Rez < 1} onto the open unit disk D that maps ~ioo to —1 and ‘+00 to Solution. ‘The desired map is obtained by composing the map just con- structed with a conformal selfmap a(w) ~ A(w~a)/(1 ~ at) of the unit202 XI_ Conformal Mapping disk. Since 2 = ~ioo goes to w = +1, we require that g(+1) = —1, and singe 2 = 100 goes to w = —L, we require that g(—1) = 1. Thus lea 0, find the conformal map w(2) of the open unit disk {|z| <1} onto the vertical strip {A < Rew < A} that sat- isfies w(0) = 0 and w/(0) > 0. Sketch the curves in the disk that correspond to vertical and horizontal lines in the strip. Find a conformal map w(z) of the strip {Imz < Rez +1 as Rez = oo, and w(z) + —1 as Rez —+ +00, Sketch the images of the straight lines {Rez = Imz +c} in the strip. What is the image of the median line (Rez = Im 2+ 1} of the strip? Find a conformal map w(2) of the right half-disk {Rez > 0, J2| <1} ‘onto the upper half-plane that maps ~i to 0, + to 00, and 0 to —1. What is w(t)? }. Let w = g(2) be the conformal map of the right aalf-disk {Rez > 0, [21 < 1} onto the entire unit disk that fixes the points 1 and +1. (a) Without computing g(:) explicitly, show that o(2) = 9(2) Hint. Argue that A(z) = 9(2) is another conformal map satisfying the same conditions, and appeal to uniqueness, (b) Use symmetry to show that g(0) = ~1. (In other words, use part (a).) (c) Find204 X1_ Conformal Mapping (=) as a composition of explicit conformal maps, and use this to check that 9(0) = —1 7, Find the conformal map of the pieslice domain {|arg 2| 0, let D be the domain obtained by slitting the upper half-plane along the vertical interval from 2 = 0 to = = ia. (a) Find a conformal map w() of D onto the entire upper half: plane such that w(z) ~ 2 as |z] + oc. Hint, Consider the preliminary map ¢ = 2°. (b) Describe how the map can be used to model the flow of water over a vertical metal sheet Iying in a flat river bed, perpen- dicular to the flow of the water. Give a rough sketch of the streamlines of the flow. 10. Find the conformal map w = f(z) of the exterior domain {|2| > 1}U{o0} onto C*\|-1, +1] such that f(o0) = oo and the argument of J (2)/2 tends to a as 2 ~ 20, where ais a fixed real number. Sketch the images of circles {|z| = r}, for r > 1, and the images of the intervals (—20, ~1] and [1,o0). What is the inverse map? (Speci the branch.) Hint. For a = 0, use the map (z + 1/2)/2 treated in (1.1) above, For the general ease, do a preliminary rotation. See also the exercises in Section IL6. 11, Show that the halfstrip {—n/2< Rez < x/2, Im2 > 0} is mapped conformally by w = sinz onto the upper half-plane. Sketch the images of horizontal and vertical lines 12. Find a conformal map of the half-strip in Exercise 11 onto the open Unit disk that maps —x/2 to ~i, x/2 to i, and 0 to +1. Where does co go under this map? 2. The Riemann Mapping Theorem ‘This section is devoted to a preliminary discussion of the Riemann mapping theorem. The proof will be postponed to the end of the chapter. The Riemann mapping theorem was first established in the following generality. by W.P. Osgood in 1900, 2. ‘The Riemann Mapping Theorem 295 Theorem (Riemann Mapping Theorem). If D is a simply connected domain in the complex plane, and D is not the entire complex plane, then there is a conformal map of D onto the open unit disk D. Concerning the hypothesis, recall from Section VIILS that a domain is simply connected if every closed curve in the domain can be deformed to a point in the domain. Several characterizations of simply connected do- ‘mains were given in Section VIILS. Roughly speaking, a domain is simply connected if the complement has no “holes.” Disks are simply connected, and annuli are not. We say that two domains are conformally equivalent if there is a conformal map of one onto the other. Thus the Riemann mapping theorem asserts that any simply connected domain in the complex plane C either coincides with C or is conformally equivalent to D, We refer to a conformal map w = (=) of D onto D as the Riemann map of D onto D. It is unique, up to postcomposing with a conformal self ‘map of D. To specify the Riemann map uniquely we must specify three real parameters. One way to do this is to specify (9) = 0 and (9) > 0 for some 29 € D. Each of the hypotheses in the Riemann mapping theorem is necessary. ‘The complex plane C cannot be mapped conformaliy onto any bounded domain, because according to Liouville’s theorem the only bounded ana- lytic functions on C are the constants. Since any closed curve in D can be deformed to a point, any closed curve in any domain conformally equiv- alent to D can also be deformed to a point, by composing deformation in D with the inverse of the Riemann map. Thus any domain conformally equivalent to D is simply connected. For a simply connected domain D in the Riemann sphere (the extended ‘complex plane), there are three possibilities. Indeed, if the domain is not the entire Riemann sphere, we can move a point in the complement of the domain to ov by a fractional linear transformation and reduce to the case of a simply connected domain in the plane, where there are two possibilities. The Riemann mapping theorem then yields the following, Corollary. A simply connected domain in the Riemann sphere is either the entire Riemann sphere, or it is conformally equivalent to the complex lane, or it is conformally equivalent to the open unit disk. Suppose now that D is a simply connected domain with Riemann map ‘¢(2) mapping D onto D. We consider the behavior of (2) as = approaches the boundary of D. For each fixed © > 0, the set {le(z)| <1 —<¢} is a ‘compact subset of D, which is at a positive distance from 9D. It follows that |y(z)| ~+ 1 as z— OD. Hence the harmonic function log|,(z)| tends to 0 as z tends to the boundary of D. The Schwarz reflection principle for harmonic funetions (Section X.3) then implies that log \p(z)| extends296 XI Conformal Mapping harmonically across any free analytic boundary arc of D. Consequently. ‘o(z) extends analytically across any free analytic boundary arc of D, and in fact we have the following. ‘Theorem. Let D be a simply connected domain in €, D # C. Then the Riemann map ¢(2) of D ontoD extends analytically across any free analytic boundary arc 7 of D, and y(=) maps one-to-one onto an arc of D. The extended function satisfies y"(2) # 0 for 2 € +, and y(z*) = 1/g(2) for 2 in ‘a neighborhood of 7, where z — 2" is the reflection across 7. Disjoint free analytic boundary arcs of D are mapped by ¢(2) to disjoint arcs of OD. For the formula for the analytic extension, see Section X.3. From the formula and the fact that |y(z)| < 1 on the side of 7 in D, we see that |i(2)| > 1 on other side of 7. ‘Thus near any point of 7 the locus where J(2)| = 1 consists only of points of the one analytic curve 7, 80 no point ‘of 7 can be a critical point of g(2). Let 20 € 7, and let @ = (zo). If Us is a small disk centered at 29, then y(U.) includes all points of D in a sufficiently small disk centered at Co. Since (2) is one-to-one on D, no other point of D is mapped by (2) to this disk centered at Go. ‘Thus no other point of 7, and no point lying on any other free analytic boundary are of D, is mapped to Go. It follows that (2) is one-to-one on 7, and further, the image of 7 is disjoint from the image of any other free analytic boundary are of D. Exercises for XI.2 1. Show that no two of the domains C*, C, and D are conformally equivalent, 2. Let (2) be a conformal map from a domain D onto the open unit disk D. For 0 0. ‘Thus the inverse function w = 9(2) of ¢(w) tends to the vertex wy as 2 € H tends to the interval [by,ai]. Now, the interval cannot have any interior points, or else g(2) ~ wo would tend to zero on the Interval, hence reflect analytically across, hence be identically zero, which Js absurd, Thus by = a). We conclude that y(w) extends continuously to the comer wp, and that ¢(w) maps the two segments in OD abutting at a ‘corner wp to two contiguous intervals on the real line meeting at the image (09) of the corner. sev) monnsaor — eme00 A. ZA hea by = 3 — ‘me Now we consider more carefully the behavior of the inverse map w = (2) of the upper half-plane onto the domain D, mapping the point ao € R to the corner wo, and mapping intervals on each side of ap onto two straight line segments in 9D terminating at wy, Let A =e", a unimodular constant298 X1_ Conformal Mapping, ‘The function w(¢) = wy + AC maps a semidisk {|¢| < ¢,1m¢ > 0} onto the part of D at the corner. We may express ¢ as a function of 2, by V/9G(z) = (g(2) — wy)"/@. Since Im ¢(z) — 0 as Imz — 0, the Schwarz reflection principle shows that ((2) is analytic at 2 = ao, and further, ag cannot be a critical point of ¢(2). Thus A ¢(2) = (2—ao) f(z) where f(2) is analytic at ag and f(ag) #0. Let h(z) be an analytic branch of f(2)°, 0 that A@ = (2 — ap)®A(2) and (ag) #0. Then from g(2) = wo + 2¢* wwe obtain (2) = wo + (2 a0)*h(2), where h(2) is analytic at ap and h(a) # 0. If we differentiate once, we eliminate wy. If we differentiate again, we obtain g'(z) = (2 — a9)" (ah(z) + O(z ~ a0), g"(2) = (z= a9)*-? (ala ~ 1)A(z) + O(z — ao) ‘and this allows us to eliminate the branch point by dividing, ate) _ a=1 g@) ~ 2 ‘Thus while g(z) has a branch point at a9, the combination of derivatives g'(2)/a'(2) extends meromorphically to ap and has a simple pole there with residue a — 1 ‘Something similar happens ifthe poitt oc is mapped to the corner 1p by g(z). We parametrize the corner by (as before, atd apply the Schwarz reflection principe, to see that X¥/%C(z) = (g(2) ~ wy)" depends analy cally on z at z = 00. In this case, \'/@¢(z) = f(z)/z, where f(z) is analytic at co and f(cc) #0. Then g(2) = uy + A(2)/2%, where h() is analytic at co and A(oo) # 0. A calculation similar to the one above shows that PO L240.) > Rims, ge) "= is analytic at oo and vanishes there. We have proved the following theorem on corners (aay + analytic, [zag] 0. a0 (3.2) ‘Theorem. Let D be a simply connected domain with a corner at w» € 0D with interior angle az, 0 0 2. If the angle at w, is 7, so that we have only an apparent vertex, then 0 —1 =0, and the corresponding factor disappears from the integral (as it should) 3, We can allow an angle of 2, corresponding to the tip of a slit. In this case we think ofthe sit as two intervals, one for each side ofthe sit. 44 We can allow intervals in @D to overlap, and we can allow one point to serve as a multiple vertex, asin the figures 5, We can allow oo to be a vertex, provided that we count the correspond- ing angle as negative. For instance, suppose 9(2) maps a; to the vertex wy = 00, with angle xay, where ~2 < ay <0. The map to the ¢-plane is then ¢(z) = 9(2)!/@ = ( —ax)¥(2), and we obtain g(2) = ( — a1)A(2) where h(2) is analytic at ay and A(a:) #0. We have (3.1) as before, and ‘we obtain the same Schwarz-Christoffel formula (3.5). This time the factor (2-4) of g'(2) is not integrable at a1, corresponding to the fact that (a1) = 00. 6. We can allow 00 to be a vertex with angle zero. This corresponds to the “end” of an infinite strip. Tn this case the map to the ¢-plane is given by ¢(z) = e%%) for some 8, and 1 1 ’ ale) = blogg = Floste a1) + A=), where h(2) is analytic at ay. A calculation reveals that (3.1) again holds. ‘We are led to the same Schwar2-Christoffel formula (3.5), with ay = 0 for the vertex at oo, 7. If ag = co is mapped to a vertex wy = oo, the Schwarz-Christoffel formula (3.5) holds, in which there is no factor corresponding to o0. The 3. The Schwarz-Christoffel Formula 301 calculations are similar to those above. The result can also be obtained by letting the parameters in the Schwarz-Christoffel formula tend to oo. (See Exercises 3 and 4.) Example. Let D be the domain obtained by cutting a vertical slit in the upper half-plane from 0 to ia on the imaginary axis. (See Exercise 1.9.) Let w = g(z) be the Schwarz-Christoffel map of the upper half-plane onto D thet sends 0 to ia and +1 to the two corners at 0. This determines (2) uniquely, since we have specified three real parameters. In the Schwarz Christoffel formula, the consecutive points are a; = —1 with angle x/2, 2 = 0 with angle 2x, and a3 = +1 with angle x/2. Thus (3.4) becomes ge) = Ale 1) Mele 4 1? = Ae VAT. This can be integrated in closed form, and after matching boundary points to determine A and the integration constant, we obtain g(z) = avi? 1, bie ‘the branch of the square root is the one that is positive on the interval 1,00). waa ae ar) 7 ° Example. Suppose D is a triangle, with vertices at 0, 1, and a point wy jn the upper half-plane, and interior angles am, 8x, and yx, where a+8+7=1. Fix a point a > 1, and let w = g(z) map the upper half- plane conformally onto D, with (0) = 0, g(1) = 1, and g(a) = w. Since ‘we have specified three real parameters, 9(2) is uniquely determined. Since (0) = 0, the Schwarz-Christoffel formula becomes (3.6) ate) = A faye ortae The constant A is determined by the condition 9(1) =02 XI Conformal Mapping Exercises for X13 1 Derive the formula (3.2) for the case that g(z) maps co to the cor- ner tp of D. Let (w) be the Riemann map of a simply connected domain D conto D. Suppose up is a comer of D with interior angle ax. Set w(C) = wo +XC®, where |A = 1. (a) Use the Schwarz reflection principle to show that y(w(C)) is analytic at ¢ = 0. (b) Show that ¢(w) has a series expansion 5” bj(w — wo)/* that converges uniformly for w € D, |w — wol <=. (c) Use (b) to show that ¢(w) has a limit as w € D tends to wo Denote by ga(z) the conformal map of H onto the triangle with vertices 0, 1, and wo given by (3.6), 80 that ga(0) = 0, ga(1) = 1 and ga(a) = tio. Show that gq(z) converges as a — +00, uniformly for |2| < R. Express the limit as an integral, and relate it to the ‘Schwarz-Christoffel map. Fix @ such that 0 0. Denote by f(z) the conformal map of onto the triangle with vertices 0, 1, and Re'®* given by (3.6), so that fr(0) = 0, fa(1) = 1, and fa(a) = Re Show that f(z) converges as R—» oo, uniformly for |z ~ a| > ce. Express the limit as an integral, and relate it to the Schware- Christoffel map. For fixed 0 << 1, show that ede w= FERRE maps the upper half-plane conformally onto a rectangle. Sketch the rectangle, using the notation ae ve at =f ogee © - [eee Show that the inverse z = 2(w) of the function defined above ex- tends to be a meromorphic function hu) on the entire complex plane. Show that hw) is az odd function, and that h(w) is doubly period with periods 4X and 2K". Locate the zeros aid the poles thn), and indicate their orders. Remark. The integrals appearing Shove ae elliptic integrals. The function h(x) isthe Jacobian elliptic anetion denoted classically by sn(u). Te maps the torus formed by identifying the opposite edges of the period rectangle two-tone onto the extended complex plane What is the Schware- Christoffel formula (3.4) for the conformal map 49(2) of the upper half-plane onto the horizontal strip {0 < Im(w) < Exercises 303 10. 0} that, maps 0 to —ce and oo to +e? By integrating, show that a(2) = g(t) + 2 Loge Use the Schwarz-Christoffel formula to find a conformal map w g(z) of the upper half-plane onto the vertical halfstrip {Imw > 0, -a < Rew 0, and let D, be the domain obtained from the upper half-plane by deleting the straight line segment from 0 to 7. (a) Find an integral expression for con- formal maps w = 9(z) of the upper half-plane onto D such that 9(0) = 7 and g(c0) = oo. (b) Show directly that if @ > 0 and 0 1} and the first quadrant {Rew > 0, Imw > 0}. Find a conformal (2) of the upper half-plane onto D satisfying (—1 Lie For c > 0, find an integral expression for a conformal map w = 9(2) of the upper half-plane onto the corridor with a turn deseribed by D = {0 0}U (0 0} In the case c= 1, find explicitly the conformal map g(z) that satis- fies 9(0) = 0, 9(1) = +00, and g(20) = 1 +4.sot XI Conformal Mapping 4, Return to Fluid Dynamics We return to fluid dynamics, to see how conformal mapping techniques can be applied to solve fluid flow problems. Following the notation of Section 111.6, we denote the complex velocity potential by f(z) = o(2) + iy(2), where o(2) is the velocity potential, and y(2) is the stream function ‘whose level sets are the flow lines of the fluid. The velocity vector field is V(z) = FG). The flow is tangent to the beundary along boundary curves fon which there are no sources or sinks. This (2) is a harmonic function ‘on D that is constant on each boundary curve with no sources or sinks, Example. The stream function (=) = Im= on the upper half-plane HE ‘corresponds to a flow parallel to the x-axis with no sources or sinks on the axis. The stream function (z) = Arg z on H corresponds to a flow from a source at the origin. We may superimpose she flows, to obtain the stream funetion U2) = erlmz+cArg2, Iz >0, which corresponds to a flow near the bank of a large river with a pipe discharging an effluent at one point on the riverbank As in Section IIL.6, our strategy for solving a flow problem on a domain D is to find a conformal map fi(w) from D to the upper half of the =-plane, in order to transfer the boundary problem to H. The transferred problem involves finding a harmonic function ¥(2) on ll that is constant on the Intervals of the real axis corresponding to the boundary’ ares of D where there are no sinks or sources. The composed function v(h(w)) is then the stream function for the flow in D. In the case that D is a polygon domain, the Schorarz-Christoffel formula gives the conformal map w = g(2) from Hi onto D, and this must be solved for the inverse function 2 = h(w) if a solution in closed form is required. Example. Consider the flow near the bank of a large river with a small parallel tributary. We represent the configuration by the upper half ofthe ‘w-plane with a horizontal slit terminating at the point ib on the positive imaginary axis, D = Hl\(ib~ 00, i6). Consider the conformal map w = g(=) of H onto D such that —1 corresponds to the tip of the slit, 0 to the source of the tributary, and ino to iae. ‘The three conditions 9(—1) = i, 9(0) = ~2c, and g{ice) = ico determine 9(2) uniquely. The function (2) maps the interval (—c0,—1] to the top edge of the slit, the interval (1,0) to the bottom edge reversing direction, and the interval (0,0) to the real axis (00,00). The Schwar2-Christoffel formula (4.3) becomes g(2) = A(z+1)z"*, corresponding to angle 2r at —1 and 0 at 0. The term involving —oe does not enter. ‘Thus 9(2) = A(=-+Log)+B. The constants Exercises 305 A and B are determined by the boundary conditions, with final result a a2) 1+2+Log2), — Imz>0. A flow function corresponding to a source in the tributary in the w-plane then corresponds to a flow function corresponding to a source at 2 = 0 in the z-plane. This is given by Arg z= Arg h(w), where h(w) is the inverse of the conformal map g(2). We superimpose this upon a flow in the river, to obtain a flow function of the form lw) = 6 Argh(w) +60 Im h(w), ‘where the constants ¢; and cy are adjusted to the magnitudes of the flows, Exercises for XI4 1. Let D be the domain obtained from the upper half-plane by deleting, the part of the imaginary axis above ib. Find a conformal map (2) of the upper half-plane onto D such that g(0) = ~o0, 9(2°} “oo, and g(—1) = i. Solve for z = A(w). With ¢(w) = arg(h(w)) fas a stream funetion, use the map to sketch the streamlines of the flow of water in a river as it passes under a vertical gate, Hint. In the Schwarz-Christoffel formula, remember to take the angles at co to be negative. An altemative method is to use an inversion to map the infinite vertical slit onto a finite slit from 0 to da. ‘The answer is a) = 3 (ve-Ze). ay = ro a wd ke Exercise | Exercise 2 Beenie 3 Ererised 4308 XI Conformal Mapping 2. Let D be the domain obtained from the vertical half-strip {Im 2 > 0, |Re2| < 1/2} by deleting the part of the imaginary axis above ia. Find a conformal map w = 9(2) of the upper half-plane onto D. Solve for z = h(w). Use the map to sketch the streamlines of the flow of water in a channel with a hairpin turn, Hint. Use the inverse sine function to map the half-strip onto an upper half-plane (see Exercise 11), and refer to Exercise 1 3, Find a conformal map w = g(2) of the upper half-plane onto the plane with two vertical slits from +1 to £1 + ioo deleted. Use the map to sketch roughly the flow of water near the mouth of a channel that empties into the middle of a large reservoir. Express the stream function for the flow in terms of the inverse z = h(w) of 9(2). 4. Let D be the upper half-plane with two vertical slits from +1+ib to £1 + 400, and let w = g(2) be the conformal map of the upper half- plane onto D satisfying w(0) =0, w(—1) = —co, and w(+1) Show that the map is symmetric about the imaginary axis, so that w(ioo) = foo. Express the map as an integral. Suppose in your integral formula that the point @ > 1 is mapped to the end 1+ i of the slit, Sketch the streamlines of the flow in the w-plane that ‘corresponds to the stream function y(2) = arg(z - a) ~ arg(z +0) in the 2-plane. What are its sources and sinks? 5. Compactness of Families of Functions We have reached a turning point. Our proof of the Riemann mapping the- orem requires some technical ingredients at a higher level of mathematical sophistication than what we have used heretofore. In this section we sum- rmarize some of the ideas involved in the Arzeli-Ascoli theorem, and we use it to prove Montel’s theorem (thesis grade). Montel’s theorem is one of the key ingredients of the proof of the Riemann mapping theorem given in the next section. For an expanded treatment of the background material, see any good introduction to mathematical analysis. Let E be a subset of the complex plane C, and let F be a family of complex-valued funetions on &. We say that F is equicontinuous at a point zo € E if for any € > 0, there is 6 > 0 such that if z € B satisfies 'z ~ zo| <6, then |f(z) ~ f(zo)| << for all f € F. This coincides with the “s.6” definition of continuity at a point 29, except that the same 5 must serve simultaneously for all functions f in the family F. We say that the family is uniformly bounded on £ if there is a constant M > 0 such that |J(2)| 0 such that |f"(z)| fm(2)) tends to 0 uniformly for 2 € Eas nym —» 00. A family F is equicontinuous with respect to the spherical metric at 2 E if for any € > 0 there is 5 > 0 such that if z € F satisfies |z — z9| < 4, then o(f(2), f(zo)) <« for all f € F. When we combine the Arzeli-Ascoli theorem with the diagonalization argument used above, we obtain the fol lowing version of the theorem, which we record for use when we return to this circle of ideas in Chapter XIV.308 XI Conformal Mapping ‘Theorem. Let D be a domain in the complex plane, and let F be a family of continuous functions from D to the extended complex plane C*. Then the following are equivalent. (1) Any sequence in F has a subsequence that converges uniformly on ‘compact subsets of D 1n the spherical metric. (2) The family F is equicontinuous at each point of D, with respect to the spherical metric. For now, our applications to complex analysis will be based on the fol- lowing preliminary version of Montel’s theorem, which was obtained by P. Montel in his thesis. We will prove a stronger version in Chapter XIV. Theorem. Suppose F is a family of analytic functions on a domain D such that F is uniformly bounded on each compact subset of D. Then every sequence in F has a subsequence that converges normally on D, that is, uniformly on each compact subset of D. If zo € D, there is r > 0 such that the closed disk {\z — 2o| A, then by Montel’ theorem, the J's have a subsequence that converges normally on D to an analytic function G(2). On account of the normal convergence, [G(2)| <1 and (G'o)] Theorem. Let D be a domain in the complex plane on which there is nonconstant bounded analytic function, and let 2 € D. Then there is an analytic function G(z) on D such that |G(z)| <1 for € D, and 7'Go)| < |G") for any analytic function f(z) on D satisfying |f(2)| <1 on D. Further, G(za) = 0 and G'(29) # 0. It remains only to establish the final statement. Let A(:) be a non- constant bounded analytic function on D. Then h(2) ~ h(z0) has a zero of finite order, say of order N, at zy. If = > 0 is tiny, the function H(z) = 2(h(2)—h{20))/(2—20)"~™ satisfies [f(2)] <1 and (20) # 0. Con- sequently, A = |G'(zo)| > 0. Finally, o(=) = (G(2) — G(20))/(1 — Glea)e) is the composition of G(z) and a conformal selfmap of A, so that g € F. From 16"(20)] Go) > Weal = Beye ‘we deduce that G(z0) = 0. ‘The extremal function G(2) is ealled the Alfors function of D. The Ahilfors function depends on 29. However, it can be shown (Exercise 9) that the Ablfors function corresponding to a fixed point 2» < D is unique, up to multiplication by a unimodular constant. The extremal value A = [G"(zo)| can be regarded as the best constant for which the Schwarz lemma (infinitesimal version) holds with respect to 2) € D. We will show in the hext section that when D is simply connected, the Ahifors function maps D conformally onto the open unit disk. Exercises for X15 1. Let {Jn(2}} be a uniformly bounded sequence of analytic functions on a domain D, and let 2 € D. Suppose that for each m > 0, 8" (29) + 0 as n+ 00. Show that f(z) —+ 0 normally on D. 2. Let {fa(2)} be a sequence of analytic functions on a domain D, and let 29 € D. Suppose Re f(z) > —C for all z € D, and f,")(z) +0 as n ~» 00 for each m > 0. Show that f(z) + 0 normally on D. Hint. Use Problem 1, which is a special case. 3, Let f(z) be a bounded analytic function on the horizontal strip {-1 < Imz < 1}, such that f(x) tends to 0 as x tends to +00. Show that for any © > 0, f(x + iy) tends to 0 as x tends to +00, uniformly for -1 +e 0, f(2) tends to 0 uniformly as z tends to 20 through the sector {= 1. (a) Show that {f,(2)} has a subsequence that converges normally to an analytic function f(z) on D. Hint. To estimate f(:), use the mean value property with respect to area (see Exercise m4). (b) Show that. fff [f(2)|de dy <1. (c) Show that if ff \fa(z)— fin(2)|dz dy — 0 as m,n —> 00, then SS \fnl2) ~ f(z)| dz dy +0 as n — 00. 9. Let D be a domain in the complex plane, and let F be the family of analytic functions f(s) on D such that |f(s)| <1 for 2 € D. A 6. Proof of the Riemann Mapping Theorem an function 9(2) is an extreme point of F if it is not the midpoint of a line segment in F, that is, if whenever 9 = (g9 + 91)/2 where 0.91 € F, then go 9 (a) Show that the function g € B is an extreme point of F if and only if the only analytic function h() satisfying |g(z)£(2)| < Lish (b) Show that the function g € F is an extreme point of F if ‘and only if the only analytic function A(z) satisfying |g(2)|-+ |A(2)| <1 is h=0. Hint. First show that if fa +4) <1, then lal + [/2 <1. (©) Suppose that there is a nonconstant bounded analytic function oon D, and fix 29 € D. Show that the Ahlfors function G(2) associated with 29 € D is an extreme point of F. (@) Show that the Ablfors function G(2) associated with z € D is unique, up to multiplication by a unimodular constant. (e) Let L = Dj4 ajeLsx be a finite linear combination of linear functionals of the form Lyx(f) = f(z), where the 2's are fixed points of D. Suppose there is a nonconstant bounded analytic function f on D such that L(f) # 0. Show that the extremal problem of maximizing ReL(f) over all f € F has a unique extremal function (2), which is an extreme point of F. 6. Proof of the Riemann Mapping Theorem Now we are ready for the proof of the Riemann mapping theorem. The kkey ingredients are Montel’s theorem, Hurwitz’s theorem, some elementary mapping properties of the square root function, and the fact that v2 is expanding on the open unit disk. We begin with the elementary mapping properties of the square root function. Suppose that D is simply connected and that D 4 C. Choose a € C\D. By the characterization of simple connectivity (Section VIIL8), there is an analytic branch g(z) of lox(=—a) in D. Then h(z) = e9)/? is an analytic branch of yz—a in D, and A(z)? = >a 40 in D. If A(z) = hea), then 21 = A(z)? + = (22)? +a = 22. Thus h(2) is univalent, and h(2) maps D conformally onto h(D). Finally, note that if wo € h(D), then ~wo ¢ h(D). Indeed, if wy = h(2o) and —w = h(2) for 20,21 € D, then 2 = A(z)? +a = ug +a = h(y)? +a = 2, which is impossible, We Lemma. Let D be a simply connected domain. Suppose a ¢ D, and let ‘n(z) be an analytic branch of Yz—a in D. Then h(2) is univalent on D, and further, h(D) is disjoint from ~h(D)..siz XI_ Conformal Mapping We remark that on the basis of the lemma we can show that any simply connected domain D # C is conformally equivalent to a subdomain of the init disk, where we have hyperbolic geometry at our disposal. Indeed, if the disk ju ~ wo| < ¢ is contained in f(D), then its negative is disjoint from h(D), $0 that |h(z) + wo| > « for 2 € D, and e/(h(2) + to) maps D conformally onto a domain contained in the unit disk. Our proof depends at least conceptually on hyperbolic geometry. Let ‘AC,£) be the hyperbolic metric in ID (Section IX.3). The function ¢?, regarded as an analytic function from D to D, is a contraction with respect to the hyperbolic metric, and it is a striet contraction on bounded subsets {Icl < 1 —£} of the hyperbolic disk, in the sense that there is ¢ < 1, depending on ¢, such that p (C2,€2) < ep(G,€) for I¢ilé| <1—e. Thus any branch of its inverse V¢ is expanding with respect to the hyperbolic metric, and it is strictly expanding on bounded subsets of the hyperbolic disk, in the sense that for any r <1, there is C > 1 such that, o(evieve) > Cole), — Iehigi sr (For a precise value for C, see Exercise IX.8.9.) The expanding property of Vi is the key to the following lemma, Lemma. Let D be a simply connected subdomain of D such that 0 € D. IfD 4D, then there is a conformal map ¥\(¢) of D onto a subdomain of D such that ¥(0) = 0 and |y"(0)| > 1 Let § € D\D, and let g be the conformal self map of D that maps b t00. Then q(D) is @ simply connected domain in the unit disk that does not contain 0. Hence there is an analytic branch h(é) of v& on g(D), and further, as shown above, his univalent on g(D). Let f be a conformal self-nap of D that maps h(g(0)) to 0, and set = fohog. Then ¥(6) is a conformal map of D onto a subdomain of B, and ¥(0) = 0. Since F and g are isometries in the hyperbolic metric, and his a strict expansion near 9(0), v is @ strict expansion near 0, that is, there is > 1 such that p(v(@),0) > CpA6,0) for [cl < Since p(v(C),0) ~ 2}8(6)| and p(C,0) ~ 2)¢| as ¢ + 0, we obtain cc < HO) OL ~ eXG,0) ic) and |y’(0)| > 1. This inequality can also be checked by expressing f and g explicitly and differentiating (Exercise 1). ‘ow we complete the proof of the Riemann mapping theorem by pro- ducing the Riemann map as an extremal function for an extremal problem similar to the problem considered in Section 2. We assume that D is a simply connected domain, D # C. Fix 29 € D, and let F be the family of univalent(!) funetions f(z) on D such that |f(2)| <1 for z € D and — WO. Exercises 313 ‘slzo) = 0. The family F is nonempty, since if h(z) maps D conformally ‘onto a bounded domain, then the function f(=) = 2((2) ~ h(29)) is in for © > 0 small. We consider the extremal problem of maximizing |f'(z0) over f € F. As before, set A = sup{if"(zo)l: fF} > 0, and let {f,(2)} be a sequence of functions in F such that |f%(20)| ~ A. By Montel’s theorem, the f,’s have a subsequence that converges normally on D to an analytic function (z). Clearly, \y(z)| < 1, and |y"(zo)| = A. ‘The functions fj are univalent, so by Hurwitz’s theorem (Section VIIL2), either ¢ is constant or y is univalent. Since y'(2o) #0, is nonconstant. and consequently, y is univalent, mapping D onto a subdomain of D. We claim that ¢(D) =D. Otherwise, we could apply the preceding Lemma to the domain y(D) and find a conformal map v(¢) of (D) onto a subdomain of D such that (0) = 0 and |y/(0)| > 1. Then yoy € F would satisty (© 9)'C20)| = lW'(O)e"(z0l > A, contradicting the definition of A. This ‘completes the proof. ‘We will give another proof of the Riemann mapping theorem in Section XV5, based on the solvability of the Dirichlet problem: Exercises for X16 1. Find explicitly the functions f(¢) and g(¢) used in the proof of the Iemma. Show by computing the derivative that |(f © A g)’(0)| > 1. 2, Let y(2) be the Riemann map ofa simply connected domain D onto the open unit disk, normalized by ¢(29) = 0 and (zo) > 0. Show that if f(2) is any analytic function on D such that |f(2)| <1 for 2 € D, then |f'(20)| < ¢(z0), with equality only when f(z) is a constant multiple of e(z). Remark. This shows that (2) is the Ablfors function of D corresponding to zo, 3, Let y(2) be the Riemann map of a simply connected domain D onto the open unit disk, normalized by i¢(z9) = 0 and ip!(zo) >. Show that if f(2) is any analytic function on D such that [f(2)| < 1 for 2 € D, then Re f’(z0) < ¢'(29), with equality only when f(z) = (2). 4, Let {Dy} be an increasing sequence of simply connected domains, and let ym be the Riemann map of Dm, onto the open unit disk D, normalized so that {2m(z0) = 0 and ‘Zf,(20) > 0 for some fixed zo € Dy. Let D be the union of the Dy's. Show that if D is the entire complex plane, then the y's are eventuially defined on any disk {|2| < R} and converge there uniformly to 0. Otherwise, D 1s simply connected and the (ys are eventually defined on each compact subset of D and converge there uniformly to the Riemann map ¢ of D onto D satisfying -(z9) = 0 and ¢!(z0) > 0si XI Conformal Mapping 5. Let {Dm} be a decreasing sequence of simply connected domains, and suppose wo € Dr for all m. Let gm(2) be the conformal map of the open unit disk D onto Dr, normalized so that gya(0) = wp and G/q(0) > 0. Show that gya(2) converges normally on D to a function (2). If the distance from uo to the boundary of Dm tends to 0, then (2) is the constant function wo, and otherwise, 9(2) maps D conformally onto some simply connected domain D. Describe D in terms of the Dr's. 6. Show that the function ¢ ++ ¢? is not a strict contraction of the hy- perbolic disk, that is, show that there is no constant c< 1 such that (2.6%) < cp(¢,€) for all ¢,¢ € D. Remark. See Exercise IX.3.9. 7. Suppose f : D -+ D is an analytic function from the unit disk into itself with a fixed point at 29 € IB. Show that the stretching at z of f(z) in the hyperbolic metric isthe same as the stretching at z of f(z) in the Euclidean metric, sm PLLC) 20) _ gg (Fl2) - 21 a oes) 7 eal [f'(20)|- XII Compact Families of Meromorphic Functions In Sections 1 and 2 we treat normal families of meromorphic functions, ‘These are families that are sequentially compact when regarded as functions with values in the extended complex plane. We give two characterizations of normal families, Marty's theorem in Section 1 and the Zaleman lemma in Section 2, From the latter characterization we deduce Montel's theorem on ‘compactness of families of meromorphic functions that omit three points, ‘and we also prove the Picard theorems. Sections 3 and 4 constitute an introduction to iteration theory and Julia sets. In Section 3 we proceed far ‘enough into the theory to see how Montel’s theorem enters the picture and to indicate the fractal nature (self-similarity) of Julia sets. In Section 4 we relate the connectedness of Julia sets to the orbits of critical points In Section 5 we introduce the Mandelbrot set, which has been called the “tnost fascinating and complicated subset of the complex plane.” 1. Marty’s Theorem In this section we consider the convergence of sequences of meromorphic functions. Our aim is to give conditions guaranteeing, compactness of families of meromorphic functions. Since functions may now assume the value oc, we must modify our notion of convergence. We do this by 1e- garding meromorphic functions on a domain D as functions from D to the extended complex plane C* = CU {oo}, which we identify with a sphere via stereographic projection (Seetion 1.3). When we identify the extended complex plane with the sphere, we refer to it also as the Riemann sphere. We will use distances on the sphere to measure distances between function values. Let o(z,1) denote the distance from z to w in the spherical metric Recall from Section IX.3 that o(2,1) is the length of the arc of the great circle on the Riemann sphere joining the points corresponding to = and w: ‘The spherical metric is invariant under the transformations corresponding, 315316 XII_Compact Families of Meromorphic Funetions to rotations of the sphere. Since the inversion 2 — 1/2 corresponds to a rotation of the sphere (by 180° around the x-axis), we have quay o(z,w) = o(1/z,1/u), wet (On any fixed bounded subset of the complex plane, the spherical metric is ‘equivalent to the Euclidean metric, 1 Blt wl < lem) < Balz—wh — lehwis Ry “while on any fixed subset of the complex plane at a positive distance from the origin, the spherical distance from = to w is comparable to the Euclidean distance between the inverse points 1/z and 1/w, 2, {wl 2 © We say that a sequence { fa(2)} of meromorphic functions on « domain D converges normally to f(z) on D if the sequence converges uniformly on compact subsets of D to f(z) in the spherical metric, that is, o(fa(2) /(2)) ‘converges to zero uniformly on each compact subset of D as n+ oo. On account of (1.1), we see that { fa(2)} converges normally to f(2) ifand only if {1/fa(2)} converges normally to 1/f(2). Here we declare 1/0 = oo and 1/00 =0, and we further allow the possibility of the constant function that is identically equal to oe. Since the spherical and Euclidean metrics are equivalent on bounded subsets of the complex plane, the definition of normal convergence given above is consistent with our earlier definition of a normally convergent sequence of analytic functions. However, now we allow the possibility of the limit being os Theorem. If the sequence (Jn(=)} of meromonphic functions on a do- main D converges normally to (2), then either f(s) is meromorphic on D or f(2) = 00. Ifa sequence {fa(2)} of analytic functions on D converges normally to f(z), then either f(z) is analytic on D or f(z) = oo. For the first statement, note that every point 2 for which f(z) # 00 has a neighborhood on which the fn's are eventually uniformly bounded and converge uniformly in the Euclidean metric. Thus f(2) is analytic on the set. where |f(=)| < co. By the same token, since 1/ f,(2) converges normally to 1/f(2), also 1/f(z) is analytic wherever f(z) # 0. Either 1/f(2) is identically zero, in which case f(z) = 00; or else the zeros of 1/4(2) are isolated in D and they are poles of f(z), in which case f(2) is meromorphic on D. For the second statement, suppose that the fn's are analytic with limit F(z) satisfying f(29) = 90 at some point. Then 1/f,(2) is a meromor- phic function with no zeros, and 1/fn(z0) + 0. Hurwitz’s theorem (Sec- 1. Marty's Theorem ait tion VIII.3) shows that 1/ f(z) converges uniformly to zero on some disk centered at zo, and consediuently,1/f(2) = 0 on the disk. By the unique. ness principle, 1/f(2) = 0 on D, and f(z) ~ oo on D. We define a family ¥ of meromorphic functions on D to be « normal family if every sequence In F has « subsequence that converges normally on D. In view of the preceding theorem, a fanily F of analytic functions on D is a normal family if and only if every sequence in F has either fa subsequence that converges uniformly on compact subsets of D t0 an analytic function ot 2 subsequence that converges uniformly on compact. subsets of D to oo. The thesis version of Montel's theorem (Section X15) can be formulated in terms of normal families. It asserts that any family of analytic functions that is uniformly bounded on each compact set is a normal fay The Arzela-Ascoli theorem (Section XI.5) provides a characterization of | normal families of meromorphie functions. "A family F of meromorphic functions is a normal femly if and only ifthe family ts equicontinuous, reqarded as functions from D, with the Euclidean metric, to the extended complex plane C>, with the spherical metric. In order to exploit this con- dition, we study the spherical metric in more detail Recall (Section IX.3) that the spherical metric is 2)d2i/(1 + |2/2). ‘The spherical length of a curve 9(0),a 1/z, by (1.1) the spherical derivative is also invariant under the inversion, (1.2) (USP = f318 XIT_ Compact Families of Meromorphie Functions This identity can also be checked by direct calculation (Exercise 1). In dealing with the spherical metric on the range of f(z), (1.2) allows us to replace f(z) near its poles by g(z) = 1/f(z), thereby reducing to the situation where f(z) is analytic. This observation is the key to the following lemma and to Marty’s theorem. Lemma. If fu(2) ~ f(2) normally on D, then f£(2) — f*(2) uniformly on compact subsets of D. Indeed, if f is analytic at 29, then fff" uniformly in some neighbor- hood of zo, 0 ff — ff uniformly in some neighborhood of 20. If f is not analytic at zo, then 1/f is analytic at zo and 1/ fe ~» 1/f normally. Again FE = (1/fa)? converges uniformly to f# = (1/f)* in Some neighborhood of 20, ‘Theorem (Marty's Theorem). A family F of meromorphic functions is normal on a domain D if and only if the spherical derivatives {*(2) £ © F} are uniformly bounded on each compact subset of D. Suppose the spherical derivatives are uniformly bounded near z) € D, say f*(z) < C for |2~ zo] 1, and let { f,(2)} be a sequence of meromorphic functions ‘on adomain D. Show that f,(2) converges normally to f(z) if and only if f,(z) converges uniformly to f(z) on compact subsets of the. ‘open set {| f(=)| 1/M} 6. Let E be a compact connected subset of the complex plane that contains more than one point. Show that the family of meromorphic functions on a domain D that omits E (that is, with range in C*\E) is a normal family of meromorphic functions. 7. Let g(z) be @ nonconstant analytic function on a domain U, let V = 9(U), and let be a family of meromorphic functions on V. Show that the family ¥ is normal on V if and only if the family of compositions {fog : f € F} is normal on U. 8, Let { fn(2)} be a sequence of rational functions that converges nor- mally to f(z) on the extended complex plane C*. Show that Jn(2) has the same degree as f(2) for n large. (See Exercise VIII4.2 for the definition of the degree of a rational function.) 9. Let F be a family of meromorphic functions on a domain D. Sup- pose there is an increasing funetion y(t), ¢ > 0, such that |f"(2)| < ‘W(|f(2))) for all z € Dand f ¢ F. Show that F is normal on D. Re ‘mark. This is Royden’s theorem. For the proof, assume that ¥(¢) is smooth and increasing, and v(t) > 142. Consider the metric in which the distance from 2 to w is the infimum of J 1/2(|¢|) Jd¢| over all smooth paths from : to w. Show that this metric is equivalent to the spherical metric. 10. Show that the family of analytic functions on a domain satisfying [f'(z)| < elf) is normal.320 XII Compact Families of Meromorphie Functions 2. Theorems of Montel and Picard As an application of Marty's theorem, we give a proof of Montel’s the- orem characterizing normal families of meromorphic functions. Another proof of Montel's theorem will be outlined at the end of Chapter XVI, based on a modular function that arises as » covering map. We begin with the following lemma. Since the converse is trivial, it actually provides a characterization of normal families: Theorem (Zaleman’s Lemma). Suppose F is a family of meromorphic functions on a domain D that is not normal. Then there are points 2» € D converging to.» point of D, numbers py >0 converging to 0, and functions fu € F such that the scaled functions gn(¢) = fu(zn + pnG) converge nor- tally: co a nonconstant meromorphic function 9(¢) on satisfying g(0) aad g'(¢) <1 for ¢€ C. The proof is tricky. By Marty's theorem, there are sequences {tw} in a ‘compact subset of D and f, € F such that fE(w_) + +00. We can assume that up -* 0. D, and we assume for convenience that the closed unit disk {lz| <1} is contained in D. Define Rn = ma f8(2)(0 l21) Since wy —+0 and f£(wq) — 00, we have Ry — 20. Suppose f£(2)(1 — el) fattains its maximum at the point z,, Rn = Fal2n)(1—lenl) Since fi(2n) > Rn, also f£(2_) 00. Define Pn = Ufkl2n) then py — 0. The disk centered at 2, of radius 1 —|2n| = pm Rn is contained in the unit disk and hence in D. We paramretrize it with a parameter ¢, by Cm ent pags [Gl < Rn Define the scaled function ga(¢) by nl6) = fnlznt pnt), — I6l < Rn. Since Ry —> 00, these functions are eventually defined en any compact set. From the chain rule (f 0 A)#(C) = f*(A(2))|h"(C)| (Exereise 1.2) with 1(C) = 2n + nG, we obtain GU) = Prfilent ont), Mel < Ra 2, Theorems of Montel and Picard 321 Now fix R > 0. Ifn is so large that Ry > R, then gn(C) is defined on the disk {|¢] < R}, and since ff(2n + Pn6)(1 ~ zn + PnG|) < Ry we have Pay T= Teal (0) SP [en + asl aR hohe PnRn Px 1—RIRy By Marty's theorem, the gy's for n large form a normal family on the disk {|2| < 2}. Passing to a subsequence, we can then assume that {gq} converges normally on € to a meromorphic function 9(¢). Since for each fixed R, 1/(1~ R/Rp) — 1, the estimate on g3(C) shows that g(C) <1 for all € € C. Since gi(0) = pa f(2n) = 1, g(0) =1. Kcr Suppose that J(2) is meromorphic for 0-< |2~ zo] 0 such that f(z) 4 ty for 0 < |z—za| <6. Thus wy is not an omitted value of f(z) at zo if and only Ifthere isa sequence, 25 —+ 29, 5 # 20, sch that f(3q) — wo. An omitted value of f(z) at 00 is defined similarly. Example. The function e!/* omits the two values 0 and co at zo = 0. The function e* omits the two values 0 and oo at oo. ‘Theorem (Montel’s Theorem). A family F of meromorphic functions on a domain D that omits three values is normal. Since normality is « local property, we can assume that the domain D is the open unit disk {/=| <1}. By composing the functions in F with a fractional linear transformation, we can assume that the omitted values are 0, 1, and oo. Since the functions in F are then analytic and nonzero on the unit disk, they have analytic roots of all orders. Let F, be the family of all 2*th roots of functions in F. Evidently Fy is normal if and only if F is normal. The functions in F, omit the values 0, 00, and all 2*th roots of unity. We will argue by contradiction. Suppose Fis not normal. Then Fi is not normal. Let Gi(C) be the entire function from the Zalcman lemma, so that GEC) 1, Gi(0) = 1, and Ge isa limit of restrictions of functions in Fe.322 XII Compact Families of Meromorphic Functions appropriately scaled. Since the functions in F, omit the 2*th roots of unity, s0 do their scaled restrictions, and from Hurwita's theorem (Section VIIL.3), so does any nonconstant normal limit. ‘Thus Gy omits the 2th roots of unity. Now, Marty's theorem shows that {Gx} is a normal family. Let G be any normal limit of a subsequence of the G's. Then GF(0) = 1, so that G is nonconstant, and by Hurwitz’s theorem again, G omits all 2¥th roots ‘of unity, this for all k > 1. Since G is an open mapping, G omits the unit circle. ‘Thus either |G| <1 or |G] > 1. Applying Liowville’s theorem to G in the first case and to 1/G in the second, we conclude that G is constant. This contradiction establishes the theorem. The following theorem is a substantial generalization of the Casorati- Weierstrass theorem. ‘Theorem (Picard’s Big Theorem). Suppose f(z) is meromorphic on punctured neighborhood {0 < |z ~ 2| < 6} of zo. If f(z) omits three values at 2, then f(z) extends to be meromorphic at 29. For the proof, we may assume that 25 =0 and that f(z) omits the val- ues 0 and oc on the punctured disk. Then f(z) is analytic on the punctured disk. Let {en} be a sequence that decreases to 0, and define Gn(2z) = f(Enz), O<|2| <6. Then {gn} omits three values, including 0 and oo. By Montel’s theorem, {gn} is a normal family. Passing to a subsequence, we can assume that n{2) converges normally to g(2) for 0 <|z| <6. Assume first that g(z) is not identically oo. Then g(z) is analytic for 0 < |z| <6. Fix p,0

0 and 0 < 9 <1, there are constants C(a, 9) > 0 with the following property. If ¥ is a family of analytic functions on the open unit disk that omits the values 0 and 1, and if |f(0)| 1. Show that {f,(2)} is a normal family on the punctured disk if and only if the singularity of f(z) at 0 is removable or a pole, 5. Let Eo,£1,£a be three disjoint compact subsets of the Riemann sphere, and let F be a family of meromorphic functions on a do- main D such that each f € F omits at least one point of each of the three sets. Show that F is a normal family. 6. Let G be the family of univalent analytic functions on a fixed domain D. (a) Show that ¢ is not normal. (b) Show that the family of functions in G that omit 0 is normal. (c) Show that the family of 0 the family of functions f € G satisfying |f"(20)| 0 such that the image of the open unit disk D under any f € S includes the open disk centered at 0 of radius r. (b) Show that the function f(2) = 2/(1 ~ 2)? belongs to S and maps D onto the complex plane slit along the negative real axis from —4 to —o. Conclude that x <4. Remark. The theorem in (2) was ist proved by P. Koebe, L. Bieberbach showed that the estimate holds with «= }, and this estimate is sharp. The theorem is known as Koebe's one-quarter theorem, and the function in (}) is referred to as the Bieberbach function.324 XII_ Compact Families of Meromorphic Functions 8. Show that there is a constant # > 0 with the following property. If f(z) is an analytic function on the open unit disk D such that £(0) = 0 and f'(0) = 1, there is a subdisk D CD such that f(2) is one-to-one on D and f(D) contains a disk of radius 3. Remark. This is Bloch’s theorem, and the optimal (largest) constant 3 is Bloch’s constant. 9. Give a proof of Royden’s theorem (Exercise 1.9) based on the Zale- ‘man lemma. 10. A family F of meromorphic functions on a domain D is normal at 29 € Dif F is normal on some open disk centered at 9. Show that F is normal at zo if and only if whenever {2,} is a sequence in D that converges to 29 and pn — 0, then every sequence in F has a subsequence {f,(2)} for which the corresponding scaled functions Gal) = fal2n+n6) converge normally to a constant (possibly 20) 1. Let F be a family of meromorphic functions on a domain D that is not normal at zo € D, and suppose that fy € F, zn — 20 and pn — 0 are such that the scaled functions gq(¢) ~ fa(zn + Px6) converge normally to a nonconstant meromorphic function 9(¢). Let G@ €C, wo = 9(¢0). (a) Show that there is a sequence & —* 0 such that Jn(Gn) = wo. (b) Show that if G is not a eritical point of (6); then f4(G) —+ 00. (¢) Show that if (2) is @ meromorphic function defined near wo such that Y(wo) ~ 20, then there is a Sequence Mo — 29 such that mq is a fixed point of Yo fn, that is, Walt) = Mn- (8) Show that if Gy is not a eritical point of g(¢), and if wy is not a critical point of y»(w), then (vo fy)*(tm) + 20. 3. Julia Sets One of the early applications of Montal’s theorem was to complex dynam ies, the study of the behavior of the iterates of an analytic or meromorphie function. Julia and Fatou used Montel’s theorem as a key tool for studying the iterates of a rational function. One of their main ideas was to un derstand the dynamical behavior of the iterates by splitting the extended complex plane C* into two invariant subsets, on one of which (the Fatow set) the iterates are well behaved, and on the other of which (the Julia set) their behavior is chaotic. We will derive some basie facts about Fatou and Julia sets in this section and the next. Let U be a domain in the extended complex plane C*, and let f(z) be an analytic map from U to U. In other words, f(z) is an analytic function on U (meromorphic if 20 € U) whose range is contained in U. For the remainder of this chapter it will be convenient to denote the nth iterate 3. Julia Sets 325 {UFC (F(z) +++) (re times) of f(2) by f(z). Danger! This should not be confused with the nth power f(2)" of f(2). Example. If f(z) =2+1, then f*(z) =2+n. Example. If f(z) = Az, then /*(2) = A"z rm Pxample. If f(2) = 24, then f*(2 Example. If /(2) is a polynomial of degree d, then f*(2) is a polynomial of degree a. In general it is not possible to express the iterates of f(z) in a simple form. An important method for obtaining effective information on the behavior of iterates is to make a “change of variable” in order to express the function in a more tractable form, such as one of the forms above, with respect to the new variable. Such a change of variable is called an analytic conjugation. It is defined formally as follows. Let f(z) bean analytic map from U into U, and g(¢) an analytic map of V into V. We say that f(z) and 9(¢) are conjugate if there is a conformal map ¢ = (2) of U onto V such that (02) = (ele), €U. Thus yo f = 90. We express this by saying that the following diagram ‘commutes: L uu vty We can regard f(2) and (¢) as the same analytic function, expressed in different coordinate systems. If y conjugates f(z) to g(¢), then ¢ also conjugates each iterate of f(z) to the corresponding iterate of 9(6), PUM) = ool), FEU 21 Further, the inverse of p(z) conjugates g(¢) to f(z), y!og = foe. Example. Every polynomial P(z) = Az4+--» of degree d > 2 is conjugate to a monic polynomial. Indeed, if we take ¢ = y(2) = ex and Q=yoPo yo}, we obtain QU) = ¢(P(2)) = eP(2) = cds! + e'44¢! + lower order. We take ¢ such that c!-#4 = 1, and we have conjugated P(z) to a monic polynomial Q(¢) of degree d.326 XIL_ Compact Families of Meromorphic Funetions Example. Consider P(z) = 2% — 2. Let 2 = (6) = ¢ + 1/6 be the ‘conformal map from {|¢| > 1} to the slit >-plane C\/~2,2]. We compute that Pony = (6+ y - Thus ¢ =U"? conjugates P(z) on C\[-2,2} to the map ¢ + exterior of the closed unit disk. From the conjugation identity ¢(J(=}) = (¢(2)) we see that f(20) = =o if and only if 9(¢(20)) = ¢(¢0), that is 2 is a fixed point of f(z) ifand only if (29) is a fixed point of o(¢). Differentiation of the conjugation identity yields '(f(2})'(2) = o/(el2)}¢"2), from which it follows that f'(z0) = 0 if and only if g{(g(zo)) = 0. We say that fixed points and eritical points are “conjugation invariants. 2 on the Example. Any fractional liner transformation f(2) with exactly two fixed points ean be conjugated to the multiplication g(¢) ~¢ fr some complex mumber 0-40. Indeed, bt 2(2\ be a fractional linear transformation that tape the Bxed points 2p and 2 of f(z) to 0 and ce, respectively. ‘Then C= ple) conjugates to the frecional linear transformation g = © Jog). Since g(¢) now has fixed points at 0 and oo, g(¢) must have ihe form AC. A aiilar argument shows that if f(2) isa fractional linear transiormaton with only one fixed point then (2) can be conjugated to the translation g(€) = ¢- 1. In tis ease we take (2) tobe the fractional Tear traneormation that maps the fed point of f(2) to 0, some other specified point 29 to 0, and f(z) to 1. (Soe Exercise 117.12) ‘Now let f(z) be a rational function, regarded as an analytic map from C* oC The Fatou set of f(z), denoted by F = F(f), consists of ll points 4% € C that have at open neighborhood WV such that the restrictions of the iterates of f(2) to WV form a normal family of analytic functions on W. Tn this case, the Fatou get off contains along with 7 all points inthe open set W, eo the Fatou set is open. The Julia set of f(s), denoted by J = J(f), is the complement of the Fatou set, J = C*\F. Thus J is a dlosed subset of C* Example. For P(2) = =? we have P*(z) = 22". ‘Thus the iterates {P"(=)} converge normally to oo on {\2| > 1) and they converge normally to 0 on {|2| < 1}. If W is any disk containing a point 29 with |zo| = 1, then the iterates P"(z) converge to both 0 and oo on nonempty open subsets of W, so the iterates do not form a normal family on W. Thus the Julia set of P(=) coincides with the unit circle, and the Fatou set of P(2) is the complement of the unit circle. ‘Theorem. The Fatou set and the Julia set of a rational function f(z) are invariant, that is, f(F) CF and f(J) < J. 3. Julia Sets 427 Since F and J are complementary sets, the theorem amounts to the assertion that f(F) CF and f-'(F) F, that is, that F is completely invariant. Thus we must show that 29 © F if and only if f(zo) € F. One direction is trivial. That f(co) € F implies 2» € F follows from the ob- servation that if a subsequence of iterates {f*~1(2)} converges normally on a disk V containing f(zo), then the compositions {f"*(z)} converge normally on the open set J~"(V) containing. 2. The less obvious diree- tion, that 29 © F implies f(z9) © F, follows from the observation that if a subsequence of iterates {f"*"(2)} converges normally on a disk U con- taining zo, then {f"*(z)} converges normally on f(U), which is an open set containing f(zo). (See Exercise 1.7.) For simplicity we focus now on a monie polynomial P() of degree d > 2. ‘The main feature of P(2) that we will exploit is its attracting fixed point at 2 = 00, Since P(z) ~ 24 as z — 00, there is R > 0 such that |P(z)) > 2[z for |z| > R. The iterates P%(z) then converge uniformly to co on the exterior domain Up = {|z| > R}. In particular, Up C F. The basin of attraction of 0, denoted by A(s0), is defined to be the set of all z such that P%(2) —+ 00. This occurs if and only if P®() © Up for large n. We denote by P-"(Up) the set of = such that P"(z) < Ur. This is the inverse image of Up under P"(=), so itis open. Since A(co) is the union of the open sets P-"(U'g), A(c0) is also open. Since P&(z) tends to 00 as k ~» 00 uniformly on P-"(Ug), each P-"(U) is contained in F. and Alco) ¢ F. Theorem. Let P(z) be a polynomial of degree d > 2. The basin of attrac- tion A(s0) of ce is an open connected subset of C* containing 00. The Julia set J of P(z) coincides with the boundary of A(oo), and it is a nonempty compact subset of C, Each bounded component of C\J is simply con- nected, ‘The proof is straightforward. First note that A(oo) is completely in- variant, that is, P(A(o0)) © A(ce) and P-1(A(oo)) < A(oo). ‘Thus P(OA(o0)) is disjoint from A(co), and P(@A(zc)) © BA(oo). ‘Then also P*(JA(co)) © Alco) for all n >'1. In particular, the iterates P"(2) are uniformly bounded on A(oc). By the maximum principle, the iterates P*(2) are uniformly bounded on the bounded components ofthe comple- ment C\9A(cc) of AA(co). Thus no bounded component of C\@A(>c) is iterated to 0, and A(co) consists of just one connected component, the unbounded component of C\PA(o0). By Montel’s theorem, the iterates P(z) form a normal family on each bounded component of the open set ©\A(co), s0 that all these components belong to the Fatou set F, and the Julia set J is contained in 2A(20). If 29 € OA(oc), ata U is any open disk containing zo, then {P"(z)) converges normally to 20 on UM A(oo). while the sequence is bounded at zo. Thus no subsequence of {P"(2)} can converge normally on U, and so 2 € J. Thus J = aA(oc). Since the328 XII Compact Families of Meromorphic Functions polynomial P(z) has degree d > 2, it has at least one fixed point zy € C. Evidently, zo ¢ A(co}, s0 A(co) # C*, and J = 8A(oo) is not empty. Finally, since (oo) ie connected, also A(co) U .4(o0) is connected, and each bounded component of the Fatou set is simply connected. (See Sec- tion VIIL8.) This completes the proof. Example. For P(z) = 23, the basin of attraction of oo is the exterior of| the unit circle, A(co) = {|2| > 1). The other component of the Fatou set is the open unit disk, which is the basin of attraction of the fixed point 0. Example. Let P(z) = 2? — 2. We have seen that P"(z) — oo for all 2€C\[-2,2), On the other hand, if -2 <2 <2, then ~2 < P(z) <2, so the interval {-2,2] is invariant under P(2) and disjoint from A(oo). Thus Aloo) = C*\[—2,2], and the Julia set is J = [-2,2]. In this case, the Fatou set coincides with A(co) ‘These examples of “smooth” Julia sets are the exception rather than the rule. It turns out that the only values of c for which the Julia set of z? + lies on a smooth curve are the values c= 0 and c = —2 treated above. Several filled-in Julia sets for other values of the parameter ¢ are depicted below. It can be shown that the Julia set of 2?—0.6 (next page) is a simple closed Jordan curve that is nowhere differentiable. This is true of all values of c in the “principal cardioid of the Mandelbrot set,” except c= 0. We ‘will show in the next section that the Julia set of 2? + 0.251 (see below) is, totally disconnected, that is, it contains no continuum. ‘ 2 Cauliflower set (c=) ‘Toray disconnected (62 0251) ‘The theorem above forms the basis for a method for producing computer images of Julia sets, called the boundary scanning method. Fix R large, fix N large, and fix a grid of points in the z-plane. For each point z in the arid, we compute the iterates P*(=) until we reach an integer n = n(z) such that |P*(z)| > R, or until we reach k = IV, in which case we set n(z) = N. 3. Julia Sets ea Quascrte Douay’ abit e=-0,12240.745i) If we color blue those points 2 for which n(2) 2. If belongs to the Julia set J of P(3) and if U is any open neighborhood of 2, then there exists V5 1 such that (a) J CUUPW)U--UP™U) Further, the inverse iterates Uje in J. P-¥(21) of any point 21 € J are dense ‘The proof of this theorem depends in an essential way on the general version of Montel’s theorem. By our hypothesis, the restrictions of the sequence of iterates {P"(:)} to U do not form a normal family. By Montel's theorem, there is at most one value that is omitted by the sequence {P"(z)} on U. If {P"(2)} does not omit any value, then (a2) Urwos Since J is compact, and each set P*(U) is open, a finite number of these open sets already cover J, and we obtain (3.1). ‘Suppose, on the other hand, there is an omitted value, call it up, so that UP"(U) = C\{wo}- Then there is no 29 # wy such that (29) = wo, oF else we would have 29 € P™(U) for some m, and wy € P"™1(U). Since the only solution of P(z) = wo is wo, Pz) — uy is « constant multiple of330 XII Compact Families of Meromorphic Functions (z—up)4, and since P(2) is assumed to be monic, P(z) = wo + (z ~ to) Now, the iterates P"(z) tend uniformly to wy on any compact subset of {lz —wol < 1}. This can be seen directly, or by making the change of variable ¢ = =~ wp, which conjugates P(2) to ¢#. In any event, wp belongs to the Fatou set of P(2), and we still obtain (8.2). Since J is compact, we then obtain (3.1) again. ‘This proves the first statement of the theorem. For the second statement, let U be an open nonempty subset of J. Then 21 € P¥(U) for some k > 1, and consequently, one of the points in P-*(21) belongs to U. This completes the proof. This theorem provides another method for producing computer images ‘of Julia sets, called the inverse iteration method. One selects a starting point 21 € J, and one calculates the inverse iterates P~¥(21) by solving the polynomial equations P¥(2) = 2; for 1 < k < NV. For this method to be effective, the degree of P(2) should be small (as d = 2). For a starting point in J, one can look among the fixed points of f(z). A rational function of degree d > 2 always has a fixed point in the Julia set (see Exercises 16 and 23). ‘The preceding theorem provides the basis for referring to Julia sets as “fractal sets,” in the sense that shapes in the Julia set reappear infinitely often at different scales near any point of the Julia set. The idea is as follows. Let 20 € J be such that the forward orbit of zo does not contain any of the critical points of P(z), and let z, be any point of J. By the theorem, we can find images P*(zo) arbitrarily close to 21. If U is a small enough disk centered at 29, then P"(2) maps U conformally onto an open set containing P"(zo). Since J is completely invariant, P*(z) maps J NU onto J P"(U). Hence J P™(U) has the same “shape” as J OU though at a different scale Exercises for XIL3 1, Show that the Julia set J and the Fatou set F of a rational function f(z) satisfy f(J) = J and f(F) =F. 2. Show that the rational function f(z) = 2?/(z? +1) is conjugate to the quadratic polynomial P(z) = 2?-+2. 3. For A,u # 0, the two maps f(z) conjugate if and onky if \ = 1 or ‘maps fixed points to fixed points. 4, Let 2 be a fixed point of f(2). Define the multiplier of the fixed point 20 to be = f'(zo). Show that the multiplier at a fixed point is a conjugation invariant, that is, if ¢ = gz) conjugates f(z) to ‘a(¢), then the multiplier of 9(¢) at the fixed point (20) is equal to ‘the multiplier of f(2) at 20. and 9(z) = uz of C* are L/y. Hint. A conjugation Exercises 331 5. A fixed point 29 of f(z) is a repelling fixed point if |f"(2o)| > 1 Show that the Julia set of a rational function f(z) contains all its repelling fixed points. 6. Show that the Julia set ofa fractional linear transformation is either ‘empty or consists of one fixed point. 7. A fixed point zo of f(2) is an attracting fixed point if|f"(zo)) <1 The basin of attraction of zo, denoted by A(z), is the set of = ‘whose iterates f"(2) converge to zo as n — oo. Show that if f(2) is 4 rational function, then A(2o) is an open subset of C* containing 29 whose boundary coincides with the Julia set. 8. Show that if f(2) is a rational funetion of degree d, then the mth iterate f(z) is a rational function of degree @” 9. Show that the Julia set of a rational function f(z) coincides with the Julia set of its mth iterate (2) 10. A point 29 is a periodic point of f(2)ift is a fixed point of f(z) for some m > 1. For such a zo, set 21 = f(20), 22 = f(z), oy = F(2m-2). Show that each z; is a fixed point of f™(z) with the same multiplier A= f’(20)--J'(zm-1). Remark. Assuming that the 2's are distinct, we define the multiplier of the cycle {20 21,.-.,2m-1} to be the multipher of f(z) at any of the points of the eycle. "Phe eycle is an attracting cycle if its multiplier 2 satisfies |A| < 1, and it isa repelling cycle if|A| > 1. The integer m is the period of the periodic point, or the length of the cycle. 11. Find all repelling cycles of the polynomial (2) 12, Show that all repelling cycles of a rational function are contained in its Julia set. 13. Find all attracting cycles of length two of the quadratic polynomial 2? +e. Show that the values of the complex parameter ¢ for which there is an attracting eycle of length two form an open disk. 14, Let (2) be a rational function. We define the basin of attraction of an attracting cycleof f(2) to consist of the points 2 € C* whose iterates {"(z) accumulate on the cycle as n —> oo. Show that the basin of attraction of an attracting cycle is an open subset of C* whose boundary coincides with the Julia set. Show that different points of an attracting cycle lie in different components of the basin, of attraction, 15, We define the multiplicity of a fixed point zy of f(z) to be the order of the zero of f(=) —= at = = 2. Show that a fixed point hasa 16, V7. 18. 19, 20, a, 23. 24 XIL_ Compact Families of Meromorphic Funetions multiplicity m > 2 if and only if its multiplier is 1. Show that the multiplicity of a fixed point is a conjugation invariant. Show that a fixed point of a rational function f(z) of multiplicity ‘m > 2 belongs to the Julia set. Show that a rational function of degree d has d +1 fixed points, counting multiplicity. Define the analytic index of a fixed point zo # co of f(z) to be the residue of 1/(2 ~ f(2)) at 20. (a) Show that if zo is a fixed point of f(z) with multiplier X £1, then the analytic index of f(z) at z9 is 1/(1—A). (b) Show that if f(2) has a fixed point of multiplicity m at 2p, then for any sinall = > 0, fo(2) = f(z) —¢ has m fixed points near zo, each of multiplicity one, for which the sum of the anal indices of f(z) tends to the analytic index of f(z) at 29 as ¢ — 0. (c) Show that the analytic index of a fixed point is a conjugation invariant, Find the fixed points and their analytic indices for the rational fune- tion f(z) = (82?-+1)/(22+3). Determine the Julia set and the Fatou set of f(2) Suppose that f(z) = z—2"*1+422"*14.0(22"2) for some integer ‘m > 1. Show that the analytic index of f(2) at the fixed point z = 0 is A. Suppose that g(z) = = — 2"*1 4 O(2"™) for some integer m > 1 Show that g(2) can be conjugated near 0 to f(z) = 2-21 + AP") 40(22"~2), where A is the analytic index of f(z) at = = 0. Hint. Try conjugating by yz) = (1 +.a2*), If cols a fixed point for f(2), we define the analytic index at oo of f(z) to be the analytic index of 1/f(1/Q) at ¢ = 0. (Danger! Tats is not the residue of =~ f(z) at 2 = 00.) Show that the sum of the analytic indices of a rational function f(z) at its fixed points in C* ist Show that if f(2) is a rational function of degree d > 2, and if all the fixed points of f(z) have multiplicity one, then f(z) has at least ‘one repelling fixed point. Hint. Use Exercise 18a, and sum the real parts of the analytic indlees at the d +1 fixed polats. Let f(z) be a rational function of degree d > 2. (a) Show that J is nonempty. (b) Show that either F is empty or F is dense in C* (c) Show that J has no isolated points. 4. Connectednese of Julia Sets 333 25. Let f(z) be a rational function of degree d > 2, Show that the repelling periodic points of f(z) are dense in the Julia set of f(z). Remark This theorem was proved independently by Fatou and Julia. Tt can be regarded as the first substantial theorem in rational iteration theory. For the proof, fill in the details of the following argument. Using the fact that J is a compact set with no isolated points, show that the points z» € J that are not in the forward orbit Of a critical point are dense in J. For such a point zo, refer to the entire function 9(¢) from Exercise 2.11. Show that there is a point Go EC such that Co is not a critical point of g(¢) and wo = 9(Go) satisfies f"(wp) = z9 for some m. Show that the points tha from Exercise 2.11 are repelling periodic points of f(z) that converge (0 2 4. Connectedness of Julia Sets The critical points and their forward orbits play a special role in complex dynamics. In this section we prove two results on connectedness of Julia sets, in which the critical points play the erucial role. While we focus on polynomials, the main ideas of the proofs are flexible and can be adapted to other situations (see the exercises). We consider a monic polynomial P(z) = 24 +0 (24-1) of degree d > 2 Let R > 0 be large, so that the circle To = {|2| = R} is contained in the basin of attraction A(co) of oe, and |P(2)| > R for |s| > R. We may assume that T; = P~(o) is “almost” a circle of radius R'/4, which is mapped dto-one by P(z) onto To. Let Us be the “annular” domain between To and I, and define Py = P~*(To) and Us = P-#(Up). Then each Uj is an open set with boundary Tk UP'k41, and P(2) maps Uy d-to-one onto Ux (counting multiplicity at eritical points). If 2 € Aloo), then either |2| > R, or there is a first integer k > 0 such that |P*(z)| > R. In the latter case, P¥(z) belongs to Up UT, and = €UkUD g. Thus A(20) is the disjoint union of the exterior disk {|z| > R). the open sets Uz, and the curves Ty separating Us and Uy. The effect of P is to map each Ty to Ty_1 and each Uy to Up334 XIL_ Compact Families of Meromorphie Functions Since there are only finitely many critical points, we can arrange by adjusting R that no critical point of P(z) lies on a curve Ty. Then P(2) is conformal at each point of [, and maps Tx d-to-one onto Ty. The composition P*(z) is then conformal at each point of Ty and maps Td to-one onto the circle Pg. Thus each T is a finite union of disjoint simple closed analytic curves. ‘The T'x’s accumulate on @A(o0) as k + o0, and since J = 0A(o0), we have J = jin Ts = jim ty Let Wi = PrM({|2| < R}). The boundary of Wy is Tr, and We is obtained from the disk {|2| < R} by removing Up Uy U-'-UUy-1 UT Since the complement of Wx is connected, each component of Wis simply connected. In view of the Riemann mapping theorem, we can think of Wi as the union ofa finite number of “analytic disks” with analytic boundaries, one domain for each closed curve in Py. Since C\Ws C A(co), We contains all the bounded components of the Fatou set F together with the Julia set J. Note also that P-"(W,) = Wiss C We. Now the stage is set, and ‘we state and prove our two theorems ‘Theorem. Let P(2) be a polynomial of degree d > 2. The Julia set J of (2) is connected if and only if the iterates of each critical point of P(2) are bounded. Suppose first that the iterates of critical points are all bounded, so that no critical point of P(z) is in A(oo). Then at each point w = P(z) € (oo) ‘we can define d analytic branches of the inverse function P~"(w), and these branches can be continued analytically along any curve in A(cc). Similarly, there are d* analytic branches of the inverse P~*(w) of the kth iterate P*(z), and P~*(w) can be continued analytically along any path in A(co), We deform Py to P; through a family of eircular closed disjoint curves Ty in Uo, 0 < #-< 1. (In fact, since Up is conformally equivalent to an annulus, we can arrange for the [y's to be simple closed analytic curves.) Suppose I, consists of just one closed curve. Then as w traverses the closed curve Ty d* times, en analytic branch of P~*(w) traverses ‘once and returns to its initial point. We deform the curve T' through the Ty's to Ts, and we follow a branch of P-M(w) d* times around each Ty Since there are no critical poirts in A(so), the branch of P~(w) moves continuously with the curve I, and cannot return to the initial value until w completes * full loops around Py. Hence each P-*(T;) consists of a single closed curve, and in particular, Cy = P~*(P) consists of a single closed curve. By induction on k we see, then, that each I, consists of ‘one closed curve, and in particular, each [', is connected. Since a limit of compact connected sets Is connected (Exercise 7), and since J is the limit of the connected sets Dn, J is connected. 4. Connectedness of Julia Sets 935 ‘To prove the converse, suppose next that there is a critical point in A(oo). Let k be the first integer such that there is a critical point in Uc, and for simplicity assume that there is only one critical point q in Us. The argument in the preceding paragraph shows that each ['; consists of just fone curve for j < k. As before, we consider the circular curves I and their inverse images P-*(T,). Let r be the smallest parameter value such that 4¢ P-*(Ty), We can assume that I is a circle for t near r. Fix <> 0 smmall. The argument in the preceding paragraph shows that P-*(I,_.) consists of just one curve, which is the boundary of a bounded simply connected domain V containing 9. Consider the inverse image P-*(I,), which includes the critical point of q. Ifthe critical point has order m, then the part of P-*(T,) near q consists of 2m-+2 analytic arcs that terminate at q and divide a disk centered at q into 2m +2 sector-like domains, as in the figure. In the sectors we altemately have P¥(z) outside Cand inside [The curve P-K(P,_-) = AV enters and then exits every other sector, as in the figure. The curves in P-¥(T',.<) enter and then exit each ‘of the remaining sectors. We can think of V as a disk, via the Riemann mapping theorem. Line segments drawn fiom q to the nearest point in OV ineach of the m-+1 sectors where AV approaches q then divide V into m +1 ‘components, Since each of these components contains an arc of P~¥(I, 4), we see that P~*(T 42) consists of at least m +1 analytic curves. Hence Wes has at least'm +1 components. Since each of these components contains curves in [) for all j > k +1, each of these components also contains points of J. and the Julia set is not connected. Jl Coe AN ial pia of orderm =2 We say that a compact subset K' of C is totally disconnected if for every 2 € K and © > 0 there is a subset E of K such that = € E, the diameter of E is less than ¢, and E is at a positive distance from K’\E. Thus K — EU(K\B) ip a decomposition of Kino two disjoint compact subsets, ‘one of which contains zp and has small diameter. At the opposite end of, the spectrum from the preceding theorem, we now have the following. ‘Theorem. Let P(z) be a polynomial of degree d > 2. If all the critical points of P(z) are iterated to oo, then the Julia set J of P(2) is totally disconnected, and the Fatou set F of P'z) coincides with the basin of attraction A(20) of.336 XII Compact Families of Meromoxphic Functions In this case we choose N’ so large that none ofthe critical points of P(=) ae in Wry. ‘Then since each component of Wy is simply connected, all d branches of P~#(w) can be defined analytically on each component of Wy Each branch has image in Wy and maps points of J to points of J. Fix 2 € J. Let V be a component of Wy that contains P*(z9) for infinitely many k’s. For such a k, let g4(w) be the branch of P~*(w) on V. satisfying ge (P§(zo)) = 20. ‘The family {gx(w)} is uniformly bounded oon V, hence a normal family of analytic funetions on the open set V. Moreover, gu(t) € Wrrsi 9 Aloe), 80 gx(u) — forall w € Wy 0 AQ). Consequently, any limit g(12) of the g¢(t)'s as k + 00 maps Way 7 Alco) into J. Since J has no interior points, 9(w) is constant, and the constant value of g(w) must be 2. Thus there is a subsequence k, for which gx, (w) is defined and converges to zo uniformly on eompact subsets of V. If wwe shrink V slightly, we obtain a compact set subset Ey of V such that P#(z9) € Ep, Eq C A(oo), and the gx,’s converge uniformly on Eo to 2 Set Ej = gx,(Eo). Then Ey is compact, 29 = gx, (P*(z9)) € E,, and the diameter of E} tends to 0. Since Ey C A(oo), also AE; C A(o0). Thus E,0 has positive distance from J\E,, and the conditions in the definition of total disconnectedness are falflled at 2p. Since 29 € J is arbitrary, the Julia set is totally disconnected Exercises for X14 1. For each of the quadratic polynomials P(c) = 2? and P(z) = 2242, sketch roughly the curves [, for 0 < k <3. Take To to be the circle {\z| = 9}, 2. Let P(2) be a polynomial of degree d > 2. Let G(z) be the Riemann map of A(oo) onto the exterior ofthe unit circle {|¢| > 1} such that G(ce) = 00. Show that ¢ = G(2) conjugates the polynomial P(z) ‘on Aco) to the map ¢# on the exterior {|¢| > 1} of the unit circle. Hint. Apply the Schwarz reflection principle to f = Go PoG-! Exercises 337 3. Let f(z) bea rational function with an attracting fixed point at zo, Let Ube the connected component of the basin of attraction of zp (see Exercise 3.7) containing 29. Suppose there is no critical point of f(z) in U except for possibly 29. (a) Show that U is simply connected. Hint. Conjugate =) to oo by a fractional linear trans- formation and assume z9 = co. Then modify the proof in the text. (b) Show that if (2) has degree d > 2, then U is conformally equiv- alent to the open unit disk, and f(z) has a critical point at zo. (c) Show that if 29 is a critical point of f(z) of order m > 1, and if G(z) is the Riemann map of U onto the unit disk D such that G(o) then ¢ = G(z) conjugates f(z) on U to the map ¢” on D. 4. Let f(z) be a rational function of degree d > 2, and suppose {o, (zo), "-"(2o)} is an attracting eyele for f(z). Show that f(2) has a 2 has at most 2d—2 at- tracting cycles. Hint. Count critical points and apply the preceding 6. Show that @ polynomial P(z) of degre: d > 2 has at amust d—1 attracting cycles in the (finite) complex plane. 7. Let {En} be a sequence of compact subsets of C that converges to a compact set E, Thus E consists of those points : for which there are 2q € Ey satisfying 2, — z. Show that if each E, is connected, then F is connected 8, Show that a compact subset A of C is totally disconnected if and only if for each 29 € K and ¢ > 0, there is a simple closed curve 7 Jn ©\K of diameter less than € such that -y separates 29 from co, that i, 29 lies in the bounded componeat of C\. 9. Show that a compact subset K of C is totally disconnected if and only if K does not contain a continuum, Strategy. For fixed 29 € K. let E(zo) be the intersection of all closed subsets E of K such that 2 € E and K\E is closed. Show that £(z9) is either a continuum or the singleton {29}. The set E(zo) is the “connected component of zp in K. 10. Let 29 be a fixed point of a polynomial P(2). Show that either {20} is a connected component of the Julia set 7, or else the connected ‘component of zo in the filled-in Julia set K = C\A(oo) contains a critieal point of P(2).338 XII_ Compact Families of Meromorphic Functions 5. The Mandelbrot Set ‘We now turn our attention to some special polynomials, the quadratic polynomials PAy=2te 2€C, where ¢ € C is a complex parameter. Let Je denote the Julia set of P.(2). ‘The only critical point of P.(z) is the origin : = 0. We consider the iterates P.(0) = ¢, P2(0) = c8 +0, P&c) = (2 +0)? +6, .... There fare two cases that can occur, correspondiag to the two theorems in the preceding section. If the iterates P(0) are bounded, then the Julia set J. is connected. Otherwise, the iterates PE(0) tend to oo, and then J. is totally disconnected. We define the Mandelbrot set MI to be the set of parameter values ¢ such that the iterates PS(0) are bounded. Thus c € M if and only if J. is connected. Note that the Mandelbrot set is « subset of parameter space {e-space) and not of dynamic space (2-space). ‘Theorem. A complex number ¢ belongs to the Mandelbrot set M if and only if |PS(0)| < 2 for all n > 1. The Mandelbrot set is a compact subset Of the closed disk {|c| < 2}. Further, C\M is connected. Suppose that je! > 2. If|2| = |e, then [2?+<| > lel? —le| = (lel lel. By the maximum principle, applied to 2/(3? +c) on the exterior of the circle, this estimate [P.(2)] > (lel ~ 1]2] persists for all = satisfying |2| > lel By iterating, starting with P.(0) = c, we obtain successively |P#(0)] > ((e)~ 1c), (P5(0)| 2 “lel — 1)¥lel, and eventually |PE(0)| > (lel —1)**le Thus P§(0) ~ 00, s0 © ¢ M. It follows that M isa subset of the closed disk {|e < 2}. Suppose next that |c| < 2 and that © ¢ M. Let n > 1 be the first integer such that |P2(0)| > 2. ‘Then |P2+1(0)| = PE(0)? + el > |PE(O)/2 — lel > (\PE(0)| — 1)|PE(0)]. We estimate P2*2(0) in the same way, and so on, 5, ‘The Mandelbrot Set 339 and we obtain |P2**(0)| > (|P2(0)| ~ 1)*|P2(0)|, which tends to +00 a8 k ~ 00. This proves the first statement in the theorem. For fixed n, the set of c € € such that |P"(0)| <2 is a closed set. ‘Thus the description of M as the set of ¢ € C such that |P2(0)| < 2 for all n shows that JM is closed, and it is bounded; hence M is compact. Let U be any bounded domain with boundary contained in M. Then |P?(0)| < 2 for all e € AU. Since P.(0) is @ polynomial in ¢, we infer from the maximum principle that |P"(0)| < 2 for all c < U. Hence Uc M. It follows that C\M has no bounded components, and so C\M consists of a single unbounded component. The theorem is proved. ‘The constant 2 in the theorem is sharp. For c= ~2, the origin is iterated in two steps to the repelling fixed point zo = 2,0 + -2 + 42 42 — In particular, the iterates PE,(0) are bounded, so -2 € M. It is straightforward to show (Exercise 3) that M meets the real axis in the closed interval [—2, 1] The theorem suggests an algorithm for obtaining computer pictures of M. We fix a large integer N, and we lay a grid of points on the disk {lel < 2}. For each point c in the grid we compute the iterates P#(0) of 0. If we reach k < N for which |PA(0)| > 4, we color the grid point sky blue; otherwise, we stop when we reach k = and we color the grid point black. The Mandelbrot set is approximated by the black grid points. By using different shades of blue to depict different escape rates, we obtain rmore attractive pictures. For an assortment of close-up pictures of various parts of the Mandelbrot set and a discussion of algorithms for obtaining the pictures, see the award-winning book of Peitgen and Richter [PR] We describe now some results on the Mandelbrot set without giving proofs. Several of the proofs are laid out as exercises, while some are beyond our scope. We focus on the set W of those parameter values ¢ such that P.(2) has an attracting cycle (other than oo). Recall from the exercises for Sections 3 and 4 the definitions of attracting fixed points, attracting cycles, and their basins of attraction. Suppose P,(=) has an attracting cycle, other than the fixed point at oo. ‘The basin of attraction of the cycle is contained in the Fatou set F, and is disjoint from A(co). Hence J, cannot be totally disconnected, and c= M. "Thus W cM. Ie ix straightforward to check that W is en open subset of the complex plane (Exercise 8). This depends on the observation that an attracting periodic point of period n isa zero of P\(z)~2, and by Hurwitz’s theorem the zeros of this polynomial vary continuously with the parame- ter. With somewhat more effort it can be proved that WC A.M, so that W is a union of connected components of the interior of M (Exercise 10) These components are called the hyperbolic components of the interior ‘of M. Each hyperbolic component corresponds to a single attracting cycle {20(¢).21(6),--- ,n-1(6)} of length m, where each n-periodie point =,(c) is an analytie function of c in the hyperbolic component.40 XIL_ Compact Families of Meromorphic Functions It turns out (see (CG}) that if U is a hyperbolic component of the in- terior of M, and if A(c) is the multiplier of the attracting cycle of P.(=). then ¢- A(c) maps U conformally onto the open unit disk in the complex plane. In particular, each of the hyperbolic components contains exactly ‘one parameter value cy such that Pa,(2) has a cycle with multiplier 0, that is, a “superattracting cyele.” We rofer to cy as the center of the hyperbolic component. The centers of the hyperbolic components cor- responding to superattracting cycles of length n ean be calculated by solv- ing P20) = 0, which is a polynomial equation in c, and discarding the solutions of PP(0) = 0 for m 2, then |PA(0)] is strictly increasing for ken. 2. Show that if |e| <2, then the Julia set J, is a subset of the closed disk {lz| <2} 3. Show that MOR = [-2.3]. Hint If -2 Me) is the Riemarn map of the cardioid onto the open unit disk, Exercises 3a 10. Find the values c for which there are superattracting cycles of length 3 and of length 4. Locate the values of ¢ on the picture of the Mandelbrot set. Show that any quadratic polynomial P(z) = Az?+Bz+C, A 40, is conjugate to the polynomial P.(2) for a unique value of ¢ Show that any quadratic polynomial P(2) = Az? + Bz +C 1s con- Jugate to the polynomial Qx(z) = Az + 2? for some cemplex num- ber A. Show that Qa(z) is conjugate to Q,(2) if and only if X= je or k= 2-1 Show that the set of parameter values ¢ such that (2) has an attracting eycle is an open subset of the complex plane that is eon- tained in.M Use the result of Exercise 4.4 to show that if Pz (=) has sn attracting cycle of length m, then P!*(0) converges to a point zo(e) of the cycle as k— 00. Show further that P"+¥(0) converges to the point 2;(€) = Pal2o(¢)) of the eycle as k -* 90, and that the points 2,(c) depend analytically on ¢, 0< j r4, hence uniformly on K. Thus f,(2) is uniformly approximable on K by polynomials in 1/(z — ¢,). Adding these approximants, we sce that f(2) is uniformly approximable on K by rational functions with poles off K There is some flexibility with respect to the location of the poles of the approximating rational functions. Lil2) Lemma. Let K bea compact subset of the complex plane, let IT be a con- nected open subset of the extended complex plane C* disjoint from K, and let 2) € U. Every rational function with poles in U can be approximated uniformly on K by rational functions with poles at zo Here we consider a polynomial in = to be a rational function with pole at co. The lemma is established by a “translation of poles” argument. We define the set V to consist of those points ¢ € U such that 1/(= ~ ¢) is approximable uniformly on K by rational functions with pole at 29. Thus iF € V, then each power 1/(= ~ ¢)* is also uniformly approximable on K by rational funetions with pole at 2. Since 29 € V, the set V is nonempty. Since U is connected, to show that V = U, it suffices to show that V is ‘open and closed in U- The “closed” assertion is easy. If ¢ € U is limit of a sequence ; € V, then 1/(2—¢) converges uniformly on K to 1/(2~), so by the definition of V, also ¢ ¢ V. The crux of the proof then is to show that V is open, and this reduces to showing that if @ € V, and if ¢ is near Co, then 1/(z ~¢) is uniformly approximable on K by polynomials in 1/(2 ~ G@). We begin with the special case Gq = co. We must show that if near oo (that is, || s large), then 1/(z—~¢) is uniformly approximableau XIII Approximation ‘Theorems ‘on K by polynomials in 2. For this, we expand 1/(2 ~ ¢) in a geometric 1 1 1a =o 7 Ima LE If |z| < C for all z € K, and if \¢| > 2C, then the kth term of the series is dominated by 1/2*. By the Weierstrass M-test, the series converges: uniformly on K, and consequently, ¢ ¢ V whenever |¢| > 2C. If Gy is nite, the proof is essentially the same, with a change of variable z+ 1/(2~G) to place G at 20. We choose « > 0 less that the distance from G to K. so that lz Gol Se for 2 K. If |¢ ~ Go| <=/2, then |(C — Go)*/(z — Go)*| < 1/2" for = © K, so the geometric series L 1 1 Be TCG) ~ = is uniformly convergent for € K., by the Weierstrass M-test. Thus all ¢ satisfying |¢ — ¢o| < ¢/2 belong to V, and V is an open set. This proves ‘the lemma, By approximating with rational functions and then using the lemma to translate the poles, we obtain immediately the following sharper version of, Runge's theorem, GG oF & ‘Theorem. Let K be a compact subset of the complex plane, and suppose that f(2) is analytic on an open set containing K. Let $ be a subset of ©*\K such that each connected component of C*\K contains a point ofS. Then f(z) can be approximated uniformly on K by rational functions with poles in 5. In particular, if K is a compact subset of the complex plane, and if the complement of K is connected, then each function analytic in a neigh borhood of A’ can be approximated uniformly on K by polynomials in = ‘There is a considerably more difficult theorem on polynomial approxima- tion, Mergelyan’s theorem, which asserts that any function that is con- tinuous on A’ and analytic on the interior of K’ can be approximated un formly on K by polynomials in z. The analogous statement for rational approximation is false. Example. There is a compact set K with empty interior, and a continuous function f(z) on K., such that f(2) eannot be approximated uniformly on K by rational functions. We outline the construction of such an example, known as a Swiss cheese set. Let $ be a sequence that is dense in the open unit disk D. Select disks Dy = {|z—cx| m, |falz) ~ fa-(2)] < 1/2" for lel < Smt Consequently, Y> |fa(2) ~ fn-1(2)| eonverges uniformly for |2| 0, pa(z) + 0 if Rez = 0, and pa(2) + —1 if Rez <0, 3. Let {=;} bea sequence of distinct points in a domain D that accumu- lates on 8D, and let {w;} be a sequence of complex numbers. Show that there is an analytic function f(z) on D such that f(2)) = wy for all j. Remark. The sequence {2} is called an interpolating sequence for analytic functions on D. 4. Let {zi} be a sequence of distinct points in a domain D that ac- cumulates on AD. Let {mg} be a sequence of positive integers, and for each , let axo,-.- sm, be complex numbers. Show that there is an analytic function f(z) on D such that f(z) = ax, for (fi(2).f2(2),fa(2)) ‘embeds D as a closed submanifold of C°, that is. such that [fi(2)] + \fal2)|-+ [fs(2)| + +00 as 2 — OD, and for each z € D there is J such that fi(z) #0. Hint. Take faz) = 1/(2—1). For j > 1, let £, be the set of z satisfying 1 ~1/} < |2| <1~1/(j +1) and €; £8 < 2x, construct f,(2) to be approximately j on E; for j odd and construct f2(2) to be approximately j on Bj for j even. Show that for any domain D in C, there exist three analytic func tions on D for which the mapping 2 > (fi(z), fa(2), falz)) em beds D as a closed submanifold of C?. (See the preceding exercise.) |. Suppose points cj € D and radii rj > 0 are such that >rj <1 and such that the disks D; = {|z— ¢)| < 73} are disjoint subdisks of D. Set K = {|z| < 1}\UD,, which is a compact set. By considering Sie $(2))d2~¥ fi-ejy(2 ~ £12) de, show that sup Ef = 1->° for any rational function f(z) with poles off K. Let D be a bounded domain whose boundary I consists of a finite number of disjoint circles, and let # = DUT. Show that a function {f(2) on E can be approximated uniformly on £ by rational functions ‘with poles off E if and only if f(z) is continuous on E and analytic ‘on D. What functions on E can be uniformly approximated on E by polynomials in 2? Let K be a compact subset of the extended complex plane C*, and let J be a subset of C*\K. Let J be the union of K and the components of C*\K that do not contain pointe of $. (a) Show that J is compact and that 2 C AK. (b) Show that if ¢ € J\K, then the funetion f(z) = 1/(z ~ ¢) is not uniformly approximable on K by rational functions with poles in S. (c) Show that a function f(z) analytic in a neighborhood of K is uniformly approximable on K by rational functions with poles in $ if and only if f(z) extends to be analytic in a neighborhood of J. (Polynomials are regarded as rational functions with poles at oo.) Let D bea bounded domain with smooth boundary , and let K be a compact subset of D. Let f(z) be a continuous function on DUT that is analytic in D. Show that the Riemann sums 3>1f(G)/(G =348 XIIL_ Approximation Theorems 2)(Gj —G-1) for the Cauchy integral fr. f(G)/(¢ ~ 2) d& converge uniformly on K to f(z) as the lengths'of the ares of between successive points ¢; of the partition tend to zero. 2. The Mittag-Leffler Theorem Recall that if f(z) is a meromorphic function with pole at zo, and the Laurent expansion of f(z) at 20 is given by 772_», ax(2 ~ 20)*, then the principal part P(s) of f(z) at 29 is the sum of the terms with the neg- ative powers, P(z) = -¢=—y ax(z— z0)*. Thus P(2) is a polynomial in 1/(z—z0), and f(z) — P(2) is analytic at zo. The Mittag-Leffler theorem as- serts that we can prescribe the poles and principal parts of a meromorphic function. ‘Theorem (Mittag-Leffler Theorem). Let D be a domain in the com- plex plane. Let {4} be a sequence of distinct points in D with no accu- ‘mulation point in D, and let Py(2) be a polynomial in 1/(2~ 2.) Then there is a meromorphic function f(z) on D whose poles are the points =, such that f(=)— P(2) i analytic at 2 Let Kp, be the set of = € D such that |2| < m and the distance from = to AD is at least 1/m. Then Kya is a compact subset of D, Kim C Kris and each component of C*\Ky, contains a point of C*\D. Let f(z) be the sum of the functions Py(z) for which 2 € Km+41\Km. There are only finitely many such functions P,(3)- By Runge's theorem, there is «rational funetion gq(z) with poles in C*\D such that |f(2)—dm(2)] $ 1/2" for 2 € Ky. Then f(z) = S| fm(2) ~ gm(2)] converges uniformly on each compact subset of D, by the Weierstrass M-test. Since 57% _yy4,lfa(2) ~ Gm(2)] is analytic on Ky for N > n, and S2¥_[fim(2) ~ Gm(z)] has poles at the points 24 that are in Ky, with the prescribed principal parts, f(z) has the prescribed poles and principal parts in D. In the case that D is the entire complex plane C, we can take Km to be the closed disk {|z| < m}. Then fym(z) is the sum of the functions P,(z) for which m < [24] 2R, then |z | > k/2,s0 the kth summand is bounded by 2/K?. ‘The technique of subtracting off terms of a power series expansion to guarantee convergence can be used to construct doubly periodic functions with double poles. Let «1 and w2 be two complex numbers that do not lie on the same line through the origin. We sketch the construction of the Weierstrass P-function P(=) associated with w; and iw. This is a dou- bly periodic meromorphic function on the complex plane. with periods w, and wa, such that the poles of P(z) are double poles at the lattice points ras, + nig, —06 < myn < oo, and the principal part of P(z) at ma + nw» is 1/(2— muy — my)? Again our first guess for constructing such a func tion might be to sum these principal parts over all integers m and n, but the sum does not converge. If we subtract the constant term of the power seties expansion at z = 0, as in the preceding example, we are led to a series that does converge, p= 45+ (mn) i009) —— (may + rai)? ‘To prove that the series converges normally, we require an estimate for the number N(B) of lattice points muy + nw in a fixed annulus {kt < |2l < k-+1)}. We claim that N(k) < ck for some constant ¢. To see this, choose 45> 0 such that the distance between any two lattice points is at least 2. We consider the 1V(k) disks of radius 6 centered at lattice points in the annulus. Since none of the disks overlap, they cover an area of x5?.V(k) We may assume that 5 <1, and then these disks are all contained in the ealarged annulus {k ~ 1 <|2| < k +2}, which has area 6rk + 3n. From 35°N(k) < Gk + Sx we obtain the asserted estimate N(k) < ok. If now |:| < R, and if k > 2R, then each summand with pole in the kth annulus c estimated easily by C/K°, where the constant C depends on R but not con k, Since 35 N(R)C/R® converges, the Weierstrass M-test shows that the series defining P(:) converges uniformly for |2| < . Hence the series converges normally on © to a meromorphic function. Further, the series converges absolutely at each 2, so that we may arrange the terms of the series in any order. For each fixed n the series350 XIII Approximation Theorems converges to a periodic function of = of period wi. Adding the appropriate constants and summing over n, we obtain P(2 + wi) = P(z). Similarly, P(z +w,) = P(z), and hence P(2) is doubly periodic Let f(2) be a meromorphic function on the complex plane, with poles at points 2. A partial fractions decomposition for f(2) is a normally converging series expansion f(=) = > P,(2), where P;(2) isa rational fune- tion whose only (finite) pole is at 2x. Thus P,() is the sum of the principal art of f(z) at z, and a polynomial. A partial fractions decomposition of ‘J(2) 1s never unique, as we can add any polynomial to one summand and subtract it from another. Example. We show that x?/sin(x2) has the partial fractions decompo- sition we < 1 en wg - Dde Set f(z) = 1/(z— k)?. If [2] < R, and if [kl > 2R, then |z — hl > [kl/2, and 1/|z — kj? < R/k?, By the Weierstrass M-test, the series converges uniformly for |2| < R. Thus f(2) is a meromorphic function on the complex plane, The functions f(2) and x2/sin®(nz) have the same poles and prinei- pal parts, so that f(s) ~1/sin®(rz) is entre. Evidently, f(z+1) = f(2), so that (2) —x2/sin®(n2) is periodic with period 1. We may focus on the period strip consisting of z = + iy for which 0 < x <1. In this strip we have 1 | 1 < Gm] ~ FeETe S Goa [kL 2, |yl 22 By the Weierstrass M-test, the series for f(z) converges uniformly and F(z) is bounded for |y| > 1 and 0 < x <1. Since each summand tends to 0 as |y| > 00, the uniform convergence implies that f(t + iy) —+ 0 as ly| + 00, 0 < 2 <1. The identity |sinz? = |sinz/? + |sinhyl? shows that also x#/sin?(xz) is bounded for |y| > 1 and tends to 0 as |y| ~» oo. Thus f(z) ~ 72/sin®(x2) is bounded for |y| > 1, hence bounded on the vertical strip {0 < x < 1}, hence by periodicity bounded on the entire complex plane. By Liouville’s theorem itis constant. Since both (=) and 3/sin®(n2) tend to 0 as |y| ~+ 00, the constant is zero, and f(2) coincides with 12/sin?(n2) Example. From the partial fractions decomposition of 7?/sin®(r2) we ean ‘obtain a partial fractions decomposition of x cot(r2) by integrating (2.1) term by term. This is justified on account of the normal convergence of the series. Since ac 1 1 L CH (tt + FF L (aes a) a> r+} ‘we obtain from (2.1) the partial fractions decomposition, o(44+3) * weot(nz) = + If we combine the terms for -tk, the constants cancel, and we obtain eot(n2) es for XTIL2 1. Use the partial fractions decomposition of x?/sin®(z2) to establish the formula 2. Establish the partial fractions decomposition 1 =-2y 4, Hint, Use tanw 3. Show that ta line sin(x2) 4, Show that = ee (-1)"(2n — 1) ae costs) j= (nH) 5. Let {2} be a sequence of distinct points such that [zs —> 00 and Slzal-"! < 00. Show that 2” Y>1/2f"(2—24) converges normally to a meromorphic function with principal part 1/(2— ze) at ze. OF 2 =U, we replace the corresponding stmmand by 1/2.) 6. Construct a meromorphic function on the complex plane whose poles are simple poles atthe Gaussian integers m-+ni with residue 1352 XIII Approximation Theorems Construct a meromorphic function on the complex plane whose poles are double poles at the points logn, m > I, with principal parts 1/(z~logn)? + 1/(2 —logn). 8. Construct a meromorphic function on the open unit disk D whose poles are simple poles at the points (1 —~2-")e"*"/", 1 < k 1, with residue 1 9. Show that I" 1/(2%—n3) converges normaly to a meromorphic funetion. Locate the poles and find the corresponding principal parts ofthe function. Express the function in terms of trigonometric functions (specifically, the cotangent function) 10. Show thatthe lattice points muy + nix, —00 < mn < oo, can be arranged in a sequence {24} such that [zu] 2 eV 11, Let {24} be a sequence of distinct points such that [2x] > eV Show that 14 x le 2 al converges normally on C and absolutely at each © C 12, Let f(z) be a doubly periodic meromorphic function on © with periods 1; and un, and let P(z) be the Weierstrass P-function as- sociated with the periods w and w. (a) Show that if the only poles of f(s) are double poles at the lattice points muy + nan =o < myn < oo, then there are constants a and b such that J() = aP(z) +b. (b} Show that if the only poles of f(2) are tiple poles at the lattice points mux + nw, -90 < m,n 1, be disjoint closed disks in D that accumulate only on the boundary of D. Suppose Qx(2) is analytic for |: — 24] > rk. Show that there is an analytic function f(z) on D\ UZ, Ex such that for each k, J(2) ~ Qel2) extends analytically to Ex 3. Infinite Products An infinite product isan expreson ofthe form [TP 75, where the py'sare complex aur, Wey tat the infate pot converges i Fa Compl enue iene ten Gals oer team eset PLO. tthe infnte product converse, we define ta vl tobe O one 3. Infinite Products: 353 of the py’s is 0; otherwise, we define it to be ‘Thus any question we might ask about infinite products ean be translated to a question about infinite series by taking logarithms. To help clarify the definition of convergent infinite product, we make several simple observations. First, if []py converges, then at most finitely many of the p's can be 0. This is because p; — 1. Second, if [Tp; ‘converges, then TL es = tim [as = Jim, papa Pm 5 pt ‘This is because exp(Log.p;) = p;- Third, we can always factor out a finite number of terms from a convergent infinite product, Tvs = mew TD = gene Finally, if an infinite product converges, and none of the factors are 0, then the product cannot be 0. Example. Consider T(E") = +n (1-4) (+4) (1-2) Since (1 +1/(2k ~ 1))(1 ~ 1/(2k)) =1, the product of the first m terms is equal to 1 if m is even, and it is equal to the last factor 1 + 1/m if m is odd. Thus the product converges to 1. tis often convenient to write p; = 14a, and to express the product as TIC +a,). If the product converges, then a; — 0, and only finitely many Of the a,'s ate equal to —1. For most purposes we can ignore the terms for which 1+ a; = 0 and work with the “tail” [[%2y(1 + a5) of the infinite product, for which a is near 0. If0 0, then St, converges if and only if > Log(1 +t) converges. This leads immediately to the following test for convergence of infinite products. ‘Theorem. Ift, > 0, then [](1+t,) converges ifand only if St; converges.354 XIII Approximation ‘Theorems Example. The infinite product fire) - e-n(-8) 6-8) et converges for a > 1 and diverges for a <1 ‘The infinite product [](1+4)) is said to converge absolutely if a, ~> 0 and S>Log(1 +a,) converges absolutely, where we sum over the terms for whicha, # —1. If [](1 + a,) converges absolutely, then > Log(1 + a,) converges, and [](1 +a,) converges. Since Log(1 +t) is analytic at w = 0 and has power series expansion Log(1 + w) = w + O(w), we see that |Log(1 +a,)| is comparable to |a;| when a; is near 0. Consequently, >| Log(1 + a,)| converges if and only if Sla;| converges. This together with the preceding theorem yield the following ‘Theorem. The infinite product [|(1 + a3) converges absolutely if and only if Da; converges absolutely. This occurs if and only if [[(1 + lasl) converges. Example. We have seen that [] (1 + (—1)**1/k) converges. However, it does not converge absolutely, on account of the divergence of the harmonic series, > 1/k = oo. Example. Consider the infinite product [](1 + i/k). Since Log(1 + i/k) = i/k+O(1/K2), the series > Log(1 +i/k) does not converge, by comparison with the harmonic series. “Consequently, [](1 + i/k) does not converge. However, since 0 < Logi + i/k| = }log(1 + 1/K2) < 1/2, the infinite product []|1+i/k| does converge. Thus absolute convergence of an infinite product is not equivalent to the convergence of the product of the absolute values of the factors. Now we turn to infinite products of functions. The Weierstrass M-test for a sum of functions is converted easily to the following test for an infinite product of functions. Theorem. Suppose that g.(r) = 1+ hy(z), k > 1, are functions on a set E. Suppose that there are constants M, > 0 such that > My < 00, and |hy(z)| < My for x € E. Then []j", gx(z) converges to []f=., gx(z) uniformly on E as m — oo, Choose a constant C such that |log(1+ w)] < Clu) for jw| < 4, and choose V such that My < } for k > N. The condition |fa(z)| < Me < 4 implies that |Log(1 + he(z))| < CM. By the Weierstrass M-test, the “Ey Log(t + h(z)) converges uniformly on E. If we exponenti- in uniform convergence of the partial products []jy g4(z) to 3. Infinite Products 355 Tle.v 9 (2) as m — oo. Since each g(x) is bounded, when we multiply by the first NV — 1 factors we obtain uniform convergence of [Tj gx(z) to TIE; ga(2) a8 m+ 00 TFG(2) = ou(2)---dm(2) is finite product of analytic functions, then by taking logarithms and diferentiting, we obtain Ge) _ HO), dale) Ge) ~ gle) > * gle) This procedure is called logarithmic differentiation. ‘The logarithmic differentiation formula also holds for uniformly convergent infinite products of analytic functions. Tt is proved by applying the formula above to finite ubproducts and passing to the limit. (3.1) ‘Theorem. Let ge(2), k > 1, be analytic functions on a domaia D such that [T_, 94(2) converges normally on D to G(2) = [1° g¢(2). Then Gz) (2) (02) ea Eg, 2D, where the sum converges normally on D. Note that the function G'(z)/G(2) has poles at the zeros of G(2). How- ever, the hypothesis implies that g4(2) + 1 uniformly on any compact subset of D, so the summands 9{(z)/gx(z) are analytic on the compact subset for k large. Since the uniform convergence ofa series is not affected by the first terms of the series, the poles do not affect the uniform conver- ‘gence. Since normal convergence of a sequence of analytic functions implies uniform convergence of the derivatives of the functions in the sequence, we ray apply (3.1) to the partial product Gma(2) = gi(z):+-m(=) and pass to the limit, to obtain (3.2). Example. Consider the infinite product f(z) = 2], (1 - 27/#)- By ‘the Weierstrass M-test, the series S>|2/?/k? converges uniformly on any bounded set. ‘Thus the infinite product converges uniformly on any bounded set, and f(2) isan entire function. Since the zeros of f(2) are simple zeros at the integers, we suspect that f(z) is related to sin(x2). ‘To check this out, we differentiate logarithmically, to obtain f(z) 1 = 1 1 1 1 Fey 72 **La > bend [G+] ‘This expression we recognize from Section 2 as the partial fractions decom- position of wcot(n2) = meos(x2)/ sin(nz). We integrate, and we see that log (2) is logsin(2) up to adding a constant. Hence f(z) = C'sin(n2). Since f(2)/z— 1 as z 0, we have C = 1/x. Thus we obtain an infinite356 XIII Approximation ‘Theorems Exercises for product expansion for the sine function, 38) sinfns) = 2 FE (1 ( “ I( Exercises for XTIL3 1, Evaluate the following. oA ras) © H0-3) © I (5) 1 1 ne a = — Le itso and og = e441 it kiseven 2 Define ak =~ Je Gh REE Show that [](1 + a,) converges, while Soax and Sa} diverge, 3. Show that if ty > 0, then [](1 +t) 1 such that litre [Ty(1 + aj) exists and is 6. Show that [](1 + ay) converges if and only if []?_,,(1-+a)) > 1 as ‘m,n — ce. Hint. Take logarithms and invoke the Cauchy criterion for series, 7. Show that if [[(1 +x) converges, then []|1 + ax] converges. 8. Suppose ay — 0. Show that the series 37 ax converges absolutely if and only if both the series S7 Arg(1 + ay) and J°Log|1 + ax) converge absolutely. 9, Show that [] (+3) 10, Show that Tf (1-4) 11, Show that if ps(2) is a polynomial of degree & such that pe(0) = 1 and pa(=) bas no zeros in the disk {|2| < K}, then [] pa (2) converges normally 12. Establish one of the following formulae, and deduce from it the other using logarithmic differentiation: eT (+ am): wa-ibt* Lapa 13. Use the infinite product expansion for sin(xz) to show that the Wallis product, lo (2h)? yy BPA BG 2n)- (2n) Beery ~ EL Tease Nena converges to /2. Use this to show that (i tim nate (nl Va we 34, Show that t> een [I (1+) snmengs vo 15, Show that + [] 2 C2) converges toa meromorphic Fanc- tion T(z) whose poles are simple poles at 0 and the negative integers, Show that (m—1)'m* Ti) = tim ~ ©) = I. ete +m Show that F(z +1) = 2P(2). Show that P(n + 1) = nl for positive integers n. Remark, The function (2) is called the gamma fune- tion, It was fist intioduced by Euler, who defined It to be the limit above. We will give an equivalent definition in the next chapter. 16. Let a4 be sequence of complex numbers, with possible repeti- tions, such that jax| <1 and ax| + 1, and consider the infinite Blaschke product defined by Be) = [] © = TT Jax] 1 mez whore the factors cortesponding £0 ay = 0 ate x (a) Suppose that. 32(1 = lax) < a0. Let E be the set of accu mulation points-on the unt circle OD of the ay's, Show that the Infinite product converges uniformly on compact subsets358 XIN Approximation Theorems of C°\E to a meromorphic function Bz), with the following properties: [B(2)| < 1 for z € D, (B(2)| =I for 2 € 9D\E, and B(2) has zeros precisely at the points (b) Show that if $7(1 — jag)) = +00, then the partial products converge uniformly on compact subsets of B to 0. (©) Suppose that f(z) is a bounded analytic function on D that is not identically zero. Show that f(2) has a factorization f(2) = D(s)g(2), where D(s) is a (Gute oF infinive) Dlaschke product, ‘and 9(3) is a bounded analytic function on B with no zeros. In particular, the zeros ay,a2,... of f(2), repeated according to multiplicity, satisfy (1 lal) < +00. 4. The Weierstrass Product Theorem The Weierstrass product theorem is a companion theorem to the Mittag- Leffler theorem. ‘The Mittag-Leffier theorem asserts that we can prescribe ‘the poles and principal parts of a meromorphic function, The Weierstrass product theorem asserts that we can prescribe the zeros and poles of a meromorphic function together with their erders, Recall that the order of a meromorphic function f(z) at a point 2» is the order of the 2er0 if f(20) = 0, and it is minus the order of the pole if f(z) has a pole at 29. If 29 is neither a pole nor « zero of f(z), the order of f() at zo is defined to be 0. Theorem (Weierstrass Product Theorem). Let D be a domain in the complex plane. Let {=} be a sequence of distinct points of D with no accumulation point in D, and let {nx} be a sequence of integers (positive or negative). Then there is a meromorphic function f(z) on D whose only zeros and poles are at the points z., such that the order of f(2) at 2 is nx: ‘The proof runs parallel to the proof of the Mittag-Lefller theorem. Let Km be the set of 2 € D such that |2| 1, and let ng be the order (possibly 2ero) of F(2) a0 "0. The prot expansion the has the oa fo = “ff (- 2)" mmo, sshere 9¢(2) is an approximation to Log(1~2/su) for || < zy], We can take ge(2) to be a partial sum for the power series expansion of Log(1 ze) at 2=0. Since bog) = ~¢-$-F—.- Locemy, we can take au(s) = - 13s ar): ‘where my is chosen large enough to guarantee convergence of the product. Exercise. Find an entire function with simple zeros at the negative integers and no other zeros. * * Solution. In this case the product 0 Gre) does not converge. However, from the estimate lz bata) Glee wwe see that (41) Il (zew ‘converges normally on the entire complex plane to an entire function with the prescribed zeros.360 XIII Approximation Theorems Exercises for XIIL4 1. Let G2) be the entire function defined by the infinite product (4). Show that 2G(2)G(—2) = sin(x2), 2. Construct an entire function that has simple zeros at the points n?, n> 0, and no other zeros. 3, Construct an entire funetion that has simple zeros on the real axis at the points n'/*, n > 0, and no other zeros. 4. Construct an entire function that has simple zeros on the positive real exis at the points v/n, n > 1, and double zeros on the imaginary axis at the points -biy/n, n> 1, and no other zeros, Construct an entire function that has simple zeros at the Gaussian integers m +ni, —00 < m,n <0, and no other zeros. 6. Find all entire functions f(z) that satisfy the fmctional equation F(22) = (1 — 22) f(z). Express the answer in terms of an infinite product. 1. Show that, 2S aa Sty g iM fie ca FS 44S] x fe, fore. 8. Let {24} be a sequence of distinct nonzero points such that [ze] > co, Let N’ > 0, and let {m4} be a sequence of positive integers such that So my|sel"¥-! < 00. Show that i ( - EY" enfin Fa + converges normally to an entire function with zeros of order ms. fat = and no other zeros. Hint. See the estimate in the preceding exercise. 9. Show that any meromorphic function f(s) on a domain D ie the quotient f(2) = 9(2)/h(2) of two analytic functions g(z) and h(=) on D. 10. Let f(z) be @ meromorphic function on a simply connected do- main D. Show that the meromorphic functions with the same zeros tand poles of the same orders as f(z) are precisely the functions of the form f(2)e4*), where A(z) is analytic on D. 11. Givea brief solution of Exercise 1.5 on interpolating sequences based, ‘on the Mittag-Lefffer theorem and the Weierstrass product theorem. XIV Some Special Functions Our aim in this chapter is to illustrate the power of complex analysis by proving a deep theorem in number theory, the prime number theoret, which does not appear at first glance to be related to complex analysis, Along the way we introduce various functions that play an important role in complex analysis. In Section 1 we introduce the gamma function T(z), which provides a meromorphic extension of the factorial function. We de. rive the asymptotic properties of the gamma function in Section 2 by view- ing it as a Laplace transform. This yields Stirling's asymptotic formula for nl. In Section 3 we study the zeta function, which is a meromorphic function whose zeros are related to the asymptotic distribution of prime numbers. In Section 4 we study Dirichlet series associated with various number-theoretic functions, thereby giving a strong hint of the fecund re- lationship between complex analysis and number theory. The proof of the prime number theorem is given in Section 5. 1. The Gamma Function ‘The gamma function P'(=) is a meromorphic function that arises frequently in complex analysis. Tt extends the factorial function from the positive integers to the entire complex plane. It is defined in the right half-plane by (1) Te) = [Petia Rez>o. ‘The integral defining T=) is absolutely convergent, and erin < rey = [eta 250. Since the integrand depends analytically on the parameter 2, the function T(z) is analytie on the right half-plane. To derive a functional equation for T=), we integrate by parts, resn= [ewwa— [ede fe 261 at.362 XIV. Some Special Functions ‘The term t¥e~t vanishes at 0 and at oo (technically, we integrate from 0 to R and take a limit as R— co), ard the identity becomes (1.2) T(z+1) = aT), Rez >0. ‘To evaluate F(z) at the positive integers, we begin with [(1) = /g° e~tdt 1, and we use the functional equat.on to obtain successively T'(2) = 1, 1(3) = 2-1, (4) = 3-2-1, and by induction (1.3) T(nti) =n ne. We declare that 0! = (1) = 1, and then (1.3) also holds for n = ‘The functional equation (1.2) allows us to extend [(2) to the left half- plane, as follows. We apply the functional equation m times to obtain T(e-+m) =(2+m—1)---(2 + 1)20(2), which we rewrite as T(z+m), TG) = Gym (ee De ‘The right-hand side is defined and meromorphic for Rez > —m, with sim- ple poles at 2 = 0, —1,--,~m-+1. By the uniqueness principle, the mero- morphic extension is unique and it satisfies the functional equation (1.2). Passing to the limit as m —+ +90, we obtain the following. ‘Theorem. The function T(z) extends to be meromorphic on the entire complex plane, where it satisfies the functional equation T(2 +1) = 3©'(). Its poles are simple poles at z ~ 0,~1,~2, Now we will express Iz) in terms of am infinite product. Define rae) = fe 2)", Rer>0,n21. I a, Since (1 = t/ny* < et and (1 = t/n" — ef for £0, we have lin Pa) = Te (The easiest way to justify passing to the limit is to appeal to Lebesgue’s dominated convergence theorem.) The substitution s = t/n yields ; (1a) rye) =o [HOt Reed nd [e-s3h Thus n= fs a-sya 1, The Gamma Function 363 If we integrate by parts in (1.4), we obtain Tale) E f'a- ora = std(1— 5)" = 22 [lea-orta = (Za) Brewer nai We repeat this step n~1 times and use the formula for Py(2), and we obtain, nent nin! Feri esas t™—) = yeaa T(z) = We rewrite this as 1a 2 z Fay = wet +2)(14+§) (142) We compare this with the infinite product with zeros at the negative inte- gers, Jew ‘which was discussed in Section XIII4. The exponential convergence factors guarantee that the product converges normally on the entire complex plane. In order to express 1/T(z) in terms of G(z), we write ne 7 Horie". fai aw = Fi(+ where 1 1 m= 14 * t5tete ‘We have seen (Section V.1) that the sequence {7} decreases to Euler's constant 7 = 0.5772... as n + 00. Thus when we pass to the limit we obtain the following, logn. ‘Theorem. Let + denote Buler’s constant. Then as) my 7H O+ews cee Initially, the identity is established only in the right half-plane. However, both sides are meromorphic, so the identity persists for all = € C, by the uniqueness principle.364 XIV_ Some Special Functions If we take logarithms in (1.5) and differentiate, we obtain Pa) 1.) SP Ly “Tey tet ree tLe Another diferentiaton leads tothe expression a Mz) Ss 1 a) ere) * es which wl tur tin the nxt ston nus anther alae ee fo ater wse which follows rom the A eae aeied fr ane) in Section XIMLI. In verme aaa erie GC) can We exprened(Ecercn XIL31) a sin(s) ~ Meese Cooking ths with (18) be sean i *einon - _28tt2) Fares = 77 66-4) ™ Since P'(1 — 2) = —2T(—2), this an rera-3)= ahs, ree Exercises for XIV-1 1, Show that P(2) has no zeros. 2. Show that for m > 0, the residue of [(=) at = = —m is (—1)"/m! 3, Show that [(4) = y. Use this to show that 1) _ 18.5.@n=1) r(n+3) = PES na an F(2n) Tin) 4, Use the Wallis product Exercise XIIL3.13) 20 show that ony OP di VR Tey ra = 2 nor (c+) ( Hint. Show first that T(2z)/T(2)T (z+ 4) has the form Ae®* by separating the product for T(2z) into even and odd terms; then evaluate A and B. tn 20-49) nose n(n) 2. Laplace Transforms 365 7. Show that Tipe (a) I Tip+a)" Remark. This integral is called the beta function, denoted by B(p.q). To establish the identity, begin by expressing ['(p)T'(q) as a double integral and converting to polar coordinates. The r-integral yields E(p +4) and the 6-integral leads to the integral above, with t= cost. pM ytd = pa>0 2. Laplace Transforms Our aim in this section is to derive an asymptotic approximation to Laplace ‘transforms of functions that do not grow too fast. This leads to an asymp- totic approximation to F(z), thence to Stirling's formula giving an asymp- totic approximation to n! Let h(s), s > 0, be a continuous or piecewise continuous function on the positive real axis. The Laplace transform of h(s) is the function (2a) (£n)(2) = [orn ds, provided that the integral converges. We will always assume that h(s) has at most exponential growth, that is, there are constants B,C such that Ins) < Ce, O B}. The estimate [tein sani < © [eereras shows that (Ch)(z) is bounded on the half-plane {Rez > B+ <} for any £>0, and (£h)(2) + 0 as Rec + 00. Example. Suppose h(s) = s* for some integer k > 0. Since ste~** —+ 0 as 8 — 00 (poiynomials grow more slowly than exponentials), there is for any = > 0 an estimate of the form [h(s)| < C.e**. Hence the Laplace transform (Cs*)(z) converges absolutely for all z satisfying Rez > 0. We compute that. (cetya) = [ersten ger [eta 0 ‘The integral on the right is P(k-+1) = kt, so (Cs*)(=) coincides with kl/2**1 when z= x > 0. By the uniqueness principle it coincides with kt/2**! on366 XIV_ Some Special Functions the right half-plane, (22) (esha) = Ao sr Rez>0, £20 Lemma. Suppose that |h(s)| < Cs%e* for some integer N > 0 and constants B,C. Then 23) (cance +in)) = 0 (yr ). as 210. ‘This follows from the obvious estimates © ate BiegM m 1 [cirri < ¢ [ee Pads = 0s = o(=t) Theorem. Suppose that \h(s)| < Ce®* for s > 0. Fix N 1. IFA(s) has the Taylor polynomial approximation 1 (24) fs) = bo bret Bea 4 PA BL O(N) (s) = 18+ oF WH) as6 04, then i (2.5) (£h)(2) as z= Rezo. ‘To see this, we let R(s) be the remainder term in the Taylor approxima- tion, so that bw hfs) = by +s +o SY ERs), 520 (v- Since |R(s)| < Cos for s > 0 near 0, there is C, > 0 such that IR(s)| < Cys, OS 800. Since the Laplace transform operates linearly, (che) = m(EIN(2) +b (ee) H+ He ‘The result now follows from (22) and (23) We wish to use the asymptotic approximation theorem to obtain asymp- totie approximations to T(z). We begin by expressing the derivative of T'(z)/T(2) as a Laplace transform, From (Cs)(z) = 1/2” we obtain ie es\(z+k) = [fe Heeds = [re Di (£s%-*) (2) + (ERY). =(e*)fsds. 2. Laplace Transforms 367 Summing over k from 0 to N, and recognizing the geometric series, we obtain Loe [er Qeertoue- fi ‘he exponentials e~*( +2) decrease to 0, and passage to the limit in the right-hand side is justified. In view of (1.6), we obtain in the limit as Noe that a0) _ [Pps an [fe pists, Rezo. ‘The integral on the right is the Laplace transform of the function sven sds gs) = ‘This function g(s) is a meromorphic function of the complex variable s, with poles at s = 2rni for n = +1,42..... The function is analytic at 0, since 1 ~e~* has a simple zero there. The power series expansion of g(s) about s = 0 is readily calculated to be 1 2 Since g(s)/s + 1 as s+ +00, an estimate of the form 0 < g(s) < Ce’, 8 2 0, holds for some constant C > 0. The theorem then yields the asymptotic approximation am) 1,14 1 ete ~ ptastasto(s): * ‘To obtain an asymptotic approximation to T(z), we must integrate twice. Except for a constant of integration, the antiderivative for any error term O(/z"), N > 2, is dominated by [lear = [8 = wate = Hence g{s) = 1+ ss) < 2m. Rez +00, ot Log 40-5 ~ Ge +0 (4), where ais @ constant of integration. If we integrate aga, we obtain 1 Lo o(2 logt(2) = 2 Log2—2+a2~} togs+ a+b +0(5 as) = loge— tas 5 Loge+ 8+ 7. +0(%). where 4 is another constant of integration. If we substitute the asymp- totic expansions for log? (2 +1) and logT(z) into the relation logl (2 + 1) = logz + logT'(z), we obtain a = 0. To evaluate 8, we compare68 XIV Some Special Functions T(2n) with P(m), which can be done with the elementary relation in Exer- cise 1.3(¢), riot (ve) = enema If we take logarithms, substitute the approximations, and do some book- keeping, we obtain 3 = log V2r. We have proved the following, ‘Theorem. ‘The gamma function F(2) satisfies 1 (26) ogt(e) = ( :) logs 2 + on viF + I +0 (5) Rez 00, ‘The asymptotic approximation to P(=) is obtained by exponentiating, o()). = Res-+s Substituting n! = nI'(n), we obtain aaa formula, ny 1 ("40+ a +0(@)) It is natural to consider the asymptotic behavior of Laplace transforms in the right half-plane as 2 = Rez — co. However, for some classes of functions it is more appropriate to consider asymptotic behavior ina sector fs |2| oe rather than in a half-plane. Let f(2) be defined and analytic in a sector $= {0 < argz < Oj}. We say that 7%» @,/2? is an asymptotic series for f(2) in the sector S if for each N'>'0, 1-4 = °(=e) fas 2 —+ co in the sector. If .a,/2? is an asymptotic series for f(). then ae = dim 28 [4(2) ~ ag ~~. SH] Hence the coefficients in the asymptotic series of f(2) are uniquely deter- mined. We refer to 57 a,/2? as the asymptotic expansion of (2) in the sector. If f(z) is analytic a¢ co, then its Laurent series is an asymptotic series for ‘J(2) in any sector, However, asymptotic series generally do not converge. (See the Exercises.) Further, it can happen that different functions have the same asymptotic expansion. For example, since 2"¢-* —» 0 as z — 00 jn any sector {Jag z| < (7/2) ~<). the function e-* has an asymptotic im the sector that is identically zero. Exercises a ‘The asymptotic approsimation theorem shows that if h(s) has at most exponential growth, ond if f(s) is infinitely differentiable at's = 0, then the Laplace transform (£h)(2) has the asymptotic expansion J> A'8}(0)/z4+1 in any sector 5, = {larg 2| < (7/2) — =}. Im any such sector there is an estimate of the form ¢,|z| < Rez < |z|: consequently, |2| —+ 00 if and only if x = Rez ~ co. In general, the series need not be an asymptotic expansion for (Ch)(z) in the entire right half-plane, regarded as a sector. However, in the case ofthe function log (2), its. Infact it can be shown that the series (2.6) is an asymptotic series for LogI'(2) in any sector {arg 2| < xe}, for any ¢ >0. Exercises for XTV.2 1. Show that if Re A > —1, then the Laplace transform of s* converges absolutely in the right half-plane, and pees) (£3)(2) Rez >0. 2. Show that if g(s) = e**h(s), then (Cg)(z) = (£h)(z — a). 3. Suppose h(s) = 0 for s <0. Show that if@ > 0 and g(s) = h(s—a), then (£9)(2) = e~*(Ch)(2), 4. Show tht ifs) is diferentible, then the Laplace transform of ts derivative h'(s) is (Ch’)(z) = 2(£h)(z) — h(0), 5. Compure (Z)" V2 and (2)" Vor (1+) for n= 2, 8 au 4 and compare the eeu with ni 6. Define the Bernoulli numbers B,, Bz. Bs, ... by So? = Bir, Baa, Bio, 23 tg a {a) Show that B, = 1/6 and By = 1/30. (b) Show that s 1, Big Boa, Bas Lt jet Spe - Feat + Beat — (c) Show that log (2) has asymptotic series Bol | Byi Dryas eer (€) Show that as = — 00 through the sector | arg] < x/2~¢, Me, 1, 1 1e 1 VEC: i maa mes +°(cR)) How many terms can be specified here using only By and Bs? log (2) ~ (2 = 1/2)log2 ~ 2+ log VIR + BE ry370 XIV_ Some Special Functions Express the function Fe =f fas a Laplace transform, and show that It extends analytically to the right half-plane. Obtain the asymptotic expansion of the function, and show that the asymptotic expansion diverges at every point. z>0, 8. Let h(s), 8 > 0, be a continuous function such that [h(s)] < Ce®* for some constants B,C > 0. Suppose that h(s) extends to be ‘analytic ina neighborhood of the origin, and suppase that the power series expansion f(s) = S7f2q Al) (0)st/kI has only a finite radius of convergence. Show thitt the asymptotic series "jg h(®)(0)/2#*? of the Laplace transform (Ch)(=) diverges at every point =, 9. Show that if f(2) is analytic at oo, then its power series expansion in descending powers of = is an asymptotic series for f(z) in any sector. 10. Show that if f(2) and g(z) have asymptotic expansions in a sector, then so does their product f(2)g(z). 11. Show that if f(2) has an asymptotic series S72 ax/2* in a sector S, beginning with the 1/2? term, then any antiderivative F(z) of f(2) has an asymptotic series in S obtained by integrating term by term. 12, Show that if f(z) has an asymptotic series S7{2pax/2* in a see- tor 5 = {4 < arg < 6}, then f'(2) has the asymptotic series X(-Ja;)/2", obtained by differentiating term by term, in any smaller sector Se = {0 +© 1, and it converges uniformly for o> 146 for any e > 0, by comparison with the numerical series S>n~(1+) ‘The function ¢(a) is a decreasing function of @ for 1 << 90, and¢(o) —+ 1 as ¢ + +00. ‘Since the harmonic series J71/n diverges, ((a) — +00 as @ +14. Note also that |¢(o + it)| < C(o) for allt. We have already met the series for ¢(2), fi yl 7 @=Ya-F For the remainder of this chapter, we will always use p to denote a prime number, p= 2,3,5,7,11,--+. The following infinite product representation of the zeta function signals a connection between C(s) and prime numbers. Theorem. IfRes > 1, then 1 L 3. 7 -3). en a Hn ¢ 7) whore the product is taken overall prime numbers p. The series I1/p"# converges absolutely for ¢ > 1, and it converges uniformly in any half-plane {a > 1+e}. Hence the product (3.1) converges Consider the geometric series 1 =p If we mutiply together the m series corresponding to primes py... ‘we obtain = ltpttp™te-, Res>1 : « Top > elt oa Fe Since every integer n> 1 has a unique representation as a product of powers of distinct primes, a summand 1/n* appears at most once in this sum, so the sum is a subsum of the series S>n~*. As we incorporate more Primes into the product, we eventually capture all terms 1% and in the limit we have +S oe = os)372 XIV Some Special Functions Next we wish to extend ¢(s) to be a meromorphic function on the entire complex plane. To do this, we represent ¢(s) as a contour integral. We make a branch cut along the positive real axis and consider the function Ht = efelost-2), 2 C\l0,00), which depends analytically on both = and the complex parameter s. Let + be the contour indicated above, following the top edge of the branch cut from +00 to ¢, detouring along a circle of radius ¢ around z = 0, and returning from £ to +90 along the bottom edge of the branch cut. We consider the function (32) 000) = 5; f Prue ‘The integral converges, and it represents an entire function of s. To evaluate it, we first assume that Re s > 1, and we express 1 peele“tors—) ols) aloe aot 1 pe elen0 mil, = atin) as. Since e* — 1 has a simple zero at z = 0, the integrand is bounded on the circle |2| =e by Ce®**~, and the second integral is bounded by CeR**—! Under our assumption that Re > 1, this integral tends to 0 with ©. Passing to the limit. we obtain nite) [2 Vie 1)) = -2isin(ns). Hence 0) = (ere The term in parentheses is 2isin(: We treat this integral by expanding 1/(e* metric series, oa [ Fe=f (Se*)= “The partial sums of the geometric seis form an increasing sequence of functions that converges uniformly on each interval [e,00). From this it is straightforsard to justify interchanging the summation and integration, (a3) 6) e-*) ina geo- 3. The Zeta Function ats (34) becomes ir = T(s)¢(s), Res >1 Substituting this into (3.3) and using F(s)P(1—s wwe obtain +/sin(xs) from (1.7), FI) rics) = Sl) #0) < Toa We have established this identity for Res > 1. However, 6(s) is an entize funetion, so we can use this identity to extend ¢(s) to be a meromorphic funetion in the entire complex plane. The extended function then satisfies the following. Theorem. For the branch of (~ we have "1 and the contour 7 described above, Tass) f (3 (3.5) G(s) = —T(— s)o(s) ‘Sat ao Since 6() is an entire function, and the poles of [(1 —s) are simple poles at s = 1,2,3,..., (s) is analytic except for possibly a simple pole at s = 1. Now, P(s) has a simple pole at s = 0 with residue 1, so P(1—s) ~ 1/(1-8) ass 1. Since Lp a ori J, 1 a) = ra[sty a} = we see that ¢(s) ~ 1/(s— 1) as s+ 1. Thus ¢(s) has a simple pole with residue I at $= 1. We collect these observations. ‘Theorem. The zeta function C(s) is a meromorphic function on the com- lex plane with only one pole, a simple pole at $= 1 with residue 1 Next we derive a functional equation for the zeta function. ‘Theorem. The function C(s) satisfies the functional equation (6) ) Fa ~ sgt ~s). For the proof, we modify the contour ¥ used to define 6(s). Fix s to be real and negative. Let 7» be the contour indicated in the figure above, obtained from 7 by replacing the segment starting at n on the top edge of, the slit and going around 0 and back along the bottom edge of the slit ton374 XIV. Some Special Functions asin Is Pen by straight line segments running counterclockwise around the boundary of the rectangle Ry, with vertices at -in + (2n +1/2)ni. Define ante = uf! 2 ge e@-1 Since |e?—1] > } on the edges ofthe rectangle, the integrand is bounded on the edges by 2n*!, By the ML-astimate, the integrals over these edges are ‘bounded by Cn®. Since s <0, this tends to 0 as n — oo, and consequently, én(s) = 0.a8 n — oo. Now, the difference [,,— f, is the integral around the closed contour consisting of the boundary of the rectangle Ay with a detour along an elongated indentation that: does not cross over the branch ‘cut. ‘The contour encloses only the poles of the integrand at the points 2nki, 0 1}. Since 1/T(z) has a product representation expressing it as an entire function, F's) has no zeros. The functional equation then shows that the only zeres of ¢(s) in the left half plane {Res < 0} are the zeros of sin(xs/2), which are simple zeros at 2, 4, ~6, .... AS mentioned above, these zeros ate called the trivial zeros of the zeta function. ‘The nentrivial zeros of the zeta function lie in the strip {0 < Res < 1}, whieh is called the eritical strip. Many outstanding mathematicians have worked on the Riemann hypothesi that the nontrivial zeros of the zeta function lie on the line {Res = 3}. It is known that ¢(s) has infinitely many zeros in the critical strip. Their asymptote distribution is known, and it is known that at least a third of ‘them lie on the critical line {Res = 3}. Further, the first zillion or so zeros fare known to lie on the critical line. Meanwhile, whether the Riemann hypothesis is true remains a famous unresolved problem. Exercises for XIV.3 1. Show that if @ is a zero of the zeta function in the critical strip, then 0 are a, 1a, and 1a. 2. Show that the function €(s) = }5(1~s)x~*/2P(s/2)¢(s) is an entire function that is symmetric with respect to the line {Res = 3}, that is, (1 —s) = (5). Show that €(s) is real when s is real and also when Res = }. Where are the zeros of &(s)? 3. By comparing ¢(o) with [O-de, show directly from the defini tion of ¢(s) that (0 —1)¢(a) + Las. 0 + 1+. 4. Show that (1 ~2'-*)¢(s) is an entire function, which is represented. by the series 1 pty et: Rese Remark. We will see in the next section that the series converges and represents the function for Res > 0. (1-2 G(s) 5. By comparing the partial fractions decomposition Lit rears) = 14 (+2) a with the series defining the Bernoulli numbers, and differentiating, derive the power series expansion a ee “Ree 7 kI<1.