Complex Function Theory

Second Edition

Donald Sarason

®AMS
AMERICAN MATHEMATICAL SOCIETY
 2000 Mathematics Subject Classification. Primary 30-01.

Front Cover: The figure on the front cover is courtesy of Andrew D. Hwang.

The Hindustan Book Agency has the rights to distribute this book in India, Bangladesh,
Bhutan, Nepal, Pakistan, Sri Lanka, and the Maldives.

For additional information and updates on this book, visit

www .ams.org/bookpages/ mbk-49

Library of Congress Cataloging-in-Publication Data

Sarason, Donald.
Complex function theory / Donald Sarason. - 2nd ed.
p. cm.
Includes index.
ISBN-13: 978-0-8218-4428-1 (alk. paper)
ISBN-10: 0-8218-4428-8 (alk. paper)
1. Functions of complex variables. I. Title.
QA331.7.S27 2007
515'.9-dc22 2007060552

Copying and reprinting. Individual readers of this publication, and nonprofit libraries
acting for them, are permitted to make fair use of the material, such as to copy a chapter for use
in teaching or research. Permission is granted to quote brief passages from this publication in
reviews, provided the customary acknowledgment of the source is given.
Republication, systematic copying, or multiple reproduction of any material in this publication
is permitted only under license from the American Mathematical Society. Requests for such
permission should be addressed to the Acquisitions Department, American Mathematical Society,
201 Charles Street, Providence, Rhode Island 02904-2294, USA. Requests can also be made by
e-mail to reprint-permission©ams. org.

© 2007 by the American Mathematical Society. All rights reserved.

Printed in the United States of America.
§ The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http: I /www.ams.org/
10 9 8 7 6 5 4 3 2 1 12 11 10 09 08 07
Contents

Preface to the Second Edition ix

Preface to the First Edition xi

Chapter I. Complex Numbers 1

§I.I. Definition of C 2
§I.2. Field Axioms 2
§I.3. Embedding of R in C. The Imaginary Unit 3
§I.4. Geometric Representation 3
§I.5. Triangle Inequality 4
§I.6. Parallelogram Equality 5
§I.7. Plane Geometry via Complex Numbers 5
§I.8. C as a Metric Space 6
§I.9. Polar Form 6
§I.10. De Moivre's Formula 7
§I.11. Roots 8
§I.12. Stereographic Projection 9
§I.13. Spherical Metric 10
§I.14. Extended Complex Plane 11

Chapter II. Complex Differentiation 13

§II. l. Definition of the Derivative 13
§II.2. Restatement in Terms of Linear Approximation 14
§II.3. Immediate Consequences 14

-
lll
lV Contents

§II.4. Polynomials and Rational Functions 15

§II.5. Comparison Between Differentiability in the Real and
Complex Senses 15
§II.6. Cauchy-Riemann Equations 16
§II. 7. Sufficient Condition for Differentiability 17
§II.8. Holomorphic Functions 17
§II.9. Complex Partial Differential Operators 18
§II.10. Picturing a Holomorphic Function 19
§II.11. Curves in C 20
§II.12. Conformality 21
§II.13. Conformal Implies Holomorphic 22
§II.14. Harmonic Functions 23
§II.15. Holomorphic Implies Harmonic 24
§II.16. Harmonic Conjugates 24
Chapter III. Linear-Fractional Transformations 27
§III.l. Complex projective space 27
§III.2. Linear-fractional transformations 28
§III.3. Conformality 29
§III.4. Fixed points 29
§III.5. Three-fold transitivity 29
§III.6. Factorization 30
§III. 7. Clircles 31
§III.8. Preservation of clircles 31
§III.9. Analyzing a linear-fractional transformation-an example 32

Chapter IV. Elementary Functions 35

§IV.l. Definition of ez 35
§IV.2. Law of Exponents 36
§IV.3. ez is holomorphic 36
§IV.4. Periodicity 37
§IV.5. ez as a map 37
§IV.6. Hyperbolic functions 38
§IV.7. Zeros of coshz and sinhz. 38
§IV.8. Trigonometric functions 39
§IV.9. Logarithms 40
§IV.10. Branches of arg z and log z. 40
Contents v

§IV.11. log z as a holomorphic function 41

§IV.12. Logarithms of holomorphic functions 42
§IV.13. Roots 43
§IV.14. Inverses of holomorphic functions 43
§IV.15. Inverse trigonometric functions 44
§IV.16. Powers 45
§IV.17. Analytic continuation and Riemann surfaces 45
Chapter V. Power Series 49
§V.l. Infinite Series 49
§V.2. Necessary Condition for Convergence 49
§V.3. Geometric Series 50
§V.4. Triangle Inequality for Series 50
§V.5. Absolute Convergence 50
§V.6. Sequences of Functions 51
§V.7. Series of Functions 51
§V.8. Power Series 52
§V.9. Region of Convergence 53
§V.10. Radius of Convergence 54
§V.11. Limits Superior 54
§V.12. Cauchy-Hadamard Theorem 55
§V.13. Ratio Test 56
§V.14. Examples 57
§V.15. Differentiation of Power Series 58
§V.16. Examples 60
§V.17. Cauchy Product 61
§V.18. Division of Power Series 63
Chapter VI. Complex Integration 65
§VI.I. Riemann Integral for Complex-Valued Functions 65
§VI.2. Fundamental Theorem of Calculus 66
§VI.3. Triangle Inequality for Integration 66
§VI.4. Arc Length 67
§VI.5. The Complex Integral 67
§VI.6. Integral of a Derivative 68
§VI.7. An Example 68
vi Contents

§VI.8. Reparametrization 69
§VI.9. The Reverse of a Curve 70
§VI.10. Estimate of the Integral 71
§VI.11. Integral of a Limit 71
§VI.12. An Example 71
Chapter VII. Core Versions of Cauchy's Theorem, and Consequences 75
§VII.I. Cauchy's Theorem for a Triangle 75
§VII.2. Cauchy's Theorem for a Convex Region 78
§VII.3. Existence of a Primitive 78
§VII.4. More Definite Integrals 79
§VII.5. Cauchy's Formula for a Circle 79
§VII.6. Mean Value Property 81
§VII.7. Cauchy Integrals 82
§VII.8. Implications for Holomorphic Functions 83
§VII.9. Cauchy Product 84
§VII.10. Converse of Goursat's Lemma 85
§VII.11. Liouville's Theorem 86
§VII.12. Fundamental Theorem of Algebra 86
§VII.13. Zeros of Holomorphic Functions 87
§VII.14. The Identity Theorem 89
§VII.15. Weierstrass Convergence Theorem 89
§VII.16. Maximum Modulus Principle 90
§VII.17. Schwarz's Lemma 91
§VII.18. Existence of Harmonic Conjugates 93
§VII.19. Infinite Differentiability of Harmonic Functions 94
§VII.20. Mean Value Property for Harmonic Functions 94
§VII.21. Identity Theorem for Harmonic Functions 94
§VII.22. Maximum Principle for Harmonic Functions 95
§VII.23. Harmonic Functions in Higher Dimensions 95

Chapter VIII. Laurent Series and Isolated Singularities 97

§VIII.I. Simple Examples 97
§VIII.2. Laurent Series 98
§VIII.3. Cauchy Integral Near oo 99
§VIII.4. Cauchy's Theorem for Two Concentric Circles 100
Contents Vll

§VIII.5. Cauchy's Formula for an Annulus 101

§VIII.6. Existence of Laurent Series Representations 101
§VIII. 7. Isolated Singularities 102
§VIII.8. Criterion for a Removable Singularity 105
§VIII.9. Criterion for a Pole 105
§VIII.10. Casorati-Weierstrass Theorem 106
§VIII.11. Picard's Theorem 106
§VIII.12. Residues 106
Chapter IX. Cauchy's Theorem 109
§IX.1. Continuous Logarithms 109
§IX.2. Piecewise C 1 Case 110
§IX.3. Increments in the Logarithm and Argument Along a Curve 110
§IX.4. Winding Number 111
§IX.5. Case of a Piecewise-C 1 Curve 111
§IX.6. Contours 113
§IX.7. Winding Numbers of Contours 114
§IX.8. Separation Lemma 115
§IX.9. Addendum to the Separation Lemma 117
§IX.10. Cauchy's Theorem 118
§IX.11. Homotopy 119
§IX.12. Continuous Logarithms-2-D Version 119
§IX.13. Homotopy and Winding Numbers 120
§IX.14. Homotopy Version of Cauchy's Theorem 121
§IX.15. Runge's Approximation Theorem 121
§IX.16. Second Proof of Cauchy's Theorem 122
§IX.17. Sharpened Form of Runge's Theorem 123
Chapter X. Further Development of Basic Complex Function 125
§X.1. Simply Connected Domains 125
§X.2. Winding Number Criterion 126
§X.3. Cauchy's Theorem for Simply Connected Domains 126
§X.4. Existence of Primitives 127
§X.5. Existence of Logarithms 127
§X.6. Existence of Harmonic Conjugates 128
§X. 7. Simple Connectivity and Homotopy 128
viii Contents

§X.8. The Residue Theorem 129

§X.9. Cauchy's Formula 130
§X.10. More Definite Integrals 130
§X.11. The Argument Principle 137
§X.12. Rouche's Theorem 138
§X.13. The Local Mapping Theorem 140
§X.14. Consequences of the Local Mapping Theorem 140
§X.15. Inverses 141
§X.16. Conformal Equivalence 141
§X.17. The Riemann Mapping Theorem 142
§X.18. An Extremal Property of Riemann Maps 143
§X.19. Stieltjes-Osgood Theorem 144
§X.20. Proof of the Riemann Mapping Theorem 146
§X.21. Simple Connectivity Again 148
Appendix 1. Sufficient condition for differentiability 151
Appendix 2. Two instances of the chain rule 153
Appendix 3. Groups, and linear-fractional transformations 155
Appendix 4. Differentiation under the integral sign 157
References 159

Index 161
Preface to the Second
Edition

As is usual in a second edition, various minor flaws that had crept into
the first edition have been corrected. Certain topics are now presented more
clearly, it is hoped, after being rewritten and/ or reorganized. The treatment
has been expanded only slightly: there is now a section on division of power
series, and a brief discussion of homotopy. (The latter topic was relegated
to a couple of exercises in the first edition.) Four appendices have been
added; they contain needed background which, experience has shown, is not
possessed nowadays by all students taking introductory complex analysis.
In this edition, the numbers of certain exercises are preceded by an
asterisk. The asterisk indicates that the exercise will be referred to later in
the text. In many cases the result established in the exercise will be needed
as part of a proof.
I am indebted to a number of students for detecting minor errors in
the first edition, and to Robert Burckel and Bjorn Poonen for their valuable
comments. Special thanks go to George Bergman and his eagle eye. George,
while teaching from the first edition, read it carefully and provided a long
list of suggested improvements, both in exposition and in typography. I owe
my colleague Henry Helson, the publisher of the first edition, thanks for
encouraging me to publish these Notes in the first place, and for his many
kindnesses during our forty-plus years together at Berkeley.

- ix
x Preface to the Second Edition

The figures from the first edition have been redrawn by Andrew D.
Hwang, whose generous help is greatly appreciated. I am indebted to Ed-
ward Dunne and his AMS colleagues for the patient and professional way
they shepherded my manuscript into print.
Finally, as always, I am deeply grateful to my wife, Mary Jennings, for
her constant support, in particular, for applying her TEX-nical skills to this
volume.

Berkeley, California
January 26, 2007
Preface to the First
Edition 1

These are the notes for a one-semester introductory course in the theory of
functions of a complex variable. The aim of the notes is to help students
of mathematics and related sciences acquire a basic understanding of the
subject, as a preparation for pursuing it at a higher level or for employing
it in other areas. The approach is standard and somewhat old-fashioned.
The user of the notes is assumed to have a thorough grounding in basic
real analysis, as one can obtain, for example, from the book of W. Rudin
cited in the list of references. Notions like metric, open set, closed set, inte-
rior, boundary, limit point, and uniform convergence are employed without
explanation. Especially important are the notions of a connected set and
of the connected components of a set. Basic notions from abstract algebra
also occur now and then. These are all concepts that ordinarily are familiar
to students by the time they reach complex function theory.
As these notes are a rather bare-bones introduction to a vast subject, the
student or instructor who uses them may well wish to supplement them with
other references. The notes owe a great deal to the book by L. V. Ahlfors
and to the book by S. Saks and A. Zygmund, which, together with the
teaching of George Piranian, were largely responsible for my own love affair
with the subject. Several other excellent books are mentioned in the list of
references.

1 The first edition was published by Henry Helson under the title Notes on Complex Function
Theory.

- xi
xii Preface to the First Edition

The notes contain only a handful of pictures, not enough to do justice

to the strong geometric component of complex function theory. The user is
advised to make his or her own sketches as an aid to visualization. Thanks
go to Andrew Hwang for drawing the pictures.
The approach in these notes to Cauchy's theorem, the central theo-
rem of the subject, follows the one used by Ahlfors, attributed by him to
A. F. Beardon. An alternative approach based on Runge's approximation
theorem, adapted from Saks and Zygmund, is also presented.
The terminology used in the notes is for the most part standard. Two
exceptions need mention. Some authors use the term "region" specifically to
refer to an open connected subset of the plane. Here the term is used, from
time to time, in a less formal way. On the other hand, the term "contour"
is used in the notes in a specific way not employed by other authors.
I wish to thank my Berkeley Math H185 class in the Spring Semester,
1994, for pointing out a number of corrections to the prepublication version
of the notes, and my wife, Mary Jennings, who read the first draft of the
notes and helped me to anticipate some of the questions students might raise
as they work through this material. She also typed the manuscript. I deeply
appreciate her assistance and support. The notes are dedicated to her.

Berkeley, California
June 8, 1994
 Chapter I

Complex Numbers

The complex number system, C, possesses both an algebraic and a topologi-

cal structure. In algebraic terms C is a field: C is equipped with two binary
operations, addition and multiplication, satisfying certain axioms (listed be-
low). It is the field one obtains by adjoining to R, the field of real numbers, a
square root of -1. Remarkably, the field so created contains not only square
roots of each of its elements, but even n-th roots for every positive integer n.
All the more remarkable is that adjoining a solution of the single equation
x 2 + 1 = 0 to R results in an algebraically closed field: every nonconstant
polynomial with coefficients in C can be factored over C into linear factors.
This is the "Fundamental Theorem of Algebra," first established by C. F.
Gauss (1777-1855) in 1799. Many different proofs are now known-over
one hundred by one estimate. Gauss himself discovered four. Despite the
theorem's name, most of the proofs, including the simplest ones, are not
purely algebraic. One of the standard ones is presented in Chapter VII.
When complex numbers were first introduced, in the 16th century, and
for many years thereafter, they were viewed with suspicion, a feeling the
reader perhaps has shared. In high school one learns how to add and multiply
two complex numbers, a+ ib and c +id, by treating them as binomials, with
the extra rule i 2 = -1 to be used in forming the product. According to this
recipe,

(a + ib) + (c + id) = (a + c) + i ( b + d)

(a + ib) (c + id) = ( ac - bd) + i (ad + be) .

Everything seems to work out, but where does this number i come from?
How can one just "make up" a square root of -1?

- 1
2 I. Complex Numbers

Nowadays it is routine for mathematicians to create new number systems

from existing ones using the basic constructions of set theory. We shall create
C in a simple direct way by imposing an algebraic structure, suggested by
the rules above for addition and multiplication, on R 2 , the set of ordered
pairs of real numbers.

1.1. Definition of C
The complex number system is defined to be the set C of all ordered pairs
(x, y) of real numbers equipped with operations of addition and multiplica-
tion defined as follows:
(x1, Y1) + (x2, Y2) = (x1 + x2, Y1 + Y2)
(x1, Y1)(x2, Y2) = (x1X2 - Y1Y2, X1Y2 + Y1X2).

I. 2. Field Axioms
The operations of addition and multiplication on C satisfy the following
conditions (the field axioms).
(i) z1 + z2 = z2 + zi, z1z2 = z2z1
(commutative laws for addition and multiplication)
(ii) z1 + (z2 + z3) = (z1 + z2) + z3, z1(z2z3) = (z1z2)z3
(associative laws for addition and multiplication)
(iii) z1 (z2 + z3) = z1z2 + z1z3,
(distributive law)
(iv) The complex number (0, 0) is an additive identity.
(v) Every complex number has an additive inverse.
(vi) The complex number (1, 0) is a multiplicative identity.
(vii) Every complex number different from (0, 0) has a multiplicative in-
verse.
Properties (i), (iv), (v), (vi) and the first identity in (ii) follow easily
from the definitions of addition and multiplication together with the prop-
erties of R. The verifications of the second identity in (ii) and of (iii) are
straightforward but somewhat tedious; they are relegated to Exercise 1.2.l
below. As for (vii), if (a, b) =I- (0, 0), then finding the multiplicative inverse
(x, y) of (a, b) amounts to solving the pair of linear equations
ax - by= 1, bx+ ay = 0
for x and y. The determinant of the system is a 2 + b2 , which is not 0, and
thus a unique solution exists: x = a/(a 2 + b2 ), y = -b/(a 2 + b2 ). In other
I.4. Geometric Representation 3

words,
-1 ( a -b )
(a,b) = a2+b2' a2+b2 .

Exercise 1.2.1. Prove that C obeys the associative law for multiplication
and the distributive law.
Exercise 1.2.2. Find the multiplicative inverses of the complex numbers
(0, 1) and (1, 1).
Exercise 1.2.3. Think of C as a vector space over R. Let c = (a, b) be in
C, and regard multiplication by c as a real linear transformation Tc. Find
the matrix Mc for Tc with respect to the basis (1, 0), (0, 1). Observe that
the map c f-+ Mc preserves addition and multiplication. Conclude that the
algebra of two-by-two matrices over R contains a replica of C.

1.3. Embedding of R in C. The Imaginary Unit

The set of complex numbers of the form (x, 0) is a subfield of C; it is the
isomorphic image of R under the map x f-+ (x, 0). Henceforth we shall iden-
tify this subfield with R itself; in particular, we shall notationally identify
the complex number (x, 0) with the real number x. Additionally, we use
the symbol i to denote the complex number (0, 1), the so-called imaginary
unit; it is one of the two square roots of -1 ( = ( -1, 0)), the other being
-i. (The reader should verify these statements directly from the definition
of how complex numbers multiply.) With these conventions, the complex
number (x, y) can be written as x + iy (or as x + yi).

1.4. Geometric Representation

Since a complex number is nothing but an ordered pair of real numbers, it
can be envisioned geometrically as a point in the coordinatized Euclidean
plane. To say it another way, each point in the plane can be labeled by a
complex number. The real numbers correspond to points on the horizontal
axis, which is thus referred to in this context as the real axis. Complex
numbers of the form i y with y real, so-called purely imaginary numbers,
correspond to points on the vertical axis, which is thus referred to as the
imaginary axis. When wishing to emphasize this geometric interpretation,
we shall refer to C as the "complex plane."
In geometric terms, the addition of two complex numbers is just vector
addition according to the parallelogram law. The geometric interpretation
of multiplication will be presented later, in Section I.9.
If z = x + iy is a complex number, then the Euclidean distance of z from
the origin is denoted by lzl and is called the absolute value (or the modulus)
4 I. Complex Numbers

of z: lzl = Jx 2 + y 2. The reflection of z with respect to the real axis is

denoted by z and is called the complex conjugate of z: z = x - iy. The
coordinates of z on the horizontal and vertical axes are denoted by Re z
and Im z, respectively, and are called the real and imaginary parts of z:
Re z = x, Im z = y. The reader should verify the following basic identities:
lzl = lzl = ../ZZ, Rez = ~(z+z), Imz = ~(z-z).
Exercise* 1.4.1. Prove that if z1 and z2 are complex numbers then z1 + z2 =
z1 + z2, z1z2 = z1z2, and lz1z2I = lz1llz2I·
Exercise* 1.4.2. Prove that the nonzero complex numbers z1 and z2 are
positive multiples of each other if and only if z1z2 is real and positive. (Note
that, in geometric terms, z1 and z2 are positive multiples of each other if
and only if they lie on the same ray emanating from the origin.)
Exercise 1.4.3. Prove that if a polynomial with real coefficients has the
complex root z, then it also has z as a root.

Note: An exercise whose number is preceded by an asterisk will be referred

to later in the text. In many cases the result established in the exercise will
be needed as part of a proof.

1.5. Triangle Inequality

If z1 and z2 are complex numbers, then Iz1 + z2 I :=::; Iz1 I + Iz2 I· The inequality
is strict unless one of z1 and z2 is zero, or z1 and z2 are positive multiples
of each other.
In view of the interpretation of addition in C as vector addition, the in-
equality expresses the geometric fact that the length of any side of a triangle
does not exceed the sum of the lengths of the other two sides. To obtain an
analytic proof we note that
(lz1I + lz21) 2 - lz1 + z21 2 = (lz1I + lz21) 2 - (z1 + z2)(z1 + z2)
= lz11 2 + lz21 2 + 2lz1llz2I - z1z1 - z1z2 - z2z1 - z2z2
= 2lz1llz2I - z1z2 - z2z1
= 2(lz1z2I - Re z1z2).
The reader will easily validate this string of equalities on the basis of some of
the identities from the preceding section (including those in Exercise I.4.1).
Now if z is any complex number it is plain from the basic definitions that
lzl 2: Re z, with equality if and only if z is real and nonnegative. From this
we conclude, in view of the equality between the extreme left and right sides
in the string above, that (I z1 I+ Iz2 I) 2 2: Iz1 + z2 l2, with equality if and only if
z 1z 2 is real and nonnegative. By Exercise I.4.2 in the preceding section, the
I. 7. Plane Geometry via Complex Numbers 5

latter happens if and only if one of z1 and z2 is 0, or z 1 and z 2 are positive

multiples of each other. This establishes the triangle inequality.
Exercise* 1.5.1. Prove that if z1, z2, ... ,zn are complex numbers then
lz1 + z2 + · · · + Znl :S lz1I + lz2I + · · · + lznl·
Exercise* 1.5.2. Prove that if z1 and z2 are complex numbers then
lz1 - z2I?: lz1l - lz2I. Determine the condition for equality.

1.6. Parallelogram Equality

If z1 and z2 are complex numbers, then
lz1 + z21 2 + lz1 - z21 2 = 2 (lz11 2 + lz21 2).

In geometric terms the equality says that the sum of the squares of the
lengths of the diagonals of a parallelogram equals the sum of the squares of
the lengths of the sides. To establish it we note that the left side equals
(z1 + z2)(z1 + z2) + (z1 - z2)(z1 - z2),
which, when multiplied out, becomes
z1z1 + z1z2 + z2z1 + z2z2 + z1z1 - z1z2 - z2z1 + z2z2.
After making the obvious cancellations, we obtain the right side.

I. 7. Plane Geometry via Complex Numbers

The two preceding sections illustrate how geometric facts can be translated
into the language of complex numbers, and vice versa. The following exer-
cises contain further illustrations.
Exercise 1.7.1. Prove that if the complex numbers z1 and z2 are thought
of as vectors in R 2 then their dot product equals Re z 1z2. Hence, those
vectors are orthogonal if and only if z1z2 is purely imaginary (equivalently,
if and only if z1z2 + z1z2 = 0).
Exercise 1.7.2. Let z1, z2, z3 be points in the complex plane, with z1 f:- z2.
Prove that the distance from Z3 to the line determined by z1 and z2 equals
1
I lz1(z2 - z3) + z2(z3 - z1) + z3(z1 - z2)I;
2 z2 - z1 1
in particular, the points z1, z2, z3 are collinear if and only if z1(z2 - z3) +
z2(z3 - z1) + z3(z 1 - z2) = 0. (Suggestion: Reduce to the case where z1 = 0
and z2 is real and positive.)
Exercise I. 7.3. Prove that if z1, z2, z3 are noncollinear points in the com-
plex plane then the medians of the triangle with vertices z1, z2, z3 intersect
i
at the point (z1 + z2 + z3).
6 I. Complex Numbers

Exercise 1.7.4. Prove that the distinct complex numbers z 1 , z 2 , z 3 are the
vertices of an equilateral triangle if and only if
zr + Z~ + Z~ = ZIZ2 + Z2Z3 + Z3Z1.
1.8. C as a Metric Space
Since C as a set coincides with the Euclidean plane, R 2 , it becomes a metric
space if we endow it with the Euclidean metric. In this metric, the distance
between the two points z1 and z2 equals Iz1 - z2 I- We can thus use in C
all of the standard notions from the theory of metric spaces, such as open
and closed sets, compactness, connectedness, convergence, and continuity.
For example, a sequence (zn)i'° in C will be said to converge to the complex
number z provided
lim Iz - Zn I = 0.
n---+ex>

Exercise 1.8.1. Prove that each of the following inequalities defines an open
subset of C:
• lzl <1,
• Rez > 0,
• lz + z21<1.
Exercise 1.8.2. Prove that the following complex-valued functions are con-
tinuous at those points of C where they are defined.
1 1
(a) f(z) = z 2 , (b) g(z) = -, (c) h(z) = - 2- .
z z -1
Exercise* 1.8.3. Prove that addition and multiplication define continuous
maps of CxC into C.

1.9. Polar Form

Let z = x + iy be a nonzero point in the complex plane, and let (r, B) be
polar coordinates for the point. Thus, r = lzl, and () is the angle between
the positive real axis and the ray from the origin through z, with the usual
sign convention. The angle () is called an argument of z, written() = arg z.
This angle is determined by z only to within addition of an integer multiple
of 27r, in other words, if() = arg z, then () + 27rk = arg z for every integer
k. (This mild misuse of the equality sign will cause no trouble in practice.)
The particular value of arg z in the interval ( -7!", 7r] is called the principal
value and denoted by Arg z. For example, Arg 1 = 0, Arg (-1) = 7r, Arg
i = I' Arg (-i) = -I·
Because x = r cos() and y = r sine, the number z can be rewritten as
z = r (cos() +i sine).
I.10. De Moivre's Formula 7

The expression on the right is the polar form of z.

Consider two nonzero complex numbers in polar form,
z1 = rl(cos01 +i sin01) and z2 = r2(cos02 +i sin02).
Their product is given by
z1z2 = rlr2[( cos 01cos02 - sin 01sin02) + i (cos 01sin02 +sin 01cos02)],
which, because of the addition formulas for the sine and cosine functions,
can be rewritten as
ziz2 = rlr2[cos (01 + 02) + i sin (01 + 02)].
The expression on the right side is in polar form, giving us a geometric
picture of the effect of complex multiplication: when one multiplies two
nonzero complex numbers, one multiplies their absolute values and adds
their arguments.
Exercise* 1.9.1. Prove that arg z = arg z- 1 = -arg z for any nonzero
complex number z.
Exercise 1.9.2. Let z 1 , z2, z3 be distinct points on the unit circle (i.e.,
IZj I = 1 for each j). Prove that
z1 Z3 - z1
arg- = 2arg ,
z2 z3 - z2

and interpret the equality geometrically. (Suggestion: arg a = 2 arg b if and

only if a b2 is real and positive.)

1.10. De Moivre's Formula

If z = r( cos 0 + i sin 0) is a nonzero complex number then, for every integer
n,
zn = rn (cos n() + i sin nO).
For n > 0, one can establish this by repeated applications of the product
formula from the last section. The case n = -1 follows from Exercise I.9.1
in the last section. Once the case n = -1 has been established, the case
n < 0 follows from the case n > 0.
A typical application: By de Moivre's formula,
cos30 + i sin30 =(cos()+ i sin0) 3
= cos3 () + 3cos 2 0 (i sinO) + 3cos0 (i sin()) 2 + (i sin()) 3
= cos 3 0 - 3 cos 0 sin 2 0 + i (3 cos 2 0 sin() - sin3 0).
Equating real and imaginary parts, we obtain the following trigonometric
identities:
cos 30 = cos 3 0 - 3 cos 0 sin2 0,
8 I. Complex Numbers

Exercise I.10.1. Use de Moivre's formula to find expressions for cos5B and
sin 58 as polynomials in cos e and sin e.
Exercise 1.10.2. Establish the formulas

cos e + i· sm· e + 1 = 2 cos e(

2 cos 2e + i· sm
· 20 ) ,
cos e+ i· sm· e - 1 = 2 i· sm
· e(
2 cos 2e + i· sm
· 20 ) .
Exercise I.10.3. Establish the formulas
1 sin(n+l)e
- + cos
2
e + cos 28 + ... + cos ne = 2 sin 2
r} '
cos Q - cos( n + l )e
sin e + sin 28 + ... + sin ne = 2 e 2
2 sin 2
(important in the theory of Fourier series) by making the substitution z =
e
cos + i sine in the identity
2 n zn+l - 1
l+z+z + .. ·+z = , (z-::/: 1).
z-1
1.11. Roots

De Moivre's formula enables us to find then-th roots of any complex number.

To illustrate, suppose we seek the cube roots of l. Any such cube root
obviously has absolute value 1, so it has the form cos a+ i sin a for some
angle a. By de Moivre's formula we must have cos 3a+i sin 3a = 1, which is
true if and only if 3a = 27rk for some integer k. The cube roots of 1 are thus
27rk . 27rk .
the numbers cos 3 +ism 3 with k an integer. Two different choices
of k yield the same root if and only if the choices are congruent modulo 3.
There are thus three distinct cube roots of 1, which we obtain, for example,
from the choices k = 0, 1, 2; besides 1 itself, they are
27r . . 27r 1 iJ3
cos 3 +i sm 3 = - 2 + - 2-,

47r . . 47r 1 iJ3

cos- +ism-= - - - - - .
3 3 2 2
By similar reasoning one sees that if z = r( cos e+ i sin B) is any nonzero
complex number and n is any positive integer, then the n-th roots of z are
the numbers
i (
rn cos
e + 27rk + i. sm. e + 27rk) ' k = 0, 1,. . ., n - l.
n n
Exercise I.11.1. Find all cube roots of i.
I.12. Stereographic Projection 9

(0,0, 1)
''

z
....... ............ . .... ''
'

Figure 1. The stereographic projection

Exercise 1.11.2. Find all eighth roots of 1.

Exercise 1.11.3. An n-th root of 1 is called a primitive n-th root if it is

not an m-th root of 1 for any positive integer m less than n. Prove that the
number of primitive n-th roots of 1 is the same as the number of positive
integers less than and relatively prime to n. Prove that one can obtain
every n-th root of 1 by taking the first through then-th powers of any given
primitive n-th root.

Exercise 1.11.4. Prove that the sum of then-th roots of 1equals0, (n > 1).

Exercise 1.11.5. Let w be an n-th root of 1 different from 1 itself. Establish

the formulas
2
1 + 2w + 3w + · · · + nw
n-1
= -n- ,
w- l
2 2n
1 + 4w + 9w 2 + · · · + n 2wn-I = _n_ - ---~
w-l (w-1)2·

1.12. Stereographic Projection

To obtain another geometric picture of C, we identify C with the horizontal
coordinate plane in R 3 . Let 8 2 denote the unit sphere (the sphere of unit
radius with center (0,0,0)) in R 3 . A given point z = x+iy in C and
the point (0, 0, 1) (the north pole of 8 2 ) determine a line in R 3 . That line
intersects 8 2 at one point other than the north pole. We denote that other
point by (~, 17, () and map C to 8 2 by sending z to (~, 17, (). The map is
clearly one-to-one, and each point of the sphere other than the north pole
is an image point.
10 I. Complex Numbers

To represent the map analytically we note that the line determined by

z and (0, 0, 1) is given parametrically by the function
(tx, ty, 1 - t), -00 < t < 00
(t = 0 corresponds to the north pole and t = 1 to z). The point (tx, ty, 1 - t)
lies on S 2 provided
t2x2 + t2y2 + (1 - t)2 = 1,
which has the two solutions t = 0 (giving the north pole) and t = 2/
(x 2 + y 2 + 1) (giving the image point (~, 77, ()). From this one sees that our
map is given by
(2Rez, 2Imz, lzl 2 - 1)
z f---t
lzl2+1
The reader should verify that the inverse map is given by
~ + i 77
(~, 77, () f---t ~-

The latter map is the stereographic projection of cartographers.

Both the map from C to S 2 and its inverse are continuous and so preserve
topological properties such as convergence and compactness. For example,
if a sequence in C converges, then so does the corresponding sequence on
S 2' and if a sequence on S 2 converges to a point other than the north pole,
then the corresponding sequence in C converges.
A few geometric features of the correspondence are easily seen from the
figure above. For example, the unit circle in C corresponds to the equator
of S 2, the interior of the unit circle corresponds to the southern hemisphere
of S 2' and straight lines in c correspond to circles on S 2 passing through
the north pole.
In Exercise III.8.1 the reader will be asked to show that a circle on
S 2 not passing through the north pole corresponds under the stereographic
projection to a circle in C.

1.13. Spherical Metric

For z1 and z2 in C, let p(z1, z2) denote the Euclidean distance between the
corresponding points on S 2. This gives a metric in C, the spherical metric,
equivalent to the Euclidean metric (i.e., yielding the same family of open
sets).
Exercise 1.13.1. Establish the following formula for the spherical metric:
2lz1-z2I
p (z1,z2 ) =
vlz1l 2 + 1vlz21 2 +1
I.14. Extended Complex Plane 11

(0, 0, 1)

Figure 2. Transformation of circles under the stereographic projection

1.14. Extended Complex Plane

In many situations it is convenient to work with the enlargement of C that
one obtains by appending a "point at infinity," corresponding to the north
pole of S 2 . We denote this added point by oo. (Set-theoretic purists can
take oo to be any object not already in C.) By the extended complex plane
we shall mean the set C = CU { oo}, to which we extend the spherical metric
by defining p( z, oo), for z in C, to be the Euclidean distance between the
point on S 2 corresponding to z and the north pole. Thus, for example, a
sequence (zn)l' in C converges to oo in this metric if and only if lznl ----> oo
(in the standard sense).
The extended complex plane is often referred to as the Riemann sphere,
after G. F. B. Riemann (1826-1866), a pioneer in our subject whose ideas
had and continue to have a profound influence on its development.
Exercise I.14.1. Establish the formula
2
p(z, oo) = Jlzl2 + 1
 Chapter 11

Complex
Differentiation

Having introduced the complex number system, we proceed to the develop-

ment of the theory of functions of a complex variable, beginning with the
notion of derivative. Although the definition of the derivative of a complex-
valued function of a complex variable is formally the same as that of the
derivative of a real-valued function of a real variable, the concept holds
surprises, as we shall see.
Generally, we shall let z denote a variable point in the complex plane;
its real and imaginary parts will be denoted by x and y, respectively.

11.1. Definition of the Derivative

Let the complex-valued function f be defined in an open subset G of C.
Then f is said to be differentiable (in the complex sense) at the point zo of
G if the difference quotient f(z) - f(zo) has a finite limit as z approaches
z - zo
zo. That limit is then called the derivative off at zo and denoted by J'(zo):

J' (zo) = lim f (z) - f (zo) .

Z--->Zo Z - ZQ

What does the last equality mean? In E-J language, it is the statement that
for every positive number E there is a positive number J such that

I f (z)z -- f (zo) - J' (zo) I < E

zo
whenever 0 < lz - zol < J.

-13
14 II. Complex Differentiation

As in calculus, the "d-notation" is also used for the derivative: if f is

differentiable at zo, then f 1 ( zo) is according to convenience denoted alterna-
. 1 b df(zo)
t1vey y ~·

NB. According to our definition, f'(zo) cannot be defined unless zo belongs

to an open set in which f is defined.

11.2. Restatement in Terms of Linear Approximation

Let the complex-valued function f be defined in an open subset of C con-
taining the point zo. Then f is differentiable in the complex sense at zo
if and only if there is a complex number c such that the function R( z) =
f(z) - f(zo) - c(z - zo) satisfies lim R(z) = 0, in which case f'(zo) = c.
Z--->ZQ Z - ZQ

The statement is obvious in view of the equality

R(z)
--=
f (z) - f (zo) -c.
z-zo z-zo
The statement says that f is differentiable at zo, with f'(zo) = c, if and only
if f is well approximated near zo by the linear function f ( zo) + c( z - zo), in
the sense that the remainder R(z) in the approximation is small compared
to the distance from zo.

11.3. Immediate Consequences

The following properties of complex differentiation are proved from the basic
definition in exactly the same way as the corresponding properties in the
theory of functions of a real variable.
(i) If f is differentiable at zo then f is continuous at zo.
(ii) If f and g are differentiable at zo, then f +g and f g also are, and
(f + g)'(zo) = f'(zo) + g'(zo) (sum rule};
(f g)'(zo) = f'(zo)g(zo) + f(zo)g'(zo) (product rule).
If in addition g(zo) #- 0, then f /g is differentiable at zo, and

( £)'(zo) = f'(zo)g(zo) - ~(zo)g'(zo) (quotient rule}.

g g(zo)
(iii) If f is differentiable at zo and g is differentiable at f(zo), then the
composite function g o f is differentiable at zo and
(go J)'(zo) = g'(f(zo))f'(zo) (chain rule}.
The proofs are left to the reader.
Exercise 11.3.1. Prove statements (i)-(iii) in detail.
II.5. Comparison of Differentiability 15

11.4. Polynomials and Rational Functions

From the definition of derivative it is immediate that a constant function is
differentiable everywhere, with derivative 0, and that the identity function
(the function f (z) = z) is differentiable everywhere, with derivative 1. Just
as in elementary calculus one can show from the last statement, by repeated
applications of the product rule, that, for any positive integer n, the function
f(z) = zn is differentiable everywhere, with derivative nzn-l. This, in
conjunction with the sum and product rules, implies that every polynomial
is everywhere differentiable: If f(z) = Cnzn+cn-1Zn-l+· · ·+c1z+co, where
co, ... , Cn are complex constants, then f'(z) = ncnzn-l+ (n- l)cn-1Zn- 2 +
· · · + C1.
A function of the form f / g, where f and g are polynomials, is called
a rational function. Such a function is defined wherever its denominator,
g, does not vanish, hence everywhere except on a finite set. The quotient
rule and the differentiability of polynomials imply that a rational function
is differentiable at every point where it is defined and that its derivative is
a rational function.

11.5. Comparison Between Differentiability in the Real and

Complex Senses
Recall that a real-valued function u defined in an open subset G of R 2 is said
to be differentiable (in the real sense) at the point (xo, Yo) of G if there are
real numbers a and b such that the function R(x, y) = u(x, y) - u(xo, Yo) -
a(x - xo) - b(y - Yo) satisfies

lim R(x, y) = 0.
(x,y)->(xo,yo) J(x - xo) 2 + (y - Yo) 2

In that case u has first partial derivatives at (xo, yo) given by

ou(xo, Yo) _ ou(xo, Yo) _ b

ox - a, oy - .

The reader will find this notion discussed in any multivariable calculus book.
To facilitate a comparison with complex differentiation, we restate the
preceding definition in complex notation: the real-valued function u in the
open subset G of C is by definition differentiable at the point zo = xo + iyo
of G if there are real numbers a and b such that the function R(z) = u(z) -
u(zo) - a(x - xo) - b(y - Yo) satisfies

lim R(z) = 0.
z->zo Z - zo
16 II. Complex Differentiation

Now suppose that f is a complex-valued function defined in the open

subset G of C, and let u and v denote its real and imaginary parts: f = u+iv.
Given a point zo = xo + iyo of G and a complex number c =a+ ib, we can
write

R(z) = f(z) - f(zo) - c(z - zo) = [u(z) - u(zo) - a(x - xo) + b(y - Yo)]
+ i [v(z) - v(zo) - b(x - xo) - a(y - Yo)]
= R1 (z) + iR2(z).

Clearly, lim R(z) = 0 if and only if lim Ri(z) = 0 and lim R 2(z) = 0.
Z->Zo z - zo Z->Zo z - zo Z->Zo z - zo

Referring to II.2, we can draw the following conclusion:

The function f is differentiable (in the complex sense) at zo if and only if

u and v are differentiable (in the real sense) at zo and their first partial
au(zo) av(zo) au(zo) av(zo)
derivatives satisfy the relations ax ------ay,
ay -------a;;-. In
that case,

f '( ) _ au(zo) .av(zo) _ av(zo) _ .au(zo)

zo - ax +i ax - ay i ay .

11.6. Cauchy-Riemann Equations

The two partial differential equations
au av au av
ax ay' ay ax
are called the Cauchy-Riemann equations for the pair of functions u, v. As
seen above, the equations are satisfied by the real and imaginary parts of a
complex-valued function at each point where that function is differentiable.

Exercise 11.6.1. At which points are the following functions f differen-

tiable?
(a) f(z) = x, (b) f(z) = z, (c) f(z) = z2 .

Exercise 11.6.2. Prove that the function f(z) = JTXYT

is not differentiable
at the origin, even though it satisfies the Cauchy-Riemann equations there.

Exercise* 11.6.3. Prove that the Cauchy-Riemann equations in polar co-

ordinates are
au av
r ar = ae'
II.8. Holomorphic Functions 17

II. 7. Sufficient Condition for Differentiability

A theorem from the theory of functions of a real variable states that if a
real-valued function of several variables has first partial derivatives, then it is
differentiable at every point where those partial derivatives are continuous.
(This can be found in any multivariable calculus book. For the convenience
of readers who have not seen a proof, one is given in Appendix I.) In
combination with the necessary and sufficient condition from II.5, this gives
the following useful sufficient condition for complex differentiability: Let
the complex-valued function f = u + iv be defined in the open subset G of
C, and assume that u and v have first partial derivatives in G. Then f is
differentiable at each point where those partial derivatives are continuous
and satisfy the Cauchy-Riemann equations.

11.8. Holomorphic Functions

A complex-valued function that is defined in an open subset G of C and
differentiable at every point of G is said to be holomorphic (or analytic)
in G. The simplest examples are polynomials, which are holomorphic in
C, and rational functions, which are holomorphic in the regions where they
are defined. Later we shall see that the elementary functions of calculus-
the exponential function, the logarithm function, trigonometric and inverse
trigonometric functions, and power functions-all have complex versions
that are holomorphic functions.
By II.5 we know that the real and imaginary parts of a holomorphic
function have partial derivatives of first order obeying the Cauchy-Riemann
equations. In the other direction, by II. 7, if the real and imaginary parts of
a complex-valued function have continuous first partial derivatives obeying
the Cauchy-Riemann equations, then the function is holomorphic.
The asymmetry in the two preceding statements-the inclusion of a con-
tinuity condition in the second but not in the first-relates to an interesting
and subtle theoretical point. The derivative of a holomorphic function, as
will be shown later (in Section VII.8), is also holomorphic, so that in fact a
holomorphic function is differentiable to all orders, and its real and imagi-
nary parts have continuous partial derivatives to all orders. We shall only
be able to prove this, however, after developing a fair amount of machinery.
Meanwhile, we shall have to skirt around it occasionally.
Although, as we have seen above, some of the basic properties of real
and complex differentiability are formally identical, the repeated differen-
tiability of holomorphic functions points to a glaring dissimilarity. There
are well-known examples of continuous real-valued functions on R that are
18 II. Complex Differentiation

nowhere differentiable. An indefinite integral of such a function is differen-

tiable everywhere while its derivative is differentiable nowhere. By taking
an n-fold indefinite integral, one can produce a function that is differen-
tiable to order n yet whose n-th derivative is nowhere differentiable. Such
"pathology" does not occur in the realm of complex differentiation.
From the basic rules of differentiation noted in Section 11.3 one sees
that if f and g are holomorphic functions defined in the same open set G,
then f + g and f g are also holomorphic in G, and f / g is holomorphic in
G\g- 1 (0). If f is holomorphic in G and g is holomorphic in an open set
containing f(G), then the composite function go f is holomorphic in G.

Exercise* 11.8.1. Let the function f be holomorphic in the open disk D.

Prove that each of the following conditions forces f to be constant: (a)
f' = 0 throughout D; (b) f is real-valued in D; (c) lfl is constant in D; (d)
arg f is constant in D.

Exercise* 11.8.2. Let the function f be holomorphic in the open set

G. Prove that the function g(z) = f(z) is holomorphic in the set G* =
{z: z E G}.

11.9. Complex Partial Differential Operators

The partial differential operators !, and gy are applied to a complex-valued
function f = u + iv in the natural way:

We define the complex partial differential operators ffz and ?z by

- a 1(0 .a)
az = -2 ax ay '
--i-

Thus,
a=
ax oza +oz'
a aya = ·(a a) .
oz - oz
i

Intuitively one can think of a holomorphic function as a complex-valued

function in an open subset of C that depends only on z, i.e., is independent
of z. We can make this notion precise as follows. Suppose the function
f = u +iv is defined and differentiable in an open set. One then has
af = ~ (au + av) + ~ (av _ au)
az 2 ax ay 2 ax {)y '
af = ~ (au _ av) + ~ (av + au) .
oz 2 ax ay 2 ax ay
II.10. Picturing a Holomorphic Function 19

The Cauchy-Riemann equations thus can be written 8Jz = 0. As this is

the condition for f to be holomorphic, it provides a precise meaning for the
statement: "A holomorphic function is one that is independent of z ." If f
is holomorphic, then (not surprisingly) f' = ~~, as the following calculation
shows:

J' = af = af + af = af.
ox oz az Bz
11.10. Picturing a Holomorphic Function
One can visualize a real-valued function of a real variable by means of its
graph, which is a curve in R 2 . A complex-valued function of a complex
variable also has a graph, but its graph is a two-dimensional object in the
four-dimensional space C x C, something ordinary mortals cannot easily vi-
sualize. A more sensible approach, if one wants to obtain a geometric picture
of a holomorphic function, is to think of the function as a map from the com-
plex plane to itself, and to try to understand how the map deforms the plane;
for example, how does it transform lines and circles?
The simplest case is that of a linear function, a function f of the form
f(z) = az + b, where a and bare complex numbers, and a I- 0 (to exclude
the trivial case of a constant function). The map z f-t az + b can be written
as the composite of three easily understood transformations:

z f-t lalz f-t az f-t az + b.

The first transformation in the chain is a scaling with respect to the origin
by the factor lal, a so-called homothetic map about the origin. The second
transformation is multiplication by the number a/lal, which is just rotation
about the origin by the angle arg a. The last transformation is translation
by the vector b. We see in particular that the linear function f(z) = az + b
maps straight lines onto straight lines and preserves the angles between
intersecting lines.
Linear functions are very special, but remember that a holomorphic
function is a function that is well approximated locally by linear functions.
If the function f is holomorphic in a neighborhood of the point zo, one
would expect it to behave near zo approximately like the linear function
z f-t f'(zo)(z - zo) + f(zo). If f'(zo) = 0 this will tell us little, but if
f'(zo) I- 0 it should say something about the "infinitesimal" deformation
produced by f near zo. As we shall see, this is indeed the case: if f'(zo) I- 0,
the holomorphic function f preserves the angles between curves intersecting
at zo. To make this precise we need some preliminaries about curves in the
complex plane.
20 II. Complex Differentiation

11.11. Curves in C
By a curve in C we shall mean a continuous function / that maps an interval
I of R into C. Thus, curves for us will always be parametrized curves.
However, we shall often speak of curves as if they were subsets of C. For
example, we shall say that the curve/ is contained in a given region of C if
the range of I is contained in that region.
Here are a few simple examples.

1. 1(t) = (1- t)z1 + tz2 (-oo < t < oo).

Here, z1 and z2 are distinct points of C. This curve is a parametrization of
the straight line determined by z1 and z2, the direction of the parametriza-
tion being from z1 to z2.

2. 1(t) =cost+ isint (0:::; t:::; 27r).

This is a parametrization of the unit circle, the circle being traversed once
in the counterclockwise direction as t moves from the initial to the terminal
point of the parameter interval [O, 27r].

3. 1(t) =cost- isint (-27r:::; t:::; 27r).

This also is a parametrization of the unit circle, but this time the circle is
traversed twice in the clockwise direction.

4. In this example, 1(t) is defined piecewise:

t, o::;t::;1,
l+(t-l)i, 1 :::: t :::: 2,
1(t) =
i + 3 - t, 2 :::: t :::: 3,
(4 - t)i, 3 :::: t :::: 4.

This is a parametrization of the square with vertices 0, 1, 1 + i, i. The

square is traversed once in the counterclockwise direction.

The curve I : I ---t C is said to be differentiable at the point to of I if

its real and imaginary parts are differentiable at to, or, what is equivalent,
if the difference quotient i(t) - i(to) approaches a finite limit as t tends to
t - to
to. That limit is then denoted by 1'(to). The curve/ is called differentiable
if it is differentiable at each of its points; it is said to be of class C 1 if it is
differentiable and its derivative, 1', is continuous.
The curve/ is said to be regular at the point to if it is differentiable at to
and 1'(to) #- 0. If I is of class C 1 and regular at each point of its interval of
definition, we call it a regular curve. The curves in the first three examples
II.12. Conformality 21

above are regular. The one in the fourth example is regular except at the
points 1, 2, 3 of the parameter interval [O, 4].
A curve"( has a well-defined direction at each point to where it is regular,
namely, the direction determined by the derivative 1' (to), referred to as the
tangent direction. We can describe that direction, for example, by specifying
the argument of 1'(t0 ), or by specifying the unit tangent vector, l~:~~~~I ·
Suppose 'YI and "12 are two curves in C, and suppose they have a point
of intersection, say 'YI (ti) = "12 (t2). Suppose further that "/j is regular at tj,
j = 1, 2. Then by the angle between "(1 and "(2 we shall mean the angle
arg "!~ ( t2) - arg "!~ (ti) (= arg "(~ (t2 h~ (ti)). In geometric terms, this is the
angle through which one must rotate the unit tangent vector to 'YI at ti to
make it coincide with the unit tangent vector to "/2 at t2. Note that the
angle depends on the order in which we take 'YI and 1 2; reversal of the order
leaves the magnitude of the angle the same but changes its sign. (To be
completely precise, perhaps we should speak of the "angle between 'YI and
"12 corresponding to the parameter values ti and t2" because the two curves
might intersect for other pairs of parameter values. This degree of precision
would not be worth the awkwardness of expression it would entail.)
Suppose that f is a holomorphic function in an open set G and that "! is
a curve in G. Then we can apply f to"( to obtain the curve f o "!· Suppose
"! is differentiable at to, and let zo = 1(to). Then the standard argument
justifying the chain rule applies to show that f o "! is differentiable at to and
that (Jo 1)'(t0 ) = f'(zoh'(to). (Details are in Appendix 2.) Thus, if"( is
regular at to and if f' (zo) #- 0, then f o "! is regular at to, and one obtains
the direction of f o "! at to from that of "! at to by adding arg f' (zo).

11.12. Conformality
Let f be a holomorphic function defined in the open subset G of C, and let
zo be a point of G such that J' (zo) #- 0. Let ')'1 and /'2 be curves such that
1'1(t1) = 1'2(t2) = zo, and such that /'j is regular at tj, j = 1, 2. Then the
angle between f o /'I and f o /'2 equals the angle between /'I and ')'2.
This statement follows immediately from the discussion preceding it,
from which one sees that

j = 1, 2.

The function f(z) = z 2 shows what can happen if the hypothesis f'(zo) -1-
0 is dropped. This function, whose derivative vanishes at the origin, trans-
forms two lines through the origin making an angle a into two lines making
22 II. Complex Differentiation

an angle 2a. On the other hand, as we shall see in a later chapter, the deriv-
ative of a nonconstant holomorphic function can vanish only on an isolated
set of points, so the angle-preservation property given by the theorem above
is the rule rather than the exception.
A map from the plane to the plane is called conformal at the point z0 if it
preserves the angles between pairs of regular curves intersecting at zo. Thus,
we can restate II.11 by saying that a holomorphic function is conformal at
each point where its derivative does not vanish.

11.13. Conformal Implies Holomorphic

We shall now show that conformal maps are necessarily holomorphic. We
begin with the simplest case, that of a linear transformation of the plane.
Linear here means linear as a transformation of the real vector space R 2 to
itself. If f is such a map then ~~ and ~~ are constants, and f is uniquely
determined by those constants plus the condition f (0) = 0. Letting a= ~~
and b = U (also constants), we see that f(z) = az + bz. Suppose this
map preserves the angles between pairs of directed lines intersecting at the
origin. We shall then prove that b = 0. We may assume that a + b i-
0 (since otherwise the transformation would send the whole real axis to
the origin) and, that done, that a i- 0 (since otherwise the map would be
anticonformal-it would reverse the angles between pairs of directed lines).
Let A be a complex number of absolute value 1. Our map sends the real line
to the directed line through the origin determined by a+ b, and it sends the
directed line through the origin determined by .\ to the one determined by
a.\ + b>... Our assumption about angle preservation thus implies that

arg (a.\+ b-X.) - arg (a+ b) = arg .\.

Since the left side in this equality equals

arg.X+arg(a+ b;)-arg(a+b),

the equality reduces to

arg (a+ b).

Now, if bi- 0 then, as ,\ traverses the unit circle, the point a+ b; (twice)
traverses the circle with center a and radius lbl, in violation of the preceding
equality, which
II.14. Harmonic Functions 23

b>..
says that a + >: lies on the ray through the origin determined by a+ b. We
can conclude that b = O; in other words, our map is given by z f--+ az, a
holomorphic function, as desired.
The preceding result will serve as a lemma for its generalization to C 1
maps. We consider a complex-valued function f = u + iv defined on an
open subset G of C. We assume that u and v have continuous first partial
derivatives. Then, if 'Y is a differentiable curve in G, the curve f o 'Y is also
differentiable. In fact, suppose zo is a point on"(, say 1(to) = zo. Let

of(zo) b = of(zo)
a
oz oz

An application of the chain rule, the details of which are in Appendix 2,

then shows that

(f o 1)'(to) = a1'(to) + b1'(to).

Hence, if f preserves the angles between pairs of regular curves intersecting
at zo, then the linear map z f--+ az + bz preserves the angles between pairs
of directed lines through the origin. By what is established above, that
means b = 0 and a i- 0. The equality b = 0 just says that the functions
u and v satisfy the Cauchy-Riemann equations at zo, which, as noted in
Section II. 7, implies that f is differentiable (in the complex sense) at zo.
Moreover, f'(zo) = EJJ~;o) =a.
The following theorem has been proved: Let f be a complex-valued func-
tion, defined in an open subset G of C, whose real and imaginary parts have
continuous first partial derivatives. If f preserves the angles between regular
curves intersecting in G, then f is holomorphic, and f' is never 0.

11.14. Harmonic Functions

The complex-valued function f, defined in an open subset of C, is said to
be harmonic if it is of class C 2 and satisfies Laplace's equation:

This equation and its higher-dimensional versions play central roles in many
branches of mathematics and physics. Of course, the complex-valued func-
tion f is harmonic if and only if its real and imaginary parts are.
24 II. Complex Differentiation

11.15. Holomorphic Implies Harmonic

A holomorphic function is harmonic, provided it is of class C 2 .
As noted in Section II.8, we shall prove later that a holomorphic function
is of class Ck for all k, at which point we can drop the proviso in the preceding
statement. To establish the proposition, let the function f = u + iv be
holomorphic and of class C 2 . By the Cauchy-Riemann equations, we have
a2u a (av) a (av) a2 u
ax 2 = ax ay = ay ax = - ay 2 '
which proves that u is harmonic. Similar reasoning proves the same result
for v, and thus f is harmonic.

11.16. Harmonic Conjugates

The reasoning in the preceding section shows that a pair of real-valued C 2
functions u and v, defined in the same open subset of C, will be harmonic
if they satisfy the Cauchy-Riemann equations:
OU
-
ov
OU
ox oy' -oy
In this situation one says that v is a harmonic conjugate of u. Phrased
differently, if u and v are real valued and of class C 2 , then v is a harmonic
conjugate of u if and only if u + iv is holomorphic.
Note that harmonic conjugates are not unique: if v is a harmonic con-
jugate of u then so are the functions that differ from v by constants. That
is essentially the extent of nonuniqueness, as we shall see later (and, in a
special case, in Exercise 11.16.4 below). A natural question is whether every
harmonic function has a harmonic conjugate. We shall eventually develop
enough machinery to deal with this question.

Friendly Advice. When beginning the study of complex analysis and faced
with a problem in the subject, the initial response of many students is to
reduce the problem to one in real variables, a subject they have previously
studied. Such a reduction can sometimes be helpful, but at other times it
can make things overly complicated. (Some of the exercises below illustrate
this point.) Try to get into the habit of "thinking complex."
Exercise 11.16.1. For which values of the real constants a, b, c, d is the
x, y) ax bx y cxy dy
function u ( = 3+ 2 + 2+ 3 harmonic? Determine a harmonic
conjugate of u in the cases where it is harmonic.
Exercise* 11.16.2. Prove that Laplace's equation can be written in polar
coordinates as
II.16. Harmonic Conjugates 25

Exercise 11.16.3. Find all real-valued functions h, defined and of class

C 2 on the positive real line, such that the function u(x, y) = h(x 2 + y 2 ) is
harmonic.
Exercise 11.16.4. Prove that, if u is a real-valued harmonic function in
an open disk D, then any two harmonic conjugates of u in D differ by a
constant.
Exercise 11.16.5. Suppose that u is a real-valued harmonic function in an
open disk D, and suppose that u 2 is also harmonic. Prove that u is constant.
Exercise 11.16.6. Prove that if the harmonic function v is a harmonic
conjugate of the harmonic function u, then the functions uv and u 2 - v 2 are
both harmonic.
Exercise 11.16. 7. Prove (assuming equality of second-order mixed partial
derivatives) that
cP 1 ( 82 82 )
8z8z = 4 8x2 + 8y2 .

Thus, Laplace's equation can be written as ::£z = 0.

Exercise 11.16.8. Prove that if u is a real-valued harmonic function then

the function ~~ is holomorphic.
 Chapter III

Linear-Fractional
Transformations

A linear-fractional transformation is a nonconstant function </> of the form

az + b
</>(z) = ---d, where a, b, c, d are complex constants. These are the
CZ+
simplest nonconstant rational functions. Linear-fractional transformations
have many interesting algebraic and geometric properties and they surface
in many different aspects of complex function theory.
It is instructive to introduce linear-fractional transformations from a
"projective" viewpoint.

111.1. Complex projective space

By C 2 we shall mean the complex vector space of two-dimensional complex
column vectors:

C 2 = { ( ~~ ) : z1 , z2 E C}.
We introduce an equivalence relation between nonzero vectors in C 2 by
declaring that two vectors shall be equivalent if they are linearly dependent,
in other words, if they are (complex) scalar multiples of each other. The
equivalence classes are thus the one-dimensional subspaces of C 2 ' with the
origin deleted. The equivalence class determined by the pair

will be denoted by [z1, z2]. The space of equivalence classes is called one-
dimensional complex projective space and denoted by CP 1 .

- 27
28 III. Linear-Fractional Transformations

There is a natural one-to-one map of CP 1 onto C, namely, the map

z1
[z1, z2] f-> - ,
z2
the right side being interpreted as oo in case z2 = 0. With this we have a
new way of thinking about the extended complex plane.

111.2. Linear-fractional transformations

A nonsingular linear transformation of C 2 to itself maps each one-dimensional
subspace of C 2 onto a one-dimensional subspace. It thus induces a map from
CP 1 to itself; that map is easily seen to be a bijection (one-to-one and onto).
In view of the correspondence between CP 1 and C, therefore, each nonsin-
gular linear transformation of C 2 determines a bijection of c with itself.
The linear transformations of C 2 are in one-to-one correspondence with
the two-by-two complex matrices, each matrix corresponding to the trans-
formation it induces via left multiplication. Suppose

M=(~ ~)
is such a matrix, and suppose that M is nonsingular, in other words, that
det M = ad - be -/: 0. Let ¢ denote the bijection of C that M induces
according to the recipe above. If z is a point of C, then

M ( ~ ) = ( ;; : ~) '

from which we conclude that ¢(z) = az +db. Thus,¢ is a linear-fractional

CZ+
transformation. (We shall always regard these transformations as acting on
- a
C. If c = 0 one has ¢( oo) = oo, and otherwise one has ¢( oo) = and
c

Certain conclusions are immediate. First, the composite of two linear-

fractional transformations is again a linear-fractional transformation. More
precisely, if ¢1 and ¢2 are the linear-fractional transformations induced by
the matrices M 1 and M2, respectively, then ¢1 o ¢2 is the linear-fractional
transformation induced by M 1M 2. Since the identity transformation of C is
the linear-fractional transformation induced by the identity matrix, it follows
that the inverse of a linear-fractional transformation is a linear-fractional
transformation: if¢ is the linear-fractional transformation induced by the
matrix M, then ¢- 1 is induced by M- 1.
Thus, the linear-fractional transformations form a group under compo-
sition. (If the concept of a group is new to you, see Appendix 3.) Under
III.5. Three-fold transitivity 29

the map that sends each matrix to its induced linear fractional transfor-
mation, this group is the homomorphic image of the group GL(2, C), the
general linear group for dimension two (alias the group of nonsingular two-
by-two complex matrices). The kernel of the homomorphism consists of the
nonzero scalar multiples of h, the two-by-two identity matrix. The group
of linear-fractional transformations is thus isomorphic to the quotient group
GL(2, C)/(C\{O} )h.

111.3. Conformality
If</> is a linear-fractional transformation and 'ljJ is its inverse, then the chain
rule gives 'l/J' (</>( z) )</>' (z) = 1 (provided </>( z) i- oc), implying that </>' is never
0. The same conclusion is easily reached directly, for if </>(z) = az + b, then
CZ +d
a short calculation gives
</>' (z)
ad - be .
=
(cz + d) 2
Hence, a linear-fractional transformation is a conformal map.

111.4. Fixed points

A linear-fractional transformation, except for the identity transformation,
has one or two fixed points.
. . az + b . .
For the proof, suppose the transformation is </>( z) = --d. It is evident
CZ+
that oo is a fixed point if and only if c = 0. The finite fixed points are the
solutions for z of the equation cz 2 + (d- a)z - b = 0. If c i- 0 that equation
has either two distinct roots or one repeated root (given by the standard
quadratic formula). If c = 0, then there is one root in cased i- a and none in
case d =a (since in the last case b-=/:- 0). In all cases, therefore, the number
of fixed points is one or two.

111.5. Three-fold transitivity

If z1, z2, z3 are three distinct points of C, and w1, w2, w3 are three distinct
points of C, then there is a unique linear-fractional transformation </> such
that </>(zj) = Wj, j = 1, 2, 3.
To establish the uniqueness we note that if the two linear-fractional
transformations </> and 'ljJ both have the required property, then 'ljJ- 1 o </> is
a linear-fractional transformation with three fixed points; hence it is the
identity, by III.4, implying that </> = 'l/J. This settles uniqueness.
As for existence, because the linear-fractional transformations form a
group, it will suffice to prove that </> exists for some particular choice of w1,
30 III. Linear-Fractional Transformations

w2, w3. (See Appendix 3 for an explanation of this assertion.) We consider

the case w1 = oo, w2 = 0, W3 = 1. It is then a simple matter to write down
the required ¢:

(z-z2)(z1 -z3)
(z-z1 )(z2-z3)
if z1, z2, z3 are all finite,
z-z2
Z3-Z2
if z1 = oo,
¢(z) Z3-Z1
z-z1 if z2 = oo,
z-z2
z-z1 if Z3 = 00.

This settles existence.

Exercise 111.5.1. Exhibit the linear-fractional transformation that maps

0, 1, oo to 1, oo, -i, respectively.

Exercise 111.5.2. Given four distinct points z1, z2, z3, z4 in C, their cross
ratio, which is denoted by (z1, z2; z3, z4), is defined to be the image of z4
under the linear-fractional transformation that sends z1, z2, z3 to oo, 0,
1, respectively. Prove that if ¢ is a linear-fractional transformation then
(¢(z1), ¢(z2); ¢(z3), ¢(z4)) = (z1, z2; z3, z4).

111.6. Factorization
It was observed in Section II.9 that every linear function (i.e., linear-fractional
transformation with constant denominator) can be written as the compos-
ite of a homothetic map, a rotation, and a translation. Those three kinds
1
of transformations, plus the transformation z f-t - , called the inversion,
z
are enough to build the general linear-fractional transformation. Indeed, if
az + b .
¢(z) = - - , and c =f- 0, then¢ can be rewritten as
CZ +d

b- ad/c a
</>(z) ----+
CZ+ d C'

from which one sees that one can obtain ¢ by applying a linear function
followed by the inversion followed by a translation. Because of the natural
isomorphism between GL(2, C)/(C\ {O} )hand the group of linear fractional
transformations, it follows that every matrix in GL(2, C) is a scalar multiple
III.8. Preservation of clircles 31

of a product of matrices of the following four kinds:

(~ ~)' k>O

(~ ~)' I-XI= 1

( 1 b
0 1
), bE C

( 0 1
1 0 )·
Exercise 111.6.1. Prove, either directly or by using the last result, that
every matrix in G L(2, C) is a product of matrices of the preceding four
kinds.

Exercise 111.6.2. Two linear-fractional transformations </>1 and </>2 are said
to be conjugate if there is a linear-fractional transformation 7./J such that
¢ 2 = 1jJ o ¢ 1 o 7./J- 1 . Prove all translations, except for translation by 0, are
mutually conjugate.

Exercise 111.6.3. Prove a linear-fractional transformation with only one

fixed point is conjugate to a translation.
1
Exercise 111.6.4. Prove the linear-fractional transformations </>1 ( z)
z
and </>2 ( z) = - z are conjugate.

III. 7. Clircles
By a clircle we shall mean a circle in C, or a straight line in C together with
the point at oo. According to Exercise III.8.1 below, clircles correspond
under the stereographic projection to circles on S 2 .

111.8. Preservation of clircles

A linear-fractional transformation maps clircles onto clircles.
This is obviously true of homothetic maps, rotations, and translations.
It will therefore suffice, by III.5, to prove it is true of the inversion. The
equation of the general clircle, or rather, the part of it in C, has the form

a Iz I2 + ,B Re z + ')' Im z + <5 = 0,
32 III. Linear-Fractional Transformations

1
where a, (3, "(, 5 are real constants. Under the map z f-----7 - , the locus of
z
the preceding equation is sent to the locus of the equation
a (3 Re z 'Y Im z
lzl2 + TzT2 - TzT2 + 5 = 0,
which can be rewritten as
5lzl 2 + (3Rez - 1Imz +a= 0,
again the equation of a clircle. This establishes the proposition.
Exercise III.8.1. Prove that the stereographic projection maps a circle on
S 2 onto a clircle, and vice versa.
Exercise III.8.2. Prove that the four distinct points z1, z2, z3, z4 of C lie
on a clircle if and only if the cross ratio (z1, z2; z3, z4) is real.

111.9. Analyzing a linear-fractional transformation-an

example
A clircle in C is determined by any three distinct points it contains. Thus,
by III.7, in order to find the image under a given linear fractional trans-
formation of a given clircle, it suffices to find the images of three points
on that clircle. One can obtain a good geometric picture of any linear-
fractional transformation on the basis of this property and two others: a
linear-fractional transformation is conformal, and, being a continuous func-
tion, it maps connected sets onto connected sets. This remark will be illus-
trated with the function ¢(z) = z + ~. (The reader is advised to supplement
z-i
the following discussion with appropriate sketches.)
First, we have immediately ¢(-i) = 0, ¢(0) = -1, ¢(i) = oo. The clircle
determined by -i, 0, i is the extended imaginary axis, and the clircle deter-
mined by 0, -1, oo is the extended real axis. Hence ¢ maps the extended
imaginary axis onto the extended real axis.
The extended imaginary axis separates C into two half-planes, the right
half-plane, Re z > 0, which we denote by P+, and the left half-plane, Re z <
0, which we denote by P_. Similarly, the extended real axis separates C into
the upper half-plane, Imz > 0, which we denote by H+, and the lower half-
plane, Imz < 0, which we denote by H_. Because¢ is a bijection of C
sending the extended imaginary axis onto the extended real axis, it must
map P+ UP_ onto H+ UH_:
¢(P+) U ¢(P-) = H+ U H_.
Because the sets P+ and P_ are connected, so are their images, ¢(P+) and
¢ ( P _). The set ¢ ( P +) thus cannot intersect both H + and H _, for if it did,
III.9. Analyzing a linear-fractional transformation-an example 33

the decomposition
¢(P+) = [¢(P+) nH+] u [¢(P+) nH_J
would be a separation of ¢ ( P+) into the union of two nonempty sets each
disjoint from the closure of the other. Hence, either ¢(P+) c H+ or
¢(P+) C H_. To determine which inclusion is correct, it suffices to note
that ¢(1) = 1 + ~ = i, which is in H+. Thus ¢(P+) c H+. By the same
1-i
reasoning, either ¢(P-) C H+ or ¢(P-) C H_. Because¢ maps C onto
C, it cannot map both P+ and P_ into H+, so it must be that ¢(P-) C H_.
Moreover, the inclusions ¢ (P+) c H + and ¢ (P _) c H _ cannot be proper,
again because ¢(C) = C. Hence ¢(P+) = H+ and ¢(P-) = H_.
The unit circle, Iz I = 1, is sent by ¢ onto a clircle that contains 0
(= ¢(-i)) and oo (= ¢(i)). By the conformality of¢, the image of the unit
circle intersects the real axis orthogonally at the origin (= ¢( -i)). Hence ¢
sends the unit circle onto the extended imaginary axis.
Let D denote the interior of the unit circle, the set lzl < 1, and E the
extended exterior, the set 1 < lzl : : ; oo. A repetition of the reasoning above
shows that either ¢(D) = P+ or ¢(D) = P_. Since ¢(0) = -1, the second
equality is correct. Reasoning again as above, we find that ¢(E) = P+.
The image under ¢ of the extended real axis is a clircle that contains
the points i (= ¢(1)), -1 (= ¢(0)) and 1 (= ¢(00)). Hence ¢ maps the
extended real axis onto the unit circle. Another repetition of the reasoning
above shows that ¢(H+) = E and ¢(H-) = D.
We can enhance the picture already obtained by determining the image
under ¢ of the coordinate grid (the families of vertical and horizontal lines).
An extended vertical line other than the imaginary axis must be sent by
¢ to a circle containing the point 1 ( = ¢( oo)) and orthogonal to the unit
circle (the image of the extended real axis). The family of extended vertical
lines is thus sent to the family of circles tangent to the real axis at the
point 1 (plus the real axis itself). By the conformality of ¢, the family of
extended horizontal lines must be sent to the family of clircles orthogonal
to the preceding ones. Thus, extended horizontal lines are sent to circles
through the point 1 orthogonal to the real axis, except for the extended
horizontal line through the point i, which is sent to the extended vertical
line through the point 1.
Exercise 111.9.1. Find the image of the half-plane Re z > 0 under the
linear-fractional transformation that maps 0, i, -i to 1, -1, 0, respectively.
Exercise 111.9.2. Find the images of the disk lzl < 1 and the half-plane
Re z > 0 under the linear-fractional transformation that maps oo to 1 and
has i and -i as fixed points.
34 III. Linear-Fractional Transformations

Exercise 111.9.3. Prove that a linear-fractional transformation maps the

half-plane Im z > 0 onto itself if and only if it is induced by a matrix with
real entries whose determinant is 1.
Exercise* 111.9.4. Prove that the linear-fractional transformations map-
A(z - zo)
ping the disk lzl < 1 onto itself are those of the form ¢(z)
zoz - 1 '
where lzol < 1 and IAI = 1.
Exercise 111.9.5. Prove that the linear-fractional transformations mapping
the disk lzl < 1 onto itself are those induced by matrices of the form

(~ ~)
with Jal 2 - lbl 2 = 1.
Exercise 111.9.6. Find the image under the linear-fractional transforma-
tion ¢(z) = z - 1 of the intersection of the disk izl < 1 with the half-plane
z+l
lmz > 0.
Exercise 111.9. 7. For the function

f(z) = (~)
z-l
2

(defined to equal 1 at z = oo and oo at z = 1), find the images of the

following sets:
(a) The extended real axis.
(b) The extended imaginary axis.
(c) The half-plane Rez > 0.
Exercise 111.9.8. Find all linear-fractional transformations that fix the
points 1 and -1. Is the group of those transformations abelian? Is it
isomorphic to any group with which you are familiar?
 Chapter IV

Elementary Functions

By the elementary functions of calculus one ordinarily means, in addition

to the rational functions, those on the following list:
1. The exponential function, ex.
2. The logarithm function, lnx.
3. The trigonometric and inverse trigonometric functions, sin x, arcsin x,
etc.
4. All functions obtainable in finitely many steps from those already
mentioned by means of the operations of addition, subtraction, multiplica-
tion, division, and composition. (An example is ~ = e~ ln(l-x 2 ).)
We have already encountered complex rational functions; they are holo-
morphic and they generalize real rational functions in an obvious way. We
shall now see that the other basic elementary functions also have complex
holomorphic versions.

IV.1. Definition of ez

To motivate the definition we follow L. Euler (1707-1783) and perform a sim-

ple manipulation with power series, disregarding questions of convergence.
One expects that, if there is a natural way of defining ez, it should maintain
the law of exponents: ez 1 +z2 = ez 1 ez 2 • Thus, writing z = x + iy as usual,
we should have ez = exeiY. As we know what ex means from real-variable
theory, it only remains to make sense of eiY. Euler's idea was to write out eiy
. x 00 xn
as a power series, and to fiddle a bit with that series. Smee e = Ln=0-1 ,
n.
(iy)n
we expect to have eiY = 2=~= 0 --1 -. Summing separately the even and odd
n.
-35
36 IV. Elementary Functions

terms of the last series, we obtain

iy - ~i2ky2k ~i2k+ly2k+l
e - ~ (2k) ! + ~ (2k + 1) !
k=O k=O
00 (-l)ky2k . 00 (-l)ky2k+l
= 2= (2k),
k=O
+ z 2=
k=O
(2k + 1),
We recognize that the right side expresses in power series form the function
cos y + i sin y.
The considerations above suggest how we should define ez, namely, by
setting
ez = ex (cos y + i sin y).
We adopt this as our definition. (The formal manipulations with power
series that led to the definition can be made rigorous, as will be clear in the
next chapter, where power series are studied systematically.)
As in real-variable theory, we write exp z instead of ez when convenient.
Exercise IV.1.1. Find the real and imaginary parts of the function exp(ez).

IV.2. Law of Exponents

The expression defining ez is in polar form: we have Iez I = ex and arg ez = y.
The law of exponents, ez 1 ez 2 = ez 1 +z 2 , follows immediately from the rule for
multiplying complex numbers in polar form.
Note that, having defined the exponential function, we can rewrite the
polar form, r(cos() + isinO), of a complex number, as reiO.

IV .3. ez is holomorphic
We prove that ez is holomorphic by verifying that its real and imaginary
parts satisfy the Cauchy-Riemann equations. This is a simple calculation.
We have Re ez = ex cosy and Im ez = ex sin y. Since

:x(excosy) =ex cosy, :y(excosy) = -exsiny

the Cauchy-Riemann equations hold. By II. 7 we can conclude that ez is

holomorphic, and from II.5 we see that its derivative is ez.
Exercise* IV.3.1. Suppose the function f is holomorphic in C and satisfies
f' = f. Prove that f is a constant times ez. (Suggestion: Consider the
function e-z f(z).)
IV.5. ez as a map 37

Exercise IV.3.2. What is the most general holomorphic function f in C

that satisfies f' = cf, where c is a constant?

IV .4. Periodicity
Because the functions cosy and sin y are periodic with period 27r, we see that
the function ez is periodic with period 27ri: ez+ 27ri = ez. In particular, since
e0 = 1, we have e 27ri = 1, a remarkable identity discovered by Euler. It is a
favorite of mathomaniacs because it contains in the same short expression
the numbers 1, 2, e, 7r, and i, arguably the five most important constants in
mathematics.
Exercise IV .4.1. Prove that if a and b are complex numbers such that
ea = ea+b, then b is an integer multiple of 27ri.

IV.5. ez as a rnap
Since Iez I = ex, the map z r--> ez sends the vertical line x = xo onto the circle
with center 0 and radius exo. As a point moves along the vertical line in the
upward direction, its image point on the circle moves in the counterclockwise
direction, making one complete circuit each time its pre-image traverses an
interval of length 27r.
Since arg ez = y, the map z r--> ez maps the horizontal line y = Yo
onto the open ray emanating from the origin that makes the angle Yo with
the positive real axis. As a point moves toward the right on the horizontal
line, its image point on the ray moves away from the origin.
Thus, the map z r--> ez sends the coordinate grid for rectangular coor-
dinates to the coordinate grid for polar coordinates. Its range is C\ { 0}, and
every point in the range has infinitely many pre-images, any two of which
differ by an integer multiple of 27ri. The restriction of the map z r--> ez
to an open horizontal strip of width at most 27r is one-to-one, and the map
sends the strip to the interior of an angle with vertex at the origin. For
example, the strip
-7r 7r
2<y<2
is sent onto the right half-plane.
Exercise IV.5.1. Describe the images under the map z r--> ez of the line
y = x and of the strip
7r 7r
x - 2 < y < x + 2·

Exercise IV.5.2. Describe the curves Iii =constant and arg f =constant
for the function
J(z) = exp(z 2 ).
38 IV. Elementary Functions

Exercise IV.5.3. Repeat Exercise IV.5.2 for the function

f (z) = exp ( z + 1 ) .
z-1

IV.6. Hyperbolic functions

The hyperbolic functions are defined in terms of the exponential function
just as in the theory of functions of a real variable:
ez + e-z ez - e-z
coshz = 2 sinhz = 2
sinhz coshz
tanh z = --h- , coth z = -.-h- ,
cos z sm z
1 1
sechz = --h-, cschz = ~h .
COS Z Sln Z
Because the function ez is holomorphic, so are the hyperbolic functions at
every point where they are defined, which is everywhere for the first two,
and everywhere the denominator in the defining expression does not vanish
for the other four. The expressions for the derivatives of the hyperbolic
functions and various identities involving the functions are easily obtained
from properties of the exponential function (Exercises IV.6.1 and IV.6.2
below).
Exercise IV.6.1. Derive the expressions for the derivatives of the hyper-
bolic functions.
Exercise IV.6.2. Derive the identities
cosh(z1 + z2) = cosh z1 cosh z2 + sinh z1 sinh z2,
sinh(z1 + z2) = sinh z1 cosh z2 + cosh z1 sinh z2.
Exercise IV.6.3. Suppose the holomorphic function fin Chas a holomor-
phic derivative and satisfies the differential equation f" = f. Prove that f
has the form f (z) = a cosh z + b sinh z, where a and b are constants.

IV. 7. Zeros of coshz and sinhz.

The zeros of the function sinh z are the solutions for z of the equation ez =
e-z, which can be rewritten as e 2 z = 1. Since e 2 z = 1 only when 2z is an
integer multiple of 27fi, we conclude that the zeros of sinh z are the numbers
n1fi with n an integer.
Similarly, the zeros of cosh z are the solutions for z of the equation
ez = -e-z which can be written as e 2 z = -1 or as e 2 z-n:i = 1 (since
' '
en:i = -1). It follows that the zeros of coshz are the numbers ( n + ~)7ri
with n an integer.
IV.8. Trigonometric functions 39

IV .8. Trigonometric functions

eiy + e-iy eiY - e-iy
If y is real then we have cosy = 2 and sin y =
2i We define
the functions cos z and sin z by replacing y by z in these expressions. Our
definitions are thus
eiz _ e-iz
COSZ= - - - - sinz= - - - -
2 2i
The other trigonometric functions are defined in terms of cosine and sine
just as in real variable theory:
sinz cosz
tanz = - - , cotz = -.- ,
cosz smz
1 1
secz = --, cscz = -.-.
cosz smz
These functions are holomorphic, the first two everywhere, the other four
everywhere their denominators are nonzero.
In previous encounters with trigonometric and hyperbolic functions, the
reader has perhaps taken note of the formal similarities between the two
function classes. The addition formula for the hyperbolic sine, for example,
is the same as that for the sine, and the addition formula for the hyperbolic
cosine differs from that for the cosine only in a minus sign. The complex
viewpoint explains these similarities, because from the basic definitions we
have the relations
. sinh iz
cos z = cosh iz, sm z = --.-.
i
Roughly speaking, one obtains the trigonometric functions by composing
the hyperbolic functions with a rotation of the complex plane through 90
degrees.
The basic properties of the trigonometric functions are easily deduced
from those of the exponential function and the hyperbolic functions. For
example, from what we already know about the latter functions, we see that
the functions cos z and sin z are periodic with period 27r, that the zeros of
sin z are the numbers n7r with n an integer, and that the zeros of cos z are
the numbers ( n + ~) 7r with n an integer. In particular, in extending the
functions sine and cosine from R to C we have not introduced additional
zeros.
Further properties are left for the reader to establish.
Exercise IV.8.1. Find all the roots of the equation cos z = 2.
Exercise IV.8.2. Establish the identities
cos z = cos x cosh y - i sin x sinh y,
40 IV. Elementary Functions

sin z = sin x cosh y + i cos x sinh y,

Icos zl 2 = cos 2 x + sinh 2 y,
I sin zl 2 = sin 2 x + sinh2 y.
Exercise IV.8.3. Describe the images of the lines Re z = constant and
Im z = constant under the map z f----t cos z.

IV.9. Logarithms
If a is a complex number, then by a logarithm of a one means any complex
number b such that eb = a. If a = 0 there is no such b, but otherwise there
are infinitely many, any two of which differ by an integer multiple of 27!' i.
For example, the logarithms of the number 1 are the numbers 2n7l'i with n an
integer, the logarithms of the number -1 are the numbers (2n + 1 )7ri with n
an integer, and the logarithms of the number i are the numbers ( 2n + ~) 7l'i
with n an integer.
Since lebl = eReb and arg eb = Im b, the number b is a logarithm of
the nonzero number a if and only if eReb = lal and Im b = arg a. Thus,
letting ln denote the natural logarithm function of calculus, we see that
the logarithms of a are the numbers ln lal + i arg a. If b is a logarithm of
a, we shall write b = log a. The particular logarithm of a whose imaginary
part is in the interval (-7!', 7l'], corresponding to the principal value of arg
a, will be denoted by Log a; i.e., Log a = ln lal + i Arg a.
Exercise IV.9.1. Find all values of cosh (log 2).
Exercise IV.9.2. Find all values of log (log i).
Exercise IV.9.3. In what sense is it true that log a1a2 = log a1 + log a2
for complex numbers a1 and a2?

IV.10. Branches of arg z and log z.

Let G be an open connected subset of C not containing the origin. By a
branch of arg z in G is meant a continuous function a in G such that, for
each z in G, the value a(z) is a value of arg z. By a branch of log z in G is
meant a continuous function l in G such that, for each z in G, the value l(z)
is a logarithm of z. For a simple example, we can take for G the region one
obtains by removing from C the interval (-oo, 0] on the real axis; in this
region the functions a(z) = Arg z and l(z) =Log z are branches of arg z
and log z, respectively.
As we shall see, depending on the shape of G, there may or may not exist
branches of arg z and log z in G. However, if there is a branch a of arg z in
G then there is a branch of log z, namely, the function l(z) = ln lzl + ia(z).
IV.11. log z as a holomorphic function 41

Similarly, if there is a branch l of log z in G then there is a branch of arg z,

namely, the function a (z) = Im l (z).
If a is a branch of arg z in G then a + 21fn is another branch for any
integer n. On the other hand, if a1 and a2 are two branches of arg z in G,
then the function ai - a 2 is continuous in G and takes only integer values,
27f
so it must be constant (because by assumption G is connected). Thus, if
there is a branch of arg z in G, then any two branches differ by a constant
integer multiple of 27r. Similarly, if there is a branch l of log z in G, then
l + 21fni is another branch for every integer n, and every branch has this
form.
An example of a region that carries no branch of arg z is given in the
exercise at the end of the section. On the other hand, if G is any open disk
that excludes the origin, then G does carry such a branch. In fact, suppose
Oo is a particular value of the argument of the center of the disk. Then the
disk is contained in the half-plane { rei(J : r > 0, Oo - ~ < () < Oo + ~}, and
we can obtain the desired branch a by letting a(z), for z in G, equal the
value of arg z in the interval ( Oo - ~, Oo + ~).
Exercise* IV.10.1. Prove that there is no branch of arg z in the region
0 < lzl < l.

IV .11. log z as a holomorphic function

If l is a branch oflog z in the open connected set G, then l is holomorphic,
and l'(z) = ~-
z
We shall verify that l is holomorphic by checking that l satisfies the
Cauchy-Riemann equations in polar coordinates, whose derivation is Exer-
cise II.6.3. Let u = Re l and v = Im l. The Cauchy-Riemann equations in
polar coordinates are

Since l is a branch of log z we have u(z) = ln lzl and v(z) = arg z, from
which it is immediate that
au 1
,
au av av
or r f)() = 0 = or , f)() = l.

The Cauchy-Riemann equations are thus satisfied, and we can conclude

by II.7 that l is holomorphic. Moreover, by II.6 we have l =
I au + i ax.
ax av
42 IV. Elementary Functions

Using the chain rule, we obtain

l'(z) = x-iy 1 1
r2 x +iy z

IV.12. Logarithms of holomorphic functions

Let G be an open connected subset of C, and let f be a holomorphic function
in G that does not assume the value 0. By a branch of log f in G one means
a continuous function g in G such that f = e9, in other words, such that,
for each z in G, the value g(z) is a logarithm of f(z). The special case of
the function f(z) = z is the one discussed in the three preceding sections.
Here in general, just as in that special case, one sees that if there is a branch
of log f in G, then one obtains the most general such branch by adding a
constant integer multiple of 27ri to any particular one.
If it happens that the range off is contained in a region in which there
is a branch l of log z, then the composite function lo f is a branch of log f;
moreover, being a composite of two holomorphic functions it is holomorphic,
and by the chain rule its derivative equals 1; .
Whether or not there is a branch of log f in G, such branches exist
locally, in the following sense. Fix a point zo in G. Since by assumption
f (zo) =I- 0, there is by the continuity of f an open disk D with center zo
such that f (D) is contained in an open disk D' that excludes the origin.
Then, as seen in Section IV.10, there is a branch l of log z in D', and by
composing l with f we obtain a branch of log f in D. In case there happens
to be a branch g of log f in all of G, the functions g and l o f differ in D by a
constant integer multiple of 27ri. In fact, we can arrange to have g = lo f by
choosing for l the particular branch of log z in D' satisfying l ( f (zo)) = g ( zo).
Since l o f is a holomorphic function with derivative 1;, we can draw the
following conclusion: If there is a branch of log f in G, then any such

branch is a holomorphic function whose derivative is 1; . Whether or not

there is a branch of log f in G, one refers to the function 1; as the logarithmic
derivative of f.
Although, as just seen, a branch of log f in G, if it exists, is locally
the composite of f with a branch of log z, it may not be globally such a
composite. A simple example is provided by the function f(z) = ez in C.
The function g( z) = z is a branch of log f, yet the range of f is C\ {0}, in
which there is no branch of log z (according to Exercise IV.10.1). Whether,
for a particular f and G, there exists a branch of log f in G, is a subtle
IV.14. Inverses of holomorphic functions 43

question which we shall only be able to deal with adequately much later, in
Chapter X.
Exercise IV.12.1. Prove that the logarithmic derivative of the product of
two holomorphic functions equals the sum of their logarithmic derivatives.

IV.13. Roots

Let G be a connected open subset of C and let J be a holomorphic function

in G that does not assume the value 0. Let n be a positive integer. By a
1
branch of Jn in G one means a continuous function h in G such that, for
each z in G, the value h(z) is an n-th root of J(z). (The presumption that
J omits the value 0 is explained by Exercise IV.13.1 below.) The discussion
I
of branches of Jn parallels that of branches of log J and will only be briefly
I
sketched. The basic facts are as follows. If there is a branch of Jn in G then
there are precisely n such branches, each obtainable from any particular one
through multiplication by an n-th root of 1. If there is a branch g of log J
in G, then the function e; is a branch of J ~ in G. Any branch of J ~ can
be obtained locally in the preceding way from a local branch of log J. If h
is a branch of J~ then h is holomorphic and ~ = :>
1
Exercise* IV.13.1. Prove that for n > 1 there is no branch of zn in the
region 0 < lzl < 1.
I
Exercise IV.13.2. Prove that if his a branch of Jn then his holomorphic
h' J'
and h = nj"
Exercise IV.13.3. Let G be the open set one obtains by removing from
C the interval [-1, 1] on the real axis. Prove that there is a branch of the
function {Z+T in G. (Suggestion: What is the image of Gunder the map
v~
z z + 1 ?)
f---t
z-1
Exercise IV.13.4. Let G be as in Exercise IV.13.3. Prove that there is a
branch of the function ~ in G.

IV.14. Inverses of holomorphic functions

The use of branches gives a way of dealing with inverses of functions that
are not one-to-one. This device appears already in the theory of functions of
a real variable. For example, the function J(x) = x 2 on R has range [O,oo),
and each point in its range, except for the origin, has two pre-images. The
"inverse function" has two branches, namely, the function g(x) = y'X (the
44 IV. Elementary Functions

positive square root), and the function -g. The function g is the bona fide
inverse of the restriction of f to the interval [0, oo), while - g is the inverse
of the restriction off to the interval (-oo, O]; both of those restrictions off
are one-to-one functions. Similarly, the sine function on R has range [-1,1],
and each point in its range has infinitely many pre-images. The so-called
inverse sine function, the function arcsin, is the inverse of the restriction of
the sine function to the interval [- ~, ~] ; it is only one of infinitely many
branches of the inverse of the sine function.
Suppose f is a holomorphic function in the open connected subset G of
C. If f is one-to-one, or, in the parlance of complex analysis, "univalent,"
then the inverse of f is the function g defined in the region f (G) by the
rule g(f(z)) = z. It turns out, as we shall see later (in Chapter X), that
if f is univalent then f(G) is an open set; in fact, that much is true under
the weaker assumption that f is nonconstant. Moreover, its derivative, f',
is never 0, and its inverse function is continuous. Granted these facts, one
easily sees, by the same reasoning one uses in calculus, that the inverse
function is holomorphic, its derivative at the point f (z) of f (G) being equal
1 .
to f'(z) (Exercise IV.14.l below).
If the holomorphic function f, defined in the open connected set G, is
not univalent, then it does not possess an inverse function in the strict sense
of the term. Nevertheless, it is still possible to speak of branches of the
inverse of f. All one needs is an open connected subset H of G on which f
is univalent; the inverse of the restriction of f to H is then called a branch
of the inverse of f. Starting with the function J(z) = ez one obtains in
this way what we called the branches of log z; starting with the function
1
f (z) = zn one obtains the branches of z:n.
Exercise* IV.14.1. Let f be a univalent holomorphic function in the open
connected set G, and let g be the inverse function. Assume that f (G) is
open, that g is continuous, and that f' is never 0. Prove g is holomorphic.

Exercise IV.14.2. Prove that a branch of the inverse of a holomorphic

function is always univalent.

IV.15. Inverse trigonometric functions

The preceding discussion, applied to any of the trigonometric functions,

explains what one means by the branches of the corresponding inverse func-
tion. Because of the relation between the trigonometric functions and the
exponential function, those branches can be related to the logarithm func-
tion.
IV.17. Analytic continuation and Riemann surfaces 45

To illustrate we consider the cosine function; the branches of its inverse

are referred to as the branches of arccos z. Let us fix a point z in the domain
of such a branch and ask what possible values the branch can have at z. If
w is such a value, then w is a solution of the equation cos w = z; in other
eiw + e-iw
words, = z, which can be rewritten as
2
e 2 iw - 2zeiw +1= 0.
This is a quadratic equation in eiw. The quadratic formula gives us two
roots (or one repeated one), namely eiw = z + ~ (recall that the
square root has two possible values if z #- ±1). The possible values of ware
therefore expressed by w = t log ( z + ~). One sees in this way that

the branches of arccos z are just the branches of ~ log ( z + ~).

Exercise IV.15.1. Find a relation between the branches of arctanz and
the logarithm function.

IV.16. Powers
If a and c are complex numbers, with a #- 0, then by the values of ac one
means the values of ecloga. For example, the values of li are the numbers
ei( 2 7rni) in other words the numbers e- 2 7rn with n = 0 ±1 ±2 · · · If c
' ' ' ' ' ' .
is an integer or the reciprocal of an integer, the preceding definition reduces
to the usual one.
On the basis of the preceding definition one can give a meaning, for any
complex number c, to the branches of zc, or, more generally, to the branches
of Jc for any holomorphic function f that omits the value 0. As this involves
no ideas not encountered earlier, the details will not be spelled out.
Exercise IV.16.1. Find all the values of (1 + i)i.
Exercise IV.16.2. Let a be a complex number of unit modulus and can
irrational real number. Prove that the values of ac form a dense subset of
the unit circle.
Exercise IV.16.3. Prove that if f is a branch of zc in an open set not
containing 0, then f is holomorphic and f' is a branch of czc-l _

IV .17. Analytic continuation and Riemann surfaces

The notion of a branch has enabled us to make sense of so-called multiple-
valued functions like log z and vz.
What we have done so far in this
direction, although it will be adequate for our purposes, is really only the
first step in a more profound treatment of multiple-valued functions. This
46 IV. Elementary Functions

more profound treatment is generally not included in an introductory course;

it relies on results we shall only develop later on. Its main ideas will be
very vaguely sketched, with emphasis on the simple example viz, in this
concluding section of Chapter IV.
A key observation is that the different branches of viz are related to one
another through a process called analytic continuation. To illustrate, con-
sider the four open disks Do, Di, D2, D3, each of unit radius, with respective
centers 1, i, -1, -i. In Do there are two branches of viz, which we call f
and g, one of which, say f, takes the value 1 at z = 1, and the other the
value -1 at z = 1. In Di there are also two branches of viz, and one of
them, which we call Ji, agrees with f in the overlap Don Di. The reader
will easily check that Ji (i) = e ¥. Similarly, in D2 there are two branches
of viz, one of which, say f2, agrees with Ji in the overlap Di n D2. One
then has h (-1) = i. We continue in this way for two more steps. The
branch of viz in D3 that agrees with h in D2 n D3 will be called h; it
31Ti
satisfies h(-i) = eT. Finally, the branch of viz in Do that agrees with h
in D3 n Do is g (not f). The two branches f and g of viz that live in Do
are thus linked by the successive branches Ji, h, h; one says that g arises
from f by analytic continuation. A precise definition of the latter term will
not be attempted-it will only be mentioned that any two branches of viz
are linked in this way, and that through analytic continuation of a branch
of viz only other branches of viz can arise.
It thus appears that the totality of the branches of viz form some kind
of organic whole, and it seems sensible to conceive of viz as the collection of
all of its branches. Except for certain technical modifications, this is what
is actually done. When viz is interpreted this way, it acquires a natural
topological structure that makes it into a surface. That surface can be
visualized as follows.
We consider two copies of C, each slit along the negative real axis; we
imagine them as stacked one above the other in real three-dimensional space.
Let A denote the one on top and B the bottom one. (It might help to
associate A with one of the branches of viz in C\ (-oo, 0] and B with the
other branch.) We join A and B, making a single surface, by identifying the
top edge of the slit in A with the bottom edge of the slit in B, and vice versa
(see Figure 3). As there are to be no additional identifications, the resulting
surface cannot actually be embedded in real three-dimensional space, but
by cheating a little we can imagine it as so embedded. Set-theoretically
the construction poses no difficulty, although we shall not worry here about
the technical details. Let S denote the resulting surface. From the way
we constructed S, out of two copies of C, we obtain a natural map from S
to C. The map is two-to-one but locally one-to-one, so by using it we can
IV.17. Analytic continuation and Riemann surfaces 47

introduce local complex coordinates in S, thereby obtaining a way to define

what it means for a complex-valued function on S to be holomorphic, or for
a function of C into S to be holomorphic. (There might seem to be trouble
at the origin, but the construction is performed in such a way that the origin
is deleted from S.) We can reinterpret the function z 2 in a natural way as a
univalent function from C\{O} onto S; that function then has a well-defined
inverse, a function mapping S to C. In this way we can think of viz, not as a
"multiple-valued function" in C, but as a bona fide (single-valued) function
on the surface S.
A similar construction is possible for other multiple-valued functions.
For example, to form a surface that carries log z as a single-valued function,
one uses countably many copies of C, each slit along the negative real axis,
indexed by all of the integers, and imagined as stacked one above the other
in three-dimensional space. One makes a single surface by identifying the
upper edge of the slit in each plane with the lower edge of the slit in the
plane immediately above it. As in the preceding example, there is a natural
way of introducing local complex coordinates on this surface.
The surfaces associated with the functions viz and log z are examples of
what are called Riemann surfaces. One can associate such a surface with
any multiple-valued holomorphic function. While for simple cases like viz
and log z the surface can be described concretely, as indicated above, in the
general case it is constructed in a natural way from the branches of the
function. The surfaces serve to make mathematically rigorous the intuitive
notion of multiple-valued function; at the same time they provide domains
on which their associated multiple-valued functions become single-valued.
In general, a Riemann surface is a surface equipped in a consistent
way with local complex coordinates. Such a surface need not arise from
a multiple-valued function. The simplest example of one that does not
is C, the extended complex plane (or Riemann sphere). To make C into
a Riemann surface one needs only to introduce complex coordinates in a
1
neighborhood of oo, which one does by means of the map z f--* - • It was
z
Riemann's brilliant insight that Riemann surfaces, and not merely the com-
plex plane itself, form the natural setting for complex analysis. The study
of Riemann surfaces is a major activity in modern complex function theory.
48 IV. Elementary Functions

Figure 3. A depiction of the Riemann surface of .,/Z, with the natural

map from the surface to C.
 Chapter V

Power Series

Power series play a central role in complex function theory because, as we

shall see in Chapter VII, every holomorphic function can be represented
locally by such series. The basic theory of complex power series, to be de-
veloped in this chapter, is largely indistinguishable from its real counterpart.

V .1. Infinite Series

Given a sequence co, CI, c2, ... of complex numbers, one says that the
infinite series L~=O Cn converges if Nlim L;:=O Cn exists and is finite. The
-HXl

number s = lim
N-><XJ
L;:=O Cn is then called the sum of the series, written
s = L~=O Cn. The number Cn is called the n-th term and the number
SN = L;:=O Cn the N-th partial sum of the series. A series that does not
converge is said to diverge.

V.2. Necessary Condition for Convergence

If the series L~=O Cn converges then lim
n---too
Cn = 0.
The proof is the same as for real series. Namely, assume the series
converges, lets be its sum, and SN its N-th partial sum. Then SN - SN-I =
CN, SO

lim CN = lim (sN - SN-I)

N---too N---too
= lim SN - lim SN-I
N---too N->oo
= s - s = 0.

-49
50 V. Power Series

(Use has been made of the fact that, if two sequences of complex numbers
converge, then the sequence of differences converges to the difference of the
limits. This is a corollary of Exercise I.8.3.)

V.3. Geometric Series

1
If c is a complex number, then the series L~=O en converges to if
1-c
lcl < 1 and diverges if lcl 2: 1.
The statement about divergence follows immediately from (V.2). The
statement about convergence is a consequence of the identity
N 1- cN+l
'"""'en= - -
L 1-c '
n=O
which holds whenever c #- 1, and which one can easily verify by multiplying
both sides by 1 - c and making cancellations on the left side. If lei < 1 then
lclN+l -----+ 0 as N -----+ oo, and the desired conclusion follows.

V.4. Triangle Inequality for Series

If the series L~=O Cn converges, then IL~=O Cn I : : ; L~=O lcn I·

In fact, let s denote the sum of the series and SN its N-th partial sum.
We have

n=O
(by Exercise I.5.1). As N -----+ oo the left side converges to Isl (by Exercise
I.5.2) and the right side converges to L~=O lcnl· The desired inequality
follows.

V.5. Absolute Convergence

The series L~=O Cn is said to converge absolutely if the series L~=O lcnl
converges. Just as for real series, the absolute convergence of a complex
series implies its convergence. In fact, letting SN as usual denote the N-th
partial sum of the series L~=O en, we have, for 0 < M < N,
N N oo

lsN - SMI = L Cn ::::; L lcnl::::; L lcnl·

n=M+l
If the series L~=O lcnl converges, then the quantity on the right side tends
to 0 as M -----+ oo, enabling us to conclude that the sequence (sN )N=l is a
Cauchy sequence. Since the metric space C is complete, the sequence of
partial sums converges, as desired.
V. 7. Series of Functions 51

V.6. Sequences of Functions

Suppose (gn)~ 0 is a sequence of complex-valued functions defined in the
open subset G of C. The sequence is said to converge on the subset S of G
if lim 9n(z) exists finitely for each z in S. If that happens, and g is the
n--+oo
limit function (i.e., g(z) = n--+oo
lim 9n(z) for each z in S), then one says that the
sequence converges uniformly on S provided sup{lg(z)-gn(z)I: z ES} tends
to 0 as n ______, oo. Stated in E-language, the sequence converges uniformly
to g on S if, for each positive number E, there exists a positive integer no
such that lg(z) - 9n(z)I < E for all n 2 no and all z in S. An equivalent
condition, which does not mention the limit function, is that the sequence
(gn)~=O be uniformly Cauchy on S, in other words, that for each positive
number E, there exist a positive integer no such that l9n(z) - 9m(z)I < E for
all m 2 no, all n 2 no, and all z in S (see Exercise V.6.1 below).
The sequence (gn)~=O is said to converge locally uniformly in G if each
point of G has a neighborhood in which the sequence converges uniformly.
A simple example, with G the open unit disk, is provided by the sequence
9n(z) = zn. If 0 < ro < 1 then this sequence converges uniformly to 0 in the
disk lzl < ro, and so it converges locally uniformly to 0 in the disk lzl < l.
However, the sequence does not converge uniformly in the disk Iz I < 1.

Exercise* V.6.1. Prove, with the notations above, that the sequence (gn)~=O
converges uniformly on S if and only if it is uniformly Cauchy on S.

Exercise V.6.2. Prove that the sequence (gn)~=O converges locally uni-
formly in the open set G if and only if it converges uniformly on each com-
pact subset of G.

V. 7. Series of Functions
Consider an infinite series L~=O fn, where the terms fn now are not complex
numbers but complex-valued functions, all defined in some open subset G of
C. As was the case with series of numbers, we define convergence for a series
of functions by referring to the sequence of partial sums. The N-th partial
sum of the series L~=O fn is the function 9N = 2:;;1= 0 fn· We say the series
converges, or converges uniformly, on the subset S of G if the sequence
(gN )'N=o of partial sums does. We say that the series converges locally
uniformly in G if the sequence of partial sums does. It is an elementary
exercise to show that if the series L~=O Ifni converges locally uniformly in
G then so does the series L~=O f n (Exercise V. 7 .1 below).
As a simple example, consider the geometric series L~=O zn. From the
discussion in Section (V.3) we know that the series converges in the disk
52 V. Power Series

1
lzl < 1 to - - , and that
1-z

1
It follows that the series L~=O zn converges locally uniformly to m
1-z
the disk lzl < 1.
Exercise* V. 7 .1. Prove, using the notations above, that the local uniform
convergence of the series L~=O Ifni in G implies the local uniform conver-
gence of the series L~=O f n in G.

Exercise* V.7.2. Prove that the series L =n=O (z-z +--1l)n converges locally
uniformly in the half-plane Re z > 0, and find the sum.

V .8. Power Series

A power series is a series of the form L~=O an(z-zo)n, where zo, ao, ai, a2, ...
are complex constants. The number zo is called the center of the series, and
the number an is called the n-th coefficient. The series converges at least
at z = zo. (Note that, when working with power series, we always use
the convention o0 = 1. Thus, when z = zo, the initial term of the series
L~=O an(Z - zor has the value ao.)
Power series can be thought of as "generalized polynomials." As we shall
see, they can often be manipulated exactly as if they were polynomials.
From this perspective, a polynomial is just a power series in which only
finitely many coefficients are nonzero, and for which therefore questions of
convergence do not arise.
If a power series converges to a function f in a given region, we shall
say that the series represents f in that region. It is important, however,
to distinguish the series from the function it represents. For example, as
shown at the end of the preceding section, the series L~=O zn represents the
1
function - - in the disk lzl < 1. However, the series is not "equal" to the
1-z
function, in a formal sense, even though we can write L~=O zn =
1- z
- 1-
for lzl < 1. The same function is represented by other power series; for
example, it is represented by the series L~=O 2-n- 1 (z + 1r
in the larger
disk lz + ll < 2, as the reader will easily verify. A power series is best
thought of as a formal sum, uniquely determined once its center and its
coefficients have been specified.
V.9. Region of Convergence 53

Exercise V.8.1. Let k be a positive integer and let zo be a point of C.

Find the power series with center zo that represents the function zk.

V.9. Region of Convergence

The region of convergence of a power series with center zo is either the
singleton { zo}, the entire complex plane, or an open disk with center zo plus
a subset of the boundary of that disk. In any open disk with center zo where
it converges, the series converges absolutely and locally uniformly.
Thus, the convergence picture for power series is relatively simple. The
preceding proposition describes the situation completely, except for the ques-
tion of how the series behaves, in the third of the three possibilities above,
on the circle forming the boundary between the region of convergence and
the region of divergence. Although that is an interesting question which has
been the subject of deep studies, it is peripheral to the present development.
In proving the proposition we shall for simplicity treat only the case
where zo = 0. It will be clear that the general case is no different. The
proposition is an immediate consequence of the following assertion: If the
power series L~=O anzn converges at a point on the circle lzl = ro, where
ro > 0, then it converges absolutely and locally uniformly in the disk lzl < ro.
We shall prove this assertion by comparing the given series with a geo-
metric series. Suppose the series L~=O anzn converges for z = c, where
lei = ro. Then lancnl ----> 0 as n ----> oo by (V.2), so the number M =
sup{lanlr0 : n = 0, 1, ... } is finite. Since

each term of the series L~=l anzn is dominated in absolute value by M

times the corresponding term of the geometric series Loon=O I!_ro In. The last
series converges locally uniformly in the disk lzl < ro; in fact, its sum in
that disk is (l _ l~/rol), and the difference between its sum and its N-th

partial sum is (lz/rolN+l) (see Section V.7). Using the Cauchy criterion
1 - lz/rol
for uniform convergence (stated in Section V.6, with the proof requested in
Exercise V.6.1) we can conclude that the series L~=O lanznl also converges
locally uniformly in the disk lzl < ro, which is the desired conclusion (in
view of Exercise V. 7 .1).
54 V. Power Series

V.10. Radius of Convergence

Let I:~=O an(z - zo)n be a power series, and let R denote the supremum of
lz - zol over all numbers z for which the series converges. The number R is
called the radius of convergence of the series. By the assertion proved in the
preceding section, if R > 0 then the series converges absolutely and locally
uniformly in the disk lz - zol < R; if R < oo then the series diverges at each
point of the region Iz - zo I > R.
Presently we shall obtain a general expression for the radius of con-
vergence of a power series in terms of its coefficients. Its proof involves a
comparison with geometric series, as did the proof of V.9. The next exercise
makes explicit the simple comparison test involved.

Exercise* V.10.1. Let I:~=O anzn and L~=O bnzn be power series with
respective radii of convergence R1 and R2. Assume that there is a positive
constant M such that lanl :S Mlbnl for all but finitely many n. Prove that
R1 2: R2. Use this test to prove that the radius of convergence of the series
°"
L...n=l n-nzn is oo ·
00

Exercise V.10.2. Prove that the series L.:;~=O nnzn has radius of conver-
gence 0.

V.11. Limits Superior

A basic notion from real analysis will now be reviewed. Given a sequence
(an)~=l of real numbers, we form the sequence

sup{ ak : k 2: n}, n = 1, 2, 3, ... ,

a nonincreasing sequence in (-oo, oo]. Being nonincreasing, the preceding

sequence converges to a point in [-oo, oo]; its limit is called the limit superior
of the original sequence and denoted by lim sup an. The notion of limit
n->oo
inferior is defined analogously and satisfies liminf an = -limsup(-an)·
n->oo n->oo
The sequence (an)]-"' converges finitely if and only if its limit superior and
limit inferior are equal and finite.
Here are a few simple examples:
(1) limsup(l + ~)
= limsup(l - ~) = 1;
n->oo n n->oo n
(2) limsup(-lt=l;
n->oo
(3) lim sup n = oo;
n->oo
(4) limsup(-n)=-oo.
n->oo
V.12. Cauchy-Hadamard Theorem 55

The following properties of limit superior can be found in any book on

real analysis. Their proofs would be good exercises for the reader unfamiliar
with the notion.
(i) If (an)l° and (f3n)l° are two sequences of real numbers, then
limsup(an + f3n)::; limsupan + limsupf3n,
n--->oo n--->oo n-+oo
provided the sum on the right is well defined (i.e., excluding the case where
one summand is oo and the other is -oo). If one of the sequences converges
then the equality holds (with the same proviso).
(ii) If (an)l° and (f3n)l° are two sequences of positive real numbers,
then
limsup(anf3n)::; (limsupan)(limsupf3n),
n--->oo n-->oo n--->oo
provided the product on the right is well defined (i.e., excluding the case
where one factor is 0 and the other is oo). If one of the sequences converges
then the equality holds (with the same proviso).
(iii) If (an)l° is a sequence of real numbers, and the real number p
satisfies p < lim sup an, then the inequality p < an holds for infinitely many
n--->oo
indices n.
(iv) If (an)l° is a sequence of real numbers, and the real number p
satisfies p > lim sup an, then the inequality p > an holds for all but perhaps
n-->oo
finitely many indices n.

V.12. Cauchy-Hadamard Theorem

The radius of convergence of the power series L:~=O an(z - zor is equal to
1

n--->oo
The proof of this theorem is essentially the same as the proof in Section
V.9. One could avoid the duplication of proofs by combining the Cauchy-
Hadamard theorem and the proposition of V.9 into a single statement. The
gain in efficiency, however, would be at the expense of digestibility.
It will obviously suffice to prove the theorem for the case zo = 0. Let
R denote the radius of convergence of the power series I:~=O anzn. We first
1 1
establish the inequality R 2". 1 , assuming that lim sup Ian In: < oo
limsuplanln: n--->oo
n-+oo
(since otherwise the inequality is trivial). Let p be any positive number
1
exceeding lim sup Ian In: . By property (iv) in the preceding section, we then
n-->oo
1
have p > lanln: except perhaps for finitely many indices n. Thus, with
56 V. Power Series

finitely many exceptions, the n-th coefficient of the series L~=O anzn is
dominated in absolute value by the n-th coefficient of the geometric series
L~=O pn zn, whose radius of convergence is ~. It follows that R 2: ~ (by
p p
1
Exercise V.10.1). Since p can be taken arbitrarily close to limsup lanl", we
n-+oo
1
obtain the desired inequality, R 2: 1 •
limsup lanl"
n-+oo

To complete the proof we establish the opposite inequality,

1
RS_ i ,

limsup lanl"
n-+oo
1
assuming that lim sup Ian In > 0. Let p be any positive number less than
n-+oo
1
lim sup Ian In:. By property (iii) in the preceding section we then have p <
n-+oo
1 1
lanl" for infinitely many indices n. Hence, if lzl = -,
p
then lanznl > 1 for
infinitely many indices n, implying that the series L~=O anzn diverges. It fol-
l 1
lows that R S_ - , and since p can be taken arbitrarily close to lim sup Ian In:,
P n-+oo
1
we obtain the desired inequality, R S_ 1 . The proof of the theo-
lim sup lanln
n--+oo
rem is complete.

Exercise V.12.1. Let the power series L~o anzn and L~o bnzn have
respective radii of convergence R1 and R2. Prove that the radius of conver-
gence of the series L~= 0 (an + bn)zn is at least the minimum of R1 and R2,
and the radius of convergence of the series L~=O anbnzn is at least R1R2
(provided the product is well defined).

Exercise* V.12.2. Prove that a power series L~=O anzn and its termwise
derivative, the series L~=l nanzn-l, have the same radius of convergence.
Exercise V.12.3. If the series L~=O anzn has radius of convergence R,
what are the radii of convergence of the series L~=O anz 2n and L~o a;zn?

V.13. Ratio Test

a
If lim _n_ exists, then that limit is the radius of convergence of the power
n-+oo an+l

series L~=oan(Z - zor. This will be deduced from the Cauchy-Hadamard

theorem. Let R denote the radius of convergence of the power series, and
V.14. Examples 57

let R' = lim I~

an+l
n---+oo
I(
R' = oo is allowed). We suppose first that R' is finite
and prove that R ::; R'. Let p be any positive number exceeding R'. Then
the inequality I~ I< p holds for all except perhaps finitely many indices
an+I
n, say for all n 2 no. We can assume ana f- 0 (since our hypothesis implies
that only finitely many of the coefficients an can vanish). Iteration of the
inequality lan+1I > p- 1lanl gives
lana+kl > P-klan I, 0 k = 1, 2, 3, ....
Hence, for n > no we have
1 1 '.':':Q_ 1
Ian In > - p n lanai".
p

The right side in the preceding inequality tends to ~ as n-----+ oo, implying
p
(in view of the Cauchy-Hadamard theorem) that
1 1 1
R = limsup lanln 2 -.
n---+oo P
Since p can be taken arbitrarily close to R', the desired inequality, R ::; R',
follows.
To complete the proof we suppose R' > 0 and show that R 2 R'. The
argument is nearly identical to the preceding one. Let p be any positive
number less than R'. Then the inequality I~
an+ I
I
> p holds for all except
perhaps finitely many indices n, say for all n 2 no. As above, we can iterate
the inequality lan+1I < p- 1 lanl to get, for n >no,
1 1 '.':':Q_ 1
lanln < - p n lanai"·
p
1
The quantity on the right side having the limit as n -----+ oo, we can
p
conclude that
1 1 1
R = lim sup Ian In ::; - .
n---+oo P
Since p can be taken arbitrarily close to R', the desired inequality, R 2 R',
follows.

V.14. Examples
When it applies, the ratio test is generally easier to use than the Cauchy-
Hadamard theorem. Often, however, the simple comparison test given in
Exercise V.10.1 suffices to determine the radius of convergence.
58 V. Power Series

oo n
Example 1. L :.n___
n=l

One can use the ratio test to show that the radius of convergence of the
series is 1. Alternatively, a comparison with the series L~=O zn shows that
the radius of convergence is at least 1. The radius of convergence is at most
1 because the series diverges at z = 1.
00

Example 2. L(n + l)zn.

n=O
Again, the ratio test gives immediately that the radius of convergence
equals 1. The same conclusion can be reached from Exercise V.12.2, since
the given series is the termwise derivative of the series L~=O zn.
oo n
Example 3. L; .
n=O n.
00

Example 4. L(n!)zn.
n=O
In each case the ratio test applies. In Example 3 the radius of conver-
gence is oo, in Example 4 it is 0.
Exercise V.14.1. Find the radius of convergence of each of the following
series.

(a)
oo
I:\;
n
n
(b) '°"' ~z3n.'
00 (

~ (3n)!
1)3
(c) f Zn!;
n
n=l n=l
00

(d) L(n!)zn!;
n=O

Exercise V.14.2. Prove that the series '°"' 00

L_..,n=l
zn converges at every point
n
of the unit circle except the point z = 1.

V.15. Differentiation of Power Series

Let the power series L~=O an(z - zor have a positive radius of convergence
R. Then the function that is represented in the disk lz - zol < R by the
series is holomorphic, and its derivative is represented in that disk by the
series L~=I nan(z - zo)n- 1 , the termwise derivative of the original series.
An immediate corollary is that the function represented in the disk lz -
zol < R by the original series is differentiable to all orders, its k-th derivative
being represented by the termwise k-th derivative of the original series. In
V.15. Differentiation of Power Series 59

particular, the value at zo of the k-th derivative of that function is k!ak. A

power series with center zo and positive radius of convergence R is thus the
Taylor series with center zo of the function it represents.
In proving the proposition, we can assume without loss of generality
that zo = 0. From Exercise V.12.2 we know that the series L~=O anzn and
L~=l nanzn-l have the same radius of convergence. Let f and g denote the
respective functions in the disk lzl < R represented by these series. Fix a
point z1 in that disk. We shall prove that J'(z1) = g(z1) by estimating the
size of the difference between f(z) - f(zi) and g(z1) as z tends to z1. Use
z - z1
will be made of the following standard algebraic identity, which holds for
any two complex numbers z and w and any positive integer n:

n-l
(z - w) L zkwn-k-l.
k=O

One proves this by multiplying out the right side and making the obvious
cancellations. (The case w = 1 of (*) appeared earlier in this chapter, in
Section V.3, in the analysis of geometric series.)
Fix a positive number p such that lz1 I < p < R. Henceforth we assume
that lzl < p. We have

f (Z) - f (Z1) - g (z1 ) = 1-

-- [~
Lanz n - ~
Lanz 1 nl
z - z1 z - z1 n=O n=O

=
~
= an [Zn -
z-z1
z]1 - nz~-ll ·

By (*) the last expression can be rewritten as

which is the same as

60 V. Power Series

Using (*) again, we have

k-1
lzk - z~I = lz - z1I I>jz~-j-l
j=O
k-1
:S lz - z1 IL lzlj lz1 lk-j-l
j=O
:S k/- 1lz - z1I-
Thus
f(z) - f(zi)
z - z1
( )
- g z1 ~ ~an[~ z~-k-'(z' - zi)]
,; ~Ian I [~ lz11•-k-' lz' - ztl]

,; ~ lanl [~Pn-k-l(kpk-llz - z11)]

= lz - z1I ~ lanl [ n(n 2- l) Pn- 2] ·

The series L oo
n=2
n(n -1)
2
lanlPn- 2 converges, by V.9 in conjunction with
Exercise V.12.2, applied twice. Letting M denote its sum, we can write our
estimate as
f(z) - f(zl) - g(z1)I :S Mlz - z1I (lzl < p).
I z - z1
. f(z) - f(z1) . . .
We conclude that hm = g(z1); m other words, f is d1fferen-
z->z1 z - z1
tiable at z1 and f'(z1) = g(z1). The proof is complete.

V.16. Examples
So far, the only power series we have summed explicitly is the series .l.::~=O zn.
We are now able to sum a few more series.
Example 1. The series .l.::~=O ~~ has radius of convergence oo. It equals its
termwise derivative. If f is the function the series represents then, by V.15,
we have f' = f. By Exercise IV.3.1 we can conclude that f is a constant
times ez. But since the functions f and ez both equal 1 at the origin, the
constant must be 1. Thus .l.::~=O ~~ = ez for all z.

In a more sophisticated approach to complex analysis than is being at-

tempted here, one uses the preceding equality to define ez. By clever use
V.17. Cauchy Product 61

of power series one then deduces all of the basic properties of the exponen-
tial function and of its relatives, the hyperbolic and trigonometric func-
tions (without assuming prior knowledge of the real-variable versions of
these functions). This approach can be found, for example, in the books
of L. V. Ahlfors, and S. Saks and A. Zygmund, cited in the list of references.
Example 2. The power series I:~=l z:
has radius of convergence l. Its
termwise derivative is the series L~=O zn. Hence, if f is the function in the
disk lzl < 1 represented by the series I:~=l z:,
then, by V.15, f'(z) = l~z·
By Exercise II.8.l(a) we can conclude that the functions f and Log l~z differ
by a constant. But both functions equal 0 at the origin, and hence they are
equal:
oo Zn 1
'"""' - = Log - (lzl < 1).
~ n 1-z
n=l

Exercise V.16.1. Find the power series with center zo that represents the
function ez.
Exercise V.16.2. What function is represented by the power series I:~=l n 2 zn?
Exercise V.16.3. Let k be a nonnegative integer. Prove that the power
series
z)k+2n
00 (-l)n ( 2
~ n!(n + k)!
has radius of convergence oo. The function represented by the series is called
the Bessel function of the first kind of order k and denoted by J k. Prove
that Jk satisfies Bessel's differential equation:
z 2 J"(z) + zj'(z) + (z 2 - k 2 )f(z) = 0.
(Bessel functions, of which those introduced above are particular exam-
ples, have been exhaustively studied. They arise in numerous questions in
mathematics and physics, for instance, in the study of a stretched circular
membrane, to cite the most popular example.)
d
Exercise V.16.4. Find an expression for dz [z-kJk(z)] in terms of Jk+1(z).

V.17. Cauchy Product

Let L~=O an(z-zor and L~=O bn(z-zo)n be two power series with the same
center. Their Cauchy product is by definition the power series 2::~ 0 cn(z -
zor whose n-th coefficient is given by Cn = L~=O akbn-k· It arises when
one forms all products aj(z - z 0)i bk(z - zo)k, adds for each n the ones with
j + k = n, and sums the resulting terms. Suppose the series I:~=O an (z - zo) n
has a positive radius of convergence R1 and the series I:~=O bn(z-zo)n has a
62 V. Power Series

positive radius of convergence R2. We shall prove that their Cauchy product
converges in the disk lz - zol < min{Ri, R2} to the product of the functions
represented by the two original series.
We can assume without loss of generality that zo = 0. Suppose lzl <
min{R1, R2}. For Na positive integer we have

(t. a;zj) (E t. bkzk) - CnZn

N

2::: ajbkzj+k - L L ajbkzj+k

0'5_j,k'5_N n=O j+k=n

L a1bkzj+k.
0'5_j,k'5_N
j+k>N

It follows that

(t, (t,a1zi) b,zk) - t. Cnz"

< L la1bkzj+kl
0'5_j,k'5_N
j+k>N

<
~<rnax{j,k}'5_N

< (~ la;zj1) (t, lb,z'1) + (t, la;z 1) c~ lb,z'1)

1

< (~ la;zj1) (t, lb·z•1) + (t, la;zj1) c~ lhz'1)

The last expression tends to 0 as N --too, because both series I::;o la1zll
and L~o Ibkzk I converge. In view of the preceding inequality, therefore, we
can conclude that
00

LCnZn =
n=O

as desired.

Exercise V.17.1. Deduce the law of exponents, e 21 e 22 = e 21 +22 , from the

power series representation of e 2 •
V.18. Division of Power Series 63

Exercise V.17.2. Consider the correspondence that associates with each

power series I:~=O anzn centered at 0 the infinite matrix (o:jk)'f'k=O defined
by
a · k for j 2: k
O:jk = { OJ- for j < k.
The correspondence is obviously linear. Prove that it is also multiplicative,
in other words, that the matrix associated with the Cauchy product of the
two series is the product of the matrices associated with the two series. (The
matrices arising in this exercise are instances of Toeplitz matrices, which is
to say that they have constant diagonals.)

V.18. Division of Power Series

Suppose the power series I:~=O bn(z - zor and I:~=O Cn(Z - zor
have pos-
itive radii of convergence and so represent holomorphic functions g and h,
respectively, in disks with center zo. Suppose also that g(zo) = bo of. 0. The
quotient f = h/ g is then holomorphic in some disk with center zo. Assume
j is represented in that disk by a power series I:~=O an(Z - zor
(an assump-
tion to be justified in VII.8). How does one find the coefficients an in terms
of the coefficients bn and Cn?
A method that always works in principle uses the Cauchy product, ac-
cording to which
n
Cn Lakbn-k> n = 0, 1, ....
=
k=O
From this we can conclude that ao = co/bo and

an= bl
O
(en - I:
k=O
akbn-k), n = 1, 2, ....

The last equality expresses an in terms of en, bo, ... , bn and ao, ... , an-l,
enabling one to determine the coefficients an recursively starting from the
initial value ao = co/bo.
Exercise V.18.1. Use the scheme above to determine the power series with
center 0 representing the function J(z) = 1/(1 + z + z 2 ) near 0. What is
the radius of convergence of the series?
 Chapter VI

Complex Integration

The integral of most use in complex function theory is a type of line integral,
usually taken around a closed curve in C. The complex line integral is intro-
duced, and its basic properties developed, in the present chapter. The first
three sections contain preliminary material on integration of complex-valued
functions of a real variable.

Vl.1. Riemann Integral for Complex-Valued Functions

The complex-valued function </> on the closed subinterval [a, b] of R is said
to be piecewise continuous if it is continuous at all but finitely many points
of [a, b] and has finite one-sided limits at each point of discontinuity. In that
case the Riemann integrals of Re</> and Im </>over [a, b] are defined, and we
define the integral of</> over [a, b] by

1b cp(t)dt = 1b Re <f>(t)dt + i 1b Im cp(t)dt.

From the linearity of the Riemann integral for real-valued functions one
obtains immediately the linearity for complex-valued functions: if </>1 and
¢2 are piecewise-continuous complex-valued functions on [a, b] and c1 and c2
are complex constants, then

Of course, one can use the Riemann integral under less restrictive con-
ditions than the piecewise continuity of the function to be integrated, but
we shall not need to do so here.

-65
66 VI. Complex Integration

Vl.2. Fundamental Theorem of Calculus

The complex-valued function </> on the interval [a, b] is said to be differen-
tiable at the point to of [a, b] if Re </> and Im </> are both differentiable at t 0 ;
we then define the derivative of</> at to by ¢'(to)= (Re¢)'(to) +i(Im<f>)'(to).
(This notion was used earlier, in Section 11.10.) The function </> is said to be
piecewise C 1 if it is continuous, it is differentiable at all but finitely many
points of [a, b], its derivative is continuous at each point where it exists,
and its derivative has finite one-sided limits at each of the finitely many
points where it fails to exist. From the fundamental theorem of calculus
for real-valued functions one obtains immediately the corresponding result
for complex-valued functions: if the complex-valued function </> on [a, b] is
piecewise C 1 ' then
1b <P'(t)dt = <f>(b) - <f>(a).

Exercise Vl.2.1. Prove that a piecewise-C 1 function has one-sided deriva-

tives at each point where it is not differentiable.

VI.3. Triangle Inequality for Integration

If <P is a piecewise-continuous complex-valued function on the interval [a, b],
then

To establish this, assume J: <f>(t)dt -=I- 0 (otherwise the inequality is triv-

ial), and let

II: <f>(t)dtl
J: <t>(t)dt .
Then

11b <f>(t)dtl >. 1b <f>(t)dt

=

=1b >.¢(t)dt = Re 1b >.¢(t)dt

= 1b Re [>.¢(t) ]dt :::; 1b [>.¢(t)[dt

= 1b l<P(t)ldt,
as desired.
VI.5. The Complex Integral 67

VI.4. Arc Length

Although the reader has no doubt already encountered the notion of arc
length, a brief review is in order. The length of the piecewise-C 1 curve
1: [a, b] ~ C is the number

L('Y) 1b l1'(t)ldt.

The reader who wishes may take the preceding equality as the definition
of L('-y). It is more natural, however, to define L('Y), for any curve 1 (not
necessarily piecewise C 1 ) to be the supremum of the lengths of all inscribed
polygonal paths. The integral expression above for L('Y), in the case where
1 is piecewise C 1 , then becomes a theorem. Details can be found in the
book of W. Rudin cited in the list of references.
One can gain an intuitive understanding of arc length by interpreting
the parameter t as time. One then thinks of 1 as describing the position
of a point moving in the plane. In case 1 is piecewise C 1 , the derivative
1' (t), at a value t where it exists, gives the instantaneous velocity of the
moving point, and l"f'(t)I gives the instantaneous speed. The total distance
the point covers between the instants t = a and t = b should thus be the
integral of b'I over [a, b].

VI.5. The Complex Integral

Let 1 : [a, b] ~ C be a piecewise-C 1 curve. A complex-valued function f
is said to be defined on 1 if it is defined on a set that contains the range of
1· In that case, the composite function f o 1 is defined. If f is continuous
and defined on 1, then f o 1 is continuous, and we define the integral of f
over 1, denoted J,, f(z)dz, by

1 f(z)dz = 1b f('Y(t)h'(t)dt.

The integral just defined has the usual linearity property: if Ji and h
are two continuous functions defined on 1, and if c1 and c2 are complex
constants, then

1 [cifi(z) + c2f2(z)]dz = c1 1 fi(z)dz + c2 l h(z)dz.

This follows immediately from the analogous property mentioned in Section

VI.l.
Another simple property is that we can partition 1 into two (or more)
subcurves and then express J,, f(z)dz as the sum of integrals over the sub-
curves. For example, if a < c < b, and if 11 = 11 [a, c] and 12 = 11 [c, b],
68 VI. Complex Integration

J J J
then ~ f(z)dz = ~1 f(z)dz + ~2 f(z)dz. This is a trivial consequence of
the corresponding property of the Riemann integral.
Sometimes we shall wish to use a symbol other than "z" to denote the
variable of integration (the "dummy" variable). A frequent choice is "(".
There is no difference between J~ f (()d( and J~ f (z)dz. Both by definition
equal J:J('"Y(t))'-y'(t)dt. (This is the same sort of flexibility in notation one
encounters when first studying integration, in calculus.)

VI.6. Integral of a Derivative

Let"( : [a, b] _____. C be a piecewise-C 1 curve. Let f be a holomorphic function
defined in an open set containing"(, and assume that J' is continuous. Then

!, J'(z)dz = J('"Y(b)) - J('"Y(a)).

In particular, if"( is closed (i.e., "f(b) = "f(a)), then J~ f'(z)dz = 0.

In fact, by definition we have

!, J'(z)dz = 1b J'('"Y(t)) "f1 (t)dt.

By the chain rule (see Appendix 2), the integrand in the integral on the right
side is the derivative of the composite function f o 'Y· The desired conclusion
thus follows from the version of the fundamental theorem of calculus given
in Section VI.2.
As mentioned in Chapter II, we shall prove later that a holomorphic
function has a holomorphic derivative, making superfluous the assumption
in the preceding proposition that J' is continuous. For the time being,
however, that assumption is needed. If f is represented by a power series,
convergent in the whole plane, say, then we can use the proposition, for we
know from Chapter V that f is then holomorphic and that f' is represented
by a power series, and hence is continuous.

VI.7. An Example
Let zo be a point of C and Ra positive real number. Define"( : [O, 27r] _____. C
by 'Y(t) = zo + Reit. Thus,"( is a parametrization of the circle lz - zol = R,
traversed once in the counterclockwise direction. If n =f- -1 then, by VI.6,
J~ (z - zordz = 0, because (z - zor is the derivative of (z-zn~l)n+I .
On the other hand, referring back to the definition of the integral, we
have
VI.8. Reparametrization 69

Exercise VI. 7.1. Prove on the basis of the last calculation that there is no
branch of arg z in the region 0 < Iz I < 1.
Exercise VI. 7.2. Derive the formula

-
1
27r
1 0
21!"
COS
2
n tdt =
1 · 3 · 5 · · · · · (2n - 1)
2 · 4 · 6 · · · · · (2n)

by integrating the function :; z

1( + :;1) 2n
around the unit circle, parametrized
by the curve 1(t) = eit (0::; t::; 27r). (Suggestion: Expand the binomial.)

Vl.8. Reparametrization
Let r : [a, b] C be a piecewise-C 1 curve. One says that the curve
-----"*
11 : [a1, bi] -----"* C is a reparametrization of r if it has the form 11 = r o /3,
where f3 is an increasing piecewise-C 1 map of [a 1, b1] onto [a, b]. In this case,
if f is a continuous function defined on r (and hence on 11), then

1 /'
f(z)dz = 1 /'1
J(z)dz.

In fact, from the definition of the complex integral,

i f(z)dz = 1b f(r(t))r'(t)dt,

1 /'1
J(z)dz = 1b
al
1
J(r(f3(s)))r'(f3(s))/3 1(s)ds.
The two integrals on the right side are equal by the change-of-variables
formula for the Riemann integral.
Thus, the curve ')' and its reparametrizations are indistinguishable for
purposes of integration. This will often save us the trouble, when we wish to
perform an integration, of mentioning a particular parametrization; it will
usually be enough to give a geometric description of the integration curve
and to specify the direction of integration.
For example, if z1 and z2 are distinct points of C, then by [z1, z2] we
shall mean the segment with endpoints z1 and z2, directed from z1 to z2.
The standard parametrization of the segment is given by the function 1(t) =
(1 - t)z1 + tz2 (0 ::; t ::; 1). If the function f is defined and continuous on
[z1, z2], then Jrz ,z
1 2 ] f(z)dz has an unambiguous meaning: it equals J 7 J(z)dz
for the ')' above or any of its reparametrizations. We need not be explicit
about a parametrization to be able to talk about f[z 1 ,z2 ] J(z)dz.
For closed curves we have even a bit more flexibility regarding parametriza-
tions, in that the choice of initial-terminal point is irrelevant for purposes
of integration. Suppose ')' : [a, b] -----"* C is a piecewise-C 1 curve with
70 VI. Complex Integration

'Y(a) = 'Y(b), and let ai be a point in (a, b). Let bi = ai + (b - a), and
define the curve 'Yi : [ai, bi] ---t C by

'Yi(t) = {'Y(t), ai ~ t ~ b
'Y(t +a - b), b ~ t ~ b1.
Thus, we obtain 'Yi by "shifting" 'Y so as to make 'Y(ai), rather than 'Y(a), the
initial-terminal point of the parametrization. The reader will easily verify
J J
that I f(z)dz = /1 f(z)dz for any function f that is defined and continuous
on 'Y·
Exercise VI.8.1. Let zi and z2 be distinct points of C. Evaluate Jf[fz1 ,z2 l zndz
and frz1,z2] zndz for n = 0, 1, 2, ....
Exercise VI.8.2.
(a) Evaluate f[-i,i] lzldz.
(b) Evaluate fc+ lzldz, where
C+ is the right half of the unit circle,
oriented counterclockwise.
Exercise VI.8.3. Let the complex-valued function f be defined and con-
tinuous in the disk Iz - zo I < R. For 0 < r < R let Cr denote the circle
lz - zol = r, with the counterclockwise orientation. Prove that
lim rf(z) dz = 27ri f(zo).
r--->O}cr Z - Zo
Exercise VI.8.4. Let f, R and Cr be as in the preceding exercise, and
assume that f is of class ci. Prove that

lim 12
r--->Or ~
r
f(z)dz = 27ri 8a~(zo).
z

(See Section II.16 for the definition of ~~.) Conclude that f is differentiable
at zo if and only if
lim 12 r
r--->O r Jcr
J(z)dz = 0.

VI.9. The Reverse of a Curve

If 'Y : [a, b]---t C is a curve, then the reverse of 'Y is the curve -')' :

[-b, -a] ---t C defined by (-'Y)(t) = 'Y(-t). It is the curve one obtains
from 'Y by reversing its direction. If 'Y is piecewise ci' then so is -')'' and
f_ 1 f(z)dz = - J, J(z)dz for any function f that is defined and continuous
on 'Y· The reader will easily verify this on the basis of the change- of-variables
formula for the Riemann integral.
VI.12. An Example 71

VI.10. Estimate of the Integral

Let "( : [a, b] - - t C be a piecewise-C 1 curve and f a continuous complex-
valued function on "f· Let M be the maximum of Iii on"(, that is,
M = max{lf("t(t))I: a St Sb}.
Then
l.l f(z)dzl < M L("f).

In fact, by VI.3 we have

l.l f(z)dzl 11b f("f(t)h'(t)dtl

< lb If b (t)) 11 'Y' (t) Idt.

The last integral is majorized by M J: l'Y'(t)ldt, which equals M L("f).

Vl.11. Integral of a Limit

Let "( : [a, b] - - t c be a piecewise C 1 curve. Let the sequence Un)~=l of
continuous complex-valued functions on "( converge uniformly on "( to the
function f. Then
lim
n-->oo
1'Y
fn(z)dz = 1'Y
f(z)dz.

In fact, let Mn denote the maximum of If - fnl on "f· By VI.10,

l.l f(z)dz - .l fn(z)dzl = l.l [f(z) - fn(z)]dzl S MnL("f).

Since lim Mn = 0 (by the hypothesis that fn --t f uniformly on "f), the
n-->oo
desired conclusion follows.

VI.12. An Example
It is now possible to illustrate how complex methods can be used to evaluate
improper Riemann integrals that are inaccessible by the usual techniques of
calculus. Many other examples of this kind will be given later, in Chapters
VII and X, after more machinery has been developed.
For a and b positive numbers, let Ra,b denote the rectangle with vertices
±a, ±a+ ib, oriented counterclockwise. By VI.6,
r e-z 2 dz = 0.
}Ra,b
72 VI. Complex Integration

We are going to hold b fixed and let a tend to oo. The preceding integral
is the sum of the integrals of e-z 2 over the four segments [-a, a], [a, a+ ib],
[a+ib, -a+ib], [-a+ib, -a]; let those integrals be denoted by fi(a), h(a),
h(a), J4(a), respectively. Obviously,

fi(a) = 1-: e-x 2 dx.

Parametrizing the segment [-a+ ib, a+ ib] by the function 'Y(t) = t + ib,
where -a ::; t ::; a, we have

h(a) = 1-"(
e-z 2 dz = -1 'Y
e-z 2dz

= -la e-(t+ib)2 dt
-a

= - 1_: eb 2 -t 2 (cos2bt - isin2bt)dt

= -eb 2 1-: e-t 2 cos 2bt dt.

(The imaginary part disappeared because e-t 2 sin 2bt is an odd function of
t.) Since I:J=l Ij(a) = 0, we obtain

eb 2 1-: e-t 2 cos2btdt = 1_: e-x 2 dx + h(a) + l4(a).

- 2 2_ 2 • 2
Because le z I = eY x , the maximum of le-z I on the segment [a, a+
ib] is e-a 2+b2 . By VI.10, therefore, lh(a)I ::; be-a 2+b2 • Exactly the same
estimate holds for J4(a). Therefore h(a) and J4(a) both tend to 0 as a
tends to oo. Taking the limit as a ------+ oo in the equality above, we thus
obtain

Now, as the reader probably knows, J_ 00 e-x dx can be found by ele-

oo 2

mentary means. The trick is to note that

Switching to polar coordinates we can rewrite the preceding double integral

as
fo fo
00 2
n: e-r 2 rdOdr,

which is easily seen to have the value 7r. Hence J~00 e-x 2 dx = .Jii. Combin-
ing this with the equality at the end of the last paragraph, we see that we
VI.12. An Example 73

have succeeded in evaluating the nonelementary integral J~00 e-t 2 cos 2bt dt:

1_: e-t 2 cos 2bt dt = yf1re-b 2 •

(The integral arises in Fourier analysis.)

Exercise VI.12.1. Prove that

fo 00
e-t 2 cos t 2 dt = ~vn V1 + J2
by integrating the function cz 2 in the counterclockwise direction around the
boundary of the region {z: lzl SR, 0 S Arg z Si}, and letting R ____. oo.
Exercise VI.12.2. Evaluate the integrals J000 cos t 2 dt and f 000 sin t 2 dt (the
Fresnel integrals) by integrating e-z 2 in the counterclockwise direction around
the boundary of the region { z : Iz I ::; R, 0 ::; Arg z ::; ~}, and letting
R ____. oo.
 Chapter VII

Core Versions of
Cauchy's Theorem,
and Consequences

Cauchy's theorem is the central theorem of complex function theory. It

states, in one of its versions, that J,,
f(z)dz = 0 whenever 'Y is a closed
curve and the function f is holomorphic in an open set containing 'Y and its
interior. In a more sophisticated version, the single curve 'Y is replaced by a
system of curves.
A general version of Cauchy's theorem is presented in Chapter IX; it re-
quires considerable preparation. In the present chapter two special cases will
be established, the case where 'Y is a triangle, and, on the basis of that, the
case where the function f is holomorphic in a convex open set containing 'Y·
This will be enough to enable us to draw far-reaching conclusions concern-
ing the behavior of holomorphic functions. Some applications to harmonic
functions are given at the end of the chapter.

VIl.1. Cauchy's Theorem for a Triangle

Let T be a triangle in C, and let f be a holomorphic function in an open set
containing T and its interior. Then fr f(z)dz = 0.
This is sometimes called Goursat's lemma, after E. Goursat, who, in the
late 1800's, was the first to recognize that, in the definition of a holomor-
phic function, one can dispense with the requirement that the derivative be
continuous.

- 75
76 VII. Core Versions of Cauchy's Theorem, and Consequences

Figure 4. Carving up the triangle T

To prove the theorem, we assume for definiteness that T has the coun-
terclockwise orientation, and we let I = fr f(z)dz. By means of seg-
ments joining the midpoints of the sides of T, we construct four triangles
Toj, j = 1, 2, 3, 4, each similar to but one-half the size of T (see Figure 4).
Orienting each Toj counterclockwise, we have

l f(z)dz =
4

~ £0
j f(z)dz.

Hence there is a j such that

Pick such a j, and let T1 denote Toj for that j.

From T1 we construct four triangles T1j, j = 1, 2, 3, 4, in the same way
that the triangles Toj were constructed from T. Orienting each T1j counter-
clockwise, we have

L 1 f(z)dz,
4
{ f(z)dz
lr1 j=l T1j
VII.1. Cauchy's Theorem for a Triangle 77

so, since lfr f(z)dzl > ~111,

1 there must be a j such that

Ilr1jr f(z)dzl ;: : :
4
2111.
1

Pick such a j, and let T2 denote T1j for that j.

Proceeding in this manner, we obtain a sequence (Tn)~=l of triangles,
each Tn being one of the four triangles obtained from Tn-1 by the process
of connecting midpoints, such that

lln f(z)dzl 2: 4~ Ill

for each n. We note that L(Tn) = 2~L(T) for each n.
Let Kn be the union of Tn and its interior. The sets Kn are compact,
they decrease as n increases, and their diameters tend to 0. Hence, their
intersection consists of a single point, which we call zo. The point zo is
either on Tor in the interior of T, and thus it is in the open set where f is
holomorphic. Hence, by II.2, we can write
f(z) = f(zo) + J'(zo)(z - zo) + R(z),

where lim R(z) = 0. By VI.6 the integral of the function f(zo)+ f'(zo)(z-
z-+zo Z - zo
zo) around Tn is 0, so
{ f(z)dz = { R(z)dz.
}Tn }Tn
We thus have
Ill :=; 4nlln R(z)dzl.

For each n, let En denote the maximum of I R( z) I for z in Tn \{zo}.

z - zo
Then lim En= 0, and for z in Tn,
n-+oo

IR(z)I :=; En diameter(Tn) :=; En L(Tn)·

Hence

lln R(z)dzl :=;En L(Tn) · L(Tn)

= :: L(T) 2 .
In combination with our earlier inequality, this gives
Ill
:=; EnL(T) 2 .
Since lim En = 0, it follows that 1 = 0, as desired.
n-+oo
78 VII. Core Versions of Cauchy's Theorem, and Consequences

VIl.2. Cauchy's Theorem for a Convex Region

Let G be a convex open subset of C and f a holomorphic function in G.
Then J,,
f(z)dz = 0 for every piecewise-C 1 closed curve 'Y in G.
We recall that a subset of C is called convex if, whenever it contains two
points z1 and z2, it contains also the entire segment with endpoints z1 and
z2. To prove the theorem we fix a point zo in G and define the function g
in G by
g(z) = { f((,)d(,.
J[zo,z]
We shall show that g is holomorphic and that g' = f. The desired conclusion
will then follow by VI.6.
Fix a point z1 in G. If z is any other point of G, then

{ f((,)d(, + { f((,)d(, + { f((,)d(, = 0,

J[zo,z1] }[z1,z] J[z,zo]
by Cauchy's theorem for a triangle (or trivially in case the points zo, z1, z
happen to be collinear). The preceding equality can be written as

g(z) - g(z1) = { J((,)d(,.

J[z1,z]
Consequently

g(z)- g(z1) - f(z1) = _1_ { [!((,)- f(z1)]dc,.

Z - Zl Z - Zl l[zi,z]

Now fix a positive number c. By the continuity of f, there is a positive

number 5 such that If((,) - f(z1)I < E whenever IC, - z1I < 5. Hence,
if lz - z1 I < 5, then the absolute value of the integrand in the preceding
integral is bounded by E, and we have

g(z) - g(z1) - f(z1)

z - z1 lz ~ z1l lz1,z] [!((,) - f(zi)]d(,

< lz ~ z1I. t:L([z1,zJ) = E.

g(z) - g(zi) . . .
This proves that lim = f(z1), rn other words, g is d1fferen-
z - z1
z---+z1
tiable at z1 and g' ( z1) = f (z1), as desired.

VIl.3. Existence of a Primitive

If the function f is defined in the open subset G of C, then by a primitive
off one means a holomorphic function in G whose derivative equals f. The
VII.5. Cauchy's Formula for a Circle 79

preceding proof shows that f has a primitive in case it is holomorphic and G

is convex. The proof in fact establishes something that seems stronger: If the
continuous complex-valued function f in the open convex set G has integral
0 around every triangle contained in G, then f has a primitive in G. This is
not in reality stronger, because, as we shall show very shortly, the derivative
of a holomorphic function is itself holomorphic, which means a function f
satisfying the hypotheses in the preceding statement is in fact holomorphic.
The preceding statement thus amounts to a converse of Goursat's lemma.
The converse is stated formally in Section VII.10.

VIl.4. More Definite Integrals

Cauchy's theorem for a convex region, although very special compared to
the general Cauchy theorem, is by itself enough for many applications. In
the following exercises it is used to evaluate improper integrals.
Exercise VII.4.1. Let 0 < b < 1. Derive the equality

1 00

_ 00
1- b + x 2
----~--~dx
(1 - b + x 2 ) 2 + 4bx 2
= 7f

by integrating the function (1 + z 2 )- 1 around the rectangle with vertices

±a, ±a+ iVb (a > 0) and taking the limit as a---+ oo. What goes wrong if
b > 1?
Exercise VIl.4.2. Let a be a positive number. Evaluate the integrals

1 00

o t +a
4
1
4 dt,
1=
o t4
t2
+ a4 dt
by integrating the function (z 2 + a 2 )- 1 in the counterclockwise direction
around the boundary of the region { z : 0 ~ Iz I ~ R, 0 ~ Arg z ~ ~}, and
taking the limit as R---+ oo.

VII.5. Cauchy's Formula for a Circle

Let C be a counterclockwise oriented circle, and let f be a holomorphic
function defined in an open set containing C and its interior. Then

f(z) = ~
27ri
r f(()z d(
le ( -
for all points z in the interior of C.
Thus, the values of f in the interior of C can be reconstructed from the
values on C. A holomorphic function "hangs together".
Cauchy's formula for a circle is a very special case of a more general
formula, to be established in Chapter X. The case of a circle is all we shall
80 VII. Core Versions of Cauchy's Theorem, and Consequences

Figure 5. Integration curves in the proof of Cauchy's formula for a circle.

need to show that holomorphic functions can be represented locally by power

series and thus are differentiable to all orders.
To establish the formula we fix a point zo in the interior of C. For 0 < E <
dist (zo, C), let Ce be the circle with center zo and radius E, oriented coun-
terclockwise. Our first task is to prove that 1 f(() d( =
C ( - ZQ
1 f(() d(
C, ( - ZQ
for every E.

The preceding equality will be deduced from Cauchy's theorem for a

convex region. For each E we subdivide the region between C and Ce into four
subregions by means of two horizontal and two vertical segments connecting
C with Ce, as in Figure 5. The boundary of each subregion is a closed curve
made up of a subarc of C, a subarc of Ce, and two of the segments, one
horizontal and one vertical. Let these boundary curves be denoted by 'Yl,
/'2, /'3, /'4, and let them be given the counterclockwise orientation. Each /'j is
contained in a convex open set in which the function /(() is holomorphic
.., -zo
(for example, in the intersection of a suitable open half-plane with an open
disk containing C in which f is holomorphic). By Cauchy's theorem for a
convex region, therefore,

1"/j (
J(() d(
- zo
0, j = 1, 2, 3, 4.
VII.6. Mean Value Property 81

Since
'"""1
4
J(() d(
L.t . ( - zo
r J(()zo d( _ ler (J(()
=
le ( -
- zo
d(,
J=l ~ €

we have the desired equality, r (f(() d( r (J(() d(. =

1e - zo 1e. - zo
In Section VI. 7 it is shown that

r -(
1
le. -zo
d( = 27ri.

This in conjunction with the preceding equality gives

~ r J(() d( -
2m, 1e ( - zo
J(zo) = _1
27ri 1e.
r f(()( -- zof(zo) d(
for each f between 0 and dist (zo, C). Let ME denote the maximum of
I J((~ =~;zo) I for ( on CE. The absolute value of the last integral is then
bounded by ME L(CE), implying that

I~ f (J(() d( -
2m le - zo
J(zo)I :::; t:ME.

As E tends to 0 so does f ME (since ME ---+ IJ'(zo)I), and the desired equality

follows.
Exercise VII.5.1. Let f and C be as in the statement of Cauchy's formula.
What is the value of { (J(() d( when z is in the exterior of C?
le -z
VIl.6. Mean Value Property
Let the function f be holomorphic in the disk Iz - zo I < R, and for 0 < r < R
let Cr denote the circle with center zo and radius r, oriented counterclock-
wise. By Cauchy's formula for a circle we have

f(zo) = ~ { J(z) dz, O< r < R.

27ri ler z - zo
Parametrizing Cr by means of the function zo + reit (0 :S t :S 27r), and
rewriting the complex integral as an integral with respect to the parameter
t, we obtain the equality

J(zo) = __!._ f 2rr f(zo + reit)dt.

27r lo
In other words, the value off at the center of Cr is the average of its values
over Cr.
82 VII. Core Versions of Cauchy's Theorem, and Consequences

Exercise VIl.6.1. Retaining the notations above, and writing z = x + iy,

prove that

f(zo) ~f
7rT
1 lz-zol<r
f(x+iy)dxdy, O<r<R;

in other words, the value of f at the center of the disk lz - zol < r is the
average of its values over the disk.

VII. 7. Cauchy Integrals

Let "( : [a, b] ---> C be a piecewise-C 1 curve, and let ¢ be a continuous
complex-valued function on 'Y· The Cauchy integral of¢ over 'Y is the func-
tion f defined in C\"( by

f(z) = 1 ¢(() d(.

')' ( - z
Cauchy's formula for a circle thus states that the Cauchy integral of a holo-
morphic function over a counterclockwise oriented circle is, inside the circle,
equal to 27ri times the function itself, provided the circle and its interior
both lie in the open set where the function is holomorphic.
We shall prove that any Cauchy integral, and hence any holomorphic
function, has local power series representations. We retain the notations
above. Fix a point zo off"(, and let R denote the distance from zo to 'Y· Let
z be a point satisfying lz - zol < R, regarded for the moment as fixed. We
. 1 . .
wnte -r-- as a geometric senes:

1
( - z
"- z

= (( -
1
zo) - (z - zo) = ( -
1 [
zo 1 -
1
c=:o
l =
~
~
(z - zo)n
(( - zo)n+l.
The series converges locally uniformly (with respect to () in the region
=
I~ :~ I < 1, and hence it converges uniformly on 'Y (a compact subset of
the preceding region). The function ¢ is bounded on "(, being continuous
there, and therefore the series
~ ¢(()(z - zo)n
~ (( - zo)n+l
n=O

converges uniformly on 'Y to

" - z
!( () .
By VI.11 we can thus integrate the
preceding series term-by-term over "(to obtain

f(z) = f 1 ((-zo)n+l
n=O 'Y
<P(()(z - zor d(, lz - zol < R.
VII.8. Implications for Holomorphic Functions 83

In other words, letting

1 'Y (( -
¢(()
zo)n+l d(,
n = 0, 1, 2, ... ,

we have shown that the power series L~=O an(Z - zor represents f in the
disk Jz - zol < R.
As shown in V.15, a function represented by a convergent power series is
differentiable to all orders, and the series is the Taylor series of the function
·t represent s. lxr
I vve thus h ave an = f(n)(zo) , w h.ich in
· comb.mat1on
· wit· h t he
n.1
expression above defining an gives an integral representation for f(n). Since
zo is an arbitrary point off"(, we can write the representation as
f(n)(z) = nl 1 ¢(() d(
• "( (( _ z )n+l '
z E C\'"V,
I

which holds for any nonnegative integer n. The upshot is that it is legitimate
to differentiate a Cauchy integral under the integral sign any number of
times.
Exercise VII. 7.1. Find explicitly the Cauchy integral of the constant func-
tion 1 over the interval [O, l].

Vll.8. Implications for Holomorphic Functions

Let f be a holomorphic function in the open set G, and let zo be a point
of G. For 0 < r < dist (zo, C\G), let Cr denote the circle with center zo
and radius r, oriented counterclockwise. Then, inside Cr, the function f
1
coincides with the Cauchy integral of -.! over Cr. This in combination
2m
with what is proved in the preceding section enables us to draw the following
conclusions.
(i) The function f is differentiable to all orders. For any positive integer
n,

f(nl(z) = n! { f(() d( Jz - zol < r <dist (zo, C\G).

2ni Jcr (( - z)n+l '

(ii) In the disk Iz - zo I < dist ( zo, C\ G), the function f is represented by
its Taylor series centered at zo.

Exercise VIl.8.1. Evaluate the following integrals, where C is the unit

circle with the counterclockwise orientation.

(a) 1 sinz
-----:IB dz,
c z
( b) r ( 2zz --12 )
Jc
3 dz.
84 VII. Core Versions of Cauchy's Theorem, and Consequences

Exercise VII.8.2. Evaluate

~
1 1211" 1 dO
27r 0 1- 2rcos0 + r 2
1 "() "()
for 0 < r < 1 by writing cos 0 = 2(e + e 1 -i ) and reducing the given integral
to a complex integral over the unit circle.

Exercise VIl.8.3. Let a and b be complex numbers such that lal < 1 < lbl.
Let C be the unit circle with the counterclockwise orientation. Evaluate
_1 r (z -br dz
27ri Jc (z - a)n
for all pairs of integers m and n.

Vll.9. Cauchy Product

Once one knows that holomorphic functions are represented locally by Tay-
lor series, it is possible to give a simplified approach to the Cauchy prod-
uct of power series. Recall from Section V.17 that the Cauchy product of
r
the two power series L~=O an ( z - zo and L~=O bn ( z - zo) n is the series
L~=O cn(z - zo)n whose n-th coefficient is given by Cn = L~=O akbn-k· In
Section V.17 it is proved in a direct way that if the first two series have posi-
tive radii of convergence Ri and R2, respectively, then their Cauchy product
converges in the disk Iz - zo I < min { R 1, R2} to the product of the functions
they represent. We can now obtain the same conclusion more simply by an
indirect argument.
To do this, let f and g denote the functions represented by the two
original series in the disks lz - zol < Ri and lz - zol < R2, respectively.
Those series are the Taylor series of the functions they represent, so that
f(n)(zo) g(n)(zo)
an = 1 and bn = 1 for each n. Define the function h in
n. n.
the disk lz - zol < min{R1, R2} by h(z) = f(z)g(z). Then his holomorphic
and so is represented in the preceding disk by its Taylor series centered at
zo. To verify that the last series is in fact the Cauchy product of the two
original series amounts to verifying the identity
h(n)(zo) _ ~ J(k)(zo) . g(n-k)(zo)
n! - ~ kl (n - k)! ' n = O, l, 2 ' · · · ·
k=O
But this is just the well-known Leibniz formula for then-th derivative of the
product of two functions, which the reader has probably encountered in real
analysis. The case n = 1 is just the product rule for differentiation, and a
straightforward induction argument yields the general case.
VII.10. Converse of Goursat 's Lemma 85

Exercise VII.9.1. If the function f is holomorphic in a neighborhood of

the origin and L~=O anzn is its Taylor series about the origin, what is the
Taylor series about the origin of the function (1 - z)- 1 f(z)?

Exercise VII.9.2. Prove that the Taylor series about the origin of the
function [Log(l - z) J 2 is
00

L 2Hn
- - z n+l
n+ 1 '
n=l
1 1 1
where Hn 1 + -2 + -3 + · · · + -n (the n- th partial sum of the harmonic
series).

VIl.10. Converse of Goursat's Lemma

Let f be a continuous complex-valued function in the open subset G of C. If
fr f(z)dz = 0 for every triangle T contained with its interior in G, then f
is holomorphic.
Because being holomorphic is a local property, it is enough to prove
this for the case where G is a disk. In that case, as pointed out in Section
VII.3, the hypotheses imply that f has a primitive, in other words, there is a
holomorphic function g in G such that g' = f. Since holomorphic functions
have holomorphic derivatives, the desired conclusion follows.
R. B. Burckel, on page 188 of his book cited in the reference list, states
that the preceding result "is (essentially) contained" in an 1886 paper of G.
Morera. In a letter to the author, Burckel explained that it is hard to pin
down what Morera actually proves because of imprecise language, but what
he seems to assume is that the function in question has integral 0 around all
closed curves in its region. That that assumption implies holomorphicity is
often referred to as Morera's theorem.
The first exercise below gives a stronger Morera-type theorem than the
one involving triangles, stronger in the sense that the one involving triangles
is easily deduced from it.

Exercise VIl.10.1. Prove that if f is a continuous complex-valued function

in the open subset G of C, and if JR
f(z)dz = 0 for every rectangle R, with
edges parallel to the coordinate axes, contained with its interior in G, then
f is holomorphic.
Exercise VII.10.2. (A version of the Schwarz reflection principle.) Let the
function f be continuous in the region { z : Iz I < 1, Im z 2: 0}, real valued
on the segment ( -1, 1) of the real axis, and holomorphic in the open set
{ z : Iz I < 1, Im z > 0}. Use the result of the preceding exercise to prove f
86 VII. Core Versions of Cauchy's Theorem, and Consequences

can be extended holomorphically to the open unit disk. (Suggestion: See

Exercise II.8.2.)

VIl.11. Liouville's Theorem

A function that is holomorphic in all of C is called an entire function. The
theorem of Liouville states that the only bounded entire functions are the
constant functions.
For the proof, assume f is a bounded entire function, say lf(z)I : : ; M for
all z. Fix a point zo in C, and for R > 0 let CR
denote the circle Iz - zo I = R,
with the counterclockwise orientation. By VII.8(i) we have

f (zo) = -21 .
f

1ri
JCR
(
f (z) ) dz.
z - zo 2

The integrand in the preceding integral is bounded in absolute value by ~,

giving the estimate

lf'(zo)I : : ; 2 ~2 L(CR) = ~.
This being true for all R > 0, we have f'(zo) = 0. Since the derivative off
vanishes identically, the function f must be constant, as desired.
Exercise VII.11.1. Let f be an entire function such that, for some positive
integer n and positive number R, If (z) / zn I is bounded for Iz I > R. Prove f
is a polynomial of degree at most n.
Exercise* VIl.11.2. Let f be a holomorphic map of the open unit disk
into itself. Prove that
lf'(z)I : : ; 1 ! lzl
for all z in the disk.
Exercise VIl.11.3. Prove that an entire function with a positive real part
is constant.

VIl.12. Fundamental Theorem of Algebra

Every nonconstant polynomial with complex coefficients can be factored over
C into linear factors.
Purely algebraic reasoning reduces this to the statement that every non-
constant polynomial with complex coefficients has a complex root. In fact,
suppose the preceding statement is known, and let p be a nonconstant poly-
nomial with complex coefficients. Let zo be a root of p. The standard
division algorithm of algebra then enables one to write p(z) = (z - zo)q(z),
where q is a polynomial whose degree is one less than that of p. The same
VII.13. Zeros of Holomorphic Functions 87

reasoning can now be applied to q, and so on, yielding the desired linear
factorization in finitely many steps.
To prove that nonconstant polynomials with complex coefficients have
complex roots we argue by contradiction. Suppose p(z) = anzn+an-lZn-l+
· · ·+ aiz + ao is a polynomial of positive degree n, with complex coefficients,
1
but without complex roots. Let f = - . Then f is an entire function. Since
p
an of= 0 we can write, for z of= 0,

p(z) = anz n ( 1 + -
an-l
- al + -ao- ) .
+ · · · + anzn-l
anz anzn
Consequently,

lp(z)I ~ lanznl (i - lan-11

lanzl
_ ... _ la1I _ ~).
lanzn-ll lanznl

The quantity in parentheses on the right will be no smaller than ~ when lzl
2
is sufficiently large, say when lzl ~ R. Hence
2
lf(z)I :S lanlRn, lzl ~ R.

On the other hand, since If I is a continuous function, it is bounded on the

compact set lzl :::; R. Thus f is a bounded entire function, and so it is
constant by Liouville's theorem. This is the desired contradiction because
the polynomial p was assumed to be nonconstant.
The proof just given is purely an existence proof; it does not provide
one with a method of locating the roots of a polynomial. The problem of
finding accurate approximations to the roots of complex polynomials is a
major theme in numerical analysis.

VIl.13. Zeros of Holomorphic Functions

Zeros of holomorphic functions can be treated like zeros of polynomials.
Suppose f is a holomorphic function in the open subset G of C, and suppose
zo is a point of G such that f (zo) = 0. The point zo is called a zero of f
of order (or multiplicity) m, where m is a positive integer, if j(n)(zo) = 0
for n = 0, ... ,m - 1, while j(m)(zo) i= 0. In that case the Taylor series for
f(m)(zo)
f about zo has the form :E~=m an(z - zo)n, where am= i= 0. We
m.1
can write f (z) = (z - zo)mg(z), where the function g, defined by

g(z) = { (z - zo)-m f(z), z E G\{zo}

am, z = zo,
88 VII. Core Versions of Cauchy's Theorem, and Consequences

is holomorphic in G. The function g, being nonzero at zo, is nonzero in

some neighborhood of z 0 , implying that zo is at a positive distance from
1- 1 ( 0) \ {zo}, the set of all the other zeros of f. In other words, zeros of
finite order of holomorphic functions are isolated.
The point zo is called a simple zero of f if it is a zero of order l. In that
case the function g defined above takes the value f'(zo) at zo. The reasoning
above shows that if zo is a zero of f of any order, then the function that
equals ( z - zo )- 1 f (z) in G\ { zo} and equals f' ( zo) at zo is holomorphic.
What about zeros of infinite order, that is, points at which f and all
of its derivatives vanish? It turns out that these exist only in trivial cases.
First, if G is connected, then f can have a zero of infinite order only if it is
the zero function. In fact, the zeros of f of infinite order form an open set
(since the Taylor series of the function centered at such a zero has all of its
coefficients equal to 0), and, by the discussion in the preceding paragraph, so
do the points of G that are not zeros off of infinite order. If G is connected,
then one of these sets must be empty. If G is not connected, and if f has a
zero of infinite order at the point zo of G, one can at least conclude that f
vanishes identically in the connected component of G containing z0 .
In contrast to the behavior of holomorphic functions, it is possible for a
real-valued C 00 function on R to have a zero of infinite order and yet not
vanish identically. A standard example is the function that equals 0 at the
origin and equals e-l/x 2 elsewhere. One can show that the origin is a zero
of infinite order of this function.
Most often in complex function theory one is concerned with holomor-
phic functions in connected open sets. The following statement summarizes
what is established above about the zeros of such functions. Let f be a
holomorphic function, not the zero function, in the connected open subset G
of C. Then each zero off is of finite order, and f- 1 (0), the zero set off,
has no limit points in G. Since an uncountable subset of C always contains
limit points of itself, one can conclude in particular that the set f- 1 (0) is
at most countable. The same is true of f- 1 (w), for any complex number w,
provided f is not the constant function w.

Exercise VIl.13.1. Prove that there is no holomorphic function f in the

open unit disk such that f(~) = 2-n for n = 2, 3, ....

Exercise Vll.13.2. Let f be a holomorphic function in the open subset G

of C. Let the point zo of G be a zero of f of order m. Prove that there is a
1
branch off m. in some open disk centered at zo.
VII.15. Weierstrass Convergence Theorem 89

VIl.14. The Identity Theorem

Let f and g be holomorphic functions in the connected open subset G of C.
If f(z) = g(z) for all z in a subset of G that has a limit point in G, then
f =g.
Thus, for example, a holomorphic function in the open unit disk is
uniquely determined by the values it assumes on the set {2-n : n = 1, 2, 3, ... }.
To prove the theorem one notes that, if f and g agree on a subset of G
that has a limit point in G, then the zero set of the function f - g has a
limit point in G, which implies by what is proved in the last section that
f - g is the zero function.
Exercise VII.14.1. Prove that there is no holomorphic function f in the

open unit disk such that f ( ~) ~ (-n~n for n ~ 2, 3, ....

Exercise VIl.14.2. (a) Let G be a nonempty, connected, open subset of
C which is symmetric with respect to the real axis. Let f be a holomorphic
function in G such that f is real-valued on G n R. Prove that f(z) = f(z)
for all z in G. (Suggestion: Use Exercise II.8.2.)
(b) Let G be as in part (a), and let f be a holomorphic function in G
such that f is real-valued on a nonempty subinterval of G n R. Prove that
f is real-valued on all of G n R.

VIl.15. Weierstrass Convergence Theorem

Let G be an open subset of C and (fk): 1 a sequence of holomorphic func-
tions in G that converges locally uniformly in G to the function f. Then
f is holomorphic, and for each positive integer n, the sequence (!kn)) : 1
converges locally uniformly in G to J(n).
Once again we have a contrast between real function theory and com-
plex function theory. The local uniform convergence on R of a sequence of
real-valued C 00 functions does not even imply the differentiability, let alone
the infinite differentiability, of the limit function. In fact, by the classical
Weierstrass approximation theorem, any continuous real-valued function on
R, even a nowhere differentiable one, is the local uniform limit of a sequence
of polynomials.
To prove the Weierstrass convergence theorem, fix a point zo in G, and
let r be a positive number such that the disk lz - zol :S r is contained in G.
Let Cr denote the circle lz - zol = r, with the counterclockwise orientation.
90 VII. Core Versions of Cauchy's Theorem, and Consequences

For each positive integer k we have, by Cauchy's formula for a circle,

fk(z) = ~
2m
1 fk(() d(,
Cr ( - Z
lz - zol < r.

As k -----t oo we have fk -----t f uniformly on Cr, so, for fixed z satisfying

Iz - zo I < r, the integrand in the preceding integral converges uniformly on
Cr to !(() . Therefore, by VI.11, the integral converges to
~-z
1
~(-z
f(() d(.
Since also fk(z) -----t f(z), we obtain

f(z) = ~
27ri
1 f(() d(,
Cr ( - Z
lz - zol < r.

Thus f is given in the disk lz - zol < r by a Cauchy integral, so it is

holomorphic in that disk. Since G is covered by such disks, f is holomorphic
in G.
Now fix a positive integer n. By VII.8(i) we have

f(n)(z) = ~
27ri
1 Cr (( -
J(() d(
z)n+l '
lz - zol < r,

and similarly for each function fk. Hence

f(n)(z) - /n)(z) = ~
k 27ri
r J(() - fk(()
Jcr (( - z)n+l
d(
'
lz - zol < r.

Let Mk denote the maximum of If(() - fk(()I for ( on Cr, k = 1, 2, ....

Then, for lz - zol < ~' the absolute value of the integrand in the preceding
2n+1Mk
integral is bounded by rn+l . The usual estimate of the integral therefore
gives us

r
lz - zol < 2·

Since Mk -----t 0, it follows that the sequence (!kn)) 00

k=l
converges uniformly to
f(n) in the disk lz - zol < ~· This establishes the local uniform convergence
of the sequence (!kn)) 00

k=l
to j(n), completing the proof of the theorem.

VIl.16. Maximum Modulus Principle

Let f be a nonconstant holomorphic function in the open connected subset
G of C. Then Ill does not attain a local maximum in G.
VII.17. Schwarz's Lemma 91

An immediate consequence is that, if K is a compact subset of G, then

Ill attains its maximum over K only at points of the boundary of K. This
is the form in which the principle is usually applied.
To establish the maximum modulus principle, we argue by contradiction,
assuming that there is a point zo of G at which Ill attains a local maximum.
Choose a positive number r such that the disk lz - zol :::; r is contained in
G, and such that lf(z)I :::; lf(zo)I for all points z in that disk. If f(zo) = 0,
then it follows immediately from the identity theorem that f is the zero
function, contrary to the hypothesis that f is nonconstant. We thus may as
lf(zo)I
well assume that f(zo) i- 0. Let A= f (zo) . By the mean value property,

f(zo) = __.!_ f 2rr f(zo + reit)dt.

27r lo
Hence

lf(zo)I = .Xf(zo) = -1 127r .Xf(zo + reit)dt

2?r 0

=Re - 1 127r .Xf(zo + reit)dt

27r 0

= 2~ fo 2
rr Re [.Xf(zo + reit) Jdt,
which we can rewrite as
1 {27r (lf(zo)I - Re[.Xf(zo + reit)] dt
)
27r lo = 0.

The integrand in the preceding integral is continuous and nonnegative, so

the vanishing of the integral implies the vanishing of the integrand. Thus
Re .Xf ( z) = If ( zo) I for z on Cr. This together with the inequality I.Xf (z) I :::;
lf(zo)I implies .Xf(z) = IJ(zo)J, in other words, f(z) = f(zo), for z on Cr.
Thus, by the identity theorem, f is the constant function f(zo), contrary to
the hypothesis that f is nonconstant. This completes the proof.
Exercise VIl.16.1. Let f be a nonconstant holomorphic function in the
connected open subset G of C. Prove that Ill can attain a local minimum
in G only at a zero off.
Exercise VIl.16.2. Give a proof of the fundamental theorem of algebra
based on the maximum modulus principle.

VII.17. Schwarz's Lemma

Let f be a holomorphic map of the open unit disk into itself such that f(O) =
0. Then If ( z) I :::; Iz I for all z in the disk. The inequality is strict at all points
92 VII. Core Versions of Cauchy's Theorem, and Consequences

other than the origin, except in the case where f has the form f ( z) = .\z
with .\ a constant of unit modulus.
To establish this we define the function g in the disk by
f(z)
g(z) = { -z-, 0 < lzl < 1
f' (0), z = 0.
Then g is holomorphic (see Section VII.13). For 0 < r < 1 the function g is
bounded in absolute value by ~ on the circle lzl = r, and hence it has the
r
same bound in the disk lzl :S: r, by the maximum modulus principle. This
being true for all r in (0, 1), the function g is bounded in absolute value by 1,
which means lf(z)I :S: lzl, as desired. Moreover, if the preceding inequality
is an equality for any z other than the origin, then 191 has a local maximum
at that z. Then, again by the maximum modulus principle, g is a constant,
.\,which means f(z) = .\z.

Exercise* VIl.17.1. Prove, under the hypotheses of Schwarz's lemma,

that If' (0) I :S: 1, and that the inequality is strict unless f (z) = .\z with .\ a
constant of unit modulus.

Exercise VIl.17.2. (Pick's lemma.) Let f be a holomorphic map of the

open unit disk into itself. Prove that, for any two points z and w in the
disk,
f(z) - f(w) z-w I
I :s: I1 - zw '
1 - f(z)f(w)
and that the inequality is strict for z # w except when f is a linear-fractional
transformation mapping the disk onto itself.

Exercise VIl.17.3. Let f be a holomorphic map of the unit disk into itself.
Prove that
lf'(w)I < 1 - lf(w)l 2
- 1- lwl2
for all w in the disk, and that the inequality is strict except when f is a
linear fractional transformation mapping the disk onto itself. Compare with
Exercise VII.11.2.

Exercise VII.17.4. Let f be a holomorphic map from the half-plane Re z >

0 into the open unit disk. Prove that

lf'(z)I :S: 1 - lf(z)l 2

2Rez
for all z in the half-plane.
VII.18. Existence of Harmonic Conjugates 93

Exercise VIl.17.5. Prove that, except for the identity function, a holo-
morphic map of the open unit disk into itself has at most one fixed point in
the disk.

VII.18. Existence of Harmonic Conjugates

Let u be a real-valued harmonic function in the convex open subset G of
C. Then there is a holomorphic function g in G such that u = Reg. The
function g is unique to within addition of an imaginary constant.
In other words, u has a harmonic conjugate in G, unique to within addi-
tion of a constant. The conclusion is true under a much weaker assumption
on G than convexity, as will be shown later on the basis of the general
Cauchy theorem.
fT\
1-0
. . we consi.d er t h e f unction
prove t h e proposition . f = au
ax - i au
ay ( = 2 au)
oz '
which is holomorphic in G (by Exercise II.16.2). As shown in the proof of
Cauchy's theorem for a convex region, the function f has a primitive in G;
that is, there is a holomorphic function gin G such that g' = f. Fix a point
zo in G. By adding a constant to g, we may assume g(zo) = u(zo). Let
ui =Reg and v1 = Img. We have
8u1 .8v1 8v1 .8u1
g' = --+i- --i--
ax ax ay ay'
and also
g' f = au - iau_
ax ay
H ence -au = --8u1 an d -au = ~.
8u1 The fi rst partia
. l d envatives
· · o f t h e function
·
ax ax ay uy
u - u 1 thus vanish throughout G, implying that u - u1 is constant. Since
u(zo) = u 1(z 0 ), the constant is 0, and u =Reg, as desired.
As for uniqueness, if two holomorphic functions in G have the same
real part, then their difference has derivative 0, by the Cauchy-Riemann
equations. In that case the functions differ by a constant.
Exercise VIl.18.1. Prove that a real-valued harmonic function u in the
open unit disk has a series representation
00

u(rei8 ) = ao + I:>n(an cosnB + bn sin nB),

n=l

where the coefficients an and bn are real numbers.

Exercise VIl.18.2. Prove that a positive harmonic function in C is con-
stant.
94 VII. Core Versions of Cauchy's Theorem, and Consequences

Exercise VII.18.3. Let u be a nonconstant real-valued harmonic function

in C.

Prove that the set u- 1 (c) is unbounded for every real number c.

VII.19. Infinite Differentiability of Harmonic Functions

Harmonic functions are of class C 00 •
Because differentiability is a local property, it will suffice to prove this
for a harmonic function u in an open disk D. We may assume without loss
of generality that u is real-valued. Then, by VII.18, we can write u = Reg
with g holomorphic in D. The desired conclusion now follows by the infinite
differentiability of holomorphic functions.

VIl.20. Mean Value Property for Harmonic Functions

Let u be a harmonic function in the open subset G of C, and let zo be a
point of G. Then

u(zo) = -
1 121!" u(zo + reit)dt, 0 < r < dist (zo, C\G).
27r 0

It will be enough to prove this for the case where u is real-valued. In

that case there is, by VII.18, a holomorphic function in the disk lz - zol <
dist (zo, C\G) whose real part is u. The mean value property for u then
follows by VII.6, the corresponding property for holomorphic functions.

VIl.21. Identity Theorem for Harmonic Functions

If G is a connected open subset of C, and if u and v are harmonic functions
in G that agree on a nonempty open subset of G, then u = v throughout G.
It will be enough to prove this for the case where u is real-valued and v
is the zero function. Suppose then that u is a real-valued harmonic function
in G that vanishes on a nonempty open subset of G. Let H be the interior
of the set u- 1 (0). Then H is nonempty and open. Suppose His not all of
G. Then, because G is connected, some point zo of G lies in the boundary
of H. Chooser > 0 such that the disk lz - zol < r is contained in G. By
VII.18 there is a holomorphic function g in this disk such that u = Reg and
g(zo) = u(zo) (= 0). Since z 0 is in the boundary of H, the intersection of H
with the disk lz - zol < r is nonempty. Thus Reg vanishes on a nonempty
subdisk of the preceding disk. On the smaller disk we have g' = 0 by the
Cauchy-Riemann equations. Hence g' = 0 in the disk lz - zol < r, by the
identity theorem for holomorphic functions. Therefore g is constant in the
disk lz - zol < r. Since g(zo) = 0, the constant is 0. We have thus shown
VII.23. Harmonic Functions in Higher Dimensions 95

that u vanishes in the disk lz - zol < r, which means zo belongs to the set
H (the interior of u- 1 ( 0)). This is a contradiction because zo was chosen to
be in the boundary of H. We can conclude that H = G, in other words, u
is the zero function, as desired.

VII.22. Maximum Principle for Harmonic Functions

Let u be a nonconstant real-valued harmonic function in the connected open
subset G of C. Then u does not attain a local maximum in G.
This is proved from the mean value property and the identity theorem
in the same way that the corresponding result for holomorphic functions is
proved in Section VII.16. Of course, by applying the maximum principle to
-u, we see that u also does not attain a local minimum in G.

VIl.23. Harmonic Functions in Higher Dimensions

The properties given in Sections VII.19 through VII.22 hold also for har-
monic functions of more than two variables. The proofs given here, however,
since they are based on complex function theory, are not available in higher
dimensions. The reader wishing to pursue harmonic functions in greater
depth can consult one of the books on the subject, for example, the one by
S. Axler, P. Bourdon and W. Ramey in the list of references.
 Chapter VIII

Laurent Series and

Isolated Singularities

Laurent series are power series that include negative as well as positive
powers of the variable. Such series arise, in particular, when one examines
holomorphic functions near what are called isolated singularities.

VIIl.1. Simple Examples

1
We have seen that the function - - is represented in the disk lzl < 1 by
1-z
. senes
t h e geometric Lm=O z n . B y wntmg
. "\""= . . -1- = - -
1 ( -1- ) , we see t h at
1
1-z z 1--z
- 1- is represented in the region lzl > 1 by a different series, the series
1-z
- L~=l z-n, a series in negative powers of z.
1
Consider the function ( )( ) , for a slightly more complicated
z-1 z-2
1 1 1
example, which can be written as - - - - - . The function - - is
z-2 z-1 z-2
represented in the disk lzl < 2 by a power series in z, and we have just seen
1
that the function - - is represented in the region Iz I > 1 by a series in
1-z
negative powers of z. Hence, by using a series in both positive and negative
. 1
powers of z, we can represent the function ( )( ) in the annulus
z-1 z-2
1 < lzl < 2.

- 97
98 VIII. Laurent Series and Isolated Singularities

Exercise VIII.1.1. Let a and b be complex numbers such that 0 < lal < lbl.
Find a series in positive and negative powers of z that represents the function
( {(
z-a z-b
) in the annulus lal < lzl < lbl.

VIIl.2. Laurent Series

A Laurent series centered at the point zo of C is, by definition, a series of
the form L~-00 an(Z - zor, where the coefficients an are complex con-
stants. The series is said to converge at the point z if both of the series
L~=O an(z - zo)n and L~=l a_n(z - zo)-n converge. In that case
N
lim ~ an(z - zor
N-+oo ~
n=-N
exists and is defined to be the sum of the series.
Let
R1 limsup la-nll/n,
n-+<XJ

n-+oo
By the Cauchy-Hadamard theorem V.12, the series L~=O an(z - zo)n con-
verges absolutely and locally uniformly in the disk lz - zol < R2, and
the series L~=l a_n(z - zo)-n converges absolutely and locally uniformly
in the region lz - zol > R1. Hence, if R1 < R2, then the Laurent series
L~=-oo an(z - zo)n converges absolutely and locally uniformly in the annu-
lus R1 < lz - zol < R2, called the annulus of convergence of the series.
Assume R1 < R2, and let f be the function to which the Laurent
series converges in the annulus R1 < lz - zol < R2, in other words, let
f(z) = lim
N-+=
L n=-N an(z -
N
zo)n. Then f is holomorphic, by the Weier-
strass convergence theorem VII.15. For R1 < r < R2, let Cr denote the
circle Iz - zo I = r, oriented counterclockwise. The series then converges
uniformly to f on Cr, enabling us to evaluate the integral fer J(z)dz by
integrating the series term-by-term. We obtain

1
Cr
f(z)dz = f 1
n=-<Xl
an
Cr
(z - zordz

= 27ria_1.
Thus
a_1 = ~
2m
r f(z)dz.
Jcr
VIII.3. Cauchy Integral Near oo 99

Applying this to the function (z - zo)-k-l f(z) in place off, where k is any
integer, we find that

ak = 21 .
7ri
r (z - zo)-k-lf(z)dz.
Jcr
The coefficients of the series are thus uniquely determined by the function
to which the series converges.

Exercise VIII.2.1. Determine the annulus of convergence of the Laurent

series I:~=-oo an 2 zn, where 0 < lal < l.

VIll.3. Cauchy Integral Near oo

We shall show that any function given by a Cauchy integral can be repre-
sented "near oo" by a Laurent series in negative powers of z - zo. The rea-
soning is essentially the same as that used in Section VII. 7 to obtain power
series representations of Cauchy integrals. The result to be established will
be used later to show that any function holomorphic in an annulus has a
Laurent series representation.
Let 'Y : [a, b] ---+ C be a piecewise-C 1 curve, and let ¢ be a continuous
complex-valued function defined on 'Y· Let f be the Cauchy integral of ¢
over 'Y, that is, the function defined in C\ "( by

J(z) = r ¢(()z d(,.

J'Y ( -
Fix a point zo in C, and let R denote the maximum of I( - zol for (on "f·
Fix any point z such that lz - zol > R. We have
1 1
( - z (( - zo) - (z - zo)

(z =zo) [i -~]
1
z-zo
00

=- L (( - zo)n-l(z - zo)-n,
n=l

the series converging locally uniformly (with respect to () in the region

I( - zo I < Iz - zo I, and hence uniformly on "(. The series
00

- L ¢(()(( - zot- 1(z - zo)-n

n=l
100 VIII. Laurent Series and Isolated Singularities

. ¢(()
thus converges urnformly to ( _ z on 'Y. We can therefore integrate the
series term-by-term over 'Y to obtain

f(z) = - f
n=l
(z - zo)-n 1(( -
I
zor- 1 ¢(()d(;

in other words, the Laurent expansion f(z) = I:;;-~-oo an(z - zo)n, with

an= - ~ (( - zo)-n- 1 ¢(()d(,

holds in the region Iz - zo I > R.

VIIl.4. Cauchy's Theorem for Two Concentric Circles

Let the function f be holomorphic in the annulus R1 < lz - zol < R2. For
R1 < r < R2, let Cr denote the circle lz-zol = r, oriented counterclockwise.
Then fer f(z)dz is independent of r (R1 < r < R2).
To prove this, let I(r) denote the preceding integral. Introducing the
parametrization t----+ zo + reit (0 :::; t:::; 27r) of Cr, we can write

I(r) = fo 2
7r f(zo + reit)ireitdt.
The integrand in the preceding integral, as a function of the two real vari-
ables r and t, has continuous partial derivatives. By a standard theorem in
calculus (whose proof is in Appendix 4), we can find d ~~) by differentiating
under the integral sign:
dI(r)
dr
We have

:r [f (zo + reit)ireit] = J' (zo+reit)ire2it + f (zo+reit)ieit = :t [f (zo + reit)eit].

Consequently
dI(r)
dr
the desired conclusion.
Exercise VIIl.4.1. Let p and q be polynomials such that deg q > deg
p + l. Let C be a circle whose interior contains all of the roots of q. Prove
that
p(z)
Je q(z) dz = 0.
r
VIII.6. Existence of Laurent Series Representations 101

VIll.5. Cauchy's Formula for an Annulus

Let the function f be holomorphic in the annulus R1 < lz - zol < R2. For
R1 < r < R2, let Cr denote the circle lz-zol = r, oriented counterclockwise.
If R1 < ri < lw - zol < r2 < R2, then

f(w) = ~ r f(z) dz -
27ri}rcr2 z - w
~ r f(z) dz.
2m}rcr1 z - w

To prove this we fix was above and define the function gin the annulus
R1 < Iz - zo I < R2 by
f(z) - f(w)
{ z I- w
g(z) = z- w '
J'(w), Z=W.

Then g is holomorphic (see Section VII.13), so by VIII.4 we have

1 Cr 1
g(z)dz = 1Cr 2
g(z)dz.

This we can rewrite as

The first integral on the right side is 27ri, by Cauchy's formula for a circle,
and the second integral is 0, by Cauchy's theorem for a convex region. The
right side therefore equals 27ri f (w), as desired.

VIll.6. Existence of Laurent Series Representations

Let the function f, as before, be holomorphic in the annulus R1 < lz - zol <
R2. Fix r1 and r2 satisfying R1 < ri < r2 < R2. By VIII.5 (with slightly
altered notation),

In other words, in the annulus ri < lz - zol < r2 we have f = h +Ji, where

h(z) = - 1- r f(() d(
27ri j Cr2 ( - z ,
lz - zol < r2;

fi(z)
102 VIII. Laurent Series and Isolated Singularities

As shown in Section VII.7, the function h has in the disk lz - zol < r2 the
power series representation J(z) = L~=O an(z - zo)n, where

an= --!--:
1fZ
r (( - zo)-n-l f(()d(.
Jcr 2

As shown in Section VIIl.3, the function Ji has in the region lz - zol > r1
the Laurent series representation f(z) = L~~-oo an(z - zo)n, where

an= --!--:
1fZ
r (( - zo)-n-l f(()d(.
Jcr 1

Consequently, f has in the annulus rl < lz - zol < r2 the Laurent series
representation f(z) = L~=-oo an(Z - zor. By VIIl.4, if n 2: 0 the integral
defining an is independent of r2 (provided, of course, that Rl < r2 < R2),
and if n < 0 the integral defining an is independent of rl (with the anal-
ogous proviso). Since r2 can be chosen arbitrarily close to R2 and ri can
be chosen arbitrarily close to R1, we can conclude that the Laurent se-
ries representation f(z) = L~=-oo an(z - zo)n holds in the entire annulus
R1 < lz - zol < R2.

VIII. 7. Isolated Singularities

By a punctured disk in C centered at zo, one means an open annulus with
center zo and inner radius 0. Such a region is defined by a pair of inequalities,
0 < lz - zol < R, where R, the radius of the punctured disk, is positive.
The case R = oo is not excluded, but in that case one usually speaks of a
punctured plane rather than a punctured disk.
Let f be a holomorphic function defined in the open subset G of C. The
point zo is called an isolated singularity off if zo is not in G but G contains a
punctured disk centered at zo. Then, as shown in the preceding section, there
is a Laurent series representation J(z) = L~=-oo an(z - zo)n converging in
the punctured disk 0 < lz-zol < R, where R is the distance from zo to C\G
(interpreted as 00 in case G = C\ {zo} ). The series L~~-00 an(Z - zor is
called the principal part of the Laurent series of f at zo. The point zo is
classified as either a removable singularity, a pole, or an essential singularity
of f according to the following scheme.
The point zo is called a removable singularity of f if an = 0 for every
negative integer n, in other words, if the principal part of the Laurent series
of f at zo is trivial. In that case one can extend f holomorphically to
the open set GU {zo} by defining f(zo) = ao. Henceforth, functions with
removable singularities will be assumed to be so extended.
An example of a removable singularity is provided by the function f(z) =
ez-:_ 1 , defined originally in C\27riZ. The function ez - 1 that appears in the
VIII. 7. Isolated Singularities 103

denominator is entire and has a simple zero at the origin, so it can be

written as ez - 1 = zg(z), where g is entire and g(O) i= 0 (in fact g(O) = 1).
1
The function - is thus holomorphic in a neighborhood of the origin, and it
g
coincides with f in a punctured disk centered at the origin. It follows that
the origin is a removable singularity of f.
The point zo is called a pole of f of order m, where m is a positive
integer, if a_m i= 0 but an = 0 for n < -m. In that case the principal
part of the Laurent series of f at zo is a rational function. An example is
provided by the function J(z) = z-m, which has a pole of order m at the
origin.
If f has a pole of order matzo, then the function g(z) = (z - zor J(z)
has a removable singularity at zo; we can extend it holomorphically to
GU {zo} by defining g(zo) = a-m· Thus J(z) = (z - zo)-mg(z) with g
holomorphic in G U { zo} and nonzero at zo. It follows that the function J
has a removable singularity at zo; its holomorphic extension to a neighbor-
hood of zo has a zero of order m at zo. In particular, if zo is a pole of f,
then lim f(z) = oo.
z---+zo

A pole of order 1 is referred to as a simple pole.

The point zo is called an essential isolated singularity of f if it is an
isolated singularity of f but is neither a removable singularity nor a pole
of f. Thus, zo is an essential singularity of f if and only if the principal
part of the Laurent series off has infinitely many nonvanishing terms (i.e.,
there are infinitely many negative integers n such that an i= 0). An example
is provided by the function e 1/z, defined in C\{O}. Its Laurent expansion
about 0 is L~=l ~ z-n. It thus has the origin as an essential singularity.
One says that the function f has an isolated singularity at oo if the
open set G where f is defined contains a punctured disk centered at oo
(that is, the exterior of some circle Jzl = R, with R 2 0). One then calls oo
a removable singularity, a pole of order m, or an essential singularity off if
the function f ( ~) has the origin as a removable singularity, a pole of order
m, or an essential singularity, respectively. For example, the point oo is a
1
removable singularity of the function - , a pole of order m of the function
z
zm, and an essential singularity of the function ez.

Exercise VIII. 7 .1. Prove that if the holomorphic function f has an isolated
singularity at zo, then the principal part of the Laurent series of f at zo
converges in C\ {zo}.
104 VIII. Laurent Series and Isolated Singularities

Exercise VIII. 7.2. Locate and classify the isolated singularities of the
following functions. (Don't ignore oo as a possible singularity.)
z5 1 1
(c) sin -.
(a) 1+z+z 2 +z 3 +z 4 ' (b) sin 2 z' z
Exercise VIII. 7.3. Prove that a rational function has no essential singu-
larities.
Exercise* VIII. 7 .4. (Partial fraction decomposition.) Let f be a noncon-
stant rational function, and let z1, z2, ... , Zp be its poles in C. Prove that f
can be written as f =Ji+ h + · · · + fp, where each fj is a rational function
whose only pole is Zj.
Exercise VIII. 7.5. Let the holomorphic function f be defined in C\F
where F is a finite set. Assume f has no essential singularities. Prove that
f is a rational function.
Exercise VIII. 7. 6. Let the holomorphic functions f and g have poles of
the same order, m, at the point zo. Prove that
lim f(z) = lim f'(z)
z--->zo g(z) z--->zo g'(z)
(a complex version of !'Hospital's rule).
Exercise VIII. 7. 7. The Bernoulli numbers Bn (n = 0, 1, 2, ... ) are defined
by
CX)

Z Bn n
'""""
(lzl < 27r).
ez - 1 = L...t --;! z
n=O
(As shown in this section, the function on the left has a removable singularity
at the origin and so is represented by a power series about the origin. The
radius of convergence of that series is 27r because the zeros of ez - 1 closest
to 0, other than 0 itself, are ±27ri.)
(a) Prove that Bo = 1, and establish the recurrence relation

Ln (n + 1) Bk
k = 0, n = 1,2, ... ,
k=O

which expresses Bn in terms of Bo, ... , Bn-l· Here, ( n: 1)

stands for the binomial coefficient 1( ( n + 1) ! )

k. n + 1- k.1
1
(b) Prove that B1 = - 2.
(c) Prove that Bn = 0 if n is odd and larger than 1.
(d) Prove that the numbers Bn are rational.
VIII.9. Criterion for a Pole 105

VIll.8. Criterion for a Removable Singularity

Let the holomorphic function f have an isolated singularity at the point zo
of C. If f is bounded in some punctured disk with center zo, then zo is a
removable singularity off.
In fact, take positive numbers M and E such that f is defined and
bounded in absolute value by Min the punctured disk 0 < lz - zol < E. Let
L~=-= an(Z - zor be the Laurent expansion off in that punctured disk.
For 0 < r < E let Cr denote the circle Iz - zo I = r, oriented counterclockwise.
For n > 0 we have
a_n = 21 .
1rZ
r (z - zo)n-l f(z)dz
Jcr
(see Section VIII.2). The integrand in the preceding integral is bounded in
absolute value by rn-l M, so
n-lM
la-nl :S r 21T L(Cr) = rn M, n = 1, 2, ....
Since r can be taken arbitrarily small it follows that a_n = 0 for n > 0, the
desired conclusion.
The preceding criterion obviously applies also in the case z0 = oo.
Exercise VIII.8.1. Show that Liouville's theorem is a corollary of the
preceding criterion.
Exercise VIII.8.2. Let the holomorphic function f have an isolated singu-
larity at the point zo of C. Assume that there are positive numbers Mand E
and a positive integer m such that lf(z)I :S Mlz-zol-m for 0 < lz-zol < E.
Prove that zo is either a removable singularity of f or a pole of order at
most m.
Exercise VIll.8.3. Stand straight with feet about one meter apart, hands
on hips. Bend at the waist, knees slightly flexed, and touch your left foot
with right hand. Straighten. Bend again and touch right foot with left hand.
Straighten. Repeat 15 times.

VIII.9. Criterion for a Pole

If the holomorphic function f has an isolated singularity at the point zo, and
if lim lf(z)I = oo, then zo is a pole of J.
z--->zo
In fact, under the given hypothesis, the point zo is a removable singular-
ity of the function g = J' by VIII.8, and g(zo) = 0. The desired conclusion
follows immediately from this.
106 VIII. Laurent Series and Isolated Singularities

VIIl.10. Casorati-Weierstrass Theorem

Let the holomorphic function f have an essential isolated singularity at the
point zo. Then, for any complex number w, there is a sequence (zn)~=l in
the region where f is defined such that lim Zn = zo and lim f (zn) = w.
n~cx:i n--+cx:>

In fact, if the conclusion were to fail for a particular w, it would follow

1
by VIII.8 that zo is a removable singularity of the function g = f _ w. But
that would mean zo is either a removable singularity off (the case g(zo) # 0)
or a pole off (the case g(zo) = 0). The desired conclusion thus follows by
contradiction.

VIll.11. Picard's Theorem

The Casorati-Weierstrass theorem reveals a wildness in the behavior of a
holomorphic function near an essential isolated singularity: in any punc-
tured disk centered at the singularity, the function comes arbitrarily close
to every complex value. A deeper theorem of E. Picard reveals more: in any
punctured disk centered at an essential isolated singularity, a holomorphic
function actually assumes every complex value infinitely often, with perhaps
one exception. In other words, except possibly for one value of w, the equal-
ity lim f(zn) = w in the conclusion of the Casorati-Weierstrass theorem
n--+=
can be replaced by f(zn) = w, n = 1, 2, .... Picard's theorem is illustrated
by the function ez, which is easily seen to assume every value other than 0
infinitely often in any punctured disk centered at oo.
Picard's theorem, as it is considerably less elementary than the theorem
of Casorati-Weierstrass, is ordinarily not proved in introductory courses.
The reader can find proofs in the books of Ahlfors, Burckel, Conway, and
Saks and Zygmund.

VIll.12. Residues
Let the holomorphic function f have an isolated singularity at the point zo
of C. The residue off at zo, denoted by resz 0 f, or by resz=zof(z), is defined
to be the coefficient of ( z - zo )- 1 in the Laurent expansion of f near zo. The
residues of a holomorphic function can be used in the evaluation of integrals
of the function. This is the content of the residue theorem, which will be
given in Chapter X. The simplest case appears already in Section VIII.2: if
C is a counterclockwise oriented circle centered at zo and contained with its
punctured interior in the region where f is defined, then

fc f(z)dz = 27ri resz 0 f.

VIII.12. Residues 107

Here are a few examples.

Example 1. The Laurent expansion of e 11z about 0 is I::~=O :;h z-n, giving
resz=oe 1 /z = 1.
z3
Example 2. f(z) = - - , zo = 1.
z-l
The Taylor expansion of z 3 about the point 1 is
z3 = 1 + 3 ( z - 1) + 3(z - 1) 2 + (z - 1) 3 .
z3
Hence, the Laurent expansion of - - about the point 1 is
z-l
z3 1
- - + 3 + 3(z - 1) + (z - 1) 2 ,
z-l z-l
giving
z3
resz=i - - 1.
z-l
z3
Example 3. J(z) = (z _ l) 2 , zo = 1.

The same calculation as used in the last example shows that the Laurent
z3
expansion of (z _ l) 2 about the point 1 is

z3 1 3
(z - 1) 2 (z - 1)2 +z- 1 + 3 + (z - 1),

giving
z3
resz=l (z _ l) 2 = 3.

Exercise* VIII.12.1. Let the functions g and h be holomorphic in an open

set containing the point zo, and assume that h has a simple zero at zo. Prove
that
g g(zo)
reSzo h = h'(zo).
Exercise VIll.12.2. Determine the residues of the following functions at
each of their isolated singularities in C.
zP
(a) - - (p, q positive integers),
1- zq
z5
(b) (z2 - 1)2'
cosz
(c) l+z+z 2 '
108 VIII. Laurent Series and Isolated Singularities

1
(d)
sinz
Exercise* VIIl.12.3. Let the function f be holomorphic in an open set
containing the point zo and have a zero at zo of order m. Prove that

resz0 ff' = m.
 Chapter IX

Cauchy's Theorem

A general version of Cauchy's theorem will be established in Section IX.10.

The necessary analytic machinery has already been developed. Still needed
are some topological preliminaries centering on the notion of winding num-
ber. A second version of Cauchy's theorem, involving homotopy, is given in
Section IX.14. This will entail some additional topological preliminaries.
At the end of the chapter an approximation theorem, Runge's theorem,
is established and used to give an alternative proof of Cauchy's theorem.
This material is not needed in the final chapter, Chapter X.

IX.1. Continuous Logarithms

Let ¢ be a continuous, nowhere vanishing, complex-valued function on the
subinterval [a, b] of R. Then there is a continuous function 'I/; on [a, b] such
that ¢ = e'l/J. The function 'I/; is unique to within addition of a constant
integer multiple of 2ni.
As for uniqueness, if 'l./;1 and 'l./;2 are two functions with the required
1
properties, then the function - . ('l./; 1 - 'l./;2) is continuous and integer valued,
2m
hence constant (since [a, b] is connected).
\Ve establish the existence of 'I/; first for the case where the range of ¢
lies in an open half-plane H bounded by a line through the origin. In that
case, as we know from Section IV.10, there is a branch l of log z in H. The
composite function 'I/; = lo¢ then has the required properties. Note that, by
suitably choosing the branch l of logz, we can arrange to make 'l./;(a) equal
to any preassigned value of log</>( a).

-109
110 IX. Cauchy's Theorem

To establish the existence of '!/; in the general case, we partition [a, b]

into subintervals [to, t1], [t1, t2], ... , [tp-1, tp], where a = to < t1 < · · · <
tp = b, such that the range of </> on each subinterval lies in an open half-
plane bounded by a line through the origin. This is easily done on the
basis of the continuity of </> and the compactness of [a, b]. By the special
case already treated, there is a continuous function 'l/; 1 on [to, t 1] such that
</> = e1/J 1 on that interval. By the same special case, there is a continuous
function 'l/;2 on [t1, t2] such that </> = e1/J 2 on that interval, and such that
'l/;2(t1) = 'l/;1(t1). Continuing in this way, we obtain on each interval [tj-1, tj]
a continuous function 'l/;j such that</> = e1/Jj there, and 'l/;j (tj-l) = 'l/Jj-l (tj_i).
The function'!/; on [a, b] that equals 'l/;j on [tj-l, tj] (j = 1, ... ,p) then has
the required properties.

IX.2. Piecewise C 1 Case

Assume that the function </> in IX.1 is piecewise C 1 . Let c be any value of
log</>( a). Then the function '!/; defined by

'l/;(t) = c + it </>' ( s)
a </>(s) ds, a:::; t :::; b,

has the required properties: it is continuous and </> = e1/J.

The argument is standard. First, we may as well assume</> is of class C 1 ,
because the piecewise-C 1 case is an easy consequence of that one. If </> is of
class C 1 ' then the function '!/; defined above is differentiable, with '!/;' = ~.
Then
(</>e-1/J)' = </>'e-1/J - </>'l/;'e-1/J = 0,

implying that the function <f>e-1/J is constant. Since </>(a) = e1/J(a), the constant
is 1, and the proof is complete.

IX.3. Increments in the Logarithm and Argument Along a

Curve
Let 'Y : [a, b] --+ C be a curve in C, and let f be a continuous complex-
valued function that is defined and nowhere vanishing on 'Y· By IX.I there
is a continuous function'!/; on [a, b], unique to within addition of a constant
integer multiple of 2ni, such that f o 'Y = e1/J. The increment in log f on
"(, denoted D.(logf,"f), is defined to be the difference 'l/;(b) - 'l/;(a). The
increment in argf on"(, denoted D.(argf,"f), is defined to be ImD.(logf,"f).
In the special case where 'Y is a closed curve, the real part of D.(log f, 'Y)
vanishes (since all logarithms of the same number have the same real part),
IX.5. Case of a Piecewise-C 1 Curve 111

1
so ~(argf,"f) = --:-~(logf,"f). Moreover, if"( is closed then ~(argf,"f) is an
i
integer multiple of 21r.
The intuition is as follows. One can imagine a perceptive point moving
along 'Y from 'Y(a) to "f(b), and, as it proceeds, keeping track in a continuous
way of a value of log f(z). The difference between the terminal and initial
values perceived by the point is the increment in log f along 'Y·
If the curve 'Y is piecewise C 1 and the function f is holomorphic in an
open set containing 'Y (and nonvanishing on "f), then by IX.2 we have

~(log f, "f) l a
b f'('Y(t))'Y'(t) d
J('Y(t)) t,
which can be rewritten

~(log f, 'Y) = 1'Y

f'(z)
f(z) dz.

If in addition 'Y is a closed curve, then the real part of ~(log f, 'Y) is 0, and
we have
~ (arg f, "() =
11
i 'Y
f'(z)
f (z) dz.

IX.4. Winding Number

Let 'Y : [a, b] ---+ C be a closed curve in C, and let zo be a point in the
complement of 'Y· The function f(z) = z - zo is then nonvanishing on
"(, so the increment ~(arg(z - zo), 'Y) is defined by the discussion in the
preceding section. That increment is an integer multiple of 21r. The integer
1
-~(arg(z - zo), 'Y) is called the winding number of "( about zo, or the
21f
index of zo with respect to "(, and denoted by ind-y(zo). If one imagines
an observer stationed at zo watching a point move once around "(, and
continuously turning so as always to face the point, then ind-y(zo) will equal
the net number of revolutions performed by the observer, counterclockwise
revolutions being reckoned as positive and clockwise revolutions as negative.
The winding number of a curve about a point, at least in simple cases, can
be read off from a picture of the curve. As we shall see shortly, it takes
a constant value on each connected component of the complement of the
curve. In the examples in Figure 6, the winding number is indicated for
each complementary component.

IX.5. Case of a Piecewise-C 1 Curve

Let 'Y: [a, b] ---+ C be a piecewise-C 1 closed curve in C, and let zo be a point
in the complement of 'Y· Applying the formula at the end of Section IX.3 to
112 IX. Cauchy's Theorem

-1 0
2

1 0

Figure 6. Winding Numbers

the function f(z) = z - zo, we obtain an integral expression for the winding
number of I about zo:

ind,,(zo) = -21 .
1ri
J ')'
- -1d z .
z - zo
The quantity on the right side of the preceding equality is easily seen
to be a continuous function of zo on C\1. (In fact, by what is proved in
Section VII.7, it is a holomorphic function of zo.) Since it is also integer
valued, it must be constant on each connected component of C\1. Because I
is bounded, only one of the components of C\1 is unbounded. As zo-+ oo,
the integral in the expression above for ind,,(zo) tends to 0 (by standard
estimates; see Section VI.10), so also ind,,(zo) -+ 0. Since ind,,(zo) is an
integer, it must actually equal 0 when zo is sufficiently large. Since it takes
a constant value on each component of C\1, it therefore equals 0 on the
unbounded component of C\ I·
Exercise IX.5.1. Prove that the properties of winding number established
in the preceding paragraph for piecewise-C 1 curves hold for general curves:
if I : [a, b] -+ C is a closed curve in C, then ind,,(zo) is constant on each
component of C\1 and 0 on the unbounded component.
IX.6. Contours 113

Exercise IX.5.2. Prove the residue theorem for rational functions: Let f
be a rational function, and let z1, ... , Zp be the poles off in C. Let"( be a
piecewise-C 1 closed curve that does not pass through any of the poles of f.
Then

-1.
2ni
1 I'
J(z)dz = '°'
p
L....t
j=l
ind,,(zj) · resz1 f.

Exercise IX.5.3. Prove that if zo is a point in C\R, then

lim - 1
2ni
R-->oo
JR -t --1 zod t
-R
1 1 1
is 2 if Im zo > 0 and - 2 if Im zo <
0. (Intuitively, the real axis winds 2
1
time around each point in the upper half-plane and - 2 time around each
point in the lower half-plane.)

IX.6. Contours
The term "contour" is commonly used in an informal way in complex func-
tion theory to refer to a closed integration path. Here we shall assign the
term a more specific meaning. By a contour we shall mean a formal sum
r = :L~= 1 nj"/j, where "(1, ... , "fp are piecewise c 1 closed curves in c and
n1, ... , np are integers. If f is a continuous complex-valued function defined
on each "/j-or, as we shall say, defined on r-we define the complex integral
off over r by

1
r
f(z)dz = L nj 1f(z)dz.
p

j=l l'j

We shall identify a piecewise C 1 closed curve 'Y with the contour l "f.
In general, we shall identify two contours if they are indistinguishable for
purposes of integration. Thus, for example, two piecewise C 1 closed curves
will be identified if one of them is a reparametrization of the other (some-
thing we have been doing all along). If 'Y is a piecewise C 1 closed curve, the
contour -1 "(will be identified with -"(, the reverse of"( (see Section (VI.9)).
If"( is defined on the parameter interval [a, b], then the contour 2"( can be
identified with the curve on the parameter interval [a, 2b - a] that takes the
value "((t) at a point t in [a, b] and the value 'Y(t - b +a) at a point t in
[b, 2b - a].
If r = :L~= 1 nnj and r' = :L~= 1 nj'Yj are two contours, we define their
sum by
p

r + r' = L (nj + njhj·

j=l
114 IX. Cauchy's Theorem

Then, if f is a continuous function defined on every "{j, we have

r
lr+r'
J(z)dz = rJ(z)dz + lrr J(z)dz.
lr 1

The algebraic structure associated with contours will not be formalized

here. With sufficient effort one can define an equivalence relation on the
family of contours and make the set of equivalence classes into an abelian
group. To carry out this program here would, however, be a digression with
little bearing on our central purpose.

IX.7. Winding Numbers of Contours

Let r = L::~=l nj"{j be a contour, and assume that each nj is nonzero. By
the complement of r, written C\r, we shall mean the set of points of C
that are in the complement of each 'Yj· (As with curves, we shall, when
convenient, speak of contours as if they were subsets of C. This will cause
no confusion in practice.) If zo is such a point, we define the winding number
of r about zo, or the index of zo with respect to r' by
p

indr(zo) = L nj ind'"Yi (zo).

j=l

As in the case of a single closed curve, we have the integral formula

~
Jrr-z - -dz.
indr(zo) = 1
2m zo
The reasoning used in Section IX.5 for single curves applies without change
to contours; it shows that indr(zo) is constant on each connected component
of C\r and equals 0 on the unbounded component.
As a simple example, let C1 be the circle lzl = 1 and C2 the circle lzl = 2,
both oriented counterclockwise. For the contour r = C 2 - C 1 we have

. {1 for 1 < lzol < 2

mdr(zo) = 0
for lzol < 1 and for lzol > 2.

We define the interior of the contour r, denoted int r, to be the set of

points zo in C\r such that indr(zo) -1- 0, and the exterior, denoted ext r, to
be the set of points zo in C\r such that indr(z0 ) = 0. Each of these sets is a
union of certain of the connected components of C\r. As those components
are open sets, the sets int I' and ext I' are open. Because the unbounded
component of C\I' is contained in ext r, the set int r is bounded. All
boundary points of int r and of ext r are contained in r.
A contour r will be called simple if indr(z0 ) is either 0 or 1 for each
point z0 in C\r.
IX.8. Separation Lemma 115

IX.8. Separation Lemma

Let G be an open subset of C and K a compact subset of G. Then there is
r
a simple contour in G\K such that Kc int c G. r
In other words, K can be separated from the complement of G by a sim-
ple contour. This is our main topological preliminary to Cauchy's theorem.
We shall prove it by forming a grid in the plane and building the required
contour out of selected segments of the grid.
Let J be a positive number less than dist(K, C\G) (= inf{lz - wl :
z E K, w E C\G} ). By means of equally spaced horizontal and vertical
lines, we subdivide C into nonoverlapping squares each of diameter J. By a
square here we mean a solid square, consisting of both boundary and interior.
The term "nonoverlapping" expresses the fact that distinct squares in the
subdivision have disjoint interiors. (Two distinct squares, if not disjoint,
share at most a common edge or a common vertex.)
Let S1, S2, ... , Sq be an enumeration of those squares in the subdivision
that intersect K. The collection {S1, S2, ... , Sq} will be denoted by S. By
the choice of J, each square in S is contained in G. For j = 1, ... , q we let
CTj denote the counterclockwise oriented boundary of the square Sj.

Each square in our subdivision has four edges. We let E be the collection
of those edges that belong to only one square in the collection S. Each edge
in E is then contained in G but disjoint from K. We give each edge in E
the orientation that it inherits from the boundary of the square in S that
contains it. We shall refer to an edge that belongs to two squares in S as
an interior edge.
We shall show first that the edges in E can be organized into a contour,
in other words, that there is a contour r such that

L 1f(z)dz
EEE E
£ f(z)dz

for any function f that is continuous on each edge in E. Subsequently we

shall show that r possesses the other required properties.
A sequence (E1, E2, ... , Ep) of edges in £, without repetitions, will be
called a chain if the terminal point of Ek-l coincides with the initial point
of Ek ( k = 2, ... , p). Such a chain will be called a cycle if, in addition, the
terminal point of Ep coincides with the initial point of Ei. If (E1, E2, ... , Ep)
is a cycle, we can build from it a piecewise-C 1 closed curve"! in an obvious
way: for k = 1, ... , p we parametrize Ek (linearly, say) by means of a func-
tion "/k on the interval [k - 1, k], and then we let"( be the function on [O,p]
116 IX. Cauchy's Theorem

C\G

Figure 7. Construction of r in the proof of the separation lemma. The

set K in this example has four components.

that coincides with '/'k on [k - 1, k] for each k. With 'Y so defined, we have

~
p

hk J(z)dz = 1 f(z)dz

for any continuous function f defined on each Ek. To construct the desired
contour r, therefore, it will suffice to show that we can decompose£ into a
disjoint union of cycles.
To accomplish the latter we note the following property of£, which one
can verify by examining all possible cases:
( *) Each vertex in our subdivision is the initial point of the same num-
ber of edges in £ as it is the terminal point of. (The number is
either 0, 1, or 2.)
From ( *) we can conclude that a maximal chain in £, that is, a chain we
cannot lengthen by appending an edge from£ at its beginning or end, must
in fact be a cycle. Since £, being finite, obviously contains maximal chains,
it contains cycles. If we remove a cycle from£ we obtain a smaller collection
of edges which will still satisfy (*). If the smaller collection is nonempty we
can extract a cycle from it, and so on. After finitely many steps we obtain
in this way the desired decomposition of £ into a disjoint union of cycles,
completing the construction of the contour r.
It remains to show that r has the required properties. That r is con-
tained in G\K is obvious. We need to show that r is simple and that
Kc intr c G.
IX.9. Addendum to the Separation Lemma 117

Suppose zo is a point in the interior of one of the squares in S, say the

square Sjo. Then

_1
27ri
1_l_dz = {1,0,
aj Z - Zo
J =Jo
J =!=Jo.
It follows that

-.L:
1
27ri
1- -1d z =
q
1.
j=l a· Z -
J
Zo

The left side in the preceding equality equals ~

{ - 1-dz. To see this,
2m Jr z - zo
note that in the sum of integrals that appears, each of the interior edges (that
is, the edges belonging to two of the squares in S) makes two contributions
which cancel (since the orientations such an edge inherits from the two
squares containing it are opposite), leaving only the contributions from the
edges in £. We thus have

_1_
27ri
r_l_dz
lr z -
zo
= 1

whenever zo is a point in the interior of one of the squares in S. The

expression on the left side is a continuous function of zo on the complement
of r, so the equality continues to hold if zo is a point off r that lies on one
of the interior edges. In particular, we have indr(zo) = 1 for all zo in K.
On the other hand, if the point zo is not in any of the squares in S, then

1-
aj zo
Z -
1-dz = 0 for every j, and, reasoning as above, we can conclude that

indr(zo) = 0. This applies, in particular, for zo in C\G.

We have thus shown that indr(zo) takes only the values 0 and 1, that it
is 1 for zo in K, and that it is 0 for zo in C\G. The proof of the lemma is
complete.

IX.9. Addendum to the Separation Lemma

Let r be the contour constructed above, and let f be a holomorphic function
in G. Then
zo EK.

This special case of Cauchy's formula will be used to deduce the general
version of Cauchy's theorem in the next section. To prove it we note first
that fa f(z)dz = 0 for each J, by Cauchy's theorem for a square (an immedi-
1
ate corollary of Cauchy's theorem for a triangle). Since I::J=l faj f(z)dz =
118 IX. Cauchy's Theorem

fr J(z)dz (by the same reasoning as used in the proof of the separation
lemma), we have fr f(z)dz = 0.
Now fix a point zo in K, and define the function g in G by

J(z) - f(zo)
g(z) = { z-zo '
z #- zo
f' (zo), z = zo.
Then g is holomorphic, so, by what was just proved, fr g(z)dz = 0. Conse-
quently

i -f(z)
-dz
r z - z0
= f(zo) i - -1 d z =
r z - z0
.
27rif(zo),

as desired.

IX.10. Cauchy's Theorem

Let G be an open subset of C, and let r be a contour contained with its
interior in G. Then fr f(z)dz = 0 for every function f that is holomorphic
in G.
In order to prove this, let K be the complement in C of the exterior of
r. Because ext r is an open set, the set K is closed. Also C\K contains
C\G as well as the unbounded component of C\r, whose complement is
bounded (by the boundedness of r). It follows that K is a compact subset
of G. Obviously r c K.
By the separation lemma and its addendum, there is a simple contour
r1 in G\K such that K C int r1 C G, and such that, for any function f
that is holomorphic in G,

z EK.

For such a function f we thus have

rJ(z)dz
Jr
= ~
2m
r r !(()zd( dz.
Jr lr 1 -,, -

In the double integral on the right, it is permissible to change the order

of integration because the integrand is a continuous function of the pair of
variables z, (. (More precisely, if one writes out the double contour integral
in terms of real parametrizing variables for the constituent curves of and r
r i, one finds that the integrands that occur are continuous functions of the
IX.12. Continuous Logarithms-2-D Version 119

parameters on the appropriate rectangles in R 2 .) Hence

rf(z)dz =
lr mlrr [!(() lrr~dz]
_2I.
1 ., - z
d(

= - r f(() indr(()d(.
lr1
The right side is 0 because r 1 is contained in the complement of K, so that
indr(() = 0 for all (on f1. This completes the proof of Cauchy's theorem.

Exercise IX.10.1. Justify in detail the change in order of integration in

the preceding proof.

Exercise IX.10.2. Explain why VII.2, Cauchy's theorem for a convex re-
gion, is a special case of IX.10, the general Cauchy theorem. (It is not a bona
fide corollary because it was used in the developments leading to IX.10.)

IX.11. Homotopy
Let G be an open subset of C. Two closed curves /'o : [O, I] ---+ G and
/'1 : [O, I] ---+ G are said to be homotopic in G if there is a continuous map
/' : [O, I] x [O, I] ---+ G satisfying
(i) l'(t, 0) = /'o(t) for all t,
(ii) 1(t, I)= 1'1(t) for all t,
(iii) 1(0, s) = 1(I, s) for alls.
Intuitively, the curves l's, 0 ::::; s ::::; I, defined by "fs(t) = 1(t, s), provide a
continuous deformation within G from "'(o to l'l · It is clear that "'(o and "fl
can only be homotopic in G if they lie in the same connected component of
G.
It will be shown that if "fo and "fl are homotopic in G then they have
the same winding number about every point in C\G. This will depend on
a two-dimensional version of the result on continuous logarithms in Section
IX.I.

IX.12. Continuous Logarithms-2-D Version

Let cf> : [O, I] x [O, I] ---+ C\{O} be a continuous function. Then there is a
continuous function 'ljJ : [O, I] x [O, I] ---+ C such that </> = e..P. The function
'ljJ is unique to within addition of an integer multiple of 27ri.
This will be deduced from IX.I, its one-dimensional counterpart. The
statement about uniqueness follows in the same way as did the corresponding
statement in IX. I.
120 IX. Cauchy's Theorem

Define ¢00 : [O, 1] ---+ C\ {O} by ¢oo(t) = ¢(t, 0). For each t in [O, 1],
define <Pt: [O, 1]---+ C\{O} by <Pt(s) = ¢(t, s). By IX.1, there is a continuous
function 7/Joo : [O, 1] ---+ C such that ¢00 = e1/Joo. By IX.l, again, there is for
each t in [O, 1] a continuous function 7./Jt : [O, 1] ---+ C such that <Pt = e1/Jt and
7./Jt(O) = 7/Joo(t). Define 7/J: [O, 1] x [O, 1] ---+ C by 'lj;(t, s) = 7/Jt(s). The equality
¢ = e1/J is immediate. It remains to prove that ·l/J is continuous.
The range of ¢ is a compact subset of C\ {O} and so has a positive
distance from 0. Choose E > 0 such that 1¢(t, s)I > E for all (t, s). Fix
to in [O, 1]. By the uniform continuity of ¢, there is a <5 > 0 such that
1¢(t, s) - ¢(to, s)I < E for alls whenever It - tol ::::; <5. Thus, for It - tol ::::; <5
we have
¢(t, s) - 11 = 1 1¢(t, s) - ¢(to, s)I < 1¢(t, s) - ¢(to, s)I < 1,
I¢(to, s) ¢(to, s) E

implying that
R ¢(t, s) 0
( s) > .
e ¢to,
Define the function x on the rectangle
R = ([to - <5, t 0 + <5] n [O, ll) x [O, 1]

by x(t, s) = Log:(~;~J). Then xis continuous, ¢(t, s)/¢(t 0, s) = ex(t,s) on R,

and x(to, s) = 1 for alls. The function ;j; on R defined by 0(t, s) = x(t, s) +
¢(to, s) is continuous and satisfies¢= E;j; on R. Also ;j;(to, 0) = 'lj;(to, 0). On
the interval [to-'5, to+<5]n[O, 1], the functions t f---; 7/J(t, 0) and t f---; {f;(t, 0) are
continuous logarithms of the same function and they agree fort = to, so they
coincide. For each tin [to - <5, to+ <5] n [O, 1], then, the functions sf---; 'lj;(t, s)
ands f---; ;j;(t, s) on [O, 1] are continuous logarithms of the same function that
agree at s = 0, so the_y coincide. We see therefore that 'ljJ coincides with the
continuous function 'ljJ on R; in particular, 'ljJ is continuous at each point of
{to} x [O, 1]. Our proof is complete.

IX.13. Homotopy and Winding Numbers

Let G be an open subset of C. Let the closed curves /'O and /'1 in G be
homo topic in G. Then ind,,0 ( z) = ind,,1 ( z) for all z in C\ G.
To establish this, take I' : [O, 1] x [O, 1] ---+ G as in the definition in
IX.11, and fix a point zo in C\G. Define the function ¢ : [O, 1] x [O, 1] ---+
C\{O} by ¢(t, s) = !'(t, s) - zo. By IX.12 there is a continuous function
7/J: [O, 1] x [O, 1] ---+ C such that¢= e1/J. Then for each sin [O, 1] we have

ind,,s (zo) = -2 1 . ( 'lj;(l, s) - 7ji(O, s)),

7ri
IX.15. Runge's Approximation Theorem 121

showing that the function s f-+ ind,,s ( zo) is continuous, hence constant. In
particular, ind,,0 (zo) = ind,,1 (zo), as desired.

IX.14. Homotopy Version of Cauchy's Theorem

Let the function f be holomorphic in the open set G, and let 'Yo and 11
be two piecewise-C 1 closed curves in G that are homotopic in G. Then
J,,
0 f(z)dz = J,,
1 f(z)dz.

This follows from IX.10 because, by IX.13, the contour 11 - 'YO has
winding number 0 about all points of C\G.

IX.15. Runge's Approximation Theorem

Let K be a compact subset of C, and let f be a holomorphic function defined
in an open set containing K. Then f can be uniformly approximated on K
by rational functions whose poles lie in C\K.
The last statement says that, for any positive number E, there is a ra-
tional function g whose poles lie in C\K such that lf(z) - g(z)I < E for
all z in K. An equivalent statement is that there exists a sequence of such
rational functions that converges uniformly on K to f. Runge's theorem
was published in 1885 (as was the more famous approximation theorem of
K. Weierstrass).
To prove Runge's theorem we first note that the family of rational func-
tions whose poles lie in any preassigned set is closed under addition: if two
functions belong to the family, then so does their sum. It follows that the
family of functions on K that can be uniformly approximated (on K) by
such rational functions is also closed under addition.
Let the function f be holomorphic in the open set G containing K. Let
r be the simple contour in G\K constructed in the proof of the separation
lemma. By the addendum to the separation lemma,

f(z) = ~
2m
r f(()
Jr ( - z
d(, z EK.

We created r by piecing together certain directed edges produced by a sub-

division of the plane into squares. Let E1, E2, ... , E 8 be an enumeration of
those edges. We then have f = 'L:j= 1 fj, where

z EK.

By the remark in the preceding paragraph, it will suffice to show that each
fj can be uniformly approximated on K by rational functions with poles in
C\K.
122 IX. Cauchy's Theorem

Fix an index j. Let the segment Ej be partitioned into subsegments

each having a length less than the distance between Ej and K. There is a
corresponding expression for fj as a sum of functions, each summand being
1
- . times the Cauchy integral of f over one of the subsegments. By the
2m
same reasoning as was just used, it will suffice to show that each summand in
that expression can be uniformly approximated on K by rational functions
with poles in C\K. A typical summand has the form

h(z) = ~
27ri
r !(() d(,
}E ( - Z
where E is the subsegment in question. Thus, the length of E is less than
the distance between E and K. Denote the preceding length by 8, and
let zo be the midpoint of E. Then, as proved in Section VIII.3, the func-
tion h is represented in the region lz - zol > 8/2 by the Laurent series
2::~~-oo an(z - zo)n, where an =-l. { (( - zo)-n-l J(()d(. The series
2m}E
converges locally uniformly in the preceding region, which contains the set
K. The series therefore converges uniformly on K. As the partial sums of
the series are rational functions whose only pole is zo, this completes the
proof of Runge's theorem.

IX.16. Second Proof of Cauchy's Theorem

With Runge's theorem one can give an alternative proof of Cauchy's theo-
rem. Let G, r, and f be as in Section IX.11, namely, G is an open subset of
C, r is a contour contained with its interior in G, and f is a holomorphic
function in G. As in the earlier proof, let K be the complement in C of the
exterior of r. Then K is a compact subset of G, and r c K.
We first note that fr g(z)dz = 0 whenever g is a rational function whose
poles lie in C\K. In fact, any such g is, to within addition of a constant,
a linear combination of functions of the form (z - zo )-n, where zo is in
C\K and n is a positive integer. (See Exercise VIII.7.4.) If n > 1 we have
fr (z - zo)-ndz = 0 because the integrand has a primitive, and if n = 1
the same conclusion holds because all points having a nonzero index with
respect to r lie in K.
By Runge's theorem, there is a sequence (gk)k=l of rational functions
with poles in C\K that converges uniformly to f on K, and so in particular
on r. Then
rf(z)dz
Jr
= lim
k-->oo
r9k(z)dz.
Jr
Since, as shown above, fr 9k(z)dz = 0 for each k, it follows that fr f(z)dz =
0, as desired.
IX.17. Sharpened Form of Runge's Theorem 123

IX.17. Sharpened Form of Runge's Theorem

Let K be a compact subset of C, and let S be a subset of C\K that contains
at least one point in each connected component of C\K. Then any function
holomorphic in an open set containing K can be uniformly approximated on
K by rational functions whose poles lie in S.
In particular, if C\K is connected we can take S to be the singleton
{ oo}, giving in this case the conclusion that any function holomorphic in an
open set containing K can be uniformly approximated on K by polynomials.
To prove the sharpened form of Runge's theorem, it will suffice to show
that any rational function whose poles lie in C\K can be uniformly ap-
proximated on K by rational functions whose poles lie in S. Because any
rational function can be written as a sum of rational functions each having
only one pole (by Exercise VIII.7.4), the desired conclusion is a consequence
of the following assertion: If the points zo of C and zb of C lie in the same
connected component of C\K, then any rational function whose only pole
is zo can be uniformly approximated on K by rational functions whose only
. I
po1e is z0 .
We treat first the case zb # oo. In that case zo and zb lie in the same
connected component of C\K, so there is a polygonal path '"Y in that com-
ponent whose initial point is zo and whose terminal point is zb. (We are
using here a basic property of connected open subsets of the plane. The
reader unfamiliar with it can establish it as Exercise IX.17.1 below.) On '"Y
we can find a finite sequence of points zo, z1, ... , Zq = zb such that, for each
j, the distance IZj-I - Zj I is less than the distance from '"'( to K. Clearly, it
will suffice to show that, for each j, any rational function whose only pole is
Zj-I can be uniformly approximated on K by rational functions whose only
pole is Zj. But a rational function whose only pole is Zj-1 has a Laurent
series representation with center Zj that converges locally uniformly to the
function in the region lz - ZJI > lzJ-1 - ZJI, a region containing K (since
lzJ-1 - ZJI <dist('"'(, K) :S dist(zj, K)). Because the given rational function
has a removable singularity at oo, the Laurent series contains no positive
powers of z-zj. Its partial sums are therefore rational functions having Zj as
their only pole, and they converge uniformly to the given rational function
on K, as desired.
It remains to treat the case zb = oo. In that case zo lies in the unbounded
component of C\K. By what has just been proved, we can then assume with
no loss of generality that K is contained in the disk lzl < lzol- In that case,
if a rational function has zo as its only pole, its Taylor series centered at
0 converges uniformly to it on K, and thus the function can be uniformly
approximated on K by polynomials. This completes the proof.
124 IX. Cauchy's Theorem

Exercise IX.1 7 .1. Prove that if two points lie in the same connected open
subset G of C, then the points can be joined by a polygonal path that lies
in G.
Exercise IX.17.2. Let G be an open subset of C and Sa subset of C\G
that contains at least one point in each connected component of C\G. Let
f be a holomorphic function in G. Prove that there is a sequence of rational
functions whose poles lie in S that converges locally uniformly to f in G.
Exercise IX.17.3. Prove that there is a sequence (Pk)k=l of polynomials
such that
1, Re z > 0
lim Pk(z) = { 0, Re z = O
k-+=
-1, Re z < 0.
 Chapter X

Further Development
of Basic Complex
Function Theory

The chapter begins with a topological notion, that of a simply connected

domain, followed by several characterizations of simply connected domains.
The term "domain" for us will mean a nonempty, connected, open subset of
C, a usage common in complex analysis. The notion of simple connectivity
is a general topological one. The definition used here for plane domains,
while arguably the most convenient for complex analysis, differs from the
general topological definition, which involves homotopy. The equivalence of
the two definitions will be established at the end of the chapter.
Next, various consequences of Cauchy's theorem will be obtained, no-
tably, the residue theorem together with examples of its use in evaluating
definite integrals, and the argument principle, which gives an integral ex-
pression for the number of zeros of a holomorphic function in a region. The
chapter culminates with a proof of the famous Riemann mapping theorem.

X.1. Simply Connected Domains

As noted above, the term "domain" is commonly used in complex function
theory to designate a nonempty, connected, open set. We shall henceforth
use the term in this sense. A domain G in C is said to be simply connected
if C\G, its extended complement, is connected.

-
125
126 X. Further Development of Basic Complex Function Theory

The reader will easily verify that every convex domain-in particular,
every open disk and every open half-plane-is simply connected. The sim-
plest example of a domain that is not simply connected is C\ {0}. A bounded
domain G is simply connected if and only if C\G, its ordinary complement,
is connected. However, the ordinary complement of an unbounded simply
connected domain can fail to be connected. The strip 0 < Im z < 1 furnishes
a simple example.

X.2. Winding Number Criterion

For the domain G in C to be simply connected, it is necessary and sufficient
that every contour in G have winding number 0 about every point in C\ G.
To establish the necessity, assume G is simply connected, and let r be a
contour in G. Since C\G is connected and contained in C\r, it is contained
in a single connected component of C\r, necessarily the unbounded compo-
nent. Hence C\G is contained in the unbounded component of C\r. Since
r has winding number 0 about every point in the unbounded component of
C\r, the desired conclusion follows.
For the other direction, assume that G is not simply connected. Then
there is a decomposition C\G =KU L, where Kand Lare closed, disjoint,
and nonempty, say with oo in L. Let H = C\L. Then His an open subset
of C, and H contains K. By the separation lemma IX.8, there is a simple
contour in H\K ( = G) that has winding number 1 about each point of K.
This establishes the sufficiency of the condition.
Exercise X.2.1. Let G be an open subset of C. Prove that C\G is con-
nected if and only if every connected component of G is simply connected.
Exercise X.2.2. Let ( Gn)~=l be a decreasing sequence of simply connected
domains such that the interior of n! Gn is nonempty. ("Decreasing" means
Gn+l C Gn for all n.) Prove that every connected component of the interior
of n! Gn is simply connected.
Exercise X.2.3. Prove that the union of an increasing sequence of simply
connected domains is simply connected.

X.3. Cauchy's Theorem for Simply Connected Domains

Cauchy's theorem states that a holomorphic function in an open set G has
integral 0 around any contour in G that has winding number 0 about every
point of C\G. By X.2, if G is a simply connected domain, the preceding
winding number condition holds automatically. Hence: If f is a holomorphic
function in the simply connected domain G, then fr f(z)dz = 0 for every
contour r in G.
X.5. Existence of Logarithms 127

X.4. Existence of Primitives

For the domain G in C to be simply connected, it is necessary and sufficient
that every holomorphic function in G have a primitive.
That holomorphic functions in convex domains have primitives was
proved, in Section VII.2, as the main step in the proof of Cauchy's the-
orem for such domains. The argument depended on Cauchy's theorem for
a triangle. Now that we have the general Cauchy theorem, we can use the
same argument in a general simply connected domain: Suppose G is simply
connected, and let f be a holomorphic function in G. Fix a point z0 in G,
J
and define the function gin G by g(z) = 1 f(()d(, where 'Y is any piecewise-
C1 curve in G that starts at zo and ends at z. Cauchy's theorem implies
that the definition of g makes sense; in other words, the integral above has
the same value for any two curves 'Y satisfying the stated conditions. One
can now prove exactly as in Section VII.2 that g is holomorphic and g' = f.
This establishes the necessity of the condition.
For the other direction, assume that G is not simply connected. Then,
by X.2, there is a contour r in G such that f - 1-dz-=/:- O for some point
Jr z - zo
1-
zo in C\G. The function - - then has no primitive in G. This establishes
z - zo
the sufficiency of the condition.

Exercise X.4.1. Let G be a simply connected domain, f a holomorphic

function in G, and ca positive number. Prove that each connected compo-
nent of the set {z E G: lf(z)I < c} is simply connected.

X.5. Existence of Logarithms

Let G be a simply connected domain and f a nowhere vanishing holomorphic
function in G. Then there is a branch oflog f in G.
To prove this, fix a point zo in G, and let a be any value of log f (zo). By
X.4, the function j has a primitive, g, in G, and we may obviously assume
that g(zo) = a. We have
(f e- 9 )' = J' e- 9 - g' fe- 9 = 0,

implying that e9 is a constant times f. Since f and e9 have the same value
at zo, the constant is 1. Thus g is a branch of log f in G, as desired.

Exercise X.5.1. Prove the converse of X.5: If the domain Gin Chas the
property that, for every nowhere vanishing holomorphic function f in G,
there is a branch of log f in G, then G is simply connected.
128 X. Further Development of Basic Complex Function Theory

Exercise X.5.2. Let G be a domain in C and fa nowhere vanishing holo-

morphic function in G. Prove that there is a branch of log f in G if and
only if there is a branch of f l/n in G for every positive integer n.

X.6. Existence of Harmonic Conjugates

Let G be a simply connected domain in C and u a real-valued harmonic
function in G. Then u has a harmonic conjugate in G, unique to within
addition of a real constant.
This is proved for convex domains in Section VII.18. The argument
there can now be applied word-for-word to the more general case of a simply
connected domain.

Exercise X.6.1. Let the domain G in C be multiply connected (i.e., not

simply connected). Prove that there is a real-valued harmonic function u in
G that has no harmonic conjugate in G.

X. 7. Simple Connectivity and Homotopy

If G is a domain in C, and if every closed curve in G is null homotopic in
G, then G is simply connected.
A closed curve in G is said to be null homotopic in G if it is homotopic
in G to a constant curve, i.e., a curve whose range is a singleton. In that
case one says the curve can be shrunk in G to a point.
To prove the statement we note that, if G has the given property, then
by IX.13 every closed curve in G has winding number 0 about every point
in C\G, which by X.2 implies that G is simply connected.
Our definition of simple connectivity for a domain G in the plane is
deficient from a topological perspective because it is not intrinsic to G: it
refers to the complement C\G. Not apparent from our definition is whether
simple connectivity is a topological property of a domain. Put in other
terms, if the domains G1 and G2 are homeomorphic (that is, in one-to-one
correspondence under a bicontinuous map), and if G1 is simply connected,
must G2 be simply connected?
According to the general definition, a topological space is called simply
connected if it is pathwise connected (any two points in the space can be
joined by a curve in the space) and every closed curve in the space is null ho-
motopic. Simple connectivity in this sense is clearly a topological property.
According to the proposition at the beginning of this section, for a domain
in C, simple connectivity in the homotopic sense implies simple connectiv-
ity in our sense. The converse implication is given in Section X.21. Thus,
X.8. The Residue Theorem 129

the definition of simple connectivity we use is equivalent to the standard

definition, in the context of plane domains.
The reader wishing to learn more about these matters is referred to the
books of R. B. Burckel and M. H. A. Newman in the reference list.

X.8. The Residue Theorem

Let G be a domain in C and r a contour contained with its interior in G.
Let the function f be holomorphic in G except for isolated singularities at
the points z1, ... 'Zp, none of which lies on r. Then
p
r f(z)dz = 27ri2:indr(zk)·reSzk(f).
lr k=l

To prove this we choose, for each singularity zk, a counterclockwise ori-

ented circle ck with center Zk, contained along with its interior in G, and
disjoint along with its interior from r and from all of the other circles. Let
nk = indr(zk), k = 1, ... ,p. We consider the domain

and the contour

p

f1 = r - LnkCk.
k=l
The function f is holomorphic in G1, and the contour f1 lies in G 1 and
has winding number 0 about each point in C\G1. By Cauchy's theorem,
fr1 f(z)dz = 0, which can be written

rf(z)dz t
lr
=
k=l
nk r f(z)dz.
lck
Since fck f(z)dz = 27ri reszk(f) for each k, the desired equality follows.
Exercise X.8.1. Let C be the unit circle with the counterclockwise orien-
tation, and let a and b be points in the interior of C. Evaluate the integral
r zk
Jc (z - a)(z - b) 2 dz,
where k is a positive integer.

Exercise X.8.2. Let p and q be polynomials such that deg q > 1 + deg p.
Prove that the sum of the residues of the rational function !!., taken over all
q
of its poles in C, is 0.
130 X. Further Development of Basic Complex Function Theory

Exercise X.8.3. Evaluate the integral

fc (z - 2)- 1 (2z + 1)- 2(3z - 1)- 3 dz,

where C is the unit circle with the counterclockwise orientation. (Suggestion:

Try to use the preceding exercise.)

Exercise X.8.4. Evaluate the integrals

r
le 2z 2 -
sinz d re 2z2 - 5z + 2 dz
5z + 2 z, Jr smz
where C is the unit circle with the counterclockwise orientation.

Exercise X.8.5. Extend the residue theorem to the case where the function
f has countably many isolated singularities in the domain G.

X.9. Cauchy's Formula

Let G be a domain in C, f a holomorphic function in G, and r a simple
contour contained with its interior in G. Then

J(zo) = ~
2m
r f(z) dz,
Jr z - zo
zo E int r.

This is an immediate consequence of the residue theorem: under the

given hypotheses, the integrand in the preceding integral is holomorphic in
G\{zo}, and its residue at zo is f(zo).
Cauchy's formula for the n-th derivative (n = 1, 2, 3, ... ) is a conse-
quence: under the given hypothesis:

f(n)(zo) = n!
27ri
r
Jr (z -
f(z) dz
zo)n+l '
zo E int r.

The deduction is contained in the discussion of Cauchy integrals in Section

VII.7.

X.10. More Definite Integrals

The exploitation of complex integration to evaluate improper Riemann in-
tegrals has been illustrated in Sections VI.12 and VII.4. Now that we have
further developed the theory, we can handle many additional integrals. The
residue theorem is an especially powerful tool for this purpose.

Example 1. 10
00 sinx
--dx.
x
X.10. More Deflnite Integrals 131

Figure 8. The contour f •,R for Example 1.

This is an important integral in Fourier analysis. It is not obvious that

the integral converges. Although the integrand is well behaved at the left
endpoint of the interval of integration, there is potential trouble at the far
end, because fo 00
Isi~ x ldx diverges (Exercise X.10.1 below). Nevertheless,
as we shall prove on the basis of Cauchy's theorem, the given integral does
converge.
For r > 0, let Sr denote the semicircle in the upper half-plane with center
0 and radius r, oriented counterclockwise. For 0 < E < R, let rE,R denote
the closed curve consisting of the interval [t, R] followed by the semicircle
SR followed by the interval [-R, -t] followed by the semicircle -SE (see
iz
Figure 8). The function 2:_, whose imaginary part on the real axis equals
z
sinx x, is holomorphic in C\ { 0}, a domain containing r €
,R and its interior.
By Cauchy's theorem,
r eiz dz = 0.
lr,,R z
(We are using here the general Cauchy theorem. By an argument like that
in Section VII.5, the perverse reader could draw the same conclusion using
only Cauchy's theorem for convex domains. Now that we have the general
Cauchy theorem, such contortions are unnecessary.)
ix
On the real axis, the real part of 2:_ is cos x, an odd function (undefined
x x
. . 1
at 0 ) , an d t h e imagmary part, as a rea y note d , is
d . -sin-x , an even function.
.
x
The preceding equality therefore implies that

€
sinx
J R
--dx = -
2i
x
eiz
-dz +
SR z
J eiz
-dz.
s. z
J
We shall now take the limit as R--+ oo and E--+ 0.
132 X. Further Development of Basic Complex Function Theory

Consider first the integral over SR· We shall prove that it tends to 0
as R ---+ oo. This is intuitively plausible, because, in the upper half-plane,
leiz I is very small except in the vicinity of the real axis. Introducing the
parametrization t f-+ Reit (0 :St :S 7r) of SR, we can write

1 eiz dz = i
· 11r eiReit dt,
SR z 0

and so

11R e:z dzl :S 17r leiReit ldt

= 11r e-Rsintdt
= 2 fo7r/2 e-Rsintdt.
Using the inequality sin t ~ 2t/7r (0 :S t :S 7r /2), we obtain

Because the right side tends to 0 as R---+ oo, we can conclude that

lim
R--><X!
r eiz dz =
lsR z
0,

as desired.
eiz - 1
Consider now the integral over S,. The function f(z) = has a
z
removable singularity at the origin; in particular, it stays bounded near the
origin. From this, one easily concludes that

lim
E-->0
r J(z)dz
JS<
= 0.

eiz 1
Since - = - + f(z) and
z z
r ~dz = 7ri
ls. z
for all t: (by a straightforward calculation), we obtain

lim
E-->0
r eiz dz = 7ri.
JS< Z
X.10. More Defi.nite Integrals 133

Combining the two limits just found with the equality ( *), we obtain

.
1Im
•-o
R-+(X)
J f
R.
smxd
-- X
x
7r
= -.
2

In other words,
{
00
sinx dx 7r

}0 x 2

Example 2. 1 00

-oo
cosx
- -4 dx.
1+x

In this example we can see at the outset, using a comparison test, that
the integral converges. (Details are left to the reader; prior knowledge of
convergence will not be used below.) As in the preceding example, we make
use of the function eiz, whose absolute value in the upper half-plane is
bounded by 1. The integrand in our integral coincides on the real axis
eiz
with the real part of the function 1 + z 4 • That function is holomorphic in
C except for simple poles at the points errin/ 4 , n = 1, 3, 5, 7 (the fourth
roots of -1).
For R > 1, let r R denote the closed curve consisting of the interval
[-R, R] followed by the semicircle SR (defined as in the preceding example).
eiz
The singularities of the function - - -4 in the interior of rR are the points
l+z
z1 = erri/ 4 and z2 = e 3rri/ 4 . By the residue theorem,

To evaluate the residues we use the result of Exercise VIII.12.1: if the func-
tions g and h are holomorphic in an open set containing the point zo and h
g g(zo)
has a simple zero at zo, then resz 0 h= h'(zo). We thus have
134 X. Further Development of Basic Complex Function Theory

reSz=z2 ( ~)-
1+z
4 -
exp(iz2)
4z
3
2

exp(-~ - ~)
1 i )
4 ( v'2 + v'2

= l (~ - ~) e-l/ v'2 (cos ~- i sin ~).

The residue at z2 is the negative of the complex conjugate of the residue at
z1, so that the sum of the two residues equals 2i times the imaginary part
of the residue at z 1. The sum of the two residues therefore equals

iv'2 -1/v'2 ( 1 . 1 )
4 e -cos v'2 - sm v'2 .

It follows that

1 - eiz nv'2
- d z = --e-l/
rR 1 + z 4 2
V2(
2 cos - 1 +sin - 1 ) .
v'2 v'2
eix
On the real axis, the imaginary part of - - -4 is an odd function. Thus,
l+x

1 eiz
--dz =
r R 1 + z4
JR -cosx
-dx +
J --dz.
- R 1 + x4
eiz
s R 1 + z4
This in combination with the preceding equality gives

JR
_R
-cosx
-dx =
1 + x4
- - e -l/ v<-
nv'2
2
12(cos-+sm-
1
v'2
. 1 ) -
v'2
Js
R
--eiz
dz.
1 + z4
We now take the limit as R ___, oo. The absolute value of the integrand in
1
the integral over SR is bounded by R 4 _ 1 , so the integral itself is bounded

in absolute value by R:~ 1 , which tends to 0 as R_ _, oo. We can conclude

that

Example 3. 100

O
1
---dx,
1 + xa
a> 1.

As in the last example, one can see at the outset, by means of a compar-
ison test, that the integral converges, although that knowledge will not be
required below. In the domain G = {reiO: r > 0, le -
~1 < n} we consider
the branch of the function za that takes the value 1 at the point z = 1. We
X.10. More Defi.nite Integrals 135

Re21ri/a

27r/a
"..
•
0 E R

Figure 9. The contour r <,R for Example 3.

1
denote this function simply by za. The function - - - is holomorphic in G
1 + za
except for a simple pole at the point zo = e7ri/a.
For r > 0, let Ar denote the circular arc { reiO : 0 < () < 2; } , oriented
counterclockwise. For 0 < E < 1 < R, let f €,R denote the closed curve con-
sisting of the interval [E, R] followed by the arc AR followed by the segment
[Re 2 7ri/a,Ee 2 7ri/a] followed by the arc -A€ (see Figure 9). By the residue
theorem,

Jrr -+ za
1 -dz = 21ri reSz=zo ( -1-) ·
<,R
1 1 + za
We can compute the residue by the same method we used in the preceding
example. We obtain
-e7ri/a
a
Thus
i 1
--dz=
r,,R 1 + za
-21riem/a
a

Now, the integral of - -1 - over the interval [E, R] equals

1 + za €
1
---dx,
1 + xa
JR
and the integral of the same function along the segment [Ee 27ri/ a, Re 27ri/ a J

equals e 27ri/a JR --
€
1-dx (as one sees by using the parametrization t
1 + xa
r-->
136 X. Further Development of Basic Complex Function Theory

te 27ri/a (t::::; t:::; R)). We can therefore rewrite the preceding equality as

( 1 - e27ri/a) JR -1 +-1 xa- dx

E
-- -27rie7ri/a
a
+ 1 -+- 1
A. 1
1
za
dz -
AR
-1- d z.
1 + za
We now take the limit as E ---+ 0 and R ---+ oo. The integral over AE clearly
tends to 0 (since the integrand stays bounded). In the integral over AR, the
1
integrand is bounded in absolute value by R a_l , so the integral itself is

bounded in absolute value by a (~7rR

a_l
) , which tends to 0 as R---+ oo. In
the limit we thus obtain

( 1 - e27ri/a) roo _l_dx

lo 1 + xa
giving (after a short calculation)

1 o
00 1
--dx
1 + xa
7r
a sin~·

Exercise X.10.1. Prove that the integral fo 00

I si: x Idx diverges.

Exercise X.10.2. Derive the formula

1 _ 00
00
-cosx
- - dx
coshx cosh~
7r

eiz
by integrating the function --h- around the rectangle with vertices - R, R, R+
cos z
7ri, -R + 7ri, and letting R---+ oo.

Exercise X.10.3. Evaluate 1 0

00 sin 2 x
--
x
2 -dx by integrating the function
1 - e 2 iz
2
z
around the curves rE,R used in Example 1, and taking the limit as t: ---+ 0
and R---+ oo.
Exercise X.10.4. Derive the formula
f 00 cos ax 7r -ab
a::=: 0, b > 0.
lo (x2 + b2)2dx = 4b3 (1 + ab)e '
Exercise X.10.5. Derive the formula
roo xa-1 dx
0 <a< 2.
lo 1 + x2 2 sin ( 7r2a ) '

Exercise X.10.6. Evaluate

1 ----de,
0
27r cos()
a - cos()
a> 1.
X.11. The Argument Principle 137

Exercise X.10. 7. Let the function f = u +iv be holomorphic in a domain

containing the closed unit disk. Derive the relations
1 12rr e·eiO + z f(eie)d() =
-2
7r o ei - z
2f(z) - f(O), lzl < 1,
~ {2rr
eie + z f(eiB)d() = f(O), lzl < 1.
27r Jo
eie - z
From these deduce Herglotz's formula,
1 12rr eie + z .
f(z) = -2 .
7r o ei 0 - z
u(ei8 )d() + iv(O), lzl < 1,
and Poisson's formula,

u(z) = -
1 12rr I1.- lzl2 iO
u(e )d(), lzl < 1.
27r 0 ei - z 12
0

(These formulas are of central importance in more advanced function the-

ory.)

X.11. The Argument Principle

Let G be a domain in C and r a simple contour contained with its interior
in G. Let f be a holomorphic function in G without zeros on r. Then the
number of zeros off in int r, taking into account multiplicities, equals

_1 1
f'(z) dz.
27ri r f(z)

From the discussion in Section IX.3 we know that

~
i
1 f'(z) dz
r f(z)
equals the increment experienced by arg f(z) as z makes one circuit of r.
The argument principle thus states that each zero off in int r accounts for
27r of this increment. Multiple zeros, as indicated in the statement of the
principle, are counted as many times as required by their multiplicities.
To establish the argument principle we first note that f can have only
finitely many zeros in int r. In fact, the set ext r is open and contains both
C\G and the unbounded connected component of C\r. The set C\ ext r
is thus a closed and bounded (i.e., compact) subset of G. Since the zero set
off has no limit points in G (see Section VII.13), there can be only finitely
many zeros of f in C\ ext r, and thus there are only finitely many zeros of
f in int r' as asserted.
Suppose z1, ... , Zp are the zeros off in int r, and let m1, ... , mp be their
respective orders. Let G1 be the domain obtained by removing from G the
138 X. Further Development of Basic Complex Function Theory

zeros off in ext r. Then rand its interior are contained in G1, where the
function f' / f is holomorphic except for simple poles at the points z1, ... , Zp·
By the residue theorem,
r f'(z)
Jr f(z)dz =
.
2m~reszk(f ).
p !'

A simple calculation (requested as Exercise VIII.12.3) shows that

f'
reszk ( f) = mk,

and the desired equality follows.

Exercise X.11.1. Evaluate

-
1 1 zn-1
dz
27fi c 3zn - 1 '
where n is a positive integer, and C is the unit circle with the counterclock-
wise orientation.

Exercise X.11.2. (Argument principle for meromorphic functions.) Let G

be a domain in C and r a simple contour contained with its interior in G.
Let the function f be holomorphic in G except for isolated poles. (Such an
f is said to be meromorphic in G.) Prove that if f has neither zeros nor
poles on r' then
_1 r
f'(z) dz
27fi Jr
f(z)
equals the number of zeros that f has in int r minus the number of poles
that f has in int r, taking account of multiplicities.
Exercise X.11.3. Let the function f be holomorphic in a domain contain-
ing the closed unit disk. Prove that the increment in arg f around the unit
circle equals 27r times the total number of zeros of f in the closed unit disk,
provided zeros on the unit circle are counted with one-half their multiplici-
ties. (Part of the problem is to make a reasonable definition of the increment
in arg f.)

X.12. Rouche's Theorem

Let G be a domain in C, Ka compact subset of G, and f and g holomorphic
functions in G such that lf(z) - g(z)I < lf(z)I for every point z in the
boundary of K. Then f and g have the same number of zeros in the interior
of K, taking into account multiplicities.
Notice that the hypotheses imply that neither f nor g vanishes on the
boundary of K. Roughly, the theorem states that the number of zeros a
X.12. Rouche's Theorem 139

holomorphic function has in a compact set is stable under small perturba-

tions of the function, provided zeros on the boundary of the set are absent.
To prove Rouche's theorem, let G 1 denote the union of K with the set
of points z in G such that IJ(z) - g(z)I < IJ(z)I. Then G1 is an open set
containing K. By the separation lemma IX.8, there is a simple contour I' in
G1 \K such that Kc int I' c G1.
For 0 :::; t :::; 1 let ft = (1 - t)f + tg, so that Jo = J and Ji = g. Let m(t)
denote the number of zeros of ft in the interior of K. From the expression
ft = J - t(J - g) it is clear that ft is nonzero wherever the inequality
If - gl < If I holds. Hence, the only zeros of ft in G1 are the zeros in the
interior of K. Therefore, m(t) equals the number of zeros of ft in int r, and
by the argument principle we have

m(t) = 1
-.
fr
J£(z)
- ( )dz.
2m r 1t z
Finally, standard estimates show that the integral on the right side of the
preceding equality is a continuous function of t (Exercise X.12.1 below).
Since m(t) depends continuously on t and is integer valued, it is constant.
In particular, m(l) = m(O), the desired conclusion.
Exercise X.12.1. Supply the argument, omitted in the preceding proof, to
establish the continuity of the function m(t).
Exercise X.12.2. How many zeros, taking into account multiplicities, does
the polynomial z 7 + 4z 4 + z 3 + 1 have in the regions Iz I < 1 and 1 < Iz I < 2?
Exercise X.12.3. Prove that, for any positive number E, the function
1
- - . +sin z has infinitely many zeros in the strip IIm zl < E.
z+i
Exercise X.12.4. Let f be a holomorphic map of the open unit disk into
itself whose range lies in a compact subset of the open disk. Prove that f
has a unique fixed point.
Exercise* X.12.5. (Hurwitz's theorem.) Let the sequence (Jnr:=l
of holo-
morphic functions in the domain G converge locally uniformly in G to the
nonconstant function f. Prove that if f has at least m zeros in G, then all
but finitely many of the functions fn have at least m zeros in G. Conclude,
as a corollary, that if every f n is univalent then f is univalent.
Exercise X.12.6. Prove that all of the zeros of the polynomial zn+cn-1Zn- 1+
· · · + c1z +co lie in the disk with center 0 and radius
\h + lcn-11 2 + · · · + lc11 2 + lcol 2 ·
Exercise X.12. 7. Prove that, if 1 < a < oo, the function z +a - ez has
only one zero in the half-plane Re z < 0, and this zero is on the real axis.
140 X. Further Development of Basic Complex Function Theory

Exercise X.12.8. Prove that, if 0 < lal < 1 and n is a positive integer,
then the equation (z - l)nez =a has exactly n roots, each of multiplicity 1,
in the half-plane Re z > 0. Prove that if lal :::; 2-n the roots are all in the
disk lz - ll < 1/2.

X.13. The Local Mapping Theorem

Let f be a nonconstant holomorphic function in the domain G. Let zo be
a point of G and let wo = f(zo). Let m be the multiplicity with which f
assumes the value wo at zo (i.e., m is the order of the zero off -wo at zo).
For every sufficiently small positive number J, there is a positive number E
such that every value w satisfying 0 < lw-wol < E is assumed by f at exactly
m distinct points in the punctured disk 0 < lz - zol < J, with multiplicity 1
at each of those points.
The theorem says that, near the point zo, the function f behaves quali-
r
tatively like the polynomial ( z - zo + wo. To prove the theorem we choose
a positive number Jo such that the closed disk Iz - zo I :::; Jo is contained
in G, and such that neither f - wo nor f' vanishes for 0 < lz - zol ~ Jo;
such a choice is possible because the zeros of a nonconstant holomorphic
function are isolated. We shall show that the desired conclusion holds for
every positive number J not exceeding Jo.
Let J be as described, and let K denote the closed disk lz - zol ~ J.
Let E denote the minimum of lf(z) - wol for z on the boundary of K; it is
positive because f does not assume the value wo in K except at zo. Fix w
such that 0 < lw - wol < E. Then for z in the boundary of K we have

IU(z) - wo) - (f(z) - w)I = lw - wol < E ~ lf(z) - wol·

It therefore follows by Rouche's theorem that the functions f -wo and f -w
have the same number of zeros in the interior of K. The function f - wo has
a zero of order m at zo and no other zeros in K. Hence, the function f - w
has a total of m zeros in the interior of K. Each of its zeros in K is of order
1 because J' is nonzero in K except possibly at zo. Thus, f assumes the
value w at exactly m distinct points in the interior of K, with multiplicity
1 at each of those points, which is the desired conclusion.

X.14. Consequences of the Local Mapping Theorem

(i) If f is a nonconstant holomorphic function in a domain G, then f

maps open subsets of G onto open sets (briefly, f is an open map).
In fact, the local mapping theorem implies that the image under f of
any disk centered at a point zo of G contains a disk centered at f (zo). The
X.16. Conformal Equivalence 141

preceding statement follows immediately from this. Note that the maximum
modulus principle is a corollary.
(ii) If the holomorphic function f has a nonvanishing derivative at the
point zo, then there is a disk centered at zo in which f is univalent.
This is the case m = 1 of the local mapping theorem. This criterion
for univalence is strictly local. The function ez, for example, has a nowhere
vanishing derivative, yet it is not globally univalent.
(iii) A univalent holomorphic Junction has a nowhere vanishing deriva-
tive.
For, by the local mapping theorem, if the derivative of a holomorphic
function vanishes at some point, then, in every neighborhood of that point,
the function fails to be univalent.
Exercise X.14.1. Let the holomorphic function f have a zero of order m
at the point zo. Prove that there is an open disk centered at zo in which
there is a univalent branch of f l/m.

X.15. Inverses
The inverse of a univalent holomorphic function is holomorphic.
In fact, let f be a univalent holomorphic function in the domain G, and
let g be the inverse function, defined in f (G). Since f is an open map,
the set f (G) is open, and, moreover, if H is any open subset of G, then
g- 1 (H) (= f(H)) is open. It follows (by a standard criterion) that the
function g is continuous. Because also f' is never 0, one can from this point
argue just as in elementary calculus to show that g is differentiable, with
'( 1 w E J(G).
g w) = f'(g(w))'
The details are left for the reader (see Exercise IV.14.1).
Exercise X.15.1. Let f be a univalent holomorphic function in the domain
G, and let g be the inverse function. Let r be a simple contour contained
with its interior in G. Use the residue theorem to derive the formula

g ( w)
1
= -27ri.
i
r
zf'(z)
f (z ) - w dz, w E f(int f).

X.16. Conformal Equivalence

Two domains G 1 and G2 in C are said to be conformally equivalent if there
is a univalent holomorphic function fin G1 such that f(G1) = G2. Such
a function f enables one to transport function theory in G1 to function
theory in G 2; it can be viewed as a change of coordinates. Because the
142 X. Further Development of Basic Complex Function Theory

class of univalent holomorphic functions is closed under composition and

under the formation of inverses, one easily sees that conformal equivalence
is an equivalence relation in the standard set-theoretic sense (it is reflexive,
symmetric and transitive).
Any open disk in C is obviously conformally equivalent to the unit disk
via a linear function. Using linear-fractional transformations one easily sees
that any open half-plane is conformally equivalent to the unit disk. The first
exercise below contains further examples of domains that are conformally
equivalent to the unit disk.
It follows from Liouville's theorem that the entire plane C is not con-
formally equivalent to the unit disk.
Exercise X.16.1. Find a univalent holomorphic map from the open unit
disk onto each of the following domains: (i) the upper half-plane; (ii) the
whole plane minus the nonpositive real axis; (iii) the strip 0 < Im z < 1;
(iv) the first quadrant; (v) the intersection of the unit disk with the upper
half-plane; (vi) the unit disk minus the segment [O, 1).
Exercise X.16.2. A univalent holomorphic map of a domain onto itself is
called a conformal automorphism of the domain. Use Schwarz's lemma to
prove that the identity function is the only conformal automorphism of the
unit disk that fixes the origin and has a positive derivative at the origin.
Exercise X.16.3. Use the result of the preceding exercise to prove that
the only conformal automorphisms of the unit disk are the linear fractional
transformations of Exercise III.9.4.
Exercise X.16.4. Prove that the only conformal automorphisms of C are
the nonconstant linear functions.
Exercise X.16.5. Prove that the function f(z) = (l ~ z) 2 (called the
Koebe function) is univalent in the open unit disk. Find the image of the
unit disk under f.

X.17. The Riemann Mapping Theorem

Any domain in C that is conformally equivalent to a simply connected do-
main must itself be simply connected. This can be seen, for example, on
the basis of the characterization of simple connectivity given in Section X.4:
the domain G in C is simply connected if and only if every holomorphic
function in G has a primitive in G. (The details are relegated to Exercise
X.17.1 below.) A remarkable converse was formulated by G. F. B. Riemann
in 1851, but not proved in full generality until 1900 by W. F. Osgood: Every
simply connected domain in C, except for C itself, is conformally equivalent
to the unit disk.
X.18. An Extremal Property of Riemann Maps 143

In Riemann's time, the modern notion of simple connectivity was yet

to be developed, so the theorem envisioned by Riemann was in a technical
sense (although not in spirit) less broad than what was eventually proved.
The proof Riemann proposed, although basically sound, contained a crucial
gap, the eventual patching of which involved major advances in the subject
of potential theory (the study of harmonic functions). About a dozen years
after Osgood's paper appeared, C. Caratheodory and P. Koebe presented
approaches to the theorem which rely on the distinctive methodology of
complex function theory, rather than on potential theory. Their methods
evolved into an extremely elegant proof which, in its final form, is due to L.
Fejer and F. Riesz. It is presented below, after a few more preliminaries.
Exercise X.1 7 .1. Prove that a domain in C is simply connected if it is
conformally equivalent to a simply connected domain.
Exercise* X.17.2. (Uniqueness of the Riemann map.) Let I and g be
univalent holomorphic functions in the unit disk, D, such that l(D) = g(D),
l(O) = g(O), and arg J'(O) = arg g'(O). Prove that I= g.
Exercise* X.1 7 .3. Assume the truth of the Riemann mapping theorem.
Let G be a simply connected domain in C, different from C, let z0 be a
point of G, and let Bo be a real number. Prove that there is a univalent
holomorphic function I mapping the unit disk onto G such that l(O) = zo
and arg l'(O) =Bo. (The function I is unique, by the preceding exercise.)

X.18. An Extremal Property of Riemann Maps

Suppose G is a simply connected domain, other than C, and suppose there
exists a univalent holomorphic function I mapping G onto the unit disk
(commonly referred to as a Riemann map of G onto the disk). Let zo =
1- 1 (0), and suppose g is any holomorphic function mapping G into the
unit disk and satisfying g(zo) = 0. The composite function go 1- 1 is then a
holomorphic map of the unit disk into itself which vanishes at the origin. By
the corollary to Schwarz's lemma given in Exercise VII.17.1, the derivative
of go 1- 1 at 0 has absolute value at most 1, in fact less than 1 unless go 1- 1
is a unimodular constant times the identity function. After an application
of the chain rule, one sees that this means lg'(zo)I S ll'(zo)I, with equality
if and only if g = >-.I with ).. a unimodular constant. The Riemann maps
of G onto the unit disk vanishing at z0 thus maximize the absolute value of
the derivative at z 0 among the family of all holomorphic maps of G into the
unit disk vanishing at zo.
The preceding discussion, although premised on the existence of Rie-
mann maps, suggests the strategy behind the existence proof to be given.
We shall consider the extremum problem of maximizing ll'(zo)I among the
144 X. Further Development of Basic Complex Function Theory

family of univalent holomorphic functions f that map G into the unit disk
and vanish at a given point zo of G. We shall prove that the extremum prob-
lem has solutions, and then, by a bit of sorcery, prove that the solutions are
the desired Riemann maps.
The discussion above, by the way, shows that a Riemann map of G onto
the unit disk is uniquely determined by the point it maps to the origin and
the argument of its derivative at that point. The reader who has worked
Exercises X.17.2 and X.17.3 in the preceding section has already been over
this ground.
Existence proofs for solutions of extremum problems often rely on some
sort of compactness property. The compactness property to be used in the
proof of the Riemann mapping theorem is given in the next section.

X.19. Stieltjes-Osgood Theorem

A locally uniformly bounded sequence of holomorphic functions, defined in a
domain in C, has a locally uniformly convergent subsequence.
Although the term "locally uniformly bounded" has not been previously
defined, the reader has no doubt guessed the definition: a family of func-
tions in a domain is said to be locally uniformly bounded if each point in
the domain has a neighborhood in which the family is uniformly bounded
(equivalently, if the family is uniformly bounded on each compact subset of
the domain).
A family of holomorphic functions in a domain is called a normal family
if every sequence from the family has either a locally uniformly convergent
subsequence or a subsequence that diverges locally uniformly to oo. The
Stieltjes-Osgood theorem can thus be restated: a locally uniformly bounded
family of holomorphic functions is a normal family. The notion of normal
family is used extensively in complex function theory.
To prove the Stieltjes-Osgood theorem, let G be our domain, and let
(Jn)':=l be our locally uniformly bounded sequence of holomorphic functions
in G. It will be enough to show that, for each closed disk in G, our sequence
has a subsequence that converges uniformly on that disk. In fact, suppose
this is known. Take a sequence (Dk)";;°=l of open disks that cover G and
whose closures are contained in G. (For example, take the collection of
open disks whose radii are rational, whose centers have rational real and
imaginary parts, and whose closures are contained in G. The collection is
countable and so can be arranged into a sequence.) By our supposition,
the sequence (Jn)r;=l has a subsequence (11,n)':=l that converges uniformly
on Di. Similarly, the sequence (11,n)r;=l has a subsequence (h,n)':=l that
converges uniformly on D2. Proceeding recursively, we obtain for each k
X.19. Stieltjes-Osgood Theorem 145

a sequence (fk,nr::=l' uniformly convergent on Dk, each sequence being a

subsequence of the preceding ones and of the original one. The diagonal
sequence (!n,n) ~=l then has the required properties: it is a subsequence
of (!n) : 1 , and it converges locally uniformly in G. This is an often used
"diagonalization"procedure, most likely familiar to the reader.
Now consider one of our disks Dk. Since Dk is contained in G, there is an
open disk containing Dk whose closure is contained in G, and hence on which
our sequence is uniformly bounded. To complete the proof it will suffice to
show that our sequence has a subsequence that converges locally uniformly
in the larger disk. A linear change of variables maps the larger disk onto the
unit disk, reducing our task to that of establishing the following assertion:
A uniformly bounded sequence of holomorphic functions in the unit disk has
a subsequence that converges locally uniformly in the disk.
To prove the assertion, let Un)'f be the given sequence, and let fn
have the power series representation L::,o an,kZk. Choose M > 0 such that
lfn(z)I:::; M for izl < 1, for all n. It is asserted that ian,kl :::; M for all n, k.
In fact, letting Cr denote the counterclockwise oriented circle with center 0
and radius r (0 < r < 1), we have

Ian,k I --11 r fn(z)

27ri JCr zk+1 dz I <
- M
rk'

from which the desired inequality follows when we let r ____, 1.

Thus, each sequence (an,k)~=l (k = 0, 1, ... ) is bounded and so has a
convergent subsequence. Now, using the diagonalization procedure in the
same way as above, we can obtain an increasing sequence (nj)~ 1 of the
natural numbers such that the sequence (anJ,k)~ 1 converges for each k.
(The details are relegated to Exercise X.19.1 below.)
Let ak = lim an ,k (k
J-"00 J
= 0, 1, ... ). We have lakl ::; M for all k, so
the power series L::, 0 akzk converges in the unit disk, say to the function
f. It will be shown that fnj ____, f locally uniformly in the disk, which will
complete the proof of the Stieltjes-Osgood theorem.
Fix r in (0, 1). We show that fnj (z) ____, f(z) uniformly for lzl ::; r. Fix
E > 0. Take a positive integer m, to be appropriately chosen later. For
146 X. Further Development of Basic Complex Function Theory

lzl :::; r we have

lf(z) - fnj(z)I = lf)ak -

k=O
anj,k)zkl
m oo
:S Llak-anj,kl +2M L rk
k=O k=m+l
m 2Mrm+l
= L lak - anj,kl + 1_ r
k=O
We now fix m so that 2Mrm+l /(1 - r) < ~'getting, for lzl :::; r,
m
lf(z) - fnj(z)I :SL lak - anj,kl + ~·
k=O
Finally, since anj,k _____, ak for each k as j _____, oo, there is a jo such that
lak - anj,kl < E/2(m + 1) for k = 0, ... , m whenever j > jo, and hence so
that lf(z) - fnj(z)I < E for lzl:::; rand j 2: jo, the desired conclusion.
Exercise X.19.1. Fill in the details for the second diagonalization proce-
dure used above by proving that, if (an,k)~=l' k = 1, 2, ... , are bounded
sequences of complex numbers, then there is an increasing sequence (nj)f= 1
of natural numbers such that the sequence (anj,k)}= 1 converges for each k.
Exercise X.19.2. Prove that the family of holomorphic functions that map
the domain G into the right half-plane, Re z > 0, is a normal family.
Exercise X.19.3. Let G be a bounded domain. Prove that the family
of holomorphic functions f in G satisfying J JG
If (x + iy) Idx dy :::; 1 is a
normal family.
Exercise X.19.4. (Vitali's theorem) Let (fn):'=l be a locally uniformly
bounded sequence of holomorphic functions in a domain G. Assume that
the sequence converges at each point of a set which has a limit point in G.
Prove that the sequence converges locally uniformly in G.
Exercise X.19.5. Let f be a bounded holomorphic function in the strip
-1 < Re z < 1. Assume that the limit lim f (iy) = c exists. Prove
y-->+oo
that, for each r in (0, 1), the limit lim
y-->+oo
f (x + i y) = c exists uniformly for
-r:::; x:::; r.

X.20. Proof of the Riemann Mapping Theorem

Let G be a simply connected domain, different from C. Our aim is to prove
that there exists a univalent holomorphic map of G onto the open unit disk,
X.20. Proof of the Riemann Mapping Theorem 147

which we shall denote by D. Fix a point zo in G, and let F be the family

of univalent holomorphic maps of G into D that vanish at z0 .
STEP 1. Our first task is to prove that F is not empty. For this it will
suffice to show that there exists a bounded, univalent holomorphic function
in G (for the composite of such a function with a suitable linear function will
belong to F). Let c be any point in C\G. Because G is simply connected,
there is by X.5 a branch l of the function log(z - c) in G. The function l is
univalent, and the image set l(G) is open. Moreover, l(G) is disjoint from
its translate by 27ri. (If l(z2) = l(z1) + 27ri, then
z2 - c =exp (l(z2)) =exp (l(zi)) = z1 - c,
a blatant contradiction.) The function (l - l(zo) - 27ri)- 1 is thus bounded,
univalent, and holomorphic in G.
STEP 2. Now that we know Fis not empty, it makes sense to consider
the problem of maximizing lf'(zo)I over the functions fin F. Let
p =sup {lf'(zo)I : f E F}.
The number p is positive because the functions in F have nonvanishing
derivatives. (The finiteness of p will be apparent momentarily.) Let (fn):'=l
be a sequence in F such that lf~(zo)I ----+ p. By the Stieltjes-Osgood theorem,
this sequence has a locally uniformly convergent subsequence. So as not to
complicate our notation needlessly, we may assume the original sequence
converges locally uniformly. Let f denote the limit function. By the Weier-
strass convergence theorem VII.15, f is holomorphic, and f~ ----+ f' locally
uniformly. Thus lf'(zo)I = p, which implies in particular that the function
f is nonconstant. Since the functions fn are univalent, it now follows by
Hurwitz's theorem (Exercise X.12.5) that f is univalent. The function f is
bounded in modulus by 1, being the limit of such functions, and it vanishes
at zo. By the maximum modulus principle, f maps G into D. The function
f thus belongs to the family F; it is a solution of our extremal problem.
STEP 3. Finally, we complete the proof of the Riemann mapping the-
orem by showing that f (G) = D. The argument involves some sleight of
hand devised by Koe be (his "Schmiegungsverfahren" ) .
We argue by contradiction, supposing f (G) is a proper subset of D. Let
a be any point in D\f(G), and define the function h1 by
f-a
h1 = 1-a -r
The function h1 is a univalent holomorphic map of G into D, being the
composite of f with a conformal automorphism of D. Also h 1 does not
vanish in G, so, because G is simply connected, there is a branch h2 of the
function VhJ. in G. It also is a univalent holomorphic map of G into D.
148 X. Further Development of Basic Complex Function Theory

Lastly, we create a function g in our family F by setting b = h2(zo) and

defining
h2 -b
g= .
1- bh2
Now, after a bit of algebra (relegated to Exercise X.20.1 below) one finds
that
g+o:)
f = ( 1 + Cig g,
2b
where o: = 1 + lbl 2 . This gives f'(zo) = ag'(zo), so p = IJ'(zo)I < lg'(zo)I
since lo:I < 1, a contradiction. This concludes the proof of the Riemann
mapping theorem.
The proof just given, while undeniably elegant, is at the same time highly
nonconstructive. It does not address the question of how to find usable
approximations to Riemann maps in concrete situations. That question is
an active topic in numerical analysis.
The Riemann mapping theorem is the starting point of the subject of
geometric function theory, a vast endeavor, very vigorous today despite its
relatively advanced age. The reader can find more information in the books
on complex analysis listed among the references.
Exercise X.20.1. Perform the algebra needed in the preceding proof to
obtain the expression for the function f in terms of the function g.
Exercise X.20.2. Assume, retaining the notations of the preceding proof,
that (fn)';:'= 1 is a sequence in F such that IJ~(zo)I -+ p and nl!._,~f~(zo)
exists. Prove that the sequence (fn):=l converges locally uniformly in G.

Exercise X.20.3. Let (Gn):, 1 be an increasing sequence of simply con-

nected domains whose union, G, is not the whole plane. Let zo be a point
of G1. For each n, let fn be the Riemann map of D onto Gn satisfying
fn(O) = zo and arg f~(O) = 0. Prove that the sequence (fn):=l converges lo-
cally uniformly in D to the Riemann map f of D onto G satisfying f(O) = zo
and argf'(O) = 0.

X.21. Simple Connectivity Again

If G is a simply connected domain in C, then every closed curve in G is null
homo topic in G.
With this we see finally that the definition we used for simple connec-
tivity, in the context of planar domains, is equivalent to the definition using
homotopy. The equivalence is nontrivial; we will use the Riemann mapping
theorem to establish the implication above. Namely, a simple argument (left
X.21. Simple Connectivity Again 149

to the reader) establishes the implication for the cases G = C and G = D.

If G -=/- C then G is conformally equivalent to D, hence homeomorphic to D.
Since a homeomorphism preserves homotopy, the validity of the implication
for D implies its validity for G.
Exercise X.21.1. Prove that a domain Gin C is simply connected if and
only if its extended boundary (the boundary relative to C) is connected.
Appendix 1. Sufficient
condition for
differentiability

Let the real-valued function u, defined in an open set containing the point
z0 of C, have first partial derivatives that are continuous at zo. Then u is
differentiable at zo.
The proof relies on the mean value theorem of calculus: If the real-valued
function f is differentiable on the interval [a, b], then there is a point c in
(a, b) such that f(b) - f(a) = f'(c)(b- a).
We write zo = xo + i YO· To avoid possible notational confusion, we
let u1 and u2 denote the partial derivatives of u in the real and imaginary
directions, respectively (u 1 = g~, u2 = g~). Using the mean value theorem
we obtain, for z = x + iy near zo,

u(z) - u(zo) = u(x, y) - u(xo, Yo)

= u(x, y) - u(x, Yo)+ u(x, Yo) - u(xo, Yo)
= u2(x, T/z)(y - Yo)+ u1(~z, Yo)(x - xo),

where T/z is between Yo and y, and ~z is between xo and x. Thus,

u(x, y) - u(xo, Yo) - u1(xo, Yo)(x - xo) - u2(xo, Yo)(y - Yo)

= [u2(X,TJz) - u2(xo,Yo)](y-yo) + [u1(~z,Yo) - u1(xo,Yo)](x - xo),

-151
152 Appendix 1. Sufficient condition for differentiability

from which it follows that

lu(x,y) - u(xo,yo) - u1(xo,Yo)(x - xo) - u2(xo,Yo)(y-yo)I
lz - zol
:S lu2(x,17z) - u2(xo,Yo)I + lu1(ez,Yo) - u1(xo,Yo)I.
As z ---* zo we have (x, 17z) ---* (xo, Yo) and (ez, Yo) ---* (xo, Yo), so by the
continuity of u1 and u2 at zo the quantity on the right side in the preceding
inequality tends to 0, which gives the desired conclusion.
Appendix 2. Two
instances of the chain
rule

1. Let the function f be holomorphic in the open set G. Let "( : I ---+ G
be a curve, differentiable at the point to of I. Then the composite curve f o '"'(
is differentiable at to, with(! o '"'!)'(to)= J'("!(to)h'(to).
We let zo = '"'!(to). By the differentiability of f at zo we have

f(z) = f(zo) + J'(zo)(z - zo) + Ri(z),

where Ri(z)/(z - zo)---+ 0 as z---+ zo. By the differentiability of'"'( at to we
have
1(t) = 1(to) + 1'(to)(t - to)+ R2(t),
where R2(t)/(t - to) ---+ 0 as t---+ to. Thus

f ("!(t)) = f ("!(to)) + !' (zo) ("!(t) - "!(to)) + Ri ("!(t))

= f('Y(to)) + J'(zoh'(to)(t - to)+ R3(t),

where

We need to show R3(t)/(t - to) ---+ 0 as t---+ to. We have

R3(t) = J' (zo) ( R2(t)) + Ri ("!(t)).

t - to t - to t - to

-153
154 Appendix 2. Two instances of the chain rule

The first summand on the right clearly tends to 0 as t ----+ to. As for the
second summand, it equals 0if1(t) = zo, and otherwise it can be written as
R1("Y(t)) (1(t) - 1(to))
1(t) - 1(to) t - to '
from which one sees that it also tends to 0 as t ----+ to.
2. Let f = u+iv be a C 1 function defined in an open subset G of C. Let
1 be a differentiable curve in G, to a point where 1 is defined, zo = 1(t0 ),
a = 8f(zo) b = 8f(zo) Then
az ' Oz .

(! o 1 )' (to) = a1' (to) + b 1' (to).

We write 1 = a+ i (3. Below, for the sake of conciseness in notation,
dependencies on the parameter t and the variables x and y are suppressed.
By the chain rule of multivariable calculus,
du("Y) = du( a, (3) = ou a'+ ou (3'
dt dt ox oy
dv("Y) = dv(a, (3) = ova'+ ov (3'.
dt dt ox oy
Hence,
(! 01)' = of a'+ of (3'
ox oy
= (of + o ~) a' + i (of _ o ~) (3'
oz oz oz oz
= of (a'+ if3') + o~ (a' - i(31 )
oz oz
_of , of----,
- oz 1 + oz 1 .
This holds in particular fort= to, which is the desired conclusion.
Appendix 3. Groups,
and linear-fractional
transformations

This appendix is aimed at readers unfamiliar with the notion of a group,

in the hope that it will help them understand some of the statements in
Chapter III.
A group is by definition a set G equipped with a binary operation

GxG___,G
(a, b) ____,ab

(the group operation) having the properties: (i) a(bc) = (ab)c for all a, b, c;
(ii) There is an element e in G (the identity element) such that ea= ae =a
for all a; (iii) Each element a has an inverse a- 1 , an element of G satisfying
a- 1 a = aa- 1 = e. As shown in Section III.2, the preceding axioms are
satisfied by the set of linear-fractional transformations equipped with the
binary operation of composition, the identity element in this case being the
identity transformation. The axioms are also satisfied by GL(2, C), the set
of nonsingular two-by-two complex matrices equipped with the operation of
matrix multiplication; the identity here is the identity matrix.
An isomorphism of a group G1 onto a group G2 is a one-to-one map J
of G1 onto G2 that preserves the group operation: J(ab) = J(a)J(b) for all
a, bin G1. If such a map J exists then G1 and G2 are said to be isomorphic.
The map of GL(2, C) onto the group of linear-fractional transformations
that sends each matrix to its corresponding transformation preserves the

-155
156 Appendix 3. Groups, and linear-fractional transformations

group operation but, while surjective, it is not one-to-one, so it is not an

isomorphism; it is what is called a homomorphism. The quotient group
GL(2, C)/(C\{O} )h mentioned in Section III.2 has as its elements equiva-
lence classes of GL(2, C), where two matrices are regarded as equivalent if
they are scalar multiples of each other (i.e., if they induce the same linear-
fractional transformation). The group operation on the quotient group is
defined in the obvious way, and the statement that the quotient group is iso-
morphic to the group of linear-fractional transformations is simple to verify.
This isomorphism was mentioned in Chapter 3 by way of orientation for
readers familiar with group theory; it is not used subsequently.
Section III.5 concerns the property that any three distinct points z1, z2,
z3 can be mapped in order to any other three distinct points w1, w2, w3
by a linear-fractional transformation. By way of a proof, the property was
established for a particular choice of w1, w2, W3 (namely oo, 0, 1), after the
statement that this would be sufficient. To justify that statement, consider
arbitrary z1, z2, z3 and arbitrary w1, w2, W3. According to the special case
treated, there is a linear-fractional transformation ¢1 mapping z1, z2, z3 to
oo, 0, 1, and there is a linear-fractional transformation ¢2 mapping w 1 , w2,
w3 to oo, 0, 1. The linear-fractional transformation ¢2 1 o ¢1 then maps z1,
z2, z3 to w1, w2, w3.
Appendix 4.
Differentiation under
the integral sign

Let the function F: [r1, r2] x [t1, t2] -> R be continuous and have a continuous
first partial derivative with respect to its first variable. Let the function
I : [r1, r2] -> R be defined by

(1) I(r) = it2

ti
F(r, t)dt.

Then I is differentiable, and

dI(r) =
dr
1t
ti
2 8F(r, t) dt.
or

This result was needed in Section VIII.4, in the proof of Cauchy's theo-
rem for a pair of concentric circles. In proving (1), we simplify the notation
by letting F1 denote the first partial derivative of F with respect to the first
variable. Fix ro in [r1, r2]· We have

(2)
I(r) - I(ro)
----- -
it2 F (ro,t )dt
1
r - ro ti

-1t
-
ti
2
[F(r, t) - F(ro, t) _ F 1 ( ro, t )] dt.
r - ro
As r ___.. r 0 , the integrand in the integral on the right tends pointwise to 0.
To obtain the desired result we need to prove that the integral tends to 0,
for which we need to estimate the rate at which the integrand tends to 0.

-157
158 Appendix 4. Differentiation under the integral sign

Fix E > 0. The function F1, being continuous on the compact set [r1, r2] x
[t1, t2], is uniformly continuous there. Hence, there is a J > 0 such that
IF1(r, t) - F1(r', t')I < E whenever the distance between (r, t) and (r', t') is
less than J. Suppose Ir - rol < J. For each t, there is by the mean value
theorem a number rt between ro and r such that
F(r, t) - F(ro, t) _ F ( )
----'--'--------'----'- - I rt' t .
r - ro
By the choice of J, this implies that

I
F(r,t)-F(ro,t) - F1 (r , t
--'-'------'------'- 0
)I < E.
r - r0
So, whenever Ir - ro I < J the integrand in the integral on the right hand side
of (2) has absolute value less than E for all t, implying that the integral has
absolute value less than E( t2 - t1). As E is arbitrary, the desired conclusion
follows.
References

1. Ahlfors, L. V., Complex Analysis, McGraw-Hill, 1979.

2. Axler, S., Bourdon, P., and Ramey, W., Harmonic Function Theory, Springer-Verlag,
1992.
3. Burckel, R. B., An Introduction to Classical Complex Analysis, Vol. 1, Birkhauser,
1979.
4. Conway, J. B., Functions of One Complex Variable, Springer-Verlag, 1978.
5. Marsden, J. E., and Hoffman, M. J., Basic Complex Analysis, W. H. Freeman, 1987.
6. Newman, M. H. A., Elements of the Topology of Plane Sets of Points, Cambridge, 1961.
7. Rudin, W., Principles of Mathematical Analysis, McGraw-Hill, 1976.
8. Saks, S., and Zygmund, A., Analytic Functions, Elsevier, 1971.

-159
Index

absolute convergence, 50 complex conjugate, 4

absolute value, 3 complex integral, 67
analytic, 1 7 complex number system, 1
analytic continuation, 45 complex partial differential operators, 18
arc length, 67 complex plane, 3
argument, 6 complex projective space, 27
argument principle, 137, 138 conformal automorphism, 142
conformal equivalence, 141
conformal map, 22
Bernoulli numbers, 104
contour, 113
bijection, 28
simple, 114
branch
convergence
function, 44
local uniform, 51
of arg z, 40
of an infinite series, 49
of log z, 40
uniform, 51
of a holomorphic function, 42
cross ratio, 30, 32
of an inverse, 44
curve, 20
of the logarithm, 42
differentiable, 20
Burckel, R. B., 85
regular, 20

Caratheodory, C., 143

De Moivre's formula, 7
Casorati-Weierstrass theorem, 106
differentation of power series, 58
Cauchy integral, 83, 99
differentiable
Cauchy product, 61, 84
in the complex sense, 13
Cauchy's formula, 130 in the real sense, 15, 151
for a circle, 79 differentiation under the integral sign, 157
Cauchy's formula for an annulus, 101
division of power series, 63
Cauchy's theorem, 75, 109, 118, 121, 122
domain, 125
for a convex region, 78
for a triangle, 75
for simply connected domains, 126 elementary function, 35
for two concentric circles, 100 entire function, 86
Cauchy-Hadamard theorem, 55 Euler, L., 35
Cauchy-Riemann equations, 16 exponential function ( ez), 35
in polar coordinates, 16 extended complex plane, 11
chain rule, 14, 153

-
clircle, 31 Fejer, L., 143

161
162 Index

field axioms, 2 locally uniformly bounded, 144

fundamental theorem of algebra, 1, 86 logarithm, 40
fundamental theorem of calculus, 66 of a continuous function, 109, 119
of a holomorphic function, 42, 127
Gauss, C. F., 1 logarithmic derivative, 42
general linear group for dimension two, 29
geometric function theory, 148 maximum modulus principle, 90
geometric series, 50 maximum principle for harmonic functions,
Goursat's lemma, 75, 85 95
Goursat, E., 75 mean value property
group, 155 for harmonic functions, 94
for holomorphic functions, 81
harmonic conjugate, 24, 93, 128 meromorphic function, 138
harmonic function, 23 modulus, 3
Herglotz's formula, 137 Morera's theorem, 85
holomorphic, 17 Morera, G., 85
homothetic map, 19 multiplicity of a zero, 87
homotopy, 119 multiply connected domain, 128
Hurwitz's theorem, 139
hyperbolic functions, 38 normal family, 144

identity theorem, 89 order

identity theorem for harmonic functions, 94 of a pole, 103
imaginary axis, 3 of a zero, 87
imaginary part, 4
imaginary unit, 3 parallelogram equality, 5
increment in the argument, 110 partial fraction decomposition, 104
increment in the logarithm, 110 partial sum, 49
index (of a point with respect to a curve), Picard's theorem, 106
111 Picard, E., 106
inverse of a holomorphic function, 43, 141 Pick's lemma, 92
inverse trigonometric functions, 44 piecewise C 1 ' 66
inversion, 30 piecewise continuous, 65
isolated singularity, 102 point at infinity, 11
essential, 102 Poisson's formula, 137
pole, 102 polar form, 6
removable, 102 pole, 102
isomorphism, 155 order, 103
simple, 103
Koebe, P., 143 polynomials, 15
potential theory, 143
Laplace's equation, 23, 25 power series, 52
in polar coordinates, 24 center, 52
Laurent series, 98 coefficient, 52
annulus of convergence, 98 primitive, 78, 127
convergence, 98 principle value of arg, 6
principal part, 102 product rule, 14
law of exponents, 35, 36, 62 purely imaginary number, 3
limit superior, 54
linear-fractional transformation, 27, 155 quotient rule, 14
conjugate, 31
factorization, 30 radius of convergence, 54
fixed points, 29 ratio test, 56
preservation of clircles, 31 rational functions, 15
three-fold transitivity, 29 real axis, 3
Liouville's theorem, 86 real part, 4
local mapping theorem, 140 reparametrization, 69
Index 163

residue, 106
residue theorem, 113, 129
reverse of a curve, 70
Riemann, 47
Riemann integral for complex-valued
functions, 65
Riemann map, 143
Riemann mapping theorem, 142, 146
Riemann sphere, 11
Riemann surface, 45
Riemann, G. F. B., 11
Riesz, F., 143
roots, 43
Rouche's theorem, 138
Runge's approximation theorem, 121, 123

Schmiegungsverfahren, 147
Schwarz's lemma, 91
separation lemma, 115
simple connectivity and homotopy, 128, 148
simply connected domain, 125
spherical metric, 10
stereographic projection, 9
Stieltjes-Osgood theorem, 144
sum rule, 14

tangent direction, 21
Toeplitz matrix, 63
triangle inequality, 4, 50
triangle inequality for integration, 66
trigonometric functions, 39

uniformly Cauchy, 51
univalent, 44

Vitali's theorem, 146

Weierstrass convergence theorem, 89

Weierstrass, K., 121
winding number, 111, 114

zero (of a holomorphic function)

order, 87
simple, 88
 ISB N 978 - 0-8218-4428-1

9 780821 844281 AMS on the Web

MBK/49 www.ams.org