Pure and Applied

Sally
The UNDERGRADUATE TEXTS 16
SERIES

Complex
Variables

Joseph L. Taylor

American Mathematical Society

Complex
Variables
 Pure and Applied
Sally
The UNDERGRADUATE TEXTS • 16
SERIES

Complex
Variables

Joseph L. Taylor

American Mathematical Society

Providence, Rhode Island
 EDITORIAL COMMITTEE
Paul J. Sally, Jr. (Chair) Joseph Silverman
Francis Su Susan Tolman

2010 Mathematics Subject Classification. Primary 30–01, 30Axx, 30Bxx, 30Dxx, 30Exx.

For additional information and updates on this book, visit

www.ams.org/bookpages/amstext-16

Library of Congress Cataloging-in-Publication Data

Taylor, Joseph L., 1941–
Complex variables / Joseph L. Taylor.
p. cm. — (Pure and applied undergraduate texts ; v. 16)
Includes bibliographical references and index.
ISBN 978-0-8218-6901-7 (alk. paper)
1. Functions of complex variables. I. Title.

QA331.7.T389 2011
515.9—dc23
2011019541

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 [email protected].


c 2011 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://www.ams.org/
10 9 8 7 6 5 4 3 2 1 16 15 14 13 12 11
Contents

Preface vii

Chapter 1. The Complex Numbers 1

1.1. Definition and Simple Properties 1
1.2. Convergence in C 8
1.3. The Exponential Function 14
1.4. Polar Form for Complex Numbers 19

Chapter 2. Analytic Functions 27

2.1. Continuous Functions 27
2.2. The Complex Derivative 34
2.3. Contour Integrals 41
2.4. Properties of Contour Integrals 47
2.5. Cauchy’s Integral Theorem for a Triangle 53
2.6. Cauchy’s Theorem for a Convex Set 60
2.7. Properties of the Index Function 66

Chapter 3. Power Series Expansions 75

3.1. Uniform Convergence 75
3.2. Power Series Expansions 82
3.3. Liouville’s Theorem 89
3.4. Zeroes and Singularities 95
3.5. The Maximum Modulus Principle 102

Chapter 4. The General Cauchy Theorems 109

4.1. Chains and Cycles 110
4.2. Cauchy’s Theorems 118

v
vi Contents

4.3. Laurent Series 124

4.4. The Residue Theorem 131
4.5. Rouché’s Theorem and Inverse Functions 137
4.6. Homotopy 141
Chapter 5. Residue Theory 153
5.1. Computing Residues 153
5.2. Evaluating Integrals Using Residues 159
5.3. Fourier Transforms 166
5.4. The Laplace and Mellin Transforms 175
5.5. Summing Infinite Series 179
Chapter 6. Conformal Mappings 185
6.1. Definition and Examples 185
6.2. The Riemann Sphere 191
6.3. Linear Fractional Transformations 197
6.4. The Riemann Mapping Theorem 202
6.5. The Poisson Integral 207
6.6. The Dirichlet Problem 213
Chapter 7. Analytic Continuation and the Picard Theorems 221
7.1. The Schwarz Reflection Principle 222
7.2. Continuation Along a Curve 226
7.3. Analytic Covering Maps 232
7.4. The Picard Theorems 236
Chapter 8. Infinite Products 245
8.1. Convergence of Infinite Products 245
8.2. Weierstrass Products 251
8.3. Entire Functions of Finite Order 257
8.4. Hadamard’s Factorization Theorem 262
Chapter 9. The Gamma and Zeta Functions 269
9.1. Euler’s Gamma Function 270
9.2. The Riemann Zeta Function 275
9.3. Properties of ζ 281
9.4. The Riemann Hypothesis and Prime Numbers 287
9.5. A Proof of the Prime Number Theorem 291
Bibliography 299
Index 301
Preface

Basic complex variables is a very popular subject among mathematics faculty and
students. There is one big theorem (the Cauchy Integral Theorem) with a somewhat
difficult proof, but this is followed by a host of easy to prove consequences that are
both surprising and powerful. Furthermore, for undergraduates, the subject is
refreshingly new and different from the analysis courses which precede it.
Undergraduate complex variables is our favorite course to teach, and we teach
it as often as we can. Over a period of years we developed notes for use in such a
course. The course is a one-semester undergraduate course on complex variables at
the University of Utah. It is designed for junior level students who have completed
three semesters of calculus and have also had some linear algebra and at least one
semester of foundations of analysis.
Over the past several years, these notes have been expanded and modified to
serve other audiences as well. We have used the expanded notes to teach a course
in applied complex variables for engineering students as well a course at the first
year graduate level for mathematics graduate students. The topics covered, the
emphasis given to topics, and the choice of exercises were different for each of these
audiences.
When taught as a one-semester junior level course for mathematics students,
the course is a transitional course between freshman and sophomore level calculus,
linear algebra, and differential equations and the much more sophisticated senior
level mathematics courses taught at Utah. The students are expected to understand
definitions and proofs, and the exercises assigned will include proofs as well as
computations. The course moves at a leisurely pace, and the material covered
includes only Chapters 1, 2, and 3 and selected sections from Chapters 4, 5, and 6.
A full year course could easily cover the entire text.
When we teach a one-semester undergraduate course for engineers using these
notes, we cover essentially the same material as in the course for mathematics
majors, but not all the proofs are done in detail, and there is more emphasis on

vii
viii Preface

computational examples and exercises. The applications in Chapter 6 are given

special emphasis.
When taught as a one-semester graduate course, we assume students have some
knowledge of complex numbers and so Chapter 1 is given just a brief review. Chap-
ters 2, 3, and 4 are covered completely, parts of Chapters 5 and 6 are covered, along
with Chapter 7. If time allows, Chapters 8 and 9 are summarized at the end of the
course. In the homework, the emphasis is on the theoretical exercises.
We have tried to present this material in a fashion which is both rigorous
and concise, with simple, straightforward explanations. We feel that the modern
tendency to expand textbooks with ever more material, excessively verbose ex-
planations, and more and more bells and whistles, simply gets in the way of the
student’s understanding of the material.
The exercises differ widely in level of abstraction and level of difficulty. They
vary from the simple to the quite difficult and from the computational to the the-
oretical. There are exercises that ask students to prove something or to construct
an example with certain properties. There are exercises that ask students to apply
theoretical material to help do a computation or to solve a practical problem. Each
section contains a number of examples designed to illustrate the material of the
section and to teach students how to approach the exercises for that section.
This text, in its various incarnations, has been used by the author and his
colleagues for several years at the University of Utah. Each use has led to improve-
ments, additions, and corrections.
The text begins, in Chapter 1, with a discussion of the fact that the real number
system is insufficient for some purposes (solving polynomial equations). The system
of complex numbers is developed in an attempt to remedy this problem. We then
study basic arithmetic of complex numbers, convergence of sequences and series of
complex numbers, power series, the exponential function, polar form for complex
numbers and the complex logarithm.
The core material of a complex variables course is the material covered here
in Chapters 2 and 3. Analytic functions are introduced in Chapter 2 as functions
which have a complex derivative. This leads to a discussion of the Cauchy-Riemann
equations and harmonic functions. We then introduce contour integrals and the
index function (winding number) for closed paths. We prove the Cauchy-integral
theorem for triangles and then for convex sets. We believe that this approach leads
to the simplest and quickest rigorous proof of a form of Cauchy’s theorem that can
be used to prove the existence of power series expansions for analytic functions.
The proof that analytic functions have power series expansions occurs early in
Chapter 3. This is followed by a wide range of powerful applications with simple
proofs – Morera’s Theorem, Liouville’s Theorem, the Fundamental Theorem of
Algebra, the characterization of zeroes and singularities of analytic functions, the
Maximum Modulus Principle, and Schwarz’s Lemma.
Chapter 4 begins with a proof of the general form of Cauchy’s Theorem and
Cauchy’s Formula. These theorems involve functions which are analytic on a general
open set. The integration takes place around a cycle (a generalization of a closed
path) which is required to have index zero about any point not in the set. We
Preface ix

go on to study Laurent series, the Residue Theorem, Rouché’s Theorem, inverse

functions and the Open Mapping Theorem. The chapter ends with a discussion of
homotopy and its relationship to integrals around closed paths. We normally do
not cover this last section in our undergraduate course, but it would certainly be
appropriate for a graduate course using this text.
Residue theory is covered in Chapter 5. We discuss techniques for computing
residues as well as a wide variety of applications of residue theory to problems
involving the calculation of integrals. We normally do not cover all of this in our
undergraduate course. We typically skip the last two sections of this chapter in
favor of covering most of Chapter 6.
Chapter 6 deals with conformal maps. We prove the Riemann Mapping The-
orem and show how it can be used to transform problems involving analytic or
harmonic functions on a simply connected subset of the plane to the analogous
problems on the unit disc. We use this technique to study the Dirichlet problem
on open, proper, simply connected subsets of C. We discuss applications to heat
flow, electrostatics, and hydrodynamics. This material is of particular interest to
students in a course on complex variables for engineering students.
The goal of Chapter 7 on analytic continuation is to prove the Picard theorems
concerning meromorphic functions at essential singularities. Along the way, we
prove the Schwarz Reflection Principle and the Monodromy Theorem, and discuss
lifting analytic functions through a covering map.
Chapters 8 and 9 are normally not covered in our one-semester undergraduate
complex variables course. Some of the topics in these chapters are, however, of-
ten covered in the graduate course. In Chapter 8 we use Weierstrass products to
construct an analytic function with a given discrete set of zeroes, with prescribed
multiplicities. This is the Weierstrass Theorem. It leads directly to the Weierstrass
Factorization Theorem for entire functions. We also prove the Mittag-Leffler Theo-
rem – which gives the existence of a meromorphic function with a prescribed set of
poles and principle parts. The final result of the chapter is the proof of Hadamard’s
Theorem characterizing entire functions of finite order. This is a key ingredient in
the proof of the Prime Number Theorem in Chapter 9.
In Chapter 9 we introduce the gamma and zeta functions, develop their basic
properties, discuss the Riemann Hypothesis, and prove the Prime Number The-
orem. Course notes by our colleague Dragan Milićić provided the original basis
for this material. It then went through several years of expansion and refinement
before reaching its present form. There is a great deal of technical calculation in
this material and we do not cover it in our undergraduate course. However, a few
lectures summarizing this material have proved to be a popular way to end the
graduate course.
Several standard texts in complex variables and related topics were useful guides
in preparing this text. These may also be of interest to the student who wishes to
learn more about the subject. They are listed in a short bibliography at the end of
the text.
 Chapter 1

The Complex Numbers

1.1. Definition and Simple Properties

The number system is a tool devised by humans to aid in the description of quan-
tities of the various things humans have to deal with. It has evolved as human
culture has evolved, beginning with something very primitive like: 1, 2, 3, many,
moving on to the natural numbers, then the integers, the rational numbers, the real
numbers and then the complex numbers.
At each stage of development, the number system was expanded in response
to the need to describe quantities that the old number system could not. For
example, the negative numbers were introduced in order to be able to describe a
loss as opposed to a gain, or moving backward rather than forward. The rational
numbers were introduced because we do not always deal with whole numbers of
things (we have 2/3 of a pie left). The real number system evolved from the
rational number system out of a need to be able to describe such things as the
length of the hypotenuse of a right triangle (this involves square roots) and the
area or circumference of a circle (this involves π).
In this course, we will assume students are familiar with the real number system
and its properties. We will define the complex number system as a needed extension
of the real number system and develop its properties. We will then go on to study
functions of a complex variable.

The Real Numbers are Insufficient. The complex number system was devel-
oped in response to the need for solutions to polynomial equations. The simplest
polynomial equation that does not have a solution in the real number system is the
equation
x2 + 1 = 0,
which has no real solution because −1 has no real square root. More generally, a
quadratic equation
(1.1.1) ax2 + bx + c = 0,

1
2 1. The Complex Numbers

where a, b and c are real numbers, formally has two solutions given by the quadratic
formula
√
−b ± b2 − 4ac
(1.1.2) x= ,
2a
but these will not be real numbers if b2 − 4ac is negative. If we could take square
roots of negative numbers, then the quadratic formula would give us solutions to
(1.1.1) for all choices of real coefficients a, b, c. To make this possible, we expand
the real number system in the following way, thus creating the complex number
system C.

Constructing C. We begin by adjoining a single new number to our old number

system R. We will denote it by i and declare it to be a square root of −1. Thus,
i2 = −1.
Our new number system is to contain both R and the new number i and it should be
closed under addition and multiplication. If it is to be closed under multiplication,
we need a number iy for every real number y. Likewise, if it is to be closed under
addition, there should be a number x + iy in our new number system for each pair
of real numbers (x, y). It turns out that this is enough. If we define the set of
complex numbers C to be the set of all symbols of the form x + iy where (x, y) is
a pair of real numbers, and if we define addition and multiplication appropriately,
then the resulting number system is a field in which every polynomial equation has
a root. We will be a long time proving the latter half of this statement, but it is
not hard to prove the first part.
To define the operations of addition and multiplication in C, we begin by noting
that, as a set, C may be identified with R2 – the set of all pairs (x, y) of real
numbers. Obviously, each pair (x, y) determines a symbol x + iy and vice versa.
This identification makes C into a vector space over R and gives us operations of
addition and scalar multiplication by reals which satisfy the usual associative and
distributive rules. The resulting operation of addition is
(x1 + iy1 ) + (x2 + iy2 ) = (x1 + x2 ) + i(y1 + y2 ).
It remains to define a product on C.
We have already declared that i2 = −1. If we also require that the associative
and distributive laws of multiplication should hold and that the multiplication of
real numbers should remain as before, then the product of two complex numbers
x1 + iy1 and x2 + iy2 must be
(x1 + iy1 )(x2 + iy2 ) = x1 x2 + ix1 y2 + iy1 x2 + i2 y1 y2 = (x1 x2 − y1 y2 ) + i(x1 y2 + y1 x2 ).
We formalize this conclusion in the following definition.
Definition 1.1.1. We define the system C of complex numbers to be the set of
all symbols of the form x + iy with (x, y) ∈ R2 , with addition and multiplication
defined by
(x1 + iy1 ) + (x2 + iy2 ) = (x1 + x2 ) + i(y1 + y2 )
and
(x1 + iy1 )(x2 + iy2 ) = (x1 x2 − y1 y2 ) + i(x1 y2 + y1 x2 ).
1.1. Definition and Simple Properties 3

A complex number of the form x + i0, with x ∈ R will be denoted simply as x.

This identifies R as a subset of C. Similarly, a complex number of the form 0 + iy
with y real will be denoted simply as iy. The numbers of this form are traditionally
called the imaginary numbers.
Note that, from the above definition, if x, y ∈ R, then
yi = (y + i0)(0 + i) = 0 + iy = iy
and so x + iy and x + yi are the same complex number. Which form is used to
describe this number is usually dictated by which looks best typographically. When
specific numbers replace x and y, the latter seems to look best. Thus, we usually
write 2 + 3i rather than 2 + i3.
Example 1.1.2. If z1 = 5 + 2i and z2 = 3 − 4i, find z1 + z2 and z1 z2 .
Solution:
z1 + z2 = 5 + 3 + (2 − 4)i = 8 − 2i,
z1 z2 = (5 · 3 − 2 · (−4)) + (5 · (−4) + 3 · 2)i = 23 − 14i.
Example 1.1.3. Show that the quadratic equation (1.1.1) has solutions which are
complex numbers.
Solution: If b2 − 4ac ≥ 0, the quadratic formula (1.1.2) tells us the solutions
are √ √
−b + b2 − 4ac −b − b2 − 4ac
and .
2a 2a
On the other hand, if b2 − 4ac < 0, then 4ac − b2 is positive and has real square
roots. By squaring both sides, and using i2 = −1, it is easy to see that
 
± b2 − 4ac = ±i 4ac − b2 .
This suggests that the solutions to the quadratic equation in this case are the two
complex numbers
√ √
b 4ac − b2 b 4ac − b2
− +i and − −i .
2a 2a 2a 2a
That these two numbers are, indeed, solutions to the quadratic equation may be
verified by directly substituting them in for x in (1.1.1). We leave this as an exercise
(Exercise 1.1.7).

Field Properties. In our definition of the product of two complex numbers,

we were guided by the desire to have the usual rules of arithmetic hold – that is,
the commutative and associative laws for addition and for multiplication and the
distributive law. Did we succeed? These are some of the properties of a field. Do
these laws actually hold in C with the operations as defined above? The following
theorem says they do.
Theorem 1.1.4. If z1 , z2 , z3 are complex numbers, then
(a) z1 + z2 = z2 + z1 – commutative law of addition;
(b) (z1 + z2 ) + z3 = z1 + (z2 + z3 ) – associative law of addition;
(c) z1 z2 = z2 z1 – commutative law of multiplication;
(d) (z1 z2 )z3 = z1 (z2 z3 ) – associative law of multiplication;
4 1. The Complex Numbers

2i . 3+2i

−1 0 1 2 3 4
−i

Figure 1.1.1. Plot of the Complex Number 3 + 2i.

(e) z1 (z2 + z3 ) = z1 z2 + z1 z3 – distributive law.

Proof. As far as the operation of addition is concerned, C is just R2 , which is a

vector space over R. Parts (a) and (b) of the theorem follow directly from this.
Part (c) is obvious from the definition of multiplication. We will prove part (d) and
leave part (e) as an exercise (Exercise 1.1.8).
If zj = xj + iyj for j= 1, 2, 3, then
(z1 z2 )z3 = ((x1 + iy1 )(x2 + iy2 ))(x3 + iy3 )
= (x1 x2 − y1 y2 + i(x1 y2 + y1 x2 ))(x3 + iy3 )
= x1 x2 x3 − y1 y2 x3 − x1 y2 y3 − y1 x2 y3
+ i(x1 x2 y3 − y1 y2 y3 + x1 y2 x3 + y1 x2 x3 ),
while
z1 (z2 z3 ) = (x1 + iy1 )((x2 + iy2 )(x3 + iy3 ))
= (x1 + iy1 )(x2 x3 − y2 y3 + i(x2 y3 + y2 x3 ))
= x1 x2 x3 − x1 y2 y3 − y1 x2 y3 − y1 y2 x3
+ i(x1 x2 y3 + x1 y2 x3 + y1 x2 x3 − y1 y2 y3 ).
Since the results are the same, the proof of (d) is complete. 

The properties described in the above theorem are some of the properties that
must hold in a field. A field must also have additive and multiplicative identities –
that is, elements 0 and 1 which satisfy
(1.1.3) z+0=z
and
(1.1.4) 1·z =z
for every element z in the field. That this holds for C follows immediately from the
fact that C is a vector space over R.
A field must also have the properties that every element z has an additive
inverse, that is, an element −z such that
(1.1.5) z + (−z) = 0,
1.1. Definition and Simple Properties 5

and every non-zero element z has a mutiplicative inverse, that is, an element z −1
such that
(1.1.6) z · z −1 = 1.
The first of these follows immediately from the fact that C is a vector space
over R. The second is nearly as easy. If z = x + iy = 0, then a direct calculation
shows that
x − iy x y
z −1 = 2 2
= 2 2
− 2 i
x +y x +y x + y2
satisfies (1.1.6). We conclude:
Theorem 1.1.5. With addition and multiplication defined as in Definition 1.1.1,
the complex numbers form a field.

Complex Conjugation and Modulus.

Definition 1.1.6. If z = x + iy is a complex number, then its complex conjugate,
denoted z, is defined by
z = x − iy,
while its modulus, denoted |z|, is defined by

|z| = x2 + y 2 .

Note that the modulus, as defined above, is just the usual Euclidean norm in
the vector space R2 . Thus, if z1 , z2 ∈ C, then |z1 − z2 | is the Euclidean distance
from z1 to z2 . The term modulus is traditional, but the terms norm and absolute
value are also commonly used to mean the same thing. We will use all three.
Note also that the two solutions of a quadratic equation with real coefficients
given in Example 1.1.3 are complex conjugates of each other. Thus, the solutions
to a quadratic equation with real coefficients occur in conjugate pairs. Quadratic
equations with complex coefficients also have roots and they are also given by the
quadratic formula. However, we cannot prove this until we prove that every complex
number has a square root. In fact, in Section 1.4 we will prove that every complex
number has roots of all orders.
For a complex number z = x + iy, the real number x is called the real part
of z and is denoted Re(z), while the number y is called the imaginary part of z
and is denoted Im(z). In graphing complex numbers using a rectilinear coordinate
system, x determines the coordinate on the horizontal axis, while y determines the
coordinate on the vertical axis.
Note that a complex number z is real if and only if z = z, and it is purely
imaginary if and only if z = −z. Note also, that if z = x + iy, then
z+z z−z
Re(z) = x = and Im(z) = y = .
2 2i
The elementary properties of conjugation and modulus are gathered together in the
next theorem.
Theorem 1.1.7. If z and w are complex numbers, then
(a) z = z;
6 1. The Complex Numbers

−z = −x+iy
. . z = x+iy

. .
−z = −x−iy z = x−iy

Figure 1.1.2. Plot of the Complex Numbers z, z, −z, and −z.

(b) zz = |z|2 ;
(c) z + w = z + w;
(d) zw = z w;
(e) |zw| = |z||w| and |z| = |z|;
(f) |z| is a non-negative real number and is 0 if and only if z = 0;
(g) | Re(zw)| ≤ |z||w|;
(h) |z + w| ≤ |z| + |w|.

Proof. We will prove (g) and (h). The other parts are elementary computations
or observations and will be left as exercises.
Parts (g) and (h) are the Cauchy-Schwarz inequality and the triangle inequality
for the vector space R2 . Versions of these inequlities hold in general Euclidian space
Rn . The proofs we give here are specializations to C of the standard proofs of these
inequalities in Rn .
To prove (g), we begin with the observation that (a) and (d) imply that
zw + zw = 2 Re(zw).
We then let t be an arbitrary real number and note that, by Parts (c), (d), and (f),
0 ≤ |zt + w|2 = (zt + w)(zt + w) = |z|2 t2 + 2 Re(zw)t + |w|2
for all values of t. This implies that the quadratic polynomial in t given by
|z|2 t2 + 2 Re(zw)t + |w|2
is never negative and, therefore, has at most one real root. This is only possible if
the expression under the radical in the quadratic formula is negative or zero. Thus,
4(Re(zw))2 − 4|z|2 |w|2 ≤ 0.
Part (g) follows immediately from this.
1.1. Definition and Simple Properties 7

Part (h) follows directly from Part (g) and the other parts of the theorem. In
fact,

|z + w|2 = (z + w)(z + w)
= |z|2 + 2 Re(zw) + |w|2
≤ |z|2 + 2|z||w| + |w|2 = (|z| + |w|)2 .

On taking square roots, we conclude |z + w| ≤ |z| + |w|, which is Part (h). 

The inequalities in the following theorem are used extensively – particularly in

the next section. The proofs are very simple and are left as an exercise (Exercise
1.1.12).

Theorem 1.1.8. If z = x + iy, then max{|x|, |y|} ≤ |z| ≤ |x| + |y|.

Inversion and Division. Recall that the inverse of a non-zero complex number
z = x + iy is
x − iy z z
z −1 = 2 = 2 = .
x +y 2 |z| zz
Stating this in the last form makes the identity zz −1 = 1 obvious.
This also suggests the right way to do complex division problems in general: to
express w/z as a complex number in standard form (as a real number plus i times
a real number), simply multiply both numerator and denominator by z. That is,
w wz wz
= = 2.
z zz |z|
The number wz is then easily put in standard form and the problem is finished by
dividing by the real number |z|2 .
1
Example 1.1.9. Express in the standard form x + yi.
2 + 3i
Solution:
1 2 − 3i 2 − 3i 2 3
= = = − i.
2 + 3i (2 + 3i)(2 − 3i) |2 + 3i|2 13 13
3 + 4i 3 − 4i
Example 1.1.10. Express and in standard form.
3 − 4i 3 + 4i
Solution:
3 + 4i (3 + 4i)2 7 24
= = − + i,
3 − 4i |3 + 4i|2 25 25
3 − 4i (3 − 4i)2 7 24
= = − − i.
3 + 4i |3 − 4i|2 25 25
8 1. The Complex Numbers

Exercise Set 1.1

1. Express (3 + i) + (2 − 7i) and (3 + i)(2 − 7i) in the standard form x + yi.

1
2. Express in the standard form x + yi.
3 + 5i
3. Express (1 + 2i)2 and (1 + 2i)−2 in the standard form x + yi.
2 − 3i
4. Express in the standard form x + yi.
3 + 2i
5. Find a square root for i.
6. If z = x + iy, express z 3 in standard form.
7. By direct substitution, prove that the two solutions to the quadratic equation
given in Example 1.1.3 really do satisfy equation (1.1.1).
8. Prove Part (e) of Theorem 1.1.4.
9. Prove Parts (a), (b) and (c) of Theorem 1.1.7.
10. Prove Parts (d), (e) and (f) of Theorem 1.1.7.
11. For z ∈ C and a ∈ R prove the following:
(a) Re(az) = a Re(z) and Im(az) = a Im(z);
(b) Re(iz) = − Im(z) and Im(iz) = Re(z).
12. Prove Theorem 1.1.8 – that is, prove that if z = x + iy, then
max{|x|, |y|} ≤ |z| ≤ |x| + |y|.
13. Graph the set of points z ∈ C which satisfy the equation |z − i| = 1.
14. Graph the set of points z ∈ C which satisfy the equation z 2 + z 2 = 2. Hint: If
z = x + iy, rewrite this equation as an equation in x and y.
15. Prove that if z is a non-zero complex number, then 1/z = 1/z and |1/z| = 1/|z|.
16. If z is any non-zero complex number, prove that z/z has modulus one.
17. Prove that every complex number of modulus 1 has the form cos θ + i sin θ for
some angle θ.
18. Prove that every line or circle in C is the solution set of an equation of the form
a|z|2 + wz + wz + b = 0,
where a and b are real numbers and w is a complex number. Conversely, show
that every equation of this form has a line, circle, point, or the empty set as its
solution set.

1.2. Convergence in C
We assume the reader is familiar with the basics concerning convergent sequences
and series of real numbers – particularly those results which follow from the com-
pleteness of the real number system, such as the fact that bounded monotone
sequences converge and the various convergence tests for series. We also assume
a familiarity with the basics of power series in a real variable. The purpose of
1.2. Convergence in C 9

this section is to extend these ideas and results to sequences and series of complex
numbers and power series in a complex variable. This is an introductory section.
A deeper study of complex power series will come later in the text.
There is no natural order relation on the complex numbers. The statement
z < w makes no sense for complex numbers z and w. However, if these numbers
happen to be real, then the inequality does make sense, because R is an ordered
field. We make heavy use of inequalities in this section, but note they are always
inequalities between real numbers. Thus, if an inequality of the form a < b occurs
in this text the numbers a and b are assumed to be real numbers, even if there is
no explicit statement to that effect.

Convergence of Sequences. A sequence of complex numbers converges if and

only if it converges as a sequence of vectors in R2 . The formal definition is the
familiar one:
Definition 1.2.1. A sequence {zn } of complex numbers is said to converge to the
number w if, for every  > 0, there exists an integer N such that
|zn − w| <  whenever n ≥ N.
In this case, we say that w is the limit of the sequence zn and write limn→∞ zn = w,
or lim zn = w, or simply zn → w.
Remark 1.2.2. There are a couple of simple observations about convergence of
sequences that will prove to be very useful.
(1) If {zn } is a sequence of complex numbers, then lim zn = w if and only if
lim |zn − w| = 0.
(2) If {an } and {bn } are sequences of real numbers with 0 ≤ an ≤ bn and if
bn → 0, then also an → 0.
These both follow immediately from the definition of limit of a sequence. The
second is one form of what is sometimes called the squeeze principle.

The first of these observations reduces the problem of showing that a sequence
of complex numbers converges to a given complex number to showing that a certain
sequence of non-negative real numbers converges to 0. This is useful because we
have many tools at our disposal to show that a sequence of non-negative numbers
converges to 0. One of the most useful of these tools is the second observation
above. The next example and the proof of the following theorem are excellent
examples of how this works.
Example 1.2.3. Prove that lim z n = 0 if |z| < 1.
Solution: Note that |z n | = |z|n follows from Part (e) of Theorem 1.1.7. If
|z| < 1 then lim |z|n = 0. That lim z n = 0, as well, follows from (1) of Remark
1.2.2.
Theorem 1.2.4. A sequence of complex numbers {zn } converges to a complex
number w if and only if {Re(zn )} converges to Re(w) and {Im(zn )} converges to
Im(w).
10 1. The Complex Numbers

Proof. By Theorem 1.1.8 we know that

| Re(zn ) − Re(w)| ≤ |zn − w|,
| Im(zn ) − Im(w)| ≤ |zn − w| and
|zn − w| ≤ | Re(zn ) − Re(w)| + | Im(zn ) − Im(w)|
for each n. The first two of these inequalities, together with Remark 1.2.2, im-
ply that if zn → w, then Re(zn ) → Re(w) and Im(zn ) → Im(w). The third
inequality, together with the fact that the sum of two sequences converging to
zero also converges to zero, implies the converse – that is, if Re(zn ) → Re(w) and
Im(zn ) → Im(w), then zn → w. 
Example 1.2.5. Show that the sequence {2−n + in/(n + 1)} converges to i.
Solution: This follows from the previous theorem and the fact that 2−n → 0
and n/(n + 1) → 1.
Example 1.2.6. Show that if {zn } and {wn } are two convergent sequences of
complex numbers with zn → z and wn → w, then zn +wn → z +w and zn wn → zw.
Solution: If zn = xn + iyn , z = x + iy, wn = un + ivn and w = u + iv, then
zn + wn = xn + un + i(yn + vn ),
z + w = x + u + i(y + v),
zn wn = xn un − yn vn + (xn vn + yn un )i
and
zw = xu − yv + (xv + yu)i.
We know that xn → x, yn → y, un → u and vn → v. We also know the rules
about limits of products, sums and differences of sequences of real numbers. These
rules imply xn + un → x + u, yn + vn → y + v, xn un − yn vn → xu − yv and
xn vn + yn un → xv + yu. Since the real and imaginary parts of zn + wn and zn wn
converge to the real and imaginary parts, respectively, of z +w and zw, the previous
theorem implies that zn + wn → z + w and zn wn → zw.

Series of Complex Numbers. A series of complex numbers is a formal sum of

the form
∞
zk = z0 + z1 + z2 + · · · + zk + · · · ,
k=0
with zk ∈ C. The series converges if its sequence of partial sums {sn } converges,
where
 n
(1.2.1) sn = zk .
k=0
In this case, if s = limn→∞ sn , then we say s is the sum of the series and write
∞
s= zk .
k=0
∞
Just as with real series, if a series k=0 zk converges, then its terms must tend
to zero. Thus, if {zk } fails to have limit 0, then the series diverges. This test for
1.2. Convergence in C 11

divergence is called the term test. Its proof for complex series is the same as the
proof for real series. We leave it as an exercise (Exercise 1.2.9).
Example 1.2.7. For what values of z does the complex geometric series
∞
zk
k=0
converge and what does it converge to?
Solution: We first note that if |z| ≥ 1, then |z|k ≥ 1 for all k and the series
diverges by the term test. For |z| < 1 we use the same trick that is used to study
the real geometric series. The nth partial sum of the series is
n
sn = zk .
k=0

If we multiply this by (1 − z), a vast cancellation of terms occurs and we obtain

(1 − z)sn = 1 − z n+1 ,
so that
1 − z n+1
sn = .
1−z
1
This sequence converges to if |z| < 1, since
1−z
 
 
sn − 1  = |z|
n
  →0
1−z |1 − z|
in this case.
∞
∞A series k=0 zk is said to converge absolutely if the series of positive terms
k=0 |zk | converges. From calculus, we know a great deal about convergence of pos-
itive termed series (comparison test, ratio test, root test, etc.), and so the following
theorem will obviously play an important role.
∞
Theorem 1.2.8. If a series k=0 zk of complex numbers converges absolutely,
then it converges. Furthermore,
∞ 
   ∞
 
 zk  ≤ |zk |.
 
k=0 k=0
∞
Proof. By hypothesis, the series k=0 |zk | converges. If zk = xk + iyk , then
|xk | < |zk | and |yk | < |zk |. By the comparison test, the two series
∞
 ∞

|xk | and |yk |
k=0 k=0
both converge. This, in turn, implies that
∞ ∞

xk and yk
k=0 k=0
converge, since an absolutely convergent series of real numbers converges. This
tells us that the real and imaginary parts of the sequence {sn } of partial sums of
12 1. The Complex Numbers

∞
k=0 zk are convergent and, by Theorem 1.2.4, so is the sequence {sn } itself. Thus,
the series converges.
It follows from the triangle inequality that
 n 
   n
 
|sn | =  zk  ≤ |zk |.
 
k=0 k=0
The inequality of the theorem follows when we pass to the limit as n → ∞ in this
inequality and use the result of Exercise 1.2.7. 
∞ 2 −1
Example 1.2.9. Prove that the complex series k=1 (k + k i) converges abso-
lutely.
Solution: By the second form of the triangle inequality (see Exercise 1.2.2)
we have
k2
|k + k2 i| ≥ k2 − k ≥ if k ≥ 2.
2
Thus,
|k + k2 i|−1 ≤ 2k−2 if k ≥ 2.
∞ −2 ∞ −2
Since
∞ the p-series k=1 k
 converges, so does k=1 2k and, by comparison,
2 −1 ∞ 2 −1
k=1 |k + k i| . Thus, k=1 (k + k i) converges absolutely.

Power Series. A complex power series is a series of the form

∞
(1.2.2) an (z − z0 )n ,
n=0
where the coefficients an are complex numbers, z0 is a complex number and z is a
complex variable. A power series of this form is said to be centered at z0 . It defines
a complex function of the complex variable z, with domain the set of those z ∈ C
for which the series converges.
Remark 1.2.10. We know from calculus that the set on which a real power series
∞

an (x − x0 )n
n=0
converges is an interval – the interval of convergence – consisting of an open interval
centered at x0 and possibly one or both of its endpoints. The power series converges
absolutely at each point of the open interval. The radius of this interval is called
the radius of convergence of the power series and is computed using the root test
or the ratio test. As we shall see in Chapter 3, a similar result is true for complex
power series, but the interval of convergence is replaced by a disc of convergence.
If for r > 0 we set
Dr (z0 ) = {z ∈ C : |z| < r} and Dr (z0 ) = {z ∈ C : |z| ≤ r},
then Dr (z0 ) is called the open disc of radius r, centered at z0 , while Dr (z0 ) is
called the closed disc of radius r, centered at z0 . Given a power series, centered at
z0 , there is a number R ≥ 0, called the radius of convergence of the power series,
such that the series converges absolutely for each z ∈ DR (z0 ) and diverges for each
z∈ / DR (z0 ). When we study power series in detail, we will prove this result and tell
how to calculate R in general. However, for most of the series we will be studying,
1.2. Convergence in C 13

there are elementary ways to show that this result holds and to find R. One simply
uses the standard convergence tests from calculus – particularly the ratio test.
Example 1.2.11. Find the radius of convergence of the series
∞
zn
(1.2.3) .
n=1
n

Solution: If we apply the ratio test to the series ∞ n=1 |z| /n, we conclude
n

that this series converges for |z| < 1 and diverges for |z| > 1. Hence, by Theorem
1.2.8, the series (1.2.3) also converges for |z| < 1. If |z| > 1, then the sequence of
terms of this series fails to converge to zero (since |z|n /n → +∞) and so the series
diverges by the term test. Thus our series converges on D1 (0) and diverges outside
of D1 (0). We conclude that it has radius of convergence 1.
Example 1.2.12. Show that the radius of convergence of the series
∞
zn
(1.2.4)
n=0
n!
is +∞ – that is, the series converges for all z.
∞
Solution: We apply the ratio test to the series n=0 |z|n /n!. We have
   n −1
|z|n+1 |z| |z|
=
(n + 1)! n! n+1
which converges to 0 for all z. Hence, by the ratio test, the power series (1.2.4)
converges for all z ∈ C. The radius of convergence is, thus, +∞.
Example 1.2.13. Find the radius of convergence of the power series
∞
an z n
n=0
where an = 2n if n is prime and an = 0 if n is not prime.
Solution: We cannot use the ratio test on this onebecause most of the terms
∞
of
∞ series
the are 0. However, if we compare the series n=0 an |z|n with the series
n=0 2 |z| – a series to which we can apply the ratio test – we conclude that the
n n

series converges for |z| < 1/2. If |z| > 1/2, then the terms of our series do not tend
to 0 and so the series diverges by the term test. We conclude that the radius of
convergence is 1/2.

Exercise Set 1.2

1. Show that the sequence (2 + ni)−1 converges to 0.

2. Prove the second form of the triangle inequality for complex numbers: | |z| −
|w| | ≤ |z − w|.
 
 1  1
3. Show that   ≤ .
z+5  |z| − 5
 √ √ n
4. Does the sequence 1/ 2 + i/ 2 converge?
14 1. The Complex Numbers

5. For which values of z does the sequence {z n } converge?

6. Prove that if a sequence {zn } converges, then it is bounded – that is, there is a
positive number M such that |zn | ≤ M for all n. Hint: Show that there is an
N such that |zn | ≤ |z| + 1 for all n ≥ N . This gives an upper bound for |zn |
for all but finitely many of the zn .
7. Prove that if {zn } is a sequence with lim zn = w, then lim zn = w and lim |zn | =
|w|.
8. For the sequence of the previous exercise, prove that lim 1/zn = 1/w provided
w = 0.
9. Prove the term test for divergence of a series.
 That is, if the sequence of terms
{zk } does not have limit 0, then the series ∞ k=0 zk diverges.
∞
10. Does the series n=0 n/(3 + 2ni) converge?
∞
11. Does the series n=1 n/(n3 + 2i) converge?

12. For which values of z does the series ∞ 2 2
n=0 1/(n + z ) converge?
∞
13. Find the radius of convergence of the power series n=0 nz n .
∞
14. Find the radius of convergence of the power series n=0 z n /3n .
∞
15. Find the radius of convergence of the power series n=0 z n /(1 + 2n ).
∞
16. Find the radius of convergence of the power series n=0 (n!/nn )z n .

1.3. The Exponential Function

There are many real-valued functions of a real variable that have natural extensions
to complex-valued functions of a complex variable. In fact, this  is true of all real
∞
functions that have convergent power series expansions. If f (x) = n=0 an (x−x0 )n
is a real power series which converges
 on an interval of radius R about x0 , then the
complex power series f (z) = ∞ a
n=0 n (z − x0 )n converges in the open disc of radius
R about x0 (Exercise 1.3.10) and serves to extend f to a function f (z) defined for
z in the disc DR (x0 ).
One of the most important examples of the use of this technique is provided
by the exponential function. We know from calculus that
∞
x xn
e = exp x = ,
n=0
n!
where the series converges to ex on the entire real line. We saw in Example 1.2.12
that this same series, with x replaced by the complex variable z, converges abso-
lutely on the entire complex plane. Thus, we get an extension of ex to a function
ez defined on C as follows:
Definition 1.3.1. For each z ∈ C, we define exp(z) = ez by
∞
z zn
e = .
n=0
n!
1.3. The Exponential Function 15

This is the complex exponential function. It has many important properties,

some expected and some surprising, as we shall see below.

The Law of Exponents. A property that we should expect of an exponential

function is the following.
Theorem 1.3.2. If z, w ∈ C, then ez+w = ez ew .

Proof. The proof uses the complex form of the binomial formula:
n
n!
(1.3.1) (z + w)n = z j wn−j .
j=0
j!(n − j)!

This is proved using induction on n. We leave it as an exercise (Exercise 1.3.11).

Proceeding with the proof of the theorem, and using (1.3.1), we have
∞
(z + w)n
ez+w =
n=0
n!
(1.3.2) ∞
 n
1
= z j wn−j .
n=0 j=0
j!(n − j)!
If we make a change of variables by setting k = n − j in the inside summation, then
(1.3.2) becomes ⎛ ⎞
∞  z j wk
⎝ ⎠.
n=0
j!k!
j+k=n
This is precisely what we get if we expand the product
⎛ ⎞ 
∞ j ∞
 k
z w
e e =⎝
z w ⎠ .
j=0
j! k!
k=0

and collect terms of degree n. Provided this operation is valid, we conclude that
ez+w = ez ew . It turns out that it is valid to expand the product of two infinite
series in this fashion if they are both absolutely convergent. We prove this in the
following lemma, which will complete the proof of the theorem. 
∞ ∞
Lemma 1.3.3. Let j=0 aj and k=0 bk be two absolutely convergent series of
complex numbers. Then
⎛ ⎞  ⎛ ⎞
∞ ∞ ∞
 
(1.3.3) ⎝ aj ⎠ bk = ⎝ aj bk ⎠ .
j=0 k=0 n=0 j+k=n

Proof. The partial sums of the series involved here are

⎛ ⎞
 J K N 
sJ = a j , tK = bk , uN = ⎝ aj bk ⎠ .
j=0 k=0 n=0 j+k=n

The left side of (1.3.3) is, by definition, (lim sJ )(lim tK ), while the right side is
lim uN . We know lim sJ and lim tK both exist since the series defining them con-
verge absolutely. We must prove that lim uN exists and equals (lim sJ )(lim tK ).
16 1. The Complex Numbers

For a given pair J, K, we let N = J + K. Then uN is the sum of all terms aj bk

for which j + k ≤ N , while sJ tK is the sum of those terms aj bk for which j ≤ J
and k ≤ K. Thus,

|uN − sJ tK | ≤ {|aj bk | : j + k ≤ N, and either j > J or k > K}
∞
 ∞
 ∞
 ∞

(1.3.4)
≤ |aj | |bk | + |aj | |bk |.
j=J+1 k=0 j=0 k=K+1
∞ ∞
The sum j=J+1 |aj | converges
∞
to zero as J → ∞ because the series
∞ j=0 aj
converges absolutely, while k=K+1 |bk | converges to 0 as K → ∞ because k=0 bk
converges absolutely.
To complete the proof, we choose for each non-negative integer N a J and K
with J + K = N , and we do this in such a way that as N → ∞ the corresponding
J and K both also tend to ∞. For example, we could choose J = K = N/2 if N is
even, and J = K + 1 = N/2 + 1/2 if N is odd. Then, from (1.3.4) it is clear that
lim |uN − sJ sK | = 0, which implies lim uN = (lim sJ )(lim sK ). 

The Exponential of an Imaginary Number. If z = x + iy, then the law of

exponents (Theorem 1.3.2) implies that
ez = ex eiy .
We understand the behavior of ex for real x from calculus. It is 1 at 0, is everywhere
positive, increasing and concave upward; it rapidly approaches +∞ as x → ∞ and
rapidly approaches 0 as x → −∞. What about eiy – the exponential of a purely
imaginary number? Here there are some surprises.
By the law of exponents, eiy e−iy = e0 = 1. Thus, 1/ eiy = e−iy . But also, for
any z ∈ C, ez = ez (Exercise 1.3.1) and, in particular, eiy = e−iy = 1/ eiy . This
implies that | eiy |2 = 1 and, consequently, | eiy | = 1. Thus, eiy is always a point on
the circle of radius 1 centered at 0 (we call this the unit circle).
A more explicit description of the number eiy comes from examination of its
power series definition. If we group together the even numbered terms (terms with
n of the form n = 2k) and the odd numbered terms (terms with n of the form
n = 2k + 1) in this power series, we derive the identity
∞ ∞
 ∞

(iy)n y 2k y 2k+1
eiy = = (−1)k +i (−1)k
n=0
n! (2k)! (2k + 1)!
k=0 k=0
= cos y + i sin y.
Thus, we have proved the following theorem, which is known as Euler’s Identity.
Theorem 1.3.4. The identity eiy = cos y + i sin y holds for all y ∈ R.

This shows that, not only is eiy a point on the unit circle, it is the point which
is reached by rotating through an angle y (measured in radians) from the initial
point (1, 0).
Example 1.3.5. Express the complex numbers e2πi , eπi and eπi/2 in standard
form.
1.3. The Exponential Function 17

Solution: By Euler’s identity, we have

e2πi = cos 2π + i sin 2π = 1,
eπi = cos π + i sin π = −1,
while
eπi/2 = cos π/2 + i sin π/2 = i.

Properties of the Exponential Function. The law of exponents and Euler’s

identity immediately imply:
Theorem 1.3.6. If z = x + iy is a complex number, then ez = ex (cos y + i sin y).

There are a number of properties of the exponential function which follow easily
from this characterization of ez . We collect them together in the following theorem
whose proof is left to the exercises.
Theorem 1.3.7. The exponential function has the following properties:
(a) ez is never 0;
(b) | ez | = eRe(z) ;
(c) | ez | ≤ e|z| ;
(d) ez is periodic of period 2πi, meaning ez+2πi = ez for every z ∈ C;
(e) ez = 1 if and only if z = 2πni for some integer n.

Complex Trigonometric Functions. If we write out Euler’s identity with y

replaced by θ and then by −θ, we obtain a pair of identities
eiθ = cos θ + i sin θ,
e−iθ = cos θ − i sin θ.
If we solve this system of equations for cos θ and sin θ, we obtain
eiθ + e−iθ
cos θ = ,
(1.3.5) 2
eiθ − e−iθ
sin θ = .
2i
This suggests defining sin z and cos z for a complex variable z in the following way:
Definition 1.3.8. For each z ∈ C, we set
eiz + e−iz
cos z = ,
2
eiz − e−iz
sin z = .
2i
These are the same functions that one would get by replacing x by z in the
power series expansions of sin x and cos x.
With sin z and cos z defined, it is easy to define complex versions of the other
trigonometric functions. For example,
sin z eiz − e−iz
(1.3.6) tan z = = −i iz .
cos z e + e−iz
18 1. The Complex Numbers

Example 1.3.9. How would you define a complex version of the arctan function?
Solution: The power series which converges to arctan x on (−1, 1) is
∞
 x2n+1
arctan x = (−1)n .
n=0
2n + 1
If we replace x by the complex variable z, we obtain a series which converges on
the disc D1 (0). We then use it to define arctan z on this disc:
∞
 z 2n+1
arctan z = (−1)n .
n=0
2n + 1

Exercise Set 1.3

1. Using the power series for ez , prove that ez = ez for each z ∈ C.

2. Express the complex numbers eπi/4 , e3πi/2 , and e13πi/6 in standard form.
√ √
3. Express 1/ 2 + i/ 2 in the form ez for some complex number z.
√
4. Find all values of z for which ez = 1 + 3 i.
5. Prove that ez is never 0 (Part (a) of Theorem 1.3.7).
6. Prove that | ez | = eRe(z) ≤ e|z| for each z ∈ C (Parts (b) and (c) of Theorem
1.3.7).
7. Verify that ez+2πi = ez for every z ∈ C (Part (d) of Theorem 1.3.7).
8. Verify that ez = 1 if and only if z = 2πni for some interger n (Part (e) of
Theorem 1.3.7).
9. For which values of z is ez a real number? For which values is it an imaginary
number?
∞ n
10. Show that if the real power series n=0 an x has radius of convergence R,
∞ n
then the corresponding complex power series n=0 an z also has radius of
convergence R (you may assume radius of convergence for a complex power
series makes sense and has the properties described in Remark 1.2.10).
11. Use induction on n to prove the complex binomial formula (1.3.1).
12. Derive the trigonometric identities
sin(z + w) = sin z cos w + cos z sin w,
cos(z + w) = cos z cos w − sin z sin w
from the law of exponents (Theorem 1.3.2) and Euler’s identity (Theorem
1.3.4).
13. Show that cos ix = cosh x and sin ix = i sinh x, where cosh and sinh are the
hyperbolic cosine and hyperbolic sine from calculus.
14. Show that
sin(x + iy) = sin x cosh y + i cos x sinh y and
cos(x + iy) = cos x cosh y − i sin x sinh y.
1.4. Polar Form for Complex Numbers 19

15. How would you define a complex version of the function log(1 + x)?
16. Verify that the power series defining arctan in Example 1.3.9 converges on the
disc D1 (0).

1.4. Polar Form for Complex Numbers

If u is a complex number of modulus 1, then u = cos θ + i sin θ = eiθ , where θ is
an angle from the positive x axis to the ray from the origin through u. This angle
is measured in radians and is the signed length of an arc on the unit circle joining
(1, 0) to u. It is positive if the arc is traversed in the counterclockwise direction
and negative if it is traversed in the clockwise direction.
If z is any non-zero complex number, then z/|z| is a complex number of modulus
1 and so it has the form eiθ for some θ. If we set r = |z|, then
z = r eiθ .
This is the polar form of the complex number z. The angle θ is called the argument
of z and is denoted arg z.
Comments:
(1) By Euler’s identity (Theorem 1.3.4), r eiθ = r(cos θ + i sin θ).
(2) The number z must be non-zero for the argument to be defined. The polar
form of z = 0 is just z = 0.
(3) For z = 0, there are infinitely many angles θ for which z = r eiθ . Given one
such θ the others are all of the form θ + 2πn, where n is an integer. Thus,
arg(z) is not a single number, but a collection of numbers that differ from
one another by multiples of 2π. We can specify a particular one of these by
insisting it lie in a particular half-open interval of length 2π – such as [0, 2π)
or (−π, π].
(4) The numbers (r, θ), where r = |z| and θ is a value of arg(z), are polar coordi-
nates for z = x + iy considered as a point (x, y) in R2 .
(5) We now have two useful ways of expressing a complex number z, the standard
form x + iy and the polar form r eiθ . It is important to be able to easily
change from one to the other.

Example 1.4.1. Find the polar form for the complex number z = 2 + 2i.
Solution: We have
 √
r = 22 + 22 = 2 2 and θ = arctan(1) = π/4.
√
Thus, z = 2 2 eπi/4 is the polar form of z.
Example 1.4.2. Put the complex number z = 2 eπi/6 in standard form x + iy.
Solution: We have
√
x = 2 cos π/6 = 3 and y = 2 sin π/6 = 1
√
and so z = 3 + i.
20 1. The Complex Numbers

r e iθ

r
θ

Figure 1.4.1. The Polar Form r eiθ of a Complex Number.

Products, Powers, and Roots. Polar form is particularly useful in dealing with
products of complex numbers, since the product has such a simple expression if the
numbers are given in polar form. If z1 = r1 eiθ1 and z2 = r2 eiθ2 , then the law of
exponents implies that
(1.4.1) z1 z2 = r1 r2 ei(θ1 +θ2 ) .
Thus, z1 z2 is the complex number whose norm (distance from the origin) |z1 z2 | is
the product of |z1 | and |z2 | and whose argument is the sum of the arguments of z1
and z2 .
Similarly, the quotient of z1 and z2 is given by
(1.4.2) z1 /z2 = (r1 /r2 ) ei(θ1 −θ2 ) .
Example 1.4.3. Find z1 z2 and z1 /z2 if z1 = 2 eπi/3 and z2 = 3 e2πi/3 .
Solution: By (1.4.1) and (1.4.2), we have
√
z1 z2 = 6 eπi = −6 and z1 /z2 = (2/3) e−πi/3 = 1/3 − i 3/3.

It is evident from (1.4.1) that the nth power of a complex number z = r eiθ is
(1.4.3) z n = r n einθ .

From this we conclude that if z = r eiθ , and we choose w = r 1/n eiθ/n , then
n
w = z. Thus, w is an nth root of z. It is not the only one, however. Since z can
also be written as z = ei(θ+2πk) for any integer k, each of the numbers
wk = r 1/n ei(θ/n+2πk/n) ,
where k is an integer, is also an nth root of z. Of course, these numbers are not all
different. Those whose arguments differ by an integral multiple of 2π are the same.
The numbers w0 , w1 , w2 , · · · , wn−1 are all distinct, but every other wk is equal to
one of these. This proves the following theorem.
Theorem 1.4.4. If z = r eiθ is a non-zero complex number, then z has exactly n
nth roots. They are the numbers
r 1/n ei(θ/n+2πk/n) for k = 0, 1, 2, · · · , n − 1.
1.4. Polar Form for Complex Numbers 21

−1/2+i 3/2

−1/2−i 3/2

Figure 1.4.2. The Cube Roots of Unity.

Note that the n numbers ei(θ/n+2πk/n) that appear in this result are evenly
spaced around the unit circle, with each successive pair separated by an angle of
2π/n.
The nth roots of the number 1 play a special role. They are called the nth roots
of unity. If we apply the above theorem in the special case where r = 1 and θ = 0,
it tells us that the nth roots of unity are the numbers
(1.4.4) e2πki/n for k = 0, 1, 2, · · · , n − 1.
Example 1.4.5. Find the cube roots of unity.
Solution: In the particular case where n = 3, (1.4.4) tells us that the roots of
unity are
e0 = 1,
√
e2πi/3 = −1/2 + ( 3/2)i, and
√
e4πi/3 = −1/2 − ( 3/2)i.
Example 1.4.6. Find the cube roots of 2i.
Solution: Since 2i = 2 eπi/2 , Theorem 1.4.4 implies that the cube roots of 2i
are
√
21/3 eπi/6 = 21/3 ( 3/2 + i/2),
√
21/3 ei(π/6+2π/3) = 21/3 e5πi/6 = 21/3 (− 3/2 + i/2), and
21/3 ei(π/6+4π/3) = 21/3 e3πi/2 = −21/3 i.

The Logarithm. If z = r eiθ , and log r is the natural logarithm of the positive
number r, then the law of exponents implies that
z = elog r+iθ .
Thus, it would make sense to define log z to be log r + iθ = log |z| + i arg z. There is
a problem with this, however. There are infinitely many possible choices for arg z,
and so log z is not well defined, just as arg z is not well defined.
The solution to this problem is to restrict θ = arg z to lie in a specific half-open
interval of length 2π.
22 1. The Complex Numbers

z iπ

iπ/2
z log(z)
0 1 2 0 1 2
−iπ/2

−iπ

Figure 1.4.3. Principal Branch of the Log Function on an Annulus.

Definition 1.4.7. Given a half-open interval I of length 2π on the line R, let

argI z, for z = 0, be the value of arg z that lies in the interval I. Then the function
defined for z = 0 by
log z = log |z| + i argI z
will be called the branch of the log function defined by I. In the special case where
I = (−π, π], this function will be called the principal branch of the log function.

The following properties of the various branches of the log function follow easily
from the definition and the properties of the exponential function. The proofs are
left to the exercises.
Theorem 1.4.8. If log is the branch of the the log function determined by an
interval I, then
(a) if z = 0, then elog z = z;
(b) if z ∈ C, then log ez = z + 2πki for some integer k;
(c) if z, w ∈ C, then log zw = log z + log w + 2πki for some integer k;
(d) log 1 = 2πki for some integer k;
(e) log agrees with the ordinary natural log function on the positive real numbers
if and only if the interval I contains 0.

Suppose the interval I defining a branch of the log function has endpoints a
and b with a < b. Then, since b − a = 2π, the polar coordinate equations θ = a
and θ = b define the same ray. This ray is called the cut line for this branch of the
logarithm. Observe that if z and w are two complex numbers with |z| = |w| > 0
which are close to each other, but on opposite sides of the cut line – say with argI z
near a and argI w near b – then log w − log z is nearly 2πi. In other words, as we
cross the cut line moving in the clockwise direction, the value of log jumps by 2πi.
Thus, log is not continuous at points on the cut line. Later we will show that it is
continuous everywhere else.
1.4. Polar Form for Complex Numbers 23

Other Functions. If we fix a branch of the complex log function, we can define

a number of other complex functions which are extensions of familiar real functions.
We mention some of these briefly.
An nth root function can be defined by setting 01/n = 0 and
(1.4.5) z 1/n = e(1/n) log z if z = 0.
√
A special case of this is the square root function defined by 0 = 0 and
√
(1.4.6) z = e(1/2) log z if z = 0.

Note that the nth root function, as defined by (1.4.5), is giving only one of the
nth roots of z. Which one is determined by the branch of the log function that is
used. The other nth roots are obtained by multiplying this one by the nth roots
of unity (Exercise 1.4.9). For example, given one square root of z, the other is
obtained by multipling it by −1.
√
Example 1.4.9. If z √ is defined using the principal branch of the log function,
analyze the behavior of z near the cut line for this branch.
Solution: For the principal branch of the log function the interval I is (−π, π]
and so the cut line is the line defined by θ = π in polar coordinates – that is, it is
the negative real axis. If z = r eiθ is just above the negative real axis, then log z is
nearly log r + iπ and
√
z = e(1/2) log z
√
is nearly i r. On the other hand, if z is just below the negative
√ axis, then √
log z is
nearly log r − iπ, (1/2) log z is close to √(log r − iπ)/2 and z is close to −i r. In
other words, as z crosses the cut line, z jumps from one square root of z to its
negative, which is the other square root of z.
√
Example 1.4.10. Analyze the function z 2 + 1, where the square root function
is defined, as above, using the principal branch of the log function.
√
Solution: The cut√ line for z is the same as for log – the half-line of negative
reals. The function z jumps from values on the positive imaginary axis to their
negatives as z crosses this line in the counterclockwise direction. The number z 2 +1
crosses the negative real half-line in the counterclockwise direction as z crosses either
(i, i∞) or (−i∞, −i) in the counterclockwise direction. √ Thus, these two half-lines
on the imaginary axes are where discontinuities of z 2 + 1 occur. As either of
these half-lines is crossed in the counterclockwise direction, a typical value of this
function jumps from a number it with t > 0 to −it (see Figure 1.4.4).

Raising a complex number to a complex power is another function that can be

defined using a branch of the log function. If z = 0 and a is any complex number,
then we set
(1.4.7) z a = ea log z .
Here, we are thinking of a as being fixed and z is the independent variable of the
function. If we want the exponent to be the variable, we would write
(1.4.8) az = ez log a .
24 1. The Complex Numbers

−it it

it −it

√
Figure 1.4.4. The Cut Line Discontinuities of z 2 + 1.

Note that for any function defined in terms of a branch log z of the log func-
tion, we can expect trouble along the cut line for log. Since, log has a jump or
discontinuity as we cross the cut line, we can expect the same for functions defined
in terms of it.
Also, we emphasize that functions defined this way depend on the choice of a
branch of the log function. If one wants such a function to agree with the standard
one on the positive real axis, then one must choose a branch of the log function
which is the ordinary natural logarithm on the positive real numbers. By Theorem
1.4.8, Part(e), this happens if and only if the interval I, in which arg(z) is required
to lie, contains 0. In particular, the principal branch has this property.

Exercise Set 1.4

√
1. Put each of the complex numbers −1, i, −i, 1 + 3 i, and 5 − 5i in polar form.
2. Put each of the complex numbers e4πi , 3 e2πi/3 , 5 e5πi/2 , and 2 e−3πi/4 in stan-
dard form.
3. Find all powers of eπi/8 . How many distinct powers of this number are there?
4. Show that (1 − i)7 = 8(1 + i) by converting to polar form, taking the seventh
power, and then converting back to standard form.
5. Using a calculator, calculate (1.2 e.5 i )n for n = 1, · · · , 6 and graph the resulting
points. Do the same for (.8 e.5 i )n .
6. Prove that if z is a number on the unit circle, then z has finitely many distinct
powers z n if and only if the argument of z is a rational multiple of 2π.
7. What are the 4th roots of unity?
8. What are the cube roots of −9?
9. Show that, given one nth root of z, the others are obtained by multiplying it
by the nth roots of unity.
10. Find argI z if (a) z = −i and I = (−π, π], (b) z = −i and I = [0, 2π), (c) z = 1
and I = [3π/2, 7π/2).
1.4. Polar Form for Complex Numbers 25

11. For the principal branch of the log function, find log(1 − i).
12. Find log(1 − i) for the branch of the log function determined by the interval
[0, 2π).
13. Prove (a) and (b) of Theorem 1.4.8.
14. Prove (c), (d), and (e) of Theorem 1.4.8.
15. Analyze the function z i defined by (1.4.7) using the principal branch of the log
function. What kind of a jump, if any, does it have as z crosses the negative
real axis?
√
16. Analyze the function 1 − z 2 , where the square root function is defined by
the principal branch of the log function. Where does it have discontinuities
(jumps)?
17. Let the square root function be √ defined by the √ principal
√ branch of the log
function. Compare the functions z 2 − 1 and z + 1 z − 1. Where are the
discontinuities of each function?
1 − z n+1
18. The identity 1 + z + z 2 + · · · + z n = was derived in Example 1.2.7.
1−z
Use this to derive Lagrange’s trigonometric identity:
1 sin (n + 1/2)θ
1 + cos θ + cos 2θ + · · · + cos nθ = + .
2 2 sin θ/2
Hint: Take the real parts of both sides in the first identity.
 Chapter 2

Analytic Functions

Complex Variables is the study of functions of a complex variable – specifically,

analytic functions of a complex variable. These are functions which have a complex
derivative in a sense that we shall define shortly. Analytic functions have amazing
properties, and deriving these properties will be the main focus of the text. The
amazing properties of analytic functions stem from the fact that the existence of
a complex derivative is a much more powerful condition than the existence of the
real derivative for a function of a real variable.
We have already seen several examples of analytic functions. The exponential
function and the sine and cosine functions are analytic functions on the entire plane.
The log and nth root functions are analytic except on the cut line for the branch
of the log function that is used in their definitions.
We must prove one difficult theorem about analytic functions (the Cauchy
Integral Theorem), and then a wealth of amazing theorems with simple proofs will
follow.
We begin this chapter with a review of the basic facts about continuous func-
tions in the plane. We then define analytic functions and prove some elementary
theorems concerning them. Next we introduce contour integration and prepare for
our assault on Cauchy’s Theorem. Finally, in Sections 2.5 and 2.6 we prove two
limited versions of this theorem. This provides us enough of the power of Cauchy’s
Theorem to allow us to move on to Chapter 3, where we develop a wide range of
applications. However, we put off proving the most general version of Cauchy’s
Theorem until Chapter 4.

2.1. Continuous Functions

Since, geometrically, C is just R2 , notions of distance, convergence of sequences, and
continuity of functions are the same for C as for R2 . In particular, a complex-valued
function of a complex variable may be regarded as a function of two real variables
with values in R2 . The student who has had a course in multivariable calculus

27
28 2. Analytic Functions

already knows what it means for such a function to be continuous. Still, we will
review the basics of continuity in this context, partly in order to set terminology
and notation that is peculiar to the complex variables context.

Open and Closed Sets. Recall that the open disc Dr (z0 ) and closed disc
Dr (z0 ), centered at z0 , with radius r > 0, are defined by
Dr (z0 ) = {z ∈ C : |z − z0 | < r} and Dr (z0 ) = {z ∈ C : |z − z0 | ≤ r}.
Open intervals and closed intervals on the real line play an important part in
calculus in one variable. Open and closed discs are the direct analogues in C of
open and closed intervals on the line. However, the geometry of the plane is much
more complicated than that of the line. We will need the concepts of open and
closed for sets that are far more complicated than discs. This leads to the following
definition.
Definition 2.1.1. If W is a subset of C, we will say that W is open if, for each
point w ∈ W , there is an open disc centered at w which is contained in W . We
will say that a subset of C is closed if its complement is open. A neighborhood of
a point z ∈ C is any open set which contains z.

It might seem obvious that open discs are open sets and closed discs are closed
sets. However, that is only because we have chosen to call them open discs and
closed discs. We actually have to prove that they satisfy the conditions of the
preceding definition. We do this in the next theorem.
Theorem 2.1.2. We have:
(a) the empty set ∅ is both open and closed;
(b) the whole space C is both open and closed;
(c) each open disc is open;
(d) each closed disc is closed.

Proof. The empty set ∅ is open because it has no points, and so the condition that
a set be open, stated in Definition 2.1.1, is vacuously satisfied. The set C is open
because it contains any open disc centered at any of its points. Thus, ∅ and C are
both open. Since they are complements of one another, they are also both closed.
To prove (c), we suppose Dr (z0 ) is an open disc and w is one of its points.
Then |w − z0 | < r and so, if we set s = r − |w − z0 |, then s > 0. Also, if z ∈ Ds (w),
then |z − w| < s and so
|z − z0 | ≤ |z − w| + |w − z0 | < s + |w − z0 | = r,
and so z ∈ Dr (z0 ) (see Figure 2.1.1). Thus, we have shown that, for each w ∈
Dr (z0 ), there is an open disc, Ds (w), centered at w, which is contained in Dr (z0 ).
By definition, this means that Dr (z0 ) is open. This completes the proof of (c).
To prove (d), we consider a closed disc Dr (z0 ). To prove that it is a closed set,
we must show its complement is open. Suppose w is a point in its complement. This
means w ∈ C but w ∈ / Dr (z0 ), and so |w − z0 | > r. This time we set s = |w − z0 | − r
and we claim that the open disc Ds (w) is contained in the complement of Dr (z0 ).
2.1. Continuous Functions 29

s w

r s r
w
z0 z0

Figure 2.1.1. Proving Theorem 2.1.2 (c) and (d).

In fact, if z ∈ Ds (w), then |z − w| < s and so, by the second form of the triangle
inequality (Exercise 1.2.2),
|z − z0 | ≥ |w − z0 | − |z − w| > |w − z0 | − s = r,
which means z is in the complement of Dr (z0 ). Thus, we have proved that each
point of the complement of Dr (z0 ) is the center of an open disc contained in the
complement of Dr (z0 ). This proves that this complement is open, hence, that
Dr (z0 ) is closed. 

The next theorem shows that the collection of all open subsets of C forms what
is called a topology for C. It says that the collection of open subsets is closed under
arbitrary unions and finite intersections.
Theorem 2.1.3. The union of an arbitrary collection of open sets is open, while
the intersection of any finite collection of open sets is open. On the other hand,
the intersection of an arbitrary collection of closed sets is closed, while the union
of any finite collection of closed sets is closed.

Proof. If z0 ∈ C, V is an arbitrary collection of open sets, and U = V is its
union, then z0 is in U if and only if it is in at least one of the sets in V. Suppose, it
is in V ∈ V. Then, since V is open, there is a disc Dr (z0 ), centered at z0 , which is
contained in V . Then this disc is also contained in U . This proves that U is open.
Now suppose {V1 , V2 , · · · , Vk } is a finite collection of open sets and
z0 ∈ U = V1 ∩ V2 ∩ · · · ∩ Vn .
Then, since each Vk is open, there exists for each k a radius rk such that Drk (z0 ) ⊂
Vk . If r = min{r1 , r2 , · · · , rn }, then Dr (z0 ) ⊂ Vk for every k, which implies that
Dr (z0 ) ⊂ U . It follows that U is open.
The proofs of the statements for closed sets follow from those for open sets by
taking complements. We leave the details to Exercise 2.1.1. 

A consequence of the above theorem is: if U is open and K is closed, then the
set-theoretic difference U \ K is open. Similarly, K \ U is closed (Exercise 2.1.2).
Example 2.1.4. If 0 < r < 1, prove that the annulus A = {z ∈ C : r < |z| < R}
is open.
30 2. Analytic Functions

E E° E

Figure 2.1.2. The Set E of Example 2.1.7, its Interior E ◦ , and Closure E.

Solution: The disc DR (0) is open, the disc Dr (0) is closed, and A is the
set-theoretic difference DR (0) \ Dr (0). Thus, by the previous remark, A is open.
Interior, Closure, and Boundary. If E is a subset of C, then E contains a
largest open subset, meaning an open subset of E that contains all other open
subsets of E. In fact, the union of all open subsets of E is open, by Theorem 2.1.3,
and is a subset of E which contains all open subsets of E. Similarly, the intersection
of all closed sets containing E is the smallest closed set containing E. Thus, the
following definition makes sense.
Definition 2.1.5. Let E be a subset of C. Then:
(a) the largest open subset of E is called the interior of E and is denoted E ◦ ;
(b) the smallest closed set containing E is called the closure of E and is denoted
E;
(c) the set E \ E ◦ is called the boundary of E and is denoted ∂E.
Recall that a neighborhood of a point z0 ∈ C is any open set containing z0 .
The proof of the following theorem is elementary and is left to the exercises.
Theorem 2.1.6. Let E be a subset of C and z an element of C. Then:
(a) z ∈ E ◦ if and only if there is a neighborhood of z that is contained in E;
(b) z ∈ E if and only if every neighborhood of z contains a point of E;
(c) z ∈ ∂E if and only if every neighborhood of z contains points of E and points
of the complement of E.
Example 2.1.7. Find the interior, closure and boundary for the set
E = {z ∈ C : |z| < 1, Im(z) ≥ 0} ∪ {−iy : y ∈ [0, 1]}.
Solution: It is immediate from the previous theorem that
E ◦ = {z ∈ C : |z| < 1, Im(z) > 0},
E = {z ∈ C : |z| ≤ 1, Im(z) ≥ 0} ∪ {−iy : y ∈ [0, 1]},
∂E = {z ∈ C : |z| = 1, Im(z) ≥ 0} ∪ [−1, 1] ∪ {−iy : y ∈ [0, 1]}.
2.1. Continuous Functions 31

Limits. We will be primarily concerned with complex-valued functions defined

on open subsets of C. However, we will also need to deal with functions whose
domain is some other kind of subset of C. For example, in the next section we will
be dealing with curves in C. A curve in C is a continuous complex-valued function
whose domain is an interval on the real line. We will also deal eventually with
functions defined on a closed disc or a circle or some other closed subset of C. The
definitions of limit and continuity we adopt here are general enough to handle all
these situations. However, as a result, they are very much domain dependent. That
is, the limit of a function at a point and whether or not the function is continuous
at a point depend very much on which set is considered to be the domain of the
function.
The definition of limit is the familiar one from calculus except for extra care
about the domain of the function and a condition concerning isolated points, as
defined below.
If E is a subset of C, then an isolated point of E is an element a ∈ E such that
there is a neighborhood of a which contains no other points of E.
Definition 2.1.8. If f is a complex-valued function with domain E, and a ∈ E
but is not an isolated point of E, then we say limz→a f (z) = L if, for every  > 0,
there is a δ > 0 such that |f (z) − L| <  whenever z ∈ E and 0 < |z − a| < δ. In
this case, we say that f has limit L at a.

The condition “z ∈ E” in this definition means that whether or not the limit
exists may depend on E. That is, given a function f defined on a set E and a
subset D of E, we can always consider a new function f |D which is f restricted
to the new domain D. It may be that the limit as z → a does not exist for f as
defined on its original domain E, but the limit of f |D does exist.
Example 2.1.9. Show that the function f (z) = z/|z| with domain D1 (0) \ {0}
does not have a limit as z → 0, but if this function is restricted to the domain
(0, 1) ⊂ R, then its limit as z → 0 is 1.
Solution: If f is considered as a function with domain D1 (0) \ {0}, then for
each z = r eiθ in this domain, f (z) = eiθ . Thus f takes on every value of modulus
one in every open disc Dδ (0) with δ < 1. So there is certainly no one number
that f (z) is approaching as z → 0. On the other hand, suppose f is restricted to
(0, 1) ⊂ R. On this interval, f is identically 1 and so its limit as z → 0 is also 1.

A deleted neighborhood of a ∈ C is a set of the form V \ {a} where V is

neighborhood of a. Using neighborhoods and deleted neighborhoods, the limit
concept may be rephrased as follows.
Theorem 2.1.10. If f is a complex-valued function with domain E, and if a ∈ E,
then limz→a f (z) = L if and only if for every neighborhood W of L there is a deleted
neighborhood V of a such that
(2.1.1) f (E ∩ V ) ⊂ W.

Proof. If limz→a f (z) = L and W is a neighborhood of L, then, since W is open

and contains L, there is an  > 0 such that D (L) ⊂ W . By the definition of limit,
32 2. Analytic Functions

there is a δ > 0 such that |f (z) − L| <  whenever z ∈ E and 0 < |z − a| < δ. This
is equivalent to the statement
f (E ∩ Dδ (a) \ {a}) ⊂ D (L) ⊂ W.
Thus, if we choose V = Dδ (a)\{a}, then V is a deleted neighborhood of a satisfying
(2.1.1). This proves the “only if” part of the theorem.
To prove the “if” part, we assume that for every neighborhood W of L there
is a deleted neighborhood V of a such that f (E ∩ V ) ⊂ W . Then, given  > 0,
let W = D (L) and let V be a deleted neighborhood of a satisfying (2.1.1). Since
V ∪ {a} is open, there exists a δ > 0 such that Dδ (a) ⊂ V ∪ {a}. Then
f (E ∩ Dδ (a) \ {a}) ⊂ f (E ∩ V ) ⊂ W = D (L).
This is equivalent to the statement that |f (z) − w| <  whenever z ∈ E and
0 < |z − a| < δ. Thus, by definition, limz→a f (z) = L. 

Continuity. The definition of a continuous function should also be familiar from

calculus.
Definition 2.1.11. A function f with domain E is said to be continuous at a ∈ E
if
lim f (z) = f (a).
z→a
If f is continuous at every point of E, then we say that f is continous on E. The
set of all functions with domain E, which are continuous on E, will be denoted
C(E).
Example 2.1.12. Show that |z| is a continuous function of z on all of C.
Solution: Let z and a be elements of C. By the second form of the triangle
inequality, we have
||z| − |a|| ≤ |z − a|.
It follows from this that limz→a |z| = |a| and, hence, that |z| is continuous at z. In
fact, given  > 0, it suffices to choose δ =  in the definition of limit, since |z −a| < 
implies ||z| − |a|| < .

The definition of continuous function is also very domain dependent. For ex-
ample, if the function f (z) = z/|z| of Example 2.1.9 is given the value 1 at 0, then
it is continuous at 0 as a function with domain [0, 1), but if we consider its domain
to be D1 (0), then it is not continuous.
The next result characterizes functions which are defined and continuous on an
open set U as those functions f on U for which f −1 preserves open sets.
Theorem 2.1.13. If f is a complex-valued function defined on an open set U ⊂ C,
then f is continuous on U if and only if f −1 (W ) is open for every open set W ⊂ C.

Proof. Suppose f is continous on U and W is an open subset of C. If a ∈ f −1 (W ),

then limz→a f (z) = f (a) ∈ W . Since W is a neighborhood of f (a), it follows from
Theorem 2.1.10 that there is a neighborhood V of a such that f (U ∩ V ) ⊂ W
(actually, Theorem 2.1.10 guarantees a deleted neighborhood V with this property,
but since f (a) ∈ W , it does not change things to put a back in the deleted neigh-
borhood and claim there is an actual neighborhood V as above). Thus, U ∩ V is
2.1. Continuous Functions 33

an open subset of f −1 (W ) containing a. We have now proved that every element

of f −1 (W ) is contained in an open set contained in f −1 (W ). The union of all such
open sets is open and is equal to f −1 (W ). Thus, f −1 (W ) is open.
To prove the converse, we suppose that f −1 (W ) is open for every open set
W . It follows that for each a ∈ U and each neighborhood W of f (a), we have
f −1 (W ) is a neighborhood of a. By Theorem 2.1.10, limz→a f (z) = f (a) and so f
is continuous at a. Since a was any point of U , f is continuous on U . 

If f and g are functions with domain a set E and they are both continuous at
a ∈ E, then f + g and f g are continuous at a, and f /g is continuous at a provided
g(a) = 0. The proofs of these facts for complex-valued functions of a complex
variable are no different than the proofs, given in calculus, of the corresponding
facts for functions of a real variable. Thus, we will accept them without further
comment.
Obviously, constants are continuous everywhere, as is the function z itself. It
follows that polynomials in z are also continuous everywhere.
If g with domain E is continuous at a and f with domain D is continuous at
b = f (a), and if g(E) ⊂ D, then the composite function f ◦ g, with domain E, is
continuous at a. This is another fact whose proof for functions of a complex variable
is no different than the proof from calculus of the analogous result for functions of
a real variable.
Example 2.1.14. Prove that if g is a continuous non-vanishing function on an
open subset U ⊂ C, then log |g| is continous on U .
Solution: The function |g(z)| is continuous on U because it is the composition
of the function g, which is continuous on U , and the function | · |, which is continous
everywhere. Also, |g(z)| has its values in the positive reals, since g is non-vanishing
on U . The function log is continuous on the positive reals and so the composition
log |g| is continuous on U .

Exercise Set 2.1

1. Prove the second statement of Theorem 2.1.3. That is, prove that the intersec-
tion of an arbitrary collection of closed sets is closed, while the union of any
finite collection of closed sets is closed.
2. Prove that if U is open and K is closed, then the set-theoretic difference U \ K
is open, while K \ U is closed.
3. Prove Theorem 2.1.6.
4. Show that the set A = {z ∈ C : Re(z) > 0} is open. Hint: You must show that
each point w ∈ A is the center of an open disc which is entirely contained in A.
5. Tell which of the following sets are open subsets of C, which are closed, and
which are neither (no proof required):
(a) {z ∈ C : 1 < |z| < 2};
(b) {z ∈ C : Im(z) = 0, 0 < Re(z) < 1};
(c) {z ∈ C : −1 ≤ Re(z) ≤ 1}, −1 ≤ Im(z) ≤ 1}.
34 2. Analytic Functions

6. Find the interior, closure, and boundary for the set {z ∈ C : 1 ≤ |z| < 2} (no
proof required).
7. Prove that w ∈ C is in the closure of a set E ⊂ C if and only if there is a
sequence {zn } ⊂ E such that lim zn = w. Thus, a set E is closed if and only if
it contains all limits of convergent sequences of points in E.
|z − z|
8. Does limz→0 f (z) exist if f (z) = with domain C \ {0}? How about if
|z|
the domain is restricted to be just R \ {0}?
9. Prove that Re(z), Im(z), and z are continuous functions of z.
10. At which points of C is the function (1 − z 4 )−1 continuous.
11. Prove that argI is continuous except on its cut line.
12. Use the result of the preceding exercise to prove that a branch of the log function
is continuous except on its cut line.
13. Use Theorem 2.1.13 to prove that if f and g are continuous functions with open
domains Uf and Ug and if g(Ug ) ⊂ Uf , then f ◦ g is continuous on Ug .
14. Prove that if f is a continuous function defined on an open subset U of C, then
sets of the form {z ∈ U : |f (z)| < r} and {z ∈ U : Re(f (z)) < r} are open.
15. Use the result of the preceding exercise to come up with an open subset of C
that has not been previously described in this text.
16. Prove that a function f with open domain U is continuous at a point a ∈ U
if and only if whenever {zn } ⊂ U is a sequence converging to a, the sequence
{f (zn )} converges to f (a).

2.2. The Complex Derivative

There is nothing surprising about the definition of the derivative of a function of
a complex variable – it looks just like the definition of the derivative of a function
of a real variable. What is surprising are the consequences of a function having a
derivative in this sense.
Definition 2.2.1. Let f be a function defined on a neighborhood of z ∈ C. If
f (w) − f (z)
lim
w→z w−z

exists, then we denote it by f (z) and we say f is differentiable at z with complex
derivative f  (z). If f is defined and differentiable at every point of an open set U ,
then we say that f is analytic on U .
Remark 2.2.2. When convenient, we will make the change of variables λ = w − z
and write the derivative in the form
f (z + λ) − f (z)
(2.2.1) f  (z) = lim .
λ→0 λ
Clearly constant functions are differentiable and have complex derivative 0,
since the difference quotient in Definition 2.2.1 is identically 0 for such a function.
2.2. The Complex Derivative 35

The first hint that there is something fundamentally different about this notion
of derivative is in the following example.

Example 2.2.3. Show that the function f (z) = z is differentiable everywhere on

C with derivative 1 and, hence, is analytic on C, but the function f (z) = z is
differentiable nowhere.
Solution: For f (z) = z, the difference quotient in (2.2.1) is
λ
= 1,
λ
which clearly has limit 1 as λ → 0 for every z. On the other hand, if f (z) = z, then
the difference quotient is
λ
= e−2iθ ,
λ
if λ = r eiθ in polar form. The limit of this function as λ → 0 clearly does not exist,
since it has a different fixed value along each ray emanating from 0. This is true
no matter what z is, and so z is nowhere differentiable.

What makes this example so surprising, at first, is that, as a function of the

two real variables x and y, z = x − iy is of class C∞ – meaning that its partial
derivatives of all orders exist and are continuous – and yet, its complex derivative
does not exist. Thus, existence of the complex derivative involves more than just
smoothness of the function.
We will soon prove that a function which has a power series expansion that
converges on an open disc is analytic on that disc. This would imply that the
exponential function, for example, is analytic on all of C. We do not have to wait,
however, to prove this fact. There is an elementary proof that ez is analytic on C.

Example 2.2.4. Prove that ez is an analytic function of z on the entire complex

plane and show that it is its own derivative.
Solution: Given an arbitrary point z ∈ C, we will show that ez has derivative
z
e at z. By the law of exponents

ez+λ − ez eλ −1
= ez .
λ λ
Thus, to show that the derivative of ez is ez we need only show that

eλ −1
(2.2.2) lim = 1.
λ→0 λ
However, if t = |λ|, inspection of the power series for eλ and et shows that
 λ   λ 
 e −1   e −1 − λ  et −1 − t
   ≤
(2.2.3)  λ − 1 =  λ  t
.

Now to show that the expression on the left has limit zero and, thus, verify (2.2.2),
we simply apply L’Hôpital’s rule to the expression on the right.
36 2. Analytic Functions

Elementary Properties of the Derivative. A simple result about derivatives

of functions of a real variable that also holds in the context of complex derivatives
is the following. The proof is elementary and is left to the exercises.
Theorem 2.2.5. If the complex derivative f  of f exists at a ∈ C, then f is
continuous at a.

The complex derivative has all of the familiar properties in relation to sums,
products, and quotients of functions. The proofs of these are in no way different
from the proofs of the corresponding results for functions of a real variable. In the
following theorem, Part (a) is trivial and we leave Parts (b) and (c) to the exercises.
Theorem 2.2.6. If f and g are functions of a complex variable which are differ-
entiable at z ∈ C, then
(a) f + g is differentiable at z and (f + g) (z) = f  (z) + g  (z);
(b) f g is differentiable at z and (f g) (z) = f  (z)g(z) + f (z)g  (z);
(c) if g(z) = 0, 1/g is differentiable at z and (1/g) (z) = −g  (z)/g 2 (z).

Parts (a) and (b) of this theorem and the fact that constant functions and the
function z are analytic on C imply that every polynomial in z is analytic on C. Of
course, since z is not analytic, we cannot expect mixed polynomials that contain
powers of both z and z to be analytic.
Parts (b) and (c) of the theorem imply that f /g is differentiable at z if f and
g are and if g(z) = 0. They also imply the quotient rule
 
f f  (z)g(z) − g  (z)f (z)
(z) = .
g g 2 (z)
The chain rule also holds for the complex derivative.
Theorem 2.2.7. If g is differentiable at a and f is differentiable at b = g(a), then
f ◦ g is differentiable at a and
(f ◦ g) (a) = f  (g(a))g  (a).

Proof. Let U be a neighborhood of b on which f is defined. We define a function

h(w) on U in the following way
⎧
⎨ f (w) − f (b)
, if w = b;
h(w) = w−b
⎩f  (b), if w = b.
Then h is continuous at b, since
f (w) − f (b)
f  (b) = lim .
w→b w−b
Also,
f ◦ g(z) − f ◦ g(a) g(z) − g(a)
(2.2.4) = h(g(z))
z−a z−a
−1
for all z in the deleted neighborhood V = g (U ) \ {a} of a. If we take the limit
of both sides of (2.2.4) and use the fact that f and h are continuous at b and g is
continuous at a, we conclude that (f ◦ g) (a) = f  (g(a))g  (a), as required. 
2.2. The Complex Derivative 37

Example 2.2.8. Suppose p(z) is a polynomial in z. Where is the function ep(z)

analytic and what is its derivative?
Solution: Since ez and p(z) both are differentiable everywhere, so is the com-
position ep(z) , by Theorem 2.2.7, and the derivative is
 
ep(z) = p (z) ep(z) .

The Cauchy-Riemann Equations. Since a function f of a complex variable

may be regarded as a complex-valued function on a subset of R2 , we can write it
in the form
(2.2.5) f (x + iy) = u(x, y) + iv(x, y),
where u and v are the real and imaginary parts of f , regarded as functions defined
on a subset of R2 . It is natural to ask what the existence of a complex derivative
for f implies about the functions u and v as functions of the two real variables x
and y. It is easy to see that it implies the existence of the partial derivatives ux ,
uy , vx and vy . In fact, it implies much more as the following discussion will show.
Recall that a function g of two real variables is said to be differentiable at (x, y)
if there are numbers A and B such that
g(x + h, y + k) − g(x, y) = Ah + Bk + (h, k),
where (h, k)/|(h, k)| → 0 as (h, k) → (0, 0). If g is differentiable at (x, y), then the
numbers A and B are the partial derivatives gx and gy at (x, y).
Suppose f is a complex-valued function defined in a neighborhood of z ∈ C. If
M = f  (z) exists, then we may write
(2.2.6) f (z + λ) − f (z) = M λ + (λ),
where (λ)/λ → 0 as λ → 0. In fact, (λ) is given by
(λ) = f (z + λ) − f (z) − M λ,
and so, the fact that (λ)/λ → 0 as λ → 0 is equivalent to the statement that f  (z)
exists and is equal to M .
If we write f, M, z, λ, and  in terms of their real and imaginary parts: f =
u + iv, M = C + iD, z = x + iy, λ = h + ik, and  = ρ + iω, then (2.2.6) becomes
(2.2.7) u(x + h, y + k) + iv(x + h, y + k) − u(x, y) − iv(x, y)
= (C + iD)(h + ik) + ρ(h, k) + iω(h, k).
On equating real and imaginary parts, this leads to the two equations
u(x + h, y + k) − u(x, y) = Ch − Dk + ρ(h, k),
(2.2.8)
v(x + h, y + k) − v(x, y) = Dh + Ck + ω(h, k).
The condition that (λ)/λ → 0 as λ →√0 implies that ρ(h, k)/|(h, k)| → 0 and
ω(h, k)/|(h, k)| → 0 (note that |(h, k)| = h2 + k2 = |λ|). Thus, we can draw two
conclusions from the existence of f  (z): (1) u and v are differentiable at (x, y), and
(2) the partial derivatives of u and v at (x, y) are given by
ux (x, y) = C, uy (x, y) = −D,
(2.2.9)
vx (x, y) = D, vy (x, y) = C.
38 2. Analytic Functions

A surprising consequence of this is that if f  exists at z = x + iy, then

ux = vy ,
(2.2.10)
uy = −vx

at (x, y). Equations (2.2.10) are the Cauchy-Riemann equations. Equations (2.2.9)
also show that if f  exists at z, then f  (z) = C + iD = ux + ivx = −i(uy + ivy ). If
we set fx = ux + ivx and fy = uy + ivy , then this can be written as f  = fx = −ify
wherever f  exists.
The above discussion shows that, at any point where f has a complex derivative,
its real and imaginary parts are differentiable functions and satisfy the Cauchy-
Riemann equations. The converse is also true: If the real and imaginary parts of f
are differentiable and satisfy the Cauchy-Riemann equations at a point z = x + iy,
then f  (z) exists. The proof of this is a matter of working backwards through the
above discussion, beginning with the assumption that u and v are differentiable at
(x, y), with partial derivatives that satisfy ux = vy = C and uy = −vx = −D. This
leads to (2.2.8), which eventually leads back to the conclusion that C + iD is the
derivative of f at z = x + iy. We leave the details to the exercises. The result is
the following theorem.

Theorem 2.2.9. If f = u + iv is a complex-valued function defined in a neigh-

borhood of z ∈ C, with real and imaginary parts u and v, then f has a complex
derivative at z if and only if u and v are differentiable and satisfy the Cauchy-
Riemann equations (2.2.10) at z = x + iy. In this case,

f  = fx = −ify .

Example 2.2.10. We already know that ez is analytic everywhere. However, give

a different proof of this by showing ez satisfies the Cauchy-Riemann equations.
Solution: With z = x + iy, we write ez = ex (cos y + i sin y). The real and
imaginary parts of ez are u(x, y) = ex cos y and v(x, y) = ex sin y. Thus,

ux (x, y) = ex cos y = vy , and

uy (x, y) = − ex sin y = −vx .

Example 2.2.11. Use the Cauchy-Riemann equations to prove that, for each
branch of the log function, log(z) is analytic everywhere except on its cut line
and has derivative 1/z.
Solution: We first prove that the principal branch of the log function is an-
alytic on the right half-plane H = {z ∈ C : Re(z) > 0}. For z ∈ H we have
z = x + iy = r eiθ where

r= x2 + y 2 and θ = tan−1 (y/x).

Thus, the principal branch of log on H is

log(x + iy) = (1/2) ln(x2 + y 2 ) + i tan−1 (y/x).

2.2. The Complex Derivative 39

Taking partial derivatives yields

∂ x
(1/2) ln(x2 + y 2 ) = ,
∂x x2 + y 2
∂ −y/x2 −y
tan−1 (y/x) = = 2 ,
∂x 1 + (y/x)2 x + y2
(2.2.11)
∂ y
(1/2) ln(x2 + y 2 ) = ,
∂y x2 + y 2
∂ 1/x x
tan−1 (y/x) = 2
= 2 .
∂y 1 + (y/x) x + y2
Thus, the Cauchy-Riemann equations are satisfied by the principal branch of the
log function on H. Furthermore
∂ x − iy 1
(log z) = log(x + iy) = 2 = .
∂x x + y2 z
Now if z is any point not on the negative real axis and not in H, then we simply
rotate z into H. That is, we choose α = ±π/2 such that eiα z ∈ H. Then
log z = log(eiα z) − iα.
Since log has derivative 1/w at w = eiα z, it follows from the chain rule that log has
derivative eiα /(eiα z) = 1/z at z. Thus, the principal branch of the log function is
analytic with derivative 1/z at any point z not on its cut line.
The analogous statement for other branches of the log function also follows
from a rotation argument, as above. That is, each such function is just the principal
branch of the log function composed with a rotation.

Harmonic Functions. In the next chapter, we will prove that analytic functions
are C∞ – that is, they have continuous complex derivatives of all orders. This, in
particular, implies that analytic functions have continuous partial derivatives of all
orders with respect to x and y. Assuming this result for the moment, we have
Theorem 2.2.12. The real and imaginary parts of an analytic function on U are
harmonic functions on U , meaning they satisfy Laplace’s equation
uxx + uyy = 0.

Proof. If f = u + iv is an analytic function, then u and v satisfy the Cauchy-

Riemann equations and so
uxx = (ux )x = (vy )x = (vx )y = (−uy )y = −uyy .
This shows that the real part of f satisfies Laplace’s equation. Since v is the real
part of the analytic function −if , it follows that v is also harmonic. Thus, both
real and imaginary parts of an analytic function are harmonic. 

If u and v are harmonic functions such that the function f = u + iv is analytic,

then we say u and v are harmonic conjugates of one another.
Example 2.2.13. Prove that u(x, y) = ex cos y is a harmonic function on all of R2
and find a harmonic conjugate for it.
40 2. Analytic Functions

Solution: The function u is the real part of f (z) = ez and is, therefore,
harmonic by the previous theorem. The imaginary part of f is v(x, y) = ex sin y,
and so this function v is a harmonic conjugate of u.

Exercise Set 2.2

1. Fill in the details in Example 2.2.4 by verifying the inequality (2.2.3) and
showing that the limit of the expression on the right is 0.
2. Prove Theorem 2.2.5.
3. Prove Part (b) of Theorem 2.2.6.
4. Prove Part (c) of Theorem 2.2.6.
5. Use induction and Theorem 2.2.6 to show that (z n ) = nz n−1 if n is a non-
negative integer.
6. Find the derivative of z 7 + 5z 4 − 2z 3 + z 2 − 1. Which results from this section
are used in this calculation?
3
7. Find the derivative of ez .
8. If we use the principal branch of the log function, at which points of C does
log z
have a complex derivative? What is its derivative at these points?
z
9. Finish the proof of Theorem 2.2.9 by showing that if f = u + iv, u and v are
differentiable at z, and u and v satisfy the Cauchy-Riemann equations at z,
then f  (z) exists.
10. Use the Cauchy-Riemann equations to verify that the function f (z) = z 2 is
analytic everywhere.
11. Describe all real-valued functions which are analytic on C.
12. Derive the Cauchy-Riemann equations in polar coordinates:
ur = r −1 vθ ,
uθ = −rvr
by using the change of variable formulas x = r cos θ, y = r sin θ and the chain
rule.
13. We showed in Example 2.2.11 that each branch of the log function is analytic
on the complex plane with its cut line removed. Use the Cauchy-Riemann
equations in polar form (previous problem) to give another proof of this fact.
14. Assuming each branch of the log function is analytic, use the chain rule to give
another prove that each such function has derivative 1/z.
15. Use the Cauchy-Riemann equations to prove that if f is analytic on an open set
U , then the function g defined by g(z) = f (z) is analytic on the set {z : z ∈ U }.
16. Verify that the function log |z| is harmonic on C \ {0} and find a harmonic con-
jugate for it on the set consisting of C with the non-positive real axis removed.
2.3. Contour Integrals 41

2.3. Contour Integrals

Integration plays a key role in this subject – specifically, integration along curves
in C. A curve or contour in the plane C is a continuous function γ from an interval
on the line into C. Such an object is sometimes called a parameterized curve and
the interval I is called the parameter interval. We will be interested in a particular
kind of curve, one whose parameter interval is a closed bounded interval which can
be subdivided into finitely many subintervals, on each of which γ is continuously
differentiable.

Smooth Curves. Let I = [a, b] be a closed interval on the real line and let
γ : I → C be a complex-valued function on I. If c ∈ I, then the derivative γ  (c) of
γ at c is defined in the usual way:
γ(t) − γ(c)
(2.3.1) γ  (c) = lim .
t→c t−c
Of course, γ is complex-valued and so this limit should be interpreted as the type
of limit discussed in Section 2.1. It can be calculated by expressing γ in terms of
its real and imaginary parts, that is, by writing γ(t) = x(t) + iy(t), where x(t) and
y(t) are real-valued functions on I. Then γ  (t) = x (t) + iy  (t) (Exercise 2.3.6).
What about the endpoints a and b of the interval I? Should we either not talk
about the derivative at the endpoints or, perhaps, use one-sided derivatives defined
in terms of one-sided limits (limit from the right at a and limit from the left at b)?
Actually, there is no need to do anything special at a and b or to exclude them. If
the domain of γ is [a, b], then our domain dependent definition of limit takes care
of the problem. If c = a, the limit as t → a in (2.3.1) only involves values of t to
the right of a, since only those are in the domain of the difference quotient that
appears in this limit. Similarly, if c = b, the limit as t → b involves only points to
the left of b. Thus, the derivatives at a and b that our definition leads to are what
in calculus would be called the right derivative at a and the left derivative at b.
The curve γ is differentiable at c if the limit defining γ  (c) exists. It is contin-
uously differentiable or smooth on I if it is differentiable at every point of I and if
the derivative is a continuous function on I. In this case we will write γ ∈ C1 (I).
Definition 2.3.1. A curve γ : [a, b] → C in C is called piecewise smooth if there
is a partition a = a0 < a1 < · · · < an = b of [a, b] such that the restriction of γ to
[aj−1 , aj ] is smooth for each j = 1, · · · , n. A curve which is piecewise smooth will
be called a path.

With appropriate choices of parameterization, familiar geometric objects in C

can be described as the image of a path.
Example 2.3.2. Find a path γ that traces once around the circle of radius r,
centered at 0, in the counterclockwise direction. Describe γ  .
Solution: The smooth path γ(t) = r eit , t ∈ [0, 2π] does the job. Its derivative
may be obtained by writing it as r(cos t + i sin t) and differentiating the real and
imaginary parts. The result is γ  (t) = r(− sin t + i cos t) = ir eit .
42 2. Analytic Functions

γ(b)
.

.
γ(a)
.
γ(t)

Figure 2.3.1. A Path in the Plane.

Example 2.3.3. Let z and w be two points in C. Find a path which traces the
straight line from z to w and find its derivative.
Solution: The path γ, with parameter interval [0, 1], defined by
γ(t) = (1 − t)z + tw = z + t(w − z),
satisfies γ(0) = z and γ(1) = w. It is a parametric form of a straight line in the
plane, and its derivative is γ  (t) = w − z.
Example 2.3.4. Find a path that traces once around the square with vertices
0, 1, 1 + i, i in the counterclockwise direction. Find γ  (t) on the subintervals where
γ is smooth.
Solution: We choose [0, 1] as the parameter interval and define a path γ as
follows (see Figure 2.3.2):
⎧
⎪
⎪ 4t, if 0 ≤ t ≤ 1/4;
⎪
⎨1 + (4t − 1)i, if 1/4 ≤ t ≤ 1/2;
γ(t) =
⎪
⎪ 3 − 4t + i, if 1/2 ≤ t ≤ 3/4;
⎪
⎩
(4 − 4t)i, if 3/4 ≤ t ≤ 1.
This is continuous on [0, 1] and smooth on each subinterval in the partition 0 <
1/4 < 1/2 < 3/4 < 1. It traces each side of the square in succession, moving in
the counterclockwise direction. On the first interval, γ  is the constant 4, on the
second it is 4i, on the third it is −4, and on the fourth it is −4i.

Riemann Integral of Complex-Valued Functions. The integral of a function

along a path will be defined in terms of the Riemann integral on an interval. This
is the familiar Riemann integral from calculus, except that the functions being
integrated will be complex-valued. This difference requires a few comments.
If f (t) = g(t) + ih(t) is a complex-valued function on an interval [a, b], where
g and h are real-valued, then we will say that f is Riemann integrable on [a, b] if
both g and h are Riemann integrable on [a, b] as real-valued functions. We then
define the integral of f on [a, b] by
 b  b  b
(2.3.2) f (t) dt = g(t) dt + i h(t) dt.
a a a
2.3. Contour Integrals 43

i 1+i

0 1

Figure 2.3.2. The Path of Example 2.3.4.

This Riemann integral for complex-valued functions has the properties one
would expect given knowledge of the Riemann integral for real-valued functions.
The next three theorems cover some of these properties.
Theorem 2.3.5. Let f1 and f2 be Riemann integrable functions on [a, b] and α
and β complex numbers. Then, αf1 + βf2 is integrable on [a, b], and
 b  b  b
(αf1 (t) + βf2 (t)) dt = α f1 (t) dt + β f2 (t) dt.
a a a

Proof. That this is true if the constants α and β are real follows directly from
expressing f1 and f2 in terms of their real and imaginary parts. Thus, to prove
b b
the theorem we just need to show that a if (t) dt = i a f (t) dt if f = g + ih is an
integrable function on [a, b]. However,
 b  b
i(g(t) + ih(t)) dt = (−h(t) + ig(t)) dt
a
 
a
 
b b b
=− h(t) dt + i g(t) dt = i (g(t) + ih(t)) dt .
a a a

This completes the proof. 

Theorem 2.3.6. If f is a function defined on [a, b] and c ∈ (a, b), then f is
integrable on [a, b] if and only if it is integrable on [a, c] and [c, b]. In this case
 b  c  b
f (t) dt = f (t) dt + f (t) dt.
a a c

Proof. This follows from the fact that the same things are true of the integrals of
the real and imaginary parts g and h of f . 
Theorem 2.3.7. If f is an integrable function on [a, b], then
  
 b  b
 
 f (t) dt ≤ |f (t)| dt.
 a  a
44 2. Analytic Functions

b
Proof. This is proved using a trick. We set w = a f (t) dt. If w = 0, there is
nothing to prove. If w = 0, let u = w/|w|. Then uw = |w| and so
   b  b
 b 
 
 f (t) dt = u f (t) dt = uf (t) dt.
 a  a a

Since this is a real number, the integral of the imaginary part of uf is zero and we
have   
 b  b  b  b
 
 f (t) dt = Re(uf (t)) dt ≤ |uf (t)| dt = |f (t)| dt. 
 a  a a a

A complex-valued function which is defined and continuous on an interval [a, b]

is clearly Riemann integrable on [a, b], since its real and imaginary parts are continu-
ous, and continuous real-valued functions on closed, bounded intervals are Riemann
integrable.

Integration Along a Path. If γ is a path, then γ  exists and is continuous on

each interval [ai−1 , ai ] in a partition a = a0 < a1 < · · · < an = b of the parameter
interval [a, b]. At the points a1 , a2 , · · · , an−1 the definition of γ  is ambiguous –
γ  (aj ) has one value from the derivative of γ on [aj−1 , aj ] and another from the
derivative of γ on [aj , aj+1 ]. In order to remove this ambiguity, we choose to define
γ  so as to be left continuous at these points. That is, at aj , we choose the value for
γ  that comes from its definition on [aj−1 , aj ]. Then γ  is well defined on I = [a, b].
If f is a complex-valued function defined and continuous on a set E containing
γ(I), then the function f (γ(t))γ  (t) is a well-defined function on I which is piecewise
continuous in the following sense: It is continuous everywhere on [a, b] except at
the partition points a1 , a2 , · · · , an−1 . It is left continuous at these points, and the
limit from the right exists and is finite at these points as well. In other words, this
function is continuous from the left everywhere on [a, b] and continuous except at
finitely many points where it has simple jump discontinuities.
A function of this type is Riemann integrable on [a, b]. To see this, first observe
that it is Riemann integrable on each subinterval [aj−1 , aj ] because, on such an
interval, the function agrees with a continuous function except at one point, aj−1 .
A continuous function on a closed interval is Riemann integrable and changing its
value at one point does not effect this fact or the value of the integral. Furthermore,
by Theorem 2.3.6, if a function is Riemann integrable on two contiguous intervals,
then it is integrable on their union. It follows that a function which is integrable
on each subinterval in a partition of [a, b] will be integrable on [a, b].
The above discussion settles the question of the Riemann integrability of the
integrand in the following definition.
Definition 2.3.8. Let γ : [a, b] → C be a path and let f be a function which is
defined and continuous on a set E which contains γ([a, b]). Then we define the
integral of f over γ to be
  b
(2.3.3) f (z) dz = f (γ(t))γ  (t) dt.
γ a
2.3. Contour Integrals 45

One may think of this definition in the following way: the contour integral on
the left in (2.3.3) is defined to be the Riemann integral obtained by replacing z by
γ(t) and dz by γ  (t)dt and integrating over the parameter interval for γ.
In practice, we will calculate contour integrals by breaking the path up into its
smooth sections, calculating the integrals over these sections and then adding the
results. That this is legitimate follows from the fact that the Riemann integral of
a function over the union of two contiguous intervals on the line is the sum of the
integrals over the two intervals.

Examples.

Example 2.3.9. Find γ
z dz if γ is the circular path defined in Example 2.3.2.
Solution: By Example 2.3.2, we have γ(t) = r eit for 0 ≤ t ≤ 2π and γ  (t) =
it
ir e . Thus,
  2π   2π
z dz = r eit ir eit dt = ir 2 e2it dt = ir 2 (cos 2t + i sin 2t) dt
γ 0 0
 2π  2π
= ir 2 cos 2t dt − r 2 sin 2t dt = 0.
0 0

Example 2.3.10. Find a path γ which traces the straight

 line from 0 to i followed
by the straight line from i to i + 1. Then calculate γ z 2 dz for this path γ.
Solution: We may choose γ to be the path parameterized on [0, 2] as follows:

it, if 0 ≤ t ≤ 1;
γ(t) =
i + t − 1, if 1 ≤ t ≤ 2.
We calculate the integrals over each of the two smooth sections of the path. On
[0, 1] we have (γ(t))2 = −t2 and γ  (t) = i. Thus, the integral over the first section
of the path is
 1  1
1
(γ(t))2 γ  (t) dt = −t2 i dt = −t3 i/30 = −i/3.
0 0

On [1, 2] we have (γ(t))2 = t2 − 2t + 2(t − 1)i and γ  (t) = 1. Thus, the integral over
the second section of the path is
 2  1
2
2 
(γ(t)) γ (t) dt = (t2 − 2t + 2(t − 1)i) dt = (t3 /3 − t2 + (t2 − 2t)i)1 = −2/3 + i.
1 0

Thus, γ
z dz = −i/3 − 2/3 + i = −2/3 + 2i/3.
2

Example 2.3.11. Find a path γ which traces once around the triangle with vertices

0, 1, i in the counterclockwise direction, starting at 0. For this path γ, find γ z dz.
Solution: A path γ with the required properties has parameter interval [0, 3]
and is given by
⎧
⎪
⎨t, if 0 ≤ t ≤ 1;
γ(t) = 2 − t + (t − 1)i if 1 ≤ t ≤ 2;
⎪
⎩
(3 − t)i if 2 ≤ t ≤ 3.
46 2. Analytic Functions

On the interval [0, 1], we have γ(t) = t and γ  (t) = 1. Hence,

 1  1
γ(t)γ  (t) dt = t dt = 1/2.
0 0

On the interval [1, 2], we have γ(t) = 2 − t − (t − 1)i and γ  (t) = −1 + i. Hence,
 2  2
γ(t)γ  (t) dt = (2t − 3 + i) dt = i.
1 1

On the interval [2, 3], we have γ(t) = t − 3 and γ  (t) = −i. Hence,
 1  3

γ(t)γ (t) dt = (3 − t)i dt = i/2.
0 2

If we add the contributions of each of these three intervals, the result is


z dz = 1/2 + i + i/2 = 1/2 + (3/2)i.
γ

Exercise Set 2.3

π
1. Find 0
eit dt.
1
2. Find 0
sin(it) dt.
 2π int imt
3. Find 0
e e dt for all integers n and m.
4. Find a path which traces the straight line joining 2 − i to −1 + 3i.
5. If z0 ∈ C, find a path which traces the circle of radius r, centered at z0 , (a)
once in the counterclockwise direction, (b) once in the clockwise direction, (c)
three times in the counterclockwise direction.
6. Prove that if γ(t) = x(t) + iy(t) is a curve defined on an interval I, with real
and imaginary parts x(t) and y(t), and if c ∈ I, then γ  (c) exists if and only if
x (c) and y  (c) exist and, in this case, γ  (c) = x (c) + iy  (c).
7. Show that if f is a smooth complex-valued function on an interval [a, b], then
b 
a
f (t) dt = f (b) − f (a).
8. Suppose γ is a path with
 parameter interval [a, b]. Use the result of the previous
exercise to show that γ 1 dz = γ(b) − γ(a).

9. Find γ z 2 dz if γ traces a straight line from 0 to w.
 
10. Find γ z −1 dz and γ z dz for the circular path γ(t) = 3 eit , 0 ≤ t ≤ 2π.

11. Find γ Re(z) dz if γ is the path of Example 2.3.11.

12. With γ as in the previous exercise, find γ Im(z 2 ) dz.
 
13. Is it generally true that Re( γ f (z) dz) = γ Re(f (z)) dz?
2.4. Properties of Contour Integrals 47

2.4. Properties of Contour Integrals

We begin this section with the question of parameter independence. To what
extent does the integral of a function along a path depend on how the path is
parameterized? The same geometric figure γ(I) may be parameterized in many
ways. For example, the top third of the unit circle may be parameterized by
 √ √
γ1 (t) = −t + i 1 − t2 , − 3/2 ≤ t ≤ 3/2, or
(2.4.1)
γ2 (t) = eit = cos t + i sin t, π/6 ≤ t ≤ 5π/6,
and these are only two of infinitely many possibilities. Does the integral of a function
over the upper third of the unit circle depend on which of these parmeterizations
is chosen?

Parameter Changes that Change the Integral. The following example shows
that some changes of parameterization do change the integral.

Example 2.4.1. Find γ1 1/z dz if γ1 (t) = r eit on [0, 2π] is the circular path of
Example 2.3.2. Does the answer change if the circle is traversed in the clockwise
direction instead, using the path γ2 (t) = r e−it on [0, 2π]?
Solution: From Example 2.3.2 we know that the path γ1 (t) = r eit has γ1 (t) =
ir eit and so the given integral is
  2π  2π
dz (r eit )
= dt = i dt = 2πi.
γ1 z 0 r eit 0

On the other hand, the derivative of γ2 = e−it is −ir e−it and so

  2π
dz
= −i dt = −2πi.
γ2 z 0

This example shows that the integral along a path depends not only on the
geometric figure that is the image γ(I) of the path, but also on the direction the
path is traversed (at the very least).
Also, traversing a portion of the curve more than once may affect the integral.
For example, if we were to go around the circle twice in Example 2.4.1, by choosing
γ(t) = e2it on [0, 2π], the result would be 4πi instead of 2πi.

The Independence of Parameterization Theorem. There is a degree to which

the integral is independent of the parameterization. Certain ways of changing the
parameterization do not effect the integral, as the following theorem shows.
Theorem 2.4.2. Let γ1 : [a, b] → C be a path and α : [c, d] → [a, b] a smooth
function with α(c) = a and α(d) = b. If γ2 is the path with parameter interval [c, d]
defined by γ2 (t) = γ1 (α(t)), then
 
f (z)dz = f (z)dz
γ2 γ1

for every function f defined and continuous on a set E containing γ1 ([a, b]) =
γ2 ([c, d]).
48 2. Analytic Functions

Proof. We have γ2 (t) = γ1 (α(t)) and, by the chain rule,

γ2 (t) = γ1 (α(t))α (t).
Thus,
  d
f (z) dz = f (γ2 (t))γ2 (t) dt
γ2 c
 d
= f (γ1 (α(t)))γ1 (α(t))α (t) dt
c
 b
= f (γ1 (s))γ1 (s) ds

a

= f (z) dz,
γ1

where the third equality follows from the substitution s = α(t). This completes the
proof. 

Note that the condition that α(c) = a and α(d) = b is essential in the above
theorem. It says that α takes the endpoints of the parameter inverval [c, d] to the
endpoints of the parameter interval [a, b] in an order preserving fashion.
Example 2.4.3. Are the integrals of a continuous function over the two paths in
(2.4.1) necessarily the same?
Solution: Yes. If we set α(t) = − cos t, then α is a smooth√function√ mapping
the parameter interval [π/6, 5π/6] to the parameter interval [− 3/2, 3/2] in an
order preserving fashion. Furthermore, γ2 = γ1 ◦ α. Thus, the above theorem
insures that the integral of a continuous function over γ1 is the same as its integral
over γ2 .

Doesn’t Example 2.4.1 contradict Theorem 2.4.2? After all, if γ1 (t) = eit on
[0, 2π] and α : [0, 2π] → [0, 2π] is defined by α(t) = 2π − t, then γ2 (t) = γ1 (α(t)) =
e−it . By Example 2.4.1 the integrals of 1/z over these two curves are different.
Doesn’t Theorem 2.4.2 say they should be the same? No. The conditions α(a) = c
and α(b) = d are not satisfied by this choice of α, since α(0) = 2π and α(2π) = 0.
In other words, this choice of α reverses the order of the endpoints of the parameter
interval rather than preserving that order.
In general, the conditions α(a) = c and α(b) = d guarantee that, overall, γ2
traverses the curve in the same direction as γ1 . If α were positive on the entire
interval, then α would be increasing on this interval and γ1 and γ2 would be moving
in the same direction at each point of the curve. If α is not positive on all of [c, d],
then there may be intervals where one path reverses direction and backtracks, while
the other path does not. These things do not affect the integral, because if a curve
does backtrack for a time, it has to turn around and recover the same ground in
order to catch up to the other curve in the end. This is an intuitive explanation;
the actual proof that the integral is unaffected is in the proof of the above theorem.
Theorem 2.4.2 leads to a strategy which, for some paths γ1 and γ2 with the
same image, yields a proof that they determine the same integral: Suppose that the
parameter intervals for the two paths can each be partitioned into n subintervals
2.4. Properties of Contour Integrals 49

in such a way that for j = 1, · · · , n, γ1 on its jth subinterval and γ2 on its jth
subinterval are related by a smooth function
 αj , as in Theorem 2.4.2. If this can
be done, then it clearly follows that γ1 f (z) dz = γ2 f (z) dz for any function f
which is continuous on a set containing γ1 (I). For this reason, Theorem 2.4.2 is
sometimes called the independence of parameterization theorem.
Remark 2.4.4. Since path integrals are essentially independent of the way the
path is parameterized, we will often describe a path without specifying a parame-
terization. Instead, we will just give a description of the geometric object that is
traced, the direction, and how many times. For example, we may describe a path
as tracing once around the unit circle in the counterclockwise direction, or tracing
once around the boundary ∂Δ of a given triangle Δ in the counterclockwise direc-
tion, or as tracing the straight line path from a complex number w1 to a complex
number w2 . In the first two cases we may simply write
 
f (z) dz or f (z) dz
|z|=1 ∂Δ
for the corresponding path integral. In the latter case, we may write
 w2
f (z) dz
w1
for the path integral along the straight line from w1 to w2 .

Closed Curves. The curves in Examples 2.3.2 and 2.3.4 both have the property
that they begin and end at the same point – that is, they are closed curves. A closed
curve γ on a parameter interval [a, b] is one that satisfies γ(a) = γ(b). A closed
curve which is a path will be called a closed path.
The famous integral theorem of Cauchy states that the integral of an analytic
function f around a closed path is 0, provided there is an appropriate relationship
between the curve γ and the domain U on which f is analytic (roughly speaking,
the curve should lie in U but not go around any holes in U ). Since the function
f (z) = z is analytic on C (as is any polynomial in z), the next example illustrates
this phenomenon.

Example 2.4.5. Find γ z dz if γ is the path of Example 2.3.4.
Solution: From Example 2.3.4 we know that the path γ(t) has values 4t, 1 +
(4t−1)i, 3−4t+i, (4−4t)i and derivatives 4, 4i, −4, and −4i on the four subintervals
of the partition 0 < 1/4 < 1/2 < 3/4 < 1. Thus, the integrals over the four smooth
pieces of our curve are
 1/4
1/4
4t · 4 dt = 8t2 0 = 1/2,
0
 1/2 1/2
(1 + (4t − 1)i) · 4i dt = (4ti − 8t2 + 4t)1/4 = i − 1/2,
1/4
 3/4 3/4
(3 − 4t + i) · (−4) dt = (−12t + 8t2 − 4ti)1/2 = −1/2 − i,
1/2
 1 1
(4 − 4t)i · (−4i) dt = (+16t − 8t2 )3/4 = 1/2.
3/4
50 2. Analytic Functions

γ2 (c)
γ1
γ1 (a) γ2

γ1 (b) = γ2 (b)

a b c

Figure 2.4.1. The Join of Two Paths.


Since these add up to 0, we have γ
z dz = 0.

The function 1/z is also analytic, except at z = 0. The circular path of Example
2.4.1 is closed and lies in the domain where 1/z is analytic. So why is the integral
not 0? Because the path goes around a hole in the domain of 1/z – it goes around
{0}.

Additivity Properties of Contour Integrals. If γ is a path with parameter

interval [a, b], then we can use Theorem 2.4.2 to change the parameter interval to
any other interval [c, d] with c < d, in a way that does not affect the image of γ or
integrals over γ. In fact, if we set
b−a
α(t) = a + (t − c),
d−c
then α is smooth, α([c, d]) = [a, b], α(c) = a and α(d) = b. Thus, γ1 (t) = γ(α(t))
defines a path γ1 with the same image as γ and, by Theorem 2.4.2, a path which
determines the same integral for continuous functions on its image. Thus, without
loss of generality, we may always assume that the parameter interval for a path is
any interval we choose.
If γ1 and γ2 are two paths so that γ1 ends where γ2 begins, then we can join
the two paths to form a single new path γ1 + γ2 . We do this as follows: If γ1 has
parameter interval [a, b], we choose a parameter interval of the form [b, c] for γ2 .
The fact that γ2 begins where γ1 ends means that γ1 (b) = γ2 (b). We define γ1 + γ2
on [a, c] by

γ1 (t) if t ∈ [a, b],
(2.4.2) (γ1 + γ2 )(t) =
γ2 (t) if t ∈ [b, c].
The path γ1 + γ2 is called the join of γ1 and γ2 .
In Example 2.4.1, changing the path from one tracing the circle counterclock-
wise to one tracing the circle clockwise had the effect of changing the sign of the
integral. As we shall see, this always happens. If γ : [a, b] :→ C is a path, denote
by −γ the path defined by
−γ(t) = γ(a + b − t).
2.4. Properties of Contour Integrals 51

Then −γ(a) = γ(b) and −γ(b) = γ(a). In fact, −γ traces the same geometric figure
as γ, but it does so in the opposite direction.
For some closed curves, such as circles, and boundaries of rectangles, triangles,
etc., there is clearly a clockwise direction around the curve and a counterclockwise
direction. If such a curve is parameterized so that it is traversed in the counter-
clockwise direction, we will say the resulting closed path has positive orientation.
If it is traversed in the clockwise direction, we will say it has negative orientation.
Clearly, if γ has positive orientation, then −γ has negative orientation.
The common starting and ending point of a closed path can be changed without
changing the integral of a function over this path. This is done by representing the
closed path as the join of two paths which connect the original starting and ending
point to the new one. One then uses part (b) of the next theorem. The details are
left to the exercises.
The next theorem states the elementary properties of path integrals having
to do with linearity and path additivity. Part (b) follows immediately from the
corresponding additivity property of the Riemann integral on the line and we have
already used it several times. We leave the proofs of (a) and (c) to the exercises
(Exercise 2.4.5).
Theorem 2.4.6. Let γ, γ1 , γ2 be paths with γ1 ending where γ2 begins, f and g two
functions which are continuous on a set E containing the images of these paths,
and a and b complex numbers. Then
  
(a) (af (z) + bg(z)) dz = a f (z) dz + b g(z) dz;
γ γ γ
  
(b) f (z) dz = f (z) dz + f (z) dz;
γ1 +γ2 γ1 γ2
 
(c) f (z) dz = − f (z) dz.
−γ γ

Part (a) of this theorem says that a path integral is a linear function of the
integrand, Part (b) says that it is an additive function of the path, while Part (c)
shows why the notation −γ is appropriate for the curve that is γ traversed in the
opposite direction.

Length of a Path. We define the length (γ) of a path γ in C in the same way
the length of a curve in R2 is defined in calculus.
Definition 2.4.7. If γ(t) = x(t) + iy(t) is a path in C with parameter interval
[a, b], then the length (γ) of γ is defined to be
 b  b

(γ) = |γ (t)| dt = x (t)2 + y  (t)2 dt.
a a

Example 2.4.8. Prove that the above definition of length yields the correct length
for a path which traces once around a circle of radius r.
Solution: The path is γ(t) = r eit , with parameter interval [0, 2π]. The deriv-
 2π
ative of γ is γ  (t) = ir eit and so |γ  (t)| = r. Thus, (γ) = 0 r dt = 2πr.
52 2. Analytic Functions

It will be important in coming sections to be able to obtain good upper bounds

on the absolute value of a path integral. The key theorem that produces such upper
bounds is the following.
Theorem 2.4.9. Let γ be a path in C and f a function continuous on a set con-
taining γ(I). If |f (z)| ≤ M for all z ∈ γ(I), then
 
 
 f (z) dz  ≤ M (γ).
 
γ

Proof. If the parameter interval for γ is [a, b], then

   b  

   
b
 f (z) dz  =  
f (γ(t))γ (t) dt ≤ |f (γ(t))γ  (t)| dt
   
γ a a
 b  b
≤ M |γ  (t)| dt = M |γ  (t)| dt = M (γ). 
a a

The next example is a typical application of this theorem.

Example 2.4.10. Show that if f is a bounded continuous function on C, and γR
is the path γR (z) = R eit for t ∈ [0, 2π], then

f (z)
(2.4.3) lim dz = 0
R→∞ γ (z − w)2
R

for each w ∈ C.
Solution: The statement that f is bounded means there is an upper bound
M for |f |. That is, |f (z)| ≤ M for all z ∈ C. We also have |z − w| ≥ |z| − |w|
by the second form of the triangle inequality. If z ∈ γ(I), then |z| = R and so
|z − w| ≥ R − |w|, which implies |z − w|−2 ≤ (R − |w|)−2 . Thus, for z ∈ γ(I), we
have the following bound on the integrand of (2.4.3):
 
 f (z)  M
 
 z − w  ≤ (R − |w|)2 .
Since (γR ) = 2πR, Theorem 2.4.9 implies that
 
 f (z) 
 dz  ≤ 2πM R .
 (z − w) 2  (R − |w|)2
γR

The right side of this inequality has limit 0 as R → ∞ and this implies (2.4.3).

Exercise Set 2.4


1. Compute γ z 2 dz if γ is any path which traces once around the circle of radius
one in the counterclockwise direction.

2. Compute γ 1/z dz if γ is any path which traces twice around the circle of radius
one, centered at 0, in the counterclockwise direction.

3. If z0 and w0 are two points of C, compute γ z dz if γ is any path which traces
the straight line from z0 to w0 once.
2.5. Cauchy’s Integral Theorem for a Triangle 53

4. Compute the integral of the previous exercise for any smooth path γ which
begins at z0 and ends at w0 .
5. Prove Parts (a) and (c) of Theorem 2.4.6.
6. Describe a smooth, order preserving function α which takes the parameter
interval [0, 1] to the parameter interval [2, 5].
7. Prove that a parameter change γ → γ ◦ α, like the one in Theorem 2.4.2, does
not change the length of a path provided α is an non-decreasing function (has
a non-negative derivative).
 
 cos z 
8. Show that  dz  ≤ 2πe if γ is a path that traces the unit circle once.
γ z
Hint: Show that | cos z| ≤ e if |z| = 1.
9. Show that if Δ is a triangle in the plane of diameter d (length of its longest
side), and if f is a continuous function on Δ with |f | bounded by M on Δ,
then  
 
 f (z) dz  ≤ 3M d.
 
∂Δ

10. Prove that γ p(z) dz = 0 if γ(t) = eit , 0 ≤ t ≤ 2π, and p(z) is any polynomial
in z (this is a special case of Cauchy’s Theorem, but do not assume Cauchy’s
Theorem in your proof).
11. Let R(z) be the remainder after n terms in the power series for ez . That is,

n ∞

zk zk
R(z) = ez − = .
k! k!
k=1 k=n+1

e−1
Prove that |R(z)| ≤ if |z| ≤ 1.
(n + 1)!

12. Prove that γ ez dz = 0 if γ(t) = eit , 0 ≤ t ≤ 2π using the previous exercise
and Exercise 10.
13. Prove that if γ is a closed path with parameter interval I = [a, b] and common
starting and ending point z = γ(a) = γ(b) and w is any other point on γ(I),
then there is another closed path γ1 with γ1 (I) = γ(I), which determines the
same integral, but has w as common starting and ending point. Hint: Use part
(b) of Theorem 2.4.6.

2.5. Cauchy’s Integral Theorem for a Triangle

The core material of any beginning Complex Variables text is the proof of Cauchy’s
Integral Theorem and the exploration of its consequences. Roughly speaking,
Cauchy’s Integral Theorem states that the integral of an analytic function around
a closed path is zero, provided the path is contained in the open set U on which the
function is analytic and does not go around any “holes” in U . Part of the problem
here is to make sense of the idea of a “hole” in an open set and to decide what it
means for a path to go around such a hole.
54 2. Analytic Functions

We take a first step toward the proof of Cauchy’s Theorem in this section by
proving it in the case where the path is the boundary of a triangle and the function
is analytic on an open set containing the triangle.
The proof of this result will make essential use of a couple of properties of
compact sets. Thus, we will precede the proof with a discussion of compact sets.

Compact Sets. A subset K of Rn is called compact if every open cover of K

has a finite subcover. Here, by an open cover of K, we mean a collection of open
sets whose union contains K. A finite subcover is then a finite subcollection of this
collection which also has union containing K.
Here we will state without proof a number of facts about compact sets. The
proofs can be found in any text on Advanced Calculus or undergraduate Real
Analysis.

Theorem 2.5.1 (Heine-Borel Theorem). A subset of Rn is compact if and only if

it is closed and bounded.

Another characterization of compact sets is obtained by using complementation

to turn the statement about open covers in the definition into a statement about
collections of closed sets. The result is the following:
A set K is compact if and only if for every collection of closed subsets of K
with empty intersection, there is a finite subcollection with empty intersection. The
same thing stated somewhat differently is:

Theorem 2.5.2. A set K is compact if and only if, whenever a collection of closed
subsets of K has the property that each finite subcollection has non-empty intersec-
tion, then the full collection also has non-empty intersection.

A consequence of the previous theorem is the following:

Corollary 2.5.3. If
A1 ⊃ A2 ⊃ · · · ⊃ An ⊃ · · ·

is a nested sequence of non-empty compact subsets of Rn , then An = ∅.

This is one of the properties of compact sets that we shall need in our proof of
Cauchy’s Theorem on a triangle. The others are as follows:

Theorem 2.5.4. If f is a continuous function, defined and continuous on a com-

pact subset K of Rn , with values in Rm , then f (K) is also compact.

A non-empty compact subset of R contains both a maximal element and a

minimal element. If f is a real-valued function, defined and continuous on a compact
set K, then f (K) is a compact subset of the line and, hence, has maximal and
minimal elements. This proves the following corollary of the previous theorem.

Corollary 2.5.5. A continuous real-valued function on a compact set has a maxi-

mal value and a minimal value.
2.5. Cauchy’s Integral Theorem for a Triangle 55

Antiderivatives. There is one case in which it is very easy to prove that the
integral of a continuous function f around a closed path is 0. This is the case where
the function f has an antiderivative.
As with functions of a real variable, an antiderivative for a function f , defined
on an open set U , is a function g such that g  = f on U .
Theorem 2.5.6. If f is a continuous function defined on an open set U , and if f
has an antiderivative g on U , then

f (z) dz = g(γ(b)) − g(γ(a))
γ

if γ is any path in U with parameter interval [a, b]. If γ is a closed path, then this
integral is 0.

Proof. Since g  = f on U , we have

  b  b
f (z) dz = f (γ(t))γ  (t) dt = g  (γ(t))γ  (t) dt.
γ a a

There is a version of the chain rule which holds for the composition of an analytic
function with a path (Exercise 2.5.4). It tells us that
(g(γ(t))) = g  (γ(t))γ  (t).
Thus,
  b
f (z) dz = (g(γ(t))) dt = g(γ(b)) − g(γ(a)),
γ a
by the complex version of the Fundamental Theorem of Calculus (Exercise 2.3.7).
If the path is closed, then γ(b) = γ(a) and the integral is 0. 

Cauchy’s Theorem. The fact that a function f has a complex derivative at

w means that, near w, it can be approximated by a linear function of z (more
precisely, by a polynomial of degree one in z). Such a function has a complex
antiderivative, as does any polynomial, and so its integral around a closed curve is
zero. Thus, a function with a complex derivative at w can be approximated near w
by a function whose integral around any closed path is zero. This is the basis for
an argument that the integral of a function with a complex derivative at w, around
a small triangle containing w, is much smaller than one would predict (using, for
example, the estimate given in Exercise 2.3.9). This is made precise in the next
lemma, which is the basis for the proof of Cauchy’s Theorem for triangles.
Lemma 2.5.7. Let f be a function which is continuous on a neighborhood of w ∈ C
and which has a complex derivative at w. Then for every  > 0, there is a δ > 0
such that  
 
 f (z) dz  < d2

∂Δ
if Δ is any triangle containing w, of diameter d ≤ δ.

Proof. Since f is continuous on a neighborhood of w, we may choose an r > 0

such that f is continuous on the open disc Dr (w).
56 2. Analytic Functions

Since f has a complex derivative at w,

f (z) − f (w)
lim = f  (w)
z→w z−w
exists. Thus, we may choose a positive δ < r such that |z − w| < δ implies that
 
 f (z) − f (w)  
(2.5.1)  − f (w) < .

 z−w 3
If we multiply (2.5.1) by |z − w|, the result is

(2.5.2) |f (z) − f (w) − f  (w)(z − w)| <|z − w| for all z ∈ Dδ (w).
3
If Δ is a triangle of diameter d ≤ δ, containing w, we set

I= f (z)dz.
∂Δ
Then
 
I= (f (w) + f  (w)(z − w))dz + (f (z) − f (w) − f  (w)(z − w))dz.
∂Δ ∂Δ
However, 
(f (w) + f  (w)(z − w))dz = 0,
∂Δ
because the integrand has a complex antiderivative: f (w)z + f  (w)(z 2 /2 − wz)
(remember w is a constant – the variable is z), and so

(2.5.3) I= (f (z) − f (w) − f  (w)(z − w))dz.
∂Δ
Then (2.5.2) implies that
|I| < d2 ,
since the modulus of the integrand in (2.5.3) is bounded by d/3 and the length of
the path ∂Δ is no more than 3d. This completes the proof. 
Theorem 2.5.8. Let f be a function which is analytic in an open set U , and
suppose that Δ is a triangle contained in U . Let ∂Δ denote the boundary of Δ,
considered as a closed path with positive orientation. Then

f (z) dz = 0.
∂Δ

Proof. We set 
I= f (z) dz.
∂Δ
Our objective is to prove that I = 0. We will do this by showing that |I| <  for
every positive number . Thus, let  be an arbitrary positive number.
We subdivide the triangle Δ into four smaller triangles by joining the midpoints
of the sides of Δ. The resulting four triangles are all similar to Δ with sides exactly
half as long as the corresponding sides of Δ (see Figure 2.5.1).
We now apply Parts (b) and (c) of Theorem 2.4.6. The sum of the integrals
around each of these smaller triangles is a sum of integrals along their edges, with
each of the edges interior to the original triangle occurring twice – once going one
2.5. Cauchy’s Integral Theorem for a Triangle 57

Figure 2.5.1. Subdividing the Triangle Δ in Theorem 2.5.8.

direction and once going the opposite direction. Thus, the contributions of these
interior edges cancel, leaving only the contributions from the edges which lie along
the boundary of the original triangle. It follows that the sum of the integrals
of f around the boundaries of these four smaller triangles is I, and so one of
these integrals must have modulus at least |I|/4. Let Δ1 denote the corresponding
triangle. In other words, Δ1 is chosen from the four subtriangles so that

|I1 | ≥ |I|/4 where I1 = f (z) dz.
∂Δ1

Note also that, if h is the diameter of Δ (which is the length of its longest side),
then Δ1 has diameter h1 = h/2.
We now repeat the above construction with Δ replaced by Δ1 . That is, we
subdivide Δ1 into four similar triangles and choose one of them, call it Δ2 , with
the property that

|I2 | ≥ |I1 |/4 ≥ |I|/4 where I2 =
2
f (z) dz,
∂Δ2
2
and with diameter h2 = h1 /2 = h/2 .
Proceeding by induction, we may choose for each n a triangle Δn , of diameter
hn , so that
Δn ⊂ Δn−1 ,

(2.5.4) |In | ≥ |I|/4n
where In = f (z)dz
∂Δn

and
(2.5.5) hn = h/2n .

The collection {Δn } is a nested sequence of closed bounded non-empty sets in

the plane (see Figure
 2.5.2) and, hence, by Corollary 2.5.3, there is a point w in
the intersection n Δn .
We now apply the previous lemma with /h2 replacing the  of the lemma. We
conclude there is a δ > 0 such that the integral of f around any triangle containing
w, of diameter d ≤ δ, is less than d2 /h2 .
58 2. Analytic Functions

Figure 2.5.2. The Nested Sequence of Triangles in Theorem 2.5.8.

We may then choose n large enough that hn = h/2n < δ. Since hn is the
diameter of Δn and w ∈ Δn , we have
h2n
|In | < .
h2
Putting this together with (2.5.5), we conclude

|In | < n .
4
Combined with (2.5.4), this yields
|I| ≤ 4n |In | < .
Since  was an arbitrary positive number, we conclude that I = 0. This completes
the proof. 

We will need a slightly stronger version of this theorem in which we allow the
possibility that there is one point in Δ where f may not have a complex derivative,
but where f is continuous.
Theorem 2.5.9. Let f, U, and Δ be as in the previous theorem except that we
assume that f is continuous on U and analytic on U \ {c} for some exceptional
point c ∈ Δ. Then we still have

f (z)dz = 0.
∂Δ

Proof. If c is a vertex of Δ, then, given  > 0, we may subdivide Δ into smaller

triangles in such a way that the one containing c has circumference less than /M ,
where M is the maximum value of |f | on Δ (Figure 2.5.3(a)). The integral of f over
the boundary of this triangle will then be less than or equal to . The integrals of
f over the boundaries of other triangles in the subdivision are all 0 by the previous
theorem (none of them contains c). As before, the integral of f over ∂Δ is the sum
of the integrals over the boudaries of the triangles in the subdivision and, hence,
has modulus less than or equal to . Since  was arbitrary, we conclude that the
integral of f over ∂Δ is zero.
If c is in Δ but is not a vertex, then the triangle can be subdivided into triangles
which do contain c as a vertex (Figure 2.5.3(b)). The integral around the boundary
2.5. Cauchy’s Integral Theorem for a Triangle 59

(a) (b)

Figure 2.5.3. Dealing with an Exceptional Point c.

of each of these is zero, so the integral around ∂Δ is also 0. This completes the
proof. 

Exercise Set 2.5

1. Prove the Bolzano-Weierstrass Theorem: If K is a compact subset of Rn , then

every sequence in K has a subsequence which converges to an element of K.
2. Use Corollary 2.5.5 to show that if K is a compact subset of C and f is a
continuous complex-valued function on K, then the modulus |f (z)| of f takes
on a maximal value at some point of K.
3. Show that if K is a compact subset of C, then there is a point z0 ∈ K of
minimum modulus – that is, a point z0 ∈ K such that
|z0 | ≤ |z| for all z ∈ K.
4. Prove that if g is analytic on an open subset U of C and γ : [a, b] → U is a path
in U , then
(g(γ(t))) = g  (γ(t))γ  (t)
for t ∈ [a, b]. Hint: The proof is very similar to the proof of Theorem 2.2.7.

5. Calculate γ z n dz if n is a non-negative integer and γ is a path in the plane
joining the point z0 to the point w0 . Hint: Use Theorem 2.5.6.

6. Show that γ p(z) dz = 0 if γ is any closed path in the plane and p is any
polynomial.

7. Calculate γ 1/z dz if γ is any path in C joining −i to i which does not cross
the half-line (−∞, 0] on the real axis. Hint: Use the result of Example 2.2.11
and Theorem 2.5.6.
60 2. Analytic Functions

8. Using the same hint as in the previous exercise, show that


1
dz = 0
γ z
if γ is any closed path contained in the complement of the set of non-positive
real numbers. Compare this with Example 2.4.1.
√ √
9. If z is defined by z = e(log z)/2 for the branch of the log function √ defined
by the condition
 √ −π/2 ≤ arg(z) ≤ 3π/2, find an antiderivative for z and
then find γ z dz, where γ is any path from −1 to 1 which lies in the upper
half-plane.
10. Prove that if f is analytic in an open set containing a rectangle R, then the
path integral of f around the boundary of this rectangle is 0.
11. Let γ be the path which traces the straight line from 1 to 1+i, then the
 straight
line from 1 + i to i and then the straight line from i to 0. Calculate γ z n dz.
12. Let Δ be the triangle with vertices 1 − i, i, and − 1 − i and S be the square with
vertices 1 − i, 1 + i, −1 + i, and − 1 − i. If f is any function which is analytic on
C \ {0}, prove that  
f (z) dz = f (z) dz,
∂Δ ∂S
where ∂Δ and ∂S are traversed in the counterclockwise direction.
13. For any pair of points a, b in C, denote the integral of a function f along the
b
straight line segment joining a to b by a f (z) dz, as in Remark 2.4.4. Suppose
f is analytic in an open set containing the triangle with vertices a, b, c. Show
that   
c b c
f (z) dz − f (z) dz = f (z) dz.
a a b
14. Show that Theorem 2.5.9 can be strengthened to conclude that the integral of
f around any triangle in U is 0 if f is continuous on U and analytic on U \ I,
where I is an interval contained in U . Hint: First consider the case where one
side of the triangle lies along the interval I.
15. If f is analytic on an open set U , then the integral of f around the boundary of
any triangle in U is 0 (Theorem 2.5.8), as is its integral around the boundary
of any rectangle in U (Exercise 2.5.10). What other geometric figures have this
property? What is the most general theorem along these lines you can think
of?

2.6. Cauchy’s Theorem for a Convex Set

A convex set C in C is a set with the property that if a and b are points in C, then
the line segment joining a and b is also contained in C.

Existence of Antiderivatives. The strategy for proving Cauchy’s Theorem for

convex sets is to prove that every analytic function on a convex set has an antideriv-
ative and then apply Theorem 2.5.6. The first step in this program is accomplished
with the following theorem.
2.6. Cauchy’s Theorem for a Convex Set 61

U z0
z
a

Figure 2.6.1. The Triangle Δ of Theorem 2.6.1.

Theorem 2.6.1. Let U be a convex open set and suppose f is a function which is
continuous on U and has the property that its integral around the boundary of any
triangle in U is zero. If a ∈ U is fixed and F (z) is defined for all z ∈ U by
 z
F (z) = f (w) dw,
a

then F (z) = f (z) for all z ∈ U .

Proof. Let [a, z] denote the line segment joining a to z, considered as a path from
a to z. For z ∈ U , this line segment lies entirely in U and so we may define a
function F (z) by
 z
F (z) = f (w) dw,
a

where by this we mean the path integral of f along the path [a, z], as in Remark
2.4.4. We will show that F  (z) = f (z). To do this, we let z and z0 be points of U
and consider the triangle Δ with vertices a, z, z0 (see Figure 2.6.1). The fact that
U is convex implies that Δ ⊂ U .
Let ∂Δ denote the boundary of Δ, considered as a contour which goes from a
to z to z0 and then to a again. Since, by hypothesis, the integral of f around the
boundary of any triangle in U is 0, we have


0= f (w) dw
∂Δ
z  z0  a
= f (w) dw + f (w) dw + f (w) dw
a z z0
 z
= F (z) − F (z0 ) − f (w) dw.
z0

We conclude that
 z
F (z) − F (z0 ) = f (w) dw.
z0
62 2. Analytic Functions

If we add and subtract the number f (z0 ) in the integrand of this integral, we obtain
 z  z
F (z) − F (z0 ) = f (z0 ) dw + (f (w) − f (z0 )) dw
z0 z
0 z
= f (z0 )(z − z0 ) + (f (w) − f (z0 )) dw.
z0
If we divide by z − z0 and subtract the first term on the right from both sides, we
get
 z
F (z) − F (z0 ) 1
(2.6.1) − f (z0 ) = (f (w) − f (z0 )) dw.
z − z0 z − z0 z0
Thus, to finish the proof that F  (z0 ) = f (z0 ), we just need to show that the
expression on the right in (2.6.1) has limit 0 as z → z0 .
Given  > 0, we may choose a δ > 0 such that |f (w) − f (z0 )| <  when
|w − z0 | < δ. This follows from the fact that f is continuous on U . If |z − z0 | < δ,
then |w − z0 | < δ for every w on the line segment [z0 , z] and so |f (w) − f (z0 )| < 
for every w on this line segment. Then
 z 
 
 (f (w) − f (z0 )) dw < |z − z0 |
 
z0
and so   z 
 1 
 (f (w) − f (z )) dw <
 z − z0 0 
z0
whenever |z − z0 | < δ. This shows that
 z
1
lim (f (w) − f (z0 )) dw = 0,
z→z0 z − z0 z
0

which completes the proof. 

Cauchy’s Integral Theorem. We now have all the tools in place to prove
Cauchy’s Integral Theorem for convex sets. This is not the most general form
of the theorem – that will come later – but it is sufficiently general to allow us to
derive a wealth of surprising consequences.
Theorem 2.6.2. Let U be a convex open set and suppose f is a function which is
analytic on U , except possibly at one point, where it is at least continuous. Then

f (z) dz = 0
γ
for every closed path in U .

Proof. By the previous theorem, and Theorem 2.5.8, f has a complex antideriva-
tive F in U . Then Theorem 2.5.6 implies that the integral of f around any closed
path is zero. 

One of the obvious applications of Cauchy’s Theorem is in proving independence

of path results for path integrals.
Corollary
 2.6.3. If U is a convex set and f is analytic on U , and a, b ∈ U , then
γ
f (z) dz is the same for all paths γ in U which begin at a and end at b.
2.6. Cauchy’s Theorem for a Convex Set 63

Figure 2.6.2. The Picture for Example 2.6.4.

The proof is left as an exercise. Another kind of path independence is illustrated

by the following example.
Example 2.6.4. Without doing any calculating, prove that the integral of 1/z
around any positively oriented ellipse with 0 inside is 2πi.
Solution: From Example 2.4.1 we know that the integral of 1/z around a
positively oriented circle centered at 0 is 2πi. Choose such a circle, with radius
small enough that the circle lies inside the ellipse. Then join the circle to the ellipse
with four line segments, two of which lie along the x-axis, and two of which lie along
the y-axis. This creates four closed paths, each of which consists of a path along
a piece on the circle followed by a line segment, followed by a piece of the ellipse
followed by a line segment leading back to the original point. Each of these closed
paths is contained in a convex open set on which 1/z is analytic and so the integral
of 1/z around each of them is 0. However, the sum of these integrals is also the
difference between the integral of 1/z around the circle and its integral around the
ellipse, because the contributions of the line segments cancel. Thus the integral of
1/z around the circle is the same as that around the ellipse and we conclude that
the latter is 2πi.

Index of a Path around a Point. Having proved Cauchy’s Integral Theorem

for a convex set, we can now prove a companion result – Cauchy’s Integral Formula
on a convex set. There are several versions of this result. The one we will present
here allows the integral to take place over a very general path, but requires that we
know how many times the path goes around a given point. The tool that measures
this is described in the following definition.
Definition 2.6.5. Let γ : I → C be any closed path in C and let z be a point of
C which does not lie on γ(I). We set

1 dw
Indγ (z) = .
2πi γ w − z
This is called the index of z with respect to γ. It is also sometimes called the
winding number of γ around z.
Theorem 2.6.6. If γ is a closed path in C with parameter interval I = [a, b], then
Indγ (z) is an integer-valued function of z defined on the complement of γ(I).
64 2. Analytic Functions

Proof. Let z0 a point in the complement of γ(I). Since γ is a closed path, we have
γ(a) = γ(b). We define a complex-valued function λ(t) on the interval a ≤ t ≤ b by
 t
γ  (s)
λ(t) = ds.
a γ(s) − z0

Then λ(a) = 0 and λ(b) = 2πi Indγ (z). If we can show that eλ(b) = 1, then the
proof will be complete, since the only numbers w with ew = 1 are the numbers
2πi n where n is an integer.
By the Fundamental Theorem of Calculus,
γ  (t)
λ (t) = ,
γ(t) − z0
while the derivative of eλ(t) is
γ  (t)
(eλ(t) ) = eλ(t) λ (t) = eλ(t) .
γ(t) − z0
It follows that
   
eλ(t) 1 γ  (t) λ(t) 
= e λ(t)
(γ(t) − z0 ) − e γ (t) = 0.
γ(t) − z0 (γ(t) − z0 )2 γ(t) − z0
Hence, eλ(t) /(γ(t) − z0 ) is a constant. We conclude that
eλ(t) eλ(a) 1
= =
γ(t) − z0 γ(a) − z0 γ(a) − z0
for every t ∈ [a, b]. If we set t = b, this gives us
γ(b) − z0
eλ(b) = .
γ(a) − z0
Since γ(a) = γ(b), it follows that eλ(b) = 1. This completes the proof. 

Cauchy’s Integral Formula for Convex Sets.

Theorem 2.6.7. Let U be a convex open set, f a function which is analytic on U
and γ : I → U a closed path in U . Then

1 f (w)
Indγ (z)f (z) = dw,
2πi γ w − z
for every point z ∈ U which does not lie on γ(I).

Proof. Consider the function g(z, w) defined for z, w ∈ U by

⎧
⎨ f (w) − f (z)
if w = z,
g(z, w) = w−z
⎩f  (z), otherwise.

For each fixed z ∈ U , this function is analytic in w everywhere on U except

possibly at w = z, but it is at least continuous at w = z. Since Theorem 2.6.2 holds
2.6. Cauchy’s Theorem for a Convex Set 65

even if the function is not analytic at some point but is continuous there, it follows
that
  
f (w) f (z)
0= g(z, w)dw = dw − dw
γ γ w − z γ w −z

f (w)
= dw − 2πi Indγ (z)f (z),
γ w −z
as long as z is not on the contour γ (note that this is required in order to write
the integral in the first line above as the difference of two integrals, since otherwise
these two integrals might not exist individually). We conclude that

1 f (w)
Indγ (z)f (z) = dw,
2πi γ w − z
as required. This completes the proof. 

This is a striking result, for it says that the values of an analytic function at
points “inside” a closed path are determined by its values at points on the path.
Here, a point is considered inside the path if the path has non-zero index at the
point.
Corollary 2.6.8. If U is a convex open set, z ∈ U and γ is a closed path in U
with Indγ (z) = 1, then

1 f (w)
f (z) = dw
2πi γ w − z
for every function f analytic on U .

Intuitively, the meaning of the hypothesis Indγ (z) = 1 in the above corollary is
that the closed path γ goes around z once and does so in the positive direction.
Cauchy’s Integral Theorem and Cauchy’s Integral Formula have a wealth of
applications. We will begin exploring these in the next chapter.
However, in order for Cauchy’s Integral Theorem, in the above form, to be us-
able, we need to be able to easily compute the index of a curve around a given point.
The last section of this chapter is devoted to developing the essential properties of
the index function which make this possible.

Exercise Set 2.6

1. Prove that a function which has complex derivative identically 0 on a convex

open set U is constant on U .

2. Calculate γ (z 2 − 4)−1 dz if γ is the unit circle traversed once in the positive
direction.

3. Calculate γ (1 − ez )−1 dz if γ is the circle γ(t) = 2i + eit .

4. Calculate γ 1/z dz if γ is any circle which does not pass through 0. Note that
the answer depends on γ.

5. Find γ 1/z 2 dz if γ is any closed path in C \ {0}.
66 2. Analytic Functions

6. Show that the principal

z branch of the log function can be described by the
formula log(z) = 1 1/w dw for z ∈
/ (−∞, 0].
7. Prove Corollary 2.6.3.
8. Without doing any calculating, show that the integral of 1/z around the bound-
ary of the triangle with vertices i, 1 − i, −1 − i is 2πi.
9. Let f be a function which is analytic on C\{z0 }. Show that the contour integral
of f around a circle of radius r > 0, centered at z0 , is independent of r.
10. Calculate Indγ (z0 ) if γ(t) = z0 + eint , t ∈ [0, 2π] and n is any integer.
11. Calculate Indγ (1 + i) if γ is the path which traces the line from 0 to 2, then
proceeds counterclockwis around the circle |z| = 2 from 2 to 2i and then traces
the line from 2i to 0. What is the answer if this path is traversed in the opposite
direction? 
ez
12. Use Cauchy’s Integral Formula to calculate dz.
|z|=1 z
13. Use Cauchy’s Formula to show that
 
1 1
dz = πi, dz = −πi.
|z−1|=1 z − 1 |z+1|=1 z − 1
2 2

14. Show that 

1
dz = 0.
|z|=3 z2 −1
Hint: Use the result of the preceding exercise.
15. Use Cauchy’s Integral Formula to prove that if f is a function which is analytic
in an open set containing the closed unit disc D1 (0), and if T = {z : |z| = 1} is
the unit circle, then |f (0)| ≤ M , where M is the maximum value of |f | on T .
16. Show that if γ is a path from z1 to z2 which does not pass through the point
z0 , then
  
1 z2 − z0
dw = log ,
γ w − z0 z1 − z0
for some branch of the log function. Note that, in the case where z1 = z2 , this
is just Theorem 2.6.6.

2.7. Properties of the Index Function

If γ is a closed path, then removing γ(I) from the plane results in a set which is
divided into a number of connected pieces. These are open sets called the connected
components of the complement of γ(I). We will prove that Indγ (z) is constant on
each of these components. Thus, to calculate Indγ (z) on a given component, one
only needs to calculate it at one point of the component.
Before proving this, we need to have a firm idea of what a connected component
is. This leads to a discussion of connected sets.
2.7. Properties of the Index Function 67

Connected Sets.

Definition 2.7.1. A set E ⊂ C is separated if there exists a pair A, B of open

subsets of C such that E ⊂ A ∪ B, A ∩ E = ∅, B ∩ E = ∅ and A ∩ B = ∅. The
pair A, B is then said to separate E. If E is not separated, then it is said to be
connected.

The union of a family of connected sets with a point in common is also connected
(Exercise 2.7.1). It follows that, if z ∈ E, then the union of all connected subsets
of E containing z is itself connected. This implies that each point of E is contained
in a maximal connected subset of E. A maximal connected subset of E is called
a connected component of E or simply a component of E. Two components of E
are either disjoint or identical, since, otherwise, their union would be a connected
set larger than one of them. Thus, the components of E form a pairwise disjoint
family of subsets of E whose union is E.
In this section we are primarily concerned with open sets and their components.
If E is open, then the sets A ∩ E and B ∩ E of Definition 2.7.1 are open subsets
of E. It follows that an open set E is separated if and only if it is the union of
two disjoint non-empty open subsets of itself. It is connected if this is not the case.
The next theorem states the essential facts regarding connected open sets that we
will need in this section.
An open set U is said to be path connected if every two points in U can be
connected by a path which lies entirely in U .

Theorem 2.7.2. Let U be an open subset of C. Then

(a) each component of U is also open;

(b) U is connected if and only if it is path connected.

Proof. We prove (b). The proof of (a) is left as an exercise.

Suppose U is connected. Given z ∈ U , let Vz be the set of points of U that are
connected to z by a path in U , and let w be some other point of U . There is an
open disc D, centered at w, which is contained in U . Since any two points in D are
connected by a line segment, either all points of D are in Vz or all points of D are
in U \ Vz . Hence, Vz and U \ Vz are open subsets of U whose union is U . Since U
is connected, one of them must be empty. Since it contains z, Vz is not empty, and
so U \ Vz must be empty. This means Vz = U and every point of U is connected to
z by a path in U . Hence, U is path connected.
Conversely, suppose U is path connected. If U = A ∪ B, where A and B are
disjoint non-empty open sets, then the function f which is 1 on A and 0 on B is a
continuous function on U , since the inverse image of any open subset of R under f
is A, B, U , or ∅, and these are all open. Now since U is path connected, there is a
path γ connecting a point of A to a point of B. Then f ◦ γ is a continuous function
on an interval I which takes on the values 1 and 0 and only these values. This is
impossible, by the Intermediate Value Theorem. The resulting contradiction shows
that there is no pair A, B as above, and so U is connected. 
68 2. Analytic Functions

By the above theorem, if K is a compact subset of C (such as the image γ(I)

of a closed path γ), then C \ K is the union of its connected components, each of
which is an open, path connected set.
Theorem 2.7.3. If K is a compact subset of C, then C \ K has exactly one un-
bounded component.

Proof. If K is a compact set, then K is closed and bounded. Since it is bounded,

it is contained in some closed disc Dr (0). Then its complement C \ K is open and
contains the complement of Dr (0). Since the complement of Dr (0) is connected, it
is contained in one of the components of C\K. This means all the other components
of C \ K are contained in Dr (0) and, hence, are bounded. 
Example 2.7.4. What are the components of C \ T , where T is the unit circle?
Solution: Clearly the sets {z ∈ C : |z| < 1} and {z ∈ C : |z| > 1} are path
connected and, hence, connected. These two connected open sets have C \ T as
union and so they must be the components of C \ T . One of them is bounded and
the other is unbounded.

Index is Constant on Components. We can now establish the result alluded

to at the beginning of this section.
Theorem 2.7.5. If γ : I → C is a closed path, then Indγ (z) is constant on each
component of C \ γ(I), and is zero in the unbounded component of C \ γ(I).

Proof. The set γ(I) is the image of a compact set under a continuous function and
so it is compact, hence, closed. Its complement C \ γ(I) is, therefore, open. Thus,
if z0 ∈ C \ γ(I), there is an open disc, centered at z0 , and contained in C \ γ(I). Let
R be the radius of one such disc. We will show that, on some smaller disc, centered
at z0 , Indγ (z) is constant.
Suppose r is a positive number less than R and z ∈ Dr (z0 ). Then
 
1 dw 1 dw
Indγ (z) − Indγ (z0 ) = −
2πi γ w − z 2πi γ w − z0
(2.7.1) 
1 z − z0
= dw.
2πi γ (w − z)(w − z0 )
Furthermore, every point w of γ(I) is at least a distance R away from z0 and a
distance R − r away from z. That is,
|w − z0 | ≥ R and |w − z| ≥ R − r.
Since |z − z0 | < r, this implies that the integrand of the last integral in (2.7.1) is
less than or equal to r/R(R − r) and, hence, that
r(γ)
| Indγ (z) − Indγ (z0 )| ≤ .
2πR(R − r)
We can make the right side of this inequality as small as we want by choosing r
sufficiently small. In particular, we can make it less than 1. However, Indγ (z) and
Indγ (z0 ) are both integers. If they differ by less than 1, then they are the same.
Thus, if r is chosen small enough, Indγ (z) is constant on Dr (z0 ).
2.7. Properties of the Index Function 69

Let A be a component of C \ γ(I), and for each integer n let Vn be the set of
points of A on which Indγ (z) = n. The above argument shows that each Vn is an
open subset of A. So is the union of all the sets Vm for which m = n. These two
sets separate A unless one of them is empty. Since A is connected, one of them
must be empty. This means that if Vn is not empty, then it is all of A. Thus, only
one of the sets Vn can be non-empty and this means that Indγ (z) is constant on A.
It remains to show that Indγ (z) = 0 on the unbounded component of C \ γ(I).
Let D be an open disc containing γ(I) and let z0 be a point outside this disc. Then
z0 is in the unbounded component of C \ γ(I). Furthermore,

1 dw
Indγ (z0 ) = =0
2πi γ w − z0
1
by Cauchy’s Integral Theorem, since D is a convex set containing γ(I) and
w − z0
is analytic on D. Since Indγ (z) is constant on each component of C \ γ(I), it must
be identically 0 on the unbounded component. 
Example 2.7.6. Calculate Indγ (z) for the path γ which traces n times around a
circle of radius r centered at z0 , where, if n is positive, this means γ traces the
circle in the counterclockwise direction, and, if n is negative, it means γ traces the
circle in the clockwise direction.
Solution: A parameterization for such a γ is γ(t) = z0 + r eint with parameter
interval I = [0, 2π]. Here γ  (t) = irn eint and so
 2π   2π
1 γ (t)dt 1
Indγ (z0 ) = = in dt = n.
2πi 0 γ(t) − z0 2πi 0
We use the previous theorem to find the index at other points z. Since the interior
of the circle traced by γ is a component of C \ γ(I), Indγ (z) must be constant on
it. Therefore, it has the value n at every z with |z − z0 | < r. The other component
of C \ γ(I) is the unbounded component {z : |z − z0 | > r}. On it, Indγ (z) = 0.
Thus, in this example, Indγ (z) is an integer n which is the number of times the
path goes around z if the path has positive orientation (goes in the counterclockwise
direction), and is the negative of this number if the path has negative orientation.
This includes the case where the path does not go around z at all, because z lies
outside the circle. Then n = 0.

Crossing a Path. For most paths, the index function can easily be computed
using the principle that if a path is crossed from right to left at a “simple” point of
the path, then the index increases by 1. We will make this statement precise below
and outline its proof. Most of the details are left to the exercises.
Definition 2.7.7. Let γ be a path with parameter interval I = [a, b] and let D be
an open disc in the plane. We will say that γ simply splits D if
(a) J = γ −1 (D) is a non-empty open subinterval of I, or (in the case where the
path is closed and γ(a) = γ(b) ∈ D) a union of two half-open subintervals
[a, c) and (d, b]; and
(b) D \ γ(J) has exactly two components.
70 2. Analytic Functions

z0 D

Figure 2.7.1. A Disc Simply Split by a Path.

Roughly speaking, a disc D is simply split by a path if the path passes just
once through D and cuts it into exactly two connected components.
In this situation, there is a way to make sense of which of the two components
is to the left of the path and which is to the right (Exercise 2.7.3).
Theorem 2.7.8. Let γ be a closed path which simply splits a disc D. Then
Indγ (z) = 1 + Indγ (w)
if z is in the left and w in the right component of D \ γ(J).
Thus, if z0 is a point of γ which is the center of a disc which is simply split by
γ, then Indγ (z) increases by 1 as z crosses γ from right to left at z0 .

The proof is left to the exercises.

The next example illustrates how to use this to compute Indγ (z) in specific
situations.
Example 2.7.9. Suppose γ is a path which traces the circle |z| = 2 once in the
counterclockwise direction beginning at 2 and then traces the circle |z − 1| = 1 once
in the counterclockwise direction. Find Indγ (z) for each z that is not in γ(I).
Solution: The components of C \ {γ(I)} are
A = {z ∈ C : |z| > 2};
B = {z ∈ C : |z| < 2, |z − 1| > 1};
C = {z ∈ C : |z − 1| < 1}.
Since A is the unbounded component, Indγ (z) = 0 if z ∈ A. Crossing γ from right
to left at 2i takes us from points of A to points of B. Hence, by the preceding
theorem, Indγ (z) = 1 for z ∈ B. Crossing γ from right to left at 1 + i takes us from
points of B to points of C and so, again by the preceding theorem, Indγ (z) = 2 if
z ∈ C.

If γ is a path with parameter interval I = [a, b], and t0 ∈ (a, b), then the
derivatives of γ as a function on [a, t0 ] and γ as a function on [t0 , b] both exist at
t0 (see Definition 2.3.1 and the discussion preceding it). They are equal if t0 is a
smooth point of the path. If t0 is not a smooth point, then γ has two derivatives
at t0 – a left derivative D γ and a right derivative Dr γ. At the endpoint a of I, γ
has only a right derivative, while at the endpoint b it has only a left derivative.
2.7. Properties of the Index Function 71

2i
A
1+i
B
C

Figure 2.7.2. The Path for Example 2.7.9.

Definition 2.7.10. If γ is a path with parameter interval I = [a, b], a point z0

on γ(I) is said to be a simple point if z0 = γ(t0 ) for exactly one parameter value
t0 ∈ (a, b) or for the two values a and b (in this case, γ is closed), and if the left
and right derivatives of γ at t0 (at a and b if z0 = γ(z) = γ(b)) are both non-zero.

In other words, a point is a simple point of a path if the path passes through
the point just once and it both approaches the point from a definite direction with
a positive speed and leaves the point in a definite direction with a positive speed.
This leads to a very useful criterion for a point on a path to be the center of a disc
that is simply split by the path.
Theorem 2.7.11. If z0 is a simple point of the path γ, then there is an open disc,
centered at z0 , which is simply split by γ.

Exercises 2.7.11 through 2.7.14 are devoted to proving this theorem in the case
where the point z0 is actually a smooth simple point of γ. In Exercise 2.7.15, the
reader is asked to modify this argument so as to prove the theorem in general.
Note that in attempting to prove the above theorem, we may assume that
z0 = γ(t0 ) for some interior point t0 of the parameter interval (otherwise we can
just reparameterize to make this the case). Also, we may assume that z0 = 0 since,
otherwise, we can just translate the curve to make this the case. Both of these
assumptions are made in Exercises 2.7.11 to 2.7.15.
Theorems 2.7.11 and 2.7.8 imply that if a closed path γ is crossed from right
to left at a simple point, then Indγ increases by 1. For most of the paths γ that
we shall encounter, all but finitely many points of γ are simple points. Exceptions
to this rule are paths which retrace parts of themselves, cross themselves infinitely
often, or come to a dead stop over some segment of the parameter interval.

Exercise Set 2.7

1. Prove that the union of a family of connected sets with a point in common is
also a connected set.
72 2. Analytic Functions

Figure 2.7.3. The Path for Exercise 2.7.9.

2. Prove Part (a) of Theorem 2.7.2. That is, prove that each component of an
open set is open.
3. Let D be an open disc and γ : I → C be a closed path which simply splits D.
Argue that, if we think of the positive direction along the curve through D as
being “up”, then it makes sense to think of one of the components into which
γ splits D as the “left” one and the other as the “right” one. Describe how to
tell which is which.
4. Suppose a closed path γ simply splits a disc D as in Figure 2.7.1. Define two new
paths γ1 and γ2 as follows: The curve γ1 agrees with γ until γ enters D. It then
departs from γ and instead traces the boundary of D in the counterclockwise
direction until it rejoins γ. It agrees with γ from that point on. The path
γ2 does the same thing except it traces the boundary of D in the clockwise
direction. For which points z inside D does Indγ (z) = Indγ1 (z)? For which
points z inside D does Indγ (z) = Indγ2 (z)? What is Indγ1 (z) − Indγ2 (z) if z is
any point inside D? Hint: Use Cauchy’s Integral Theorem and Example 2.7.6.
5. Use the results of the previous exercise to prove Theorem 2.7.8.
6. Prove that if γ is a closed path whose complement has just two components
and if γ has at least one simple point, then Indγ (z) = ±1 on the bounded
component.
7. In Example 2.7.9 how would the answers differ if the inner circle is traced in
the clockwise direction rather than the counterclockwise direction?
8. If a path γ traces a figure eight once, what are the possibilities for Indγ (z) in
the two bounded components of the complement of the figure eight?
9. Determine the value of Indγ (z) in each of the components of C \ γ(I) if γ is the
curve of Figure 2.7.3.
10. Suppose γ1 and γ2 are closed paths and z is not on either path. Show that
Indγ1 +γ2 (z) = Indγ1 (z) + Indγ2 (z) and Ind−γ1 = − Indγ1 (z). Hint: Use Theo-
rem 2.4.6.
11. Let γ(t) = x(t) + iy(t) be a path with parameter interval I = [a, b] and t0 a
point in (a, b) at which the path is smooth and simple. For further simplicity,
assume γ(t0 ) = 0 (we can always achieve this by translating the path). Prove
that, even though γ  (t0 ) may not exist, the second derivative of the function
h(t) = |γ(t)|2 = x2 (t) + y 2 (t)
does exist at t0 and equals 2|γ  (t0 )|2 .
2.7. Properties of the Index Function 73

12. Let γ and t0 be as in the previous exercise. Prove that there is an inter-
val (c, d) ⊂ (a, b), containing t0 , such that |γ(t)| is strictly decreasing on
[c, t0 ] and strictly increasing on [t0 , d]. Show that this implies that, if δ =
min{|γ(c)|, |γ(d)|}, then γ(t) crosses each circle of radius less than δ, centered
at 0, exactly once for t ∈ (c, t0 ) and exactly once for t ∈ (t0 , d).
13. Let γ, t0 and (c, d) be as in the previous exercise. Prove that there is an open
disc D, centered at 0, such that γ −1 (D) is an open subinterval J of (c, d). Hint:
Begin by showing you can choose D small enough that it contains no points
γ(t) for t ∈/ (c, d); then use the result of the previous exercise.
14. With γ, t0 , (c, d), and D as in the previous exercise, prove that D is simply
split by γ. This proves Theorem 2.7.11 in the case of a smooth simple point.
15. How would the argument outlined in the previous four exercises need to be
modified to prove Theorem 2.7.11 for a point which is a simple point of γ, but
not a smooth point? Note that, in this case, γ(t) will have two derivatives at
t0 – one from the left and one from the right.
 Chapter 3

Power Series Expansions

In this section we present several striking applications of Cauchy’s theorems. The

first of these is the existence of local power series expansions for analytic functions.
This leads to a number of other results, including the Fundamental Theorem of
Algebra and detailed information about the zeroes and singularities of analytic
functions.
Before we show that analytic functions have power series expansions we need
to develop a deeper understanding of convergence issues for power series. The first
section of the chapter is devoted to this task.

3.1. Uniform Convergence

We would like to be able to integrate and differentiate power series term by term.
This is shown to be legitimate in the case of real power series in the typical advanced
calculus or foundations of analysis course. The key to doing this is to show that
power series converge uniformly on certain sets. We will give a brief development
of these ideas in the context of complex-valued functions of a complex variable.
Definition 3.1.1. Let {fn } be a sequence of functions defined on a set S ⊂ C.
Then
(a) the sequence {fn } converges pointwise to the function f on S if, for each
z ∈ S, the sequence of numbers {fn (z)} converges to the number f (z) – that
is, for each z ∈ S and each  > 0, there is an N such that
|fn (z) − f (z)| <  for all n ≥ N;
(b) the sequence {fn } converges uniformly on S if for each  > 0 there exists an
N such that
|fn (z) − f (z)| <  for all n≥N and all z ∈ S.

There is a subtle but crucial difference between statements (a) and (b) in the
above definition: In (b), given , there must be an N that works for all z ∈ S. In

75
76 3. Power Series Expansions

(a), for each z there must be an N , but N depends on z, in general, and there may
not be an N that works simultaneously for all z ∈ S.
The importance of uniform convergence stems primarily from two facts that
are proved below and are extensively used thereafter: (1) The limit of a uniformly
convergent sequence of continuous functions is also continuous; and (2) the integral
along a path of the limit of a uniformly convergent sequence of continuous functions
is the limit of their integrals.
Theorem 3.1.2. If E is a subset of C and {fn } is a sequence of continuous func-
tions on E which converges uniformly on E to a function f , then f is also contin-
uous on E.

Proof. Let z0 be a point of E. Given  > 0, we choose N such that


n ≥ N implies |f (z) − fn (z)| < for all z ∈ E.
3
We can do this because {fn } converges uniformly to f on E.
We next choose a δ > 0 such that

|fN (z) − fN (z0 )| < whenever z ∈ E and |z − z0 | < δ.
3
We can do this because fN is continuous on E.
Then |z − z0 | < δ and z ∈ E imply
|f (z) − f (z0 )| ≤ |f (z) − fN (z)| + |fN (z) − fN (z0 )| + |fN (z0 ) − f (z0 )|
  
< + + = .
3 3 3
We conclude that limz→z0 f (z) = f (z0 ), and so f is continuous at z0 . Since z0 was
a general point of E, f is continuous on E. 
Theorem 3.1.3. If γ : I → C is a path and {fn } is a sequence of continuous
functions on γ(I) which converges uniformly on γ(I) to f , then
 
(3.1.1) lim fn (z) dz = f (z) dz.
n→∞ γ γ

Proof. Given  > 0, we choose N such that

|f (z) − fn (z)| < /(γ) for all n ≥ N, z ∈ γ(I).
Then Theorem 2.4.9 implies that, for n ≥ N ,
    
   
 f (z) dz − fn (z) dz  =  (f (z) − fn (z)) dz  ≤  (γ) = 
    (γ)
γ γ γ

and this proves (3.1.1). 

Example 3.1.4. If fn (z) = |z|n , then prove that the sequence {fn } converges
pointwise on D1 (0) but not uniformly. Show that it does converge uniformly on
any disc Dr (0) with r < 1.
Solution: If |z| < 1, then |z|n → 0. If |z| = 1, then |z|n → 1. Thus, the
sequence converges pointwise and the limit function f (z) is 0 if |z| < 1 and 1 if
|z| = 1. The convergence is not uniform because the limit function is not continuous.
3.1. Uniform Convergence 77

.8

.6

.4
x2
.2 x4
x8
x 16
x 32
0
0 .2 .4 .6 .8 1

Figure 3.1.1. The Sequence {xn } does not Converge Uniformly on [0, 1].

The fact that the convergence is not uniform can also be seen directly: No
matter how large n is chosen, we can always find a z with |z| < 1 such that
|fn (z) − f (z)| = |z|n ≥ 1/2. In fact, (1/2)1/n is such a z. Thus, the condition for
uniform convergence fails to hold for  = 1/2.
On the other hand, if z ∈ D r (0) with r < 1, then |z| ≤ r and |z|n ≤ r n . Given
 > 0, if N is chosen larger than log / log r, then n ≥ N implies
|fn (z) − f (z)| = |z|n ≤ r n < .
Since N was chosen independent of z, the convergence is uniform on Dr (0).

The real analogue of the above example is the sequence {xn } on [0, 1], which is
illustrated in Figure 3.1.1.

Uniform Convergence of Series. We say that an infinite series ∞ k=0 fk (z) of
functions, defined on a set E, converges uniformly on E if the sequence of partial
sums {sn } converges uniformly on E, where we recall that

n
sn (z) = fk (z).
k=0

There is a very useful criterion which insures uniform convergence of such a se-
quence. This is the Weierstrass M -test:
Theorem 3.1.5 (Weierstrass M -Test). Let
∞

(3.1.2) fk (z)
k=0
78 3. Power Series Expansions

be an infinite series of functions defined on a set E. If there is a convergent series

of non-negative numbers
∞

(3.1.3) Mk
k=0

such that |fk (z)| ≤ Mk for all k and all z ∈ E, then (3.1.2) converges uniformly on
E.

Proof. The comparison test, comparing (3.1.2) to (3.1.3), shows that, for each
z ∈ E, the series (3.1.2) converges. Let s(z) be the number it converges to, and let
sn (z) denote the nth partial sum of (3.1.2). Then
∞
 ∞

(3.1.4) |s(z) − sn (z)| ≤ |fk (z)| ≤ Mk .
k=n+1 k=n+1

Since the series (3.1.3) converges, given  > 0, we can choose N such that the right
side of (3.1.4) is less than  for all n ≥ N . Then (3.1.4) implies that |s(z)−sn (z)| < 
for all n ≥ N and all z ∈ E. Since N was chosen independently of z, this shows
that the convergence is uniform. 
∞ k 2
Example 3.1.6. Show that the series k=1 z /k converges uniformly on the
closed unit disc D1 (0).

Solution: We have |z k /k2 | ≤ 1/k2 if |z| ≤ 1. Furthermore, ∞ 2
k=1 1/k con-
verges,because it is a p-series with p = 2. Hence, by the Weierstrass M -test, the
∞
series k=1 z k /k2 converges uniformly on D1 (0).

Uniform Convergence of Power Series. In Section 1.2 we stated without proof

that a complex power series converges on a certain open disc and diverges at all
points in the complement of the corresponding closed disc. The radius of this disc
is called the radius of convergence of the power series. We are now prepared to
prove this result and give a formula for the radius of convergence.
The theorem that does this uses the notion of lim sup of a sequence. This is
defined as follows.
If S is a non-empty set of real numbers, then an upper bound for S is a number
M such that s ≤ M for every s ∈ S. If there is an upper bound for S, then S is said
to be bounded above. The completeness axiom for the real number system states
that each non-empty set of real numbers S that is bounded above, has a least upper
bound L. This means L has two properties: (1) it is greater than or equal to each
element of S, and (2) it is less than every other number with this property. We
will denote the least upper bound of a non-empty set S which is bounded above by
sup(S). If the set is non-empty, but not bounded above, we set sup(S) = ∞. The
notion of inf or greatest lower bound is defined analogously, but the inequalities are
all reversed.
Definition 3.1.7. If {ak } is a sequence of real numbers, then lim sup{ak } is the
limit of the non-increasing sequence {un } defined by
un = sup{ak : k ≥ n}.
3.1. Uniform Convergence 79

The sequence {un } of this definition is non-increasing because as n increases,

the set of numbers {ak : k ≥ n} gets smaller. The sequence {un } may not, however,
be bounded below and so lim sup{ak } may be −∞. Also, the un could all be +∞,
in which case lim sup{ak } = +∞.
Of course, there is an analogous notion, lim inf, which uses inf instead of sup
in the above definition.
The lim sup and lim inf of a sequence {ak } always exist (but may be infinite),
even though lim ak itself may not exist. In fact the sequence has a limit (which may
be infinite) if and only if lim sup ak = lim inf ak . In this case lim ak is this common
value.

Theorem 3.1.8. Given a power series ∞ k=0 ck (z − z0 ) , let
k

 −1
R = lim sup |ck |1/k .

Then the series converges absolutely if z ∈ DR (z0 ) and diverges if z ∈ / DR (z0 ).

Furthermore, it converges uniformly on each closed disc Dr (z0 ) with r < R. Thus,
R is the radius of convergence of the given power series.

Proof. Since we can always make a change of variables which replaces z − z0 by

z, we may as well assume that z0 = 0.
By definition, lim sup |ck |1/k is the limit of the non-increasing sequence {un }
where
un = sup{|ck |1/k : k ≥ n}.
If r < R, we choose a number t with r < t < R. Then t−1 > R−1 = lim un . This
implies that for large enough n the numbers un are less than t−1 . If n is one such
integer, then, for all k ≥ n,
|ck |1/k < t−1 and, hence, |ck | < t−k .
Now if |z| ≤ r, then this implies that
 r k
(3.1.5) |ck z k | < for all k ≥ n.
t
∞
Since |r/t| < 1, the geometric series k=n (r/t)k converges. Then the Weierstrass
∞ k
M -test implies that the series k=n c k z converges uniformly in the disc D r (0).
∞
The same is true of the original series k=0 ck z k , since the convergence or uniform
convergence of a series is unaffected by the first n terms if n is fixed.
Since the series converges on each closed disk Dr (0), with radius less than R,
it converges at each point z in the open disc DR (0).
It remains to prove that the series diverges at each z with |z| > R. Given such
a z, we have
|z|−1 < lim un .
This implies that, for each n, there is a k > n with
|z|−1 < |ck |1/k , so that |ck z k | > 1.
But this means that there is a subsequence of the sequence of terms {ck z k } con-
sisting of numbers with modulus greater than 1. Since the sequence of terms does
not converge to 0, the series diverges by the term test (Exercise 1.2.9). 
80 3. Power Series Expansions

The above theorem has the following corollary, the proof of which is left as an
exercise.
Corollary 3.1.9. If f has a power series expansion about z0 which converges on
the disc DR (z0 ), then f is continuous on this disc.
∞ k
Example 3.1.10. Prove that if  k=0 ck z is a power series with radius of con-
∞ k−1
vergence R, then the power series k=1 kck z also has radius of convergence
R.
Solution: If we multiply the second series by z, the set on which the series
converges does not change, and so its radius of convergence does not change. The
∞
resulting series is k=0 kck z k . Let R1 be its radius of convergence. By the previous
theorem,
 −1
R = lim sup |ck |1/k ,
and  −1  −1
R1 = lim sup |kck |1/k = lim sup k1/k |ck |1/k .
The sequence k1/k has limit 1 (Exercise 3.1.11), and so the factor k1/k does not
effect the lim sup (Exercise 3.1.12). Hence, R1 = R.

Theorem 3.1.8 and Theorem 3.1.3 combine to prove that it is legitimate to

integrate a power series term by term.

Theorem 3.1.11. Let f (z) = ∞ k=0 ck (z − z0 ) , where the radius of convergence
k

of this power series is R. Then

 z ∞
 ck
(3.1.6) f (w) dw = (z − z0 )k+1
z0 k+1
k=0

for all z ∈ DR (z0 ).


Proof. If we set sn (z) = nk=0 ck (z − z0 )k for each positive integer n, then
 z n
ck
sn (w) dw = (z − z0 )k+1
z0 k +1
k=0

because the integral is linear and we know how to integrate (z − z0 )k . To finish

the proof of (3.1.6), we just need to take the limit of both sides and use Theorem
3.1.3 to bring the limit inside the integral on the left. Of course, we need to know
that the convergence of {sn } to f is uniform on [z0 , z]. This, however, follows
from Theorem 3.1.8, since the interval [z0 , z] is inside the closed disc D r (z0 ), where
r = |z| < R. 

Exercise Set 3.1

1. Show that the sequence {1/(nz)} converges uniformly to 0 on every set of the
form {z : |z| ≥ r} for fixed r > 0, but it does not converge uniformly on
{z : z = 0}.
3.1. Uniform Convergence 81

2. Show that the sequence {sin(x/n)} converges uniformly to 0 on any interval of

the form [0, k], but it does not converge uniformly on [0, ∞).
3. Show that the sequence {arctan(nx)} converges pointwise but not uniformly
on R.
4. Use the Weierstrass M -test (but not Theorem 3.1.8) to show that the series
∞
 zk
converges uniformly on DR (0) for each R > 0.
k!
k=1
∞
 k+z
5. Use the Weierstrass M -test to show that the series converges uni-
k3 + 1
k=1
formly on D1 (0).
6. Show that for each r > 0 the series
∞
1
k2 − z
k=0

converges uniformly on the set

Er = {z : |z| ≤ r, z = k2 for k = 0, 1, 2, · · · }.
∞
7. Prove that the series k=1 k−z converges uniformly on each set of the form
{z ∈ C : Re(z) > s}, with s > 1. The function to which it converges is called
the Riemann Zeta Function.
8. If the series of the previous exercise is differentiated term by term, does the
resulting series still converge uniformly on {z ∈ C : Re(z) > s} if s > 1?
9. For each n find sup{1 + (−1)k + 1/k : k ≥ n}.
∞
10. Find the radius of convergence of the power series k=0 (2 + (−1)k )k z k .
11. Prove that lim k1/k = 1.
12. Prove that if {ak } and {bk } are two sequences of non-negative numbers with
lim ak = a and lim sup bk = b, then lim sup ak bk = ab.
13. Prove Corollary 3.1.9.
∞
14. Can a power series of the form k=0 ck (z − 1)k converge at z = 3 and diverge
at z = 0? Why?
∞
1
15. Using the power series expansion = (−1)k wk , find a power series
1+w
 z k=0
1
expansion for dw about 0. What is the radius of convergence of this
0 1+w
power series? What function does it converge to?
16. If a function E(z) is defined on C by
 z
e−w dw,
2
E(z) =
0

find a power series expansion for E(z) about 0. Where does this power series
converge?
82 3. Power Series Expansions

3.2. Power Series Expansions

A function of a real variable can be differentiable, even infinitely differentiable on an
interval and still not have a convergent power series expansion in that interval. For
functions of a complex variable, the situation is quite different. We will prove that
every analytic function has convergent power series expansions about each point of
its domain. In fact, we will prove that a function f is analytic if and only if it has
such expansions. First, we show that a function which has a power series expansion
on a disc is analytic on that disc. This involves showing that we can differentiate
power series term by term.

Differentiating Power Series. The next theorem and its corollaries concern a
function f defined by a convergent power series
∞

(3.2.1) f (z) = cn (z − z0 )n
n=0

with radius of convergence R.

Theorem 3.2.1. If f is defined as above, then f is analytic on DR (z0 ) and f  has
a convergent power series expansion
∞

f  (z) = ncn (z − z0 )n−1 ,
n=1

which converges to f  on DR (z0 ).

Proof. If f has a power series expansion (3.2.1) with radius of convergence R, let
g be the function which we hope turns out to be the derivative of f – that is, we
set
∞
g(z) = ncn (z − z0 )n−1 .
n=1

This series has the same radius of convergence as the series for f (see Example
3.1.10) and so it converges on DR (z0 ) and converges uniformly on any smaller
closed disc. By Corollary 3.1.9, g is continous on DR (z0 ), and by Theorem 3.1.11,
 z ∞
g(w) dw = cn z n = f (z) − f (z0 )
z0 n=1

on DR (z0 ). Theorem 2.6.1 tells us this function is an antiderivative for g(z). Since
f (z0 ) is a constant, this means f  = g on DR (z0 ), as required. 

Example 3.2.2. Find a power series expansion for the principal branch of the log
function about the point z0 = 1.
Solution: We know that the derivative of log(z) is 1/z (Exercise 2.2.14). We
also know that, since
1 1
= ,
z 1 − (1 − z)
3.2. Power Series Expansions 83

the power series expansion of 1/z about z0 = 1 is

∞ ∞
1  
(3.2.2) = (1 − z)n = (−1)n (z − 1)n .
z n=0 n=0

Since we can differentiate term by term, a series which has this as its derivative is
the series
∞ ∞
n (z − 1) (z − 1)n
n+1
(3.2.3) (−1) =− (−1)n−1 .
n=0
n+1 n=1
n
The series (3.2.2) and (3.2.3) both have radius of convergence 1. If f (z) denotes
the sum of series (3.2.3), then f  (z) = 1/z = log (z). It follows that f and log
differ by a constant. Since they both have the value 0 at z = 1, they are the same.
Therefore, (3.2.3) is the power series expansion of log(z) about 1.

Using Theorem 3.2.1, the following can be proved in the same way as its real
variable counterpart. The details are left to the exercises.
Corollary 3.2.3. If f has a power series expansion (3.2.1) about z0 , with radius
of convergence R, then it has derivatives of all orders on DR (z0 ). Its kth derivative
is
∞
 n! cn
(k)
f (z) = (z − z0 )n−k .
(n − k)!
n=k
In particular, its kth derivative at z0 is given by
f (k) (z0 ) = k! ck .

This immediately implies:

Corollary 3.2.4. If f has a power series expansion (3.2.1) about z0 , with positive
radius of convergence, then it has only one such expansion. In fact, the coefficients
{cn } for such an expansion are uniquely determined by the equations
f (n) (z0 )
cn = .
n!
Power Series Expansions of Analytic Functions. We are now ready to present
the most important application of Cauchy’s theorems – the proof of the existence
of local power series expansions of analytic functions.
Theorem 3.2.5. Let f be analytic in an open set U and suppose Dr (z0 ), r > 0, is
an open disc contained in U . Then there is a power series expansion for f
∞

f (z) = cn (z − z0 )n ,
n=0

which converges to f (z) on Dr (z0 ). Furthermore, the coefficients of this power

series are the numbers

1 f (w)
(3.2.4) cn = dw,
2πi |w−z0 |=s (w − z0 )n+1
where s is any number with 0 < s < r.
84 3. Power Series Expansions

z t s
r
z0 U

Figure 3.2.1. Setup for the Proof of the Existence of Power Series Expansions.

Proof. If 0 < t < s < r, |w − z0 | = s, and |z − z0 | ≤ t, then

 
 z − z0  t
 
 w − z0  ≤ s < 1,
and so we have
 −1  ∞  n
w − z0 z − z0 z − z0
(3.2.5) = 1− = ,
w−z w − z0 n=0
w − z0
where
∞ the geometric series on the right is dominated by the constant geometric series
n
n=0 (t/s) . By Theorem 3.1.5, this implies that (3.2.5) converges uniformly as a
function of z ∈ Dt (z0 ) and also as a function of w ∈ ∂Ds (z0 ).
If we multiply (3.2.5) by f (w)/(w −z0 ) and integrate around the circle of radius
s, then, since the series converges uniformly in w, we may integrate term by term.
Using Cauchy’s Integral Formula, this yields

1 f (w)
f (z) = dw
2πi |z0 −w|=s w − z
∞
  
1  f (w)
= dw (z − z0 )n .
|w−z0 |=s (w − z0 )
2πi n=0 n+1

This gives us a power series expansion of f on Ds (z0 ) with coefficients given by

(3.2.4).
Now given any z ∈ Dr (z0 ), we may choose s such that |z − z0 | < s < r. With
this choice of s, the above series is defined and converges at z. However, it follows
from Cauchy’s Integral Theorem that the integrals defining the coefficients of this
series do not depend on the choice of s (see Exercise 2.6.9). Thus, we have a power
series expansion for f which converges on Dr (z0 ), with coefficients given by (3.2.4).
This completes the proof. 

Corollary 3.2.6. If f is analytic on an open set U , then f has derivatives of all

orders on U and they are all analytic.
3.2. Power Series Expansions 85

z0 U z
z1

Figure 3.2.2. Shift to an Expansion About a Different Point of U .

Proof. Given any point z0 ∈ U , there is a disc Dr (z0 ) centered at z0 which is

contained in U . On this disc, f has a convergent power series expansion. By
Corollary 3.2.3 f has derivatives of all orders on this disc. These derivatives are
analytic because each of them has a complex derivative. Therefore, f has analytic
derivatives of all orders on all of U . 

It is important to emphasize that the power series expansion of an analytic

function f about a point z0 converges in the largest open disc, centered at z0 , that
is contained in the domain U of f . In general, it will not converge at other points
of U . To obtain a power series expansion of f that converges in a neighborhood
of a point z outside this disc, we have to shift to a power series expansion about a
different point of U – a point z1 with the property that the largest disc centered
at z1 and contained in U contains the point z (see Figure 3.2.2). For example,
the power series expansion of log z about z = 1 given in Example 3.2.2 has radius
of convergence 1, and so it does not converge at z = i. There is a power series
expansion of log z about z = i (Exercise 3.2.7), but it is a different power series
than the one about z = 1.
Example 3.2.7. Show that if f is a function which is analytic in an open set U ,
z0 ∈ U , and f (z0 ) = 0, then g(z) = f (z)/(z − z0 ) can be given a value at z0 which
makes it analytic on U .
Solution: By the above theorem f has a power series expansion about z0
which converges in the largest open disc D which is centered at z0 and contained
in U . Since f (z0 ) = 0, the constant term of this series is 0. Hence, this expansion
has the form
f (z) = c1 (z − z0 ) + c2 (z − z0 )2 + · · · + cn (z − z0 )n + · · · .
If we give the function g(z) the value c1 at z = z0 , then it agrees on D with the
sum of the power series
c1 + c2 (z − z0 ) + c3 (z − z0 )2 + · · · + cn (z − z0 )n−1 + · · · .
Since it has a power series expansion on D, g is analytic on D. Also, g is analytic
in U \ {z0 } since f is analytic, and z − z0 is analytic and non-vanishing on this set.
86 3. Power Series Expansions

A function, defined on the union of two open sets and analytic on each of them, is
clearly also analytic on their union; and so g is analytic on U .

Cauchy’s Estimates. We now know that a function which is analytic in an

open disc Dr (z0 ) has a power series expansion about z0 which converges in that
disc. We also have integral formulas (3.2.4) for the coefficients of this power series.
However, we also have the formulas
f (n) (z0 )
(3.2.6) cn =
n!
for these coefficients from Corollary 3.2.4. Combining these yields integral formulas
for the derivatives of an analytic function.
Theorem 3.2.8. Let f be analytic in an open set containing the closed disc DR (z0 ).
Then,

(n) n! f (w)
(3.2.7) f (z0 ) = dw
2πi |w−z0 |=R (w − z0 )n+1
for n = 0, 1, 2, · · · .

Proof. We just need to observe that if D R (z0 ) ⊂ U , with U open, then there is an
open disc Dr (z0 ) with DR (z0 ) ⊂ Dr (z0 ) ⊂ U (Exercise 3.2.4). We can then apply
Theorem 3.2.5 with s = R to obtain a power series expansion of f with coefficients
given by (3.2.4). Then Corollary 3.2.3 relates these coefficients to the derivatives
of f at z0 . 

This leads to a very powerful tool. By estimating the size of the integrands
in this formula, we can get estimates on the size of the derivatives of f . These
estimates are called Cauchy’s estimates.
Theorem 3.2.9 (Cauchy’s Estimates). If f is analytic on an open set containing
the closed disc DR (z0 ), and if |f (z)| ≤ M on the boundary of this disc, then
n!M
(3.2.8) |f (n) (z0 )| ≤
Rn
for n = 0, 1, 2, · · · .

Proof. We use the previous theorem. Since |w − z0 | = R and |f (w)| ≤ M for w on

the path |w − z0 | = R, the integrand of (3.2.7) is bounded by M/Rn+1 . The length
of the path is the circumference of a circle of radius R and so it is 2πR. Thus,
Theorem 2.4.9 implies that
n! M n!M
|f (n) (z0 )| ≤ n+1
2πR = n
2π R R
for each non-negative integer n. 

This theorem will provide the crucial step in the proof of Liouville’s Theorem
in the next section.
3.2. Power Series Expansions 87

Example 3.2.10. Find upper bounds on the derivatives at 0 of a function f which

is analytic on the unit disc D1 (0) and has modulus bounded by one on this disc.
Also find bounds on the moduli of coefficients in the power series expansion of this
function about 0.
Solution: We apply Cauchy’s estimates. If r < 1, then f is analytic in the
open set D1 (0), which contains Dr (0). Since |f (z)| ≤ 1 on D1 (0), we may choose
M = 1 and R = r in Cauchy’s estimates. We conclude
n!
|f (n) (0)| ≤ n .
r
However, since r was any positive number less than 1, we may pass to the limit as
r → 1 and conclude that
|f (n) (0)| ≤ n!.
By (3.2.6), the corresponding estimate on the power series coefficients is
|cn | ≤ 1.

Morera’s Theorem. This is a very handy tool for showing that a function is
analytic.
Theorem 3.2.11 (Morera’s Theorem). Let f be a continuous function defined on
an open set U . If the integral of f is 0 around the boundary of every triangle that
is contained in U , then f is analytic in U .

Proof. Theorem 2.6.1 says that a function f that is continuous on a convex open
set U , and has the property that its integral around any triangle in U is 0, has
a complex antiderivative g in U . However, the fact that g  = f on U means, in
particular, that g is analytic on U . But then f is analytic on U by Corollary 3.2.6.
The hypothesis that U is convex in the above argument is not necessary, since
every open set is a union of convex open sets (open discs, in fact). Thus, we can
apply the argument of the previous paragraph to each open disc contained in U
and conclude that f is analytic on each of them. In particular, it has a derivative
at each point of U and, hence, is analytic on U . 

An example of how Morera’s Theorem is used is provided by the following

theorem.
Theorem 3.2.12. Let {fn } be a sequence of analytic functions on an open set U
and suppose this sequence converges uniformly to f on each compact subset of U .
Then f is analytic on U .

Proof. Since fn → f uniformly on each compact subset of U , f is continuous on

U . The convergence is uniform, in particular, on ∂Δ for every triangle Δ contained
in U . Given such a triangle Δ, Theorem 3.1.3 implies that
 
f (z) dz = lim fn (z) dz.
∂Δ n→∞ ∂Δ
Since each fn is analytic, Theorem 2.5.8 implies

fn (z) dz = 0
∂Δ
88 3. Power Series Expansions

for each n. We conclude that


f (z) dz = 0,
∂Δ

for every triangle Δ ⊂ U . By Morera’s Theorem, f is analytic on U . 

Exercise Set 3.2

1
1. Use the power series expansion for about 0 to find the power series
1−z
1
expansion of about 0.
(1 − z)2
√
2. Find a power series expansion of 1 + z about 0, where the square root function
is defined in terms of the principal branch of the log function. What is the radius
of convergence of this series?
3. Prove Corollary 3.2.3.
4. Prove that if DR (z0 ) is a closed disc contained in an open set U , then there is
an open disc Dr (z0 ) such that DR (z0 ) ⊂ Dr (z0 ) ⊂ U .
5. Show that if f is analytic on an open set U , then, as a function on R2 , it is C∞
– that is, its partial derivatives of all orders exist and are continuous.
1
6. The function has a power series expansion about z0 = 0. Without finding
cos z
the series, show that its radius of convergence is π/2.
7. Find the power series expansion of the principal branch of the log function
about the point z = i. There are several ways to do this, one of which is really
easy (see Example 3.2.2).
sin z
8. Use power series methods to show that the function which is when z = 0
z
and 1 when z = 0 is analytic on the whole complex plane.
9. If f is analytic and not identically 0 on a disc Dr (z0 ), show that there is a
non-negative integer k and a function g, which is also analytic in Dr (z0 ), such
that f (z) = (z − z0 )k g(z) and g(z0 ) = 0.
10. Prove that if f is analytic on the disc DR (z0 ) and |f (z)| ≤ M on DR (z0 ), then
|f  (z0 )| ≤ M/R.
11. Suppose p(z) = a3 z 3 + a2 z 2 + a1 z + a0 is a polynomial of degree 3. If |p(z)| ≤ 1
on the unit circle {z : |z| = 1}, then show that |a3 | ≤ 1.
12. Suppose f is analytic in an open set U . Also, suppose z ∈ U and the distance
from z to the complement of U is d. If |f (w)| ≤ M for all w ∈ U , find estimates,
similar to Cauchy’s estimates, on the size of |f (n) (z)| in terms of M and d.
13. Suppose f is analytic on a disc Dr (z0 ) and unbounded (there is no M such
that |f (z)| ≤ M on Dr (z0 )). Then prove that the radius of convergence of the
power series expansion of f about z0 is r.
3.3. Liouville’s Theorem 89

14. Use Morera’s Theorem to show that if f is continuous on an open set U and
analytic on U \ E, where E is either a point or a line segment, then f is actually
analytic on all of U .
15. Use Cauchy’s estimates to prove that if {fn } is a sequence of analytic functions
on an open set U , converging uniformly to f on each compact subset of U , then
(k)
{fn } converges uniformly to f (k) on each compact subset of U .
16. Use Morera’s Theorem to prove that if U is an open subset of C, I = [a, b] is
an interval on the real line, and g(z, t) is a continuous function on U × I which
is analytic in z for each t ∈ I, then the function
 b
f (z) = g(z, t) dt
a

is analytic in U .

3.3. Liouville’s Theorem

Liouville’s Theorem is simple to state, very easy to prove (given what we know at
this point), and extremely powerful. It concerns entire functions, where an entire
function is a function which is analytic in the entire complex plane. It also concerns
bounded functions, where a function is bounded on a set E if there is a positive
constant M such that |f (z)| ≤ M for every z ∈ E. If f is bounded on its domain,
we simply say it is bounded. Thus, a bounded entire function is a function which is
analytic and bounded on C.

Theorem 3.3.1 (Liouville’s Theorem). The only bounded entire functions are the
constant functions.

Proof. The reader who did Exercise 3.2.10 of the preceding section has nearly com-
pleted the proof of Liouville’s Theorem. The exercise states a simple consequence
of the Cauchy estimates: If a function f is analytic on a disc of radius R, centered
at z0 , and if |f (z)| ≤ M for all z in this disc, then
M
(3.3.1) |f  (z0 )| ≤ .
R

If f is bounded and entire, then f (z) is analytic on the entire plane and |f (z)|
is bounded by some number M on the entire plane. This implies that (3.3.1) holds
for all positive numbers R and all z0 ∈ C. If we take the limit as R → ∞ in (3.3.1),
we conclude that |f  (z0 )| = 0 for all z0 ∈ C. In other words, the derivative of f is
identically zero. This implies f is a constant (see Exercise 2.6.1). 

One has to see the consequences of this theorem to appreciate its power. In
the remainder of this section and the exercises we will introduce a number of these.
Others will occur later.
90 3. Power Series Expansions

The Fundamental Theorem of Algebra. At the very beginning of the text, we

promised that we would prove that every non-constant polynomial with complex
coefficients has a complex root. This is the Fundamental Theorem of Algebra.
Before proving this theorem, it will be convenient to introduce limits at infinity.
Definition 3.3.2. If f is a function defined on an unbounded set E (so E contains
points of arbitrarily large modulus) and L ∈ C, then we say
lim f (z) = L
z→∞
if, for every  > 0, there is an R > 0 such that |f (z) − L| <  whenever |z| > R and
z ∈ E.

This concept satisfies the same basic rules as other kinds of limits: limit of the
sum is sum of the limits, limit of the product is product of the limits, limit of the
quotient is quotient of the limits if the denominator does not have limit zero, etc.
These facts, as well as the fact that
1
lim =0
z→∞ z
(Exercise 3.3.1), are used in the proof of the Fundamental Theorem of Algebra.
Before proving that theorem, we prove the following simple result.
Theorem 3.3.3. If a function f is defined and continuous on the entire plane and
if limz→∞ f (z) exists, then f is bounded on C.

Proof. If limz→∞ f (z) = L, there exists an R > 0 such that |f (z) − L| < 1
whenever |z| > R. By the triangle inequality, this implies
|f (z)| < |L| + 1 if |z| > R.
Since f is continuous on C and DR (0) is closed and bounded, hence compact, f is
bounded on DR (0). Since f is bounded on DR (0) and on its exterior, it is bounded
on all of C. 
Theorem 3.3.4 (Fundamental Theorem of Algebra). Every non-constant complex
polynomial has a complex root.

Proof. Let p(z) = an z n + an−1 z n−1 + · · · + a1 z + a0 be a non-constant complex

polynomial of degree n. Then n ≥ 1 and an = 0. We will show the assumption
that p has no root leads to a contradiction.
If p has no root, then p(z) = 0 for every z ∈ C. This implies that 1/p is an
entire function. We will show that it is also bounded.
If we define a function h(z) by
zn 1
h(z) = = ,
p(z) an + an−1 z −1 + · · · + a1 z n−1 + a0 z −n
then
1 h(z)
(3.3.2) = n
p(z) z
for z = 0. Furthermore
lim h(z) = 1/an
z→∞
3.3. Liouville’s Theorem 91

and so
1 h(z)
lim = lim = 0.
z→∞p(z) z→∞ z n
Since the limit of 1/p exists at infinity, the previous theorem implies that 1/p is
bounded on all of C. So Liouville’s Theorem implies that it is a constant. In
fact, the constant must be zero, since 1/p(z) has limit 0 at ∞. This is clearly a
contradiction, since 1/p(z) cannot take on the value 0 on C. We conclude that p(z)
must have a root. 

The Fundamental Theorem of Algebra has a number of important consequences.

We will discuss only a few of them.

Factoring Polynomials. A polynomial is said to factor completely if it can be

written as a product
p(z) = b(z − z1 )(z − z2 ) · · · (z − zn )
of linear factors. Here, n is the degree of p and the numbers z1 , z2 , · · · , zn are
the roots of p. The roots need not all be distict. If a root occurs k times in this
factorization, it is said to be a root of mutiplicity k. Necessarily, the number of
roots is the degree of the polynomial if each root is counted as many times as its
multiplicity.
Corollary 3.3.5. Each complex polynomial factors completely.

Proof. The proof is by induction on the degree n of the polynomial. Obviously

a polynomial of degree 0 or 1 factors completely. If every polynomial of degee n
factors completely and p is a polynomial of degree n + 1, then we use the previous
theorem to assert that p has a root – call it zn+1 . It follows that p factors as
P (z) = (z − zn+1 )q(z),
where q is a polynomial of degree n. By assumption, q factors as
q(z) = b(z − z1 )(z − z2 ) · · · (z − zn ).
Then p factors as
p(z) = b(z − z1 )(z − z2 ) · · · (z − zn )(z − zn+1 ).
This completes the induction step and finishes the proof of the corollary. 

Eigenvalues of Matrices. An eigenvalue for an n × n matrix A is a complex

number λ such that the matrix λI − A is singular (has no inverse). Here, I is the
n × n identity matrix.
Corollary 3.3.6. An n × n matrix A has at least one complex eigenvalue.

Proof. The determinant det(λI − A) of λI − A is a polynomial in λ of degree n.

This is the characteristic polynomial of A. By Kramer’s Rule, λI − A is singular
for a given λ if and only if det(λI − A) = 0 – that is, if and only if λ is a root
of the charcteristic polynomial. By the Fundamental Theorem of Algebra, the
characteristic polynomial has a root. Thus, A has an eigenvalue. 
92 3. Power Series Expansions

The above result is the essential ingredient in the proof that every complex
matrix can be put in upper triangular form (Exercise 3.3.14).
 
1 2
Example 3.3.7. Find the eigenvalues of the matrix .
−1 3
Solution: The characteristic polynomial of this matrix is
 
λ−1 −2
det = λ2 − 4λ + 5.
1 λ−3
By the Quadratic Formula, the roots of this polynomial are 2 ± 2i. Thus, the
eigenvalues of the above matrix are 2 + 2i and 2 − 2i.

Differential Equations. The Fundamental Theorem of Algebra also has appli-

cations to Differential Equations. A homogeneous constant-coefficient differential
equation is an equation of the form
(3.3.3) an y (n) + an−1 y (n−1) + · · · + a1 y  + a0 y = 0.
Example 3.3.8. Prove that each homogeneous constant-coefficient differential
equation has a solution of the form y = eλx , where λ is a complex number. Which
numbers λ yield solutions?
Solution: We simply plug y = eλx into the equation (3.3.3) and see if there
are values of λ that yield solutions. Since (eλx ) = λ eλx , the result is
(an λn + an−1 λn−1 + · · · + a1 λ + a0 ) eλx = 0.
The polynomial in parentheses is called the auxiliary polynomial for (3.3.3). Clearly
eλx is a solution of (3.3.3) if and only if λ is a root of this polynomial. Since every
non-constant polynomial has a root, equation (3.3.3) has a solution of the form
y = eλx .
If the auxiliary polynomial has distinct roots λ1 , · · · , λn , then the general so-
lution to equation (3.3.3) is a linear combination of the solutions eλj x .

If the coefficients of (3.3.3) are real numbers and we are looking only for real so-
lutions to this differential equation, then we exploit the fact that the non-real roots
of a polynomial with real coefficients occur in conjugate pairs (Exercise 3.3.10). If
λ = α + iβ and λ = α − iβ are roots of the auxilliary equation for (3.3.3), then

eλx + eλx
eαx cos βx = and
2
e λx
− eλx
eαx sin βx =
2i
both give real solutions to (3.3.3), as does any linear combination
C eαx cos βx + D eαx sin βx = eαx (C cos βx + D sin βx),
where C and D are arbitrary real constants.
Example 3.3.9. Find all solutions to the differential equation y  − 2y  + 5y = 0.
Then find all real solutions.
3.3. Liouville’s Theorem 93

Solution: The auxiliary equation is λ2 − 2λ + 5 = 0, which has solutions

λ = 1 ± 2i. Thus, the general solution to the differential equation is

y = A e(1+2i)x +B e(1−2i)x ,
where A and B are complex constants. The general real solution is
y = ex (C cos 2x + D sin 2x),
where C and D are real constants.

Characterization of Polynomials. In the proof of Liouville’s Theorem, we only

use Cauchy’s estimate on the first derivative of an analytic function on a disc. Here
we present a generalization of Liouville’s Theorem that uses Cauchy’s estimates on
higher derivatives.

Theorem 3.3.10. An entire function f is a polynomial of degree at most n if and

only if there are positive constants A and B such that
(3.3.4) |f (z)| ≤ A + B|z|n
for all z ∈ C.

Proof. We will prove that every polynomial satisfies an inequality of the form
(3.3.4). We leave the converse as an exercise in the application of Cauchy’s esti-
mates.
Suppose p(z) = an z n + an−1 z n−1 + · · · + a1 z + a0 is a polynomial of degree at
most n. Then
p(z)
lim = an .
z→∞ z n

If we apply the definition of limit, using  = 1, we conclude that there is an R > 0

such that |z| > R implies
 
 p(z) 
 − a 
n  < 1.
 zn
It follows from the triangle inequality that
|p(z)|
< |an | + 1 if |z| > R.
|z|n
If we set B = |an | + 1, then
|p(z)| < B|z|n if |z| > R.

This gives a bound on |p| on the complement of the closed disc DR (0). A closed
disc of finite radius is a compact set and so p is bounded on DR (0) – say, |p(z)| ≤ A
on this disc. If we combine this with our bound on the complement of D R (0), we
conclude that
|p(z)| ≤ A + B|z|n
on all of C, as required. 
94 3. Power Series Expansions

Exercise Set 3.3

1
1. Prove that lim = 0.
z→∞ z
2. We say limz→∞ f (z) = ∞ if for each K > 0, there is an M > 0 such that
|f (z)| > K whenever |z| > M . Prove that limz→∞ f (z) = ∞ if and only if
limz→∞ 1/f (z) = 0.
3. Show that if f is an entire function and limz→∞ f (z) = ∞, then f must have
a zero somewhere in C. Hint: See the previous exercise.
4. Prove that if f is an entire function which satisfies |f (z)| ≥ 1 on the entire
plane, then f is constant.
5. Prove that if an entire function has real part which is bounded above, then the
function is constant.
6. Prove that if an entire function f is not constant, then its range f (C) is dense in
C – meaning that the closure of f (C) is C. Hint: If f (C) is not dense, then there
is a point z0 and a disc Dr (z0 ), centered at z0 , such that f (C) ∩ Dr (z0 ) = ∅.
7. Show that if f is an entire function and |f (z)| ≤ K|z| for all z ∈ C, where K is
a positive constant, then f (z) = Cz for some constant C.
8. Show that if f is an entire function and |f (z)| ≤ K| ez | for all z ∈ C, where K
is a positive constant, then f (z) = C ez for some constant C.
9. Finish the proof of Theorem 3.3.10 by using Cauchy’s estimates to prove that
the only entire functions f that satisfy an inequality of the form |f (z)| ≤
A + B|z|n for all z ∈ C are polynomials of degree at most n.
10. Show that if p is a polynomial with real coefficients, then the non-real roots of
p occur in conjugate pairs. That is, show that if w is a root, then so is w.
11. Find the roots of the polynomial p(z) = z 2 − 2z + 2.
 
1 1
12. Find the eigenvalues of the matrix .
−1 1
13. Find all solutions of the differential equation y  − 4y  + 13y = 0. Then find all
real solutions.
14. Use Corollary 3.3.6 to prove that if A is an n × n matrix, then A is conjugate
to an upper triangular matrix.
15. Use the result of Exercise 10 to prove that every polynomial with real coeffi-
cients factors as a product of polynomials of degree at most 2, also with real
coefficients.
16. Are there any non-constant entire functions f that satisfy an inequality of the
form
|f (z)| ≤ A + B log |z| for all z with |z| ≥ 1,
where A and B are positive constants?
3.4. Zeroes and Singularities 95

3.4. Zeroes and Singularities

The existence of power series expansions leads to a great deal of information about
the local structure of analytic functions. For example, Exercise 3.2.9 makes an
assertion that is important enough to be expanded and restated as a theorem.
Theorem 3.4.1. If f is a function which is analytic in a open set U , then for each
z0 ∈ U , exactly one of the following statements is true:
(1) there is an open disc, centered at z0 , on which f is identically 0;
(2) there is a non-negative integer k, an open disc Dr (z0 ), and a function g,
analytic on U , such that
(3.4.1) f (z) = (z − z0 )k g(z) for all z ∈ Dr (z0 )
and g(z) is non-zero at each point of Dr (z0 ).

Proof. The function f has a power series expansion

∞

(3.4.2) f (z) = cn (z − z0 )n
n=0

which converges in a disc DR (z0 ) of positive radius R. Some of the coefficients cn

may be zero. If they are all zero, then f is identically zero on DR (z0 ) and so (1)
holds in this case.
If the coefficients cn are not all zero, there is a smallest n for which cn = 0 –
call it k. Then (3.4.2) becomes
∞
 ∞

(3.4.3) f (z) = cn (z − z0 )n = (z − z0 )k cn+k (z − z0 )n ,
n=k n=0

where ck = 0. We may then define g by

⎧ ∞
⎪
⎪
⎨ cn+k (z − z0 )n , if z ∈ DR (z0 );
g(z) = n=0
⎪
⎪
⎩ f (z) , if z ∈ U \ {z0 }.
(z − z0 )k
This is a consistent definition, since on DR (z0 ) \ {z0 } the functions f (z)/(z − z0 )k
and ∞ n=0 cn+k (z − z0 ) are equal.
n

Since g is continuous and g(z0 ) = ck = 0, there is a positive r < R such that

|g(z) − g(z0 )| < |ck | whenever |z − z0 | < r. This implies g(z) = 0 if |z − z0 | < r. In
other words, g(z) = 0 for every z ∈ Dr (z0 ). In this case, (2) holds. 

Zeroes of Analytic Functions. By a zero of a function, we mean a point in

its domain at which it vanishes (has value 0). The previous theorem leads to an
important and somewhat surprising result about the zeroes of an analytic function
f on a connected open set: Unless f is identically zero, each zero of f has no
neighboring zeroes. This is made precise in Part (b) of the following theorem.
Theorem 3.4.2. Let f be a function which is analytic on a connected open set U
and is not identically zero. Then
96 3. Power Series Expansions

(a) for each z0 ∈ U , there is a non-negative integer k, an r > 0, and a function g,

analytic in U , such that
f (z) = (z − z0 )k g(z) for all z ∈ U,
and g(z) has no zeroes on Dr (z0 );
(b) each z0 ∈ U is the center of an open disc Dr (z0 ) ⊂ U in which there are no
zeroes of f except possibly z0 itself;
(c) the set of zeroes of f is at most countable.

Proof. Theorem 3.4.1 shows that, at each point z0 of U , there are two possibilities:
(1) there is a disc centered at z0 in which f is identically 0, and (2) there is a
disc centered at z0 in which f has no zeroes except possibly at z0 itself. Let Vj ,
j = 1, 2 be the set of points z0 for which the jth possibility is the one which occurs.
Obviously, V1 and V2 are both open sets, V1 ∩ V2 = ∅, and V1 ∪ V2 = U . Thus,
either V1 or V2 is empty, since, otherwise, they would separate the connected set
U . If V2 = ∅, then f is identically 0 on U . Since f is not identically 0, we conclude
that V1 = ∅. Since the second of the possibilities in Theorem 3.4.1 is the only one
that occurs in this situation, we conclude that (a) and (b) above both hold at every
z0 ∈ U .
We can modify the disc Dr (z0 ) in statement (a) of the theorem so that it has a
center (which may no longer be z0 ) with rational coordinates and a rational radius
and still has the property that it contains z0 and has no zeroes other than possibly
z0 . We simply choose a point z0 with rational coordinates, and a positive rational
number ρ such that |z0 − z0 | < ρ < r/2. Then z0 ∈ Dρ (z0 ) ⊂ Dr (z0 ) and so Dρ (z0 )
contains z0 but no other zeroes of f . Since we may do this for each zero of f in U
and since there are only countably many discs with rational centers and rational
radii, we conclude that f can have only countably many zeroes in U . 

Recall that an isolated point of a set E is a point which is contained in an open

disc which contains no other points of E. Thus, if Z(f ) denotes the set of zeroes of
an analytic function f on its domain U , then Part (b) of the above theorem implies
that each point of Z(f ) is an isolated point of Z(f ). Actually it says something
much stronger. It says that every point z0 of U (whether in Z(f ) or not) is the
center of a disc containing no points of Z(f ) other than z0 . There is a term to
describe subsets with this property:
Definition 3.4.3. Let U be an open subset of C and E a subset of U . We say
that E is a discrete subset of U if every point z0 of U has a neighborhood which
contains no points of E other than possibly z0 itself.

Thus, Part (b) of the above theorem says that the zero set Z(f ) of a non-
constant analytic function on a connected open set U is a discrete subset of U .
It turns out that a subset of an open set U is discrete if and only if no sequence
of distinct points of E converges to a point of U (see Exercise 3.4.1). Of course,
there may be sequences in E which converge to points not in U .
Theorem 3.4.2 has the following easy but important consequence. We leave the
proof as an exercise (Exercise 3.4.2).
3.4. Zeroes and Singularities 97

Theorem 3.4.4 (Identity Theorem). Suppose f and g are two analytic functions
with domain a connected open set U . If f (w) = g(w) at each point w of a non-
discrete subset E of U , then f (z) = g(z) at every point of U .

If the f of Theorem 3.4.2 actually has a zero at z0 , then the integer k is positive.
We call it the order of the zero of f at z0 .
Example 3.4.5. What is the order of the zero of the function f (z) = cos z − 1 at
0? What is the function g of Part (a) of Theorem 3.4.2 in this case if z0 = 0?
Solution: If we subtract 1 from the power series expansion of cos z about 0,
we obtain
z2 z4 z 2n
cos z − 1 = − + + · · · (−1)n + ···
2!
 4! 2n! 
1 z2 z 2n−2
= z2 − + + · · · (−1)n + ··· .
2! 4! 2n!
We conclude that the order of the zero of cos z − 1 at 0 is 2 and the function g of
Part (a) of Theorem 3.4.2 is the function given by the power series in parentheses
above. This power series has infinite radius of convergence by the ratio test.
Theorem 3.4.6. If an analytic function g is not zero at a point z0 in its domain,
then in some neighborhood V of z0 there is an analytic function h such that g(z) =
eh(z) in V .

Proof. To find such an h, we simply choose a branch of the log function that does
not have g(z0 ) on its cut line. Then the set on which this branch of the log function
is analytic is an open set W which contains g(z0 ). If we set V = g −1 (W ) and define
h on V by
h(z) = log(g(z)),
then V is a neighborhood of z0 and g(z) = eh(z) on V . 

If we combine this result with Theorem 3.4.2, we have:

Theorem 3.4.7. Let f be an analytic function on an open set U and let z0 be
a point of U such that f is not identically zero in a neighborhood of z0 . Then
there exist a non-negative integer k, a neighborhood V ⊂ U of z0 , and an analytic
function h on V , such that
f (z) = (z − z0 )k eh(z)
for all z ∈ V .

Isolated Singularities. If U is an open set, z0 ∈ U , and f is a function which is

analytic on U \ {z0 }, then f is said to have an isolated singularity at z0 . If f can be
given a value at z0 which makes it analytic everywhere on U , then the singularity
is said to be removable.
Theorem 3.4.8. If f has an isolated singularity at z0 and is bounded in some
deleted neighborhood of z0 , then z0 is a removable singularity of f .
98 3. Power Series Expansions

Proof. Suppose f is analytic and bounded on U \ {z0 }. If we define a function g

by g(z) = (z − z0 )2 f (z) for z = z0 and g(z0 ) = 0, then g is differentiable at z0 . In
fact,
g(z) − g(z0 )
g  (z0 ) = lim = lim (z − z0 )f (z) = 0.
z→z0 z − z0 z→z0
That this limit is 0 is proved as follows. Let M be a bound on |f (z)| on U \ {z0 }.
Then |(z − z0 )f (z)| ≤ M |z − z0 | on U . Since limz→z0 M |z − z0 | = 0, it follows that
limz→z0 (z − z0 )f (z) = 0 as well.
The function g is differentiable at every point of U \ {z0 } and, by the above, is
also differentiable at z0 . Thus, it is analytic on all of U . Since g(z0 ) = g  (z0 ) = 0,
the first two terms of its power series expansion about z0 are 0. It follows that
we may factor (z − z0 )2 out of every term of its power series expansion and write
g(z) = (z − z0 )2 h(z) for an analytic function h defined by a power series in a disc
centered at z0 . Clearly h and f are the same in this disc, except at z0 , where f is
not defined. Thus, setting f (z0 ) = h(z0 ) serves to define f at z0 in such a way that
it becomes analytic on all of U . 
cos z − 1
Example 3.4.9. Show that f (z) = has a removable singularity at 0.
z2
Solution: This follows immediately from the factorization of cos z −1 obtained
in Example 3.4.5. If f is given the value −1/2 at z = 0, it becomes analytic on the
entire plane.

There are two types of isolated singularities that are not removable. A function
f defined on U \ {z0 } of the form
g(z)
f (z) = ,
(z − z0 )k
where g is analytic on U , g(z0 ) = 0, and k is a positive integer, is said to have a
pole of order k. If the order of the pole is 1, then it is called a simple pole. An
isolated singularity which is not a pole and is not a removable singularity is called
an essential singularity.
Example 3.4.10. If U is an open set and z0 ∈ U , then analyze the singularities of
a function of the form f /g, where f and g are analytic on U .
Solution: The singularities of f /g in U are all isolated because the zeroes of
g are isolated. If we factor f and g as in Theorem 3.4.2, then
f (z) = (z − z0 )j p(z) and g(z) = (z − z0 )k h(z),
where j and k are the orders of the zeroes of f and g at z0 and p and h are
analytic in U and non-vanishing in some neighborhood of z0 . Then f (z)/g(z) =
(z − z0 )j−k p(z)/h(z) with p(z)/h(z) analytic and non-vanishing in a neighborhood
of z0 . The point z0 is a removable singularity for f /g if j ≥ k and, otherwise, is a
pole of order k − j.
1
Example 3.4.11. Analyze the singularities of the function f (z) = .
1 − ez
Solution: The denominator of this fraction has zeroes at the points {2πki}
for k an integer. Each of these is a zero of order 1 because the derivative of 1 − ez
3.4. Zeroes and Singularities 99

is − ez and this is non-zero for every z and, in particular, is non-zero at the points
{2πki}. It follows from the preceding example that each of these points is a simple
pole for f .

Essential singularities are quite wild. In fact, the Big Picard Theorem states
that a function with an essential singularity at z0 takes on every complex number
but one as a value in every open disc centered at z0 . We will prove the Big Picard
Theorem and its little brother – the Little Picard Theorem – in Chapter 7. The
following is a very much weaker statement than the Big Picard Theorem, but it is
still enough to show that an analytic function behaves very wildly near an essential
singularity.
Theorem 3.4.12. If f is analytic in U \{z0 } and has an essential singularity at z0 ,
then for every open disc D, centered at z0 and contained in U , the set f (D \ {z0 })
has closure equal to the entire complex plane.

Proof. Suppose there is a disc D, centered at z0 and contained in U and there is

some complex number w which is not in the closure of f (D \ {0}). Then there is
an r > 0 such that f (D \ {z0 }) ∩ Dr (w) = ∅. This means |f (z) − w| ≥ r for every
z ∈ D \ {z0 }. Then
 
 1  1
 
 f (z) − w  ≤ r for all z ∈ D \ {z0 }.
Thus, 1/(f (z)−w) is analytic and bounded on D \{z0 } and, hence, has a removable
singularity at z0 . In other words, there is an analytic function h on D such that
1
= h(z)
f (z) − w
on D \ {z0 }. If h has no zero at z0 , set k = 0; otherwise, let k be the order of the
zero of h at 0. Then
h(z) = (z − z0 )k g(z)
for some analytic function g on D which does not have a zero at z0 . Solving for f ,
we find
1 w(z − z0 )k + 1/g(z)
f (z) = w + =
h(z) (z − z0 )k
in some disc containing z0 where g is non-vanishing. Since the numerator is analytic
in this disc, f has a pole of order k at z0 (or a removable singularity if k = 0).
Since neither of these things is true, we conclude that there is no w which fails to
be in the closure of f (D). This completes the proof. 

The following theorem gives an easy way to identify whether a given isolated
singularity is removable, a pole, or essential.
Theorem 3.4.13. Let f be an analytic function with an isolated singularity at z0 .
Then
(a) f has a removable singularity at z0 if and only if limz→z0 f (z) exists and is
finite;
(b) f has a pole at z0 if and only if limz→z0 f (z) = ∞;
(c) f has an essential singularity at z0 if and only if limz→z0 f (z) does not exist,
even as an infinite limit.
100 3. Power Series Expansions

Proof. The function f has a removable singularity at z0 if and only if it can be

given a value w at z0 which makes it analytic in a neighborhood of z0 . In this case,
limz→z0 f (z) = w, since f with its new value at z0 must be continuous at z0 .
On the other hand, if f has a finite limit as z → z0 , then f is bounded in a
neighborhood of z0 and, by Theorem 3.4.8, the singularity at z0 is removable. This
proves (a).
g(z)
If f has a pole at z0 , then f (z) = for some k > 0 and some g
(x − z0 )k
analytic in a neighborhood of z0 with g(z0 ) = 0. Then limz→z0 g(z) = g(z0 ) = 0
and limz→z0 (z − z0 )k = 0. It follows that limz→z0 f (z) = ∞.
On the other hand, if limz→z0 f (z) = ∞ the singularity is not removable since
f does not have a finite limit as z → z0 . It cannot be essential either, since the
previous theorem says that, if it were essential, f would take on values arbitrarily
close to any given complex number in every open disc centered at z0 . Thus, f could
not have limit ∞ at z0 if the singularity were essential. This proves (b).
If limz→z0 f (z) does not exist, even as a finite limit, then by (a) and (b) f does
not have a removable singularity or a pole. Thus, it has an essential singularity.
Conversely, if f has an essential singularity at z0 , then it does not have a removable
singularity or a pole. Then, also by (a) and (b), f does not have a finite limit at
z0 nor does it have limit ∞. This proves (c). 

Example 3.4.14. Analyze the singularity of the function f (z) = e1/z at z = 0.

Solution: This is an essential singularity. The function f (z) = e1/z takes on
the value 1 at all points of the form z = (2πn)−1 and the value en at all points of
the form z = n−1 . Thus, f (z) approaches 1 as z approaches 0 along one sequence
of points and it approaches ∞ as z approaches 0 along another sequence of points.
Thus, f certainly does not have a limit, finite or infinite, as z → 0.

Meromorphic Functions. For many reasons, it is important to study functions

which are analytic on an open set U except on a subset of U consisting of points
where poles occur. Necessarily, such a subset is a discrete subset of U .

Definition 3.4.15. Let U be an open set and E a discrete subset of U . If f is a

function which is analytic on U \ E and has a removable singularity or a pole at
each point of E, then f is called a meromorphic function on U .

One reason the set of meromorphic functions on U is interesting is that it is a

field, if U is a connected open set. Obviously, we can add and multiply meromorphic
functions and the results are still meromorphic. Just as obviously, the appropriate
commutative, associative and distributive laws hold and there are zero and identity
elements and additive inverses. All of these things are also true of the class of
analytic functions on U . However, in the class of meromorphic functions we also
have the last field axiom satisfied: every non-zero element has a multiplicative
inverse.

Theorem 3.4.16. If U is a connected open set and f is a meromorphic function

on U which is not identically zero, then 1/f is also meromorphic.
3.4. Zeroes and Singularities 101

Proof. By Theorem 3.4.2 the set Z(f ) of zeroes of f is a discrete subset of U . By

definition, the set P (f ) of poles of f is a discrete subset of U . It follows that the
set E = Z(f ) ∪ P (f ) is also a discrete subset of U . The function 1/f is analytic on
U \ E and so the only thing we need to establish is that at points of E it has only
removable singularities or poles.
Let z0 be a point of E. Since, E is discrete, there is a disc D, centered at z0
in which z0 is the only point of E. If z0 ∈ E, then f is analytic everwhere in D
except at z0 where is has either a zero or a pole.
If f has a zero of order k at z0 , then it factors as (z − z0 )k g(z) where g is
analytic and non-vanishing in D. Obviously then 1/f (z) = (z − z0 )−k /g(z) has a
pole of order k at z0 .
If f has a pole of order k at z0 , then f factors as f (z) = (z − z0 )−k h(z) where
h is analytic and non-vanishing in D. Then 1/f (z) = (z − z0 )k /h(z) has a zero of
order k at z0 and, hence, a removable singularity at z0 . 

Exercise Set 3.4

1. Prove that a set E is a discrete subset of an open set U if and only if no

sequence of distinct points of E converges to a point in U .
2. Prove Theorem 3.4.4.
3. Is there a function, analytic on C, which is 0 on the set of points {1/n} for
n a positive integer and not identically 0? Justify your answer. What if the
function is only required to be analytic on C \ {0}?
4. If f (z) = sin z − z, find the order of the zero of f at 0. Then give the factor-
ization of f at 0, as in Theorem 3.4.1.
5. If f (z) = cos z − 1 + z 2 /2, find the order of the zero and the factorization at 0
as in the previous exercise.
6. Give an example to show that Theorem 3.4.2 no longer holds if we drop the
hypothesis that U is connected.
7. Prove that if f is an analytic function on a connected open set U and if K is
a compact subset of U , then there can be at most finitely many zeroes of f in
K.
8. Show that if f is an analytic function with a zero of order k at z0 , then there
is a neighborhood V of z0 and an analytic function g on V such that f = g k
on V and g  (z0 ) = 0.
9. Suppose f and g are analytic functions on an open set U , z0 ∈ U , and f (z0 ) =
g(z0 ) = 0. Show that

f (z) f  (z)
lim = lim 
z→z0 g(z) z→z0 g (z)
and that this limit exists if ∞ is allowed as a possible value.
10. Suppose U is a connected open set and z0 ∈ U . Prove that if f is a non-
constant analytic function on U \ {z0 } and f takes on a certain value c at least
once in every deleted neighborhood of z0 , then f has an essential singularity
at z0 .
102 3. Power Series Expansions

11. Prove that if f is a function which is analytic in the exterior of the closed disc
Dr (0) and if limz→∞ f (z) = 0, then f has a power series expansion of the form
∞

f (z) = an z −n ,
n=1

which converges on the set {z ∈ C : |z| > r}. Hint: Consider the function
g(z) = f (1/z).

In the next five problems, analyze each singularity z0 of f . Is it removable, a

pole, or essential? If it is a pole, what is its order? If it is removable, what
value should you give the function at z0 to make it analytic?

1
12. f (z) = .
z − z3
13. f (z) = sin 1/z.
ez −1 − z
14. f (z) = .
z2
1 1
15. f (z) = z − .
e −1 z
log z
16. f (z) = , where log is the principal branch of the log function.
(1 − z)2

3.5. The Maximum Modulus Principle

This is another important application of Cauchy’s theorems. Before stating it, we
state and prove a technical lemma that will be used in its proof. The lemma states
that if the average value of a continuous function on an interval is as large, in
modulus, as all values of the function on the interval, then the function must be
constant.
Lemma 3.5.1. Let f be a continuous complex-valued function defined on an in-
terval I = [a, b] on the real line. If |f (t)| ≤ M on I, where
 
 1  b 
 
(3.5.1)  f (t) dt = M,
b − a a 

then f is a constant of modulus M on I.

Proof. If we choose a complex number u of modulus 1 such that

 b  
 b 
 
u f (t) dt =  f (t) dt ,
a  a 

then (3.5.1) may be written as

 b
(3.5.2) (M − uf (t)) dt = 0.
a
3.5. The Maximum Modulus Principle 103

Let uf = g + ih, where g and h are real-valued continuous functions on I. Note

that |f (t)| ≤ M implies that g(t) ≤ M and, hence, that M − g(t) ≥ 0. Equation
(3.5.2) implies
 b
(M − g(t)) dt = 0.
x a
However, a (M − g(t)) dt is a differentiable function of x with non-negative deriva-
tive M − g(x) on I and, hence, is a non-decreasing function. Since it is 0 at a and b,
it must be identically zero. This implies that its derivative M − g(x) is identically
0.
We now have that the real part of uf = g + ih is the constant M . However, we
also have that
M 2 ≥ |f (t)|2 = |uf (t)|2 = g 2 (t) + h2 (t) = M 2 + h2 (t),
and this implies that h(t) ≡ 0 on I. Thus, f is the constant u−1 M , which has
modulus M . 

The Maximum Modulus Theorem.

Theorem 3.5.2. If f is analytic on a connected open set U and |f | has a local
maximum at z0 ∈ U , then f is constant on U .

Proof. If f has a local maximum at z0 ∈ U , then we may choose an r > 0 such

that Dr (z0 ) ⊂ U and such that |f (z0 )| is a maximum for |f (z)| on Dr (z0 ). Then
Cauchy’s Integral Theorem tells us that

1 f (z)
f (z0 ) = dz
2πi |z−z0 |=r z − z0
(3.5.3)  2π
1
= f (z0 + r eit ) dt
2π 0
Since |f (z0 + r eit )| ≤ |f (z0 )| for each t, by our choice of r, we have the hypotheses
of the previous lemma satisfied with M = |f (z0 )|, [a, b] = [0, 2π], and f (z0 + r eit )
playing the role of the function f of the lemma. Based on the lemma, we conclude
that f is a constant c on the circle z0 + r eit . Since this circle is a non-discrete
subset of U , it follows from the Identity Theorem (Theorem 3.4.4) that f is the
constant c on all of U . 

Recall that a subset of the plane is compact if it is both closed and bounded.
Suppose U is an open set which is bounded. Then its closure U is both closed and
bounded and, hence, is compact. Suppose U is connected. If f is a continuous
complex-valued function on U , then the continuous real valued function |f (z)| has
a maximum value on U . If f is also analytic on U and non-constant, then the
previous theorem implies that this maximum cannot occur at a point of U . This
means it must occur only on the boundary ∂U . This proves the following corollary
of Theorem 3.5.2.
Corollary 3.5.3. Suppose U is a connected, bounded, open subset of C. If f is
a function which is continuous on U , analytic on U , and non-constant, then the
maximum value of |f (z)| on U is attained on ∂U and nowhere else.
104 3. Power Series Expansions

The typical example of a set U of the type described in the above corollary is
an open disc D r (z0 ) of finite radius.
Example 3.5.4. Find where the function f (z) = z 2 − z attains its maximum
modulus on the closed unit disc.
Solution: By Corollary 3.5.3, the maximum occurs only on the unit circle
{z = eit : t ∈ [0, 2π]}. Thus, we need to find where the maximum modulus of
the function |f (eit )| occurs for t ∈ [0, 2π]. This is equivalent to finding where the
square of the function has a maximum. Thus, we wish to maximize the function
h(t) = | e2it − eit |2 = | eit −1|2 = 2 − 2 cos t.
Clearly, the maximum occurs at t = π and only there. Thus, the maximum of
|z 2 − z| on the closed unit disc is 2 and it occurs only at z = −1.

Schwarz’s Lemma. Schwarz’s Lemma is a nice application of the Maximum

Modulus Theorem. It will be quite useful later in the theory of conformal mappings.
Lemma 3.5.5 (Schwarz’s Lemma). Let f be analytic on D1 (0), with f (0) = 0 and
|f (z)| ≤ 1 for every z ∈ D1 (0). Then,
(3.5.4) |f (z)| ≤ |z| for all z ∈ D1 (0)
and |f  (0)| ≤ 1. If |f  (0)| = 1, then there is a constant c of modulus one such that
f (z) = cz.

Proof. Since f (0) = 0, Theorem 3.4.2 implies that f (z) = zg(z), where g is also
analytic in D1 (0). Since |f (z)| ≤ 1 on D1 (0), it follows that
1
|g(z)| ≤ on the circle |z| = r
r
for each r < 1. The Maximum Modulus Theorem implies that this inequality also
holds inside the disc of radius r. Since this is true of each r < 1, we conclude that
|g(z)| ≤ 1 on all of D1 (0) and this implies (3.5.4).
The inequality |f  (0)| ≤ 1 follows from (3.5.4), since
f (z)
f  (0) = lim = lim g(z) = g(0).
z→0 z z→0

If |f  (0)| = 1, then |g(0)| = 1 and this is the maximum value of g on D1 (0).

The Maximum Modulus Theorem says this cannot happen unless g is a constant c.
Then f (z) = cz. 

We now give a simple application of Schwarz’s Lemma, which is a precursor

of its later use in conformal mapping theory. Let U and V be open sets in C. A
bi-analytic map from U to V is an analytic function f : U → V with an analytic
inverse f −1 : V → U . That is, f −1 ◦ f (z) = z for every z ∈ U and f ◦ f −1 (w) = w
for every w ∈ V .
Theorem 3.5.6. The only bi-analytic maps from the unit disc to itself that take
0 to 0 are of the form f (z) = cz for a constant c of modulus 1. That is, the only
bi-analytic maps of the unit disc onto itself which fix 0 are the rotations.
3.5. The Maximum Modulus Principle 105

Proof. If f : D1 (0) → D1 (0) is bi-analytic with inverse f −1 , then both f and f −1

satisfy the hypotheses of Schwarz’s Lemma. Thus,
(3.5.5) |f  (0)| ≤ 1 and |(f −1 ) (0)| ≤ 1.
However, it follows from the chain rule, applied to the composition f −1 ◦ f (z) = z,
that
1
(f −1 ) (0) =  .
f (0)
Combining this with (3.5.5), we conclude that |f  (0)| = 1. By Schwarz’e Lemma,
f (z) = cz for some constant c of modulus 1. 

Harmonic Functions. Theorems about analytic functions, like Cauchy’s For-

mula and the Maximum Modulus Theorem, imply things about harmonic functions.
This is due to the fact that, locally at least, each harmonic function is the real part
of an analytic function:
Theorem 3.5.7. Let u be a function which is of class C2 and harmonic on a convex
open set U . Then u has a harmonic conjugate v on U . That is, there is a harmonic
function v on U such that the function f = u + iv is analytic on U .

Proof. Consider the function g = ux − iuy . It is C1 , therefore differentiable, and

satisfies the Cauchy-Riemann equations, since
uxx = −uyy and uxy = uyx .
Thus, g is analytic in U . Since U is convex, Theorem 2.5.6 implies that g has an
antiderivative h in U . This means h is analytic in U and h = g. If h = w + iv,
with w and v real, then
ux − iuy = g = h = wx + ivx = wx − iwy .
On equating real and imaginary parts in this equation, we conclude that
ux = w x ,
uy = w y .
It follows that w = u + c for some real constant c. Thus, f = h − c = u + iv is an
analytic function in U with real part u. 

This theorem leads to two important results about harmonic functions: a max-
imum principle for harmonic functions and an integral formula.
Theorem 3.5.8. If u is a harmonic function on a connected open set U and u has
a local maximum at some point z0 ∈ U , then u is constant on U .

Proof. Let V be a convex neighborhood of z0 with V ⊂ U . The previous theorem

implies that u has a harmonic conjugate v on V . Then, f = u + iv is analytic on
V , as is the function
g(z) = ef (z) .
Since |g(z)| = eu(z) , if u has a local maximum at z0 , then so does |g(z)|. By
the Maximum Modulus Theorem, this implies that g is constant on V . But if g
is constant on the connected open set V , then f is also constant on V (Exercise
3.5.10). Hence u is constant on V .
106 3. Power Series Expansions

The completion of the proof involves showing that if a harmonic function on

a connected open set U is constant on a non-empty open subset of U , then it is
constant on all of U . We leave this as an exercise (Exercise 3.5.12). 

The next theorem shows that a harmonic function u has the mean value prop-
erty: the value of u at a point is equal to its mean value over any circle centered at
the point.
Theorem 3.5.9. If u is harmonic on an open set U and Dr (z0 ) ⊂ U , then
 2π
1
(3.5.6) u(z0 ) = u(z0 + r eit ) dt.
2π 0

Proof. Choose R > r such that DR (z0 ) ⊂ U . Then Theorem 3.5.7 implies that u
is the real part of an analytic function f on DR (z0 ). The Cauchy Integral Formula
applied to f and the path γ(t) = z0 + r eit yields
 2π
1
f (z) = f (z0 + r eit ) dt.
2π 0
Equation (3.5.6) follows from this by equating real parts. 
Example 3.5.10. Find a harmonic congugate for the harmonic function x2 − y 2 .
Solution: We recognize x2 − y 2 as the real part of the analytic function z 2 =
(x + iy)2 . The imaginary part of this function is 2xy and so 2xy is a harmonic
conjugate of x2 − y 2 .
Example 3.5.11. Prove that
x
u(x, y) =
x2 + y 2
is harmonic on C \ {0} and find a harmonic conjugate for it.
Solution: We simply observe that u(x, y) is the real part of
x − iy 1
= .
x2 + y 2 z
Therefore, u is harmonic and has
−y
x2 + y 2
as a harmonic conjugate.

Exercise Set 3.5

1. Find where the function z 2 − 1 attains its maximum modulus on the closed
unit disc.
2. Find where the function | ez | attains its maximum value on the closed unit
disc.
3. Find where the function (z − 1)2 attains its maximum modulus on the triangle
with vertices at 0, 1 + i, 1 − i.
3.5. The Maximum Modulus Principle 107

4. Show that if f is a non-constant analytic function on a connected open set U

and if f has no zeroes on U , then there are no points of U where |f (z)| has a
local minimum.
5. Show that if f is a non-constant, continuous function on D1 (0), which is an-
alytic on D1 (0) and |f (z)| = 1 for all z on the unit circle, then f has a zero
somewhere in D1 (0).
6. Prove that if f is a non-constant entire function, and c > 0 is a constant such
that the closed set K = {z ∈ C : |f (z)| ≤ c} is bounded, then the open set
U = {z ∈ C : |f (z)| < c} contains at least one zero of f .
7. Supppose f is an analytic function on the unit disc D1 (0). Prove that if
|f (z)| ≤ 1 on D1 (0) and f has a zero of order 2 at 0, then |f (z)| ≤ |z|2 for all
z ∈ D1 (0).
8. If f is a harmonic function u on a connected open set U , prove that any two
harmonic conjugates for u must differ by a constant.
9. Prove that if U is a connected open set with compact closure U , and u is a
continuous function on U which is harmonic on U , then u attains its maximum
and minimum values on ∂U . Also, prove that if u attains either its maximum
or its minimum at a point of U , then it is constant.
10. Prove that if f is a continuous function on a connected open set V and if ef (z)
is constant on V , then f is also constant on V .
2z − 1
11. Show that f (z) = is a bi-analytic map from the unit disc onto itself
z−2
which takes 0 to 1/2. Hint: Show that |f (z)| = 1 on the unit circle and
conclude from this that f maps the unit disc into itself. Then show it has an
inverse function which also maps the unit disc into itself by directly solving
the equation w = f (z) for z as a function of w.
12. Prove that if a harmonic function on a connected open set U is constant on a
non-empty open subset of U , then it is constant on all of U .
13. Find a harmonic conjugate for u(x, y) = ex cos y on C.
14. Find a harmonic conjugate for u(x, y) = 1/2 log(x2 + y 2 ) on the complement
of the non-positive real axis in C.
15. Give an example of an open subset U of C and a harmonic function on U
which has no harmonic conjugate on U .
 Chapter 4

The General Cauchy

Theorems

In this chapter we extend Cauchy’s theorems in two ways: (1) we remove the
condition that the open set U be convex; and (2) we replace the path γ by a more
general type of object called a cycle.
In making these improvements, we finally come to grips with an issue which
we have avoided so far. This issue concerns the topology of the plane. Specifically,
there is a phenomenon that occurs in the plane which does not occur in the line.
On the line, every connected open set is an open interval, which certainly has no
gaps or holes. However, in the plane, there are connected open sets which do have
holes. An example is an open annulus of the form

A = {z ∈ C : r < |z| < R},

with 0 ≤ r < R. The integral of an analytic function on A around a closed path in

A may fail to be zero. In fact, the integral of 1/z around any positively oriented
circle in A, centered at 0, is 2πi, not 0, even though 1/z is analytic on A. Thus,
Cauchy’s Integral Theorem, as stated in Theorem 2.6.2, is not valid if we simply
drop the condition that U be convex. The problem here is that there is a hole in
the domain A on which f is analytic (the missing disc of radius r) and the path
goes around this hole. In our final version of Cauchy’s Theorem we will avoid this
problem, not with a hypothesis on the shape of U , but rather with a hypothesis that
ensures that the path γ does not go around any holes in U . We do this by adding
the hypothesis that the path γ has index 0 around every point in the complement
of U .
We begin by introducing a compact and efficient notation for expressing an
otherwise complicated sum of integrals over a number of different paths. This is
the language of chains and cycles, as used in the study of homology in algebraic
topology. Using this language will allow for relatively simple statements and proofs
of theorems that otherwise could be, at least notationally, quite complicated.

109
110 4. The General Cauchy Theorems

4.1. Chains and Cycles

In many calculations involving path integrals of an analytic function, we may in-
tegrate the same function over several paths and add the results, or add some of
the results and subtract others. We may integrate over the same path a number of
times. We may break a single path up into a collection of several paths, integrate
over each and then add the results. We may replace a path by one which simply
reverses the direction of travel, resulting in an integral which is the negative of the
original. We have already seen examples of some of these things (Examples 2.6.4
and 2.3.4 come to mind). The language of chains and cycles, introduced below, will
provide us with a compact and efficient way to deal with these kinds of integrals.
In this section, all paths will have parameter interval [0, 1].

1-Chains. An abelian group is a set X with a single operation (x, y) → x + y :

X × X → X which is associative and commutative, has an identity (denoted by 0)
and in which every element x has an inverse (denoted by −x). So an abelian group
satisfies the addition axioms of a vector space, but it is not assumed that there is
also a scalar multiplication operation.
Let U be an open subset of C. We will construct an abelian group called the
group of 1-chains in U from the set of all paths γ : [0, 1] → U as follows: we define
a 1-chain in U to be a finite formal linear combination, with integral coefficients,

p
Γ= m j γj ,
j=1

of distinct paths γj in U . We agree that a chain remains the same if we either

throw out or add in some summands mj γj with mj = 0.
The operation in our group of chains is addition. We add two chains in the
obvious way: if Γ and Λ are chains and γ1 , γ2 , · · · , γn are the paths that occur in
summands of either Γ or Λ, then

p 
p
Γ= n j γj and Λ = m j γj
j=1 j=1

for coefficients {nj } and {mj } some of which may be zero. Then we define

p
Γ+Λ= (nj + mj )γj .
j=1

The operation of addition, so defined, is clearly associative and commutative and

so the set of 1-chains forms an abelian group.
In group theory, the group of chains, as defined above, is known as the free
abelian group generated by the set of paths in U .
A path itself is a 1-chain (a 1-chain where there is only one summand and its
coefficient is 1). However, often it is useful to break up a path into pieces that form
the summands of a 1-chain. The original path and the 1-chain formed by its pieces
are equivalent in a sense that will be discussed later in the section.
4.1. Chains and Cycles 111

Example 4.1.1. Show how to use a path in U , in the sense of the previous chapter,
to produce a 1-chain in U which is a linear combination of smooth paths.
Solution: Suppose γ : [a, b] → U is a closed path. Let
a = t0 < t1 < · · · < tn = b
be a partition such that γ is smooth on each subinterval. Let zj = γ(tj ) for
j = 1, · · · , n. For j = 1, · · · , n let γj be a reparameterization of the restriction of
γ to [tj−1 , tj ] which changes the parameter interval to [0, 1]. If we define a chain Γ
to be the formal sum
n
Γ= γj ,
j=1
then this is a chain consisting of smooth paths.

Integration Over Chains. We introduced 1-chains in order to integrate over

them. The definition of the integral is as follows:
Definition 4.1.2. Let

p
Γ= m j γj
j=1
be a 1-chain. With I = [0, 1], we set

Γ(I) = γj (I).
mj =0

If f is a continuous function on a set E ⊂ C with Γ(I) ⊂ E, then we define the

integral of f over Γ by
 
p 
(4.1.1) f (z) dz = mj f (z) dz.
Γ j=1 γj

It is an immediate consequence of the definition of the integral over a 1-chain,

that if Γ and Λ are two 1-chains and f is a function defined and continuous on a
set containing both Γ(I) and Λ(I), then
  
(4.1.2) f (z) dz = f (z) dz + f (z) dz.
Γ+Λ Γ Λ

Definition 4.1.3. Suppose Γ and Λ are chains with Γ(I) and Λ(I) both subsets
of a set E ⊂ C. We will say that Γ and Λ are E-equivalent if
 
f (z) dz = f (z) dz
Γ Λ
for every continuous function f on E. If Γ(I) = Λ(I) = E and Γ and Λ are
E-equivalent, then will will simply say that Γ and Λ are equivalent.
Example 4.1.4. Let Δ be the triangle with vertices {a, b, c} listed in counterclock-
wise order. Recall that ∂Δ denotes the path which traverses the boundary of Δ
once in the counterclockwise direction. Show that ∂Δ and Γ = [a, b] + [b, c] + [c, a]
are equivalent, as are ∂Δ and Λ = [a, b] + [b, c] − [a, c].
Solution: Clearly ∂Δ(I) = Γ(I) = Λ(I) – each of them is just the boundary of
the triangle Δ. Also, it follows from Theorem 2.4.6, Parts (b) and (c), that ∂Δ, Γ,
112 4. The General Cauchy Theorems

and Λ all determine the same integral for continuous functions on the boundary of
Δ. Thus, by definition, the three chains are equivalent.

0-Chains. Just as the group of 1-chains is the free abelian group generated
by paths, the group of 0-chains in U is the free abelian group generated by the
singleton subsets of C (sets {z} containing a single point z ∈ C) – that is, it is the
group of formal linear combinations,


p
mj {zj },
j=1

with integral coefficients, of distinct singleton subsets of U . Again, linear combi-

nations which differ only by summands with 0’s as coefficients are all identified.
Addition is defined as before, and the result is another abelian group.
A path γ determines both a 1-chain (the chain with γ as its only summand and
a 0-chain {γ(1)} − {γ(0)}. This 0-chain is called the boundary of γ. Notice that
it is the 0 element of the group of 0-chains if and only if {γ(1)} = {γ(0)} – that
is, if and only if γ is a closed path. Both the boundary of a path in this sense and
the concept of a closed path generalize to the context of chains. The result is as
follows:

The Boundary Map and Cycles.

Definition 4.1.5. If U is an open subset of C, we define the boundary map ∂ from

1-chains in U to 0-chains in U by

⎛ ⎞

p 
p
∂⎝ m j γj ⎠ = (mj {γj (1)} − mj {γj (0)}).
j=1 j=1

This may not result in a linear combination of distinct singleton sets, but
we fix this, combining terms involving the same singleton subset by adding their
coefficients. In order for ∂ to determine a well defined map from 1-chains to 0-
chains, one must show that it preserves the identifications made in our definition
of 1-chains. This simply requires that any summand in a 1-chain that has zero
as coefficient is sent by ∂ to a 0-chain with all zeroes as coefficients, and this is
obviously true.
Once we know that ∂ is well defined, it is quite evident from the definition that it
is a group homomorphism ∂ from 1-chains to 0-chains (meaning ∂(Γ+Λ) = ∂Γ+∂Λ).
This ensures that ker ∂ = {Γ : ∂Γ = 0} is closed under addition and taking additive
inverses and, hence, is a subgroup of the set of 1-chains in U . Elements of this
4.1. Chains and Cycles 113

Figure 4.1.1. A Cycle which is a Sum of Closed Paths.

subgroup are called cycles in U – that is:

Definition 4.1.6. A 1-chain Γ in U is called a 1-cycle if ∂Γ = 0.

As we shall see below, a cycle is a chain whose summands can be fit together

to form a series of loops (closed paths).
Given a chain Γ, there are always many other chains which are equivalent to it.
This is due to the fact that, as in the previous example, paths in a chain may be
partitioned into sums of subpaths without changing the integral, and several paths
may be joined to form a single path if their endpoints correspond in the right way.
This is what is going on in the following theorem.
Theorem 4.1.7. If Γ is a 1-cycle, then there is a 1-cycle Λ which is equivalent to
Γ and which is a sum of closed paths.

Proof. We will make a succession of changes to Γ, none of which will change Γ(I)
or the integral (4.1.1).
We first express Γ as a linear combination of paths in which the coefficients are
all 1 or −1. That is, each term mj γj is replaced by a sum of mj copies of γj if mj
is positive and by a sum of −mj copies of (−1)γj if mj is negative. This does not
change the chain Γ.
The next step is to replace each summand of the form (−1)γj with the path
−γj , where we recall that −γj (t) = γj (1 − t), so that −γj is γj traversed in the
reverse direction. Theorem 2.4.6, Part (c), implies that this change results in a
cycle Γ̃ which is equivalent to Γ.
At this point we have replaced Γ with a cycle Γ̃ which is a simple sum of some
number of paths. Say the number is n. We next show that if these paths are not
all closed paths, then we can replace Γ̃ with a sum of fewer than n paths without
changing the integral.
Suppose γj is a path in Γ̃ which is not closed. Then γj (0) = γj (1). Since
∂ Γ̃ = 0, γj (1) must be equal to γk (0) for some path γk in Γ̃; otherwise there would
be no term to cancel with {γj (1)} in the expression for ∂Γ. We may join γk and
γj , as in (2.4.2), to form a new path which begins at γj (0) and ends at γk (1). By
Theorem 2.4.6, Part (b), replacing γj and γk in Γ̃ by this new path does not change
the integral. Thus, if its summands are not all closed paths, we may replace Γ̃ with
114 4. The General Cauchy Theorems

a path with fewer summands, which determines the same integral. This forms the
basis for an induction argument, which is carried out below.
If a cycle consists of a single path γ, then the condition
∂γ = {γ(1)} − {γ(0)} = 0
implies γ is a closed path. Suppose we know that any cycle which can be written
as a sum of fewer than n paths may be replaced by an equivalent cycle which is a
sum of closed paths. Then the argument of the previous paragraph shows that any
cycle which can be written as a sum of n paths is equivalent to some cycle which
is a sum of closed paths. By induction, every cycle is equivalent to a sum of closed
paths. 

Index of a Cycle. The notion of index of a path around a point can be extended
to cycles as follows.
Definition 4.1.8. If Γ is a 1-cycle in C and z ∈ C is a point which does not lie on
Γ(I), then we set

1 1
IndΓ (z) = dw.
2πi Γ w − z
The number IndΓ (z) is called the index of Γ around z.

By Theorem 4.1.7 the cycle Γ may be replaced by one which is a sum of closed
paths without changing Γ(I) or the integral defining IndΓ (z). Thus, IndΓ (z) is a
finite sum of numbers of the form Indγ (z), where γ is a closed path. This, together
with Theorems 2.6.6 and 2.7.5 proves the first three parts of the following theorem.
The last part follows immediately from (4.1.2).
Theorem 4.1.9. Let Γ be a 1-cycle in C. Then
(a) IndΓ is an integer-valued function defined on the complement of Γ(I);
(b) IndΓ is constant on each component of C \ Γ(I);
(c) IndΓ is 0 on the unbounded component of C \ Γ(I);
(d) if Λ is also a cycle, and z ∈
/ Γ(I) ∪ Λ(I), then
IndΓ+Λ (z) = IndΓ (z) + IndΛ (z).

Example 4.1.10. For each integer j, let

1 2πit
γj (t) = j + e , t ∈ [0, 1].
4
For a positive integer N set
γ(t) = (N + 1/2) e2πit , t ∈ [0, 1]
and

N
Γ1 = γj , Γ = γ − Γ1 .
j=−N

Find Indγ (z), IndΓ1 (z), and IndΓ (z) for each value of z which is an integer.
4.1. Chains and Cycles 115

... ...
... ...
γ−N γ−2 γ−1 γ0 γ1 γ 2 γN

Figure 4.1.2. The Paths Making up the Cycle Γ of Example 4.1.10.

Solution: The paths γ and γj are circles traversed once in the counterclockwise
direction. For such a path we know the index is 1 inside the circle and 0 outside
the circle. Thus, if k is an integer,

1, if |k| ≤ N ;
Indγ (k) =
0, if |k| > N ;
and 
1, if k = j;
Indγj (k) =
0  j.
if k =
Thus,


k
1, if |k| ≤ N ;
IndΓ1 (k) = Indγj (k) =
j=−k
0, if |k| > N ;
and so,
IndΓ (k) = Indγ (k) − IndΓ1 (k) = 0
for every integer k.

Homologous Cycles. In the general Cauchy Theorem of the next section, the
condition that U be convex is replaced by a condition on the path or cycle over which
the integration takes place. The condition is that the path or cycle be homologous
to 0 in the sense of the following definition.
Definition 4.1.11. Let U be an open subset of C and let Γ and Λ be 1-cycles in
U . We will say that Γ and Λ are homologous in U if
IndΓ (z) = IndΛ (z)
116 4. The General Cauchy Theorems

γ2 γ1

γ3 γ4

Figure 4.1.3. The Paths Making up the Cycle Γ of Exercise 4.1.6.

for every z in the complement of U . We will say that Γ is homologous to 0 in U if

IndΓ (z) = 0

for every z in the complement of U .

For a closed path γ, intuitively, the statement that γ is homologous to 0 in U

means that γ does not go around any points of the complement of U – that is, it
does not go around any “holes” in U . For cycles, the meaning is more complicated.
For example, the cycle Γ of Example 4.1.10 is homologous to 0 in the complement
of the integers in C.
Note that two 1-cycles Γ and Λ are homologous if and only if Γ−Λ is homologous
to 0.

Example 4.1.12. If γ1 (t) = e2πit and γ2 (t) = 2 e2πit for t ∈ [0, 1], show that
Γ = γ2 − γ1 is a 1-cycle which is homologous to 0 in U = {z ∈ C : 1/2 < |z| < 3},
but is not homologous to 0 in the set V = U \ {3/2}.
Solution: The curve γ1 has index 1 about points inside the unit disc D1 (0),
and index 0 about points exterior to the closed unit disc, while γ2 has index 1
about points in D2 (0) and 0 about points exterior to the closure of this disc. The
complement of U is
D1/2 (0) ∪ (C \ D3 (0)).
Since
Indγ1 = Indγ2 = 1 on D1/2 (0)
and
Indγ1 = Indγ2 = 0 on C \ D3 (0),
we conclude that Γ = γ2 − γ1 is homologous to 0 in U .
However, at the point z = 3/2, γ1 has index 0 and γ2 has index 1. Since 3/2 is
in the complement of V = U \ {3/2}, Γ is not homologous to 0 in V .
4.1. Chains and Cycles 117

Exercise Set 4.1

1. Let a, b, c be three points in C and let γ1 be a path beginning at a and ending

at b, γ2 a path beginning at a and ending at c, and γ3 a path beginning at b
and ending at c. Is γ1 + γ2 + γ3 a cycle? How about γ1 − γ2 + γ3 ?
2. Consider the following directed line segments: γ1 from −1 to −1 + i, γ2 from
1 to 1 + i, γ3 from −1 to 0, γ4 from 0 to 1, γ5 from −1 + i to i, γ6 from i to
i + 1, and γ7 from 0 to i. Show that Γ = γ1 + γ2 − γ3 + γ4 + γ5 − γ6 − 2γ7 is a
cycle.
3. Find a sum of closed paths which is equivalent to the cycle of the previous
exercise.
4. For t ∈ [0, 1], let γ1 (t) = 1 + e2πi(t+1/2) and γ2 (t) = −1 + e2πit . Find a single
closed path γ with parameter interval [0, 1], such that γ is equivalent to γ1 −γ2 .
5. Let f be a function which is analytic on the open unit disc D and γ1 and γ2
two closed  paths in D each of which has index 1 about 0. If Γ = γ1 − γ2 ,
compute Γ f (z)/z dz?
6. Let R ⊂ C be a rectangle and let γ be a path which traverses the boundary
of R once in the positive direction. Partition R into four subrectangles Rj ,
j = 1, 2, 3, 4, by joining each pair of opposite sides of R with a perpendicular
line. Let γj be a path which traverses the boundary of Rj once in the positive
4
direction. Let Γ = j=1 γj and E = Γ(I). Show that γ and Γ are E-equivalent
but not equivalent (see Definition 4.1.3).
7. For t ∈ I = [0, 1], let

γ1 (t) = 4 e2πit ,
γ2 (t) = i + e2πit ,
γ3 (t) = −i + e2πit .

If Γ = γ1 − γ2 − γ3 , then find IndΓ (z) for z in each of the components of

C \ Γ(I). Is Γ homologous to 0 in C \ {i, −i}?
8. Find IndΓ (z) for z in each component of C \ Γ(I) if Γ is the cycle of Exercise
4.1.2.
9. Find IndΓ (z) for z in each component of C \ Γ(I) if Γ = 2γ1 − γ2 − γ3 with
γ1 , γ2 , γ3 the positively oriented circles of radii 4, 2, and 2, centered at 0, −1,
and 1, respectively.
10. Let U be an open set. Prove that if Γ and Γ1 are cycles that are U -equivalent,
then they are homologous in U .
11. Let U be an open subset of C and S = {z1 , · · · , zn } a finite subset of U . Find
a cycle Γ which is homologous to 0 in U and which has index 1 about each
point of S.
12. If n is an integer, let γ(t) = e2nπit and γ1 (t) = 2 e2πit for t ∈ [0, 1]. Show that
the cycle γ − nγ1 is homologous to 0 in A = {z ∈ C : 0 < |z| < 3}.
13. Find a cycle Γ which is homologous to 0 in the annulus {z : 1 < |z| < 4}, but
has index 1 about every point in the annulus {z : 2 < |z| < 3}.
118 4. The General Cauchy Theorems

4.2. Cauchy’s Theorems

We will first prove the general version of Cauchy’s Formula and then use it to prove
the general version of Cauchy’s Theorem. This is the reverse of the order in which
we proved the analogous results on convex sets.

A Continuity Lemma. Let f be an analytic function on an open set U . In the

proof of the Cauchy Integral Formula for convex sets (Theorem 2.6.7) a key role
was played by the function
⎧
⎨ f (w) − f (z)
if w = z;
(4.2.1) g(z, w) = w−z
⎩f  (z) if w = z.
For each fixed z, this function is clearly a continuous function of w everywhere on
U , and is analytic on U except possibly at w = z. This was enough for our purposes
in the proof of Theorem 2.6.7. We will use the same function in the proof of our
more general Cauchy Integral Theorem, but we need to know that it satisfies a
stronger continuity condition. We need that it is continuous as a function of two
complex variables on
U × U = {(z, w) : z ∈ U, w ∈ U }.
This is equivalent to the condition that g be continuous on U × U as a function of
four real variables with values in R2 = C. It means simply that
(4.2.2) lim g(z, w) = g(z0 , w0 )
(z,w)→(z0 ,w0 )

for all (z0 , w0 ) ∈ U × U . Here, the limit has the usual -δ definition, using the
appropriate definition of the distance between (z, w) and (z0 , w0 ), which is

|(z, w) − (z0 , w0 )| = |z − z0 |2 + |w − w0 |2 .
This is the same as the Euclidean distance between these two points, considered
as points of R4 . It is easy to see that this type of limit satisfies all the usual limit
rules regarding sums, products, quotients, etc.
Lemma 4.2.1. If g(z, w) is defined by (4.2.1), then g is a continuous function of
two complex variables on U × U .

Proof. We must prove that (4.2.2) holds at every point (z0 , w0 ) ∈ U × U . If

z0 = w0 , then this follows from the continuity of f and the quotient rule for limits.
In the case where z0 = w0 , we have g(z0 , z0 ) = f  (z0 ). Thus, in order to prove
(4.2.2) we just need to prove
(4.2.3) lim g(z, w) = f  (z0 ).
(z,w)→(z0 ,z0 )

Since f  is continuous on U , given  > 0, we may choose δ > 0 such that

Dδ (z0 ) ⊂ U and
(4.2.4) |f  (z) − f  (z0 )| <  whenever |z − z0 | < δ.
4.2. Cauchy’s Theorems 119

If |(z, w) − (z0 , z0 )| < δ, then both z and w are in Dδ (z0 ), as is any point λ on the
line segment joining them. If z = w, integrating over this line segment yields
 w
f (w) − f (z) = f  (λ) dλ,
z
and so   
 1 w 
 
|g(z, w) − f (z0 )| =  (f (λ) − f (z0 )) dλ < .
 
w−z z
If z = w, then |g(z, w) − f  (z0 )| = |f  (z) − f  (z0 )| < . Hence, g is continuous at
(z0 , z0 ). This completes the proof. 

Cauchy’s Integral Formula.

Theorem 4.2.2 (Cauchy’s Integral Formula). Let U be any open subset of C, f a
function which is analytic on U , and Γ a 1-cycle in U which is homologous to 0 in
U . Then

1 f (w)
(4.2.5) IndΓ (z)f (z) = dw,
2πi Γ w − z
for every point z ∈ U which does not lie on Γ(I).

Proof. Let h be the function defined by


h(z) = g(z, w) dw,
Γ
where g is the function of the previous lemma. The fact that g is continuous on
U × U as a function of two complex variables implies that h is continuous on U .
If z ∈
/ Γ(I), then z = w for all w ∈ Γ(I) and so the integral defining h can be
broken apart as
  
f (w) f (z) f (w)
h(z) = dw − dw = dw − 2πi IndΓ (z)f (z).
Γ z−w Γ z−w Γ z−w
Thus, to prove the theorem, we simply need to prove that h is identically zero in
U . In the proof of Theorem 2.6.7, we proved h(z) = 0 using the Cauchy Theorem
in a convex set. Here, since U is not convex, and Γ may not be a single closed path,
we must proceed differently.
We will show that h(z) = 0 for all z ∈ U by showing that it is analytic on U
and then extending it to an entire function which is bounded and has limit 0 at
infinity. Then Liouville’s Theorem will imply that it is identically 0.
If Δ is a triangle contained in U , we have
  
h(z)dz = g(z, w) dwdz
(4.2.6) ∂Δ
∂Δ Γ
= g(z, w) dzdw.
Γ ∂Δ
That the order of integration can be reversed follows from the fact that, ultimately,
each iterated integral is a sum of iterated Riemann integrals, over the rectangle
I × I in R2 , of functions continuous on the rectangle. The fact that the order of
integration can be reversed in each of these integrals is then Fubini’s Theorem from
calculus.
120 4. The General Cauchy Theorems

For each fixed z ∈ U , g(z, w) is an analytic function of w ∈ U , except at w = z,

where it is at least continuous. Thus, by Theorem 2.5.8 the last integral in (4.2.6) is
0. Hence, the integral of h around each triangle in U is 0. It follows from Morera’s
Theorem that h is analytic in U .
Now let V be the set of all points z in the complement of Γ(I) for which
IndΓ (z) = 0. This is an open set, since it is a union of some of the components of
C \ Γ(I). It is also a set which contains the complement of U , by hypothesis. So
U ∪ V = C. If z ∈ V ∩ U , then

1 f (z)
dw = f (z) IndΓ (z) = 0.
2πi Γ z − w
Since z is not on Γ(I), we may break the integral defining h apart and write
  
f (w) f (z) f (w)
(4.2.7) h(z) = dw − dw = dw
Γ w − z Γ w − z Γ z −w
for z ∈ U ∩ V . This means we can extend the definition of h to all of C by defining
f (w)
it to be dw on V . That is, the original definition of h on U and this
Γ z −w
new definition of h on V agree on the overlap U ∩ V . The resulting function h is
analytic on all of C and, hence, is an entire function. On the unbounded component
of C \ Γ(I), h(z) is given by (4.2.7) and, from this, it is easy to see that
(4.2.8) lim h(z) = 0.
z→∞

We conclude that h is a bounded entire function and, hence, is a constant, by Li-

ouvilles’s Theorem. By (4.2.8) this constant must be zero. Thus, h(z) is identically
zero and this immediately implies (4.2.5) for z ∈ / Γ(I). 

1
Example 4.2.3. Find dz if Γ = γ2 − γ1 , where γ1 = (1/2) e2πit and
Γ z −z
2

γ2 = 2 e2πit for t ∈ I = [0, 1].

Solution: The cycle Γ has index 0 at all points outside the annulus bounded
by the two circles γ1 (I) and γ2 (I) and index 1 at points inside this annulus. If
we set f (z) = 1/z, then f is analytic in U = C \ {0} and the cycle Γ is a cycle
homologous to 0 in U . Thus, the Cauchy Integral Formula tells us that
 
1 f (z)
dz = dz = 2πif (1) = 2πi.
Γ z −z Γ z−1
2

The Cauchy Integral Theorem. The Cauchy Integral Theorem follows easily
from Theorem 4.2.2.
Theorem 4.2.4. If f is analytic in the open set U and if Γ is a 1-cycle in U which
is homologous to 0 in U , then

f (z) dz = 0.
Γ

Proof. Let z0 be any point in U which is not on Γ(I) and define

g(z) = f (z)(z − z0 ).
4.2. Cauchy’s Theorems 121

Then g(z) is also analytic in U and g(z0 ) = 0. Theorem 4.2.2, applied to the
function g, tells us that
 
g(z)
f (z) dz = dz = 2πi IndΓ (z0 )g(z0 ) = 0.
Γ Γ z − z0
This completes the proof. 

1
Example 4.2.5. Find dz if Γ is the cycle of Example 4.1.10.
Γ sin(πz)
Solution: By Example 4.1.10, Γ has index 0 at every integer k and so it is
homologous to 0 in U = C \ Z, where Z is the set of integers. The function sin(πz)
vanishes exactly on the set of integers (Exercise 4.2.1) and so 1/ sin(πz) is analytic
in U . It follows from the above theorem that the integral in question is 0.
Simple Closed Paths. Although we have stated the general Cauchy theorems
in terms of integration over 1-cycles, they also hold if the integration is over an
ordinary closed path. After all, a closed path is just a particularly simple 1-cycle.
Classically, these theorems are stated and proved for integrals over a particular kind
of closed path – a simple closed path.
A simple closed curve is a closed curve which does not intersect itself except at
the endpoints of the parameter interval I = [a, b]. That is,
Definition 4.2.6. A closed curve γ, defined on [a, b], is called a simple closed curve
if γ(s) = γ(t) for a ≤ s < t ≤ b unless s = a and t = b.
The Jordan Curve Theorem says that if γ is a simple closed curve, then the
complement of γ(I) has two components – one unbounded, and one bounded. Here
we will prove a weak version of the Jordan Curve Theorem – one that is sufficient
for our purposes.
Recall from Definition 2.7.10 that a simple point of a path γ is a point that the
path passes through just once and at which the left and right derivatives of γ are
both non-zero.
Definition 4.2.7. A closed path will be called a simple closed path if each of its
points is a simple point.
Thus, a simple closed path is a simple closed curve which is piecewise smooth
with non-zero left and right derivatives at each point. Our weak version of the
Jordan Curve Theorem is the following.
Theorem 4.2.8. If γ is a simple closed path, then C \ γ(I) has exactly two compo-
nents – a bounded component on which Indγ (z) = ±1 and an unbounded component
on which Indγ (z) = 0.

Proof. By Theorems 2.7.8 and 2.7.11, we may choose for each point z ∈ γ(I) an
open disc Dz , centered at z, such that Dz \ (Dz ∩ γ(I)) has exactly two components
– a left component Lz and a right component Rz , and Indγ (z) is one unit greater
on Lz than on Rz (see Figure 4.2.2).
It is not difficult to see that two sufficiently close points of γ(I) have discs with
overlapping left components and overlapping right components. It follows that the
122 4. The General Cauchy Theorems

Figure 4.2.1. A Simple Closed Path.

Figure 4.2.2. Some of the Discs Dz in the Proof of Theorem 4.2.8.

union of all the left components is a connected open set L, while the union of all
the right components is a connected open set R (see Exercise 4.2.11). Furthermore,
Indγ (z) = Indγ (w) + 1 if z ∈ L and w ∈ R. It follows that L and R are disjoint.
The union of all the discs Dz is an open set U containing γ(I) and, in fact, U is
the union of the disjoint sets γ(I), L, and R.
Now every component of the complement of γ(I) has a subset of γ(I) as its
boundary (Exercise 4.2.12). This implies that every such component must have
non-empty intersection with U and, hence, must meet either L or R. But since L
and R are connected, a component which meets one of them must, in fact, contain
it. It follows that there are only two components of the complement of γ(I) –
one containing L and one containing R. One of these must be the unbounded
component and Indγ is zero on it. The other one is a bounded component and Indγ
must be plus or minus one on it, depending on whether it contains L or R. This
completes the proof of the theorem. 

From the proof of the above theorem, it follows that Indγ (z) = 1 on the bounded
component of C\γ(I) if this component contains L. Intuitively, this means that the
bounded component is on the left as the path is traversed in the positive direction.
In this case, we say that the path has positive orientation. The other possibility is
that Indγ (z) = −1 on the bounded component, which implies that this component
contains R – i.e., that the bounded component is on the right as the path is traversed
in the positive direction. In this case, we say that the path has negative orientation.
4.2. Cauchy’s Theorems 123

If γ is a simple closed path, we will call the bounded component of C \ γ(I) the
inside of γ and the unbounded component the outside of γ.
We may now apply Theorems 4.2.2 and 4.2.4 in the case where the cycle Γ is
a single simple closed path γ to obtain the Cauchy Theorem and Formula in their
classical forms.
Theorem 4.2.9. If γ is a simple closed path and f is a function analytic in an
open set U containing both γ(I) and its inside, then

(4.2.9) f (w) dw = 0
γ

and

1 f (w)
(4.2.10) f (z) = dw
2πi γ w−z
for each z on the inside of γ(I).

Proof. We know that Indγ (z) = 0 on the outside of γ, and this contains the
complement of U (since U contains γ(I) and its inside). Thus, the hypothesis of
Theorem 4.2.2 is satisfied, and so we know that (4.2.9) is true and that

1 f (w)
Indγ (z)f (z) = dw
2πi γ w − z
for z on the inside of γ. Then (4.2.10) follows if we just observe that Indγ (z) = 1
on the inside of γ. 

Exercise Set 4.2

1. Verify the fact, used in Example 4.2.5, that sin(πz) = 0 if and only if z is an
integer. Hint: We know this if z is real, but how do we know there are no
zeroes of sin z except those on the real line?
2. Let Γ = γ1 − γ2 − γ3 , where γ1 , γ2 , and γ3 are positively oriented circles with
radii 5, 1, and 1 and centers 0, −2, and 3, respectively. Find

1
dz
Γ (z + 2)(z − 3)
without calculating the integrals around the individual paths γi .
3. Check your answer to the previous problem by calculating

1
dz
γj (z + 2)(z − 3)

for each of the individual paths γj . Hint: The integral around γ1 may be
calculated by using a partial fraction decomposition of the integrand.
4. If γ is any simple closed path, with −2 and 3 inside γ, find

1
dz.
γ (z + 2)(z − 3)
124 4. The General Cauchy Theorems

5. For the cycle Γ of Exercise 2, find


1
dz
Γ z(z + 2)(z − 3)
by applying the Cauchy Integral Formula for the cycle Γ and the function
1
f (z) = .
(z + 2)(z − 3)
6. If γ1 and γ2 are positively oriented circles, centered at 0, with radii 1 and 2,
respectively, and if Γ = γ2 −γ1 , find the integrals of ez /(z −1/2), ez /(z −3/2),
and ez /(z − 3) around Γ.
7. If γ is a simple closed path with −1 and 1 on its inside and f is an entire
function, show that

f (z)
2−1
dz = πi(f (1) − f (−1)).
γ z
8. If γ is a simple closed path with 0 on its inside and all other integral multiples
of π on its outside, find 
1
dz.
γ sin z

9. For positive numbers r < R, set γ1 (t) = r e2πit and γ2 (t) = R e2πit for t ∈ [0, 1]
and set Γ = γ2 − γ1 . Prove that if f is analytic in an open set containing the
set {z : r ≤ |z| ≤ R}, then

1 f (w)
(4.2.11) f (z) = dw
2πi Γ w − z
for every z with r < |z| < R.
10. For the cycle Γ and function f of the previous problem, what is the integral
(4.2.11) if |z| < r or |z| > R?
11. Let γ be a simple closed path and, for each z ∈ γ(I), let Dz , Lz and Rz be the
sets used in the proof of Theorem 4.2.8. Prove that if z and w are two points
of γ(I) such that Dz ∩ Dw ∩ γ(I) = ∅, then Lz ∩ Lw = ∅ and Rz ∩ Rw = ∅.
Show that this implies that the union of all the Lz is a connected set as is the
union of all the Rz .
12. Prove that if K is a closed subset of C, then each component of C \ K has its
boundary contained in K.

4.3. Laurent Series

In this section, we will show that a function which is analytic on an annulus has
a special kind of power series expansion. We first make a preliminary observation
about functions which have limit 0 at infinity.
Recall that a neighborhood of ∞ is any open set which contains the complement
of some closed bounded disc. If f is a function which is analytic in a neighborhood
of ∞, then we say that f vanishes at infinity if
lim f (z) = 0.
z→∞
4.3. Laurent Series 125

Lemma 4.3.1. If h is a function which is analytic in the exterior of the disc Dr (z0 )
and vanishes at ∞, then the function

h(1/w + z0 ), if w = 0;
q(w) =
0, if w = 0;
is analytic in D1/r (0).

Proof. The number 1/w + z0 is in C \ Dr (z0 ) if and only if 1/|w| > r, that is, if
and only if w ∈ D1/r (0). So the domain of q is D1/r (0). Clearly q is analytic on
the complement of {0} in this disc.
The fact that limz→∞ h(z) = 0 implies that limw→0 q(w) = 0. Thus, q is
continuous at 0. It follows from the Removable Singularity Theorem (Theorem
3.4.8) that it is analytic on all of D1/r (0). 

Analytic Functions on an Annulus. An open annulus centered at z0 is a set

of the form
(4.3.1) A = {z ∈ C : r < |z − z0 | < R},
where 0 ≤ r < R ≤ ∞. The numbers r and R are the inner and outer radii of A,
respectively.
For the annulus A of (4.3.1), and numbers s and S with
r < s < S < R,
consider two paths γs and γS in A, which traverse the circles |w − z0 | = s and
|w − z0 | = S once in the positive direction. If we define a cycle Γ by
Γ = γS − γs ,
then ⎧
⎪
⎨0, if |z − z0 | > S;
IndΓ (z) = 1, if s < |z − z0 | < S;
⎪
⎩
0, if |z − z0 | < s.
In particular, this means that Γ is homologous to 0 in A. From this and the general
Cauchy theorems, it follows that if f is a function which is analytic in A, then
⎧
 ⎪
⎨0, if |z − z0 | > S;
1 f (w)
dw = f (z), if s < |z − z0 | < S;
2πi Γ w − z ⎪
⎩
0, if |z − z0 | < s.
In other words,
 
1 f (w) 1 f (w)
(4.3.2) dw = dw
2πi γS w−z 2πi γs w−z
if s and S are on the same side of |z − z0 | and
 
1 f (w) 1 f (w)
(4.3.3) f (z) = dw − dw
2πi γS w − z 2πi γs w − z
if s < |z − z0 | < S. This is the basis for the following result.
126 4. The General Cauchy Theorems

Theorem 4.3.2. If f is an analytic function on an annulus A of the form (4.3.1),

then there exists a unique way of writing f as
f (z) = g(z) − h(z) for z ∈ A,
where g is an analytic function on DR (z0 ) and h is an analytic function on the
complement of Dr (z0 ), which vanishes at infinity.

Proof. We define the function g on DR (z0 ) as follows: If |z − z0 | < R, we choose

S such that |z − z0 | < S and r < S < R. We choose a closed path γS which traces
the circle |w − z0 | = S, once in the positive direction; and we set

1 f (w)
g(z) = dw.
2πi γS w − z
It follows from (4.3.2) that this does not depend on the choice of S as long as
|z − z0 | < S < R.
Similarly, we define the function h on C \ Dr (z0 ) as follows. If |z − z0 | > r, we
choose s such that r < s < R and s < |z − z0 |; we choose a closed path γs which
traces the circle |w − z0 | = s once in the positive direction; and we set

1 f (w)
(4.3.4) h(z) = dw.
2πi γs w − z
Again, by (4.3.2), this does not depend on the choice of s.
By (4.3.3)
f (z) = g(z) − h(z)
if z ∈ A. We emphasize that z was an arbitrary point of A, and that g and h do
not depend on the choices for the circles γS and γs .
It follows from Morera’s Theorem that g is analytic in the open disc DR (z0 )
and h is analytic in the exterior of the closed disc Dr (z0 ) (see Exercise 3.2.16).
Also, for a fixed s with r < s < R, |z − z0 | > s, and γs as above,
|z − w| ≥ |z| − |z0 | − |w − z0 | = |z| − |z0 | − s if w ∈ γs (I),
and so  
 f (w)  M
 
 z − w  ≤ |z| − |z0 | − s if w ∈ γs (I),

where M = sup{|f (w)| : w ∈ γs (I)}. From this bound on the integrand in (4.3.4),
it follows that
Ms
|h(z)| ≤
|z| − |z0 | − s
and so, since |z0 | and s are fixed,
lim h(z) = 0,
z→∞

as required.
The uniqueness of g and h follows from Liouville’s Theorem. In fact, suppose
we also have
f = g1 − h1
4.3. Laurent Series 127

S
z
R
r s
z0
γs γS

Figure 4.3.1. The Paths γS and γs in the Proof of Theorem 4.3.2.

with g1 analytic on DR (z0 ), h1 analytic on C \ Dr (z0 ), and h1 vanishing at infinity.

Then
g − g1 = h1 − h on A.
This means that the analytic function g − g1 on DR (z0 ) and the analytic function
h1 − h on C \ Dr (z0 ) agree on the intersection of these two open sets. It follows that
they define an entire function on the plane which vanishes at infinity. By Liouville’s
Theorem, the only such function is the identically 0 function. Thus, g = g1 and
h = h1 . 

Example 4.3.3. Find a decomposition like that in the previous theorem if the
function f is
1
f (z) =
(z − 1)(z − 2)
and the annulus A is A = {z ∈ C : 1 < |z| < 2}.
Solution: This is just the partial fraction decomposition of f . We set
1
g(z) = for |z| < 2,
z−2
and
1
h(z) = for |z| > 1.
z−1
Then g and h are analytic, f = g − h on A, and h vanishes at infinity.

Laurent Series Expansion. A function which is analytic on an annulus has a

special kind of series expansion called a Laurent Series expansion.
Theorem 4.3.4. Let f be a function which is analytic on the annulus A of (4.3.1).
Then f has a unique series expansion of the form
∞

(4.3.5) f (z) = cn (z − z0 )n ,
n=−∞

which converges to f at all points of A and converges uniformly on compact subsets

of A.
128 4. The General Cauchy Theorems

Proof. We write f = g − h with g and h as in the previous theorem. Since g is

analytic on DR (z0 ), it has a power series expansion
∞

g(z) = cn (z − z0 )n ,
n=0
which converges on DR (z0 ) and converges uniformly on compact subsets of this
disc.
By Lemma 4.3.1 the function q(w) = h(1/w + z0 ) is, if given the value 0 at
w = 0, an analytic function of w in the disc D1/r (0), and it vanishes at w = 0.
Therefore, it has a power series expansion
∞

q(w) = bn wn
n=1
which converges uniformly on compact subsets of D1/r (0). If we set cn = −b−n for
negative integers n and make the substitution w = (z − z0 )−1 , then this becomes
−1

h(z) = − cn (z − z0 )n .
n=−∞
On combining the series for g and h, using the fact that f = g − h in A, we obtain
(4.3.5). The series converges uniformly on compact subsets of A because the series
for g and h both have this property.
The uniqueness of the expansion follows from the uniqueness statements in
Theorem 4.3.2 and Corollary 3.2.4. 
Example 4.3.5. For the function f and annulus A of Example 4.3.3 find the
Laurent series expansion for f on A.
1 1/2
Solution: We have f = g − h, where g(z) = = − has power
z−2 1 − z/2
series expansion
∞
1  zn
g(z) = −
2 n=0 2n
1 1/z
in D2 (0), and h(z) = = has an expansion
z−1 1 − 1/z
∞
1
h(z) =
n=1
zn
on C \ D1 (0). Thus, the function f = g − h has Laurent expansion
−1
 ∞  
1
f (z) = n
(−1)z + − n+1 z n
n=−∞ n=0
2

in A = D2 (0) ∩ (C \ D1 (0)).
1
Example 4.3.6. The function f = of the previous exercise is also
(z − 1)(z − 2)
analytic on the annulus
B = {z ∈ C : 0 < |z − 1| < 1},
4.3. Laurent Series 129

which is centered at 1. Find its Laurent expansion in this annulus.

1 1
Solution: Again, f = g − h, with g(z) = and h(z) = . Here g is
z−2 z−1
analytic in the disc D1 (1), with power series expansion
∞

g(z) = − (z − 1)n ,
n=0

while h(z) is analytic in C \ {1} with limit 0 at infinity. Since it is already a power
of (z − 1), it needs no further expansion. Thus, the Laurent expansion of f in the
annulus B is
∞
 ∞

−1
f (z) = − (z − 1) − (z − 1)
n
=− (z − 1)n .
n=0 n=−1

Example 4.3.7. Find the Laurent expansion of e1/z in the annulus C \ {0}.
Solution: Since f (z) = e1/z is analytic except at 0 and has limit 1 at infinity,
in the decomposition f = g −h of Theorem 4.3.2 the function g is 1 and the function
h is 1 − e1/z . The Laurent expansion is

0
zn
1/z
e = .
n=−∞
|n|!

The coefficients of the Laurent expansion of a function analytic in an annulus

are given by an integral formula similar to the one for the power series expansion
of a function analytic in a disc (Theorem 3.2.5).

Theorem 4.3.8. If A = {z ∈ C : r < |z − z0 | < R}, f is analytic in A and

r < s < R, then the Laurent series (4.3.5) for f in A has coefficients given by

1 f (w)
(4.3.6) ck = dw.
2πi |w−z0 |=s (w − z0 )k+1

Proof. With f given by (4.3.5), we have

 ∞ 
1 f (w)
(4.3.7) dw = c n (w − z0 )n−k−1 dw.
2πi |w−z0 |=s (w − z0 )k+1 n=−∞ |w−z0 |=s

A path which traverses the circle |w − z0 | = s once in the positive direction is given
by γ(t) = z0 + s eit , for 0 ≤ t ≤ 2π. Then

γ  (t) = si eit and γ(t) − z0 = s eit .

Thus,
  2π
(w − z0 )n−k−1 dw = isn−k ei(n−k)t dt.
|w−z0 |=s 0

This is 0 if n = k and 2πi if n = k. Combined with (4.3.7), this proves the

theorem. 
130 4. The General Cauchy Theorems

Exercise Set 4.3

1. In Examples 4.3.5 and 4.3.6, Laurent expansions for

1
f (z) =
(z − 1)(z − 2)
are given in two different annuli. Find a third annulus in which this function
is analytic and find its Laurent expansion there. Are there other such annuli?
2. Find the Laurent expansion of
1
f (z) = in the annulus A = {z ∈ C : 0 < |z| < 1}.
z − z2

3. Find the Laurent expansion for f (z) = z 3 e1/z in C \ {0}.

4. Find the Laurent expansion for f (z) = z −3 ez in C \ {0}.
5. Find the Laurent expansion for
log(1 + z)
f (z) = in C \ {0}.
z4
6. Find the Laurent expansion for
z
f (z) = in the annulus A = {z ∈ C : 0 < |z − i| < 2}.
z2 + 1
7. Find the Laurent expansion for f (z) = sin(1/z) in C \ {0}.
8. Find the Laurent expansion for f (z) = ez + e1/z in C \ {0}.
9. Find the Laurent expansion for f (z) = ez+1/z in C \ {0}. Hint: You will need
the binomial formula.
10. Prove that if f is analytic in a disc Dr (z0 ) except at z0 , where it has a pole of
order k, then, in the annulus {z : 0 < |z − z0 | < r}, the Laurent expansion for
f has only finitely many terms with negative exponent and the most negative
exponent that appears is −k.
11. Suppose f is analytic in the annulus A = {z ∈ C : R < |z|} and satisfies the
inequality |f (z)| ≤ |z|k in this set. Then prove that the Laurent expansion of
f in A has no terms with positive exponent greater than k.
1
12. If cn is the nth coefficient in the Laurent series expansion of in the
sin z
annulus {z : 0 < |z| < π}, show that cn = 0 if n is even or if n < −1.
13. With cn as in the previous exercise, find c−1 , c1 , and c3 Hint: use long division
to divide sin z into 1.
1
14. For the function of the previous two exercises, how do the Laurent
sin z
coefficients change when we compute the Laurent expansion in the annulus
{z : π < |z| < 2π} instead of the annulus {z : 0 < |z| < π}? Hint: Use
Cauchy’s formula to compute the difference in formula (4.3.6) if the radius s
of the circle of integration is changed from one with 0 < s < π to one with
π < s < 2π.
4.4. The Residue Theorem 131

4.4. The Residue Theorem

In this section we return to the study of analytic functions with isolated singulari-
ties, as discussed in Section 3.4. Recall that a function f has isolated singularities
on an open set U if it is analytic on U except on a discrete subset.
In particular, if a function f is analytic on an open set containing the annulus
Dr (z0 ) \ {z0 }, then it has an isolated singularity at z0 . Since Dr (z0 ) \ {z0 } is an
annulus centered at z0 , f has a Laurent expansion of the form
∞
(4.4.1) f (z) = cn (z − z0 )n .
n=−∞

The coefficient c−1 of (z − z0 )−1 in (4.4.1) plays a special role in complex

analysis.
Definition 4.4.1. If f is a function which is analytic on A = Dr (z0 )\{z0 }, with an
isolated singularity at z0 , then the coefficient of (z −z0 )−1 in the Laurent expansion
of f on A is called the residue of f at z0 and is denoted by Res(f, z0 ).

The residue of f at z0 is given by a simple integral formula, which follows

immediately from Theorem 4.3.8.
Theorem 4.4.2. Let f be an analytic function with an isolated singularity at z0 ∈
C. Let R > r > 0 be numbers with f analytic on DR (z0 ) \ {z0 }. Then

1
Res(f, z0 ) = f (z) dz.
2πi |z−z0 |=r

This leads to the following important application of the general Cauchy Theo-
rem.
Theorem 4.4.3 (Residue Theorem). Let f be a function which is analytic on U \E,
where U is an open subset of C and E is a discrete subset of U . If γ is a closed
path in U \ E which is homologous to 0 in U , then
(a) there are only finitely many points of E at which Indγ is non-zero;
(b) if these points are {z1 , z2 , · · · , zn }, then
 n
1
(4.4.2) f (z) dz = Indγ (zj ) Res(f, zj ).
2πi γ j=1

Proof. We first prove (a). Recall from the proof of Theorem 2.7.5 that if r is
chosen so that γ(I) ⊂ Dr (0), then the bounded components of C \ γ(I) are also
contained in Dr (0). Also, Indγ (z) is non-zero only on certain bounded components
of C \ γ(I). It follows that the union of γ(I) and the components of its complement
where Indγ (z) is non-zero is a bounded set K. The set K is also the complement
of the union of the components of C \ γ(I) on which Indγ = 0, and so it is closed.
Thus, K is compact.
Since γ is homologous to 0 in U , every point of the complement of U is a point
where Indγ is 0. This implies K ⊂ U . Since the singularities of f form a discrete
subset E of U , we may choose for each point of U an open disc, centered at that
point and contained in U , which either contains one singularity (the center of the
132 4. The General Cauchy Theorems

zj
γj

Figure 4.4.1. The Paths γ and γj in the Proof of Theorem 4.4.3.

disc) or no singularities. This collection of open discs covers K (since its union
is U ), and so some finite subcollection also covers K. But this means there are
only finitely many singularities of f in K, and, hence, there are only finitely many
singularities of f at which Indγ is nonzero. This completes the proof of (a).
Let z1 , z2 , · · · , zn be the singularities of f at which Indγ is nonzero. For each
of these points zj , we choose an rj > 0 such that Drj is contained in the open set
U \ γ(I). We choose r > 0 such that r < min{r1 , · · · , rn } and the discs Dr (zj ) are
non-overlapping. We then set

(4.4.3) mj = Indγ (zj )

and define a 1-cycle Γ by


n
Γ=γ− m j γj ,
j=1

where γj (t) = zj + r e2πit for t ∈ [0, 1] (we may assume that γ also has [0, 1] as its
parameter interval).
Now each zj is a point where Indγ is non-zero, and so is every point in the open
disc Dr (zj ). Hence the closure D r (zj ) of this disc is contained in K ⊂ U . This
means that the complement of U is contained in the complement of Dr (zj ). Thus,
Indγj (z) = 0 on the complement of U . Since this is true for every j and is also true
of γ, we have that IndΓ (z) = 0 on the complement of U – that is, Γ is homologous
to 0 in U .
Note, f is not analytic in U and so the general Cauchy Theorem does not yet
apply. However, we also have that

Indγ (zj ) = mj = mj Indγj (zj )

while
mj Indγj (zk ) = 0 for k = j.
4.4. The Residue Theorem 133

It follows that Γ is also homologous to 0 in U \ E. Now U \ E is a set on which f

is analytic, and so, by the general Cauchy Integral Theorem,
  
n 
0= f (z) dz = f (z) dz − mj f (z) dz.
Γ γ j=1 γj

Now (4.4.2) follows from this, Theorem 4.4.2, and (4.4.3). This completes the
proof. 

The Residue Theorem has a vast number of applications. We will explore

many of these in the remainder of this section and the next chapter. The next
three examples give just a taste of how the Residue Theorem is used to calculate
integrals.
Example 4.4.4. Let f be a function of the form
g(z)
f (z) = ,
z − z0
where g is analytic in an open set containing z0 . Show that
(4.4.4) Res(f, z0 ) = g(z0 ).
Solution: The function g(z) is analytic in some disc centered at z0 and so
it has a power series expansion in this disc with constant term g(z0 ). Hence the
Laurent series expansion of f (z) about z0 has g(z0 ) as the coefficient of (z − z0 )−1 .
By definition, this means that (4.4.4) holds.
Example 4.4.5. Let γ be a simple closed path with 1 and 2 inside γ. Compute

z+1
dz.
γ (z − 1)(z − 2)

Solution: By the Residue Theorem, this integral is

2πi(Res(f, 1) + Res(f, 2)),
z+1 z+1
where f (z) = . The function g(z) = is analytic in a disc
(z − 1)(z − 2) z−2
g(z)
centered at 1 and f (z) = . Hence, by the previous example,
z−1
Res(f, 1) = g(1) = −2.
z+1
By the same reasoning, if h(z) = , then
z−1
Res(f, 2) = h(2) = 3.
We conclude that

z+1
dz = 2πi(3 − 2) = 2πi.
γ (z − 1)(z − 2)
Example 4.4.6. If γ is a simple closed path with 0 and π inside γ and all other
multiples of π outside γ, find 
1
dz.
γ sin z
134 4. The General Cauchy Theorems

1
Solution: The integrand f (z) = has isolated singularities inside γ at 0
sin z
and π. The function
z
g(z) =
sin z
has a removable singularity at 0 and the value of the resulting analytic function at
0 is 1. Thus,
Res(f, 0) = 1.
Since sin(z) = − sin(z − π), the function
z−π z−π
h(z) = =−
sin z sin (z − π)
has a removable singularity at z = π. The value at π of the resulting analytic
function is −1. Thus,
Res(f, π) = −1.
It follows from the Residue Theorem that the integral of f around γ is zero.

Counting Zeroes and Poles. Recall that a meromorphic function on an open

set U is a function which is analytic on U except on a discrete subset E where it
has poles.
A simple pole is a pole of order 1. If f is a meromorphic function in an open
set U , then, as we will show in the proof of the next theorem, the function f  /f has
a simple pole at each zero and at each pole of f . Furthermore, these simple poles
have a special form which leads directly to an integral formula for the sum of the
number of zeroes minus the number of poles, counting multiplicity, surrounded by
a closed path in U .
Theorem 4.4.7. If f is a meromorphic function on U and z0 ∈ U , then
Res(f  /f, z0 ) = k,
where k is the order of the zero of f at z0 , or minus the order of the pole of f at
z0 , or 0 if f has no zero or pole at z0 .

Proof. We may factor f as

f (z) = (z − z0 )k g(z),
where g is meromorphic on U and has no zero or pole at z0 . The integer k is
positive, negative, or zero, depending on whether f has a zero, a pole, or neither
at z0 . Then
f  (z) = k(z − z0 )k−1 g(z) + (z − z0 )k g  (z),
and so
f  (z) k g  (z)
(4.4.5) = + .
f (z) z − z0 g(z)
If Dr (z0 ) is a disc contained in U , which contains no zero or pole of f except
possibly z0 , then when we integrate (4.4.5) around the boundary of this disc and
divide by 2πi, we get
Res(f  /f, z0 ) = k,
since g  /g is analytic on an open set containing Dr (z0 ) and, thus, has integral 0. 
4.4. The Residue Theorem 135

If we combine this with the Residue Theorem, the result is the following theo-
rem.
Theorem 4.4.8. Let f be a meromorphic function defined in a open set U and let
γ be a closed path in U which is homologous to 0 in U . Assume there are no zeroes
or poles of f on γ(I) and the zeroes and poles of f at which Indγ is not zero occur
at the points z1 , z2 , · · · , zn . Then
n  
1 f (z)
(4.4.6) mj kj = dz,
j=1
2πi γ f (z)

where, for j = 1, · · · , n, the integer kj is the order of the zero of f at zj or minus

the order of the pole at zj , and mj = Indγ (zj ).
Corollary 4.4.9. Let U , f , {z1 , · · · , zn }, {k1 , · · · kn }, and γ be as in Theorem
4.4.8. If we compose f with γ to form a new path f ◦ γ, then
n
(4.4.7) mj kj = Indf ◦γ (0),
j=1

where mj = Indγ (zj ).

Proof. If the parameter interval of γ is [a, b], we have

  b
1 1 1 (f ◦ γ) (t)
Indf ◦γ (0) = dz = dt
2πi f ◦γ z 2πi a f ◦ γ(t)
 b   
1 f (γ(t))γ  (t) 1 f (z)
= dt = dz.
2πi a f (γ(t)) 2πi γ f (z)

By Theorem 4.4.8, this last integral is equal to nj=1 mj kj . 

If the path γ is simple, then the previous theorem and corollary are simpler.
Corollary 4.4.10. Let f be a meromorphic function on the open set U . If γ is a
simple closed path in U , with its inside contained in U , which does not pass through
a zero or pole of f , and {z1 , · · · , zn } is the set of zeroes and poles of f inside γ,
then
 n  
1 f (z)
(4.4.8) kj = dz = Indf ◦γ (0),
j=1
2πi γ f (z)
where, for j = 1, · · · , n, kj is the order of the zero or minus the order of the pole
at zj .

Proof. Since the inside of γ is contained in U , γ is homologous to 0 in U . Thus,

Theorem 4.4.8 and the previous corollary apply. Since Indγ is 1 on the inside of γ,
it is 1 at each zj . Equation (4.4.8) follows. 
Example 4.4.11. Suppose f is a meromorphic function in a convex open set U .
Suppose that the only zero of f in U occurs at z1 and has order k, and the only
pole of f in U occurs at z2 and also has order k. Prove that there is an analytic
logarithm of f defined in V = U \ [z1 , z2 ]. That is, prove that there is an analytic
function g on V such that f (z) = eg(z) for every z ∈ V .
136 4. The General Cauchy Theorems

Solution: If γ is any closed path in V , then γ is homologous to 0 in U , since

U is convex. Since z1 and z2 are connected by a line segment in the complement
of V , they lie in the same component of the complement of γ(I). This implies
Indγ (z1 ) = Indγ (z2 ). Then, by Theorem 4.4.8,
 
1 f (w)
dw = Indγ (z1 )k − Indγ (z2 )k = 0.
2πi γ f (w)
It follows from this that, if z0 is a fixed point and z is a variable point of V , and
γz is a path in V which begins at z0 and ends at z, then

f  (w)
h(z) = dw
γz f (w)

is independent of which path γz is chosen. Furthermore, we know that h is an

f  (w)
antiderivative of , by Theorem 2.5.6. Thus, h = f  /f . This implies
f (w)
(f e−h ) = f  e−h −f h e−h = 0,
and, hence, that f = C eh for some non-zero constant C. Then
g(z) = h(z) + log(C)
has the required properties, where log is any branch of the log function.

Exercise Set 4.4

1
1. If f (z) = , find the residue of f at each of its singularities.
2z 2 − 5z + 2 
1
2. If γ is a simple closed path with 0 and 3 inside γ, find dz.
γ z − 3z
2

ez
3. Find .
|z|=2 z − 1
2

4. If γ is a simple closed path with 0 inside, but no other multiples of 2πi inside,
find 
1
dz.
γ e −1
z

5. If γ is any simple closed path with 0, 1, and 2 inside, find


1
dz.
γ z(z − 1)(z − 2)

3z 2 − 1
6. Find dz.
|z|=2 z − z
3

7. Find tan z dz.

|z|=π
log z
8. Find dz.
|z−5|=4 sin z
9. Let f be a meromorphic function on C with a zero of order 1 at z = 1 and a
pole of order 1 at z = −1 and no other zeroes or poles. If γ is a simple closed
4.5. Rouché’s Theorem and Inverse Functions 137

path which does not pass through 1 or −1, what are the possible values for

f  (z)
and what determines which value is achieved?
γ f (z) 
10. Use Theorem 4.4.8 to derive a formula for γ cot z dz, where γ is any closed
path which does not pass through an integral multiple of π.
z+1
11. Prove that there is an analytic logarithm for defined in the open set
z−1
U = C \ [−1, 1].
12. Example 4.4.11 is only one of many possible theorems concerning the existence
of analytic logarithms that can be proved using Theorem 4.4.8. Invent and
prove another one.

4.5. Rouché’s Theorem and Inverse Functions

Corollary 4.4.10 leads to a proof of Rouché’s Theorem, which is a very useful tool
for analyzing the zeroes of an analytic function.
Theorem 4.5.1 (Rouché’s Theorem). Let f and g be analytic in an open set U
and let γ be a simple closed path in U , with its inside contained in U , and with
parameter interval I. If f has no zero on γ(I), and
(4.5.1) |f (z) − g(z)| ≤ |g(z)| on γ(I),
then f and g have the same number of zeroes, counting order, inside γ.

Proof. If f has no zeroes on γ(I), then (4.5.1) implies that g also has no zeroes
on γ(I). If we set h(z) = f (z)/g(z), then h is meromorphic in U and has no zeroes
or poles on γ(I). The inequality (4.5.1) implies that the curve h ◦ γ satisfies
|h ◦ γ(t) − 1| ≤ 1
for all t ∈ I. It follows from this that 0 is in the unbounded component of C\h◦γ(I).
Hence,
Indh◦γ (0) = 0.
The path γ is homologous to 0 in U and so we may apply Corollary 4.4.9. It tells
us that the number of zeroes minus the number of poles of h inside γ, counting
order, is 0. However, this is the number of zeroes of f minus the number of zeroes
of g inside γ, counting order. Hence, these two numbers are the same. 
Example 4.5.2. How many zeroes, counting order, does the polynomial
4z 5 − z 3 + z 2 − 2
have inside the unit circle?
Solution: We apply Rouché’s Theorem with f (z) = 4z 5 − z 3 + z 2 − 2 and
g(x) = 4z 5 . On the unit circle |z| = 1, we have
|f (z) − g(z)| = | − z 3 + z 2 − 2| ≤ |z|3 + |z|2 + 2 = 4 = |g(z)|.
By Rouché’s Theorem, f and g have the same number of zeroes inside the unit
circle. Since g(z) = 4z 5 has 5 zeroes, counting order, inside this circle, so does f .
138 4. The General Cauchy Theorems

Example 4.5.3. Let {fn } be a sequence of non-vanishing analytic functions on a

connected open set U . If {fn } converges uniformly to f on each compact subset of
U , prove that f is either also non-vanishing on U or is identically 0.
Solution: Assume f is not identically 0. Let z0 be any point of U . We will
prove that f (z0 ) = 0.
We may choose a closed disc Dr (z0 ) on which f has no zeroes except possibly
at z0 . This is possible because f is analytic and not identically zero and, hence, has
a discrete set of zeroes in U . Since f is non-vanishing on the boundary of Dr (z0 ),
it has a minimum modulus m > 0 on ∂Dr (z0 ). Since fn → f uniformly on Dr (z0 ),
there is an N such that
n≥N implies |f (z) − fn (z)| < m ∀ z ∈ Dr (z0 ).
Since m ≤ |f (z)| for all z ∈ ∂D r (z0 ), Rouché’s Theorem implies that f and fn have
the same number of zeroes in Dr (z0 ) if n ≥ N . Since fn has no zeroes, f has no
zeroes in Dr (z0 ). In particular, f (z0 ) = 0.

The Inverse Mapping Theorem. An application which demonstrates the util-

ity of Rouché’s Theorem is the following proof of the Inverse Function Theorem for
analytic functions.
A function from a set U to a set V is said to be one-to-one if f (z) = f (w) for
every pair of distinct points z, w ∈ U . It is said to be onto if f (U ) = V – that
is, if every point of V is the image of some point of U . A function f : U → V is
one-to-one and onto if and only if it has a well-defined inverse function f −1 . This
is defined by the condition that f (z) = w for z ∈ U if and only if f −1 (w) = z. This
is the same as saying that
f −1 ◦ f (z) = z for all z ∈ U, and
−1
f ◦f (w) = w for all w ∈ V.
The inverse function for a continuous function f may or may not be continuous; if
it is, we say that f has a continuous inverse function on U . Similarly, if U and V
are open, f : U → V is analytic, and f −1 is also analytic, then we say that f has
an analytic inverse function on U .
Typically, analytic functions are not one-to-one and, hence, do not have inverse
functions on their full domains. However, they quite often have local analytic
inverse functions in the sense of the following definition.
Definition 4.5.4. If f is analytic in U and z0 ∈ U , then we say that f has a local
analytic inverse at z0 if there are neighborhoods V of z0 and W of w0 = f (z0 ) such
that f is a one-to-one map of V onto W and its inverse function f −1 : W → V is
analytic.

Theorem 4.5.5 (Inverse Mapping Theorem). If f is analytic on U , z0 ∈ U ,

and f  (z0 ) = 0, then f has a local analytic inverse f −1 at z0 . Furthermore, the
1
derivative of f −1 is (f −1 ) (w) =  −1 .
f (f (w))
4.5. Rouché’s Theorem and Inverse Functions 139

f
V
U −1
f

Figure 4.5.1. A Function f and its Inverse Function f −1 .

Proof. We set w0 = f (z0 ). If f  (z0 ) = 0, then the function f (z) − w0 also has non-
vanishing derivative at z0 and, hence, it has a zero of order 1 at z0 . Furthermore,
by Theorem 3.4.2 there is an r > 0 such that, in the disc Dr (z0 ), this is the only
zero of f (z) − w0 .
We choose δ such that 0 < δ < r and f  (z) = 0 for all z ∈ Dδ (z0 ). This is
possible because f  is continuous and so f −1 (C \ {0}) is an open set containing z0 .
We next let γ be the circle
γ(t) = z0 + δ e2πit , t ∈ [0, 1].
Then f (z)−w0 has no zero on the compact set γ(I). This means that the minimum
value  of |f (z) − w0 | in γ(I) is positive. Then if |w − w0 | <  and z ∈ γ(I), we
have
|(f (z) − w) − (f (z) − w0 )| = |w − w0 | <  ≤ |f (z) − w0 |.
By Rouché’s Theorem, the two functions of z, f (z) − w and f (z) − w0 , have the
same number of zeros, counting order, in Dδ (z0 ). Since, f (z) − w0 has one zero in
this disc, so does f (z) − w, and this is true for each w ∈ D (w0 ).
We conclude from the above that, for each w ∈ D (w0 ), there is exactly one
z ∈ Dδ (z0 ) such that f (z) = w. In other words, if we set
W = D (w0 ) and V = f −1 (W ),
then f : V → W is one-to-one and onto, and, hence, has a well-defined inverse
function f −1 .
We claim that f −1 is continuous on W . By Theorem 2.1.13, to show this,
we only need to show that its inverse function, f , takes open sets to open sets.
However, we just showed that an analytic function f on an open set U , which has
a non-zero derivative at a point z0 , takes U to a set which contains a neighborhood
of f (z0 ). This result applies equally well to any open set containing any point z
of V , since f  (z) = 0 for z ∈ V . Thus, f takes any open subset of V to an open
subset of W and this implies that f −1 : W → V is continuous.
We next show that the inverse function is analytic and has the indicated de-
rivative. If w and w1 are two points of W , with z = f −1 (w) and z1 = f −1 (w1 ) the
corresponding points of V , then w = f (z), w1 = f (z1 ), and
 −1
f −1 (w) − f −1 (w1 ) f (z) − f (z1 ) 1 1
lim = lim =  =  −1 .
w→w1 w − w1 z→z1 z − z1 f (z1 ) f (f (w1 ))
140 4. The General Cauchy Theorems

1
This shows that (f −1 ) (w1 ) exists and equals . Of course, the fact that
f  (f −1 (w 1 ))
z → z1 as w → w1 follows from the continuity of f −1 . 

The hypothesis that f  (z0 ) = 0 in the above theorem is necessary. In fact, we

have the following theorem, the proof of which is left as an exercise (Exercise 4.5.9).
Theorem 4.5.6. If f is analytic in U , z0 ∈ U , and f  (z0 ) = 0, then f has no
local analytic inverse at z0 . In fact, if f  (z0 ) = 0, then f is not one-to-one in any
neighborhood of z0 .
Example 4.5.7. At which points z0 is it true that there is an analytic inverse
function for sin z? What is the derivative of the inverse function at w = sin(z) for
each such point z?
Solution: By Theorems 4.5.5 and 4.5.6, sin has an analytic inverse exactly at
those points z0 for which sin z0 = 0. Since, sin z = cos z, the points we are looking
for are those which are not odd multiples of πi/2. The derivative of the inverse
function sin−1 at w = sin z for any such point z is
1 1 1
 −1 = −1 = .
sin (sin w) cos(sin w) cos z
Since cos2 z = 1 − sin2 z = 1 − w2 , cos z will be some square root of 1 − w2 . Thus,
we may write
1
(sin−1 ) (w) = √
1 − w2
with the understanding that we are using some branch of the square root function.
Which branch is being used depends on the two points z0 and w0 = sin z0 and
the neighborhood on which the inverse function is defined. For example, √ for z0 =
w0 = 0, we must use a branch of the square root function for which 1 = 1, since
cos 0 = 1. On the other hand, if z0 √= π and w0 = 0, then we must use a branch of
the square root function for which 1 = −1, since cos π = −1.

The Open Mapping Theorem. If U is an open subset of C, then an open

mapping from U to C is a function f : U → C such that f (V ) is open for every
open subset V ⊂ U .
Theorem 4.5.8 (Open Mapping Theorem). A non-constant analytic function on
a connected open set U is an open mapping of U to C.

Proof. Let f be a non-constant analytic function defined on a connected open set

U . If V is any open subset of U , we need to show that f (V ) is open. We will do
this by showing that for each z0 ∈ V there is a neighborhood of f (z0 ) contained in
f (V ).
Let w0 = f (z0 ) and let k be the order of the zero of f (z)−w0 at z0 . By Exercise
3.4.8, there is a neighborhood V1 of z0 (which we may assume is contained in V )
and an analytic function g on V1 such that, on V1 ,
f (z) − w0 = g k (z) and g  (z0 ) = 0.
The previous theorem implies that there is a neighborhood V2 of z0 , with V2 ⊂ V1
such that g(V2 ) is open. In fact, we may assume g(V2 ) is an open disc D (0) centered
4.6. Homotopy 141

at g(z0 ) = 0. Then the image of V2 under g k is the open disc Dk (0), and so the
image of V2 under f is f (z0 ) + Dk (0). Since this is a neighborhood of f (z0 ) which
is contained in f (V ), the proof is complete. 

Exercise Set 4.5

1. How many zeroes, counting order, does the polynomial 3z 7 − z 3 + 1 have in

the open unit disc?
2. How many zeroes, counting order, does the polynomial z 5 − 4z 3 + z − 1 have
in the open unit disc?
3. Prove that if 0 ≤ C ≤ 1/ e, then the equation z = C ez has exactly one solution
in the unit disc.
4. Prove that if h and g are analytic functions defined in a neighborhood of the
closed unit disc and if h has k zeroes in the open unit disc and no zeroes on
the unit circle, then there is a δ > 0 such that h + λg also has k zeroes in the
open unit disc for all positive numbers λ < δ.
5. Prove that if {fn } is a sequence of one-to-one analytic functions on a connected
open set U , and if fn → f uniformly on compact subsets of U , then f is either
also one-to-one or is a constant.
6. Prove that if f is analytic in an open set U and has a zero of order k at a
point z0 in U , then there is a neighborhood V of z0 and a neighborhood W
of 0, such that the equation f (z) = w has exactly k solutions in V for each
w ∈ W \ {0}.
7. At which points in the plane does the function f (z) = 2z 3 + 3z 2 − 12z + 6 have
a local analytic inverse?
8. Without using any knowledge of log functions, use Theorem 4.5.5 to decide at
which points z0 ∈ C it is true that the function f (z) = ez has a local analytic
inverse. Also, use Theorem 4.5.5 to compute the derivative of this inverse
function.
9. Prove Theorem 4.5.6.
10. Prove that if f is an analytic function defined on an open set U and if f is
one-to-one on U , then f has an analytic inverse function f −1 : W → U , where
W is the open set f (U ).
11. Prove that if f is an entire function with the property that f −1 (B) is bounded
whenever B is bounded, then f is onto (that is, f (C) = C). Hint: Show that
f (C) is both open and closed.

4.6. Homotopy
This section is devoted to developing tools for calculating the index of a closed
path (or more generally a cycle) about a point. This is crucial, since both the
hypotheses and conclusions of the Cauchy theorems involve the index. The tools
we will develop involve ideas from algebraic topology applied in the plane.
We begin by extending the index function to closed curves which are not nec-
essarily piecewise smooth. The key to doing this is to show that all closed paths
142 4. The General Cauchy Theorems

sufficiently near a given closed curve have the same index, and then to show that
every closed curve can be approximated arbitrarily closely by closed paths. The
common value of the index of all closed paths near a given closed curve γ is then
defined to be the index of γ. The key to the first part of this program is the fol-
lowing theorem, which is closely related to Rouché’s Theorem and has a similar
proof.
Theorem 4.6.1. Suppose z ∈ C and γ1 and γ2 are two closed paths which do not
pass through z and which satisfy
|γ1 (t) − γ2 (t)| ≤ |γ2 (t) − z|
for all t ∈ I. Then Indγ1 (z) = Indγ2 (z).

Proof. By translating γ1 and γ1 by z and using the fact, obvious from the definition
of Ind, that Indγ (z) = Indγ−z (0), we may assume that z = 0. Then our hypothesis
is that γ1 and γ2 are two paths which do not pass through 0 and which satisfy
|γ1 (t) − γ2 (t)| ≤ |γ2 (t)|.
We define a new path γ by
γ1 (t)
γ(t) = .
γ2 (t)
This path does not pass through 0, and it satisfies
|γ(t) − 1| ≤ 1.
In other words, γ(I) lies in the closed unit disc of radius 1 centered at 1 and does
not contain 0. It follows from this that 0 is in the unbounded component of C\γ(I).
Hence,
Indγ (0) = 0.
To complete the proof, we will show that
(4.6.1) Indγ (0) = Indγ1 (0) − Indγ2 (0).
In fact,
  b 
1 1 1 γ (t)
Indγ (0) = dz = dt,
2πi γ z 2πi a γ(t)
and so (4.6.1) follows from the calculation
γ γ  γ2 − γ1 γ2 γ γ
= 1 = 1 − 2.
γ γ1 γ2 γ1 γ2
This completes the proof. 

Index for Closed Curves. Recall that a curve is just a continuous function
γ : I → C, where I is a closed bounded interval – no differentiability is assumed.
For simplicity in the following discussion, we will just work with curves for which
the parameter interval I is [0, 1].
There is a notion of distance between two curves parameterized on I = [0, 1].
Definition 4.6.2. The distance between curves γ1 and γ2 on I is defined to be
||γ1 − γ2 || = sup |γ1 (t) − γ2 (t)|.
t∈I
4.6. Homotopy 143

With this notion of distance, we will show that every curve can be approximated
arbitrarily closely by curves which are piecewise smooth, in fact, piecewise linear.
Here, by a linear curve, we mean a curve of the form
γ(t) = u + tv
for fixed complex numbers u and v and t ranging over some parameter interval.
Such a curve traces a straight line segment. The linear curve which traces the line
segment from w to z and has parameter interval [a, b] is given by
b−t t−a
(4.6.2) γ(t) = w+ z
b−a b−a
(see Exercise 4.6.1).
A curve on I is piecewise linear if there is a partition 0 = t0 < t1 < · · · < tn = 1
of I such that γ is linear on each subinterval [tj−1 , tj ]. Obviously, piecewise linear
curves are piecewise smooth and, hence, are paths.
Theorem 4.6.3. If γ : I → C is a curve, then for each  > 0 there is a piecewise
linear curve γ̃ such that ||γ̃ − γ|| < .

Proof. Since γ is continuous and I is a closed bounded interval, γ is uniformly

coninuous on I. This means that, given  > 0, there is a δ > 0 such that
|f (s) − f (t)| <  whenever s, t ∈ I, |s − t| < δ.
We choose a partition 0 = t0 < t1 < · · · < tn = 1 of I such that tj − tj−1 < δ. By
(4.6.2) the linear curve γj , with parameter interval [tj−1 , tj ], which traces the line
segment from γ(tj−1 ) to γ(tj ) is given by
tj − t t − tj−1
γj (t) = γ(tj−1 ) + γ(tj ) for t ∈ [tj−1 , tj ]
tj − tj−1 tj − tj−1
Note that, for j = 1, · · · , n,
γj (tj ) = γ(tj ) = γj+1 (tj ).
This implies that the linear curves γj , defined on the subintervals [tj−1 , tj ], fit
together to form a continuous piecewise linear curve γ̃ on I = [0, 1]. Also, for each
j,
 
 tj − t t − tj−1 
|γ(t) − γj (t)| =  (γ(t) − γ(tj−1 )) + (γ(t) − γ(tj ))
tj − tj−1 tj − tj−1
tj − t t − tj−1
≤ |γ(t) − γ(tj−1 )| + |γ(t) − γ(tj )|
tj − tj−1 tj − tj−1
tj − t t − tj−1
≤ + =
tj − tj−1 tj − tj−1
if t ∈ [tj−1 , tj ]. Since this is true for every j and since γ̃(t) = γj (t) for t ∈ [tj−1 , tj ],
we conclude that ||γ − γ̃|| < . This completes the proof. 

The next theorem tells us that, given a closed curve γ and a point z not on
γ(I), all closed paths sufficiently near γ have the same index about z.
144 4. The General Cauchy Theorems

Theorem 4.6.4. If γ is a closed curve and z ∈ C, a point not on γ(I), then there
is a δ > 0 such that if γ1 and γ2 are paths with ||γ − γj || < δ for j = 1, 2, then
Indγ1 (z) = Indγ2 (z).

Proof. This is an application of Theorem 4.6.1. We choose

1
δ= inf |γ(t) − z|,
3 t∈I
so that
|γ(t) − z| ≥ 3δ for all t ∈ I.
If ||γ − γj || < δ for j = 1, 2, then
|γ(t) − γj (t)| < δ for all t ∈ I, j = 1, 2,
and so
|γj (t) − z| ≥ |γ(t) − z| − |γ(t) − γj (t)| ≥ 2δ for all t ∈ I, j = 1, 2.
It follows that
|γ1 (t) − γ2 (t)| ≤ |γ(t) − γ1 (t)| + |γ(t) − γ2 (t)| < 2δ ≤ |γ1 (t) − z|
for all t ∈ I. By Theorem 4.6.1, Indγ1 (z) = Indγ2 (z). 

This leads to the following definition.

Definition 4.6.5. Let γ be a closed curve in C and z ∈ C a point not on γ(I). We
define Indγ (z) in the following way: we choose a δ as in Theorem 4.6.4. Then all
closed paths within a distance δ of γ have the same index about z. By Theorem
4.6.3, there are closed paths within a distance δ of γ. We define Indγ (z) to be the
common value of Indγ1 (z) for all such paths γ1 .

With this definition, Theorem 4.6.4 implies that Indγ (z) is a locally constant
function of the closed curve γ. That is, the following theorem holds.
Theorem 4.6.6. If z ∈ C and γ is a closed curve in C such that z ∈ / Γ(I), then
there is a δ > 0 such that if γ1 is any other closed curve in C and ||γ − γ1 || < δ,
then
Indγ (z) = Indγ1 (z).

Proof. We choose δ as in Theorem 4.6.4. Then all closed paths within a distance δ
of γ have the same index around z, and this is the index of γ around z, by definition.
If γ1 is a closed curve with ||γ − γ1 || < δ, set
δ1 = δ − ||γ − γ1 ||.
Then all closed paths within a distance δ1 of γ1 are within a distance δ of γ, and
so they all have the same index about z as γ does. By definition, this is also the
index of γ1 about z. 
4.6. Homotopy 145

Homotopy. Let U be an open set in C and let γ0 and γ1 be two closed curves in
U with parameter interval [0, 1]. The curves are said to be homotopic if it is possible
to continuously deform γ0 to γ1 while remaining in U . This is made precise in the
following definition.
Definition 4.6.7. Two closed curves γ0 and γ1 in U , with parameter interval [0, 1],
are said to be homotopic in U if there is a continuous function h, defined on the
square [0, 1] × [0, 1] in R2 , with values in U , such that
(a) h(0, t) = γ0 (t) for all t ∈ [0, 1];
(b) h(1, t) = γ1 (t) for all t ∈ [0, 1]; and
(c) h(s, 0) = h(s, 1) for all s ∈ [0, 1].

In the above definition, for each fixed s, the function γs (t) = h(s, t) defines a
curve γs in U . Parts (a) and (b) say that when s = 0 and 1, these curves are the
original curves γ0 and γ1 . Part (c) says that each of the curves γs is closed.
The family of curves {γs } determined by a homotopy h, as in the above defini-
tion, is a continuous one-parameter family of curves in the sense that the following
theorem is true.
Theorem 4.6.8. If {γs } is the family of curves determined by a homotopy, as in
Definition 4.6.7, then for each s0 ∈ I and each  > 0, there is a δ > 0 such that
||γs − γs0 || <  whenever |s − s0 | < δ.

Proof. This proof uses the fact that a continuous function on a compact set such
as the square I × I, is uniformly continuous. For the continuous function h, this
means that given  > 0 there is a δ > 0, such that
|h(s, t) − h(s0 , t0 )| <  whenever |(s, t) − (s0 , t0 )| < δ,
for any pair of points (s, t), (s0 , t0 ) ∈ I × I. In particular, in the case where t = t0
this says that if |s − s0 | < δ, then
|γs (t) − γs0 (t)| <  for all t ∈ I,
which implies ||γs − γs0 || < , and this completes the proof. 

We are now in a position to prove the homotopy theorem for the index function.
Theorem 4.6.9. If γ0 and γ1 are two closed curves in U which are homotopic in
U , then
Indγ0 (z) = Indγ1 (z)
for every z ∈ C \ U .

Proof. Let {γs }, s ∈ I be the family of curves determined by a homotopy h joining

γ0 to γ1 in U . Since z ∈ C\U , none of these curves passes through z and so Indγs (z)
is defined for every s ∈ I.
If s0 ∈ I, then Theorem 4.6.6 implies that there is an  > 0 such that
Indγs (z) = Indγs0 (z) whenever ||γs − γs0 || < .
Then Theorem 4.6.8 implies that there is a δ > 0 such that
||γs − γs0 || <  whenever |s − s0 | < δ.
146 4. The General Cauchy Theorems

Figure 4.6.1. Some Paths γs for a Homotopy of a Circle to a Square.

We conclude that Indγs (z) is constant in the open interval (s0 − δ, s0 + δ). Since s0
is an arbitrary point of I, we conclude that the set on which Indγs (z) takes on any
given value is an open set. Since I is connected, and the union of these open sets
is I, there can be only one of them that is non-empty. Thus, Indγs (z) is constant
on I and, in particular, Indγ1 (z) = Indγ0 (z). 

If γ0 and γ1 are homotopic closed paths in an open set U , then the above
theorem implies that the cycle Γ = γ1 − γ0 has index 0 about every point of the
complement of U . Hence, Γ is homologous to 0. The general Cauchy Theorem then
says that
  
0= f (z) dz = f (z) dz − f (z) dz
Γ γ1 γ0

for every function f which is analytic on U . This proves the homotopy version of
Cauchy’s Theorem.

Theorem 4.6.10. If U is an open subset of C, γ0 and γ1 are homotopic closed

paths in U , and f is an analytic function on U , then
 
f (z) dz = f (z) dz.
γ1 γ0

Example 4.6.11. Prove that the circle of radius 2 centered at 0, with its standard
parameterization, is homotopic in C \ {0} to the piecewise linear curve which traces
once in the positive direction around the square with vertices at 1, i, −1, −i (see
Figure 4.6.1).
Solution: We parameterize the circle by γ0 (t) = 2 e2πit , t ∈ [0, 1] and the
square by
⎧
⎪
⎪ (1 − 4t) + 4ti, if 0 ≤ t ≤ 1/4;
⎪
⎨(4t − 1) + (2 − 4t)i, if 1/4 ≤ t ≤ 1/2;
(4.6.3) γ1 (t) =
⎪
⎪ (4t − 3) + (2 − 4t)i, if 1/2 ≤ t ≤ 3/4;
⎪
⎩
(4t − 3) + (4t − 4)i, if 3/4 ≤ t ≤ 1.
4.6. Homotopy 147

We define a homotopy h between the two by simply letting h(s, t) be the point
h(s, t) = (1 − s)γ0 (t) + sγ1 (t)
on the line segment joining γ0 (t) to γ1 (t). The resulting point h(s, t) is always on
or outside the boundary of the square and so it is never 0. For 0 ≤ t ≤ 1/4, the
formula for h is
h(s, t) = 2(1 − s) cos t + s(1 − 4t) + [2(1 − s) sin t + 4st]i.
The formula for h if t is in one of the other subintervals of the partition 0 < 1/4 <
1/2 < 3/4 < 1 is also easily calculated using (4.6.3).
Example 4.6.12. Show why the curves γ0 and γ1 of the previous exercise cannot
be homotopic in V = C \ {3/2}.
Solution: The point 3/2 is in the complement of V , and the curve γ0 has
index 1 about 3/2. However, the curve γ1 has index 0 about 3/2, since 3/2 is in
the unbounded component of C \ γ1 (I). By Theorem 4.6.9, the two curves cannot
be homotopic in V .
Simply Connected Sets.
Definition 4.6.13. A connected open set U is said to be simply connected if every
closed curve in U is homotopic to a point (that is, a constant curve).
Example 4.6.14. Prove that a convex open set U is simply connected.
Solution: Fix a point z0 ∈ U . If γ : I → U is a closed curve in U , we define a
homotopy between γ and the constant curve z0 as follows:
h(s, t) = (1 − s)γ(t) + sz0 .
Note that, for each t ∈ I, h(x, t) is on the line segment joining γ(t) to z0 and,
hence, is in U . Clearly, h is continuous on I × I, h(0, t) = γ(t), h(1, t) = z0 , and
h(s, 0) = h(s, 1), since γ(0) = γ(1). This establishes that h is a homotopy between
γ and the constant curve z0 . Since this can be done for every closed curve γ in U ,
the set U is simply connected.
We have talked about an open set in C possibly having a “hole”. What do we
mean by a hole in an open set? This is usually understood to mean a bounded
component of the closed set C \ U . However, we have not explored the ideas of
connectedness and connected components for closed sets so far in this text. A few
words concerning these concepts now will make the statement and proof of the next
theorem much easier to understand.
Recall that a set S is separated by a pair of open sets U , V if U ∩ S and V ∩ S
are both non-empty, have empty intersection, and have union equal to S. If there
is no pair of open sets which separate S, then S is connected. A component of S is
a maximal connected subset of S.
If the pair U , V separates S and S happens to be closed, then A = S \ (C \ V )
is closed and is equal to U ∩ S. Similarly, B = S \ (C \ U ) is closed and equal to
V ∩ S. Thus, a closed set S is separated by a pair of open sets U, V if and only
if it is separated by a pair of its closed subsets A, B in the sense that A, B is a
disjoint pair of closed, non-empty subsets of S with union equal to S. In this case,
A = U ∩ S and B = V ∩ S.
148 4. The General Cauchy Theorems

In the next theorem we will need the following lemma concerning components
and separation of a close subset of C. The proof is left to the exercises.
Lemma 4.6.15. Let S be a closed subset of C. If C is a component of S, then C
is bounded if and only if C is contained in a closed bounded subset A of S such that
B = S \ A is also closed.

Note that the set B of the above lemma could be empty. If it is empty, then S
itself is bounded. If it is not empty, then the pair A, B separates S.
The next theorem shows that simply connected open sets are connected open
sets with no holes and that they are exactly the sets on which all the things we
would like to be true in complex analysis are, in fact, true.
Theorem 4.6.16. Let U be a connected open subset of C. Then the following
statements are equivalent.
(a) U is simply connected;
(b) every cycle in U is homologous to 0;
(c) U has no holes, that is, C \ U has no bounded components;
(d) Γ f (z) dz = 0 for every cycle Γ in U and every analytic function f on U ;
(e) every analytic function on U has an antiderivative;
(f) every harmonic function on U has a harmonic conjugate;
(g) every non-vanishing analytic function f on U has an analytic logarithm;
(h) every non-vanishing analytic function f on U has an analytic square root.

Proof. We can only give an incomplete proof at this point. We will prove that each
of the statements (a) through (h) implies the next statement on the list. However,
the proof that (h) implies (a) will have to wait until we prove the Riemann Mapping
Theorem in Chapter 6.
A constant path γ(t) ≡ c in U has γ  (t) ≡ 0, and so its index about any point
other than c is 0. In particular, it is homologous to 0 in U . Since each cycle is
equivalent to a sum of closed paths, it follows from Theorem 4.6.9 that (a) implies
(b).
We next prove that (b) implies (c). Suppose (c) fails, that is, suppose there is a
bounded component C of S = C \ U . Then, by the previous lemma, C is contained
in a closed bounded subset A of S such that B = S \ A is also closed.
The set A is compact. It is contained in some bounded open rectangle R as
well as in the open set C \ B. Thus, A is contained in the bounded √ open set
V = R \ (C \ B). We fix a point z0 ∈ A. Let δ > 0 be chosen so that 2 δ is smaller
than the distance from A to the complement of V . Suppose the open rectangle R
is defined by inequalities
a < Re(z) < b, c < Im(z) < d.
Let a = s0 < s1 < · · · < sn = b and c = t0 < t1 < · · · < tn = d be partitions of
[a, b] and [c, d] into subintervals of length at most δ. We are careful to choose these
so that Re(z0 ) is not one of the sj ’s and Im(z0 ) is not one of the tk ’s.
Now each of the rectangles
Rjk = {z : sj−1 ≤ Re(z) ≤ sj , tk−1 ≤ Im(z) ≤ tk }
4.6. Homotopy 149

U
A
Γ1
V

Figure 4.6.2. Creating a Cylce Γ1 Surrounding a Hole.

which meets A also lies in V . Let Γ be the cycle which is the sum of all the paths
∂Rjk for which Rjk ∩ A = ∅. Now if an edge of one of these rectangles meets A,
then this edge belongs to two of the rectangles whose boundaries make Γ. The edge
occurs with opposite orientation in the two paths and, hence, their contributions
will cancel out in any integral around Γ as well as in the expression for ∂Γ. This
means that if Γ1 is the chain which is left after all edges which meet A are thrown
out of Γ, then Γ1 is still a cycle, and if f is a function continuous on Γ(I), then
 
(4.6.4) f (z) dz = f (z) dz.
Γ Γ1

Note that Γ1 (I) is contained in V \ A which is contained in U (see Figure 4.6.2).

Since z0 is in a rectangle Rpq whose boundary contributes to Γ, but z0 is not on
this boundary, we may calculate the index of Γ around z0 . By the Cauchy Integral
Theorem, the contributions of the paths ∂Rjk different from ∂Rpq are all 0 and so
 
1 1 1 1
IndΓ (z0 ) = dz = dz = 1.
2πi Γ z − z0 2πi ∂Rpq z − z0

Thus, in view of (4.6.4), we have

IndΓ1 (z0 ) = 1.

Since Γ1 is a cycle in U , we conclude that (b) fails. This completes the proof that
(b) implies (c).
To show that (c) implies (d), suppose γ is a closed path in U . If V is a bounded
component of C\γ(I), then A = V ∩(C\U ) is a closed set since it is the intersection
of C \ U with the complement of the union of the other components of C \ γ(I).
The set B = (C \ U ) \ A is also closed since it is equal to C \ (U ∪ V ). By Lemma
4.6.15, if A is non-empty, then it contains a bounded component of C \ U . Thus,
if (c) holds, then A must be empty – that is, V ∩ (C \ U ) = ∅ for every bounded
component V of C \ γ(I). Then z ∈ C \ U implies z is in the unbounded component
of C\γ(I) and so Indγ (z) = 0. The hypotheses of the Cauchy Theorem are satisfied
and so (d) holds. Thus, (c) implies (d).
150 4. The General Cauchy Theorems

If (d) holds, then we may define an antiderivative g for the analytic function f
on U by fixing a point z0 ∈ U and setting

g(z) = f (w) dw,
γz

where γz is any path in U beginning at z0 and ending at z. That this is independent

of the path chosen follows from (d). That it yields an antiderivative for f follows
from Theorem 2.6.1. Thus, (d) implies (e).
That (e) implies (f) follows if we notice that, in the proof of Theorem 3.5.7,
the only use of the hypotheses that U was convex was to ensure that an analytic
function has an antiderivative.
If f is analytic and non-vanishing on U , then log |f | is a harmonic function on
U . Thus, if (f) is true, then log |f | is the real part of an analytic function g on U .
Then eg and f are two analytic functions on U with the same modulus |f |. This
means f e−g has constant modulus and, hence, is a constant a, by the Maximum
Modulus Theorem. If we choose a logarithm b for the non-zero number a, then
a = eb and
f = eg+b .
In other words, g + b is an analytic logarithm for f . This proves that (f) implies
(g).
If f has an analytic logarithm h, then eh/2 is an analytic square root for f .
Thus, (g) implies (h).
We have now proved that each of the statements (a) through (g) implies the
next statement on the list. To complete the proof we must show that (h) implies
(a). This is done in Corollary 6.4.6 of Chapter 6. 

Example 4.6.17. Prove that U = C \ D1 (0) is not simply connected.

Solution: The point 0 is in the complement of U and γ(t) = 2 e2πit defines
a curve in U which has index 1 about 0. Thus, γ is a closed curve in U which is
not homologous to 0. Since (c) of the preceding theorem fails to hold for U , we
conclude that U is not simply connected.

Homotopy for Non-Closed Curves. There is a type of homotopy for non-closed

curves that have a fixed starting point z0 and a fixed ending point z1 .
Definition 4.6.18. If γ0 and γ1 are curves in U , parameterized on I = [0, 1],
and if γ0 (0) = γ1 (0) = z0 and γ0 (1) = γ1 (1) = z1 , then γ0 and γ0 are said to be
homotopic curves joining z0 to z1 in U if there exists a continuous function
h:I ×I →U
such that
h(0, t) = γ0 (t), for all t ∈ I,
h(1, t) = γ1 (t), for all t ∈ I,
h(s, 0) = z0 , for all s ∈ I, and
h(s, 1) = z1 , for all s ∈ I.
4.6. Homotopy 151

As with a homotopy for closed curves, a homotopy of the above type determines
a continuous one-parameter family of curves {γs } by
γs (t) = h(s, t).
In this case, the curves in the family all begin at z0 and end at z1 . For this type
of homotopy, we have the following results, which follow easily from the preceding
material, and whose proofs are left as exercises.
Theorem 4.6.19. If γ0 and γ1 are homotopic paths in U connecting z0 to z1 , then
 
f (z) dz = f (z) dz
γ0 γ1
for every function f which is analytic in U .
Theorem 4.6.20. Let U be a connected open set and let z0 , z2 be points of U .
Then U is simply connected if and only if any two curves in U connecting z0 to z1
are homotopic in U .

Exercise Set 4.6

1. Show that the path defined by (4.6.2) is, as claimed in the text, a linear curve,
with parameter interval [a, b], which traces the line segment from w to z.
2. Find ||γ − γ1 || if γ(t) = e2πit , and γ1 (t) = 2 cos(2πt) + 3i sin(2πt), for t ∈ [0, 1].
3. If γ(t) = e2πit for t ∈ [0, 1], describe a piecewise linear path γ1 such that
||γ − γ1 || < π/8.
4. Show that the curves γ and γ1 of Exercise 2 are homotopic in C\{0} by finding
an explicit homotopy in C \ {0} between them.
5. Explain why the curves γ and γ1 of the previous exercise cannot be homotopic
in the set C \ {2i}.
6. Which of the following open sets in C are simply connected:
(a) C \ {0}, (b) C \ [−∞, 0],
(c) D2 (0) \ [−1, 1], (d) D1 (0) \ [0, 1),
(e) C \ ({0} ∪ {1}), (f) D2 (0) \ D1 (0)?
7. For the set C \ {0}, show by example that (b), (c), and (d) of Theorem 4.6.16
all fail.
8. Also show by example that (e), (f), and (g) of Theorem 4.6.16 fail for the set
C \ {0}.
9. Prove Theorem 4.6.19.
10. Prove Theorem 4.6.20.
11. Prove that if U and V are homeomorphic open subsets of C, and if U is
simply connected, then so is V . Here, U and V are homeomorphic if there is a
one-to-one continuous function f of U onto V which has a continuous inverse
function.
12. Prove Lemma 4.6.15 – that is, prove that if S is a closed subset of C and C
is a component of S, then C is bounded if and only if it contains a bounded
closed subset A for which S \ A is also closed.
 Chapter 5

Residue Theory

The Residue Theorem (Theorem 4.4.3) has a wide range of applications. This
chapter is devoted to exploring some of them. We begin with a section on techniques
for computing residues.

5.1. Computing Residues

Recall the discussion of residues in Section 4.4. If f is analytic in a neighborhood
of z0 , except at z0 itself, then f has an isolated singularity at z0 . In this case, it
has a Laurent expansion
∞

f (z) = an (z − z0 )n
n=−∞

in D \ {z0 } for some disc D, centered at z0 . The residue of f at z0 , denoted

Res(f, z0 ), is then the coefficient a−1 of (z − z0 )−1 in this expansion. Computing
the residue can be easy. It can also be hard.
If the singularity of f at z0 is a pole of order k, then there is a formula for
Res(f, z0 ) which makes it, in principle, computable. If f has a pole of order k at
z0 , then it has the form
g(z)
f (z) = ,
(z − z0 )k
where g is analytic and non-vanishing in a neighborhood of z0 .
g(z)
Theorem 5.1.1. If f (z) = , where g is analytic in a neighborhood of z0 ,
(z − z0 )k
then Res(f, z0 ) is the coefficient of degree k − 1 in the power series expansion of g
about z0 . That is,
g (k−1) (z0 )
Res(f, z0 ) = .
(k − 1)!

153
154 5. Residue Theory

Proof. Since g is analytic in a neighborhood of z0 , Theorem 3.2.5 implies that it

has a power series expansion
∞

g(z) = bn (z − z0 )n ,
n=0

where, by Theorem 3.2.1,

g (n) (z0 )
bn = .
n!
The Laurent expansion of f is then given by
∞

f (z) = b0 (z − z0 )−k + · · · + bk−1 (z − z0 )−1 + bn (z − z0 )n−k .
n=k

Thus, the residue of f at z0 is

g k−1 (z0 )
Res(f, z0 ) = bk−1 = ,
(k − 1)!
as claimed. 

The first couple of cases of the preceding theorem are worth highlighting. We
do this in the following corollary, which follows immediately from the theorem.
Corollary 5.1.2. Given a function g, analytic in a neighborhood of z0 ,
g(z)
(a) if f (z) = , then Res(f, z0 ) = g(z0 );
z − z0
g(z)
(b) if f (z) = , then Res(f, z0 ) = g  (z0 ).
(z − z0 )2
z2 + 1
Example 5.1.3. Find Res(f, 1) if f (z) = .
(z − 1)(z 2 − 2z + 5)
Solution: The function f is of the form
g(z)
f (z) = ,
z−1
where
z2 + 1
g(z) = .
z 2 − 2z + 5
Since g is analytic in a neighborhood of 1, Corollary 5.1.2(a) implies that
1
Res(f, 1) = g(1) = .
2
sin z
Example 5.1.4. Find Res(f, 0) if f (z) = .
z2
Solution: This is clearly a situation where Corollary 5.1.2(b) applies, with
g(z) = sin z. Thus,
Res(f, 0) = g  (0) = cos 0 = 1.
We could also have done this one directly, by writing out the power series expansion
for sin z about 0, and dividing by z 2 to obtain the Laurent series expansion of f .
5.1. Computing Residues 155

Long Division. Often the function f , whose residue we wish to compute at

some z0 , is of the form
p(z)
f (z) = ,
h(z)
where p and h are analytic functions in a neighborhood of z0 with known power
series expansions about z0 , and h has a zero of order k at z0 . In this situation, we
can always derive a formula for Res(f, z0 ) using Theorem 5.1.1. We simply write h
as h(z) = (z − z0 )k q(z), where q is analytic in a neighborhood of z0 and q(z0 ) = 0.
Then
g(z) p(z)
(5.1.1) f (z) = , where g(z) = .
(z − z0 )k q(z)
We then apply Theorem 5.1.1. Of course, to apply this theorem, one has to com-
pute g (k−1) (z0 ) (or, equivalently, the coefficient of (z − z0 )k−1 in the power series
expansion of g about z0 ), and this may be quite difficult. Rather than doing this
directly, through repeated differentiation, it is possible to compute the power se-
ries coefficients of g using power series methods – specifically, the method of long
division of power series.
Note that p and q are both analytic functions in a neighborhood of z0 and,
since the power series expansions of p and h about z0 are known, the same thing
is true of the power series expansion of q about z0 . Since q(z0 ) = 0, g = p/q is also
analytic in a neighborhood of z0 and, hence, has a power series expansion in some
disc centered at z0 . The method of long division is an algorithm for finding the
power series coefficents of p/q in terms of those of p and q.
For simplicity, in our discussion of long division, we will work with power series
in z – that is, power series centered at 0. Power series centered at some point
z0 other than 0 can always be put in this form with a change of variables. The
coefficient formulas we derive below are not affected by such a change.
Suppose that p has a zero of order at least m at 0 (m might be 0 and it might
be less than the actual order of the zero of p at 0). Also suppose that q(0) = 0.
Then p and q have power series expansions
p(z) = am z m + am+1 z m+1 + am+2 z m+2 + · · · + an z n + · · · ,
(5.1.2)
q(z) = b0 + b1 z + b2 z 2 + · · · + bn z n + · · · ,
which converge in some disc Dr (0), with r > 0. Since q(0) = b0 = 0, q(z) is non-
vanishing in some disc Dδ (0), with 0 < δ < r. Then the quotient p/q is analytic
in Dδ (0), with a zero of order at least m at 0, and has a convergent power series
expansion
p(z)
(5.1.3) = cm z m + cm+1 z m+1 + cm+2 z m2 + · · · + cn z n + · · · .
q(z)
How can we determine the coefficients cn if we know the coefficients an and
bn ? A method often taught in calculus for doing this is the method of long division
of power series. This is a recursive method that determines the coefficients an one
after another. It is described as follows: We have
p(z) am m p(z) − (a0 /b0 )z m q(z)
(5.1.4) = z + ,
q(z) b0 q(z)
156 5. Residue Theory

where the fraction on the right has the same denominator as the original fraction,
but the numerator has changed and is now a power series with lowest order term
of degree at least m + 1, whereas the original numerator had lowest order term of
degree at least m. If we write out explicitly the polynomials involved, then 5.1.4
becomes
p(z) am z m + am+1 z m+1 + am+2 z m+2 + · · ·
=
q(z) b0 + b1 z + b2 z 2 + · · ·
(5.1.5)
am m (am+1 − b1 am /b0 )z m+1 + (am+2 − b2 am /b0 )z m+2 + · · ·
= z + .
b0 b0 + b1 z + b2 z 2 + · · ·
This tells us that the first coefficient in (5.1.3) is
(5.1.6) cm = am /b0 .
The next coefficient cm+1 is obtained by repeating the procedure on the new fraction
that appears on the right in the above equation. This procedure may be repeated
as often as is needed to obtain a given coefficient cn . For example, in Exercises
5.1.16 and 5.1.17 you are asked to use this procedure to show that
am+1 am b1
(5.1.7) cm+1 = − 2
b0 b0
and
am+2 am+1 b1 am b21 am b2
(5.1.8) cm+2 = − + − 2 .
b0 b20 b30 b0

Residue of a Quotient. We now return to the problem of finding the residue

at z0 of a quotient, f = p/h, where h has a zero of order k at z0 . We define q and
g = p/q as in (5.1.1). It follows from Theorem 5.1.1 that:
p(z)
Theorem 5.1.5. Suppose f (z) = , where p(z) and h(z) are analytic in a
h(z)
neighborhood of z0 and h has a zero of order k at z0 . If we write h(z) = (z−z0 )k q(z)
where q is analytic in a neighborhood of z0 and q(0) = 0, then
Res(f, z0 ) = ck−1 ,
where ck−1 is the coefficient of the degree k − 1 term in the power series expansion
of g = p/q about z0 . The numbers ck−1 , may be computed using long division on
the quotient p/q; that is, they may be computed through repeated use of (5.1.5).

For small k, the expression for ck−1 , given by carrying out the method described
in the above theorem, is reasonably simple. However, the complexity increases
rapidly with increasing k.
When k = 1, the above theorem and (5.1.6), with m = 0, say that if
p(z) p(z)
f (z) = = ,
h(z) (z − z0 )q(z)
with q(z0 ) = 0, then
a0 p(z0 )
Res(f, z0 ) = c0 = = .
b0 q(z0 )
5.1. Computing Residues 157

This is just Corollary 5.1.2(a). We can restate it in terms of the original functions
p and h as follows: since h(z) = (z − z0 )q(z), we have
q(z0 ) = h (z0 )
and so
Corollary 5.1.6. Let p and h be analytic in a neighborhood of z0 , where h has a
zero of order 1 at z0 . Then
p(z0 )
Res(p/h, z0 ) = .
h (z0 )
Example 5.1.7. Find Res(f, 0) if
ez
f (z) = .
sin z
Solution: Since sin z has a zero of order 1 at z = 0, we may apply Corollary
5.1.6 with p(z) = ez and h(z) = sin z. Then
e0
Res(f, 0) = = 1.
cos 0
Example 5.1.8. Find Res(f, 0) if
1
f (z) = .
ez −1 − z
Solution: Here the denominator has a zero of order 2. We set p(z) = 1 and
z2 z3
h(z) = ez −1 − z = + + ··· .
2 3!
Then h(z) = z 2 q(z) where
1 z
q(z) = + + ··· .
2 3!
By Theorem 5.1.5, the residue we seek is the coefficient of the degree 1 term in the
power series expansion of 1/q, and this may be obtained by long division. In fact,
we have
1 z/3 + · · ·
=2− = 2 − (2/3)z + · · · ,
1/2 + z/3! + · · · 1/2 + z/3! + · · ·
and so the residue is −2/3.
cot z
Example 5.1.9. Find the residue at 0 of f (z) = .
z2
Solution: We proceed as in Theorem 5.1.5. The function f has a pole of order
3 at 0. We set p(z) = cos z, h(z) = z 2 sin z, and
h(z) sin z z2 z4
q(z) = 3
= =1− + − ··· .
z z 6 120
We then use long division to find the first three power series coefficients, c0 , c1 and
c2 of p/q. By Theorem 5.1.5, the residue we seek will then be the coefficient c2 . We
158 5. Residue Theory

have,

p(z) cos z 1 − z 2 /2 + · · ·
= −1 =
q(z) z sin z 1 − z 2 /6 + · · ·
(−1/2 + 1/6)z 2 + · · ·
=1+
1 − z 2 /6 + · · ·
−(1/3)z 2 + · · ·
=1+
1 − z 2 /6 + · · ·
= 1 − (1/3)z 2 + · · · .

Thus, c0 = 1, c1 = 0, and c2 = −1/3. We conclude that Res(f, 0) = −1/3.

Exercise Set 5.1

z+2
1. Find Res(f, 0) and Res(f, 1) if f (z) = .
z(z − 1)(z + 4)
cos z
2. Find Res(f, 0) and Res(f, 2) if f (z) = .
2z − z 2
1/z
3. Find Res(f, 0) if f (z) = e .
4. Find Res(f, 0) if f (z) = cot z.
5. Find Res(f, π) if f (z) = cot z.
6. Find Res(f, 0) if f (z) = cot2 z.
ez
7. Find Res(f, 0) if f (z) = .
cos z − 1
1
8. Find Res(f, 0) if f (z) = 2 .
z sin z
1
9. Find Res(f, 0) if f (z) = .
z − log(1 + z)
sinh z
10. Find Res(f, 0) if f (z) = .
z4
(log z)3
11. Find Res(f, i) if f (z) = 2 .
z +1
12. Use long division of power series to find the power series expansion about 0 of
sin z
tan z = through terms of degree five.
cos z
13. Prove that if f has a simple pole at z0 and g is analytic in a neighborhood of
z0 , then Res(gf, z0 ) = g(z0 ) Res(f, z0 ).
14. Show by example that it is not true that Res(gf, z0 ) = g(z0 ) Res(f, z0 ) if g is
analytic in a neighborhood of z0 and f has a pole of degree greater than 1 at
z0 .
15. Prove that a meromorphic function which is even and has an isolated singu-
larity at 0 has residue 0 at 0.
16. Derive formula (5.1.7).
17. Derive formula (5.1.8).
5.2. Evaluating Integrals Using Residues 159

5.2. Evaluating Integrals Using Residues

The Residue Theorem (Theorem 4.4.3) tells us that if f is a function which is
analytic in a set U , except at a discrete set of isolated singularities, then we can
evaluate its integral around a closed path γ in U , which is homologous to 0 in U
and does not hit any of these singularities, provided we can find the residues of f
at singularities where γ has non-zero index. As we shall see, this turns out to be a
very practical method for evaluating a wide variety of integrals.

Sines and Cosines. Because of the identities

eiθ + e−iθ eiθ − e−iθ
cos θ = , sin θ = ,
2 2i
the cosine and sine functions may be regarded as the restrictions to the unit circle
(parameterized by z = eiθ , 0 ≤ θ ≤ 2π) of the meromorphic functions
z + z −1 z − z −1
and .
2 2i
This means that an integral of the form
 2π
(5.2.1) f (θ) dθ,
0
where f is a rational expression in cos θ and sin θ, may be reformulated as a path
integral around the unit circle of a certain meromorphic function of z. We simply
let g(z) be the function of z obtained by replacing the functions cos θ and sin θ,
which make up f , with the functions (z + z −1 )/2 and (z − z −1 )/2i. It will then be
true that
g(eiθ ) = f (θ).
With γ(θ) = eiθ , we then have
  2π
g(z)
(5.2.2) dz = f (θ) dθ.
γ iz 0

The integral on the left is 2πi times the sum of the residues of g(z)/iz inside the
unit circle, provided g(z)/iz has no poles on the unit circle itself.
 2π
dθ
Example 5.2.1. Find .
0 2 + sin θ
Solution: If we replace sin θ in the integrand by (z − z −1 )/2i, the resulting
function is
1 2iz
= .
2 + (z − z −1 )/2i 4iz + z 2 − 1
This is the g of equation (5.2.2). We divide this by iz and integrate around the
unit circle to get our answer. Thus, we need to evaluate the integral
 
2 2
2 + 4iz − 1
dz = √ √ dz.
|z|=1 z |z|=1 (z + (2 − 3)i)(z + (2 + 3)i)
Here we have factored z 2 + 4iz − 1 by finding its roots using the Quadratic Formula.
The only pole of the integrand that is inside the unit circle is the one at the point
160 5. Residue Theory

√
(−2 + 3)i. The residue at this pole is found using Corollary 5.1.2(a). That is, we
simply evaluate the function
2
√
z + (2 + 3)i
√
at the pole z = (−2 + 3)i. The resulting residue is
√
2 3
√ =− i.
2 3i 3
By the Residue Theorem, the integral we seek is this number times 2πi. Thus,
 2π √
dθ 2 3
= π.
0 2 + sin θ 3

This technique may be used to evaluate any integral of the form (5.2.1), where
f (θ) is a rational function of sin θ and cos θ with a denominator which does not
vanish for θ ∈ [0, 2π].

Improper Integrals. Improper integrals of the form

 ∞
(5.2.3) f (t) dt
−∞

can often be evaluated using residues. Here we are assuming that f is a function
which is Riemann integrable on each finite subinterval [a, b] of R.
There are two senses in which such an integral may converge. If the expression
 x
f (t) dt
−y

has a limit as x and y approach +∞ independently, then the improper integral is

said to converge. This means, given  > 0, there is an M such that the integral
is within  of the limit whenever x ≥ M and y ≥ M . This is equivalent to the
convergence of the two one-sided improper integrals
 ∞  0
f (t) dt and f (t) dt.
0 −∞

On the other hand, if the limit of the symmetric integral,

 x
lim f (t) dt,
x→∞ −x

exists, then the integral (5.2.3) is said to converge in the principal value sense and
the above limit is called the principal value of the integral. This is weaker than
ordinary convergence of the integral – that is, (5.2.3) may converge in the principal
value sense and not converge in the ordinary sense.
An improper integral (5.2.3) is said to converge absolutely if the integral of
|f | converges. This implies the convergence of (5.2.3) and, in fact, we have the
following integral analogue of the comparison test for convergence of series.
5.2. Evaluating Integrals Using Residues 161

Theorem 5.2.2. If f and g are continuous functions on the line, with g(t) ≥ 0
and |f (t)| ≤ g(t) for all t ∈ R, then the integral of f on R converges and
 ∞   ∞
 
 f (t) dt ≤ g(t) dt.

−∞ −∞
provided the integral of g on R converges.

Proof. Assuming the integral of f on R converges, we will

 0 prove the convergence of
∞
the improper integral 0 f (t) dt. The convergence of −∞ f (t) dt follows from the
same argument. We will use the fact that, if a function h on (0, ∞) has the property
that limn→∞ h(xn ) exists for every increasing sequence {xn } converging to infinity,
then the limits of all such sequences are the same number L and limx→∞ h(x) = L
(Exercise 5.2.11).
Let {xk }∞
k=1 be an increasing sequence of positive numbers converging to infin-
ity. We set x0 = 0 and, for k ≥ 1, set
 xk  xk
ak = f (t) dt, bk = g(t) dt.
xk−1 xk−1

Then the hypothesis that |f (t)| ≤ g(t) implies that |ak | ≤ bk for all n. Furthermore,
for each positive integer n,
 xn 
n  xn n
f (t) dt = ak and g(t) dt = bk .
0 k=1 0 k=1

Thus, the convergence of the ∞improper integral of g(t) implies the convergence of
the series of positive
 terms k=1 bk . This, in turn, implies the absolute convergence
of the series ∞ a
k=1 k and the inequality
∞ 
   ∞
 
(5.2.4)  ak  ≤ bk .
 
k=1 k=1
x
This means that, if h(x) = 0 f (t) dt, then the sequence {h(xn )} has a limit for ev-
ery increasing sequence {xn } converging to infinity. By the remark in the first para-
graph, limx→∞ h(x) exists and is the common limit of all the sequences {h(xn )}.
Hence the improper integral of f exists and is, by definition, this number L.
It follows from (5.2.4) that
 ∞  ∞
  ∞
 
 f (t) dt  = |L| ≤ b = g(t) dt.
  k
0 k=1 0

0
The same argument can be used to show that −∞ f (t) dt exists and satisfies
the inequality  0   0
 
 f (t) dt ≤ g(t) dt.
 
−∞ −∞
The theorem follows on combining these results. 

If a function f is Riemann integrable on each finite subinterval of the real line,

and its improper integral over (−∞, ∞) converges absolutely, then we will say that
f is absolutely integrable on (−∞, ∞).
162 5. Residue Theory

−r r

Figure 5.2.1. The Path γr .

Evaluating Integrals on the Line. Often the integrand f of an integral of the

form (5.2.3) is the restriction to the real line of a function which is analytic, except
for a discrete set of singularities, on an open set containing either the closed upper
half-plane {z : Im(z) ≥ 0} or the closed lower half-plane {z : Im(z) ≤ 0}. If this
function has no singularities on the real line and it decreases rapidly enough as
z → ∞, then the integral can be approximated arbitrarily closely by the integral of
f around a semicirclular closed path in the appropriate half-plane. Such an integral
can be evaluated using residues.
Specifically, suppose f is analytic in U , except on a discrete subset of U , where
U is an open set containing the closed upper half-plane. Suppose also that p, R
and C are positive numbers, with p > 1, such that, for Im(z) ≥ 0,
(5.2.5) |f (z)| ≤ C|z|−p when |z| > R.
This ensures that the improper integral (5.2.3) converges (Exercise 5.2.12).
Let γr be the path which begins at −r on the real line, traverses the interval
[−r, r] and then returns to −r along the semicircle which is the upper half of the
circle |z| = r (see Figure 5.2.1).
We have
  r  π
(5.2.6) f (z) dz = f (x) dx + f (r eit )ir eit dt.
γr −r 0

If r > R, then (5.2.5) implies that |f (r e )ir e | ≤ Cr 1−p and so

it it
 π 
 
 f (r eit
)ir e it
dt  ≤ πCr 1−p .
 
0
Since p > 1, the right side of this inequality has limit 0 as r → ∞. We conclude
from this and (5.2.6) that
  r  ∞
lim f (z) dz = lim f (x) dx = f (x) dx.
r→∞ γr r→∞ −r −∞

Now, by the Residue Theorem, γr f (z) dz is 2πi times the sum of the residues
of f at singularities inside γr . The condition (5.2.5) ensures that there are no
singularities of f in the upper half-plane outside the circle |z| = R. This implies
there are only finitely many singularieties in the upper half-plane. It also implies
5.2. Evaluating Integrals Using Residues 163

that the integral is independent of r, for r > R, and is equal to 2πi times the sum
of the residues of f at this finite set of singularities.
If f is analytic in an open set containing the lower half-plane, then all of the
above still holds with the following changes: the lower half-plane replaces the upper
half-plane, the path γr is replaced by its reflection through the x-axis, and, since
the new path has negative orientation, the resulting integral is −2πi times the sum
of the residues of f in the lower half-plane.
This proves the following theorem:
Theorem 5.2.3. Let H be either the closed upper or lower half-plane. If f is
analytic, except at a discrete set of singularities, in an open set containing H, has
no singularities on R, and there are positive numbers R, C, and p > 1 such that
(5.2.5) is satisfied for z ∈ H, then
 ∞ 
m
f (x) dx = σ(H)2πi Res(f, zj ),
−∞ j=1

where z1 , z2 , · · · , zm are the singularities of f in the half-plane H, and σ(H) = 1 if

H is the upper half-plane and −1 if H is the lower half-plane.
 ∞
x2
Example 5.2.4. Find 4
dx.
−∞ 1 + x
Solution: We use the previous theorem with
z2 z −2
f (z) = 4
= −4 .
1+z z +1
If we fix an R > 1, we have
|z|−2
|f (z)| ≤ if |z| ≥ R,
1 − R−4
and so condition (5.2.5) is satisfied with p = 2 and C = (1 − R−4 )−1 .
The poles of f occur at the 4th roots of −1, and these are
√ √
2 2
z1 = (1 + i), z2 = (−1 + i),
√2 2√
2 2
z3 = (−1 − i), z4 = (1 − i).
2 2
Thus
z2
f (z) = .
(z − z1 )(z − z2 )(z − z3 )(z − z4 )
Only z1 and z2 are in the upper half-plane. We evaluate the residue of f at each
of these points using Corollary 5.1.2(a). This means, for j = 1, 2, we evaluate at zj
what is left of f when z − zj is removed from the denominator. The results are
√
√ (1 + i)2 2
Res(f, z1 ) = 2 =− (1 + i),
(2)(2 + 2i)(2i) 8
√
√ (−1 − i)2 2
Res(f, z2 ) = 2 = (1 − i).
(−2)(2i)(−2 + 2i) 8
164 5. Residue Theory

√
2
The sum of these is − i and, when multiplied by 2πi, this yields
4
 ∞ √
x2 2
4
dx = π.
−∞ 1 + x 2

Symmetries. It can be useful to exploit symmetries of the integrand in com-

puting integrals using residues.
 ∞
1
Example 5.2.5. Find dx.
0 (1 + x2 ) 2
Solution: Since the integrand is an even function, its integral over the entire
real line is twice its integral from 0 to ∞. Thus,
 ∞ 
1 1 ∞ 1
2 2
dx = dx.
0 (1 + x ) 2 −∞ (1 + x2 )2
We evaluate the latter integral using Theorem 5.2.3. We set
1 1
f (z) = = .
2
(1 + z ) 2 (z − i) (z + i)2
2

The only pole in the upper half-plane is the one at i, and this is a pole of order
2. Corollary 5.1.2(b) applies, and it tells us that the residue of f at this pole is
obtained by differentiating
1
g(z) = ,
(z + i)2
and evaluating at z = i. The result is
−2 i
g  (i) = 3
=− .
(2i) 4
Thus,  ∞
1 1 −i π
2 2
dx = · 2πi · = .
0 (1 + x ) 2 4 4
In the next example, we exploit the fact that the integrand is unchanged if we
rotate coordinates by an angle π/n.
 ∞
1
Example 5.2.6. Find dx.
0 1 + x2n
Solution: The integrand is even, so we could just proceed as in the preceding
example, replacing this integral by one over all of R and then applying Theorem
5.2.3. However, this would mean evaluating residues at each of the n poles in the
upper half-plane. There is a better way. We set
1
f (z) =
1 + z 2n
and note that
(5.2.7) f (eπi/n z) = f (z)
for each z ∈ C. This means f has the same values along the ray {eπi/n x : 0 ≤ x}
as it does along the ray [0, +∞). To exploit this, we choose r > 1 and use the path
γr which traverses the boundary of the sector indicated in Figure 5.2.2. That is,
5.2. Evaluating Integrals Using Residues 165

r e πi/n

e πi/2n

Figure 5.2.2. Path for Example 5.2.6.

the path γr traverses the interval [0, r], then the arc r eit , 0 ≤ t ≤ π/n, and then
returns to 0 along the interval [r eπi/n , 0].
Then
  r  π/n  r
f (z) dz = f (x) dx + f (r e )ri e dt −
it it
f (eπi/n x) eπi/n dx
γr 0 0 0
 r  π/n
= (1 − eπi/n ) f (x) dx + f (r eit )ri eit dt.
0 0
The second integral on the right clearly goes to 0 as r → ∞ and the integral on the
left is 2π Res(f, z1 ), where z1 = eπi/2n is the only pole of f inside γr . Thus,
 ∞
Res(f, z1 )
f (x) dx = 2πi .
0 1 − eπi/n
All that remains is to evaluate this residue.
Since z12n = −1, we have
z 2n + 1 = z 2n − z12n = (z − z1 )(z 2n−1 + z 2n−2 z1 + · · · + z12n−1 ).
Corollary 5.1.2(a) says that, to find the residue of f at z1 , we simply evaluate the
inverse of the second factor on the right at z = z1 . The result is
1 z1 eπi/2n
Res(f, z1 ) = 2n−1 = − 2n = − 2n ,
2nz1
and this implies
 ∞
1 2πi eπi/2n π/2n
dx = − = .
0 1 + x2n 2n 1 − eπi/n sin(π/2n)

Exercise Set 5.2

 2π
dθ
1. Find .
0 5 − 4 cos θ
166 5. Residue Theory

 2π
dθ
2. Find .
0 ∞ 10 + 6 sin θ
1
3. Find 2 + 2x + 2
dx.
x
−∞ ∞
x2
4. Find 2 2
dx.
0 ∞ (1 + x )
x
5. Find 2 + 2x + 2)2
dx
 ∞
−∞ (x
1
6. Find dx (see Example 5.2.6).
0∞ 1 + x6
1
7. Find dx using contour integration over the boundary of the sector
0 1 + x3
defined in polar coordinates by {r eiθ : 0 ≤ r ≤ R, 0 ≤ θ ≤ 2π/3} with R > 1
(see Example
 ∞ 5.2.6).
eix
8. Find dx.
1 + x2
−∞ ∞  ∞
cos(x) eix
9. Find 2 2
dx by first finding 2 2
dx and and then taking
−∞ (1 + x ) −∞ (1 + x )
the real
 ∞ part of the result.
cos x
10. Find dx.
0 1 + x2
11. Prove the fact, used in the proof of Theorem 5.2.2, that if h is a function on
(0, ∞) and limk→∞ h(xk ) exists for every increasing sequence {xk } of positive
numbers converging to ∞, then these limits are all the same number L and
limx→∞ h(x) = L.
12. Prove that an improper integral of the form (5.2.3) converges if f is a contin-
uous function on R which satisfies an inequality |f (x)| ≤ C|x|−p for |x| > R,
where C and R are positive constants, and p > 1.
13. Suppose f is a continuous function on (0, ∞) (which may have a singularity at
s
 ∞ Show how to use the substitution t = e to convert the improper integral
0).
0
f (t) dt into an improper integral on (−∞, ∞). Use this to prove a theorem
 ∞ to Theorem 5.2.2 for improper integrals on (0, ∞).
analogous
dt
14. Find . Hint: Use the substitution of the previous exercise.
0 t + t log2 t

5.3. Fourier Transforms

The technique of evaluating integrals using residue theory is particularly useful in
the study of integral transforms such as the Fourier, Laplace, and Mellin trans-
forms. Often the integrals involved cannot be evaluated using standard integration
techniques from calculus.

The Fourier Transform. The Fourier transform is widely used in Differential

Equations, Physics, and Engineering. Its generalizations are the subject of a ma-
jor area of mathematical research, with applications that impact every branch of
mathematics.
5.3. Fourier Transforms 167

ic

−a b

Figure 5.3.1. Path γ of Theorem 5.3.2.

Definition 5.3.1. If f is a function defined on the real line and t ∈ R, we set

 ∞
1
fˆ(t) = √ f (x) e−itx dx,
2π −∞
whenever this improper integral exists. The resulting function of t is called the
Fourier transform of f .

Note that if f (x) is an absolutely integrable function of x on (−∞, ∞), then

so is e−itx f (x) for each t, since these two functions have the same absolute value.
Thus, the Fourier transform fˆ(t), for such a function, exists at every t ∈ (−∞, ∞).
We will show how to use residue theory to evaluate Fourier transforms of func-
tions that are analytic on C, except at a discrete set of singularities, and that vanish
at infinity. The key to this technique is the fact that | e−itz | = ety vanishes rapidly
as y = Im(z) → ∞ if t < 0, while it vanishes rapidly as y → −∞ if t > 0. In the
first case, we are able to express the integral defining fˆ as a sum of residues in the
upper half-plane, and in the second case we express it as a sum of residues in the
lower half-plane. The final result is the following theorem.
Theorem 5.3.2. Let f be a function analytic on C, except at a discrete set of
singularities, with no singularities on the√real line and with limz→∞ f (z) = 0.
Then fˆ(t) exists if t = 0. It is equal to 2π i times the √ sum of the residues of
f (z) e−izt in the upper half-plane if t < 0, and is equal to − 2π i times the sum of
the residues of f (z) e−izt in the lower half-plane if t > 0.

Proof. We give the proof in case t < 0. The case t > 0 is the same except that
the lower half-plane is used instead of the upper half-plane.
Since limz→∞ f (z) = 0, f has only finitely many singularities. For positive
numbers a, b, and c let γ be the path which traverses the boundary of the rectangle
with vertices −a, b, b + ic, −a + ic in the positive direction (see Figure 5.3.1).
If a, b, and c are large enough, all singularities of f in the upper half-plane will
be inside γ. In particular, there will be no singularities on γ(I). Then,
  b  c
−izt −itx
f (z) e dz = f (x) e dx + i f (b + iy) e−it(b+iy) dy
γ −a 0
(5.3.1)  b  c
− f (x + ic) e−it(x+ic) dx − i f (−a + iy) e−it(−a+iy) dy.
−a 0
168 5. Residue Theory

We pass to the limit, first as c → ∞, and then as a and b approach ∞ independently.

We will show that each of the last three integrals on the right in (5.3.1) vanishes
when we do this.
Since |f (z)| is a continuous function on the complement of its set of singularities,
and it has limit 0 at ∞, it has a maximum value on each line which does not hit
one of the singularities. Let M1 denote its maximum on the vertical line Re(z) = b,
M2 its maximum on the horizontal line Im(z) = c, and M3 its maximum on the
vertical line Re(z) = −a. These numbers are functions of b, c, and a, respectively,
and they each have limit 0 as a, b, and c approach ∞, because limz→∞ f (z) = 0.
For the third integral on the right in (5.3.1), we have
   b
 b 
 −it(x+ic) 
 f (x + ic) e dx ≤ M2 etc dx = M2 etc (b + a).
 −a  −a

Since t < 0, this has limit 0 as c → ∞ for fixed values of a and b.

For the second integral on the right in (5.3.1) we have
 c   c
  M1 tc
 f (b + iy) e −it(b+iy)
dy  ≤ M1 ety dy = (e −1).
  t
0 0

This has limit −M1 /t as c → ∞. Then, since M1 has limit 0 as a, b approach ∞,

the same is true of this integral. The argument that the fourth integral on the right
in (5.3.1) has limit 0 as first c, then a, b approach ∞ is the same.
Of course, the integral on the left in (5.3.1) does not change once a, b, and c
are large enough that the rectangle they determine contains all the singularities of
f in the upper half-plane. Once this is true, its value is 2πi times the sum of the
residues at these singularities. Thus, after passing to the limit as first c, then a and
b approach ∞, (5.3.1) becomes
n  b
2πi Res(f, zj ) = lim f (x) e−itx dx
(a,b)→(∞,∞) −a
j=1
(5.3.2)  ∞ √
= f (x) e−itx dx = 2πfˆ(t),
−∞

where z1 , · · · , zn are the singularities of f in the upper half-plane. On dividing by

√
2π, the proof of the theorem is complete in the case t < 0.
Note that the existence of the limit in (5.3.2) is exactly what is meant by the
convergence of the improper integral defining fˆ(t).
The proof in the case t < 0 is almost identical. The difference is: the number
c is negative. Then the path γ lies in the lower half-plane with negative rather
that positive orientation. This results in the left side of (5.3.1) being −2πi times
the sum of the residues of f e−itz in the lower half-plane for large enough a, b, and
−c. 
1
Example 5.3.3. Find the Fourier transform of f (x) = .
1 + x2
Solution: We set
1
f (z) = .
1 + z2
5.3. Fourier Transforms 169

√
If t < 0, the previous theorem applies, and it tells us that fˆ(t) is 2π i times the
sum of the residues of e−itz f (z) in the upper half-plane. Now f has only one pole
in the upper half-plane and that is at i. The residue of e−itz f (z) at this point can
be computed using Corollary 5.1.2(a), since
1
f (z) = .
(z + i)(z − i)
The residue is obtained by evaluating e−itz /(z + i) at z = i. The result is

et
,
2i
and so 
π t
fˆ(t) = e if t < 0.
2
−itz
√ For t > 0, we compute the residue of e f (z) at −i and then multiply by
− 2π i. The result is 
π −t
fˆ(t) = e if t > 0.
2
At t = 0 we have  ∞ 
1 1 π
fˆ(0) = √ 2
dx = ,
2π −∞ 1 + x 2
by the case n = 1 of Example 5.2.6 (or simply by noting that arctan is an anti-
derivative for the integrand).
Putting the cases t < 0, t > 0, t = 0 together, we conclude that

π −|t|
fˆ(t) = e for all t ∈ R.
2
 ∞
cos x
Example 5.3.4. Show that 2
dx = π/ e .
−∞ 1 + x
Solution: The integral we seek is the real part of the integral
 ∞ −ix
e
2
dx,
−∞ 1 + x
√ 1
which is 2π times the Fourier transform of evaluated at t = 1. By the
1 + x2
previous example, this is π/e.
1
Example 5.3.5. Find the Fourier transform of f (x) = .
x+i
Solution: We set
1
f (z) = .
z+i
This function is meromorphic, with no poles on the real axis, and it vanishes at
infinity. Thus, Theorem 5.3.2 applies.
The only pole of f occurs at z = −i. Since there are no poles in the upper
half-plane, Theorem 5.3.2 tells us that fˆ(t) = 0 if t < 0.
170 5. Residue Theory

√
For t > 0, fˆ(t) is − 2π i times the residue of f (z) e−itz at z = −i. This
residue is e−t and so

0, if t < 0;
fˆ(t) = √ −t
− 2π i e , if t > 0.
Note the jump discontinuity at t = 0.
What about the value of fˆ at t = 0? We have
 ∞
1
fˆ(0) = √ f (x) dx.
2π −∞
We can attempt to evaluate this using the fact that f (z) has log(z + i) as an
antiderivative, given any branch of the log function. If we use the principal branch
of log, then the line Im(z) = i is contained in its domain of definition, and so the
real line is contained in the domain on which log(z + i) is defined and has f (z) as
its derivative. We conclude that
 b
f (x) dx = log(b + i) − log(−a + i)
−a
1
= (log(b2 + 1) − log(a2 + 1)) + i(arg(b + i) − arg(−a + i)).
2
The limit as (a, b) → (∞, ∞) of this expression does not exist, since its real part
approaches +∞ as b → ∞ with a fixed and approaches −∞ as a → ∞ with b
fixed. Thus, the improper integral defining fˆ(0) does not converge. However, the
symmetric integral
 a
f (x) dx = i(arg(a + i) − arg(−a + i))
−a

has limit −πi as a → ∞. In this case, we say that the integral has principal value
−πi, even though it is not a convergent improper integral. If we define fˆ(0) using
this value for the integral, then we have

π
fˆ(0) = − i,
2
which is exactly half way between limt→0+ fˆ(t) and limt→0− fˆ(t).

The next example involves a different technique – no residues – but the Cauchy
theorems are still involved.

Example 5.3.6. Find the Fourier transform of the normal distribution function
1
h(x) = √ e−x /2 .
2

2π

Solution: We have
 ∞
1
e−(x
2
+2itx)/2
hˆ(t) = dx.
2π −∞
5.3. Fourier Transforms 171

it

−a b

Figure 5.3.2. Path for the Fourier Transform in Example 5.3.6.

By completing the square in the exponent, we can write this as

 ∞
1
e−(x+it) /2 e−t /2 dx
2 2
hˆ(t) =
2π −∞
(5.3.3) 
e−t /2 ∞ −(x+it)2 /2
2

= e dx.
2π −∞

We will use Cauchy’s Integral Theorem to show that the last integral above is
independent of t.
Fix a, b > 0 and consider the path γ that traces once in the positive direction
around the boundary of the rectangle with vertices at −a, b, b + ti, −a + ti (see
Figure 5.3.2).
The integral of e−z /2 over this path is 0 by Cauchy’s Theorem. Hence,
2

 b  t
e−x /2 dx + e−(b+iy) /2 i dy
2 2
0=
−a 0
(5.3.4)  b  t
e−(x+it) e−(−a+iy)
2 2
− /2
dx − /2
i dy.
−a 0

The second of these integrals satisfies the following estimate:

 t 
 
 e −(b+iy)2 /2
i dy  ≤ |t| ey2 −b2 ,
 
0
which implies that it has limit 0 as b → ∞. A similar estimate on the fourth integral
in (5.3.4) shows that it has limit 0 as a → 0. It follows that
 ∞  ∞
−(x+it)2 /2
e−x /2 dx,
2
e dx =
−∞ −∞

which means that the integral on the left is independent of t and can be evaluated
by evaluating the integral on √ the right. The integral on the right is a standard
calculus problem. Its value is 2π, and it is obtained by expressing the square of
the integral as a double integral over the plane and then evaluating this using polar
coordinates. We leave it as an exercise (Exercise 5.3.1).
In view of (5.3.3) and the above calculation, we have
e−t /2 √ e−t /2
2 2

hˆ(t) = 2π = √ = h(t).
2π 2π
172 5. Residue Theory

In other words, the normal distribution function is its own Fourier transform.

The Fourier Inversion Formula. We define the inverse Fourier transform fˇ of

a function f on R to be the Fourier transform of f followed by reflection through
the origin – that is:

 ∞
1
(5.3.5) fˇ(x) = fˆ(−x) = √ eitx f (t) dt.
2π −∞

Under reasonable conditions, at points x where f is continuous, we can recover

f (x) from the Fourier transform fˆ by applying the inverse Fourier transform to
fˆ. That is, at points of continuity for f ,

 ∞
1
(5.3.6) f (x) = fˆˇ(x) = √ eitx fˆ(t) dt.
2π −∞

There are several versions of this result, depending on the hypotheses that
are assumed satisfied by f and fˆ. They all have rather technical proofs that are
complicated by the fact that we are working with improper integrals. One fairly
standard version assumes that both f and fˆ are absolutely integrable. A proof of
this version can be found in [8]. We will not attempt to prove any of the versions of
this result here. Instead, we will just illustrate the formula in a couple of examples.
Example
 5.3.7. By Example 5.3.3 the function f (x) = 1/(1 + x2 ) has fˆ(t) =
−|t|
π/2 e as Fourier transform. Compute the inverse Fourier transform of fˆ and
verify that the Fourier Inversion Formula holds in this case.
Solution: We have
 ∞ 
1
fˆˇ(x) = √ eixt π/2 e−|t| dt
2π −∞
 0  ∞ 
1
= e(ix+1)t dt + e(ix−1)t dt
2 −∞ 0
 
1 1 1 1
= − = 2 = f (x).
2 ix + 1 ix − 1 x +1

Indented Contours. The next example involves a technique in which we inte-

grate only part of the way around a circle centered at a pole of the integrand. The
theorem which allows us to compute such integrals is the following:
Theorem 5.3.8. If f is analytic in DR (0) \ {0}, has a simple pole at 0, and γ is
the path γ(t) = r eit , θ1 ≤ t ≤ θ2 , r < R, then

lim f (z) dz = i(θ1 − θ2 ) Res(f, 0).
r→0 γ

Proof. Since f has a simple pole at 0, its Laurent series expansion in DR (0) has
the form
a−1
f (x) = + a0 + a1 z + · · · + an z n + · · · ,
z
5.3. Fourier Transforms 173

where a−1 = Res(f, 0). The integral of the first term of this expansion over γ is
 θ2
ia−1 dθ = i(θ1 − θ2 ) Res(f, 0),
θ1
while the integral of the nth term for n ≥ 0 is
 θ2
ian r n+1 ei(n+1)θ dθ,
θ1

which has absolute value less than or equal to |an |r n+1 (θ2 − θ1 ). Since this has
limit 0 as r → 0, the proof is complete. 
Example 5.3.9. Given real numbers a < b, find the Fourier transform of the
function f which is equal to 1 for a ≤ x ≤ b and is equal to 0 otherwise. Verify
that the Fourier Inversion Formula holds in this case.
Solution: By the definition of f ,
 b
1 i e−ibt − e−iat
fˆ(t) = √ e−itx dx = √ .
2π a 2π t
Then fˆˇ is given by the improper integral
 ∞ it(x−b)
i e − eit(x−a)
fˆˇ(x) = dt.
2π −∞ t
We may write this first as
 −r it(x−b)  ∞ it(x−b) 
i e − eit(x−a) e − eit(x−a)
lim dt + dt ,
r→0 2π −∞ t r t
and then as
 −r  ∞ 
i eit(x−b) eit(x−b)
lim dt + dt ,
r→0 2π −∞ t r t
(5.3.7) −r  ∞ it(x−a) 
i eit(x−a) e
− lim dt + dt ,
r→0 2π −∞ t r t
provided these limits exist.
We evaluate the two limits in (5.3.7) using an indented contour. We use the
contour described in Figure 5.3.3 in the case where x − b > 0 in the first limit and
where x − a > 0 in the second limit. Since the integrands are analytic except at 0,
the integral around this contour is 0. As in the proof of Theorem 5.3.2, the integrals
along the vertical lines and the horizontal line at z = ic have limit 0 as L, R, and c
approach infinity. By the previous theorem, the integral around the semicircle has
limit iπ as r → 0. It follows that the first of the two limits in (5.3.7) is 1/2 if x > b,
while the second limit is 1/2 if x > a.
For the cases x < b and x < a we use the reflection of the contour in Figure
5.3.3 through the x-axis. This has the result of changing the sign of the integral
around the semicircle. Thus, the limits are −1/2 for the case x < b and the case
x < a.
Of course, when x = b, the first limit is limr→0 (ln r − ln r) = 0. The same result
holds for the second limit when x = a.
174 5. Residue Theory

ic

−L −r r R

Figure 5.3.3. Indented Path Used in Example 5.3.9.

We conclude that fˆˇ(x) = 1 if a < x < b and is 0 otherwise. Note that this is
equal to f (x) except at the points a and b where f is not continuous.

Exercise Set 5.3

1. Prove that
 ∞ √
e−x
2
/2
dx = 2π
−∞

by expressing the square of the integral as a double integral over R2 , which

can then be evaluated using polar coordinates.
x
2. Find the Fourier transform of f (x) = .
1 + x2
1
3. Find the Fourier transform of f (x) = .
1 + x4 
π −x
4. Find the Fourier transform of f if f (x) = 0 for x < 0 and f (x) = e for
2
x ≥ 0.
1
5. Find the Fourier transform of f (x) = 2 .
x + 4x + 5
sin x
6. Find the Fourier transform of f (x) = using the method of indented
x
contours as in Example 5.3.9.
7. If f and f  are both continuous and have Fourier transforms which exist at t,
and if f has limit 0 at both ∞ and −∞, prove that the Fourier transform of
f  at t is itfˆ(t). Hint: Use integration by parts.
∞
x sin x
8. Find dx.
−∞ 1 + x2
9. If h is the normal distribution function, λ > 0, and hλ (x) = h(x/λ), then find
(hλ )ˆ(t).
10. Verify that the Fourier Inversion Formula holds for the function hλ of the
previous exercise.
5.4. The Laplace and Mellin Transforms 175

5.4. The Laplace and Mellin Transforms

Two other integral transforms of great importance in applications are the Laplace
and Mellin transforms. Both are strongly related to the Fourier transform, and
residue theory plays a similar role in computations involving these transforms.

The Laplace Transform. The Laplace transform of a function f on [0, ∞) is

defined to be
 ∞
(5.4.1) F (λ) = e−xλ f (x) dx
0

on the set of complex numbers λ for which this integral exists. If u is a real number
such that e−xu f (x) is an absolutely integrable function of x, then the function
e−xλ f (x) will also be integrable for all λ with Re(λ) ≥ u and the integral F (λ) will
be an analytic function of λ on the open half-plane {λ ∈ C : Re(λ) > u} with limit
0 as λ → 0 in this half-plane. In fact, since
 ∞
F (s + it) = e−ixt e−xs f (x) dx,
0
√
the Laplace transform of f at s + it is just 2π times the Fourier transform of
the function g(x) which is e−xs f (x) for x ≥ 0 and is 0 for x < 0. If the Fourier
Inversion Formula applies to g and gˆ, then it results in a formula for recovering f
from its Laplace transform F . Specifically, with λ = s + it,
  s+i∞
exs ∞ 1
(5.4.2) f (x) = F (s + it) eixt dt = F (λ) exλ dλ.
2π −∞ 2πi s−i∞
We will call this the inverse Laplace transform of F .
On the open half-plane where F is defined and analytic, the integral in (5.4.2)
is independent of s and, if x < 0, it has limit 0 as s → ∞. If F can be extended
to a function which is analytic on C except at finitely many singularities, all to the
left of Re(λ) = u, then we may calculate the integral in (5.4.2) using residues. That
is, we compute

1
F (λ) eixλ dλ,
2πi γ
where γ traverses the boundary of a rectangle with interior containing the singu-
larities of F and with right edge along the vertical line Re(λ) = u. As in the proof
of Theorem 5.3.2, this integral is independent of the rectangle chosen and, on the
one hand, is equal to the integral on the right in (5.4.2) and, on the other hand, is
equal to the sum of the residues of F (λ) exλ . Thus, if f is continuous on (0, ∞) and
has a Laplace transform F which is analytic except at finitely many singularities,
then

f (x) = Res(F (λ) exλ , λi ),
where {λi } is the set of singularities of F .
λ
Example 5.4.1. Find the inverse Laplace transform of F (λ) = .
(λ − 2)(λ + 1)2
176 5. Residue Theory

Solution: The function F (λ) exλ has a simple pole at λ = 2 and a pole of
order 2 at λ = −1. By Theorem 5.1.2
2 e2x 2
Res(F (λ) exλ , 2) = = e2x ,
(2 + 1)2 9
while  
d λ eλx  3x − 2 −x
Res(F (λ) e , −1) =
xλ
 = e .
dλ λ − 2 λ=−1 9
Thus, f (x) = 19 [(3x − 2) e−x +2 e2x ] for x ≥ 0.

The Mellin Transform. If f is a function defined on the positive real numbers,

then the function of t defined by
 ∞
(5.4.3) f (x)xt−1 dx,
0
with t > 0, is called the Mellin transform of f . Of course, this improper integral
may converge for only certain values of t and possibly for no values of t. Using
residue theory, we will derive some conditions on f and t which ensure this integral
does converge, and give a formula for computing it.
We assume that f is analytic on C except at a finite set of singularities, none of
which lie on the positive real axis. We define a function z t−1 , analytic on C \ [0, ∞),
in the following way: We cut the plane along the positive real axis and denote by
log(z) for z = 0 the branch of the log function determined by restricting arg(z) to lie
in the interval [0, 2π]. Note that, since we allow both arg(z) = 0 and arg(z) = 2π,
this results in there being two values of log at each point on the positive real axis.
Think of the cut along (0, ∞) as resulting in two copies of the positive real axis –
an upper edge (arg(z) = 0) and a lower edge (arg(z) = 2π). Our log function is
the ordinary real-valued natural log function along the upper edge of the cut, and
is this plus 2πi along the lower edge of the cut. We then define
z t−1 = e(t−1) log z for z = 0.
In the next theorem, we compute the integral (5.4.3) by using path integration on
the function f (z)z t−1 .
Theorem 5.4.2. Let f be a function analytic on C, except at finitely many sin-
gularities, none of which are on the positive real axis, and let z1 , z2 , · · · , zn be the
non-zero singularities of f . Suppose t is not an integer and the function |f (z)||z|t
has limit 0 as z → 0 and as z → ∞. Then the Mellin transform of f at t exists
and is given by
 ∞
e−πti 
n
(5.4.4) f (x)xt−1 dx = −π Res(g, zj ),
0 sin(πt) j=1

where g(z) = f (z)z t−1 on C \ [0, +∞) and z t−1 is defined as above.

Proof. The path γ for our path integral is the path which goes along the upper
edge of the cut from  > 0 to R, then along the circle γR of radius R centered at
the origin, then along the lower edge of the cut from R to  and finally around the
origin along the circle γ of radius  (see Figure 5.4.1).
5.4. The Laplace and Mellin Transforms 177

γR

γε

Figure 5.4.1. Path for Computing the Mellin Transform.

Let z1 , z2 , · · · , zn be the non-zero singularities of f and assume that R is large

enough and  small enough that all of these singularities lie inside the circle of
radius R and outside the circle of radius . Then γ has index 1 about each zj . If
we set
g(z) = f (z)z t−1 for z ∈ / [0, +∞),
then, by the Residue Theorem, we have
n 
2πi Res(g, zj ) = f (z)z t−1 dz
j=1 γ
 R 
(5.4.5) = f (x)xt−1 dx + f (z)z 1−t dz
 γR
 R 
− e2πi(t−1) f (x)xt−1 dx − f (z)z t−1 dz.
 γ

If, as R → ∞ and  → 0, the integrals around γR and γ have vanishing limits,

then

n  ∞
2πi Res(g, zj ) = (1 − e 2πit
) f (x)xt−1 dx.
j=1 0

If t is not an integer, we may solve for the integral. This yields

 ∞
e−πit 
n
(5.4.6) f (x)xt−1 dx = −π Res(g, zj ).
0 sin(πt) j=1

It remains to show that the conditions on f (z)z t−1 ensure the integrals around
the two circles in (5.4.5) have limits 0 as R → ∞ and  → 0. If M (R) is the
maximum value of |f (z)| on the circle of radius R, then
 
 
 f (z)z t−1
dz  ≤ 2πM (R)Rt .
 
γR

Hence, this integral will have limit 0 as R → ∞ if

lim |f (z)||z|t = 0.
z→∞
178 5. Residue Theory

Similarly, if M () is the maximum value of f on the circle of radius , then

 
 
 f (z)z t−1
dz  ≤ 2πM ()t ,

γ

and so this integral will have limit 0 as  → 0 if

lim |f (z)||z|t = 0.
z→0

This completes the proof. 

1
The calculation of the Mellin transform for the function is a key step in
1+x
the developement of the properties of the gamma function in Chapter 9.
1 π
Example 5.4.3. Show that the Mellin transform of f (x) = is .
1+x sin(πt)
Solution: According to the preceding theorem, the Mellin transform of this
function f will exist if 0 < t < 1, since it is for these values of t that
|z|t |z|t
lim = 0 = lim .
z→0 |1 + z| z→∞ |1 + z|

The only singularity of f (z) occurs when z = −1, and the corresponding residue,
Res(g, −1) for g(z) = f (z)z t−1 , is
(−1)t−1 = eπ(t−1)i = − eπti .
Thus, by Theorem 5.4.2, if 0 < t < 1, then
 ∞ t−1
x π
dx = .
0 1+x sin(πt)

Exercise Set 5.4

1. Find the Laplace transforms of the functions f (x) = 1 and g(x) = x. On what
subset of C are these defined?
2. Find the Laplace transform of the function f of Example 5.4.1. Does it coincide
with the function F of this example?
λ
3. Find the inverse Laplace transform of F (x) = 2 .
(λ + 4)
1
4. Find the inverse Laplace transform of F (x) = .
1 + λ2
5. If f is continuous on [0, ∞), differentiable on (0, ∞), f (0) = a, and the Laplace
transform F of f exists for Re(λ) > u, what is the Laplace transform of f  in
terms of F ?
6. If f satisfies the differential equation f  + af  + bf = g on (0, ∞) and f and
f  are continuous on [0, ∞), describe the Laplace transform F of f in terms of
the Laplace transform G of g and the initial values (f (0) and f  (0)) of f and
f  . Hint: Use the result of the previous exercise.
7. Can a non-zero constant function be the Laplace transform of a function f on
[0, ∞)?
5.5. Summing Infinite Series 179

1
8. Find the Mellin transform of f (x) = .
1 + x2
1
9. Find the Mellin transform of f (x) = .
 ∞ 2/3 1 + x3
x
10. Find dx.
0 1 +x
11. Show
 that if h(y) = f (ey ), then the Mellin transform of f can be expressed as
∞
h(y) ety dy. Note the similarity to the inverse Fourier transform.
−∞
12. Verify the second equality in (5.4.5).

5.5. Summing Infinite Series

In this section we demonstrate a method for using the Residue Theorem to calculate
the sum of an infinite series of the form
∞
(5.5.1) f (n),
n=−∞

where f is an analytic function with isolated singularities in the complex plane.

The idea is this: If we can find an analytic function g such that g has a simple
pole with residue 1 at each integer n, then f g will have a simple pole with residue
f (n) at each integer n (see Exercise 5.1.13). If we can also choose an expanding
sequence of simple closed paths {γN } such that each singularity of f g is contained
in γN for sufficiently large N , and such that the integral of f g around γN has limit 0
as N → ∞, then the Residue Theorem will imply that the sum of all residues of f g
constitutes a convergent series with sum 0. If f has only finitely many singularities
and none of them are integers, then the sum (5.5.1) will necessarily converge to
the negative of the sum of the residues of f g at the singularities of f . If some of
the singularities of f occur at integers, then the result will need to be adjusted
accordingly.
This program actually works under reasonable hypotheses on f , if g is chosen
well. A particularly good choice is g(z) = π cot(πz).

Some Properties of the Cotangent. Since sin(πz) has a zero of order one at
every integer and no other zeroes on C, the meromorphic function
cos(πz)
cot(πz) =
sin(πz)
has a simple pole at each integer and no other poles. The residue of this function
is 1/π at each n (Exercise 5.1.5). Thus,
Lemma 5.5.1. The function π cot(πz) has a simple pole with residue 1 at each
integer, and no other poles.

The function π cot(πz) is obviously not bounded near ∞, since it has a pole
at each integer. However, it turns out that there is a sequence of closed paths,
converging to ∞, on which this function is bounded. The paths in question are the
180 5. Residue Theory

(N+1/2)(−1+i) (N+1/2)(1+i)

(N+1/2)(−1−i) (N+1/2)(1−i)

Figure 5.5.1. Path γN .

paths γN , where N is a positive integer and γN traces once in the positive direction
around the square with vertices at
(N + 1/2)(1 + i), (N + 1/2)(−1 + i), (N + 1/2)(−1 − i), and (N + 1/2)(1 − i)
(see Figure 5.5.1).
Lemma 5.5.2. There is a positive number R such that | cot(πz)| ≤ 2 on γN (I) for
each N ≥ R.

Proof. By Definition 1.3.8, we have

cos(πz) eπiz + e−πiz
cot(πz) = = i πiz
sin(πz) e − e−πiz
e 2πiz
+1 e2πix e−2πy +1
= i 2πiz = i 2πix −2πy ,
e −1 e e −1
where z = x + iy. Thus,
 2πix −2πy 
e e +1 
| cot(πz)| =  2πix −2πy .
e e −1 
If x = N + 1/2 for an integer N , then e2πix = −1 and | cot(πz)| ≤ 1. Thus,
| cot(πz)| ≤ 1 on the vertical sides of each γN . On the horizontal sides, where y is
fixed at ±(N + 1/2), the maximum of | cot(πz)| occurs when e2πix = 1, that is, at
integer values of x. Thus, the maximum is
 ±(2N +1)π 
e +1 

 e±(2N +1)π −1  .
Since this expression has limit 1 as N → ∞, there is an M such that it is less than
2 when N > M . 

Lemma 5.5.3. With γN as above,


π cot(πz)
(5.5.2) =0
γN z
for each positive integer N .
5.5. Summing Infinite Series 181

π cot(πz)
Proof. The function h(z) = is analytic inside γN except at integer
z
points, where it has poles. By the Residue Theorem
 N
π cot(πz)
(5.5.3) = Res(h, n).
γN z
n=−N

Since h is an even function, its residue at 0 is 0 (Exercise 5.1.15 ). The poles of h

at non-zero integers are simple poles and may be calculated using Corollary 5.1.6.
The result is that the residue of h at n is 1/n. Since this is an odd function of n
and the range of summation in (5.5.3) is symmetric about the origin, the integral
in (5.5.2) is 0. 

A Summation Theorem. We now have enough information about the function

g(z) = π cot(πz) to prove the following summation theorem.
Theorem 5.5.4. Suppose f is analytic on C except at a finite set
E = {z1 , z2 , · · · , zn }
of isolated singularities. Also suppose that there exist positive constants R and M
such that
M
(5.5.4) |f (z)| ≤ for |z| ≥ R.
|z|
Then
  m
lim f (n) = − Res(πf (z) cot(πz), zj ),
N →∞
n∈ZN
0
j=1

where Z0N is the set of integers in [−N, N ] which do not belong to E.

Proof. The singularities of πf (z) cot(πz) occur at the integers and at the points
z1 , · · · , zn . At an integer n not in E, this function has residue f (n), since cot(πz)
has a simple pole with residue 1 at n and f has no singularity at n. If N is large
enough so that the zj are all inside γN , then the Residue Theorem implies
  m
1
(5.5.5) πf (z) cot(πz) dz = f (n) + Res(πf (z) cot(πz), zj ).
2πi γN N j=1
n∈Z0

Thus, to prove the theorem, we simply need to show that the integral on the left
converges to 0 as N → ∞.
Choose R and M large enough that (5.5.4) holds, the inequality in Lemma
5.5.2 holds, and |zj | ≤ R for each j. Then f is analytic in the annulus
A = {z : R < |z| < ∞}
and vanishes at infinity. It therefore has a Laurent expansion, in this annulus,
involving only negative powers of z. Say,
c−1 c−2 c−n
f (z) = + 2 + ··· + n + ··· .
z z z
By Lemma 5.5.3, 
π cot(πz)
dz = 0.
γN z
182 5. Residue Theory

and so
  
1 1 c−1 
(5.5.6) πf (z) cot(πz) dz = π f (z) − cot(πz) dz.
2πi γN 2πi γN z
However, the function
c−1 c−2 c−n
f (z) − = 2 + ···+ n + ···
z z z
q(z)
has the form in A, where q is analytic in A and has limit c−2 at infinity.
z2
This implies that, for some R1 > R, the function |q| is bounded by some positive
constant M1 on {z : |z| ≥ R1 }. Then, by Lemma 5.5.2, the integrand of the integral
on the right in (5.5.6) is bounded above by 2M M1 /|z|2 . Since |z| > N on γN and
the length of the path γN is 8N + 4, we have that
  
 1  2M M1 (8N + 4)
 πf (z) cot(πz) dz  ≤ ,
 2πi N2
γN

which converges to 0 as N → ∞. This completes the proof. 

∞
1 π2
Example 5.5.5. Prove that 2
= .
n=1
n 6
Solution: We apply the preceding theorem with f (z) = 1/z 2 . It clearly satis-
fies the hypothesis of the theorem. The only singularity of this function is the one
at 0, and so the theorem tells us that
 1  
π cot(πz)
= − Res , 0 ,
n2 z2
n∈Z0

cot z
where Z0 is the set of non-zero integers. The residue of at 0 is −1/3 by
z2
Example 5.1.9. Since the coefficient of z in the Laurent expansion of π cot(πz) is
π 2 times the coefficient of z in the Laurent expansion of cot z, we conclude that
π cot(πz)
has residue π 2 /3 at 0. Then,
z2
∞
1 1  1 π2
= = .
n=1
n2 2 n2 6
n∈Z0

Exercise Set 5.5

2πi
1. Prove that, like π cot(πz), the function has a simple pole with residue
e2πiz −1
1 at each integer and it has no other poles.
∞
1
2. Find .
n=1
1 + n2
∞
1
3. Find .
n=1
(n − i)2
5.5. Summing Infinite Series 183

∞
1
4. Find .
n=1
n4

N
1
5. Prove that if w ∈ C \ Z, then π cot πw = lim .
N →∞ n+w
n=−N
6. Assuming the result of the previous exercise, prove that if w ∈ C \ Z, then
∞
1  1
π cot πw = + .
w n=1 w2 − n2
π
7. Prove that the function has a pole at each integer n with residue (−1)n
sin(πz)
and no other poles.
∞

8. Derive a method for summing a series of the form (−1)n f (n), where f is a
−∞
meromorphic function with a finite number of poles. Hint: Use the preceding
exercise and the method of Theorem 5.5.4.
 Chapter 6

Conformal Mappings

If U and V are open subsets of the complex plane, then just how different are U
and V from the standpoint of the study of analytic functions? If there is a analytic
map h : U → V with an analytic inverse h−1 : V → U , then the answer is, “they
are not different at all.” In this case, we shall say that h is a conformal equivalence
between U and V , and that U and V are conformally equivalent.
If h : U → V is a conformal equivalence, then each analytic function g on
V gives rise to an analytic function g ◦ h on U , and each analytic function f on
U gives rise to an analytic function f ◦ h−1 on V . This establishes a one-to-
one correspondence between the analytic functions on U and those on V . This
correspondence preserves sums and products and exponentials and logarithms and
roots. The same formulas define one-to-one correspondences between meromorphic
functions on U and on V , and between functions with isolated singularities on U and
on V . Similarly, paths, closed paths, and homotopies in U all have corresponding
objects in V . Indices of paths are preserved by this correspondence.
Clearly then, it is important to know when two open sets U and V are con-
formally equivalent. That is the subject of this chapter. The main theorem of the
chapter is the Riemann Mapping Theorem, which says that every connected, simply
connected, proper open subset of C is conformally equivalent to the open unit disc.

6.1. Definition and Examples

The word conformal refers to maps between regions of the plane that are angle
preserving, in a sense to be defined below. It turns out these are exactly the maps
which are analytic with a non-vanishing derivative.

Angle Preserving Maps. Let U and V be open sets in C. A mapping h : U → V

is said to be conformal if it preserves angles between smooth curves at each point
of U . We will make this statement precise below.

185
186 6. Conformal Mappings

Figure 6.1.1. The Angle θ Between Two Paths.

If h(x + iy) = u(x, y) + iv(x, y), for z = x + iy ∈ U , then we may consider h to

be an R2 -valued function of two real variables x and y with coordinate functions
u(x, y) and v(x, y). Let z0 = (x0 , y0 ) be a point of U . If h is differentiable at z0 ,
as a function in this sense, then its differential dh(z0 ) is the linear transformation
from R2 to R2 whose matrix is
 
ux uy
(6.1.1) dh =
vx vy
evaluated at z0 . If γ(t) = (x(t), y(t)) is a differentiable curve in U with γ(t0 ) = z0 ,
then γ  (t0 ) = (x (t0 ), y  (t0 )) is the tangent vector to the curve at the point z0 . The
chain rule implies that
(h ◦ γ) (t0 ) = dh(z0 ) · γ  (t0 ),
where (h ◦ γ) (t0 ) is the tangent vector to the curve h ◦ γ at h(z0 ), and the product
on the right represents the linear transformation dh(z0 ) applied to the vector γ  (t0 ).
Using the matrix for dh, this is given in terms of vector-matrix multiplication as
   
u x uy x
(h ◦ γ) = ,
vx vy y
where the partial derivatives are evaluated at z0 while (h◦γ) , x and y  are evaluated
at t0 . Thus, the matrix dh(z0 ) transforms the tangent vector to γ at z0 into the
tangent vector to the image curve h ◦ γ at h(z0 ).
Definition 6.1.1. We say a mapping h, as above, is conformal at z0 if its differential
dh at z0 is non-singular and angle preserving, in the sense that if a and b are any
two non-zero vectors in R2 , then the angle between dh(z0 )a and dh(z0 )b is the same
as the angle between a and b. If h : U → V is conformal at each point of U and
maps onto V , then we say it is a conformal map of U onto V .
Theorem 6.1.2. A complex-valued function h, defined in a neighborhood of z0 ∈ C,
is conformal at z0 if and only if it has a complex derivative at z0 and this derivative
is non-zero.

Proof. If h has a non-zero complex derivative at z0 , then its partial derivatives

satisfy the Cauchy-Riemann equations at z0 and are not all 0. Hence, the differential
6.1. Definition and Examples 187

(6.1.1) has the form

 
ux uy
dh = ,
−uy ux
with the partial derivatives evaluated at z0 and the vector (ux , uy ) non-zero. If we
put this vector in polar form (r cos θ, r sin θ), with r = 0, then this matrix becomes
    
r cos θ r sin θ r 0 cos θ sin θ
(6.1.2) = .
−r sin θ r cos θ 0 r − sin θ cos θ

That is, if h has a complex derivative at z0 , then the differential of h at z0 is rotation

through an angle θ followed by expansion or contraction by a positive factor r. A
linear transformation of this form is clearly non-singular and preserves the angle
between any two vectors, since both vectors get rotated by the same angle θ. Hence,
h is conformal at z0 .
Conversely, suppose h is conformal at z0 . Let θ be the angle from the vector
(1, 0) to its image under dh(z0 ). Let
 
cos θ sin θ
A=
− sin θ cos θ

be the matrix which rotates every vector through an angle θ. Then A−1 · dh(z0 ) is
an angle preserving matrix which sends (1, 0) to a positive multiple (r, 0) of itself.
Since it preserves angles between vectors, it must also send every other vector to a
multiple of itself. We claim that, in fact, it multiplies each vector by r. Suppose
(0, 1) is sent to (0, t). Then the vector (1, 1) is sent to (r, t) and, if this is to be a
multiple of (1, 1), we must have r = t. This implies that A−1 · dh(z0 ) is r times the
identity matrix. Hence, dh(z0 ) has the form (6.1.2). This implies that h satisfies
the Cauchy-Riemann equations at z0 and, hence, has a complex derivative at z0 .
This derivative is non-zero because dh(z0 ) is non-singular. 

Note that a conformal map h from U onto V need not be one-to-one. In fact,
the exponential function ez is a conformal map of C onto C which is not one-to-one.
However, the following is true, based on the Inverse Mapping Theorem (Theorem
4.5.5).

Theorem 6.1.3. If h is a one-to-one conformal map of U onto V , then h has an

inverse function h−1 : V → U which is also a conformal map.

Proof. The function h : U → V has a well-defined inverse function h−1 : V → U

because h is one-to-one and onto. By Theorem 4.5.5, since h is analytic and has
non-vanishing derivative at each point of U , its inverse function is analytic with
non-vanishing derivative on all of V . 

As mentioned in the introduction, an analytic map from U to V with an analytic

inverse is called a conformal equivalence between U and V . By the preceding
theorem, a map of U onto V is a conformal equivalence if and only if it is a one-
to-one conformal map.
188 6. Conformal Mappings

z 1+z
i
1−z

Figure 6.1.2. Conformal Map from Disc to Upper Half-Plane.

Examples of Conformal Maps. It can be useful to have a substantial inventory

of conformal maps between various regions in the plane. Here we describe a short list
of such maps. Many others can be constructed using similar ideas or by combining
some of these.
Example 6.1.4. Find a conformal equivalence between the unit disc and the right
half-plane.
Solution: Let
1+z
g(z) = .
1−z
Then g is defined and analytic on C \ {1}. If w = −1, the equation
1+z
w=
1−z
can be uniquely solved for z. The solution is
w−1
z= .
w+1
It follows that g is a one-to-one analytic map of the set C\{1} onto the set C\{−1}
with an analytic inverse function
w−1
g −1 (w) = .
w+1
If we multiply both numerator and denominator of the fraction defining g by
1 − |z|2
1 − z, we see that the real part of g(z) is . Hence, g has positive real part
|1 − z|2
if and only if |z| < 1. Thus, g maps the unit disc onto the right half-plane. Since
it has an inverse function, it is a conformal equivalence from the unit disc onto the
right half-plane.
Example 6.1.5. Find a conformal equivalence between the unit disc and the upper
half-plane.
Solution: Since multiplication by i has the effect of rotating points of the plane
by π/2 in the counterclockwise direction, we obtain a conformal equivalence h of
the unit disc to the upper half-plane by multiplying the function g of the previous
example by i. Thus,
1+z
h(z) = i
1−z
6.1. Definition and Examples 189

z
√ i 1+z
1−z

Figure 6.1.3. Conformal Map from Unit Disc to 1st Quadrant.

z 2 1+z
π log 1−z

Figure 6.1.4. Conformal Map from Disc to −1 < Im(z) < 1.

is the equivalence we seek (see Figure 6.1.2).

Example 6.1.6. Find a conformal equivalence from the unit disc to the first quad-
rant of the plane.
Solution: The upper half-plane consists of all points which, in polar coor-
dinates, have the form z = r eiθ with r > 0 and 0 < θ < π, while the open
first quadrant consists of points of this form with 0 < θ < π/2. It follows that
√
the principal branch of the square root function (which sends r eiθ to r eiθ/2 for
−π < θ < π) is a conformal equivalence from the upper half-plane to the first quad-
rant with inverse function w → w2 . It follows from this and the previous example
that 
1+z
h(z) = i
1−z
is a conformal equivalence from the unit disc to the first quadrant, provided we use
the principal branch of the square root function.

Example 6.1.7. Find a conformal equivalence from the unit disc onto the strip
S = {z : −1 < Im(z) < 1}.
Solution: The principal branch of the log function is a conformal equivalence
of the open right half-plane onto the strip {z : −π/2 < Im(z) < π/2}. Thus, we
190 6. Conformal Mappings

z sin(z)

−1 1
−π/2 π/2

Figure 6.1.5. Conformal Map from Half-Strip to Half-Plane.

can construct a conformal equivalence from the unit disc to S by composing the
following series of conformal maps: First, we map the unit disc to the right half-
1+z
plane using z → , as in Example 6.1.4; we follow this by the map log from
1−z
the right half-plane to the strip {z : −π/2 < Im(z) < π/2}; and finally, we divide
by π/2. The composition of these maps is
2 1+z
h(z) = log ,
π 1−z
which is the required conformal equivalence.
Example 6.1.8. Find a conformal equivalence from the set
A = {z : −π/2 < Re(z) < π/2, Im(z) > 0}
to the upper half-plane.
Solution: Recall from Exercise 1.3.14 that the function sin z can be written
as
(6.1.3) sin(x + iy) = sin x cosh y + i cos x sinh y,
while its derivative cos z can be written as
(6.1.4) cos(x + iy) = cos x cosh y − i sin x sinh y.
The first of these formulas can be used to prove that sin is one-to-one on the set
A (Exercise 6.1.9), while the second can be used to prove that its derivative, cos z,
is non-vanishing on A (Exercise 6.1.10). Therefore, sin is a one-to-one conformal
map of A onto some open set B.
To identify B we note that sin takes the interval [−π/2, π/2] on the real line
to the interval [−1, 1]. It takes the half-lines {z : Re(z) = ±π/2, Im(z) > 0} to the
half-lines [1, ∞) and [−∞, −1]. Hence, it must take A to either the upper or lower
half-plane. Since it takes i to i sinh 1 and sinh 1 > 0, it must take A to the upper
half-plane. Hence, sin is the required conformal equivalence.
6.2. The Riemann Sphere 191

Exercise Set 6.1

1. If r eiθ ∈ C and the map z → r eiθ z is considered as a linear transformation

from R2 to R2 , then what is its corresponding matrix?
2. Find a conformal equivalence from the half-plane {z : Re(z) > 1} to the unit
disc D.
3. If D is the unit disc, find a conformal equivalence from D \ {0} to C \ D.
4. Find a conformal equivalence from the open unit disc to the set
W = {z : 0 < arg(z) < π/4}.
5. Show that the image of the upper half-disc {z : |z| < 1, Im(z) > 0}, under the
mapping
1+z
z→
1−z
of Example 6.1.4, is the open first quadrant.
6. Find a conformal equivalence from the upper half-disc of the previous exercise
to the unit disc.
7. For each a with 0 < a < 2π, find a conformal equivalence from the set {z : 0 <
arg z < a} to the upper half-plane.
8. Find a conformal equivalence from the upper half-plane to C \ (−∞, 1].
9. Prove that sin z is one-to-one on the strip A of Example 6.1.8. Hint: Show
that sinh and cosh are increasing functions on {y ∈ R : y > 0} and then use
(6.1.3).
10. Prove cos z is never 0 on the strip A of Example 6.1.8. Hint: Show that sinh
and cosh are non-vanishing functions on {y ∈ R : y > 0} and then use (6.1.4).
11. Prove that the composition h ◦ g of a conformal map g : U → V with a
conformal map h : V → W is conformal.
12. Prove that an analytic function f on an open set U in the plane is a conformal
map from U to f (U ) if and only if it has a local analytic inverse (in the sense
of Definition 4.5.4) at each point of U .

6.2. The Riemann Sphere

In Lemma 4.3.1 we saw how the change of variables z = 1/w + z0 transforms a
function which is analytic in the exterior of a disc centered at 0, and has a limit at
∞, into a function which is analytic in a disc centered at z0 . This suggests that, in
some sense, ∞ can be made to play the same role that ordinary points of C play in
analytic function theory. In fact, it turns out that the set C ∪ {∞} can be treated
as a space with both a topological and an analytic structure in which ∞ is in no
way different from other points. This space is called the Riemann Sphere. It is
the proper setting for the theory of conformal mappings and so it is time for us to
introduce it and discuss its properties.

Construction of the Riemann Sphere. The Riemann Sphere S 2 has several

descriptions. We begin with the simplest of these.
192 6. Conformal Mappings

We denote by S 2 the set C ∪ {∞}, where ∞ denotes an abstract point that

is adjoined to C. We want to emphasize that ∞ is to be thought of as a point of
this space no different than any other. In fact, we will work with two coordinate
systems on S 2 . In one of these, the coordinate function is z – a function analytic
at every point except z = ∞. In the other, the coordinate function is w = 1/z – a
function analytic at every point except z = 0, including the point at ∞ (this will
follow once we define what it means for a function to be analytic at ∞).
An open disc in S 2 , centered at ∞, is defined to be the set of all points with
w coordinate satisfying |w| < r for a fixed r > 0. Equivalently, it is the set of all
points with z coordinate satisfying |z| > 1/r. An open set in S 2 is a set U such that
every point of U is the center of some open disc contained in U . This is the familiar
definition of open set in case U ⊂ C. It now also makes sense if U contains ∞, since
we have defined the notion of an open disc centered at ∞. With this definition, a
neighborhood of ∞ is the union of {∞} with an open subset of C which contains
the complement of a closed bounded disc centered at 0.
With the topology (the collection of open sets) defined, we can talk about limits
of functions from a subset of S 2 to a subset of S 2 , and about the continuity of such
functions. These notions are defined using open sets (i.e., neighborhoods of points)
rather than epsilons and deltas, as suggested by Theorem 2.1.10 and Definition
2.1.11. Thus, if f : U → S 2 is a function defined on an open set U ⊂ S 2 , then
lim f (z) = L
z→z0

if and only if for each neighborhood W of L, there is a deleted neighborhood V of

z0 such that V ⊂ U and f (V ) ⊂ W . This leads to definitions for the meaning of
the statements limz→z0 f (z) = ∞ and limz→∞ f (z) = L, that are consistent with
the meanings attached to them in preceding chapters.
A function f , defined on an open set U ⊂ S 2 , with values in S 2 , is continuous
at z0 ∈ U if and only if limz→z0 f (z) = f (z0 ). As in Theorem 2.1.13, f is continuous
at every point of its domain U if and only if the inverse image, under f , of every
open set is open.

Analytic Functions on the Riemann Sphere. In what follows, the phrase an-
alytic at z0 , applied to a function on a subset of S 2 , will mean defined and analytic
in a neighborhood of z0 .
We know what it means for a function to be analytic at a finite point of S 2 .
What does it mean for a function f (z) to be analytic at ∞? We make sense of this
by, again, using the coordinate function w = 1/z, which is 0 at z = ∞.
Definition 6.2.1. We will say that a complex-valued function f (z) is analytic
at ∞ if f (1/w) is analytic at w = 0 (meaning defined and analytic in a deleted
neighborhood of 0 with a removable singularity at 0).

Thus, a function f (z) is analytic at ∞ if f (1/w), when given the appropriate

value at 0, is complex differentiable with respect to w for w in a neighborhood of
0. Note that, at points where z and w = 1/z are both finite, differentiability with
respect to z and differentiability with respect to w are equivalent, although the two
derivatives are not the same (Exercise 6.2.1).
6.2. The Riemann Sphere 193

1−z
Example 6.2.2. Show that f (z) = is analytic at ∞ if we set f (∞) = −1.
1+z
Solution: According to the above definition, we must show that f (1/w) is
analytic at w = 0. However,
1 − 1/w w−1
f (1/w) = =
1 + 1/w w+1
and this is, indeed, analytic at w = 0.

A way to think of S 2 that is suggested by the use of the z and w coordinate

functions is to regard it as two copies of C glued together along C\{0} by identifying
the point z in one copy of C \ {0} with the point 1/z in the other copy.

Analytic Functions from S 2 to S 2 . We will now discuss analytic functions

defined on an open subset of S 2 with values in S 2 rather than in C. There is some
question as to what this means. That is, we know what an analytic function from
an open subset of S 2 to C is, but what does it mean for a function to be analytic
at a point where it has the value ∞? This is only a problem if we insist on using
the z coordinate function on the image space of the function f . If we use the w
coordinate function, instead, then it is clear that f should be considered analytic
at a point z0 where f (z0 ) = ∞ if and only if 1/f is analytic at z0 .
1−z
Example 6.2.3. Show that the function is analytic at z = −1 as a function
1+z
from S 2 to S 2 .
Solution: Since f (−1) = ∞, to show that f is analytic at −1 as an S 2 -valued
function, we must show that 1/f is analytic at −1. In fact,
1 1+z
=
f (z) 1−z
and this is analytic at z = −1.

The function 1/z that has played a major role in the discussion to this point,
can not only be thought of as a new coordinate function on S 2 , it can also be
thought of as a mapping of S 2 to itself which interchanges 0 and ∞. We will call
this mapping s. Thus,
s(z) = 1/z,
for all z ∈ S 2 (including 0 and ∞, where, of course, 1/0 = ∞ and 1/∞ = 0).
According to our definitions, a function f is analytic at infinity if and only if
f ◦ s is analytic at 0, while f is analytic at a point z0 where it has the value ∞ if
and only if s ◦ f is analytic at z0 .
Theorem 6.2.4. The mapping s is a conformal equivalence of S 2 onto itself, that
is, it is a one-to-one analytic map of S 2 onto itself with an analytic inverse map.

Proof. Clearly s is one-to-one and onto and is its own inverse function, since
s ◦ s(z) = z. It is also analytic at every point, with the possible exceptions of 0
and ∞. However, s(1/w) = w is analytic at w = 0 and so s is analytic at z = ∞.
Similarly, 1/s(z) = z is analytic at z = 0 and so s is analytic at 0. This completes
the proof. 
194 6. Conformal Mappings

A meromorphic function f (z) on C is said to have a pole of order n at ∞ if

f (1/w) has a pole of order n at 0 (we include the case where the pole is of order
0 – i.e. the case of a removable singularity). The next theorem shows that the
meromorphic functions on C, with poles at ∞, are the analytic functions from S 2
to S 2 .
Theorem 6.2.5. Each meromorphic function on C with a pole at ∞ defines an
analytic function f˜ from S 2 to S 2 , and each analytic function from S 2 to S 2 arises
in this way.

Proof. Certainly a meromorphic function with a pole at ∞ can be considered as

a function f˜ from S 2 to S 2 – we let f˜(z) = f (z) at points of C which are not poles
of positive order and assign the value ∞ to f˜ at each point (including ∞) at which
f has a pole of positive order.
If f (z) has a pole of order 0 at ∞, then f (1/w) has a removable singularity
at 0 and we assign f˜(∞) the value which makes f (1/w) analytic at w = 0. The
resulting function f˜ : S 2 to S 2 is analytic at each point of C where f is analytic.
At a pole z0 ∈ C of f (z), 1/f (z) is analytic, and this implies that f˜(z) is analytic.
At the point ∞, f (z) has a pole as a meromorphic function on C if and only if
f (1/w) has a pole at 0. This happens if and only if f˜ is analytic at ∞ as a function
from S 2 to S 2 .
Thus, each meromorphic function f on C with a pole at ∞ extends to be an
analytic function f˜ from S 2 to S 2 . It remains to show that every analytic function
from S 2 to S 2 arises in this way.
If g is an analytic function from S 2 to S 2 , then at points z0 of C where g(z0 ) =
∞, g is an analytic complex-valued function. At a point z0 of C where g(z0 ) = ∞,
1/g(z) is analytic and, hence, g(z) has a pole. Therefore, the restriction f of g to
C is a meromorphic function. At ∞, g(1/w) is either analytic or has a pole and
this implies that f has a pole at ∞. This completes the proof. 

Complex Projective Space. The Riemann Sphere can also be described in a

way that treats all points alike from the beginning, so that there is no special point
∞. This is its description as Complex Projective Space of dimension 1 as described
below.
Let C2 denote the complex two-dimensional vector space consisting of all or-
dered pairs (z, w) of complex numbers. We define a space P 1 (C) by starting with
the space of non-zero vectors in C2 and then identifying any two of these that dif-
fer only by a scalar factor. That is, we identify (z1 , w1 ) and (z2 , w2 ) if there is a
complex scalar λ = 0 such that (z2 , w2 ) = (λz1 , λw1 ).
The set {(z, 0) : 0 = z ∈ C} is a single point p of P 1 (C), since any two vectors in
this set are scalar multiples of one another. Similarly, the set {(0, w) : 0 = w ∈ C}
is a single point q of P 1 (C). If we remove p from P 1 (C), then there is a well-defined
map
z
φ : P 1 (C) \ {p} → C, where φ(z, w) = .
w
Note that φ defines a well-defined map on P 1 (C) \ {p} because it sends equivalent
points (z, w) and (λz, λw) to the same point z/w. Similarly, there is a well-defined
6.2. The Riemann Sphere 195

map
w
ψ : P 1 (C) \ {q} → C, where ψ(z, w) = .
z
Both φ and ψ are one-to-one and onto. The inverse map for φ sends the complex
number z to the equivalence class containing (z, 1), while the inverse map for ψ
sends w to the equivalence class containing (1, w). Then, on C\{0}, the composition
ψ ◦ φ−1 is defined and it is, in fact,
1
ψ ◦ φ−1 (z) = .
z
This means that P 1 (C) may also be thought of as two copies of C glued together
along C \ {0} by identifying the point z in one copy of C \ {0} with the point 1/z in
the other copy. Definitions of continuity and complex differentiability for functions
on P 1 (C) may then be defined at points on each copy of C. These definitions will
agree on the overlap because 1/z is an analytic function of z on C \ {0}. The
space P 1 (C), with topology and notion of analytic function as defined above, is
one-dimensional complex projective space.
It is apparent that our description of P 1 (C) is just another description of the
Riemann Sphere S 2 .

Stereographic Projection. Why do we call S 2 the Riemann Sphere? It does

not appear to look like a sphere at all. However, there is a standard construction
that shows how a plane with the point at infinity adjoined can be made to look like
a sphere.
Consider a (hollow) sphere S in R3 . If one point is removed from S, the result
is a space which can be mapped in a one-to-one fashion onto the plane. To see this,
imagine S sits in R3 = {(x, y, t) : x, y, t ∈ R}, above the xy-plane and is tangent to
it at the origin. Thus, the t-axis meets the sphere at the origin and one other point
p. Then every line through p, that is not tangent to the sphere, meets the sphere
S at exactly one point u, other than p, and also meets the xy-plane at exactly one
point v. The function ρ which sends the point u to the point v is a one-to-one
map of the sphere with p removed onto the xy-plane. This function is called the
stereographic projection of the sphere to the plane. The standard topology on the
sphere is such that ρ and its inverse function are both continuous. This means that
a set is open in S \ {p} if and only if its image in the plane is open. A one-to-one
onto map with this property is called a homeomorphism between the two spaces
involved, and its existence means that the two spaces are equivalent as topological
spaces.
Note that, if S is thought of as a globe representing the earth and p is thought
of as the north pole, then the spherical open discs on the sphere, centered at p,
are just the sets consisting of all points above a certain north latitude. Each such
disc is taken by the stereographic projection ρ onto the exterior of a closed disc,
centered at 0, in the xy-plane.
This discussion suggests that the sphere S, as a topological space, is homeo-
morphic to a space constructed from the plane by adding a single point and then
defining the topology in such a way that exteriors of closed discs centered at 0 are
deleted neighborhoods of ∞. This is how we constructed the Riemann Sphere.
196 6. Conformal Mappings

Figure 6.2.1. Stereographic Projection.

Exercise Set 6.2

1. Prove that if f is defined in a neighborhood of z0 ∈ C \ {0} , then f (z) is

differentiable with respect to z at z0 if and only if f (1/w) is differentiable
with respect to w at w0 = 1/z0 . What is the relationship between the two
derivatives?
1
2. Show that the function f (z) = 2 is analytic at ∞ if it is given the value
z +5
0 at ∞.
3. Show that the function f (z) = 2z − 1 is not analytic at ∞ as a complex-valued
function, but it is analytic at ∞ as an S 2 -valued function.
4. Show that ez is not analytic at ∞ – either as a complex-valued function or as
an S 2 -valued function.
z4
5. If f (z) = 2 is considered as an analytic function from S 2 to S 2 , where
z −1
are the poles of f and what are their orders?
6. Prove that a polynomial of degree n determines an analytic function from S 2
to S 2 which has a pole of degee n at ∞ and no other poles.
7. Show that if f is an analytic function from S 2 to S 2 which has no poles on C,
then f is a polynomial.
8. Find a conformal equivalence from S 2 to S 2 which sends 1 to ∞, −2 to 0, and
0 to 6.
9. Show that every open cover of S 2 has a finite subcover – that is, show that S 2
is compact as a topological space (see Section 2.5).
10. Show that if f is a non-constant analytic function from S 2 to S 2 and w ∈ S 2 ,
then f takes on the value w at most finitely many times. If w = ∞, this means
that f has only finitely many poles. If w = ∞, this means that f (z) − w has
only finitely many zeroes.
6.3. Linear Fractional Transformations 197

11. Can we consider sin z as an analytic function from S 2 to S 2 ? Justify your

answer.
12. Show that if f is a non-constant analytic function from S 2 to S 2 , then f is
onto. That is, f takes on each value p ∈ S 2 at least once.
13. Show that S 2 is connected. That is, show that S 2 cannot be expressed as the
disjoint union of two non-empy open subsets.
14. If f : S 2 → S 2 is an analytic function
 which has no pole or zero at ∞, then
show that, for sufficiently large R, |z|=R f (z) dz = 0.
15. If f : S 2 → S 2 is a non-constant analytic function, w ∈ S 2 , and n(f, w) is
the number of zeroes of f (z) − w, counting multiplicity, if w = ∞, and is the
number of poles of f , counting multiplicity, if w = ∞, then prove that n(f, w)
is constant as a function of w. Hint: Use the result of previous exercise and
Rouché’s Theorem.
16. Prove that every analytic function from S 2 to S 2 is a rational function – that
p(z)
is, a function of the form , where p and q are polynomials in z.
q(z)

6.3. Linear Fractional Transformations

A linear fractional transformation or Möbius transformation is a mapping of the
form
az + b
(6.3.1) h(z) = ,
cz + d
where ad − bc = 0. Note that, if the condition ad − bc = 0 is not satisfied, then h
is the constant function b/d.
The right way to think about a linear fractional transformation is to regard it as
a map from S 2 to S 2 , where ∞ is taken to limz→∞ h(z) = a/c and the point −d/c
is taken to ∞. In fact, as we shall see below, the linear fractional transformations
are exactly the conformal equivalences from S 2 to itself.

Conformal Automorphisms of S 2 . By a conformal automorphism of an open

subset U of S 2 we mean a one to one conformal map of U onto itself. The composi-
tion of two conformal automorphisms of U is clearly also a conformal automorphism
of U , as is the inverse function of a conformal automorphism of U . Thus, the con-
formal automorphisms of an open set U form a group under the operation of com-
position. We are particularly interested in the group of conformal automorphisms
of S 2 itself.
We will show that the conformal automorphisms of S 2 are exactly the linear
fractional transformations – that is, the transformations of the form (6.3.1). There
are two types of linear fractional transformations that will be particularly useful in
the argument. These are the affine transformations, that is, transformations of the
form
L(z) = az + b,
198 6. Conformal Mappings

and the inversion transformation s, given by

1
s(z) = .
z
Theorem 6.3.1. Each linear fractional transformation is a conformal automor-
phism of S 2 . Conversely, each conformal automorphism of S 2 is either an affine
transformation or a composition L1 ◦ s ◦ L2 , where L1 and L2 are affine transfor-
mations. Hence, each conformal automorphism of S 2 is a linear fractional trans-
formation.

Proof. We show that each linear fractional transformation h has an inverse func-
tion which is also a linear fractional transformation. In fact, if
az + b
h(z) = ,
cz + d
and we solve the equation w = h(z) for z, the result is
−dw + b
z= = g(w).
cw − a
This suggests that g is an inverse function for h. In fact, if we calculate h ◦ g, then
for all w ∈ S 2 except w = ∞ and w = −a/c we get
(ab − cd)w
h ◦ g(w) = .
ab − cd
Since ab − cd = 0, we conclude that h ◦ g(w) = w. The cases w = ∞ and w = −a/c
can be checked separately (Exercise 6.3.1), and so h ◦ g(w) = w for all w ∈ S 2 . A
similar calculation shows that g ◦ h(z) = z for all z ∈ S 2 . Hence, h has an inverse
function h−1 = g : S 2 → S 2 , and it is also a linear fractional transformation.
Since h has an analytic inverse function, defined on all of S 2 , it is a conformal
automorphism of S 2 . Thus, each linear fractional transformation is a conformal
automorphism of S 2 .
To prove the converse, we let f be a conformal automorphism of S 2 and proceed
to show that it has the form claimed in the theorem. We first consider the case
where f (∞) = ∞. By Theorem 6.2.4, s(z) = 1/z is a conformal automorphism of
S 2 . Thus, the composition
1
q(z) = s ◦ f ◦ s(z) =
f (1/z)
is also a conformal automorphism of S 2 , and it takes 0 to 0. Thus, q is a one-to-one
analytic function with a zero at 0. This zero must be of order one since a conformal
map has non-vanishing derivative at each point. This implies that f (1/z) has a
pole of order 1 at z = 0 and is analytic and finite at every other point of S 2 . Then
f (z) has a pole of order 1 at ∞ and is analytic and finite everywhere on C. It
follows that z −1 (f (z) − f (0)) is analytic and finite everywhere on S 2 . Hence, it is
constant by Liouville’s Theorem. If this constant is a and b = f (0), then f is the
affine transformation f (z) = az + b. Such a transformation is of the form (6.3.1),
with c = 1 and d = 0, and, hence, is a linear fractional transformation.
We next consider the case where f (∞) = k = ∞. If we let
1
p(z) = ,
z−k
6.3. Linear Fractional Transformations 199

then p is a linear fractional transformation and, hence, a conformal automorphism

of S 2 . The composition
1
p ◦ f (z) =
f (z) − k
is then a confromal automorphism of S 2 which takes ∞ to ∞. By the previous
paragraph, it has the form az + b. This implies that
1 akz + bk + 1
f (z) = +k = .
az + b az + b
This is the composition of the affine transformation z → az + b, followed by s,
followed by the affine transformation z → z + k. Since this composition is also a
linear fractional transformation, the proof is complete. 

Lines and Circles. By a circle in S 2 we mean either a circle in C or a line in

C together with the point {∞}. That is, lines are to be thought of as circles in S 2
which pass through ∞.
Theorem 6.3.2. Each linear fractional transformation transforms a circle in S 2
to a circle in S 2 .

Proof. By the previous theorem, each linear fractional transformation is either

affine or the composition of affine transformations and s. Thus, it suffices to show
that s and each affine transformation map circles in S 2 to circles in S 2 .
Each affine transformation L(z) = az + b obviously takes lines to lines and
circles to circles, since, if a = r eiθ , then L is just rotation through an angle θ,
followed by an expansion or contraction by a factor r, followed by a translation by
b. To finish the proof, we must show that s also takes circles in S 2 to circles in S 2 .
Every line or circle in the plane is the solution set to an equation of the form
(6.3.2) a|z|2 + wz + wz + b = 0,
where a and b are real numbers and w is a complex number. Conversely, every
equation of this form has a line, circle, point, or the empty set as its set of solutions
(Exercise 1.1.18). Given a line or circle with equation (6.3.2), the transformation s
takes a point on this line or circle to a point z which satisfies the equation
a|z|−2 + wz −1 + wz −1 + d.
On multiplying through by |z|2 , this becomes
a + wz + wz + d|z|2 ,
which is another equation of the form (6.3.2). Hence, its solution set is a line, circle,
point, or the empty set. Clearly, only the first two are possible images of a line or
circle under s. This completes the proof. 

The preceding result can be very useful in making a quick determination of the
image of a set under a linear fractional transformation.
Example 6.3.3. Find the image of the unit disc under the transformation
2z
h(z) = .
z−i
200 6. Conformal Mappings

Solution: We have h(1) = 1 + i, h(−1) = 1 − i, and h(i) = ∞. Since h

must take the unit circle to a line or a circle, it evidently takes the unit circle to
the vertical line Re(z) = 1. Since the unit circle is the boundary of the open unit
disc, h must take this disc to a set having the line Re(z) = 1 and the point ∞
as its boundary. Thus, the image must be one of the half-planes {z : Re(z) < 1},
{z : Re(z) > 1}. Since h(0) = 0, the image must be {z : Re(z) < 1}.
A linear fractional transformation h is determined by the images of any three
distinct points in S 2 under h.
Theorem 6.3.4. Given two ordered triples {w1 , w2 , w3 } and {z1 , z2 , z3 } of dis-
tinct points of S 2 , there is exactly one linear fractional transformation h such that
h(wj ) = zj , j = 1, 2, 3.

Proof. We first show that there is a linear fractional tranformation that takes a
given ordered triple of complex numbers {w1 , w2 , w3 } to {0, 1, ∞}. In fact,
(w2 − w3 )(z − w1 )
h(z) =
(w2 − w1 )(z − w3 )
does the job.
Now suppose we have a triple {w1 , w2 , w3 } of distinct points of S 2 , where one
of the points is ∞. Then we can still find an h taking this triple to {0, 1, ∞}. For
example, if w3 = ∞, then
z − w1
h(z) =
w2 − w1
will suffice.
Now to get a linear fractional transformation h which takes {w1 , w2 , w3 } to
{z1 , z2 , z3 }, we choose h1 taking {w1 , w2 , w3 } to {0, 1, ∞} and h2 taking {z1 , z2 , z3 }
to {0, 1, ∞} and then set h = h−1 2 ◦ h1 .
It remains to prove that h is unique. Suppose g is another linear fractional
transformation that takes {w1 , w2 , w3 } to {z1 , z2 , z3 }. With h = h−1 2 ◦ h1 , as in the
previous paragraph, we have that f = h2 ◦g◦h−1 1 is a linear fractional transformation
that takes {0, 1, ∞} to {0, 1, ∞}. If
az + b
f (z) = ,
cz + d
then f (0) = 0 implies that b = 0, f (∞) = ∞ implies that c = 0, and f (1) = 1
implies that a/d = 1. We conclude that f is the identity transformation – that is,
f (z) = z. This, in turn, implies that
g = h−1 −1
2 ◦ f ◦ h1 = h2 ◦ h1 = h. 

Automorphisms of the Unit Disc. We next determine the conformal automor-

phisms of the open unit disc and study their properties. These maps will play a
key role in our proof of the Riemann Mapping Theorem. We begin by identifying
which linear fractional transformations map the unit disc to itself.
Let D = D1 (0) be the open unit disc and let w be a point of D. Then the
linear fractional transformation
z−w
(6.3.3) hw (z) =
1 − wz
6.3. Linear Fractional Transformations 201

maps w to 0, and 0 to −w. It also maps the unit circle to itself since, if |u| = 1,
then u−1 = u, and so
     
 u − w   u − w  u − w
 = = 
 1 − wu   u−1 − w   u − w  = 1.

Since hw maps the unit circle to itself and a point, w, inside the unit circle to
another point, 0, inside the unit circle, it must map the unit disc D onto itself (it
has to map it to some open set in S 2 which has the unit circle as boundary and
there are only two choices: D and the complement of its closure). It follows that
hw is a conformal automorphism of D.
Suppose that h is some other conformal automorphism of D that takes w to 0.
Then h ◦ h−1w is a conformal automorphism of D that takes 0 to 0. By Theorem
3.5.6 there is a complex number u of modulus one such that h ◦ h−1w (z) = uz. On
replacing z by hw (z), this implies that h(z) = uhw (z). Thus, we have proved the
following theorem.
Theorem 6.3.5. The conformal automorphisms of the unit disc are the linear
fractional transformation of the form
z−w
h(z) = u = uhw (z),
1 − wz
where |u| = 1 and |w| < 1.

The following is an elementary calculation and is left as an exercise (Exercise

6.3.8). It will be used in the proof of the Riemann Mapping Theorem.
Theorem 6.3.6. With w ∈ D and hw as above,
1
hw (0) = 1 − |w|2 and hw (w) = .
1 − |w|2

Exercise Set 6.3

1. Prove that h ◦ g(w) = w for the transformations h and g of Theorem 6.3.1 in

the exceptional cases w = ∞ and w = a/c.
2. Find a linear fractional transformation that takes the line Re(z) = 1 to the
circle of radius one centered at −1.
3. What is the image of the unit disc under the linear fractional transformation
2iz
h(z) = ?
z−1
4. What is the image of the disc of radius 1, centered at 1 + i, under the linear
fractional transformation
1+z
h(z) = ?
1−z
5. What is the image of the disc of radius 1/2, centered at 1/2, under the linear
fractional transformation of the previous exercise?
202 6. Conformal Mappings

1+z
6. The linear fractional transformation h(z) = takes the open unit disc D
1−z
to the right half-plane. Show that h sends the set consisting of those points in
D that lie outside another circle which intersects the unit circle at right angles
and passes through the point z = 1 to a set of the form
{z : Re(z) > 0, Im(z) < a} or {z : Re(z) > 0, Im(z) > −a}
for some positive a. What is a?
7. A certain linear fractional transformation h takes the ordered triple of points
{a1 , b1 , c1 } on circle number one to the ordered triple {a2 , b2 , c2 } on circle
number two. Just by looking at these points, how can you tell whether h maps
the inside of circle one to the inside or to the outside of circle two?
8. Prove Theorem 6.3.6.
9. Prove that, for each w ∈ D, the mapping hw of (6.3.3) has h−w as its inverse
transformation.
10. Find a conformal automorphism of D which takes 1/2 to 0 and has derivative
3i/4 at z = 0.
11. Find a conformal automorphism of D which takes 1/2 to i/3 and has positive
derivative at 1/2.
12. If w ∈ D, prove that among all analytic functions that map D into D, map w
to 0, and have a positive derivative at w, the one with the largest derivative
is the function hw of (6.3.3). Hint: Use Schwarz’s Lemma (Lemma 3.5.5).
13. Each non-singular 2 × 2 complex matrix
 
a b
A=
c d
determines a linear fractional transformation
az + b
φA (z) = .
cz + d
Show that this correspondence A → φA is a group homomorphism – that is,
show that it takes the product AB of two matrices A and B to the composition
φA ◦ φB of the corresponding linear fractional transformations.
14. What is the kernel of the map A → φA of the preceding exercise – that is, for
which matrices A is φA the identity transformation z → z?

6.4. The Riemann Mapping Theorem

In this section we prove the Riemann Mapping Theorem, which says that every
proper, simply connected, open subset of C is conformally equivalent to the unit
disc.
The proof proceeds as follows: Given a proper, simply connected, open subset
U ⊂ C, we fix a point z0 ∈ U and consider the family F of all one-to-one conformal
maps of U into the unit disc D which map z0 to 0. We first show that this set is
non-empty. We then show that if a given f ∈ F does not map U onto D, then there
is a g ∈ F with |g  (z0 | > |f  (z0 )|. Finally, we show that there is an h ∈ F whose
derivative at z0 has maximum absolute value. Such an h must map U onto D.
6.4. The Riemann Mapping Theorem 203

The last step in this program – showing that there is an h ∈ F with derivative of
maximal modulus at z0 – involves showing that any sequence in F has a subsequence
which converges to a function which is also in F. This means we must prove that
F is a normal family. We take up this task first.

Normal Families.
Definition 6.4.1. Let U be an open subset of C. A collection F of analytic func-
tions defined on U is called a normal family if each sequence in F either converges
uniformly to infinity on each compact subset of U or has a subsequence which
converges uniformly on each compact subset of U to an analytic function.

Normal families containing sequences which converge to infinity will not occur
in this chapter, but they will play a role in the next chapter in the proof of the Big
Picard Theorem.
Here we encounter another one of the amazing properties of analytic functions:
It takes very little for a family of analytic functions to be a normal family. In fact,
it is enough that the family be uniformly bounded, where F is said to be uniformly
bounded on U if there is an M such that |f (z)| ≤ M for every z ∈ U and every
f ∈ F.
Theorem 6.4.2 (Montel’s Theorem). Given an open set U ⊂ C, each uniformly
bounded sequence of analytic functions on U has subsequence which converges uni-
formly on each compact subset of U . Hence, each uniformly bounded family of
functions on U is a normal family.

Proof. Let {fn } be a uniformly bounded sequence of functions on U . We must

show that this sequence has a subsequence that converges uniformly on each com-
pact subset of U .
We begin by choosing an enumeration z1 , z2 , · · · , zn , · · · of the points of U with
rational coordinates. Since F is uniformly bounded, there is an M > 0 such that
|f (z)| ≤ M for all z ∈ U and all f ∈ F. Then {fn (z1 )} is a sequence of complex
numbers, bounded in modulus by M . It therefore has a convergent subsequence.
We denote by {f1n } the corresponding subsequence of {fn }. Then {f1n (z2 )} is
a bounded sequence, and so it too has a convergent subsequence. We denote the
corresponding subsequence of {f1n } by {f2n }. Continuing in this way we inductively
construct a sequence of subsequences of {fn }:
f11 , f12 , ··· f1n , ···
f21 , f22 , ··· f2n , ···
· · ··· · ···
(6.4.1) · · ··· · ···
fk1 , fk2 , ··· fkn , ···
· · ··· · ···
· · ··· · ···
with each sequence a subsequence of the preceding sequence, and {fkn (zj )}n a
convergent sequence of numbers for each k and each j ≤ k. Then the diagonal
sequence {fnn } is a subsequence of {fn } which converges at every zj .
204 6. Conformal Mappings

Let gn = fnn . We will show that {gn } converges uniformly on every compact
subset of U – not just pointwise at each rational point zj .
If w ∈ U , we choose r > 0 such that the closed disc D2r (w) is contained in U .
Then for each z ∈ Dr (w) we have
Dr (z) ⊂ D2r (w) ⊂ U.
By Cauchy’s estimates, each f ∈ F satisfies
M
|f  (z)| ≤ .
r
Then, for any two points z, z  ∈ Dr (w) we have
 z 
  M
(6.4.2) |f (z) − f (z  )| =  f  (λ) dλ ≤ |z − z  |.
z  r
Note that the bound on the right in this estimate is independent of the function
f ∈ F. In particular, it holds for each function in the sequence {gn }.
r
Given  > 0, we choose δ = . If z ∈ Dr (w), there are points with rational
3M
coefficients arbitrarily close to z. Let zj be one such point with |z − zj | < δ. By
(6.4.2),
M r 
|gn (z) − gn (zj )| < = .
r 3M 3
We next choose N such that

|gn (zj ) − gm (zj )| < whenever n, m ≥ N.
3
We can do this because {gn (zj )} is a convergent, hence Cauchy, sequence for each
j. Then
|gn (z) − gm (z)| ≤ |gn (z) − gn (zj )| + |gn (zj ) − gm (zj )| + |gm (zj ) − gm (z)| < ,
if n, m ≥ N . This proves that gn is uniformly Cauchy on the disc Dr (w) and,
hence, converges uniformly on this disc.
If K is a compact subset of U , then the discs of the form Dr (w), where w ∈ K
and r is chosen so that D2r (w) ⊂ U , form an open cover of K. Since K is compact,
it is covered by finitely many of these discs. Since {gn } converges uniformly on
each of these discs, it also converges uniformly on K. 

The Riemann Mapping Theorem. Let U be a non-empty, connected, proper

subset of C with the property that every non-vanishing analytic function on U has
an analytic square root. This property is satisfied, for example, if U is simply
connected (Theorem 4.6.16). We fix a point z0 ∈ U and let F denote the set of
one-to-one conformal maps f of U into the open unit disc D, such that f (z0 ) = 0.
The proof of the Riemann Mapping Theorem consists of several steps leading to
the conclusion that there is an element of F which maps onto D.
Lemma 6.4.3. The set F is non-empty.

Proof. Since U is a proper subset of C, there is a point λ ∈ C which is not in U .

Then the function f (z) = z − λ is non-vanishing on U and, hence, has an analytic
square root g. Since f is one-to-one and non-vanishing, so is g and, hence, g is a
one-to-one conformal map whose image does not contain 0.
6.4. The Riemann Mapping Theorem 205

w
f
0
z0

Figure 6.4.1. The Mapping f of Lemma 6.4.4.

Since g is analytic, its image g(U ) is open by the Open Mapping Theorem
(Theorem 4.5.8). It follows that we can find a closed disc Dr (w0 ) ⊂ g(U ) with
0 < r < ∞. Since g 2 = f and f is a one-to-one function, the image of g cannot
contain a non-zero point and its negative, since then these two points would be
taken to the same point by g 2 . Thus no point of the reflection through the origin,
Dr (−w0 ), of Dr (w0 ) is contained in g(U ). It follows that
|g(z) + w0 | > r
for all z ∈ U , and this means that the function
r
p(z) =
g(z) + w0
is a one-to-one conformal map of U into the unit disc. If p(z0 ) = w, then we can
compose p with the conformal automorphism hw of the unit disc (6.3.3) to obtain
a one-to-one conformal map hw ◦ p of U into D which takes z0 to 0. Thus, F is not
empty. 

Lemma 6.4.4. Let U , z0 , and F be as above. If f ∈ F and f does not map U onto
D, then there is a g ∈ F with
|g  (z0 )| > |f  (z0 )|.

Proof. Let w ∈ D be a point which is not in the image of f (see Figure 6.4.1).
Recall from the previous section that the map hw of (6.3.3) is a conformal automor-
phism of the unit disc which sends w to 0. Thus, hw ◦ f (z) = 0 for all z ∈ U . As
in the proof of the previous lemma, this means that hw ◦ f has an analytic square
root q. If q(z0 ) = λ, then λ2 = w, q 2 = hw ◦ f , and
hw (0)  (1 − |w|2 )  (1 − |λ|4 ) 
q  (z0 ) = f (z0 ) = f (z0 ) = f (z0 ).
2q(z0 ) 2λ 2λ
The function q is a one-to-one conformal map of U into the unit disc, but q(z0 ) is
not 0 – it is λ. To get an element of F we compose q with hλ . The resulting map
g = hλ ◦ q belongs to F and satisfies
(1 − |λ|4 )  (1 + |λ|2 ) 
(6.4.3) g  (z0 ) = hλ (λ)q  (z0 ) = f (z0 ) = f (z0 ).
2λ(1 − |λ|2 ) 2λ
206 6. Conformal Mappings

Since 0 < (1 − |λ|)2 = 1 − 2|λ| + |λ|2 , we have 2|λ| < 1 + |λ|2 . Since a conformal
map has non-vanishing derivative, |f  (z0 )| > 0. It then follows from (6.4.3) that
|g  (z0 )| > |f  (z0 )|. 
Theorem 6.4.5 (Riemann Mapping Theorem). Let U be a proper, connected open
subset of C with the property that every non-vanishing analytic function on U has
an analytic square root. Then there is a conformal equivalence of U onto the unit
disc D.

Proof. Let z0 ∈ U and F be as above. We know that F is non-empty by Lemma

6.4.3. We set
m = sup{|f  (z0 )| : f ∈ F}.
Then m is either +∞ or a positive number, since every conformal map has non-
vanishing derivative. We will show that there is an element h of F such that
|h (z0 )| = m. Clearly, in view of the previous lemma, such an h maps U onto D.
We choose a sequence {fn } of elements of F such that
lim |fn (z0 )| = m.
n→∞
Since F is a uniformly bounded family of analytic functions, it is a normal family
by Theorem 6.4.2. Hence, there is a subsequence of {fn } which converges uniformly
on compact subsets of U to a function h. It follows from Cauchy’s estimates that
the derivatives of the functions in this subsequence converge to h (Exercise 3.2.15),
and so |h (z0 )| = m. Since m = 0, h is not a constant. Since each fn is one-to-one,
it follows that h is also one to one (Exercise 4.5.5). Since fn (z0 ) = 0 for every
n, we also have h(z0 ) = 0. Hence, h is an element of F whose derivative at z0
has modulus m. By the previous lemma, h must map U onto D. Hence, h is a
conformal equivalence of U to D. 

A conformal equivalence h from an open set U to an open set V is, in particular,

a homeomorphism – that is, a one-to-one continuous function from U onto V with
a continuous inverse function. Such a function h takes a curve γ in U to a curve
h ◦ γ in V . Its inverse function h−1 takes curves in V to curves in U . Clearly
homotopies are taken to homotopies by these maps. It follows that if there is a
homeomorphism between two open sets, then one is simply connected if and only
if the other is also. This implies the following corollary of the Riemann Mapping
Theorem.
Corollary 6.4.6. If U is a connected open set in the plane with the property that
every non-vanishing analytic function on U has an analytic square root, then U is
simply connected.
This corollary completes the proof of Theorem 4.6.16.

Exercise Set 6.4

1. Is the set of all conformal automorphisms of the unit disc a normal family?
If so, then every sequence of such maps has a convergent subsequence. Is the
limit necessarily also a conformal automorphism of the unit disc?
6.5. The Poisson Integral 207

2. If U is a connected open subset of C and z0 is a point of U , prove that the set

of all analytic functions f on U with positive real part and with f (z0 ) = 1 is
a normal family.
3. Suppose U is a connected open set and V is a simply connected open set which
is not the whole plane. If z0 ∈ U and w0 ∈ V are fixed, let F be the family
of all analytic functions from U to V which send z0 to w0 . Prove that F is a
normal family.
4. Show by example that it is not true that every uniformly bounded family F
of continuous functions on an interval on the line has the property that every
sequence in F has a convergent subsequence.
5. Prove that the complex plane C is not conformally equivalent to the unit disc
even though it is simply connected. Why doesn’t this contradict the Riemann
Mapping Theorem?
6. Is an open rectangle in the plane conformally equivalent to the open unit disc
D?
7. Is the open right half-plane, with the interval [1, ∞) removed, conformally
equivalent to the unit disc D?
8. Is the open right half-plane, with the interval [1, 2] removed, conformally equiv-
alent to the unit disc D?
9. Is the punctured disc D1 (0)\{0} conformally equivalent to the right half-plane
{z : Re(z) > 0}?
10. Given one conformal equivalence h from an open set U to the unit disc D, can
you describe all the others in terms of h?
11. With F as in the proof of Theorem 6.4.5, is it necessarily true that a conformal
equivalence h from U to D that takes z0 ∈ U to 0 has the property that |h (z0 )|
is maximal among the numbers |f  (z0 )| for f ∈ F?
12. Prove that if f is analytic and satisfies |f (z)| < 1 on the right half-plane, and
if f (1) = 0, then  
z − 1

|f (z)| ≤  
z + 1
for all z in the right half-plane.

6.5. The Poisson Integral

Let U be a bounded open set in the plane and let g be a continuous function on
∂U . The Dirichlet problem for U and g is the problem of finding a function u which
is continuous on U , harmonic on U , and equal to g on ∂U . The function g is called
the boundary function for the problem. A uniqueness theorem for solutions to the
Dirichlet problem follows easily from the results of Section 3.5 (see Exercise 6.5.2):
Theorem 6.5.1. If U is a bounded open subset of C and g is a continuous function
on ∂U , then there is at most one solution to the Dirichlet problem for U and g.

If U is simply connected and has a simple closed curve as its boundary, then
the Dirichlet problem always has a solution. One proof that this is so, uses the
Riemann Mapping Theorem. This proof proceeds as follows: We exhibit an explicit
solution to the Dirichlet problem for the unit disc D and any continuous function
208 6. Conformal Mappings

on its boundary. The Riemann Mapping Theorem tells us there is a conformal

equivalence h : U → D. We use this map to show that the Dirichlet problem on U
is equivalent to the Dirichlet problem on D.
The difficulty with the above approach is that it requires that the conformal
equivalence h : U → D extends to a one-to-one continuous map U → D with a
continuous inverse. In fact, this is true and we will state it as a theorem, but we
do not include a proof. A proof can be found in Walter Rudin’s text [8].
Theorem 6.5.2. If U is a simply connected open set with a simple closed curve
as boundary, and if h : U → D is a conformal equivalence, then h extends to a
continuous function from U to D with a continuous inverse.

Assuming this theorem, we will proceed with the proof that the Dirichlet prob-
lem always has a solution for a set U as above. In this section we deal with the
first step – the case where U is the unit disc D. The next section is devoted to
showing how to use the Riemann Mapping Theorem to attack the problem on more
complicated sets.

The Poisson Kernel. Recall the linear fractional transformation from Example
6.1.4:
1+z
(6.5.1) .
1−z
This is a one-to-one conformal map of the unit disc onto the right half-plane.
The Poisson kernel is constructed from the real part of this function. This was
calculated in Example 6.1.4 by multiplying numerator and denominator in (6.5.1)
by the conjugate of the denominator. The result is
 
1+z 1 − |z|2
(6.5.2) Re = .
1−z |1 − z|2
If z = r eiθ is the polar form of z, then
|1 − z|2 = |1 − r eiθ |2 = 1 − 2r cos θ + r 2
and so the expression on the right in (6.5.2) becomes
1 − r2
(6.5.3) Pr (θ) = .
1 − 2r cos θ + r 2
This may also be written as
∞

Pr (θ) = r |n| einθ
−∞
by Exercise 6.5.1.
The Poisson kernel is the function
1 − r2
Pr (θ − t) =
1 − 2r cos(θ − t) + r 2
(6.5.4)    it 
1 + e−it z e +z
= Re = Re .
1 − e−it z eit −z
It produces a harmonic function on D when it is integrated with respect to t against
a continuous function on the boundary of D.
6.5. The Poisson Integral 209

Theorem 6.5.3. If D is the open unit disc, g is a continuous real-valued function

on ∂D, and u is defined by
 2π
1
u(r eiθ ) = g(eit )Pr (θ − t) dt,
2π 0
then u is a harmonic function of z = r eiθ on D.

Proof. Since g is real-valued, the function u is the real part of

 2π
1 eit +z
(6.5.5) f (z) = g(eit ) it dt.
2π 0 e −z
This function is analytic on the complement of the unit circle (Exercise 3.2.16) and,
in particular, on D. Hence, its real part u is harmonic on D. 

The integral defining the harmonic function u in the above theorem is called
the Poisson integral.

Boundary Values. We will show that the function u, defined in the above
theorem, has the property that
lim u(z) = g(w)
z→w
for each w on the unit circle. This implies that we get a continuous extension of
u to the closed unit disc D if we define u(w) to be g(w) at each w ∈ ∂D. Before
proving this, we need to establish some properties of the function Pr (θ).
Lemma 6.5.4. The Poisson kernel Pr (θ − t) of (6.5.4) satisfies
 π
1
Pr (θ − t) dt = 1
2π −π
for every r < 1 and every θ.

Proof. We have
 π   π 
1 1 eit +z
(6.5.6) Pr (θ − t) dt = Re dt .
2π −π 2π −π eit −z
We can easily evaluate this using residue theory. In fact,
 π it  π
1 e +z 1 1 + e−it z it
dt = i e dt
2π −π e −z
it 2πi −π eit −z

1 1 + z/w
= dw.
2πi |w|=1 w − z
The last expression can be evaluated by applying the Residue Theorem to the
function
1 + z/w w+z
= .
w−z w(w − z)
This has residue −1 at w = 0 and residue 2 at w = z. Hence,
 π it
1 e +z
dθ = 2 − 1 = 1.
2π −π eit −z
Thus, its real part is also 1. In view of (6.5.6), the proof is complete. 
210 6. Conformal Mappings

Lemma 6.5.5. If g is a continuous function on the unit circle, then

 π
1
(6.5.7) lim− g(eit )Pr (θ − t) dt = g(eiθ ).
r→1 2π −π

Furthermore, the convergence is uniform in θ.

Proof. The idea of the proof is this: When r is near 1, Pr (θ − t) is nearly 0 for all
t ∈ [−π, π] except near θ, where the function has a large spike. This means that
only values of g(eit ) for t near θ contribute significantly to the integral (6.5.7). The
proof below makes this claim precise.
The function Pr (θ) is a positive, even function of θ. Since its denominator is
increasing on the interval [0, π], this function is decreasing as θ moves away from 0
in either direction. It follows that, if η is a fixed number between 0 and π, then
(6.5.8) Pr (θ − t) ≤ Pr (η) when η ≤ |θ − t| ≤ π.

Since g is continuous on the unit circle, which is compact, g is uniformly con-

tinuous. Thus, given  > 0, we may choose η > 0, which does not depend on θ,
such that
(6.5.9) |g(eit ) − g(eiθ )| < /2 whenever |θ − t| < η.

It follows from Lemma 6.5.4 that

 π  π
1 1
g(e )Pr (θ − t) dt − g(e ) =
it iθ
(g(eit ) − g(eiθ ))Pr (θ − t) dt.
2π −π 2π −π
The absolute value of this expression is less than or equal to
 π
1
|g(eit ) − g(eiθ )|Pr (θ − t) dt.
2π −π
We split this integral into the sum of an integral over [θ − η, θ + η] and integrals
over [−π, θ − η] and [θ + η, π]. The first summand may be estimated using (6.5.9)
and Lemma 6.5.4:
 θ+η  θ+η
1  1 
(6.5.10) |g(eit ) − g(eiθ )|Pr (θ − t) dt < Pr (θ − t) dt ≤ .
2π θ−η 2 2π θ−η 2
The integrals over [−π, −θ0 ] and [θ0 , π] are estimated using (6.5.8). If M is an
upper bound for g on the circle, then
 π
1
|g(eit ) − g(eiθ )|Pr (θ − t) dt
2π θ+η
(6.5.11)  π
1
< Pr (η) |g(eit ) − g(eiθ )| dt ≤ M Pr (η),
2π θ+η
and the integral over [−π, θ − η] satisfies the same estimate.
Since 0 < η < π,
1 − r2
lim− Pr (η) = lim− = 0,
r→1 r→1 1 − 2r cos η + r 2
because the numerator has limit 0 and the denominator has limit 2 − 2 cos η = 0.
6.5. The Poisson Integral 211

We choose δ > 0 such that


|Pr (η)| < if 1 − δ < r < 1.
4M
Then (6.5.11) implies that
 π
1  
(6.5.12) |g(eit ) − g(eiθ )|Pr (θ − t) dt ≤ M = .
2π θ+η 4M 4

Putting (6.5.10) together with (6.5.12) and the analogous estimate for the in-
tegral on [−π, θ − η], we conclude that
  π 
 1 
 iθ 
g(e )Pr (θ) dθ − g(e ) < 
iθ
 2π
−π

for 1 − δ < r < 1 and for every θ ∈ [−π, π]. This proves the Theorem. 

The next theorem tells us that the Poisson integral yields a solution to the
Dirichlet problem for the unit disc and an arbitrary continuous boundary function
g.
Theorem 6.5.6. If g is a real-valued continuous function on the unit circle and u
is defined on the closed unit disc D by
 2π
1
(6.5.13) iθ
u(r e ) = g(eit )Pr (θ − t) dt
2π 0
if r < 1 and by u(eiθ ) = g(eiθ ) on the unit circle, then u is continuous on D and
harmonic on D.

Proof. In view of Theorem 6.5.3, the only thing left to prove is that
(6.5.14) lim u(z) = g(eiθ0 ) for all θ0 .
z→eiθ0

The previous lemma almost says this, but not quite. The limit there is taken only
along radial lines, and this is not exactly what is meant by (6.5.14). We need to do
a bit more work.
Because the convergence in Lemma 6.5.5 is uniform in θ, given  > 0 we may
choose a δ1 > 0 such that

|u(r eiθ ) − g(eiθ )| ≤
2
for all θ and all r with 1 − δ1 < r < 1. We may also choose δ2 such that

|g(eiθ ) − g(eiθ0 )| <
2
whenever |eiθ − eiθ0 | < δ2 . If δ = min{δ1 , δ2 }, then |r eiθ − eiθ0 | < δ implies both
|r eiθ − eiθ | < δ1 and | eiθ − eiθ0 | < δ2 , from which it follows that
|u(r eiθ ) − g(eiθ0 )| ≤ |u(r eiθ ) − g(eiθ )| + |g(eiθ ) − g(eiθ0 )| < .
This proves (6.5.14). 
212 6. Conformal Mappings

Poisson’s Formula. Suppose g is a continuous function defined on the circle

of radius R centered at z0 and we wish to solve the Dirichlet problem for DR (z0 )
and g. By using the conformal map z → z0 + Rz, which takes D to DR (z0 ), we
can transform this problem into a Dirichlet problem on the unit disc. The solution
given by Theorem 6.5.3 may then be pulled back to DR (z0 ) using the inverse map
z → (z − z0 )/R. The result is a solution to the Dirichlet problem for DR (z0 ) and
g given by
 2π
1 (R2 − r 2 )
(6.5.15) u(r eiθ ) = g(z0 + R eit ) 2 dt.
2π 0 R − 2rR cos(θ − t) + r 2
We leave the details to Exercise 6.5.8.
This result may be used to prove Poisson’s Integral Formula for a harmonic
function on an open set U :
Theorem 6.5.7. Let u be a real-valued harmonic function defined on an open set
U in the plane. If D R (z0 ) ⊂ U , then
 2π
1 (R2 − r 2 )
u(z) = u(z0 + R eit ) 2 dt
2π 0 R − 2Rr cos(θ − t) + r 2
  2π 
1 R eit +z
= Re u(z0 + R eit ) it dt
2π 0 R e −z
for all z = z0 + r eiθ ∈ DR (z0 ).

The proof is left as an exercise (Exercise 6.5.9)

Note that the second equality in the above theorem expresses u as the real part
of an analytic function in the disc DR (z0 ).

Exercise Set 6.5

1. Prove that the Poisson kernel may also be written as

∞

Pr (θ) = r |n| einθ .
−∞

2. Prove the uniqueness theorem (Theorem 6.5.1) for solutions of the Dirichlet
problem.
3. By inspection, find a solution to the Dirichlet problem on D if the boundary
function g is g(eit ) = cos t. Use your answer to evaluate
 2π
1 (1 − r 2 ) cos t
dt.
2π 0 1 − 2r cos(θ − t) + r 2
4. Show that the Poisson kernel Pr (θ − t) is a harmonic function of z = r eiθ on
the open unit disc.
5. Equation (6.5.5) expresses the harmonic function u, given by the Poisson in-
tegral, as the real part of an analytic function on D. Use this equation to find
an integral formula for a harmonic conjugate to u.
6.6. The Dirichlet Problem 213

6. Show that if g is a positive, continuous function on the unit circle, then the
analytic function f defined by equation (6.5.5) has positive real part and its
imaginary part vanishes at z = 0.
7. Let f be analytic on the open unit disc D. If the real part of f is positive and
extends to be continuous on D, then prove that f is a function of the form
given in the previous exercise plus an imaginary constant.
8. Supply the details of the proof that (6.5.15) gives a solution to the Dirichlet
problem for DR (z0 ) with boundary function g.
9. Prove Theorem 6.5.7.
10. Show that if u is a real-valued, non-negative harmonic function on an open set
containing DR (z0 ), then

R−r R+r
u(z0 ) ≤ u(z) ≤ u(z0 )
R+r R−r
for 0 ≤ r = |z − z0 | < R. This is Harnack’s Inequality. Hint: Use Theorem
6.5.7.
11. If u is harmonic and non-negative on D and u(0) = 1, give upper and lower
bounds for u(1/2). Hint: Use the result of the previous exercise.
12. Prove that if un is a sequence of positive harmonic functions on a connected
open set U , and if un (z0 ) → 0 at one point z0 ∈ u, then un (z) → 0 at every
point z of U . Hint: Use the result of Exercise 10.

6.6. The Dirichlet Problem

The Dirichlet problem is of great importance in physics and engineering. Here we
briefly discuss some of the applications that make this so. We then move on to
showing how to use conformal equivalence to reduce the Dirichlet problem for a set
U and boundary function g to a simpler problem which, hopefully, we know how
to solve.

Heat Flow. The termperature T at points of a flat plate represented by a region

U in the plane satisfies the heat equation
 2 
∂T ∂ T ∂2T
(6.6.1) = kΔT = k + 2 .
∂t ∂2x ∂ y

If the set U is bounded and if each point w of its boundary is maintained at

a constant temperature T0 (w), then, over time, the temperature distribution on
the entire plate will approach an equilibrium state T̃ – one that does not change
with time. This means ∂ T̃ /∂t = 0. The heat equation (6.6.1) then implies that
T̃ , as a function of x and y satisfies Laplace’s equation. Hence, T̃ is a harmonic
function on U which agrees with the boundary function T0 on ∂U . Thus, to find the
equilibrium temperature on U that will result from maintaining the temperature
on the boundary at T0 means solving the Dirichlet problem for U and T0 .
214 6. Conformal Mappings

Electrostatics. An electric field on a plane region U is a vector function

E(z) = (E1 (z), E2 (z))
which, for each z ∈ U , describes the force that would be exerted on a charged
particle located at z. This vector function has the form
 
∂φ ∂φ
E = −∇φ = − ,
∂x ∂y
for a real-valued function φ on U called the potential function for the electric field
E. A static electric field (one that is not changing in time) satisfies
∂E1 ∂E2
div E = + = 0.
∂x ∂x
This means its potential function φ satisfies Laplace’s equation and is, therefore,
harmonic.
If the boundary of U is maintained at a potential given by the boundary func-
tion ψ on ∂U , then the problem of finding the resulting potential at points of U is
the Dirichlet problem for U with boundary function ψ.

Hydrodynamics. In the study of two-dimensional fluid flow for non-viscous

fluids, the velocity of the fluid at a given point z is a vector q(z) = (q1 (z), q2 (z)).
If the fluid is incompressible and the flow is irrotational, so that
div q = 0 and curl q = 0,
then the vector q is the gradient ∇φ of a potential function φ which satisfies
Laplace’s equation and, hence, is a harmonic function. A typical fluid flow problem
is not, however, a Dirichlet type problem. The typical boundary condition for fluid
flow is that the flow along the walls of the container should be parallel to the walls;
that is, the component ∂φ/∂n of ∇φ normal to the wall should be 0. The problem
of finding a harmonic function which satisfies this condition on a given region is
called the Neumann problem.

Existence of Solutions. Given a bounded simply connected open set U with a

simple closed curve for its boundary, the Riemann Mapping Theorem and Theorem
6.5.2 tell us that there is a homeomorphism h : U → D which, on U , is a conformal
map of U onto D. This allows us to prove the existence of a solution to the Dirichlet
problem for U in the following way.
If g is a continuous real-valued function on ∂U , then g ◦ h−1 is a continuous
real valued function on ∂D. By Theorem 6.5.6, g ◦ h−1 is the restriction to ∂D
of a function u which is continous on D and harmonic on D. Then u ◦ h is a
continuous function on U which equals g on ∂U . It is also harmonic on U , since the
composition of an analytic function followed by a harmonic function is harmonic
(Exercise 6.6.11). This proves that the Dirichlet problem for U and g has a solution
for every continuous function g on ∂U .
The Dirichlet problem also makes sense for unbounded open sets. Suppose U is
an unbounded, simply connected, open subset of C with closure U a proper subset
6.6. The Dirichlet Problem 215

of C. If we choose a point w ∈ C \ U , then the linear fractional transformation

1
h(z) =
z−w
is a conformal automorphism of S 2 which takes U to a bounded simply connected
open subset V of C. If V has a simple closed curve as boundary, then we will say
that U has a simple closed curve as boundary in S 2 . If this is the case, then the
Dirichlet problem can be solved for V and any continuous function on ∂V . The
map h is a homeomorphism from the closure of U in S 2 to the closure of V and so
it can be used to transform a continuous function g on the boundary of U in S 2 to
a continuous function g ◦ h−1 on the boundary of V . A solution to the Dirichlet
problem for V and g ◦h may then be transformed back to a solution to the Dirichlet
problem for U and g by composing with h.
The following theorem summarizes the preceding discussion.
Theorem 6.6.1. Let U be a proper simply connected open subset of C with a simple
closed curve as its boundary in S 2 . Then each continous function on the boundary
of U in S 2 is the restriction of a continuous function on the closure of U in S 2
which is harmonic on U .
Remark 6.6.2. The hypothesis that g is continuous in the above theorem is not
really necessary if one is willing to accept somewhat less in the conclusion. For
example, if g is bounded and continuous except at a finite set of discontinuities,
then Lemma 6.5.3 continues to hold, while in Lemma 6.5.5 and Theorem 6.5.6, the
conclusions about limits and continuity of u at boundary points continue to hold
at the points where g is continuous. This implies that the previous theorem can
be modified to cover the case where g is bounded and continuous except at a finite
set of discontinuities. The conclusion is then that g is the restriction to ∂U of a
function which is harmonic on U and continuous on U except at those points of ∂U
where g is not continuous.
Finding Explicit Solutions. The previous theorem tells us that solutions to
the Dirichlet problem exist if U is simply connected with a simple closed curve as
boundary, but it does not give us an explicit expression for the solution in terms
of elementary functions. It is too much to hope for this in general, but there are
situations where a conformal equivalence between U and some other set V reduces
the problem to one where we can find an explicit solution, often by inspection. This
is another way in which conformal maps aid in the study of the Dirichlet problem.
We will demonstrate this with a series of examples.
Example 6.6.3. Let U be the strip {z : −1 < Im(z) < 1}. Find a solution
to the Dirichlet problem for U and the boundary function which is 0 on the line
{z : Im(z) = −1} and 2 on the line {z : Im(z) = 1}.
Solution: This one is easy. The function 1 − iz is analytic on C and so its real
part
1 + Im(z) = 1 + y
is harmonic on C. The restriction of this function to {z : −1 ≤ Im(z) ≤ 1} is
continuous, is harmonic on {z : −1 < Im(z) < 1}, and has the required values on
the boundary of this set.
216 6. Conformal Mappings

2
2
h

0
0

Figure 6.6.1. The Picture for Example 6.6.4.

Example 6.6.4. Find a function on the closed unit disc which is continuous except
at ±1, is harmonic on the open unit disc, and is 0 on the lower half of the unit
circle and 2 on the upper half of the unit circle.
Solution: In Example 6.1.7 we showed that the function
 
2 1+z
h(z) = log
π 1−z
is a conformal equivalence from the unit disc to the strip {z : −1 < Im(z) < 1}.
This map takes the upper semicircle to the line Im(z) = 1 and the lower semicircle
to the line Im(z) = −1. It follows that a solution to the problem may be obtained
by composing the solution to the previous problem with h (see Figure 6.6.1). This
yields  
2 1+z
u(z) = 1 + Im(h(z)) = 1 + arg ,
π 1−z
so that  
2 2y
u(x, y) = 1 + tan−1
π 1 − x2 − y 2
is our solution. Here, for points on the unit circle x2 + y 2 = 1, we interpret this
expression to mean its limit as the point (x, y) is approached from within the unit
disc. Note that the resulting function u is, indeed, continuous on the closed disc,
except at the points ±1 where the boundary function fails to be continuous.
√
Example 6.6.5. Let g be the function on R which is 1 − x2 for −1 < x < 1 and
is 0 for all other values of x. Find a harmonic function on the upper half-plane
which has g as its boundary function and which has limit 0 as z → ∞ in the upper
half-plane.
Solution: We use the results of Example 6.1.8. There we showed that sin z is
a conformal equivalence of the strip A = {z : −π/2 < Re(z) < π/2, Im(z) > 0}
onto the upper half-plane. If we compose g with sin z, we get the function q on ∂A
which is cos x on the real line between −π/2 and π/2 and is 0 on the vertical lines
Re(z) = ±π/2. The real part e−y cos x of eiz is a harmonic function on A which
has q as boundary function and has limit 0 as z → ∞ in A. Hence, we obtain a
solution u to our problem by setting
−1
f (z) = ei sin (z)
and u(z) = Re(f (z))
for z in the upper half-plane. We can simplify this.
If w = sin−1 (z), then sin w = z and cos2 w = 1 − z 2 . Since, for every z in the
open upper half-plane, 1 − z 2 is in the domain of the principal branch of the square
6.6. The Dirichlet Problem 217

i z z+1/z

-1 1 -2 0 2

Figure 6.6.2. The Picture for Example 6.6.6.

√ √
root function (denoted · ), we have cos w = ± 1 − z 2 . Since cos w has positive
real part if w ∈ A (see (6.1.4)), and the principal branch of the √
square root function
takes values in the right half-plane, we conclude that cos w = 1 − z 2 . Then

eiw = cos w + i sin w = 1 − z 2 + iz,
and so 
f (z) = 1 − z 2 + iz.
Then our solution u is
 
u(z) = Re 1 − z 2 + iz
(6.6.2)   
1 
= −y + 1 − x2 + y 2 + (1 − x2 + y 2 )2 + 4x2 y 2 .
2
This last equality is left as an exercise (Exercise 6.6.13).

Example 6.6.6. Let B be the open upper half-plane with the upper half of the
closed unit disc removed. That is,
B = {z : |z| > 1, Im(z) > 0}.
Solve the Neumann problem for B: Find a function u on B which is harmonic on
B and has normal derivative 0 at every boundary point of B.
Solution: The same problem on the upper half-plane has an easy solution: the
function Re(z) = x is harmonic on the upper half-plane and its gradient (1, 0) is
parallel to the real axis at each point. Thus, its normal derivative on the real axis
is 0. To get a solution to our original problem, we need only compose this solution
with a conformal equivalence of B to the upper half-plane. In fact
1
z→z+
z
is such a map (Exercise 6.6.9). Thus, our solution is
 
1
u(z) = Re z +
z
x
=x+ 2 .
x + y2
Any constant multiple of this solution is also a solution. If this solution represents
the potential function for a fluid flow problem, then the corresponding velocity
218 6. Conformal Mappings

vector field for the flow would be its gradient

 
y 2 − x2 2xy
∇u(x, y) = 1 + 2 ,− 2 .
(x + y 2 )2 (x + y 2 )2

Exercise Set 6.6

1. Solve the Dirichlet problem for the upper half-plane and the boundary function
which is 1 on the positive real axis and 0 on the negative real axis. Hint: Think
about the function arg(z) on the upper half-plane.
2. Use the result of the preceding exercise to find a different way of solving the
problem in Example 6.6.4 .
3. Solve the Dirichlet problem on the first quadrant with boundary function g
which is 1 on the real axis and 0 on the imaginary axis.
4. Solve the Dirichlet problem on the half-disc {z : |z| < 1, Im(z) > 0} with
boundary function which is 0 on the line [−1, 1] and 1 on the semicircle where
|z| = 1. Hint: Use the result of Exercise 6.1.5.
5. Solve the Dirichlet problem for the open set consisting of the points inside the
unit circle and outside a circle C which intersects the unit circle at right angles
at the points 1 and i. The boundary function is the function which is 1 on the
part of the boundary which lies on the unit circle and is 0 on the part which
lies on C. Hint: see Exercise 6.3.6.
1 1
6. Show that , for x ∈ R, is the real part of . Use this to solve the
1 + x2 1 − ix
1
Dirichlet problem on the upper half-plane with boundary function on
1 + x2
the real line and limit 0 as z → ∞.
7. Modify the approach used in the preceding problem to find a solution to the
1
Dirichlet problem on the lower half-plane for the boundary function .
1 + x2
8. For each w ∈ C show that the equation w = z + 1/z has two solutions for z,
one is the inverse of the other, and they are distinct if and only if w = ±2.
Show that the solution with |z| ≥ 1 has imaginary part with the same sign as
Im(w), while if |z| < 1, then z and w have imaginary parts with opposite sign.
9. Show that the transformation z → z + 1/z is a conformal equivalence of the
exterior of the unit circle onto the complex plane with the interval [−2, 2]
removed. Also show that the restriction of this transformation to

{z : |z| > 1, Im(z) > 0}

is a conformal equivalence from this set to the upper half-plane. Hint: Use the
result of the previous exercise.
10. Find a solution to the Neumann problem for the set consisting of the exterior
of the unit disc. Hint: Use the result of Exercise 9.
11. Let U and V be open sets in the plane and let h : U → V be analytic. Prove
that u ◦ h is harmonic on U for every harmonic function u on V .
12. Suppose U is a bounded open set and g is a continous function on ∂U . Prove
that if u is a solution to the Dirichlet problem on U with boundary function
6.6. The Dirichlet Problem 219

g, then, for any real constants a and b, au + b is a solution to the Dirichlet

problem on U with boundary function ag + b.
13. Prove the second equality in (6.6.2). Hint: Express 1 − z 2 in polar form:
1 − z 2 = r eiθ = r(cos θ + i sin θ),
with r = |1 − z 2 | and θ = ± cos−1 (Re(1 − z 2 )/r).
 Chapter 7

Analytic Continuation and the

Picard Theorems

It often happens in solving complex variables problems, particularly where the

solution is obtained as the sum of a series, that we are led to an analytic function
f defined, as far as we know, just on a certain open set U . Yet it may turn out
that f can be extended to be an analytic function on a much larger open set. For
example, the power series
∞

zn
n=0

converges on the open unit disc D to a function analytic on D. In fact, this function
is
1
f (z) = ,
1−z
on D, and it can obviously be extended to a function, given by the same algebraic
expression, which is analytic on the much larger set C \ {1}.
According to the Inverse Function Theorem, a function f which is analytic on
an open set U and which has a non-zero derivative at a point z0 ∈ U has an analytic
inverse function defined in some neighborhood W of f (z0 ). Although the theorem
assures us of an inverse function just in some possibly very small neighborhood
of f (z0 ), we would naturally like to know on just how large a set can the inverse
function be defined.
There are many problems of this nature, problems where a solution in the form
of an analytic function is known or is known to exist locally, in a neighborhood of
a point, and the question arises as to just how much we can enlarge the domain
on which this function is defined, is analytic, and is still a solution to the original
problem. Questions of this type are questions of analytic continuation.
One method of analytic continuation, using the Schwarz Reflection Principle,
will at the end of the chapter lead to the existence of a special analytic function (an

221
222 7. Analytic Continuation and the Picard Theorems

Figure 7.1.1. Continuation by Reflection.

analytic covering map) from the unit disc onto C \ {0, 1}. Using this function, one
can turn an entire function with two distinct points of the plane not in its range
into an entire function with range contained in the unit disc. Such a function must
be constant. This leads directly to the proofs of the Big and Little Picard theorems.

7.1. The Schwarz Reflection Principle

One way to extend an analytic function to a larger domain, is to take advantage of
some kind of symmetry operation. The Schwarz Reflection Principle does exactly
this by making use of the symmetry operation z → z, which sends a point to its
reflection through the x axis.

Schwarz Reflection. Figure 7.1.1 illustrates the situation described in the next
theorem. In what follows, if V is a subset of C, then V − will denote {z : z ∈ V }.
Theorem 7.1.1 (Schwarz Reflection Principle). Let U be an open set which is
symmetric about the x-axis R. Let A = U ∩ R and let V be the part of U in the
open upper half-plane. If f is a function which is continuous on V ∪ A, analytic on
V , and real-valued on A, then f can be continued to a function analytic on all of
U by setting its continuation equal to f (z) on V − .

Proof. The function g, defined by

g(z) = f (z),
−
is analytic on the set V (Exercise 2.2.15).
Note that V − is the part of U which lies in the open lower half-plane, since
U is symmetric about the x-axis. Note also that the function g is continuous on
V − ∪ A, since f is continuous on V ∪ A.
The fact that f is real on A implies that f and g agree on A and, hence, that
they define a single continuous function on U = V ∪ A ∪ V − . This function is
analytic on U , except possibly at points of A.
7.1. The Schwarz Reflection Principle 223

Now A is an open subset of the real line and, hence, is a union of open line
segments. Let L be one of these line segments. Then the set W = V ∪ L ∪ V − is
an open subset of U and f is continuous on W and analytic on W except possibly
at points of L. However, it is a simple consequence of Morera’s Theorem that a
function which is continuous on an open set W and analytic on W except possibly
at the points of a line segment in W is actually analytic on all of W (Exercise
3.2.14). Since f is analytic on each of the open sets W = V ∪ L ∪ V − for L one of
the line segments making up A, it is anaytic on the union of these sets, and U is
this union. This completes the proof. 

Continuation Across a Simple Analytic Curve. Recall from our discussion of

the Riemann Sphere S 2 in Section 6.3 that lines in C can be thought of as circles
in S 2 that pass through the point at infinity. In fact, the set of all lines and circles
in C is just the set of all circles in S 2 .
Definition 7.1.2. An analytic curve is the image under a conformal equivalence
h : W → V , between domains in S 2 , of a set I ⊂ W which is a circle or an open
arc on a circle in S 2 .
Of course, by using a linear fractional transformation to move W , we can always
assume that the circle in the above definition is R ∪ {∞} and I is either an open
interval on the line or is R ∪ {∞} itself. In the latter case, the curve will be a simple
closed curve.
In the proof of the Schwarz Reflection Principle, we used the fact that a function
which is continuous on an open set U and analytic on U except possibly at the points
of a line segment in U is actually analytic on all of U (Exercise 3.2.14). We can
strengthen this as follows:
Theorem 7.1.3. If f is a function which is continuous on an open set U and
analytic on U except possibly at the points of a simple analytic curve C in U , then
f is actually analytic on all of U .

Proof. By removing a point p from C, if necessary, we obtain a simple analytic

curve C1 which is not a closed curve. Then there is a domain V containing C1 , a
domain W which contains an open interval L ⊂ R, and a conformal equivalence
h : W → V which maps L onto C1 . We may assume that V ⊂ U . Let h−1 : V → W
denote the inverse function of h.
The composition f ◦h is a continuous function on W which is analytic on W \L.
It is, therefore, analytic on all of W . Since, on V ,
f = f ◦ h ◦ h−1 ,
we conclude that f is analytic on all of V . Since it is analytic on each of two open
sets, V and U \ C, whose union is U \ {p}, f is actually analytic on all of U \ {p}.
Since it is continuous at p, it is analytic on all of U , by Theorem 3.4.8. 
Example 7.1.4. Prove that a function f which is continuous on an open set U
and analytic on U \ C, where C is an open arc of a circle, is actually analytic on
all of U .
Solution: An open arc of a circle is a simple analytic curve. Thus, by the
above theorem, f is analytic on all of U .
224 7. Analytic Continuation and the Picard Theorems

Reflection Through an Analytic Curve. The symmetry consisting of reflection

through the x-axis played the key role in the Schwarz Reflection Principle. It turns
out there is a similar reflection symmetry for each simple analytic curve and it leads
to a similar reflection principle. To describe this requires the notion of a conjugate
analytic function.
A function f on an open set U is said to be conjugate analytic if f is analytic
on U , where
f (z) = f (z).
Definition 7.1.5. Suppose U is a domain in C and C is a simple analytic curve
in U such that U \ C has two connected components V and W . If ρ is a conjugate
analytic function from U to U which fixes each point of C and interchanges V and
W , and if
(7.1.1) ρ ◦ ρ(z) = z for every z ∈ U,
then ρ is called a reflection through C defined on U .

Note that (7.1.1) implies that ρ is its own inverse function on U . If κ(z) = z,
then κ ◦ ρ and ρ ◦ κ are both analytic functions and are inverse functions of one
another. This implies that κ◦ρ is a one-to-one conformal map. Thus, each reflection
defined on U is the conjugate of a conformal equivalence from U to U − .
Theorem 7.1.6. If C is any simple analytic curve in C, then there is a reflection
ρ through C defined on some domain V containing C. This reflection is unique
in the sense that another reflection through C, defined on a neighborhood V1 of C,
must be equal to ρ on the connected component of V ∩ V1 which contains C.

Proof. Since C is a simple analytic curve, there is a conformal equivalence

h : W → V,
where W is a domain in S which meets R ∪ {∞} in a set L which is either R ∪ {∞}
2

or a line segment in R, V contains C, and C is the image of L under h.

Note that W ∩ W − also has L as its intersection with R ∪ {∞}. If W ∩ W −
is not connected, then exactly one of its connected components contains L. Thus,
we may as well assume that W − = W , since, otherwise, we can replace W by the
connected component of W ∩ W − containing L without affecting its intersection,
L, with R ∪ {∞}, or the fact that its image under h is a domain containing C.
With the assumption that W − = W , conjugation defines a reflection κ through
L on W (κ(z) = z). We then get a reflection ρ through C on V by setting
ρ = h ◦ κ ◦ h−1
on V , where h−1 : V → W is the inverse function for h. The mapping ρ is conjugate
analytic because it is the composition of an analytic function h with a conjugate
analytic function κ ◦ h−1 = h−1 (Exercise 7.1.7). It also satisfies
ρ ◦ ρ(z) = h ◦ κ ◦ h−1 ◦ h ◦ κ ◦ h−1 (z)
= h ◦ κ ◦ κ ◦ h−1 (z) = z
on V , and
ρ(z) = z
7.1. The Schwarz Reflection Principle 225

on C, because κ(z) = z on L and h−1 maps C to L. Thus, ρ is a reflection through

C defined on V .
If σ is any other reflection through C, defined on V1 , then σ ◦ ρ(z) is an analytic
function of z on V ∩ V1 , since the composition of two conjugate analytic functions
is analytic (Exercise 7.1.8). However, it is also equal to z on C, because this is
true of both σ(z) and ρ(z). But then it must be equal to z on all of the connected
component of V ∩ V1 containing C. This implies ρ = σ −1 = σ on this component.
Thus, ρ is unique in the sense described in the theorem. 

It is now a simple matter to prove that a version of the Schwarz Reflection

Principle holds for reflection through a simple analytic curve.

Theorem 7.1.7. If U is a domain, C ⊂ U is a simple analytic curve with U \ C

having two components V and W , and ρ is a reflection through C defined on U ,
then any function f which is continuous on V ∪ C, analytic on V , and real-valued
on C can be continued to an analytic function defined on all of U by setting its
continuation equal to f (ρ(z)) on W .

Proof. The proof is the same as the proof of the Schwarz Reflection Principle. We
set
g(z) = f (ρ(z))

on W ∪ C and note that g is continuous on W ∪ C and analytic on W because it

is the composition of two conjugate analytic functions, f and ρ (Exercise 7.1.8).
The functions g and f agree on C because ρ fixes points of C and f and g are
real-valued on C. Hence, they define a single function on U which is continuous on
U and analytic on U \ C. By Theorem 7.1.3, this function is analytic on all of U .
Since it agrees with f on V ∪ C, it is an analytic continuation of f to U . 

Example 7.1.8. What is the reflection through an arc on the unit circle and what
does the previous theorem say about continuing an analytic function across such
an arc?
Solution: The map

1 1 iθ
ρ(z) = or ρ(r eiθ ) = e
z r
is defined and conjugate analytic on C \ {0}. It satisfies ρ ◦ ρ(z) = z. It also fixes
each point on the unit circle since, for z on the unit circle, |z|2 = zz = 1, which
implies z = 1/z. It follows that, for any domain U which meets the unit circle in
an arc C and is taken to itself by ρ, the unique reflection through C, defined on U ,
is the map ρ.
For such a U and C, the previous theorem implies that any function analytic
on the part V of U lying on one side of C, continuous on V ∪ C, and real-valued
on C can be analytically continued to an analytic function on all of U .
226 7. Analytic Continuation and the Picard Theorems

Exercise Set 7.1


1. The power series ∞ n
n=0 (n+1)z defines an analytic function g on the open unit
disc. What is the largest domain in the plane to which g can be analytically
continued?
2. Suppose that the hypotheses for the Schwarz Reflection Principle (Theorem
7.1.1) are satisfied for f , U , V and A, except that the function f is purely
imaginary on A rather than purely real. Is it still true that f can be analyt-
ically continued to all of U ? If so, prove it, and describe the formula for the
continuation of f to V − .
3. If L is the line with equation ax + by = 0, find a formula for the map which is
reflection through
√ L, and verify that this map is conjugate analytic.
4. The function z 2 − 1 is defined and analytic on the set V consisting of the
open right half-plane with the interval (0, 1] removed, provided we use the
principal branch of the log function to define the square root (this is because
z 2 − 1 lies in the complement of the non-negative real axis if z ∈ V ). This
function can be extended to be continuous on the closed right half-plane with
the interval [0, 1] removed. Show how to use the Schwarz Reflection Principle
to extend this function to an analytic function on C \ [−1, 1].
5. Formulate and prove a version of the Schwarz Reflection Principle (Theorem
7.1.1) for meromorphic functions.
6. Formulate and prove a version of the Schwarz Reflection Principle for harmonic
functions.
7. Prove that the composition f ◦ g of an analytic function f with a conjugate
analytic function g is conjugate analytic.
8. Prove that the composition f ◦g of two conjugate analytic functions is analytic.
9. Find a formula for the reflection through the circle of radius r centered at z0 .
10. Prove that the curve in the plane with equation y = x2 is a simple analytic
curve.
11. Prove the following form of the Schwarz Reflection Principle: Suppose C1 and
C2 are circles and the corresponding reflections through these circles are ρ1
and ρ2 . If U is a domain containing C1 which is symmetric under ρ1 , and V is
the part of U which lies on one side of C1 (inside or outside), then any analytic
function f on V which is continuous on V ∪ C1 and maps C1 to C2 may be
analytically continued to a function defined on all of U by setting it equal to
ρ2 ◦ f ◦ ρ1 on ρ1 (V ) (the part of U which lies on the other side of C1 ).
12. The power series ∞ n=0 z
n!
defines an analytic function f on the open unit
disc. Prove that f cannot be analytically continued to any larger open set.
Hint: Consider the values of f along rays of the form z = r e2πip/q , where p
and q are integers.

7.2. Continuation Along a Curve

Suppose f is a function which is analytic on a domain U . One way to try to extend
f to a larger domain is to consider its power series expansion about a point z0 near
7.2. Continuation Along a Curve 227

r
z
.
0 D

U U∪D

Figure 7.2.1. Continuation to a Disc Overlapping the Boundary.

the boundary of U . Suppose this power series converges on disc D, centered at z0 ,

of radius r. Then it defines an analytic function g on D. Suppose that r is greater
than the distance from z0 to the boundary of U , so that D overlaps the boundary
of U , and suppose that U ∩ D is connected. We know the power series expansion
converges to f on any disc centered at z0 which is contained in U and, hence, g
must agree with f on such a disc. It follows from the Identity Theorem (Theorem
3.4.4) that g agrees with f on all of U ∩ D. Thus, we succeed in extending f to an
analytic function on the larger domain U ∪ D by setting it equal to g on D.
It may be possible to continue the above procedure, extending the domain on
which f is analytic in various directions to ever larger domains until something gets
in the way. This could be a singularity where f blows up or it could be a place where
to extend the domain in a given direction would cause the function to be defined in
a new way on a set where it was already defined. We can imagine extending along
different “fingers” from U and having these eventually overlap resulting in different
definitions for f on the overlap. It seems that this process has the potential to
become quite complicated, maybe hopelessly so.
The concept of analytic continuation along a curve and the Monodromy The-
orem serve to bring order to this process and to show that it is not so complicated
as it seems. We will get to these ideas shortly, but first we need to introduce some
useful terminology.
An analytic function element, (f, D), is an analytic function f defined on an
open disc D. We will say that it is an analytic function element at w if w ∈ D.
Two analytic function elements (f1 , D1 ) and (f2 , D2 ) at w are said to be equivalent
at w if f = g on D1 ∩ D2 . In particular, an analytic function element (f, D) at w
is equivalent to each function element (f1 , D1 ) where w ∈ D1 ⊂ D and f1 is the
restriction of f to D1 . In other words, the disc in an analytic function element at
w is not important as long as it contains w and the function f is analytic on it.

Definition 7.2.1. Let γ : [0, 1] → C be a curve and (f0 , D0 ) an analytic function

element at z0 = γ(0). Suppose there exist
(a) a partition 0 = t0 < t1 < · · · < tn+1 = 1 of [0, 1]; and
228 7. Analytic Continuation and the Picard Theorems

Figure 7.2.2. Continuation Along a Curve.

(b) a sequence of function elements

(f1 , D1 ), (f2 , D2 ), · · · , (fn , Dn ),
with γ([tj , tj+1 ]) ⊂ Dj for j = 0, · · · , n and fj = fj+1 on Dj ∩ Dj+1 for
j = 0, · · · , n − 1.
Then we will say that (fn , Dn ) is an analytic continuation of (f0 , D0 ) along γ (see
Figure 7.2.2).

Although the function element (fn , Dn ) of the above definition, seems to de-
pend on the choice of partition 0 = t0 < t1 < · · · < tn+1 = 1 of [0, 1] and sequence
(f1 , D1 ), (f2 , D2 ), · · · , (fn , Dn ) of function elements, it turns out that, up to equiv-
alence, it is actually independent of these choices.
Theorem 7.2.2. Given a curve beginning at z0 and ending at w and an analytic
function element (f0 , D0 ) at z0 , any two analytic continuations of (f0 , D0 ) along γ
are equivalent as analytic function elements at w.

Proof. We first observe that, given an analytic continuation of (f0 , D0 ) along γ,

involving a particular partition 0 = t0 < t1 < · · · < tn+1 = 1 of [0, 1] and corre-
sponding function elements (f, D) = (f1 , D1 ), (f2 , D2 ), · · · , (fn Dn ), we can replace
the original partition with any given refinement 0 = s0 , s1 , · · · , sm+1 = 1 and mod-
ify the sequence of function elements so that this new set of data still determines
the same analytic continuation of (f0 , D0 ). In fact, if [tj , tj+1 ] is partitioned into
k subintervals in the new partition, we simply assign (fj , Dj ) to each of these new
subintervals (relabeling them in accordance with the labeling of the subintervals in
the new partition). Obviously, this does not change the function element assigned
to the last subinterval.
Given two analytic continuations of (f0 , D0 ) along γ, by passing to a common
refinement of the corresponding partitions, we may assume that the two continua-
tions are defined using the same partition
0 = s0 < s1 < · · · < sm+1 = 1.
Suppose one of them is defined from the sequence of function elements
(f1 , D1 ), · · · , (fm , Dm ),
7.2. Continuation Along a Curve 229

while the other is defined from

(f˜1 , D̃1 ), · · · , (f˜m , D̃m )
(we set (f˜0 , D̃0 ) = (f0 , D0 )). If we set Vj = Dj ∩ D̃j for j = 1, · · · , m, then, for
each j, Vj is a connected (in fact, convex) open set containing γ(tj ) and γ(tj+1 ).
To prove the theorem we need to show that fm = f˜m on Vm . In fact, we will prove
by induction on j that fj = f˜j on Vj for each j.
We have f0 = f˜0 on all of V0 by definition. Suppose j < m and fj = f˜j on Vj .
We have fj+1 = fj on Dj ∩ Dj+1 and f˜j+1 = f˜j on D̃j ∩ D̃j+1 . Since γ(tj ) is in
both of these open sets, it follows that fj+1 = f˜j+1 in a neighborhood of γ(tj ) and,
hence, on all of Vj+1 , since Vj+1 is connected. This completes the induction step
and finishes the proof. 

The next theorem is almost obvious, but there is some work to be done to show
the required chain of overlapping discs can be chosen as in Definition 7.2.1. We
leave the details to the exercises (Exercise 7.2.1).
Theorem 7.2.3. Suppose (f0 , D0 ) is a function element at z0 and γ is a curve
joining z0 to w. If there is an open set U , containing D0 and γ(I), and an analytic
function f on U such that f = f0 on D0 , then (f0 , D0 ) can be analytically continued
along γ.

The converse of the above theorem is not true. It may seem that it should be
true, since the union of the overlapping discs in Definition 7.2.1 is an open set U
containing γ(I), and it appears that the functions fj fit together to define a single
function, on this union, that agrees with f0 on D0 . However, this is not the case,
due to the fact that the curve γ may cross itself, as in Figure 7.2.2, and the crossing
point may be contained in two of the discs Dj and Dk . The corresponding functions
fj and fk may not be equivalent as function elements at this point. In fact, the
curve might be a closed curve, so that z0 = γ(0) and w = γ(1) are the same point.
In this case the analytic continuation of (f0 , D0 ) along γ leads to another function
element (fn , Dn ) at z0 . This function element need not be equivalent to (f0 , D0 ).
Example 7.2.4. Give an example of the phenomenon referred to in the previous
paragraph.
Solution: Let log denote the principal branch of the log function and consider
the analytic function element (log, D1 (1)). If γ is the unit circle traversed once
in the counterclockwise direction beginning at γ(0) = γ(1) = 1, then (log, D1 (1))
can be analytically continued along γ (Exercise 7.2.2), but the resulting function
element on the final disc will differ from that on the first disc by 2πi.
There is no domain containing the unit circle to which (log, D1 (1)) can be
analytically continued. If there were such a domain U and a continuation f of
(log, D1 (1)) to U , then f would have to agree with the principal branch of the log
function on all of U \ [−∞, 0], but this function has a 2πi jump discontinuity across
the half-line [−∞, 0] and certainly cannot be analytically continued across it.

The previous example is closely related to the following question: If γ1 and γ2

are two curves joining z0 to w, (f0 , D0 ) is an analytic function element at z0 , and
230 7. Analytic Continuation and the Picard Theorems

(f0 , D0 ) can be analytically continued along γ1 and along γ2 , then are the resulting
function elements at w equivalent? Not necessarily, as the following example shows:
Example 7.2.5. Can analytic continuations of a function element at z0 along
different paths from z0 to w lead to non-equivalent function elements at w?
Solution: Let the initial function element be (log, D1 (1)), where log is the
principal branch of the log function, z0 = 1, and w = −1. Clearly this can be
continued along the curve γ1 consisting of the upper half of the unit circle traversed
from 1 to −1 and along the curve γ2 consisting of the lower half of the unit circle
traversed from 1 to −1 . However, the resulting function elements at −1 = γ1 (1) =
γ2 (1) differ by 2πi.

The Monodromy Theorem. The next theorem gives conditions under which
we can be sure that the analytic continuations of a function along two curves from
z0 to w are necessarily equivalent function elements, in contrast to the preceding
example.
Theorem 7.2.6 (Monodromy Theorem). Let U be a connected open set in C, z0
and w points of U , and (f0 , D0 ) an analytic function element at z0 , with D0 ⊂ U .
Suppose
(a) (f0 , D0 ) can be analytically continued along every curve in U ; and
(b) γ0 and γ1 are homotopic curves in U joining z0 to w.
Then the continuations of f along γ0 and γ1 are equivalent function elements at w.

Proof. We refer to terminology and notation developed in Section 4.6.

A homotopy from γ0 to γ1 in U determines a continuous one-parameter family
of curves {γs } from z0 to w in U . The continuity of this family means that, given
 > 0, there is a δ > 0 such that
||γs − γr || <  whenever |s − r| < δ.
The analytic function element (f0 , D0 ) has an analytic continuation along each
of the curves γs , by hypothesis. Denote the terminal function element for the
continuation along γs by φs . We claim that, for each r ∈ [0, 1], there is a δ > 0
such that φs is equivalent to φr whenever |s − r| < δ.
Let 0 = t0 < t1 < · · · < tn+1 = 1 be a partition and (f1 , D1 ), · · · , (fn , Dn ) a
sequence of function elements defining φr = (fn , Dn ) as an analytic continuation
of (f0 , D0 ) along γr . Then
γr ([tj , tj+1 ]) ⊂ Dj for j = 0, · · · , n.
For each j = 0, · · · , n, let j be the distance from the compact set γr ([tj , tj+1 ])
to the boundary of the disc Dj . If ||γs − γr || < j , then it will also be true that
γs ([tj , tj+1 ]) ⊂ Dj . Thus, if  = min{0 , · · · , n }, and we choose δ > 0 such that
||γs − γr || <  whenever |s − r| < δ,
then, for each s with |s − r| < δ, the partition 0 = t0 < t1 < · · · < tn+1 = 1 and
the sequence of function elements (f1 , D1 ), · · · , (fn , Dn ) also defines (fn , Dn ) as an
analytic continuation of (f0 , D0 ) along γs . Since, by the previous theorem, any
other continuation of (f0 , D0 ) along γs is equivalent to this one, we conclude that
φr is equivalent to φs . This proves that φs is equivalent to φr whenever |r − s| < δ.
7.2. Continuation Along a Curve 231

The remainder of the proof is a standard connectedness argument. We define

a function h on [0, 1] by setting h(s) = 0 if φs is equivalent to φ0 , and h(s) = 1
otherwise. Then, by the result of the previous paragraph, h is a continuous function
on [0, 1]. If h(1) = 1, then, since h(0) = 0, the Intermediate Value Theorem implies
that h must take on every value between 0 and 1. Since this is not the case, we
conclude that h(1) = 0. This means that φ1 is equivalent to φ0 . 

Example 7.2.7. Consider the function element φ consisting of the principal branch
of the log function and the disc D1 (1). Set

A = {z ∈ C : |z| > 0, −π/2 < arg(z) < 3π/2}.

Show that the analytic continuations along any two curves from 1 to −1 in A have
equivalent terminal function elements at −1.
Solution: The branch of the log function defined by restricting arg(z) to lie
in the inteval (−π/2, 3π/2) agrees with the principle branch of the log function
on D1 (1). By Theorem 7.2.3, φ can be analytically continued along any curve in
A. The set A is simply connected and so any two curves in A from 1 to −1 are
homotopic. In view of the Monodromy Theorem, continuations of φ along two such
curves will have equivalent terminal function elements at −1.

The Monodromy Theorem allows us, in the next theorem, to conclude that if
a function element can be analytically continued along every curve in an open set
U and if the open set is simply connected, then there is an analytic continuation of
f to all of U .

Theorem 7.2.8. Suppose U is a simply connected open set and (f0 , D0 ) is an

analytic function element at z0 ∈ U (with D0 ⊂ U ). If (f0 , D0 ) can be analytically
continued along every curve in U , then there is a function f which is analytic on
U and equal to f0 on D0 .

Proof. If z ∈ U , then since any two curves from z0 to z in U are homotopic in

U , the Monodromy Theorem implies that any two terminal elements of analytic
continuations of (f0 , D0 ) along curves from z0 to z in U will be equivalent and,
hence, will determine the same analytic function in some neighborhood of z. Hence
all such analytic continuations determine the same function value at z. We define
f (z) to be this value.
Clearly f0 (z) = f (z) on D0 . It remains to prove that f is analytic on U . Let
w be a point of U , let γ be a curve in U joining z0 to w, and let (Dn , fn ) be
the terminal function element of some continuation of f along γ. If z ∈ Dn , then
(Dn , fn ) is also the terminal element of a continuation of (f0 , D0 ) along a curve γ1
from z0 to z in U – we simply extend γ to a curve γ1 , ending at z, by joining γ
with the line segment from w to z. It follows that f (z) = fn (z) on all of Dn , not
just when z = w. Since fn is analytic on Dn , f is also analytic on Dn . Since w was
an arbitrary point of U , we conclude that f is analytic everywhere on U . 
232 7. Analytic Continuation and the Picard Theorems

Exercise Set 7.2

1. Prove Theorem 7.2.3.

2. Prove that if log is the principal branch of the log function and D0 is any disc
centered at 1 and not containing 0, then (log, D0 ) can be analytically continued
along any curve γ in C \ {0}.
3. Prove that any analytic continuation of the element (log, D0 ) of the previous
exercise, along a curve γ in C \ {0}, yields a function element (f, D) at γ(1)
which is some branch of the log function restricted to D. That is, prove that
it satisfies ef (z) = z on D.
In the next four exercises U is a domain and (g0 , D0 ) is a function element in U
which can be analytically continued along a curve γ in U to a function element
(g, D) in U .
4. If f is an entire function, prove that (f ◦ g0 , D0 ) can be analytically continued
along γ to (f ◦ g, D).
5. If f is an entire function, h is a function analytic on U and f ◦ g0 = h on D0 ,
then f ◦ g = h on D.
6. If a linear differential equation on U , with analytic coeficients, has g0 for a
solution on D0 , then it has g for a solution on D.
7. If u is a harmonic function on U and u = Re(g0 ) on D0 , then u = Re(g) on D.
In the next three exercises, use Theorem 7.2.8 to give a proof, different from the
one in Theorem 4.6.16, that the indicated property holds for a simply connected
domain U . In each case, show there is a solution in a disc D0 centered at a point
z0 ∈ U and then show that this solution can be analytically continued along any
curve in U .
8. Suppose g is a non-vanishing analytic function on U . Prove that g has an
analytic square root on U .
9. Prove that every analytic function on U has an analytic antiderivative.
10. Prove that every real-valued harmonic function on U is the real part of an
analytic function on U .

7.3. Analytic Covering Maps

If U , V and W are open subsets of C and
h:V →W and f :U →W
are analytic maps, then we say that f lifts through h if there is an analytic map
g:U →V with f = h ◦ g
(see Figure 7.3.1). If h is a conformal equivalence from V to W , then f trivially
lifts through h. In this case, h has an analytic inverse function h−1 : W → V and
we can simply set
g = h−1 ◦ f : U → V.
7.3. Analytic Covering Maps 233

g h
f
U W

Figure 7.3.1. Lifting a Map f Through Another Map h.

There is another class of maps h for which a kind of lifting result holds. This
is the class of analytic covering maps.
Definition 7.3.1. An analytic map h : V → W is called an analytic covering map
if, for each w0 ∈ W , there is a neighborhood A of w0 , contained in W , such that
h−1 (A) is the disjoint union of a collection {Bj } of open subsets of V with the
property that, for each j, h is a conformal equivalence of Bj onto A.
Example 7.3.2. Show that z → ez is an analytic covering map from C to C \ {0}.
Solution: The function exp, defined by exp(z) = ez , is certainly an analytic
map of C onto C \ {0}. We claim that each disc D in C \ {0} has inverse image
exp−1 (D) consisting of a disjoint union of open sets on each of which exp is a
conformal equivalence onto D.
Given a disc D ⊂ C \ {0}, there is a ray from 0 to ∞ disjoint from D (e.g.,
take the ray from 0 to ∞ which passes through the center of D and rotate it by the
angle π). This means that there is an angle θ0 such that every z for which ez ∈ D
satisfies
θ0 + 2n π < Im(z) < θ0 + 2(n + 1)π
for some integer n. Thus, exp−1 (D) is a disjoint union of open sets
Bn = exp−1 (D) ∩ {z : θ0 + 2n π < Im(z) < θ0 + 2(n + 1)π}.
On each Bn , exp is a conformal equivalence from Bn to D, with inverse function
equal to the branch of the log function for which arg takes values between θ0 + 2n π
and θ0 +2(n+1)π (see Figure 7.3.2). Thus, by definition, exp is an analytic covering
map.

An analytic covering map h : V → W is, in particular, a conformal map, since,

for each point z0 of V it is a conformal equivalence of a neighborhood of z0 onto a
neighborhood of h(z0 ). However, not every conformal map is a covering map.
Example 7.3.3. Give an example of a conformal map h of a region V onto a region
W such that h is not an analytic covering map.
Solution: We use the exponential map again. However, this time we restrict
its domain to be the set
V = {z : −π < Im(z) < 2π}.
The image of this map is still C \ {0}, and it is clearly a conformal map. However,
any disc D ⊂ C \ {0}, centered at 1, has an inverse image which is a disjoint union
of two open sets – one which lies in the strip −π < Im(z) < π and one which lies
234 7. Analytic Continuation and the Picard Theorems

(θ0+2π)i
B0
D
θ0 i exp
B−1 θ0

(θ0−2π)i
B−2
(θ0−3π)i
B−3

Figure 7.3.2. The Exponential as a Covering Map.

2πi
πi exp D

−πi

Figure 7.3.3. A Conformal Map Which is Not a Covering Map.

in the strip π < Im(z) < 2π. The map exp is a conformal equivalence of the first of
these onto D, but only maps the second one onto the lower half of D (see Figure
7.3.3). Since this problem persists no matter how small a disc we choose centered
at 1, the map exp is not an analytic covering map from V to C \ {0}.

Lifting Through an Analytic Covering Map.

Theorem 7.3.4. Let h : V → W be an analytic covering map and U a simply
connected domain in C. Then each analytic function f : U → W can be lifted
through h. That is, there is an analytic map g : U → V with f = h ◦ g.

Proof. We fix a z0 ∈ U and let w0 = f (z0 ). We then choose a neighborhood A0

of w0 such that h−1 (A0 ) is a disjoint union of open sets on each of which h is a
conformal equivalence onto A0 . We choose one of these, call it B0 , and denote by
h−1
0 the inverse of h : B0 → A0 . We then choose an open disc D0 , centered at z0
and contained in f −1 (A0 ). On this disc, we define a function g0 by
g0 = h−1
0 ◦ f.

This function serves to lift f through h, but only on the set D0 . The main work of
the proof is to show that the analytic function element (g0 , D0 ) can be analytically
continued along any curve in U beginning at z0 . If we can do this, then the Mon-
odromy Theorem implies that (g0 , D0 ) can be analytically continued to an analytic
7.3. Analytic Covering Maps 235

function defined on all of U (since U is simply connected). This function g serves

to lift f through h on all of U , since it satisfies the identity
f =h◦g
on D0 and, hence, on all of U , by Theorem 3.4.2.
So it remains to prove that (g0 , D0 ) can be analytically continued along any
curve in U beginning at z0 . Let γ : [0, 1] → U be such a curve. Let S be the subset
of I = [0, 1] consisting of those points s such that (g0 , D0 ) can be analytically
continued along γ as far as s. Clearly, if s ∈ S and 0 ≤ s1 ≤ s, then s1 ∈ S as
well. Also, if s ∈ S, then there is a chain of open discs along γ restricted to [0, s]
that serves to analytically continue (g0 , D0 ) along this curve to γ(s). If s < 1, then
there is an s1 > s such that s1 is also in the last disc in this chain of open discs.
This implies that (g0 , D0 ) can also be analytically continued along γ restricted to
[0, s1 ] and, hence, that s1 is also in S. We conclude that S is either all of [0, 1], in
which case the proof is complete, or it is a half-open interval of the form [0, r) with
r = sup S ≤ 1.
If S is a half-open interval [0, r), we let zr = γ(r) and choose a neighborhood
A of f (zr ) in V such that h−1 (A) is a disjoint union of open sets Bj , each of which
is conformally equivalent to A under h. We choose an open disc D ⊂ U such that
f (D) ⊂ A. This is possible because f is continuous. We also choose an s ∈ [0, r)
such that w = f (γ(s)) belongs to D. This is possible because D is open and f ◦ γ
is continuous.
Because s ∈ S, the function element (g0 , D0 ) may be analytically continued
along γ to a function element (gn , Dn ) with γ(s) ∈ Dn . Since f = h ◦ g0 on D0 , the
same thing will be true of the function element (gn , Dn ). That is,
f = h ◦ gn on Dn .
In particular, w = f (γ(s)) = h(gn (γ(s))). This means that gn (γ(s)) belongs to
h−1 (A) and, hence, to exactly one of the sets Bj , call it Bk . We define a new
function element (gn+1 , Dn+1 ) by choosing Dn+1 = D and setting gn+1 equal to
the composition of f : D → A with an inverse function for h : Bk → A. This new
function element certainly satisfies
f = h ◦ gn+1 on Dn+1 ,
but does it agree with gn on Dn ∩ Dn+1 ? It does because, not only is the point
gn (γ(s)) in Bk , but all of gn (Dn ∩Dn+1 ) is contained in Bk ; otherwise, the connected
open set Dn ∩ Dn+1 would be separated by gn−1 (Bk ) and the union of the sets
gn−1 (Bj ) for j = k. But this means that the two inverse functions for h used in the
definitions of gn and gn+1 agree on f (Dn ∩ Dn+1 ), which implies that gn and gn+1
agree on Dn ∩ Dn+1 . But now, since γ(r) is an interior point of Dn+1 , this implies
that (g0 , D0 ) can be analytically continued to γ(r) and, hence, that r ∈ S. This
contradicts the assumption that S has the form [0, r). The only other possiblility
is that S = [0, 1], and this means that (g0 , D0 ) can be analytically continued along
γ. 

The preceding theorem is, on the one hand, a powerful application of the Mon-
odromy Theorem and, on the other hand, is the key ingredient in our proofs of the
Picard theorems in the next section.
236 7. Analytic Continuation and the Picard Theorems

Exercise Set 7.3

1. Is the function h(z) = z 2 an analytic covering map of C onto C? Is it an

analytic covering map of C \ {0} onto C \ {0}? Justify your answers.
2. Give a proof, using Theorem 7.3.4, that if U is a simply connected open set
and f is a non-vanishing analytic function on U , then there is an analytic
logarithm of f – that is, an analytic function g on U such that f = eg .
3. Prove that if h : V → W is an analytic covering map from a connected open
set U to a simply connected open set V , then h is a conformal equivalence.
4. Prove that if h : V → W is an analytic covering map and U is any simply
connected open subset of W , then h−1 (U ) is a disjoint union of open sets on
each of which h is a conformal equivalence onto U .
5. Is there an analytic covering map from the unit disc to C \ {0}? Justify your
answer.
6. Prove that if h : V → W is an analytic cover and γ : [0, 1] → W is a curve,
then γ can be lifted through h in the sense that there is a curve λ : [0, 1] → V
such that γ = h ◦ λ.
7. Let p(z) be a polynomial of degree n in z and let w be a point of C. Prove
that the polynomial p(z) − w fails to have n distinct roots (that is, it has a
repeated root) if and only if w = p(z0 ) for some point z0 at which p (z0 ) = 0.
8. Let p(z) be a polynomial of degree n in z and let S be the set of all points
w ∈ C at which the polynomial p(z) − w fails to have n distinct roots. If
W = C \ S and V = p−1 (W ), prove that p : V → W is an analytic covering
map. Hint: Use the Inverse Mapping Theorem and the preceding exercise.
9. Prove that the polynomial p(z) = z 3 − 3z is an analytic covering map from
C \ {−1, 1} to C \ {−2, 2}.
10. Use the previous exercise to prove that if g is an analytic function on a simply
connected domain U and if g does not take on the values −2 and 2, then there
is an analytic function f on U which satisfies f 3 (z) − 3f (z) = g(z) on U .

7.4. The Picard Theorems

The Little Picard Theorem states that a non-constant entire function takes on every
value in C except possibly one. This is a vast generalization of Liouville’s Theorem
and an impressive application of the analytic continuation techniques we have been
developing in this chapter. The Big Picard Theorem states that an analytic function
with an essential sigularity at z0 takes on every complex value but one infinitely
often in every neighborhood of z0 .
The strategy for proving the Picard theorems is to construct an analytic cov-
ering map h from the unit disc to the plane with two points removed. The Little
Picard Theorem follows easily from this since any analytic function from the plane
to the plane with two points removed will lift through this covering map to an ana-
lytic function from the plane to the unit disc. By Liouville’s Theorem the only such
functions are constants. The Big Picard Theorem uses the existence of h along with
Montel’s Theorem. The construction of h relies heavily on analytic continuation
7.4. The Picard Theorems 237

C2
C1

Figure 7.4.1. Reflection Through an Arc of a Perpendicular Circle.

by reflection through a circle (Example 7.1.8 – the Schwarz Reflection Principle for
circles). We begin by proving a few simple facts about reflection through a circle.

Reflection Through a Circle. Suppose C1 and C2 are two circles in the plane
that intersect one another in two points. If the tangents to the two circles are
perpendicular at each of these two points, then we will say that the circles meet at
right angles (Figure 7.4.1). Actually, it turns out that if the tangents are perpen-
dicular at one of the points of intersection, then they are also perpendicular at the
other point of intersection (Exercise 7.4.1).
Theorem 7.4.1. If circles C1 and C2 meet at right angles, and A is the arc of C1
that lies inside C2 , then reflection through A maps C2 onto itself.

Proof. There is a linear fractional transformation h that maps C1 onto R ∪ {∞}

and maps the inside of C1 onto the upper half-plane. We can choose which point
on C1 is sent to ∞ and we choose it so that it is not on A; in particular, it is not
one of the points where C1 and C2 intersect.
The fact that linear fractional transformations are conformal maps (angle pre-
serving maps) and the fact that C1 and C2 meet at right angles mean that the
image of C2 under h meets the real line at two points and it has vertical tangents
at these points. The image of a circle under a linear fractional transformation is
either a line or a circle, and so h(C2 ) must be a circle which meets the real axis at
right angles. This implies that the real axis passes through a diameter of h(C2 ). It
follows that reflection through the x-axis, maps the inside of the circle h(C2 ) onto
itself.
Now reflection through the arc A may be described as h composed with reflec-
tion through the horizontal diameter of h(C2 ) followed by h−1 . Why? Well, this
certainly describes a reflection through A defined on the inside of C2 and, by the
uniqueness part of Theorem 7.1.6, this is the only reflection through A.
Since reflection through the x-axis takes the inside of h(C2 ) onto itself, reflection
through A takes the inside of C2 onto itself. 

Theorem 7.4.2. With C1 , C2 as above, the reflection through C1 takes any other
circle C, which meets C2 at right angles, to another circle which meets C2 at right
angles or to a line through the center of C2 .
238 7. Analytic Continuation and the Picard Theorems

Figure 7.4.2. The Curvilinear Triangle V0 .

Proof. A reflection through a circle is a linear fractional transformation followed

by conjugation (Exercise 7.1.9). It therefore takes a circle C to either a circle or a
line.
The reflection through C1 takes the inside of C2 onto itself. It follows that the
image of C under this reflection still meets C2 in two points.
A conformal map preserves angles, while conjugation reverses the orientation
of any angle – that is, it takes an angle to its negative. It follows that a reflection,
which is a conformal map followed by conjugation, is also an angle reversing map.
This implies that if two curves have perpendicular tangents at a point of inter-
section, then the same will be true of their images under a reflection. Therefore,
reflection through C1 takes C to a circle or line which also meets C2 at right angles.
The only way the image of C could be a line is if it passes through the center of
C2 . Thus, the reflection of C through C1 is another circle which meets C2 at right
angles or is a line through the center of C2 . 

√ √
An Analytic Covering Map. The three points −1, 1/2 + i 3/2, 1/2 − i 3/2
are equidistant points on the unit circle C. According to Exercise 7.42, we can join
each pair of these points with an arc of a circle that meets C at right angles. These
three arcs then bound an open curvilinear triangle V0 (see Figure 7.4.2).
The open set V0 is clearly simply connected. By the Riemann Mapping Theo-
rem, there is a conformal equivalence h : V0 → H, where H is the upper half-plane.
By Theorem 6.5.2, the map h extends to a continous map of the closure of V0 to
the closure of H in S 2 , which takes the boundary of V0 to the boundary of H in S 2 .
By composing with a linear fractional transformation, if necessary,
√ we √may choose
this map in such a way that it takes the points −1, 1/2 − i 3/2, 1/2 + i 3/2 to the
points ∞, 0, 1.
The function h is real-valued on each of the three “edges” of V0 . By the circle
version of the Schwarz Reflection Principle, h can be continued by reflection across
each of these edges. This results in h being defined and analytic in the set V1
described in Figure 7.4.3. Note that each of the three new cells on which h is
defined in this way is contained in the unit disc D (Theorem 7.4.1) and is bounded
by three circles that meet the unit circle C in right angles (Theorem 7.4.2).
7.4. The Picard Theorems 239

Figure 7.4.3. The Sets V1 , V2 and V3 . of Theorem 7.4.3.

Note also that, while h maps V0 to the upper half-plane, it maps each of the
new cells of V1 to the lower half-plane, since each is a reflection of V0 through one
of its edges.
We can now analytically continue h to a still larger domain V2 by reflecting
across each of the circular arcs that bound V1 . This results in h being defined in
the set V2 represented in Figure 7.4.3. Note that h now maps each of the new cells
in V2 into the upper half-plane. We can clearly continue this process by induction
to create an increasing sequence {Vn } of open sets to which h may be analytically
continued. The union of these open sets is D (Exercise 7.4.5), and so the result is
an analytic function h which maps the open unit disc D onto a subset of the plane
which contains the upper and lower open half-planes and the intervals (−∞, 0),
(0, 1), and (1, ∞) on the real line. The points ∞, 0, and 1 are not in the image
because every triangular cell that occurs in the above construction has all of its
vertices on the unit circle. Thus, h is an analytic map of D onto C \ {0, 1}.
Theorem 7.4.3. The analytic map h : D → C \ {0, 1}, described above, is an
analytic covering map.

Proof. Think of the disc D as being partitioned into light cells and dark cells,
separated by arcs of circles, as in Figure 7.4.3. The dark cells are mapped into the
upper half-plane by h and the light cells are mapped into the lower half-plane. Each
arc separating two of these cells is mapped to one of the intervals I1 = (−∞, 0),
I2 = (0, 1), I3 = (1, ∞).
For j = 1, 2, 3, let Pj be the union of the open upper and lower half-planes and
the open interval Ij . Then each Pj is an open set consisting of S 2 with a closed arc
removed.
Let Δ be any open disc in C \ {0, 1}. Then Δ lies entirely inside Pj for at
least one j. We fix a j for which this is true. The inverse image of Pj under h
consists of the union of all the light and dark open cells together with some of the
arcs separating them – those arcs which h maps to Ij . The collection of these arcs
is pairwise disjoint, and each one of them separates exactly one light cell from one
dark cell. The union of the arc and the two cells it separates is an open set on
which h is a conformal equivalence onto Pj . No two distinct open sets of this form
can overlap since, if they did, the overlap would have to be a cell with boundary
containing two arcs mapping to Ij under h. Let {Uk }∞ k=1 be the collection of open
240 7. Analytic Continuation and the Picard Theorems

sets of this form. This is a pairwise disjoint collection of open sets with union equal
to h−1 (Pj ). Also, 
h−1 (Δ) ⊂ Uk ,
k
and so h−1 (Δ) is the disjoint union of the open sets Bk = h−1 (Δ) ∩ Uk . The map h
is a conformal equivalence of Bk onto Δ for each k. It follows that h : D → C\{0, 1}
is an analytic covering map. 

The Little Picard Theorem.

Theorem 7.4.4 (Little Picard Theorem). If f is an entire function and there are
two distinct points in the plane that are not in the image of f , then f is constant.

Proof. If f does not take on the values z0 and z1 , then the function
f (z) − z0
z1 − z0
does not take on the values 0 and 1. Thus, we may as well assume that f itself has
this property. Then f is an analytic function from the simply connected set C to
C \ {0, 1}. Since the map h : D → C \ {0, 1} of the previous theorem is an analytic
covering map, Theorem 7.3.4 implies that f may be lifted through h to an analytic
function g : C → D such that
f = h ◦ g.
However, this means that g is a bounded entire function and, hence, is a constant,
by Liouville’s Theorem. This, of course, forces f to also be a constant. 

The Big Picard Theorem. The proof of the big Picard Theorem depends on
the following theorem. Recall the definition of a normal family (Definition 6.4.1).
Theorem 7.4.5. Let U be a connected subset of the plane. Then the set of analytic
functions on U with values in the set C \ {0, 1} is a normal family.

Proof. We will prove the theorem in the case where U is an open disc Δ. The
proof that the theorem for general U follows from this special case will be left to
the exercises.
Let Δ be any open disc and let z0 be its center. Let F denote the set of analytic
functions on U with values in C \ {0, 1}. Let {fn } be a sequence in F. We will show
that this sequence converges uniformly to ∞ on compact subsets of Δ or it has
a subsequence which converges uniformly on compact subsets of Δ to a function
analytic on Δ.
Suppose {fn } does not converge uniformly to ∞ on compact subsets of Δ.
Then there is a disc Δ1 with the same center and with Δ1 ⊂ Δ and an R > 0 such
that fn (Δ1 ) has elements with modulus less than R for infinitely many n. This
means that we can choose a subsequence of {fn } which itself has no subsequence
converging uniformly to ∞ on Δ1 . Without loss of generality, we may replace
{fn } by this subsequence. That is, we may assume that {fn } has no subsequence
converging uniformly to ∞ on Δ1 .
Let h : D → C \ {0, 1} be the analytic covering map of Theorem 7.4.3. Then,
since Δ is simply connected, Theorem 7.3.4 implies each fn can be lifted through
7.4. The Picard Theorems 241

h to an analytic function gn : Δ → D with fn = h ◦ gn . Furthermore, since the set

V1 in Figure 7.4.3 maps onto C \ {0, 1} under h, we can choose the gn in such a
way that gn (z0 ) ∈ V1 .
The sequence {gn } is a uniformly bounded sequence of analytic functions on
Δ and, hence, by Montel’s Theorem (Theorem 6.4.2), it has a subsequence which
converges uniformly on compact subsets of Δ to an analytic function g.
The function g has its image in D. However, unless g is constant, this image
must be an open set (by the Open Mapping Theorem – Theorem 4.5.8) and, hence,
must be contained in D. In this case f = h ◦ g is an analytic function on Δ with
values in C\{0, 1} and is the uniform limit on compact subsets of Δ of a subsequence
of {fn }.
On the other hand, if g is constant, then this constant must lie in the closure
of V1 , since each gn (z0 ) lies in V1 . This means the constant is either in D or is one
of the six points of V 1 which lie on the unit circle. This means that f = h ◦ g has
constant value 0, 1, or ∞.
Thus, each sequence {fn } has a subsequence which converges uniformly on
compact subsets of Δ to ∞ or to an analytic function. Since there is no subsequence
converging to ∞, it has a subsequence converging uniformly on compact subsets to
an analytic function. This completes the proof of the theorem in the case where U
is a disc. The general case follows from Exercise 7.4.11. 

Theorem 7.4.6 (Big Picard Theorem). Let f be a function which is analytic in a

neighborhood U of z0 except at z0 itself, where it has an essential singularity. Then
f takes on every value but one infinitely often in every neighborhood of z0 .

Proof. We may as well assume that z0 = 0. If f does not have the property stated
in the conclusion of the theorem, then there is a disc Dr (0) on which f fails to take
on at least two complex values. We may assume these values are 0 and 1, since
otherwise we may compose f with a linear fractional transformation which takes
the two values missed by f to 0 and 1.
Thus, we may assume f is an analytic function from Dr (0) \ {0} to C \ {0, 1}
with an essential singularity at 0. We define a sequence of functions with the same
properties by setting
fn (z) = f (z/n) for n = 1, 2, · · · .
By the previous theorem, this sequence converges uniformly on compact subsets of
Dr (0) to ∞ or it has a subsequence converging uniformly on compact subsets of
Dr (0) to an analytic function. In the first case, 1/f is bounded and, hence, has
a removable singularity at 0. This means it extends to an analytic function with
a zero of some finite order at 0. This is impossible, since it implies that f has
a pole at 0 rather than an essential singularity. The second case implies that f
itself is bounded and, hence, has a removable singularity at 0. This also violates
the hypothesis that f has an essential singularity at 0. Thus, our assumption that
f misses two values in some disc centered at 0 has led to a contradiction. This
completes the proof. 
242 7. Analytic Continuation and the Picard Theorems

Exercise Set 7.4

1. Prove that if two circles intersect in two points and their tangents are perpen-
dicular at one point of intersection, then the tangents are also perpendicular
at the other point of intersection.
2. Prove that, given two distinct points on a circle C1 , there is a unique circle C2
which meets C1 at right angles at these two points.
3. Prove that if Cr and CR are circles of radius r and R, respectively, which are
tangent at a point, then the reflection of CR through Cr is a circle of radius
rR
2R + r
if neither circle is inside the other and is
rR
2R − r
if Cr is inside CR .
4. The set Vn in the construction preceding Theorem 7.4.3 has boundary consist-
ing of a set of circular arcs of different sizes with endpoints on the boundary
of the unit disc. Prove by induction that the largest of these circular arcs has
radius (2n + 1)−1 R0 , where R0 is the radius of each of the three circular arcs
comprising the boundary of V0 . Hint: Use the calculation of the preceding
exercise.
5. Prove that the union of the open sets Vn , described in the paragraph preceding
Theorem 7.4.3, is the unit disc D. Hint: Use the result of the preceding
exercise.
6. Prove that if f and 1/f are both entire functions, then the image of f is exactly
C \ {0}.
7. Prove that if f is a non-constant entire function and b2 = 4ac, then the function
g(z) = af 2 (z) + bf (z) + c
must have a zero.
8. Is there a non-constant analytic function from the plane with one point re-
moved to the plane with two points removed? Justify your answer.
9. If U is a simply connected open set, show that the set of all analytic functions
on U with values in C \ {0, 1} is a normal family (see Section 6.4).
10. Suppose r < 1, R < 1 and h is a conformal equivalence from the annulus
AR = {z : R < |z| < 1} to the annulus Ar = {z : r < |z| < 1}. Suppose
also that h extends to a continuous map from AR to Ar which takes the unit
circle to itself and the disc of radius R to the disc of radius r. Prove that
h can be continued to a conformal equivalence of the unit disc onto the unit
disc that takes 0 to 0. Conclude from this that R = r. Hint: Do the analytic
continuation through a series of steps involving reflection through a circle, as
in Exercise 7.1.11; the first of these steps involves reflecting the annulus AR
through the circle of radius R.
11. Prove that if F is a family of analytic functions on a connected open set U
and if for each disc Δ in U the family of restrictions of elements of F to Δ is
a normal family, then F itself is a normal family.
7.4. The Picard Theorems 243

12. Prove that an entire function which is not a polynomial takes on every complex
value but one infinitely often.
 Chapter 8

Infinite Products

As is amply demonstrated by power series expansions, a highly useful technique

in complex analysis is to express an analytic function as an infinite sum of much
simpler functions. Likewise, it can be useful to express an analytic function as
an infinte product of simpler functions. This is especially true in the study of the
zeroes of analytic functions. For example, a polynomial p of degree n can be written
as a product
n
p(z) = a (z − zj ),
j=1

where {z1 , z2 , · · · , zn } are the zeroes of p. It turns out that product expansions
of a similar type (but with infinitely many factors) are possible for other analytic
functions.
Since the exponential function converts sums to products, we can expect that
the theory of infinite products will be closely related to the theory of infinite sums.

8.1. Convergence of Infinite Products

In the following discussion, log will be the principal branch of the log function.

Definition 8.1.1. If {uk } is a sequence of complex numbers and


n
(8.1.1) pn = uk ,
k=1

then we will say that the infinite product

∞

(8.1.2) uk
k=1

converges to the complex number p if limn→∞ pn = p.

245
246 8. Infinite Products

Theorem 8.1.2. If {uk } is a sequence of complex numbers, then the infinite product
(8.1.2) converges to a non-zero number p if and only if the infinite sum
∞

(8.1.3) log uk
k=1

converges to a number λ. In this case, p = eλ . Furthermore, if the infinite series

converges absolutely, then the infinite product is unchanged by a rearrangement of
the factors.

Proof. We have to be careful here, because the log function converts products to
sums only up to ±2πi. However, it is true that log uv = log u + log v if u and v
have positive real parts since, in this case, log u and log v have imaginary parts in
(−π/2, π/2) and log uv has imaginary part in (−π, π). Thus, log uv and log u+log v
cannot differ by a non-zero multiple of 2πi.
We define the partial products pn as in (8.1.1). If p = limn→∞ pn exists and is
non-zero, then
lim log(pn /p) = 0,
n→∞

since log is continuous at 1. In particular, there is an N such that

−π/4 < Im(log(pn /p)) < π/4 whenever n ≥ N.

It follows that pn /pm = (pn /p)(pm /p)−1 is in the right half-plane for n, m ≥ N . In
particular, un+1 = pn+1 /pn is in the right half-plane for n ≥ N . Thus,

log(pn+1 /pN ) = log((pn /pN )un+1 ) = log(pn /pN ) + log un+1

whenever n ≥ N . This equation and an induction argument beginning with n = N

show that
n
log(pn /pN ) = log uk
k=N +1

for all n > N . Since the left side of this equality converges as n → ∞, so does the
right side. This implies the convergence of the series (8.1.3).
Conversely, if this series converges and we let

n
λn = log un
k=1

be its nth partial sum, then the sequence {λn } converges to a number λ. Since

p n = eλn ,

and the exponential function is continuous, the sequence {pn } converges to eλ .

If the series (8.1.3) converges absolutely, then each of its rearrangements con-
verges to the same number. It follows that each rearrangement of the infinite
product (8.1.2) also converges to the same number. 
8.1. Convergence of Infinite Products 247

Uniform Convergence of Products. We will be primarily interested in infinite

products of analytic functions. In this situation, whether or not the product con-
verges uniformly is of critical importance. We say that the infinite product of a
sequence of functions {uk } converges uniformly on a set S if the sequence pn of
partial products converges uniformly on S.
Theorem 8.1.3. Let uk be a sequence of complex-valued functions defined and
bounded on a set S. If the series
∞

log uk (z)
k=1

converges uniformly to λ(z) on S, then the infinite product

∞

uk (z)
k=1

converges uniformly to eλ(z) on S.

Proof. Let λn (z) be the nth partial sum of the infinite sum and pn (z) the nth
partial product of the infinite product. Then the uniform convergence of the series
on S implies that λn (z)−λ(z) converges uniformly to 0 on S. Since the exponential
function is continuous at 0, this implies that
pn (z)
= eλn (z)−λ(z)
p(z)
converges uniformly to 1.
The fact that each pn is bounded on S implies that Re(λn ) is bounded above
on S. This and the uniform convergence imply that Re(λ) is bounded above on
S, which implies that p(z) = eλ(z) is bounded on S. It follows that pn = (pn /p)p
converges uniformly to p on S. 

Theorem 8.1.4. Let {ak (z)} be a sequence of complex-valued functions defined on

a set S. If the series
∞

(8.1.4) |ak (z)|
k=1

converges uniformly on S, then the infinite product

∞

(8.1.5) (1 + ak (z))
k=1

converges uniformly on S. Each rearrangement of the infinite product converges to

the same function. If the infinite product converges to p(z), then each zero of p(z)
is a zero, with the same order, of some finite product of the factors 1 + ak (z).

Proof. If |w| < 1/2, then (Exercise 8.1.1)

2
(8.1.6) |w| ≤ | log(1 + w)| ≤ 2|w|.
3
248 8. Infinite Products

If the series (8.1.4) converges uniformly on S, then there is a K such that |ak (z)| ≤
1/2 for k ≥ K and for all z ∈ S. If we use (8.1.6) with w = ak , it follows that one
of the two series
∞ ∞
| log(1 + ak (z))| and |ak (z)|
k=K k=K
converges uniformly on S if and only if the other one does also. Hence, if (8.1.4)
converges uniformly, then
∞
log(1 + ak (z))
k=K
converges uniformly and absolutely. By the previous two theorems, this implies the
uniform convergence of
∞

(1 + ak (z))
k=K
to a function on S with no zeroes. It follows that (8.1.5) converges uniformly on S,
the limit is unaffected by rearrangements of the factors, and each of its zeroes is a
zero, with the same order, of the product of the factors 1 + ak (z) for k < K. 
Example 8.1.5. Prove that the infinite product
∞

(8.1.7) (1 − z 2 /k2 )
k=1

converges uniformly on each bounded subset of C.

Solution: We have | − z 2 /k2 | ≤ R2 /k2 for all z in the disc DR (0). Since the
positive termed series
∞
 R2
k2
k=1
converges, it follows that the series
∞
 |z 2 |
k2
k=1

converges uniformly on DR (0). Hence, by the previous theorem, the infinite product
(8.1.7) also converges uniformly on DR (0) for each R and, hence, on each bounded
subset of C.

Logarithmic Derivative of a Product. If an analytic function f on an open set

U has an analytic logarithm g on U – that is, if f = eg on U with g analytic – then
g  = f  /f . The expression f  /f is independent of which logarithm is chosen for f .
Furthermore, as long as f is not identically zero on any component of U , f  /f exists
(as a meromorphic function on U ) even if f does not have an analytic logarithm on
U . Note that f cannot have an analytic or even a meromorphic logarithm in any
neighborhood of a point where it has the value zero (Exercise 8.1.4).
Definition 8.1.6. Let f be an analytic function on an open set U and suppose
that f is not identically 0 on any component of U . Then the meromorphic function
f  /f is called the logarithmic derivative of f on U .
8.1. Convergence of Infinite Products 249

Logarithmic derivative is quite a well-behaved notion. The logarithmic deriv-

ative of the product of two functions is the sum of their logarithmic derivatives
(Exercise 8.1.5). Furthermore, the following theorem states that logarithmic de-
rivative is preserved by uniform limits. The proof is left to the exercises (Exercise
8.1.6).
Theorem 8.1.7. Let {fn } be a sequence of analytic functions on a connected open
set U . If this sequence converges uniformly to f on U , then the sequence {fn /fn }
converges uniformly to f  /f on compact subsets of U \ S, where S is the set of
zeroes of f .

When applied to infinite products, this immediately implies the following corol-
lary.
Corollary 8.1.8. Let {uk } be a sequence of analytic functions on a connected open
set U . If the product
∞

f (z) = uk (z)
k=1
converges uniformly on compact subsets of U to a function f which is not identically
0, then the infinite sum
∞
uk (z)
uk (z)
k=1

converges uniformly to f /f on compact subsets of U \ S, where S is the set of
zeroes of f .
Example 8.1.9. Show that the function
∞ 
 
z2
f (z) = πz 1− 2
k
k=1

has a logarithmic derivative which can be written as

∞
f  (z) 1  2z
(8.1.8) = +
f (z) z z 2 − k2
k=1
or as
f  (z) 
n
1
(8.1.9) = lim .
f (z) n→∞ z−k
k=−n

Solution: Note that the infinite product in the expression for f converges
uniformly on each compact disc in the plane by Example 8.1.5.
By the previous theorem,
∞ ∞
f  (z) 1  −2z/k2 1  2z
= + = + .
f (z) z 1 − z /k
2 2 z z 2 − k2
k=1 k=1

This proves (8.1.8). Since

2z 1 1
= + ,
z 2 − k2 z−k z+k
250 8. Infinite Products

the nth partial sum of the series (8.1.8) can be rewritten as the sum that appears
in (8.1.9).

The logarithmic derivative of f , as computed in the above example, will be

used in the exercise set to prove that f (z) = sin πz. That is,
∞  
z2
(8.1.10) sin(πz) = πz 1− 2 .
k
k=1

Exercise Set 8.1

1. Prove that if w is a complex number with |w| ≤ 1/2, then

2
|w| ≤ | log(1 + w)| ≤ 2|w|.
3
2. Does the infinite product
∞  
1
1+
k
k=1
converge? How about the product
∞ 
 
1
1 + 3/2 ?
k=1
k
3. Show that the infinite product
∞ 
 z  z/k
1− e
k
k=1

converges uniformly on compact subsets of the plane.

4. Prove that if f is analytic on U and has a zero at z0 ∈ U , then there is no
meromorphic function g defined in a neighborhood V of z0 such that f = eg
on V .
5. Prove that the logarithmic derivative of the product f g of two analytic func-
tions is the sum of the logarithmic derivative of f and the logarithmic derivative
of g. Also prove the analogous statement for the quotient f /g.
6. Prove Theorem 8.1.7. Hint: First prove that it is true on any disc in U on
which f has no zeroes.
7. Prove that the logarithmic derivative of a meromorphic function f on C is also
a meromorphic function on C and is odd (even) if f is odd (even).
8. If f is the function defined in Example 8.1.9, prove that the logarithmic de-
rivative of f is an odd meromorphic function which is periodic of period 1.
Observe that the logarithmic derivative of sin πz has the same properties.
9. Prove that if f is the function of the previous exercise, and we set
sin(πz)
g(z) = ,
f (z)
then g is an entire function with no zeroes and, hence, has a logarithm h which
is entire. Then, sin(πz) = f (z) eh(z) .
8.2. Weierstrass Products 251

10. Prove that if f , g and h are the functions of the previous exercise, then the
logarithmic derivative of g is h (z) = π cot πz − f  (z)/f (z).
11. With h as above, prove that h is bounded on the strip 0 ≤ Re(z) ≤ 1 (use
(8.1.8)). Show that this implies it is bounded on the entire plane and, hence,
is constant.
12. With h as above, prove that h (0) = 0 and, hence, that h is identically 0 and
h is a constant. Then use the fact that limz→0 z −1 sin z = 1 to show that this
constant is 0. Conclude that
∞  
z2
sin(πz) = πz 1− 2 .
k
k=1

8.2. Weierstrass Products

In this section we will show that, given any sequence of points of an open set U ⊂ C,
with no limit point in U , there is an analytic function on U with exactly the points
of this sequence as its zeroes, with each zero having order equal to the number of
times it appears in the sequence. The analytic function will be constructed as an
infinite product of certain simple functions, each of which has exactly one zero.
These simple functions are constructed as follows.
For p = 0, 1, 2, · · · we define entire functions Ep (z) by E0 (z) = 1 − z and
2
/2+···+z p /p
Ep (z) = (1 − z) ez+z for p > 0.
Note that z + z /2 + · · · + z /p is the pth partial sum for the power series expansion
2 p

of − log(1 − z) about z = 0 and so, although Ep (1) = 0, the sequence Ep (z) will
converge uniformly to (1−z)(1−z)−1 = 1 on each disc of radius less than 1 centered
at 0. More precisely:
Theorem 8.2.1. Each Ep (z) is an entire function with the following properties:
(a) the only zero of Ep (z) occurs at z = 1;
(b) if |z| ≤ 1, then |Ep (z) − 1| < |z|p+1 .

Proof. Part (a) is obvious. To prove Part (b), we note that the derivative of
1 − Ep (z) is (Exercise 8.2.1)
(1 − Ep (z)) = −Ep (z) = z p ez+z
2
/2+···+z p /p
(8.2.1) .
Since this has a zero of order p at z = 0, the function 1 − Ep (z) has a zero of order
p + 1 at z = 0.
The function (8.2.1) has a power series expansion about 0 with all of its coeffi-
cients non-negative real numbers, since this is true of the exponential function and
the function z + z 2 /2 + · · · + z p /p. It follows that the function
1 − Ep (z)
h(z) =
z p+1
also has non-negative real numbers as coefficients for its power series expansion
about 0. This implies that the maximum value achieved by |h(z)| for |z| ≤ 1 is
h(1) = 1. Part (b) follows from this. 
252 8. Infinite Products

If f is an analytic function on U , then we will say that a sequence {zk } ⊂ U is

a list of the zeroes of f counting multiplicity if each zk is a zero of f , and if each
zero w of f occurs m(w) times in this sequence, where m(w) is the order of the
zero w.
Let {zk } be a sequence of non-zero complex numbers converging to ∞. The
next theorem shows how to use scaled versions of the functions Ep to construct an
entire function with this sequence as a list of its zeroes counting multiplicity. The
resulting product is called a Weierstrass product.
Theorem 8.2.2. Let A be a subset of C. If {zk } is a sequence of non-zero complex
numbers and {pk } is a sequence of integers such that
∞ 
 
 r pk +1
(8.2.2)   < ∞ for all r > 0,
 zk 
k=1

then the Weierstrass product

∞

(8.2.3) f (z) = Epk (z/zk )
k=1

converges uniformly on compact subsets of C to an entire function which has {zk }

as a list of its zeroes counting multiplicity.

Proof. By part (b) of the previous theorem, we have

 pk +1
z
|Epk (z/zk ) − 1| ≤   if |z| ≤ |zk |.
zk
The condition |z| ≤ |zk | must be satisfied for all sufficiently large k if the series
(8.2.2) converges. The theorem follows by applying Theorem 8.1.4 with ak (z) =
Epk (z/zk ) − 1. 

The Weierstrass Theorem.

Theorem 8.2.3. If {zk } is any sequence of complex numbers converging to infinity,
then there is an entire function with {zk } as a list of its zeroes counting multiplicity.

Proof. Suppose m of the zk are equal to 0 (m might be 0). We may as well

assume these are the first m terms of the sequence. Then {zk }∞
k=m+1 is a sequence
of non-zero complex numbers.
If R > 0, then, since zk → ∞, there is a K > m such that |zk | > 2R for all
k ≥ K. Then the series
  z k
 
 zk 
k=m+1
converges uniformly on |z| ≤ R by comparison with the geometric series with
ratio 1/2. Thus, the hypotheses of the previous theorem are satisfied if we choose
pk = k − 1 for each k. The resulting Weierstrass product
∞

Ep (z/zk )
k=m+1
8.2. Weierstrass Products 253

converges uniformly on compact subsets of C to an entire function which has

{zk }∞
k=m+1 as a list of its zeroes counting multiplicity. Then
∞

f (z) = z m Ep (z/zk )
k=m+1

is an entire function with {zk }∞

k=1 as a list of its zeroes counting multiplicity. 
Example 8.2.4. Find an entire function which has a zero of order k at each positive
integer k.
Solution: We construct a sequence
1, 2, 2, 3, 3, 3, 4, 4, 4, 4, · · ·
in which each positive integer n appears n times and the terms are arranged in
increasing order. Then, for this sequence {zk } and a given positive integer p, we
have
∞ ∞ ∞
1 1 1
= n = .
|zk |p+1 n=1
n p+1
n=1
n p
k=1
If we choose p = 2, then the right side is the convergent series
∞
1
2
.
n=1
n
The Weierstrass product for the sequences {zk } and {pk = 2} is
∞ 
 k
2 2
(1 − z/k) ez/k+z /(2k ) .
k=1

By Theorem 8.2.2 this infinite product converges to an entire function with the
required zeroes.

Weierstrass Factorization. The Weierstrass Theorem for the plane leads im-
mediately to the Weierstrass Factorization Theorem for entire functions:
Theorem 8.2.5. Let f be an entire function which is not identically zero. Let m
be the order of the zero of f at 0, and let {zk } be a list of the non-zero zeroes of
f counting multiplicity. Then there exist non-negative integers p1 , p2 , · · · and an
entire function h such that
∞

f (z) = eh(z) z m Epk (z/zk ).
k=1

The sequence {pk } may be chosen in any way which satisfies ( 8.2.2).

Proof. The product

∞

m
g(z) = z Epk (z/zk )
k=1
converges uniformly on compact sets if {pk } is chosen such that (8.2.2) holds (pk =
k − 1 is one choice which always works, but there may be better choices for a given
f ). Furthermore, the resulting function g has the same zeroes as f with the same
254 8. Infinite Products

multiplicities. Thus, f g −1 is an entire function with no zeroes (after removable

singularities are removed). It follows that
f g −1 = eh
for some entire function h. The theorem follows from this. 

In many cases, the sequence {pk } can be chosen to be constant.

Example 8.2.6. Find a Weierstrass factorization for sin(πz).
Solution: This function has a zero of order 1 at each integer and has no other
zeroes. Since
∞
 1
< ∞,
k2
k=1
the condition (8.2.2) holds if we choose pk = 1 for every k. Then the above theorem
tells us that
 
sin(πz) = eh(z) z E1 (z/k) = eh(z) z (1 − z/k) ez/k ,
k=0 k=0

where the product is over all non-zero integers k. Note that if the factors for k and
−k in this product are paired, the result is
(1 − z/k) ez/k (1 + z/k) e−z/k = 1 − z 2 /k2 .
We conclude from Exercise 8.1.12 that eh(z) = π, that

sin(πz) = πz (1 − z/k) ez/k
k=0

is a Weierstrass factorization of sin(πz), and that this factorization is equivalent to

the factorization
∞

sin(πz) = πz (1 − z 2 /k2 ).
k=1

The General Weierstrass Theorem. If C is replaced by an arbitrary non-

empty, proper open subset of S 2 , the analogue of Theorem 8.2.3 holds with only a
slightly more complicated proof.
Theorem 8.2.7. Let U be a non-empty, proper open subset of S 2 . If {zk } is any
sequence of points of U with no limit points in U , then there is an analytic function
f on U with {zk } as a list of its zeroes counting mulltiplicity.

Proof. Either U or its image under some linear fractional transformation will con-
tain ∞. Thus, we may as well assume ∞ ∈ U . Then the complement of U in S 2 is
a compact subset K of the plane.
Since {zk } has no limit point in U , the distance between zk and K must ap-
proach 0 as k → ∞. It follows that we may choose a sequence {wk } of points of K
such that lim |zk − wk | = 0.
We set  
∞
 zk − wk
f (z) = Ek .
n=1
z − wk
8.2. Weierstrass Products 255

The product converges uniformly on compact subsets of U , since lim |zk − wk | = 0

implies the uniform convergence on compact subsets of U of
∞  
 zk − wk k+1
  .
 z − wk 
k=1

The function f is analytic in U and has {zk } as a list of its zeroes counting multi-
plicity. 

Meromorphic Functions. On a connected open set U , the set of analytic func-

tions forms an integral domain – that is, it is a commutative ring with the property
that the product of two elements is zero if and only if one of them is zero. The set
of meromorphic functions of U forms a field – that is, a commutative ring in which
every non-zero element has an inverse. The next theorem shows that the field of
meromorphic function is actually the quotient field of the ring of analytic functions.
That is, every meromorphic function is the quotient f /g of two analytic functions.
Theorem 8.2.8. If U is a connected open subset of C, then each meromorphic
function on U has the form f /g, where f and g are analytic on U and g is not
identically zero.

Proof. Let h be a meromorphic function on U and let {zk } be a sequence consisting

of the poles of h, with each zk listed as many times as the order of the pole at zk .
By Theorem 8.2.7 there is an analytic function g on U with {zk } as a list of its
zeroes counting multiplicity. Then, after removing removable singularities, f = gh
is an analytic function on U . Thus, h = f /g with f and g analytic on U . 

The Mittag-Leffler Theorem. The Weierstrass Theorem (Theorem 8.2.7) gives

the existence of an analytic function with a specified list of zeroes counting multi-
plicity. The Mittag-Leffler Theorem is a companion theorem. It gives the existence
of a meromorphic function with a specified list of poles and principal parts. We
prove it only for discs, although it is true for general open sets.
Theorem 8.2.9. Let R be a positive number or ∞. Let S be a discrete set of
points of DR (0) and {hw : w ∈ S} a set of polynomials with no constant terms.
Then there exists a meromorphic function f with a pole at w with principal part
hk ((z − w)−1 ) for each w ∈ S and with no other poles.

Proof. We choose an increasing sequence of radii {rn } with rn → R, we let S1 be

the subset of S which lies in Dr1 (0) and, for n > 1, we let
Sn = {w ∈ S : rn−1 < |w| ≤ rn }.
Then, for each n, 
gn (z) = hk ((z − w)−1 )
w∈Sn
is a meromorphic function on the plane with a pole at w with the required principal
part for each w ∈ Sn and with no other poles.
We might hope to construct the function we are after by simply taking the
infinite sum of the functions gn . Unfortunately, there is no reason to think this
sequence should converge on DR (0). However, we can modify each gn , without
256 8. Infinite Products

changing its poles and principal parts, in such a way as to end up with an infinite
series which does converge.
For each n > 1, the function gn is analytic on an open set containing the closed
disc Drn−1 (0). Hence, it is the uniform limit on this closed disc of its power series
at 0. It follows that there is a polynomial pn such that
|gn (z) − pn (z)| < 2−n for |z| ≤ rn−1 .
If we set f1 = g1 and fn = gn − pn for n > 1, then, for each m > 1, the series
∞

fn (z)
n=m+1

converges uniformly to an analytic function on Drm (0). This means that

∞

f (z) = fn (z)
n=1

is defined as a meromorphic function on Drm (0) and has the required poles and
principal parts at those points of S which lie in this disc. Since this is true for each
m, and lim rm = R, f is meromorphic on all of DR (0) and has the required poles
and principal parts. 

Exercise Set 8.2

2 p
1. Show that the derivative of Ep (z) is z p ez+z /2+···+z /p .
2. Compute the logarithmic derivative of Ep (z).
3. Find an entire
√ function (given by a Weierstrass product) that has a zero of
order 1 at n for n = 1, 2, 3, · · · and no other zeroes.
4. Find an entire
√ function (given by a Weierstrass product) that has a zero of
order 2 at n for n = 1, 2, 3, · · · and has no other zeroes.
5. Find an entire function (given by a Weierstrass product) that has a zero of
order n at n2 for n = 1, 2, 3, · · · and has no other zeroes.
6. If f is an entire function, show that f = g n for some entire function g if and
only if the order of each zero of f is divisible by n.
7. Suppose f is an entire function such that {zk } is a list of its non-zero zeroes
counting multiplicity and suppose that
∞
 1
< ∞.
|zk |
k=1

Describe the simplest Weierstrass factorization of f .

8. Suppose f is an odd entire function, the order of the zero at 0 is m, and {zk }
is a list of the other zeroes of f counting multiplicity. Show that m is positive
and odd. If
∞
1
< ∞,
|zk |2
k=1
8.3. Entire Functions of Finite Order 257

prove that f has a factorization of the form

∞ 
 
z2
m h(z)
f (z) = z e 1− 2 ,
zk
k=1

where h is an entire function.

9. Show that, if U is any non-empty open subset of the plane, then there is an
analytic function on U which cannot be extended to be analytic on any larger
open set. Hint: Use the general Weierstrass Theorem to construct an analytic
function on U with a lot of zeroes.
10. Prove that, given a sequence {zk } of complex numbers converging to infinity
and a sequence {nk } of integers, there is an entire function f with given values
for f and its derivatives up to order nk at zk for each k. Hint: Use the
Mittag-Leffler and Weierstrass Theorems together.
11. Prove that if f1 and f2 are two entire functions with no common zeroes, then
there exist entire functions g1 and g2 such that
g1 f1 + g2 f2 = 1.
Hint: Use the Mittag-Leffler Theorem to show that an entire function g2 can
be chosen so that, at each zero of f1 , the function 1 − g2 f2 has a zero of order
at least as large.
12. Let f1 , f2 , · · · , fn be entire functions. Show that there are entire functions
h1 , h2 , · · · , hn , and u such that fj = uhj for j = 1, · · · , n and the functions
h1 , h2 , · · · , hn have no common zeroes. Hint: Use the Weierstrass Theorem.
13. Let f1 , f2 , · · · , fn be entire functions with no common zero. Use induction and
the preceding two exercises to show that there are entire functions g1 , g2 , · · · , gn
such that
g1 f1 + g2 f2 + · · · + gn fn = 1.
14. Those who are familiar with commutative ring theory may want to do this
exercise. Let E be the ring of entire functions. Show that the following ring-
theoretic properties of E are consequences of the preceding two exercises:
(a) every finitely generated ideal of E is a principal ideal;
(b) every finitely generated maximal ideal of E is of the form
Mw = {f ∈ E : f (w) = 0} for some w ∈ C.
15. The conclusions of the last four exercises actually hold for the ring of analytic
functions on any open subset of the plane. However, to prove them all in this
generality would require a stronger form of the Mittag-Leffler Theorem than
the one proved here. Prove these results for the largest class of open sets that
you can using the machinery developed in this text.

8.3. Entire Functions of Finite Order

Definition 8.3.1. An entire function f is said to be of finite order if there is a
number t such that
|f (z)| ≤ e|z|
t
258 8. Infinite Products

for all z with |z| sufficiently large. The infimum of all such numbers t is called the
order of f .
p
For each non-negative integer p, the function ez is an entire function of finite
order p. More generally:
Example 8.3.2. Show that eh(z) is an entire function of finite order p if h is a
polynomial of degree p.
Solution: If t > p, then limz→∞ |z|−t |h(z)| = 0. This implies that there is an
R > 0 such that
|h(z)| < |z|t for |z| > R.
Then
| eh(z) | ≤ e|z|
t
(8.3.1) for |z| > R.
Since such a statement is true for all t > p, by definition eh(z) has finite order at
most p.
On the other hand, if t < p, then limz→∞ |z|−t |h(z)| = +∞. Hence, there is no
R for which (8.3.1) holds. We conclude that the order of f is at least p and, hence,
is equal to p.

Non-Vanishing Entire Functions of Finite Order. It turns out that the func-
tions eh(z) of the preceding example are the only entire functions of finite order
which are non-vanishing.
To prove this, we will need the following theorem of Borel-Carathéodory relating
the growth of the real part of an analytic function to the growth of the the absolute
value of the function.
Theorem 8.3.3. Suppose 0 < r < R and let g be a function analytic on an open
set containing DR (0). Then
2r R+r
|g(z)| ≤ sup{Re(g(w)) : |w| = R} + |g(0)| if |z| ≤ r.
R−r R−r

Proof. We suppose first that g(0) = 0. We set

m = sup{Re(g(w)) : |w| = R}.
Note that the Mean Value Theorem for harmonic functions implies that m ≥ 0.
If |w| = R and u = Re(g(w)), then u ≤ m and
u − 2m ≤ u ≤ 2m − u.
Thus, |u| ≤ |2m − u|, from which it follows that
|g(w)| ≤ |2m − g(w)|,
since the numbers g(w) and 2m − g(w) have the same imaginary parts of the same
magnitude and have real parts u and 2m − u, respectively.
We conclude from the above that the function
g(z)
h(z) =
z(2m − g(z))
8.3. Entire Functions of Finite Order 259

satisfies the inequality

1
|h(w)| ≤ for |w| = R.
R
Since the analytic function h has a removable singularity at 0, this inequality holds
throughout the disc DR (0) by the Maximum Modulus Theorem. Thus,
|g(z)| 1
≤ whenever |z| = r,
r|2m − g(z)| R
which implies
r
|g(z)| ≤ (2m + |g(z)|).
R
If we collect terms involving |g(z)| on the left and divide by 1 − r/R, the result is
2r
|g(z)| ≤ m for |z| ≤ r.
R−r

This concludes the proof in the case where g(0) = 0. This general case follows
from applying this result to the function g0 (z) = g(z) − g(0). The details are left
to the exercises. 

Theorem 8.3.4. An entire function f with no zeroes has finite order p if and only
if p is a non-negative integer and f has the form

f (z) = eh(z) ,

where h is a polynomial of degree p.

Proof. In view of Example 8.3.2, we need only show that every non-vanishing
entire function f of finite order p has the above form.
Since f has no zeroes and the plane is simply connected, there is an entire
function h such that
f (z) = eh(z) for all z ∈ C.
Since f has finite order p, for each t > p there is an M > 0 such that

eRe(h(z)) = |f (z)| ≤ e|z|

t
for |z| ≥ M.

This implies
Re(h(z)) ≤ |z|t for |z| ≥ M.
We apply the previous theorem with r > M and R = 2r to conclude

|h(z)| ≤ 2|z|t + 3|h(0)| if |z| = r.

Since this is true for all r > M , Exercise 3.3.9 implies that h must be a polynomial
of degree at most t. Since t was an arbitrary number greater than p, we conclude
that h is a polynomial of degree at most p. If it were a polynomial of degree less
than p, then f would have order less than p. Hence, the degree of the polynomial
h is exactly p. This, of course, implies that p is a non-negative integer. 
260 8. Infinite Products

Canonical Products. Given a sequence {zk }, we let μ be the inf of the numbers
t such that
∞
1
(8.3.2) < ∞.
|zk |t
k=1
If there is no such t, then we set μ = ∞. The number μ is called the exponent of
convergence for the sequence {zk }.
If {zk } has finite exponent of convergence μ, then we can write down a con-
vergent Weierstrass product (8.2.3), using {zk }, in which the sequence {pk } is a
constant p. We choose p to be the smallest integer such that μ < p + 1. Then the
condition
∞
1
(8.3.3) <∞
|zk |p+1
k=1
is satisfied. Hence, by Theorem 8.2.2, the Weierstrass product
∞
(8.3.4) f (z) = Ep (z/zk )
k=1

converges. This is called the canonical product for the sequence {zk }.
The significance of the choice of p made for the canonical product is that, with
this choice, the resulting product is an entire function with order λ equal to the
exponent of convergence μ of the sequence {zk }. The next theorem yields part of
what is needed to prove this. The remainder of the proof will come in the next
section.
Theorem 8.3.5. The canonical product for a sequence {zk }, with finite exponent
of convergence μ, is an entire function of finite order λ ≤ μ.

Proof. We choose p to be the smallest integer such that μ < p + 1, and let t be
any number in the range μ < t < p + 1.
We claim that there is a positive constant A such that
t
(8.3.5) |Ep (z)| ≤ eA|z|
for all z.
If |z| ≤ 1/2, this follows from (8.1.6) with w = Ep (z) − 1 and Theorem 8.2.1.
These combine to show that
| log Ep (z)| ≤ 2|z|p+1 ≤ 2|z|t ,
and this implies (8.3.5) holds with A = 2.
If |z| > 1/2, then |z|k ≤ 2t−k |z|t , and so

p
Re(z k )
log |Ep (z)| = log |1 − z| +
k
k=1

p
≤ |z| + |z|k ≤ (p + 1)2t |z|t .
k=1
t
Thus, (8.3.5) holds with A = (p + 1)2 in this case.
8.3. Entire Functions of Finite Order 261

To prove the theorem, we note that, if f is given by the canonical product

(8.3.4), then by (8.3.5),
∞

eA|z/zk | = eB|z| ,
t t
|f (z)| ≤
k=1

where
∞

B=A 1/|zk |t .
k=1

The series in this expression converges because t is larger than the exponent of
convergence μ.
Since for any s > t, we have B|z|t ≤ |z|s for |z| sufficiently large, it follows that
f has finite order at most t. Since t was an arbitrary number strictly between μ
and p + 1, we conclude that f has order at most μ. 

One might guess, based on Theorem 8.3.4, that the order of an entire function
of finite order must be a non-negative integer. This is not the case, as is shown by
the following example.
Example 8.3.6. Find an entire function with finite order 1/2.
Solution: The function
∞  
sin πz z2
= 1− 2
πz k
k=1

has order 1 (Exercise 8.3.3). It seems reasonable that if we replace z 2 by z in this

product, then the result would be an entire function of order 1/2. In fact, the
resulting function has a zero of order 1 at k2 for each positive integer k and
∞
 1
<∞
(k2 )t
k=1

for every t > 1/2 and for no smaller values of t. Hence, the sequence {1/k2 } has
exponent of convergence 1/2. Since 0 is the smallest integer p such that 1/2 < p+1,
the preceding theorem implies that the canonical product
∞ 
 z 
f (z) = 1− 2
k
k=1

is an entire function of finite order at most 1/2.

In fact, it is easy to directly compute the order of f if we note that
√
sin π z
f (z) = √ .
π z
The expression on the right is entire and is independent of the choice of the square
root function because the function (πz)−1 sin πz is an even function. It is easy to
see from this that, since (πz)−1 sin πz has order 1, f has order 1/2 (Exercise 8.3.5).
262 8. Infinite Products

Exercise Set 8.3

1. Finish the proof of Theorem 8.3.3 by showing that, if it is true in the case
where g(0) = 0, then it is true in general.
2. Show that a polynomial has finite order 0.
3. Show that sin z, z −1 sin z, and cos z all have finite order 1.
4. If f is an entire function of order λ(f ), k is a non-negative integer, and g(z) =
f (z k ), then prove that λ(g) = kλ(f ), where λ(g) is the order of g. √
5. Prove that if g(z) is an even entire function of finite order λ and f (z) = g( √ z),
then f is an entire function of finite order λ/2. In particular, show that cos z
has order 1/2.
6. Prove that the order of the sum or product of two entire functions is less than
or equal to the maximum of the orders of the two functions.
7. What is the order of the entire function esin z ?
8. Suppose f is an entire function which satisfies the inequality |f (z)| ≤ |z||z| for
|z| sufficiently large. Prove that f has finite order at most 1.
9. Find the exponent of convergence of the following sequences: {2k }, {kr } (r >
0), {log k}.
10. Given an arbitrary non-negative real number μ, show that there is a sequence
of complex numbers {zk } with exponent of convergence μ.
11. Does the order of an entire function necessarily have to be the same as the
exponent of convergence of its sequence of zeroes? Justify your answer.

8.4. Hadamard’s Factorization Theorem

Our goal in this section is to complete the characterization of entire functions of
finite order λ. We will prove a theorem of Hadamard which asserts that every such
function factors as a power of z times a canonical product of order at most λ times
the exponential of a polynomial of degree at most λ. The key ingredient in the
proof is Jensen’s Formula relating the density of the zeroes of an entire function to
the rate of growth at infinity of the function.

Jensen’s Formula.
Theorem 8.4.1. If f is analytic in an open set containing the disc Dr (0), f has
no zeroes on the boundary of this disc, f (0) = 0, and z1 , z2 , · · · , zn are the zeroes,
counting multiplicity, of f in Dr (0), then
   2π
|f (0)|r n 1
log = log(|f (r ei θ)|) dθ.
|z1 | · |z2 | · · · |zn | 2π 0

Proof. We first prove this in the case where r = 1. We divide f by a product of

linear fractional transformations which preserve the unit circle and have zeroes at
the points zi . This yields a function
1 − z1z 1 − z2 z 1 − zn z
g(z) = f (z) ··· .
z − z1 z − z2 z − zn
8.4. Hadamard’s Factorization Theorem 263

This function is analytic and non-vanishing in an open set containing the closed
unit disc D, and has the same modulus on the unit circle as does f . Thus, g has an
analytic logarithm in an open set containing D. Then log |g(x)| is the real part of
an analytic function in this set and, hence, is harmonic. The Mean Value Theorem
for harmonic functions implies that
   2π
|f (0)| 1
(8.4.1) log = log |g(0)| = log |f (eiθ )| dθ.
|z1 | · |z2 | · · · |zn | 2π 0

To prove the theorem for general r, it suffices to apply (8.4.1) with f replaced
by the function f (rz). If f has zeroes at z1 , z2 , · · · , zn in the disc Dr (0), then
f (rz) has zeroes z1 /r, z2 /r, · · · , zn /r in the unit disc D. Thus, the equation of the
theorem follows directly from (8.4.1) applied to f (rz). 

This leads to the following estimate on the number of zeroes of an entire function
inside a disc Dr (0).
Theorem 8.4.2. If f is an entire function with |f (0)| = 1, n(r) is the number
of zeroes of f inside a disc Dr (0), and M (2r) is the supremum of |f (z)| on the
boundary of D2r (0), then
log M (2r)
n(r) ≤ .
log 2

Proof. Let n = n(r) and m = n(2r), and let z1 , z2 , · · · , zm be the zeroes of f

inside the disc D2r (0) ordered so that |zj | ≤ |zk | for j ≤ k. Then Jensen’s Theorem
with r replaced by 2r implies that
 
 2r 2r 2r 2r 

log f (0) ··· ··· ≤ log M (2r).
z1 z2 zn zm 
Since
2r 2r
2< if j≤n and 1< if n < j ≤ m,
|zj | |zj |
this implies that
log |f (0)2n | ≤ log M (2r), or log |f (0)| + n log 2 ≤ log M (2r).
Since log |f (0)| = 0, it follows that
log M (2r)
n≤ ,
log 2
which completes the proof. 

Zeroes of Functions of Finite Order. The preceding theorem has the following
consequence for entire functions of finite order.
Theorem 8.4.3. Let f be an entire function of finite order λ and with f (0) = 0.
Let {zk } be a list of the zeroes of f , counted according to multiplicity and indexed in
the order of increasing modulus, and let μ be the exponent of convergence of {zk };
then μ ≤ λ.
264 8. Infinite Products

Proof. We claim that, for each t > λ, there are constants N , C > 0 and q > 1
such that
(8.4.2) |zk |t ≥ Ckq for all k ≥ N.
Assuming this, we conclude that the series
∞
 1
|zk |t
k=1

converges for all t > λ, by comparison with the series

∞
 1
,
kq
k=1

which converges for q > 1. This, in turn, implies the exponent of convergence μ is
at most λ.
To complete the proof, we must verify the claim concerning (8.4.2). In doing
this, we may as well assume that |f (0)| = 1, since, if this is not so, we may make it
so by replacing f by f divided by a constant times a power of z. Such a replacement
will have no effect on whether the above claim is true.
Let rk = |zk |. Since the zeroes are indexed in such a way that the modulus is
a non-decreasing function of k, there are at least k zeroes of f with modulus less
than or equal to rk . By Theorem 8.4.2,
log M (2rk )
k≤ ,
log 2
where M (2rk ) is the sup of |f (z)| on the circle |z| = 2rk .
We choose s with λ < s < t. Since f has order λ, there is an R such that
rk ≥ R implies
s
M (2rk ) ≤ e(2rk ) .
Hence, for rk ≥ R,
(2rk )s
k≤ .
log 2
This implies
(log 2)t/s t/s
rkt ≥ k = Ckq ,
2t
where
(log 2t/s ) t
C= and q = > 1.
2t s
This is true provided rk = |zk | > R. However, since lim zk = ∞, there is an N such
that k > N implies |zk | > R. This completes the proof. 

The above theorem, when combined with Theorem 8.3.5, yields the following
corollary.
Corollary 8.4.4. The canonical product for a sequence {zk } with exponent of con-
vergence μ has finite order λ = μ.
8.4. Hadamard’s Factorization Theorem 265

Hadamard’s Theorem. In the proof of the next theorem, we will need the
following estimates on the size of the inverse Ep−1 (z) of the function Ep (z).
Lemma 8.4.5. If p is a non-negative integer, p ≤ t ≤ p + 1, and z ∈ C, then there
is a constant A such that
1 t
(8.4.3) ≤ eA|z|
|Ep (z)|
if |z| ≥ 2 or |z| ≤ 1/2.

Proof. If |z| ≥ 2, then |1 − z| ≥ 1 and |z k /k| ≤ |z|t for k ≤ p. Hence,

|Ep−1 (z)| = |1 − z|−1 | e−z−z

2
/2−···−z p /p t
| ≤ ep|z|
and so (8.4.3) holds with A = p in this case.
On the other hand, the definition of Ep (z) and the Taylor series expansion of
log(1 − z) lead to
∞
 ∞

log Ep (z) = log(1 − z) + z k /k = − z k /k,
k=1 k=p+1

and so if |z| ≤ 1/2, then

∞

log |Ep−1 (z)| = − Re(log Ep (z)) ≤ |z|p+1 |z|k ≤ 2|z|p+1 ≤ 2|z|t .
j=0

Thus, (8.4.3) holds with A = 2 in this case. If we choose A = max{2, p}, then
(8.4.3) holds in both cases. 

We are now in a position to prove Hadamard’s Theorem characterizing entire

functions of finite order. This will be used in the proof of the Prime Number
Theorem in the next chapter.
Theorem 8.4.6. If f is an entire function of order λ, and p is the smallest integer
such that p + 1 > λ, then f factors as
∞

(8.4.4) f (z) = z m eh(z) Ep (z/zk ),
k=1

where m is the order of the zero of f at 0, {zk } is a list of the other zeroes of f
counting multiplicity, and h(z) is a polynomial of degree at most p.

Proof. According to the Weierstrass Factorization Theorem (Theorem 8.2.5) f has

a factorization of the form (8.4.4), where h is an entire function. Thus, the only
thing to be proved is that h is a polynomial of degree at most p. This will follow
from Theorem 8.3.4 if we can show that the function
f (z)
g(z) = eh(z) = ∞
zm j=1 Ep (z/zk )
has finite order at most λ.
266 8. Infinite Products

Let t be any number with λ < t ≤ p + 1 and let r ≥ 1 be any radius which is
not one of the numbers |zk |. We factor g(z) as g(z) = g1 (z)g2 (z), where

(8.4.5) g1 (z) = f (z)z −m Ep−1 (z/zk )
|zk |≤2r

and

(8.4.6) g2 (z) = Ep−1 (z/zk ).
|zk |>2r

Suppose |z| = 4r = R. Then |z/zk | ≥ 2 for all k with |zk | ≤ 2r. By the
previous lemma, there is a positive constant A1 such that

eA1 |z/zk | .
t
|g1 (z)| ≤ |f (z)|
|zk |≤2r

Since f has finite order λ, for sufficiently large r we have

|f (z)| ≤ e|z|
t

and, hence,
t
(8.4.7) |g1 (z)| ≤ eB1 r ,

where
 ∞

 1
t
B1 = 4 1 + A1 .
|zk |t
k=1

The infinite series in this expression converges by Theorem 8.4.3. Since g1 (z) is
an entire function (once the removable singularities at the zk with |zk | < 2r are
removed), if the inequality (8.4.7) holds for |z| = 4r = R, it must hold for all z in
the disc |z| ≤ R, by the Maximum Modulus Principle. In particular, this inequality
holds for all z with |z| = r.
Also if |z| = r, then |z/zk | < 1/2 if |zk | > 2r, and the previous lemma implies
that there is a constant A2 such that

eA2 |z/zk | ≤ eB2 r ,
t t
(8.4.8) |g2 (z)| ≤
|zk |>2r

where
∞
 1
B2 = A 2 .
|zk |
k=1

If we set B = B1 + B2 and combine (8.4.7) and (8.4.8), we obtain

t
|g(z)| ≤ eB|z| .

Since t is an arbitrary number larger than λ and less than or equal to p + 1, g has
order at most λ. This completes the proof. 
8.4. Hadamard’s Factorization Theorem 267

Exercise Set 8.4

1. What does Hadamard’s Factorization Theorem say about an entire function

of order λ < 1? √
2. If a non-constant entire function of finite order λ has zeroes at the points i n,
what are the possible values for λ?
3. Show that an even entire function of order 1 has the form
 z2

Cz m 1− 2 ,
zk
k
where m is even, C is a non-zero constant, and the sequence
{z1 , −z1 , z2 , −z2 , · · · , zk , −zk , · · · }
is a list of the zeroes of f counting multiplicity.
4. What is the exponent of convergence for the sequence of zeroes in the preceding
exercise?
5. State and prove the analogues of the previous two exercises for odd entire
functions of order 1.
6. Prove that if f is an entire function of order λ and λ is not an integer, then f
has infinitely many zeroes.
7. Under the hypotheses of the preceding exercise, prove that f takes on every
complex value infinitely many times.
8. Prove that if f and g are entire functions of finite order λ and if f (zk ) = g(zk )
on a sequence which satisfies
 1
=∞
|zk |t
k=1
for some t > λ, then f (z) = g(z) identically.
9. Use the previous exercise to prove that if two functions of finite order agree at
the points of the sequence {log n}∞ z
n=1 , then they agree identically. Thus, e is
the only entire function of finite order which has the value n at the point log n
for n = 1, 2, · · · .
10. Find an entire function which has a zero at each point of the sequence {log n}
for n = 1, · · · , ∞. Does it have finite order?
11. Suppose f is an entire function of finite order λ and μ is the exponent of
convergence of the list of zeroes of f . Prove that if μ < λ, then λ is an integer.
12. Is there an entire function of order 3/2 which has the integers as its list of
zeroes, counting multiplicity?
 Chapter 9

The Gamma and Zeta

Functions

This chapter is devoted to developing some of the properties of two special func-
tions of a complex variable – the gamma function and the zeta function. These
functions are of great importance in modern mathematics. The developement of
their properties provides a very instructive practical application of many of the
techniques developed in the preceding chapters.
The zeta function is the subject of one of the most famous unsolved problems
in mathematics – the Riemann Hypothesis. This conjecture arose from Riemann’s
attempt to settle an old conjecture concerning the rate of growth of the number
π(x) of primes less than or equal to x as the positive number x increases. In the
process, Riemann developed (but did not completely prove) a formula for π(x). This
formula involves the zeroes of the zeta function in the strip 0 < Re(z) < 1, and its
study led Riemann to conjecture that all these zeroes lie on the line Re(z) = 1/2.
If true, this would have been helpful in both the proof of Riemann’s Formula and
its use in analyzing the growth of π(x).
The methods introduced by Riemann eventually led to proofs by others of the
result on the growth of π(x) that he was seeking. This result is now known as
the Prime Number Theorem. These proofs use information about the location of
the zeroes of the zeta function, but not, of course, the information proposed in the
Riemann Hypothesis, since it has never been proved.
One of the reasons this chapter is included in the text is so that we may describe
the Riemann Hypothesis and its connection to the Prime Number Theorem. For
completeness, we conclude the chapter with a proof of the Prime Number Theorem.
This proof makes strong use of the results on infinite products presented in the
previous chapter.
We begin the chapter with a discussion of the gamma function.

269
270 9. The Gamma and Zeta Functions

9.1. Euler’s Gamma Function

We define Euler’s gamma function for Re(z) > 0 by the integral formula
 ∞
(9.1.1) Γ(z) = e−t tz−1 dt.
0

Of course, Γ is defined by an improper integral and so we must show that this

integral actually converges if Re(z) > 0. In fact, it not only converges, but the
resulting function of z is analytic.
Theorem 9.1.1. The integral (9.1.1) converges and defines an analytic function
Γ(z) for Re(z) > 0.

Proof. For 0 < r < s we define a function Γr,s on the right half-plane by
 s
Γr,s (z) = e−t tz−1 dt.
r

The function e−t tz−1 = e−t+(z−1) log t is continuous as a function of (t, z) in [r, s]×C
and is analytic in z for each fixed value of t. By Exercise 3.2.16, Γr,s is analytic
on the entire plane. We will show that as s → ∞ and r → 0 the functions Γr,s
converge uniformly on each strip of the form
S = {z : a ≤ Re(z) ≤ b} with 0 < a < b.
The limit function is then necessarily analytic on the right half-plane and is, by
definition, Euler’s function Γ.
If x = Re(z), then
| e−t tz−1 | = e−t tx−1 .
Thus, if z is in the strip S, then
(9.1.2) | e−t tz−1 | ≤ ta−1 for t ≤ 1.
If t ≥ 1, then
| e−t tz−1 | ≤ e−t tb−1 on S.
The function e−t tb+1 is continuous and has limit 0 at infinity. It is, therefore,
bounded on [1, ∞), by a positive number K. Thus,
(9.1.3) | e−t tz−1 | ≤ Kt−2 for t ≥ 1.
−2
Since t is integrable on (0, 1] and Kt is integrable on [1, ∞), inequalities (9.1.2)
a−1

and (9.1.3) imply that the improper integrals of e−t tz−1 on (0, 1] and on [1, ∞) both
exist (see Theorem 5.2.2). Hence, the improper integral defining Γ exists for each
z ∈ S.
To show that Γ(z) is analytic, we will show that Γr,s converges uniformly to Γ
on each strip of the form S as r → 0 and s → ∞. In fact, from (9.1.2) and (9.1.3)
we conclude
 r  ∞
|Γ(z) − Γr,s (z)| ≤ ta−1 dt + Kt−2 dt ≤ r a /a + K/s.
0 s

Given  > 0 the right side of this inequality is less than  whenever r < (a/2)1/a
and s > 2K/. It follows from this that Γr,s (z) converges uniformly to Γ(z) for
z ∈ S as r → 0 and s → ∞. This completes the proof. 
9.1. Euler’s Gamma Function 271

Analytic Continuation of Gamma. We will continue Γ to a meromorphic func-

tion defined on the entire plane. The key to doing this is the fact that Γ satisfies a
functional equation, as specified in the following theorem.
Theorem 9.1.2. The gamma function satisfies the functional equation
Γ(z + 1) = zΓ(z)
for all z in the right half-plane.

Proof. We have  ∞
Γ(z + 1) = e−t tz dt.
0
Integrating by parts with u = tz and dv = e−t dt yields
 ∞
∞
Γ(z + 1) = −tz e−t 0 + e−t ztz−1 dt = zΓ(z). 
0

Corollary 9.1.3. If n is a positive integer, then Γ(n) = (n − 1)!.

We leave the proof of this corollary as an exercise (Exercise 9.1.2).

Theorem 9.1.4. The gamma function has a meromorphic continuation to the
complex plane which has simple poles at the points {0, −1, −2, · · · }.

Proof. We prove by induction that Γ has a meromorphic continuation with the

indicated poles and satisfying the functional equation Γ(z + 1) = zΓ(z) on the set
{z : Re(z) > −n} for n = 0, 1, 2, · · · . This is trivially true for n = 0. If it is true
for n, then
Γ(z + 1)
Γ(z) =
z
defines Γ on {z : Re(z) > −n − 1} in a fashion which is consistent with its definition
on the smaller set {z : Re(z) > −n}, because of the functional equation. Clearly
the poles of this continuation are as required and the functional equation continues
to hold. 

Zeroes of Gamma. It turns out that Γ has no zeroes. To prove this requires de-
riving another functional equation. The derivation involves a pair of computational
lemmas.
We define Euler’s beta function, B(z, w), by
 1
B(z, w) = (1 − s)z−1 sw−1 ds,
0

for z and w with positive real part.

Lemma 9.1.5. If z and w have positive real parts, then
Γ(z)Γ(w)
= B(z, w).
Γ(z + w)
272 9. The Gamma and Zeta Functions

Proof. For z with Re(z) > 0, the substitution t = u2 leads to

 ∞  ∞
z−1 −t
u2z−1 e−u du.
2
Γ(z) = t e dt = 2
0 0
Then for two points z, w with positive real parts we have
 ∞ ∞
e−(u +v ) u2z−1 v 2w−1 du dv.
2 2
Γ(z)Γ(w) = 4
0 0
We pass to polar coordinates, setting u = r cos (θ) and v = r sin (θ). Then
 ∞  π/2
e−r r 2(z+w)−2 cos2z−1 (θ) sin2w−1 (θ) r dr dθ
2
Γ(z)Γ(w) = 4
0 0
 ∞  π/2
e−r r 2(z+w)−1 dr · 2
2
=2 cos2z−1 (θ) sin2w−1 (θ) dθ
0 0
 π/2
= Γ(z + w) · 2 cos2z−1 (θ) sin2w−1 (θ) dθ.
0
2
The substitution s = sin (θ) leads to
 π/2  1
2 cos2z−1 (θ) sin2w−1 (θ) dθ = (1 − s)z−1 sw−1 ds.
0 0
The latter expression is Euler’s beta function B(z, w). Thus,
Γ(z)Γ(w)
= B(z, w). 
Γ(z + w)
We will use the above identity in the case w = 1 − z to derive a functional
equation for Γ. To do this, we will need to evaluate B(z, 1 − z). This involves the
following integral evaluation.
 ∞ −x
t π
Lemma 9.1.6. If x ∈ (0, 1), then dt = .
0 1 + t sin πx
Proof. We essentially proved this back in Chapter 5 where we discussed the Mellin
1
transform. In fact, the integral in the theorem is just the Mellin transform of
1+t
evaluated at 1 − x. In Example 5.4.3, as a special case of Theorem 5.4.2, we proved
1
that the Mellin transform of is
1+x
 ∞ t−1
x π
dx = .
0 1+x sin πt
If the roles of t and x are reversed, this becomes
 ∞ x−1
t π
dt = .
0 1 + t sin πx
The identity of the theorem is then obtained by replacing x by 1 − x and using the
identity sin(π − πx) = sin πx. 
Theorem 9.1.7. The gamma function satisfies the functional equation
π
Γ(z)Γ(1 − z) = .
sin πz
9.1. Euler’s Gamma Function 273

Proof. If we set z = x and w = 1 − x for x ∈ (0, 1), then since Γ(1) = 1, Lemma
9.1.5 implies
 1
Γ(x)Γ(1 − x)
Γ(x)Γ(1 − x) = = B(x, 1 − x) = (1 − s)x−1 s−x ds.
Γ(1) 0
t
Then the substitution s = leads to
t+1
 ∞ x−1  ∞ −x
t t−x dt t
Γ(x)Γ(1 − x) = 1− −x 2
= dt.
0 t+1 (t + 1) (t + 1) 0 1+t
We use the previous lemma to evaluate the last integral and conclude that
π
Γ(x)Γ(1 − x) = .
sin πx
Since this identity holds for x ∈ (0, 1), the Identity Theorem implies that it contin-
ues to hold when x is replaced by any complex number z for which the functions
involved are defined. 

This theorem has the following corollary, the proof of which is left as an exercise
(Exercise 9.1.4).
Corollary 9.1.8. The gamma function has no zeroes.

Product Formula for Γ. The fact that e−t = limn→∞ (1−t/n)n can be exploited
to express Γ as an infinite product of the sort studied in the previous chapter. The
first step in deriving this formula is to show the following:
Theorem 9.1.9. The identity
 n
Γ(x) = lim (1 − t/n)n tx−1 dt
n→∞ 0
holds for all x > 0.

Proof. The function e−s −1 + s is 0 at s = 0 and has a positive derivative 1 − e−s

for s > 0. It is, therefore positive for s > 0. Thus, 1 − s ≤ e−s for s > 0. With
s = t/n this implies 1 − t/n < e−t/n for t > 0 and, on taking nth powers,
(1 − t/n)n ≤ e−t .
Furthermore, an elementary calculus argument (Exercise 9.1.8) shows that
1
(9.1.4) e−t −(1 − t/n)n ≤ ,
ne
for t ≥ 0.
If we fix a > 0, then
 n  a  ∞
Γ(x) − (1 − t/n)n tx−1 dt ≤ (e−t −(1 − t/n)n )tx−1 dt + e−t tx−1 dt
0 0 a

for n > a. The first term on the right converges to 0 as n → ∞ by (9.1.4) and the
second term can be made less than any given  by choosing a large enough, because
the improper integral defining Γ converges. 
274 9. The Gamma and Zeta Functions

Theorem 9.1.10. The entire function 1/Γ can be represented as the infinite prod-
uct
∞

1 1 + z/k
=z ,
Γ(z) (1 + 1/k)z
k=1
where this product converges uniformly on each disc of finite radius.

Proof. The integral in the previous theorem may be evaluated using a repeated
application of integration by parts (Exercise 9.1.9). The result is
nx n!
Γ(x) = lim .
n→∞ x(x + 1) · · · (x + n)

for x > 0.
If we invert this, divide both numerater and denomenator by n!, and note that
nx = n−1 x
k=1 (1 + 1/k) , we obtain
∞
 1 + x/k
1
=x
Γ(x) (1 + 1/k)x
k=1

for x > 0.
If we can show that this infinite product converges, not just for x > 0, but
uniformly on each disc of finite radius in the complex plane, then the result will be
an entire function which agrees with the entire function 1/Γ(z) on the positive real
axis. This implies the two entire functions agree on all of C. Thus, the proof will
be complete if we can show that
∞
 1 + z/k
(9.1.5) z
(1 + 1/k)z
k=1

converges uniformly on each compact disc. This product is very nearly a Weierstrass
product, as studied in the previous chapter. This fact can be used to prove the
uniform convergence on compact discs. The details are left to Exercise 9.1.10.


Exercise Set 9.1

1. Show that, for z real and positive, Γ(z) is the Mellin transform of a certain
function. What function? (see Section 5.4).
2. Prove that Γ(n) = (n − 1)! if n is a positive integer.
Γ(z + n + 1)
3. Prove that z(z + 1)(z + 2) · · · (z + n) = .
Γ(z)
4. Prove Corollary 9.1.8.
(−1)n
5. Prove that the residue of Γ(z) at −n is for n = 0, 1, 2, · · · .
 ∞ n!
−rt z−1
6. Prove that, for r > 0 and Re(z) > 0, 0 e t dt = r −z Γ(z).
−π
7. Prove that Γ(z)Γ(−z) = .
z sin πz
9.2. The Riemann Zeta Function 275

1
8. Prove that e−t −(1 − t/n)n ≤ for all t ∈ [0, n]. Hint: Show that the
ne
−t
maximum of the function h(t) = e −(1 − t/n)n on [0, n] occurs at a point t0
where h(t0 ) = e−t0 t0 /n. Then show that this number is less than or equal to
1
.
ne
9. Using integration by parts, prove that if x > 0, then
 n n
t nx n!
1− tx−1 dt = .
0 n x(x + 1) · · · (x + n)
10. Prove that the infiinite product
∞
 1 + z/k
1
=z
Γ(z) (1 + 1/k)z
k=1

converges uniformly on each compact disc. Hint: Show that

1 + z/k
= (1 + z/k) e−z/k eak z ,
(1 + 1/k)z
−z/k
where ak = 1/k − log(1 + 1/k). Then
 show that k (1 + z/k) e is a
convergent Weierstrass product and k |ak | converges.

9.2. The Riemann Zeta Function

The Riemann zeta function is defined on the set {z : Re(z) > 1} by the infinite
series
∞

(9.2.1) ζ(z) = n−z .
n=1
−z −x
If z = x + iy, then |n | = n , and so this series converges uniformly absolutely
on each set of the form {z : Re(z) ≥ r} for r > 1. It follows that ζ(z) is defined
and analytic on the set {z : Re(z) > 1}.

A Product Formula for the Zeta Function. Let {p1 , p2 , p3 , · · · } be the set of
prime numbers written in increasing order. Then we have
∞
Theorem 9.2.1. For Re(z) > 1, ζ(z) = n=1 (1 − p−z
n )
−1
.

Proof. The fact that the infinite product converges for Re(z) > 1 follows from
∞ Re(z)
Theorem 8.1.4 and the fact that n=1 pn converges if Re(z) > 1. Then
∞
 ∞
 
ζ(z)(1 − 2−z ) = n−z − (2n)−z = n−z ,
1 1 n∈S1

where S1 is the set of odd natural numbers. An induction argument using the same
technique then shows that for each natural number k

k 
ζ(z) (1 − p−z
n ) = n−z ,
n=1 n∈Sk
276 9. The Gamma and Zeta Functions

where Sk is the set of natural numbers not divisible by any of the first k primes.
Since the right side of this equation has limit 1 as k → ∞, the theorem follows. 

This theorem has the following two corollaries. We leave the proofs to the
exercises (Exercises 9.2.2 and 9.2.3).
Corollary 9.2.2. There are infinitely many primes.
Corollary 9.2.3. The zeta function has no zeroes in the region Re(z) > 1.

The Function ξ. Our next goal is to extend the zeta function to be a meromor-
phic function on the entire plane. We will do this by expressing the zeta function
in terms of the gamma function and a certain entire function ξ.
We begin the development of ξ by making the substitution t = n2 s2 π in the
formula (9.1.1) defining Γ. The result is
 ∞
ds
e−n s π s2z .
2 2
Γ(z) = 2n2z π z
0 s
If we divide by n2z π z and sum over n = 1, 2, 3, · · · , we obtain
∞  ∞
−z ds
e−n s π s2z
2 2
(9.2.2) ζ(2z)Γ(z)π = 2 if Re(z) > 1.
n=1 0
s
We will use the result of Exercise 9.2.7 to prove that it is legitimate to move the
summation inside the integral in the expression on the right. We estimate the size
of each integrand in this series as follows:
 
 −n2 s2 π 2z−1 
 ≤ e−ns s2 Re(z)−1 .
2
e s
The functions on the right are positive and their sum is
∞
s2 Re(z)−1
e−ns s2 Re(z)−1 =
2
(9.2.3) .
n=1
es2 −1
For each z, this series converges uniformly on each closed subinterval of (0, ∞).
2
Furthermore, since es −1 ≥ s2 , if Re(z) > 2, the function on the right in (9.2.3) is
less than or equal to s2 Re(z)−3 and, hence, has finite integral over [0, 1] if Re(z) > 2.
2 2
Since es −1 ≥ es /2 if s ≥ 1, the function on the right in (9.2.3) is less than or
equal to 2 e−s sRe(z)−1 on [1, ∞) and, hence, has finite integral on [1, ∞). It follows
2

that this function has finite integral on [0, ∞). Thus, by the result of Exercise 9.2.7,
the sum can be taken inside the integral in (9.2.2). Doing so yields
 ∞ ∞
−z ds
e−n s π s2z
2 2
(9.2.4) ζ(2z)Γ(z)π = 2 for Re(z) > 2.
0 n=1
s

If we replace z by z/2 and set

∞

e−n
2 2
s π
(9.2.5) H(s) = ,
n=1
then (9.2.4) may be rewritten as
 ∞
ds
(9.2.6) ζ(z)Γ(z/2)π −z/2 = 2 H(s)sz for Re(z) > 2.
0 s
9.2. The Riemann Zeta Function 277

The function ξ is obtained by multiplying this expression by z(z − 1)/2. Thus, for
Re(z) > 2,
 ∞
z(z − 1) −z/2 ds
(9.2.7) ξ(z) = ζ(z)Γ(z/2)π = z(z − 1) H(s)sz .
2 0 s

The Poisson Summation Formula. We pause to prove a technical result about

Fourier transforms (see Section 5.3). It will be used in the upcoming proof that ξ
extends to an entire function.
Theorem 9.2.4. If f is a continuous function on R with the property that the
series ∞ f (x + 2πn) converges absolutely and uniformly for x ∈ [−π, π] and
∞
n=−∞
the series n=−∞ fˆ(n) converges absolutely, then
∞
 ∞
1 
f (2πn) = √ fˆ(n),
n=1
2π n=1
where fˆ is the Fourier transform of f .

Proof. We set g(θ) = ∞ n=1 f (θ + 2πn) whenever θ ∈ [−π, π], and then integrate
this function against the Poisson kernel (see Section 6.5)
 ∞
1 − r2
Pr (θ) = = r |n| einθ
1 − 2r cos θ + r 2
−∞

over the interval [−π, π]. The result is

 π  π ∞

g(θ)Pr (θ) dθ = f (θ + 2πn)Pr (θ) dθ
−π −π −∞

 π
∞
= f (θ + 2πn)Pr (θ) dθ
−∞ −π
∞  (2n+1)π

(9.2.8) = f (θ)Pr (θ) dθ
−∞ (2n−1)π
 ∞
= f (θ)Pr (θ) dθ
−∞
∞

√
= 2π r |n| fˆ(n).
−∞

Note that the third step in this calculation uses the hypothesis that the series
defining g converges uniformly absolutely .
As r → 1 the integral on the left in (9.2.8) converges to
∞

2πg(0) = 2π f (2πn),
n=1
by Lemma 6.5.5, while the sum on the right converges to
∞
√ 
2π fˆ(n)
−∞
278 9. The Gamma and Zeta Functions

∞
since, by hypothesis, the series −∞ fˆ(n) converges absolutely. We conclude that
∞
 ∞
1
f (2πn) = √ fˆ(n),
n=−∞
2π n=−∞
as required. 

The Symmetry of ξ. We can now prove that ξ extends to be an entire function

and that it satisfies the symmetry relation ξ(z) = ξ(1 − z). We first derive a
symmetry relation for H.
Lemma 9.2.5. The function H satisfies the relation
(9.2.9) H(s−1 ) = sH(s) + (s − 1)/2.

Proof. Note that H(s) = (G(s) − 1)/2, where

∞
 ∞

e−n e−n
2 2 2 2
s π s π
G(s) = =1+2 .
−∞ n=1

1
If g(x) = √ e−x /2 is the normal distribution function from Example 5.3.6, then
2

2π
∞

√ √
G(s) = 2π g(ns 2π).
−∞

Note that the function g is its own Fourier transform,

√ by Example 5.3.6. This
implies that the function f , defined by f (x) = g(xs/ 2π), satisfies the hypotheses
of the previous theorem. If we apply that theorem to f , we conclude that
∞
 ∞
1 
G(s) = f (2πn) = √ fˆ(n).
−∞
2π −∞
A change of√variables in
√ the integral defining the Fourier transform shows that
fˆ(n) = s−1 2πgˆ(ns−1 2π). Thus,
∞
 √ √
G(s) = s−1 2πg(ns−1 2π) = s−1 G(s−1 ).
−∞

The identity (9.2.9) follows from this. 

The consequence for the function ξ is the following:

Theorem 9.2.6. The function ξ extends to an entire function which is symmetric
about the point z = 1/2; that is, ξ(1 − z) = ξ(z).

Proof. If we break the integral on the right side of (9.2.6) into an integral over
[1, ∞) and an integral over [0, 1] and make the substitution s → s−1 in the latter
integral, the result is
 ∞  ∞
ds ds
H(s)sz + H(s−1 )s−z .
1 s 1 s
9.2. The Riemann Zeta Function 279

Using (9.2.9), this becomes

 ∞  ∞
ds ds
H(s)sz + (sH(s) + (s − 1)/2)s−z
s s
1
 ∞ 1
 ∞ 
ds ds 1 ∞ (1−z) ds
= H(s)sz + H(s)s(1−z) + (s − s−z ) .
1 s 1 s 2 1 s
Since  ∞
1 ds 1/2
(s(1−z) − s−z ) = if Re(z) > 1,
2 1 s z(z − 1)
we conclude that
 ∞
ds
(9.2.10) ξ(z) = 1/2 − z(1 − z) H(s)(sz + s1−z ) for Re(z) > 2.
1 s
However, the right side of this equation is defined and analytic on the entire plane,
since H(s) times any power of s is absolutely integrable on [1, ∞). It is also obvi-
ously symmetric about z = 1/2. Thus, ξ has an extension to the whole plane with
the required properties. 

Meromorphic Extension of ζ. The next theorem is an immediate consequence

of the preceding theorem and (9.2.7).
Theorem 9.2.7. The function ζ has a meromorphic extension to the plane given
by the formula
2π z/2 ξ(z)
(9.2.11) ζ(z) = .
z(z − 1)Γ(z/2)

It is useful to note that the above formula can be put in a slightly different
form by using Theorem 9.1.2. This theorem, with z replaced by z/2, implies that
(z/2)Γ(z/2) = Γ(z/2 + 1).
Then (9.2.11) becomes
π z/2 ξ(z)
(9.2.12) ζ(z) = .
(z − 1)Γ(z/2 + 1)

Exercise Set 9.2

1. Show that limx→∞ ζ(x + iy) = 1 and the convergence is uniform in y.

2. Use Theorem 9.2.1 to prove Corollary 9.2.2.
3. Use Theorem 9.2.1 to prove Corollary 9.2.3.
4. If z = s + it with s > 1, prove that
 
 1 
 
 ζ(z)  < ζ(s).

Hint: Use Theorem 9.2.1.

280 9. The Gamma and Zeta Functions

5. Use the result of the previous exercise to prove that

 
 1 
 
 ζ(z)  < ζ(2)

if Re(z) ≥ 2.
∞
6. Let u(t) = n=1 un (t) be the sum of a series of positive continuous func-
tions on (0, ∞) and suppose this series converges uniformly on closed bounded
intervals of (0, ∞). Prove that
 ∞  ∞  ∞
un (t) dt = u(t) dt.
n=1 0 0

Hint: Either
both sides are infinite or one of them is finite.
∞
7. Let h(t) = n=1 hn (t) be the sum of an infinite series of continuous functions
on (0, ∞) and suppose
|hn (t)| ≤ un (t) for all n, t,
∞
where n=1 un is a positive termed series which satisfies the conditions of the
previous exercise. Prove that the improper integral of h on R converges and
 ∞ ∞  ∞

h(t) dt = hn (t) dt.
0 n=1 0

8. Show that the Poisson Summation Formula can, under appropriate hypotheses
on f and fˆ, be reformulated as
∞
 ∞

√
f (n) = 2π fˆ(2πn).
n=−∞ n=−∞

9. Use the form of the Poisson Summation Formula derived in the previous ex-
ercise to show that
∞
1 e2π +1
= π .
−∞
1 + n2 e2π −1
1
Hint: The Fourier transform of is calculated in Example 5.3.3.
1 + x2
The next five exercises outline an alternative to the approach used in Theorem 9.2.7
to prove that ζ extends to be meromorphic in the plane.
10. Formula (9.2.4) was developed using the substitution t = n2 sπ in the integral
defining Γ. Use the substitution t = ns in a similar way to derive the formula
 ∞
1 z−1
ζ(z)Γ(z) = s ds for Re(z) > 1.
0 es −1
11. For complex numbers w and z, define (−w)z−1 to be e(z−1) log(−w) , where log
is the principal branch of the log function. Show that this function is analytic
except for a cut on the positive real line, that its limit as w approaches the
positive real number s from above is e(z−1)(log s−πi) , and that its limit as w
appproaches s from below is e(z−1)(log s+πi) .
9.3. Properties of ζ 281

γr

Figure 9.2.1. Contour for Exercise 9.2.12.

12. Using this definition for (−w)z−1 , consider the contour integral

1
(9.2.13) η(z) = w −1
(−w)z−1 dw,
γr e
where γr is the contour indicated in Figure 9.2.1, r < 2π is the radius of
the indicated circle, and the two horizontal lines are a distance  ≤ r above
and below the positive real axis. Prove that this integral exists for all z, is
independent of r and , and defines an entire function η(z).
13. By passing to the limit as  → 0 and r → 0 in the integral 9.2.13, prove that
η(z) = −2πi sin(πz)Γ(z)ζ(z) if Re(z) > 1.
14. Use the previous exercise and an identify involving Γ to prove that
1
ζ(z) = − Γ(1 − z)η(z) for Re(z) > 1.
2πi
Conclude that ζ has a meromorphic extension to the plane with a single simple
pole at z = 1.

9.3. Properties of ζ
The expressions (9.2.11) and (9.2.12) for ζ and the properties of Γ and ξ lead to a
wealth of information about ζ. Ultimately, this information will lead to a proof of
the Prime Number Theorem in the last section of this chapter.

Zeroes and Poles.

Theorem 9.3.1. The zeta function has a simple pole at z = 1, with residue 1, and
no other poles.

Proof. The function Γ(z/2 + 1) has no zeroes, and the function ξ is entire. Thus,
(9.2.12) implies that the only pole of ζ is at z = 1 and it is a simple pole. The √fact
that the residue is 1 follows from the fact, proved in the exercises, that Γ(1/2) = π,
and from (9.2.10), which implies ξ(1) = 1/2. 
282 9. The Gamma and Zeta Functions

Theorem 9.3.2. The zeta function has a zero of order 1 at each negative even
integer.

This is Exercise 9.3.1.

Corollary 9.3.3. The zeta function has no zeroes outside the strip
0 ≤ Re(z) ≤ 1
except the ones that occur at negative even integers.

Proof. Except for the zeroes of ζ that occur at negative even integers (due to the
poles of Γ), the functions ζ and ξ have the same zeroes. Since ζ has no zeroes in
the region Re(z) > 1 by Corollary 9.2.3, and since ξ is symmetric about 1/2, it
follows that ζ has no zeroes outside the strip 0 ≤ Re(z) ≤ 1 except the negative
even integers. 

This result can be strengthened to exclude the existence of zeroes of ζ on the

lines Re(z) = 0, 1. The proof uses the following lemma:
Lemma 9.3.4. For Re(z) > 1 there is an analytic logarithm for ζ(z), defined by
 ∞
p−mz
(9.3.1) log ζ(z) = .
p m=1
m

The derivative of this function is

 ∞
ζ  (z)
(9.3.2) =− p−mz log p.
ζ(z) p m=1

Here, in both equations, the summation on the left is over all primes p.

Proof. If Re(z) > 1, then |p−z | < 1/2 for all primes p and so log(1 − p−z ) is
defined and analytic if log is the principal branch of the log function. Furthermore,
by Theorem 9.2.1 we have for Re(z) > 1
 
 
−z
(9.3.3) exp − log(1 − p ) = (1 − p−z )−1 = ζ(z).
p
 −z
Hence, − p log(1 − p ) is an analytic logarithm for ζ(z) on Re(z) > 1. If we
expand this function in a power series in p−z , the result is (9.3.1). On differentiating
(9.3.1), we obtain (9.3.2). 

Theorem 9.3.5. The zeta function has no zeroes outside the strip
0 < Re(z) < 1
except those which occur at negative even integers.

Proof. We first note that the inequality

(9.3.4) 0 ≤ 3 + 4 cos θ + cos 2θ
9.3. Properties of ζ 283

holds for all real numbers θ due to the identity

3 + 4 cos θ + cos 2θ = 2(1 + cos θ)2 ,
which follows from cos 2θ = 2 cos2 θ − 1.
If z = x + iy, then Re(p−mz ) = p−mx cos(−my log p). It follows from (9.3.4)
that, for x > 1,
0 ≤ 3p−mx + 4 Re(p−mx−myi ) + Re(p−mx−2myi ),
which, when combined with (9.3.1), implies that, for x > 1,
3 Re(log ζ(x)) + 4 Re(log ζ(x + iy)) + Re(log ζ(x + i2y)) ≥ 0,
or, on exponentiating,
(9.3.5) |ζ(x)|3 |ζ(x + iy)|4 |ζ(x + i2y)| ≥ 1.
We divide both sides of this inequality by x − 1 and write the result in the form
 4
3  ζ(x + iy)  1
(9.3.6) |(x − 1)ζ(x)|  |ζ(x + i2y)| ≥ .
x−1  x−1
Since ζ(z) has a simple pole at z = 1, (1 − z)ζ(z) has a removable singularity at
z = 1. This implies that the first factor on the left in (9.3.6) is bounded as x → 1.
Since there are no other poles of ζ, the factor on the right is also bounded as x → 1,
provided y = 0. If ζ has a zero at 1 + iy, then the middle factor is also bounded as
x → 1. Since the right side of (9.3.6) is not bounded as x → 1, we conclude that
there can be no zero of ζ at z = 1 + iy.
Now that we know that there are no zeroes of ζ on the line Re(z) = 1, we use
the fact that the set of zeroes of ζ in the strip 0 ≤ Re(z) ≤ 1 is symmetric about
the line Re(z) = 1/2 (since these are also the zeroes of ξ) to conclude that there
are no zeroes on the line Re(z) = 0. In view of Corollary 9.3.3, this completes the
proof. 

Estimate on the growth of ξ. Integration by parts in the integral appearing

in (9.2.10) leads to
 ∞
ξ(z) = 1/2 + H(1) + H  (s)((1 − z)sz + zs1−z ) ds
 ∞
1
ds
= 1/2 + H(1) + s2 H  (s)((1 − z)s(z−1) + zs−z ) .
1 s
Another integration by parts leads to
 ∞

ξ(z) = 1/2 + H(1) + 2H (1) + (s2 H  (s)) (sz−1 + s−z ) ds.
1
If we differentiate (9.2.9) and set s = 1, the result is
1
+ H(1) + 2H  (1) = 0.
2
Thus,
 ∞
(9.3.7) ξ(z) = (s2 H  (s)) (sz−1 + s−z ) ds.
1
284 9. The Gamma and Zeta Functions

Since
(sz−1 + s−z ) = s−1/2 (sz−1/2 + (s−z+1/2 )
= 2s−1/2 cosh ((z − 1/2) log(s)),
(9.3.7) can be rewrittten as
 ∞
(9.3.8) ξ(z) = (s2 H  (s)) s−1/2 cosh ((z − 1/2) log(s)) ds.
1

Theorem 9.3.6. If ξ is expanded as a power series in z − 1/2, the coefficients of

the power series are all real and non-negative.

Proof. A direct calculation using (9.2.5) shows that

∞

(s2 H  (s)) s−1/2 = (4n4 π 2 s4 − 6n2 πs2 )s−1/2 e−n
2 2
s π
.
n=1

The terms of this series are clearly positive for s ≥ 1 and so the function itself is
positive. Also, the power series coefficients in the expansion of cosh w about 0 are
real and non-negative. Since log(s) ≥ 0 for s ≥ 1, the theorem follows (see Exercise
9.37). 

Theorem 9.3.7. There is a constant R such that |ξ(1/2 + z)| ≤ r r for all z ∈ C
with |z| = r > R.

Proof. Since ξ(z + 1/2) has a power series expansion in z with real non-negative
coefficients, its maximum absolute value on any disc Dr (0) is achieved at z = r.
However, if n is an integer such that 1/2 + r ≤ 2n ≤ 5/2 + r, then by (9.2.7) and
the fact that ξ is increasing on the positive real line (Exercise 9.3.2),
ξ(1/2 + r) ≤ ξ(2n) = n(2n − 1)ζ(2n)Γ(n)π −n.
Now ζ is decreasing on (1, ∞). Thus,
ζ(2n) ≤ ζ(2) if n ≥ 1,
and
Γ(n) = (n − 1)!
if n is a positive integer. Thus,
ξ(1/2 + r) ≤ 2n n! ζ(2) ≤ 2ζ(2) nn+1 ≤ r r
if r is sufficiently large (since n ≤ 5/4 + r/2) (Exercise 9.3.9). 

The following corollary is a direct consequence of the above theorem and the
results of sections 8.3 and 8.4. We leave the details to the exercises.
Corollary 9.3.8. The function ξ is an entire function of finite order at most
1. Consequently, the zeroes of ξ (and, hence, the zeroes of ζ that lie in the strip
0 < Im(z) < 1) form a sequence with exponent of convergence at most 1.
9.3. Properties of ζ 285

A Product Expansion for ξ. In what follows, it will be convenient to make the

change of variables w = z − 1/2. Then

ξ(z) = ξ(1/2 + w).

Since ξ is symmetric about 1/2, the function ξ(1/2 + w) is symmetric about 0.

By Corollary 9.3.8, ξ(1/2 + w) is an entire function of finite order at most 1.
Hence, the Hadamard Factorization Theorem applies with λ = 1. It tells us that ξ
has a factorization of the form

ξ(1/2 + w) = eq(w) (1 − w/σ) ew/σ ,
σ

where q is a polynomial of degree at most 1, and the product is over all zeroes σ of
ξ(1/2 + w). However, the zeroes of this function are symmetric about 0 and so, if a
zero σ appears in this product, then so does its negative. The exponential factors
ew/σ and e−w/σ cancel and so

ξ(1/2 + w) = eq(w) (1 − w/σ)
σ

as long as it is understood that the factors involving a given σ and its negative −σ
are to be grouped together. If this is done, then the product expansion becomes

ξ(1/2 + w) = eq(w) (1 − w2 /σ 2 ).
Im(σ)>0

It is this product that actually converges (see the discussion of the product expan-
sion of sin(πz) in Example 8.2.6).
Now ξ(1/2 + w) is an even function of w (symmetric about 0) and so is the
above product. It follows that the polynomial q must also be an even function.
Since the only even polynomials of degree at most 1 are constants, we conclude
that

(9.3.9) ξ(1/2 + w) = c (1 − w/σ)
σ

for some constant c.

If we recall that w = z − 1/2 and σ = ρ − 1/2 and we use the identity
  
(z − 1/2) z 1/2
(9.3.10) 1− = 1− 1+
(ρ − 1/2) ρ ρ − 1/2
(Exercise 9.3.10), then (9.3.9) becomes
 z

1/2

ξ(z) = 1− 1+ .
ρ
ρ ρ − 1/2

Since the product

 1/2

(9.3.11) c1 = c 1+
ρ
ρ − 1/2
286 9. The Gamma and Zeta Functions

converges as long as the factors involving ρ − 1/2 and its negative are grouped
together (Exercise 9.3.11), we obtain an infinite product expansion
 z

ξ(z) = c1 1− .
ρ
ρ

By evaluating at z = 0, we see that c1 = ξ(0) = 1/2. This proves the following

theorem.
Theorem 9.3.9. The function ξ has the following infinite product expansion which
converges uniformly on each disc of finite radius:
 
1 z
ξ(z) = 1− ,
2 ρ ρ

where the numbers ρ are the zeroes of ξ(z) and the factors are arranged so that ρ
and 1 − ρ are grouped together. The product converges uniformly on each disc of
finite radius.

This, in turn, implies that the zeta function has the following infinite product
expansion:
Theorem 9.3.10. The zeta function satisfies
 
1 π z/2  z
ζ(z) = 1− ,
z(z − 1) Γ(z/2) ρ ρ

where the product is over the zeroes ρ of ζ in the strip 0 < Re(z) < 1, and the
factors involving ρ and 1 − ρ are grouped together in the infinite product.

Exercise Set 9.3

1. Prove that ζ has a zero of order 1 at each negative even integer.

2. Prove that ξ is increasing on the positive real line.
3. Show that ξ(0) = 1/2 even though (9.2.7) suggests it should be 0. Why is
there no contradiction here?
4. Calculate ζ(0) using (9.2.12).
5. Calculate ζ(2) directly from the definition of ζ, using results from Section 5.5.
6. Prove Corollary 9.3.8 (see Section 8.3 and Exercise 8.3.8).
7. Suppose f is an entire function and the coefficients of the power series expan-
sion of f about 0 are all non-negative. Show that if g(t) and h(t) are positive
continuous functions on [1, ∞), then
 ∞
g(t)f (zh(t)) dt
1

is also an entire function of z with non-negative coefficients in its power series

expansion about 0, provided this integral exists for all positive real values of
z.
9.4. The Riemann Hypothesis and Prime Numbers 287

8. Show that  ∞
dt
e−t tz
1 t
is an entire function of z with all non-negative coefficients in its power series
expansion about 0.
9. Verify the claim made in the last sentence of the proof of Theorem 9.3.7 – that
is, show that 2ζ(2)nn+1 ≤ r r if r is sufficiently large and n ≤ 5/4 + r/2.
10. Prove the identity (9.3.10).
11. Show why the product in (9.3.11) converges if the terms are grouped as indi-
cated.

9.4. The Riemann Hypothesis and Prime Numbers

The Riemann Hypothesis is the conjecture that all zeroes of the zeta function in
the strip 0 < Re(z) < 1 lie on the line Re(z) = 1/2. The significance of this
conjecture lies in its connection with the problem of estimating the density of the
prime numbers in the set of natural numbers. In this section we will discuss some
of the history of the two problems and attempt to illustrate the connection between
them without going into too much computational detail. For a more comprehensive
and detailed account of the subject see the book by H. M. Edwards [4].
We let π(x) denote the number of primes less than or equal to the positive real
number x. It was recognized by Riemann that there is a connection between the
rate of growth of π(x) as x increases and the zeroes of the zeta function. That there
is some connection between the growth of π(x) and the zeta function is seen in the
proof that there are infinitely many primes (Exercise 9.2.2).
Based on experimental evidence, Gauss and Legendre conjectured that
π(x)
(9.4.1) lim log x = 1.
x→∞ x
π(x)
This means that the fraction of the natural numbers up to x that are prime, ,
x
1
is asymptotic to in the sense that their ratio has limit 1.
log x
In his famous 1859 paper Riemann introduced a formula for π(x). For a certain
constant c, the function Li is defined to be
⎧
⎨0, x ≤ 2;
 x
Li(x) = dt
⎩ + c, x > 2.
2 log t
The formula of Riemann for π(x) is then
∞
 ∞

(9.4.2) π(x) = Li(x1/n ) + Li(xρ/n ) + other,
n=1 ρ n=1

where the “other” terms are of lower order in x, and ρ ranges over all zeroes of ζ
in the strip 0 < Re(z) < 1. Note that the sums over n are actually finite sums for
each fixed x and ρ, since x1/n and xρ/n are less than 2 if n is large enough.
288 9. The Gamma and Zeta Functions

An integration by parts argument shows that the first (and presumably domi-
nant) term in the expansion (9.4.2) may be rewritten as
 x
x dt
Li(x) = − 2
+ c.
log x 2 (log t)
x
The second term in this expression, when divided by , has limit 0 at infinity
log x
(Exercise 9.4.1). Thus, if the remaining terms in Riemann’s Formula for π(x), when
x
divided by , also have limit 0 at infinity, and if it is legitimate to take the limit
log x
inside the sum in the first term, then (9.4.1) follows.
In dealing with the terms involving the zeroes ρ of ζ in Riemann’s Formula, it
would be useful if the zeroes of ζ in the strip 0 < Re(z) < 1 all satisfied Re(z) < r
for some r < 1. Riemann suspected that this was true and, in fact, he conjectured
that all such zeroes actually lie on the line Re(z) = 1/2. This is the Riemann
Hypothesis.
Actually, Riemann did not give a complete proof of his formula (9.4.2) for
π(x) and, in fact, he did not even prove that the infinite series in this formula
converges. Both facts were eventually proved, but the difficulties involved in these
proofs and in determining the contribution of the terms involving the zeroes of ζ to
the asymptotic behavior of π(x) led to the introduction of another function ψ(x)
which also measures the density of primes and which satisfies a simpler and more
natural formula analogous to (9.4.2).
Eventually, Hadamard and de la Vallée-Poussin in 1896 proved (9.4.1). It is
now known as the Prime Number Theorem. The proofs of Hadamard and de la
Vallée-Poussin as well as other classical proofs of this result are based on the fact
that there are no zeroes of the zeta function on the line Re(z) = 1. We will present
one such proof in the next section.

The Function ψ. As mentioned above, the function π(x) is closely related to

another function ψ(x) which also measures the density of primes and which has a
more straightforward connection to the zeta function. The following equation gives
two ways of expressing this function:

 
(9.4.3) ψ(x) = log p = mp log p,
pm ≤x p≤x

where the first sum is over all prime powers pm ≤ x. For a given prime p, the term
log p appears in this sum as many times as there are positive powers of p less than
or equal to x. This number is the mp which appears in the second sum. It can also
be described as the largest number m such that pm ≤ x.
We will show that if the function ψ satisfies limx→∞ ψ(x)/x = 1, then the
Prime Number Theorem follows.
Lemma 9.4.1. Let x and y be real numbers greater than 1. Then
ψ(x) ψ(x)
≤ π(x) ≤ y + .
log x log y
9.4. The Riemann Hypothesis and Prime Numbers 289

Proof. We have
  log p
π(x) = π(y) + 1≤y+ if y < x,
(9.4.4) log y
y<p≤x y<p≤x
π(x) ≤ y if y ≥ x,
where the sums are over primes p in the indicated range.
By definition,
 
(9.4.5) ψ(x) = mp log p = log pmp .
p≤x p≤x

The sum on the right satisfies the inequalities

 
log p ≤ log pmp ≤ π(x) log x,
y<p≤x p≤x

and so, by (9.4.5), 

log p ≤ ψ(x) ≤ π(x) log x.
y<p≤x
The left side of this inequality, combined with (9.4.4), yields
ψ(x)
π(x) ≤ y + .
log y
The right side, when divided by log x, yields
ψ(x)
≤ π(x).
log x
This completes the proof of the lemma. 
ψ(x) π(x)
Theorem 9.4.2. If lim = 1, then lim log x = 1.
x→∞ x x→∞ x
Proof. Given x > 1, we use the previous lemma with
x
y= .
(log x)2
According to that lemma,
ψ(x) x ψ(x)
≤ π(x) ≤ + ,
log x (log x) 2 log x − 2 log log x
and so
log x 1 x log x
1 ≤ π(x) ≤ + .
ψ(x) log x ψ(x) log x − 2 log log x
The first term on the right has limit 0 as x → ∞, while the second term has limit
1. This is because
ψ(x)
lim = 1,
x→∞ x
by hypothesis, while
log x
lim = 1.
x→∞ log x − 2 log log x
The latter statement is left as an exercise (Exercise 9.4.4). It follows that
log x
lim π(x) = 1,
x→∞ ψ(x)
290 9. The Gamma and Zeta Functions

and, from this, that

π(x)
lim log x = 1.
x→∞ x
This completes the proof. 

The above result says that the Prime Number Theorem will follow if we can
prove that limx→∞ ψ(x)/x = 1. There are direct proofs of this; however, they
involve serious difficulties with improper integrals and conditionally convergent se-
ries. It turns out that these difficulties can be made to disappear if we follow a
similar approach, but use the integral of ψ rather than ψ itself as the main focus
of attention. Thus, we set
 x
(9.4.6) φ(x) = ψ(u) du.
1
This is another function which measures the density of primes. Furthermore, the
Prime Number Theorem follows from an appropriate estimate on its asymptotic
behavior.

ψ(x)
Properties of φ. It turns out that, lim = 1, from which the Prime
x→∞ x
Number Theorem follows, provided
φ(x) 1
(9.4.7) lim 2
= .
x→∞ x 2
This is proved using a kind of reverse L’Hôpital’s Rule. Specifically:
Lemma 9.4.3. Let f be a positive, increasing function on [1, ∞) and suppose r > 0.
Then 
f (x) r+1 x
lim = lim f (u) du,
x→∞ xr x→∞ xr+1 1

provided the limit on the right exists and is finite.

Proof. If we were to assume that the limit on the left exists, then the equality
would follow from applying L’Hôpital’s rule to the limit on the right. However,
we are assuming only that the limit on the right exists, and so we must proceed
differently (see Exercise 9.4.6).
Using the fact that f is increasing and positive, we conclude that, for α < 1
and β > 1,
 x  βx
1 1
f (u) du ≤ f (x) ≤ f (u) du.
(1 − α)x αx (β − 1)x x
On dividing by xr , this becomes
 x  βx
1 f (x) 1
f (u) du ≤ ≤ f (u) du.
(1 − α)xr+1 αx xr (β − 1)xr+1 x
x
If we set F (x) = 1 f (u) du, then this can be rewritten as
F (x) − F (αx) f (x) F (βx) − F (x)
≤ r ≤ ,
(1 − α)xr+1 x (β − 1)xr+1
9.5. A Proof of the Prime Number Theorem 291

or as
   
1 F (x) F (αx) f (x) 1 F (βx) F (x)
− αr+1 ≤ ≤ β r+1 − .
1−α x r+1 (αx)r+1 x r β−1 (βx)r+1 xr+1
If we set
F (x) F (αx) F (βx)
L = lim = lim = lim ,
xr+1
x→∞ x→∞ (αx)r+1 x→∞ (βx)r+1
then the above inequality implies that
1 − αr+1 f (x) f (x) β r+1 − 1
L ≤ lim inf r ≤ lim sup r ≤ L.
1−α x→∞ x x→∞ x β−1
The lemma then follows from this on taking the limit as α and β approach 1, since
1 − αr+1 β r+1 − 1
both and have limit r + 1. 
1−α β−1
This leads directly to the following theorem. The details are left to the exercises.
φ(x) π(x)
Theorem 9.4.4. If lim = 1/2, then lim log x = 1.
x→∞ x2 x→∞ x

Exercise Set 9.4


log x x dt
1. Prove that lim 2
= 0.
x→∞ x 2 (log t)
2. Prove Part (a) of Lemma 9.5.1.
3. Prove Part (b) of Lemma 9.5.1.
log x
4. Prove that lim = 1.
x→∞ log x − 2 log log x
5. Use Lemma 9.4.3 to prove Theorem 9.4.4.
6. Find functions f and g on R such that
f (x) f  (x)
lim exists, but lim  does not.
x→∞ g(x) x→∞ g (x)

9.5. A Proof of the Prime Number Theorem

In view of Theorems 9.4.2 and 9.4.4, to prove the Prime Number Theorem, it
suffices to show that limx→∞ φ(x)/x2 = 1/2. The strategy for doing this involves
expressing φ(x) as an integral involving ζ  /ζ. This is where the zeroes of the zeta
function come in.
The integral formula relating ζ  /ζ and φ is derived from the series expansion
(9.3.2) and the following integral formula.
Lemma 9.5.1. Suppose p(z) is a non-constant polynomial and b a real number
such that no zero of p lies on the line Re(z) = b. If y > 1, let A = {z1 , z2 , · · · , zn }
be the set of zeroes of p(z) that lie to the left of the line Re(z) = b. Then
 b+i∞ z n
1 y
(9.5.1) dz = Res(y z /p(z), zk ).
2πi b−i∞ p(z)
k=1
292 9. The Gamma and Zeta Functions

If the set A is empty, then the integral is zero.

If y < 1, a similar formula holds, the only differences being: A is replaced by the
set of zeroes to the right of Re(z) = b and the expression on the right is multiplied
by −1.

Proof. In Chapter 5 we showed how to use residue theory to calculate the Fourier
transforms of certain functions (Theorem 5.3.2). The integral that appears in (9.5.1)
is actually the Fourier transform of a function to which Theorem 5.3.2 applies. To
see this, we write
yz ez log y yb
= = eit log y
p(z) p(z) p(b + it)
for z = b + it. Then
 b+i∞ z  ∞
1 y 1 1
dz = f (t) eit log y dt = √ fˆ(− log y),
2πi b−i∞ p(z) 2π −∞ 2π
where f is the restriction to the real line of the meromorphic function
yb
f (z) = .
p(b + iz)
The function f has limit 0 at infinity since p is a non-constant polynomial. Thus,
by Theorem 5.3.2, if y > 1, then
 b+i∞ z 
1 y
dz = i Res(f (z) eiz log y , w),
2πi b−i∞ p(z)
w∈B

where B is the set of poles of f in the upper half-plane. Since

y b+iz
f (z) ez log y = ,
p(b + iz)
a calculation of the effect on a residue of the change of variables z → b+iz (Exercise
9.5.8) shows that
 b+i∞ z 
1 y
dz = Res(f (z) eiz log y , λ),
2πi b−i∞ p(z)
λ∈A

where A = {λ = b + iw : w ∈ B} – that is, A is the set of zeroes of p that lie to the

left of the line Re(z) = b. This completes the proof in the case y > 1. The proof in
the case y < 1 proceeds in the same way. 

Example 9.5.2. Prove that if b > 0 and y > 0, then

 b+i∞ 
1 yz 1 − 1/y, if y > 1;
dz =
2πi b−i∞ z(z + 1) 0, if y < 1.

Solution: Since b > 0, by the previous lemma, if y > 1, the integral is the
yz
sum of the residues of at 0 and −1, which is 1 − 1/y. If y < 1, the lemma
z(z + 1)
implies that the integral is 0, since z(z + 1) has no zeroes to the right of Re(z) = 0.
9.5. A Proof of the Prime Number Theorem 293

Theorem 9.5.3. If x > 0 is not a power of a prime and b > 1, then

 b+i∞
1 xz+1 ζ  (z)
φ(x) = − dz.
2πi b−i∞ z(z + 1) ζ(z)

xz+1
Proof. We multiply equation (9.3.2) by and integrate. The result is
z(z + 1)

1 b+i∞
xz+1 ζ  (z)
− dz
2πi b−i∞ z(z + 1) ζ(z)
(9.5.2)   b+i∞  z 
 1 x x log p
= dz ,
pm
2πi b−i∞ pm z(z + 1)

provided the integrals exist and the integral can be moved inside the summation
on the right. Assuming these things for the moment, we have, by the previous
example,
 b+i∞ 
1 xz+1 ζ  (z)
− dz = (x − pm ) log p.
2πi b−i∞ z(z + 1) ζ(z)
pm ≤x
x
The expression on the right is φ(x) = 1 ψ(u) du (Exercise 9.5.1), and so the proof
will be complete if we can verify that the integrals in (9.5.2) exist and the integral
can be brought inside the summation on the right.
The integrand corresponding to n = pm on the right in (9.5.2) is less than or
equal in modulus to
log n xb+1
nb b2 + t2
on the vertical line z = b + it. This has  the form cn f (t), where f is a positive
n
integrable function of t on (−∞, ∞) and 1 cn is a convergent series of positive
numbers (see Exercise 9.5.2). By Exercise 9.5.4 this implies that the series of
integrals on the right in (9.5.2) converges and it converges to the integral on the
left. 

A Series Expansion of φ. The Prime Number Theorem will follow directly

from the following infinite series expansion of φ.
Theorem 9.5.4. There are constants A and B such that
∞
x2  x1−2k  xρ+1
φ(x) = − − − Ax + B,
2 2k(2k − 1) ρ
ρ(ρ + 1)
k=1

where ρ ranges over the zeroes of ζ in the strip 0 < Re(z) < 1.

Proof. The integral that appears in Theorem 9.5.3 can also be evaluated by using
the infinite product expansion of Theorem 9.3.10. This theorem implies that the
logarithmic derivative of ζ can be written as
ζ  (z) 1 1 log π Γ (z/2)  1
= − + − + ,
ζ(z) 1−z z 2 Γ(z/2) ρ
z−ρ
294 9. The Gamma and Zeta Functions

where, in the last sum, terms involving ρ and 1 − ρ must be grouped together for
the series to converge. The product formula for 1/Γ given in Theorem 9.1.10 leads
to
∞  
Γ (z/2) 1  1 1
− = + − log(1 + 1/k) .
Γ(z/2) z 2k + z 2
k=1
Thus,
∞   
ζ  (z) 1 log π  1 1 1
=− + + + log(1 + 1/k) + .
ζ(z) z−1 2 z + 2k 2 ρ
z − ρ
k=1

This simplifies significantly if we subtract ζ  (0)/ζ(0):

∞    
ζ  (z) ζ  (0) 1 1 1 1 1
− = −1 − + − + + ,
ζ(z) ζ(0) z−1 z + 2k 2k ρ
z−ρ ρ
k=1

or
 ∞ 
ζ  (z) z z z ζ  (0)
(9.5.3) =− − + + .
ζ(z) z−1 2k(z + 2k) ρ
ρ(z − ρ) ζ(0)
k=1

1 xz+1
We next multiply equation (9.5.3) by and integrate along the line
2πi z(z + 1)
Re(z) = b > 1, obtaining
(9.5.4)
 b+i∞
1 xz+1 ζ  (z)
φ(x) = − dz
2πi b−i∞ z(z + 1) ζ(z)
 b+i∞  b+i∞ ∞
1 xz+1 1 xz+1
= dz + dz
2πi b−i∞ (z − 1)(z + 1) 2πi b−i∞ 2k(z + 2k)(z + 1)
k=1
 b+i∞   b+i∞ 
1 xz+1 1 ζ (0) xz+1
− dz − dz.
2πi b−i∞ ρ ρ(z − ρ)(z + 1) 2πi b−i∞ ζ(0) z(z + 1)

Assuming for the moment that the integral can be taken inside each of the
infinite sums, the result is
 b+i∞
1 xz+1 ζ  (z)
φ(x) = − dz
2πi b−i∞ z(z + 1) ζ(z)
 b+i∞ ∞  b+i∞
1 xz+1 1 xz+1
= dz + dz
2πi b−i∞ (z − 1)(z + 1) 2πi b−i∞ 2k(z + 2k)(z + 1)
k=1
 1  b+i∞ xz+1 1
 b+i∞ 
ζ (0) xz+1
− dz − dz.
ρ
2πi b−i∞ ρ(z − ρ)(z + 1) 2πi b−i∞ ζ(0) z(z + 1)

Each of these integrals can be evaluated using Lemma 9.5.1. This leads to
∞
x2  x1−2k − 1  xρ+1 − 1 ζ  (0)
φ(x) = − − − x,
2 2k(2k − 1) ρ
ρ(ρ + 1) ζ(0)
k=1
9.5. A Proof of the Prime Number Theorem 295

or
∞
x2  x1−2k  xρ+1
φ(x) = − − − Ax + B,
2 2k(2k − 1) ρ
ρ(ρ + 1)
k=1
 ∞ 
ζ  (0) 1 1
where A = and B = + .
ζ(0) 2k(2k − 1) ρ
ρ(ρ + 1)
k=1
It remains to prove that the integral can be taken inside the infinite sums in
(9.5.4). The kth term of the first sum is
xz+1
(9.5.5) .
2k(z + 2k)(z + 1)
The numerator of this fraction is bounded on the vertical line Re(z) = b. With
z = b + it, we estimate the middle factor of the denominator as follows:
|z + 2k|2 = t2 + (b + 2k)2 = t2 + b2 + 4bk + 4k2
≥ t2 + b2 + 4k2 = |z|2 + (2k)2 ≥ 4|z|k.
Thus,
|z + 2k| ≥ 4|z|1/2 k1/2 .
The right factor of the denominator satisfies |z +1| ≥ |z| since z = b+it has positive
real part. Hence, the fraction(9.5.5) has modulus less than or equal to a constant
times |z|−3/2 k−3/2 . It follows that, in the first infinite sum the integral of each term
over Re(z) = b exists and the integral of the sum is the sum of the integrals and
the latter sum is absolutely convergent (see Exercise 9.5.4).
The term involving ρ of the second infinite sum in (9.5.4) is
xz+1
(9.5.6) .
ρ(z − ρ)(z + 1)
The numerator is bounded by xb+1 on Re(z) = b. With z = b + iy, ρ = β + iγ, and
c = b − 1, we estimate the denominator as follows: Since |z − ρ| ≥ (|y − γ| + c)/2
and |z + 1| ≥ (|y| + b + 1)/2 ≥ (|y| + c)/2, we have
|ρ(z − ρ)(z + 1)| ≥ |γ|(|y − γ| + c)(|y| + c)/4.
Thus, the modulus of (9.5.6) is less than or equal to
1
g(y) = 4xb+1 |γ|−1 h(y) where h(y) = .
(|y − γ| + c)(|y| + c)
If we divide the real line into three subintervals by cutting at y = 0 and y = γ,
then, on each of these subintervals, the absolute values in h can be eliminated, and
the integral of h with respect to y can be evaluated using the method of partial
fractions. The result (Exercise 9.5.5) is that the integral of h over each of the
unbounded subintervals is
|γ|−1 log(|γ|/c + 1),
while the integral of h over the bounded subinterval is
2(|γ| + 2c)−1 log(|γ|/c + 1) ≤ 2|γ|−1 log(|γ|/c + 1).
Thus, the integral of g over (−∞, ∞) is less than or equal to
16xb+1 |γ|−2 log(|γ|/c + 1)
296 9. The Gamma and Zeta Functions

and, since log(|γ|/c) ≤ (|γ|/c)1/2 , this integral is less than or equal to

(9.5.7) 16c1/2 xb+1 |γ|−3/2 .
Since, by Corollary 9.3.8, the sequence of roots ρ = β + iγ has exponent of conver-
gence at most 1, the same is true of the sequence of imaginary parts γ. It follows
that the terms (9.5.7) are the terms of a convergent series. Thus, Exercise 9.2.7,
modified to cover integrals over (−∞, ∞), implies that the integral of the sum is
the sum of the integrals for the series of terms given by (9.5.6). 

The Prime Number Theorem.

Theorem 9.5.5. If π(x) is the number of primes less than or equal to x, then
log x
lim π(x) = 1.
x→∞ x
Proof. By Theorems 9.4.4 and 9.4.2, it suffices to prove that limx→∞ φ(x)/x2 =
1/2.
By Theorem 9.5.4,
∞
φ(x) 1  x−1−2k  xρ−1 A B
(9.5.8) = − − − + 2.
x 2 2 2k(2k − 1) ρ
ρ(ρ + 1) x x
k=1
Each of the infinite sums in this expression involves only negative powers of x and
each of them is absolutely convergent at x = 1. It follows that both infinite series
converge uniformly in x on [1, ∞). Thus, in taking the limit of φ(x)/x2 as x → ∞,
we may take the limit inside the infinite sums. Since each term on the right side of
(9.5.8) has limit 0 except the term 1/2, the theorem is proved. 

Exercise Set 9.5

x 
1. Verify that 1 ψ(u) du = pm ≤x (x − pm ) log p, where p is prime and m is a
positive integer.
2. Prove that if p and r are arbitrary positive numbers, there is a constant C
such that logp (t) ≤ Ctr for all t > 1.
3. Give a direct proof of Lemma 9.5.1, using residue theory, but without inter-
preting the integal as a Fourier transform and applying Theorem 5.3.2.
4. For each n, let gn be be a continuous function
∞ on (−∞, ∞) satisfying |gn (t)| ≤
cn f (t), where f ≥ 0 is integrable and 1 cn is a convergent series of positive
numbers. Use Exercise 9.2.7 to prove that
∞  ∞
  ∞ 
∞
gn (t) dt = gn (t) dt.
n=1 −∞ −∞ n=1

5. Verify the claims made near the end of the proof of Theorem 9.5.4 regarding
the integral of
1
h(y) =
(|y − γ| + c)(|y| + c)
over each of the subintervals of (−∞, ∞) created by cutting at y = 0 and
y = γ.
9.5. A Proof of the Prime Number Theorem 297

6. Prove that log(s + 1) ≤ s1/2 for s ∈ (0, ∞). This was used near the end of the
proof of Theorem 9.5.4.
7. Verify the statement about taking the limit inside  the integral in the proof
of Theorem 9.5.5. That is, prove that if a series ∞ n=1 un (x) of functions on
[1, ∞) converges uniformly absolutely on [1, ∞), then
∞ ∞
lim un (x) = lim un (x),
x→∞ x→∞
n=1 n=1
provided each limit on the right converges.
8. Prove that if f is a function analytic in an open set containing b + iw and
g(z) = f (b + iz), then g is analytic in an open set containing w and
Res(g, w) = −i Res(f, b + iw).
Bibliography

[1] Ahlfors, L., Complex Analysis, 2nd ed., McGraw-Hill, 1966.

[2] Ash, R., Complex Variables, Academic Press, 1971.
[3] Brown, J. W. and Churchill, R. V., Complex Variables and Applications, McGraw-
Hill, 2009.
[4] Edwards, H. M., Riemann’s Zeta Function, Dover, 1974.
[5] Jeffrey, A., Complex Analysis and Applications, CRC Press, 1992.
[6] Lang, S., Complex Analysis, Addison-Wesley, 1977.
[7] Marsden, M. H. and Hoffman, M. J., Basic Complex Analysis, 3rd ed., W. H. Free-
man, 1999.
[8] Rudin, W., Real and Complex Analysis, McGraw-Hill, 1987.
[9] Saff, E. B. and Snider, A. D., Fundamentals of Complex Analysis for Mathematics,
Science, and Engineering, 3rd ed., Prentice Hall, 2003.

299
Index

E ◦ , 30 limz→a f (z), 31
Ep (z), 251
Li(x), 287 abelian group, 110
P 1 (C), 194 free, 110
S 2 , 192 absolute convergence, 11
Z(f ), 96 absolute value, 5
C, 2 absolutely integrable function, 161
affine transformations, 198
C2 , 194
analytic at ∞, 192
Γ(z), 270
analytic continuation, 221
Indγ (z), 64
across an analytic curve, 223
Res(f, z0 ), 131 along a curve, 227
arctan z, 18 analytic covering map, 233
E, 30 analytic curve, 223
cos z, 17 analytic function, 34
ez , 14 from S 2 to S 2 , 193
(γ), 52 higher derivatives, 86
exp(z), 14 local factorization, 95, 98

γ
f (z) dz, 45 power series expansion, 84
log z, 21 zeroes, 96
F, 205 analytic function element, 227
∂E, 30 angle preserving maps, 185
∂, 112 annulus, 125
antiderivative, 55
φ(x), 290
existence of, 61
π(x), 269, 287
arctan function, 18
ψ(x), 288
auxiliary polynomial, 92
sin z, 17
√
z, 23 bi-analytic map, 105
tan z, 17 binomial formula, 15, 18
ξ(z), 277 Bolzano-Weierstrass Theorem, 59
ζ(z), 275 Borel-Carathéodory Theorem, 258
az , 23 boundary, 30
z a , 23 boundary map, 112

301
302 Index

bounded function, 89 convergence, 8

of improper integrals, 160
canonical product, 260 of infinite products, 246
order, 264 uniform, 247
Cauchy’s estimates, 87 of sequences, 9
Cauchy’s Integral Formula, 65 of series, 10
classical form, 123 convex set, 61
for convex sets, 65 cos function, 17
general form, 119 cotan function
Cauchy’s Integral Theorem, 54 properties, 179
classical form, 123 counting zeroes and poles, 134
for a triangle, 57 curve, 41
for convex sets, 63 closed, 49
general form, 120 differentiable, 41
Cauchy-Riemann equations, 38 linear, 143
Cauchy-Schwarz inequality, 6 parameterized, 41
chain, 110 piecewise linear, 143
chain rule, 36 piecewise smooth, 41
characteristic polynomial, 92 simple closed, 121
circles in S 2 , 199 smooth, 41
circles meeting at right angles, 237 curvilinear triangle, 238
closed set, 28 cycle, 113
closure, 30
compact set, 54 de la Vallée-Poussin, 288
comparison test, 11 deleted neighborhood, 31
complex derivative, 34 derivative, 34
properties, 36 logarithmic, 248
complex numbers, 2 of a curve, 41
field properties, 3 of a power series, 82
graphing, 5 differentiable, 34
why needed, 1 differential equations, 92
complex powers, 23 exponential solution, 92
Complex Projective Space, 194 Dirichlet problem, 208
component, 148 solution on D, 211
conformal automorphism, 197 discrete subset, 97
of S 2 , 198 distance between curves, 143
of the unit disc, 201 division in C, 7
conformal equivalence, 185, 187
conformal map, 185, 186 eigenvalues of matrices, 92
examples, 188 electrostatics, 214
has complex derivative, 186 entire function
conjugate pairs, 5 of finite order, 258
conjugation in C, 5 with no zeroes, 259
connected component, 67 equivalent, 111
connected set, 67 E-equivalent, 111
continuous function, 32 essential singularity, 98
alternate characterization, 32 Euclidean distance, 5
contour, 41 Euclidean norm, 5
converge Euler’s identity, 16
absolutely, 160 exponent of convergence, 260
pointwise, 75 exponential function, 14
uniformly, 75 law of exponents, 15
Index 303

properties, 17 properties, 67
inf, 79
factoring polynomials, 91 infinite products
finite order convergence, 246
of an entire function, 258 uniform convergence, 247
Fourier Inversion Formula, 172 integral
Fourier transform, 167 over a 1-chain, 111
of normal distribution, 170 over a path, 45
using residues, 167 Riemann, 43
function φ, 290 integrating using residues, 159
function ψ, 288 interior, 30
functions of finite order, 258 inverse Fourier transform, 172
zeroes, 263 inverse Laplace transform, 175
Fundamental Theorem of Algebra, 90 inversion in C, 7
inversion transformation, 198
gamma function, 270
isolated point, 31, 96
has no zeroes, 273
isolated singularity, 98
infinite product formula, 274
meromorphic continuation, 271 Jensen’s Formula, 262
geometric series, 11 join of two paths, 50
greatest lower bound, 79 Jordan Curve Theorem, 121
group homomorphism, 112
kernel, 202
Hadamard, 288
Hadamard’s Theorem, 265 Laplace transform, 175
harmonic conjugate, 40 Laplace’s equation, 39
existence, 105 Laurent series expansion, 127
harmonic function, 39, 105 least upper bound, 78
maximum principle, 106 length of a path, 52
mean value property, 106 lifting a map, 232
Harnack’s Inequality, 213 through a covering map, 234
heat equation, 214 liminf, 79
heat flow, 214 limit of a function, 31
Heine-Borel Theorem, 54 limit of a sequence, 9
homeomorphic sets, 152 limsup, 79
homeomorphism, 195 linear fractional transformation, 197
homologous cycles, 115 Liouville’s Theorem, 89
homologous to 0, 115 log function, 21
homotopic curves, 145 branches, 22
homotopy, 145 is analytic, 39
hydrodynamics, 214 principal branch, 22
properties, 22
Identity Theorem, 97 logarithm of a function, 97
imaginary part, 5 logarithmic derivative, 248
improper integral, 160 long division of power series, 155
converge absolutely, 160
principal value, 160 Maximum Modulus Theorem, 103
indented contours, 172 Mellin transform, 176
independence of path, 63 using residues, 176
index, 64 meromorphic function, 101
for curves, 144 factorization, 255
locally constant, 144 inverse of, 101
of a cycle, 114 with pole at ∞, 194
304 Index

Mittag-Leffler Theorem, 255 ratio test, 11

Möbius transformation, 197 real part, 5
modulus, 5 reflection
Monodromy Theorem, 230 through an arc, 225
Montel’s Theorem, 203 through a circle, 237
Morera’s Theorem, 87 through a curve, 224
removable singularity, 98
neighborhood, 28 Residue Theorem, 131
neighborhood of ∞, 192 residues, 131
Neumann problem, 214 applications
norm, 5 evaluating integrals, 159
normal family, 203 Fourier transform, 167
integrals on the line, 162
open set, 28
Mellin transform, 176
parameter independence, 48 summing series, 181
path, 42 computation, 153
closed, 49 of quotients, 156
simple closed, 121 Riemann Hypothesis, 287, 288
path connected, 67 Riemann integral, 43
path integral, 45 Riemann Mapping Theorem, 206
additivity and linearity, 51 Riemann Sphere, 191
parameter independence, 48 analytic functions on, 192
Picard Theorems, 99 as C ∪ ∞, 192
Big, 241 open sets, 192
Little, 240 topology of, 192
pointwise convergence, 75 Riemann zeta function, 275
Poisson root functions, 23
Formula, 212 root test, 11
Integral, 209 roots of a complex number, 20
integral roots of unity, 21
boudary values, 210 Rouché’s Theorem, 142
kernel, 209
Summation Formula, 277 Schwarz Reflection Principle, 222
polar form, 19 for reflection through a curve, 225
pole Schwarz’s Lemma, 104
of order k, 98 separated, 148
simple, 98 separated set, 67
polynomial open set, 67
characterization, 93 series of complex numbers, 10
factoring, 91, 95 simple closed curve, 121
growth, 93 simple closed path, 121
root, 90 inside, 122
power series, 12 outside, 122
kth derivative, 83 simple point of a path, 71, 121
coefficients, 83 simply connected, 147
derivative of, 82 simply split, 70
integration of, 80 sin function, 17
powers of a complex number, 20 singularity
Prime Number Theorem, 288, 296 essential, 98
principal value, 160 isolated, 98
removable, 98
radius of convergence, 12, 78, 79 smooth curve, 41
Index 305

square root function, 23 of infinite products, 247

stereographic projection, 195 of series, 77
summing series using residues, 181 upper bound, 78
sup, 79
symmetry of ξ, 278 vanish at infinity, 124

tan function, 17 Weierstrass

term test, 11 factorization, 253, 254
topology, 29 product, 252
triangle inequality, 6 Theorem, 252
trigonometric functions, 17 Weierstrass M -test, 78

uniform convergence, 75 zeroes and singularities, 95

and continuity, 76 zeta function, 275
and integration, 76 meromorphic extension, 279
of analytic functions, 88 product formula, 275
zeroes, 276
Titles in This Series
Volume
16 Joseph L. Taylor
Complex Variables
15 Mark A. Pinsky
Partial Differential Equations and Boundary-Value Problems with Applications,
Third Edition
14 Michael E. Taylor
Introduction to Differential Equations
13 Randall Pruim
Foundations and Applications of Statistics: An Introduction Using R
12 John P. D’Angelo
An Introduction to Complex Analysis and Geometry
11 Mark R. Sepanski
Algebra
10 Sue E. Goodman
Beginning Topology
9 Ronald Solomon
Abstract Algebra
8 I. Martin Isaacs
Geometry for College Students
7 Victor Goodman and Joseph Stampfli
The Mathematics of Finance: Modeling and Hedging
6 Michael A. Bean
Probability: The Science of Uncertainty with Applications to Investments,
Insurance, and Engineering
5 Patrick M. Fitzpatrick
Advanced Calculus, Second Edition
4 Gerald B. Folland
Fourier Analysis and Its Applications
3 Bettina Richmond and Thomas Richmond
A Discrete Transition to Advanced Mathematics
2 David Kincaid and Ward Cheney
Numerical Analysis: Mathematics of Scientific Computing, Third Edition
1 Edward D. Gaughan
Introduction to Analysis, Fifth Edition
The text covers a broad spectrum between basic and advanced
complex variables on the one hand and between theoretical

Photograph by Christina M. Taylor

and applied or computational material on the other hand.
With careful selection of the emphasis put on the various
sections, examples, and exercises, the book can be used in a
one- or two-semester course for undergraduate mathematics
majors, a one-semester course for engineering or physics
majors, or a one-semester course for first-year mathematics
graduate students. It has been tested in all three settings at the
University of Utah.
The exposition is clear, concise, and lively. There is a clean and modern approach
to Cauchy’s theorems and Taylor series expansions, with rigorous proofs but no
long and tedious arguments. This is followed by the rich harvest of easy conse-
quences of the existence of power series expansions.
Through the central portion of the text, there is a careful and extensive treat-
ment of residue theory and its application to computation of integrals, conformal
mapping and its applications to applied problems, analytic continuation, and the
proofs of the Picard theorems.
Chapter 8 covers material on infinite products and zeroes of entire functions.
This leads to the final chapter which is devoted to the Riemann zeta function, the
Riemann Hypothesis, and a proof of the Prime Number Theorem.

For additional information

and updates on this book, visit
www.ams.org/bookpages/amstext-16

AMS on the Web

AMSTEXT/16 www.ams.org

Sally
The This series was founded by the highly respected
mathematician and educator, Paul J. Sally, Jr.