You are on page 1of 525

Function Theory of

One Complex Variable


THIRD EDITION

Robert E. Greene
Steven G. Krantz

Graduate Studies
in Mathematics
Volume 40

American Mathematical Society

Function Theory of
One Complex Variable
THIRD EDITION

Robert E. Greene
Steven G. Krantz

Graduate Studies
in Mathematics
Volume 40

+J:1t,"

"'

.en"'=

'""NDEI>'

American Mathematical Society


Providence, Rhode Island

Editorial Board

Steven G. Krantz
David Saltman (Chair)
David Sattinger
Ronald Stern
2000

Mathematics Subject Classification. Primary 30-01; Secondary 30-02, 30-03.

For additional information and updates on this book, visit

.ams.org/bookpages/gsm-40

www

Library of Congress Cataloging-in-Publication Data


Greene, Robert Everist, 1943F\.tnction theory of one complex variable
p. cm. -

( Graduate

Robert E. Greene and Steven G. Krantz.-3rd ed.

studies in mathematics, ISSN 1065-7339; v. 40 )

Includes bibliographical references and index.

( alk.

ISBN 0-8218-3962-4

paper )

1. functions of complex variables.

I. Krantz, Steven G.

( Steven

George ) , 1951-

II. Title.

III. Series.

Q A331.7.G75

2006
2005057188

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, sy stematic copying, or multiple reproduction of any material in this publication
is permitted only under license from the American Mathematical Society.

Requests for such

permission should be addressed to the Acquisitions Department, American Mathematical Society,


201 Charles Street, Providence, Rhode Island 02904-2294, USA. Requests can also be made by
e-mail to reprint-permissionams. org.
Second Edition
Third Edition

2002 by the American Mathematical Society.

2006 by the American Mathematical Society. All rights reserved.

The American Mathematical Society retains all rights


except those granted to the United States Government.
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

11 10 09 08 07 06

To Paige and Randi

Contents

Preface to the Third Edition


Preface to the Second Edition
Preface to the F irst Edition
Acknowledgments
Chapter 1.

Fundamental Concepts

xiii
xv
xvii
xix
1

1.1.

Elementary Properties of the Complex Numbers

1.2.

Further Properties of the Complex Numbers

1.3.

Complex Polynomials

1.4.

Holomorphic Functions, the Cauchy-Riemann Equations,


and Harmonic Functions

1.5.

Real and Holomorphic Antiderivatives

Exercises
Chapter 2.

10

14
17
20

Complex Line Integrals

29

2.1.

Real and Complex Line Integrals

29

2.2.

Complex Differentiability and Conformality

34

2.3.

Antiderivatives Revisited

40

2.4.

The Cauchy Integral Formula and the Cauchy


Integral Theorem

43
-

Vll

viii

Contents

2.5.

The Cauchy Integral Formula: Some Examples

2.6.

An Introduction to the Cauchy Integral Theorem and the


Cauchy Integral Formula for More General Curves

Exercises
Chapter 3.

50

53
60

Applications of the Cauchy Integral

69

3.l.

Differentiability Properties of Holomorphic Functions

69

3.2.

Complex Power Series

74

3.3.

The Power Series Expansion for a Holomorphic Function

81

3.4.

The Cauchy Estimates and Liouville's Theorem

84

3.5.

Uniform Limits of Holomorphic Functions

88

3.6.

The Zeros of a Holomorphic Function

90

Exercises
Chapter 4.
4.l.

94
Meromorphic Functions and Residues

105

The Behavior of a Holomorphic Function Near an


Isolated Singularity

105

4.2.

Expansion around Singular Points

109

4.3.

Existence of Laurent Expansions

113

4.4.

Examples of Laurent Expansions

119

4.5.

The Calculus of Residues

122

4.6.

Applications of the Calculus of Residues to the


Calculation of Definite Integrals and Sums

4.7.

Meromorphic Functions and Singularities at Infinity

Exercises
Chapter 5.

128
137
145

The Zeros of a Holomorphic Function

157

5.l.

Counting Zeros and Poles

157

5.2.

The Local Geometry of Holomorphic Functions

162

5.3.

Further Results on the Zeros of Holomorphic Functions

166

5.4.

The Maximum Modulus Principle

169

5.5.

The Schwarz Lemma

171

Exercises

174

Contents

ix

Chapter 6.

Holomorphic Functions as Geometric Mappings

179

6.1.

Biholomorphic Mappings of the Complex Plane to Itself

180

6.2.

Biholomorphic Mappings of the Unit Disc to Itself

182

6.3.

Linear Fractional Transformations

184

6.4.

The Riemann Mapping Theorem: Statement and


Idea of Proof

189

6.5.

Normal Families

192

6.6.

Holomorphically Simply Connected Domains

196

6.7.

The Proof of the Analytic Form of the Riemann


Mapping Theorem

Exercises
Chapter 7.

198
202

Harmonic Functions

207

7.1.

Basic Properties of Harmonic Functions

208

7.2.

The Maximum Principle and the Mean Value Property

210

7.3.

The Poisson Integral Formula

212

7.4.

Regularity of Harmonic Functions

218

7.5.

The Schwarz Reflection Principle

220

7.6.

Harnack's Principle

224

7.7.

The Dirichlet Problem and Subharmonic Functions

226

7.8.

The Perron Method and the Solution of the


Dirichlet Problem

7.9.

Conformal Mappings of Annuli

Exercises
Chapter 8.

236
240
243

Infinite Series and Products

255

8.1.

Basic Concepts Concerning Infinite Sums and Products

255

8.2.

The Weierstrass Factorization Theorem

263

8.3.

The Theorems of Weierstrass and Mittag-Leffler:


Interpolation Problems

Exercises
Chapter 9.

266
274

Applications of Infinite Sums and Products

279

Contents

9.l.

Jensen's Formula and an Introduction to


Blaschke Products

279

9.2.

The Hadamard Gap Theorem

285

9.3.

Entire Functions of Finite Order

288
296

Exercises
Chapter 10.

Analytic Continuation

299

10.l.

Definition of an Analytic Function Element

299

10.2.

Analytic Continuation along a Curve

304

10.3.

The Monodromy Theorem

307

10.4.

The Idea of a Riemann Surface

310

10.5.

The Elliptic Modular Function and Picard's Theorem

314

10.6.

Elliptic Functions

323
330

Exercises
Chapter 11.

Topology

11.l.

Multiply Connected Domains

11.2.

The Cauchy Integral Formula for Multiply


Connected Domains

11.3.

11.5.

335

338

Holomorphic Simple Connectivity and Topological


Simple Connectivity

11.4.

335

343

Simple Connectivity and Connectedness of


the Complement

344

Multiply Connected Domains Revisited

349
352

Exercises
Chapter 12.

Rational Approximation Theory

363

12.l.

Runge's Theorem

363

12.2.

Mergelyan's Theorem

369

12.3.

Some Remarks about Analytic Capacity

378
381

Exercises
Chapter 13.
13.l.

Special Classes of Holomorphic Functions

Schlicht Functions and the Bieberbach Conjecture

385
386

xi

Contents

13.2.

Continuity to the Boundary of Conformal Mappings

392

13.3.

Hardy Spaces

401

13.4.

Boundary Behavior of Functions in Hardy Classes


[An Optional Section for Those Who Know
Elementary Measure Theory]

Exercises
Chapter 14.

406
412

Hilbert Spaces of Holomorphic Functions, the Bergman


Kernel, and Biholomorphic Mappings

415

14.1.

The Geometry of Hilbert Space

415

14.2.

Orthonormal Systems in Hilbert Space

426

14.3.

The Bergman Kernel

431

14.4.

Bell's Condition R

438

14.5.

Smoothness to the Boundary of Conformal Mappings

443

Exercises
Chapter 15.

446

Special Functions

449

15.1.

The Gamma and Beta Functions

449

15.2.

The Riemann Zeta Function

457

Exercises
Chapter 16.

467

The Prime Number Theorem

471

16.0.

Introduction

471

16.1.

Complex Analysis and the Prime Number Theorem

473

16.2.

Precise Connections to Complex Analysis

478

16.3.

Proof of the Integral Theorem

483

Exercises

485

APPENDIX A: Real Analysis

487

APPENDIX B: The Statement and Proof of Goursat's Theorem

493

References

497

Index

501

Preface to the Third


Edition

This third edition follows the overall plan and even the specific arrangement
of topics of the second edition, but there have been substantial changes in
matters of detail. A considerable number of the proofs, especially in the later
chapters, have been corrected, clarified, or simplified. Many of the exercises
have been revised, and in many cases the exercises have been rearranged to
make for greater consistency and less duplication. The mathematical roads
that this new edition follows are the same as before, but we hope that the
ride is considerably smoother.
We are indebted to Harold Boas and Gerald B. Folland for their ex
tremely careful reading of the second edition in the course of their using the
book as a text.

They provided far more suggestions and corrections than

we had any right to expect of anyone but ourselves, and to the extent that
this edition is superior to the previous, it is very largely to that extent that
we are in their debt. Any remaining errors are, of course, our responsibility.
Rahul Fernandez brought mathematical expertise, typesetting skills, and
a great deal of patience to the daunting task of taking our heavily marked
and indeed sometimes scribbled-upon manuscript of the second edition and
making this third one. We are grateful to him for his efforts. We also thank
the publishing staff of the American Mathematical Society for their willing
ness to undertake a third edition and for their support in general.

Robert E. Greene and Steven G. Krantz

Xlll

Preface to the Second


Edition

In this second edition, the exposition of topological matters in Chapter 11


has been revised; the prime number theorem has been given a clearer and
shorter proof; and various small errors and unclear points throughout the
text have been rectified.
We have been heartened by an enthusiastic response from our read
ers.

We are indebted to many of our colleagues and students for sugges

tions.

We especially thank Harold Boas, Robert Burckel, Gerald Folland,

and Martin Silverstein, who each scrutinized the original text carefully and
contributed valuable ideas for its improvement. The authors are indebted
to Rahul Fernandez for carefully correcting the

TEX

file of the entire text.

Robert E. Greene and Steven G. Krantz

xv

Preface to the First


Edition

T his book is a text for a first-year graduate course in complex analysis. All
material usually treated in such a course is covered here, but our book is
based on principles that differ somewhat from those underlying most intro
ductory graduate texts on the subject.
First of all, we have developed the idea that an introductory book on
this subject should emphasize how complex analysis is a natural outgrowth
of multivariable real calculus. Complex function theory has, of course, long
been an independently flourishing field. But the easiest path into the subject
is to observe how at least its rudiments arise directly from ideas about
calculus with which the student will already be familiar.

We pursue this

point of view both by comparing and by contrasting complex variable theory


with real-variable calculus.
Second, we have made a systematic attempt to separate aTJ.alytical ideas,
belonging to complex analysis in the strictest sense, from topological con
siderations.

Historically, complex analysis and topology grew up together

in the late nineteenth century. And, long ago, it was natural to write com
plex analysis texts that were a simultaneous introduction to both subjects.
But topology has been an independent discipline for almost a century, and it
seems to us only a confusion of issues to treat complex analysis as a justifica
tion for an introduction to the topology of the plane. Topological questions
do arise naturally, of course; but we have collected all of the difficult topo
logical issues in a single chapter (Chapter 11), leaving the way open for a
more direct and less ambivalent approach to the analytical material.

xvii

xviii

Preface to the First Edition

Finally, we have included a number of special topics in the later chap


ters that bring the reader rather close to subjects of current research. These
include the Bergman kernel function, HP spaces, and the Bell-Ligocka ap
proach to proving smoothness to the boundary of biholomorphic mappings.
These topics are not part of a standard course on complex analysis (i.e.,
they would probably not appear on any qualifying examination), but they
are in fact quite accessible once the standard material is mastered.
A large number of exercises are included, many of them being routine
drill but many others being further developments of the theory that the
reader can carry out using the hints and outlines provided.

One of the

striking features of complex analysis is its interconnectedness: Almost any


of the basic results can be used to prove any of the others.

It would be

tedious to explore all these implications explicitly in the text proper. But
it is important, educational, and even rather entertaining for students to
follow these logical byways in the exercises.
We hope that the distinctive aspects of our book will make it of inter
est both to student and instructor and that readers will come to share the
fascination with the subject that led us to write this book.

Robert E. Greene and Steven G. Krantz

Acknowledgments

Over the years, a number of students, friends, and colleagues have used or
read drafts of this text. As a result, they have contributed many suggestions
and have helped to detect errors.
We are pleased to thank Lynn Apfel, Harold Boas, Anthony Falcone,
Robert Burckel, H. Turgay Kaptanoglu, Judy Kenney, Peter Lampe, Holly
Lowy, Anne Marie Mosher, Michelle Penner, Martin Silverstein, Kevin Stock
ard, Kristin Toft, W ilberd van der Kallen, and David Weiland. Urban Ce
grell used the manuscript in his complex analysis class in Ume, Sweden.
Both he and his students, Anders Gustavsson, Per Jacobsson, Weigiang Jin,
Klas Marstrom, and Jonas W iklund, gave us several useful ideas.
We thank Jeanne Armstrong for typing an early draft of the book.

R.E.G.
S.G.K.

xix

Chapter 1

Fundamental Concepts

1.1. Elementary Properties of the Complex Numbers


We take for granted the real numbers, which will be denoted by the symbol

R Then we set
consist of

IR2

IR2 = {(x,y) : x

IR , y

IR}.

The complex numbers tC

equipped with some special algebraic operations. Namely, one

defines

(x,y) + (x',y')
(x, y) (x',y')

(x + x',y + y') ,
(xx' - yy',xy' + yx').

You can check for yourself that these operations of+ and

are commutative

and associative.
It is both conventional and convenient to denote

by

i.

We also adopt the convention that, if

(1, 0)

a E IR, then

by

and

(0, 1)

a (x,y) =(a, 0) (x, y) =(ax, ay).

(x,y) can be written in one and only one way


x 1 + y i with x,y E R We usually write the number even more
as x + iy. Then our laws of addition and multiplication become

Then every complex number


in the form
succinctly

(x + iy) + (x' + iy')


(x + x') + i(y + y'),
(x + iy) (x' + iy') = (xx' - yy') + i(xy' + yx').
Observe that

i i = -1.

Moreover, our multiplication law is consistent with

the real multiplication introduced in line


The symbols z, w,

(* ) .

are frequently used to denote complex numbers.

Unless it is explicitly stated otherwise, we always take z


u

iv , ( = e + i'f/.

The real number

= x + iy

is called the real part of z and is

1.

x =Re z.
y =Imz.
written

Likewise

Fundamental Concepts

y is called the imaginary part of z and is written

x -iy is by definition the conjugate of the complex


x + iy. We denote the conjugate of a complex number z by the
symbol z. So if z =x + iy, then z =x - iy.
Notice that z + z =2x, z - z =2iy. You should verify for yourself that
The complex number

number

z + w,
ZW.
A complex number is real (has no imaginary part) if and only if
imaginary (has zero real part) if

z=

z = z.

It is

(x, y) to (0, 0)
modulus (or absolute value) of
z = x + iy, and we write lzl= Jx2 + y2. Notice that
The ordinary Euclidean distance of

is

call this number the

Jx2 + y2 .

We also

the complex number

z. z =x 2 + y 2 =lzl2
You should check for yourself that the distance from
Verify also that

lz wl = izl lwl

w is lz - wl.
I zl izl and

to

(square both sides). Also Re

I Im zl lzl.
Let 0 =0 + iO. If z EC, then z + 0 = z. Also, letting -z = -x - iy, we
notice that z + (-z) =0. So every complex number has an additive inverse.
Since 1 = 1 + iO, it follows that 1 z =z 1 = z for every complex number
z. If izl i= 0, then izl2 i= 0 and
z.

z
izl2
=
= 1.
izl2 izl2

So every nonzero complex number has a multiplicative inverse. It is natural


to define

1/z to be the multiplicative inverse z/izl2

to define

1
zw
z
-=z-=-w
w lwl2

You can also see that

for

of

z and, more generally,

w i= 0 .

z/w=z/w.

Observe now that multiplication and addition satisfy the usual distribu
tive, associative, and commutative (as previously noted) laws. So C is

afield.

It follows from general properties of fields, or you can just check directly,

unique additive inverse and every nonzero


unique multiplicative inverse. Also C contains a copy

that every complex number has a


complex number has a

of the real numbers in an obvious way:

IR

1-+

x + iO EC.

1.2. Further Properties

It is easy to see that this identification respects addition and multiplication.


So we can think of C as a field extension of R It is a larger field which
contains the field R
It is not true that every polynomial with real coefficients has a real root.

p(x)

For instance,

x2 + 1

has no real roots. Historically, already existing

strong interest in the complex numbers was further stimulated by Gauss's


proof of the long-suspected but still remarkable fact, known as the funda
mental theorem algebra, that every polynomial with complex coefficients
has a complex root.

This basic fact about the complex numbers makes it

natural to study the class of all polynomials with complex coefficients. The
point of view of complex analysis, however, is that there is a much larger

( known

class of functions

as the holomorphic functions ) which share many

of the most useful properties of polynomials.

Moreover, the class of holo

morphic functions is closed under certain operations

( such

as division by a

nonvanishing function ) for which the collection of polynomials is not closed.


The holomorphic functions can be easily identified by a simple differential
condition. These are the functions which we shall study in this book.
In this book we are presupposing that the reader has had some experi
ence with the arithmetic of complex numbers; the material we are presenting
at first is in the nature of a review. The first few exercises at the end of the
chapter provide some drill in basic calculations with the complex numbers.
The later exercises introduce some arithmetic properties that, though they
are elementary, may not be familiar . The reader should examine these ex
ercises carefully since the properties introduced will play an important role
in what follows.

1.2. Further Properties of the Complex Numbers


We first consider the complex exponential, which we define as follows:

(1)

If

is real, then

as in calculus.

(2)

If

iy is pure imaginary, then

ez

(3)

If

x+ iy,

(2)

and

(3)

eiy =cosy+ i siny.

then

ez
Parts

ex+iy =ex

( cosy+ i siny ) .

of the definition, due to Euler, may seem somewhat

arbitrary. We shall now show, using power series, that these definitions are

1. Fundamental Concepts

perfectly natural. We shall wait until Section 3.2 to give a careful presenta

tion of the theory of complex power series. So the power series arguments
that we are about to present should be considered purely formal and given
primarily for motivation.
Since, as was noted in

we have

(1),

ex=

oo

x
L-:r'

j=O J.

then it is natural to attempt to define

00

'"' z1
ez - --:r
j=O J.

If we assume that this series converges in some reasonable sense and that
it can be manipulated like the real power series with which we are familiar,
then we can proceed as follows:
If z =

iy, y

E 'then

(iy)J
.

,
J

j=O

.
y2
iy3
y4
+ iy 2! 3! + 4!
y6
iy5
iy7
yB
.
+
+
8! + ..
5! 6T 7!
l _ y2
yB _
y4
y6
. ..
+
+
_

2!

4!

6!

8!

y7
y3
y5
. ..
+i y +
7! 5!
3!
cos y + i sin y.

)
)

By formal manipulation of series,it is now easily checked,using the definition

( * ) ,that

ea+b= eaeb,

T hen for z = x

+ iy

IL,.
any a ' b E ""

we have

ez

ex+iy exeiy
ex ( cos y + i sin y),
=

giving thus a formal "demonstration" of our definition of exponential.


To stress that there is no circular reasoning involved here, we reiterate
that the

definition

of the complex exponential is that, for z

ez = ex ( cos y + i sin y).

+ iy,

1.2. Further Prop erties

The justification with power series is, at this point, to be taken as intuitive;
it will be made precise in Section 3.6.
A consequence of our definition of the complex exponential is that if
E C,
l I
Figure

1.1).

1,

then there is a number e, 0 e

<

2rr,

such that

Here 0 is the (signed) angle between the positive

ei() (see

axis and the

ray o.
Now if

is any nonzero complex number, then


z

where

/ l zl

has modulus

l zl

1.

(i;1)

l zl C

Again letting 0 be the angle between the

real axis and O,

where r
number
If

l zl .

This form is called the polar representation for the complex

z.

k is an integer,
ei()

then

cos e + i sin e

cos(O +

2krr) + i sin(O + 2krr)

ei(o+2k1r) .
So a nonzero complex number has infinitely many different representations

of the form rei0, r

>

0, e ER

In order to make optimal use of the polar representation, we need to


know that our exp function behaves in the expected fashion relative to mul
tiplication, namely that

This fact follows from part

(3)

of our definition of the exponential func

tion, provided we assume the addition formulas for sine and cosine. Sine
and cosine will be put on a precise analy tic basis later on in this text, via
power series, and the addition formulas will be checked in that way. While
the traditional geometric proofs of these formulas are correct, they depend
on the rather difficult process of setting up trigonometry strictly with the
Euclidean-axiomatic-geometry setting.
As a consequence, if

1, 2, ..., then
(ez )

enz.

times

This simple formula has elegant applications, as the following examples


show.

1. Fundamental Concepts

eie

-- --

cos 9 + i sin 9
/

I
I
I
I
I
\

'

'
'

.... --

'
'

\
I
I

I
I

/
/

'

_,,,

Figure 1.1

EXAMPLE 1.2.1. To find all sixth roots of 2, we let rei9 be an arbitrary

sixth root of 2 and solve for r and 0. If

( rei9 ) 6
or

r6ei69

ei O

2 . eiO
,

2116 E JR and ()
then it follows that r
0 solve this equation. So the real
i
2116 is a sixth root of 2. This is not terribly surprising,
number 2116 e O
=

but we are not finished.


We may also solve

r6ei69

Hence
r

2116

()

2. e21ri.

27r/6

7r/3.

This gives us the number


21i6e;/3

21 i6 coS11"

/3 + isin "/3)

as a sixth root of 2. Similarly, we can solve

r6ei69

2 . e41ri

r6ei69

r6ei69

r6ei69

2 . e61ri
7r
2 . e8 i

'

'

2. el07ri

to obtain the other four sixth roots of 2:


21/6

(-

+ iv'3
2

'

2 1 !

0 )
+i

1.2. Further Properties

-21 6
/ '

21/6

(-! -iv'3)
0-i)
2

'

21/6

Further effort along these lines results only in repetition of the six roots we
have found.
These are in fact all the sixth roots. You can see this by noting that
6
2 and 60
2 implies that r6
27rk, for some k, or by noting that
(rei8)
=

the division algorithm for polynomials guarantees that z6 - 2 has at most

six distinct roots. Division for polynomials with complex coefficients works
just the same way as for polynomials with real coefficients. Another way
to see that the number 2 has no more than six sixth roots is to check by
calculation the following assertion: if we call the six sixth roots of 2 that we
have found so far by the names a1, ..., a , then
6
z6 - 2
Thus z6

(z -

a1

) (z -

2 if and only if z is one of the Gj.

EXAMPLE 1.2.2. To find all the cube roots of


r3ei38

r3ei38

It follows that the cube roots are

ei7r/6

re"

e'"'i'

reiB

ei37r/2

i,

6).

we write

ei7r/2'

i
i
(v'3 i!)
(- i)
-i.
=

r3ei38

reiB

(z

ei57r/2,
ei97r/2.

'

The (nonunique) angle 0 associated to a nonzero complex number z is

called its argument and is written arg z. For instance, arg(l + )

it is also correct to write arg(l + )

7r/4. But

97r/4, l77r/4, -77r/4, and so forth. We

generally choose the argument 0 to satisfy 0 0 < 27r unless circumstances


dictate that it is convenient to do otherwise. Because of the nonuniqueness
of arg z, the notation has to be used with caution. One should properly
speak only of an argument for z, or of the set of arguments for z, but for
historical reasons the arg z notation has been retained; it is up to the user
of the notation to be sure that the ambiguity does not lead to errors. The

1. Fundamental Con cepts

"multi-valued" nature of the function arg

z reflects deep properties of C, as

we shall see later.


Arguments are additive, subject to the ambiguity which has already

z = rei(J and w = sei'l/J, then


z w = reiO sei'l/J =(rs) ei(o+'l/J)_
under multiplication, moduli multiply while arguments add.

been noted. That is, if

Thus,

Now we need to record a few inequalities. These are familiar in form,


but are new in the context of

Proposition

Proof.

(rather than

IR).

z, w E C, then
lz + wl :'.S lzl + lwl.

1.2.3 (Triangle inequality). If

We calculate that

lz + wl2 =(z + w) (z + w)
= lzl2 + lwl2 + wz + zw
=lzl2 + lwl2 + 2Re (zw)
:'.S \zl2 + lwl2 + 2lzl lwl
=(lzl + lwl)2.

By induction, it is easy to see that


n

L Zj :'.S L \zil
j=l
j=l
Another useful variant of the triangle inequality is

lzl ;:::: lwl - lz - wl.


The proof is to write

lwl = lz + (w - z)I :'.S lzl + lw - zl,


which is equivalent to the asserted inequality.

Proposition 1.2.4 (The Cauchy-Schwarz Inequality).


and w1, ... , Wn are complex numbers, then

w
L.J J J

j=l

Proof.

For any complex number

j=l

j=l

:'.SL lzil2 L lwil2>.,


n

:'.S L lzi - >.wj \ 2.


j=l

If z1,

... , Zn

1.2. Further Properties

But this last line

L(lzjl2 + IAl2lwjl2 - 2Re(zj"Xwj))


j=l
n

Since

[Notice

j=l

j=l

j=l

is arbitrary, we may take

A LJ=lZjWj
= Lnj=l IW312
LJ=l lwj12 -=/= O;
A

that we may safely assume that

is nothing to prove.

The choice of

otherwise there

is not merely dictated by ingenious

caprice: It happens to minimize the expression on the right side of

**

as

you can check with calculus.

A, ( )
2 21LJ=lZjWjl2
L
J=lZjWj
l
I
O::; lzjl2 +
LJ=l lwjl2 LJ=l lwjl2

With the above choice of

becomes

**

or

j=l

j=l

::; L lz112 L lw112


D

as desired.

EXAMPLE 1.2.5. If
see this, fix

2::}:1 lzjl2

converges, then

a positive integer. Then

2::}:1 lzjl/j

converges.

To

t lz!I

j=l

Now the first sum is bounded above, independent of


second sum is known from calculus
n --t
n.

( use

n,

by hypothesis. The

the integral test

to converge as

oo; hence those partial sums are also bounded above, independent of

Thus

is bounded above, independent of

tlz!I
j=l J
n.

Hence the series converges.

1. Fundamental Concepts

10

1.3. Complex Polynomials


In the calculus of real variables, polynomials are the simplest nontrivial
functions. The purpose of this section is to consider complex-valued poly
nomials of a complex variable, with the idea of seeing what new features
appear.

Later we shall use the discussion as motivation for considering

more general functions.


There are several slightly different ways of looking at polynomials from
the complex viewpoint. One way is to consider polynomials in x and y,
(x, y) E R2, with complex coefficients: for example, (2 +i)xy +3iy2 +5x2.
Such polynomials give functions from R2 to C, which we could equally well

think of as functions from

C,

to

with (x, y) determined by z

x+

iy. Another kind of polynomial that we can consider is complex-coefficient


polynomials in the complex variable z, for example, i+(3+i)z+5z2. These

also give functions from

C.

to

A polynomial in z gives rise naturally to

a polynomial in x and y by substituting z

x + iy and expanding.

For

instance
i +(3 +i)z +5z

i +(3 +i)(x +iy) +5(x +iy)

2
2
i +3x-y +ix +3iy +5x + lOixy - 5y
2
2
i +(3 +i)x +(3i - l)y +5x +(10i)xy - 5y .

It is an important and somewhat surprising fact that the converse of this


expansion process does not always work:. there are many polynomials in x
and y that cannot be written as polynomials in z. Let us consider a specific
simple example: the polynomial x itself. If it were true that
x

P(z)

P(x +iy)

for some polynomial P(z) in z, then P would have to be of first degree. But
a first degree polynomial az+b

ax+iay+b cannot be identically equal to

x, no matter how we choose a and b in

C ( see

Exercise 35). What is really

going on here?
One way to write a polynomial in x and y in complex notation is to use
the substitutions

where z

z +z
=

--

=
z-z

x-iy as in Section 1.1. The point of the previous paragraph is

that, when a polynomial in x, y is converted to the z, z notation, then there


will usually be some z's in the resulting expression, and these z's may not
cancel out.
For example,
2
2
x +y

(;) ()
z

11

1.3. Complex Polynomials

z2 zz z2 z2 zz z2
4+2+4-4+2-4
z z.
You can check for yourself that there is no polynomial expression in
without any z's, that equals

x2 + y2:

z,

the occurrence of z is required.

Of course, sometimes one can be lucky and there will not be any z's:

x2

y2 + 2ixy

( z ; ) 2 ( z ) 2 + 2i ( z ; ) ( z )
z

z2 zz z2 z2 zz z2 2i(z2 - z2)
4+2+4+4-2+4+
22i
2
z .
In this example, all the z terms cancel out.
We have gone into perhaps painful detail here because in fact this feature,
that sometimes z's occur and sometimes not, is the key to complex analysis.
W hat complex analysis is about is functions, not necessarily polynomials,

z but not on z in the same sense that x2-y2 +2ixy depends


z but not on z. The question is how to extend this idea from polynomials

that depend on
on

to more general functions in a precise way.


F irst let us specify what type of function

we will consider:

R2 is open and f : U
R is a continuous
function, then J is called c1 (or continuously differentiable) on u if of/ ox
and of joy exist and are continuous on U. We write f E C1(U) for short.
More generally, if k E {O, 1, 2, ... }, then a function f on U is called Ck (k
times continuously differentiable) if f and all partial derivatives of f up to
and including order k exist and are continuous on U. We write in this case
f E Ck(U). In particular, a c0 function is just a continuous function.
A function f
u
C is called Ck if both u and v are Ck.
+ iv : U

Definition 1.3.1. If

---t

---t

Recall now that if


that

f: U

does not depend on

does not depend on

if and only if

U is connected, then we have


of/ ox = 0 on U. Likewise f

of/ oy =

0 on U. We would like to have

---t

is

if and only if

C1

and

a similar notion of differentiation to ascertain when


either

or z. For motivation, recall that

oy

0,

does not depend on

1. Fundamental Con cepts

12

We would like a similar circumstance to occur for the variables z and z


(instead of x and

!__ f
and

!__ f
oz

If z =x

(!__
2

0z

iy,

We define, for

y).

ox

(!__
2

i!__

oy

i!__

ox + oy

z =x -

: U
+ iv

=u

) f 2 (
=

) f 2 (
=

ov
+
ox
oy
ov

ox

oy

Ca C1 function,

) 2!_ (

ou

ou

---t

) 2!_ (
+

ov
ox

ou
oy

ov

ou
+
ox
oy

then one can check directly that

iy,

0
oz

z =1 ,

oz

0
azz =o,

0
oz

Notice the apparent sign reversal:

z =x +

iy

but

z =x -

iy

but

z = 0,
z = 1.

0
oz
0
Oz

1 0
2 ox

1 0
2 ox

--

i
-

0
-

2 oy

'

--.

2 oy

W hile at first puzzling, the reversal is justified by the equations

( )
*

It is straightforward to calculate that o/oz, o/az are linear:

oF
oG
,
b
(aF + bG) =a
oz
oz + oz
oF
oG
b
bG) =a
(aF
+
+
oz
oz
az
o

for any a, b E C and C1 functions F, G. It is also easy, if tedious, to verify


the Leibniz Rules:

o
oz

(F G) =

oF
oz

o
oF
0z(F G) = 0z

G+F
G+F

oG

oz ,

oG
0z .

You should work these routine facts out carefully for yourself in order to
gain facility with this new notation.
For the purposes of complex analysis, the partial differential operators

o/oz and o/Oz play a fundamental role. As an example of the use of these
new operators, we shall now obtain information about complex polynomials.
If

j, k, f, m

are nonnegative integers, then it is immediate from

( )( )
oe

om

ozl

ozm

j(j

1)

( )
*

that

.k
(z1z )

(j

- f_ +

1)k(k - 1)

(k - m + 1). zj-e-zk-m

1.3.

if

13

Complex Polynomials

f :::; j and m :::; k.

if either f, >

or

Also

()
()
(z1-zk)
aze
azm

m > k. In particular,

( ae )( -am ) f.-m
(z z ) - f. m..
_

azm

aze

I.

Now we can prove the basic fact which motivated our new notation:

Proposition 1.3.2. If

"'\'""

L__. aemzf.-zm
then p contains no term with m > 0 (that is, p contains no
-)
p(z,z

is a polynomial,
z terms) if and only if 8p/8z
Proof. If

aem

0 for all

= 0.

m > 0, then
p(z) l:aeoi
=

and

Conversely, if

whenever

8p/0z

8p
az

= 0, then

"'\'""

L__. aeofzf.
ae+m
azeazmP

-1 az az =

:=

0.

1. But the three formulas preceding the proposition show

that

evaluated at 0 is

f!m!aem

We think of a polynomial
in a sense, "independent of
when

m > 0,

z";

I: aez

with no terms containing

as being,

but the precise meaning is just that

or, equivalently, that

8p/oz

aem

= 0.

We leave the next assertion, which is a similar application of the mono


mial differentiation formulas, as an exercise.

Proposition 1.3.3. If

p(z, -z)
q(z,z)
are polynomials, and if p(z, z)

f,m.

I: aem i -zm ,
Lbemizm

q(z, z) for all z, z, then aem

The next section extends these ideas from polynomials to

C1

bem for all


functions.

14

1. Fundamental Concepts

1.4. Holomorphic Functions, the Cauchy-Riemann


Equations, and Harmonic Functions
Functions

f which

satisfy

(8/az)f

0 are the main concern of complex

analysis. We make a precise definition:

Definition 1.4.1. A continuously differentiable (C1) function


defined on an open subset U of

f:

is said to be holomorphic if

of=
Oz

at every point of U.

Remark: Some books use the word "analytic" instead of "holomorphic."


Still others say "differentiable" or "complex differentiable" instead of "holo
morphic." The use of "analytic" derives from the fact that a holomorphic
function has a local power series expansion about each point of its domain.
The use of "differentiable" derives from properties related to the Cauchy
Riemann equations and conformality. These pieces of terminology, and their
significance, will all be sorted out as the book develops.
If

f is

any complex-valued function, then we may write

u +iv,

where

and v are real-valued functions. For example,

z2 = (x2 - y2) + i(2xy);


in this example

u = x2 -y2 and v= 2xy.

The following lemma reformulates

Definition 1.4.1 in terms of the real and imaginary parts of

f:

f U C defined
on an open subset U ofC is holomorphic if, writing f(z) = u(x, y) +iv(x, y),
with z = x + iy and real-valued functions u and v, we have that u and v

Lemma 1.4.2. A continuously differentiable function

satisfy the equations

au

av

ax

8y

au

and

8y

av

ax

at every point of U.

Proof. The assertion follows immediately from the definition of holomor

phic function and the formula

8f ! 8u
=
2

Oz

The equations

au

ax

av
8y

av

and

!_
2

8v
ax

au

8u
8y

av

ax
8y
ax
{)y
are called the Cauchy-Riemann equations. The proof of the following easy
result is left as an exercise for you:

1.4. Holomorphic and Harmonic Functions

Proposition 1.4.3. If

f : U

Riemann equations, then

--..

C is

15

C1 and if f satisfies the Cauchy

of
of
. of
i
=
=
8y
{)z - ox -

on U.
The Cauchy-Riemann equations suggest a further line of investigation which
is of considerable importance. Namely, suppose that

u v
and

are

C2

func

tions which satisfy the Cauchy-Riemann equations. Then

a (au) = a (av)

ox
and

ox

ox

{)y

a (au) = a (av) .
8y

{)y

- {)y

ox

Exploiting the standard theorem on the equality of mixed partial derivatives


(that

a2v/axay =a2v/oyox),

we obtain

82u 82u
8x2 + {)y2 =0.

A similar calculation shows that

82v
82v
+
8x2 {)y2 =0.
You should check this last equation as an exercise.
These observations motivate a definition:

Definition 1.4.4. If
harmonic if

C is open and

The operator

C2(U),

then

a2 a 2
8x2 {)y2

is called

+-

is called the Laplace operator, or Laplacian, and is denoted by the symbol

6.. We write

u C 2,
(
(
:z ! u :x - :y ) :x :y ) u =6.u.

Notice that if

then

+i

1.

16

Fundamental Concepts

Similarly,

4u = 4
+ i
Oz az
2
ax
ay

(ax

i u
ay

6.u.

We have seen that the real and imaginary parts of a holomorphic function
are harmonic. It is natural to ask whether the converse is true. That is, if

u E C2(U)
U

is harmonic and real-valued, does there exist a holomorphic f on

u? It is possible to
when u is a polynomial.

such that Ref

answer is "yes"

give an elementary proof that the

This we now do. The general case

will be discussed in the next section.

Lemma 1.4.5. If u(x, y) is a real-valued polynomial with 6.u


there exists a (holomorphic) polynomial Q(z) such that Re Q u.

0,

then

Proof. We begin by writing

u(x,y)

u(x, y)

in terms of

( z+z z-z
)
--,
2

P(z,z)

and

where Pis a polynomial function. The hypothesis

z:

e-m
'"""
L.,, aemz z ,

6.u

0 becomes

ae+ m
0
aze azmp =
whenever e 2'.
words,

aem # 0

and m 2'.

only if e

1.

P(z, z)

It follows that

0 or

0. Thus

'"""
e '"""
-m
aoo + L.,, aeoz + L.,, aomz
m2'.l
e::::1
aoo

aoo

and

aeo

0 for e, m 2'.

1.

In other

m
aoo + L aeoi+ L aomz .
m2'.l
e::::1

Since Pis real-valued, in other words P

Thus

aem

aoe

P,

we have

- '""" -e '""" - m
aoo + L.,, aeoz + L.,, aomz .
m2'.l
e::::1

for all e 2'. 1

(see Proposition

1.3.3).

In conclusion,

u(z)

and

P(z, z)

aoo + L aeoze + L aomzm


m2'.l
e::::1
Re

(aoo + 2

Re

(Q(z)),

L aeoi)
e::::1

is the holomorphic polynomial required.

1.5.

17

Real and Holomorphic Antiderivatives

1.5. Real and Holomorphic Antiderivatives


In this section we want to treat in greater generality the question of whether

u is the real part of a holomorphic function


F = u+iv, then the Cauchy-Riemann equations

a real-valued harmonic function

F.

Notice that if we write

say that

av
ax

au
ay'

av
ay
In short, once

u is given,

the existence of

v ( and

avI ax and avI ay are completely

then

These in turn determine

au
ax.

up to an additive constant.

hence of

F)

amounts to solving a familiar problem

of multivariable calculus: Given two functions

au/ax, respectively ) ,
avjay= g?

and

determined.

Thus determining

f and g ( in this case -au/ay


v such that av/ax= f and

can we find a function

A partial solution to this problem is given by the following theorem.


We shall see later that the practice, begun in this theorem, of restricting
consideration to functions defined on rectangles is not simply a convenience.
In fact, the next theorem would actually be false if we considered functions
defined on arbitrary open sets in C

Theorem 1.5.1. If J, g a re

R= {(x,y)

( see

Exercise

52).

C1 functions on the rectangle


E

IR2: Ix-al< 8,ly-bl< t:}

and if
af
ag
on R,
ay
ax
C2(R) such that

then there is a function h E

and

on R. If f and g are real-valued, then


Proof. For

(x,y)

(1.5.1.1)

we

may take h to

be rea l- va lued a lso.

E R, set

h(x,y) =

1x f(t,b)dt+1Y g(x,s)ds.

By the fundamental theorem of calculus,

ah
(x,y) = g(x,y).
ay

(1.5.1.2)

1.

18

This is half of our result. To calculate

oh/ox,

Fundamental Concepts

notice that, by the funda

mental theorem of calculus,

rx
f(t, b) dt = f(x, b).
ox la
0

Moreover, since

g is C 1 ,

(1.5.1.3)

the theorem on differentiation under the integral

sign (see Appendix A) guarantees that

[Y 0
[Y
g(x, s) ds
g(x, s) ds =
,
ox l b
l b ox
0

which by

(1.5.1.1)

[Y 0
f(x, s) ds
lb oy
= f(x, y) - f(x, b)
=

(by the fundamental theorem of calculus). Now

oh/ox= f.

we see that
and

g are.

Since

: = f EC

h E C2(R).

(1.5.1.4)
(1.5.1.2)-(1.5.1.4)

(R) ,

oh
= g EC 1 (R),
oy
It is clear from

give that

(1.5.1.2)

that

is real-valued if

f
D

It is worth noting that, while we constructed

h using integrals beginning

at

(a, b) (the coordinates of the center of the square), we could have used
any (ao, bo) E R as our base point. This changes h only by an additive
constant. Note also that Theorem 1.5.1 holds for R an open disc: The only
special property needed for the proof is that for some fixed point Po E R
and for any point Q E R the horizontal-vertical path from Po to Q lies in
R. This property holds for the disc if we choose Po to be the center.
R is an open rectangle (or open disc) and if u is a
real-valued harmonic function on R, then there is a holomorphic function F
on R such that Re F u.
Corollary 1.5.2. If

Proof. Notice that

satisfy

f= -

of
oy

og
ox

OU

g=
ox
on

R.

1.5. Real and Holomorphic Antiderivatives

[This is equivalent to the hypothesis that

19

6u = 0.]

1
C , the
C2 function v on n

Since f, g E

theorem now guarantees the existence of a real-valued


satisfying

av
ax
av
ay
But this says that

F(z)

!=-

au
g= ax

u(z) + iv(z)

satisfies the Cauchy-Riemann equa

tions on R, as desired.

We have thus given the desired characterization of harmonic functions as


real parts of holomorphic functions, at least when the functions are defined
on an open rectangle or open disc. You will want to check for yourself in
detail that Theorem 1.5.l and Corollary 1.5.2 do indeed hold when R is
replaced by an open disc (or by any convex open set).
Theorem 1.5.1 is essentially an antidifferentiation theorem.

Given our

new notation, it would be aesthetically pleasing if we could use it to solve

F is holomorphic, can we find


aHIaz = F? We formulate the affir

the following antidifferentiation problem: If


a holomorphic function

such that

mative answer as our next theorem:

Theorem 1.5.3. If U

C is either an open rectangle or an open disc and

if F is holomorphic on U, then there is a holomorphic function

on U such

that aH/az = Fon U.

Proof. Write

F(z) = u(z) + iv(z).

Notice that if we set f

=u

and

g = -v

function

h1 such

then, by the Cauchy-Riemann equations, we know that

a f ag
ay =ax
_

Hence Theorem 1.5.1 guarantees the existence of a real

C2

that

ah1
=
ax

=u '

ah1
v
ay = g=- .

J = v and g = u (again notice that


guarantee that a jIay = a9 Iax) to obtain

Next we apply Theorem 1.5.1 with


the Cauchy-Riemann equations
a real

C2

function

( 1.5.3. l)

such that

ah
2
g
= u.
=
ay

1. Fundamental Concepts

20
H(z)

But then

h1 (z)+ih2(z) is C2 and
ah2
ax

ah1
ay

- =v=--.

Therefore

H satisfies the

Also

Cauchy-Riemann equations, so

(
(

H is holomorphic.

a
1
a
.a
-H = - - - i - (h1 +1h.2)
2 ax
az
ay
ah2 ah1
ah1 ah2
=
+
+
2 ax
ay
2 ax
ay
1
i
= "2 ( u+ u ) + "2 (v+v) = F.
Theorem

1.5.3

)
D

will prove to be crucial in the succeeding sections, for it

is the key to the success of complex line integration.

Exercises

1. Write each of the following

x+iy:
1
(a) -i
4+i
(b)
6 - 3i

(. 1)
;

(c)
i 6
(d) (2i - 4)2
(e ) i 4n+3 , n

(f}

as

complex numbers in the standard form

(
v; )
(..;'2 )8
-

'

v'2

(g)

2+2i

2. Find the real and imaginary parts of

(a) (i+ 1)2 (i - 1 )


i+ 1
(b)
i- 12
i
( c ) i 3 - 4i+ 6

21

Exercises

z
(d) z2+ 1
z2

( e ) z-1
(f) z4 +2z+ 6

3. Find the modulus of

( a ) (2- i) 2 (4+ 6i)

3- i

(b) (6+2i)3
(c ) ( v'3+ i) . ( v'3-i)
+2
i-2

(d)
(e)

1)

(i+

(i+2)

(i+ 3)

4. Prove that, for any complex numbers

and

w,

lz+ wl2 lzl2+ lwl2+ 2Re (z w),


lz+ wl2+ lz-wl2 2lzl2+ 2lwl2.

z such that z2

5. Find all complex numbers

such that

w4

6. Prove that

l-

z-w 2
1- zw

p(z)
j ::=:; n.

7. Let
0 ::=:;

i. Find all complex numbers

l.
(1 -l zl2)(1- lwl2)
11 -zwl2

a o + a1z +

Prove that if

z w =I- 1.

n
+ anz have real coefficients:

zo

provided

satisfies

p(zo)

0, then also

aj E IR for

p(zo)

0. Give

a counterexample to this assertion in case not all of a0, a1, ... , an are
real.

8. A field

is said to be ordered if there is a distinguished subset

with the following properties:

(i)
(ii)

if a, b E P, then a+ b E P and a

P F

b E P;

if a E F, then precisely one of the following holds:

a EP

- a EP

or

or

0.

P IR is taken to
P C which
1 E P so that -1 P.

{x

be

O}.

makes C ordered.

Prove that there is no choice of

(Suggestion: First show that

i E P or -i E P.)

E IR :

x >

Verify that IR is ordered when

Then ask whether

9. Prove that the function

<P(z)
maps the set D

C: Im z

>

O}.

{z

EC:

lzl

<

l-z
1+ z
1} one-to-one

onto the set U

The map is called the Cayley transform.

{z

22

Exercises

10. Let U

{z EC:

Im z >

O}.

Let
QZ +{3

1/J(z) =

"(Z +8

where a, {3, "(, 8 are real numbers and a8-{3"f > 0. Prove that 1/J : U---+ U
is one-to-one and onto. Conversely, prove that if

u(z)
with

a, b, c , dEC and u:

az+b
cz+d

U---+ U one-to-one and onto, then

a, b, c, dare

real (after multiplying numerator and denominator by a constant) and

ad- be>
11. Let S

0.

{z EC: 1/2

<

lzl

12. Compute all fifth roots of

-1,

<

2}.

Let

(z)

1 +i, all cube


-J3/2+i/2.

z+ (1/z).

roots of

-i,

Compute (S).
all sixth roots of

and all square roots of

13. Convert each of the following complex numbers to polar form. Give

possible polar
( a ) J?,+i
(b) JJ- i
( c ) -6+6i
(d) 4 - 8 i
( e ) 2i
(f) -3i
(g) -1
(h) -8+'Tri

all

forms of each number.

14. Each of the following complex numbers is in polar form. Convert it to

the rectangular form

+iy :

3ei7r
4ei7r/4
4ei277r/4
6eil07r/3
ei7r/l2
(f) e-i37r/8
(g) 7e -i7r/6
(h) 4e-i207r/6

( a)
(b)
(c)
(d)
(e)

15. The Cauchy-Schwarz inequality can be proved by induction. Try it.


16. The Cauchy-Schwarz inequality is a consequence of Lagrange's identity:
n

'""" z w
J J
j=l

j=l

j=l

23

Exercises

Prove Lagrange's identity, and from this deduce the Cauchy-Schwarz


inequality.

17. Find necessary and sufficient conditions on

{Zj}, { Wj}

for equality to

hold in the Cauchy-Schwarz inequality. Give a geometric interpretation


of your answer.

[Hint:

See Exercise 16.]

18. Give necessary and sufficient conditions on

z,w

for equality to hold in

the triangle inequality. Give a geometric interpretation of your answer.

19. Fix

n a positive integer. Suppose that

satisfying

n
LZjWj
j= I

for all w1, ... ,Wn E C such that

z1,... , Zn

are complex numbers

L::j= 1 lwj 12

1. Prove that

1. Formulate and prove a converse assertion as well.

20. The identity

ei9

=cosO+isinO is called Euler's formula.

LJ=I lzj 12

Use it to prove

DeMoivre's formula:
(cos 0+isinOt = cos nO+isin nO.
Use this last identity to derive formulas for cos 0 /2 , sin 0 /2.

21. Find all roots of

z2+z+ 1= 0.

Write them in polar form.

22. Give geometric interpretations for addition and multiplication of com


plex numbers (in terms of vectors).

23. Prove that the function

24.
25.

lzl
is continuous in the sense that if Zj---+ z (where this is defined to mean
that lzj - zl---+ 0) then lzjl---+ lzl (i.e., lzjl - lzl---+ 0).
Prove: If z is a nonzero complex number and k is a positive integer
exceeding 1, then the sum of the kth roots of z is zero.
Define a relation on IR by YI ,...., Y , YI,Y E IR, if and only if Y1 -y =27rn
2
2
2
t--+

for some integer n. Prove:

(i) ,...., is actually an equivalence relation.


(ii) eiYi = eiY2 if and only if YI ,...., Y2

26. Write each of the following complex polynomials in real notation:

( a)
(b)
(c)
(d)
(e)

F(z,z)=z3 - z2z
F(z,z) (z+z)2
F(z,z)=z2z+z2z
F(z,z)=z3
F(z,z)=z4
=

27. Write each of the following polynomials as a polynomial in

( a ) F(x,y)= (x - y2)+ i(y2+x)

and z.

24

Exercises

{b) F(x, y)
(c) F(x, y)
{d) F(x, y)

(x2y - y2x)+i(2xy)
(x2 - y2)+i(2xy)
(x3 - 3xy2)+i(-3x2y+y3)

28. Compute each of the following der ivatives:

a
(4z2 - z3)
( a) az
a
(b) -(z-2+z2z-3)
Oz
83
(c) -- (3z2z4 - 2z3z+z 4 - z-5)
8x28y
as
6z2)
2 z
{d) az3az2 (z - z+4z -

29. Compute each of the following der ivatives:

( a)

:z (x2 - y)

a
2
{b) az (x+y )
84
(xy2)
(c)
8z8z3
a2
7z)
z2 z3
{d) azaz (z - z+
30. Let f : C --+ C be a polynomial.

Suppose further that

8f
az
and

8f
=

Oz
for all

31. Let F

z
:

C Prove that

C --+ C be

constant.

a polynomial. Suppose that

a2
F =0
8z2
for all

C Prove that

F(z, z)
for some polynomials

32. Suppose that F:

z G(z)+ H(z)

G, Hinz ( i.e.,

C--+ C is

they depend only on

a polynomial. Suppose further that

F2

= 0.

What can you say about F?

33. Prove that if

z).

a
8x

a
8y

L=a-+b-

Exercises

25

and

(with

M=c-+dox
oy

a, b, c, d E C)

and if

Lz = 1,
Mz=:O,

Lz= 0,

Mz=

1,

then it must be that

L= (:x -i :y ) = :z
and

M= +i =.
oy
oz
2 ox

and

g2

are the only ones possible.

C1 function on an open set U

C,

then prove that

In other words, the definitions of

34. If

f is a

gz

a-

oz1

O-

ozf.

35. Prove, using only algebraic methods, that


that depends only on

z.

p(z)= x

is not a polynomial

Prove a similar assertion for

36. Write

and

oz

q( z)=x2.

Oz

in polar coordinates.

37. Let h be the set of all harmonic functions.


operations

(+, -,

...;-

) is

h closed?

Under which arithmetic

38. Answer Exercise 37 with h replaced by 1t, the set of all holomorphic
functions.

39. Let F be a

real-valued holomorphic polynomial.

Prove that F is identi

cally constant.

40. Let F be a harmonic function. Prove that F is harmonic.


41. Let U

{ z I z - zo I
:

be an open set.

::;

r}

Let

zo

U and

>

0 and assume that

U. For j a positive integer compute

21f

__!__ f (z0 + rei9) j d()


271" lo
and
21f
.
1
(zo + rei9)1 d().
271" 0
Use these results to prove that if u is a harmonic polynomial on U,
21f
1 {
u(zo + rei9) d()=u(zo).
271" lo

then

Exercises

26

J are both holomorphic on a connected


prove that f is identically constant.
Prove that if f is holomorphic on U C, then
2
6(lfl2) 4

42. If
43.

and

1 1

44. Prove that if

is holomorphic on U C and

6(lflP)
45. Prove that if

P2lflP-2

open set UC C, then

I1

is nonvanishing, then

any p > 0.

is harmonic and real-valued on U C and if

is non

vanishing, then
any p 2: 1.
be a

C2,

Prove that

h(z)

46. Let

47. Prove that if

real-valued harmonic function on an open set U C.


=

is

ou/oy +i8u / 8x

C2,

is holomorphic on U.

holomorphic, and nonvanishing, then log lfl is

harmonic.

48. Give an explicit description of all harmonic polynomials of second de


gree. Can you do the same for the third degree?

49. Let

C1 functions and assume that fog is defined.


version ( s ) of the chain rule: With w
g(z)
a
of og
of a9
(f g
+

f and g

following

be

Prove the

aw . oz aw . oz '
of og
of a9
(f 0 g) aw 0z +aw 0z
0z

How do these formulas simplify if (i) f and g are both holomorphic, (ii)
only f is holomorphic, (iii) only g is holomorphic, (iv) J and g are both
holomorphic? [Suggestion: Write everything out in real and imaginary
oz
a

parts and use the real variable chain rule from calculus.]

50. Let F be holomorphic on a connected open set U C. Suppose that

G1, G2

are holomorphic on U and that

8G1

oz Prove that

51. Let

(v1, v2)

that

G1 - G2 =constant.

8G2

- oz .

be a pair of harmonic functions on a disc U C. Suppose

8v1

8v2
ax

oy
Prove that

(v1, v2)

harmonic function

and

8v1

8v2

ax + oy

0.

is the gradient ( i.e., the vector

(oh/ox, oh/oy))

of a

27

Exercises

52. The function


Prove that

J(z)

1/z

is holomorphic on U

{z EC: 1

<

does not have a holomorphic antiderivative on U.

izl

2}.
[Hint: If
<

there were an antiderivative, then its imaginary part would differ from

arg z by a constant.

53. Let U C be an open set. Suppose that U 2

{ eit

<

27r}.

Let

be a holomorphic function on U which has a holomorphic antiderivative.


Prove that

1
[Hint:

2rr

( eit )

eit

dt

0.

Use the chain rule and the fundamental theorem of calculus.

f be a holomorphic function on an open set U C and assume


f has a holomorphic antiderivative F. Does it follow that F has a
holomorphic antiderivative? [Hint: Consider the function 1/ z2 on the
domain C \ {O}. Refer to Exercise 52.]

54. Let

that

55. Let U1, U2 be open sets in C and assume that U1 n U2 is nonempty and
connected. Let

be holomorphic on U

antiderivative on U1 and
prove that

U1 UU2. If f has a holomorphic

has a holomorphic antiderivative on U2, then

f has a holomorphic antiderivative on

U. Show by example

that the hypothesis on the intersection of the two domains is necessary.

56. Let U1 U2 U3

C be connected open sets and define


00

Let

be a holomorphic function on U. Suppose that, for each

a holomorphic antiderivative on Uj . Prove then that


antiderivative on all of U.

57. Prove that there is a constant C, independent of


complex numbers and if

j,

flu

has

f has a holomorphic

n, such that if

{ Zj } are

Llzj11,
j=l

then there is a subcollection

{ ZJi, ... , Z j }
k

{ z1, ... , Z }
n

such that

L Zjm C.
m=l

Can you find the best constant C?

[Suggestion:

Look at the arguments

of the nonzero Zj. At least one third of these arguments lie within

of each other.

27r/3

28

Exercises

58. Let

z1, z2, ... be a countable set of distinct

complex numbers. If lzj-Zkl

is an integer for every j, k (the integer may depend on j and k), then

prove that the


*

{Zj}

lie on a single straight line.

if Re z, Im z are integers.
?
Approximately how many lattice points are contained in {z : lzl < R}

59. A complex number

is called a

lattice point

That is, can you obtain an estimate

asymptotically as
1.
im

[Hint:

--+

no. of lattice pts. in

for this number which is sharp

{z : lzl

<

F(R)
z

R}

= 1 ?

{z : lzl < R}, then the unit square


{z : I z I < R + J2/2}. So the number
in {z : lzl < R} does not exceed n(R + J2/2)2
R + J2/2}. Similar logic about areas can be used to
is in

is contained in

of lattice points

{z : I z I

F(R)

+oo in the following sense:

If a lattice point

centered at
area of

<

establish a lower bound.]

60. Express the Laplacian in polar coordinates.

Chapter 2

Complex Line Integrals

2.1. Real and Complex Line Integrals


In the previous chapter, we approached the question of finding a function
with given partial derivatives by integrating along
directions only.

vertical

The fact that the horizontal derivative is

vertical derivative is

{)I ay

horizontal
{)I ax and the

and

then made the computations in Section 1.5 obvi

ous. But the restriction to such integrals is geometrically unnatural. In this


section we are going to develop an integration process along more general
curves. It is in fact not a new method of integration at all but is the process
of line integration which you learned in calculus.

Our chief job here is to

make it rigorous and to introduce notation that is convenient for complex


analysis.
First, let us define the class of curves we shall consider. It is convenient to
think of a curve as a (continuous) function 'Y from a closed interval
into

object ;y

[a, b]

C Although it is frequently convenient to refer to the geometrical

{'Y (t)

: t

[a., b]},

most of our analysis will be done with the

function 'Y It is often useful to write

1( t)

h1 ( t), 'Y2 ( t))

depending on the context.


is called

simple closed if

or

1( t)

'Y1 (t) + h2 ( t),

closed if 1 ( a ) = 1(b). It
and 1(a) 1( b). Intuitively, a

The curve 'Y is called

1l[a,b)

is one-to-one

simple closed curve is a curve with no self-intersections, except of course for


the closing up at

a., t

b.

In order to work effectively with "f, we need to impose on it some differ


entiability properties. Since 'Y is defined on a

closed interval,

this requires a

new definition.

29

2. Complex Line Integrals

30

</J : [a,b]
JR is
</JE C1([a,b]), if

Definition 2.1.1. A function

tiable

(or

C1), and

we write

---+

(a) </J is continuous on [a,b];


(b) 'exists on (a,b);
( c ) 'has a continuous extension

called

continuously differen

[a,b].

to

In other words, we require that


lim '(t)

lim '(t)

and

t-+a+

t-.b-

both exist.
The motivation for the definition is that if

</J E C1([a,b])

and

</J

is real

valued, then

<P(b)- (a)

lim

(</J(b- t: ) - </J(a + t: ) )

lim

E-+O+

E-+0+
b

b-E

a+E

<P'(t) dt

1 <P'(t) dt.

So the fundamental theorem of calculus holds for

Definition 2.1.2. A curve


if both

-y1

[a,b], and

and

'Y2

'Y : [a,b]

are. The curve is

</JE C1([a,b]).

C is said to be continuous on [a,b]


continuously differentiable (or C1) on

---+

we write

'YE C1([a,b]),
if

'Y1,'Y2

are continuously differentiable on

we will write

We also write

-y'(t)

for

Definition 2.1.3. Let

1/;1(t) + i 1/;2(t).

d-y
dt
d-y / dt.
1/; : [a,b]

---+

d-y1
dt

[a,b].

Under these circumstances

. d-y2
.
i dt

C be continuous on

[a,b].

Write

1/;(t)

Then we define
b

1 1/;(t) dt 1 1/;1(t) dt i 1 1/;2(t) dt.


=

Now Definitions 2.1.2 and 2.1.3, together with the comment following
Definition 2.1.1, yield that if

'YE C1([a,b])
b

-y(b)- -y(a)

is complex-valued, then

1 -y'(t) dt.

2.1.

Real and Complex Line Integrals

31

We state now a result that is, in effect, the fundamental theorem of


calculus along curves. [This result is probably familiar to you, from real
variable calculus, in terms of the line integral of the gradient of a function
along a curve.]

Let U C be open and let


and f E C1( U) , then

Proposition 2.1.4.

curve. Iff:

U---+

f('Y(b)) - f('Y(a))
Proof. Since f o 'Y
chain rule. That is,

lb (
a

'Y

[a, b] ---+

{)f
d"(
d"(
('Y(t)) dt1 +
('Y(t)) dt2
{)

{)f
ox

U be

a C1

) dt.

(t)

C1([a, b]), the result follows from ( * ) above and the


D

The expression on the right of (t) is what we usually call the "line
integral of the gradient of f along 'Y" The proposition as stated is about
real-valued functions. But, by dealing with Ref and Imf separately, we can
extend the proposition to complex-valued f without change. Our main goal
in this section is to derive an analogue of (t) in which complex numbers,
particularly complex-valued functions, play a more vital role.
For motivational purposes, let us consider what happens to (t), thus
extended to complex-valued functions, when f
u + iv is holomorphic.
Then we have
=

b
l
= ('Y(t)). dd"ft1 (t) + i ('Y(t)). dd"fit (t)
d
d
+ ('Y(t)). 2 (t) + i ('Y(t)). 2 (t) dt.

f('Y(b)) - f('Y(a))
a

8u

av

OX

OX

Since, by the Cauchy-Riemann equations,


8u
ox

8v

{)y

the integrand may be rewritten

{)u
ox

and

([

8v

8y

ox ,

as

d
('Y(t)). "f1 (t)
dt
+i

8u

8v
('Y(t)).
ox

d
('Y(t)). "f1 (t) +
ox
dt
av

{)u
ox

+i

8v
ox

[
)
I
')'(t)

d
"f2 (t)
dt

8u
ox

d
('Y(t)). "f2 (t)
dt

. d"f1 (t) + id"f2 (t)


dt
dt

( ** )

32

2. Complex Line Integrals

since
Thus

f is
( ** )

d-y
of
b(t) ) .
(t)
ox
dt
d-y
of
b(t) ) .
(t)
oz
dt
holomorphic. [ Note here

that we have used Proposition

1.4.3.]

becomes

r b of

f ( 'y ( b ) ) - f ( 'y ( a ) ) = la

d-y
( t ) dt,
dt

d-yifdt + id-y2fdt.
We have established formula ( *** ) only for holomorphic f; but we shall
want to use integrals like the one in ( *** ) for any f E C1 ( U ) . This consid

where, as earlier, we have taken

d-y/dt

oz

( 'y ( t ) ) .

to be by definition

eration motivates the following definition:

Definition 2.1.5. If U C is open,

-y : [a, b]

---+

: U

U is a C1 curve, then we define

1 F ( z ) d z= f
lr
la
The multiplicative

---+

C is continuous on U, and

the complex line integral

F ( 'y ( t ) )

d-y
dt.
dt

here is to be interpreted as multiplication of complex

numbers.
The payoff for this definition is the following elegant formula:

Proposition 2.1.6. Let U C be open and let

-y : [a, b]

---+

U be a C1

curve. If f is a holomorphic function on U, then

f('y( b ) ) - f ( 'y ( a ) ) =

J af
lr
( z ) dz.
oz

Proof. Rewrite ( *** ) using Definition 2.1.5.

This latter result plays much the same role for holomorphic functions as
does the fundamental theorem of calculus for functions from IR to R The
whole concept of complex line integral is central to our further considerations
in later sections.

We emphasize that Proposition

holomorphic functions.

[ Check

2.1.6

is valid only for

for yourself that it fails for

example. ]

F(z) =

z, for

We conclude this section with some easy but useful facts about integrals.

Proposition 2.1. 7. If>:

[a, b]

---+

C is continuous, then

2.1. Real and Complex Line Integrals

33

Proof. The point of the following proof is to reduce the inequality to the
analogous inequality for real functions that we know from advanced calculus.
Let

a=

l (t) dt.
b

a = 0, then there is nothing to prove. So assume that a i- 0 and set


TJ = a/lal. Let 'lj;(t) = Re(TJ (t)). Then 'lj;(t) ITJ(t)I= ITJl 1(t)I = 1(t)I.
If

Hence

ll b (t) dt l = lal = TJ a= TJ lb (t) dt = lb TJ(t) dt.


lb TJ(t) dt = lb Re(TJ(t)) dt = lb 7f;(t) dt lb 1(t)I dt.

But the far right side is real since the far left side is real. Hence

Proposition 2.1.8. Let

C1

curve, then

Uc:;:;

C be open and

C0(U).

I 1-rJ f(z) dz l (tE[a,b] IJ("/(t))I)


sup

where

f("t) =

If "I:

lb 1; (t) I dt.

(d-yifdt,d12/dt)

[a, b] --+ U is

f ("/)

Note that f("I) is the length of "f in the usual sense since
length of the real tangent vector

ld1 /dt(t)I is the


-y(t).

of-y at the point

Proof. We have

Ii

b
b
l
t
dt

t
l
l
(
(
))
(
(
))I
l l f "l ; l c b] f "l ) I; I dt

f(z) dz =

by Proposition 2.1.7.

The last result of this section tells us that the calculation of a complex
line integral is independent of the way in which we parametrize the path.
This fact proves very useful in calculations.
Proposition 2.1.9. Let

function. Let

Uc:;:;

"I : [a, b] --+ U

a one-to-one, onto, increasing

U--+

Ca continuous

curve. Suppose that :

is

C be an open set and F:

be a

C1

C1

function with a

C1

[c, d] --+ [a, b]

inverse. Let

;y = "Io.

Then

Proof. Exercise. Use the standard change of variable formula from calculus.
D

2. Complex Line Integrals

34

Proposition
function

2.1.9

f along

implies that one can use the idea of the integral of a

a curve 'Y when the curve 'Y is described geometrically but

without reference to a specific parametrization. For instance, "the integral


of z counterclockwise around the unit circle

{z E

C:

IzI

1}"

is now

a phrase that makes sense, even though we have not indicated a specific
parametrization of the unit circle. Note, however, that the direction counts:
The integral of z counterclockwise around the unit circle is 27ri. If the direc
tion is reversed, then the integral changes sign: The integral of z clockwise
around the unit circle is -27ri!

Geometric, parameter-less specification of

curves plays an important role in making what follows conveniently non


pedantic. The reader should endeavor to become accustomed to this style
as soon as possible.

2.2. Complex Differentiability and Conformality


The goal of our work so far has been to develop a complex differential and
integral calculus. We recall from ordinary calculus that when we differentiate
functions on lll2, we consider partial derivatives and directional derivatives,
as well as a total derivative. It is a very nice byproduct of the field structure
of C that we may now unify these ideas in the complex case.
F irst we need a suitable notion of limit. The definition is in complete
analogy with the usual definition in calculus:
Let U Cbe open,

PE

U, and

g:

g(z)

f,

\ {P}

- Ca function. We say

that
lim

Z->P

fE

C,

if for any E > 0 there is a '5 > 0 such that when


then

lg(z) - fl

zE

U and 0

<

lz - Pl

<

'5,

< E.

f is a complex-valued function on an open set U


that f is continuous at P if limz_,p f(z)
J(P).

In a similar fashion, if
and

PE

U , then we say

Now let

f be a function on the open set

U in Cand consider, in analogy

with one variable calculus, the difference quotient

f(z) - f(zo)
z - zo
for

zo -=/= zE

U. In case
lim

Z->ZQ

complex derivative by
then certainly

f is

Z - ZQ

f has a complex derivative at zo. We denote the


J'(zo). Observe that if f has a complex derivative at
continuous at zo.

exists, then we say that

zo,

f(z) - J(zo)

2.2. Complex Differentiability and Conformality

35

Figure 2.1

The classical method of studying complex function theory is by means


of the complex derivative.

We take this opportunity to tie up the (well

motivated) classical viewpoint with our present one.

Theorem 2.2.1. Let U


U. Then

f' exists

C be an open set and let f be holomorphic on

at each point of U and

J'(z)
for all

zE

Proof. Fix

8f
az

(where 8f /8z is denned as in Section 1.3).

zo E

U. If

z is near to zo
-y(t)

(1

(see Figure 2.1), then we may define

- t) zo + tz

and 'Y: [0,1]---+ U.


By Proposition 2.1.6,

f(z) - f(zo)
=

Therefore

f(z) - f(zo)
z - zo

f('Y(l)) - f('Y(O))
J 8f dz
1-r az
f 1 8f ( (t)) d-y dt
lo az 'Y . dt
f l 8f ( (t)) . (z - zo) dt.
lo az 'Y

2. Complex Line Integrals

36

af
{ 1 af
('y(t)) - az (zo) dt
(zo) dt +l
lo az
o az
1
af
af
r af
(z ) dt.
('y(t))(zo)
az o
az
+lo az

r1 af

But

b(t) - zol

tlz - zol

lz - zol

all

[O, l].

Let E > 0. If 8 > 0 is so small that the continuous function

af
l -(
az

whenever l w

when

- zol

lz- zol

<

<

a1
)- (zo)
az

8, then, in particular,

l a8zf ('y(t))- a8z1 (zo)

satisfies

< E

< E

8. Hence, by Proposition 2.1.7,

111 [ ('y(t))- (zo)] d l


11 l ('y(t))- (zo) I dt
lol Edt
t

<

<

It follows that

1.

E.

f(z) - f(zo)
Z- zo

1m -----

z--+zo

exists and equals

(8f /8z)(zo).

Notice that, as a result of Theorem 2.2.1, we can


J' for

af I az

when

( and

often will ) write

is holomorphic. The following result is a converse of

Theorem 2.2.1.

Theorem 2.2.2. If f
of U, then

C1 (U)

and

f has a complex derivative at each point

is lwlomorphic on U. In particular, if a continuous, complex

valued function

on U has a compl ex derivative at each point and if J' is

continuous on U, then

is holomorphic on U .

Proof. We need only check the Cauchy-Riemann equations. This is done by

letting z approach

z0

in two different ways ( first in the horizontal direction,

and then in the vertical direction ) as follows:

1.

Im

z--+zo

f(z)- f(zo)
----Z - Zo

2.2. Complex Differentiability and Conformality

37

f(zo+h) - f(zo)

lim
h--+O , h

real

u(xo+ h, Yo)+iv(xo+ h, Yo) - u(xo, Yo) - iv(xo, Yo)

1.

Im

h--+O

--

1.

---

u(xo+ h, Yo) - u(xo, Yo)

Im

--

h--+O

1.

v(xo+ h, Yo) - v(xo, y0)

+i

+z Im

h--+O

(xo,yo)

----

v
X

(xo,Yo)

On the other hand,

1.
Im
z--+zo

f(z) - f(zo)
Z - zo

f(zo+ih) - f(zo)
ih

lim
h--+O , h real

-----

lim
h--+O

u(xo, Yo+ h) - u(xo, y0)

1.
1 iv(xo, Yo+ h) - iv(xo, Yo)
+Im h
h--+O i
av
au.
+( -i.)
.
ay (xo,Yo) ay (xo,Yo)
-

Comparing the two expressions for limz--+zo U (z)

(-i)

au
av
+
ay (xo,yo) ay (xo,yo)

au
ax (xo,yo)

av
,
ay (xo,yo)

- f(zo))/(z - zo)

av
au
+i
ax (xo,yo)
ax (xo,yo)

Equating real and imaginary parts yields

au
a
Y (xo,Yo)

=
-

av

a
X

gives

(xo,yo)

These are the Cauchy-Riemann equations, as required.


The second statement in the theorem also follows since the continuity of

J'

plus the fact that

!'

au
av
+i
ax
ax

av
au
-i
ay
ay

imply the continuity of the real first partial derivatives of

f'

exists everywhere and is continuous, then

since the Cauchy-Riemann equations hold.


It is perfectly logical to consider an
ative at each point of

U without the

fE

I
C (U)

u and v.

Thus if

and is holomorphic

which possesses a complex deriv

additional assumption that

f E C1 (U).

2. Complex Line Integrals

38

Under these circumstances

and

still satisfy the Cauchy-Riemann equa

tions. To see this, note that we did not use the continuity of the derivative
in the proof of Theorem 2.2.2.

Historical Perspective:
1

text of C

Our study of holomorphic functions is in the con

functions. For such functions, we have two equivalent criteria

for the property of being holomorphic: one involves the Cauchy-Riemann


equations, the other the existence of the "complex derivative", that is, the
existence of

f(z + h)
h

.
11m

IC3h_,O

f(z)

Theorem 3.1.1 will tell us that such a function, which starts out being only
C1 in the definition, is in fact necessarily C00.
It was a matter of great interest classically to determine whether the
hy pothesis that the functions be C

could be weakened. In fact, E. Goursat

proved that it could: a function with a complex derivative at each point is


holomorphic.
It is alway s important in principle to know the optimal circumstance
under which a theorem holds, but in practice these weaker hy potheses never
come up as such. We have therefore in this text relegated the Goursat result
to an historical role.

For completeness, the proof of Goursat 's celebrated

result will be outlined later in Appendix B. The reader may safely omit
reading Appendix B with no loss of continuity. In any case, it should be
deferred until after Chapter 3, since it depends on ideas introduced there.
We conclude this section by making some remarks about "conformal
ity." Stated loosely, a function is conforma l at a point
"preserves angles" at

PE C

if the function

and "stretches equally in all directions" at

P.

Holo

morphic functions enjoy both properties:

Theorem 2.2.3. Let f be holomorphic in


w1, w2 be complex numbers of unit modulus.

a neighborhood of

P E C.

Let

Consider the directional deriva

tives

DwJ(P) =

f(P + tw1)
t->0
t

Dw2f(P) =

J(P + tw2)
t->0
t

and

lim

lim

f(P)

J(P)

Then

(2.2.3.1}: IDwJ(P)I
(2.2.3.2}: if IJ'(P)I f:.

IDw2f(P)I;

0, then the directed angle from w1 to w2 equals

the directed angle from

DwJ(P) to Dw2f(P).

2.2.

Complex Differentiability and Conformality

39

Proof. Notice that

f(P + twj) - f(P) . twj


t
twj
J'(P) Wj, j 1,2.

lim

t-+O

The first assertion is now immediate and the second follows from the usual
geometric interpretation of multiplication by a nonzero complex number,

rei9, r f. 0, multiplies lengths by r and rotates


angle ().
D

namely, that multiplication by

( around

the origin ) by the

Exercise

12

2.2.3, which in
(2.2.3.2) holds at P,

asks you to prove a converse to Theorem

either of statements (2.2.3.1) or


f has a complex derivative at P. Thus a C1 function that is conformal
( in either sense ) at all points of an open set U must possess the complex
derivative at each point of U. By Theorem 2.2.2, f is holomorphic if it is

effect asserts that if


then

ci.
It is worthwhile to consider Theorem

2.2.3

expressed in terms of real

f u+i v , where u, v are real-valued functions.


f(x + iy), and hence u and v, as functions of the real
variables x and y. Thus f, as a function from an open subset of C into C,
can be regarded as a function from an open subset of R2 into IR2 With
f viewed in these real-variable terms, the first derivative behavior of f is
functions. That is, we write

Also we consider

described by its Jacobian matrix:

( )

Recall that this matrix, evaluated at a point


linear transformation that best approximates
Now the Cauchy-Riemann equations for

(x0, Yo) , is the matrix of the


f(x, y) - f(xo, Yo) at (xo, Yo).

mean exactly that this matrix

has the form

Such a matrix is either the zero matrix or it can be written as the product

( v'a2+b2

of two matrices:

v'a2+b2

) (
.

cos()

- sin()

sin()

cos()

for some choice of () E R One chooses () so that


cos()
sin()

a
v'a2+b2'
b

40

2. Complex Line Integrals

Such a choice of 0 is possible because

2+ b 2 1.
(Ja2+2b ) (Ja2+2b ) =
a

Thus the Cauchy-Riemann equations imply that the (real) Jacobian of f


has the form

( ). (:; -::)
for some >.

IR, >.

>

0, and some 0

IR.

Geometrically, these two matrices have simple meanings. The matrix

cos e

sinO

sin e
cosO

is the representation of a rotation around the origin by the angle 0. The


matrix

( )
IR.2
( ). (:; -::)
IR.2

is multiplication of all vectors in

by >.. Therefore the product

represents the same operation on


complex number >.eiO.

as does multiplication on C by the

Notice that, for our particular (Jacobian) matrix

we have

( )2

2+
=
>. (:) () =IJ'(z)I,
in agreement with Theorem 2.2.3.

2.3. Antiderivatives Revisited


It is our goal in this section to extend Theorems 1.5. 1 and 1.5.3 to the
situation where f and g (for Theorem

1.5.1)

and

(for Theorem

.5.3)

have isolated singularities. These rather technical results will be needed for
our derivation of what is known as the Cauchy integral formula in the next
section. In particular, we shall want to study the complex line integral of

F(z) - F(zo)
z - zo

2.3. Antiderivatives Revisited

41

()(

Figure 2.2

when

F is holomorphic on U and z0 E U is fixed. Such a function is certainly


C1 on U\ { zo}. But it is a priori known only to be continuous on the entire set
U ( if it is defined to equal F'( zo) at zo). We need to deal with this situation,

and doing so is the motivation for the rather technical refinements of this
section.
We begin with a lemma about functions on R

Lemma 2.3.1. Let (n, [3) <;;;;IR be an open interval and let H: (n, [3) --+IR,
F : (n, [3) --+ IR be continuous functions. Let p E (n, [3) and suppose that

dH/dx exists and equals F(x) for all xE (n,{3) \ {p}. See Figure 2.2.
(dH/dx)(p) exists and (dH/dx)(x) F(x) for all xE (n,{3).

Then

Proof. It is enough to prove the result on a compact subinterval

(n, /1)

that contains

[a, b]

of

in its interior. Set

K(x)

H(a) +

F(t)dt.

K'(x) exists on all of [a, b] and K'(x) F(x) on both [a, p) and (p, b].
K and H differ by constants on each of these half open intervals.
Since both functions are continuous on all of [a, b], it follows that K - H is
constant on all of [a, b]. Since (K - H)(a)
0, it follows that K = H.
D

Then

Thus

Theorem 2.3.2. Let


and let

PE U.

Let

<;;;; C be either an open rectangle or an open disc

f and g be continuous,

real-valued functions on

are continuously differentiable


hypothesis is made

on U \ {P} (note that no


at the point P). Suppose further that

af
ay

Then there exists a C1 function

at

every point

of

U (including

h: U

--+IR such that

'

the point

ah
ay

-=

P).

Proof. As in the proof of Theorem 1.5 .1, we fix a point

U and

(ao, bo)

define

h(x, y)

iao f(t, bo) dt


x

which

ag
on U \ {P}.
ax

ah
ax

differentiability

[Y g (x, s) ds.
}bo

ao + iboE

42

2. Complex Line Integrals

------------------,

__________________

F igure

2.3

Then the fundamental theorem of calculus tells us that

ah
= (x, y)
ay g
.
The difficulty, just as in the proof of Theorem
Certainly the arguments of Theorem

1.5.1,

1.5.1 apply

is with

verbatim in the present

situation provided that the vertical line segment from


not contain P (Figure
For such

(x, y)

ah/ax.

(x, bo)

to

(x, y)

does

2.3).

we obtain (as in Theorem

1.5.1)

[Y !

! fY

(x, s)ds =
g(x, s)ds
uX }bo g
}bo uX

[Y f(x, s)ds

}bo us

f(x, y) - f(x, bo).


Furthermore,

a
f(t, bo) dt = f(x, bo)
uX ao

for

any (x, y).

1x

So

a
h(x, y) = f(x, y)
ax
provided that the situation is as in Figure 2.3.
Now for the general case.
continuous,

F(x) = f(x, y)

For

fixed, the function

is continuous, and

d
H(x) = F(x),
dx

H(x) = h(x, y)

is

2.4. Cauchy's Theorem and Formula

except possibly at

43

P1 (the first coordinate of P). By Lemma 2.3.1,

d
H(x) = F(x)
dx
for all

x;

hence

ah
=

ax

f
D

as desired.

As a corollary to Theorem 2.3.2 we immediately obtain an analogue to


Theorem 1.5.3.

C be either an open rectangle or an open disc.


Let P E U be fixed. Suppose that F is continuous on U and holomorphic
on U\ {P}. Then there is a holomorphic Hon U such that a H/ a z = F.

Theorem 2.3.3. Let U

Proof. The proof is a verbatim transcription of the proof of Theorem 1.5.3


except that we use Theorem 2.3.2 instead of Theorem 1.5.1.

Theorem 2.3.3 is the main and indeed the only point of this section. We
shall apply it in the next section as follows:
on

f will be a holomorphic function

U, and z EU will be fixed. We shall consider

F(()
Such an

f(() - f(z)
(- z
f'(z)

is, of course, continuous on

for

(EU\ {z}

for

z.

U.

We leave it as an exercise for the reader to check-before proceeding to


the next section-that analogues of Lemma 2.3.1 and Theorems 2.3.2 and

2.3.3 hold for the case of

finitely many

"singular points".

2.4. The Cauchy Integral Formula


and the Cauchy Integral Theorem
In this section we shall study two results which are fundamental to the
development of the rest of this subject: the Cauchy Integral Formula and
the Cauchy integral theorem. The Cauchy integral formula gives the value
of a holomorphic function at each point inside a circle in terms of the values
of the function

on

the circle (later we shall learn that "circle" here may

be replaced by a more general closed curve, subject to certain topological


restrictions). We proceed now with our study of the Cauchy integral formula.
The Cauchy integral formula is so startling that one may at first be
puzzled by it and find it unmotivated. However, as the theory develops, one
can look at it in retrospect and see how natural it is. In any case it makes

2. Complex Line Integrals

44

possible the extremely rapid development of a very powerful theory, and a


certain patience with the abruptness of its proof will be richly rewarded.

P is a
r a positive real number, then we let D(P,r)= { z E
D(P,r)= {z EC: lz - Pl-::; r}. These sets are called,

It will be useful to have at our disposal the following notation: If


complex number, and

C:

lz -Pl< r} and

respectively, open and closed discs in the complex plane. The boundary of

aD(P, r), is

the disc,

The boundary

the set of

aD(P,r)

simple closed curve "Y :

[O, 1]

such that

of the disc
--->

lz -Pl =r.

D(P,r)

can be parametrized as a

C by setting

1(t)= p + re2n-it.
We say that this "Y is the boundary of a disc with

counterclockwise

orienta

tion. The terminology is justified by the fact that, in the usual picture of

C, this curve "Y runs counterclockwise. Similarly, the curve

er : [O, 1]

--->

defined by

er(t)=P + re-2n-it
D(P,r) with clockwise orientation. Recall
aD(P,r) do not depend

is called the boundary of the disc


that,

as

already explained, line integrals around

on exactly which parametrization is chosen, but they do depend on the


direction or orientation of the curve. Integration along

er

gives the negative

of integration along "Y. Thus we can think of integration counterclockwise


around

aD(P,r) as

a well-defined integration process-that is, independent

of the choice of parametrization-provided that it goes around in the same


direction as our specific parametrization "Y.
We now turn to a fundamental lemma:

Lemma 2.4.1. Let "Y be the boundary of a disc


plane, equipped with counterclockwise
the circle

aD(zo,r).

Then

27ri

Proof. To evaluate

"Y

D(zo, r) in the complex


orientation. Let z be a point inside

1
--

(-z

d(=l.

i ( z d(,

we could proceed by direct computation.

However, this leads to a rather

messy (though calculable) integral. Instead, we reason as follows:


Consider the function

l
I(z)= j __ d(,
f-r (-z
defined for all

with

lz - zol < r.

We shall establish two facts:

2.4. Ca u chy's Theorem and Formula

(i) f(z)

is independent of

(ii) f(zo)

z;

27ri.

Then we shall have


To establish

45

f(z)

27ri for all

with

lz -zol

r, as required.

<

note that (by the theorem on differentiation under the

(i),

integral sign in Appendix A ) we can differentiate the formula for f(z) to see
that

:zf(z) i :z ( ( z ) d( i
=

Also

O d(

0.

:zf(z) i :z ( ( z ) d( i ( ( z)2 d(.


=

Now the function

1
( f---+ (( - z)2

is holomorphic on C

\ {z}. Moreover this

function

Hence, by Proposition

o
-f(z)
oz
since

1(0)

1( 1).

function is the ( derivative of the

-1
( f---+ (-z
-- .

2.1.6,

1
d(
"Y((-z)2

-1
1( 1)-z

-1
1(0)-z

[This looks like a trick because we keep changing which

variable, (or z, is actually being varied, but some thought will convince you

that each step is legitimate.]

D(z0, r) , the function f(z) is holomor


of /oz = 0 on D(zo, r). By calculus, the
(i).

So now we have shown that, on


phic; moreover
function

1:

and

is constant, proving

To prove
[O, 27r]

of /oz

---+

(ii) is
1(t)

C,

an easy direct computation:


=

rcost+ irsint+

zo.

Then

We write the circle as

1' (t)

-rsint+ ircost

and

f(zo)

J__
l d(

f_y (-zo

102

71'

-rsint + ircost
dt
. .
rcost + i rsmt

{ 271'
i dt
lo

This completes the proof.

27ri.

2. Complex Line Integrals

46

Theorem 2.4.2 The Cauchy integral formula . Suppose that U is

an

open

f is a holomorphic function on U. Let zo E U and let


D(z0, r) U. Let / : [O, 1] C be the C1 cur ve
zo+ r cos(27rt) + ir sin (27rt) . Then, for each z E D(zo, r),
J f(() d(.
f(z)
27ri !, ( -z

set in C and that

r
1

0 be such that

>

(t)

Before proving the theorem, we remark briefly on some of its striking


features. Many more such features will appear later.
First note that

f(z)

f on the
zo. (The curve / has

is entirely determined by the values of

curve /, or, equivalently, the circle of radius


image precisely equal to this circle.

about

Of course, nothing like this kind of

unique determination works in general for an arbitrary

real variables. For example, the function [1

C1 function of two

- (x2 + y2W

the circle of radius 1 about 0 E C, any positive integer

is identically 0 on

k,

but is not zero

inside the circle. This latter function in complex notation is [1

-zz]k, which

is of course not holomorphic since

az[l -zz]k

k[l -zz]k-1(-z)

o.

We shall see later that the converse of Theorem 2.4.2 also holds: if
by the Cauchy integral formula, then

f is holomorphic.

f is given

A second important property of the formula is the extremely simple

z occurs in the integral (the right-hand side of the formula ) .


z in this way suggests the possibility of differentiating
under the integral, since obviously f(( ) /(( - z) is a holomorphic function of
z, z E D(zo, r), for each fixed ( E 8D(zo, r). This matter will be considered
way in which

The occurrence of

in detail in Section 3.1.


Now the proof:

D(zo, r+E) U. Fix a point z E D(zo, r+E).


H: D(zo, r + E)
C such that
f(( = (z) if f= z
8H
(
a(
f'(z)
if (
z.
Since 1 ( 0 )
1 ( 1 ) , Proposition 2.1.6 ( with ( in place of z) shows that
f(( = (z) d(.
H'(() d(
0
H(1 ( l )) - H(r(O))
Proof. Choose

>

0 such that

Then, by Theorem 2.3.3, there is a holomorphic function


=

Thus

J f(z)
J
!, ( -z d( !,(J(()
-z d(.
=

2.4. Cauchy's Theorem and Formula

Since

47

f(z)
J
d( = f(z) J _I_ d(,
f-r(-z
f-r(-z

it remains only to see that

i ( z d(=

But that is Lemma 2.4.1. Now formula

(*)

2tri.
gives the desired conclusion.

In the course of the proof of Theorem 2.4.2 there occurred


the following fact, which is known

as

( implicitly )

the Cauchy integral theorem. This is

the second main result of the present section:

Theorem 2.4.3 ( The Cauchy integral theorem ) . If f is a ho lomo rphic func


tion on an open disc U in the complex plane, and if 'Y :
curve in U with

1(a)

1(b),

[a, b]

i f(z) dz=

on U. Since

1(a) = 1(b),
0

U is a C1

0.

Proof. By Theorem 1.5.3, there is a holomorphic function

G' = f

--+

then

--+

C with

we have that

= G('Y(b) ) -G('Y(a) ) .

By Proposition 2.1.6, this equals

i G'(z) dz= i f(z)dz.

Check for yourself that the proof of the Cauchy integral theorem is valid
if the open disc is replaced by an open rectangle.
We shall be using complex line integrals so much in what follows that
it will often be convenient to write them in a simplified notation. We shall
write expressions like

f(z) dz.
J
fov(a,r)
This is to be interpreted

as

follows: the integration is to be formed over

a C1 simple, closed path 'Y whose image in the complex plane is


the circle bounding

D(a, r ) ,

with counterclockwise orientation.

8D(a, r ) ,

[For a region

like a rectangle we would like to use a union of C1 curves-or what is called


a piecewise C1 curve. Such curves will be defined precisely in Section

2.6.]

The various possibilities for choosing such a parametrization 'Y all yield the
same value for the integral

( Proposition

2.1.9 ) .

In particular, the "default" orientation for a curve is always the coun

terclockwise orientation; this orientation is sometimes called the "positive"


orientation.

2. Complex Line Integrals

48

All of the known proofs of the Cauchy integral formula are rather tech
nical in nature.

This is of necessity, because there are certain topological

strictures that must be addressed. Since the Cauchy integral formula and
Cauchy integral theorem form the bedrock of the subject of complex anal
y sis, we have given a complete and discursive treatment of these theorems.
W hile each step in the proofs of these theorems is quite transparent, the
aggregate picture may be somewhat obscure. For this reason we shall now
give a more "quick and dirty" treatment of these topics. This new "proof"
is only correct when the curves being treated are very nice, and when no
topological issues intervene; we note that making it precise is difficult. But
it provides a way to connect the Cauchy integral theorem rather directly to
real-variable calculus. You should not, however, regard it as a substitute for
the rigorous proof given earlier.

1 : [O, 1] C is a simple closed, continuously


1(0) = 1(1), and the curve does not intersect

Let us then suppose that


differentiable curve.

Thus

---->

itself elsewhere. Assume the Jordan curve theorem; that is, assume that C \

1( [ 0, 1]) consists of two open,

connected sets with

1( [0, 1]) as their common

boundary. Let U denote the (bounded) open region that is the interior of

Thus, abusing notation slightly,

1=

8U.

Our basic tool will be one of the multivariable versions of the fundamen

tal theorem of calculus (for which see [ THO] ). This is a "standard" result,
though it is seldom established rigorously in texts. See also Appendix A.

[ Green's

THEOREM

P and
on U

theorem] .

Let U,

be

as

above.

If

are functions that are continuously differentiable

1( [ 0, 1]) U

U (i.e., on the curve together with its

interior), then

1 p dx - Q dy=

fr

Jrlur oQox - oy dxdy.


{)

Let us apply Green's theorem to derive immediately a version of the


Cauchy integral theorem.

Let

be a function that is holomorphic in a

neighborhood of U. We would obtain a version of the Cauchy integral the


orem if we could prove that

i f(z) dz=
Writing

f=

iv

i f(z) dz
= i(

0.

as usual, we calculate that

iv)(dx + idy)

2.4. Cauchy's Theorem and Formula

49

i u dx (-v) dy i iv dx u dy
(Gi;;en)
[a(-v) - au] dxdy i [au - av] dxdy.
ff ax ay
ff ax ay
=

(t)

Of course each of the integrands on the right is zero, by the Cauchy-Riemann


equations. That completes our "proof" of this version of the Cauchy integral
theorem, having duly noted the rather imprecise version of Green's theorem
that we have used here.
More interesting is the Cauchy integral
if

EU, then

f(z) =

formula.

We wish to show that,

J f(() d(.

2ni f-y (
z
-

Again we use Green's theorem, noting first that Green's theorem applies just
as well to a domain bounded by

finitely many continuously

differentiable,

simple closed curves 11, ... , /k (with the "inside" curves oriented clockwise).
Let

0 be so small that

>

\ D(z, E ) .

D(z, E )

u. Consider the region uf

The function

(
(() = f( )
(-z

is holomorphic on a neighborhood of UE. We apply Green's theorem to

fau, <P(() d(.

just because

The result (as in

(t))

is that

J <P ( d(=O,
( )
!au.
<P

(*)

is holomorphic.

But aUE has two components, namely the boundary of U (i.e., the curve

/)

and the boundary of the little disc. So this last equation just says that

1 (() d( = 1
(() d(.
!aD(z,E)

1-r

Let us do some calculations with the right-hand side of the last equation.
Now

1
(() d( = 1
!aD(z,E)
!aD(z,E)
J
laD(z,E)
=

Observe that

f(()
d(
(-Z
f(z) d( + J
f(()-f(z)
d(
(-z
!aD(z,E) (-z

I+ II.

f(()-f(z)
(-z

50

2. Complex Line Integrals

is bounded by some constant on

D(z, 1: )

since the function in the integrand

[!(() - f(z)]/(( - z) is continuous on U \ {z} and has a limit as ( z, ( E


\ {z}. Since the boundary curve 8D(z, 1: ) has length 27rE, the expression
II does not exceed 27rEC; hence II
0 as E 0. Thus the term of interest
--t

--t

--t

is I.
But
I= f(z)
by Lemma

2.4.l.

laD(z,l)

d(
-z

= f(z) 27ri

'>

Putting together our information, we find (letting

2m

--t

0)

that

J J(() d(

fr ( - z

= f(z).

This is the desired result.

2.5. The Cauchy Integral Formula: Some Examples


The Cauchy integral formula will play such an important role in our later
work that it is worthwhile to see just how it works out in some concrete
cases.

In particular, in this section we are going to show how it applies

to polynomials by doing some explicit calculations. These calculations will


involve integrating some infinite series term by term.

It is not very hard

to justify this process in detail in the cases we shall be discussing.

But

the justification will be presented in a more general context later, so for


the moment we shall treat the following calculations on just a formal basis,
without worrying very much about convergence questions.
To simplify the notation and to make everything specific, we shall con
sider only integration around the unit circle about the origin.

/:

[O, 27r]

--t

So define

C by 1(t) =cos t+ i sin t. Our first observations are the follow

ing:

i ( d(
(2) i c d(
k

(1)

=0 if k E Z, k

# -1;

=27ri.

Actually, we proved

(2)

by a calculation in the previous section, and we

shall not repeat it. But we repeated the conclusion here to contrast it with
formula

(1).

To prove
Then

(1),

set

fk(()

fk is holomorphic on

(1 + k ) -1 ( k+1, for k an integer unequal to -1.


C \ {0}. By Proposition 2.1.6,

0= fk(r(27r)) - fk(r(O)) =

18f d( frJ (
fr (
k

d(.

2.5. The Cauchy Integral Formula: Some Examples

51

This argument is similar to an argument used in Section 2.4. Of course,

(1)

can also be established by explicit computation:

i(

kd

2rr

If k +

fo (
fo (
2rr

cost+ isint k

- sint+ icost dt

()

cost+ isint k+ 1 dt.

is positive, then

( ) fo
*

2rr

[ cos

)]

k + l t+ isin k + l t dt

0.

Here we have used the well-known DeMoivre formula:

cost+ isint

cos nt + isin nt ,

which is easily proved by induction on n and the usual angle-addition for


mulas for sine and cosine
Notice that

since

Hence, if k +
2rr

fo (

(
<

Exercise 20 of Chapter

cost+ isint -1

cost+ isint

0, then

27r
}{0

cos2 t+ sin2 t

1.

cost - isint) ( k+ 1) dt

{ 27r [ cos
Jo

o.

cost - isint

cost - isint

cost+ isint k+l dt

1).

((

))

((

) )] dt

- k + l t - isin - k + l t

Note: We have incidentally shown here that the DeMoivre formula holds

also for negative n .

Now let us return to the Cauchy integral formula for a polynomial


N

P(z)

n=O

The formula gives

l l

for z <

P(z)
1,

2m

n
anz .

i (P(()
'Y

- z

with/ the parametrized boundary of the disc. Now

(-z

1
=

( 1 - z /(

52

2. Complex Line Int egrals

Since
than

1 for
ICI
1. Thus, by
=

( on 'Y and since

<

lzl

1,

the absolute value

+oo

1
1- z;c

lz/(I

is less

7),

the geometric series formula (Exercise

(z) k

L (

k= O

Hence

J P(() d(

P() ( m) d(

f-y (-z

(
+oo

J
h

P(() k
z
(k+l

d(.

Now, assuming (as we discussed earlier) that we can integrate term by term,
we have

By

(1)

and

(2),
if

0 k N

if

>

N.

Thus

i ---P(()
-=-; d(
'Y

27ri L akzk
k=O

27riP(z).

This directly verifies the Cauchy integral formula, by way of a calcula


tion, in the case the function in question is a holomorphic polynomial and

'Y is the unit circle. (The case of 'Y being an arbitrary circle is left to the
reader.)
The process just given does more than make the Cauchy integral formula
plausible.

It also suggests that there is a close relationship between the

Cauchy integral formula and functions that are sums of powers of

z,

even

possibly infinite sums. Formally, the same arguments would apply even if

P(z)

were given by a power series

L anzn
00

n=O
instead of

P(z)

being a finite sum (polynomial). We shall see later that this

formal relationship is more than accidental: Holomorphic functions will be


seen to be, in a very precise sense, those functions which can be represented
(at least locally) by power series.

Exact statements and proofs of these

assertions will be given in Section 3.3.

2.6. More General Curves

53

2.6. An Introduction to the Cauchy Integral Theorem and

the Cauchy Integral Formula for More General Curves

For many purposes, the Cauchy integral formula for integration around cir
cles and the Cauchy integral theorem for rectangular or disc regions are
adequate. But for some applications more generality is desirable. To for
mulate really general versions requires some considerable effort that would
lead us aside (at the present moment) from the main developments that we
want to pursue. We shall give such general results later, but for now we
instead present a procedure which suffices in any particular case (i.e., in
proofs or applications which one is likely to encounter) that may actually
arise. The procedure can be easily described in intuitive terms; and, in any
concrete case, it is simple to make the procedure into a rigorous proof. In
particular, Proposition 2.6.5 will be a precise result for the region between
two concentric circles; this result will be used heavily later on.
F irst, we want to extend slightly the class of curves over which we per
form line integration.

Definition 2.6.1. A piecewise C1 curve

'Y [a, b]

a < b, a, b E , is a
continuous function such that there exists a finite set of numbers a1 a2
ak satisfying a
a and ak b and with the property that for every
1
I
1 j k 1, 'Y [
l is a C1 curve. As before, 'Y is a piecewise C1 curve
a1 ,a1+1
(with image) in an open set U if 'Y ( [a, b]) U.

C,

Intuitively, a piecewise C1 curve is a finite number of C1 curves attached


together at their endpoints. The natural way to integrate over such a curve
is to add together the integrals over each C1 piece.

Definition 2.6.2. If U C is open and

'Y [a, b]

curve in U (with notation as in Definition

2.6.1)

U is a piecewise C1

and if

continuous, complex-valued function on U, then

C is a

i f(z) dz= Li
-Y

where

a1' a2, ... , ak

J=l

f(z) dz,

-Yl[aj.aJ+iJ

are as in the definition of "piecewise C1 curve."

In order for this definition to make sense, we need to know that the
sum on the right side does not depend on the choice of the

a1

or of k. This

independence does indeed hold, and Exercise 33 asks you to verify it.
It follows from the remarks in the preceding paragraph together with the
remarks in Section

2.1

about parametrizations of curves that the complex

line integral over a piecewise C1 curve has a value which is independent of


whatever parametrization for the curve is chosen, provided that the direction

2. Complex Line Integrals

54

of traversal is kept the same. This (rather redundant) remark cannot be


overemphasized. We leave as a problem for you (see Exercise 34) the proof
of the following technical formulation of this assertion:
Lemma 2.6.3. Let 1
Let
with

[a, b] ----+

U, U open in C,

: [c, d] ----+ [a, b] be a piecewise C1

(c)=a

and

Then the function

(d)=b. Let f:
[c, d] ----+ U

'Yo :

U----+ C
is

be a

piecewise C1 curve.

strictly monotone increasing function

be a

continuous function on U.

piecewise C1 curve and

J f(z) dz= J f(z) dz.

1'-r

1'-ro

The foregoing rather technical discussion now allows us to make, without


ambiguity, statements about such items as "the integral of f(z) counterclock
wise around the square with vertices 0, 1, 1 + i, and i." One now needs only
a piecewise C1 path, a direction to traverse it, and, of course, a function to
integrate along it.
Not surprisingly, we now have a generalization of Proposition 2.1.6 (the
fundamental theorem of calculus for complex line integrals) :
Lemma 2.6.4. If f: U----+ C is
is

a holomorphic function

f(I(b)) - f(I(a))=
Proof.

wise

and if/:

[a, b] ----+

piecewise C1 curve, then

C1

i J'(z) dz.

The result follows from Proposition 2.1.6 and the definition of piece
curve.
D

We are now ready to introduce the observation that is the basis for our
extensions of the Cauchy integral formula and the Cauchy integral theorem.
It is an absolutely central idea in this subject: Namely, the integral

i f(z) dz,

f holomorphic, does not change if a small (localized) change is made in the


curve/.

In particular, we consider the following setup. Let f : U ----+ C be a


holomorphic function on an open set U, and 'Y a piecewise C1 curve in
U. Suppose that is another piecewise C1 curve in U that coincides with
'Y except on an interval where both 'Y and have images which lie in a
disc contained in U. See Figure 2.4 (the portion of that differs from 'Y is
indicated with a dotted curve) .
Then we claim that

i f(z) dz=i f(z) dz.

2.6. More General Curves

55

Figure

2.4

To see this, recall that on the disc there is (by Theorem


phic function

2.4)

H'(z) f(z). So integrating from p to q (see Figure


H(q) - H(p), and so does integrating from p to q along

with

along yields

1 . 5. 3) a holomor

'Y The rest of the integration along 'Y and also gives equal results since the
two curves coincide there. So the entire integrals in formula ( * ) are equal.
This seemingly small observation opens up a large vista of techniques and
results. Let us consider a specific example:

Proposition 2.6.5. If

\ {O}

'Yr describes the circle of radius

--+

C is a holomorphic function, and if

around 0, traversed once around counter

clockwise, then, for any two positive numbers

/'r1

f(z)

dz= i

r1 < r2,

J(z) dz.

l'r2

Remark: Note that the statement of the proposition is not a trivial asser
tion: In general each integral is not zero-consider the example J (z)

1 / z.

Proof. To prove the proposition, we notice that "fr1 can be changed to "fr2
by a sequence of operations of the sort described just before Proposition

2.6.5. A

sequence of allowable deformations is illustrated in Figure

2.5.

You should think carefully about this figure to be sure that you grasp the
idea. Notice that each successive deformation takes place inside a disc, as
required, but that the entire deformation process could not be performed in
just one step inside a disc which misses the origin.

56

2. Complex Line Integrals

Figure 2.5
The same proof establishes the following result:

Proposition 2.6.6. Let 0

{z
r

<

r1

C : r
<

r2

<
<

<

lzl < R}. Let J


R and if for each j

<
:

< oo and define the annulus

A=

C be a holomorphic function.

-+

If

the curve rr1 describes the circle of radius

rj around 0, traversed once counterclockwise, then we have

J(z) dz=

rl

J(z) dz.

r2

Perhaps all this looks a bit strange at first.

Although our proofs of

Propositions 2.6.5 and 2.6.6 are thoroughly rigorous, they are specific for
circular paths. It turns out to be quite a bit of trouble to state and prove a
result for more general curves (even though the intuitive content is obvious).
We shall formulate a result on general curves later on in Chapter 11.
Let us try for now to put Propositions 2.6.5 and 2.6.6 into perspective by
describing the general situation in which this type of result works: Suppose
that f : u

-+

c is a holomorphic function and 'Y and are piecewise C1

closed curves in U. W hen can we be sure that

J(z) dz t J(z) dz
=

The answer is that we can be sure that equality holds if 'Y can be continuously
deformed to through other closed curves in U (remember that, at this
stage, this is an intuitive discussion!)-that is, if 'Y can be continuously

57

2.6. More General Curves

Impossible

Figure 2.6
moved around in U, staying closed, until it coincides with . Of course this
is not always possible. For instance, if U

= C \ {O},

then a circle around

the origin cannot be continuously deformed in U to a circle that does not

encircle the origin-see Figure 2.6.

Later we shall return to these matters in detail.

For the moment, we

conclude with an application of these methods to obtain an extension of the


Cauchy integral formula .

EXAMPLE 2.6. 7. Let

---4

C be holomorphic. Let/ be a piecewise C1,

simple closed curve in U and let

P be as in

Figure 2. 7. The key property we

require of the curve/ is that it can be continuously deformed, in U, to the


curve

8D(P, 1: )

for E very small. In applications, this property would need

to be verified explicitly for the particular curve 1: at this time we have no


general result that would guarantee this deformation property.
is always true if the "interior" of / is contained in U , and if

Though it
is in that

"interior," the proof of these assertions in general terms requires a great


deal of topological work. But, for a particular given/, it is usually easy to
verify what is required. Suppose for the moment that this verification has
actually been done.
Then

2m

J(() d
( = J(P).
(-P

The justification for this assertion is that the curve/ may be continuously
deformed, by a sequence of operations as in the discussion preceding Propo
sition 2.6.5, to the curve C =

{z

EC:

lz - Pl= E}.

But then we have

l
27ri

1 J(() d( =
(-z

l
27ri

1 f(() d(.
le ( - z

This last, by the Cauchy integral formula, equals

f(P).

58

2. Complex Line Integrals

Figure 2.7

ez

Figure 2.8

EXAMPLE 2.6.8. To calculate

_1 1 f(() d(
27ri h ( - z
for

holom orphic on U, where "'(, U are

"Y can be written

as

as

in Figure 2.8, we notice that

11 followed by 12 where "Yi and 12 are illustrated in

2.6. More General Curves

59

'Y2

Figure 2.9

Figure 2.9. Then we have

f(()
f(()
f(()

J
d( = J
d( + J
d( = O + f(z).
2m fr ( - z
27rt l-n ( - z
2m fr2 ( - z
These pictorial examples are for intuitive illustration only.

In actual

practice, one would need to have the curve/ given by a precise formula in
order to carry out the deformation process as described or indeed to calculate
the integrals at all.
We conclude this section with an intuitive description of general forms
of the Cauchy integral formula and Cauchy integral theorem:

THEOREM: Let
Then

f U--+
:

C be holomorphic and

i f(z) dz=

for each piecewise C1 closed curve 'Y in


deformed in

U open.

that can be

through closed curves to a closed curve

U.
D(z,r)

lying entirely in a disc contained in

In addition, suppose that

U. Then

J f(() d(= f(z)


2m fr ( - z
U \ {z} that can
U\ {z} to 8D(z,r) equipped

for any piecewise C1 closed curve 'Y in


be continuously deformed in

with counterclockwise orientation.

60

2. Complex Line Integrals

You are invited to convince yourself intuitively that these general forms
are in fact valid! Again, these discussions serve, for now, as an invitation to
the subject. A more rigorous treatment is given in Chapter

11.

Exercises
1. Let

(i.e.

'Y

be holomorphic on an open set U which is the interior of a disc

or a rectangle. Let
is a

'Y : [O, l] ......

closed curve).

U be a C1 curve satisfying

"f(O) = 'Y(l)

Prove that

i f(z) dz= 0.

2.

[Hint: Recall that f has a holomorphic antiderivative.]


Let f be holomorphic on an open set U which is the interior of a disc
or a rectangle. Let p, q E U. Let 'Yj : [a, b] ...... U, j = 1, 2, be C1 curves
such that 'Yj(a) = p, 'Yj(b) = q, j = 1,2. Show that
J f dz= J f dz.

lr1

lr2

3. Let U C be an open disc with center


If

zE

U, then define

'Yz

0.

Let

f be holomorphic

on U.

to be the path

'Yz(t) = tz , 0::; t::; 1.


Define

F(z) = J f(() d(.

Prove that

lrz

F is a holomorphic antiderivative

for

f.

4. Compute the following complex line integrals:

(a) J dz
f-r z

where

'Y

is the unit circle [center

(0, O)]

with counter

clockwise orientation;

(b)

iz

z2z dz

where

'Y

is the unit square [side 2, center

(0,O)]

with

clockwise orientation;

z
(c) J dz where 'Y is the triangle with vertices 1+Oi,0+i, 0 - i,
8
fr + z2
and 'Y is equipped with counterclockwise orientation;
(d) J dz
where 'Y is the rectangle with vertices 3 i with clock
f-r 8 + z
wise orientation.

5. Evaluate

'"Y zJ dz,

for every integer value of j, where

'Y

is a circle with

counterclockwise orientation and whose interior contains

0.

61

Exercises

6. If U i;;;;; IR2 is an open set, then a vector field on U is an ordered pair


of continuous functions a(z) = (u(z),v(z)) on U. We can think of a as
assigning to each z E U the vector (u(z),v(z)) (hence the name). We
along a C1 path 'Y :

define the line integral of Q

f, 1b (
a=

Here the

u("!(t)),v('Y(t))

[a, b]

---+

u by the formula

"(1(t) dt .

denotes the dot product of vectors.

Notice that this is a

direct generalization of Proposition 2.1.4, where a(z) was taken to be


gradf(z). You have seen integrals like (*) in your multivariable calculus
course.
Define
n1(z)= (x,y) ,
n (z) = (x2, y2) ,
2
a3(z) = (y2,x2) ,
a4(z) = (-x,-y) .
Let 'Yi(t) = 4e211"it,

"(2(t)

annulus. Calculate

= e-211"it, 0 t l. Note that "(1, "1

Qj +

'-r1

Qj,

bound an

= 1,2,3,4.

'-r2

Assume that each aj represents the velocity of a fluid flow and give a
physical interpretation for the value of this sum of integrals,

= 1, 2,3,4.

7. Derive a formula for the partial sum SN of the geometric series

Lo aJ,

where a is a fixed complex number not equal to 1. In case lnl < 1, derive
a formula for the sum of the full series.

8. Suppose that the function F possesses a complex derivative at P. Prove


directly from the definition that the quantities 8F/8x, (1/i)(8F/8y),
8F/oz all exist at P and are equal.

9. Assume that F : U

( a)

(b)

---+

C is C1 on an open set U i;;;;; C

Let P E U. Assume that F is angle preserving at P in the sense of


statement (2.2.3.2). Prove that F'(P) exists.
Let P E U. Assume that F "stretches equally in all directions at P"
in the sense of statement (2.2.3.1). Prove that either F'(P) exists
or (F)'(P) exists.

10. Prove that if U i;;;;; C is open, and if f : U


at each point of U, then

---+

C has a complex derivative

is continuous at each point of U.

11. Prove that if F is holomorphic in a neighborhood of P

C, then F

reverses angles at P (part of the problem here is to formulate the result


refer to Theorem 2.2.3).

62

Exercises

12. Formulate and prove a converse to Theorem


discussion in the text.] (Cf. Exercise

2.2.3. [Hint:

Refer to the

9.)

13. In analogy with calculus, formulate and prove a sum, a product, a quo
tient, and a chain rule for the complex derivative.

14. If f and

g'

g are C1 complex-valued functions on an open set U, if f' and


U, and if J'(z) g'(z) for all z E U, then how are f and g

exist on

related?

15. Let

C1 on an open disc U with center 0. Assume


U \ {0}. Prove that u is the real part of a holo

be real-valued and

that

is harmonic on

morphic function on U.

16. Let

g
U \ {O},

C be an open disc with center 0. Suppose that both f and

U \ {O}. If 8J/8z 8g/8z on


g differ by a constant.
Give an example to show that Lemma 2.3. l is false if F is not

are holomorphic functions on

then prove that f and

17.

assumed

to be continuous at p.

18. Compute each of the following complex line integrals:

i ( d(
(b) i
d(
(
( + 4)
+ i)
( ) i : d(
( 2
(d) i ( ( + 4) d(
( ) i ( d(
3
(f) i :(: 7' ) d(
) S
(
(a)

where "f describes the circle of radius

with center 0

and counterclockwise orientation;

_ 1

where "f describes the circle of radius

with

center 0 and counterclockwise orientation;


where 'Y is a circle, centered at 0, with radius 5, oriented

clockwise;

where "f is the circle of radius

clockwise orientation;

where 'Y is the circle of radius

and center 0 with

and center 0 with counter

clockwise orientation;

where 'Y is the circle with center 2+i and radius

with clockwise orientation.

19. Suppose that


every

D(z, r )

C is an open set. Let

U and 'Y

clockwise orientation) and all

F (w)
Prove that

F is

FE C0(U).

Suppose that for

the curve surrounding this disc (with counter

holomorphic.

wE D(z, r )

it holds that

F
J (() d(.

27ri 1-r ( - w

Exercises

63

Figure 2.10

20. Prove that the Cauchy integral formula ( Theorem

C0(D) (here,

assume only that FE


holomorphic on

21. Let

D.

of course,

{ J j J(()

is valid if we

D is some disc) , and Fis

{z : lzl = 1}.

be a continuous function on

2.4.2)

Define, with/= the

unit circle traversed counterclockwise,

F(z) =

(z)

27fi

22.
23.

D(O, 1)? [Hint:

IzI = 1

if

lzl

<

1.

f(z) z.]
Let FE C(D(O, 1)) and holomorphic on D(O, 1). Suppose that IF(z)I
1 when lzl = 1. Prove that IF(z)I 1 for z E D(O, 1).
Let f(z)
z2 . Calculate that the integral off around the circle 8D(2, 1)

Is F continuous on
*

1-r ( - z d(

if

Consider

given by

fo21r f(2

ei8

) d(}

is not zero. Yet the Cauchy integral theorem asserts that

1
f(() d( = o.
lan(2 ,1)
Give an explanation.

24. Use the Cauchy integral theorem to prove that if 11, 12 are the two
contours depicted in Figure 2.10, if Fis holomorphic in a neighborhood
of

D,

and if P is as in the figure, then

j
l-r1

F(()

(-

d( =

j (F(() d(.
l-r2 - p

64

Exercises

'

y
'Y

1
-

\W

Figure

25.

(a)

2.11

Let 'Y be the boundary curve of the unit disc, equipped with coun
terclockwise orientation. Give an example of a C1 function j on a
neighborhood of D(O, 1) such that

i J(() d(

0,

but such that f is not holomorphic on any open set.

(b)

Suppose that f is a continuous function on the disc D(O,


satisfies

for all 0

1
J(() d(
laD(O,r)
< r <

1.

1)

and

Must f be holomorphic on D(O, 1)?

26. Let "( be the curve describing the boundary of the unit box (see Figure

2.11),

with counterclockwise orientation. Calculate explicitly that

i zkdz
if k is an integer unequal to

-1. [Hint:

0
Calculating these integrals from

first principles is liable to be messy. Look for an antiderivative, or think

31.]
Figure 2.12.

about using a different curve. See also the hint for Exercise

27. Do Exercise

26 with 'Y replaced by the triangle in

28. Calculate explicitly the integrals

J
z3dz
laD(8i,2)
'
1
(z-i)2dz.
laD(6+i,3)

Exercises

65

0 + i

1 - i

-1 - i

Figure

29. Calculate

2.12

l
1
d(
_ 1
27ri laD(O,I) ( + 2 '
l
_1 1
d(
27ri laD(0,2) ( + 1
__

__

explicitly.

28 with the aid of the Cauchy integral

30. Calculate the integrals in Exercise

formula and Cauchy integral theorem.

0. Prove that if 'Yl and 'Y2


are counterclockwise oriented squares of center 0 and side length s1 and s2,

31. Let

be holomorphic on

C \ {0}.

Let

s1, s2

>

respectively, then

1
_ . 1 f(() d(
2 7ri 1'-r1

1
_ . 1 f(() d(.
2 7ri 1'-r2

not assuming that the sides of these squares are parallel


to the coordinate axes! ) [Hint: The function f has a holomorphic anti
derivative on {z
x + iy: y > O} and {z
x + iy: y < O}. Use these

( Note:

We are

to evaluate the integrals over the top and bottom halves of the given
curves.

'Y be the counterclockwise oriented unit square in C ( with center


0) and let be the counterclockwise oriented unit circle. If F is
holomorphic on C \ {O}, then prove that
1
1
_ . J F(() d( _ . J F(() d(.
2 7ri }
2 7ri f-r

32. Let

at

[See the hint from Exercise 31].

66

Exercises

33. Prove that the sum used in Definition 2.6.2 to define the integral over a
I
piecewise C curve is independent of the choice of aj and k.
34. Prove Lemma

2.6.3.
2.6.4.

35. Give a detailed proof of Lemma


36. Calculate

where "( is the circle

1
1
(
t
27l'i "I (( - 1) (( - 2i) d ,
with center 0, radius 4, and

counterclockwise ori

entation.

37. Calculate

where 'Y is the


entation.

1
(2+ (
('
t
27l'i
- 2i) (( + 3) d
"I ((
circle with center 1, radius , and
5

counterclockwise ori

38. Verify that

1
1
d(
d(
t
t
27l'i "11 (( - 1) (( + 1) = 27l'i "12 (( - 1) (( + 1) '
8D(l, 1)

where 'YI is

equipped with counterclockwise orientation and

"(2 is 8D(-1, 1) equipped with clockwise orientation.

Notice that these

curves cannot be continuously deformed to one another through U

C\ {-1, 1},

1/(( -1) ((+1) is holomorphic ( at least


intuitively ) . Discuss why this example

the domain on which

you should be able to see this

does not contradict the Cauchy integral formula as discussed in this


chapter.

39. Let "( be the unit circle equipped with clockwise orientation. For each
real number

.A,

give an example of a nonconstant holomorphic function

Fon the annulus

{z: 1/2

<

lzl

<

2}

such that

1
- . F(z) dz= .A.
271''1, }
"I
40. Let 'YI be the curve

8D(O, 1)

and let

"(2

be the curve

8D(O, 3) ,

both

equipped with counterclockwise orientation. Note that the two curves


taken together form the boundary of an annulus. Compute

( a)
(b)
( c)

t (2+5(d(-t (2+5(d( '


27l't
27rt
( -2
(-2
2-2
2-2
1
1
(
(
- t -- (-- t -- ( '
27l'i
( d 27l'i
( d
1
1
( 3 - 3( - 6
(3 - 3 ( - 6
(
t
t
27l'i
(((+ 2)(( + 4) d 27ri (((+ 2) (( + 4) d(
"12

"12

"12

"ll

"11

"fl

Exercises

67

We think of these differences of integrals

as

the ( oriented ) complex line

integral around the boundary of the annulus. Can you explain the an
swers you are getting in terms of the points at which the functions being
integrated are not defined?
*

41. Call a curve

piecewise linear if

it is piecewise C1 and each C1 subcurve

describes a line segment in the plane. Let U C be an open set and let

'Y : [O, 1]
U be a piecewise C1 curve. Prove that there is a piecewise
linear curve : [O, 1]
U such that if F is any holomorphic function
--t

--t

on U, then

1 j F(() d(= 21 j F(() d(.


f-r
f

2rri
.

42. Let

rri

f: C
C be C1. Suppose that for
'Y: [a, b]
C satisfying 'Y(a) = 'Y(b) it holds
--t

--t

i f(z) dz=

any piecewise linear curve


that

0.

F on C such that (8/8z)F


[Hint: See Exercise 3.]

Prove that there is a holomorphic


Conclude that
*

43. If

is holomorphic.

f.

is a holomorphic polynomial and if

1
f(z)zi dz= 0 , j = 0, 1, 2, ... ,
laD(O,l)
then prove that
*

f = 0.

44. Let U C be an open set. Let

f: U

--t

C be a C1 function such that

f"Y f(z) dz= 0 for every curve 'Y such that {'Y(t)}

with its interior, is in U. Prove that

is a circle which, along

is holomorphic.

[Hint:

Refer to

the form of Green's theorem that appears in Appendix A. In fact this


method can be used to prove a version of Morera's theorem for translates
and dilates of

any fixed,

smooth, simple closed curve; cf.

[STE] .]

Chapter 3

Applications of the
Cauchy Integral

3.1. Differentiability Properties of Holomorphic Functions


The first property that we shall deduce from the Cauchy integral formula
is that holomorphic functions, which by definition have

( continuous )

first

partial derivatives with respect to x and y, have in fact continuous deriva


tives of all orders. This property again contrasts strongly with the situation
for real functions and for general complex-valued functions on C, where a
function can have continuous derivatives up to and including some order
but not have a

(k + l)st

Theorem 3.1.1. Let

U.

Then

f E C00(U).

( 8z ) k f(z)

derivative at any point

U C

z E D(P, r) ,
(

is continuous on

D(P, r)

and

f(() 1 d(,
J
(
-z
27ri 11(-Pl=r ( ) k +

Proof. Notice that, for

8D(P, r) .

be an open set and Jet

Moreover, if

( Exercises 58

Also,

f(()

(-w

0, 1,

2,

then

...

f(()
(-z

I( - zl 2'.

--

be holomorphic on

z E D(P, r) ,
k

k,

59).

the function

f------T

It follows that

and

r-

lz -Pl

>

0 for all (

8D(P, r) .

--+ (f(()
-z
--

69

3. Applications

70

as w ----+ z uniformly over (


algebra, it follows that
1
h
uniformly over (
lim

h-+O

of the

Cauchy Integral

8D(P, r). From this assertion, and elementary

(() ) ----+ f(()


( (-(zf(() h) - (f-z
((-z)2
+

8D(P, r) ash----+

f(z + h) -f(z)
h

0.

Therefore

f(()
1
- f(() d(
lim - .!. J
(-Z
h-+O 27ri h 11(,-Pl=r (-(z + h)
l
_
27ri

J
lim .!.
h
Jlc.-Pl=r h-+O

(() ) d( .
( (-(zf(() h) - (f-z
+

[Because the limit occurs uniformly, it was legitimate to interchange the


order of the integral and the limit-see Appendix A.] Thus
lim

h-+O

f(z + h) -f(z)
h

Therefore J'(z) equals

1
_
27ri

f(()
J
Jlc.-Pl=r ((-z)2 d(.

f(()
1 1
d(.
_
27ri Jlc.-Pl=r ((-z)2

This same argument applied to f' shows that f' itself has a complex deriv
ative at each point of D(P, r), given by the formula
(J'(z))'

f(()
J
d(.

2m Jlc.-Pl=r ((-z)3

The function on the right-hand side is a continuous function of z: This


follows from the fact that
f(()
f(()
----+
((-w)3
((-z)3
---

---

uniformly in ( E 8D(P, r) as w ----+ z, as earlier. So f' has a continuous


derivative on D(P, r). By Theorem 2.2.2, the function J' is holomorphic.
In particular, f' E C1 (U) so f E C2 (U). Repeating this argument ( k -1)
times demonstrates the existence of and formula for (8/8z)k f(P). It follows
then that f E Ck(U) for each k 1, 2, 3, .... Therefore f E C00(U).
D
=

The proof of Theorem 3.1.1 amounts to nothing more than a careful


justification of the formal process of "differentiating under the integral sign."
This argument should not be taken lightly. Contrast it with the incorrect
application of "differentiation under the integral " which is considered in
Exercise 3.
Recall that a holomorphic function satisfies (a I &z) f = 0 on the domain
of definition of f. Hence if D1, D2, . .., Dk are partial derivatives and if even

3.1. Differentiability Properties

71

a I az, then D1 D2 ... Dk f = o. Thus only pure holomor


derivatives (8/8z)k of f can be nonzero. We frequently denote these

one of them is
phic

derivatives by

f', f",

and so forth.

This viewpoint y ields again that

f'

is

holomorphic.

\orollary
i'norphic.

Proof.

If f

3.1.2.

Since

C is holomorphic,

is c=, it is certainly

then f' : U

C is holo

C2. So its mixed partial derivatives of

second order commute. As a result,

(!')
Oz

( )

8
az t
az
-

( )

.
az t
Oz

But the last expression in parentheses is identically zero since


morphic. So we see that the

C1

function

f'

is holo

satisfies the Cauchy-Riemann

equations; hence it is holomorphic.

It is worth noting for later purposes that, when we proved Theorem

3.1.1,

we actually proved in effect a stronger statement as well; namely we

have the following:

Theorem

If</> is a continuous function on { ( : I( - Pl

3.1.3.

the function

f given by

f(z)

is defined and holomorphic on

- Pl
instance, if P

I(

=
=

0,

r }, then

</> (
J

( ) d(
2m 11(-Pl=r ( - z

(*)

D(P, r).

Note that the limiting behavior of the quantity


circle

f(z) as z approaches the

might not have any apparent relationship with

1 , </> (

()

(,

then

f(z) =

0,

</>.

For

D(O, 1). [ Obviously


</> on 8D(O, 1).]

on the interior of the disc does not "match up" with

However, in the special case that


of a function

matters are different.

2.4.2)

</> is already

the restriction to the circle

which is holomorphic on a neighborhood of

D(P, r),

In this case the Cauchy integral formula

then

( Theorem

guarantees that

</>(() d(
J
27ri 11(-Pl=r ( - Z
l

Trivially, the limit of the

h(()
J
d( h(z).
27ri 11(-Pl=r ( - Z
integral coincides with </>
h as z
8D(P, r).
=

Thus, in these special circumstances, the holomorphic function produced by

( )
*

"matches up" with the boundary data

</>.

It must be clearly understood

that this happy circumstance fails in general.


The fact that a holomorphic function, which is required by our definition

to be continuously differentiable, is actually c= sometimes enables one to

72

3. Applications of the Cauchy Integral

Figure 3.1
see that functions are holomorphic under surprisingly weak conditions. The
following classical result illustrates this point:
(Morera). Suppose that f : U
C is a continuous function
on a connected open subset U of C. Assume that for every closed, pjecewfse-
C1 curve 1: [O, 1]
U, 1(0)
1(1), it holds that

Theorem 3.1.4

----t

----t

Then

f is holomorphic on U.

i !(() d(

0.

Fix Po EU. Define a function F: U


C as follows: Given Q EU,
choose a piecewise C1 curve 'I/; : [O, 1]
U with 'l/;(O)
Po and 'l/;(1) Q.
Set
Proof.

----t

----t

F(Q)

t f (() d(.

We claim that the condition that the integral of f around any piecewise C1
closed curve be 0 yields that

t J(() d( i f(() d(
=

for any two curves 'I/;, T from Po to Q (see Figure 3.1). Thus F is unam
biguously defined. To prove the claim, note that the curve consisting of 'I/;

3.1. Differentiability Properties

73

T-1 (T traced in reverse ) is a piecewise C1


integral of f over the curve is 0. Thus

followed by
the

0=

1 f(() d(
J

closed curve. Hence

1 f(() d( + 1 f(() d(,


'71
Jtf;

so that

t f(() d( i f(() d(
=

'

and the claim is established.


Now we will show that the function Fis continuously differentiable and
satisfies the Cauchy-Riemann equations. We proceed as follows:

P = (x, y) E U and 'I/; a piecewise C1 curve connecting Po to


(x, y). Choose h E R so small that (x + h, y) E U and so that the segment
fh(t) = (x + t, y), 0 :::; t :::; h, lies in U. Thus fh describes the line segment
connecting (x, y) to (x + h, y). Let 'l/Jh be the piecewise C1 curve consisting
of 'l/J followed by fh. Then
Fix a point

1 f(() d( - 1 J(() d(
Jt/Jh
Jtf;
1 f(() d(
hh

F(x + h, y) - F(x, y)

lh f(P

Denote the real part of F by

U and

s)ds.

the imaginary part by V. Then taking

real parts of the last equation yields

U(x + h, )
Letting

h---+

U(x)

h
l
l J(P
Re

s)ds

ll

Re f( P +

0 on both sides of this equation reveals that

and equals Ref( p )

( we

s)ds.

[a/ax]U(p) exists

have of course used the fundamental theorem of

calculus on the right-hand limit ) .


Similar calculations yield that

au
ay

= -Imf,

av
ax

= lmf,

av
ay

=Ref.

Thus the first partial derivatives of Fexist as claimed. These partial deriva
tives are continuous, by inspection, since

f is.

Finally we see that Fsatisfies

the Cauchy-Riemann equations:

au av
au
ax .. ay-rta ay
/

av
.
ax

3. Applications of the Cauchy Integral

74

Thus F is holomorphic.
holomorphic. But F'

f,

Now Corollary

3.1.2

guarantees that F' is also

proving our assertion.

The conclusion of Morera's theorem actually follows under the weaker

hypothesis thatf f(() d(


0 for any piecewise C1 curve 'Y that consists of
,,
a succession of horizontal and vertical line segments (e.g., the boundary of a
=

rectangle). That this hypothesis suffices can be seen by careful examination


of the given proof. The only really new observation needed is that, given

p, q E U, with U a connected open set in C, there is a curve from p to q


consisting of horizontal and vertical line segments: See Exercises 5 and 7. It
is also possible to use just triangles in the hypothesis of Morera's theorem.
Morera's theorem and its proof give a new view of the question of finding
holomorphic antiderivatives for a given function. The theorem says that
itself must be holomorphic. The

f
f

proof reveals that a holomorphic function


f(() d( 0 for all piecewise C1 closed

has a holomorphic antiderivative iff

,,
curves in U. This sufficient condition for the existence of an antiderivative

is also necessary (Theorem

2.4.3).

The vanishing off

,,

J( ) d(

for all closed

'Y is sometimes referred to as "path independence." For,( as we saw in the


proof of Morera's theorem, the vanishing is equivalent to the equation

i J(()
for every pair of curves 'lj;,

d(

i J(()

d(

which have the same initial and terminal points.

Thus the value off f(() d( depends only on the initial and terminal points
,,
of "(-it is independent of the particular path that 'Y takes.

3.2. Complex Power Series


The theory of Taylor series in real-variable calculus associates to each infin
itely differentiable function
at each point of

JR,

from JR to JR a formal power series expansion

namely
oo

L.,,,

J(n) (p)
I
n.

n=O

(x

- )n
p ,

pER

There is no general guarantee that this series converges for any

x other than

p. Moreover, there is also no general guarantee that, even if it does

converge at some

x I- p,

its sum is actually equal to

J(x).

An instance of

this latter phenomenon is the function

J(x)

e-l/x2

if

x I- 0
x 0.
=

This function can be easily checked to be C00 on JR with

f' (0)

f" (0)

J(O)

. . . (use l'Hopital's rule to verify this assertion). So the

3.2. Complex Power Series

75

Taylor expansion off at 0 is

0+

Ox + Ox2 + Ox3

which obviously converges for all

+ ... ,

f(x)

with sum = 0. But

is 0 only if

0. (An example of the phenomenon that the Taylor series need not even

x = p is given in Exercise 64.) The familiar functions of


ex, and so forth-all have convergent power series. But
most C00 functions on JR do not. Real functions f that have, at each point
p E JR, a Taylor expansion that converges to f for all x near enough to p are
called real analytic on R
converge except at

calculus-sin, cos,

In holomorphic function theory, it is natural to attempt to expand


functions in powers of

z.

We would expect to need

powers only and not z

powers since our original definition of holomorphic functions was designed


to rule out any z's (in polynomial functions in particular). Compared to the
real-variable case just discussed

( C00

functions from

to expand a holomorphic function in powers of


fact, we shall see that if

-t

set U and if P E U, then the formal

converges for all


all

f(z)

JR

to

JR),

the attempt

works remarkably well. In

is a holomorphic function on an open

expansion at P ,

in some neighborhood of P and it converges to

near P. Specifically, if

converges to

for all

z in

>

0 is such that D(P, r )

f(z)

for

U, then the series

D(P , r). This expansion shows a sense in which

holomorphic functions are a natural generalization of polynomials in

z.

We prove these assertions in the next section. First, we need to intro


duce complex power series formally and establish a few of their properties.
We assume the reader to be familiar with real power series at the level of
freshman calculus.
A

sequence

of complex numbers is a function from

{l, 2,... }

to

(sometimes it is convenient to renumber and think of a sequence as a function

{ O, 1, 2,. .. } to C). We usually write a sequence as {a1, a2,. .. } or


{ak}b1 (resp. {ao, a1,... } or {ak}b0). The sequence is said to converge
to a limit f E C if for any f. > 0 there is an No such that k ;::::: No implies
lak - fl < f.. It is frequently useful to test convergence by means of the
Cauchy criterion:

from

Lemma 3.2.1. Let

{ak}b1

be a sequence of complex numbers. Then

{ak}

converges to a limit if and only if for each f. > 0 there is an No such that

j, k;:::::

No implies

laj - akl

< E.

76

3. Applications of the Cauchy Integral

Proof. If

k No

{ak}

implies

f, then, given E > 0, there is an No


E/2. Then if j, k No, it follows that

converges to

lak -fl

<

laJ - akl .::::: laJ - fl + If - akl

<

E/2 + E/2

such that

E.

{ak} satisfies the criterion in the lemma, then set ak


o:k + if3k with o:k, f3k real. Now {o:k} is a Cauchy sequence of real numbers
since lo:k - O:j I .:::; lak - aj I. By the completeness of the real numbers, the
sequence {o:k} thus has a limit o:. Likewise the sequence {/3k} has a limit /3.
We conclude that the original complex sequence {ak} has the limit o:+if3.
0
Conversely, if

Definition 3.2.2. Let


at

P)

PE

C be fixed. A

complex power series

(centered

is an expression of the form


00

I:ak(z - P)
k=O
where the

( Note:

{ak}0

are complex constants.

This power series expansion is only a formal expression. It may or

any sense.)
k
If Lo ak(z - P) is a complex power series, then its Nth partial sum
is by definition given by SN (z)
2::-::f=o aj (z - P)J. We say that the series
converges to a limit f(z) at z if SN(z) -----t P(z) as N -----too.
k
It follows from the Cauchy criterion that 2::::%=o ak(z -P) converges at
z if and only if for each E > 0 there is an No such that m j No implies

may not converge-in

k
I:ak(z - P)
k=j
From this we see that a
that

ak(z - P)

-----t

necessary condition

<

f.

for

L ak(z - P)

to converge is

0.

Exercise: Show that this condition is

not sufficient

for convergence.

k
Lo ak(z - P) converges at some z, then the
each w E D(P, r), where r
lz - Pl. See Figure 3.2.

Lemma 3.2.3 (Abel). If


series converges at

k
L ak(z - P)

Proof. Since

k -----t oo. So the latter is a bounded


lakl .:::; Mr-k for some ME JR.
Now, for each k,

ak(w - P)

k
ak(z - P) -----t 0
k .:::;
lak(z - P) l M

converges, it follows that


sequence; that is,

l .:::; lakllw - Pl

k.:::;

( lw Pl ) k

as

or

77

3.2. Complex Power Series

Figure 3.2

Since w E D(P, r), it holds that (lw - Pl /r) < 1. Therefore the series of
nonnegative terms

fc )
w

Pl

k=O

converges: It is a geometric series with ratio less than 1 (in absolute value).
Thus the series
00

k=O
converges, since it converges absolutely.

The lemma tells us that the domain of convergence of a complex power


series is (except for some subtleties about its boundary) a disc.

We now

make this observation more explicit:

Definition 3.2.4. Let

2:::%':0 ak(z

sup{lw - Pl :

the

L ak(w - P)k

radius of convergence
disc of convergence.

is called the

- P)k be a power series. Then


converges}

of the power series. We will call D(P, r)

Note that the radius of convergence may be +oo; in this case, we adopt
the convention that D(P, oo )
case, D(P, r)

0,

C It may also happen that

O; in this

since it cannot be that lw - Pl < 0. We can reformulate

Lemma 3.2.3 as follows:

Lemma 3.2.5. If :L%':o ak(z - P)k is a power series with radius of conver
gence r, then the series converges for each w E D(P, r) and diverges for each

3. Applications of the Cauchy Integral

78

such that

l w - Pl > r. (If r = +oo, then b y convention n o

satisfies

l w - Pl> r.)
Note that the convergence or divergence question for w with l w - Pl = r is
left open. For such w, the convergence or divergence has to be investigated
directly on a case-by-case basis.
Much of the value of the concept of radius of convergence comes from
the fact that the radius of convergence can be determined from the coeffi
cients { ak } of the power series. This determination is as follows:
Lemma 3.2.6 (The root test). The radius of convergence of the power
series Lo a k(z - P)k is
(a)

iflim supk---++oo lak ll/k > 0, or


(b)

+oo

iflim supk---++oo lak ll/k= 0.


We first treat the case lim supk---++oo l ak ll/k> 0. Set
a= lim sup lak ll/k.
k---++oo
If lz - Pl > l/a, then lz - Pl = c/a, some c > 1. It follows, for infinitely
many k, that lak ll/k> a/c. For such k,
Proof.

lak(z - P)k l = (lak ll/k)k lz - Pl k= (lak ll/k)k(c/a)k> (a/c)k(c/a)k= 1.


Hence I: ak(z ,-- P)k diverges because its terms do not have limit zero.
If lz - Pl < l/a, then lz - Pl = d/a, some 0 < d < 1. Let c satisfy
d < c < 1. Then, for all sufficiently large k, lak ll/k < a/c. Thus, for such k,
k
lak(z - P)l k= (lak ll/k)k lz - Pl k < (a/c)k(d/a)k=

( )

Since I:0(d/c)k is an absolutely convergent (geometric) series, it follows


that I: ak(z - P)k converges.
The case lim supk---++oo lak ll/k= 0 is treated similarly and is left to the
reader (see Exercise 19) .
D
The convergence properties of a power series at points within its radius
of convergence are better than just simple convergence. First of all, the series
Lo lak l lz - Plk converges for each z E D(P, r), with r being the radius of
convergence of I: ak(z - P)k. (The series converges absolutely, i.e., the sum
of the absolute values converges. As already noted, if Lo l ak l , Cl.k E re,

3.2. Complex Power Series

converges, then

79

Lo O:k converges.

But the converse is not true in general! )

Moreover, the convergence is uniform on each disc D ( P,

R), 0:::; R

<

r.

The

precise definition of uniform convergence is as follows:

Definition 3.2. 7. A series

Lo fk(z)

formly on a set E to the function


that if

N N0,

g(z)

of functions

if for each

>

fk(z)

converges uni

0 there is an

No such

then
N

L fk(z)- g(z)

<

for all z EE.

k=O
The point is that
depending on

No does not

( but

not on

z)

depend on

z EE:

There is, for each

that works for all

E,

an

No

z EE.

We take this opportunity to record the uniform Cauchy criterion for


series:

Definition 3.2.8. Let


series is said to be
integer

No such

Lo fk(z)

be a series of functions on a set E. The

uniformly Cauchy

that if m

j No,

if, for any

>

0, there is a positive

then

1n

L fk(z)

k=j

<

for all

z EE.

If a series is uniformly Cauchy on a set E, then it converges uniformly


on E ( to some limit function ) . See Appendix A. From this it follows that if

L lfk(z)I

is uniformly convergent, then

L fk(z)

is uniformly convergent ( to

some limit ) .
Now we return to our main theme, the convergence of a power series.
Inspection of the proof of Lemma 3.2.3 reveals the following:

Proposition 3.2.9. Let

Lo ak(z - P)k

be a power series with r adius

r. Then, for any number R with 0 :::; R < r, the


Lo lak(z - P)kl converges uniformly on D ( P, R). In particular, the
Lt ak(z - P)k converges uniformly and absolutely on D ( P, R).
of convergence

series
series

Our principal interest in complex power series is to be able to establish


the convergence of the power series expansion of a holomorphic function;
the convergence should be

to the original function.

It is important first to

note that any given function has at most one power series expansion about
a given point P. For this, we shall use the following lemma.

Lemma 3.2.10. If a power series


00

l: a1(z - P)l

j=O

Applications of the Cauchy Integral

3.

80

has radius of convergence r > 0, then the series defines a C00 function f(z)
on D(P, r). The function f is holomorphic on D(P, r). The series obtained
by termwise differentiation k times of the original power series,
00

(3.2.10.1)
LJ(j - 1) (j - k + l)aJ (z -P)J-k,
j=k
converges on D(P, r), and its sum is [8/8z]k f(z) for each z E D(P, r).

Proof. We shall first show that f(z) has a complex derivative at each z

D(P, r).

W ith
lim

fixed,

f(z + h) - f(z)

lim

h--+O
By Proposition 2.1.6,

(aj(z + h)j -ajzj)


So

ll(aj(z

In particular, for

1h
1

Now

lhl

!(r - lzl),
.

lzl))J-l

fo1 h j z

aj

h)j -ajzj)

(aj(z + h)1-ajz1)

LJlaJl(!(r

_!_ (a (z + h)j -a zj)


h J
J

h--+O L.,,
j=O

( )
*

th)j-l dt.

jlaJl(lzl + lhl)j-1 .

jlaJI 2(r + lzl)

) j-1

converges (use the root test).

Hence, by the

Weierstrass M-test (see Appendix A), the series on the right-hand side of
(*

converges uniformly in

for

small. So, in the right-hand side of ( *

the summation and limit can be interchanged (cf. Appendix A). Hence
exists and equals

:L:'; o jajzJ-1.

f1 (z)

This series also has radius of convergence

(by Lemma 3.2.6). The conclusions of the lemma now follow inductively.

Now we have:

Proposition 3.2.11. If both series L:'; o aj(z

- P)J and :L:'; o bj(z - P)J


and if:l::'; o aj(z-P)J :L:'; o bj(z-P)J

converge on a disc D(P, r) , r > 0,


on D(P, r) , then aj = bj for every j.
Proof. Let f(z)

:L:'; o aj(z - P)J.

is infinitely differentiable on

We know from Lemma 3.2.10 that

D(P, r) and

(:z ) k f(z) f j (j
=
J

-1)

(j - k + l)aj(z - P)J-k_

3.3. Power Series Expansions

So

81

( : ) k (P)=k!ak
J

or

k J(P)
(/z)
ak= k!
L b1(z - P)j =J(z),

Since it also holds that

00

j=O

it follows from the same argument that

k J(P)
(/z)
bk= k!
for every k. As a result, ak=bk \:/k.

3.3. The Power Series Expansion


for a Holomorphic Function

As previously discussed, we first demonstrate that a holomorphic function


has a convergent complex power series expansion (locally) about any point
in its domain. Note that since a holomorphic function is defined on an
arbitrary open set U while a power series converges on a disc, we cannot
expect a single power series expanded about a fixed point
to converge to
Jon all of U.

f
D(P, )
f (okf/k)(P) (z - P)k

Theorem 3.3.1. Let U


Let

PE

be an open set and Jet

U and suppose that

be holomorphic on U.

U. Then the complex power series

k=O

has radius of convergence at least

r.

It converges to

J(z)

on D(P,

).

Proof. Recall that from Theorem 3.1.1 we know that


f is C00 So the
coefficients of the power series expansion make sense. Given an arbitrary
'
z
r ) , we shall now prove convergence of the series at this z. Let r
be a positive number greater than lz - l but less than r so that

E D(P,

P
D(P, ) D(P, )
zE D(P, )
Assume without loss of generality that P = (this simplifies the notation
considerably, but does not change the mathematics) and apply the Cauchy
integral formula to f on D(P, ) Thus for zE D(P, ) = D(O, ) we have
f(() d(
J(z) = J
2m Ji<l=r' ( - z
'
r

'
r

'

'

'
r

82

3. Applications of the Cauchy Integral

2m Ji

f(()
d(
(l=r' ( 1 z. (-1

J f(() )z. CI)kd(.


27l'Z Ji(l=r' (
k=O
last equality is true because z
l l < r', 1(1 r',
=

Notice that the

so

(* )
hence

I: ( z (-1 )k is a convergent geometric series expansion which converges to

1/(1

(-1 )

(see Exercise 7 of Chapter

uni ormly on {( 1(1


f
:

integral.

r'}.

2).

Moreover, the series converges

This fact allows us to switch the sum and the

This step is so crucial that we write it out rather carefully (see

Appendix A). Set

SN(z, ()

2::=0(z C1 )k.

Then we have

f(() lim SN(z, ()d(


( N->oo
( ()

! SN(z , ()d(
lim J
27l'i N->oo Ji(l=r' (

J
2m Ji

(l=r'

by uniform convergence. Now this last expression equals

lim J
2m N->oo L.,,

l.:=0 xl(l=r'

f(() (z . C1)kd(
(

since finite sums always commute with integration. The last equation equals

f(()
k+ d(.
(l=r' ( I

J
L zk
N-+oo
2m Ji
lim

k=O

f zk_
l
27ri

J f(()
+l d(
Ji(l=r' (k
1 ak
f
zk
2=
k! azk (o)
k=O
k=O
oo

by Theorem

3.1.1.

( **)

We have proved that the power series expansion


converges, and that it converges to

f(z),

as desired.

**

of

about P

Let us separate the forest from the trees in this (very important) proof.
The entire discussion from line

( )
*

to line

**

was to justify switching the

sum and the integral. The rest was trivial formal manipulation, using the

3.3. Power Series Expan sions

83

,,-- .......
I/

I\

\..

I
I

...... _

Figure 3.3

1/(( - z)

idea of writing

as

and expanding the second factor as a geometric series.


If we are given a holomorphic function

PE U,

f on

an open set U and a point

then it is easy to estimate a minimum possible size for the radius r

of convergence of the power series expansion of


the distance of

P to C \ U.

about P. Namely let

Then the theorem implies that

power series will converge at least on

D(P, f).

r.

be

Thus the

It may or may not converge

on a larger disc. See Figure 3.3.

EXAMPLE 3.3.2. Let

f : D(O, 2)
f about 0

power series expansion of

C be given by

f(z)

lj(z

2i).

The

is

00

:_.)2i)-k-l ( -1)kzk.
k=O
D(O, 2) and on no larger set. We could have
J(z) is unbounded as D(O, 2) 3 z -2i.
on D(O, 2 + t: ) , some E > 0, then f would be

This series converges precisely on

predicted this since the function


If the power series converged
bounded near

z =

-2i,

and that is false.

3. Applications of the Cauchy Integral

84

I
I

I
I
I

------

;"'/
/
/

'\

'

',

\
'\

....
,

'\
\
\

'

'

I
I

I
I

/
,,
----- """

__

..... _

-2i

\
\
\

Figure 3.4
The power series expansion of f about z

1 is

00

L(2i + 1)-k-1(-1)k(z- 1)k.

k=O

This series converges precisely on D(l,

v'5)

and on no larger set. In partic

ular, the convergent power series gives a way to extend the domain of the
function given by

(*)

so that it will be holomorphic on D(O, 2) U D(l,

v'5)

( of course we know that f is holomorphic on that set anyway-indeed f is

holomorphic everywhere except at z

-2i).

Again, we might have pre

dicted the radius of convergence because of the bad behavior off near -2i.
See Figure 3.4.

It should be noted that complex power series give an easy way to define

holomorphic functions other than polynomials. Examples are

oo

sinz

and

.
z2j+l
(-1)1
(2j + l)!

.
z21
L(-1)1- ' ) I'
(2J .
j=O
oo

cosz

Observe that each of these functions, when restricted to z real, coincides


with its familiar real-variable analogue.

3.4. The Cauchy Estimates and Liouville's Theorem


This section will establish some estimates for the derivatives of holomor

phic functions in terms of bounds on the function itself.

The possibility

3.4. Cauchy Estimates, Liouville's Theorem

85

of such estimation is a feature of holomorphic function theory that has


no analogue in the theory of real variables.

fk(x)
l; but

sin

kx, x

IR, k

Z, are for all

For example:

The functions

bounded in absolute value by

their derivatives at 0 are not bounded since

f(O)

k.

The reader

should think frequently about the ways in which real function theory and
complex holomorphic function theory differ. In the subject matter of this
section, the differences are particularly pronounced.

Theorem 3.4.1 (The Cauchy estimates). Let


phic function on an open set U,

D(P, r), r > 0,


k=l,2,3 ... we

is contained in U.
have

Proof. By Theorem

Set M

: U

---+

be a holomor

U, and assume that the closed disc

= supzED(P r) lf(z)I.
,

Then for

3.1.1,

fJk J
(P)
fJzk

J(()
d(.
27ri Jl<:-Pl =r (( P)k+l

_!}__ J

Now we use Proposition 2.1.8 to see that

l fJuzk fk (P)
!)

M k!
k!
Iii
. 27rr sup
.
::;
k+
- Pl l
27r
r
rk
l<-Pl= I(

::; -

Notice that the Cauchy estimates enable one to estimate directly the
radius of convergence of the power series

Namely
lim sup

k-++oo

for any

such that

(k)(P) 1/k
k!

D(P, r) <;;;

which equals

[
is at least

1/(1/r)

M.

k
k ! 1 1/
rk- - k.'
k-++oo -

.
::; hm sup

1
r

U. In particular, the radius of convergence,

l/k -1
(k)
j (P)
lim sup
,
k'
k-++oo

for all

D(P, r) <;;; U. Hence the radius of


P to C \ U (this result
perspective in Theorem 3.3.1).

such that

convergence is at least as large as the distance from


was considered from a different

The derivative bounds of Theorem

3.4.1

have some remarkable conse

quences. The first one is the fact that a holomorphic function on

that is

bounded in absolute value is in fact constant. We shall need the following


lemma to prove this:

3. Applications of the Cauchy Integral

86

f
afIaz 0
f
f
af/az = 0.
aj/az=0. aj /ax= aj jay= 0.
f
entire
f
ez
z, z
f(z) = /
z =0.
Lemma 3.4.2. Suppose that
open set u c. If

Proof. Since

is a holomorphic function on a connected

on U, then

is holomorphic,

But we have assumed that

Thus

is said to be

all of C, that is,

: C

poly nomial is entire,


z

So J is constant.

A function

is const ant on u .

----+

if it is defined and holomorphic on

C is holomorphic. For instance, any holomorphic

is entire, and sin

cos

are entire.

is not entire because it is undefined at

The function

[In a sense that

we shall make precise later, this last function has a "singularity'' at


question we wish to consider is:

O.]

The

"Which entire functions are bounded?"

This question has a very elegant and complete answer as follows:

Theorem 3.4.3 (Liouville's theorem). A bounded entire function is con


s t ant

PE

0.

Proof. Let J be entire and assume that I J


C and let

>

(z) I

The result is

az f(P) I :::; r
l
r 0,
(P) =0.

Ml!

Since this inequality is true for every

Since

>

= E D(P, r).

:::; M for all

We apply the Cauchy estimate for k

C. F ix a

1 on

we conclude that

a1=0.
az -

was arbitrary, we conclude that

By the lemma,

is constant.

Liouville's theorem is so important that it is worthwhile to sketch a

aj /az(P) =0.
akf (P)
l 8zk J :::; ---:;:k
ak /azk(P) = 0 P
f P
f

slightly different approach to the proof:


that

M k!

Hence
at

For a fixed P, we showed above

We can similarly see from the Cauchy estimates that

for all

for all r >

0.

power series converges to

0.

and all k 2: 1 . Thus the power series of

contains only the constant term-all other terms are


on all of C, it follows that

Since the

is constant.

The observations just given imply a generalization of Liouville's theorem:

87

3.4. Cauchy Estimates, Liouville's Theorem

Theorem 3.4.4. If f : C

-> C

is an entire function and if for some real

number C and some positive integer k it holds that

for all

with

lzl

>

1,

then f is a polynomial in

k
k
(8 +f./8z +f.)J(O)

Proof. Exercise. Show that

of degree at most k.

0 for all f

0 by Cauchy's

>

estimates, so that the power series of f about 0 has only finitely many

terms.

One of the most elegant applications of Liouville's theorem is a proof


of what is known as the fundamental theorem of algebra:

Theorem 3.4.5. Let


Then

p(z)

be a nonconstant (holomorphic) polynomial.

has a root. That is, there exists an

Proof. Suppose not. Then


then

lp(z)I-> +oo.

Liouville's theorem,

rithm) we may divide

p1

C such that

p(o)

0.

g(z)
l/p(z) is entire. Also when lzl ->
1/lp(z)I-> 0 as lzl-> oo; hence g is bounded.
constant; hence p is constant. Contradiction.
=

is

p has degree k
z- o into p with
p(z)

Here

Thus

If, in the theorem,

p1

is a polynomial of degree k-

for some polynomial

P2(z)

1,

By

then (by the Euclidean algo

no remainder to obtain

( z- o) Pi (z).
1

. If k-

has a root f3 . Thus Pl is divisible by

p(z)

oo,

(z -

1,

then, by the theorem,

(3) and we have

(z- o)(z- (3) P2(z)

of degree k- 2. This process can be continued

until we arrive at a polynomial

Pk

of degree

O;

that is,

Pk

is constant. We

have derived the following:

Corollary 3.4.6. If p(z) is a holomorphic polynomial of degree k, then there


are k complex numbers

o1, ... Ok

(not necessarily distinct) and a nonzero

constant C such that

p(z)

(z -

01

) (z - ok).

It is possible to prove the fundamental theorem of algebra directly,


without using results about holomorphic functions as such: see Exercise 34.
Another proof of the fundamental theorem is explored in Exercise 35.
The fundamental theorem is but one of many applications of complex
analysis to algebra and number theory.

88

3.

Applications of the Cauchy Integral

3.5. Uniform Limits of Holomorphic Functions


We have already seen that a convergent power series
morphic function ( Lemma

3.2.10).

( in

z)

defines a holo

One can think of this fact as the assertion

that a certain sequence of holomorphic functions, namely the finite partial


sums of the series, has a holomorphic limit. This idea, that a limit of holo
morphic functions is holomorphic, holds in almost unrestricted generality.

Let f : U ---+ C, j
1, 2, 3 ... , be a sequence of holo
j
morphic functions on an open set U in C. Suppose that there is a func tion
f : U ---+ C such that, for each compact subset E of U, the sequence fj IE
converges uniformly to !IE- Then f is holomorphic on U . (In particular,
f E C00(U).)

Theorem 3.5.1.

Before beginning the proof, we again note the contrast with the real
variable situation. Any continuous function from Ill to Ill is the limit, uni
formly on compact subsets of Ill, of some sequence of poly nomials: This is
the well-known Weierstrass approximation theorem. But, of course, a con
tinuous function from Ill to Ill certainly need not be real analytic, or even

C00

The difference between the real-variable situation and that of the the

orem is related to the Cauchy estimates. The convergence of a sequence of


holomorphic functions implies convergence of their derivatives also. No such
estimation holds in the real case, and a sequence of

C00

functions can con

verge uniformly without their derivatives having any convergence properties


at all ( see Exercise

and

[RUDI]).

We shall give one detailed proof of Theorem

3.5.1 and

sketch a second.

The first method is especially brief.

P E U be arbitrary. Then choose r > 0


{fj} converges to f uniformly on D(P,r)
continuous, f is also continuous on D(P,r). For any

Proof of Theorem 3.5.1. Let


such that

D(P,r) U.

and since each

z E D(P,r),

fj is

f (z )

Since

lim

J-+00

fj ( z)

fj(() d(
1
11( Pl=r ( - Z
fj(() d(
1
lim

2m Jic.-Pl=r j-+oo ( - z
1
f(() d(.
1
27ri 1lc. -Pl=r ( - z
lim

j-+oo 27ri

89

3.5. Uniform Limits of Holomorphic Functions

The interchange of integral and limit is justified by the fact that, for z fixed,
J1(() / (( - z) converges to f(() / (( - z) uniformly for ( in the compact set
{(:I(- Pl= r}-again see Appendix A.
By Theorem 3.1.3, f(z) is a holomorphic function of
Hence f is holomorphic on U.

on

D(P, r).

Another closely related way to look at Theorem 3.5. l is to ma


_ ke the
following observation: Since {!1} is uniformly Cauchy on D(P, r) and hence
in particular on 8D(P, r), it follows from the Cauchy estimates that { fj } is

uniformly Cauchy on D(P, ro) for any ro < r. It is a standard result (see
Appendix A) that if a sequence of functions with continuous first partial
derivatives converges uniformly and if the partial derivatives also converge
uniformly, then the limit of the sequence has continuous first partials; also
the (first) partial derivatives of the limit are the limit of the corresponding
partial derivatives. In the present setup, this implies that f has continuous
first partials and that these must satisfy the Cauchy-Riemann equations. So
we see again that f must be holomorphic.

Notice that this second method of proof of Theorem 3.5.1 also gives a
proof of the following corollary:
Corollary 3.5.2. If fj,

f, U

E {O, 1, 2, ...} we have


k

are

as

in the theorem, then for any integer

k
k
(z
f
)
1
:
)
---; ( :z ) f(z)
(z

uniformly on compact sets.

Proof. Let E be a compact subset of U . Then there is an r

for each
the set

z EE, the closed disc D(z, r) is contained in U.


Er=

is a compact subset of U. For each


that

So, for all

z EE,

> 0 such that,


F ix such an r. Then

(u D(z,r))
zEE

z EE, we have from the Cauchy estimates

we see that

( aaz ) k (!Ji(z)- f12(z))

:'.S;

k sup
r (,EEr

l!Ji(()-!12(()1.

90

3. Applications of the Cauchy Int egral

)1 , )2

The right-hand side goes to 0 as

{!j }

since Er is compact, so that


converges uniformly on Er. Thus {(a/ oz ) k fj } is uniformly Cauchy on

oo,

E, as required.

The ideas being considered in this section can also be used to enhance
our understanding of power series. A power series
00

L aj (z - P)j

j=O

is defined to be the limit of its partial sums

N
SN(z)

L aj(z

- P)l.

j=O

Of course the partial sums, being polynomials, are holomorphic on any disc

D(P, r). If the disc of convergence of the power series is D(P, r0), then let
f denote the function to which the power series converges. Then for any
0 < r < ro we have that
SN(z) - J(z),
uniformly on

f(z)

D(P, r).

By Theorem

is holomorphic on

D(P, ro).

3.5.1,

we can conclude immediately that

Moreover, by the corollary, we know that

This shows that a differentiated power series has a disc of convergence at


least as large as the disc of convergence ( with the same center ) of the original
series, and that the differentiated power series converges on that disc to the
derivative of the limit of the original series. In fact, the differentiated series
has exactly the same radius of convergence as the original series as one sees
from Lemma

3.2.6. ( Exercise:

3.6. The Zeros of


Let

Check this, recalling that limk-++oo

1.)

Holomorphic Function

be a holomorphic function. If

out that

lfk

is not identically zero, then it turns

cannot vanish at too many points. This once again bears out the

dictum that holomorphic functions are a lot like polynomials. The idea has
a precise formulation as follows:

Theorem 3.6.1. L et U
be holomorphic.

{ Zj }1

Let Z

\ {zo }

C be a connected open set and let

{z

such that

Zj

f(z)

- zo, then J

O }.
=

If there are a zo

0.

U
E

Z and

3.6. The Zer os of a Holomorphic Function

91

Let us formulate Theorem 3.6.1 in topological terms. We recall that a

zo is said to be an accumulation point of a set Z if there is a sequence


Z \ { zo} with limj-.oo
zo. Then Theorem 3.6.l is equivalent to
the statement: If f : U --t C is a holomorphic function on a connected open
set U and if Z = { z E U : f ( z )
0 has an accumulation point in U, then
f =- 0.
point

Zj}

Zj

There is still more terminology attached to the situation in Theorem

3.6.1. A set Sis said to be

D(s, 1: )

n S=

}.

{s

discrete if for each s

E Sthere is an

1:

>

0 such that

People also say, in an abuse of language, that a discrete

set has points which are "isolated " or that S contains only "isolated points."
Theorem 3.6.1 thus asserts that if

is a nonconstant holomorphic function

on a connected open set, then its zero set is discrete or, less formally, the
zeros of

f are isolated. It is important to realize that Theorem 3.6.1 does not


f can have accumulation points

rule out the possibility that the zero set of


in C

\ U;

in particular, a nonconstant holomorphic function on an open set

U can indeed have zeros accumulating at a point of


function

f(z)

sin ( l / ( 1

is holomorphic on U

- z))

on the set

{1

: n

nn

oU. For
D(O, 1 )

example, the
and vanishes

1, 2, 3, ...

}.

Plainly Z has no accumulation points in U; however the point 1 E oU

accumulation point of Z.

is an

Proof of Theorem 3.6.1. We first claim that, under the hypotheses of the
theorem,

(8/oz)n f(zo)

0 for every nonnegative integer

n. If this is not the

case, let no be the least nonnegative integer n such that

(oz )no f(zo) -/=


Then we have, on some small disc

f(z)

00

J=nu
Therefore the function

g(z)

D(zo, r),

(()azjJ f(zo))

the power series expansion

(z - zo)l
j!

defined by

=-

j f(zo) (z - zo)J-no
)
(!_
_!
oz
L

J=no
is holomorphic on

0.

D(zo, r) and g(zo) -/= 0 since ( gz ) 110 f(zo) -/= 0.

Notice that

the indicated power series has the same radius of convergence as that for
itself. Furthermore,
of

g, g(z0)

g(z1)

0 for l

1, 2, 3, .... But then, by the continuity

0. This contradiction proves our claim.

92

3. Applications of the Cauchy Integral

The just-established claim implies that


E

{z

(8/8z)j J(z)

EU:

0 for all

j}

is not empty. It is also relatively closed in U. This is so because each of the


sets Ej

{z

EU :

(8/8z)J f(z)

O},

being the inverse image of a closed

set under a continuous function, is closed; and E


Finally, E is open.

To see this, notice that if P E E and r

selected so that D ( P, r ) U, then we have for all

f(z }
Since

flD(P,r)

nJ EJ .

t, (:J

/ (P }

(z

rP)J

>

0 is

E D(P, r ) that

"0.

0, all derivatives off vanish at all points of D ( P, r ) . Hence

D ( P, r ) E. Therefore E is open.

We have proved that E is an open, closed, and nonempty subset of U.


Since U is connected, it follows that E

U, proving the theorem.

Corollary 3.6.2. Let U C be a connected open set and D ( P, r ) U. If

is 1101omorphic on U and

JID(P,r)

0, then

0 on U.

Corollary 3.6.3. Let U C be a connected open set. Let


phic on U. If

f =g.

{z

EU:

J(z)

g(z)}

f, g

be holomor

has an accumulation point in U, then

f, g be
or g = 0 on U.

Corollary 3.6.4. Let U C be a connected open set and let

holornorphic on U. If f

g =0

on U, then either

0 on U

f ( P ) f= 0, then (since f is continuous)


f(z) f= 0 when z E D(P,r ) . Thus glD(P,r) = 0.

Proof. If P E U is a point where


there is an r

>

0 such that

By Corollary 3.6.2,

g = 0.

Corollary 3.6.5. Let U C be connected and open and let

be holomor

phic on U. If there is a P EU such that

for every

j,

then

J = 0.

Proof. There is a (small) D ( P, r ) U. Thus, for

f(z)
By Corollary 3.6.2,

f : (

j=O

0 on U.

P)

(z

!P)J

z
=

E D ( P, r ) ,

0.
D

3.6. The Zeros of a

93

Holomorphic Function

In Sections 1.2 and 3.3 we introduced definitions of sin


ez, for all

z,

cos

z,

and

E C. These were chosen in such a way that they agreed with

the usual calculus definitions when

z ER

One pleasing aspect of Corollary

3.6.3 is that such extensions are unique, as we state explicitly in the following
corollary :

Corollary 3.6.6. If f an d g are entire holomorpllic fu nctions


g(x) for all x E IR C, then J = g.

and

if f(x)

Proof. The real line, considered as a subset of C, has an accumulation point


in C. (Indeed, every point of IR is an accumulation point.)

Corollary 3.6.6 also shows that functional identities that are true for
all real values of the variable are also true for complex values of the variable.
For instance,
sin2

+ cos2

z=1

for all

z EC.

To check this in detail from our present point of view, we note that the
holomorphic function on C given by

g(z)
is identically

sin2

z + cos2 z

on R By Corollary 3.6.6, it follows that

on all of C.

Similarly, one can show that


cos 2z
sin

2z

cos2

sin2

2 cos z sin z

z EC ,

z,

z EC,

and so forth.
Corollary 3.6.6 also clarifies the situation regarding the function ez. In
Section 1.2 we defined ez to be ex(cos y + i siny) if

z=

x +iy,

assuming the

usual (real-variable) definitions for ex, sin x, and cos x. It is an easy exercise
to check that this function is holomorphic on C-by verifying the Cauchy
Riemann equations, for example (see Exercise
the right function values on IR since

49). This definition of ez gives


cos 0 + i sin 0 = 1. Thus our definition

of ez, which might have seemed artificial, is now revealed

as

in fact the

only possible holomorphic extension of the exponential function to all of C.


Moreover, the function defined by

Lo zl/j!,

which is holomorphic on all

of C (because the series has radius of convergence +oo), must be the same
as the function ex (cosy+i siny), because the two holomorphic functions are
equal on IR. Of course, one can also see from Theorem 3.3.1 that the power
series (centered at

0)

for ex(cosy +i siny) is

calculation of the derivatives at the origin.

I::0zJ/j!

by direct and easy

We also get "for free" the usual identities for the exponential function:

94

3. Applications

of the Cauchy Integral

z EC.
The way to prove that
for all z, EC is slightly more subtle:
For fixed both
and
are holomorphic functions of
Thus to
check that
for all
with z fixed, it suffices to show that
for all z, x E R. Now for fixed (but arbitrary) x E JR,
and
are both holomorphic functions of z, and these two holomorphic
functions agree when z E JR. Thus they are equal for all z, and we are done.
ez

z,

ew

ez+w

ez+w

ez

ez

ez+w

ew

ew

w.

w,

ez+x = ez ex
ez

ez+x

ex

The fact that identities on JR persist as identities on C was in past times


dignified by an elegant name: "The principle of the persistence of functional
relations." This name has fallen into disuse in our instance here, where the
whole phenomenon is now perceived as just a special case of holomorphic
functions having isolated zeros. But the idea remains important in relating
real (analytic) functions on JR to their holomorphic extensions to C and it
becomes more profound in situations involving general analytic continuation,
in the sense that we shall discuss in Chapter 10.

Exercises

1.

It was shown (Corollary 3.5.2) that if fj are holomorphic on an open set


U C and if
f uniformly on compact subsets of U, then

fj

--.

a
a
azfj az1
__...

uniformly on compact subsets of U. Give an example to show that if the


word "holomorphic" is replaced by "infinitely differentiable", then the
result is false.
2.

Let 'Y : [O, 1]

--.

be any C1 curve. Define


f(

z)

i ( z d( .

Prove that f is holomorphic on C \ )' where )'


{'Y( t) : 0 t 1}.
In case 1(t) t, show that there is no way to extend f to a continuous
function on all of C.
,

3.

4.

Explain why the following string of equalities is incorrect:


d2
d2
-1
tldt
log Ix tldt
dt.
log
Ix
2
dx
(x t)2
dx 2

11
-1

11
-1

11
-1

{fj}

Use Morera's theorem to give another proof of Theorem 3.5.1: If


is
a sequence of holomorphic functions on a domain U and if the sequence

Exercises

5.

6.
7.

8.
9.

10.
11.

95

converges uniformly on compact subsets of U to a limit function f, then


f is holomorphic on U.
( a ) Prove that if U C is open and connected and if p, q E U, then
there is a piecewise C1 curve from p to q consisting of horizontal
and vertical line segments. [Hint: Show that, with p E U fixed,
the set of points q E U that are reachable from p by curves of the
required type is both open and closed in U.]
(b) Deduce from ( a ) that the hypothesis f.., f(() d( = 0 in Morera's
theorem (Theorem 3.1.4) must be assumed only for closed curves
made up of horizontal and vertical line segments, as discussed in
the text after Theorem 3.1.4.
(c ) Let f be continuous on the entire plane and holomorphic on the
complement of the coordinate axes. Prove that f is actually holo
morphic on all of C.
Do Exercise 5 with "coordinate axes" replaced by "unit circle." What
if "coordinate axes" or "unit circle" is replaced by "smooth curve"?
Show that the conclusion of Morera's theorem still holds if it is assumed
only that the integral of f around the boundary of every rectangle in U
or around every triangle in U is 0. [Hint: It is enough to treat U a disc
centered at the origin. Then it suffices to show that the integral from
(0, 0) to (x, 0) followed by the integral from (x, 0) to (x, y) equals the
integral from (0, 0) to (0, y) followed by the integral from (0, y) to (x, y).
This equality follows from using two triangles.]
Show that in Exercise 7, "triangle" can be replaced by "disc." [Hint:
Use Green's theorem-cf. Exercise 44 of Chapter 2.]
Let Z:o akxk and Z:o bkxk be real power series which converge for
lxl < 1. Suppose that l:o akxk = l:o bkxk when x = 1/2, 1/3, 1/4 . . . .
Prove that ak =bk for all k.
Find the complex power series expansion for z2/(1 z2)3 about 0 and
determine the radius of convergence (do not use Taylor's formula).
Determine the disc of convergence of each of the following series. Then
determine at which points on the boundary of the disc of convergence
the series converges.
( a ) 2:3 kzk
(b) 2:2 klogk(z + l)k
( c ) 2:2(1og k)10gk(z 3)k
(d) l:0p(k) zk where p is a polynomial
( e ) 2:1 3k(z + 2i)k
(f) 2:2 24zk (Hint: Use summation by parts [RUDI].)
k
(g) l:o ke-kzk
-

Exercises

96

12.

(h) I:1 -b(z - 5)k


(i) I:1 k - kzk
Let g : [O, 1] ---t IR be a continuous function. Let E > 0. Prove that there
IR such that lg(x) - h(x)I < E for
is a real analytic function h: [O, 1]
all 0 x 1. [Hint: Think about Weierstrass's theorem. ]
---t

13. Let

f : (-1, 1)

IR be c=. Prove that

---t

is real analytic in some

neighborhood of 0 if and only if there is a nonempty interval

M > 0 such that i(d/dx)k f(x)I


all x E ( -fJ, fJ).

a constant
and

Mk k!

(-fJ, fJ) and


{1, 2, ... }

for all k E

14. Discuss the convergence of the power series

15. Give an example of a power series


complex value of

Lo akzk

which converges for every

and which sums to zero for infinitely many values of

but which is not the identically zero series.

16. Give an example of a nonconstant power series


verges for every value of

lxl

<

1.

But

which con

but which sums to zero for no value of z.

J(x)

17. The power series expansion for


only when

Lo akzk

1/(1 +x2) about x

0 converges

is real analytic on all of R. W hy does the

power series not converge at all values of IR?

18. Suppose that

: IR

---t

analytic. Prove that

f!)

f2 is real analytic, and f3 is real


analytic. ( Warning: Beware the zeros of

IR is continuous,

is real

19. Prove the case of Lemma 3.2.6 where


lim sup ia

k-++co

kil

/k

0.

20. Find the power series expansion for each of the following holomorphic
functions about the given point
3.3.l and

(ii)

by using the

(i)

proof of

by using the
Theorem

3.3.1.

of convergence of each series.

( a)
(b)
(c)
(d)

f(z) 1/(1 + 2z),


J(z) z2/(4- z),
J(z) = l /z,
f(z) (z - 1/2)/(1 - (1/2)z),
=

0.

i.

2 - i.
0.

statement of Theorem
Determine the disc

Exercises

21.

97

Prove that the function

J(z)

00

I: r1z(21)
10

is holomorphic on
is a

(2N)th

D(O, 1)

D(O, 1). Prove that if w


IJ'(rw)I = +oo. Deduce that J

and continuous on

root of unity, then limr_,1-

D(O, 1)

cannot be the restriction to

of a holomorphic function defined

on a connected open set that is strictly larger than

22.

D(O, 1).

Prove a version of l'Hopital's rule for holomorphic functions: If

J(z)
z-.P g(z)
lim

is an indeterminate expression for f and

g holomorphic (you must decide

what this means), then the limit may be evaluated by considering

of/oz
z-+P og/oz
lim

Formulate a precise result and prove it.

23.

TRUE or FALSE: Let f be holomorphic on


is a holomorphic polynomial on
polynomial on

24.

D(O, 1).

TRUE or FALSE: Let

aj

>

D(O, 1).

and assume that f2

Then f is also a holomorphic

j
L a1z is convergent on
l.:(aj + E)z1 is convergent

0, j = 1, 2, .... If

D(O, r) and if E > 0 is sufficiently


on D(O, r') for some 0 < r' < r.
25.

D(O, 1)

small , then

Define a notion of "real analytic function" of two real variables. Prove


that a holomorphic function of a complex variable is also a real analytic
function of two real variables.

26.

The functions fk(x) =sin kx are

[-1, 1],

C00

and bounded by

on the interval

yet their derivatives at 0 are unbounded.

Contrast this situation with the functions

fk(z) =sin kz on the unit


(o/oz)fk(O). Why are

disc. The Cauchy estimates provide bounds for


these two examples not contradictory?

27.

28.

Suppose that f and g are entire functions and that g never vanishes. If
IJ(z)I lg(z)I for all z, then prove that there is a constant C such that
J(z) =Cg(z). What if g does have zeros?
Let U
Let

<C be an open set.

Let f: U--+

<C be holomorphic and bounded.

U. Prove that

I:{ (P) I ; lfl,

where

is the distance of

to

s p

<C \ U .

98

Exercises

29. Suppose that

l D(O, 1)
:

---+

is holomorphic and that

Ill 2.

Derive

an estimate for

30. Let

C,

be an entire function and

k,

not depending on

C.

Prove that there is a constant

such that

( : ) kl(P) Ck!.
z

Can you improve this estimate? Is there necessarily a polynomial p(k)


such that

( : ) kl(P) l (k)I
p

k.

31. Fix a positive integer

l is an entire function such that,


of l is a polynomial. Prove that l is

Suppose that

for some k, the kth derivative l(k)


a polynomial.

l is bounded and holomorphic on C \ {0}.


is constant. [Hint: Consider the function g(z)
z2 l(z)
to apply Theorem 3.4.4.]
(a) Show that if l: D(O, r ) ---+ C is holomorphic, then

32. Suppose that

33.

l l(O)I
[Hint:

v 1r r

The function

{
ll(x, y)l2 dxdy
JD(O,r)

J2 is holomorphic too.

Prove that f
and endeavor

112

Use the Cauchy integral

formula to obtain

for

(b)

<

_.!_ { 21r l2(sei8)d0


27r l
o

<

integrate in
Let U

r.

12(0)

Multiply both sides by a real parameter

from

0 to r. ]

be an open set and let

l is holomorphic on
s p Il l

and

be a compact subset of U.

Show that there is a constant C (depending on U and

if

K)

such that

U, then

(fu ll(x, y)l2dxdy) 112

34. Complete this sketch of another proof of the Fundamental Theorem of


Algebra:
If p(z) is a nonconstant polynomial and p(z) never vanishes, then
h(z) = 1/p(z) is an entire function (hence, in particular, it is continuous)
which vanishes at oo. Therefore h has a maximum at some point

Po

C.

99

Exercises

Examine the Taylor expansion of h about Po to see that this conclusion


is impossible.

35. Complete the following outline to obtain R. P. Boas's proof [ORE] of


the fundamental theorem of algebra:
If p(z)
'I: =O akzk is a polynomial of degree N 2:
vanish, then let

Q(z)

ak z k

(t ) (t )
k=O

akzk

k=O

which does not

Q has degree 2N 2: 2 and Q takes real values on the real axis. Of


Q is never zero on IR. so the polynomial Q must be of one sign on
real axis. Say that Q > 0 on IR.. Now, with D the unit disc,
J
J z2N-1dz
dz
d()
Q(2 cos 0)
-i !a v zQ(z + z-1)
-i !a v Q(z)

Then

course
the

111"
-11"

where

Q(z) = z2N Q(z + z-1) .

Now

is a nonvanishing polynomial;

hence the integrand is holomorphic. Therefore

(* )

contradiction.

36. Let p(z)

ao +a1z + +anzn, an i- 0. P rove that if R >


large and if lzl
R, then lp(z)I 2: lanl Rn/2.

37. Let

0 is sufficiently

{Pj}

0 . This is a

be holomorphic polynomials, and assume that the degree of Pj

does not exceed N, all j and some fixed N. If

{Pj}

converges uniformly on

compact sets, prove that the limit function is a holomorphic polynomial

of degree not exceeding N.

38. If

fj :

U---+ C are holomorphic and

l!J I

2-j,

then prove that

00

L:Jj(z)

j=O

converges to a holomorphic function on U.

39. Let

cp: D(O,

1)---+ D(O, 1)

be given by

cp(z),

'Pj(z)

cpocpj-1(z),

and so forth. Suppose that

and

z + a2z2 + .... Define

cpocp(z),

{ 'Pj}

converges uniformly on compact sets.

cp?

are holomorphic, and if

the following outline to prove that

(i)

cp1(z)
cp2(z)

What can you say about

40. If

cp(z)

The result is easy when

f og is well-defined,
f og is holomorphic:

is a polynomial.

then complete

Exercises

100

(ii)
(iii)

Any

is locally the uniform limit of polynomials.

{!jog}
41. Let

fn

--t

f(zo).

fog.

f,

then

fn
f uniformly
If U 2 {zn} and Zn
zo E U, then prove that
the fj are also holomorphic and if 0 < k E Z, then

be continuous on the open set

and let

--t

--t

If

prove that

converges uniformly on compact sets to

converges uniformly on compact sets to

on compact sets.

fn(zn)

42. Let

{!j}

If the sequence

( :z )

fn(zn)

( :z )

--t

f(zo).

be holomorphic on a neighborhood of

not identically zero.

Prove that

D(P, r ) .

D(P, r ) .

Suppose that

is

has at most finitely many zeros in

43. Let E be a closed subset of R Prove that there is a continuous, real


valued function

on JR such that

{x
[Hint:

JR:

f(x)

=O} = E.

The complement of E is an open set, hence a countable union of

disjoint open intervals.]

* 44. If

f : D(O, 1)

--t

phic, then prove that

* 45. Suppose that

f2

is a function,

is holomorphic, and

f3

is holomor

is holomorphic.

is holomorphic on all of
lim

n--+oo

( Z)
a

and that

f(z)

exists, uniformly on compact sets, and that this limit is not identically
zero. Then the limit function Fmust be a very particular kind of entire
function. Can you say what kind ?

[Hint:

If Fis the limit function, then

F is holomorphic. How is F' related to F?]

fj be
pointwise

* 46. Let

U t;;;; C. If fj converges
f on U, then prove that f is holomorphic
of U. [Hint: Use the Baire category theorem to

holomorphic functions on an open set


to a limit function

on a dense open subset

locate points in a neighborhood of which the convergence is uniform.]

* 47. This exercise is for those who know some functional analysis. Let
be a bounded open set. Let

X = {!
If

X, then define

C(U) : f

llfll

is holomorphic on

=sup IJI.

U} .

U t;;;; C

101

Exercises

11 11 is a Banach
for any fixed PE U and any k E {O, 1, 2, ... } it holds

Prove that
that

equipped with the norm

is a bounded linear functional on


*

that the map

[)k f
(P)
[)zk
X.

48. Let U be a connected open subset of C and let


on U. Prove that the zero set of

space. Prove

f . 0

be holomorphic

is at most a countably infinite set.

49. Check directly, using the Cauchy-Riemann equations, that the function

ez
*

50. Let

(!

ex(cosy + isiny) is entire.

f and g be
g)(x) x.

entire functions. Suppose that when

Prove that

is the inverse of
*

51. Suppose that

J'(O)

1.

borhood of
function
of
*

and

x E JR.

it holds that

are one-to-one and onto and that

g.
is a holomorphic function on

Consider whether

D(O, 1)

by "inverting" the power series of

defined near

with

with

f(O)

0 and

is one-to-one and onto on some neigh

g(! (z))

to find a holomorphic

for all

in a neighborhood

0.

52. Prove that the composition of two real analytic functions (when the
composition makes sense) is real analytic. Be careful of the domain of
convergence of the composition.

53. Formulate and prove results to the effect that the sum, product, and quo
tient of functions defined by convergent power series are also functions
which are defined by convergent power series. Be careful of the domain
of convergence of the sum, product, or quotient.

(Suggestion:

This can

be done by thinking about holomorphic functions; no manipulation of


series is needed.)
*

54. Prove the following very simple version of the Cauchy-Kowalewski the
orem:
Consider the differential equation
'
y +
where

a(x), b(x)

b(x)

have convergent power series expansions in

a neighborhood of

L%:o akxk

a(x)y

0E R

Then there is a solution

power series converges) in a neighborhood of


*

55. Let

y(x)

of the differential equation which is defined (the

0.

f: (-1, 1) JR. be C00 Suppose that J(k)(x) > 0 for all x E (-1, 1)
k E {O, 1, 2, . .. }. Prove that f is real analytic on (-1, 1). [Hint:

and all

Examine the remainder term. See [BOA] for a detailed discussion.]


*

56. Prove that

mation

L%:i zk /k

converges when

by parts-see [RUDl].]

lzl

1, z i-

l.

[Hint:

Use

sum

102

Exercises

g : [O, 1] x re -t re is continuous and dif


x and y, with 8g/8x and 8g/8y continuous in all three

* 57. Suppose that the function

ferentiable in

variables jointly. Further assume that


a

az
for all values oft and

g(t, z)

z. Prove that
1
J(z)
g(t, z)dt
=

is entire. As a special case, notice that if

g(z)

C1 [O,1], then

11 <P(t)eitzdt

is entire. In this last example, show that we need only assume

continuous, not C1.

<P to be

0 :c:; k E Z. Give an example of a function f: D(O, 1) -t re such that


f E Ck(D) but f Ck+1(D). [Hint: The differentiability need only fail

* 58. Let

at a single point.]

f as in Exercise 58 with the additional property


that f is not ( k + 1) times differentiable at any point. [Hint: Let
be a suitable nowhere differentiable function on [O, 1]. Let <I> be a kth
antiderivative of , normalized so that <I>(j) (0) = 0, 0 :c:; j :c:; k. Define
f(z) <I>(lzl2).]
* 60. Complete the following outline to show that there is a C00 function
f: IR-t IR such that f(t) f= 0 for each t E (-1,1) but f(t) 0 if t :c:; -1
or t 1.
(a) Verify that the function
* 59. Give an example of an

<P(x)
is

{0

x :c:; 0
if x > 0
if

e-l/x2

C00 on all of IR. [Hint: Use l'Hopital's rule.]

(b) Verify that the function


J(x)

<P(x + 1) <P(-x + 1)

satisfies the desired conclusions.


* 61. What can you say about the zero sets of real analytic functions on IR2?

What topological properties do they have?

Can they have interior?

[Hint: Here, by "real analytic function" of the two real variables x and
y, we mean a function that can be locally represented as a convergent
power series in x and y. How does this differ from the power series
representation of a holomorphic function?]

Exercises

62.

63.

64.

65.

103

Let f be holomorphic on a neighborhood of D(O, 1). Assume that the


restriction off to D(O, 1) is one-to-one and f' is nowhere zero on D(O, 1).
Prove that in fact f is one-to-one on a neighborhood of D(O, 1).
Let ao, ai, a2, . . . be real constants. Prove that there is a C00 function
f defined on an open interval centered at the origin in such that the
Taylor series expansion off about 0 is Lo ajxj. This result is called
E. Borel's theorem. [ Hint: Construct C00 "bump" functions cpj-as in
Exercise 60-with the property that cp(j) (0) = 1 and all other derivatives
at the origin are 0. Arrange for the supports of these functions (i.e., the
sets where they do not vanish) to shrink rapidly as j ---+ +oo. Find a
way to add these together. See [BOA] or [FED] for further details.]
Is there a similar result for holomorphic functions in a neighborhood
of 0 EC?
Give an example of a C00 function f : (-1, 1) ---+whose Taylor expan
sion about 0 converges only at 0. [ Hint: Apply Exercise 63 to get a C00
function f with j(n)(O) = (n ! ) 2.]
Do Exercise 43 with the word "continuous" replaced by "C00".

Chapter 4

Meromorphic
Functions and Residues

4.1. The Behavior of a Holomorphic Function


Near an Isolated Singularity
In the proof of the Cauchy integral formula in Section 2.4, we saw that it is
often important to consider a function that is holomorphic on a punctured
open set U

\ { P}

C. The consideration of a holomorphic function with

such an "isolated singularity" turns out to occupy a central position in much


of the subject. These singularities can arise in various way s. Perhaps the
most obvious way occurs as the reciprocal of a holomorphic function, for
instance passing from z] to

1/ z], j

a positive integer.

More complicated

examples can be generated, for instance, by exponentiating the reciprocals


of holomorphic functions: for example,

e1/z, z i- 0.

In this chapter we shall study carefully the behavior of holomorphic


functions near a singularity.

In particular, we shall obtain a new kind of

infinite series expansion which generalizes the idea of the power series ex
pansion of a holomorphic function about a (nonsingular) point. We shall in
the process completely classify the behavior of holomorphic functions near
an isolated singular point.
Let U

C be an open

set and

U. Suppose that f

holomorphic. In this situation we say that f has an


(or

isolated singularity)

at

P.

: U \ { P} --+ C is
isolated singular point

The implication of the phrase is usually just

that f is defined and holomorphic on some such "deleted neighborhood"


of

P.

The specification of the set U is of secondary interest; we wish to

consider the behavior of f "near

P" .
-

105

4. Meromorphic Functions and Residues

106

There are three possibilities for the behavior of

f nearP that are worth

distinguishing:

(i) lf(z)I

D(P, r ) \ {P} for some r

is bounded on

U; that is, there is some

M for allz

>

and some M

D(P, r) \ {P}.

D(P, r)
> 0 such that lf(z)I
>

with

E U n

(ii)

limz-.P

lf(z)I =

(iii)

Neither

(i) nor (ii)

+oo.
applies.

Of course this classification does not say much unless we can find some

f related to (i), (ii), and (iii). We shall prove momen


(i) holds, then f has a limit atP which extends f so that

other properties of
tarily that if case

it is holomorphic on all of U. It is commonly said in this circumstance that

f has a removable singularity at P. In case (ii), we will say that f has a pole
atP. In case (iii), f will be said to have an essential singularity at P. Our
goal in this and the next section is to understand (i), (ii), and (iii) in some
further detail.

Theorem 4.1.1 (The Riemann removable singularities theorem).

D(P, r ) \ {P}---+ C
(1)

limz-.P

L et

f:

be holomorphic and bounded. Then

f(z)

(2) the function

exists;

f:D(P, r ) ---+ C
f(z) =

{ f(z)f(()

defined by

lim
c,-.P

if
if

z =I p
z =P

is holomorphic.
Remark: Notice that, a priori, it is :r;_ot even clear that limz-.P f(z) exists
or, even if it does, that the function

f has any

regularity at

beyond just

continuity.

Proof of Theorem 4.1.1. Define a function

( )=
gz
We claim that

{ (z -P)2 f(z)

C1(D(P, r)).

if
if

g:D(P, r ) ---+ C by

z ED(P, r ) \ {P}
z=P.

Assume this claim for the moment.

Since

ag = a
!=0
(z -P)2 f(z) + (z -P)28Oz
az
on D(P, r) \ {P}, continuity of the first partial derivatives implies
8g/0z = 0 onD(P, r) and hence that g is holomorphic onD(P, r).
Oz

that

4.1. Behavior near an Isolated Singularity

107

M such that l f(z)I ::::;; M for all


z E D(P, r) \ {P}. Hence lg(z)I ::::;; Mlz-Pl 2 on D(P, r). It follows that the
power series expansion of g about P has the form
00
g(z) L:a1(z-P)1 .
j=2
By hypothesis, there is a constant

r by
00
L a1(z - P)1-2.
j=2

This series has radius of convergence at least

H(z)
Then

phic extension

In conclusion, the

f of f to D(P, r).

It remains to show that

D(P, r) and equals g(z)/(z - P)2


function H gives the desired holomor

is a holomorphic function on

f(z) when z =/= P.

then

Theorem 3.3.1. Set

g is C1. This is apparent on D(P, r) \ {P} since


C00). Now notice that

is holomorphic ( hence

og
OX

( since f

(P)

lim
lR3h-+O

g(P + h) -g(P)
h

lim
h-+O

h f(P + h)

is bounded ) . Likewise

og
(P)
oy

0.

Thus to check the continuity of the partial derivatives of


that

(z)
lim
z-+P uX

we need to see

(z).
lim
z-+P uy

We check the first; the second is similar.


By hypothesis, there is a constant

{P}. If zo E D(P, r /2) \ {P},


D(zo , lzo-Pl), yield that

Since

is holomorphic at

I (zo)I ::::;; lzo Pl

zo ,

Proposition 1.4.3 yields that

I (zo)I ::::;; I zo

Therefore

M such that If(z)I ::::;; M on D(P, r) \

then the Cauchy estimates, applied on the disc

og
(zo)
ox

I l
=

og
(zo)
oz

pI

l2(zo-P)f(zo) + (zo-P)2f' (zo)I

::::;; 2lzo-Pl M + lzo-Pl 2

M
lzo - Pl.

-t

og
-(P)
ox

108

as zo

4.

P.

Meromorphic Functions and Residues

This completes the proof of the theorem.

The behavior described by Theorem 4.1.1 is so startling that some


examples are in order.
Consider the function f(z) sin(l/lzl). Then f E C00(C\
{O}) and f is bounded, yet f cannot be extended even continuously to
D(O , 1). Thus, as usual, the C00 case contrasts with the holomorphic case.
EXAMPLE 4.1.2.

Set f(z) = e1/z. Then f is holomorphic on D(O, 1) \ {O}.


Let us consider its behavior near 0. We claim: For any o: E C \ {O} and
any f. > 0, there is a z with 0 < lzl < f. such that e1/z = o:. To see this,
choose first a complex number w such that ew = o:. Then ew+2nik = o: for all
k = 0, 1, 2, 3 .... Fork large, w+2nik I- 0, and we may set Zk 1/(w+2nik).
Then
EXAMPLE 4.1.3.

Also limk_,+ Zk 0. So we can choose a Zk with lzkl < f.. The point of this
00
example is that e1/z behaves very wildly near 0. It oscillates enough to hit
each value except 0 infinitely often on even a small punctured disc around
0. In particular f is not bounded in any neighborhood of 0.
=

The wild behavior exhibited by e1/z is in fact typical of holomorphic


functions near an essential singularity. The following result makes this ex
plicit. For the statement of the result, recall that a set S i:;;;; C is defined to
be dense in C if the closure of S in C is actually all of C.
(Casorati-Weierstrass). If f: D(P, ro ) \ {P}
C is holo
morphic and Pis an essential singularity off, then J(D(P, r ) \ {P}) is dense
in C for any 0 < r < ro.
Theorem 4.1.4

Proof. In fact, the occurrence of r in the statement of the theorem, as


given, is redundant (for purposes of emphasis). It is enough to prove the
assertion for r ro.
=

Suppose that the statement of the theorem fails for


is a .X E C and an f. > 0 such that
lf(z) - -XI

ro.

Then there

> f.

for all z E D(P, ro ) \ {P}. Consider the function g : D(P, ro ) \ { P}


C
defined by
1
g(z)
J(z) - .X
Then g is holomorphic on D(P,ro) \ {P} since f(z) - A is nonvanishing on
D(P,ro) \ {P}. Moreover,
1
lg(z)I < f.

4.2. Expansion around Singular Points

for all

D(P, ro) \ {P} by ( * ) .

109

By the Riemann removable singularities

g: D(P, r0) ----+ C such that g(z)


D(P, ro) \ {P}. Notice that the only point in D(P, ro) at
vanish is at P itself. Thus we may solve the equation

theorem, there is a holomorphic function

g(z)

for all

which

g could

g(z)
z

away from the point

g(z)

l
=

f(z)

,\

and obtain

f(z)

,\+ g(z)"

Now there are two possibilities: Either

g(P)

0 or

g(P) =/= 0.

In the first

case,
lim

z--+P

In other words,

lf(z)I

lim

z--+P

has a pole at

I,\+ g(l ) I +oo.


=

P [case (ii)

for the three possibilities for a

(** ) is
f has a removable singularity

singularity ] . In the second case, we see that the right-hand side of


holomorphic on all of

D(P, ro);

at

P.

P,

and that is a contradiction.

in other words,

In either case, we see that

does not have an essential singularity at

Now we have seen that, at a removable singularity

P,

a holomorphic

f on D(P, ro) \ {P} can be continued to be holomorphic on all of


D(P, ro). Also, near an essential singularity at P, a holomorphic function
g on D(P, ro) \ {P} has image which is dense in C. The third possibility,
that h has a pole at P , has yet to be described. This case will be examined
function

further in the next section.


Functions with poles and singularities at an isolated point
amenable to rather pleasant looking series expansions about

P are always
P ( these series

are called "Laurent series" ) . These will be considered in the next section as
well.

4.2. Expansion around Singular Points


To aid in our further understanding of poles and essential singularities, we
are going to develop a method of series expansion of holomorphic functions
on

D(P, r ) \ {P}.

Except for removable singularities, we cannot expect to

expand such a function in a power series convergent in a neighborhood of

P,

since such a power series would define a holomorphic function on a whole

neighborhood of

P,

including

itself.

A natural extension of the idea of

power series is to allow negative as well as positive powers of

(z

P).

This

extension turns out to be enough to handle poles and essential singularities

110

4. Meromorphic Functions and Residues

both. That it works well for poles is easy to see; essential singularities take
a bit more work. We turn now to the details.

Laurent series on D(P, r)

is a (formal) expression of the form

+oo

j=-oo

aj(z - P)J.

Note that the individual terms are each defined for all

D(P, r) \ {P}.

To discuss Laurent series in terms of convergence, we must first make


a general agreement as to the meaning of the convergence of a "doubly
infinite" series
and

L.t::._00 aj.

We say that such a series

converges if L.t aj

L.t:t a_j converge in the usual sense. In this case, we set

You can check easily that


only if for each E >

I (L, =-k Oj)-a I <

0
E.

L.: aj converges to a complex number a if and

there is an N >

such that, if f N and k N, then

It is important to realize that f and k are independent

here. [In particular, the existence of the limit limk-++oo L.


imply in general that

L.: aj converges.

See Exercises

t:-k aj does not

10

and

11.]

W ith these convergence ideas in mind, we can now present the analogue
for Laurent series of Lemmas 3.2.3 and 3.2.5 for power series.

Lemma 4.2.1.

If'L.t::._00 aj(z - P)J

and if lz1 -Pl< lz2-Pl ,


lz - Pl< lz2- Pl.
Refer to Figure

4.1

at z1 =/= P and at z2 =/= P


converges for all z with lz1 -Pl<

converges

then the series

for an illustration of the situation described in the

Lemma.

L.t::._00 aj(z2-P)J converges, then the defini


tion of convergence of a doubly infinite sum implies that L.t aj(z2- P)J
converges. By Lemma 3.2.3, L.t aj(z-P)J then converges when lz-PI<

Proof of Lemma 4.2.1. If

L..j+oo
=l a_j(z1 -P)-J.
L..j+oo
=-oo aj(z1 -P)J converges, then so does
lz2-PI. If
Since 0< lz1 -Pl< lz-PI, it follows that 11/(z-P)I< ll/(z1 -P)I. Hence
j
Lemma 3.2.3 again applies to show that L.j:t a-j(z-P)- converges. Thus
L.: aj(z -P)J converges when lz1 - Pl< lz - Pl< lz2- Pl.
D

From Lemma 4.2.1, one sees easily that the set of convergence of a
Laurent series is either a set of the form

{z : 0 ::::; r1 < lz - Pl <


r 1 or lzl

together with perhaps some or all of the points satisfying lzl

r2},
=

r2,

111

4.2. Expansion around Singular Points

,,.

.,., -

I
I

I
I
I
\

--

..... ......_z2
'

.,., --- .....

'

I
I

',_
'

'

' .....

l
.,.,

\
I
I
I

\'

'

\
I
I

'

....... ____ _,,,,

Figure 4.1

{z

: 0

lz -Pl < +oo}, together with perhaps


some or all of the points satisfying lzl
r1.
An open set of the form
{z : 0 r1 < lz -Pl < r2 < +oo} or {z : 0 r1 < lz -Pl} is called an
annulus centered at P. We shall let
or a set of the form

r1 <

D(P, +oo)

{z: lz-PI

<

+oo}

and

D(P,O)

{P}

so that all annuli can be written in the form

D(P, r2 ) \ D(P, r1 ) ,

r1 < r2

+oo.

In precise terms, the "domain of convergence" of a Laurent series is given


by the following lemma:

Lemma 4.2.2. Let

j=-oo
be a doubly infinite series that converges at (at least) one point. There are
unique nonnegative numbers r1 and r2 (r1 or r2 may be

+oo)

such that the

z with r1 < lz -Pl < r2 and diverges


lz -Pl < r1 or lz -Pl > r2. Also, if r1 < r r < r2, then
'L,j::_00 laj(z -P)JI converges uniformly on {z : r lz -Pl rD and
consequently 'L,j::_00 aj(z - P)J converges uniformly and absolutely there.

series converges absolutely for all


for

with

The first statement follows immediately from Lemma 4.2.1.

The second

statement follows from the corresponding result for power series combined
with the argument used to prove Lemma 4.2.1.

112

4. Meromorphic Functions and Residues

EXAMPLE 4.2.3. The Laurent series

oo l
I: z2

j=-10

#0

lzl

converges absolutely for all 0 <

<

1,

diverges for

izl

>

1,

and converges

absolutely at all points of the unit _circle.


The Laurent series

zl
L,,

J=-(X)

j2

#0

- oo )

(same summands, but the lower limit is

{z: izl

l}.

The Laurent series

converges only on the circle

50

I: 21(z + i)j
j=-00

lz +ii > 1/2.


also diverges on the circle { z : I z + iI

converges absolutely for

It diverges on the disc


=

1/2}.

D(-i, 1/2) and

It follows from the definition of convergence of a Laurent series and from


Lemma

4.2.1

that the function defined by a Laurent series on its annulus

of convergence is the uniform limit, on compact subsets, of a sequence of


holomorphic functions.

Therefore the infinite sum is itself a holomorphic

function on that annulus (by Theorem

3.5.1).

Our project in the next section

is to prove the converse: that any holomorphic function on an annulus is


given by a Laurent series that converges on that annulus (it may in fact
converge on an even larger region). Right now we can prove that there is at
most one such expansion:

Proposition 4.2.4 (Uniqueness of the Laurent expansion). Let 0 :-::;

r2 :-::; oo.
D(P, r1)
j E 7!..,

If the Laurent series "Lt::_00 a1(z - P)l

to a function

f,

a1
In particular, the

then, for any

D(P, r2) \
r2, and each

converges on

satisfying

r1 <

<

f .
_l 1
(() d(.
27ri 11<-Pl=r (( - P)1+1

a1 's are uniquely determined by f.

Proof. The series converges uniformly on the circle


follows that

r1 <

d(
11<-Pl=r (( - P)1+1
(()

{z

I z - PI

ak(( - P)k-1-1]

[
11<;-Pl=r k=-oo
J

d(

r}.

It

4.3. Existence of Laurent Expansions

113

+oo
(( - P)k-j-Id(.
2: a J
k=-oo k 11(-Pl=r

Here we have commuted the sum and the integral using a result from Ap
pendix A.

( cf. Section 2.5)


0
f k - - 1 # -1 ( i.e.,
P)k-j-Id(=
2m if k- J - 1 = -1 ( i.e.,

By explicit calculation

((

11(-Pl=r

Hence

i1(-Pl=r

k#

k=

j)
j).

+ d(= 2maj.
(( - P) i
f(()

4.3. Existence of Laurent Expansions


We turn now to establishing that convergent Laurent expansions of functions
holomorphic on an annulus do in fact exist. We will require the following
result.

Theorem 4.3.1

( The

that 0 r1 < r2 +oo and that f: D(P,r2) \D(P,ri)


Then, for each s1,s2 such that r1 < s1 < s2 < r2 and
D(P,s1), it holds that

f(z) =

D(P,s2) \ D(P,s1).

9z(() =
Then

9z

--4

f(() d( 1

1
27l'Z 1l(-Pl=s2 ( - Z
27ri 11(-P l =si

Proof. Fix a point

D(P,r1),

Cauchy integral formula for an annulus . Suppose

C is holomorphic.

each

f(()

(-

Define, for

D(P,s2) \

d(.
E

D(P,r2) \

f(() - f(z)

(- Z

f'(z)

is a holomorphic function of

(, (

(= z.
D(P,r2) \ D(P,r1)

( by

Riemann removable singularities theorem .


Now we consider the integrals

i(-Pl=s1 gz(() d(
and

1l(-Pl=s2
By the considerations in Section

1l(-Pl=s2

2.6,

9z(() d(.

these two_ integrals are equal. So

9z(() d( - J

11(-Pl=si

9z ( ( ) d(

the

114

4.

Meromorphic Functions and Residues

Figure 4.2

f(() - f(z) d(

li<-Pl=s2

- Z

f(() d(

li<-Pl=s2 ( - Z
=

Now

li<-Pl=s2 ( - Z

li<-Pl=s1

f(z) d(

f(z) J

li<-Pl=s2 ( - Z

- Z

f(z) d(

f(() d(

li<-Pl=s1

Hence

f(() - f(z) d(.

- Z

li<-Pl=s1

li<-Pl=s2

f(z) d(.

- Z

1
-- d(
(-Z

27rif(z)

by the Cauchy integral formula for the constant function 1 on D(P, r2 ) (or
by direct calculation).
Also

Ji<-Pl=s1

f(z) d(

- Z

f(z) J

li<-Pl=s1

1
__ d(
(-Z

0.

This can be seen from the Cauchy integral theorem (Theorem 2.4.3 ) since
1/(( - z) is holomorphic for ( E D(P, lz - Pl) and {( : I( - Pl s1}
D(P, lz - Pl). See Figure 4.2.
So

27rif(z)
as

desired.

li<-Pl=s2

f(() d( - J

- Z

li<-Pl=s1

f(() d(,

- Z

115

4.3. Existence of Laurent Expansions

Now we have our main result:


Theorem 4.3.2 (The existence of Laurent expansions). IfO r1

and

f D(P,r2) \ D(P,r1)

numbers aj such that

converges

<

r2

oo

- C is holomorphic, then there exist complex

+oo
L aj(z - P)j
j=-oo

on D(P,r2) \ D(P,r1)

to

f.

converges absolutely and uniformly on

Ifr1

< s1 < s2 <

r2,

D(P, s 2) \ D(P, s1 ) .

then the series

Proof. If 0 r1 < s1 < lz - Pl < s2 < r2, then the two integrals on the
right-hand side of the equation in Theorem 4.3.1 can each be expanded in a
series. For the first integral we have

1l(,-Pl=s2

!(() d(
(-Z

l
J(
.
p ( _d(
1(,-P - p

1l(,-Pl=s2

1lc.-Pl=s2

(z - P)J
J(()
d(
( - p j=O (( - P)J

J(()(z - P)J d(

1lc.-Pl=s2

(( - P)]+l

where the geometric series expansion of


1

- (z - P)/(( - P)

converges because lz - Pl/I( - Pl = lz - Pl/s2 < l. In fact, since the value


of l(z - P)/(( - P)I is independent of(, for I( - Pl = s 2, it follows that the
geometric series converges uniformly.
Thus we may switch the order of summation and integration (see Ap
pendix A) to obtain

1lc.-Pl=s2
For

s1 <

f(() d(
(-Z

J(()
J
(

+id( ) (z - P)l.
j O 1lc.-Pl=s2 (( - P)J

lz - Pl, similar arguments justify

11(,-Pl=si

J(()
d(
(-Z

the formula

_ 11(,-Pl=si

J(() . l
__d(
1 - = Z - p

- J

J(()
z-P

11(,-Pl=si

(( - P)J

(z - P)J d(

11 6

4. Meromorphic Functions and Residues

j=O [llJ(-Pl=s1
- I:

f(() ( ( - PF d( (z - P)

=-1
f(()
JL Ji - d(
1
(( P)J+l
J=-oo I( Pl-s1 -

[1

Thus

j=-1
f(()
L Ji1 .=(( P)J+l
J oo I(. Pl-s1 -

27rif(z)

[1
L [t

as

(z - P F .

d( (z - PF.

l
l

d( (z - PF'
((- P)J+l

oo

j- i

J(()

j=O l(-Pl-s2
-

desired. Notice that the results of Lemma 4.2.2 (see also the facts about

power series in Section 3.2) imply that the convergence of the series is uni
form as asserted in the theorem. This completes the proof.

At first, the series expansion derived at the end of the proof of Theorem
4.3.2 might not seem to fit the desired conclusion: for it seems to depend
on the choice of s1 and s2. However, Proposition 2.6.5 yields that in fact
the series expansion is independent of s1 and s2. In fact, for each fixed
j

0, 1, 2, ..., the value of

llj

1
_

27ri

1
11( Pl= r
-

!(()
J+ d(
(
( - P) l

is independent of r provided that r1 < r < r2.


Yet another way to look at the independence from s1, s2 of the coef
ficients of the Laurent series is by way of Proposition 4.2.4-see Exercise
2 6.

D(P, r ) \ {P}

Now let us specialize what we have learned about Laurent series ex


pansions to the case of

D(P, r ) \ {P}

---t

holomorphic, that is, to

holomorphic functions with an isolated singularity:


Proposition 4.3.3. If f:

unique Laurent series expansion


00

f(z)

---t

is holomorphic, then

aj (z - P)

aj =

1
2 7rZ.

has a

D(P, r ) \ {P}. The convergence is uniform


D(P, r ) \ {P}. The coefficients are given by
j=-oo

which converges absolutely for z E


on compact subsets of

8D(P,s) ((- P)J+l

f(()

d(,

any 0 < s < r.

117

4.3. Existence of La urent Expansions

There are three mutually exclusive possibilities for the Laurent series

4.3.3:

of Proposition

(i) aj
(ii) for
(iii)

0 for all j < O;

some k

neither

>

0, aj

nor

(i)

0 for all -oo < j < -k;

applies.

(ii)

These three cases correspond exactly to the three types of isolated


singularities that we discussed in Section

is a removable singularity; case

case

(iii)

occurs if and only if

(ii)

4.1:

Case

(i)

occurs if and only if

occurs if and only if

is a pole; and

is an essential singularity. These assertions

are almost obvious, but not quite. Therefore we shall discuss them in detail,
each implication in turn.

Removable Singularity: The Laurent series is a power series


P which converges on D(P,r). This power series con
verges to a holomorphic function on D(P, r) which equals f on
D(P,r) \ {P}; hence lfl is bounded near P.
Removable Singularity ::::} (i): Assuming that Pis removable, we need
to see that aj
0 for j < 0, where f(z)
I: aj(Z - P)J is the Lau

rent expansion for f. If f is the holomorphic extension of f to P


(Theorem 4.1. l), then f has a power series expansion I:: bj(z - P)J
j
(i)

::::}

centered at

that converges on

D(P,r)

since

f is

the uniqueness for Laurent series (Proposition


j 0 and

(ii)

::::}

Pole:

aj

If k

D(P,r).
4.2.4), aj
bj

holomorphic on

0 for j < 0.

By
for

0 and

>

+oo
f(z)

L aj(z - P)l

j=-k
with

a_k -=/=

l/(z)I

2':

0, then

lz - Pl-'

Now

l-I -

+oo
''

lim
z->P

aj(z - P)J+k

j=-k+l
+k is a power series with a positive
because L:: l k+I aj(z - P)J
radius of convergence (Exercise 22 asks you to verify this assertion).
Hence

lf(z)I
for all

k
lz - Pl- (lakl - lakl/2)

close enough to P. It follows that limz_,p lf(z)I

+oo.

118

4. Meromorphic Functions and Residues

Pole:::::} (ii):
less than

Since limz_,p IJ(z)I

such that

g
Then limz_,p

+oo, there is a positive numbers

is never zero on

1/f: D(P,s) \ {P}

g(z)

Consider

C.

By the Riemann removable singularities

0.

D(P, s) \ {P}.

theorem, the function

H(z)

is holomorphic on

{ g(z)

on
at

D(P, s).

D(P, s) \ {P}
P

Hence, for some unique

positive

it

holds that

H(z)

(z - P)m Q(z)

D(P, s) such that Q(P) =/: 0.


D(P, s) is at P, it follows that Q
does not vanish on all of D(P, s). Therefore 1/Q(z) is a holomorphic
function on D(P, s) and has a convergent power series expansion
for some holomorphic function

on

In fact, since the only zero of Hon

+oo

::::>j (z - P)J.

j=O
For 0

<

Iz - Pl
f(z)

<

s,
1
H(z)

1
(z - P)-m . _ _
Q(z)
(z - P)-m
+oo

j=-m

(I:
1=0

bj(z - P)j

bj+ (z - P)j.
m

By the uniqueness of the Laurent expansion, this last series coin

cides with the expansion of

on

{z

: 0 <

lz - Pl

< r

},

and

(ii)

holds.

(iii)

:?

Essential Singularity:

The equivalence of

(iii)

and

being

an essential singularity follows immediately since they are the only


remaining possibilities.
If a function

has a Laurent expansion


00

f(z)

j=-k

aj(z - P)J

4.4. Examples of Laurent Expansions

for some

order k

119

a_k =I
that f has

0 and if

>

at P. Note

f has a pole of
k at P if and only if

0, then we say that


a pole of order

(z - P)k f(z)
is bounded near

but

(z - P)k-l f(z)
is not.

4.4. Examples of Laurent Expansions


In this section we consider several examples which illustrate techniques for
calculating Laurent expansions. First, some terminology:
pole at

P,

expansion of

around

principal part of f

the

00

f(z)

for

near

When

has a

it is customary to call the negative power part of the Laurent

P,

then the

aj(z-P)j

at P. That is, if

j=-k
principal part of f at P is
-1
L aj(z-P)J.
j=-k

Next, we give an algorithm for calculating the coefficients of the Laurent


expansion:

Proposition 4.4.1. Let f be holomorphic on D(P, r) \ {P} and suppose


that f has a pole of order k at P. Then the Laurent series coefficients aj of
f expanded about the point P, for j -k,-k+1,-k+2, ..., are gi ven by
=

(a8 ) k+j ((z-P)k f)

the formula

aj
Proof.

1
(k+j)!

This is just a direct calculation with the Laurent expansion off and

is left as an exercise (Exercise


EXAMPLE 4.4.2. Let

f(z)

Laurent series expansion for

(i)

I z=P

Let

g(z)

29)

z/(z -1) on D(l, 2) \ {1 }.


f about 1 in two ways:

(z-1) f(z).
1 is

Then

1+ (z -1).

Therefore

f(z)

g(z)
z -1

We shall derive the

g is holomorphic and

expansion about

g(z)

for you.

(z -1)-1+1.

its power series

120

4. Meromorphic Functions and Residues

Thus the principal part of

(ii)

at

(z - 1)-1.

is

Using Proposition 4.3.3, we have

aj

If j ;::::

-1,

1
_
27ri
1
_
27ri

(j

aj

<

-1,

(*)

then the Cauchy integral formula tells us that this last

line is

If j

f(().
d(
(
JaD(l,1) ( - l)J+l
J
( . d
(.
(
!av(1,1) ( 1)3+2

j 1
(
8) +
+ 1)!. a
(()
(

if j

-1, 0

otherwise.

then the integrand in

0. Thus

J (z)

(=l

.
+oo
""" aj (z - 1)3
L.J
j=-oo

(* )

is holomorphic on D ( 1, 2) so

1 (z - 1 )-1 + 1.

J (z) ez /[ (z - i)3 (z + 2)2 ] . Clearly f is holomorphic


except at z
i and z -2. Now limz-+i l (z -i)2 J (z)I +oo and (z -i)3 J (z)
is bounded near i, so J has a pole of order 3 at i. Likewise, f has a pole of
order 2 at -2. So the principal part of f at i is of the form
-1
L aj (z - i)j
j=-3
EXAMPLE 4.4.3. Let

and the principal part at

-2

is of the form

-1

j=-2

aj(z+2)J.

a_ =f. 0 since limz_.i(z - i)3 J (z) =f. 0. But we


3
a-2 , a_1, and so forth except by computing
the second case, a-2 =f. 0 but we know nothing

In the first case we know that

can draw no conclusions about


them explicitly. Likewise, in
about

a_1

without explicit computation.

To compute the principal part of

a_

f about i,

(8/oz)-3+3 z - i)3 J (z))


=

(z + 2)2 z=

we use the formula

z=

121

4.4. Examples of Laurent Expansions

(2+ i)2"
(8 /8z)-2+3 ( - i)3f(z))

it

a_2

:z Cz : 2)2 l z=i
z

(z+2)2

l z=i

ez - ez . 2(z+2)

(z+2)4

I z=i

\
l z=i
2
1 0 ( ez )
2 8z2 (z+2)2 I =i
z
1 8 ( zez )
2 oz (z 2)3 I =i
z
(2+ i)3.

(8 /oz)-1+3 z - i)3f(z))

a_1

(z+2)3 (zez + ez ) zez 3(z+2)2

(z+2)

ei

1
.
2 (2+ i)4
Thus the principal part of

z=i

f at i is

(z - i)-1.
(z - i)-2+
(z - i)-3 +
(2+ i)3
(2+ i)2
2(2+ i)4
EXAMPLE 4.4.4. Let

xl <

1,

IYI

<

1}.

f(z)

ez / sin z.

Restrict attention to U

{x + iy

Since
sin z

2i

eix-y

e-ix+y

2i
it is clear that sin z can vanish only if

e-y

ieix-y I

ie-ix+y I

eY

or

y
But when

0.

0, then sin z reduces to the familiar calculus function

sinx, which vanishes only when x is an integral multiple of 71". Thus, on U,


sin z vanishes only at

0. So

is holomorphic on U except possibly at

4.

122

Meromorphic Functions and Residues

(8/8z) sin z = cos z and cos(O) -=f. 0, it follows that sin z has a
zero of order 1 at z = 0. Since e0
1 -=f. 0, it follows that the functionf has
a pole of order 1 at 0. We compute the principal part:

0. Since

(i_)-l+l ( (z . ez)
=

O)
.
filnZ

8z
1.
z ez
im -

z--+O

sinZ

lim

z(l + z +
)
z - z3 /3! +

z--+O

1.

So the principal part off at 0 is

1
1 z- .

4.5. The Calculus of Residues


In the previous section we focused special attention on functions that were

D(P, r ) \ {P}.
: D(P, r ) \ {P} -+

holomorphic on punctured discs, that is, on sets of the form


The terminology for this situation was to say that f

C had an isolated singularity at

P.

It turns out to be useful, especially

in evaluating various types of integrals, to consider functions which have


inore than one "singularity" in this same informal sense.

More precisely,

we want to consider the following general question: Suppose that f

{P1, P2, ..., Pn}-+ C is

U\

U C with
P1, P2, ..., Pn removed. Suppose further that
1
'Y: [O, 1] -+ U \ {P1, P2, ..., Pn} is a piecewise C closed curve. Then how is
f related to the behavior off near the points P1, P2, ..., Pn?
a holomorphic function on an open set

finitely many distinct points

"'Y

The first step is naturally to restrict our attention to open sets

for which

"'Yf

is necessarily 0 if

P1, P2, ...Pn

are removable singularities

of f. W ithout this restriction we cannot expect our question to have any


reasonable answer. We thus introduce the following definition:

Definition 4.5.1. An open set

(abbreviated h.s.c.) if
f:

U-+ C,

UC

is holomorphically simply connected

is connected and if, for each holomorphic function

there is a holomorphic function F:

The connectedness of

U-+ C

such that F'

U is assumed just for convenience.

=f.

The important

part of the definition is the existence of holomorphic antiderivatives F for

each holomorphicf on U. The following statement is a consequence of our

. earlier work on complex line integrals:

Lemma 4.5.2. A

if and only if for

connected open set U is holomorphically simply connected


holomorphic functionf : U -+ C and each piecewise

each

4.5.

The Calculus of Residues

123

1
C closed curve r in U,

=0.

Note: Holomorphic simple connectivity turns out to be equivalent to another


idea, namely to the property that every closed curve in U is continuously
deformable to a point, in the sense discussed briefly in Section 2.6. This
latter notion is usually called just "simple connectivity." In Chapter 11, we
shall prove that this topological kind of simple connectivity is equivalent
to holomorphic simple connectivity. Then we can drop the word "holomor
phic" and just refer to simple connectivity unambiguously. But, since we
are deferring the proof of the equivalence, it is best to use the terminology
"holomorphic simple connectivity" for now to avoid circular reasoning. Fi
nally, we note that the phrase "holomorphic simple connectivity" is special
to this book and is used here primarily for expository convenience. In stan
dard mathematical terminology, the phrase "simple connectivity" is used
only to refer to the deformability of all closed curves to a point, and the
phrase "holomorphic simple connectivity" is not in general use at all.
We know from previous work that discs, squares, and C itself are all
holomorphically simply connected (see Exercise 45 for further examples).
On the other hand, D(O, 1) \ {O} is not: The holomorphic function 1/z has
no holomorphic antiderivative on that set.
We are now prepared to state the theorem on integrals of functions
that answers the question posed at the beginning of this section:
Theorem 4.5.3

(The residue theorem). Suppose that U C is an h.s.c.


open set in C, and that P1, ..., Pn are distinct points of U. Suppose that
f : U \ {P1, ..., Pn} ---+ C is a holomorphic function and r is a closed,
1
piecewise C curve in U \ {P1, ..., Pn} Set
Rj

the coefficient of (z - Pj)-1


in the Laurent expansion off about Pj .

Then
(4.5.3.1)
Before giving the proof of the theorem, we shall discuss it and its context.
The result just stated is used so often that some special terminology is
commonly used to simplify its statement. First, the number Rj is usually
called the residue off at Pj , written Res1(Pj). Note that this terminology
of considering the number Rj attached to the point Pj makes sense because
Res1(Pj) is completely determined by knowing fin a small neighborhood of

124

4.

Meromorphic Functions and Residues

Pj. In particular, the value of the residue does not depend on what type the
other points Pk, k i- j, might be, or on how f behaves near those points.
The second piece of terminology associated to Theorem 4.5.3 deals with
the integrals that appear on the right-hand side of (4.5.3.1).
Definition 4.5.4. If 'Y : [a, b] ----t C is a piecewise C1 closed curve and if
P rf. -;y 'Y([a, b]), then the index of 'Y with respect to P, written Ind7(P), is
defined to be the number
1
1
-- d(.
_
27ri 'Y ( - p
=

The index is also sometimes called the "winding number of the curve 'Y
about the point P." This is appropriate because, as we shall see in Chapter
11, Ind7(P) coincides with a natural intuitive idea of how many times 'Y
winds around P, counting orientation. For the moment, we content ourselves
with proving that Ind7(P) is always an integer.
Lemma 4.5.5. If 'Y:

[a, b] ----t C \ {P} is a piecewise C1 closed curve, and if


P is a point not on (the image of) that curve, then
1
_
27ri

'Y

1
1
--d( =- _
(-P
27ri

lb 'Y('Yt)'(t)

-P

dt

(*)

is an integer.
Proof. Consider the function

g(t)

g: [a, b]

('Y(t) - P) exp

----t

C defined by (with exp A=- eA)

(-1t 'Y'(s)/['Y(s) - ] s)
Pd

The function g is continuous; indeed it is piecewise C1 on [a, b], with a


continuous derivative at each point at which 'Y has a continuous derivative.
We want to prove that g is constant. To do so, it suffices to show that
g' (t) 0 for every t at which "(1(t) exists and is continuous. We compute g'
using Leibniz's Rule:
=

g'(t)

'Y'(t) exp

(- 1t 'Y(s)'Y'(s) s)
_

+('Y(t) - P)

-'Y'(t)
exp
'Y(t) P
_

( - 1t 'Y(s)'Y'(s) s)
_

This last expression simplifies to


exp

(-1t "f1(s)/('Y(s)

- P) ds

Thus g is constant. Now we evaluate at t


'Y(a) "f( b). We have
=

('Y(a) - P) e0

g(a)

g( b)

b'(t) - 'Y'(t)]

a and t

0.

b, using the fact that

125

4.5. The Calculus of Residues

(r(b) - P) exp
(r(a) - P) exp
It follows that
exp

(-1b 1'(s)/(r(s) -

- J: ..J:)(Pds

and hence

in equation

(-1b --/(s)/(r(s)
(-1b 1'(s)/(r(s) -

- P)

P)

ds

must be an integer multiple of

)
ds) .

ds

P)

1
27ri.

So the quantity

is an integer.

The fact that the index is an integer-valued function suggests (as we


shall discuss further in Chapter

11)

that the index counts the topological

winding of the curve I' Note in particular that a curve which traces a circle

about the origin

k times

in a counterclockwise direction has index

respect to the origin; a curve which traces a circle about the origin
in a clockwise direction has index

-k with

k with
k times

respect to the origin.

Using the notation of residue and index, the residue theorem's formula
becomes

if= 27ri t

j=l

1'

Res1(Pj) Ind,,(Pj)

People sometimes state this formula informally as "the integral of


/' equals

27ri

around

times the sum of the residues counted according to the index of

/' about the singularities." Since the index of/' around a given singularity
is an integer, the "counted according to" statement makes good sense.
Now we turn to the (long delayed) proof of the residue theorem:

Proof of the residue theorem (Theorem 4.5.3).


teger

1, 2, . .. , n,

expand

For each positive in

in a Laurent series about Pj. Let

s j(

be the

principal part-that is, the negative power part-of the Laurent series at
Pj. Note that the series defining

s j is a convergent power series in (

Pj ) -1

and defines a holomorphic function on C \ {Pj}. Thus we can write

f- L
f- L

with both
function
each

(!- (s1 ++Sn) ) + (s1 + + sn),


L

Sj and
Sj defined on all of U\ {Pi, ... , Pn} Moreover, the
Sj has removable singularities at each of the points Pj because

sk, k-=/= j, is holomorphic at Pj and

f-

Sj has Laurent expansion with

no negative powers at Pj. Since the set U is h.s.c., we may thus conclude
that

1 26

4. Meromorphic Functions and Residues

i iL

or

f=

It remains to evaluate

s1.

I: Sj = I: Sj. Fix j, and write s1(z)


2::1 al_ (z - P1)-k. (Note that al_1 = Rj, by definition of R1.) Because
k
-y([a, b]) is a compact set and P1 fl. --y([a, b]), the series s1 converges uniformly

on -y; therefore

"I

Sj(()d(

"I

"I

f aJ_k 1 (( - P1)-k d(= i al_1(( - Pj)-1d(.


"I

"I

k= l
The last equality is true because, for k 2 , the function (( - P1)-k has an
antiderivative on C \ { P1} and Proposition 2.1.6 then tells us that (( -P1)-k
integrates to zero.
Thus

s1(() d(

27ri Res1(P1) lnd'Y(P1).

The proof is complete.

In order to use the residue theorem effectively, we need a method for


calculating residues. (We could also do with a quick and easy way to find
the index, or winding number; see Chapter 11.)
Proposition 4.5.6. L et f be a function with a pole of order k at P. Then

Res1(P)=
Proof. This is the case

) ((

:
l
!
(
)
(k
z
k-I

j=

- P)k f(z)

z=P

-1 of Proposition 4.4.1.

EXAMPLE 4.5. 7. Let 'Y be as in Figure 4.3 (as usual, 'Y

function and ;y its image) and let

z
f(z)= (z + 3i )3 (z + 2
)2

[O, 1]

---+

C is a

Clearly f has a pole of order 3 at -3i and a pole of order 2 at -2 . Now


1 a2
z
Res f ( - 3 i ) =
2! 8z2 (z + 2)2 z=-3i

)I
(
.1
2_
(
1
( 2-z )I
(z + 2)2

2! 8z

z . 2( z + 2)

(z + 2)4

2! 8z

(z + 2)3

2_ (z + 2)3
2!

z=-3i

z=-3i
(-1)- ( 2- z) 3(z + 2)2
(z + 2) 6
z=-3i

4.5. The Calculus of Residues

127

Figure 4 3
.

1 2z
2!

(z + 2)4

-6i - 8

z=- i
3

2! (-3i + 2)4.

;, ! Cz :

Resi(-2)

3i)3) lz=-2

(z + 3i)3

1-

2
3(z + 3i)

(z + 3i)6
-2z + 3i

(z + 3i)4
4 + 3i
=

z=-2

z=-2

(3i - 2)4 .
Thus

1 f(z) dz=
h

-3i - 4
(-3i + 2)4

.
Ind-y(-3i) +

Since Ind-y(-3i) = 3 and Ind-y(-2)


1
27ri

-2

f(z) dz=

4 + 3i
(3i

2)4

we obtain

-15i - 20
(3i - 2)4 .

Ind-y(-2).

128

4. Meromorphic Functions and Residues

4.6. Applications of the Calculus of Residues to the


Calculation of Definite Integrals and Sums
One of the most classical and fascinating applications of the calculus of

residues is the calculation of definite (usually improper) real integrals. It


is an over-simplification to call these calculations, taken together, a "tech
nique": It is more like a

collection of techniques. We can present only several

instances of the method and ask you to do lots of problems to sharpen your

skills.
The main interest of the method which we are about to present is that
it allows us to calculate many improper integrals which are not tractable
by ordinary techniques of calculus. However, we shall begin (for simplicity)
with an example which

could

in principle be done with calculus by using

partial fractions.

EXAMPLE 4.6.1. To evaluate

00

-oo

we "complexify" the integrand to

1
dx,
1+x4

f ( z)

'YR

1/(1+z4)

and consider the integral

1
dx.
1+Z 4

--

Hx

See Figure 4.4. Note that since lim __,00 J0a


4 dx and limb__,00 Jb l x4 dx
a
exist separately, the improper integral exists and there is no harm in evaluating the integral

as

l x4 dx, with R

-t

oo.

Now part of the game here is to choose the right piecewise


"contour"

'YR.

curve or

The appropriateness of our choice is justified (after the fact)

by the calculation which we are about to do. Assume that R

'Yh(t)
'Yk(t)

t+iO
R eit

if
if

Call these two curves, taken together,


Now we set U

-1/../2 i/../2, P4
on U \ {P1, ... , P4}
-

1.

Define

- R S:tS: R ,

0 5:

'Y

>

or

t 5:

1r.

'YR

C, P1
1/../2 +i/../2, P2
-1/../2 + i/../2, P3
1/../2 - i/J2; thus J(z) 1/(1 +z4) is holomorphic
=

and the residue theorem applies.

On the one hand,

21 9

4.6. Applications of the Calculus of Residues

Curve

-R

4.4

Figure

where we sum only over the poles off which lie

shows that

and

'YR

inside

'Y An easy calculation

1 .
= 4(-/1 J2+1 i/J2)3 - 4(1 - J21 + i J2)
.1
1 : z4 dz=2i ( -) ([ + i ) +( - + i )]
es

1(P2)

Of course the index at each point is

Therefore

(*)

On the other hand,

1 z.
1 z+ i --d
1 z=i --d
i-r --d
4
4
z
z
z4
1
+
+
1
+
1
-rh
-rh

Trivially,

1 dz=JR _ 1 dt J00 -1-dt


-1+z4
1+ t4
-R 1+ t4
hk
+oo.
J

l_

--t

-oo

as R--+

That is good because this last is the integral that we wish to

evaluate. Better still,

hh

dz
1+ z

I I
-rh 1+ z

{length bk)} sup

/ 1.

130

4. Meromorphic Functions and Residues

( Here

we use the inequality

ll +z 4 1

lzl4 - 1.)

1
h l +z 4

J __ dz ---+ 0

Finally,

(* ) , (** ) ,

and

(*** )

as

Thus

R---+

oo.

taken together yield

1
--dz
"I 1 + z4
1
1
lim J -- dz + lim J -- dz
R-.oo h 1 + z4
R-.oo hh 1 + z 4
00
1
-- dt+O.
-oo 1 + t 4

. i

hm

R-.oo

This solves the problem: The value of the integral is rr /

v'2.

In other problems, it will not be so easy to pick the contour so that


the superfluous parts ( in the above example, this would be the integral over

1h)

tend to zero, nor is it always so easy to prove that they

do tend

to zero.

Sometimes, it is not even obvious how to complexify the integrand.

EXAMPLE 4.6.2. We evaluate

J00 s 1 dx
x

by using the contour /'R as in

Figure 4.4. The obvious choice for the complexification of the integrand is

eixe-Y + e-ixey
eiz + e-iz
=
1 +z2
1 +z2
le-YI ::::; 1 on f'R but le-ixeyl = leYI becomes
f(z) =

Now

leixe-YI

on /'R when

quite large

is large and positive. There is no evident way to alter the

contour so that good estimates result. Instead, we alter the function! Let

g(z) = eiz / ( 1 +z2).

Now, on the one hand

( for R

J g(z) =

hR

>

2rri

2rri
On the other hand, with

1),

Res9(i) Ind"IR(i)
( J =
2

1h(t) = t, -R::::; t::::; R,

and

1h(t) = Reit, 0::::; t::::;

rr, we have

J g(z) dz= J g(z) dz+ J g(z) dz.

hR
Of course

i'Yk

hh

l -eix
dx
1 +x 2

oo

g(z) dz---+

-oo

as

R---+

oo.

131

4. 6. Applications of the Calculus of Residues

-R

Figure 4.5

Also,

j g(z) dz

f-rk

Thus

100

.:::; length (/'

cosx

_00 1 + x

2dx

--

k)

lgl .:::; 7rR

sup

7k

1
---+
R2 - 1

= 1_00= 1 eixx2dx =
Re

--

Re

R---+

as

oo.

( 7r ) = 7r
e

-.

Next is an example with a more subtle contour.

EXAMPLE 4.6.3. Let us evaluate


that

( sin x) / x

f00 sinxxdx.
-oo

Before we begin, we remark

is bounded near zero; also, the integral converges at

an improper Riemann integral

by integration by parts.

oo

( as

So the problem

makes sense. Using the lesson learned from the last example, we consider

g(z) eiz /z. However the pole of eiz /z is at z


lies on the contour in Figure 4.4. Thus that contour may not
instead use the contour
R in Figure 4.5. Define
the function

k(t)
'h(t)
(t)
k(t)
Clearly

t
eit / R
t
Reit

if
if
if
if

-R.:::; t.:::; -1/R,


7r .:::; t .:::; 27r'
1/R.:::; t.:::; R,
0 .:::; t .:::; 7r.

0 and that

be used. We

132

4. Meromorphic Functions and Residues

On the one hand, for

R > 0,

t g(z) dz=

On the other hand,

2niR es9 ( 0 ) Ind(O) = 2ni 11 = 2ni.

1=
g(z) dz----) Im

Im J g(z) dz+Im J
Jh
fk

+= sinx
1
dx =
x
oo x

eix

Furthermore,

<

fk

ftt,y<v'R

as

R----) oo.

-00

J g(z) dz

(*)

g(z) dz

=A+B.

Now

<

length('kn {z:

<

4VR

Also
B

( )

<

length(k n {z:

<

nR

So

as

----) 0

-v'R
T

( )

VR}) sup{[g(z)[: z E 'k,Imz < VR}

y <

y >

----)

1 g(z) dz

Ji

Finally,

R----) oo.

VR}) sup{[g(z)[: z E 'k, y


as

>

VR}

R----) oo.

----) 0

R----) oo.

as

J g(z) dz

]7t
As R----)

oo,

this tends to
=i

In summary, ( * )

-(

127r ldt

= ni
****

as

R----) oo.

) y ield

2n =Im

f, g(z) dz=Im L tiR g(z) dz


j=l

( **** )

4.6. Applications of the Calculus of Residues

-1/R

-R

IIR

Figure

-+Im

oo

Thus

oo

7r

eix

as

R-+ oo.

sinx
d

oo

4.6

-dx+7r
x
=

133

--

oo

x.

Exercise: Suppose that we had decided to do the last example with the

with 'ji,(t) e-it / R, 7r ::::; t ::::; 27r.


4.6. Yet the technique still works to

contour ob tained from by replacing


Then has no poles inside! See Figure

evaluate the integral: Prove this assertion.


In the next example (somewhat similarly to the last one), we cannot
make the spurious part of our integral disappear; so we incorporate it into
the solution.

EXAMPLE 4.6.4. Consider the integral

100 --2dx.
xl/3

l+x

We complexify the integrand by setting


on the simply connected set U

f(z)

\ { iy

: y ::::;

z1/3 /(1 + z

0},

).

Notice that,

when

4.6,

reiO,

-7r /2

< () <

37r /2.

z1/3 is
r113ei0/3

the expression

unambiguously defined as a holomorphic function by setting z113

We use the contour displayed in Figure

just as in the exercise following Example

4.6.3.

We must do this since

z1/3 is not a well-defined holomorphic function in any neighborhood of

Let us use the notation of Example

4.6.3.

Clearly

f,

f(z) dz-+

100
0

tl/3

2dt.
1 +t

--

0.

134

4. Meromorphic Functions and Residues

Of course that is good, but what will become of the integral over k? We
have

tl/3
1-R-1/R 1--dt
+t 2
(-t)I/3
J1/RR --dt
l+t2
eirr/3tl/3dt
J1/RR --l+t2

z113 !).
t /3
1 f(z)dz+l f(z)dz-t (1+ (+ y'3i)) r = l 2dt
}0 1+t
J
Jk
( by

Thus

our definition of

On the other hand,

J( ) dz
z

::;

7r

as

So

as

The calculus of residues tells us that, for

2 7r Res1

>

Ind

. J

x(+i)
Finall y,

R-too.

R-too
R RR2 I/31-t
-2rr ---. eit
i
(eit e/ R)it I/3 -J J(z)dz
1
rr- 1+0 2 / R2 R dt
J
R -too.
R-4/3 1rr 1+eei42t/it3/ R2dt-t
R-too.
L f(z)dz 0 + i) f /::2dt
R 1,
J f(z)dz= i
(i) R (i)
fR
hi ( e';:')
J
]

and

as

()
*

and

**

taken together y ield

1= --dt=tl/3
2
0 1+ t

7r

y'3"

as

135

4.6. Applications of the Calculus of Residues

In the next example we exploit a contour of a different kind and use


some other new tricks as well.
While the integral

EXAMPLE 4.6.5.

(00

dx
x2 + 6x + 8

Jo
can be calculated using methods of calculus, it is enlightening to perform
the integration by complex variable methods. Notice that if we endeavor to
use the integrand f(z)= 1/(z2 + 6z + 8) together with the idea of the last
example, then there is no "auxiliary radius" which helps. More precisely,
(rei11)2 + 6rei11 + 8 is a constant multiple of r2 + 6r + 8 only if () is an
integer multiple of 2rr. The following nonobvious device is often of great
utility in problems of this kind. Define log z on U = C \ { x : x O} by
{log(reill)=(log r)+iO when 0 < () < 2rr,r > 0}. Here log r is understood to
be the standard real logarithm. Then, on U, log is a well-defined holomorphic
function (why?).
We use the contour 'f/R displayed in Figure 4. 7 and integrate the func
tion g(z)=log z/(z2 + 6z + 8). Let
'flk(t) = t + i/J2R,
'fl(t) = Reit,

1/J2R::; t::; R,

Oo::; t::; 2rr - Oo,

where Oo=Oo(R)= sin-1(1/(RJ2R)),


'flh(t)

R - t - i/J2R, O::; t::; R - 1/J2R,

'fl(t) = e-it/..fR, rr/4::; t::; 7rr/4.


Now

J g(z)dz=2rri(Res9(-2) 1 + Res9(-4)1)

'1i R

( log(-2) + log
(-4) )
__ _ _
-2
2
. ( log 2 + rri log 4 + rri )
=2m
+ ----

=2rri

=-rri log 2 .
Also, it is straightforward to check that

J g(z)dz -+O,

'1i h

J g(z)dz -+O,

'1i Jt

-2

(* )

136

4. Meromorphic Functions and R esidues

Ri

Curve TIR

{R

Tlk

-R

ill

-i

{R

-Ri

Figure 4.7
as

Thus

---t

oo.

The device that makes this technique work is that, as


log(x + i/v'2R) - log(x

- i/v'2R)---t -27ri.

dt
1 g(z)dz+ 1 g(z)dz---t-27ri f 00
2
l
t
+
6t + s
o
h,k
h,'h

---t

oo,

(***)

Now (*),(**),(***) taken together yield

1
(00
dt
g .
=
lo t2 +6t+8 210 2

The next, and last, example we only outline. Think of it as an exercise


with a lot of hints.
EXAMPLE 4.6.6.

We sum the series

00

L j27r2 _
j=l

x2

1, 2, ... let
using contour integration. Define cot( = cos(/ sin(. For n
r be the contour ( shown in Figure 4.8) consisting of the counterclockwise
oriented square with corners { (1 i) ( n+ ) 7r }. For z fixed and n > Jzl
we calculate using residues that
=

_1 J cot(
d
27ri Jrn (((- z) ( =

t j7r(j7r- z)+ t j7r(j7r+ z)


j=l

1
cot z
+---2
z
z

j=l

4. 7. Singularities at Infinity

137

n1ti
'

-n1t

n7t

,,

'
'
-

-n1ti

rn

Figure 4.8

W hen n

lzl ,

it is easy to estimate the left-hand side in modulus by

Thus we see that

+
. j7r(j7r - z ) 6 j7r(j7r + z )
J=l
J=l
We conclude that

6
j=l

j27r2

cotz

cotz

z2

-z- + z2

-z - + z2

or

00

z
27r2 z2
j=l J

-2cot z +

2
z

This is the desired result.

4. 7. Meromorphic Functions and Singularities at Infinity


We have considered carefully functions which are holomorphic on sets of
the form

D(P,r) \ {P}

is an open set in C and

or, more generally, of the form

U.

U \ {P},

where

Sometimes it is important to consider the

possibility that a function could be "singular" at more than just one point.
The appropriate precise definition requires a little preliminary consideration
of what kinds of sets might be appropriate as "sets of singularities". Recall
from Section 3.6:

4.

138

Definition

4. 7.1. A set Sin

is a positive number

Meromorphic Functions and Residues

<C is discrete if and only if for each z


z) such that

Sthere

(depending on S and on
Sn

D(z, r )

{z}.

We also say in this circumstance that S consists of isolated points.


Now fix an open set U; we next define the central concept of meromor
phic function on U.

Definition
function f :
(a)
(b)
(c)

4. 7.2. A
U \ S --+

meromorphic function f
<C such that

on U

with singular set Sis

the set S is closed in U and is discrete,


the function
open in

<C),

for each

{z},

is holomorphic on U\S (note that U\Sis necessarily

r > 0 such that D(z, r) U and Sn D(z, r)


flv(z,r)\{z} has a (finite order) pole at z.

Sand

the function

For convenience, one often suppresses explicit consideration of the set


Sand just says that

is a meromorphic function on U. Sometimes we say,

informally, that a meromorphic function on U is a function on U which is


holomorphic "except for poles." Implicit in this description is the idea that

P is a pole of f
D(P, r) around P such that f is holomorphic
pole at P. Back on the level of precise language,

a pole is an "isolated singularity." In other words, a point


if and only if there is a disc
on

D(P, r) \ {P}

and has a

we see that our definition of a meromorphic function on U implies that, for

P E U, either there is a disc D(P, r) U such that f is holomorphic


on D(P, r) or there is a disc D(P, r) U such that f is holomorphic on
D(P, r) \ {P} and has a pole at P.
each

Meromorphic functions are very natural objects to consider, primarily


because they result from considering the (algebraic) reciprocals of holomor
phic functions. We shall prove this carefully now, partly just to illustrate
how the definitions work:

Lemma 4. 7.3. If U is a connected open set in <C and if f


holomorphic function with f . 0, then the function
F:

U\

{z: f (z)

--+

<C is

O}--+ <C

defined by F(z)
1/ f(z), z E U \ {z E U : f (z)
O}, is a meromorphic
function on U with singular set (or pole set) equal to {z E U : f (z) 0}.
=

Remark:

In actual practice, we say for short that "1/ f is meromorphic."

This is all right as long as the precise meaning, as stated in the conclusion
of the lemma, is understood.

4. 7. Singularities at Infinity

139

Proof of Lemma 4.7.3. The set Z::::::: {z E U : f(z)


O} is closed in U
by the continuity off. It is discrete by Theorem 3.6. 1. The function F is
obviously holomorphic on U \ Z (cf. Exercise 67). If P E Z and r > 0 is
such that D(P, r) U and D(P, r) n Z {P}, then F is holomorphic on
D(P, r) \ {P}. Also,
=

lim IF(z)I

z-+P

since limz_,p lf(z)I

0.

l
z-+P lf( Z)I
lim

+oo
=

Thus FID(P,r)\{P} has a pole at P.

It is possible to define arithmetic operations on the whole set ofmero


morphic functions on U so that this set becomes a field in the sense of
abstract algebra. This fact is fundamental in certain directions ofdevelop
ment ofthe subject ofcomplex analysis, but as it happens we shall not need
it in this book. The interested reader can pursue the topic in Exercises 63
and 64.
It is quite possible for a meromorphic function on an open set U to have
infinitely many poles in U. It is easy to see that if U is an open set, then
there are infinite sets S U such that S is closed in U and S is discrete
(see Exercises 66 and 68). Such a set could in principle be the pole set
of a meromorphic function on U; namely S satisfies condition (a) of the
definition ofmeromorphic function. We shall see later that every such set S
actually is the pole set ofsome meromorphic function (Section 8.3). Exercise
66 exhibits a specific example of a meromorphic function on D(O , 1) with
infinitely many poles. The function 1/ sin z is an obvious example on C.
Our discussion so far of singularities of holomorphic functions can
be generalized to include the limit behavior of holomorphic functions as
lzl --+ +oo. This is a powerful method with many important consequences.
Suppose, for example, that f : C --+ C is an entire function. We can asso
ciate to f a new function G: C \ {O} --+ C by setting G(z) f(l/z). The
behavior ofthe function G near 0 reflects in an obvious sense the behavior
off as lzl --+ +oo. For instance,
=

lim lf(z) I
lzJ-++oo

+oo

ifand only ifG has a pole at 0. This idea is useful enough to justify a formal
definition (for a more general kind off):

Definition 4.7.4. Suppose that f : U --+ C is a holomorphic function on


an open set U C and that, for some R > 0, U :2 {z : lzl > R}. Define
G : {z: 0 < lzl < 1/R}--+ C by G(z) f(l/z). Then we say that
=

(i) f has a removable singularity at oo ifG has a removable singularity


at O;

4.

140

(ii) f
(iii) f

has a

pole at oo

has an

at

if

Meromorphic Functions and Residues

O;

has a pole at

essential singularity

at

oo

if

G has

an essential singularity

0.

G around 0, G(z)
for f which converges

The Laurent expansion of


mediately a series expansion

G(l/z)

Lanz-n

lzl

>

yields im

R, namely

La-nzn.
-oo

-oo

The series

for

+oo

+oo

f(z)

E: anzn,

E: a_nzn is called the Laurent expansion of J around oo.

It follows from our definitions and from the discussion following Proposition

4.3.3

that

has a removable singularity at

series has no
pole at
of

positive

powers of

oo

if and only if the Laurent

with nonzero coefficients. Also

has a

oo if and only if the series has only a finite number of positive powers
f has an essential singularity at oo if

with nonzero coefficients. Finally,

and only if the series has infinitely many positive powers.


We can apply these considerations to prove the following theorem:

Theorem 4.7.5.

Suppose that f : C --+ C is an entire function. Then


lf(z)I
+oo (i.e., J has a pole at oo) if and only if J is a non
constant polynomial.

limlzl-++oo

The function f has


constant.
Proof. Since

removable singularity at oo if and only if f is

is entire, the power series of

f(z),

00

f(z)
converges for all

Lanzn ,
n=O

C. Hence
00

G(z)
for all

\ {O}.

J(l/z)

Lanz-n
n=O

( Proposition
4.2.4) gives that this is the only possible Laurent expansion of G around 0.
Hence the Laurent expansion of f around oo is ( not surprisingly ) the original
E

The uniqueness for Laurent expansions

power series

n=O
As noted,

has a pole at

oo

if and only if this expansion contains only a

finite number of positive powers. Hence

is a polynomial. The function

4. 7. Singularities at Infinity

141

has a removable singularity at oo if and only if this expansion contains no


positive powers; in this case,

f(z)

ao.

The converses of these statements are straightforward from the definitions.

This theorem shows that nontrivial holomorphic functions, which we


defined initially as a generalization of polynomials in z, actually tend to differ
from polynomials in some important ways. In particular, the behavior of a
nonconstant entire function at infinity, if the function is not a polynomial,
is much more complicated than the behavior of polynomials at infinity: A
nonpolynomial entire function must have an essential singularity at oo.
It is natural to extend the idea of meromorphic functions to include
the possibility of a pole at oo. Of course, we want to require that the pole
at oo be an isolated singularity.

f is a meromorphic function defined on


<C such that, for some R > 0, we have U 2 {z : lzl >
R}. We say that f is meromorphic at oo if the function G ( z ) = f( 1/z) is
meromorphic in the usual sense on {z : lzl < 1/ R}.

Definition 4. 7.6. Suppose that


an open set U <;::;;

R'

The definition as given is equivalent to requiring that, for some

has no poles in

{z

<C: lzl

>

R'}

and that

f has

The following theorem elucidates the situation for the case U

Theorem 4. 7.7. A meromorpl1ic function

on

<C

>

R,

a pole at oo.
=

<C.

which is also meromor

phic at oo must be a rational function (i.e., a quotient of polynomials in


Conversely, every rational function is meromorphic 011

<C

z) .

and at oo.

Remark: It is conventional to rephrase the ( first sentence of the ) theorem


by saying that the only functions that are meromorphic in the "extended
plane" are rational functions.

f has a pole at oo, then it has ( by definition


of pole at oo ) no poles in the set {z E <C : I z I > R} for some R > 0. Thus
all poles ( in <C) of f lie in {z : lzl ::::; R}. The poles of f ( in <C) form, by
Proof of Theorem 4.7.7. If

definition of a meromorphic function, a discrete set that is closed in <C.


Now a closed

( in <C),

discrete set

subset of the plane must be finite.

( If

that is contained in a compact

it were infinite, then

an accumulation point, which would belong to


would contradict the discreteness of
finitely many poles of

S.)

since

would have

is closed; and that

We conclude that there are only

in the finite part of the plane.

142

4. Meromorphic Functions and Residues

Ifinstead f is finite-valued at oo-say that f(oo)

=a EC-then there
{z : lzl > R} on which f is finite
valued, and indeed satisfies lf(z) - al < 1, say. In this case, too, the poles
off lie in {z: izl ::::; R}; hence there are only finitely many of them.

is a "neighborhood of oo" of the form

In either of the two cases just discussed-pole at oo or not-we let

P1 , ... , Pk denote the poles in the finite part of the plane. Thus there are
positive integers n1, . . . , nk such that
F(z)

(z - Pi)n1

(z - Pktk f(z)

has only removable singularities in C. Observe that f is rational ifand only


if F is. So it is enough for us to consider the entire function F, with the
removable singularities in the finite part of the plane filled in.
We consider three possibilities:

(i) F has a removable singularity at oo. Then F is constant (by The


orem 4.7.5), and we are done.
(ii) F has a pole at oo. Then F is a polynomial by Theorem 4.7.5, and
we are done.

(iii) F has an essential singularity at oo. This is impossible, for the


Laurent expansion of f about oo has only finitely many positive
powers, hence so does the expansion for F about oo.

The discussion in this section suggests that the behavior at infinity ofa
holomorphic function is describable in the same terms as behavior near any
other isolated singularity. Pursuing this notion a little further, one might
expect that the so-called extended plane, namely the set (['.U { oo}, could be
considered as homogeneous, in some sense; that is, the point oo would have
the same status as the points of C.

Specifically, replacing

z by 1/z (the

process by which we have treated singularities at infinity ) could be thought


ofas acting on all of(['.U { oo}, with 0 corresponding to oo and oo to 0, while
finite

z E C, z #

0, correspond to

1/z. This mapping can be easily seen to

be meromorphic on all of (['.U {oo} in an obvious sense.


One can push this viewpoint a little further as follows. First, we define
a subset U of (['.U {oo} to be open if

(1) u n (['. is open in (['.


and

(2) if oo EU , then, for some R

>

0, the set

{z E (['.: izl

>

R, z EC} is

contained in U.
With this idea ofopen sets in (['.U { oo}, the mapping of(['.U { oo} to (['.U { oo}
given by

1-+

1/z is continuous (i.e., the inverse of an open set is open).


1-+ 1/z is its own inverse, it follows that this mapping

Because the mapping z

4. 7.

Singularities at Infinity

143

is a homeomorphism-the image of a set is open if and only if the set itself


is open.
W ith the topology in C U

{ oo}

that we have just described, and with

the associated idea of continuity, we can reformulate our entire concept of a

f : D ( P, r) \ { P } --+ C with
f : D(P, r) --+ CU { oo}
with J(P)
oo . Then the fact that f has a pole at P corresponds exactly to
the continuity of the extended f at the point P. Notice that in this case 1/ f

meromorphic function: A holomorphic function


a pole at

( E C) can be thought of as a function

is C-valued and holomorphic on some neighborhood of P. This motivates one


to think of meromorphic functions as holomorphic functions with values in

F:
CU { oo}

CU { oo}. One simply defines a continuous function

open subset U of CU

to be holomorphic if

( 1) F

F-1 (IC)

{ oo}

that has values in

--+

CU { oo} on an

is holomorphic

and

(2) p-1({00})

is a discrete set in U and

open neighborhood of

1/F

is holomorphic on some

p-1 ({ oo} ).

From this viewpoint, the idea of a meromorphic function as we have consid


ered it previously, seems much more natural: A pole is just a point where
the function attains the value oo and the function is holomorphic in a neigh
borhood of that point. The set CU

{ oo}

is often denoted by

C.

This may seem a bit formalistic and tedious. But, in practice, it turns
out to be remarkably convenient in many contexts to regard oo in CU

{ oo}

as just like every other point, in exactly the manner that we have indicated.
Further details will be explored as the book develops.
As an exercise, the reader should prove that any rational function can
be considered as a holomorphic function from C U

{ oo}

to C U

{ oo } .

The Riemann sphere as just described is what is known in topology as


the "one-point compactification" of the complex plane C. Intuitively, what
we did here topologically was, in effect, to gather up the complex plane
(think of it as a large piece of paper) and turn it into a sphere by adding a
plug (the point oo ) . See Figure

4.9.

The topology we specified on Sis just

the same topology that the unit sphere in

space would inherit if we used

as open sets the intersections of open sets in three-dimensional Euclidean


space with the sphere.
Now we present a precise version of this intuitive idea. It is known as
stereographic projection. Refer to Figure

4.10

as you read along.

Imagine the unit sphere in three-dimensional space S

JR3

: x2

y2

z2

1},

cut by the

x-y

{ (x, y, z)

plane as shown in the figure. We

144

4.

Meromorphic Functions and Residues

00

Figure

Figure

shall think of the point denoted by P

4.9

4.10

(0, 0, 1)

as

the "point at infinity."

The idea now is to have a geometric scheme for mapping the plane to the
sphere.
Let z

(x, y)

iy

line connecting P with z.

be a point of the plane. Let f be the Euclidean

Denote the unique point where f intersects the

sphere as 7r ( z) . The point 7r ( z ) is called the

stereographic projection

of z to

the sphere.
It is easy to see that

except the

{ 7r ( z)

: z E

C}

consists of all points of the sphere

pole P. The points of the plane that get mapped near to P by 7r

are points with very large absolute value. Thus it is natural to think of P
as

the "point at infinity."

A set in

CU { oo}

is open in the sense we defined earlier if and only

if the corresponding set in S is open. More precisely, define a mapping


of

CU { oo}

to S by z

t-t

( )

7r z ,

z E

C, II(oo)

(0, 0, 1).

Then

II

II

is a

145

Exercises

II

homeomorphism, that is,

continuous. The proof that

is continuous, one-to-one, and onto, and

II

II-1 is

is a homeomorphism is left as an exercise.

Exercises

R(z)

p and q are
{P1, P2, ... , Pk}
and suppose that f has a pole at each of the points P1, P2, ... , Pk. Finally

1. Let

R(z)

be a rational function:

holomorphic polynomials. Let

p(z)/q(z)

where

be holomorphic on C \

assume that

lf(z)I:::; IR(z)I
for all

z at

which

stant multiple of

f(z)
R. In

R(z)

and

are defined.

particular,

Prove that

is rational.

[Hint:

is a con

Think about

f(z)/ R(z) .]
2. Let

f: D(P, r ) \ {P}

C be holomorphic. Let U

f(D(P, r ) \ {P} ) .

Assume that U is open (we shall later see that this is always the case if

is nonconstant). Let g: U

singularity at

P,

C be holomorphic. If

has a removable

does go f have one also? What about the case of poles

and essential singularities?

3. Let

be holomorphic on U

singularity at

P,

PE U

Aj

{!

has an essential

1/f

have at

has a removable singularity or a pole at

P?

P?

C,U open. Let


holomorphic on U

\ {P}: f

satisfies case (j) at

where (j) refers to singularities of types

(iii)

open. If

then what type of singularity does

What about when

4. Let

\ {P}, PE U, U

(i)

removable,

P},

(ii)

pole, or

essential singularity. Is Aj closed under addition? Multiplication?

Division?

5. Let

0. Classify each of the following as having a removable singu

larity, a pole, or an essential singularity at

1
( a ) -,
z
. 1
{b) sm -,
z
1
( c ) 3 - cosz,
z
{d) z. el/z. e-l/z2,
(e)
{f)

sin z

z
cosz

P:

146

Exercises

k k
'"'
(g) L..,, 00
k= 2 2 z
z3

6. Prove that

00

converges for
larity at

LT(2i). z-j
j=l
:f. 0 and defines a function which has an essential singu

0.

7. Suppose that we attempt to prove a version of the Riemann removable


singularities theorem in which the word "holomorphic" is replaced by

"C00". If we attempt to imitate the proof of Riemann's theorem given in


the text, what is the first step at which the proof for C00 breaks down?
8. Let

D(P, r) \ {P}.

Prove the following two refined versions of Rie

mann's theorem:

( a)
*

{b)

If

f is

holomorphic on

and limz_,p(z

continues holomorphically across


If

f is

then

[Hint:

holomorphic on

and if

L IJ(z)l2dxdy

f extends

- P) J(z)
U).

<

holomorphically across P.

7rE

JD(Q,E)

polar coordinates centered at

Q;

l(Q,E)

IF(z)l2dxdy

D(Q,E), then

IF(z)l2dxdy

by using the Cauchy integral formula for


observe that

0, then

oo,

First show that if F is a holomorphic function on

IF(Q)l2

(to all of

F2 and

writing the integral in

cf. Exercise 33 in Chapter 3. In detail,

1 ( fo IF2(rei9)1 dO) rdr


f 2rr
1 l1 F2(rei9)d0 1 rdr

fo 127rF2(Q)I rdr
2rr

>

7rE2IF2(Q)I

Use this to show that Fas in the statement of the problem has at most a
simple pole or removable singularity at
Then show that

9. Prove that if

J IFl2

(not an essential singularity).

+oo if F has a pole.]

f: D(P,r) \ {P}

P,
D(P,r) \ {P}

Chas an essential singularity at

then for each positive integer N there is a sequence

{Zn}

147

Exercises

with limn-+oo Zn= p and

[Informally, we can say that


of

1/(z

"blows up" faster than any positive power

- P) along some sequence converging to P.]

10. Give an example of a series of complex constants

limN-++oo L =-N an exists but

E: an does

E: an

such that

not converge.

11. Give an example of a (formal) doubly infinite series

E: anzn

such
n
that there is a z f:- 0 for which the limit limN-++oo L =-N anz exists
n
but such that E: anz fails to converge for that same z.

12. A Laurent series converges on an annular region. Give examples to show

that the set of convergence for a Laurent series can include some of the
boundary, all of the boundary, or none of the boundary.
13. Calculate the annulus of convergence (including any boundary points)

for each of the following Laurent series:


( a)
2 j zJ,

(b)
(c)

14.

2:;:_00
1
"' 00=D 4-j zJ + UJ
"' -:--=-oo 3JzJ '
UJ
E;: 1 zJ /j2,

(d) E;:-oo,#OzJ/jJ,
( e ) 2: -oo zJ/IJI! (O! = 1),
Cr) E;:- 0 J2zJ.
2
Let fi,f2, ... be holomorphic on D(P,r) \ {P} and suppose that each
fj has a pole at P. If limj -+oo fj converges uniformly on compact subsets
of D(P, r) \ {P}, then does the limit function f necessarily have a pole
at

P?

Answer the same question with "pole" replaced by "removable

singularity" or "essential singularity."


15. Make the discussion of pp.

117-118

completely explicit by doing the

16:

following proofs and continuing with Exercise

( a)

Prove that if

P,
(b)
(c )

then

value of

1/ f
1/ f

has a removable
at

D(P,r) \ {P} and J has a pole at


singularity at P with the "filled in"

is holomorphic on

P equal to

0.

Let k be the order of the zero of

1/ f

P.

at

Prove that (z

- P)k f

has a removable singularity at P.


Conversely show that if

g is holomorphic

not bounded, and if there is a

bounded, then

< m E

D(P, r) \ {P}, if g is
such that (z - Prg is

on

has a pole at P. Prove that the least such m is

precisely the order of the pole.

15 to prove that the classification of isolated


(i), (ii), (iii) is equivalent to the following:

16. Use the result of Exercise

singularities given by

Exercises

148

A function f holomorphic on

D(P,r) \ {P}

satisfies one and only

one of the following conditions:

(i) f is bounded near P;


(ii) f is unbounded near P but there is a
(z - Prf is bounded near P;
(iii) T here is no m satisfying (ii).

positive integer

such that

17. Use formal algebra to calculate the first four terms of the Laurent series
expansion of each of the following functions:

( a ) tanz = (sinz/cosz) about


(b) ez /sinz about 0,
( c ) ez /( 1 - ez) about 0,
(d) sin (l/z) about 0,
( e ) z(sinz)-2 about 0,
(f) z2 (sinz)-3 about 0.

rr/2,

For each of these functions, identify the type of singularity at the point
about which the function has been expanded.

18. Let f : C---> C be a nonconstant entire function. Define g( z)

f(l/z).

Prove that f is a polynomial if and only if g has a pole at 0. In other


words, f is transcendental (nonpolynomial ) if and only if g has an es
sential singularity at 0.

19. Iff is holomorphic on

P, then

D(P,r) \ {P}

and has an essential singularity at

what type of singularity does f2 have at

P?

20. Answer Exercise 19 with "essential singularity" replaced by "pole" or


by "removable singularity."

21. Prove that if f :

P, then ef
22. Let {a1 :

D(P,r) \ {P}

--->

C has a nonremovable singularity at

has an essential singularity at


=

converges on

P.

0, 1, 2, ...} be given. F ix k

D(O,r) for some r >

0, then L,

D(O,r).

>

0. Prove: If L,

j a1zl

+k converges on
k+l a1zl

23. Let f be holomorphic on

D(P,r) \ {P} and suppose that f has a pole


P of order precisely k. What is the relationship between the Laurent
k
expansion off at P and the Taylor coefficients of (z - P ) f about P ?

at

24. Show that the Laurent expansion of


f(z)
about

1
=

ez-1

0 has the form

(-l)k-1

k=l

Bk k
2
( 2k )!z '

149

Exercises

for some (uniquely determined) real numbers


the

Bernoulli numbers.

B2k.

The

B2k

are called

Calculate the first three Bernoulli numbers.

25. Prove that if a function f is holomorphic on D(P, r)


essential singularity at P, then for any integer

\ { P} and has an

the function (z - Pr

f(z) has an essential singularity at P.

26. Use Proposition 4.2.4, together with the ideas in the proof of Theorem
4. 3.2, to give another proof of the independence of the expansion in
Theorem 4. 3 .2 from

s1, s2 .

27. Calculate the first four terms of the Laurent expansion of the given
function about the given point.

In each case, specify the annulus of

convergence of the expansion.

( a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
(i)

J(z)= csc z about P =

f(z) :== z/(z + 1)3about P = -1


f(z)= z/[(z - l)(z - 3)(z

- 5)]
- 5)]
f(z)= z/[(z - l)(z - 3)(z - 5)]

J(z)= z/[(z - l)(z - 3)(z

about P

about P =3
about P

f(z)= csc z about P = rr

J(z)= sec z about P = rr/2


f(z)=ez/z3about P =O
J(z)=e11z /z3about P =

28. A holomorphic function f on a set of the form { z : I z I

>

R}, some R > 0,

is said to have a zero at oo of order k if f(1/z) has a zero of order k at

0.

Using this definition as motivation, give a definition of

pole

of order

k at oo. If g has a pole of order k at oo, what property does 1/g have
at oo? What property does 1/g(l/z) have at O?

29. Derive the algorithm in Proposition 4.4. l for calculating the Laurent
coefficients of a function from the one using the Cauchy theory that was
given in Proposition 4.3.3.

30. Calculate the stereographic projection explicitly in terms of the coordi


nates of three-dimensional space.

31. Prove that if z, w in the complex plane correspond to diametrically op


posite points on the Riemann sphere (in the stereographic projection
model), then z

w= -1.

32. If z, w are points in the complex plane, then the distance between their
stereographic projections is given by

d=

2lz

-wl
2
J(l + lzl )(1 + lwl2)

Prove this. The resulting metric on C is called the

33. Compute each of the following residues:

spherical metric.

150

Exercises

z2

( a)

Res1(2i),

f(z) =

(b)

Resi(-3),

f(z) =

(c)

Res1( i + 1),

f(z)

(d)

Res1(2),

f(z) =

(e)

Res1(-i),

(f)

Res1(0),

(g)

Res1(0),

(h )

Res1( 7r),

J(z) = 2
z z + i)2 '
cotz
f(z) =
z(z + 1) '
sinz
f(z) =
3
z (z-2)(z + 1)
cotz
J(z) = 2
z z + 1)

'

(z-2i)(z + 3)
z2 + 1

(z + 3 2z '
e
.
1)3
(z
iz

'

(z + l)(z-2)
cotz

'

'

34. Use the calculus of residues to compute each of the following integrals:
1
a .
f(z)dz where f(z) =z/[(z + l)(z + 2i)],
27rt
1
.
f(z)dz where f(z) =ez /[(z + l) sinz],
(b)
27rt

( )

laD(0,5)

laD(0,5)

( c ) 2m
1

laD(0,8)

(d)

(e)
(f)

(g )

2m

ez
f(z)dz where f(z) =
z(z + 1)(z + 2)
-y

(i)

and r is the nega-

tively ( clockwise) oriented triangle with vertices 1 i and -3,


ez
1
and / is
.
f(z)dz where f(z) =
. 2
2m -y
(z + 3i z + 3 2 z + 4)

) (

) (

the negatively oriented rectangle with vertices 2 i, -8 i,


cosz
1
and 'Y is as in
.
- .
f(z)dz where f(z) =
'Y
2
+
1 2 z + i)
z z
27ri

Figure 4.11,
1
f(z)dz where
27ri 'Y

4.12,

(h)

f(z)dz where f(z) =cotz/[(z - 6i)2 + 64],

1
- .
27ri

f(z) =

) (

sinz
z(z + 2i)3

and

is as in Figure

eiz
and r is the positively
f(z)dz where f(z) = .
'Y
( smz)( cosz)

i
i

( counterclockwise) oriented quadrilateral with vertices 5 i, 10,


1
.
f(z)dz where f(z) = tanz and r is the curve in Figure
27rt
'Y
4.13.

151

Exercises

Figure 4.11
y

Figure 4.12
y

Figure 4.13

Exercises

152

35. Prove the following assertion: Let P1, ..., Pk be distinct points in an
open set U. Let a1, ..., ak be any complex numbers. Then there is a
meromorphic function

on U with poles at P1, ..., Pk (and nowhere

1, . . ., k.

else) such that Res1(Pj)=aj,J =

36. Prove the following: If a

C and if k

there is a holomorphic function

Z is a positive integer, then

on D(P, r)

\ {P }

such that

has a

pole of order k at P and Res1(P )=a.

37. Suppose that

\ {P} --t C is continuous and that f is holo


\ {P}. Assume that f has an essential singularity

: D(P, r)

morphic on D(P, r)

at P . What is the value of


coefficients of

38. Suppose that


and

2m

JaD(P,r)

f?
f

and

f(z) dz

in terms of the Laurent

are holomorphic on a neighborhood of D(P, r)

has simple zeros at P1, P2, ... , P k E D(P, r). Compute

1
_
27l'i

J
laD(P,r)

f(z)
dz
g( Z )

f(P1), ..., f(Pk) and g' (P1), ..., g' (Pk). How
mula change if the zeros of g are not simple?

in terms of

39. Let

be holomorphic in a neighborhood of 0. Set

Res9k(O)=0 for k=0,

40. Let

f(z)=e(z+I/z).

1, 2, ...,

then prove that

Res1(0)=

f(z). If

0.

.
L
k=O k!(k + l)!

41. This exercise develops a notion of residue at


First, note that if

r,

9k(z)=z-

Prove that

00

0 < s <

does your for

oo.

is holomorphic on a set D(O, r)

faD(O,s) g(z) dz

then "the residue at O" = 2 i

particular coefficient of the Laurent expansion of

\ {O}

and if

picks out one

about 0, namely it

equals a-1 If g is defined and holomorphic on {z : lzl > R}, then the
residue at oo of g is defined to be the negative of the residue at 0 of
2
H(z) = z- g(l/z) (Because a positively oriented circle about oo is
negatively oriented with respect to the origin and vice versa, we defined
the

residue of g

negative of the residue of Hat 0.) Prove


z in the Laurent expansion
also that the definition of residue of g at oo

at oo to be the

that the residue at oo of g is the coefficient of


of

on

{z: lzl

>

R}. Prove

remains unchanged if the origin is replaced by some other point in the


finite plane.

1 53

Exercises

y
-5 I
\

'
\

'

I
'

'

./

'

15
I
I

-- _,,.

..... _

Figure 4.14

42. Refer to Exercise 41 for terminology. Let R( z) be a rational function


(quotient of polynomials).

Prove that the sum of all the residues (in

cluding the residue at oo ) of R is zero. Is this true for a more general


class of functions than rational functions?

43. Refer to Exercise 41 for terminology. Calculate the residue of the given
function at oo.
(a) J(z) = z3 -

{b) f(z)
{c) f(z)
{d) f(z)
(e)
{f)
(g)
{h)
44.

(a)

=z

2
7z + 8

ez

5)2ez
= p(z)ez, for pa polynomial
p(z)
.
J(z) =
, where p and q are polynomials
q(z)
f(z) =sin z
f(z) =cot z
ez
J(z) =
, where pis a polynomial
p(z)
= (z +

Compute lnd,.(i) for the curve 'Y shown in Figure 4.14. [Hint: First
show that the integral

5 equals
{b)
(c)
45.

(a)

1
--. along the line segment from

f(

- i

-5 to

the integral along the dotted curve in the figure.]

Compute Ind,.( -i) for the curve 'Y in part

(a).

Compute lnd7(i) for the curve 'Y shown in Figure 4.15, any E > 0.
Show that if U1 U2 U3

so is U =

and if each Uj is h.s.c., then

LJ Uj. [Hint: Let f be a given holomorphic function on


U. Choose 91 an antiderivative of J on U1. Then choose 92 an
antiderivative of J on U2. By adding a suitable constant to 92, we
may arrange that 91 = 92 on U1. Continue inductively.]

Exercises

154

-2 + (1 + E) i

.---------+---y-------, 2 + (1 + E) i

-2 + (1

L....----+--' 2 + (1

- E)

- E)

Figure 4.15

(b)

Deduce from

connected.

(a)

that any half plane in C is holomorphically simply

Use the calculus of residues to calculate the integrals in Exercises 46-58:


46.

7
4 .
48.
49.
50.
51.
52.
53.
54.
55.
56.

7
5 .

1
o

+00

+oo

--dx
1 +x4
1

J oo --dx
1 +x4
+00 xl/4
--dx
1
1 +x3
o
+=
1
r
dx
}0 x3 + x+ 1
+00 1
--dx
1
1 +x3
o
+00 xsinx
--dx
1
1 +x2
o
00
fo p( ) dx where p(x) is any polynomial with no zeros
COSX

+=_x
_ dx
J_
sinhx
o00 xI/3
--5 - dx
J-oo-l+x
+oo sin2 x
J oo --dx
x2
+oo x4
- dx
J oo -1 +xlO
oo
e-ix

on the nonnegative real axis

J-oo Jx+ 2i + Jx+ 5i dx


11"
d()
58. J
-71" 5 + 3cos ()

Exercises

155

Use the calculus of residues to sum each of the series in Exercises 59-61.
Make use of either the cotangent or tangent functions (as in the text)
to introduce infinitely many poles that are located at the integer values
that you wish to study.
00

59.
60.

L: k4+ 1

k=O

I:
00

J=-oo

61.

00

j3+2

'"""' L__

j4+4

J=O

62. When a is not a real integer, prove that


00

k=-oo

71"2

(k + a)2

2
sin 7ra

63. Let U be an open set in the extended plane. Let M denote the collection

of all meromorphic functions on U. Prove that M is closed under addi


tion and subtraction.

(Note that part of the job is to find a definition

for these operations: You should formulate your definition in terms of


the Laurent expansions of the functions.)

64. Let U be an open set in the extended plane. Let M denote the collec

tion of all meromorphic functions on U. Prove that M is closed under

multiplication and division. Refer to Exercise 63 for a hint.

65. Give an alternative proof of the second statement of Theorem 4.7. 7 using
Liouville's theorem.

66. Prove that the function 1/sin (l/(1-z)) has infinitely many poles in the
unit disc

D(O, 1).

Enumerate all the poles explicitly.

67. Prove the statement contained in the third sentence of the proof of
Lemma 4.7.3.

68. Let U be any nonempty open subset of C Prove that U contains an


infinite, relatively closed subset which is discrete, and for which every
boundary point of U is a limit point.

[Hint:

Let

S1

be a set of points

lz - w l 1 and which is maximal

in U such that, for each z, w E

S1,

with respect to this property.

Of course you will need to prove that

such a set exists. Then find S2 a set of points in U such that, for each
z, w E S2, lz - w l 1/2 and which is maximal with respect to this
property. Continue defining Sn similarly for all positive integers n. Let

Sn

consist of those elements of

Sn

that have distance at most 2/n from

Exercises

156

U and distance at least

1/(2n)

from C

U. The set S

=Un Sn

will

do the job.]

69. Let U be any bounded, convex open set in C. Prove that U is h.s.c.
Work a little harder to show that a region in the shape of an S is h.s.c.
Finally, look at a region in the shape of a T.

Chapter 5

The Zeros of

Holomorphic Function

In Chapter 3, we concentrated on properties of holomorphic functions that


were essentially analytic in nature: Estimation of derivatives and conver
gence questions dominated the scene. In this chapter, we shall be concerned
more with questions that have a geometric, qualitative nature rather than
an analytical, quantitative one. These questions center around the issue of
the local geometric behavior of a holomorphic function.

Questions of this

type are, for instance: Does a nonconstant holomorphic function on an open


set always have an open image? W hat conditions on the local geometry of
a holomorphic function might force it to be constant?
These questions and many related ones have strikingly elegant answers.
We begin with some results on the zero set of a holomorphic function.

5.1. Counting Zeros and Poles

f : U re is a holomorphic function on a connected, open set


U re and that D(P, r) U. We know from the Cauchy integral formula
that the values off on D(P, r) are completely determined by the values off
on aD(P, r). In particular, the number and even the location of the zeros off
in D(P, r) are determined in principle by f on 8D(P, r). But it is nonetheless
a pleasant surprise that there is a simple formula for the number of zeros of
f in D(P, r) in terms of f (and f') on 8D(P, r). In order to construct this
Suppose that

formula, we shall have to agree to count zeros in a particular fashion. This


method of counting will in fact be a generalization of the notion of counting
-

157

158

5. The Zeros of a Holomorphic Function

the zeros of a polynomial according to multiplicity. We now explain the


precise idea.
Let f : U
C be holomorphic as before, and assume that f has zeros
but that f is not identically zero. Fix zo E U such that f(zo) 0. Since the
zeros off are isolated, there is an r > 0 such that D(zo, r) U and such
that f does not vanish on D(zo,r) \ {zo}.
Now the power series expansion off about zo has a first nonzero term
determined by the least positive integer n such that J(n) (zo) -=/=- 0. [Note that
n 1 since f(zo) 0 by hypothesis.] Thus the power series expansion off
about zo begins with the nth term:
--t

f(z)

00

1{}Jf

'' -:-- -. (zo)(z - zo)3.


]! ozJ

J=n

Under these circumstances we say that f has a zero of ordern (or multiplicity
1, then we say that zo is a simple zero off.
The concept of zero of "order n," or "multiplicity n," for a function f
is so important that a variety of terminology has grown up around it. It
has already been noted that when the multiplicity n
1, then the zero is
sometimes called simple. For arbitrary n, we sometimes say that "n is the
/3, so that, for some
order of zo as a zero of f. " More generally if f(zo)
n 1, the function f( ) - /3 has a zero of order n at zo, then we say either
that "f assumes the value /3 at zo to order n" or that "the order of the value
/3 at zo is n." When n > 1, then we call zo a multiple point of the function
n) at zo. When n

f.

The next lemma provides a method for computing the multiplicity n


of the zero zo from the values of f, f' on the boundary of a disc centered at
zo.
Lemma 5.1.1. If f is holomorphic on a neighborhood of a disc D(z0, r)
and has a zero of order n at zo and no other zeros in the closed disc, then

271"i

J
!'(() d(
Ja D(zo,r) f(()

n.

[Notice that f does not vanish on8D(zo, r) so that the integrand is a con
tinuous function of (.]
Proof.

f(z)

By hypothesis,

00

18Jf

'' -:-- -. (zo)(z - zo)3


J !8zJ

J=n

(z - zot

[00

l{}Jf

'' -:---. (zo)(z - zo)3 J ! ozJ

J=n

n .

The expression in brackets [ ] is a power series with positive radius of


convergence, indeed the same radius of convergence as the power series

5.1.

for

Counting

and

Poles

159

itself (as noted earlier).

Hence the expression in brackets defines a

Zeros

H(z), on some open disc centered at zo. Since


H(z) = f(z)/(z - z0)n when z =f zo, the function H can in fact be taken
to be defined (and is holomorphic) on the domain of f, and hence on a
neighborhood of D(zo,r).
holomorphic function, say

Notice that

H(zo)
For

D(zo,r) \ {zo},

anf
(zo) =f 0.
n. uzn
1

-1

f'(()
(( - zo)nH'(() +n(( - zor-1H(()
n
H( )
= ' ( +
=
f(()
(( - zo)nH(()
H(()
( - zo
--

By continuity,

--

is nowhere zero on a neighborhood of

H is nonzero on a neighborhood of

--.

D(zo,r) \ {zo}, hence

D(zo,r). It follows that


D(zo,r). Thus

the function

H'/His holomorphic on a neighborhood of

J'(() d( =

Jlr:-zol=r f(()

H'(()
_n_d(
d( + j
Jlr:-zol=r ( - Zo
Jlr:-zol=r H(()

O+n

1(-zol=r

27l'in.

-- d(

( - Zo

This completes the proof of the lemma.

The lemma just established is an important special case of the general


"counting zeros" formula that we wish to establish. We shall generalize the

zo of f
center of the circle of integration; second, we want to allow several
zeros of f to lie inside the circle of integration. [We could also consider more
lemma in two way s: First, it is artificial to assume that the zero

is the

general curves of integration, but we postpone that topic.]

that f: U---+ C is holomorphic on an open set


and that D(P,r) U. Suppose that f is nonvanishing on 8D(P,r)
and that z1, z2,..., Zk are the zeros of f in the interior of the disc. Let ne
be the order of the zero of f at ze. Then

Proposition 5.1.2. Suppose


U C

J'(() d(=

--

2m l<:-Pl=r f(()

-.

Proof. Set

ne.

160

for

5. The Zeros of a Holomorphic Function

z EU\ {z1, z2, ..., z } .


k
H(( )

For each

j E {1, 2, ... , k}

we write

f(()

((

1
. II
.
- Zj)ni f,#j (( - Zt)ne

The second term is clearly holomorphic in a neighborhood of Zj while the


first is holomorphic by the argument in the proof of Lemma

5.1.1.

Thus

H is holomorphic on a neighborhood of D(P, r). Also, by design, H has no


zeros in D(P, r). Calculating as in the proof of the lemma gives
H'(()

J'(()
f(()

H(()

.
t
( - Zf___
l=l __!!!_

Now H' / H is holomorphic on an open set containing D(P, r). As a result,


integration of equation ( * ) yields

!'(()

-d(

1(-Pl=r f(()

o+

1 d
k
k
i -_( 27ri:L:nt.
L:nt
l= l
l=l 1(-Pl=r ( Zf

The formula in the proposition, which is often called the

principle,

argument

is both useful and important. For one thing, there is no obvious

reason why the integral in the formula should be an integer, much less the
crucial integer that it is (see Exercise 9 for more on this matter).
The integral

_1

!'(() d(

27ri 11(-Pl=r f(()

can be reinterpreted as follows: Consider the


1(t)
Then

_1

f(P

C1

closed curve

+ reit), t E [O, 27r].

!'(() d(

27ri 11(-Pl=r f(()

_1

r2rr 1'(t)dt

27ri Jo 1(t)

as you can check by direct calculation. The expression on the right is just
the index of the curve 1-with the notion of index that we defined in Section

4.5. Thus, the


{( : I( - Pl

number of zeros off (counting multiplicity) inside the circle

r}

is equal to the index of /.

This number, intuitively

speaking, is equal to the number of times that the f-image of the boundary

circle winds around

If

in C.

is a (real-valued) continuously differentiable (not necessarily holo

morphic) function on a neighborhood of D(P, r) into


can set

9(t)

g(P+ reit),O::; t::;

integral

!(9)

2 \ {O}

C\

{O},

we

271", and associate tog the integer-valued

1
-.

27ri

12rr _!L_
' (t)
( ) dt .
0

g t

161

5.1. Counting Zeros and Poles

Even when g is not holomorphic, the integer

1(9)

tells us something about

the behavior of g about the point P-that is, it tells us about the tendency
of g to "wrap around" at P.

Thus

1(9)

has something to do with how

often points near 0 are the images, under the function g, of points near
P.

This perspective is the starting point for the topological idea of the

(local) degree of a differentiable mapping. Indeed, from this viewpoint the


argument principle is a special case of a general principle of topology that
the total of the local degrees can be computed by the behavior of g on the
boundary. A detailed treatment of this subject is beyond the scope of the
present book, but the matter will be clearer after we have talked about
homotopy in Chapter 11 (see also Exercise 9 in this chapter).
The argument principle can be extended to yield information about
meromorphic functions, too.

We can see that there is hope for this by

investigating the analogue of Lemma 5.1.1 for a pole:

Lemma 5.1.3. If f: U
on U

\ {Q}

\ {Q}

-t

<C is a nowhere zero holomorphic function

with a p ole of order n at

and if D(Q,r) U, then

f'(()
d(
JaD(Q,r) f(()

1 J
_
27ri

-n

Proof. By the discussion at the end of Section 4.3,

(z - Q) n f(z)

has a

Q. Denote the associated holomorphic function on a


neighborhood of D(Q,r) by H [i.e., H(z)
(z-Q) n f(z) on a neighborhood
of D(Q,r) \ {Q}]. Then, for ( E D(Q,r) \ {Q},
removable singularity at

Q) n-1J(() + (( Q) nJ'(()
(( - Q) nJ(()
n
'
+ f (()
.
( - Q f(()

H'(()
H(()

((

Now we integrate both sides over {) D(Q,r) ; since

H'/ H is holomorphic, the

integral on the left side vanishes. We obtain

f'(()
8D(Q,r) f(()

-- d(

-n

i8D(Q,r) ( - Q

--d(

-27rm.
.

Just as with the argument principle for holomorphic functions, this new
argument principle, together with Lemma 5.1.1, gives a counting principle
for zeros and poles of meromorphic functions:

Theorem 5.1.4 (Argument principle for meromorphic functions). Suppose


that

is a meromorphic function on an open set U <C, that

D( P, r)

162

5. The Zeros of a Holomorphic Function

on 8D(P, r ) . Then

and that f has neither poles nor zeros

1 1
!'(()
27ri fa D( P,r ) f(() d(

f;

nj

{;
q

mk,

where n1, n2, ... , np are the multiplicities of the zeros

D(P, r ) and
in D(P, r ) .

m1, m2, ... , mq

z1, z2, ... ,

are the orders of the poles

Zp of f in

w1, w2, ... , Wq

Proof. Exercise for the reader.

off

5.2. The Local Geometry of Holomorphic Functions


The argument principle for holomorphic functions (the formula of Proposi
tion

5.1.2) has a consequence which is one of the most important facts about

holomorphic functions considered as geometric mappings:


Theorem 5.2.1 (The open mapping theorem). If f : U

---+

C is a noncon

stant holomorphic function on a connected open set U, then J(U) is an open


set in C.
Before beginning the proof of the theorem, we discuss its significance. The
theorem says, in particular, that if U C is connected and open and if
f : U

---+

C is holomorphic, then either J(U) is a connected open set (the

nonconstant case) or J(U) is a single point. There is no analogous result for


2

C00, or even real analytic functions from C to C (or from IR to IR

) . As an

example, consider the function

g:C

---+

l z l 2.

The domain of g is the entire plane C , which is certainly open and connected.
The set g(C) , however, is

{x+iO: IR 3 x 2'.: O}

which is not open as a subset

of C. The function g is in fact real analytic, but of course not holomorphic.


Note, by contrast, that the

holomorphic function

g:C

---+

has image the entire complex plane (which is, of course, an open set). More
significantly, every open subset of C has image under g which is open.
In the subject of topology, a function f is defined to be continuous
if the inverse image of any open set under f is also open.
where the

{J definition makes sense, the

In contexts

- {J definition is equivalent

to the inverse-image-of-open-sets definition. Functions for which the direct


image of any open set is open are called "open mappings" (note that this

5.2. Local Geometry of Functions

163

concept is distinct from continuity). Theorem 5.2.l says that a nonconstant


holomorphic function on a connected open set
the name of Theorem

5.2.1

C is an open mapping;

is derived from this property.

Proof of Theorem 5.2.1. We need to show that, given Q E

J(U),

there

f(U)
Select PE U with J(P)
Q. Set g(z)
J(z) - Q. Since g(z) vanishes
at z
P and is nonconstant, there is an r > 0 such that D(P, r ) U and
g does not vanish on D(P, r ) \ {P}. Suppose that g vanishes at P to order

is a disc D(Q,

n for somen

l.

Then the argument principle asserts that

1 J
!'(()
_
i
27l' fav(P,r) f(()

-Q

The nonvanishing of
imply that there is an

g(()
E >

on 8D(P,

point of

f(U),

E,

on 8D(P,

r .

J(U). Since Q is an arbitrary


would establish the openness of J(U).

the disc D(Q,

this claim

and the compactness of 8D(P,

0 such that

lg(()I >
We claim that, for this

d( =n.

lies in

To prove the claim, we define the function


N(z)

l
j
(' () d(,
_
27l'i fav(P,r) f () - z
(

zE

D(Q,

Nate that the denominator of the integrand does not vanish since if
D(Q,

E ,

then

If(() - zl If(() - QI

l z - QI

> E

Then N is a continuous--indeed a C00-function

lz - QI > 0.
of z E D(Q, E )

(see Ap

pendix A). But N is integer-valued, since (according to Proposition


it counts the zeros of

J(-) - z

in D(P,

r .

5.1.2)
and

Since N is both continuous

integer-valued, it must be constant. But we already know that N(Q)


so that N must be constantly equal ton on D(Q,

g(()

Since n
=

1,

J(() - z

is proved.

we see that, for each fixed

zE

D(Q,

E .

n,

zE

Therefore D(Q,

E .

D(Q,

vanishes at some point(s) of D(P,

takes on each value

zE

r .

E ,

the function

In other words,

J(U),

and our claim

The proof that we have just presented actually provides us with con
siderable additional information about the local behavior of a holomorphic
function. The intuitive idea is

as

follows: Fix Q in the image of the holomor

phic function f. If the holomorphic function f takes the value Q at the point
P to order k, then f behaves very much like the function (z) Q+(z-P)k
=

164

near

P.

Zeros of a

Holomorphic Function

This statement is true in the sense that every point

borhood of Q has k preimages near


is

The

5.

in

neigh

while the only preimage of Q near

itself (to kth order). In view of the fact that the power series expansion

for f about

will have the form


00

J(

z)

Q+

L:aj(z

P)J,

j=k
this fact is not surprising.
Let us now give a precise formulation of the idea we have been dis
cussing:

Suppose that f : U -> C is a nonconstant ho lorn orphic


function on a connected open set U such that P E U and f(P) = Q with
order k. Then there are numbers 8, t: > 0 such that each q E D(Q,t:) \ {Q}
has exactly k distinct preimages in D(P, 8) and each preimage is a simple
point of f.
Theorem 5.2.2.

We begin the proof of the theorem with a lemma about the multiple points
of a holomorphic function.

Let f : U -> C be a nonconstant holomorphic function on a


connected open set U C. Then the multiple points off in U are isolated.

Lemma 5.2.3.

Proof. Since f is nonconstant, the holomorphic function J' is not identically


zero. But then Theorem 3.6.1 tells us that the zeros of f' are isolated. Since
any multiple point p of f has the property that J'(p)

0,

it follows that the

multiple points are isolated.

> 0 such
{P } is a simple point off. Now choose 8, E > 0
as follows: Take 0 < 8 < 81 such that Q t/. f(D(P, 8) \ {P }). Choose
E > 0 such that D(Q,t:) f(D(P, 8)) and so that D(Q, t:) does not meet
J (8D(P, 8)); this choice is possible because f(D(P, 8)) is open by Theorem
5.2.1 and because f is continuous. Then, by the reasoning in the proof of
Theorem 5.2.1, for each q E D(Q, t:) \ {Q},

Proof of Theorem 5.2.2. According to the lemma, there is a 81

that every point of D(P, 81) \

1
.

27ri

!'(() d(
- q
f(()
8D(P,8)

[order off at

By Proposition 5.1.2, if PI,p2, ... , Pe are the zeros


ni, n2, ... , ne are their orders, then

ni + n2 +

81 ,

+ ne

P]

off() - q

>

in

0.
D(P, 8),

and

k.

each nj
1 for j
1, ... , f. It follows that f
must equal k. That is, the point q has precisely k preimages p1,p2, ... ,pk,
But, by the choice of

<

each of which is a simple point of f.

165

5.2. Local Geometry of Functions

It is important for the reader to understand clearly that none of the


results being discussed here asserts anything like the claim that the image of
every disc is a disc or that the preimage of every disc is a disc. W hat
is that the image of a disc about

will always

contain

is true

f(P).

a disc about

This statement of openness is precisely what we need for the applications


that follow.
Special interest is attached to the case of Theorem

J'(P) #

in which

0. In this case, f is locally one-to-one in the precise sense, de

rived in Theorem
in

5.2.2

D(P, 8).

can define a function

1-1 : D(Q, E )

1-1 (w )
Of course

that each point of

5.2.2,

D(Q, E )

has exactly one preimage

One says, in this situation, that f is locally invertible. Also we

--+

C by

the unique preimage of

in

D(P, 8).

may have other f-preimages elsewhere in U, but we ignore these

here. It is natural to ask if

1-1 is holomorphic on D(Q, E) .

Furthermore, in

fact, this is always true: "The local inverse of a locally invertible holomorphic
function is holomorphic" is the way that we usually describe this fact in
summary language.
In order to prove that

Q.

continuous at

f.1

>

0 so that

J-1(D(Q, fl))

can suppose that


(Theorem

1-1 is holomorphic,

For this, suppose that

5.2.1),

a <

D(P, a) .

the previous

the image f

(D(P, a) )
D(Q, fl)

fl

choose

to be less than the previous

fl

we shall first prove that it is

0 is given. We need to choose

W ithout loss of generality, we

Now, by the open mapping theorem

8.

for some

>

a >

0 we know that

E.

is an open set containing

f(D(P, a)) .

Q,

so

We can and shall

Now each point in

D(Q, E)

and

in D(Q, fl) has only one preimage in D(P, 8), while each point in
D(Q, Ei ) has at least one preimage in D(P, a) since J(D(P, a)) ;;;> D(Q, fl).
So 1-1 (D(Q, E )) D(P, a) as required. Thus 1-1 is continuous at Q.
1
Next we show that 1-1 is complex differentiable at Q. For this, we

fortiori

need to show that

1-l(Q')-1-l(Q)
Q'-Q
Q'--+Q,Q''IQ
lim

exists. We note that

1-l(Q')-1-l(Q)
Q'-Q

where we have written

P'

f-1(Q'),
r

P11 '!!!p

Q'-Q
1-l(Q')-J-l(Q)
J(P')-J(P)
'
P '-P

I(
1
I

and of course

J(P ')-J(P )
P '- p

1-1(Q)

P.

But

166

5. The Zeros of a Holomorphic Function

f'(P). Moreover, f'(P) =/:- 0 by hypothesis.


that P' --+ P as Q' --+ Q. Therefore

exists and equals


of

1-1

implies

exists and equals

The continuity

1/f'(P).
P and Q in these
D(Q, 8 ) and set P' f-1(Q'), then we can
that 1-1 on D(Q, E ) is continuous at Q' as

Finally, note that there is nothing magical about


arguments. If we choose

Q'

repeat the reasoning to show

[This point is slightly trickier than it sounds;

well.

We also obtain that

(f-1 )'(Q')

1/f'(P').

think about it carefully.

Notice that

f'(P')

is nonzero,

because of the way that 8 is chosen in Theorem 5.2.2. So at last we con


clude that

1-1

is complex differentiable everywhere on

continuous complex derivative

(f-1 ) '.

So

1-1

D(Q, E )

and has a

is holomorphic by Theorem

2.2.2.
This looks almost too good to be true. But oddly enough, for once, the
same argument works in one-variable real calculus: You can run through
it for yourself to prove that if
is nowhere zero in

u-1)1(y)

( a, b),

1/f'(f-1(y)).

then

f:
f

( a, b) --+

JR. is a

C1 function and if f'(x)


f((a, b)) to ( a, b) and

is invertible from

Actually, the real-variable case is even a little

better than the complex case, because you do not have to shrink the do
main and range to obtain a well-defined function

1-1.

5.3. Further Results on the Zeros of Holomorphic Functions


In the previous sections of this chapter, we have developed a detailed un
derstanding of the local behavior of holomorphic functions, that is, of their
behavior in a small neighborhood of a particular point. The methods we
used, and especially Proposition 5.1.2, can be applied in a wider context
to the "global behavior" of a holomorphic function on its whole domain of
definition. In this section we state and prove two important results of this
sort.

( Rouche's theorem ) . Suppose that f, g: U--+ C are holo


fu n ctions on an open set U C. Suppose also that D(P, r ) U
and that, for each ( E 8D(P, r ) ,

Theorem 5.3.1
morphic

If(() - g(()I
Then

_1

!'(() d(
1
27l'i fav(P,r) f(()

<

lf(()I + lg(()I.
_1

1
27l'i fav(P,r)

g'(() d(.
g(()

5.3.

167

Further Results About Zeros

That is, the number of zeros off in D(P, r ) counting multiplicities equals
the number of zeros ofg in D(P, r ) counting multiplicities.
Before beginning the proof of Rouche's theorem, we note that the (at first

f(() nor g(() can vanish


f nor g vanishes identically; moreover,

strange looking) inequality(*) implies that neither


on

8D(P, r ) .

In particular, neither

off'/ f and of g/' g on 8D(P, r ) are defined.


Also, (*) implies that the function f(()/g(() cannot
{x + iO: x:::; O} for any ( E 8D(P, r ) . If it did, say
f(()
= .x < 0
g(()
for some ( E 8D(P, r ) , then
f(()
-1
\.X - 1\
g(()
-.X + 1
f(()
+ '
g(()
l
the integrals

l l

hence
This equality

take a value in

If(() - g(()\ = If(()\+ \g(()\.


contradicts ( *).

The observation in the last paragraph implies immediately that, for all

t E [O, 1]

and every

( E 8D(P, r ) , the complex number


tf(() + (1 - t)g(()

is not zero. This fact will play a crucial role in the proof of the theorem.

Proof of Rouche's theorem. Define a parametrized family of holomor

ft : U
C, t E [O, 1], by the formula ft(()
tf(() +
(1 - t)g((). We have already noted that ft is nonzero for t E [O, 1] and
( E 8D(P, r ) . Consider the integral
f (()
1
d(, t E [O, 1].
j
It:= 27ri J'a n(P,r) ft (()
Then It is a continuous function oft E [O, 1] (we use here the facts that
phic functions

----+

the denominator of the integrand does not vanish and that the integrand
depends continuously on

t).

By Proposition

5.1.2,

It takes values in the

integers; therefore It must be constant. In particular, Io= I1. That is what


we wished to prove.

Remark: Rouche's theorem is often stated with the stronger hypothesis


that

If(() - g(()\

<

\g(()\

5. The Zeros of a Holomorphic Function

168

for

E 8D ( P, r ) . Rewriting this hypothesis as

l fg(((())_l l <l ,
we see that it says that the image 'Y under
the disc D ( l,

1).

f/ g

of the circle 8D ( P, r ) lies in

Our weaker hypothesis that lf(()-g(()I< lf(()l+lg(()I has

f (()/ g(() lies in the set C\ { x+iO : x O}.


Either hypothesis implies that the image of the circle 8D ( P, r ) under f has
the same "winding number" around 0 as does the image under g of that
circle ( see Section 4.5). This point is especially clear from an intuitive point
of view in the case of the If(() - g(()I< [g(()I assumption: If you walk your
dog ( on a leash ) in a park with a flag pole in the middle, and if you keep the
the geometric interpretation that

leash always shorter than your distance to the flag pole, then you and your
dog must have gone around the flag pole the same number of times when
you both come back to the starting point of the walk. This is the underlying
geometric interpretation of the theorem.

EXAMPLE 5.3.2. Let us determine the number of roots of the poly nomial

f(z)
f to

z7 + 5z3 - z - 2 in the

unit disc. We do so by comparing the function

the h olomorphi c function

g(z)

5z3

on the unit circle. For

lzl

we

have

lf(z) - g(z)I
By Rouche's theorem,

lz7 - z - 21<4< If(()!+ [g(()[.

and

have the same number of zeros, counting

multiplicity, in the unit disc. Since

has three zeros, so does

f.

Rouche's theorem provides a useful way to locate approximately the


zeros of holomorphic functions which are too complicated for the zeros to be
obtained explicitly. As an illustration, we analyze the zeros of a nonconstant
polynomial
P ( z)
If

-2
zn + an-IZn 1 + an-2Zn +

is sufficiently large

( say R

> max

+ ao.

[l , n max1::;j::;n-1 [aj[])

and

lzl

R,

then

n-1 + an-2Zn-2 + . . . + ao
Ian-IZ
I l
< .
z
l ln
Thus Rouche's theorem applies on D ( O, R) with f(z)
zn and g(z) P (z) .
We conclude that the number of zeros of P ( z ) inside D ( O, R), counting mul
tiplicities, is the same as the number of zeros of zn inside D(O R), counting
=

multiplicities, namely n. Thus we recover the fundamental theorem of alge


bra

( Theorem 3.4.5).

Incidentally, this example underlines the importance

of counting zeros with multiplicities: The function zn has only one root in
the na"ive sense of counting the number of points where it is zero.

5.4.

The Maximum Modulus Principle

169

A second useful consequence of the argument principle is the following


result about the limit of a sequence of zero-free holomorphic functions.

Suppose that U c;;;; C is a connected


open set and that {fj} is a sequenc of nowhere vanishing holomorphic func
tions on U. If the sequence U1} converges uniformly on compact subsets of
U to a (necessarily holomorphic) limit function Jo, then either Jo is nowhere
vanishing or Jo = 0.

Theorem 5.3.3 (Hurwitz's theorem).

Jo is not identically zero


PE U. Chooser> 0 so small that D(P,r) c;;;; U
and Jo does not vanish on D(P,r) \ {P}. By Proposition 5.1.2,
JM() d(
_1 1
( )
27ri 11(-Pl=r Jo(()
is a positive integer, equal to the order of the zero of J at P.
Proof. Seeking a contradiction, we suppose that

but that

Jo(P)

0 for some

On the other hand, each

equals 0 since each


in line

fj

(** )

J1 is

Jj(() d
_1 1
(
27ri 11(-Pl=r J1(()

( ** )

nowhere vanishing. Also, as

converge to the integral in line

(* )

j ----;

(we use here the fact that

Jo uniformly and Jj ----; J uniformly on {( : I( -Pl

----;

the integrals

oo,

r}; see Appendix

A). This is a contradiction, for of course a sequence of zeros cannot converge

to a positive integer.

The reader should contrast the Hurwitz theorem just proved with the
real-variable situation. For instance, each of the functions

1/j, j

1, 2, ...,

x2 +

is nonvanishing on the entire real line. But they converge,

uniformly on compact sets, to the limit function

Jo has

fj ( x )

a double zero at

J0 ( x)

2. The

function

0. So the real-variable analogue of Hurwitz's

theorem fails.
However, if we "complexify" and consider the functions fj (z)
and the limit function

Jo(z) z2, then we see


J1 are not zero free:
=

that

compact sets. But now the

J1 ----; Jo,

fj

coalesce into the double zero of

viewpoint, the double zero of

z2

z2+1 /j

uniformly on

Each has two zeros and the

situation is clearly consistent with Hurwitz's theorem. As


distinct zeros of each

j ----;

Jo,

oo,

the two

and from this

at the origin arises naturally in the limit

process.

5.4. The Maximum Modulus Principle


Consider the C00 function
Notice that

<

lg(z)I 2

on the unit disc given by

and that

g(O)

2. The

g(z)

lzl2.

function assumes an

170

5. The Zeros of a Holomorphic Function

interior maximum at

0. One of the most startling features of holomorphic

functions is that they cannot behave in this fashion: In stating the results
about this phenomenon, the concept of a connected open set occurs so often
that it is convenient to introduce a single word for it.

Definition 5.4.1. A domain in


domain is a connected open set
for all

zEU .

Theorem 5.4.2

( The

such that there is an R > 0 with

maximum modulus principle ) . Let

f be a holomorphic function on U. If there is a


f
P
(
l
)I lf(z)I for all zEU, then f is constant.

main. Let
that

Proof. Assume that there is such a

P.

If

>

lf(P)I.

This is a contradiction. Hence

lzl <

C be a do
PEU such

point

f(U) is
f(U) with

is not constant, then

open by the open mapping principle. Hence there are points

1(1

A bounded

is a connected open set.

of

is a constant.

Here is a consequence of the maximum modulus principle that is often


useful:

Corollary 5.4.3
domain.

Let

( Maximum

modulus theorem ) . Let

be a continuous function on

Then the maximum value of Ill on

be a bounded

that is holomorphic on

(which must occur, since

U.

is closed

and bounded) must occur on au.

Proof. Since

lfl

is a continuous function on the compact set

U,

then it

must attain its maximum somewhere.


If

is constant, then the maximum value of

in which case the conclusion clearly holds.


maximum value of

Ill on

the compact set

If

IJI occurs

at every point,

is not constant, then the

cannot occur at

PEU,

by the

theorem; hence the maximum occurs on au .

Here is a sharper variant of the theorem, which also follows immedi


ately:

Theorem 5.4.4. Let


function on
then

U.

be a domain and let

If there is a point

PEU

at which

IfI

be a holomorphic

has a local maximum,

is constant.

P such that the


P ( by the definition of local max
imum ) . The theorem then tells us that f is constant on V. By Corollary
3.6.2 ( and the connectedness of U), f is constant on U.
D
Proof. Let V be a small open neighborhood of the point
absolute maximum of

Ill on

V occurs at

Holomorphic functions { or, more precisely, their moduli ) can have inte
rior minima. The function

f(z)

z2

has the property that

0 is a global

5.5. The Schwarz Lemma

171

minimum for lfl. However, it is not accidental that this minimum value is
0:
Proposition 5.4.5. Let f be holomorphic on a domain U C. Assume
that f never vanishes. If there is a point PE U such that \f(P)I lf(z)I
for all zE U, then f is constant.
Proof. Apply the maximum modulus principle to the function g(z)

1/f(z).
D

It should be noted that the maximum modulus theorem is not always


true on an unbounded domain. The standard example is the function f(z)
exp(exp(z)) on the domain U
{z
x + iy : -7r/2 < y < 7r/2}. Check
for yourself that Ill
1 on the boundary of U. But the restriction of f to
the real number line is unbounded at infinity. The theorem does, however,
remain true with some additional restrictions. The result known as the
Phragmen-Lindelof theorem is one method of treating maximum modulus
theorems on unbounded domains (see [RUD2]).
=

5.5. The Schwarz Lemma

This section treats certain estimates that bounded holomorphic functions on


the unit disc necessarily satisfy. At first sight, these estimates appear to be
restricted to such a specific situation that they are of limited interest. But
even this special situation occurs so often that the estimates are in fact very
useful. Moreover, it was pointed out by Ahlfors [AHLI] that these estimates
can be interpreted as a statement about certain kinds of geometric structures
that occur in many different contexts in complex analysis. A treatment of
this point of view that is accessible to readers who have reached this point
in the present book can be found in [KRA3]. This section presents the
classical, analytic viewpoint in the subject.
Proposition 5.5.1 (Schwarz's lemma).

Let

be holomorphic on the unit

disc. Assume that

( 1 ) lf(z)I 1
(2) f(O) 0.

for all

z,

lf(z)I \zl and lf'(O)I 1.


If either lf(z)I
\z\ for some z i- 0 or if \f'(O)\ 1, then f is a rotation:
f(z) = az for some complex constant a of unit modulus.
Then

Proof. Consider the function g(z) f(z)/z. This function is holomorphic


on D(O, 1 ) \ {O}. Also limz_,o g(z) f'(O). So if we define g(O) J'(O), then
=

5. The Zeros of a Holomorphic Function

1 72

g is continuous on D(O, 1). By the Riemann removable singularities theorem


(Theorem

4.1.1),

g is then holomorphic on all of D(O, 1).

Restrict attention to the closed disc D(O, 1

::; 1/(1::; 1/(1 )


::; 1

On the boundary of this disc, lg(z)I


theorem then implies that lg(z)I
Letting

).

on D

lzl for some z

then lg(z)I

I- 0,

for

>

and small.

1.

).

D ( 0, 1). In other words,

f'(O) so we also see that lf'(O)I

the proof of the first half of the theorem.


If lf(z)I

The maximum modulus

-1: on this entire disc D(O, 1

o+ then yields that I g(z)I

E --->

lf(z)I ::; lzl. Now g(O)

::; 1.

That completes

Since lg(z)I ::;

on the

entire disc, we conclude from the maximum modulus principle that g(z) is
a constant of modulus
If instead lf'(O)I

1.

Leto: be that constant. Then f(z)

1,

then, as previously noted, lg(O)I

o:z.

1.

Again, the

maximum modulus principle implies that g is a constant of absolute value

1,

and again we see that f is a rotation.

Schwarz's lemma enables one to classify the invertible holomorphic self


maps of the unit disc. [These are commonly referred to as the ''conformal
self-maps" of the disc.] We shall treat that topic in Section 6.2.
We conclude this section by presenting

generalization of the Schwarz

lemma, in which we consider holomorphic mappings f : D


discard the hypothesis that f(O)

0.

Pick lemma.

--->

D, but we

This result is known as the Schwarz

Theorem 5.5.2 (Schwarz-Pick). Let f be holomorphic on the unit disc


with lf(z)I
b

::; 1

for all z E D(O, 1).

Then, for any a E D(O, 1) and with

f(a), we have the estimate

::; = 11 11 : .

lf'(a)I
Moreover, if f(ai)

b1 and f(a2)

b2 , then

There is also a uniqueness statement for the Schwarz-Pick theorem; we shall


discuss it later.
Proof of Theorem 5.5.2. If c

D(O, 1), then set

rPc(z)
Notice that

z-c
=

-_-.

1-

--

1- CZ
_

CZ

I <1

173

5.5. The Schwarz Lemma

if and only if

lz - c l2 < 11- ez l2
if and only if

lz l2 - 2Reez + lcl2 < 1- 2Reez + lcl21z l2


if and only if

if and only if

lz l < 1.
It follows that

<Pc

is a holomorphic mapping from

Elementary algebra shows that


Now, for the function

Then

</Jc

D(O, 1)

to itself.

is invertible and that its inverse is

<P-c

given in the theorem, we consider

satisfies the hypotheses of the Schwarz lemma

We conclude that

( Proposition 5.5.1).

lg'(O)I 1.
This inequality, properly interpreted, will give the first conclusion of the
lemma:
Note that

a(O) =1 - lal2 and (b) =1/(1 - lbl2). Therefore

12:lg'(O)I=1(b)l IJ'(a)l 1'__a(O)I= (l

?'(a)I.

-l 2
l

The first conclusion of the lemma follows.


For the second conclusion, let us take

h =b1
Schwarz's lemma implies that

lb1

</J-a1

lh(z )I lz l. That

is,

f 0 <P-a1 (z )I lz l

or

Taking

z =a

gives

Writing this out, with the definition of the </J's, gives the desired conclusion.

174

5. The Zeros of a Holomorphic Function

There is a "uniqueness" result in the situation of Theorem


corresponds to the last conclusion of Proposition

IJ'(a)I
then the function

1 - lbl2
1 - lal2

5.5.l.

5.5.2

that

If either

I 1b2 -b1bb1 I I 1a2- -a1aa1 I '

or

is a conformal self-mapping (one-to-one, onto holomor

phic function) of D(O,

1)

to itself.

The proof of this statement consists in reducing it to the corresponding

5.5.1,

statement for Proposition

5.5.2.

using the methods of the proof of Theorem

Since we consider the conformal self-maps of the disc in Section

6.2,

we

shall ask the reader to consider these uniqueness statements as an exercise

at that time.

Exercises
1. Let

be holomorphic on a neighborhood of D(P, r ) . Suppose that

not identically zero on


zeros in

2. Let

f, g

zeros at

D(P, r ) .

D(P, r ) .

Prove that

be holomorphic on a neighborhood D(O,

Pi, P2, ... , Pk

boundary circle of D(O,

D(O,

1),

1
-.
27ri

1)

is

has at most finitely many

1).

Assume that

and no zero in 8D(O,

1).

has

Let"( be the

traversed counterclockwise. Compute

i f'(z)
f(z)

"f

g(z)dz.

3. Give another proof of the fundamental theorem of algebra as follows:


Let

P(z)

be a nonconstant polynomial. Fix a point Q E C. Consider

P'(z)
_1_ 1
dz.
27ri la n( Q,R ) P(z)
Argue that, as R----+ +oo, this expression tends to a nonzero constant.

4. Without supposing that you have any prior knowledge of the calculus
function ex, prove that
00

ez

L
!
k=O

never vanishes by computing (ez)' / ez, and so forth.

5. Let

fj

D(O,

1)

----+

C be holomorphic and suppose that each

least k roots in D(O,

1),

counting multiplicities.

Suppose

fj has at
that fj
f
----+

uniformly on compact sets. Show by example that it does not follow that

has at least k roots counting multiplicities. In particular, construct

examples, for each fixed k and each, 0 k, where

has exactly

175

Exercises

roots. What simple hypothesis can you add that will guarantee that
have at least k roots? (Cf. Exercise 8.)

does

: D(O, 1) - C be holomorphic and nonvanishing.


Prove that
has a well-defined holomorphic logarithm on D(O, 1) by showing that
the differential equation

6. Let

azg(z)
8

f'(z)
f(z)

has a suitable solution and checking that this solution


7. Let U and V be open subsets of C. Suppose that

phic, one-to-one, and onto. Prove that


on V.

f:

8. Let

-f 1

U - C be holomorphic. Assume that

a (P r).

g
f a (P r) ,

f:

g does the job.

U - V is holomor

is a holomorphic function

(P, r)

fis
and g is

U and that

nowhere zero on D ,
Show that if is holomorphic on U
sufficiently uniformly close to on D ,
then the number of zeros
of in D
equals the number of zeros of in D
(Remember
to count zeros according to multiplicity.)

9. Define

(P, r)
f(z) z
=

(P, r).

z. Let 'Y be the counterclockwise oriented boundary of

D(O, 1). Show that, in some sense,

of zeros off in D(O, 1) counting multiplicities] =/:-

2m f-y

f'(() d(.
f(()

[H n t : You must first decide on an interpretation of the integral


remember that this is a
statement. A good idea would
be to set

f-y

real-variable
f'(() d( - {2rr [(d/dt ) f(eit)] dt.
f(eit)
f(()
lo

This would make sense and would equal the integral as usually defined
in case is holomorphic. Then you have also to figure out what 'mul
tiplicity' means in the present context: See the discussion later in this
exercise.] What is actually true in general is that if E C1(D(O, 1)) and
does not vanish on D O, 1), then, with the integral interpreted as we
have just indicated,

a (

I 2i i (] d(I I#

of zeros off in D(O, 1) counting multiplicities .

f
f(P)
a:
P.

To "prove" this inequality, suppose that


vanishes to first order
at a point
E D(O, 1) [i.e.,
0 but \1
=/:- 0]. Then, by
Taylor's formula, there are complex constants and {3 such that
+ {3(z
+(error) when is near
If 'Y is a small curve

a:(z - P)

f(P)

- P)

f(z)

f
f

176

Exercises

surrounding P, then calculate

l
_
27ri
The answer will be

or

1,

1 !'(() d(

f(()

depending on whether o: and

{3

are zero or

nonzero. Now treat the general case for simple zeros by breaking D(O,
up into smaller regions, some of which surround the zeros of
of which do not.

What additional idea is needed in case

of multiplicity greater than one?

[Hint:

f and
f has

)1

some
zeros

In order to study this problem

successfully, you will need to come up with a notion of multiplicity of


a smooth (not necessarily holomorphic) function
isolated point P where

f(P) = 0.

of x and y at an

The basic idea is that the multiplicity

at a point is obtained by choosing a small circle around the point and


evaluating the "generalized integral"

1
27ri

J'

ff

as already discussed. The reference [MIL] will help you to come to grips
with these ideas. Treat this as an open-ended exercise.]

10. Estimate the number of zeros of the given function in the given region
u.

( a)
(b)
(c)
(d)
(e)
(f)

f(z) = z8 + 5z7 20,


f(z) = z3 - 3z2 + 2,
f(z) = z10 + lOz + 9,
f(z) = z10 + lOzez+l 9,
f(z) = z4e- z3 + z2 /6- 10,
f(z) = z2ez - z,

U = D(0 , 6 )

U = D(O,

)1
U = D(O, 1)
U = D(O, 1)
U = D(O, 2)
U = D(O, 2)

11. Imitate the proof of the argument principle to prove the following for
mula: If

: U

--->

C is holomorphic in U and invertible, P

U, and if

D(P, r ) is a sufficiently small disc about P, then

l
1-1(w)= _
27ri

(f'(() d(
j
laD(P,r) f(() - W

for all win some disc D(J(P), r1 ) , r1 >

0 sufficiently small.

Derive from

this the formula

l
u-1 )'(w)= _
27ri

(f'(()
1
d(.
laD(P.r) (!(() - w)2

Set Q = J(P). Integrate by parts and use some algebra to obtain

u-l)'(w)=-1
27ri

1
1
laD(P,r) f(() Q
-

) (
.

1
-

w-Q
-1 d(.
f(() -Q

(*

Exercises

177

Let ak be the kth coefficient of the power series expansion of

Q:

the point

f-1 ( w )
(* )

Then formula
this!

L ak(w - Q)k.
00

k=O

to obtain

27ri

(n
12. Suppose that

U.

1
1
d(
laD(P,r) [J(() - Q]n
1
a n-l (( - P)n
1 )!
8(
[!(() Q]n (=P

( )

f'

zeros inside

U?

Can you

( Here the closed convex hull of a


S is the intersection of all closed convex sets that contain S.) [ Hint:

If the zeros of

f'. ]

14. Let

- 1

is a polynomial on C, then the zeros off' are contained in

the closed convex hull of the zeros of

of

(n
f'?

has

conclude anything about the zeros of

is holomorphic and has n zeros, counting multiplicities,

Can you conclude that

13. Prove: If

Lagrange's formula.

This is called

set

about

may be expanded and integrated term by term prove

1
__

inside

1-1

Pt(z)

are contained in a half plane V, then so are the zeros

be a polynomial in

Suppose that

f.

Pt(z)

t, 0

for each fixed value of

is continuous in

<

<

1.

in the sense that

Pt(z)
and each

a1(t) is continuous.

Z is closed in C

[O, 1 ].

If

L a1(t)zl

j=O

Let Z =

Pt0(zo)

{(z , t) : Pt(z)

0 and

O}.

(8/8z) Pt0(z)

By continuity,

z z
= o

# 0,

then

show, using the argument principle, that there is an <: > 0 such that for

to there is a unique z E D(zo, 1: )


Pt0( ) vanishes to order k at zo?

sufficiently near

can you say if

with

Pt(z)

0. What

15. Complete the following outline for an alternative argument to see that
the function

( a)

( b)
(c)

N(z)

in the proof of the open mapping theorem is constant:

Calculate that

J'(()
J
d(.
27ri !aD(P,r) (!(() - z)2
integrand is the derivative, in ( ,

!._N(z)
8z

1
-

z, the
function H(() on a neighborhood of D(P, r).
It follows that (8/8z)N(z) = 0 on D(P, r).
For fixed

of a holomorphic

178

Exercises

(d)

N(z)

is constant.

16. Prove that if f: U

----t

C is holomorphic, PE U, and f'(P)

0, then f

is not one-to-one in any neighborhood of P.

17. Prove: If f is holomorphic on a neighborhood of the closed unit disc


D and if f is one-to-one on 8D, then f is one-to-one on D.

[Note:

Here

you may assume any topological notions you need that seem intuitively
plausible. Remark on each one as you use it.]

18. Let

Pt(z)

ao(t) + a1(t)z +

an(t)zn

be a polynomial in which the

coefficients depend continuously on a parameter tE


if the roots of

Pto

Prove that

are distinct (no multiple roots), for some fixed value

of the parameter, then the same is true for


to

( -1, 1).

to-provided that
Pto

the degree of

Pt

Pt

when tis sufficiently close

remains the same as the degree of

Chapter 6

Holomorphic Functions
as Geometric Mappings

Like Chapter

5,

this chapter is concerned primarily with geometric questions.

While the proofs that we present will of course be analytic, it is useful to


interpret them pictorially. The proofs of the theorems presented here arose
from essentially pictorial ideas, and these pictures can still serve to guide our
perceptions. In fact geometry is a pervasive part of the subject of complex
analysis and will occur in various forms throughout the remainder of the
book.
The main objects of study in this chapter are holomorphic functions
h : U

V, with U, V open in C, that are one-to-one and onto.

---->

a holomorphic function is called a

conformal

(or

biholomorphic)

Such

mapping.

The fact that his supposed to be one-to-one implies that h' is nowhere zero
on U [remember, by Theorem
point P E U , then h is

(k +

5.2.2,

that if h' vanishes to order

1)-to-1 in a small neighborhood of P].

result, h-1 : V----> U is also holomorphic (see Section


h: U

---->

5.2).

at a
As a

A conformal map

V from one open set to another can be used to transfer holomorphic

functions on U to V and vice versa: That is,


and only if
only if g

k ;:::: 0

f : V

---->

C is holomorphic if

his holomorphic on U; and g : U----> C is holomorphic if and

h-1 is holomorphic on V.

Thus, if there is a conformal mapping from U to V, then U and V are


essentially indistinguishable from the viewpoint of complex function theory.
On a practical level, one can often study holomorphic functions on a rather
complicated open set by first mapping that open set to some simpler open
set, then transferring the holomorphic functions as indicated.

179

6. Holomorphic Functions

180

as

Geometric Mappings

The object of the present chapter is to begin the study of the biholo
morphic equivalence of open sets in C. We shall return to the subject in
later chapters.

6.1. Biholomorphic Mappings of the Complex Plane to Itself


The simplest open subset of C is C itself. Thus it is natural to begin our
study of conformal mappings by considering the biholomorphic mappings of

C to itself. Of course, there are a great many holomorphic

functions

from

C to C, but rather few of these turn out to be one-to-one and onto. The
techniques that we use to analyze even this rather simple situation will intro
duce some of the basic ideas in the study of mappings. The biholomorphic
mappings from C to C can be explicitly described as follows:

Theorem 6.1.1. A function

f: C-+ C is a conformal mapping


numbers a, b with a =/:- 0 such that

if there are complex

J(z)

az + b , z

C.

One aspect of the theorem is fairly obvious: If


the map

az

r-;

if and only

a, b

C and

a =/:- 0,

then

is certainly a conformal mapping of C to C. In fact

one checks easily that

r->

(z

b) /a is

the inverse mapping. The interesting

part of the theorem is that these are in fact the only conformal maps of C
to C. To see this, fix a conformal map off : C-+ C. We begin with some
lemmas:

Lemma 6.1.2. The holomorphic function


lim

lzl-++oo
That is, given

lf(z)I

>

>

0,

IJ(z)I

satisfi.es

+oo.

there is a number C >

such that if

lzl

>

C, then

1/E.

Proof. This is a purely topological fact, and our proof uses no complex
analysis as such.
The set

{z: lzl

1/E}

is a compact subset of C. Since

is holomorphic, it is continuous.
Therefore S

set is compact.

1-1({z : lzl

1/E})

is compact.

Heine-Borel theorem (see [RUDl]), S must be bounded.


positive number C such that S

{z: lzl

Taking contrapositives, we see that if


of

1-1({z: lzl

lwl

>

Thus there is a

w is not an element
element of {z: lzl 1/E}.
0

result.

Using the concepts introduced in Section 4.7, Lemma

By the

C, then

1/E}). Therefore J(w) is not an


lf(w)I > 1/E. That is the desired

thought of as saying that

C-+ C

C} .

In other words,

1-1 :

Also the continuous image of a compact

is continuous at oo

with value

oo.

6.1.2

can be

181

6.1. Bilwlomorphic Mappings of the Plan e

Further understanding of the behavior of

f(z) when z has large absolute

value may be obtained by applying the technique already used in Chapter


z E <C such that z -=/= 0
f(l/z) -=/= 0, a function g(z)
1/f(l/z). By Lemma 6.1.2, there is a
number C such that lf(z)I > 1 if izl > C. Clearly g is defined on {z : 0 <
lzl < 1/C}. Furthermore, g is bounded on this set by 1. By the Riemann
removable singularities theorem, g extends to be holomorphic on the full disc
D(O, 1/C) {z: lzl < 1/C}. Lemma 6.1.2 tells us that in fact g(O) 0.

4 to talk about singularities at oo. Define, for all

and

f : <C ---+ <C is one-to-one, it follows that g


D(O, 1/C). In particular, g'(O) cannot be 0. Since

Now, because
on the disc

0-=/= Jg'(O)I

lim

lzl-+O+

g(z) - g(O)
Z

A > 0 such

we see that there is a constant

lim

lzl-+O+

is one-to-one

g(z)
I I
Z

that

Jg(z)I 2': Alzl


for

sufficiently small.

original function

We next translate this to information about the

f.

Lemma 6.1.3. There are numbers B,

lf(z)I

<

D > 0 such that, if lzl > D, then


BlzJ.

Proof. As noted above, there is a number

Jg(z)I 2': AlzJ.

If

izl > 1/8,

8 >0

such that if

lzl

<

8,

then

lf(z)I

1
Jg(l/z)i

Thus the lemma holds for any B

> 1/A

and

1/8.

The proof of the theorem is now easily given. By Theorem 3. 4.4,


a polynomial of degree at most
Clearly

then

1,

that is,

f(z)

is one-to-one and onto if and only if

az + b for
a -=/= O; thus

some

a, b

Theorem

f
E

is

<C.

6.1.1

is proved. A shorter if less self-contained proof can be given using Theorem

4.7.5: see Exercise 35.


One part of the proof just given is worth considering in the more general
context of singularities at oo, as discussed in Section 4. 7. Suppose now that
his holomorphic on a set

{z: lzl >a},


lim

lzl-++oo

for some positive a, and that

Jh(z)i

+oo.

g(z) = 1/h(l/z) is defined and holomorphic on


17 > 0. Also, by the same reasoning as above, g
to D(O, 17) with g(O)
0. If we do not assume in

Then it remains true that

{z : 0

<

lzl

<

77},

some

extends holomorphically
advance that

is one-to-one, then we may not say (as we did before) that

182

6. Holomorphic Functions

g'(O) -:/= 0.
n

But

as

Geometric Mappings

is not constant, since his not, so there is a positive integer

and a positive number

such that

lg(z)I Alzln
for all

6.1.2)

that

lzl

with

sufficiently small. It then follows (as in the proof of Lemma

lh(z)I
for

l zl

sufficiently large.

1
lzl
A n

This line of reasoning, combined with Theorem

3.4.4 recovers Theorem 4. 7. 5: If h: C ---+ C is a holomorphic function such


that
lim

lzl---++oo

lh(z)I

+oo,

then h is a polynomial.
The reasoning presented here is essentially just a specific application
of the techniques and results of Section 4. 7.
6.2. Biholomorphic Mappings of the Unit Disc to Itself
In this section the set of all conformal maps of the unit disc to itself will
be determined. The determination process is somewhat less natural than in
the last section, for the reader is presented with a "list" of mappings, and
then it is proved that these are all the conformal self-maps of the disc (i.e.,
conformal maps of the disc to itself). This artificiality is a bit unsatisfying;
later, when we treat the geometric structure known as the Bergman metric
(Chapter

14) ,

we shall be able to explain the genesis of these mappings.

Our first step is to determine those conformal maps of the disc to the
disc that fix the origin. Let D denote the unit disc.

Lemma 6.2.1. A holomorphic function

: D---+ D that satisfies

f(O)

is a conformal mapping of D onto itself if and only if there is a complex


number

w with lwI

1 such that
f(z)

wz for all z

ED.

In other words, a conformal self-map of the disc that fixes the origin must
b e a r ot ati on.
Proof. If

E C and

lwl

1,

then clearly the function

conformal self-map of the disc: The inverse mapping is

f(z) = wz
z/w.

is a

i---+

To prove the converse, suppose that f : D---+ D is a conformal self-map


of the disc that fixes the origin.
lemma (Proposition

Let

1-1

: D ---+ D. By the Schwarz

5.5.1),
lf'(O)I 1

and

lg'(O)I 1.

6.2. Biholomorphic Mappings of the Unit Disc

183

However the chain rule (see Chapter 1, Exercise 49 and Chapter 2, Exercise
13) implies that
1 (f og)'(O) J'(O) g'(O).
It follows from this equation and from ( * ) that lf'(O)I = 1 and lg'(O)I
1.
The uniqueness part of the Schwarz lemma then yields that f(z) = f'(O)z.
D
This, with w f'(O), is the desired conclusion.
=

It is not immediately clear that there are any conformal self-maps of


the disc that do not fix the origin, but in fact some of these were already
introduced in Section 5.5 in connection with the Schwarz-Pick lemma. For
convenience, we write them out again:
(Construction of Mobius transformations) .

Lemma 6.2.2

1,

we define

<Pa(z)
Then each
Proof.

<Pa

For

EC,

lal

<

z-a
1-az

_
.

is a conformal self-map of the unit disc.

See the first part of the proof of Theorem

5.5.2.

The biholomorphic self-mappings of D can now be completely charac


terized. Notice before we begin that the composition of any two conformal
self-maps of the disc is also a conformal self-map of the disc.
Theorem 6.2.3. Let

f:

lwl

1, lal

<

f is a
a, w with

D be a holomorphic function. Then

conformal self-map of D if and only if there are complex numbers


such that, for all

f(z)

ED,
=

w <Pa(z) .

Proof. That functions f of the given form are conformal self-maps of the
disc has already been checked. For the converse, let f be any conformal self
map of the disc. Set b f(O) and consider the map <Pb of: D D, which
is certainly a conformal self-map of the disc. Observe that (<Pb of)(O) 0.
Therefore Lemma 6.2.1 applies, and there exists an w EC such that lwl 1
and <Pb of(z) wz, all z ED. Thus
=

<P;1(wz)

f(z)

<P-b(wz)
wz+b
1+bwz
z+bw-1
=

where we have used the fact that lwl


f(z)

----

1+bw-1z'
1 to see that bw-1

w <Pa(z),

bW

bw. Thus

6. Holomorphic Functions

184

where a

as

Geometric Mappings

-bw-1.

Notice that our proof also shows that any conformal self-map of the disc
has the form

f(z)

cPa(wz) for some lwl

and lal

<

1.

You can easily

verify by direct calculation that the composition of any two maps of the
form given in the theorem, or of any two maps of the form given in the last
sentence, or of one of each, gives in turn a map that can be written in either
of these special forms. Actually, it is easy to see that the set of conformal
self-maps of any open set U C forms a group under composition. So the
calculations just discussed had to be successful: Once the precise form of
all the conformal self-maps is determined (as in Theorem

6.2.3),

then it was

necessarily the case that compositions and inverses would have that form
too. The group of conformal self-maps of an open set U C is called the

automorphism group of U, and the conformal self-maps themselves are called


automorphisms of U. The automorphisms of the unit disc are usually called
Mobius transformations (after A. F. Mobius, 1790-1860).
Note that the case a

for a corresponds to the case of rotations

treated in the first lemma.

6.3. Linear Fractional 'fransformations


The automorphisms (i.e., conformal self-mappings) of the unit disc D are
special cases of functions of the form
z

az+b

1----t

-- ,
CZ+d

a,b,c,d EC.

It is worthwhile to consider functions of this form in generality. One restric


tion on this generality needs to be imposed, however; if ad- be

0,

then the

numerator is a constant multiple of the denominator if the denominator is


not identically zero. So if ad- be =

0,

then the function is either constant or

has zero denominator and is nowhere defined. Thus only the case ad-be i-

is worth considering in detail.

Definition 6.3.1. A function of the form


z
is called a

az+b

1----t

-CZ+d

ad - be i-

0,

linear fractional transformation.

Note that (az + b) /(cz + d) is not necessarily defined for all z E C


Specifically, if c i-

0,

then it is undefined at z
lim

Z---+-d / c

az+b
CZ+d

-d/c. In case c i-

0,

+oo.

This observation suggests that one might well, for linguistic convenience,

adjoin formally a "point at oo" (as in Section

4.7)

to C and consider the

185

6.3. Linear Fractional Transformations

value of (az + b)/(cz + d) to be oo when z = -d/c (c =f. 0). It turns out


to be convenient to think of a transformation z
(az + b)/ (cz + d) as a
mapping from <CU { oo } to <CU { oo } (again along the line of ideas introduced
in Section 4. 7). Specifically, we are led to the following definition:
1--->

Definition 6.3.2. A function f : <CU { oo }


<CU { oo } is a linear fractional
transformation if there exist a, b, c, d E <C, ad - be =f. 0, such that either
-t

(i) c=0, J(oo)=oo, and f (z)= (a/d)z + (b/d) for all z

<C,

or
(ii) c =f. 0, f (oo)= a/c, J(-d/c)=oo, and f (z)= (az + b)/(cz + d) for
all z E <C, z =f. -d/c.
It is important to realize that, as before, the status of the point oo
is entirely formal: We are just using it as a linguistic convenience, to keep
track of the behavior of f (z) both where it is not defined as a map on <C
and to keep track of its behavior when lzl
+oo. The justification for the
particular devices used is the fact that
-t

(1) limlzl-++oo f (z)= f (oo),


(2) limz-+-d/c l f (z) I=+oo [c =f.

O;

case (ii) of the definition].

These limit properties of f can be considered as continuity properties of f


from <CU { oo } to <CU { oo } using the definition of continuity that comes from
the topology on <CU { oo } introduced in Section 4. 7. There the topology was
introduced by way of open sets. But it is also convenient to formulate that
same topological structure in terms of convergence of sequences:
Definition 6.3.3. A sequence {pi } in
(notation limi-+oo Pi=Po) if either

<CU

{ oo } converges to Po

<CU

{ oo }

(1) Po=oo and limi-++oo IPil=+oo where the limit in this expression
is taken for all i such that Pi E <C
or
(2) Po E <C, all but a finite number of the Pi are in <C and limi-+oo Pi=Po
in the usual sense of convergence in <C.

This definition of convergence coincides with the definition in terms


of metric spaces or topological spaces, with the ideas of distance and open
set introduced in Chapter 4. Stereographic projection puts <C U { oo } into
one-to-one correspondence with the two-dimensional sphere S in IR3, S =
{(x,y,z) E JR3: x2 + y2 + z2 = 1 } in such a way that convergence of the
sequence is preserved in both directions of the correspondence. For this
reason, <C U { oo } is often thought of as "being" a sphere and is then called,
for historical reasons, the Riemann sphere, as already discussed in Section
4.7.

6. Holomorphic Functions

186

as

Geometric Mappings

In these terms, linear fractional transformations become homeomor


phisms of CU

{ oo}

( Recall that a homeomorphism is,

to itself.

by definition,

a one-to-one, onto continuous mapping with a continuous inverse. )

Theorem 6.3.4. If
formation, then

1-1

: CU

{ oo}

: CU

{ oo}

CU

{ oo} is

a linear fractional trans

f is a one-to-one and onto continuous function.


CU { oo} is a linear fractional transformation.

Also,

If g
then

: C U { oo} C U { oo} is also a linear fractional transformation,


f o g is a linear fractional transformation.

Proof. The continuity of


that
c

is clear from the previous discussion.

To see

f(z) = (az+b)/(cz+d), c# 0 ( the case


exercise ) . The inverse of f can be computed

is one-to-one, suppose that

= 0, d # 0

is left as an

formally as follows: Set

az+b
w- CZ+d"
-

Solving for

z,

we find that, formally, the inverse of

Z=
To define the inverse of

is

-dw+b
cw- a

---

precisely, set

-dz+b
CZ - a
d

g ( z)
g(oo)

if

z -=I a I c ,

g(a/c)

00.

g is a linear fractional transformation, and one can easily check directly


f o g and g o f are both equal to the identity map from C U { oo} to

Then
that

itself.
Similarly, one can compute the formal composition of the linear frac
tional mappings

z 1--+ (az+b)/(cz+d)

A ( (az+b)/(cz+d) ) +B
C ( (az+b)/(cz+d) ) +D

and

1--+

(Az+B)/(Cz+D) :

(Aa+Bc)z+(Ab+ Bd)
(aC+cD)z+ (bC+Dd)

Note that

(Aa+Bc) (bC+Dd) - (Ab+Bd) (aC+cD) =(AD - BC) (ad - be)# 0.


Thus

Zt-+

(Aa+Bc)z+ (Ab+Bd)
(aC+cD)z+ (bC +Dd)

-----

is a formal linear fractional transformation, and the associated linear frac


tional transformation of CU
composition.

{ oo}

to itself is easily verified to be the actual

6.3.

Linear Fractional Transformations

187

For instance, the image of oo under the first transformation is a/c (in
case c i- 0) and the image of a/c under the second is
A(a/c) +B

Aa +Be

C(a/c) + D

aC + cD

if aC +cD i- 0. Also, if aC +cD i- 0, then the image of oo under the formal


composition is also (Aa +Bc)/(aC +cD). The checking of the various cases,
D
and so forth, is left as an exercise.
The reader who does actually check all the cases and possibilities in the
composition proof just described will become convinced that the simplicity
of language obtained by adjoining oo to C (so that the composition and
inverse properties of the theorem hold) is well worth the trouble. Certainly
one does not wish to consider the multiplicity of special possibilities (c
0, c i- 0, aC + cD i- 0, aC + cD 0, etc.) that arise every time composition
is considered.
=

In fact, it is worth summarizing what we have learned in a theorem.


F irst note that it makes sense now to talk about a homeomorphism from
CU { oo} to CU { oo} being conformal: This just means that it (and hence
its inverse) are holomorphic in our extended sense. If <P is a conformal
map of C U { oo} to itself, then, after composing with a linear fractional
transformation, we may suppose that maps oo to itself. Thus , after
composition with a linear fraction transformation, is linear (cf. Theorem
6. 1 . 1 ). It follows that itself is linear fractional. The following result
summarizes the situtation:

function is a conformal self-mapping of CU { oo} to


itself if and only if <P is linear fractional.

Theorem 6.3.5. A

We turn now to the actual utility of linear fractional transformations


(beyond their having been the form of automorphisms of D in Theorem
6.2.3). The most frequently occurring use is the following result, already
noted as Exercise 9 in Chapter 1:
Theorem 6.3.6 (The (inverse) Cayley transform). The

linear fractional
transformation z
(z i)/(z + i) maps the upper half plane {z: Imz > O}
conformally onto the unit disc D {z: lzl <l}.
f-t

Proof. The proof is a formal calculation:

z i
<l
z +i
-_

if and only if

lz - il2 < lz + il2

188

6.

if and only if

Holomorphic Functions

2 Re ( zi)

<

Geometric Mappings

-2 Re ( zi)

if and only if
4Re( zi)

<

>

0.

if and only if
Imz

as

0
0

Calculations of the type just given are straightforward but tedious. It is


thus worthwhile to seek a simpler way to understand what the image under
a linear fractional transformation of a given region is. For regions bounded
by line segments and arcs of circles, the following result gives a method for
addressing this issue:
Theorem 6.3. 7. Let C be the set of subsets of CU
circles and

(ii)

sets of the form LU

{ oo}

{ oo}

consisting of

(i)

where L is a line in C Then every

linear fractional transformation <P takes elements of C to elements of C.

Sets consisting of a line L union { oo} are thought of as "generalized


circles." Thus the theorem says that linear fractional transformations take
circles to circles, in the generalized sense of the word.
Proof of Theorem 6.3. 7.

Every linear fractional transformation is a com


position of transformations of the ( formal) forms z
az, z
z+b, z
1/ z
( exercise). Thus it is enough to verify the statement of the theorem for each
of these transformations.

The conclusion is obvious for mappings of the form z


az ( the dila
tion of a generalized circle is still a generalized circle) and z
z + b ( the
translation of a generalized circle is a generalized circle). Thus it remains
to check the result for the inversion z
1/ z. Assume that z i- 0, oo.

Consider an arbitrary circle x2+y2+ax+ (3y+"Y = 0. We set w = 1/ z


with w =u +iv. Then
u

x=Rez=Re( l/w)= 2
u +v2,
-v
y = Imz = Im ( 1/w) = 2
.
u +v2

Substituting into the equation of the circle and simplifying, we find that
v
1
u
2 - (3 2
2 + "Y = o.
2
2+Q 2
u +v
2
2
Since lwl =u +v2 i- 0,

u +v

u +v

1 +nu - (3v + 1(u2+v2) = 0.


If "Y i- 0, then this is the equation of a ( standard) circle. Thus the inversion
of the original circle is indeed a circle. If "Y 0, then this is the equation of
=

6.4. The Riemann Mapping Theorem

189

a line. In this case the image of the point at infinity under the inversion is
the origin and the image of the origin (which lies on the original circle) is

the point at infinity. So the image is a generalized circle.

To illustrate the utility of this theorem, we return to the Cayley transformation

i)

i.

(1-i)/(1 + i)

z-i
z+i

--.

1, the

point 1 is sent

-i, and the point -1 is sent to

(-l-i)/(-1 +

Under this mapping the point


to the point

1--4

oo

is sent to the point

Thus the image under the Cay ley transform (a linear fractional

transformation) of three points on IR U { oo} contains three points on the unit


circle. Since three points determine a (generalized) circle, and since linear
fractional transformations send generalized circles to generalized circles, we
conclude that the Cayley transform sends the real line to the unit circle. Now
the Cayley transform is one-to-one and onto from CCU

{ oo}

to CCU

{ oo }.

By

continuity, it either sends the upper half plane to the (open) unit disc or to
the complement of the closed unit disc. The image of i is 0, so in fact the
Cay ley transform sends the upper half plane to the unit disc.
The argument just given, while wordy, is conceptually much simpler
and quicker than the sort of calculation that we have given previously. Some
other applications of this technique are given in the exercises.

6.4. The Riemann Mapping Theorem:


Statement and Idea of Proof
The conformal equivalence of open sets in CC is generally not very easy to
compute or to verify, even in the special case of sets bounded by (generalized)
circular arcs. For instance, it is the case (as will be verified soon) that the
square

{z
x + iy: x
l l < l,IYI < l} is conformally equivalent to
{z: z
l l < l}. But it is quite difficult to write down explicitly
=

unit disc

the
the

conformal equivalence. Even when a biholomorphic mapping is known (from


other considerations) to exist, it is generally quite difficult to find it. Thus
there is a priori interest in demonstrating the existence of the conformal
equivalence of certain open sets by abstract, nonconstructive methods. In
this section we will be concerned with the question of when an open set

U CC is conformally equivalent to the unit disc.


Several restrictions must obviously be imposed on U. For instance, U
obviously cannot be CC because every holomorphic function f : U
would then be constant (Liouville's theorem).

__,

Also, U must be topologi

cally equivalent (i.e., homeomorphic) to the unit disc D since we are in fact

demanding much more in asking that U be conformally equivalent. For in


stance, D could not be conformally equivalent to

{z : 1

<

IzI

<

2}

since it

6. Holomorphic Functions as Geome t ric Mappings

190

is not topologically (i.e., homeomorphically) equivalent.

[Although at the

moment we have no precise proof that the two are not homeomorphic, this
assertion is at least intuitively plausible.]
Surprisingly, the two restrictions just indicated (that U not be C and
that U be topologically equivalent to

D)

are not only necessary but they

are sufficient to guarantee that U be conformally equivalent to the disc

D.

To give this notion a more precise formulation, we first define formally


what we want to mean by topological (homeomorphic) equivalence (reiter
ating the definition in Section 6.3, prior to Theorem 6.3.4):

Definition 6.4.1. Two open sets U and V in C are homeomorphic if there


is a one-to-one, onto, continuous function
continuous. Such a function

l: U

--;

V with l-1 : V

--;

U also

is called a homeomorphism from U to V.

[It is a fact that the inverse of a continuous, one-to-one, onto function


from one open subset of IR2 to another is always continuous.
condition in the definition of homeomorphism that

l-1

Thus the

be continuous is

actually redundant. But this "automatic continuity" for the inverse function
is rather difficult to prove and plays no role in our later work. Thus we shall
skip the proof; the interested reader may consult an algebraic topology text.]

Theorem 6.4.2 (The Riemann mapping theorem). IfU is an open subset of


C, U -f. C and if U is homeomorphic to D, then U is conformally equivalen t
to D.
The proof of the Riemann mapping theorem consists of two parts. First,
one shows that if a domain U is homeomorphic to the disc, then each holo
morphic function on U has an antiderivative (recall that this property is
called "holomorphic simple connectivity"-see Section 4.5). The argument
to prove this assertion is essentially topological: and indeed the conclusion
will be deduced (in Chapter

11)

from a topological condition called simple

connectivity that appears to be weaker than homeomorphism to the disc,


though it is in fact equivalent to it for open sets in C.

The second step

uses the first to construct a one-to-one holomorphic mapping from U into


the unit disc. We then construct a mapping from U to

that is both one

to-one and onto by solving an "extremal problem"; that is, we shall find
a function that maximizes some specific quantity (details will be provided
momentarily).
The second part of the proof of the Riemann mapping theorem will
occupy us for the rest of this chapter (Theorem 6.6.3-the analytic form
of the Riemann mapping theorem). The first, essentially topological, part
of the proof will be given separately in Chapter

11.

Before presenting the

(second part of the) proof, let us discuss the significance of the R.iemann
mapping theorem and of some of the issues that it raises.

191

6.4. Tlie Riemann Mapping Theorem

First, it is important to realize that homeomorphism is, in general,


quite far from implying biholomorphic equivalence:

EXAl\IPLE 6.4.3. The entire plane CC is topologically equivalent to the unit


disc D. An equivalence is given by
f(z)=

1 +11.
z

However, as previously noted, CC is not conformally equivalent to D.


In the next chapter we shall learn that two annuli { z : r1 < I z I < r2 } and
{z :

s1

<

I z I < s2} are conformally equivalent if and only if

r2 / r1

s2 / s i.

Yet two such annuli are always homeomorphic (exercise). Thus we see that
the Riemann mapping theorem is startling by comparison.
The idea of the proof of the Riemann mapping theorem comes from
the Schwarz lemma. Namely, one can think of this lemma in the following
form: Consider a holomorphic (not necessarily one-to-one and onto) function
f: D------> D with f(O)= 0. Then fis conformal (i.e., one-to-one and onto) if
and only if IJ'(O)I is as large as possible, that is, if and only if
lf'(O)I = sup{lh'(O)I : h: D------> D, h(O)= 0, h holomorphic}.
The special attention paid to the point 0 here is inessential.

In fact if

f: D------> D is a holomorphic function such that


lf'(P)I= sup{lh'(P)I : h: D------> D, h(P) = P, h holomorphic},
then f must be conformal.

(Exercise:

Prove this, and find an analogous

statement to cover the case when f maps P to Q and Q -=/:- P.)


If U satisfies the hypotheses of the Riemann mapping theorem, then
one might look for a conformal mapping of U to D by choosing a point
P E U and then looking for a holomorphic function f : U ------> D such that
f(P) = 0 and lf'(P)I is maximal. It turns out to be technically simpler to
consider only functions fthat are assumed in advance to be one-to-one, and
we shall take advantage of the resulting simplification.
Several questions arise about our plan of attack. First, do there exist

any one-to-one holomorphic functions J: U ------> D? Assuming that there are


such functions, is the set of values {IJ'(P)I} bounded?

Also, is the least

upper bound attained? That is, is there a holomorphic function f from U


to D such that f(P) = 0 and IJ'(P)I is as large as possible?
The latter question is of particular historical interest because Riemann
in fact neglected in his proof to verify that a similar sort of maximum was
assumed by some particular function. Thus his theorem was in doubt for
some time until what has become known as "Dirichlet's principle" gave a

6. Holomorphic Functions as Geometric Mappings

192

method to fill the gap. The modern approach to the Riemann mapping the
orem involves a convergence idea for holomorphic functions called "normal
families," which we develop in the next section.
Note that the issue of the boundedness of the derivatives of
is easily dispatched.
that

D(P, r) <:;;: U.

For let

: U

Then, of course,

lf'(P)I 1/r.

estimates imply that

D, f(P)
IJ(z)I 1 for
____,

0. Choose

>

at

0 such

all z; hence the Cauchy

Thus we have an a priori upper bound

on the size of derivatives off at P. In the next section we turn our attention
to establishing the existence of an
that

actually equals

the

least

whose derivative

J'(P)

has modulus

upper bound of the absolute values of such

derivatives.

6.5. Normal Families


The results of the previous section have already suggested that there is spe
cial interest in holomorphic functions that satisfy some extremal property.
The special nature of such extremal problems is this: For any E > 0 there
will be a holomorphic function

( e.g.,

ff

that comes within E of being the extremum

the least upper bound in our special case from the previous sect.ion).

fj such that the fj come ever closer to being the


but one would like to assert that (some subseque nce of) the fj
actually converge to an "extremal function" f0. ( By analogy, consider the
problem of finding the "shortest" curve between two points P, Q on the
unit sphere in space. There will be curves /j joining P, Q whose length is
within 1/j of the greatest lower bound on the lengths. But one would like
Thus there is a sequence

extremum;

to extract a subsequence of the rj that actually converges in some sense to


a length-miniminizing curve rO It is not a priori obvious that this can be
done. Furthermore, if one considers the sphere with one point removed, then
there are actually pairs of points for which this "extremal problem'' has no
solution-the "shortest curve" would necessarily pass through the missing
point!)
We now develop a versatile set of tools that allows us to extract

con

vergent subsequence from a fairly general family of holomorphic functions.


We begin with a definition that will be useful in the formulation of our main
result.

Definition 6.5.1. A sequence of functions


to

converge normally

to a limit function

fj on an open set U <:;;: C is said


Jo on U if {fj} converges to f0

uniformly on compact subsets of U. That is, the convergence is normal if for


each compact set K <:;;: U and each E > 0 there is a J > 0 ( depending on K
and t ) such that, when j > J and z EK, then

lfj(z) - fo(z)I

<

E.

193

6.5. Normal Families

F igure 6.1

EXAMPLE 6.5.2. The functions


disc D to the function
on all of D to

Jo,

fo(z)

fj(z)

zj

converge normally on the unit

0. The sequence does

not converge

uniformly

but it does converge uniformly on each compact subset of

D.
Theorem 6.5.3

( Mantel's

theorem, first version ) . Let F

family of holomorphic fonctions on an open set U C


is a con stan t AI > 0 such that. for

all z

U an d

lfo(z)I :SM
Then, for every

sequence {fj}

all f n

{f0}oEA be

S uppose
E

that there

F,

F, there is a s u bs eq u en ce

verges normally on U to a limit (holomorphic) function

Jo.

{!Jk}

that con

Mantel's theorem is a remarkable compactness statement. For contrast


in the real-variable situation, note that the functions
bounded by the constant 1 on the interval

[O, 27T],

{ sin kx}k=I

are all

yet there is no convergent

subsequence-not even a pointwise convergent subsequence, much less a


normally convergent one.

( holomorphic ) functions with common domain


U. We say that F is a normal family if every sequence in F has a subse
quence that converges uniformly on compact subsets of U, that is, converges
normally on U.
Let F be a family of

The proof of Mantel's theorem uses powerful machinery from real anal
ysis. The Ascoli-Arzela theorem, one of the main tools, is treated in Appen
dix A. An alternative proof of Mantel's theorem, not m;ing the Ascoli-Arzela
theorem, is given in the exercises.

194

6. Holomorphic Functions as Geometric l\Jappings

U.

Proof of Montel's theorem. Fix a compact set K


larger compact set L

T/

>

0, any two points

z, w

Choose a slightly

with its interior L containing K. Then, for some

EK with lz - wl

17 have the property that the

<

6.1. Since L is compact,


D( e, r) U. But then, for

line segment connecting them lies in L. See Figure


there is a number
any

f E F,

>

0 such that if f E L, then

we have from the Cauchy estimates that

If ()I r

c I.

Observe that this inequality is uniform in both e and


Fix an

J E F.

If

lz - wl

<

line segment connecting

[O. 1]

r1 and if"(:
to

w,

---->

<C is

f.

Now let

Z,

E K.

a parametrization of the

then we have

lf(z) - f(w)I

lfh(l)) - f('(Y O))I

l.ll J'('Y(t)). "f1(t)dtl


1
C1 [ b'(t)idt

<

.fo

C1lz -wl.
Thus, for any

z, w EK

and any

f EF

we have that, if

IJ(z) - f(w)I Clz In particular, F is an equicontinuous family

lz - w l

<

TJ,

w l.

( this

concept is defined in Ap

pendix A ) . The Ascoli-Arzela theorem thus applies (again see Appendix A)


to show that any sequence

{fj}

F has a subsequence

uniformly on K.

{!jJ

that converges

Now we need an extra argument to see that a subsequence can be found


that converges uniformly on every compact subset of

U ( the

has to work for every compact set ) . Thus let K 1 K2


0

sets such that Kj KJ+1 for each

I<j

[ In

case U

{z

EU

and

distance( z, <C

<C, letting Kj

D(O. j)

we have just given, there is a sequence

LJj Kj

{fj}

same sequence

be compact

U ( e.g., we could let

\ U) 1 /j }

would do.]

D( 0, j)) .

Then, by the argument

F such that

{/;}

converges

uniformly on K1. Applying the argument again, we may find a subsequence


of this sequence-call it

{!jk } -that converges uniformly on K2.

Continuing

in this fashion, we may find for each index ma subsequence of the

( m - l)th

subsequence that converges uniformly on Km. Now we form a final sequence


by setting

6.5.

195

Normal Families

93

fjk3'

etc.
Then the sequence

{gJ}

is a subset of

sequences that we formed above. So


Now any compact subset L of

F and is a subsequence of each of the


{gj} converges uniformly on each K1.
K1

must be contained in some

since

LJ Kj

is an open cover of L, which necessarily has a finite subcover. Therefore the


sequence 9J converges uniformly on any compact subset of

U.

The proof just presented did not use the full force of the hypothesis
that

IJ(z)I ::;

M for all

f EF

and all

z EU.

of this sort for all z in each comt>act subset

In fact, all we need is a bound

K,

and the bound may depend

on I<. Thus we are led to the following definition:


Definition 6.5.4. Let
We say that

U C
K U
E K:

be a family of functions on an open set

is bounded on compact sets if for each compact set

there is a constant M

MK

such that, for all

IJ(z)I ::;

J EF

and all

M.

The remarks in the preceding paragraph yield the proof of the following
improved version of Theorem

6.5.3:

Theorem 6.5.5 ( Mantel's theorem, second version ) . Let U C be an open

and let F be a family of holomorphic functions on U that is bounded


on compact sets. Then for every sequence {fj} F there is a su bsequence
U1k} that converges normally on U to a limit (necessarily holomorphic)
function Jo.
set

EXAMPLE 6.5.6. We consider two instances of the application of Mantel's


theorem.

(A)

Consider the family


take

1)

{z1}1

of holomorphic functions. If we

to be any subset of the unit disc, then

is bounded

( by

so Mantel's theorem guarantees that there is a subsequence that

converges uniformly on compact subsets. Of course, in this case it


is plain by inspection that any subsequence will converge uniformly
on compact sets to the identically zero function.
The family

F fails to be bounded on compact sets for any U that

contains points of modulus greater than l. Thus neither version of


Mantel's theorem would apply on such a
convergent sequence in

(B)

F
lz/JI ::;

Let

{z/j}1

M for all

on C

for such a

U.

In fact there is no

U.

Then there is no bound M such that

and all

z EC

But for each fixed compact

subset I< C there is a constant MK such that

lz/JI

<

MK for all

6.

196

Holomorphic Functions as G eom etric Mappings

j and all z EK. (For instance, MK= sup{lzl : z EK} would do.)

Therefore the second version of Mantel's theorem applies. Indeed

the sequence {z/j}1 converges normally to 0 on C

Let us conclude this section by settling the derivative-maximizing holo

morphic function issue that was raised earlier in Section 6.4.

Proposition 6.5.7. Let Uc::;; tC be any open set. Fix a point P E U. Let F
be a family of holomorphic functions from U into the unit disc

P to 0. Then there is a 11 0lomorphic function Jo : U


limit of a sequence {JJ}, JJ E F, such that

---->

that take

that is the 11or111al

IJMP)I l!'P)I

for all J E F.

Proof. We have already noted in Section

6.4

that the Cauchy estimates

give a finite upper bound for IJ'(P)I for all J E F. Let


.\ = sup{IJ'(P)I : J E F}.

By the definition of supremum, there is a sequence {JJ} c::;; F such that


IJ (P)I----> .\.But the sequence {fj} is bounded by

since all elements of F

take values in the unit disc. Therefore Mantel's theorem applies and there

is a subsequence {JJk} that converges uniformly on compact sets to a limit


function Jo.

By the Cauchy estimates, the sequence {f k (P)} converges to JQ(P).

Therefore IJQ(P)I =.\ as desired.

It remains to observe that, a priori, Jo is known only to map U into

the closed unit disc

5.4.3)

D(O, 1). But the

maximum modulus theorem (Corollary

implies that if Jo(U) n {z : lzl =

modulus

1).

of modulus

1} i- 0,

then Jo is a constant (of

Since Jo(P) = 0, the function Jo certainly cannot be a constant

1.

Thus Jo(U) C

D(O, 1).

6.6. Holomorphically Simply Connected Domains


The derivative-maximizing holomorphic function from an open set U to

the unit disc (as provided by Proposition

6.5.7)

turns out to have some

interesting special properties when U resembles the disc in a certain sense.

The type of resemblance that we wish to consider is called 'holomorphic

simple connectivity', a concept that we introduced in Section

4.5.

As noted

there, this terminology is a temporary one that we use for convenience; it


is not used universally (in particular, it is not found in most other texts on

complex function theory).

The concept of holomorphically simply connected' (h.s.c.) arises nat

urally in our present context, as it did in Chapter

4.

However, it turns out

6.6. Holomorphically Simply Connected Domains

197

to be implied by a more easily verified topological condition that is called

simple connectivity. Simple connectivity is in fact a universally used mathe


matical concept; we shall discuss it in detail in Chapter 11 and demonstrate
there that simple connectivity implies holomorphic simple connectivity. Re
call the definition of holomorphic simple connectivity:

U C is holomorphically simply
f : U ---> C, there is a holomor
function satisfying F' (z)
f(z).

Definition 6.6.1. A connected open set

connected if, for each holomorphic function


phic antiderivative F-that is, a

EXAMPLE 6.6.2. As established in Chapter 4, open discs and open rectan


gles are holomorphically simply connected.

U2 . . . are holomorphically simply connected sets, then


U LJ [i is also holomorphically simply connected ( Chapter 1,
Exercise 56). In particular, the plane C is holomorphically simply connected.
If U1

their union

Theorem 6.6.3 ( Riemann mapping theorem:

analytic form ) . If

holomorphically simply co11nected open set in C and

U -=f. C,

then

is a

is

conformally eq ui val ent to tl1e unit disc .


The proof of this result involves some intricate constructions.
present it in the next section.

We

The rest of the present section is devoted

to the consideration of some preliminary results that are needed for the
proof. These results also have considerable intrinsic interest.

Lemma 6.6.4 ( The holomorphic logarithm lemma ) . Let U be


If

phically simply connected open set.

nowhere zero on

U,

f : U ---> C

holomor

is holomorphic and

then there exists a holomorphic function h on

that

Proof. The function

eh= J

on

J'(z)/ f(z)

such

U.

is holomorphic on

U,

because

is

U. Since U is holomorphically simply connected, there is a


U---> C such that h'(z) J'(z)/ f(z) on U. F ix a point zo EU.

nowhere zero on
function h:

Adding a constant to h if necessary, we may suppose that

eh(zo)

f(zo).
e" =

satisfies g'

After this normalization, we can now demonstrate that


It is enough to show that

g(z)

f(z)e

h(

:.)

then g is necessarily a constant. Since we have arranged that

g(z)

=
(z) :z (!(z)e-h(z))
f
8 (z) (z) f(z) (
az

conclude that

on

U.

0 on

g(zo)

U.

For

1, we

1. This is equivalent to what we want to prove. Now

ah
( z )e
az

h(:.) )

6. Holom orphic

198

e-h(z) f(z)

Functions as Geometric Mappings

( afj:az (z))
_

0
D

by the way that we constructed h.

is lwlomorphically simply connected and J : U --->


<C \ {O} is lwlomorphic, then there is a function g: U---> <C \ {O} such that
Corollary 6.6.5. If U

f(z)
for all

[g(z)]2

U.

Proof. Choose has in the lemma. Set

g(z)

eh(z)/2.

It is important to realize that the pm;sibility of taking logarithms, or


of taking roots, depends critically on the holomorphic simple connectivity
of U. There are connected open sets (such as the annulus

{z: 1

for which these processes are not possible even for the function

lzl < 2})


(
f z) = z.
<

6.7. The Proof of the Analytic Form of


the Riemann Mapping Theorem
Let U be a holomorphically simply connected open set in <C that is not equal
to all of <C. Fix a point
F

{! : J
f

U and set
is holomorphic on U.

is one-to-one,

J(P)

J: U

___,

D,

= O}.

We shall prove the following three assertions:

(1)

Fis nonempty.

(2)

There is a function

Jo

E Fsuch that

IJMP)I
(3)

If

=sup
hEF

lh'(P)I.

is any element of Fsuch that

maps U onto the unit disc D.


The proof of assertion

(1)

lg'(P)I

suphEF

is by direct construction.

lh'(P)I,

then

Statement

(2)

is almost the same as Proposition 6.5.7 (however there is now the extra
element that the derivative-maximizing map must be shown to be one-to
one). Statement
conclusion of

(3)

(3)

is the least obvious and will require some work: If the

is assumed to be false, then we are able to construct an

element g E Fsuch that

lfJ'(P)I

>

lg'(P)I.

Now we turn to the proofs.

6. 7. Proof of the Riemann Mapping Theorem

199

Proof of

(1). If U is bounded, then this assertion is easy: If we simply let


a= 1/(2sup{lzl: z EU } ) and b = -aP, then the function f (z) = az +b is

in F.
If

U -/= C, there
nonvanishing on U, and U

is unbounded, we must work a bit harder.

is a point Q tf.

U.

The function qJ( z ) =

z-

Q is

Since

is holomorphically simply connected. Therefore there exists a holomorphic

h2 = .

h must be one-to-one since <P


is. Also there cannot be two distinct points z 1, z2 E U such that h(zi) =
-h(z2) [otherwise (zi) = (z2)]. Now h is a nonconstant holomorphic
function: hence an open mapping. Thus the image of h contains a disc
D (b , r ) . But then the image of h must be disjoint from the disc D(- b , r) .

function

such that

Notice that

We may therefore define the holomorphic function


r

f(z) =
2(h(z)
Since

lh(z)

- (-b ) I

h is one-to-one, so is

f.

for

z E U,

+b)

it follows that

Composing

maps

to D. Since

with a suitable automorphism of

D ( Mobi u s transformation), we obtain a function that is not only one-to

one and holomorphic with image in the disc. but also maps

f E F.

to 0. Thus

Proof of

(2).

Proposition 6.5. 7 will yield the desired conclusion once it

has been established that the limit derivative-maximizing function is itself


one-to-one. Suppose for notation that the

fj E F

converge normally to

Jo,

with

IJMP)I =sup lf'(P)I.


fE:F

We want to show that

Jo

is one-to-one into D.

specifically Hurwitz's theorem (Theorem

5.3.3),

The argument principle,


will now yield this conclu

s10n:

EU. Consider the holomorphic functions 9J (z) = fj (z) U\ { b } . Each fJ is one-to-one: hence each 9J is nowhere
vanishing on U \ {b } . Hurwitz's theorem guarantees that either the limit
function fo(z) - fo (b) is identically zero or is nowhere vanishing. But, for a
function h E F, it must hold that h'(P) -/= 0 because if h' (P) were equal to

fj (b)

Fix

point b

on the open set

zero, then that h would not he one-to-one. Since F is nonempty, it follows

that suphEF lh'(P)I > 0. Thu s the function Jo, which satisfies IJ(P)I
s up h EF lh'(P )I, cannot have !MP) = 0 and Jo cannot be constant. The only
=

fo(z) - fo(b) is nowhere zero on U \ { b } . Since


each b EU, we conclude that Jo is one-to-one.
D

possible conclusion is that


this statement holds for

200

6. Holomorphic Functions as Geometric Mappings

Proof of

(3).

Let

that the image of

F and suppose that there is a point

does not contain

(z)
Here we have composed
is one-to-one

( see

Section

R.

Set

D such

g(z) - R .
1 - g(z)R

with a transformation that preserves the disc and

6. 2).

Note that is nonvanishing.

The holomorphic simple connectivity of U guarantees the existence of

a holomorphic function 'ljJ

-4 C

such that

'l/J2

Now 'l/.J is still one-to

one and has range contained in the unit disc. However, it cannot be in F
since it is nonvanishing. We repair this by composing with another Mobius

transformation: Define

tjJ(z) - 'l/J(P)

p(z)
=

Then

p(P)

0, p

1 - 'l/J(z)'l/J(P)

p
of p

maps U into the disc, and

F. Now we will calculate the derivative

actually larger in modulus than the derivative of

is one-to-one.
at

P and
P.

Therefore

show that it is

at

We have

(1 - lw(P)i2) 'l/J'(P) - ('l/J(P) - 'l/J(P))(- 'l/J'(P)


(1 - 1(P)i2)2
1
I
(P).
1 - l'l/!(P)l2 . t/J

p'(P)

Also

2'1/J(P) . w'(P)
But

g(P)

'(P)

R
(1 - g(P)R)g'(P) - (g(P) - R) (- g'(P) ) .
(1 - g(P)R)2

O; hence

2'1/:(P). 'l/J'(P)

(1 - IRl2)g'(P).

We conclude that

pI(P)

1 - IRl2 Ip
1
.
g( )
1 - lw(P)l2 2'1/!(P)
1 - IRl2 Ip
1
.
1 - 1(P)I 2'1/J(P) g ( )
1
1 - IRl2 I
.
g (P)
IR
1 - I 2'1/J(P)
1 + IRI
g'(P).
2'!/J(P)

201

6. 7. Proof of the Riemann Mapping Theorem

1 + IRI > 1 ( since R -1- 0)


(1 + IRl)/(2J[RI) > 1, that

However

l/(P)I
g

Thus, if the mapping

11/;(P)I

and

>

It follows, since

lg'(P)I.

of statement

at the beginning of the sec

(3)

tion were not onto, then it could not have property


absolute value of the derivative at

J[RI.

P.

(2),

of maximizing the
D

We have completed the proofs of each of the three assertions and hence
of the analytic form of the Riemann mapping theorem.
The proof of statement

(3)

may have seemed unmotivated. Let us have

another look at it. Let

(z)

T(z)

z-R
1 - Rz'

z-1/J(P)
=

1 -'!j,i(P)z

and

S(z) = z2.
Then, by our construction, with

/l-1 o SoT-1:

-1oSoT-1op
h p.

Now the chain rule tells us that

lg'(P)I
Since

h(O)

0,

and since

lh'(O)l l/(P)I.

is not a conformal equivalence of the disc to

itself, the Schwarz lemma will then tell us that


But this says that

lg'(P)I

<

l/(P)I,

lh'(O)I

must be less than

giving us the required contradiction.

1.

202

Exercises

Exercises

1. Does there exist a holomorphic mapping of the disc onto


holomorphic mapping z 1---t

2. Prove that if

i)

C? [ Hint:

takes the upper half plane onto

is entire and one-to-one, then

Use the fact that

must be linear.

The

C.]

[ Hint:

is one-to-one to analyze the possibilities for the

oo.]

singularity at

3. If n

(z

is a domain, let Aut ( f!) be the collection of all conformal

mappings of n to n. We call Aut ( f! ) the automorphism group of n.


Prove that Aut( f!) is a group if the group operation is composition of
functions. Topologize this group as follows: A sequence

{fJ}

Aut ( f!)

converges if it does so uniformly on compact sets. Give an example of a


bounded domain for which Aut( f!) is not compact.

4. Refer to Exercise 3 for terminology. Let f!1 and f!2 be domains in


Suppose that cl> : f!1

f!2 is a conformal map.

---->

C.

Using cl>, exhibit a

relation between Aut ( f!i ) and Aut ( f!2 ).

5. Let n

C \ {O} .

Give an explicit description of all the biholomorphic

C\ {P1, ..., Pk}. Give an explicit description

self-maps of n. Now let n be

of all the biholomorphic self-maps of n.

6. Let n

C \ { z : lzl

[ Hint:

:S

1}. Determine

all biholomorphic self-maps of n.

The domain n is conformally equivalent to

7. Refer to Exercise 3 for terminology. Let n


a, b

C, a i= O}.

{z

: 0 <

jzj

C. Then Aut( n)

<
=

1}.]
{az + b :

Use the binary operation of functional composition.

This group is sometimes called the "a plus b" group. A group is called

nilpotent if there is an ]\,/


the identity.

[g, h]
[g, h]
"a

>

0 such that any commutator of order 111 is

[ Here a commutator in a group

ghg-1h-1 for g, h

G is an element of the form

G; a commutator of order k is an element

where h is already a commutator of order k -

l.]

Show that the

plus b" group is not nilpotent.

8. Let U

{z

Im z >

O}. Calculate

all the biholomorphic self-maps

of U.

9. Refer to Exercise 3 for terminology. Let G be the automorphism group


of

D(O, 1).

Let K G be given by

K
Here

Po

{po

: 0 :S () < 271" }.

is rotation through an angle of e. Prove that K is a compact

subgroup of G.

203

Exercises

10. Let n C be

a bounded domain and let

{ </Jj}

be a sequence

of conformal mappings of n to n. Prove that some subsequence of S


converges normally to a holomorphic function from n into C. [Contrast
this result with Exercise

3.]

11. Let n C be a bounded, simply connected domain. Let 1 : n

___,

2 :n ---> D be conformal maps. How are 1 and 2 related to each


1
[Hint: Look at 2 o ; it is a conformal self-map of the disc.
Now use ideas from Section 6.2.]

and

other?

12. Let n be a simply connected domain in C and let

P and Q be distinct
1 and 2 be conformal self-maps of n. If 1(P)= 2(P)
2(Q), then prove that 1 = 2.

points of n. Let

1(Q)

and

{z : 1/2 < lzl < 2}


{z : 1/2 < lzl < 2}. [Hint: If such an J
existed, then J' would equal 1/ z, so that J J' =fa 0, where ,, is the unit
circle with counterclockwise orientation.] (Cf. Exercise 52, Chapter 1.)

13. Prove that there is no holomorphic function f on


such that ef(z)

for all z E

14. A holomorphic function f


ef(z)

(a)

z for

--->

C is called a "branch of log z" on U if

all z E U. Prove that

there is a branch of log z defined on any open disc not containing


the origin;

(b)

(an open half line

O));

emanating from

(c )

\ ( { 0} U

there is a branch of log z defined on C

there is no branch of log z defined on any open set U containing

{z:lzl=l};
(d)

if there is a continuous function g


for

zE

:U

--->

C such that eg(z)

U, then g is necessarily holomorphic and hence a branch of

log z on U in the sense already defined.

15. A holomorphic function F

(3 E IR,
D(P, r )

--->

C is called a "branch of

D(P, r ) U,
e8f on D(P, r ) .

if, for each disc


such that F

there is a branch

[Hint:

z8"

on U,

of log z on

Prove that if there is a branch

of z13 defined on an open set U that contains


be an integer.

{ z: I z I

1},

then /] must

If {3 is not an integer. then the "local" logs of

associated to the branch on each disc must fit together to give a branch
of log z defined in a neighborhood of

{z:lzl

1}.]

16. Let n be a bounded domain in C such that C \ n consists of only finitely


many connected components, no one of which is a single point.
Complete the following outline to prove that there is a conformal
map of n onto a domain which consists of a disc with finitely many
closed sets removed; in fact each of these closed sets will have the spe
cial form of an open set bounded by a single

C1

(even real analytic)

curve, together with that boundary curve. [Along the way, impose any
topological hypotheses that you require-this is a "heuristic exercise,"

204

Exercises

as far as topology goes. Also, you may assume, as will be proved in


Chapter 11, that a bounded set U such that C \ U is connected is holo
morphically simply connected.]
(i) Fill in the holes in n to create a holomorphically simply connected
domain 0. Let 1 be a conformal mapping of 0 onto the disc D.
(ii) Let P be a point exterior to n and in one of the holes. Let ry be
the "inversion" z l/(z - P) which sends P to oo.
(iii) Apply step (i) to the domain ry(!1) to obtain a conformal map 2.
(iv) Repeat steps (ii) and (iii) for each of the holes in n.
(v) Manufacture the desired conformal map from the maps J.
1 7. Let n be a bounded domain and let be a conformal mapping of n

to itself. Let P E n and suppose both that (P) = P and '(P) = 1.


Prove that must be the identity. [Hint: Write the power series
(z) = P + (z - P) + higher order terms

and consider o , o o , .... Apply Cauchy estimates to the first


nonzero coefficient of the power series for after the term 1 (z - P).
Obtain a contradiction to the existence of this term.]

18. Let n be a bounded, holomorphically simply connected domain in C Fix

a point PE n. Let 1 and 2 be conformal mappings of n to D. Prove


that if 1 (P) = 2(P) and sgn( (P)) = sgn((P)), then 1 = 2. Here
sgn(o) = o/iol (definition) whenever o is a nonzero complex number.

19. Let n be a holomorphically simply connected planar domain and let be

1
a conformal mapping of n to D. Set P = - (0). Let f: n -----t D be any
holomorphic function such that f(P) = 0. Prove that lf'(P)I :S l<P'(P)I.

20. Let {fa} be a normal family of holomorphic functions on a domain U.

Prove that {f} is a normal family.

21. Let a be a complex number of modulus less than one and let

L(z) =

z.
1- az

Define L1 =L and, for j 1, LJ+1 =Lo Lj.


Prove that limLj exists, uniformly on compact subsets of D(O, 1),
and determine what holomorphic function it is.
22. Let n be a domain in C and let F = {fa} be a family of holomorphic

maps on n, all taking values in the upper half plane {z : Im z > O}. Prove
that F is a normal family, provided that one broadens the definition of
'normal family' to allow for subsequences that converge uniformly on
compact subsets to the constant value oo. [Hint: Consider the family
obtained by composing the fa with a fixed conformal mapping of the
upper half plane to the unit disc.]

Exercises

205

23. Suppose that Un} is a uniformly bounded family of holomorphic func

tions on a domain 0. Let {zk}1 0 and suppose that Zk


zo E 0.
Assume that limn-.oo fn(zk) exists for k
1, 2, .... Prove that the full
sequence {Jn} converges uniformly on compact subsets of 0.
____,

tC be a bounded domain and let U1} be a sequence of holo


morphic functions on 0. Assume that

24. Let 0

lo l

f1 ( z)l dxdy

<

< oo,

where C does not depend on j. Prove that U1} is a normal family.


[ Hint: Use the Cauchy inequalities to show that lf(z)l2 does not exceed
2
the mean value of 111 on a small disc centered at z and contained in
U-see Exercise 8 in Chapter 4. Then deduce that the fj are locally
uniformly bounded.]
25. The collection of all linear fractional transformations forms a group un
der composition. Prove that this group is nonabelian.

Exercises 26-30 are about the cross ratio, which is defined in Exercise
26.

a2, a3,a4 are three distinct points in tC U { oo} = C, then prove that
there is one and only one LFT (linear fractional transformation) which
sends these points to 1, 0, and oo, respectively. Write a formula for this
LFT (you will have to consider separately the case when one of the aj 's
is oo ) .
If a1 is another extended complex number, distinct from the points
a2,a3, a4, then (a i, a2,a3,a4) is defined to be the image of a1 under
the LFT constructed above. This 4-tuple is called the cross ratio of

26. If

a1, a2,a3,a4.

27. Prove that ifT is an LFT, then


28. Prove that

(Ta1,Ta.2,Ta3,Ta4)

(a1,a2,a3,a4).

(a1,a2,a3,a4) is real if and only if a1,a2,a3,a4 lie on a gen

eralized circle.
29. Using the result of Exercise 28, prove that an LFT carries circles and

lines into circles and lines.

a and a* to be symmetric with respect to the cir


cle or line determined by a1,a2,a3 (these three points being distinct) if
(a,ai, a2,a3) (a*,ai,a2,a.3). Prove that if a,a* are symmetric with re
spect to the line/circle through a1,a.2,a3 and if A is an LFT, then Aa and
Aa* are symmetric with respect to the line/circle through Aa1, Aa2, Aa3.
Give a geometric interpretation of symmetry in the case that the three
points ai,a2,a3 determine a line.

30. Define two points

206

Exercises

31. Let C1 and C2 be concentric circles. Show that the ratio of the radii of
these circles is unchanged by any linear fractional transformation which
sends them to two new concentric circles. [More precisely, the ratio of
the larger radius to the smaller is unchanged after application of the
linear fractional transformation.]

32. Construct a linear fractional transformation that sends the unit disc to
the half plane that lies below the line x + 2y

4.

33. Complete the following outline to obtain an alternative proof that the
Mobius transformations <Pa map the unit disc conformally to the unit
disc:

lzl=1,

Check that <P-a is the inverse of <Pa Hence <Pa must be one-to-one.
If

z = =1
l1 :z1 l=;I .

then calculate that

Thus <Pa maps the unit circle to the unit circle.

Since <Pa(a)

0, it

follows from connectedness of the unit disc that <Pa maps the disc D into
itself. But, replacing a by -a, we see that <P-a maps D into D. Hence
<Pa maps D onto D. That is the desired conclusion.

34. Provide an alternative proof of Mantel's theorem by using a connectivity


argument: Find a subsequence that converges at a point. Then find a
subsequence of that subsequence that converges on a disc about that
point-by making each coefficient of the power series expansions of the
functions converge. That is, for each fixed n, one makes the sequence of
nth coefficients converge for the functions in the subsequence. Then use

P(z)

connectivity to spread the result to larger and larger sets.

35.

(a)

Prove: If

number of values o: E C, the equation

1.

distinct solutions. In particular, if P : <C


the degree of P is

(b)

Use part

(a)

P(z) =

is a polynomial of degree k then for all but a finite

--+

o: has exactly

<C is one-to-one, then

to derive Theorem 6.1.1 from Theorem 4.7.5.

Chapter 7

Harmonic Functions

Let F be a holomorphic function on an open set U C Write F


where u and

u +iv,

are real-valued. We have already observed that the real part

u satisfies a certain partial differential equation known as Laplace

a2

82

8x2 + 8y2

's

equation:

0.

(Of course the imaginary part v satisfies the same equation.) In this chapter
we shall examine systematically those C2 functions that satisfy this equation.
They are called harmonic functions.
There are two main justifications for this study: First, the consideration
of harmonic functions illuminates the behavior of holomorphic functions: If
two holomorphic functions on a connected open set have the same real part,
then the difference of the functions attains only imaginary values and, by the
open mapping theorem, the difference is therefore an imaginary constant.
Thus the real part (or the imaginary part) of a holomorphic function already
contains essentially complete information about the holomorphic function
itself. Second, harmonic functions are the most classical and fundamental
instance of solutions of what are known as linear, elliptic partial differential
equations. The detailed consideration of any general results from the theory
of partial differential equations far exceeds the scope of this text. But it is
important that harmonic function theory fits into this larger context (see

[KRA2]

for more on these matters).

One of the most fascinating features

of basic complex function theory is that it is the birthplace of a number


of other branches of mathematics. The theory of elliptic partial differential
equations is one of these.

207

7. Harmonic Functions

208

7.1. Basic Properties of Harmonic Functions


Recall from Section 1.4 the precise definition of harmonic function:

Definition 7.1.1. A real-valued function u : U --+IR on an open set U <::;; C

is harmonic if it is C2 on u and

u=O,
where the Laplacian u is defined by

( ::2 ::2)
+

u.

This definition applies as well to complex-valued functions. A complex


valued function is harmonic if and only if its real and imaginary parts are
each harmonic (Exercise 16). There is hardly any reason, at least iny this
text, to consider complex-valued har nic functions in their own right; we
seldom do so.
The first thing that we need to check is that real-valued harmonic
functions really are just those functions that arise as the real parts of holo
morphic functions--at least locally.

(We shall see later that certain com

plications arise when the harmonic function is defined on a set that is not
holomorphically simply connected; for now we confine ourselves to the disc.)

Lemma 7.1.2. If u: V--+IR is a harmonic function on a disc V, then there


is a holornorphic function F : V--+ C such that Re F = u on V.
Proof. (Corollary 1.5.2. For convenience, we recall the proof.) We want to
find av: V --+ IR such that

u + iv: V--+ C
is holomorphic.

Note that a C1 function v

V --+ IR will make u + iv

holomorphic if and only if

av

au

ax

ay

and

av
ay

These are, of course, the Cauchy-Riemann equations. To determine whether


such avexists, we wish to apply Theorem 1.5.l. We need to check that

: ( -) :x ( ) .
y

This condition does indeed hold by hy pothesis:

6u

( ::2 ::2)
+

u= 0

on V. So Theorem 1.5.1 applies and a suitablev exists.

209

7.1. B asic Properties of Harmonic Functions

Note that v is uniquely determined by

u except for an additive constant:

The Cauchy-Riemann equations determine the partial derivatives of v and


hence determine v up to an additive constant.
determination, up to a constant, of v by

One can also think of the

in another way, as noted briefly

in the introductory remarks to this chapter: If


that

is holomorphic, then H

u +iv

i(v

- v)

is another function such

(u +iv) - (u +iv)

is a

holomorphic function with zero real part; hence its image is not open. Thus

H must be a constant, and v and


function v such that

u.

u +iv

Corollary 7.1.3. If
U C, then

C00

differ by a constant. Any (harmonic)

is holomorphic is called a

--+

harmonic conjugate of

IR is a harmonic function on an open set

Proof. It suffices to check that, for each disc V U, the restriction

C00

For each such V, there is a holomorphic function Fv : V

that ReFv

ul v

Since Fv is

C00,

then so is

--+

uI v

is

C such
D

u.

Lemma 7.1.2 actually applies to any holomorphically simply connected


open set. The most convenient proof in this more general case proceeds by
noticing that if f

u +iv
!'

is holomorphic, then

au
ax

i av

ax

au - au
i .
ax

ay

The second equality comes, of course, from the Cauchy-Riemann equations.


Thus we can find f' (and hence, by integration, f) directly from
details follow:

u.

The

Lemma 7.1.4. If U is a holomorphically simply connected open set and


if

: U

--+

IR is a harmonic function, then there is a

harmonic function v such that

u +iv

: U

--+

C2 (indeed

C00)

C is holomorphic.

Proof. Consider the function H given by


H

=au -iau

The real and imaginary parts are


harmonic function). Moreover,

because

6u

0. Also

ax

C1

ay

since

.
u

is

C2

(by the definition of

7. Harmonic Functions

210

by the equality of mixed partial derivatives for a C2 function. Hence the


function H is holomorphic, since it is C1 and its real and imaginary parts
satisfy the Cauchy-Riemann equations.
Because U is holomorphically simply connected, there is a holomorphic
<C with F' =H. Write F = u +iv, so that

function F : U---+

ou
'
F =
ox

+i

ov
ox

ou
ox

ou
oy ,

where the second equality holds by the Cauchy-Riemann equations. Since

F' =H =

ou/oy.

ou/ox - iou/oy, it follows that ou/ox =ou/ox and ou/oy =


u and u differ by a constant c. Hence F - c =u +iv is also

Thus

holomorphic and Re (F - c) = u as required.

7.2. The Maximum Principle and the Mean Value Property


The fact that a harmonic function is locally the real part of a holomorphic
function has a number of important consequences. Two of these will be the
subject of this section. We first discuss the maximum principle, which is
in a sense the analogue of the maximum modulus principle for holomorphic
functions that we discussed in Section

5.4.

Theorem 7.2.1 (The maximum principle for harmonic functions). If u :


U---+ is harmonic on a connected open set U and if there is a point Po EU
with the property that u(Po) =supQEU u(Q), then u is constant on U.
Proof. Let M = {P E U

u(P) = sup(EU u(()}.

We shall show that

M is open and (relatively) closed in U. Hence, since M is nonempty by


hypothesis and U is connected, M is all of U .
For the openness, suppose that P is a point (possibly distinct from
Po) with u(P) =sup(EU u((). Choose a disc V =D(P, r ) that is centered
at P and contained in U. Let H : V ---+ <C be a holomorphic function with
ReH
u. Define F(z) = eH(z)_ Then IF(P)I
sup(ED IF(()!. By the
maximum modulus principle for holomorphic functions (Theorem 5.4.2),
=

the function F must be constant on V. Hence u is constant on V (since


This shows that the set M = {P: u(P) =sup(Euu(()} is
open. The continuity of u implies that M is closed in U. As noted , we may

u =log IF!).

thus conclude from the connectedness of U that M =U and therefore u is

constant.

Corollary 7.2.2 (Minimum principle for harmonic functions). If u: U---+


is a harmonic function on a connected open set U

<C

and if there is a point

Po EU such that u(Po) =inf QEU u(Q), then u is constant on U.

Proof. Apply the theorem to -u.

211

7.2. The Maximum Principle

An important and intuitively appealing consequence of the maximum


principle is the following result (which is sometimes called the boundary
maximum principle). Recall that a continuous function on a compact set
assumes a maximum value. W hen the function is harmonic, the maximum
occurs at the boundary in the following precise sense:

Corollary 7.2.3. Let U C be a bounded, connected open set and let u


be a continuous, real-valued function on the closure U of U that is harmonic
on U. Then
maxu

maxu.
8U

Proof. By the remarks preceding the statement of the corollary, there is


a

U at which u(P) is maximal. If

U, then u is constant by the

theorem and the conclusion of the corollary is certainly true. If


then the conclusion of the corollary is, of course, also true.

8U,

Corollary 7.2.4. Let U C be a bounded connected open set and let u


be a continuous real-valued function on the closure U of U that is harmonic
on U. Then
minu

minu.
8U

Proof. Apply the preceding corollary to -u.

We turn now to a property of harmonic functions that is analogous


to the Cauchy integral formula.

It allows us to ascertain the value of a

harmonic function u at the center of a disc from its values on the boundary.
[Later on we shall learn a technique for determining the value of u at any
point of the disc from its values on the boundary.]

Theorem 7.2.5 (The mean value property). Suppose that u: U---? IR is a


harmonic function on an open set U C and that

>

0. Then
u(P)

Proof. Choose s

>

u(P)

D(P, s )

+iv(P)

1211"

u(P

with

U for some

+ rei0)d0.

D(P, s ) U. Let H : D(P, s ) ---? C be a


+iv. By the Cauchy integral formula
respect to the circle 'Y that bounds D(P, r),

such that

holomorphic function with H


applied to H on

1
27r

D(P, r)

H(P)

_1 J H(()
27ri h ( - p d(
1
H(P + rei0)
.
ire10dO
27fi 0 (P + re10) - P

1211"

212

7. Harmonic Functions

___!__ f2n u(P + rei9) + iv(P + rei9)d(}


27r lo
___!__ f2n u(P + rei9)d(} + i_!_ f2n v(P + rei9)d(}.
27r lo
27r lo
Taking real and imaginary parts yields

u(P)

___!__ f2n u(P + rei9)d(}

27r lo

as desired.
If u1 :

D(O, 1)

JR. and u2 : D(O, 1)


D(O, 1)

each of which is harmonic on

l zl

1},

then u1

JR. are two continuous functions,


oD(O, 1) = { z :

and if u1 = u2 on

u2. This assertion follows from the boundary maximum

principle (Corollary

7.2.3)

applied to u1 - u2 and u2 - u1. Thus, in effect,

D(O, 1) is
completely determined by its values on D(O, l)\D(O, 1) = oD(O, 1). Theorem
7.2.5 makes thi8 determination explicit for the point z = 0: u (O) is the mean
value (average) of u on oD(O, 1). It is therefore natural to a8k for an explicit
formula for the values of u at other points of D(O, 1), not just the point 0.
a harmonic function u on

D(O, 1)

that extends continuously to

There is such a formula, and it is the subject of the next section.

7 .3. The Poisson Integral Formula


We begin with a lemma collecting results proved in Section
morphic mappings of the disc to itself.

6.2 about biholo

In the present context, one might

think of these facts as saying, in effect, that any result that can be proved
about the point 0 in the unit disc can be translated to a statement about
other points in the disc as well.

Lemma 7.3.1.

L et a E D(O, 1) .

Then the holomorphic function

<Pa(z)

z-a

-_1- az

has the following properties:

(1) <Pa

is holomorphic and invertible on a neighborhood of D(O,

(2) <Pa : D(O, 1 ) D(O, 1)


( 3 ) (a)-l = cP-a;
(4) <Pa(a) = 0.

Proof. See Section

then u

is one-to-one and onto ;

6.2.

Lemma 7.3.2. Ifu: U


H is harmonic.

1) ;

JR. is harmonic and if H:

U is holomorphic,

213

7.3. The Poisson Integral Formula

Proof. We calculate that

b.(uoH)

since

u is

harmonic. [Note here that we have used the chain rule in complex

notation, as treated in Exercise 49 of Chapter l.]

One can also prove Lemma 7.3.2 by using the fact that a C2 function is
harmonic if and only if it is locally the real part of a holomorphic function
(Exercise 32).
The following theorem shows how to calculate a harmonic function on
the disc from its "boundary values," that is, its values on the circle that
bounds the disc. This extends Theorem 7.2.5 in the way we had predicted
at the end of Section 7.2.

Theorem 7.3.3 (The Poisson integral formula).

Let u

---+

JR be a

harmonic function on a neighborhood of D ( O, 1 ) . Then, for any point

D ( O, 1 ) ,
-

u(a) - 211"

271"

u(ei

!f;

J aJ 2

.!f; dv1.
12

I a - ei

Remark: The expression

Ja l 2
211" Ja - ei!f;J2
1

is called the
the formula

1 -

Poisson kernel for the unit disc. It is often convenient to rewrite


in the theorem by setting a= JaJeilJ = rei!J. Then the theorem

says that
.

u(rei0) = 211"

271"

1 -r2

u(ei'lf;)

l-2rcos(O-'lj;)+r2

The Poisson kernel now has the form

Pr (0

'ljJ)

1 -r2

2_

211" 1

2rcos(O-'ljJ) + r2'

d'ljJ.

214

7. Harmonic Functions

with

Proof of Theorem 7.3.3. We apply the mean value property to the har

monic function u

<P-a

u(a)

This y ields

u o <P-a(O) =

1211"u(<P-a(ei9))d0.

271" 0
We wish to express this integral in terms ofu directly, by a change of variable.
-

This change is most transparent if we first convert the integral to a complex


line integral. Namely, we rewrite ( * ) as

(*)

_1 _
27ri
_1
27ri

{ 211"u(<P-a(ei9)) _ iei9 d (}
ei9
lo
u(<P-a(())
J
d(.

!aD{0,1)

(=<Pa(). This is a one-to-one, C1 transformation on a neigh


D(O, 1), hence in particular on 8D(O, 1). Also <Pa(8D(O, 1)) =

Now let
borhood of
8D(O,

1)

and

Therefore

u() '
_1 r
a()d
27ri laD{0 , 1) <Pa()
1 { 211"
u(eitf>)
27ri lo (eitf> a)/(1 aeitf>)

u(a) =

1211"u(eitP)

271" 0

1
(

la12
. it/> d'l/J
ie
ae1'1'1.)2

_
-

1 - la l2
d'lf;.
I e1'ti> a 12
-

This completes the proof of the theorem.

As noted in Chapter 3, the Cauchy integral formula not only reproduces


holomorphic functions but also creates them: If f is any continuous function
on 8D(O,

1),

then

F(z)
is holomorphic for

D(O,

connection between f and


limit" of the function

0 on D(O,

F.

F;

1).

11"
{2

2m lo

J(()

d(

But, of course, there is in general no direct

one cannot hope to recover fas the "boundary

For instance, if J(()

1/(

on 8D(O,

1),

then

1).

For harmonic functions the situation is different. The Poisson integral


formula both reproduces and creates harmonic functions (the second of these

215

7.3. The Poisson Integral Formula

properties will be verified below). But, in contrast to the holomorphic case,


there is a simple connection between a continuous function

and the created harmonic function

on

D.

on

8D(O, 1)

The following theorem states

this connection precisely. The theorem is usually called "the solution of the
Dirichlet problem for the disc."

It is a special case of a standard type of

boundary value problem in partial differential equations [KRA2].

Theorem

7.3.4 (Solution of the Dirichlet problem for the disc). Let

a continuous function on

u(z)

27r

8D(O, 1).

127r f(en/i. )

1 - I z 12
. d'ljJ
I z - ei1/il2

if
if

is continuous on

Proof (Part I).

u(z)

D(O, 1)

To see that

and harmonic on
is harmonic in

D(O, 1)

8D(O, 1).

D(O, 1).

D(O, 1),

we write

The first integral is a holomorphic (hence harmonic) function on


because 6.

g(
z

since

be

Define

f(z)
Then

482 /8zaz

2
ei
{ 7r f(eifJ) . fJ d(}
ei8 - z
lo

ei8 /(ei8-z)

D(O, 1)

and

is holomorphic in

2
{ 7r f(eifJ) g
z
lo

z.

( ei.8ei-fJ z ) d(}

0,

The second integral is harmonic. This

follows from the fact that the integrand is the conjugate of a holomorphic
function and one can differentiate under the integral sign, just as for the first
integral. The last integral yields a constant, which is certainly a harmonic
function.
Thus the function

is a sum of harmonic functions and hence is har-

monic.

Proof (Part II).

The proof of the continuity assertion involves a good deal

of complicated estimation, so it might be a good idea first to understand


the rough idea of the proof. Think of the Poisson integral formula as saying
that the value

u(rei8)

of

at an interior point

rei8

is a weighted average of

the values off on the boundary circle, where the weight Pr(O- 'lj!) is always

216

7.

Harmonic Functions

Po= eieo

= reie

7.1

Figure

1. It is easy to see that, for () fixed,


1/J i= (). Thus, if r is very close to
boundary, then Pr ( () - 1/J) is close to 0 except

positive and has total weight equal to


the limit limr_, 1- Pr ( ()

1,

so that

when

1/J

reiO

- 1/J)

equals 0 if

is close to the

is close to (). So we are taking a weighted average with most of the

weight concentrated near the point


that

u(rei0)

converges to

f(eiO)

as

eiO.

This makes it natural to believe

1-.

--t

The following proof makes this

intuition precise:

To verify the continuity assertion in detail, note that, since v ( z )

harmonic on a neighborhood of

1
Now if Po
Figure

7 .1),

eiOo

8D

it holds for each z E

1
27r

1 - lzl2
I z - e1 1/ll2 d'lj;.

1271'
0

is fixed and if z

reiO

D(O, 1)

1 is

that

is near to Po

( see

then

where we have used


Let M

( )

v z

D(O, 1),

(* ) .

maxaD(O,l)

lfl.

on 8D(O, 1), we may select


f(eit)I < t/2. Now choose z
=

f is uniformly continuous
6 > 0 such that if Is - ti < 6, then lf(eis) reiO so near to Po that 1 e - eo l < 6/3, r 1/2,

Choose

>

0. Since

7.3.

217

The Poisson Integral Formula

and ll -rl < 82E/(1000M). Then, from (**),

d'ljJ
__!__27r 1{1/J:l1/J-9ol<8} j f(ei9o)-f(ei1/J) I ll - rei(9-1/J
)l2
1 r2

d'ljJ
__!__271" 1{1/J:l1/J-9ol2'.8} j f(ei9o)-f(ei1/J) I ll-rei(9
-1/J)l2
1

<

lu ( P ) - u ( z ) I

r2

I +II.

Then term I can be estimated rather simply by


I

1
E
1 -r2
-
.
d'ljJ
27r {1/J:l1/J-9old} 2 I 1 -rei(0-1/J) 12
271"
E 1
1 -r2
-d'ljJ
2 271"
ll-rei(9-1/J)l2

< <
=

1
1
o

by ( * ).
To estimate II, note first that for 0 :S lal :S 7r, 1-cos a a2 /20. This
is elementary (see Exercise 80). Then by our choice of 8 and r we have
ll -rei(0-1/J)l2

>

>

(1 - r)2 +2r(l - cos(O-'lj;))


2r(l - cos(O-'l/J))
(0 -'lj;)2
r
20

Thus

1 -r2
1
d'ljJ.
40M
27r {1/J:l1/J-9ol2'.8}
(0 -'l/J)2
Now we combine the inequalities I O - O l < 8/3 and l'l/J -Ool 8 to obtain
o
that, for 'ljJ in the domain of integration,

<II -

28
I O-'l/J I 3

Thus the integral that estimates the term II can itself be estimated to yield
1 360M [271"
(1 +r)(l -r)d'ljJ
271" 48"2 J
o
__!__ 720M . 82E
<
271"
27r 482
lOOOM
< 2
Combining the estimates on terms I and II yields that
II

<

lu ( P )

That is what we wished to prove.

( ) I < E.

- u z

218

7. Harmonic Functions

As an exercise, you should verify

(using a change of variables) that if

is harmonic on a neighborhood of D(P, R), then, for z E D(P,

R),

f 27r R2 - l z - Pl2
i1/J
u (P +Re )d'lf;.
i
2
1/J
l ( z - P) Re l
27r lo
This is the analogue of Theorem 7.3.3 for the disc D(P, R). The analogue of
_

u z

( ) -

Theorem 7.3.4 follows routinely as well.

7.4. Regularity of Harmonic Functions


The goal of this section is to prove a converse of Theorem 7.2.5. We shall
prove that a continuous function with the mean value property is necessarily
harmonic and, consequently, C00 In fact, we shall prove something slightly
stronger. For this purpose we introduce a convenient piece of terminology.
(This terminology is for internal, temporary use only; it is not standard
usage.)

Definition 7.4.1. A continuous function h: U---+ JR on an open set U C


has the small circle mean value
there is an

f.p

>

(SCMV) property if,

0 such that D(P,


h(P)

f.p) U and

for each point

for every 0

PE

U,

< f. < Ep,

21r
2_ f h(P + Eei0)d0.
27r lo

The SCMV property allows the size of

f.p

to vary arbitrarily with P.

Theorem 7.4.2. If h : U ---+ JR is a continuous function on

an

open set U

with the SCMV property, then h is harmonic.

Corollary 7.4.3. If { hj} is a sequence of real-valued harmonic functions


that converges uniformly on compact subsets of U to a function h: U---+ JR,
then h is harmonic on U.

Proof of Corollary 7.4.3, given Theorem 7.4.2. The function his ob

r) is a closed disc in
21r
2_ f hj(P+rei0) d(}
27r lo

viously continuous. Also, if D(P,


hj(P)

U, then

for each j by Theorem 7.2.5. It follows that


h(P)

21r
2_ f h(P +reiO) d(}
27r lo

by uniform convergence. Thus hhas the SCMV property. So his harmonic.

7.4. Regularity of Harmonic Functions

219

This corollary can also be proved directly, without recourse to the the
orem, by using the fact that each hj satisfies the Poisson integral formula
on closed discs in U and hence that hdoes also ( Exercise

28). See Exercise

29 for yet another approach.


Proof of Theorem 7.4.2. The proof proceeds by way of a lemma that is

interesting in its own right:


Lemma 7.4.4. If g : V

-t

is a continuous function on a connected open

set V with the SCMV property and if there is a point

g(Po)

supQEV

then

g(Q),

Po

E V such that

is constant on V.

Proof of Lemma 7.4.4. As in the proof of Theorem


s

supQEV

because

Po

g(Q) and

define M

{z

EV:

g(z)

EM. Also M is closed in V because

7.2.1, we set

}.

Then M is nonempty

g is

continuous. We shall

show that M is open.


For this, let

E M. Choose

property. Then, for 0


s

f.p

as in the definition of the SCMV

< f. < f.p,

g(P)

s.

Since the first and last expressions are equal, the inequality must be an
equality. Therefore
holds for all 0

g(P

< f. < f.p,

f.ei8

D(P, f.p)

for all 0 () 27r. Since this identity


M. Hence M is open.

Since M is open in V, closed in V, and nonempty, the connectedness


of V implies that M

V. Thus

= s

as required.

Proof of Theorem 7.4.2. Let V be an open disc such that V U. By

Theorem

7.3.4, there is a harmonic function

function uv on V defined by
uV (z)

{ h(z)(z)
uv

is continuous. We want to show that h

if
if
=

z
z

uv : V

-t

such that the

EV
E av

uv on V as well as on

For this purpose, we consider the continuous function w


on V. Now w vanishes on

av.

av.
=

uv

Also, it satisfies the SCMV property on V

because both hand uv do. It follows from Lemma

7.4.4 that w 0 on V:
D would
occur in D so that wwould be a positive constant on D (contradicting the
fact that wis continuous on D and equal to 0 on aD).
If supw > 0, then, because w

0 on

aD,

the maximum of won

220

7.

Harmonic Functions

By applying this reasoning to -w, we find also that -w


Thus w = 0 on

D and

hence h

::; 0

on D.

uv on D. In particular, h is harmonic on

D.
Notice that the disc D was chosen arbitrarily in U and h was shown to
be harmonic on D. It follows that h is harmonic on all of

U.

The arguments in this section are important techniques.

They are

worth mastering. They will arise again later when we consider subharmonic
functions.

7.5. The Schwarz Reflection Principle


Before continuing our study of harmonic functions as such, we present in
this section an application of what we have learned so far. This will already
illustrate the importance and power of harmonic function theory and will
provide us with a striking result about holomorphic functions.
The next lemma is the technical key to our main result:

Lemma 7.5.1.

Let V be a connected open set in C Suppose that V n


( real axis )
{x E JR: a< x < b}. Set U
{z E V: Imz > O}. Assume
v : U
JR is harmonic and that, for each ( E V n (real axis),
=

---+

lim

Set

U3z---.(

v(z)

0.

{ z : z E U}. Denn e
vz
( )

v(z)

if

if

-v(z)

if

z EU
z E n ( real
z EU.

Then vis harmonic on the open set U U

U U {x E JR: a< x < b}.

Proof. It is obvious from the definitions that

{x E JR

: a<

<

b}.

axis )

To see that

v is continuous on W
v is harmonic on W, we
=

shall check the SCMV property.


If

P EU,

then the SCMV property at

in a neighborhood of

-v(z)

for all z in a neighborhood of

is harmonic

( exercise ) .

In this case, choose

x < b}.

Then for 0 <

is clearly valid because v


v

P EU we note that
=

and that the function z

---+

-v(z)
P. It

So the SCMV property again follows for these

remains to check the assertion for

a<

and vis harmonic. For the case

P E {x E JR :

a<

x < b}.

Ep > 0 so small that D(P, Ep) U U


E < Ep, it holds that
v(P)

__!___ f21f v(P + Eei8)de


lo

27r

U U { x E JR:

221

Principle

7.5. The Schwarz Reflection

x=a

Figure
because v(x +

iy)

- v ( x - iy).

{ 27r v(P + Eei0) d(}


lo

7.2

In detail,

1 1r v(P + EeiO)
1r
+ 1 v(P + Eei(H7r))
1 1r v(P + Eei0) fo'Tr v(P + Eei(7r-O))
d(}

d(}

d(} -

d(}

0
[since

J01r v(P+Eei(7r-O)) d(}

J01r v(P+Eei0) d(} by change of variables].

v satisfies the SCMV property and is therefore harmonic,

as

claimed.

Thus

The lemma provides a way of extending a harmonic function from a


given open set to a larger open set. Ths method is known as the Schwarz
reflection principle.

One can think of U as the reflection of U in the real

axis, and the definition of


of reflecting the function

on

fJ

as the correspondingly appropriate idea

See Figure

v.

7.2.

We shall see in a moment that

there is a similar method of reflecting holomorphic functions, too. Before


we go on to that result, it is worth noting how extraordinary the fact is
that

on U and

v on fJ

fit together to be C00 (For comparison purposes

try extending the function

f(x)

x2

on [O, 1] to be odd on [-1, 1].) This is

clearly a very special property of harmonic functions.


Later on, we shall see that the smoothness of a harmonic function at
a boundary point is directly controlled by the smoothness of the boundary
function itself (e.g., Section

14.5).

real axis is approached in Lemma

The boundedness of derivatives as the

7.5.1

is a special case of these estimates,

called "Schauder estimates," for elliptic partial differential equations.

7.

222

Harmonic Functions

The principal result of this section is a closely related extension prin


ciple for holomorphic functions.

Theorem 7.5.2

( Schwarz reflection principle for holomorphic functions ) .


connected open set in C such that V n ( the real axis )
{x E
IR : a< x < b} for some a, b E R Set U
{ z E V : Im z > 0 }. Suppose that
F : U -+ C is holomorphic and that
Let V be a

lim lmF ( z )
U3z-+x

for each x E IR with a< x < b. Defi.ne fJ {z EC: z EU}. Then there is a
holomorphic function G on U U fJ U {x E IR: a< x < b} such that GI u F.
In particular, <P(x) = limu3z-+x ReF ( z ) exists for each x
x + iO E (a, b).
Also
F (z)
if z EU
G(z)
<P(x) + iO if z E Lx E IR: a< x < b}
F (z)
if z EU.
=

Before beginning the proof proper of the Schwarz reflection principle


for holomorphic functions, let us make a few preliminary remarks.
of all, recall that the function z

-+

F ( z ) on

fJ

First

is holomorphic, as one sees

either by checking the Cauchy-Riemann equations or by noting that F ( z )

E fJ,

has, locally about P

a power series expansion in ( z - P ) ; hence F ( z )

has a local power series expansion about P in


the proof lies in showing that F ( z ) on

{x E IR:

( z - P ) . Thus the burden of


fJ and F ( z ) on U "fit together" along

a < x < b} to form the holomorphic function G. The surprising


thing here ( in contrast with the lemma ) is that at first we do not even know
that F ( z ) on fJ and F on U fit together continuously at {x E IR : a< x < b}
because we have not assumed in advance that limu3z-+x ReF ( z ) exists at
any x E IR: This is part of what we have to prove.
What is clear at once, however, is that if there is any holomorphic

U U fJ U {x E IR: a < x < b}, then it must be


fl. The reason is that H( z ) is a holomorphic function
on U U fJ U {x E IR : a < x < b} and agrees with H on {x : a < x < b}
( since His real there ) . It follows that H( z ) H( z ) F ( z ) for all z E fJ by

extension H of F to all of
that H( z )

F ( z ) on

Theorem 3.6.1.

x E R,
{x ER: a< x < b}.

Proof of Schwarz reflection for holomorphic functions. Let

a<

x<

b.

Choose E > 0 such that

D(x, E )

U U fJ U

Set
v (z)

ImF ( z ) '

E D(x, E )

nu.

7.5.

The Schwarz Reflection Principle

223

Then limz-+t v(z) = 0 for all t E {s E : a < s < b} n D(x, t ) . Also vis
harmonic on D(x, t ) n U. By Lemma 7.5.1, there is a harmonic function v
on all of D(x, t ) such that v von D(x, t ) n U. Choose a harmonic function
u such that u + iV is holomorphic on D(x, t ) .
=

On

D(x, t ) n U,
Im (F - (u + iV))

(ImF) - v
(ImF) - v

0.
Since only a constant holomorphic function can have an everywhere zero
imaginary part o( n a connected open set), we conclude that F= (u+iv)+
C, C a real constant, on D(x, t ) n U. Set Go = u + iv+ C so that Go is
holomorphic on all of D(x, t ) and Go= F on D(x, t ) n U.
Thus we see that the original holomorphic function F has an analytic
extension Go to D(x, t ) . The function ,\ : z Go(z) is also holomorphic
on D(x, t ) . Since Go is real-valued on D(x, t ) n t( he real axis), this new
function ,\ equals Go on D(x, t ) n t( he real axis). Hence Go(z) = Go(z) on
all of D(x, t ) . This shows that the function G defined in the statement of
D
the theorem is holomorphic on U U fJ U {x E :a< x < b}.
We take this opportunity to note that Schwarz reflection is not simply a
fact about reflection in lines. Since lines are conformally equivalent b
( y way
of linear fractional transformations) to circles, it is also possible to perform
Schwarz reflection in a circle w
( ith suitably modified hypotheses). More
is true: One can conformally map any real analytic curve locally to a line
segment; so, with some extra effort, Schwarz reflection may be performed in
any real analytic arc.
Some of these last assertions are explored in the exercises. Meanwhile,
here is a brief example to illustrate the utility of reflection in circles.
Let F be a continuous function on D(O, 1) such that F
is holomorphic on D(O, 1). Suppose that there is an open arc I 8D(O, 1)
such that F l 1 0. Then F = 0 on D(O, 1).
If I is all of 8D(O, 1), then this result follows from the maximum mod
ulus theorem (Corollary 5.4.3). So we may assume that there is some
point of 8D(O, 1) that is not in I. After a rotation, we may assume that
point to be -1. Let : D(O, 1) -t U be the inverse Cayley transform,
</Jz
( )= i(l - z)/(1 + z), where U {z EC: Im z > 0}. Then G = F o -1
is holomorphic on U, continuous on U, and vanishes on a segment J (!)
in the real axis. Let U U be an open half disc with 8U nJ real axis J.
Then we may Schwarz-reflect G to a holomorphic function G on U U U U J.
EXAMPLE 7.5.3.

224

7. Harmonic Functions

But then G is a holomorphic function with zero set having an interior accu
mulation point (any point of
therefore G

J),

hence

0. It follows that G

0 on U;

0 on U. Therefore the original function F is identically zero.

7.6. Harnack's Principle


Since harmonic functions are the real parts of holomorphic functions (at least
on holomorphically simply connected open sets), it is natural to expect there
to be properties of harmonic functions that are analogues of the normal
families properties of holomorphic functions .

The subject of the present

section is a principle about harmonic functions that is closely related to


Montel's theorem for holomorphic functions (Theorem

6.5.3).

This result,

called Harnack's principle, will be used in the next section to study what is
known as the Dirichlet problem: the problem of finding a harmonic function
with specified limits at all boundary points.

Proposition 7.6.1 (The Harnack inequality). Let

be a nonnegative, har

monic function on a neighborhood of D(O, R). Then, for any

D(O, R),

R-lzl
R lzl .
u(O).
u(O) ::; u(z) ::; +
R-lzl
R + lzl .
Proof. Recall the Poisson integral formula for

u(z)
Now

1
- 27r

(* )

and

(** )

on

D(O, R) :

1211' u(Rei'l/J ) R2-lzl2 d1/I.


i, -z 12
IRe'li'

R2-lzl2
R2 - lzl2
::;
IRei'l/J -zl2 (R - lzl)2

From

R+ lzl
R - lzl

we see that

R+ lzl 2_ {211'
<
u(z) u(ReiO)d()
R-lzl 27r lo

R+ lzl
. u(O).
R-lzl

That is one of the desired inequalities; the other is proved similarly.

We state as a corollary a version of Harnack's inequalities on discs that


are not necessarily centered at the origin. The proof is easy and is left to
the reader.

Corollary 7.6.2. Let

be a nonnegative, harmonic function on a neigh

borhood of D(P, R). Then, for any

D(P, R),

R - lz-Pl
R
-Pl .
. u(P) < u(z) < + lz
u(P).
R-lz - Pl
R+ lz-Pl

7.6. Harnack's Principle

225

Theorem 7.6.3 (Harnack's principle). Let u1 ::; u2 ::; . . . be harmonic


functions on a connected open set U C. Then either Uj

-T

o n compact sets or there is a harmonic function u o n U such

oo uniformly
that Uj -Tu

uniformly o n compact sets.


Remark: You can look at the Harnack principle this way: If u1 ::; u2 ::; ...
is an increasing sequence of harmonic functions on U and if,
{ Uj(z)} is bounded, then there is a harmonic u such that Uj

for just one z,


-Tu uniformly

on compact sets. This version makes it clear how surprising the result really
is: At first sight, it seems unlikely that the boundedness of the sequence of
functions at just one point would force boundedness at every other point.

Proof of Harnack's principle. If P is a point in U at which Uj(P)


then, for some ]o, Uj0 (P)

>

0. Thus, for some r

>

-Too,

0, D(P, r) U and Uj0 is

every where positive on D(P, r). Then Harnack's inequality implies that, for
z E D(P,r/2), we have
uj(z) 2:
Thus Uj

-T+oo

on

r - r/2
r+r/2

uj(P)

1
=

uj(P)

/ +oo.

D(P, r/2).

On the other hand, if Q E U is a point at which Uj(Q) tends to a


finite limit f and if D(Q, s) U, then Harnack's inequality implies for
z E D(Q, s/2) that, as
Uj(z) - uk(z) ::;

j, k -T+oo, j

s+s/2
s - s/2

>

k,

(uj - uk)(Q)

3(uj(Q) - uk(Q))

-T

0.

Thus Uj converges uniformly (to a harmonic function) on D(Q, s/2).


We have shown that both the set on which Uj

/ +oo

is open and also

the set on which Uj tends to a finite limit is open. Since U is connected, one
of these sets must be empty. This verifies the alternative stated in Harnack's
principle.
Finally, any compact K U may be covered either by finitely many

D(P, r/2)

or finitely many

D(Q, s/2)

(depending on which case occurs).

Hence we obtain uniform convergence on K to a necessarily harmonic limit


function (see Corollary 7.4.3) or uniform convergence to +oo.

Harnack's principle can also be proved by normal families methods. In


outline, one proceeds as follows: As in the argument just given, one needs
to show that the sets

U1

{z EU: lim uj(z)

<

+oo}

+oo}

and

U2

{z EU: lim uj(z)

7. Harmonic Functions

226

are both open. These statements are local in the sense that a set is open if
and only if its intersection with each open disc is open. This suggests that
to prove Harnack's principle one should first treat the case where U is a disc

D(P, r).

(The details of how to pass from the disc case to the general case

will be discussed later.)


To deal with the case where the domain U
functions
on

Vj D(P, r)

D(P, r).

---+

with

vj(P)

The holomorphic functions

D(P, r), choose harmonic


Uj + ivj holomorphic

D(P, r) since for each point p of K we have


le-ui(p)-ivj(P)I::; e-infKui::; e-infKu1.

each fixed compact subset K of

According to Mantel's theorem (Theorem

{e-(uik+ivik)}

0 and with

e-(ui+ivj) are uniformly bounded on


=

6.5.3),

there is a subsequence

D (P, r) to
e-(u
j
)
i+iv is nowhere
a holomorphic function F0. Since each of the functions
vanishing, it follows that the limit function Fo
D(P, r)
C is either
that converges uniformly on compact subsets of

---+

nowhere zero or is identically zero (Hurwitz's theorem, Theorem

5.3.3).

Ujk is everywhere finite. Moreover, since on


any fixed compact set K the sequence { e-(uik+ivik)} converges uniformly
{ Ujk} converges uniformly to
to Fo, it follows that { - log le-(uh +ivJk) I}
- log ! Fo l . Since the sequence { Uj} is monotone, the uniform convergence on
K of the subsequence { Ujk} implies the uniform convergence of the entire
sequence { Uj}. In the case Fo = 0, it follows similarly that { Uj} tends to
In the former case, limk_,00

+oo uniformly on compact sets. This completes the proof when the domain

is a disc.
The general case, for arbitrary connected U, can be obtained from the
disc case by noting first that the sets U1 and U2 are open, as already noted.
Thus either lim

Uj(z)

<

+oo for all

z EU or lim Uj(z)

+oo for all

z EU. In

either situation, one then gets the uniformity of the convergence on compact

subsets of U by noting that every compact subset K of U is contained in the


union of a finite number of discs

D ( P, r) such that D ( P, 2r) U. Each set


D ( P, 2r) , so that convergence is uniform

Kn D ( P, r) is a compact subset of

on Kn

D ( P, r)

by our disc argument. Thus uniformity on K follows.

7.7. The Dirichlet Problem and Subharmonic Functions


Let U C be an open set, U

-=/= C.

Let f be a given continuous function

on au. Does there exist a continuous function


and

is harmonic on U?

If

taken together are called the

on U such that

exists, is it unique?

Dirichlet problem

l au

These two questions

for the domain U. It has

many motivations from physics (see [COH], [RES]). For instance, suppose
that a flat, thin film of heat-conducting material is in thermal equilibrium.
That is, the temperature at each point of the film is constant with passing

7. 7. The Dirichlet Problem

227

time (the system is in equilibrium). Then its temperature at various points


is a harmonic function.

au

Physical intuition suggests that if the boundary

of the film has a given temperature distribution f :

au

---+

R, then

the temperatures at interior points are uniquely determined. Historically,


physicists have found this intuition strongly compelling, although it is surely
not mathematically convincing.
From the viewpoint of mathematical proof, as opposed to physical intu
ition, the situation is more complicated. Theorem 7.3.4 asserts in effect that
the Dirichlet problem on the unit disc always has a solution. Furthermore,
it has only one solution corresponding to any given boundary function f,
because of the (boundary) maximum principle------Corollary 7.2.3: If

u2

are both solutions, then

so that

u1 - u2

u1 - u2

u1

and

is harmonic and is zero on the boundary,

0. While this reasoning demonstrates that the Dirichlet

problem on a bounded open set U can have at most one solution, it is also
the case that on more complicated domains the Dirichlet problem may not
have any solution.

EXAMPLE 7.7.1. Let U

D(O, 1) \ {O}.

f(z)

{ 1 1
0

Then

D(O, 1)

that if

u(z)

u(z)

u(z')

au.

If there is a continuous

this assertion, note that

u( ei8z)

Set

u must be radial:

on

That

lzl
lz'I The reason for this is
u(ei8z) for any fixed () E R (To see

whenever

is a solution, then so is

1} U {O}.

which solves the Dirichlet problem, then

is, it must be that

IzI = 1
z o.

Then f is a continuous function on

au= {z: lzl

is certainly harmonic and still equals f on

the boundary.) The Dirichlet problem has a unique solution in this setup,
by reasoning using the maximum principle as noted in the paragraph just
before this example. So it must be that

u(z)

u(ei8z).

Now it is straightforward if tedious to check, using the chain rule, that


the Laplace operator

a2 I ax2 + a2 I ay2,

takes the form

O
hence

1a
a
1 a2
+

r 2 ae2 .
u is independent of (),

Then, because our function

when written in polar coordinates,

)
r
(
ar ar

au )
ar (r ar ;
:r (r :r u)
r#
=

1a

when

so that

au
ar

228

7. Harm onic Functions

for some real constant C.


Thus
u

C logr+ D

for some real constant D.


However, there is no choice of C and D that will allow such a

to agree

with the given f on the boundary of U. This particular Dirichlet problem


cannot be solved.
Thus

some

conditions on au are necessary in order that the Dirichlet

problem be solvable for u. It will turn out that if au consists of "smooth"


curves, then the Dirichlet problem is always solvable.

The best possible

general result (which we shall not prove) is that if each connected component
of the boundary of U contains more than one point, then the Dirichlet
problem can always be solved. Thus, in a sense, the example just given is
the typical example of a domain in Con which the Dirichlet problem has, in
general, no solution. In the next section we shall formulate and prove that,
for quite a large class of domains, the Dirichlet problem can be solved for
arbitrarily given (continuous) "boundary values". In the remainder of the
present section we develop the tools that we shall need to prove this.
We first consider the concept of subharmonicity. This is a complex
analytic analogue of the notion of convexity that we motivate by considering
convexity on the real line. For the moment, fix attention on functions from

JR to JR.
On the real line, the analogue of the Laplacian is the operator

d2 / dx2.

The analogue of real-valued harmonic functions (i.e., the functions annihi


lated by this operator) are therefore the linear ones. Let S be the set of
continuous functions f :

JR

JR such that, whenever I

[a, b]

JR and h

is a real-valued harmonic function with f(a) ::; h(a) and f(b) ::; h(b), then
f(x) ::; h(x) for all x

J. (Put simply, if a harmonic function h is at least

as

large as f at the endpoints of an interval, then it is at least as large as f on


the entire interval.) Which functions are in S? The answer is the collection
of all convex functions (in the usual sense). Refer to F igure 7.3. [Recall here
that a function f : [a, b]
and 0 ::; >. ::;

JR is said to be convex if, whenever c, d

[a, b]

1, then
f((l - >.)c + >.d) ::;

(1 - >.)f(c) + >.f(d).]

These considerations give us a way to think about convex functions without


resorting to differentiation.
Our definition of subharmonic function on a domain in C (or
motivated by the discussion in the preceding paragraph.

JR2) is

229

7. 7. The Dirichlet Problem

h(x)

F igure 7.3

Definition 7.7.2. Let U C be an open set and fa real-valued continuous


function on U. Suppose that for each D(P, r) U and every real-valued

h defined
f ::::; h on oD(P, r), it holds
subharmonic on U.
harmonic function

on

neighborhood of D(P, r) which satisfies

that f ::::;

EXAMPLE 7. 7.3. Any harmonic function

on D(P, r). Then f is said to be

on an open set U C is certainly

r) U and h is harmonic and greater than or


equal to u on oD(P, r) , then u - h ::::; 0 on oD(P, r). By the maximum
principle, we conclude that u - h ::::; 0 or u ::::; h on D(P, r). Thus u is

subharmonic.

For if D(P,

subharmonic.
As an exercise, use a limiting argument to see that subharmonic func
tions may also be characterized by comparing them to functions
on any given closed disc D(P,

r) U

It turns out that a function f

and

harmonic

continuous

on D(P, r) .

IR which is

C2

is subharmonic

if and only if 6.f 2'. 0 everywhere. This is analogous to the fact that a

C2

function on (an open set in) IR is convex if and only if it has nonnegative
second derivatives everywhere. For the proofs, see Exercises 41 and 69. The
next proposition will allow us to identify many subharmonic functions which
are only continuous, not

C2,

so that the 6.f 2'. 0 criterion is not applicable.

Proposition 7.7.4. Let f: U

IR be continuous. Suppose that, for each

D(P, r) U,

f(P ) ::::; -

271"

Then f is subharmonic.

127!"
0

f(P +

rei6 ) d0.

7.

230

Harmonic Functions

Figure

7.4

Conversely, if f : U ---+ is a (continuous) subharmonic function and


if D( P, r ) U, then the inequality ( * ) holds.
Proof. Suppose that ( * ) holds for every P, r but
there is a disc D(Q, s )

of D(Q, s ) such that

f is not subharmonic .

Then

and a harmonic function hon a neighborhood

f(zo) > h(zo)


f - hon D(Q, s ) .

h on 8D(Q, s ) but

D(Q, s ) . Consider the function

for some

zo

g is less than or equal to zero on the boundary of the disc


we set M
maxD( s) g and K
{z E D(Q, s ) : g(z)
Q

The function

g(zo) > 0. If
,
M}, then it follows that K is a compact subset of D(Q, s ) . In particular,

but

K is not all of D(Q,

7.4)

s .

Let w be a boundary point of K. Then (see Figure

TJ) at which g is less


g is less than Mon an open arc J of 8D(w, TJ).

for some small T/ > 0 there is a point of 8D(w,

than M. By continuity, then,


Therefore, for this

TJ,

2._ f21T g(w + rJei9)d(}


27r lo
But the left side of this inequality is

<

g(w).

231

7. 7. The Dirichlet Problem

(since

is harmonic). We conclude that

{
27r f(w + T/ei8)d0 < f(w),
lo

1
27r

and this inequality contradicts our hypothesis.

f be subharmonic on U and let D(Q, s) U be


8D(Q, s) --+ R be the Poisson kernel for D(Q, s).

For the converse, let


fixed. Let

P : D(Q, s) x
P > 0. Let> 0.

Recall that

h(z) =

Then

fo27r P(z, ei8)[f(Q


D(Q, s),
8D(Q, s). It

sei8) + ]d0

D(Q, s), and h(()


h(() > f(() for ( E
the continuity of the functions f and h) when o is small
enough. Buth is harmonic on a neighborhood of D(Q, s - o). Thus, by the
subharmonicity hypothesis, f hon D(Q, s - o). In particular,

defines a harmonic function on

continuous on

f(() + > f(()


8D(Q, s - o) (by

follows that

Letting

--+

for

f(Q)

h(Q)

__!___ f27r [f(Q + sei8) + ]d0.


lo

27r

o+ yields the desired inequality.

EXAMPLE 7.7.5. Let

To see this, let D ( W,

F: U--+ C be holomorphic.
s) U. Then, by the mean

IFI is subharmonic.

Then

value property (Theorem

7.2.5 ) ,

IF(Q)I

I__!___ lof27r F(Q


27r

By the proposition,

IFI

sei8)d0

F(z)

f:

EXAMPLE 7.7.6. Let

IFI will
zk, k EN.

f2

o J)(P)

not be harmonic when Fis harmonic.

of

D(P, r)

(f(P ))

<

<

Let

R --+ R be

is subharmonic. For example,

is subharmonic if

To see this, notice that if


(

__!___ f27r IF(Q + sei8)ld0.


lo

27r

U --+ R be subharmonic.

nondecreasing and convex. Then


subharmonic. Also

is subharmonic.

Notice that, in general,


For example, take

is everywhere nonnegative.

U, then

( fo27r J(P rei8)d0)


__!___ lof27r (f(P rei8))d0
__!___ lof27r oJ)(P rei8)d0.

27r

27r

el

is

7. Harmonic Functions

232

Here the second inequality follows from the convexity of <P (this is known as
Jensen's inequality-see Exercise 66). By the proposition, we conclude that
<Po f is subharmonic.
The last example raises a subtle point, which is worth making explicit.
If f is harmonic and <P is convex, then (with only a small change) the ar
gument we have just given shows that <Po f is subharmonic. However, if
f is only subharmonic, then <P must be both convex and nondecreasing in
order for <P o f to be subharmonic. For example, the square of a subhar
monic function f is not necessarily subharmonic [here <P(x) = x2] if f takes
negative values at some points. But if f is every where nonnegative, then
f2 is subharmonic [the function <jJ(x)
x2 is nondecreasing on [O,+oo), but
of course not on all of (-oo, +oo)]. The reader is invited to consider this
matter in further detail in Exercise 49.
=

An immediate consequence of the sub-mean value property is the max


imum principle for subharmonic functions:
Proposition 7.7.7 (Maximum principle). If f is subharmonic on U, and if
there is a P EU such that f(P ) 2: J(z) for all z EU, then J is constant.
Proof. This is proved by using the argument used to prove Lemma 7.4.4,

along with the sub-mean value inequality of Proposition 7.7.4: We need only
note that it is actually the inequality, rather than the equality, of the SCMV
property that is needed in the proof of Lemma 7.4.4. We leave the details
of the assertion as an exercise (Exercise 43 ) .
D
Note in passing that there is no "minimum principle" for subharmonic
functions. Subharmonicity is a "one-sided" property; in the proof of the
maximum principle this assertion corresponds to the fact that Proposition
7.7.4 gives an inequality, not an equality.
Here are some properties of subharmonic functions that we shall need
in the sequel. The third of these explains why subharmonic functions are a
much more flexible tool than holomorphic, or even harmonic, functions.

(1) If Ji,f2 are subharmonic functions on U, then so is Ji+ f2.


(2) If Ji is subharmonic on U and a > 0 is a constant, then nfi is
subharmonic on U.

(3) If fi,h are subharmonic' on U, then g(z)

max{f1(z),f2(z)} is

also subharmonic on U.

Since properties (1) and (2) are clear from Proposition 7.7.4, we shall
prove only property (3). Let Ji ,f2 be subharmonic on U and let D(P, r )
U. Then

7. 7. The Dirichlet Problem

233

max{f1(P), h(P)}

{ fo2rr rei8)d0, fo2rr h(P


[fi(P+rei8),h(P+rei0) dO.
1 r rr
2
Jo
]

<

max

<

fi(P +

rei8)d()

max

271'

This proves that max{f1, h} is subharmonic.


The next concept that we need to introduce is that of a barrier. Namely,
we want to put a geometric-analytic condition on the boundary of a domain
which will rule out the phenomenon that occurs in the first example of this
section (in which the Dirichlet problem could not be solved). The definition
of a barrier at a point PEau is a bit technical, but the existence of a barrier
will turn out to be exactly the hypothesis needed for the construction (later)
of the solution of the Dirichlet problem.

Definition 7.7.8.
b

--t

(i)
(ii)
(iii)
(iv)

IR. a

C be an open set and PEau.


at P if

Let U

barrier for U

is continuous;

is subharmonic on U;

We call a function

b :S O;
{zEau: b(z) = O} = {P}.

Thus the barrier

singles out

in a special function-theoretic fashion.

EXAMPLE 7.7.9. Let U = D(O, 1) and

b(z) = x- 1

is a barrier for U at

P,

= 1+iOEau. Then the function


x

=Re z.

PE U be the
r = IPI = IP-OI and notice that
7. 5. Let Oo = arg P, 0 :::; Oo < 271'. Then the function

EXAMPLE 7.7.10. Let U

C be

where

a bounded open set. Let

point in U which is furthest from 0. Let


U D( 0,

r).

See Figure

z
is a barrier for U at

1----+

Re(e

i Oo z)- r

P.

EXAMPLE 7.7.11. Let U

C be

an open set. Let

PE au. Suppose that


P to some point

there is a closed segment I, of positive length, connecting

QEC\ U

such that In (U) = {P}. Then the function


Zt----+

z-P
z-Q

--

234

7. Harmonic Functions

Figure 7.5

maps C \I to C \ J, where J is a closed (infinite) ray emanating from the


origin. Therefore

has a well-defined branch on


one-to-one map of

U into C.

<:;;; C \I and it extends to be a continuous,

Notice that

'l/J(U) is contained in a closed half


'lf;(U) to the upper half plane and
'l/J(U) to the unit disc with po 'l/J(P ) 1.

plane. Finally, a suitable rotation Po maps


the Cayley transform maps po o
In summary, the function
unit disc and sends

Po

'If; maps

U one-to-one into

the closed

to 1. Thus we have reduced the present example to

the first barrier example (the disc). That is, there exists a barrier for

at

P.
It follows from the last example that if U is bounded by a smooth curve
1
(say of smoothness class C ), then every point of au has a barrier.

EXAMPLE 7. 7.12. Let

>

<:;;; C be an open set and

P E au.

If, for some

0, there is a barrier at

P: If b1 is the barrier for


by
b(z)
is a barrier at

P for UnD(P,r), then there is a barrier for U at


Un D(P,r), then the function b: U
R defined

-E
max(-E,b1(z))

P for U if E

>

if

if

EU\ D(P,r)
EUnD(P,r)

0 is chosen small enough (check this assertion as

an exercise). Thus the concept of the existence of a barrier at


the sense that it depends only on how

P is local, in
U looks in a neighborhood of P. This

235

7. 7. The Dirichlet Problem

property greatly simplifies the task of finding barriers for boundary points
of domains with complicated structures.
In the next section, we are going to use the existence of barriers at
boundary points as the condition for the existence of a solution for the
Dirichlet problem. Therefore it would be a good idea to try to see that a
barrier does not exist in the case where we already know that the Dirichlet
problem cannot be solved-namely for the center of the punctured unit disc
(the example discussed at the beginning of this section).
Concretely, we should try to see that there cannot be a function b on
D(O, 1) such that

(i)
(ii)
(iii)
(iv)

bis continuous on D(O, 1);


bis subharmonic on D(O, 1) \ {O};
b
{z

on D(O, l);

D(O, 1): b(z)

O}

{O}.

The thing to note here is that, if such a b exists, then the function
b(z)

27r

{27r
o b(ei9z) d()
J

also has properties (i)-(iv). Also, for any a > 0, the function abhas these
properties as well. Notice that (iii) and (iv) together then imply that bis
a negative constant on 8D(O, 1). Thus if any barrier b exists for the point
0,

then there is another barrier B such that B

-1 and such that

8D(0,1)

B(ei9z) B(z) for all(); that is, Bis rotationally invariant. We shall now
show that that is impossible.
=

For this purpose, consider the functions

Hr(z)

B(r) + 1
log !z! - 1 ,
1ogr

O<r<l.

Each function Hr is harmonic on D(O, 1) \ {O} and has the same values as
Bon {z: !z! 1} and on {z: !z! r}. Since Bis subharmonic so is B-Hr
from which it follows that, for each z E {z: r lzl 1},
=

B(z) Hr(z).
Thus for each fixed z

D(O, 1), z =I=

0,

we see that

B(z) lim Hr(z).


r-+O+
But, for z =I=

fixed,

B(r) + 1
1og I z I
r-+O+ log r
1.im

7. Harmonic Fu nc tio ns

236

because IB(r)I is bounded and I log rJ---+ +oo. As a result,


B(z) :S -1
for all z

E D(O 1) \ {O}.
,

This contradicts the fact that limz-+O B(z) =0.

Thus the domain D(O, 1) \

{O} has no barrier at the origin.

7.8. The Perron Method and the Solution of the Dirichlet


Problem

In this section we shall show that, on a large class of domains U, the Dirichlet
problem can be solved. The result that we shall prove is not the best possible,
but it is sufficient for most purposes. More refined theorems will be discussed
in the remarks at the end of the section and in the exercises. See also [TSU]
for a complete treatment of these matters. The methodology of this section
is due to 0. Perron (1880-1975).
The solution to the Dirichlet problem is constructed by solving an
extremal problem (just as in the Riemann mapping theorem). It turns
out that a proof of the Riemann mapping theorem can be obtained as a
corollary of the Dirichlet problem (which proof is in fact closer in spirit to
the original proof due to Riemann and Dirichlet). See Exercise 73 for some
of the details.
Theorem 7.8.1. L e t
that

be a bounded, c onnec ted open subset of C suc h

P E au. Then the Dirichlet problem c an


f is a c ontinuous function on au, then
c ontinuous on U, harmonic on U, suc h that u I au = f.

has a barrier bp for eac h

always be solved on
there is a function u

U.

That is, if

The function u is uniquely determined by these c onditions.

Proof. There is no loss of generality in assuming that f is real-valued. As


already noted, the final (uniqueness) statement is an immediate consequence
of the boundary maximum principle (Corollary 7. 2 .3).

To begin the proof of the existence of a suitable harmonic function u,


set
S = {'1/1: 'l/J is subharmonic on

U and

lim sup'lj;(z):::;
U3z-+P

f(P), VP Eau}.

Notice that au is compact so that f is bounded below by some real constant


m. Thus the function 'l/J(z) =mis an element of S. In particular, Si- 0.
Define, for each z

E U,
u(z) =sup 'l/J(z).
1/;ES

We claim that u solves the Dirichlet problem for


proof is broken into three parts:

f and the domain U.

The

237

7.8. The Perron Method

Part 1. The function u is bounded above:


M
max(E8U /(().
U : 'l/;(z) 2': M + E}. Seeking

letting

Let

To see this, we begin by

'I/; E S, E

>

0, and let E{

{z

a contradiction, we assume that E{ is

nonempty.
We claim that

E{

is closed in C Assuming the claim for the

moment, we complete the proof as follows. Since E{ lies in


bounded, hence compact. So

'I/; assumes a

U,

it is

maximum at some point

EE{. But this maximum value is at least M +E, and 'I/; is less than
M + E off E{; therefore 'l/;(p) is a maximum for 'I/; on all of U . By
the maximum principle, 'I/; is constant. Furthermore, that constant,
since p E E{, must equal or exceed M + E. This contradicts the
membership of 'I/; in S (the boundary limits of 'I/; must not exceed
f, which in turn lies below M + E).
This contradiction implies that E{ is empty. But E{
0 for all
E > 0 means that 'I/;(() :S M for all ( E U. This is what we wished
p

to prove.
It remains to see that E{ is closed. Let z

E cE{.

There are three

possibilities:

E C\U: In this case z has a neighborhood which is contained


in C \ U <::;; C \E{.
(b) z E fJU: In this case, because f(z) :S M, the definition of S
implies that there is a neighborhood V of z (in C) such that
'l/;(z) < M + E for all z E v nu. So v nE{
0.
(c) z E U \ E{: In this case , 'l/;(z) < M + E and (since 'I/; is
continuous) there is a neighborhood V of z with V <::;; U on
which 'I/;(() < M + E. These cases together show that C \E{ is
open or E{ is closed.

(a)

Part 2. The function u is harmonic:


let p

E D(P, r).

Let D(P, r) <::;; U and moreover


u, there exists a sequence { 'l/;j}1 <::;;
u(p). Let

By definition of

S such that limj_,00 'l/;j(P)

'11n( Z )

max { 'l/;1 ( Z ) , .. .

, 'l/Jn( Z )}

E U. Then, by property (3) of subharmonic functions


7. 7, '11n is subharmonic on U and '111 :S '112 :S .... Let

for each z
Section

cl>n(z)

_
-

Then

{ '11n

(z)

cl>n

is subharmonic

if

E U \ D(P, r)
E D(P, r) .

sub-mean value property)

cl>n E S (since it agrees with '11n near the boundary


U). Furthermore, cf>1 :S cf>2 :S ... on all of U by the maximum

and of course
of

'11nlaD(P r)
,
on U (by the

Poisson integral of

if

in

principle. F inally,

238

7. Harmonic Functions

for each n. It follows that limn--+oo <I>n(P)


u(p). But the crucial
thing to notice is that, by Harnack's principle, the functions {<I>n}
converge to a harmonic function <I> on the disc D(P, r) (since they
do not go to oo at p). Moreover, <I> ( P)
u(P) by (*). We shall show
in a moment that <I>
u on the entire disc D(P, r). Then it will
follow that u is harmonic near P. Since P was chosen arbitrarily,
this will complete Part 2.
To show that <I> u on D(P, r), pick another point q E D(P, r).
By definition of u, there exists a sequence {Pi }1 S such that
limj--+oo Pi (q) u(q). Let Pi(z) max{pj(z), 'l/ij(z)} for z E U. Let
=

A n(z)

max{p1(z), ..., Pn(z)}

for each z E U. Then, again by property (3) of subharmonic func


tions in Section 7.7, A n is subharmonic on U and A1 A2 . . .
and A n(q)
u(q). Define

Hn(z)

(z)
Poisson integral of AnlaD(P r)

{ An

if z E U \ D(P, r)
if z E D(P, r) .

Then, as in the case of the <I>n 's, the sequence { Hn} has a harmonic
limit function H on D(P, r). Also, H(q)
u(q). Finally, by design,
=

<I>(z) H(z) u(z) on D(P, r) .


Hence H (p)
u (p) because <I> (p)
u(p). Thus the harmonic
function <I> - H satisfies <I> - H 0 on aD(P, r) and (<I> - H)(p)
0. By the maximum principle, <I> = H. This means that for any
q E D(P, r) the harmonic function H of our construction agrees
with the function <I> for p. In other words, the harmonic function <I>
of our construction agrees with H, and hence with u, at each point
of D(P, r). But then u is harmonic on D(P, r). Since P and r were
arbitrary, we conclude that u is harmonic.
=

f(w): Fix w E au. To


prove the assertion, we exploit the barrier at w. Let E > 0. Since

Part

3. For each w Eau, limu3z--+w u(z)

U is bounded, the boundary of U is compact. By the consequent

uniform continuity off, choose 0 > 0 such that if o., (3 E au and


< o, then lf(o.) - J(/3)1 < E. Let bw be a barrier at w.
On the set U \ D(w, o)' the barrier bw has a negative maximum
. Define the function

lo. - /31

g(z)

J(w) + E +

bw (z)

(M - J(w)),

where M is the maximum of Ill on au. The function -g is sub


harmonic since bw is such and -(1/)(M - f(w)) 0. Now for

7.8. The Perron Method

E 8U n

239

D(w, 8)

it holds trivially that

g(() f(w) + > [!(()


For

] +

( E 8U \ D(w, 8), it is clear from


g(() f(w) + + (M - f(w))

f(().

the definition of that

+ > f(().

The maximum principle for subharmonic functions now shows that

any 'ljJ ES satisfies 'ljJ


Namely, 'ljJ

g is

on U (refer to the argument in Part

subharmonic and has lim supu3z- ( 'ljJ

all E 8U; hence 'ljJ

in U. Thus u

g0

lim supu(z)
U3z--+w

g(w)

1).

g) 0

for

on U. As a result,

f(w) + .

This is half of what we need to prove.

For the other half, consider the function

g(z)

f(w)

We shall show that

g ES;

(z)
bw
(M

then u(z)

lim inf u(z)


U3z--+w

g(w)

+ f(w)).
g(z)

J(w)

for z EU and

This will finish the proof, for then

J(w) -
and, since

lim inf u(z)


U3z--+w

lim supu(z)
U3z--+w

f(w) +

J(w). Thus the


> 0 was arbitrary, limu3z--+w u(z)
function which is u on U and f on 8U is continuous on U. The

proof will then be complete.


To see that

g ES, first notice that g is a subharmonic function


+ f(w)) 0 and bw is subharmonic. If ( E 8U n

since -(1/)(M

D(w, 8),

then

and

[!(() + ] - = f(();
if ( E 8U \ D(w, 8), then g(() -M
< f(().
g(() f(w)

<

we are done.

So

g ES

and

Notice that the construction of u, and the proof that u is harmonic,

uses nothing about

or U except that

could in principle be identically


and the continuity of

+oo).

is bounded above (otherwise u

The barrier at a boundary point

check the continuity of u at

w.

As pointed out earlier, the existence of a barrier at

property": It depends only on how U behaves near


check whether u is necessarily continuous at

in a neighborhood of

w.

w,

at that point, are used-one point at a time-to

w.

w strictly

for U is a "local

So, in effect, one can

in terms of local data

240

7. Harmonic Functions

7.9. Conformal Mappings of Annuli

The Riemann mapping theorem tells us that, from the point of view of
complex analysis, there are only two conformally distinct domains that are
homeomorphic to the disc: the disc and the plane. Any other domain home
omorphic to the disc is biholomorphic to one of these. It is natural then to
ask about domains with holes. Take, for example, a domain U with precisely
one hole. Is it conformally equivalent to an annulus?
This question is too difficult for us to answer right now in full generality,
partly because we have not rigorously formulated the concept of having just
one hole. [Incidentally, the answer to the question is "yes": Every open set
in C that is topologically equivalent to an annulus is biholomorphic to an
open set of the form {z E C: r1 < lzl < r2}, 0 ::; r1 < r2 ::; +oo. See
Exercise 76 for further ideas on this question.] We content ourselves for
the moment with addressing a simpler question that is in the same spirit.
Namely, we will classify all annuli up to conformal equivalence. [We treat
the case 0 < r1 < r2 < +oo. See Exercise 77 for the r1 0 and/or r2
+oo
cases.] Notice that if c > 0 is a constant, then, for any r1 < r2, the annuli
=

A1

{z: r1

<

lzl

<

r2} and A2

{z: cr1

<

lzl

<

cr2}

are biholomorphically equivalent under the mapping z 1----t cz. The surprising
fact that we shall prove is that these are the only circumstances under which
two annuli (with 0 < r1 < r2 < +oo) are equivalent:
Theorem 7.9.1. Let, for R1, R2

A1

>

1,

{z EC: 1

<

lzl

<

Ri}

{z E <C: 1

<

lzl

<

R2}.

and
A2

Then A1 is conformally equivalent to A2 if and only if R1

R2.

Proof. The "if" part is obvious.

For the "only if" part, suppose that

>: A1

--t

A2

1
is a biholomorphic equivalence. If K in A2 is compact, then - (K) is also
1
compact (since - is continuous). It follows that if a sequence {w1} in
A1 converges to the boundary (in the obvious sense that it has no interior
accumulation point), then so does the sequence {(w1)} in A2. Moreover,
we claim that if lw1I --t 1, then either, for all such sequences, l(w1)I --t 1 or
l(w1)l --t R2. In the first of these cases, we further claim that if lw1I --t R1,
then l(w1)I --t R2; in the second case we claim that if lw1I --t R1, then
1(Wj) I --t 1. The verifications of these claims are elementary but tricky, and
we defer them until the end of the proof.

7.9. Conformal Mappings

of Annuli

241

After composing <P with a reflection if necessary, we may suppose that

and
Consider the function
h(z) =log lzl log R2 - log l<P(z)I log R1.
This function is harmonic on A1 and extends continuously to Ai . Namely,
if we set h(z) = 0 when z E A1 \ A1, then his continuous on A1, as one
sees immediately from the indicated behavior of <Pas the boundary of A1 is
approached. By the maximum principle, h= 0.
Solving the equation h= 0 for <Pthus yields that
l<P(z)I =lzlf3,
where (3 = log R2/ log R1. Let D(P, r ) A1 be a disc. Then the function
F(z) = zf3 can be made well-defined and holomorphic on D(P, r ) if we
set F(z) = ef3g(z), where g is a holomorphic function on D(P, r ) such that
z = eg(z). In other words, Fis a branch of zf3. [See Exercise 15, Chapter 6,
for details of the idea of "branches" of zf3 as used here.] Also (z)/ F(z)
is holomorphic on D(P, r ) and has unit modulus. By the open mapping
theorem, <P(z)/ F(z) is a constant of modulus one on D(P, r ) . That is,
<P(z) =a zf3
on D(p, r ) for some a EC, lal =1.
Since the computation in the last paragraph can be performed on any
D(P, r ) A1 and since is continuous, it follows that (1/a), for some
a E C with lal = 1 and (3 = log R2/ log R1, is a "branch" of zf3 on the
entire annulus. It follows then that (3 must be a (nonnegative) integer, since
otherwise no such branch exists (see Exercise 15, Chapter 6). Finally, the
only possible nonnegative integer value for (3 is 1; otherwise would not be
one-to-one.
D

Thus <Pcan only be a rotation and therefore R1 =R2.

Proof of the claim. We have already noted that if A1 3 lwjl


8A1, then
A2 3 l<P(wj)I
8A2. In particular, if E > 0 is small, then ({z: 1 < Jzl <
1 + E}) does not intersect {z: lzl = (1 + R2)/2}.
----t

----t

For all j large, <P(wj) must be in a fixed component of {z: lzl # (1 +


R2)/2}: <P(wj) cannot jump back and forth between {z: (1 + R2)/2 < lzl <
R2} and {z: 1 < lzl < (l+R2)/2}. The reason is that ({z: 1 < lzl < l+E})
does not intersect {z : lzl = (1 + R2)/2} and, for J1,J2 large, WJi and Wj2

7. Harmonic Functions

242

can be connected by a curve in {z: 1 < lzl < 1 + t:}; hence so can (wj1)
and (w12) be connected by a curve in {z: lzl-/- (1 + R2)/2}.
Thus one sees that either

for all sequences {Wj} with limj_,00 lwj I


Hm 1(wj)I
J--->00
for all sequences with limj--+oo lwj I
that

or

R2

l.

As noted, by composing with an inversion if necessary, one can suppose

for all sequences Wj with limj_,00 lwj I

l.

Under the assumption in the last paragraph, the same reasoning shows
that either
_lim 1(wj) I 1
J--->00
for all sequences {Wj} with limj--+oo lwj I R1 or
=

Hm 1(wj)I R2
J--->00
for all such sequences. We claim that this first apparent possibility cannot
in fact occur.
=

If it did, then the function 11 would attain a maximum in (the interior


of) its domain, so that would be constant, hence not one-to-one. This
D
establishes the claim.
The classification of planar domains up to biholomorphic equivalence
is a part of the theory of Riemann surfaces (see Section 10.4). For now, we
comment that one of the startling classification theorems (a generalization
of the Riemann mapping theorem) is that any bounded planar domain with
finitely many "holes" is conformally equivalent to the unit disc with finitely
many closed circular arcs, coming from circles centered at the origin, re
moved. [Here a "hole" in the present context means a bounded, connected
component of the complement of the domain in C, a concept which coincides
with the intuitive idea of a hole.] An alternative equivalent statement is that
any bounded planar domain with finitely many holes is conformally equiv
alent to the plane with finitely many closed line segments, perpendicular to
the real axis, deleted. The analogous result for domains with infinitely many
holes is true when the number of holes is countable. When the number of
holes is uncountable, the situation is far more complicated and indeed it is
unclear what ty pes of "model domains" would be appropriate. (This whole

Exercises

243

line of thought then comes to involve delicate considerations in axiomatic


set theory.)

Exercises

1. Use Liouville's theorem for holomorphic functions to prove Liouville's

theorem for harmonic functions: If u is harmonic and bounded on all of


C, then u is identically constant (cf. Exercise 19).
2. Let U=

{z

C: 1

<

IzI

<

2}.

Define

u(x + iy) = log(x2 + y 2).


Check that u is harmonic on U . Prove that u does not have a well
defined harmonic conjugate on all of U.
3. Give an example of a nonsimply connected domain U

C and a har
monic function u on U such that u does have a well-defined harmonic
conjugate on U.

4. Let

u be a harmonic function on a domain U. Suppose that v1 is a


harmonic conjugate for u and that v2 is a harmonic conjugate for u.
How are v1 and v2 related?

5. This exercise uses differential forms (see [RUDI] for a review).

Given a differential w= Adx + Bdy, define


prove the following statements:

( a) * ( *w )
(b) If f is

(c )

*W

= -Bdx +Ady. Now

-w.

holomorphic and f = u + iv, then dv = *du. [Hint: By


definition, du= (8u/8x)dx+(8u/8y)dy, and similarly for dv. Then
the conclusion follows from the definition of the * operator and the
Cauchy -Riemann equations.]
Let U be a domain. If h : U---+ IR is a harmonic function, then there

is a harmonic function h : U---+ IR such that h+ih is holomorphic on


U (i.e., his a harmonic conjugate of h) if and only if f"Y *dh= 0 for

every piecewise C1 closed curve r in U . [Hint: Given the integral


condition, pick a point Po E U and define h(q) = f"Y *dh for a

piecewise C1 curve r from Po to q. Note that h, so defined, is


independent of the choice of 'Y Show that h + ih is holomorphic.]
*

6. Use the ideas of Exercise 5 to prove the following statement:

If h : {z : 0 < lzl < 1} ---+ IR is harmonic, then there is


one and only one real number A such that the function z

Exercises

244

h(z) - A In I zI

{z : 0 < I zI

set

[Hint:

0}

is the real part of a holomorphic function on the


1} .

<

There is a harmonic function

such that

function

hb

hu

on U

+ ihu is holomorphic on U.

on V =

{z : 0

<

lzl

O}

1,Imz <

<

{z : 0

<

lzl

such that

holomorphic on V. What happens when you endeavor to fit


together to obtain a suitable harmonic function on all of U

lzl

1 } ?]

<

* 7. Show that if

F : {z : 0

<

I zI

M(r)

for

<

<

1,

(a)

1}

h + ihb is
hu and hb

{z : 0

<

----+ C is holomorphic, then the function

2_ f 27r F(rei0) d(),


27r lo

is constant independent of

Laurent expansion about

* 8.

<

1, Im z >

<

Also, there is a harmonic

0.]

r. [Hint:

Expand

in a

Combine the results of Exercises 6 and 7 to prove that if a harmonic

: {z : 0 < lzl < 1 }


IR is bounded, then h = Re F for
{z : 0 < lzl < 1 }
C holomorphic. [Hint: If h is
0
bounded, then so is J027r h(rei ) d(), 0 < r < 1. Therefore, with
h(z) =A In lzl +Re F(z), deduce from Exercise 7 that A= O. ]
(b) Prove that if a holomorphic function F : {z : 0 < lzl < 1 } C has
Re F bounded, then F has a removable singularity at 0.
( c ) Combine ( a ) and (b) to prove the removable singularities theorem
for harmonic functions: If h : {z : 0 < I z I < 1} ----+ IR is harmonic
and bounded, then there is a harmonic function h : {
: lzl <
1}
IR such that h = h on {z : 0 < lzl < 1 } , that is, "h extends
harmonically across the point O."
Use part ( c ) of Exercise 8 to provide another proof that there is no
harmonic function on {z : 0 < lzl < 1 } with boundary value ( s ) 1 as
lzl 1 and boundary value 0 as lzl 0 ( as discussed in the text using
another method ) .
function

some

----+

----+

----+

----+

* 9.

----+

----+

10. Use Montel's theorem, together with material from this chapter, to prove
the following:

If U C is a domain, and if F =

of harmonic functions such that

lf0(z)I

::;

< oo

{!0}

for all

is a family

U, then

there is a sequence f0j that converges uniformly on compact subsets of U.

[Hint:

Reduce to the case of functions on a disc. Then choose functions

90 such that f0 +i90 is holomorphic.

The holomorphic functions

are then bounded and bounded away from

ef"' +ig"

u is continuous on D and harmonic on D and if u


vanishes on an open arc in 8D, then u
0.
If u is real-valued and harmonic on a connected open set, and if u2 is
also harmonic, then prove that u is constant.

11. TRUE or FALSE: If

12.

O.]

245

Exercises

13. If

u is

complex-valued and harmonic on a connected open set, and if u2

is harmonic, then prove that either


(Compare Exercise

14. If

u is holomorphic or u is holomorphic.

1 2.)

is harmonic and nonvanishing, and if

1/u is

harmonic, then what

can you say about u?


n

15. For each

<Pn harmonic on D and continuous on


n
<Pn ( eiO) =ei O, 0 :::; () :::; 27!'. Can the same be

E Z give a function

the closure of

D such

that

done for holomorphic functions?

16. Prove: If

is a complex-valued harmonic function, then the real and

the imaginary parts of


function

u is

u are harmonic.

Conclude that a complex-valued

harmonic if and only if u is harmonic.

17. Give two distinct harmonic functions on C that vanish on the entire real
axis. Why is this not possible for holomorphic functions?

18. Let

u : U----+ C

D(P, r) U. Verify the following two


property as direct corollaries of the version

be harmonic and

variants of the mean value

of the mean value property proved in the text.

( a)
(b)

u(P) = !r faD(P,r) u() ds(), where ds


2
8D(P, r).
u(P)= * JD(P,r) u(x, y) dxdy.

is arc length measure on

u : C ----+ IR

19. Prove that there is no nonconstant harmonic function


that

u(z) :::;

0 for all z EC

such

20. Use the open mapping principle for holomorphic functions to prove an
open mapping principle for

22.
23.
24.

fun

ons.

z)/(1 - z)].
Calculate v(z). What limiting value does v have as D(O 1) 3 z----+ 1?
Prove that if u is harmonic on a connected domain U C and if u = 0
on some small disc D(P, r) U, then u = 0 on all of U.
If H is a nonvanishing holomorphic function on an open set U C,
then prove that log IHI is harmonic on U.
Let be a partial differential operator of the form
a2
a2
a2
=a 8 2 + b 8 2 + c
axay
y
x
with a, b, c constants. Assume that commutes with rotations in the
sense that whenever J is a C2 function on C, then (!) Po=(! po)
for any rotation po(z)=ei0 z. Prove that must be a constant multiple

21. Let the function

v(z)

real-valued harmonic

be the imaginary part of exp[(

'

of the Laplacian.

25. Compute a formula analogous to the Poisson integral formula, for the
region U

= {z

Im

>

0}

formally to the unit disc.

(the upper half plane), by mapping U con

246

Exercises

26. Compute a Poisson integral formula for U

{z : lzl

1, Im z

<

>

O}

by

mapping U conformally to the disc.

27. Let

P(z, () be the Poisson kernel for the disc. If you write z rei8 and
ei1/!, then you can relate the formula for the Poisson kernel that was
=

given in the text to the new formula

P(z, ()

1 1(12
271' lz

1 1 lzl2
271' lz (12
-

Do so.
Now calculate

in

z.

zP(z, ()

for

lzl2
(12

to see that

is harmonic

{hJ} converges uniformly on


compact subsets of U , then prove that the limit function ho is harmonic.
[Hint: Use the fact that, on each closed disc in U, ho satisfies the Poisson
integral formula because each of the h j does.]

28. If

hi, h2 , ... are harmonic on

D(O, 1)

29. Prove the result of Exercise

U C and if

28 by using

Mantel's theorem together with

the fact that the limit of a sequence of holomorphic functions, uniformly

[Hint:

on compact sets, is holomorphic.


the sequence

{Uj}

converges to

u0,

If Uj + ivj are holomorphic and

then the sequence of functions

eu1+ivj

is bounded on compact sets, and hence has a convergent subsequence.


If the limit is

F,

then limj Uj

ln

IFll

u be a positive harmonic function on the unit disc and suppose that


u(O) =a. How large can u(3/4) be? How small can it be? What is the

30. Let

best possible bound? What function realizes that bound?

31. Let K be a compact subset of the unit disc. If

is a point of K and if

is a positive harmonic function on the disc, then let

that there is a constant

C>0

such that for all

u( a) . Prove

K it holds that

a/C u(z) Ca.


(Note that

32. If

will depend on a and K, but it should

not depend

on

u.)

is a continuous real-valued function on the closure of the unit disc

which is harmonic on the interior of the disc, then prove that

is the

real part of the holomorphic function

h(z)
33. Let

1
271'

12 11" eei8
0

e1

+
-

z i
u(e 8) dO.
z

be a continuous function on an open set U C and let

Suppose that
on all of U.

u is harmonic on

U\

{P}.

Prove that

u is in

U.

fact harmonic

247

Exercises

C be a continuous function which is C2 in the second


variable and which satisfies tlzf(t, z)
0 for every t and every z. Define

34. Let

f: [O, 1]

u(z)
Prove that

fo1 f(t, z)dt.

is harmonic on C.

35. It is natural to wonder whether or not harmonic functions can be char


acterized by a mean value property over squares or triangles. Explain
why the circle is the only curve for which T heorem 7.2.5 could hold.
36. Recall that the function

z/lzl2

is the natural notion of reflection in

directly a

the unit circle. Given this fact, formulate and prove

version of

the Schwarz reflection principle for reflection in the unit circle. Do not
use any of the results of Section 7.5.
37. Classify all conformal self-maps of an annulus A=
38. Let

h(z)

lzl

<

<

R}.

be holomorphic on a neighborhood of the closed unit disc.

Assume further that

{z: r

lh(z)I

when

z lies in the unit circle.

Prove that

is a rational function.

h is entire, the restriction of h to the real axis is real, and the restriction
of h to the imaginary axis is imaginary, then prove that h(-z)
-h(z)
for all z.

39. If

40. Is there a version of Harnack's principle for a

decreasing

sequence of

harmonic functions? If so, formulate and prove it. If not, give a coun
terexample.
41. Prove that if

tlf

0 on U.

C2
[Hint:
is

on an open set U and

6.f
P and see

To show that

variable Taylor expansion of

at

0 at

is subharmonic, then

P,

write out the two

what conditions the Taylor

coefficients must satisfy in order for Lemma 7.4.4 to hold for all small
circles around
42. Let

fj

P.]

(See Exercise 69 for the converse statement.)

be subharmonic functions on an open set U. Suppose that

converges uniformly on compact sets to a function

on U. Is

{!j}

neces

sarily subharmonic?
43. Formulate and prove a maximum principle for subharmonic functions
(see Proposition 7.7.7).
44. If

is harmonic on U c:;::; C, then

satisfies a

minimum principle

as

well as a maximum principle (Corollary 7.2.2). However, subharmonic


functions do not satisfy a minimum principle. Illustrate this claim.
45. If U is a domain, the boundary of which is the disjoint union of finitely
many simple closed

C1

curves, each with the property that the domain

lies only on one side of the curve at each point, then prove that every
point of {)U has a barrier.

[Hint:

T he question is local, i.e., one need

2 8
4

Exercises

only look at a neighborhood of a given boundary point.

Notice that

there is a line segment emanating from the boundary point which is


exterior to the domain.]

46. Let

be a real-valued

harmonic

function on U C and let :R

--+

be a convex function (do not assume that is nondecreasing!). Prove


that o

is subharmonic.

l: U

47. Let U C be a connected open set. Let


Suppose further that V is open and F

lo F

that

48. Let

--+

--+

R be subharmonic.

U is holomorphic. Prove

is subharmonic. What happens if F is only harmonic?

l: U

--+

F : D(O, 1)

--+

R be a function.
U it holds that

Suppose that for every holomorphic

lo F

is subharmonic. Prove that

is

subharmonic.

49. Give an example of a continuous


monic but

50. If

l is

l is

on

D(O, 1)

such that

12

is subhar

not.

lllP is
that lllP

holomorphic on an open set U and p > 0, then prove that

subharmonic. Now suppose that

l is

merely harmonic. Prove

is subharmonic when p 1 but that in general it fails to be subharmonic


for p < 1.

51. Give an example of a domain U C, U

point

-f- C,

such that

no boundary

of U has a barrier. But prove that, for any bounded domain U,

there is a point in au that has a barrier.

52. Let U be a bounded, holomorphically simply connected domain with the


property that each P E au has a barrier. Let be a positive, continuous
function on au. Prove that there is a holomorphic function
that

Ill

is continuous on u and

Ill

l on U such
[Hint: Use

restricted to au equals.

the logarithm function.]

53. The Perron method for solving the Dirichlet problem is highly noncon
structive. In practice, the Dirichlet problem on a domain is often solved
by conformal mapping. Solve each of the following Dirichlet problems
by using a conformal mapping to transform the problem to one on the
disc.

( a) 0

is the first quadrant. The boundary function equals 0 on the

positive real axis and y on the positive imaginary axis.

(b) 0

is the upper half of the unit disc. The boundary function is


i0
(e )=0 ,

(c )

001!",

l-x
(x)=1!"- -, -1x1 .
2
0 is the strip {z : 0 < Rez < 1}. The boundary function is iden
tically 0 on the imaginary axis and is identically 1 on the line
{z:Rez=l}.

Exercises

54.

249

If U
{z EC: lm z > O},P E U, and
U \ D (P, r ) . For which P, P, r,T is it the
equivalent to UP r?
=

<

case

ImP, define UP,r


that UP,r is conformally

<

U1

55. Let

D (O, 1 ) \D (3/4, r ) and U2


< r < 1/4 and 0 < s

D (O, 1 ) \D (-1/2, s). Of course we


1/2. Give necessary and sufficient
biholomorphic to U2. Give explicit

assume that 0

<

conditions on r and s so that U1 is


formulas for all possible biholomorphisms.

F on U {z:
z E u tends to au. [Hint:
You will not necessarily be able to write F explicitly. Note that F
should vanish somewhere since otherwise log IFI would be
0 by the

56. Prove that there is a nonconstant holomorphic function

l zl

<

<

2}

such that

IF(z)I

---+

whenever

maximum-minimum principle for harmonic functions. Think about find


ing log

F.

IFI

and using the

operation of Exercise 5 to find an associated

Alternatively, look at the image of

f-+

z/-/2

-/2/ z.

Refer to

Exercise 59 . ]

57. Prove that the region C\ { x + iO

: -1 :::;

:::; 1} is conformally

equivalent

to an annulus.

58. Let n consist of a disc with a closed segment, interior to the disc, re
moved. Prove that n is conformally equivalent to an annulus.

59. Let n

{z: 1/2

W of 0 under ?

lzl < 2}. Define ( z ) z + 1/z. What is the image


If w E W, then how many elements does the set -1 ( w )
<

have?

60. Use the Poisson integral formula to show that a harmonic function must
be

C00,

indeed real analytic.

61. Let G and H be functions that are continuous on


on

D.

and holomorphic

Prove that if Re G and Re H agree on 8D, then

imaginary constant on

F-G

is an

D.

62. Prove an analogue of the Cauchy estimates for a harmonic function.


63. Suppose that in the statement of the Schwarz reflection principle the
region

is holomorphically simply connected, that

is nonvanishing,

and that we have the limiting behavior limu3z-+(real axis) arg F (z )

for some fixed real constant c. Formulate and prove a suitable version
of the Schwarz reflection principle for this situation.
*

64. Let

F be a

for all

continuous function on C and suppose that

E C (C)

that vanish outside a compact set

will depend on ) . Prove that

is

C00

J F /':-,. dx

( the compact set


[This is Weyl's

and harmonic.

lemma.]
65. TRUE or FALSE:

If f

is subharmonic on a domain

U,

then so is

lfl.

250

Exercises

66. Prove that if f is a continuous, real-valued function on the interval

is a convex function on R, then


b
b
<P
f(x)dx
f
d
ba
b a ( o )(x) x.
[Hint: First treat the case when is linear. Then use the

a <

b,

[a,b] ,

and if

fact that
is the pointwise maximum of its supporting tangent lines.] This is a
version of Jensen's inequality for integrals.
*

U C be a bounded, connected region whose boundary consists


of finitely many simple closed C2 curves. Then it is a fact (which we

67. Let

shall not prove here, but see [KRAl]) that there is a Poisson kernel on
U. That is, there is a continuous function P(z, () on U x au such that
for any u E C(U) which is harmonic on U it holds that
u(z)

) (() ds(()
{ P(z,(u
lau

for all z

U.

Here ds is the element of arc length on au. Prove that P(z,() must be
strictly positive on U x au. [Hint: Think about barriers at boundary
points of n.]
*

68. Verify that if

n(ei9)

einO, 0 ()
<Pn(reiO)

271", n

<

Z, then

rlnl einO

is the unique harmonic extension of <Pn to D. Define

00
P(rei9)

n=-oo

<Pn(rei9).

Verify that the series converges uniformly on compact subsets of the


unit disc. Compute that the Poisson kernel

P(rei9,ei11i)

(1/27r)P(rei(0-1/I)) .

This computation is motivated by the fact that every differentiable on


aD(O, 1) can be expressed as (ei9) L:_00an einO. So the harmonic
extension of to D ought to be
=

00

<P(reiO)

n=-oo
*

69. Let

(a)

f: U

an rlnl einO.

R be a C2 function on an open set in U C.


Recall that if 6f > 0 at a point P, then f cannot have a local
maximum at P. Use this observation to deduce that if 6f > 0
everywhere on U, then f is subharmonic. [Hint: If his harmonic,
then 6(! - h) > 0 too.]

251

Exercises

{b)

If

70.

2'.:

everywhere, then, for each

everywhere. Use a limiting argument and

6.f 2'.: 0

0, 6.(f + Elzl2)
part ( a ) to deduce
>

>

that

f is subharmonic.
If f U
IR. is merely continuous, we might call f strictly subharmonic
if whenever D(P, r) U, then
if

6.f

everywhere, then

-t

f(P) <
For

C2

1
27r

r2rr
lo f(P + rei6)d0.

functions, is this equivalent to the assertion that

6.f

>

O? Does

one definition imply the other? Can you think of a definition that applies

6.f > 0 when f is C2? [Hint:


If fE C2, then 6.f > 0 if and only if 6.(f - 6lzl2) 2'.: 0 for some 6 > 0.]
Give an example of a harmonic function u on an open set U C such
that = { zE U u(z)
0} has an interior accumulation point yet u is
not identically 0. How large can be?
Suppose that f is C2 and subharmonic on the unit disc. For 0 < r < 1
to continuous functions and is equivalent to

71.

72.

define

rr
2_ f 2 f(rei6)d0.
27r lo
Prove that Mis a nondecreasing function of r. [Hint:
M(r)

Differentiate under

the integral sign and apply the divergence theorem together with the fact
from Exercise 41 that
*

6.f

2'.:

0.]

73. Complete the following outline to show that Perron's solution of the
Dirichlet problem can be used to give another proof of the Riemann
mapping theorem. F ix a simply connected domain 0 C. For simplic
ity, let us assume that n is bounded and that the boundary of n consists
of a single

C1

itively obvious"

PE
( a)
{b)
(c)

0.
Let

I'(z)

( You are authorized to take for granted any "intu


topological properties that you need here. ) F ix a point

curve.

=log lz

- Pl.

Let be the restriction of r to the boundary of 0. Let

be the

solution of the Dirichlet problem with boundary data -.


Set

u(z)

function

I'(z) + m(z). Then u is harmonic, except at P. [The


u(z), which of course depends on the choice of P , is called
=

the "Green's function" for O; the usual notation for the function

( d)
{e)

u(z) is G(z, P).]


f -OU I 8y dx + OU I ax dy is path-independent modulo
f-y is a multiple of 27r for any closed 'Y lying in 0 \ { P},

that we call

The integral

27r, that is,

the multiple depending on 'Y


The function

f(x, y) =exp [u + i(f -8u/8y dx + 8u/8x dy)]

is well

defined and homomorphic. Here the value of the integral at a point

252

Exercises

Q E n means the integral from some fixed base point P0 i- P to


Q along any curve from P0 to Q, this integral being defined up to
multiples of 271".
(f) Verify that f takes boundary limits of modulus 1 at all boundary
points of n. Conclude that f: n----+ D.
(g) Use the argument principle to check that f is one-to-one and onto.
74.

Let nI
{z E <C : 1 < lzl < ai} and n2
{z E <C: 1 < lzl < a2}.
Assume that a2 < aI and assume that : nI----+ n2 is a conformal map.
Derive a contradiction as follows.
Define 'YI
, 12
o 'YI, and in general 'Yj
'Yj-I o 'YI.
Then a subsequence of { 'Yj} must converge normally to a limit
function g.
Show that the image of g cannot contain any open set. On the other
hand, g cannot be constant. [Hint: Notice that (og)(z) g(z), but is
not the identity function. Use the fact that the integral 1) (() h1 (() d(
around the curve {(: 1(1 (1 + ai)/2} is 27ri for all j.]
=

* 75.

Let U {z E <C: r < lzl < R}. Let f11 (z) zn for n 1, 2, 3, .... For
which values of r and R is {Jn} a normal family? [Hint: Be sure to use
the notion of normal family that allows a sequence of functions to con
verge to oo, uniformly on compact sets, as well as the notion, introduced
in the text, of convergence uniformly to a finite-valued function.]

* 76.

In the heuristic topological spirit of Exercise 73, carry out the following
outline of how to prove that a bounded domain U with boundary consist
ing of two simple, closed CI curves (the domain being homeomorphic to
an annulus) is biholomorphic to a region of the form { z : 1 < lzl < R}.
( a ) There is a harmonic function u : U ----+ IR such that the limit of u
is 1 at the outer boundary and the limit of u is zero at the inner
boundary.
(b) Choose a fixed curve 'Y in U that goes once around the hole coun
terclockwise. Set
au
au
w
--dx+-dy.
ay
ax

Show that w i- 0 by applying the argument principle to e"+i .f w


"!
if w 0.
"/
( c ) Choose a constant ,X such that .Xw 271". Then show that e"+i .f w
"!
is well defined (cf. Exercise 73 for the meaning here) and that this
function is a holomorphic, one-to-one, onto mapping from U to
{ z : 1 < I z I < eA}.
=

77.

Prove that {z E
equivalent to {z

<C:
E

0 < lzl < R < +oo}, some R > 0, is conformally


0::; rI < lzl < r2 < +oo} if and only if r1

<C

253

Exercises

0. Prove that {z E C : 0 < lzl < +oo} is conformally equivalent


to {z E C : r1 < lzl < r2} if and only if r1
0 and r2
+oo.
Prove that {z E C : 0 < r3 < lzl < +oo} is conformally equivalent
to {z E C : 0 < r1 < lzl < +oo} for all choices of r1,r3 > 0 and is
conformally equivalent to {z EC: 0 < lzl < r2} for all 0 < r2 < +oo.
=

78.

Prove that Perron's solution of the Dirichlet problem (Theorem 7.8.1)


still works if one uses the following more general definition of a barrier:
If U C is an open set and P E au, then a function b : U ----+ is
called a barrier at P if (i) b is subharmonic on U, (ii) b < 0 on U, (iii)
lim supu3(-.Q b(() < 0 if Q Eau, Q-/- P, (iv) limu::i(-+P b(() 0.
( a ) Formulate the concept of the small circle sub-mean value property
analogous to the small circle mean value property of Definition
7.4.1.
(b) Use the technique of the proof of Theorem 7.4.2 to prove the fol
lowing result: If u : U ----+is a continuous function with the small
circle sub-mean value property, then u is a subharmonic function
(cf. the proof of Proposition 7.7.7).
=

79.

80.

( a ) Integrate the inequality cos x 2: 1/2 over [O,7r/3] to get 1 - cos x


x2/4 for x E [0,7r/3].
(b) Use the fact that 1- cos x is increasing to show that (1- cos x)/ x2
l/27r2 for x E [7r/3,7r].
(c ) Deduce that 1 cos x 2: x2/20 for all x E [O,7r].
-

2:
2:

Chapter 8

Infinite Series and


Products

8.1.

Basic Concepts Concerning Infinite Sums and Products

If Ji, h, ... are functions on an open set U C, then we may study the
convergence properties of
00

Lli
l
j=

from the point of view of the convergence of its partial sums. In particular,
the normal convergence concept for sequences (Definition 6.5.1) is to be
applied to series in the following way: The series converges normally if its
sequence of partial sums

SN(z)

L fj(z), N

1, 2, ...

j=l

converges normally in U. In case the functions fj are holomorphic, then the


function
00

f(z)

Ll fj(z)
j=

will then be holomorphic because it is the normal limit of the


which is holomorphic).

SN'S (each of
-

255

8. Infinite Series and Products

256

There is a corresponding Cauchy condition for normal convergence of


a series: The series
00

I: fj(z)
j=l
is said to be

compact sets if for each compact K U

uniformly Cauchy on

and each E > 0 there is an N > 0 such that for all L 2: M 2: N it holds that
L

fj(z)
jL
=M

< E

for all

K.

[Notice that this is just a reformulation of the Cauchy condition of Definition

3.2.8 for the sequence of partial sums

SN(z).]

It is easy to see that a series

that is uniformly Cauchy converges normally to its limit function.


Now we turn to products. One of the principal activities in complex
analysis is to construct holomorphic or meromorphic functions with certain
prescribed behavior.

For some problems of this type, it turns out that

infinite products are more useful than infinite sums. If, for instance, we
want to construct a function which will vanish on a certain infinite set
then we could hope to find a function
then multiply the

fj 's

fj

which vanishes at

{aj},

aj for each j and

together. This process requires that we make sense

of the notion of "infinite product."


We begin with infinite products of complex numbers and then adapt
the ideas to infinite products of functions. For reasons which will become
apparent momentarily, it is convenient to write products in the form

where

aj E C

The symbol

TI

stands for multiplication. We want to define

( * ) to converge.
partial products PN of ( * ) to be

what it means for a product such as


F irst define the

PN

IT (1 +aj)
j=l

(1 +ai) (1 +a2)

..

(1 +aN ) .

We might be tempted to say that the infinite product


00

IT(l+aj)
j=l
converges if the sequence of partial products

{ PN} converges.

nical reasons, a different definition is more useful.

But, for tech

8.1. Basic Concepts

Definition

8.1.1.

257

An infinite product

00

II (1+aj)
j=l
is said to converge if

(1)
(2)

only a finite number aj1, ..., aJk of the aj 's are equal to
if No

>

0 is so large that aj #

-1 for j > No, then

-1;

N
lim
N-++oo

II (1+aj)
j=No+l

exists and is nonzero.


If

f11 (1+aj) converges, then we define its value to be

[fi ]
j=l

(1+aj)

oo

ft (1+

aj).

No+l

Remark: If

f11 (1+aj) converges, then limM,N-+oo IT (1+aj) exists and


equals 1 (Exercise 1). This property would fail if we allowed products with
N
lim
N-++oo

II (1+aj)

0.

j=No+l

Notice also that the value of a convergent infinite product is inde

pendent of the choice of No, subject to the definition (exercise for the

f1(1 + aj) converges, then limN-++oo IJf (1 + aj) exists and


equals the value of IJ(l +aj). But the converse is not true: The existence
of limN-++oo rrf (1+aj) does not imply the convergence of IT(l+aj) even
when no aj equals -1. A counterexample is given by the case when all aj
equal -1/2.
Lemma 8.1.2. If 0 x < 1, then
1+x ex 1+2x.
(8.1.2.1)
reader).

If

Proof. First note that


+oo

1 oo 1
I:< I: 2j l

j=2 J

j=2

Then, using the Taylor expansion for

00

1+x I: '

j=O J.

ex,

i.

(00 )
we have

1+x+x I: --:-r1
j=2 J.

1+2x.

258

8. Infinite Series and Products

Corollary 8.1.3. If

a3 E C, la3I < 1,

then the partial product

PN for

00

satisfies

(8.1.2.1), we have 1+ia31 elail; hence


PN=(1+la1I) (1+laNI) exp(la1 I ++laNi).
By the second part of Eq. (8.1.2.1),
1+la3I=1+2(21 la3I) e
Proof. By the first part of Eq.

Therefore

Corollary 8.1.4. If

then

00

IT(1+la3I)

j=l

converges.
Proof. Let

00

Lla3l=M.

Then, by Corollary
Since

8.1.3,

j =l

PN expM.

P1 P2 ,the sequence of partial products converges to a nonzero


D

limit.

Corollary 8.1.5. If
00

IT(l +la3I)

j=l
converges, then

converges.

8.1. Basic Concepts

259

Proof. By Corollary

8.1.3,
PN

exp

(l;I)

So the convergence and hence the boundedness of the monotone sequence

{ PN}

implies the boundedness and hence the convergence of the monotone

sequence

u:.:

la3I}.

Now, because of our experience with series, we expect that "absolute


convergence implies convergence." For series this assertion follows trivially
from the Cauchy condition and the inequality

I: O'.j I:
j=M

j=M

1a31.

The analogous assertion for products is somewhat less obvious and requires
the following technical lemma.

Lemma 8.1.6. Let

PN

a3

EC. Set

II (1 +

j=l

a3) ,

PN

II (1 +

la31).

j=l

Remark: By renumbering, we could also prove the following variant:

M
-1 + II (1 +
j=N+l

a3)

M
-1 + II (1 +
j=N+l

la31).

This observation will be useful below.

Proof of Lemma 8.1.6. The desired estimate follows from expanding

PN

1 +monomial

terms consisting of products of the

a3

's

and noting that

PN

After subtracting

1 +absolute
1

values of the same monomials.

in both cases, the desired inequality becomes just the

statement that the absolute value of a sum does not exceed the sum of the
absolute values.

260

8. Infinite Series and Products

Theorem 8.1. 7. If the infinite product


00

converges, then so does

By Corollary 8.1.5, 2::::}=1 lajIconverges if f1}=1 (1 +lajI) does. There


fore limj-++oo lajl 0. In particular, for some No, it holds that aj i- -1 if
Proof.

>No.

Write, for

J >No,

QJ
If M

>N >N0,

II
j=No+l
then

(1 + aj) and QJ

J
=

II
j=No+l

(1 + lajl).

II (1 + aj) - 1
j=N+l
M
IQNI II (1 + lajl) - 1
j=N+l
by the remark following Lemma 8.1.6. This last is
M
NI
IQ
II (1 + lajl) - 1 IQ M - QNI
j=N+l
IQNI

Thus convergence of the sequence { QN} implies the convergence of


the sequence {QN}. Now, by the remark just after Definition 8.1.1, choose
M >No+ 1 so large that -1 + f1Z.(1 + lajl) <for all N > M. Then, for
all N > M, by Lemma 8.1.6, 1- 1 +f1Z.(1 + aj)I <so If1Z.(1 + aj)I >
Hence
M-1
-+oo II (1 + aj)
Nlim
No+I
l M-1
>
(1 + aj) > o .
2 II
No+l
D
So f1(1 + aj) converges, as desired.

8.1. Basic Concepts

Corollary

261

8.1.8. If

then

00

converges.

Remark:

Corollary

products.

It is so important that it is worth restating in a standard al

8.1.8

is our standard convergence result for infinite

ternative form: If

00

L:ll-ajl<oo,
j=l
then

00

II aj
j=l
converges. The proof is just a change of notation.
We now apply these considerations to infinite products of holomorphic
functions.

Theorem

8.1.9. Let U C C be open .

Suppo se that

fj

: U

--+

C are

holomorphic an d that
00

L:l!JI
j=l

(8.1.9.1)

converges un iformly on compact sets. Then the sequence of partial products


N

FN(z)

II (1 + fj(z))
j=l

(8.1.9.2)

converges un iformly on compact sets. In particular, the limit of these partial


products defin es a holomorphic fun ction F on U.
The fun ction van ishes at a point zo

U if an d on ly if

fj (zo)

-1 for

some j. The multiplicity of the zero at zo is the sum of the multiplicities of


the zeros of the fun ction s 1 +

Remark:

at zo.

For convenience, one says that the product

verges uniformly on

(a)

fj

a set E if

it converges for each z in E

IJ(l + fj(z))

con

262

8. Infinite Series and Products

and

(b)

the sequence

fj(z)).

{IJ (1+ fj(z))} converges uniformly on E to IJr'(l+

Then the theorem can be summarized as follows:

2::1 lfjI converges uniformly on compact sets, then the


product IJr'(l + fj(z)) converges uniformly on compact
If

sets.

Proof of Theorem 8.1.9. First we consider the uniform convergence. Fix


a compact subset K U. Notice that, by the uniform convergence of

Lj lfjI

on K, the partial sums of the series are uniformly bounded on K by some


constant C. Therefore by Corollary

8.1.3 the partial products FN of

00

II (1 + l!JI)
j=l
are uniformly bounded on K by ec.
Let 0 < <

l. Choose L so large that if M 2:: N 2:: L, then


M

L lfj(z)I <

z EK.

for all

j=N

If M 2:: N 2:: L, then, by Lemma

IFM(z) - FN(z)I

8.1.6 and Corollary 8.1.3,

M
II
(1 + lfj(z)I) - 1
z
(
)I
IFN
j=N+l
N

<

II (1 + lfj(z)I)
j=l

<

Since e

-1

--+

ec

(e

( [. t ] )
3=N+I

1) .

o+, it follows that the sequence

Suppose that

F(zo)

zo. By
Jo such that

0 for some

of infinite products, there is a

lfj(z)I -1

exp

{FN(Z)} is uniformly Cauchy.


definition of the convergence

lim

N-++oo
is nonvanishing at

zo.

II (1 + fj(z))
io+l

By the part of the theorem that has already b een

proved, this limit is holomorphic. In particular, it is continuous and hence

8.2. The Weierstrass Factorization Theorem

263

zo.

is nonvanishing in some neighborhood V of

Now

oo

rr (1 + fj(z))= II (1+11(z)) x N-+lim+oo II (1+11(z)).


1

jo+l

Since the second factor is holomorphic and nonvanishing on V, the state


ments about the zeros of

nri( l + fj(z))

and their multiplicities follows by

inspection of the first factor.

8.2. The Weierstrass Factorization Theorem


One of the most significant facts about a polynomial function

p(z)

of

is that it can be factored:


k

p(z)=c

IT (z-a1)
j =l

Among other things, such a factorization facilitates the study of the zeros
of

p. In

this section we shall show that in fact any function holomorphic

on all of

can be factored in such a way that each multiplicative factor

possesses precisely one zero


on all of

C ( called

an

( of

first order ) .

entire function)

Since a function holomorphic

can have infinitely many zeros, the

factorization must be an infinite product in at least some cases. So we have


to worry about convergence. Our multiplicative factors will usually have to

(z-a1) The
Weierstrass elementary factors.

be something a bit more sophisticated than


we use are called the

basic factors that

To obtain these factors, we define

Eo(z)=1-z
and for

1p

Z we let

( :

Ep(z)= (1- z) exp z+


Of course each

Ep

Ep

is close to

+ ..+

is holomorphic on all of

hinges on the following technical lemma.


sense,

if

I zI

C.

;)

The factorization theory

This lemma says that, in some

is small. This assertion is not surprising since

( :
z+

+ ...+

;)

is the initial part of the power series of - log( l -z). Thus

(1-z)exp z+
might b e expected t o b e close t o

fo r

+ ..+ ;)
2

small ( and

large ) .

264

8. Infinite Series and Products

Lemma 8.2.1. If lzl 1, then

ll - Ep(z)I lzlP+1.
Proof. The case p

0 is trivial so we assume p 1. Write


00

Ep(z)
We claim that

b1

b2

+L

n=l

bp

bnzn.
bn 0

0 and

for

>

p. Assume the

claim for the moment.


Then
00

Ep(l)

+L

n=p+l

bn

so that
00

I: lbnl
n=p+l

i.

lzl 1,

Now we can estimate, for

00

lzlp+l

n=p+l
lzlp+l

<

bnzn-p-1

00

n=p+l
p+l
lzl .

lbnl

This completes the proof, except that we must verify the claim.
Notice that, by direct calculation,

E;(z)

(z +

-zexp

+ )

By the power series expansion for exp, all

-z'

Gn

<>nZn

(I+ )

are positive. This last expres

sion equals
00

-zP

(-an-p)zn.
n=p+l

+L

Comparing this with


00

E(z)

L nbnzn-l,

n=l

Ep(O)

1,

8.2. The Weierstrass Factorization Theorem

265

yields the claim.


Theorem 8.2.2. Let

{ aj }f=1

be a sequence of nonzero complex numbers

with no accumulation point in the complex plane (note, however, that the

aj 's

need not be distinct). If

for every

>

{Pj}

are positive integers that satisfy

C: r

+l

<

(8.2.2.1)

oo

"

0, then the infinite product

converges uniformly on compact subsets of C to an entire function F. The


zeros of F are precisely the points

{ aj }, counted

with multiplicity.

given r > 0, there is an N > 0 such that if n N, then


lanl > r (otherwise the an's would have an accumulation point in D(O, r)).
Thus, for all n N and z E D(O,r), Lemma 8.2.l tells us that
Pn+l
Pn+l

l
P
n
Proof. For a

Thus

IE n (: ) l l : J

ILI

11E,. ( : J 1
-1

<

oo

Uniform convergence of the series on D(O, r) follows from the Weierstrass


M-test. Thus the infinite product

converges uniformly on D(O,r) by Theorem 8.1.9. Since r was arbitrary, the


infinite product defines an entire function f.
The assertion in the theorem about the zeros off follows immediately
D
from Theorem 8.1.9.
Corollary 8.2.3. Let
accumulation point.
precisely equal to

{ an}=
l

be any sequence in the plane with no finite

Then there exists an entire function

{an}=

with zero set

(counting multiplicities).

We may assume that a1, ... , am are zero and all the others nonzero.
Let r > 0 be fixed. There is an N > m such that, when n N, then
lanl > 2r. But then
Proof.

8. Infinite Series and Products

266

so the hypotheses of the theorem are satisfied with

Pn

1. Thus

Zm n=IT+l En-1 ( a: )
m

is the entire function that we want.

Theorem 8.2.4

( Weierstrass factorization theorem ) . Let f be an entire


function. Suppose that f vanishes to order m at 0, m 0. Let {an} be the
other zeros of f, listed with multiplicities. Then there is an entire function

g such that

f(z)

zm

el>l

.!! En-I ().

Proof. By the corollary, the function

h(z)
f,

has the same zeros as


nonvanishing.

Since

holomorphic logarithm

Zm n=l
IT En-1 (:n)

counting multipicities.

e9

f /h

Thus

( holomorphically ) simply
g on C ( Lemma 6.6.4). That
is

is entire and

connected,

f /h

has a

is,

h
D

as desired.

8.3. The Theorems of Weierstrass and Mittag-Leffler:


Interpolation Problems
In the previous section, we were able to construct holomorphic functions

C ( Corollary 8.2.3). By modifying the ideas there,


any open set U C. That is one of the main things that

with prescribed zeros in


we can do this on

we do in this section.
The only necessary condition that we know for a set
zero set of a function

f holomorphic on

U is that

{aj}

{aj}

U to be the

have no accumulation

point in U. It is remarkable that this condition is also sufficient: That is the


content of Weierstrass's theorem.

Theorem 8.3.1

( Weierstrass ) .

Let U

be any open set. Let

a1, a2, ...

be a finite or infinite sequence in U (possibly with repetitions) which has no


accumulation point

in

U. Then there exists a holomorphic function

whose zero set is precisely

{aj}

Proof. First observe that if

{aj}

is infinite.

on U

(counting multiplicities).

{aj}

is finite, then the required holomorphic

function is given by the polynomial


that

p(z)

f1j(z - aj)

So we may assume

8.3. Weierstrass and Mittag-Leffler Theorems

267

It simplifies the proof if we think of U


linear fractional transformation

1/(z

- p)

C U {oo} and apply a


p E U that is not in

for some

the set where the function we are constructing is supposed to vanish. Then

oo E U

and all the boundary points of U are in the finite part of the plane.

Thus the new setup is as follows:

(1) u cc;
1'

(2) C\ U is compact;
(3) {aj}1 U {oo} U;
(4) {oo} n {aj} 0.
=

By hypothesis, the accumulation points of

{ai}

are all in 8U . Hence any

compact subset of U contains only finitely many of the


By property
point

ai EC\ U;

(2)

aj has a(not
lai -ail, then

above, each

if we set

dj

as

aj 's.

necessarily unique) nearest

j -+ 00.

Now let K be a compact subset of U. Then the distance of K to C \ U is


positive: Say that

lz -wl
In particular,

all

8 > 0,

lz -ail

8 for all

z E K,

z EK

w EC\ U.

all

and all j. Thus by(*) we have, for

all j exceeding some Jo, that

dJ < lz -aI
J
2

In other words,

aj
I ajz -aj
- I ---

<

or a11

z E K, J.

.
> Jo

We now apply Theorem 8.1.9 and Lemma 8.2.1 with


by

(ai -aj)/(z - ai)

j and

replaced

Thus

f(z)

- j
)
IT Ej ( ajz -aj

j=l

converges uniformly on K. Since K U was arbitrary, the function


holomorphic on U and has the desired properties.
happens at the point

oo.]

[Exercise:

is

Check what

We next want to formulate a result about maximal domains of existence


of holomorphic functions.
subsets of the plane.

But first we need a geometric fact about open

8. Infinite Series and Products

268

F igure

Lemma

8.3.2.
A

infinite set

8.1

Let UC C be any open set. Then there exists a countably

!-

U such that

(1) A has no accumulation point in U;


(2) every PE 8U is an accumulation point

of A.

Remark:

You considered this construction earlier in Exercise

Proof of Lemma

8.3.2.

wEC\U
inf

/z

- wl

is well defined, positive, and finite-valued on U. For j

For each j let


of the form

Qj

of Chapter 4.

Since C \ U is nonempty, the function

d(z)

Ai

68

{z E

U: 4i+I

<

1, 2, . .. ,

let

1
d(z) 4i }.

be the collection of open boxes in the plane with vertices

1
1
1
:i
, :i )
,
)
,
,

;
i
i
),
)
( , (
(
(

for all k, l E Z (see F igure

1
,

8.1).

For each fixed j, the collection of such boxes is pairwise disjoint. The

union of their closures is C. Enumerate the open boxes in

Qj

{%}1

Qj

and write

8.3.

26 9

Weierstrass and Mittag-Leffler Theorems

Now, for each

1, 2, 3, ... and each

with closure intersecting A ,

the center of lies in U . This follows from the fact that some point of the

C \ U 1/41+1 while the distance of the center


1/8j. Moreover, the distance of such a center point
to C \ U is not more than 1/4j-l by a similar argument: A point of the
closure of is within 1/41 of C \ U and again the center is within 1/8j of

closure of has distance to

to any point of

is <

that point.
Let

PJ

%,

be the center of

paragraph. Let

D1

hence so is the set

for each as described in the previous

be the set of these

PJ.

Clearly

Dj

is countable and

00

We claim that A satisfies properties

For property

(1),

note that if

(1) and (2) of Lemma 8.3.2.


satisfy la - a'I < 1/161, then

a, a' EA

a,a' E

LJ Dj.

j?_jo
'
So d(a) < 1/41-1 and d(a ) < 1/4jo-l.
For property (2), let p Eau and E > 0, with E < 1/4 for convenience.
The disc D(P, E) contains a point z E U, by definition of boundary point,
and d(z) < E < 1/4. Choose j 1 such that
1
1
4)+1

Then

z E A1. There

P.

d(z) 41.

is a the closure of which contains

this square belongs to

of

<

D.

This point of

D .is

within

z.

So the center of

E + 1/8j < E + d(z)

<

2E
D

The following corollary is particularly striking. It implies that every

open U C

1-

is the "domain of existence" of a holomorphic function. We

need a piece of terminology: If

p is said to be

D(P, r)

f is holomorphic on U

JID(P,r)nU

I-

C and P

Eau, then

D(P, r) and a holomorphic f on


JID(P,r)nu In effect, the Jon D(P, r) is an

regular for f if there is a disc

such that

"analytic continuation" of the function

analytic continuation in Chapter

10.]

on U. [We shall deal in detail with

Now we have the following corollary:

Let U C be a connected, open subset with U f- C.


There is a function f holomorphic on U such that no P Eau is regular for
f.
Corollary 8.3.3.

8. Infi.nite Series and Products

270

Proof. Let A U satisfy the conclusions of the lemma. By property


of the lemma we may apply the Weierstrass theorem to A
obtain a holomorphic

f on

{ aj}

(2)

on U to

U whose zero set is precisely A.

PE oU were regular, then there would be a disc D(P, r ) and


holomorphic j on D(P, r ) such that D(P,r)nu
flv(P,r)nu But the zeros
<?.!" J accumulate at P so the zeros of f would also accumulate at P, forcing
f = 0. Therefore J = 0, a contradiction.
0
If some

Another important corollary of Weierstrass's theorem is that, for any


open U, the field generated by the ring of holomorphic functions on U is the
field of all meromorphic functions on U. In simpler language:

Corollary 8.3.4. Let U C be open. Let m be meromorphic on U. Then

there are holomorphic functions

J, g

m( z )

on U such that

J( z)
.
g ( z)

Proof. Let ai, a2, . . . be the poles of m, listed with multiplicity.

By the

Weierstrass theorem, there is a holomorphic g on U with zero set precisely

{ aj}. Then Riemann's removable singularities theorem say s that the


function f = m g is holomorphic on U. Then
the set

E U and r > 0 is
D( z, r ) is essentially a

In Corollary 8.3.4 it should be noticed that if z


small enough, then the statement that m

f/g

on

tautology. W hat is of interest is that a quotient representation can be found


globally on U.
Since it is possible to prescribe zeros of a holomorphic function on any
open U, then of course we can prescribe poles-since
where

J has

1/ f has poles exactly

zeros. But we can do better: W ith a little extra work we can

prescribe the negative power portion of the Laurent series on any discrete
subset of U. For the proof of this result, we need a lemma on "pole-pushing"
which is useful in many other contexts as well.

Lemma 8.3.5 (Pole-pushing lemma). Let


-1

A( z )

a, (3E C.

L aj ( z - a)j

j=-M

Denne

8.3.

Weierstrass and Mittag-Leffler Theorems

for some M 2: l. Fix


Laurent expansion

number

In - .Bl- Let f.

r >

271

>

0.

Then there is

finite

B(z)

L bj(z - ,B)i,

some N 2: 1, K 2: -1, such that


IA(z) - B(z)I <

j=-N
f.

for all

z EC\ D(,B, r ) .

Proof. Consider the Laurent expansion for A


that converges uniformly to A

expar;_ded about the point f3


on any set of the form C\ D (,B, r ) for r > Ia -

.Bl. Now take B to be a sufficiently large partial sum of this expansion.

It is instructive to notice that this lemma can be proved in a more


elementary fashion by observing that

(z - a)-1

(z

- ,B)

-1 .

- (a - ,B)(z - ,B)-1

1
00

(z - m-1 I: ((a - .B)(z - m-1)

j=O

{z: lz-/31 > ln-,BI}. This


- a)-1 may be substituted into(*) to obtain an expansion

which converges uniformly on compact subsets of


expansion for(z
about

(3.

We now formulate the basic result on prescribing pole behavior, known


as the Mittag-Leffler theorem, in two different (but equivalent) ways: one
qualitative and the other quantitative. We leave it as an exercise for you to
check the equivalence of the two versions; we shall prove only the second.

Theorem 8.3.6 (Mittag-Leffler theorem, first version). Let U C be any


open set. Let a1, a2,. . . be a finite or countably infinite set of distinct
elements of U with no accumulation point in U. Suppose, for each j, that Uj
is a neighborhood of <lj and that <lj fl. Uk if k-::/= j. Further assume, for each
j, that mj is a meromorphic function defined on Uj with a pole at <lj and
no other poles. Then there exists a meromorphic m on U such that m - mj
is holomorphic on Uj for every j and which has no poles other than those
at the <lj.
Theorem 8.3.6a (Mittag-Leffler theorem, second version). Let U C be
any open set. Let a1, a2, . . . be a finite or countable set of distinct elements
of U, having no accumulation point in U. Let SJ be a sequence of Laurent
polynomials (or "principal parts"),

-1

si(z)

(=-p(j)

a (z - aj)e .

272

8. Infinite Series and Products

Then there is a meromorphic function on U whose principal part at each a1


is

Sj

and which has no other poles.

Proof (of the second version). If there are only finitely many a/s, then
the sum of the Sj gives a suitable meromorphic function. So we shall assume
that there are infinitely many a1 's. As in the proof of Werstrass's theorem,
we shall assume without loss of generality that U c C and that oo E U
"I

(Exercise 23). For each j let aj E C \ U be a (not necessarily unique)


nearest point to O'.j. We set d1 lii1 - a1 I
=

For each j, apply the pole-pushing lemma to find a polynomial t1( z ) in


negative powers 0f( z - aj) such that

We claim that
00

l:(s1(z) - t1( z ))
j=l
is the meromorphic function we want. All that needs to be checked is uniform
convergence on compact subsets of U \ { a1}.
Just as in the proof of Weierstrass's theorem, we know that d1
0.
Fix a closed disc
r) U \ { 1}. Choose J so large that j J implies
----?

D(a,

2d1 <dist

(n(a,r),C\u).

Da

Then, of course, ( , r) U \ D(a1,2d1) for j J, so(*) applies for all


z E
r). Thus

D(a,

ls1(z) - t1( z )I <Tj for

z E

D(a,

r .

The Weierstrass M-test now gives uniform convergence of (**) on

D(a, r ) .

D(a,

Since
r ) U \ { 1} was arbitrary, the asserted uniform conver
gence is proved. [Notice that, in the infinite sum ]s1 - t1), the terms s1
contribute the poles and the terms tj force convergence.]
D
The theorems of Weierstrass and Mittag-Leffler can be combined to
allow specification of a finite part of the Laurent series at a discrete set of
points. We first need an algebraic result, which is in effect a consequence of
Laurent series expansions(here, of e0 + ... divided by g) .

Lemma

8.3. 7. Let U C be any open set. Let

numbers

eo, ... , ep

be given. Suppose that

U and let complex

is holomorphic on U and that

8.3. Weierstrass and Mittag-Leffler Theorems

273

g has a zero of order p + 1 at o:. Then there is a Laurent polynomial


v(z)

-1

j=-p-1

b_j(z - o:)i

such that
g(z)

v(z)

eo + e1(z - o:) +

+ ep(z - o:)P

+ higher order terms.

Proof. Assume without loss of generality that o:


g(z)

(t)
=

0. Then

+l
+2
+ ...
+ Cp+2zP
Cp+1zP

for small values of z and


v(z)

b1z

-1

+ b2z

-2

1
+ bp+1z -P- .

Our job is to solve for the b/s, subject to the condition (t). But
g(z)

v(z)

(cp+1bp+1) + (Cp+2bp+l + Cp+1bp)z


+(cp+3bp+l + Cp+2bp + Cp+1bp-1)z

+ ... + (c2p+1bp+l + C2pbp + ... + Cp+1b1)zP + ....

Clearly the equations


Cp+1bp+l

eo,

Cp+2bp+l + Cp+lbp
Cp+3bp+l + Cp+2bp + Cp+lbp-1

C2p+lbp+l + C2pbp +

+ Cp+lbl

may be solved in succession for bp+l bp, ... , b1.

Theorem 8.3.8. Let U C be an open set and let 0:1, 0:2, . . . be a finite or
countably infinite set of distinct points of U having no interior accumulation
point. For each j let there be given

an

N(j)
sj(z)

expression
j
l
al (z - o:J) ,

l=-M(j)

with M(j), N(j) ;:::: O. Then there is a meromorphic function


morphic on U

on U, holo

{o:j}, such that if -M(j) N(j), then the gth Laurent

coefficient of m at O:j is

Proof. By Weierstrass's theorem there is a holomorphic function h(z) on


U with a

ero of order M(j) at O:j and no others. Let


sj(z)

h(z)

Sj(z)

274

8. Infinite Series and Products

and let

N(j)+M(j)

Sj(z)=

a(z-o:j)f

L
f=O

be the N(j) + M (j) order Taylor expansion of

Sj

about

O:j

Again, by

Weierstrass's theorem, there is a holomorphic g on U with zeros of order


N(j) + M (j) +

O:j,

at

each j, and no others.

functions

By the lemma there are

-1
Vj(Z)

L.::

f=-N(j)-M(j)-1

such that

g(z)

vj(z)

Sj (z) +(higher order terms).

(t)

k(z) on U such
But then g(z) k(z) will have

Now Mittag-Leffier's theorem gives a meromorphic function


that

k has principal part Vj (z) at O:j, each j.

no poles and by (+) will have N(j) + M (j) degree Taylor polynomial at

Sj (z),

equal to

O:j

each j. T herefore the meromorphic function

g(z) k(z)
h(z)
D

will satisfy the conclusion of the theorem.

Exercises
1. Let

{an}

C. Assume that no

an

-1.

Prove that the product

00

IT (1 +an)

n=l

converges (and the product is not equal to zero) if and only if the partial
products satisfy the product Cauchy condition: If
K

>

0 such that if N M K, then

IT (1 +an) - 1

< E.

n=M
2. Let

{aj}

\ {-1}.

Prove that if

lim

n---+oo
exists and is nonzero, then

aj

IT(l + aj)

j=l
-t

0.

E >

0, then there is a

275

Exercises

3. If

lzl

<

R,

then prove that

R2n z2n
R2n

oo

+ )
(
n=O

II
4. If

bn 1
>

for all

n,

_!!:___
R-z

then prove that

n
converges if and only if

L log

n bn
]1(1+(-l)n)

5. Determine whether

< oo.

converges. Do the same for

6. For which values of p ER does

converge? (See Exercise

4.)

7. Calculate

explicitly.

8. Prove that if

Ej laj I

< oo

and if

is any permutation of the positive

integers, then
00

II

00

(1+

j=l

aj)

=II

j=l

(1+

au(j))

9. Prove that, in general, the result of Exercise 8 fails if the hypothesis

E laj I

< oo

is replaced by a weaker condition like

E aj <

oo.

10. Let f be entire and have a first-order zero at each of the nonpositive
integers. Prove that

/(z)

z. e()

for some entire function g.

J1 [ (1+}) ]
e-/i

276

Exercises

11. (Weierstrass's sigma function) Let

a/ (3 R If f is entire and has a


jn+ k(3, j, k E Z, then show that

II
f(z)=eg(z).z
.
(J,k)#(O,O)

{(

1-

for some entire function g.

C be nonzero and such that

first-order zero at each of the points

exp
[ja : k(3+ (ja : k(3) ] }
ja : k(3 )
2

12. Prove that

00

1
1
71"2 n oo (z - n)2.

(sin 7rZ

[Hint:

a, (3

)2

The left and right sides differ by an entire function which is

bounded.]

13. For which

z does

00

converge?

14. Suppose that

L Inn - f3nl

< 00.

Then determine the largest open set of

for which

II Z - O!n
00

n=l Z - f3n
converges normally.

15. Prove that


cot 7rZ =

00

2z
1 1
- -+ L z2 -1.2
7r z j=l

Conclude that

00

sin7rz
16. The case
Wallis:

1/2 of

7rZ

II [l - n-2z2].

n=l

( * ) in Exercise

15

gives the product formula of

1
2. 4. 6 .. 2k .
{if =
v 2 k11! 1. 3. 5 ... (2k - 1) J2k+ i"

Verify this formula.

17. Use the formula for cot 7rZ in Exercise


explicitly.

18. Prove Weierstrass's theorem (Theorem


theorem (Theorem

8.3.6). [Hint:

15

to sum L

8.3.1) by
f' / f.]

Consider

1/j2

and L

1/j4

using Mittag-Leffier's

Exercises

277

19. Prove that


sin z

sin 7rZ

7r

L.,,
n=l

1)

sin n

z2 - n2

20. Suppose that 91, 92 are entire functions with no common zeros. Prove
that there are entire functions

Ji, h

such that

Ji . 91 + h . 92
[Hint:

Use Theorem 8.3.8 to choose

1.

so that

zeros of 91 to at least the same order as 91 .]

21. Let

tC

f-

be any open set. Let

{ Cj}

8U

1 - h 92

be dense in

is zero at the

8U.

Let

{ aj}

be

the sequence

Choose

{ bj}

such that

L laj - bj I
14

Exercise

< oo.

asked you to consider the convergence of

f(z)

IT

j=l

bj
.
z- a
J
z-

Show that in the present circumstances

P E 8U

is regular for

f.

is holomorphic on

and no

22. Reformulate Theorem 8.3.6 so that the last sentence reads "...such that

m/mj

is holomorphic on

Uj.''

Prove this new version of the Mittag

Leffier theorem. Be sure that your new formulation is neither trivially


true nor trivially false.

C is a meromorphic function with a pole at


f U
P E U, P -=/= 0, P -=/= z.
(a) Investigate how the Laurent series "principal part" (negative pow
ers of f expanded at P) is related to the principal part of f ( 1/ z)
at 1/ P.
(b) Use your results from part (a) to justify the claim in the proof of

23. Suppose that

Theorem 8.3.6a that we can suppose without loss of generality that


oo

24.

(a)

is in the domain of the desired meromorphic function.

Investigate the possibility of a direct proof of the Mittag-Leffler the


orem (Theorem 8.3.6) for

=a bounded domain in

poles" not to infinity but to (the closest point of)


note that

8U,

1/ ( z - P)

tC

by "pushing

8U.

(Namely,

can be expanded in a power series around

which converges uniformly on compact subsets of

Q,

U.)

Use finite partial sums to make things converge when you take an
infinite sum of "principal parts" at interior points.

278

Exercises

(b) Do the same for the Weierstrass theorem (Theorem 8.3.1).

Chapter 9

Applications of Infinite
Sums and Products

9.1. Jensen's Formula and an Introduction to Blaschke


Products
In the previous chapter, we determined that the behavior of the zero set of
a holomorphic function is essentially arbitrary: Except for the fact that the
zeros cannot accumulate at any point of the domain of the function, they
can otherwise be specified at will. The subject of the present chapter is, by
contrast, a sequence of results which in effect control the behavior of the
zeros when some hypotheses are imposed about the general behavior of the
function.

Roughly speaking, we might summarize these results as saying

that, in order for a function to have a great many zeros, it must grow fairly
fast at the boundary of its domain of definition.
One obvious instance to consider is the case of polynomial functions:
To have more zeros is to have higher degree, which in turn implies faster
growth at infinity. This last observation is really too simple to be interesting
in terms of the general theory, but the spirit of the example has far-reaching
implications.
The first situation that we shall investigate in detail concerns functions
holomorphic on the unit disc. Later in the chapter, we shall also consider
functions holomorphic on all of

(i.e., entire functions). Now we turn to

the details.
Suppose that g is a nowhere vanishing holomorphic function on a neigh
borhood of

D(O, 1).

Then log

191

is harmonic and the mean value property

279

280

9. Applications of Infinite Sums and Products

1 1271"

gives that
log lg(O)I

27r

log lg(ei0)ld0.

This formula gives quantitative information about the size of


terms of the size of

at the point

and vice versa.

on

8D

in

Our first main goal

in this section is to generalize the formula to a disc of any radius and to a


function that has zeros.
First, we need means for manipulating the zeros of a holomorphic func
tion on the disc. What we want is an analogue for the disc of the factors

(z-a) that are used when we study


Blaschke factors:

is called the
If

[The

D(O,

1),

poly nomials on C. The necessary device

then we define the Blaschke factor

Ba(z)

= 1-z -aza

-_-.

Blaschke factors should look familiar because they appeared in a dif

ferent guise as Mobius transformations in Sections 5.5 and

1).

6.2.]

function Ba is holomorphic on a neighborhood of


It has a simple zero at a and no others. It also satisfies the property
D(O,
that IBa(z)I
if z E 8D.

Proposition 9.1.1. The

Proof. If

lzl

D(O, 1/lal)

;;2

1
Ba
lzl=1
Izzi = 1
z
= =1.
ll - l l ;z l 1;=;1

< 1/lal, then


- az -=/= 0 so
D(O,
The assertion about the

Finally, if

1).
lzl 1,
=

then

IBa(z)I

Theorem 9.1.2

( Jensen's

and

formula ) . Let

hood of D(O, r) and suppose that


in

D(O, r),

that

is holomorphic on

zeros of Bis obvious.


so that

f(O) -=/= 0.

be holomorphic on a neighbor

Let

a1, ..., ak

be the zeros of f

counted according to their multiplicities. Then

I :j I 217r Jor27r
k

log lf(O)I +

log

Proof. Suppose first that


function

I Uj I

<

for all

is holomorphic on a neighborhood of
order

at

Uj

log lf(rei0)ld0.

and no other zeros; also

the function

g(z)

j.

Notice that for

j/

1, ...

=1 lzl= r.

D(O, r). The


IBa /r(z/r)I
j

= Tik=l Bfa(z)r(z/r)

, k the

function has a zero of


if

Therefore

281

9.1. Jensen's Formula and Blaschke Products

D(O, r)

is holomorphic on a neighborhood of

(* )

Thus

applies to g and we have


k

log lf(O)I

1
""" log
IBa J ;r(O)I =
271"
j=l

1211"
0

and has no zeros in

D(O, r).

log lf(rei6)ld0

(9.1.2.1)
But, for each

j,

and
log IBaj/r(rei6/r)I =log 1 =0.
So Eq.

(9.1.2.1)

becomes

log If (0) I +
The general case, when

log

J= l

larl

I I
a!_1

may be

r,

f211" log If(rei6)IdO.


27r loo
follows by continuity (Exercise

Corollary 9.1.3 (Jensen's inequality). With

15).

as in the theorem,

f211" log lf(rei6)1 dO.


27r l o
Each of the terms log (r/ l aJ I) on the left of Jensen's formula is non
log lf(O)I

Proof.

negative.

We now turn to what is a recurring theme in complex function theory:


The growth rate of a holomorphic (or meromorphic) function gives informa
tion on the distribution of its zeros. For the unit disc, the result will be an
application of Jensen's formula.

Theorem 9.1.4. If

D(O, 1)

and

ai, a2, ...

is a nonconstant

bounded

holomorphic function

on

are the zeros of f (counted according to their multi

plicities), then

00

L(l - laJ I)

j=l

f (0) i= 0. Since { aj} is a countable set, we can


1 but arbitrarily near 1 such that laj I i= r for all j. We
Theorem 9.1.2 to f on D(O, r). Then we have
r
n( )
l 211" log lf(rei6)ld0.
log lf(O)I + L log !_
(9.1.4.1)
a1
.J=l
27r 0

Proof. Assume first that


find numbers
apply

< oo.

<

I I

282

9. Applications of Infinite Sums and Products

Here n ( r ) is the number of zeros inside the disc D(O, r ) . Since f is bounded,
say IJI ::; M on D(O, 1), we conclude from letting r
1- in Eq. (9.1.4.1)
that
___.

00
f;

For any real number


-log a

-log ( 1- (1

::; logM -loglf(O)I.


lajl

log

a E

(0, 1), we know that

))

(1 -a ) +

(1
2

a 2

( 1 -a ) 3 + . . . .
3

Thus
log

( )

-log a

>

a.

As a result,

is finite because

1
I: 1og

is finite.

In case J vanishes at 0 to order m, then the preceding argument applies


to the function f(z)/zm and the zeros at 0 contribute only m terms of 1 to
0
the series.
It is remarkable that the converse of the preceding theorem holds (one
might have expected that there would be additional restrictions that need
to be imposed on the aj).
Theorem 9.1.5. If {aj} D(O, 1)

(with possible repetitions) satisfies

00

2:( 1- la j l)

< oo

j=l
and no

aj

0,

then there is a bounded holomorphic function on

which has zero set consisting precisely of the

aj 's,

D(O, 1)

counted according to

their multiplicities. Specifically, the infinite product

00 - aJ
-

IT I Baj (z)
j=l laJ
converges uniformly on compact subsets of D(O, 1) to a bounded holomorphic
function

B(z).

The zeros of

their multiplicities.

are precisely the

aj 's,

counted according to

283

9.1. Jensen's Formula and Blaschke Products

For the convergence it is enough, by Theorem 8.1.9, to fix


and to check that
Proof.

11+ 1::1Ba;(z)I

<

<

z D (O, r),
+ajz-lajJ2
1I + Iaa13 I Baj(z) I IJajl-lajJajz
IajI(1 - Zllj) 1
(Jaji+zaj)(l-Jaji)
-zaj) I
I(1 lajl(l
+r)(l-Jaji)
JajJ(l - r)
When j is large enough, then lajl
1/2 (because the condition Lj(l IajI) entails limj IajI 1) so that the last line is
2 ( ) (1-IajI).
By the Weierstrass M-test, the convergence of Lj(l-JajJ) now implies
that
f 11+ J : J Baj(Z) I
converges uniformly on D( r).
The statement about the zeros of B is obvious from Theorem 8.1.9.
The assertion that
JB(z)I 1
is clear since each IBaj(z) I l.

converges uniformly on D (O, r). Now, for

<

2:

< oo

J=l

0,

Remark: A

Blaschke product is an expression of the form

zm. IT al Baj(z),
where
is a nonnegative integer. When L::(l -lajI)
guarantees that the product converges.
f
D (O, 1)
{ aj}
00

j=l

Corollary 9.1.6. Suppose that

< oo,

the theorem

is a bounded holomorphic function on

vanishing to order m 2: 0 at 0 and

are its other zeros listed

with multiplicities. Then

f(z) zm { ft-1:11Baj(z) } F(z)


=

J=l

(9.1.6.1)

9. Applications of Infi.nite Sums and Pr odu cts

284

where Fis a bounded holomorphic function on D ( O,

lf (z)I

sup
zED(O,l)
In other words,

sup
zED(O,l)

1), Fis

zero free, and

IF (z)I.

is the product of a Blaschke product and a nonvanisl1ing

function.
Proof. We

define F

by

F (z)

J(z)
=

zm

[f11

(aj/lajl) Ba1(z)

Then, by the Riemann removable singularities theorem, F is holomorphic


on D ( O,

1),

and clearly F is zero free.

Since
00

lzml

IT -1 :11Ba1 (z)

j=l

on D, it follows that
sup IF (z)I sup IJ (z)I.
zED
zED
For the reverse inequality, let N be a positive integer and define

FN (z)
Let E > 0. Since BN E C ( D) , we may select

ro

f (z)
zm. BN (z)

<

'

such that if

ro

<

<

1,

then

IBN (rei9)1

>

E.

But then, by the maximum modulus principle,


sup IFN (z)I
zED

sup
IFN (z)I
zED, lzl>ro
<

lf (z)I
E
zED 1

r0 m sup
-

----+

as

ro

----+

sup
zED

lf (z)I
1 E
-

. Since E was arbitrary, we see that


sup IFN(z)I sup lf (z)I.
zED
zED

Since FN converges uniformly on compacta to F, the inequality


sup IF (z)I sup IJ (z)I
zED
zED
follows.

9.2. The Hadamard Gap Theorem

285

9.2. The Hadamard Gap Theorem


In this brief section we present a technique of Ostrowski for producing se
ries which exhibit a phenomenon called "over-convergence"

( to

be defined

below ) . It was discovered by J. Hadamard that these series produce holo


morphic functions on the disc for which no PE 8D is regular ( in the sense
of Section

( or

8.3).

The series are examples of what are called "gap series"

"lacunary series" ) . These are series that are formed by deleting many

terms from a series formed by a regular pattern. They have various uses in
analysis. For instance, they can be used to construct continuous, nowhere
differentiable functions

( see

Exercise

13).

Let us begin with a concrete example


Consider the power series

Ln z2n.

( cf.

Exercise

lzl < 1. Thus the power series defines a holomorphic


1). We claim that no point of 8D is regular for F.
F'(rw)

3).

since

1)

when

To see this, consider

Chapter

This series converges absolutely and uni

formly on compact subsets of the unit disc D ( O,

D ( O,

21,

for

r E (0, 1)

and

function

F(z)

on

for

w such that w2

a fixed positive integer N. Now

N-l
F'(rw)

L 2nr2n-1W2n-1 + L 2nr2n-1W2n-1
n=l
oo

n=N

N-1

2::.: 2nr2n-1w2"-1 + _!_ 2::.: 2n. r2n-1 . i.


oo

n=l

n=N

Thus
lim

r->1-

IF'(rw)I

+oo

since, for N fixed, the first sum is bounded while the second tends to +oo

( each

summand is positive real and tends to a limit greater than

1).
N
1,
We conclude that F' is unbounded near all w E 8D ( O, 1) with w2
some positive integer N. But the set of all such w is dense in 8D ( O, 1). Thus
extension of F to a neighborhood of a point in 8D ( O, 1) is impossible.
=

It is natural to try to generalize this example as much as possible. But


the process is not easy. The proof of the following theorem uses a very clever
trick, which at first might seem unmotivated. Try to see the elements of our
example lurking in the background:

9. Applications of Infinite Sums and Products

286

Theorem 9.2.1

( Ostrowski-Hadamard ) . Let 0 < p1 < p2 <

and suppose that there is a

PJ+l
Pi

A >

be integers

1 such that

>A

for

1,2, ....

(9.2.1.1)

Suppose that, for some sequence of complex numbers { ai}, the power series

f(z)

00

L ajzPi

(9.2.1.2)

j=l

has radius of convergence 1. Then no point of an is regular for f. {Here a


point p of an is called regular if f extends to be a holomorphic function on
an open set containing n and also the point P.}
Proof. Seeking a contradiction, we suppose that some P

for

f;

without loss of generality we take

loss of generality, since changing

wz, lwl

to

=
=

an is regular

1. ( This is indeed without


1, gives a series of the same

form, with coefficients having the same absolute values as before; so the new

1, by Lemma 3.2.6.) Then there is a


n(l, t: ) and a holomorphic Fon U = n(O, 1) U n( l, t: ) such that

series also has radius of convergence


disc

Fln(o,1 )nD(l ,)
Choose an integer k

>

t/J(l)

l; and if

lt/J(z)I

fln(o,1 )nD(l ,)'

0 such that (k+ 1)/k <

t/J(z)
Notice that

lzl

and define

(zk+zk+l ).

1 but

z # 1,

zk z+ ll
2l l l

<

then we have

zk 2
2l l

1.

t/,J(n) is a compact subset of U. It follows by continuity of t/J that there


n(O, 1 + 6) such that t/,J(n(o, 1 + 6)) C U. Note also that 1 E
t/,J(n(O, 1+6)).
So

is a disc

Define

G(z)

F(t/J(z)),

n(O, 1+6).

Expand G in a power series about 0:


00

n=O

Compare this formula with what is obtained by substituting

zk+1 )/2

into the power series for F=

( Fo t/J)(z)

L aj
J

on

n(O, 1):

1
zk+ zk+1
2
2
1

) Pi

t/J(z)

(zk+

9.2. The Hadamard Gap Theorem

Notice that the

jth

term of this series contributes powers of

zk Pi
while the

(j+ l)st

287

z(k+l)Pi

to

term contributes powers of

ranging from

(9.2.1.3)

ranging from

to

(9.2.1.4)

(9.2.1.1) and the choice of k guarantees that (k+l)pj < k(Pj+ I ) so


that the powers appearing in Eq. (9.2.1.3) are distinct from those appearing
in Eq. (9.2.1.4). As a result,
But Eq.

(k+l)PN

L aj(1/i(z))Pi

j=O

L
i=O

Cfi.

The right side converges as N -t oo on the disc D(O, 1+6). Hence so does
the left side. In other words,
00

L ajwPi

j=O
converges for all
all

1/i(D(O, 1+6)). In particular this series converges for

in a neighborhood of

1,

so its radius of convergence is not

1.

This

contradicts our hypothesis.

In general, a point P in the boundary of the (unit) disc can be a


regular point for the holomorphic function f (on the disc) without the power
series for f about 0 converging in a neighborhood of P. A good example is

f(z)

1/(1

z).

Then all P E 8D

\ {1}

are regular but the radius of

convergence of the power series is


exhibited in the proof of Theorem
Theorem

9.2.1

1. This explains why the phenomenon


9.2.1 is called over-convergence.

applies in particular to our explicit example of a non

continuable analytic function. The series


00

F(z)

L: z2i

j=O

satisfies the hypotheses of the theorem with A

3/2,

for instance. Hence

no point of 8D is regular for f, as we have already checked directly using


an ad hoc argument. The classical terminology for a function f which is not
regular at any point of 8D is that f has 8D as its "natural boundary."
For comparison purposes, recall that in Section

8.3

we showed by dif

ferent methods that if U C is any open subset of C, not equal to C, then


there is a holomorphic f on U that is not the restriction to U of a holomor
phic function defined on a larger open set, that is, f cannot be "continued"
to a larger (connected) open set. Such an f has 8U as its natural boundary

9. Applications of Infinite Sums and Products

288

in the sense already indicated. But gap series give a simple and direct set
of examples in the special case when U is a disc.

9.3. Entire Functions of Finite Order


This section contains the version for entire functions of a basic idea that we
alluded to earlier: The rate of growth of a function controls the distribution
of its zeros. We have already seen this theme occur in our study of bounded
holomorphic functions in the disc. But the entire function theory is harder.
The Hadamard factorization theorem proved in this section is a step
ping stone, both in technique and in content, to the still more advanced
topic known as the Nevanlinna theory of distribution of values of holomor
phic functions. We do not explore Nevanlinna theory in this book, but a
good exposition may be found in [NEV]. This section is not required for the
reading of the remainder of this book. But it represents an important aspect
of classical function theory, and the philosophical principle that it illustrates
is useful.
Let us say that an entire function f is of finite

a, r

>

order if there are numbers

0 such that
for

lzl

>

r.

In other words, f is of finite order if it grows exponentially for large values


of

z.

We let A

.X(J) be the infimum of all numbers

The number A is called the

order of

for which ( * ) is true.

f. Notice, for instance, that sin z is of

order 1, while exp(exp z) is not of finite order.


Our aim is to see what A say s about the rate at which the zeros of J
tend to infinity. If

{ aj}

are the zeros of f, then we measure this rate by a

sum of the form

The infimum of all K's for which this sum is finite is an indicator of how
many zeros of f are in

D(O, r)

when

is large. If there are a great many

zeros of f, then we would have to take a bigger (negative) power to make


the series converge. For example, if the entire function f has simple zeros
at 1, 2, 3, ..., then any K

>

0 would make the series converge. If another

entire function g has simple zeros at 1,

J2, J3, ... ,

then K would have to

be greater than 1. To facilitate this discussion, we let


of zeros in

D(O, r)

of our entire function f.

We begin with a technical lemma:

n(r)

be the number

9.3. Entire Functions of Finite Order

Lemma 9.3.1. If

maxlzl=r lf(z)I, 0 <

f is an entire
r < oo. Then

289

function with

( log2 ) n(r)

0 =log lf(O)I =
where

n 2r )

log

k=l

ai,...,an(2r) are the


la1I la2I

Assume that

la I

n(r}

so that

z::: log (2 )

k=l

k=I

<

2r
ak

log -

_!._ f
27r l
o

27r

log lf(2rei8)1 dB

log M(2r).

<

n(r)

2
{ 7r log f(2rei8)1 dB,
l

D(O,2r) counted with multiplicity.


ai,...,an r ) E D(O,r). Hence

L log

<

M(r)

27r l
o

I I
n(2r} 2r
z::: I a I
k
k=I

n(r}

Equivalently,

(2 )

zeros off in

1, let

logM(2r).

Proof. By Jensen's formula,

f(O)

log2 logM(2r).
D

Theorem 9.3.2. Let

f(O)

1.

If ai, a2, ...

be an entire function of fi.nite order ,\ and satisfying

are the zeros off listed with multiplicities, then


00

L lanl-.X-1 < oo.

n=l

Proof. We know from Lemma 9.3.1 that

( log 2) n(r)

Since

has finite order ..\,

M(r)
for any small f > 0 and

oo. For

<

r>.+ / 2
e

sufficiently large. Thus

n(r)r-(ME)
as

log M(2r).

1
)
--r-(ME
log2

. (2r)ME/2

sufficiently large it follows that

n(r)

r.x+E.

290

9. Applications of Infinite Sums and Products

Assuming without loss of generality that

{a1, ...,aj}, we

have

j
for

large and

'5

>

n( lai l

Ja1I Ja2I . .. so that D(O, Jaji )

+ '5)

( Jai l

+
+ '5)A E

arbitrarily small. Letting

we find that

'5
0,
laj1-(A+l) j-(A+l)/(A+E)

Selecting E =

1/2,

say, now shows that the series

converges.
In Theorem

8.2.2

and what followed we showed how to write a factor

ization of an entire function in a fashion that made its zeros explicit ( the
Weierstrass factorization). In the case that the entire function f has finite
order A, a variant of Theorem

9.3.2

with

gether with the discussion in Section


written in the form

8.2,

lanl-A-l

replaced by

Janl-A-f,

to

implies that the function f can be

n
[Here [] is the greatest integer function.] That is, the factorization of The
orem

8.2.2

can be taken to have

Pn

[A]

"greatest integer

all n. This particular factorization will be called the

factorization

off. We also set

P(z)

A" for

Weierstrass canonical

IJ E[Aj(a/zn).
n

Now that we have examined what A(f) says about the zeros of f, we
will turn to the other part of the Weierstrass factorization: the function g.
Several technical lemmas are needed for this purpose.

Lemma 9.3.3. Let f be


Let p >

an entire function of finite order

A with f (0) =

- 1 and let {aj} be the zeros of f (listed with


Ja i l Ja2I
. Then, for any z EC,
n(r)
1
1.
L: -p
im
ak+l ( r2 - akz ) p = 0 .
r->+oo
k=l

Suppose that

1.

multiplicities).

Proof. Let

be fixed and

for these a's we have

>

2Jzl. Of course a1,... , an(r)

Jr2 - akzl 2:: r2 - r (r/2)

r2/2.

lie in

D(O, r)

and

9.3.

291

Entire Functions of Finite Order

Hence
laklp+llr2- llkZ1-p-l

rP+l

( ) p l ( ) p l

(9.3.3.1)

Now f being of order A says that


IJ(rei9)1 er>-+
for any small and r large enough. Thus, by Lemma 9.3.1,
(log 2)n(r)r-p-l

<

Combining this with

n(r)

Eq.

(logM(2r))r-p-l
log

(e(2r)>-+) r-p-l

2A+ErA+E-p-l.

(9.3.3.1) gives

2 p+l 2.x+p+E+l
+
1
1
1
a
r.X+E-p-1.
kz)-pn(r)
(r2
t
a:
I::

2
r

1
If we take (p + 1 - A)/2, then the exponent of r is negative so that the
D
last line tends to zero as r
+oo.

(-)

Lemma 9.3.4. If f is entire of fi.nite order A and f(O)


p >A- 1, pan integer, and z EC fixed,

7r

as

2_ f 2 2rei9(rei9- z)-P-2 log lf(rei9)1 d(}


2Tr lo

Proof.

1, then, for

+oo.

The function

has residue

</>(w)

at z E D(O,r). Hence

(w- z)P+2

fav(O,r) </>(w) dw =

0.

In parametrized form,

As a result,

(9.3.4.1)
I 2 h27r 2rei9(rei9- z)-p-2 log IJ(rei9)1 d(} I
I 2 h27r(2rei9(rei9 - z)-p-2)(log lf(rei9)1- logM(r)) d(} I
=

292

9. Applications of Innnite Sums and Products

1
271"

1211" 2r ( r-2 ) -p-2 (log M(r) - log lf(rei8)1)dO.


0

By Jensen's inequality,
1 { 211"
log lf(rei8)1d()2: log lf(O)I
271" l
o

0.

Thus Eq. (9.3.4.1) is less than or equal to


2P+3r-P-1 log M(r).
Since f is of finite order

(9.3.4. 2)

.X,

log M(r) .:::; rA+{


for any small

and r sufficiently large. So Eq. (9.3.4.2) is

With

(p

2p+3r.X+{-p-l.

+ 1 - .X)/2, the expression tends to


S

as

r---+

oo.

f be a nonconstant entire function of E.nite order


.X and suppose that f(O)
1. Let {al, a2, ... } be the zeros off listed with
multiplicities. Suppose that I a1 I S I a2I S
. If p > .X 1 is an integer,
Proposition 9.3.5. Let

[ ]

then

dP f'(z)
dzP f(z)

Proof.

Let r

>

I rr(z--ajaJz)
L,,
- '"'log
( )

1
p
- ! L (aj - z)P+l
J

(9 3. 5 .1)

2 lz l. The Poisson-Jensen formula (Exercise 1) says that


2

n r

log lf(z)I

J=l

+_!__27!" lro 211" Re ( rereill z ) log lf(reill)I d().


i
I o
o

- z

By logarithmic differentiation in z,

z)
f(z)

( )

!'(

n r

( )

n r

L,, z - aj )-1 '"'L,, aj (r 2 - ajz )-1


''(
J=l
J=l
27r
f 2rei8(reill - z)-2 log lf(rei8)1dO.
271" l
o
Differentiate both sides of the equation p times to obtain
=

+_!__

( d ) p [
'( )]
_

dz

f(z)

( )

n r
_
=

p.I''
L,, ( aJ
J=l

z )-p-1

l
+p.I '"'L,, aJp+ (r2
( )

n r

J=l

aJz)-p-1

9.3. Entire Functions of Finite Order

293

By Lemma 9.3.3, the second sum on the right tends to zero as


Likewise Lemma 9.3.4 says that the integral tends to

as

r -+

r -+

+oo.

+oo. The
D

result follows.

The following lemma is not quite a special case of Proposition 9.3.5


because we do not know a priori that the infinite product in the Weierstrass

canonical factorization is order .X. But the result is easy because of the work
we have already done in proving Proposition 9.3.5.

Lemma 9.3.6. If f is an entire function of finite order .X, f (0)

P(z)

IT

n=l

E[-'J

Proof. If

[P'(z)]
P
(z)

-p!
L
n

.X

>

-1

we have

- z)P+l

(9.3.6.1)

PN

is the Nth partial product , then it is clear that

dP
dzP

[Pfv(z)]-- N
a

PN(z)

p.

1, and if

(a:)

is the associated product, then for any integer

dP
dzP

- z)P+1

( error ) .

(9.3.6.2)

By Proposition 9.3.5 the sum on the right converges to the full sum on the
right of Eq. (9.3.6.1) and the error becomes

when

(p + l)st

derivative of a polynomial of degree

then

-+

p.

>

Since

.X - 1 , since it is the

PN P
-+

normally,

normally so that the expansion on the left of Eq. (9.3.6.2)

Pfv P'

converges normally to that on the left of Eq. (9.3.6.1).

Theorem 9.3. 7. If f is an entire function of finite order .X and f(O)

1,

then the Weierstrass canonical product

f(z)
has the property that

eg(z) P(z)

is a polynomial of degree less than or equal to .X.

Proof. Differentiate

f(z)
The result is

Let

>

.X

f'(z)
f(z)

- 1.

eg(z)

(z) .

P
P'
P

(z) +

Differentiating both sides

(z)

(z)

times and applying Proposition

9.3.5 and Lemma 9.3.6 gives

-p!

L(

dP+l
(z) p
- z)P+l = dzP+l g
- !L

- z)P+l.

(9.3.7.1)

294

9. Applications of Infinite Sums and Products

It follows that

and hence

dP+l
g(z) =
dzP+l

is a holomorphic polynomial of degree not exceeding p.

The purely formal proof of Theorem


hard work was.

( Lemma 9.3.4)

perhaps obscures where the

The point was that the logarithmic factorization lemma

(9.3.7.1),

(9.3.5.1)

actually converges. This enabled us to

and the rest is automatic.

f(O)
f

Notice that the hypothesis

9.3.7

9.3.7

is a formal triviality, but the finite order condition is needed

to see that the series in Eq.


write Eq.

in Theorem

9.3.2

and Theorem

vanishes to order m at 0, then

is in effect superfluous since if

(z)
f(z) = J
zm
will still have finite order A and the theorems may be applied to
normalized so that

f(o)

9.3.2

Theorems

and

f,

suitably

1.

9.3.7 taken together amount to the classical

Hada

mard factorization theorem. In order to formulate the result in its classical


form, we need the classical terminology. If

rank p

the zeros off, then we define the

f is entire of finite order and { aj}

off to be the least ( nonnegative )

integer such that

JanJ-(p+l)

< oo.

anO

Notice that Theorem


Theorem

9.3.7

9.3.2

guarantees that such a p exists. We know from

and the subsequent remark that

J(z)
where

zm eg(z) P(z),

is a polynomial. Note that the rank of the product is now p. If q

is the degree of

g,

then we define the

genus

off to be the maximum of p

and q. Now the Hadamard factorization theorem simply says the following:
Theorem 9.3.8. An entire function of finite order ..X is also of finite genus

and
5: ..X.
The order A is also controlled from above by the genus ( see Exercise

12).

We shall now derive a few interesting consequences about value distri

bution theory, that is, the study of how many times an entire function can
assume any given value.
Theorem 9.3.9. Let

be an entire function of finite order A <t Z. Then

there are infinitely many distinct

Zj EC

such that

f(zj)

c.

295

9.3. Entire Functions of Finite Order

Proof. We may assume that c = 0. Now suppose that the assertion is false.
1
Then 1- ( {O})
{a1, ..., aN} for some N E Z and
=

f(z)

By Hadamard's theorem,
if E

>0

and Jzl

>>Jail

(z - aN).

is a polynomial of degree not exceeding>.. But

for all j, then

Jeg(z)J
z

and for a sequence of

eg(z) (z - ai)

= ) f(z)J

I I1j=1(z - aj)I

with Jzl

elziH

oo

Jeg(z) I 2'. elzl"'-.


It follows that the order of

e9

is>..

But it is clear that the order of


contradicting the hypothesis that>.

Theorem 9.3.10.

Let

e9

g. Hence q

is just the degree q of

Z.

=>.,
0

be a nonconstant entire function of fi.nite order.

Then the image of f contains all complex numbers except possibly one. If,

in addition,

has nonintegral order, then

infinitely many times.

assumes each of these values

Proof. We need only prove the first statement. If the image off omits two
complex numbers

a1

and a2, then we consider

never vanishes, we may write

f(z) - a1
for some entire function
a polynomial. Since

g(z)

and Hadamard's theorem guarantees that

f(z) - a1
log(a2 - ai). So if /3

- a1; thus g omits the value(s)


a1 - a2, then the polynomial g(z) - /3

fundamental theorem of algebra unless

this function

eg(z)

omits the value a2, then

a2

makes

f - a1. Since

is

omits the value


E

C with e/3

never vanishes. This contradicts the

is constant. But

constant, contradicting our hypothesis.

being constant

Notice that a polynomial has order 0. So it is excluded from Theorem


9.3.9, as it should be. But the first part of Theorem 9.3.10 certainly applies
to polynomials.

10) asserts
9.3.10 is true for any entire function. The
"Great Theorem" of Picard (again see Chapter 10) says in effect that the
conclusion of Theorem 9.3.10 holds near any essential singularity of a holo
The "Little Theorem" of Picard (to be studied in Chapter

that the first part of Theorem

morphic function. The proofs of Picard's theorems use a different technique


from the ones used in the present chapter.

The new technique is part of

a more geometric approach; and, while it produces a stronger result than

9. Applications of Infinite Sums and Products

296

Theorem 9.3.10, there are other aspects of the results of the present section
that cannot be recovered using the geometric methods of Chapter 10.

Exercises
1.

r)

(The Poisson-Jensen formula): Let zo E D(O,


be fixed. Let f be
holomorphic on a neighborhood of D(O,r). Let a1,... ,an be the zeros
off in D(O,
and assume that no zeros lie on aD(O,
Let </>(z)
( 2z+ rzo)/(r+ zzo). Apply Jensen's formula tof </>and do a change
of variable in the integral to obtain the formula

r)

logf
l (zo)+
l

L: log

k=I

Ir

- akzo

r2
(zo- ak )

1
27r

=-

271"

Re

(reiO )
rei
r),

forf holomorphic in a neighborhood of D(O,

+ zo
- zo

r).

rei

logf(
l

.9

)l d0

f(zo) =/= 0.

2.

Give an alternative proof of the Poisson-Jensen formula (Exercise 1) by


imitating the proof of Jensen's formula but using the Poisson integral
for D(O, instead of the mean value property.

3.

Calculate the genus of cos JZ, sin2 z, sin(z2).

4.

-aj/lajl necessary in Theorem 9.1.5?


Let {aj} D satisfy Lj 1-lajl < oo and let B(z) be the corresponding

5.

r)

Why are the factors

Blaschke product. Let PE aD. Prove that B has a continuous extension


to P if and only if P is not an accumulation point of the aj 's.
6.

Construct a convergent Blaschke product B(z) such that no PE aD is a


regular point for B (i. e. , the holomorphic function B does not continue
analytically past any point of aD).

7.

Is the functional composition of two Blaschke products a Blaschke prod


uct? [Hint: Read the definition of "Blaschke product" carefully.]

8.

Let B be a convergent Blaschke product. Prove that


sup IB(z)I

1.

zED

9.

10.

Let B1,B2,.. . be Blaschke products. Suppose that {BJ } converges nor


mally to a nonconstant holomorphic function B0 on D. Is B0 a Blaschke
product? [Hint: Read the definition of Blaschke product carefully.]
Let

{an}

<C

satisfy
00

L 1 1- lanll

n=l

< oo.

297

Exercises

Discuss the convergence of

in the entire plane. Is oo a pole, an essential singularity, or a regular


point?

11. Let

<jJ

be a continuous function on

[a, b].

Let 0: ER Prove that

is an entire function of finite order. Can you compute the order of f?


Does it depend on ?

12. Prove that finite genus implies finite order.


to show that

(1

log

lzl) log IE(z)I lzl+I

[Hint:

Use Taylor's formula

so that by induction

IP(z)I (2 + 1) lzl+I L lanl--l.


n

T he sharp relation between the genus and the order,\ is,\+l.]

13. Complete the following outline to construct a continuous, nowhere dif


ferentiable function (notice the role of gap series in this argument):
Define
2 -j sin(2Jx).

(a)
{b)
(c)

f(z)

Verify that f is continuous on the real line.

(d)

Fix x. Analyze the Newton quotient

Pick

h>
h 2-J.

0 small and choose a positive integer J such that 2-J-I

f as

J(x)

I:

OjJ-I

(e)

L:j:0

rj sin(2jx) +

tends to

O;

J(x)] / h by

writing

I: rj sin(2Jx) + rJ sin(2Jx).

j>J

Determine from your analysis in


unbounded as

[f(x + h)

<

{d)

that the Newton quotient is

thus f is not differentiable at x.

14. Here is a second way to see that the power series

has no regular point in

oD.

You should provide the details.

Notice that the series converges uniformly on


continuous on
holomorphic

oD.

{ z : I z I 1}; hence f is

If eiOo E oD were regular for f, then there would be a

j on some D(eiOo, E ) such that j


j on oD, so J(ei9)

But this necessitates f

f on

D(eiOo, E ) n D(O,

1).

would be a differentiable

Exercises

298

function near ()

Oo. However

J(ei(J)

00

I: rne2ni(J
n=O

is the Weierstrass nowhere differentiable function (see Exercise 13 or


[KAT]). That is a contradiction.

15. Recall that

J11

log lxldx is a convergent improper integral: -1/log lxl

has an "integrable singularity" at 0. Use this fact to fill in the details


of the proof of Jensen's formula (Theorem 9.1.2) for the case that the
holomorphic function f has zeros with modulus

f(a)

0, lal

r.

For this, you will need

a, f(z) (z - a)kh(z) with


h(a) =I= 0, k 2: 1, so that log Jf(z)J k log Jz - aJ +log Jh(z)J. Then note
that log Jz - aJ, with z restricted to the boundary circle and with z near
a, behaves like log lzJ along the real axis, namely it has an (absolutely)
to note that if

r,

then, near
=

integrable singularity.

Chapter 10

Analytic Continuation

10.1. Definition of an Analytic Function Element


Suppose that V is a connected, open subset of C and that

Ji : V ---+ C and
C are holomorphic functions. If there is an open, nonempty subset
U of V such that Ji =hon U, then Ji =hon all of V. Put another way, if

h: V

---+

we are given an

f to

f holomorphic on U,

then there is at most one way to extend

V so that the extended function is holomorphic. [Of course there might

not even be such an extension: If V is the unit disc and U the punctured
disc, then the function

f(z)

1/z

does not extend. Or if U is the plane

with the nonpositive real axis removed, V


-tr

< () <

tr,

C, and

f(rei9)

r112ei9/2,

then again no extension from U to V is possible.]

This chapter deals with the question of when this (loosely described)
extension process can be carried out, and in particular what precise meaning
can be given to enlarging the set V as much as possible. One might hope
that, given U and

f:

---+

C holomorphic, then there would be a uniquely

determined maximal V 2 U to which

f extends holomorphically.

This turns

out not to be the case. However, there is a complete theory concerning these
questions; this theory is the subject of the present chapter.
We introduce the basic issues of "analytic continuation" by way of three
examples:

EXAMPLE 10.1.1. Define

f(z)

00

Lzj.
j=O

This series converges normally on the disc D


for

lzl

>

1.

{z EC: lzl

<

l}. It diverges

Is it safe to say that D is the natural domain of definition for


-

299

300

10. Analytic Continuation

f ( refer

to Section 9.2 for this terminology ) ? Or can we "continue"

to a

larger open set?


We cannot discern easily the answer to this question simply by ex
amining the power series.

Instead, we should sum the series and observe

that

f(z)
This formula for

f agrees

z.

with the original definition off as a series; however

the formula makes sense for all


made more precise later,

1
=

\ { 1}.

In our new terminology, to be

an analytic continuation to C

f has

Thus we see that the natural domain of definition for


large set C

\ { 1}.

\ { 1}.
is the rather

However, the original definition, by way of a series, gave

little hint of this fact.

EXAMPLE

10.1.2.

fo00

Consider the function

r(z)

e-1e-t

dt.

This function is known as the gamma function of Euler; it is discussed in

15.1.

detail in Section

(1)

Let us make the following quick observations:

The term tz-l has size

the origin will be integrable

(2)

tRez-1. Thus
when Re z > 0.

w-11

the "singularity" at

Because of the presence of the exponential factor, the integrand


will certainly be integrable at infinity.

( 3)

The function

is holomorphic on the domain Uo =

The functions

with b
sign

( or

l/a,

1b

e-le-t

Re z >

0}:

dt,

are holomorphic by differentiation under the integral

use Morera's theorem ) , and

these integrals as

is the normal limit of

f(z)

a --+ o+.

The given definition of

r( z) makes no sense when


0. Can we conclude from
definition of r is U0?

improper integral diverges at


the natural domain of

{z

Re z ::::;

0 because the

this observation that

Let us examine this question by integrating by parts:

r(z)

100
0

e-1e-t

dt

.
=

1
z

-ee-t

00 z1 100
I
+

An elementary analysis shows that, as long as Re z >


vanish ( in the limit ) . Thus we see that

0,

tze -t

dt.

the boundary terms

10.1. Analytic Function Elements

30 1

Now, whereas the original definition of the gamma function made sense on

Uo,

this new formula

sense on
integral

U1

as

{z:

( which agrees with the old one on Uo) actually


> -1} \ {O}. No difficulty about the limit
limit tends to o+ occurs if z E { z : Re z > -1}.

Rez

the lower

makes
of the

We can integrate by parts once again and find that

r(z)= z(z 1 1) 100


+

This last formula makes sense on

t':+le-t dt.

U2 = { z :

Re z

>

-2} \ {O. -1}.

Continuing this process, we may verify that the gamma function, orig
inally defined only on

z f. 0, -1, -2, . . .}.

Uo,

U = {z EC :

can be "analytically continued" to

There are poles at

0, -1, -2, . .. .

In the first two examples, the functions are given by a formula that
makes sense only on a certain open set; yet there is in each case

device for

extending the function to a larger open set. Recall that, by our uniqueness
results for analytic functions, there can be at most one way to effect this
"analytic continuation" process to

fixed, larger

( connected )

open set. In

the next example, we learn about possible ambiguities in the process when
one attempts continuation along two different paths.

EXAMPLE 10.1.3. Consider the function

D(l, 1/2)

by

f(rei0) =r112e1012.

f(z),

initially defined on the disc

Here it is understood that

-7r/6 < () < 7r/6.

This function is well defined and holomorphic; in fact it is the function


usually called the

principal branch

of

Imagine analytically continuing

10. 1.

y'z.
f to

Note that
a

second disc, as shown in Figure

This is easily done, using the same definition

we continue to a third disc

( Figure 10. 2) ,

[f(z)]2=z.
f(re10) = r112ei0/2.

If

and so on, we end up defining

the square root function at z= -1. See Figure

10.3.

Indeed, we find that

J( -l)=i .
However, we might have begun our analytic continuation process

10. 4. If we continued the process to z= -1,


J(-1) = -i.

shown in Figure
found that

a.<;

we would have

Thus we see that the process of analytic continuation can be ambigu


ous. In the present example, the ambiguity is connected to the fact that a
holomorphic square root function cannot be defined in any neighborhood of
the origin ( see Exercise

1 4,

Chapter

6).

These examples illustrate that an analytic function can sometimes be


continued to a domain of definition that is larger than the initial one. How
ever, this continuation process can result in inconsistencies in the sense that
the value of a continuation at a given point may depend on which contin
uation is involved, in effect, on how the point was reached. In each of the

10. Analytic Continuation

302

Figure 10.1

Figure 10.2

Figure 10.3

examples, the continuation process was effected differently by a device spe


cific to the function.

This raises two general and rather vague questions:

10.1. Analytic Function Elements

303

Figure 10.4

How can we carry out analytic continuation in general? Furthermore, how


can we determine when it can be carried out unambiguously?
It is because of these questions that we must take a detailed and tech
nical approach to the process of analytic continuation. Even making the
questions themselves precise takes some thought.
Definition 10.1.4. A

is a disc

D(P, r )

function element is

an ordered pair (f,

U),

where

and f is a holomorphic function defined on U.

U) and (g, V) be function elements. We say that


(g, V) is a direct analytic continuation of (f, U) if Un V 1= 0 and f and g are
equal on Un V. Obviously (g, V) is a direct analytic continuation of (f, U)
if and only if (f, U) is a direct analytic continuation of (g, V).
Definition 10.1.5. Let (f,

If (/1,

U1), ..., (fk, Uk)

are function elements and if each

Uj-l Uj- 1), j


analytic continuation of (/ 1, U1).

direct analytic continuation of

(fk, Uk)

is an

Clearly

(/ 1, U1)

(fk, Uk)

(fj, Uj)

is a

2, ..., k, then we say that

(!1, U1) if and only if


Also if (fk, Uk) is an ana

is an analytic continuation of

(fk, Uk)
U1) and (fk+f., Uk+e) is an analytic continuation
of (fk, Uk) via a chain (fk, Uk), (fk+1. uk+i), ... 'Uk+t.. uk+f.), then stringing
the two chains together into (/ 1, U1), ..., (fk+1., Uk+e) exhibits (fk+e, Uk+e)
as an analytic continuation of (f 1, U1). Obviously (f, U) is an analytic con
is an analytic continuation of

lytic continuation of (/1,

tinuation of itself.
Thus we have an equivalence relation on the set of function elements.
The equivalence classes induced by this relation are called

functions.

(global} analytic

However, a caution is in order: Global analytic functions are not

yet functions in the usual sense, and they are not analytic in any sense that

10. Analytic Continuation

304

we have defined as yet. Justification for the terminology will appear in due
course.
Notice that the initial element

(f, U)

(!1,

Ui)

uniquely determines

the global analytic function, or equivalence class, that contains it.

But a

global analytic function may include more than one function element of the
form

(f, U) for a fixed disc U.

Indeed, a global analytic function

may

have in effect more than one value at a point of C: Two function elements
(!1,

U)

and

(h, U)

can be equivalent even though

is the center of the disc

U.

Ji(P) i- h(P),

where

If f denotes the global analytic function

(f, U), then we call (.f, U) a branch off (see Exercises 14


Chapter 6). Example 10.1.3 illustrates a global analytic function

corresponding to
and

15

in

(namely, the square root function) with two distinct branches centered at
the point

-1.

Logarithms of z again illustrate the point:

EXAMPLE 10.1.6. Let

D(2, 1)

and let

be the holomorphic function

log z. Here log z is understood to be defined as log lzl +

i arg z,

and -rr /6 <

arg z < rr /6. As in the preceding example, the function element


be analytically continued to the point

-1 + iO

(f, U)

can

in (at least) two different

ways, depending on whether the continuation is along a curve proceeding


clockwise about the origin or counterclockwise about the origin.
In fact, all the "branches" of log z, in the sense of Exercise

14, Chapter
log I z I + iarg z branch on

6, can be obtained by analytic continuation of the


D(2, 1). Thus the usual idea of branches of log z coincides

with the general

analytic continuation terminology just introduced.


In some situations, it is convenient to think of a function element as a
convergent power series. Then the role of the open disc
domain of convergence of the power series.

is played by the

This is a useful heuristic idea

for the reader to bear in mind. From this viewpoint, two function elements
(!1,

U)

and

(h, V)
P)

the same point

at a point

(such that

and

should be regarded as equal if

V are discs centered at


Ji = h on U n V. This

identification is convenient and we shall introduce it formally a little later.


As our examples have already indicated, the question of nonambigu
ity of analytic continuation is bound up with questions of planar topology.
These, in turn, are handled by the concept of homotopy-continuous de
formation of curves. Therefore, in practice, it is useful to think of

continuation along a curve.

analytic

That is the topic of the next section.

10.2. Analytic Continuation along a Curve


Definition 10.2.1. Let -y
tion element with

-y(O)

[O, 1]

-t

C be a curve and let

the center of the disc

(Figure

(f, U) be a func
10.5). An analytic

305

10.2. Analytic Continuation along a Curve

y(l)

Figure

10.5

continuation of(!, U) along the curve 'Y is a collection of function elements


Ut, Ut), t E [O, 1], such that
(1) (Jo, Uo) (!, U);
(2) for each t E [O, 1], the center of the disc Ut is 'Y(t), 0 :<:::; t :<:::; 1;
(3) for each t E [0, 1], there is an E > 0 such that, for each t 1 E [O, 1]
with lt 1 - ti < E, it holds that
( a ) 'Y(t 1) E Ut and hence Ut1 n Ut /:- 0;
ft1 on Ut1 n Ut [so that Ut1, Ut1) is a direct analy tic
(b) ft
continuation of (ft, Ut)] .
Refer to Figure 10.6.
=

Proposition

10.2.2.

Let(!,

U)

be a function element with

center P. Let 'Y be a curve such that "f(O)


tions of(!,

U)

a disc having

P. Any two analytic continua

U1) is the terminal


if (h, U 1) is the terminal
then Ji and Ji are equ al

along 'Y agree in the following sense: If(!1,

element of one analytic continuation

Ut, Ut) and


(h, Ut),

element of another analytic continuation

on

U1 n U1 .

Proof. Let S denote the set of

t0

[O, 1] such that, for all

[O,

t o],

the

ft are equal on Ut n Ut. [Note that Ut n Ut is open, since


Ut are; the intersection is also nonempty since 'Y(t) lies in both

functions ft and
both Ut and

sets .] We intend to use a "continuity argument" (or connectivity argument)


to see that S

[O,

1].

The idea is to show that S is nonempty, S is closed

in [O, 1], and Sis open in [O, l]. It will then follow that S
First, we know that Sis nonempty since
let

to

S and take

[O,

t o] .

Then

0E

[O,

1].

S. To see that Sis closed,

S by definition.

Thus to show

10. Analytic Continuation

306

Figure 10.6
that S is closed, we need only check that if {tj} is an increasing sequence
of elements of S, then the limit point t' lies in S as well. For this, choose a
positive number 1 for Ut'' Ut') and another positive number 2 for (h1' ut')
as in the definition of analytic continuation along "I Choose j so large that
lt1 - t'I < min(E1,E2). Then

"1(t1) E Utj nUtj nut' nut'


Also, since t1 ES,it holds that ft1

ft1 on Ut1 n UtJ' Therefore

on

Ut ). n Ut ). nut' nut'.

Thus

ft1

ft1

on

Hence t' ES.


To see that S is open,consider the complement of S in [O,1]. Arguments
similar to those just given suffice to show that this complement is closed in
[O,l]. In more detail, to see that [O,1] \ S is closed, it is enough to show
that if {t1} is a convergent sequence in [O, 1] \ S, then limtj E [O, 1] \ S.
Set t0
lim t1. Now the fact that tj S implies, by definition of S, that
=

there is an Sj E [O,1] with Sj tj and fs1 . Ts1 on Us1 nUs1 . Passing to a


subsequence if necessary, we can assume that lim Sj exists; call the limit so.
Clearly s0 t0 sinc:._sj t1 for 11 j. To show tha to E [O,1] \_,it suffices

to show that fso . !so on Us0 n Us0 But if !so =}so on Us0 n Uso, then the
_
same reasoning as used earlier shows that fs1
fs1 on Us 1 n Us1 for j large
enough. This conclusion contradicts the choice of the sj's.
=

The Monodromy Theorem

10.3.

307

This completes the proof of the continuity argument: S is nonempty,


open, and closed. Thus S

[O, 1]

and we are done.

Thus we see that the analytic continuation of a given function element


along a given curve is essentially unique, if it exists.

From here on, to

avoid being pedantic, we shall regard two analytic continuations (ft, Ut) and

(ft, Ut)

as "equal," or equivalent, if ft

ft

on

Ut n Ut

for all

t.

With this

terminological convention (which will cause no trouble), the proposition says


exactly that analytic continuation of a given function element along a given
curve is unique.

It should be stressed, however, that this is a uniqueness

statement only-not an existence statement.


The most elementary method for trying to continue a given function
element

(J, U)

(with

D(P, r ))
'Y(t1) of 'Y

following: Choose a point

U.

along a curve

that is in

Calculate the power series off at

'Y(t1).

'Y

starting at

is the

but near the boundary of

With any luck, this power series

has a radius of convergence larger than the distance of

U,
'Y

'Y(t1)

to the edge of

so that the series can be used to define function elements further along
One might hope to continue this process.
The procedure described in the last paragraph works quite effectively

for the function element

(J, U)

where

for any curve "f, beginning at, say,


point

0.

U
D(l, 1/2) and f(z)
.Ji and
iO that does not pass through the
0, then we would find that the radii
=

If the curve did terminate at

of convergence of our power series were always small enough so that the
chain of discs never engulfed
continuation at
If

0.

0.

In short, there is an obstruction to analytic

U is a connected open set in C

andf is a holomorphic function on

U,

then, by an "analytic continuation off," we mean an analytic continuation


of some function element

(J, V)

where

V U.

In practice no confusion

should arise.

10.3. The Monodromy Theorem


The fundamental issue to be addressed in the present section is this:

Let P and Q be points in the complex plane. Let (!, U) be a


function element such that U is a disc centered at P. If")'1, ")'2 are
two curves that begin at P and terminate at Q, then does the
terminal element of the analytic continuation of(!, U) along l'l
(supposing it exists) equal the terminal element of the analytic
continuation of(!, U) along ")'2 (supposing it exists)?
In the third example of Section

10.1,

there was an informal treatment

of an instance where in fact agreement does not occur.

It turns out that

the reason that such a failure could occur is that the curve

'Yo(t)

ei7rt,

308

:=::; t :=::;

1,

and the curve /'I (t )

e - i 7r t,

10.

Analytic Continuation

1,

cannot be continuously

deformed to each other, keeping the endpoints fixed, so that all intermediate
curves admit analytic continuation of the original function element.
enemy, as already indicated, is the origin:
function element for

viz

The

It is not possible to define a

on a disc centered at the origin.

Thus it is not

possible to do analytic continuation along a curve that passes through zero.


Furthermore , there is no way continuously to deform l'o into /'I without
incorporating at least one curve that passes through the origin.
The monodromy theorem gives a precise formulation of the conditions
that will avoid the situation in Example

10.1.3.

F irst we need a definition:

Definition 10.3.1. Let W be a connected open set in C Let l'o : [ O, 1]


W
and l'l : [ O, 1]
W be curves. Assume that l'o ( O )
1'1 (0)
P and that
l'o (1)
/'I (1)
Q. We say that l'o and /'I are homotopic in W (with fixed
----t

----t

endpoints) if there is a continuous function


H : [ O,

1]

[O, 1]

----t

such that

(1)

H ( O, t )

l'o ( t ) for all t E [ O,

(2)

H (1, t )

1'1 (t ) for all t

(3)

H ( s,

0)

P for alls

(4)

H ( s,

1)

Q for alls E [ O,

1];

[ O, 1];

[O, 1];
1].

Then H is called a homotopy of the curve ')'o to the curve /'I (with fixed
endpoints).

[ Note:

Since we are interested only in homotopies with fixed endpoints, we

shall sometimes omit the phrase "with fixed endpoints" in the remainder of
our discussion.]
Intuitively, we think of a homotopy Has follows. Let Hs ( t )
Then condition

(1)

says that Ho is the curve /'O Condition

H1 is the curve l'l Condition

Condition

(4)

(3)

(2)

H ( s, t ) .

says that

says that all the curves Hs begin at P.

says that all the curves Hs terminate at Q. The homotopy

amounts to a continuous deformation of l'o to l'l with the endpoints fixed


and all curves in the process restricted to lie in W.
We introduce one last piece of terminology:

Definition 10.3.2.

Let W be a connected open set and let

function element in W. Let P be the center of

U.

unrestricted continuation in W if there is an analytic

of

(J, U)

(J, U) be a
(J, U) admits
continuation (ft, Ut)

We say

along I' for every curve I' that begins at P and lies in W.

10.3. The Monodromy Theorem

309

One situation, in practice the primary situation, in which the question


raised at the beginning of this section always has an affirmative answer is
given by the following theorem:

Theorem 10.3.3 (The monodromy theorem). Let W C be a connected

(!, U) be a function element, with U W. Let P denote the


U. Assume that (J, U) admits unrestricted continuation
in W. If /'o, 1'1 are each curves that begin at P, terminate at some point
Q, and are homotopic in W, then the analytic continuation of (J, U) to Q
along 'Yo equals the analytic continuation of (J, U) to Q along 1'1.
open set. Let

center of the disc

Proof. Write

fs,t

for the analytic function element at H(s,

analytic continuation of

1-+

fs,t,

(J, U)

along the curve

H(s,

1-+

t).

t)

obtained by

In other words,

t
Ji,1

for s fixed, is an analytic continuation along the curve

In this notation, the desired conclusion of the theorem is that

1-+

t).
fo,1.

H(s,

We shall prove this assertion by a method similar to the "continuity


argument" used in the proof of Proposition 10.2.2. Namely, set
S

{s E [O, 1] : for all,\ E [O, s],

f>...1

fo,i}.

Clearly S is nonempty since 0 E S. If we can prove that S is (relatively)


open in [O, 1] and also that S is closed, then the connectedness of [O, 1] will
imply that S

[0,1]. Then it follows that 1 ES and hence

!1,1

fo,1,

as

we must show in order to establish the theorem.


To prove that S is open in [O, 1], suppose that
that fixed point

so,

so

there is an E > 0 such that, for each

of convergence of the power series expansion of

fso,t

ES. Note that, for

E [O, 1], the radius

around

H(so, t)

is at

least E. This assertion follows easily from the compactness of [O, 1] and the
definition of analytic continuation along curves (exercise). Now choose

so small that, for all s E (so


that I H(s,

t)

- H (so,

t) I

- 8, so + 8)

n [O, 1] and all

- 8, so+ 8)
fs,t

> 0

.fs,t

along H(s,

t),

n [O, 1], is obtained by setting

the unique function element at H(s,


a direct analytic continuation of

the direct analytic continuation of fso,O

t)

that is

fso,t

(J, U), the uniqueness


of analytic continuation (Proposition 10.2.2) implies that h,t
fs,t for all
t E [O, 1]. Now h,1 is the direct analytic continuation of !so.I fo,1 Hence
fs,1
fo,1 Thus every s E (so - 8,so + 8) n [O, 1] belongs to S, and S is
Since

fs.o is

E [O, 1], it holds

< E. This is possible by the uniform continuity of

H. Then it is easy to see that an analytic continuation


s E (so

indeed open in [O, 1].


To show that S is closed in [O, 1], suppose that {Sj} is a convergent
sequence of points in S. Let

so

limj Sj. By the same argument that we

10. Analytic Continuation

310

8
fs,l is a direct analytic
( so - 8, so + 8) n [O, 1],

just used, we can choose


then

Sj

So

so

S and Sis closed.

- 8, so + 8) n [O, 1],
continuation of fso,l Since, for j large enough,
fsJ.I
fo,1
it follows that, for such j, fso,l
>

0 such that if s E ( so

As noted, the facts that Sis nonempty, open in [O, 1], and closed imply

the conclusion of the theorem.

Corollary 10.3.4. Let W

set. Assume further

that W is

that any two curves

C be a connected open
topologically simply connected, in the sense

that begin at the same point and end at the same point (possibly different
from the initial point) are homotopic.

Assume that (!,

U)

admits unre

stricted continuation in W.

Then there is a globally defined holomorphic

function F on W that equals

on

U.

In view of the monodromy theorem, we now can understand specifically


how it can be that the function y'Z, and more generally the function log z,
cannot be analytically continued in a well-defined fashion to all of

C\

{0}.

The difficulty is with the two curves specified in the second paragraph of
this section: They are not homotopic

in the region C \

{O} .

10.4. The Idea of a Riemann Surface


In this section we give an intuitive description of the concept of what is called
a Riemann surface . We content ourselves with a descriptive treatment, and
we refer to [AHL2, pp. 287-290] and [GUN] for an introduction to the sheaf
theoretic approach, which is closely related to the idea of global analytic
functions that we have already discussed. See also [SPR] for an approach
via the geometric idea of a surface.
The idea of a Riemann surface is that one can visualize, or make geo
metric in some sense, the behavior of function elements and their analytic
continuations. At the moment, a global analytic function is an analytic ob
ject. As defined in Section 10.1, a global analytic function is the set of all
function elements obtained by analytic continuation along curves (from a
base point P E

C)

of a function element (!,

U)

at P.

Such a set, which

amounts to a collection of convergent power series at different points of the


plane

C,

does not seem very geometric in any sense. But in fact it can be

given the structure of a surface, in the intuitive sense of that word, quite
easily.

(The precise and detailed definition of what a "surface" is would

take us too far afield: We shall be content here with the informal idea that
a surface is a two-dimensional object that locally "looks like" an open set in
the plane. A more precise definition would be that a surface is a topological
space that is locally homeomorphic to

C.)

10.4. The Idea of a Riemann Surface

311

The idea that we need is most easily appreciated by first working with a
few examples. Consider the function element

(J, U)

defined on

D(l, 1)

by

z
where r >

0 and -7r /2

< () <

rei9 f---+ rl/2ei9/2 '

7r /2

makes the rei9 representation of

D(l, 1) unique. This function element


z
1 that we have already discussed.

The functional element

be analytically continued along every curve / emanating from

C \ {0}.

is the "principal branch of ,fi" at

in

(J, U)

can

and lying

Let us denote by R (for "Riemann surface") the totality of all

function elements obtained by such analytic continuations.

Of course, in

set-theoretic terms, R is just the global analytic function ,Ji, just as we


defined this concept earlier. All we are trying to do now is to "visualize" R
in some sense.
Note that every point of R "lies over" a unique point of
function element

(J, U)

R is associated to the center of

function element at a point of

7r : R

--t

C \ {0}

C \ {O}.

C \ {O}: A
U; (J, U) is a

So we can define a "projection"

by

7r( (J, U))

the center of the disc

U.

This is just new terminology for a situation that we have already discussed.

7r of R is two-to-one onto C \ {0}. In a neighborhood


C \ {O}, there are exactly two holomorphic branches of ,Ji.
[If one of these is (J, U), then the other is ( f, U). But there is no way to
decide which of (J, U) and ( f, U) is the square root in any sense that can
be made to vary continuously over all of C \ {0}.] We can think of R as a
The projection

of a given

"surface" in the following manner:


Let us define neighborhoods of "points"
neighborhood of

(J, U)

(J, U)

in R by declaring a

to be

{ (Jp , Up) : p E U

and

(Jp , Up)

is the direct

analytic continuation of

(J, U)

to

p} .

This new definition may seem formalistic and awkward.

7r : R

But it has the

C \ {O} locally one-to-one. Every


(J, U) has a neighborhood that maps under 7r one-to-one onto an open subset
of C \ {O}. This gives a way to think of R as being locally like an open set

attractive property that it makes

--t

in the plane.
Let us try to visualize R still further. Let
Then

7r-1(W)

C\ {z

x+iO: x O}.

decomposes naturally into two components, each of which is

an open set in R. [Since we have defined neighborhoods of points in R, we


naturally have a concept of what it means for a set to be open in R as well:
A set is open if it contains a neighborhood of each of its points.] These two

312

10. Analytic Continuation

Figure

__

10.7

__

Figure

10.8
71"-i(W)
7r-1(W), the

components are "glued together" in R itself: R is connected, while


is not.

Note that, on each of the connected components of

projection

7r

is one-to-one.

All of this language is just a formalization of

W, there are two holomorphic branches of viz-namely


rei8 1--t -ri/2ei8/2, -7!" < () < 7r.
Each of the components of 7r-i(W) can be thought of as a "copy" of
W, since 7r maps a given component one-to-one onto W. See Figure 10.7.
How are these "copies", say Qi and Q2, glued together to form R? We join
the second quadrant edge of Qi to the third quadrant edge of Q2 and the
second quadrant edge of Q2 to the third quadrant edge of Qi. Of course
the fact that, on

rei8

1--t

rl/2ei8/2

and

these joins cannot be simultaneously performed in three-dimensional space.


So our picture is idealized. See Figure

10.8.

Tacitly in our construction of

R, we have restored the negative real axis (but not the point
We have now constructed a surface, known

as

0).

the "Riemann surface for

the function viz." T his surface that we have obtained by gluing together the
two copies of

is in fact homeomorphic to the topological space that we

made from R (the set of function elements) when we defined neighborhoods


in R.

So we can regard our geometric surface, built from gluing the two

copies of

together, and the function element space R

as

being the same

thing, that is, the same surface.


Since

7r

: R

-t

C \ {O}

is locally one-to-one, we can even use this

projection to describe what it means for a function F : R


morphic. Namely, F is holomorphic if F
for each open set U in R with

7r

7r-i : 7r( U)

-t

-t

to be holo

is holomorphic

one-to-one on U. W ith this definition in

10.4. The Idea of a Riemann Surface

313

mind, ../Z becomes a well-defined, "single-valued" holomorphic function on


R. Namely, if (!, U) is a function element in R, located at P E C \ {O},
that is, with Tr(j, U)
P, then we set
=

F((f, U))

f(P).

In this setup, F2((f, U))


7r(f, U) [since f2(z)
z]. Therefore F is the
square root function, in the sense described. There are similar pictures for
ylz-see Figure 10.9.
=

It will require some time, and some practice, for you to become ac
customed to this kind of mathematical construction. To help you to get
accustomed to it further, let us now discuss briefly the Riemann surface
for "log z." More precisely, we begin with the "principal branch" rei8 1---t
log r + iO defined on D( 1, 1) by requiring that -Tr /2 < () < 7r /2, and
we consider all its analytic continuations along curves emanating from 1.
We can visualize the "branches" here by noting that, again with W =
C\ ( {O}U the negative real axis), 7r-l (W) has infinitely many components
each a copy of W-on which 7r maps one-to-one onto W. Namely, these
components are the "branches" of log z on W:
rei8

1---t

log r + iO + 2Trik,

Z,

where -Tr < () < Tr. Picture each of these (infinitely many) images stacked
one above the other (Figure 10.10). We join them in an infinite spiral, or
screw, so that going around the origin (counted counterclockwise in C \ { 0})
corresponds to going around and up one level on the spiral surface. This is
the geometric representation of the fact that, when we analytically continue
a branch of log r + i() + 2Trik once around the origin counterclockwise, then
k increases by 1 (Figure 10.11). This time there is no joining of the first and
last "sheets." The spiral goes on without limit in both directions.
The idea that we have been discussing, of building surfaces from func
tion elements, can be carried out in complete generality: Consider the set of
all analytic function elements that can be obtained by analytic continuation
(along some curve in C) of a given function element (!, U). This is what
we called earlier a complete or global analytic function. Then this set of
function elements can actually (and always) be regarded as a connected sur
face: There is a projection onto an open set in C obtained by sending each
function element to the point of C at which it is located. This projection is
a local identification of the set of function elements with part of C, and so it
in effect exhibits the set of function elements as being two-dimensional, that
is, a surface; after this observation, everything proceeds as in the examples.
The reader is invited to experiment with these new ideas in Exercises 1-4,
8, and 17, in which there are more complicated function elements than ../Z
and log z.

10. Analytic Continuation

314

-------==-

--------==-

-------==-

::=:::
_
::>

_____

Figure 10.9

Figure 10.10

10.5. The Elliptic Modular Function and Picard's Theorem


Introductory Remarks
Liouville's theorem tells us that a bounded, entire function is constant. But,
as we in effect showed earlier, a stronger result holds:
Theorem.

Let f : C - C be a nonconstant holomorphic function.


Then the range of f must be dense in C.

10.5.

Picard's Theorem

315

Figure 10.11
Proof. If the range off is not dense, then there is a disc

D(P, r )

Range J

D(P, r ) such that

0. Define
g(z)

r
=

f(z) - p

Then g is entire and lg(z)I :::; 1 for all z. By Liouville's theorem, g is constant
and hence so is f.
0
In fact a much more precise and striking result is true; it is known as
the "little theorem" of Picard:
Theorem 10.5.1. If the

range of a holomorphic function f : C


two points of C, then J is constant.

---+

C omits

The entire function f(z)


ez shows that an entire function can omit
one value (in this case, the value 0). But the theorem says that the only
way that it can omit two values is if the function in question is constant.
=

There are many ways to prove Picard's little theorem. We shall use
the classical method of constructing a special function known as the elliptic
modular function. This rather lengthy construction will occupy most of the
remainder of this section. There are briefer methods (see [LANG]) which are
simultaneously more elementary and more cryptic. There are also methods
using the ideas of Nevanlinna theory (see [SAZ]). The method presented
here is easily understood and also gives the reader a glimpse of what are
called Fuchsian groups and of functions that commute with a group action.
Our discussion rather closely follows that in [RUD2].

316

10. Analytic Continuation

The Action of the Modular Group


Consider the set H of all conformal self-mappings of the upper half
plane U

{z

Im z >

0}.

The set of such mappings is closed under

composition, each element has an inverse, and the group law (composition)
is associative; these are obvious assertions. So H is indeed a group.
Informally, we can determine the structure of this group as follows (for
a formal determination, use Theorem

6.2.3

and the Cayley transform). If

E H, then by Schwarz reflection, f may be analytically continued to a


;._
mapping of C to itself. (The informality comes from not exactly knowing

that the Schwarz reflection applies!) An examination of the reflection oper


ation reveal that the extended function (still called J) will be a conformal
self-map of C. We know (see Theorem 6.3.5) that f must therefore be given
by a linear fractional transformation
f(z)

az + b
.
CZ+ d

Since

and

must all be real (after multiplication of all four numben; by the same

must preserve the real axis, it is not difficult to verify that

a, b, c,

constant if necessary, which leaves the function f unchanged). Writing f


f(z)

[ac I z 12 +

bd + (be+ ad) Re z] + i [Im z (ad - be)]


,
lcz + d l2

we see that if we want {z: lmf(z) >


have that

ad - be> 0.

{z: Imf(z) >


if

O}.

as

Furthermore, if

O}

{z: lm z >

ad - be> 0,

O},

then we must

then f ( {z : Im z >

0})

That is, f maps the upper half plane to itself if and only

ad - be> 0.
Therefore a complete description of the group H is that it is the set of

all linear fractional transformations


f(z)
such that

a, b, c, d

JR and

az + b
CZ+ d

ad - be > 0.

As noted, this group can be de

termined formally by using our complete description of the conformal self


mappings of the disc (Section

6.2)

and transferring this information to U by

means of the Cayley transform C(z)


We now focus on the so-called

i(l -

z)/(1 + z).

modular group.

This is the subgroup G

of H consisting of those linear fractional transformations


f(z)
with

a, b, c,

and

real

integers

and

az + b
CZ +d
ad - be

1. Check for yourself that

the composition of two elements of G is still an element of G and that the

317

10.5. Picard's Theorem

inverse of such an

is

1-1 ( z )

dz - b
--__
-cz + a'

hence still a member of G.

A subgroup K of G is called nontrivial if it contains elements other


than the identity.

Definition 10.5.2. A function f

---> C

there is a nontrivial subgroup K G such that, for all

(f

)(z)

modular function if
E K and all z E U,

is called a

f(z).

In words, a modular function is one that is invariant under the action


of some nontrivial subgroup of G.

The Subgroup r
For our purposes it is useful to concentrate on the subgroup of G that
is generated by the two elements

(z)

z
22 + 1

w(z)

and

z + 2.

It is traditional to denote this group by r. The functions and w are plainly


elements of G. By the subgroup generated by and w we mean the intersec
tion of all subgroups of G containing these two elements. This is the same
1
as the group that is obtained by considering all finite products of, w,- ,
1
and w- in any order. The linear fractional transformations that belong to

r have the form


Zt---->

with

a, d

az + b
--CZ + d

odd integers and b, c even integers, as you can check for yourself-

see Exercise

16.

It is useful at this point to proceed by analogy. The additive group Z


(the integers) acts on the real line IR. in a natural way: To each integer

there corresponds a mapping

c/>m : JR.
given by

<Pm(x)

x+m.

--->

The mappings

JR.

<Pm form

a group under composition

which is canonically isomorphic to the original group Z. This group action


has what is known as a

fundamental domain given by I


[O, 1). W hat we
m(/) is distinct, the sets are pairwise disjoint,
IR.. We think of the action of the group { m }mEZ as
=

mean by this is that each set


and their union is all of

moving the interval I around to tile, or fill up, the real line.
In an analogous fashion, we shall have a fundamental domain W for
the action of r on the upper half space:

{z

x + iy: -1 x

<

1, l2z + ll 2: 1, l2z - ll

>

1, y > O } .

10. Analytic Continuation

318

Figure 10.12

This domain is shown in Figure 10.12. Notice that it is bounded on the left
and right by the vertical lines

1 and

-1. It is not bounded above,

but it is bounded below by the upper halves of the circles 8D(-1/2, 1/2)
and 8D(l/2, 1/2). The region

contains both the left vertical bounding

line and the left semicircle; however it does not contain the right vertical
bounding line nor the right semicircle. Note that

contains no points of

the real axis.


Let us formalize our assertion that

W is

a fundamental domain for the

action of r.

Proposition 10.5.3.
(1)
(2)

The modular group has the fallowing properties:

If f,g are distinct elements off, then

f(W) n g(W) = 0.

LJ f(W) = U.
fEf

Proof. To prove (1), notice that it suffices to show that (g-1of) (W) nW=
0. In other words, if e-/=- h Er, then h(W) n W= 0. Write
h(z) =

az+b
.
cz + d

We now consider three cases:

(a)

c = 0. Then

(b)

a and d are integers, we conclude that


a= d= 1. It follows that h(z) = z + 2n for some nonzero integer
n (check the form of , w). But W is only two units wide in the
real direction. Therefore h(W) n W = 0.
=

ad =

1. Since

2d. Then, since

ad - be =

1, we see that

d( a - 2b) = 1. It follows that d = 1; thus

ad -

2bd = 1 or

2. Hence, since

Picard's

10.5.

319

Theorem

d(a- 2b)

1, we see that h( z )

(again recall the form of and

(W)

that

( z )+2m,

w).

where mis some integer

But a Himple calculation shows

D(l/2, 1/2)-after all, the boundary of

of four generalized circles, so we need only look at where

consists

maps

those four circles. Also, (real) translation by an even integer will


move

W to the corresponding component of the complement


W . Again, it follows that h(W) n vV 0.

image of

(c ) c =f.
z

and c =f. 2d.

of the

Then it will follow that

lc z + di

> 1 for all

E vF. If we accept this claim for the moment, then we may use

the formula

l m h(z)
(remember that ad-be

z
lcz +dl2
Im

1) to see that Im h( z ) < Im z for

z EW.

If

there were a z E W nh( W ), then we could apply a similar argument


to

h-1

and conclude that Im

Imh-1(h( z ))

< Im h( z ). These

two inequalities are contradictory and demonstrate that

W are

h(W) and

disjoint.

lc z +di 1
D( -d/ c, 1 / i c l) n W =f. 0. The definition of W
of the points -1,0, 1 must lie in D(-d/ c, 1 / i cl).

It remains to verify the claim. If the claim were false, then


for some

z E W.

But then

now Hhows that at least one

lcw + di

Thus

< 1 for

our hypotheses about

equaling one of -1, 0, 1. But, for such

and

d,

the quantity

cw +d

w,

under

is an odd integer whose

absolute value cannot be less than 1. That is a contradiction. So we have


established the claim that

lc z +di

> 1 for all

This completes the proof of part


Now we prove part

h Er.

(2).

(1)

z E W.

of Proposition 10.5.3.

Let M equal the union of all the sets

h(W),

Clearly M U. Note two things about M:

(i)

The set M contains all sets of the form

wk ( z )

(ii)

Of course,

z + 2k.

Since maps the circle

{z

wk(W), k E Z.

E C : l2z - II

{z

E C : I 2 z + 1I

1} onto the circle

1}, we see that M contains every

z EU

that

satisfies, for all m E Z,

l2z - (2m + 1)1

1.

(t)

w E U . We shall find a point z EW such


h Er. Notice that if n > 0, then there are
only finitely many pairs c, d such that lcw + di o. As a result,
we may select an element h Er such that lcw +di is minimal. By
equation ( * ) , we conclude that
Now fix a point

that

h(z)

for some

Im h( w )

Im h( w)

320

10.

for all

h E r.

Let

z = h( 111 ) .

becomes
for all

11-1

Analytic Continuation

Then the last displayed inequality

Im h(z):::; Im z

h E r.

We apply this last line to

h=

w
n. Note that the first of these is given by
o
w

-n and to

'T} -2n
-n ( )=
w

TJ
2ry-4n+l

h=

( ** )

and the second is given by


-1

Using our formula

-n

(*)

(IJ

)=

'T} -2n

_2ry+4n+l

for Imh together with

may thus conclude that, for all

l2z-4n+lll
It follows that

/i-l

( t);

( ** )

and

( *** ) ,

we

l2z-4n-ll 1.

and

satisfies

E f, we find that

n E Z,

hence

z E M.

Since

=h-1 (z) and

EM.
D

That completes the proof.

The Modular Function


Now we come to the construction of the particular modular function, known
as the

elliptic modular function,

theorem .

that we shall use to prove Picard's little

In the exercises, we shall explore applications of this modular

function to normal families and to other topics in complex function theory.

10.5.4. With r and W


holomorphic 011 U such that

Theorem
>.

( 1) ). o h = ). for every h
(2)

>.

as

chosen above, there exists a function

E f;

is one-to-one on W;

(3) the range

N of>.

is

\ { O, 1};

(4) the function >. cannot be analytically continued to any open set
that strictly contains U.
Proof. Let us denote the right half of

W' ={ z

EW

W
:

by W':

Re z

>

0}.

By the Riemann mapping theorem there is a one-to-one, onto, holomorphic


function

W'

____,

U,

g(O) =0, g(l) = 1, g(oo) =oo. This g extends to


W' onto U. [Remember that U is

be a one-to-one, continuous mapping of

conformally equivalent to the disc via the Cayley mapping.

We are using

Caratheodory's theorem that a conformal mapping of a region bounded


by a Jordan curve to the disc extends continuously and one-to-one to the

10.5. Picard's Theorem

321

boundaries of the respective domains; this theorem will be given independent


treatment in Section

13.2.]

Now the Schwarz reflection principle may be applied to reflect the func
tion

in the

axis: We set

a continuous function on

\ {x +iO: x O}.

g(-x + iy) = g(x +iy).

Further,

This extends

which is a holomorphic mapping of

g is

one-to-one on Wand

g(W)

to be

W onto
C \ {O, 1}.

Now observe that, by the nature o f the Schwarz reflection and the
definitions of and

w,

we have that

g(-l+iy)=g(l+iy)=g(w(-l+iy)) ,

O<y<oo,

and

We could at this step analytically continue


flections in

8W.

g to all of U

by repeated re

This program is carried out in [ LANG ] , for instance. Instead

we exploit the action of the group r on U . Set


for all

E h(W),

This function>. is well defined because any given


only one

all

h E f.

lies in

h(W) for one

and

h Er.

Since the function


lishes property

( 3).

has range C

\ { 0, 1},

then so does >.. That estab

The periodicity property

definition of>.. That >. is one-to-one on

( 1)

is immediate from the

is immediate from the construc

tion. Also we know automatically that >. is holomorphic on the


each of the sets

h(W).

interior of
3 that a

But we know from Exercise 6 of Chapter

function holomorphic off a smooth curve and continuous across the curve is
holomorphic across the curve. It follows that >. is holomorphic on all of U.
It remains to check property

(4).

Observe that the set of numbers X =

{b/d: b, d E Z, b even, d odd } .

{ h(O)

h E r}

equals the set

Thus Xis dense in JR, the topological bound

ary of U. If >. could be analytically continued to a neighborhood of some


boundary point p of U, then p would be an accumulation point of the zero
set of >. and hence >. would be identically zero.

This is plainly not true.

Thus>. cannot be continued analytically to any open set larger than U.

The Little Picard Theorem


Now we come to the main application of the modular function for this
section. A formal statement of the result is this:

322

10. Analytic Continuation

Theorem 10.5.5 (Picard's little theorem). Let


suppose that the range of
Then

be an entire function and

om i ts two distinct complex values a and

(3.

must be identically constant.

Proof. The idea of the proof is this: We try to write


of some holomorphic function K
that a = 0, /3 =

1).

<C

U with..\

____,

____,

Since K is entire, it follows that K

as the composition

<C
:

\ {O, 1}

<C

____,

(assuming

U is constant

and hence so is J.
For the details, begin by noticing that we can take a

0 and f3 = 1, by
[f ( z) - a ] / [(3 - a] . Let be a disc lying in <C \ { 0, 1}.
Then there is an open set V <;;;; U (indeed there are infinitely many such Vas
soon as there is one) such that..\(V) = and..\ is one-to-one on V. Observe
that V will intersect at most two of the fundamental regions h(W). Let p
=

simply replacing J with

denote the inverse of

.Alv

If ' is another disc in <C

\ { 0, 1}

that has a nontrivial intersection

with , then it will have corresponding to it a domain

V' n V =/= 0

and..\ maps

V'

<;;;; U such that

conformally onto'. The inverse mapping

V'

be a direct analytic continuation of

p.

p'

will

This process may be continued.

Now we begin the construction. There is a disc


that

Eo,

f(Vo)

Vo with center 0 such


Eo <;;;; <C \ {O, 1}. Choose po as above, defined on
Set k =Poof on Vo so that ..\ok
f.

lies in some disc

to be the inverse of..\.


If 'Y

[O, 1]

____,

f o'Y is compact.
Vi as above such

<C is a curve beginning at 0, then of course the image of


We can cover the image of 'Y with a finite chain of discs
that each

may choose functions

Pi,

lies in a disc i <;;;; <C

\ { 0, 1}.

Also, we

defined on i, that are inverse to ..\. Each

be inverse to ..\, and each

(Pi-1.Ei-1).

f(Vi)

(pi,i)

Thus we obtain an analytic continuation K of

Thus the function element

Pi

is to

is to be a direct analytic continuation of

(k, Vo)

(k, Vo)

along

'Y

has the property of unrestricted an

alytic continuation in the domain consisting of the entire complex plane.


Of course the complex plane is simply connected. Therefore, by the mon
odromy theorem, the function element continues to all of <C. F inally, note
that all of the function elements

(Pi of, Vi)

take values in the upper half

plane U. Hence so does K.


Now ..\ ok = J on

Vo

by construction.

Hence ..\ o K

J on <C.

But

K maps <C to U, so K is constant (by the first theorem in this section).


Therefore

is constant.

That concludes the proof.

Remark: It is worth noting that the naive approach of analytically con


tinuing the function element

p-

(p, Eo)

cannot work. Apart from the fact that

=..\ is infinite-to-one, it is also the case that the (potential) domain for

10.6. Elliptic Functions

p is

the set

C \ { 0, 1}

323

which is not simply connected. W hat we do instead

in the proof is to continue analytically the function element


this context we are working on the domain

(po

f, Vo); in

and the monodromy theorem

does apply.
The following theorem , known as the big theorem of Picard, strengthens
Theorem 10.5.5 (the little theorem).

Theorem 10.5.6 (Picard's big theorem). Let U be a region in the plane,

U, and suppose that f is holomorphic on U

singularity at

P.

If

f. >

\ {P}

and has an essential

0, then the restriction of f to Un

D(P, E ) \ {P}

assumes all complex values except possibly one.

A discussion and proof of the big theorem can be found in [KRA3].


Compare Theorem 10.5.6 with the theorem of Casorati and Weierstrass
(Theorem 4.1.4), which says that , in a deleted neighborhood of an essential
singularity, a holomorphic function assumes a dense set of values. Picard's
big theorem refines this to "all values except possibly one."
W hat is the connection between the big theorem and the little theorem?
A nonconstant entire function cannot be bounded near infinity, or else it
would be bounded on

and hence constant. So it has either a pole or an

essential singularity. In the first instance, the function is a polynomial (see


Section 4. 7).

But then the fundamental theorem of algebra tells us that

the function assumes all complex values .

In the second instance, the big

theorem applies at the point oo and implies the little theorem.

10.6. Elliptic Functions


It is a familiar idea that the sine and cosine functions are periodic functions
of a real variable:

for all real x.

sin(x+

2n)

sin x,

cos(x+

2n)

cos x

Actually, these properties hold for all complex numbers z

as well (in Section 3.6 we referred to the fact that such identities remain
necessarily true when the real variable is replaced by a complex variable as
the "principle of the persistence of functional relations"):
sin(z +

2n)

sin z ,

cos(z +

2n)

cos z.

We say that sine and cosine are periodic with period

2n.

The exponential

function ex of the real variable xis not periodic, but the complex exponential

10. Analytic Continuation

324

ez

function

is, with period

27ri:

For all

<C,

It is reasonable to ask whether a function could exhibit both phenomena


at once, that is, could a function

have two periods, one real and one

imaginary? Specifically, could one have, for all

f(z + n )

<C,

f(z),
f(z)

f(z +i/])

for fixed nonzero real numbers n and (3? We shall see momentarily that if

is holomorphic, then this double periodicity phenomenon implies that

is constant. But it will turn out that there are nonconstant meromorphic

f : <C

functions

<C

----->

{ oo}

that do satisfy both periodicity properties at

once. Such functions are called

elliptic funct'ions.

We shall concentrate our discussion on the case when the periods are
a =

and

i/3

i.

One could discuss more general ideas: arbitrary n and /3,

of course, or even functions satisfying


where

w1

and

w2

f(z+wi)

f(z) and f(z+w2)

f( z ) ,

are arbitrary nonzero complex numbers. But the particular

case that we shall discuss in detail illustrates most of the interesting features
of such "double periodicity" in general.
periods

w1, w2

with

wif w2

If a meromorphic function has

not real, then the periods are, in a sense to be

made precise later, not dependent; this situation can be analyzed using the
same methods that we are about to use for the case of periods

w1/w2
Exercise 18).

case, where

( see

1, i.

The other

is real, is much less interesting and also less important

For the remainder of the section, we shall consider only those doubly
periodic meromorphic functions

for all

with periods

J(z + 1)

f(z),

f(z + i)

f(z)

1, i:

<C.

The first question to ask about such functions is whether there exist
any nonconstant examples. The answer is not at all apparent. W hen

is

entire, the double periodicity hypothesis implies that, for each z, there is an
element :Zwith 0::; Rez:S
means that

1,0::;

Imz:S

such that

f(z)

f(z).

But this

is bounded and hence constant.

In order to find an example of a meromorphic, doubly periodic func


tion, we shall try just to write one down.
functions with specified poles
riodic

with

artificial as it

Using the idea of constructing

( Section 8.3), we shall look for a doubly pe


poles at { m + ni : m, n E Z}. This special choice is not as
at first appears. Since f has a pole, we can suppose after a

10.6.

325

Elliptic Functions

translation of coordinates that there is a pole at the origin. Then, by the


double periodicity, it would follow that f has poles at { m+ni : m, n E Z}.
It turns out that there is a good reason not to seek an J with simple poles at
{m + n'i} and no other poles; namely there do not exist any such functions
(see Exercise 19).
Instead, we try
f(z)

'""

m,nEZ

(z

1
m

ni)2

Unfortunately, this series does not converge absolutely for any fixed value
of z. The reason is as follows: We can think of the sum (except for one
important term, which we omit) as
=

h=I max(lml,lnl)=h

Then, for max(lml, lnl)

I (z

1
m ni)2

h, h large, the term


(z

1
m n-i)2
-

has absolute value of the order of magnitude 1/h2. But, for fixed h, the
number of such terms is of order h. So the summands of the (outside) sum
over h have order 1/h. But the series

diverges.
We need to modify the series ( * ) so that the terms have order 1/h3.
We have in fact already learned the trick for doing this (Section 8.3). We
look at the series

Since

I
I

(m+ni)2 (z m ni)2
(z m ni)2(m+ni)2
z2+2z ( m+ni )
(z m ni)2(m + ni)2
-

326

10. Analytic Continuation

is of order

1 / h3,

for

1
+
z2

fixed and h large, it follows that,

[ (z-m-ni)2- (m : ni)2]
1

L
m,nEZ

(m,n)f(O,O)

converges absolutely and uniformly on compact sets to a meromorphic func


tion

P(z)

with double poles at

{ m + ni : m, n

Z}.

The trouble is that it is not at all obvious that

is in fact doubly

periodic. Our original, nonconvergent, sum

m,nEZ

1
(z-m-ni)2

is formally doubly periodic, but it does not give a function


does not converge ) . Our new sum

P(z)

m,nEZ

(m,n)f(O,O)

( since

1
1
1
+
(z-m-ni)2 - (m+ni)2
z2

does give a well-defined meromorphic function; but if we replace

( for

instance ) , then we obtain

m,nEZ

(m,n)f(O,O)

by

z+ 1

1
1
1
+
(z-(m-l)-ni)2 (m+ni)2
(z+l)2"

The same "terms" occur: as

m, n

run over all integers, the summands

1
(z-(m- 1 )-ni)2
run over the same set of terms.

l/(m + ni)2 in

the series

different ways.

and

1
(z-m-ni)2

But they are paired up with the terms

Of course we may not rearrange the terms

since

m,nEZ

1
(m+ni)2

(m,n)f(0,0)
and

L
(z-m-ni)2
1n,n
are definitely not absolutely convergent separately. It is the specific pairing
up, that we have given above, that makes the series for

P(z)

rn,nEZ

(m,n)f(O,O)
absolutely convergent.

1
(z-m-ni)2 (m

P,

: ni)2] + :2

10.6. Elliptic Functions

327

The way out of this dilemma is to think about the derivative of P. We


can find the derivative by differentiating the series for P term by term

z not

( for

a pole of P) :

P'(z)

"'"'

rn,n

-2

(z- m-ni) 3

Notice that this is "legal" since the series for P is absolutely and uniformly
convergent on compact subsets of the complement of the pole set of

[ Check

for yourself that the series representation for

P'

P.

is absolutely and

uniformly convergent in the same sense. ]


Now P'(z) is obviously formally doubly periodic: Replacing z by z + 1
or z +

i gives

back the very same series. Hence

91 (z)

P(z + 1 ) - P(z)

92(z)

P(z + i) - P(z)

and
have derivatives that are

0, that is, the functions 91 and 92 are constants.

To determine what the constants are, we note that

91 ( -1 / 2 )
But

P is

P(l/2) - P(-1/2) .

clearly an even function:

P(z)

P(-z)

for all z E C-just look

at the series for P(z) and P(-z) , matching up

[(z- m- ni) 2 - (m : ni) 2]

with

Thus 91

[(-z - (-- n'i)) 2 - (-m ni) 2]

0. Similarly, setting

doubly periodic.

-i / 2 gives 92

As noted, the function P(z) is even, that is,

0. Thus

is truly

P(-z) P(z) for all z.


P'(-z) -P'(z) for
at the formulas for P(z) and
=

On the other hand, the function P' (z) is odd, that is,
all z. Both these assertions follow from looking
P' (z) , respectively. ( Actually, the derivative of
an odd function--see Exercise 14. )

an even function is always

P' could be expressed as


P. The reason for this will become apparent as the story
unfolds. Because P is even while P' is odd, it is impossible for P' to be equal
to a polynomial in P. But (P') 2 is an even function, thus it is conceivable
that (P') 2 could be a polynomial in P. Indeed, this is actually true, as we
We next consider the question of whether

a polynomial in

shall now show.


First let us look at the situation in a neighborhood of
function P' has a Laurent expansion around z

0. The

0 which begins with -2/z3,

10. Analytic Continuation

328
contains no other negative powers of
powers of

z.

Namely,

z,

and contains only odd

-2
P (z)= 3 + C1z+ C3z3+
z
I

( positive )

This form for the Laurent series follows from the formula

-2
.
(z - m - m)3

P'(z)=

m,n

and the fact that

P'

is odd. As a result,

-2
P (z)= 3 (1+Aiz4+A3z6+
z
I

Thus

(P'(z) ) 2=

-Ci

Ai= -- , ....
2

with

(1+2A1z4+2A3z6+

).

Now, by similar reasoning,

P(z)=

1
(1+B1z4+ ..
z2

where the omitted terms are sixth or higher degree

( the

absence of the

term in the parentheses requires some explanation-see Exercise

1
P 2(z)= 2 (1+2B1z4 +
4

20).

z2

Hence

and

1
3
P (z)= 6 (1+3B1z4+
z

),

with the omitted terms, in both formulas, being of degree six or higher.
Thus, if we set

C1= -8A1+12B1 , then F(z) has no negative powers in its Laurent


0. Now F(z) is meromorphic on C and holomorphic on
C \ {m+ ni: m, n E Z}. Moreover F(z) is doubly periodic, since P and P'
are. Finally, F is holomorphic across z= O; we chose C1 so that this would

with

expansion around

happen.

Thus Fis holomorphic across all points m+ni, by its double periodicity.
So F is entire, and bounded since it is doubly periodic. Thus it is constant.
In other words,

for some constant

C 2,

with

C1

as before.

329

10.6. Elliptic Functions

We may think of this as a differential equation for

P,

and we may

"solve" it by writing

P'(z)
=

y'4P3(z) - C1P(z) + C2

Hence

zo

I J4w3 -dwC1w

+ C2

This symbolic manipulation can actually be made precise (Exercise

22).

This calculation shows, at least in outline form, how the "elliptic" function

is related to the integration of integrands of the form

dw
which are the sort that arise in computing the arc length of an ellipse (Exer
cise

15).

In fact the name "elliptic function" arose historically because some

of these functions turn up in connection with the specific integrals involved


in evaluating the arc length of an ellipse.

This is not of direct relevance

here. The interested reader may consult [SIE].


Of course we can find many more doubly periodic functions by just
writing down polynomials (with constant coefficients) in the two functions

P(z)

and

P'(z).

That is,

G(z)

m2'.0,n2'.0
Because

(P'(z))2

Cmn(P'(z))m(P(z)t.

is already a polynomial in

P(z),

any such

G can be written

as

P'(z) (polynomial

in

P) +(some

other polynomial in

P).

What is really surprising is that these generate all the doubly periodic
meromorphic functions (with periods

1 and i)

that exist: Every such function

either has this form or is a quotient of these. In particular, to borrow the


language of algebra for a moment, the field of doubly periodic meromorphic
functions on

with periods

and

is au algebraic extension (by adjoining

P') of a purely transcendental extension of C

by the single function

says that the field has transcendence degree

1.)

would lead us too far afield.

P.

(One

The proof of this assertion

However, this striking fact led historically

to what is known as the algebraic-geometric theory of compact Riemann


surfaces and later to the development of a whole field of transcendental
algebraic geometry. The interested reader may pursue this subject further
in, for example, [SIE].

Exercises

330

Exercises

1.

(a) Show that there is an analytic function element (f, U), U D(O, E)
for some E > 0, with cos J(z) z for all z E D(O, E) and f(O)
?T/2.
=

Show that f is unique except for the choice of E, that is, (f1, U1)
and (h, U2) satisfy Ji = h on U1 n U2 if (f1, U1) and (h, U2) are
as described. [Hint: Notice that cos' z
-sin z and sin(?T/2) f- 0.
So the discussion after the proof of Theorem 5.2.2 applies.]
(b) Show that the function element of part (a) admits unrestricted
continuation on C \ { -1, 1}.
( c ) Recall that the Riemann surface associated to a function element
(f, U) is the set of all analytic function elements obtainable as ana
lytic continuations along some curve emanating from P, P being the
center of U, with this set interpreted geometrically. The Riemann
surface associated to the function element of part (a) is called "the
Riemann surface of arccos" (or of cos-1). Justify this terminology
by proving that if h : D(zo, 6) ---t C, 6 > 0, zo E C, is holomorphic
and satisfies cos h(z) = z, all z E D(zo, 6), then zo E C \ { -1, 1}
and the function element (h, D(zo, 6)) is obtainable by continuing
the element (f, U) of part (a) along some curve. [Hint: Write
cosw = (eiw + e-iw)/2. So if cosw = z, then one can solve for eiw
algebraically.]
=

2. Exercise

1, along with the y'z and log z Riemann surfaces, suggests that

one might try to find Riemann surfaces for the (local) "inverse functions"
of other holomorphic functions. Carry out this construction of "k-1" if
k(z) z + (l/z).
3. Same as Exercise 2, for "m-1" if m(z)
(1 - z 2) 2.
=

4. Same as Exercise

2, for "p-1" if p(z)

z3 + z.

5. What does the monodromy theorem tell us about the function log z?
6. What does the monodromy theorem tell us about the function

..Ji?

7. In view of the ideas discussed ill this chapter, discuss the problem of

unambiguously assigning a time of day to each point on the face of the


earth. Discuss the International Date Line.
8. If you compactify the Riemann surface of the "inverse function" of

a polynomial of degree k (as in Exercises 2-4) by adding in "branch


points", what compact surfaces can you get for k
2, 3? For general
=

331

Exercises

k?

[Hint:

See [SPR] for how to compute the Euler characteristic in the

general case.]

9. Consider the holomorphic function


f(z)

00

1
= L 72'
z1.
J
j=l

Determine the largest open set to which f can be analytically continued.


Can you calculate a closed formula for f?

[If

not , then present the

function in as near to a closed form as you can.]

10. Repeat Exercise 9 for the function

.
00
J(z)=I:
z3

j=l

11. Consider the analytic function

g(z)=

100
0

e -4
1+t

dt.

For which values of z does this formula define gas an analytic function?
Can we use integration by parts to continue the function analytically to
a larger open set?

12. Let

oo

z2j
g(z)=:L7+i

J
l
j=

Determine the disc of convergence for this series. Find the largest open
set to which

g can be analytically continued.

13. Let be a continuous function with compact support on

(0, oo ) JR

(that is, is continuous on all of JR, but vanishes off some compact
set). Define the function

F(z)=

fo00 cp(t)etz dt.

For which values of z is this function well defined and holomorphic?


Can one use integration by parts to extend the domain of the function?
What is its maximal domain of definition?

14. The derivative of an odd function is an even function. The derivative


of an even function is an odd function.

Formulate these statements

precisely, both in the real variable context and in the complex variable
context. Then prove them.

15. Write down the integral that represents the arc length of the ellipse

ax2 + by2

1.

Relate this expression to the so-called elliptic integrals

that occur at the end of Section

10.6.

332

Exercises

16.

Prove that the linear fractional transformations that belong to r have


the form
az + b
Zf---+ -CZ + d
with a, d odd integers and b, c even integers. [Hint: Try induction on
the number of terms in the product of generators.]

17.

Think about the Riemann surfaces of the following "functions" (function


elements). You will need to find subsets of re on which the functions
have well-defined "branches." These will be obtained by removing line
segments or half lines ("branch cuts"). The Riemann surface is then
obtained by "gluing" sheets along the points over the branch cuts:
(a) y'(z - l)(z + 1) [Hint: There are branches on re\ [-1, +1].]
(b) Jz(z - l)(z - 2) [Hint: Use re\ ([O, 1] U [2 , +oo)).]
( c ) V'z(z - 1) [Hint: Three "sheets," branch cuts from 0 to 1 and
from 1 to +oo.]
( d) \;'z(z - 1)(z - 2) [Hint: Three sheets, branches defined over re\
([O, 1] u [1, 2]).]
( e ) log(sin z) [Hint: Infinitely many sheets, branch points at 7rk, k E
Z.]

18.

(a) Prove that if a., /3 E JR, a.=!= 0, fJ=!= 0, then { m a +n/3: m,n E Z}
is either dense in JR or has the form {kw : k E Z} for some w E JR.
[Hint: Consider
inf{ l(ma. +n/3)1 m, n E Z and ma+n/3=!= O}
and distinguish the two cases: (i) infimum > 0 and (ii) infimum
:

= 0. The former occurs when a.//3 is rational, the latter when


a.//3 is irrational; but this information is not needed to prove the
requested result.]
(b) Consider the functions f : re ___. re that are meromorphic and doubly
periodic with real periods a., /3, that is,

f(z +a.)= f(z +/3)= f(z)


for all z E tC. Show that this set of functions consists either of the
constants only or of all the meromorphic functions f : re ---; re such
that f(z +w) = f(z), all z Ere, where w is as in part (a).
19.

Suppose that f : re ___. re u { oo} is a doubly periodic meromorphic


function, with periods 1 and i, that is holomorphic on re \ { m +ni :
m, n E Z}. Prove that the residue off at each m +ni, m, n E Z, is
zero. [Hint: Integrate f around the unit square centered at m +ni, i.e.,
with sides {x + iy: x = m 1/2, n - 1/2 :<:::: y :<:::: n + 1/2}; {x + iy: y
n 1/2, m - 1/2 :<:::: x :<:::: m + 1/2}; etc. Use double periodicity to show
that this integral vanishes.]
=

Exercises

20.

21.

333

Explain why the Laurent series for the P-function around 0 has 0 con
stant term. [Hint: Look at the limit as z - 0 of the series for P(z) 2
(1/z ) obtained from the definition of P.]

Prove that the image under P of C \ {rn + ni : m, n E Z} is all of C


[Hint: Note that double periodicity shows that inf IP(z)I is attained at
some point of { z : z x + iy, 0 :S :T :S 1, 0 :S y :S 1} .]
=

22.

Suppose that w -y(t) is a curve in C along which w3 - C1w + C2 does


not vanish, where C1, C2 are the constants such that
2
(P'(z))
4P3(z) - C1P(z) + C2.
=

Choose, using Exercise 21, a point zo in C \ {m + n'i : m, n E Z} such


that P(zo) -y(O). Then:
(a) Prove that there is a unique curve r in C\ {rn+ni: m, n E Z} such
that P(f(t)) -y(t). [Hint: Since 4w3 - C1w + C2 is nonzero along
'Y, the function P' is nonzero at any point z with P( z ) E image 'Y.
So P has a "local inverse" along -y.]
(b) Prove that, with r as in part (a), we have
1
1
f(l) - r(O) = r(l) - z0
dw,
fr J4w3 - C1w + C2
=

where the branch of J4w3 - C1w + C2 along 'Y is determined by


{./4-y(0)3 - Cn(O) + C2 P'(r(O)) P'(zo).
=

[Hint: The -y-integral is the same as fr l, by change of variable.]

Chapter 11

Topology

11.1. Multiply Connected Domains


Let

U C

be called a

be a connected open set. In this chapter, such an open set will

domain.

[O, 1]

of all curves/ :
once again

11,/2

( as

Fix a point
---t

in Section

such that

10.3)

U.

1(0)

Consider the collection


=

1(1)

C(U)

the relationship between two such curves

given by the property of being homotopic ( at

domain

P. We want to consider

P)

to each other in the

U.

Recall that this means the following: There is a continuous function

H : [O, 1]

[O, 1]

(1) H(O, t)
(2) H(l, t)
(3) H(s, 0)

---t

such that

/o(t) for all t E [O, 1];


/1 (t) for all t E [O, 1];
H(s, 1) P for alls E [O, 1].

The property of being homotopic is an equivalence relation on

C.

We leave

the details of this assertion to the reader. Let the collection of all equivalence
classes be called rt.
Now we can define a binary operation on 1t which turns it into a group.

are curves in C. Then we want to define/ to be


/followed by the curve . Although the particular parametrization

Namely, suppose that/,


the curve

has no significance, for specificity we define/

(r. )(t)

{ 1(2t)(2t

We invite the reader to check that


equivalence classes and

(ii)

1)
(i)

if
if

: [O, 1]

---t

C by

0 :S t :S 1/2
1/2 < t :S 1.

this operation is well defined on

it is associative on equivalence classes-even

335

11.

336

though ('y )
homotopic.

<5

is not literally equal to 1

Topology

( 8), the two are indeed

Second, we declare the identity element to be the equivalence class [e]


containing the curve that is constantly equal to P. If [] is any equivalence
class in Ji, then [c] [Ji] [p] [e] [11], as is easy to see.
Finally, let 1 EC. The curve 1-1 is defined by 1 - 1 (t) 1 ( 1 - t). Both
curves have the same image, and both curves begin and end at P; but 1-1

is ''1 run backwards," so to speak. Check for yourself that the operation of
taking inverses respects the equivalence relation. So the operation is well
defined on equivalence classes. Therefore the inversion operation acts on Ji.
Also, one may check that bl b]-1
b]-1 bl [e] (Exercise 20).

Thus Ji, equipped with the operation , forms a group. This group is
usually denoted by rr1 (U) and is called the first homotopy group of U or,
more commonly, the fundamental group of U. [In conversation, topologists
just call it rr1, or "pi-one. "] The fundamental group is independent, up to
group isomorphism, of the base point P with respect to which it is calculated
(Exercise 22). Thus one can speak of the fundamental group of U.

Let '11 : U ---+ V be a continuous mapping that takes the point P E U to


the point Q E V. Then '11 induces a mapping '11* from rr1 (U) at P to rri(V)
at Q by

Of course one needs to check that the mapping '11* is well defined, and we
leave the details of that argument to Exercise 26 (or see [MCC] or [GG]).
In fact the mapping '11* is a group homomorphism. It can be checked that
if '11 : U ---+ V and <I> : V ---+ W, then

Now suppose that '11 is a homeomorphism; that is, '11 is continuous,


one-to-one, and onto, and has a continuous inverse. Then w-1 also induces
a map of homotopy groups, and it follows from ( * ) that '11* is a group
isomorphism of homotopy groups.
If a domain has fundamental group consisting of just one element (the
identity), then it is called (topologically) simply connected. [We have ear
lier studied the concept of a domain being analytically, or holomorphically,
simply connected. It turns out that the analytic notion and the topologi
cal notion are equivalent when the topological space in question is a planar
domain. This assertion will be proved in Section 11. 3. ] If a domain has a
fundamental group consisting of more than one element, then it is called
multiply connected.

11.1. Multiply

337

Connected Domains

It is immediate from our definitions that a simply connected domain


and a multiply connected domain cannot be homeomorphic. In detail, sup
pose that <I> were a homeomorphism of a simply connected domain U with a
multiply connected domain V. Then <I>* would he a group isomorphism of the
groups 7r1(U) and 7r1(V). But 7r1(U) consists of a single element and 7r1(V)
contains more than one element. This is impossible, and the contradiction
establishes our assertion.
Here is a more profound application of these ideas. Let U be the unit
disc. Then U is simply connected. To see thi8, let P
(0, 0) be the base
point in U. If 'Y(t) (/'1(t), /'2 (t)) is any curve in C(U), then the mapping
=

H(s, t )

((1 - sh1(t), (1 - sh2(t))

is a homotopy of 'Y with the constantly-equal-to-? mapping. This shows that


has just one equivalence class, that is, that 7r1(U) is a group containing
just one element.
1i

Now let V be the annulus {z: 1 < lzl < 3}. Let the base point be Q
(2, 0). Then the important elements of C arc the mappings /'j : [O, 1] ____, V
defined by
'Yj(t) 2e2nijt, j E Z.
=

Notice that 'Yo is the curve that is constantly equal to Q. Also 'Yl is the curve
that circles the annulus once in the counterclockwise direction. F\irthermore,
'Y-l is the curve that circles the annulus once in the clockwise direction. In
general, /'j circles the annulus j times in the counterclockwise direction when
j is positive; it circles the annulus ljl times in the clockwise direction when
j is negative.
It is intuitively plausible, but not straightforward to prove rigorously,
that if j # k, then /'j is not homotopic to 'Yk (at the point 2); this assertion
1
will follow from Corollary 11.2.5, to be proved in the next section. [If bj'Yk" ]
1
is [e], then Corollary 11.2.5 implies that f (1/z) dz over 'YJ'Yk" equals 0, while
1
the integral is easily computed to be 27ri(j - k). Therefore bj 'Yk" ]
[e]
if and only if j
k.] Thus there are (at least) countably many distinct
homotopy classes. It is easy to see that bi] b1]
b2] and, in general,
bi] bj]
bJ] bi]
bJ+il when j 2: 0. Also 'Y-1
bi]-1. It requires
some additional work to prove that every closed curve at the base point Q is
homotopic to one of the 'YJ (and to only one, since the 'YJ 's are not homotopic
to each other)-see Exercises 27, 28, and 29. Once this is shown, it follows
that b1] generates 7r1 (V) as a cyclic group of countably many elements. In
summary, 7r1 (V) is group-theoretically isomorphic to the additive group Z.
=

It follows, in particular, that the annulus is not homeomorphic to the


unit disc: This assertion follows simply from the fact that /'l is not ho
motopic to the constant curve 'Yo, for instance (or by direct comparison of

11. Topology

338

homotopy groups). One can think of the annulus as a domain with one
hole in it. Similarly, there is an obvious, intuitive idea of a domain with
k holes, k 0 (k
0 is just the case of a domain homeomorphic to the
disc). It turns out that any planar domain with k holes is homeomorphic to
any other planar domain with k holes; also, a domain with k holes is never
homeomorphic to a domain with f holes if k =I- f. This statement is difficult
to make precise, and even more difficult to prove. So we shall leave this
topic as a sort of invitation to algebraic topology, with some further ideas
discussed in Section 11.5.
=

11.2. The Cauchy Integral Formula for


Multiply Connected Domains

In previous chapters we treated the Cauchy integral formula for holomorphic


functions defined on holomorphically simply connected (h.s.c.) domains. [In
Chapter 4 we discussed the Cauchy formula on an annulus, in order to aid
our study of Laurent expansions.] In many contexts we focused our attention
on discs and squares. However it is often useful to have an extension of the
Cauchy integral theory to multiply connected domains. While this matter
is straightforward, it is not trivial. For instance, the holes in the domain
may harbor poles or essential singularities of the function being integrated.
Thus we need to treat this matter in detail.
The first stage of this process is to notice that the concept of integrating
a holomorphic function along a piecewise C1 curve can be extended to apply
to continuous curves. The idea of how to do this is simple: We want to define
fl' J, where 'Y : [a, b] --t C is a continuous curve and f is holomorphic on a
neighborhood of 'Y ([a, b]). If a
< ak+I = b is a subdivision
a1 < a2 <
of [a, b] and if we set 'Yi to be 'Y restricted to [ai, ai+i], then certainly we
would require that

Suppose further that, for each i


1, ... , k, the image of 'Yi is contained in
some open disc Di on which f is defined and holomorphic. Then we would
certainly want to set
=

J f

1-ri

Fi('Yi (ai+1)) - Fi('Yi (ai )),

where Fi : Di --t C is a holomorphic antiderivative for J on Di. Thus fl' f


would be determined. It is a straightforward if somewhat lengthy matter
(Exercise 36) to show that subdivisions of this sort always exist and that the
resulting
definition of ,e f is independent of the particular such subdivision
.
:rl'
used. This definition of integration along continuous curves enables us to

11.2. The Cauchy Integral Formula

339

use, for instance, the concept of index for continuous closed curves in the
discussion that follows. [We take continuity to be part of the definition of
the word "curve" from now on, so that "closed curve" means continuous
closed curve, and so forth.]
is important:

This extension to continuous, closed curves

A homotopy can be thought of

as

a continuous family of

curves, but these need not be in general piecewise C1 curves, and it would
be awkward to restrict the homotopy concept to families of piecewise C1
curves.
Definition 11.2.1. Let

be a connected open set. Let 'Y:

for all points PE

[O, 1]

-----+

homologous to 0 if Ind-y(P)

continuous closed curve. We say that 'Y is

be a
=

C \ U.

The intuition here is just this: Suppose that 'Y is a simple closed curve
in

U.

Intuitively, one expects the complement of 'Y to have two components,

a bounded "interior" and an unbounded "exterior",

as

indeed it does (see

the Jordan curve theorem in [WHY]). If the simple closed curve 'Y encircles
a hole in

U,

then it is not homologous to zero. For if c is a point of

C\U

that lies in that hole, then the index of 'Y with respect to c will not be zero.
On the other hand, if 'Y does not encircle any hole, then any point in
would lie

outside the closed curve. Also, the index of

C\U

'Y with respect to that

point would be 0-see Exercise 9. Thus, intuitively, a (simple) closed curve

is expected to be homologous to 0 in
any hole in

if and only if it does not encircle

U.

Definition 11.2.2. A connected open set

closed curve in

U is homologically trivial if

every

is homologous to 0.

It is natural to wonder whether a curve that is homotopic to a point


is then necessarily homologous to zero, or vice versa. In fact it is easy to
see that a "homotopically trivial curve" 'Y is indeed homologous to zero (cf.
Exercise 23 and also Exercise 30).

U C is a connected open set and 'Y is a closed curve


U, that is homotopic to the constant curve at P, then
to 0. In particular, if U is simply connected, then it is

Lemma 11.2.3. If
in

U,

based at P E

'Y is homologous

homologically trivial.
Proof. Let

H(s, t) be a homotopy of 'Y to a point PE U. Set H5(t)

U.

Let c be a point in the complement of


Ind-y(c)
by definition. Rewrite this

1
.

2m

'Y

Then

1
--

d(

as

1
Io=-.
2m

Ho

--d(.
( c
-

H(s, t).

40

11. Top ol ogy

It is easy to check that the value of


ls

1
1
__ J _ _ d(
27ri JHs (
- c

is a continuous function of s; since it is integer-valued, it must be constant.


Whens is sufficiently near to

1,

then the curve

H(s,

) will be contained in a

small open disc contained in U and centered at P. But then it is immediate


that /8

0. Since the expression is constant in

s,

it follows that Io

0.

As noted, a closed curve that is homotopic to a constant curve is ho


mologous to 0. But the converse is not true in general. It is not always true
for individual curves that a curve that is homologous to zero is homotopic
to a point. What is true is that if every curve in the domain U is homolo

gous to 0, then every curve in the domain is also homotopic to a point i. e. ,

homotopic to zero . We shall say something about this matter in Section

11.4.

In particular, we shall prove that a connected open set U <;;: C is

( topologically )

simply connected if and only if it is homologically trivial see

11.4.1).

Theorem

The main result of this section is the following theorem.

The proof

that we present uses an idea of A. F. Beardon, as implemented in AHL2 .

Theorem 11.2.4

( Cauchy

integral theorem for multiply connected do

mains . Let U <;;: C be a connected open set and suppose that .f is holomor

phic on U. Then

i .f(z) dz=

for any curve 'Y in U that is homologous to zero.


Combining Lemma 11.2.3 with Theorem

11.2.4

y ields the following

corollary:

Corollary 11.2.5

: U

--t

( Homotopy

ven;ion of the Cauchy integral theorem . If

C is 110lomorphic and i f 'Y :

[O, 1]

--t

C is a closed curve at P

that is homotopic to the constant curve at P, t h en

f'Y J ( z) dz

0.

This corollary can be proved directly, without reference to Theorem

11.2.4;

a direct proof is given in Exercises 35-37.

Proof of Theorem 11.2.4. It is convenient to treat first the case of U


bounded. Let 'Y be a curve that lies in U. Let > 0 be the distance of the
image of 'Y to C

U. Let 0 < 8 < . 2.

Cover the plane by a mesh of closed squares, with sides parallel to the
axes, disjoint interiors, and having side length 8. Let

{ Qj }f=1

be those closed

squares from the mesh that lie entirely in U. Let Uli denote the interior of

11.2. The Cauchy Integral Formula

341

'

'

_,.

\
u

\
I

'"

\
\

(
I

I
"

"'

"'
i\
I J
\.Y

I'\

\
'

I
I

I"
.. \
\

r-..

Figure 11.l

the union of the finitely many closed squares

Q1, ... , QI<. Clearly,

from the

choice of 8, the image of 'Y lies in U15. Let C15 denote the boundary of U15.
Refer to Figure 11.1.
We orient C15 as follows:

QJ with the
QJ meet, then

Equip each

orientation. W hen the sides of two of the

counterclockwise
integration along

those two sides cancels. The edges that remain-that is, whose integrations
do not cancel out-comprise the boundary C15 of U15. Each of those edges is
oriented, and their orientations are consistent. Take that as the orientation
of C8, so that integration over C15 is defined.
Let

QJ.

U15. Then c lies in some square

Also, there must be a point x E

segment joining c to x lies entirely in


not intersect U15. Now Ind1(x)

=0

Q
Q

U15. In particular

conclude that Ind1(c)

=0

which is not one of the

that does not lie in U.

The line

and therefore, in particular, does

by hypothesis. By the continuity of the

integral, it also must be the case that Ind1(c)


element of U

= 0.

Now c was an arbitrary

could be an arbitrary element of C15. We

for every c E C15.

Suppose that z E U15 is any point.


lies in the interior of some square

QJo.

Suppose also at the outset that z


[The case that

lies on the edge of

some square will follow afterward by the continuity of the integral.] Then,
by the Cauchy integral formula for squares, and for
squares that make up U15,

1
_
27ri

J f(() d( =
J0Q1 ( - z

{ f(z)
O

if
if

Q1

J =J o
J i- Jo.

any of the chosen

11. Topology

342

Summing over

j,

and noting the cancellation of the integrals over edges of

the squares that lie in the interior of U0, we see that

J
27ri lc

!(()

( - z

d(= f(z).

Now integrate both sides in a complex line integral over z E 'Y We


obtain

J f(z) dz= J

fr

fr

J !) )
(
27ri lc
6

.,

d(

dz.

We may apply the version of Fubini's theorem that appears in Appendix


A, since the integrand is a jointly continuous function of both its arguments
(see Exercise

39).

Therefore we find that

J f(z) dz= J J(()


J
27ri fr
lc
6

fr

.,

dz

d(.

But of course the integral inside the parentheses is just the (negative of
the) index of the curve 'Y about the point ( E C0. This has already been
established to be 0. As a result,

f(z) dz= 0

as desired.
In case U is not bounded, we may take the intersection of U with a
large disc that contains 'Y Then the proof proceeds essentially as before,
with the additional observation that Ind'Y(z)= 0 for z outside this disc and
hence Ind'Y(z)= 0 for all z in the complement of Un (the disc).

As discussed earlier, we are looking for a Cauchy integral formula for


multiply connected regions; in such a formula, the index ought to be used to
take into account the possibility that the curve of integration goes around
one or more holes. Such a formula in fact follows directly, as it did in the
case of the disc and the square, from the Cauchy integral theorem:

Theorem 11.2.6 (Cauchy integral formula for multiply connected domains).


Let U C be a connected open set. Let 'Y be a closed curve, with image in
U, that is homologous to zero. If z is a point of U, and if f is holomorphic
on U, then
Ind'Y(z) f(z)=

J
27ri Ir

f(()
( - z

d(.

In particular, the formula holds for any closed curve 'Y that is homotopic to
a constant curve.

11.3. Simpl e Connectivity

343

Proof. The proof of the first statement is obtained by applying Theorem

11.2.4 to the function

1-t

[!(() - J(z)]/(( - z).

The details are left as an

exercise. The second statement follows from the first by Lemma 11.2.3, or
by applying Corollary 11.2.5 to the function

1-t

[f(() - f(z)]/(( - z).

11.3. Holomorphic Simple Connectivity


and Topological Simple Connectivity
In Section 6.7 we proved that if a proper subset U of the complex plane
is holomorphically (analytically) simply connected, then it is conformally
equivalent to the unit disc. At that time we promised to relate holomorphic
simple connectivity to some more geometric notion. This is the purpose of
the present section.
It is plain that if a domain is holomorphically simply connected, then
it is

(topologically) simply connected. For the hypothesis implies (by the

analytic form of the Riemann mapping theorem, Theorem 6.6.3) that the

domain is either conformally equivalent to the disc or to C. Hence it is cer

tainly topologically equivalent to the disc. Therefore it is simply connected


in the topological sense.
For the converse direction, we notice that if a domain U is

( topolog

ically) simply connected, then simple modifications of ideas that we have


already studied show that any holomorphic function on U has an antideriv

--> C is holomorphic. Choose a point P E U. If


1
1'1' 1'2 are two piecewise C curves from p to another point Q in u' then

ative: Suppose that

,c
J'-n

J(z) dz

,c
J'-y2

J(z) dz.

This equality follows from applying Corollary

11.2.5 to the (closed) curve made up of l'l followed by "1'2 backwards" (this
curve is homotopic to a constant, since U is simply connected).
F(Q)

f7 J(z) dz

Define

for any curve I' from P to Q. It follows, as in the proof

of Morera's theorem (Theorem 3.1.4), that Fis holomorphic and F'

U.

f on

Thus one has the chain of implications for a domain U c C but unequal
to C: homeomorphism to the disc D =>

(topological) simple connectivity

=>

holomorphic simple connectivity => conformal equivalence to the disc D

=>

homeomorphism to the disc D, so that all these properties are logically

equivalent to each other.

[Note:

It is surprisingly hard to prove that "simple

connectivity" => "homeomorphism to D" without using complex analytic


methods.] This in particular proves Theorem 6.4.2 and also establishes the
following assertion, which is the result usually called the Riemann mapping
theorem:

11. Top ology

344

11.3.1 (Riemann mapping theorem).


ply connected open subset of C, then either U
equivalent to the unit disc D.

Theorem

is a connected, sim
C or U is conformally

If U
=

11.4. Simple Connectivity and


Connectedness of the Complement
If '"Y is a closed curve in C, then the winding number, or index, Ind,.(a) is a
continuous function of the point a

E C\ '"Y.

In particular, if'"'(

[O, 1]

----+

U is a

closed curve in an open and connected set U, then Ind,. (a) is continuous on

C \ U.

It follows that the winding number Ind,.(a) is constant on each con

nected component of C\ U. In particular, if C1 is an unbounded component


of C \ U, then Ind1(a)
then

C\ U

0 for all a

E C1

(cf. Exercise

9).

If U is bounded,

has exactly one unbounded component. Hence if U is bounded

and C \U has only one component, then t hat component is unbounded and
Ind1(a)

0 for all a

EC \ U.

11.2, we saw that if U is simply connected, then Ind,.(a) 0


EC\ U (Lemma 11.2.3). It is natural to suspect that there is some
relationship between the two possible reasons why Ind,.(a) = 0 for all a E
C\ U , one reason being that U is simply connected, the other that C\ U has
In Section

for all a

only one component. [This intuition is increased by noting that the annulus
is not simply connected and its complement has two components. Moreover
the curves which are not homotopic to constants are ones that, intuitively,
go around the bounded component of the complement of the domain.]

It

is, in fact, true that these conditions are related and indeed equivalent for
bounded domains. The following theorem makes the expectation precise:

11.4.1. If U is a bounded, open, connected subset ofC, then the


following properties of U are equivalent:
Theorem

(a)

is simply connected;

(b) C \ U is connected;
( c ) for each closed curve '"Yin

and

Ind,.(a)

0.

(a) implies ( c ) . Also, (b)


C \ U is connected, then
Ind,.(a), being continuous and integer-valued, is constant on C \ U. But
Ind,.(a)
0 for all a E C \ U with lal > sup{lzl : z E U} by Exercise 9.
Therefore ( c ) holds if (b) does.
We already know, from Section

implies

(c)

11.2,

EC\ U,

that

by the reasoning already noted:

If

To prove that

4.5.2

to see that

(c)

(c)

implies

(a),

combine Theorem

11.2.4

with Lemma

implies holomorphic simple connectivity. Then, as ex

plained in the second paragraph of Section

11.3,

simple connectivity follows.

11.4.

Connec t edness

of t he

345

Complement

This section

(a)

Figure 11.2

The remaining assertion, that ( c ) =? (b) (which will complete the


proof), is the most difficult. We prove this next. By now the logic of the
proof is probably obscure. Figure 11.2 outlines the reasoning.
The following proposition gives the critical step ( c )

Proposition 11.4.2. If U

is

=?

(b):

a bounded, connected open set and ifC \ U is


Ind-y(a) -1- 0

not connect ed , t hen t here is a closed curve 'Y lying in U such t hat
for

some a EC \ U.

Proof of Proposition 11.4.2. If the closed set C \ U is not connected,


then it is expressible as the union of two disjoint, nonempty, closed sets
C1 and C2. [In detail: By the definition of connectedness, the fact that
C \ U is not connected means that C \ U
C1 U C2 where C1, C2 are
disjoint, nonempty, and closed in C \ U in the relative topology of C \ U.
But, since C \ U is a closed set in C, the sets C1 and C2 are necessarily
closed in C also.] Precisely one of these two sets can be unbounded, since
{z EC: lzl 2 1 + supwEU lwl} is connected (Exercise 18(a)). Let C2 be the
unbounded set and let C1 be the bounded one. Choose a E C1. Now define
=

inf{lz - wl : z E C1 and w E C2 }.

Notice that d > 0 because C1, being closed and bounded, is compact and
C2 is closed (Exercise 17).
Now cover the plane with a grid of squares of side d/10 (and sides
parallel to the coordinate axes) and suppose without loss of generality that
a is the center of one of these squares. Now look at the set S of those closed
squares from the grid that have a nonempty intersection with C1. Note that
no closed square in the set S can have a nonempty intersection with C2
because if a 1 x 1 square intersected both C1 and C2, then there would be

346

11. Topology

points in

C1 and C2

J2d/1 0

that are within distance

(i.e., the length of the

diagonal of that square). That would be a contradiction to the choice of d.

S1 be the set of squares in the collection S which can be reached


a by a chain of squares in S, each having an edge
in common with the previous one (Figure 1 1 .3). Orient all the squares in S1
as shown in Figure 1 1 .4 (counterclockwise), and consider all the edges which
do not belong to two squares in S1. This set of edges is clearly the union of
Now let

from the square containing

a finite number of oriented, piecewise linear, closed curves rJ

1, . . , k,
.

in U; the orientations of these curves are determined by the orientations of


the squares, with the curves disjoint except possibly for isolated vertices in
common-Figure

1 1 .5. [Exercise:

Show by example how a rj can occur that

S.]

is not the entire boundary of a component of

Now consider the sum of the integrals over the oriented squares:

I: _1 1

This sum equals

QES1

27ri

__

JaQ ( - a

d,

because for the square Qa containing

l
1 1
d(
27ri JaQ a ( - a

__

__

we have

1'

while for all the other squares we have

l
1 1
d(
27ri JaQ ( - a

__

__

o.

On the other hand, because a common edge of two squares in

S1

is

counted once in one direction and once in the other, the sum

1
d(
"'""' _1 1
27ri JaQ ( - a
QE
__

"'""' _1 1 1 d( .
27ri !, ( - a

J
__

In particular, by ( * ) , it must be that

d, -1=0
1
I:
2m, fr1 ( - a
_
_

Hence at least one term of this sum is nonzero, that is, some 'YJ is a closed
curve in U with nonzero winding number about the point
the desired conclusion.

a E

C1.

This is
0

The reader should think carefully about the argument just given. The
point, philosophically, is that the use of the squares enables us to make a
specific problem out of what at first appears to be a rather nebulous intu
ition. In particular, it is not necessary to know any results about boundaries
of general regions in order to establish the proposition.

11.4. Connectedness of the Complement

I/

I..; ...

...... ....
.....

....

'

347

"'

!'..

-.....

.....

'

other part of

'

S(notS1l

'

{Shaded}
S1

-,
7

region

)
I/

I/

"'
_ .....

......

Figure 11.3

Figure 11.4

It is interesting to note that an independently proved result in Chapter

12 will show directly that

(b)

implies

( a)

in Theorem 11.4.1. We already

know that if U is analytically simply connected, then it must be simply con


nected. Moreover, if every holomorphic function f on U has a holomorphic
antiderivative, then U is holomorphically simply connected by definition.

11. Top ology

348

Figure

Finally we know that

11.5

has a holomorphic antiderivative if

i f(z)dz=

for every piecewise smooth closed curve 'Y in U.


The advantage of condition

(** )

is that it is closed under normal limits.

In particular, we shall apply the following result, which is


theorem that will be proved independently in Section

corollary of a

12.1 ( see

Exercise

38):

Proposition 11.4.3
connected open set
tion

f:

---+

( Corollary of Runge's theorem ) . If U is a bounded,


in C with C \ U connected, then each holomorphic func

C is the limit, uniformly on compact subsets of U, of some

sequence { Pj} of polynomials.


This theorem is a generalization of the fact that a holomorphic function
on a disc can be approximated uniformly on compact sets by the partial sums
of its power series.
Now assume

(b)

in Theorem

11.4.1.

For each fixed, closed curve"(,

i P(()d(=
for any polynomial

P ( because P certainly has an antiderivative ) .

i f(()d(

Therefore

11.5. Multiply Connected Domains Revisited

349

for each function f that is the uniform limit on 'Y of a sequence of polyno
mials. In particular, if C \ U is connected, then

i J(() d(

for every holomorphic function f: U---+ C Thus

(b)

implies

as required.

( a)

11.5. Multiply Connected Domains Revisited


We developed earlier the intuitive idea that a (topologically) simply con
nected domain was one with "no holes." This chapter has given that notion
a

precise form:

(i)

the absence of a hole in the sense of there being no points

in the complement for a closed curve in the domain to wrap around is ex


actly the same as

(ii)

the absence of a hole in the sense of there being no

bounded component of the complement of the (bounded) domain. But what


about domains that

do have

holes, that is , that

do have nonempty

bounded

components of their complement?


It is natural, in case the complement C \ U of a bounded domain U has
a finite number

of bounded components, to say that U has

holes. [As

before, C \ U has exactly one unbounded component.] In more customary


if less picturesque terminology, U is said to have

connectivity k + 1

(i.e., the

number of components of C \ U). Then it is also tempting to ask whether


the number of holes can be determined by looking at rr1 (U). For
know this to be true: U has no holes, that is,

0, we

0 if and only if rr1 (U)

is trivial, that is, rr1(U) equals the group with one element (the identity).
But can we tell the difference between a domain with, say, two holes and a
domain with three holes just by examining their fundamental groups? We
know that the fundamental groups will then both be nontrivial, but do they
somehow encode exactly how many holes there are?
The answer to this question is "yes." But this assertion is not so easy to
prove. The whole situation is really part of the subject of algebraic topology,
not of complex analysis. Therefore we do not want to go into all the details.
But, just so you will know the facts, we shall summarize the key points that
topologists have established:
If U is a bounded domain with

holes (i.e., C \ U has

components), then U is homeomorphic to the open disc with


moved. The disc with

withe points removed if

k bounded
k points re

points removed is not homeomorphic to the disc

k f:- P.

Indeed, if

k f:- e,

then the disc with

points

removed has a different fundamental group than the disc with e points re
moved. [Here "different" means that the two groups are not isomorphic as
groups.]

11. Topology

350

We can actually describe what these fundamental groups are. For this,
recall that the

free group F

generated by a set A (or the

free group on A) is

defined to be the set of all finite "words" made up of elements of A, that is,
finite sequences with 1 exponents:

a tia i .. .a ;i.
Here

a-1

is arbitrary but finite. A word is not allowed if

(same

a)

or if

a-i

is followed by

a+i

is followed by

a+i.

These words are combined-this is the group operation-by juxtapo


sition:

(a tia i .. a ;/ ) . (b tib i .. b!I) = a tia i . . a ib tib i .. b !i


(with

a's , b's EA),

subject to the rule that if am=

b , and if am and b
i
i

ap

pear with oppositely signed exponents, then they shall be "cancelled." If, in
this situation,

am-i

and b2 are equal but have oppositely signed exponents,

then they are to be cancelled as well; and so forth. For instance,

(For convenience,

(aia2a3) (a3ia2iai) = aiai.


aiai is usually written a i.) It is

easy to see that the

allowable words and the composition described form a group. In an obvious


sense, the group just described is generated by the elements of

with no

"relations" assumed: hence it is called the "free group" generated by A.


W ith this terminology in mind, we can now describe the fundamental
groups of domains with holes: The fundamental group of a bounded domain
with k holes is (isomorphic to) the free group on (a set with) k generators.
It is not

quite obvious

that the free group

isomorphic to the free group

Fl

Fk

on k generators is not

on f generators when k

f. To see that

these groups are not isomorphic, the thing to do is to look at

and

Flf[Fl,Ff.].
(Recall from group theory that if G is a group, then the

[G, G]

of

commutator subgroup

is the subgroup generated by the elements

{gig2g!ig ;ii: gi ,g2 E G}.)


You can check for yourself that

is isomorphic to the additive group

{(xi ,... ,xk ) E JR.k:

Xj

E Z,

all

j}.

351

11.5. Multiply Connected Domains Revisited

From this, it is easy to see that

Fk/[Fk,Fk]
is not isomorphic to

F,j[F1.,F1.]

if k

=I=

f. So

Fk

Ft.

cannot be isomorphic to

either.
It turns out that a closed curve

'Y ( at

a base point ) in a domain

is

homologous to 0 in the sense we have already defined if and only if

["!]

[7r1(U),7r1(U) ] .

For a bounded domain, with k holes, k 2, the group


equivalently,

7rI (U)

Fk

[7rI (U),7rI (U) ] =I= O;

is not commutative in this case.

So, for k 2,

there are always curves that are homologous to 0 but not homotopic to the
constant curve at the base point. This point is discussed in more detail in
the exercises.
Even a little more is true: Define an equivalence relation on the closed
curves

'Y

starting and ending at the base point P by

'YI ""' 'Y2


Q E C \ U. Equivalently, 'YI !!., "12 ( read "(1
1
I
is homologous to "12 ) if 'YI 'Y2 is homologous to 0, where 'Y2 (t)
'Y2 (l t),
as in finding homotopy inverses ( and denotes the composition of curves
defined in Section 11.1). The equivalence classes relative to this equivalence
relation are called homology classes ( actually, homology classes at P-the
base point usually is not important and is therefore ignored ) . We shall write
b]h for the homology class of "f.
if Ind,,1

( Q)

Ind,,2 ( Q) for every

We can define a group structure on the set of homology classes by


defining a composition

+h:
bI]h +h b2]h ['YI . 'Y2]h,

where

is the composition of paths used to defined

7r1.

We used a plus

sign for this composition because the group of homology classes is always
abelian, that is,

['Y2]h +h bI]h
This is actually easy to see since Ind,,1 .,,2 ( Q)
Ind,,2.,,1 ( Q). The group of
homology classes of curves is denoted HI (U).
The group o f homology classes HI and the group o f homotopy classes
7rI are related:
HI (U) 7rI (U) /[7rI (U),7rI (U) ].
['YI]h +h b2]h

W ith this striking fact , we conclude our brief tour of the key ideas of ele
mentary algebraic topology and invite you to consult [ MAS] or

[GRH]

for

a more detailed view of these matters. Much of what we have said can be

11. Topology

352

generalized to arbitrary topological spaces, and it is worthwhile to find out


how.

Exercises

1. Provide the details of the proof of the Cauchy integral formula (using

11.2.4) for
[( - z]
(z)]/
[!(() - f
z E U.

Theorem

2. A set S

multiply connected domains, using the fact that


is a holomorphic function of

for each fixed

ffi.N is called

there is a continuous
'Y(O)

( EU

P and 'Y(l)

path connected if, for any two points


function (i.e., a path) 'Y : [O, 1]
S
-+

[Hint:

E S,

such that

Q. Prove that any connected open set in

is in fact path connected.

ffi.

Show that the set of points S that can

be connected by a continuous path to a fixed point P in


and closed in

P, Q

is both open

U.]

{z EC: 1/2 < lzl < 2}. Consider the two paths 'Yi(t) e27rit
e47rit, 0 -:::; t -:::; 1. Prove that 'YI and 'Y2 are not homotopic.
and 'Y2 (t)
[Hint: You could use the index as a homotopy invariant. Or you could
construct your own invariant in terms of crossings of the segment { O+it :
1/2 < t < 2}.]

3. Let

4. Is the union of two topologically simply connected open sets in the plane
also topologically simply connected? How about the intersection? What
if the word "plane" is replaced by "Riemann sphere"?

5. Let

U1 U2
Define U
LJ1 U1.

C be

topologically simply connected open sets.

Prove that U is then topologically simply connected.

What if the hypothesis of openness is removed?

6. Use your intuition to try and guess the form of

7r1 (the first homotopy

group) of the set consisting of the plane minus two disjoint closed discs.

7. Let S be the unit sphere in ffi.3. Calculate the first homotopy group of
S.

[Hint:

First show that any closed curve is homotopic to a closed

curve that is a finite union of arcs of great circles. Do so by subdividing


an arbitrary given closed curve and using the fact that two curves with
the same endpoints, both curves lying in a fixed open hemisphere, are
homotopic with fixed endpoints.]

8. Refer to Exercise

for terminology. Show that if a planar set is path

connected, then it must be connected. Show that the set

[-1, 1]} U { (x, sin 1/x) : x E (0, 1]} is

{ (0, y) : y E

connected but not path connected.

Exercises

353

9. Suppose that A <;;;; C is a connected, unbounded subset ofC and that r is

a closed curve in C that does not intersect A. Show that Ind"Y (a) 0 for
all a E A. [Hint: The quantity Ind"Y(a) is a continuous, integer-valued
function on the connected set A and is hence constant on A. Also, for
a E A, with lal large enough, the index Ind"Y(a) is smaller than 1 and
hence must be equal to O.]
=

10. Let

S be a subset of the plane that is compact and topologically simply


connected. Does it follow that the interior of S is topologically simply
connected? [Hint: Look at the complement of the set.]

11. Let T be an open subset of the plane that is topologically simply con

nected. Does it follow that the closure of T is topologically simply


connected?
12. Let f(z)

1/[(z + l)(z - 2)(z + 3i )]. Write a Cauchy integral formula


for f, where at least a part of the integration takes place over the circle
with center 0 and radius 5, and lzl < 5.

13. Let g(z)

tanz. Write a Cauchy integral formula for g, where at least


a part of the integration takes place over the circle with center at 0 and
radius 10, and lzl < 10.
=

14. Let

S and T be topologically simply connected subsets of the plane.


Does it follow that S \Tis topologically simply connected?

15. Let

S be a topologically simply connected subset of C and let J be a


complex-valued, continuous function on S. Does it follow that J(S) is
topologically simply connected?

16. Let

S be a subset of C and let f : C --+ C be an onto, continuous


1
function. Does it follow that 1- (S) is topologically simply connected
if S is?

17. Prove that if C1,

C2 are disjoint, nonempty closed sets in C and if C1 is


bounded, then inf{lz wl : z E C1, w E C2} > 0.
-

18.

(a) Assume that the domain U is bounded. Prove that there can be
just one connected component of the complement of U that is un
bounded. [Hint: {z EC: lzl 1 + supwEU lwl} is connected.]
(b) Assume further that the complement of the bounded domain U
has only finitely many connected components, and let C1 be the
union of the bounded components of the complement. Let C2 be
the unbounded component of the complement. Define
d

inf{ lz - wl : z E C1 and w E C2}.

Use Exercise 17 to prove that d


19. Show that part
U is bounded.

>

0.

(a) of Exercise 18 fails if we remove the hypothesis that

354

20.

Exercises

Let Ube a domain and P E U. Let/be a closed curve in U thatbegins


and ends at P.

Show that "Y

,-1 and ,-1 / are homotopic to the

constant curve at P. [Hint: For the first of these,

H(s,t)

21.

1(2ts)

if

0:::;t:::;1/2

1(2s(l-t))

if

1/2

<

t:::;1.

Let Ube a domain. F ix a point P E U. Let /1, 12,/3be closed curves


based at P. Show that if /1 is homotopic to /2 and/2 is homotopic to
/3 , then 1

is homotopic to /3. [Hint: If Hl is a homotopy from /1 to

/2 and H2 is a homotopy from/2 to/3, then define

H( 8' t)

22.

H (2s,t)
1
H2(2s- l,t)

0:::;s:::;1/2

if
if

1/2

Let Ube a domain. Suppose that r : [O, 1]


with r(O)

P E U and f(l)

curve/: [O,1]

---+

U with 1(0)

by

(r-1 . 1. r)(t)

<

---+

s:::;1.

U is a continuous curve

Q E U. For each continuous, closed

1(1)

P, define r-1 / r: [O,1]

r(l-4t)

if

---+

o:::;t:::;1/4

1(2(t-1/4))

if

1/4

<

t:::; 3/4

r(4(t - 3/4))

if

3/4

<

t:::;1.

Show that r-1 / r is a continuous curve that starts and ends at

( a)

Q.
Show that, with r fixed, the equivalence class [r-1 "Y f] of r-1 "Y. r

(b)

in rr (U)based at Q depends only on the equivalence class [r] of/

in rr (U) based at P.

(c )

Show that the mapping

[r]

[r-1 . 1 . r]

f--+

is a group homomorphism.
Show that the composition of the group homomorphisms [r]

(d)

[r-1. "Y. r] followedby [<5]

f--+

f--+

[r-1 . <5 . r] (for <5 a closed curve at Q)

is the identity mapping from rr (U) based at P to itself.

Show that rr (U)based at P is isomorphic to rr (U)based at Q for

(e)

all points P, Q in U.

23.

Let U

Cbe

a connected open set and/: [O, 1]

---+

U a closed curve in

U. We define the statement "1 is freely homotopic to a constant in U"


to mean: There is a continuous function H: [O,1]

(1)

H(l, t) is a constant , independent oft E [O,1];

[O,1]

(2)

---+

H(s,

U such that

0.)

H(s,1)

1(t) for all t E [O,l]. (One thinks of


as a family of closed curves , the curve corresponding to s
0

for alls E [O,1]; and

H(s,

(3)

H(O,t)

being/ and the curve corresponding to s

1being a constant.) Clearly,

355

Exercises

"f(O) to the constant curve at 'Y(O), in the sense


11.1, then 'Y is freely homotopic to a constant.
Prove the converse: If 'Y is freely homotopic to a constant, then 'Y is
homotopic at 'Y(O) to a constant. [Hint: The bottom edge of the square
if

'Y

is homotopic at

introduced in Section

can be deformed with fixed endpoints to the curve formed by the other
three sides in succession. The H-image of the three-sided curve is itself
homotopic at

'Y(O)

to the constant curve at

7rI ( U)

24. In the text we defined

for any (open) domain

struction can be applied to define

P E X.

with

Here

"f(O).]

7rI (X, P)

is a "base point."

through, and show that if

U.

The same con

for any topological space


Think this generalization

is path connected (see Exercise 2-the

definition there applies to any topological space), then

7rI (X, Q)

are isomorphic groups for all

P, Q E X

7rI (X, P)

and

(see Exercise 22 for a

clue).

25. Let

and T be topologically simply connected sets in the plane. Does

it follow that

26. Let

\JI

--t

Tis topologically simply connected?


be a continuous mapping from a topological space

to a topological space

Y.

Let

"induces" a group homomorphism

P
\JI*

X. Show that \JI


7rI (X, P) to 7rI(Y, \JI ( P)) by

be a point of
from

w*(['Y]) = [w o "f]. [Hint: Part of the problem, of course, is to show


W* is well defined, i.e., that if 'YI ,....., "f2, then W*(['YI]) = W*(['Y2]).]
27. Let

(a)

{z EC: 1

<

lzl

<

that

3}.

Show that every closed curve


homotopic at the point

'Y: [O, 1]

--t

with

2 to a closed curve

'Y(O) = "f(l) = 2 is
::Y: [O, 1] --t U such

<::;: {z: lzl = 2 } .


'YI, 'Y2 ( as in part (a)) are homotopic at the point 2 in
U, then 91 and ::Y2 are homotopic at the point 2 in {z : I z I = 2 } .
I
( c ) Deduce that 7rI(U) is isomorphic to 7rI(8 ), where 81 = {z: lzl =
2 }.
that

(b)

9([0, 1])

Show that if

28. The goal of this exercise is to prove that


group of integers Z under addition. Here

7rI(81) is isomorphic to the


81 = {z E C: zl = 1}, that

is, the unit circle in the plane.

(a)

'Y(O) = 1. Show
A-y : [O, 1] --t JR.
2
such that A-y(O) = 0 and e 71"iA-y ( t ) = 'Y(t) for all t E [O, l]. In
accordance with our previous terminology we call A-y(l) the winding
number of 'Y [Hint: For existence, subdivide the interval [O, 1] into
subintervals with a partition 0 = to < tI < t2 <
< tk = 1 such
that, for each j = 0, 1, .. . , k-1, the image 'Y([tj, tj+ ]) is contained
1
in some arc of the unit circle having length less than 7r. Select
Let

'Y : [O, 1]

--t

81

be a continuous curve with

that there is one and only one continuous function

"branches" of the angle function on each such circular arc, and

356

Exercises

-1

Figure 11.6
patch these together, adjusting by multiples of 27r when necessary,
to make them fit together continuously.]
(b) Suppose that 11 and 12 are as in part ( a ) and that, moreover,
11 (1)
12 (1)
1. Prove that 11 and 12 are homotopic at 1 (as
1
closed curves in S ) if and only if A-y1 (1)
A-y2 (1). [Hint: If /I
and 12 are homotopic, then An (1)
A-y2 (1) because the winding
number cannot "jump" when the curve is deformed continuously.
For a homotopy when A-y1 (1) A-y2 (1), use
=

exp[27ri((l - s)A-y1 (t) + sA-y2(t))]


for s, t E [O, 1].
1
( c ) Conclude that 7r1(S )

Z.

29.

Use the ideas and results of Exercises 27 and 28 to show that 7r1 ( { z E
C: R1 < [z[ < R2}) Z for any R1,R2 such that 0:::; R1 < R2:::; +oo.
[Hint: See Exercise 27.]

30.

The purpose of the present exercise is to construct a curve that is ho


mologous to zero but not homotopic to zero. The construction is rather
elaborate. You should consider it as an open-ended invitation to learn
more about topology.
An open interval I JR can be subdivided into infinitely many
subintervals with the property that the endpoints of the subintervals
accumulate only at the endpoints of I. For instance, the interval (-1, 1)
can be subdivided into (0, 1/2), (1/2, 3/4), (3/4, 7/8), ... and (-1/2, O),
(-3/4, -1/2), (-7/8, -3/4), .... With this subdivision process, which
2
can be scaled to apply to any interval, we construct a subset of JR. as
follows:
Begin with (-1, 1) subdivided as described in the last paragraph.
Attach to each point that occurs as an endpoint in the subdivision a ver
tical, open segment centered at that point and having length one (Figure

Exercises

357

+
I

Figure

11.6).

11.7

Now subdivide each of these vertical segments as in the preceding

paragraph. On each vertical segment J, attach at each subdivision point


a small horizontal open segment, centered on J; the segments J should
be so short that no two horizontal segments intersect.
Continue this process by subdividing each new horizontal segment,
and adding even smaller vertical segments that are centered on the hor
izontal segments at each subdivision point.
are shown in Figure

Some steps of the process

11.7.

The result of repeating this construction infinitely many times is an


"infinite graph" G (in the sense of graph theory) with four edges meeting
at each vertex. Now define a map

7r :

2 whose image is the union

G-----+

of two circles that osculate at a point. For specificity, let the two circles
be

{(x,y) E 2: (x+ 1)2+y2=1}


Denote this union by

S = 81 U 82.

{(x,y)

2: (x-1)2+y2=1}.

We define the map as follows: All

vertices in the graph get mapped to the common point (0, 0) of the two
circles. A horizontal segment between two adjacent vertices of G is to
be "wrapped" counterclockwise around

82.

A vertical segment between

two adjacent vertices of G is to be "wrapped" counterclockwise around

S1. Now prove the following:


(a) The mapping 7r just described is a continuous function from G (with
the topology induced from 2) onto S.
(b) If t
1( t) E 2, t E [O, 1], is a continuous curve with 1( t) lying
in S for all t E [O, 1] and with 1(0 ) = (0, 0), then there is a unique
curve t
::Y(t) E G such that 9(0) = (0, 0) and 7r(9(t)) = 1(t) for
all t E [O, 1]. [Hint: Look carefully at how you did Exercise 28 (a)
-

f->

and imitate that idea.]

Exercises

358

( c ) If ')'1,')'2 are two curves as in part (b) and if b1 ]


b2 ] E 7r1(S),
then 11 (1) 12 (1).
(d) If ')'1, 1'2 are as in parts (b), ( c ) and if 11 (1) 12 (1), then b1 ] b2 ]
in 7r1 (S).
( e ) Deduce that if o1 is a curve that goes once counterclockwise around
S2 and o2 is a curve that goes once counterclockwise around S1
(with Oj(O) Oj(l) (0, 0), j 1, 2), then
=

[01 02] f= [02 01].


1
1
Conclude from ( e ) that [o1 02 o;_- 02 ] f= [e].
2
Show that there is a continuous mapping R: JR \ {(-1, 0), ( +1, O)}
- S with the property that R is restricted to S is the identity.
1
1
Deduce from (g) that [01 02 o;_- 02 ] f= [e] in the homotopy
1
2
group 7r1 (JR \ {(-1, 0), ( +1, O)}). That is, show that 01 02 o1 o21
2
is not homotopic to a constant mapping in JR \ {(-1, 0), ( +1, 0)}.
[Hint: Such a homotopy, if it existed, could be composed with R to
give a homotopy taking place in the union of the two circles S.]
1
Show that 01 0201102 is homologous to zero in JR2\{(-1, 0), (1, O)}.

(f)
(g)
(h)

(i)
31.

Let U {z: JzJ < 1, z f= 1/2, z f= -1/2}. Let f: U - C be holomor


phic. Prove that there is a holomorphic function F : U - C such that
F'
f if and only if
=

li(-1/21=1/4

J(() d(

and

li(+l/21=1/4

f(() d(

0.

* 32. Let U be a connected open set in C Define an equivalence relation


,....., on the set of holomorphic functions on U by Ji ,....., h if there is a
holomorphic function F : U - C such that F'
J1 - h on all of U.
Prove the following statements about ,...:..,
( a ) The relation ,....., is indeed an equivalence relation.
(b) The set of equivalence classes induced by ,....., forms a vector space
over C under the operations [f] + [g]
[f + g] and o[f]
[of],
oEC
( c ) Use Exercise 31 to prove that, if U
{z : lzJ < 1, z f= 1/2, z f=
-1/2}, then the linear space constructed in (b) is a two-dimensional
vector space over the field C of scalars.
** (d) Use Runge's theorem to convince yourself that if U is a bounded,
connected open set in C such that C \ U has k + 1 components,
then the vector space has dimension k over C.
=

* 33. Let U be a connected open set in C. Define an equivalence relation


,....., on the set of all harmonic functions from U to JR by h1 ,....., h2 if
h1 - h2 : U - JR has a harmonic conjugate (i.e., h - h2
Re F for
1
some holomorphic function F : U - JR). Prove the following statements:
=

Exercises

( a)
(b)

359

The relation "' is indeed an equivalence relation.


The set of equivalence classes forms a vector space over the scalar

[h 1 ]

field IR with operations

[h2]

[h 1

h2]

and a [h] =

[ah] ,

a ER

(c )

Show that the dimension of this new vector space is

D(O, 1) \ {z1, ..., zk},

that is, the disc with

Uj : z

the functions

k for the domain

points removed. Do this by showing that

ln

!z - Zjl, j

1, ..., k,

form a basis.

[Hint:

Refer to Exercise 6 of Chapter 7 for the case of one deleted point.]

* 34. Let

be a connected open set in C.

"' on the set of all

( real )

Define an equivalence relation

C00 differentials

( i.e.,

differentials with C00

coefficients ) w =
if w1 - w2

A(x, y)dx+B(x, y)dy with 8B/8x = 8A/8y by w1 "' w 2


df for some C00 function f : U R Prove the following
-+

statements:

(a)
(b)

The relation "' is indeed an equivalence relation.


The equivalence classes under this relation form a vector space over

IR in a natural way ( refer to Exercises

(c)

The vector space has dimension

32 and 33).
k when the region U is a disc with k

points deleted. What differentials can you use as an explicit basis?

* 35. Prove the following result, which is commonly known as the Lebesgue
covering lemma: If K is a compact set in a Euclidean space Rn, n

1,

{U.x, .X E A} is a collection of open sets in Rn such that K C


U.xEA U_x, then there is an E > 0 such that, for each k EK, there is a set
U.x with B(k, t: ) c U_x. Here B(y, r ) is the Euclidean ball with center y
and radius r. [Hint: Suppose not. Then there is, for j = 1, 2, 3, ... , a
kj EK such that B(kj, 1/j ) fails to be contained in any U_x. Choose a
subsequence {k]t} that converges to a point, call it ko, in K. Then derive
a contradiction from the fact that ko E U_x0 for some index .X o.] Note
and if

that the Lebesgue covering lemma actually applies to compact metric


spaces in general, by the same proof.

* 36. Suppose that

is a connected, open set in C,

morphic function, and


is

not

program to define

( a)

-y : [O, 1]

-+

assumed to be piecewise C

for

0, 1, 2,

C is a holo

-+

is a continuous curve

( note

that "f

Carry out in detail the following

f-y J:

Choose, by applying Exercise


contained in

).

f : U

35,

a positive integer

... , N -1, -y([j/N, (j + 1) /N] )

U.

such that,

some open disc

Dj

360

Exercises

(b)

Let Fj : Dj

C be

a holomorphic function such that F

Dj. Then each

=f

on

N-1

i'Y f= L [
=
j=O

(c)

Fj(r((j + 1)/N)) - Fj(r(j/N)) .

Suppose that 0
t o < t1 <
< tM
1 is any partition of [O, 1]
with the property that 'Y([tj, tj+iD C some open disc Dj contained

in U. Verify that
M-1

if= L [
j=O

'Y

F (r(tj+1)) - F (r(tj)) .

In particular, verify that the definition of

(d)

'Y

in part

(b)

is inde

pendent of the choice of the integer N.


Check that if 'Y is a piecewise C1 curve, then the definition of

J-y
(b) agrees with the definition from

that we have formulated in part

Chapter 2; that is, check that the new definition gives the same
1
J(r(t)) r'(t) dt.

answer as
*

37. Construct an alternative proof of Corollary 11.2.5 by completing the


following outline:
Let H : [O, 1] x [O, 1] Ube a homotopy through closed curves,
that is, H is continuous and H(s, 0)

Suppose that
that

: U

H(s, 1) for all s

[O, 1].

is holomorphic. We want to prove

(0,.) f = .) f
=
=
( 1,

Choose, using Exercise 35, a positive integer N so large that,


for each j, k with j

0, 1, 2, ..., N -1, k

holds that H [j/N, (j + 1)/N]

0, 1, 2, ..., N -1, it

[k/N, (k+ l)/Nl

is contained

in some open disc contained in U. Then the counterclockwise


integral of

around the curve given by the H-image of the

boundary of the square [j/N, (j + 1)/N]

[k/N, (k + 1)/N]

equals 0.
Sum over j, k and note that the inside edges cancel to deduce
that the integral of

around the boundary of [O, 1]

[O, 1]

equals 0. Then deduce the desired conclusion of the theorem.

38.

( a)

(b)

Prove:

If U is a bounded, connected open set in

connected, then the set K6

c5 > 0 sufficiently small,

{z

U : dist(z,

is compact and

\ K6

\ U)

with

c5},

\U

with

is connected.

Deduce Proposition 11.4.3 from Corollary 1 2.1.2 in the next chap


ter.

Exercises

361

39. Justify in detail the application of Fubini's Theorem in the discussion


preceding Theorem 11.2.6.

For this, you will need to think carefully

about the definition of the integral of a holomorphic function along a


continuous curve given earlier.

Chapter 12

Rational
Approximation Theory

12.1. Runge's Theorem


A

rational function is, by definition, a quotient of polynomials. A function


C to C is rational if and only if it is meromorphic on all of C (Theorem

from

4.7.7). A

rational function has finitely many zeros and finitely many poles.

The polynomials are those rational functions with pole only at oo. Let K

C be compact and let J : K


conditions is

--t

C be a given function on K. UndeE_ what

f the uniform limit of rational functions with poles in C \ K?

There are some obvious necessary conditions:

(1)

J must be continuous on K;

(2)

J must be holomorphic on K, the interior of K.

It is a striking result of Mergelyan that, in case C \ K has finitely many


connected components, then these conditions are also sufficient. We shall
prove Mergelyan's theorem in the next section.

In the present section we

shall establish a slightly weaker result-known as Runge's theorem-that is


considerably easier to prove.

12.1.1 (Runge). Let K C be compact. L!t f be holomorphic


on a neighborhood of K. Suppose that P is a subset of C \ K containing one
point from each connected component of C \ K. Then for any t > 0 there is
a rational function r(z) with poles in P such that

Theorem

sup

lf(z) - r(z)I

<

t.

zEK
-

363

12. Rational Approximation Theory

364

Figure 12.1

<C be

Corollary 12.1.2. Let K

compact with

<C \

K connected.

Let

be holom orphic on a n eighborh ood of K. Then for any E > 0 there is a

holom orphic polynomial

p(z) such that


sup
K

lp(z) - f(z)I

< E.

The corollary follows from the theorem by taking P

{ oo}.

We shall prove Theorem 12.1.1 later in this section, but first we shall
give some applications and explain the idea of the proof.

EXAMPLE 12.1.3. There is a sequence of holomorphic polynomials Pj such


that Pj

1 normally in the upper half plane, Pj ___, -1 normally in the


lower half plane, and Pj ___, 0 uniformly on compact subsets of the real axis.
___,

Proof. Let
Kk

{z

<C:

::::; I Im zl ::::; k, I Re zl ::::; k

{z E <C: Im z

See Figure 12.1.


Let

fk(z)

-1

if
if
.l f

Imz > A
IImzl < A
I mz < -i
2k .

0, Re

zl

::::; k }.

12.1. Runge's Theorem

365

Figure 12.2

Then

fk is

holomorphic on a neighborhood of

Kk and

\ Kk is

connected.

So Corollary 12.1.2 applies, and there is a holomorphic polynomial


that
sup IPk(z)
Kk

The sequence

{Pk}

- fk(z)I

Pk such

k
r .

<

has the desired properties.

EXAMPLE 12.1.4. There is a sequence of holomorphic polynomials Pk such


0 pointwise on C \ {x + iO: x 2: O}.
Pk(O) = 1 for all k and Pk

that

_____,

Proof. For k = 1, 2, 3, ... , let

[ ( ) { z

Lk = D(O,k ) \ D

0,

C :

1
- 2k

< Im

<

0, Re

> 0

}]

and

See Figure 12.2. Define

fk(z) =
Then

fk is

if

IzI

<

41k

if

I zI

>

31k .

holomorphic on a neighborhood of

Kk,

Corollary 12. 1.2 applies. So there is a polynomial


sup
Kk

Define

lqk(z) - fk(z)I

<

r .

\ Kk is

qk such

connected, and

that

366

12. Rational Approximation Theory

Then

Pk(O) = 1

and

sup IPk(z)

- fk(z)I

<

sup

IPk(z) - qk(z)I

+sup

Kk

Kk
<

lqk(z) - fk(z)I

Kk

rk + sup lqk(z) - fk(z)I


zEKk

Thus the sequence


Examples

Pk

12.1.3

has all the required properties.

12.l.4 provide a counterpoint to the normal fam


6.5. There we developed some intuition that se

and

ilies material in Section

quences of holomorphic functions tend to converge uniformly on compact


sets (i.e., normally). By contrast, the sequences in these examples converge
pointwise, but most definitely do not converge normally on C. The point
is that the sequences in the examples are not bounded uniformly on com
pact sets, so that Montel's theorem (Theorem

6.5.3)

is irrelevant to these

situations. Thus one can think of the examples as illustrating the essential
nature of the boundedness hypothesis of Montel's theorem.
We turn now to the proof of Theorem

12.l.l.

The essential idea in the

proof of Runge's theorem is to apply the Cauchy integral theorem to f and


then to approximate the integral by its Riemann sums. The construction
used is closely related to ideas used in Chapter
of Theorem

11.2.4.

11,

for example the proof

This is best perceived if we first prove a preliminary

version of the theorem in which we do not worry about the exact location of
the poles. The poles can be handled afterwards with the technique of "pole
pushing" (cf. Lemma

8.3.5).

Proposition 12.1.5. Let K C be compact and let U be an open set


containing K. If f is holomorphic on U and E > 0, then there is a rational
function

r(z)

with all its poles in C \ K, with simple poles only and limit 0

at oo, s u ch that
sup IJ(z)

zEK

- r(z)I

< E.

Proof. We may assume that U is bounded. Let


a=

inf{lz - wl: z EK, w EC\ U} > 0.

Choose a positive integer N such that

2-N+l

<

and consider the grid in

C consisting of closed boxes with vertices


k
+1 k
j
k +1
+ 1 k +1
2N'2N ' '2N ' 2N'2fl ' '2fl

(j
Let

) (j

) (

) (j

be the set of all such boxes which intersect K. See Figure

12.3.

367

12.1. Runge's Theorem

Figure 12.3

For any Q E g and

(j. 8Q we see that (since Q

_1 j f(()
d(
27l"i JaQ ( - z

{ f(z)
O

if
if

z
z

U)

EQ
(j. Q.

Here each Q is equipped with the positive, counterclockwise orientation.


If z EK is fixed, z not in the boundary of any Q, then there is precisely
one Q such that z E Q. Thus

I:
1 J(() d(
27l"t Ja
- Z
Q(

QE9

t(z) .

But the integration over any edge shared by two Q's occurs twice in this
sum and with opposite orientations. Thus the integrations cancel. If 1 is the
oriented union of the remaining edges, then the image ::Y of '"'( lies in U \ K
and

27ri lr
See Figure 12.4.

f(()

(-z

d(

(12.1.5.1)

f(z).

Now the integral on the left of Eq. (12.1.5.1) is a continuous function


of z, so Eq. (12.1.5.1) persists for all z EK (not just those in K\LJ (J 8Q).
QE
The set ::Y consists of finitely many linear segments (let the segments be
parametrized by 'Yj: [O, 1]---+ C,j
1, ... ,n ). Thus Eq. (12.1.5.1) becomes
=

[1 f('Yj(t)) 1j (t) dt

1 (t) - z
27ri
l

J=

}0

f(z).

(12.1.5.2)

Since K is disjoint from ;:y, each integrand in Eq. (12.1.5.2) is jointly


continuous in z and t. Thus one can verify directly that the Riemann sums

12. Rational Approximation Theory

368
G

Figure

for each of the n integrals in Eq.


As a result, there are partitions 0
n

M(j}

LL

1=1 t=l

tr

<

(tt.))
1 e j(). 6. - f (z)
1
11(te) - z
f(

(12.1.5.3) is
'"'/j (t ) E C \ K as desired.

The proof of Theorem

12.1.1

converge uniformly for

(12.1.5.2)

But each summand in Eq.


pole only at

12.4

t1M(j)'j

<

< E,

1, ... , n

zEK.

such that

(12.1.5.3)

a rational function of z with simple

in its full generality requires the following

slight refinement of the "pole-pushing" Lemma

Lemma 12.1.6

all

z E K.

8.3.5.

K C be compact. Let
PE C \ K and let U be the connected component of C \ K which contains
P. If E > 0 and QE U, then there is a rational function q(z) with pole only

at Q

( Generalized

pole-pushing ) . Let

such that

sup
zEK

I _!_p - q(z) I

< E.

Z -

Proof. We will show that the set G U of points

which satisfy the

conclusion of the lemma is open, nonempty, and has open complement in U.


This will imply that G

U, completing the proof.

The set G is certainly nonempty since

PE G.

The set G is open since if Q E G and

Lemma

8.3.5

applies to show that

Q'E G.

I Q - Q'I

<

dist ( Q, K), then

12.2. Mergelyan 's Theorem

369

The set U \ G is open since if Q E U \ G and if Q' E U satisfies


IQ - Q'I
be in G.

< dist(Q, K), then Q' cannot be in G, for if it


[Exercise: Work out the situation oo E U.]

were, then Q would

This completes the proof.

Proof of Theorem
tion

r(z)

By Proposition

12.1.1.

12.1.5,

there is a rational func

with all poles in C \ K, with simple poles and limit 0 at infinity,

such that
sup lf(z)

- r(z)I < :.. .


2

zEK

We can write

G-j

"'

rz
'
( ) = L.....t
f3 -z
j=l J
f]j is in C\K. Let Uj be the connected component of C\K that
f3j Let Pj E P n Uj . By Lemma 12.1.6 there is a rational function
a pole only at Pj such that

where each
contains

Qj

with

sup

zEK

1
- - Qj I <
1f3j 2Nlajl
E

Define

r(z) = L O'.jQj
j=l
Then
sup lr(z) -f(z)I:::; sup lr(z) -r(z)I +sup lr(z)

zEK

zEK

zEK

- f(z)I

12.2. Mergelyan's Theorem


It is easy to see that the proof of Runge's theorem in Section
relied on the hypothesis that

12.1 genuinely

is holomorphic in a neighborhood of K.

After all, we needed some place to put the contour 'Y This leaves open the
question whether conditions

(1)

and

(2)

at the begining of Section

12.1 are

sufficient for rational approximation on K. When C \ K has finitely many


components, the conditions are sufficient: This is the content of Mergelyan's
theorem which we prove in this section.

It is startling that a period of

sixty-seven years elapsed between the appearance of Runge's theorem

( 1885)

12. Rational Approximation Theory

370

and that of Mergelyan's theorem (1952), for Mergelyan's proof involves no


fundamentally new ideas. The proof even uses Runge's theorem. The sixty
seven year gap is even more surprising if one examines the literature and sees
the large number of papers written on this subject during those years. The
explanation is probably that people thought that Mergelyan's theorem was
too good to be true. It took the world by surprise and superseded a huge
number of very technical partial results that had been proved by others.
We shall concentrate first on proving the simplest form of Mergelyan's
theorem, using elementary and self-contained methods (the proof that we
give here follows the steps in [RUD2]). Afterwards, in this section and in
Section 12.3, we shall explore further results of the same type.

Theorem 12.2.1 (Mergelyan). L et

K C

be compact and assume that

\K

is connected. Letf E

Then for any

>

C(K)

lp(z) - J(z)I<

0 there is a holomorphic polynomial


sup
zEK

p(z)

be holomorphic on the interior

of K.

such that

E.

The proof is most readily grasped if it is broken into several lemmas: In this
way, the main ideas can be identified.

oo

In the next lemma, and in later discussion in this chapter, it is con

oo.

venient to have a concept of the derivative at

of a holomorphic funtion

oo
J: {z: lzl R}
oo.
J'(oo) :zf(l/z)lz=O'
{z: < lz - Pl<
P,
J'(P) = !!__dz ()I
J .
z
oo,
( )

with removable singularity at

This is a natural extension of the idea we

hav! already considered of regarding

as essentially like any other point

of <C. More precisely, suppose that

function with a removable singularity at

>

--->

C is a holomorphic

We define

and similarly for higher derivatives off. Moreover, iff :

R}

--->

C is a holomorphic function with a pole at

then we set

Note that these conventions continue to make the chain rule apply.
example, the function

1/(2z) has derivative 1/2 at

t--t

inverse function w t--t 1 / 2w


at

oo

has derivative 2 at 0.

For

while its (same)

So the compositions,

in either order, which are the identity function, and must have derivative 1
and 0, respectively, can do so consistently with the chain rule, since

=1.

Lemma 12.2.2. Let

E ;

C\E

--->

D satisfying(

C be compact, connected, and contain more than

oo)

c;_ne point. Assume that C

\ E is connected.

I'(oo)I

Then any conformal mapping <P :

= 0 has the property that

2: (diam E) /4.

Mergelyan 's Theorem

12.2.

Proof. Let a =

function

'(00)

371

f- 0. Then
'l/J(z)

(-1)'(0) =

1/a. If T

E E,

then the

Q
=

<P-I(z) - T

is holomorphic and one-to-one on D and satisfies 'lj;(O) = 0, 'l/J'(O) = 1 (exer


cise using Laurent series). It will be proved in Section 13.1 (independent of
the results of the present section) that such a function 'ljJ must have image
that contains D(O, 1/4) . If f- T is another point ofE, then is not in the
image of-1; hence a/( - T) is not in the image of 'lj;.
Thus

or

T) l l
I

lnl
Since , T

E E were arbitrary,

Tl

we conclude that
diamE

Note: See Exercise 21 for the proofthat

C\E is conformally equivalent

IQI

to D.

12.2.3. L et E D(P, r) C be a compact, connected set with

diameter at least r/2. Assume that C \ E is connected. Then there is a


family </Jc;, ( ED(P, r) , of holomorphic functions

Lemma

<;: C\ E--4C
satisfying
(1) 1 <;(z)I < 600/r for all z EC\ E,
(2) l<P<;(z) - l/(z - ()I < (5000r2) /(lz - (13) for all z EC\ (EU{(}),
(3) 0 e function ((, z) = </Jc; (z) is jointly continuous in z and(, z E
C\ E, ( E D(P, r) .
Proof. Assume for simplicity that

P = 0. Let
:C\ E--4D
be a conformal mapping with (00) 0. Then define a= '(00) and
=

h(z) = -<j;(z) .
Q

By Lemma 12.2.2 we know that the image of h lies in D(O, 8/r) . Let f3 =
h" ( oo ) /2 and define

</J c; (z) = h(z)

(( - {3) h2(z) .

372

12. Rational Approximation Theory

The Cauchy estimates tell us that


lf31

:S 8r.

Thus, for ( E D(O, r),zEC\E, we have

I <t>dz) I
This is property

::;

8
+ 9r
r

(8)2 600
r
r

( 1).

To prove property

2 r we have

= -- +
z-(
1

It follows that

Thus, by definition of <Pc, z),


(

(z-()2

(+

z2

--

( + =

or

we expand h(z) about (E D(O, r). For lz -(I

(2),

h(z)

<

>

( )
-=(3
h"

00

= f3 - ( .

17 (z) = <t>c. (z)

1
- --

z-(

(z-()3

is bounded whenz is large. It follows that r7 has a removable singularity at


z = oo; that is, 17 is holomorphic on C \ E. Ifz E C \ E and z
l l < r, then
lz - (I < 2 r and
By property

( 1)

this is

:S 5000r2.

The maximum modulus theorem now implies that

l11(z) I :S 5000r2

or

5000r2

for allzEC\E.
:S
</> ( )z-(
lz-(l 3
c,z
This is property (2). The joint continuity assertion in property
mediate from the definition of ((,z).
Lemma

12.2.4. If is continuously differentiable on C and

compact set, then

( )=
</>z

(where (

+ iry)

_!_Jr { (8/8()(() dd17


7f

for allzEC.

Jc

(-z

(3)

is im
D

vanishes off a

12.2.

373

Mergelyan 's Theorem

Proof. Let {z: (z)

i=- O} D(O, R). Then Green's theorem (see Appendix


A) gives, for any z E D(O, R), that
</J(z)

(()

J
2rri

laD(O,R)

- z

d(

_!_ff,D(O,R)

(8/8()(()

7r

(-z

dd17.

The first integral vanishes since = 0 on 8D(O, R). Since R can be taken to
be as large as we please, the result follows.
D
Lemma

12.2.5. Define

A(z)
Th en, for any 0

< r < oo,

(1 - lzl2) 2

if lzl 1
if lzl > 1.

the function
Ar(z)

3 =

rrr2

(z)

A r

satisfies the following:


(1)

Jfc Ar(() dd17

1.

(2) Ar E C(C).
(3) If PE C and f is holomorphic on D(P, r ) then

JJ f(P

()Ar(() dd17

Proof. Note first that

Jfc >..( () dd17

rr

f(P).

Now property (2) follows by direct verification. Also property (6) follows
from property (4) since the support of 8Ar/8( is contained in D(O, r ) . Prop
erty (4) follows from the observation that

IV7>..I
together with the chain rule.

2(1 - lzl2) 2lzl

374

12. Rational Approximation Theory

For property (1) we use a change of variables:

Jk Ar(() ddry r2 Jfc Ar r( dd17 Jk Ai(() ddry


(

To establish property (3) we use polar coordinates with


typographical convenience:

f(P - ()Ar(() ddry

1.

a 3/7r
=

for

for fo2rr f - sei9 Ar sei9 s d()ds


r-2a for >..(s/r)s fo2rr f - sei9) d()ds
r-2a for >.. s /r s27r ds
27r a fol >.. s s ds
a Jk )...(() ddry
(P

(P

J(P)

J(P)

( )

J(P)

J(P).
Finally property ( 5) holds because

r f 8 Ar dxdy
JJe 8 x

{
JR

and likewise for

j 2r 88Axr dxdy { Ar(2r, y) - Ar(-2r, y) dy


JR
2r
r f 8 Ar dxdy.
Jle ay
=

Now we proceed as follows. Given an f E C(K) which is holomoro

phic on K, we extend f continuously to all of C (by the Tietze extension


theorem-see Appendix A). The extension may be taken to have compact
support (i.e., the closure of the set where the extended function is nonzero
is compact). Continue to designate the extended function by the letter f. If
8 > 0, we define the modulus of continuity off to be

w(8)

sup

lz-wl-S.5

lf(z) - f(w)I.

Since f vanishes off a compact set, it follows that f is uniformly continuous.


0.
Therefore lim6-+0+ w(8)
For

>

0 define

F(z)

Jk Ar(z - () f(()ddry.
K) r}

Let U
{z E K: dist(z, C \
>
and H (suppF) \ U (where suppF,
the support ofF, is the closure of the set where F does not vanish).
=

12.2. Mergelyan 's Theorem

375

Lemma 12.2.6. The function

FE C(C);
(2) F(z)
f(z)
(3) lf(z)-F(z)I

satisfi.es

(1)

for all

(4)

.::;: w(r);
(z) .::; 4n(r)

(5) F(z)

-
7r

z EU;
for all

z EC;

(aF/a{,)(() d d
'f
J./H (-z ry,
F

Proof. The function

for all

z EC

has compact support because

and

Ar

do.

The

continuom; differentiability follows from

8F
ax

1.

F(x + h, y)-F(x, y)

r { Ar(z
r { Jle ,\r(z
Jle

1111
h-+O

lim

+h

h-+O

Iim

(-Ar(z-() f(() ddry

+ h-()

h-+O

- Ar(z-() f( ddry
()

(see Appendix A). This last expression equals

and likewise

aF
ay

" { 8 r(z-() f(() ddry.


J.le ay
.X

Next, we obtain property (2) of Lemma


Lemma

12.2.5

since if

For property

F(z) =

(3)

z EU,

then

12.2.6

from property

f is holomorphic on D(z, r ) .

we write

f[ Ar(z-()f(() ddrJ = J[ f(z-(),\r(() ddry.

It follows that

IF(z) - f(z)I

Ilic (f(z-()-f(z)),\r(() ddT} I

(where we have used property (1) of Lemma

12.2.5);

.::;: JI w(r) r(( ) ddT}

this is

.X

(since the integrand has support contained in


=

w(r).

D(O,

r))

(3)

of

12. Rational Approximation

376

Theory

Next,

ff 8;; (z - ()f(() ddT/


ff 7 (()f(z - () dd11
ff 7 ((){f(z - () - f(z)} ddT/

oF
oz

(by property

(5) of Lemma 12.2.5). Hence

I':: I ff I 7 (()I w(r) dd11 ; w(r)


4

by property

(6) of Lemma 12.2.5. This gives property (4) of Lemma 12.2.6.

Finally, Lemma 12.2.4 says that

F(z)

-
7r

{ ( oF/o()(()
dd11
le ( - z

J
r

since F E c; but property (2) of Lemma 12.2.6 says that oF Io(


and on C\(supp F). Thus property (5) of Lemma 12.2.6 follows.

0 on

Final Argument in the Proof of Mergelyan 's Theorem. Cover H by


finitely many discs D(P1, r'), ..., DjPk, r'), where r' is slightly larger than
r, such that no Pj lies in K. Since C \ K is connected, there is for each j a

piecewise linear path 'Yi contained in C\ K which connects Pj to oo. Let ::Yi
be the image of 'Yi each j. Then j =::Yin D(Pj,r/2) has the property that

1, ... , k. (Here
diam Ej r/2, Ej n K 0 and C\ Ej is connected, all j
if fj n D(Pj,r/2) is not connected, we take its initial component.)
=

Nov:_ Lemma 12.2.3 supplies for each j a function


and

z EC\ Ej, such that


1(z)i
1

'
5000r

t Jr f
j=l j
7r

Hj

( ** )

efin:_ Hj H n D(Pj,r) and make disjoint sets by setting H1


H2 \ H1, and so forth. Define

G(z)

E D(Pj, r)

6 0
<

I ((z) - z - ( I < lz - (13


j

(z), (

H1, H2

l!_ i((, z ) dd11.


0(

Then it is clear from the construction of the </) that G is holomorphic on


the complement of LJ;=l Ej. In particular, G is holomorphic on an open

12.2. Mergelyan 's Theorem

377

neighborhood V of K. We claim that


desired provided that
For

approximates f on K as closely as

is sufficiently small.

EV,

IG(z)-F(z)I

k { J1

I:

1
-

. l
J=

1T'

(by property

(5)

<tJ(. c,, z) dd'f/ 1 J1H . 88F(,

8F
HJ. 8(,

--=.

of Lemma

(by property
<

(4)

12.2.6).

4w(r)
r
+

k "

I:

1
- dd'f/
Z - (

This, in turn, is

k r \q}(c,,
J

J=l
.

of Lemma

--=.

1T'

:::; . 47rw(r) I:
7r

1
z) - __ dd'f/
z-(,

r
}H

12.2.6)

600
r.
dd'f/
JH1nD(z,2r) r

J
: r2 dd'f/
4w; r) t r f
J}Hj\D(z,2r) I (I3
j=l

o o

O(w(r)) ,

j=l

(where we have used

(*)

( ** ) above, and we have estimated the


1/ lz - (I in the obvious way by O(w(r))).

and

coming from the integral of

last expression is dominated by

2400w(r)
r2

(area

D(z, 2r))

20000rw(r)

Ji

lz-(l2r I Z - ( l3

dd'f/

term
This

O(w(r)) .

The estimation of this last integral is an easy exercise with polar coordinates.
Thus we obtain

IG(z)-F(z)I

:::; 96007rw(r) + 200007rw(r) + O(w(r))


296007rw(r)

O(w(r)),

all

EV.

But we already know that

IF(z)-J(z)I :::; w(r),


G

Now

all

z.

is holomorphic in a neighborhood of K so we may apply Runge's

theorem to obtain a holomorphic polynomial satisfying

IG(z) - p(z)I

<

w(r),

all

EK.

Combining the last three lines gives

IJ(z) - p(z)I :::; 29602 7r w(r)

Since

w(r)

O(w(r)),

is as small as we please, provided

theorem is proved.

all

EK.

is small enough, Mergelyan's


0

12.

378

Rational Approximation Theory

We leave it as an exercise for the reader to check that our proof actually
yields the following more general result.

Theorem 12.2. 7. L e t K

be

compact and suppose that C \ K has only


0

finitely many connected components. If f E C(K) is holomorphic on K and


if E > 0, then there is a rationa l function r(z) with poles in C \ K such t h a t
sup

zEK

lf(z) - r(z)I <

E.

Indeed, only the final argument in the proof needs to be modified so


that sets

Ej

are selected in each component of

C \ K.

The rest of the proof

remains exactly as given.


The proof of Mergelyan's theorem that we have presented here is based
on the exposition in [RUD2] and on some unpublished notes of T. W .
Gamelin [GAM3].

12.3. Some Remarks about Analytic Capacity


Our purpose here is to give a few definitions and elementary results about a
concept called analytic capacity and to connect this idea with the proof in
Section 12.2. The point is that capacity theory is the natural framework in
which to consider problems in "rational approximation theory" -the theory
of approximation of a holomorphic function by rational functions.
If

E C

is compact, then we let UE be the collection of functions

holomorphic on

C\E

the analytic capacity of

and such that

f(oo)

and sup

lf(z)I :::;

1. Define

to be

-y(E)

sup

fE/AE

lf'(oo)I.

The set function 'Y measures in some sense the "capacity" of

C\E

to admit

nonconstant, bounded holomorphic functions.

EXAMPLE 12.3.1. If

-y(E)

{zo}

is a set consisting of a single point, then

0.
f E UE. Then the Riemann removable singularities
f continues analytically to all of C Since f is bounded,
constant. Thus J' (oo)
0.

To see this, let


theorem implies that
it follows that

is

The same argument as in Example 12.3.1 shows that any finite set has
zero analytic capacity. It would be incorrect, however, to try to deduce this
assertion from some "subadditivity'' property of analytic capacity; it is not
known in general whether analytic capacity is subadditive in the sense that

-y(E1 U E2) :::; -y(E1) + -y(E2).

It is easy, however, to see that 'Y is monotone.

12.3.

Some Remarks about Analytic Capacity

Proposition 12.3.2.

Proof.

379

If E1 E2 Care compact, then

This is immediate from the inclusion


D

Let E be a compact subset of C. If E C is connected, if C \ E


is connected, and if E contains more than one point, then E is called a
continuum. Any compact, convex set containing more than one point is a
continuum. So is the image of any continuous curve : [O, 1] ----+ C that
is one-to-one. [It requires proof that C \ ( [0, 1]) is connected in this case:
The proof involves techniques of topology that are beyond the scope of the
present book. See [WHY].] The next proposition gives us, in principle, a
1!chnique for calculating the analytic capacity of a continuum. Recall that
C \E is conformally equivalent to the unit disc if E is a continuum (Exercise

21).

If E C is a copact continuum, then 'Y(E)


where
is
a
conformal
mapping ofC\E to the unit disc D satisfying
oo
f
lf'( )I
f(oo) O.
Proposition 12.3.3.

Proof.

If g

UE

is a "competitor", then consider


go f-1

h: D----+ D.

Notice that h(O)


0 so Schwarz's lemma says that lh(z)I :::; lzl or lg(z)I :::;
lf(z)I. As a result,
=

lg'(oo)I
Proposition 12.3.4.

in fact equal to

lim lzg(z)I:::; lim lzf(z)I


z---+oo

z---+oo

lf'(oo)I.

The analytic capacity of a closed disc of radius

is

r:

'Y(D(P, r ) )
Proof.

= r.

Since D(P, r ) is a compact continuum, we need only notice that


f(z)

r
=

is a conformal mapping of C \ D(P, r ) to D which satisfies f(oo)


f'(oo) r, the conclusion follows by Proposition 12.3.3.
=

Theorem 12.3.5. IfE

C is a compact contin uum , th en


diamE
4

:::; 'Y(E) :::; diamE.

0.

Since
D

12. Rational Approximation Theory

380

Proof. Let f : C\E


12.2.2 tells us
If

D be a conformal map taking oo to 0. Then Lemma


that IJ'(oo)I 2:: ( diamE ) / 4. This proves the left inequality.
---7

diamE, then E is contained in some disc

D(P,r)

of radius

r.

Thus, by Propositions 12.3.2 and 12.3.4,

1(E)::; 1(D(P,r))

r.

The astute reader will have noticed that we called upon Lemma 12.2.2
to prove Theorem 12.3.5. Of course Lemma 12.2.2 was the critical estimate
in the proof of Mergelyan's theorem.
produce

( easy )

As an exercise, you may wish to

examples to see which of the inequalities in Theorem 12.3.5

is sharp.
Notice that the existence of the sets Ej at the end of the proof of
Mergelyan's theorem guarantees that, for each

1(D(P,4r) \ K)

PE 8K,

( some j)

>

1 (Ej )

>

( diamEj ) / 4

>

r/8.

The fact that C \Ej supports holomorphic functions with large derivative
at oo makes possible the construction of the

which in turn leads to the

construction of the function G that is near to f but holomorphic on a neigh


bor hood of K.
The preceding discussion is intended to provide motivation for the fol
lowing capacity-theoretic version of Mergelyan's theorem:

Theorem 12.3.6. Let K C be compact. Suppose that there is a C

such that, for every t5

>

0 and

>

every PE aK,

1(D(P, 15) \ K) 2:: C <5.

Then any f E C (K ) which is holomorphic on K an be uniformly approximated on K by rational functions with poles in C \ K.
A detailed exposition of Mergelyan's theorem and related ideas from

this point of view are given in [ GAMl, GAM2] .

We conclude with an elegant corollary, which is immediate from The


orems 12.3.5 and 12.3.6. The corollary is in fact one of Mergelyan's formu
lations of special cases of his theorem.

Corollary 12.3.7. If K C is compact and if the comp on ents Uj ofC\K

satisfy diam Uj 2:: <50


0

>

0 for all

j, then any f E C (K ) which is holomorphic

on K can be uniformly approximated on K by rational functions with poles


off K.

Exercises

381

Exercises
1. Let K

D(4, 1) U D(-4, 1) and L


D(4i, 1) U D(-4i, 1). Construct
U1} of entire functions with the property that fj -----; 1 uni
on K and fj -----; -1 uniformly on L.
=

a sequence
formly

2. Construct a sequence

of entire functions with the following prop

{gj}

erty: For each rational number q that lies strictly between 0 and

{gj} converges
O}.

sequence

{reiq1r : r

hj -----; 1 uniformly on
hj does not converge
[O, 1)

-->

such that limt-->1functions

hj

the

uniformly to q on compact subsets of the set

>

3. Construct a sequence

4. Let 'Y :

{h1}

{h1}

of entire functions with the property that

compact subsets of the open right half plane, but


at any point of the open left half plane.

re be a non-self-intersecting, piecewise linear curve

1(t)

Prove that there is a sequence of entire

oo.

such that limj--.oo lh1b(t))I

---; 0 uniformly on compact subsets of re

1/(1 - t) for each t


\ b( t) : 0 :::;: t < 1}.
=

but

5. Complete the following outline to construct a holomorphic function on


the disc with pathological boundary behavior.

< R1 < R2, E > 0, and if f is holomorphic on a neighborhood of


D(O R2), then construct a holomorphic polynomial p(z) such that
(i) lp(z)I < E for lzl :S: R1;
(ii) for each () E [O, 27r) there exist points s1, s2 E [R1, R2] such
that lf(s1ei8) + p(s1ei8)I < E and lf(s2ei8) + p(s2ei8)I > 1/E.
(b) Let 0 < R1 < R2 <
l. Use ( a ) to define inductively a
sequence of holomorphic polynomials Pj (z) such that
(i) IP1(z)I :S: 2 j for lzl :S: Rj;
(ii) for each() E [O, 27r) there exist points s1, s2 E [Rj, Rj+i] such

(a)

If 0

-->

that

LPt(s1ei8)

<

j
T ,

f=l

j
LPt(s2ei8)

>

1
2 .

f=l

(c )

Conclude that
00

LPt(z)
=1

converges normally to a holomorphic function

on

D.

382

Exercises

( d)

Conclude that the holomorphic function


that for no () E

[O, 271") does

f in (c)

has the property

J(rei9)

lim

r-+1exist.

6. Use Runge's theorem to prove the following result:

If n <:;;; tC is an open set and if f : n--+ tC is holomorphic, then


ere exists a sequence { rj} of rational functions with poles in
tC \ n such that rj
f normally on n.
Let E
[a, b] <:;;;JR be a line segment. Calculate r(E) explicitly.
Let E be the closure of {z E D : Rez > O}. Compute r(E) explicitly.
--+

1.
8.

9. Prove that if Ei, E2 are disjoint compact sets which can be separated
by a line (with n Ei

for

r E1 u E2) :::;

1, 2),

then

'Y(E1 ) + r(E2).

It is a famous open problem to determine whether analytic capacity


satisfies such a subadditivity property in complete generality.

10. Prove that if E <:;;; tC is any nonempty compact set, then there must exist
an f E UE such that lf'(oo)I
r(E).
=

11. Compute explicitly the analytic capacity of each of the following sets:

D(O, 2) \ D(O, 1),


(b) 8D(O, 1),
(c) 8D(O, 1) n {z : Rez 2'. O}.
(a)

12. If E <:;;; tC is compact, then define VE to be the collection of those con


tinuous functions on
and satisfy

'Y( oo)

which are holomorphic on

C \ E,

bounded by

1,

0. We define the continuous analytic capacity of E

to be
sup

fEVE

IJ'(oo)I.

Compute the continuous analytic capacity of

( a)

D(P, r),
(b) a line segment [a, b],
(c) a point,
( d) a circle {z E tC : Iz - PI r},
( e ) {z E tC: lzl :::; 1, (Rez) (Imz)
a closed disc

13. Refer to Exercise

12.

O}.

Can you prove a version of Theorem

12.3.5

for

continuous analytic capacity?

14. If E <:;;; tC is compact, then let


relating r(E) to -y(2E).

2E

{ 2z : z E E}. Derive a formula

383

Exercises

Pk(z) that converges at


every point of the plane to some function g and so that g takes the value
zero on some dense set and g takes the value 1 on another dense set.
[Hint: Use the Baire category theorem to see that the pointwise limit of
continuous functions must be continuous on a dense set.]

15. Prove that there is no sequence of polynomials

J is holomorphic
in U \ E and bounded, then what capacity-theoretic condition on E will
allow you to prove that f continues analytically to U?

16. Suppose that U S: re is open and ES: U is compact. If

17. Prove that there is an entire function

lim

f such that

f(x)

+oo,

f(x)
lim f(iy)
y-++oc

-oo,

x -+ + oc

lim

x-oo

lim

y-+-OC
*

f(iy)

1,
-1.

D(aj,rj) be pairwise disjoint closed discs in D(O, 1) such that the


union of discs LJ,1=1 D(aj, Tj) is dense in D(O, 1). Let K
D(O, 1) \
(LJ,i= 1 D(aj, rj)). Prove that such discs can be chosen so that L j rj < 1
and that in this case the conclusion of Mergelyan's theorem fails on K.
(This is the famous "swiss cheese" example of Alice Roth-see [GAM2].)

18. Let

D(aj,rj) be pairwise disjoint closed discs in D(O, 1) which accumu


late only at an. Is it possible to select aj, Tj so that the set D(O, 1) \
(LJ_i=1 D(aj, Tj)) = K satisfies the hypothesis of Theorem 12.3.6? [Hint:
See [GAM2].]

19. Let

20. Prove that there is a holomorphic function

f on

D such that for each

fixed Oo E [O, 271') it holds that f ( {reiOo : 0 r < 1}) is dense in tC.
[Hint: Refer to the techniques explained in Exercise 5.]
21. Prove that, if E is a compact, connected, proper subset ofre that con

tains more thn one point and such that the complement re \ E is con
nected, then re \ E is biholomorphic to D via the following steps.
( a) Without loss of generality, oo E E.
{b) EU {z Ere: lzl 2: R} is connected for each R 2: 0.
( c ) Fix a point Po not in E. By Theorem 11.4.1, if R is large, then the
connected component of {z Ere: z cf. E, lzl < R} that contains Po
is simply connected.
{d) The union over R > Oof the sets in part ( c ) is simply connected
and this union equals re \ E.
(e) Apply Riemann mapping (Theorem 11.3.1).

Chapter 13

Special Classes of
Holomorphic Functions

Complex function theory is a prototypical example of a successful mathe


matical theory: A class of analytic objects is defined by simple and natural
considerations-in this case, holomorphic functions-and the objects turn
out to have remarkable and profound properties. These properties in turn
illuminate other parts of mathematics. Complex analysis provides insight
into real-variable calculus, for example, as in the evaluation of definite in
tegrals by residues in Chapter

4.

It also enables one to prove topological

results, such as that a topologically simply connected domain in the plane


is homeomorphic to the unit disc (Chapter

11).

Such connections to other

areas are a strong indicator of mathematical vitality.


Successful mathematical theories tend to have a strong component of
purely internal development as well: Special topics are heavily investigated
for their own intrinsic interest, whether or not they have any obvious rele
vance to other areas or even to the main lines of the general development of
the theory from which they grew. Many times these technical explorations
produce results that have important applications and interest in other con
texts. The fact is that one never knows what will arise, or what will be of
value, until one gets there.
Section

13.1

is devoted to some specialized topics, as just described,

that arise in complex analysis. Sections

13.2-13.4 are also devoted to special

questions, but the utility of these questions may perhaps be more evident.
The boundary extension of conformal mappings (see also Chapter

14)

is

critical in transferring function theory from a simply connected domain to


a canonical domain-namely, the disc.

Less obvious is the fact that the


-

385

13. Special Classes of Holomorphic Functions

386

theory of Hardy spaces has profound connections with Fourier series and
other parts of harmonic analysis; in the last thirty years, Hardy space theory
has wrought a considerable transformation on all of real analysis.

13.1.

Schlicht FUnctions and the Bieberbach Conjecture

A holomorphic function

is one-to-one.

normalizations

schlicht

on the unit disc

is usually called

We are interested in such one-to-one

f(O)

0,

f'(O)

1.

schlicht

if

that satisfy the

In what follows, we restrict the word

to mean one-to-one with these normalizations. Such a function has

a power series expansion of the form

f(z)

00

L aJ

j=2

j
z .

If 0 :::; () < 27r, then the Koebe functions

fo(z)

are schlicht functions which satisfy

(1

Jaji

+
=

z
eiOz)2
j for all j. It was a problem of long

standing to show that the Koebe functions are the only schlicht functions
> 1, Jaji
1. It was further conjectured that for
f it holds that Jaj I :::; j. This last problem was known as the
Bieberbach conjecture. Both of these problems were solved in the affirmative
by Louis de Branges in 1984.

such that, for some j

any schlicht

In the present section, by way of introducing the subject, we will prove


two of the oldest results in the theory of schlicht functions. The first is that

Ja2J :::;

2. The second is the Koebe

1/4

of a schlicht function always contains

theorem to the effect that the image

D(O, 1/4).

This result of Koebe finds

applications in many other parts of complex analysis: For instance, we used


it in Section 12.2 to prove Mergelyan's theorem and in our study of analytic
capacity. [The proof of Koebe's theorem that we present in this chapter is
independent of those earlier considerations.] The proofs in schlicht function
theory are very striking for their geometric character. The reference [DUR]
contains further results on the theory of schlicht functions.

Lemma 13.1.1. Let

be schlicht and write

h(z)
For 0 <

<

1
f(z)

1
;

<

JzJ

00

L bnzn.

n=O

1, Jet
Ar

{z : r

<

1}.

13.1. The B ie b er b ach Conjecture

387

Then there is a constant 7J > 0, independent ofr, such that, for all sufficiently
small r, h( Ar) is contained in an ellipse Er with major axis

( t bi r) J
2 (t- lb1lr) J
+I

2a= 2
and minor axis
2{3=

1 + ryr3

i+

Proof. It is convenient to assume that

b1

h(z) by e-i9h(ei9z) for some real()


to arrange that bo= 0. Write

replace

h(z)

1
=

ryr3.

is real and nonnegative (if not,


to make it so). Also translate

+ b1z+ <j;(z),

where is an "error term" defined by this equation, which vanishes to order

at 0.
Set

'
a =

I/r+ b1r, {3' = I/r - b1r.

Writing

z=

into its real and imaginary parts, we have

h(reiO)=

((t

+ rb1

) (( t

cosO+ e + i

reiO

+ rb1

and breaking

sinO+Im

Thus

(Reh(rei9))2 (Imh(rei9))2
+
'
{3' 2
a 2
Since vanishes to order

_
-

cos()+

at 0, and since

we see that

I I
I; I

'

) (

R e 2
a,

'
a ,

7Jor3'

7Jor3

{3'

2'.

.
sm

for

small and some constant


In conclusion,

7J

>

0.

{3'

1/(2r) when r

7Jo > 0. Hence


(Imh(rei9))2
(R e h( rei9))2
+
'2
{3'2
a
(I cosOI + 1Jor3)2 +(I sinOI + ry0r3)2
<
1 + ryr3

for some constant

{}- Im

is small,

388

13. Special Classes of Holomorphic Functions

Thus h(oD(O,r)) lies in the specified ellipse. Since h(D(O,r)) is a neighbor


hood of oo, and since h is one-to-one, it follows that h(Ar) lies inside the
ellipse.
0
Lemma 13.1.2 (Lusin area integral). Let 0 <:;:; C be a domain and
C a one-to-one holomorphic function. Then (0) is a domain and

area((0)) =

in

---t

1'(z)l2 dxdy.

Proof. We may as well suppose that the areas of 0 and (0) are finite:

The general result then follows by exhaustion. Notice that if we write =


u +iv= ( u., v) = F, then we may think of (z) as
F: (x, y)

(u(x, y), v(x, y)),

an invertible C2 mapping of 0 <:;:; JR2 to F(O) <:;:; JR2. The set F(O) is open
by the open mapping theorem (Theorem 5.2.1); it is also connected since it
is the image of a connected set. Therefore F(O) is a domain. The Jacobian
of F is

also

ouov
det Jac F = ox oy

ov OU
--.
ox oy

By the Cauchy-Riemann equations we obtain


det JacF =

( ou) + ( ov ) =
2

ox

ox

I I2.

Thus the change of variables theorem (see Appendix A) gives

<t>(rl)

as

1 dxdy = { det JacF dxdy = { 1'(z)i2 dxdy

Jn

Jn

desired.

The Lusin area integral arises frequently in complex and harmonic anal
ysis. We use it in the next section to study the boundary behavior of con
formal mappings. We use it now to prove one of the oldest results (due to
Gronwall) in the subject of schlicht functions.
Theorem 13.1.3 (The area principle). If f is schlicht and if

h( z )

.
1
1
00
=; + L bjzl,
f(z)
O

J=

13.1. The

389

Bieberhach Conjecture

then
00

I:jlbj12::; i.
j=l
Proof. We use the notation in Lemma

(1+1]T3) 2:

7r

r2

- lb1l2r2 (1+1]T3)

2: area(h(Ar))
by Lemma
obtain

31 .1..
2

31 .1..
1 Now

7ra

/3

area( Er)

{ lh'(z)l2 dxdy
}Ar

We write this last expression in polar coordinates to

Of course, when we multiply out the square and integrate in(}, then all the
cross terms vanish (since

;,(1+r3)

J eikO d(}
")>

2n

7r

[(

if

k =/= 0).

s-3+

1
- +
1
r2

00

Hence

)
}

J 2lb;l2s2j-l

ds

jlb1l2(l - r2J)

or

+1r'rJT 2:

7r

or

00

7r

LJlb1l2(l - r2j)
j=l

1 1]T 2: I:jlb1l2(1 - r2j)


+
j=l
for any large N. Letting r

--4

o+ gives
N

1 2: I:jlb112.
j=l
Letting N

--4 oo

Lemma 13.1.4. If f

f(z2).

finally yields the result.

is schlicht, then there is a schlicht g such that g2(z)

D
=

390

13. Special Classes of Holomorphic Functions

Proof. Write
and that

r(z)

f(z)

=z

(z),

holomorphic on D. Notice that

has a holomorphic
g(z) = z r(z2). Then

is nonvanishing on D. Thus

on D satisfying

r(O)

g2(z)
as required. Notice that

= 1. Define

(O)

square root

= z2r2(z2) = z2(z2) = f(z2)

g'(O)

= 0 and

g(O)

=1

(*)

= r(O) = 1.

g is one-to-one. If g(a) = g(b), then f(a2)


2
so
= b . Thus a = b. If a = b, then we are done. If a = -b, then
the definition of g shows that g(a) = -g(b) whence g(a)
g(b) = 0. Since r
0
is nonvanishing, this forces a = b = 0.
We need to check that

f(b2)

a2

Theorem 13.1.5. If f is schlicht, with


00

f(z)

L a1z1,

= z+

j=2

Proof. By Lemma 13.1.4, there is a schlicht function

g2(z).

such that

Write
1

g(z)

1
=-+

00

bz1.
L
1
.

]=0

Then the area principle gives


00

I:jlb112::;
j=l

1;

hence

Now
1

(
g z)

1
2bo
=2+ -+

(b + 2b1) + ( higher

order terms )

while
1

f(z2)

z2 + a2z4 + ...
1

z2

1
2 (1

- ( -a2z2 -

- a 2Z 2 +

).

Since these two expansions must be equal, we find that

b + 2b1

= 2b1 = -a

bo

= 0 and

f(z2)

13.l. The Bieberbach Conjecture

391

or

la2I

2l b 1 I :-::;2.

Theorem 13.1.6 (The Kobe 1/4 theorem). If f is schlicht, then

1)) 2 D(O, 1/4).

J(D(O,

Proof. If

J(D(O,

1)), then we construct the auxiliary function

g(z)

f(z)
f(z)

1-ct

(you should take note of this trick: It is a useful one). We claim that
schlicht. Assuming the claim for the moment, we apply Theorem
see that

But

g(z)

I
f(z). (

g" 0
<2.
( 2 )

--

1+

f(z)

(z + a2z2 +
z

Hence we have, from Eqs.

la2I :-::; 2,

J2 (z)

+ ...

(1 + ;- + )

(13.1.6.2)

(13.1.6.1) and (13.1.6.2), that

l a2 1

we conclude that

1 1

or

J(D(O,

:-::;2.

:-::;4

In conclusion,

C\

is

(13.1.6.1)

(a2 ) z2
+

Since

13.1.5 to

1))

{ z: lzl l}

whence
J(D(O,
To verify the claim that
that g ( O )

0, g' (O )

1)) 2 D 0, 1/4) .

is schlicht, note that Eq.

(13.1.6.2) shows

1. Also the function g, being the composition of J

with a linear fractional transformation, is one-to-one.

13. Special Classes of Holomorphic Functions

392

Remark: Notice that if

and

of

J'(O)

f is a ( normalized )

schlicht function, so that

by the inverse function theorem

1,

( see [RUDI]),

f(O) =

the image

will contain an open neighborhood of the origin. The same conclusion

also follows from the open mapping principle.

Koebe's theorem estimates

the size of that neighborhood.

13.2. Continuity to the Boundary of Conformal Mappings


Let

01, 02 be domains in C

and let

c.p: 01

--+

02 be a conformal ( one-to-one,

onto, holomorphic ) mapping. It is of both aesthetic and practical interest


to consider whether

802).

c.p extends

continuously to

801 ( and,

One might expect that, the nicer the boundaries of

likewise,

01

and

1
c.p- to
02, the

more likely the existence of a continuous extension.


It turns out that a continuous extension to the boundary exists virtually
all the time, and the extension is more regular when the boundary is more
regular. Here are two extreme examples:

EXAMPLE

13.2.1.

The conformal map

c.p: D(O, 1)--+ {z E D(O, 1):


given by

lm z

>

O}

/2
(i(l-z))l
l+z
( )=
c.pz
l/2
(i(l-z))
1- i
l+z
1+ i

------

extends continuously but not differentiably to 1. Of course the boundary


of

{ z E D( 0, 1) :

Im z >

0}

is only piecewise differentiable: It has corners at

1.
EXAMPLE

13.2.2.

Any conformal map of the disc to the disc has the form
o
z--+ ei o .

some Oo E

[O, 27r)

and a E

D.

z- a
1-az'

---

Such a map continues holomorphically past

every point of 8D. Notice that 8D is real analytic.


These two examples are indicative of some general phenomena.

We

mention three:

(1)

If

801, 802

are Jordan curves

( simple closed
801.

curves ) , then

c.p ex

tends one-to-one and continuously to

(2)

If

801, 802

are c=, then

c.p extends

fashion that the extension is C00 on

(3)

801, 802 are real analytic,


801.

If

then

one-to-one to

801

in such a

01.

c.pcontinues holomorphically past

13.2. Boundary Continuity of Conformal Mappings

Assertion
Section

14.5.

(1)

is proved in this section.

Assertion

(3)

Jordan

Assertion

(2)

is proved in

follows from the Schwarz reflection principle

15.

and is considered in Exercise


Recall that a

393

simple closed) curve is a continuous function


1( s)
1(t) if and only if {s, t} {O, 1}. Set

(or

/ : [O, 1] - C such that


{T(t) : 0 :S t :S 1}. It is a celebrated result of Camille Jordan that
i
C \ i is the union of two disjoint open sets. One of these is unbounded and
the other (called the region bounded by /) is topologically equivalent to the
=

disc. In practice, there is no trouble (in any particular case, with the curve
given explicitly) verifying the Jordan curve theorem. The general result is
rather difficult and is best left to a topology text such as [WHY]. In this
chapter, we shall take the Jordan curve theorem for granted and prove the
next result in maximum generality:

Theorem

13.2.3

(Caratheodory). Let fh, 02 be bounded domains in

each of which is bounded by a single Jordan curve.

C,

If t.p : 01 - 02 is a

conformal (one-to-one, onto, holomorphic) mapping, then t.p extends con


tinuously and one-to-one to 801. That is, there is a continuous, one-to-one
function (p: 01 - 02 such that

4?101

t.p.

Remark: Clearly the domains 01, 02 in the theorem are each simply con
nected since each has complement with just one connected component. It
will be clear from the proof that in fact Caratheodory's theorem is valid
for conformal maps of domains that are bounded by finitely many Jordan
curves.

Proof of Carathedory's theorem. The proof will be broken up into sev


eral lemmas, which will fill out the rest of the section. First, let us remark
that the result fails completely for maps t.p : 01 - 02 which are one-to-one
and onto but not holomorphic: Even if t.p is infinitely differentiable in the
real variable sense, it may not extend continuously to even one boundary
point. [Exercise

16 explores this point.]

The extension-to-the-boundary situation changes if one or both of the


domains is unbounded.

D - {z : Im z

>

Consider, for instance, the Cayley transform t.p :


i (l - z)/(l + z). [See Exercise
for

O} given by ip(z)

17

more on this case.]


We shall in fact first prove the result in the special case that 01 is the
unit disc D. At the end we shall derive the general result from this special
one by the application of a few simple tricks.

Now let us set up the proof. Let 1 : [O, ] - D, 2 : [O, ] - D be


two curves satisfying 1(1)

2(1)

1,

1([0,

1))

D, 2([0,

1))

D.

Our aim is to see that limt---.1- ip(1(t)) and limt---.1- ip(2(t)) exist and are

394

13. Special Classes of Holomorphic Functions

13.1

Figure

equal. Once this result is established, we may define

ij)( 1)

to be this limit

(since it is independent of the choice of curve). Then cj; is automatically a


continuous extension of

an follows in the
n U an is easy to

'P to n U {1}.

The extension to the other points of

same way, and the continuity of the whole extension to


check.

Thus let 1, 2 be as above and

13.1.

Now

V has

finite area and

Notice that if

'P-l (K) n

is compact.

Equivalently,

n),

then

{'P(zj)}

n(l, 1/2). See Figure


'P(V) fh
then (since 'P-l is continuous),
if { Zj} is a sequence in n that

n(O, 1)

has finite area M since

02 is compact,

diverges to the boundary of


a point of

'P(V)

(i.e., has no subsequence that converges to

diverges to a02. In particular,

have compact closure in 02. Figure

13.2

suggests what

'P(V)

'P(V)

does not

looks like. The

standard mathematical terminology for this general situation is to say that

'P

is a

proper map. [A

continuous map is said to be

proper if

the inverse

image of every compact set is compact.]

V centered at 1 E an and rotated


an r small, there is a Oo
(}0 (r) such

Let us introduce polar coordinates on


by 7r from the usual orientation. Given

-Oo <
Figure 13.3.

that

(} <

Oo

describes the corresponding r-constant arc in V. See

Let
'Yr:

[-Oo(r), Oo(r)]

---+

V,

(}

1 - rei6.

Lemma 13.2.4. L et fr be the length of 'Po 'Yr Notice that

!Oo(r)
fr=

-Oo(r)

i'P'(l - rei6)ird0.

13.2. Boundary Continuity of Conformal Mappings

Figure 13.2

-6rfr)

<

Figure 13.3

Then

1 1/2 (fr)2

--

is fi.nite.

7rT

dr

<

0rfr)

395

396

13. Special Classes of Holomorphic Functions

Proof.

[1-8B0o r de] rrr dr


As a result,

1 1/2 (fr)2 dr

< oo.

--

rrr

As a consequence of the lemma, because


origin, it must be that there is a sequence Tj

l/r

----+

is not integrable at the

0 such that

fr1

----+

o+.

Lemma 13.2.5. For each such Tj, the limits


lim

8--+0o(ri)-

(1 - rjei8 )

and

exist.

Proof. Each

fr1

is finite.

It follows from the definition of limit that the

indicated endpoint limits exist.

The next lemma is a result about metric topology, rather than complex
analy sis as such.

Lemma 13.2.6. Let

77( o) '

that if a,

E T

with l a

having endpoints a,

bl

o, then there is one and only one arc of T

whose diameter is

Proof. It is convenient to parameterize


Let

'ljJ(eit)

an . There is a function
2
77( o) ----+ 0 as 0 ----+ o+' such

be the curve that describes

defined for all sufficiently small 0 > 0, with

77( o) .

with respect to the unit circle.

'ljJ is a one-to-one, onto


'ljJ has a continuous inverse.
11/J(() -1/J((')I Oo, then I( -('I < 2.

be such a parametrization of T. Note that

map of compact Hausdorff spaces.

Therefore

Let Oo > 0 be so small that whenever

11/J(() -1/J((')I 80 we let a be the unique


E C : Iz I
1} having endpoints (, (' . Let
p
'ljJ o a. As usual, we identify p with its image. By the continuity of 1/J- 1 ,
diamp----+ 0 uniformly for 11/J(() -1/J((')I--+ 0.
For(,(' in the circle with

shorter arc of the circle S


=

{z

13.2. Boundary Continuity of Conformal Mappings

80,

If 0 < 6 <

then we set

17( 8)
Then

17(8)

0 as (j

-4

397

sup{ diam p :

81

0. Let 0 <

-4

17( 81)

<

l'I/!(() - !((')I

<

80

15}.

be so small that

1 .

2 diam T.

Then the statement of the lemma holds for

8 '.'::: 81.

Definition 13.2. 7. Let notation be as in the proof of the last lemma. Let

a, b

E T with

la -

bl sufficiently small. The unique arc of

a, b and having dia meter

'.'::: 11(la - bl) is

called the

with endpoints

smaller arc of T

that joins

a to b. We denote this arc by Tab

{ TJ} selected

Lemma 13.2.8. With

as

in the paragraph preceding Lemma

13.2.5, th ere are for each j two possibilities:


802 3

aj

lim

(J-.0{)

=I=

cp(l - rjei9)

lim

o-.-ot

cp(l - rjei0)

bj E 802

(13.2.8.1)

or
802 3 lim

0-.00

cp(l - rjci0)

:=

lim

o-.-ot

PJ

cp(l - rjei9)
(13.2.8.2)

802 .

See F igure 13.4.

Proof. This is immediate from Lemma 13.2.5 and the properness of


Let us write r rj

smaller of

cp 0 'Yrj"

In the case of Eq.

( 13.2.8. l),

cp.

let Tj be the

the two boundary arcs of 02 connecting aj to bj. Then r r1 U Tj

forms a simple closed curve.


In the case of Eq.
member that

cp

r rJ U

(13.2.8.2),

is one-to-one).

In either case, r rj u Tj or r,.j u


For each

let Yj

{1 - reiO

{Pj}

0 <

{PJ}

surrounds a region
<

j,
wo

Proof. For each

wo

Wj.

Then

<P(YJ)

Tj
D \ Yj. Now fix
;p(zo) for some zo E D.

let
=

Wj.

an index j.
Either

zo E Yj, then we are done by connectivity.


cp(Tj) Wj. We will show that this is impossible

If

wj 02.

rj, IOI < Oo(r)}. Then for each j


w1 or cp(Yj) 02 \ M-'J.

there are only two possibilities: Either cp(Yj)

Lemma 13.2.9. For j sufficiently large,

is a Jordan curve in 02 (re

zo

If instead
for j large.

Select a point

Tj
zo

or
E

zo E Yj.
Tj, then

13.

398

Special Classes of Holomorphic Functions

Figure 13.4
Now
area[<p(T1)]

area[02] - area [<p(Vj)]


2
area[02] l<t?'l dxdy

---+ area [02].


In the penultimate line we have used the Lusin area theorem and in the last
line the fact that nj Vj (/J.
=

Let f1
fri. Certainly la1 - b1 I ::; f1. Thus, by Lemma 13.2.6, Tj has
diameter ::; 17(1) This last quantity tends to 0 by Lemma 13.2.6.
Let D1 be a disc of radius 1 + r1( f1) and having center a1. Then the
entire Jordan curve r Tj u Tj lies in Dj by the above estimates. Hence wj
lies inside Dj.
=

In conclusion,
area [Wj] ::;

7r

( f1 + 17( f1) ) ---+ 0

as j---+ oo. But we have proved that area(<p(T1))---+ area02. Thus it cannot
be that <p(T1) W1 for j large. We conclude that <p(T1) 02 \ W1; hence
D
<p(Vj) W1.

As j---+ +oo,
diam [Wj] ---+ 0

Lemma 13.2.10.

Proof.

and

Immediate from the proof of the preceding lemma.

the curves l : [O, 1] ---+ D satisfy t(l)


1, 2, then
lim <p(2(t)).
lim <p(1(t))

Lemma 13.2.11. If

t([O, 1)) D, f,

and

t->l-

t->l-

(Notice that we are asserting that each limit exists and that they are equal.)

399

13.2. Boundary Continuity of Conformal Mappings

Proof. Let

0. Choose J so large that diam

>

S1

(W1)

<

and that the

S1 <
cp(1(t)) E W1. Likewise, if S2
is sufficiently large, then S2 < t <1 implies that l2(t)
l l < r1; hence
cp(2(t))E W1. Therefore lcp(1(t)) - cp(2(t))I <
conclusion of Lemma 13.2. 9 holds.

l1 (t)

t < 1 implies that

If

is sufficiently large, then

< r1; hence

E.

This proves that

where { Q} is the singleton n Wj .

Lemma 13.2.11 provides the continuous extension of cp to 8D: If PE


8D, then choose a curve 'Y
(pP
( )

limt-.i-

cp('Y(t)).

[O, 1] ---+ D, -yl


( )

P,-y([O, 1))

D. Define

The last lemma guarantees that the limit exists

and is independent of the choice of "(. The continuity is nearly obvious;


think about why i:pP
( j) ---+ (pP
( ) when PjE 8D and Pj---+ PE 8D.

Lemma 13.2.12. L e t
D. L e t

;y 8D

F: D---+ C

be an open arc. If

be continuous 011 D and holomorphic on

Fl1

is constantly equal to c, then

c.

Remark: This result was proved in Example 7.5.3. Now we present another
proof.

Proof. We can assume that


composing

Fl1

0. If

F is not identically zero,

F with a Mobius transformation,

then, by

we can suppose that F(O)

i-

0.

Then Jensen's inequality yields

2n
IF(O)I _!.__ f log IF(rei6)ld0
27f lo
for all but countably many r <1 . As r ---+ 1 - , this leads to a contradiction.
-

oo <log

D
Lemma 13.2.13. The continuous extension (p of

cp

to D is one-to-one on

D.

Proof. Since (pD


( )

0, (p8
( D)

80, and

'Plv

is one-to-one, it is enough

to check that (pis one-to-one on 8D. Suppose that P, P'E 8D and (p(P)

(p(P') . Consider the two radial arcs running from 0 out to P and to P' . Call
these two arcs a and

'/3,

respectively. Our hypothesis implies that

is a closed Jordan curve. Let


( U
If D \a

'/J)

1 U2,

U U

be the planar region bounded by this curve.

then either

cp(U1)

13.5.
Say without loss of generality that
arc bounding U1. Then

cp ( i)

cp(a U '/J)

cp(Ui)

or

W.

See Figure

W. Let i be the circular


W n 802 is the singleton

802. But

cp(U2)

400

13. Special Classes of Holomorphic Functions

Figure 13.5

c.p(P)
c.p(P'). Hence 0 is constant on
c.p is constant-a clear contradiction.
=

the entire arc)'. By Lemma 13.2.8,

Finally, Lemma 13.2.11, the subsequent remark, and Lemma 13.2.13

c.p :

give the desired continuous, one-to-one extension of a conformal map


D----+ f12 to D.
Consider now the general case. If nI,

n2

are bounded, simply connected

nI

be

be a conformal mapping.

If

domains in C, each bounded by a Jordan curve, then let


a conformal mapping and let c.p2 : D ----+
c.p

: f11 ----+ f12

n2

'PI

D ----+

is any conformal mapping, then consider the diagram in Figure

13.6.
The function 'PI extends to a continuous map 0I of D to nI. Also,
01 lav is continuous, one-to-one, and onto anI. It follows that ( 01 lavr1 is
well defined and continuous ( exercise ) . Thus 011 : nI ----+ D is one-to-one,
onto, and continuous. A similar statement holds for
Next, W

c.p2

c.p o ip1

c.p2

: 02 ----+

D.

is a conformal map of D to D; hence W is a

rotation composed with a Mobius transformation. It follows that W extends


continuously and one-to-one to a function

: D----+ D. Finally,

defines a continuous and one-to-one extension of

c.p

This completes the proof of Theorem 13.2.3.

to

OI.

02

011
0

13.3. Hardy Spaces

401

Figure

13.6

13.3. Hardy Spaces

If < p < oo, then we define HP(D) to be the class of those functions
holomorphic on the disc and satisfying the growth condition

O<r<

sup

(-12 Jof21r lf(rei8)1Pd0)

/p

< oo.

(*)

7r

It is convenient to use the notation llJllHP to denote the quantity on the left
of ( * ) .

We also let H00(D) be the class of bounded holomorphic functions


on D, and we let llJllHoo denote the supremum off on D. [The notation
originates from the use of 00 for bounded functions, with the essential
supremum norm, in real analysis.]

In Section
we studied the H00 functions and characterized their
zero sets. We also discovered that every H00 function f has a factorization

9.1

f(z)

F(z) B(z)

where B(z) is a Blaschke product having the same zeros (counting mul
tiplicities) as f and F is a nonvanishing holomorphic function such that

llFllH00

llJllH00

It is clear that for any p we have HP(D) 2 H00(D). G. H. Hardy and


F. Riesz, in a series of papers published in the early part of the twentieth
century, discovered that HP(D) functions share many of the properties of

Special Classes of Holomorphic Functions

13.

402

H=(D)

functions.

In particular, it turns out that the zero sets of

H00(D)

functions are ust the same as those for


torization such as
present section.

(** )

is valid.

HP(D)

functions, and that a fac

We shall investigate these matters in the

We begin with some remarks about subharmonicity. Let

f be holomor
f is not zero,

phic on an open set U. In a neighborhood of any point where


log

lfl

is harmonic.

Therefore

lflP

exp p log lfl) is subharmonic when

p > 0 since the function exp is convex and nondecreasing cf. Example
and Exercise 50, Chapter
monic because

IJIP

7).

If

does have zeros, then

is subharmonic off the zero set of

sub-mean value property holds at the zeros of


since

IfIP

and the small circle

by the continuity of

IJIP,

is holomorphic on a neighborhood of a closed disc

D(P, r), and that his harmonic on D(P, r).


8D(P,
r). Then the maximum principle implies
IfIP
entire closed disc D(P, r).

that his continuous on

Also assume that h 2':


that

7.7.6

is still subhar

2': 0 every where.

Now suppose that

D(P, r),

lflP

IfIP::;

hon the

on

We will apply these observations in the next lemma:

Lemma 13.3.1

then

Proof. Let

( Hardy ) .

Let

be

holomorphic on D.

If 0

<

be the solution to the Dirichlet problem on

IfIP
lf1P l aD(O,r2)"
IfIP::; g on D(O, r2). Hence
f 21r
f 21r
(}
lo lf(r1eiO)IP d ::; lo g(r1eiO) d(}

boundary data
follows that

Then, since

Lemma 13.3.2.

If 0

then

P2

Proof. For any fixed p, Pl

<

p < P2 , define

fp(z)

oo,

(1

r2

<

1,

D(O, r2) with


D(O, 1), it

is subharmonic on

1
27r
1 lf(r2eiO)IP

<

<

<

27rg(O)
27r
g(r2eiO) d(}

Pl

<

r1

1
,
z)l/p

d(}.

403

13.3. Hardy Spaces

where the power is defined via the principal branch of the logarithm (this
makes sense because Re (1 - z ) >

0).

Also

Now for

<

7r/2

()

0/2

we have sin() ;::::

Hence the integral does not exceed

4 . 2PifPr -pifp
fp

since pifp < 1. Thus

HP2,

11r/2 0-pifp

HP1 (D).

00

fP <f. H .]
HP for different p

Thus

< 00

Another computation shows that

since p2fp > 1 (Exercise 18). [In case P2

since clearly

d()

(use Taylor's theorem).

oo

fp <f.

there is nothing to check

are distinct. To see the inclusion relation, we

use Jensen's inequality with exponent P2/p1 > 1. [There are two Jensen
inequalities-one from complex analysis and one from measure theory. This
time we are using the latter: See Appendix A .] We obtain

(27r)(PrP1)/p2
If

HP2,

(fo

27r

lf(reiO)IP2 d()

then the supremum over

so is that of the left. So

HP1

) pifp2

of the right side is finite, hence

We record now a technical fact:

Lemma 13.3.3. Ifp

>

0,

then there is a positive


logx

for all x

>

Cp x P

constant Gp

0.

Also, if 0

<x <

1, then there is a constant Co such that


1 -x

Co log(l/x).

such that

404

13.

Special Classes of Holomorphic Functions

The first inequality is clear if x :S 1. Now choose Cp 1/p. Then


the derivative for x 2'. 1 of log x does not exceed the derivative of CpxP.
Conclude by integrating.
Proof.

The proofof the second inequality is similar and we omit it.

If f E HP, 0 < p ::::; oo, and if a1, a2, ... are the zeros off
(listed with multiplicities), then 2:: 1 ( 1 - la1I) < oo.

Lemma 13.3.4.

Proof. We may as well suppose that la 1 I :S la2I :S


If f has a zero
of order m at zero, then notice that j( z )
f( z ) / zm is zero free and, by
Lemma 13.3.1, llfllHP
llJllHP In short, we may suppose without loss of
generality that f does not vanish at 0. We may also, by Lemma 13.3. 2,
restrict attention to 0 < p :S 1.
=

Choose 0 < r < 1 such that no la1I =rand let n(r) be the number of
zeros off in D(O, r). Then, by Jensen's formula,
loglf(O)I

2
1og !_ + 2- f 7r loglf(rei0)ldB.

I:. l a1 I

211" lo

J=l

But Lemma 13.3.3 says that, since p

>

0, we have

log lf(reiO)I ::::; Cplf(reiO)IP'

where Cp is a constant depending on p. Thus


n(r)

1
log !_ :S Cp- r
L
211" lo
aJ
.
J=l

I I

Therefore

f; I:; I

n(r)

log

hence (letting r

---?

2rr

lf(rei0)1P dB

loglf(O)I.

<'.

c,11f11, -log If (0) I ;

:S

Cpllflli-Ip -loglf(O)I.

1-)

00

j=l

.7

L log fal

Now the second statement of the preceding lemma gives the result.

Theorem 9. 1.5 gives a remarkable converse to Lemma 13.3.4: If {aj}


D satisfies
L 1 -la1I < oo ,
j

then there is an H00 function (indeed a Blaschke product, and certainly an


element of HP) with zero set {a1}. We shall use this fact now to obtain a
factorization theorem for HP functions:

13.3. Hardy Spaces

405

Theorem 13.3.5. If f

HP(D),

then there is a convergent Blaschke prod

uct B and a nonvanishing holomorphic F

on D

such that

f =FB.
Furthermore,

llFllHP

llJllHP

Proof. As we noted in the remarks preceding the statement of the theorem,


there is a Blaschke product
as

f.

with the same zeros, counting multiplicities,

Thus the Riemann removable singularities theorem implies that

F(z) =
is holomorphic on all of
theorem.

< oo.

Since the case

D. It remains to prove the final


p = oo has been treated already,

assertion of the
we assume that

Write

and

N
BN(z) = IT
J=l

Notice that

( by

f(z)
B(z)

Lemma

FN

J/BN

al
J

satisfies

13.3.1)
=

lim

r-.i-

( since BN

Ba1 (z).

1
-

27!'

1271" lf(re'.8)1P
o

d()

D and of modulus 1 on 8D). Now if 0 < r < 1 is


FN(rei8) ...... F(rei8) uniformly in (), so ( see Appendix A )

is continuous in

fixed, then clearly

<

Here we have used

(* ) .

This is half of the desired equality.

For the reverse inequality, note that

111.

IBI

l; hence

IF I

If/Bl>
D

406

13. Special Classes of Holomorphic Functions

13.4. Boundary Behavior of Functions in Hardy Classes

[AN OPTIONAL SECTION FOR THOSE WHO KNOW


ELEMENTARY MEASURE THEORY]
To treat 8D as a measure space, notice that there is a natural correspondence

[O,

between

27r)

and 8D:

\JI :

()

I-+

ei().

If S 8D, then Sis said to be measurable if and only if

w-1(8) is

Lebesgue

measurable. Also the measure of S is defined to be the Lebesgue measure


of

w-1(8).

(In practice the correspondence is written out explicitly so there

is little chance of confusion.) A measurable

< oo,

f on 8D is

said to be in V, 0

<

if

(2 12rr lf(eiO)IPd()) 1/p

I IfI ILP.

The expression on the left is denoted by

p < 1:
II!+ gll :S llfll + llgll in

not a norm when

< oo.

(Strictly speaking, this is

It fails then to satisfy the subadditivity property


general. However, it is common to use the norm

notation any way.)

Theorem 13.3.5 shows us that HP spaces are a natural generalization


of H00; Lemma 13.3.2 guarantees that there are many more functions in
HP than in H00, any

< oo.

We now use Blaschke products to extend the

theory further: We shall investigate boundary behavior.


first to study

H2.

Lemma 13.4.1. If p(z)


0

<

<

It is convenient

1, then

'L,f=0 ajzj

is a holomorphic polynomial and if

Proof. By direct calculation, using the fact that

{ 27r eimO d()


lo

if m-=/= 0.

Lemma 13.4.2. If f is holomorphic on D, then


1

27T
where

aj

is the

jth

{27r f reiO 2 d()


)1
(
l0o l

00

L laj 12r2j,
j=O

coefficient of the power series expansion of f about 0.

13.4. Classes of Holomorphic Functions

Proof.

407

For r fixed,
N

L a1(reiO)j
uniformly in () as

j=O

--t

J(reiO)

--t

So

oo.

d()

(see Appendix A)

(by Lemma

13.4.1 )
00

L la112r2j.

j=O

Corollary 13.4.3. With notation as in the lemma,

I: la1l2

H2

if and only if

< oo.

Proposition 13.4.4. For

H2(D),

let

fr(ei9) = f(rei9),
Then there is a unique function

0 <

2(8D)

lim llfr - JllL2


r-.1Furthermore,

llJllL2

llJllH2

g(z)

where
By Lemma

13.4.2

<

1.

such that

0.

limr_,1- llfrllL2

(: z) -

f(z)

H2.

the last line equals


00

L
j=O

(2 )
1

j
-

1
la112r .

( 13.4.4.1 )

13. Special Classes of Holomorphic Functions

408
Here the
Let

a/ s are the coefficients in the expansion


E

>

0.

Since

L laj l 2 =A<

off.

oo,

we may choose J so large that


00

"""'

j=J+I
Now choose

0<

ro <

so near

lajl 2 < 2

( -2 ) - 1 -E
r

<

that if r < r1 < r <


2

2A

for

< J. <
1-

1,

then

J.

With J, r1, r so chosen we see that Eq. (13.4.4.1) is less than


2
00
J
E
"""' E I Uj I 2
"""' 1 . I U 12 < E . A
j

E.
+
+
2A
2A
j=O
j=J+I

Thus

Ur }o<r<l

--t 1-.

is Cauchy in 2 and it follows that there is a unique

llf - frllL2

L2(8D) such that

--t

as r

The last statement of the proposition is immediate since

111112
which by Lemma

lim
r--+ i - llfrllL2

13.3.l

D
llfllH2
Iff E H2(D) and f E L2(8D) is as in Proposition 13.4.4,
=

Theorem 13.4.5.
then

f(z)

J !(( ) d(,

27l'Z !av Z

'>

for all

z ED.

Proof. Apply the Cauchy integral formula to the function


which is holomorphic on D: For any

f(rz)
Now for

fr(z)

z fixed in D we have

z ED we have
J fr((z) d(.

2m !av
-

1
1- lzl
So for

Iz I <

r<

we have

fr, 0 <

r <

1,

(13.4.5.1)

13.4. Classes of Holomorphic Functions

1 J
_

fr ( ( )
d(
27ri !av ( - z

1
27ri

__

J f(() d(
!av (-z

12_ {21r (!r(ei9)-f(ei9))


27r h

<

( where

(2-27r lor

27r

409

ei9
.
d()
ei9-z

lfr(ei(J)- f(ei9)12 de

(
2 ) 1/2
27r
,
) 2_ r I . I
1/2

27r lo

ei0 -z

d()

we have used the Cauchy-Schwarz inequality )

llfr-!112

Thus, as r --t

1-,

Theorem

13.4.5

oo.

1
--t 0
- lzl

(13.4.5.1)

Eq.

f(z)
HP, 0 < p <

as r --t

1-.

becomes

J f(() d .

(
2m
-z

is the sort of result we want to establish for every

A rather technical measure-theoretic result is required first.

<pj, c.p are measurable on 8D, <pj, c.p


a.e., and ll'PjllLP --t ll'PllLP, then ll'Pj-'PllLP --t 0.

Lemma 13.4.6. If

Proof. Let

f. >

E is less than 6,

!av (

0.

Choose fJ >

then

0 such

P( 8D)

, 'Pj --t c.p

that if the Lebesgue measure

fe 1c.p(ei9)1P d()

<

m(E)

of

f..

F <;;; [O, 27r) with m([O, 27r) - F) <


j
--t
c.p
uniformly
on
F.
Choose
J so large that if j J , then
'P
l'P1(ei9)-c.p(ei9)1 < f. for all() E F.
Finally, choose K so large that if j K, then J ll'Pjlli, - ll'Plli, I < f..
P
P
By Egorov's theorem, there is a set

fJ and

If

max ( J,

K),

fo

27r

then we have

l'Pj(ei9) - c.p(ei9)1P d()

L l'Pj(eiO)-c.p(eiO)IP d()
r l'Pj(ei(J)-c.p(ei(J)IP d()
lcF
r d(} lcFr 2Pl'Pj(ei(J)IP d(}
lF
r 2Plc.p(ei9)1P d(}
lcF
+

<

f.p

13. Special Classes of Holomorphic Functions

410

1cp 2Plcp(ei8)1Pd()
+2P 11 l'Pj(ei8)1P - lcp(ei8)1PdB i
F
+ 1 2Plcp(ei8)1PdO.
F
I'Pj fancpl'PjlP - lcplPdO I E. IJF l'PjlP - lcplPdO I
lfcp l'PjlP - lcplPdO I
ll'Pj - 'PllLP
27rEP +

<

By the choice of 6 and F, the first and last integrals are each majorized
by

2PE.

Now

large since

uniformly on F. Thus

---4

is large. Altogether,

consider those

HP

Proposition

13.4.7.

limr---.1-

HP, 0

<

<

oo,

is small when

in two stages.

D
First we

which are nonvanishing.


If 0 < p < oo, f E

then there is a function

lllllLP

is small for

0.

---4

We extend the theory to

(1)
(2)
(3)

But

<

llJllHP;
Ill - frllLP

LP(8D)

HP(D),

and

is nonvanishing,

such that

= O;

if p 1, then

f(z) =

J l(() d(,

2m lan ( z

all

z ED.

f has no zeros in D and D is simply connected, then g(z) =


l
JP 2(z) is well defined on D and g E H2. Hence there is, by Theorem 13.4.5,
a g E L2(8D) satisfying 9r ---4 g in 2. We may conclude that there is
a subsequence 9 ri converging pointwise almost everywhere to g. But then
there is a function l such that Ill = 1912/P almost everywhere and fr1 ---4 l
Proof. Since

almost everywhere.
Also

llllltP

ll'fllli2

119111"2

llfllp

llfr1 llLP ---4 llJllHP = lllllLP So Lemma 13.4.6 applies and we


may conclude that llfri - lllLP ---4 0 as j ---4 oo. The uniqueness of l follows
from the uniqueness of g and the construction. Equation ( * ) follows for
D
p 1 just as it did in Theorem 13.4.5 for p = 2.
In particular,

Notice that for

LP (8D), p

<

1, the integral in equation

(* )

does

not make any a priori sense from the point of view of Lebesgue integration
theory:

A "function" in

LP,

0 < p < 1, need not be locally integrable.

There are more advanced techniques, using the theory of distributions, for
understanding this integral. We shall not treat them here.

411

13.4. Classes of Holomorphic Functions

We now present one of the most striking applications of Blaschke prod


ucts:

13.4.7

It will yield that Proposition

holds even when

HP

has

zeros.

Lemma 13.4.8. Let 0 <p <oo. If f


and

fi, h

HP ,

1) + F =Ji+ f2.

then

f =Ji+ h

where

fj

HP

are nonvanishing.

Proof. Write

f = F B = F (B

Theorem 13.4.9. Let 0 <p <oo and


of Proposition 13.4. 7 hold for

Proof. Use Lemma

to Ji and to

13.4.8

f.

to write

f E HP(D).

f = Ji + f2.

Then all the conclusions

Apply Proposition

13.4.7

separately. The rest is obvious.

It should be noted that the case

13.4.7 and Theorem 13.4.9.

p = oo

is excluded from Proposition

f E H00,
fr are bounded and continuous on 8D. Moreover, llfr - jllH'"' ---+ 0
mean that fr ---+ j uniformly. It would follow that j is continuous,
The results are false in this case: W hen

then the
would

which is clearly false in general. There exist more delicate methods which
produce, for

j---+ f

H00(D),

an

L00(8D)

such that

lljllv"' = llf II Hoo


]

pointwise a.e. Good references are HOF or GAR .

An immediate corollary of Proposition

13.4.7

and Theorem

and

13.4.9

is

the following:

Corollary 13.4.10. Let 0 <p <oo. If f


Proof. Write

FP

H1

F B

as usual. Then

we have

We see that

FP(z) =
2rri
=

O;

hence

HP

and

P =0

P
J
F (() d( = 0
laD ( - Z

0 a.e., then

f = 0.

almost everywhere. Since

for all z.

f = 0.

The final result of this section is a remarkable strengthening of this


corollary.

Theorem 13.4.11. If f
positive measure, then

HP,p

f = 0.

>

Proof. By dividing out a power of


Jensen's inequality,
-oo

<log

lf(O)I

0, and

j vanishes

on a set E

z,

we may suppose that

1
271'

1211'

log

lf(rei6)1 d()

f(O) f=

8D

of

0. By

412

Exercises

Figure 13.7

for any

8D(O, )

such that f has no zeros on

As

->

1- through such

values, the right-hand side of this expression tends to

2__ f27r log jj(ei11)1 d(}


lo

21f
a clear contradiction.

-oo,

Exercises

1. Prove that if f

HP for
sup

O<r<l

some p > 0, then

f27r max{log IJI, O} d(}


lo

< oo.

2. Prove that the converse to Exercise 1 is not true.


3. Consider the domain U C whose boundary is

comb, as in Figure

13. 7. The domain is simply connected, and the Riemann mapping the
orem guarantees that there is a conformal map of

onto U.

Prove

that this map cannot be extended continuously and one-to-one to the


boundary.

4. There is a conformal mapping from the unit disc to the first quadrant.
Calculate the mapping explicitly.

How smooth is this mapping at the

boundary point P that is mapped to the origin?

5. Let f be holomorphic on the unit disc


is a harmonic function u on

D such

D.

that

Let p > 0. Suppose that there

IJIP::;

u.

Prove that f E

HP.

Exercises

413

6. Does the result of Exercise 5 suggest a method for defining

HP

functions

on an arbitrary open set U <;;;; C?

7. Suppose that
to

f is an 1

function on

Does it follow that F E

D.

8D and that Fis its Cauchy integral


H 1 ? What condition can you put on f

that will make your answer affirmative?

8. Assume that
J

HP(D), 0 < p
g, h E H2P(D).

h, where

< oo.

Prove that it is possible to write

9. Let S be the family of all schlicht functions. Fix an integer k


that there is a number Ck > 0 such that if

coefficient of its Taylor expansion about 0, then

> 1. Prove
S and ak is the kth

lakl Ck.

10. Is the number of schlicht functions countable or uncountable?


11. Is there a largest Hardy space? Is there a smallest one?
12. Fix 0
13. If

<

< oo.

HP

Under what arithmetic operations is

and g E

H00 ,

then prove that

f g

HP(D)
HP.

closed?

What other

statements can you make about products of Hardy space functions?

14. Let 0

<

< oo.

Give an example of an

Give an example of an
*

HP

function that is schlicht.

HP

function that is not schlicht.

15. Let U be a bounded domain in C with real analytic boundary (the


boundary is locally the graph of a real function that is given by a conver
gent power series). Assume that U is simply connected. Let

f: D

___,

be a conformal mapping, as provided by the Riemann mapping theorem.


Prove that

[Hint:
f

that

has an analytic continuation to a neighborhood W of

D.

First, the problem is local and has nothing to do with the fact
is one-to-one and onto. What is important is that the map take

the boundary to the boundary. By a change of coordinates, reduce to


the case when both boundaries are flat. Then apply Schwarz reflection. ]

16. Give an example of a one-to-one, onto, C00 mapping of the disc

to

itself, with a C00 inverse, such that the mapping does not extend even
continuously to any boundary point.

[Hint:

The map should oscillate

near the boundary. ]

17. Give an example of a conformal mapping of the disc

to a simply

connected, unbounded domain with smooth boundary such that the


mapping is
domain

not smooth to the boundary. [Hint:


the upper half plane. ]

Take as your unbounded

18. Do the calculation to complete the proof of Lemma 13.3.2, that is, the
calculation to show

fp rf. HP2

there.

Chapter 14

Hilbert Spaces of
Holomorphic
Functions, the
Bergman Kernel, and
Biholomorphic
Mappings

14.1. The Geometry of Hilbert Space


One of the most far-reaching developments in twentieth-century mathemat
ics has been the sy stematic exploitation of the idea of infinite-dimensional
vector spaces-especially vector spaces of functions.

Anything like a com

plete exploration of these ideas would take us far outside the scope of the
present book.

But it turns out that the part of this general theory that

applies specifically to vector spaces of holomorphic functions can be pre


sented quite quickly.

Of course, the full resonance of these ideas would be

apparent only in the context of a presentation of the entire theory. But the
holomorphic case, which is fairly simple, can also be used as an introduc
tion to and basic instance of the general theory. The reader familiar with
Hilbert space theory will find here some new applications. The reader un
familiar with Hilbert space theory will at least see the theory in action in
some interesting and relatively accessible instances.

415

14. Hilbert Spaces of Holomorphic Functions

416

Stefan Bergman (1895-1977) was one of the most remarkable and orig
inal complex analysts of the twentieth century. He conceived of the idea of
considering Hilbert spaces of holomorphic functions on a region in complex
space ( either in the plane or in higher dimensions ) . His investigations led
to the definition of the Bergman kernel, which is a reproducing kernel for
holomorphic functions. In turn the Bergman kernel leads to the definition
of a holomorphically invariant metric called the Bergman metric. This met
ric is a generalization of the Poincare metric on the disc ( see [KRA3] for
further details ) and is also one of the first examples of what has come to be
known as a Kahler metric. We shall explore some of these ideas, in a simple
context, in the present chapter.
S. Bell and E. Ligocka developed the idea, in the early 1980's, of using
the Bergman kernel to study the boundary behavior of biholomorphic map
pings ( see [ BEK] ) . In the function theory of one complex variable, these are
just the familiar conformal mappings. We shall describe the Bell-Ligocka
approach to the subject in this chapter.
The basic idea here is to try to establish, in certain infinite-dimensional
vector spaces, a kind of geometry that is analogous to standard Euclidean
geometry. For this, one needs an appropriate idea of inner product.
Definition

14.1.1. If V is

vector space over e, then a function

(-,):Vxv_,e
is called

positive-definite (nondegenerate) Hermitian inner product if

(1) (v1 +v2,w) = (v1,w) + (v2,w), all v1,v2,w EV;


(2) (o:v, w ) = o:(v,w), all o: Ee, v, w, EV;
(3) (v,w)

(4) (v,v) 0,
(5) (v, v)

all v,w EV;


all v EV;

(w, v),

0 if and only if v

0.

Here are some simple examples of complex vector spaces equipped with
nondegenerate Hermitian inner products:

14.1.2. Let V en with its usual vector space structure. If


(111, ... , Vn), W = (w1, . .. , Wn ) are elements of en, then let

EXAMPLE

Tl

(v, w)

L Vj'Wj
j=l

The verification of properties (1)-(5) in Definition 14.1.1 is straightforward.


EXAMPLE 14.1.3. Let V be the set of all continuous, complex-valued func
tions on [O, 1], with addition and scalar multiplication defined in the usual

14.1. The

way. If

Geometry

f, g EV,

of Hilbert

417

Space

then let

(J,g) =
The verification of properties

1
f(x)g(x)dx.

fu

(1)-(5)

14.1.1

in Definition

follows the same

algebraic pattern as in the previous example, except that the implication

(J, J) = 0

only if

requires a brief extra argument using continuity

(exercise).
EXAMPLE

ter

13.J

14.1.4.

= H2(D).

Let V

[The space

H2(D) was

defined in Chap

f, g E H2(D),

Let addition and scalar multiplication be as usual. If

associate to them the boundary functions

13.4.9).

f, g

E L2(8D)

(see Theorem

Define

(J, g)= 2_
{2Tr f(eiO)g(eiO) d(}.
27r lo
The verification of properties (1)-(5) of Definition 14.1.1 is immediate, ex
cept that again the fact that (J, J) = 0 implies f
0 requires special
consideration. One notes that J lf21 =0 implies f= 0 a.e.; hence f
0.
=

Here are a few elementary properties of a complex vector space equipped


with a positive-definite Hermitian inner product:

Proposition 14.1.5.
(1) ( V, W

If Vis

W )= ( V, W
2

(2) ( v, aw) =a(v, w ) ,

as

1)

all

above, then

( V, W2), all V, W
n EC, v, w EV;

(3) (O,v) = (v,O) =0, all v EV;


(4) ( v + w, v + w) = (v,v) + ( w, w)
Proof.

14.1.1.
14.1.1.
If

v, w

Property
Property
As for

(3)

is immediate from parts

and

(4):

is immediate from parts


If
=

v EV,

, W EV;
2

(v, w) ,

(1)
(2)

(v, 0) = (0, v)
EV,

+ 2Re

(1)
(2)

all

and
and

v, w EV.

(3)
(3)

of Definition
of Definition

then

(v - v,v) = (v,v) - (v,v) =0.

then

(v + w, v + w) = (v,v) +
= (v, v) +
= (v,v) +

(v,w) + ( w, v) + ( w, w)
(v,w) + (v, w) + ( w, w )
( w, w) + 2 Re (v,w) .

If V is a complex vector space and


inner product on

V,

( , )

a nondegenerate Hermitian

then we define the associated norm by

llvll=

14. Hilbert Spaces of Holomorphic Functions

418

(recall, for motivation, the relation between norm and inner product on

JR.").

Fundamental to the utility of this norm is that it satisfies the Cauchy

Schwarz inequality:

Proposition 14.1.6. If v, w EV, then

l(v, w )I :=:; llvll llwll


v -/- 0, w -/- O; otherwise there is nothing
=
v
l
2,/1
2
llw l ,"( l(v, w )I . Then o:-/- 0,/1-/- 0. Choose
ll ll
eiO
(
w,
v)
"/. For r E JR.,
that

Proof. We may as well suppose


to prove. Let a=
() E

[O, 271')

such

0 < (v - r e i O .w , v

r e i O ,w )

(v,v) +r2(w,w)-2Re (rPi9(w,v))


o:+ /1r2 - 2q.
Let

r = "f / fJ

to obtain

"12 212
O<
- o:+--/1
fJ
or

as desired.

Corollary 14.1.7 (Triangle inequality).

Ifv,w EV, then

111 +wll llvll + llwll


Proof. We have

llv +wll2

(v + w, v + w )
( v, v ) + ( 'w, w) +(v, w ) + ( w, v)
< llvll2 + llwll2 + llvll llwll + llwll llvll
(llvll + llwll)2.
D

Taking the square root of both sides gives the result.


The triangle inequality tells us that we can use

II II

to define a reason

able notion of distance on V that will make Va metric space: If


then define the distance between

and w to be

llv - wll

v,w

E V,

We now use the

notion of distance in considering properties of sequences.

), II 11 are as above and {vJ}1 Vis a sequence in V, then


{ Vj} satisfies the Cauchy condition if for any f > 0 there is a
J > 0 such that j, k J implies llvj - vkll < E. We say that V is complete
if every Cauchy sequence { Vj} has a l:imit v E V : T hat is, for every E > 0
there is a J > 0 such that j J implies llvj -vii < E.
If V, ( ,

we say that

14.1.

419

of Hilbert Space

The Geometry

For the purposes of doing analysis, it is best to work in a complete


space. The fundamental process in analysis is taking limits, and one wants
to know that every Cauchy sequence has a limit.

F inite-dimensional vec

tor spaces are complete no matter what inner product norm is assigned
to them

( see

Exercise

Infinite-dimensional spaces with inner-product

17).

norms need not be complete, and completeness must be mandated and / or


proved in order to obtain spaces on which mathematical analysis can be
successfully studied. These general remarks have been provided to motivate
one to consider the mathematical structures which we now define.

Definition 14.1.8. A complex vector space V with positive-definite Her

( , ),

mitian inner product

II II ,

is called a

which is complete under the associated norm

Hilbert space.

EXAMPLE 14.1.9. As in Example

let

14.1.2,

j =l

j=l

This norm provides the usual notion of distance ( or metric )

llv - wll

L lv1 - w112
j=l

We know that ten, equipped with this metric, is complete.


EXAMPLE 14.1.10. Let

A2(0)
Define, for

f, g

{!

C be a domain. Let

holomorphic on n:

A2(0),
(f,g)

The fact that

lo lf(z)l2 dxdy

< oo }.

lo f(z)g(z)dxdy.

lf(z)g(z)I = lf(z)l lg(z)I

:S

lf(z)l2 + lg(z)l2 implies that this


A2(0) is

integral ( over the open set n) converges absolutely. We claim that


a Hilbert space when equipped with this inner product.

0,

To see this, let K n be compact. Select


such that

D(w, r)

whenever

wE

we have that

lh(w)I

rrr j

D(w,r)

::::: f
rrr

JD(w,r)

h(z) dxdy

lh(z)I. 1 dxdy

>

K
. For all

0, depending on Kand

wE

Kand

A2(0),

420

14. Hilbert Spaces of Holomorphic Functions

;,

:S
rrr

(
, (J,

:S

D(w,r)

V rr r

:S

r;;;

v rr r

lh(z)l2dxdy

D(w,r)

) (
)
112

112
f
12dxdy
jD(w,r)

112

lh(z)l2dxdy

(14.1.10.1)

llhll

A2(0) is a Cauchy sequence, then let E > 0 and choose J


j, k 2'. J, then llf1 - hll < E y'rrr. Then, by Eq. (14.1.10.1)
h= f1 - f1.:, we have

Now if U1}

so large that when


applied to

sup

wEK

f.

lf1(w) - fk(w)I

:S

r;;;

y7rr

11!1 - hll <

E.

Thus U1} converges uniformly on compact sets of n to a limit function

Of course

Then

f is holomorphic.

To see that

A2(0),

fix a c