Robert E. Greene
Steven G. Krantz
Graduate Studies
in Mathematics
Volume 40
Function Theory of
One Complex Variable
THIRD EDITION
Robert E. Greene
Steven G. Krantz
Graduate Studies
in Mathematics
Volume 40
+J:1t,"
"'
.en"'=
'""NDEI>'
Editorial Board
Steven G. Krantz
David Saltman (Chair)
David Sattinger
Ronald Stern
2000
.ams.org/bookpages/gsm40
www
( Graduate
( alk.
ISBN 0821839624
paper )
I. Krantz, Steven G.
( Steven
George ) , 1951
II. Title.
III. Series.
Q A331.7.G75
2006
2005057188
acting for them, are permitted to make fair use of the material, such as to copy a chapter for use
in teaching or research.
The paper used in this book is acidfree 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
Contents
Fundamental Concepts
xiii
xv
xvii
xix
1
1.1.
1.2.
1.3.
Complex Polynomials
1.4.
1.5.
Exercises
Chapter 2.
10
14
17
20
29
2.1.
29
2.2.
34
2.3.
Antiderivatives Revisited
40
2.4.
43

Vll
viii
Contents
2.5.
2.6.
Exercises
Chapter 3.
50
53
60
69
3.l.
69
3.2.
74
3.3.
81
3.4.
84
3.5.
88
3.6.
90
Exercises
Chapter 4.
4.l.
94
Meromorphic Functions and Residues
105
105
4.2.
109
4.3.
113
4.4.
119
4.5.
122
4.6.
4.7.
Exercises
Chapter 5.
128
137
145
157
5.l.
157
5.2.
162
5.3.
166
5.4.
169
5.5.
171
Exercises
174
Contents
ix
Chapter 6.
179
6.1.
180
6.2.
182
6.3.
184
6.4.
189
6.5.
Normal Families
192
6.6.
196
6.7.
Exercises
Chapter 7.
198
202
Harmonic Functions
207
7.1.
208
7.2.
210
7.3.
212
7.4.
218
7.5.
220
7.6.
Harnack's Principle
224
7.7.
226
7.8.
7.9.
Exercises
Chapter 8.
236
240
243
255
8.1.
255
8.2.
263
8.3.
Exercises
Chapter 9.
266
274
279
Contents
9.l.
279
9.2.
285
9.3.
288
296
Exercises
Chapter 10.
Analytic Continuation
299
10.l.
299
10.2.
304
10.3.
307
10.4.
310
10.5.
314
10.6.
Elliptic Functions
323
330
Exercises
Chapter 11.
Topology
11.l.
11.2.
11.3.
11.5.
335
338
11.4.
335
343
344
349
352
Exercises
Chapter 12.
363
12.l.
Runge's Theorem
363
12.2.
Mergelyan's Theorem
369
12.3.
378
381
Exercises
Chapter 13.
13.l.
385
386
xi
Contents
13.2.
392
13.3.
Hardy Spaces
401
13.4.
Exercises
Chapter 14.
406
412
415
14.1.
415
14.2.
426
14.3.
431
14.4.
Bell's Condition R
438
14.5.
443
Exercises
Chapter 15.
446
Special Functions
449
15.1.
449
15.2.
457
Exercises
Chapter 16.
467
471
16.0.
Introduction
471
16.1.
473
16.2.
478
16.3.
483
Exercises
485
487
493
References
497
Index
501
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.
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 scribbledupon 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.
Xlll
tions.
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
xv
T his book is a text for a firstyear 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
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
One of the
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.
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
R Then we set
consist of
IR2
IR2 = {(x,y) : x
IR , y
IR}.
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.
(1, 0)
a E IR, then
by
and
(0, 1)
i i = 1.
(* ) .
iv , ( = e + i'f/.
= x + iy
1.
x =Re z.
y =Imz.
written
Likewise
Fundamental Concepts
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
Jx2 + y2 .
We also
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
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
to define
1
zw
z
=z=w
w lwl2
for
of
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,
IR
1+
x + iO EC.
p(x)
For instance,
x2 + 1
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
( such
as division by a
(1)
If
is real, then
as in calculus.
(2)
If
ez
(3)
If
x+ iy,
(2)
and
(3)
then
ez
Parts
ex+iy =ex
( cosy+ i siny ) .
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.
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.
)
)
( * ) ,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),
=
definition
+ iy,
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,
<
2rr,
such that
ei() (see
ray o.
Now if
where
/ l zl
has modulus
l zl
1.
(i;1)
l zl C
where r
number
If
l zl .
z.
k is an integer,
ei()
then
cos e + i sin e
cos(O +
ei(o+2k1r) .
So a nonzero complex number has infinitely many different representations
>
0, e ER
(3)
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
Euclideanaxiomaticgeometry setting.
As a consequence, if
1, 2, ..., then
(ez )
enz.
times
1. Fundamental Concepts
eie
 
cos 9 + i sin 9
/
I
I
I
I
I
\
'
'
'
.... 
'
'
\
I
I
I
I
/
/
'
_,,,
Figure 1.1
( 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
=
r6ei69
Hence
r
2116
()
2. e21ri.
27r/6
7r/3.
21 i6 coS11"
/3 + isin "/3)
r6ei69
2 . e41ri
r6ei69
r6ei69
r6ei69
2 . e61ri
7r
2 . e8 i
'
'
2. el07ri
(
+ iv'3
2
'
2 1 !
0 )
+i
21 6
/ '
21/6
(! iv'3)
0i)
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)
=
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 
r3ei38
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.
'
7r/4. But
Thus,
Proposition
Proof.
(rather than
IR).
z, w E C, then
lz + wl :'.S lzl + lwl.
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.
L Zj :'.S L \zil
j=l
j=l
Another useful variant of the triangle inequality is
w
L.J J J
j=l
Proof.
j=l
j=l
If z1,
... , Zn
Since
[Notice
j=l
j=l
j=l
A LJ=lZjWj
= Lnj=l IW312
LJ=l lwj12 =/= O;
A
is nothing to prove.
The choice of
otherwise there
**
as
A, ( )
2 21LJ=lZjWjl2
L
J=lZjWj
l
I
O::; lzjl2 +
LJ=l lwjl2 LJ=l lwjl2
becomes
**
or
j=l
j=l
as desired.
EXAMPLE 1.2.5. If
see this, fix
2::}:1 lzjl2
converges, then
2::}:1 lzjl/j
converges.
To
t lz!I
j=l
( use
n,
by hypothesis. The
to converge as
oo; hence those partial sums are also bounded above, independent of
Thus
tlz!I
j=l J
n.
1. Fundamental Concepts
10
C,
to
x+
C.
to
x + iy and expanding.
For
instance
i +(3 +i)z +5z
2
2
i +3xy +ix +3iy +5x + lOixy  5y
2
2
i +(3 +i)x +(3i  l)y +5x +(10i)xy  5y .
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
C ( see
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
=

=
zz
(;) ()
z
11
z2 zz z2 z2 zz z2
4+2+44+24
z z.
You can check for yourself that there is no polynomial expression in
without any z's, that equals
x2 + y2:
z,
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+42+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,
that depend on
on
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
f: U
if and only if
of/ oy =
t
is
if and only if
C1
and
oy
0,
12
!__ 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
iy,
0
oz
z =1 ,
oz
0
azz =o,
0
oz
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
( )
*
oF
oG
,
b
(aF + bG) =a
oz
oz + oz
oF
oG
b
bG) =a
(aF
+
+
oz
oz
az
o
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
( )( )
oe
om
ozl
ozm
j(j
1)
( )
*
that
.k
(z1z )
(j
 f_ +
1)k(k  1)
(k  m + 1). zjezkm
1.3.
if
13
Complex Polynomials
if either f, >
or
Also
()
()
(z1zk)
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.
that
evaluated at 0 is
f!m!aem
We think of a polynomial
in a sense, "independent of
when
m > 0,
z";
I: aez
as being,
8p/oz
aem
= 0.
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
C1
14
1. Fundamental Concepts
f which
satisfy
(8/az)f
f:
is said to be holomorphic if
of=
Oz
at every point of U.
f is
u +iv,
where
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 realvalued functions u and v, we have that u and v
au
av
ax
8y
au
and
8y
av
ax
at every point of U.
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 CauchyRiemann equations. The proof of the following easy
result is left as an exercise for you:
Proposition 1.4.3. If
f : U
..
C is
15
of
of
. of
i
=
=
8y
{)z  ox 
on U.
The CauchyRiemann equations suggest a further line of investigation which
is of considerable importance. Namely, suppose that
u v
and
are
C2
func
a (au) = a (av)
ox
and
ox
ox
{)y
a (au) = a (av) .
8y
{)y
 {)y
ox
a2v/axay =a2v/oyox),
we obtain
82u 82u
8x2 + {)y2 =0.
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
+
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
u? It is possible to
when u is a polynomial.
answer is "yes"
0,
then
u(x,y)
u(x, y)
in terms of
( z+z zz
)
,
2
P(z,z)
and
z:
em
'"""
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
Thus
aem
aoe
P,
we have
 '""" e '"""  m
aoo + L.,, aeoz + L.,, aomz .
m2'.l
e::::1
(see Proposition
1.3.3).
In conclusion,
u(z)
and
P(z, z)
(aoo + 2
Re
(Q(z)),
L aeoi)
e::::1
1.5.
17
F.
say that
av
ax
au
ay'
av
ay
In short, once
u is given,
the existence of
v ( and
then
au
ax.
up to an additive constant.
hence of
F)
au/ax, respectively ) ,
avjay= g?
and
determined.
Thus determining
Theorem 1.5.1. If J, g a re
R= {(x,y)
( see
Exercise
52).
and if
af
ag
on R,
ay
ax
C2(R) such that
and
(x,y)
(1.5.1.1)
we
may take h to
E R, set
h(x,y) =
1x f(t,b)dt+1Y g(x,s)ds.
ah
(x,y) = g(x,y).
ay
(1.5.1.2)
1.
18
oh/ox,
Fundamental Concepts
rx
f(t, b) dt = f(x, b).
ox la
0
Moreover, since
g is C 1 ,
(1.5.1.3)
[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)
=
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 realvalued if
f
D
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 horizontalvertical 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
realvalued harmonic function on R, then there is a holomorphic function F
on R such that Re F u.
Corollary 1.5.2. If
satisfy
f= 
of
oy
og
ox
OU
g=
ox
on
R.
19
6u = 0.]
1
C , the
C2 function v on n
Since f, g E
av
ax
av
ay
But this says that
F(z)
!=
au
g= ax
u(z) + iv(z)
tions on R, as desired.
Given our
such that
Theorem 1.5.3. If U
on U such
Proof. Write
=u
and
g = v
function
h1 such
a f ag
ay =ax
_
C2
that
ah1
=
ax
=u '
ah1
v
ay = g= .
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
CauchyRiemann 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
Exercises
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
(
v; )
(..;'2 )8

'
v'2
(g)
2+2i
21
Exercises
z
(d) z2+ 1
z2
( e ) z1
(f) z4 +2z+ 6
3 i
(b) (6+2i)3
(c ) ( v'3+ i) . ( v'3i)
+2
i2
(d)
(e)
1)
(i+
(i+2)
(i+ 3)
and
w,
z such that z2
such that
w4
6. Prove that
l
zw 2
1 zw
p(z)
j ::=:; n.
7. Let
0 ::=:;
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
(i)
(ii)
if a, b E P, then a+ b E P and a
P F
b E P;
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.
i E P or i E P.)
E IR :
x >
<P(z)
maps the set D
C: Im z
>
O}.
{z
EC:
lzl
<
lz
1+ z
1} onetoone
{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 onetoone and onto. Conversely, prove that if
u(z)
with
a, b, c , dEC and u:
az+b
cz+d
a, b, c, dare
ad be>
11. Let S
0.
{z EC: 1/2
<
lzl
1,
<
2}.
Let
(z)
z+ (1/z).
roots of
i,
Compute (S).
all sixth 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
+iy :
3ei7r
4ei7r/4
4ei277r/4
6eil07r/3
ei7r/l2
(f) ei37r/8
(g) 7e i7r/6
(h) 4ei207r/6
( a)
(b)
(c)
(d)
(e)
'""" z w
J J
j=l
j=l
j=l
23
Exercises
{Zj}, { Wj}
for equality to
[Hint:
z,w
19. Fix
satisfying
n
LZjWj
j= I
z1,... , Zn
L::j= 1 lwj 12
1. Prove that
ei9
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.
z2+z+ 1= 0.
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+
( 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
=
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)
a
(4z2  z3)
( a) az
a
(b) (z2+z2z3)
Oz
83
(c)  (3z2z4  2z3z+z 4  z5)
8x28y
as
6z2)
2 z
{d) az3az2 (z  z+4z 
( 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.
8f
az
and
8f
=
Oz
for all
31. Let F
z
:
C Prove that
C + C be
constant.
a2
F =0
8z2
for all
C Prove that
F(z, z)
for some polynomials
z G(z)+ H(z)
G, Hinz ( i.e.,
C+ C is
F2
= 0.
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,
L= (:x i :y ) = :z
and
M= +i =.
oy
oz
2 ox
and
g2
C,
34. If
f is a
gz
a
oz1
O
ozf.
z.
p(z)= x
is not a polynomial
36. Write
and
oz
q( z)=x2.
Oz
in polar coordinates.
(+, ,
...;
) is
h closed?
38. Answer Exercise 37 with h replaced by 1t, the set of all holomorphic
functions.
39. Let F be a
cally constant.
{ z I z  zo I
:
be an open set.
::;
r}
Let
zo
U and
>
21f
then
Exercises
26
42. If
43.
and
1 1
is holomorphic on U C and
6(lflP)
45. Prove that if
P2lflP2
I1
is nonvanishing, then
any p > 0.
is non
vanishing, then
any p 2: 1.
be a
C2,
Prove that
h(z)
46. Let
is
ou/oy +i8u / 8x
C2,
is holomorphic on U.
harmonic.
49. Let
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.]
G1, G2
8G1
oz Prove that
51. Let
(v1, v2)
that
G1  G2 =constant.
8G2
 oz .
8v1
8v2
ax
oy
Prove that
(v1, v2)
harmonic function
and
8v1
8v2
ax + oy
0.
(oh/ox, oh/oy))
of a
27
Exercises
J(z)
1/z
is holomorphic on U
{z EC: 1
<
izl
2}.
[Hint: If
<
there were an antiderivative, then its imaginary part would differ from
arg z by a constant.
{ eit
<
27r}.
Let
1
[Hint:
2rr
( eit )
eit
dt
0.
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
U. Show by example
56. Let U1 U2 U3
Let
j,
flu
has
f has a holomorphic
n, such that if
{ Zj } are
Llzj11,
j=l
{ ZJi, ... , Z j }
k
{ z1, ... , Z }
n
such that
L Zjm C.
m=l
[Suggestion:
of the nonzero Zj. At least one third of these arguments lie within
of each other.
27r/3
28
Exercises
58. Let
is an integer for every j, k (the integer may depend on j and k), then
{Zj}
if Re z, Im z are integers.
?
Approximately how many lattice points are contained in {z : lzl < R}
is called a
lattice point
asymptotically as
1.
im
[Hint:
+
{z : lzl
<
F(R)
z
R}
= 1 ?
is contained in
of lattice points
{z : I z I
F(R)
If a lattice point
centered at
area of
<
Chapter 2
vertical
vertical derivative is
{)I ay
horizontal
{)I ax and the
and
object ;y
[a, b]
{'Y (t)
: t
[a., b]},
1( t)
simple closed if
or
1( t)
closed if 1 ( a ) = 1(b). It
and 1(a) 1( b). Intuitively, a
1l[a,b)
is onetoone
a., t
b.
closed interval,
this requires a
new definition.
29
30
</J : [a,b]
JR is
</JE C1([a,b]), if
tiable
(or
C1), and
we write
+
called
continuously differen
[a,b].
to
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
bE
a+E
<P'(t) dt
1 <P'(t) dt.
y1
[a,b], and
and
'Y2
'Y : [a,b]
</JE C1([a,b]).
+
we write
'YE C1([a,b]),
if
'Y1,'Y2
we will write
We also write
y'(t)
for
1/;1(t) + i 1/;2(t).
dy
dt
dy / dt.
1/; : [a,b]
+
dy1
dt
[a,b].
. dy2
.
i dt
C be continuous on
[a,b].
Write
1/;(t)
Then we define
b
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 complexvalued, then
1 y'(t) dt.
2.1.
31
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)
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
realvalued functions. But, by dealing with Ref and Imf separately, we can
extend the proposition to complexvalued f without change. Our main goal
in this section is to derive an analogue of (t) in which complex numbers,
particularly complexvalued functions, play a more vital role.
For motivational purposes, let us consider what happens to (t), thus
extended to complexvalued 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
8v
{)y
{)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
( ** )
32
since
Thus
f is
( ** )
dy
of
b(t) ) .
(t)
ox
dt
dy
of
b(t) ) .
(t)
oz
dt
holomorphic. [ Note here
1.4.3.]
becomes
r b of
f ( 'y ( b ) )  f ( 'y ( a ) ) = la
dy
( t ) dt,
dt
dyifdt + idy2fdt.
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
dy/dt
oz
( 'y ( t ) ) .
to be by definition
y : [a, b]
+
: U
1 F ( z ) d z= f
lr
la
The multiplicative
+
C is continuous on U, and
F ( 'y ( t ) )
dy
dt.
dt
numbers.
The payoff for this definition is the following elegant formula:
y : [a, b]
+
U be a C1
f('y( b ) )  f ( 'y ( a ) ) =
J af
lr
( z ) dz.
oz
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.
holomorphic functions.
[ Check
2.1.6
example. ]
F(z) =
z, for
We conclude this section with some easy but useful facts about integrals.
[a, b]
+
C is continuous, then
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
Hence
But the far right side is real since the far left side is real. Hence
C1
curve, then
Uc:;:;
C be open and
C0(U).
where
f("t) =
If "I:
lb 1; (t) I dt.
(dyifdt,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
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:;:;
U+
Ca continuous
is
be a
C1
C1
function with a
C1
inverse. Let
;y = "Io.
Then
Proof. Exercise. Use the standard change of variable formula from calculus.
D
34
Proposition
function
2.1.9
f along
{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!
PE
U, and
g:
g(z)
f,
\ {P}
 Ca function. We say
that
lim
Z>P
fE
C,
lg(z)  fl
zE
U and 0
<
lz  Pl
<
'5,
< E.
In a similar fashion, if
and
PE
U , then we say
Now let
f(z)  f(zo)
z  zo
for
zo =/= zE
U. In case
lim
Z>ZQ
complex derivative by
then certainly
f is
Z  ZQ
zo,
f(z)  J(zo)
35
Figure 2.1
f' exists
J'(z)
for all
zE
Proof. Fix
8f
az
zo E
U. If
z is near to zo
y(t)
(1
 t) zo + tz
f(z)  f(zo)
=
Therefore
f(z)  f(zo)
z  zo
f('Y(l))  f('Y(O))
J 8f dz
1r az
f 1 8f ( (t)) dy dt
lo az 'Y . dt
f l 8f ( (t)) . (z  zo) dt.
lo az 'Y
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].
af
l (
az
whenever l w
when
 zol
lz zol
<
<
a1
) (zo)
az
8, then, in particular,
satisfies
< E
< E
<
<
It follows that
1.
E.
f(z)  f(zo)
Z zo
1m 
z+zo
(8f /8z)(zo).
af I az
when
( and
Theorem 2.2.1.
Theorem 2.2.2. If f
of U, then
C1 (U)
and
valued function
continuous on U, then
is holomorphic on U .
letting z approach
z0
1.
Im
z+zo
f(z) f(zo)
Z  Zo
37
f(zo+h)  f(zo)
lim
h+O , h
real
1.
Im
h+O

1.

Im

h+O
1.
+i
+z Im
h+O
(xo,yo)

v
X
(xo,Yo)
1.
Im
z+zo
f(z)  f(zo)
Z  zo
f(zo+ih)  f(zo)
ih
lim
h+O , h real

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

(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)
au
a
Y (xo,Yo)
=

av
a
X
gives
(xo,yo)
J'
!'
au
av
+i
ax
ax
av
au
i
ay
ay
f'
U without the
fE
I
C (U)
u and v.
Thus if
and is holomorphic
f E C1 (U).
38
and
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
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
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
P.
Holo
a neighborhood of
P E C.
Let
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;
DwJ(P) to Dw2f(P).
2.2.
39
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,
( around
Exercise
12
2.2.3, which in
(2.2.3.2) holds at P,
ci.
It is worthwhile to consider Theorem
2.2.3
Also we consider
( )
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()
a
v'a2+b2'
b
40
2+ b 2 1.
(Ja2+2b ) (Ja2+2b ) =
a
( ). (:; ::)
for some >.
IR, >.
>
0, and some 0
IR.
cos e
sinO
sin e
cosO
( )
IR.2
( ). (:; ::)
IR.2
we have
( )2
2+
=
>. (:) () =IJ'(z)I,
in agreement with Theorem 2.2.3.
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
41
()(
Figure 2.2
when
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
(n, /1)
that contains
[a, b]
of
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
PE U.
Let
f and g be continuous,
realvalued functions on
af
ay
at
every point
of
U (including
h: U
'
the point
ah
ay
=
P).
U and
(ao, bo)
define
h(x, y)
which
ag
on U \ {P}.
ax
ah
ax
differentiability
[Y g (x, s) ds.
}bo
ao + iboE
42
,
__________________
F igure
2.3
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
(x, y)
ah/ax.
(x, bo)
to
(x, y)
does
2.3).
1.5.1)
[Y !
! fY
(x, s)ds =
g(x, s)ds
uX }bo g
}bo uX
[Y f(x, s)ds
}bo us
a
f(t, bo) dt = f(x, bo)
uX ao
for
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
is continuous, and
d
H(x) = F(x),
dx
H(x) = h(x, y)
is
except possibly at
43
d
H(x) = F(x)
dx
for all
x;
hence
ah
=
ax
f
D
as desired.
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(()
Such an
f(()  f(z)
( z
f'(z)
for
(EU\ {z}
for
z.
U.
finitely many
"singular points".
on
44
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,
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)
[O, 1]
such that
of the disc
>
lz Pl =r.
D(P,r)
can be parametrized as a
C by setting
1(t)= p + re2nit.
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
er : [O, 1]
>
defined by
er(t)=P + re2nit
D(P,r) with clockwise orientation. Recall
aD(P,r) do not depend
as
er
aD(P,r) as
aD(zo,r).
Then
27ri
Proof. To evaluate
"Y
1

(z
d(=l.
i ( z d(,
l
I(z)= j __ d(,
fr (z
defined for all
with
lz  zol < r.
(i) f(z)
is independent of
(ii) f(zo)
z;
27ri.
45
f(z)
with
lz zol
r, as required.
<
(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.
1
( f+ ((  z)2
is holomorphic on C
function
Hence, by Proposition
o
f(z)
oz
since
1(0)
1( 1).
1
( f+ (z
 .
2.1.6,
1
d(
"Y((z)2
1
1( 1)z
1
1(0)z
variable, (or z, is actually being varied, but some thought will convince you
1:
and
is constant, proving
To prove
[O, 27r]
of /oz
+
(ii) is
1(t)
C,
rcost+ irsint+
zo.
Then
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
27ri.
46
an
open
r
1
0 be such that
>
(t)
f(z)
f on the
zo. (The curve / has
about
C1 function of two
 (x2 + y2W
is identically 0 on
k,
zz]k, which
az[l zz]k
k[l zz]k1(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
The occurrence of
>
0 such that
Thus
J f(z)
J
!, ( z d( !,(J(()
z d(.
=
Since
47
f(z)
J
d( = f(z) J _I_ d(,
fr(z
fr(z
i ( z d(=
(*)
2tri.
gives the desired conclusion.
as
( implicitly )
1(a)
1(b),
[a, b]
i f(z) dz=
on U. Since
1(a) = 1(b),
0
U is a C1
0.
G' = f
+
then
+
C with
we have that
= G('Y(b) ) G('Y(a) ) .
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
D(a, r ) ,
8D(a, r ) ,
[For a region
2.6.]
The various possibilities for choosing such a parametrization 'Y all yield the
same value for the integral
( Proposition
2.1.9 ) .
48
All of the known proofs of the Cauchy integral formula are rather tech
nical in nature.
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
realvariable calculus. You should not, however, regard it as a substitute for
the rigorous proof given earlier.
Thus
>
itself elsewhere. Assume the Jordan curve theorem; that is, assume that C \
boundary. Let U denote the (bounded) open region that is the interior of
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
1( [ 0, 1]) U
interior), then
1 p dx  Q dy=
fr
Let
i f(z) dz=
Writing
f=
iv
i f(z) dz
= i(
0.
iv)(dx + idy)
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)
EU, then
f(z) =
formula.
J f(() d(.
2ni fy (
z

Again we use Green's theorem, noting first that Green's theorem applies just
as well to a domain bounded by
differentiable,
simple closed curves 11, ... , /k (with the "inside" curves oriented clockwise).
Let
0 be so small that
>
\ D(z, E ) .
D(z, E )
The function
(
(() = f( )
(z
just because
(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)
1r
Let us do some calculations with the righthand 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
D(z, 1: )
t
t
is I.
But
I= f(z)
by Lemma
2.4.l.
laD(z,l)
d(
z
= f(z) 27ri
'>
2m
t
0)
that
J J(() d(
fr (  z
= f(z).
But
/:
[O, 27r]
t
So define
ing:
i ( d(
(2) i c d(
k
(1)
=0 if k E Z, k
# 1;
=27ri.
Actually, we proved
(2)
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
0= fk(r(27r))  fk(r(O)) =
18f d( frJ (
fr (
k
d(.
51
(1)
i(
kd
2rr
If k +
fo (
fo (
2rr
cost+ isint k
 sint+ icost dt
()
is positive, then
( ) fo
*
2rr
[ cos
)]
k + l t+ isin k + l t dt
0.
cost+ isint
cos nt + isin nt ,
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
1).
((
))
((
) )] dt
 k + l t  isin  k + l t
Note: We have incidentally shown here that the DeMoivre formula holds
P(z)
n=O
l l
for z <
P(z)
1,
2m
n
anz .
i (P(()
'Y
 z
(z
1
=
( 1  z /(
52
Since
than
1 for
ICI
1. Thus, by
=
<
lzl
1,
+oo
1
1 z;c
lz/(I
is less
7),
(z) k
L (
k= O
Hence
J P(() d(
P() ( m) d(
fy (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).
'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.
z,
even
possibly infinite sums. Formally, the same arguments would apply even if
P(z)
L anzn
00
n=O
instead of
P(z)
53
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.
'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,
'Y [a, b]
2.6.1)
U is a piecewise C1
and if
C is a
i f(z) dz= Li
Y
where
J=l
f(z) dz,
Yl[aj.aJ+iJ
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
54
[a, b] +
U, U open in C,
(c)=a
and
(d)=b. Let f:
[c, d] + U
'Yo :
U+ C
is
be a
piecewise C1 curve.
be a
continuous function on U.
1'r
1'ro
a holomorphic function
f(I(b))  f(I(a))=
Proof.
wise
and if/:
[a, b] +
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,
55
Figure
2.4
2.4)
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}
+
/'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 zeroconsider 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
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
Figure 2.5
The same proof establishes the following result:
{z
r
<
r1
C : r
<
r2
<
<
<
<
:
A=
C be a holomorphic function.
+
If
J(z) dz=
rl
J(z) dz.
r2
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
+
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
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},
4
P be as in
8D(P, 1: )
Though it
is in that
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}.
l
27ri
1 J(() d( =
(z
l
27ri
1 f(() d(.
le (  z
f(P).
58
Figure 2.7
ez
Figure 2.8
_1 1 f(() d(
27ri h (  z
for
as
as
59
'Y2
Figure 2.9
f(()
f(()
f(()
J
d( = J
d( + J
d( = O + f(z).
2m fr (  z
27rt ln (  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=
U open.
that can be
U.
D(z,r)
U. Then
60
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
or a rectangle. Let
is a
closed curve).
U be a C1 curve satisfying
"f(O) = 'Y(l)
Prove that
i f(z) dz= 0.
2.
lr1
lr2
zE
U, then define
'Yz
0.
Let
f be holomorphic
on U.
to be the path
Prove that
lrz
F is a holomorphic antiderivative
for
f.
(a) J dz
fr z
where
'Y
(0, O)]
with counter
clockwise orientation;
(b)
iz
z2z dz
where
'Y
(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
fr 8 + z
wise orientation.
5. Evaluate
'"Y zJ dz,
'Y
is a circle with
0.
61
Exercises
f, 1b (
a=
Here the
u("!(t)),v('Y(t))
[a, b]
+
u by the formula
"(1(t) dt .
"(2(t)
annulus. Calculate
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.
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.
9. Assume that F : U
( a)
(b)
+
+
C, then F
62
Exercises
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'
exist on
related?
15. Let
be realvalued and
that
is harmonic on
morphic function on U.
16. Let
g
U \ {O},
17.
assumed
to be continuous at p.
i ( d(
(b) i
d(
(
( + 4)
+ i)
( ) i : d(
( 2
(d) i ( ( + 4) d(
( ) i ( d(
3
(f) i :(: 7' ) d(
) S
(
(a)
with center 0
_ 1
with
clockwise;
clockwise orientation;
clockwise orientation;
D(z, r )
U and 'Y
F (w)
Prove that
F is
FE C0(U).
holomorphic.
wE D(z, r )
it holds that
F
J (() d(.
27ri 1r (  w
Exercises
63
Figure 2.10
C0(D) (here,
21. Let
D.
of course,
{ J j J(()
is valid if we
{z : lzl = 1}.
be a continuous function on
2.4.2)
F(z) =
(z)
27fi
22.
23.
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
*
1r (  z d(
if
Consider
given by
fo21r f(2
ei8
) d(}
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,
j
lr1
F(()
(
d( =
j (F(() d(.
lr2  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,
(b)
for all 0
1
J(() d(
laD(O,r)
< r <
1.
1)
and
26. Let "( be the curve describing the boundary of the unit box (see Figure
2.11),
i zkdz
if k is an integer unequal to
1. [Hint:
0
Calculating these integrals from
31.]
Figure 2.12.
about using a different curve. See also the hint for Exercise
27. Do Exercise
J
z3dz
laD(8i,2)
'
1
(zi)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.
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
( Note:
We are
to evaluate the integrals over the top and bottom halves of the given
curves.
32. Let
at
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.
1
1
(
t
27l'i "I ((  1) ((  2i) d ,
with center 0, radius 4, and
counterclockwise ori
entation.
37. Calculate
1
(2+ (
('
t
27l'i
 2i) (( + 3) d
"I ((
circle with center 1, radius , and
5
counterclockwise ori
1
1
d(
d(
t
t
27l'i "11 ((  1) (( + 1) = 27l'i "12 ((  1) (( + 1) '
8D(l, 1)
where 'YI is
C\ {1, 1},
39. Let "( be the unit circle equipped with clockwise orientation. For each
real number
.A,
{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
( a)
(b)
( c)
"12
"12
"ll
"11
"fl
Exercises
67
as
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?
*
piecewise linear if
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
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=
0.
43. If
is holomorphic.
f.
1
f(z)zi dz= 0 , j = 0, 1, 2, ... ,
laD(O,l)
then prove that
*
f = 0.
f: U
t
f"Y f(z) dz= 0 for every curve 'Y such that {'Y(t)}
is holomorphic.
[Hint:
Refer to
any fixed,
[STE] .]
Chapter 3
Applications of the
Cauchy Integral
( continuous )
first
(k + l)st
U.
Then
f E C00(U).
( 8z ) k f(z)
U C
z E D(P, r) ,
(
is continuous on
D(P, r)
and
f(() 1 d(,
J
(
z
27ri 11(Pl=r ( ) k +
8D(P, r) .
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
fT
It follows that
and
r
lz Pl
>
0 for all (
8D(P, r) .
+ (f(()
z

69
3. Applications
70
h+O
of the
Cauchy Integral
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)  (fz
+
h+O
f(z + h) f(z)
h
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

71
one of them is
phic
derivatives by
f', f",
and so forth.
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
(!')
Oz
( )
8
az t
az

( )
.
az t
Oz
C1
function
f'
is holo
3.1.1,
Theorem
3.1.3.
the function
f given by
f(z)
 Pl
instance, if P
I(
=
=
0,
r }, then
</> (
J
( ) d(
2m 11(Pl=r (  z
(*)
D(P, r).
1 , </> (
()
(,
then
f(z) =
0,
</>.
For
2.4.2)
</> is already
D(P, r),
then
( Theorem
guarantees that
</>(() d(
J
27ri 11(Pl=r (  Z
l
h(()
J
d( h(z).
27ri 11(Pl=r (  Z
integral coincides with </>
h as z
8D(P, r).
=
( )
*
</>.
72
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.
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/;
73
followed by
the
0=
1 f(() d(
J
so that
t f(() d( i f(() d(
=
'
1 f(() d(  1 J(() d(
Jt/Jh
Jtf;
1 f(() d(
hh
F(x + h, y)  F(x, y)
lh f(P
U and
s)ds.
U(x + h, )
Letting
h+
U(x)
h
l
l J(P
Re
s)ds
ll
Re f( P +
( we
s)ds.
[a/ax]U(p) exists
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.
au av
au
ax .. ayrta ay
/
av
.
ax
74
Thus F is holomorphic.
holomorphic. But F'
f,
Now Corollary
3.1.2
f
f
,,
curves in U. This sufficient condition for the existence of an antiderivative
2.4.3).
,,
J( ) d(
i J(()
for every pair of curves 'lj;,
d(
i J(()
d(
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.
JR,
namely
oo
L.,,,
J(n) (p)
I
n.
n=O
(x
 )n
p ,
pER
x other than
converge at some
x I p,
J(x).
An instance of
J(x)
el/x2
if
x I 0
x 0.
=
f' (0)
f" (0)
J(O)
75
0+
Ox + Ox2 + Ox3
+ ... ,
f(x)
is 0 only if
0. (An example of the phenomenon that the Taylor series need not even
calculussin, cos,
z.
( C00
functions from
t
f(z)
JR
to
JR),
the attempt
expansion at P ,
near P. Specifically, if
converges to
for all
z in
>
f(z)
for
z.
sequence
{l, 2,... }
to
from
{ak}b1
{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
Proof. If
k No
{ak}
implies
converges to
lak fl
<
<
E/2 + E/2
such that
E.
P)
PE
C be fixed. A
(centered
I:ak(z  P)
k=O
where the
( Note:
{ak}0
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
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.
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.
k
L ak(z  P)
Proof. Since
ak(w  P)
k
ak(z  P) t 0
k .:::;
lak(z  P) l M
l .:::; lakllw  Pl
k.:::;
( lw Pl ) k
as
or
77
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.
We now
2:::%':0 ak(z
sup{lw  Pl :
the
L ak(w  P)k
radius of convergence
disc of convergence.
is called the
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,
O; in this
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
78
such that
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 casebycase 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)
+oo
( )
converges, then
79
Lo O:k converges.
R), 0:::; R
<
r.
The
Lo fk(z)
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)
<
k=O
The point is that
depending on
No does not
( but
not on
z)
depend on
z EE:
E,
an
No
z EE.
No such
Lo fk(z)
uniformly Cauchy
that if m
j No,
>
0, there is a positive
then
1n
L fk(z)
k=j
<
for all
z EE.
L lfk(z)I
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:
Lo ak(z  P)k
series
series
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.
l: a1(z  P)l
j=O
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)Jk,
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,
ll(aj(z
In particular, for
1h
1
Now
lhl
!(r  lzl),
.
lzl))Jl
fo1 h j z
aj
h)j ajzj)
(aj(z + h)1ajz1)
LJlaJl(!(r
h+O L.,,
j=O
( )
*
th)jl dt.
jlaJl(lzl + lhl)j1 .
) j1
Hence, by the
Weierstrass Mtest (see Appendix A), the series on the righthand side of
(*
converges uniformly in
for
the summation and limit can be interchanged (cf. Appendix A). Hence
exists and equals
:L:'; o jajzJ1.
f1 (z)
(by Lemma 3.2.6). The conclusions of the lemma now follow inductively.
Now we have:
is infinitely differentiable on
D(P, r) and
(:z ) k f(z) f j (j
=
J
1)
(j  k + l)aj(z  P)Jk_
So
81
( : ) k (P)=k!ak
J
or
k J(P)
(/z)
ak= k!
L b1(z  P)j =J(z),
00
j=O
k J(P)
(/z)
bk= k!
for every k. As a result, ak=bk \:/k.
f
D(P, )
f (okf/k)(P) (z  P)k
PE
be holomorphic on U.
k=O
r.
It converges to
J(z)
on D(P,
).
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
2m Ji
f(()
d(
(l=r' ( 1 z. (1
so
(* )
hence
1/(1
(1 )
integral.
r'}.
2).
SN(z, ()
2::=0(z C1 )k.
Then we have
! SN(z , ()d(
lim J
27l'i N>oo Ji(l=r' (
J
2m Ji
(l=r'
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.
( **)
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
**
sum and the integral. The rest was trivial formal manipulation, using the
83
,, .......
I/
I\
\..
I
I
...... _
Figure 3.3
1/((  z)
idea of writing
as
PE U,
f on
P to C \ U.
D(P, f).
r.
be
Thus the
f : D(O, 2)
f about 0
C be given by
f(z)
lj(z
2i).
The
is
00
:_.)2i)kl ( 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
z =
2i,
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
k=O
v'5)
ular, the convergent power series gives a way to extend the domain of the
function given by
(*)
v'5)
2i).
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
oo
sinz
and
.
z2j+l
(1)1
(2j + l)!
.
z21
L(1)1 ' ) I'
(2J .
j=O
oo
cosz
The possibility
85
fk(x)
l; but
sin
kx, x
IR, k
For example:
The functions
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.
is contained in U.
have
Proof. By Theorem
Set M
: U
+
be a holomor
= 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
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
l/k 1
(k)
j (P)
lim sup
,
k'
k++oo
for all
such that
3.4.1
that is
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
on U, then
is holomorphic,
Thus
is said to be
: C
So J is constant.
A function
is const ant on u .
+
cos
are entire.
The function
O.]
The
PE
0.
>
(z) I
The result is
az f(P) I :::; r
l
r 0,
(P) =0.
Ml!
Since
>
= E D(P, r).
C. F ix a
1 on
we conclude that
a1=0.
az 
By the lemma,
is constant.
aj /az(P) =0.
akf (P)
l 8zk J :::; :;:k
ak /azk(P) = 0 P
f P
f
M k!
Hence
at
for all
0.
0.
Since the
is constant.
87
Theorem 3.4.4. If f : C
> C
for all
with
lzl
>
1,
then f is a polynomial in
k
k
(8 +f./8z +f.)J(O)
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.
p(z)
lp(z)I> +oo.
Liouville's theorem,
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
p1
is a polynomial of degree k
P2(z)
1,
By
no remainder to obtain
( z o) Pi (z).
1
. If k
p(z)
oo,
(z 
1,
Pk
of degree
O;
that is,
Pk
is constant. We
o1, ... Ok
p(z)
(z 
01
) (z  ok).
88
3.
3.2.10).
( in
z)
defines a holo
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 wellknown 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 realvariable situation and that of the the
C00
and
[RUDI]).
3.5.1 and
sketch a second.
D(P,r) U.
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
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).
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 CauchyRiemann 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
are
as
k
k
(z
f
)
1
:
)
; ( :z ) f(z)
(z
for each
the set
z EE,
(u D(z,r))
zEE
we see that
:'.S;
k sup
r (,EEr
l!Ji(()!12(()1.
90
)1 , )2
{!j }
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
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,
3.2.6. ( Exercise:
1.)
Holomorphic Function
be a holomorphic function. If
out that
lfk
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 }
{z
such that
Zj
f(z)
 zo, then J
O }.
=
If there are a zo
0.
U
E
Z and
91
Zj}
Zj
D(s, 1: )
n S=
}.
{s
E Sthere is an
1:
>
0 such that
set has points which are "isolated " or that S contains only "isolated points."
Theorem 3.6.1 thus asserts that if
on a connected open set, then its zero set is discrete or, less formally, the
zeros of
\ U;
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, ...
}.
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)
f(z)
00
J=nu
Therefore the function
g(z)
D(zo, r),
(()azjJ f(zo))
(z  zo)l
j!
defined by
=
j f(zo) (z  zo)Jno
)
(!_
_!
oz
L
J=no
is holomorphic on
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
92
{z
(8/8z)j J(z)
EU:
0 for all
j}
{z
EU :
(8/8z)J f(z)
O},
f(z }
Since
flD(P,r)
nJ EJ .
t, (:J
/ (P }
(z
rP)J
>
0 is
E D(P, r ) that
"0.
D ( P, r ) E. Therefore E is open.
is 1101omorphic on U and
JID(P,r)
0, then
0 on U.
f =g.
{z
EU:
J(z)
g(z)}
f, g
be holomor
f, g be
or g = 0 on U.
holornorphic on U. If f
g =0
on U, then either
0 on U
>
0 such that
By Corollary 3.6.2,
g = 0.
be holomor
for every
j,
then
J = 0.
f(z)
By Corollary 3.6.2,
f : (
j=O
0 on U.
P)
(z
!P)J
z
=
E D ( P, r ) ,
0.
D
93
Holomorphic Function
z,
cos
z,
and
z ER
3.6.3 is that such extensions are unique, as we state explicitly in the following
corollary :
and
if f(x)
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 all of C.
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 (realvariable) definitions for ex, sin x, and cos x. It is an easy exercise
to check that this function is holomorphic on Cby verifying the Cauchy
Riemann equations, for example (see Exercise
the right function values on IR since
as
in fact the
Lo zl/j!,
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)
I::0zJ/j!
We also get "for free" the usual identities for the exponential function:
94
3. Applications
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
Exercises
1.
fj
.
a
a
azfj az1
__...
.
z)
i ( z d( .
3.
4.
11
1
11
1
11
1
{fj}
Exercises
5.
6.
7.
8.
9.
10.
11.
95
Exercises
96
12.
13. Let
f : (1, 1)
t
a constant
and
Mk k!
for all k E
Lo akzk
lxl
<
1.
But
which con
J(x)
Lo akzk
0 converges
: IR
t
f!)
IR is continuous,
is real
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.
( a)
(b)
(c)
(d)
0.
i.
2  i.
0.
statement of Theorem
Determine the disc
Exercises
21.
97
J(z)
00
I: r1z(21)
10
is holomorphic on
is a
(2N)th
D(O, 1)
and continuous on
D(O, 1)
22.
D(O, 1).
J(z)
z.P g(z)
lim
of/oz
z+P og/oz
lim
23.
24.
D(O, 1).
aj
>
D(O, 1).
j
L a1z is convergent on
l.:(aj + E)z1 is convergent
0, j = 1, 2, .... If
D(O, 1)
small , then
26.
[1, 1],
C00
and bounded by
on the interval
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
Let f: U+
U. Prove that
where
is the distance of
to
s p
<C \ U .
98
Exercises
l D(O, 1)
:
+
Ill 2.
Derive
an estimate for
30. Let
C,
k,
not depending on
C.
such that
( : ) kl(P) Ck!.
z
( : ) kl(P) l (k)I
p
k.
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
formula to obtain
for
(b)
<
<
integrate in
Let U
r.
12(0)
from
0 to r. ]
l is holomorphic on
s p Il l
and
be a compact subset of U.
if
K)
such that
U, then
Po
C.
99
Exercises
Q(z)
ak z k
(t ) (t )
k=O
akzk
k=O
Then
course
the
111"
11"
where
Now
is a nonvanishing polynomial;
(* )
contradiction.
37. Let
0 is sufficiently
{Pj}
0 . This is a
{Pj}
converges uniformly on
38. If
fj :
l!J I
2j,
00
L:Jj(z)
j=O
39. Let
cp: D(O,
1)+ D(O, 1)
be given by
cp(z),
'Pj(z)
cpocpj1(z),
and
cpocp(z),
{ 'Pj}
cp?
(i)
cp1(z)
cp2(z)
40. If
cp(z)
f og is welldefined,
f og is holomorphic:
is a polynomial.
then complete
Exercises
100
(ii)
(iii)
Any
{!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
and let
t
t
If
prove that
on compact sets.
fn(zn)
42. Let
{!j}
If the sequence
( :z )
fn(zn)
( :z )
t
f(zo).
be holomorphic on a neighborhood of
Prove that
D(P, r ) .
D(P, r ) .
Suppose that
is
on JR such that
{x
[Hint:
JR:
f(x)
=O} = E.
* 44. If
f : D(O, 1)
t
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:
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
* 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
[)k f
(P)
[)zk
X.
space. Prove
f . 0
be holomorphic
49. Check directly, using the CauchyRiemann equations, that the function
ez
*
50. Let
(!
f and g be
g)(x) x.
Prove that
is the inverse of
*
J'(O)
1.
borhood of
function
of
*
and
x E JR.
it holds that
g.
is a holomorphic function on
Consider whether
D(O, 1)
defined near
with
with
f(O)
0 and
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
54. Prove the following very simple version of the CauchyKowalewski the
orem:
Consider the differential equation
'
y +
where
a(x), b(x)
b(x)
a neighborhood of
L%:o akxk
a(x)y
0E R
55. Let
y(x)
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
mation
L%:i zk /k
converges when
by partssee [RUDl].]
lzl
1, z i
l.
[Hint:
Use
sum
102
Exercises
ferentiable in
az
for all values oft and
g(t, z)
z. Prove that
1
J(z)
g(t, z)dt
=
g(z)
C1 [O,1], then
11 <P(t)eitzdt
<P to be
* 58. Let
at a single point.]
<P(x)
is
{0
x :c:; 0
if x > 0
if
el/x2
<P(x + 1) <P(x + 1)
[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
Chapter 4
Meromorphic
Functions and Residues
\ { P}
1/ z], j
a positive integer.
More complicated
e1/z, z i 0.
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
isolated singularity)
at
P.
: U \ { P} + C is
isolated singular point
P.
P" .

105
106
distinguishing:
(i) lf(z)I
is bounded on
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
+oo.
applies.
Of course this classification does not say much unless we can find some
other properties of
tarily that if case
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.
D(P, r ) \ {P}+ C
(1)
limz.P
L et
f:
f(z)
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.
( )=
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.
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
107
r by
00
L a1(z  P)12.
j=2
H(z)
Then
phic extension
In conclusion, the
f of f to D(P, r).
is a holomorphic function on
then
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.
(z)
lim
z+P uX
we need to see
(z).
lim
z+P uy
Since
is holomorphic at
zo ,
I (zo)I ::::;; I zo
Therefore
og
(zo)
ox
I l
=
og
(zo)
oz
pI
M
lzo  Pl.
t
og
(P)
ox
108
as zo
4.
P.
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.
=
ro.
Then there
> f.
for all
109
g(z)
for all
which
g could
g(z)
z
g(z)
l
=
f(z)
,\
and obtain
f(z)
,\+ g(z)"
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
P [case (ii)
(** ) is
f has a removable singularity
D(P, ro);
at
P.
P,
in other words,
P,
a holomorphic
P are always
P ( these series
are called "Laurent series" ) . These will be considered in the next section as
well.
D(P, r ) \ {P}.
P,
neighborhood of
P,
including
itself.
(z
P).
This
110
both. That it works well for poles is easy to see; essential singularities take
a bit more work. We turn now to the details.
+oo
j=oo
aj(z  P)J.
Note that the individual terms are each defined for all
D(P, r) \ {P}.
L.t::._00 aj.
converges if L.t aj
0
E.
there is an N >
L.: aj converges.
See Exercises
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.
4.1
converges
Lemma.
L..j+oo
=l a_j(z1 P)J.
L..j+oo
=oo aj(z1 P)J converges, then so does
lz2PI. If
Since 0< lz1 Pl< lzPI, it follows that 11/(zP)I< ll/(z1 P)I. Hence
j
Lemma 3.2.3 again applies to show that L.j:t aj(zP) 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
r2},
=
r2,
111
,,.
.,., 
I
I
I
I
I
\

..... ......_z2
'
'
I
I
',_
'
'
' .....
l
.,.,
\
I
I
I
\'
'
\
I
I
'
Figure 4.1
{z
: 0
r1 <
D(P, +oo)
{z: lzPI
<
+oo}
and
D(P,O)
{P}
D(P, r2 ) \ D(P, r1 ) ,
r1 < r2
+oo.
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)
with
The second
statement follows from the corresponding result for power series combined
with the argument used to prove Lemma 4.2.1.
112
oo l
I: z2
j=10
#0
lzl
<
1,
diverges for
izl
>
1,
and converges
zl
L,,
J=(X)
j2
#0
 oo )
{z: izl
l}.
50
I: 21(z + i)j
j=00
1/2}.
4.2.1
3.5.1).
r2 :::; oo.
D(P, r1)
j E 7!..,
to a function
f,
a1
In particular, the
D(P, r2) \
r2, and each
converges on
satisfying
r1 <
<
f .
_l 1
(() d(.
27ri 11<Pl=r ((  P)1+1
r1 <
d(
11<Pl=r ((  P)1+1
(()
{z
I z  PI
ak((  P)k11]
[
11<;Pl=r k=oo
J
d(
r}.
It
113
+oo
((  P)kjId(.
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.
By explicit calculation
((
11(Pl=r
Hence
i1(Pl=r
k#
k=
j)
j).
+ d(= 2maj.
((  P) i
f(()
Theorem 4.3.1
( The
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
D(P,r1),
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
i(Pl=s1 gz(() d(
and
1l(Pl=s2
By the considerations in Section
1l(Pl=s2
2.6,
9z(() d(.
9z(() d(  J
11(Pl=si
9z ( ( ) d(
the
114
4.
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
 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
and
f D(P,r2) \ D(P,r1)
converges
<
r2
oo
+oo
L aj(z  P)j
j=oo
on D(P,r2) \ D(P,r1)
to
f.
Ifr1
r2,
D(P, s 2) \ D(P, s1 ) .
Proof. If 0 r1 < s1 < lz  Pl < s2 < r2, then the two integrals on the
righthand 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
 (z  P)/((  P)
1lc.Pl=s2
For
s1 <
f(() d(
(Z
J(()
J
(
+id( ) (z  P)l.
j O 1lc.Pl=s2 ((  P)J
11(,Pl=si
J(()
d(
(Z
the formula
_ 11(,Pl=si
J(() . l
__d(
1  = Z  p
 J
J(()
zP
11(,Pl=si
((  P)J
(z  P)J d(
11 6
j=O [llJ(Pl=s1
 I:
f(() ( (  PF d( (z  P)
=1
f(()
JL Ji  d(
1
(( P)J+l
J=oo I( Pls1 
[1
Thus
j=1
f(()
L Ji1 .=(( P)J+l
J oo I(. Pls1 
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(Pls2

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
llj
1
_
27ri
1
11( Pl= r

!(()
J+ d(
(
(  P) l
D(P, r ) \ {P}
D(P, r ) \ {P}
t
f(z)
t
is holomorphic, then
aj (z  P)
aj =
1
2 7rZ.
has a
f(()
d(,
117
There are three mutually exclusive possibilities for the Laurent series
4.3.3:
of Proposition
(i) aj
(ii) for
(iii)
some k
neither
>
0, aj
nor
(i)
applies.
(ii)
case
(iii)
(ii)
4.1:
Case
(i)
is a pole; and
are almost obvious, but not quite. Therefore we shall discuss them in detail,
each implication in turn.
::::}
centered at
that converges on
D(P,r)
since
f is
(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
lI 
+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)
+oo.
118
Pole:::::} (ii):
less than
such that
g
Then limz_,p
is never zero on
g(z)
Consider
C.
0.
D(P, s) \ {P}.
H(z)
is holomorphic on
{ g(z)
on
at
D(P, s).
D(P, s) \ {P}
P
positive
it
holds that
H(z)
(z  P)m Q(z)
on
+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
on
{z
: 0 <
lz  Pl
< r
},
and
(ii)
holds.
(iii)
:?
Essential Singularity:
The equivalence of
(iii)
and
being
f(z)
j=k
aj(z  P)J
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
(z  P)k f(z)
is bounded near
but
(z  P)kl f(z)
is not.
P,
expansion of
around
principal part of f
the
00
f(z)
for
near
When
has a
P,
then the
aj(zP)j
at P. That is, if
j=k
principal part of f at P is
1
L aj(zP)J.
j=k
the formula
aj
Proof.
1
(k+j)!
This is just a direct calculation with the Laurent expansion off and
f(z)
(i)
I z=P
Let
g(z)
29)
(z1) f(z).
1 is
Then
1+ (z 1).
Therefore
f(z)
g(z)
z 1
g is holomorphic and
expansion about
g(z)
for you.
(z 1)1+1.
120
(ii)
at
(z  1)1.
is
aj
If j ;::::
1,
1
_
27ri
1
_
27ri
(j
aj
<
1,
(*)
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.
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.
2
is of the form
1
j=2
aj(z+2)J.
a_1
a_
f about i,
(z + 2)2 z=
z=
121
(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)
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
eixy
eix+y
2i
it is clear that sin z can vanish only if
ey
ieixy I
ieix+y I
eY
or
y
But when
0.
0. So
4.
122
(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.
1
1 z .
D(P, r ) \ {P}.
: D(P, r ) \ {P} +
P.
More precisely,
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
"'Y
for which
"'Yf
is necessarily 0 if
(abbreviated h.s.c.) if
f:
U+ C,
UC
The connectedness of
U+ C
=f.
The important
Lemma 4.5.2. A
each
4.5.
123
1
C closed curve r in U,
=0.
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.
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 righthand 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:
'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
(1t 'Y'(s)/['Y(s)  ] s)
Pd
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)
_
(1t "f1(s)/('Y(s)
 P) ds
('Y(a)  P) e0
g(a)
g( b)
b'(t)  'Y'(t)]
a and t
0.
125
(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
)
ds) .
ds
P)
1
27ri.
So the quantity
is an integer.
11)
winding of the curve I' Note in particular that a curve which traces a circle
k times
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
Using the notation of residue and index, the residue theorem's formula
becomes
if= 27ri t
j=l
1'
Res1(Pj) Ind,,(Pj)
27ri
around
/' 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:
1, 2, . .. , n,
expand
s j(
be the
principal partthat is, the negative power partof the Laurent series at
Pj. Note that the series defining
Pj ) 1
f L
f L
with both
function
each
Sj and
Sj defined on all of U\ {Pi, ... , Pn} Moreover, the
Sj has removable singularities at each of the points Pj because
f
no negative powers at Pj. Since the set U is h.s.c., we may thus conclude
that
1 26
i iL
or
f=
It remains to evaluate
s1.
on y; therefore
"I
Sj(()d(
"I
"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(
Res1(P)=
Proof. This is the case
) ((
:
l
!
(
)
(k
z
kI
j=
 P)k f(z)
z=P
1 of Proposition 4.4.1.
z
f(z)= (z + 3i )3 (z + 2
)2
[O, 1]
+
C is a
)I
(
.1
2_
(
1
( 2z )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
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
.
Indy(3i) +
2
f(z) dz=
4 + 3i
(3i
2)4
we obtain
15i  20
(3i  2)4 .
Indy(2).
128
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
partial fractions.
00
oo
1
dx,
1+x4
f ( z)
'YR
1/(1+z4)
1
dx.
1+Z 4

Hx
as
l x4 dx, with R
t
oo.
'YR.
curve or
'Yh(t)
'Yk(t)
t+iO
R eit
if
if
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
=
21 9
Curve
R
4.4
Figure
shows that
and
'YR
inside
1 .
= 4(/1 J2+1 i/J2)3  4(1  J21 + i J2)
.1
1 : z4 dz=2i ( ) ([ + i ) +(  + i )]
es
1(P2)
Therefore
(*)
1 z.
1 z+ i d
1 z=i d
ir d
4
4
z
z
z4
1
+
+
1
+
1
rh
rh
Trivially,
l_
t
oo
as R+
hh
dz
1+ z
I I
rh 1+ z
/ 1.
130
( Here
ll +z 4 1
lzl4  1.)
1
h l +z 4
J __ dz + 0
Finally,
(* ) , (** ) ,
and
(*** )
as
Thus
R+
oo.
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
v'2.
1h)
do tend
to zero.
J00 s 1 dx
x
Figure 4.4. The obvious choice for the complexification of the integrand is
eixeY + eixey
eiz + eiz
=
1 +z2
1 +z2
leYI ::::; 1 on f'R but leixeyl = leYI becomes
f(z) =
Now
leixeYI
on /'R when
quite large
contour so that good estimates result. Instead, we alter the function! Let
( for R
J g(z) =
hR
>
2rri
2rri
On the other hand, with
1),
Res9(i) Ind"IR(i)
( J =
2
and
rr, we have
hR
Of course
i'Yk
hh
l eix
dx
1 +x 2
oo
g(z) dz+
oo
as
R+
oo.
131
R
Figure 4.5
Also,
j g(z) dz
frk
Thus
100
cosx
_00 1 + x
2dx

k)
sup
7k
1
+
R2  1
= 1_00= 1 eixx2dx =
Re

Re
R+
as
oo.
( 7r ) = 7r
e
.
( sin x) / x
f00 sinxxdx.
oo
by integration by parts.
oo
( as
So the problem
makes sense. Using the lesson learned from the last example, we consider
k(t)
'h(t)
(t)
k(t)
Clearly
t
eit / R
t
Reit
if
if
if
if
0 and that
be used. We
132
R > 0,
t g(z) dz=
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
( )
y <
y >
)
1 g(z) dz
Ji
Finally,
R) oo.
>
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
( **** )
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
100 2dx.
xl/3
l+x
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
4.6.3.
4.6.3.
Clearly
f,
f(z) dz+
100
0
tl/3
2dt.
1 +t

0.
134
Of course that is good, but what will become of the integral over k? We
have
tl/3
1R1/R 1dt
+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)dzt (1+ (+ y'3i)) r = l 2dt
}0 1+t
J
Jk
( by
Thus
our definition of
J( ) dz
z
::;
7r
as
So
as
2 7r Res1
>
Ind
. J
x(+i)
Finall y,
Rtoo.
Rtoo
R RR2 I/31t
2rr . eit
i
(eit e/ R)it I/3 J J(z)dz
1
rr 1+0 2 / R2 R dt
J
R too.
R4/3 1rr 1+eei42t/it3/ R2dtt
Rtoo.
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
**
1= dt=tl/3
2
0 1+ t
7r
y'3"
as
135
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 welldefined 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,
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
Ri
Curve TIR
{R
Tlk
R
ill
i
{R
Ri
Figure 4.7
as
Thus
t
oo.
 i/v'2R)t 27ri.
dt
1 g(z)dz+ 1 g(z)dzt27ri f 00
2
l
t
+
6t + s
o
h,k
h,'h
t
oo,
(***)
1
(00
dt
g .
=
lo t2 +6t+8 210 2
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) ( =
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 ,
+
. 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
D(P,r) \ {P}
U.
U \ {P},
where
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
is a positive number
Sthere
(depending on S and on
Sn
D(z, r )
{z}.
Definition
function f :
(a)
(b)
(c)
4. 7.2. A
U \ S +
meromorphic function f
<C such that
on U
<C),
for each
{z},
Sand
the function
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,
D(P, r) \ {P}
and has a
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:
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
lim IF(z)I
z+P
0.
l
z+P lf( Z)I
lim
+oo
=
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):
4.
140
(ii) f
(iii) f
has a
pole at oo
has an
at
if
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
G(l/z)
Lanzn
lzl
>
yields im
R, namely
Lanzn.
oo
oo
The series
for
+oo
+oo
f(z)
E: anzn,
It follows from our definitions and from the discussion following Proposition
4.3.3
that
series has no
pole at
of
positive
powers of
oo
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
Theorem 4.7.5.
limlzl++oo
f(z),
00
f(z)
converges for all
Lanzn ,
n=O
C. Hence
00
G(z)
for all
\ {O}.
J(l/z)
Lanzn
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
power series
n=O
As noted,
has a pole at
oo
4. 7. Singularities at Infinity
141
f(z)
ao.
R'
has no poles in
{z
<C: lzl
>
R'}
and that
f has
on
<C
>
R,
a pole at oo.
=
<C.
<C
z) .
and at oo.
( in <C),
discrete set
( If
S.)
since
would have
142
=a ECthen 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.
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)
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 socalled 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
z E C, z #
0, correspond to
>
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+
4. 7.
Singularities at Infinity
143
{ oo}
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
F:
CU { oo}
open subset U of CU
to be holomorphic if
( 1) F
F1 (IC)
{ oo}
+
CU { oo} on an
is holomorphic
and
(2) p1({00})
open neighborhood of
1/F
is holomorphic on some
p1 ({ oo} ).
{ 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 } .
4.9.
4.10
JR3
: x2
y2
z2
1},
cut by the
xy
{ (x, y, z)
144
4.
00
Figure
Figure
4.9
4.10
(0, 0, 1)
as
The idea now is to have a geometric scheme for mapping the plane to the
sphere.
Let z
(x, y)
iy
stereographic projection
of z to
the sphere.
It is easy to see that
except the
{ 7r ( z)
: z E
C}
are points with very large absolute value. Thus it is natural to think of P
as
A set in
CU { oo}
CU { oo}
to S by z
tt
( )
7r z ,
z E
C, II(oo)
(0, 0, 1).
Then
II
II
is a
145
Exercises
II
II
II1 is
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:
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
3. Let
be holomorphic on U
singularity at
P,
PE U
Aj
{!
has an essential
1/f
have at
P?
P?
\ {P}: f
(iii)
open. If
4. Let
\ {P}, PE U, U
(i)
removable,
P},
(ii)
pole, or
Division?
5. Let
1
( a ) ,
z
. 1
{b) sm ,
z
1
( c ) 3  cosz,
z
{d) z. el/z. el/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). zj
j=l
:f. 0 and defines a function which has an essential singu
0.
D(P, r) \ {P}.
mann's theorem:
( a)
*
{b)
If
f is
holomorphic on
and limz_,p(z
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)
Q;
l(Q,E)
IF(z)l2dxdy
D(Q,E), then
IF(z)l2dxdy
0, then
oo,
IF(Q)l2
(to all of
F2 and
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
f: D(P,r) \ {P}
P,
D(P,r) \ {P}
{Zn}
147
Exercises
1/(z
E: an does
E: an
such that
not converge.
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.
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)
(b)
(c)
14.
2:;:_00
1
"' 00=D 4j 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?
117118
16:
( a)
Prove that if
P,
(b)
(c )
then
value of
1/ f
1/ f
has a removable
at
is holomorphic on
P equal to
0.
1/ f
P.
at
Prove that (z
 P)k f
g is holomorphic
bounded, then
< m E
D(P, r) \ {P}, if g is
such that (z  Prg is
on
singularities given by
Exercises
148
A function f holomorphic on
D(P,r) \ {P}
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:
rr/2,
For each of these functions, identify the type of singularity at the point
about which the function has been expanded.
f(l/z).
P, then
D(P,r) \ {P}
P?
P, then ef
22. Let {a1 :
D(P,r) \ {P}
>
converges on
P.
0, 1, 2, ...} be given. F ix k
0, then L,
D(O,r).
>
0. Prove: If L,
j a1zl
+k converges on
k+l a1zl
at
1
=
ez1
(l)k1
k=l
Bk k
2
( 2k )!z '
149
Exercises
Bernoulli numbers.
B2k.
The
B2k
are called
\ { P} and has an
the function (z  Pr
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.
( a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
(i)
 5)]
 5)]
f(z)= z/[(z  l)(z  3)(z  5)]
about P
about P =3
about P
>
0.
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.
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)
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 (z2)(z + 1)
cotz
J(z) = 2
z z + 1)
'
(z2i)(z + 3)
z2 + 1
(z + 3 2z '
e
.
1)3
(z
iz
'
(z + l)(z2)
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)
) (
) (
Figure 4.11,
1
f(z)dz where
27ri 'Y
4.12,
(h)
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
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
1, . . ., k.
C and if k
on D(P, r)
\ {P }
such that
has a
: D(P, r)
morphic on D(P, r)
2m
JaD(P,r)
f?
f
and
f(z) dz
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
40. Let
f(z)=e(z+I/z).
1, 2, ...,
Res1(0)=
f(z). If
0.
.
L
k=O k!(k + l)!
r,
9k(z)=z
Prove that
00
0 < s <
oo.
faD(O,s) g(z) dz
\ {O}
and if
about 0, namely it
equals a1 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
at oo to be the
on
{z: lzl
>
R}. Prove
1 53
Exercises
y
5 I
\
'
\
'
I
'
'
./
'
15
I
I
 _,,.
..... _
Figure 4.14
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
(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 =
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)
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
Jool+x
+oo sin2 x
J oo dx
x2
+oo x4
 dx
J oo 1 +xlO
oo
eix
Exercises
155
Use the calculus of residues to sum each of the series in Exercises 5961.
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
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
64. Let U be an open set in the extended plane. Let M denote the collec
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/(1z)) has infinitely many poles in the
unit disc
D(O, 1).
67. Prove the statement contained in the third sentence of the proof of
Lemma 4.7.3.
[Hint:
Let
S1
be a set of points
S1,
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
Sn
Exercises
156
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
Questions of this
157
158
f(z)
00
1{}Jf
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.
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
J=n
(z  zot
[00
l{}Jf
J=n
n .
5.1.
for
Counting
and
Poles
159
Zeros
Notice that
H(zo)
For
D(zo,r) \ {zo},
anf
(zo) =f 0.
n. uzn
1
1
f'(()
((  zo)nH'(() +n((  zor1H(()
n
H( )
= ' ( +
=
f(()
((  zo)nH(()
H(()
(  zo

By continuity,

H is nonzero on a neighborhood of
.
the function
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
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
J'(() d(=

2m l<:Pl=r f(()
.
Proof. Set
ne.
160
for
For each
j E {1, 2, ... , k}
we write
f(()
((
1
. II
.
 Zj)ni f,#j ((  Zt)ne
5.1.1.
Thus
J'(()
f(()
H(()
.
t
(  Zf___
l=l __!!!_
!'(()
d(
1(Pl=r f(()
o+
1 d
k
k
i _( 27ri:L:nt.
L:nt
l= l
l=l 1(Pl=r ( Zf
principle,
argument
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(
_1
f(P
C1
closed curve
!'(() d(
_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 1with the notion of index that we defined in Section
r}
speaking, is equal to the number of times that the fimage of the boundary
If
in C.
9(t)
integral
!(9)
2 \ {O}
C\
{O},
we
1
.
27ri
12rr _!L_
' (t)
( ) dt .
0
g t
161
1(9)
the behavior of g about the point Pthat is, it tells us about the tendency
of g to "wrap around" at P.
Thus
1(9)
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
Lemma 5.1.3. If f: U
on U
\ {Q}
\ {Q}
t
f'(()
d(
JaD(Q,r) f(()
1 J
_
27ri
n
(z  Q) n f(z)
has a
Q) n1J(() + (( Q) nJ'(()
((  Q) nJ(()
n
'
+ f (()
.
(  Q f(()
H'(()
H(()
((
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:
D( P, r)
162
on 8D(P, r ) . Then
1 1
!'(()
27ri fa D( P,r ) f(() d(
f;
nj
{;
q
mk,
D(P, r ) and
in D(P, r ) .
Zp of f in
off
5.1.2) has a consequence which is one of the most important facts about
+
C is a noncon
+
) . As an
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}
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
In contexts
 {J definition is equivalent
163
5.2.1
C is an open mapping;
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.
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 .
this claim
0 such that
lg(()I >
We claim that, for this
d( =n.
lies in
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
lz  QI > 0.
of z E D(Q, E )
(see Ap
J()  z
in D(P,
r .
5.1.2)
and
g(()
Since n
=
1,
J(()  z
is proved.
zE
D(Q,
E .
n,
zE
Therefore D(Q,
E .
D(Q,
zE
r .
E ,
the function
In other words,
J(U),
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
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+(zP)k
=
164
near
P.
Zeros of a
Holomorphic Function
The
5.
in
neigh
itself (to kth order). In view of the fact that the power series expansion
for f about
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:
We begin the proof of the theorem with a lemma about the multiple points
of a holomorphic function.
Lemma 5.2.3.
0,
> 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},
1
.
27ri
!'(() d(
 q
f(()
8D(P,8)
[order off at
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
<
165
will always
contain
is true
f(P).
a disc about
J'(P) #
in which
rived in Theorem
in
5.2.2
D(P, 8).
11 : D(Q, E )
11 (w )
Of course
5.2.2,
D(Q, E )
+
C by
in
D(P, 8).
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
J1(D(Q, fl))
11 is holomorphic,
5.2.1),
a <
D(P, a) .
the previous
the image f
(D(P, a) )
D(Q, fl)
fl
choose
fl
8.
for some
>
a >
0 we know that
E.
f(D(P, a)) .
Q,
so
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 11 (D(Q, E )) D(P, a) as required. Thus 11 is continuous at Q.
1
Next we show that 11 is complex differentiable at Q. For this, we
fortiori
1l(Q')1l(Q)
Q'Q
Q'+Q,Q''IQ
lim
1l(Q')1l(Q)
Q'Q
P'
f1(Q'),
r
P11 '!!!p
Q'Q
1l(Q')Jl(Q)
J(P')J(P)
'
P 'P
I(
1
I
and of course
J(P ')J(P )
P ' p
11(Q)
P.
But
166
11
implies
The continuity
1/f'(P).
P and Q in these
D(Q, 8 ) and set P' f1(Q'), then we can
that 11 on D(Q, E ) is continuous at Q' as
Q'
well.
(f1 )'(Q')
1/f'(P').
Notice that
f'(P')
is nonzero,
11
(f1 ) '.
So
11
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 onevariable real calculus: You can run through
it for yourself to prove that if
is nowhere zero in
u1)1(y)
( a, b),
1/f'(f1(y)).
then
f:
f
( a, b) +
JR. is a
is invertible from
better than the complex case, because you do not have to shrink the do
main and range to obtain a welldefined function
11.
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
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
8D(P, r ) .
In particular, neither
l l
hence
This equality
take a value in
The observation in the last paragraph implies immediately that, for all
t E [O, 1]
and every
is not zero. This fact will play a crucial role in the proof of the theorem.
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,
If(()  g(()\
<
\g(()\
168
for
l fg(((())_l l <l ,
we see that it says that the image 'Y under
the disc D ( l,
1).
f/ g
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
g(z)
5z3
lzl
we
have
lf(z)  g(z)I
By Rouche's theorem,
and
f.
2
zn + anIZn 1 + an2Zn +
is sufficiently large
( say R
> max
+ ao.
[l , n max1::;j::;n1 [aj[])
and
lzl
R,
then
n1 + an2Zn2 + . . . + ao
IanIZ
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
=
( Theorem 3.4.5).
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.
169
but that
Jo(P)
0 for some
fj
(** )
J1 is
Jj(() d
_1 1
(
27ri 11(Pl=r J1(()
( ** )
(* )
j ;
;
the integrals
oo,
to a positive integer.
The reader should contrast the Hurwitz theorem just proved with the
realvariable situation. For instance, each of the functions
1/j, j
1, 2, ...,
x2 +
Jo has
fj ( x )
a double zero at
J0 ( x)
2. The
function
theorem fails.
However, if we "complexify" and consider the functions fj (z)
and the limit function
that
J1 ; Jo,
fj
z2
z2+1 /j
uniformly on
j ;
Jo,
oo,
the two
process.
<
lg(z)I 2
and that
g(O)
2. The
g(z)
lzl2.
function assumes an
170
interior maximum at
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.
zEU .
Theorem 5.4.2
( The
main. Let
that
P.
If
>
lf(P)I.
lzl <
C be a do
PEU such
point
f(U) is
f(U) with
1(1
A bounded
of
is a constant.
Corollary 5.4.3
domain.
Let
( Maximum
be a continuous function on
be a bounded
that is holomorphic on
U.
is closed
Proof. Since
lfl
U,
then it
Ill on
If
IJI occurs
at every point,
cannot occur at
PEU,
by the
U.
If there is a point
PEU
at which
IfI
be a holomorphic
is constant.
Ill on
V occurs at
Holomorphic functions { or, more precisely, their moduli ) can have inte
rior minima. The function
f(z)
z2
0 is a global
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
Let
( 1 ) lf(z)I 1
(2) f(O) 0.
for all
z,
1 72
4.1.1),
).
on D
then lg(z)I
I 0,
for
>
and small.
1.
).
E >
::; 1.
That completes
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.
1,
o:z.
1.
Again, the
1,
0.
Pick lemma.
>
D, but we
::; 1
::; = 11 11 : .
lf'(a)I
Moreover, if f(ai)
b1 and f(a2)
b2 , then
rPc(z)
Notice that
zc
=
_.
1

1 CZ
_
CZ
I <1
173
if and only if
lz  c l2 < 11 ez l2
if and only if
if and only if
lz l < 1.
It follows that
<Pc
Then
</Jc
D(O, 1)
to itself.
<Pc
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)I.
l 2
l
h =b1
Schwarz's lemma implies that
lb1
</Ja1
lh(z )I lz l. That
is,
f 0 <Pa1 (z )I lz l
or
Taking
z =a
gives
Writing this out, with the definition of the </J's, gives the desired conclusion.
174
IJ'(a)I
then the function
1  lbl2
1  lal2
5.5.l.
5.5.2
that
If either
or
1)
to itself.
5.5.1,
5.5.2.
6.2,
we
at that time.
Exercises
1. Let
2. Let
f, g
zeros at
D(P, r ) .
D(P, r ) .
Prove that
D(O,
1),
1
.
27ri
1)
is
1).
Assume that
1).
has
Let"( be the
i f'(z)
f(z)
"f
g(z)dz.
P(z)
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
5. Let
fj
D(O,
1)
+
1),
counting multiplicities.
Suppose
fj has at
that fj
f
+
uniformly on compact sets. Show by example that it does not follow that
has exactly
175
Exercises
roots. What simple hypothesis can you add that will guarantee that
have at least k roots? (Cf. Exercise 8.)
does
6. Let
azg(z)
8
f'(z)
f(z)
f:
8. Let
f 1
a (P r).
g
f a (P r) ,
f:
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).
2m fy
f'(() d(.
f(()
fy
realvariable
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#
f
f(P)
a:
P.
a:(z  P)
f(P)
 P)
f(z)
f
f
176
Exercises
l
_
27ri
The answer will be
or
1,
1 !'(() d(
f(()
{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.
[Hint:
f and
f has
)1
some
zeros
f(P) = 0.
of x and y at an
1
27ri
J'
ff
as already discussed. The reference [MIL] will help you to come to grips
with these ideas. Treat this as an openended exercise.]
10. Estimate the number of zeros of the given function in the given region
u.
( a)
(b)
(c)
(d)
(e)
(f)
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
>
U, and if
l
11(w)= _
27ri
(f'(() d(
j
laD(P,r) f(()  W
0 sufficiently small.
Derive from
l
u1 )'(w)= _
27ri
(f'(()
1
d(.
laD(P.r) (!(()  w)2
ul)'(w)=1
27ri
1
1
laD(P,r) f(() Q

) (
.
1

wQ
1 d(.
f(() Q
(*
Exercises
177
Q:
the point
f1 ( 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 nl ((  P)n
1 )!
8(
[!(() Q]n (=P
( )
f'
zeros inside
U?
Can you
If the zeros of
f'. ]
14. Let
 1
of
(n
f'?
has
13. Prove: If
Lagrange's formula.
This is called
set
about
1
__
inside
11
Pt(z)
be a polynomial in
Suppose that
f.
Pt(z)
t, 0
is continuous in
<
<
1.
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
sufficiently near
with
Pt(z)
0. What
15. Complete the following outline for an alternative argument to see that
the function
( a)
( b)
(c)
N(z)
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.
t
0, then f
[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
Pto
Prove that
( 1, 1).
toprovided that
Pto
the degree of
Pt
Pt
Chapter 6
Holomorphic Functions
as Geometric Mappings
Like Chapter
5,
>
conformal
(or
biholomorphic)
Such
mapping.
The fact that his supposed to be onetoone implies that h' is nowhere zero
on U [remember, by Theorem
point P E U , then h is
(k +
5.2.2,
>
5.2).
at a
As a
A conformal map
k ;:::: 0
f : V
>
C is holomorphic if
h1 is holomorphic on V.
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.
functions
from
C to C, but rather few of these turn out to be onetoone 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:
J(z)
az + b , z
C.
az
r;
if and only
a, b
C and
a =/: 0,
then
r>
(z
b) /a is
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:
lzl++oo
That is, given
lf(z)I
>
>
0,
IJ(z)I
satisfi.es
+oo.
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 holomorphic, it is continuous.
Therefore S
set is compact.
11({z : lzl
1/E})
is compact.
{z: lzl
11({z: lzl
lwl
>
Thus there is a
w is not an element
element of {z: lzl 1/E}.
0
result.
By the
C, then
C+ C
C} .
In other words,
11 :
is continuous at oo
with value
oo.
6.1.2
can be
181
and
Now, because
on the disc
0=/= Jg'(O)I
lim
lzl+O+
g(z)  g(O)
Z
A > 0 such
lim
lzl+O+
is onetoone
g(z)
I I
Z
that
sufficiently small.
original function
f.
lf(z)I
<
If
8 >0
such that if
lzl
<
8,
then
lf(z)I
1
Jg(l/z)i
> 1/A
and
1/8.
then
1,
that is,
f(z)
az + b for
a =/= O; thus
some
a, b
Theorem
f
E
is
<C.
6.1.1
lzl++oo
Jh(z)i
+oo.
{z : 0
<
lzl
<
77},
some
extends holomorphically
advance that
182
6. Holomorphic Functions
g'(O) :/= 0.
n
But
as
Geometric Mappings
such that
lg(z)I Alzln
for all
6.1.2)
that
lzl
with
lh(z)I
for
l zl
sufficiently large.
1
lzl
A n
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 selfmaps 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) ,
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.
f(O)
w with lwI
1 such that
f(z)
wz for all z
ED.
In other words, a conformal selfmap of the disc that fixes the origin must
b e a r ot ati on.
Proof. If
E C and
lwl
1,
f(z) = wz
z/w.
is a
i+
Let
11
5.5.1),
lf'(O)I 1
and
lg'(O)I 1.
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.
=
Lemma 6.2.2
1,
we define
<Pa(z)
Then each
Proof.
<Pa
For
EC,
lal
<
za
1az
_
.
5.5.2.
f:
lwl
1, lal
<
f is a
a, w with
f(z)
ED,
=
w <Pa(z) .
Proof. That functions f of the given form are conformal selfmaps 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 selfmap 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)
<Pb(wz)
wz+b
1+bwz
z+bw1
=

1+bw1z'
1 to see that bw1
w <Pa(z),
bW
bw. Thus
6. Holomorphic Functions
184
where a
as
Geometric Mappings
bw1.
Notice that our proof also shows that any conformal selfmap of the disc
has the form
f(z)
and lal
<
1.
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
selfmaps 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 selfmaps 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 selfmaps of an open set U C is called the
az+b
1t
 ,
CZ+d
a,b,c,d EC.
0,
then the
0,
has zero denominator and is nowhere defined. Thus only the case adbe i
az+b
1t
CZ+d
ad  be i
0,
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,
4.7)
185
<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
O;
<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.
6. Holomorphic Functions
186
as
Geometric Mappings
{ oo}
to itself.
by definition,
Theorem 6.3.4. If
formation, then
11
: CU
{ oo}
: CU
{ oo}
CU
{ oo} is
Also,
If g
then
To see
= 0, d # 0
is left as an
az+b
w CZ+d"

Solving for
z,
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.
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
Zt+
(Aa+Bc)z+ (Ab+Bd)
(aC+cD)z+ (bC +Dd)

{ oo}
6.3.
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
Theorem 6.3.5. A
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}.
ft
z i
<l
z +i
_
if and only if
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
(ii)
{ oo}
{ oo}
consisting of
(i)
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
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
i)
i.
(1i)/(1 + i)
zi
z+i
.
1, the
point 1 is sent
(li)/(1 +
14
oo
Thus the image under the Cay ley transform (a linear fractional
{ 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.
{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
__,
cally equivalent (i.e., homeomorphic) to the unit disc D since we are in fact
{z : 1
<
IzI
<
2}
since it
190
[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)
D.
l: U
;
V with l1 : V
;
U also
l1
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.]
11)
toone 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 6.6.3the analytic form
of the Riemann mapping theorem). The first, essentially topological, part
of the proof will be given separately in Chapter
11.
(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
1 +11.
z
s1
<
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 onetoone and onto) function
f: D> D with f(O)= 0. Then fis conformal (i.e., onetoone 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
(Exercise:
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.
D, f(P)
IJ(z)I 1 for
____,
0. Choose
>
at
0 such
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
derivatives.
( e.g.,
ff
the least upper bound in our special case from the previous sect.ion).
extremum;
con
converge normally
to a limit function
lfj(z)  fo(z)I
<
E.
193
F igure 6.1
Jo,
fo(z)
fj(z)
zj
not converge
uniformly
D.
Theorem 6.5.3
( Mantel's
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
Jo.
{!Jk}
that con
[O, 27T],
{ sin kx}k=I
are all
The proof of Mantel's theorem uses powerful machinery from real anal
ysis. The AscoliArzela theorem, one of the main tools, is treated in Appen
dix A. An alternative proof of Mantel's theorem, not m;ing the AscoliArzela
theorem, is given in the exercises.
194
U.
T/
>
z, w
Choose a slightly
EK with lz  wl
<
f E F,
>
If ()I r
c I.
J E F.
If
lz  wl
<
[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
<
.fo
C1lz wl.
Thus, for any
z, w EK
and any
f EF
we have that, if
lz  w l
<
TJ,
w l.
( this
concept is defined in Ap
{fj}
F has a subsequence
uniformly on K.
{!jJ
that converges
U ( the
I<j
[ In
case U
{z
EU
and
distance( z, <C
<C, letting Kj
D(O. j)
LJj Kj
{fj}
same sequence
be compact
\ U) 1 /j }
would do.]
D( 0, j)) .
F such that
{/;}
converges
Continuing
( m  l)th
6.5.
195
Normal Families
93
fjk3'
etc.
Then the sequence
{gJ}
is a subset of
since
LJ Kj
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.
K,
U C
K U
E K:
there is a constant M
MK
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:
(A)
1)
{z1}1
of holomorphic functions. If we
is bounded
( by
(B)
F
lz/JI ::;
Let
{z/j}1
M for all
on C
for such a
U.
In fact there is no
U.
and all
z EC
lz/JI
<
MK for all
6.
196
j and all z EK. (For instance, MK= sup{lzl : z EK} would do.)
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
>
that take
IJMP)I l!'P)I
for all J E F.
6.4
take values in the unit disc. Therefore Mantel's theorem applies and there
5.4.3)
modulus
1).
of modulus
1} i 0,
1.
Thus Jo(U) C
D(O, 1).
6.5.7)
4.5.
As noted
4.
197
U C is holomorphically simply
f : U > C, there is a holomor
function satisfying F' (z)
f(z).
their union
analytic form ) . If
U =f. C,
then
is a
is
We
to the consideration of some preliminary results that are needed for the
proof. These results also have considerable intrinsic interest.
nowhere zero on
U,
f : U > C
holomor
is holomorphic and
that
eh= J
on
J'(z)/ f(z)
such
U.
is holomorphic on
U,
because
is
nowhere zero on
function h:
eh(zo)
f(zo).
e" =
satisfies g'
g(z)
f(z)e
h(
:.)
g(z)
=
(z) :z (!(z)eh(z))
f
8 (z) (z) f(z) (
az
conclude that
on
U.
0 on
g(zo)
U.
For
1, we
ah
( z )e
az
h(:.) )
6. Holom orphic
198
eh(z) f(z)
( afj:az (z))
_
0
D
f(z)
for all
[g(z)]2
U.
g(z)
eh(z)/2.
{z: 1
for which these processes are not possible even for the function
{! : J
f
U and set
is holomorphic on U.
is onetoone,
J(P)
J: U
___,
D,
= O}.
(1)
Fis nonempty.
(2)
There is a function
Jo
E Fsuch that
IJMP)I
(3)
If
=sup
hEF
lh'(P)I.
(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 derivativemaximizing map must be shown to be oneto
one). Statement
conclusion of
(3)
(3)
lfJ'(P)I
>
lg'(P)I.
199
Proof of
in F.
If
U /= C, there
nonvanishing on U, and U
is a point Q tf.
U.
z
Q is
Since
h2 = .
function
such that
Notice that
f(z) =
2(h(z)
Since
lh(z)
 (b ) I
h is onetoone, so is
f.
for
z E U,
+b)
it follows that
Composing
maps
to D. Since
one and holomorphic with image in the disc. but also maps
f E F.
to 0. Thus
Proof of
(2).
fj E F
converge normally to
Jo,
with
Jo
is onetoone into D.
5.3.3),
s10n:
EU. Consider the holomorphic functions 9J (z) = fj (z) U\ { b } . Each fJ is onetoone: 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
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
=
200
Proof of
(3).
Let
(z)
Here we have composed
is onetoone
( see
Section
R.
Set
D such
g(z)  R .
1  g(z)R
6. 2).
4 C
such that
'l/J2
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
is onetoone.
at
P and
P.
Therefore
show that it is
at
We have
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
However
l/(P)I
g
11/;(P)I
and
>
It follows, since
lg'(P)I.
of statement
(3)
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)
(z)
T(z)
zR
1  Rz'
z1/J(P)
=
1 '!j,i(P)z
and
S(z) = z2.
Then, by our construction, with
/l1 o SoT1:
1oSoT1op
h p.
lg'(P)I
Since
h(O)
0,
and since
lh'(O)l l/(P)I.
lg'(P)I
<
l/(P)I,
lh'(O)I
1.
202
Exercises
Exercises
2. Prove that if
i)
C? [ Hint:
must be linear.
The
C.]
[ Hint:
oo.]
singularity at
3. If n
(z
{fJ}
Aut ( f!)
>
C.
5. Let n
C \ {O} .
6. Let n
C \ { z : lzl
[ Hint:
:S
1}. Determine
C, a i= O}.
{z
: 0 <
jzj
C. Then Aut( n)
<
=
1}.]
{az + b :
This group is sometimes called the "a plus b" group. A group is called
[g, h]
[g, h]
"a
>
ghg1h1 for g, h
l.]
8. Let U
{z
Im z >
O}. Calculate
of U.
D(O, 1).
Let K G be given by
K
Here
Po
{po
: 0 :S () < 271" }.
subgroup of G.
203
Exercises
10. Let n C be
{ </Jj}
be a sequence
3.]
___,
and
other?
P and Q be distinct
1 and 2 be conformal selfmaps of n. If 1(P)= 2(P)
2(Q), then prove that 1 = 2.
points of n. Let
1(Q)
and
for all z E
(a)
z for
>
(b)
O));
emanating from
(c )
\ ( { 0} U
{z:lzl=l};
(d)
zE
:U
>
(3 E IR,
D(P, r )
>
C is called a "branch of
D(P, r ) U,
e8f on D(P, r ) .
there is a branch
[Hint:
z8"
on U,
of log z on
{ z: I z I
1},
then /] must
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}.]
C1
curve, together with that boundary curve. [Along the way, impose any
topological hypotheses that you requirethis is a "heuristic exercise,"
204
Exercises
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.
21. Let a be a complex number of modulus less than one and let
L(z) =
z.
1 az
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
24. Let 0
lo l
f1 ( z)l dxdy
<
< oo,
Exercises 2630 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 4tuple is called the cross ratio of
26. If
a1, a2,a3,a4.
(Ta1,Ta.2,Ta3,Ta4)
(a1,a2,a3,a4).
eralized circle.
29. Using the result of Exercise 28, prove that an LFT carries circles and
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 <Pa is the inverse of <Pa Hence <Pa must be onetoone.
If
z = =1
l1 :z1 l=;I .
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 <Pa maps D into D. Hence
<Pa maps D onto D. That is the desired conclusion.
P(z)
35.
(a)
Prove: If
1.
(b)
Use part
(a)
P(z) =
+
o: has exactly
Chapter 7
Harmonic Functions
u +iv,
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]
207
7. Harmonic Functions
208
is harmonic if it is C2 on u and
u=O,
where the Laplacian u is defined by
( ::2 ::2)
+
u.
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.)
u + iv: V+ C
is holomorphic.
av
au
ax
ay
and
av
ay
: ( ) :x ( ) .
y
6u
( ::2 ::2)
+
u= 0
209
is holomorphic, then H
u +iv
i(v
 v)
(u +iv)  (u +iv)
is a
holomorphic function with zero real part; hence its image is not open. Thus
u.
u +iv
Corollary 7.1.3. If
U C, then
C00
is holomorphic is called a
+
harmonic conjugate of
C00
that ReFv
ul v
Since Fv is
C00,
then so is
+
uI v
is
C such
D
u.
u +iv
!'
is holomorphic, then
au
ax
i av
ax
au  au
i .
ax
ay
u.
The
: U
+
u +iv
: U
+
C2 (indeed
C00)
C is holomorphic.
=au iau
because
6u
0. Also
ax
C1
ay
since
.
u
is
C2
7. Harmonic Functions
210
function F : U+
ou
'
F =
ox
+i
ov
ox
ou
ox
ou
oy ,
F' =H =
ou/oy.
Thus
5.4.
u =log IF!).
constant.
<C
211
maxu.
8U
8U,
minu.
8U
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.]
>
0. Then
u(P)
Proof. Choose s
>
u(P)
D(P, s )
+iv(P)
1211"
u(P
with
U for some
+ rei0)d0.
such that
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
u(P)
27r lo
as desired.
If u1 :
D(O, 1)
l zl
1},
then u1
and if u1 = u2 on
principle (Corollary
7.2.3)
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)
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) .
<Pa(z)
za
_1 az
(1) <Pa
then u
6.2.
1) ;
U is holomorphic,
213
b.(uoH)
since
u is
harmonic. [Note here that we have used the chain rule in complex
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.
Let u
+
JR be a
D ( O, 1 ) ,

u(a)  211"
271"
u(ei
!f;
J aJ 2
.!f; dv1.
12
I a  ei
Ja l 2
211" Ja  ei!f;J2
1
is called the
the formula
1 
says that
.
u(rei0) = 211"
271"
1 r2
u(ei'lf;)
l2rcos(O'lj;)+r2
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
<Pa
u(a)
This y ields
u o <Pa(O) =
1211"u(<Pa(ei9))d0.
271" 0
We wish to express this integral in terms ofu directly, by a change of variable.

(*)
_1 _
27ri
_1
27ri
{ 211"u(<Pa(ei9)) _ iei9 d (}
ei9
lo
u(<Pa(())
J
d(.
!aD{0,1)
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

1),
then
F(z)
is holomorphic for
D(O,
0 on D(O,
F.
F;
1).
11"
{2
2m lo
J(()
d(
1/(
on 8D(O,
1),
then
1).
215
on
D.
on
8D(O, 1)
this connection precisely. The theorem is usually called "the solution of the
Dirichlet problem for the disc."
Theorem
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
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
g(
z
since
be
Define
f(z)
Then
482 /8zaz
2
ei
{ 7r f(eifJ) . fJ d(}
ei8  z
lo
ei8 /(ei8z)
D(O, 1)
and
is holomorphic in
2
{ 7r f(eifJ) g
z
lo
z.
( ei.8eifJ z ) d(}
0,
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
monic.
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,
so that
when
1/J
reiO
 1/J)
equals 0 if
is close to the
u(rei0)
converges to
f(eiO)
as
eiO.
1.
t
intuition precise:
harmonic on a neighborhood of
1
Now if Po
Figure
7 .1),
eiOo
8D
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
( )
v z
D(O, 1),
(* ) .
maxaD(O,l)
lfl.
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
d'ljJ
__!__27r 1{1/J:l1/J9ol<8} j f(ei9o)f(ei1/J) I ll  rei(91/J
)l2
1 r2
d'ljJ
__!__271" 1{1/J:l1/J9ol2'.8} j f(ei9o)f(ei1/J) I llrei(9
1/J)l2
1
<
lu ( P )  u ( z ) I
r2
I +II.
1
E
1 r2

.
d'ljJ
27r {1/J:l1/J9old} 2 I 1 rei(01/J) 12
271"
E 1
1 r2
d'ljJ
2 271"
llrei(91/J)l2
< <
=
1
1
o
by ( * ).
To estimate II, note first that for 0 :S lal :S 7r, 1cos a a2 /20. This
is elementary (see Exercise 80). Then by our choice of 8 and r we have
ll rei(01/J)l2
>
>
Thus
1 r2
1
d'ljJ.
40M
27r {1/J:l1/J9ol2'.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 )
( ) I < E.
 u z
218
7. Harmonic Functions
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
( ) 
f.p
>
f.p) U and
for every 0
PE
U,
21r
2_ f h(P + Eei0)d0.
27r lo
f.p
an
open set U
r) is a closed disc in
21r
2_ f hj(P+rei0) d(}
27r lo
U, then
21r
2_ f h(P +reiO) d(}
27r lo
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
t
g(Po)
supQEV
then
g(Q),
Po
E V such that
is constant on V.
supQEV
because
Po
g(Q) and
define M
{z
EV:
g(z)
7.2.1, we set
}.
Then M is nonempty
g is
continuous. We shall
E M. Choose
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.ei8
D(P, f.p)
V. Thus
= s
as required.
Theorem
function uv on V defined by
uV (z)
{ h(z)(z)
uv
if
if
=
z
z
uv : V
t
EV
E av
uv on V as well as on
av.
av.
=
uv
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,
220
7.
Harmonic Functions
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.
They are
worth mastering. They will arise again later when we consider subharmonic
functions.
Lemma 7.5.1.
+
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.
{x E JR
: a<
<
b}.
axis )
To see that
v is continuous on W
v is harmonic on W, we
=
P EU,
in a neighborhood of
v(z)
is harmonic
( exercise ) .
x < b}.
P EU we note that
=
+
v(z)
P. It
a<
P E {x E JR :
a<
x < b}.
27r
U U { x E JR:
221
Principle
x=a
Figure
because v(x +
iy)
 v ( x  iy).
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(7rO))
d(}
d(}
d(} 
d(}
0
[since
as
claimed.
Thus
on
fJ
See Figure
v.
7.2.
on U and
v on fJ
f(x)
x2
14.5).
7.5.1
7.
222
Harmonic Functions
Theorem 7.5.2
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.
=
+
F ( z ) on
fJ
First
E fJ,
{x E IR:
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}.
a<
x<
b.
D(x, E )
U U fJ U
Set
v (z)
ImF ( z ) '
E D(x, E )
nu.
7.5.
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 realvalued 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 Schwarzreflect 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;
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.
be a nonnegative, har
D(O, R),
Rlzl
R lzl .
u(O).
u(O) ::; u(z) ::; +
Rlzl
R + lzl .
Proof. Recall the Poisson integral formula for
u(z)
Now
1
 27r
(* )
and
(** )
on
D(O, R) :
R2lzl2
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()
Rlzl 27r lo
R+ lzl
. u(O).
Rlzl
D(P, R),
R  lzPl
R
Pl .
. u(P) < u(z) < + lz
u(P).
Rlz  Pl
R+ lzPl
225
T
oo uniformly
that Uj Tu
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.
>
>
Too,
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).
j, k T+oo, j
s+s/2
s  s/2
>
k,
(uj  uk)(Q)
3(uj(Q)  uk(Q))
T
0.
/ +oo
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)
U1
<
+oo}
+oo}
and
U2
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
Vj D(P, r)
D(P, r).
+
with
vj(P)
{e(uik+ivik)}
0 and with
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
+
5.3.3).
+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)
<
z EU or lim Uj(z)
z EU. In
either situation, one then gets the uniformity of the convergence on compact
Kn D ( P, r) is a compact subset of
on Kn
D ( P, r)
=/= C.
is harmonic on U?
If
on U such that
exists, is it unique?
Dirichlet problem
l au
many motivations from physics (see [COH], [RES]). For instance, suppose
that a flat, thin film of heatconducting material is in thermal equilibrium.
That is, the temperature at each point of the film is constant with passing
227
au
au
+
R, then
u2
so that
u1  u2
u1  u2
u1
and
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.
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
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}.
IzI = 1
z o.
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).
a2 I ax2 + a2 I ay2,
O
hence
1a
a
1 a2
+
r 2 ae2 .
u is independent of (),
)
r
(
ar ar
au )
ar (r ar ;
:r (r :r u)
r#
=
1a
when
so that
au
ar
228
C logr+ D
to agree
some
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.
JR
[a, b]
JR and h
is a realvalued harmonic function with f(a) ::; h(a) and f(b) ::; h(b), then
f(x) ::; h(x) for all x
as
[a, b]
1, then
f((l  >.)c + >.d) ::;
(1  >.)f(c) + >.f(d).]
JR2) is
229
h(x)
F igure 7.3
h defined
f ::::; h on oD(P, r), it holds
subharmonic on U.
harmonic function
on
that f ::::;
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
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,
D(P, r) U,
f(P ) ::::; 
271"
Then f is subharmonic.
127!"
0
f(P +
rei6 ) d0.
7.
230
Harmonic Functions
Figure
7.4
f is not subharmonic .
Then
h on 8D(Q, s ) but
for some
zo
The function
g(zo) > 0. If
,
M}, then it follows that K is a compact subset of D(Q, s ) . In particular,
but
7.4)
s .
TJ,
<
g(w).
231
(since
{
27r f(w + T/ei8)d0 < f(w),
lo
1
27r
P : D(Q, s) x
P > 0. Let> 0.
Recall that
h(z) =
Then
sei8) + ]d0
continuous on
follows that
Letting
+
for
f(Q)
h(Q)
27r
F: U+ C be holomorphic.
s) U. Then, by the mean
IFI is subharmonic.
Then
7.2.5 ) ,
IF(Q)I
By the proposition,
IFI
sei8)d0
F(z)
f:
IFI will
zk, k EN.
f2
o J)(P)
of
D(P, r)
(f(P ))
<
<
Let
R + R be
is subharmonic if
27r
U + R be subharmonic.
is subharmonic.
is everywhere nonnegative.
U, then
27r
27r
el
is
7. Harmonic Functions
232
Here the second inequality follows from the convexity of <P (this is known as
Jensen's inequalitysee 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.
=
along with the submean 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 "onesided" 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.
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
233
max{f1(P), h(P)}
<
max
<
fi(P +
rei8)d()
max
271'
Definition 7.7.8.
b
t
(i)
(ii)
(iii)
(iv)
IR. a
Let U
barrier for U
is continuous;
is subharmonic on U;
We call a function
b :S O;
{zEau: b(z) = O} = {P}.
singles out
b(z) = x 1
is a barrier for U at
P,
=Re z.
PE U be the
r = IPI = IPOI and notice that
7. 5. Let Oo = arg P, 0 :::; Oo < 271'. Then the function
C be
where
r).
See Figure
z
is a barrier for U at
1+
Re(e
i Oo z) r
P.
C be
QEC\ U
zP
zQ

234
7. Harmonic Functions
Figure 7.5
U into C.
Notice that
Po
'If; maps
U onetoone into
the closed
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.
>
P E au.
0, there is a barrier at
E
max(E,b1(z))
P for U if E
>
if
if
EU\ D(P,r)
EUnD(P,r)
P is local, in
U looks in a neighborhood of P. This
235
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 solvednamely 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)
on D(O, l);
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,
8D(0,1)
B(ei9z) B(z) for all(); that is, Bis rotationally invariant. We shall now
show that that is impossible.
=
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 BHr
from which it follows that, for each z E {z: r lzl 1},
=
B(z) Hr(z).
Thus for each fixed z
0,
we see that
fixed,
B(r) + 1
1og I z I
r+O+ log r
1.im
7. Harmonic Fu nc tio ns
236
E D(O 1) \ {O}.
,
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 (18801975).
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
always be solved on
there is a function u
U.
That is, if
U and
lim sup'lj;(z):::;
U3z+P
f(P), VP Eau}.
E U,
u(z) =sup 'l/J(z).
1/;ES
The
237
letting
Let
'I/; E S, E
>
0, and let E{
{z
nonempty.
We claim that
E{
'I/; assumes a
U,
it is
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{.
possibilities:
(a)
E D(P, r).
By definition of
'11n( Z )
max { 'l/;1 ( Z ) , .. .
, 'l/Jn( Z )}
for each z
Section
cl>n(z)
_

Then
{ '11n
(z)
cl>n
is subharmonic
if
E U \ D(P, r)
E D(P, r) .
and of course
of
'11nlaD(P r)
,
on U (by the
Poisson integral of
if
in
principle. F inally,
238
7. Harmonic Functions
A n(z)
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,
=
Part
lo.  /31
g(z)
J(w) + E +
bw (z)
(M  J(w)),
E 8U n
239
D(w, 8)
] +
f(().
+ > f(().
g is
in U. Thus u
g0
lim supu(z)
U3z+w
g(w)
1).
g) 0
for
on U. As a result,
f(w) + .
g(z)
f(w)
g ES;
(z)
bw
(M
then u(z)
g(w)
+ f(w)).
g(z)
J(w)
for z EU and
J(w) 
and, since
lim supu(z)
U3z+w
f(w) +
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
or U except that
+oo).
w.
in a neighborhood of
w.
w,
w.
w strictly
for U is a "local
240
7. Harmonic Functions
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 1t 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
R2.
>: 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.
of Annuli
241
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 welldefined 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
onetoone.
D
t
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
or
R2
l.
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.
=
Exercises
243
Exercises
{z
C: 1
<
IzI
<
2}.
Define
C and a har
monic function u on U such that u does have a welldefined harmonic
conjugate on U.
4. Let
( a) * ( *w )
(b) If f is
(c )
*W
w.
Exercises
244
h(z)  A In I zI
{z : 0 < I zI
set
[Hint:
0}
<
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
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
<
is constant independent of
* 8.
<
1, Im z >
<
0.]
r. [Hint:
Expand
in a
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 =
lf0(z)I
::;
< oo
{!0}
for all
is a family
U, then
[Hint:
ef"' +ig"
12.
O.]
245
Exercises
13. If
u is
14. If
u is holomorphic or u is holomorphic.
1 2.)
1/u is
E Z give a function
the closure of
D such
that
16. Prove: If
u is
u are 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
be harmonic and
( a)
(b)
u : C + IR
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
v(z)
realvalued harmonic
'
of the Laplacian.
25. Compute a formula analogous to the Poisson integral formula, for the
region U
= {z
Im
>
0}
246
Exercises
{z : lzl
1, Im z
<
>
O}
by
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
=
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
28. If
D(O, 1)
U C and if
28 by using
[Hint:
{Uj}
converges to
u0,
eu1+ivj
F,
then limj Uj
ln
IFll
30. Let
is a point of K and if
C>0
u( a) . Prove
K it holds that
32. If
not depend
on
u.)
is the
h(z)
33. Let
1
271'
12 11" eei8
0
e1
+

z i
u(e 8) dO.
z
Suppose that
on all of U.
u is harmonic on
U\
{P}.
Prove that
u is in
U.
fact harmonic
247
Exercises
34. Let
f: [O, 1]
u(z)
Prove that
is harmonic on C.
z/lzl2
directly a
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 selfmaps of an annulus A=
38. Let
h(z)
lzl
<
<
R}.
{z: r
lh(z)I
when
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
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
6.f
P and see
To show that
at
0 at
is subharmonic, then
P,
coefficients must satisfy in order for Lemma 7.4.4 to hold for all small
circles around
42. Let
fj
P.]
on U. Is
{!j}
neces
sarily subharmonic?
43. Formulate and prove a maximum principle for subharmonic functions
(see Proposition 7.7.7).
44. If
satisfies a
minimum principle
as
C1
lies only on one side of the curve at each point, then prove that every
point of {)U has a barrier.
[Hint:
2 8
4
Exercises
Notice that
46. Let
be a realvalued
harmonic
+
is subharmonic.
l: U
lo F
that
48. Let
+
+
R be subharmonic.
U is holomorphic. Prove
l: U
+
F : D(O, 1)
+
R be a function.
U it holds that
lo F
is
subharmonic.
50. If
l is
l is
on
D(O, 1)
such that
12
is subhar
not.
lllP is
that lllP
l is
point
f C,
such that
no boundary
Ill
is continuous on u and
Ill
l on U such
[Hint: Use
restricted to au equals.
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
(b) 0
(c )
001!",
lx
(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
<
U1
55. Let
assume that 0
<
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
l zl
<
<
2}
such that
IF(z)I
+
whenever
F.
IFI
f+
z//2
/2/ z.
Refer to
Exercise 59 . ]
: 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 ?
have?
60. Use the Poisson integral formula to show that a harmonic function must
be
C00,
D.
and holomorphic
imaginary constant on
FG
is an
D.
is nonvanishing,
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
E C (C)
is
C00
J F /':,. dx
and harmonic.
lemma.]
65. TRUE or FALSE:
If f
is subharmonic on a domain
U,
then so is
lfl.
250
Exercises
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.
*
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.]
*
n(ei9)
einO, 0 ()
<Pn(reiO)
271", n
<
Z, then
rlnl einO
00
P(rei9)
n=oo
<Pn(rei9).
P(rei9,ei11i)
(1/27r)P(rei(01/I)) .
00
<P(reiO)
n=oo
*
69. Let
(a)
f: U
an rlnl einO.
251
Exercises
{b)
If
70.
2'.:
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.
6.f
>
O? Does
one definition imply the other? Can you think of a definition that applies
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)
curve.
=log lz
 Pl.
be the
u(z)
function
the "Green's function" for O; the usual notation for the function
( d)
{e)
that we call
The integral
is well
252
Exercises
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
'YjI 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 finitevalued 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
77.
Prove that {z E
equivalent to {z
<C:
E
<C
253
Exercises
78.
79.
80.
2:
2:
Chapter 8
8.1.
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
f(z)
Ll fj(z)
j=
SN'S (each of

255
256
I: fj(z)
j=l
is said to be
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.
SN(z).]
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},
where
aj E C
The symbol
TI
( * ) to converge.
partial products PN of ( * ) to be
PN
IT (1 +aj)
j=l
(1 +ai) (1 +a2)
..
(1 +aN ) .
IT(l+aj)
j=l
converges if the sequence of partial products
{ PN} converges.
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;
N
lim
N++oo
II (1+aj)
j=No+l
[fi ]
j=l
(1+aj)
oo
ft (1+
aj).
No+l
Remark: If
II (1+aj)
0.
j=No+l
pendent of the choice of No, subject to the definition (exercise for the
If
1 oo 1
I:< I: 2j l
j=2 J
j=2
00
1+x I: '
j=O J.
ex,
i.
(00 )
we have
1+x+x I: :r1
j=2 J.
1+2x.
258
Corollary 8.1.3. If
a3 E C, la3I < 1,
PN for
00
satisfies
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.
limit.
Corollary 8.1.5. If
00
IT(l +la3I)
j=l
converges, then
converges.
259
Proof. By Corollary
8.1.3,
PN
exp
(l;I)
{ PN}
sequence
u:.:
la3I}.
I: O'.j I:
j=M
j=M
1a31.
The analogous assertion for products is somewhat less obvious and requires
the following technical lemma.
PN
a3
EC. Set
II (1 +
j=l
a3) ,
PN
II (1 +
la31).
j=l
M
1 + II (1 +
j=N+l
a3)
M
1 + II (1 +
j=N+l
la31).
PN
1 +monomial
a3
's
PN
After subtracting
1 +absolute
1
statement that the absolute value of a sum does not exceed the sum of the
absolute values.
260
>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
Corollary
261
8.1.8. If
then
00
converges.
Remark:
Corollary
products.
8.1.8
ternative form: If
00
L:llajl<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
Suppo se that
fj
: U
+
C are
holomorphic an d that
00
L:l!JI
j=l
(8.1.9.1)
FN(z)
II (1 + fj(z))
j=l
(8.1.9.2)
U if an d on ly if
fj (zo)
1 for
Remark:
at zo.
verges uniformly on
(a)
fj
a set E if
IJ(l + fj(z))
con
262
and
(b)
the sequence
fj(z)).
sets.
Lj lfjI
00
II (1 + l!JI)
j=l
are uniformly bounded on K by ec.
Let 0 < <
L lfj(z)I <
z EK.
for all
j=N
IFM(z)  FN(z)I
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) .
Suppose that
F(zo)
zo. By
Jo such that
0 for some
lfj(z)I 1
exp
lim
N++oo
is nonvanishing at
zo.
II (1 + fj(z))
io+l
263
zo.
Now
oo
jo+l
nri( l + fj(z))
p(z)
of
p(z)=c
IT (za1)
j =l
Among other things, such a factorization facilitates the study of the zeros
of
p. In
on all of
C ( called
an
( of
first order ) .
entire function)
(za1) The
Weierstrass elementary factors.
Eo(z)=1z
and for
1p
Z we let
( :
Ep
Ep
is close to
+ ..+
is holomorphic on all of
if
I zI
C.
;)
( :
z+
+ ...+
;)
(1z)exp z+
might b e expected t o b e close t o
fo r
+ ..+ ;)
2
small ( and
large ) .
264
ll  Ep(z)I lzlP+1.
Proof. The case p
Ep(z)
We claim that
b1
b2
+L
n=l
bp
bnzn.
bn 0
0 and
for
>
p. Assume the
Ep(l)
+L
n=p+l
bn
so that
00
I: lbnl
n=p+l
i.
lzl 1,
00
lzlp+l
n=p+l
lzlp+l
<
bnznp1
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
+ )
z'
Gn
<>nZn
(I+ )
sion equals
00
zP
(anp)zn.
n=p+l
+L
E(z)
L nbnznl,
n=l
Ep(O)
1,
265
{ aj }f=1
with no accumulation point in the complex plane (note, however, that the
aj 's
for every
>
{Pj}
C: r
+l
<
(8.2.2.1)
oo
"
{ aj }, counted
with multiplicity.
l
P
n
Proof. For a
Thus
IE n (: ) l l : J
ILI
11E,. ( : J 1
1
<
oo
{ an}=
l
{an}=
(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.
266
Pn
1. Thus
Zm n=IT+l En1 ( a: )
m
Theorem 8.2.4
g such that
f(z)
zm
el>l
h(z)
f,
Since
holomorphic logarithm
Zm n=l
IT En1 (: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.
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
Theorem 8.3.1
( Weierstrass ) .
Let U
in
{aj}
{aj}
is infinite.
on U
(counting multiplicities).
{aj}
p(z)
f1j(z  aj)
So we may assume
267
1/(z
 p)
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.
(1) u cc;
1'
(2) C\ U is compact;
(3) {aj}1 U {oo} U;
(4) {oo} n {aj} 0.
=
{ai}
ai EC\ U;
(2)
aj has a(not
lai ail, then
above, each
if we set
dj
as
aj 's.
j + 00.
lz wl
In particular,
all
8 > 0,
lz ail
8 for all
z E K,
z EK
w EC\ U.
all
dJ < lz aI
J
2
In other words,
aj
I ajz aj
 I 
<
or a11
z E K, J.
.
> Jo
j and
replaced
Thus
f(z)
 j
)
IT Ej ( ajz aj
j=l
oo.]
[Exercise:
is
Check what
268
F igure
Lemma
8.3.2.
A
infinite set
8.1
!
U such that
of A.
Remark:
Proof of Lemma
8.3.2.
wEC\U
inf
/z
 wl
Qj
of Chapter 4.
d(z)
Ai
68
{z E
U: 4i+I
<
1, 2, . .. ,
let
1
d(z) 4i }.
1
1
1
:i
, :i )
,
)
,
,
;
i
i
),
)
( , (
(
(
1
,
8.1).
For each fixed j, the collection of such boxes is pairwise disjoint. The
Qj
{%}1
Qj
and write
8.3.
26 9
the center of lies in U . This follows from the fact that some point of the
to any point of
is <
that point.
Let
PJ
%,
be the center of
paragraph. Let
D1
PJ.
Clearly
Dj
is countable and
00
For property
(1),
note that if
a, a' EA
a,a' E
LJ Dj.
j?_jo
'
So d(a) < 1/411 and d(a ) < 1/4jol.
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.
of
<
D.
This point of
D .is
within
z.
So the center of
<
2E
D
open U C
1
p is said to be
D(P, r)
f is holomorphic on U
JID(P,r)nU
I
C and P
Eau, then
such that
10.]
270
f on
{ aj}
(2)
on U to
J, g
m( z )
on U such that
J( z)
.
g ( z)
By the
E U and r > 0 is
D( z, r ) is essentially a
f/g
on
J has
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 "polepushing"
which is useful in many other contexts as well.
A( z )
a, (3E C.
L aj ( z  a)j
j=M
Denne
8.3.
number
In  .Bl Let f.
r >
271
>
0.
Then there is
finite
B(z)
L bj(z  ,B)i,
j=N
f.
for all
z EC\ D(,B, r ) .
(z  a)1
(z
 ,B)
1 .
 (a  ,B)(z  ,B)1
1
00
j=O
(3.
1
si(z)
(=p(j)
a (z  aj)e .
272
Sj
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
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
D(a,
z E
D(a,
r .
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 MittagLeffler 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
numbers
eo, ... , ep
273
1
j=p1
b_j(z  o:)i
such that
g(z)
v(z)
eo + e1(z  o:) +
+ ep(z  o:)P
(t)
=
0. Then
+l
+2
+ ...
+ Cp+2zP
Cp+1zP
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)
eo,
Cp+2bp+l + Cp+lbp
Cp+3bp+l + Cp+2bp + Cp+lbp1
C2p+lbp+l + C2pbp +
+ Cp+lbl
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)
on U, holo
coefficient of m at O:j is
h(z)
Sj(z)
274
and let
N(j)+M(j)
Sj(z)=
a(zo:j)f
L
f=O
Sj
about
O:j
Again, by
O:j,
at
functions
1
Vj(Z)
L.::
f=N(j)M(j)1
such that
g(z)
vj(z)
(t)
k(z) on U such
But then g(z) k(z) will have
no poles and by (+) will have N(j) + M (j) degree Taylor polynomial at
Sj (z),
equal to
O:j
g(z) k(z)
h(z)
D
Exercises
1. Let
{an}
C. Assume that no
an
1.
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
>
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,
R2n z2n
R2n
oo
+ )
(
n=O
II
4. If
bn 1
>
for all
n,
_!!:___
Rz
n
converges if and only if
L log
n bn
]1(1+(l)n)
5. Determine whether
< oo.
4.)
7. Calculate
explicitly.
8. Prove that if
Ej laj I
< oo
and if
integers, then
00
II
00
(1+
j=l
aj)
=II
j=l
(1+
au(j))
E laj I
< oo
E aj <
oo.
10. Let f be entire and have a firstorder zero at each of the nonpositive
integers. Prove that
/(z)
z. e()
J1 [ (1+}) ]
e/i
276
Exercises
II
f(z)=eg(z).z
.
(J,k)#(O,O)
{(
1
exp
[ja : k(3+ (ja : k(3) ] }
ja : k(3 )
2
00
1
1
71"2 n oo (z  n)2.
(sin 7rZ
[Hint:
a, (3
)2
bounded.]
z does
00
converge?
L Inn  f3nl
< 00.
for which
II Z  O!n
00
n=l Z  f3n
converges normally.
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  n2z2].
n=l
( * ) in Exercise
15
1
2. 4. 6 .. 2k .
{if =
v 2 k11! 1. 3. 5 ... (2k  1) J2k+ i"
8.3.6). [Hint:
15
to sum L
8.3.1) by
f' / f.]
Consider
1/j2
and L
1/j4
using MittagLeffier's
Exercises
277
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:
1.
so that
21. Let
tC
f
{ 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.
f(z)
IT
j=l
bj
.
z a
J
z
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.''
24.
(a)
=a bounded domain in
8U,
1/ ( z  P)
tC
by "pushing
8U.
(Namely,
Q,
U.)
Use finite partial sums to make things converge when you take an
infinite sum of "principal parts" at interior points.
278
Exercises
Chapter 9
Applications of Infinite
Sums and Products
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 farreaching
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
the details.
Suppose that g is a nowhere vanishing holomorphic function on a neigh
borhood of
D(O, 1).
Then log
191
279
280
1 1271"
gives that
log lg(O)I
27r
log lg(ei0)ld0.
at the point
on
8D
in
is called the
If
[The
D(O,
1),
Ba(z)
= 1z aza
_.
1).
6.2.]
Proof. If
lzl
D(O, 1/lal)
;;2
1
Ba
lzl=1
Izzi = 1
z
= =1.
ll  l l ;z l 1;=;1
Finally, if
1).
lzl 1,
=
then
IBa(z)I
Theorem 9.1.2
( Jensen's
and
formula ) . Let
D(O, r),
that
is holomorphic on
f(O) =/= 0.
be holomorphic on a neighbor
Let
a1, ..., ak
be the zeros of f
I :j I 217r Jor27r
k
log lf(O)I +
log
I Uj I
<
for all
is holomorphic on a neighborhood of
order
at
Uj
log lf(rei0)ld0.
the function
g(z)
j.
j/
1, ...
=1 lzl= r.
= Tik=l Bfa(z)r(z/r)
, k the
Therefore
281
D(O, r)
is holomorphic on a neighborhood of
(* )
Thus
log lf(O)I
1
""" log
IBa J ;r(O)I =
271"
j=l
1211"
0
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,
15).
as in the theorem,
Proof.
negative.
Theorem 9.1.4. If
D(O, 1)
and
is a nonconstant
bounded
holomorphic function
on
plicities), then
00
L(l  laJ I)
j=l
< oo.
<
I I
282
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;
log ( 1 (1
log
a E
))
(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.
00
2:( 1 la j l)
< oo
j=l
and no
aj
0,
aj 's,
D(O, 1)
counted according to
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.
aj 's,
counted according to
283
11+ 1::1Ba;(z)I
<
<
z D (O, r),
+ajzlajJ2
1I + Iaa13 I Baj(z) I IJajllajJajz
IajI(1  Zllj) 1
(Jaji+zaj)(lJaji)
zaj) I
I(1 lajl(l
+r)(lJaji)
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 ( ) (1IajI).
By the Weierstrass Mtest, the convergence of Lj(lJajJ) 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.
<
2:
< oo
J=l
0,
Remark: A
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
< oo,
the theorem
J=l
(9.1.6.1)
284
lf (z)I
sup
zED(O,l)
In other words,
sup
zED(O,l)
1), Fis
IF (z)I.
function.
Proof. We
define F
by
F (z)
J(z)
=
zm
[f11
(aj/lajl) Ba1(z)
1),
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.
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

285
( to
be defined
( or
8.3).
"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).
Ln z2n.
( cf.
Exercise
3).
since
1)
when
Chapter
D ( O,
21,
for
r E (0, 1)
and
function
F(z)
on
for
w such that w2
Nl
F'(rw)
L 2nr2n1W2n1 + L 2nr2n1W2n1
n=l
oo
n=N
N1
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
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.
=
286
Theorem 9.2.1
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
for
f;
wz, lwl
to
=
=
an is regular
form, with coefficients having the same absolute values as before; so the new
Fln(o,1 )nD(l ,)
Choose an integer k
>
t/J(l)
l; and if
lt/J(z)I
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.
is a disc
Define
G(z)
F(t/J(z)),
n(O, 1+6).
n=O
zk+1 )/2
( Fo t/J)(z)
L aj
J
on
n(O, 1):
1
zk+ zk+1
2
2
1
) Pi
t/J(z)
(zk+
jth
zk Pi
while the
(j+ l)st
287
z(k+l)Pi
to
ranging from
(9.2.1.3)
ranging from
to
(9.2.1.4)
(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
in a neighborhood of
1,
1.
This
f(z)
1/(1
z).
Then all P E 8D
\ {1}
9.2.1
F(z)
L: z2i
j=O
3/2,
8.3
we showed by dif
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.
a, r
>
0 such that
for
lzl
>
r.
z.
We let A
order of
{ aj}
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
>
D(O, r)
n(r)
be the number
Lemma 9.3.1. If
f is an entire
r < oo. Then
289
function with
( log2 ) n(r)
0 =log lf(O)I =
where
n 2r )
log
k=l
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
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).
f(O)
log2 logM(2r).
D
f(O)
1.
n=l
( log 2) n(r)
Since
M(r)
for any small f > 0 and
oo. For
<
r>.+ / 2
e
n(r)r(ME)
as
log M(2r).
1
)
r(ME
log2
. (2r)ME/2
n(r)
r.x+E.
290
{a1, ...,aj}, we
have
j
for
large and
'5
>
n( lai l
+ '5)
( Jai l
+
+ '5)A E
we find that
'5
0,
laj1(A+l) j(A+l)/(A+E)
Selecting E =
1/2,
converges.
In Theorem
8.2.2
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
8.2,
lanlAl
replaced by
JanlAf,
to
n
[Here [] is the greatest integer function.] That is, the factorization of The
orem
8.2.2
Pn
[A]
"greatest integer
factorization
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.
A with f (0) =
Suppose that
1.
multiplicities).
Proof. Let
be fixed and
>
r2/2.
lie in
D(O, r)
and
9.3.
291
Hence
laklp+llr2 llkZ1pl
rP+l
( ) p l ( ) p l
(9.3.3.1)
<
n(r)
Eq.
(logM(2r))rpl
log
(e(2r)>+) rpl
2A+ErA+Epl.
(9.3.3.1) gives
2 p+l 2.x+p+E+l
+
1
1
1
a
r.X+Ep1.
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.
()
7r
as
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)p2 log IJ(rei9)1 d(} I
I 2 h27r(2rei9(rei9  z)p2)(log lf(rei9)1 logM(r)) d(} I
=
292
1
271"
By Jensen's inequality,
1 { 211"
log lf(rei8)1d()2: log lf(O)I
271" l
o
0.
(9.3.4. 2)
.X,
With
(p
2p+3r.X+{pl.
as
r+
oo.
[ ]
then
dP f'(z)
dzP f(z)
Proof.
Let r
>
I rr(zajaJz)
L,,
 '"'log
( )
1
p
 ! L (aj  z)P+l
J
(9 3. 5 .1)
n r
log lf(z)I
J=l
 z
By logarithmic differentiation in z,
z)
f(z)
( )
!'(
n r
( )
n r
+_!__
( d ) p [
'( )]
_
dz
f(z)
( )
n r
_
=
p.I''
L,, ( aJ
J=l
z )p1
l
+p.I '"'L,, aJp+ (r2
( )
n r
J=l
aJz)p1
293
as
r +
r +
+oo.
+oo. The
D
result follows.
canonical factorization is order .X. But the result is easy because of the work
we have already done in proving Proposition 9.3.5.
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
dP
dzP
[Pfv(z)] N
a
PN(z)
p.
1, and if
(a:)
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
then
+
p.
>
Since
.X  1 , since it is the
PN P
+
normally,
Pfv P'
1,
f(z)
has the property that
eg(z) P(z)
Proof. Differentiate
f(z)
The result is
Let
>
.X
f'(z)
f(z)
 1.
eg(z)
(z) .
P
P'
P
(z) +
(z)
(z)
p!
L(
dP+l
(z) p
 z)P+l = dzP+l g
 !L
 z)P+l.
(9.3.7.1)
294
It follows that
and hence
dP+l
g(z) =
dzP+l
( Lemma 9.3.4)
(9.3.7.1),
(9.3.5.1)
f(O)
f
9.3.7
9.3.7
in Theorem
9.3.2
and Theorem
(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.
Hada
rank p
JanJ(p+l)
< oo.
anO
9.3.7
9.3.2
J(z)
where
zm eg(z) P(z),
is the degree of
g,
genus
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).
bution theory, that is, the study of how many times an entire function can
assume any given value.
Theorem 9.3.9. Let
Zj EC
such that
f(zj)
c.
295
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).
Jeg(z)J
z
eg(z) (z  ai)
= ) f(z)J
I I1j=1(z  aj)I
with Jzl
elziH
oo
e9
is>..
Theorem 9.3.10.
Let
e9
g. Hence q
Z.
=>.,
0
Then the image of f contains all complex numbers except possibly one. If,
in addition,
Proof. We need only prove the first statement. If the image off omits two
complex numbers
a1
f(z)  a1
for some entire function
a polynomial. Since
g(z)
f(z)  a1
log(a2  ai). So if /3
this function
eg(z)
a2
makes
f  a1. Since
is
C with e/3
is constant. But
being constant
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
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)
r)
logf
l (zo)+
l
L: log
k=I
Ir
 akzo
r2
(zo ak )
1
27r
=
271"
Re
(reiO )
rei
r),
+ zo
 zo
r).
rei
logf(
l
.9
)l d0
f(zo) =/= 0.
2.
3.
4.
5.
r)
7.
8.
1.
zED
9.
10.
{an}
<C
satisfy
00
L 1 1 lanll
n=l
< oo.
297
Exercises
11. Let
<jJ
be a continuous function on
[a, b].
(1
log
[Hint:
so that by induction
(a)
{b)
(c)
f(z)
(d)
Pick
h>
h 2J.
f as
J(x)
I:
OjJI
(e)
L:j:0
rj sin(2jx) +
tends to
O;
J(x)] / h by
writing
I: rj sin(2Jx) + rJ sin(2Jx).
j>J
[f(x + h)
<
{d)
oD.
oD.
{ z : I z I 1}; hence f is
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
J11
f(a)
0, lal
r.
r,
then, near
=
integrable singularity.
Chapter 10
Analytic Continuation
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,
not even be such an extension: If V is the unit disc and U the punctured
disc, then the function
f(z)
1/z
< () <
tr,
C, and
f(rei9)
r112ei9/2,
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:
+
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:
f(z)
00
Lzj.
j=O
lzl
>
1.
{z EC: lzl
<
l}. It diverges
299
300
f ( refer
to a
that
f(z)
This formula for
f agrees
z.
1
=
\ { 1}.
an analytic continuation to C
f has
\ { 1}.
\ { 1}.
is the rather
EXAMPLE
10.1.2.
fo00
r(z)
e1et
dt.
15.1.
detail in Section
(1)
(2)
tRez1. Thus
when Re z > 0.
w11
the "singularity" at
( 3)
The function
The functions
with b
sign
( or
l/a,
1b
elet
Re z >
0}:
dt,
these integrals as
f(z)
a + o+.
{z
Re z ::::;
0 because the
r(z)
100
0
e1et
dt
.
=
1
z
eet
00 z1 100
I
+
0,
tze t
dt.
30 1
Now, whereas the original definition of the gamma function made sense on
Uo,
sense on
integral
U1
as
{z:
Rez
the lower
makes
of the
t':+let dt.
U2 = { z :
Re z
>
Continuing this process, we may verify that the gamma function, orig
inally defined only on
Uo,
U = {z EC :
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.
D(l, 1/2)
by
f(rei0) =r112e1012.
f(z),
principal branch
of
10. 1.
y'z.
f to
Note that
a
( Figure 10. 2) ,
[f(z)]2=z.
f(re10) = r112ei0/2.
If
10.3.
J( l)=i .
However, we might have begun our analytic continuation process
shown in Figure
found that
a.<;
we would have
1 4,
Chapter
6).
302
Figure 10.1
Figure 10.2
Figure 10.3
303
Figure 10.4
is a disc
D(P, r )
function element is
U),
where
If (/1,
(fk, Uk)
is an
Clearly
(/ 1, U1)
(fk, Uk)
(fj, Uj)
is a
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
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
yet functions in the usual sense, and they are not analytic in any sense that
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
But a
global analytic function may include more than one function element of the
form
may
have in effect more than one value at a point of C: Two function elements
(!1,
U)
and
(h, U)
U.
Ji(P) i h(P),
where
corresponding to
and
15
in
(namely, the square root function) with two distinct branches centered at
the point
1.
D(2, 1)
and let
i arg z,
1 + iO
(f, U)
can
14, Chapter
log I z I + iarg z branch on
is played by the
for the reader to bear in mind. From this viewpoint, two function elements
(!1,
U)
and
(h, V)
P)
at a point
(such that
and
analytic
y(O)
[O, 1]
t
(Figure
(f, U) be a func
10.5). An analytic
305
y(l)
Figure
10.5
Proposition
10.2.2.
Let(!,
U)
U)
a disc having
on
U1 n U1 .
t0
[O,
t o],
the
functions ft and
both Ut and
[O,
1].
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 by definition.
Thus to show
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
on
Ut ). n Ut ). nut' nut'.
Thus
ft1
ft1
on
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.
=
10.3.
307
[O, 1]
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
(J, U)
(with
D(P, r ))
'Y(t1) of 'Y
U.
along a curve
that is in
'Y(t1).
'Y
starting at
is the
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
(J, U)
where
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
=
of convergence of our power series were always small enough so that the
chain of discs never engulfed
continuation at
If
0.
0.
U,
(J, V)
where
V U.
In practice no confusion
should arise.
10.1,
the reason that such a failure could occur is that the curve
'Yo(t)
ei7rt,
308
:=::; t :=::;
1,
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
Thus it is not
10.1.3.
t
1]
[O, 1]
t
such that
(1)
H ( O, t )
(2)
H (1, 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:
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)
Condition
(4)
(3)
(2)
H ( s, t ) .
says that
Definition 10.3.2.
U.
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.
309
Proof. Write
fs,t
analytic continuation of
1+
fs,t,
(J, U)
H(s,
1+
t).
t)
obtained by
In other words,
t
Ji,1
1+
t).
fo,1.
H(s,
f>...1
fo,i}.
!1,1
fo,1,
as
so,
so
fso,t
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
t)
 H (so,
t) I
 8, so + 8)
 8, so+ 8)
fs,t
> 0
.fs,t
along H(s,
t),
t)
that is
fso,t
fs.o is
so
310
8
fs,l is a direct analytic
( so  8, so + 8) n [O, 1],
Sj
So
so
 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
that W is
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.
U)
admits unre
stricted continuation in W.
on
U.
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} .
C)
U)
at P.
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.
take us too far afield: We shall be content here with the informal idea that
a surface is a twodimensional 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.)
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
< () <
7r /2
C \ {0}.
in
(J, U)
can
and lying
Of course, in
(J, U)
7r : R
t
C \ {0}
C \ {O}.
C \ {O}: A
U; (J, U) is a
by
U.
This is just new terminology for a situation that we have already discussed.
of a given
(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} .
7r : R
t
in the plane.
Let us try to visualize R still further. Let
Then
7r1(W)
C\ {z
x+iO: x O}.
312
Figure
__
10.7
__
Figure
10.8
71"i(W)
7r1(W), the
projection
7r
is onetoone.
rei8
1t
rl/2ei8/2
and
10.8.
R, we have restored the negative real axis (but not the point
We have now constructed a surface, known
as
0).
the function viz." T his surface that we have obtained by gluing together the
two copies of
So we can regard our geometric surface, built from gluing the two
copies of
as
7r
: R
t
C \ {O}
7r
7ri : 7r( U)
t
t
to be holo
is holomorphic
313
F((f, U))
f(P).
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 1t
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), 7rl (W) has infinitely many components
each a copy of Won which 7r maps onetoone onto W. Namely, these
components are the "branches" of log z on W:
rei8
1t
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 twodimensional, 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 14,
8, and 17, in which there are more complicated function elements than ../Z
and log z.
314
==
==
==
::=:::
_
::>
_____
Figure 10.9
Figure 10.10
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
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
+
C omits
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
{z
Im z >
0}.
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
az + b
.
CZ+ d
Since
and
must all be real (after multiplication of all four numben; by the same
a, b, c,
[ac I z 12 +
ad  be> 0.
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
a, b, c, d
JR and
az + b
CZ+ d
ad  be > 0.
6.2)
i(l 
z)/(1 + z).
modular group.
a, b, c,
and
real
integers
and
az + b
CZ +d
ad  be
317
inverse of such an
is
11 ( z )
dz  b
__
cz + a'
> C
(f
)(z)
modular function if
E K and all z E U,
is called a
f(z).
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.
with
a, d
az + b
CZ + d
odd integers and b, c even integers, as you can check for yourself
see Exercise
16.
c/>m : JR.
given by
<Pm(x)
x+m.
>
The mappings
JR.
<Pm form
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 } .
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
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
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
W is
action of r.
Proposition 10.5.3.
(1)
(2)
f(W) n g(W) = 0.
LJ f(W) = U.
fEf
Proof. To prove (1), notice that it suffices to show that (g1of) (W) nW=
0. In other words, if e/= h Er, then h(W) n W= 0. Write
h(z) =
az+b
.
cz + d
(a)
c = 0. Then
(b)
ad =
1. Since
ad  be =
1, we see that
ad 
2bd = 1 or
2. Hence, since
Picard's
10.5.
319
Theorem
d(a 2b)
1, we see that h( z )
(W)
that
( z )+2m,
w).
consists
maps
image of
(c ) c =f.
z
of the
lc z + di
E vF. If we accept this claim for the moment, then we may use
the formula
l m h(z)
(remember that adbe
z
lcz +dl2
Im
z EW.
If
h1
Imh1(h( z ))
< Im h( z ). These
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).
z E W.
But then
lcw + di
Thus
< 1 for
and
d,
the quantity
cw +d
w,
under
lc z +di
h Er.
(2).
(1)
z E W.
of Proposition 10.5.3.
h(W),
(i)
wk ( z )
(ii)
Of course,
z + 2k.
{z
wk(W), k E Z.
E C : l2z  II
{z
E C : I 2 z + 1I
z EU
that
1.
(t)
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
111
Analytic Continuation
Im h(z):::; Im z
h E r.
h=
w
n. Note that the first of these is given by
o
w
n and to
'T} 2n
n ( )=
w
TJ
2ry4n+l
h=
( ** )
n
(*)
(IJ
)=
'T} 2n
_2ry+4n+l
l2z4n+lll
It follows that
/il
( t);
( ** )
and
( *** ) ,
we
l2z4nll 1.
and
satisfies
E f, we find that
n E Z,
hence
z E M.
Since
EM.
D
theorem .
Theorem
>.
( 1) ). o h = ). for every h
(2)
>.
as
E f;
is onetoone on W;
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}.
W'
____,
U,
We are using
321
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}.
Further,
This extends
g is
onetoone 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
8W.
g to all of U
by repeated re
E h(W),
all
h E f.
lies in
and
h Er.
( 3).
has range C
\ { 0, 1},
( 1)
h(W).
interior of
3 that a
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).
{ h(O)
h E r}
322
(3.
1).
<C
U with..\
____,
____,
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 onetoone on V. Observe
that V will intersect at most two of the fundamental regions h(W). Let p
=
.Alv
\ { 0, 1}
V' n V =/= 0
and..\ maps
V'
V'
p.
p'
will
Eo,
f(Vo)
[O, 1]
____,
f o'Y is compact.
Vi as above such
Pi,
\ { 0, 1}.
Also, we
(Pi1.Ei1).
f(Vi)
(pi,i)
Pi
is to
(k, Vo)
(k, Vo)
along
'Y
Vo
by construction.
Hence ..\ o K
J on <C.
But
is constant.
p
(p, Eo)
=..\ is infinitetoone, it is also the case that the (potential) domain for
p is
the set
C \ { 0, 1}
323
(po
f, Vo); in
does apply.
The following theorem , known as the big theorem of Picard, strengthens
Theorem 10.5.5 (the little theorem).
singularity at
P.
If
f. >
\ {P}
D(P, E ) \ {P}
sin(x+
2n)
sin x,
cos(x+
2n)
cos x
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.
2n.
The exponential
function ex of the real variable xis not periodic, but the complex exponential
324
ez
function
27ri:
For all
<C,
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 constant. But it will turn out that there are nonconstant meromorphic
f : <C
functions
<C
>
{ oo}
elliptic funct'ions.
We shall concentrate our discussion on the case when the periods are
a =
and
i/3
i.
w1
and
w2
f(z+wi)
f( z ) ,
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
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
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
with
artificial as it
10.6.
325
Elliptic Functions
'""
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
I (z
1
m ni)2
1
m ni)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
is of order
1 / h3,
for
1
+
z2
[ (zmni)2 (m : ni)2]
1
L
m,nEZ
(m,n)f(O,O)
P(z)
{ m + ni : m, n
Z}.
is in fact doubly
m,nEZ
1
(zmni)2
P(z)
m,nEZ
(m,n)f(O,O)
( since
1
1
1
+
(zmni)2  (m+ni)2
z2
( for
m,nEZ
(m,n)f(O,O)
by
z+ 1
1
1
1
+
(z(ml)ni)2 (m+ni)2
(z+l)2"
m, n
1
(z(m 1 )ni)2
run over the same set of terms.
l/(m + ni)2 in
the series
different ways.
and
1
(zmni)2
since
m,nEZ
1
(m+ni)2
(m,n)f(0,0)
and
L
(zmni)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
(zmni)2 (m
P,
: ni)2] + :2
327
z not
( for
a pole of P) :
P'(z)
"'"'
rn,n
2
(z mni) 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
P'
P.
is absolutely and
i gives
91 (z)
P(z + 1 )  P(z)
92(z)
P(z + i)  P(z)
and
have derivatives that are
91 ( 1 / 2 )
But
P is
P(l/2)  P(1/2) .
P(z)
P(z)
with
Thus 91
0. Similarly, setting
doubly periodic.
i / 2 gives 92
0. Thus
is truly
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 functionsee Exercise 14. )
a polynomial in
0. The
328
contains no other negative powers of
powers of
z.
Namely,
z,
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
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+
).
P(z)=
1
(1+B1z4+ ..
z2
( the
absence of the
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
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,
C 2,
with
C1
as before.
329
P,
and we may
"solve" it by writing
P'(z)
=
y'4P3(z)  C1P(z) + C2
Hence
zo
I J4w3 dwC1w
+ C2
22).
This calculation shows, at least in outline form, how the "elliptic" function
dw
which are the sort that arise in computing the arc length of an ellipse (Exer
cise
15).
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)
with periods
and
1.)
P.
(One
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 cos1). 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 + eiw)/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 "k1" if
k(z) z + (l/z).
3. Same as Exercise 2, for "m1" if m(z)
(1  z 2) 2.
=
4. Same as Exercise
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
331
Exercises
k?
[Hint:
general case.]
00
1
= L 72'
z1.
J
j=l
[If
.
00
J(z)=I:
z3
j=l
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
(0, oo ) JR
(that is, is continuous on all of JR, but vanishes off some compact
set). Define the function
F(z)=
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.
10.6.
332
Exercises
16.
17.