Complex Analysis Christer Bennewitz

COMPLEX ANALYSIS
Fall 2006
Christer Bennewitz
Mnne ifrn hgre zoner

analytiska funktioner
svaret nu dig finna lta
p oddlighetens gta?
F. Lffler
c 2006 by Christer Bennewitz

Copyright
Preface
These notes are basically a printed version of my lectures in complex
analysis at the University of Lund. As such they present a limited view
of any of the subject matters brought up, caused by the time constraints
one is faced by in a series of lectures. The core of the subject, presented
in Chapter 3, is very strongly influenced by the treatment in Ahlfors
Complex Analysis, one of the genuine masterpieces of the subject. Any
reader who wants to find out more is advised to read this book.
Mathematical prerequisites are in principle the mathematics courses
given in the first two semesters in Lund. Most importantly, this includes a reasonably complete discussion of analysis in one and several
variables and basic facts about series of functions including absolute
and uniform convergence. A course in topology is also useful, but not
essential. Primarily, a familiarity with the concept of a connected set
is of use.
Egevang, August 2006
Christer Bennewitz
Contents
Preface
Chapter 1. Complex functions

1.1. The complex number system
1.2. Polar form of complex numbers
1.3. Square roots
1.4. Stereographic projection
1.5. Mbius transforms
1.6. Polynomials, rational functions and power series
1
1
4
5
7
10
16
Chapter 2. Analytic functions

2.1. Conformal mappings and analyticity
2.2. Analyticity of power series; elementary functions
2.3. Conformal mappings by elementary functions
23
23
27
33
Chapter 3. Integration
3.1. Complex integration
3.2. Goursats theorem
3.3. Local properties of analytic functions
3.4. A general form of Cauchys integral theorem
3.5. Analyticity on the Riemann sphere
37
37
40
45
48
51
Chapter 4. Singularities
4.1. Singular points
4.2. Laurent expansions and the residue theorem
4.3. Residue calculus
4.4. The argument principle
53
53
56
59
65
Chapter 5. Harmonic functions

5.1. Fundamental properties
5.2. Dirichlets problem
73
73
79
Chapter 6. Entire functions

6.1. Sequences of analytic functions
6.2. Infinite products
6.3. Canonical products
6.4. Partial fractions
6.5. Hadamards theorem
85
85
87
90
94
96
Chapter 7. The Riemann mapping theorem

iii
101
iv
CONTENTS
Chapter 8. The Gamma function
105
CHAPTER 1
Complex functions
1.1. The complex number system
Recall that a group (G, ) is a set G provided with a binary operation 1 satisfying the following properties:
(1) For all elements x, y and z G holds (x y) z = x (y z).
(associative law)
(2) There exists a neutral element e G with the properties xe =
e x = x for every x G.
(3) Every element x G has an inverse x1 with the properties
x x1 = x1 x = e.
Exercise 1.1. Show that a set provided with an associative binary
operation can have at most one neutral element.
Hint: Show that if the set has a left neutral element and a right
neutral element, they must coincide.
Exercise 1.2. Show that if a set has an associative binary operation with neutral element, then any element of the set has at most one
inverse.
Hint: Show that if an element has a left inverse and a right inverse,
then these must coincide.
A group may also have the property
(4) For all elements x and y G holds x y = y x. (commutative
law)
in which case the group is called commutative or Abelian (after Niels
Henrik Abel (18021829)). Familiar examples of Abelian groups are
(Z, +), the integers under ordinary addition; (R, +), the real numbers under addition; (Rn , +), the set of n-tuples of real numbers under
(vector) addition; and (R \ {0}, ), the non-zero real numbers under
multiplication. As an example of a non-Abelian group, consider the
set of all rotations around lines through the origin in 3-dimensional
space; the binary operation is the ordinary composition of maps. The
reader should check these examples carefully; in particular, find the
neutral elements and inverses in these groups.
1That
is, a map : G G G, so that for every pair of elements x, y of G,

there is a unique element of G denoted by x y.
1
1. COMPLEX FUNCTIONS
A field (F, +, ) is a set F provided with two binary operations +

and , such that (F, +) is an Abelian group and, if 0 denotes the neutral
element of this group, also (F \ {0}, ) is an Abelian group. In addition
the distributive laws
(
(x + y) z = x z + y z,
x (y + z) = x y + x z.
hold for all elements x, y and z F . It is usual to denote the neutral
element of (F \ {0}, ) by 1.
Exercise 1.3. Prove that in any field F holds 0 x = x 0 = 0
for all x F (as always, 0 denotes the neutral element of the group
(F, +)).
Exercise 1.4. Prove that a field does not have any non-zero divisors of zero, i.e., if xy = 0, then either x = 0 or y = 0.
Familiar examples of fields are (Q, +, ), the rational numbers under
ordinary addition and multiplication, and (R, +, ). We shall show, in
this section, that there is precisely one reasonable way of making the
Euclidean plane into a field. By introducing Cartesian coordinates
this plane may be identified with the Abelian group (R2 , +), and we
will make this into a field by extending the usual multiplication of an
element of R2 by a real number. The resulting field is the field C of
complex numbers.
To see how to make the definition, assume we have already managed
to construct our field C. Then there is a multiplicative neutral element,
which we will for the moment denote by 1, to distinguish it from the
real number 1. We may identify R with the set of real multiples of 1
(explain!) and may therefore consider R as a subset of C. Let e be
an element of R2 which is linearly independent of 1, so that 1, e is a
basis in R2 . Any element z C may then be written z = x1 + ye
with real numbers x and y. In particular, there are real numbers a
and b such that e2 = a1 + be so that z 2 = (x2 + ay 2 )1 + (2xy + by 2 )e
(note that 1 1 = 1, e 1 = e). Now clearly z 2 is real if y = 0 (since
actually z itself is, by the identification above). But z 2 will also be real
2
2
if x = 2b y. We then get z 2 = (a+ b4 )y 2 . We can not have a+ b4 0 by
q
q
2
2
Exercise 1.4 since then (z y a + b4 )(z + y a + b4 ) = 0, but neither
2
of the factors is 0 unless their e-component y = 0. Hence a + b4 < 0.

q
2
If we set y = 1/ (a + b4 ) we therefore get z 2 = 1.
Roughly, we have seen that if we can define a multiplication in
2
R which makes it into a field with addition being the ordinary vector
addition, then there exists an element the square of which is 1 (rather,
the additive inverse of the multiplicative neutral element). We pick one
such element (we will see later that there are precisely two), denote it
1.1. THE COMPLEX NUMBER SYSTEM
by i and call it the imaginary unit. If we use 1, i as a basis we may

therefore write any element in the plane as x1 + yi with real x, y. For
convenience we will actually write it x + iy from now on.
It is important to note that we have not yet shown that it is possible
to make a field of the plane; we have just seen that if it is possible,
then we may identify the x-axis with the real numbers and the y-axis
with the multiples of an element, the square of which is 1.
Exercise 1.5. Show that if we calculate with symbols x+iy, where
x and y are real numbers, according to the usual rules for adding and
multiplying numbers and in addition use i2 = 1, then all the requirements for a field are satisfied.
From now on the field we have constructed is denoted by C and
called the field of complex numbers. Note that the field of real numbers
is an ordered field. This means that we have a relation < defined among
the real numbers such that
(1) If x and y R, then exactly one of x < y, y < x and x = y is
true.
(2) Sums and products of positive (i.e., > 0) numbers are positive.
We have not introduced anything similar for the complex numbers for
the simple reason that it can not be done.
Exercise 1.6. Show that in an ordered field squares of non-zero
elements are always > 0. Use this to show that if it were possible to
make C into an ordered field, then both 1 > 0 and 1 > 0, and hence
also 0 > 0, a contradiction.
As a final note to this first section, the fact that the Euclidean plane
can be made into a field is extremely useful in all areas of mathematics
and its applications. Since we live in a 3-dimensional (at least) world, it
would, from the point of view of applications, be very useful if we could
make 3-dimensional space into a field as well. In the early part of the
nineteenth century, this is exactly what the famous Irish mathematician
W. R. Hamilton tried, unsuccessfully, to do.
Exercise 1.7. Try to show that Hamilton was doomed to fail. To
simplify things, you may require that the complex plane should be
a 2-dimensional restriction of the 3-dimensional field. Show that the
existence of divisors of zero can not be avoided.
Hamilton succeeded (1843) to introduce a multiplication in R4 which
makes this into a field, with the minor defect that the multiplicative
group is not Abelian (such a structure is called a skew field ). Hamilton
called his structure the quaternions; this structure actually strongly
hints that it would be profitable, in physics, to consider the world 4dimensional, with time as the fourth dimension.
Exercise 1.8. Consider the set of symbols x + iy + ju + kv, where

x, y, u and v are real numbers, and the symbols i, j, k satisfy i2 =
j 2 = k 2 = 1, ij = ji = k, jk = kj = i and ki = ik = j. Show
that using these relations and calculating with the same formal rules
as in dealing with real numbers, we obtain a skew field; this is the set
of quaternions.
1.2. Polar form of complex numbers
In the complex number z = x + iy the real number x is called the
real part of z, x = Re z, and the number y is called the imaginary
part of z, y = Im z. There is of course nothing imaginary whatever
about the imaginary part; the reasons for this curious appellation are
historic. If we introduce the notation z for the complex number x iy,
called the complex conjugate of z, we see that Re z = 12 (z + z) and
Im z = 2i1 (z z). In particular, z is real (i.e., has imaginary part 0)
precisely if z = z. If z has real part 0, so that z = z, one calls z
purelyp
imaginary. We define the absolute value |z| of z = x + iy to be
|z| = x2 + y 2 . This is of course the ordinary length of z, considered
as a vector in the plane, provided we draw 1, i as orthonormal vectors.
A very useful observation is that zz = |z|2 .
Exercise 1.9. Show this and that for any complex numbers z and
w we have
(1) z + w = z + w,
(2) zw = z w,
(3) |zw| = |z||w|.
It is worth remarking how one carries out division by a complex
number. Since the complex numbers constitute a field, every non-zero
complex number has a multiplicative inverse, i.e., we can divide by it;
namely, if z 6= 0 and w are complex numbers, then there is a unique
complex number u, denoted wz , such that zu = w. The question is,
how does one write the quotient on the standard form as real part plus
i times imaginary part. To see how, multiply through by z to obtain
|z|2 u = zw. Since |z|2 6= 0 we can divide by this (real) number, and so
u = zw/|z|2 . So, to write w/z on standard form, multiply numerator
and denominator by z.
Exercise 1.10. Write
1+2i
3+4i
on standard form.
The geometric interpretation of addition is already familiar, since

this is the ordinary vector addition in the plane. To get a geometric
picture of multiplication, we introduce polar coordinates in the plane in
the following way. If z 6= 0, then z/|z| is located somewhere on the unit
circle; hence we can find an angle such that z/|z| = cos + i sin . We
may therefore write z on polar form as z = |z|(cos + i sin ) where
is called the argument of z and is denoted = arg z. It is unfortunate,
1.3. SQUARE ROOTS
but extremely important that arg z is NOT uniquely determined by

z; adding any integer multiple of 2 to gives another, equally valid,
value for arg z. When one therefore speaks of the argument for a
complex number, one means one of the infinitely many possible values
of the argument. Another, less serious ambiguity, is that we have not
assigned an argument to the number 0; it is usual to allow any real
number whatsoever as a valid argument for 0.
Now suppose z = |z|(cos + i sin ) and w = |w|(cos + i sin )
are complex numbers. Then zw = |z||w|(cos cos sin sin +
i(cos sin + sin cos )) = |zw|(cos( + ) + i sin( + )) according
to the addition formulas for sin and cos. Thus, when calculating the
product of two complex numbers the absolute values are multiplied and
the arguments are added. In particular, multiplication by a complex
number of absolute value 1 is equivalent to a rotation with an angle
equal to the argument of the given number.
Exercise 1.11. Write the number z = 3 + i on polar form and

then calculate z 13 on standard form.
1.3. Square roots
Working with real numbers it is possible to find the square root
of any non-negative number; to obtain a unique number the square
root is required to be non-negative as well. After introducing complex
numbers we can, for any given real number, find a real or complex
number whose square is the given number. Of course, not much would
be gained unless we could actually find the square root of any complex
number as well. This means that we would like to be able to find a
solution to z 2 = w for any complex number w. Suppose w = u + iv
and let z = x + iy (in situations like this it is always assumed that u,
v, x and y are real numbers). Since z 2 = x2 y 2 + 2ixy we need to
solve the nonlinear system
(
x2 y 2 = u,
(1.1)
2xy = v.
in two real unknowns x and y. Squaring and adding the two equations
we get, after extracting a (real) square root, that x2 + y 2 = u2 + v 2

(this simply expresses the fact that |z|2 = |w|, which has to be true in
view of Exercise 2.1). Together with the first equation this shows that
q
x = 1 ( u2 + v 2 + u),
q2
(1.2)
y = 1 ( u2 + v 2 u).
2
Note that all the expressions within square roots are non-negative no
matter what u and v are, so these are ordinary real square roots. (1.2)
therefore give all possible solutions of (1.1), and it is easily verified that
the first equation is actually satisfied, whereas the second is satisfied

if and only if one chooses the right combination of signs, so that there
are actually always precisely two distinct complex numbers z satisfying
z 2 = w, unless w = 0 in which case z = 0 is the only solution. Since
a quadratic equation can be solved by extracting square roots one now
easily sees that any quadratic equation with complex coefficients always
has a complex root. In fact, if counted by multiplicity there are always
exactly two roots (we will return later to the concept of multiplicity
for a root).
We have seen that we can always extract square roots of a complex
number w, and that there are always (unless w = 0) exactly two such
numbers. The question arises:
Which of the two possibilities are we
to denote by the symbol w? Since the complex numbers are not

ordered there is no simple answer to this question, as in the real case.
To analyze the situation we write
p w = |w|(cos + i sin ) on polar form.
2
If z = w, then clearly |z| = |w|, and if is an argument for z, then
2 must be an argument for w. The simplest choice for is therefore
to set = /2. Which number z we get this way obviously depends
on the choice of , which is only determined up to an integer multiple
of 2. If we add 2 to we will add to , which will replace z by
z. Adding or subtracting further multiples of 2 to will not yield
any more values for z, so we have again seen that there are exactly
two square roots of any non-zero number. We can write any complex
number w on polar form with an argument in the interval <
and choosing the argument of the square root to be = /2 we will
get 2 < 2 .
This is one way of assigning a unique value to the square root of
any complex number. Considering z as a function of w this is called
the principal branch of the square root; if w is a non-negative real
number it obviously coincides with the usual real square root. The
values of the principal branch of the square root are all in the right
half plane, i.e., they have non-negative real part. There are, however,
other ways of choosing a branch of the square root that are sometimes
more convenient. On may for example restrict to the interval 0
< 2, which will give the argument of the square root in the interval
0 < , i.e., this branch of the root has all its values in the upper
half plane.
Why can one not, once and for all like in the case of the real square
root, choose a particular branch and stick to it? The reason is problems
with continuity. Suppose we have a nice curve in the w-plane which
intersects the negative real axis. If we take the square root of this, using
the principal branch, the image of the curve in the z-plane will jump
from a point on the negative imaginary axis to a point on the positive
imaginary axis; we have lost the continuity of the curve. Another choice
of branch might solve the problem for a particular curve, but it is clear
1.4. STEREOGRAPHIC PROJECTION
that no choice of branch will be suitable for all curves. Since there is
no choice of branch
which will work best in all situations one must not
use the notation without specifying which branch of the square root
one is talking about.
The need to deal with several different branches occurs for all kinds
of other complex functions and is a major complicating factor in the
theory. There is a sophisticated and completely satisfactory solution
to the problem, namely the introduction of the concept of a Riemann
surface. Unfortunately we can not go into that here.
1.4. Stereographic projection

Since we have a notion of distance (i.e., d(z, w) = |z w|) in C we
may view C as a metric space. It is clear that this space is complete
in the sense that any Cauchy sequence converges; to see this note that
since | Re z| |z| and | Im z| |z| | Re z| + | Im z| for any z C
it follows that if zj = xj + iyj , j = 1, 2, . . . is a Cauchy sequence in
C, then xj , j = 1, 2, . . . and yj , j = 1, 2, . . . are Cauchy sequences in
R. Furthermore, if xj x R and yj y R as j , then
xj + iyj x + iy C as j . Thus the completeness of C follows
from that of R.
From the point of view of topology, it would be even better if C
were compact, i.e., any open cover of C should have a finite subcover.
This is not true, however, as can be seen by considering the open cover
of C consisting of all open balls |z| < R centered at 0, which obviously
has no finite subcover. One can make C compact without changing
its topology by adding (at least) one ideal point and modifying the
metric. This one-point compactification of the complex plane is very
important in the theory of functions of a complex variable and we will
give a very enlightening geometric interpretation of it in this section.
Imagine C as the x1 x2 -plane in R3 and let S2 be the unit sphere;
it will intersect C along the unit circle. Call the point (0, 0, 1) on the
sphere the North pole N (so that (0, 0, 1) is the South pole). We can
map C in a one-to-one fashion onto S2 \ {N } by mapping z C onto
the point (x1 , x2 , x3 ) S2 such that the straight line connecting z with
N goes through (x1 , x2 , x3 ). This map is called stereographic projection
and has many interesting properties, as we shall see. In this connection
S2 is called the Riemann sphere.
It is nearly obvious that this stereographic projection is a bi-continuous map, using the topology induced by the metric of R3 . To make
absolutely sure, let us find the mapping explicitly. The line through
N and z = x + iy C is (x1 , x2 , x3 ) = (0, 0, 1) + t(x, y, 1). The
intersection with S2 is given by t satisfying t2 (x2 + y 2 ) + (1 t)2 = 1
which gives t = 0, i.e., N , and the more interesting t = 2/(x2 + y 2 + 1).
Figure 1. Stereographic projection

We therefore get
2 Re z
x1 = 2
|z| + 1
2 Im z
x2 = 2
.
|z| + 1
|z|2 1
x3 = 2
|z| + 1
Since |z|2 + 1 = 2/(1 x3 ) by the third equation the inverse is easily

seen to be given by
x1 + ix2
z=
.
1 x3
It is clear that these maps are both continuous (note that (x1 , x2 , x3 )
S2 \ {N } so x3 6= 1). We may now introduce a new metric in C by
setting the distance between points in C equal to the Euclidean distance
between their image points on S2 .
Exercise 1.12. Show that this metric is given by
d(z, w) = 2
(|z|2
|z w|
.
+ 1)1/2 (|w|2 + 1)1/2
Also show that the distance between the image of z and N is
2
.
(|z|2 +1)1/2
In view of Exercise 1.12 we may now add to C an ideal point ,

the image of which under stereographic projection is N . This new set
is called the extended complex plane and we denote it by C . Using the
metric of Exercise 1.12 in C the extended plane becomes homeomorphic to the Riemann sphere with the topology of Euclidean distance.
Since S2 is compact, so is the extended plane; we have compactified
the plane. For the statement of the next theorem, note that a circle
in S 2 is the intersection of S 2 by a non-tangential plane, and any such
(non-empty) intersection is a circle.
1.4. STEREOGRAPHIC PROJECTION
Theorem 1.13. The image of a straight line in C under stereographic projection is a circle through N , with N excluded. The image
of a circle in C under stereographic projection is a circle not containing
N . The inverse image of any circle on S2 is a straight line together
with if the circle passes through N , otherwise a circle.
Proof. Since a straight line in the x1 x2 -plane together with N
determines a unique plane, the intersection of which with S2 is the
image of the straight line we only need to consider the case of a circle
in C. If it has center a and radius r its equation is |z a|2 = r2 or
1 +ix2
|z|2 2 Re(az) + |a|2 = r2 . Substituting z = x1x
into this, using that
3
2
2
2
x1 + x2 + x3 = 1 and x3 6= 1, we get 1 + x3 2x1 Re a 2x2 Im a +
(1 x3 )(|a|2 r2 ) = 0 which is the equation of a plane. Conversely,
a circle on the Riemann sphere is determined by three distinct points.
The inverse images of these three points determine a circle in C. The
image of this circle is clearly the original circle.
In view of this theorem we will by a circle in the extended plane

mean either a line together with , or an actual circle.
A map is called conformal if it preserves angles and their orientation
(we will give a more exact definition in Chapter 2.1). A surface is given
an orientation by assigning to each point a normal direction which
varies continuously with the point. For example, the usual orientation
of C is given by letting at each point the normal point upwards, i.e., in
our present picture in the direction of the x3 -axis. Similarly, we may
give the Riemann sphere an orientation by letting the normal point
towards the origin.
The angle between two smooth curves in an oriented surface at a
point of intersection of the curves is the angle between the tangents
at the point. There are two such angles, the sum of which is . If
the curves are given in a certain order, the positively oriented angle
between them is that angle through which one has to turn the first
tangent vector so as to coincide with the second tangent vector, turning counterclockwise as seen from the normal to the surface. A strict
definition would of course have to be freed from such obviously intuitive
geometric concepts, but we will not attempt this here.
Theorem 1.14. Stereographic projection is conformal.
Proof. Consider two curves intersecting at z and their tangents
at z in C. Together with N the tangents determine two planes that
intersect the Riemann sphere in two circles through N . The tangents
to the circles at N are in these planes and also in the plane through N
parallel to C. It follows that they are parallel to the original tangent
vectors so that viewed from inside the sphere they give rise to an angle
equal to but of opposite orientation to the original angle.
10
The circles intersect also at the image of z on the sphere, and are
tangent to the images of the curves there. The angles at the two points
where the circles intersect are equal but of opposite orientation by
symmetry (the two angles are images of each other under reflection in
the plane through the origin and parallel to the normals of the planes
of the circles). The theorem now follows.
Although the proof above is very geometric in nature it is actually not difficult to make it analytic, using the fact that stereographic
projection is a differentiable map, but we will not do that here.
1.5. Mbius transforms
A Mbius transform (also called a linear fractional transformation)
az + b
is a non-constant mapping of the form z 7 f (z) =
for complex
cz + d
numbers a, b, c and d. To begin with we consider this defined in C
except, if c 6= 0, for z = d/c. The fact that the mapping is nonconstant means that (a, b) is not proportional to (c, d). This can be
expressed by requiring ad bc 6= 0 which is always assumed from now
on. Clearly we get the same mapping if we multiply all the coefficients
a, b, c, d by the same non-zero number so that although the mapping
is determined by the matrix ( ac db ) any non-zero multiple of this matrix
gives the same mapping. The requirement ad bc 6= 0 means that
the determinant is 6= 0 so multiplying by an appropriate number we
may always assume that the determinant is 1. This determines the
coefficients up to a change in sign of all of them.
It is clear that if c = 0, then f (z) as z . On the other
hand, if c 6= 0, then f (z) as z d/c and f (z) a/c as z .
We may therefore extend the definition of f to all of the extended plane
C in such a way that the extended function is a continuous function
of C into C . We will always consider Mbius transforms as defined
in the extended plane, or equivalently on the Riemann sphere, in this
way. We have the following interesting proposition.
Proposition 1.15. If f and g are Mbius transforms corresponding to the matrices A and B, then the composed map f g is a Mbius
transform corresponding to the matrix AB.
Exercise 1.16. Prove Proposition 1.15.
Since the set of all non-singular 2 2 matrices is a group under
matrix multiplication, it follows that so are the Mbius transforms.
This means that any Mbius transform has an inverse which is also a
Mbius transform.
Exercise 1.17. Find all Mbius transforms T for which T 2 = T .
1.5. MBIUS TRANSFORMS
11
Among other things this means that a Mbius transform is a homeomorphism of the extended plane onto itself, i.e., a continuous oneto-one and onto map whose inverse is also continuous. But Mbius
transforms have more surprising properties. Recall that we by a circle
in the extended plane mean either an actual circle in the plane or a
straight line together with .
Theorem 1.18. Mbius transforms are conformal and circle-preserving, i.e., any circle in the extended plane is mapped onto a circle
in the extended plane.
Proof. The theorem is obvious for certain simple special cases,
namely a translation z 7 z + b, a rotation z 7 az where |a| = 1
and a dilation z 7 az where a > 0. It is therefore also true for a
multiplication z 7 az where 0 6= a C, since this is the composite
a
of the rotation z 7 |a|
z and the dilation z 7 |a|z. Composing a
multiplication with a translation the theorem follows for a linear map
z 7 az + b where a 6= 0, and hence for any Mbius transform for which
1
c = 0. If c 6= 0 we have az+b
= bcad
+ ac so that this map is the
cz+d
c cz+d
composite of three maps, the first and last being linear and the middle
map is the inversion z 7 1/z. The theorem therefore follows if we can
prove it for an inversion.
If the image of z under stereographic projection is (x1 , x2 , x3 ), then
3
2
1 +ix2
so that 1/z = x1x
= (1 x3 ) xx12ix
we have z = x1x
Since
2 .
3
1 +ix2
1 +x2
(x1 , x2 , x3 ) is on the unit sphere we have x21 + x22 = 1 x23 so that
1 ix2
. Therefore 1/z is the inverse stereographic projection of
1/z = x1+x
3
(x1 , x2 , x3 ). The map that takes (x1 , x2 , x3 ) into this is a rotation
around the x1 -axis by an angle . This is obviously a circle-preserving
and conformal map, and since we know that also stereographic projection is circle-preserving and conformal it follows that the inversion has
the desired properties. The proof is now complete.
Exercise 1.19. Prove the theorem by calculation, not using stereographic projection.
Note that since removing a circle from the extended plane leaves
a set with exactly two components, and since Mbius transforms are
continuous in the extended plane, the interior of any circle in the plane
is mapped either onto the interior or onto the exterior, including ,
of another circle. This follows from the fact that continuous maps
preserve connectedness.
Exercise 1.20. Prove the statement above in detail.
Sets that are left invariant under a mapping are obviously important
characteristics of the map. For a Mbius transform one may for example ask which circles it leaves invariant, or conversely, which Mbius
transforms leave a given circle invariant. We will consider some such
12
problems later. Right now we will instead ask for fixpoints of a given
transform, i.e., points left invariant by the map. By our definition of
the image of , this is a fixpoint if and only if the map is linear. A
linear map z 7 az + b also has the finite fixpoint z = b/(1 a), except
if a = 1. Thus, a translation which is not the identity has only the
fixpoint , but any other linear map which is not the identity has
exactly one finite fixpoint as well. For a Mbius transform z 7 az+b
cz+d
with c 6= 0 the equation for a fixpoint becomes z(cz + d) = az + b which
is a quadratic equation. It therefore has either two distinct roots or a
double root. We have therefore proved the following proposition.
Proposition 1.21. A Mbius transform different from the identity
has either one or two fixpoints, as a map defined on the extended plane.
Exercise 1.22. Find the fixed points of the linear transformations
z
2z
3z 4
z
w=
, w=
, w=
, w=
.
2z 1
3z 1
z1
2z
In particular, a Mbius transform that leaves three distinct points
invariant is the identity. It also follows that there can be at most
one Mbius transform that takes three given, distinct points into three
specified, distinct points. Because, if there were two, say f and g, then
f 1 g would be a transform different from the identity and leaving the
given points invariant. Conversely, we will prove that there actually
always exists a Mbius transform that takes the given points into the
specified ones. To see this, define the cross ratio of four distinct points
z0 , z1 , z2 , z3 in C by
z0 z2 z1 z2
(z0 , z1 , z2 , z3 ) =
/
z0 z3 z1 z3
when all the points are finite. If one of them is , the cross ratio is
defined as the appropriate limit of the expression above. The following
proposition follows by inspection.
Proposition 1.23. Suppose z1 , z2 , z3 are distinct points in C . The
unique Mbius transform taking these points to 1, 0, in order is z 7
(z, z1 , z2 , z3 ).
It is now clear that to find the unique Mbius transform taking the
distinct points z1 , z2 , z3 into the distinct points w1 , w2 , w3 in order, one
simply has to solve for w in (w, w1 , w2 , w3 ) = (z, z1 , z2 , z3 ).
Exercise 1.24. Find the Mbius transformation that carries 0, i,
i in order into 1, 1, 0.
Exercise 1.25. Show that any Mbius transformation which leaves
R {} invariant may be written with real coefficients.
Exercise 1.26. Show that the map z 7 z1
maps the right halfz+1
plane (i.e., the set Re z > 0) onto the interior of the unit circle.
13
Two points z and z are said to be symmetric with respect to R if

z = z. If T is a Mbius transform that maps R {} onto itself, then
according to Exercise 1.25 one may write T with real coefficients. It
follows that T z and T (z ) are symmetric with respect to the real axis
if and only if z and z are. To generalize the concept of symmetry with
respect to the real axis to symmetry with respect to any circle in the
extended plane we make the following definition.
Definition 1.27. Let be a circle in C . Two points z and z are
said to be symmetric with respect to if there is a Mbius transform
T which maps onto the real axis for which T (z ) = T z.
By the reasoning just before the definition it is clear that this is a
genuine extension of the notion of conjugate points and that z and z
are symmetric with respect to precisely if T (z ) = T z for any Mbius
transform T that takes to the real axis. For, if T and S both take
onto the real axis and T (z ) = T z, then U = ST 1 maps the real axis
onto itself so that S(z ) = U T (z ) = U (T z) = U T z = Sz. There is
therefore for every z precisely one point z so that z, z are symmetric
with respect to . A similar calculation proves the next theorem.
Theorem 1.28. Suppose S is a Mbius transform that takes the
circle C onto the circle 0 C . Then the points z and z are
symmetric with respect to if and only Sz and S(z ) are symmetric
with respect to 0 .
Proof. If T maps onto the real axis, then U = T S 1 maps
0 onto the real axis. But U S(z ) = T (z ) and U Sz = T z so that
U S(z ) = U Sz if and only if T (z ) = T z. The theorem follows.
In short, Theorem 1.28 says that symmetry is preserved by Mbius

transforms. The next theorem allows us to calculate the symmetric
point to any given z and circle.
Theorem 1.29. If is a straight line, then z and z are symmetric
with respect to precisely if they are each others mirror image in .
If is a genuine circle with center a and radius R, then a and are
symmetric with respect to . If z is finite and 6= a, then z and z are
symmetric precisely if (z a)(z a) = R2 .
Proof. If is a straight line it is mapped onto the real axis by a
translation or a rotation and these transformations obviously preserve
mirror images.
zaR
If is a circle with center a and radius R the map z 7 i za+R
takes
onto the real axis (since a + R 7 0, a R 7 and a iR 7 1).
Now a and are mapped onto i and i respectively, so they are a
symmetric pair. If z has neither of these values a simple calculation
shows that z and z are mapped onto conjugate points precisely if
(z a)(z a) = R2 .
14
In particular the fact that the center of a circle and are symmetric
with respect to the circle are often very helpful in trying to find maps
that take a given circle into another.
Exercise 1.30. Find the Mbius transform which carries the circle
|z| = 2 into |z + 1| = 1, the point 2 into the origin, and the origin
into i.
Exercise 1.31. Find all Mbius transforms that leave the circle
|z| = R invariant. Which of these leave the interior of the circle
invariant?
Exercise 1.32. Suppose a Mbius transform maps a pair of concentric circles onto a pair of concentric circles. Is the ratio of the radii
invariant under the map?
Exercise 1.33. Find all circles that are orthogonal to |z| = 1 and
|z 1| = 4.
We will end this section by discussing conjugacy classes of Mbius
transforms.
Definition 1.34. Two Mbius transforms S and T are called conjugate if there is a Mbius transform U such that S = U 1 T U .
Conjugacy is obviously an equivalence relation, i.e., if we write
S T when S is conjugate to T , then we have:
(1) S S for any Mbius transform S.
(reflexive)
(2) If S T , then T S
(symmetric)
(3) If S T and T W , then S W .
(transitive)
It follows that the set of all Mbius transforms is split into equivalence
classes such that every transform belongs to exactly one equivalence
class and is equivalent to all the transforms in the same class, but to
no others.
Exercise 1.35. Prove the three properties above and the statement
about equivalence classes. What are the elements of the equivalence
class that contains the identity transform?
The concept of conjugacy has importance in the theory of (discrete)
dynamical systems. This is the study of sequences generated by the
iterates of some map, i.e., if S is a map of some set M into itself,
one studies sequences of the form z, Sz, S 2 z, . . . where z M . This
sequence is called the (forward) orbit of z under the map S. One
is particularly interested in what happens in the long run, e.g., for
which zs the sequence has a limit (and what the limit then is), for
which zs the sequence is periodic and for which zs there seems to be
no discernible pattern at all (chaos). Note that if S = U 1 T U , then
S n = U 1 T n U so that all maps in the same conjugacy class behave
qualitatively in the same way, at least with respect to the properties
15
listed above. It therefore seems natural to try to find, in each conjugacy

class, some particularly simple map for which the questions above are
particularly simple to answer. In other words, one looks for a canonical
representative in each equivalence class. We will carry out this for the
case of Mbius transforms.
If S = U 1 T U and z is a fixpoint of S, then U z is a fixpoint of T
since T U z = U Sz = U z. If S has only one fixpoint z0 we may choose
V so that V z0 = . Then V SV 1 has only the fixpoint and is
therefore a translation z 7 z + b for some b 6= 0. If we set U = 1b V it
follows that U SU 1 z = z + 1. If S has two fixpoints z1 and z2 we may
choose U so that U z1 = 0 and U z2 = . Then T = U SU 1 has the
fixpoint , so it is linear, T z = az + b, and it also has the fixpoint 0,
so b = 0. Now set, for 6= 0,
z + 1 for = 1 ,
T z =
z for 0 6= 6= 1 .
We have then proved most of the following theorem.
Theorem 1.36. For every Mbius transform S different from the
identity there exists 6= 0 such that S T . If T T , then either
= or = 1/.
Proof. It only remains to prove the last statement. But this is
clear if = 1, since this is the only value for which T has just one
fixpoint. We may therefore assume that and are both 6= 1 (and of
course non-zero). But if U T = T U and U z = az+b
we obtain
cz+d
az + b
az + b
=
cz + d
cz + d
for all z. Since ad bc 6= 0 we can not have d = c = 0. If d 6= 0, setting
z = 0 gives b/d = b/d so that b = 0 and therefore a 6= 0. If now c 6= 0,
setting z = d/c we get on the right but not the left. It follows
that c = 0 and (5.2) becomes = . On the other hand, if d = 0 we
must have c 6= 0 and so z = gives a/c = a/c. It follows that a = 0.
In this case (5.2) becomes = 1/ and the proof is complete.
(1.3)
What we have proved is that each conjugacy class different from

the class of the identity contains one of the operators T and also
T1/ , but no other operators of this form. We may therefore with any
Mbius transform S associate the corresponding unique (non-ordered)
pair (, 1/) of reciprocal complex numbers, called the multiplier of S.
The multiplier is thus a conjugacy invariant. Note that some T leave
the interior of certain circles in the extended plane invariant. Namely,
T1 leaves all halfplanes above or below a horizontal line invariant. If
> 0 (but 6= 1), then T leaves all halfplanes bounded by a line through
the origin invariant. Finally, if || = 1 but 6= 1, then T leaves the
interiors and exteriors of any circle concentric with the origin invariant.
16
On the other hand, if is neither positive nor of absolute value 1

there is no disk which is invariant under T . Show this as an exercise!
The transforms in the conjugacy class of T1 are called parabolic, those in
the conjugacy class of T for some > 0 but 6= 1 are called hyperbolic
and those in the conjugacy class of T for some 6= 1 with || = 1
are called elliptic. The reason for these names will be clear from the
result of Exercise 1.37. The remaining Mbius transforms are called
loxodromic. This is because they are conjugate to a T for which the
sequence of iterates z, T z, T2 z, . . . lie on a logarithmic spiral, which
under stereographic projection becomes a curve known as a loxodrome.
Exercise 1.37. Suppose that the coefficients of the transformation
az + b
Sz =
cz + d
are normalized by adbc = 1. Show that S is elliptic if 0 (a+d)2 < 4,
parabolic if (a + d)2 = 4, hyperbolic if (a + d)2 > 4 and loxodromic in
all other cases. Hint: The determinant and the trace a + d of a matrix
( ac db ) is invariant under conjugation by an invertible matrix.
Exercise 1.38. Show that a linear transformation which satisfies
S n = S for some integer n is necessarily elliptic.
Exercise 1.39. If S is hyperbolic or loxodromic, show that S n z
converges to a fixpoint as n , the same for all z which are not equal
to the other fixpoint. The exceptional fixpoint is called repelling, the
other one attractive. What happens when n ? What happens
in the parabolic and elliptic cases?
Exercise 1.40. Find all linear transformations that are rotations
of the Riemann sphere.
Hint: The antipodal point to a point on the unit sphere is obtained by
multiplication by 1. Use the fact that an antipodal pair is mapped
onto an antipodal pair by a rotation.
1.6. Polynomials, rational functions and power series
We define a polynomial to be a complex-valued function p of a
complex variable given by a formula p(z) = an z n + an1 z n1 + +
a1 z + a0 where the coefficients a0 , . . . an are complex numbers, an 6= 0,
and n is a non-negative integer, called the degree of the polynomial,
deg p. The function identically equal 0 is also a polynomial, of degree
. The sum of two polynomials of degrees n and m is a polynomial
of degree max(n, m). The product of two polynomials of degrees n
and m is a polynomial of degree n + m. The division algorithm says
that if p and q are polynomials, then there are unique polynomials k
and r with deg r < deg q such that p = kq + r. From this follows the
factor theorem which states that if p(a) = 0, then z a divides p. The
1.6. POLYNOMIALS, RATIONAL FUNCTIONS AND POWER SERIES
17
proof is simply the observation that since p(z) = k(z)(z a) + r where

r is constant (of degree < 1), then r = 0 if and only if p(a) = 0. It
is of course possible that the quotient k is also divisible by z a. If j
is the largest integer such that (z a)j divides p, then j is called the
multiplicity of a as a zero of p.
It also follows from the factor theorem that two polynomials p, q
for which p(z) = q(z) for all z C have to be identical, i.e., have the
same coefficients.
A very important fact about polynomials (which is only true if we
consider polynomials in the complex domain) is the fundamental theorem of algebra which says that any non-constant polynomial has a
zero. We will prove this later, but assume it for the present. Combining the fundamental theorem of algebra with the factor theorem it
easily follows that if we add up the multiplicities of all the zeros of a
polynomial p (count the zeros with their multiplicities), the sum will
be the degree of p.
Also for complex functions the concepts of limit and continuity
are of central importance. However, since complex numbers are just
vectors in R2 , where we in addition has defined a multiplication, we
can take these concepts over from the calculus of several real variables.
For reference we nevertheless state the definitions
Definition 1.41. Suppose f is a complex-valued function of either
a real or complex variable, with domain R or C.
If a is a point in the closure of , we say that limza f (z) = A
if A is a complex number such that for every > 0 there is a
> 0 with the property that |f (z) A| < whenever z
and 0 < |z a| < .
If a we say that f is continuous at a if limza f (z) = f (a).
All the standard calculation rules for limits and continuity familiar
from calculus continue to hold in this context, with exactly the same
proofs, so we will not dwell on this. We also remind the reader of the
concept of uniform convergence for a sequence of functions.
Definition 1.42. Suppose f and f1 , f2 , . . . are complex-valued
function of either a real or complex variable, with domain R
or C. If K we say that fj f uniformly on K if for every
> 0 there is a real number N such that |fj (z) f (z)| < for all
z K if j N .
As a function in C a polynomial is continuous; this follows easily
since constant polynomials and the polynomial z obviously are continuous, and any other polynomial can be built up from these by multiplications and additions so the continuity follows from the standard
calculation rules for limits.
A rational function is a quotient r(z) = p(z)/q(z) where p and q are
polynomials and q not identically 0 (if q is constant r is a polynomial).
18
It follows that r is continuous as a function in C in all points which

are not zeros of q. We may assume that p and q have no common nonconstant polynomial factors (the common divisor to two polynomials
of largest degree can always be found by a purely algebraic device, the
Euclidean algorithm). Hence p and q have no common zeros. It follows
that r(z) as z tends to any zero of q. As z we have r(z) 0
if deg p < deg q and r(z) if deg p > deg q. If deg p = deg q, then
r(z) a/b where a and b are the highest order coefficients of p and q
respectively.
A power series is a series
X
(1.4)
an (z a)n
n=0
where a, a0 , a1 , a2 , . . . are given complex numbers and z a complex variable. In many respects such series behave like polynomials of infinite
order and that is actually how they were viewed until the end of the
19:th century. The very first question to ask is of course: For which
values of z does the series converge? In order to answer this question
we make the following definition.
Definition 1.43. Let the radius of convergence for (1.4) be
R = sup{r 0 | a0 , a1 r, a2 r2 , . . . is a bounded sequence } .
Then R is either a number 0 or R = .
The explanation for the definition is in the following theorem.
Theorem 1.44. For |z a| > R the series (1.4) diverges and for
|z a| < R it converges absolutely. The convergence is uniform on
every compact subset of |z a| < R.
In order to prove the theorem we need a few results which should
be well known in the context of functions of a real variable.
Theorem 1.45. An absolutely convergent complex series is convergent.
Proof. For any complex number z we
P have | Re z| |z| and
| Im z| |z| | Re z| + | Im P
z|. Hence, if P |an | is convergent, then
by comparison the real series
Re an and
Im an are absolutely convergent, to x and y say. The theorem now follows from
|
N
X
n=0
an x iy| |
N
X
n=0
Re an x| + |
N
X
Im an y| 0 as N .
n=0
The next theorem is the complex version of what is usually known

under the silly name of Weierstrass M-test.
19
Theorem 1.46. Let A be a subset of C and f1 , f2 , . . . a sequence of

complex functions defined
Pon A and such that |fn (z)|
P an for all z A
and n = 1, 2, . . . . If
a
converges,
then
n=0 n
n=0 fn (z) converges
uniformly in A.
P
fn (z) converges absolutely
Proof. By Theorem 1.45 the series
for every z A; call the sum s(z). Then
|s(z)
N
X
n=0
fn (z)| = |
X
n=N +1
fn (z)|
X
n=N +1
|fn (z)|
an .
n=N +1
The last member does not depend on z and tends to 0 as N . The

theorem follows.
Proof of Theorem 1.44. If |z a| > R then an (z a)n , n =

0, 1, 2, . . . is an unbounded sequence and hence can not converge to 0.
Hence the power series diverges.
If r < R, then there exists > r such that an n , n = 0, 1, 2, . . . is
a bounded sequence; let C be a bound. Then if |z a| r we have
|an (z a)n | |an |rn = |an n |(r/)n C(r/)n . Since a geometric
series with quotient 0 r/ < 1 is convergent, the theorem follows
from Theorem 1.46 (any compact subset of |z a| < R is a subset of
|z a| r for some r < R).
Here is the complex version of another well known theorem.

Theorem 1.47. Suppose f1 , f2 , . . . is a sequence of continuous,
complex functions converging uniformly to f on the set M . Then f
is continuous on M .
The proof is word for word the same as in the case of real functions so we will not repeat it here. We have the following corollary of
Theorems 1.44 and 1.47.
Corollary 1.48. If R is the radius of convergence of (1.4), then
(1.4) is a continuous function of z for |z a| < R.
Proof. The partial sums of a power series are polynomials and
therefore continuous. Since any z in the disk |z a| < R is an interior point of a compact subset of the disk the claim follows from
Theorems 1.44 and 1.47.
So far we have said nothing about convergence on the boundary of

the circle of convergence. There is a good reason for this; nothing much
can be said in general. One can have divergence at every point of the
circle, convergence at some points and divergence at others or one can
have absolute convergence at every
P point ofn 2the circle. A general result
by Carleson (1966) says that if
n=0 |an R | converges, then (1.4) will
converge almost everywhere on the circle, in the sense of Lebesgue
integration. On the other hand, there are examples (the first one given
20
by Kolmogorov in 1926) for which an Rn 0 such that (1.4) diverges

for every point on the circle.
P n
Exercise 1.49. Show that
z diverges at every point of its
P n=0
n
circle of convergence, that n=0 z /n converges
but not all
P forn some
2
points on its circle of convergence and that
z
/n
converges
abn=0
solutely for all points on its circle of convergence.
It is often possible to find the radius of convergence for a given
power series by inspection and use of the definition. As an aid in cases
where this might be difficult we have the following two theorems.
Theorem 1.50. lim |an |1/n = 1/R. This is to be interpreted by
n
using the conventions 1/0 = and 1/ = 0.

Here we have defined lim cn = lim supn cn = limn supkn ck
n
for a real sequence c0 , c1 , . . . .
Proof. Let L = lim |an |1/n . If r < 1/L, then |an |1/n < 1/r for all
n
sufficiently large n. Hence |an rn | < 1 for such n, so the sequence an rn ,

n = 0, 1, 2, . . . is bounded. Hence 1/L R.
If r > 1/L, then there exists , r > > 1/L, so that |an |1/n >
1/ for infinitely many n. Hence |an rn | = |an n |(r/)n > (r/)n for
infinitely many n. Since (r/)n the sequence an rn , n = 0, 1, 2, . . .
can not be bounded and so 1/L R and the proof is complete (check
the cases L = 0 and L = separately).
Theorem 1.51. If limn |an /an+1 | exists it is equal to R.

Proof. Let L = limn |an /an+1 |. If r < L, then r < |an /an+1 |
for all sufficiently large n. It follows that |an+1 rn+1 | < |an rn | for such
n so that an rn , n = 0, 1, 2, . . . , being of eventually decreasing absolute
value, is bounded. Hence L R.
On the other hand, if r > L, choose so that r > > L. Then
there exists N such that > |an /an+1 | for n N . It follows for
such n that |an+1 n+1 | > |an n | > > |aN N |. We therefore have
|an rn | = |an n |(r/)n > |aN N |(r/)n as n (we may assume
aN 6= 0 since this must be true eventually for limn |an /an+1 | to
exist). It follows that L R so the theorem is proved.
21
Exercise 1.52. Find the radius of convergence for the following

power series:
X
X
2 2
(a)
nz
(b)
nn z n
(c)
n=0
X
n=0
z
(n!)2
X
zn
(e)
ln n
n=2
(g)
(i)
(k)
X
n=1
X
n=0
(d)
z
arctan n
(h)
cos(n/4)z n
(j)
(l)
n=1
(s)
(u)
(w)
(y)
()
n=0
X
n=1
n=0
X
n=1
(n)
(n2 + 2n )z n
2n + 2n n
z
n
(2n + (2)n + 1)z n
2n
zn
zn
1 + 2 + + n
zn
(r)
(t)
X
n=1
n=0
n=1
n=1
(v)
n=0
(sin n)z n
n=0
(arctan n1 )z n
n=1

X
2n n
z
n
n n
32n+1
zn
X
2n
z 2n
(p)
n
n=1
X
zn
n=0
zn
X
(n + 2)3
n=0
X
21/n
n=1
()
n2
n=0
n=0
(q)
X
2n
n=1
n1/n z n
2n z n
n=0
(f)
(m)
( n2 + 1 n2 1)z n
(o)
n=0
(1 + 21 + + n1 )z n
2
zn
n2
z 4n
2n + n2
X
nn
zn
n!
n=1

X
a n
(z)
z
n
n=0
(x)
()
X
(3n + 1)!
n=0
(h)
X
n=1
(n!)4
2
zn
(1 + n1 )n z n
CHAPTER 2
Analytic functions
2.1. Conformal mappings and analyticity
Definition 2.1. A map f : C, where is an open subset of
C, is called conformal if it satisfies the following conditions:
(1) As a map from a subset of R2 into R2 , f is differentiable.
(2) f preserves angles of intersection between smooth curves.
(3) f preserves orientation in the sense that the determinant of
the total derivative of the map is > 0.
To explain the definition in more detail, note that if z = x + iy,
where x and y are real, then f (z) = u(x, y) + iv(x, y) where u and v are
real-valued functions of two real variables,
so the action of f can also be

u(x,y)
x
described by the mapping ( y ) 7 v(x,y) . The first condition of the
definition then says that this map should be differentiable. Recall that
this implies that the partial derivatives ux , uy , vx and vy exist and that
the chain rule can be applied when composing with other differentiable
maps. Also recall that the existence of the partials is not enough to
guarantee differentiability, but if the partials are continuous, then the
map is differentiable.
We measure the angle between two non-zero vectors and Rn
h,i
, where h, i is the usual scalar product and
by the expression kkkk
k k the Euclidean norm (the actual angle is arccos of this). If t 7
(t) = 1 (t) + i2 (t) is a differentiable curve
in
, then its tangent
10
0
vector is or, expressed as a column vector, 0 . The image f of
2
under f is another differentiable
curve.
According
to the chain rule
0
1
ux uy
its tangent vector is J 0 where J = ( vx vy ) is the Jacobi matrix or
2
total derivative of the map. The second point of the definition then
means that the linear map given by the Jacobi matrix maps any two
vectors onto two vectors which make the same angle as the original
u u
vectors. The third point simply means that the Jacobian | vxx vyy | =
ux vy uy vx 0 in .
Exercise 2.2. Show that the map z 7 z satisfies the two first
points of Definition 2.1, but reverses the orientation (i.e., the Jacobian is < 0). Such a map is called anti-conformal. Show that any
anti-conformal map is of the form z 7 f (z) where f is conformal.
23
24
2. ANALYTIC FUNCTIONS
This shows that there is really no need to study anti-conformal maps

separately from conformal maps.
We have the following basic theorem.
Theorem 2.3. Suppose f = u + iv is conformal in . Then the
partials of u and v satisfy the Cauchy-Riemann equations
(
ux = v y ,
(2.1)
uy = vx .
Conversely, if ( uv ) satisfy the Cauchy-Riemann equations, the corresponding map is differentiable and its Jacobi matrix does not vanish at
any point in , then the map is conformal.
The slight asymmetry in the statement of Theorem 2.3 will be removed later; in fact, we will later show (see the discussion after Corollary 4.23) that the Jacobi matrix can not vanish in any domain where
the map is conformal.
Proof. Suppose f is conformal and let and be the column
vectors in the Jacobi matrix. Since multiplication by the Jacobi matrix
preserves angles the vectors ( 10 ) and ( 01 ) are mapped onto orthogonal
vectors, i.e., and are orthogonal. Similarly, the vectors ( 11 ) and
1
( 1
) are mapped onto orthogonal vectors. Since the scalar product of
2
+ and is kk2 kk
same
vy it follows that and also have the
ux
length, so that ( vx ) = uy . The Jacobian is therefore (u2x + u2y ).
To preserve orientation we must choose the plus sign. It follows that
any conformal map satisfies the Cauchy-Riemann equations.
Conversely, if the map satisfies the Cauchy-Riemann equations, is
differentiable,
and has non-vanishing Jacobi matrix, then this matrix
p 2
2
is ux + uy O where O is an orthogonal matrix with determinant one,
i.e., a rotation. The map is therefore conformal.
Exercise 2.4. Show that the map z 7 z 2 is conformal in any open

set not containing the origin.
We will now connect the geometric notion of a conformal map with
the analytic notion of complex derivative. We first need a definition.
Definition 2.5. A complex-valued function f defined in an open
subset of C is said to be differentiable at a if
f (z) f (a)
lim
za
za
exists. The limit is called the derivative of f at a and is denoted by
f 0 (a).
All the elementary properties of derivatives that we know from the
theory of a real function of one variable continue to hold, with essentially the same proofs. We collect some such properties in the next
theorem.
2.1. CONFORMAL MAPPINGS AND ANALYTICITY
25
Theorem 2.6. Suppose that f is differentiable at a. Then

(1) f is continuous at a.
(2) Cf is differentiable at a with derivative Cf 0 (a) for any constant C.
(3) If g is differentiable at a, then so is f + g, f g and, if g(a) 6= 0,
f /g and
(f + g)0 (a) = f 0 (a) + g 0 (a)
(f g)0 (a) = f 0 (a)g(a) + f (a)g 0 (a)
(f /g)0 (a) = (f 0 (a)g(a) f (a)g 0 (a))/g(a)2
(4) If g is differentiable at f (a), then g f is differentiable at a
and the chain rule (g f )0 (a) = g 0 (f (a))f 0 (a) is valid.
(5) If f 0 (a) 6= 0 and the inverse f 1 is defined in a neighborhood
of f (a) and is continuous at b = f (a), then the inverse is
differentiable at b and (f 1 )0 (b)) = 1/f 0 (a).
(6) Polynomials and rational functions are differentiable where
they are defined (as functions in C) and their derivatives are
calculated in the same way as in the case of real polynomials
and rational functions.
Exercise 2.7. Prove Theorem 2.6.
Exercise 2.8. Show that any branch of
z 6= 0 and calculate the derivative.
z is differentiable for
We will later prove (see Corollary 4.23) that if f is differentiable

in a neighborhood of a, then the assumption f 0 (a) 6= 0 implies all the
other assumptions of Theorem 2.6 (5). We are now ready to state the
second main result of this section.
Theorem 2.9. f = u + iv has a complex derivative at z = a + ib
if and only the map ( xy ) 7 u(x,y)
is differentiable and the Cauchyv(x,y)
Riemann equations (2.1) are satisfied at (a, b). We also have f 0 (z) =
ux (x, y) + ivx (x, y) = vy0 (x, y) iu0y (x, y).
Proof. For f to be differentiable with derivative a+ib at z = x+iy
means that
|f (z + w) f (z) (a + ib)w| = |w|r(w) where r(w) 0 as w 0 .
for ( uv ) to be differentiable at (x, y) with a Jacobi matrix
Similarly,
a b
b a satisfying the Cauchy-Riemann equations means that

u(x + h, y + k) u(x, y)
a
b
h
= k(h, k)kk(h, k)k
v(x + h, y + k) v(x, y)
b a
k
where (h, k) 0 as (h, k) 0. But if f = u + iv and w = h + ik
the left hand sides of these two relations are equal so the theorem
follows.
26
If g is a complex-valued function of a real variable with real and

imaginary parts u and v respectively, we say that g is differentiable if u
and v are, and define g 0 = u0 +iv 0 . Using the equivalence in Theorem 2.9
it then follows from the chain rule for vector-valued functions of several
variables that if g is a complex-valued, differentiable function of one
variable with range in the domain of a complex differentiable function
f , then the chain rule dtd f (g(t)) = f 0 (g(t))g 0 (t) is valid.
There are some alternative ways of expressing the Cauchy-Riemann
equations which are sometimes used. If we view f as a function of
x = Re z and y = Im z it is clear that the Cauchy-Riemann equations
are equivalent to fx0 + ify0 = 0. Note also that this means that if the
complex derivative f 0 exists, then f 0 = fx0 = ify0 .
The differential of f as a function of (x, y) is df = fx0 dx + fy0 dy, in
particular dz = dx + idy and dz = dx idy. We can therefore write
df = 12 (fx0 ify0 )dz + 12 (fx0 +ify0 )dz and for this reason one introduces the
= 12 (fx0 ify0 ) and f
= 12 (fx0 + ify0 ). The Cauchy-Riemann
notation f
z
z
= 0, and then f
= f 0.
equations may then be expressed as f
z
z
We also have df = f
dz + f
dz, so if we introduce the holomorphic
z
z
f
differential by f = z dz and the anti-holomorphic differential
by f = f
dz we have d = + , and the Cauchy-Riemann equations
z
may also be written as f = 0. An analytic function is therefore a
solution of the homogeneous equation (pronounced d-bar equation).
This also means that f is analytic if df = f = f
dz = f 0 (z) dz.
z
Definition 2.10. A function f : M C where M C is called
analytic in M if it is defined and differentiable in some open set containing M .
Note that to say that f is analytic in a means more than just having
a derivative in a; f has to be differentiable in a whole neighborhood of
a. A function which is analytic in an open set is often said to be
holomorphic in , and the set of functions which are holomorphic in
is often denoted H().
Practically always the only domains of analyticity that are of interest are connected. There are two notions of connectivity in common use,
arcwise connectivity which is used in calculus, and the more general
notion of connectivity from topology. Since we are always considering
open domains, it makes no difference whether you use one or the other,
since they are equivalent for open sets. For convenience, we will use
the word region to denote an open, connected subset of the complex
plane (or, occasionally, of the Riemann sphere).
We end by a simple result that we will use in the next section.
Theorem 2.11. Suppose f is analytic in a region and that f 0 (z) =
0 for all z . Then f is constant in .
2.2. ANALYTICITY OF POWER SERIES; ELEMENTARY FUNCTIONS
27
We will actually prove much stronger results later; in fact it will be

enough to assume that the zeros of f 0 has a point of accumulation in
for the conclusion to be valid.
Proof. If z and w C is sufficiently close to z, then the line
segment between z and w is entirely in . For 0 t 1 we then obtain
d
f (z + t(w z)) = f 0 (z + t(w z))(w z) = 0 using the remark after
dt
the proof of Theorem 2.9. Thus Re f and Im f are constant on the line
segment. In particular, f (z) = f (w) so that f is locally constant. Now
pick a and let A = {z | f (z) = f (a)}. Then A is open by
what we just saw. But also \ A is open for the same reason. Since
a A we have A 6= . Since is connected we therefore must have
\ A = , i.e., A = . In other words, f is constant.
2.2. Analyticity of power series; elementary functions

We will first continue the study of power series begun in Chapter 1.6. First of all, if a power series really behaves like a polynomial
of infinite order, then we should be able to differentiate the series like
a finite sum, i.e., term by term, and actually obtain the derivative of
the sum of the series.
In order to prove this, we first note that the usual derivative and
integral of a function of one variable extends to the case of a complexvalued function of a real variable in an obvious manner. If f is such
a function, with real and imaginary
parts
Rb
R b u and v, we
R b simply define
0
0
0
f (t) = u (t) + iv (t) and a f (t) dt = a u(t) dt + i a v(t) dt. This
means that we define f as differentiable respectively integrable if its
real and imaginary parts have these properties.
immediately follows that the fundamental theorem of calculus
R It
t
d
f = f (t) holds also for complex-valued, continuous functions f .
dt a
It is also more or less obvious that the usual calculation rules for derivatives and integrals continue to hold. In particular, the integral is interval additive and linear, so that
Zb
f+
f=
a
Zb
Zc
Zb
Zb
(f + g) =
Zb
f +
g
a
for a < c < b and arbitrary constants and if f and g are both
integrable on [a, b]. We also have the triangle inequality
(2.2)
Zb
Zb
f (t) dt |f (t)| dt.

a
28
This is less obvious, but follows from

Zb
Zb Zb
Re ei f = Re(ei f ) |f |
a
Rb
by choosing = arg( a f ).
As already mentioned we also have the chain rule dtd f (g(t)) =
0
f (g(t))g 0 (t) if f is analytic and g is a differentiable complex-valued
function of a real variable. Thus, if f is analytic in a region containing
the line segment connecting z and z + h, then dtd f (z + th) = hf 0 (z + th)
R1
so that h1 (f (z + h) f (z)) = 0 f 0 (z + th) dt if the derivative is continuous. An immediately consequence is the following lemma.
Lemma 2.12. Suppose f is analytic with continuous derivative in
a compact set K containing the line segment connecting z and z + h
where h 6= 0. Then we have | h1 (f (z + h) f (z)| supK |f 0 |.
Proof. By the triangle inequality we obtain
f (z + h) f (z) Z1
Z1
= f 0 (z + th) dt |f 0 (z + th)| dt sup |f 0 |.

h
K
0
Exercise 2.13. Prove the theorem without assuming f 0 to be continuous.

Hint: Use the mean value theorem on Re(ei f (z + th)).
We can now state our theorem about differentiating power series.
P
k
Theorem 2.14. If the series f (z) =
k=0 ak (z a) has convergence radius R, then f has derivatives of all orders for |z a| < R.
The derivatives are calculated by term by term differentiation, and
the resulting
radius of convergence R. In particular,
P series all have
0
k1
f (z) = k=1 kak (z a) .
Proof. We will prove the statement for the first derivative. The
statement for the higher derivatives then follows immediately. Clearly
X
g(z) =
kak (z a)k1
k=1
P
k
has
the
same
radius
of
convergence
as
k=1 kak (z a) and since
k
k 1 as k , it follows from Theorem 1.50 that g has the same
radius of convergence as
Pf .
k1
converges, and
If r < R the series
k=1 k|ak |r
X
1
ak
(f (z + h) f (z)) =
((z + h a)k (z a)k ).
h
h
k=1
29
Now fix z, |z a| < r. Then the terms of this series are continuous
functions of h, with value kak (z a)k1 for h = 0. By Lemma 2.12
the terms have absolute value less than k|ak |rk1 if |z a| < r and
|z + h a| r, so according to Theorems 1.46 and 1.47 the sum is
a continuous function of h in |z + h a| r. For h 6= 0 its value
is h1 (f (z + h) f (z)) and for h = 0 the value is g(z). Thus f is
differentiable and f 0 (z) = g(z) for any z satisfying |z a| < R.
We will use Theorem 2.14 to introduce some more elementary funcP

zk
tions. It is clear that the series
k=0 k! converges for all z so that the
following definition is meaningful.
Definition 2.15. For any z C, let
P
zk
(1) ez =
k=0 k! ,
iz
iz
(2) sin z = e e
,
2i
eiz +eiz
(3) cos z =
.
2
These are all analytic functions in the whole plane. Such a function
is called entire. From the definition follows immediately that e0 = 1,
d z
d
sin 0 = 0 and cos 0 = 1. Furthermore, dz
e = ez , dz
cos z = sin z
d
and dz sin z = cos z. It also follows that sin is odd (sin(z) = sin z)
and cos even (cos(z) = cos z) and that we have the power series
P
P
(1)k 2k+1
(1)k 2k
expansions sin z =
and cos z =
k=0 (2k+1)! z
k=0 (2k)! z .
Theorem 2.16. The functions of Definition 2.15 satisfy the following functional equations:
(1) ez+w = ez ew , for any complex numbers z and w.
(2) sin(z + w) = sin z cos w + cos z sin w,
(3) cos(z + w) = cos z cos w sin z sin w
for any complex numbers z and w.
Note that the particular case w = z of (1) shows that ez ez = 1
so that ez 6= 0 for all z C.
Proof. Given w C, let f (z) = ez ez+w . This is an entire function with derivative f 0 (z) = ez ez+w + ez ez+w = 0 so it is constant
by Theorem 2.11. Setting z = 0 we obtain ez ez+w = ew for all z and
w. The special case w = 0 shows that ez ez = 1 so that ez = 1/ez .
The first formula now follows; the other formulas follow immediately
from this by inserting the definitions of sin and cos in the formulas to
the right.
Theorem 2.17. We have |ez | = eRe z . There exists a smallest real

number > 0 such that sin = 0, and this number satisfies 2.8 < <
3.2. The exponential function has period 2i, the functions sin and cos
period 2. All other periods are integer multiples of these.
30
Proof. First note that ex+iy = ex (cos y + i sin y) by Theorem 2.16.

Since the coefficients in the power series for ez are all real it follows
that as a function of a real variable, ex is real-valued. It is also > 0,
since it is continuous, never = 0 and e0 = 1 > 0. Since it is also its own
derivative it follows that it is strictly increasing (and strictly convex).
For similar reasons cos and sin are real-valued for real arguments and
since
(2.3)
cos2 z + sin2 z = 1
(take w = z in Theorem 2.16 (3)) it follows that cos y + i sin y is a

point on the unit circle for y R. Hence |ez | = eRe z .
We next note that if a real, continuous and non-constant function
is periodic, then all its periods are integer multiples of its smallest
positive period. First of all, since y is a period if and only if y is,
there are positive periods if there are any. Next, if there are arbitrarily
small periods > 0, then given x and > 0 we can find a period a,
0 < a < and an integer p such that |ap x| < . Now f (0) = f (ap)
and by continuity f (ap) f (x) as 0, so f is constant. Also
note that the set of periods of f is closed, since if yj y and all yj
are periods, then f (y + x) = lim f (yj + x) = f (x), so that also y is a
period. If f is non-constant it therefore has a smallest positive period
a, and if b is another period, then for any integer q, b aq is a period.
But if q is the integer quotient and r the remainder when dividing b
by a, then 0 r = b aq < a. So, a can not be the smallest positive
period unless r = 0.
If now w is a period for the exponential function so that ez = ez+w
for all z, we see that this is equivalent to ew = 1. Taking absolute values
it follows that Re w = 0. Setting w = iy we see that y is a real period
for both sin and cos. Note that neither of these functions can have
non-real periods since we immediately obtain from Theorem 2.16 (2)
respectively 2.16 (3) that w is a period of either of these functions if
and only if sin w = 0 and cos w = 1. By (2.3) the first of these relations
follows from the second, which may be rewritten as (eiw 1)2 = 0. This
is true if and only if iw is a period for ez . Therefore, sin and cos have
the same periods, they are real and y is the smallest positive period
of the trigonometric functions if and only if it is the smallest positive
number for which cos y = 1.
Now cos y = 1 2 sin2 y2 according to Theorem 2.16 (3) and (2.3).
It follows that y is the smallest positive number for which cos y = 1
if and only if y2 is the smallest positive zero of sin. According to (2.3)
we must then have cos y2 = 1 and there can be no smaller positive
numbers for which cos takes the value 1. Now cos y2 = 2 cos2 y4 1 so
that y2 has this property if and only if y4 is the smallest positive zero
for cos. It now only remains to show that cos actually has a smallest
positive zero, and to estimate its value.
31
Since cos is continuous the set of its non-negative zeros is a closed

set and therefore has a smallest element if it is non-empty. By (2.3)
we have cos x 1 for real x and integrating this from 0 to x > 0 four
2
3
times we get in turn sin x x, 1 cos x x2 , x sin x x6 and
4
2
x2
1 + cos x x24 . It follows that for x > 0 we have 1 x2 cos x
2
4
2
1 x2 + x24 (this may also be deduced from the fact that the power
series for cos is an alternating series).
first positive zeros of the
p The
two polynomials are 2 > 1.4 and 6 2 3 < 1.6 respectively. It

follows that cos has a smallest positive zero which is in the interval
(1.4, 1.6). The proof is now complete.
One may easily continue to define strictly all the usual (real) functions of elementary calculus and prove all the usual properties of them.
We will assume this done; in particular x is the arclength of the arc of
the unit circle beginning at 1 and ending at eix = cos x + i sin x, so x is
the angle between the rays through these points starting at 0. We will
also use the common properties of the inverse tangent function.
If we want to extend the definition of the logarithm to the complex domain, we should find the inverse of the exponential function.
However, since the exponential function is periodic it has no inverse
unless we restrict its domain appropriately (cf. the definition of the
inverse trigonometric functions). To see how to do this, let us attempt
to calculate the inverse of the exponential function, i.e., to solve the
equation z = ew for a given z.
We first note that we must assume z 6= 0, since the exponential
function never vanishes. Taking absolute values we find that |z| = eRe w
so that Re w = ln |z|, where ln is the usual natural logarithm of a
z
positive real number. We also obtain |z|
= ei Im w . Now cos 0 = 1 and
cos = 1 and since cos is continuous, it takes all values in [1, 1] in
z
the interval [0, ]. Since Re |z|
[1, 1] we can find x [0, ] such that
z
z
cos x = Re |z| . It follows that sin x = Im |z|
. Changing the sign of x
changes the sign on sin x but leaves cos x unchanged. Therefore either
z
eix or eix equals |z|
.
We may therefore solve the equation for w given any z 6= 0. If w1
and w2 are two solutions, it follows that ew1 w2 = 1, so that w1 and
w2 differ by an integer multiple of 2i. We call any permissible value
of Im w an argument for z, and denote any such number by arg z.
We should therefore define log z = ln |z| + i arg z. To get an actual
(single-valued) function, we must make particular choices of arg z for
each z. We shall see later that in order to be able to this and obtain
a continuous function, we can not allow all of C \ {0} in the domain.
Intuitively it is clear that we must choose the domain such that there
are no closed curves in it that go around the origin, since following
such a curve we would have changed the argument continuously by an
32
integer multiple of 2 when we arrive back at the starting point. This

leads to the following concept.
Definition 2.18. A connected subset of the Riemann sphere is
called simply connected if its complement is connected.
If is a region where we want to define a single-valued, continuous
argument function, it must not contain 0 or , and to exclude the
possibility of having a closed curve in that winds around 0, we
should exclude from a connected set containing both 0 and . Now
suppose we have selected a region which is simply connected in C and
does not contain 0, and one of the possible arguments for some point
in . It seems plausible that this should determine a single-valued,
continuous logarithm in . We shall show later (see Theorem 3.19)
that this is the case; we call such a function a branch of the logarithm.
The most important example is obtained when one chooses to
be C with the non-positive part of the real axis removed, and fixes
the argument at 1 to be 0. This is called the principal branch of the
logarithm. The argument of any number in is determined by the
requirement that it is in (, ). The notation Log with a capital L is
sometimes used for this branch.
Another important case is when one instead removes the non-negative real axis and fixes the argument at 1 to be . The argument
is then in the interval (0, 2). Other choices are obtained when one
removes from C any smooth, non self-intersecting curve starting at
0 and ending at . In any case, it is not possible to talk about the
complex logarithm without specifying which branch one is dealing with.
Theorem 2.19. Any branch of the logarithm is analytic with derivative 1/z.
Proof. For any z = x + iy we have log(x + iy) = u(x, y) + iv(x, y)
where u(x, y) = 21 ln(x2 + y 2 ) and v(x, y) = arctan xy + k, where k is
some integer except if x = 0 in which case v(x, y) = 2 arctan xy + k.
By continuity the same value of k has to be used in any sufficiently
small neighborhood of z. Differentiating we therefore get ux (x, y) =
y
y
x
x
, uy (x, y) = x2 +y
2 , vx (x, y) = x2 +y 2 and vy (x, y) = x2 +y 2 so that
x2 +y 2
the Cauchy-Riemann equations are satisfied. Since the partials are all
continuous for (x, y) 6= (0, 0), the function is analytic by Theorem 2.3.
1
The derivative is ux + ivx = xxiy
2 +y 2 = z so the proof is complete.

Exercise 2.20. Prove Theorem 2.19 by use of Theorem 2.6 (5).
We are now able to define arbitrary powers of any complex number
w 6= 0. We set wz = ez log w , where log is some branch of the logarithm,
giving rise to a branch of the power. By varying the branch, there is
therefore in general infinitely many values of the power; e.g., ii = ei log i
and since log i = ln 1 + i arg i the possible values of log i are i( 2 + 2k),
2.3. CONFORMAL MAPPINGS BY ELEMENTARY FUNCTIONS
33
where k is an integer. Hence the possible values of ii are e 2 2k .

There are therefore infinitely many possible values (note that they are
all real!). In some cases the situation is simpler, however; if w is real
> 0 one always uses the principal branch of the logarithm so that wz
for real z is the elementary exponential function with base w.
One can of course also view the exponent as fixed and the base as
the variable; these are the power functions z 7 z . If is an integer
it is clear that the choice of branch for the logarithm is irrelevant; the
function coincides with the elementary concept of a power function. If
n
is rational = m
where m > 0 and n are integers with no common
factors, there are exactly m possible values of z for each z 6= 0; one
usually says that there are m branches. This agrees with our discussion
of the square root in Chapter 1.3. If is irrational or non-real, however, there are always infinitely many branches of the power. Different
powers are said to be of the same branch if they are defined through
the same branch of the logarithm.
Theorem 2.21. Any branch of z is analytic (in its domain) with
derivative z 1 , using the same branch.
Proof.
d log z
e
dz
= z e log z = e(1) log z .
If is real and > 0 the power function is also defined for z = 0,

with value 0 and if = 0 the power function is the constant 1.
2.3. Conformal mappings by elementary functions
We will here only give some examples of mappings induced by power
functions and by the exponential function and their combinations with
Mbius transforms.
Suppose R. That a branch of w = z is defined in an open set
means that z = e log z for an appropriately chosen branch of the
logarithm. Note that those z for which |z| = r are mapped onto
|w| = r so that circular arcs centered at the origin are mapped onto
(other) circular arcs centered at the origin. Similarly, if z is on a ray
arg z = we have arg w = so rays from the origin are mapped onto
other rays from the origin. Also, angles at the origin are multiplied
by a factor so that the map is certainly not conformal there unless
= 1. This is true even if is an integer so that z is well defined in
the whole plane. Note that the derivative vanishes at 0 then.
These observations show that a wedge domain, bounded by two rays
from the origin making the angle may be mapped onto a half plane
by applying a branch of z where = /. More generally, any region
with a corner at the origin may have this corner straightened out by
applying an appropriate power function. Since any region bounded by
two intersecting circular arcs may be mapped onto a wedge by a Mbius
transform taking the points of intersection to 0 and respectively, any
34
such region may be mapped onto a half plane by composing a Mbius

transform and a power function.
Exercise 2.22. Construct a conformal mapping that takes the region
(
|z + 3| < 10,
|z 2| < 5,
onto the interior of the first quadrant.
Exercise 2.23. Map the region
0 < arg z < / ,
0 < |z| < 1 ,
onto the interior of the unit circle ( 1/2).

Since ez = ex (cos y + i sin y) if z = x + iy it is clear that the
exponential function takes any line parallel to the real axis into a ray
from the origin. Similarly, any vertical line segment is taken into a
circular arc centered at the origin and with angular opening equal to
the length of the line segment. This means that an infinite strip parallel
to the real axis, i.e., a region of the type a < Im z < b, is mapped onto a
wedge domain by the exponential function. A half infinite strip defined
by a < Im z < b, Re z < c is similarly mapped onto a circular sector
centered at the origin.
Exercise 2.24. What is the image of the region 0 < Im z < 2
under the map w = ez ?
Exercise 2.25. Construct a conformal mapping of the region
|z 1| > 1,
|z| < 2,
onto the upper half plane.
Exercise 2.26. Construct a conformal map of the region
/4 < arg z < /4 ,

0 < |z| < R ,
(R > 1)
onto the interior of the unit circle, so that z = 1 is mapped onto the
origin. Calculate the length of the arc of the unit circle which is the
image of the arc
/4 arg z /4 ,
|z| = R .
Exercise 2.27. What is the image of the unit disk under the map
w = F (z) = (z + 1/z)/2, F (0) = .
Hint: Introduce W = w1
and Z = z1
.
w+1
z+1
2.3. CONFORMAL MAPPINGS BY ELEMENTARY FUNCTIONS
35
Exercise 2.28. Map the region a < arg z < b, where 0 < a <
b < 2, conformally onto the complex plane with the positive real axis
removed.
Exercise 2.29. Find a conformal map that takes
|(1 i)z + (1 + i)z| < 2 ,

=
(1 + i)z + (1 i)z > 0 ,
onto
|z 1 i| < 2 ,
0
=
i(z z) > 2 .
Exercise 2.30. Consider the conformal map given by cos z. What
are the images of lines parallel to the real and imaginary axes? What
is the image of the strip < Re z < ?
CHAPTER 3
Integration
3.1. Complex integration
Complex integration is at the core of the deeper facts about analytic
functions. Here we will discuss the basic definitions.
Let be a piecewise differentiable curve in C. This means a complexvalued, continuous function defined on a compact real interval which is
continuously differentiable except at a finite number of points, where
at least the left and right hand limits of the derivative exist. Thus it
is described by an equation z = z(t) where a t b for some real
numbers a and b and z 0 is continuous except for a finite number of jump
discontinuities. For convenience we will in the sequel call such a curve
an arc.
If f is a continuous, complex-valued function of a complex variable
defined on an arc , then the composite function f (z(t)) is continuous
and we make the following definition.
Z
Definition 3.1.
Zb
f (z(t))z 0 (t) dt.
f (z) dz =
If you know
R about line integrals and f = u + iv, z = x + iy you will
realize that f (z) dz is the line integral
Z
Z
u dx v dy + i v dx + u dy,
but we will not use this. It is, however, very important that the
complex integral is independent of the parametrization of the arc .
This means the following. A change of parameter is given by a piecewise differentiable, increasing function t(s) mapping an interval [c, d]
onto [a, b]. The usual change of variables formula then shows that
R
Rd
f (z) dz = c f (z(t(s)))z 0 (t(s))t0 (s) ds. Here z 0 (t(s))t0 (s) is, by the
chain rule, the derivative of z(t(s)), so that the definition of the complex integral gives the same value whether we parametrize by z(t) or
z(t(s)).
Note that the arc has an orientation, in that it begins at z(a) and
ends at z(b). If t(s) is a decreasing piecewise differentiable function,
mapping [c, d] onto [a, b], then the equation z = z(t(s)) will give a
parametrization of the opposite arc to , which we denote by , in
37
38
3. INTEGRATION
that the initial pointRis now z(t(c)) =Rz(b) and the final point z(t(d)) =
z(a). Thus we have f (z) dz = f (z) dz.
It is clear by the definition that the integral is linear in f , and also
that if we divide an arc into two sub-arcs 1 and 2 by splitting
the parameter interval into two subintervals with
no common
interior
R
R
points (keeping the correct orientation), then f (z) dz = 1 f (z) dz +
R
f (z) dz. It is now an obvious step to consider the sum of two (or
2
more) arcs 1 and 2 even if they are not sub-arcs of another arc, and
define the integral over such a sum as the sum of the integrals over the
individual terms. Such a formal sum of arcs is called a chain. Given arcs
1 , . . . , n we may integrate over chains of the form = a1 1 + +an n ,
where the coefficients
a1 , . . .R, an are arbitrary integers, indicating that
R
the integral j f enters in f with the coefficient aj . If aj = 0 the
arc j can of course be left out of .
Note that our notation for the opposite of an arc makes sense, in
that integrating over amounts to integrating over (1). Very often
we will integrate over closed arcs. This means an arc where the initial
and final points coincide. A simple arc is one without self-intersections;
for a closed arc this means no self-intersections apart from the common
initial and final point.
There is also a triangle inequality for complex integrals. From the
definition of integral and the triangle inequality (2.2) it immediately
follows that
Z
Z
f (z) dz |f (z)| |dz|,
Rb
where the last integral is defined by |f (z)| |dz| := a |f (z(t)||z 0 (t)| dt
and is called an integral
R with respect
R b 0 to arc-length. The reason for
this is, of course, that |dz| = a |z (t)| dt gives the length of the arc
. If you dont know this already, you may take it as a definition of
length. Note that a very similar calculation to the one we did earlier
shows that an integral with respect to arc length is independent of the
parametrization, and in this case also of the orientation of the arc.
Example 3.2. Suppose is the circle |z a| = r, oriented by
running through it counter-clockwise. A parametrization is z(t) =
a + reit , 0 t 2. We obtain
z 0 (t) = ireit so that |z 0 (t)| = r. The
R 2
length of the circle is therefore 0 rdt = 2r, as expected.
It is possible to integrate along more general curves than those
that are piece-wise differentiable, so called rectifiable curves. There
is seldom any reason to do this in complex analysis, however. In fact,
when integrating analytic functions the integral is, as we shall see later,
independent of small changes in the path we integrate over, so it is
practically always enough to consider piece-wise differentiable arcs.
3.1. COMPLEX INTEGRATION
39
The elementary way of calculating integrals of a function of a real

variable is by finding a primitive of the integrand. This method can
be used also for complex integrals. Suppose that a continuous function
f has a primitive, i.e., a function F analytic on a continuously differentiable curve such that F 0 = f . Suppose z = z(t), a t b, is a
parametrization of . Then
Z
(3.1)
Zb
f (z) dz =
Zb
0
(F (z(t)))0 dt
f (z(t))z (t) dt =
a
b
= F (z(t)) a = F (z1 ) F (z0 ),
where z0 = z(a) is the initial and z1 = z(b) the final point of . If is

just piece-wise continuously differentiable the same formula holds; one
only has to use (3.1) on each differentiable piece and add the resulting
formulas. The evaluations of the primitive at the intermediate points
will then cancel, and we obtain (3.1) again.
So far the theory of analytic functions closely parallels the theory
of functions of a real variable. This is quite misleading, as we shall
see in the next section. The first indication that the theory of analytic
functions is very different from one-variable real analysis comes when
one asks the question of which functions f of a complex variable have a
primitive. This turns out to require that f is analytic, but not even this
is enough. There are also requirements on the nature of the domain of
f , and these questions are a central theme for the theory of analytic
functions. The starting point is the following theorem.
Theorem 3.3. Suppose f is continuous
in a region . Then f has
Z
a primitive F in if and only if
f (z) dz = 0 for every closed arc
. It is enough if this is true for arcs made up solely of vertical

and horizontal line segments.
Proof. If F is a primitive ofR f in and a closed arc with initial
and final points z1 = z0 , then f (z) dz = F (z1 ) F (z0 ) = 0 since
z1 = z0 .
Conversely, if the integral
R along closed arcs vanishes, pick a point
z0 and define F (z) = f , where is an arc in starting at z0
and ending at z. This gives an unambiguous definition of F , since if
is another such arc, then the arc is a closed arc in . Thus the
integral along has the same value as the integral along . We may
restrict ourselves to arcs of the special type of the statement of the
theorem, since in an open, connected set every pair of points may be
connected by an arc of this kind in (show this as an exercise!).
It now remains to show that F is a primitive of f in .
40
3. INTEGRATION
Writing z = x + iy with real x, y we shall calculate the partial

derivatives of F with respect to x and y. To do this, let h R be
so small that the line segment
between z and z + h is contained in .
Rh
Then F (z + h) F (z) = 0 f (z + t) dt. This is seen by choosing an arc
starting at z0 and ending at z to calculate F (z), and then calculating
F (z + h) by adding to the line segment connecting z to z + h, which
we parametrize by z(t) = z + t, 0 t h.
By the fundamental theorem of calculus, differentiating with respect
to h gives Fx0 (z) = f (z). Similarly, considering F (z + ih) F (z) =
Rh
i 0 f (z + it) dt we obtain Fy0 (z) = if (z). Thus the Cauchy-Riemann
equation Fx0 + iFy0 = 0 is satisfied, and F 0 = Fx0 = f , so that F is a
primitive of f .
3.2. Goursats theorem

In this section we shall begin to explore properties of analytic functions which show them to be very different in nature to differentiable
functions of a real variable.
We first prove a fundamental theorem by Goursat (1905). We then
consider integrals along the boundary of a rectangle. A rectangle is of
course a set defined by inequalities a Re z b, c Im z d, and the
boundary consists of four line segments with endpoints at the points
a + ic, b + ic, b + id and a + id. The boundary is therefore a closed arc,
and we orient it by running through the vertices in the order described,
ending up finally with a + ic again. This means we run through the
boundary in the direction which has the interior of the rectangle to
the left of the boundary. This orientation of the boundary is called
positive.
Theorem 3.4. Suppose f is analytic in a closed rectangle ( i.e., in
an open set containing the rectangle)
Z and let be the positively oriented
boundary of the rectangle. Then
f (z) dz = 0.
Proof. Let R be the rectangle and I be the value of the integral.

Now divide R into four congruent rectangles by one horizontal and one
vertical cut, and let the integrals over the positively oriented boundaries
of the sub-rectangles be I j , j = 1, . . . , 4. A common side to two of the
rectangles will then be given opposite orientation in the corresponding
integrals. It follows that I = I 1 + I 2 + I 3 + I 4 , since the contributions
from integrating over the cuts will cancel. Thus the absolute value
of at least one of the I j will be |I|/4. Let R1 be a corresponding
sub-rectangle and I1 the associated integral, so that |I| 4|I1 |. We
can now repeat the process with R1 , and then repeat this process indefinitely. We obtain a nested1 sequence R1 , R2 , . . . of rectangles and
1i.e.,
each rectangle is contained in the previous one
3.2. GOURSATS THEOREM
41
a corresponding sequence I1 , I2 , . . . of integrals such that |I| 4n |In |

for n = 1, 2, . . . .
The sequences of lower left corner real and imaginary parts in Rn
are both increasing, because the rectangles are nested, and bounded
from above, because all rectangles are contained in R. It follows that
the sequence of lower left corners converge to a point w R. Let d be
the diameter of R, i.e., the length of the diagonal. Then it is clear that
the diameter of Rn is dn = 2n d, so that given any neighborhood of w,
Rn will be contained in this neighborhood for all sufficiently large n.
(w)
Now f is differentiable at w, so that | f (z)f
f 0 (w)| < if z
zw
is sufficiently close to w. Denoting the expression inside the absolute
value signs by (z) we obtain f (z) = f (w) + f 0 (w)(z w) + (z)(z w),
where |(z)| < if z is in a sufficiently small neighborhood of w. Choose
n so large that Rn is contained in such a neighborhood, and let n be
the positively oriented boundary of Rn , so that
Z
In = f (z) dz
n
Z
=
Z
f (w) dz +
Z
0
f (w)(z w) dz +
n
(z)(z w) dz.
n
Now, the constant f (w) has primitive f (w)z and the first order polynomial f 0 (w)(z w) has primitive 21 f 0 (w)(z w)2 , so that the first two
integrals in the second line vanish. The third integral is estimated as
follows:
Z
Z
(z)(z w) dz |z w||dz| dn Ln ,
n
where Ln is the length of n and dn as before is the diameter of Rn .

The estimate follows by the triangle inequality and since |z w| dn ,
both of z and w being in Rn . However, we have dn = 2n d, and it is
equally clear that Ln = 2n L, where L is the length of the boundary
of R. Thus we have |I| 4n |In | dL. Since > 0 is arbitrary, it
follows that I = 0.
We will also need a slight extension of Goursats theorem.

Corollary 3.5. Suppose f is analytic in a closed rectangle R except for at an interior point p, where (z p)fZ(z) 0 as z p. If
is the positively oriented boundary of R, then
f (z) dz = 0.
Proof. Let > 0 and R0 R be a square centered at p and so

small that |(z p)f (z)| < for z R0 . If 0 is the positively oriented
42
3. INTEGRATION
boundary of R0 we obtain
Z
Z
f (z) dz
0
|dz|
8.
|z p|
The last inequality is due to the facts that |z p| `/2 if ` is the side
length of R0 , and that the length of 0 is 4`.
Now extend the sides of R0 until they cut R into 9 rectangles, one
of which is R0 . RThe other
R 8 satisfy the assumptions of Theorem 3.4.
It follows that | f | = | 0 f | 8, and since > 0 is arbitrary the
integral over must be 0.
We can now prove a first version of a fundamental theorem known

as the Cauchy integral theorem.
Corollary 3.6 (Cauchys integral theorem for a disk). Suppose
f is analytic in an open disk D, except possibly at a point p where
(z p)f (z) 0 as z p. Then f has a primitive
in D \ {p}, and for
Z
every closed curve in D \ {p} we then have
f (z) dz = 0 .
Proof. In view of Theorem 3.3 it is enough to show that f has a

primitive in D \ {p}.
Let z0 be a fixed point in D with both Re z0 6= Re p and Im z0 6=
Im p. We may also assume that the center of D has both real and
imaginary parts closer to those of p than to those of z0 . Let z 6= p be
another point in D.
Suppose first that the boundary of the rectangle with opposite corners at z0 and z is in D and does not contain p. We then define F (z) as
the integral of f along first the horizontal side of the rectangle starting
at z0 , and then the vertical side ending at z. It is clear, reasoning as
in the proof of Theorem 3.3, that Fy0 (z) = if (z). However, by Corollary 3.5 F (z) will have the same value if we first integrate along the
vertical side starting at z0 and then along the horizontal side ending
at z, and with this definition we see that Fx0 (z) = f (z), so that F is a
primitive of f wherever it is defined.
It remains to define F at points for which p is on the boundary of
the rectangle, or one of the corners of the rectangle is outside D. Then
first note that we could have started our path of integration by first
moving vertically, then horizontally and finally vertically again until
we reach z, and the horizontal path may be chosen anywhere between
Im z and Im z0 , as long as the path stays in D and doesnt contain p.
This change will not affect the value of F in view of Corollary 3.5.
Suppose now that either p is on the horizontal side ending at z,
or else that the other endpoint of this side is outside D. We then
define F (z) just as before and obtain Fy0 (z) = if (z). However, when
calculating Fx0 we modify our path by first following the horizontal side
3.2. GOURSATS THEOREM
43
starting at z0 some distance, then following a vertical path until we

reach the horizontal side ending at z, and then following this side until
we reach z. This can be done so that the path is inside D and does
not contain p. The value of the integral will again equal F (z) because
of Corollary 3.5, and we get just as before that Fx0 (z) = f (z).
It is clear that a similar construction will work if p is on the vertical
side ending at z, or if this side is not in D. Thus F is a well defined
analytic function in D \ {p} with derivative f . You should draw a
picture of the various cases and convince yourself that the construction
will give an unambiguous definition of F !
The conclusion of Corollary 3.6 can not be drawn with weaker assumptions on f at the point p. To illustrate this, let f (z) = 1/(z p)
which is analytic in any disk centered at p, except at z = p, and let be
the positively oriented boundary of such a circle. We may parametrize
by z(t) = p + reit , 0 t 2. Then z 0 (t) = ireit so that
Z2 it
Z2
Z
ire dt
dz
=
= i dt = 2i 6= 0.
(3.2)
zp
reit
This example is actually more crucial than is immediately obvious, and

we use it as the basis for the notion of index or winding number of a
point with respect to a closed arc.
Definition 3.7.
(1) A cycle is a chain (a formal sum of arcs) which may be written
as a sum of finitely many closed arcs.
(2) The index of a point p
/ with respect to a cycle is
Z
1
dz
n(, p) =
.
2i
zp
Note that the range of is a compact set, being finite union of continuous images of the compact parameter intervals, so its complement
is open. An open set may be split into open, connected components 2.
Clearly there is precisely one unbounded component in the complement
of .
Lemma 3.8. The index has the following properties.
(1) n(, p) is always an integer.
(2) n(, p) = n(, p).
(3) n(1 + 2 , p) = n(1 , p) + n(2 , p) if 1 and 2 are both cycles
not containing p.
(4) n(, p) is constant as a function of p in any connected component of the complement of the range of .
2This
is true whether you think of connected as meaning arcwise connected, or

use the more general concept of connectedness introduced in topology.
44
3. INTEGRATION
(5) n(, p) = 0 for all p in the unbounded component of the complement of the range of .
Proof. Let z = z(t), a t b, be a parametrization of a closed
R t 0 (s) ds
for t [a, b]. Then g(b) = 2i n(, p) and
arc and set g(t) = a zz(s)p
0
0
g (t) = z (t)/(z(t) p) so that the derivative of h(t) = eg(t) (z(t) p)
is identically 0. We have h(a) = z(a) p, so h is constant equal to
z(a) p. For t = b we obtain eg(b) (z(b) p) = z(a) p. Since is a
closed arc we have z(b) = z(a) 6= p so that eg(b) = 1. Thus g(b) is an
integer multiple of 2i. Since a finite sum of integers is an integer this
proves (1).
(2) and (3) are obvious from the definition, and it is also obvious that n(, p) depends continuously on p
/ (give detailed reasons
yourself!). But a continuous, real-valued function in a region assumes
intermediate values, so since the index is integer-valued, (4) follows.
Finally, it is clear that n(, p) 0 as p , and since n(, p) is
independent of p for p in the unbounded component of the complement
of , (5) follows.
A circle has a complement consisting of exactly two components,

and since we saw in (3.2) that the index of the center of a positively
oriented circle is 1, all other points in the open disk will also have index
1 with respect to the boundary circle.
Theorem 3.9 (Cauchys integral formula). Suppose f is analytic
in an open set D for which the conclusion of Corollary 3.6 is correct,
and that is a cycle in D. Then, if p
/ ,
Z
1
f (z) dz
n(, p)f (p) =
.
2i
zp
In particular, this is true if D is a disk.

(p)
Proof. Put g(z) = f (z)f
. Then g is analytic in D \ {p} and
zp
(z p)g(z)
= f (z) f (p) 0 as z p since f is continuous at p.
R
Thus g(z) dz = 0 by Corollary 3.6. But by the definition of g this
R (z)
R dz
means that fzp
dz = f (p) zp
. The theorem follows.
For the special

R f (z) case when is a positively oriented circle we obtain
1
f (p) = 2i zp dz when p is inside the circle. When p is outside
the circle, the integral equals 0. The situation for analytic functions is
therefore radically different than for differentiable functions of a real
variable, since Cauchys integral formula shows that if you know the
values of an analytic function on a circle, then all the values inside the
circle are determined. We shall see many more instances of how the
behavior of an analytic function in one place determines the behavior
in other locations.
3.3. LOCAL PROPERTIES OF ANALYTIC FUNCTIONS
45
Note that so far we only know the conclusion of the theorem in

the case when D is a disk. More general regions will be discussed in
Section 3.4.
3.3. Local properties of analytic functions
We start with a useful result about analytic dependence on a parameter in certain integrals.
Lemma 3.10. Suppose f is continuous on a circle with equation
|z p| = r. Then the function
Z
1
f ()
(3.3)
g(z) =
d
2i
z
is analytic in the corresponding open disk |z p|P

< r. In fact, we may
k
expand the function in a power series g(z) =
k=0 ak (z p) with
radius of convergence Rat least equal to r. The coefficients in the series
f (z)
1
are given by ak = 2i
dz.
(zp)k+1
zp
Proof. The denominator in the integral is z = ( p)(1 p
).
The reciprocal of this is the sum of a convergent geometric series since
| zp
| < 1, z being closer to p than . A partial sum of this series has
p
the sum
n1
X
k=0
( p)
z p k
p
= ( p)
1 ( zp
)n
p
1
zp
p
1
(z p)n
.
z ( p)n ( z)
Solving for 1/( z) and inserting in (3.3) we obtain

Z
n1
X
(z p)n
f () d
k
ak (z p) +
g(z) =
,
2i
( p)n ( z)
k=0
where ak are given in the statement of the theorem. We can estimate

the absolute value of the last term by
z p n 1 Z |f ()|
|d|
r
2
| z|
which obviously tends to 0 as n since |z p| < r. The lemma

follows.
Essentially all results about the local behavior of analytic functions,

i.e., properties valid in a neighborhood of a point of analyticity, can be
deduced from the following theorem, which is an easy consequence of
Lemma 3.10.
46
3. INTEGRATION
Theorem 3.11. Suppose f is analytic in a disk |z p| < R. Then

f has derivatives
of all orders and one may expand f in a power series
P
f (z) =
a
(z
p)k , with radius of convergence at least equal to R.
k=0 k
R
f (z)
1
We have ak = f (k) (p)/k! = 2i
dz, where is any positively
(zp)k+1
oriented circle centered at p such that f is analytic in the corresponding
closed disk.
Proof. Let be the circle |z p| = r, 0 < r < R. If z is inside
the circle Cauchys integral formula gives
Z
1
f ()
f (z) =
d.
2i
z
We may now apply Lemma 3.10. We may choose r as close to R as

we wish, so the radius of convergence is at least R. Since any power
series may be differentiated term by term as many times as we wish,
the differentiability follows. This also shows that f (k) (p) = k!ak .
If the largest open disk centered at p in which there is an analytic

function that agrees with f near p has radius R ( ), then the radius
of convergence of the power series is R. But we can not have strict
inequality here, since f then has an analytic extension to a larger disk.
We conclude that the circle of convergence has at least one singularity
of f on its boundary. In particular, if f is entire, it may be expanded
in a power series around any p C, and the radius of convergence will
always be infinite.
Another observation is that if all derivatives of a function analytic in
a disk vanishes at the center of the disk, then the function is identically
zero in the disk, since all coefficients in the power series vanish. We
can generalize this.
Theorem 3.12. Suppose f is analytic in a region and that all
derivatives of f vanish at a point p . Then f vanishes identically.
Proof. The set of points where all derivatives vanish is, as we just
saw, open. But so is the set of points where at least one derivative does
not vanish, since all derivatives are continuous. Thus is the union of
two disjoint open sets, one of which therefore has to be empty3. The
theorem follows.
The power series is called the Taylor series for f at p and the
formula
Z
n1
X
(z p)n
f () d
k
,
f (z) =
ak (z p) +
2i
( p)n ( z)
k=0
3We
use the topological notion of connectedness. Modify the proof yourself to

apply to the notion of arcwise connectedness
3.3. LOCAL PROPERTIES OF ANALYTIC FUNCTIONS
47
obtained from Lemma 3.10 is known as Taylors formula with n terms

and remainder.
Theorem 3.11 gives integral formulas for the derivatives of an analytic function at the center of a disk. This may be generalized.
Corollary 3.13. Suppose f is analytic in a region for which
Cauchys theorem is valid, and that p . Then the derivatives of f
at p are given by
Z
n!
f ()
(n)
(3.4)
f (p) =
d,
2i
( p)n+1
where is any cycle in \ {p} and such that n(, p) = 1.

In particular, (3.4) is true for any positively oriented circle containing p and such that f is analytic in the corresponding closed disk.
P
k
Proof. Suppose f (z) =
k=0 ak (z p) near p. The function
Pn1
g(z) = (f (z) k=0 ak (z p)k )/(z p)n is analytic in , since this
is obvious away from p, and near p it follows from the power series
expansion, which also shows that g(p) = an = f (n) (p)/n!.
Now (z p)kn1 has a primitive (z p)kn /(k n) in \ {p} for
k < n so applying Cauchys integral formula to g, all the terms from
the sum contribute 0 to the integral. The corollary follows.
We are also able to give a kind of converse to the Cauchy integral

theorem which is sometimes useful.
Theorem
3.14 (Morera). Suppose f is continuous in a region
R
and f (z) dz = 0 for all cycles in . Then f is analytic in . It
is actually enough if each point of has a neighborhood such that the
condition is satisfied when is the boundary of any rectangle contained
in the neighborhood.
Proof. The assumption shows that f has a primitive in a neighborhood of every point of , according to Theorem 3.3, or with the
less restrictive assumptions, according to the proof of the Cauchy integral theorem. Thus f is near each point the derivative of an analytic
function, so it is itself differentiable in , i.e., analytic.
Corollary 3.15. Zeros of an analytic function not identically 0

are isolated points in the domain of analyticity.
to
Proof. Suppose f is analytic at p and f (p)
P = 0. According
k
a
(z
p)
.
Since
Theorem 3.11 we may expand f in power series
k=0 k
f (p) = 0 the first term in the series vanishes, and if n is the first index
for whichP
an 6= 0 we obtain f (z) = (z p)n g(z), where g is the analytic
k
function
k=0 an+k (z p) , so that g(p) = an 6= 0. The positive integer
n is called to order or multiplicity of the zero p.
48
3. INTEGRATION
Since g is continuous and g(p) 6= 0 there is a neighborhood of p in

which g doesnt vanish. Since (z p)n only vanishes for z = p it follows
that there is a neighborhood of p in which p is the only zero of f .
Note that the fact that the zeros do not accumulate anywhere in
the domain of analyticity does not prevent them from accumulating at
some point of the boundary of the domain. An example is sin(1/z),
which is analytic in z 6= 0 and has zeros 1/(k), k = 1, 2, . . . , which
accumulate at 0.
A fundamental theorem for entire functions is named after Liouville.
Theorem 3.16 (Liouville). Suppose f is an entire function such
that |f (z)| C|z|n for all sufficiently large |z|. Then f is a polynomial
of degree n. In particular, if f is bounded in all of C, then f is
constant.
Proof. Suppose is a circle centered at 0 of radius r, and consider (3.4) for p = 0. If M (r) = sup|z|=r |f (z)| we obtain |f (k) (0)|
k!rk M (r), k = 0, 1, 2, . . . . These estimates are called Cauchys estimates. Our assumption is that M (r) Crn for large r, so that
|f (k) (0)| k!Crnk if r is large enough. As r we obtain f (k) (0) =
0 for k > n, so that the Taylor expansion of f is a polynomial of degree
n.
The fact that the only bounded, entire functions are constants is
often very useful. We can for example now give a very simple proof of
the fundamental theorem of algebra.
Theorem 3.17 (Fundamental theorem of algebra). Any non-constant polynomial has at least one zero.
Proof. Suppose P is a polynomial without zeros. Then 1/P (z) is
an entire function, and we shall see that it is bounded, so that Liouvilles theorem will show it to be constant.
If P (z) = an z n + an1 z n1 + + a0 with an 6= 0 we may write
P (z) = z n (an + an1 /z + + a0 /z n ). Here the expression in brackets
tends to an as z . Since z n if n > 0 we have 1/P bounded
for large |z|. Thus P is constant. The theorem follows.
3.4. A general form of Cauchys integral theorem

The aim of this section is to prove a version of Corollary 3.6 valid in
more general regions than disks. Note that as soon as we do this we also
have a more general version of Cauchys integral formula, Theorem 3.9.
We remind the reader that a chain is a formal sum of arcs, which can
be integrated over by integrating over each term separately, and adding
the results. Similarly, a chain which may be written as a sum where
all terms are closed arcs is called a cycle.
3.4. A GENERAL FORM OF CAUCHYS INTEGRAL THEOREM
49
We also remind the reader about the notion of a simply connected

set, according to Definition 2.18. This property is closely connected
with the notion of index.
Theorem 3.18. A region is simply connected if and only if
n(, p) = 0 for all cycles and all points p
/ .
Proof. Suppose the complement of (with respect to the extended plane) is connected. Then the complement is contained in
the unbounded region determined by , so Lemma 3.8 (5) shows that
n(, p) = 0 if p
/
Conversely, if { is not connected, we can write it as the disjoint
union of two closed sets4, one of which contains , and the other
therefore being bounded. If the bounded set is K, it is compact and
will therefore have a smallest distance d > 0 to the other part of the
complement. It is left to the reader to prove this.
Let p K and cover the whole plane by a net of closed squares with
side d/2, such that p is the center of one of the squares. Clearly only
finitely many of the squares have at least one point in common with K,
since K is bounded. Let these squares have P
positively oriented boundaries 1 , . . . , n and consider the cycle =
j . Clearly n(, p) = 1
since p is in exactly one of the squares. It is also clear that K,
since the diameters of the squares are < d and they contain points from
K. Now, certain sides of the squares occur twice in , being common
to two adjacent squares. Any side that has a point in common with
K is of this type, and the contributions from these sides in an integral
cancel, since they are run through in opposite directions. Removing
these sides will therefore not change , and then . It follows that
if is not simply connected, then indices for points outside with
respect to cycles in are not always 0.
We shall use this characterization of simply connected regions to

prove the following general version of Cauchys theorem.
Theorem 3.19 (Cauchys integral
theorem). If f is analytic in a
R
simply connected region , then f (z) dz = 0 for any cycle .
Proof. We will show that the assumptions imply that f has a
primitive in . This follows if we can show that the integral of f along
a cycle in consisting only of vertical and horizontal line segments
always vanishes, since then the integral from a fixed point z0 to z along
a path of this type is independent of the particular path, so that we
obtain a well defined primitive in the usual way.
If is such a cycle, extend all line segments in indefinitely. We
obtain a rectangular net consisting of some rectangles with positively
oriented boundaries 1 , . . . , n , and some unbounded regions. We may
4This
is a topological fact
50
3. INTEGRATION
assume n > 0, and pick a point pj in the interior

of each rectangle.
P
0
We shall first show that is the cycle =
n(, pj )j . It is clear
by construction that n( 0 , pj ) = 0 and also that n( 0 , p) =
0 for every point p in one of the unbounded regions determined by
the net, since these points are obviously all in the unbounded regions
determined by respectively 0 .
We shall show that no side of any rectangle is in 0 . Suppose to
the contrary that a side of the rectangle bounded by j is contained in
0 with coefficient a 6= 0. There are at least one region determined
by the net in addition to the rectangle j which have some part of
on its boundary. Let p be a point in in such a region. Now is not
contained in 0 aj , so that the indices of p and pj are the same
with respect to this cycle. But by construction the indices are actually
0 and a, respectively, so that a = 0, and 0 is the empty cycle.
Next we prove that all j for which n(j , pj ) 6= 0 bound rectangles
contained in . For suppose p is in the closed rectangle, but not in
. Then n(, p) = 0, since is simply connected. On the other hand,
the line segment connecting p and pj does not intersect , so p and pj
are in the same component of the complement of , and therefore have
the same index with respect to . It follows that n(, pj ) = 0 unless
the rectangle bounded
by j is contained in . It follows by Goursats
R
theorem that j f = 0 for all j for which j is part of . Thus also
R
f = 0.
We end this section with a very important consequence of the previous theorem.
Corollary 3.20. Suppose f is analytic and has no zeros in a
simply connected region . Then one may define a branch of log(f (z))
in .
Proof. Since f has no zeros in the function f 0 (z)/f (z) is analytic in so that Cauchys integral theorem applies to it. According
to Theorem 3.3 there is therefore a primitive g of this function defined
0 (z)
d
in , and dz
(f (z)eg(z) ) = f 0 (z)eg(z) f (z) ff (z)
eg(z) = 0, so that
f (z)eg(z) = C, where C 6= 0 since neither f nor the exponential function vanishes. Thus we may find A C so that eA = C. It follows that
f (z) = eg(z)+A , so that g(z) + A is a branch of log(f (z)).
Since one may define a branch of the logarithm one may also define
branches of any power function in . We shall use this in proving the
Riemann mapping theorem in Chapter 7.
Remark 3.21. To obtain a version of Cauchys theorem valid in
arbitrary regions we would have to discuss homology of cycles, and we
will abstain from this. We sometimes have to deal with regions which
are not simply connected, but the cycles we integrate over are then
3.5. ANALYTICITY ON THE RIEMANN SPHERE
51
always very simple and explicitly given and therefore never cause any
problem.
For example, suppose f is analytic in a circular ring defined by
0 r0 < |z a| < R0 and suppose r0 < r < R < R0 and let be
the cycle consisting of the two circles |za| = r and |za|
= R, the first
R
negatively and the second positively oriented. Then f (z) dz = 0, and
R f (z) dz
1
for any w satisfying
we also have Cauchys formula f (w) = 2i
zw
r < |w a| < R.
To see this, let 1 be the vertical ray going upwards from a and
2 the opposite ray. Then \ 1 is simply connected. Let 1 be a
positively oriented cycle obtained by taking the part of in the set
Im z Im a and connecting
R the pieces by radial line segments. Then
Theorem 3.19 shows that 1 f = 0.
Similarly, the region \ 2 is simply connected, and if 2 is the
part of in Im z Im a, made into a positively
oriented cycle by
R
adding radial line segments, we also have 2 f = 0. But = 1 + 2 ,
since the radial line segments will be run through twice and in opposite
directions.
It is also easy to see that if Im w > Im a, then n(1 , w) = 0 and
n(2 , w) = 1 so that n(, w) = 1. The reader should modify the construction for other locations of w to see that n(, w) = 1 as soon as
r < |w a| < R.
3.5. Analyticity on the Riemann sphere
Viewing analytic functions as defined on the Riemann sphere, where
all points including look the same, one should be able to define
analyticity at infinity. This leads to the following definition.
Definition 3.22. Suppose f is analytic in a neighborhood |z| >
r > 0 of Then we say that f is analytic at if z 7 f (1/z), which
is analytic in 0 < |z| < 1/r, extends to a function analytic also at 0.
Similarly, if f is analytic in a neighborhood of a and f (z) as
z a it would be tempting to say that f is analytic at a if 1/f (z)
extends to a function analytic at a. We will not use this terminology
since it may lead to confusion, but it is a perfectly reasonable point of
view. In fact, in the next section we will show that if f (z) as
z a, then 1/f (z) always has an analytic extension to a.
Any Mbius transform is in this sense analytic everywhere on the
Riemann sphere, and the reader should should carry out the simple
verification, and also show that the same is true for any rational function.
CHAPTER 4
Singularities
4.1. Singular points
An isolated singularity of a complex function f is a point a such
that it has a neighborhood O with f analytic in O \ {a} (a so called
punctured neighborhood of a). In some cases a is an isolated singularity
simply because we do not know that f is analytic there, or that f is
not analytic at a but will become so provided we assign the correct
value to f (a). In that case a is said to be a removable singularity for
f . A typical example would be z 7 sinz z which is not defined at 0,
but where it is clear from the power series expansion of sin z that the
function becomes entire once we assign it the value 1 at the origin. The
main fact about removable singularities is contained in the following
theorem.
Theorem 4.1. Suppose that f is analytic in a punctured neighborhood of a. Then a is a removable singularity for f if and only if
(z a)f (z) 0 as z a.
Thus the singularities we allowed in Corollaries 3.4, 3.6 are actually
removable, and may be ignored.
Proof. The only if part of the theorem is trivial, since in that
case f must have a finite limit at a. To prove the other direction, let
and be the positively oriented boundaries of disks centered at a
and such that f is analytic in the punctured disks. If is the smaller
disk f is analytic in the ring-shaped region between and , so by
Remark 3.21
Z
Z
1
1
f ()
f ()
f (z) =
d
d
2i
z
2i
z
if z is in the ring-shaped region. Note that the first integral is analytic

in the disk bounded by according to Lemma 3.10. If we can show
that the integral over is zero we have therefore proved the theorem,
since we may remove the singularity at a by defining f (a) to be the
value of the first integral at z = a.
Actually, the integral over does not depend on the radius of the
disk it bounds, as long as that radius is smaller than |z a|. To show
that the integral is 0 it is therefore sufficient to show that its limit as the
radius tends to 0 is 0. To see this, let > 0 and choose > 0 so small
53
54
4. SINGULARITIES
that |( a)f ()| < if | a| < . Then, if the radius of is r < and
r < |z a|/2, we obtain | z| |z a|| a| = |z a|r |z a|/2
so that
Z
Z
f ()
4
|d|
z d | a|| z| |z a| .
The proof is now complete.
Let us now consider an arbitrary isolated singularity a for f . Then

one of the following three cases must obtain:
(1) There is a real number such that |z a| f (z) 0 as z a.
(2) There is a real number such that |z a| f (z) as z a.
(3) Neither of the first two cases hold.
Consider first case (1). If 1, then f has a removable singularity
at a by Theorem 4.1. Otherwise, if n is the largest integer < we
have (z a)n+1 f (z) 0 as z a. By Theorem 4.1 it follows that the
function (z a)n f (z) has a removable singularity at a. This function
may have a zero at a, but ignoring the trivial case when f is identically
zero, we may lower the value of n until (z a)n f (z) has a non-zero
value at a. If n 0 it follows that f is analytic at a.
n > 0 and the
PIf
n
power series expansion around a of (z a) f (z) is k=0 ak (z a)k it
follows that
1
X
X
k
bk (z a)k ,
bk (z a) +
f (z) =
k=n
k=0
where bk = an+k . The first sum above is called the singular part of f
at a. Note that the singular part is analytic everywhere (even at )
except at a. Therefore, if we subtract the singular part from f we get a
function which is analytic wherever f is, and also at a. Subtracting the
singular part at a therefore removes the singularity at a. The fact that
the singular part, in this case, consists of a finite sum of very simple
functions makes this type of singularity rather harmless. It is called a
pole of order n.
A pole of order n > 0 is characterized by the fact that (z a)n f (z)
has a non-zero limit as z a, just as a zero of order n is characterized
by the fact that (z a)n f (z) has a non-zero limit as z a. Note
that f (z) as z approaches a pole so that 1/f has a removable
singularity there. We may therefore view a pole as a point where f
is analytic with the value ; this agrees completely with our point
of view when we discussed functions analytic on the Riemann sphere.
Also note that poles, like zeros, are isolated points. We finally note
that if f has a pole or zero of order |n| at a, then case (1) holds exactly
if > n and case (2) holds exactly if < n.
Now let us consider case (2). If n is the smallest integer , then
(z a)n f (z) as z a so that (z a)n /f (z) has a removable
4.1. SINGULAR POINTS
55
singularity at a. It is clear that this function has a zero at a, say of

order k > 0. It is then clear that (z a)n+k f (z) has a removable
singularity with a non-zero value at a. Therefore, if n + k 0 f has a
removable singularity at a, and otherwise f has a pole of order n + k
at a. So, also in case (2) we have at worst a pole at a.
Unless we have case (3) we therefore have at worst a pole at a
and a singular part consisting of a finite linear combination of negative
integer powers of z a. Conversely, this can not be the case in case
(3) since a pole or a regular point immediately puts us in the cases
(1) and (2). We call the singularity at a essential when we have case
(3). It clearly is a less simple situation, since we can not have a finite
singular part in this case. We shall see in the next section that there
actually is a singular part, but it has infinitely many terms. Another
indication of how complicated the behavior of an analytic function is
near an essential singularity is given by the following theorem.
Theorem 4.2 (Casorati-Weierstrass). The range of the restriction
of an analytic function to an arbitrary punctured neighborhood of an
essential singularity is dense in C.
Proof. Suppose f is analytic in the punctured neighborhood
of a, and that there is a complex number b such that all values of
f in has distance at least d > 0 from b. Consider the function
g(z) = (f (z) b)1 . It is analytic in and bounded by 1/d there.
By Theorem 4.1 it therefore has a removable singularity at a so that
1/g(z) has at most a pole at a (if g has a zero of order n at a, then the
pole has order n). So, f (z) = b + 1/g(z) has at worst a pole at a.
Example 4.3. The function e1/z , z 6= 0, has an essential singularity
at 0. To see this, note that if z 0 along the positive real axis
the function tends to , so the function can not have a removable
singularity at 0. On the other hand, e1/z 0 as z 0 along the
negative real axis, so the origin can not be a pole either. The only
remaining possibility is an essential singularity. Note that by the usual
1/z
power
=
Pseries1 expansion for the exponential function we have e
1 + k=1 k!zk . Hence this function actually has a singular part, but it
consists of infinitely many terms.
Let us end this section by a short discussion of poles at infinity.
Naturally f is said to have a pole of order n at if z 7 f (1/z)
has a pole of order n at 0. It therefore has a singular part which is
a polynomial p(1/z) of order n in 1/z. In particular, it follows that
f (1/z) p(1/z) has a removable singularity at 0, so is bounded there.
It follows that f (z) p(z) is bounded at infinity. The singular part of a
function which has a pole of order n at infinity is therefore a polynomial
of order n.
56
4. SINGULARITIES
Definition 4.4. A function is said to be meromorphic in a region

if it is analytic in except for poles at certain points.
Suppose f is meromorphic in the extended plane C . Since the
extended plane is a compact set, f can only have a finite number of
poles; by Bolzano Weierstrass theorem there would otherwise be a
point of accumulation of poles in the extended plane. This would have
to be a non-isolated singularity. We may therefore subtract the singular
parts for all the poles from f and will then be left with a function
analytic in the extended plane. In particular, a bounded function. By
Liouvilles theorem it will have to be constant. We have proved the
following theorem.
Theorem 4.5. A function is meromorphic in the extended plane if
and only if it is rational.
As a special case it follows that an entire function which is not a
polynomial has an essential singularity at . The elementary functions
ez , cos z and sin z therefore have essential singularities at .
4.2. Laurent expansions and the residue theorem
In this section we will give an expansion generalizing the power
series expansion of an analytic function. In particular we will see that
a function has a singular part at any isolated singularity, analogous to
what we discussed in the previous section, but now possibly consisting
of infinitely many terms. We will then use this expansion to prove
the residue theorem, which gives a particularly simple way to calculate
many complex integrals. We will finally apply this to several types of
real integrals that are difficult or impossible to calculate by elementary
means.
Consider a function f which is analytic in a region containing the
ring 0 R0 < |z a| < R1 . The case R0 = 0 corresponds to the
case when we have an isolated singularity at a. If R0 < r < R < R1 ,
then it follows from Remark 3.21 that the Cauchy integral formula
holds in the form
Z
Z
f ()
1
f ()
1
d
d
f (z) =
2i
z
2i
z
|a|=R
|a|=r
for any z satisfying r < |z a| < R. If we set

Z
1
f ()
1
f1 (z) =
d,
f2 (z) =
2i
z
2i
|a|=R
f ()
d,
z
|a|=r
then f (z) = f1 (z)+f2 (z) for such values of z. However, by Lemma 3.10
f1 is analytic in |z a| < R. It follows that f2 is analytic in r <
|z a| < R. Actually, f2 is analytic in |z a| > r, even at ,
which is seen similarly to the proof of Lemma 3.10. In fact, setting
4.2. LAURENT EXPANSIONS AND THE RESIDUE THEOREM
57
z = a+1/w we may write the denominator in f1 as z = a1/w =

(1 ( a)w)/w, and since |(z a)w| < 1 the reciprocal of this is
the sum of a convergent geometric series, and reasoning just as in the
proof of Lemma 3.10 we obtain
Z
X
wk+1
f2 (a + 1/w) =
f ()( a)k d,
2i
k=0
|a|=r
a power series in w which converges for |w| < 1/r, corresponding to

|z a| > r. It follows that f2 is analytic in |z a| > r, including
z = . Setting
Z
1
f ()
(4.1)
ak =
2i
( a)k+1
|a|=r
also for k = 1, 2, . . . we can write this as f2 (z) =

Adding up we obtain the following theorem.
P1
k=
ak (z a)k .
Theorem 4.6 (Laurent expansion). Suppose f is analytic in 0

R0 < |z a| < R1 . Then f has a Laurent expansion around a
of the form
X
ak (z a)k ,
f (z) =
k=
converging at least for R0 < |z a| < R1 , where the coefficients ak are

given by (4.1).
P
k
The singular part of f at a is 1
k= ak (z a) and is analytic
for |z a| > R0 , including at . In particular, if a is an isolated singularity for f , the singular part expansion converges everywhere except
at z = a. The difference of f and its singular part is analytic wherever
f is and also for |z a| < R1 .
a1
Note that wherever the series converges the function f (z) za
=
P
ak
k
k+1
,
k6=1 ak (z a) is the derivative of the function
k6=1 k+1 (z a)
so that its integral along any closed curve in the domain of convergence
is 0. It follows that the integral of f around a positively oriented circle
in R0 < |z a| < R1 is equal to 2ia1 . The coefficient a1 in the
Laurent expansion of f around a is called the residue of f at a, since it
determines what remains after integration around a closed curve. We
will denote the residue of f at an isolated singularity a by Res f (a).
This is of course 0 unless a actually is a singularity of f . A slight
generalization of the above gives the following important theorem.
Theorem 4.7 (Residue theorem). Suppose is a simply connected

region, and that f is analytic in except for a finite number of isolated
singularities. Then, if is a cycle in not passing through any of the
58
singularities,
4. SINGULARITIES
Z
f (z) dz = 2i
n(, z) Res f (z).
Proof. Subtract from f the singular parts for all singularities.

This leaves a function analytic in , so that its integral is zero. As
we saw above, the singular parts are analytic outside the corresponding singularity, and removing the term with index 1 the rest of the
singular part has a primitive defined outside the singularity, so that
their integrals vanish. It remains only to integrate the terms of index
1 for each singularity, which gives the result by the definition of the
index.
In all our applications of the residue theorem we will choose the

cycle so that the indices of all the singularities with respect to are
either 1 or 0.
A formula for the residue at an isolated singularity is of course given
by (4.1) for k = 1. Actually, this formula is not of much practical
value; on the contrary, one tries to find the residues without integration
and then uses this to evaluate integrals. It is clear, however, that for
this to be possible we need methods not involving integration to find
residues. No such generally applicable method is known in the case
of an essential singularity, even though there are of course many cases
when we will know the Laurent expansion, as we saw in the case of
e1/z . The situation is different in the case of a pole, and we have the
following theorem.
Theorem 4.8. Suppose that f has a pole of order n at a. Then
1
dn1
n
Res f (a) = (n1)!
limza dz
n1 ((z a) f (z)). In particular, for a simple
pole the residue is limza (z a)f (z). If f = p/q where p and q are
analytic at a, p(a) 6= 0 and q has a simple zero at a, then the residue
at a is p(a)/q 0 (a). Similarly, if q has a double zero at a the residue is
(4.2)
6p0 (a)q 00 (a) 2p(a)q 000 (a)

.
3(q 00 (a))2
Similar, even more complicated, formulas hold for higher order poles.
P
k
Proof. According to assumption f (z) =
k=n ak (z a) for z
n
close to a, so g(z) = (z a) f (z) has a removable singularity at a and
a1 is the coefficient of (z a)n1 in the corresponding power series
expansion. But this coefficient is g (n1) (a)/(n 1)! and since g (n1) is
continuous at a the first claim follows. If now q has a simple zero at
za
a, then (z a)p(z)/q(z) = p(z) q(z)q(a)
p(a)/q 0 (a) since q(a) = 0,
q 0 (a) 6= 0. Finally, if q has a double zero at a, then q(z) = (z a)2 q2 (z)
where q2 (a) = q 00 (a)/2 and q20 (a) = q 000 (a)/6, as is easily verified. Hence
4.3. RESIDUE CALCULUS
59
((z a)2 f (z))0 = (p(z)/q2 (z))0 = (p0 (z)q2 (z) p(z)q20 (z))(q2 (z))2 . Letting z a (4.2) follows, and the final claim is left to the reader to
verify.
The conclusion of all this is that simple poles cause little problem
in determining the residue, whereas higher order poles are considerably
more messy to deal with. In the next section we shall see how one may
use the residue theorem to calculate certain real integrals.
4.3. Residue calculus
In this section we shall see how one may use the residue theorem
to calculate certain real integrals. We will only discuss a few types of
integrals that can be handled; many others exist. R
2
1. Let us first consider an integral of the form 0 (cos , sin ) d.
Here (x, y) is a rational function of two variables with no poles for
(x, y) on the unit circle. If we think of this integral as the result of calculating an integral around the unit circle by the parametrization z = ei ,
0 2, we note that by Eulers formulas we have cos = 21 (z+1/z)
and sin = 2i1 (z1/z). Furthermore, dz = iei d so that d = i dz/z.
R
The integral therefore equals i |z|=1 ( z+1/z
, z1/z
) dz/z. The inte2
2i
grand is rational and it only remains to find those poles that are inside
the unit circle and evaluate their residues.
R 2 d
Example 4.9. Consider the integral 0 a+cos
where a > 1. Set
R
i
ting z = e as above, the integral equals i |z|=1 (a+ 12 (z+1/z))1 dz/z
which after simplification becomes
Z
dz
2i
.
z 2 + 2az + 1
|z|=1
The zeros of the denominator are z = a a2 1. Since their

product is 1,
precisely one root is inside the unit circle; a being > 1
this root is a2 1 a. Since the pole comes from a simple zero in the
denominator, we can use the method of Theorem 4.8 to calculate the
residue. The residue is therefore the value of (2z + 2a)1 at the root.
By the residue theorem the original integral therefore equals a2
2 1 .
R
2. We next consider an integral (x) dx where is a rational function with no real poles and the degree of the denominator at least 2
higher than the degree of the numerator, so that the integral certainly
converges. To calculate this using residue calculus, let be a half circle in the upper half plane, centered at the origin and with radius R,
together with the real line segment [R, R]. We give positive orientation. For R sufficiently large, all theRpoles of whichP
are in the upper
half plane will be inside so that (z) dz = 2i Im z>0 Res (z).
60
4. SINGULARITIES
On the other hand, along the part of which is a half circle we can
estimate the integral by
Z
Z
|dz|
2
= 2 sup |z 2 (z)|/R 0
(z) dz sup |z (z)|
2
|z|
|z|=R
|z|=R
|z|=R
Im z>0
|z|=R
Im z>0
as R , since z 2 (z) is bounded for large values of |z|, by the

assumption on the degree of . It follows that
Z
X
(x) dx = 2i
Res (z) .
Im z>0
R
2
Example 4.10. Consider the integral x4x+1 dx which satisfies
all the requirements above. The poles of the integrand are given by the
zeros of its denominator, so they are the roots of z 4 + 1 = 0. Setting
z = Rei we easily see that the roots are zk = ei(/4+k/2) , k = 0, 1, 2, 3.
The roots in the upper half plane are the two first ones. Since the zeros
z2
1
are simple ones, the residues are obtained by evaluating 4z
3 = 4z at
these points. It follows that
Z
x2
i/4
3i/4
dx
=
2i(e
+
e
)/4
=
2/2 .
x4 + 1
R
3. We next consider an integral (x)eiax dx where a is real and
a rational function without real poles. This is the Fourier transform of
the function at a. We assume that the degree of the denominator of
is higher than the degree of the numerator. This does not guarantee
absolute convergence of the integral, but as we shall see it does imply
conditional convergence if a 6= 0. In the calculations below we shall
assume that a > 0, but the case a < 0 can be treated very similarly.
This is done either by replacing the upper half plane by the lower half
plane in the considerations below, or else by first making the change of
variable t = x in the integral, which has the effect of replacing a by
a.
Let A, B and C be positive real numbers. We consider a contour
starting with a segment [A, B] of the real axis, continuing with a
vertical line segment from B to B + iC, then a horizontal line segment
from B+iC to A+iC and finally a vertical line segment from A+iC
to A. If A, B and C are sufficiently large, this rectangle will contain
all poles of which are in the upper half plane so that
Z
X
(z)eiaz dz = 2i
Res((z)eiaz ) .
Im z>0
Our assumptions guarantee that z(z) is bounded for |z| sufficiently

large, say |z| > R. Let a corresponding bound be M . If we parametrize
61
the vertical line segment from A to A + iC by z = A + it, 0 t C

the absolute value of the corresponding integral may be estimated by
ZC
Z
M
M
M
at
e dt
,
eat dt =
|A + it|
A
aA
0
provided A > R. Similarly, the integral over the other vertical side
M
may be estimated by aB
provided B > R. Note that these estimates
are independent of C. Assuming that C > R and parametrizing the
upper side of the rectangle by z = t + iC, B t A, we can
similarly estimate the corresponding part of the integral by
ZA
M
M (A + B)
.
eaC dt
| t + iC|
CeaC
This clearly tends to 0 as C . It follows that

ZB
M 1
1
iax
iaz
(x)e dx 2i
Res((z)e
( + ).
a A B
Im z>0
A
This shows that the original integral indeed converges, at least conditionally, and that its value is given by the residues in the upper half
plane.
R
Example 4.11. Consider the integral cos(x)
dx, where is real.
x2 +1
First note that the cos function is even. We may therefore replace
by || without affecting the value of the integral. Next, note that the
R ix||
integral is the real part of xe 2 +1 dx which we may evaluate using
the method above, and then take the real part of. Actually, since it is
easily seen that the integrand of the imaginary part is an odd function,
the imaginary part is zero anyway. Note, however, that we can not
evaluate the present integral, or integrals similar to it, by considering
the residues of cos(z)
, since the function cos(z) is large for large | Im z|,
z 2 +1
in both upper and lower half planes.
According to our deliberations above, we have
Z ix||
X
eiz||
e
dx = 2i
Res 2
.
x2 + 1
z +1
Im z>0
In the upper half plane there is only one singularity, a simple pole at
1 iz||
z = i. Thus, the residue is obtained by evaluating 2z
e
at z = i.
e||
This gives 2i so that we obtain
Z
cos(x)
dx = e|| .
x2 + 1
62
4. SINGULARITIES
Example 4.12. We will consider one more example of this type of

integral,
R sin x with an added difficulty. The integral we want to evaluate is
dx. According to our present strategy we ought to relate this
x
to the residues of eiz /z. Unfortunately this function has a pole at the
origin; note that there is no such problem with sinx x , which is an entire
function. This is immediately seen from the power series expansion. To
circumvent the difficulty we modify the path so that the line segment
[A, B] on the real axis is replaced by the two line segments [A, r]
and [r, B], connected by a half circle in the upper half plane, centered
at the origin and with radius r. If denotes this half circle, but run
through counterclockwise, estimates of the same kind as before show
that
Z
Z
X
eiz
sin x
eiz
dx = Im
dz + 2i
Res
.
x
z
z
Im z>0
|x|>r
Since there are no poles in the upper half plane we only need to consider
the integral overR. If we parametrize by z = rei , 0 , the
integral
i 0 exp(irei ) d which tends to i as r 0. It follows
R equals
that sinx x dx = .
R
4. Next consider the integral 0 x (x) dx where 0 < < 1 and is
rational with no positive real poles. For convergence we must assume
that the degree of the denominator in is at least 2 more than the
degree of the numerator. Similarly, we may allow at most a simple
pole at the origin. If we want to relate the value of this integral to the
residues of the function z (z), note that we now have a branch point
at the origin. However, instead of causing difficulties this is actually
what will allow us to evaluate the integral.
Suppose we choose the branch of z where the plane is cut along
the positive real axis and for which 0 < arg z < 2. This means that
z = e log z where log is the corresponding branch of the logarithm. As
we approach a point x > 0 on the real axis from above we then obtain
the usual real power x . However, as we approach the same point from
below we instead obtain e(ln x+i2) = ei2 x .
Intuitively, we would therefore like to choose a contour which
starts at r > 0 on the upper edge of the real axis, continues to R >
r, then follows the circle with radius R and centered at the origin,
counterclockwise, until we reach R on the lower edge of the real axis,
then back to r, still on the lower edge, and finally along the circle
with radius r and centered at the origin, clockwise, until we reach the
initial point again (you will need to draw a picture of this). The two
contributions from integrating along the real axis will not cancel since
the power has different values along the upper and lower edges. The
catch is, there is no such thing as an upper or lower edge of the positive
real axis; in fact, since we have cut the plane along the positive real
axis, we cant integrate along it at all.
63
The problem can be avoided in the following way. First cut the
plane along some ray from the origin other than the positive real axis.
In the remaining part of the plane define a branch of z by requiring its
values to be real on the positive real axis. Now pick a contour, starting
at r > 0, continuing to R > r, then along the circle of radius R as
before, but stop before you reach the branch cut and go back along a
ray , which does not contain any pole of , until you reach the circle
with radius r again. Finally, continue clockwise along this circle until
you reach the point r again.
Next, pick another branch of z by cutting the plane along a ray
which comes before , counted counterclockwise from the positive real
axis. Fix the branch by requiring its values on to coincide with the
values of the earlier branch. This will give the branch the value ei2 x
for a real, positive x. We integrate the new branch along a contour
which starts at R, continuous along the positive real axis to r, then
follows the circle with radius r clockwise until it reaches the ray ,
then follows this ray outwards until it reaches the circle with radius R.
Finally, it follows this circle counterclockwise until it reaches the point
R on the positive real axis again.
If we add the two integrals constructed above, the contributions to
them along the ray will cancel, and the total effect will be exactly
as if we could integrate as we originally did, letting z have different
values along the upper and lower edges of the positive real axis. In
view of this, we will not commit any errors of we think of this as being
possible.
Now let us estimate the integrals along the circles. Note that |z | =
|z| whatever branch we use (as long as is real). Our assumption on
the degree of means that z 2 (z) is bounded, say by M , for large |z|.
If R is sufficiently large we therefore get
Z
Z
z (z) dz M R
|dz| = 2M R1 0
|z|=R
|z|=R
as R . Similarly, z(z) is bounded, say by m, for |z| sufficiently

small. If r > 0 is sufficiently small we therefore get
Z
Z
|dz| = 2mr 0
z (z) dz mr
|z|=r
|z|=R
as r 0. The
along the positive real axis together contribute
R R integrals
i2
(1 e
) r x (x) dx. It follows that
Z
x (x) dx = (1 ei2 )1 2i
0
X
z6=0
Res(z (z)) .
64
4. SINGULARITIES
Note that it is extremely important that one uses the branch of z

where the plane is cut along the positive real axis and 0 < arg z < 2.
R dx
Example 4.13. Consider the integral 0 xx2 +1
, 0 < < 1. Using
z
the appropriate branch the function z2 +1 has simple poles at i, so the
residues at these points are the values of z2z at i.

The integrand satisfies all the conditions above, so the integral
equals (1 ei2 )1 (i (i) ). We have to choose the argument of i
to be /2 and that of i to be 3/2 so i (i) = ei/2 ei3/2 =
ei (ei/2 ei/2 ). Similarly, 1 ei2 = ei (ei ei ).
The second factor equals (ei/2 ei/2 ) (ei/2 + ei/2 ) =
i/2
(e
ei/2 )2 cos(/2). It follows that
Z
x dx
=
.
2
x +1
2 cos(/2)
0
This is actually true for 1 < < 0 too, since in that case we may write
x+1
the integrand as x(x
2 +1) where now 0 < +1 < 1 so all the assumptions
are satisfied. But the residues remain the same since z +1 /z = z , using
the same branches of the powers. The formula is of course also true for
= 0, for elementary reasons.
R
5. We finally consider an integral 0 (x) ln x dx, where again is
rational without positive real poles. We still need to assume that the
degree of the denominator in is at least 2 more than that of the
numerator. In contrast to integrals of type 4, however, we can no longer
allow a pole at the origin. On the other hand, we use the same contour,
justifying the use of different values of the logarithm along the upper
and lower edges of the positive real axis as before. If we consider
(z) log z, using the branch of the logarithm where 0 < arg z < 2, its
values at x > 0 on the upper edge of the positive real axis is (x) ln x,
where ln is the usual real logarithm. For x > 0 on the lower edge
we instead get (x)(ln x + 2i). The difference is therefore 2i(x),
so we will not get the integral we are looking for. So, instead we
consider the function (log z)2 (z) which is (ln x)2 (x) on the upper
and (ln2 x + 4i ln x 4 2 )(x) on the lower edge of the positive real
axis. The difference is therefore 4i(x) ln x + 4 2 (x). It follows
that
Z
Z
1X
(x) ln x dx + i (x) dx =
Res((log z)2 (z)) .
2 z6=0
0
If, as we normally assume, has real coefficients we can therefore calculate the desired integral by taking the real part of the right hand
side. Otherwise, we would first have to calculate the second integral
by integrating (z) log z as in our first attempt.
4.4. THE ARGUMENT PRINCIPLE
65
R x
Example 4.14. Consider the integral 0 xln2 +1
dx which satisfies
the assumptions above. Using the appropriate branch in the plane cut
z)2
along the positive real axis, the function (log
has simple poles at i
z 2 +1
2
so the residues are the values of (log2zz) at these points. The sum of
the residues is therefore 2i1 ((i/2)2 (i3/2)2 ) = i 2 which is purely
imaginary. It follows that
Z
ln x
dx = 0 .
x2 + 1
0
Incidentally, by taking the imaginary part, we also get

but there is of course an easier way of getting this.
R
0
dx
1+x2
= /2,
Example 4.15. As a final example we consider an integral of type

5, but with an added difficulty. Since ln x has a simple zero at x = 1
it ought to be possible to allow a simple pole of at 1. This causes
problems, however, since the branch of the logarithm we use on the
lower edge of the positive real axis does not have a zero at 1. To
circumvent the difficulty, we replace the part of the integral along the
lower edge between 1 + r and 1 r by a half circle of radius rR > 0 in the
x
lower half plane, centered at 1. As an example, consider 0 xln2 1
dx.
The integral along the half circle is then
Z2
i
(log(1 + rei ))2 i

re d 2 3 i
(1 + rei )2 1
as r 0 .
Note that the logarithm approaches 2i as r 0! The integrand has a

simple pole at z = 1 and nowhere else, so the sum of the residues is
(log(1))2
= 2 /2. It follows that
2
Z
ln x
1
2
lim
4i 2
+ 4 2
dx = 2 3 i + 2i 2 /2 = 3 i .
r0
x 1
x 1
x0
|x1|r
Taking imaginary parts and dividing by 4 we finally obtain

Z
2
ln x
dx =
.
x2 1
4
0
4.4. The argument principle

The following theorem is a simple consequence of the residue theorem.
Theorem 4.16. Suppose that f is meromorphic in a simply connected region and that is a cycle in .
66
4. SINGULARITIES
Assume further that f has zeros a1 , a2 , . . . , an and poles b1 , b2 , . . . , bk

in , each repeated according to multiplicity and none of them on .
Then
Z 0
n
k
X
X
1
f (z)
dz =
n(, aj )
n(, bj ).
2i
f (z)
j=1
j=1
We normally choose so that the index of each zero and pole with
respect to equals one, and then the right hand side becomes the
difference between the number of zeros and the number of poles in ,
each counted by multiplicity.
Proof. If f (z) = (z a)n g(z) where n is a non-zero integer, g is
analytic near a and g(a) 6= 0, then f 0 (z) = n(za)n1 g(z)+(za)n g 0 (z)
0 (z)
0 (z)
n
so that ff (z)
= za
+ gg(z)
. The last term is analytic near a, so the residue
of the left hand side at a is n. Since |n| is the multiplicity of a as a zero
respectively pole, the theorem follows from the residue theorem.
The theorem is usually known as the argument principle,

R f 0 (z) for the
following reason. If is a closed arc, the integral f (z) dz equals
R dz
, as is easily seen using the definition of the integral through a
f z
parametrization. Thus the integral is 2i n(f , 0). But this is the
variation of the argument of z as z runs through f . To see this, note
that the principal logarithm is a primitive of 1/z away from the negative
real axis. Now, f may intersect the negative real axis at certain
points; assume for simplicity that they are finitely many. At every
such intersection we have to add or subtract 2 from the argument of
z, depending on whether we intersect from below or above. Between
two intersections we may calculate the integral by using the principal
branch of the logarithm. Adding everything up, the real parts will
cancel, and what remains is an integer multiple of 2i, in other words
i times the variation of the argument along the curve. Clearly the
variation of the argument of z along f is the same as the variation
of the argument of f (z) along .
The integral is therefore (i times) the variation of the argument of
f (z) as z runs through . Since one can often find the variation of
argument without calculating the integral, this gives information on
the number of zeros or poles in a region. Used this way, the argument
principle is of great importance to many applications in control theory
and related subjects. We give a few examples of how this is done.
Example 4.17. We wish to find the number of zeros in the right
half plane of the polynomial p(z) = z 5 + z + 1.
If z = iy is purely imaginary p(z) = iy(y 4 + 1) + 1 has real part 1,
so is never zero. If is the argument of p(iy) we have tan = y(y 4 + 1)
which tends to + as y + and as y . Running
through the imaginary axis from iR to iR for a large R > 0 the
67
argument thus decreases by nearly . On a large circle |z| = R we

have p(z) = z 5 (1 + z 4 + z 5 ), where the second factor is nearly one, so
that its argument varies very little, whereas the argument of the first
factor increases by 5 as we follow the circle in a positive direction
from iR to iR. The variation of argument of p along the boundary of
the large halfdisk is therefore nearly 4, and since it must be an integer
multiple of 2 it is exactly 4. There are therefore exactly two zeros
inside the halfcircle if it is sufficiently large. In other words, there are
precisely two zeros in the right half plane!
Example 4.18. We wish to find the number of zeros in the first
quadrant of the polynomial f (z) = z 4 z 3 + 13z 2 z + 36.
First note that there are no zeros on either the real or imaginary
axes since for z = x R we have
1
47
143
>0
x4 x3 + 13x2 x + 36 = (x2 + 1)(x )2 + x2 +
2
4
4
and for z = iy, y R we have
z 4 z 3 + 13z 2 z + 36 = y 4 13y 2 + 36 + i(y 3 y).
The imaginary part vanishes only for y = 0 and y = 1, neither of
which is a zero for the real part. Now let be the line segment from
0 to R > 0, followed by a quarter circle of radius R centered at 0
and ending at iR, and finally the vertical line segment from iR to
0. For R sufficiently large, all the zeros in the first quadrant will be
inside , so we only need to calculate the variation of argument for
the polynomial along . Since f > 0 on the real axis, the argument
stays equal to 0 along the horizontal part of . For |z| = R we write
f (z) = z 4 (1 z1 + 13
z13 + 36
). Note that the bracketed expression
z2
z4
tends to 1 as R so its argument varies only a little around 0. The
argument of the first factor varies 4 times the variation of the argument
of z, i.e., by 4 2 = 2. So, along the circular arc the argument varies
close to 2.
It remains to find the variation of the argument along the imaginary
y 3 y
axis. If denotes the argument of f (z), then tan = y4 13y
2 +36 . For
y = 0 this is 0, and for y we get tan 0. The argument
variation along the vertical part of is therefore close to some integer
multiple of . To go from one multiple to the next, tan will have to
become in between. This happens at the zeros of y 4 13y 2 + 36 =
(y 2 9)(y 2 4) = (y + 3)(y + 2)(y 3)(y 2). The first two factors
stay positive for y 0 so the denominator in tan passes from positive
to negative as y decreases through 3, and from negative to positive as
y decreases past 2. In both these points the numerator is positive, so
tan passes from + to as y decreases through 3 and then from
back to + as y decreases through 2.
Hence, if we start at y = R for a large value of R, the variation in
argument along the vertical line segment is close to 0. Therefore, for
68
4. SINGULARITIES
large R > 0 the variation in argument of f along is close to 2, and

since it has to be an integer multiple of 2, it is exactly 2. There is
therefore exactly one zero of f in the first quadrant.
A useful consequence of the argument principle is the following
theorem.
Theorem 4.19 (Rouchs theorem). Suppose f and g are analytic
in a simply connected region and that is a cycle in such that
n(, z) is 0 or 1 for every z .
Also assume that |f (z) g(z)| < |f (z)| for z . Then f and g
have the same number of zeros, counted with multiplicity, enclosed by
( i.e., for which the index with respect to is 1).
Proof. The inequality shows that neither f nor g can have a zero
, then the zeros for F are the zeros for g
on . If we set F (z) = fg(z)
(z)
and the poles for F are the zeros for f . We therefore need to show that
F has the same number of zeros and poles, i.e., that the variation of
argument of F along is 0. Note that this is true even if f and g have
common zeros so that there is some cancellation in F .
However, by assumption |F (z) 1| < 1 for z . Hence F has
all its values on in the disk with radius 1 centered at 1, which does
not contain the origin. Hence the variation of argument is 0 (give a
detailed motivation, for example using the principal logarithm).
Example 4.20. We shall determine the number of zeros in the right

half plane of the function g(z) = a z ez , where a > 1.
It is clear that the function f (z) = a z has only the zero z = a,
which is in the right half plane. If is a positively oriented half circle
in the right half plane, with radius R and centered at the origin, this
zero is inside as soonpas R > a. For z = iy on the imaginary
axis we have |f (z)| =
a2 + y 2 a > 1, and on the circular arc
i
i
we have |f (Re )| = |Re a| R a > 1 if R > 1 + a. But
|f (z) g(z)| = |ez | = e Re z 1 for z in the right half plane. Hence
|f (z) g(z)| < |f (z)| for z as soon as R > 1 + a. Therefore, by
Rouchs theorem, g has exactly one zero in the right half plane, and
this zero has absolute value 1 + a.
The next theorem demonstrates a very important topological property of an analytical map.
Theorem 4.21. Suppose f is analytic at z0 and that f (z0 ) = w0
with multiplicity n, i.e., f (z)w0 has a zero of multiplicity n for z = z0 .
Then, for every sufficiently small > 0 there exists a > 0 such
that if |a w0 | < , then f (z) = a has exactly n roots (counted with
multiplicity) in |z z0 | < .
Proof. Since zeros of analytic functions are isolated, we may require > 0 to be so small that z0 is the only point in |z z0 | where
69
f (z) w0 = 0. If = min |f (z) w0 | it follows that the integral

|zz0 |=
1
2i
f 0 (z)
dz
f (z) a
|zz0 |=
is continuous as a function of a for |a w0 | < . But it is also an

integer, so it must be constant in this disk. Since it equals n for a = w0
the theorem follows from the argument principle.
We restate the most important part of the conclusion of Theorem 4.21 as the open mapping theorem.
Corollary 4.22. Suppose f is analytic in some region and not
constant. Then f is an open mapping, i.e., the images of open sets are
open.
Proof. If z0 is in the domain of f , then by Theorem 4.21 the image
of any sufficiently small neighborhood of z0 contains a neighborhood of
f (z0 ). Hence f is an open mapping.
Note that n = 1 in Theorem 4.21 exactly if f 0 (z0 ) 6= 0, and that

n = 1 means that the inverse function f 1 is defined in |z w0 | < . By
Corollary 4.22 the inverse function has the property that the inverse
image of an open set under f 1 is open; in other words, the inverse is
continuous. But by Theorem 2.6 (5) this implies that f 1 is analytic,
with (f 1 )0 (z) = 1/f 0 (f 1 (z)) (note that the denominator is 6= 0 here).
We therefore also have the following corollary.
Corollary 4.23. If f 0 (z0 ) 6= 0, then f maps a neighborhood of z0
conformally and topologically ( i.e., continuously and with continuous
inverse) onto a neighborhood of f (z0 ).
It remains to see what type of mapping we have in a neighborhood
of a point z0 where f 0 (z0 ) = 0. We have one very well known example;
the function z 7 z n where n is an integer > 1. This function has an
n-fold zero at z = 0, and the image of a neighborhood of 0 covers a
neighborhood exactly n times. This is, in fact, what happens in general.
To see this, consider a function f such that f (z) w0 has a zero of
order n at z0 . We may then write f (z) = w0 + (z z0 )n g(z), where
g is analytic where f is, and g(z0 ) 6= 0. According top
Corollary 3.20
n
we may therefore define a single-valued branch h(z) of g(z) which is
analytic in a neighborhood of z = z0 .
d
Note that dz
(z z0 )h(z) = h(z)+(z z0 )h0 (z) which equals h(z0 ) 6=
0 for z = z0 . The function z 7 (z z0 )h(z) therefore maps a neighborhood of z0 conformally onto a neighborhood of 0. We may therefore
view f (z) = w0 + ((z z0 )h(z))n as a composite of this function, of
the function z 7 z n , and a translation. It follows that the image of a
70
4. SINGULARITIES
small neighborhood of z0 under f covers a neighborhood of w0 exactly

n times.
We turn now from these general considerations to a very useful and
very specific result.
Theorem 4.24 (Maximum principle). Suppose f is analytic in a
region . If |f | has a (local) maximum in , then f is constant.
A variant of this states that if f is analytic in a compact set, then
the maximum of |f | on the set is taken on the boundary unless f is
constant. This follows from Theorem 4.24 and the fact that a function
continuous on a compact set, in this case |f |, takes a maximum value.
Proof. Suppose f is not constant. According to the open mapping
theorem, given any neighborhood O of z0 , all values in a sufficiently
small neighborhood of f (z0 ) are taken in O. Some of these values will
be further from the origin than f (z0 ), so |f (z0 )| can not be a local
maximum value of |f |.
A rather special, but as it turns out, very useful, consequence of

the maximum principle is the following.
Theorem 4.25 (Schwarz lemma). Suppose f is analytic in |z| < 1,
that |f (z)| < 1 and f (0) = 0. Then |f (z)| |z| for |z| < 1, |f 0 (0)| 1,
and if equality occurs in either of these inequalities, then f (z) = cz for
some c with |c| = 1.
Proof. The function g(z) = f (z)/z has a removable singularity
at 0; we must set g(0) = f 0 (0). For |z| = R < 1 we have |g(z)| <
|z|1 = 1/R so by the maximum principle we have |g(z)| < 1/R for
|z| < R. Given any z with |z| < 1 we therefore have |g(z)| < 1/R
for all R, |z| < R < 1. Letting R 1 we get |g(z)| 1 in the unit
disk. The maximum principle finally tells us that if we have equality
anywhere, i.e., a local maximum of |g|, then g is constant. The theorem
follows.
Schwarz lemma has a very important application in determining

to what extent conformal maps are unique. Later we shall show that
any simply connected region can be mapped conformally and bijectively onto the unit disk. This immediately shows that any two simply
connected regions may be mapped conformally and bijectively onto
each other, since one may first map both conformally and bijectively
onto the unit disk, and then compose the inverse of one map with the
other map. The resulting function then maps one region onto the other
conformally and bijectively.
It is clear that uniqueness questions can also be answered if they
can be resolved for the special case of a map onto the unit disk. It is
immediately clear that if there is a conformal map of onto the unit
disk, then we can pick any point z0 and require it to be mapped to
71
0. For, by assumption there is a conformal map of onto the unit disk;

suppose the image of z0 is w0 . We can then find a Mbius transform
that maps the unit disk onto itself and takes w0 to 0. Composing the
original map with this Mbius transform we obtain a map of which
takes z0 to 0. Is this map unique?
Suppose f and g both map conformally onto the unit disk, and
both map z0 onto 0. Then f g 1 maps the unit disk onto itself
and keeps 0 fixed. By Schwarz lemma |f g 1 (w)| |w|. But setting
z = f g 1 (w) this means |g f 1 (z)| |z|. On the other hand,
Schwarz lemma again tells us that |g f 1 (z)| |z| so that in fact
equality holds throughout the unit disk. A final use of Schwarz lemma
tells us that f g 1 (z) = cz where |c| = 1.
0 (z )
0
Note that c = (f g 1 )0 (0) = fg0 (z
so that if we specify the argument
0)
of the derivative at z0 as well, the map is unique. A particular case is of
course when is the unit disk itself; it follows that the only automorphisms of the unit disk (bijective conformal maps of the unit disk onto
itself) are the Mbius transforms with this property. More generally,
given any two regions that are circles or half planes, the only bijective
conformal maps of one onto the other are Mbius transforms. Similar
statements can be made with respect to the other special regions for
which we found explicit conformal maps (wedges, infinite strips, etc.).
CHAPTER 5
Harmonic functions
5.1. Fundamental properties
Suppose f is analytic in some region and u, v are its real and
imaginary parts, so that f (x + iy) = u(x, y) + iv(x, y). Then u and v
are harmonic in , according to the following definition.
Definition 5.1. A function u defined in an region C is called
harmonic if it is twice continuously differentiable in and satisfies
2
2
u = 0, where = x
2 + y 2 .
This follows since u, v satisfy the Cauchy-Riemann equations
ux = v y ,
uy = vx .
Since f is infinitely differentiable, we can differentiate the first equation
with respect to x, the second with respect to y, and add the results to
obtain u = 0, using that vxy = vyx . Similarly one shows that v is
harmonic.
If a function u, harmonic in , is given, then another harmonic
function v is called a conjugate function to u in if u + iv is analytic
in . Note that if u has a conjugate function in some region, then it
is determined up to an additive real constant. For suppose u + iv and
u + i
v are both analytic. Then so is the difference i(v v) which has
real part 0. It follows that the imaginary part v v is constant (this
follows from the Cauchy-Riemann equations, but also directly from the
open mapping theorem).
Note that if v is the harmonic conjugate of u, then u is the harmonic conjugate of v, since v iu = i(u + iv) is analytic if u + iv
is. A harmonic function does not necessarily have a conjugate
function
p
2
2
defined in all of its domain; consider for example ln x + y which is
the real part of any branch of the logarithm and therefore harmonic in
R2 \ {(0, 0)}. It can not have a conjugate function in this set, because
that would imply that we could define a single-valued branch of the
logarithm in the plane with just the origin removed. But we cant. On
the other hand, locally there is always a conjugate function. In fact,
the following theorem holds.
Theorem 5.2. If u is harmonic in a disk, then it has a conjugate
function there.
73
74
5. HARMONIC FUNCTIONS
R x (x0 , y0 ) is the center of the disk and set v(x, y) =

R y Proof. Suppose
u (x, t) dt x0 uy (t, y0 ) dt for any (x, y) in the disk. Note that v is
y0 x
well defined since we are only evaluating u at points in the disk. By the
fundamental theorem of calculus we have vy = ux and differentiating
under the integral sign we obtain
Zy
vx (x, y) =
uxx (x, t) dt uy (x, y0 )

y0
Zy
=
uyy (x, t) dt uy (x, y0 ) = uy (x, y),

y0
using the fact that u is harmonic. So, v is a harmonic conjugate of

u.
Since analytic functions are infinitely differentiable we immediately

obtain the following corollary.
Corollary 5.3. Harmonic functions are infinitely differentiable.
There is a much more general version of the theorem, which states
that any function harmonic in a simply connected region has a harmonic conjugate there. However, since this follows from Exercise 5.5
below and the Riemann mapping theorem, which we will prove later,
we will not attempt a proof here.
Corollary 5.4. Suppose f is analytic in and u is harmonic in
the range of f . Then u f is harmonic in .
Proof. In a neighborhood of any point in its domain u has a conjugate function, so it is the real part of some analytic function g defined
near the point. Since the composite g f is analytic, its real part u f
is harmonic.
Exercise 5.5. Suppose u is harmonic in the region and that one

can find a bijective conformal mapping of onto the unit disk. Show
that u has a harmonic conjugate in .
The next theorem is also a simple corollary of Theorem 5.2, but it
is so important it is a theorem anyway.
Theorem 5.6 (Mean value property). Suppose u is harmonic in
the open disk centered at z with radius R, and continuous in the closed
disk. Then
Z2
1
u(z + Rei ) d .
u(z) =
2
0
5.1. FUNDAMENTAL PROPERTIES
75
Proof. In the open disk u is the real part of an analytic function

f , by Theorem R5.2. If 0 < r < R Cauchys integral formula implies
f ()
1
that f (z) = 2i
d. Parametrizing the circle by = z +rei ,
|z|=r z
0 2, gives
1
f (z) =
2
Z2
f (z + rei ) d .
0
Taking the real part of this gives the desired formula with R replaced
by r. By the continuity of the integrand, however, we may now let
r R and so obtain the desired result.
Clearly one can calculate mean values in the above sense for any
continuous function. Interestingly enough, any continuous function
having the mean value property has to be harmonic (and is therefore
also infinitely differentiable). We will show this in Theorem 5.11.
Theorem 5.7 (Maximum principle). Suppose u is continuous on
the closure of a bounded region and satisfies the mean value property
in . Then u takes its largest and smallest value in on , and if
either is assumed in an interior point, then u is constant.
Proof. Suppose a and sup = u(a). There is a disk |z a| <
R contained in , and u(a + rei ) u(a) for all and 0 < r < R.
If there is strict inequality for some choice of r, , Rthen there is strict
2
1
inequality in a neighborhood by continuity, and 2
u(a + rei d <
0
u(a), violating the mean value property.
Thus the set M = {z | u(z) = u(a)} is open, as is the complement {z | u(z) 6= u(a)} by continuity. Since is connected and
M 6= it follows that M = , so that u is constant.
Since u satisfies the mean value property if u does, the statement
about smallest value follows as well.
Harmonic functions satisfy the mean value property, so the theorem

applies to harmonic functions. We obtain a corollary, which is also
referred to as the maximum principle.
Corollary 5.8. Suppose u is harmonic and not constant in a
region . Then u has no local extrema in .
Proof. By Theorem 5.7 u is constant in a neighborhood of a local
extremum point a. Consider the set
M = {z | u() = u(a) for in a neighborhood of z}.
Clearly M is open. But if zj M , zj z , then any neighborhood
of z contains a disk where u is identically u(a). Therefore, near z the
function u is the real part of an analytic function which is constant on
an open set and therefore is constant. It follows that z M so M is
76
also relatively closed in . Since is connected and M 6= , it follows

that M = , i.e., u is constant in .
A problem of great importance both for the theory of harmonic

functions and their applications is Dirichlets problem. It concerns the
possibility of finding a function harmonic in a given region, continuous
on its closure, and taking prescribed values on the boundary. There are
also other, more general formulations which will not concern us here.
Note that if we can solve Dirichlets problem for some region , and
if we can find a conformal map of onto some other region which
extends continuously as an invertible map of the closure of onto
the closure of , then by Corollary 5.4, we can also solve Dirichlets
problem for the region . We first note:
Theorem 5.9. If Dirichlets problem has a solution for a bounded
region , then it is unique.
Proof. Suppose u and v are harmonic in , continuous in the closure and agree on . Then u v is harmonic in and vanishes on the
boundary. But according to Theorem 5.7 it takes both its largest and
smallest value on the boundary; we therefore have u = v throughout
.
To prove the existence of a solution is much harder, and requires

additional assumptions. We will here give a solution for the simple case
when is a disk centered at the origin. In the next section we will
show the existence of a solution in much more general circumstances.
We start by assuming that we have a function u, harmonic in |z| <
R and continuous in |z| R. We should like to express the values of
u in the interior of the disk in terms of its values on the boundary.
The mean value property gives us such a formula for the center of the
circle. An obvious way of trying to get a formula for other interior
points would be to use a Mbius transform to map the unit disk onto
the given disk in such a way as to map the origin to a given point a in
R+a
the disk. The map T () = R R+a
, which has inverse = R Rza
2 az , does
exactly that. Hence
1
u(a) = u(T (0)) =
2
Z2
1
u(T (e )) d =
2i
u(T ())
||=1
Setting z = T () in this integral gives (note that zz = R2 )

Z
1
R2 |a|2 dz
u(a) =
u(z)
.
2i
|z a|2 z
|z|=R
d
.
5.1. FUNDAMENTAL PROPERTIES
77
Setting z = Rei finally gives Poissons integral formula

(5.1)
1
u(a) =
2
Z2
u(Rei )
R2 |a|2
d .
|Rei a|2
What we have just seen is this: If Dirichlets problem for the disk
|z| < R has a solution, it must be given by Poissons integral formula.
2 |a|2
z+a
Note that R|za|
2 = Re za and that for |a| < R the integral
(5.2)
1
2
Z2
u(Rei )
Rei + a
d
Rei a
is an analytic function of a, as is seen by differentiating under the

integral sign. The real part of this integral is Poissons integral, so that
the imaginary part is a conjugate harmonic function to u in |z| < R.
But (5.2) is an analytic function whether u is harmonic or not, as long
as it behaves well enough on the boundary |z| = R for us to be allowed
to differentiate under the integral sign. Continuity is certainly enough.
It follows that Poissons integral represents a harmonic function for any
function u defined and continuous on |z| = R. We denote this function
by Pu , so that we know that Pu = u in the disk if u is known to be
harmonic in the interior and continuous on the closed disk.
If u is only defined and continuous on the boundary we still know
that Pu is harmonic in the interior. To show that Pu solves Dirichlets
problem, it only remains to show that it assumes the correct boundary
values. First note that, since a constant is harmonic, the integral of
1 R2 |a|2
the Poisson kernel 2
is 1 for all a, |a| < R. Since the Poisson
|Rei a|2
kernel is also positive, it follows that
1
|Pu (a) u(Rei )|
2
Z2
|u(Rei ) u(Rei )|
R2 |a|2
d .
|Rei a|2
Given > 0 we may find > 0 so that |u(Rei ) u(Rei )| < for
< < + . The integral over [0, ] [ + , 2] (if = 0,
over [, 2 ]) clearly tends to 0 as a Rei , and the integral over
[ , + ] (respectively [0, ] [2 , 2]) is < . It follows that
|Pu (a) u(Rei )| 0 as a Rei so that actually Pu tends to the
correct boundary values. We have proved the following theorem.
Theorem 5.10. Suppose u is a continuous function defined on |z| =
R. Then the function which equals Pu (z) for |z| < R and u(z) for
|z| = R is harmonic in |z| < R and continuous in |z| R.
In the process of solving Dirichlets problem we also obtained (5.2)
which expresses the values of a function analytic in the disk |z| < R in
78
terms of the boundary values of its real part, in the case when these are
assumed continuous. This is a well known theorem by H. A. Schwarz.
Theorem 5.11. Suppose u is continuous in a region C and
has the mean value property there. Then u is harmonic.
Proof. Let |z z0 | < R be an open disk with closure contained in
and Pu the Poisson integral applied to u( + z0 ). Then Pu ( + z0 ) is
harmonic in the disk so that Pu u satisfies the mean value property
in the disk and is continuous in its closure. Therefore Pu u satisfies
the maximum principle Theorem 5.7 in the closed disk. But Pu u
vanishes on the boundary of the disk and is therefore identically 0.
Thus Pu = u in the disk, so that u is harmonic.
We finally consider the reflection principle. In order to formulate

the theorem, let us call a region symmetric with respect to the real
axis if for each z it contains z if and only if it contains z. We denote
the intersection of with the real axis by and the part of which
is in the (open) upper half plane by + .
Theorem 5.12 (Reflection principle). Suppose v is continuous in
+ , vanishes on and is harmonic in + . Then v has a harmonic extension to satisfying the symmetry v(z) = v(z). If v is
the imaginary part of a function f analytic in + , then f has an analytic extension to satisfying f (z) = f (z).
Proof. If we define the extension of v by setting v(z) = v(z)
for z {z | Im z < 0} it is clear that v is continuous in and
harmonic except possibly on . Let p be an arbitrary point on . We
need to show that v is harmonic in a neighborhood of p. Let R > 0 be
so small that the disk |z p| R is contained in , and let Pv be the
Poisson integral corresponding to this disk, extended by continuity to
the boundary of the disk. Then Pv is harmonic in |z p| < R and we
will be done if we can prove that Pv coincides with v there.
Now Pv vanishes on the real diameter of |z p| < R because of
the symmetry in v, and the boundary values of Pv on |z p| = R
coincide with those of v, by Theorem 5.10. Hence the function Pv v,
which is harmonic in the half disk |z p| < R, Im z > 0, has vanishing
boundary values in this half disk. By the maximum principle Pv = v
in the half disk, so Pv is a harmonic extension of v to the whole disk,
and obviously has the same symmetry as v. It follows that Pv coincides
with v in the disk, so that v is harmonic there.
Now suppose f is analytic in + with imaginary part v there. Consider a disk as before with center on . In this disk v has a harmonic
conjugate u so that g = u + iv is analytic in the disk. Now g(z) is
also analytic in the disk and the function g(z) g(z) is analytic in the
disk, has zero imaginary part and vanishes on the real axis. It follows
5.2. DIRICHLETS PROBLEM
79
that this function is identically zero so that g has the appropriate symmetry. Since f g has zero imaginary part in the upper half circle it
is a real constant there. It follows that f can be extended analytically
as claimed.
5.2. Dirichlets problem

In this section we will solve the Dirichlet problem by Perrons
method. Recall first that one version of Dirichlets problem is the following:
Find a function u harmonic in a given region such
that u(z) f () as 3 z where f is a given
function on .
It is not hard to see that this problem can not be solved in general
without assumptions both on the boundary values f and the nature
of the boundary . We will impose such conditions later. Perrons
method, like many other methods for solving Dirichlets problem, consists in converting the problem of finding a solution to a maximization
problem. To explain how, we need to make the following definition.
Definition 5.13. A real-valued function v, defined and continuous in a region , is called subharmonic if for every u harmonic in a
of the function v u satisfies the maximum principle in
subregion
.
means that v u
That v u satisfies the maximum principle in
has no maximum in unless it is constant. The following theorem

gives a more concrete characterization of subharmonicity.
Theorem 5.14. A continuous function v is subharmonic in if
and only if
Z2
1
v(z0 )
v(z0 + rei ) d
2
0
whenever the disk |z z0 | r is contained in .

Proof. If the inequality holds, then it holds also for v u since u
has the mean value property. But the inequality is all that is needed
to prove the maximum principle (cf. the proof of Theorem 5.7) so that
one direction of the theorem follows.
Conversely, if v u satisfies the maximum principle for every harmonic u we may for u pick the Poisson integral Pv belonging to v on
the circle |z z0 | = r. Then v(z) Pv (z) approaches 0 as z approaches
the circle from its interior (Theorem 5.10). By the maximum principle
v Pv 0 in the disk; in particular, for z = z0 we obtain the desired
inequality.
We list some elementary properties of subharmonic functions.
80
(1) If v is subharmonic in , then so is kv for any non-negative

constant k.
(2) If v1 and v2 are subharmonic in , then so is v1 + v2 .
(3) If v1 and v2 are subharmonic in , then so is max(v1 , v2 ).
(4) If v is subharmonic in , D is a disk whose closure is in
and Pv is the Poisson integral corresponding to this disk with
boundary values given by v, put v = Pv in D and v = v in
\ D. Then v is subharmonic in .
The first two properties are immediate consequences either of the
definition or of Theorem 5.14. The other properties are only a little
less obvious.
Proof of (3). Let v = max(v1 , v2 ) and suppose v u has a max , where u is harmonic in .
We may assume
imum at z0
v(z0 ) = v1 (z0 ). We then have
v1 (z) u(z) v(z) u(z) v(z0 ) u(z0 ) = v1 (z0 ) u(z0 )
It follows, first that v1 u is constant, and then from the
for z .
same inequality that v u is constant. Hence v is subharmonic.
Proof of (4). By Theorem 5.10 v is continuous. We have v Pv

in D so v v throughout . Since Pv is harmonic and v subharmonic
it follows that v is subharmonic except possibly on D. But if v u
has a maximum at a point on D, then so has v u so that v u
is constant. But then it follows that also Pv u and hence v u is
constant.
Note that any harmonic function is also subharmonic. It follows

by the maximum principle that it is greater than any subharmonic
function with smaller boundary values. If we therefore let F denote
the set of all functions v subharmonic in which have the additional
property that lim v(z) f () for every , then the solution of
3z
Dirichlets problem, if it exists, ought to be the largest element of F.

To make sure that F is not empty we now assume that f is bounded,
|f ()| M for all . It follows that any constant M is
in F, so F is definitely not empty. A less important, but convenient,
assumption we will make is that also is bounded. We now set
u(z) = sup v(z) ,
z,
vF
expecting this to be the solution of Dirichlets problem, if it exists. In

fact, with no further assumptions, u is harmonic in .
Lemma 5.15. The function u defined above is harmonic in .
To be able to prove Lemma 5.15 we need the following important
lemma.
81
Theorem 5.16 (Harnacks principle). Suppose u1 , u2 , . . . is an increasing sequence of functions harmonic in a region . Then either
un + locally uniformly in , or else un converges locally uniformly to a function u which is harmonic in .
Proof. Suppose u is harmonic in a closed disk |z z0 | . The
Poisson integral formula then states that for z in the open disk
1
u(z) =
2
Z2
0
2 r 2
u(z0 + ei ) d ,
i
2
|e (z z0 )|
where r = |z z0 |. Since r |ei (z z0 )| + r by the triangle

inequality the first factor in the integral can be estimated by
r
2 r 2
+r
i
2
+r
|e (z z0 )|
r
If now u is non-negative in the disk we obtain Harnacks inequality
r
+r
u(z0 ) u(z)
u(z0 ) ,
+r
r
by the Poisson integral formula and the mean value property. Now
suppose r /2. Then Harnacks inequality shows that
1
u(z0 ) u(z) 3u(z0 ) .
3
Now consider the sequence u1 , u2 , . . . . Since the sequence is increasing it has a pointwise limit everywhere in , which is either finite or
+. If n > m the function un um is positive and harmonic in
so we can apply Harnacks inequality to it. It follows from (5.3) that
if un (z0 ) +, then un + uniformly in a neighborhood of z0 .
It also follows that the set where un tends to + is an open subset
of . Similarly, if un (z0 ) has a finite limit, then the limit is finite in
a neighborhood of z0 so the set where the limit is finite is also open.
Since is connected it follows that either un tends locally uniformly
to + in , or else the limit function u is finite everywhere. Applying
(5.3) to un um and letting n we get
(5.3)
0 u(z) um (z) 3(u(z0 ) um (z0 ))

so that the convergence is locally uniform. Finally, to see that u is
harmonic we may apply the Poisson integral formula to un over any
circle contained in and take the limit under the integral sign, by
uniform convergence. It follows that locally u is given by its Poisson
integral so that u is harmonic. The proof is complete
Lemma 5.17. Suppose v is subharmonic in and for some constant

K we have lim v(z) K for every . Then v K in .
3z
82
Proof. If > 0 there is a neighborhood of where v < K + .

It follows that the set E = {z | v(z) K + } is closed and since
it is bounded (as a subset of the bounded set ), it is in fact compact.
If E 6= it follows that v has a maximum in E, which will also be an
interior maximum in . It would follow that v is constant K+ which
contradicts the assumption about the boundary behavior. Hence E is
empty, and since > 0 is arbitrary the desired conclusion follows.
Proof of Lemma 5.15. First note that by Lemma 5.17 v M
for all v F. It follows that u is finite everywhere in . Now let
z0 . We may then choose a sequence v1 , v2 , . . . from F such that
vn (z0 ) u(z0 ). We also have vn (z0 ) u(z0 ), n = 1, 2, . . . . Now
let Vn = max(v1 , . . . , vn ). By property (3) of subharmonic functions
Vn F and vn (z0 ) Vn (z0 ) u(z0 ) so we have Vn (z0 ) u(z0 ). In
addition the sequence V1 , V2 , . . . is increasing. Now choose a disk D
containing z0 and whose closure is in and let Vn equal Vn outside
D and the Poisson integral of Vn over D in D. By property (4) of
subharmonic functions also Vn F so Vn u and it is Vn by the
maximum principle. Hence Vn (z0 ) u(z0 ) and V1 , V2 , . . . is increasing.
Since Vn is harmonic in D we may apply Harnacks principle, and since
Vn (z0 ) u(z0 ) < it follows that Vn U locally uniformly in D,
where U is a harmonic function for which U (z0 ) = u(z0 ).
Now let z1 be an arbitrary point of D. As before we can then find
a sequence w1 , w2 , . . . in F such that wn (z1 ) u(z1 ). If we set wn =
max(wn , vn ) we still have elements of F, the limit at z1 is unchanged
and we also have wn vn . We continue similar to what we did above,
n equal to Wn outside D and
setting Wn = max(w
1 , . . . , wn ) and then W
equal to the corresponding Poisson integral inside D. The sequence
1, W
2 , . . . is then in F, harmonic in D, increasing and W
n (z1 )
W
n Vn so that W
n (z0 ) u(z0 ). As before
u(z1 ). We also have W
n U1 locally uniformly, where U1 is
it follows that in D we have W
harmonic, U1 (z1 ) = u(z1 ), U U1 and also U1 (z0 ) = u(z0 ) = U (z0 ).
The harmonic function U U1 is therefore non-positive but 0 in z0 .
By the maximum principle it is constant and therefore identically 0.
It follows that U (z1 ) = u(z1 ). Since z1 is an arbitrary point of D it
follows that U = u in D so that u is harmonic in a neighborhood of
every point z0 . The proof is complete.
To deal with the question whether u assumes the desired boundary

values we need to introduce the concept of a barrier function.
Definition 5.18. A barrier for at a point is a function
w harmonic in and continuous in , and such that w() = 0 but w
is strictly positive in all other points of .
The following lemma reduces the question of whether u takes the
desired boundary values to the question of finding barriers.
83
Lemma 5.19. Suppose f is continuous at a point 0 and there

is a barrier for at 0 . Then u(z) f (0 ) as 3 z 0 .
Proof. We will show that we have
that
lim u(z) f (0 ) + and
3z0
lim u(z) f (0 ) for every > 0 from which the theorem
3z0
follows.
Let > 0 and choose a neighborhood O of 0 such that |f ()
f (0 )| < for O . Furthermore, let w0 be the minimum
of w over the (compact) set \ O. By the properties of w we have
w0 > 0. Now put V (z) = f (0 ) + + w(z)
(M f (0 )). Then V is
w0
harmonic in and continuous in the closure. For O we have
V () f (0 ) + > f (). For \ O we have w() w0 so we
get V () M + > f (). If v F and we therefore have
lim (v(z) V (z)) < 0 so by Lemma 5.17 v V in . It follows that
3z
also u V in so that
lim u(z) V (0 ) = f (0 ) + .
3z0
To prove the other inequality, set W (z) = f (0 ) w(z)

(M +f (0 )).
w0
Again W is harmonic in and continuous in the closure. For O
we have W () f (0 ) < f () and for \O we have w() w0
so that we get W () M < f (). It follows that W F so
that W u. Hence lim u(z) W (0 ) = f (0 ) . The proof is
complete.
3z0
It is sometimes easy to find a barrier. For example, suppose a point

has a supporting line, i.e., a line which intersects the closure of
only in , and let be the direction of the line, chosen so that is
to the left of it. Then Im(ei (z ) is a barrier for at . Show this!
If is strictly convex, then every boundary point has a supporting line
so there is a barrier for at every boundary point. To state a more
general result, we make the following definition.
Definition 5.20. A region is said to have the segment property
at a boundary point if there exists a line segment exterior to except
that one endpoint is .
A continuous curve without self-intersections is called a
free boundary arc of if every point on is the center of a disk which
is split in exactly two components by . It is called one-sided if one
of the components is always in and the other not.
It is clear that if is a free onesided boundary arc of and has
a normal at a point , then has the segment property at ; one
only has to choose a sufficiently short piece of the exterior normal.
Lemma 5.21. A region has a barrier at any boundary point where
it has the segment property.
84
Proof. Suppose has the segment property at and that

the other endpoint of the corresponding segment is p. We can then
choose a complex number a such that the segment is mapped onto
, the image of being 0. Using
the negative real axis by z 7 a z
zp
q
the principal branch of the root it is then obvious that Re a z
is a
zp
barrier for at .
We collect our results about Dirichlets problem in the following

theorem.
Theorem 5.22. Suppose is a bounded region having the segment
property at each of its boundary points. Then Dirichlets problem has
a unique solution in for arbitrary boundary values f continuous on
.
Proof. We have proved everything claimed in this section except
the uniqueness; but this is Theorem 5.9.
CHAPTER 6
Entire functions
6.1. Sequences of analytic functions
In this section we shall consider sequences of analytic functions
which are uniformly convergent. We will use the notation H() for
the functions holomorphic (analytic) in the region C. By a region
we will always mean an open, connected set. Recall that we say that
a sequence f1 , f2 , . . . of real or complex-valued functions defined on a
set E is uniformly convergent on E to another function f defined on E
provided that for each > 0 we can find a number N such that if n N
then |fn (z)f (z)| < for every z E. If one introduces the maximumnorm k kE by setting kf kE = supzE |f (z)| the uniform convergence
of fn to f on E is equivalent to kfn f kE 0 as n . When
dealing with functions defined in an open set C (or Rn ) one
often talks about locally uniform convergence. A sequence of functions
f1 , f2 , . . . defined in is said to converge locally uniformly to f in
if every x has a neighborhood in which the sequence converges
uniformly to f . Equivalently, this means that fn f uniformly on
every compact subset of . This is an immediate consequence of the
Heine-Borel lemma.
Exercise 6.1. Show this equivalence!
Recall that in Flervariabelanalys it is proved that the uniform limit
of continuous functions is continuous. It immediately follows that the
same is true of locally uniform limits of continuous functions (explain
why this is obvious!). When dealing with analytic functions one can
say a lot more.
The main result of the section is the following.
Theorem 6.2. Suppose fn H(), n = 1, 2, 3, . . . and that fn f
(j)
locally uniformly in . Then f H(). Furthermore, fn f (j)
locally uniformly in for j = 1, 2, 3, . . . .
Proof. Let be a positively oriented circle such that the corresponding closed disk is contained in . For z in the open disk we then
have (Corollary 3.13)
Z
j!
fn (w) dw
(j)
.
(6.1)
fn (z) =
2i
(w z)j+1
85
86
6. ENTIRE FUNCTIONS
Since fn fR uniformly on the closed disk the integral on the right

f (w) dw
converges to (wz)
For j = 0 the left hand side
j+1 as n .
converges to f (z), so that f satisfies the Cauchy integral formula; thus
by Lemma 3.10 f is analytic in a neighborhood of every point of so
that f H(). By uniform convergence the right hand side of (6.1)
converges (pointwise) to f (j) (z). Suppose has radius r. I claim that
this convergence is uniform for z in the disk of radius r/2 concentric to
, which would prove locally uniform convergence and thus finish the
proof.
To verify the claim, note that for z in the sub-disk and w we
have |z w| r/2 so that
Z
Z
fn (w) dw
f (w) dw
j+1
(w z)
(w z)j+1
Z
(r/2)j1 |fn (w) f (w)| |dw| 2r(r/2)j1 kfn f k .
Since fn f uniformly on this shows the uniform convergence.
Theorem 6.2 was first proved by Weierstrass in a slightly different

formulation which we state as a corollary.
PCorollary 6.3. Suppose f1 , f2 , . . . are all in H() and the series
k=1 fk converges locally uniformly on . Then the series converges
to a function in H(), it may be differentiated termwise any number
of times, and the differentiated series all converge locally uniformly in
.
This is obviously equivalent to Theorem 6.2. We prove one more
result (by A. Hurwitz) on uniform convergence.
Theorem 6.4. Suppose fn H() for n = 1, 2, . . . and that fn
f locally uniformly on as n . Suppose furthermore that none of
the functions fn assume the value w in . Then neither does f , unless
f is constant (= w).
Proof. Replacing fn by fn w and f by f w we may as well
assume that w = 0. Assume that f is not identically zero. We must
then prove that f has no zeros in . We know, since f H(), that
the zeros of f are isolated, so any point of is the center of a closed
disk contained in and such that f has no zeros on the boundary
circle. If is the positively oriented boundary of such
R a0 disk, then the
1
number of zeros of f in the open disk is given by 2i f /f , so we need
to show that any such integral is 0.
Since z 7 |f (z)| is continuous and compact, |f | assumes a minimum m on which is > 0 since f has no zeros on . Since fn f
6.2. INFINITE PRODUCTS
87
uniformly on we have |fn | m/2 on for all sufficiently large n. So,

for z and sufficiently large n we have
|
1
1
|f (z) fn (z)|
2
|=
2 kfn f k .
fn (z) f (z)
|fn (z)f (z)|
m
Thus 1/fn 1/f uniformly on , soRby TheoremR 1.2 it follows that

fn0 /fn f 0 /f uniformly on . Thus fn0 /fn f 0 /f . But all the
integrals on the left equal 0, because fn are all zero-free in . It follows
that the limit is also 0, and the proof is complete.
As an almost immediate consequence we have the following interesting corollary about so called univalent functions. A univalent function
is an injective (one-to-one) analytic function.
Corollary 6.5. Suppose fn H(), n = 1, 2, . . . , and fn f
locally uniformly in . If all fn are univalent, then so is f , unless it is
constant.
Proof. Assume f is not constant. Then, if f (z0 ) = w, we must
show that f (z) w 6= 0 for z \ {z0 }. Setting gn (z) = fn (z) fn (z0 )
we have gn f w locally uniformly in . Since by assumption gn
does not vanish in \ {z0 }, neither does f w by Theorem 6.4.
Exercise 6.6. Show that for any > 0 there exists N such that all
P (1)k 2k+1
Taylor polynomials of sin x (partial sums of
x
) of degree
(2k+1)!
at least N has exactly one zero in ( , + ).
Exercise 6.7. A famous theorem by Weierstrass states that any
function continuous on a real interval [a, b] is the uniform limit of a
sequence of polynomials. Why does this not contradict Theorem 6.2?
6.2. Infinite products
Any analytic function may be expanded in a power series centered
at any point of the domain of analyticity; the radius of convergence is
such that on the boundary of the disk of convergence there is at least
one singularity of the function. If the function is analytic everywhere
in C, the radius of convergence is therefore infinite. Such a function
is called entire (in Britain often also integral ). A power series used to
be viewed as a polynomial of infinite order, especially if the radius of
convergence is infinite. The reason is of course that many properties
of polynomials have their counterpart for entire functions.
One of the more fundamental properties of a polynomial is that,
according to the fundamental theorem of algebra and the factor theorem, it may be factored into a product of first degree polynomials,
each of which vanishes at one of the zeros of the polynomial.
If p is a
Qn
polynomial of degree n one usually writes p(z) = A k=1 (z zk ), where
z1 , z2 , . . . are the zeros of p, repeated according to multiplicity, and A
88
6. ENTIRE FUNCTIONS
is the highest order coefficient of p. Clearly this does not generalize

to entire functions; a polynomial of infinite degree can hardly have a
highest order coefficient. But one may also write
Y
z
(6.2)
p(z) = Bz j
(1 ) ,
zk
z 6=0
k
where B is the coefficient of the non-zero term in p with lowest degree,

and j is the multiplicity of z = 0 as a zero of p (so that j = 0 if
p(0) 6= 0). As we shall see in the next section this expansion has a
generalization to arbitrary entire functions. In this section we shall
prepare the ground for this by considering
Qinfinite products.
What meaning should one assign to Q k=1 Ak ? The obvious answer
is to consider the partial products Pn = nk=1 Ak and then assign to the
infinite product the value limn Pn if the limit exists. This is almost
right, but note that the limit is 0 if just one factor is zero, completely
independent of the values of all the other factors. This does not seem
reasonable, so one makes the following modified definition.
Q
Definition 6.8. The infinite product
k=1 Ak is said to converge
to P if
(1) The sequence of partial products converge to P .
(2) There are only a finite number of zero factors in the product,
and the sequence of partial products obtained by excluding
these factors converge to a non-zero number.
If Pn P 6= 0 as n it follows that An = Pn /Pn1 P/P = 1
as n , so the factors in a convergent product always tend to 1. It
is therefore convenient to write infinite products on the form
Y
(6.3)
(1 + ak ) ,
k=1
so that the necessary condition for convergence just derived takes the
following form.
Proposition 6.9. A necessary (but not sufficient) condition for
convergence of the infinite product (6.3) is that ak 0 as k .
Since a sequence has a non-zero limit precisely if the sequence of
logarithms has a finite limit it is natural to compare the infinite product
(6.3) with the series with terms log(1 + ak ). Recall that the principal
branch of the logarithm is Log z = ln |z|+i arg z, where < arg z .
Theorem 6.10. If ak 6= 1, k = 1, 2, . . . , then the infinite product
(6.3) converges if and only if the series
X
(6.4)
Log(1 + ak )
k=1
converges. Here Log denotes the principal branch of the logarithm.
6.2. INFINITE PRODUCTS
89
Proof. Since the terms of a convergent series must tend to 0, and

by Proposition 2.2, we must have ak 0 if either the product (6.3) or
the series (6.4) converges. If Sn denotes the partial sum of the series
we have Pn = eSn so that the convergence of the product follows from
that of the series.
Conversely, assume that Pn P 6= 0 and choose a branch of the
logarithm which is continuous in a neighborhood of P . Then log Pn
log P . We have Sn = log Pn + 2kn i, where kn is an integer. Thus
Sn Sn1 = log Pn log Pn1 + 2(kn kn1 )i. But since ak 0 the
imaginary part of Sn Sn1 tends to 0. Since also log Pn log Pn1 0
it follows that kn kn1 0 so that, since kn is an integer, all kn
equal a fixed integer k from a certain value of n on. This means that
Sn log P + 2ki so the proof is complete.
We will, by definition, sayQ

that the product (6.3) is absolutely convergent if the infinite product
k=1 (1+|ak |) converges. By Theorem 2.3
this is equivalent to the convergence of the positive series
(6.5)
log(1 + |ak |) .
k=1
Log(1+z)
z
Noting that
1 as z 0 (using the principal branch of the
logarithm) it follows by a standard comparison theorem that the series
(6.4) (omitting
Pterms for which ak = 1) is absolutely convergent if
and only if
k=1 ak is absolutely convergent (note that if either of
the two series are convergent, then we must have ak 0Pas k
). In particular it follows that (6.5) converges if and only
k=1 |ak |
converges. So we have proved the following proposition.
Proposition
6.11. The product (6.3) converges absolutely if and
P
only if
k=1 ak converges absolutely. This is also equivalent to the
series (6.4) converging absolutely, after omitting (the finite number of )
terms for which ak = 1.
We now turn to the case when the factors of (6.3) are functions of
z C. By inspection of the proofs it is clear that all the results obtained so far remain true if we replace convergence by locally uniform
convergence. So by Theorem 1.2, if ak H() for every
P k, then (6.3)
converges locally uniformly to a function in H() if
|ak | converges
locally uniformly in . In particular, by Weierstrass majorization theorem (Weierstrass M -test
P in most English language books) it follows
that this is the case if
kak kK converges for every compact K .
We can now return to the problem of generalizing the polynomial
factorization (6.2) to an arbitrary entire function. Suppose that we
have an entire function for which 0 is a zero of multiplicity j which
also has other zeros a1 , a2 , . . . , repeated according to multiplicity. By
analogy with (6.2) our candidate for this function would then be a
90
6. ENTIRE FUNCTIONS
Q
z
constant multiple of z j
k=1 (1 ak ). This may not be so, however.
First of all, there are entire functions with no zeros at all. One example
is ez ; more generally, eg(z) is such a function for any entire function g.
We would certainly have to allow such a factor in front of the product
to obtain a generally valid factorization. Furthermore, for the product
P 1
to converge absolutely for some z 6= 0 we must require that
|ak |
converges; this may not always hold, although it is true that we always
have ak as k (Exercise P
6.12). For example, the function
sin(z) has zeros 0, 1, 2, . . . and
1/k is divergent. A little more
effort is therefore required to obtain a general factorization formula for
entire functions. We will carry this out in the next section.
Exercise 6.12. Prove that if a1 , a2 , . . . are the zeros of an entire
function, repeated according to multiplicity, then ak as k .
6.3. Canonical products
Consider first the case of an entire function f with only finitely
many non-vanishing zeros a1 , . . . , an , as always counted with multiplicities. If the multiplicity of 0 as a zero is j 0 it is clear that
j
h(z) = f (z)z /
n
Y
(1
k=1
z
)
ak
is an entire function without zeros. Thus also h0 (z)/h(z) is entire,

so it has an entire primitive g. Differentiating h(z)eg(z) we obtain
0 (z)
h0 (z)eg(z) h(z) hh(z)
eg(z) = 0 so that h is a constant multiple of eg .
By adding an appropriate constant to g, if necessary,
we may assume
Qn
z
g
j g(z)
that h = e . Thus we obtain f (z) = z e
k=1 (1 ak ) for some
entire function g. If f has infinitely many zeros the same reasoning
gives the representation
(6.6)
f (z) = z j eg(z)
(1
k=1
z
),
ak
with an entire function g, provided that the infinite product converges

P
locally uniformly. This is ensured, by the previous chapter, if
|ak |1
converges.
Example 6.13. The function f (z) = sinz z , where z is any branch

of the root, has zeros (k)2 , k = 1, 2, . . . and no others. It is an entire
P (1)k k
function since it has a power series expansion
z . This follows
(2k+1)!
immediately from the expansion of sin z. From this Euler proved his
famous formula
X
1
2
(6.7)
=
n2
6
n=1
6.3. CANONICAL PRODUCTS
91
using the following argument. According to (6.2) the coefficient of z in

any polynomial with constant coefficient 1 is the negative of the sum
of the reciprocals of the zeros of the polynomial. The power series for
f has constant coefficient 1 and the coefficient
of z is 1/6. The sum
P
2
of the reciprocals of the zeros of f is
(k)
. Hence (6.7) follows.
k=1
Well. . .
The argument can be quite easily justified if we assume that we
know that
sin z Y
z
(6.8)
=
(1 2 2 ) .
k
z
k=1
It is at least clear
the product converges absolutely and locally
P that
2
uniformly since
1/k converges.
P The partial product Pn is a polynomial and its z-coefficient is nk=1 (k)2 . This is the value of Pn0 (0),
and according to Theorem 6.2 this converges to the derivative at 0 of
the infinite product. Assuming (6.8) the power series for f shows that
this is 1/6 and so Euler would be vindicated.
Instead of proving (6.8) we will show the equivalent statement
sin z Y
z2
(6.9)
(1 2 ) ,
=
z
k
k=1
obtained by replacing z in (6.8) by 2 z 2 . Since the infinite product
converges absolutely locally uniformly we at least know from (6.6) that
Y
z2
sin z
g(z)
(1 2 ) ,
=e
z
k
k=1
where g is entire. Taking the logarithmic derivative of both sides we
obtain
X
1
1
0
+
,
(6.10)
cot z 1/z = g (z) +
zk z+k
k=1
and differentiating once more we obtain
X
2
1
00
=
g
(z)
+
.
2
2
(z
+
k)
sin z
k=
However, both the infinite sum and the left hand side are of period 1
here, so g 00 also has period 1. Writing z = x + iy it is easy to see that
both the infinite sum and the left hand side tend to 0 locally uniformly
in x as y . Thus the same is true of g 00 which is therefore bounded
and entire, hence by Liouvilles theorem constant; since g 00 (iy) 0 as
y + the constant is 0. Therefore g 0 is constant. But from (6.10)
follows that g 0 is odd, so that also g 0 vanishes. Thus g is constant, so
that (6.9) holds apart from a constant factor. Since both sides tend to
1 as z 0 this factor is 1, and we have finally established (6.9), and
thereby Eulers formula.
92
6. ENTIRE FUNCTIONS
Exercise 6.14. Justify all unproved claims at the end of Example 6.13.
What is one to do to obtain a factorization for an entire function
where the sum of the reciprocals of the zeros is not absolutely convergent? The idea is to replace the factor 1 azk in the product by
(1 azk )epk (z) , where pk is an entire function which promotes convergence without introducing new zeros. As we shall see, one can always
choose pk to be a polynomial. Convergence is obtained by choosing pk
so that (1 azk )epk (z) is sufficiently close to 1, so the ultimate choice
would be log(1 azk ). Unfortunately this is not an entire function.
It is therefore natural to attempt to choose pk as a Taylor polynomial of this function of sufficiently high degree. Now for the princiP z k
pal branch of the logarithm Log(1 z) =
k=1 k and the series
Pn z k
converges for |z| < 1. In fact, if we set Ln (z) =
k=1 k we have
Log(1 z) = Ln (z) + Rn (z), where an easy estimate gives
X
X
|z|k
|z|n+1
|Rn (z)|
|z|k =
k
1 |z|
k=n+1
k=n+1
for |z| < 1. According to Proposition 6.11 the product

(6.11)
(1
k=1
z pk (z)
)e
ak
converges absolutely and locally uniformly in z precisely if the series

(6.12)
X
z
{Log(1 ) + pk (z)}
ak
k=1
does (check this carefully!). We assume now |z| < R. There are only
finitely many factors in (6.11) for which |ak | < 2R (Exercise 2.5) so excluding these factors from the product will not affect convergence. We
may thus also assume that |ak | 2R. If we choose pk (z) = Lnk (z/ak )
the absolute value of the term in (6.12) is |Rnk (z/ak )| and may therefore be estimated by 21nk , using the facts that R/|ak | 1/2 and
1 R/|ak | 1/2. We conclude that (6.12) converges absolutely and
uniformly
P n in |z| < R if we can choose nk for every k such that the series
k converges. A obvious choice that works is n
k = k. Since
k=1 2
R is arbitrary we conclude that the choice pk (z) = Lk (z/ak ) makes
(6.11) absolutely and locally uniformly convergent. We have proved
the following theorem by Weierstrass.
Theorem 6.15. There exists an entire function with arbitrarily prescribed non-vanishing zeros a1 , a2 , . . . (repeated according to multiplicity), provided they are either finitely many or else ak as k .
6.3. CANONICAL PRODUCTS
93
Every entire function with these and no other zeros may be written
Y
z L (z)
(6.13)
f (z) = z j eg(z)
(1 )e nk ak ,
ak
a 6=0
k
P
k
where g is an entire function, Ln (z) = nk=1 zk , and nk , k = 1, 2, . . .
are certain (sufficiently large) positive integers. A possible choice is
nk = k.
The theorem has a very important corollary concerning meromorphic functions. Recall that a function f is called meromorphic in if
it is analytic in except for isolated singularities which are poles.
Corollary 6.16. Every function which is meromorphic in the
whole plane is the quotient of two entire functions.
Proof. If f is meromorphic in the whole plane we may, according
to Theorem 3.3, find an entire function g so that all the poles of f are
zeros of g, and with the same multiplicities. Thus h = f g is an entire
function and f = h/g.
The expansion (6.13) becomes particularly interesting

Pif one may
choose nk = h independent of k. This is the case if
|Rh (z/ak )|
converges absolutely
uniformly
for
|z|
R
for
any
R.
Since
ak
P
P
h+1
h+1
h+1
this happens if (R/|ak |)
=R
1/|ak |
converges. In other
words, if the zeros do not tend too slowly
to
infinity.
Suppose now
P
h+1
that h is the smallest integer for which
1/|ak |
converges (so that
h 0). Then
Y
z
(1 )eLh (z/ak )
ak
k=1
is called the canonical product associated with the sequence a1 , a2 , . . . ,
and the integer h is called the genus of the canonical product. If possible we use the canonical product in the expansion (6.13). In that case
the expansion becomes uniquely determined by f . If it then happens
that g is a polynomial, one says that the function f has finite genus,
and the genus of f is the degree of g or the genus of the canonical
product, whichever
is the largest. This means for example that the
function sin z/ z considered in Example 6.13 and with the product
expansion (6.8) is of genus 0.
Example
P 6.17. The function
P sin z has all the integers as its zeros,
and since 1/n diverges but 1/n2 converges we obtain an expansion
of the form
Y
z
sin z = zeg(z)
(1 )ez/k .
k
k6=0
If we group the factors for k together and compare the result to (6.9)
it follows that g is the constant log . Consequently, sin z is of genus
94
6. ENTIRE FUNCTIONS
1 and has the canonical expansion

Y
z
(6.14)
sin z = z (1 )ez/k
k
k6=0
Exercise 6.18. If f has genus h, what is the possible range for the
genus of z 7 f (z 2 )?
Exercise 6.19.PLet a1 , a2 , . . . be a sequence satisfying 0 < |ak | < 1
for all k for which
k=1 (1 |ak |) converges. Show that the product
Y
ak ak z
|ak | 1 ak z
k=1
(a so called Blaschke product) converges to a function holomorphic in

the unit disk with the given sequence as zeros.
6.4. Partial fractions
As we have seen a meromorphic function is the quotient of two entire
functions, and thus the analogue of a rational function. A fundamental fact about rational functions is that they allow a partial fractions
expansion. In fact, if r(z) = p(z)/q(z) where p and q are polynomials
without common factors, then one may write
r(z) = g(z) +
n
X
k=1
Pk (
1
)
z ak
where g and all Pk are polynomials, a1 , . . . , an the different zeros of

q, and deg Pk = nk where nk is the multiplicity of ak as a zero of q.
1
Note that Pk ( za
) is the singular part of r at ak as a meromorphic
k
function. For a function meromorphic in the whole plane one would
therefore expect a similar expansion, where now g is entire and n may
be infinite. This leads to Mittag-Lefflers theorem, although the sum
has to be slightly modified to ensure convergence.
Theorem 6.20 (Mittag-Leffler). Let a1 , a2 , . . . be a sequence converging to and let Pk be polynomials without constant terms. Then
there are functions meromorphic in the whole plane with poles precisely
1
). The most general such
at ak and corresponding singular part Pk ( za
k
meromorphic function may be written
(6.15)
X
f (z) = g(z) +
(Pk (
k=1
1
) qk (z)) ,
z ak
where g is entire and qk suitably chosen polynomials.

Proof. If ak = 0 we choose qk = 0. If ak 6= 0 the function h(z) =
1
Pk ( za
) is analytic at 0 and we will choose for qk the corresponding
k
6.4. PARTIAL FRACTIONS
95
Taylor polynomial of degree nk . If is the circle || = |ak |/2 and z a

point inside the circle we then have
Z
nk
nk
X
X
1
h(k) (0) k
1
zk
h(z)
z =
} d .
h(){
k+1
k!
2i
k=0
k=0
Summing the geometric series we obtain

Z
z nk +1
h() d
h(z) qk (z) =
.
2i
( z) nk +1
1
Supposing |Pk ( za
)| Mk for |z| = |ak |/2 we obtain
k
|h(z) qk (z)| 3Mk
2|z| nk +1
|ak |
for |z| |ak |/3. Consider a disk |z| R. There are only finitely
many of the ak with |ak | < 3R, and it is clear that after removing the
corresponding
terms (6.15) converges uniformly in |z| R if the series
P
2R nk
Mk ( |ak | ) converges. We may consider this a power series in R, and
it will then have infinite radius of convergence if the terms tend to 0 for
)nk
every R > 0. Choosing nk log Mk we have Mk ( |a2Rk | )nk ( 2eR
|ak |
0 as k since ak . Thus the sum in (6.15) represents a
meromorphic function with the same singular parts as f in all poles,
so the theorem follows.
Example 6.21. The function

P 1 cot z has simple poles with residue
1 at every integer. Since
does not converge, we must include a
zk
convergence term in the partial fractions expansion.
P 1The constant
P term
1
z
in the Taylor expansion of zk
is k1 , and since ( zk
+ k1 ) =
k(zk)
P 1
converges like
, we must have
k2
1 X 1
1
(6.16)
cot z = g(z) + +
+
z k6=0 z k k
with an entire function g. Comparing this with (6.10) it is clear that
g 0. Differentiating the expansion we obtain the partial fractions
expansion
X
1
2
.
=
2
sin z k= (z k)2
As a final example, consider the function sinz which has simple poles
at the integers, with residue 1 at even and 1 at odd integers. The
partial fractions expansion is thus of the form
(1)k
= g(z) +
sin z
zk
k=
96
6. ENTIRE FUNCTIONS
with an entire function g. The series is not absolutely convergent here,

which would be ensured by choosing qk = (1)k+1 k1 for k 6= 0. However,
P
(1)k
the alternating series
converges, so the series is locally unik=1
k
formly convergent (away from the integers) as it stands. To determine
g, note that the sum may be rewritten
1
1 X
1
+
+
(1)k
z k=1
zk z+k
1 X 1
1
1
1
1
= +
+
+
z k=1 z 2k z + 2k
z 1 k=1 z 1 2k z 1 + 2k
Thus comparing to (6.16) we obtain
z
(z 1)
g(z) = cot
cot
.
2
2
2
2
sin z
But the right hand side is easily seen to vanish identically, so we have
the partial fractions expansion
(1)k
=
sin z k= z k
Exercise 6.22. Evaluate
1
.
(z + k)2 + a2
k=
6.5. Hadamards theorem

In this section we will prove a fundamental theorem by Hadamard
connecting the growth rate at infinity of an entire function with the
distribution of its zeros. As we know, the genus of an entire function
gives information about the distribution of its zeros a1 , a2 , . . . , since if
the genus P
is h the function either has only finitely many zeros, or else
the series
1/|ak |h+1 converges. We now introduce a measure for the
growth at infinity of an entire function f . First denote by M (r) the
maximum of |f (z)| on the circle |z| = r.
Definition 6.23. The order of an entire function is defined by
log log M (r)
.
= lim
r
log r
+
This means that is the smallest number such that |f (z)| < e|z|
for any given > 0 as soon as |z| is sufficiently large. Consequently,
polynomials have order 0, ez and sin z have order 1, ep(z) has order n
z
if p is a polynomial of degree n, and ee has infinite order. Note that
whereas the genus is always a natural number (or infinity), the order
may be anynon-negative number (or infinity); for example, the entire
function sinz z we discussed earlier has order 1/2 (show this!).
6.5. HADAMARDS THEOREM
97
Theorem 6.24 (Hadamard). The genus h and order of an entire

function satisfy h h + 1.
The proof needs a bit of preparation. Recall that the real and imaginary parts of an analytic function of z = x+iy are harmonic functions.
This means in particular that log |f (z)| is a harmonic function wherever f is analytic and 6= 0, since in a neighborhood of such a point one
may define a branch of log f (z), which has real part log |f (z)|. Furthermore, if u is harmonic in a neighborhood of |z| , then it satisfies
the Poisson integral formula (see (5.1))
Z2 2
|z|2
1
u(z) =
u(eit ) dt
2
|z eit |2
0
forR|z| < . In particular, we have the mean value property u(0) =

2
1
u(eit ) dt.
2 0
If f is analytic in the disk |z| and never 0, we can apply
Poissons integral formula to log |f (z)|. If f has zeros inside the circle
we instead obtain Poisson-Jensens formula.
Theorem 6.25. Suppose f H() where contains the disk |z|
and that f has only the zeros a1 , . . . , an in |z| < , and no zeros on
|z| = . Then the Poisson-Jensen formula
(6.17)
Z2
n
X
2 ak z
1
eit + z
+
log
log |f (z)| =
Re it
log |f (eit )| dt
(z
a
2
e
z
k)
k=1
0
is valid if |z| < is not one of the zeros. In particular, if f (0) 6= 0 we

have Jensens formula
Z2
n
X
1
log |f (0)| =
log
+
log |f (eit )| dt
|ak | 2
k=1
0
2 a
Proof. Note that if |z| = , then (zakkz) =

value 1. Hence, if we set
n
Y
2 ak z
,
F (z) = f (z)
(z ak )
k=1
z zak
zak
has absolute
then F has no zeros in |z| < and |F (z)| = |f (z)| for |z| = . Thus
(6.17) follows on applying Poissons integral formula to log |F (z)|.
We can now turn to Hadamards theorem.
Proof of Theorem 6.24. AssumeP
first that the entire function
f has finite genus h. This means that
1/|ak |h+1 converges, where
a1 , a2 , . . . are the zeros of f . The exponential factor in (6.13) is clearly
of order h, and since the order of a product clearly does not exceed
98
6. ENTIRE FUNCTIONS
the order of the factors, we need only consider

canonical product.
Q the
Rh (z/ak )
Using the notation from page 92 it is P (z) = e
. To estimate
the size of this we shall prove that
| Re Rh (z)| (2h + 1)|z|h+1
(6.18)
for all z. This is true for h = 0, since log |1 z| log(1 + |z|) |z|.
By the definition of Rh it is obvious that we have
| Re Rh (z)| | Re Rh1 (z)| + |z|h
(6.19)
for all z. If | Re Rh1 (z)| (2h 1)|z|h then clearly (6.18) follows if
|z| 1. But if |z| < 1 we have the estimate (1|z|)| Re Rh (z)| |z|h+1
from page 92. Multiplying (6.19) by |z| and adding we get | Re Rh (z)|
|z|| Re Rh1 (z)| + 2|z|h+1 from which again (6.18) follows by the induction assumption. We can now estimate
log |P (z)| =
X
X
( Re Rh (z/ak )) (2h + 1)|z|h+1
1
|ak |h+1
which shows that the order of P (z) is at most h + 1.

Conversely
we have to prove that if the function f has finite order
P
then
1/|ak |h+1 converges, where h is the integer part of . If the
number of zeros of f in |z| < is denoted n(), then applying Jensens
formula for the disk |z| 2 we obtain
1
n() log 2
2
Z2
log |f (2eit )| dt log |f (0)| ,
0
where we have ignored the terms coming from zeros satisfying

|ak | < 2. Given > 0 the integrand is here bounded by + for
sufficiently large , so if we order the zeros according to size |a1 |
|a2 | . . . we have k n(|ak |) |ak |+ for large k. Thus we have
a bound 1/|ak |h+1 1/k (h+1)/(+)
P . If we choose so small that =
(h + 1)/( + ) > 1 the series
1/k converges, so the genus of the
canonical product is at most h.
We finally need to show that the function g in the exponential
factor in (6.13) is a polynomial of degree h. To this end, note that if
f = u + iv is analytic, then f 0 = u0x + ivx0 = u0x iu0y according to the
Cauchy-Riemann equations. Applying x

i y
to (6.17) we therefore
obtain
n()
n()
X
1
f 0 (z) X 1
ak
=
+
+
2
f (z)
z ak k=1 ak z 2
k=1
Z2
0
2eit
log |f (eit )| dt .
it
2
(e z)
6.5. HADAMARDS THEOREM
99
Differentiating this h times gives

n()
X
dh f 0 (z)
h!
=
(6.20)
h
dz f (z)
(ak z)h+1
k=1
n()
X
k=1
h! ak h+1
1
+
(2 ak z)h+1 2
Z2
2(h + 1)!eit
log |f (eit )| dt .
it
h+2
(e z)
We will show that the two last terms tend to 0 as . Note

first that the integral vanishes if f is constant, so that the integral is
unchanged if we divide f by M (). If |z| /2 the absolute value of
the integral is therefore at most a constant multiple of
Z2
h1
log
M ()
dt .
|f (eit )|
R 2
1
By Jensens formula we have 2
log |f | log |f (0)| and since by
0
log(M ())
assumption h+1 0 as , it follows that the integral in
(6.20) vanishes as . Similarly, the penultimate term in (6.20)
may, for |z| /2 be estimated by n()/h+1 which, as we have already
seen, tends to 0 as . It follows that
X
h!
dh f 0 (z)
=
.
h
dz f (z)
(ak z)h+1
k=1
If we write f (z) = eg(z) P (z), where P is the canonical product, then
dh P 0 (z)
(h+1)
clearly the sum to the right is dz
(z) =
h P (z) , so that it follows that g
0. Thus g is a polynomial of degree at most h, and the proof is finally
complete.
As an indication of the power of Hadamards theorem, we have the

following corollary.
Corollary 6.26. An entire function of non-integer order assumes
every finite value infinitely many times.
Proof. Since f (z) and f (z) w obviously have the same order,
as functions of z, for every complex number w, it is enough to show
that the function f has infinitely many zeros if it is of non-integer
order. If f only has finitely many zeros, then the canonical product
is a polynomial and thus of order 0. Thus f is a polynomial times ep
where p also is a polynomial (the genus being finite by Theorem 6.24).
If p has degree n, then clearly f has order n, which is an integer. The
corollary follows.
Note that the most useful way to interpret Theorem 6.24 is as a

factorization theorem for functions of finite order. If the order is not
100
6. ENTIRE FUNCTIONS
an integer, the genus, and thus the form of the factorization, is uniquely
determined, whereas there is an ambiguity if the order is an integer.
Exercise 6.27. Let f be entire and M (r) as before. Suppose
lim log rM (r) = A is finite and not 0. Show that f is of order , but
r
that the existence of the limit does not follow from assuming f to have
order . An entire function for which A is finite and > 0 is said to be
of order and normal type. Extend Corollary 6.26 to show that an
entire function of finite order has infinitely many zeros unless it is of
integer order and normal type.
CHAPTER 7
The Riemann mapping theorem

In this chapter we will prove the Riemann mapping theorem by a
limiting procedure. We will then need to know that the sequence of
mappings constructed, or at least a subsequence of it, has a limit. To
see this, the sequence needs to have a compactness property, analogous
to the Bolzano-Weierstrass theorem for sequences of numbers. The
appropriate concept is given by the following definition.
Definition 7.1. A family (i.e., a set) F of analytic functions defined on a region is called normal if every sequence of functions in
F has a subsequence locally uniformly convergent in .
Exercise 7.2. Prove this equivalence (use the Heine-Borel theorem)!
The main result about normal families is the following characterization.
Theorem 7.3. A family F of functions analytic on a region is
normal if and only if it is locally equibounded.
Here locally equibounded means that for each compact subset E of
there is a constant KE such that |f (z)| KE for every f F and
z E. Equivalently, every point in has a neighborhood E such that
this holds. The proof of Theorem 7.3 is a fairly simple consequence of
a more general compactness theorem by Arzela and Ascoli. Before we
can state this theorem we need to make a definition.
Definition 7.4. A family F of complex valued functions defined
in a complex region is called locally equicontinuous if for every > 0
and compact subset E of there is a > 0 such that |f (z) f (w)| <
for every f F and all z, w E satisfying |z w| < .
Note that as given in the definition above depends only on F, E
and . In other words, it does not depend on the particular function f
we are dealing with.
Theorem 7.5 (Arzela-Ascoli). Suppose f1 , f2 , . . . is a sequence of
complex-valued functions defined on a region C, and assume the
sequence is locally equibounded and equicontinuous in . Then there is
a locally uniformly convergent subsequence.
101
102
7. THE RIEMANN MAPPING THEOREM
Proof. The set of points in with rational real and imaginary

parts is countable and dense in . That the set is countable means
that there is a sequence z1 , z2 , . . . consisting precisely of these points,
and that it is dense means that any neighborhood of any point in contains a point from the sequence z1 , z2 , . . . . Consider now the sequence
f1 (z1 ), f2 (z1 ), f3 (z1 ), . . . of complex numbers. This is a bounded sequence since the set {z1 } is compact, so by the Bolzano-Weierstrass
theorem it has a convergent subsequence, given by evaluating a subsequence f11 , f12 , f13 , . . . of f1 , f2 , . . . at z1 ; call the limit f (z1 ). The
sequence f11 (z2 ), f12 (z2 ), f13 (z2 ), . . . is again bounded, so we can find
a subsequence f21 , f22 , f23 , . . . of f11 , f12 , f13 , . . . which converges when
evaluated at z2 ; call the limit f (z2 ). Since a subsequence of a convergent sequence converges to the same thing as the sequence itself, we
still have limn f2n (z1 ) = f (z1 ). Continuing in this fashion we get
a sequence of sequences fk1 , fk2 , fk3 , . . . , k = 1, 2, . . . such that each
sequence is a subsequence of the ones coming before it, and such that
limn fkn (zj ) = f (zj ) for j k. Now consider the diagonal sequence
f11 , f22 , f33 , . . . . This is a subsequence of the sequence fj1 , fj2 , fj3 , . . .
from its j:th element onwards, so limk fkk (zj ) = f (zj ) for any j. We
shall finish the proof by showing that in fact f11 , f22 , f33 , . . . converges
locally uniformly on .
Let a compact subset E of and a number > 0 be given. By local
equicontinuity we can then find > 0 so that |fnn (z) fnn (w)| < /3
for z, w E and |z w| < . Now consider the open cover of E given
by the balls of radius and centered at zj , j = 1, 2, . . . . This is a
cover since z1 , z2 , . . . is dense in . By the Heine-Borel theorem there
is a finite number of balls, say centered at z1 , z2 , . . . , zk which already
cover E. Given z E we can therefore find zj with j k such that
|z zj | < and therefore get
|fnn (z) fmm (z)|
|fnn (z) fnn (zj )| + |fnn (zj ) fmm (zj )| + |fmm (zj ) fmm (z)|
< /3 + |fnn (zj ) fmm (zj )| + /3.
By Cauchys convergence principle (for complex numbers) and our construction it follows that for every j there is a number Nj such that
|fnn (zj ) fmm (zj )| < /3 if n, m > Nj . If we choose N as the largest
of N1 , . . . , Nk it follows that
|fnn (z) fmm (z)| <
if n and m > N .
Using the other direction of Cauchys convergence principle it follows

that f (z) = limn fnn (z) exists for every z , and letting m
in the expression above we get |fnn (z) f (z)| for every z E if
n > N . This shows that fnn f locally uniformly in .
103
Proof of Theorem 7.3. It is clear by Theorem 7.5 that all we

have to do is show local equicontinuity of F. So let z0 and choose
r > 0 such that the closed disk with radius 2r and centered at z0 is in
. The boundary of the disk is a compact subset of so we can find
a uniform bound M on this set for all f F , by assumption. If z and
w are in the disk B(r, z0 ) with radius r and center z0 we obtain
Z
1
1
1
|f (z) f (w)| =
f ()
d
2i
z w
|z0 |=2r
|z w|
=
2
f ()
d
( z)( w)
|z0 |=2r
M |z w|
2r2
|d| =
2M
|z w|
r
|z0 |=2r
r
makes
since | z| > r, | w| > r. It follows that choosing = 2M
|f (z) f (w)| < if z and w B(r, z0 ) and |z w| < . The local
equicontinuity of the family F follows and the theorem is therefore a
corollary to Theorem 7.5.
Theorem 7.6 (Riemann mapping theorem). Given a simply connected region which is not the entire complex plane C and a point
z0 there is precisely one univalent conformal map f of onto the
unit disk such that f (z0 ) = 0 and f 0 (z0 ) > 0.
Note that Liouvilles theorem shows that it is not possible to map
the entire plane C conformally onto the unit disk; the only bounded
entire functions are the constants.
Proof. We have already proved the uniqueness in Chapter 4.4
after Schwarz lemma (p.7071). To see how to get existence, note that
if g solves the problem and f is a map of into the unit disk mapping
z0 onto 0 and with positive derivative at z0 , then f g 1 satisfies the
conditions of Schwarz lemma so |(f g 1 )0 (0)| 1. Calculating the
derivative we see that this means that f 0 (z0 ) g 0 (z0 ). If we have
equality it follows from Schwarz lemma that f g.
Now let F be the family of univalent functions f analytic in such
that f (z0 ) = 0, |f (z)| 1 for z and f 0 (z0 ) > 0. We just saw that
if our problem has a solution it is the element of F which maximizes
the derivative at z0 . To complete the proof along these lines we need
to: (1) Show that F is not empty, (2) See that F has an element
f maximizing the derivative at z0 and, finally, (3) Show that this f
actually solves the mapping problem.
(1) Since is not all of C there is a (finite) point a
/ .
Since is
simply connected we can define a single-valued branch h of z a in
. Clearly h can not take the value w if it somewhere takes the value
104
w. But by the open mapping theorem there is a disk |w h(z0 )| <

contained in the image h(). It follows that |h(z) + h(z0 )| for all
z ; in particular 2|h(z0 )| . The function
1
h(z) h(z0 )
2
= h(z0 )
h(z) + h(z0 )
h(z0 ) h(z) + h(z0 )
0
h (z0 )
maps z0 to 0 and is bounded by 4|h(z0 )|/. Its derivative at z0 is 2h(z
.
0)
If we now put
|h0 (z0 )| h(z0 ) h(z) h(z0 )
g(z) =
4 |h(z0 )|2 h0 (z0 ) h(z) + h(z0 )
it follows that g is univalent, g(z0 ) = 0, |g(z)| 1, and g 0 (z0 ) > 0 so
that g F. Hence F 6= .
(2) Since all elements of F have their values in the unit disk it
follows that F is an equibounded family, and therefore by Theorem 7.3
a normal family. Now let B = supf F f 0 (z0 ) so that 0 < B .
We can then find a sequence f1 , f2 , . . . in F so that fj0 (z0 ) B as
j . Since F is normal we can find a locally uniformly convergent
subsequence; call the limit function f . It is then clear that f 0 (z0 ) = B
so that actually B < and f is not constant. By Corollary 6.5 f is
univalent. It is clear that f (z0 ) = 0 and f has its values in the closed
unit disk; but by the open mapping theorem the values are then in the
open unit disk.
(3) We need to prove that f () is the unit disk. Suppose to the
contrary that w0 is in the unit disk but w0
/ f (). Since is simply
connected we may define a single-valued branch of
s
f (z) w0
G(z) =
.
1 w0 f (z)
ww0
Since the Mbius transform w 7 1w
preserves the unit disk, the
0w
function G maps univalently into the unit disk. To obtain a member
of F we now set
|G0 (z0 )| G(z) G(z0 )
.
F (z) = 0
G (z0 ) 1 G(z0 )G(z)
It is again clear that F has its values in the unit disk and maps z0 to 0.
2
0 )|
The derivative at z0 is easily calculated to be F 0 (z0 ) = B 1+|G(z
>B
2|G(z0 )|
so that F F. But this contradicts the definition of B.
Note that it is no accident that we get F 0 (z0 ) > f 0 (z0 ); this just
expresses the fact that the inverse of the map f 7 F takes the unit
disk into itself with 0 fixed so that Schwarz lemma shows that the
derivative at 0 is < 1 (clearly the map is no rotation).
CHAPTER 8
The Gamma function

In earlier courses you may have encountered the function
Z
(z) = tz1 et dt ,
0
the Gamma function. The integral converges locally uniformly in z

for Re z > 0, since the absolute value of the integrand is tx1 et if
z = x + iy. If 0 < r x R this shows that on the interval (0, 1] the
integrand may be estimated by tr1 , the integral of which converges on
(0, 1]. Similarly, on the interval [1, ) the integrand may be estimated
by tR1 et = tR1 et/2 et/2 . Here the first factor tends to 0 as t
and is therefore bounded on [0, ), say by M , so the integrand may
be estimated by M et/2 which has convergent integral. It follows that
is analytic in Re z > 0, since the integrand is analytic.
Integration by parts shows that the functional equation
(8.1)
(z + 1) = z (z)
is valid for Re z > 0 (check this). Since clearly (1) = 1 it follows by induction that (n + 1) = n! for natural numbers n, so one may view the
gamma-function as an extension of the factorial to non-natural numbers. Another very important consequence of (8.1) is that it allows one
to extend analytically to the left of Re z = 0. If is already defined
in z + 1 we may define (z) = z1 (z + 1). Clearly this works as long
as z 6= 0. By induction we may therefore define everywhere except
at the non-positive integers. In these points the extended gammafunction has simple poles. In this way the gamma-function is extended
to a meromorphic function in the whole complex plane, with poles at
0, 1, 2, . . . and nowhere else.
Exercise 8.1. Calculate the residues of at the non-positive integers!
To obtain a product expansion of , let us first construct
an entire
P
function
with
simple
zeros
where
has
poles.
Since
1/n
diverges
P
2
but
1/n does not, we set
F (z) = z
z
(1 + )ez/k .
k
k=1
105
106
8. THE GAMMA FUNCTION
Clearly F (z)F (z) = z 2

shows that
k6=0 (1
kz )ez/k so comparison with (6.14)
z sin z
.
It is also clear that F (z + 1) has the same zeros, apart from z = 0, as

F (z), so we have
(8.2)
(8.3)
F (z)F (z) =
F (z)/z = e(z) F (z + 1)
with an entire function . To determine we take the logarithmic

derivative of both sides to obtain
X
X
1
1
1
1
1
0
(
) = (z) +
+
(
).
z+k k
z + 1 k=1 z + 1 + k k
k=1
If we replace k by k + 1 in the first sum we obtain, after simplification
X
1
1
0 (z) =
(
)1 .
k
k
+
1
k=1
Since the series telescopes to the sum 1 we have 0 (z) = 0 so that
is constant. To determine the value of , we note that F (z)/z 1 as
z 0 so we obtain from (8.3) that 1 = e F (1). But the n:th partial
product of F (1) is
n
X
1
23
n + 1 1 1 1
2
n
...
e
= (n + 1) exp(
),
12
n
k
k=1
P
so that = limn ( nk=1 k1 log n). The constant is called Eulers
constant and equals approximately 0.5772. As far as I know it is not
known whether is rational (though it seems unlikely). If we set
G(z) = ez /F (z) we have the expansion
ez Y
z
(8.4)
G(z) =
(1 + )1 ez/k ,
z k=1
k
and from (8.3) follows
G(z + 1) = zG(z) ,
the same functional equation that satisfies. One might now guess
that G = . We will show this, which is surprisingly difficult. Note
that we obtain from (8.2) and the functional equation the so called
reflection formula
.
G(z)G(1 z) =
sin z
Since F has no poles the function G has no zeros, and since it has
the same poles as , the function (z)/G(z) is entire. If we can show
that it is bounded, then by Liouvilles theorem it is constant and since
(1) = G(1) we would be done. Note that by the functional equations
that G and satisfy we have (z + 1)/G(z + 1) = (z)/G(z) so that
107
/G is periodic with period 1. We therefore only need to bound /G

in a period strip, say 1 Re z 2. But it is immediately clear that,
in this strip, | (x + iy)| (x) so is bounded in the strip by the
maximum of in the real interval [1, 2]. We now need a lower bound
for G(x+iy) for 1 x 2 and |y| large. Such a bound can be obtained
from Stirlings formula
(8.5)
G(z) = 2z z1/2 ez eJ(z) ,
where J(z) 0 as z in a half-plane Re z c > 0. We will
prove this formula later; for the moment let us show that it implies the
desired lower bound for G and hence the identity of G and . If (8.5)
is true we obtain, for z = x + iy,
log |G(z)| = 12 log 2 x + (x 12 ) log |z| y arg z + Re J(z) .
All terms are here bounded from below except y arg z, which is at
least bounded from below by |y|/2. It follows that /G is bounded
in the period strip by a constant multiple of e|y|/2 . For a function of
period 1 this is enough to show boundedness, since such a function may
1
be viewed as a function of = e2iz , the possible values of z = 2i
log
differing by integers. As a function of the function /G has isolated
singularities at 0 and , but our bound on /G is e| log |||/4 , i.e., for
small || a multiple of ||1/4 and for large || a multiple ||1/4 . Thus
both singularities are removable (see the following exercise), /G is
bounded, and we are done.
Exercise 8.2. Recall that if f is analytic with an isolated singularity at z = w, then the singularity is removable if (and only if)
(z w)f (z) 0 as z w. State a similar condition for singularities
at infinity.
Hint: Look at the discussion just before Theorem 4.2.
Let us now turn to Stirlings formula, so assume Re zP
> 0. Accord1
1
ing to (8.4) the logarithmic derivative of G is z1
1 ( z+k k )
and differentiating once more we get
d G0 (z) X
1
=
.
dz G(z)
(z + k)2
k=0
For fixed z in the right half-plane the terms of the sum are the residues
cot
. Note that
in the right half-plane of the function H() =
(z + )2
= z is not in the right half-plane, and sincos is analytic and equals
1 at 0. Thus the residue at 0 is z12 . By periodicity the residue at k is
1
; thus the residues of H are as stated.
(z+k)2
Now let be the contour consisting of a rectangle with corners iY
and n+ 21 iY , except for avoiding = 0 by a small semicircle of radius
R
1
H.
r centered at 0, such that 0 is inside the contour. Consider 2i
108
P
1
This is independent of r for small r and equals nk=0 (z+k)
2 . On the
horizontal sides the factor cot tends uniformly to i as Y , and
the other factor (z + )2 tends uniformly to 0, so the corresponding
integrals also tend to 0. Our contour now consists of two infinite vertical
lines, apart from the little semicircle. On the line Re = n+ 21 the factor
cot is bounded, independentlyR of the integer n, so the corresponding
integral is less than a multiple of Re =n+ 1 |z+|2 d Im which tends to
2
0 as n . The integrals over the straight line parts of the remaining
part of the contour may be written
1
Zr
1
cot(i)
d
2
(i + z)
2
cot(i)
d
(i + z)2
1
=
2
cot(i)
1
1
d ,
(i z)2 (i + z)2
and the integral over the semi-circle tends to 2z12 as r 0 so we finally

obtain
Z
d G0 (z)
1
1
4iz
(8.6)
= 2+
cot(i) 2
d .
dz G(z)
2z
2
( + z 2 )2
0
Exercise 8.3. Verify all calculations and claims above!

2
, and
Using Eulers formulas we may write i cot(i) = 1+ exp(2)1
the part of the integral coming from the term 1 has the value 1/z. In
this way we obtain
d G0 (z)
1
1
= + 2+
dz G(z)
z 2z
Z
( 2
4z
d
.
2
2
2
+z ) e 1
We need to integrate this twice to obtain Stirlings formula. A first

integration gives, for Re z > 0,
G0 (z)
1
= C + log z
G(z)
2z
Z
2
d
2
.
2
2
+z e 1
Give a justification for changing the order of integration in the integral!

To integrate once more we first make an integration by parts in the
1
log(1e2 )
integral. Noting that a primitive of the second factor is 2
we obtain
Z 2
Z
d
1
z2
2
=
log(1 e2 ) d ,
2
2
2
2
2
2
+z e 1
( + z )
0
109
so that another integration gives

1
log G(z) = D + Cz + (z 12 ) log z
Z
2
z
log(1 e2 ) d .
+ z2
The last term (including the minus-sign) we define to be

1
J(z) =
Z
2
1
z
log
d ,
2
+z
1 e2
so it only remains to show that J(z) has the claimed behavior and to
determine the constants of integration C, D. But we have | 2 + z 2 | =
|z i||z + i| c|z| if Re z R c > 0 so the integral over [N, ) may be
1
estimated by the integral c
log 1e12 d which is convergent and
N
therefore < for sufficiently large N . But if |z| > N we can estimate
R
log 1e12 d, which tends to 0
the integral over (0, N ] by (|z||z|
2 N 2 ) 0
as z . Thus J(z) 0 if z in Re z c > 0. The functional
equation for G may be expressed log G(z + 1) = log z + log G(z), at
least if z > 0. Substituting (8.6) in this gives, after simplification,
1
1
C = (z + ) log(1 + ) + J(z) J(z + 1) .
2
z
Letting z + it follows that C = 1. To determine D we substitute
(8.6) in the reflection formula G(z)G(1 z) = / sin z for z = 12 + iy
to obtain, after simplification,
= (eD )2
cosh y
exp(1 + iy(log( 12 + iy) log( 12 iy)) + J( 12 + iy) + J( 12 iy)),
where the logarithms have their principal value. Further simplification
gives
(eD )2 = 2 exp(1 + 2y arctan(2y) J( 12 + iy) J( 12 iy))/(ey + ey )
1
= 2 exp(y 2y arctan 2y
+ 1 J( 12 + iy) J( 12 iy))/(ey + ey )
2 as y + .
Since G(x) > 0 for x > 0 it follows that eD = 2 so we have finally

proved Stirlings formula for G. Since this implies the identity of G
and we have also proved the reflection formula
.
(z) (1 z) =
sin z
and Stirlings formula
(z) = 2z z1/2 ez eJ(z) .
110
Exercise 8.4. Verify the calculations above. Then show that the
integrand in J(z) may be developed as a finite sum of odd powers of
1/z plus a remainder and that the result may be integrated to yield an
expansion
n
X
Ak
J(z) =
+ Jn (z) ,
2k1
z
k=1
where the remainder Jn (z) may be estimated by a constant multiple of
1/z 2n+1 for large z satisfying Re z c > 0. Also show that for fixed
z the remainder Jn (z) has no limit as n . An expansion of this
kind is called an asymptotic expansion (as z in Re z c > 0).
One may express the constants Ak explicitly in terms of the so called
Bernoulli numbers.

Complex Analysis Christer Bennewitz

Uploaded by

Copyright:

Available Formats

Complex Analysis Christer Bennewitz

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Complex Analysis Christer Bennewitz

Uploaded by

Copyright:

Available Formats

COMPLEX ANALYSIS

Mnne ifrn hgre zoner

c 2006 by Christer Bennewitz

Chapter 1. Complex functions

Chapter 2. Analytic functions

Chapter 5. Harmonic functions

Chapter 6. Entire functions

Chapter 7. The Riemann mapping theorem

Chapter 8. The Gamma function

is, a map : G G G, so that for every pair of elements x, y of G,

A field (F, +, ) is a set F provided with two binary operations +

of the factors is 0 unless their e-component y = 0. Hence a + b4 < 0.

1.1. THE COMPLEX NUMBER SYSTEM

by i and call it the imaginary unit. If we use 1, i as a basis we may

Exercise 1.8. Consider the set of symbols x + iy + ju + kv, where

The geometric interpretation of addition is already familiar, since

1.3. SQUARE ROOTS

but extremely important that arg z is NOT uniquely determined by

Exercise 1.11. Write the number z = 3 + i on polar form and

we get, after extracting a (real) square root, that x2 + y 2 = u2 + v 2

the first equation is actually satisfied, whereas the second is satisfied

to denote by the symbol w? Since the complex numbers are not

1.4. STEREOGRAPHIC PROJECTION

1.4. Stereographic projection

Figure 1. Stereographic projection

Since |z|2 + 1 = 2/(1 x3 ) by the third equation the inverse is easily

Also show that the distance between the image of z and N is

In view of Exercise 1.12 we may now add to C an ideal point ,

1.4. STEREOGRAPHIC PROJECTION

In view of this theorem we will by a circle in the extended plane

1.5. MBIUS TRANSFORMS

1.5. MBIUS TRANSFORMS

Two points z and z are said to be symmetric with respect to R if

In short, Theorem 1.28 says that symmetry is preserved by Mbius

1.5. MBIUS TRANSFORMS

listed above. It therefore seems natural to try to find, in each conjugacy

What we have proved is that each conjugacy class different from

On the other hand, if is neither positive nor of absolute value 1

1.6. POLYNOMIALS, RATIONAL FUNCTIONS AND POWER SERIES

proof is simply the observation that since p(z) = k(z)(z a) + r where

It follows that r is continuous as a function in C in all points which

The next theorem is the complex version of what is usually known

1.6. POLYNOMIALS, RATIONAL FUNCTIONS AND POWER SERIES

Theorem 1.46. Let A be a subset of C and f1 , f2 , . . . a sequence of

The last member does not depend on z and tends to 0 as N . The

Proof of Theorem 1.44. If |z a| > R then an (z a)n , n =

Here is the complex version of another well known theorem.

So far we have said nothing about convergence on the boundary of

by Kolmogorov in 1926) for which an Rn 0 such that (1.4) diverges

using the conventions 1/0 = and 1/ = 0.

sufficiently large n. Hence |an rn | < 1 for such n, so the sequence an rn ,

Theorem 1.51. If limn |an /an+1 | exists it is equal to R.

1.6. POLYNOMIALS, RATIONAL FUNCTIONS AND POWER SERIES

Exercise 1.52. Find the radius of convergence for the following

so the action of f can also be

This shows that there is really no need to study anti-conformal maps

Exercise 2.4. Show that the map z 7 z 2 is conformal in any open

2.1. CONFORMAL MAPPINGS AND ANALYTICITY

Theorem 2.6. Suppose that f is differentiable at a. Then

We will later prove (see Corollary 4.23) that if f is differentiable