Lectures on complex analysis

Alexander Kupers

April 30, 2020

Abstract
These are the collected lecture notes for Math 113.

i
 Contents

1 Complex numbers and holomorphic functions 1

1.1 Complex numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 The algebra of complex numbers . . . . . . . . . . . . . . . . . . . 2
1.1.2 The geometry of complex numbers . . . . . . . . . . . . . . . . . . 4
1.2 Holomorphic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Examples of holomorphic functions . . . . . . . . . . . . . . . . . . 7
1.2.2 Integration of holomorphic functions . . . . . . . . . . . . . . . . . 8
1.3 The Riemann zeta-function . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Functions on the complex plane 11

2.1 The complex plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 Recollection: the algebra of complex numbers . . . . . . . . . . . . 11
2.1.2 The topology of complex numbers . . . . . . . . . . . . . . . . . . 11
2.2 Holomorphic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.1 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.2 Complex differentiation . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.3 The Cauchy–Riemann equations . . . . . . . . . . . . . . . . . . . 16

3 Integration and Goursat’s theorem 18

3.1 Integration along curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1.1 Examples of integrals . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2 Goursat’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2.1 Goursat’s theorem for triangles . . . . . . . . . . . . . . . . . . . . 23
3.2.2 A down-to-earth argument in support of Goursat’s theorem . . . . 25

4 Cauchy’s integral formula 27

4.1 Holomorphic functions have primitives on disks . . . . . . . . . . . . . . . 27
4.2 Computing some integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.3 Cauchy integral formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

5 Applications of Cauchy’s integral formula 35

5.1 Cauchy integral formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

ii
 Contents iii

5.1.1 Liouville’s theorem and the fundamental theorem of algebra . . . . 35

5.2 Holomorphic functions are analytic . . . . . . . . . . . . . . . . . . . . . . 36
5.2.1 Power series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.2.2 Analytic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.2.3 Analytic continuation . . . . . . . . . . . . . . . . . . . . . . . . . 40

6 Morera’s theorem and its applications 42

6.1 Morera’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.2 Constructing holomorphic functions by limits . . . . . . . . . . . . . . . . 42
6.2.1 Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.2.2 Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.2.3 Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.2.4 Pointwise convergence . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.3 Schwarz reflection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

7 Poles 49
7.1 Poles and zeroes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
7.1.1 Isolated singularities . . . . . . . . . . . . . . . . . . . . . . . . . . 49
7.1.2 Zeroes and poles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
7.1.3 Laurent series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7.2 The residue formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
7.3 Integrals using the residue formula . . . . . . . . . . . . . . . . . . . . . . 53
7.3.1 Zeroes and poles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
7.4 The residue formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

8 Singularities 58
8.1 Removable singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
8.2 Essential singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
8.2.1 Application: the automorphisms of C . . . . . . . . . . . . . . . . 60
8.3 Meromorphic functions and the complex projective plane . . . . . . . . . 60
8.3.1 Application: the automorphisms of CP 1 . . . . . . . . . . . . . . . 62

9 The argument principle 64

9.1 The argument principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
9.2 Rouché’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
9.3 The open mapping theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 66
9.4 The maximum modulus principle . . . . . . . . . . . . . . . . . . . . . . . 66

10 Homotopy invariance and the logarithm 68

10.1 Integrals along paths are homotopy-invariant . . . . . . . . . . . . . . . . 68
10.2 The logarithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
10.2.1 Logarithms of holomorphic functions . . . . . . . . . . . . . . . . . 72

11 Pre-midterm recap 73
11.1 Holomorphic functions and their properties . . . . . . . . . . . . . . . . . 73
11.2 Constructing holomorphic functions . . . . . . . . . . . . . . . . . . . . . 74
iv Contents

11.3 The classification of singularities . . . . . . . . . . . . . . . . . . . . . . . 74

11.4 Evaluation of integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
11.5 Integral formula’s and growth behavior of holomorphic functions . . . . . 76

12 Growth of zeroes 77
12.1 The mean value property . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
12.2 Jensen’s formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
12.3 Functions of finite order . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
12.4 An application of Jensen’s formula to polynomials . . . . . . . . . . . . . 82

13 Weierstrass’ construction 84
13.1 Infinite products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
13.1.1 The product formula for the sine . . . . . . . . . . . . . . . . . . . 85
13.2 Weierstrass’ construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

14 Hadamard factorization theorem 89

14.1 Hadamard factorization theorem . . . . . . . . . . . . . . . . . . . . . . . 89
14.1.1 Preparatory lemmas . . . . . . . . . . . . . . . . . . . . . . . . . . 90
14.1.2 The proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
14.2 Picard’s little theorem for functions of finite order . . . . . . . . . . . . . 93

15 The Gamma function 94

15.1 The Gamma function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
15.1.1 Analytic extension . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
15.1.2 Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
15.1.3 Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
15.1.4 Product formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
15.2 Mellin transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

16 The Riemann zeta function 100

16.1 The Riemann zeta function . . . . . . . . . . . . . . . . . . . . . . . . . . 100
16.1.1 The theta function . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
16.1.2 Expressing ζ in terms of θ . . . . . . . . . . . . . . . . . . . . . . . 102
16.1.3 Estimates on the line Re(s) = 1 . . . . . . . . . . . . . . . . . . . . 104

17 The zeroes of the Riemann zeta function 106

17.1 Recollection on the prime number theorem . . . . . . . . . . . . . . . . . 106
17.1.1 The Riemann zeta function as a product . . . . . . . . . . . . . . . 106
17.2 The trivial zeroes of Riemann zeta function . . . . . . . . . . . . . . . . . 108
17.3 The Riemann zeta function has no zeroes on Re(s) = 1 . . . . . . . . . . . 108
17.3.1 Some additional estimates . . . . . . . . . . . . . . . . . . . . . . . 110

18 Tchebychev’s functions 111

18.1 Tchebychev’s functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
18.2 Relating Tchebychev’s functions to ζ . . . . . . . . . . . . . . . . . . . . . 113
18.3 Proving the asymptotics of ψ1 . . . . . . . . . . . . . . . . . . . . . . . . . 115
 Contents v

19 The proof of the prime number theorem 116

19.1 The asymptotics of ψ1 (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
19.2 Zeroes of the Riemann zeta function and error terms for the prime number
theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

20 Conformal equivalences 121

20.1 Conformal equivalences . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
20.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
20.2.1 The half-plane is conformally equivalent to the unit disk . . . . . . 123
20.2.2 Further examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
20.3 The Riemann mapping theorem . . . . . . . . . . . . . . . . . . . . . . . . 125

21 The Schwartz lemma and conformal automorphisms 126

21.1 Schwartz lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
21.2 Conformal automorphisms of the upper half-plane and unit disk . . . . . 127
21.2.1 Conformal automorphisms . . . . . . . . . . . . . . . . . . . . . . . 127
21.2.2 The conformal automorphisms of D . . . . . . . . . . . . . . . . . 127
21.2.3 The conformal automorphisms of H . . . . . . . . . . . . . . . . . 129

22 Several complex variables 133

22.1 Holomorphic functions of several complex variables . . . . . . . . . . . . . 133
22.2 The Hartogs extension theorem . . . . . . . . . . . . . . . . . . . . . . . . 134
22.3 Domains of holomorphy and convergence of power series . . . . . . . . . . 135

23 Final recap 138

23.1 Midterm material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
23.2 Zeroes of entire functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
23.2.1 Jensen’s formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
23.2.2 Functions of finite order . . . . . . . . . . . . . . . . . . . . . . . . 140
23.3 The Gamma function, Riemann zeta function, and the prime number
theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
23.4 Conformal equivalences . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

Bibliography 142
 Chapter 1

Complex numbers and holomorphic functions

In this first chapter I will give you a taste of complex analysis, and recall some basic
facts about the complex numbers. We define holomorphic functions, the subject of this
course. These functions turn out to be much more well-behaved than the functions you
have encountered in real analysis. We will mention the most striking such properties that
we will prove during the courses. We also give many examples of holomorphic functions,
ending with the most exciting example: the Riemann zeta-function.
We will go over this material in more depth in the coming lectures. However, you can
take a look at [SS03, Section 1.1], [MH87, Section 1.1 and 1.2] for more details about the
complex numbers.

1.1 Complex numbers

Complex numbers have their origin in attempts to understand the roots of polynomials.
By the sixteenth century, mathematicians had found formulas for the solutions of the
equation
p(x) = 0,
when p is a quadratic or cubic polynomial. For example, when p(x) = ax2 + bx + c such
solutions x0 are given by the well-known quadratic formula
√
−b ± b2 − 4ac
x0 = .
2a
When b2 − 4ac ≥ 0 the graph of p intersects the x-axis, and the above formula can be
interpreted as giving the x-coordinate of these intersections. When b2 − 4ac √ < 0 there
are no such intersections, which is reflected by the fact that at first sight b2 − 4ac is
meaningless when the expression b2 − 4ac, called the discriminant is negative. This
suggests one way to interpret the quadratic formula: if the discriminant is positive it
gives two roots, if the discriminant is zero it gives a single root, and if it is negative it
gives no roots.
However, it was observed that if one introduced a new number i satisfying i2 = −1
to the number system, and let square roots of negative numbers be multiples of i, they
not only appeared to give roots for the polynomial p when the discriminant is negative,

1
2 Chapter 1 Complex numbers and holomorphic functions

but could be fruitfully manipulated to solve various other algebraic problems. That
is, the transition from real to complex numbers gives the quadratic formula a useful
interpretation even when the discriminant is negative. Throughout the seventeenth and
eighteenth centuries these novel complex numbers took on increasingly important roles
in mathematics, until eventually in the nineteenth they were considered as being equal
in status to the real numbers.

1.1.1 The algebra of complex numbers

Let us introduce some notation which we will use throughout the course. When a is
a real number, we use the notation a ∈ R. A number of the form bi with b ∈ R is called
an imaginary number, and a number of the form a + bi is called an complex number. In
particular, through the course, real numbers and imaginary numbers are subsets of the
complex numbers. Given a complex number a + bi, a is its real part and b its imaginary
part. Observe we can record a + bi as a pair (a, b) of real numbers. In fact, we shall take
this as the starting point of our formal definition.

Definition 1.1.1. The set C of complex numbers is the set of all pairs (a, b) ∈ R2 . We
will write a pair (a, b) as a + bi.

z = a + bi
bi •

·
a real axis
imaginary axis

Figure 1.1 A point in the complex plane, its real part giving its x-coordinate and its imaginary
part its y-coordinate.

What makes numbers useful is that one can perform algebraic operations with them.
Knowing how to define such operations for the real numbers, we can define addition and
multiplication of complex numbers:

(a + bi) + (c + di) := (a + c) + (b + d)i

(a + bi) · (c + di) := (ac − bd) + (ad + bc)i.

 1.1 Complex numbers 3

The latter rule is easy to reconstruct if you recall that multiplication of complex numbers
should be an extension of that of real numbers, and that we required i to have the
property that i2 = −1. We will often drop the · from the notation of multiplication.
These addition and multiplication rules have the same formal properties as the real
numbers: the only non-obvious one is that non-zero complex numbers have multiplicative
inverses. To see this is the case, let’s try to solve

(a + bi) · (c + di) = (ac − bd) + (ad + bc)i = 1.

If we attempt to solve the linear equations

ac − bd = 1 and ad + bc = 0

for c, d in terms of a, b, we get

a − bi
(a + bi)−1 = c + di = . (1.1)
a2 + b2
In other words, the complex numbers are a field. This means that they satisfy the
following axioms (this spells out the “obvious” formal properties I didn’t spell out above):
Proposition 1.1.2. (C, +, ·, 0, 1) is a field, that is, has the following properties:
(i) a + b = b + a,
(ii) (a + b) + c = a + (b + c),
(iii) a + 0 = a = 0 + a,
(iv) for all a ∈ C, there exists b ∈ C such that a + b = 0,
(v) ab = ba,
(vi) (ab)c = a(bc),
(vii) 1a = a = a1,
(viii) for all a ∈ C \ {0}, there exists b ∈ C such that ab = 1,
(ix) (a + b)c = ab + bc and a(b + c) = ab + ac.
As a exercise, you should identify which of these properties are referred to as “com-
mutativity,” “associativity,” and “distributivity.”
In fact, the complex numbers are best type of field; they are algebraically closed. This
means that every non-constant polynomial p with coefficients in C has a root in C, i.e. for
every polynomial p which is not constant there exists a z0 ∈ C such that p(z0 ) = 0. This
is a fact that we will be able to prove quite early in this course using the techniques of
complex analysis:
Theorem 1.1.3 (Fundamental theorem of algebra). If p(z) = an z n + · · · + a1 z + a0 is a
polynomial with complex coefficients which is not constant (i.e. n > 0), then there is a
complex number z0 such that p(z0 ) = 0.
This should be surprising. It is not clear at all that adding the solutions ±i of the
single equation x2 + 1 = 0 means you can now also solve other quadratic polynomials, or
polynomials of higher degree, let alone polynomials with complex coefficients.
4 Chapter 1 Complex numbers and holomorphic functions

1.1.2 The geometry of complex numbers

The real numbers have two more important properties: a notion of absolute value,
and an order. Complex numbers inherit the former, a notion of absolute value, but not
the latter (as you’ll see on the homework).

The norm
To define the distance between complex numbers, we use a geometric interpretation
of complex numbers. Indeed, using our definition of complex numbers as pairs of real
numbers we can draw the complex numbers C as a plane: the complex plane. The real
numbers lie on the x-axis and the imaginary numbers lie on the y-axis. Under this
identification the complex a + bi corresponds to the vector (a, b) ∈ R2 . and addition of
complex numbers is vector-addition.
The absolute value of a real number is its distance from the origin. This makes it clear
how we should define the absolute value of a complex number, which we will occasionally
also refer to as the norm. It should defined as the distance to the origin:
p
|a + bi| := a2 + b2 .

Observe that |z| = 0 if and only if z = 0, as behooves a norm. It interacts with addition
and multiplication as follows:

Proposition 1.1.4. | − | : C → R≥0 is a norm, that is, has the following properties:
(i) |z| = 0 if and only if z = 0,
(ii) |zw| = |z||w|,
(iii) |z + z 0 | ≤ |z| + |z 0 |.

As usual, one can define distances in terms of absolute values. Just like the distance
between two real numbers x, x0 is given by |x − x0 |, we should think of |z − z 0 | as the
distance between the two complex numbers z, z 0 . It is just the Euclidean distance in the
complex plane.

The complex conjugation

It is helpful to express the norm in terms of another operation on complex numbers,
complex conjugation. This is given by sending i to −i, or more precisely by

a + bi := a − bi.

That such an operation exists and is useful, is not surprising: when formally adding the
solution i of x2 + 1 = 0 to the real numbers, you can’t distinguish between i and −i, and
from this should result a symmetry of the complex numbers.1
1
In fact, these are all the symmetries of the complex numbers as a field extension of the real numbers.
That is, the Galois group Gal(C/R) is Z/2 with non-identity element given by complex conjugation.
 1.1 Complex numbers 5

We have that |z|2 = z z̄. From this we get the following expression for the multiplicative
inverse:
z̄
z −1 = 2 ,
|z|
which you should compare to (1.1).

The polar representation of complex numbers

When we were thinking of complex numbers as elements in the plane, we were using
cartesian coordinates. We can also use polar coordinates: recording the distance to the
origin and the angle to the positive x-axis. A non-zero complex number a + bi has polar
coordinates
(distance, angle) = (r, θ) =: (|z|, arg(z)).
That is, the first is the norm |z| of a complex number z, the second is the so-called
argument arg(z). In other words, I am defining these by demanding that

z = |z|(cos(arg(z)) + i sin(arg(z))).

The argument is a bit more subtle than the norm: if we can take arg(z) = θ in the
above formula, we can also take it to be θ + k2π for any k ∈ Z. The way we resolve
this ambiguity is by declaring the argument to be an element of the quotient R/2πZ,
i.e. identifying two real numbers when they differs by a multiple of 2π. This will continue
to be a source of headaches, especially when trying to define logarithms.

z = a + bi
•

arg(z)

Figure 1.2 The circle in the complex plane around the origin and through z has radius |z|.

Addition of complex numbers was easy to represent in cartesian coordinates, but mul-
tiplication not so much so. For polar coordinates the situation is reversed: multiplication
is easy to represent, but addition not so much. This is encoded by the following lemma,
the second part of which is proven by the addition formulas for sines and cosines:

Lemma 1.1.5. We have that |zw| = |z||w|, arg(zw) ≡ arg(z) + arg(w) (mod 2πZ).
6 Chapter 1 Complex numbers and holomorphic functions

This gives us some insight into the geometry of multiplication by a complex number
z: it scales the complex plane by |z| and rotates it by angle arg(z).

1.2 Holomorphic functions

The main objects of interest in this course are complex-valued functions defined on a
some subset of the complex plane:
f : C ⊃ Ω −→ C.
We do not care about arbitrary functions: we care about those that are complex-
differentiable. These are called holomorphic. This definition concerns the behavior of
f (zi ) as zi → z0 ∈ Ω and for this to be reasonable we should be able to approach z0 from
all directions. That is, Ω should be open, i.e. for each point in Ω we can find a little disk
around it which is also contained in Ω.
Definition 1.2.1. Let Ω be open, then f : C ⊃ Ω −→ C is complex-differentiable at
z0 ∈ Ω if there exists an a ∈ C such that
|f (z0 + h) − f (z0 ) − ah|
lim = 0.
h→0 |h|
We then write a = f 0 (z0 ), and say a is the derivative of f at z0 .
Definition 1.2.2. We say that f is holomorphic if it is complex-differentiable at all
z0 ∈ Ω.
This implies f is continuous, but at first sight does not imply that f 0 is continuous.
However, holomorphic functions have many miraculous properties, the first of which is
the following:
(1) If f is holomorphic then not only f 0 continuous, but it is again holomorphic.
By induction, if f is once complex-differentiable then f is infinitely many times
complex-differentiable!
Let me try to dispel some of the mystery of this astounding rigidity property of
holomorphic functions. Using cartesian coordinates, the subset Ω ⊂ C can be thought
of as a subset of R2 and the holomorphic function f : C ⊃ Ω → C gives a function
R2 ⊃ Ω → R2 . Explicitly, this is the function
R2 ⊃ Ω −→ R2
(x, y) 7−→ (Re(f (x + yi)), Im(f (x + yi))).
We will see that if f is holomorphic, this is a differentiable function of two variables. That
is, at each z0 = x0 + y0 i it has a total derivative at z0 = x0 + y0 i, a (2 × 2)-matrix with
real entries, which gives the best linear approximation to the function. For an arbitrary
differentiable function of two variables, this matrix can be anything. For a holomorphic
function, this matrix comes from multiplication by the complex number f 0 (z0 ) and is
given by a composition of rotation and scaling. Shearing or scaling by differing amounts
in different directions are not allowed. That the total derivative is so restricted at each
point is what gives holomorphic functions their rigidity.
 1.2 Holomorphic functions 7

1.2.1 Examples of holomorphic functions

From the definition of a holomorphic function, one can directly check that functions

1 : C −→ C
z 7−→ 1,

z : C −→ C
z 7−→ z,

are holomorphic. Indeed, we will see that the rules of complex differentiation of holomor-
phic functions are the same as those for ordinary differentiation of differentiable functions.
The derivatives of the above functions are just 10 = 0 and z 0 = 1.
Similarly, a sum of holomorphic functions is holomorphic with derivative the sum of
the derivatives, and a product of holomorphic functions is holomorphic with derivative
given by the product rule. Using finite linear combinations of products of these basic
functions, we can form any polynomial:
Example 1.2.3. Every polynomial with complex coefficients is holomorphic.
We can form more complicated functions using infinite sums. Within its radius of
convergence, the power series ∞ n
P
n=0 an z define a holomorphic function.

Example 1.2.4. The exponential function can be extended from real numbers to the
cmoplex numbers by the familiar everywhere-convergent power series:
∞
X 1 n
ez = z .
n=0
n!

This will our definition. From this we can define the cosine and sine:

ez + e−z eiz − e−iz

cos(z) = , sin(z) = ,
2 2i
as well as other trigonometric functions. Using these definitions, we can write the polar
representation of a non-zero complex number z more conveniently as:

z = |z|ei arg(z) .

In fact, another special property of holomorphic functions which we shall prove in

this course, is that locally holomorphic functions can be defined using power series:
(2) If f is holomorphic, then for each z0 ∈ Ω there is a disk of radius  > 0 around z0
such that f can be described on this disk by a convergent power series
∞
X
an (z − z0 )n .
n=0
8 Chapter 1 Complex numbers and holomorphic functions

That is, all holomorphic functions are analytic. This is in contract with differentiable
functions, e.g. the function

g : R −→ R
( 2
e−1/x if x > 0,
x 7−→
0 otherwise,

is not given by a power series near x = 0.

1.2.2 Integration of holomorphic functions

Since holomorphic functions are continuous functions, we can try to integrate them.
The complex plane gives us more freedom to define such integrals than the real line: it is
unambiguous what integrating g : R → R from a real number a to a larger real number
b means, but for a holomorphic function f : C → C we need to give a path γ from a
complex number a ∈ C to a complex number b ∈ C to make sense of the integral. That
is, we can write
Z b
g(x)dx
a

but should write Z

f (z)dz.
γ

One can then ask how this depends on the path γ from a to b. Here we meet the
most important property of holomorphic functions for physics and engineering:
(3) If γ0 can be deformed into γ1 through a family of paths γt with the same endpoints
and depending continuously on t, then
Z Z
f (z)dz = f (z)dz.
γ0 γ1

This gives a very powerful method to compute integrals, since you can pick γ to make
your life as easy as possible.
Example 1.2.5. After the first couple of weeks you will be able to use this to compute
that Z 1
log(sin(πz))dz = − log(2).
0

Example 1.2.6 (Contour integration). If γ(0) = γ(1) and γ can be deformed in Ω to a

constant map while keeping its endpoints fixed, then
Z
f (z)dz = 0.
γ

This is surprising, as γ could be some large circle.

 1.3 The Riemann zeta-function 9

1.3 The Riemann zeta-function

A much more interesting example of a holomorphic function is the Riemann zeta-

function. For Re(s) > 1 this is defined as2
∞
X 1
ζ(s) = .
n=1
ns

To see that difficulties occur with this definition when Re(s) ≤ 1, observe that for s = 1
we get the divergent harmonic series ∞ 1
P
n=1 n .
Remark 1.3.1. It is in fact an instance of a general “point-counting function” construction
in algebraic geometry, applied to the scheme Spec(Z), the algebro-geometric counterpart
of the integers.
The Riemann zeta-function actually extends to a holomorphic function Ω := C\{1} →
C. At z = 1, it blows up in the sense that |ζ(s)| → ∞ as s → 1. That such an extension
exists is not a special property of Riemann zeta function, but a general property of
holomorphic functions. We will give precise formulations during the course, but for now
let us phrase it as:
(4) Holomorphic functions defined on small subsets Ω0 ⊂ C admit unique extensions to
much larger subsets Ω ⊂ C.
The behavior of the zeroes of Riemann ζ-function in Ω is closely related to the
distribution of the prime numbers. This is made more plausible from the following
1 P∞ sk
equivalent definition ζ(s) for Re(s) > 1: using that 1−p s = k=0 p , unique prime
factorization gives (we are skipping some justifications about the manipulations of infinite
products and sums)
∞ ∞
!
1 1
p−sk
X Y X Y
ζ(s) = = = .
n=1
ns p prime k=0 p prime
1 − p−s

For example, let us deduce the following from what we’ve asserted about the ζ-function
so far:

Proposition 1.3.2. There are infinitely many primes.

Proof. We use that |ζ(s)| → ∞ as s → 1, and give a proof by contradiction. Indeed, if

there were only finitely many primes p1 , . . . , pr , then our product expansion for Re(s) → 1
is finite
1 1
ζ(s) = −s · · ·
1 − p1 1 − p−s
r

which converges to a large but finite value as s → 1.

A much more subtle statement about the behavior of ζ-function is:

Theorem 1.3.3. ζ(s) has no zeroes satisfying Re(s) ≥ 1.

2
We switched from z to s as notation for our complex variable here. Using s for the Riemann
zeta-function has been the convention since Riemann’s paper.
10 Chapter 1 Complex numbers and holomorphic functions

This has implications for the prime-counting function

π : R≥0 −→ R
x 7−→ #{primes less than or equal to x}.

A somewhat complicated argument, which relies a lot on the special properties of

integrals of holomorphic functions, deduces from this statement about the Riemann zeta
function the following statement about the prime-counting function:

Corollary 1.3.4 (Prime number theorem). We have that

x
π(x) ∼ ,
log(x)

where f (x) ∼ g(x) means f (x)/g(x) → 1 as x → ∞.

A much stronger error term on the prime number theorem would follow if we knew
that the only zeroes of ζ(s) lie on the line Re(s) = 1/2. This is the famous Riemann
hypothesis. It is one of the Millennium Prize problems, and you get a million dollars
if you solve it.3 You can find a plot of the values of ζ(s) on the line Re(s) = 1/2 at
http://mathworld.wolfram.com/pdf/posters/Zeta.pdf.

Conjecture 1.3.5 (Riemann hypothesis). All zeroes of ζ(s) which are not −2k for k > 0
lie on the line Re(s) = 1/2.

This has consequences for the error term in the prime number theorem (we will not
prove this in the course):

Corollary 1.3.6. If the Riemann hypothesis is true, then

√
 
π(x) − x  < 1 x log(x)
 
for x ≥ 2657.
 log(x)  8π

3
See https://www.claymath.org/millennium-problems/riemann-hypothesis. I would recommend
against trying to do so until you have learned more mathematics, as this problem has withstood more
than a century of attacks by some of the best mathematicians.
 Chapter 2

Functions on the complex plane

Last lecture we outlined some of the topics of the course. Today we recall the complex
plane and holomorphic functions, and establish rigorously their basic properties. This is
[SS03, Sections 1,1, 1.2].

2.1 The complex plane

As we discussed last time, the objects of interest in this course are in particular
complex-valued functions defined on subsets of C. Let’s discuss the complex plane C in
more detail.

2.1.1 Recollection: the algebra of complex numbers

Recall that the complex numbers are of the form a + bi with a, b ∈ R and i2 = −1.
Here a = Re(a + bi) is the real part and b = Im(a + bi) is the imaginary part. As a set
we can identify them with R2 , a + bi ∈ C corresponding to (a, b) ∈ R2 . The addition and
multiplication on C follow by demanding that (commutativity, associativity, distributivity,
and unitality), and that the equation i2 = −1 holds.
With these operations, C is a field: every non-zero z ∈ C has a multiplicative inverse
given by
z̄
z −1 = 2 ,
|z|
√
with a + bi = a − bi the complex conjugate, and |a + bi| = a2 + b2 the absolute value
(or norm).
In terms of the complex conjugate, we have

z + z̄ z − z̄
Re(z) = , Im(z) = .
2 2i

2.1.2 The topology of complex numbers

Using the absolute value (or norm), we will make C into a metric space. Using this,
we make sense of convergence, as well as open, closed, and compact subsets.

11
12 Chapter 2 Functions on the complex plane

Distance
The absolute value | − | gives a notion of distance on C. In fact, it provides a metric
on C: we define the distance between two points z, w ∈ C to be

|z − w|,

and this satisfies (i) |z − w| ≥ 0 with equality if and only if z = w, (ii) the triangle
inequality |z − w| ≤ |z − y| + |y − w|. Some useful additional properties are

|Re(z)| ≤ |z|, |Im(z)| ≤ |z|, and ||z| − |w|| ≤ |z − w|.

The latter follows from the triangle inequality, as we have |z| = |(z − w) − (−w)| ≤
|z − w| + |w| and similarly |w| = |(z − w) − z| ≤ |z − w| + |z|.
Remark 2.1.1. Since the distance on C defined by absolute value coincides with the
Euclidean distance upon identifying it with R2 , the properties of C and R2 as metric
spaces coincide.

Convergence
Convergence of sequences is defined as for the reals:

Definition 2.1.2. A sequence {zn }n∈N of complex numbers converges to z ∈ C when

|zn − z| → 0 as n → ∞.

Recall this means that for all  > 0 there is an N ∈ N such that |zn − z| <  for all
n > N . We say that z is the limit of the sequence {zn }n∈N , which is unique. Using the
inequalities for the absolute given above, one easily deduces that

Lemma 2.1.3. A sequence {zn }n∈N of complex numbers converges to z if and only if
the sequences of real numbers {Re(zn )}n∈N , {Im(zn )}n∈N converge to Re(z) and Im(z)
respectively.

Sometimes you don’t know the limit z yet, but you still want to check a sequence
converges. The notion of a Cauchy sequence takes care of that:

Definition 2.1.4. A sequence {zn }n∈N of complex numbers is a Cauchy sequence if

|zn − zm | → 0 as n, m → ∞.

Recall this means that for all  > 0 there is an N ∈ N such that |zn − zm | <  for all
n, m > N . As in the previous lemma, {zn }n∈N is a Cauchy sequence if and only if the
sequences of real numbers {Re(zn )}n∈N , {Im(zn )}n∈N are.
The real numbers are complete: every Cauchy sequence converges. By the previous
observation, this implies the same for the complex numbers (this is the first instance of
Remark 2.1.1):

Theorem 2.1.5. C is complete, i.e. every Cauchy sequence converges.

 2.1 The complex plane 13

Open and closed subsets

The open disk of radius r around z0 is given by

Dr (z0 ) := {z ∈ C | |z − z0 | < r}.

This is the prototypical example of an open subset:

Definition 2.1.6. A subset Ω ⊂ C is open if for each z0 ∈ C there exists a r > 0 such
that Dr (z0 ) ⊂ Ω.

Example 2.1.7. C and ∅ are both open, as is {z ∈ C | Re(z) > 0}. A good intution is
that a subset defined by a strict inequality tends to be open.

Lemma 2.1.8. Open subsets have the following properties:

· If {Ωi }i∈I is a collection of open subsets, then i Ωi is open.
S
Tk
· If {Ω1 , . . . , Ωk } is a finite collection of open subsets, then i=1 Ωi is open.

Definition 2.1.9. A subset Ω ⊂ C is closed if it is the complement of an open subset.

By taking complement in the above examples and lemma, we get:

Example 2.1.10. C and ∅ are both closed, as is {z ∈ C | Re(z) ≤ 0}. A good intution is
that a subset defined by a non-strict inequality tends to be closed.

Lemma 2.1.11. Closed subsets have the following properties:

· If {Ωi }i∈I is a collection of closed subsets, then i Ωi is xlosed.
T
Sk
· If {Ω1 , . . . , Ωk } is a finite collection of closed subsets, then i=1 Ωi is closed.

Closed subsets have a useful characterization in terms of convergent sequences, which

maybe explains the terminology “closed”: you can’t create a larger subset by adding
limit points.

Lemma 2.1.12. A subset Ω ⊂ C is closed if and only if each convergent sequence

{zn }n∈N with zn ∈ Ω has its limit in Ω.

Compact subsets
Particularly nice subsets are those are closed and bounded, in the sense that the
diameter
diam(Ω) := sup |z − w|
z,w∈Ω

is bounded.
The Bolzano–Weierstrass theorem says a subset of R2 is closed and bounded if and
only if it is compact, in the sense that every sequence has a convergent subsequence. By
Remark 2.1.1, the same holds for C:

Theorem 2.1.13. A subset Ω ⊂ C is closed and bounded if and only if every sequence
{zn }n∈N has a convergent subsequence.
14 Chapter 2 Functions on the complex plane

Remark 2.1.14. There is an equivalent formulation of compactness in terms of open

covers, see [SS03, Theorem 1.3]. This is the Heine–Borel theorem.
Next lecture we will use the following fact, which we prove as an exercise with the
definitions:
Definition 2.1.15. Suppose that Ω1 ⊃ Ω2 ⊃ Ω3 ⊃ · · · is a collection of non-empty
compact subsets whose diameter goes to 0. Then ∞
T
n=1 Ωn consists of a single point.

Proof. We first show the intersection is non-empty. Pick a point zn in each Ωn . Then
{zn }n∈N is a Cauchy sequence because the diameters of the Ωn go to 0: for n, m > N ,
|zn − zm | ≤ diam(ΩN ) and this goes to 0 as N → ∞. Thus it has a limit z ∈ C.
Since zn ∈ ΩN for n ≥ N and ΩN is closed, z ∈ ΩN . Because N was arbitrary here,
z∈ ∞
T
n=1 Ωn .
We next show by contradiction there is only a single point in the intersection. Suppose
that z, z 0 ∈ ∞ 6 z 0 . Then |z − z 0 | is non-zero, and hence bigger than some
T
n=1 Ωn and z =
 > 0. Let N be such that diam(ΩN ) < . Then z and z 0 can’t both be in ΩN , which
gives the contradiction.

2.2 Holomorphic functions

Now that we have discussed more carefully some of the properties of the complex
plane, we can revisit the definition of holomorphic functions.

2.2.1 Continuity
Having a metric on C, we can define continuity for complex-valued functions defined
on subsets of C. It is the straight-forward generalization of continuity for real-valued
functions defined on subsets of R.
Definition 2.2.1. A function f : C ⊃ Ω → C is continuous at z ∈ Ω if
lim f (z + h) = f (z).
h→0

We say that f is continuous if it is continuous at all z ∈ Ω.

In other words, for all  > 0 there exists a δ > 0 such that |h| < δ implies |f (z + h) −
f (z)| < . This has a reformulation in terms of sequences: it is continuous at z ∈ C if
and only if for all sequence {zn }n∈N in Ω converging to z we have limn→∞ f (zn ) = f (z).
By Remark 2.1.1, a function f : C ⊃ Ω → C is continuous if and only if the corre-
sponding function F : R2 ⊃ Ω → R2 of two variables, given by
(x, y) 7−→ (Re(f (x + yi)), Im(f (x + yi))),
is continuous. This makes clear that if f is continuous, so is the real-valued function
z 7→ |f (z)|. We can then use the fact that every continuous real-valued function attains
a minimum and maximum on each compact subset of Rn to deduce that:
Proposition 2.2.2. If f : C ⊃ Ω → C is continuous, then z 7→ |f (z)| is bounded on each
compact subset Ω0 ⊂ Ω and moreover attains a minimum and maximum on Ω0 .
 2.2 Holomorphic functions 15

2.2.2 Complex differentiation

Like we generalized continuity from R to C, we generalize differentiability from R to
C.

Definition 2.2.3. A function f : C ⊃ Ω → C is complex-differentiable at z ∈ Ω if there

is an a ∈ C such that
f (z + h) − f (z)
lim = a.
h→0 h
We then say that a is the derivative of f at z, and set f 0 (z) := a.
If Ω is open, we say that f is holomorphic if it is complex-differentiable at all z ∈ Ω.
If Ω is not open, we say that f is holomorphic if it extends to a holomorphic function on
a open subset Ω0 containing Ω.

In other words, for all  > 0 there exists a δ > 0 such that 0 < |h| < δ implies
|f (z + h) − f (z) − ah|/|h| < . Observe that h is a complex number, and so approaches
z from all directions.
Example 2.2.4. For c ∈ C, the constant function f given by z 7→ c defined on all of C is
holomorphic: if we take a = 0, then we have

|f (z + h) − f (z)|/|h| = 0 for all h > 0.

The identity function g given by z 7→ z defined on all of C is also holomorphic: if we

take a = 1, then we have

|g(z + h) − g(z) − h|/|h| = |z + h − z − h|/|h| = 0 for all h > 0.

Similarly, we can explicit check that each polynomial an z n + · · · + a1 z + a0 defines

a holomorphic function C → C with derivative given by nan z n−1 + · · · + a1 . This also
follows from the previous two examples and the forthcoming Lemma 2.2.7.
Non-example 2.2.5. The function f given by z 7→ z̄ is not holomorphic. For example,
moving over the real axis we get

f (a) − f (0) a
lim = lim = 1,
R3a→0 a R3a→0 a

but moving over the imaginary axis we get

f (bi) − f (0) −bi

lim = lim = −1.
R3b→0 bi R3b→0 bi

A helpful reformulation says that f is linear up to an error term:

Lemma 2.2.6. f is complex-differentiable at z with derivative f 0 (z) if and only if

f (z + h) = f (z) + f 0 (z)h + φ(z + h)h

and |φ(z + h)| → 0 as h → 0

16 Chapter 2 Functions on the complex plane

Proof. The direction ⇐ is easy, as the difference quotient is just

|f (z + h) − f (z) − f 0 (z)h| |φ(z + h)h|
= = |φ(z + h)|,
|h| |h|
which indeed goes to 0 as h → 0.
For the converse, we define φ(z + h) by
( f (z+h)−f (z)−f 0 (z)h
h if h 6= 0,
φ(z + h) =
0 if h = 0.

That its absolute value goes to 0 as h → 0 is exactly saying that f is complex-differentiable

at z with derivative f 0 (z).
The complex derivative has the same properties as the real derivative:

Lemma 2.2.7. We have that:

· If f is complex-differentiable at z, then f is continuous at z.
· If f, g are holomorphic on Ω, so is f + g. Furthermore (f + g)0 = f 0 + g 0 .
· If f, g are holomorphic on Ω, so is f g. Furthermore (f g)0 = f 0 g + f g 0 .
· If f : Ω → U ⊂ C and g : U → C are holomorphic, then g ◦ f is holomorphic on Ω.
Furthermore
(g ◦ f )0 (z) = g 0 (f (z))f 0 (z).

2.2.3 The Cauchy–Riemann equations

Even though f : C ⊃ Ω → C is continuous if and only if the corresponding function
F : R2 ⊃ Ω → R2 , Non-example 2.2.5 shows that it is not the case that f is holomorphic
if and only if F is a differentiable function of two variables. Indeed, for z 7→ z̄ the
corresponding function of two variables is (x, y) 7→ (x, −y) and this is certainly a
differentiable function of two variables with total derivative everywhere given by
" #
1 0
.
0 −1

So what is the relationship between holomorphic functions and differentiable functions

of two variables? We mentioned one such relationship during last lecture: for F to be
differentiable at ~z = (a, b) there needs to exists a (2 × 2)-matrix A such that

||(F (~z + ~h) − F (~z) − A~h||

lim = 0.
~h→~0 ||~h||
If F comes from a complex-valued function which is complex-differentiable at the point
a + bi, then A needs be a product of a scaling and a rotation. These are given by
" # " #
λ 0 cos(θ) sin(θ)
0 λ − sin(θ) cos(θ),
 2.2 Holomorphic functions 17

and thus satisfies a = d and b = −c in terms of its entries

" #
a b
A= .
c d

In other words, if we write

F (x, y) = (u(x, y), v(x, y)),

we get the Cauchy–Riemann equations

∂u ∂v ∂u ∂v
= , =− .
∂x ∂y ∂y ∂x
Thus we have shown one direction of the following result:

Theorem 2.2.8. A differentiable function F : R2 ⊃ Ω → R2 comes from a holomorphic

function f : C ⊃ Ω → C if and only if F satisfies the Cauchy–Riemann equations.

I will not prove the other direction here, instead look at pages 10–13 of [SS03].
 Chapter 3

Integration and Goursat’s theorem

Last time we defined holomorphic function. Today we study their integrals along
piecewise-smooth curves. In particular, we will establish Goursat’s theorem on the
integral of a holomorphic function along a triangle. These are [SS03, Section 1.3, 2.1].

3.1 Integration along curves

Given a holomorphic function f : C ⊃ Ω → C, we can attempt to define its integral

along curves in Ω. We first need a definition of “curve” suitable to our purposes:

Definition 3.1.1. A parametrized curve is a function γ : [a, b] → Ω. It is smooth if

γ : [a, b] → Ω is differentiable with continuous derivative. It is piecewise-smooth if it is
continuous and there exist a = a0 < a1 < · · · < an = b such that z(t) is smooth on the
intervals [ai , ai+1 ],

a
•

b•

Figure 3.1 A piecewise-smooth curve in C.

18
 3.1 Integration along curves 19

We want to allow different parametrizations, i.e. consider two piecewise-smooth

parametrized curves
γ : [a, b] → Ω, γ̃ : [c, d] → Ω
as giving rise to the same piecewise-smooth curve if there exists a piecewise-smooth
increasing homeomorphism τ : [a, b] → [c, d] such that γ̃ = γ ◦ τ .
Remark 3.1.2. We want it be increasing, instead of just any piecewise-smooth homeomor-
phism, because we want to preserve the orientation of the interval. We will soon see that
if one uses a reparametrization that changes the orientation, the integral changes by a
sign. This is easily verified by hand, but the more subtle reason is given in Math 132: if
you reverse the orientation on the interval, the induced orientation on the boundary is
also reversed, and hence you ought to reverse the limits a and b in order to get the same
answer out.
Example 3.1.3. The line segment
γ : [0, 1] −→ C
t 7−→ ta,
is a smooth parametrized curve from 0 to a. We could have run through the same line
segment at twice the speed, obtaining a different parametrization
γ̃ : [0, 1/2] −→ C
t 7−→ 2ta.
Definition 3.1.4. If γ : [a, b] → Ω is a smooth parametrized curve and f : Ω → C is
continuous, we define
Z Z b
f (z)dz := f (γ(t))γ 0 (t)dt,
γ a
where the integral of a complex-valued function is given by integrating the real and
imaginary parts separately.
If γ : [a, b] → Ω is instead piecewise-smooth, we define this as
Z n Z ai
f (γ(t))γ 0 (t)dt.
X
f (z)dz :=
γ i=1 ai−1

The term γ 0 (t) should remind you of the Jacobian in multivariable calculus, and
serves to make the value of the integral independent of the choice of parametrization.
Indeed, the change-of-variables theorem and chain rule imply that if γ̃ = γ ◦ τ then
we have (we are assuming everything is smooth for the sake convenience, to work with
piecewise-smooth functions just insert some sums)
Z b Z d
0
f (γ(t))γ (t)dt = f (γ(τ (s)))γ(τ (s))τ 0 (s)ds
a c
Z d
= f (γ(τ (s)))(γ ◦ τ )0 (s)ds
c
Z d
= f (γ̃(s))γ̃ 0 (s)ds.
c
20 Chapter 3 Integration and Goursat’s theorem

Example 3.1.5. In particular, the following is well-defined:

Z
length(γ) = |γ 0 (t)|dt,
γ

intuitively just the length of γ in R2 .

Now that we’ve shown it to be well-defined, it makes sense to state:

Lemma 3.1.6. Integration has the following properties:

R R R
(i) It is linear, i.e. γ af (z) + bg(z)dz = a γ f (z)dz + b γ g(z)dz.

orietation gives a sign: if γ − is γ run backwards, then

R
(ii) Reversing
R γ− f (z)dz =
− γ f (z)dz.

(iii) There is an inequality

Z 
 
 f (z)dz  ≤ sup |f (γ(t))| · length(γ).
 
γ t∈[a,b]

Proof. Part (i) follows from the corresponding property of the integral of real-valued
function, and (ii) follows because once we fix an parametrization, changing the orientation
amounts to precomposing γ by t 7→ −t contributing a −1 to the derivative. For part we
use the integral version of triangle inequality (obtained from the usual triangle inequality
by some manipulation of Riemann sums)
Z  Z
 b  b
f (γ(t))γ 0 (t)dt ≤ |f (γ(t))||γ 0 (t)|dt
 

 a  a
Z b !
≤ sup |f (γ(t))| |γ 0 (t)|dt
a t∈[a,b]
Z b
= sup |f (γ(t))| |γ 0 (t)|dt
t∈[a,b] a

= sup |f (γ(t))| · length(γ).

t∈[a,b]

3.1.1 Examples of integrals

This setup makes everything sound more complicated than it actually is. Let us do
some examples:
 3.1 Integration along curves 21

Example 3.1.7. Let us integrate ez along the boundary of the unit square (positively
oriented, i.e. ran through counter-clockwise)

i 1+i
<

>
0 1

It is the sum of four integrals, obtained by parametrizing each of the four line segments
in the obvious manner:
Z 1
bottom edge: ez dz = e1 − e0 ,
0
Z 1
right edge: e1+iz idz = e1+i − e1 ,
0
Z 1
top edge: e1+i−z (−1)dz = ei − e1+i ,
0
Z 1
left edge: ei−iz (−i)dz = e0 − ei .
0
Adding these up gives 0.
This is as expected: ez is holomorphic on all of C and since the square can be
contracted to a point in C, one of the special properties of holomorphic functions
discussed in the first lecture tells us that the answer should be the same as integration
0
over the constant path. This vanishes because γconst (t) = 0 for all t.
Example 3.1.8. Let’s continue with some easy integrals over the unit circle C1 (positively
oriented, i.e. ran through counter-clockwise):

>

We shall use the convenient parametrization of this circle in terms of polar representations
of complex number as
γ(θ) = eiθ θ ∈ [0, 2π].
22 Chapter 3 Integration and Goursat’s theorem

Firstly, we have that

Z Z 2π h i
1dz = ieiθ dθ = e2πi − e0 = 0.
C1 0

Secondly, we have that

Z 2π Z 2π
1 h 4πi
Z i
zdz = eiθ ieiθ dθ = ie2iθ dθ = e − e0 = 0.
C1 0 0 2

I’ll leave it to you to verify that the same happens for all powers z n , n ≥ 0: their integrals
over C1 vanish.
What the previous two examples have in common is the following: if f has a primitive
F on Ω, i.e. F 0 (z) = f (z), then γ f (z)dz = 0 when γ is closed. This is because whenever
R

f has a primitive and γ is a piecewise-smooth curve from a to b, we have

Z Z t1
f (z)dz = f (γ(t))γ 0 (t)dt
γ t0
Z t1
= F 0 (γ(t))γ 0 (t)dt
t0
Z t1
= (F ◦ γ)0 (t)dt
t0
= F (γ(t1 )) − F (γ(t0 ))
= F (b) − F (a).

Here a just happens to equal b.

Example 3.1.9. Let us next consider the most important integral in complex analysis. It
concerns the holomorphic function

1/z : C \ {0} −→ C
z 7−→ 1/z,

and is along the positively-oriented unit circle we saw before:

>
 3.2 Goursat’s theorem 23

Parametrizing as before, we get:

Z 2π
1 1
Z
dz = ieiθ dθ
C1 z 0 eiθ
Z 2π
= idθ
0
= 2πi.

This doesn’t contradict the miraculous property discussed in the first lecture, which
says that integrals along homotopic paths are equal: the circle can’t be contracted within
the domain Ω, which has a hole at the origin. Observe that this computation implies
that f does not have a primitive on C \ {0}!

3.2 Goursat’s theorem

3.2.1 Goursat’s theorem for triangles

Theorem 3.2.1 (Goursat). Suppose that f : C ⊃ Ω → C is holomorphic. Let T ⊂ Ω ⊂ C
be a solid triangle. Let ∂T be the piecewise-smooth curve obtained by positively orientation
the boundary of T (i.e. moving over it in a counterclockwise fashion). Then we have that
Z
f (z)dz = 0.
∂T

Proof. We can subdivide our triangle T into four congruent triangles, each half the size:

If we orient the boundaries of the four triangles T11 , T12 , T13 , T14 positively as well, then
the integrals over paired edges cancel out, and we get
Z 4 Z
X
f (z)dz = f (z)dz.
∂T i
i=1 ∂T1

This implies
4 Z
Z  X   Z 
    
f (z)dz  ≤ f (z)dz  ≤ 4  f (z)dz ,
    
 i1

∂T i
i=1 ∂T1 ∂T1
   

for some i1 (of course the one whose integral has largest absolute value). If we perform
the subdivision procedure n times, we get
Z  Z 
n
   
 f (z)dz  ≤ 4 
 f (z)dz  (3.1)
∂Tnin

∂T
24 Chapter 3 Integration and Goursat’s theorem

for some triangle Tnin in the nth subdivision.

Observe that T1i1 ⊃ T2i2 ⊃ T3i3 ⊃ · · · is a nested collection of non-empty compact
subsets, so its intersection contains a unique point z0 . Near z0 we can write f as
f (z) = f (z0 ) + f 0 (z0 )(z − z0 ) + φ(z)(z − z0 ),
with |φ(z − z0 )| → 0 as z → z0 . Since integrals are linear
Z Z Z
f (z)dz = f (z0 ) + f 0 (z0 )(z − z0 )dz + φ(z)(z − z0 )dz.
∂Tnin ∂Tnin ∂Tnin

The first of these vanishes by direct calculation, so we get that it’s equal to
Z
φ(z)(z − z0 )dz,
∂Tnin

whose absolute value can be estimated as

sup |φ(z)(z − z0 )| · length(∂Tnin ).
z∈∂Tnin

Now, supz∈∂Tnin |φ(z)(z − z0 )| ≤ supz∈∂Tnin |φ(z)| · supz∈∂Tnin |(z − z0 )|, and the latter is
≤ diam(∂Tnin ) = 2−n diam(T ). On the other hand, length(∂Tnin ) = 2−n length(∂T ). We
conclude that
Z 
  diam(T )length(∂T )
 φ(z)(z − z0 )dz  ≤ sup |φ(z)| .

∂Tnin z∈∂Tnin
4n

Combining this with (3.1), we get

Z 
 

 f (z)dz  ≤ sup |φ(z)|diam(T )length(∂T ),
∂T i
z∈∂Tnn

and since |φ(z)| → 0 as z → z0 , supz∈∂Tnin |φ(z)| → 0 as z → z0 as well.

Remark 3.2.2. Observe it was crucially important that f was defined on all of T , not just
near ∂T , as the subdivided triangles have boundaries which eventually become dense in
T.
A triangle is a slightly unusual shape, but other shapes are easily decomposed into
them or approximated by them. For example, we can decompose a rectangle into two
triangles:

If we oriented the boundaries of the triangles positively, the integrals along the diagonal
cancel out and what remains is the integral over the positively-oriented boundary of the
rectangle. Thus by applying Goursat’s theorem to both triangles we get:
 3.2 Goursat’s theorem 25

Corollary 3.2.3. Suppose that f : C ⊃ Ω → C is holomorphic. Let R ⊂ Ω ⊂ C be a

solid rectangle. Let ∂R be the piecewise-smooth curve obtained by positively orientation
the boundary of R. Then we have that
Z
f (z)dz = 0.
∂R

3.2.2 A down-to-earth argument in support of Goursat’s theorem

Let us give a argument that, though not rigorous, gives some intuition for why
Goursat’s theorem for rectangles, here Corollary 3.2.3, follows from the Cauchy–Riemann
equations. These equations we deduced by observing that multiplication by a complex
numbers geometrically is given by a composition of scaling and rotation
We start similarly to the prove of Goursat’s theorem, with the observation that we
can approximately fill out our rectangle with squares whose sides have some small length
 > 0:

As before, If we integrate over the positively-oriented boundaries of each of the squares

then all of the internal edges cancel out, and we get something that will approximate
well the integral over the boundary of the rectangle as  → 0:
Z X Z
f (z)dz ≈ f (z)dz.
∂(rectangle) squares ∂(square)

The number of squares grows as 1/2 , so what we need to prove that is the integral
over the boundary of each of the squares grows as a larger power of  than 2 as  → 0.
For example, if it grew as 3 , then our integral (which is independent of ) would grow as
3 /=  as  → 0 and hence need to be 0. Make this estimate for the integral over the
boundary of each of the squares, we label the midpoints of the edges and the center of
the square as
C

D E B

and use that the integral over a square can be approximated well by:

f (A) + f (B)i − f (C) − f (D)i = (f (A) − f (C)) + i(f (B) − f (D)).
26 Chapter 3 Integration and Goursat’s theorem

For general differentiable functions f (x, y) = u(x, y) + iv(x, y)), f (A) − f (C) and
f (B) − f (D) are of order ; they are approximately

∂u ∂v ∂u ∂v
− (E) ·  − i (E) ·  and (E) ·  + i (E) · 
∂y ∂y ∂x ∂x

respectively. The results is that the integral is of order 2 ; combined with the 1/2 squares
we see that we expect a finite but non-zero value to result.
Now we finally use that f is holomorphic. This means that u and v satisfy the
Cauchy–Riemann equations:
∂u ∂v ∂u ∂v
= , =− .
∂x ∂y ∂y ∂x
Multiplying out our approximation to the integral we get
∂u ∂v ∂u ∂v
− (E) · 2 − i (E)2 + i (E) · 2 − (E) · 2
∂y ∂y ∂x ∂x
∂u ∂v ∂v ∂u
   
2 2
= − (E) − (E)  + i − (E) + (E) 2 ,
∂y ∂x ∂y ∂x
separating the real and imaginary parts. But the Cauchy–Riemann equations tell us
these cancel!
 Chapter 4

Cauchy’s integral formula

Last time we proved that the integral of a holomorphic function over the boundary of a
triangle or rectangle vanishes. Today, we use this to prove that holomorphic functions on
disks (and several other shapes) have primitives, and use this to compute some integrals.

4.1 Holomorphic functions have primitives on disks

Let time we proved that if f : C ⊃ Ω → C is a holomorphic function and T, R ⊂ Ω

are a triangle or rectangle, then we have that
Z Z
f (z)dz = 0 and f (z)dz = 0.
∂T ∂R

We will now use this to prove that if f is holomorphic on the open unit disk
D1 (0), then F has a primitive on D1 (0). That is, there exists a holomorphic function
F : C ⊃ D1 (0) → C such that F 0 (z) = f (z).
Our strategy is to mimic the fundamental theorem of analysis for real-valued
Rx
functions
of a single variable: a primitive of a continuous f is given by F (x) = 0 f (x)dx. In our
case there is a similar prefered linear path from the origin to z ∈ D1 (0), and Goursat’s
theorem will be used to show that integrating along this gives a primitive.

Theorem 4.1.1. If f is holomorphic on D1 (0), then there exists a holomorphic function

F : C ⊃ D1 (0) → C such that F 0 (z) = f (z).

Proof. For z ∈ D1 (0), let γz be the smooth curve from 0 to z given by

γz : [0, 1] −→ D1 (0)
t 7−→ tz.

Then we define F : D1 (0) → C by

Z
F (z) := f (w)dw.
γz

27
28 Chapter 4 Cauchy’s integral formula

What we need to show is that F is holomorphic and that F 0 (z) = f (z). Do so, we
study the differrence quotient
Z Z
F (z + h) − F (z) = f (w)dw − f (w)dw.
γz+h γz

From the picture

z + h•
z•

we see that there is a triangle with side γz and γz+h with opposite orientations, as well
as new side γz+h,z given the line segment from z to z + h. By Goursat’s theorem the
integral over its boundary vanishes, so we get that
Z Z  Z 
   
f (w)dw − f (w)dw =  f (w)dw.
   

 γz+h γz   γz+h,z 

We know that we can write f (z+u) = f (z)+f 0 (z)u+ψ(z+u)u with limu→0 |ψ(z+u)| =
0. Thus we can make the following estimate (note that the changes of variables from z
to u means we replace γz+h,z by γh )
Z  Z  Z 
0
     

 f (w)dw − f (z)h ≤  f (z)wdw +  ψ(z + w)wdw
γh γh γh
1
≤ |f 0 (z)||h|2 + sup |ψ(z + w)w| · length(γh )
2 w∈γh
!
1 0
= |h| |f (z)||h| + sup |ψ(z + w)w|
2 w∈γh

Since limu→0 |ψ(z + u)| = 0, it is bounded. This means we can estimate supw∈γh |ψ(z +
w)w| as being ≤ C|h| as h → 0. We conclude that
R 
 γz+h,z f (w)dw − f (z)h
 
1 0
 
lim ≤ lim |f (z)||h| + C|h| = 0.
h→0 |h| h→0 2

We conclude that F (z) is holomorphic with derivative f (z).

 4.1 Holomorphic functions have primitives on disks 29

Remark 4.1.2. There is another reasonable choice of piecewise-smooth path from 0 to z:

z•

By Goursat’s theorem, if we define F using this path instead, we get the same function.
Since holomorphic functions with a primitive have the property that integrals over
closed curves vanish, we conclude that:

Corollary 4.1.3 (Cauchy’s theorem). Suppose C ⊂ Ω is a circle whose interior is also

contained in Ω and f : C ⊃ Ω → C is holomorphic, then
Z
f (z)dz = 0.
C

Proof. Since Ω is open, there is a slightly large disk Dr (z0 ) containing C. By scaling and
translation, we may assume that Dr (z0 ) = D1 (0). Now apply the previous theorem.
This argument is not specific to disks. In essence, we only use one property of Ω: in
each path component of Ω there is a piecewise-linear path from some basepoint z0 to z,
which is unique up to replacing one or two edges of a triangle by the remaining two or
one edges. This is not a great definition, and we will give a much better one later, but it
clearly applies to a collection of useful shapes (see p. 42 of [SS03]):
· sectors,
· semi-circles,
· indented semi-circles,
· “keyholes”,
· squares,
· rectangles,
· parallelograms,
· regular polygons.
As a consequence, if such a shape is contained in Ω the integral over its boundary
vanishes.
30 Chapter 4 Cauchy’s integral formula

4.2 Computing some integrals

We can use this to compute some interesting integrals:

Example 4.2.1. We will prove that for ξ ∈ R
Z ∞
−πξ 2 2 −2πixξ
e = e−πx dx.
−∞

2
In other words, we prove that e−πx is its own Fourier transform (up to some constants,
depending on your definition of the Fourier transform). For ξ = 0, this gives the famous
formula Z ∞
2
1= e−πx dx.
−∞

We will assume this is already known.

2
For the proof we integrate f (z) = e−πz over the positively-oriented boundary of the
rectangle with corners R, R + iξ, −R − iξ, −R:

−R − iξ R + iξ

−R R

Our intention is to eventually let R → ∞.

Since f is holomorphic, by the version of Cauchy’s theorem for rectangles this integral
vanishes. The integral over the right sides is given by
Z ξ
2 +−2πiRy+πy 2
e−πR idy,
0

2
and thus its absolute value can be estimated as ≤ Ce−πR with C independent of R.
Thus by letting R → ∞ we can make this arbitrarily small. Thus we have that
Z ∞ Z ∞
−πx2 2 −2πixξ+πξ 2
0= e dx − e−πx dx.
−∞ −∞

Using that we know the first integral, this gives

Z ∞
2 2 −2πixξ
eπξ e−πx dx = 1,
−∞

and the desired formula follows.

 4.2 Computing some integrals 31

Example 4.2.2. We will prove that

Z ∞
sin(x) π
dx = .
0 x 2
eiz
Here we will integrate the holomorphic function f (z) = z over the positively-oriented
boundary of the following indented semi-circle:

−R  R
−

This vanishes by the version of Cauchy’s theorem for indented semicircles. The point is
now that the integrals over the two line segments can be combined as the integral over
the single line segment [, R] of the function
eix e−ix sin(x)
− = 2i .
x x x
The integral over the large semicircle can be estimated by observing |eiz | = e−Im(z)
so that |(eiz )/z| ≤ (e−Im(z) )/|z|, and hence we get
Z  Z
 eiz  π
dz  ≤ e−R sin(x) dx.


 large semi-circle z  0

For  > 0 there is a S > 0 such that if R > S, then e−R sin(x) <  except on the intervals
[0, ] and [1 − , 1]. Since e−R sin(x) ≤ 1, we can thus estimate under these conditions the
integral by 3. In particular, it vanishes as R → ∞.
We can write eiz /z = 1/z + E(z) where E(z) is bounded near 0, using the Taylor
expansion ez = ∞ n
P
n=0 z /n!. Thus the integral over the little semi-circle can be written
as
1
Z Z Z
1/z + E(z)dz = dz + E(z)dz.
small semi-circle small semi-circle z small semi-circle

Since E(z) is bounded, the absolute value of the right integral can be estimate as ≤ Cπ,
so goes to 0 as  → 0. On the other hand, the right integral can be computed as
Z π
1 1
Z
dz = − ieiθ dθ = −πi.
small semi-circle z 0 eiθ
32 Chapter 4 Cauchy’s integral formula

Letting  → 0 and R → ∞, we thus get that

Z ∞
sin(x)
−πi + 2i = 0,
0 x

and the result follows.

Example 4.2.3. You should read Example 2 on p. 44 of [SS03] to see a proof of
Z ∞
1 − cos(x) π
dx = .
0 x2 2

4.3 Cauchy integral formula

We will give one of the most important formula’s in complex analysis, from which all
the miraculous facts mentioned in the first lecture are deduced:

Theorem 4.3.1 (Cauchy’s integral formula). Suppose that f : C ⊃ Ω → C is holomorphic

and C ⊂ Ω is a circle whose interior is also contained in Ω. Then for any z in the
interior of C, we have that
1 f (ζ)
Z
f (z) = dζ.
2πi C ζ − z

Proof. Without loss of generality C is the unit circle. We will integrate along the following
path:

z•
 4.3 Cauchy integral formula 33

Since g(ζ) = fζ−z(ζ)

is holomorphic on a neighborhood of the interior of this path, the
integral over it vanishes. Letting the little corrider close up, the integrals over both sides
cancel out, and we get
f (ζ)
Z Z Z Z
0= g(ζ)dζ − g(ζ)dζ = dζ − g(ζ)dζ,
big circle little circle C ζ −z little circle

where the sign comes from the fact that in our path the little circle appears negatively-
oriented. So it suffices to prove that
Z
g(ζ)dζ = 2πif (z).
little circle

This follows by writing g near z as

f (ζ) f (ζ) − f (z) f (z)
g(w) = = + .
ζ −z ζ −z ζ −z
Since f is holomorphic, the first one is bounded when ζ is near z (as it converges to
f 0 (z)). Thus as the radius of the little circle goes to 0, the integral goes to

f (z)
Z
dζ
little circle ζ −z
and we can explicitly evaluate this as
Z 2π
f (z)
ieiθ dθ = 2πif (z).
0 eiθ
Observe that what we are integrating is a function of z which is infinitely many
differentiable. This leads to the following corollary, the only subtlety being that we need
to move the derivative inside the integral.

Corollary 4.3.2. If f : C ⊃ Ω → C is holomorphic, then it is infinitely many times

complex differentable. Moreover, if C ⊂ Ω is a circle whose interior is also contained in
Ω and contains z, then
n! f (ζ)
Z
f (n) (z) = dz
2πi C (ζ − z)n+1
Proof. The proof is by induction on n, the case n = 0 being Cauchy’s theorem. We start
by using the induction hypothesis to write the difference quotient as

f (n−1) (z + h) − f (n−1) (z) (n − 1) 1 1 1

Z  
= f (ζ) − dz.
h 2πi C h (ζ − z − h) (ζ − z)n

Writing A = 1/(ζ − z − h) and B = 1/(ζ − z), we see that we need to understand

An − B n . This can be written as (A − B)(An−1 + An−2 B + · · · + B n−1 ), with A − B =
h/((ζ − z − h)(ζ − z)), so we get

(n − 1) 1 h
Z h i
f (ζ) An−1 + An−2 B + · · · + B n−1 dz.
2πi C h (ζ − z − h)(ζ − z)
34 Chapter 4 Cauchy’s integral formula

As h → z the integrand converges absolutely uniformly, and thus we can take the limit
inside. Moreover, both A and B converge to 1/(ζ − z). The conclusion is that the
different quotient converges as h → 0, and that it converges to the following value:

(n − 1) 1 h n n! f (ζ)
Z Z
f (ζ) n−1
dz = dz.
2πi C h (ζ − z)(ζ − z) (ζ − z) 2πi C (ζ − z)n+1

This completes the induction step.

This implies that we can estimate f (z) and the derivatives f (n) (z) in terms of the
values of f on a circle:

Corollary 4.3.3 (Cauchy’s inequalities). Given f , C of radius R, and z as above, we

have that
n!
|f (n) (z)| ≤ n supζ∈C |f (ζ)|.
R
R
Proof. We use the estimate | γ g(z)dz| ≤ length(γ)·supζ∈γ |g(ζ)|. Here length(C1 ) = 2πR
and  
 f (ζ)  1
supζ∈C  n+1
= n+1 supζ∈γ |g(ζ)|.
(ζ − z)  R
 Chapter 5

Applications of Cauchy’s integral formula

Last time we proved Cauchy’s integral formula. Today we deduce several consequences
from this, which requires a discussion of power series.

5.1 Cauchy integral formula

Let us start with some recollections:

Theorem 5.1.1 (Cauchy’s integral formula). Suppose that f : C ⊃ Ω → C is holomorphic

and C ⊂ Ω is a circle whose interior is also contained in Ω. Then for any z in the
interior of C, we have that
1 f (ζ)
Z
f (z) = dζ.
2πi C ζ − z
Observe that what we are integrating is a function of z which is infinitely many times
differentiable, which implied:

Corollary 5.1.2. If f : C ⊃ Ω → C is holomorphic, then it is infinitely many times

complex differentable. Moreover, if C ⊂ Ω is a circle whose interior is also contained in
Ω and contains z, then
n! f (ζ)
Z
f (n) (z) = dz
2πi C (ζ − z)n+1

This implies that we can estimate f (z) and the derivatives f (n) (z) in terms of the
values of f on a circle:

Corollary 5.1.3 (Cauchy’s inequalities). Given f , C of radius R, and z as above, then

if z is the center of C we have that

n!
|f (n) (z)| ≤ supζ∈C |f (ζ)|.
Rn

5.1.1 Liouville’s theorem and the fundamental theorem of algebra

Let us give a first application of this:

35
36 Chapter 5 Applications of Cauchy’s integral formula

Theorem 5.1.4 (Liouville). If f : C → C is analytic and bounded, then it is constant.

Proof. Suppose the bound is |f (ζ)| ≤ C. Then the Cauchy inequality for a circle of
radius R centered as z and n = 1 gives
B
|f 0 (z)| ≤ .
R
Letting R → ∞ we see that f 0 (z) = 0. Since C is connected, this implies that f is
constant.
We now prove something promised in the first lecture: C is algebraically closed.

Corollary 5.1.5 (Fundamental theorem of algebra). If p(z) = an z n + · · · + a1 z + a0 is

not constant, then it has a root in C.

Proof. If p is not constant we may assume an 6= 0. Then we can estimate

|p(z)| ≥ |an ||z|n − (|an−1 ||z|n−1 + · · · + |a0 |),

which goes to ∞ as |z| → ∞. Thus if p has no root, it stays a bounded distance from
0 and hence 1/p(z) is a bounded holomorphic function. By Liouville’s theorem it is
constant and hence so is p. This is of course not true, by the estimate we just gave.

Corollary 5.1.6. Every polynomial p(z) = an z n + · · · + a1 z + a0 with an =

6 0 can be
written as
an (z − w1 ) · · · (z − wn )
for some wi ∈ C.

Proof sketch. This is an induction over n, the case n = 0 being obvious. By the
fundamental theorem of algebra we can find a root w1 of p(z). This implies that
p(z) = (z − w1 )q(z) for q of degree n − 1, and we can apply the induction hypothesis.
For more details, see [SS03, Corollary 4.7].

5.2 Holomorphic functions are analytic

5.2.1 Power series

We would like to have more examples of holomorphic functions, generalizing expo-
nentials and trigoniometric functions. These are defined using power series, the infinite
sum variation of polynomials.
For example, the exponential function is defined by
∞
X zn
ez := .
n=0
n!

This sum makes sense, and in fact converges absolutely: |z n /n!| = |z|n /n!, and from
|z|n
the usual real-valued exponential we know that ∞ |z| < ∞. Thus the above
P
n=0 n! = e
 5.2 Holomorphic functions are analytic 37

formula defines a function C → C, and we will soon see it is holomorphic with derivative
computed term by term. The result is that it’s equal to its own deritative:
∞ ∞
z n−1 z n−1
(ez )0 =
X X
n = = ez .
n=0
n! n=1
(n − 1)!

Power series don’t need to converge everywhere. The classical example is the geometric
series
∞
1 X
= zn,
1 − z n=0
which converges absolutely for |z| < 1. In general, there is a critical radius inside of
which a power series converges and outside of which it diverges (on the circle itself the
behavior is very complicated).

Theorem 5.2.1. Let

∞
X
an z n
n=0

be a power series, and define R ∈ [0, ∞] by1

 
1/R = lim sup |an |1/n = lim sup{|am |1/m | m ≥ n} .
n→∞ n→∞

Then the power series converges absolutely for z with |z| < R and diverges for some z
with |z| > R.

The number R is called the radius of convergence.

1
Example 5.2.2. For 1−z , the radius of convergence is given by 1/R = lim supn→∞ 11/n = 1,
as expected.

Proof of Theorem 5.2.1. We will only prove the statement about convergence, which is
the only one we shall be using. We shall also assume that R 6= 0, ∞, and set L = 1/R.
For |z| satisfying |z| < R, choose  > 0 such that (L + )|z| = r < 1. By definition
|an |1/n < L +  for n sufficiently large, so

|an ||z|n ≤ (L + )n |z|n < rn .

P∞ n, P∞ n
By comparison to the geometric series n=0 r the series n=0 an z converges absolutely.

Observe that the candidate derivative of ∞ n ∞ n−1 , and its radius

P P
n=0 an z is n=0 nan z
of convergence is the same. A great trick is that the radius of convergence is not affected
by multiplication with z, which makes nan the coefficient of z n instead of (n + 1)an+1 ,
and so it suffices to prove that:

Lemma 5.2.3. We have that lim supn→∞ |nan |1/n = lim supn→∞ |an |1/n .
1
If the right hand side is 0, take R = ∞. If the right hand side is ∞, take R = 0.
38 Chapter 5 Applications of Cauchy’s integral formula

Proof. We claim that if limn→∞ bn = b > 0 and lim supn→∞ cn = c, then

lim sup bn cn = bc.

n→∞

We assume c 6= ∞, and we may without loss of generality assume that all cn are positive
(this is certainly the case when we want to apply the result). For  > 0 we have that
(b − )cn ≤ bn cn ≤ (b + )cn for n large enough, taking lim sup we get

(b − )c ≤ lim sup bn cn ≤ (b + )c.

n→∞

Since  was arbitrary, we can let it go to 0 and the claim follows.

It is easy to see that lim supn→∞ |n|1/n = 1,2 so we get that lim supn→∞ |nan |1/n =
lim supn→∞ |an |1/n .

In particular, ∞ n−1 has the same radius of convergence as ∞ n

P P
n=0 nan z n=0 an z . There
is a general fact that you can exchange differentation and infinite sums, as long as the
infinite sums which appear converge absolutely. A proof can be found in [SS03, Theorem
2.7]:

Theorem 5.2.4. If ∞ n
P
n=0 an z has P
radius of convergence R, then it is holomorphic on
the open disk DR (0) with derivative ∞n=0 nan z
n−1 .

Since the radius of convergence of ∞ n−1 is the same as that of ∞ n

P P
n=0 nan z n=0 an z ,
iterating this theorem we see that we can differentiate as many times as we want:
P∞ n
Corollary 5.2.5. Within its radius of convergence, n=0 an z is infinitely many times
complex differentiable.

So far we discussed power series expansions about the origin. You can expand around
z0 ∈ C by studying ∞ n
n=0 an (z − z0 ) instead.
P

5.2.2 Analytic functions

Definition 5.2.6. A function f : C ⊃ Ω → C is analytic if for each z0 ∈ Ω we can a disk
DR (z0 ) ⊂ Ω and a power series
∞
X
an (z − z0 )n
n=0

with radius of convergence > R, so that

∞
X
f (z) = an (z − z0 )n for z ∈ DR (z0 ).
n=0

This sum is called a power series expansion, or Taylor series, for f at z0 .

Example 5.2.7. exp(z), sin(z), and cos(z) are analytic on all of C. So is any polynomial.
2
For example, take log to get 1/n log(n) and observe logarithms grow slower than linear functions.
 5.2 Holomorphic functions are analytic 39

Since being complex-differentiable is a local property, Theorem 5.2.1 implies that any
analytic function is holomorphic. It follows from Cauchy’s formula that the converse is
also true:

Theorem 5.2.8. Suppose f : C ⊃ Ω → C is holomorphic and there is a closed disk

DR (z0 ) ⊂ Ω, then f has a power series expansion
∞
X
f (z) = an (z − z0 )n for z ∈ DR (z0 ),
n=0

f (n) (z0 )
with an = n! . In particular this power series has radius of convergence > R.

Proof. Our starting point is

1 f (ζ)
Z
f (z) = dζ,
2πi C ζ −z
with C a circle of radius R +  around z0 . We write
1 1 1
= ,
ζ −z ζ − z0 1 − z−z
ζ−z0
0

and use the geometric series to write this is as

∞ 
1 X z − z0 n

.
ζ − z0 n=0 ζ − z0

This converges absolutely uniformly when z−z

ζ−z0 < 1, so in particular when |z − z0 | ≤ R.
0

Hence we interchange the sum and integral in Cauchy’s integral formula to obtain
∞ n
1 f (ζ) z − z0
X Z 
f (z) = dζ
n=0
2πi C ζ − z0 ζ − z0
∞
1 f (ζ)
X Z 
= dζ (z − z0 )n
n=0
2πi C (ζ − z0 )n+1
∞
X f (n) (z0 )
= (z − z0 )n .
n=0
n!

The last equation used the integral expression for the nth derivative.
In this theorem is hidden a rather strong statement: if the radius of convergence of a
power series
∞
X
an (z − z0 )n
n=0

representing f is R, then it is impossible to extend f to some open neighborhood of

DR (z0 ): if you could, then the radius of convergence of the power series would have to
be larger.
40 Chapter 5 Applications of Cauchy’s integral formula

5.2.3 Analytic continuation

Having established two of the miraculous properties of holomorphic functions (being
infinitely many times differentiable, and being analytic), we give a precise statement of
the claim that “many holomorphic functions extend uniquely to larger domains.”
This will be a consequence of the following result, which says that the global behavior
of a holomorphic function is determined locally, as long as the domain is connected:
Theorem 5.2.9. Suppose that Ω is connected and there is a sequence of distinct points
{zn }n∈N in Ω with a limit point z ∈ Ω. Suppose that f : C ⊃ Ω → C is holomorphic, then
if f (zn ) = 0 for all n ∈ N, then f (z) = 0 for all z ∈ Ω.
Proof. Consider the power series expansion f (w) = ∞ n
n=0 an (w − z) around z. We will
P

prove that all an vanish. This follows by observing that if n is smallest such that an 6= 0,
then f (w) = an (w − z)n (1 + g(w)) with g(w) → 0 as z → 0. This is contradicts z being
a limit point of zeroes.
Now consider the subset int(f −1 (0)). This is by definition open. It is also closed,
because if a sequence z10 , z20 , . . . ∈ int(f −1 (0)) converges to z 0 ∈ Ω, then by the previous
argument f vanishes on an open disk DR0 (z 0 ). But a non-empty subset of a connected Ω
which is both open and closed must be all of Ω.
Example 5.2.10. If f vanishes on a non-empty open subset of a connected Ω, it has to be
identically zero. This means bump functions can never be holomorphic.
Corollary 5.2.11. Let Ω and {zn }n∈N be as above. Suppose that f, g : C ⊃ Ω → C are
holomorphic, then if f (zn ) = g(zn ) for all n ∈ N, then f (z) = 0 for all z ∈ Ω.
Proof. Apply the previous theorem to f − g.
This gives a way of extending holomorphic functions uniquely; a procedure called
analytic continuation. For example, if f : Ω : C and g : Ω0 → C are holomorphic functions
on connected open subsets, Ω ∩ Ω0 6= ∅, and f = g on Ω ∩ Ω0 , then
Ω ∪ Ω0 −→ C
(
f (z) if z ∈ Ω,
z 7−→
g(z) if z ∈ Ω0
is holomorphic. This is not the interesting part: by the corollary this is the only possible
extension of f to Ω ∪ Ω0 . In fact, in this argument only one of Ω or Ω0 needs to be
path-connected.
It is this technique that we will use to extend ζ(s) defined as a sum for Re(s) > 1, to
C \ {1}.
Example 5.2.12. Suppose that f : C ⊃ Ω → C has as Tayler series at z0 ∈ Ω with
convergence radius R. Then ∞ n
n=0 an (z − z0 ) is a holomorphic function on DR (z0 ). This
P

disk may very well be larger than Ω. Since DR (z0 ) is path-connected, we can thus extend
f uniquely to Ω ∪ DR (z0 ) by
(
f (z) if z ∈ Ω,
f (z) = P∞
n=0 an (z − z0 )n if z ∈ DR (z0 ).
 5.2 Holomorphic functions are analytic 41

We can keep iterating this construction of extensions using Taylor series. Does this
end? The answer is subtle: a holomorphic function may not have a “maximal” domain
of definition. Eventually, we will see that the logarithm provides an example of this.
 Chapter 6

Morera’s theorem and its applications

Last time we gave two applications of Cauchy’s integral formula: we proved that
holomorphic functions are analytic, and we used this to construct unique extensions of
holomorphic functions (“analytic extension”). This is Sections 5.1–5.4 of Chapter 2 of
[SS03].

6.1 Morera’s theorem

Two lectures ago we proved that every holomorphic function f : C ⊃ Ω → C on an

open disk Ω has a primitive F , i.e. a holomorphic F satisfying F 0 = f . This was proven
by constructing
R
F using Goursat’s theorem, which says that for every triangle
R
T ⊂ Ω we
have ∂T f (z)dz = 0. In fact, we only used that f is continuous and that ∂T f (z)dz = 0
for each triangle T . This leads to the following result.

Theorem 6.1.1 (Morera’s theorem). Suppose we are given Ra continuous function f : C ⊃

Ω → C, such that for every triangle T ⊂ Ω we have ∂T f (z)dz = 0. Then f is
holomorphic.

Proof. Since being holomorphic is a local property, we may assume that Ω is a disk.
By the preceding remark, there is a holomorphic F : C ⊃ Ω → C such that F 0 = f .
Since every holomorphic function is infinitely many times complex-differentiable, F 0 is
holomorphic and hence so is f .

We shall use this to recognize that a continuous function is holomorphic.

6.2 Constructing holomorphic functions by limits

6.2.1 Sequences
Let’s start by constructing holomorphic functions by as limits of sequences:
Question 6.2.1. Suppose that f0 , f1 , f2 , · · · : C ⊃ Ω → C are holomorphic and the limit
limn→∞ fn (z) exists for all z ∈ Ω. When is the function limn→∞ fn holomorphic?

42
 6.2 Constructing holomorphic functions by limits 43

Example 6.2.2. A special case of this appeared in our discussion of power series: the
power series ∞
P n Pm n
n=0 an z is the limit of the partial sums sm = n=0 an z when it coverges.
This happens only within its radius of convergence R, given by 1/R = lim inf |an |1/n .
That is, where the partial sums converges we get a holomorphic function.

Definition 6.2.3. A sequence f0 , f1 , f2 , . . . : C ⊃ Ω → C of functions converges uniformly

to f on a subset K ⊂ Ω if for all  > 0 there exists an N such that for all n ≥ N we have
|fn (z) − f (z)| <  for all z ∈ K.

We don’t need to know f beforehand, and may replace convergent sequences by

Cauchy sequences. That is, f0 , f1 , f2 , . . . : C ⊃ Ω → C of functions converges uniformly
on a subset K ⊂ Ω if for all  > 0 there exists an N such that for all n, m ≥ N we have
|fn (z) − fm (z)| <  for all z ∈ K.

Theorem 6.2.4. Suppose that f0 , f1 , f2 , · · · : C ⊃ Ω → C are holomorphic and converge

uniformly on all compact subsets K ⊂ Ω. Then f = limn→∞ fn is holomorphic.
R
Proof. By Morera’s theorem, it suffices to prove that ∂T f (z)dz = 0 for all triangles
T ⊂ Ω. Now we use a fact from real analysis (whose proof you do not need to know),
which says that uniform convergence is a sufficient condition for interchanging limits and
integrals: Z Z Z
f (z)dz = lim fn (z)dz = lim fn (z)dz.
∂T ∂T n→∞ n→∞ ∂T

But each of the terms in the sequence on the right hand side vanish: fn is holomorphic
so Goursat’s theorem applies. The limit is then also 0, and the result follows.
Remark 6.2.5. This should be very surprising: a limit of differentiable real-valued functions
need not be differentiable again. For example, by the Weierstrass approximation theorem
any continuous function can be approximated by polynomials uniformly on compacts.
We got the complex-differentiability of f = limn fn through a rather complicated
chain of arguments. Thus it is not clear what its derivative is. It turns out to be what
you expect it to be, by the following result:

Proposition 6.2.6. Suppose that f0 , f1 , f2 , · · · : C ⊃ Ω → C are holomorphic and

converge uniformly on all compact subsets K ⊂ Ω. Then the derivatives f00 , f10 , f20 , · · ·
converge uniformly on compacts to f 0 .

Proof. Let’s start with some reductions: by replacing fn with fn − f we may assume
that f = 0 and hence f 0 = 0. Since every compact subset can be covered by finitely
many closed balls and we can translate, it suffices to take K = Dr (0).
Our goal is now to prove that supz∈Dr (0) |fn0 (z)| → 0 as n → ∞. To get this estimate,
we observe that since Ω is open, there is a δ > 0 such that Dr+δ (z0 ) ⊂ Ω. As fn converge
uniformly to 0 on Dr+ (0), this follows once we prove the estimate

sup |fn0 (z)| ≤ C sup |fn (z)|

z∈Dr (0) z∈Dr+δ (0)

for some constant C > 0.

44 Chapter 6 Morera’s theorem and its applications

But this follows from the Cauchy inequality applied to a circle C of radius r > δ
around z:
1 1
|fn0 (z)| ≤ sup |fn (z)| ≤ sup |fn (z)|
r z∈C δ z∈Dr+ (0)

when we take C = 1δ .

By induction the same holds for the higher derivatives: the sequence of holomorphic
(k) (k) (k)
functions f0 , f1 , f2 , · · · converges uniformly on compact subsets to f (k) .

6.2.2 Sums
Pm
Taking the functions to be the partial sums sm = n=0 fn , we obtain:

Corollary 6.2.7. Suppose that f0 , f1 , f2 , · · · : C ⊃ Ω → C are holomorphic and the sum

P∞ P∞
n=0 fn converges uniformly on all compact subsets K ⊂ Ω. Then f = n=0 fn is
holomorphic.

Proof. By definition, convergence of the sum means that the partial sums sm converge
as m → ∞. Now apply Theorem 6.2.4.

As a consequence of the propositions, we see that we can compute the derivative of a

sum as in the corollary, by taking the sum of the derivatives of the terms. The higher
derivatives are similar.

Example 6.2.8. For Re(s) > 1, the Riemann ζ-function

∞
X 1
ζ(s) =
n=1
ns

is defined by a sum. We claim this converges uniformly on all half-planes {s ∈ C |

Re(s) ≥ 1 + } for  > 0. Indeed, we have
 
1 1 1
 =
 ns  Re(s)
≤ 1+ ,
n n

which is a convergent sum as  > 0. For example, you can estimate from above by
Z ∞ ∞
1 1

1+ dt = 1 + −(1 + )  = 2 + .
1 t1+ t 1

Since every compact subset of {s ∈ C | Re(s) > 1} is contained in one of these half-
planes, the sum defining the Riemann ζ-function is uniformly convergent on compacts.
By the Corollary ζ(s) is holomorphic on {s ∈ C | Re(s) > 1}.
 6.2 Constructing holomorphic functions by limits 45

6.2.3 Integrals
Since integrals are similarly limits of finite sums, now Riemann sums instead of partial
sums, we get a similar result for holomorphic functions defined by integrals:

Corollary 6.2.9. Suppose that F : C × [0, 1] ⊃ Ω × [0, 1] → C has the property that (i)
for
R1
each s ∈ [0, 1], F (−, t) : C ⊃ Ω → C is holomorphic, (ii) F is continuous. Then
0 F (z, t)dt defines a holomorphic function Ω → C.
1 PN
Proof sketch. Apply Theorem 6.2.4 to the Riemann sums N k=1 f (z, k/N ). See pages
56-57 of [SS03] for the details.
Of course we may replace [0, 1] by any other closed interval [a, b]. As before, the
derivative F 0 of F will be as expected, 01 ∂F∂z
(z,t)
R
dt. The higher derivatives are similar.
Example 6.2.10. For Re(s) > 0, the Gamma function
Z ∞
Γ(z) = e−t tz−1 ds
0

is defined by an integral. Let us sketch why it is holomorphic, details will appear later
in this course. Firstly, we claim that uniformly on compacts in {s ∈ C | Re(s) > 0}, we
have that Z 1/
lim e−t tz−1 dt.
→0 

If so, we can use Theorem 6.2.4 to prove that Γ is holomorphic if we can prove that the
function (z, t) 7−→ e−t tz−1 satisfies the condition of the previous Corollary. But its clear
that (i) it’s holomorphic for fixed t ∈ [, 1/], and (ii) continuous.
The Gamma function will play a role when studying the Riemann ζ-function. It also
appears in the formula for the volume of a ball B1 of radius 1 in Rn :

π n/2
vol(B1 ) = .
Γ(n/2 + 1)

6.2.4 Pointwise convergence

Suppose we drop the condition that fn → f uniformly on compacts, and only demand
that fn (z) → f (z). This was our original question, and we can answer it with some
sledgehammers from real analysis and complex analysis: the Baire category theorem and
Montel’s theorem. We will prove the latter in a later lecture.

Theorem 6.2.11 (Corollary of the Baire category theorem). If K ⊂ C is compact and

we can write K = ∞
S
j=1 Kj as a countable union of compact subsets, then some Kj is
somewhere dense in K (i.e. its closure contains an open disk).

Theorem 6.2.12 (Corollary of Montel’s theorem). If a sequence f0 , f1 , f2 , . . . : C ⊃

Ω → C is a sequence of functions such that for all compacts K ⊂ Ω there is a B > 0 such
that |fn (z)| < B for all z ∈ K. Then the sequence has a subsequence which converges
uniformly on compacts.
46 Chapter 6 Morera’s theorem and its applications

Theorem 6.2.13. If f0 , f1 , f2 , . . . : C ⊃ Ω → C is a sequence of holomorphic functions

that converges pointwise to a function f . Then f is holomorphic on a dense open subset
of Ω.

Proof. Let Dr (z) be a closed disk in Ω. Since z and r are arbitrary, it suffices to prove
that f is holomorphic on an open subset of Dr (z).
Take K = Dr (z) and define

Kj = {z ∈ Dr (z) | |fn (z)| ≤ j for all n ≥ 0.}.

Since the fn (z) converges to f (z), they are bounded and ∞

S
j=1 Kj = K. Furthermore,
each Kj is an intersection n {z ∈ Dr (z) | |fn (z)| ≤ j} of closed subsets. Hence Kj is
T

closed, and since it also bounded it is compact. By the Baire category theorem some
Kj has the property that Kj contains an open disk Dr0 (z 0 ). Thus the restrictions of
f0 , f1 , f2 , . . . to any compact are in absolute value bounded by j.
By Montel’s theorem, there is a subsequence which converges on compacts, necessarily
to f . By Theorem 6.2.4, f is holomorphic on Dr0 (z 0 ).

6.3 Schwarz reflection

Since being holomorphic is a local condition, if f : C ⊃ Ω → C and f˜: Ω̃ → C are

holomorphic functions and f = f˜ on Ω ∩ Ω̃, then
(
f (z) if z ∈ Ω,
z 7−→
f˜(z) if z ∈ Ω̃,

is holomorphic. We can think of this as an extension of f from Ω to Ω ∪ Ω̃. This extension

is unique if either Ω or Ω̃ is connected.
Today we give a different result for producing extensions of holomorphic functions:
the Schwarz reflection principle. Let Ω ⊂ C be an open subset symmetric with respect
to reflection in the real axis, i.e. z ∈ Ω if and only z ∈ Ω. Then the subsets Ω+ ⊂ {z ∈
C | Im(z) ≥ 0} and Ω− ⊂ {z ∈ C | Im(z) ≤ 0} are mapped to each other by complex
conjugation, and they intersect in Ω+ ∩ Ω− ⊂ R.
Let f + : C ⊃ Ω+ → C and f − : C ⊃ Ω− → C be continuous functions which coincide
on Ω+ ∩ Ω− = Ω ∩ R. Then the formula
(
f + (z) if z ∈ Ω+ ,
f (z) :=
f − (z) if z ∈ Ω− ,

defines a continuous function on Ω.

Lemma 6.3.1. With notation as above, further suppose that f + is holomorphic on

Ω+ ∩ {z ∈ C | Im(z) > 0} and f − is holomorphic on Ω+ ∩ {z ∈ C | Im(z) > 0}. Then f
is also holomorphic.
 6.3 Schwarz reflection 47

Proof. It is clear that f is complex-differentiable at z ∈ Ω \ R. To prove it is complex-

differentiable at z0 ∈ Ω ∩ R, we may restrict our attention to an open ball D1 (z0 ) ⊂ Ω.
Ω+ , then ∂T f (z)dz =
R
Let us take a triangle T ⊂ Ω. If T lies in the interior of
f + (z)dz, which vanishes by Goursat. If T lies in Ω+ , ∂T f (z)dz is the limit of
R R
R∂T + + −
∂Ti f (z)dz with Ti in the interior of Ω and also vanishes. The same is true for Ω , so
it suffices to restrict our attention to T which intersect both the interior of Ω and Ω− .
+

In that case, we can write the integral

Z
f (z)dz
∂T

as a sum of the integral over the boundaries of two or three triangles, each of which
vanishes. The following is a typical example:

Thus we have proven that the integral over the boundaries of all triangles vanishes,
and conclude by Morera’s theorem that f is holomorphic on D1 (z0 ) as well.

Since there is no condition on the real axis apart from continuity, we can use this to
easily construct examples:

Theorem 6.3.2 (Schwarz reflection principle). Let Ω be as above, and suppose that
f + : Ω+ → C is continuous satisfying (i) its restriction to Ω ∩ R is real-valued, and (ii)
its restriction to Ω+ ∩ {z ∈ C | Im(z) > 0} is holomorphic. Then the function
(
f + (z) if z ∈ Ω+ ,
f (z) :=
f + (z) if z ∈ Ω− ,

is also holomorphic.

Proof. Since f is real-valued on the real axis, f + (z) = f + (z) for z ∈ R. Furthermore,
to verify f + (z) is holomorphic, we use that if near z0 the function f has a Taylor series
P∞ n +
n=0 an (z − z0 ) then f (z) is given near z0 by

∞
X
an (z − z0 )n .
n=0

This has the same radius of convergence as the original Taylor series, and hence exhibits
f + (z) as a holomorphic function near z0 .
48 Chapter 6 Morera’s theorem and its applications

Observe that the extension f of f + to Ω is unique. If f˜ is another extension, it must

necessary coincide with f on Ω+ ∩ {z ∈ C | Im(z) > 0}. Since this is open and meets
every connected component of Ω, this means that f˜ = f on each connected component
of Ω and hence they are equal.
Remark 6.3.3. There are straightforward generalizations of this result: we can replace
the condition that f takes real values on the real axis by the condition that it takes
imaginary values: then f + (z) needs to be replaced by −f + (z). We can similarly replace
the real axis by the imaginary axis, or even the circle of radius 1 around the origin. The
latter will be an exercise.
Example 6.3.4. Suppose that f : C → C takes real values on the real axis, and imaginary
values on the imaginary axis. Then we can apply the Schwarz reflection principle twice
to the restriction of f to {z ∈ C | Re(z) ≥ 0 and Im(z) ≥ 0} and get a holomorphic
function: 


 f (z) if Re(z) ≥ 0 and Im(z) ≥ 0,

f (z) if Re(z) ≥ 0 and Im(z) ≤ 0,

g(z) =


 −f (−z) if Re(z) ≤ 0 and Im(z) ≥ 0,

−f (−z) if Re(z) ≤ 0 and Im(z) ≤ 0.


Indeed, this is arranged to be continuous and holomorphic on each of the open quadrants.
Observe that g satisfies g(−z) = −g(z), i.e. g is odd.
Since f and g coincide on {z ∈ C | Re(z) ≥ 0 and Im(z) ≥ 0}, they must be equal
and hence f must be odd. Thus we have proven that if f : C → C takes real values on
the real axis, and imaginary values on the imaginary axis, it must be odd!
 Chapter 7

Poles

We give now discuss holomorphic functions with singularities, i.e. there are some
points where they are not defined. We start the classification of these
R
singularities, and
discuss the case of poles. Then we explain how to evaluate integrals γ f (z)dz along loops
γ that contain a pole in their interior. This is Section 1–2 of Chapter 3 of [SS03].

7.1 Poles and zeroes

7.1.1 Isolated singularities

Definition 7.1.1. A point z0 ∈ C is an isolated singularity of a holomorphic function
f : Ω → C if Ω contains a punctured disk Dr (z0 ) \ {z0 }.
There are three types of isolated singularities:
· removable singularities, where |f (z)| is bounded as z → z0 ,
· poles, where |f (z)| goes to ∞ as z → z0 ,
· essential singularities, all other cases.
Today we will talk about the second case, which we will approach by studying 1/f (z).
We will talk about removable and essential singularities next lecture.

7.1.2 Zeroes and poles

Definition 7.1.2. A point z0 ∈ Ω is a zero of a holomorphic function f : Ω → C if
f (z0 ) = 0.
Suppose Ω is connected and z0 is a zero of a holomorphic function f : Ω → C. Unless
it is 0 everywhere, it can’t vanish in a neighborhood of z0 . In fact, something stronger is
true: there is an open neighborhood U of z0 in Ω, n ≥ 1, and a non-vanishing holomorphic
function g : U → C such that
f (z) = (z − z0 )n g(z) for z ∈ U .
This is proven by observing that on a neighborhood of z0 , we have
∞
X
f (z) = an (z − z0 )n .
n=0

49
50 Chapter 7 Poles

Letting n be the lowest integer such that an 6= 0 (which exists since f is not 0 everywhere),
we can take ∞ X
g(z) = 1 + an+m (z − z0 )m .
m=1
Furthermore, fixing U the integer n and function g are unique.
Definition 7.1.3. We call n the multiplicity of the zero z0 (or say z0 is a zero of order
n).
Example 7.1.4. Recall that
eπiz + e−πiz
cos(πz) = .
2
This means cos(πz) = 0 is equivalent to eπiz = −e−πiz . Using the fact that the exponential
function has no zeroes, this happens when e2πiz = −1. This is the case when z = k + 1/2
for k ∈ Z. Thus cos(πz) has infinitely many zeroes, exactly at all half-integers on the
real line.
We claim they all have multiplicity 1. Let’s prove this at z = 1/2: as cos(π(z −1/2)) =
− sin(πz), we may equivalently consider the multiplicitive of sin(πz) at the origin. But
the Taylor series sin(πz) = πz + higher order terms starts with a linear term.
Let us apply this to poles, and for that purpose we consider a holomorphic function
f : Ω → C such that Dr (z0 ) \ {z0 } ⊂ Ω for some r > 0.
Definition 7.1.5. We say f as above has a pole at z0 if there is an r > 0 such that f is
non-zero on Dr (z0 ) \ {z0 }, and the function
(
1/f (z) if z ∈ Dr (z0 ) \ {z0 },
z 7−→
0 if z = z0 ,

is holomorphic.
Writing 1/f for the above function, it has a 0 of some order n at z0 and hence we
can write
1/f (z) = (z − z0 )n g(z)
with g non-vanishing. The integer n is unique, and we say z0 is a pole of order n. A pole
of order 1 is said to be simple.
1
Example 7.1.6. The function cos(πz) has poles of order 1 at each half-integer. The function
z−1/2
cos(πz)has poles at each half-integer except 1/2 (there it will turn out to have a removable
singularity).
We then get that:
Proposition 7.1.7. If f has a pole of order n at z0 , we can write
1
f (z) = h(z)
(z − z0 )n
on Dr (z0 ) \ {z0 } with h a non-vanishing holomorphic function Dr (z0 ) → C.
 7.1 Poles and zeroes 51

P∞
Writing h(z) = n=0 bn (z − z0 )n , we get

a−n a−1
f (z) = n
+ ··· + + H(z),
(z − z0 ) z − z0

where a−n+i = bi and H : Dr (z0 ) → Ω is holomorphic.

The terms
−a−n a−1
+ ··· +
(z − z0 )n z − z0
are the principal part, and a−1 is the residue of f at z0 . Let us record this more
prominently:
resz0 f := a−1 .

Example 7.1.8. We can compute the residue as

1 dn−1
resz0 f = lim (z − z0 )n f (z).
z→z0 (n − 1)! dz n−1

7.1.3 Laurent series

P∞ j,
If we were to expand H by a power series H(z) = j=0 aj z we would get

∞
X
f (z) = aj z j ,
j=−n

the Laurent series of f around the pole z0 .

More generally, a Laurent series can have infinitely many negative powers. These
exist as series expansions for holomorphic functions on annuli. For 0 < r < R, let
Ar,R = {z ∈ C | r < |z| < R} be an annulus around the origin.

Theorem 7.1.9. Suppose that f : Ω → C is holomorphic and Ar,R ⊂ Ω. Then we have

that
∞
X
f (z) = an z n
n=−∞

on Ar,R , where the series converges absolutely.

Proof sketch. Using a “slit annulus contour” we see that for some small δ > 0 we can
write
1 f (ζ) 1 f (ζ)
Z Z
f (z) = dζ − dζ.
2πi CR+δ z − ζ 2πi Cr−δ z − ζ
1
We can expand the term z−ζ = − ζ1 1−z/ζ
1
in the first integral as a geometric series, and
1 1 1
the term z−ζ = z 1−ζ/z in the second integral as a geometric series. The first gives the
positive powers of z, the second the negative powers.
52 Chapter 7 Poles

7.2 The residue formula

Recall that if f : Ω → C is holomorphic and C ⊂ Ω is a circle whose interior is also

contained in Ω, then we have Z
f (z)dz = 0.
C
Next suppose that it is the interior of C is also contained in Ω with the exception of a
pole z0 . Then we have the following:
Theorem 7.2.1. Let f , C be as above, then
Z
f (z)dz = 2πi resz0 f.
C

Proof. Consider a “keyhole contour” as follows:

z0 •

By Cauchy’s theorem, the integral over this vanishes. Letting the distance δ between the
parallel line segments go to 0, we get that
Z Z
f (z)dz − f (z)dz = 0,
C C

where C is the tiny circle of radius  around z0 that appears when δ → 0.

We conclude that
a−n a−1
Z Z
f (z)dz = n
+ ··· + + H(z)dz,
C C (z − z0 ) z − z0
which is the equal
R
to the sums of the integrals of each of the terms on the right hand side.
The integral C H(z)dz vanishes because H is holomorphic on the interior of C . The
other integrals we can compute explicitly using the parametrization θ 7→ z0 + eiθ of the
circle C :
Z 2π (
a−j 2πia−1 if j = 1,
Z
i(j−1)θ
dz = a−j ie dθ =
C (z − z0 )j 0 0 otherwise.
 7.3 Integrals using the residue formula 53

This proves the theorem, once we recall that resz0 f is by definition a−1 .
More generally, if there are k poles z1 , . . . , zk in the interior of C we can use a
“multiple keyhole contour” to prove:
Corollary 7.2.2. Let f , C be as above, then
Z k
X
f (z)dz = 2πi reszj f.
C j=1

More generally, we can replace C by your favorite toy contour; rectangles, keyholes,
indented semicircles, etc.

7.3 Integrals using the residue formula

The residue formula is one of the most powerful methods to evaluate integrals.
Example 7.3.1. Let us compute Z ∞
1
2
dx.
−∞ 1 + x
We shall take a contour given by a semi-circle SR of radius R > 1 in the upper half-plane:

1
Since f (z) = 1+z 2 has two poles, at ±i, this contains a single pole at z0 = i. The residue
is given by writing
1 1 1
2
= ,
1+z z+iz−i
which by expanding the first term around i gives us the Laurent series
−i 1
f (z) = + higher order terms
2 z−i
around i. Thus the residue at i at −i/2, and we get
1
Z
dz = 2πi · −i/2 = π.
SR 1 + z2
We can write the integral over SR as the sum of the integral
Z R
1
dx
−R 1 + x2
54 Chapter 7 Poles

+
and the integral over the upper semicircle SR , which we can estimate as being at most
 
 1 
+
length(SR ) · sup 
  . = πR √ 1 ,
+ 1 + z
2
1 + R4

z∈SR

which goes to 0 as R → ∞. Thus letting R → ∞ we get that

Z ∞
1 1
Z
dx = lim dz = π.
−∞ 1 + x2 R→∞ SR 1 + z2
Example 7.3.2. Let’s come up with a crazy integral you can do to impress your friends.
z
We take f (z) = ez and integrate over the circle C1 of radius 1 around the origin. On the
one hand, we have a pole with residue 1 at the origin so we get
Z
f (z)dz = 2πi.
C

On the other hand, we can evaluate it by parametrizing the circle by θ 7→ eiθ to get
Z 2π eiθ
e
Z
f (z)dz = ieiθ dθ
C 0 eiθ
Z 2π
= iecos(θ)+i sin(θ) dθ
0
Z 2π
= iecos(θ) (cos(sin(θ) + i sin(sin(θ))) dθ.
0

Taking imaginary parts of both expressions we get

Z 2π
2π = ecos(θ) cos(sin(θ))dθ.
0

It’s unlikely you’d be able to evaluate the right hand side without knowing its origin.
Example 7.3.3. Let us prove compute the average of the squared absolute value of a
polynomial over the unit circle: for a polynomial

p(z) = an z n + · · · + a1 z + a0 ,

we will compute Z 2π
1
|p(eiθ )|2 dθ.
2π 0
Since we are on the unit circle C1 , z̄ = z −1 , so we have

|p(z)|2 = p(z)p(z) = p(z)p(z −1 )

with
p(z) = an z n + · · · + a1 z + a0 .
That is, we have
Z 2π
1 1 1
Z
iθ
|p(e )| dθ =2
p(z)p(z −1 ) dz.
2π 0 2πi C1 z
 7.3 Integrals using the residue formula 55

Pn
We now observe that p(z)p(z −1 ) z1 has pole at 0, with residue given by j=0 aj aj . Using
the residue theorem, we conclude that
Z 2π
1
|p(eiθ )|2 dθ = |an |2 + · · · + |a1 |2 + |a0 |2 .
2π 0

Here is a fun consequence by estimate the integral: the coefficients of a polynomial

satisfy:
 2
2 1 −1 1 
length(C1 ) · sup p(z)p(z )  = sup |p(z)|2 .
 
|ai | ≤ 
2π |z|=1 z |z|=1

There are three types of isolated singularities:

· removable singularities, where |f (z)| is bounded as z → z0 ,
· poles, where |f (z)| goes to ∞ as z → z0 ,
· essential singularities, all other cases.
Today we will talk about the second case, which we will approach by studying 1/f (z).
We will talk about removable and essential singularities next lecture.

7.3.1 Zeroes and poles

Definition 7.3.4. A point z0 ∈ Ω is a zero of a holomorphic function f : Ω → C if
f (z0 ) = 0.

Suppose Ω is connected and z0 is a zero of a holomorphic function f : Ω → C. Unless

it is 0 everywhere, it can’t vanish in a neighborhood of z0 . In fact, something stronger is
true: there is an open neighborhood U of z0 in Ω, n ≥ 1, and a non-vanishing holomorphic
function g : U → C such that

f (z) = (z − z0 )n g(z) for z ∈ U .

This is proven by observing that on a neighborhood of z0 , we have

∞
X
f (z) = an (z − z0 )n .
n=0

Letting n be the lowest integer such that an 6= 0 (which exists since f is not 0 everywhere),
we can take
∞
X
g(z) = 1 + an+m (z − z0 )m .
m=1

Furthermore, fixing U the integer n and function g are unique.

Definition 7.3.5. We call n the multiplicity of the zero z0 (or say z0 is a zero of order
n).

Let us apply this to poles, and for that purpose we consider a holomorphic function
f : Ω → C such that Dr (z0 ) \ {z0 } ⊂ Ω for some r > 0.
56 Chapter 7 Poles

Definition 7.3.6. We say f as above has a pole at z0 if there is an r > 0 such that f is
non-zero on Dr (z0 ) \ {z0 }, and the function
(
1/f (z) if z ∈ Dr (z0 ) \ {z0 },
z 7−→
0 if z = z0 ,

is holomorphic.

Writing 1/f for the above function, it has a 0 of some order n at z0 and hence we
can write
1/f (z) = (z − z0 )n g(z)
with g non-vanishing. The integer n is unique, and we say z0 is a pole of order n. A pole
of order 1 is said to be simple. We then get that:

Proposition 7.3.7. If f has a pole of order n at z0 , we can write

1
f (z) = h(z)
(z − z0 )n

on Dr (z0 ) \ {z0 } with h a non-vanishing holomorphic function Dr (z0 ) → C.

P∞
Writing h(z) = n=0 bn (z − z0 )n , we get
−a−n a−1
f (z) = n
+ ··· + + H(z),
(z − z0 ) z − z0

where a−n+i = bi and H : Dr (z0 ) → Ω is holomorphic.

The terms
−a−n a−1
n
+ ··· +
(z − z0 ) z − z0
are the principal part, and a−1 is the residue of f at z0 . Let us record this more
prominently:
resz0 f := a−1 .
Example 7.3.8. We can compute the residue as

1 dn−1
resz0 f = lim (z − z0 )n f (z).
z→z0 (n − 1)! dz n−1

7.4 The residue formula

Recall that if f : Ω → C is holomorphic and C ⊂ Ω is a circle whose interior is also

contained in Ω, then we have Z
f (z)dz = 0.
C
Next suppose that it is the interior of C is also contained in Ω with the exception of a
pole z0 . Then we have the following:
 7.4 The residue formula 57

Theorem 7.4.1. Let f , C be as above, then

Z
f (z)dz = 2πi resz0 f.
C

More generally, if there are k poles z1 , . . . , zk in the interior of C we have (which

includes the case k = 0):

Corollary 7.4.2. Let f , C be as above, then

Z k
X
f (z)dz = 2πi reszi f.
C j=1

More generally, we can replace C by your favorite toy contour; rectangles, keyholes,
indented semicircles, etc.
 Chapter 8

Singularities

We finish the classification of singularities, discussing removable and essential singu-

larities. This is Section 3 of Chapter 3 of [SS03].

8.1 Removable singularities

Suppose that f : Ω → C has an isolated singularity at z0 , i.e. Dr (z0 ) \ {z0 } ⊂ Ω.

Definition 8.1.1. We say a singularity z0 of f is a removable singularity if we can

extend f to a holomorphic function Ω ∪ {z0 }.

Such an extension is unique if it is exists, since f (z0 ) is determined by f (z) as z → z0

by continuity.

Theorem 8.1.2. |f (z)| is bounded as z → z0 if and only if z0 is a removable singularity.

Proof. The direction ⇐ follows from continuity. For the direction ⇒, we take a circle
C ⊂ Ω ∪ {z0 } centered at z0 . It suffices to extend f to the interior of C. We will prove
that
1 f (ζ)
Z
int(C) 3 z 7−→ dζ
2πi C z − ζ
is holomorphic and agrees with f on int(C) \ {z0 }.
That it is holomorphic follows from a previous theorem we had about defining
holomorphic functions by integrals, applied to the integral

f (z0 + eiθ )
Z 2π
1
z→
7 ieiθ dθ
2πi 0 z − z0 − eiθ

obtained by parametrizing the circles.

To prove this integral agrees with f at z 6= z0 in the interior of C, we use a “double
keyhole contour” to get that

1 f (ζ) 1 f (ζ) 1 f (ζ)

Z Z Z
dζ = dζ + dζ
2πi C z−ζ 2πi Cδ z−ζ 2πi Cδ0 z−ζ

58
 8.2 Essential singularities 59

where Cδ is a tiny circle of radius δ > 0 around z and Cδ0 is a tiny circle of radius δ > 0
around z0 . On the one hand, by the Cauchy integral formula, the first term is f (z). On
the other hand, we can estimate the second term as
 
 f (ζ) 
≤ length(Cδ0 ) sup 2π 
 .
0
z∈Cδ z −ζ

Since f is bounded by z → z0 and ζ stays away from z0 , the term within | − | is bounded
as z → z0 . Hence we can estimate the right hand side as ≤ 2πδC, which goes to 0 as
δ → 0.

We can use this to characterize poles:

Corollary 8.1.3. |f (z)| → ∞ as z → z0 if and only if z0 is a pole.

Proof. For direction ⇐, recall that if f has a pole as z0 , then 1/f has a zero at z0 . Thus
1/|f (z)| → 0 as z → z0 which is equivalent to |f (z)| → ∞ as z → z0 .
For the direction ⇒, observe that 1/f (z) will be bounded near z0 . Hence the
singularity is removable, and by continuity the function f extends to have a zero at
z0 .

8.2 Essential singularities

It remains to discuss the “garbage” case of essential singularities.

Definition 8.2.1. We say a singularity z0 of f is an essential singularity if it is neither

removable nor a pole.

The surprise is that we can still say anything about this case:

Theorem 8.2.2 (Casorati–Weierstrass). If f : Dr (z0 ) \ {z0 } → C is holomorphic with

an essential singularity at z0 , then f (Dr (z0 ) \ {z0 }) is dense in C.

Proof. We give a proof by contradiction. If the image is not dense, then there exists a
w ∈ C and δ > 0 such that |f (z) − w| > δ for all z ∈ Dr (z0 ) \ z0 . Then the function

1
z 7−→
f (z) − w

is holomorphic and bounded by 1/δ as z → z0 . Hence its singularity at z0 is removable.

If it extends to have non-zero value at z0 , then z0 was a removable singularity of f , which
is ruled out by the hypothesis. If it extends to have a zero, then z0 was a pole of f , which
is ruled out by the hypothesis.
60 Chapter 8 Singularities

8.2.1 Application: the automorphisms of C

Theorem 8.2.3. If f : C → C is a bijective holomorphic map with holomorphic inverse,
then f (z) = az + b for some a, b ∈ C with a 6= 0.

This will follow from the following stronger result:

Proposition 8.2.4. f f : C → C is an injective holomorphic map, then f (z) = az + b

for some a, b ∈ C with a =
6 0.

The proof of this proposition in turn will rely on a result which we will prove in the
next lecture:

Theorem 8.2.5 (Open mapping theorem). If Ω is connected and f : Ω → C is holomor-

phic and not constant, then f maps open sets to open sets.

Proof of Proposition 8.2.4. The function f (1/z) has a removable singularity, pole, or
essential singularity at 0. We will prove it has a pole, and then deduce from this that f
is a polynomial of degree 1.
First we rule out that it is a removable singularity. Since f is injective, it can’t be
bounded and hence f (1/z) is not bounded as z → 0.
Next we rule out that that it is an essential singularity. Write g(z) = f (1/z), which
is also injective. If g has an essential singularity at 0, consider the value u := g(1). By
the open mapping theorem, g(D1/2 (1)) is open in C, so contains Dδ (u) for some δ > 0.
Since g is injective, this implies that g(D1/2 (0) \ {0}) avoids Dδ (u). But this contradicts
the Casorati–Weierstrass theorem applied to D1/2 (0).
Hence f (1/z) has a pole at 0. This means we can write it as
a−n a−1
f (1/z) = + ··· + + H(z)
zn z
near 0. This implies that
f (z) − (a−n z n + · · · + a−1 z)
is bounded, and hence constant by Liouville’s theorem. Thus f (z) is a polynomial. We
claim that the only injective polynomials are those of degree 1. For suppose that f is
injective of degree n > 1. Then it must have a single zero of multiplicity n. Thus f (z) =
(z − z0 )n and this takes the same value all points z0 + e2πi·k/n for k = 0, 1, . . . , n − 1.

8.3 Meromorphic functions and the complex projective plane

Removable singularities are uniquely removable, so we may safely assume our holo-
morphic functions don’t have them. Essential singularities are hard, so it is reasonable
to restrict our attention to holomorphic function that only have poles:

Definition 8.3.1. A holomorphic function f : C \ {z1 , z2 , . . .} → C is said to be mero-

morphic if the set {z1 , z2 , . . .} is discrete and each singularity zi of f is a pole.
 8.3 Meromorphic functions and the complex projective plane 61

By considering the singularity of g(1/z) at the origin, we see that g either has a
removable singularity, pole, or essential singurality at infinity. Again, we focus our
attention on the case without essential singularities:

Definition 8.3.2. A meromorphic function is said to be meromorphic in the extended

complex plane if it has a removable singurality or pole at infinity.

We can think of such a function in terms of complex manifolds. An f which is

meromorphic in the extended complex plane extends to a function fˆ: CP 1 → CP 1 on
the extended complex plane
CP 1 = C ∪ {∞}

as follows:
· fˆ(zi ) = ∞,
· if f has a removable singularity at ∞, fˆ(∞) = limz→∞ f (z).
· if f has a pole at ∞, fˆ(∞) = ∞.
The notation reflects that CP 1 can also be though of as the complex projective plane,
whose points complex lines through the origin in C2 . It is homeomorphic to S 2 and can
be covered by two copies of C:

z 7−→ z and z 7−→ 1/z,

the latter with the understanding that 1/0 = ∞. These are the charts making CP 1 into
a complex 1-dimensional manifold. A function CP 1 → CP 1 is holomorphic if it is with
respect to these coordinates:

Definition 8.3.3. A function fˆ: CP 1 → CP 1 is holomorphic if its restrictions

· z 7→ fˆ(z) to fˆ−1 (CP 1 \ {∞}) \ {∞},
· z 7→ fˆ(1/z) to fˆ−1 (CP 1 \ {∞}) \ {0},
· z 7→ 1/fˆ(z) to fˆ−1 (CP 1 \ {0}) \ {∞},
· z 7→ 1/fˆ(1/z) to fˆ−1 (CP 1 \ {0}) \ {0},
are holomorphic.

Lemma 8.3.4. If a function f is meromorphic on the extended complex plane, its

extension fˆ: CP 1 → CP 1 is holomorphic. Conversely, the restriction of a holomorphic
map fˆ: CP 1 → CP 1 to fˆ−1 (C) \ {∞} is meromorphic on the extended complex plane.

Since CP 1 is compact and holomorphic functions are quite rigid, you might imagine
there are not many functions which are meromorphic in the extended complex plane:

Theorem 8.3.5. If f is a meromorphic in the extended complex plane then it is a

rational function, i.e. a quotient of two polynomials.
62 Chapter 8 Singularities

Proof. Near each pole zi ∈ C, we can write f as

f (z) = fk (z) + gk (z)

where fk (z) is the principal part, a polynomial in 1/(z − zi ), and gk is holomorphic near
zi . Similarly, around ∞ we can write

f (1/z) = f˜∞ (z) + g̃∞ (z)

where f˜∞ (z) is the principal part, a polynomial in 1/z, and g̃∞ is holomorphic near the
origin. Now set f∞ (z) = f˜∞ (1/z), a polynomial in z, and g∞ (z) = g̃∞ (1/z)
Observe that since f is holomorphic near ∞, it can only have finitely many poles in
C. Thus the function
n
X
F = f − f∞ − fk
k=1

is holomorphic. By constructing it only has removable singularities, both at the zi ’s and

∞. Thus we can extend it to a holomorphic C → C which is bounded. By Liouville it is
constant and we conclude that
n
X
f = constant + f∞ + fk .
k=1

Thus f is a sum of a polynomial and finitely many polynomials in 1/(z −zi ). By combining
denominators, we see it is rational.

Corollary 8.3.6. Up to a multiplicative constant, functions which are meromorphic in

the extended complex plane are determined by their zeroes and poles (with multiplicity).

8.3.1 Application: the automorphisms of CP 1

We saw above that the automorphisms of C are given by z 7→ az + b for a, b ∈ C and
a 6= 0. The extended complex plane has more automorphisms, because we can move ∞.

Theorem 8.3.7. If fˆ: CP 1 → CP 1 is a bijective holomorphic map with holomorphic

inverse, then f (z) = az+b
cz+d for some a, b, c, d ∈ C with ad − bc 6= 0.

The maps z 7→ az+b

cz+d are called Möbius transformations. If only depends on a, b, c, d
up to rescaling by a non-zero complex numbers.
If ad = bc, then it is constant as

az + b a(cz + d) ad − bc a
= − = .
cz + d c(cz + d) c(cz − d) c

If we record the coefficients as a (2 × 2)-matrix

" #
a b
,
c d
 8.3 Meromorphic functions and the complex projective plane 63

az+b a0 z+b0
then the composition of cz+d with c0 z+d0 has coefficients given by the matrix multiplication
" #" #
a0 b0 a b
.
c0 d0 c d

The conclusion is that the Möbius transformation form a group under composition,
isomorphic to the group PGL2 (C) of invertible (2 × 2)-matrices with complex entries up
to rescaling.
Proof of Theorem 8.3.7. Suppose fˆ(∞) 6= ∞. Then fˆ(∞) = x for some x ∈ C. The
1
Möbius transformation g given by z 7→ z−x sends x to ∞, so g ◦ fˆ sends ∞ to ∞. Thus
it restricts to an automorphism of C and hence must be a Möbius transformation h given
by z 7→ az + b. This means that fˆ must have been the Möbius transformation g −1 ◦ h.
Remark 8.3.8. This theorem says that the automorphisms of CP 1 as a 1-dimensional
complex manifold are given by a Lie group. This is in contrast with the automorphisms
of it as a 2-dimensional smooth manifold; the diffeomorphisms of S 2 form a infinite-
dimensional Lie group.
 Chapter 9

The argument principle

The argument principle is a technique to count zeroes and poles of a meromorphic

function within a circle. We then give a number of applications. This is Section 4 of
Chapter 3 of [SS03].

9.1 The argument principle

Suppose we want to recover whether f has a zero at z0 , and what multiplicity it has,
using an integral. If the multiplicity is n, then we can write

f (z) = (z − z0 )n g(z)

with g non-vanishing near z0 . Then we have

f 0 (z) n g 0 (z)
= + ,
f (z) z − z0 g(z)

where the term g 0 (z)/g(z) is holomorphic near z0 . In other words, it has a pole at z0
with residue n.
Remark 9.1.1. This is an example of a general phenomena:

(f1 f2 )0 f 0 f2 + f1 f20 f0 f0
= 1 = 1 + 2.
f1 f2 f1 f2 f1 f2

Similarly, if f has a pole of order n near z0 then we can write

f (z) = (z − z0 )−n h(z)

and thus
f 0 (z) −n h0 (z)
= + .
f (z) z − z0 h(z)
In other words, it has a pole at z0 with residue −n.

64
 9.2 Rouché’s theorem 65

Theorem 9.1.2 (Argument principle). Suppose that f is meromorphic on Ω containing

a circle C and its interior. If on C, f has neither zeroes nor poles, then
1 f 0 (z)
Z
dz = #{zeroes of f in C} − #{zeroes of f in C},
2πi C f (z)
where the zeroes and poles are counted with multiplicity.

Proof. Apply the residue theorem and the above observation that f 0 (z)/f (z) has a pole
with residue n at each zero of f of multiplicity n, and a pole with residue −n at each
pole of f with order n.
A similar result holds for toy contours. The argument principle says that contour
integrals of f 0 /f can be used to count zeroes and poles. We will three applications of
this idea.

9.2 Rouché’s theorem

The first concerns the relationship of the zeroes and poles of f to those of small
pertubations of f .

Theorem 9.2.1 (Rouché). Suppose that f and g are holomorphic on Ω containing a

circle and its interior. Then if

|f (z)| > |g(z)| for all z ∈ C,

both f and f + g have the same number of zeroes inside C.

Proof. Consider the family of holomorphic functions given by

ft (z) = f (z) + tg(z) with t ∈ [0, 1].

Let nt denote the number of zeroes of ft within C; the hypothesis tells us ft has no
zeroes on C, so this is given by
1 ft0 (z)
Z
nt = dz.
2πi C ft (z)
This is an integer-valued function depending continuously on t, so must be constant. We
conclude that n0 = n1 .
The same is true for toy contours.
Example 9.2.2. We will prove that the function f (z) = z + 2 − ez has exactly one zero in
the half-plane {z ∈ C | Im(z) < 0}. To do so, it suffices to prove that it has one zero in a
sufficiently large semicircle SR in this half plane, of radius R centered at the origin 0.
Since z + 2 has a unique zero in the interior of this semicircle when R > 1, we can
apply Rouché to this contour with f (z) = z + 2 and g(z) = −ez . It then suffices to
observe that
|z + 2| ≥ 2 > 1 ≥ | − ez | = e−Re(z) .
66 Chapter 9 The argument principle

9.3 The open mapping theorem

Recall that a function f : Ω → C is open if it sends open subsets to open subsets.

Theorem 9.3.1. If Ω is connected and f : Ω → C is holomorphic and non-constant,

then f is open.

If we didn’t assume Ω was connected, we just need f to be non-constant on all

connected components.

Proof. By restricting f to an open subset of interest, it suffices to prove that f (Ω) is

open. Suppose w0 = f (z0 ), then we must prove that all points w near w0 must also be
in the image of f .
Consider the functoin f (z) − w. We want to use Rouché to prove that this has a zero.
To do so, we write

f (z) − w = (f (z) − w0 ) + (w0 − w) = F (z) + G(z).

Since f is not constant, we know that f (z) = w0 + an (z − z0 )n (1 + h(z)) for some

n with an = 6 0 and h(z0 ) = 0s. That is, F (z) = an (z − z0 )n (1 + h(z)). In particular,
F (z) 6= 0 on the circle Cδ given by |z − z0 | = δ for δ > 0 sufficiently small. Take δ > 0
small enough so that Cδ and its interior are in Ω. Since Cδ is compact, there is an  > 0
such that |F (z)| >  on Cδ .
Then the hypothesis for Rouché’s theorem apply whenever |G(z)| = |w0 − w| < .
Thus for |w0 − w| <  we have that F (z) = f (z0 ) − w0 and F (z) + G(z) = f (z) − w have
the same number of zeroes within Cδ . As F (z) has at least one, so does F (z) + G(z).
We conclude from this that w is in the image of f .

Last lecture we saw an application of the open mapping theorem: the classification of
injective holomorphic functions f : C → C; these are polynomials of degree 1.

9.4 The maximum modulus principle

The final application asks whether |f | can have local maxima:

Theorem 9.4.1 (Maximum modulus principle). If Ω is connected and f : Ω → C is

holomorphic and non-constant, then |f | can’t attain a maximum in Ω.

Proof. If it did, this would violate the open mapping theorem.

Corollary 9.4.2. If Ω is connected with compact closure Ω and f : Ω → C is continuous

and holomorphic in the interior Ω, then

max |f (z)| ≤ max |f (z).

z∈Ω z∈∂Ω

Proof. Since Ω is compact, the continuous function |f | attains a maximum somewhere in

Ω. The maximum modulus principle rules out that this occurs in the interior.
 9.4 The maximum modulus principle 67

Example 9.4.3. Suppose Ω contains the closed unit disk, and that |f (z)| = 1 when |z| = 1.
By the maximum modulus principle |f (z)| < 1 when |z| < 1.
We claim that f must have a zero in the unit disk. If it didn’t, then 1/f would be
holomorphic on a neighborhood of the unit disk and since |f (z)| < 1 when |z| < 1 would
violate the maximum modulus principle.
Now recall for |w| < 1 the Blaschke factor
w−z
βw : z 7−→ .
1 − w̄z
It is holomorphic near the closed unit disk, preserves the unit circle, interchanges 0 and
w, and satisfies βw ◦ βw = id. By applying the above argument to βw ◦ f , we see that
βw ◦ f must have a zero, and hence βw ◦ βw ◦ f = f must take the value w on the unit
disk.
By letting w vary, we conclude the following result: each f as above has the entire
closed unit disk in its image.
 Chapter 10

Homotopy invariance and the logarithm

We finally finish the proof that integrals of holomorphic functions along paths are
homotopy-invariant. This implies that holomorphic functions have primitives on simply-
connected subsets. We then study the logarithm and its primitives. This is Sections 5
and 6 of Chapter 3 of [SS03].

10.1 Integrals along paths are homotopy-invariant

We now give the most general form of Cauchy’s theorem, which dispenses with the
notions of toy contours. Recall that Cauchy’s theorem says that under certain conditions
Z
f (z)dz = 0,
C

and that this is certainly the case if f admits a primitive on a neighborhood of C. Thus
finding the most general form of Cauchy’s theorem is closely related to the finding the
most general statement about the existence of primitives.
When we defined primitives in previous proofs, we did so by integrating along
piecewise-linear paths. By Goursat’s theorem this was independent of the choice of
piecewise-linear path. To define primitives more generally, we need a more general
condition under which integrals are independent of the chosen path.

Definition 10.1.1. Two (piecewise smooth) curves γ0 , γ1 : [a, b] → Ω with γ0 (a) = α =

γ1 (a) and γ0 (b) = β = γ0 (b) are said to be homotopic rel endpoints if there exists a
continuous map G : [a, b] × [0, 1] → Ω such that each γs := G|[a,b]×{s} is piecewise smooth.

It is not hard to see that among paths parametrized by [a, b] with the same endpoints,
homotopy is an equivalence relation.

Theorem 10.1.2. If f : Ω → C is holomorphic and γ0 , γ1 are homotopic rel endpoints,

then Z Z
f (z)dz = f (z)dz.
γ0 γ1
R
Proof. The idea is to show that the integrals γt f (z)dz are all equal. Thus we only need
to show that the values don’t change if paths are close to each other. We will do so by

68
 10.1 Integrals along paths are homotopy-invariant 69

writing the difference between them as a sum of integrals of holomorphic functions with
a primitive along closed paths.
Since [a, b] × [0, 1] is compact, so is its image K under G. Thus we can find an  > 0
such that for all z ∈ K, the disk D3 (z) ⊂ Ω. (For details of this statement in point-set
topology, see [SS03, p. 94].) Again using that [a, b] × [0, 1] is compact, we see that G
is uniformly continuous. Thus there is a δ > 0 such thatR |γs1 (t) − γs−2R(t)| <  for all
t ∈ [a, b] whenever |s1 − s2 | < δ. We will now prove that γs f (z)dz = γs f (z)dz. By
1 2
covering [0, 1] with finitely intervals of size δ, this proves the theorem.
To do so, we choose disks D0 , . . . , Dn of radius , points z0 , . . . , zn+1 on γs1 , and
points w0 , . . . , wn+1 on γs2 , so that the disks cover both curves and zi , zi+1 , wi , wi+1 ∈ Di .
We may assume z0 = α = w0 and zn+1 = β = wn+1 . The conditions on the radii of the
Di imply that Di ⊂ Ω.
Now we recall that f has a primitive Fi on Di . Since primitives differ by a constant,
Fi+1 (zi ) − Fi (zi ) = Fi+1 (wi ) − Fi (wi ). This implies

Fi+1 (zi+1 ) − Fi+1 (wi+1 ) = Fi (zi ) − Fi (wi ). (10.1)

Now we have that

Z Z n
X n
X
f (z)dz − f (z)dz = (Fi (zi+1 ) − Fi (zi )) − (Fi (wi+1 ) − Fi (wi ))
γs1 γs2 i=0 i=0
Xn
= (Fi+1 (zi+1 ) − Fi+1 (wi+1 )) − (Fi (zi ) − Fi (wi ))
i=0
= (Fn (zn+1 ) − Fn (wn+1 )) − (F0 (z0 ) − F0 (w0 ))
= (β − β) − (α − α) = 0.

The third equality used a telescoping of most of the sum, using (10.1).

Corollary 10.1.3. If a closed path γ is homotopic to a constant path, then

Z
f (z)dz = 0.
γ

Definition 10.1.4. Ω ⊂ C is simply-connected is if it is connected and any pair of paths

with the same endpoints is homotopic.

Remark 10.1.5. Taking the endpoints to be equal, we see that on simply-connected Ω

every closed curve is homotopic to a constant path. Conversely, if Ω is connected and
has the property that every closed curve is homotopic through closed paths to a constant
path, then Ω is simply-connected.
Example 10.1.6. An open disk is simply-connected by linear interpolation. More generally,
every open convex set is.
Example 10.1.7. The split plane C \ {(−∞, 0]} is simply-connected. Here one can linearly
interpolate the radius and argument independently.
70 Chapter 10 Homotopy invariance and the logarithm

Example 10.1.8. The punctured plane C \ {0} is not simply-connected. For example,
going from 1 to −1 across the upper-half plane is not homotopic to going from 1 to −1
across the bottom half-plane. One can prove this by integrating 1/z along semicircle and
obtaining a different answer.
The use of this definition is the following:
Theorem 10.1.9. A holomorphic function with simply-connected domain Ω has a prim-
itive.
Proof. We follow the proof of the existence of a primitive on a disk. It starts by defining
the primitive: we fix a point z0 ∈ Ω and we define F : Ω → C by
Z
F (z) := f (w)dz,
γz0 ,w

where γz0 ,w is a path from z0 to w. Such a path exists since being connected implies
path-connected, and the value is independent of the path by Theorem 10.1.2.
To check this is a primitive, we need to verify it is holomorphic with derivative f .
However, if we take h such that z + h is in a disk around z ∈ Ω which is also contained
in Ω, then the concatenation of γz0 ,z , the straight line segment η from z to z + h and
γz0 ,z+h run in opposite direction, is a closed curve. This is homotopic to a constant path,
hence the integral over it vanishes. We conclude that
Z
F (z + h) − F (z) = f (w)dw.
η

The proof now continues as in the case of a disk to show that

F (z + h) − F (z)
lim = f (z),
h→0 h
proving that F is holomorphic and F 0 = f .
Corollary 10.1.10. If f is holomorphic with simply-connected domain Ω, then for any
closed path γ we have Z
f (z)dz = 0.
γ

Remark 10.1.11. This also follows by combining Corollary 10.1.3 with the definition of
simply-connectedness.

10.2 The logarithm

We have long ignored the logarithm, as its usual definition as the primitive
R
of 1/z
does not make sense: its domain C \ {0} is not simply-connected, and as C1 1/zdz = 2πi
with C1 the circle of radius 1 around the origin, 1/z can’t have a primitive on C \ {0}.
Our solution will be to restrict the domain.
More down-to-earth, giving z = reiθ it seems reasonable to define

log(z) = log(r) + iθ.

 10.2 The logarithm 71

But then one realizes that θ is only well-defined up to addition of 2π. If we try to fix a
choice, we realize as one passes around the origin counterclockwise 2π is added to θ. As
we don’t want to lose continuity by restricting the imaginary part to [0, 2π), the solution
is again to restrict the domain so we can’t go around the origin.
What domain Ω should we restrict to? The previous section tells us it should suffice
to take Ω simply-connected.

Theorem 10.2.1. If Ω is simply-connected and satisfies 1 ∈ Ω and 0 ∈ / Ω, then there is

a holomorphic function logΩ : Ω → C such that (i) elogΩ (z) = z, and (ii) for r real near 1,
logΩ (r) is the usual real-valued logarithm log(r).

Proof. The function logΩ is the primitive of 1/z with the property that logΩ (1) = 1.
This exists by the previous section.
To check (i) holds, we should prove that ze− logΩ (z) = 1. By constructing this is true
at z = 1, and since Ω is connected it suffices to prove the derivative of ze− logΩ (z) vanishes.
But this is
d  − logΩ (z)  1
ze = e− logΩ (z) + z e− logΩ (z) = 0.
dz z
To check (ii) holds, we use that the primitive of 1/z is constructed by integrals along
paths. Near 1 on the real axis, we can take a path that moves over the real axis and get
the ordinary logarithm as a result.
The argument also works when we drop the condition that 1 ∈ Ω, though we can’t
state (ii) anymore. However, we can instead decide to fix the value of the logarithm near
any other point z0 ∈ Ω.
Example 10.2.2. Taking Ω = C \ {z | Re(z) ≥ 0}, we get the principal branch of the
logarithm and we drop Ω from the notation. We claim that for z = reiθ with −π < θ < π
this is given by
log(z) = log(r) + iθ.
This is proven by integrating for 1 along the path in Figure 8 of [SS03]. It has a Taylor
series expansion given as follows:
∞
X zn
log(1 + z) = (−1)n for |z| < 1.
n=1
n

To prove this, observe it agrees with the principal branch on the positive real axis and
has the same derivative.
Note that log(z1 z2 ) 6= log(z1 ) + log(z2 ); this is impossible when demanding the
imaginary parts lies between −π and π. Indeed, if the value of the addition has imaginary
part exceeding πi, we ought to subtract 2πi.
In keeping with the previous example, we refer to the logΩ as other branches of the
logarithm.
Example 10.2.3. We can now define z α when α ∈ C and z ∈ / {z ∈ C | Re(z) ≤ 0} by
α
taking z = e α log(z) . This uses the principal branch. If we want to define this for z with
Re(z) ≤ 0 we need to pick a logarithm on a different simply-connected Ω avoiding 0.
72 Chapter 10 Homotopy invariance and the logarithm

10.2.1 Logarithms of holomorphic functions

Using Theorem 10.2.1 we can also take logarithms of functions.

Theorem 10.2.4. If f is a non-vanishing holomorphic function with simply-connected

domain Ω, then there exists a holomorphic function g : Ω → C such that f (z) = eg(z) .
The function g is unique up to addition of 2πi.
0
Proof. Take g to be a primitive of ff with constant chosen such that f (z)e−g(z) = 1.
d
Then we prove as before that dz (f (z)e−g(z) ) = 0 to conclude to that f (z) = eg(z) .
 Chapter 11

Pre-midterm recap

Today we do a recap of the material covered by the midterm exam. It covers Chapters
1-3 of [SS03] except Sections 2.5.5 and 3.5, 3.6, and 3.7.

11.1 Holomorphic functions and their properties

The objects of interest to this course are holomorphic functions, which are maps
f : C ⊃ Ω → C with Ω ⊂ C open, which are everywhere complex-differentiable. This
means that for all z ∈ Ω there exists a f 0 (z) ∈ C such that

f (z + h) − f (z)
lim = f 0 (z).
h→0 h
Using the real and imaginary parts to consider f as a function R2 ⊃ Ω → R2 , it is not
just differentiable but its derivatives satisfy the Cauchy–Riemann equations.
This condition is much stronger than you might suspect at first, as we proved it has
the following consequences:
· Every holomorphic function is infinitely many times complex-differentiable.
· Every holomorphic function is analytic, that is, for z near z0 ∈ Ω a holomorphic
function f is equal to its Taylor expansion
∞
X 1 dn f
f (z) = an (z − z0 )n with an = (z0 ).
n=0
n! dz n

· The integral of a holomorphic function along a path only depends on its homotopy
class rel endpoints. That is, if we have a continuous map G : [a, b] × [0, 1] → Ω such
that G|[a,b]×{t} is piecewise-smooth for all t ∈ [0, 1] then
Z Z
f (z)dz = f (z)dz.
γ0 γ1

As a consequence, integrals along closed paths that can be contracted to a point

vanish.
R
For example, if C ⊂ Ω is a circle whose interior is also contained in Ω, then
C f (z)dz = 0 (Cauchy’s theorem).

73
74 Chapter 11 Pre-midterm recap

· Holomorphic functions often extend uniquely to larger domains. A concrete instance

of this is the result that if Ω ⊂ Ω0 are path-connected and g : Ω0 → C extends
f : Ω → C, then g is unique.
Another instance of this is the Schwarz reflection principle, which says that if Ω is
symmetric with respect to complex conjugation, and f : Ω ∩ {z | Im(z) ≥ 0} → C is
continuous and holomorphic on Ω ∩ {z | Im(z) > 0}, then f extends uniquely to Ω.

11.2 Constructing holomorphic functions

How do get holomorphic functions? We saw several ways of constructing them:

· We have verified some elementary functions are holomorphic on all of C: polynomials,
ez , cos(z), sin(z). Some other elementary functions are holomorphic on subsets of
C: 1/z on C \ {0}, and the principal branch log(z) of the logarithm on C \ {z |
Re(z) ≤ 0}.
· Holomorphic functions are closed under the usual constructions: addition, scaling,
multiplication, dividing, composition.
· If a sum of holomorphic functions converge uniformly on compacts, the result is
again holomorphic. Taylor series are a special case of this, and the construction of
ez is a particular case of this.
This is a special case of a limit of holomorphic functions which converges uniformly
on compacts having a holomorphic limit. That result not only applies the above
result about sums, but also to integrals: if F : Ω × [a, b] → C is continuous and
each F |Ω×{s} is holomorphic, then z 7→ ab F (z, s)ds is holomorphic.
R

11.3 The classification of singularities

Maybe the most important example an complex analysis is the function 1/z : C\{0} →
C. It is the prototypical example of a function with a pol, the most interesting type
of singularity that a holomorphic function can have. We say that f : Ω → C has a
singularity at z0 if D (z0 ) ⊂ Ω but z0 ∈
/ Ω. That is, f is not defined at z0 but is defined
at all nearby points. We proved there are three types of singularities:
(i) removable singularities: this is the case where you can extend f holomorphically
over z0 , and is characterized by f being bounded near z0 . The prototypical example
is the singularity at 0 of the constant function 1 : C \ {0} → C.
(ii) poles: this is the case where f is given by
a−n a−1
f (z) = n
+ ··· + + g(z) with an 6= 0,
(z − z0 ) z − z0
with g : Ω ∪ {z0 } → C holomorphic. Here n is the order of the pole, and a−1 is the
residue of f at z0 .
(iii) essential singularities: these are the singularities which are not removable or poles.
The prototypical example is the singularity at 0 of e1/z : C \ {0} → C. The Casorati–
Weierstrass theorem says that the image under f of a small disk around an essential
singularity is dense.
 11.4 Evaluation of integrals 75

11.4 Evaluation of integrals

The classical application of complex analysis is the evaluation of integrals. This is

done through the residue theorem, which has Cauchy’s theorem as a special case: if
f : Ω → C and γ is a toy contour in Ω such that f is defined on its interior except at
some poles z1 , . . . , zn , then we have
Z n
X
f (z)dz = 2πi reszi (f ).
γ i=1
Usually, γ will depend on some parameter R that we will want to take infinity, in
which case some parts of the paths won’t contribute. It’s best to do an example.
Example 11.4.1. Let us evaluate
x2
Z ∞
dx.
−∞ 1 + x4
z 2
We will apply the residue theorem to the function f (z) = 1+z 4 , integrated over the toy
contour given by the path [−R, R] ⊂ R closed up by a semi-circle in the upper half plane.
We will assume R > 1 and denote this SR .
The poles of f (z) are at the roots of unity
z1 = eπ/4i , z2 = e3π/4i , z3 = e5π/4i , z4 = e7π/4i
and SR encircles the first two of these. The residue at one of these points is given by
writing
z2 (z − zi )2 + 2zi (z − zi ) + zi2
= .
1 + z4 (z − zi ) j6=i (z − zj )
Q

We conclude that the residue at zi is

zi2
j6=i (zi − zj ).
Q

To evaluate this, we observe that j6=i (zi − zj ) is what you get when you evaluate the
Q

derivative of 1 + z 4 at zi , so it is 4zi3 . Thus the residue is 4z1 i .

The conclusion is that
z2 2πi −π/4i π
Z
4
dz = (e + e−3π/4i ) = √
SR 1 + z 4 2
for any R > 1.
We want to let R → ∞, and verify that the integral over the semicircle then vanishes.
Indeed, its absolute value can be estimate by
 
 z2  R2
sup · length(semicircle) ≤ πR → 0 as R → ∞.
 
4 1 + R4
 
z∈semicircle  1 + z 

We conclude that
x2
Z ∞
π
4
dx = √ .
−∞ 1 + x 2
This shows the importance of being able to find poles and residues. This is not
discussed systematically in [SS03], instead you can look at [MH87, Section 4.1].
76 Chapter 11 Pre-midterm recap

11.5 Integral formula’s and growth behavior of holomorphic functions

The important properties of holomorphic functions given above were proven by

integral formula’s. The most important of this is the Cauchy integral formula:

n! f (ζ)
Z
(n)
f (z) = dζ,
2πi C (ζ − z)n+1

with C ⊂ Ω a circle whose interiors is also contained in Ω and contains z. (You should
try to deduce this from the Taylor expansion and the residue theorem.)
This is the basis of an important estimate: if C is a circle as above, of radius R
centered at z, then
n!
|f (n) (z)| ≤ n · supζ∈C |f (ζ)|.
R
Applied to a bounded function defined on all of C, this says that |f 0 (z)| ≤ C/R. Letting
R → ∞ we deduce that f 0 (z) = 0 so f is constant. This is Liouville’s theorem.
Theorems with a similar flavor are the argument principle and maximum modules
principle:
· The argument principle says that if γ is a toy contour in Ω such that f is defined
on its interior except at some poles z1 , . . . , zn , then if f has no zeroes on γ we have

1 f 0 (z)
Z
dz = #zeroes of f in γ − #poles of f in γ,
2πi γ f (z)

where zeroes and poles are too be counted with multiplicity, resp. order.
This implies Rouché’s theorem, which says that a sufficiently small perturbation of
a holomorphic function doesn’t change the number of zeroes it has within a toy
contour.
· The maximum modulus principle says that if f is holomorphic, then |f | can’t have
local maxima unless it is locally constant.
This implies the open mapping theorem, which says that f takes open subsets to
open subsets unless it is locally constant.
 Chapter 12

Growth of zeroes

Today we start the next part of the course: the study of holomorphic functions
whose domain is all of the complex plane C. We will answer three questions about such
functions:
(1) What do the sets of zeroes of such functions look like?
(2) How do these functions grow as |z| → ∞?
(3) To what extent are such functions determined by their zeroes?
This is Sections 5.1 and 5.2 of [SS03]. The application to polynomials is from [Mah60].

12.1 The mean value property

For later use, we give another expression for the Taylor series coefficients an in
∞
X
f (z) = an (z − z0 )n .
n=0

Theorem 12.1.1. If the Taylor series above converges on DR (z0 ), we have

1 2π Z
an = f (z0 + reiθ )e−inθ dθ
2πrn 0
for n ≥ 0 and 0 < r < R. If we have n < 0, the right hand side vanishes.
f ( n)(z0 )
Proof. Recalling that an = n! , we can use the Cauchy integral formula for the nth
derivative to get
1 f (ζ)
Z
an = dζ.
2πi Cr (z0 ) (ζ − z0 )n+1
Let us parametrize the circle as θ 7→ z0 + reiθ . Then we get
1 2π f (z0 + reiθ )
Z
an = ireiθ dθ
2πi 0 (z0 + reiθ − z0 )n+1
1 2π
Z
= f (z0 + reiθ )r−(n+1) e−i(n+1)θ reiθ dθ
2π 0
Z 2π
1
= f (z0 + reiθ )e−inθ dθ.
2πrn 0

77
78 Chapter 12 Growth of zeroes

For n < 0, we get that the expression in the theorem is given by

Z 2π
1
Z
−inθ
iθ
f (z0 + re )e dθ = rac12πi f (ζ)(ζ − z0 )−n−1 dζ
2πrn 0 Cr (z0 )

which vanishes by Cauchy’s theorem when n < 0 or equivalently −n − 1 ≥ 0.

Taking n = 0, we get:

Corollary 12.1.2 (Mean-value property). If f : DR (z0 ) → C is holomorphic, then

Z 2π
1
f (z0 ) = f (z0 + reiθ )dθ
2π 0

for 0 < r < R.

We can deduce from this a mean-value property for harmonic functions by taking the
real part, as every harmonic function is the real part of a holomorphic function.

12.2 Jensen’s formula

As usual, DR and CR are the open disk and circle of radius R > 0 around the origin.
Let Ω ⊂ C contain DR and f : Ω → C be holomorphic. If f is not constant and f has
no zeroes on CR , it can have only finite many zeroes in DR : otherwise the set of zeroes
would have an accumulation point in DR and f would have to vanish on the connected
components of Ω containing DR .

Theorem 12.2.1 (Jensen’s formula). Let Ω ⊂ C contain DR and f : Ω → C be holomor-

phic. Suppose that f is not constant, f (0) 6= 0, and f has no zeroes on CR . If z1 , . . . , zr
are the zeroes of f inside CR counted with multiplicity, then
r Z 2π
|zk | 1
X  
log |f (0)| = log + log |f (Reiθ )|dθ. (12.1)
k=1
R 2π 0

You should think of this as a function that constricts the location of zeroes of f in
terms of the values of f at the origin and the circle CR .

Proof. There are four steps:

Step 1 The first observation is that f1 , f2 : Ω → C satisfy the hypothesis, then so does
their product f1 f2 . Furthermore, we claim that if f1 and f2 satisfy the conclusions
of the theorem, then so does f1 f2 . For the left hand side of (12.1) this is easy:
log |f1 (0)f2 (0)| = log |f1 (0)| + log |f2 (0)|. The same additivity holds for the integral
in the right hand side of (12.1):
Z 2π Z 2π Z 2π
1 1 1
log |f1 (Reiθ )f2 (Reiθ )|dθ = log |f1 (Reiθ )|dθ+ log |f2 (Reiθ )|dθ.
2π 0 2π 0 2π 0
 12.2 Jensen’s formula 79

Finally, the set of zeroes of f1 f2 is the union of the sets of zeroes of f1 and f2 : when
z1 , . . . , zr are the zeroes of f1 and z10 , . . . , zs0 are the zeroes of f2 , z1 , . . . , zr , z10 , . . . , zs0
are the zeroes of f1 f2 . Thus we have that the sum
r  s
|zj0 |
!
|zk |
X  X
+
k=1
R j=1
R

is the sum that appears in the right hand side of (12.1) for f1 f2 .
Step 2 Next note that
f (z)
g(z) =
(z − z1 ) · · · (z − zr )
is bounded near z1 , . . . , zr . Hence these singularities are removable and g gives a
holomorphic function Ω → C without zeroes in CR . Writing
f (z) = (z − z1 ) · · · (z − zr )g(z),
we see it suffices to prove (12.1) for linear functions, i.e. z − zi , or functions that
have no zeroes within CR , i.e. g.
Step 3 We first prove (12.1) for functions g that have no zeroes within CR . In this case,
we have to prove that
Z 2π
1
log |g(0)| = log |g(Reiθ )|dθ.
2π 0
To do so, recall that since g is non-vanishing and holomorphic on the simply-
connected set DR+ , we can write g(z) = eh(z) with h : DR+ → C holomorphic.
Rephrased in this property, we have to prove that
Z 2π
1
Re(h(0)) = Re(h(Reiθ ))dθ.
2π 0
But this is obtained by taking the real part of the mean-value property for h with
z0 = 0 and r = R:
1 2π
Z
h(0) = h(Reiθ )dθ.
2π 0
Step 4 We finally prove (12.1) for a linear function z − z0 for z0 ∈ DR . We must show
that
|z0 | 1 2π
  Z
log |z0 | = log + log |Reiθ − z0 |dθ.
R 2π 0
Rewriting, this is equivalent to
Z 2π
0= log |eiθ − z0 /R|dθ.
0

Writing a = z0 /R and substituting −θ for θ, this is equivalent to

Z 2π
log |1 − aeiθ |dθ = 0.
0

This follows by applying the mean-value property to the harmonic function log |1 −
az|, which vanishes at the origin.
80 Chapter 12 Growth of zeroes

We can reformulate Jensen’s formula in terms of the function

n(r) : R≥0 −→ Z≥0

r 7−→ #{zeroes of f in Dr counted with multiplicity},

which counts the number of zeroes by radius.

Proposition 12.2.2. Let Ω ⊂ C contain DR and f : Ω → C be holomorphic such that

f (0) 6= 0. Then for 0 < r < R we have
Z R Z 2π
n(r) 1
dr = log |f (Reiθ )|dθ − log |f (0)|.
0 r 2π 0

Proof. By Jensen’s formula, it suffices to prove that

Z R r
n(r) R
X  
dr = log ,
0 r k=1
|zk |

where as before z1 , . . . , zr are the zeroes of f counted with multiplicity.

Since n(r) = rk=1 ηk (r) with ηk (r) = 1 if r > |zk | and 0 otherwise, it suffices to
P

prove that Z R
ηk (r) R
 
dr = log .
0 r |zk |
R R ηk (r) RR 1
This follows since 0 r dr = |zk | r dr = log(R) − log(|zk |).

12.3 Functions of finite order

Using the version of Jensen’s formula given in Proposition 12.2.2, we will prove
a stronger result about the locations of zeroes for the following restricted class of
holomorphic functions:

Definition 12.3.1. For ρ > 0, a holomorphic f : C → C has order of growth ≤ ρ if there

ρ
exist A, B such that |f (z)| ≤ AeB|z| .

Definition 12.3.2. A holomorphic f : C → C is of finite order if it has order of growth

≤ ρ for some ρ > 0. Its order of growth is given by the infimum over all ρ such that f
has order of growth ≤ ρ.
3
Example 12.3.3. The function ez has order of growth 1, the function ez has order of
growth 3, and polynomials have order of growth 0 (the infimum over all positive numbers).

Theorem 12.3.4. If f : C → C is holomorphic with order of growth ≤ ρ, then

(i) n(r) ≤ Crρ for some C > 0 and r sufficiently large,
(ii) if z1 , z2 , . . . denote the zeroes of f not equal to 0, then for all s > ρ we have
∞
X 1
< ∞.
|z |s
k=1 k
 12.3 Functions of finite order 81

Proof. We start with the first part. Without loss of generality f (0) 6= 0; otherwise we
may replace f (z) by f (z)/z ` with ` the order of the zero at f at the origin. This only
changes n(r) by a constant. We have
Z R Z 2π
n(x) 1
dx = log |f (Reiθ )|dθ − log |f (0)|.
0 x 2π 0
We will estimate n(r) by the left hand side, and use the growth condition to estimate the
right hand side.
For the former, by integrating over a smaller interval, we get
Z 2r Z 2π
n(x) 1
dx ≤ log |f (Reiθ )|dθ − log |f (0)|.
r x 2π 0
Since n(x) is increasing, we have
Z 2r Z 2r
n(x) 1
dx ≥ n(x) dx = n(r)(log(2r) − log(r)) = n(r) log(2).
r x r x
ρ
For the latter, the growth estimate |f (z)| ≤ AeB|z| gives
Z 2π Z 2π Z 2π
ρ
log |f (Reiθ )|dθ ≤ log(AeB(2r) )dθ = log(A) + B(2r)ρ dθ ≤ C 0 rρ
0 0 0

for some C0
> 0 absorbing various constants and r sufficiently large.
Combined, we get
n(r) log(2) ≤ C 0 rρ − log |f (0)|,
so that picking an appropriate constant C > 0, we get
n(r) ≤ Crρ
for r sufficiently large.
For the second part, we use the first part to estimate the sum over all roots with
absolute value ≥ 1 (there are only finitely many with absolute value < 1):
 
∞
|zk |−s = |zk |−s 
X X X

|zk |≥1 j=0 2j ≤|zk |<2j+1
∞
2−js n(2j+1 )
X
≤
j=0
∞
0
2−js 2(j+1)ρ
X
≤C
j=0
∞
≤ C0
X
2(ρ−s)j < ∞.
j=0

In the first inequality, we use that |zk |−s ≤ 2−js for 2j ≤ |zk | ≤ 2j+1 . In the second, we
used the growth estimate (modifying the constant C to C 0 to absorb the fact that our
estimate only holds for r sufficiently large). In the third, we used that 2−js 2(j+1)ρ =
2(ρ−s)j+ρ and absorbed the 2ρ into C 0 to get C 00 . The final geometric series converges
because s > ρ.
82 Chapter 12 Growth of zeroes

Example 12.3.5. The condition s > ρ in the last part can’t be improved: take f (z) =
sin(πz). It has order of growth ≤ 1, as f (z) = 2i 1
(eiπz − e−iπz ) implies |f (z) ≤ Ceπ|z| .
(In fact, it is equal to 1 since it grows exactly as ex on the imaginary axis.) However, the
zeroes are at n ∈ Z with order 1 and we indeed have that
∞ ∞
X 1 X 1
s
= 2
|z |
k=1 k n=1
ns

and this converges if and only if s > 1.

12.4 An application of Jensen’s formula to polynomials

What happens when we apply Jensen’s formula to a polynomial p(z) = an z n + · · · +

a1 z + a0 ? We might as well assume an = 1 for simplicity and need a0 6= 0 to get p(0) 6= 0.
Let z1 , . . . , zn be the zeroes of p, counted with multiplicity, and assume that |zi | = 6 1 for
all i. That is, p has no zeroes on the circle C1 . Let z1 , . . . , zr denote the zeroes of p with
absolute value < 1, i.e. those in D1 . Then Jensen’s formula says
r Z 2π
X 1
log |a0 | = log |zk | + log |f (eiθ )|dθ.
k=1
2π 0

Proposition 12.4.1. We have that

2n
Z 2π
log(|a0 | + · · · + |an |) ≤ log |f (eiθ )|dθ.
2π 0

Proof. Reorganizing Jensen’s formula, we get

1 2π
 
 a0 
Z
log 
  = log |f (eiθ )|dθ.
z1 · · · zr  2π 0
a0
Let us now recall that a0 is the product of all zeroes, so z1 ···zr = zr+1 · · · zn , and we get
n Z 2π
X 1
log |zk | = log |f (eiθ )|dθ.
k=r+1
2π 0

Since log |zi | < 0 if |zi | < 1, we see that for any subset I ⊂ {1, . . . , n}, we have
X n
X
log |zi | ≤ log |zk |,
i∈I k=r+1

and hence Z 2π
X 1
log |zi | ≤ log |f (eiθ )|dθ. (12.2)
i∈I
2π 0
P Q
Now observe that the coefficient as is equal to the sum |I|=n−s i∈I zi (as usual,
the empty product is 1). Thus we have that (with |an | = 1)
X Y
|a0 | + · · · + |an | ≤ |zi |.
I⊂{1,...,n} i∈I
 12.4 An application of Jensen’s formula to polynomials 83

Now the logarithm is concave, i.e. log(x + y) ≤ log(x) + log(y), so we have

X X
log(|a0 | + · · · + |an |) ≤ log |zi |.
I⊂{1,...,n} i∈I

1 R 2π
Each of the 2n terms log |zi | can be estimated by log |f (eiθ )|dθ, so we conclude
P
i∈I 2π 0
that
2n
Z 2π
log(|a0 | + · · · + |an |) ≤ log |f (eiθ )|dθ.
2π 0

From (12.2), we can deduce another consequence. By the triangle inequality, we

have that |f (eiθ )| ≤ |a0 | + · · · + |an |. Thus we can estimate the right hand side as
≤ log(|a0 | + · · · + |an |). This directly gives

Lemma 12.4.2 (Feldman’s inequality). For every I ⊂ {1, . . . , n}, we have

X
log |zi | ≤ |a0 | + · · · + |an |.
i∈I

1 R 2π
Remark 12.4.3. We can also use this observation to bound 2π 0 log |f (eiθ )|dθ from
above and below as follows:
Z 2π
1 1
log(|a0 | + · · · + |an |) ≤ log |f (eiθ )|dθ ≤ log(|a0 | + · · · + |an |).
2n 2π 0
 Chapter 13

Weierstrass’ construction

We continue our study of holomorphic functions whose domain is all of the complex
plane C. Our goal for this lecture and the next one is the Hadamard factorization
theorem, which describes an entire function of finite order in terms of its zeroes. That is,
it is a generalization of the factorization of polynomials to holomorphic functions which
do not grow too fast.
Today we prepare by studying infinite products, and Weierstrass construction of an
entire function with prescribed zeroes. This is Sections 5.3 and 5.4 of [SS03].

13.1 Infinite products

You may have seen the expression

∞
!
sin(πz) Y z2
=z 1− 2 .
π n=1
n
In this section we make sense of such infinite products, and explain how to deduce this
as an example.
Q∞
Definition 13.1.1. An infinite product n=1 (1 + an ) converges if the sequence of partial
products N
Q
n=1 (1 + an ) does.

Proposition 5.3.1 of [SS03] gives a sufficient condition, which is proven by taking the
logarithm after discarding the finitely many terms with |an | ≥ 1/2:
Proposition 13.1.2. If ∞
Q∞
n=1 |an | < ∞, then the infinite product
P
n=1 (1+an ) converges.
Furthermore, it is 0 if and only if at least one of the terms is 0.
We will need one fact from its proof though: the rate of convergence for the product
depends on that of the sum.
This has the following consequence, which is line with previous results about infinite
sums and integrals:
Theorem 13.1.3. Suppose that {Fn }n≥1 is a sequence of holomorphic functions Ω → C.
Suppose that there exist cn > 0 such that ∞n=1 cn < ∞ and |Fn (z) − 1| < cn for all
P

z ∈ Ω.

84
 13.1 Infinite products 85

Then we have that the product ∞

Q
n=1 Fn (z) converges uniformly on Ω to a holomorphic
function F (z). For z ∈ Ω, F (z) = 0 if and only if at least one the terms Fn (z) is 0. If
F (z) 6= 0, we have
∞
F 0 (z) X Fn0 (z)
= .
F (z) F (z)
n=1 n

Proof. We may write Fn (z) = 1 + an (z), where ∞ n=1 |an (z)| < ∞ for all z ∈ Ω. The
P

first two parts that follows from the previous proposition; it is uniform because the rate
of convergence of the sum is controlled by the rate of convergence of ∞
P
n=1 cn .
For the third part, we recall that if a sequence GN (z) of holomorphic functions
converges uniformly on compacts to G(z), then the derivatives G0N (z) converge uniformly
on compacts to G0N (z). We shall apply this to GN (z) = N
Q
n=1 Fn (z), which are all
non-zero on K and hence have absolute value which is uniformly bounded below. Thus
G0N (z)/GN (z) converges uniformly on compacts to G0 (z)/G(z). Now we just need the
fact that
N
G0N (z) X Fn0 (z)
= .
GN (z) n=1 Fn (z)

13.1.1 The product formula for the sine

Let us now prove that
∞
!
sin(πz) Y z2
=z 1− 2 .
π n=1
n

 by G(z)
Let us denote the left hand side and the right hand side by P (z). Our first
 z2 
observation is that since ∞ |z| 2 P∞ n2 < ∞ for all z ∈ C, the infinite product
P
n=1  n2  = n=1
in P (z) converges uniformly to a holomorphic function. Furthermore, P (z) is non-zero if
and only z ∈/ Z, and we have
∞
P 0 (z) 1 X 2z
= + .
P (z) z n=1 z − n2
2

G0 (z)
We will show that the same expression holds for G(z) . If so, we would have
0  0
P (z) P (z) P (z) G0 (z)
 
= − = 0.
G(z) G(z) P (z) G(z)

This means that P (z) = cG(z) on C \ Z for some constant c, and by continuity this is
true everywhere. By evaluating the derivative at 0, we get

P 0 (0) = 1, G0 (0) = 1,
Q∞  z2

so we see that c = 1. The second is easy; for the first observe that P̃ (z) = n=1 1− n2
holomorphic, so P 0 (0) = P̃ (0). It is easy to see that P̃ (0) = 1.
86 Chapter 13 Weierstrass’ construction

Thus, it remains to prove that

∞
G0 (z) 1 X 2z
= + .
G(z) z n=1 z − n2
2

0
Observe that GG(z)
(z)
= π cot(πz) = π cos(πz)
sin(πz) . We claim that F (z) = π cot(πz) satisfies the
following conditions:
(i) F is defined on C \ Z,
(ii) F (z + 1) = F (z) when z ∈ C \ Z,
1
(iii) F (z) = z + F0 (z) with F0 (z) holomorphic near 0,
(iv) F (z) is odd,
(v) F (z) is bounded on |Re(z)| ≤ 1/2, |Im(z)| ≥ 1.
In fact, (i), (ii), and (iii) imply that all singularities are simple poles with residue
1. Properties (i), (ii) and (iv) are clear for π cot(πz), and (iii) follows using the Taylor
expansion sin(πz) = πz(1 + g(z)) with g(0) = 0.
We claim that
∞
1 X 2z
H(z) = +
z n=1 z − n2
2

satisfies the same properties. Properties (i), (iii), and (iv) are clear. To see (ii) holds,
we observe that z 22z 1 1
, so the right hand side is limN →∞ N 1
P
−n2
= z+n + z−n n=−N z+n .
Replacing z by z + 1 just shifts the indices (up to a small error term from the boundary
terms).
We claim that
∆(z) = F (z) − H(z)
is bounded. This means it extends to a constant function, which is zero by (iv). By
(ii), it suffices to prove it is bounded on the strip {z ∈ C | Re(z) ≤ 1/2}. By (iii), we
have that ∆ is bounded for z in the strip with |Im(z)| ≤ 1. By (v) it is bounded on
|Re(z)| ≤ 1/2, |Im(z)| ≥ 1, as both F (z) and H(z) are bounded.
We will give the outline for (v) for F (z) and H(z) now, for details see pages 143–144
of [SS03]. For F (z) this follows by writing for z = x + iy

e−2πy + e−2πix
cot(πz) = i ,
e−2πy − e−2πix
and observe that as y → ±∞ this remains bounded. For H(z) this follows by writing
∞
1 X 2(x + iy)
H(z) = + ,
x + iy n=1 x2 − y 2 − n2 + 2ixy

and observe that if |x| ≤ 1/2 and y > 1, we can estimate Rits absolute value by A +
∞
B ∞ y y
P
n=1 y 2 +n2 . The sum can be estimated by the integral 0 y 2 +x2 dx which can be
shown to be independent of y by substituting yx for x. In conclusion, for y > 1 it is
bounded. The argument for y < −1 is similar.
 13.2 Weierstrass’ construction 87

13.2 Weierstrass’ construction

The goal of the Weierstrass construction is, given a sequence {an }n∈∞ with |an | → ∞,
to construct a holomorphic function which vanishes exactly at the an . Note that the an
are allowed to be repeated, say k times, in which case we want our function to have a
zero at an of multiplicity k.
This uses the so-called canonical factors.

Definition 13.2.1. For k ≥ 0, we define

2 /2+···+z k /k
E0 (z) := 1 − z, and Ek (z) := (1 − z)ez+z for k ≥ 1.

These are essentially 1 − z, with more and more control over the growth of 1 − Ek (z)
for |z| small:

Lemma 13.2.2. If |z| ≤ 1/2, then |1 − Ek (z)| ≤ c|z|k+1 for some constant c > 0
independent of k.

Proof. For |z| ≤ 1/2, we have a convergent power series expression

∞
X
log(1 − z) = − z n /n.
n=1

Thus we get
2 /2+···+z k /k
Ek (z) = elog(1−z)+z+z = ew
∞
X
with w = − z n /n.
n=k+1

Since |z| ≤ 1/2 we can estimate |w| by

∞
X
|w| ≤ |z|k+1 |z|n−k−1 /n.
n=k+1

Using the crude estimate |z|n−k−1 /n ≤ 1/2n−k−1 , we can estimate this as

∞
2−j = 2|z|k+1 .
X
≤ |z|k+1
j=0

Thus in particular |w| ≤ 1.

Since ew − 1 = w(1 + F (w)) with F (0) = 0. Since w 7→ |1 + F (w)| is continuous, it
attains a maximum on w with |w| ≤ 1. Thus we have that

|1 − Ek (z)| = |1 − ew | ≤ c0 |w| ≤ c|z|k+1

with c = 2c0 .

Proposition 13.2.3. Given a sequence {an }n∈∞ with |an | → ∞ as n → ∞, there exists
an entire function f such that the zeroes of f are exactly the an (counted with multiplicity).
88 Chapter 13 Weierstrass’ construction

Proof. Suppose 0 occurs m times among the an , and henceforth suppose all an are
non-zero. Then we take ∞ Y
f (z) = z m En (z/an ).
n=1

If suffices to prove that f (z) has the desired properties on the disk DR . There are two
types of canonical factors, (i) those with |an | ≤ 2R, (ii) those with |an | > 2R. We can
then write the product as
   
Y Y
z m En (z/an ) ·  En (z/an) .
|an |≤2R |an |>2R

The first term provide the correct zeroes on the disk DR : a zero of order m at the origin,
and from each canonical factor En (z/an ) a simple zero at an (if an appears r times in
the sequence, then we have r canonical factors provide a simple zero at an and hence we
get a zero of order r at an ).
If thus remains to show that second term is holomorphic and non-vanishing on DR .
If |an | > 2R and |z| < R, then |z/an | ≤ 1/2. Hence by the previous lemma we have

|1 − En (z/an )| ≤ c|z/an |n+1 ≤ c/2n+1

with c independent on n. Thus the infinite product

Y
En (z/an )
|an |>2R

converges to a holomorphic function on |z| < R. It does not vanish on DR since none of
the terms in the product does.

Theorem 13.2.4. Given a sequence {an }n∈∞ with |an | → ∞ as n → ∞, if g is an

entire function such that the zeroes of g are exactly the an (counted with multiplicity)
then g(z) = f (z)eh(z) with f as in the previous proposition and h entire.

Proof. Consider fg(z)

(z) . This is holomorphic on C \ {an | n ≥ 1}. Since the zeroes of g
cancel out of those of f , it has removable singularities only and hence extends to an
entire function. This function vanishes nowhere, so is of the form eh(z) . The conclusion
follows.
 Chapter 14

Hadamard factorization theorem

We continue our study of holomorphic functions whose domain is all of the complex
plane C. We now finish the proof of the Hadamard factorization theorem, which describes
an entire function of finite order in terms of its zeroes. This uses the Weierstrass
construction of a holomorphic function with specified zeroes. This is Section 5.5 of [SS03].

14.1 Hadamard factorization theorem

Last lecture we proved that when f has a zero of order m at the origin and further
zeroes exactly given by {an }n≥1 with an → ∞, then we have that
∞
Y
f (z) = eg(z) z k En (z/an ),
n=1

2 n
with En the canonical factor (1 − z)ez+z /2+···+z /n .
The Hadamard factorization theorem improves on this when f is of finite order.
ρ
Recall that we say f has order of growth ≤ ρ if |f (z)| ≤ AeB|z| for some constants
A, B > 0, and the order of growth of f is the infimum over all such ρ’s.

Theorem 14.1.1 (Hadamard factorization theorem). Suppose f is entire and has order
of growth ρ satisfying k ≤ ρ ≤ k + 1 with k ≥ 0 an integer. If f has a zero of order m at
the origin and further zeroes exactly given by {an }n≥1 with an → ∞, then we have that
∞
Y
f (z) = eP (z) z m Ek (z/an ),
n=1

with P a polynomial of degree ≤ k.

sin(πz)
Example 14.1.2. Recall that π has order of growth 1. The above theorem then tells
us that
sin(πz) Y
= eaz+b (1 − z/n)
π n∈Z

89
90 Chapter 14 Hadamard factorization theorem

for some constant a, b. Collecting the terms 1 − z/n and 1 + z/n, we get
∞
!
az+b
Y z2
e z 1− 2 .
n=1
n

In fact, from the product expression for the sine we know that in fact a = 0 = b. To
prove this, one evaluates the first and third derivatives at 0.
Example 14.1.3. Suppose that f is entire and of order ρ not an integer. We claim it must
have infinitely many zeroes. If not, then by Theorem 14.1.1 we have
r
Y
f (z) = eP (z) z m Ek (z/an )
n=1

2 k
with P a polynomial of degree k. Since eP (z) as well as Ek (z) = (1 − z)ez+z /2+···+z /k
are of order ≤ k, it is easy to estimate that f (z) as above must be of order ≤ k. This
contradicts the fact that it is of order ρ > k.

14.1.1 Preparatory lemmas

The main technical input of Weierstrass’ construction are estimate on the canonical
factors. Since the Hadamard factorization theorem is an improvement of this, it is not
surprising that we need to prove better estimates on the canonical factors.

Lemma 14.1.4. We have that

k+1
|Ek (z)| ≥ e−c|z| for |z| ≤ 1/2, some c ≥ 0,
0 k
|Ek (z)| ≥ |1 − z|e−c |z| for |z| ≥ 1/2, some c ≥ 0.

Proof. For the first part, recall that |z| ≤ 1/2, we can use the power series expansion for
log(1 − z) = ∞ n
n=1 −z /n to write
P

Pk P∞
z n /n − z /n
Ek (z) = elog(1−z)+ n=1 =e n=k+1 n = ew ,

for w = − ∞ n Now use that |ew | ≥ e−|w|

 , as well as the expression |w| =
P
n=k+1 z /n.

k+1  ∞ n−k−1 P∞ n−k−1
|z| . Since z 7→  n=k+1 z
P
 n=k+1 z  is a continuous function, it attains
 

a maximum c when |z| ≤ 1/2.

For the second part, we simply write
 2 /2+···+z k /k

|Ek (z)| = |1 − z| ez+z .
 

 2 /2+···+z k /k
 2 k
Now observe that ez+z  ≥ e−|z+z /2+···+z /k| and that |z + z 2 /2 + · · · + z k /k| ≤
 

c0 |z|k for some constant c0 ≥ 0 by the type of estimates we used for the proof of the
Liouville theorem.
We shall use this to estimate the infinite product of canonical factors away from some
disks:
 14.1 Hadamard factorization theorem 91

Lemma 14.1.5. For s with ρ < s < k + 1 we have

∞
 
Y  s
Ek (z/an ) ≥ e−C|z|
 

 
n=1

on the complement of a union of disks centered at the an of radius |an |−k−1 .

Proof. As in the Weierstrass construction, we write

   
∞
Y Y Y
Ek (z/an ) =  Ek (z/an ) ·  Ek (z/an )
n=1 |an |≤2|z| |an |>2|z|

and estimate the two terms in the product separately.

For the second term, we have
 
 
 Y  Y

 Ek (z/an ) = |Ek (z/an )|
|an |>2|z|  |an |>2|z|
k+1
e−c|z/an |
Y
≥
|an |>2|z|
P
−c|z|k+1 |an |−k−1
≥e |an |>2|z|

Now observe that |an |−k−1 = |an |−s |an |s−k−1 and since |an | > 2|z| (in fact |an | > |z|
would have sufficed) and s < k + 1, this can be estimate as ≤ C1 |an |−s |z|s−k−1 . Since
|an |−s converges by the assumption on the order of growth of f , we can estimate
P
nP
C 0 |an |>2|z| |an |−k−1 as ≤ C2 |z|s−k−1 . The result is that
 
 Y 
−C2 |z|s
 
E k (z/a n ≥e
) .
 

|an |>2|z| 

(This did not use that z was in the complement of the disks centered at an and radius
|an |−k−1 .)
For the first term, we have the estimate
     
 Y 
0 k
e−c |z/an |  ,
  Y Y

 Ek (z/an ) ≥  |1 − z/an | · 
|an |≤2|z|  |an |≤2|z| |an |≤2|z|

and we will investigate bothPterms in the product separately. For the latter one,
0 k |a |k
Q −c0 |z/an |k = e−c |z| |an ≤2|z| n
|an |≤2|z| e we can make an estimate as above to see
−C |z| s
it is ≥ e 3 .
The more subtle one is the former one, |an |≤2|z| |1 − z/an |. We will take σ such
Q

ρ < σ < s. When z is distance at least |an |−k−1 from an , we have that

|1 − z/an | ≥ |an |−k−2 ,

92 Chapter 14 Hadamard factorization theorem

P
−(k+2) n log |a |
and thus the product is ≥ e |an |≤2|z| . This is something we can estimate in
terms of the function n(r), counting the number of zeroes (with multiplicity) within a
circle of radius r:
X
(k + 2) log |an | ≤ (k + 2)n(2z) log |2z| ≤ C4 |z|σ log |2z| ≤ C5 |z|s .
|an |≤2|z|

s
In the end, we get |an |≤2|z| |1 − z/an | ≥ e−C5 |z| .
Q

Combing all these results, we get

∞ 
Y  s s s s
Ek (z/an ) ≥ e−C2 |z| e−C3 |z| e−C5 |z| = e−c|z|
 

 
n=1

with c = C2 + C3 + C5 , as long as z avoids the disks centered at an of radius |an |−k−1 .

Since the radii |an |−k−1 sum to a finite sum, they can’t possibly exhaust each of the
intervals [R, R + 1] of possible radii. Thus we see that:

Corollary 14.1.6. There exist a sequence r1 , r2 , . . . of radii with ri → ∞, such that

∞
 
Y  s
Ek (z/an ) ≥ e−C|z|
 

 
n=1

when |z| = ri .

14.1.2 The proof

Proof of Theorem 14.1.1. That the infinite product
∞
Y
E(z) = z m Ek (z/an )
n=1

requires nothing but the estimates of the previous section: when n is large enough
|z/an | ≤ 1/2 and we have |1 − Ek (z/an )| ≤ c|z/an |k . Since the sum ∞ n=1 |an |
−k−1
P

converges by our previous results on the zeroes of functions of order < k + 1, this infinite
product converges to an entire function. Further, it exactly has the same zeroes as f
(counted with multiplicity).
Instead, our fancy estimates come in when we try to understand the entire functon
f (z)/E(z), which is non-vanishing. Hence we can write it as f (z)/E(z) = eg(z) for some
entire function g. We must soch that g is a polynomial P of degree ≤ k. However, on
each circle of radius ri around the origin, we have
|f (z)| 0 s
|eg(z) | ≤ ≤ eC |z| when |z| = ri ,
|E(z)|
s
since f is of order ρ < s and we have |E(z)| ≥ e−C|z| . Since |eg(z) | = eRe(g(z)) , it
suffices to prove that when Re(g(z)) ≤ C 0 |z|s , then g is polynomial of degree ≤ k (recall
k ≤ ρ < s < k + 1).
 14.2 Picard’s little theorem for functions of finite order 93

This shall use a consequence of the formula for an which we used to derive the mean
value property: if h(z) = ∞ n
n=0 an z , then for n ≥ 0 we have that
P

Z 2π
1
an = h(reiθ )e−inθ dθ.
2πrn 0

When we take n > 0 to be negative in the right hand side, the integral vanishes. In
1 R 2π
particular, the complex conjugate of the case with −n gives that 2πr n 0 h(reiθ )e−inθ dθ
is 0. If n > 0, we add this to the expression in the case n, we get
an = an + 0
1 2πZ
1
Z 2π
= n
h(reiθ )e−inθ dθ + h(reiθ )e−inθ dθ
2πr 0 2πrn 0
Z π
1
= Re(h)(reiθ )e−inθ dθ.
2πrn 0
1 π 0 s −inθ
R
We may as well subtract a term 2πr n 0 C r e dθ, since this is 0 anyway: for n > 0
we get Z π
1
an = (Re(h)(reiθ ) − C 0 rs e−inθ dθ.
2πrn 0
Taking r to be one of the ri given above, we get
Z 2π Z 2π
1 0 s −inθ 1
|an | ≤ n iθ
|(Re(h)(re ) − C r e |dθ ≤ n C 0 rs − u(reiθ )dθ,
πr 0 πr 0
C0
and we compute the right hand side is π rs−n − π2 Re(a0 )rn . If n ≥ k + 1 > s, this goes
to 0 as ri → ∞. Thus we conclude that g(z) = a0 + a1 z · · · + ak z k , i.e. it is a polynomial
of degree ≤ k.
Remark 14.1.7. For n = 0 we get a different result in the above argument:
Z π
1
2Re(a0 ) = Re(h)(reiθ )dθ.
2πrn 0

14.2 Picard’s little theorem for functions of finite order

Picard’s little theorem says that if f is entire function which omits two values, then
f is constant. We shall this under the assumption that f is of finite order.
Theorem 14.2.1. If f is entire function of finite order which omits two values, then f
is constant.
Proof. If f omits a, then f (z) − a is still of finite order and never vanishes. By the
Hadamard factorization theorem we have f (z) − a = ep(z) for a polynomial p(z). From
the fundamental theorem, if p has degree ≥ 1 then it is surjective and hence ep(z) only
omits the value 0. This would contradict the assumption that f omits two values, so p
must have degree 0. We conclude that f (z) − a is constant and hence so is f .
2
Example 14.2.2. The function ez − sin(z) has finite order (in fact 2), so all points z ∈ C
except possibly two are in its image.
 Chapter 15

The Gamma function

Having finished the study of entire functions, we now move on the Riemann zeta
function and its applications to prime numbers. As preparation, we study the Γ-function;
the “simplest” function with simple poles at 0, −1, −2, . . ., which appears in the functional
equation for the Riemann zeta function. This is Section 6.1 of [SS03].

15.1 The Gamma function

When we discussed the construction of holomorphic functions by limits, such as sums

or integrals, we proved that the formula
Z ∞
Γ(s) := e−t ts−1 dt
0

defines a holomorphic function on {s ∈ C | Re(s) > 0}: the Gamma function. I

recommend your read [SS03, Proposition 6.1.1] to refresh your memory, but the strategy
is to write Z  Z Z1/ ∞
Γ(s) = e−t ts−1 dt + e−t ts−1 dt + e−t ts−1 dt,
0  1/

show that the middle term is holomorphic for all  when Re(s) > 0 and show that the
other two terms go to 0 as  → 0.
We will today extend this to C \ {0, −1, −2, . . .}, necessarily uniquely, and discuss its
most important properties. In the next section we will describe how the Gamma function
is a particular instance of a Mellin transform, applied to e−t .

15.1.1 Analytic extension

To extend Γ, we make the following observation:

Proposition 15.1.1. When Re(s) > 0, we have Γ(s + 1) = sΓ(s).

d −t s
Proof. By integrating dt (e t ) we get
Z 1/ Z 1/ Z 1/
d
e−1/ (1/)−s − e− s = (e−t ts )dt = − e−t ts dt + s e−t ts−1 dt.
 dt  

94
 15.1 The Gamma function 95

As  → 0, the left hand side goes to 0 and the right hand side to −Γ(s + 1) + sΓ(s). This
proves the formula.
Example 15.1.2. In particular, we have

Γ(n + 1) = n Γ(n) = n(n − 1) Γ(n − 1) = · · · n! Γ(1),

∞
and since Γ(1) = 0∞ e−t dt = −e−t 0 = 1, we see that Γ(n + 1) = n!. In other words,
R 

the Gamma function is a holomorphic extension of the factorial function.

Theorem 15.1.3. The function Γ(s) has a unique extension to a meromorphic function
on C, with simple poles at 0, −1, −2, . . . and ress=−n Γ(s) = (−1)n /n!.

Proof. Observe that for m ≥ 1, we have that

Γ(s + m)
(s + m − 1)(s + m − 2) · · · s

is meromorphic on {s ∈ C | Re(s) ≥ −m} with poles at 0, −1, . . . , −(m − 1), and

coincides with Γ(s) on {s ∈ C | Re(s) ≥ −m}. Furthermore, for the residue at n ∈
{0, −1, . . . , −(m − 1)} simply observe that it is given by factoring out the term 1/(s − n)
and evaluating at −n:
Γ(−n + m)
ress=−n Γ(s) =
(m − n − 1)!(−1)(−2) · · · (−n)
(m − n − 1)! (−1)n
= = .
(m − n − 1)!(−1)(−2) · · · (−n) n!
All these extensions agree on the overlaps of their domains, as these are connected
subsets of C and all of them coincide with Γ(s) as defined by integral on Re(s) > 0. We
obtain a well-defined extension as
Γ(s + m)
s 7−→ if Re(s) > −m and s 6= 0, −1, −2, . . .
(s + m − 1)(s + m − 2) · · · s

We shall denote this extension also by Γ(s).

Remark 15.1.4. On page 163 of [SS03], a different explanation for the extension is given.
We write Z 1 Z ∞
Γ(s) = e−t ts−1 dt + e−t ts−1 dt.
0 1
P∞ tn
Expanding e−t
as and integrating the terms we get that the first term is equal
n=0 n!
P∞ (−1)n
to n=0 n!(n+s) . This gives an expression
∞
(−1)n
Z ∞
e−t ts−1 dt
X
Γ(s) = +
n=0
n!(n + s) 1

at first valid for Re(s) > 0. However, by doing some estimates one can show that this
combination of an infinite sum and integral defines a meromorphic function on all of C!
96 Chapter 15 The Gamma function

15.1.2 Symmetry
Recall that we studied the function sin(πs)
π , which has a simple zero at all integers
n ∈ Z. In some sense 1/Γ(s) is “half” of the function:

Theorem 15.1.5. For all s ∈ C we have

π
Γ(s) Γ(1 − s) = .
sin(πs)

Proof. It suffices to prove that this is true for s real satisfying 0 < s < 1, as the uniqueness
for extensions of holomorphic functions then implies it is true everywhere.
For 0 < s < 1 we have that
Z ∞
Γ(1 − s) = e−u u−s du = t ∈∞
0 e
−vt
(vt)−s dv
0

by substituting v = tu. Thus we get

Z ∞
Γ(1 − s)Γ(s) = e−t ts−1 Γ(1 − s)ds
0
Z ∞  Z ∞ 
−t s−1 −vt −s
= e t t e (vt) dv dt
Z0∞ Z ∞ 0

= e−t(1+v) v −s dv dt
Z0∞ Z0∞
= e−t(1+v) v −s dt dv
0 0
Z ∞ −s
v
= dv.
0 1+v
−s (s−1)x
Now we observe that by substituting v = ex , we have 0∞ 1+v v ∞ e
R R
dv = −∞ 1+ex dx. We
evaluate this in an earlier lecture as an example of Cauchy’s theorem, and the value is
π π
sin(π(1−s)) , which is equal to sin(πs) as the sine function is odd.
√
Example 15.1.6. Take s = 1/2 to get Γ(1/2)2 = π. Hence Γ(1/2) = 2 as Γ(s) is easily
seen to be positive on the positive real line.

15.1.3 Growth
We shall next study the growth of 1/Γ(s):

Theorem 15.1.7. The function 1/Γ(s) extends to an entire function with simple zeroes
at 0, −1, −2, . . . and no other zeroes. We have that
 
 1  C2 |s| log |s|
 Γ(s)  ≤ C1 e
 

for some constants C1 , C2 > 0.

 15.1 The Gamma function 97

Proof. For the first part we write

1 sin(πs)
= Γ(1 − s)
Γ(s) π
and observe that the zeroes of the sines cancel out the poles of Γ(1 − s) at 1, 2, . . .. The
zeroes at 0, −1, −2, . . . remain.
For the growth estimate, we shall use that
∞
(−1)n
Z ∞
1 sin(πs) X sin(πs)
= + e−t t−s dt.
Γ(s) π n=0
n!(n + 1 − s) π 1
 
We first estimate the second term:  sin(πs)
π  ≤ Ce
π|s| and
 
Z ∞  Z ∞
−t −s 
e−t t|Re(s)| dt.
 

 e t dt ≤
1 1
Now pick an integer n such that |Re(s)| ≤ n ≤ |Re(s)| + 1 and observe
Z ∞ Z ∞
−t |Re(s)|
e t dt = e−t tn dt
1 1
= Γ(n) = n!
≤ nn = en log n
≤ e(|Re(s)|+1) log(|Re(s)|+1) .
Now using |Re(s)| ≤ |s|, we get that the second term can be estimated by
 sin(πs) ∞ −t −s 
 Z 
 e t dt ≤ Ceπ|s| e(|s|+1) log(|s|+1) ≤ C 0 eC20 |s| log |s|
 π  1
1
for suitable C1 , C2 > 0.
00
Thus it remains to prove that the first term to bounded by C100 eC2 |s| . There are two
1
cases: if |Im(s)| ≥ 1, then | n+1−s | ≤ 1 and we can estimate it as
∞
  
 sin(πs) X (−1)n   sin(πs) 
≤ e
  
−

 π n!(n + 1 s)  π 
n=0

and we have a further estimate | sin(πs)| ≤ Ceπ|s| so this has the desired bound.
The harder case is |Im(s)| ≤ 1. Fix an integer k such that k − 1/2 ≤ Re(s) ≤ k + 1/2.
1
If k ≤ 0, then | n+1−s | ≤ 1 again and we do the same estimate as above. If k ≥ 1, then
we split the sum as
∞
sin(πs) X (−1)n sin(πs) X (−1)n sin(πs)
= (−1)k−1 + .
π n=0
n!(n + 1 − s) (k − 1)!(k − s)π n6=k−1 n!(n + 1 − s) π
000
Both of these are bounded by C1000 eC2 |s| ; for the second term this is done similarly as
1
before by estimate n+1−s and for the first term we use that the zero of sin(πs) cancels
the pole of 1/(k − s).
Corollary 15.1.8. 1/Γ(s) is of order ≤ 1.
Proof. It suffices to observe that for any  > 0, C|z| log |z| ≤ C 0 |z|1+ . Thus 1/Γ(s) is of
order ≤ 1 +  for any  > 0.
98 Chapter 15 The Gamma function

15.1.4 Product formula

Since 1/Γ(s) is a entire function of finite order whose zeroes we know, the Hadamard
factorization theorem gives a product expression for it. In particular, its order is ≤ 1
and it has simple zeroes at 0, −1, −2, . . .. Thus we get
∞
(1 + s/n)e−s/n
Y
1/Γ(s) = eas+b s (15.1)
n=1

for constants a, b ∈ C to be determined.

 PN 
1
Proposition 15.1.9. We may take b = 0 and a = γ := limN →∞ − log(N ) + n=1 n .

That the limit in this proposition exists is not completely obvious, see page 167 of
[SS03]; γ is called the Euler–Mascheroni constant. It is a famous open conjecture that it
is irrational.
Proof. Since the residue of Γ(s) at s = 0 is 1, we have that lims→0 sΓ(s) = 1. On the
other hand, the limit as s → 0 of (15.1) is e−b so b must be a multiple of 2πi; we may
take 0.
Since Γ(1) = 1, we may evaluate (15.1) at s = 1 to get
∞
(1 + 1/n)e−1/n .
Y
ea
n=1

We can write the product as a limit of N → ∞ of the exponential of N n=1 log(1 + 1/n) −
P
PN
n=1 1/n. Writing log(1 + 1/n) as log(1 + n) − log(n), half of the terms cancel to give
log(1+N )− N
P
n=1 1/n. This is almost the correct description to give the Euler–Mascheroni
constant as N → ∞, except we have log(1 + N ) instead of log(N ). To see this doesn’t
matter, write log(1+N ) as log(N )−log(1+1/N ) and that limN →∞ log(1+1/N ) = 0.
Example 15.1.10. Recall the product expansion
∞ ∞
!
sin(πs) Y z2 Y
=z 1− 2 =z (1 − z/n) (1 + z/n) .
π n=1
n n=1

In the expression
1 sin(πs)
= ,
Γ(s)Γ(1 − s) π
1
the terms z and 1 + z/n for n ≥ 1 come from Γ(s) and the terms 1 + z/n for n ≥ 1 from
1
Γ(1−s) .

15.2 Mellin transforms

The Mellin transform of a suitable function (e.g. continuous and quickly decaying) is
Z ∞
dt
M(f )(s) = f (t)ts .
0 t
 15.2 Mellin transforms 99

Thus the Gamma function is nothing but the Mellin transform of e−x . There is a precise
sense in which the Mellin transform is a Fourier transform.
The ordinary Fourier transform of a periodic function f : R → C
Z 1
F(f )(n) = f (t)ei2πnt dt
0

relates functions on R/Z (i.e. periodic functions) to functions on Z (i.e. the Fourier
coefficients). Closely related is the Fourier transform of non-periodic functions
Z ∞
F(f )(s) = f (t)eist dt,
−∞

relations functions on R to functions on R.

To see how these are related and what role the Mellin transform plays, we give a more
general perspective. In the first case we consider the group (R/Z, +). The characters,
i.e. homomorphism λ : (R/Z, +) → U (1) := {z ∈ C | |z| = 1} are given by t 7→ ei2πnt for
some n ∈ Z. It is a result of Pontryagin that because G = (R/Z, +) is locally compact
abelian group, so is the set of Ĝ (the Pontryagin dual of G). Similarly, in the second
case we consider the group (R, +). Its characters, i.e. homomorphism λ : (R, +) → C are
given by t 7→ ei2πst for some s ∈ R.
Thus in general a Fourier transform relates functions on a locally compact abelian
group to functions on its Pontryagin dual. The transformation is given by sending
f : G → C to the function Z
Ĝ 3 λ 7−→ f (g)λ(g)dµ,
G
where µ is a G-invariant measure (unique up to scaling).
The Mellin transform is then the case G = (R>0 , ·). Just like (R, +) it is its own
Pontryagin dual, with the character corresponding to s ∈ iR given by t 7→ ts = eis log t ∈
U (1) (there is no issue about branches of the logarithm since t ∈ R+ ). The invariant
measure is dtt . In other words, the slogan is:
The Mellin transform is the Fourier transform for the multiplicative group of
positive real numbers.
 Chapter 16

The Riemann zeta function

Last lecture we extended the Gamma function to a meromorphic function on C and

discussed its properties. Today we do the same for the Riemann zeta function.1 This
serves as preparation for the eventual proof of the prime number theorem. This is Section
6.2 of [SS03].

16.1 The Riemann zeta function

When we discussed the construction of holomorphic functions by infinite sums, we

gave the Riemann zeta function as an example: for Re(s) > 1 the following sum converges
uniformly on compacts
∞
X 1
ζ(s) =
n=1
ns
and hence defines a holomorphic function. We shall extend it to a meromorphic function,
with a simple pole at s = 1. (Recall from the first lecture that having a spole at s = 1
implied that there are infinitely many primes.) The argument will relate ζ to Γ and a
new function θ.

16.1.1 The theta function

Let us consider the function

θ : R>0 −→ R
∞
2
e−πn t .
X
t 7−→
n=−∞

2
To verify it is well-defined, we estimate |e−πn t | ≤ e−π|n|t , so the sum can be estimated
 ∞  ∞
 X 
−πn2 t  1
(e−πt )n = 1 + 2
X
e ≤1+2 .

1 − e−πt

 
n=−∞ n=1
1
Since Γ is a capital Greek letter we capitalize “Gamma” when referring to the Gamma function, but
since ζ is lowercase we do not capitalize “zeta” when referring to the Riemann zeta function.

100
 16.1 The Riemann zeta function 101

This shows that the sum is uniformly convergent on the sets t > , and hence defines a
continuous function. A more refined estimate shows it is smooth, but we will not use
this.

Proposition 16.1.1 (Functional equation for θ). For t > 0 we have

√
θ(t) = 1/ t θ(1/t).
2
Proof. Consider the holomorphic function f (z) = e−πtz . Within the rectangular contour
RN with corners N + 1/2 − i, N + 1/2 + i, −N − 1/2 + i, −N − 1/2 − i the function
f (z)/(e2πiz − 1) has poles at −N, −(N − 1), · · · , N − 1, N with residues f2πi
(n)
. By the
1
residue theorem and canceling a 2πi , we get

f (z)
Z X
dz = f (n).
RN e2πiz −1 |n|≤N

2
The integrals over the vertical line segments can be estimated as ≤ Ce−πtN so go to 0
as N → ∞. We thus get that
Z ∞ Z ∞
X f (x − i) f (x + i)
f (n) = 2πi(x−i)
dx − dx.
n∈Z −∞ e −1 −∞ e2πi(x+i)−1

Now observe that |e2πi(x−i) | > 1 and |e2πi(x+i) | < 1 so we can write
∞ ∞
1
= e−2πi(x−i) e−2πni(x−i) = e−2πni(x−i) ,
X X
e2πi(x−i) − 1 n=0 n=1

∞
1 X
=− e2πni(x+i) .
e2πi(x+i) − 1 n=0

Substituting these and exchanging sums and integrals (justified by the absolute
convergence of either), we get
∞ Z ∞ ∞ Z ∞
f (x − i)e−2πni(x−i) dx.
X X X
f (n) = f (x + i)e2πni(x+i) dx +
n∈Z n=0 −∞ n=1 −∞

R∞ −πx2 e−2πixξ dx 2
Now we recall an integral we did: −∞ e = e−πξ . Investigating a term
in the first sum, we get
Z ∞ Z ∞
2 +2πni(x+i)
f (x + i)e2πni(x+i) dx = e−πt(x+i) dx.
−∞ −∞
√
Let us substitute y = t(x + i) to get this is equal to

1
Z ∞
2
√ 1 √ 2 1 2
√ e−πy e2πin/ ty
dy = √ e−(n/ t) = √ e−n /t .
t −∞ t t
102 Chapter 16 The Riemann zeta function

Similarly, we obtain an expression

Z ∞
1 2
f (x − i)e−2πni(x−i) dx = √ e−n /t .
−∞ t

The conclusion is that

2 X 1 2 /t 1
e−πn t = √ e−n
X
θ(t) = = √ θ(1/t).
n∈Z n∈Z
t t

Remark 16.1.2. If we have done Chapter 4 of [SS03], this would have been an easily
consequence of the Poisson summation formula in Fourier theory.
We will also need two estimates:
(i) |θ(t) − 1| ≤ Ce−πt as t → ∞ because
∞ ∞
−πt −π(n2 −1)t −πt
e−πnt
X X
|θ(t) − 1| ≤ e 2 e ≤e
n=1 n=0

and the latter sum can be bounded by a constant for t ≥ 1.

(ii) Combining this with the functional equation we get that as t → 0,
√
|θ(t)| ≤ C/ t.

16.1.2 Expressing ζ in terms of θ

Proposition 16.1.3. for Re(s) > 1 we have
Z ∞
−s/2 1
π Γ(s/2)ζ(s) = us/2−1 (θ(u) − 1)du.
2 0

Proof. We start on the right hand side and observe that

∞
θ(u) − 1 2
e−πn u .
X
=
2 n=1

The estimates given above justify that we can exchange the sum and the integral to get
Z ∞ ∞ Z ∞
1 1 2
us/2−1 e−πn u du.
X
s/2−1
= u (θ(u) − 1)du = =
2 0 n=1
2 0

t
If we substitute u = πn2
in the terms, we get
Z ∞
1 1
ts/2 e−t dt = π −s/2 Γ(s/2) .
(πn )−s/2
2
0 ns

Summing this over n ≥ 1, we get exactly π −s/2 Γ(s/2)ζ(s).

 16.1 The Riemann zeta function 103

It is a good idea to consider π −s/2 Γ(s/2)ζ(s) instead of Γ(s), and we give it a name
ξ(s) := π −s/2 Γ(s/2)ζ(s),
at the moment defined for Re(s) > 1.2 The following is not only the basic input for
the extension of ζ, but expresses a symmetry property similar to that of the Gamma
function:
Theorem 16.1.4. The function ξ extends to a meromorphic function with poles at
s = 0, 1 and satisfies
ξ(s) = ξ(1 − s)
for all s ∈ C.
Proof. As the above proposition tells us, the function ψ(u) := (θ(u) − 1)/2 is closely
related to the xi function. The functional equation θ(u) = u−1/2 θ(1/u) gives one for ψ:
1 1
ψ(u) = u−1/2 ψ(1/u) + 1/2
− .
2u 2
We use this to rewrite ξ(s) for Re(s) > 1, using the trick of splitting the integral into
the sections [0, 1] and [1, ∞), familiar from the Gamma function:
Z ∞
1
ξ(s) = us/2−1 (θ(u) − 1)du
2 0
Z ∞
= us/2−1 ψ(u)du
0
Z 1 Z ∞
s/2−1
= u ψ(u)du + us/2−1 ψ(u)du.
0 1

Now we use the functional equation in the first term, so it is:

Z 1
1 1
 
s/2−1 −1/2
u u ψ(1/u) + 1/2 − du
0 2u 2
us/2−1/2 us/2−1
Z 1
= us/2−3/2 ψ(1/u) + − du
0 2 2
Z ∞
1 1
= − + v −s/2−1/2 ψ(v)dv
s−1 s 1

by substituting v = 1/u.
The end result is that
Z ∞
1 1 
ξ(s) = − + us/2−1 + u−s/2−1/2 ψ(u)du. (16.1)
s−1 1−s 1

We claim that the integral defines an entire function. To see this is case, we first observe
that
[1, R] × C −→ C
(s, u) 7−→ (us/2−1 + u−s/2−1/2 )ψ(u)
2
See https://terrytao.wordpress.com/2008/07/27/tates-proof-of-the-functional-equation/
for a high-level explanation why ξ is better than ζ.
104 Chapter 16 The Riemann zeta function

 
is continuous and holomorphic when we fix s. Thus s 7→ 1R us/2−1 + u−s/2−1/2 ψ(u)du
R

defines a holomorphic function. Now we let R → ∞ and need to check uniform convergence
on compacts to see that the result is still holomorphic. But if |s| ≤ D then
Z ∞    Z ∞ 
−s/2−1/2
s/2−1
u|s|/2+1 + u|s|/2+1/2 Ce−πu du
 

 u +u ψ(u)du ≤
R
ZR∞
≤ C 0 uD+1 e−πu du < ∞
R

since e−πu outgrows any polynomial.

Therefore (16.1) gives a meromorphic extension of ξ to C, with evident simple poles
at s = 0, 1. The functional equation is easily read off from this, especially when we write
us/2−1 + u−s/2−1/2 = us/2−1 + u(1−s)/2−1 .

Corollary 16.1.5. The function ζ extends to a meromorphic function with a simple

pole at s = 1.

Proof. Writing ξ for the extension given above, we can define an extension of ζ as

π s/2
ζ(s) = ξ(s),
Γ(s/2)

recalling that 1/Γ is an entire function. Furthermore, 1/Γ(s/2) has simple zeroes at
0, −1, −2, · · · , so one of the two poles of ξ(s) cancels out and only the one at s = 1
remains.

16.1.3 Estimates on the line Re(s) = 1

We now give a different argument which extends ζ to Re(ζ) > 0. Its purpose is to
give an estimate on the line Re(s) = 1, which will be used to prove some results about
primes later.
|s|
Proposition 16.1.6. There are entire function δn (s) such that (i) |δn (s) ≤ nRe(s)+1
, (ii)
we have
N Z N N
X 1 1 X
− dx = δn (s)
n=1
ns 1 x
s
n=1

for Re(s) > 0.

Proof. To make (ii), we need to define

Z n+1
1 1
δn (s) := − dx.
n ns xs
This is entire by our usual results about defining holomorphic functions by integrals. To
get the estimate, we start with
   
1 1   1 1 
 ns − xs  ≤  nRe(s) − xRe(s) 

 16.1 The Riemann zeta function 105

we note that by the mean-value theorem for f (x) = x−Re(s) , we have

x =−Re(s) = f (x) = f (n) + (x − n)f 0 (c) = n−Re(s) − (x − n)Re(s)c−Re(s)−1 .
for some c ∈ [n, x]. Reorganizing and estimating |x−n| ≤ 1, |Re(s)| ≤ |s| and c−Re(s)−1 ≤
n−Re(s)−1 , we get
|s|
 
1 1 
 ns − xs  ≤ nRe(s)+1


for n ≤ x ≤ n + 1.
Corollary 16.1.7. For Re(s) > 0 we have
∞
1 X
ζ(s) − = δn (s)
s − 1 n=1
with right hand side holomorphic.
Proof. The estimate on the δn (s) proves that the sums converges uniformly for Re(s) ≤ 
for any  > 0, so it remains to observe that the equation follows by letting N → ∞ in (ii)
and evaluating the integral.
We can now give our estimate on ζ(s) near the line Re(s) = 1.
Lemma 16.1.8. Write s = σ + iτ with σ, τ ∈ R. Then for all σ0 ∈ [0, 1] and  ∈ (0, 1)
there exists constants c , c0 > 0 such that
· |ζ(s)| ≤ c |τ |1−σ0 + for σ ≥ σ0 and |τ | ≥ 1 − ,
· |ζ 0 (s)| ≤ c0 |τ | if σ ≥ 1 and |τ | ≥ 1.
2 |s|
Proof. We start with the observation that |δn (s)| ≤ nσ in addition to |δn (s)| ≤ ≤nσ+1
.
Thus we have for ρ ≥ 0 to be fixed later that
ρ  1−ρ
|s| 2

|δn (s)| = |δn (s)|ρ |δn (s)|1−ρ = .
nσ+1 nσ
2|s|ρ
Estimate from above by σ0 replacing σ and simplifying, we get this is ≤ nσ0 +ρ
.
1
Now we take ρ = 1 − σ0 +  and use ζ(s) = s−1 + ∞
P
n=1 δn (s) to get

 1 
 ∞
 + 2|s|1−σ0 +
X 1
|ζ(s)| ≤ 

1+
.
1−s 
n=1
n
 
 1 
Since the sum converges and  1−s  is bounded when |τ | ≥ 1 − , part (i) follows.
By adjusting the constant to take care of the compact part where |τ | ≤ 2, we may
assume that |τ | ≥ 2. Part (ii) follows from part (i) using the Cauchy integral formula
Z 2π
0 1
ζ (s) = ζ(s + reiθ )dθ
2πi 0
where r is a circle of radius /2. This doesn’t contain a pole as  ∈ (0, 1). We may apply
(i) with σ0 = 1 − /2 to get
Z 2π
0 1
|ζ (s)| ≤ c/2 (|τ | − /2) dθ ≤ c0 |τ | .
2π 0
 Chapter 17

The zeroes of the Riemann zeta function

Last lectures we extended the Gamma function and Riemann zeta function to mero-
morphic functions on C and we discussed their basic properties. Today we start using
this to eventually prove the prime number theorem. This is Section 7.1 of [SS03].

17.1 Recollection on the prime number theorem

One of the highlights of this course is the application of complex analysis to the study
of the prime-counting function

π : R≥0 −→ R
x 7−→ #{primes less than or equal to x}.

In particular, we will leverage the fact that ζ(s) does not vanish on the line Re(s) = 1 to
prove the following result:

Theorem 17.1.1 (Prime number theorem). We have that

x
π(x) ∼ ,
log(x)
f (x)
where f (x) ∼ g(x) means that limx→∞ g(x) = 1.

It’ll take several lectures to prove this, and today we’ll only prove the input on the
zeroes of the Riemann zeta function on the line Re(s) = 1. Let us start with making
precise the relationship between the primes and the Riemann zeta function.

17.1.1 The Riemann zeta function as a product

We start with a fact that may seems obvious, but requires a bit of estimates to make
precise.

Proposition 17.1.2. For Re(s) > 1 we have that

Y 1
ζ(s) = .
primes p
1 − p−s

106
 17.1 Recollection on the prime number theorem 107

1.095

1.090

1.085

1.080

1.075

1.070

1.065

2 × 10 6 4 × 10 6 6 × 10 6 8 × 10 6 1 × 10 7

Figure 17.1 The quotient π(x)/(x/ log(x)) for x ≤ 107 . Note the convergence is not very fast.

Proof. We first observe that the right hand side indeed defines a holomorphic function.
We will use the criterion that ∞
Q
n=1 fn (z) of holomorphic
P∞
functions gives a holomorphic
function if there exist constants cn > 0 such that n=1 cn < ∞ and |fn (z) − 1| ≤ cn .
To check this estimate on Re(s) ≥ 1 + , we start noting that in this case we use the
geometric series expansion
∞
1 X 1
−s
= .
1−p n=1
pns
So we can make the estimates
∞ ∞
p−(1+)
p−nRe(s) ≤ p−(1+)n = ≤ Cp−(1+) ,
X X
|fp (z) − 1| ≤
n=2 n=2
1 − p−(1+)
1
with C = 1−2−(1+) . The sum over p can then be bounded by the sum over all n, which
indeed converges.
Now that we know that both the left and right hand side are holomorphic, it suffices
to prove the identity holds when s is real and s > 1.
Let us now deduce the equality from a pair of estimates. Since every number ≤ N
can be written uniquely as a product of primes ≤ N , we have
N
1 Y  1 1
X 
≤ 1+ + · · · + Ns
n=1
ns p≤N
ps p
Y 1
≤
p≤N
1 − p−s
Y 1
≤ .
p 1 − p−s
P∞
Letting N → ∞, we get that n=1 n1s ≤ p 1−p1 −s .
Q

Hence it suffices to establish the reverse inequality. This follows by letting N → ∞

in the similar estimate
∞
Y  1 1 1
 X
1 + s + · · · + Ns ≤ .
p≤N
p p n=1
ns
108 Chapter 17 The zeroes of the Riemann zeta function

Recall that our convergence criteria for infinite products also says it vanishes if and
only if one of the terms vanish; thus ζ(s) 6= 0 when Re(s) > 1.

17.2 The trivial zeroes of Riemann zeta function

To get the first properties of ζ(s), and in particular finding its zeroes on the real line,
we use the functional equation for ξ(s) = π −s/2 Γ(s/2)ζ(s). This was ξ(s) = ξ(1 − s) and
hence expands to
π −s/2 Γ(s/2)ζ(s) = π −(1−s)/2 Γ( 1−s
2 )ζ(1 − s),

so that
Γ( 1−s
2 )
ζ(s) = π s−1/2 ζ(1 − s).
Γ(s/2)
From this we can deduce the following:

Theorem 17.2.1. The only zeroes of ζ(s) outside the critical strip 0 ≤ Re(s) ≤ 1 are at
−2, −4, −6, · · · .

Proof. It suffices to only consider the subset Re(s) ≤ 0. Then neither π s−1/2 and ζ(1 − s)
vanishes. The enumerator Γ( 1−s 2 ) has neither zeroes nor poles there, and the denominator
Γ(s/2) has poles at s = −2, −4, −6, . . .. This implies the result.

The zeroes given in the above theorem are called the trivial zeroes. There are many
zeroes known in the critical strip, but all of them lie on the line Re(s) = 1/2. It is an
open problem whether this is always true:

Conjecture 17.2.2 (Riemann hypothesis). All non-trivial zeroes of the Riemann zeta
function lie on the line Re(s) = 1/2.

17.3 The Riemann zeta function has no zeroes on Re(s) = 1

We shall now rule out the existence of zeroes on the right edge of the critical strip;
by the functional equation this also rules out zeroes on the left edge.

Theorem 17.3.1. The Riemann zeta function has no zeroes on the line Re(s) = 1.

Of course, since ζ(s) has a pole at s = 1 there are no zeroes near 1. This is thus
really about 1 + iτ with |τ | large.

Lemma 17.3.2. If Re(s) > 1 then

X p−ms ∞
cn n−s
X
log(ζ(s)) = =
p,m m n=1

with cn = 1/m if n = pm and 0 otherwise.

 17.3 The Riemann zeta function has no zeroes on Re(s) = 1 109

Proof. Since ζ(s) has no zeroes on the simply-connected domain Re(s) > 1, log(ζ(s)) is
a well-defined holomorphic function. We leave it to the reader to prove that the middle
and right sums also define holomorphic functions on Re(s) > 1. By analytic continuation,
it then again suffices to prove the identities for s real.
Both are easy: for the first we write
 
1 
− log(1 − p−s ),
Y X
log(ζ(s)) = log  −s
=
primes p
1 − p primes p
P∞ m /m.
and now use that for |x| ≤ 1/2 we have that − log(1 − x) = m=1 x The second is
tautologically true, by the definition of the coefficients cn .
Lemma 17.3.3. If σ, τ ∈ R with σ > 1, then
log |ζ 3 (σ)ζ 4 (σ + iτ )ζ(σ + 2iτ )| ≥ 0.
Proof. We start with the observation that
3 + 4 cos(θ) + cos(2θ) = 2(1 + cos(θ))2 ,
so in particular the left hand side is non-negative.
Now we write
log |ζ 3 (σ)ζ 4 (σ + iτ )ζ(σ + 2iτ )|
= 3 log |ζ(σ)| + 4 log |ζ(σ + iτ )| + log |ζ(σ + 2iτ )|
= 3Re(log(ζ(σ))) + 4Re(log(ζ(σ + iτ ))) + Re(log(ζ(σ + 2iτ )))
∞
cn n−σ (3 + 4 cos(t log n) + cos(2(t log n)))
X
=
n=1
≥ 0.
Proof of Theorem 17.3.1. This is a proof by contradiction; we assume that ζ(1 + iτ ) = 0.
We may assume τ 6= 0 since there is a pole at s = 1.
Since the zero must be of order at least 1, we get
|ζ(σ + iτ )|4 ≤ C(σ − 1)4 as σ → 1.
Since the pole at s = 1 is simple, so
|ζ(σ)|3 ≤ C 0 (σ − 1)−3 as σ → 1.
Since ζ is holomorphic at σ + 2iτ , we have that
|ζ(σ + 2iτ )| ≤ C 00 as σ → 1.
Inserting this in the previous lemma, we get that
|ζ 3 (σ)ζ 4 (σ + iτ )ζ(σ + 2iτ )| → 0 as σ → 1.
This means that its logarithm must become negative, contradicting the previous lemma.
110 Chapter 17 The zeroes of the Riemann zeta function

17.3.1 Some additional estimates

In fact, we will need a quantative version of Theorem 17.3.1, bounding 1/|ζ(s)| from
above when s = σ + iτ with σ ≥ 1 and |τ | ≥ 1.

Proposition 17.3.4. For  > 0 we have 1/|ζ(s)| ≤ c0 |τ | when s = σ + iτ with σ ≥ 1

and |τ | ≥ 1.

Proof. By the lemma, we have that

|ζ 3 (σ)ζ 4 (σ + iτ )ζ(σ + 2iτ )| ≥ 1.

So we have that
1 1
|ζ 4 (σ + iτ )| = .
|ζ 3 (σ)| |ζ(σ + 2iτ )|
Now use that ζ has a simple at s = 1, and that last lecture we prove the estimate
|ζ(σ + iτ )| ≤ c |τ | . Thus we can estimate the right hand side for σ > 1 as

≥ C(σ − 1)−3 |τ |− .

Taking fourth roots we get

|ζ(σ + iτ )| ≥ C 0 (σ − 1)−3/4 |τ |−/4 . (17.1)

By the mean value theorem and the estimate |ζ 0 (σ + iτ )| ≤ c0 |τ | from the last lecture,
we have
|ζ(σ 0 + iτ ) − ζ(σ + iτ )| ≤ C 00 |σ 0 − σ||τ | .
Let us set A = (C 0 /(2C 00 ))4 .
Now there are two cases:
(i) The first case is σ − 1 ≥ A|τ |−5 we have from (17.1) we get

|ζ(σ + iτ )| ≥ A0 |τ |−4 ,

which upon replacing 4 by  given by desired estimate.

(ii) The second case is σ − 1 ≤ A|τ |−5 . We take σ 0 − 1 = A|τ |−5 with the triangle
inequality
|ζ(σ + iτ )| ≥ |ζ(σ 0 + iτ )| − |ζ(σ + iτ ) − ζ(σ 0 + iτ )|.
Above we estimate the second right hand term by C 00 |σ 0 − σ||τ | , which it ≤
C 00 |σ 0 − 1||τ | .
Applying (17.1) for σ 0 we get

|ζ(σ + iτ )| ≥ C 0 (σ 0 − 1)3/4 |τ |−/4 − C 00 |σ 0 − 1||τ | .

Now A has been chosen such that C 0 (σ 0 − 1)3/4 |τ |−/4 = 2C 00 |σ 0 − 1||τ | , and
we get that the right hand side is C 00 |σ 0 − 1||t| ≥ A00 |τ |−4 using the fact that
σ 0 − 1 = A|τ |−5 .
 Chapter 18

Tchebychev’s functions

Having proven that ζ(s) has no zeroes when Re(s) ≥, we will use this to prove the
prime number theorem. This is Section 7.2 of [SS03].

18.1 Tchebychev’s functions

To relate the prime counting function

π : R≥0 −→ R
x 7−→ #{primes ≤ x}
to the Riemann zeta function we will use two functions ψ and ψ1 introduced by Tcheby-
chev.
Definition 18.1.1. We define the Tchebychev ψ-function by
ψ : R≥0 −→ R
X
x 7−→ log p.
pm ≤x

It is easy to give alternatively expressions for ψ(x), both of which we shall use later.
Firstly, we can write it in terms of the von Mangoldt Λ-function
Λ : N>0 −→ R
(
log(p) if n = pm ,
n 7−→
0 otherwise,
as follows: X
ψ(x) = Λ(n).
1≤n≤x

Secondly, we can write it terms of the floor function buc as follows:

X  log(x) 
ψ(x) = log(p).
p≤x
log(p)

As sounds plausible, the asymptotics of ψ are related to those of π.

111
112 Chapter 18 Tchebychev’s functions

Proposition 18.1.2. If ψ(x) ∼ x, then π(x) ∼ x/ log(x).

Proof. The strategy is to bound π(x) log(x)/x as x → ∞ by 1 from above and below.
Let us first give the lower bound:
X  log(x)  X log(x)
ψ(x) = log(p) ≤ log(p) = π(x) log(x),
p≤x
log(p) p≤x
log(p)

so we have that ψ(x)/x ≤ π(x) log(x)/x. As x → ∞ the left hand side goes to 1, so we
get 1 ≤ lim inf x→∞ π(x) log(x)/x.
For the upper bound, we fix an 0 < α < 1 (with the intention of letting it go to 1)
and write
X  log(x)  X X
ψ(x) = log(p) ≥ log(p) ≥ log(p)
p≤x
log(p) p≤x xα ≤p≤x
X
≥ log(xα ) = (π(x) − π(xα )) log(xα ).
xα ≤p≤x

From this we get that

ψ(x) + απ(xα ) log(x) απ(x) log(x)
≥ .
x x
Since xα ≥ π(xα ) and αxα−1 log(x) → 0 as x → ∞, the left hand side goes to 1 as
x → ∞. We conclude that 1 ≥ lim supx→∞ απ(x) log(x)/x. Letting α → 1, we get the
desired upper bound.
It will turn out to be more convenient to study an integrated version of ψ:

ψ1 : R≥1 −→ R
Z x
x 7−→ ψ(t)dt.
1

Observe we may as well have had the integral start at 0 instead of 1, as ψ(t) = 0 for
t < 1.

Proposition 18.1.3. If ψ1 (x) ∼ x2 /2, then ψ(x) ∼ x.

Proof. As ψ(x) is increasing, we have that for 0 < α < 1 < β (for the sake of convenience,
we will also assume αx ≥ 1) we can compare average values on the intervals [αx, x] and
[x, βx] to the value at x:
Z x Z βx
1 1
ψ(t)dt ≤ ψ(x) ≤ ψ(t)dt.
(1 − α)x αx (β − 1)x x

1
We can write the left hand side as (1−α)x (ψ1 (x) − ψ1 (αx)). Thus we get
!
1 ψ1 (x) α2 ψ1 (αx) ψ(x)
− ≤ .
(1 − α) x2 (αx)2 x
 18.2 Relating Tchebychev’s functions to ζ 113

1
Letting x → ∞ the left hand side goes to 1−α (1/2 − α2 /2) = (1 + α)/2, so we get
(1 + α)/2 ≤ lim inf x→∞ ψ(x)/x. Letting α → 1 we get the lower bound of 1 for the lim inf.
A similar argument with the right hand side gives an upper bound lim supx→∞ ψ(x)/x ≤
1.
Remark 18.1.4. In fact, the following are equivalent:
· π(x) ∼ x/ log(x),
· ψ(x) ∼ x,
· ψ1 (x) ∼ x2 /2.
Remark 18.1.5. Another way to think of f (x) ∼ g(x) is that there is a function c(x) such
that f (x) = c(x)g(x) and c(x) → 1 as x → ∞. Let’s apply this to extract an interesting
consequence from π(x) ∼ x/ log(x): taking logarithms we get

log(π(x)) = log(x) − log log(x) + log(c(x)).

Dividing by log(x), we see that

log(π(x)) log log(x) log(c(x))

=1− + ,
log(x) log(x) log(x)

and the right hand side goes to 1 as x → ∞, as log log(x) grows much slower than log(x)
and log(c(x)) already goes to 0. We conclude that log(π(x)) ∼ log(x).
This means that we can deduce from π(x) ∼ x/ log(x) that π(x) log(π(x)) ∼ x.
Substituting the nth prime pn for x, we get n log(n) ∼ pn . In other words, we deduce an
asymptotic for the nth prime number.

18.2 Relating Tchebychev’s functions to ζ

The von Mangoldt Λ-function will provide the link between ψ and ψ1 on the one
hand, and ζ on the other hand. Recall that for Re(s) > 1 we have

X p−ms X e−m log(p)s

log(ζ(s)) = = ,
m,p m m,p m

so taking the derivative we get

∞
ζ 0 (s) X
−ms
X Λ(n)
=− log(p)p =− .
ζ(s) m,p n=1
ns

We shall use this expression to prove that ψ1 (x) can be computed by integrals of a
function involving ζ 0 (s)/ζ(s) over vertical lines in the complex plane.

Proposition 18.2.1. For c > 1 we have

xs+1 ζ 0 (s)
Z c+i∞
1
ψ1 (x) = − ds.
2πi c−i∞ s(s + 1) ζ(s)
114 Chapter 18 Tchebychev’s functions

We first compute a simpler contour integral:

Lemma 18.2.2. For c > 1 we have

(
as
Z c+i∞
1 0 if a ∈ (0, 1),
ds =
2πi c−i∞ s(s + 1) 1 − 1/a if a ∈ [1, ∞).

1 R∞ 1
Proof. The integral converges because we can estimate its absolute value as ≤ C 2π −∞ (t+1)2 dt,
1 C 0
using |as | = ac is independent of s and an estimate | s(s+1) | ≤ (t+1)2 when s lies on the

line Re(s) = c with c > 1.

There are two cases: the first is a ∈ [1, ∞), in which case β = log α ≥ 0. Then we
write
esβ
f (s) = ,
s(s + 1)
which has a pole at s = 0 with residue 1 and at s = −1 with residue −e−β = −1/α. We
then take a contour given by a semi-circle S(T ) with vertical line segment [c − iT, c + iT ]
and a semi-circle on the left encircling both poles (at least if T is large enough). The
residue theorem gives
1
Z
f (s)ds = 1 − 1/a.
2πi S(T )
The result then follows when we prove that the integral of the semi-circle goes to 0 as
T → ∞. This follows from the estimate that |s(s + 1)| ≥ C 0 T 2 when T is large and s is
on this semi-circle, and |eβs | ≤ eβc as Re(s) ≤ c. Thus we can estimate

eβc
Z 
 
 f (s)ds ≤ πT → 0 as T → ∞.

semicircle C 0T 2

For a ∈ (0, 1) we take a similar contour but with semicircle to the right. This will
contain no poles, so it again suffices to prove that the integral over the semi-circle goes
to 0 as T → ∞. This uses that now β < 0 and hence |eβs | ≤ eβc as Re(s) ≥ c.
0
Proof of Proposition 18.2.1. Recalling that ζζ(s)
(s)
=− ∞ Λ(n)
P
n=1 ns , let us interchange the
sum and integral in
xs+1 ζ 0 (s)
Z c+i∞
1
− ds
2πi c−i∞ s(s + 1) ζ(s)
to get
∞
(x/n)s
Z c+i∞
X x
Λ(n) ds.
n=1
2πi c−∞ s(s + 1)

(The interchange of sum and integral is justified by e.g. dominated convergence, and
some estimates.) We can use the above lemma to identify this with
∞
X X ∞ X
X
x Λ(n)(1 − n/x) = Λ(n)(x − n).
n=1 x/n≥1 n=1 x/n≥1
 18.3 Proving the asymptotics of ψ1 115

We will now identify this with ψ1 (x). To do so, recall that

∞
X
ψ(t) = Λ(n)fn (t)
n=1

with fn (t) taking the value 1 on [0, n] and 0 otherwise. Thus we have that
Z x
ψ1 (x) = ψ(t)dt
0
∞
X Z x
= Λ(n)fn (t)dt
n=1 0
X∞
= Λ(n)(x − n).
n=1

18.3 Proving the asymptotics of ψ1

Let’s regroup: our goal is to prove that ψ1 (x) ∼ x2 /2, which implies π(x) ∼ x/ log(x),
using the following two facts: (I) for c > 1 we have

xs+1 ζ 0 (s)
Z c+i∞
1
ψ1 (x) = − ds,
2πi c−i∞ s(s + 1) ζ(s)

and (II) ζ(s) has no zeroes on the line Re(s) = 1 (or rather the estimates we gave on
ζ 0 (s) and 1/ζ(s) near that line).
Of course, the strategy will be to let c → 1. The main difficulty is the pole of ζ(s)
at s = 1, which will give the term x2 /2. The estimates will serve to rule out other
contributions. We will do so in the next lecture.
 Chapter 19

The proof of the prime number theorem

Having related ζ to ψ1 , we will now use estimate on the Riemann zeta function near
the line Re(s) = 1 to prove the prime number theorem. This is the end of Section 7.2 of
[SS03]. We then discuss some stronger results relating the zeroes of the Riemann zeta
function to the distribution of the prime numbers.

19.1 The asymptotics of ψ1 (x)

Recall that we want to prove π(x) ∼ x/ log(x), which we showed is implied by the
estimate ψ1 (x) ∼ x2 /2, with ψ1 (x) given by

xs+1 ζ 0 (s)
Z c+i∞
1
ψ1 (x) = − ds.
2πi c−i∞ s(s + 1) ζ(s)
To do so we will use the following estimate, a consequence of estimates in the previous
three lectures:
Lemma 19.1.1. For fixed η > 0, there is a constant A > 0 such that for s = σ + iτ with
σ ≥ 1 and |τ | ≥ 1 we have  0 
 ζ (s)  η
 ζ(s)  ≤ A|τ | .
 

Proof. The estimate we proved earlier were that, with same conditions on s, for any
 > 0 we can find constants A0 , A00 such that:
· |ζ 0 (s)| ≤ A0 |τ | ,
· |1/ζ(s)| ≤ A00 |τ | .
Taking  = η/2 and A = A0 A00 gives the result.
The strategy is to let c → 1 in a smart way.
Theorem 19.1.2. We have that ψ1 (x) ∼ x2 /2.
Proof. Let us introduce the notation
xs+1 ζ 0 (s)
F (s) = − .
s(s + 1) ζ(s)

116
 19.1 The asymptotics of ψ1 (x) 117

The above estimate for |ζ 0 (s)/ζ(s)| implies the estimate |F (s)| ≤ C|τ |η−2 .
We first claim that the integral over the line Re(s) = c equals that of the contour
γ(T ) with T large to be picked later given by:

1 + iT c + iT

s = 1•

To see this, we observe that F (s) has no poles in the rectangle with corners (1+iT, c+iT 0 ),
T 0  T . Hence by Cauchy’s theorem the integral over the boundary of that rectangle
vanishes. To get the result we want, we need to prove that the integral over the top
edge vanishes at T 0 → ∞. This is justified by the above estimate, taking η = 1/2 say.
A similar argument works for the rectangle with corners (1 − iT, c − iT 0 ), T 0  T . We
conclude that
1 1
Z Z
ψ1 (x) = F (s)ds = F (s)ds.
2πi Re(s)=1 2πi γ(T )
We now pass to the integral over the contour γ(T, δ) given by

1 − δ + iT 1 + iT

•s = 1

γ1

Here δ > 0 is chosen small enough so that the rectangle with corners (1 − δ − iT, 1 + iT )
contains no zeroes of ζ. This is possible as there are no zeroes of ζ on the line Re(s) = 1.
Since F (s) has a simple pole at s = 1 with residue −x2 /2, the residue theorem tells
us that
1 1
Z Z
2
F (s)ds − x /2 = F (s)ds.
2πi γ(T ) 2πi γ(T,δ)
118 Chapter 19 The proof of the prime number theorem

We now decompose the left hand side into integrals over the five straight line segments,
γ1 , . . . , γ5 (consecutively ordered, starting at the bottom), and estimate each of these.
· Firstly, by adjusting T we will make
Z  Z 
F (s)ds ≤ x2 /2 F (s)ds ≤ x2 /2.
   
 
 
γ1 γ5

We will give the argument in the case of γ5 only, as the case of γ1 is identical.
Observe that |xs | = x2 here, so that by the estimate F (s) ≤ C|τ |−3/2 we get
Z  Z ∞
1
F (s)ds ≤ C 0 x2 dt ≤ C 00 x2 T −1/2 ,
 


γ1 T t3/2

so letting T be large enough we get the desired estimate. (At this point we also fix
our δ > 0.)
· On γ3 we have that |x1+s | = x2−δ so that we get a crude estimate (using the fact
that remaining terms in F (s) are bounded):
Z 
F (s)ds ≤ C 000 x2−δ .
 


γ3

· On γ4 (as before, γ2 will be similar), we again use that all terms except x1+s are
bounded because they are continuous to get
Z  Z 1
F (s)ds ≤ C 0000 x1+σ dσ
 


γ4 1−δ
= C (x / log(x) − (x2−δ / log(x)) ≤ C 0000 x2 / log(x).
0000 2

In the end we get an estimate

 
   1 Z 
2 
ψ1 (x) − x /2 =  F (s)ds ≤ x2 /2 + C 000 x2−δ + C 0000 x2 / log(x).
  
 2πi γ(T,δ) 

Dividing by x2 /2, this implies that

 
2ψ1 (x)/x2 − 1 ≤  + 2C 000 x−δ + 2C 0000 1/ log(x)
 

So as x → ∞, we can make lim supx→∞ 2ψ1 (x)/x2 − 1 arbitrary small. This completes
 

the proof that ψ1 (x) ∼ x2 /2.

19.2 Zeroes of the Riemann zeta function and error terms for the prime
number theorem

The following is much stronger estimate on ψ1 (x) (which we will not prove):

x2 X xρ+1
ψ1 (x) = − − E(x),
2 ρ ρ(ρ + 1)
 19.2 Zeroes of the Riemann zeta function and error terms for the prime number theorem
119

where ρ runs over the zeroes of the Riemann zeta function in the critical strip, and
x1−2k
E(x) = c1 x + c0 + ∞
P
k=1 2k(2k−1) is an essentially linear term collecting the trivial zeroes.
Observe that |xρ+1 | = xRe(ρ)+1 . That there are no zeroes on the lines Re(s) = 1 thus
says that none of the terms has grows as quickly as x2 /2. Under the Riemann hypothesis,
Re(s) = 1/2 so we get even better error terms. There is an issue that maybe the infinitely
many terms, each not growing as quickly as x2 /2, could add up something that does. To
understand whether this can happen, let us make a further observation on the Riemann
zeta function.
Proposition 19.2.1. (s − 1)ζ(s) is an entire function with order of growth ≤ 1.
Sketch of proof. That it is entire follows from the fact that ζ(s) has a unique simple pole
at s = 1. For the sake of estimate the order of growth, we may ignore the term (s − 1).
There are a couple of estimates to do: Let us first consider the situation when
Re(s) ≥ 1/2 and |s| is large. Firstly, since |ns | = nRe(s) , ζ(s) is bounded on Re(s) ≥ 2.
Secondly, from a previous lecture, we know that if s = σ + iτ with σ ≥ 1/2 and |τ | ≥ 1,
then ζ(s) ≤ C|t|1/2+ for any  > 0. Thus in general case we can estimate |ζ(s)| ≤ A+B|s|
under the above condition.
Let us use the function equation for ξ(s) = π −s/2 Γ(s/2)ζ(s) the Riemann xi function,
given by ξ(s) = ξ(1 − s), to write
Γ((1 − s)/2)
ζ(s) = π s−1/2 ζ(1 − s).
Γ(s/2)
Let us recall that for Re(s) > 0, Γ(s) is bounded by Γ(Re(s)) which can be estimated by
0
CeC Re(s) log(Re(s)) and that 1/Γ(s) has order of growth 1. Thus for Re(s) ≤ 1/2 and |s|
large, we have that
0 1+
|ζ(s)| ≤ CeC |s| (A + B|s|)
for any  > 0. This shows that ζ(s) has order of growth ≤ 1.
1
< ∞ for all . This rules out that the sum
P
In particular, we have that ρ |ρ|1+
P xρ+1
ρ ρ(ρ+1) grows too quickly: under the Riemann hypothesis
 
X xρ+1 
 ≤ Cx3/2 .
 


ρ ρ(ρ + 1) 

In fact, the Riemann zeta function and the Riemann xi function both haves order
of growth exactly 1, which we can use to show that ζ has infinitely many zeroes in the
critical strip. Let us define F (s) = (1/2 − s)(1/2 + s)ξ(s + 1/2), which has removable
singularities and hence is entire.
Lemma 19.2.2. There is a holomorphic function G such that G(s2 ) = F (s + 1/2), with
order of growth 1/2.
Proof. Observe that the function equation ξ(s) = ξ(1 − s) translates to F (s + 1/2) =
F (−(s + 1/2)), so s 7→ F (s + 1/2) is even. This implies that the existence of G; we read
off from Taylor expansion that there are no odd degree terms. It is easy to see that G
must then have half the order of growth of F .
120 Chapter 19 The proof of the prime number theorem

Proposition 19.2.3. The Riemann zeta function has infinitely many zeroes in the
critical strip.

Proof. We proved before that a fractional order of growth implies that there are infinitely
many zeroes, so G has infinitely many zeroes and hence so has (1/2 − s)(1/2 + s)ξ(s + 1/2)
and thus ξ(s). Now recall that ξ(s) has no zeroes outside the critical strip; π s/2 Γ(s/2)ζ(s)
doesn’t for Re(s) > 1 because none of the terms do, and by the functional equation the
same is true for Re(s) < 0. Thus ξ must have infinitely many zeroes in the critical strip,
and hence so does ζ(s).
 Chapter 20

Conformal equivalences

We have mostly been ignoring the role played by the domain Ω of a holomorphic
function. If we mentioned its properties, e.g. whether it is connected or simply-connected,
they appeared at conditions in results about holomorphic functions. In this and the next
lecture we will study the geometry of Ω from the perspective of complex analysis. This
is Section 8.1 of [SS03].

20.1 Conformal equivalences

Two open subsets U, V ⊂ C are the same from the point of view of complex analysis
if there exists a holomorphic bijection f : U → V between them. We will soon see that
inverse is also holomorphic. This is the correct notion, because it means that f induces
a bijection between holomorphic functions V → C and U → C. In this lecture we give
some first examples, and state the Riemann mapping theorem.

Definition 20.1.1. A conformal map is a bijective holomorphic function f : U → V .

This terminology comes from the early days of complex analysis: we saw in one of
the first lectures

Proposition 20.1.2. If f : U → V is holomorphic and injective, then f 0 (z) 6= 0 for all

z ∈ U . More, f −1 : f (U ) → U is also holomorphic.

Proof. The proof is by contradiction, so suppose that f 0 (z0 ) = 0. Then for z near z0 we
can write
f (z) − f (z0 ) = ak (z − z0 )k + G(z)
for some k ≥ 2, ak 6= 0 and G holomorphic vanishing to order k + 1.
Recall Rouché’s theorem says that if F and G are holomorphic on an open set
containing a circle C and its interior, and |F (z)| > |G(z)| for all z ∈ C, then F and
F + G have the same number of zeroes inside C.
As the zero z0 of f 0 is necessarily isolated, we can fix a sufficiently small circle C around
z0 such that (a) z0 is the only point at which f 0 vanishes inside C, (b) |ak (z−z0 )k | > |G(z)|
for all z ∈ C. We will take G(z) as above and F (z) = ak (z − z0 )k − w, with w non-zero

121
122 Chapter 20 Conformal equivalences

small enough so that it is still true that |F (z)| > |G(z)| for z ∈ C. Then Rouché’s
theorem says that because F has at least 2 zeroes in C, so does

F (z) + G(z) = f (z) − f (z0 ) − w.

Since the zero z0 of f 0 is isolated, on a small disk around z0 the derivative of

f (z) − f (z0 ) − w only vanish at z0 . However, as w 6= 0 the two zeroes of f (z) − f (z0 ) − w
within C are not at z0 and hence distinct. This proves that f (z) takes the value f (z0 ) + w
twice and hence is not injective.
Without loss generality f (U ) = V , and we write g := f −1 . We will show that g is
holomorphic at w0 = f (z0 ) ∈ V , which is unique by the first part of this proposition. To
compute the relevant difference quotient, we take w = f (z) ∈ V near w0 . We assume
w 6= w0 and write
−1 −1
g(w) − g(w0 ) w − w0 f (z) − f (z0 )
 
= = .
w − w0 g(w) − g(w0 ) z − z0

Since f 0 (z0 ) 6= 0, we can let z → z0 to see that this converges. We conclude that g is
holomorphic at w0 with derivative 1/f 0 (z0 ).

Corollary 20.1.3. If f is a conformal map, then f −1 is also a conformal map.

Thus the following is an equivalence relation on open subsets of C:

Definition 20.1.4. We say that U, V ⊂ C are conformally equivalent if there is a

conformal map f : U → V .

We just saw that if f is injective then f 0 is non-vanishing. The converse is not true,
e.g. z 7→ z 2 on C \ {0}. However, it is still the case that f is locally injective:

Theorem 20.1.5 (The holomorphic inverse function theorem). If f : U → V has an

everywhere non-vanishing derivative, then for around every z0 ∈ U there exists a open
disk D such that f |D is injective.

Proof. We intend to apply Rouché again. Now, for z near z0 we can write

f (z) − f (z0 ) = a1 (z − z0 ) + G(z)

with G(z) vanishing to order 2 at z0 . We will take C small enough such that |a1 (z −z0 )| >
|G(z)| for all z ∈ C, G(z) as above and F (z) = a1 (z − z0 ) − w, with w non-zero small
enough so that it is still true |F (z)| > |G(z)| for z ∈ C. Then Rouché’s theorem says
that because F has a single zero in C, so does

F (z) + G(z) = f (z) − f (z0 ) − w.

This proves that f (z) takes the value f (z0 ) + w only a single time on the interior D of C
when w is close enough to 0, and hence is injective near z0 .
 20.2 Examples 123

20.2 Examples

To build some intuition for conformal equivalence, we give a few examples of confor-
mally equivalent subsets of C.

20.2.1 The half-plane is conformally equivalent to the unit disk

Our first examples are the upper half plane

H := {z ∈ C | Im(z) > 0},

and the open unit disk

D := {w ∈ C | |w| < 1}.

The maps we will use are examples of Moebius transformation or fractional linear
transformations, which we saw appear as isomorphisms of CP 1 before and which we will
study in more detail in the next lecture.

Theorem 20.2.1. The map

F : H −→ D
i−z
z 7−→ ,
i+z

is a conformal equivalence with inverse

G : D −→ H
1−w
w 7−→ i .
1+w

Proof. It is obvious that F and G are holomorphic, but we must verify that both maps
have the desired codomain. For F , this follows because −i is further from z than i is, so
|i − z| < |i + z|. For G, we may equivalently check that Re( 1−w
1+w ) > 0 for w ∈ D. Writing
w = x + iy we have

1−w 1 − x − iy
   
Re = Re
1+w 1 + x + iy
(1 − x − iy)(1 − x + iy)
 
= Re
(1 + x)2 + y 2
1 − x − y2
2
= >0
(1 + x)2 + y 2

because x2 + y 2 < 1.
124 Chapter 20 Conformal equivalences

Finally we verify that G(F (z)) = z (leaving F (G(w)) = w for the book):
i−z
1− i+z
G(F (z)) = i i−z
1+ i+z
i + z − (i − z)
=i
i+z+i−z
2z
=i
2i
= z.

1.0

0.5

-1.0 -0.5 0.5 1.0

-0.5

-1.0

Figure 20.1 The images under F of [−20, 20] (blue), [−20 + i, 20 + i] (orange), [−20 + 2i, 20 + 2i]
(green).

20.2.2 Further examples

Some of these examples are pictured in Figure 1 on [SS03, p. 213].
Example 20.2.2 (Translations). For a ∈ C, the map

z 7−→ z + a

has the effect of translating. Its inverse is given by z 7→ z − a. For example, it gives a
conformal equivalence from D = D1 (0) to D1 (c). When a is real, it gives a conformal
equivalence from H onto itself.
Example 20.2.3 (Dilation and rotation). For c ∈ C \ {0}, the map

z 7−→ cz
 20.3 The Riemann mapping theorem 125

has the effect of dilating by |c| and rotating by arg(z). It inverse is given by z 7→ c−1 z.
For example, when |c| = 1 it gives a conformal equivalence from D onto itself. When c is
real, it gives a conformal equivalence from H onto itself.
Example 20.2.4 (Powers). The map

S = {z ∈ C | 0 < arg(z) < π/n} −→ H

z 7−→ z n

is a conformal equivalence from the sector S onto the upper half plane. Its inverse is
z 7→ z 1/n (whose definition requires a choice of branch of logarithm, we take the one
where we delete the positive real axis). In fact, this works for αn with 0 < α < 2.
Example 20.2.5 (Logarithm and exponential). Taking the principal branch of the loga-
rithm, the map

C \ {z ∈ R | z ≤ 0} −→ {z ∈ C | 0 < Im(z) < π}

z 7−→ log(z)

gives a conformal equivalence between a complex plane with the negative real axis deleted,
and the vertical strip of complex numbers whose imaginary part lies in (0, 1). Its inverse
is of course z 7→ ez .
We can restrict to intersection of the domain with D; its codomain is then the
half-strip {z ∈ C | 0 < Im(z) < π, and Re(z) < 0}.

20.3 The Riemann mapping theorem

The Riemann theorem is a deep result in complex analysis, which describes all open
subsets U ⊂ C which are conformally equivalent to the open unit disk D. Recall that U
is simply-connected if every closed curve is homotopy to a constant map.

Theorem 20.3.1 (Riemann mapping theorem). Suppose that U is connected, simply-

connected and not equal to C. Then U is conformally equivalent to D. Moreover, if we fix
z0 ∈ U then there is a unique conformal map f : U → D such that f (z0 ) = 0 and f 0 (z0 )
is a positive real number.

Let us recall why there is no conformal equivalence C → D. If this existed, we would

have a bounded entire function, which is constant by Liouville’s theorem and hence not
injective.
It is a purely topological argument that if U is conformally equivalent to D, then it
needs to be simply-connected.

Lemma 20.3.2. If we have a conformal equivalence f : U → D, then U is simply-

connected.

Proof. If γ is a closed curve in U , then f (γ) is a closed curve in D. Since D is simply-

connected, there is a homotopy H from f (γ) to a constant map. Then f −1 H is a
homotopy from f −1 (f (γ)) = γ to a constant map.
 Chapter 21

The Schwartz lemma and conformal

automorphisms

In the last chapter we discussed conformal equivalence of open subsets of C; U and

V are conformally equivalent if there exists an bijective holomorphic function U → V .
Today we discuss what happens when U is equal to V , in the case that U = D or V = D.
Along the way we prove a very useful lemma. This is Section 8.2 of [SS03].

21.1 Schwartz lemma

The Schwartz lemma is an extremely useful result about holomorphic maps D → D:

Theorem 21.1.1 (Schwartz lemma). Suppose that f : D → D is a holomorphic map with

f (0) = 0. Then we have
(i) |f (z)| ≤ |z| for all z ∈ D and equality holds if and onlny if z 7→ c · z with |c| = 1,
(ii) |f 0 (0)| ≤ 1 and equality holds if and only f is given by z 7→ c · z with |c| = 1.
In both case, c = f 0 (0).

Proof. For part (i), we start with the observation that since f has a zero at 0, f (z)/z
has a removable singularity at 0 and hence extends to a holomorphic function. This
extension takes the value f 0 (0) at the origin. Since |f (z)| ≤ 1 we have that
 
 f (z)  1
 z  ≤ |z| .
 

By the maximum modulus principle, the same inequality holds for any w with |w| ≤ |z|:
 
 f (w)  1
 w  ≤ |z| .
 

Letting |z| → 1 we obtain | f (w)

w | ≤ 1 or equivalently |f (w)| ≤ |w|. This completes the
first part of (i). For the second part, observe that if |f (z0 )| = |z0 | then f (z)
z attains a
local maximum in its interior and needs to be constant, necessarily given by some c of
absolute value 1.

126
 21.2 Conformal automorphisms of the upper half-plane and unit disk 127

For part (ii), recall that the value of f (z)/z at 0 is equal to f 0 (0) and hence in absolute
value ≤ 1. If it is equal to 1, we can once again apply the maximum modulus principle
as above.

21.2 Conformal automorphisms of the upper half-plane and unit disk

21.2.1 Conformal automorphisms

Let U ⊂ C be a open subset, then we define

Aut(U ) = {conformal equivalences f : U → U }.

This is a group under composition: its unit is provided by the identity function, we know
that if f is a conformal equivalence so if f −1 , and if f1 , f2 : U → U are holomorphic
bijections then so is their composition f2 ◦ f1 .

Lemma 21.2.1. If U and V are conformally equivalent, then there is an isomorphism

of groups Aut(U ) ∼
= Aut(V ).
Proof. Let g : U → V be a conformal equivalence, then we have a map

g ∗ : Aut(V ) −→ Aut(U )
f 7−→ g −1 ◦ f ◦ g.

This is a bijection because it has (g −1 )∗ as an inverse. It is a group homomorphism

because
g ∗ (idV ) = g −1 ◦ g = idU ,
g ∗ (f2 ◦ f1 ) = g −1 ◦ f2 ◦ f1 ◦ g = (g −1 ◦ f2 ◦ g) ◦ (g −1 ◦ f1 ◦ g) = g ∗ (f2 ) ◦ g ∗ (f1 ). .
Example 21.2.2. Some time ago we studied the conformal automorphisms of C, and
proved that it is given by translation, dilation, and rotation:

Aut(C) = {z 7→ az + b | a 6= 0}.

21.2.2 The conformal automorphisms of D

We next use the Schwarz lemma to determine the conformal automorphisms of D.
Last lecture we saw that the rotations

z 7−→ cz for |c| = 1

are examples. Further examples are given by the Blaschke factors for |α| < 1

βα : D −→ D
α−z
z 7−→ .
1 − αz
You have proven its properties in the very beginning of the course, and we’ve used it
a couple of times since. Let us recall these properties below and give alternative proofs:
128 Chapter 21 The Schwartz lemma and conformal automorphisms

Lemma 21.2.3.
(i) βα is indeed a holomorphic map D → D,
(ii) β interchanges α and 0,
(iii) βα ◦ βα = idD and hence it is bijective.
Proof. Instead of a direct computation, we can prove (i) by observing that βα is holo-
morphic on an open neighborhood of D and that βα (1) = α−1
1−α has absolute value 1. By
the maximum modulus principle we thus must have |βα (z)| < 1 if |z| < 1.
For (ii), we simply compute that
βα (α) = 0 βα (0) = α.
For (iii), we observe that βα ◦ βα is a map D → D which fixes the origin. Hence we
can apply the Schwartz lemma; to prove our result we need to show that the derivative
at 0 is 1. By the chain rule, this is the product of the derivatives of βα at 0 and α:
−1 α−z
βα0 (z) = +α .
1 − αz (1 − αz)2
1
So we get βα (0) = −1 + |α|2 and βα (α) = − 1−|α| 2 , which multiply to 1.

Thus we can get many automorphisms by composing Blaschke factors with rotations.
It doesn’t really matter in which order we do this, because
α − cz αc̄ − z
βα (cz) = =c = cβαc̄ (z).
1 − ᾱcz 1 − ᾱcz
Theorem 21.2.4. Every conformal equivalence of D is given by
z 7−→ cβα (z)
for c with |c| = 1 and α with |α| < 1.
Proof. Suppose we are given a conformal automorphism f : D → D with f (0) = α. Then
g = βα ◦ f is another conformal automorphism fixing the origin. Applying the Schwartz
lemma, we get that
|g(z)| ≤ |z|.
1 Also applying the Schwartz lemma to g −1 , we get
|g −1 (w)| ≤ |w|,
and writing w = g(z) we see
|z| ≤ |g(z)|.
We conclude that |g(z)| = |z| and hence g is a rotation by the Schwartz lemma.
Observing that β0 (z) = −z is the only case when a rotation is a Blaschke factor, we
see that the c and α in previous theorem are almost unique: if f (0) = 0, then there
is a unique representation of f as z 7→ cz. If f (0) = α, then βα ◦ f has a unique such
representation and we conclude that f is z 7→ cβαc̄ (z) with unique |c| = 1 and 0 < |α| < 1.
Setting α = 0 we get:
 21.2 Conformal automorphisms of the upper half-plane and unit disk 129

Corollary 21.2.5. The only conformal automorphisms of D which fix the origin are
rotations.

Example 21.2.6. We can use the classification of conformal automorphisms of D to prove

the uniqueness part of the Riemann mapping theorem: this said that given a connected
simply-connected open subset U ( C and z0 ∈ U there is a unique conformal equivalence
f : U → V such that f (z0 ) = 0 and f 0 (z0 ) is real and positive. For uniqueness, suppose
that f, f˜: U → D both have these properties. Then f˜ ◦ f : D → D is a conformal map
fixing the origin, hence a rotation, with positive real derivative at the origin, hence the
identity map.
Setting z = 0, we see that (of course this only uses the Blaschke factors):

Corollary 21.2.7. Aut(D) acts transitively on D.

We still need to determine the group structure on the set (c, α) of coefficients of
a conformal equivalence z 7→ cβα (z). We will do so by determining the conformal
automorphisms of H instead.

21.2.3 The conformal automorphisms of H

Since there is a conformal equivalence

F : H −→ D
i−z
z 7−→
i+z
we get an isomorphism
∼
F ∗ : Aut(D) −→ Aut(H).
In principle, one can use this to compute Aut(H). We’ll take a slightly different route,
first constructing many elements of Aut(H) and then using the above result to verify we
have found all. Suppose we are given a (2 × 2)-matrix with real entries
" #
a b
M=
c d

with det(M ) = ad − bc = 1. Then we can define a map, called a fractional linear

transformation,

fM : H −→ H
az + b
z 7−→ .
cz + d
Clearly if two matrices differ by −id, they give the same fractional linear transformation.
This is in fact the only way that fM can be equal to fM 0 . I’ll leave this as an exercise.

Theorem 21.2.8. Every conformal equivalence of H is given by fM , where M is unique

up to replacing it by −M .
130 Chapter 21 The Schwartz lemma and conformal automorphisms

Proof. We first verify that fM maps H to H: if Im(z) > 0, then

az + b (az + b)(cz + d)
   
Im = Im
cz + d |cz + d|2
!
ac|z|2 + bd + adz + bcz
= Im
|cz + d|2
(ad − bc)Im(z)
= > 0.
|cz + d|2
h i
a0 b0
a b
Next we prove that fM1 ◦ fM2 = fM1 M2 with M1 = c d , M2 = c0 d0
:: it is given by
0 0
a ac0 z+d
z+b
0 + b
fM2 fM1 (z) = 0 0
c ac0 z+d
z+b
0 + d

a(a0 z + b0 ) + b(c0 z + d0 )
=
c(a0 z + b0 ) + d(c0 z + d0 )
(aa0 + bc0 )z + (ab0 + bd0 )
=
(ca0 + dc0 )z + (cb0 + dd0 )
= fM1 M2 (z).

This shows that each fM is a conformal equivalence with inverse fM −1 .

Next we show that the fM act transitively on H. It suffices to prove at every z ∈ H
can be mapped to i. We first note that setting d = 0 so that bc = −1, we get
az + b Im(z)
 
Im =
cz |cz|2
so by adjusting c we can make the imaginary part i. That is, we shall take
" #
0 −c−1
M1 = .
c 0

Then fM1 (z) = b + i for some b ∈ R, and we can move this to i by applying fM2 with
" #
1 −b
M2 = .
0 1

Next we investigate what happens when we transfer fMθ ∈ Aut(H) with

" #
cos(θ) sin(θ)
Mθ =
− sin(θ) cos(θ)

to Aut(H) using the inverse G : D → H of F given by z 7→ i 1−z

1+z . Since fMθ (i) =
cos(θ)i+sin(θ)
= i, G∗ (fMθ ) fixes the origin and thus is a rotation. Its derivative at the
− sin(θ)i+cos(θ)
origin can be computed using the chain rule as

(G∗ (fMθ )0 (0) = F 0 (i)fM

0
θ
(i)G0 (0) = e2iθ
 21.2 Conformal automorphisms of the upper half-plane and unit disk 131

with
1 1−z
G0 (z) = −i −i so G0 (0) = −2i
1+z (1 + z)2
1 i−z 2i
F 0 (z) = − − so F 0 (i) = −
i + z (i + z)2 .

0 cos(θ) cos(θ)z + sin(θ)(z)

fM (z) = + sin(θ)
θ
− sin(θ)z + cos(θ) (− sin(θ)z + cos(θ))2
0 cos(θ) sin(θ)i
so fM (i) = + = e2iθ .
θ
− sin(θ)i + cos(θ) − sin(θ)i + cos(θ)
Thus G∗ (fMθ ) is rotation by e2iθ . Applying the inverse F ∗ of G∗ we see that all rotations
go to conformal equivalences of the form fMθ .
Under the map F : D → H, the origin goes to i. If f : H → H is a conformal equivalence
sending i to z, we can compose with an fM to get a new conformal equivalence g := fM ◦f
which fixes i. Then under G∗ : Aut(H) → Aut(D) this gets mapped to a conformal
equivalence of D fixing the origin, i.e. a rotation by e2iθ . But rotation z 7→ e2iθ z goes
under F ∗ : Aut(H) → Aut(D) to fMθ . This implies that g = fMθ and hence f = fM −1 Mθ .
For uniqueness, suppose that fM1 = fM2 . By composing with fM moving fM1 (i) =
z = fM2 (i) to i, we get that fM M1 = fM M2 , both fixing the origin. By the above argument,
this is equal to a rotation matrix fMθ . We conclude that M M1 Mθ−1 = M M2 Mθ−1 , up to
a sign, and hence M1 = M2 up to a sign.
Example 21.2.9. If an automorphism f of H fixes two points it is the identity. Indeed, by
conjugating with an automorphism sending f (z) to i, we may assume it fixes i. It hence
corresponds to a rotation of D which fixes three points. But if a rotation fixes two points
it is the identity.
The advantage of the previous proof is that allows us to understand Aut(H) as a
group. It is given by:

Definition 21.2.10. PSL2 (R) is the quotient group of (2 × 2)-matrices with real entries
" #
a b
c d

and determinant 1, under composition, by the central subgroup ±id.

So what we have just shown is that

Aut(D) ∼
= Aut(H) = PSL2 (R).

Example 21.2.11. An alternative description of PSL2 (R) is as follows. Let SU(1, 1) denote
the group of (2 × 2)-matrices with entries in C which fix the Hermitian form
*" # " #+
z1 w
, 1 = z 1 w1 − z 2 w2 ,
z2 w2
132 Chapter 21 The Schwartz lemma and conformal automorphisms

under composition. (A Hermitian form is like a bilinear form over C, expect it is linearity
in the second entry involves a complex conjugation.)
We claim that PSL2 (R) =∼ SU(1, 1). Indeed, setting
" #
1 0
J= ,
0 −1

t
h i
αβ
the condition is that A JA = J. Writing A = γ δ , this gives
" # " #
|α|2 − |γ|2 αβ − γδ 1 0
= ,
γα − δβ −|δ|2 + |β|2 0 −1

so we get γ = β, δ = α and |α|2 − |β|2 = 1. That is,

" #
α β
A=
β α

We can construct a homomorphism

SU(1, 1) −→ Aut(D)
" #
αz + β
 
α β
7−→ z 7→ .
β α βz + α

I’ll it to leave to verify that this preserves the unit disk. That it is a homomorphism is
exactly the same computation as before. By picking α = eiθ and β = 1 we can realize
β−z
all rotations. By picking α = −1 and |β| < 1, we get z 7→ 1−βz and hence can realize
all Blaschke factors. Thus SU(1, 1) → Aut(D) is surjective. I’ll again leave injectivity to
you, but it is essentially because α and β can be recovered from the value and derivative
of the automorphism at the origin.
Thus we have isomorphisms of groups
∼ ∼ ∼
SU(1, 1) −→ Aut(D) −→ Aut(H) ←− PSL2 (R).

What is the explicit isomorphism?

 Chapter 22

Several complex variables

We end with a short preview of functions of several complex variables. This is not
part of the material for the exam. Our reference is [Ran03].1

22.1 Holomorphic functions of several complex variables

As the name suggests, functions of several complex variables are complex-valued

functions defined on an open subset Ω of Cn .

Definition 22.1.1. A function f : Cn ⊃ Ω → C is holomorphic if it is holomorphic in

each variable separately.

As in the case n = 1, the most important properties of such functions are derived
from an integral formula. Let’s study this in the case n = 2 for concreteness. For
z = (z1 , z2 ) ∈ C2 we can fix z1 and apply the Cauchy integral formula in the second
variable to write
1 f (z1 , ζ2 )
Z
f (z1 , z2 ) = dζ2 ,
2πi |ζ2 −a2 |<r2 ζ2 − z2
for those z = (z1 , z2 ) satisfying |ζ2 − a2 | < r2 . Let us now also apply the Cauchy integral
formula in the first variable:
1 f (ζ1 , ζ2 )
Z Z
f (z1 , z2 ) = dζ2 dζ1 ,
(2πi)2 |ζ1 −a1 |<r1 |ζ2 −a2 |<r2 (ζ1 − z1 )(ζ2 − z2 )

for those z = (z1 , z2 ) satisfying |ζj − aj | < rj for j = 1, 2. Defining the polydisk
P (z, r) ⊂ C2 to be {(z1 , z2 ) | |ζj − zj | < rj for j = 1, 2} (this is just a product of disks
in each copy of C), we can write this as

1 f (ζ1 , ζ2 )
Z
f (z1 , z2 ) = dζ1 dζ2 ,
(2πi)2 ∂0 P (a,r) (ζ1 − z1 )(ζ2 − z2 )

with ∂0 P (z, r) = {(z1 , z2 ) | |ζj − zj | = rj for j = 1 or j = 2} (this looks like a product

of circles). Thus we see that the Cauchy integral formula goes through, but with disks
1
I have posted it to Canvas.

133
134 Chapter 22 Several complex variables

generalized in a somewhat surprising manner: not to higher-dimensional disks but to

polydisks.
From this expression we get several generalizations of results we proved for holomorphic
functions:
· By differentiating under the integral sign, we obtain that the partial derivatives of
f are again holomorphic functions of two variables.
· Similarly, we obtain that locally f can be written as a Taylor series of two variables
X 1 ∂i∂j f
f (z1 , z2 ) = (a1 , a2 ) · (z1 − a1 )i (zj − aj )j .
i,j
i!j! ∂z1i ∂z2j

This at least converges on some polydisk P (a, r) around a = (a1 , a2 ), but later we
will see that understanding the precise domain of convergence of such power series
is more subtle.
· As a consequence, the uniqueness of extensions holds. Using power series, this is
easily deduced from the fact that if f vanishes on an infinite set of points which has
a convergence point, it vanishes on the entire component of its domain containing
the convergence point.
In other words, most of the theory for holomorphic functions generalizes. In the next
section I will discuss one feature of the high-dimensional setting which is very different,
which leads to a new phenomena which we discuss in the third section.

22.2 The Hartogs extension theorem

The most important feature distinguishing several complex variables from a single
complex variable concerns the existence of extensions of holomorphic functions:

Theorem 22.2.1 (Hartogs extension theorem). Suppose that B is a bounded connected

open subset of Cn with connected boundary ∂B. If n ≥ 2 then every holomorphic function
f defined on a connected open neighborhood of ∂B has a unique holomorphic extension
to B.

This is false for n = 1: f (z) = 1/z is holomorphic near ∂D1 (0) but does not extend
to D1 (0). Let’s prove a lemma, which is also false for n = 1 (with same counterexample),
that illustrates why n ≥ 2 is important:

Lemma 22.2.2. Suppose that K ⊂ Cn is compact and n ≥ 2. Then every bounded

holomorphic function Cn \ K is constant on the unbounded component of Cn \ K. In
particular it extends to all of Cn .

Proof. Let’s again assume n = 2. Suppose that |z| < R for z ∈ K. Then whenever
|z| ≥ R we have that z1 7→ f (z1 , z2 ) is a bounded entire function, hence constant by
Liouville’s theorem. This means that for |z| ≥ R, f is independent of the second variable.
Reversing the role of z1 and z2 , we see that f is also independent of the first variables.
This shows that f is constant on |z| ≥ R. By the uniqueness of extensions, we see that f
is constant on all of C2 \ K.
 22.3 Domains of holomorphy and convergence of power series 135

To understand the proof in general, we recall a trick we used before to define Laurent
expansions of holomorphic functions f defined on an open neighborhood of the annulus
A = {z ∈ C | r ≤ |z| ≤ R}. If we apply the residue theorem to ζ 7→ fζ−z(ζ)
with a contour
given by almost completing the two circles CR and Cr and connecting them by a strip of
width δ, and δ → 0, we get that

1 f (ζ) 1 f (ζ)
Z Z
f (z) = dζ − dζ. (22.1)
2πi CR ζ −z 2πi Cr ζ −z

The first term defines holomorphic function on DR (0) Similarly, the second term z 7→
1 R f (ζ)
2πi Cr ζ−z dζ defines a holomorphic function on C \ D r (0) and is bounded as |z| → ∞.
Thus if Lemma 22.2.2 had held, the latter would in fact extend to all of C and we would
have obtained our desired extension.
The idea of proof of Theorem 22.2.1 is that (22.1) generalizes to several variables,
through the Bochner–Martinelli formula. As in this case Lemma 22.2.2 does hold, this
proves the result for annuli, and for the general case, we just need to observe that we
can replace the boundaries of higher-dimensional annuli with the boundaries of a more
complicated shape adapted to B.
Example 22.2.3. The Hartogs extension theorem implies that in several complex variables
all isolated singularities are removable. Here an isolated singularity such that f is defined
on a neighborhood over it but not at the point itself, and removable means that the
function in question extends over it. In fact, it just suffices that the singularities we
are considering contained in some bounded subset. Indeed, it is codimension 1 analytic
subsets instead of points that play the role of singularities. As soon as a singularity is
codimension ≥ 2 it is removable.

22.3 Domains of holomorphy and convergence of power series

The curious consequence of the Hartogs extension theorem is that there are subsets
Ω ⊂ Cn such that every holomorphic function f : Ω → Cn extends to some strictly larger
open subset. An Ω for which this is not the case has a name:

Definition 22.3.1. A domain of holomorphy is a subset Ω ⊂ Cn such that at least one

holomorphic function f : Ω → C cannot be extend to any strictly larger open subset.

It turns out to be equivalent to say that for all b ∈ ∂Ω there exists a holomorphic
function fb : Ω → C which does not extend to an open subset containing b.
Example 22.3.2. Every open subset is a domain of holomorphy when n = 1: for b ∈ ∂Ω
1
take fb = z−b . This is the reason we did not see domains of holomorphy appearing in
our course; every open subset is one. Products of open subsets of C are domains of
holomorphy for the same reason.
Example 22.3.3. Convex subsets Ω are domain of holomorphy. To see this, observe that
by translating and rotation, we may assume that b = 0 and Ω lies in Re(z1 ) > 0 (this
uses convexity). Then 1/z1 does not extend over b.
136 Chapter 22 Several complex variables

There is an interesting connection between domains of holomorphy to convergence

of power series. Let us assume the power series we are investigating is centered at the
origin, i.e. of the form i,j aij z1i z2j . Our leading question is: what can we say about the
P

interior Ω of the subset of Cn where this converges absolutely? We call this the domain
of convergence.
If r1 , r2 > 0 are such that
sup |ai,j |r1i r2j < ∞,
i,j
P i jconverges absolutely on the polydisc P (r) = {(z1 , z2 ) ∈ C2 |
then the sum i,j ai,j z1 z2
|zi | < ri }. Indeed, we can then estimate our sums as
 
X  X
j
X
i  i j
C|z1 /r1 |i |z2 /r2 |j ,
 

 a i,j z 1 z 2 ≤ |a i,j ||z 1 | |z 2 | ≤
 i,j  i,j i,j

which converges. Thus we conclude that

Lemma 22.3.4. If p = (p1 , p2 ) lies in the domain of convergence Ω, then Ω contains

the polydisk P (|p|) with |p| = (|p1 |, |p2 |).

This is quite reminiscent of the single-variable situation. However, there is a second

more subtle property. Suppose we have two points p, q ∈ C2 in the domain of convergence.
Then we can pick M such that

sup |aij ||p1 |i |p2 |j ≤ M sup |aij ||q1 |i |q2 |j ≤ M.

i,j i,j

It is then easy to see that when we pick λ ∈ [0, 1] and define s ∈ C2 by

s = (s1 , s2 ) = (|p1 |λ |q1 |1−λ , |p2 |λ |q2 |1−λ )

we have that
sup |aij ||s1 |i |s2 |j ≤ M.
ij

Thus the power series converges then and hence on the entire polydisk P (s). We conclude
that:

Lemma 22.3.5. The domain of convergence Ω has the property that

{(log |z1 |, log |z2 |) | (z1 , z2 ) ∈ Ω} ⊂ R2

is convex.

We will say that Ω is logarithmically convex if it satisfies the condition of the previous
lemma.

Theorem 22.3.6. The following are equivalent for Ω satisfying the condition in Lemma
22.3.4:
· Ω is a domain of convergence for some power series.
 22.3 Domains of holomorphy and convergence of power series 137

· Ω is logarithmically convex.
· Ω is a domain of holomorphy.

Example 22.3.7. Since the unit disk {(z1 , z2 ) | |z1 |2 + |z2 |2 < 1} ⊂ C2 satisfies the
condition in Lemma 22.3.4 and is logarithmically convex, it is a domain of holomorphy
and the domain of convergence for some power series (which one?). Of course, we already
knew the former because it is convex.
 Chapter 23

Final recap

To help you prepare for the final, let me recall what material we covered in the second
half of the course. We covered Chapters 5, 6, and 7 of [SS03] as well as Sections 8.1 and
8.2. Before the midterm, we covered Chapters 1, 2, 3, with the exception of Sections
2.5.5 and 3.7.

23.1 Midterm material

This course was about holomorphic functions:

Definition 23.1.1. For Ω ⊂ C an open subset, f : Ω → C is holomorphic if it is

complex-differentiable at each point in z ∈ Ω.

In the first part of the course we proved that these functions are very well-behaved,
and in particular have the following basic properties:
· If f : Ω → C is holomorphic, it is infinitely many times complex differentiable. In
particular, its derivative is again holomorphic.
· If f : Ω → C is the limit of a sequence of holomorphic functions which converges
on uniformly on compact subsets, it is holomorphic. There are similar results for
sums and integrals.
· Every holomorphic function f is locally equal to a convergent power series
P
n=0 an (z−
z0 )n .
· If f, g : U → C are holomorphic and agree on an open subset V , they agree on all
connected components of U which meet V .
· If f : Ω → C is holomorphic
R
and C ⊂ Ω is toy contour whose interior is also
contained in Ω, then C f (z)dz = 0.
· More generally, if γ0 , γ1 : [0, 1] → Ω are piecewise smooth paths with the same
endpoint which areRhomotopic while R
keeping the endpoints fixed and f : Ω → C is
holomorphic, then γ0 f (z)dz = γ1 f (z)dz.
We obtained more advanced results in Chapter 3: the argument principle, Roucheś
theorem, the open mapping theorem, and the maximum modulus principle. However,

138
 23.2 Zeroes of entire functions 139

there is one particular result I want to highlight, because of its prominent role in later
chapters: the residue theorem.
Recall that if f : Ω → C has isolated singularities, these come in three types: (i)
removable ones, (ii) poles, (iii) essential singularities. We are only interested in the case
of poles; removable ones we will remove and essential ones are too hard. If f : Ω → C
only has poles as isolated singularities we say it is meromorphic. If z0 is a pole f , then
near z0 f can be written as
a−n an−1 a−1
n
+ n−1
+ ··· + + g(z),
(z − z0 ) (z − z0 ) z − z0
with g(z) a holomorphic function which is defined not only near z0 but also at z0 . The
coefficient a−1 is called the residue of f at z0 .
Theorem 23.1.2 (Residue theorem). Suppose that f : Ω → C is meromorphic and
C ⊂ Ω is a toy contour whose interior is also contained in Ω and which avoids the poles
of f . Then we have that
Z X
f (z)dz = 2πi resz0 (f ).
C
poles in int(C)

This is a powerful tool for computing integrals, but also for proving general results
about meromorphic functions.

23.2 Zeroes of entire functions

The residue theorem even has applications for entire functions, i.e. holomorphic
functions f : C → C. Indeed, if f is not constant equal to 0, then z 7→ 1/f (z) will have a
pole wherever f has a zero. We used it to study what can be said about the set of zeroes
of an entire function.

23.2.1 Jensen’s formula

The most important estimate in this topic is Jensen’s formula:
Theorem 23.2.1 (Jensen’s formula). Suppose f is holomorphic near DR (0), f (0) 6= 0,
and f doesn’t vanishes on CR . Then if z1 , . . . , zr are the zeroes of f within CR counted
with multiplicity, we have
r Z 2π
|zk | 1
X  
log |f (0)| = log + log |f (Reiθ )|dθ.
k=1
R 2π 0

This has a consequence for the zero-counting function

n : R>0 −→ N
r 7−→ circles in Dr
Corollary 23.2.2. If z1 , . . . , xr are the zeroes of f within DR , then
Z R Z 2π
n(r) 1
dr = log |f (Reiθ )|dθ − log |f (0)|.
0 r 2π 0
140 Chapter 23 Final recap

23.2.2 Functions of finite order

We say more if we impose a growth condition on the entire function f : it is of order
ρ
≤ ρ if |f (z)| ≤ AeB|z| for some A, B > 0.

Theorem 23.2.3. If f is an entire function of order ≤ ρ, then n(r) ≤ Crρ for some
C > 0 and r sufficiently, and if z1 , z2 , . . . are the zeroes of f which are not equal to 0
then for s > ρ we have
∞
X 1
< ∞.
k=1 k
|z |s

In fact, such functions are nearly determined by their zeroes, as polynomials are.
Recall the canonical factors
(
1−z if k = 0,
Ek (z) := 2 k
(1 − z)ez+z /2+···+z /k if k ≥ 1.

(You can remember these are roughly (1 − z)e− log(1−z) , but using only partial Taylor
expansions for − log(1 − z).) These are versions of 1 − z with more control on their
growth behavior.

Theorem 23.2.4 (Hadamard factorization). Suppose f is entire and has order of growth
ρ. Let k be an integer with k ≤ ρ0 < k + 1. If a1 , a2 , . . . denote the non-zero zeroes of f ,
then
∞
Y
P (z) m
f (z) = e z Ek (z/an ).
n=1

Here P is a polynomial of degree ≤ k, and m is the order of the zero of f at 0 (this order
can be 0).
sin(πz)
Example 23.2.5. The function π has order 1, and its Hadamard factorization simplifies
to
∞
sin(πz) Y
=z (1 − z 2 /n2 ).
π n=1

(The sine function has order 1, but the exponentials in the canonical factor for the zeroes
n and −n cancel.)

23.3 The Gamma function, Riemann zeta function, and the prime num-
ber theorem

The next goal was to prove the following result about the distribution of the prime
numbers. This is phrased in terms of the prime counting function

π : R>0 −→ R
x 7−→ #{primes p ≤ x}.
 23.4 Conformal equivalences 141

π(x)
Theorem 23.3.1. We have that π(x) ∼ x/ log(x), which means that limx→∞ x/ log(x)
exists and is equal to 1.
You interpret this as π(x) as growing like x/ log(x) asymptotically. This result is
proven by studying a pair of special holomorphic functions, the most important of which
is the Riemann zeta function
∞
X 1 Y 1
ζ(s) = = for Re(s) > 1.
n=1
ns primes p
1 − p−s

We proved it has several properties:

· It extends uniquely to a holomorphic function ζ : C \ {1} → C with a simple pole
at s = 1.
· It satisfies a functional equation, best phrased in terms of ξ(s) = π −s/2 Γ(s/2)ζ(s):
ξ(s) = ξ(1 − s).

· The only zeroes outside the critical strip 0 < Re(s) < 1 are the “trivial zeroes” at
s = −2, −4, −6, . . .. In particular, there are no zeroes on Re(s) = 1.
· Moreover, we can bound 1/ζ(s) and ζ 0 (s) near the line Re(s) = 1 in terms of
|Im(s)|.
· The Riemann hypothesis, one of the most famous open conjectures, says that the
only non-trivial zeroes lie on the line Re(s) = 1/2.
Let us focus our attention on the functional equation: it is used all over the place,
e.g. in the continuation of ζ(s) to all of C\{1} and in the results about zeroes. Furthermore,
it involves another function that we studied: the Gamma function
Z ∞
Γ(s) = e−t ts−1 dt for Re(s) > 0,
0

which extends uniquely to a holomorphic function Γ : C \ {0, −1, −2, . . .} → C with simple
poles at the non-positive integers.

23.4 Conformal equivalences

At the end, we shifted our attention from holomorphic functions to their domains.
We studied the following equivalence relation on open subsets U ⊂ C:
Definition 23.4.1. We say that U, V ⊂ C are conformally equivalent if there is a
holomorphic bijection U → V .
To see this is indeed an equivalence relation, we proved that if f : U → V is a
holomorphic bijection, then its inverse f −1 : V → U is also holomorphic. For example,
the upper half-plane H is conformally equivalent to the open unit disk D, and in fact
the Riemann mapping theorem (which we didn’t prove) says that any connected simply-
connected open subset U ( C is conformally equivalent to D. This makes it interesting
to study the conformal automorphisms of D (or equivalently, of H). Using the Schwartz
lemma, we proved these are given by compositions of rotations and Blaschke factors.
 Bibliography

[Mah60] K. Mahler, An application of Jensen’s formula to polynomials, Mathematika 7

(1960), 98–100. MR 124467 77
[MH87] Jerrold E. Marsden and Michael J. Hoffman, Basic complex analysis, second
ed., W. H. Freeman and Company, New York, 1987. MR 913736 1, 75
[Ran03] R.M. Range, Complex analysis: A brief tour into higher dimensions, Amer.
Math. Monthly 110 (2003), 89–108. 133
[SS03] Elias M. Stein and Rami Shakarchi, Complex analysis, Princeton Lectures in
Analysis, vol. 2, Princeton University Press, Princeton, NJ, 2003. MR 1976398
1, 11, 14, 17, 18, 29, 32, 36, 38, 42, 45, 49, 58, 64, 68, 69, 71, 73, 75, 77, 84, 86,
89, 94, 95, 98, 100, 102, 106, 111, 116, 121, 124, 126, 138

142