Complex Analysis

Dr I Petridis
April 3, 2013

Contents
1 Review of Complex Numbers
1.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Polar Form of Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Roots of Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1
1
4
5

2 Limits and Continuity

2.1 Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 (Topology) Open sets in the plane . . . . . . . . . . . . . . . . . . . . . . . .
2.3 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7
7
10
13

Differentiation
18
3.1 Mostly results which are the same as in Analysis 2 . . . . . . . . . . . . . . . 18
3.2 The Cauchy-Riemann Equations . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 (Topology) Convex Sets, Paths and Polygonal Paths . . . . . . . . . . . . . 24
3.4 When if f holomorphic? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5 Harmonic and Conjugate Harmonic Functions. New Differential Operators 34

4 Power Series
4.1 Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 lim sup and lim inf . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 Important Theorems and Results about Complex Power Series
4.4 Examples - Finding the Radius of Convergence of Power Series

.
.
.
.

5 Mapping Properties of Complex Functions. Conformal Maps

5.1 Mapping properties of ez . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Properties of log(z) . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.3 Mapping Properties of w = z 2 . . . . . . . . . . . . . . . . . . . . . .
5.3.1 Plotting Level Curves . . . . . . . . . . . . . . . . . . . . . .
5.3.2 Plotting backwards . . . . . . . . . . . . . . . . . . . . . . .
5.4 Conformal Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.5 Linear Fractional Transformations . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.

37
37
38
39
45

.
.
.
.
.
.
.

49
49
50
51
51
52
53
55

6 Integration Along Curves (Contour Integration)

60
6.1 Curves: Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.2 Important definitions and useful results for later on . . . . . . . . . . . . . . 62
6.3 Properties of Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.4 (T) Cauchys Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
6.5 (T) The Fundamental Theorem of Line Integrals and an Important Corollary 71
6.6 (T)Gousats Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.7 (T) The Antiderivative Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.8 Toy contours . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.9 (T)Cauchys Integral Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.10 (T) Cauchys Formula for f and higher derivatives . . . . . . . . . . . . . . 92
6.11 (T)Cauchys Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.12 (T)Liouvilles Theorem and The Fundamental Theorem of Algebra . . . . . 101
6.13 Power Series Expansion of Holomorphic Functions . . . . . . . . . . . . . . . 106
6.14 Singularities and Zeroes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.15 Residues and The Residue Theorem . . . . . . . . . . . . . . . . . . . . . . . 119
6.16 Liouvilles Theorem: Generalisation . . . . . . . . . . . . . . . . . . . . . . . . 127
6.17 (T) The Principle of Analytic Continuation . . . . . . . . . . . . . . . . . . . 128

6.18 (M) Solving integrals of the type

p(x)
q(x) dx

. . . . . . . . . . . . . . . . . . . . 133

p(x)
6.19 (M) Solving integrals of the type p(x)
q(x) cos(ax)dx or q(x) sin(ax)dx . . 135

6.19.1 Jordans Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

2

6.20 (M) Solving integrals of the type F (sin t, cos t)dt . . . . . . . . . . . . . . 140
0

6.21 (T) Laurent Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

6.22 (T) Casorati-Weierstrass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
6.23 The Mean Value Property and the Maximum Modulus Principle . . . . . . 161

Abstract
Unofficial material. Use with caution :D!

Chapter 1
Review of Complex Numbers
1.1

General

Complex numbers can be represented in the form

z = x + iy
where x, y R and i2 = 1
x = R(z)
y = I(z)
The Real numbers are a subset of the Imaginary numbers
R = {z C s.t. y = 0}
The Purely Imaginary numbers are the numbers z = iy, where y R
Definition 1 (Addition of Complex Numbers)
If z1 = x1 + iy1 and z2 = x2 + iy2 , then
z1 + z2 = (x1 + x2 ) + i(y1 + y2 )
Definition 2 (Multiplication of Complex Numbers)
If z1 = x1 + iy1 and z2 = x2 + iy2 , then
z1 z2 = (x1 + iy1 )(x2 + iy2 ) = x1 x2 + ix1 y2 + iy1 x2 y1 y2
Example 1
Suppose that z = x + iy. Then
1
x
iy
= 2
2
2
z x +y
x + y2
1

Check:

x
iy
1

)
z = (x + iy) ( 2
z
x + y 2 x2 + y 2
x2
y (y)
y
x
= 2

+
i
(x
+
y
)
x + y 2 x2 + y 2
x2 + y 2
x2 + y 2
=1

The Complex Field C

Commutativity: z1 + z2 = z2 + z1 and z1 z2 = z2 z1
Associativity: (z1 + z2 ) + z3 = z1 + (z2 + z3 ) and (z1 z2 )z3 = z1 (z2 z3 )
Identity z + 0 = z and z.1 = z
Inverses z + (z) = 0 and z. z1 = 1 for z =/ 0.
Distributive Law: (z1 + z2 )w = z1 w + z2 w
Definition 3 (Powers of i)

i if n 1(mod4)

1 if n 2(mod4)
in =

i if n 3(mod4)

1 if n 0(mod4)
Definition 4 (Complex Conjugate)
If z = x + iy, then its complex conjugate is defined as z = x iy
Important!
We say that z is purely imaginary iff z = z
We say that z is real iff z = z
Definition 5
z+z
2
zz
I(z) =
2i

R(z) =

Definition 6 (Absolute value of a complex number)

The modulus, or absolute value of a complex number is given by

z = x2 + y 2
Proposition 1
.
2

1. z1 + z2 = z1 + z2
2. z1 z2 = z1 .z2
3. z.z = z2
4. z1 z2 = z1 z2
5. zz21 =
6.

1
z

z1
z2

z
z2

if z2 =/ 0

if z =/ 0

Proof. .
Part 3
z.z = (x + iy)(x iy) = x2 + y 2 + i(yx xy) = x2 + y 2 + 0i = x2 + y 2 = z2
Part 4
z1 z2 2 = (z1 z2 )(z1 z2 ) = z1 z2 z1 z2 = z1 z1 .z2 z2 = z1 2 z2 2 z1 z2 = z1 z2
Part 6
z
1 x iy
= 2
= 2
2
z x +y
z
Example 2 (Proof that if z0 is a root of a polynomial, then z0 is also a root)
Let f (z) = an z n + an1 z n1 + . . . + a1 z + a0 be a polynomial with an , an1 , . . . , a0 R
Let z0 be a root of the polynomial. Then z0 is also a root.
Proof. Since z0 is root, we have that
0 = f (z0 ) = an z0n + an1 z0n1 + . . . + a1 z0 + a0
Taking conjugates of both sides, we get that
0 = an z0n + an1 z0n1 + . . . + a1 z0 + a0
= an z0n + an1 z0n1 + . . . + a1 z0 + a0
= an (z0 )n + an1 (z0 )n1 + . . . + a1 z0 + a0 since all the ai are real
= f (z0 ) z0 is a root of f (z)

Inequalities The following are important inequalities for working with complex numbers
1. The Triangle Inequality:
z w z + w z + w
2. z Rz
3. z Iz
Proof. of the Triangle Inequality
Consider z + w2 = (z + w)(z + w) = (z + w)(z + w) = zz + zw + wz + ww
= z2 + zw + wz + w2 = z2 + zw + zw + w2 = z2 + 2R(zw) + w2
z2 + 2zw + w2 = z2 + 2zw + w2 = z2 + 2zw + w2
= (z2 + w2 )
z + w < z + w by taking the square roots

1.2

Polar Form of Complex Numbers

Since we can represent x and y in polars as x = r cos() and y = r sin(), we have that
z = x + iy = r cos() + ir sin() = r(cos + i sin )
Here

r = z is called the modulus of the complex number

y
= tan is called the argument of the complex number, written as arg(z)
x
Note" is not uniquely defined. If we demand that < , then we say that is the
principle argument .
Proposition 2 (Multiplying Numbers in Polar Form)
Take z1 = r1 (cos 1 + i sin 1 ) and z2 = r2 (cos 2 + i sin 2 )
Then
z1 z2 = r1 r2 (cos 1 + i sin 1 )(cos 2 + i sin 2 )
= r1 r2 (cos 1 cos 2 sin 1 sin 2 + i(cos 1 sin 2 + cos 2 sin 1 ))
= r1 r2 (cos(1 + 2 ) + i sin(1 + 2 )) by the theorem of trig addition
So

z1 z2 = z1 z2 and arg (z1 z2 ) = arg(z1 ) + arg(z2 ) up to multiples of 2

4

Corollary 1 (De Moivres Theorem)

If z = r(cos + i sin ) then z n = rn (cos(n) + i sin(n))
Geometric Interpretation of Complex Numbers Addition and Multiplication
.
Addition
It is the same as vector addition - do it with the parallelogram and triangle laws.

z2

z1 + z2

z2 - z1
z1

Multiplication
Add the arguments and multiply the moduli.
z1z2
|z1z2|

|z2|

1.3

1+2

|z1|

Roots of Complex Numbers

Our aim is compute the nth roots of equations of the form

z n = a , where z C and a C
5

Write
z = (cos + i sin ) = ei
Then
z n = n (cos(n) + i sin(n)) = n ein
Write a as

a = a(cos + i sin ) = aei where the is argument of a

So now we have
n (cos(n) + i sin(n)) = a(cos + i sin ) or n ein = aei
So comparing the moduli and arguments we get that
1

n = a = a n
+ 2m
n
Note that we need to restrict m so that we dont get infinitely many solutions. We take
0 m (n 1).
n = + 2m for m Z =

Example 3
Find the 4th roots of 1.
First of all, we write 1 as
1 = 1(cos 0 + i sin 0) = ei(0+2n)
And z 4 as
z n = r4 (cos(4) + i sin(4)) = r4 ei4
Since r4 = 1, we conclude that r = 1. And for the arguments we have that
4 = 0 + 2n =

2n
for n = 0, 1, 2, 3
4

Computing these explicitly gives

n = 0 1 = 0 z1 = 1ei0 = 1
2

n = 1 2 =
=
z2 = 1ei 2 = (cos( ) + i sin( )) = (0 + 1i) = i
4
2
2
2
4
n = 2 3 =
= z3 = 1ei = (cos() + i sin()) = (1 + 0i) = 1
4
3
3
6 3
3
=
z4 = 1ei 2 = (cos( ) + i sin( )) = (0 1i) = i
n = 3 4 =
4
2
2
2
Note that the roots all lie on the unit circle, and are equally spaced. In, general, the roots
lie on a circle and are equally spaced. There are as many of them as the degree of z.

Chapter 2
Limits and Continuity
2.1

Sequences

Definition 7 (Convergence of a sequence of complex numbers)

The sequence < zn > of complex numbers converges to w C iff
> 0, N s.t. n > N zn w <
Definition 8 (Cauchy sequence)
The sequence < zn > is a Cauchy sequence if
> 0, N s.t. n, m > N zn zm <
Theorem 1 (The combination theorem)
If lim zn = w and lim wn = v then
n

lim (zn + wn ) = w + v
n

lim (zn wn ) = wv
n

The proofs are the same as in Analysis 1 and so we dont give them here.
Proposition 3
Let
Then

zn = xn + iyn where xn = R(zn ) and yn = I(zn )

lim zn = w if and only if

lim xn = R(w)
n

lim yn = I(w)
n

Proof. Part 1:
Assuming that lim(zn ) = w, we have that
> 0, N s.t. n > N zn w <
Consider
R(zn ) R(w)
We have that
R(zn ) R(w) = R(zn w) zn w < R(zn ) R(w) <
So given > 0, choose the same N as for lim(zn ) = w and then we have that n > N ,
R(zn ) R(w) <
which implies that
lim xn = R(w)

We now repeat this proof for the second part. We consider this time
I(zn ) I(w)
We have that
I(zn ) I(w) zn w < I(zn ) I(w) <
So given > 0, we choose the same N as in the definition we are assuming and we have
that n > N ,
I(zn ) I(w) <
which implies that
lim yn = I(w)

Part 2:
Assume that

R(zn ) R(w) as n
I(zn ) I(w) as n

So we have that

> 0, N1 s.t. n > N1 R(zn ) R(w) <

2

> 0, N2 s.t. n > N2 I(zn ) I(w) <

2
Now choose N = max (N1 , N2 ), so that for n > N , both inequalities above hold.
Then > 0, n > N , we have

zn w = R(zn w)2 + I(zn + w)2

2 2
<
+ =
2
2
which means that lim zn = w
n

Corollary 2
If lim (zn ) = w , then
n

lim zn = w
n

lim zn = w
n

Note: zn =

(R(zn ))2 + (I(zn ))2

(R(w))2 + (I(w))2 = w

Proposition 4 (Cauchy convergent)

If zn is convergent sequence, then it is Cauchy. The proof is the same as in Analysis 1.
If zn is Cauchy sequence, then it is convergent. And we actually need to prove this as it
is not necessarily the same.
Cauchy convergent. Assume that < zn > is Cauchy, i.e. we have that
> 0, N s.t. n, m > N zn zm <
Now, we have that
zn zm R(zn zm ) = R(zn ) R(zm )
and thus

R(zn ) R(zm ) zn zm < i.e. R(zn ) R(zm ) <

which means that R(zn ) is a Cauchy sequence of reals. By the General Principle of
Convergence, it converges. Call its limit a, i.e.
lim R(zn ) = a

We repeat the argument for the imaginary part:

zn zm I(zn zm ) = I(zn ) I(zm )
and thus

I(zn ) I(zm ) zn zm < i.e. I(zn ) I(zm ) <

which means that R(zn ) is a Cauchy sequence of reals. By the General Principle of
Convergence, it converges. Call its limit b, i.e.
lim I(zn ) = b

But now we can conclude that

lim = lim (R(zn ) + I(zn )) = a + ib

i.e. the sequence is convergent.

9

2.2

(Topology) Open sets in the plane

Definition 9 (Open Disk)

The open disk centred at z0 with radius > 0 is defined by
D(z0 , ) = {z C s.t. z z0 < }
So thats everything in the disk, including the centre, but without the boundary, i.e.
without the circle.
Definition 10 (Punctured open disk)
The punctured (open) disk centred at z0 with radius > 0 is defined by
D (z0 , ) = {z C s.t. 0 < z z0 < }
So here we are excluding both the centre and the boundary of the disk.
Finally, we have
Definition 11 (Closed Disk)
The closed disk centred at z0 with radius > 0 is defined by
D(z0 , ) = {z C s.t. z z0 }
So this set includes the disk with its centre and boundary.
Definition 12 (Open Set)
A set S is called open if for all z S, we can find an open disk D(z, r) such that
D(z, r) S
Remark: r depends on z.
Method 1 (Showing that something is an open set)
.
1. We consider a point z is the set S which we want to prove to be open. We put an
open disk D around this point, i.e. with this point at its centre, i.e. we take the
disk to be D(z, r), where r depends on z. Note that we generally need to find some
expression for r since otherwise we wont be able to prove much.
2. With pick a point w in this disk and we write an expression for w, i.e. w z < r,
by our definition of an open disk.
3. We try to prove that this point also lies in the set S by looking at the properties of
the elements of S and trying to deduce them from whatever expression we got from
w z < r
10

Example 4 (The open disk is an open set)

We consider the open disk D(z0 , ). We need to show that we can find another open disk,
say D(z, r) which is inside the original disk.
So our new disk is centred at some point z. We first need to write an expression for its
radius, r.
Suppose that we picked the biggest disk centred at z which is included in the disk centred
at z0 , so that the diameter of the disk centred at z is the same as the radius of the disk
centred at z0 .

Im

r=e-|z-z0|

z0

|z-z0|

Re

Then we have that

= r + z z0 r = z z0

First all, the radius r is positive, which we can see as follows:

Since z D(z0 , ), we know that z z0 < . But then r = z z0 > 0.
What we want to prove now is that
w D(z, r) w D(z0 , )
where w is any point in the disk D(z, r) which we constructed.
s we have
w D(z, r) w z < r = z z0
and need to show
w D(z0 , ) w z0 <
Start from the one we want to show
w z0 = w z0 + z z = (w z) + (z z0 )
w z + z z0 < r + z z0
= z z0 + z z0 =
11

So
w z0 <
which is what we wanted to show.
Example 5
Show that the set
A = {z C; R(z) > 0}
is open
We start by picking any point z A and drawing a disk around it. Let the radius of the
disk be R(z)

r = Re(z)

One possibility for such

a disk; it is big enough
so that it is touching the
imaginary axis and thus
the radius is Re(z)

Well show that D(z, R(z)) A

First of all, the disk has positive radius, since z A R(z) > 0.
So now we take a point w D and we are aiming to show that w A.
So we are assuming
w D(z, R(z)) w z < R(w)
and trying to prove that
w A R(z) > 0

12

We have
R(w z) w z < R(z)
= R(w) R(z) w z < R(z)
R(w) R(z) < R(z)
R(z) R(w) R(w) R(z) < R(z) by the triangle inequality
R(z) R(w) < R(z)
R(w) < 0
R(w) > 0
Example 6 (A set which is not open)
The set
B = {z C; R(z) 0}
This set includes the imaginary axis. So suppose we took a point z B which lies on
the imaginary axis. Then any disk we draw around this point is not going to be fully
contained in the set B, in fact exactly half if the disk will be outside B.
Generally, any similar set, with a non-strict inequality, is not open. We need to have a
strict inequality. Then if we draw is disk around a point very close to the, say, imaginary
axis, the disk is still contained in the set, even though its radius might be infinitesimal.

2.3

Functions

In the following definitions and theorems, we are assuming that

f C
where is a subset of C. We are also going to write are real part of the function f (z) as
u(x, y) and the imaginary part as v(x, y)
Example 7
f (z) = z 2 f (x + iy) = (x + iy)2 = x2 + y 2 + 2ixy
so f (x + iy) = u(x, y) + iv(x, y)
where u(x, y) = R(f ) = x2 + y 2 and v(x, y) = I(f ) = 2xy
Definition 13 (Limit of a function)
lim f (z) = w0

zz0

> 0, > 0 s.t. 0 < z z0 < and z f (z) w0 <

Remark: 0 < z z0 < , z can be written as z D (z0 , ).

13

Definition 14 (Continuity at a point)

We say that f is continuous at z0 if
z0 and lim f (z) = f (z0 )
zz0

Remark: The limit is unique.

Definition 15 (Continuity in an interval)
f is continuous on if f is continuous at z0 for all z0
Theorem 2
lim f (z) = w0 ( lim R(f ) = R(w0 )) ( lim I(f ) = I(w0 ))

zz0

zz0

zz0

Proof. Part 1:
We are assuming
lim f (z) = w0 > 0, > 0 s.t. z 0 < z z0 < f (z) w0 <

zz0

To prove that
lim R(f ) = R(w0 )

zz0

we consider
R(f w0 ) = R(f ) R(w0 ) f (z) w0
But then, given > 0, for the same > 0 as in the definition we are assuming, we have
that
0 < z z0 < R(f ) R(w0 ) < R(f ) R(w0 ) as z z0
Similarly,
I(f w0 ) = I(f ) I(w0 ) f (z) w0
But then, given > 0, for the same > 0 as in the definition we are assuming, we have
that
0 < z z0 < I(f ) I(w0 ) < I(f ) I(w0 ) as z z0
Part 2:
Assume:

lim R(f ) = R(w0 )

zz0

lim I(f ) = I(w0 )

zz0

So

> 0, 2 > 0 s.t. 0 < z z0 < 2 I(f ) I(w0 ) <

2
Choose = min{1 , 2 } > 0. Then, for 0 < z z0 < , both inequalities above hold. So we
have

0 < z z0 < R(f ) R(w0 ) < and I(f ) I(w0 ) <

2
2
> 0, 1 > 0 s.t. 0 < z z0 < 1 R(f ) R(w0 ) <

14

So consider now
f (z) w0 R(f (z)) R(w0 ) + I(f (z)) I(w0 ) <

+ =
2 2

So we have shown that

> 0, > 0 s.t.
0 < z z0 < f (z) w0 < so lim f (z) = w0
zz0

Corollary 3
f is continuous at z0 iff

R(f ) is continuous at z0
I(f ) is continuous at z0

Proposition 5 (The Combination Theorem)

lim (f (z) + g(z)) = lim f (z) + lim g(z)

zz0

zz0

zz0

lim (f (z).g(z)) = ( lim f (z)).( lim g(z))

zz0

zz0

lim (

zz0

zz0

lim f (z)
f (z)
zz0
)=
if lim(g(z)) =/ 0
g(z)
lim g(z)
zz0

Proposition 6 (Sequential Definition of continuity of a function)

We say that f is continuous at z0 iff for all sequences < zn > with lim zn = z0 , we have
n
that
lim f (zn ) = f (z0 )
n

The proof is the same as in Analysis 1, and is not examinable.

Proposition 7
Let p(z) = an z n + an1 z n1 + . . . + a1 z + a0 be a polynomial
Then it is continuous on the whole complex plane.
Proof. Let z0 C Then
lim (p(z)) = lim (an z n + an1 z n1 + . . . + a1 z + a0 )

zz0

zz0

= lim (an z n ) + lim (an1 z n1 ) + . . . + lim (a1 z) + lim (a0 ) by The Algebra of limits
zz0

zz0

= an lim z + an1 lim z

zz0

zz0

zz0

n1

zz0

+ . . . + a1 lim z + a0 since the ai are constants

zz0

= an ( lim z) + . . . + a1 ( lim z) + a0 by the Combination Theorem

zz0

zz0

= an z0n + . . . + a1 z0 + a0
= p(z0 )
15

Corollary 4
If R(z) is a rational function, i.e.
R(z) =

P (z)
where P, Q K[x]
Q(z)

then R(z) is continuous whenever Q(z) =/ 0

Example 8
f (z) = Arg(z) for z =/ 0 and < Arg(z)
If z0 > 0 for z R, then f (z) is continuous at z0 . But if z0 < 0 for z0 R, then f (z) is
not continuous at z0 . Why is this so? We learned in Analysis 1 that, using the sequential
definition, we can prove that a function is discontinuous at a point z0 if we can find two
sequences converging to this point, e.g.
lim zn = z0

lim wn = z0

But with
lim f (zn ) =/ lim f (wn )
We are going to consider the sequences
zn = z0 (cos(n ) + i sin(n ))
where n is the sequences which tends to , and which n <
wn = z0 (cos(n ) + i sin(n ))
where n is the sequence which tends to and which n >

16

|z0|
z0
|z0|

n
0
n

So we have that
zn z0 (cos() + i sin()) = z0 = z0
wn z0 (cos() + i sin()) = z0 = z0
But
f (zn ) = Arg(zn ) = n
f (wn ) = Arg(wn ) = n
and =/ . So the function is not continuous at z0

17

Chapter 3
Differentiation
3.1

Derivatives and Important Results

Definition 16 (Derivative)
Let be an open set in C and let z0 . Let f C. If
f (z) f (z0 )
exists, we call it the derivative of f at the point z0
zz0
z z0
lim

We write
f (z0 ) = lim

zz0

f (z) f (z0 )
z z0

and we say that f is

holomorphic
analytic
differentiable
at the point z0
The alternative definition is
f (z0 ) = lim
h0

f (z + h) f (z)
h

Note that h here is a complex number! So we can approach 0 from everywhere in the
complex plane, including from the real or imaginary axis. In other words, for a complex
function to be differentiable, it must be true that the limit exists and it unique no matter
where we approach z0 from.
Similarly, z0 is complex number.
Remark: Since is open and z0 , > 0 s.t. D(z0 , ) . Therefore, if z D(z0 , ),
then f (z) makes sense. i.e. for h < , z = z0 + h

18

Im

0
h

Re

Terminology
Definition 17
If f is holomorphic for all z0 , we say that f is holomorphic in .
Definition 18
If S is any subset of C, we say that f is holomorphic on S if we can find open s.t. S
s.t. f is holomorphic on
Definition 19
If f is holomorphic on all C, then we say that f is entire or integral function.
Example 9
Find the derivative of
f (z) = z 2
We have
z 2 z02
f (z) f (z0 )
= lim
= lim (z + z0 ) = z0 + z0 = 2z0
zz0 z z0
zz0
zz0
z z0

f (z0 ) = lim

So f (z0 ) = 2z0 f (z) = 2z

Example 10
Find the derivative of
f (z) = z
z0 + h z0
f (z0 + h) f (z0 )
= lim
h0
h
h
z0 + h z0
h
= lim
= lim
h0
h0 h
h
But we said that h C, so this limit must exist and be unique to matter what h is. Lets
pick two options for h and show that they wont give us the same answer.
f (z0 ) = lim
h0

19

If h R. then we know that h = h. So we have that

h
h
= lim = lim 1 = 1
h0 h
h0
h0 h

lim

If h is purely imaginary, then we know that h = h, and so

h
h
= lim
= lim 1 = 1
h0 h
h0 h
h0

lim

So we get different limits and hence the limit is not unique, which implies that the
derivative doesnt exist. So the function is not differentiable.
Proposition 8
If f is holomorphic at z0 , then f is continuous at z0 .
Proof. Consider
f (z) = f (z0 ) =

f (z) f (z0 )
.(z z0 )
z z0

Let z z0 . By definition
f (z) f (z0 )
= f (z0 ) since we are assuming that f is holomorphic
zz0
z z0
lim

And also,
lim (z z0 ) = 0

zz0

So, taking the limit of the LHS, we get that

lim f (z) f (z0 ) = 0 f (z) f (z0 ) as z z0

zz0

Theorem 3 (The Combination Theorem)

Let f, g he holomorphic on . Then
f + g is holomorphic on and (f + g) = f + g
f g is holomorphic on and (f g) = f g + f g
f f g f g
f
If g(z0 ) =/ 0 , then is holomorphic at z0 and ( ) =
g
g
g2
Theorem 4 (The Chain Rule)
If f V and g V C are holomorphic, then g f C is holomorphic and
(f g) (z) = f (g(z))g (z)

20

Remarks:
f (z) = c f (z) = 0
f (z) = z n f (z) = nz n1
If p(z) = an z n + an1 z n1 + . . . + a1 z + a0 then p (z) = nan z n1 + (n 1)an1 z n2 + . . . + a1
Example 11
Another function which is not holomorphic.
f C C, f (z) = z2
Consider
f (z0 + h) f (z0 ) z0 + h2 z0 2 (z0 + h)(z0 + h) z0 z0
=
=
h
h
h
(z0 + h)((z0 ) + (h)) z0 z0 + hz0 + hz0 + hh z0 z0
=
=
h
h
hz0 + hz0 + hh
h
= z0 + z0 + h
=
h
h
To find the derivative, we need to take the limit of this expression as h 0. However,
once again, considering different h leads to different results and to the conclusion that f
is not holomorphic since the limit is not unique.
For example, consider h R. Then, since h = h, we have that
lim (z0 +
h0

h
+ h) = z0 + z0 + h
h

On the other hand, if h is purely imaginary, so that h = h, we have that

lim (z0 +
h0

h
+ h) = z0 z0 h
h

Since the results are different, the function is not holomorphic!

Note that there is one point where the function is holomorphic, and that is at 0, since, if
z0 = 0, the limit is just 0, so it does exist.
Definition 20 (Partial Derivatives)
Recall the definition and notation for partial derivatives from Methods 2: If u R2 R,
then
u
u(x0 + h, y0 ) u(x0 , y0 )
(x0 , y0 ) = lim
= ux (x0 , y0 )
h0
x
h
u
u(x0 , y0 + h) u(x0 , y0 )
(x0 , y0 ) = lim
= uy (x0 , y0 )
h0
y
h

21

3.2

The Cauchy-Riemann Equations

Theorem 5 (Cauchy-Riemann equations)

Let f (z) = u(x, y) + iv(x, y)
where z = x + iy and u(x, y) = R(f ) and v(x, y) = I(f ).
Assume that f is holomorphic at z0 and that u and v have partial derivatives at (x0 , y0 ).
Then
v
u
(x0 , y0 ) + i (x0 , y0 )
f (z0 ) =
x
x
v
u
=
(x0 , y0 ) i (x0 , y0 )
y
y
Hence, u and v satisfy the following equations known as The Cauchy-Riemann Equations
v
u
(x0 , y0 ) =
(x0 , y0 )
x
y
v
u
(x0 , y0 ) = (x0 , y0 )
y
x
Remarks:
The CRE show that if we have two functions u, v R2 R, then it is very unlikely
that u + iv is holomorphic. So the CRE impose restrictions on u and v.
The CRE are necessary but not sufficient conditions for u + iv to be holomorphic.
Example 12
Let the function
f (z) = z 2 = (x + iy)2 = x2 y 2 + 2ixy
So here

u(x, y) = x2 y 2 and v(x, y) = 2xy

And they satisfy the Cauchy-Riemann equations:

v
u
= 2x =
= 2x
x
y
u
v
= 2y =
= 2y
y
x
Proof. In the following proof we are going to use the fact that
lim (R(f (z))) = R ( lim f (z))

zz0

zz0

lim (I(f (z))) = I ( lim f (z))

zz0

zz0

22

We are going to consider two possibilities for h. We are going to consider h real and h
purely imaginary and so arrive at the two forms of the derivative of f .
Part 1: h R
Consider
f (z0 + h) f (z0 ) u(x0 + h, y0 ) + iv(x0 + h, y0 ) u(x0 , y0 ) iv(x0 , y0 )
=
h
h
Consider only the real part of this expression
R(

f (z0 + h) f (z0 )
u(x0 + h, y0 ) u(x0 , y0 )
)=
h
h

We take te limit as h 0 of both sides and we have that

u(x0 + h, y0 ) u(x0 , y0 )
f (z0 + h) f (z0 )
) = lim
h0
h
h
f (z0 + h) f (z0 )
u
R (lim
)=
(x0 , y0 ) by defintion
h0
h
x
u
R (f (z0 )) =
(x0 , y0 )
x

lim R (
h0

Similarly, from the imaginary part we get that

I(

f (z0 + h) f (z0 )
v(x0 + h, y0 ) v(x0 , y0 )
)=
h
h

We take te limit as h 0 of both sides and we have that

v(x0 + h, y0 ) v(x0 , y0 )
f (z0 + h) f (z0 )
) = lim
h0
h0
h
h
f (z0 + h) f (z0 )
v
I (lim
)=
(x0 , y0 ) by defintion
h0
h
x
v
(x0 , y0 )
I (f (z0 )) =
x

lim I (

Combining the two results together, we get that

f (z0 ) = Rf (z0 ) + iIf (z0 ) =

u
v
(x0 , y0 ) + i (x0 , y0 )
x
x

Part 2: h is purely imaginary

Take h = it, since h is purely imaginary. Consider then
f (z0 + h) f (z0 ) u(x0 , y0 + t) + iv(x0 , y0 + t) u(x0 , y0 ) iv(x0 , y0 )
=
h
it
Again, we first take the real part of this expression. We have that
R(

f (z0 + h) f (z0 )
v(x0 , y0 + t) v(x0 , y0 )
)=
h
t
23

Then again we take the limit as h 0 t 0 and we get that

v(x0 , y0 + t) v(x0 , y0 )
f (z0 + h) f (z0 )
) = lim
t0
h0
h
t
f (z0 + h) f (z0 )
v
R (lim
)=
(x, y0 )
h0
h
y
v
(x, y0 )
Rf (z0 ) =
y

lim R (

And for the imaginary part, we have that

I(

f (z0 + h) f (z0 )
u(x0 , y0 + t) + u(x0 , y0 )
)=
h
t

We take the limit as h 0 t 0 and we get that

f (z0 + h) f (z0 )
u(x0 , y0 + t) + u(x0 , y0 )
) = lim
t0
h0
h
t
f (z0 + h) f (z0 )
u(x0 , y0 + t) u(x0 , y0 )
I (lim
) = lim
t0
h0
h
t
u
If (z0 ) = (x, y0 )
y

lim I (

Combining these two, we get that

f (z0 ) = Rf (z0 ) + iIf (z0 ) =

u
v
(x0 , y0 ) + i (x0 , y0 )
y
y

Finally, the Cauchy-Riemann equations come from comparing the real and imaginary
parts of the two results we have for f (z0 )

3.3

(Topology) Convex Sets, Paths and Polygonal Paths

Definition 21 (Line Segment)

The line segment between a and b is written as [a, b] and is defined as
[a, b] = {z = ta + (1 t)b, t [0, 1]}
Definition 22 (Convex Set)
A set S is called convex if a, b S, the line segment [a, b] is in S. i.e.
[a, b] S
Example 13
The set A = {z R(z) > 0} is convex
24

Proof. As before, with open sets, we are going to proceed by picking points in the given
set which form, in this case, the line segment and then picking a random point in the line
segment and showing that it lies in the set. So we have:
Let a, b A. Then R(a) > 0 and R(b) > 0. We are going to show that [a, b] A. That is,
we pick any point z [a, b] and we want to show that z A, i.e that R(z) > 0
Since z [a, b], t [0, 1] s.t. z = ta + (1 t)b, so we can write
R(z) = R(ta + (1 t)b) = R(ta) + R((1 t)b) = tR(a) + (1 t)R(b)
But since t [0, 1], we have that (1 t) > 0. Also, we are assuming that R(a) > 0 and
R(b) > 0, and so the whole expression above is positive. That is, R(z) > 0 and so we can
conclude that z A
Example 14
The disk D(z0 , r) is convex
Proof. Let a, b D(z0 , r) So we have that
a z0 < r and b z0 < r
To show that [a, b] D(z0 , r), we pick a point z [a, b] and we need to show that
z D(z0 , r), or that is to show that z z0 < r. So,
z [a, b] t [0, 1] s.t. z = ta + (1 t)b
Now,
z z0 = ta + (1 t)b z0 = ta + (1 t)b (t + (1 t))z0
= ta tz0 + (1 t)b (1 t)z0 = t(a z0 ) + (1 t)(b z0 )
t(a z0 ) + (1 t)(b z0 ) = ta z0 + (1 t)b z0
< t.r + (1 t)r = r
So z z0 < r, which is what we wanted to prove.
Definition 23 (Polygonal Path)
A polygonal path from a to b is a union of a finite sequence of line segments starting at
a and finishing at b; i.e. a finite sequence of points z0 , z1 , . . . , zn s,t, z0 = a, zn = b and
[z0 , z1 ] [z1 , z2 ] . . . [zn1 , zn ]

25

zn = b
z2
z3

a = z0

...
.....

zn-1

z1

Definition 24 (Polygonally Connected Set)

A set S is called polygonally connected if a, b S, we can find a polygonal path lying
completely in S.
Example 15
Any convex set is polygonally connected
Example 16
Here is an example of a set which is not convex and not polygonally connected
D(0, 1) D(1, 1)
Even though we can draw as many polygonal paths inside the set, we can also find ones
which dont lie entirely in the set, while having their endpoints in the set.

This path doesnt lie in the set

This path lies in the set

1
These paths (segments) lie
in the set

26

Definition 25 (Domain (Region))

A set is called a domain or a region if it is open and polygonally connected.
Theorem 6
Let f be holomorphic in a domain . Then
If f (z) = 0, z , then f = const, i.e. k C s.t. f (z) = k, z
If k 0 s.t. f (z) = k z , then f = const, i.e. c C s.t. f (z) = c, z .
Remark: needs to be polygonally connected. As an example of a set which is
not polygonally connected and doesnt satisfy the conditions of the theorem, take =
D(0, 1) D(3, 1). Let f (z) = 1 on the disk D(0, 1) and f (z) = 1 on the disk D(3, 1).
Then f = 1 but the function is not constant.

Proof. Part 1
First of all, we are allowed to use the CRE, since f is holomorphic, so the assumption
that f (z) = 0 gives us that
u u v v
=
=
=
=0
x y x y
Now consider the following diagram:
x =x0
y =y0

We have that
u(x, y0 ) = const depending on y0 , since

27

u
=0
x

v
=0
x
u
u(x0 , y) = const depending on x0 , since
=0
y
v
v(x0 , y) = const depending on x0 , since
=0
y
Now consider the following:
Let z0 . Since is open, we can find r0 > 0 s.t. D(z0 , r0 ) . Take z, w D(z0 , r0 ).
such that [z0 , w] is a horizontal segment and [w, z] is a vertical segment. Now, on the
horizontal segment, the y-value is constant and on the vertical segment, the x-value is
constant.
So we have that, for the horizontal line segment [z0 , w]:
v(x, y0 ) = const depending on y0 , since

u(z0 , y) = u(w, y) and v(z0 , y) = v(w, y) where y is fixed, so we can write u(w) = u(z0 ) and v(w) = v(z0 )
This implies that
f (w) = f (z0 )
And similarly, from the vertical line segment [w, z], we have that:
u(x, z) = u(x, w) and v(x, z) = v(x, w) where x is fixed, so we can write u(w) = u(z) and v(w) = v(z)
This implies that
f (w) = f (z)
So (f (w) = f (z)) (f (w) = f (z0 )) f (z) = f (z0 ) = cosnt . So we conclude that the
function is constant of the disk D(z0 , r).
But this only proves the result for the disk in our domain, and we want to prove that f
is constant everywhere in the domain .
So we consider the following: Since is open, we can connect any point on it to any other
point with a polygonal path. So we can get from any a to any b via a polygonal
path. We are going to cover the whole path with overlapping disks and prove that since
f is constant on all of the disks, it has to be the same constant everywhere and thus the
same constant in the whole of .

28

Let the disks be

D(z0 , r0 ) with r0 > 0

D(z1 , r1 ) with r1 > 0

D(zn , rn ) with rn > 0

We construct a sequence of zi on the polygonal path, with ri > 0 s.t.

n

D(zi , ri ) convers the polygonal path

i=1

Moreover, all the disks are in , i.e. i, D(zi , ri ) . Thus we know from the previous
argument that f = const on these disks, say
f (z) = ci , z D(zi , ri )
But since the disks overlap, i.e.
D(zi , ri ) D(zi+1 , ri+1 ) =/
this is in fact the same constant; i.e. ci = ci+1 , i
Part 2
We are assuming that k 0 s.t. f (z) = k, z . So we can write
u(x, y) + iv(x, y) = k u2 + v 2 = k 2
Differentiating w.r.t x, we get
2u

v
u
v
u
+ 2v
= 0 u
+v
=0
x
x
x
x

Differentiating w.r.t y, we get

2u

u
v
u
v
+ 2v
= 0 u
+v
=0
y
y
y
y

Suppose that k =/ 0. We consider the system

uux + vvx = 0
uuy + vvy = 0 uvx + vux = 0 from the CRE
This is a system in unknowns ux , vx and we can solve it by Cramers Rule. We have

ux =

0 v

0 u

u v

v u

0
0
0
=
= 2 = 0 as k =/ 0
2
2
2
v
(v + v ) k

u2

29

u 0

v 0

vx =

u v

v u

(v 2

0
0
= 2 =0
2
+ u ) k

So the conclusion is that ux = vx = 0 and thus

f (z) = ux + ivx = 0 f is const

3.4

When if f holomorphic?

Theorem 7
Let f (z) = u(x, y) + iv(x, y) be defined on an open set containing z0 = x0 + iy0 . Assume
the following:
u and v are continuous on

u u v v
x , y , x , y

exist and are continuous on

The CRE hold at z0

Then, f is holomorphic at z0
Reminder of the MVT:
If f [a, b] R is differentiable on (a, b) and continuous on [a, b], then (a, b) s.t.
f (a) f (b) = f ()(b a)
Proof. Since is open and z0 , r > 0 s.t. D(z0 , r)
(z0 )
Take some z D(z0 , r). Our aim will be to look at lim f (z)f
zz0
zz0
We have that
f (z)f (z0 ) = u(x, y)+iv(x, y)u(x0 , y0 )iv(x0 , y0 ) = u(x, y)u(x0 , y0 )+i[v(x, y)v(x0 , y0 )]
where u D(z0 , r) R.
We polygonally connect z0 to z, where z is some point in D. So we can write
f (z) f (z0 ) = u(x, y) u(x0 , y0 ) + i[v(x, y) v(x0 , y0 )]
= u(x, y) u(x, y0 ) + u(x, y0 ) u(x0 , y0 ) + i[v(x, y) v(x, y0 ) + v(x, y0 ) + v(x, y0 )]
We are going to apply the MVT 4 times to the 4 pairs of terms in the latter expression.
MVT on u(x, y0 ) u(x0 , y0 ). Remember that u is now a single-variable function since we
are holding y0 fixed, which is why we can apply the MVT. We have that
Jx where Jx = [x0 , x] or [x, x0 ], s.t. u(x, y0 ) u(x0 , y0 ) =
30

u
(, y0 )(x x0 )
x

MVT on u(x, y) u(x, y0 ). u is again single variable as we are holding x fixed. We have
Jy where Jy = [y, y0 ] or [y0 , y], s.t. u(x, y) u(x, y0 ) =

u
(x, )(y y0 )
y

MVT on v(x, y0 ) v(x0 , y0 ); holding y0 fixed.

Jx where Jx = [x0 , x] or [x, x0 ], s.t. v(x, y0 ) v(x0 , y0 ) =

v
( , y0 )(x x0 )
x

MVT on v(x, y) v(x, y0 ); holding x fixed.

Jy where Jy = [y, y0 ] or [y0 , y], s.t. v(x, y) v(x, y0 ) =

v
(x, )(y y0 )
y

We now take all of these results and plug them into the above expression of f (z) f (z0 ).
We get
f (z)f (z0 ) =

u
v
v
u
(x, )(yy0 )+ (, y0 )(xx0 )+i [ (x, )(y y0 ) +
( , y0 )(x x0 )]
y
x
y
x

We rewrite this by adding and subtracting the same terms as:

u
u
u
(x, )
(x0 , y0 )] (y y0 ) +
(x0 , y0 )(y y0 )
y
y
y
u
u
u
(x0 , y0 )] (x x0 ) +
(x0 , y0 )(x x0 )
+ [ (, y0 )
x
x
x
v
v
v
+ i ([ (x, ) (x0 , y0 )] (y y0 ) + (x0 , y0 )(y y0 ))
y
y
y
v
v
v
(x0 , y0 )] (x x0 ) +
(x0 , y0 )(x x0 ))
+ i ([ ( , y0 )
x
x
x

f (z) f (z0 ) = [

Which we rearrange as
u
u
(x0 , y0 )(y y0 ) +
(x0 , y0 )(x x0 )
y
x
v
v
+ i ( (x0 , y0 )(y y0 ) +
(x0 , y0 )(x x0 ))
y
x
u
u
+ ( (x, )
(x0 , y0 )) (y y0 ) A
y
y
u
u
+ ( (, y0 )
(x0 , y0 )) (x x0 ) B
x
x
v
v
+ i ( (x, ) (x0 , y0 )) (y y0 ) C
y
y
v
v
+ i ( ( , y0 )
(x0 , y0 )) (x x0 ) D
x
x

f (z) f (z0 ) =

31

And by the CRE, this is equal to

v
u
u
v
(x0 , y0 )(y y0 ) +
(x0 , y0 )(x x0 ) + i (x0 , y0 )(y y0 ) + i (x0 , y0 )(x x0 )
x
x
x
x
+A+B+C +D
v
u
(x0 , y0 ) ((x x0 ) + i(y y0 )) +
(x0 , y0 ) (i(x x0 ) (y y0 )) + A + B + C + D
=
x
x
u
v
=
(x0 , y0 )(z z0 ) + i (x0 , y0 )(z z0 ) + A + B + C + D
x
x
f (z) f (z0 ) u
v
A+B+C +D

=
(x0 , y0 ) + i (x0 , y0 ) +
z z0
x
x
z z0

f (z) f (z0 ) =

We want to consider what happens as z z0 and we expect to get that

f (z) f (z0 )
= ux (x0 , y0 ) + ivx (x0 , y0 )
zz0
z z0
lim

But to get this, we need to show that

A+B+C +D
0 as z z0
z z0
Consider just
We have that

A
z z0
u
( u
A
y (x, ) y (x0 , y0 )) (y y0 )

=
z z0
z z0
u
( u
y (x, ) y (x0 , y0 )) (z z0 )

z z0
u
u
= (x, )
(x0 , y0 )
y
y

But this goes to 0 as z z0 for the following reason:

z(x,y)
(x,)
z0

(x,y0)

32

We had that [y, y0 ] or [y0 , y] and as z z0 , we also have that x x0 and y y0 , so

as z z0 , y y0 and so the length of the interval Jy 0. And since u
y is continuous at
(x0 , y0 ) by assumption, we have that

u
u
(x, )
(x0 , y0 ) 0
y
y

Thus, the terms

A
gives no contribution as we take the limit
z z0
The proofs for the terms involving B, C and D are similar.
Thus,
f (z) f (z0 )
= ux (x0 , y0 ) + iv(x0 , y0 ) f (z0 ) = ux (x0 , y0 ) + iv(x0 , y0 )
zz0
z z0
lim

Example 17
Is the following function holomorphic?
f (z) =

x2

y
x
i 2
for z =/ 0
2
+y
x + y2

We need to show that the partial derivatives for u and v exist and are continuous and
satisfy the CRE. Here
u(x, y) =
So
ux =
vx =

x
y
and
v(x,
y)
=

x2 + y 2
x2 + y 2

x2 + y 2 2x2
y 2 x2
2yx
=
and
u
=
y
(x2 + y 2 )2
(x2 + y 2 )2
(x2 + y 2 )2

2xy
2xy
x2 + y 2 2y 2
x2 y 2
=
and
v
=

y
(x2 + y 2 )2 (x2 + y 2 )2
(x2 + y 2 )2
(x2 + y 2 )2
So ux = vy and uy = vx The CRE are satisfied

All the partial derivatives are continuous as long as (x, y) =/ 0. We conclude by the
Theorem, that f is holomorphic for z =/ 0
Note that
x iy
z
z
1
f (z) = 2
= 2=
=
2
x +y
z
z.z z
Well eventually prove that f , f exist on if f is holomorphic on and this will
imply that
2u 2u 2v 2v 2u 2v
,
,
,
,
,
x2 x2 x2 y 2 xy xy
exist and are continuous
33

3.5

Harmonic and Conjugate Harmonic Functions. New

Differential Operators

Definition 26 (Harmonic Functions)

We say that the function u R2 R is harmonic if
2u 2u
+
=0
x2 y 2
Example 18
If f is holomorphic, show that u = Rf and v = If are harmonic.
Since f is holomorphic, we can use the CRE. We have that
2u
2v
=
u
=
(u
)
=
(v
)
=
xx
x x
y x
x2
yx
2u
2v
=
u
=
(u
)
=
(v
)
=

yy
y y
x y
y 2
xy
As mixed partials are equal, we have that
2u 2u
+
= 0 u is harmonic
x2 y 2
Similarly,

2u
2v
=
v
=
(v
)
=
(u
)
=

xx
x
x
y
x
x2
yx
2u
2v
=
v
=
(v
)
=
(u
)
=
yy
y
y
x
y
y 2
xy

So again we get that

2v 2v
+
= 0 v is harmonic
x2 y 2

Definition 27 (Conjugate Harmonic)

If u and v are harmonic and satisfy the CRE, we say that they are conjugate harmonic
Example 19
Let u = 2x x3 + 3xy 2
Find its conjugate harmonic. Find f holomorphic s.t. R(f ) = u. Write f as a function of
z.
Part 1: Checking that u is harmonic.
ux = 2 3x2 + 3y 2 uxx = 6x
uy = 6xy uyy = 6x
34

So uxx + uyy = 0 u is harmonic

Part 2: Finding the conjugate harmonic of u.
We are going to find v, the conjugate harmonic of u by using the CRE.
ux = vy = 2 3x2 + 3y 2 v(x, y) = (2 3x2 + 3y 2 )dy = 2y 3x2 y + y 3 + c(x)
Differentiating this w.r.t. x, we get that
vx = 6xy +

d
c(x)
dx

But we also have that

uy = vx = 6xy vx = 6xy
d
d
c(x)
c(x) = 0 c(x) = k
dx
dx
So v(x, y) = 2y 3x2 y + y 3 + k

And thus 6xy = 6xy +

Part 3: Finding f
We already showed in the previous example that for f holomorphic, the functions u and
v s.t. f = u + iv are conjugate harmonic. So the function we are looking for is
f (z) = u + iv
= (2x x3 + 3xy 2 ) + i(2y 3x2 y + y 3 + k) = (2x + i2y) + (x3 + iy 3 ) + (3xy 2 + i(3x2 y)) + ki
= 2z (x + iy)3 + ki = 2z z 3 + ki
Definition 28
v
f (u + iv) u
=
=
+i
x
x
x
x
v
f (u + iv) u
=
=
+i
y
y
y
y
Definition 29

= (
i )
z 2 x
y

= (
+i )
z 2 x
y
If f is holomorphic, then

f
z (z0 )

= f (z0 ).

Example 20
Show that the CRE for f = u + iv are equivalent to

35

f
z

=0

f 1 f
f
1 u
v
u
v
= (
+i )= (
+i
+ i(
+ i ))
z 2 x
y
2 x
x
y
y
v
u v
1 u v
1 v u
1 u
+i
+i
)= (
) + i(
+
)
= (
2 x
x
y y
2 x y
2 x y
1 v v
1 v v
= ( ) + i(

) from the CRE

2 y y
2 x x
=0

36

Chapter 4
Power Series
4.1

Review

Example 21 (Power Series)

A power series from Analysis 1:

zn
n=0

We consider the partial sums of this power series

N

zn =
n=0

1 z N +1
for z =/ 1
1z

We have two possibilities:

If z < 1, then the series converges since lim z n = 0. Then we say that the sum of the
n
series is

1
zn =
1z
n=0
If z > 1, then lim z n = . Therefore, z N +1 is not convergent and so the series diverges.
n

Definition 30
Power series centred at 0 are of the form

an z n where an C, z C
n=0

Definition 31
Power series centred at a are of the form

an (z a)n where a C, z C, an C
n=0

Definition 32
If R is the radius of convergence, then D(0, R) is the disk of convergence for power series
centred at 0 and D(a, R) for power series centred at a.
37

4.2

lim sup and lim inf

Take a sequence < bn > with bn 0. Define

cN = sup {bN , bN +1 , bN +2 , . . .} = sup bn
nN

i.e. we are ignoring the first N 1 terms when finding the supremum.
If < bn > is bounded, then cN R, cN 0
If < bn > is unbounded, then cN =
We also have that
{bN +1 , bN +2 , . . .} {bN , bN +1 , bN +2 . . .} sup{bN +1 , bN +2 , . . .} sup{bN , bN +1 , bN +2 . . .}
So cN +1 cN , i.e. < cN > is a decreasing sequence with numbers 0 or (bounded below
by 0). Therefore, it is convergent, i.e. lim cN exists.
N
Remark: If < bn > is unbounded, then all cN = and lim cN =
N

Definition 33
lim sup bn = lim cN = lim sup bn

Example 22
Find the lim sup of the sequence
bn = 3 +

N nN

(1)n
n

b1 = 3 1
b2 = 3 + 1/2
b3 = 3 1/3,
b4 = 3 + 1/4
b5 = 3 1/5
supn1 bn = b2
supn2 bn = b2 = 3 + 1/2
supn3 bn = b4

1
supn2k bn = b2k = 3 + 2k
1
supn2k1 bn = b2k = 3 + 2k
So we can write < cN >= {3 + 1/2, 3 + 1/4, 3 + 1/4, 3 + 1/6, 3 + 1/6, . . .}. This is a decreasing
sequence and
lim cN = 3 lim sup bn = 3
N

Remark: If lim bn = l, then lim supn bn = l

n
Similarly, we can define
Definition 34
lim inf bn = lim inf bn
n

N nN

38

Remark: If lim bn = l, then lim inf n bn = 1

n

Lemma 1
lim inf

bn+1
bn+1
lim inf n bn lim sup n bn lim sup
bn
bn

Corollary 5

bn+1
= l lim n bn = l
n
n bn
lim

4.3

Important Theorems and Results about Complex

Power Series

Definition 35 (Radius of Convergence with Hadamards formula)

n=0

n=0

The radius of convergence of the power series an (z a)n or an z n is given by

Hadamards Formula
R=

1
lim sup

an

Remark: If lim an = l, then R = 1l .

This is Cauchys Root Test, which says that

If an z n ,then we consider

an z n =

an z lz as n

n=0

And the root test says that

If l < 1, then the series converges
If l > 1, then the series diverges
Note that the sequence < an > may not be a convergent sequence. This is when we use
Hadamards Formula, as the sequence < aN > is not the same as < an > and may be
convergent. So if other tests are no good, use the lim sup and Hadamard. (Hadamards
formula and Cauchys Root test involving lim sup are useful when we want to apply the
root test but have a divergent sequence, i.e. one which might involve something like
(1)n )
Theorem 8
There exists a unique R [0, ] (i.e. we can define R to be 0 or , too) s.t. for the
power series

an z n
n=0

we have
39

 if z < R, then the power series converges absolutely

if z > R, then the power series diverges
Similarly, for the power series

an (z a)n
n=0

if z a < R, then the series converges absolutely

if z a > R, then the series diverges
Remark: It is delicate to find out what happens on z = R for centre at 0, or at
z a = R for centre at a.
Proof. Define
R=

1
lim sup

and for simplicity assume that R =/ 0,

an

Let cN = supnNn an (this is a decreasing sequence).

Let l = lim sup n an = lim cN .
N
Part 1
We need to show that if z C with z < R = 1l l <
absolutely.
We pick > 0 s.t. l + <

1
z

1
z ,

then an z n converges
n=0

(l + )z < 1. Call r = (l + )z, so that r < 1.

Since l = lim cN and l + > l, N s.t. cN < l +

N

So then we have that

cN = sup n an < l + n an < l + , n N an < (l + )n

nN

So now we consider the series

an z n = an zn (l + )n zn = rn
nN

nN

nN

nN

But we showed that r < 1, so the series rn is the geometric series and it converges.
nN

So by the Comparison Test, the series an z n converges, i.e. the series an z n connN

nN

verges absolutely.
And since adding a finite terms (the terms for n < N ) doesnt affect convergence of the
series, we conclude that

an z n converges absolutely
n=0

40

Part 2
If z > R then the series diverges.

We know for the nth term test for convergence that an z n converges < an z n > 0
n=0

as n . So we are going to prove the contrapositive of the statement above, i.e. we are

going to show that the sequence < an z n >

/ 0 an z n diverges. To do this, we will
n=0

find an increasing sequence of natural numbers an1 < an2 < . . . with n1 < n2 < . . . such that
ank z nk > 1
As before, we have that
1
1
1

=
l >
n
l lim sup an
z

z > R =
We can find an > 0 s.t. l >

1
z

1 < (l )z

Now, c1 = sup n an > l

n1

a2 , 3 a3 , . . .}.

n1 s.t. n1 an1 > l

cn1 +1 = sup n an > l

so l is not an upper bound for { a1 ,

nn1 +1

n2 s.t. n2 n1 + 1 with n2 an2 > l

cn2 +1 = sup n an > l

nn2 +1

n3 n2 + 1 s.t.

n3

an3 > l

We continue in this way and deduce that

nk
ank > l
So that

nk

ank > l ank > (l )nk ank z nk > [(l )z]

So by the nth term test, the series diverges.

Example 23
A limit from Analysis last year: revision
lim

1
n
n = lim n n = lim (elog n ) n

= lim e
n

log n
n

n
lim ( log
)
n

= en

41

= e0 = 1

nk

>1

Theorem 9
The power series

an z n for z < R
n=0

defines a holomorphic function in the disk of convergence D(0, R).

Moreover, its derivative is given by the power series

nan z n1
n=1

obtained by differentiating the original series term by term.

Moreover, this series has the same radius of convergence as the original, R. So we have
the holomorphic function and its derivative:

f (z) = an z n
n=0

f (z) = nan z n1
n=1

Remark: The power series an (z a)n are holomorphic inside z a < R

n=0

Proof. The radius of convergence of

1

f (z) = an z n is given by R =
n=0

lim sup

an

Then the radius of convergence of

nan z n1 is given by R =
n=1

lim sup

=
n(an )

1
1

=
=R

lim sup n n n an lim sup n an

So these two series have the same radius of convergence.

Now define g(z) = nan z n , with z < R

n=1

Fix z0 D(0, R), which is the disk of convergence of the power series.
We are aiming to show that f (z0 ) = g(z0 ) i.e, that the second power series is what we
get by differentiating the first.

42

z0

Let z D(z0 , ) for << 1, s.t. D(z0 , ) D(0, R).

Aim: Show that
f (z) f (z0 )
lim
g(z0 ) = 0
zz0
z z0
Label z0 = r < R.
N

n=0

n=0

Define the partial sums SN (z) = an z n ( s.t. SN as n )

Define the tail as EN (z) = an z n ( s.t. EN 0 as n )
n>N

We also define the partial sums and the tail of g(z0 ) as

N

nan z0n1 and nan z0n1 respectively

n=0

n>N

And now we consider

f (z) f (z0 )
SN (z) + EN (z) SN (z0 ) EN (z0 ) N
g(z0 ) =
nan z0n1 nan z0n1
z z0
z z0
n=0
n>N
=(

SN (z) SN z0 N
EN (z) EN (z0 )
nan z0n1 ) + ( nan z0n1 ) + (
)
z z0
z z0
n=0
n>N

We are going to consider all the terms grouped as in the brackets.

Part 1 Consider first nan z0n1 .
n>N

Since z0 < R, the radius of convergence of g(z), the series

nan z0n1 converges, so N1 s.t. N > N1 the tails nan z0n1 <
n=0

n>N

43

Part 2 Consider the terms

We have that

EN (z)EN (z0 )
zz0

EN (z) EN (z0 )
=
z z0

an z n an z0n

n>N

an (z n z0n )

n>N

n>N

=
z z0
z z0
z n z0n
= an (z n1 + z n2 z0 + . . . + z0n1 )
= an
z

z
0
n>N
n>N

Now, we had that z D(z0 , ) and this implies that z z0 < . We also defined that
z0 = r. Combining these two and applying the triangle inequality, we get that
z < r +
So if we consider the general form of each term appearing in the expansion above, we
have that
z ni z0i1 = zni z0 i1 < (r + )ni ri1 < (r + )ni (r + )i1 = (r + )n1
Now plug everything back into the series (with modulus signs) to form an inequality:
an (z n1 z0 + z n2 z02 + . . . + z0n1 ) nan (r + )n1
n>N

n>N

The series

n=0

n=0

nan (r+)n1 converges as (r+) < R , which is the radius of convergence of nan z n1
But we know that the tail of a convergent series 0, so we conclude that
N2 s.t. N > N2 an (z n1 + . . . + z0n1 ) <
n>N

Now take max {N1 , N2 } and fix N > max N1 , N2 , The we have for this N (from Part 1 and
Part 2) that
EN (z) EN (z0 )

< and nan z0n1 <

z z0
3
3
n>N
Part 3 Consider the term

SN (z)SN (z0 )
zz0

We have that

nan z0n1
n=0

n=0

n=1

SN (z) = an z n , which is a polynomial and its derivative is SN

(z) = nan z n1
N
SN (z) SN (z0 )

= SN
(z0 ) = nan z0n1
zz0
z z0
n=1

So lim

So that 1 > 0 s.t. 0 < z z0 < 1

44

SN (z) SN (z0 )

SN
(z0 ) <
z z0
3

The last part of the proof is combining all together, we get that 1 > 0 s.t. 0 < z z0 <
1
f (z) f (z0 )

g(z0 ) < + + =
z z0
3 3 3
Thus

f (z) f (z0 )
lim
= g(z0 ) = nan z0n1
zz0
z z0
n=0

4.4

Examples - Finding the Radius of Convergence of

Power Series

Example 24
Find the radius of convergence of the power series

n2 3n z n
n=0

We use the ratio test:

bn+1
(n + 1)2 3.3n z.z n
3(n + 1)2 z
=

bn
n2 3n z n
n2
3(n + 1)2
=
z 3z as n
n2

So if 3z < 1 z < 31 , we have that the series converges absolutely. If, on the other
hand, 3z > 1 z > 13 , the series diverges. The radius of convergence is R = 31 .
What happens at 13 ?
n
We plug in z = 13 into n2 3n z n and we get n2 3n ( 31 ) = n2 as n .
But we know that if the series converges, then the sequence of the general terms tends to
0, while here at z = 31 tends to . So the series cannot converge for z = 31 . We conclude
that it converges only inside the disk with radius 13 , but not on the circle.
Example 25
What is the radius of convergence of

(3 + (1)n )n z n
n=0

We could try applying the root test, which gives us

n
(3 + (1)n )n z n = (3 + (1)n )z = (3 + (1)n )z
But this diverges as n , since the sequence (3 + (1)n ) diverges. We can see this
by choosing 2 subsequences, one of the odd terms and of the even terms. The fist one
45

converges to 4, while the second one converges to 2.

However, we can apply Hadamards Formula, according to which
R=
where an = (3 + (1)n )n , so
is even and to 2 if n is odd.
For lim sup, we define

1
lim sup

an

an = 3 + (1)n and we showed that this is equal to 4 is n

cN = sup(3 + (1)n ) = 4
nN

lim sup

an = lim cN = lim(4) = 4
N

We conclude that
R=

1
4

Example 26 (The exponential)

Find the radius of convergence of
zn
n=0 n!

ez = exp(z) =
We use the Ratio Test
n+1

z
(n+1)!
z n+1
n!
n!
z
bn+1
= zn = n .
= z
=
0 as n

bn
z (n + 1)!
(n + 1)n! (n + 1)
n!

So the series converges z C. We conclude that R =

Example 27 (Cosine)
Find the radius of convergence of
(1)n z 2n
(2n)!
n=0

cos(z) =
Again we use the ratio test

RR (1)n+1 z2n+2 RR
bn+1 RRRR (2n+2)! RRRR
(2n)!
z

= RRR (1)n z2n RRR = z

=
0 as n
bn
(2n + 2)! (2n + 2)(2n + 1)
RRR
RRR
(2n)!
R
R
So again the radius of convergence is R =
Example 28 (Sine)
Find the radius of convergence of
(1)n z 2n+1
n=0 (2n + 1)!

sin(z) =

46

Ratio Test:
RR (1)n+1 z2n+3 RR
z2
bn+1 RRRR (2n+3)! RRRR
(2n + 1)!
= RRR (1)n z2n+1 RRR = z 2
=
0 as n

bn
(2n + 3)! (2n + 2)(2n + 3)
RRR (2n+1)! RRR
R
R
So R =
Example 29 (Exp)
From the power series definition of ex , we have that
iz (iz)2 (iz)3
(iz)n
=1+ +
+
+ ...
1!
2!
3!
n=0 n!

eiz =
But we also have that

(1)n z 2n+1
(1)n z 2n
+i
(2n)!
n=0 (2n + 1)!
n=0
2
4
6
3
z
z
z
z
z5 z7
= (1 + + . . .) + i (z + + . . .)
2! 4! 6!
3! 5! 7!

eiz = cos(z) + i sin(z) =

We also have that

(1)n z 2n+1
(1)n z 2n
i
(2n)!
n=0 (2n + 1)!
n=0
z2 z4 z6
z3 z5 z7
= (1 + + . . .) i (z + + . . .)
2! 4! 6!
3! 5! 7!

eiz = cos(z) + i sin(z) = cos(z) i sin(z) =

This is how we get Eulers Formulae

eiz + eiz
eiz eiz
and sin z =
2
2i
Another thing we can show about the exponential function is that its derivative is the
same as itself. We have that
cos(z) =

k
z
nz n1 z n1
=
= (by relabelling)
= ez
n=1 (n 1)!
n=1 n!
k=0 k!

(ez ) =

Lemma 2
Let f be holomorphic in a region with f (z0 ) = f (z), z .
Then k C s.t. f (z) = kez , z .
Proof. We are going to show that

f (z)
=k
ez

by differentiating it.

f (z)
f (z)ez (ez ) f (z) f (z)ez ez f (z)
)
=
=
=0
ez
(ez )2
e2z

47

Example 30
Show that
ez+w = ez ew , z, w C
Proof. Fix w C and define f (z) = ez+w . Then f (z) = (z + w) ez+w = ez+w = f (z)
Using the lemma, k C with ez+w = k.ez , z C. Plugging in z = 0, we get
ew = ke0 = k.1 = k
So ez + w = ez .ew
Useful to remember:

i
e 2 = cos( ) + i sin( ) = i
2
2
i
e = cos() + i sin() = 1 ei + 1 = 0
e

i3
2

z+2i

= i
= ez e2i = ez (cos(2) + i sin(2)) = ez

48

Chapter 5
Mapping Properties of Complex
Functions. Conformal Maps
5.1

Mapping properties of ez

For a complex function f (z) C, f C C we get four dimensions which cannot be

plotted easily. The way we plot complex functions is by introducing a z-plane and a
w-plane. Then for z in the z-plane, we compute using f (x + iy) = ex (cos y + i sin y) the
value of w which we represent on the w-plane. We plot the following:
The z = 0 value gives ez = 1
Horizontal lines of the type z = x + iy0 : We get f (z) = ex (cos y0 + i sin y0 ), which has
a fixed argument y0 and a variable length as x varies. Hence, horizontal lines are
mapped onto rays emanating from the origin with angle y0 .
Vertical lines of the type z = x0 + iy: We get f (z) = ex0 (cos y + i sin y), which has a
fixed modulus x0 and a variable argument. Hence, vertical lines are mapped onto
circles centred at the origin with radius ex0 .

x+iy0

w=f(z) =exp(z)

exp(x0)

x0+iy
49

v
0

y0

We notice that a grid system of horizontal and vertical lines maps to a series of circles
and rays (centred and the origin and emanating from the origin). Thus the picture in
the w-plane is a grid in polar coordinates. Thus orthogonality of the original vertical and
horizontal lines is preserved (except at the origin).

Note that f (z) = ez is not injective in general because ez+2i = ez . However, if we pick y0 ,
y1 with y0 y1 < 2, then it is injective. Consider the following proof:
Suppose that ez1 = ez2 with z1 = x1 + iy1 and z2 = x2 + iy2 . Then ez1 = ex1 and ez2 = ex2 .
Then ez1 = ex1 (cos y1 + i sin y1 ) = ez2 = ex2 (cos y2 + i sin y2 ) ex1 = ex2 x1 = x2
and y1 = y2 + 2k. Hence, if y2 y1 < 2 we have that y1 = y2 and thus the function is
injective.
In fact, we have that ez maps bijectively {z C 0 < I(z) < } in the z-plane to {w
C I(w) > 0}, i.e. the horizontal strip of width 2 is mapped onto the upper half plane
bijectively. This is the hyperbolic upper half plane.
y

5.2

w=exp(z)
x

v
u

Properties of log(z)

Since ez is not injective over the whole complex plane, its inverse may not exist. However,
if we restrict domains, then it is possible that is does.
Given w C/{0}, we find all solutions of ez = w. To do this we let w = wei . Then
ez = w ez = w(cos() + i sin()) ex (cos y + i sin y) = w(cos() + i sin())
ex = w and y = + 2k for k Z
z = log w + i arg(w)
50

which is not unique in its imaginary part as it is determined up to 2k. This is why we
define the principle logarithm.
Definition 36 (Principle Logarithm)
The principle branch of the complex logarithm is
Log(w) = log w + iArg(w)
With this definition, the principle logarithm is the inverse of exp(z)
Remark: The principle branch of the logarithm is not continuous on {z z R 0},
where Arg(w) = .
The function Log(w) is holomorphic on C/{z, z 0}
Naturally, being the inverse function, log reverses the action of exp. So we get

5.3

Mapping Properties of w = z 2.

Here we have
z 2 = (x2 y 2 ) + 2ixy u(x, y) = x2 y 2 and v(x, y) = 2xy

5.3.1

Plotting Level Curves

To visualise this, we are going to use the technique of level curves.

Fix u = c. We get that u(x, y) = x2 y 2 = c. Taking different values for c give
us different rectangular hyperbolae. If c = 0, we get two lines from the origin. So
rectangular hyperbolae in the z-plane are mapped onto lines u = c, i.e. vertical lines
in the w-plane.
Fix v = k. We get that v(x, y) = xy = k. Taking different values for k give us different
hyperbolae. If k = 0, we get the origin. So hyperbolae in the z-plane are mapped
onto lines v = k, i.e. horizontal lines in the w-plane.
51

We claim that in the z-plane our level curves from the sets u = c, v = k intersect orthogonally. (i.e. their tangents at the points of the intersection are perpendicular). We can
prove this:
Let x2 y 2 = c and 2xy = k. Suppose y is a function of x. Then differentiating w.r.t. x,
we get
y
2xy = k 2y + 2xy = 0 y =
x
x
2
2

x y = c 2x 2yy = c y =
y
x0
Hence, if (x0 , y0 ) is a point of intersection, the slopes are y0 and xy00 . So the product of
the slopes is 1 and thus the curves are orthogonal.

5.3.2

Plotting backwards

We can also reverse our process, i.e. we can look at what fixed shapes in the z-plane give
in the w-plane.
The circles z = rei . Then f (z) = z 2 = (rei )2 = r2 [cos(2) + i sin(2)]. So we get a
circle again but this time its radius is r2 rather than r and the circle is traced at
twice the angular speed of the original one, i.e. going around the original circle once
corresponds to going around the new circle twice.
The rays with angle for a given value of . We get that f (z) = f () = 2 . So we
get another ray in the same direction but with increased size.
The shaded sector in the w-plane is subtended by an angle twice the size of the
angle in the z-plane.
The general problem: Given two regions G and , is there a holomorphic function
f G which is bijective and f 1 G is holomorphic. If it exists, find it.
Definition 37 (Path/Curve)
A path or a curve in a region is a continuous function [a, b] C i.e. (t) = x(t)+iy(t)
where x [a, b] R and y [a, b] R are continuous functions and x(t) + iy(t)
If x and y exist t [a, b], then we say that is smooth.
We define (t) = x (t) + iy (t).
Take a curve parametrised by t and find a fixed point t0 . Then (t0 ) = x (t0 ) + iy (t0 ).
0)
. So we can
Then the vector (t0 ) is parallel to he tangent line at at t0 , lim (t)(t
tt0
tt0

write the equation o f the tangent line. We know that arg( (t0 )) measures its inclination,
i.e.
y (t0 )
slope of tangent line =
= tan(arg( (t0 )))
x (t0 )
Now we take this curve and apply to it a mapping w = f (z). Then the curve becomes
c(t) = f ((t)) for t [a, b]
Assume that f is holomorphic and we can write it as f = u + iv. Then
c(t) = f ((t)) = u((t)) + iv((t)) = u(x(t), y(t)) + iv(x(t), y(t)) for (t) = x(t) + iy(t)
52

d
u dx u dy
v dx v dy
dc d
= u(x(t), y(t)) + i v(x(t), y(t)) =
+
+ i(
+
)
dt dt
dt
x dt y dt
x dt y dt
u
v dx
u
v dy
dx
v
u dy
=(
+i )
+(
+i )
= f (z0 ) + (
+i )
x
x dt
y
y dt
dt
x
x dt
dy
dx
= f (z0 ) + if (z0 ) = f (z0 )(x (t0 ) + iy (t0 ))
dt
dt
= f (z0 ) (t0 )
Then we can find that
c (t0 ) = arg[f (z0 ) (t0 )] = arg( (t0 ))

5.4

Conformal Maps

Theorem 10
Given a complex function f s.t. f is holomorphic at z0 and f (z0 ) =/ 0, then f preserves
angles at z0 .
This means that if 1 and 2 are two smooth curves through z0 and f (1 ), f (2 ) are the
images through f (z0 ), then the angle between 1 and 2 is the same as the angle between
f (1 ) and f (2 ).
Definition 38 (Conformal Map)
If f G preserves angles and
f (z) f (a)
za
z a

lim

exists a , then f is called a conformal map.

Remark 1: If f is holomorphic on and f (z) =/ 0 on , then f is a conformal map. Note
that this is the criterium we normally use to check if a map is conformal, rather than the
official definition which involves taking the limit.
Remark 2: Domain restrictions can impact mappings. For example, consider again the
mapping f (z) = z 2 . We have that f (z) = 2z and so f (z) =/ 0 for z =/ 0. Thus, f is
conformal of C/{0} but not on the whole of C.
Example 31 (Mapping properties of f (z) = z1 )
Consider the function f (z) = z1 . State the domain over which it is holomorphic and show
that it maps circles and lines to circles and lines.
1
We have that f C {} C {}, f (z) = z1 is bijective if we set 10 = and
= 0.
We can write f (z) as f = z 1 ei .
If z = 1, we get f (z) = ei f (z) = 1.
If z < 1, we get f (z) > 1
If z > 1, we get f (z) < 1
53

So everything inside the circle of radius 1 in the z-plane is mapped onto the exterior of
the circle of radius 1 in the w-plane and vice versa. The circle itself is mapped onto the
same circle.
y

f=1/z

Now, circles in the z-plane are given by the equation

a(x2 + y 2 ) + bx + cy + d = 0 where a, b, c, d R for a =/ 0
In fact all lines and circles are of this form (see discussion below).
We want to turn this equation into one in variables z and z. We have:
z+z
x2 + y 2 = z2 = zz. x = R(z) = z+z
2 and y = I(z) = 2i . So we can write the above equation
as
z+z
z+z
+c
+d=0
azz + b
2
2i
We now want to find the equation of this circle in the z-plane in the w-plane, i.e. after
we have applied to it the function w = f (z) = z1 . So we write w = z1 z = w1 and then
the equation becomes:
a

b 1 1
c 1 1
w+w
1
ww
+ ( + ) + ( ) + d = 0 a + b
+c
+ dww = 0
ww 2 w w
2i w w
2
2i

Since u =
is

w+w
2

and v =

ww
2i

in the w-plane, where w = u + iv we get that the final equation

a + bu cv + d(u2 + v 2 ) = 0

which is of the same form as the equation we had for the lines and circles in the z-plane.
We conclude that f = z1 maps lines and circles onto lines and circles.
The full classification is as follows:
a(x2 + y 2 ) + bx + cy + d = 0
Circle not passing through (0, 0) a, d =/ 0 is mapped to circle not passing
through (0, 0)
Circle passing thought (0, 0) a =/ 0, d = 0 is mapped to line not passing through
(0, 0)
54

 Line not passing through (0, 0) a = 0, d =/ 0 is mapped to circle passing through

(0, 0)
Line passing through (0, 0) a, d = 0 is mapped to line passing through (0, 0)
Note how lines arent necessarily mapped onto lines and circles arent necessarily mapped
onto circles.
Example 32
Let w = f (z) w = az + b, where a =/ 0. Show that w is a bijection f C C and that is
maps lines to lines and circles to circles.
Injective:
f (z1 ) = f (z2 ) az1 + b = az2 + b z1 = z2 , for a =/ 0
Surjective:
Given w C we solve w = az + b z =

wb
which is uniquely determined
a

So the function is bijective.

To show that lines are mapped to lines and circles are mapped to circles, we write a = rei
, so that r = a and = arg(a). Then w = rei z + b. Comparing z with rei z + b, we see
that all points are only rotated and scaled and translated. Thus lines are mapped to lines
and circles to circles.

5.5

Linear Fractional Transformations

We are going to analyse solution to

f (z) = w =

az + b
cz + d

for ad bc =/ 0 Such functions are called linear fractional transformations.

Case 1: c = 0, d =/ 0. Then
az + b
a
b
w=
= ( )z +
d
d
d
This is in the same form as the linear function we considered in the example above, so we
know that it maps lines to lines and circles to circles.
Case 2: c =/ 0.
We are going to first show that w = f (z) is not a constant function.
w=

b
d
az + b a z + ab a z + dc dc + ab a
a
1 ad + dc
a c
=
=
=
(1
+
) = (1 +
)
d
d
d
cz + d c z + c c
c
c
ac z + dc
z+ c
z+ c

Since ad bc =/ 0, the fractional term appears and it is dependent on z, and hence the
function is not constant.
Note that the function is not defined for z = dc .
55

We are now going to show that it is bijective and find the inverse function.
Using the relation
az + b
w=
cz + d
we solve for z to get
w(cz + d) = az + b cwz az = b dw z =

dw + b
cw a

so we get a unique solution for w =/ ac . Hence, w = f (z) is injective.

We define f ( dc ) = and f () = ac . Then f is defined for all C {}, i.e. f C {}
C {} is surjective.
Thus w = f (z) is bijective and its inverse is given by
f 1 (w) =

dw c
dw + c
=
cw a
cw + a

Example 33
Let f and g be linear fractional transformations given by
w = f (z) =

az + b
a w + b
and g(w) =
cz + d
c w + d

Where ad bc =/ 0 and a d b c =/ 0.
Compute g f and show that it is a linear fractional transformation.
We have that
(g f )(z) = g(f (z)) =

a az+b
a (az + b) + b (cz + d) (a a + b c)z + (a b + b d)
cz+d + b
=
=

c (az + b) + d (cz + d) (c a + d c)z + (c b + d d)

c az+b
cz+d + d

c
/0
which is a linear fractional transformation for cc a+d
b+d d =
Note: If we represent f by the coefficient matrix

a b
F =[
]
c d
and g by the coefficient matrix
a b
G = [ ]
c d
Then g f is represented by GF .
Linear fractional transformations form a group. Also, det(GF ) = det(G)det(F ) = (a d
b c )(ad bc) =/ 0
Linear fractional transformations maps lines and circles to lines and circles. We can
see this as follows: We write the linear fractional transformation in the form
w=

az + b a
1 bc ad
= (1 +
)
cz + d c
ac z + dc
56

Then
zz+

d
= Z , which is a linear map
c

1
= W , which is a transfomation we have previously looked at
Z
a 1
W + 2 (bc ad)W , which is a linear map
c c
Hence, linear fractional transformations are compositions of up to 3 known transformations which map lines and circles to circles and lines. Hence, they do the same.
Z

Definition 39 (fixed point)

If f is a map and f (z) = z, then z is called a fixed point of f .
Example 34
Find all the fixed points of the linear fractional transformation w =
We are looking for z s.t.
az + b
=z
cz + d
Case 1: c = 0

az+v
cz+d ,

where adbc =/ 0.

b
a
b
a
z + = z ( 1) z + = 0
d
d
d
d
If

a
d

b/d
=/ 1, then we have a unique solution z = a/d1

If ad = 1 and b = 0, then 0 + zb = z. This is always true, i.e. z C, z is a fixed point

a
and we have w = az+0
0z+d = d z = z this is the identity transformation.
If

a
d

= 1 and b =/ 0, then there is no solution, i.e. no fixed points.

Case 2: c =/ 0.
az + b
= z az + b = z(cz + d) = cz 2 + dz 0 = cz 2 + (d a)z b
cz + d

(d a) (d a)2 + 4bc
z =
2c
So w has up to 2 fixed points.
The conclusion is that if f is a linear fractional transformation and f =/ Id, it has at most
2 fixed points.
Corollary 6
If f is a linear fractional transformation and f (z1 ) = z1 , f (z2 ) = z2 , f (z3 ) = z3 , with
z1 , z2 , z3 distinct, then f (z) = z, z C.
Example 35
Given z1 , z2 , z3 C distinct and w1 , w2 , w3 C distinct, find all linear fractional transformations T with T (zi ) = wi
57

Consider the following transformation T . We claim that it is the one we are looking for.
ww1
ww2
w1 w2
w1 w3

zz2
zz3
z1 z2
z1 z3

w w1 z1 z2 z z2 w1 w2
.
=
.
w w2 z1 z3 z z3 w1 w3

(w w2 )(z1 z2 )(z z3 )(w1 w3 ) = (z z2 )(w1 w2 )(w w3 )(z1 z3 )

Plugging in z = z3 into the LHS, we get 0, so the RHS must also be 0. But since zi
are distinct, it must be true that if z3 = 0, then z2 =/ 0 and z1 =/ 0 and since w1 =/ w2 ,
the only term in the RHS that can be 0 is (w w3 ). So we conclude that w = w3 ,
i.e. z = z3 w = w3 .
Plugging in z = z2 into the LHS, we get 0, so the RHS must also be 0. Since w1 =/ w3
and the zi are also distinct, the only term that we can get to be 0 from the RHS is
(w w2 ). So we conclude that z = z2 w = w2
Plugging in z = z1 in the LHS, we get 0. We get
(w w2 )(z1 z2 )(z1 z3 )(w1 w3 ) = (z1 z2 )(w1 w2 )(w w3 )(z1 z3 ) = 0
w1 w2
(w w2 )(w1 w3 ) = (w1 w2 )(w w3 ) w w2 =
(w w3 )
w1 w3
This is linear, so it has only one solution. This is true for w = w1 , which is the
unique solution. So once again we get that z = z1 w = w1
Note: T is unique. Suppose that it wasnt and that we could find another transformation
S(zi ) = wi .
Consider S T 1 (wi ) = S(T 1 (wi )) = S(wi ) = wi . So S T 1 has three fixed points and it
is a linear fractional transformation S T 1 = Id S = T .
Example 36
Determine all linear fractional transformations f H D(0, 1), where H = {z C I(z) >
0} and D(0, 1) = {w C w < 1}.
Solution:
We are looking for a transformation of the form f (z) = az+b
/ 0.
cz+d with ad bc =

H
D(0,1)

f(l)

We are looking for linear fractional transformations of the form f (z) = az+b
/ 0.
cz+d with adbc =
First of all H is bounded by the real axis, call it l. Thus by the Inertia principle and since
58

we are looking for a bijective mapping, as l.f.t. are continuous, we deduce that f (l) is a
circle, which is the boundary of the disk D(0, 1), the unit circle.
Then we consider the behaviour of the mapping at the points z1 = 0, z2 = 1, z3 = . to
find restrictions of a, b, c, d.
f (0) =

a.0+b
c.0+d

b
d

Since 0 l, f (0) = 1 db = 1 b = d

f () = ac . Since l, f () = 1 ac = 1 a = c.
But we have that ad bc =/ 0, a =/ 0, and c =/ 0, b =/ 0 and d =/ 0. The modulus of ac is
1, so we can write it is complex form as ac = cos() + i sin() = ei . Thus we have
z + ab
a z + ab
i
=e
f (z) =
c z + dc
z + dc
For convenience, label z0 = ab and z1 = dc (note that this z1 is different from the
one above.) Then we have
z z0
f (z) = ei
z z1
0

Since 1 l, we have f (1) = ei 1z

1z1 = 1
From this we can deduce the following:

1z0
1z1

= 1 1 z0 = 1 z1

1 z0 = 1 z1 1 z0 2 = 1 z1 2 (1 z0 )(1 z0 ) = (1 z1 )(1 z1 )
(1 z0 )(1 z0 ) = (1 z1 )(1 z1 ) 1 z0 z0 + z0 2 = 1 z1 z1 + z1 2
(z0 + z0 ) + z0 2 = (z1 + z1 ) + z1 2
But z0 = ab = dc = z1 z0 2 = z1 2 . So we get that
(z0 + z0 ) = (z1 + z1 ) R(z0 ) = R(z1 )
From here we have that R(z0 ) = R(z1 ) and z0 = z1 z0 = z1 or z0 = z1 . But
z0 =/ z1 , so we deduce that z0 = z1
The final answer is thus
f (z) = ei

59

z z0
z z0

Chapter 6
Integration Along Curves (Contour
Integration)
6.1

Curves: Definitions

Definition 40 (Parametrised Curve)

A parametrised curve is a continuous map
[a, b] C
given by
(t) = x(t) + iy(t)
such that the functions
x [a, b] R and y [a, b] R are continuous
(a) is the initial point and (b) is the final/end/terminal point
Definition 41
The reverse curve for [a, b] C is [a, b] C such that
(t) = (a + b t)
So t a + b t is a linear function
a a + b a = a and b a + b b = a
So [a, b] [a, b]
Definition 42 (Simple Curve)
A curve is called simple is its image doesnt cross itself. i.e.
If s < t and (s) = (t) then s = a, t = b
(so the curve may meet itself at the starting or terminal point only, e.g. like a circle)
60

crosses

doesnt cross

doesnt cross

Definition 43 (Closed Curve)

The curve is closed if (a) = (b)
Definition 44 (Simple Closed)
The curve is simple closed if it is simple and closed.
Definition 45 (Smooth Curve)
The curve is smooth if (t) = x (t) + iy (t) exists and is continuous.
Remark: At a and b we use one-sided derivatives
(a) = lim
h0

(a + h) (a)
h

(b + h) (b)
h0
h

(b) = lim

Definition 46 (Path, Piecewise Smooth)

The curve is a path if it is piecewise smooth, which means that is continuous and
there exist points a0 , a1 , a2 , . . . , an with
a = a0 < a1 < a2 < . . . < an1 < an = b
such that [ai , ai+1 ] C is smooth for i = 0, 1, 2, . . . , n 1.
So consider the following:

The derivatives from the left and right at the corners are not the same so is not differentiable there. So it is not smooth. However, it is piecewise smooth.

61

Definition 47 (Contour)
The curve is a contour if it is a simple closed path.
Example 37 (Parametrising a circle)
Consider the circle with positive orientation.
Cr (z0 ) = {z C z z0 = r}
It is parametrised by
z(t) = z0 + r(cos t + i sin t) = z0 + reit where 0 t 2
The same curve with negative orientation is parametrised by
z(t) = z0 + reit with 0 t 2

6.2

Important definitions and useful results for later on

Definition 48 (Length of a curve)

The length of the curve (t) = x(t) + iy(t) with a t b is given by
b

L() =

(x (t))2 + (y (t))2 dt = (t) dt

Theorem 11 (Jordans Curve Theorem)

Let be a contour which is closed and simple.
Consider the complement of in C, i.e.
C {z z = (t); t [a, b]}
Then C is the union of two disjoint regions, 1 and 2 , with 1 bounded and 2 unbounded.
C = ([a, b]) 1 2
with 1 the interior of and 2 the exterior of

inisde the contour

interior
outside the
contour
exterior

62

Definition 49
Let F [a, b] C s.t. F (t) = U (t) + iV (t) with U, V [a, b] R
Assume that U and V are piecewise smooth.
b

F (t)dt = U (t)dt + i V (t)dt

a

Obviously,
b

R F (t)dt = U (t)dt = R(F (t))dt

a
a
a
We have that

(F (t) + G(t))dt = F (t)dt + G(t)dt

a

If c = + i, then
b

cF (t)dt = c F (t)dt
a

Proof. of the last part

First of all
cF (t) = ( + i)(U (t) + iV (t)) = (U (t) V (t)) + i (V (t) + U (t))
So that

cF (t)dt = U (t) V (t) + i V (t) + U (t)

a

On the other hand

b

c F (t)dt = ( + i) U (t)dt + i V (t)dt

a

a
a
b

= U (t)dt V (t)dt + i U (t)dt + V (t)dt

a
a

a
a
b

= U (t) V (t) + i V (t) + U (t)

a

We conclude that

c F (t)dt = cF (t)dt
a

63

Lemma 3 (Very Important)

RRR b
RRR
b
RRR
RRR
RRR F (t)dtRRR F (t)dt
RRR a
RRR a
Proof.
b

Let I = F (t)dt C
a

The polar form of I is I = Iei .Then we have that I = Iei . So

RRR b
RRR
b
b
RRR
RRR
i
i
i
RRR F (t)dtRRR = I = Ie = e F (t)dt = e F (t)dt
RRR a
RRR
a
a
Where the last integral is real since we took the modulus sign at the beginning. So
b

R ei F (t)dt = ei F (t)dt
a
a
RRR
b
RRRR b
R
i
i
R
R
= R (e F (t)) dt RR R (e F (t)) dtRRRR
RRR
RRR
a
R
Ra
b

b
i

R(e

F (t)) dt = ei F (t) dt

a
b

= F (t)dt since ei = 1
a

Note: We got the last step using the fact from Analysis 2:
RRR b
RRR
b
RRR
RRR
RRR f (x)dxRRR f (x)dx
RRR a
RRR a

In the following we are assuming that:

C is a piecewise-smooth curve given by
z(t) = x(t) + iy(t) with a t b
f is a function defined on an open set U containing the image of C.
f z is piecewise continuous on [a, b]

64

Definition 50
b

f (z)dz = f (z(t))z (t)dt

a

Note the following

b

f (z) = u(x, y) + iv(x, y) f (z)dz = [(u(x(t), y(t)) + iv(x(t), y(t))) . (x (t) + iy (t))] dt
a

C
b

= [u(x(t), y(t))x (t) v(x(t), y(t))y (t)] dt

a
b

+ i [u(x(t), y(t))y (t) + v(x(t), y(t))x (t)] dt

a

udx vdy + i udy + vdx

C

6.3

Properties of Line Integrals

Proposition 9
Let be a parametrised curve and z(t) = x(t) + iy(t) with a t b. Denote the reverse
curve by = z(a + b t) = z(t), with a t b. Then
f (z)dz = f (z)dz

Proof.
b

f (z)dz = f (z(t))z (t)dt

a

d
= f (z(a + b t))z (a + b t) (a + b t)dt = f (z(a + b t))z (a + b t)(1)dt
dt

Now substitute u = a + b t so that du = dt and at t = a, u = b and t = b, u = a. So we

have
u=a

f (z(u))z (u)(1)(1)du = f (z(u))z (u)du

u=b

b
b

= f (z(u))z (u)du = f (z(t))z (t)dt by ranaming the parameter

= f (z)dz

65

Proposition 10
Let 1 and 2 be two parametrised curves with 1 (b) = 2 (a), i.e. 2 follows 1 . Then we
have the curve 1 2 , which is denoted by 1 + 2 and
f (z)dz = f (z)dz + f (z)dz

1 +2

Proposition 11
If f (z) M on the curve , i.e.
f (z(t)) M t [a, b] then
RRR
RRR
RRR
RR
RRRR f (z)dz RRRRR M L()
RRR
RRR
Proof. Consider
RRR RR b
RRR
RRR
RRR RRR
RRR
R

RRR f (z)dz RRR = RRR f (z(t))z (t)dtRRRR

RRR
RRR RRR a
RRR
R
R R
R
b

f (z(t))z (t)dt M z (t)dt

a
b

= M z (t)dt = M L()
a

Proposition 12 (Comparing Parametrisations)

.

(r)
r

z()

66

 [c, d] [a, b] is bijective with continuous partial derivatives and (r) > 0
(c) = a, (d) = b and t = (r) dt = (r)dr
since is bijective and increasing, z parametrises the curve with the same orientation.
The new parametrisation is
z [c, d] C
Now
b

f (z)dz = f (z(t))z (t)dt

a
d

= f (z((r)))z ((r)) (r)dr with the cahnge of variables t = (r)

c
d

= f ((z )(r))((z ) (r))dr

c

Example 38
Compute the integral
I = z 2 dz
C

where C is the curve

2+i

2 A

We need to give a parametrisation to the curve:

We can actually find easily the equation of C, it is y = x2 . So we can parametrise by
z(x) = x + i

x
for 0 x 2
2

Or we can set x = 2t so that 0 x 2 0 t 1. Then y =

parametrise by
z(t) = 2t + it

67

x
2

= t and then we can

We are going to use the second parametrisation.

z(t) = 2t + it z (t) = 2 + i where 0 t 1
f (z) = z 2 = x2 y 2 + 2ixy = (2t)2 t2 + 2i.2t.t = (4t2 t2 ) + i4t2 = 3t2 + 4it2
f (z(t)).z (t) = (3t2 + 4it2 )(2 + i) = t2 (3 + 4i)(2 + i)
So we have that
1
2

1
2

2
z dz = t (3 + 4i)(2 + i)dt = (3 + 4i)(2 + i) t dt
0

1
1
= (3 + 4i)(2 + i)[t3 ]10 = (3 + 4i)(2 + i)
3
3
1
i
2
11
= (6 4) + (3 + 8) = + i
3
3
3
3
Now consider the same integral evaluated on the horizontal segment [0, A], called 1 and
the vertical segment [A, B], called 2 . We are going to show that we get the same result
since C = 1 + 2 i.e. well show that
2
2
2
z dz = z dz + z dz

Consider first 1 .

z(x) = x + 0i = x where 0 x 2
f (z(x))z (x) = x2 .1 = x2

Then

1 3 8
2
z dz = x dx = 3 2 = 3
2

Consider now 2 .

z(y) = 2 + iy where 0 y 1 and z (y) = i

Then

1
2

2
z dz = (2 + iy) .idy = (4 + 4iy y )idy
2

0
1

= (4i 4y iy 2 )dy = i ([4y

0

y2 1
y3 1
] + [4i ] )
3 0
2 0

1
1
11
11
+ 4i ) = i ( + 2i) = i 2
3
2
3
3
The sum of the two integrals gives us
= i (4

8
11 2
11
2+i = +i
3
3 3
3
which is what we got before.
68

Example 39 (Extremely Important Result!)

Let C be the circle Cr (0) traversed anticlockwise. Then

0 if n =/ 1
z dz =

2i if n = 1
C
n

Proof. We parametrise the circle by

z(t) = reit with 0 t 2
So that z (t) = ireit .
Then f (z(t))z (t) = (reit )n .reit .i and so
2
it n

it

n+1 nit it
f (z)dz = 2(re ) re idt = r e e .idt
0
2

= r

0
2
n+1 (n+1)it

idt = r

n+1

i e(n+1)it dt

Case 1 n = 1.
Then we have that n + 1 = 0 and
2

n+1

i e

(n+1)it

2
0

dt = r i e dt = 1.i. 1 = 2i

Case 2 n =/ 1.
2

n+1

i e(n+1)it dt = rn+1 i
0

rn+1
rn+1
1
2
[e(n+1)it ]0 =
(e2i(n+1) e0 ) =
(1 1) = 0
i(n + 1)
(n + 1)
(n + 1)

Example 40
Find
zdz
C

over

-1

-C

69

We parametrise the curve as

z(t) = eit for 0 t

z (t) = ieit

This is the reverse of the given curve, so we consider the integral with a - sign.

zdz = z(t)z (t)dt = (eit ) . (ieit ) dt

= idt = i 1dt = i[t]0 = i

0

Theorem 12 (Greens Theorem)

Let C be a simple closed curve bounding a region R. Let C be traversed anticlockwise.
Let P (x, y) and Q(x, y) be continuous functions on C and R, with continuous partial
derivatives. Then
Q P
P dx + Qdy = ( x y ) dxdy
C

6.4

(T) Cauchys Theorem

Theorem 13 (Cauchy-Gousat (or just Cauchy))

If f is holomorphic on a simple closed curve C and its interior, then
f (z)dz = 0
C

Remark: No assumptions are made on the continuity of f (z) or ux , uy , vx , vy .

Proof. Let f (z) = u(x, y) + iv(x, y) , where u = R(f ) and v = I(f ). Also, dz = dx + idy .
So we have that
f (z)dz = (u + iv)(dx + idy) = (udx vdy) + i(udy + vdx)
C

= (udx vdy) + i (udy + vdx) to which we now apply Greens Theorem to get
C

v u
v u
= (

) dxdy + i ( +
) dxdy
x y
y x
R
R
= (vx uy )dxdy + i (vy + ux )dxdy
R

But we are assuming that f is holomorphic, hence the CRE hold. So we have that
ux = vy uy + ux = 0
uy = vx vy ux = 0
70

And so the integrals become

R 0dxdy + i R 0dxdy = 0
Thus
f (z)dz = 0
C

6.5

(T) The Fundamental Theorem of Line Integrals

and an Important Corollary

Definition 51 (Antiderivative)
Let f be continuous on a region and assume that there exists a holomorphic function
F on with F (z) = f (z).
Then F is called the antiderivative or primitive of f .
Theorem 14 (The Fundamental Theorem of Calculus of Line Integrals)
If f is continuous on a region , F is an antiderivative of f in and is a path joining
the two points z1 , z2 , then
f (z)dz = F (z2 ) F (z1 )

i.e. the value of the integral is independent of the path chosen and depends only on the
endpoints.
Proof. Let be parametrised by z(t) for a t b. Assume that is smooth, i,e, that
z (t) is continuous.
b

f (z)dz = f (z(t))z (t)dt by definition

a
b

= F (z(t))z (t)dt
a
b

=
a

dF
(z(t))dt by the Chain Rule
dt
b

= [F (z(t))]a = F (z(a)) F (z(b))

= F (z2 ) F (z1 ) since z(a) = z1 and z(b) = z2

71

We made an assumption that the curve is smooth. What if it was piecewise smooth?
It turns out that the result still holds.
Suppose the piecewise smooth curve consists of the curves C1 , C2 , . . . , Cn such that the
curves are connected at the corners a = a0 , a1 , a2 , . . . , an = b.

z0 = z(a0) =z(a)

zn = z(an) =z(b)
Cn

C3

C1

C2

Then we have that

f (z)dz = f (z)dz + f (z)dz + . . . + f (z)dz

C1

C2

Cn

= [F (z(a1 )) F (z(a0 ))] + [F (z(a2 )) F (z(a1 ))] + . . . + [F (z(an )) F (z(an1 ))]

= F (z(an )) F (z(a0 )) since all the other terms in this telescopic sum cancel out
= F (z2 ) F (z1 )
as before.
Corollary 7
If is closed inside the region and f is continuous on and has an antiderivative F
on , then
f (z)dz = 0

Example 41
We already showed that

Cr (0)

1
dz = 2i
z

Since the curve we are integrating is closed and we dont get 0, the integrand doesnt have
an antiderivative on C/{0}
But what about Log(z), the principle logarithm, given by Log(z) = log z+iArg(z), where
< Arg(z) ?
This is actually not an antiderivative, since the definition of antiderivative requires it
to be differentiable, and it is not even continuous everywhere. In particular, Log is not
continuous on the negative real axis, and it requires a cut there.
72

Then what about another log? For example, consider

log(z) = log z + iArg(z) with

< Arg(z)
2
2

Now this requires a cut on the negative imaginary axis, since this new log is not continuous there. In fact, not matter how we define the complex logarithm, it will always require
a cut, and therefore, we cannot find an antiderivative of the function z1 in any disk around
the origin.
Example 42
Suppose we wanted to evaluate the same integral on another curve, like the curves C1 and
C2 . Then
1
1
z dz = i and z dz = i
C1

C2

So the integral depends on the curve we have chosen.

log creates discontinuities, since the argument cant be made continuous around the circle.
On the other hand, By Cauchys Theorem,
f (z)dz = 0

So consider a similar example, but with a function which satisfies the conditions of the
theorem and a curve, which is not a circle, but still closed.
we take
1 , 2 to have the same endpoint but we are traversing them in the opposite direction, so
f (z)dz = f (z)dz
1

Now suppose that the anticlockwise curve goes along 1 and comes back along 2 . Call
this curve . We have that

z2

z1 -2
73

 = 1 + (2 )
And then Cauchys Theorem gives
0 = f (z)dz = f (z)dz + f (z)dz

= f (z)dz f (z)dz
1

=0
Back to the example with z1 , we have that
= C1 + (C2 )
1
1
1
z dz = z dz z dz = i (i) = 2i

6.6

C1

C2

(T)Gousats Theorem

Theorem 15 (Gousats Theorem)

Let be a triangular contour with its interior contained in some domain U and let f (z)
be holomorphic on U . Then
f (z)dz = 0

Proof
Setting
Let ABC be the triangle with contour (0) , i.e. (0) is the curve ABC
Call ABC T (0) .
Let d(0) be the diameter of T (0) , i.e. the length of the longest side of ABC.
Let p(0) be the perimeter of ABC.
Bisect each side of the triangle ABC, such that AD = DC, BF = F C, AE = EB, as
shown. Connect the points E, D, F to form a triangle with vertices the points E, D, F .
74

A
E

Then we have that

ED = 12 BC, DF = 12 AB, EF = 12 AC
Now, the lines ED, DF and EF divide the original triangle into 4 smaller triangles. Call
(1)
them Tk , with k = 1, 2, 3, 4. We have
(1)
(1)
T1 with boundary 1
(1)
(1)
T2 with boundary 2
(1)
(1)
T3 with boundary 3
(1)
(1)
T4 with boundary 4
(1)
(1)
i.e. Tk has boundary k for k = 1, 2, 3, 4
Step 1

A
E

T(1)1

T(1)2

T(1)4

T(1)3

We are going to find the contour integral around each of the 4 new triangles we now
have.
Around ADE
f (z)dz = f (z)dz + f (z)dz + f (z)dz
(1)

DA

AE

ED

Around CDF
f (z)dz = f (z)dz + f (z)dz + f (z)dz
(1)

CD

DF

75

FC

Around BF E
f (z)dz = f (z)dz + f (z)dz + f (z)dz
(1)

BF

FE

EB

Around DEF
f (z)dz = f (z)dz + f (z)dz + f (z)dz
(1)

DE

EF

FD

Now, we have that

f (z)dz = f (z)dz
DE

ED

f (z)dz = f (z)dz
DF

FD

f (z)dz = f (z)dz
EF

FE

And this means that

f (z)dz + f (z)dz + f (z)dz + f (z)dz
(1)

(1)

(1)

(1)

= f (z)dz + f (z)dz + f (z)dz + f (z)dz + f (z)dz + f (z)dz

CD

DA

AE

EB

BF

FC

= f (z)dz
(0)

Our aim now is to show that this equals 0.

Step 2
We so far managed to express f (z)dz as a sum of 4 contour integrals, but this
(0)

is not very useful. We would like to somehow compare them, so that we end up with
an inequality between 2 integrals. However, we cannot compare the integrals themselves
since they are complex integrals. So what we do instead is compare the moduli. We ask
ourselves the question: Which f (z)dz has the biggest modulus? Since we picked an
(1)

arbitrary triangle to begin with, we cant explicitly answer this question. It could be any
(1)
of the 4 triangles, so we just pick some number j s.t. 1 j 4 and assume that Tj is
the triangle with greatest modulus of the contour integral around j . That is, we have
(1)

RRR
RRR
RRR RRR
RRR
RRR
RRR RRR
max RRRR. f (z)dz RRRR = RRRRR. f (z)dz RRRRR
k=1,2,3,4 RR
RRR RR
RRR
RRRR k(1)
RRR RRR j(1)
RRR
R
76

And now we can use this in the following way.

RRR
RRR
RRR RRRR
RR
RRR
RRR RRR 4
RRR. f (z)dz RRR = RRR f (z)dz RRRRR
RRR
RRR RRRk=1
RRR
RR (0)
RR RRR k(1)
RRR
RRR
RRR
RRR
4 RR
RRR
RR. f (z)dz RRRR (by the triangle inequality)
RRR
k=1 RRR (1)
RRR k
RRR
RRR
RRR
RRR
RRR
R
R
4 RRR. f (z)dz RRRRR (*)
RRR (1)
RRR
RRR j
RRR
Step 3
Now take the triangle with subscipt j and call it T (1) . Call its diameter d(1) , its
perimeter p(1) and its boundary (1) .
We can relate the diameter and the perimeter of this triangle to the diameter and perimeter of the original triangle. Since the way we obtained the triangles was by bisecting the
sides of the original triangle, we know that the largest side of the triangle T (1) is half the
largest side of the triangle T (0) and similarly for the perimeter. That is, we have:
d(1) = 21 d(0) and p(1) = 21 p(0) .
Now we concentrate only on the triangle T (1) and do exactly the same as with the
triangle T (0) . We bisect all the sides and connect the points which bisect the sides in
(2)
(2)
(2)
(2)
order to form 4 new triangles inside of T (1) . Call them T1 , T2 , T3 , T4 . Call their
(2) (2) (2) (2)
contours 1 ,2 ,3 ,4 .
A
E

T(1)1
T(1)4

T(1)3

T(2)1

T(2)4

T(2)2

T(2)3

Then again we choose the largest of the four new triangles, Tj , such that
(2)

RRR
RRR
RRR
RRR
RRR
RRR
RRR
RRR
RRR. f (z)dz RRR 4 RRRR. f (z)dz RRRR
RRR
RRR
RRR
RRR
RRR (2)
RRR
RR
RR (1)
R j
R
77

And now we call Tj

d(2) . Then

(2)

just T (2) . Call its boundary (2) , its perimeter p(2) and its diameter
1
p(2) = p(1) =
2
1
d(2) = d(1) =
2

and

1 (0)
p
4
1 (0)
d
4

RRR
RRR
RRR
RRR
RRR
RRR
RRR
R
RRR. f (z)dz RRR 4 RRR. f (z)dz RRRRR (*)
RRR
RRR
RRR
RRR
RR (1)
RR
RR (2)
RR

Now from the two (*)s, we can compare the moduli of the original contour integral
and the one around (2) . We have
RRR
RRR
RRR
RRR
RRR
RRR
R
R
R
RRR. f (z)dz RRR 42 RRRR. f (z)dz RRRRR
RRR
RRR
RRR
RRR
RR (0)
RR
RR (2)
RR
Step 4
Keep repeating the process. We once again divide the biggest new triangle into 4 smaller
triangles and compare their perimeters, diameters and moduli of the contour integrals
around their boundaries. We end up with
d(n) =

1 (0)
d
2n

p(n) =

1 (0)
p
2n

and, most importantly,

RRR
RRR
RRR
RRR
RRR
R
R
RRR
R
RRR. f (z)dz RRR 4n RRRR. f (z)dz RRRRR
RRR
RRR
RRR
RRR
RR
RR (n)
RR
RR (0)
Step 5
Well show that the RHS of the above inequality tends to 0 and from there reach the
statement of the theorem.
Now we notice that we kept taking triangles that are within bigger triangles, such that the
smaller triangles are always contained in the bigger triangles. We can make an analogy
with nested intervals, where the triangles are our "intervals". This implies that there
exists a unique point z0 which is contained in all triangles T (n) .
The point obviously also belongs to the original triangle T (0) and f (z) is holomorphic at
z0 . So the derivative at z0 of f (z) exists and we have
f (z) f (z0 )
zz0
z z0
f (z) f (z0 ) f (z0 )(z z0 )
lim
=0
zz0
z z0

f (z0 ) = lim

78

Call the numerator of the second fraction , i.e. we define

(z) = f (z) f (z0 ) f (z0 )(z z0 )
so that lim

(z)
zz0 zz0

= 0.

Now, holomorphic continuous, so we can use the , definition of continuity at a

point to get
(z)
<
 > 0 , > 0 s.t. 0 < z z0 <
z z0
Choose n large enough so that d(n) < , so 0 < z z0 d(n) < (**)
Step 6
RRR
RRR
R
R
Were interested in RRR. f (z)dz RRRRR, so it makes sense to want to look at f (z)dz.
RRR (n)
RRR
(n)
Now
(z) = f (z) f (z0 ) f (z0 )(z z0 ) f (z) = f (z0 ) + f (z0 )(z z0 ) + (z)
Integrating both sides, we get

f (z)dz = f (z0 )dz + f (z0 )(z z0 )dz + (z)dz

(n)

(n)

(n)

(n)

Since f (z0 )dz has f continuous and has an antiderivative and were integrating
(n)

around a closed curve, it gives is just 0.

Similarly for f (z0 )(z z0 )dz. Its antidervative is f (z0 ) 21 (z z0 )2 . So this integral
(n)

also gives us 0.

Thus, we end up with

f (z)dz = (z)dz
(n)

(n)

Now we take the moduli of both sides (and the equality still holds)
RRR
RRR
RRRR RRRR
RRR
R
R
R
R
R
RRR. f (z)dz RRR = RRR. (z)dz RRRRR
RRR
RRR RRR
RRR
RR RR (n)
RR
RR (n)

79

But we also have that

RRR
RRR
RRR
R
RRR. (z)dz RRRRR max (z).L( (n) )
RRR
RRR z (n)
RR
RR (n)
= max (z).p(n) (since the length of the triangualar curve is its perimeter)
z (n)

. max z z0 .p(n) (follows from (**) above)

z (z)

.d(n) .p(n) (again look at (**))

1
1
= . n d(0) . n p(0)
2
2
And so we have that

RRR
RRR
RRR
R
1
1
RRR. f (z)dz RRRRR . d(0) . p(0)
RRR
RRR
2n
2n
RR
RR (n)

which in turn means that

RRR
RRR
R
RRR
1
1
RRR. f (z)dz RRRRR 4n (. d(0) . p(0) )
RRR
RRR
2n
2n
RR
RR (0)
or,

RRR
RRR
RRR
R
RRR. f (z)dz RRRRR .d(0) .p(0)
RRR
RRR
RR
RR (0)

The diameter and perimeter are positive real numbers and this is inequality is true for
any  > 0. So the RHS gets arbitrarily close to 0, so it must be true that
f (z)dz = 0
(0)

QED
Example 43 (Application of Gousats Theorem)
Let f be holomorphic on the rectangular path R and its interior. Then
f (z)dz = 0
R

Proof. We split the rectangle into 2 triangles as shown:

80

T2

T1

We know from Gousats Theorem that

f (z)dz = 0 and f (z)dz = 0

T1

T2

f (z)dz = 0

T1 +T2

Now, we have that

f (z)dz = f (z)dz + f (z)dz + f (z)dz
T1

AB

BC

CA

and
f (z)dz = f (z)dz + f (z)dz + f (z)dz
T2

CD

DA

AC

So

T1 +T2

f (z)dz = 0 f (z)dz + f (z)dz + f (z)dz + f (z)dz + f (z)dz + f (z)dz = 0

AB

BC

CA

CD

DA

AC

f (z)dz + f (z)dz + f (z)dz + f (z)dz + f (z)dz f (z)dz = 0

AB

BC

CA

CD

DA

f (z)dz + f (z)dz + f (z)dz + f (z)dz = 0

AB

BC

CD

DA

f (z)dz = 0
R

Example 44 (The Fourier Transform of the Gaussian)

Prove that

x 2ix
e
dx = e
e

Proof. First of all, we notice that if we take = 0, we get the standard Gaussian

x
dx = e0 = e0 = 1
e

Now let > 0. We are going to use the following contour

81

CA

Im

-R+i
C4

-R

C1

C3
i

R+i

C2

Re

Call the whole curve R. Then R = C1 + C2 C3 C4 , where C3 and C4 are the curves
C3 and C4 but with reversed orientation, so that the curve R is in the positive direction
overall.
The parametrisations along C1 , C2 , C3 , C4 are as follows:
C1 Take z(t) = t for R t R. and z (t) = 1
C2 Take z(t) = R + it for 0 t . and z (t) = i
C3 Take z(t) = i t for R t R. but we are going to use the reverse (positive) of C3 , so take
C3 by z(t) = i + t for R t R. and z (t) = 1
C4 Take z(t) = R it for 0 t , but we take the reverse curve
C4 by z(t) = R + it for 0 t . and z (t) = i
The function we are going to consider along this contour is
f (z) = ez

This function is holomorphic on R and its interior, so we can apply Cauchys Theorem:
f (z)dz = 0 0 = f (z)dz + f (z)dz + f (z)dz + f (z)dz
R

C1

C2

C3

C4

We are going to consider each of the integrals and we are going to let the sides of the
rectangle grow to , thus making each line segment infinite.
Part 1 C1
We have that
2
f (z(t))z (t) = f (t).1 = et
R

f (z)dz = e
C1

t2

dt et dt as R

But et dt = 1, as we already showed. So the contribution from this integral as R

is 1.
2

82

Part 2 C2
Here

f (z(t))z (t) = f (R + it).i = e(R+it) .i

(R+it)2

f (z)dz = e

idt = ie(R

C2

2 t2 +2iRt)

dt

Consider the modulus of this integral. We have

RRR
RRR

RRR
RRR
2
2
(R2 t2 +2iRt)
dt = eR et e2iRt dt
RRR f (z)dz RRR ie
RRR
RRR
0
RC2
R 0

R2 t2

= e

R2

dt = e

t
e dt
0

The last integral is something we cant evaluate but it turns out we dont even need to,
because we are only interested in what happens as we let R . Doing this doesnt
2
change the integral, but eR 0 as R .
Thus

R2

t
e dt 0 as R
0

So we conclude (by the Sandwich Theorem) that

f (z)dz = 0 as R
C2

So this integral gives no contribution.

Part 3 C3 .
Since we are evaluating the whole integral over R in the clockwise direction, we take
f (z)dz = f (z)dz
C3

C3

So with the positive orientation, we have:

f (z(t))z (t) = f (i + t).1 = e(t+i)
R

f (z)dz = e
C3

(t+i)2

2 2 +2it)

dt = e(t

= e

t2 2 2it

= e

2 2it

t
e
R

R
2

= e ex e2ix dx
R

dt

x 2ix
e
dx as R
e

83

dt

Part 4 C4
Similar calculations to the ones for C2 show that the contribution from this integral as
R is 0.
Part 5 Combining all together:
0 = f (z)dz + f (z)dz + f (z)dz + f (z)dz
C1

C2

C3

C4

We let R and we get that

0=1+0e

x 2ix
e
dx + 0
e

So we have that

0=1e

x 2ix
e
dx
e

e ex e2ix dx = 1

ex e2ix dx = e

This is what we wanted to prove. Note that we can take real and imaginary parts to get
that:

e = ex cos(2x)dx

0 = ex sin(2x)dx

6.7

(T) The Antiderivative Theorem

The following theorem tells us when a function has an antiderivative.

Theorem 16 (The antiderivative theorem)
Let be a convex region. Let f (z) be continuous on with f (z)dz = 0, triangular

contours contained in . (note that we proved this for f (z) holomorphic.)

Fix a and define
F (z) = f (w)dw
[a,z]

Then F is a primitive of f (z) on (i.e. F (z) = f (z), z )

84

z
[a,z]
a

Proof. Fix z and > 0 s.t. D(z, ) .

Take h C s.t. h < so that z + h D(z, )
Now, let a be a point in which is outside of the disk we just constructed. Connect a to
the points z and z + h to form a triangle like this one:

[z,z+h]

|h|

z+h

[a,z+h]

[a,h]

Now we use Gausats Theorem on the triangle with vertices a, z, z + h. So he have:

f (w)dw +
[a,z]

f (w)dw +

[z,z+h]

[z+h,a]

= f (w)dw +

f (w)dw

[a,z]

[z,z+h]

[a,z+h]

Here we recognise
f (w)dw = F (z)
[a,z]

and

f (w)dw = 0

f (w)dw = F (z + h)

[a,z+h]

85

f (w)dw

by the statement of the theorem. So we have:

F (z) +

f (w)dw F (z + h) = 0

[z,z+h]

Now, our aim is to show that

F (z + h) F (z)
= f (z)
h0
h

lim
So we will work with

F (z + h) F (z) =

f (w)dw

[z,z+h]

to show that.
Consider
F (z + h) F (z)
1
f (z) =
h
h

f (w)dw f (z)

[z,z+h]

1
h

f (w)dw

1
h

[z,z+h]

f (z)dw

[z,z+h]

(Here we used a trick - we represented f (z) as an integral)

1
=
(f (w) f (z)) dw
h
[z,z+h]

Lets consider now the modulus

RR
RRR
R
F (z + h) F (z)
1 RRRR
RRR. (f (w) f (z)) dwRRRRR

f (z) =
RRR
h
h RRR
RR [z,z+h]
RR
1
.h. max f (w) f (z) , where h is the length of the line segment
w[z,z+h]
h
max f (w) f (z)
w[z,z+h]

Now, since f is continuous at z, we have, by the , definition of continuity at a point,

 > 0, 1 > 0 s.t. w z < 1 f (w) f (z) < 
By making the disk as small as possible, we can take h < 1 so that
max f (w) f (z) < 

w[z,z+h]

and so it tends to 0 as  0, which proves that

F (z + h) F (z)
f (z) as  0
h

86

Corollary 8
On an open disk a holomorphic function f has a primitive (the disk is a convex region)
if is a closed curve inside the disk,
f (z)dz = 0

Corollary 9
Let f be holomorphic on an open set U containing a circle C and its interior. Then
f (z)dz = 0
C

sketch proof. Let C be a circle of radius slightly bigger than that of C. Then we can
apply the previous corollary to the disk C bounded by C .
Example 45
is any closed curve in C. Then
z
e dz = 0

since ez is holomorphic (entire)everywhere in C

Example 46
For n N,
n
z dz = 0

Note: We proved this result before when considering the general case for all n.
Example 47
is the circle z = 2 traversed anticlockwise. Then

cos(z). cosh(3z + )
dz = 0
(z 2 + 16)(z 3 28)

We notice that there are possible discontinuities arising from potential 0s in the denominator. However, these 0s dont occur in the region we are working over.
z 2 + 16 = 0 z = + 4i

2i
4i
3
3
3
z 3 28 = 0 z = 28, 28e 4 , 28e 3
All of these points are outside the disk of radius 2.

87

6.8

Toy contours

The following are the kinds of contours we are going to work with. They are called, in
order:
circle (or semi-circle), triangle, rectangle, keyhole (circle), keyhole (rectangle), parallelogram, sector, indented region

6.9

(T)Cauchys Integral Formula

Theorem 17 (Cauchys Integral Formula)

Let f be holomorphic on an open set containing a closed disk with boundary the circle
C = CR (a) traversed anticlockwise. Let z0 C. Then
f (z0 ) =

1
f (z)
dz

2i
z z0
C

88

DR(a)
a

Proof. We are going to work with the following keyhole contour , :

The big almost-circle is C , traversed anticlockwise. It is missing an infinitesimal part of
its perimeter of length . The full circle is labelled C.
The small almost-circle is , , traversed clockwise. It is centred at z0 and has radius .
It is missing a part of its perimeter of infinitesimal length . The full circle is labelled
The line segment l1 is the one going inwards from the big circle to the small circle.
The line segment l2 is the one going outwards from the small circle to the big circle.
is the width of the corridor.

a
1,

Now, F (z) =

f (z)
zz0

0
z

1,

is holomorphic inside , , and so, by Cauchys Theorem,

F (z)dz = 0
,

We now evaluate the same integral by splitting the contour into the 4 constituent parts
listed above and we get that
0 = F (z)dz + F (z)dz + F (z)dz + F (z)dz
C

l1,

l2,

89

Lets consider first the integrals along the two vertical line segments. We let 0, and
the two segments become essentially the same segment. But since the original segments
had different directions, the two integrals sum to 0:
F (z)dz + F (z)dz 0 as 0
l1,

l2,

Similarly, we consider the integral along C as 0, the circle becomes a closed circle.
Same for the integral around , which become . That is, we have:
F (z)dz F (z)dz as 0
C

F (z)dz F (z)dz as 0
,

So overall, as 0, we end up with

0=
C

f (z)
f (z)
dz +
dz
z z0
z z0


The integral around C is the one we want to still have in the end, but we need to
manipulate the other integral to get the f (z0 ) in the formula. Now, we do the following
trick:
f (z) f (z0 ) + f (z0 )
f (z) f (z0 )
f (z0 )
f (z)
dz =
dz +
dz
z z dz =
z z0
z z0
z z0
0

We are going to consider these two integral individually and show that one of them gives
us the LHS of the formula, while the other one vanishes.
Consider first the second integral.
f (z0 ) is a constant so we can write
f (z0 )
1
z z dz = f (z0 ) z z dz
0
0

And we can evaluate this integral by parametrising with z(t) = z0 + eit , 0 t 2, so that
dz = ieit and
2
f (z0 )
1
it
z z dz = f (z0 ) eit ie dt
0

= f (z0 ) idt
0

= 2if (z0 )
90

Now consider the integral

f (z) f (z0 )
dz
z z0

We recognise something in the integrand. It is that

f (z) f (z0 )
= f (z0 )
zz0
z z0
lim

Note that the derivative exists since the function is holomorphic.

Now this implies that
E > 0, > 0 s.t. 0 < z z0 <

f (z) f (z0 )
f (z0 ) < E
z z0

We can make the following choices: Let E = 1, fix > 0 and take < .
We now consider the modulus of the integral with the intention to show that it is zero.
RRR
RR RR
RRR
RRR
R
f (z) f (z0 ) RRRR RRRR f (z) f (z0 )

RRR.
dz RRR = RRR
f (z0 ) + f (z0 )dz RRRRR
z z0
z z0
RRR
RRR RRR
RRR
R
R R
R
f (z) f (z0 )
f (z0 ) + f (z0 )
2. max
zz0 =
z z0
f (z) f (z0 )
2 max
f (z0 ) + f (z0 )
zz0 =
z z0
< 2 (1 + f (z0 ))
0 as 0
So, combining all together, we have
0=
C

f (z)
1
f (z)
dz 2if (z0 ) f (z0 ) =
dz

z z0
2i
z z0
C

Example 48
Let C be the circle z = 2. Evaluate, using Cauchys Integral Formula,
z
(9 z 2 )(z + i) dz
C

Let z0 = i and f (z) =

z
9z 2 .

Then

f (z)
z + i dz = 2if (i)
C

= 2i

i
2i(i)
=
=
2
9 (i)
9 (1) 5
91

Example 49
With the same circle, consider

cos z
z 2 + 1 dz
C

Failed attempt:

cos z
cos z
z 2 + 1 dz = (z + i)(z i) dz
C

z
so take f (z) = cos
zi and z0 = i. This wont work since f is not a holomorphic function
inside the circle of radius 2, since it has a singularity at i.
Correct attempt: By using partial fraction, we rewrite the integral as

1
cos z
1
cos z
cos z
z 2 + 1 dz = 2i z + i dz + 2i z i dz
C

Now we can apply Cauchys Integral Formula with f (z) = cos z, to get
2i
2i
cos(i) +
cos(i)
2i
2i
= cos(i) + cos(i)
ei.i + ei.i
ei.i + ei.i
=
+
2
2
=0
=

6.10

(T) Cauchys Formula for f and higher derivatives

Theorem 18 (Cauchys Formula for f)

f (z0 ) =

1
f (z)
dz

2i
(z z0 )2
C

Proof. We start with

f (z0 + h) f (z0 )
h
with the intention to take te limit as h later.

92

We use Cauchys Integral Formula to write this as

f (z0 + h) f (z0 )
h
1
[ 2i

f (z)
z(z0 +h) dz

1
2i

f (z)
zz0 dz]

h
f (z)
f (z)
1
(

) dz
=

2ih
z (z0 + h) z z0
C

z z0 z + (z0 + h)
1
f (z) (
) dz

2ih
(z (z0 + h))(z z0 )
C

1
f (z)
dz

2i
(z z0 )(z (z0 + h))
C

We would like to show that when h 0,

f (z)
f (z)
1
1
dz
dz

2i
(z z0 )(z (z0 + h))
2i
(z z0 )(z z0 )
C

If we can show this, then we have proved the theorem.

So now we need to prove:
lim
h0

or,

f (z)
f (z)
dz =
dz
(z z0 h)(z z0 )
(z z0 )2
C

f (z)
f (z)
(z z h)(z z ) dz (z z )2 dz 0
0
0
0
C

Once again, we will use the technique of looking at the modulus of the integral when we

93

dont know what to say about the integral itself. So we consider

RRR
RRR RRR
f (z)
f (z)
RRR
RRR RRR
dz
dz
RRR
R=R
(z z0 )2 RRRR RRRR
RRR C (z z0 h)(z z0 )
C
R RC
RRR
R
= RRRR
RRR
RC
RRR
R
= RRRR
RRR
RC
RRR
R
= RRRR
RRR
RC

RRR
1
1
f (z)
R
(

)] dz RRRR
[
RRR
(z z0 ) z z0 h z z0
R
RR
R
f (z) z z0 z + z0 + h
R
(
)] dz RRRR
[
z z0 (z z0 h)(z z0 )
RRRR
RRR
h
f (z)
R
(
)] dz RRRR
[
RRR
z z0 (z z0 h)(z z0 )
R
RRR
h
f (z)
R
(
)] dz RRRR
[
2
RRR
(z z0 ) (z z0 h)
R
f (z)h

(2R). max
za=R (z z0 )2 (z z0 h)

where 2R is the perimeter of the

circle of radius R and a is its center
Now we only need to consider (zz0 f)2(z)h
(zz0 h) . For this, we have the following diagram:

C
z

where a is the center of the circle we are considering, R is the radius and z0 is a point inside
the disk, z1 is a point on the circle. Were interested in the distance z z0 . for any z on
the circle. Now, the shortest distance from a fixed z0 to any point z1 on the circle is along
the diagonal through a, and this distance is d = z1 z0 . We can drop the subscript 1, so
that d = z1 z0 = min z z0 (as we chose z1 to be the point closest to z0 ). So now we have:

94

RRR
RRR
f (z)
f (z)h
f (z)
R
RRR
dz
dz RRRR (2R). max

RRR
2
RRR
za=R (z z0 )2 (z z0 h)
(z z0 )
RRR C (z z0 h)(z z0 )
C
R
f (z)h
(2R). max 2
zC d .z z0 h
f (z)h
(2R). max 2
zC d .(d h)
since z z0 h z z0 h d h
0 as h 0
Theorem 19 (Cauchys Formula for the nth derivative)
f (n) (z0 ) =

f (z)
n!
dz

2i
(z z0 )n+1
C

See proof in homework.

Example 50
Show that

1 cos(x)

dx =
2
x
2

This integral has nothing to do with complex numbers at first glance nad it takes some
experience to be able to notice the complex function and contour to use to arrive at the
result.
Here we will consider
1 eiz
f (z) =
z2
and we are going to integrate over the following contour:
The upper semi-circle of radius R, but with a small semi-circular bit cut off from the
contour very close to 0, i.e. a distance from 0.
Our contour consists of the following:
The curve R+ , i.e. the big semi-circle traversed anti-clockwise from R to R.
The line segment [R, ], traversed from left to right
The curve , i.e. the small semi-circle traversed clockwise from to
The line segment [, R], traversed from left to right
iz

Now, we have chosen f (z) = 1e

z 2 , which is a holomorphic function in our region and
hence, by Cauchys Theorem, we have
f (z)dz = 0

95

So we can write
0 = f (z)dz

= f (z)dz +
+
R

f (z)dz + f (z)dz + f (z)dz

[R,]

[,R]

Now we are going to consider each term individually.

Term 1
f (z)dz

[R,]

We can parametrise the curve [R, ] with z(x) = x, R x

So z (x) = 1 and f (z(x)) = f (x)
Thus

1 eix
1 eiz
=
dx

z2
x2
[R,]

Now lets substitute x = u (for reasons which will become clear later), so that du = dx
and at x = R we have u = R and at x = we have u = . And so we have the integral

1 eiu
1 eiu
1 eiu
du
=

du
=
du

u2
u2
u2
R

And now just replace u with x again to get

R

1 eix
dx
x2

Term 2

f (z)dz
[,R]

We can parametrise the curve [, R] with z(x) = x, x R

So z (x) = 1 and f (z(x)) = f (x)
So we get
R

1 eix
dx
x2

96

Now the sum of Term 1 and Term 2 is

R

1 eix
1 eix + 1 eix
1 eix
dx
+
dx
=
dx

x2
x2
x2

2 (eix + eix )
dx
x2

2 2(cos x)
dx
x2

= 2

1 cosx
dx
x2

Now as we increase the size of the bigger circle, i.e. we let R and decrease the size
of the smaller circle. i.e. we let 0, we effectively are trying to cover the whole upper
complex plane, and thus to obtain the result we are looking for.
So we consider what happens as R and 0. We get that
R

1 cos x
1 cos x
dx

2
dx
2

x2
x2

Term 3
f (z)dz =
+
R

+
R

1 eiz
dz
z2

A common trick is to consider the modulus of the integral if we seems incapable of saying
anything about the integral itself.
So consider
RRR
R
RRR 1 eiz RRRRR
1 eiz
RR (max
dz
) .R where R is the perimeter of the semi-circle
RRRR
R
RRR
z=R
z2
z2
RRR +
RR
RR
1 eiz
= R max
,as z 2 = R2
z=R
R2

= max 1 eiz (*)

R z=R
Now we consider just 1 eiz and try to deduce something about it.
1 eiz = 1 ei(x+iy) = 1 eix ey
1 + eix ey = 1eix ey
= 1 + ey since eix = 1
97

Now, we know that y 0 as we are integrating over a contour located in the upper halfplane. Thus, ey 1 and so 1 + ey 1 + 1 = 2.
Then (*) becomes
RR
RRR
RRR
2
1 eiz RRRR
dz RRR .2 =
RRR.
2
z
R
RRR R
RRR
R
R R
And as R , we have that 2
R 0 and hence
f (z)dz 0
R

So there is no contribution from this integral.

Term 4
f (z)dz

First of all, we have that is the curve traversed clockwise. So we let C be the same
curve traversed anticlockwise and we have that this curve can be parametrised by
z(t) = eit for 0 t
So we have
f (z)dz = f (z)dz =

1 eiz
dz
z2

Now again we want to consider what happens as 0. However, here parametrising the
curve wont help as it will complicate the problem. Its not obvious how to proceed but

98

here is a nice trick: We use the expansion of eiz . We have that

2

1 eiz

z2

dz =

(iz)
1 (1 + iz + (iz)
2! + 3! + . . .)

z2

dz

i
i2 i3 z i4 z 2
= ( +
+
+ . . .) dz
z
2! 3!
4!
C

= i
C

= i
0

i2 i3 z i4 z 2
1
dz + E(z)dz where E(z) = +
+
+ ...
z
2! 3!
4!
C

ieit
E(z)dz
+
eit
C

where we converted the line integral using the parametrisation from earlier.

= i idz + E(z)dz
0

i [iz]0

+ E(z)dz
C

= + E(z)dz
C

At this point we notice that we can actually solve the problem if we can show that
E(z)dz 0 as 0.
C

Again we are going to consider the modulus of the integral.

RRR
RRR
RRR
R
RRR E(z)dz RRRRR max E(z)
z=
RRR
RRR
RC
R
What can we say about E(z)?
E(z) =

r
i i3 z i4 z 2
i
+
+
+ . . . = z r2
2! 3!
4!
r=0 r!

We can apply the ratio test

ir+1 z r1 r!
iz r1
1
. r r2 =
=
z 0 as r
r2
(r + 1)! i z
(r + 1)z
(r + 1)

So we get that this series has an infinite radius of convergence and therefore E(z) is
continuous on D(0, ) E(z) is bounded on D(0, ), say E(z) M for some
constant M . We now have that
RRR
RRR
RRR
R
RRR E(z)dz RRRRR M 0 as 0
RRR
RRR
RC
R
99

So the total contribution as 0 of

f (z)dz
C

is .
Step 5: Combining all together, we get that

0 = 2
0

1 cos(x)
dx + 0 + 0 + ()
x2

2
0

6.11

1 cos(x)
dx =
x2

1 cos(x)
dx =
2
x
2

(T)Cauchys Inequalities

Theorem 20 (Cauchys Inequlities)

f (n) (z0 )

[n! maxzz0 =R f (z)]

Rn

Remark:
maxzz0 =R f (z) exists. This is the same as maxt[0,2] f (z0 + Reit ), since
f (z0 ) + Reit is a continuous function, f [0, 2] R, so it achieves a maximum (by
Analysis 1)
Proof. First of all, by Cauchys Integral Formula, we have that
f (n) (z0 ) =

n!
f (z)
dz

2i
(z z0 )n+1
C

We want to end up with and inequality for the modulus of f (n) (z0 ), we stick in the

100

modulus signs:
f

(n)

RRR
RRR
f (z)
RRR
RRR n!
dz
(z0 ) = RR
R

RRR 2i
(z z0 )n+1 RRRR
C
R
R
RRR
RRR
n! RR
f (z)
R
RRR
dz RRRR
=
n+1
RRR
2 RRR (z z0 )
RC
R
f (z)
n!
.(2R). max
zz0 =R (z z0 )n+1
2
f (z)
= n!R. max
zz0 =R Rn+1
n!
= n max f (z)
R zz0 =R

Definition 52 (Entire)
If f is holomorphic on the whole complex plane, then f is called entire.

6.12

(T)Liouvilles Theorem and The Fundamental Theorem of Algebra

Theorem 21 (Liouvilles Theorem)

If f is entire and bounded, then f = constant.
Proof. To find that f is constant we are going to prove that f (z) = 0.
Lets use Cauchys inequality on an integral about the circle with radius R, centered at
z0 . We have
f (z)
f (z) max
zz0 =R
R
Now, since f is bounded on C, we can find a constant M > 0 such that
f (z0 )

M
R

And now we note that, as R , this goes to 0, and, hence

f (z0 ) = 0
Now, we took an arbitrary circle to begin with and a point z0 to be its centre. But as we
let R , we are essentially covering the whole complex plane, and so we can take any
z0 , i.e. f (z) = 0, z C. The result follows.
Example 51 (An application of Cauchys Inequality and Liouvilles Theorem)
Let f be entire and suppose we can find M , R0 , s.t.
f (z) M z1/2 , z, s.t. z > R0
101

i.e. we can bound the function outside and on the disk of radius R0
Claim: f (z) is a constant everywhere.
Proof. Fix z0 C. To show that f (z0 ) = 0, consider Cauchys Inequality:
f (z)
M z1/2
= max
zz0 =R
zz0 =R
R
R

f (z0 ) max

Now we want to say something about z. Lets express it as z = z z0 + z0 and the we

can apply the triangle inequality to get
z z0 + z0 z z0 + z0
Now, if z is on the circle of radius R, then z z0 = R and so
z z0 + z0 = R + z0
So for any z inside the disk,
z z0 + z0 < R + z0
So we get
z R + z0 z1/2 (R + z0 )1/2
So now we can go back and say
M z1/2 (R + z0 )1/2
f (z)
= max

zz0 =R
zz0 =R
R
R
R

f (z0 ) max
And this goes to 0 as R .
f (z) = const

f (z0 ) = 0. Therefore, f (z) = 0, z C.

Theorem 22 (The Fundamental Theorem of Algebra)

Every non-constant polynomial (with constant coefficients) has a root in C.
Proof. (By contradiction)
Assume that
p(z) = an z n + an1 z n1 + an2 z n2 + + a1 z + a0
1
We will show that f (z) is entire
has no roots in C. We are going to look at f (z) = p(z)
and bounded and conclude, by Liovelles Theorem, that f (z) = k. This, in turn would
imply that p(z) = k1 , so our polynomial is also constant. But this is a contradiction.

Step 1 (f is entire)

1
= f (z) is entire
p(z)

Now, the denominator is never zero, since our assumption is that p(z) has no roots.
f is entire.
102

Step 2 (f is bounded)
First of all, here is some intuition into the proof to follow:
Let z = R, where R is large (eventually going to ).
p(z) = an z n + an1 z n1 + + a1 z + a0
an z n = an Rn
an1 z n1 = an1 Rn1

a0
Since R is a large number, we expect that the first term above will be much bigger than
the others. That is, the behaviour of the polynomial is pretty much determined by the
leading term as we let R .
R4

R3

End of intuition. Here is how we actually prove this:

Fix R and assume z = R.

p(z) = an z n + an1 z n1 + + a1 z + a0
an z n an1 z n1 + + a1 z + a0 by the triangle inequality
Now,

an1 z n1 + + a1 z + a0 an1 z n1 + an2 z n2 + + a1 z + a0

= an1 Rn1 + an2 Rn2 + + a1 R + a0

So plugging this back above, we get

p(z) an Rn an1 Rn1 an2 Rn2 a0
= an Rn (1
Now
lim (

an1 1 an2 1
a0 1

)
2
an R
an R
an Rn

an1 1 an2 1
a0 1

)=0
2
an R
an R
an Rn
103

The Principle of Inertia says that R0 > 0 such that R, R > R0 ,

(1

a0 1
1
an1 1 an2 1

)>
2
n
an R
an R
an R
2

expansion

1
1/2
R

R0

So for R > R0 ,

1
2
1
2
p(z) an Rn

2
p(z)
an Rn an R0n
With this, we have proved that f is bounded for R > R0 , so now we only need to show that
it is also bounded in the disk of radius R0 , i.e. in D(0, R0 ). Now, this disk is contained
2
in the square with sides 2R0 , i.e. D(0, R0 ) [R0 , R0 ] . [R0 , R0 ] (= [R, R0 ] )
But on this square, f is bounded. This is analogous to "continuous functions on closed
1
is continuous on the box
intervals are bounded", from Analysis 1. This is so because p(z)
2

[R0 , R0 ] , and therefore it is bounded on it.

2R0
0

104

R0

So,

1
p(z)

is bounded on D(0, R0 ) by, say, k. So

1
k, z D(0, R0 )
p(z)

So now he have inequalities for f both inside and outside he disk of radius R0 , i.e. we
have 2 inequalities covering essentially the whole complex plane. Lets turn them into a
single inequality by taking M = min (k, an2Rn ).
0
Then f M on the whole complex plane. So we have proved that f is bounded.
Hence, we have now proved that f is both entire and bounded, which implies (Liouvilles
Theorem) that f =constant and so p(z) = constant. This is a contradiction, since the
statement of the the theorem requires p(z) to be nonconstant.
Corollary 10
Every polynomial over C of degree n 1 can be factorised as
p(z) = an (z w1 )(z w2 )(z wn )
where w1 , . . . , wn are the roots of the polynomial and there are exactly n of them.
Proof. By the Fundamental Theorem of Algebra, we can find a root w1 for the polynomial.
i.e. we can write
p(z) = (zw1 )Q(z)+R(z) , where Q(z) is the quotient and R(z) is the remainder of the division
Since w1 is a root, we have p(w1 ) = 0, so
0 = p(w1 ) = 0.Q(w1 ) + R R = 0
So
p(z) = (z w1 )Q(z)
Now deg(Q(z)) = n 1 and we can write
p(z) = (z w1 )(z w2 ).Q1 (z), where deg(Q1 (z)) = n 2
Continue in this way until
p(z) = (z w1 )(z w2 ) . . . (z wn ).C , where C is a constant quotient
Now compare
p(z) = an z n + an1 z n1 + + a1 z + a0 = C(z w1 ) . . . (z wn1 )(z wn )
So z n has coefficient an on the LHS, while it has a coefficient C on the RHS. Therefore,
an = C.

105

6.13

Power Series Expansion of Holomorphic Functions

Theorem 23
Let f be holomorphic on an open set containing the closed disk D(z0 , R). Then f has a
power series expansion at z0 .
z D(z0 , R)

f (z) = an (z z0 )n
n=0

where
an =

1
f (n) (z0 )
=
n!
2i

CR (z0 )

f (w)
dw
(w z0 )n+1

Proof. By Cauchys Integral Formula

f (z) =

1
2i

f (w)
dw
wz

1
2i

f (w)
dw
w z0 (z z0 )

1
2i

f (w)
1
.
zz0
w z0 1 wz
0

CR (z0 )

CR (z0 )

CR (z0 )

Now we notice that the term

1
zz
z wz0

is the sum of the geometric series

(
n=0

z z0 n
)
w z0

zz0
wz0 ,

for r =
if r < 1.
Now, looking at our picture: The ws lie on the circle, while z is inside the circle, and so
is z0 , which is the centre. So we have that z z0 < w z0 and so r is indeed less than 1
and the sum we wrote above is valid.

z0

z
w

106

Therefore, now we have

f (z) =

1
2i

f (w)
dw
wz

1
2i

f (w) (z z0 )n
.
w z0 n=0 (w z0 )n

CR (z0 )

CR (z0 )

f (w)
1
n
=
(w z )n+1 dw(z z0 )
2i
0
n=0

CR (z0 )

which is what we wanted to prove. However, we still need to prove that the last step we
did was legitimate. We need to show that
N

n=0
CR (z0 )

f (w)
dw.(z z0 )n
(w z0 )n+1

CR (z0 )

f (w)
(z z0 )n dw as N
n+1
(w

z
)
0
n=0

Consider the modulus of the difference

RRR
RRR

RRR N
RR
f (w)
f
(w)
RRR
(z z0 )n dw
(z z0 )n dwRRRR
n+1
n+1
RRRn=0
RRR
(w z0 )
n=0 (w z0 )
CR (z0 )
RR CR (z0 )
RR
RRR
RRR

N
RR
RRR
f (w)
n
RRR
= RRRR. ( )
(z

z
)
dw
0
n+1
RRR
RRR
(w

z
)
0
RR
RR CR (z0 ) n=0 n=0
RRR
RRR

RRR
RR
f (w)
n
RRR
(z

z
)
dw
= RRRR.
0
n+1
RRR
RRR
N +1 (w z0 )
RR
RR CR (z0 )
n
f (w)(z z0 )
2R. max

wz0 =R n>N (w z0 )n+1

f (w)(z z0 )n

wz0 =R n>N
(w z0 )n+1
f (w)
z z0 n
2R max
.

wz0 =R w z0 n>N w z0
f (w)
z z0 n
= 2R max
.

wz0 =R
R n>N w z0
f (w)
. rn
= 2R max
wz0 =R
R n>N
2R max

= 2 max f (w) rn
wz0 =R

n>N

where
rn =
n>N

107

rN +1
1 r

so

z z0
<1
w z0
lim rN +1 = 0

r =

and so

rN +1
=0
N 1 r
lim

Thus we have shown that

n=0

n=0

But also,
N

n=0

n=0

And these two imply that

=
n=0

n=0

Theorem 24 (Moreras Theorem)

Let f be continuous on a region and
f (z)dz = 0

for all , closed contours in , with interior in .

Then f is holomorphic.
Proof. Let D(z, r) be a disk (a convex set) inside . For f continuous on a convex set,
we can construct a primitive function F on D(z, r) such that F = f .

D(z,r)

r>0

108

F (w) = f (u)du
[z,w]

So F in holomorphic and by the previous theorem, F has a power series expansion.

Therefore, it is infinitely many times differentiable. But f = F , so f is also infinitely
many times differentiable, i.e. it is holomorphic.
Discussion
If f is holomorphic on D(z0 , r), r > 0, we can express f as a power series centred at z0 as

f (z) = an (z z0 )n
n=0
(n)

where an = f n!(z0 ) . Then we have 2 cases:

Case 1: If an = 0, n, then f (z) = 0, z D(z0 , r)
Case 2: If f (z) =/ 0 z D(z0 , r), then an which are not 0.
Let N be the smallest such n for which aN =/ 0, while a0 = a1 = = aN 1 = 0. Then
f (z) = aN (z z0 )N + aN +1 (z z0 )N +1 + . . .
= (z z0 )N (aN + aN +1 (z z0 ) + aN +2 (z z0 )2 + . . . )
= (z z0 )N g(z)
Here g(z) has the same coefficients as the series of f (z), so g(z) has the same radius of
convergence r > 0, so g(z) is holomorphic on D(z0 , r) g is continuous on D(z0 , r).
lim g(z) = g(z0 ) = aN =/ 0

zz0

By the Inertia Principle. > 0 s.t. g(z) =/ 0. for z D(z0 , ).

So
f (z) = (z z0 )N g(z)
with g(z) holomorphic and non-vanishing on D(z0 , ).

z0

109

Theorem 25
If f is holomorphic on D(z0 , r), r > 0 and f doesnt vanish identically (i.e. it is not the
case that f (z) = 0, z D(z0 , r) ), then > 0, and a non-vanishing holomorphic function
g(z) on D(z0 , ) and also a unique N 0 with
f (z) = (z z0 )N g(z) , z D(z0 , )
Proof. We have already given most of the proof in the previous discussion. The only thing
that remains to be proved in uniqueness.
Assume
f (z) = (z z0 )N g(z)
and
f (z) = (z z0 )k h(z)
with g and h holomorphic and non-vanishing on some D(z0 , ), > 0. Then N = k. Why
is this true?
Suppose that N > k. Then we can cancel out (z z0 )k from
f (z) = (z z0 )N g(z) = (z z0 )k h(z)
to get

(z z0 )N k g(z) = h(z) ,where N k > 0

Then, taking the limits of both sides, we get that

lim (z z0 )N k g(z) = lim h(z)
zz0

zz0

This implies that

0.g(z0 ) = h(z0 ) h(z0 ) = 0
but we were assuming that h is non-vanishing. This is a contradiction.
The case k > N gives a similar contradiction.
Hence, k = N

6.14

Singularities and Zeroes

Definition 53
If N 1, we say that f (z) has a zero at z0 of order N .
If N = 1, we say that f (z) has a simple zero at z0
Definition 54 (Removable Singularity)
If f (z) is defined and holomorphic on the punctured disk D (z0 , r) = {z; 0 < z z0 < r}
and one can define f (z0 ) such that f is holomorphic on the whole disk D(z0 , r), then z0
is called a removable singularity.

110

Theorem 26
Suppose that f (z) is analytic in the region obtained by omitting a point a from
the region . A necessary and sufficient condition that there exists an analytic
function in which coincides with f (z) in is that
lim(z a)f (z) = 0
za

The extended function is uniquely determined.

(Theorem form Ahlros, p 124)
This theorem is useful for actually checking whether something is a removable
singularity or not. In class we mentioned that if lim f (z) = , where a is the
za
singularity means that the singularity is not removable. However, the theorem
above provides a rigorous check for removable singularities. Note that as long as
lim f (z) is computable, and not , we can safely (but non-rigorously) conclude
za
that the singularity is removable. Note also that, by computing the above limit,
if it exists, we can find the value of g(a), i.e. the value we want to assign to
the function at a removable singularity (where g is our "extended" f , as in the
definition above. )
The following are a few examples done first in the way we proved them in class
(Step 1) and then applying the theorem above (Step 2), and then we find the value
of g(a) (Step 3) in Example 1 only.
Example 52
f C {0} C, f (z) = z
Part 1:
Define g C C s.t. z C {0}, g(z) = f (z) and g(z) = 0 for z = 0
Then g(z) = z, z C, and hence, g is entire. So we have a removable singularity at 0.
Part 2:
Consider
lim(z 0)z = lim z 2 = 0
z0

z0

So 0 is indeed a removable singularity.

Part 3:
To find the actual value we want to take for g(0), we compute:
lim z = 0
z0

So we take g(0) = 0.
Note that in this case it was obvious what value we wanted to take for g(0) but in some
more complicated cases, computing the limit is actually required.

111

Example 53
f C {0} C, f (z) =

1
z

Part 1: The point where we have a singularities is 0.

1
=
z0 z

lim f (z) = lim

z0

(note: not formal proof of infinite limit. See definition below) So we dont have a removable singularity at 0.
Part 2:
1
lim(z 0) = lim(1) = 1 =/ 0
z0
z z0
So we dont have a removable singularity at 0.
Part 3:
This is the same as part 1, so in class we used this as proof that the singularity is not
removable.
Example 54
f (z) =

z+1
+ 1)

z 3 .(z 2

The singularities are at 0, i, i.

Part 1:
At 0:
z+1
z0 z 3 (z 2 + 1)
1
z+1
= (lim 3 ) . (lim 2
)
z0 z
z0 z + 1
= .(1) =

lim f (z) = lim

z0

We dont get a finite value for the limit, so there is no removable singularity.
At i:
z+1
zi z 3 (z 2 + 1)
z+1 1
1
= lim 3
zi z
z+izi
z+1
1
1
= (lim 3 ) . (lim
) . (lim
)
zi z
zi z + i
zi z i
= ..
=

lim f (z) = lim

zi

112

So again we dont have a removable singularity.

Exactly similar calculation shows that for z = i, we also get
lim f (z) =

zi

And so we dont have any removable singularities in this example.

Part 2:
Lets apply the theorem to the 3 points in turn:
z+1
= =/ 0
+ 1)

lim(z 0)f (z) = lim

z0 z ( z 2

z0

z+1
= =/ 0
+ i)
z+1
= =/ 0
lim (z + i)f (z) = lim 3
zi
zi z (z i)
that the points are not removable singularities.
(note: not formal proofs of infinite limit. See definition below)
lim(z i)f (z) = lim
zi

zi

z 3 (z

Example 55
An additional example
sin z
z
Applying the Theorem to check if we have a singularity at z = 0:
f (z) =

lim(z 0)f (z)

= lim z.

z0

= lim sin z
z0

z0

sin z
z

z3 z5
= lim (z + + . . . )
z0
3! 5!

=0
Hence, 0 is a removable singularity. To check the value of g(0), we compute the
limit:
3
5
z z3! + z5! + . . .
lim f (z) = lim
z0
z0
z
z2 z4
= lim 1 + . . .
z0
3! 5!
=1
113

So we take g(0) = 1. And thus we define

g C C = f C {0} C, z C {0}
and g(z) = 1 for z = 0, so g is holomorphic everywhere.

Definition 55
We define
lim f (z) =

zz0

to mean

M > 0, > 0 s.t. 0 < z z0 < f (z) > M

Definition 56 (Isolated singularity)

If f is holomorphic on the punctured disk D (z0 , r) = {z, 0 < z z0 < r}, then z0 is called
an isolated singularity.
Additional 3
Easy way of thinking about this: If we have a function with more than 1 singularity,
we want those to be relatively further apart to be considered isolated singularities
(note how the word "isolated" makes sense). This means that around any of the
singularities, we can draw a punctured disk with an arbitrarily small radius such
that the disk doesnt contain any other singularities.
Example 56
f (z) = Log(z) = Logz + iArg(z), < Arg(z)
There is a singularity at z0 = 0, but this is not an isolated singularity as no matter how
small an open disk we put around 0, f still wont be holomorphic on it since Log requires
a cut from to 0, which goes through any circle centred at the origin, and so f is not
holomorphic there.
Example 57
f (z) =

1
sin( z )

First lets find all the singularities:

sin( ) = 0
= arcsin(0) n = for n N
z
z
z
So all the points z = n1 for n N are singularities.
There is also a singularity at z = 0
114

Claim: Only z = n1 are an isolated singularities.

Consider the following argument: The numbers z = n1 all lie on the real axis and are all
between -1 and 1. Since n N, we cant cover the whole line segment [1, 1] with numbers
z = n1 , since there are also irrational numbers in this interval. But this means that around
any of the numbers n1 we can draw some circle such that the interior of the circle doesnt
include any other singularities but the one at the centre. And since were interested in
f being holomorphic on a punctured disk, we can just take the disk to be, centred at a
given n1 , with radius extending to the singularity to the left and to the right of the one we
are considering. But by the definition of isolated singularity, this means that all of the
points z = n1 are isolated singularities.
On the other hand, consider the singularity at z = 0. If we draw any disk around z = 0,
this disk will contain some other singularities (at least 2) of the form z = n1 . So f cant
be made holomorphic on any punctured disk with 0 at its centre. So 0 is not an isolated
singularity.
Definition 57 (Pole)
We say that f has a pole at z0 if f is defined at some punctured disk D (z0 , r) C and
1
D (z0 , r) C
f
can be defined at z0 by

1
(z0 ) = 0
f

and is holomorphic on D(z0 , ) for > 0

Additional 4
A meromorphic function on an open subset D of the complex plane is a function
that is holomorphic on all D except a set of isolated singularities, which are poles
for the function.
Note that removable singularities are just a type of isolated singularities. Here
is the full classification:
Definition 58 (Isolated singularity)
Let U be an open subset of a Riemann surface and let z0 U .
A holomorphic function f U {z0 } C is said to have an isolated singularity at z0 .
Types of isolated singularities:
Removable singularity
A pole:
If lim f (z) = , then the point z0 is said to be a pole of f (z) and we set
zz0

115

f (a) = .
An essential singularity:
This is an isolated singularity which is neither removable nor a pole.
An important point is that a "pole" is actually the same thing as a removable
singularity, if we think of our function as a map that takes values on the Riemann
sphere (which is the complex plane with a point at added; the complex structure
near comes from the map z z1 ).
So a function that has a removable singularity or a pole at z0 doesnt have a
"real" singularity there at all; rather, we can extend the function to an analytic
or meromorphic function in z0 . If we cannot extend the function in this way, the
singularity is indeed "essential"; i.e., we cannot get rid of it.
(End of additional)
Theorem 27
If f has a pole at z0 , then a small > 0 and a holomorphic function h(z) on D(z0 , )
which doesnt vanish of D(z0 , ) and there exists a unique N N with
f (z) = (z z0 )N h(z)
where z D(z0 , )
Proof. f1 can be written as (z z0 )N g(z) with N 1, by the previous Theorem.
Here g(z) is holomorphic and it doesnt vanish on the disk D(z0 , )
f (z) = (z z0 )N

1
g(z)

1
Set h(z) = g(z)
. As g(z) is holomorphic and g(z =/ 0) on the disk, h(z) is also holomorphic
and not equal to 0 on the disk.

Definition 59 (Order of the pole)

The N which appears in the proof above is called the order of the pole.
If N = 1, then the pole is called a simple pole.
Additional 5
In the following examples we use the fact that we already found the isolated, nonremovable singularities of the function and we are trying to show that they are in
1
fact poles by trying to represent the function in the form f (z) = (z z0 )N g(z)
as
in the theorem above. (We never use the definition of the pole as being a limit.)
By writing the function in this form, we can see what the order of the pole is.
Note that it seems like we are using the theorem, so we are assuming that we
already know the poles and that they are the non-removable isolated singularities,
and yet we end up proving that these points are the poles. This seems like a wrong
1
approach. The theorem states that we can represent f as f (z) = (z z0 )N g(z)
if we already know that the point z0 is a removable singularity but we use it to
116

prove that. Adopt this approach for the purposes of the course but check with the
other definitions when possible.
(End of additional)
Example 58
1
z
We have seen that 0 is an isolated singularity and its non-removable.
We can write
f (z) = (z 0)1 .1
f (z) =

Where our h(z) = 1 is holomorphic and never 0. So 0 is a pole, and it is in fact a simple
pole.
Example 59
z+1
+ 1)
We found that f has non-removable singularities at 0, i, i. Lets look at each of them:
At 0:
We can represent f as
z+1
f (z) = (z 0)3 . 2
z +1
z+1
where our function h(z) = z2 +1 . This function is holomorphic on C{+i, i}. To fulfill the
conditions of the theorem, we need to find a D(0, ) s.t. h(z) is holomorphic on it and
doesnt vanish on it. How do we find such a disk? Well, the denominator of h vanishes
at i, i so we can take the radius of the disk to be 1. i.e. take the disk D(0, 1).
So 0 is a pole of order 3.
f (z) =

z 3 (z 2

At i:
We can represent f as

-i
f (z) = (z i)1 .
117

z+1
z 3 (z + i)

where the function h(z) = z3z+1

(z+i) . This function is holomorphic on C {0, i}. The disk
we can put around i s.t. h is holomorphic and non-vanishing on it is D(i, 1)
So i is a simple pole.

i
0
At i:
We can represent f as:
f (z) = (z + i)1 .

z+1
i)

z 3 (z

Where the function h(z) = z3z+1

(zi) . This function is holomorpic on C {0, i}. The disk we
can put around i s.t. h is holomorphic and non-vanishing on it is D(i, 1).

0
-i

Theorem 28
If f has a pole of order N at z0 , then we can write
f (z) =

aN
aN +1
a1
+
+

+
+ G(z)
(z z0 )N (z z0 )N 1
(z z0 )

with aN , aN +1 , . . . , a1 uniquely determined and G(z) holomorphic in some D(z0 , ) for

>0
Proof. We have
f (z) = (z z0 )N .h(z)
118

with h(z) holomorphic and different from 0 on D(z0 , ).

h(z) = A0 + A1 (z z0 ) + A2 (z z0 )2 + . . .
So,
f (z) =

A0
A1
A2
AN 1
+
+
+ +
+ G(z)
N
N
1
N
2
(z z0 )
(z z0 )
(z z0 )
(z z0 )

Set
aN = A0 , aN +1 = A1 , aN +2 = A2 , . . . a1 = AN 1

6.15

Residues and The Residue Theorem

Definition 60 (Residue)
The numerator of the term

a1
zz0 ,

a1 is called the residue of f at the pole z0 . We write

res(f, z0 ) = a1

Definition 61
a1
z z0
is called the principal part.
Theorem 29
If z0 is a simple pole of f (z), the residue is given by
res(f, z0 ) = lim (z z0 )f (z)
zz0

If z0 is a pole of order N , then the residues is given by

res(f, z0 ) =

1
d(N 1)
lim (N 1) (z z0 )N f (z)
(N 1)! zz0 dz

Proof. Part 1: N = 1
Since f has a simple pole,

a1
+ G(z)
z z0
And multiplying through by the denominator, we get
f (z) =

(z z0 )f (z) = a1 + G(z)(z z0 )
And now we take the limits of both sides as z z0 . We get
lim [(z z0 )f (z)] = lim [a1 + G(z).(z z0 )]

zz0

zz0

= a1 + lim (z z0 )G(z)
zz0

= a1 + G(z0 ) lim (z z0 ) , as G(z) is continuous

zz0

= a1 + 0 , since lim (z z0 ) = 0
zz0

119

Thus,
a1 = lim [(z z0 )f (z)]
zz0

Part 2: N > 1
(z z0 )N f (z) = aN + aN +1 (z z0 ) + aN +2 (z z0 )2 + . . . + a1 (z z0 )N 1 + (z z0 )N G(z)
Differentiating once, we get
d [(z z0 )N f (z)]
= aN +1 +2aN +2 (zz0 )+. . .+(N 1)a1 (zz0 )N 2 +N (zz0 )N 1 G(z)+(zz0 )N .G (z)
dz
Differentiating again, we get
d2 [(z z0 )N f (z)]
= 2aN +2 + 6aN +3 (z z0 )
dz 2
+ . . . + (N 2)(N 1)a1 (z z0 )N 3 + N (N 1)(z z0 )N 2 G(z) + 2N (z z0 )N 1 G (z) + (z z0 )N G (z)
We continue in this way until the N 1th derivative:
dN 1 [(z z0 )N f (z)]
= (N 1)(N 2)(N 3) . . . 1a1 + terms containing at least one power of (z z0 ),
dz N 1
which is multiplied either by G or a higher derivative of G
and all of these other terms go to 0 as z z0 . So we have
lim [

zz0

dN 1 (z z0 )N f (z)
] = (N 1)!a1
dz N 1

Thus
a1 =

1
dN 1
(z z0 )N f (z)
N
(N 1)! dz

Theorem 30 (The Residue Theorem)

Let f be holomorphic on an open set containing a toy contour and its interior except
for the poles z1 , . . . , zk inside . Then
k

f (z)dz = 2i res(f, zi )
i=1

This is called The Residue Formula.

Proof. Step 1
First of all, pick at region with contour along which to integrate. Let the points z1 ,
z2 , . . ., zk be the poles of the function we want to integrate.

120

f (z) is clearly not holomoprpic on this region as it is not even defined on the poles (they
are not removable singularities). However, f (z) is holomorpic on the keyhole region ,
since it is a region that excludes all singularities. So we have that
f (z)dz = 0

by Cauchys Theorem.
We represent the keyhole region as follows:
We are traversing the whole contour anticlockwise, which means that all of the small
almost-circles around the singularities are traversed clockwise.
Each small almost-circle is called C (zj ), where zj is a pole, 1 j k, and the pole is at
the centre of the circle.

The width of the corridor going to the keyholes is very small, 0.

When we let the width of the corridor tend to 0, i.e. 0, we essentially make the two
line segments which make up each corridor become the same line segment; but since the
original two line segments were traversed in the opposite directions, the integrals along
them actually cancel out. The contribution of the integrals along all the corridors is thus

121

0.
Then we have
f (z)dz

f (z)dz

C (z1 )

f (z)dz

C (z2 )

f (z)dz = f (z)dz = 0

C (zk )

Note that the signs come from the fact that we are traversing the small circles clockwise.
Step 2
We want to show that for each j, 1 j k, we have that
f (z)dz = 2ires(f, zj )

C (zj )

Now, we have
f (z) =

a1
an
+ ... +
+ G(z)
N
(z zj )
z zj

where G(z) is holomorpic in D(zj , ). Taking the integral on both sides, we get

C (zj )

f (z)dz =
C (zj )

aN
dz + . . . +
(z zj )N

C (zj )

a1
dz +
z zj

G(z)dz

C (zj )

Now,
G(z)dz = 0
C

since G(z) is holomorphic on D(zj , ), by Cauchys Theorem.

All the other integrals which have in the denominator a power of (z zj ) higher than 1 are
also 0, since we can find an antiderivative for all of them. For example, the antiderivative
of
aN
(z z )N dz
j

C (Zj )

is

aN
(z zj )N +1
N + 1
So the only non-zero term comes from
F=

C (zj )

a1
dz
z zj

and it is in fact equal to 2i.a1 , by Cauchys Integral Formula.

So we get
f (z)dz = 2ia1 = 2i.res(f, zj )
C (zj )

122

Step 3
When we go back to the expression we ended up with in Step 1, we get that
f (z)dz =

f (z)dz +

C (z1 )

f (z)dz + . . . +

C (z2 )

f (z)dz

C (zk )

= 2i.res(f, z1 ) + 2i.res(f, z2 ) + . . . + 2i.res(f, zk )

k

= 2i. res(f, zj )
j=1

Example 60
1
9 + z2
The isolated singularities are 3i, 3i. Lets look at the them in turn, find if they are poles
and of what order and find h(z) and a disk D on which h is holomorpic and non-vanishing
and calculate the residues.
What is the radius of convergence of the power series of f centred at 0? Centred at i?
Calculate

1
9 + x2 dx
f (z) =

1. Singularity 3i:
We can write f as

1
z + 3i
and it holomorphic on the disk D(3i, 6).
f (z) = (z 3i)1 .

where h(z) =

1
z+3i

3i
-3i

123

So we have a pole at 3i which is a simple pole. We can calculate the residue as

res(f, 3i) = lim (z 3i).f (z)
z3i

= lim (z 3i)
z3i

= lim

z3i

1
1
z 3i z + 3i

1
z + 3i

1
1
=
3i + 3i 6i

2. Singularity 3i:
We can write f as

1
z 3i
and it is holomorphic on the disk D(3i, 6).
f (z) = (z + 3i)1 .

where h(z) =

1
z3i

3i
-3i

So we have a simple pole at 3i. We can calculate the residue as

res(f, 3i) = lim (z + 3i).f (z)
z3i

= lim (z + 3i)
z3i

1
1
z 3i z + 3i

1
z3i z 3i
1
1
=
=
3i 3i
6i
3. Radius of convergence of f centred at 0 is R = 3 since f is holomorphic on D(0, 3)
We can represent f as a power series.
= lim

1
9 + z2
1
=
2
9 (1 + z9 )
1
=
2
9 (1 z9 )

f (z) =

124

and we recognise the geometric series

z 2 n 1 (1)n z 2n
1
(
) =

9 n=0
9
9 n=0
9n
with
an =

(1)n
9n+1

This is a geometric series as long as

z2
< 1 z < 3
9

Which confirms our result.

Note that we didnt need to use the definition for the power series of f given as

f (z) = an z n with an =
n=0

f (n) (0)
n!

4. Radius of convergence of f centred at i is R = 2 since f is holomorphic on D(i, 2).

5.

1
9 + x2 dx

First of all, note that we can calculate this integral using techniques from Analysis 2/
Methods 1. Lets try it this way:
M

1
1
lim
dx

9 + x2 dx = L,M

9 + x2
L

1
x M
1

[arctan ( )]
= ( ( )) =
L,M 3
3 L
3 2
2
3
lim

Now we are going to solve the same integral with complex analysis, using the same techniques as in the previous week but this time the integral on the contour that we pick
wont be 0, because the function we are going to choose to integrate will have singularities.
So start with the function
1
f (z) =
9 + z2
Take the contour to be (as in the picture) the upper semi-circle from R to R. Now, since
we are taking just the upper semi-circle, only some of the poles will lie in it.

125

3i
-R

-3i

The two poles are at 3i and 3i, and only the one at 3i lies within our contour. So we
can calculate the integral on with the Residue Formula. It is
1
9 + z 2 dz = 2i.res(f, 3i)

1
= 2i.
6i

= , as expected
3
And now lets compare this with the integrals along the curves out of which is made,
i.e.
1. The line segment [R, R] parametrised by z(x) = x for R x R.
2. The semi-circular curve R parametrised by z(t) = Reit for 0 t
So we have
f (z)dz = f (z)dz + f (z)dz
R

[R,R]

Lets first calculate the second integral on the RHS. We have the parametrisation z(x) = x,
so z (x) = 1 and thus f (z(x)).z (x) = f (x). So we have
R

f (z)dz = f (x)dx

[R,R]

R
R

1
dx
9 + x2

1
dx as R
9 + x2

And the other integral is

RRR
RRR
RRR
RRR
1
1
RRR.
dz
R

(R).
max

R
zR 9 + z 2
RRR 9 + z 2 RRRR
R R
R
1
1
= (R) max
(R) 2
0 as R
z=R,y0 9 + z 2
R 9
126

We got from to the last line by applying the triangle inequality to

9 + z 2 z 2 9 = R2 9
So the contribution from this line integral is 0 and thus we get the result we expected.

6.16

Liouvilles Theorem: Generalisation

We saw an application of Liouvilles Theorem: If f is entire and bounded, then

f (z) M z1/2 , z, z > R f = const
And we said that this is fact true for any fractional power of z which is less than 1.
Then what happens if we take it to be greater than 1? The following example generalised
Liouvilles Theorem.
Example 61
Let f be entire and assume that for some M > 0, N N,
f (z) M (1 + z)N , z C f is a polynomial of degree less than or equal to N
Proof. Since f is entire, it has a power series at 0 with an infinite radius of convergence.

f (z) = an z n
n=0

with
an =

f (n) (0)
n!

So
f (z) = a0 + a1 z + a2 z 2 + + aN z N + aN +1 z N +1 + . . . ...
Our aim will be to show that all the terms after aN z N are 0, i.e. an = 0 n > N . So
consider the following
maxz=R f (z)
f (n) (0)

, by Cauchys Inequality
n!
Rn
maxz=R M (1 + z)N M (1 + R)N
=

Rn
Rn

an =

Now what happens to M (1+R)

as R ? N is a fixed number s.t. n > N , so the
Rn
denominator tends to infinity faster that the numerator. So the whole expression tends
to 0 and thus we have proved that
an = 0, n > N
Therefore, f is a polynomial of degree less than or equal to N .
127

6.17

(T) The Principle of Analytic Continuation

Theorem 31 (The Principle of Analytic Continuation)

Let f be holomorphic in a region , and let wk for k = 1, 2, 3, . . . and z0 , with zo =/ wk ,
k N and
lim wk = z0
k

(Such a point z0 is called a limit point in the region)

Moreover, we assume that f (wk ) = 0, k N.
Then
f (z) = 0, z
Proof. f is holomorphic at z0 , so f has a power series

f (z) = an (z z0 )n
n=0

and it converges inside the radius of convergence R, i.e. for z z0 < R.

Assume that f (z) is not identically 0 on the disk z z0 < R. We have seen that there
exists a minimum N s.t. aN =/ 0 and then f (z) = (z z0 )N g(z), with g(z) holomorphic
and not-vanishing on the disk D(z0 , ), so f cannot have other zeroes on the disk D(z0 , )
apart from the z0 . (which we get from (z z0 )N = 0 z z0 = 0 z = z0 ).
But then we have the following picture:

wj

z0

w3

w1

w2

w4

w5

While f (z) =/ 0 on the disk D(z0 , ) {z0 }, contained in out region , and z0 is the single
zero, there are also other zeroes - some of the wj s. So z0 is not in fact the only zero in
the disk. The argument written rigorously reads:
This is contradiction as wk z0 , so we can find a wk D(z0 , ) with wk =/ z0 and f (wk ) = 0
128

by assumption. This implies that in fact f (z) = 0, z D(z0 , ).

Thus we have proved by contradiction that f is identically 0 in a disk centred at z0 .

f (z) = bn z z0 n
n=0

where
bn =

f (n) (z0 )
=0
n!

since f is zero on D(z0 , ).

To prove that f is identically 0 everywhere in the region , we do the following: We pick
a point z1 which lies inside the disk D(z0 , ) close to the boundary and we draw a disk
around z1 or radius 1 . Then f is identically zero on this new disk. We then pick a point
z2 inside the disk D(z1 , 1 ) close to the boundary and draw a circle around it. Then f is
identically 0 on this new disk. If we continue in this way, we can connect the point z0 to
any point inside the region and thus show that f is identically 0 in the whole region.

some random point we

want to connect to

z0

Corollary 11
If f and g are holomorphic on a region and f (z) = g(z), z D(a, ) . (or even
f (wk ) = g(wk ), k N , with < wk > having a limit point z0 ), then
f (z) = g(z) in the whole of

129

130

Example 62

f (z) = z n , z < 1
n=0

Let = {z, z < 1}. f is holomorphic on

Now, the function

1
1z
is holomorpic everywhere except for 1, so on = {z C {1}}
But the functions f and g are connected, since
g(z) =

zn =
n=0

1
for z < 1
1z

So for
f C
g C
with , we say that g is the analytic continuation of f from the smaller domain
to the larger domain .
Example 63
Does there exist a holomorphic function on D(0, 1) with
1
f ( ) = (1)n , n N
n
No. f is not even continuous on this disk. Consider
1
1
lim f ( ) = f (0) , since lim 0
n
n
n

So
lim (1)n = f (0)

But this limit doesnt even exists. f is not continuous and thus certainly not holomorphic.
Example 64
Find all holomophic functions on D(0, 1) with
1
n
f( )=
n
n+1
There exists at most one such function. If there existed another such function then
1
1
f ( ) = g ( ) , n N
n
n

131

But n1 0, n1 =/ 0; 0 is a limit point on D(0, 1). Therefore, f (z) = g(z) z D(0, 1) by

the corollary of The Principle of Analytic Continuation
To find the function, we set z = n1 n = z1 and we get
f (z) =

1
z
1
z

+1

1
z
1+z
z

1
z+1

Example 65
The power series

z2 z3 z4
+ + ...
2
3
4
has radius of convergence 1, so it represents a holomorphic function in the disk D(0, 1).
We recognise this power series as
z

log(1 + z) = z

z2 z3 z4
+ + ...
2
3
4

As in a previous example, here log is holomorphic in a bigger domain that the one of the
power series. The function requires a cut from to 1, but is continuous everywhere
else. i.e. on C {x, x 1}, log(1 + z) is holomorphic.
Thus log(1 + z) is the analytic continuation of the power series.
What about the power series expansion around z0 = 1? Wed need to work them out from
first principles by using
f (z0 ) = f (z0 )(z z0 ) +
Where

f (z0 )
f (z0 )
(z z0 )2 +
(z z0 )3 + . . .
2
3!

f (1) = log(2)
1
1
f (1) =
(1) =
1+z
2
1
1
f (1) =
(1) =
2
(1 + z)
4
2
f (1) =
8
(n 1)!
(n 1)!
f (n) (1) = (1)n1
= (1)n1
n
(1 + n)
2n

so around the point z0 = 1,

(1)n1 (n 1)n
2n .n
n1

log(1 + z) = log(2) +

132

6.18

(M) Solving integrals of the type

0

p(x)
q(x) dx

Method 2 (Solving integrals)

of the type

p(x)
q(x) dx

or

p(x)
q(x) dx
0

where p(x) and q(x) are real polynomials with no common factors and q(x) has no real
roots.
We use the upper or lower semi-circular contour, traversed in the positive sense. The
only residues we need to consider are the ones inside the contour, so generally only half of
them, if the function is even. This saves work when calculating the residues and applying
the Residue Theorem.
Example 66
Calculate

2x2 1
x4 + 5x2 + 4 dx
0

First of all, this is an even function, so it is in fact equal to

1
2x2 1
dx
2 x4 + 5x2 + 4

and this is the function we are going to aim to get.

Lets work with the upper semi-circular contour and the function
f (z) =

2z 2 1
z 4 + 5z 2 + 4

First we need to find the poles of the function. These occur when the denominator is 0,
so
z 4 + 5z 2 + 4 = (z 2 + 1)(z 2 + 4)
= (z i)(z + i)(z 2i)(z + 2i)
z = i, i, 2i, 2i
In fact the only poles that lie within the region we are considering are i and 2i.

133

Next we calculate the residues:

Res(f, i) = lim(z i)f (z)
zi

2z 2 1
zi
(z i)(z + i)(z 2i)(z + 2i)
2
2z 1
= lim
zi (z + i)(z 2 + 4)
2i 1
=
(i + 1)(1 + 4)
3
1
=
=
2i + 3 2i
= lim(z i)

Res(f, 2i) = lim (z 2i)f (z)

z2i

2z 2 1
z2i
(z 2i)(z + 2i)(z 2 + 1)
2z 2 1
= lim
z2i (z + 2i)(z 2 + 1)
8 1
3
=
=
(4i)(4 + 1) 4i

= lim (z 2i)

Then, by the Residue Theorem, we have

f (z)dz = 2i (Res(f, i) + Res(f, 2i))

= 2i (
=

1 3
+ )
2i 4i

Now, as usual, we evaluate the same integral by splitting the contour into pieces, integrating and then comparing results.
The contour consists of the line segment [R, R], parametrised by z(x) = x, R < x < R.
and the semi-circle given by R , parametrised by z(t) = Reit , 0 < t <
Lets first evaluate the line segment integral:
R

f (z)dz = f (x)dx

[R,R]

R
R

2x2 1
dx
x4 + 5x2 + 4

2x2 1
dx ,as R
x4 + 5x2 + 4

134

This is what we want ot get in the end, so we expect the contribution from the other
integral to be 0. We are going to prove this by considering the modulus of the integral:
RRR
RRR
RRR
R
RRR. f (z)dz RRRRR R max f (z)
z=R
RRR
RRR
R R
R
2z 2 1
= R max 4
z=R z + 5z 2 + 4
2z 2 + 1
R ( 2
) from the Triangle Inequality
(z + 1)(z 2 + 4)
= R (

2R2 + 1
)
(z 2 + 1)(z 2 + 4)

For the final step, we need the Triangle Inequality again

z 2 + 1 z 2 1 = R2 1
z 2 + 4 z 2 4 = R2 4
So we get

RRR
RRR
RRR
R
2R2 + 1
RRR f (z)dz RRRRR
0 as R
RRR
RRR (R2 1)(R2 4)
RR
R
So the integral along R is indeed 0.
So we end up with

2x2 1

1
dx =

4
2
2
x + 5x + 4
2

And so

2x2 1
x4 + 5x2 + 4 dx =
0

6.19

(M) Solving integrals of the type

or

p(x)
q(x)

sin(ax)dx

Method 3 (Solving integrals)

of the type

p(x)
q(x) cos(ax)dx

p(x)
q(x) sin(ax)dx

135

p(x)
q(x)

cos(ax)dx

For a > 0, we use the upper semi-circular contour and the function
f (z) =

p(z) iaz
e
q(z)

For a < 0, we use the lower semi-circular contour and the same function.
The thing which is useful in choosing this function is that when we consider the modulus
of the integral along the semi-circular arch, and when we bound it above, we can use the
fact that
eiaz = eia(x+iy) = eiax .eay
= eiax eay = 1.eay 1 since a > 0, y 0
Example 67
Show that

cos x
(x2 + 1)2 dx = e

We consider the function

eiz
(z 2 + 1)2
and the semi-circular contour from R toR, where the arch is called R .
The function has poles at z 2 + 1 = 0 z = i, i, but only one of them is inside the
region we are considering, so we only need to compute the residue at i. However, this
is a double pole of f , so we need to apply the appropriate formula when computing the
residue. We have
d
res(f, i) = lim (z i)2 f (z)dz
zi dz
d eiz
= lim
zi dz (z + i)2
ieiz (z + i)2 2(z + i)eiz
= lim
zi
(z + i)4
eii (i + i)2 2(i + i)eii
=
(i + i)4
ie1 (4) 4ie1
=
16
8ie1 i
=
=
16
2e
So we have that, by the Residue Theorem,
f (z) =

eiz
i
(z 2 + 1)2 dz = 2ires(f, i) = 2i 2e = e

Now we calculate the same integral over the curve split into the line segment [R, R]
and the semi-circular arch R , so that

=
f (z)dz + f (z)dz
e
R

[R,R]

136

Consider first the integral of R :

RRR
RRR
RRR
R
RRR f (z)dz RRRRR R max f (z)
z=R
RRR
RRR
Rr
R
Not, consider just
eiz

f (z) = 2
(z + 1)2
We aim to bound the numerator below and the denominator above, so we have
eiz = ei(x+iy) = eixy = eix ey = ey 1 , since y 0
z 2 + 1 z 2 1 = z2 1 = R2 1 on z = R
So we have

RRR
RRR
R
RRR
1
RRR f (z)dz RRRRR R 2
0 as R
(R 1)2
RRR
RRR
R
RR
So the contribution of the integral on the semi-circular arch is 0. Now the one on the line
segment:
Let z(x) = x s.t. R x R and f (z(x))z (x) = f (x). So we get
R

[R,R]

f (z)dz =

eix
dx
(x2 + 1)2

R
R

cos x + i sin x
dx
(x2 + 1)2

R
R

sin x
cos x
sin x
cos x
dx + i
dx
dx + i
dx
=
2
2
2
2
2
2
(x + 1)
(x + 1)
(x + 1)
(x2 + 1)2

So we end up with

sin x

cos x
(x2 + 1)2 dx + i (x2 + 1)2 dx = e

Comparing real and imaginary parts gives us

cos x

(x2 + 1)2 dx = e

and

sin x
(x2 + 1)2 dx = 0

where the question didnt ask fo the second part, but we got it anyway. The result isnt
surprising as the function we are integrating is odd, so the area under the graph to the
right of the x-axis cancels the area to the left of the x-axis.
137

6.19.1

Jordans Lemma

Lemma 4 (Jordans Lemma)

Let R be the curve z = Reit with 0 t .
Suppose that
max f (z) 0 as R
zR

and take a > 0. Then

lim eiaz f (z)dz = 0

Proof. Consider the segment (the RHS part of it only. We are going to consider the LHS
of it later. One diagram just because I was too lazy to make two separate ones)

iR

CR

-R

We use the parametrisation z(t) = Reit , where 0 t

and z (t) = iReit . We have

RRR
RRR /2
RRR
R
it
RRR. eiaz f (z)dz RRRRR = eiaRe f (Reit )iReit dt
RRR
RRR
R CR
R 0
/2
it

eiaRe f (Reit )iReit dt

0
/2
2

eaR t f (Reit )R dt
0

where we got to the last line from the previous one using
it

eiaz = eiaRe = eiaR cos taR sin t

2t

= eiaR cos t eaR sin t = 1.eaR sin t eaR as a > 0

138

So weve got
/2

/2

aR 2 t

2t

it

f (Re )R dt R max f (z) eaR dt

zR

2t

eaR 2
= R max f (z) [
] after integrating the above
zR
aR 2 0

= max f (z) ( eaR )

zR
2a 2a

=
max f (z)(1 eaR ) 0 as R since
2a zR
max f (z) as R by the statement of the teorem
zR

To prove the lemma, it remains to prove that the integral along the other part of the
curve also goes to 0 as R . But this integral is
RRR
RRR

RRR
RRR
iaz
RRR e f (z)dz RRR eaR sin t Rf (Reit )dt
RRR
RRR
RRCR
RR /2

R max f (z) eaR sin t dt

zR

/2
/2

= R max f (z) eaR sin t dt

zR

since

/2

aR sin t

e
0

dt = eaR sin t dt
/2

by symmetry The proof is identical afterwards.

Example 68
Claim:

u for 0 < u

2
We proved this with differentiation in a homework. Here is another argument:
sin(u)

139

sin(u)

pi/2

Since (sin u) = sin u 0 on 0 < u 2 , g(u) = sin(u) is concave downwards.

We draw the following graph with two secant lines: Slope of the green secant segment =
sin(u)sin(0)
= sin(u)
u0
u
2
10
Slope of the blue secant segment = 0
=

2
Since the first slope is obviously bigger than the second one, we have
sin(u) 2
2u

sin(u)
u

6.20

(M) Solving integrals of the type F (sin t, cos t)dt

0

Method 4 (Solving integrals)

of the type

F (sin t, cos t)dt

0

We use the unit circle as our contour and take the parametrisation
z = eit , 0 t 2
Then

z = ieit , so dz = ieit dt = izdt

And we write

eit + eit z + z 1
=
2
2
eit eit z z 1
sin t =
=
2i
2i

cos t =

So we end up with
2

F (sin t, cos t)dt = F ((

0

z=1

140

z z 1
z + z 1 dz
),(
))
2i
2
iz

Example 69
Prove that

2
dt
, where 1 < a < 1
1 + a sin t =
(1 a2 )
0
Taking
a sin t = a

z z 1
2i

we get, using the method above

2

1
1
dz
1 + a sin t dt =
zz 1 iz
1 + a 2i
0
z=1

2i
dz
=
1
2i + a(z z ) iz
z=1

2dz
=
2iz + az 2 a
z=1

2
=
2
az + 2iz a
z=1

We have different cases for a in the given interval. If a = 0, then the problem reduces to
2dz
2
1
2iz = z dz = 2i 2i = 2
z=1

z=1

If a =/ 0, then we are back to the integral again and we solve it as usual with the Residue
Theorem, but first we need to find the poles which lie inside our contour, the unit circle.
The poles are

2i + / (4 4a(a))
2
az + 2iz a = 0 z1,2 =
2a

2
2i + / 4 + 4a
i + / 1 + a2
=
=
2a

a
2
i + / i 1 a
1 + / 1 a2
=
= i(
)
a
a
where we wrote the roots as purely imaginary in the last step, since 1 < a < 1, a =/ 0.
Which of these 2 roots lie inside the unit circle? We will prove that

1 + 1 + a2
z1 =
a

141

lies inside the unit disk, z 1

1 + 1 + a2
1 + 1 a2

1
1
a
a

1 1 a2

1 1 1 a2 a
a

(1 + a a2 ) 2 1 a2 < a2 1 + 1 a2 a2 < 2 1 a2

2(1 a2 ) < 1 a2 1 a2 < 1

which is true.
Where is the other root? Applying Vietas formuli on the original equation, we get that
z1 .z2 = a
a = 1. So the product of the roots is -1 and so the modulus of the product is 1.
We can use this fact to determine if the other root lies inside the unit circle.
z1 z2 = 1 z2 =
But z1 < 1 and so
z2 =

1
z1

1
>1
z1

so the other root is outside the unit circle. We can now compute the residue of the simple
pole z1 . Note that we can factorise f as
f=

2
a(z z1 )(z z2 )

res(f, z1 ) = lim (z z1 )
zz1

2
a(z z1 )(z z2 )

2
zz1 a(z z2 )
2
=
a(z1 z2 )
= lim

Now
z1 z2 = i (

1 +

1 a2 1 1 a2
2i 1 a2

)=
a
a
a

So
res(f, z1 ) =

2
1

=
2i 1 a2 i 1 a2

Then the Residue Theorem gives

1
2
=
f (z)dz = 2ires(f, z1 ) = 2i
i 1 a2
1 a2
z=1

142

6.21

(T) Laurent Expansion

Theorem 32 (Laurent Expansion)

Let f be holomorphic on {z; R1 < z z0 < R2 } with 0 R1 < R2 . Then we have an
expansion

f (z) = an (z z0 )n on the above set

n=

where
an =

1
2i

Cr (z0 )

f (w)
dw
(w z0 )n+1

where R1 < r < R2

Note that the theorem is very similar to the power series expansion of f , except that in
this case, we are working over the annular region and we allow negative powers in (z z0 )
in the sum.
The proof is also very similar to the proof of the Taylor expansion.
Proof. In this proof we are going to include two circles, of radii r1 and r2 inside the big
circle ., s.t. the new circles lie between the circle of radius R1 and the circle of radius
R2 , as in the picture below:

R1
r
1
As we let the width of the corridor of the keyhole go to 0, the contribution of the integral
over the corridors goes to 0.
So, we have
R1 < r1 < z z0 < r2 < R2
And now we are going to apply Cauchys Integral formula of the integral over , which is
the circle of radius r2 Note that we are traversing the circle of radius r2 in the positive
143

sense, and the circle of radius r1 in the negative sense, since we go from one to the other
via the corridors. So we get
f (z) =

f (w)
1
dz

2i
wz

1
=
2i

f (w)
1
dw
wz
2i

wz0 =r2

f (w)
dw
wz

wz0 =r1

+ 0 which is the contribution of the corridors

Consider first w z0 = r2 . We have that

z z0
<1
w z0

since the ws lie on the circle. Now we have

1
1
=
w z w z0 + z0 z
1
1
=
z0 z
w z0 1 + wz
0
1
1
=
zz0
w z0 1 wz
0
1
z z0 n
=
)
(
w z0 n0 w z0
Thus

wz0 =r2

f (w)
dw =
wz
n0

f (w)
dw(z z0 )n
(w z0 )n+1

wz0 =r2

Consider now w z0 = r1 , where we have that

We have

w z0
<1
z z0

1
1
=
w z w z0 + z0 z
1
1
=
.
0
z z0 1 wz
zz0
=

Thus

zz0 =r1

f (w)
dw =
wz
k0

1
w z0 k
)
(
z z0 k0 z z0
f (w)(w z0 )k (z z0 )(1+k) dw

wz0 =r1

144

Set k + 1 = n, so that k 0 n < 0, so we can rewrite the above as

n<0
wz0 =r1

f (w)
dw.(z z0 )n
(w z0 )n+1

So we end up with
f (z) =

1
f (w)
1
f (w)
f (w)
n
n

dw =
dw(z

z
)

dw(z

z
)

0
0

2i
wz
2i n0
(w z0 )n+1
(w z0 )n+1
n<0

wz0 =r2
wz0 =r1

The final step is to show that this is valid when integrating over Cr (z0 ), where R1 < r < R2
By Cauchys Theorem, as long as the function we are integrating is holomorphic (which
it is in this case), we have

Cr (z0 )

Cr (z0 )

f (w)
dw
(w z0 )n+1

f (w)
dw = 0
(w z0 )n+1

f (w)
dw =
(w z0 )n+1

f (w)
dw
(w z0 )n+1

f (w)
dw
(w z0 )n+1

f (w)
dw = 0
(w z0 )n+1

f (w)
dw =
(w z0 )n+1

f (w)
dw
(w z0 )n+1

Cr1 (z0 )

Cr2 (z0 )

Cr (z0 )

Cr (z0 )

Cr1 (z0 )

Cr2 (z0 )

Where r is greater than r1 and r2 , but still less than R2 .

So now we have
f (z) =

1
f (w)
1
f (w)
f (w)
n
n

dw
=
dw(z

z
)

dw(z

z
)

0
0

2i w z
2i n0
(w z0 )n+1
(w z0 )n+1
n<0

wz0 =r
wz0 =r

Which is the same as

f (z) = an (z z0 )n on the above set

n=

where
an =

1
2i

Cr (z0 )

f (w)
dw
(w z0 )n+1

Example 70
Find the Laurent expansion of
g(z) =

z
(1 + z)(1 z)

for
1. z < 1
145

2. z 1 > 2
Solution: Part 1
In z < 1, the function doesnt have a vanishing denominator, so g is holomorphic on
z < 1 and so we write the Taylor series. We need to write it in the form

g(z) = an z n with an =
n=0

1
2i

C1 (0)

f (w)
g (n) (0)
dw
=
(w 0)n+1
n!

But we dont actually need to work from these definitions (in general), since we can just
consider the geometric series. We are first going to split the function into partial fractions:
B
z
A
+
=
1 + z 1 z (1 + z)(1 z)
We find that A =

1
2

and B = 12 . So we have
1
1
1
1
)+ (
)
g(z) = (
2 1+z
2 1z

which we can write as a geometric series

1
1
1
1
g(z) = (z)n + (z)n = ((z)n + (z)n ) = ((1)n+1 + 1) z n
2 n=0
2 n=0
2 n=0
2 n=0
and this is the final answer.
Solution: Part 2
We need to find the expansion outside the disk z 1 > 2

-1

The centre of the disk is at 1, so the Laurent expansion should be of the form

g(z) = an (z 1)n
n=

So we need to express the function in terms on (z 1) before doing anything else to it.
2
The condition z 1 > 2 means that z1
< 1.
1
Consider first 1+z . We can write it as
1
1
1
=
=
2
1 + z 2 + z 1 (z 1) ( z1
+ 1)
146

Thus we can write

z
z1+1
=
2
(1 + z)(1 z) (z 1) ( z1
+ 1) (1 z)
1
(z 1)
+
=
2
2
(z 1) ( z1 + 1) (1 z) (z 1) ( z1 + 1) (1 z)
1
1
=
+
2
2
2
(z 1) ( z1 + 1) (z 1) ( z1
+ 1)
n

2
1
2 n
(1)
(
)

(
)
=

z 1 n=0 z 1
(z 1)2 n=0 z 1

g(z) =

= (2) (1)(z 1)

n1

n=0

(2)k (z 1)k2
k=0

Set k 2 = n 1 k = n 1 + 2 = n + 1 k = n 1 so we have that

g(z) = (2)n (1)(z 1)n1 (2)n1 (z 1)n1

n=0

n=1

So need to start from n = 1. The term for n = 0 from the first series is

g(z) = [(2)n (1) (2)n1 ] (z 1)n1

n=1

1
z1 .

So we have

1
z1

Note that when we write this out explicitly, the terms in the expansion start at
Example 71
Find

1
1+z .

log x
(1 + x2 )2 dx
0

Let
f (z) =

log z
(1 + z 2 )2

The contour we are going to consider is the intended upper semi circle. Now, if we took
the principle logarithm, it would not be defined on part of our contour - the negative real
axis. This is why we dont choose the principle log, but instead we restrict the argument
to lie between 2 and 3
2 , i.e. we have chosen the cut for the log to be on the negative
imaginary axis. Now we only have one discontinuity to worry about, and this is at the
origin, which is why we take the intended semi-circle. So now we have a well-defined
contour along which to integrate. First we consider the poles of f (z). They occur at
(z 2 + 1)2 = 0 (z + i)2 (z i)2 = 0 z = + i

147

So the points i and i are poles of order 2. However, only the pole at i lies in our contour.
So we only need to find the residue there. We have
d
((z i)2 f (z))
zi dz
d (log(z))(z i)2
= lim (
)
zi dz
(z + i)2 (z i)2

res(f, i) = lim

d log(z)
(
)
dz (z + i)2
1
(z + i)2 2 log(z)(z + i)
= lim ( z
)
zi
(z + i)4
1
(i + i)2 2 log(i)
)
=(i
(i + i)4
1
2i 2 log(i)
= i
(2i)3
2 2 log(i)
=
8(i)
2 2 log i 2arg(i)
=
8i
2 i
=
8i
= lim
zi

So by the Residue Theorem, we have that

2 i
2 i
f (z)dz = 2i ( 8i ) = 4

Again we are going to evaluate the same integral again by integrating along the contour
we have chosen. We split the contour into 4 parts:
the line segment [R, r] parametrised by z(x) = x for R x r
the line segment [r, R] parametrised by z(x) = x for r x R
the semi-circle R
the semi-circle r
Lets look first at the integral around the segment [r, R]. We have
R

log(x)
f (z)dz = (1 + x2 )2 dx
[r,R]

and this is ok as it is. Lets now look at the integral on the other line segment. Its
important to note that on the other segment, we are working on the negative real axis,
148

which means that log(x) is not the usual real logarithm, but it is actually a complex
logarithm, given by
log(x) = log x + i
Since the argument on the negative real axis is . Here x is negative, s we leave it a
modulus. So now lets consider the integral with its parametrisation:
r

log x
1
log x + i
dx =
dx + i
f (z)dz =
2
2
2
2
(1 + x )
(1 + x )
(1 + x2 )2

[R,r]

Make the substitution x = u, so that u is now a positive number. The limits of integration
become, at x = R, u = R and at x = r, u = r. Also, dx = du and so we now have
r

log(u)
log x
(x2 + 1)2 dx = (u2 + 1)2 (du)
R

=
r

log(u)
du
(u2 + 1)2

where the - sign of du cancels with the - from flipping the limits of integration
But u can just be swapped for another variable, so lets pick x, so that we have
R

log(x)
(x2 + 1)2
r

So in total, we have from the two integrals over the line segments,
R

log(x)
log(x)
i
(x2 + 1)2 dx + R (x2 + 1)2 dx + (x2 + 1)2 dx
r

We let R and r 0 and we get

0

log(x)

2
dx + i
dx
2
2
(x + 1)
(1 + x2 )2
0

Now lets look at the integral over gaR . We have

RRR
RRR
RRR
RRR
log(z)
log(z)
RRR.
dz
R

R
max

R
zR (1 + z 2 )2
RRR (1 + z 2 )2 RRRR
R R
R
We have that

log(z) = logz + iarg(z)

log z + iarg(z)
log(R) + on z = R
149

Here we used the triangle inequality, the fact that the largest argument that z can have
on the upper semi-circle is and that i = 1. Similarly, we bound the denominator by
using the triangle inequality
1 + z 2 z 2 1
= z2 1 = R2 1 on z = R
Putting this together, we get that
RRR
RRR
R
RRR
log(R) +
log(z)
dz RRRRR R (
) 0 as R
RRR
2
2
(R2 1)2
RRR
RRR (1 + z )
R
RR
Where the last limits is obtained by applying LHopitals Rule on R log(R) and then by
taking standard limits.
Now for the last integral, over r , we have that
RRR
RRR
R
RRR
log(z)

RRR f (z)dz RRRRR r max

zr (1 + z 2 )2
RRR
RRR
R
Rr
First of all, we can assume that r is very small, and in particular that it is less than 1
(otherwise the pole at i wouldnt lie in the region we are considering). We bound the
numerator in the same way as before
log(z) = log z + iagr(z) log(r) + i on z = r
For the denominator, we use again the triangle inequality, but in this case we dont write
1 + z 2 z 2 1 = r2 1
since we are assuming that r < 1 and we dont want to bound the modulus of the integral
above with a negative number as it wouldnt tell us much about what is happening. So
instead we take the other possibility of the triangle inequality. We have that
1 + z 2 1 z 2 = 1 r2
Substituting this all back we end up with
RRR
RRR
RRR
R
log(r) +
RRR. f (z)dz RRRRR r
0 as r 0
(1 r2 )2
RRR
RRR
R r
R
Where the last limit can again be calculated with LHopitals Rule.
Since the contributions of the last two integrals are zero as R and r 0, we get
overall that

log(x)
1
2 i 2
2
dx
+
i
dx
=
=
+
i
2
(x2 + 1)2
(1 + x2 )2
8i
4
2
0

150

The last step is to compare the real and imaginary parts. We get the answer we want
from the real part:

log(x)

log(x)

2
dx =

dx =
2
2
2
2
(x + 1)
2
(x + 1)
4
0

The Argument Principle Part 1

If f is holomorphic and has a zero of order N at a point z0 , then we can write
f (z) = (z z0 )N g(z)
where g(z) is a holomorphic function at some small disk D(z0 , ) for > 0 and g(z) =/ 0
for z D(z0 , ).
Differentiating this expression for f , we get
f (z) = N (z z0 )N 1 g(z) + (z z0 )N g (z)
and dividing through by f , we get that
N
g (z)
f (z) N (z z0 )N 1 g(z) + (z z0 )N g(z)
=
=
+
f (z)
(z z0 )N g(z)
z z0 g(z)
Here z D(z0 , ) and the function

g (z)
g(z) is
(z)
of ff (z)

holomorphic on D because g(z) =/ 0. Then we

can see that z0 is a pole of order 1

N
Its residue is N , which is the principle part of (zz
.
0)
Part 2
Suppose that f has a pole of order m at w0 . Then we can write
f (z) = (z w0 )m h(z)
where h is holomorphic on D(w0 , ) and h(z) =/ for z D(w0 , ). Then we have that
m
h (z)
f (z) m(z w0 )m1 h(z) + (z w0 )m h (z)
=
=
+
f (z)
(z w0 )m h(z)
z w0 h(z)

(z)
is holomorphic as h(z) =/ 0. Thus
where hh(z)
m.
We put this formally into the

f (z)
f (z)

has a simple pole at w0 with residue

Theorem 33 (The Argument Principle)

Suppose that f is holomrphic on an open set containing the toy contour C and its interior
with the exception that f can have a finite number of poles at f (z) in its interior.
Suppose that f (z) =/ 0 for z C.
Let N be the number of zeroes of f (z) which are inside C, counted with their multiplicities
and let P be the number of poles of f inside C counted with their multiplicities. Then
1
f (z)
dz = N P

2i
f (z)
C

151

z1

wk

w2

zk

z2

w1

Proof. We use the Residue Theorem on

f (z)
f (z)
This is holomorphic inside C for z not a pole or a zero of f (z).
Let the poles of f be at w1 , w2 , . . . , ws and let the zeroes be at z1 , z2 , . . . , zk .
Then
f (z)
f (z) dz = 2i residues
C

= multiplicities of the zeroess of f multiplicities of the poles of f

=N P

Theorem 34 (Rouches Theorem)

Suppose that f and g are holomorphic on an open set containing a toy contour C and its
interior.
Assume that
f (z) > g(z), for z C
Then f and f + g have the same number of zeroes inside C.
Note: We are considering zeroes inside C.
Proof.
f (z) > g(z) 0 for z C f (z) =/ 0 for z C
(f + g)(z) = f (z) + g(z) f (z) g(z) > 0 for z C
Therefore,

f (z) + g(z) =/ 0 on C

152

Call Nf +g the number of zeroes of f + g. Let Nf be the number of zeroes of f . (There are
no poles.)
Applying the Argument Principle, we get
Nf =

f (z)
1
dz
2i f (z)
C

Nf +g

1
(f + g) (z)
dz
2i (f + g)(z)
C

Then we get that

Nf +g Nf =

(f + g )(z) f (z)
1
(

) dz

2i
f (z) + g(z) f (z)
C

1
(f + g )f (f + g)f
(z)dz
2i
f (f + g)
C

g f gf
1
=
(z)dz
2i f (f + g)
C

Now we notice the following

g f gf
f2
=
g
1+ f
1 + fg
(g f gf )f

(1 + fg )

f 2 (f + g)

g f gf
f (f + g)

which is exactly what we are trying to integrate above. So we can write

(1 + fg )
1
Nf +g Nf =
(z)dz
2i (1 + fg )
C

And now we notice that we actually have

(log(1 + g/f )) =

(1 + g/f )
(1 + g/f )

so we have found an antiderivative for the integral above and the contour is closed, and
hence, by the Antiderivative Theorem, the integral is 0.
It remains only to show that the antiderivative we have found is actually legitimate.
Now, log(1 + g/f ) is the composition of the functions h(z) = 1 + g/f and H(z) = log(z).
We are going to show that the function h never takes value on the negative real axis, so
the composition of h and H is a well-defined holomorphic function.
By the statement of the theorem, we have that g(z) < f (z). We can use this as follows
g(z)
<1
f (z)
g
(1 + (z)) 1 < 1 h 1 < 1
f

g(z) < f (z)

153

but this is the equation of the disk centred at 1 on the real axis, with radius 1. So all the
values of h are inside this disk. In particular, h doesnt touch the negative real axis. So
we can just take the standard cut for the principal logarithm and then the composition
of h and H is a well-defined function and it is the antiderivative we want.

1+f/g
0

Example 72
How many zeroes of
z 7 4z 3 + z 1 = 0
are inside the unit disk? What about the disk or radius 2?
We have, on z = 1,
z 7 = 1, 4z 3 = 4, z = 1, 1 = 1
Let
f (z) = 4z 3
g(z) = z 7 + z 1
Then
g(z) = z 7 + z 1 z 7 + z + 1 = 1 + 1 + 1 = 3
f (z) = 4
so
g(z) < f (z)
Therefore, f (z) = 4z 3 has the same number of zeroes inside z = 1 as does f (z) + g(z) =
4z 3 + z 7 + z 1
Now, f (z) = 0 4z 3 = 0 z = 0 with multiplicity 3. Therefore z 7 4z 3 + z 1 has
3 zeroes inside z = 1.
Where are the other 2 roots? Lets try the circle with radius 2:
On z = 2,
z 7 = 128, 4z 3 = 32, z = 2, 1 = 1.
Since 128 > 32 + 2 + 1 = 35, take
f (z) = z 7
g(z) = 4z 3 + z 1
154

Then
g(z) = 4z 3 + z 1 4z 3 + z + 1 = 32 + 2 + 1 = 35
f (z) = z 7 = 128
So g(z) < f (z) and so f and f + g have the same number of zeroes in side z = 2. Now
f (z) = 0 z 7 = 0 z = 0
with multiplicity 7. Therefore
f (z) + g(z) = z 7 4z 3 + z 1
has the same number of root inside z = 2 and f , i.e. there are 7 roots.
Thus all the roots of the equation lie inside the disk of radius 2.
Note that this doesnt mean they dont lie in a smaller disk, maybe of radius 1.5.
In fact we can ask ourselves the question: Where are the roots located in the until disk?
We know that complex roots occur in conjugate pairs and we have 3 roots in the unit
disk. So we are either going to have two complex roots which are conjugates and one real
root, or 3 real roots.
Example 73
Show that

xa
1 + x dx = sin a , for 0 < a < 1
0

We are going to work with

z a
1+z
And the following contour: The keyhole contour, which avoids the positive real axis and
the origin. In this region, we can take the cut for the logarithm to be on the positive real
axis, i.e. we take 0 < arg(z) < 2. Then the log is holomorphic inside the contour we are
working with. Now, we have that
f (z) =

elog z = z
(elog z )a = z a = ea log z
Our contour consists of the following:
L1 is the segment from r + i to R + i, i.e. the segment [r + i, R + i]. We parametrise
it by z(x) = x for r x R
L2 is the inverse of L1 , i.e. the segment [r i, R i].
R is the bigger circle (the full circle), where we parametrise by z(t) = Reit
r is the smaller circle (the full circle) where we parametrise by z(t) = reit
Now, computing the integral with the Residue Theorem, we get that
z a
) = lim ea log z
z1
z1
z+1
a log (1)
a log 1+i
=e
=e
= eai

res(f, 1) = lim ((z + 1)

155

and so
ai
f (z)dz = 2ie

Now we are going to consider what happens to our integral on the contour we have chosen,
as we let the width of the corridor to go to 0.
On L1 , we have that
R

f (z)dz
r

L1

ea log(x)
dx , as 0
1+x

this is so, because in this region, the x-values are positive, so the logarithm is the usual
real logarithm, so we have that
z a = ea log(z) = ea(log z+iarg(z)) ea log(z) = ea log(x) as 0 when x 0
However, on L2 , x is still positive so we dont get any problems there, but we are approaching the positive real axis from below, so the argument of the logarithm doesnt
tend to 0 but to 2. So we have that
z a = ea log(z) = ea(log z+iarg(z)) ea(log(z)+i2) = ea log(x) eai2 as 0
Thus, we get from these two integrals that
R

f (z)dz + dz = (1 e

2ia

L1

)
r

L2

a log x
1+x

Consider now the curve R . We have that

RRR
RRR
RRR
R
z a
RRR f (z)dz RRRRR 2R max

zR 1 + z
RRR
RRR
RR
R
z a
2R max
zR R 1
since the triangle inequality gives us that 1 + z z 1 = R 1 on z = R.
We also have that
z a = ea log(z) = eR(a log z) = ea log z
= za = Ra on z = R
as log(z) = log z + iagr(z) and arg(z) is the imaginary part, so R(a log(z)) = a log z
Thus we have that
RRR
RR
RRRR f (z)dz RRRRR 2R Ra = 2i R1a 0 as R , since 0 < a < 1
RRRR
RRRR
R1
R1
RRR
RR
Finally, consider the integral around e . We have that
RRR
RRR
RRR
R
z a
ra
ra
r1a
RRR f (z)dz RRRRR 2r max
2r
= 2r
= 2r
0 as r 0
z=r 1 + z
1r
1r
1r
RRR
RRR
Rr
R
156

by the same arguments as before, but using the other triangle inequality.
Altogether we have that
R

(1 e

2ia

)
r

xa
dx = 2ieai
1+x

which becomes, as we let R and r 0

(1 e

2ia

)
0

xa
dx = 2ieai
1+x

Therefore,

2ieai ai
2i
2i

xa
= ai ai =
=
1 + x dx = 1 e2ia e
e
e
2i(sin a) sin a
0

Theorem 35 (The Fundamental Theorem of Algebra - Proof with Rouches Theorem)

Every non-constant polynomial of degree n has n roots, counted with their multiplicities.
Proof.
P (z) = an z n + an1 z n1 + . . . + a1 z + a0
Suppose that z = R, where R is large. Let
f (z) = an z n
g(z) = an1 z n1 + . . . + a1 z + z0
Then how many zeroes does f (z) have inside of z = R? We set f (z) = an z n , where an =/ 0,
so the zeroes of f are given by
an z n = 0 z n = 0 z = 0 with multiplicity n
So f has n zeroes. Consider now g. We have that
g(z) = an1 z n1 + . . . + a1 z + a0
an1 z n1 + an2 z n2 + . . . + a1 z + a0 by the triangle inequality
= an1 Rn1 + . . . + a1 R + a0 since here z = R
< an Rn = f (z)
This means that, by Rouches theorem, f + g has the same number of roots as f , which
has n roots (counted with their multiplicities).
Now we just need to prove that the last step above was legitimate, i.e. we need to prove
that
an1 Rn1 + . . . + a1 R + a0 < an
157

Dividing though by Rn , we get that

an1

1
1
+ . . . + a0 n < an
R
R

But

1
1
+ . . . + a0 n ) = 0
R
R
and therefore, by the Inertia Principle, we have that, for R large enough,
lim (an1

an1

1
1
1
+ an2 2 + . . . + a0 n < an
R
R
R

but since we started with a large R, this result definitely holds. This proves the last step
we needed to justify and thus the proof of the theorem is complete.
Theorem 36 (Riemanns Theorem of Removable Singularities)
Let be an open set and let f {z0 } C be holomorphic.
Assume that f is bounded on {z0 }.
Then f has a removable singularity at z0 .
Proof. Since f is holomorphic of the annulus 0 < z z0 < R, we have a Laurent expansion
for f (z). That is, z with 0 < z z0 < R, we have

f (z) = an (z z0 )n
n=

with
an =

1
2i

Cr (z0 )

f (w)
dw
(w z0 )n+1

Our aim will be to show that for n < 0, an = 0, and thus deduce that the Laurent expansion
actually starts from 0 rather than from
We have that f is bounded. Therefore, we can find an M such that f (w) M , w .
Fix n < 0. Then
RRRR
RRRR
f (w)
R
RR 1
R
dwRR
an = RR
RRR 2i (w z0 )n+1 RRRRR
RR
RR Cr (z0 )
1
f (w)

.2r max
zCr (z0 ) w z0 n+1
2
M
M
r. n+1 = n 0 for r 0 since n > 0
r
r
Therefore, an = 0 for n < 0.
So for the Laurent series, we get that

an (z z0 )n = a0 + a1 (z z0 ) + . . .
0

158

and this converges inside z z0 < R. Lets call this power series g(z). Then g is holomorphic and g(z0 ) = a0 ., while g(z) = f (z), z =/ z0 .
So we can define f (z0 ) = g(z0 ) = a0 , and by doing this, we have defined a value that the
function can take at the singularity at z0 . Therefore, its a removable singularity.
This leads to the following
Corollary 12
The function f has an isolated singularity at z0 , which is a pole if and only if
lim f (z) =

zz0

Proof. Part 1
z0 is a pole lim f (z) =
zz0

Assume that z0 is a pole of f . Then for some small punctured disk D (z0 , ), we can write
f (z) = (z z0 )m h(z)
where m is the order of the pole. Here h is holomorphic on the whole disk and it is also
non-vanishing.
f (z) =

1
1
h(z) =
h(z) as z z0
m
(z z0 )
z z0 m

since

1
= +
zz0 z z0 m
lim

and

lim h(z) = h(z0 ) =/ 0 as h(z) =/ 0 on D

zz0

Part 2
lim f (z) = z0 is a pole

zz0

Looking at g =

1
f

D (z0 , ) C.
g(z) =

1
0 as z z0
f (z)

Now, since z0 is an isolated singularity of g(z) and g(z) is bounded on the disk, Riemanns
Theorem says that z0 is a removable singularity. So
lim g(z) = 0 we must define g(z0 ) = 0

zz0

Then z0 is a zero of f1 , and so, by the definition of a pole, f has a pole at z0 .

This leads to the following
159

Classification of Isolated Singularities

removable singularities
poles
essential singularities (everything else)
Example 74
1
The function f (z) = e z has an essential singularity at 0.
Suppose that z is real and that we are approaching 0 from the right (i.e. from the positive
real axis, going to the left). Then
lim+

z0

1
1
= lim+ e z =
z0
z

Now suppose that z is still real, but we take the limit from the other side of the real axis,
i.e. we are approaching 0 from the left. We have that
lim

z0

1
1
= lim e z = 0
z0
z

We can keep going. For example, we could take z to be purely imaginary, i.e. z = ix, and
take x to be positive at approach 0 from the positive imaginary axis, i.e. from above. We
get that
1
1
i
e z = e ix = e x = 1 = const
so the limit is also 1, which is a constant. Thus we see that we get different limits as we
approach the singularity from different directions. This is weird!

6.22

(T) Casorati-Weierstrass

Theorem 37 (Casorati-Weierstrass)
Suppose z0 is an essential singularity of f (z). Let r be small enough for f to be holomorphic on D(z0 , r). Given w C and > 0, z D (z0 , r) s.t. f (z) w < .
Then we say that f (D (z0 , r)) is dense in C
Proof. (by contradiction)
We are going to prove the negated statement, by contradiction.
[ > 0, z D (z0 , r) s.t. f (z) w < ] = 0 s.t. z D (z0 , r), f (z) w 0
1
.
Consider the function g D (z0 , r) C s.t. g(z) = f (z)w
The denominator is never 0 since f is holomorphic on D (z0 , r) and f (z) =/ w because
f (z) w 0 , we can deduce that g is holomorphic on D (the denominator is not 0).
1
Look at g(z) = f (z)w
10 , z D .
Therefore, since g is bounded on the punctured disk, by Riemanns Theorem, g has a
removable singularity at z0 (we can define g(z0 ) and g is continuous and holmorphic at

160

z0 ).
g(z) =

1
1
1
f (z) w =
f (z) = w +
f (z) w
g(z)
g(z)

Case 1: g(z0 ) = 0
lim f (z) = lim w +

zz0

zz0

1
1
1
= lim (w) + lim
= lim (w) +
=
zz0 g(z)
zz0
g(z) zz0
g(z0 )

since lim (w) is a fixed number, while

zz0

1
g(z0 )

= .

Therefore, z0 is a pole of f (z).

Case 2: g(z0 ) =/ 0.
g is holomorphic on the whole disk.
We define f (z0 ) = w + g(z10 ) . The f becomes holomorphic at z0 , i.e. z0 is removable.
So in one case, we get an isolated singularity, while in the other case we get a pole. This
is a contradiction.
Note: We cant plot functions close to an essential singularity. It goes everywhere.

6.23

The Mean Value Property and the Maximum Modulus Principle

Discussion
Let f be holomorphic and non-constant on D(z0 , r) and r < r0 . Let C be the circle
z z0 < r and assume Cauchys Integral Formula
f (z0 ) =

f (z)
1
dz

2i
z z0
C

z0

r0
r

We are observing the integral only on the points on the circle. Summing up the values
on the circle is the same as integrating along the circle, and since we are diving by 2
(which is the length of the circle), we get an average value of the function on the circle.
Moreover, the average turns out to be the same as the value of the function at z0 . This
161

is always the case.

If we know the value of f (z0 ), we know the average of the function and vice versa.
This is called The Mean Value Property
Now consider the following discussion which will form part of the proof for the Maximum
Modulus Principle.
Parametrise with z(t) = z0 + reit where 0 t 2. We get that
2

f (z0 + reit ) it
1
1
ir dt =
f (z0 + reit )dt
f (z0 ) =

it
2i
re
2
0

RRR
RRR
2
R
RRR 1
it
f (z0 ) = RR f (z0 + re )dtRRRR
RRR
RRR 2
0
R
R
2

1
f (z0 + reit ) dt

2
0

This is true for any radius r [0, r0 ].

Assume that f (z) f (z0 ), z with z z0 = r. Then we have that
2

1
f (z0 + reit )dt
f (z0 )
2
0
2

1
f (z0 )dz = f (z0 )
2
0

1
f (z0 ) =
f (z0 + reit )dt

2
0

0 =

1
f (z0 + reit )dt f (z0 )
2
0

2
it

0 = f (z0 + re )dt f (z0 )dt

0
2

0 = (f (z0 + reit ) f (z0 )) dt

0

So we have an integral whose value is 0 and we are also assuming that f (z) f (z0 )
f (z0 + reit ) f (z0 ) 0. (it is also continuous). So this actually implies that t [0, 2],
we have that
f (z) = f (z0 ), z with z z0 = r, r [0, r0 ]
i.e.
f (z) = f (z0 ), z D(z0 , r0 )
162

But this is a contradiction since we are assuming that the function is not constant. The
contradiction comes from the assumption that f (z) f (z0 ). So we conclude that if f
is holomorphic and non-constant in the neighbourhood D(z0 , r0 ) at z0 , then there exists
at least one point z in it such that f (z) > f (z0 )
Theorem 38 (The Maximum Modulus Principle)
If f is holomorphic and non-constant in a region , then f (z) has no maximum value
inside .
Proof. We prove the theorem by contradiction.
If we assume that z0 s.t. f (z) f (z0 ), z D(z0 , r) , then we get that f
is constant on D(z0 , r) by the previous discussion. Then by The Principle of Analytic
Continuation, we can conclude that f is constant on the whole of . This contradicts the
assumption of the theorem. Thus, we reject this possibility.
Assume that = D(z0 , r0 ). Assume that f is holomorphic on D(z0 , r) and continuous on
D(z0 , r0 ) = {z z z0 R0 }. Since f (z) is continuous on D(z0 , r0 ), it is going to achieve
a maximum there.
Set M = maxzz0 r0 f (z), i.e. w0 D(z0 , r0 ) s.t. M = f (w0 ).
By the max modulus principle, w0 / D(z0 , r0 ) but w z0 = r0 , i.e. w0 is on the circle.
w0
z0

163

r0