Complex Analysis

Complex Analysis II (Epiphany)
Katie Gittins & Stephen Harrap1
(Epiphany terms starts at section 6.3)
2022-2023
Contents
1 The Complex Plane and Riemann Sphere 1

1.1 Complex numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Exponential and trigonometric functions . . . . . . . . . . . . . . 5
1.3 Logarithms and complex powers . . . . . . . . . . . . . . . . . . . 6
1.4 The Riemann Sphere and extended complex plane . . . . . . . . . 11
2 Metric Spaces 14
2.1 Metric spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Open and closed sets . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Convergence and continuity . . . . . . . . . . . . . . . . . . . . . 22
2.4 Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3 Complex Differentiation 33
3.1 Complex differentiability . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Cauchy-Riemann equations . . . . . . . . . . . . . . . . . . . . . . 34
3.3 Connected sets and zero derivatives . . . . . . . . . . . . . . . . . 38
3.4 The angle-preserving properties of holomorphic functions . . . . . 40
3.5 Biholomorphic maps . . . . . . . . . . . . . . . . . . . . . . . . . 43
4 Möbius transformations 46
4.1 Definition and first properties of Möbius transformations . . . . . 46
4.2 Fixed points, the cross-ratio, and the three points Theorem . . . . 49
4.3 Circles and lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.4 The Riemann sphere, revisited . . . . . . . . . . . . . . . . . . . . 55
4.5 Möbius transformations preserving the upper half plane or unit disc 56
4.6 Finding biholomorphic maps between domains . . . . . . . . . . . 60
5 Notions of Convergence and Power Series 64

5.1 Pointwise and uniform convergence . . . . . . . . . . . . . . . . . 64
5.2 Locally uniform convergence . . . . . . . . . . . . . . . . . . . . . 69
2
CONTENTS 3
5.3 Complex power series . . . . . . . . . . . . . . . . . . . . . . . . . 71
6 Complex integration over contours 77

6.1 Definition of contour integrals . . . . . . . . . . . . . . . . . . . . 77
6.2 The Fundamental Theorem of Calculus . . . . . . . . . . . . . . . 83
6.3 First version of Cauchy’s Theorem . . . . . . . . . . . . . . . . . . 89
6.4 Cauchy’s Integral Formula . . . . . . . . . . . . . . . . . . . . . . 94
7 Features of Holomorphic Functions 96

7.1 The Cauchy-Taylor Theorem and CIF for derivatives . . . . . . . 96
7.2 Liouville’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 100
7.3 The Maximum Modulus Principle . . . . . . . . . . . . . . . . . . 102
7.4 Analytic continuation and the Identity Theorem . . . . . . . . . . 104
7.5 Harmonic functions and the Dirichlet problem . . . . . . . . . . . 107
8 General form of Cauchy’s Theorem and C.I.F. 112

8.1 Winding number and simply connected sets . . . . . . . . . . . . 112
8.2 General Forms of Cauchy’s Theorem and Integral Formula . . . . 116
9 Holomorphic functions on punctured domains 118

9.1 Laurent series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
9.2 Classification of isolated singularities . . . . . . . . . . . . . . . . 123
9.3 Removable singularities . . . . . . . . . . . . . . . . . . . . . . . . 124
9.4 Poles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
9.5 Essential singularities . . . . . . . . . . . . . . . . . . . . . . . . . 128
10 Cauchy’s Residue Theorem 129

10.1 The Residue Theorem . . . . . . . . . . . . . . . . . . . . . . . . 129
10.2 Calculation of real integrals . . . . . . . . . . . . . . . . . . . . . 133
10.2.1 Part 1: Rational functions . . . . . . . . . . . . . . . . . . 133
10.2.2 Part 2: Principal value integrals. . . . . . . . . . . . . . . 137
10.2.3 Part 3: Indented contours. . . . . . . . . . . . . . . . . . . 141
10.3 The Argument Principle and Rouché’s Theorem . . . . . . . . . . 146
1
These notes are essentially the notes of Sabine Bögli & Stephen Harrap from 2021–2022,
which in turn were essentially those of Michael Magee from 2018-2019, in which it says: Chapters
1-5 are an evolution of notes of Stephen Harrap, which were in turn based on original notes of
Jens Funke. Chapters 6 onwards have gone though a similar evolution through Jens, Thanasis
Bouganis, Michael & Stephen.
Chapter 1
The Complex Plane and Riemann

Sphere
1.1 Complex numbers

A complex number z is a quantity of the form z = x + iy, where x, y are real numbers
and i is the imaginary unit. We denote by C the set of all complex numbers.
Algebra
We can add, subtract and multiply complex numbers: If z1 = x1 + iy1 and z2 = x2 + iy2
then
z1 ± z2 := (x1 ± x2 ) + i(y1 ± y2 ),
z1 z2 := (x1 x2 − y1 y2 ) + i(x1 y2 + x2 y1 ).
Notice that addition simply corresponds to adding the individual components. In gen-
eral we denote by Re(z) = x the real part of z, and by Im(z) = y the imaginary part
of z. By the definition of multiplication we have i2 = −1, and using this we see that
multiplication corresponds to ‘multiplying out the brackets’: (x1 + iy1 )(x2 + iy2 ) =
(x1 x2 + i2 y1 y2 ) + i(x1 y2 + x2 y1 ).
We can also divide complex numbers. For z2 6= 0 (here we use the shorthand
0 = 0 + 0i) we have
z1 x1 + iy1 (x1 + iy1 )(x2 − iy2 ) x1 x2 + y1 y2 x2 y1 − x1 y2

= = = +i ∈ C.
z2 x2 + iy2 (x2 + iy2 )(x2 − iy2 ) x22 + y22 x22 + y22
The quantity we used to make the denominator real is important. In general, for
z = x + iy we call z̄ := x − iy the complex conjugate of z.
1
CHAPTER 1. THE COMPLEX PLANE AND RIEMANN SPHERE 2
We immediately have a multiplicative inverse

1 x y
z −1 := = 2 2
−i 2 .
z x +y x + y2
WARNING: While most of the nice properties of R hold in C, we do not have notions
of ≤ <, ≥ or >; the set C is not ‘ordered’ and expressions like z1 < z2 have no meaning.
How do we visualise the complex numbers?

There are various ‘models’ for the complex numbers. The most commonly used/most
intuitive is to think of C as a copy of R2 equipped with a map
R2 × R2 → R2 : ((x1 , y1 ), (x2 , y2 )) 7→ (x1 x2 − y1 y2 , x1 y2 + x2 y1 ).
So, a copy of R2 with a way of multiplying1 (and dividing) vectors! Indeed there is an
obvious bijection f : R2 → C given by f ((x, y)) = x + iy. As a result, we often draw
complex numbers on the usual (x, y)-plane: such a picture is called an Argand diagram
(see Figure 1.1).
Figure 1.1: Argand diagram.
OnpR2 there is a natural notion of size, and we use it in C: we call the quantity
|z| := x2 + y 2 the modulus or absolute value of z (= x + iy).
Lemma 1.1 (Important Properties of Complex numbers).
1. z1 z2 = 0 ⇐⇒ z1 = 0 or z2 = 0,
√
2. |z| = z z̄,
z+z̄ z−z̄
3. Re(z) = 2 and Im(z) = 2i ,
1
The multiplication of complex numbers is commutative and associative. The fact that such a
multiplication exists in 2 dimensions is truly remarkable: there is no such multiplication on R3 , not
even if we relax the condition that it is commutative! The search for an associative multiplication on
R3 lead Hamilton to discover (a non-commutative) one on R4 instead which was his famous discovery
of the quaternions.
4. z −1 = z̄
|z|2
.
Remark 1.2. Property 1. is very important, and makes C an integral domain (see Algebra
II).
Polar form/coordinates of a complex number

Now we have a notion of distance, as in R2 we can implement a change of variables
z(x, y) → z(r, θ). Let r = |z| and let θ denote the anticlockwise angle measured from
the real axis. (Angles measured clockwise will be considered negative.) We call θ the
argument of z (for z 6= 0) and write arg(z) = θ. We then have the following polar
coordinates for z:
z = r(cos θ + i sin θ),
which we write in shorthand as z = reiθ (see Figure 1.2).
Figure 1.2: Polar coordinates.

√ √ √ √ √
For example, i = eiπ/2 and 1 + i = 2eiπ/4 . We have | − 1/ 2 − i 3/ 2| = 2, so
√ √ √ √ √ √
−1/ 2 − i 3/ 2 = 2(−1/2 − i 3/2) = 2e−i2π/3 .
Note that arg(z) is only defined up to multiples of 2π; for example i = eiπ/2 = ei5π/2 =
e−i3π/2 . Strictly speaking arg(i) = π/2+2πk, for any k ∈ Z (and so arg is a one-to-many
function!). As a result, we need to be careful; we choose a fixed interval in which to
express the argument: the principal value of arg(z) is the value in the interval (−π, π]
and will be denoted Arg(z). So Arg(i) = π/2 and Arg(−1) = π for example.
Lemma 1.3 (Properties of argument). We have the following properties of the argu-
ment:
1. arg(z1 z2 ) = arg(z1 ) + arg(z2 ) mod 2π
2. arg(1/z) = − arg(z) mod 2π
3. arg(z̄) = − arg(z) mod 2π .

When we say two real numbers are equal mod2π we mean they differ by an integer
multiple of 2π.
It is nice to have a geometric picture of what the algebraic operations on complex
numbers mean.
Lemma 1.4. Geometrically, multiplication in C is given by a dilated rotation; i.e., if
z1 = r1 eiθ1 and z2 = r2 eiθ2 then
z1 z2 = r1 r2 (cos(θ1 + θ2 ) + i sin(θ1 + θ2 )) = r1 r2 ei(θ1 +θ2 ) .
In particular, multiplying by z2 constitutes an (anticlockwise) rotation of z1 by θ2 de-

grees, followed by a dilation with factor r2 . Furthermore, addition represents a transla-
tion and conjugation represents a reflection in the real axis. Taking the real or imaginary
part of a complex number z represents a projection of z onto the real or imaginary axis
respectively.
Proof. By the standard double angle formula we have
z1 z2 = r1 r2 (cos θ1 + i sin θ1 ) (cos θ2 + i sin θ2 )

= r1 r2 ((cos θ1 cos θ2 − sin θ1 sin θ2 ) + i(sin θ1 cos θ2 + sin θ2 cos θ1 ))
= r1 r2 (cos(θ1 + θ2 ) + i sin(θ1 + θ2 )) .
The geometric interpretations of addition, conjugation, and real/imaginary parts are

fairly obvious (see Figure 1.3).
Figure 1.3: Geometric interpretation of conjugation and taking real and imaginary
parts.
Corollary 1.5. 1. |z1 z2 | = |z1 | |z2 |,
2. De Moivre’s formula: (cos θ + i sin θ)n = cos(nθ) + i sin(nθ).

The modulus also has the following important properties.
1. (Triangle inequality) |z1 + z2 | ≤ |z1 | + |z2 |
2. |z| ≥ 0 and |z| = 0 ⇐⇒ z = 0
3. max(|Re(z)|, |Im(z)|) ≤ |z| ≤ |Re(z)| + |Im(z)|
Remark 1.6. The first two properties above along with Corollary 1.5 make the modulus
a norm on C (see later, Definition 2.2).
Simple complex functions and geometry (examples)

• We can also use functions to define regions in the complex plane. Consider the
set of points z which satisfy the inequality |z − i| < |z + i|. This is precisely the
points in C whose distance to i is strictly smaller than their distance to −i. Thus,
the inequality represents the upper half plane H := {z ∈ C : Im(z) > 0}.
• Note that the equation |z − i| = 1 represents a circle centred at i of radius 1.
1.2 Exponential and trigonometric functions

Definition 1.7 (Complex exponential). We define the complex exponential func-
tion exp : C → C by
exp(z) := ex (cos y + i sin y). (z = x + iy)
As shorthand we write exp(z) = ez .

zn
Remark 1.8. We will see later that ez = ∞
P
n=0 n! as in real analysis. We could have
started with this as the definition instead.
Proposition 1.9. We have the following properties of the complex exponential function:
1. ez 6= 0 for all z ∈ C,
2. ez1 +z2 = ez1 ez2 ,
3. ez = 1 ⇐⇒ z = 2πik for some k ∈ Z,
4. e−z = 1/ez ,
5. |ez | = eRe(z) .
Proof. Most are straightforward to check. 3. is very important: exp(z) = 1 ⇐⇒

ex cos y = 1 and ex sin y = 0. Since ex > 0, the latter is equivalent to sin y = 0 and so
y = nπ for some n ∈ Z. Thus we have exp(z) = 1 ⇐⇒ ex cos(nπ) = 1 ⇐⇒ ex (−1)n =
1 ⇐⇒ n is even and ex = 1 ⇐⇒ x = 0 and y = 2kπ (k ∈ Z).
Corollary 1.10. We have exp(2πi) = 1 and exp(πi) = −1. The latter is Euler’s
formula.
Corollary 1.11. The complex exponential function is 2πi-periodic; that is, exp(z +
2kπi) = exp(z) for any k ∈ Z.
Remark 1.12. The above implies exp is determined entirely by the values it takes in any
horizontal strip of width 2π in the complex plane. Note also that exp is ‘unbounded’,
since by 5. the modulus | exp(z)| gets arbitrarily large as Re(z) increases.
Definition 1.13 (Trigonometric functions). All as functions from C → C, we define

1 iz 1
sin(z) := (e − e−iz ) cos(z) := (eiz + e−iz )
2i 2
1 1
sinh(z) := (ez − e−z ) cosh(z) := (ez + e−z )
2 2
(For z = x real these coincide with the real functions. We will eventually get power
series expressions for them from the one for the exponential function.)
Remark 1.14. All the usual double angle formulae and equations relating the functions
hold: e.g., cos2 (z)+sin2 (z) = 1. Additionally, notice that we have cosh(iz) = cos(z) and
cos(iz) = cosh(z), while sinh(iz) = i sin(z) and sin(iz) = i sinh(z). All four functions
are unbounded.
1.3 Logarithms and complex powers

We use the notation C∗ = C − {0}, i.e. the set of nonzero complex numbers.
Lemma 1.15. [Inverting the exponential function] For every w ∈ C∗ , the equation
ez = w (1.1)
has a solution z. Furthermore, if we write w = |w|eiφ with φ = Arg(w), then all

solutions to (1.1) are given by
z = log |w| + i(φ + 2kπ) for k ∈ Z. (1.2)
Here, log |w| is the usual natural logarithm of the real number |w|. Note that there are
infinitely many solutions.
Proof. If z is of the form as in (1.2) for some given k ∈ Z, then
ez = elog |w|+i(φ+2kπ) = elog |w| ei(φ+2kπ) = elog |w| ei(φ) = |w|ei(φ) = w,

by Proposition 1.9, Part 2. and Corollary 1.10. Thus, z is a solution.

To see all solutions are of the given form, first write z = x + iy and assume ez = w.
Since ex eiy = ez = w = |w|eiφ , we have |ez | = ex = |w|. Thus x = log |w|. Moreover,
dividing both sides by |w| we have eiy = eiφ and so ei(y−φ) = 1. It follows from
Proposition 1.9, Part 3. that all solutions are given by i(y − φ) = 2kπi for some k ∈ Z;
in other words, y = φ + 2kπ.
We now come to an important topic called branch cuts.

First we give a high-level description of why branch cuts are necessary. We have just
described for fixed w0 6= 0, exactly what the possible numbers are such that ez = w0 .
These values of z could all reasonably be called log(w0 ). Can we make this choice of
solution to ez = w0 vary nicely as we move w0 a little bit? 2 . Of course, we will run
into trouble at 0, since ez = 0 has no solutions. Bearing this in mind, could we at
least define a ‘log’ function that is ‘continuous’ on C \ {0}? Since we don’t know the
definition of continuous yet, let us just ask that the function should not jump abruptly
when we move from a point to a nearby one.
The answer is no, and let’s see why by trying to come up with one.
Let’s suppose w0 = 1 and we pick a solution to ez = w0 . The obvious one is z = 0
so let’s pick that. Now let w be close to 1. If we’ve found a solution to ez = w then we
know from Lemma 1.15 that it must be of the form
z = log |w| + i(Arg(w) + 2πk)

for some k ∈ Z. Since w is close to 1, we know log |w| is close to 0 and Arg(w) is close
to 0. Now, if k is not zero, then z will not be close to 0 (any integer that is not zero has
absolute value at least one!). So k must be zero for z to be close to 0. We just argued
that if we move w a little, and if we want log to be ‘continuous’, we must choose the
argument of w ‘continuously’.
Now let’s take our idea a little further. If we move w on a path beginning at 1,
following the unit circle anticlockwise, until we reach 1 again, what happens to our
solutions to ez = w if we are choosing them continuously as above? In other words,
what happens to the argument of w? Since we are moving anticlockwise and we are
choosing the argument continuously, it is increasing as we go around the unit circle.
So just before we complete the circle, the argument of w is just below 2π. This is a
big problem, since it means there are values of w just below 1 on the unit circle where
we have been forced to set log(w) very close to 2πi. On the other hand, we began by
assuming log(1) = 0. So the way we have tried to do things, our log function is going
to have a jump discontinuity below 1.
No matter what way we try to define log continuously on C \ {0}, we will run into
a similar problem. On the other hand, if Rθ is any ray of the form
2
To be precise, we would like the solution to ez = w to vary as a continuous function of w; see later
for the precise definition of continuous function.
Rθ = {reiθ : r ∈ R, r ≥ 0} ⊂ C
then it is possible to define a continuous function log(z) on C \ Rθ . (One can think of
cutting out this ray as cutting out the points where log will have a jump discontinuity).
Definition 1.16 (Complex logarithm functions). For any two real numbers θ1 < θ2
with θ2 − θ1 = 2π, let arg be the choice of argument function with values in (θ1 , θ2 ].
Then the function
log(z) := log |z| + i arg(z)

is called a branch of logarithm. It has a jump discontinuity along the ray Rθ1 = Rθ2 .
This ray is called a branch cut.
If we choose arg(z) = Arg(z) ∈ (−π, π], then we obtain a branch of logarithm called
the principal branch of log . We write Log for this principal branch: it is given by
the formula
Log(z) := log |z| + iArg(z).
The principal branch of logarithm has a ‘jump discontinuity’ along the ray given by the
non-positive real axis R≤0 .
Remark 1.17. • Any time one talks about a function called log, one has to declare
which branch of log we use. This is normally done simply by stating the interval
(θ1 , θ2 ] where arg(z) lives.
• As soon as we define continuous functions, we will easily be able to see that the
branch of log corresponding to arg(z) ∈ (θ1 , θ2 ] is continuous on C \ Rθ1 .
• The principal branch, Log, agrees with the natural logarithm log on the real line;
that is, for x > 0 we have Logx = log x. For this reason we will always use the
principal branch unless otherwise stated.
Lemma 1.18. [Properties of logarithms] We have the following properties when using
any given branch of logarithm:
1. elog z = z for any z ∈ C \ {0}, but,
2. in general, log(zw) 6= log z + log w, and
3. in general, log(ez ) 6= z.
We would now like to define functions giving powers of complex numbers. We already
know from the world of real numbers that to define a function giving for example, a
square root of a positive real number, we have to make a choice of whether to take
the positive or negative root. To take a root of a complex number, we have to make a
similar choice, but we have more freedom.
Definition 1.19 (Complex powers). For w ∈ C fixed, by choosing any branch of log
we can define a branch of the function z 7→ z w by the expression
z w := exp(w log z).
For example, if w = 1/n and we use the principal branch we get
z 1/n = e(log |z|+iArg(z))/n = |z|1/n eiArg(z)/n .
Warning: different branches of log can give different power functions! So we must
always specify which branch of log we are using.
Remark 1.20. Now that we have defined complex powers, we should check that our
exponential function matches up with the concept of ‘raising e to the power z’ for a
suitable choice of log e. The natural choice of log e is 1. Then, e raised to the power z
should agree with computing exp(z log e) = exp(z) as we expected.
Finding values of log and complex powers (examples)

(a) Using√the principal branch of log, we find log(1 i) and (1 − i)1/2 . We have
√ −−iπ/4
|1 − i| = 2 and Arg(1 − i) =√−π/4. Thus, 1 − i = 2e . Therefore, Log(1 − i) =
log |1 − i| + iArg(1 − i) = log 2 − iπ/4, and
√ √ π √

1 1 π π
1/2
= 2e−i 8 .
4 4
(1−i) = exp Log(1 − i) = exp log 2 − i = exp log 2 − i
2 2 4 8
(b) Using the principal branch and the previous example
π √ √
(1 − i)i = exp(iLog(1 − i)) = exp + i log 2 = eπ/4 ei log 2 .
4
(c) Again, using the principal branch
√ √

1/2 1
2 = exp log 2 = exp(log 2) = 2.
2
What about the other root? It comes from using a different branch: if we let log be the
branch of logarithm corresponding to arg(z) ∈ (π, 3π] then log(z) = log |z| + i(Arg(z) +
2π), so we have log(2) = log 2 + i2π and
√ √ √

1/2 1
2 = exp (log 2 + i2π) = exp(log 2 + iπ) = 2eiπ = − 2.
2
Remark 1.21. All nth roots can be found this way (see Sheet 1 Q23). In particular, for
z ∈ C∗ all nth roots of z are of the form

1/n 1/n Arg(z) 2kπi
z = |z| exp i + for k = 0, . . . , n − 1.
n n
Visualizing complex functions

The ‘graph’ of a complex-valued function f : C → C is 4-dimensional, so difficult to
visualise - we certainly can’t draw it. We can employ other techniques to get a grasp
on complex functions:
• We can graph the real-valued function |f | : C → R. For example, consider the
complex function cos z. When z = x is purely real, we have that |f (z)| = | cos(x)|
is obviously just the modulus of the real cosine function. But for z = iy purely
complex we have |f (z)| = | cos(iy)| = | cosh y| = cosh y. So in the imaginary
direction f simply looks like cosh!
• It is often useful to visualise complex functions by considering how they map
regions of the complex plane. Consider the image of the ‘right half-plane’ HR :=
{z ∈ C : Re(z) > 0} under the map f (z) = z 2 . Note that HR = {z ∈ C : −π/2 <
Arg(z) < π/2}. If z = reiθ ∈ HR then z 2 = zz = r2 ei2θ has argument 2θ ∈
(−π, π). Thus, f maps HR to C \ R≤0 , where R≤0 denotes the negative real axis
(including 0), see Figure 1.4. The map is onto, since for every w = seiφ ∈ C \ R≤0
(so φ ∈ (−π, π)) we can find z ∈ HR such that f (z) = w; namely we can choose
√
z = seiφ/2 ∈ HR .
Figure 1.4: Geometric picture of how f (z) = z 2 transforms the right half-plane.
Similarly, the left half plane HL = {z ∈ C : Re(z) < 0} is mapped to C \ R≤0 .

Moreover, f maps both the strictly positive imaginary axis iR>0 and the strictly
negative imaginary axis iR<0 (given by iR>0 = {iy ∈ C : y > 0} and iR<0 =
{iy ∈ C : y < 0} respectively) to the strictly negative real axis R<0 .
Adding the observation that f (0) = 0, we have that f (z) = z 2 in essence maps
C to two copies of itself (except for the origin, which is only attained once in the
image - remember this different behaviour at the origin later!)
• Branches of log. Let log be the branch of logarithm corresponding to arg(z) ∈
(θ1 , θ2 ]. Then log maps C \ Rθ1 to the infinite horizontal strip
{z ∈ C : θ1 < Im(z) ≤ θ2 }.
Infinite rays emanating from 0 map to horizontal lines, and circles centred at zero,
minus their interection with Rθ1 , map to vertical line segments (see Figure 1.5).
Figure 1.5: Geometric picture of how f (z) = log(z) transforms rays and circles.
1.4 The Riemann Sphere and extended complex plane

It is very useful at various points of the course to extend the complex plane by adding a
point ‘at infinity’. To do this, we create a new object called ‘infinity’, denoted ∞, and
consider the set
Ĉ := C ∪ {∞}.
At the moment, we have accomplished nothing really. What will be useful later is that
we can think of the point ∞ as being glued ‘nicely’ onto C. The correct way to do this
is by introducing the Riemann sphere.
The Riemann sphere

Consider the unit sphere S 2 := {(x, y, s) ∈ R3 : x2 + y 2 + s2 = 1} in R3 and consider
a copy of C embedded in R3 by identifying C = R2 with the (x, y)-plane. Explicitly, a
point x + iy ∈ C corresponds to the point (x, y, 0) ∈ R3 .
Let N = (0, 0, 1) ∈ S 2 denote the ‘north pole’. For any point v ∈ S 2 \{N }, there is a
unique straight line LN,v passing through N and v. Since v 6= N , this line is not parallel
to the (x, y)-plane. Hence it intersects the (x, y)-plane in a unique point (x, y, 0). This
corresponds to the point x + iy ∈ C. We have defined a map P : S 2 \ {N } → C by
P (v) = x + iy in the notation of the preceding discussion. The map P is called the
stereographic projection (from the north pole).
What is the formula for stereographic projection? Let (x, y, s) ∈ S 2 \ {N }. Note
that s 6= 1. The equation of the line passing through the point (x, y, s) ∈ S 2 and the
North Pole N = (0, 0, 1) ∈ S 2 is given by
      
x 0 x
γ(t) = N + y  − N  t = 0 +  y  t, (t ∈ R).
s 1 s−1
1
This clearly intersects the plane when t = 1−s . Thus
x iy
P : (x, y, s) 7→ + .
1−s 1−s
It is possible to find an inverse to P : given any point z ∈ C, draw the straight line
passing through N and z (thinking of C as the (x, y)-plane as before). This straight line
passes through S 2 in exactly one point. Hence P is a bijection that identifies S 2 \ {N }
with C.
Now we have identified C with S 2 \ {N } via the map P , it gives us a natural way
to view the added point ∞ of Ĉ. It should correspond to adding back in the north pole
to S 2 \ {N }. In other words, we should think of Ĉ simply as the entire sphere S 2 !
In fact, one can show that we have the following correspondences/mappings:
In S 2 In Ĉ (1.3)
N ←→ ∞
S ←→ 0
Equator ←→ Unit circle {z ∈ C : |z| = 1}
(open) Southern hemisphere ←→ D := {z ∈ C : |z| < 1}
(open) Northern hemisphere ←→ Ĉ \ B 1 (0) = Ĉ \ {z ∈ C : |z| ≤ 1}
Algebraically:
x + iy
(x, y, s) ←→ (Stereo. Proj.)
1−s
1
2Re(z), 2Im(z), |z|2 − 1

←→ z (Inverse Stereo. Proj.)
|z|2+1
Remark 1.22. Note that we could have used the south pole S = (0, 0, −1), rather than N ,
to define the projection. In that case we would have the correspondence (x, y, s) 7→ x+iy
1+s
(and you can check that the map f (z) = 1/z̄ takes x+iy
1−s to x+iy
1+s ).
Definition 1.23. The Riemann sphere is the unit sphere S 2 ⊂ R3 along with the
stereographic projections from the north and south pole.
Remark 1.24. Later in your studies you might learn that the Riemann sphere is a
special example of a Riemann surface. The purpose of considering the two stereographic
projection maps as part of the definition is that any point in S 2 is in the domain of one
of the projections, so informally speaking, the maps allow us to think of a region nearby
to any point in S 2 as a region inside C.
Chapter 2
Metric Spaces
2.1 Metric spaces

Now we have another way of thinking about Ĉ - as a sphere in R3 - it looks like there
might be (at least!) two natural ways to measure the distance between two points z
and w in the extended complex plane (or indeed in C).
• The modulus |z − w| of their difference in C.
• The distance in R3 between their stereographic representatives on the sphere S 2 .
In both cases, the set is the same, but there is a different notion of distance. A metric
space is a set together with a ‘distance’ function that satisfies certain axioms.
Definition 2.1 (Metric spaces). A metric space is a set X together with a function
d : X × X → R≥0 such that for all x, y, z ∈ X
• (D1) Positivity. d(x, y) ≥ 0 and d(x, y) = 0 ⇐⇒ x = y;
• (D2) Symmetry. d(x, y) = d(y, x);
• (D3) Triangle inequality. d(x, y) ≤ d(x, z) + d(z, y).
The function d is called a metric and we will often denote a metric space by (X, d).
Examples of Metrics
1. The metric induced by the modulus function | . | on R or C. We can define a
distance function d on R × R or C × C by the formula d(x, y) = |x − y|. This
metric satisfies (D1)-(D3) by the properties 1. and 2. of the modulus we gave just
after Corollary 1.5.
14
CHAPTER 2. METRIC SPACES 15
2. The Euclidean norm on Rn or Cn For vectors x in Rn (or Cn ), the function

v
u n
uX
d(x, y) := kx − yk2 = t |xi − yi |2
i=1
is a metric. It is easy to check properties (D1) − (D3). Note that this norm comes
from an inner-product. For example, for n = 2, the real Euclidean norm on R2
√
comes from the usual dot product kxk2 = x · x and the complex Euclidean p norm
2
on C comes from the inner product hz, wi = z1 w̄1 + z2 w̄2 ; that is kzk2 = hz, zi.
3. • (a) Metrics induced from inner products in vector spaces

More generally, given any finite dimensional real vector space V with a (pos-
itive definite) inner product h . i, then
p
d(v, w) : = kv − wk = hv − w, v − wi (v, w ∈ V )
is a metric. Properties (D1) and (D2) are obvious, property (D3) follows
from Cauchy-Schwarz: | hv, wi | ≤ kvk · kwk - see sheet 2.
• (b) Metrics induced from norms in vector spaces
Even more generally, so long as a vector space has a ‘nice’ notion of the ‘size’
of each vector, we can define a metric in the obvious way. Such a notion is
in generality referred to as a norm:
Definition 2.2. [Norms and normed vector spaces] Given any real or com-
plex vector space V , a function k . k : V → R≥0 is a norm if it satisfies (for
v, w ∈ V )
– (N1) kvk ≥ 0 and kvk = 0 ⇐⇒ v = 0;
– (N2) kλvk = |λ| · kvk for λ ∈ R or C;
– (N3) kv + wk ≤ kvk + kwk (the triangle inequality).
Note that (N3) implies kv − wk ≥ | kvk − kwk | (the reverse triangle inequal-
ity). A vector space equipped with a norm is called a normed vector space.
The metric given by d(v, w) := kv − wk then always defines a metric (it is
easy to check properties (D1)-(D3)). In particular, since the modulus func-
tion on C is a norm, the metric we get from the modulus function comes
from a norm.
4. `p -norm on Rn or Cn (p ≥ 1)
The above suggests that a vector space could be home to many different norms
(so many different metrics). But, not all norms arise from an inner product as in
(3a); for example, for vectors x in Rn or Cn , the function

v
u n
uX
kxkp := tp
|xi |p
i=1
defines a norm for every p ≥ 1, called the `p -norm. But, for p 6= 2 (the Euclidean
norm) it does not arise from an inner product. When p P= 1, the `1 -norm is simply
given by the sum of the size of the components kxk1 = ni=1 |xi | and is sometimes
referred to as the Taxicab norm.
5. `∞ -norm on Rn (or Cn )
The function
kxk∞ := max |xi |
i=1,...,n
also defines a norm, called the `∞ -norm (or the sup-norm), thus it also defines
a metric. It is in some sense the ‘limiting notion’ of the `p norms.
6. Riemannian (chordal) metric on Ĉ

Let f : Ĉ → S 2 be the (inverse of the) stereographic projection. Then the function
d(z, w) := kf (z) − f (w)k2 (z, w ∈ Ĉ),
where k . k2 is Euclidean norm in R3 (so, the `2 -norm), is a metric on Ĉ. It is
called the Riemannian metric (or chordal metric). Note that with respect to
this metric, the distance from 0 to i is the same as the distance from i to ∞, for
example!
7. Discrete metric
Let X be a non-empty finite set. Then (for x, y ∈ X) the function
(
0 if x = y,
d(x, y) :=
1 if x 6= y,
defines a metric, called the discrete metric. It is easy to check (D1)-(D3). In

this case, (X, d) is called a discrete metric space.
8. Function spaces
There are many of these, such as the space X = C([a, b]) of continuous functions
on an interval [a, b]. The function
kf k := max |f (x)|
x∈[a,b]
defines a norm, and thus a metric (see Analysis III for more examples).
9. Subspace metric
Any non-zero subset Y ⊂ X of a metric space X is itself a metric space with
respect to the same metric (this is easy to check). The metric restricted to the set
Y is then called the subspace metric [There is actually much more than meets
the eye with this metric - see sheets 2 and 3.]
2.2 Open and closed sets

Since we have a general notion of distance in any metric space X, we can define balls
in the space. These will be the key to understanding the topology of the space.
Definition 2.3 (Balls in a metric space). Let (X, d) be a metric space, x ∈ X and let
r > 0 be a real number. Then:
• The open ball Br (x) of radius r centred at x is
Br (x) := {y ∈ X : d(x, y) < r}.
• The closed ball B̄r (x) of radius r centred at x is
B̄r (x) := {y ∈ X : d(x, y) ≤ r}.
Visualizing balls (examples)

1. Let X = C and d(z, w) = |z − w|, then B1 (0) = D = {z : |z| < 1} as before.
More generally Br (z0 ) is the usual ball of radius r around z0 , not including its
boundary circle. B̄r (z0 ) is the ball of radius r around z0 , including its boundary
circle. This is the most important example from the point of view of Complex
Analysis.
2. Let us consider the unit ball B1 (0) in R2 with respect to the `p -norms, for p = 1, 2
and ∞.
For p = 2 the unit ball B1 (0) is the usual Euclidean ball - so the inside of the unit
circle centred at the origin. For p = ∞, the equation max{|x|, |y|} < 1 (for (x, y) ∈
R2 ) clearly defines the interior of a square with vertices (1, 1), (−1, 1), (1, −1) and
(−1, −1).
For the `1 -norm a little care is needed. We are interested in the points (x, y) ∈ R2
with |x| + |y| < 1. In the 1st quadrant this means y < 1 − x, in the 2nd it means
y < 1 + x, in the 3rd we have y > −1 − x and in the 4th its y > x − 1. Thus, the
unit ball is the interior of a diamond with vertices (1, 0), (0, 1), (−1, 0) and (0, −1).
Definition 2.4. [Open/closed sets in a metric space] Let (X, d) be a metric space.
Then:
• A subset U ⊆ X is open (in X) if for every point x ∈ U there exists > 0 such
that B (x) ⊂ U .
• A subset U ⊆ X is closed (in X) if its complement X \ U is open.
Remark 2.5. Sets in a metric space can be open and closed at the same time! For
example, the empty set ∅ and the whole metric space X are both open and closed. Such
sets are referred to as ‘clopen’.
Lemma 2.6. [Open balls are open] In a metric space, the open ball Br (x) is open!
Proof. Let y ∈ Br (x) with d(x, y) = s (and so s < r). We need to show there exists
> 0 such that B (y) ⊆ Br (x). Simply take = r − s > 0. Then for every z ∈ B (y)
we have
(D3)
d(x, z) ≤ d(x, y) + d(y, z) < s + = r.
Thus, z ∈ Br (x) as required.
Remark 2.7. It can also be shown that in a metric space the closed ball B r (x) is closed
(see Sheet 2).
Open sets (examples/warnings)

1. All of the previously encountered subsets H, D, C∗ and C\R≤0 of the complex plane
are open (see sheet 2). The 1st quadrant Ω1 := {z ∈ C : Re(z) > 0, Im(z) > 0} is
open. To see this, for z ∈ Ω1 consider the ball Br (z) where r = min(Re(z), Im(z))/2.
(
0 if x = y.
2. Let X be a discrete metric space, so d(x, y) := Then, for
1 if x 6= y.
x ∈ X and r > 0 we have
(
{x} if r ≤ 1.
Br (x) :=
X if r > 1.
Therefore, (by Lemma 2.6) every singleton {x} is an open set with respect to the
discrete metric. Moreover, the complement X \ {x} is also open, since for any
y ∈ X \ {x} (that is, any y 6= x in X) and any r < 1 the open ball Br (y) = {y}
is contained in X \ {x}. Thus, all balls are clopen with respect to the discrete
metric! In fact, any subset Y ⊆ X of a discrete metric space is clopen!
3. Sets don’t have to be either open or closed. For example, [0, 1) is neither open
nor closed in R (with respect to the standard metric | . |) - simply check the point
x = 0 in [0, 1) and the point x = 1 in the complement (−∞, 0) ∪ [1, ∞). However,
recall that any subset of a metric space is itself a metric space (w.r.t the same
metric) - the subspace metric. Thus the pair ([0, 1), | . |) is a metric space - but
then (by the remark after Definition 2.4) the set [0, 1) is open!
Key: Open and closed sets are really relative notions, depending on the ambient
space (as well as the metric).
Notation: When we say a subset of R or Rn or C are open/closed, we will mean
with respect to the standard norms | . | and k . k2 and | . | respectively. Most sets
we encounter do not simply look like open/closed balls, so it will be useful to have
rules for union and intersection:
Lemma 2.8. [Unions and intersections of open sets] Let (X, d) be a metric space.
Then:
1. Arbitrary unions of open sets are open; that is

[
Ui is open, for any (possibly infinite) collection of open sets Ui .
i∈I
2. Finite intersections of open sets are open; that is

n
\
Ui is open, for any finite collection of open sets Ui .
i=1
S
Proof. 1. Let x ∈ i∈I Ui . Then, by definition, it must be contained in the set Uj
for some j ∈ I. Since Uj is openSthere must exist a ball B (x) centred at x lying
in Uj . But then B (x) ⊆ Uj ⊆ i∈I Ui as required.
2. Let x ∈ ni=1 Ui . By definition x ∈ Ui for every i = 1, . . . , n. But, since they are

T
all open, for every Ui there must exist ri > 0 such that Bri (x) ⊂ Ui . Now simply
take = min (ri ). Then for every i we have B (x) ⊆ Bri (x) and so
i=1,...n
n
\ n
\
B (x) ⊆ Bri (x) ⊆ Ui .
i=1 i=1
Corollary 2.9 (Unions and intersections of closed sets). Let (X, d) be a metric space.
Then:
1. Finite unions of closed sets are closed.
2. Arbitrary intersections of closed sets are closed.

Proof. De Morgan’s laws together with Lemma 2.8.
We have the following remarks about open and closed sets.

• An infinite intersection of open sets is not necessarily open (see Sheet 2). Similarly,
an infinite union of closed sets is not necessarily closed: e.g., the union of closed
intervals in R,
∞
[ 1 1
, 1− = (0, 1) is open in R.
i i
i=1
• The next generalisation of a metric space you will encounter (see Topology III)
is called a Topological space T . There, the only stipulations are the existence of
open sets such that
(i) ∅ and T are open ; (ii) Lemma 2.8 holds.
We have a hierarchy:
Inner-product space =⇒ Normed space =⇒ Metric space =⇒ Topological space.
• Why have we been looking at examples in R2 rather than C? It turns out that
both these spaces are ‘topologically equivalent’, that is, they have the same open
sets - this is obvious since the complex modulus is essentially just the Euclidean
norm on R2 .
As we have seen, some sets are neither open nor closed. It will be useful to ask what
the largest possible open set is inside a given set. Similarly, what is the smallest closed
set containing a given set?
Definition 2.10 (Interior points, closure, boundary, exterior). Let A be a subset of a
metric space (X, d).
• The interior A0 of A is defined by
A0 := {x ∈ A : there exists an open set U ⊆ A such that x ∈ U }.
• The closure Ā of A is the complement of the interior of the complement:
Ā := {x ∈ X : U ∩ A 6= ∅ for every open set U with x ∈ U }.

• The boundary ∂A of A is the closure without the interior:
∂A := Ā \ A0 = X \ (A0 ∪ (X \ A)0 ) .

• The exterior Ae of A is the complement of the closure:
Ae := X \ Ā = X \ (A0 ∪ ∂A) = (X \ A)0 .

• Don’t confuse closure with conjugation! Closure concerns sets in any metric space,
conjugation concerns points in C.
• Clearly the interior and exterior are open and clearly the boundary is closed. The
closure is also closed (see sheet 2) - in fact, it is often defined more simply as
Ā := X \ (X \ A)0 , from which the closedness is obvious - it is the complement of
an open set.
• The boundary matches our naive notion. Broadly speaking, the interior of a set
consists of all the points that are not on its ‘edge’, and to form the closure of
a set you simply add all the missing edge points. Indeed, we have the following
additional properties of a subset A ⊂ X (see sheet 2):
[
(a) A is open ⇐⇒ ∂A ∩ A = ∅ ⇐⇒ A = A0 ; In fact A0 = U;
U ⊆A
U open
\
(b) A is closed ⇐⇒ ∂A ⊆ A ⇐⇒ A = Ā; In fact Ā = F;
A⊆F
F closed
That is, the interior A0 is the largest open set contained in A and the closure Ā
is the smallest closed set containing A. Convince yourself that all the definitions
reflect your intuitive notions, say, for the plane!
• In Rn or Cn for simple sets we only have to replace strict inequality with equality
(or vice versa) to obtain the closure (or interior). For example, for A = {z ∈ C :
1 < |z| ≤ 3}, we have A0 = {z ∈ C : 1 < |z| < 3}, Ā = {z ∈ C : 1 ≤ |z| ≤ 3}, and
∂A = {z ∈ C : |z| = 1} ∪ {z ∈ C : |z| = 3} (see Figure 2.1).
Similarly
{z ∈ C : 1 < Re(z) ≤ 3, |Im(z)| < 1} = {z ∈ C : 1 ≤ Re(z) ≤ 3, |Im(z)| ≤ 1}
and
{z ∈ C : 1 < Re(z) ≤ 3, |Im(z)| < 1}0 = {z ∈ C : 1 < Re(z) < 3, |Im(z)| < 1}.
Figure 2.1: The set A with its interior and closure.
In fact, Br (x) = B̄r (x) for any ball in Rn or Cn .

However, this is not true in every metric space (see sheet 2) - there are metric
spaces for which Br (x) 6= B̄r (x); that is, the smallest closed set containing the
open ball Br (x) is not the closed ball B̄r (x)! [Hint: what if the open ball is already
closed!?]
2.3 Convergence and continuity

With our general notion of distance (a metric) comes a natural notion of convergence.
Definition 2.11. [Limits and convergence in a metric space] We say a sequence {xn }
in a metric space (X, d) converges to x ∈ X if we have
lim d(xn , x) = 0.
n→∞
That is,
for every > 0 there exists N ∈ N such that d(xn , x) < for every n > N.
We write “xn → x as n → ∞”, or “ lim xn = x”.

n→∞
Convergent sequences (example)

As mentioned, the chordal metric on Ĉ is d(z, w) = kf (z) − f (w)k2 , where k . k2 is the
Euclidean norm on R3 and f is the inverse Stereographic projection given by
2Re(z) 2Im(z) |z|2 − 1

f (z) = , , .
|z|2 + 1 |z|2 + 1 |z|2 + 1
Show that with respect to this metric the sequence {ki}k∈N in Ĉ converges to ∞ ∈ Ĉ.
Since |ki| = k and f (∞) = (0, 0, 1) we have
k2 − 1

2k
d(ki, ∞) = kf (ki) − f (∞)k2 = 0, 2 , 2 − (0, 0, 1)
k +1 k +1 2

2k 2
= 0, 2 ,− 2
k +1 k +1 2
s 2 2
2k −2
= + −→ 0 as k → ∞.
k2 + 1 k2 + 1
Thus, the sequence indeed converges to ∞. This is quite an odd notion as we are
used to saying sequences ‘diverge’ if they tend to infinity. The key is that convergence
depends on the metric being used.
Limits in C with the standard metric (Very important!).

The above definition of limit in a metric space also gives us an example of limits in C
with the standard metric. This says that if zn is a sequence of complex numbers, then
limn→∞ zn = z if and only if
‘for all > 0, there exists N > 0 such that for all n ≥ N , |zn − z| < ’.
Note this is the same definition as in Analysis I, but replacing the absolute value on
the real line by the modulus on C. Importantly, by the same proofs as in Analysis I,
limits in the complex plane follow the COLT rules.
Furthermore there is a very important link between convergence in the complex
plane and real convergence (see Analysis I for the proof). Let {zn }n∈N be a sequence of
complex numbers zn = xn + iyn . Then, for any fixed z0 = x0 + iy0 ∈ C we have
lim zn = z0 ⇐⇒ lim xn = x0 and lim yn = y0 .

n→∞ n→∞ n→∞
In other words, the sequence {zn } converges iff the real sequences {Re(zn )} and {Im(zn )}
converge (see Sheet 3).
Let’s return to our general setting of metric spaces and prove some properties of
limits.
Lemma 2.12. [Limits and open sets] Let (X, d) be a metric space. Then:
1. A sequence can have at most one limit.

2. We have
lim xn = x ⇐⇒ ∀ open U with x ∈ U, ∃ N ∈ N such that ∀n > N xn ∈ U.

n→∞
Hence the notion of a limit in a metric space can be stated in terms only of its
open sets.
Proof. 1. Assume lim xn = x and lim xn = y. Then we have for each n, by the
n→∞ n→∞
triangle inequality d(x, y) ≤ d(x, xn ) + d(xn , y), so taking the limit as n → ∞
gives
d(x, y) ≤ lim d(x, xn ) + lim d(xn , y) = 0 + 0 = 0,
n→∞ n→∞
so d(x, y) = 0, hence x = y by property (D1) of metric spaces.
2. (⇒): Assume lim xn = x and that U is open with x ∈ U . By definition there
n→∞
exists r > 0 such that Br (x) ⊆ U and an N ∈ N such that d(xn , x) < r for every
n > N . Thus, xn ∈ Br (x) ⊆ U for every n > N .
(⇐): Let > 0. We wish to find N ∈ N such that d(xn , x) < for n > N . Consider
the ball B (x). It is open and contains x, and so there exists N ∈ N such that
xn ∈ B (x) for every n > N . This is precisely the statement that d(xn , x) < for
n > N.
Remark 2.13. The key to completing proofs of this type is to write down the definitions
in your assumptions and also write precisely what you need to prove. Usually, doing
this leads very quickly to the proof.
We can now define what it means for a function between two metric spaces to be
continuous (this will incorporate many of the functions we have already encountered;
e.g., f : C → C; R → C; or C → R).
Definition 2.14 (Continuity). A map f : (X1 , d1 ) → (X2 , d2 ) between two metric
spaces is called continuous at x0 ∈ X1 if
∀ > 0 ∃δ > 0 such that ∀ x ∈ X1 we have d1 (x, x0 ) < δ ⇒ d2 (f (x), f (x0 )) < .
We say a function f is continuous on X1 if it is continuous at every point x0 ∈ X1 .

[Note that this is exactly the same as in Analysis I.]
Remark 2.15. Equivalently, one could write
∀ > 0 ∃δ > 0 such that x ∈ Bδ (x0 ) ⇒ f (x) ∈ B (f (x0 )),
where the first ball is in X1 and the second ball is in X2 .

Lemma 2.16. [Basic properties of continuous functions]
1. Products, sum, quotients of real/complex valued continuous functions on a metric

space X are continuous. E.g., if f : X → C and g : X → C are continuous, then
f + g and f g and f /g are continuous (where defined).
2. Compositions of continuous functions are continuous. I.e., if f : X1 → X2 and

g : X2 → X3 are continuous maps between metric spaces, then g ◦ f : X1 → X3 is
continuous.
Proof. Almost word-for-word from Analysis I.
Examples of continuous functions on the complex plane (with the stan-

dard metric)
• The identity function is continuous.
• Constant functions are continuous.
• The functions Re, Im : C → R are continuous.
• The complex conjugation z 7→ z̄ is continuous as a map from C → C.
• The modulus function z 7→ |z| is continuous as a map from C → R.
• All of exp, sin, cos, sinh and cosh are continuous on C, as are all polynomials.
• If arg is the choice of argument function with values in (θ1 , θ2 ] then arg is contin-
uous on C\Rθ1 (recall Rθ1 is the ray with angle θ1 )
• If log is a branch of log corresponding to an argument function as above, then log

is continuous on C\Rθ1 .
As with limits, it will be useful to restate continuity in terms of open sets. First, recall
that for any function f : X1 → X2 and any set U ⊆ X2 we define the preimage f −1 (U )
of U under f by f −1 (U ) := {x ∈ X1 : f (x) ∈ U }.
Theorem 2.17. [Continuity via open sets] Let X1 and X2 be metric spaces. Then:
f : X1 → X2 continuous ⇐⇒ f −1 (U ) is open in X1 for every open set U in X2

⇐⇒ f −1 (F ) is closed in X1 for every closed set F in X2 .
Proof. For open sets:

• (⇒): Let U be open in X2 , and pick x ∈ f −1 (U ). Since U is open there exists

> 0 such that B (f (x)) ⊆ U (since f (x) ∈ U ). But f is continuous, so by
definition there exists δ > 0 such that if y ∈ Bδ (x) we have f (y) ∈ B (f (x)).
Thus f (y) ∈ U , and so y ∈ f −1 (U ); since this is true for every y ∈ Bδ (x) we have
shown Bδ (x) ⊆ f −1 (U ) and so the set f −1 (U ) is open.
• (⇐): Let x ∈ X1 and > 0. We need to find a δ > 0 such that y ∈ Bδ (x) ⇒
f (y) ∈ B (f (x)). First notice that (by Lemma 2.6) the ball B (f (x)) is open. By
assumption, the preimage f −1 (B (f (x))) of this ball is also open. The point x
must be in this preimage (because the centre f (x) is certainly in B (f (x))). Then,
since the preimage is open there must exist an open ball Bδ (x) around x contained
in f −1 (B (f (x))). But, this is precisely the statement that y ∈ Bδ (x) ⇒ f (y) ∈
B (f (x)) as required.
For closed sets see sheet 3.
Remark 2.18. • From the proof we see that we can be slightly more precise than
the statement of the theorem. We have, for example
f : X1 → X2 continuous ⇐⇒ f −1 (U ) is open in X1 for every open set U inX2

at x ∈ X1 containing f (x)
• Note that for the (⇐) direction we only actually needed the fact that f −1 (B) was
open for any open ball B in X2 . It turns out that the open balls ‘generate’ all the
open sets (via Lemma 2.8) - see Topology/Analysis III!
• Note that, by the second part of Theorem 2.17, if a function f : X1 → X2 is

continuous then f −1 ({x}) is closed for any x ∈ X2 .
As with limits, it turned out that continuity depends only upon the open sets in the
respective metric spaces. This means we can use the continuity of known functions to
prove the openness of very complicated sets.
Showing sets are open using continuity (examples)

• Show the following set is open:
p
U = {(x, y) ∈ R2 : (x2 + y 2 ) sin3 ( x2 + 7) > 2}.
Well, the function

p
f : R2 → R : (x, y) 7→ (x2 + y 2 ) sin3 ( x2 + 7)
is continuous by Lemma 2.16, because it is the product/composition of real valued

continuous functions. Moreover,
U = {(x, y) ∈ R2 : f ((x, y)) > 2} = f −1 ((2, ∞)).
Since (2, ∞) is open in R (see sheet 2), the set U is the preimage of an open set
under a continuous map and by Theorem 2.17 it is therefore open.
• We can actually do a little more using the following useful properties of the preim-
age (from Analysis I):
Useful properties of preimage
– f −1 (A ∪ B) = f −1 (A) ∪ f −1 (B).
– f −1 (A ∩ B) = f −1 (A) ∩ f −1 (B).
– f −1 (A \ B) = f −1 (A) \ f −1 (B).
Show the following set is open:
U = {(x, y) ∈ R2 : xy > 1, x2 + y 2 > 3}.
Both
f (x, y) = xy and g(x, y) = x2 + y 2
are continuous as functions R2 → R, and U = f −1 ((1, ∞)) ∩ g −1 ((3, ∞)).
Since f and g are continuous and both (1, ∞) and (3, ∞) are open in R, the
preimages f −1 ((1, ∞)) and g −1 ((3, ∞)) are open (by Theorem 2.17). By Lemma
2.8 we have that U is open, since it is the intersection of two open sets.
Showing functions are not continuous using open sets (example)

• We can also use openness to prove a function is not continuous. Indeed, for
f : X1 → X2 , if there exists an open set U in X2 such that f −1 (U ) is not open in
X1 , then f is not continuous. For example, the function f : R2 → R defined by
(
xy
2 2, if (x, y) 6= 0,
f (x, y) = x +y
0 otherwise,
isn’t continuous at (0, 0). Why? Consider the preimage f −1 ((−, )). Claim:
This preimage is not open for sufficiently small.
First note that the preimage in question contains (0, 0) since f ((0, 0)) = 0. To
show the preimage is not open it is enough to show that any open ball in R2
centred at (0, 0) is not contained in f −1 ((−, )): Let < 1/4, say, and for any
δ > 0 consider the ball Bδ ((0, 0)) centred at (0, q 0). The point (δ/2, δ/2) is in
2 2
Bδ ((0, 0)) since k( 2 , 2 ) − (0, 0)k2 = k( 2 , 2 )k2 = δ4 + δ4 = √δ2 < δ. But
δ δ δ δ
δ δ
·

δ δ 2 2 1
f , = δ2 δ2
= > ,
2 2 2
4 + 4
so (δ/2, δ/2) is not in f −1 ((−, )). Thus, for any δ > 0 the ball Bδ ((0, 0)) is not
contained in f −1 ((−, )) and so this preimage is not open.
Since (−, ) is open in R it follows from Theorem 2.17 that f is not continuous.
Why preimages?
Note that the use of preimages in Theorem 2.17, rather than images, is important.
The same result is not true of images. E.g., the function f (z) = |z| is continuous as a
function C → R, but it maps an open set in the complex plane f : D → [0, 1) to an
interval that is neither open nor closed in R.
Note that this f is actually a bijection from R≥0 to R≥0 , and in R≥0 the interval
[0, 1) is open! So, is the problem that we need the function to be bijective? No. For
example, consider the metric spaces X1 = [0, 1) ∪ [2, 3] and X2 = [0, 2] with the usual
(subspace) metric coming from the absolute value on R. Define
(
x, if x ∈ [0, 1).
f : X1 → X2 : x 7→
x − 1, if x ∈ [2, 3].
It is easy to check that f is a bijection and is continuous on its domain: (Continuity

is trivial for x 6= 2. For x = 2, pick > 0, then for any 0 < δ < 1 we have Bδ (2) =
[2, 2 + δ). Note that f (2) = 1 and so B (f (2)) = (1 − , 1 + ). To show f is continuous
we must therefore find a δ so that f (x) ∈ (1 − , 1 + ) if x ∈ [2, 2 + δ). Simply pick any
δ < , for then:
x ∈ [2, 2 + δ) ⇒ f (x) = x − 1 ∈ [1, 1 + δ) ⊂ (1 − δ, 1 + δ) ⊂ (1 − , 1 + ),
as required.) But, the set [2, 3] is open in X1 (see sheet 3) and its image f ([2, 3]) = [1, 2]
is not open in X2 .
Thus, we genuinely do need to use preimages. When can we use images of contin-
uous functions to preserve properties of the sets in question? When can we find the
maximum/minimum value taken by a function on a set? It turns out a key concept is
that of compactness.
2.4 Compactness
Definition 2.19 (Compactness). A non-empty subset K of a metric space X is called
(sequentially) compact if for any sequence {xk }k∈N in K there exists a convergent
subsequence {xnk }k∈N with limit in K.
Note that the initial sequence in the definition does not have to converge. But what
if it does? What is the link between a convergent sequence and its subsequences?
Lemma 2.20. If {xk }k∈N is a convergent sequence in a metric space X, then any
subsequence converges to the same limit.
Proof. Assume xk → x. Let {xnk }k∈N be a subsequence (here n1 < n2 < · · · ). We

know that for every > 0 there exists N ∈ N such that xk ∈ B (x) for every k > N . So
simply note that nk > k, whence xnk ∈ B (x) for k > N . Thus, xnk → x.
It seems quite daunting to have to check every sequence in a set for convergent sub-
sequences, so it will be useful to re-express compactness in terms of our basic building
blocks; that is, open and closed sets. However, exploring openness doesn’t seem par-
ticularly fruitful; for example, consider the set (0, 1) in R; the sequence {1/n} lies in
(0, 1), but its limit is 0 ∈
/ (0, 1). Moreover, any subsequence must have the same limit
0 by Lemma 2.20, and so (0, 1) is not compact. Let’s try closedness.
Proposition 2.21. [Closed sets and limits of sequences] We have
F ⊂ X is closed ⇐⇒ Every sequence in F which converges in X has its limit point in F.

(that is, if xn ∈ F and xn → x for some x ∈ X, then x ∈ F.)
Proof.
• (⇒): Assume F is closed and let {xn }n∈N be a sequence with xn ∈ F that converges
to x ∈ X. We wish to show x ∈ F . For a contradiction, assume x ∈ / F ; that is,
assume x ∈ X \ F . We know X \ F is open, so by definition there must exist an
open ball B (x) centred at x and contained in X \ F . But xn → x, so there exists
N ∈ N such that xn ∈ B (x) for n > N . Thus xn ∈ X \ F for n > N (that is,
xn ∈
/ F for n > N ), which is a contradiction.
• (⇐): We need to show X \ F is open. Let x ∈ X \ F . We wish to show we can

always find a ball B (x) centred at x and contained in X \ F . If for some n ∈ N we
have B1/n (x) ⊆ X \ F we are done. Otherwise, pick an element xn ∈ B1/n (x) ∩ F
for each n. But then xn → x and xn ∈ F , but by assumption x ∈ / F . This
contradicts the premise of the proof.
Corollary 2.22. [Relationship between compactness and closedness]
1. Compact sets are closed.
2. Any closed subset of a compact subset is compact.
Proof. 1. If F is compact and {xk }k∈N is a convergent sequence with xk ∈ F and limit
x ∈ X, then by definition there must exist a convergent subsequence {xnk }k∈N
with limit x0 ∈ F . But, by Lemma 2.20, these limits must be the same; i.e.,
x = x0 . Thus x ∈ F and by Proposition 2.21 the set F is closed.
2. Assume F ⊆ K is closed and K is compact. Let {xk }k∈N be any sequence in

F . Since each xk ∈ K there must exist a convergent subsequence {xnk }k∈N with
limit x ∈ K. But, by Proposition 2.21 this limit must be in F . This shows F is
compact.
Are all closed sets compact? No: e.g., [0, ∞) is closed in R, but xn = n has no
convergent subsequence. The problem here is that [0, ∞) is unbounded.
Definition 2.23 (Bounded sets). A subset A ⊆ X of a metric space X is bounded if
there exists R > 0 and x ∈ X such that A ⊆ BR (x).
Lemma 2.24. [Compact sets are bounded] Let K ⊆ X be a compact subset of a metric
space X. Then K is bounded.
Proof. We employ a contrapositive argument. Assume K is not bounded and fix x ∈ K.

For each k ∈ N we can therefore find a point xk ∈ K such that d(xk , x) ≥ k (since
Bk (x) does not contain K). However, the sequence {xk }k∈N cannot have a convergent
subsequence; given any potential limit point x0 of xnk we have
(D3)
d(xnk , x0 ) ≥ d(xnk , x) − d(x, x0 ) ≥ nk − d(x, x0 ) → ∞ as k → ∞.
So, we know that compact sets are closed and bounded. Is that enough? Recall, the
theorem of Bolzano-Weierstrass from Analysis I. It states precisely that the closed
and bounded intervals [a, b] in R are compact.
Theorem 2.25. [Heine-Borel for Rn ]
A subset K of Rn is compact ⇐⇒ K is closed and bounded.
Remark 2.26. Heine-Borel does not hold for arbitrary metric spaces.
The proof is simply induction on n, starting from the base ‘n = 1’ case for subsets
of R. For this reason (and since this is a Complex Analysis course), we will prove it for
C and leave the proof in higher dimensions for the enthusiastic reader - the statement
for C corresponds to that of R2 in the statement of Theorem 2.25, since C with its
standard metric can be identified with R2 with its standard metric.
Theorem 2.27 (Heine-Borel for C).
A subset K of C is compact ⇐⇒ K is closed and bounded.
Proof. • (⇒): This has been done via Corollary 2.22 and Lemma 2.24.
• (⇐): Applying Bolzano-Weierstrass, we see that Heine-Borel holds for subsets of

R.
Next, let K ⊆ C be a closed and bounded subset of C. It is very easy to show that
if K is bounded then there exists R > 0 such that K ⊆ BR (0). Let R > 0 be such
a number (so that |z| < R for every z ∈ K) and let {zk }k∈N be a sequence in C with
zn = xn + iyn . Note that |xn | < R and |yn | < R. We wish to show there is a convergent
subsequence with limit in K.
The real interval [−R, R] is closed and bounded, so by Heine-Borel for R the interval
[−R, R] is compact. Notice that xk ∈ (−R, R) ⊂ [−R, R]. Since {xk }k∈N is a sequence
in [−R, R] there must be a convergent subsequence {xnk }k∈N with limit in [−R, R], say
xnk → x ∈ [−R, R].
Consider the corresponding complex subsequence {znk }k∈N , where znk = xnk +
iynk . Its imaginary part, the real sequence {ynk }k∈N , also lies in the compact set
[−R, R]. So, it also must have a convergent subsequence {ynmk }k∈N (so a subsequence
of a subsequence!!) which converges to some y ∈ [−R, R].
Finally, take the (sub)subsequence {znmk }k∈N , where znmk = xnmk + iynmk . The
imaginary part converges to y and, by Lemma 2.20, the real part converges to x. Then,
by the facts we discussed about convergence of complex sequences, the subsequence
{znmk }k∈N converges with limit x + iy. Since K is closed by assumption, it follows from
Proposition 2.21 that this limit x + iy is in K. This proves K is compact.
Remark 2.28. • The complex plane C is not compact with respect to the standard
metric; e.g., the sequence {ik}k∈N has no convergent subsequence. (Neither is
Rn .)
• The Riemann sphere S 2 in R3 is compact as a subset of R3 (with the usual

Euclidean metric). Hence Ĉ = C ∪ {∞} is compact with respect to the chordal
metric. (One way to prove this formally will be Theorem 2.30 below).
• The orthogonal group O(n) and the unitary group U(n) are compact; SLn (R) and
GLn (R) are not (the same holds for matrices with entries in C) - see sheet 3.
Finally, before stating the connection between continuous functions and compact
sets, we restate continuity in terms of convergent sequences:
Lemma 2.29. A function f : X → Y between two metric spaces is continuous at x ∈ X

if and only if
lim f (xn ) = f (x) for every convergent sequence {xn }n∈N in X with xn → x.
n→∞
Proof. See sheet 3.
Recall (from Analysis I) that non-empty compact sets in R have a minimal and
maximal element. [This is because f (x) = |x| is continuous - the statement you have
seen says continuous functions on compact sets attain their max/min.]
Theorem 2.30. Let f : X → Y be a map between two metric spaces. Then,
K ⊂ X is compact and f is continuous =⇒ the image f (K) is compact in Y.
In particular, for Y = R, any continuous real-valued function on a metric space X

attains minima and maxima on compact sets.
Proof. Let {yk }k∈N be a sequence in f (K), say with f (xk ) = yk . We wish to show it
has a convergent subsequence with limit in f (K). Since K is compact, there must be
a convergent subsequence {xnk }k∈N (of the sequence {xk }k∈N ) with some limit x in K.
Since f is continuous, we have by Lemma 2.29 that xnk → x implies ynk = f (xnk ) →
f (x). Thus {ynk }k∈N has limit f (x), and because x ∈ K we must have f (x) ∈ f (K).
Remark 2.31. It follows that if K is a compact subset of C, then Re(z), Im(z) and |z|
all attain maximum and minimum values on K. In term 2, we will find out a stronger
statement under the additional assumptions that f is (complex) differentiable and K is
‘nice’; then the maximum modulus of a function occurs on the boundary of K. This is
called the maximum modulus theorem.
Chapter 3
Complex Differentiation
3.1 Complex differentiability

Differentiation of a complex function is defined in a similar way to that in R, except
using complex limits.
Definition 3.1 (Complex differentiability). A function f : U → C defined on an open

set U in C is (complex) differentiable at z0 ∈ U if
f (z) − f (z0 )
lim exists.
z→z0 z − z0
We call this limit the derivative of f at z0 and write f 0 (z0 ) for the limit, i.e.
f (z0 + h) − f (z0 )
f 0 (z0 ) = lim .
h→0 h
Remark 3.2. • In the second formulation the quantity h is a complex number (not
a real number), so the limit must exist from every direction. The second is often
the more useful expression of the two.
• Note that if a function f is complex differentiable at z then it is continuous at z.
Differentiating complex functions from first principles (examples)

1. Rules for differentiating polynomials are the same. For example, consider f (z) =
z 2 on C. For any z ∈ C we have
(z + h)2 − z 2 z 2 + 2hz + h2 − z 2
lim = lim = lim (2z + h) = 2z.
h→0 h h→0 h h→0
Thus, f is differentiable on C and f 0 (z) = 2z as expected.
33
CHAPTER 3. COMPLEX DIFFERENTIATION 34
2. Consider f (z) = z̄. For it to be differentiable we must obtain the same limit from
every direction. But, considering limits from the purely real and purely imaginary
directions, for every z ∈ C we have
z + h − z̄ h
lim = lim = 1,
h→0 h h→0 h
h∈R h∈R
yet
z + ih − z̄ −ih
lim = lim = −1.
h→0 ih h→0 ih
h∈R h∈R
Since z was arbitrary this shows f is not differentiable anywhere.
3. As in the real case, sums/products/quotients of complex differentiable functions
are complex differentiable where defined (e.g., all polynomials/rational functions).
In particular, the product and quotient rules hold for complex derivatives.
4. Composition of differentiable functions are complex differentiable where defined.
In particular, the chain rule holds for complex derivatives. The proofs of 3.
and 4. here are almost identical to those from Analysis I, so are excluded.
5. Generally, non-constant purely real/imaginary functions are not complex differen-
tiable ; e.g., Re(z), Im(z), |z| are nowhere differentiable as functions from C → C.
3.2 Cauchy-Riemann equations

We have a basic way of determining whether a function is complex differentiable, but
for more complicated functions we wish to find some nicer conditions. We consider a
complex valued function f , defined for z = x + iy in some set U ⊆ C, as a function of
the two real variables x and y. We write
f (z) = u(x, y) + iv(x, y)
and call Re(f ) = u(x, y) the real part of f and Im(f ) = v(x, y) the imaginary part of f .
For example, if f (z) = z 2 then z 2 = (x + iy)2 = x2 − y 2 + i2xy, so u(x, y) = x2 − y 2
and v(x, y) = 2xy.
When considered as real functions R2 → R there are naturally two partial derivatives
associated with each of u and v:
∂ u(x + h, y) − u(x, y) u(x, y + h) − u(x, y)
ux (x, y) = u(x, y) := lim , uy (x, y) := lim ,
∂x h→0 h h→0 h
v(x + h, y) − v(x, y) v(x, y + h) − v(x, y)
vx (x, y) := lim , vy (x, y) := lim .
h→0 h h→0 h
Note that the limits above are standard real limits in R (so h is real).
Proposition 3.3. [Cauchy-Riemann equations] Let f = u+iv be complex differentiable

at z0 . Then the real partial derivatives ux , uy , vx , vy exist at z0 and satisfy the Cauchy-
Riemann equations:
ux (z0 ) = vy (z0 ) uy (z0 ) = −vx (z0 ).
Furthermore, the derivative of f at z0 can be written as
f 0 (z0 ) = ux (z0 ) + ivx (z0 ) = vy (z0 ) − iuy (z0 )

= ux (z0 ) − iuy (z0 ) = vy (z0 ) + ivx (z0 ).
Proof. Since f is complex differentiable at z0 = x0 + iy0 , the limit in the definition must
exist and agree no matter which way we approach z0 . Choosing the purely real and
purely imaginary directions we have
f (z0 + h) − f (z0 ) f (z0 + ih) − f (z0 )

lim = f 0 (z0 ) = lim .
h→0 h h→0 ih
h∈R h∈R
We may split each limit into real and imaginary parts, by standard properties of complex
limits. Since, when h is real,
f (z0 + h) = f ((x0 + h) + iy0 ) = u(x0 + h, y0 ) + iv(x0 + h, y0 ), and

f (z0 + ih) = f (x0 + i(y0 + h)) = u(x0 , y0 + h) + iv(x0 , y0 + h),
the above reads

u(x0 + h, y0 ) − u(x0 , y0 ) v(x0 + h, y0 ) − v(x0 , y0 )
lim + i lim = f 0 (z0 )
h→0 h h→0 h
1 u(x0 , y0 + h) − u(x0 , y0 ) i v(x0 , y0 + h) − v(x0 , y0 )
= lim + lim .
i h→0 h i h→0 h
In other words, ux (z0 ) + ivx (z0 ) = f 0 (z0 ) = (1/i)uy (z0 ) + vy (z0 ). Since 1/i = −i the
statement follows by comparing real and imaginary parts.
Returning to the function f (z) = z 2 , note that the C-R equations indeed hold: We
have u(x, y) = x2 − y 2 and v(x, y) = 2xy and so
ux = 2x, vy = 2x, uy = −2y, vx = 2y.
Remark 3.4. [Warning] Note that Proposition 3.3 provides a way of showing a function is
not differentiable at a given point. It cannot be used to prove a function is differentiable.
I.e., it is only a necessary condition, not a sufficient one.
It turns out, by considering a complex function f : C → C as a real function

f : R2 → R2 : (x, y) 7→ (u(x, y), v(x, y)), that we have precisely
f is complex differentiable ⇐⇒ f is real differentiable + C-R equations hold;
so indeed we do have a converse:
Theorem 3.5. Let f = u + iv be defined on an open subset U of C. Assume the

partial derivatives ux , uy , vx , vy exist, are continuous, and satisfy the Cauchy-Riemann
equations at z0 ∈ U . Then f is complex differentiable at z0 .
Proof. Omitted. (If you’re interested to see a proof of this theorem, then you can find
one in Section 5.6, page 59, of the book “Introduction to Complex Analysis” by H. A.
Priestley, Second Edition, Oxford University Press, 2009.)
Determining complex differentiability via Cauchy-Riemann (examples)

1. Let f (z) = exp(z) = ex cos y + iex sin y. Then
ux = ex cos y, vy = ex cos y, uy = −ex sin y, vx = ex sin y.
All these functions are continuous as real functions (see Calculus I/AMV II) and
the C-R equations hold. Thus, by Theorem 3.5, exp is differentiable everywhere
in C and by Proposition 3.3
exp0 (z) = ux + ivx = ex cos y + iex sin y = exp(z).
2. By the chain rule, f (z) = eiz is differentiable and f 0 (z) = ieiz . Since they are just
sums of exp, all of the functions sin, cos, sinh, cosh are differentiable everywhere
in C and you can verify
sin0 (z) = cos z, cos0 (z) = − sin z, sinh0 (z) = cosh z, cosh0 (z) = sinh z.
Similarly, all polynomials/rational functions are differentiable with the usual for-
mulae: e.g., for a0 . . . an complex;
(an z n + · · · + a2 z 2 + a1 z + a0 )0 = nan z n−1 + · · · + 2a2 z + a1 .
For the branch of log corresponding to arguments in (θ1 , θ2 ], the function log(z) is
differentiable in C \ Rθ1 , i.e. at all points outside the branch cut. At these points,
the derivative is given by log0 (z) = 1/z (see Sheet 4 Q7 where this is proved for
the principal branch).
3. f (z) = sin(z)/z 2 is differentiable everywhere except z = 0, since it is the quo-

tient of two complex differentiable functions (and is not defined when z 2 = 0).
Furthermore,
0 (cos z)(z 2 ) − (sin z)(2z) z cos z − 2 sin z
sin(z)/z 2 =

= .
(z 2 )2 z3
4. Let f (z) = f (x + iy) = (x3 + 3x2 y − y 3 − x2 − 2y 2 ) + i(−x3 + 3xy 2 − y 3 + 4xy + 3y).

Then,
ux = 3x2 + 6xy − 2x, vy = 6xy − 3y 2 + 4x + 3,

uy = 3x2 − 3y 2 − 4y, vx = −3x2 + 3y 2 + 4y.
Thus, the partial derivatives exist and are continuous everywhere. We have uy =
−vx everywhere, but ux = vy if and only if 3x2 − 2x = −3y 2 + 4x + 3; that√is,
when (x − 1)2 + y 2 = 2. So, f is differentiable only on the circle of radius 2
centred at 1 in the complex plane!
Holomorphicity
In example 4. above, the function is only differentiable on a 1-dimensional subset of C.
This is a similar situation to a real function on R being differentiable only at a single
point - not a very interesting function to work with from an analytical perspective. It
will be useful for us to consider functions that are differentiable on genuine 2-dimensional
sets in C, for this will allow us to (later) express the functions using Taylor series. Such
functions will turn out to have some quite remarkable properties.
Definition 3.6 (Holomorphic functions). A function f : U → C defined on an open set

U ⊂ C is holomorphic on U if it is complex differentiable at every point in U .
We say f is holomorphic at z0 ∈ U if it is holomorphic on some open ball B (z0 );

(in other words, if there exists > 0 such that f is complex differentiable at every point
in B (z0 )).
Remark 3.7. • By the previous examples, exp, trig functions, and polynomials are
holomorphic on all of C. Logarithms and complex powers are holomorphic at
points outside their branch cuts.
• The function in the previous example 4. is nowhere holomorphic despite being

differentiable on a circle. To see this, simply note that any (open) ball in C
centred at a point z on this circle must contain a point w not on the circle. The
function is not complex differentiable at w, so is not holomorphic at z. Thus, being
holomorphic is a strictly stronger property than being complex differentiable.
3.3 Connected sets and zero derivatives

We explore what it means for the derivative to be zero for a complex function. Consider,
for example, the function f : U → C, where U = {z ∈ C : |z| = 6 1} and
(
1 if |z| < 1,
f (z) =
2 if |z| > 1.
The set U is open, the function f is clearly holomorphic on U and has zero derivative
on U . However, it takes two different values - it is not constant. The problem is that
f lives on two ‘unconnected’ subsets of the complex plane. We wish to find conditions
on the set U under which we can conclude f is constant (as we are used to for real
functions).
First, we need some terminology:
Definition 3.8. [Paths & path-connectedness]
1. A path or curve (from a ∈ C to b ∈ C) is a continuous function γ : [0, 1] → C

with γ(0) = a and γ(1) = b. We say the path/curve is closed if a = b (in this
case, the endpoints of the path join up).
2. A path/curve is smooth if it is continuously differentiable. (Our definition of the

derivative of a function from R → C is the ‘obvious’ one.)
3. We say a subset U ⊆ C is path-connected if for every pair of points a, b ∈ U

there exists a smooth path from a to b such that γ(t) ∈ U for every t ∈ [0, 1].
Remark 3.9. • [Advanced/Warning] These definitions are not universal to all of

mathematics. Outside of complex analysis, people often use ‘smooth’ to refer
to the property of having infinitely many derivatives. Also, usually people use
‘path-connected’ to mean you can connect two points simply by a continuous
curve. We used a stronger definition to make our life easier.
Intuitively, a subset is path-connected if it has ‘one piece’; i.e., if you can draw a line
between any two points in the set without the pencil leaving the subset (or the paper!).
Definition 3.10 (Domains). A domain D is an open, path-connected subset of C.

Some people call domains regions.
Showing sets are domains (examples)

1. We know both C and Br (z) (for any z ∈ C and r > 0) are open. They are
also path-connected: Simply choose the line segment between a and b given by
γ(t) = a + (b − a)t. This is clearly a smooth path, so C and Br (z) are domains.
2. The set C \ R≤0 , on which the principal branches of log and complex powers are
defined, is path-connected: Choose arbitrary a, b ∈ C \ R≤0 . If a lies on the
positive real axis then the line segment again works (since it avoids the negative
real axis). Otherwise, we could try letting
γ1 (t) = |a|eiArg(a)(1−t) , γ2 (t) = |a| + (b − |a|)t.
The first path takes a to |a|, avoiding R≤0 by tracing around the circle of radius
|a| centred at the origin. The second joins |a| to b via a straight line. See Figure
3.1. (You can reparametrize to make t run from 0 to 1 on the combined path if
you like.) But the combined path running through γ1 then γ2 is not differentiable
at the point where the paths γ1 and γ2 meet! One possibility for a differentiable
path joining the points a and b would be the arc of the circle through a and b that
does not cross R≤0 .
We know C \ R≤0 is also open, so it is a domain.
Figure 3.1: A depiction of the paths when a is on the real line, and when a is not on
the real line.
3. The set {z ∈ C : |z| =

6 1} is not a domain. It is open, but it is not path-connected.
Pick a point a with |a| < 1 and b with |b| > 1. We cannot draw a continuous path
from a to b without crossing the circle |z| = 1; i.e., you cannot draw a continuous
path between the points without leaving the set.
Previously we have seen that there is a chain rule for the composition of two complex
differentiable functions. There is also the chain rule for the composition of a complex
differentiable function and a smooth path:
Lemma 3.11. [Chain rule] Let U ⊂ C be an open set, f : U → C be a holomorphic

function on C and γ : [0, 1] → U be a smooth path. Then for t0 ∈ [0, 1] we have
(f ◦ γ)0 (t0 ) = f 0 (γ(t0 ))γ 0 (t0 ).

Proof. Similar to the proof of the chain rule in Analysis I.
Theorem 3.12. Let f : D → C be holomorphic on a domain D ⊆ C. If f 0 (z) = 0 for

every z ∈ D then f is constant on D.
Proof. Since D is path-connected, it is enough to show f is constant on any smooth
path γ lying in D. In other words we wish to show f ◦ γ is a constant function of t.
We have (f ◦ γ)0 (t) = f 0 (γ(t))γ 0 (t) = 0 by the chain rule. Since f = u + iv, we have
0 = (f ◦ γ)0 (t) = (u ◦ γ)0 (t) + i(v ◦ γ)0 (t),

which means both (u ◦ γ)0 (t) and (v ◦ γ)0 (t) are equal to 0 for all t. Hence u and v are
constant along γ, so f is too.
Remark 3.13. The same statement holds if “f 0 = 0” is replaced with “f is purely

real/imaginary”, or “f has constant real/imaginary part”, or “f has constant modulus”
(see Sheet 4). So none of Re(z), Im(z), |z| or Arg(z) are holomorphic anywhere (since
they are non-constant real-valued functions).
3.4 The angle-preserving properties of holomorphic func-

tions
Given a smooth path γ : [0, 1] → C, we can talk about the tangent vector to γ at a
point z0 = γ(t0 ) on the path.
Definition 3.14 (Conformal maps). We say a (real differentiable) map f : D → C on
a domain D ⊆ C is conformal at z0 if it preserves the angle and orientation between
any two tangent vectors at z0 . This is exactly the same as saying that it preserves the
angle and orientation between any two smooth curves passing through z0 . We say that
f is conformal if it is conformal at all points in D.
It turns out that there is a close relationship between holomorphic and conformal
maps.
Lemma 3.15. [Holomorphic maps are conformal] A holomorphic map f with f 0 (z0 ) 6= 0
is conformal at z0 .
Proof. Take a smooth path γ : [0, 1] → C that passes through z0 , and consider the
tangent vector to γ at the point z0 = γ(t0 ) on the path. What happens to this tangent
vector under the application of a holomorphic function f ?
The path f ◦ γ is the path obtained by mapping γ under f . So to calculate what
happens to the tangent vector γ 0 (t0 ) under f , we should calculate the tangent vector
(f ◦ γ)0 (t0 ). This is done using the chain rule:
(f ◦ γ)0 (t0 ) = f 0 (γ(t0 ))γ 0 (t0 ) = f 0 (z0 )γ 0 (t0 ).
This describes what happens to tangent vectors geometrically: f transforms the tangent
vector γ 0 (t0 ) by multiplying it by the complex number f 0 (z0 ).
But we understand what multiplication by complex numbers means geometrically:
if f 0 (z0 ) 6= 0 then multiplcation by f 0 (z0 ) is dilation (by |f 0 (z0 )|) followed by a rotation
(by Arg(f 0 (z0 ))).
We now notice the following: both dilations and rotations preserve the angles and
orientations between vectors! This completes the proof.
Determining where maps are conformal (examples)

• Let f (z) = z 2 . Here, f 0 (z) = 2z so f 0 (z) = 0 ⇐⇒ z = 0 and so f is conformal on
C∗ = C \ {0} (since it is holomorphic everywhere). We don’t yet know whether f
is conformal at the origin, but we can check: we know that f 0 (z) = 2z so f 0 (0) = 0
and hence the action of f on tangent vectors at the origin is to multiply by 0,
hence sending any tangent vector to zero.
• Let f (z) = 32 z 3 + 12 (1 − 2i)z 2 − iz + 2 − 1. This function is clearly holomorphic as

it is just a polynomial. We have
f 0 (z) = 2z 2 + (1 − 2i)z − i = (2z + 1)(z − i),
which has zeros at − 21 and i, so f is conformal on C \ {− 12 , i}. [Note, we can

explicitly check what f does to tangent vectors at each point in this set: For
example, at z = i/2 we have f 0 (i/2) = (i + 1)(−i/2) = (1 − i)/2 = √12 e−iπ/4 .
Thus, at z = i/2 the function f rotates tangent √ vectors by Arg(f 0 (i/2)) = −π/4
degrees, then dilates by a factor of |f 0 (z)| = 1/ 2.]
Rather than having to then check two paths through z = 1/2 and z = i respec-
tively, to determine whether f is conformal there, it will be convenient to have
a complete description in terms of holomorphicity. The converse to Lemma 3.15
holds:
Proposition 3.16. [Conformal maps are holomorphic] Let D be a domain. If f is

conformal at z0 ∈ D then f is complex differentiable at z0 and f 0 (z0 ) 6= 0. Therefore, if
f is conformal on D, then f is holomorphic on D and f 0 (z) 6= 0 for all z ∈ D. Thus
f is conformal on D ⇐⇒ f is holomorphic with f 0 (z) 6= 0 for all z ∈ D.
Proof. (Non-examinable) Sketch: we know f = u + iv is real differentiable when con-

x u(x,y)
sidered as a map f : ( y ) 7→ v(x,y) . The total differential of this map at a point

u (x ,y ) u (x ,y )
z0 = x0 + iy0 (see AMV II) is the Jacobian matrix Dz0 = vxx(x00,y00) vyy(x00,y00) . To pre-
serve angles it must be of the form c cos θ − sin θ for some c ∈ R . Thus D

sin θ cos θ 6=0 z0 is of the
form ab −b

a
and the C-R equations are satisfied at z0 . Since f is real differentiable by
assumption, C-R equations are actually enough to show that f is complex differentiable
at z0 (we have not stated exactly this result before, but it holds). Since det Dz 6= 0 we
have f 0 (z0 ) = ux (x0 , y0 ) + ivx (x0 , y0 ) 6= 0.
Remark 3.17. [Advanced] The idea of this proof, in reverse, can be adapted to give a
more sophisticated proof of the CR equations!
Determining where maps are conformal (examples continued)

• Let f (z) = xy + iy 2 . Then ux = y, vy = 2y, uy = x, vx = 0. So, the C-R equations
only hold at y = x = 0. Hence f is not conformal at any point z0 6= 0, since it
cannot be complex differentiable at z0 (as C-R. equations fail to hold), and hence
it is not conformal at z0 by Proposition 3.16. What happens for z0 = 0? There,
f 0 (0) = ux (0, 0) + ivx (0, 0) = 0 + i0 = 0. So f cannot be conformal at 0 either by
Proposition 3.16. Thus, f is nowhere conformal.
Visualising conformal maps

Here is a useful result, helpful in visualising the action of conformal maps:
Corollary 3.18. Any conformal map maps orthogonal grids in the (x, y)-plane to or-
thogonal grids.
Remark 3.19. [Important warning] We must allow our grids to be made up of

arbitrary smooth curves, not necessarily straight lines (Corollary 3.18 is false if they are
required to be made of straight lines).
How to visualise conformal maps (examples)

• Let f (z) = z 2 = x2 − y 2 + i2xy and consider the simple grid in the (x, y)-plane
made up of lines parallel to the real/imaginary axes, separated by distance 1.
Given a line (not passing through the origin) in such a grid, say x = a (6= 0),
the function f takes this line to the points (a2 − y 2 , 2ay) in the (u, v)-plane. If
v 2
u = a2 − y 2 and v = 2ay, then v 2 = 4a2 y 2 and so u = a2 − ( 2a ) and the image
defines a parabola.
v 2
Similarly, the line y = b (6= 0) is taken to the parabola u = ( 2b ) − b2 in the (u, v)-
plane. By sketching these on the same axes we see that the parabolas indeed cross
at right angles.
• For the same example, consider the level curves u(x, y) = a and v(x, y) = b with
b
a, b 6= 0. These trace out the curves y 2 = x2 − a and y = 2x respectively. By
sketching these curves on a graph (for, say, u = ±1, v = ±1) we see that they are
Figure 3.2: The image of the perpendicular lines x = 1, y = 1 is a pair of parabolas

that cross at right angles.
perpendicular. (They have to be, since f (z) is conformal on C − {0} and they
map to perpendicular straight lines.)
3.5 Biholomorphic maps

We wish to build a dictionary of conformal maps as a way of getting back and forth
between domains of the complex plane. To do this, we need our maps to be invertible.
Definition 3.20 (Biholomorphic maps). Let D and D0 be domains. We say that
f : D → D0 is biholomorphic if f is holomorphic, a bijection, and the inverse f −1 :
D0 → D is also holomorphic. A biholomorphic map f is called a biholomorphism.
When f as above exists, we say that the domains D and D0 are biholomorphic and
∼
→ D0 .
write f : D −
Remark 3.21. [Advanced] It is possible to prove that if f as above is holomorphic and
bijective, then it is automatic that its inverse is holomorphic.
Finding biholomorphisms (examples)

1. The function exp : C → C∗ is not biholomorphic since it is not injective (e.g.,
ez = ez+2nπi ). How can we make it biholomorphic? We simply need to restrict it
to a smaller domain where it will be injective. It is not too hard to check that
exp(z1 ) = exp(z2 )
if and only if Re(z1 ) = Re(z2 ) and Im(z1 ) − Im(z2 ) ∈ 2πZ. So if we rule out
different z1 and z2 whose imaginary parts differ by an non-zero integer multiple
of 2π then we might be ok. We can do this by setting
D := {z ∈ C : Im(z) ∈ (−π, π) }.
If z = x + iy ∈ D then exp(z) = ex eiy so the principal argument of exp(z) is

never equal to π. On the other hand, it is not hard to check that exp maps D
onto C − R≤0 . Moreover, by our choice of D, exp is injective on D. So exp is
a bijection between D and C − R≤0 . Therefore it has an inverse function, and
we already know what it is: the principal branch of Log. The function Log maps
C − R≤0 to D and is differentiable at every point of C − R≤0 , so it is holomorphic
on C − R≤0 . In summary, when restricted to D, exp is biholomorphic and hence
D and C − R≤0 are biholomorphic.
2. Let f (z) = z 2 . We have seen that this function is holomorphic on C. However, as

in the previous example, it is not injective since e.g. f (1) = f (−1) = 1. Note that
f (z1 ) = f (z2 ) with z1 6= z2 if and only if z1 = ±z2 , so we may be able to create
a biholomorphic function by restricting f to a domain that never simultaneously
contains z and −z. Such a domain is the right half plane HR := {z ∈ C : Re(z) >
0}. We saw in a previous example that f maps HR to C − R≤0 . In fact this is a
bijection, with inverse given by
1
f −1 : C − R≤0 → HR , f −1 (z) := exp( Log(z)).
2
Thus f −1 is holomorphic, since it is a composition of holomorphic functions, and
∼
hence f gives a biholomorphism f : HR − → C − R≤0 .
∼
3. The ‘affine’ linear maps z 7→ az + b (for a ∈ C∗ , b ∈ C) are biholomorphic C −
→ C.
As expected we can compose biholomorphic maps to construct new ones.
Lemma 3.22 (Automorphism groups). Let D ⊂ C be a domain. The set of all biholo-
∼
morphic maps f : D −→ D from D to itself forms a group under composition. We call
this group the automorphism group of D and denote it by Aut(D).
Proof. We need only check the group properties:
• The identity map Id : z 7→ z is always in Aut(D), and is clearly biholomorphic.
• We know composition of functions is associative, so f ◦ (g ◦ h) = (f ◦ g) ◦ h, for

holomorphic maps.
• f −1 is the inverse of f since f ◦ f −1 = Id = f −1 ◦ f . Moreover, f −1 is in

Aut(D) since it is itself holomorphic by assumption and has holomorphic inverse
(f −1 )−1 = f .
• Aut(D) is closed: (f ◦ g) is holomorphic (by the chain rule) and its inverse
(f ◦ g)−1 := g −1 ◦ f −1 is also the composition of two holomorphic functions,
so holomorphic.
Remark 3.23. Note that example 1. shows conformal maps are not necessarily biholo-
morphic. On the other hand, it turns out (see Sheet 5) that all biholomorphic maps are
conformal. [Whilst the converse is not true in general, conformal maps turn out to be
‘locally’ biholomorphic.]
Chapter 4
Möbius transformations
4.1 Definition and first properties of Möbius transforma-

tions
In this chapter we study a very special class of functions that play a truly fundamental
role in mathematics, and have many beautiful properties. Recall that GL2 (C) is the
collection of matrices

a b
GL2 (C) = : a, b, c, d ∈ C, ad − bc 6= 0
c d
= {2 × 2 complex matrices with nonzero determinant}.
Also recall the extended complex plane Ĉ = C ∪ {∞} from Chapter 1.

a b
Definition 4.1 (Möbius transformations). Given any matrix T = ∈ GL2 (C)
c d
we can define a function
MT : C → Ĉ
by the formula
az + b
MT (z) =
cz + d
if cz + d 6= 0, and if cz + d = 0, then we set MT (z) = ∞. The function MT is called a
Möbius transformation.
Remark 4.2. • [Advanced] The reason for excluding matrices for which det T = 0 is
because if det T = 0 and at least one of c and d is non-zero, so that the definition
makes sense, then T has rank 1, and this implies that (a, b) = λ(c, d) for some
λ ∈ C. But this would give that for z with cz + d 6= 0, we would have MT (z) = λ,
46
CHAPTER 4. MÖBIUS TRANSFORMATIONS 47
so the transformation would be constant. This is not a very interesting function

to think about!
• For T ∈ GL2 (C), by taking complex square roots, we can find a number such that
k 2 = det T . Then
az b
az + b k + k
MT (z) = = cz d
= M k1 T (z)
cz + d k + k
and
1 1
det( k T ) = det T = 1.
k2
Thus, we may scale any T ∈ GL2 (C) to obtain T 0 = T
k such that det T 0 = 1, and
M T = MT 0 .
It is slightly annoying that at the moment, the domain of a Möbius transformation
is not the same as its codomain. This would prevent us from composing two Möbius
transformations. However, it is possible to extend the definition to describe where
∞ ∈ Ĉ gets mapped to: we declare that
(
a
if c 6= 0
MT (∞) = c .
∞ if c = 0
Now we have MT : Ĉ → Ĉ.
Example4.3. 1. f (z) = z −1 is a Möbius transformation corresponding to the ma-

0 1
trix . Note that |z| < 1 and z 6= 0 if and only if |f (z)| = |z −1 | = |z|−1 >
1 0
1, so f maps the punctured unit ball B1 (0) − {0} onto the outside of the closed
unit ball. We have f (0) = ∞ and f (∞) = 0, so f interchanges these two points.
Figure 4.1: Image of B1 (0) − {0} under f .


z−i 1 −i
2. The Cayley map f (z) = z+i corresponds to the matrix . We have
1 i
f (z) ∈ B1 (0) ⇐⇒ |f (z)| < 1 ⇐⇒ |z+i| > |z−i| ⇐⇒ z ∈ H = {z ∈ C : Im(z) > 0}.
Hence f maps the upper half plane to the open unit ball centered at 0. Moreover,
f (∞) = 1 and f (−i) = ∞.
Figure 4.2: Image of upper half-plane under Cayley map.
Lemma 4.4. The set of Möbius transformations form a group under composition. Fur-
thermore,
1. MT1 ◦ MT2 = MT1 T2 .

2. (MT )−1 = MT −1 .

1 0
3. MT = Id ⇐⇒ T =t (t ∈ C∗ ).
0 1
Proof. See Sheet 6.
Remark 4.5. Recall from Linear Algebra that GL2 (C) forms a group under matrix
multiplication. Lemma 4.4 says more than that the Möbius transformations form a
group. It says that the mapping
T 7→ MT
is a group homomorphism between GL2 (C) and the Möbius transformations! In other
words, composing Möbius transformations is basically just multiplying matrices!

a b
Lemma 4.6. Let T = ∈ GL2 (C). If c = 0, the Möbius transformation MT
c d
gives a biholomorphic map
∼
MT : C −→ C.
If c 6= 0, then MT gives a biholomorphic map

−d ∼ nao
MT : C − −
→C− .
c c
Proof. If c = 0 then a, d 6= 0 since det T 6= 0. Hence

az + b a b
MT (z) = = z+
d d d
is just an affine linear map which is holomorphic. It is a bijection with inverse given
by (MT )−1 = ad z − ab , which is holomorphic by the same token. So MT gives a
∼
biholomorphic map C − → C.
If c 6= 0 let’s check that MT is complex differentiable at every z 6= −dc . We can
calculate
a(cz + d) − c(az + b) det(T )

MT0 (z) = = .
(cz + d)2 (cz + d)2
Therefore the derivative exists at all points of C − −d

c and hence MT is holomorphic
there. It is bijective with the inverse given by the inverse Möbius transformation,
which
is holomorphic
a by the same argument. Hence MT is a biholomorphism from
∼
C − −d c −→ C − c .
Corollary 4.7. A Möbius transformation MT is conformal at all z ∈ C with MT (z) 6=

∞.
Proof. We know that biholomorphic maps are conformal, and MT is a biholomorphism
on C minus the points that map to ∞. Alternatively, one can check that in both cases
of the proof of Lemma 4.6, the derivative never vanishes.
Corollary 4.8. Any Möbius transformation is a bijection from Ĉ to Ĉ.

Proof. By Lemma 4.4 MT has an inverse MT −1 , so that MT MT −1 = MT −1 MT as maps
from Ĉ to Ĉ, so MT is a bijection.
4.2 Fixed points, the cross-ratio, and the three points

Theorem
Let T ∈ GL2 (C) and MT be a Möbius transformation. Then a point z is a fixed point
of MT if MT (z) = z.
Lemma 4.9. Let T ∈ GL2 (C). If MT : Ĉ → Ĉ is not the identity map, then MT has
at most 2 fixed points in Ĉ. In other words, if a Möbius transformation has three fixed
points in Ĉ, then it is the identity.
Proof. We split into two cases.
First suppose MT (∞) = ∞. From the definition, this can only happen if c = 0, so
MT preserves C, and for z ∈ C, MT (z) = ad z + db with a, d 6= 0 (because det T 6= 0).
Such an affine linear map has at most one fixed point as follows:
• If a = d then b 6= 0 since we assume MT is not the identity. Then MT is a

translation by b/d which has no fixed points.
• If a 6= d then MT has a unique fixed point in C given by z0 = b/(d − a).
Now suppose MT (∞) 6= ∞. Then any fixed points of MT are in C. Suppose z0 ∈ C is
such that MT (z0 ) = z0 . We have
az0 + b
M  (z
0) = z0 ⇐⇒ = z0 ⇐⇒ cz02 + (d − a)z0 − b = 0.

a b  cz0 + d
c d
There are at most two roots of this quadratic equation, so at most two fixed points of
MT in Ĉ.
We introduce an important quantity called the cross-ratio.

Definition 4.10. Given four distinct points z0 , z1 , z2 , z3 ∈ C, the cross-ratio of these
points is denoted (z0 , z1 ; z2 , z3 ) and defined by
(z0 − z2 )(z1 − z3 )
(z0 , z1 ; z2 , z3 ) := .
(z0 − z3 )(z1 − z2 )
We can extend the definition to the case that one of the points is ∞ by removing all
differences involving that point, for example,
(z1 − z3 )
(∞, z1 ; z2 , z3 ) := .
(z1 − z2 )
We can use the cross ratio to prove that a Möbius transformation is uniquely deter-
mined by how it acts upon any three given points in Ĉ.
Theorem 4.11. [Three points Theorem] Let {z1 , z2 , z3 } and {w1 , w2 , w3 } be two sets of
three ordered distinct points in Ĉ. Then there exists a unique Möbius Transformation
f such that f (zi ) = wi for i = 1, 2, 3.
Proof.
Existence: Consider the functions
F (z) := (z, w1 ; w2 , w3 ), G(z) := (z, z1 ; z2 , z3 ).
These are Möbius transformations with the properties that F (w1 ) = 1, F (w2 ) = 0, F (w3 ) =
∞, and G(z1 ) = 1, G(z2 ) = 0, G(z3 ) = ∞. Therefore F −1 ◦ G maps each zi to wi .
Uniqueness: Assume that there are two such maps, say f and f˜. Then the Möbius
transformation H := f −1 ◦ f˜ satisfies H(zi ) = f −1 (f˜(zi )) = f −1 (wi ) = zi . This shows H
has three fixed points, z1 , z2 , z3 , and by Lemma 4.9 it must therefore be trivial. Thus,
f = f˜.
One has the following beautiful and fundamental fact about Möbius transformations:
Proposition 4.12. [Möbius transformations preserve cross-ratio] Möbius transforma-

tions preserve the cross ratio: if z0 , z1 , z2 , z3 are four distinct points in Ĉ, and f is a
Möbius transformation, then
(f (z0 ), f (z1 ); f (z2 ), f (z3 )) = (z0 , z1 ; z2 , z3 ). (∗)
Proof. Let wi = f (zi ). Let F and G be the functions defined in the proof of Theorem
4.11. Recall that F −1 ◦ G mapped each zi to wi . So does f . Since there is a unique
Möbius transformation with this property, we must have f = F −1 ◦ G. Rearranging,
F ◦ f = G. But this is just the identity stated after applying both functions to z0 :
(f (z0 ), f (z1 ); f (z2 ), f (z3 )) = F ◦ f (z0 ) = G(z0 ) = (z0 , z1 ; z2 , z3 ).
Finding a Möbius transformation from three points (example)

Find the unique Möbius transformation f : Ĉ → Ĉ mapping the points {1, −1, i} to the
points {0, ∞, 1}. Our method is simply to rearrange (∗): Notice that
f (z)
f (z) − w w −1 −1
lim = lim w1 = = 1,
|w|→∞ w1 − w |w|→∞ w − 1 −1
so (∗) reduces to

w1 −w3 z − z2 z1 − z3
1· f (z)−w3 =
z − z3 z1 − z2

0−1 z − (−1) 1−i
⇐⇒ f (z)−1 =
z−i 1 − (−1)
−2(z − i)
⇐⇒ f (z) − 1 =
(z + 1)(1 − i)
(−1 − i)z + (1 + i)
⇐⇒ f (z) =
(1 − i)z + (1 − i)

−1 − i z−1
⇐⇒ f (z) =
1−i z+1
−iz + i
⇐⇒ f (z) = .
z+1
Möbius transformations methods

This leads us to a general strategy to find a Möbius transformation from how it acts on three points:
Simply notice that since the cross ratio is preserved,

f (z) − w2 w1 − w3 z − z2 z 1 − z3
= ,
f (z) − w3 w1 − w2 z − z3 z 1 − z2
and to find f we need only rearrange this equation.

There is also a general strategy to find the image of a region D under a Möbius transformation MT :
1. Find the image MT (∂D) of the boundary ∂D.
2. Find the image MT (z0 ) of a point z0 ∈ D in the interior.
3. The region D0 bounded by MT (∂D) and containing MT (z0 ) is precisely the image
of D under MT , and
∼
MT : D −→ D0 = MT (D).
4.3 Circles and lines

The following property of Möbius transformations is very useful when deciding how
they map different regions of the complex plane.
Proposition 4.13. Möbius transformations map circles and lines in Ĉ to circles and
lines in Ĉ.
Remark 4.14. [Technical] We consider any line to pass through infinity, so that the
above makes sense. By circles in Ĉ we mean simply circles in C.
In order to talk about what happens to circles and lines under Möbius transforma-
tions we need to know the equations of circles and lines. Of course, we already know
the equation of a circle or line, but it turns out these can both be described by the same
type of equation using complex numbers.
A circle of centre α and radius r is given by the equation
|z − α|2 = r2
which can be rewritten as
r2 = (z − α)(z̄ − ᾱ) = z z̄ − αz̄ − ᾱz + αᾱ.

Introducing a new parameter β = αᾱ − r2 ∈ R we have the equation
z z̄ − αz̄ − ᾱz + β = 0.
This is the general equation of a circle, given by parameters α ∈ C and β ∈ R such that
|α|2 − β = r2 > 0.
What about lines? Well a line can always be written as a bisector:
` = {z ∈ C : |z − w1 | = |z − w2 |}
with w1 6= w2 ∈ C. This means
(z − w1 )(z̄ − w¯1 ) = (z − w2 )(z̄ − w¯2 )

or
z z̄ − w1 z̄ − w¯1 z + w1 w¯1 = z z̄ − w2 z̄ − w¯2 z + w2 w¯2 .

Rearranging this, and writing α = w1 − w2 6= 0 and β = w1 w¯1 − w2 w¯2 ∈ R, we have
−αz̄ − ᾱz + β = 0.
This is the general equation of a line in C, given by parameters α ∈ C and β ∈ R such
that α 6= 0.
Therefore we have proved
Lemma 4.15. [Equation of circles and lines in C] Given γ, β ∈ R and α ∈ C, the

equation
γz z̄ − αz̄ − ᾱz + β = 0
describes a circle if γ = 1 and (|α|2 − β) > 0, and a line if γ = 0 and α 6= 0. Conversely,
any circle or line can be described by an equation of this form.
Proof
of Proposition 4.13. Consider an arbitrary Möbius transformation MT where T =
a b
. We can assume by previous remarks that det T = 1. If c = 0, then MT is
c d
affine linear and is easily seen to preserve circles and lines (since rotations, dilations,
and translations obviously do). So we can assume c 6= 0.
Note that
!
az + b caz + cb a(cz + d) cb − ad a 1 a 1 1
MT (z) = = = + = − = − .
cz + d c(cz + d) c(cz + d) c(cz + d) c c(cz + d) c c2 z + dc
Thus MT is just some linear maps composed with the function f (z) = 1/z, called an
inversion. Since linear maps preserve circles and lines, it is therefore enough to consider
the action of f (z) = 1/z on lines and circles. Moreover, we have f −1 (z) = 1/z = f (z)
so it is enough to consider one direction.
Let X be a circle or line. By Lemma 4.15, X is described by an equation γz z̄ −

αz̄ − ᾱz + β = 0 with γ, β ∈ R. We claim that f (X) is described by a similar equation.
Indeed, if z ∈ f (X) and z 6= 0, then f (z) = f −1 (z) ∈ X and this happens if and only if

1 1 1 1
γ −α − ᾱ + β = 0 ⇐⇒ βz z̄ − αz − ᾱz̄ + γ = 0.
z z z z
Finally, let’s check this is the equation of a circle or a line.

First consider the case β = 0. If we started with a line we have α 6= 0. If we started
with a circle we know |α|2 = |α|2 − β > 0 so we have α 6= 0 again. Therefore, the new
equation describes a line.
Next, consider the case β 6= 0. Then we can divide by β to get z z̄ − αβ z − ᾱβ z̄ + βγ = 0.
This looks like the equation of a circle, but we have to check
α γ
|| ||2 − > 0 (4.1)
β β
If we began with a line, then γ = 0 and α 6= 0 so this is clearly true. If we began with a
circle, then γ = 1. If β < 0 then (4.1) is obvious since both terms are nonnegative and
γ/β < 0. If β > 0 then (4.1) follows from |α|2 − β > 0.
We we use the term circline to refer to an object that is either a circle or line.
Hence the previous Proposition could have been stated ‘Möbius transformations pre-
serve circlines’. The fact that Möbius transformations preserve circles and lines is very
powerful when combined with:
Remark 4.16. Any three distinct non-colinear points z1 , z2 , z3 ∈ C uniquely determine
a circle in C passing through those points. Any two distinct points uniquely determine
a line passing through those points.
So to find out where a circle is mapped under a Möbius transformation, one simply
needs to check where three points on the circle are mapped!
Finding the image of regions under Möbius transformations (example)

Find the image of the unit disc D := B1 (0) = {w ∈ C : |w| < 1} under the Möbius
2 + 2i −2 − 6i
transformation corresponding to the matrix T = . To determine
1 −1 − 2i
the image we look at what happens to the boundary: we first find the image of the unit
circle centred at the origin. Consider what happens to the four points on the unit circle,
(2+2i)z−(2+6i)
1, i, −1, −i under the map MT (z) = z−(1+2i) . Under MT :
(2 + 2i) − (2 + 6i) −4i

1 7→ = = 2;
1 − (1 + 2i) −2i
(2 + 2i)i − (2 + 6i) −4 − 4i
i 7→ = = 4;
i − (1 + 2i) −1 − i
−(2 + 2i) − (2 + 6i) −4 − 8i
−1 7→ =
−1 − (1 + 2i) −2 − 2i
By Proposition 4.13 the unit circle must be mapped to a circle, and the three calculations
above tell us that the image of the unit circle must be the circle centred at z = 3 of
radius 1. Note, the action of MT is not something so simple as a rotation; for example
we have
−(2 + 2i)i − (2 + 6i) −8i 12 + 4i
−i 7→ = = , which is near the image of -1 !!
−i − (1 + 2i) −1 − 3i 5
What about the interior of this circle? Let’s pick a point in the unit disc and see
where it is taken. The point z = 0 is an obvious choice. We have
(2 + 2i)0 − (2 + 6i) 2 + 6i (2 + 6i)(1 − 2i) 14 + 2i
MT (0) = = = = ,
0 − (1 + 2i) 1 + 2i 5 5
which lies inside this circle in question. Thus, by continuity MT maps D to the interior
of the circle centred at z = 3 of radius 1 (it cannot map another point z0 ∈ D to
somewhere outside this circle as the image of the path from 0 to z0 would have to
cross the boundary, but Möbius transformations are conformal). Continuity (and the
existence of an inverse) tell us the map must be onto and so
∼
MT : D −
→ {z ∈ C : |z − 3| < 1}.
4.4 The Riemann sphere, revisited

Since we saw earlier that we can identify Ĉ with the Riemann sphere S 2 by stereo-
graphic projection from the north pole, we can now think of Möbius transformations as
bijections from S 2 to S 2 ! In fact, our definitions were carefully chosen so that Möbius
transformations give continuous bijections from S 2 to S 2 .
This is a very fruitful perspective that can unify our picture of Möbius transfor-
mations. I encourage you to try to think about Möbius transformations in this way
whenever you can!
Remark 4.17. It turns out that the definition of holomorphicity can be extended1 to
functions from S 2 to S 2 . In fact, Möbius transformations give biholomorphic maps from
1
The trick is to use the stereographic projection maps to transfer your function to C, and use the
definition of holomorphicity there.
S 2 to S 2 . Even better: the Möbius transformations give all the biholomorphic maps
from S 2 to S 2 .
What do circles and lines in Ĉ correspond to in the Riemann sphere? The answer
is just circles! We can now add some more to our table of correspondences in (1.3):
In S 2 In Ĉ
N ←→ ∞
S ←→ 0
Geometrically:
Circle not through N ←→ Circle
Circle through N ←→ Line
This also justifies our earlier convention that we think of all lines as passing through
infinity.
Remark 4.18. [Advanced] One other nice fact is that stereographic projection itself is
conformal! This means two tangent vectors to the sphere at the same point are mapped
to two tangent vectors in C (at the same point) with the same angle.
z−i
Example 4.19. Consider the map f : Ĉ → Ĉ, f (z) = i z+i . Using stereographic
ˆ
projection, we can think of this as a transformation f of the sphere S 2 . What is it?
We have that fˆ(N ) should correspond to f (∞) = i , fˆ(0, 1, 0) should correspond to
f (i) = 0 and fˆ(0, 0, −1) should correspond to f (0) = 1i = −i. So fˆ maps N, (0, 1, 0), (0, 0, −1)
to (0, 1, 0), (0, 0, −1), (0, −1, 0). We might guess that fˆ is simply a rotation of 90 de-
grees about the x-axis taking the ‘back’ hemisphere to the bottom/south hemisphere,
and this turns out to be correct!
To see why this is true, let’s cheat slightly and assume that the rotation that we’ve
guessed corresponds to a Möbius transformation. Then the Möbius transformation is
determined by what it does to three points, and we’ve already checked that our guess
agrees with the map f at three points ∞, i and 0.
4.5 Möbius transformations preserving the upper half plane

or unit disc
Notation: For a domain D ⊂ C, denote by Mob(D) the set of all Möbius transformations
f such that f (D) = D, i.e. f maps D to D.
Proposition 4.20. [H2H] Every Möbius transformation mapping H to H is of the form
MT with T in the group
SL2 (R) := {T ∈ Mat2 (R) : det T = 1}.
Conversely, every such Möbius transformation maps H to H, and hence gives a biholo-
morphism from H to H.
In more concise terms, we have
f ∈ Mob(H) ⇐⇒ f = MT with T ∈ SL2 (R).
Remark 4.21. This gives us a group homomorphism SL2 (R) → Mob(H) mapping T 7→
MT and hence also a group homomorphism SL2 (R) → Aut(H).
Proof. Any Möbius transformation f : H → H must map the boundary of H to the

boundary of H. Since this boundary is precisely the real line, f : R ∪ {∞} → R ∪ {∞}.
In particular, it must take the ordered set {1, 0, ∞} to {x1 , x2 , x3 } for some xi ∈ R∪{∞}.
Consider the cross-ratio; assuming xi 6= ∞ we have

f (z)−x2

x1 −x3
z−0 1−∞
f (z)−x3 x1 −x2 = = z
z−∞ 1−0
⇐⇒ (f (z) − x2 )(x1 − x3 ) = z(f (z) − x3 )(x1 − x2 )
x3 (x1 − x2 )z + x2 (x3 − x1 )
⇐⇒ f (z) = .
(x1 − x2 )z + (x3 − x1 )
T ∈ GL2 (R).
Thus, all coefficients arereal, so
a b
Furthermore, if T = ∈ GL2 (R) and z = x + iy, then
c d

az + b (az + b)(cz̄ + d)
Im (MT (z)) = Im = Im
cz + d |cz + d|2

adz + bcz̄
= Im
|cz + d|2
(ad − bc)y y det T
= 2
= .
|cz + d| |cz + d|2
We have z ∈ H ⇐⇒ y > 0 so
MT (z) ∈ H ⇐⇒ T ∈ GL2 (R), det T > 0
and so we can replace T by a real matrix of determinant 1 by scaling T by a real

number.
Proposition 4.22. [D2D] Every Möbius transformation from the unit disk D to itself
is of the form MT with T in the set

α β 2 2
SU(1, 1) := T = : α, β ∈ C, det T = |α| − |β| = 1 .
β̄ ᾱ
Conversely, every such Möbius transformation maps D to D and hence gives a biholo-
morphic automorphism of D.
In more concise terms, we have
f ∈ Mob(D) ⇐⇒ f = MT with T ∈ SU(1, 1).
Remark 4.23. • We obtain a group homomorphism
SU(1, 1) → Mob(D) T 7→ MT
and hence also a group homomorphism SU(1, 1) → Aut(D).
• [Advanced] The group SU(1, 1) is not the Special Unitary Group SU(2). It is
actually the set of matrices T which preserve the quadratic form hz, wi = z1 w1 −
z2 w2 over C; so hT z, T wi = hz, wi for T ∈ SU(1, 1) and z, w ∈ C.
Proof. (⇒): Let MT : D → D be a Mob trans and consider the following picture:
f
H → H
MC MC
↓ ↓
M
D →T D

1 −i
where MC is the Cayley Map, so C = .
1 i
We have that f := MC −1 ◦ MT ◦ MC is a Möbius transformation from H to H. By Prop
−1
4.20 (H2H) f = MS with S ∈ SL2 (R), and by Lemma 4.4 we have S = C T C.
we have
a b
Let S = with ad − bc = 1 and a, b, c, d ∈ R. Then you can calculate that
c d

−1 1 (a + d) + i(b − c) (a − d) − i(c + b)
T = CSC = .
2 (a − d) + i(c + b) (a + d) − i(b − c)
Clearly this is of the required form. Finally, we have
det T = det(CSC −1 ) = (det C)−1 · det S · det C = det S = 1
as required.
(⇐): Conversely, if T ∈ SU(1, 1) then the same calculation in reverse shows that the
matrix S := C −1 T C is in SL2 (R). Thus MS : H → H is a Möbius transformation by
Prop 4.20 (H2H), and by Lemma 4.4 the map MT := MC ◦ MS ◦ MC−1 is a Möbius
transformation H → H.
We can actually say much more about what they look like:
Corollary 4.24 (D2D*).

1. Every Möbius transformation f from the unit disk D to itself can be written
z − z0
f (z) = eiθ ,
z0 z − 1
for some angle θ and z0 ∈ D the unique point such that f (z0 ) = 0.
2. All Möbius transformations of the unit disk to itself for which f (0) = 0 are rota-
tions about 0.
Remark 4.25. The map g(z) := zz−z 0
0 z−1
in the above swaps 0 and z0 and is an “involution"
(see Sheet 6); that is, g ◦ g = Id. Furthermore, the map z 7→ eiθ z is a rotation. So
all Möbius transformations of the unit disk are given by an involution followed by a
rotation.
Proof. 1. By Proposition D2D we have

 
az + b a z − − ab
f (z) = = −  ,
b̄z + ā ā − āb̄ z − 1
so z0 = −b/a. Moreover, since − āa = 1, we must have − aā = eiθ for some
θ ∈ (−π, π]. All that remains is to check that z0 ∈ D. Since |a|2 − |b|2 = 1, we
have
2
2 b |b|2 1
|z0 | − 1 = − −1 = −1 = − < 0,
a |a|2 |a|2
and so |z0 | < 1 as required.
2. We have

iθ −z0
f (0) = 0 ⇐⇒ e =0 ⇐⇒ z0 = 0 ⇐⇒ f (z) = −eiθ z,
−1
and so f is a rotation.
Finding automorphisms (example)

Find a Möbius transformation f from the closed unit disc onto the closed unit disc
taking 2i to 0 and −i to 1. Since 2i 7→ 0, by Corollary D2D* with z0 = 2i we have
!
i
−

iθ z 2 iθ 2z − i
f (z) = e =e
− 2i z − 1 −iz − 2
for some θ. Since −i 7→ 1, we have

iθ −2i − i
1=e ⇐⇒ 1 = eiθ i ⇐⇒ eiθ = −i.
−1 − 2
Thus

2z − i 2z − i 2z − i
f (z) = −i =i = (multiplying top and bottom by − i).
−iz − 2 iz + 2 z − 2i
Finding the image of geometric shapes under Möbius transformation (ex-

ample)
Let F be the geometric figure made up of a line segment from 0 to −1, a clockwise
circular arc (tracing out the unit circle) from −1 to i, then a line segment from i to 0.
Find the image of F under the Cayley Map MC (z) = z−i z+i .
Each section is a segment of a circle or a line, so must be taken to a segment of a
circle or a line by Proposition 4.13. First, let us check where the three ‘vertices’ go. We
have
−i −1 − i (1 + i)2 i−i
0 7→ = −1; −1 7→ = = i; i 7→ = 0.
i −1 + i 2 i+i
Let’s look at where the line segment from 0 to i goes:
i
i 2 −i −3i 1
7→ i
= = − ,
2 2 +i i 3
so the line segment from 0 to i must map to the circle/line from −1 to 0, passing through
− 13 ; that is, it is the line segment from −1 to 0 on the real axis.
Now, if we wanted we could do the same for the other two sections; pick a point on
each remaining line/circular arc and see where it maps to, thus determining whether
each image is a line segment or a circular arc. However, there is a much quicker method:
We know Mob trans are conformal and so preserve angles/orientation - so tracing round
the shape we must have the same angles in the image of the figure. So, travelling from
MC (0) = −1, when we reach MC (i) = 0 we must turn anticlockwise π/2 degrees and
head towards MC (−1) = i. Thus, the next section must simply be the straight line
from 0 to i. Similarly, the final section must be a circular arc from i back to −1. Thus,
the image is just F . Actually, letting D = {z ∈ D : π/2 < Arg(z) < π} be the interior
of F , one can check that D maps to D and and so MC −1 (and therefore MC ) is actually
in Aut(D).
4.6 Finding biholomorphic maps between domains

1. Find a biholomorphism from D := {z ∈ D : Im(z) < 0} to H. We build the map
in various stages from simpler known maps.
• Step 1: We know the Cayley transform MC is a map from H to D, so its

inverse MC −1 maps D to H. This is a good thing to start with as it has the
right image. To find the image of D under MC −1 first consider how it acts
on two segments of the boundary:
iz+i
– The line segment from −1 to 1 (through 0). We have MC −1 = −z+1 so
−i + i i+i 0+i
MC −1 (−1) = = 0, MC −1 (1) = = ∞, MC −1 (0) = = i.
1+1 −1 + 1 0+1
Thus, the line segment from −1 to 1 (through 0) is taken to the line
segment from 0 to ∞ (through i); so, the nonnegative imaginary axis.
[You could also have just checked that “−d/c” = −1 is on the line, so it
must be taken to another line.]
– The circular arc from −1 to 1 (through −i). We have
−i2 + i
MC −1 (−i) = = 1.
i+1
Thus, the circular arc from −1 to 1 (through −i) is taken to the line
segment from 0 to ∞ (through 1); so, the real axis. [Instead, we could
just have used conformality to deduce that this was the image - the angle
and its orientation at z = −1 must be preserved, so the positive real axis
had to be the image.]
Combined, this tells us that the image of D under MC −1 is the first quadrant
Ω = {w ∈ C : 0 < Arg(w) < π/2} - by conformality the interior must stay
on the ‘same side’ of each line segment. [You could instead explicitly check
what happens to an element in D, say z = −i/2.]
• Step 2: We now need a biholomorphic map from Ω to H. We already know
∼
of one from earlier: the map g : z 7→ z 2 : Ω −
→ H. Since compositions of
biholomorphic maps are biholomorphic, this gives us the map we want; let
f := g ◦ MC −1 then
∼
f :D − → H.
We can write down the map f explicitly by composing the formulae of the func-
tions:
iz + i 2

iz + i
f (z) = g ◦ MC −1 (z) = g = .
−z + 1 −z + 1
Figure 4.3: The sequence of transformations to obtain f .
2. Find the image of D\R≤0 := {z ∈ D : −π < Argz < π} under the map h(z) = z 1/2 .
Thus, find a biholomorphic map from D \ R≤0 to the unit disc D.
We have (using the principal branch)
√ √

iθ 1/2 iθ/2 θ π π
z = re (θ ∈ (−π, π), 0 < r < 1) ⇐⇒ z = re ∈ (− , ), 0 < r < 1 .
2 2 2
We know the complex power function is biholomorphic on C − R≤0 , so

∼
h : D \ R≤0 −
→ {z ∈ D : −π/2 < Argz < π/2}
maps D \ R≤0 biholomorphically to the (open) right half of the unit disc.
We also know rotations are biholomorphic on C and so r(z) = e−π/2 z = −iz maps
the right half of the unit disc biholomorphically to the lower half of the unit disc.
Using the previous example f from 1. we can now write down a biholomorphic
map from D := {z ∈ D : Im(z) < 0} to H, and then map H (via the Cayley Map)
∼ ∼
to D. Namely, since MC (z) = z−i → D we have f˜ : D \ R≤0 −
z+i : H − → D, where
2
i(−iz 1/2 )+i
−(−iz 1/2 )+1
−i (z 1/2 + i)2 − i(iz 1/2 + 1)2
f˜(z) : = (MC ◦ f ◦ r ◦ h)(z) = 2 = ,
i(−iz 1/2 )+i
+i (z 1/2 + i)2 + i(iz 1/2 + 1)2
−(−iz 1/2 )+1
which you can simplify further if you wish.
3. Translations are biholomorphic on C so for example

∼
f (z) = z − i : {w ∈ C : Im(w) − 1 > Re(w)} −
→ {w ∈ C : Im(w) > Re(w)}.
Since
π 1−i ∼
g(z) = e−i 4 z = √ z : {w ∈ C : Im(w) > Re(w)} −
→ H
2
Figure 4.4: The sequence of transformations to obtain f .
we have that f˜ := g ◦ f satisfies

∼
f˜(z) = {w ∈ C : Im(w) − 1 > Re(w)} −
→ H.

Explicitly, f˜(z) = 1−i
√
2
(z − i).
Chapter 5
Notions of Convergence in
complex analysis and power series
5.1 Pointwise and uniform convergence

Now we have good knowledge of basic continuous/holomorphic functions and their prop-
erties, we would like to be able to construct new continuous functions from them by
taking limits.
Given a sequence {fn }n∈N of functions fn on a subset of a metric space (say, on a
region in C), we want the limit function f (x) := limn→∞ fn (x) (if it exists) to behave
“nicely"; i.e., we want it to carry over useful properties of the functions fn . One
important case is that of an infinite series of functions
∞
X N
X
fn := lim fn ;
N →∞
n=0 n=0
∞
X
for example, a power series an (z − z0 )n in C. First we need a concrete notion of
n=0
what it means for a sequence of functions between two metric spaces to converge. Our
initial naive idea is the following version:
Definition 5.1 (Pointwise convergence). Let (X, dX ) and (Y, dY ) be two metric spaces.
A sequence of functions {fn }n∈N : X → Y converges pointwise (on X) to f if every
x ∈ X the limit function f (x) := limn→∞ fn (x) exists in Y . In other words, we have
∀ x ∈ X and ∀ > 0, there exists N ∈ N such that ∀ n > N, dY (fn (x), f (x)) < .
Note that N depends on x ∈ X.
64
CHAPTER 5. NOTIONS OF CONVERGENCE AND POWER SERIES 65
Remark 5.2. Note that for any given x ∈ X we have that fn (x) is just a sequence of
points in Y . This definition is precisely what we get by applying Definition 2.11 to the
sequence fn (x) at every point x ∈ X independently.
The problem with pointwise convergence (examples)

1. Consider the sequence of functions fn = xn on [0, 1]. It is easy to see that sequence
is pointwise convergent on [0, 1] with limit
(
0, if x < 1.
f (x) =
1, if x = 1.
This is a sequence of continuous, differentiable functions on a compact set, but
the limit function is not continuous!
2. We encounter this issue again in the complex plane. Let fn : C → C : z 7→ z n .
We split into three cases.
|z| < 1: Pick z ∈ D and > 0. Then, we can certainly find N ∈ N such that |z|N <
(for example, take any N > log / log |z|). Thus, for every n > N we have
|fn (z) − 0| = |z|n < |z|N < ,
and so limn→∞ fn (x) = 0 in D.
|z| = 1: When |z| = 1 the point z rotates around the unit circle ∂D by Arg(z) every
iteration. For any z 6= 1 this sequence clearly doesn’t converge, but for z = 1
we have limn→∞ fn (z) = limn→∞ 1 = 1.
|z| > 1: Here, the value |z|n is unbounded, so the limit does not exist.
To conclude, the sequence fn (z) is not pointwise convergent on C, however, notice
that it is pointwise convergent on D ∪ {1} with limit function
(
0, if z ∈ D.
f (z) =
1, if z = 1.
Again, this is not continuous.
3. Was the problem that the region where the function converges wasn’t open? From
the previous examples it seems like this could be the case (it was fine on the
interior of balls). No! For example, the sequence of continuous functions fn (x) =
arctan(nx) converges pointwise on all of R to the not-continuous function

π/2,
 if x > 0.
f (x) = 0, if x = 0.

−π/2, if x < 0.

Key: Pointwise convergence does not preserve continuity. We need a better notion.
Definition 5.3 (Uniform convergence). We say a sequence of functions {fn }n∈N : X →

Y converges uniformly (on X) to (the limit function) f if we have
∀ > 0, ∃N ∈ N such that ∀n > N d(fn (x), f (x)) < , ∀ x ∈ X.
Note that N here does not depend on the specific choice of x ∈ X - the same N works
for all of them!
Remark 5.4. Uniform convergence trivially implies pointwise convergence (to the same
limit).
The following is a generalization of a theorem from Analysis I.
Theorem 5.5. [Uniform limits of continuous functions are continuous] Let (X, dX )
and (Y, dY ) be two metric spaces and let {fn }n∈N : X → Y be a sequence of continuous
functions that converges uniformly to f on X. Then f is continuous on X.
Proof. We will show f is continuous at any given x0 ∈ X. Let > 0. Since fn → f

uniformly we have that

∃N ∈ N such that ∀n>N dY (f (x), fn (x)) < ∀ x ∈ X. (5.1)
3
We know each fn is continuous at x0 , therefore there exists δ > 0 such that

∀ x with dX (x, x0 ) < δ, we have dY (fn (x), fn (x0 )) < . (5.2)
3
Pick some fixed n > N . Then, for any x with dX (x, x0 ) < δ we have
(D3)
dY (f (x), f (x0 )) ≤ dY (f (x), fn (x)) + dY (fn (x), f (x0 ))
(D3)
≤ dY (f (x), fn (x)) + dY (fn (x), fn (x0 )) + dY (fn (x0 ), f (x0 ))
(5.1)
< + dY (fn (x), fn (x0 )) +
3 3
(5.2)
< + + = .
3 3 3
Thus f is continuous at x0 .
We would like to develop criteria to determine when sequences of complex valued

functions converge uniformly. Here is the first, if we already know the sequence converges
pointwise:
Lemma 5.6. [Test for uniform convergence] Let fn : X → C be a sequence of functions

converging pointwise to a limit function f .
1. If |fn (x) − f (x)| ≤ sn for every x ∈ X, where {sn }n∈N is some sequence in R>0
(independent of x) with limn→∞ sn = 0, then fn converge uniformly to f on X.
2. If there exists a sequence xn ∈ X such that |fn (xn ) − f (xn )| ≥ c for some positive
constant c, then fn does not converge uniformly to f on X.
Proof. Very easy.
1. Let > 0 and x ∈ X. Since sn → 0 there exists N ∈ N with sn < for n > N .
Thus, for n > N we have |fn (x) − f (x)| ≤ sn < .
2. The statement obviously implies the negation of uniform continuity; it states that
there exists (= c) such that for all sufficiently large n ∈ N there is a point
x (= xn ) ∈ X for which |fn (x) − f (x)| ≥ .
Checking for uniform convergence (example)

Consider the sequence of functions fn (z) = ez + n1 and gn (z) = ez + nz . It is easy to see
that both functions converge pointwise to the exponential function f (z) = ez : for every
fixed z ∈ C we have limn→∞ (ez + n1 ) = ez = limn→∞ (ez + nz ).
Notice that fn → f uniformly because for every z ∈ C we have

z 1 1
|fn (z) − f (z)| = e + − ez = ;
n n
so we may take sn = 1/n in Lemma 5.6 part 1..

However gn 6→ f uniformly - the giveaway is that in writing

z z |z|
|gn (z) − f (z)| = e + − ez = ;
n n
the difference depends on |z|, and |z| is unbounded in C. So, we simply notice that
upon taking zn = n that |gn (zn ) − f (zn )| = 1 and by Lemma 5.6 part 2. there is no
uniform convergence.
However, for ρ > 0 the sequence gn does converge uniformly in any ball Bρ (0) in
the complex plane. For then |gn (z) − f (z)| = |z|/n < sn := ρ/n.
As mentioned, we are mainly interested in the convergence of infinite series.

Theorem 5.7. [Weierstrass M-test] Let fn : X → C be a sequence of functions such

that
X∞
|fn (x)| ≤ Mn for all x ∈ X and Mn < ∞.
n=1
Then
∞
X
fn converges uniformly on X to some limit function f : X → C.
n=1
Proof. Excluded - see similar proof from Analysis I. Instead, we will prove a stronger
result later.
Determining the convergence of series via the M-test (example)

Show that
∞
X |2z|3n
32n n2
n=1
converges uniformly on D and let f (z) be its limit function. Is f (z) continuous on D?
Note that when |z| ≤ 1 we have
n
|2z|3n 23n 8 1 1
2n 2
≤ 2n 2 = 2
< 2.
3 n 3 n 9 n n
We know n∈N 1/n2 converges, so taking Mn = 1/n2 the Weierstrass M-test implies
P
the function converges uniformly to some limit function f . Furthermore, since every fn
was continuous, then so is f (by Theorem 5.5).
Have we found our ideal definition of convergence? Indeed, we record one other
crucial property of uniform convergence from Analysis I:
Theorem 5.8. Assume a sequence of functions fn : [a, b] → R converge uniformly on

an interval [a, b] to some function f , and that {fn } are all continuous. Then
Z c Z c
lim fn (x) dx = f (x) dx, for all c ∈ [a, b].
n→∞ a a
Remark 5.9. If the convergence is not uniform all bets are off! We will heavily use this
in term 2 for contour integrals.
Everything looks good, but there is a slight issue to think about.
The problem with uniform convergence (an important example)

Let us return to the sequence of functions fn (z) = z n . We saw that it converges
pointwise on the (open) unit disc D to the function f (z) = 0. Is this convergence
uniform?
Let’s see if we can find a sequence such that |fn (zn ) − f (zn )| = c for some c > 0.
For simplicity, let’s just try to find a sequence of positive real numbers: We want
c = |fn (zn ) − f (zn )| = |(zn )n − 0| = (zn )n ,
so simply take zn = c1/n . We need this sequence to be in D so let’s set c = 1/2. Then
with zn = (1/2)1/n ∈ D, by construction we have
1/n !n n/n
1 1 1
|fn (zn ) − f (zn )| = −0 = = .
2 2 2
It follows from Lemma 5.6 part 2. that the convergence is not uniform. BUT, notice
that the limit function f (z) = 0 is trivially continuous on all of D - our notion can’t
even conclude that the constant function is continuous for a very basic example - we
have in some sense been too restrictive!
Key: Uniform convergence is actually too restrictive in terms of preservation of conti-

nuity.
5.2 Locally uniform convergence

Definition 5.10 (Locally uniform convergence). Let {fn }n∈N be a sequence of functions
in a metric space X. We say {fn } converges locally uniformly (on X) to (the
limit function) f , if for every x ∈ X there exists an open set U ⊂ X containing x on
which {fn }n∈N converges uniformly to f .
Theorem 5.11. Let {fn }n∈N be a sequence of continuous functions, which converges
locally uniformly on X to a limit function f . Then f is continuous on X.
Proof. Locally uniform convergence gives for any x ∈ X that the sequence converges
uniformly on some open U containing x. Hence f is continuous on U by Theorem 5.5.
Hence f is continuous at x. Since x was arbitrary, f is continuous.
Remark 5.12. This is in many ways the “right" notion for convergence of a sequence of
functions. It was championed by Weierstrass. In term 2, we will see that the limit of a
locally uniform convergent sequence of holomorphic functions is again holomorphic!!
Locally uniform but not uniform convergence (example)

Again, consider fn (z) = z n . We know this sequence converges pointwise to f = 0 on
the (open) unit disc D, but not uniformly. However, the convergence is locally uniform.
For w ∈ D, we can find r < 1 such that w ∈ Br (0) (this will be our open set U ).
Then for all z ∈ Br (0), we have |z n | < rn and limn→∞ rn = 0. Hence by Lemma 5.6
part 1. (with sn := rn ), we have uniform convergence in Br (0).
Note the ‘counter example’ (1/2)1/n from above fails since limn→∞ (1/2)1/n = 1, so
at some point the sequence (1/2)1/n leaves the ball Br (0).
Remark 5.13. Proving locally uniform convergence on every ball Br (0) in D (for 0 <
r < 1) is enough to prove that the limit function f is continuous on all of D. In fact,
to show a limit function f is continuous on a set X, it turns out to be enough to prove
uniform convergence on all compact sets in X:
Theorem 5.14. [Local M-test] Let X be a metric space and let fn : X → C be a
sequence of continuous functions such thatP for any y ∈ X, there is an open U ⊂ X
∞
containing y Pand constants Mn > 0 with n=1 Mn < ∞ and |fn (x)| ≤ Mn for all
∞
x ∈ U . Then n=1 fn converges locally uniformly to a continuous function on X.
Proof. [Hard but worthwhile!] If we prove ∞
P
n=1 fn converges locally uniformly on X,
then it converges
P to a continuous function on X by Theorem 5.11. So we need to show
that if Fk = kn=1 fn , then the sequence Fk converges locally uniformly on X, i.e. we
want to prove:
There is a function F : X → C such that for any y ∈ X there is an open U with
y ∈ U ⊂ X and with Fn converging uniformly to F on U .
Step 1. The sequence is pointwise convergent. Given y ∈P X, the hypothesis implies
that there are constants Mn > 0 such that |fn (y)| ≤ Mn and ∞ n=1 Mn converges. Note
that
k
X k
X k
X
|Fk (y)| = fn (y) ≤ |fn (y)| ≤ Mn
n=1 n=1 n=1
Since as k → ∞, the RHS converges, it must be bounded, say by L, so for all k,

|Fk (y)| ≤ L and hence the sequence Fk (y) is bounded. Hence it is contained in some
large closed ball, which is compact by Heine-Borel, and hence there is a subsequence
Fkj (y) that converges to some value F (y). We want to prove the whole sequence also
converges to F (y).
Also note that for kj ≥ k
kj kj kj
X X X
|Fkj (y) − Fk (y)| = fn (y) ≤ |fn (y)| ≤ Mn .
n=k+1 n=k+1 n=k+1
Taking the limit as j → ∞, both the LHS and RHS converge, and we get
∞
X
|F (y) − Fk (y)| ≤ Mn .
n=k+1
Now taking the limit as k → ∞ of both sides, the RHS goes to zero (any sequence that
converges must have tails that go to 0) and we obtain
lim |F (y) − Fk (y)| = 0.

k→∞
So Fk (y) → F (y) as k → ∞. Repeating this for all y, we get that Fk → F pointwise on

X. (I.e. we didn’t need the subsequences.)
Step 2. Locally uniform convergence. Given y as in the statement we want to
prove, let U be the open set provided by hypothesis of the Theorem, and Mn be the
provided constants for this U .
By the same estimate as we used before, we have for all x in U (this is the key
point!) and for all ` > k
`
X
|F` (x) − Fk (x)| ≤ Mn .
n=k+1
P∞
Taking the limit
P∞as ` → ∞ we obtain |F (x) − F k (x)| ≤ n=k+1 Mn for all x ∈ U . Now
since limk→∞ n=k+1 Mn = 0, the test for uniform convergence (Lemma 5.6 part 1.)
tells us that Fk → F uniformly on U , as required.
5.3 Complex power series

A (complex) power series is an expression of the form
∞
X
an (z − c)n
n=0
with an and c complex numbers. Recall that from Analysis I we have:
Theorem 5.15. There are three cases:

∞
X
1. an (z − c)n converges only for z = c. (R = 0)
n=0
2. There exists R > 0 (radius of convergence) such that

∞
X
• an (z − c)n converges absolutely for |z − c| < R.
n=0
(We call BR (c) the disc of convergence.)

∞
X
• an (z − c)n diverges for |z − c| > R.
n=0
(Anything can happen on the circle |z − c| = R).

∞
X
3. an (z − c)n converges absolutely for all z ∈ C. (R = ∞)
n=0
Proof. See Analysis I.
Remark 5.16. [Recap of Ratio and Root tests] The radius of convergence is usually
determined using the ratio test or the root test (see Analysis I) - it is the real number
for which the “limit value in the test is 1”.
We have the following key result:
∞
X
Theorem 5.17. A power series an (z − c)n with radius of convergence 0 < R ≤ ∞
n=0
converges uniformly on any ball Br (c) with 0 < r < R. This implies the power series is
locally uniformly convergent on its disc of convergence.
Proof. The second statement follows from the first since every w ∈ BR (c) is contained
in some Br (c) with r < R.
So we will prove the convergence is uniform on Br (c) given r < R. Consider
the point z0 = c + r. We have |z0 − c| = r < R, so z0 ∈ BR (c) and the se-
∞
X ∞
X
ries an (z0 − c)n = an rn converges absolutely by Theorem 5.15; in other words
n=1 n=1
∞
X
|an | rn < ∞. So, with Mn = |an | rn it follows that |an (z−c)n | ≤ Mn for all z ∈ Br (c)
n=1
∞
X
and Mn < ∞. Thus, the conditions of the M-test (Theorem 5.7) are satisfied and
n=0
∞
X
the series an (z − c)n converges uniformly on Br (c).
n=1
Remark 5.18. Power series do not converge uniformly in the entire disc of convergence
BR (c). The good news is that this is not necessary because we only need locally uniform
convergence to conclude continuity!
Showing a power series converges (example)

zn
For example, consider the series ∞
P
n=0 n! . Applying the ratio test, we have
|z n+1 |/(n + 1)! |z|

L = lim = lim = 0 < 1,
n→∞ |z n |/n! n→∞ n + 1
for every z ∈ C; so R = ∞ and the series converges absolutely in C. In fact, by Theorem

5.17 it converges locally uniformly in the whole complex plane to a continuous function.
This series looks familiar, and actually agrees with the exponential function ez we’ve
defined. We’ll be able to prove this later in the course.
Differentiation/integration of power series

Now we have an idea of what it means for a power series to converge, and where it
converges, we would like to know when a power series represents a holomorphic function.
In other words, we want to know when we can differentiate (and integrate) power series.
We have the following generalization of a result from Analysis I for real power series.
Proposition 5.19. [Term by term differentiation or integration preserves the radius of
∞
X
convergence] Let an (z − c)n be a power series with radius of convergence 0 < R ≤ ∞.
n=0
Then the formal derivatives and anti-derivatives
∞ ∞
X X an
nan (z − c)n−1 and (z − c)n+1
n+1
n=1 n=0
have the same radius of convergence R.

Remark 5.20. By ‘formal’ we mean that the above series (our guess for the derivative
and anti-derivative) both define convergent series in the disc BR (c).
So, we know the series with the expected formulae for derivative and anti-derivative
both converge, but do they genuinely represent the derivative and anti-derivative of the
original series?
Theorem 5.21. [Power series can be differentiated term by term in their disc of conver-
∞
X
gence] Let an (z − c)n be a power series in C, with radius of convergence 0 < R ≤ ∞,
n=0
and let f : BR (c) → C be the resulting limit function. Then f is holomorphic on BR (c)
with
∞
0
X
f (z) = nan (z − c)n−1 (5.3)
n=1
for z ∈ BR (c).
Proof. For simplicity, assume c = 0 (the case of general c is essentially the same).
f (z) − f (w)
We wish to show for each w ∈ BR (c) that lim exists and agrees with
z→w z−w
the expression in (5.3). Since convergence of f (z) is absolute in the disc of convergence
(meaning we can reorder sums), we have
∞
X ∞
X
f (z) − f (w) = an (z n − wn ) = an (z − w) qn (z),
n=1 n=1
n−1
X
where qn (z) = wk z n−1−k . So, for z 6= w
k=0
∞
f (z) − f (w) X
= an qn (z) =: h(z).
z−w
n=1
Note the series makes sense even at w, so we view h as being defined there too.
We claim that the series defining h(z) converges to a continuous function on BR (0).
We’ll prove this using the local M-test. Given z0 ∈ BR (0), choose r < R such that
w, z0 ∈ Br (0). We need constants Mn for this set Br (0) that control the terms an qn (z)
defining h. We have for z ∈ Br (0)
n−1
X n−1
X
k n−1−k
| an qn (z)| = an w z ≤ |an | |w|k |z|n−1−k
k=0 k=0
n−1
X
< |an | rk rn−1−k = n |an | rn−1 =: Mn .
k=0
P∞ P∞
We have n=1 Mn = n=0 n|an |rn−1 which converges, since by Proposition 5.19 the
∞
X
series nan z n−1 has radius of convergence R, and so converges absolutely on BR (0),
n=1
in particular at the point r. It follows from the Local M-test that the series defining h
converges locally uniformly to a continuous function on BR (0). Hence
∞ ∞ n−1 ∞
f (z) − f (w) X X X X
lim = lim h(z) = h(w) = an qn (w) = an wk wn−1−k = n an wn−1
z→w z−w z→w
n=1 n=1 k=0 n=1
as required.
Corollary 5.22. A power series f as in Theorem 5.21 with positivePradius of conver-

gence R can be differentiated infinitely many times and f (k) (z) := ∞ n
n=k k! k an (z −
c)n−k for z ∈ BR (c). This implies f (k) (c) = k! ak .
Corollary 5.23 (Power series can be integrated term by term in their disc of conver-
gence). A power series f as in Theorem 5.21 with positive radius of convergence has
a holomorphic
P∞ antiderivative F : BR (c) → C, that is, F 0 (z) = f (z), and F is given by
an n+1
F (z) := n=0 n+1 (z − c) for z ∈ BR (c).
Convergence of familiar power series (examples)

1. The expected power series for sin(z), cos(z), sinh(z) and cosh(z) all converge lo-
cally uniformly to continuous functions on C, and the derivatives/anti-derivatives
match those expected. We will see later that the series genuinely do represent the
functions in question.
More generally, when can a holomorphic function be represented via a power
series? What about log(z)? It is only defined (and is holomorphic) on C \ R≤0
(and is not continuous if extended to the negative real line), so certainly cannot
be defined by a power series on D, say. Maybe it works if we change the point
at which the series is expanded about (i.e., change the constant c from zero).
∞
X zn
Indeed, one can show that (−1)n+1 converges on D. We expect this to
n
n=1
be the power series converging to log(1 + z), but what was so special about the
taking a power series about the point c = 1? It turns out, nothing. We will
find out that the holomorphic functions are precisely the functions that can be
locally represented by power series at every point in their domain. As a result,
we will see by Theorem 5.21 that every holomorphic function is infinitely many
times complex differentiable!
∞
X
2. Consider the geometric series z n . By the ratio test this sum converges when
n=1
|z| < 1, so the radius of convergence is R = 1 and the series converges to a
continuous function on its disc of convergence D. To find the continuous function
the series converges to we need to consider the pointwise limit. Notice that
∞ N
!
1 − z N +1

X
n
X
n 1
z = lim z = lim = , (for |z| < 1).
N →∞ N →∞ 1−z 1−z
n=1 n=1
This limit function is defined and is continuous on all of C \ {1}. So, in some sense
the convergence of the series in the complex plane is limited to the unit disc D
because it can’t pass the ‘pole’ at z = 1.
3. We may determine the convergence of new power series by substitution. For

example, notice that |z| < 1 ⇐⇒ |z 2 | < 1 ⇐⇒ | − z 2 | < 1, so by the
∞
X
substitution z 7→ z2 we have z 2n converges locally uniformly on D to 1
1−z 2
,
n=1
∞
X
and similarly, by the substitution z 7→ −z 2 we have (−1)n z 2n converges locally
n=1
1
uniformly to 1+z 2
.
These examples give us real insight into the reasons for convergence of the corre-
sponding real power series. Consider the graph of the real function y = 1/(1−x2 ).
It appears obvious why its real interval of convergence is the unit interval - there
are asymptotes at x = ±1 that we can’t ‘get past’ continuously. But on the other
1
hand y = 1+x 2 is a nice smooth looking graph everywhere on the real line - so
why on earth is its interval of convergence also restricted to the unit interval?
We can now see the answer - the interval of convergence of the real power series
is restricted by the disc of convergence of the corresponding complex power se-
ries!!! The issue being the poles in the complex plane at z = ±i that we couldn’t
see when considering only the real version of the function. So hidden inside the
real power series of nice continuous real function is actually some meaningful and
significant complex analysis.
Key: Complex analysis can give us new information about real functions!
Chapter 6
Complex integration over

contours
6.1 Definition of contour integrals

Our first aim is to give a meaning to the following
Z
f (z) dz,
γ
where f : U → C is a complex function on an open set U , and γ is a curve in U . This

will be called the integral of f along the curve γ.
Our first step towards the above is to start by considering complex valued functions
of a real variable. That is,
f : [a, b] → C
where [a, b] ⊂ R. We note that such a function can be written as
f = u + iv, ∀t ∈ [a, b],

where u, v : [a, b] → R and given by u = Re(f ) and v = Im(f ). Moreover we note that
f is continuous if and only if both u and v are continuous. We then define
Z b Z b Z b
f (t) dt := u(t) dt + i v(t) dt ∈ C.
a a a
R R
b Rb b Rb
We note that Re a f (t) dt = a u(t) dt and Im a f (t) dt = a v(t) dt.
77
CHAPTER 6. COMPLEX INTEGRATION OVER CONTOURS 78
Example 6.1. We consider the function f (t) = t + it. That is u(t) = v(t) = t. We
compute
Z 1 Z 1 Z 1 Z 1
f (t) dt = (t + it) dt = t dt + i t dt =
0 0 0 0
1 1
t2 t2

1 1
= +i = +i .
2 0 2 0 2 2
Before going on we check our definition of integrating complex valued functions of

a real parameter is complex linear (as it should be!).
Lemma 6.2.
Rb
1. Let f1 and f2 be continuous functions from [a, b] to C. Then a (f1 (t) + f2 (t))dt =
Rb Rb
a f1 (t)dt + a f2 (t)dt.
Rb
2. For any complex number c ∈ C, and continuous function f : [a, b] → C, a cf (t)dt =
Rb
c a f (t)dt.
Proof. (Simple, included for completeness). For part 1, write fj = uj + ivj where uj
and vj are the real and imaginary parts of fj . Then using the definition, and the known
linearity of real integrals,
Z b Z b
(f1 (t) + f2 (t))dt = (u1 (t) + iv1 (t) + u2 (t) + iv2 (t))dt
a a
Z b
= (u1 (t) + u2 (t) + i(v1 (t) + v2 (t))dt
a
Z b Z b
= (u1 (t) + u2 (t))dt + i (v1 (t) + v2 (t))dt
a a
Z b Z b Z b Z b
= u1 (t)dt + u2 (t)dt + i v1 (t)dt + v2 (t)dt
a a a a
Z b Z b Z b Z b
= u1 (t)dt + i v1 (t)dt + u2 (t)dt + i v2 (t)dt
a a a a
Z b Z b
= f1 (t)dt + f2 (t)dt.
a a
For part 2 we write f = u + iv where u and v are real, and write c = x + iy. Then
using linearity of real integrals as before,
Z b Z b
cf (t)dt = (x + iy)(u(t) + iv(t))dt
a a
Z b
= (xu(t) − yv(t) + i(yu(t) + xv(t)))dt
a
Z b Z b
= (xu(t) − yv(t))dt + i (yu(t) + xv(t)))dt
a a
Z b Z b Z b Z b
=x u(t)dt − y v(t)dt + i y u(t)dt + x v(t)dt
a a a a
Z b Z b
= (x + iy) u(t)dt + i v(t)dt
a a
Z b
=c f (t)dt.
a
Recall from Definition 3.8 that a smooth curve in C is a continuously differentiable

function γ : [0, 1] → C. In fact we can be more general and consider curves γ : [a, b] → C.
For short, we will say that such curves are C 1 .
Remark 6.3. Let us write γ(t) = u(t) + iv(t) with u, v : [a, b] → R. Then the derivative
γ 0 is defined as
γ 0 (t) := u0 (t) + iv 0 (t).
One needs to be careful for the meaning of the condition on smoothness at the end-
points. That is, there we demand that the one-sided derivative exists and is continuous
from the one side. That is, γ 0 (b) := limh→0− u(b+h)−u(b)
h + i limh→0− v(b+h)−v(b)
h exists
0 0
and limt→b− γ (t) = γ (b).
Examples.
1. An example of a C 1 curve is
γ : [0, 2π] → C, γ(θ) = reiθ ,
where r > 0. (Here we use the letter θ to denote the parameter since it is more
often used to denote angles.) We note that this is nothing else than the circle
with center the origin and radius r. Moreover we note that as the parameter θ
runs from 0 to 2π we run the curve on a anti-clockwise direction. It is easy to see
that this is C 1 since
γ 0 (θ) = (reiθ )0 = (r cos0 (θ)) + ir sin0 (θ)) = rieiθ .

2. Another example is given by
γ : [0, 2π] → C, γ(θ) = re−iθ ,
where r > 0. We note the graph of this curve is exactly the same as above, namely
the circle centred at the origin and radius r. However here we run the curve on
the clockwise direction!
3. We now define the curve

γ : [0, 2] → C,
(
t + it, 0 ≤ t ≤ 1
γ(t) :=
t + i 1 ≤ t ≤ 2.
It is easy to see that γ consists of a straight line connecting the origin and the
point 1 + i followed by another line connecting the point 1 + i and the point 2 + i.
Note that this curve is not C 1 since the derivative at the point t = 1 does not
exist (from the left it is 1 + i and from the right it is 1).
We are now ready to give the following definition.
Definition 6.4. Let U ⊂ C be an open set, and let f : U → C be a continuous function.

Let γ : [a, b] → U ⊂ C be a C 1 -curve. Then we define the integral of f along the curve
γ by
Z Z b
f (z) dz := f (γ(t))γ 0 (t) dt.
γ a
Remark 6.5. Note the function f is of a complex variable z. Moreover f (γ(t)) makes
sense since γ([a, b]) ⊂ U . Furthermore the condition that γ is C 1 guarantees the exis-
tence of γ 0 (t). Finally the function g(t) := f (γ(t))γ 0 (t) is a function of the real variable
Rb
t, so we have already defined a g(t) dt above.
Basic Properties. We now note the following basic properties. Assume f is continu-
ous.
1. We have Z Z Z
(f1 (z) + f2 (z)) dz = f1 (z) dz + f2 (z) dz
γ γ γ
Indeed we have by using Lemma 6.2,

Z b Z b Z b
0 0
(f1 (γ(t)) + f2 (γ(t)))γ (t) dt = f1 (γ(t))γ (t) dt + f2 (γ(t))γ 0 (t) dt
a Za Z a
= f1 (z) dz + f2 (z) dz.

γ γ
2. For a c ∈ C we have Z Z
cf (z) dz = c f (z) dz.
γ γ
Indeed,
Z Z b Z b Z
cf (z) dz := cf (γ(t))γ 0 (t) dt = c f (γ(t))γ 0 (t) dt = c f (z) dz
γ a a γ
where we used Lemma 6.2 for the middle equality.
3. Given γ : [a, b] → C we define the curve (−γ) : [−b, −a] → C, by (−γ)(t) := γ(−t).
Then we have Z Z
f (z) dz = − f (z) dz.
−γ γ
See Problem Sheet 9, Question 2.
Example 6.6.
dz, where γ(θ) = reiθ with 0 ≤ θ ≤ 2π, with r > 0.

R
1. We compute the integral γ
Z Z 2π Z 2π Z 2π
iθ 0 iθ
dz = (re ) dθ = rie dθ = ri eiθ dθ =
γ 0 0 0
Z 2π Z 2π
ri cos(θ) dθ + i sin(θ) dθ = ri(0 + 0) = 0.
0 0
2. We consider the curve
γ : [−π/2, π/2] → C, γ(θ) = 2eiθ

R
and we compute the integral γ z̄ dz. We have
Z Z π/2 Z π/2 Z π/2
z̄ dz = (2eiθ )(2eiθ )0 dθ = 2e −iθ iθ
2ie dθ = 4i dθ = 4πi.
γ −π/2 −π/2 −π/2
In the last example we may parametrise the same curve differently. For example we
may consider δ : [0, 1] → C where δ(t) = 2eπi(t−1/2) . It is easily seen that both γ
and δ parametrise the very same curve, namely the semi-circle joining −2iRand 2i and
lying entirely at the positive part of the x-axis. We may then ask whether γ f (z) dz =
R
δ f (z) dz. It turns out that this is indeed the case, as the following lemma shows.
Lemma 6.7 (Reparametrisation of curves). Let U ⊂ C be an open set, f : U → C be
continuous, and let γ : [a, b] → U be a C 1 curve. If φ : [a0 , b0 ] → [a, b] with φ(a0 ) = a
and φ(b0 ) = b is continuously differentiable and we define
δ : [a0 , b0 ] → C, δ := γ ◦ φ (composition),
then we have Z Z
f (z) dz = f (z) dz.
γ δ
Proof. We have
Z Z b0
f (z) dz = f (δ(t))δ 0 (t) dt
δ a0
Z b0
= f (γ(φ(t)))(γ(φ(t))0 dt
a0
Z b0
= f (γ(φ(t)))γ 0 (φ(t))φ0 (t) dt
a0
We change the variable s := φ(t) and get that ds = φ0 (t)dt, and we use the fact that
φ(a0 ) = a and φ(b0 ) = b. That is, the integral above is equal to
Z b Z
f (γ(s))γ 0 (s) ds = f (z) dz.
a γ
In the last example above we have, with φ(t) = π(t − 21 ) that δ = γ ◦ φ. The next
step is to consider more general curves, and not only C 1 curves.
Definition 6.8 (Contours). Let γ : [a, b] → C be a curve, and suppose that there exist
a = a0 < a1 < a2 < . . . < an−1 < an = b such that the curves γi : [ai−1 , ai ] → C,
i = 1, 2, . . . , n defined by γi (t) := γ(t) for t ∈ [ai−1 , ai ] are C 1 curves. Then we say that
γ is a piecewise C 1 -curve, or contour .
For a contour γ as above, we then define
Z n Z
X
γ i=1 γi
This type of integral is called a contour integral. It is the main object of study of
this chapter and plays a cental role in Complex Analysis.
Remark 6.9. [Technical] Here one should check that the above is well-defined. That is,
we may find a different a = a00 < a01 < a02 < . . . < a0m−1 < a0m = b, and γj0 to establish
that γ is a contour. But then it is easy to show that
n Z
X m Z
X
i=1 γi j=1 γj0
We will need one more definition regarding curves. We will need to add them. That
is, if γ : [a, b] → C and δ : [c, d] → C are two contours with γ(b) = δ(c) then we define
the contour γ ∪ δ : [a, b + d − c] → C
(
γ(t), a ≤ t ≤ b
(γ ∪ δ)(t) :=
δ(t + c − b), b ≤ t ≤ b + d − c.
Then it is easy to see that

Z Z Z
f (z) dz = f (z) dz + f (z) dz.
γ∪δ γ δ
6.2 The Fundamental Theorem of Calculus

The Fundamental Theorem of Calculus in Analysis I tells us that we can calculate
real integrals by finding an antiderivative of the quantity being integrated. There is a
version of this theorem for contour integrals and, as you might expect, it is very useful
in practice.
Theorem 6.10 (Complex Fundamental Theorem of Calculus (FTC)). Let U ⊂

C be an open set and let F : U → C be holomorphic with continuous derivative f . Then
for any contour γ : [a, b] → U we have
Z
f (z) dz = F (γ(b)) − F (γ(a)).
γ
In particular if γ is closed, that is γ(a) = γ(b), then we have that

Z
f (z) dz = 0.
γ
Proof. First we consider the case of γ being a C 1 curve. Write F = u + iv. We have
Z Z b
F 0 (z) dz = F 0 (γ(t))γ 0 (t) dt
γ a
Z b Z b Z b
0 0
= (F (γ(t))) dt = (u(γ(t))) dt + i (v(γ(t)))0 dt
a a a
= [u(γ(t))]ba + i [v(γ(t))]ba = F (γ(b)) − F (γ(a)).
where here we note that (F (γ(t)))0 = dF (γ(t))

dt and for the second to last equality we
used the Fundamental Theorem of Calculus from real analysis.
We can now extend the proof to any contour. Indeed all we need to observe is that
(with notation as above)
Z n Z
X n
X
0
F (z) dz = 0
F (z) dz = [F (γ(t))]aaii−1 = F (γ(an ))−F (γ(a0 )) = F (γ(b))−F (γ(a)).
γ i=1 γi i=1
Remark 6.11.
• [Looking ahead] Later we will prove that if a function F is holomorphic, then

its derivative f is automatically continuous. So that assumption can be dropped
from the FTC.
• Note that Theorem 6.10 states that if two curves γ1 : [a, b] → C and γ2 : [a0 , b0 ] → C
have the same endpoints, that is γ1 (a) = γ2 (a0 ) and γ1 (b) = γ2 (b0 ) then
Z Z
f (z) dz = f (z) dz
γ1 γ2
under the same hypothesis on F as in Theorem 6.10.
Example 6.12. (Important) We now compute the integral

Z Z
z n dz := z n dz, n ∈ Z
|z|=r γ
where γ(θ) = reiθ , 0 ≤ θ ≤ 2π. We first consider the case n 6= −1. Then we have
0 r
z n+1 z n+1
Z Z Z
n n
z dz = z dz = dz = = 0,
|z|=r γ γ n+1 n+1 r
n+1
that is we apply the theorem with F (z) = zn+1 and take U = C \ {0}, which clearly
contains the curve γ.
However we cannot do the same with n = −1. Indeed, a natural candidate to
consider as F (z) would be the logarithm. But we have to make sure that the open set
U where this is defined does include the curve γ. For example the principal branch of
the logarithm will not do, since we remove the negative real axis, and the contour γ
goes through it. Actually if we turn to the actual definition of the integral we see that
Z Z 2π Z 2π
1 −1 −iθ iθ
dz = r e rie dθ = i dθ = 2πi 6= 0.
γ z 0 0
Note that the integral is not zero. That is, we can conclude that there is no open set
U that contains γ such that log(z) is well-defined, since if it were so, by the theorem
above we would get that the integral had to be zero, which is not!
Our next step is to address the following question:
Problem 6.13. Given a function f : U → C, provide sufficient conditions such that

there exists a holomorphic function F : U → C, with derivative F 0 = f . Such an F is
usually called the antiderivative or primitive of f . In other words, is there a converse
to the second part of the Fundamental Theorem of Calculus?
For this we need to be able to estimate the modulus of a contour integral, that is,
R
bound γ f (z) dz from above. We start with a definition.
Definition 6.14 (Length of a contour). Let γ : [a, b] → C be a contour. We define the

length of γ by
Z b
L(γ) := |γ 0 (t)| dt.
a
We now prove
Lemma 6.15 (The Estimation Lemma). Let f : U → C be continuous and γ : [a, b] → U

be a contour. Then Z
f (z) dz ≤ L(γ) sup |f |,
γ γ
where
sup |f | := sup{|f (z)| : z ∈ γ}.
γ
Proof. We start by proving the following: Given a continuous function g : [a, b] → C,

we have that Z b Z b
g(t) dt ≤ |g(t)| dt. (6.1)
a a
Rb
Indeed if we write a g(t) dt = reiθ , with r ≥ 0 then we have that
Z b Z b
−iθ
g(t) dt = r = Re(r) = Re e g(t) dt
a a
Z b Z b
−iθ
= Re e g(t) dt = Re e−iθ g(t) dt
a a
Z b Z b
−iθ
≤ e g(t) dt = |g(t)| dt
a a
Using (6.1) we can now conclude the proof of the lemma. Indeed,
Z Z b Z b
f (z) dz = f (γ(t))γ 0 (t) dt ≤ f (γ(t))γ 0 (t) dt,
γ a a
where we used the above observation for g(t) = f (γ(t))γ 0 (t). But then we have that
Z b Z b
0
f (γ(t))γ (t) dt ≤ sup |f | γ 0 (t) dt = sup |f |L(γ).
a γ a γ
Example 6.16. We consider γ : [0, π/2] → C given by γ(θ) = 2eiθ . Then we get an
upper bound for Z
z+4
3
dz
γ z −1
R π/2
We have L(γ) = 0 |(2eiθ )0 | dθ = π, and for z ∈ γ, that is |z| = 2 we have
z+4 |z + 4| |z| + 4 6
3
= 3 ≤ 3
= .
z −1 |z − 1| ||z| − 1| 7
Here we used the triangle inequality for the numerator and the reverse triangle inequality
for the denominator. Then, by Proposition 6.15,
Z
z+4 6π
3
dz ≤ .
γ z −1 7
We are now ready to answer the problem stated above. The next theorem gives a
converse to the second part of the FTC.
RTheorem 6.17 (Converse to FTC). Let f : D → C be continuous on a domain D. If

γ f (z) dz = 0 for all closed contours γ in D, then there exists a holomorphic F : D → C
such that
F 0 (z) = f (z).
Proof. We fix a point a0 ∈ D. Then for any other point w ∈ D, by definition of D being
a domain there is always a smooth path connecting a0 and w and we write γw for such
a path. Then we define the following function
Z
F (w) := f (z) dz.
γw
Step 1. F (w) doesn’t depend on the contour we use. We first check that F (w) does
not depend on the choice of contour connecting a0 and w. To see this, if γ̃(w) is another
contour from a0 to w, then we can consider the closed contour C = γ(w) ∪ (−γ̃(w))
obtained by following γ(w) then γ̃(w) in the reverse direction. Then
Z Z Z Z
0= f (z) dz = f (z) dz = f (z) dz − f (z) dz
C γ(w)∪(−γ̃(w)) γ(w) γ̃(w)
where the first equality is from the hypothesis that integrals of f over closed contours
are zero. This gives
Z Z
f (z) dz = f (z) dz
γ(w) γ̃(w)
showing that F (w) does not depend on the choice of γ(w).

Step 2. F is holomorphic with derivative f . We now claim that F is holomorphic
and moreover its derivative is f , that is, for every w ∈ D we have that F 0 (w) exists and
F 0 (w) = f (w), i.e.,
F (w + h) − F (w)
lim = f (w) w ∈ D.
h→0 h
For any given w ∈ D we fix an r > 0 such that Br (w) ⊂ D. This ball exists since D
is open. Then for any h ∈ C with |h| < r we consider the straight line δh that connects
the point w to w + h. A parametrisation of such a line is given by
δh : [0, 1] → D, t 7→ w + th.
We now consider the contour γw ∪ δh . This is a contour from a0 to w + h, contained in

D, so
Z Z Z Z
F (w + h) = f (z) dz = f (z) dz + f (z) dz = F (w) + f (z) dz.
γw ∪δh γw δh δh
R
It is easily seen that δh f (w) dz = hf (w) and hence we may rewrite the above equation
as Z
F (w + h) = F (w) + hf (w) + (f (z) − f (w)) dz,
δh
or for h 6= 0
F (w + h) − F (w)
Z
1
− f (w) = (f (z) − f (w)) dz.
h h δh
That is,
F (w + h) − F (w)
Z Z
1 1
− f (w) = (f (z) − f (w)) dz = (f (z) − f (w)) dz
h h δh |h| δh
and by Lemma 6.15 (note here it is essential that f is continuous) we have that the
above quantity is bounded by
1
≤ L(δh ) sup |f (z) − f (w)|.
|h| z∈δh
But a simple calculation shows that L(δh ) = |h| and hence we have that
F (w + h) − F (w)
− f (w) ≤ sup |f (z) − f (w)|.
h z∈δh
Now we take the limit h → 0, and we observe that the curve δh becomes just the point
w. (More rigorously, every z ∈ δh has |z − w| ≤ h.) Since f is continuous, we get that
sup z∈δh |f (z) − f (w)| → 0 as h → 0. Hence we have that F (w+h)−F
h
(w)
− f (w) → 0 as
h → 0. That is,
F (w + h) − F (w)
lim = f (w).
h→0 h
Since w was any point in D we have established the theorem.
Remark 6.18. The above theorem established the existence of a holomorphic anti-
derivative of f . However there may be more than one such F , after all we make quite a
few choices in the construction of F above. For example we could have picked a different
a0 . However if there is another F̃ such that F̃ 0 (w) = f (w) for all w ∈ D, then we would
have that F 0 (w) = F̃ 0 (w) or equivalently (F − F̃ )0 (w) = 0. But we know from Theorem
3.12 that if a function defined over a domain has a zero derivative we have that the
function is just a constant. That is, there is some c ∈ C such that F̃ = F + c.
6.3 First version of Cauchy’s Theorem

In the preceding section (see Important Example 6.12) we encountered the curious fact
that if n 6= −1, the contour integral of f (z) = z n around the unit circle is 0. Perhaps
when we integrate a holomorphic function over a closed contour we always get 0?
No: We’ve also seen in the previous section that despite the function f (z) = z1 being
holomorphic on C∗ = C − {0} the contour integral of f around the unit circle, going
anticlockwise, gives 2πi, which is not 0. As we shall later discover, the problem here is
that C∗ has a hole inside the contour at z = 0, and f does not extend to a holomorphic
function on C (i.e., ‘across the hole’).
The main goal of the current section to prove a version of Cauchy’s Theorem.
Informally, this says
‘If f is a holomorphic
R function on a domain with no holes, and γ is a closed contour
in the domain, then γ f (z) dz = 0’.
This theorem will be stated precisely later and we will not prove it in full generality
here. Instead we will make a compromise and prove it now for a certain nice class of
domain; namely, so-called starlike domains.
Definition 6.19. A domain D is called starlike if there exists a point a0 ∈ D such
that for any other point b ∈ D the straight line connecting a0 to b lies entirely in D.
Examples: An example of a starlike domain is trivially the whole of C, where we could
take as a0 any point, for example the origin. Another one is Br (a), with a0 for example
equal to the centre a of the ball.
Non-Examples: A non-example of starlike domain is C∗ , or Br (a) − {a}, the open

ball with the centre removed.
Here is the form of Cauchy’s Theorem we will prove in full detail:

Theorem 6.20. [Cauchy’s Theorem for starlike domains]
Let D be a starlike domain, and let f : D → C be holomorphic. Then if γ ⊂ D is any
closed contour we have Z
f (z) dz = 0.
γ
The same conclusion holds if f is holomorphic on D − S, where S is a finite set of
points, and f is continuous on D.
Remark 6.21. [Technical] The final part of the statement, concerning the case that f is
holomorphic outside a finite set S, will be needed in certain proofs later on. However,
it is not the main point one should take away.
Example 6.22. We have that

ez sin2 (z)
Z
dz = 0,
|z|= 12 ez 2
z 2
since the function f (z) = e sinz2 (z) is holomorphic in C (a starlike domain) - it is the
e
composition/product/quotient of holomorphic functions - and |z| = 12 is a closed curve.
Strategy of proof of Cauchy’s Theorem:
Step A. First, we will prove the following result that tells us what happens when we
integrate holomorphic functions over boundaries of triangles:
Lemma 6.23. Let U be an open set, f : U → C a holomorphic function, and let
∆ ⊂ U be any triangle in U . Then
Z
f (z) dz = 0.
∂∆
The same conclusion holds if f is holomorphic on U −S where S is a finite set of points,

and f is continuous on U .
Remark 6.24. Here ∂∆ is the boundary of the triangle. That is, if we name the vertices
of the triangle ∆ by A, B and C then we define ∂∆ to be the contour AB + BC + CA.
Remark 6.25. If we are not given a parametrization of a closed contour, we always go
anti-clockwise around it (if the contour does not cross itself then it is intuitively clear
what this means).
Step B. Then we will prove
Lemma
Z 6.26. Let D be a starlike domain and f : D → C be a continuous function.
If f (z) dz = 0 for all triangles ∆ in D then there exists a holomorphic function
∂∆
F : D → C such that F 0 (z) = f (z) for all z ∈ D.
Note this looks very much like the Converse to the Fundamental Theorem of Cal-
culus (Theorem 6.17), but with slightly different hypotheses (we only need to integrate
over triangles, but we need the additional condition that the domain is starlike).
Step C. The two preceding lemmas are then combined as follows to establish Cauchy’s
Theorem (Theorem 6.20).
Proof of Cauchy’s Theorem. Let f be a continuous on the domain D and assume that
f is holomorphic on D, except at a finite set of points. Then by Lemma 6.23
Z
f (z) dz = 0
∂∆
for any triangle ∆ ⊂ D. But then by Lemma 6.26 there exists an anti-derivative
F : D → C such that F 0 (z) = f (z). So by the FTC (Theorem 6.10), using that f is
continuous, we have that
Z Z
f (z) dz = F 0 (z) dz = 0,
γ γ
for any closed contour, as required.
We now return to Steps A & B to prove the respective lemmas.
Proof of Lemma 6.23 when S is empty. [Challenging but fun!] Our assumptions are
that we have a holomorphic function f on an open set U , and a triangle ∆ ⊂ U .
We start by dividing the triangle ∆ into 4 smaller triangles ∆(1) , ∆(2) ,∆(3) ,∆(4) by
considering the straight lines that connect the mid points of the sides of the triangle ∆.
Then it is easy to see that we obtain
Z 4 Z
X
f (z) dz = f (z) dz
∂∆ i=1 ∂∆(i)
since the inner sides of the new triangles will be traversed twice in opposite direction and
hence cancel each other. By taking the absolute value and using the triangle inequality
we obtain
Z X 4 Z
f (z) dz ≤ f (z) dz .
∂∆ i=1 ∂∆(i)
By setting j ∈ {1, 2, 3, 4} Rto be the index associated

R with the maximal integral (that
is, the index j such that ∂∆(j) f (z) dz ≥ ∂∆(i) f (z) dz for all i = 1, 2, 3, 4) we may
rewrite the above inequality as
Z Z
f (z) dz ≤ 4 f (z) dz .
∂∆ ∂∆(j)
Moreover by construction of the triangles we have that L(∂∆(j) ) = 12 L(∂∆). We now

set ∆1 := ∆(j) and repeat this process. In this way we obtain a sequence of triangles
∆ ⊃ ∆1 ⊃ ∆2 ⊃ . . . ⊃ ∆n ⊃ . . .
where we have that Z Z

n
f (z) dz ≤ 4 f (z) dz ,
∂∆ ∂∆n
and
1
L(∂∆n ) = L(∂∆).
2n
In particular we have that L(∂∆n ) → 0, as n → ∞. That is, the triangle

T ∆n gets smaller
and smaller, and eventually in the limit is just a point.1 That is, n∈N ∆n = {w} for
some point w ∈ U .
R R
We now note that ∂∆n 1 dz = ∂∆n z dz = 0 since the integrals are over closed
contours and the integrand functions have anti-derivatives. Hence we have that
Z Z
f (z) dz = f (z) − f (w) − (z − w)f 0 (w) dz,
∂∆n ∂∆n
since the integral of the terms that are linear in z give zero (because ω, f (ω) and f 0 (ω)
are just constants).
We then consider the function
f (z) − f (w)
g(z) := − f 0 (w), z ∈ U \ {w}
z−w
and we set g(w) = 0. Then we see that g is a continuous function on U (since f is
holomorphic at w). In particular we may rewrite the above equation as
Z Z Z
0
f (z) dz = f (z) − f (w) − (z − w)f (w) dz = (z − w)g(z) .
∂∆n ∂∆n ∂∆n
Therefore, we have
Z Z
n
f (z) dz ≤ 4 (z − w)g(z) dz .
∂∆ ∂∆n
Using the Estimation Lemma (Lemma 6.15) we obtain that

Z
f (z) dz ≤ 4n L(∂∆n ) sup |z − w||g(z)| = 2n L(∂∆) sup |z − w||g(z)|
∂∆ z∈∂∆n z∈∂∆n
1
But it is easy to see that supz∈∂∆n |z − w| ≤ L(∂∆n ) = 2n L(∂∆), and so
1
sup |z − w||g(z)| ≤ L(∂∆) sup |g(z)|.
z∈∂∆n 2n z∈∂∆n
All together gives Z

f (z) dz ≤ L(∂∆))2 sup |g(z)|.
∂∆ z∈∂∆n
Taking the limit n → ∞ (note the left hand side is independent of n!) we have that
supz∈∂∆n |g(z)| → |g(w)| = 0. That is,
Z
f (z) dz ≤ 0
∂∆
so it is zero.
1
This is a famous property of ‘complete’ metric spaces known as Cantor’s intersection theorem.
Proof of Lemma 6.23 for general finite sets S. [Non-examinable] Let us first assume that
the set S consists of a single point w. If w 6∈ ∆ there is not much to prove, since we
apply what we proved when S is empty. So we assume that w ∈ ∆. We now consider
any other triangle ∆0 such that w ∈ ∆0 and ∆0 ⊂ ∆. Pick any subdivision of ∆ to
smaller triangles such that ∆0 is one of the triangles of this subdivision, and w is in the
interior of ∆0 . That is [
∆= ∆i
i
where ∆i may intersect only at the edges, and for some i we have ∆i = ∆0 . Then
Z XZ
f (z) dz = f (z) dz
∂∆ i ∂∆i
f (z) dz for all the ∆i 6= ∆0 . Hence

R
But by what we proved before we have that ∂∆i
Z Z
∂∆ ∂∆0
But since f is continuous on ∆0 we have that | ∂∆0 f (z) dz| ≤ L(∂∆0 ) sup z∈∂∆0 |f (z)|.
R
That is, Z
f (z) dz ≤ L(∂∆0 ) sup z∈∂∆0 |f (z)|
∂∆
But we
R can make ∆0arbitrarily small, that is we may let L(∂∆0 ) → 0. But then we get
that ∂∆ f (z) dz = 0.
If now the set S consists of more than one point, then we can divide the initial triangle
∆ to smaller triangles, where each one contains only one point of the finite set. Then
we can repeat the argument above to obtain the general result.
Finally we outline the proof of Lemma 6.26.
Proof of Lemma 6.26. [Non-examinable] The proof is nearly identical to the proof of
Theorem 6.17, but we now pick as a0 the defining point in the starlike domain that
has the property that for any b ∈ D the straight line connecting a0 to b lies entirely
in D. In particular, we define F (ω) in the same way but take γw specifically to be the
straight line in D between a0 and ω (skipping Step 1). For Step 2 we start as before,
but note that γw+h ∪ (−δh ) ∪ (−γw ) is a triangle and so the integral around it is zero
by assumption. In particular, in Step 2 we have
Z Z Z Z
F (ω +h) := f (z) dz = f (z) dz + f (z) dz = f (z) dz,
γw+h γw+h ∪(−δh )∪(−γω ) γw ∪δh γw ∪δh
and Step 2 then proceeds as before.

6.4 Cauchy’s Integral Formula

Cauchy’s integral formula tells us the following remarkable fact: the values of a holo-
morphic function in an open disc are determined by the values of the function on the
boundary! In other words, we can reconstruct the whole function in a 2-dimensional
disc by just knowing the function on the 1-dimensional circle that bounds the disc. Let’s
see how this works.
Theorem 6.27. [Cauchy’s Integral Formula (C.I.F.)] Let B := Br (a) for some
a ∈ C and r > 0. Assume that f : B → C is holomorphic. Then for every w ∈ B and ρ
such that |w − a| < ρ < r we have
Z
1 f (z)
f (w) = dz,
2πi |z−a|=ρ z − w
or equivalently Z
f (z)
dz = 2πif (w).
|z−a|=ρ z−w
Proof. We consider the auxiliary function
(
f (z)−f (w)
z−w , z 6= w
g(z) := 0
f (w), z = w
We note that this function is clearly holomorphic on B − {w} since there it is simply
the quotient of a holomorphic function and a non-zero holomorphic function. However,
it is also continuous at the point w because
f (z) − f (w)
lim g(z) = lim = f 0 (w) = g(w),
z→w z→w z−w
using here the fact that f is holomorphic at z = w. Hence by (the technical final
statement in) Cauchy’s Theorem for starlike domains 6.20 we have that
Z
g(z) dz = 0.
|z−a|=ρ
That is, (note that w does not lie on the curve |z − a| = ρ)
f (z) − f (w)
Z
dz = 0
|z−a|=ρ z−w
or equivalently Z Z
f (z) f (w)
dz = dz.
|z−a|=ρ z−w |z−a|=ρ z−w
We now observe that by using the geometric series expansion we may write
∞ n ∞
(w − a)n

1 1 1 X w−a X
= w−a = = .
z−w (z − a)(1 − z−a )
(z − a) z−a (z − a)n+1
n=0 n=0
|w−a| |w−a|
Moreover, we notice that the common ratio satisfies w−a
z−a = |z−a| = ρ < 1, since
|w − a| < ρ. In particular we have that the geometric series converges uniformly on
|z − a| = ρ (see Theorem 5.17) and
∞
(w − a)n
Z Z
f (w) X
dz = f (w) dz.
|z−a|=ρ z−w |z−a|=ρ (z − a)n+1
n=0
Because the series converges uniformly, we can exchange summation and integration
(see Theorem 5.8), and so obtain
Z ∞ Z
f (w) X 1
dz = f (w)(w − a)n .
|z−a|=ρ z−w |z−a|=ρ (z − a)n+1
n=0
But we have seen in Important Example 6.12 (at least for a = 0 but the same calculation
holds for any a) that
(
0, n 6= 0,
Z
1
n+1
=
|z−a|=ρ (z − a) 2πi, n = 0.
In particular we have that in the series above all terms are zero, except for n = 0. But
this is just f (w)2πi. Hence putting all together we conclude that
Z Z
f (z) f (w)
dz = dz = 2πif (w).
|z−a|=ρ z − w |z−a|=ρ z − w
Chapter 7
Features of Holomorphic
Functions
7.1 The Cauchy-Taylor Theorem and CIF for derivatives

We learned in back in Section 5.3 that if a complex power series has a positive (or
infinite) radius of convergence then it can be differentiated in its disc of convergence.
Hence power series with nonzero radii of convergence give rise to holomorphic functions
in their disc of convergence. One might wonder whether all holomorphic functions are
given by power series, and the answer is yes!
Theorem 7.1. [Cauchy-Taylor Theorem] Let f : U → C for some open set U , be

holomorphic. Then for any r > 0 such that Br (a) ⊂ U , f is given by a power series
that converges on Br (a):
∞
X
f (z) = cn (z − a)n , for z ∈ Br (a),
n=0
with Z
1 f (z)
cn = dz,
2πi |z−a|=ρ (z − a)n+1
for any 0 < ρ < r. The power series is called the Taylor series of f around a.
Proof. First fix ρ. For any w with |w − a| < ρ < r we have by the C.I.F. (Theorem 6.27)
∞
(w − a)n
Z Z
1 f (z) 1 X
f (w) = dz = f (z) dz
2πi |z−a|=ρ z−w 2πi |z−a|=ρ (z − a)n+1
n=0
∞ Z !
X 1 f (z)
= (w − a)n ,
2πi |z−a|=ρ (z − a)n+1
n=0
96
CHAPTER 7. FEATURES OF HOLOMORPHIC FUNCTIONS 97
where for the second equality we used the geometric series expansion, and for the last
equality we can exchange summation and integration thanks to the uniform convergence
of the geometric series on |z − a| = ρ. In particular we have established that for any
w ∈ Bρ (a) we have the following convergent expression:
∞ Z !
X 1 f (z)
f (w) = (w − a)n .
2πi |z−a|=ρ (z − a)n+1
n=0
1
R f (z)
Setting cn := 2πi |z−a|=ρ (z−a)n+1 , we have obtained a power series expansion for f (w)
when w ∈ Bρ (a).
To see that the constants cn we obtained don’t depend on ρ, note we have by
Corollary 5.22
f (n) (a)
cn =
n!
which doesn’t depend on ρ. Hence, since we would have got the same power series
whichever ρ we started with, and they all converge, the power series we obtained con-
verges on any Bρ0 (a) with 0 < ρ0 < r and hence converges on all of Br (a).
Functions that posses a Taylor series expansion in a disc Br (a) with r > 0 are called
analytic at a. Functions are called analytic if they are analytic at a for all points a in
their domain. We knew from Theorem 5.21 that all analytic functions are holomorphic,
but now Theorem 7.1 tells us that all holomorphic functions are analytic! For this rea-
son, some sources use the term ‘holomorphic function’ and ‘analytic function’ inter-
changeably.
We immediately obtain a nice extension of C.I.F. that allows us to compute all
derivatives of a function by performing contour integrals.
Theorem 7.2. [Cauchy’s Integral Formula for derivatives] Let B = Br (a) and
f : B → C holomorphic. Then for any 0 < ρ < r we have
Z
f (z) 2πi (n)
n+1
dz = f (a).
|z−a|=ρ (z − a) n!
Proof. By Theorem 7.1 we know that f has a convergent power series representation
in B given by
∞
X
f (z) = cn (z − a)n ,
n=0
1
R f (z)
where cn = 2πi |z−a|=ρ (z−a)n+1 dz. But on the other hand we also have that cn =
f (n) (a)
n! ,since we know this about power series (Corollary 5.22). Hence equating the two
expressions we obtain the statement of the theorem.
Remark 7.3. Combining Theorems 7.1 and 7.2 (or alternatively combining Corollary 5.22
and Theorem 7.1) tells us another important fact: If f is holomorphic on a ball
B = Br (a) then for all z in B one has the convergent expression
∞
X f (n) (a)
f (z) = (z − a)n ,
n!
n=0
which is the form of the Taylor expansion that we are used to.
Example 7.4. Consider the function f (z) = ez . We know by the Cauchy-Riemann

equations that f is holomorphic in C (see example 1. on page 36 after Theorem 3.5).
Therefore its Taylor series is convergent on all of C. We also calculated using the C-R
equations that f 0 (z) = ez , hence f (n) (z) = ez for all n ≥ 0. Thus f (n) (0) = e0 = 1 for
all n and
∞ ∞
X f (n) (0) n
X zn
f (z) = z = .
n! n!
n=0 n=0
We’ve finally proved that the two definitions of the exponential function are the same!
Remark 7.5. [Difference between real and complex analysis] The Cauchy-Taylor theorem
told us roughly that a function with a complex derivative at every point of a ball has a
Taylor series on the ball. This is far from being true in Real Analysis! Everyone should
know the following example: The real function
(
e−1/x x > 0
f (x) =
0 x≤0
has derivatives of all orders. We also have that f (n) (0) = 0 for all n ∈ N. In particular
the function f does not possess a Taylor series expansion since if it had one about x = 0
then
∞
X
f (x) = cn xn x ∈ (0 − , 0 + )
n=0
(n)
for some > 0, but that would mean cn = f n!(0) = 0 for all n ∈ N (by Remark 7.3).
However, f (x) is not identically zero in a neighbourhood of the origin. So there is no
Taylor series for it at x = 0.
As another direct corollary of the Cauchy-Taylor theorem we obtain
Corollary 7.6. [Holomorphic functions have infinitely many derivatives] If U ⊂ C is

an open set and f : U → C is holomorphic, then f has derivatives of all orders on U
and they are all holomorphic.
Proof. Indeed for any a ∈ U we take an r > 0 such that Br (a) ⊂ U . Then by the
theorem above we have that
∞
X
f (z) = cn (z − a)n , z ∈ Br (a).
n=0
But then by Corollary 5.22 we have f 0 (z) = ∞ n−1 for z ∈ B (a). That
P
n=1 ncn (z − a) r
is, around any point a of U we have a Taylor series representation and its derivative
exists, and it has again a Taylor series representation. That is, f 0 is holomorphic. We
can repeat now replacing f with f 0 .
Remark 7.7. [Another difference between real and complex analysis] This Corollary
marks another big difference between complex and real analysis. To see this, consider
that the real function f (x) = |x|xn , x ∈ R, is exactly n-times differentiable at x = 0
but the derivative f (n+1) (x) is not defined at x = 0.
Corollary 7.6 allows us to prove the following nice theorem, which can be viewed as
a converse to Cauchy’s Theorem.
Theorem 7.8.Z [Morera’s Theorem] Let D ⊂ C be a domain and f : D → C be
continuous. If f (z) dz = 0 for all closed contours γ in D, then f is holomorphic.
γ
Proof. By Theorem 6.17, there exists an F : D → C, holomorphic, such that F 0 (z) =

f (z). But we now by Corollary 7.6, F 0 is also holomorphic.
Examples We can use C.I.F. for derivatives to compute some integrals1 .

1. We compute Z
sin z
.
|z|=1 z2
This integral cannot be computed directly with C.I.F.. However we can apply
C.I.F. for derivatives (Theorem 7.2) with f (z) = sin z, ρ = 1, r anything bigger
than 1, a = 0 and n = 1. Then we obtain
Z
sin z 2πi 0
2
= f (0) = 2πi.
|z|=1 z 1!
2. We now compute
ez
Z
dz.
|z|=3 z 2 (z − 1)
1
although later we will have a more powerful tool called Cauchy’s Residue Theorem.
We first find the partial fraction transformation

1 a b c a(z − 1)z + b(z − 1) + cz 2
2
= + 2+ =
z (z − 1) z z z−1 z 2 (z − 1)
(a + c)z 2 + (b − a)z − b
= .
z 2 (z − 1)
That is, b = −1, a = b = −1 and c = −a = 1. Hence
1 −1 −1 1
= + 2 +
z 2 (z − 1) z z z−1
and we may write
ez ez ez ez
Z Z Z Z
2
dz = dz − 2
dz − dz.
|z|=3 z (z − 1) |z|=3 (z − 1) |z|=3 z |z|=3 z
But the first and the third can be computed using C.I.F. and they are equal to
2πie1 and 2πie0 = 2πi respectively and the second we can use C.I.F. for derivatives
(Theorem 7.2) to conclude it is 2πie0 . Hence the integral is equal to 2πi(e − 2).
7.2 Liouville’s Theorem

The next big theorem is Liouville’s theorem. Before stating this, let us explain some
terminology. We say that a function f : C → C is entire if f is holomorphic on the
whole of C. We say that a function f : C → C is bounded if there exists some positive
real number M such that |f (z)| ≤ M for all z ∈ C.
Theorem 7.9. [Liouville’s Theorem] Every bounded entire function is constant.
Proof. Let f : C → C be entire, and assume there exists an M ∈ R such that |f (z)| <
M for all z ∈ C. Consider any w ∈ C. Then we have for any ρ > |w|, by C.I.F.
(Theorem 6.27) and the Estimation Lemma (Lemma 6.15)
Z Z
1 f (z) f (z)
|f (w) − f (0)| = dz − dz
2π |z|=ρ z − w |z|=ρ z
Z
1 1 1
= f (z) − dz
2π |z|=ρ z−w z
Z
1 f (z)
= w dz
2π |z|=ρ z(z − w)
|w|
Z
f (z)
= dz
2π |z|=ρ z(z − w)
|w| |f (z)|
≤ · (2πρ) · sup .
2π |z|=ρ |z||z − w|
We then use the boundedness of f and the reverse triangle inequality respectively to
conclude that the latter quantity is
1 |w|M
≤ |w| ρ M sup ≤ .
|z|=ρ |z||z − w| ρ − |w|
The final term here tends to zero as ρ → ∞ and we picked ρ arbitrarily. Thus we have
f (w) = f (0). This holds for any w ∈ C. Hence f is constant.
Remark 7.10. Note that the assumption on the holomorphicity cannot be dropped in
the statement of Louiville’s theorem. For example, it is easy to find bounded continuous
functions from C to C that are not constant. For example, we could take f (x + iy) =
sin(x) + sin(y), or f (x + iy) = sin(x) + i sin(y) In fact: there are bounded non-constant
functions from C to C that have all real derivatives, are ≡ 1 inside B1 (0) and are ≡ 0
outside B2 (0). These are called ‘bump-functions’ and are often used in real analysis.
Liouville tells us there are no holomorphic bump-functions.
An immediate consequence of Liouville’s theorem is the Fundamental Theorem of Al-
gebra. This is very nice: suddenly Complex Analysis has given us a result in Algebra!
Theorem 7.11. [Fundamental Theorem of Algebra] Every non-constant polyno-

mial with complex coefficients (i.e P (z) = ad z d + ad−1 z d−1 + . . . + a1 z + a0 with ai ∈ C
and ad 6= 0) has a complex root. That is, there exists an z0 ∈ C such that P (z0 ) = 0.
Proof. Clearly, since we are taking d ≥ 1 we have that |P (z)| → ∞ as |z| → ∞ (see
Problem Sheet 12, question 9a). In particular there exists some R > 0 such that
|P (z)| > 1 for all |z| > R. We are going to argue by contradiction. That is, we will
assume that there exists no z0 ∈ C such that P (z0 ) = 0, and derive a contradiction.
1
We define the function f (z) := P (z) . Then under our hypothesis, we have that f (z)
is entire, since P (z) is itself entire and has no roots. We now show that such a function is
also bounded. Indeed, since f (z) is holomorphic it is also continuous, and so in particular
(by Theorem 2.30) it is bounded on the compact set B R (0) = {z ∈ C : |z| ≤ R}; that
is, on the open disc/ball BR (0) and its boundary. Moreover, by the very definition of R
we have that f (z) is bounded also on the complement of the above set (since there
|P (z)| > 1 implies |f (z)| = 1/|P (z)| < 1). Hence f (z) is bounded on C. This means
we can apply Liouville’s Theorem (Theorem 7.9) to f and conclude that f (z) must be
constant. But that means that P (z) has to be a constant polynomial. Contradiction.
7.3 The Maximum Modulus Principle

We now present one last direct application of Cauchy’s Integral Formula.
Theorem 7.12. [Local maximum modulus principle] Let B := Br (a), and f :

B → C holomorphic. If for every z ∈ B we have
|f (z)| ≤ |f (a)|
then f is constant on B.
Proof. Step 1: The modulus of f is constant.

Pick any w ∈ Br (a). We’ll prove |f (w)| = |f (a)|, so the modulus of f is constant.
Let ρ = |w − a|, and note ρ < r. Parametrize the curve |z − a| = ρ through ω
by γ(t) = a + ρe2πit with t ∈ [0, 1]. Recall from inequality (6.1) in the proof of the
Estimation Lemma (Lemma 6.15) that for any continuous function g : [a, b] → C, we
have Z b Z b
g(t) dt ≤ |g(t)| dt.
a a
Taking g(t) = f (γ(t)) we can use C.I.F. to get
Z Z 1 Z 1
1 f (z) 1 f (γ(t)) 0 1 f (γ(t))
|f (a)| = dz = γ (t)dt = 2πiρe2πit dt
2π γ z−a 2π 0 γ(t) − a 2π 0 ρe2πit
Z 1
= f (γ(t)) dt
0
Z 1
≤ |f (γ(t))| dt
0
Z 1
≤ |f (a)| dt = |f (a)|.
0
Since we began and ended at the same thing, all inequalities above are equalities. The
only way the last inequality can be an inequality (since f is continuous) is if |f (z)| =
|f (a)| for all z with |z − a| = ρ, and in particular, |f (w)| = |f (a)|.
Step 2: f is constant.
We know |f (z)| = c for some c ∈ C and all z ∈ B. If c = 0, then we have that
|f (z)| = 0 and hence we should have that f (z) = 0, that is constant. If we assume that
2
c 6= 0 then we may write f (z)f (z) = c2 , and so f (z) = fc(z) . That is both f and f¯ are
holomorphic. By the Cauchy-Riemann (C-R) equations we obtain that such a function
has to be constant. Indeed if we write f (z) = u(x, y) + iv(x, y) with z = x + iy, x, y ∈ R,
then f (z) = u(x, y) − iv(x, y). The C-R equation for the first function read ux = vy
and uy = −vx . And the ones for the second read ux = −vy and uy = vx . From this it
follows that ux = vx = 0. But then by Proposition 3.3 we have f 0 (z) = ux + ivx = 0.

Hence f is constant by the Zero Derivative Theorem (Theorem 3.12).
The next step is to extend the above theorem to any domain. Indeed we have the
following,
Theorem 7.13. [Maximum Modulus Principle] Let D be a domain, and f : D → C
holomorphic. If there exists an a ∈ D such that
|f (z)| ≤ |f (a)|, ∀z ∈ D
then f is constant.
We need to know the following fact from the Topology of metric spaces (see Problem
Sheet 3 Q13(ii)).
Fact 7.14. [Path-connected implies connected] If U ⊂ C is path connected and open,
then it is not possible to write U = U1 ∪ U2 where U1 and U2 are disjoint, open, and
nonempty. In particular, domains are connected.
Proof of Theorem 7.13. Let U1 be the set of points z of D where f (z) = f (a). Then
U1 is nonempty. We wish to show U1 is open. Given z ∈ U1 , there is a disc Br (z) ⊂ D
since D is open. Pick w ∈ Br (z). Then since |f (w)| ≤ |f (a)| = |f (z)| and z is the
centre of Br (z), the Local Maximum Modulus Principle (Theorem 7.12) tells us that f
is constant on Br (z). Hence |f (w)| = |f (a)| and so Br (z) ⊂ U1 . Thus U1 is open.
Let U2 = D − U1 . Then we can write U2 = f −1 (C − {f (a)}), and by Theorem 2.17
this set must be open since f is holomorphic (so continuous) and C − {f (a)} is open.
Since D is path-connected, U2 must be empty by Fact 7.14 (otherwise U1 and U2
would be disjoint, open, and nonempty sets whose union is D). So U1 = D and f (z) =
f (a) for all z ∈ D.
Example 7.15. Find the maximum of |z 2 + 2z − 3| on B1 (0) = {z ∈ C : |z| ≤ 1}.

First we explain why such a maximum value exists. We set f (z) = z 2 + 2z − 3. the
function |f | : B1 (0) → R is continuous, and hence since B1 (0) is compact, the function
|f | attains a maximum value. By the Maximum Modulus Principle (Theorem 7.13)
this value cannot be attained in B1 (0) since the function f (z) is holomorphic, and not
constant. Hence the function |f | attains its maximum value at the boundary of the unit
disc, that is for some z with |z| = 1, and such a z we may write z = eit , with t ∈ [0, 2π].
We now note
|z 2 + 2z − 3|2 = (z 2 + 2z − 3)(z 2 + 2z − 3) =
(e2it + 2eit − 3)(e−2it + 2e−it − 3) = 14 − 3e−2it − 3e2it − 4eit − 4e−it =
14 − 6 cos(2t) − 8 cos(t) = −12 cos2 (t) − 8 cos(t) + 20.
−1
By Calculus, the polynomial −12x2 − 8x + 20 attains its maximum at x = cos(t) = 3 ,
and hence the maximum value of |z 2 + 2z − 3|2 is 64 2 √8
3 and so of |z + 2z − 3| is 3 .
7.4 Analytic continuation and the Identity Theorem

We now move to discuss another interesting application of the fact that holomorphic
functions are analytic (that is, around each point can be represented by a power series).
Let us start by considering the following example. Define the function
∞
X
f (z) := z n , z ∈ B1 (0).
n=0
We know that such a function is holomorphic in B1 (0). We may ask:

Problem 7.16. Is there a holomorphic function g(z) defined on a larger domain D that
contains B1 (0) and with the property that f (z) = g(z), ∀z ∈ B1 (0)?
1
By our knowledge of the geometric series we may guess g(z) = 1−z , which is holo-
morphic in C \ {1}, a set that certainly contains B1 (0). We know that f (z) = g(z) for
all z ∈ B1 (0). Furthermore, we may ask the question, is this g unique? In order to
address these questions we start by studying zeros of holomorphic functions.
Let f : Br (a) → C be a holomorphic function, and write
∞
X
f (z) = cn (z − a)n ,
n=0
for its Taylor series at a (which exists and converges by Cauchy-Taylor, Theorem 7.1).
If we assume that f is not the zero function, then not all of the cn ’s are zero. Let us set
m := min{n ≥ 0 : cn 6= 0}.
Then we may write
∞
!
X
m n−m
f (z) = (z − a) cn (z − a) ,
n=m
and if we set h(z) := ∞ n−m = c + c

P
n=m cn (z − a) m m+1 (z − a) + . . . we have that h(a) =
cm 6= 0 (and h itself, being a convergent power series, is holomorphic by Theorem 5.21).
Note that f (a) = 0 if and only if m > 0.
Definition 7.17 (Orders of zeros). We say f : Br (a) → C has a zero of order
m at a if for some m > 0 there exists a holomorphic h : Br (a) → C such that
f (z) = (z − a)m h(z) and h(a) 6= 0.
One can check (see Problem Sheet 13) that f has a zero of order m at a if and only if
f (a) = f (1) (a) = f (2) (a) = . . . = f (m−1) (a) = 0

and f (m) (a) =
6 0. For that reason, one may use the latter property as an alternative
definition of a zero of order m.
Example 7.18. Let f (z) = z(ez − 1), and take a = 0. Then we have that
∞
! !
X zn X z n−1
2
f (z) = z −1 =z .
n! n!
n=0 n=1
Hence f has a zero of order 2 at a = 0. And indeed f 0 (z) = (ez − 1) + zez and
f (2) (z) = ez + ez + zez . And so f (0) = f 0 (0) = 0 and f (2) (0) = 2 6= 0. Note that f also
has zeros at a = 2πin for integers n 6= 0, and one can check they are order 1.
Next, we make an important observation. If for a holomorphic function f we have

f (a) 6= 0, then since f is continuous at z = a there exists ρ > 0 such that f (z) 6= 0 for
all z ∈ Bρ (a). That is, if f is non-zero at z = a then then it is also non-zero ‘nearby’ a.
So, any point at which f is non-zero can be isolated from all zeros of f . Surprisingly,
this is also true for the zeros:
Proposition 7.19. [Principle of isolated zeros] Let f : Br (a) → C be holomorphic

and not identically zero. Then there exists an 0 < ρ ≤ r such that
f (z) 6= 0 ∀z ∈ Bρ (a) − {a}.
In particular, this holds when f (a) = 0; that is, all the zeros of a holomorphic function
must be isolated from one another.
Proof. As observed above, the statement holds when f (a) 6= 0 by continuity. The
interesting case is when f (a) = 0; that is, when a is a zero of f . Since f is not
identically zero, we have, by definition, that f (z) = (z − a)m h(z), where m > 0 is the
order of the zero, and h(a) 6= 0. But h is holomorphic, and in particular continuous at
z = a, so as before we have that there exists an ρ > 0 such that h(z) 6= 0 for z ∈ Bρ (a).
But clearly (z − a)m is also non zero in Bρ (a) − {a}, and hence so is f .
We can now prove another remarkable result.
Theorem 7.20. [Uniqueness of analytic continuation] Let D0 ⊂ D be non empty

domains, and f : D0 → C be holomorphic. Then there exists at most one holomorphic
g : D → C such that
f (z) = g(z), ∀z ∈ D0 .
The function g, if it exists, is called the analytic continuation of f to D.
Proof. Let g1 , g2 : D → C be analytic continuations of f to D. We set h(z) := g1 (z) −

g2 (z). Then we have that h(z) = 0 for all z ∈ D0 . We claim that actually h(z) = 0 for
all z ∈ D. For this we consider the following sets
D0 := {w ∈ D : ∃ r > 0 s.t. h(z) = 0, ∀z ∈ Br (w)}

and
D1 := {w ∈ D : h(n) (w) 6= 0 for some n ≥ 0} = ∪∞

n=0 {w ∈ D : h
(n)
(w) 6= 0 }.
Clearly D0 ⊂ D0 so D0 is nonempty. But we can also see that D0 is open: If

w ∈ D0 , then h is zero on some ball Br (w). We claim Br (w) ⊂ D0 . Pick z ∈ Br (w),
then there is a smaller ball centered at z contained in Br (w) (since Br (w) is itself open)
on which h is of course zero. Thus Br (w) ⊂ D0 .
Also, D1 is open by Lemma 2.8 since it is the union of (h(n) )−1 (C − {0}), which are
open by Theorem 2.17 (since each h(n) is holomorphic, hence continuous, and C − {0}
is open).
Next, note that if w ∈/ D1 then h(n) (w) = 0 for all n. Since h is holomorphic it has a
power series representation on some disc centered at w (by Cauchy-Taylor), which must
then be identically 0 on that disc (by Corollary 5.22). Hence w ∈ D0 .
On the other hand, if z ∈ D1 then h must have a non-trivial Taylor series about ω
(at least one coefficient is non-zero). It follows from Proposition 7.19 that h cannot be
zero on any ball centred at w. In other words, w ∈ / D0 . This shows D0 ∪ D1 = D
and that this union is disjoint.
Now by Fact 7.14, D1 must therefore be empty. Hence D = D0 , and so h(z) = 0 for
all z ∈ D. That is, g1 (z) = g2 (z) for all z ∈ D.
Corollary 7.21. Let f and g be two holomorphic functions defined on a domain D,

such that for some disc Br (a) ⊂ D we have that f (z) = g(z) for all z ∈ Br (a). Then
f (z) = g(z) for all z ∈ D.
Actually, even a weaker condition is enough to guarantee the above equality. For
this we introduce the notion of non-isolated point.
Definition 7.22. Given a set S ⊂ C, we will say that a point w ∈ S is
• an isolated point of S if there exists > 0 such that B (w) ∩ S = {ω};
• a non-isolated point of S if for every > 0 there exists an z ∈ S with z 6= w

such that z ∈ B (w).
Note, a point w ∈ S is non-isolated if we can get arbitrarily close to w by using other

points of S. So, in other words, w ∈ S is non-isolated iff it is a limit point of some
sequence of points in S − {w} (see Problem Sheet 14).
Theorem 7.23. [Identity Theorem] Let f, g : D → C be holomorphic on a domain D.

If the set S := {z ∈ D : f (z) = g(z)} contains a non-isolated point, then f ≡ g on D.
Proof. Let w ∈ S be a non-isolated point. Write h(z) := f (z) − g(z). Then h(w) = 0
and we see that w is not an isolated zero for the holomorphic function h. Indeed we
cannot find a disc around a such that h(z) 6= 0 for all z in that disc, since we can
arbitrary close to w by using point in S, namely zeros of h. By Proposition 7.19, the
function h must then be identically zero on some ball Bρ (w) ⊂ D. But then f (z) = g(z)
for all z ∈ Bρ (w) and so by Corollary 7.21 we have that f = g on D.
1

Example 7.24. Let f : C → C be a holomorphic function, and assume that f n =
sin n1 for all n ∈ Z, n ≥ 1. From this we can show that f (z) = sin(z) for all z ∈ C.

Indeed, note that since both sin and f are holomorphic on C, they are continuous, so

1 1
f (0) = lim f = lim sin = sin(0),
n→∞ n n→∞ n
by Lemma 2.29. Hence
1
S = { z ∈ C : f (z) = sin(z)} ⊃ {0} ∪ { : n ∈ Z, n ≥ 1 }.
n
Since 0 is a non-isolated point of {0} ∪ { n1 : n ∈ Z, n ≥ 1 }, 0 is a non-isolated point
of S. Therefore by the Identity Theorem (Theorem 7.23), f ≡ sin on C.
7.5 Harmonic functions and the Dirichlet problem

We now study the relationship between holomorphic functions (which we understand
quite well now) and certain important real functions called harmonic functions.
Definition 7.25 (Harmonic functions). A real-valued function u : D → R on a domain
D ⊆ C is harmonic if u(x, y) has continuous second order partial derivatives and
uxx + uyy = 0.
The equation uxx + uyy = 0 is called Laplace’s equation. The first observation is
that holomorphic functions hide harmonic functions in their real and imaginary parts.
Proposition 7.26. Let f = u + iv : D → C be holomorphic on a domain D ⊆ C. Then
u and v are both harmonic functions. That is,
uxx + uyy = 0, vxx + vyy = 0.
Proof. We know from Proposition 3.3 that the first partial derivatives of u and v must
exist. Moreover, this Proposition also gives various formulae for the derivative f 0 in
terms of the these partial derivatives, e.g., f 0 = ux − iuy = vy + ivx . However, Corol-
lary 7.6 implies that the nth complex derivative f (n) exists and is holomorphic for all
n ≥ 0. In particular, the derivative f 0 must be continuous and so then must all the 1st
partial derivatives of u and v. Repeating this argument shows that all partial derivatives
of u and v of all orders exist and are continuous. Moreover, by the Schwartz-Clairault
Theorem (from Calculus I) we have symmetry of the second partial derivatives:
uxy = uyx , vxy = vyx .

Also, by Proposition 3.3, the first partial derivatives of u and v satisfy the C-R equations
ux = vy and uy = −vx . By differentiating the first C-R equation with respect to x and
the second C-R equation with respect to y we get
uxx = vyx = vxy and uyy = −vxy .
Adding these together gives uxx + uyy = vxy − vxy = 0.
Similarly, by differentiating the first C-R equation with respect to y and the second
C-R equation with respect to x we get
vyy = uxy and vxx = −uyx = −uxy .
Adding these together gives vxx + vyy = uxy − uxy = 0.
Showing functions are harmonic (examples)

(i) Let f (z) = ez = ex cos y + iex sin y. Then u(x, y) = ex cos y and v(x, y) = ex sin y,
so as expected we have
uxx = ex cos y and uyy = −ex cos y,
plus
vxx = −ex sin y and vyy = ex sin y.
Thus u and v are harmonic
(ii) Let f (z) = x2 + y 2 (= |z|2 ). Then v(x, y) = 0 always satisfies Laplace’s equation,
but u(x, y) = x2 + y 2 does not (since uxx = 2 = uyy and so uxx + uyy = 4). So u
is not harmonic.
(iii) Consider u : C → R defined by u(z) = u(x, y) = x2 − y 2 + 3x. We have
ux = 2x + 3, uy = −2y, so uxx = 2, uyy = −2.
Thus, u is harmonic.
This raises an interesting question: given a harmonic function on a domain, does it come
from a holomorphic function? Is the converse to Proposition 7.26 true? The answer
is yes, but only on domains without ‘holes’. Recall that our favorite types of domains
without holes are the starlike domains.
Theorem 7.27. [Existence of harmonic conjugates] If D is a starlike2 domain and
u : D → R is harmonic, then there exists a harmonic function v : D → R such that
f = u + iv is holomorphic on D. The function v is called a harmonic conjugate of u.
The harmonic conjugate v is unique up to adding a real constant.
2
As should be clear from the proof, the condition of the domain being starlike can be relaxed in this
statement when we prove a stronger version of Cauchy’s Theorem; namely it can be replaced with the
condition on the domain that appears in our strengthening of Cauchy’s Theorem.
We will use the following proposition, which is interesting in its own right.
Proposition 7.28. Let f : D → C be a holomorphic function on a starlike domain D.

Then f has a holomorphic antiderivative on D. In other words, there is a holomorphic
function F : D → C such that F 0 (z) = f (z) for all z ∈ D.
R
Proof. By Cauchy’s Theorem for starlike domains (Theorem 6.20), we have γ f (z)dz =
0 for all closed contours γ ⊂ D. Hence by the converse to the Fundamental Theorem
of Calculus (Theorem 6.17), f has a holomorphic antiderivative on D.
Proof of Theorem 7.27. Motivation: Suppose we had found such a holomorphic f =

u + iv. By Proposition 3.3, we would have f 0 = ux − iuy , and by Corollary 7.6 this
function would be holomorphic. Let us try to reverse the logic of this.
Step 1. Construct f 0 :
This means we should define a function g = ux − iuy : D → C. We want to prove g is
holomorphic. First observe that the Cauchy Riemann equations for g hold:
(ux )x = uxx = −uyy = (−uy )y

(ux )y = uxy = uyx = −(−uy )x .
Since u is harmonic, its second partial derivatives are continuous, which means the
first partial derivatives of the real and imaginary parts of g are continuous. Hence by
Theorem 3.5, g is holomorphic.
Step 2. Construct f :
By Proposition 7.28, g has a holomorphic antiderivative F = U + iV . Since F 0 = g, we
have F 0 = Ux − iUy = g = ux − iuy . This implies Ux = ux and Uy = uy , or
(U − u)x = (U − u)y = 0.
This implies the function (U − u) has vanishing first partial derivatives. This means it
is a (real) constant, so U = u + c. Letting v = V we have that f = u + iv = F − c is a
holomorphic function and we have established the existence of the harmonic conjugate v.
Since f is holomorphic, it follows from Proposition 7.26 that the harmonic conjugate
v is harmonic. If ṽ is another harmonic conjugate, then f = u + iv and f˜ = u + iṽ are
both holomorphic, so i(v − ṽ) = f − f˜ is a purely imaginary holomorphic function. An
easy exercise using the Cauchy-Riemann equations shows that the only purely imaginary
holomorphic functions on a domain are constant. Hence v − ṽ is a real constant, proving
the statement that harmonic conjugates are unique up to adding real constants.
Finding harmonic conjugates (examples)

Continuing the previous example (iii), we have that u(x, y) = x2 − y 2 + 3x is harmonic
on C. Let’s find its harmonic conjugate v(x, y).
By the first C-R equation we know that v must satisfy vy = ux = 2x + 3. Integrating

with respect to y we get
v(x, y) = (2x + 3)y + g(x),
where g is a function of x independent of y. By differentiating this with respect to x
and applying the second C-R equation, we have
2y + g 0 (x) = vx = −uy = 2y.
Thus g 0 (x) = 0, so g(x) = c is constant and we’ve found our harmonic conjugates:
v(x, y) = 2yx + 3y + c (for some c ∈ R).
We can check that the corresponding function
f (x, y) = u(x, y) + iv(x, y) = (x2 − y 2 + 3x) + i(2xy + 3y + c)
is holomorphic. Indeed, with z = x + iy we have 3x + 3iy = 3z, and z 2 = (x + iy)2 =
x2 −y 2 +2ixy, so our corresponding holomorphic function f is simply f (z) = z 2 +3z +ic.
The Dirichlet Problem

The following problem has a huge and famous history - it is very important in theoretical
physics, particularly in the case D = B1 (0).
Problem 7.29. Let D be a domain in C with closure D̄ with boundary ∂D and let
g : ∂D → R be a continuous function. Find a continuous function µ : D̄ → R such that
µ is harmonic on D and matches g on ∂D.
In other words, find a continuous µ such that its second partial derivatives exist, are
continuous, and satisfy µxx + µyy = 0 on D, and for which µ = g.
∂ D̄
Example 7.30. Let D1 = {x + iy : 2 < y < 5}. Then D̄1 = {x + iy ∈ C : 2 ≤ y ≤ 5}.
Let (
4 if y = 2,
g(x, y) =
13 if y = 5.
Let us try to solve the Dirichlet problem for this boundary function. To make a guess
at the form of µ, first notice that g is independent of x. It is therefore likely that µ
is also independent of x and so µx = 0 (since it must match g on the boundary). If
µx = 0 then 0 = µxx = −µyy , because we want µ to be harmonic. Integrating twice
with respect to y gives us the ansatz3 :
µ(x, y) = ay + b (for some a, b, ∈ R).
We need µ to match g on the boundary, so must have a5 + b = 13 and a2 + b = 4. Thus
a = 3 and b = −2, and µ(x, y) = 3y − 2 is a solution to the Dirichlet Problem on D1 .
3
Ansatz means guess.
How about nastier regions?
Proposition 7.31. Suppose that f = u + iv : D → C is holomorphic on a domain

D ⊆ C, and suppose that µ is harmonic on f (D). Then, µ̃ := µ ◦ f = µ(u, v) is
harmonic on D.
Proof. Apply the chain/product rules and use the C-R equations (see Sheet 14).
(examples continued.)
Example 7.32. We modify the previous example and consider the domain D2 formed
by rotating D1 = {x + iy ∈ C : 2 < y < 5} anticlockwise by π/4 √ degrees. We ask for a
harmonic √function µ̃ such that µ̃ = 13 on the line y = x + 5 2 and µ̃ = 4 on the line
y = x + 2 2. Immediately, we have D1 = f (D2 ), where f (z) = e−iπ/4 z. We know that
f is holomorphic and that µ(x, y) = 3y − 2 is harmonic on f (D2 ). By Proposition 7.31
the function µ̃(x, y) := µ(f (x, y)) is harmonic on D1 . It is easy to check that f takes
the boundary lines of the new region to the boundary lines of the region from (i), and
you can check that µ̃ matches the values taken on the new boundary lines.
Note that
1 1 1
f (x, y) = e−iπ/4 (x + iy) = √ (1 − i)(x + iy) = √ (x + y) + i √ (y − x).
2 2 2
So, our solution is

1 1 1 3
µ̃(x, y) = µ √ (x + y), √ (y − x) = 3 √ (y − x) − 2 = √ (y − x) − 2.
2 2 2 2
Chapter 8
General form of Cauchy’s

Theorem and C.I.F.
8.1 Winding number and simply connected sets

In this chapter we discuss how we can improve the statements of Cauchy’s Theorem (for
starlike domains) and the CIF. We will widen the class of domains on which Cauchy’s
Theorem applies, and widen the class of contours for which the CIF applies.
We first want to introduce the notion of the winding number. Consider the curves
γ1 (t) = e2πit with t ∈ [0, 1] and γ2 (t) = e2πit with t ∈ [0, 2]. Even though their graph is
the same, namely the circle of radius one centered at the origin, the second is tracing
this graph twice. We would like to define a quantity that keeps track of this.
Definition 8.1 (Winding number). Assume we are given a curve γ : [a, b] → C of the
form
γ(t) = w + r(t)eiθ(t) ,
where w ∈ C, and r, θ : [a, b] → R are (continuous) piecewise C 1 functions with r(t) > 0
for all t ∈ [a, b]. We define
θ(b) − θ(a)
I(γ; w) := ,
2π
and call it the winding number or index of the curve γ around the point w.
Note that w 6∈ γ since r(t) > 0. In the examples above we have that I(γ1 ; 0) = 1
and I(γ2 ; 0) = 2. Actually one can show that if γ is closed then I(γ; w) ∈ Z. Indeed,
since γ is closed, we have that γ(a) = γ(b) and hence r(a)eiθ(a) = r(b)eiθ(b) and so
θ(b) − θ(a) ∈ 2πZ. That is, I(γ, w) ∈ Z. We now state without proof the following.
Theorem 8.2. Let γ : [a, b] → C be a contour, and w ∈ C with w 6∈ γ. Then there
exists r, θ : [a, b] → R, (continuous) piecewise C 1 , and r(t) > 0 such that
γ(t) = w + r(t)eiθ(t) .
112
CHAPTER 8. GENERAL FORM OF CAUCHY’S THEOREM AND C.I.F. 113
Actually it is not hard to see that r(t) is unique, and is given as |γ(t) − w|, that is,
measures the distance of the point γ(t) from the point w. The function θ is unique
up to an additive constant, but this does not affect the definition of I(γ; w) since we
measure differences.
So now we see that we can define the winding number for any contour γ and a point
w 6∈ γ. The relationship between winding number and complex integration is given by
the following lemma.
Lemma 8.3. Let γ : [a, b] → C be a closed contour, and w 6∈ γ. Then
Z
1 dz
I(γ; w) = .
2πi γ z − w
Proof. We write γ(t) = w + r(t)eiθ(t) , by Theorem 8.2. Then we have that
γ 0 (t)
Z Z b Z b 0
dz r (t) 0
= dt = + iθ (t) dt =
γ z−w a γ(t) − w a r(t)
[ln(r(t)) + iθ(t)]ba = i(θ(b) − θ(a)) = 2πiI(γ; w),

where we have used the fact r(a) = r(b) since γ is closed.
Often we will need to establish that I(γ; w) is zero. For this we have the following
proposition.
Proposition 8.4. Let D be a starlike domain. If γ is any closed contour in D and
w∈/ D then I(γ; w) = 0.
1
Proof. The function f (z) = z−w is holomorphic on D since w 6∈ D. And since γ is
closed in this domain, by Cauchy’s Theorem for starlike domains and Lemma 8.3
Z
1 1
I(γ; w) = dz = 0.
2πi γ z − w
This property is so useful, we use it to make the following definitions.

Definition 8.5. Let U ⊂ C be an open set. Then
1. A closed contour γ in U is called homologous to zero in U if
I(γ; w) = 0, ∀w 6∈ U.
2. The open set U is called simply connected if every closed contour in U is homolo-
gous to zero in U .
Informally, a set is simply connected if it has no ‘holes’. This is the condition we

seek to replace ‘starlike’ with in the statement of Cauchy’s Theorem.
Comment: [Advanced] The more usual definition of a simply connected set D is
that every continuous closed curve in D can be continuously shrunk to a point. The
mathematically precise way to state this is that for every continuous γ : [0, 1] → D
with γ(0) = γ(1) = a, there is a continuous map F : [0, 1] × [0, 1] → D such that
F (0, t) = γ(t) for t ∈ [0, 1], F (x, 0) = F (x, 1) = a for all x, and F (1, t) = a for
all t ∈ [0, 1]. The map F is called a homotopy. This is the beginning of the part
of Mathematics called Algebraic Topology. This definition involving shrinking loops
coincides with Definition 8.5, although proving this is outside the scope of the course.
Example 8.6. Proposition 8.4 tells us that all starlike domains are simply connected.
For example, C is simply connected, and Br (a) is simply connected. This matches our
intuition about these sets not having holes.
Example 8.7. An example of a non simply connected set is an annulus
A := {z ∈ C : α < |z| < β},
for all 0 ≤ α < β ∈ R. Indeed if we take the curve γ(t) = ρe2πit with any ρ ∈ (α, β), then
we have that γ ⊂ A and if we consider the origin 0 6∈ A we have that I(γ; 0) = 1 6= 0
by Cauchy’s Integral Formula. That is, the curve γ is not homologous to zero in A.
Now we have seen how our domains can be generalized, we introduce the notion of
a cycle, which extends that of a closed contour.
Definition 8.8 (Cycle). A cycle Γ, defined in an open set U , is a finite collection of

closed contours γ1 , . . . , γn in U . We denote a cycle via the formal sum of
Γ := γ1 + γ2 + . . . + γn .
We say that a point w ∈ C does not lie in Γ if w 6∈ γi for all i = 1, 2, . . . , n. For such a
point we define the winding number of a cycle Γ around w by
n
X
I(Γ; w) := I(γi ; w).
i=1
Moreover we define Z n Z
X
f (z) dz := f (z) dz.
Γ i=1 γi
Finally we say that the cycle Γ is homologous to zero in U if for every w 6∈ U we

have that I(Γ; w) = 0.
Note, there is no actual operation of addition in te formal sum, it is

simply notation. The curves in the cycle are independent of each other and there
is no need to have any relation between them. The γi can be disjoint, intersecting, or
copied on top of themselves with the same or reverse orientation. Note that a cycle
extends the notion of a closed contour. In particular a closed contour γ is an example
of a cycle.
Example 8.9. [Important] We consider the annulus A := {z ∈ C : r < |z − a| < R}
for 0 ≤ r < R ≤ ∞. Moreover we consider two curves γ1 (t) := a + ρ1 e−2πit and
γ2 (t) = a + ρ2 e2πit with t ∈ [0, 1], with r < ρ1 < ρ2 < R. We define the following cycle
in A:
Γ := γ1 + γ2 .
We claim that Γ is homologous to zero in A. Let w 6∈ A. Then we have
Z Z
1 1 1 1
I(Γ; w) = I(γ1 ; w) + I(γ2 ; w) = dz + dz.
2πi γ1 z − w 2πi γ2 z − w
1 1
R R
If |w − a| ≥ R then we have by Cauchy’s Theorem that γ1 z−w dz = γ2 z−w dz = 0,
andR so I(Γ; w) = 0. If on Rthe other hand |w − a| ≤ r, then we have by CIF that
1 1 1 1
2πi γ1 z−w dz = −1 and 2πi γ2 z−w dz = 1 (for γ1 we get a minus sign since we traverse
the circle clockwise). That is, again we have I(Γ; w) = −1 + 1 = 0.
One nice situation that will occur many times is that we are given a contour that
doesn’t cross itself.
Definition 8.10 (Simple closed curve). A closed curve γ : [a, b] → C is called simple
if γ(t1 ) = γ(t2 ) for any t1 < t2 implies that t1 = a and t2 = b (i.e no self-crossing or
back-tracking ).
We now state (without proof) a few facts about simple closed curves. We start with
the Jordan Curve Theorem.
Theorem 8.11. [Jordan Curve Theorem] Let γ ⊂ C be a simple closed curve. Then
its complement C\γ, is a disjoint union of two domains, exactly one of which is bounded.
We will call the bounded domain the interior of γ, and we will be denoting it by Dγint ,
and the other will be called the exterior, and we will be denoting it by Dγext . In particular
for a point w ∈ C and not in γ, we will say that w lies inside γ if w ∈ Dγint and outside
if w ∈ Dγext .
Given a simple closed contour γ (so, a simple closed curve that is also piecewise C 1 )
it is possible to put an orientation on γ such that for all w ∈ C \ γ, we have that
(
1, w ∈ Dγint ,
I(γ; w) =
0, w ∈ Dγext .
That is, we pick a way to trace the curve γ such as the bounded component Dγint is
always on the left of the curve. We then say that the curve γ is positively oriented. This
is the same as going anticlockwise around the curve.
8.2 General Forms of Cauchy’s Theorem and Integral For-

mula
We are now ready to state without proof the general form of Cauchy’s Theorem and
Cauchy’s Integral Formula.
Theorem 8.12. Let D be a domain, and f : D → C holomorphic. Then for every

cycle Γ in D, which is homologous to zero in D, and for every w ∈ D − Γ we have
General Form of Cauchy’s Theorem:

Z
f (z) dz = 0
Γ
and
General Form of Cauchy’s Integral Formula:

Z
f (z)
dz = 2πiI(Γ; w)f (w).
Γ −w
z
Example 8.13. This tells us if D is simply connected and f : D → C is holomorphic

then Z
f (z) dz = 0,
γ
for any closed contour γ in D. In particular, since we know starlike domains are simply
connected by Proposition 8.4, we reobtain the version of Cauchy’s Theorem for starlike
domains (Theorem 6.20) that we proved rigorously as a special case.
Example 8.14. Let us see why the above theorem also gives a generalization of CIF
that we have seen. Indeed if we take D = Br (a), the open disc, and we consider the
closed contour γ(t) = a + ρe2πit , t ∈ [0, 1], with 0 < ρ < r, and w is a point with
|w − a| < ρ then the above theorem gives that
Z
f (z)
dz = 2πiI(γ; w)f (w) = 2πif (w).
γ z−w
Theorem 8.12 takes a much nicer form when γ is a simple closed curve. Given a
simple closed curve γ, we will say that a function f is holomorphic on Dγint ∪ γ if there
exists a domain D such that Dγint ∪ γ ⊂ D and that f is holomorphic on D. We note
that in particular we have that γ is homologous to zero in D since for every point w 6∈ D
we have that w 6∈ Dγint ∪ γ and so w ∈ Dγext and hence I(γ; w) = 0.
Theorem 8.15. Let γ be a simple closed positively oriented contour, and f a holomor-
phic function on Dγint ∪ γ. Then we have
Cauchy’s Theorem for simple closed contours:

Z
f (z) dz = 0
γ
and
Cauchy’s Integral Formula for simple closed contours:

Z
f (z)
For w ∈ Dγint dz = 2πif (w).
γ z −w
Proof. We just apply Theorem 8.12, by taking Γ equal to γ here. Note that I(γ; w) = 1
since w ∈ Dγint .
Example 8.16. Let γ be the boundary of the square with vertices in 1−i, −1−i, −1+i
and 1 + i, traced anti-clockwise. Consider the integral
Z
cos(z)
2
dz.
γ z(z + 2)
The contour√γ is a simple closed positively oriented contour and the roots of z 2 + 2,
which are ± 2i, lie outside γ. It follows that the function f (z) = cos(z)
z 2 +2
is holomorphic
int
on Dγ ∪ γ. Thus, by the CIF for simple closed contours (with ω = 0) we have
Z Z
cos(z) f (z) cos(0)
dz = dz = 2πi · f (0) = 2πi · 2 = πi.
γ z(z 2 + 2) γ z 0 +2
Chapter 9
Holomorphic functions on
punctured domains
We now have a very good understanding of how holomorphic functions behave on do-
mains with ‘no holes’, and how to integrate them around contours. In this chapter we
will study the behaviour of functions f : D → C on a domain D that we know are
holomorphic on D except for at a finite set of points. The simplest example of this is
when we are given a holomorphic function f on a punctured ball Br∗ (a) = Br (a) − {a}.
How might f behave near a? This problem leads to very interesting mathematics and
also practical applications.
9.1 Laurent series

Definition 9.1 (Laurent Series). A Laurent series is an infinite series of the form
∞
X
cn (z − a)n (9.1)
n=−∞
where the cn and a are complex numbers. The point a P is called the centre of the
−1 n
P∞ n=−∞ cn (zn− a) is called
Laurent series. Given a Laurent series as in (9.1), the sum
the principal part of the Laurent series. The sum n=0 cn (z − a) is called the
analytic part of the Laurent series.
A Laurent series converges at a point z ∈ C if and only if its principal and analytic
parts converge at z. Note, a Laurent series is a power series precisely when it has zero
principal part, and it is not defined at z = a unless this is the case. We now ask where
a Laurent series can converge. For this we define for 0 ≤ r < R ≤ ∞ and a ∈ C the set
Ar,R (a) := {z ∈ C : r < |z − a| < R}
118
CHAPTER 9. HOLOMORPHIC FUNCTIONS ON PUNCTURED DOMAINS 119
and call this an annulus of centre a and internal/external radii r, R.

Remark 9.2. Since we allow r = 0 or ‘R = ∞’, our definition of annulus Ar,R (a) includes
the sets BR ∗ (a), C∗ , and C − B (a).
r
Proposition 9.3. Given a Laurent series ∞ n

P
n=−∞ cn (z − a) that is not a power series
(i.e. cn 6= 0 for some n < 0), then either
1. the Laurent series never converges, or else;
2. there exist 0 ≤ r < R ≤ ∞ such that the series converges absolutely when r <
|z − a| < R and diverges when either |z − a| < r or |z − a| > R. In this case, the
annulus Ar,R (a) is called the annulus of convergence of the Laurent series.
Proof. By definition, the Laurent series converges absolutely if and only if both of
∞
X −1
X
F1 (z) = cn (z − a)n , F2 (z) = cn (z − a)n
n=0 n=−∞
converge absolutely. The analytic part F1 (z) is a power series that we already under-
stand (via Theorem 5.15). If F1 (z) only converges when z = a, then the Laurent series
never converges since F2 is not defined at a. Otherwise, let 0 < R ≤ ∞ be the radius
of convergence of F1 .
For the principal part F2 , introduce a new variable w = (z − a)−1 . Then
−1
X ∞
X
n
F2 (z) = cn (z − a) = c−n wn = F̃ (w)
n=−∞ n=1
is a power series in the variable w with center 0. If this power series converges only at
w = 0 then F2 never converges. So suppose that doesn’t happen and let 0 < R0 ≤ ∞
be the radius of convergence of this power series. Let
(
1/R0 if 0 < R0 < ∞
r= .
0 if R0 = ∞
Then the series F2 (z) = −1 n 0

P
n=−∞ cn (z − a) converges absolutely when 0 < |w| < R ,
i.e. when |z − a| > r, and diverges when |w| > R0 , i.e. when |z − a| < r. This proves
either of case 1 or 2 occur in the statement of the proposition.
Corollary 9.4. Assume that a Laurent series has annulus of convergence Ar,R (a). Then
the Laurent series converges uniformly on any annulus Aρ1 ,ρ2 (a) with r < ρ1 < ρ2 < R.
Hence Laurent series converge locally uniformly in their annulus of convergence and so
Laurent series represent holomorphic functions in their annulus of convergence.
Proof. We use the same notation as in the previous proof. By Theorem 5.17 we know
that F1 (z) converges locally uniformly on Bρ2 (a) and F̃ (w) = F2 (z), as a power series
in w with radius of convergence R0 , converges uniformly on |w| < ρ−1 −1
1 , since ρ1 <
0
1/r = R , i.e. it converges uniformly when |z − a| > ρ1 . Hence both halves of the
Laurent series converge uniformly on Aρ1 ,ρ2 (a). A similar argument for holomorphicity
follows via Theorem 5.21.
Proposition 9.5. [Uniqueness of Laurent series] Let f (z) = ∞ n

P
n=−∞ cn (z − a) be a
Laurent series with annulus of convergence containing Ar,R (a), a ∈ C, 0 ≤ r < R ≤ ∞.
Then the constants cn are unique and are given, for any ρ ∈ (r, R), by
Z
1 f (z)
cm = dz, (m ∈ Z).
2πi |z−a|=ρ (z − a)m+1
In particular, the Laurent series expansion of f in Ar,R (a) is unique.
Remark 9.6. The corresponding statement about the uniqueness of constants in a power
(n)
series follows from the fact (for power series) that cn = f n!(a) . Since any Laurent series f
as above is not defined at z = a (so certainly isn’t differentiable there), we have to take
a slightly different approach.
Proof. For any m ∈ Z we have

Z Z ∞
f (z) X
m+1
dz = cn (z − a)n−m−1 dz
|z−a|=ρ (z − a) |z−a|=ρ n=−∞
X∞ Z
= cn (z − a)n−m−1 dz,
n=−∞ |z−a|=ρ
where the interchange

P∞of summationnand integration is justified by the uniform conver-
gence of the series n=−∞ cn (z − a) on |z − a| = ρ, by Corollary 9.4 and Theorem 5.8.
But by the usual combination of Cauchy’s Theorem/FTC (n > m), CIF (n = m) &
FTC (n < m) we have that
Z (
n−m−1 2πi, n = m
(z − a) dz = .
|z−a|=ρ 0 n 6= m.
Hence we have that the only sum that is not equal to zero in the last summation is the
one with n = m. That is,
Z
f (z)
m+1
dz = 2πi · cm (m ∈ Z).
|z−a|=ρ (z − a)
This shows two things. Firstly, that the integral in question does not depend on ρ.
Secondly, since the integral is entirely determined by f , if we had started with another
Laurent series f (z) = ∞ 0 n

P
n=−∞ cn (z − a) representing f on Ar,R (a) then by the same
argument we would have
Z
0 1 f (z)
cm = dz = cm
2πi |z−a|=ρ (z − a)m+1
and so the coefficients in a Laurent series are unique.
Theorem 9.7. [Holomorphic functions on annuli have Laurent series]

Let A = Ar,R (a) with 0 ≤ r < R ≤ ∞ be an annulus and let f : A → C be holomorphic.
Then there are unique cn in C such that
∞
X
f (z) = cn (z − a)n
n=−∞
for all z ∈ A, and the annulus of convergence of the Laurent series contains A. The
series above is called the Laurent series of f on A.
Proof. First fix ρ1 , ρ2 with r < ρ1 < ρ2 < R. We will first show that f has a convergent
Laurent series expansion on Aρ1 ,ρ2 (a). To that end, consider the cycle in A given by
Γ = γ1 + γ2 , where
γ1 (t) = a + ρ1 e−2πit , γ2 (t) = a + ρ2 e2πit , t ∈ [0, 1].
We have seen (see Important Example 8.9 above) that this cycle Γ is homologous to zero
in A. We may therefore apply Theorem 8.12 (General Form of CIF) to get an expression
R f (z)
for f (ω) for any ω ∈ A − Γ; namely 2πi · I(Γ; ω) · f (w) = Γ z−w dz. Moreover for
w ∈ Aρ1 ,ρ2 (a), we have by Lemma 8.3, Cauchy’s Theorem, and the CIF, that
Z Z
1 1 1 1
I(Γ; w) = I(γ1 ; w) + I(γ2 ; w) = dz + dz = 0 + 1 = 1.
2πi γ1 z − w 2πi γ2 z − w
Thus, we obtain for w ∈ Aρ1 ,ρ2 (a) that

Z Z Z
1 f (z) 1 f (z) 1 f (z)
f (w) = dz = dz + dz.
2πi Γ z − w 2πi γ1 z − w 2πi γ2 z − w
1
R f (z) 1
R f (z)
We set f1 (w) = 2πi γ1 z−w dz and f2 (w) = 2πi γ2 z−w dz. That is, we have proved
f (w) = f1 (w) + f2 (w) for w ∈ Aρ1 ,ρ2 (a). We will show that f1 (ω) gives rise to a con-
vergent principal part of a Laurent series, and f2 (ω) gives rise to a convergent analytic
part of a Laurent series. Together, this will produce a convergent Laurent series for f
on Aρ1 ,ρ2 (a).
We consider first f2 (w). For this we can apply the same argument as we saw to
prove the Cauchy-Taylor theorem. Namely, for any z ∈ γ2 we may write
∞
1 1 X (w − a)n
= w−a = ,
z−w (z − a)(1 − z−a )
(z − a)n+1
n=0
and use the fact that the geometric series is uniformly convergent on |z − a| = ρ2 , since
|w−a|
on this contour we have w−a
z−a = ρ2 < 1. Then
∞
(w − a)n
Z Z
1 f (z) 1 X
f2 (w) = dz = f (z) =
2πi γ2 z−w 2πi γ2 (z − a)n+1
n=0
∞ Z
X 1 f (z)
(w − a)n ,
2πi γ2 (z − a)n+1
n=0
where the interchange of summation andR integration is justified by thePuniform conver-
1
gence of the series. We set cn := 2πi f (z)
γ2 (z−a)n+1 , and hence f2 (w) = ∞ n=0 cn (w − a)
n
can be represented by a convergent power series on Aρ1 ,ρ2 (a).

For f1 we need to argue slightly differently. We write for z ∈ γ1
1
X (z − a)m ∞
1 1
=− = − w−az−a = − .
z−w w−z 1 − w−a (w − a)m+1
m=0
z−a
This series converges uniformly on |z − a| = ρ1 since there we have that w−a < 1. In
particular we have that
∞
(z − a)m
Z Z
1 f (z) 1 X
f1 (w) = dz = − f (z) dz
2πi γ1 z−w 2πi γ1 (w − a)m+1
m=0
∞ Z
X 1 f (z)
= − dz (w − a)−m−1 .
2πi γ1 (z − a)−m
m=0
We now set n = −m − 1, so that the above sum can be written as

−1
X
f1 (w) = cn (w − a)n ,
n=−∞
1
R f (z)
where cn = − 2πi γ1 (z−a)n+1 dz. This shows that f2 (ω) can be expressed as the principal
part of a convergent Laurent series on Aρ1 ,ρ2 (a).
Putting the expansions P∞for f1 (ω) and f2n(ω) together proves that we have a convergent
Laurent series f (w) = n=−∞ cn (w − a) valid for all w ∈ Aρ1 ,ρ2 (a). Finally, if had
r < ρ01 < ρ1 < ρ2 < ρ01 < R and we were to exchange ρ1 , ρ2 with ρ01 , ρ02 and repeat the
same argument as above, we would get a Laurent series for f valid on Aρ01 ,ρ02 (a). In
particular, this new Laurent series would give us another Laurent series for f on the
subset Aρ1 ,ρ2 (a). But by Proposition 9.5 this Laurent series would have to be the same
as the original one, and converge on all Aρ01 ,ρ02 (a). Hence the Laurent series for f we
obtained on Aρ1 ,ρ2 (a) actually converges on all of Ar,R (a), and is unique.
9.2 Classification of isolated singularities

Suppose that f is a holomorphic function on a punctured ball BR ∗ (a) = B (a) − {a}.
R
In this situation we say that f has an isolated singularity at a. What can happen
nearby to a? Note that since BR ∗ (a) is the ‘annulus’ A
0,R (a), we know that f has a
(unique) Laurent series P ∗
expansion in BR (a).
Let us write f (z) = ∞ n
n=−∞ cn (z − a) for the unique Laurent series representation
of f in BR∗ (a). Exactly one of the following cases must occur:
1. f has a removable singularity at z = a. If cn = 0 for all n ≤ −1, that is

the principal part of f is zero, then we say that
P f has a removable singularity at
z = a. In this situation we have that f (z) = ∞ n=0 cn (z − a) n for all z ∈ B ∗ (a).
R
z
Example 9.8. As an example we can take f (z) = e z−1 in BR
∗ (0). Then, by
uniqueness the Laurent series expansion is equal to

P∞ z n ∞
− 1 X z n−1
f (z) = n=0 n! = .
z n!
n=1
That is, there is no principal part (no negative powers of z). In this case, we can
actually say a lot more. We know that power series define holomorphic functions
wherever they converge, and one can check (using the ratio test) that the power
z n−1
series g(z) = ∞
P
n=1 n! converges on all of C. In particular, the function g(z) is
holomorphic on BR (0) and matches the values of f (z) on BR ∗ (0) ⊂ B (0). Hence,
R
by Theorem 7.20 we know g is the analytic continuation of f to BR (0). The
function g(z) has value 1 at z = 0, and so we may extend f (z) to a holomorphic
function on all of BR (0) by declaring f (0) = 1.
2. f has a pole at z = a. If there exists a k > 0 such that c−k 6= 0 and cn = 0 for
all n < −k then we say that f has a pole of order k at z = a. Notice that this
means the principal part is not zero, but contains only finitely many nonzero terms.
z
Example 9.9. As an example we can take f (z) = e z−1 2 on B1∗ (0). Then by
uniqueness the Laurent series is given by
P∞ z n ∞
− 1 X z n−2 1
f (z) = n=0 2n! = = z −1 + + . . . .
z n! 2
n=1
That is, f has a pole of order 1 at 0. We cannot analytically extend this function
by ‘plugging the hole’ as in the previous case.
3. f has an essential singularity at z = a. Finally, when there are infinitely

many n < 0 such that cn 6= 0 then we say that f has an essential singularity at
z = a. That is the principal part of f is an infinite series.
Example 9.10. As an example we can take f (z) = e1/z . Then by uniqueness we

have that the Laurent expansion in B1∗ (0) is given by
∞
X z −n
f (z) = .
n!
n=0
Hence there are infinitely many negative powers of z in the Laurent series. Sin-
gularities of this final type are extremely hard to deal with.
Next we will study the above three cases in some more detail.
9.3 Removable singularities

We will first show that having a removable singularity is the same as having a holomor-
phic extension across the singularity (that is, there exists an analytic continuation of f
to BR (a)), which explains the name ‘removable singularity’.
Lemma 9.11. Let f be a holomorphic function on BR ∗ (a). Then f has a removable
singularity at a if and only if f extends to a holomorphic function on BR (a).

Proof. If f extends to a holomorphic function f˜ on BR (a), then by Cauchy-Taylor
(Thm 7.1) f˜ is given by a power series on BR (a). This power series must agree with the
Laurent series of f on BR ∗ (a) since Laurent series are unique (Proposition 9.5). Hence
f has a removable singularity.

If f has a removable singularity at z = a then its Laurent series is given by a
convergent power series on BR ∗ (a). This means the disc of convergence of the power
∗
series must contain BR (a). Moreover, by Proposition 5.15 the disc of convergence of this
power series must in fact contain BR (a). Hence, this power series defines a holomorphic
extension of f to BR (a).
The following proposition gives a useful analytic condition for a function to have a
removable singularity.
Proposition 9.12. Let f be a holomorphic function on BR∗ (a) with R > 0 and a ∈ C.
Then f has a removable singularity at a if and only if
lim (z − a)f (z) = 0.

z→a
Proof. ( =⇒ ) We first assume that f has a removable singularity at z = a. Then

by Lemma 9.11 f extends to a holomorphic function f˜ on BR (a). In particular, f˜ is
continuous on BR (a) and so lim (z − a)f (z) = lim (z − a)f˜(z) = (a − a)f˜(a) = 0.
z→a z→a
( ⇐= ) We now assume that lim z→a (z − a)f (z) = 0. We have by Proposition 9.5,
P∞
for any n ∈ Z and ρ ∈ (0, R), f = n=−∞ cn (z − a)n with
Z
1 f (z)
cn = dz.
2πi |z−a|=ρ (z − a)n+1
so by the Estimation Lemma (Lemma 6.15)
1 f (z) (z − a)f (z)

0 ≤ |cn | ≤ 2πρ sup n+1
= ρ sup n+2
2π |z−a|=ρ (z − a) |z−a|=ρ (z − a)
1 1
≤ρ sup |(z − a)f (z)| = sup |(z − a)f (z)| .
ρn+2 |z−a|=ρ ρn+1 |z−a|=ρ
This tends to zero as ρ → 0 for all n ≤ −1, so by squeezing, cn is zero for negative n.
Hence f has a removable singularity at a.
ez −1 ez − 1
Example 9.13. We consider f (z) = z . Then lim z= lim ez − 1 = 0, so f
z→0 z z→0
has a removable singularity at z = 0 as we have already seen.
Proposition 9.12 has the following consequence due to Riemann.
Theorem 9.14. [Riemann extension theorem] Let f : BR ∗ (a) → C be holomorphic
and bounded. Then f has a removable singularity at z = a.

∗ (a) then
Proof. If |f (z)| ≤ M for all z ∈ BR
0 ≤ |(z − a)f (z)| ≤ |z − a|M → 0
as z → a, so by squeezing, limz→a (z − a)f (z) = 0. Hence by Proposition 9.12, f has a

removable singularity at a.
9.4 Poles
We now prove the following proposition giving equivalent conditions for f to have a
pole at a.
∗ (a) → C be holomorphic. The following are equivalent:
Proposition 9.15. Let f : BR
(i) f has a pole of order k at z = a,
(ii) f (z) = (z − a)−k g(z) where g : BR (a) → C is holomorphic and g(a) 6= 0.
(iii) There exists an 0 < r ≤ R and h : Br (a) → C holomorphic, with a zero at z = a

of order k such that
1
f (z) = , z ∈ Br∗ (a).
h(z)
Proof. (i) implies (ii): We can write f (z) = ∞ n

P
n=−k cn (z − a) , with c−k 6= 0. We
X∞ X∞
set h(z) := (z − a)k f (z) = cn (z − a)n+k = cm−k (z − a)m . In particular, the
n=−k m=0
Laurent series of h is a power series (so z = a is a removable singularity of h) and h can
be extended to a holomorphic function g : BR (a) → C by Lemma 9.11. Finally, we have
g(a) = c−k 6= 0 and on BR ∗ (a) we have f (z) = (z − a)−k h(z) = (z − a)−k g(z) as required.
(ii)
P∞ implies (i): Since g(z) is holomorphic, by Cauchy-Taylor we may write g(z) =
a (z − a)n with a = g(a) 6= 0. Then for z ∈ B ∗ (a) we have
n=0 n 0 R
∞
X ∞
X ∞
X
−k −k n n−k
f (z) = (z−a) g(z) = (z−a) an (z−a) = an (z−a) = am+k (z−a)m .
n=0 n=0 m=−k
P∞
Writing cm := am+k , we have c−k = a0 6= 0 and we see that f = m=−k cm (z − a)m
has a pole of order k at z = a.
(ii) implies (iii): Since g(z) is holomorphic and g(a) 6= 0, by the Principle of Isolated
Zeros (Proposition 7.19), there exists some 0 < r < R such that g(z) 6= 0 for z ∈ Br (a).
1
Hence h(z) := (z − a)k g(z) is holomorphic in Br (a), and since 1/g(a) 6= 0 we have that h
has a zero of order k at z = a. But then for all z ∈ Br∗ (a) we have that
1
= (z − a)−k g(z) = f (z).
h(z)
(iii) implies (ii): Since h has a zero of order k at a we may write h(z) = (z − a)k h1 (z)
for some holomorphic h1 with h1 (a) 6= 0. Moreover, since 1/h(z) = f (z) is holomorphic
in Br∗ (a) we must have that h(z) 6= 0 on Br∗ (a), and it follows that h1 (z) 6= 0 for
all z ∈ Br (a). Hence, the function g1 (z) := 1/h1 (z) is holomorphic on Br (a) with
k
g1 (a) 6= 0. Moreover, for z ∈ Br∗ (a) we have (z − a)k f (z) = (z−a) 1
h(z) = h1 (z) = g1 (z).
Thus, the function g1 satisfies the required properties on the smaller ball Br (a). Finally,
since g1 (z) is holomorphic on Br∗ (a) ⊂ BR ∗ (a) and matches the function (z − a)k f (z),
which is holomorphic on the larger domain BR ∗ (a), we can extend g to a holomorphic

1
function g on BR (a) with the same properties by Proposition 7.20.
Example 9.16. Let us find all zeros, poles and removable singularities of the function
tan(z) sin(z)
f (z) = = .
z z cos(z)
We note that the zeros of the numerator and at z = nπ where n ∈ Z. Moreover the
possible singularities can be at the zeros of the denominator, that is at (n + 1/2)π for
n ∈ Z (the zeros of the cosine) and z = 0. We first check the zeros. If n 6= 0 then for
z = nπ we have that the denominator is non zero and the numerator vanishes, so they
are indeed zeros of f . To compute their order we just need to notice that f (nπ) = 0
and
cos(z) sin(z)(cos(z) − z sin(z)) 1
f 0 (nπ) = − = 6= 0.
z cos(z) (z cos(z))2 z=nπ nπ
That is, f has zeros of order one at z = nπ for n ∈ Z6=0 . For z = 0 both numerator
sin(z)
and denominator vanish. But then lim zf (z) = = 0. That is, z = 0 is a
z→0 cos(z)
removable singularity of f by Proposition 9.12. Finally for the rest, we see that
1
z = (n + 1/2)π are zeros of first order of cos(z), and hence also for the function f (z) .
Hence, by Proposition 9.15(iii) we have that there the function f has poles of order
one at z = (n + 1/2)π.
∗ (a) → C be holomorphic. Then f has a pole at z = a if
Proposition 9.17. Let f : BR
and only if
lim |f (z)| = ∞.
z→a
Proof. Assume f has a pole at z = a. Then by Proposition 9.15 (ii),
|f (z)| = |(z − a)−k | · |g(z)|, (g(a) 6= 0)
where g is holomorphic. In particular limz→a |f (z)| = limz→a |(z − a)−k | · |g(z)| = ∞
since limz→a |(z − a)−k | = ∞ and g(a) 6= 0. Conversely, if limz→a |f (z)| = ∞ then there
exists r > 0 such that f (z) 6= 0 for all z ∈ Br∗ (a). Hence f1 is holomorphic on Br∗ (a).
z−a 1
Moreover 0 ≤ f (z) = |z − a| · |f (z)| → 0 · 0 = 0 as z → a, and so by Proposition 9.12
1
we have that f has aremovable singularity at z = a. By Lemma 9.11 there exists
1
h(z) holomorphic in Br (a), such that h(z) = f (z) for z ∈ Br∗ (a). As limz→a |f (z)|
1
=0
1
we have h(a) = 0. By Proposition 9.15 (iii) we conclude that f (z) = h(z) has a pole
at z = a.
9.5 Essential singularities

By Proposition 9.12 and Proposition 9.17 we now conclude that for an f as above, if the
limit limz→a |f (z)| exists in the broader sense that it is either in R or equal to ∞ , then
f either has a pole or a removable singularity at a. That is, the limit does not exist (in
this broader sense) when f has an essential singularity at z = a. Actually even more
can be said about the behaviour of f around z = a when a is an essential singularity.
Theorem 9.18. [Casorati-Weierstrass] Let f : BR ∗ (a) → C be holomorphic, with an
essential singularity at a. Then for all w ∈ C and for all 0 < r < R and all > 0 there
exists a z ∈ Br∗ (a) such that
f (z) ∈ B (w).
Proof. We argue by contradiction. Assume not, then there exists w ∈ C, 0 < r < R
and > 0 such that
f (z) 6∈ B (w), ∀z ∈ Br∗ (a).
1
We consider the function g(z) := f (z)−w . Note that this function is holomorphic in Br∗ (a)
and since |f (z) − w| ≥ for every z ∈ Br∗ (a), we have that g(z) is bounded there, since
1 1
f (z)−w ≤ . By the Riemann extension theorem (Theorem 9.14), g has a removable
singularity at z = a and hence extends to a holomorphic function on Br (a). We have for
z ∈ Br∗ (a) that f (z) = w+ g(z)
1
= w g(z)+1
g(z) . If g(a) 6= 0 then f has a removable singularity
at a by Proposition 9.12 (since limz→a (z − a)f (z) = (a − a)(wg(a) + 1)/g(a) = 0). If,
g(z)
on the other hand, g(a) = 0, the function h(z) := wg(z)+1 is holomorphic at z = a and
has a zero at z = a (since h(a) = g(a)/(wg(a) + 1) = 0/(0 + 1) = 0), so f = 1/h has a
pole at z = a by Proposition 9.15(iii). This is a contradiction to f having an essential
singularity at a.
Actually, a stronger result is true, the proof of which is beyond the scope of the course.
Theorem 9.19 (Big Picard Theorem). Let f : BR ∗ (a) → C be holomorphic, with an
essential singularity at z = a. There is some b ∈ C such that for all r with 0 < r < R,
f (Br∗ (a)) ⊃ C − {b}.

1
Example 9.20. One can see from the example e z that we really do sometimes need
1
to remove a point b from C. In this case that point is b = 0, since e z is never zero for
all values of z. Then the big Picard Theorem says that for all r (no matter how small
ones) we have that
1
e z : Br∗ (0) → C − {0}
is surjective.
Chapter 10
Cauchy’s Residue Theorem
10.1 The Residue Theorem

We will now see how we can use the machinery of the Laurent series expansion to
compute contour integrals.
The following definition that captures functions that are not too far away from being
holomorphic.
Definition 10.1 (Meromorphic). Let D be a domain. We say a function f is mero-
morphic on D if f is holomorphic on D − S, where S ⊂ D has no non-isolated points,
and f has a pole at each element of S.
Remark 10.2. Let γ be a simple closed contour. Then we say that a function f is
meromorphic on Dγint ∪ γ if there exists a domain D that contains Dγint ∪ γ, and f
is meromorphic on D. Note, a meramorphic function can only have finitely many
isolated singularities inside a simple closed contour - roughly speaking, the Jordan
Curve Theorem (Theorem 8.11 this region is bounded, so cannot contain infinitely
many isolated singularities without at least one of them being a non-isolated point.
Definition 10.3 (Residues). Let D be a domain. If f is meromorphic on D with a
pole at a, and Laurent expansion
∞
X
f (z) = cn (z − a)n
n=−k
around z = a, then the residue of f at a is denoted by Resz=a (f ) and defined by
Resz=a (f ) := c−1 .
We now ready to state one of the big theorems of the course. It is very powerful in
its applications.
129
CHAPTER 10. CAUCHY’S RESIDUE THEOREM 130
Theorem 10.4. [Cauchy’s Residue Theorem for simple closed contours] Let γ
be a positively oriented simple closed contour, and let f be meromorphic on Dγint ∪ γ.
Assume that f has no poles on γ, and has a finite number of poles1 inside γ, say
{a1 , a2 , . . . , am }. Then
Z m
X
f (z) dz = 2πi Resz=aj (f ).
γ j=1
Rules for calculating residues.

The above theorem indicates that in order to compute a complex integral of a mero-
morphic function along a simple closed contour, all we need to do is to find the location
of the poles inside the contour and then compute the residues of the function at those
poles. Before we see the proof of Cauchy’s residue theorem, and look at examples of it
in action, it will be useful to develop ways to calculate residues that are more efficient
than determining the whole Laurent series.
1. (Linear combinations) If f and g both have poles at z = a and A, B are complex
numbers, then Resz=a (Af +Bg) = AResz=a (f )+BResz=a (g) for A, B ∈ C. That’s
clear from the definition (we just add the Laurent series).
2. (Cover up rule for a pole of order 1) If f has a pole of order one (simple
pole) at z = a then Resz=a (f ) = limz→a (z − a)f (z). Indeed we have that
∞
X ∞
X
lim (z − a)f (z) = lim (z − a)( cn (z − a)n ) = lim cn (z − a)n+1 = c−1 .
z→a z→a z→a
n=−1 n=−1
Example 10.5. Let f (z) = z 21−9 = (z−3)(z+3)

1
. Then f has simple poles at z = ±3
and the residues can be calculated by the cover up rule, for example
z−3 1 1
Resz=3 (f ) = lim = lim = .
z→3 (z − 3)(z + 3) z→3 z+3 6
g(z)
3. (Simple zero on denominator) If f (z) = h(z) , with g, h holomorphic at z = a,
g(a) 6= 0 and h has a zero of order one (i.e. a simple zero), then
g(a)
Resz=a (f ) = .
h0 (a)
Indeed, we know that by Proposition 9.15, f has a pole of order one at z = a. So
we may write
X∞
f (z) = cn (z − a)n .
n=−1
1
Actually, this condition is always satisfied for meromorphic functions - see Remark 10.2.
Moreover we may write h(z) = ∞ n

P
n=1 dn (z − a) with d1 6= 0 since
P h has a zero of
order one at z = a. That is, h(z) = (z −a)h1 (z) where h1 (z) = ∞ d
n=1 n (z −a)n−1 ,
and h1 (a) = d1 6= 0. Hence

∞
g(z) X
(z − a) = (z − a)f (z) = cn (z − a)n+1 ,
h(z)
n=−1
or
∞
g(z) X
= cn (z − a)n+1 .
h1 (z)
n=−1
Evaluating at z = a we obtain
g(a)
= c−1 .
h1 (a)
Finally, note that h0 (a) = d1 = h1 (a).
Example 10.6. Consider the function f (z) = sin1 z . We’ll calculate the residue of
f at 0. Indeed, since sin(0) = 0 and sin0 (0) = cos(0) = 1, sin(z) has a simple zero
at 0 by Proposition 9.15. Hence we have Resz=0 (f ) = sin10 (0) = 11 = 1.
g(z)
4. (Poles of higher orders) If f (z) = (z−a) k for some k > 0 and g holomorphic
at z = a (possibly with g(a) = 0), then
g (k−1) (a)
Resz=a (f ) = .
(k − 1)!
P∞
Indeed if we write g(z) = n=0 cn (z − a)n we have that
∞
X ∞
X
f (z) = cn (z − a)n−k = cm+k (z − a)m .
n=0 m=−k
We are clearly interested in the constant ck−1 as that corresponds to the expo-
(k−1) (a)
nent −1. But then we have Resz=a (f ) = ck−1 = g (k−1)! , by the derivative formula
for the coefficients of the Taylor series (see Remark 7.3 or Corollary 5.22).
Proof of the Residue Theorem, and some examples.

We will now prove Cauchy’s Residue Theorem for simple closed contours.
Proof of Theorem 10.4. Let S = {a1 , . . . , am } be the poles of f in Dγint so that f is

holomorphic on (Dγint ∪ γ) \ S. By Theorem 9.7, for each pole aj there exists a Laurent
series for f about z = aj on some punctured ball around aj . Let
−1
X
fj (z) = cn,j (z − aj )n
n=−kj
be the principal part of f in the Laurent series expansion around z = aj . Since each
fj (z) is just a finite sum of powers of reciprocals of (z −aj ), it follows from Theorem 7.20
that each fj (z) extends to a unique holomorphic function f˜j on C − {aj }. In particular,
the contour integral of f˜j (z) around the contour γ exists and is well-defined.
We calculate that for n ≤ −1
(
2πi; n = −1 (by CIF for Simple Closed Contours)
Z
(z−aj )n dz =
γ 0; n < −1 (by FTC, as (z − a)−n has an antiderivative on Dγint ∪ γ).
So we get
Z −1
X Z
f˜j (z)dz = cn,j (z − aj )n dz = 2πic−1,j = 2πiResz=aj (f ).
γ n=−kj γ
Consider the auxiliary holomorphic function g : (Dγint ∪ γ) \ S → C defined by

m
X
g=f− f˜j .
j=1
At each point z = aj , the corresponding principal part f˜j cancels. Moreover, the other
functions f˜i (for i 6= j) are holomorphic at z = aj . It follows that the Laurent series of
g at z = aj has no principal part and g therefore has a removable singularity at each aj .
By Lemma 9.11, g therefore extends to a holomorphic function g̃ on Dγint ∪ γ. Therefore
by Cauchy’s Theorem for Simple Closed Contours (Theorem 8.15),
 
Z Xm Z Z
f − ˜
fj dz = g(z)dz = g̃(z)dz = 0.

γ j=1 γ γ
Finally, by rearranging we have

Z m Z
X
f (z)dz = f˜j (z)dz
γ j=1 γ
Xm
= 2πiResz=aj (f ).
j=1
This proves the theorem.

Example 10.7. Let us compute the integral

Z
1
dz,
|z|=ρ sin z
for 0 < ρ < π.

The function f (z) = sin1 z is meromorphic on C with poles at the zeros of sin z, i.e.
at kπ for k ∈ Z. The only pole inside the contour is at 0. We calculated (in Rule 3.
example) that Resz=0 (f ) = 1. Therefore
Z
1
dz = 2πiResz=0 (f ) = 2πi.
|z|=ρ sin z
Example 10.8. Let us compute the integral

ez
Z
2 3
dz
γ z +z
where γ is any positively oriented simple closed contour with 0 ∈ Dγint and −1 ∈ Dγext .
z ez
We write f (z) = z 2e+z 3 = z 2 (z+1) . That is, f is meromorphic on C with poles at
z = 0 and z = −1. By Theorem 10.4 (Residue theorem for simple closed contours),
ez
Z
2 3
dz = 2πiResz=0 (f ).
γ z +z
ez
But Resz=0 (f ) is equal (by Rule 4 above) to g 0 (0) where g(z) = z+1 . So we calculate
ez (z+1)−ez
g 0 (z) = (z+1)2
and hence g 0 (0) = 0. Hence the integral is equal to zero.
10.2 Calculation of real integrals

10.2.1 Part 1: Rational functions
We now turn to one of the very important applications of complex analysis, namely
computation of real integrals. We start with the following example.
Example 10.9 (Integrals of rational functions of sin and cos). Show that
Z 2π
dθ 2π
=√ , −1 < a < 1.
0 1 + a sin(θ) 1 − a2
Solution. The equality is clear if a = 0, so we may now assume that a 6= 0. We
write
eiθ − e−iθ
sin(θ) = ,
2i
and hence Z 2π Z 2π
dθ dθ
= iθ −e−iθ
0 1 + a sin(θ) 0 1 + ae 2i
Now we claim that the last integral is equal to the complex integral
Z
1 dz
z−z −1 iz
, γ(θ) = eiθ , θ ∈ [0, 2π].
γ 1+a 2i
Indeed by applying the very definition of the complex integral we have

Z Z 2π Z 2π
1 dz 1 1 iθ dθ
z−z −1 iz
= eiθ −e−iθ ieiθ
ie dθ = iθ −iθ .
γ 1+a 2i 0 1+a 2i 0 1 + a e −e 2i
That is, it is enough to compute the complex integral,

Z
1 dz
z−z −1 iz
.
|z|=1 1 + a 2i
After some easy algebraic manipulations we get that,

Z Z
1 dz 2
z−z −1 iz
= 2 2i
dz
|z|=1 1 + a 2i |z|=1 a(z + a z − 1)
2
We set f (z) = a(z 2 + 2i z−1)
. The poles of f (z) are at the zeros of the quadratic polynomial
a
on the denominator. In particular we may write
2
f (z) = ,
a(z − z1 )(z − z2 )
√ √
2 2
where z1 = −1+ a1−a i and z2 = −1− a1−a i. Since |a| < 1, we have that
√
1+ 1 − a2
|z2 | = > 1,
|a|
and since z1 z2 = −1 we have |z1 ||z2 | = 1 and so |z1 | < 1. By applying the residue
theorem we obtain Z
f (z) dz = 2πiResz=z1 (f ).
|z|=1

But Resz=z1 (f ) = Resz=z1 a(z−z12)(z−z2 ) = a(z12−z2 ) , where we applied Rule 2 (cover
up rule) for computing residues. Hence
Z 2π Z
dθ 2 1 2π
= f (z) dz = 2πi = 2πi √ =√ .
0 1 + a sin(θ) |z|=1 a(z1 − z2 ) i 1 − a2 1 − a2
Remark 10.10. This method can be used to compute real integrals that involve sin and
cos. Indeed if we want to compute an integral of the form
Z 2π
F (sin(θ), cos(θ)) dθ,
0
P (x,y)
where F (x, y) = Q(x,y) is some rational function on x and y (i.e. P and Q are polyno-
mials in two variables), then we can perform the “change of variable” z = eiθ to obtain
(the explanation is as in the example) dθ = dz
iz and
z − z −1 z + z −1
sin(θ) = , cos(θ) = ,
2i 2
to obtain
2π
z − z −1 z + z −1
Z Z
dz
F (sin(θ), cos(θ)) dθ = F , .
0 |z|=1 2i 2 iz
Then we proceed by computing the complex integral using the Residue Theorem.
Example 10.11 (Integrating rational functions). Evaluate the integral

∞ R
x2 x2
Z Z
dx := lim dx.
0 x6 + 1 R→∞ 0 x6 + 1
2
Solution. We consider the complex function f (z) := z 6z+1 . It is meromorphic on C with
poles at the zeros of the denominator. These are

π 2kπ
zk = exp(i + ), k = 0, 1, 2, 3, 4, 5.
6 6
Note that they are distinct, so f has simple poles there. Moreover note that none of
them are real. Let now R > 1 (anything larger than the (largest) absolute value of the
poles will do), and consider the D-shaped simple closed contour defined as
γR = LR ∪ CR ,
where LR is the straight line running from the point −R to R, and CR is the contour
defined as CR (θ) = Reiθ with θ ∈ [0, π]. Note that the D-shaped contour γR (for any
R > 1) includes the three roots z0 , z1 , z2 , and none of the poles of f (z) are on this
contour. By the residue theorem we have that
Z 2
X
f (z) dz = 2πi Resz=zk (f ).
γR k=0
We can compute the residues by applying the third rule, since the function z 2 does not
vanish at zk and z 6 + 1 has a zero of order one there. Hence
zk2 1
Resz=zk (f ) = 5 = 3.
6zk 6zk
We conclude that,
Z 2
X 2πi 1 1 1 πi 1 1 1
f (z) dz = 2πi Resz=zk (f ) = ( + + ) = ( − + ) = π/3.
γR 6 z03 z13 z23 3 i i i
k=0
By the very definition of γR (see Remark 6.9) we have that,

Z Z Z
f (z) dz = f (z) dz + f (z) dz = π/3. (10.1)
γR LR CR
R RR
But LR f (z) dz = −R f (x) dx. We now claim that
Z
lim f (z) dz = 0.
R→∞ CR
We have by the Estimation Lemma (Lemma 6.15)

Z
f (z) dz ≤ L(CR ) sup |f (z)| = πR sup |f (z)|.
CR z∈CR z∈CR
But for z ∈ CR , we have that |z| = R, and so
z2 |z|2 R2
|f (z)| = = ≤ ,
z6 + 1 |z 6 + 1| R6 − 1
where we used the reverse triangle inequality to get |z 6 − 1| ≥ ||z|6 − 1| ≥ |R6 − 1|.
Hence
R3
Z
f (z) dz ≤ π 6 ,
CR R −1
which clearly tends to zero as R → ∞. Hence after taking the limit in (10.1) we have
Z R Z
lim f (x) dx + lim f (z) dz = π/3,
R→∞ −R R→∞ CR
R
and since limR→∞ CR f (z) dz = 0,
Z R
lim f (x) dx = π/3.
R→∞ −R
RR RR
Now we note that f (−x) = f (x), i.e. f is even. Hence −R f (x) dx = 2 0 f (x) dx, and
hence Z R
lim f (x) dx = π/6.
R→∞ 0
This finishes the current example.

Remark 10.12. The above method can be used to compute integrals of the form
Z R
p(x)
lim dx,
R→∞ −R q(x)
where deg(q) ≥ deg(p) + 2 and q(x) has no real roots. Indeed we just set f (z) := p(z)
q(z)
and consider as above the D-shaped contour γR = LR ∪ CR ., with R larger than the
absolute value of all complex rootsR of q(z). The condition on the degrees of p and q
allow us to establish that limR→∞ CR f (z) dz = 0.
10.2.2 Part 2: Principal value integrals.

Before we proceed to the next example, we pause for a minute to discuss a bit more
the notion of improper integrals. For a real function f : R → R we define the improper
integral Z ∞ Z r Z 0
f (x) dx := lim f (x) dx + lim f (x) dx.
−∞ r→∞ 0 s→∞ −s
We say that the limit exists in R, that is the integral converges, if both limits exist in R.
We note that this is not always equal to
Z r
lim f (x) dx.
r→∞ −r
R∞
Indeed if we compute −∞ x dx we see that this is
r 0 r 0
x2 x2 r2 s2
Z Z
= lim x dx + lim x dx = lim + lim = lim + lim − ,
r→∞ 0 s→∞ −s r→∞ 2 0
s→∞ 2 −s
r→∞ 2 s→∞ 2
but neither of the final two limits exist, so the improper integral is undefined. On the
other hand Z r
lim x dx = lim [x2 /2]r−r = lim 0 = 0
r→∞ −r r→∞ r→∞
Rr
exists, and it is zero. In general the limit limr→∞ −r f (x) dx is called the Cauchy
R∞ R∞
Principal Value of −∞ f (x) dx and it is denoted by P.V. −∞ f (x) dx. Clearly if
R∞ R∞
−∞ f (x) dx exists, then so exists P.V. −∞ f (x) dx and they are equal (by COLT). In
the case of f (x) = f (−x), that is f is even, then we have that
Z ∞ Z r Z r
P.V. f (x) dx = lim f (x) dx = 2 lim f (x) dx
−∞ r→∞ −r r→∞ 0
Z r Z 0
= lim f (x) dx + lim f (x) dx
r→∞ 0 s→∞ −s
Z ∞
= f (x) dx,
−∞
so (when f is even) if the Cauchy Principal Value exists then so exists also the improper
integral, and they are equal.
The following lemma is useful to know for some examples. We will often use it in
place of the Estimation Lemma for evaluating the contour integral around the curved
part CR of a D-shaped contour.
Lemma 10.13. [Jordan’s Lemma] If for some r > 0 f is holomorphic in D := {z :
|z| > r} and zf (z) is bounded in D, then for any α > 0 we have that
Z
lim f (z)eiαz dz = 0,
R→∞ CR
where CR (θ) = Reiθ with θ ∈ [0, π].

Proof. By the assumption there exists an M > 0 such that |zf (z)| ≤ M for all z with
|z| > r. Pick a z with |z| = R > r. Then,
exp(iαReiθ ) = eiαR(cos(θ)+i sin(θ)) = e−αR sin(θ) eiαR cos(θ) = e−αR sin(θ) .
We note that
2θ
sin(θ) ≥ , θ ∈ [0, π/2].
π
Indeed, if we consider the curve of sin(θ) we see that the line y = 2θ
π goes through the
points (0, 0) and (π/2, 1). But the function y = sin(θ) is convex in the interval [0, π/2]
hence the line must be always under the graph of sin(θ). We now can establish the
following, Z π
π
e−αR sin(θ) dθ < ,
0 αR
for any R > 0.
First we note that since sin(π − θ) = sin(θ) we have that
Z π Z π/2
−αR sin(θ)
e dθ = 2 e−αR sin(θ) dθ.
0 0
But then
Z π/2 Z π/2 2αRθ π π
2 e−αR sin(θ) dθ ≤ 2 e− π dθ = 2 (1 − e−αR ) < .
0 0 2Rα αR
Finally, combining these we have that

Z Z π
f (z)eiαz dz = f (Reiθ ) exp(iαReiθ )iReiθ dθ
CR 0
Z π
≤ f (Reiθ )Reiθ · exp(iαReiθ ) dθ
0
Z π
≤M exp(iαReiθ ) dθ
Z0 π
π
=M exp(−αR sin(θ)) dθ < M .
0 αR
In particular if we let R → ∞ then we get that

Z
lim f (z)eiαz dz = 0.
R→∞ CR
Example 10.14 (Principal value integrals, use of Jordan’s Lemma). Compute the
integral Z ∞
x sin(x)
2
dx
−∞ x + 1
Solution. Note that the integrand is even, hence the value of this integral is equal to
Z ∞ Z R
x sin(x) x sin(x)
P.V. dx = lim dx.
−∞ x2 + 1 R→∞ −R x2 + 1
We consider the function f (z) = z 2z+1 = (z−i)(z+i)

z
. For R > 1, we consider the complex
iz
R
integral γR f (z)e dz, where γR = LR ∪ CR is the D-shaped curve we considered in the
previous example. The reason for this is that
Z Z R Z R Z R
iz
f (z)e dz = f (x)(cos(x)+i sin(x)) dx = f (x) cos(x) dx+i f (x) sin(x) dx,
LR −R −R −R
and, after equating imaginary parts, we have

Z R Z
iz
lim f (x) sin(x) dx = lim Im f (z)e dz .
R→∞ −R R→∞ LR
To calculate the latter integral, we first have by Cauchy’s Residue Theorem that
Z
f (z)eiz dz = 2πiResz=i eiz f (z) ,

γR
since the only pole of f (z)eiz that lies inside the curve γR is for z = i. By Rule 2
(cover-up rule) for calculating residues we have that
2
iei e−1
Resz=i eiz f (z) = lim (z − i)f (z)eiz =

= ,
z→i 2i 2
and so
e−1
Z
f (z)eiz dz = 2πi · = πie−1 .
γR 2
In particular (by splitting the contour into its two parts) we obtain
Z Z Z
iz
f (z)e dz + iz
f (z)e dz = f (z)eiz dz = πie−1
LR CR γR
Thus, we have that

Z R Z Z
iz −1 iz
f (x) sin(x) dx = Im f (z)e dz = πe − Im f (z)e dz .
−R LR CR
If we now take the limit R → ∞ we obtain,

Z ∞ Z
x sin(x) −1 iz
2
dx = πe − lim Im f (z)e dz .
−∞ x + 1 R→∞ CR
We now claim that limR→∞ CR f (z)eiz dz = 0, and hence in particular

R
Z
iz
lim Im f (z)e dz = 0.
R→∞ CR
This will follow from Jordan’s Lemma (Lemma 10.13) with α = 1. Let us check that
the function f (z) = z 2z+1 meets the requirements. Indeed, with any fixed r > 1 we have
that for any z with |z| = R > r,
z2 R2 1
|zf (z)| = ≤ = .
2
z +1 2
R −1 1 − R12
√ √
In particular if we pick r = 2 thenR we have that |zf (z)| ≤ 2 for all |z| > 2.
Hence indeed we have that limR→∞ CR f (z)eiz dz = 0 in the example above, and
R ∞ x sin(x) −1
−∞ x2 +1 dx = πe . This finishes the current example.
R∞ x sin(x)
Remark 10.15. The same method that we used to compute the integral −∞ x2 +1 dx
can be used to calculate integrals of the form
Z ∞ Z ∞
P.V. f (x) sin(αx) dx, P.V. f (x) cos(αx) dx,
−∞ −∞
when f is meromorphic in C with finitely many poles, none of which is real, and f
satisfies the requirements of Jordan’s Lemma. Examples of such functions are f (x) =
p(x)
q(x) , where p and q are polynomials with deg(q) ≥ deg(p) + 1 and q(x) has no real roots.
Note that, we can take R sufficiently large (larger than the absolute value of any of the
finitely many poles) and consider the integral
Z
f (z)eiαz dz,
γR
where γR = LR ∪ CR the usual D-shaped contour. This can be calculated using the
Residue Theorem. Then we proceed as in the example using the fact that
Z Z R Z R
iαz
f (z)e dz = f (x) cos(αx) dx + i f (x) sin(αx) dx,
LR −R −R
and using Jordan’s Lemma to show that limR→∞ CR f (z)eiαz dz = 0. Then we equate
R
real and imaginary parts to obtain the values of the integrals.
10.2.3 Part 3: Indented contours.

The following lemma lets us compute even more real integrals.
Lemma 10.16. [Indentation Lemma] Suppose that g is a meromorphic function

on C that has a simple pole at 0 and consider the contour C (θ) = eiθ for θ ∈ [0, π].
Then
Z
lim g(z) dz = πiResz=0 (g).
→0 C
Proof. We consider the Laurent expansion of g(z) around the isolated singularity z = 0,
∗ (0). We have
which is valid on some punctured ball BR
∞
X
g(z) = an z n
n=−1
where a−1 = Resz=0 (g). Then

a−1
g(z) = + h(z),
z
where h(z) is holomorphic. Moreover, so long as < R (so that the contour is contained
in the annulus of convergence of the Laurent series) we have
Z Z Z Z
a−1 1
g(z) dz = + h(z) dz = a−1 dz + h(z) dz.
C C z C z C
We now first claim that Z

lim h(z) dz = 0
→0 C
Indeed since h(z) is holomorphic on BR (0), it is in particular continuous, and hence

|h(z)| is bounded on the compact set B r (0) for any fixed r < R, by Theorem 2.30. In
particular, for such a fixed r > 0 there exists M > 0 such that |h(z)| ≤ M for all z ∈ C
with ≤ r (so the upper bound of h on C does not depend on ). We may assume
≤ r < R from here on since we are calculating a limit as → 0. Therefore, by the
Estimation Lemma Z
h(z) dz ≤ πM,
C
which goes to zero as → 0. Hence we established the claim. We now compute
Z Z π Z π
1 1 iθ
dz = iθ
ie dθ = i dθ = πi.
C z 0 e 0
Putting these together, we obtain

Z
lim g(z) dz = πia−1 = πiResz=0 (g).
→0 C
Example 10.17 (Use of indented contours). Evaluate the integral

Z ∞
sin(x)
dx.
0 x
Solution. Note that the strategy used in the last example does not apply here. The
iz
reason is of course that the function g(z) := ez has a pole at z = 0. In particular we
cannot consider the D-shaped contour of the previous example, since then a pole will
lie on the curve. We are going to define a different contour. We consider 0 < ρ < R
and then we define the curve γρ,R as
γρ,R = L2 ∪ C
fρ ∪ L1 ∪ CR ,
where
1. L2 is the straight line running from −R to −ρ,
fρ := −Cρ , with Cρ (θ) = ρeiθ , θ ∈ [0, π],

2. C
3. L1 is the straight line running from ρ to R,
4. CR (θ) = Reiθ , θ ∈ [0, π].
Contours of this form as usually called indented. For simplicity we write just γ for
iz
γρ,R . Note, that since the only pole of g(z) = ez is at z = 0, and this does not lie inside
or on the curve, we have by Cauchy’s Theorem for simple closed contours that
Z
g(z) dz = 0.
γ
And hence Z Z Z Z
g(z) dz + g(z) dz = g(z) dz − g(z) dz. (∗)
L1 L2 Cρ CR
However we have that

R −ρ
eix eix
Z Z Z Z
g(z) dz + g(z) dz = dx + dx =
L1 L2 ρ x −R x
R
eix e−ix
R
Z Z
dx − dx =
ρ x ρ x
R
eix − e−ix
Z Z R
sin(x)
dx = 2i dx.
ρ x ρ x
We want to consider the limits ρ → 0 and R → ∞ in (∗). For this we first show that
Z Z
1 iz
lim g(z) dz = lim e dz = 0.
R→∞ CR R→∞ CR z
1
Indeed this follows directly by Jordan’s Lemma for the function f (z) = z (and α = 1).
Clearly |zf (z)| = 1 is bounded. Hence we obtain that the limit is zero.
We now compute the limit Z
lim g(z) dz.
ρ→0 Cρ
For this we use the Indentation Lemma (Lemma 10.16). Indeed g(z) has a simple pole
iz
at 0, and the residue is, by the cover up rule, Resz=0 (g) = limz→0 z ez = limz→0 eiz = 1.
Hence the Indentation Lemma gives
Z
lim g(z) dz = πi.
ρ→0 Cρ
Now taking the limits ρ → 0 and R → ∞ in (∗) we obtain

Z ∞
sin(x)
2i dx = πi
0 x
and hence Z ∞
sin(x) π
dx = .
0 x 2
This finishes the current example.
Example 10.18 (Indented contours and branch cuts). We now consider another ex-
ample, that can be handled by using similar ideas as above. We show that
Z ∞
log(x) π
2 2
dx = (log(2) − 1).
0 (x + 4) 32
Solution. Note that the principle branch of log is not defined/not holomorphic
on the negative real axis. This is a problem with regards to our D-shaped contours
(whether they are indented or not). To alleviate this, we consider the function
log(z)
f (z) = ,
(z 2 + 4)2
where the branch of the logarithm is given by
log(z) = log |z| + i arg(z),
with |z| > 0 and −π/2 < arg(z) < 3π/2. That is, the domain of definition is the complex
plane with the negative imaginary axis removed, i.e. D = C − iR≤0 . The function f
is then meromorphic on D, with a single pole z = 2i (the other −2i lies outside the
domain D). We choose 0 < ρ < 2 < R and can now consider the same indented contour
γ := γρ,R as in the previous example. Note the only pole in D is included within the
simple closed contour γ, and hence by the Residue Theorem we have
Z
f (z) dz = 2πiResz=2i (f ),
γ
where  
log(z)
log(z) (z+2i)2
Resz=2i (f ) = Resz=2i = Resz=2i  =
(z 2 + 4)2 (z − 2i)2
0
log(z) π 1 − log(2)
= +i ,
(z + 2i)2 z=2i 64 32
where we have used the 4th rule to compute the residue. Hence we obtain
π2
Z
π
f (z) dz = (log(2) − 1) + i ,
γ 16 32
and so,
π2
Z Z Z Z
π
f (z) dz + f (z) dz = (log(2) − 1) + i + f (z) dz − f (z) dz. (∗∗)
L2 L1 16 32 Cρ CR
But we have Z Z R
log(x)
f (z) dz = dx,
L1 ρ (x2 + 4)2
and Z Z −ρ
log(x)
f (z) dz = dx.
L2 −R (x2 + 4)2
Observe now that for x < 0 we have x = |x|eiπ , and hence the above integral can be
rewritten as, Z −ρ Z R
log(|x|) + iπ log(x) + iπ
2 2
dx = dx,
−R (x + 4) ρ (x2 + 4)2
Putting together we have
Z Z Z R Z R
log(x) dx
f (z) dz + f (z) dz = 2 + iπ .
L1 L2 ρ (x2 + 4)2 ρ (x2 + 4)2
taking real parts in equation (∗∗) we have that

!
Z R Z Z
log(x) π
2 = (log(2) − 1) + Re f (z) dz − Re f (z) dz (∗ ∗ ∗)
ρ (x2 + 4)2 16 Cρ CR
And we consider the limits ρ → 0 and R → ∞. We first Claim:

Z
lim f (z) dz = 0,
ρ→0 Cρ
and hence so also the real part of it. Indeed since | arg(z)| ≤ π for all z ∈ Cρ , we have
by the Estimation Lemma that for any ρ sufficiently small (i.e. ρ < 1 will do so that
log ρ < 0)
| log |z|| + | arg(z)| − log(ρ) + π

Z
log(|z|) + i arg(z)
f (z) dz ≤ πρ sup ≤ πρ sup ≤ .
Cρ z∈Cρ (z 2 + 4)2 z∈Cρ (|z| 2 − 4)2 (4 − ρ2 )2
π(πρ−ρ log(ρ))
But limρ→0 (4−ρ2 )2
= 0. Indeed by (real) l’Hôpital’s rule we have that
log(ρ) 1/ρ
lim ρ log(ρ) = lim = lim = − lim ρ = 0.
ρ→0 ρ→0 1/ρ ρ→0 −1/ρ2 ρ→0
So the claim is established.

Similarly we have that Z
lim f (z) dz = 0,
R→∞ CR
and hence so also the real part of it. Indeed for R large enough, by a similar argument
as the one above, the Estimation Lemma gives
Z
log(R) + π
f (z) dz ≤ πR
CR (R2 − 4)2
which goes to zero for R → ∞. Indeed we may write

π log(R)
log(R) + π R + R
lim πR = lim π ,
R→∞ (R2 − 4)2 R→∞ (R − R4 )2
log(R) 1/R
and observe via l’Hôpital (or “powers beat logs”) we have lim = lim = 0.
R→∞ R R→∞ 1
Hence putting this all together, equation (∗ ∗ ∗) gives
Z ∞ Z R Z
log(x) log(x) 1 log(z) π
dx = lim lim dx = dz = (log(2) − 1).
0 (x2 + 4)2 R→∞ ρ→0 ρ (x2 + 4)2 2 γ (z 2 + 4)2 32
This finishes the current example, which is our final example of Complex Analysis
solving real integrals.
10.3 The Argument Principle and Rouché’s Theorem

We have now seen various ways to use the residue theorem to compute integrals. We
now show a theoretical application of the residue theorem which is also known as the
argument principle. We start with the following lemma.
Lemma 10.19. Let f be meromorphic on a domain D with a zero or pole of order

0 (z)
k > 0 at a ∈ D. Then the function ff (z) has a simple pole at a and
(
f0

k if f has a zero at z = a;
Resz=a =
f −k if f has a pole at z = a.
Proof. Assume f has a zero of order k > 0 at a. Then we may write
f (z) = (z − a)k g(z), g(a) 6= 0

∗ (a), R > 0, and g is holomorphic on B (a). But then
for all z in some BR R
f 0 (z) k(z − a)k−1 g(z) + (z − a)k g 0 (z) k g 0 (z)

= = +
f (z) (z − a)k g(z) z−a g(z)
∗ (a). Since g(z) is holomorphic on B (a) and g(a) 6= 0, g0
for all z ∈ BR R g has a removable
singularity at a and so we have
0
g0

f k
Resz=a = Resz=a + = k.
f z−a g
Similarly one shows the case of f having a pole of order k > 0 at z = a, where using
Proposition 9.15 we now write
f (z) = (z − a)−k g(z),
with g holomorphic around a and g(a) 6= 0. Then
f 0 (z) −k(z − a)−k−1 g(z) + (z − a)−k g 0 (z) k g 0 (z)

= = − + ,
f (z) (z − a)−k g(z) z−a g(z)
and the remainder of the proof is identical.
z
Example 10.20. Consider f (z) = ez , which has a pole of order one at z = 0. We have
z z 0 (z) z z)
f 0 (z) = e z−e
z2
and hence ff (z) = z(ez 2z−e
ez
= z−1 1
z = − z + 1, which has clearly a pole of
order one at z = 0 and the residue is equal to −1.
Observation: Note that in general when f is a meromorphic function, then the function
0 (z)
h(z) := ff (z) is also meromorphic, and its poles are at
• the zeros and poles of the function f (by Lemma 10.19),
• the poles of the function f 0 (z). But these are precisely the poles of f (z).
Indeed if z = a isPa pole of f 0 (z), then it has to be also
P∞a pole for f (z), since
otherwise f (z) = ∞ n=0 n c (z − a)n , and hence f 0 (z) =
n=1 ncn (z − a) n−1 , and
hence z = a could not be a pole for f 0 (z). Conversely, we have seen that poles of
f are always poles of f 0 on (see Problem Sheet 16 Q7).
0
So we summarize: The poles of h(z) = ff (z)
(z)
are precisely the zeros and poles of f (z),
and by Lemma 10.19 they are all simple. We now prove the following theorem, which
is also known as the argument principle.
Theorem 10.21. [The Argument Principle] Let γ be a simple closed contour, pos-
itively oriented, and f meromorphic on Dγint ∪ γ. Assume f has no zeros or poles on γ,
and write Zf for the number of zeros of f in Dγint (counted with multiplicities), and Pf
for the number of poles of f in Dγint (counted with multiplicities). Then
Z 0
1 f (z)
dz = Zf − Pf .
2πi γ f (z)
Remark 10.22. When we say ‘counted with multiplicities’ we mean that when we have
a zero of order k, we count k towards Zf . Similarly for a pole of order k we count k
towards Pf .
0
Proof. We set h(z) := ff (z)(z)
. Then h is a meromorphic function. By the previous
observation, the poles of h are precisely the zeros and poles of f , and since by assumption
none of these zeros or poles lie on γ we can apply the residue theorem to h to obtain
m
f 0 (z)
Z Z X
dz = h(z) dz = 2πi Resz=ai (h),
γ f (z) γ i=1
where ai are the poles of h inside γ. Again, these are precisely the zeros and poles of f ,
so by Lemma 10.19 we have that
(
ki if ai is a zero of order ki of f ;
Resz=ai (h) =
−ki . if ai is a pole of order ki of f .
Hence we obtain that indeed,

f 0 (z)
Z
1
dz = Zf − Pf .
2πi γ f (z)
Example 10.23. We consider the function
(z − 3)3 (z − 1)7 z 3
f (z) =
(z − i)4 (z + 4)5 (z − 3i)7
and we take γ(θ) = 27 eiθ ; θ ∈ [0, 2π]. Then clearly we have that the number of ze-
ros inside γ (counted with multiplicity) is Zf = 3 + 7 + 3 = 13 and the number of
poles inside γ (counted with multiplicity) is Pf = 4 + 7 = 11. So Zf − Pf = 2 and
1
R f 0 (z)
2πi γ f (z) dz = 13 − 11 = 2. Indeed, one can check that
f 0 (z) 3 7 3 4 5 7
= + + − − − ,
f (z) z − 3 z − 1 z z − i z + 4 z − 3i
and so by the Residue Theorem (and linearity, Rule 1, for residues) we get that indeed
Z 0
1 f (z)
dz = 3 + 7 + 3 − 4 − 7 = 2.
2πi γ f (z)
Observation: Let us now explain why Theorem 10.21 is called the argument principle.
Given a simple closed contour γ : [a, b] → C and a meromorphic function f such that
none of the zeros or poles of f are on γ, we define the closed contour (perhaps not
simple) Γf by
Γf := f ◦ γ ⊂ C − {0};
that is, the contour Γf is the composition of the contour γ followed by the function f .
Note that this is well-defined since none of the zeros or poles of f are on γ. Let’s say
we wish to know how many times this contour circles around a point in the plane, say,
the point z = 0. Then we see that
Z Z b Z 0
dw 1 f (z)
= f 0 (γ(t))γ 0 (t) dt = dz.
Γf w a f (γ(t)) γ f (z)
But we know from Lemma

R dw 8.3 that the first integral can be related to the winding
1
number of Γf via 2πi Γf w = I(Γf ; 0). Hence the Argument Principle tells us precisely
that this winding number may be written in the form
I(Γf ; 0) = Zf − Pf .
So, if we now recall that I(Γf ; 0) measures the number of times the contour Γf circum-
navigates the point z = 0, we see that quite remarkably the statement of the Argument
Principle is equivalent to saying that this number is obtained simply by summing the
number of zeros of f and taking away the number of poles of f (all counted with
multiplicities) that lie inside the curve γ.
Clearly, knowing the location of the zeros of a function is therefore very useful. To
aid us in this we can now prove the following important theorem.
Theorem 10.24. [Rouché’s theorem] Let γ be a simple closed contour, and let f, g
be holomorphic functions on Dγint ∪ γ. Suppose that
|f (z) − g(z)| < |g(z)| ∀z ∈ γ.
Then f (z) and g(z) have the same number of zeros (counted with multiplicities) inside γ.
Proof. Since both f and g are holomorphic on Dγint , they have no poles on γ. We set
h(z) := fg(z)
(z)
. We observe that we also have that h has no poles on γ. Indeed the poles
would have to be at zeros of g, but g has no zeros on γ: if g(z0 ) = 0 for some z0 ∈ γ, then
we have that |f (z0 )| < 0, which is not possible. Similarly, h has no zeros on γ, since
these would precisely be the zeros of f on γ, but f has no zeros on γ: if f (z0 ) = 0, then
we have that |g(z0 )| < |g(z0 )|. Moreover, h is the quotient of two holomorphic functions,
and any poles of h must be isolated (since they are zeros of the denominator g, and any
zeros of g are isolated by the Principle of Isolated Zeros, Proposition 7.19), and so h is
meromorphic on Dγint ∪ γ. Thus, we can apply the Argument Principle to all three of
the functions h, f and g.
Next, we notice that
f 0 (z) f (z)g 0 (z)
1
Z
h0 (z) 1
Z
g(z) − g 2 (z) 1
Z
f 0 (z)
Z
g 0 (z)

dz = f (z)
dz = dz − dz .
2πi γ h(z) 2πi γ 2πi γ f (z) γ g(z)
g(z)
So, by the Argument Principle (Theorem 10.21) applied to f and g we have

Z 0 Z 0
1 f (z) 1 g (z)
dz = Zf ; dz = Zg .
2πi γ f (z) 2πi γ g(z)
Thus
h0 (z)
Z
1
dz = Zf − Zg .
2πi γ h(z)
On the other hand, by the observation preceding the statement of this theorem, we also
have by the Argument Principle applied to h that
Z 0
1 h (z)
dz = I(Γh ; 0),
2πi γ h(z)
where Γh = h ◦ γ. The proof will therefore be complete upon establishing that we have
I(Γh ; 0) = 0. To that end, note that the premise
|f (z) − g(z)| < |g(z)| ∀z ∈ γ
implies
f (z) g(z) 1 |g(z)|

|h(z) − 1| = − = |f (z) − g(z)| < =1 ∀z ∈ γ.
g(z) g(z) |g(z)| |g(z)|
That is, we have Γh ⊂ B1 (1). Finally, since 0 6∈ B1 (1) and B1 (1) is starlike, it follows
from Proposition 8.4 that we indeed have I(Γh ; 0) = 0. We conclude Zf = Zg .
Example 10.25. Let us see how one can use Rouché’s Theorem to obtain information
on the location of zeros of functions. Consider the polynomial P (z) = z 4 + 6z + 3.
We know that this polynomial has 4 zeros in C (by a consequence of the Fundamental
Theorem of Algebra, Theorem 7.11). We are now able to say something more about
the location of these zeros. We consider the curve γ : |z| = 2. Then we note that if we
set g(z) = z 4 we have for all z ∈ γ,
|g(z)| = |z|4 = 16 > 15 = 6|z| + 3 ≥ |6z + 3| = |P (z) − g(z)|.
In particular by Rouché’s Theorem P and g have the same number of zeros within the
curve γ. But g(z) = z 4 clearly has only one zero, occurring at z = 0, and this zero
appears with multiplicity four (i.e., it is a zero of order 4). Hence we conclude that
the number of zeros of P inside γ, counted with multiplicity, is 4. Now we consider the
curve γ : |z| = 1, and set g(z) = 6z. Then for any z ∈ γ we have
|g(z)| = 6|z| = 6 > 4 = 1 + 3 ≥ |z|4 + 3 ≥ |z 4 + 3| = |P (z) − g(z)|.
And again by the theorem above we have that P (z) and 6z have the same number of
zeros inside the curve γ. But clearly 6z has only one zero inside γ, namely at z = 0
again. Hence we conclude that P (z) has one zero of modulus less than 1, and has three
zeros (counted with multiplicity) of modulus between 1 and 2.
Example 10.26. We consider the function
f (z) = cos(πz) − αz m , α > eπ , m ∈ N.
We will show that this function has m zeros (counted with multiplicities) in the unit
disc. We start by setting g(z) := −αz m , and we consider the curve γ : |z| = 1. Then,
for all z ∈ γ we have |g(z)| = α > eπ . On the other hand, for all z ∈ γ we have
eiπz + e−iπz |eiπz | + |e−iπz |

|f (z) − g(z)| = | cos(πz)| = ≤ .
2 2
Since z ∈ γ, if we write z = x + iy, we have that x ∈ [−1, 1] and y ∈ [−1, 1]. In

particular we have that for all z ∈ γ,
|eiπz | = |eiπ(x+iy) | = e−πy ≤ eπ ,
and
|e−iπz | = |e−iπ(x+iy) | = eπy ≤ eπ .
Hence we have that for all z ∈ γ,
|f (z) − g(z)| ≤ eπ < α = |g(z)|.

By Rouché’s Theorem we have that the functions f (z) and g(z) have the same number
of zeros, counted with multiplicities, inside the unit disc. But clearly g(z) = αz m has
only the z = 0 as a zero, and this with multiplicity m. Hence f has also m zeros,
counted with multiplicity, inside the unit disc.
Lastly, we use Rouché’s Theorem to derive one further important (and remarkable)
result regarding holomorphic functions. Recall the most powerful result from metric
spaces that we have seen, Theorem 2.30, which stated that the image of a compact set
under a continuous function is compact. We can do better for holomorphic functions.
Theorem 10.27. [Open Mapping Theorem] Let D be a domain and f : D → C be

holomorphic and non-constant. Then f maps open sets in D to open sets in C. More
precisely, if U ⊂ D is open, then f (U ) is open.
Remark 10.28. The analogue of Theorem 10.27 is not true in Real Analysis. Indeed if
we consider the function f (x) = x2 , then, f (R) = [0, ∞), which is not open (see also,
the discussion on page 28).
Proof of Theorem 10.27. Note that f is not constant on U ; otherwise it would be con-
stant on some open ball in U (by the definition of openness) and then by analytic
continuation it would be constant on D (see Corollary 7.21).
We want to show f (U ) is open. This means given any b ∈ f (U ), we need to find a
δ > 0 such that Bδ (b) ⊂ f (U ). Unravelling this, for the δ we find we need to show that
for any b̃ ∈ Bδ (b) there is some ã in U with f (ã) = b̃.
Step 1: Finding δ: Since b ∈ f (U ), there is some a ∈ U such that f (a) = b. Let

g(z) = f (z) − b so that g(a) = f (a) − b = b − b = 0 and z = a is a zero of g. Then as
f is non-constant, g is too, and by the Principle of Isolated Zeros (Proposition 7.19),
we know that this zero of g at z = a is isolated. This means there is some > 0 with
B (a) ⊂ U and g(B∗ (a)) ⊂ C − {0}. In particular, since g is continuous on the compact
circle of half the radius

γ(t) = a + eit , t ∈ [0, 2π],
2
and is never zero on γ, the function |1/g(z)| is continuous and bounded on this circle. By
Theorem 2.30 the function |1/g(z)| must be bounded above on the circle. In particular,
there exists some δ > 0 such that
|g(z)| ≥ δ ∀ z ∈ γ.
This is the δ we will use.
Step 2: Showing this δ works (Finding ã from b̃): Now, given b̃ ∈ Bδ (b), we note that
|g(z) − (b̃ − b) − g(z)| = |b̃ − b| < δ ≤ |g(z)| ∀ z ∈ γ.
Therefore by Rouché’s Theorem (Theorem 10.24) the function h(z) := g(z) − (b̃ − b) has
the same number of zeros (counted with multiplicities) inside B/2 (a) as the function g.
Since g has precisely one zero inside this ball (at z = a) we know that h does too. So
there is some ã ∈ B/2 (a) with
0 = h(ã) = g(ã) − (b̃ − b) = f (ã) − b − (b̃ − b) = f (ã) − b̃.
So f (ã) = b̃ as required.
This completes our journey through Complex Analysis.

Complex Analysis

Uploaded by

Complex Analysis

Uploaded by

Complex Analysis II (Epiphany)

Katie Gittins & Stephen Harrap1

(Epiphany terms starts at section 6.3)

1 The Complex Plane and Riemann Sphere 1

5 Notions of Convergence and Power Series 64

5.3 Complex power series . . . . . . . . . . . . . . . . . . . . . . . . . 71

6 Complex integration over contours 77

7 Features of Holomorphic Functions 96

8 General form of Cauchy’s Theorem and C.I.F. 112

9 Holomorphic functions on punctured domains 118

10 Cauchy’s Residue Theorem 129

The Complex Plane and Riemann

1.1 Complex numbers

z1 x1 + iy1 (x1 + iy1 )(x2 − iy2 ) x1 x2 + y1 y2 x2 y1 − x1 y2

We immediately have a multiplicative inverse

How do we visualise the complex numbers?

R2 × R2 → R2 : ((x1 , y1 ), (x2 , y2 )) 7→ (x1 x2 − y1 y2 , x1 y2 + x2 y1 ).

Figure 1.1: Argand diagram.

Polar form/coordinates of a complex number

Figure 1.2: Polar coordinates.

2. arg(1/z) = − arg(z) mod 2π

3. arg(z̄) = − arg(z) mod 2π .

z1 z2 = r1 r2 (cos(θ1 + θ2 ) + i sin(θ1 + θ2 )) = r1 r2 ei(θ1 +θ2 ) .

In particular, multiplying by z2 constitutes an (anticlockwise) rotation of z1 by θ2 de-

z1 z2 = r1 r2 (cos θ1 + i sin θ1 ) (cos θ2 + i sin θ2 )

The geometric interpretations of addition, conjugation, and real/imaginary parts are

Corollary 1.5. 1. |z1 z2 | = |z1 | |z2 |,

2. De Moivre’s formula: (cos θ + i sin θ)n = cos(nθ) + i sin(nθ).

1. (Triangle inequality) |z1 + z2 | ≤ |z1 | + |z2 |

2. |z| ≥ 0 and |z| = 0 ⇐⇒ z = 0

3. max(|Re(z)|, |Im(z)|) ≤ |z| ≤ |Re(z)| + |Im(z)|

Simple complex functions and geometry (examples)

• Note that the equation |z − i| = 1 represents a circle centred at i of radius 1.

1.2 Exponential and trigonometric functions

exp(z) := ex (cos y + i sin y). (z = x + iy)

As shorthand we write exp(z) = ez .

2. ez1 +z2 = ez1 ez2 ,

3. ez = 1 ⇐⇒ z = 2πik for some k ∈ Z,

Proof. Most are straightforward to check. 3. is very important: exp(z) = 1 ⇐⇒

Definition 1.13 (Trigonometric functions). All as functions from C → C, we define

1.3 Logarithms and complex powers

has a solution z. Furthermore, if we write w = |w|eiφ with φ = Arg(w), then all

z = log |w| + i(φ + 2kπ) for k ∈ Z. (1.2)

Proof. If z is of the form as in (1.2) for some given k ∈ Z, then

ez = elog |w|+i(φ+2kπ) = elog |w| ei(φ+2kπ) = elog |w| ei(φ) = |w|ei(φ) = w,

by Proposition 1.9, Part 2. and Corollary 1.10. Thus, z is a solution.

We now come to an important topic called branch cuts.

z = log |w| + i(Arg(w) + 2πk)

log(z) := log |z| + i arg(z)

z w := exp(w log z).

For example, if w = 1/n and we use the principal branch we get

z 1/n = e(log |z|+iArg(z))/n = |z|1/n eiArg(z)/n .

Finding values of log and complex powers (examples)

Visualizing complex functions

Similarly, the left half plane HL = {z ∈ C : Re(z) < 0} is mapped to C \ R≤0 .

1.4 The Riemann Sphere and extended complex plane

The Riemann sphere

2.1 Metric spaces

• The modulus |z − w| of their difference in C.

• The distance in R3 between their stereographic representatives on the sphere S 2 .

• (D1) Positivity. d(x, y) ≥ 0 and d(x, y) = 0 ⇐⇒ x = y;

• (D2) Symmetry. d(x, y) = d(y, x);

• (D3) Triangle inequality. d(x, y) ≤ d(x, z) + d(z, y).

2. The Euclidean norm on Rn or Cn For vectors x in Rn (or Cn ), the function

3. • (a) Metrics induced from inner products in vector spaces

(3a); for example, for vectors x in Rn or Cn , the function

6. Riemannian (chordal) metric on Ĉ

defines a metric, called the discrete metric. It is easy to check (D1)-(D3). In

2.2 Open and closed sets

• The open ball Br (x) of radius r centred at x is

Br (x) := {y ∈ X : d(x, y) < r}.

• The closed ball B̄r (x) of radius r centred at x is

B̄r (x) := {y ∈ X : d(x, y) ≤ r}.

Visualizing balls (examples)

• A subset U ⊆ X is closed (in X) if its complement X \ U is open.

Open sets (examples/warnings)

∀ > 0 ∃δ > 0 such that x ∈ Bδ (x0 ) ⇒ f (x) ∈ B (f (x0 )),

x ∈ [2, 2 + δ) ⇒ f (x) = x − 1 ∈ [1, 1 + δ) ⊂ (1 − δ, 1 + δ) ⊂ (1 − , 1 + ),